Cambridge University Press

978-1-108-42266-6 — Cognitive Discourse Analysis

Thora Tenbrink
Frontmatter
More Information

Cognitive Discourse Analysis

Analysing language data systematically and looking closely at how people

formulate their thoughts can reveal astonishing insights about the human
mind. Without presupposing specific subject knowledge, this book gently
introduces its readers to theoretical insights as well as practical principles
for systematic linguistic analysis from a cognitive perspective. Drawing on
Thora Tenbrink’s twenty years’ experience in both linguistics and cognitive
science, this book offers theoretical guidance and practical advice for doing
Cognitive Discourse Analysis. It covers areas of analysis as diverse as atten-
tion, perspective, granularity, certainty, inference, transformation, communi-
cation, and cognitive strategies, using inspiring examples from many different
projects. Simple techniques and tools are used to allow readers new to the
subject easy ways to apply the methods, without the need for complex
technologies, whilst the cross-disciplinary approach can be applied to a
diverse range of research purposes and contexts in which language and
thought play a role.

thora tenbrink is a Reader in Cognitive Linguistics at Bangor University.

Her previous books include Space, Time, and the Use of Language (2007) and
three co-edited books on spatial language and cognition.

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Cognitive Discourse Analysis

An Introduction

Thora Tenbrink
Bangor University

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

University Printing House, Cambridge CB2 8BS, United Kingdom

One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre,
New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.

[Link]
Information on this title: [Link]/9781108422666
DOI: 10.1017/9781108525176
© Thora Tenbrink 2020
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2020
Printed in the United Kingdom by TJ International Ltd, Padstow Cornwall
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Tenbrink, Thora, 1969– author.
Title: Cognitive discourse analysis : an introduction / Thora Tenbrink, Bangor
University.
Description: Cambridge, UK ; New York : Cambridge University Press, 2020. |
Includes bibliographical references and index.
Identifiers: LCCN 2019030553 (print) | LCCN 2019030554 (ebook) | ISBN
9781108422666 (hardback) | ISBN 9781108525176 (epub)
Subjects: LCSH: Discourse analysis – Psychological aspects. | Cognition.
Classification: LCC P302.8 .T4856 2020 (print) | LCC P302.8 (ebook) |
DDC 401/.41–dc23
LC record available at [Link]
LC ebook record available at [Link]
ISBN 978-1-108-42266-6 Hardback
ISBN 978-1-108-43684-7 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Contents

List of Figures page vii

List of Tables viii
Preface ix
Acknowledgements x

Introduction 1
1 Background and Scope 7
1.1 How Language Relates to Thought 8
1.2 Areas of Application for CODA 12
1.3 Reliability, Generalisability, and Limits of CODA 26

2 Language as a Representation of Thought 32

2.1 Research on the Use of Language to Represent Thought: A Brief Overview 33
2.2 How Psycholinguists Analyse Language Use in Relation to Cognition 37
2.3 How Cognitive Psychologists Analyse Language Use in Relation to
Cognition 41

3 Resources 57
3.1 Cognitive Linguistics 57
3.2 Discourse Analysis 72
3.3 Functional Grammar 83

4 Identifying Cognitive Orientation 92

4.1 Attention 92
4.2 Perspective 103

5 Identifying Cognitive Depth 117

5.1 Granularity 117
5.2 Certainty 128

6 Identifying Cognitive Constructiveness 142

6.1 Inference 142
6.2 Transformation 155

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

vi Contents

7 Using Language to Convey Thoughts 168

7.1 Communication 169
7.2 Cognitive Strategies 181

8 CODA Procedures 194

8.1 Preliminary Considerations 195
8.2 Experimental Design 197
8.3 Verbal Data Elicitation 201
8.4 Data Preparation 208
8.5 Practical Steps of Data Analysis 210
8.6 Qualitative Insights 220
8.7 Quantitative Patterns 222

9 Beyond CODA 226

9.1 Triangulation: Combinations with Other Methods 226
9.2 Practical Purposes within and outside Academia 240
9.3 Conclusion 248

Register of Linguistic Features 251

References 255
Index 269

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Figures

1.1 Furniture item arrangement used in Andonova et al. (2010)

and Tenbrink et al. (2011b). page 17
3.1 Profiled meanings in Langacker (2002:6). 65
4.1 The different profiles of semantically similar sentences
(following Langacker, 1986). 96
4.2 Screenshot of a tunnel during the virtual journey in the tunnel
task used in Schönebeck et al. (2001), followed by a set of
arrows shown when emerging from the tunnel. 108
5.1 Example for paths chosen by novices and experts (Tenbrink
et al., 2011a). 139
6.1 Screenshot from the film in Tenbrink et al. (2016), showing
a tilted view with artificial horizon. 154
7.1 An example for a dialogue situation modelled by COR (Shi
et al., 2010). 177
8.1 Screenshot of Excel-based analysis done in Tenbrink and Taylor
(2015). 215
8.2 Spreadsheet for the analysis in Tenbrink and Taylor (2015). 223
9.1 A process model of writing (Flower & Hayes, 1981). 235

vii

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Tables

3.1 Examples of Chinese characters that represent their meaning

to a degree. page 59

viii

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Preface

This book is the result of twenty years of engagement with a seemingly simple
idea: we use language to express our thoughts. At first glance, you might find
this rather trivial: isn’t this what language is for, what we do all the time? But
then you might wonder how much research is really out there that examines
discourse in order to identify people’s thoughts. There is some, but there could
be more – after all, what could be more fascinating than the human mind? This
train of thought is what motivated me to look more closely into methodology
and develop Cognitive Discourse Analysis (CODA).
Initially inspired by a few fascinating studies that used language to access
central human concepts, I started experimenting with this approach myself.
Over the years, I benefited tremendously from collaborating with experts
across various disciplines, who shared the desire to learn something specific
about the human mind – and trusted me enough to believe my suggestion that
language provides a unique way of accessing it. Each of my collaborators has
contributed in their own way to the refinement of CODA as it is presented here.
I am grateful to all of them, as well as to the generations of students who have
done linguistic analysis in my classes and tutorials, creating ever-new perspec-
tives on how language represents thought. Based on this personal experience
and the host of relevant literature now available, I strongly believe that CODA
is useful for gaining insights about the human mind across disciplines. I hope
this book will help to spread the method and develop it further.

ix

© in this web service Cambridge University Press [Link]

Cambridge University Press
978-1-108-42266-6 — Cognitive Discourse Analysis
Thora Tenbrink
Frontmatter
More Information

Acknowledgements

I am greatly indebted to my colleagues, friends, and collaborators for their

continuous support as this book took shape. Several people have read and
commented on every single chapter; others provided critical feedback on
individual parts. Many, many thanks go to Holly Taylor, whose inspiration,
collaboration, and friendship have been absolutely invaluable, right from the
very early stages of using the method. My heartfelt thanks further extend to
Jago (Anwen Williams) for her reliable constructive criticism and support
throughout, and likewise to Josie (Sarah J. Ryan), Harry (Yu Hui), Saskia
Kuliga, Katja Egorova, Sharon Miller, Colin Burns, Asterios Chardalias,
Charlotte Danino, Davood Gozli, Rick Moore, Katharina Stoltmann, and
Phillip Wadley. I am sure the book wouldn’t be the same without their feedback
and inspiring comments. I also thank Bangor University for granting me
a teaching-free semester to finish this book, and I am grateful for support by
the publisher, Cambridge University Press, particularly Andrew Winnard and
Stephanie Taylor.

© in this web service Cambridge University Press [Link]

 Preface

This book is the result of twenty years of engagement with a seemingly simple
idea: we use language to express our thoughts. At first glance, you might find
this rather trivial: isn’t this what language is for, what we do all the time? But
then you might wonder how much research is really out there that examines
discourse in order to identify people’s thoughts. There is some, but there could
be more – after all, what could be more fascinating than the human mind? This
train of thought is what motivated me to look more closely into methodology
and develop Cognitive Discourse Analysis (CODA).
Initially inspired by a few fascinating studies that used language to access
central human concepts, I started experimenting with this approach myself.
Over the years, I benefited tremendously from collaborating with experts
across various disciplines, who shared the desire to learn something specific
about the human mind – and trusted me enough to believe my suggestion that
language provides a unique way of accessing it. Each of my collaborators has
contributed in their own way to the refinement of CODA as it is presented here.
I am grateful to all of them, as well as to the generations of students who have
done linguistic analysis in my classes and tutorials, creating ever-new perspec-
tives on how language represents thought. Based on this personal experience
and the host of relevant literature now available, I strongly believe that CODA
is useful for gaining insights about the human mind across disciplines. I hope
this book will help to spread the method and develop it further.

ix

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 14:05:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 Introduction

There is a paradox here – a delicious one – which I cannot resolve: if there is

indeed a fundamental difference between experience and description,
between direct and mediated knowledge of the world, how is it that language
can be so powerful? Language, that most human invention, can enable what,
in principle, should not be possible. It can allow all of us, even the congeni-
tally blind, to see with another person’s eyes.
Oliver Sacks, The Mind’s Eye (2010:240)

Imagine you are trying to find your friend who is waiting for you in the cafeteria
of a large hospital. The building is complex, and you don’t see any signs that
would help you find the way. You ask the receptionist, who advises, ‘Well, try
the big staircase over there, it must be there somewhere!’ What do you make of
this? Almost certainly, you’d be well aware (and probably very surprised) that
the receptionist doesn’t really know where the cafeteria is. Nevertheless, you
might try the staircase as advised.
But – how do you know this? Where does your awareness of the recep-
tionist’s lack of knowledge come from? Can you read her mind? She never
mentioned any uncertainty explicitly, and you probably wouldn’t be able to
state her ignorance as a given fact. Nevertheless, your knowledge of language,
and of the principles of conventional language use, enable you to make this
inference. The receptionist’s uncertainty was communicated implicitly, not
explicitly – not in what was said, in terms of the information conveyed, but
in how it was said, in terms of linguistic features.
With a closer look, you will find evidence in the receptionist’s short answer
to support your intuition. To start with, you may note what was not said: she
didn’t, as might be expected, give you precise directions to the cafeteria.
Instead, there are various indications of vagueness and uncertainty. The initial
discourse marker well provides a first hint, followed by the term try, which
points to an activity that should not be necessary while following an expertly
given route description. Referring to the main staircase as ‘the big staircase
over there’ seems odd, and the modal verb must indicates a degree of uncer-
tainty. Finally, the term somewhere isn’t sufficiently specific to serve as expert
guidance. Quite likely, the receptionist’s utterance also sounded hesitant with

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2 Introduction

regard to prosody and fluency, which you would recognise as additional subtle
indicators.
Any adult native speaker of a language will intuitively recognise features
such as these. In everyday life, we absorb this kind of implicit information
subtly, swiftly, and effortlessly. This book demonstrates how to use these
insights and intuitions systematically for research on human cognition, based
on a wealth of findings on the relationship between language and thought.
Cognitive Discourse Analysis – CODA – is a method designed to reveal how
speakers’ thoughts and concepts are reflected in language, both explicitly and
implicitly. CODA looks at linguistic data, at spoken or written language (i.e.
discourse) produced in situations relevant to thinking. Closely related to tradi-
tional discourse analysis methods, CODA means doing discourse analysis with
a focus on concepts and thought processes.
In a very basic sense, speaking is based on thinking. Without concepts and
thoughts there would be no language. Sounds and noises may be produced
without cognition, but meaning cannot. Also, language seems to be the most
direct medium available to humans for expressing our thoughts; almost inevi-
tably, we will answer the everyday question ‘What are you thinking?’ through
language rather than any other medium.
Linguistic research, across various subfields, is accumulating an ever-
increasing wealth of insights about the ways in which language relates to
cognition. In cognitive linguistics, various theories address the relationship
between language and the mind (e.g. Evans, 2009a, 2009b; Langacker, 2000;
Talmy, 2000, 2007; Tomasello, 2003). In particular, lexicogrammatical struc-
tures in language appear to be systematically related to cognitive structures and
processes. This structural fact carries over to principles of language in use: the
way we think is related to the way we talk. This is true both generally in terms
of what we can do with language (or with a particular language in contrast to
another), and specifically with respect to what we actually do.
The relationship between language and thought may not be simple, but it is
nevertheless systematic. For instance, how a speaker describes a scene reflects
what they are currently attending to in the scene, in contrast to other aspects of
the scene that are not mentioned or remain in the background. Thus, the car
next to the tree conveys a different focus of attention to the tree next to the car,
even though the spatial relationship is identical. In both cases, other objects in
the scene, and aspects such as the car’s colour or the type of tree, are not
represented at all.
Linguistic choices reflect crucial aspects of the speakers’ concepts and strate-
gies at the time of speaking, along with their relevance in a communicative
context. Some of these strategies may be intentional, but most of our linguistic
choices while speaking happen too quickly to be completely conscious. Because
of these phenomena, discourse analysis provides a good pathway to access

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 Introduction 3

various layers of cognition, assuming the necessary expertise regarding relevant

features of language. This book offers the conceptual tools needed for this. It will
enable researchers to reveal aspects of human thought through a close analysis of
language.
CODA primarily uses language as data that more or less directly expresses
thoughts and concepts: for example, when people describe a complex scene or
event, think aloud while solving a problem or making a difficult decision, or
explain a complicated process to a friend. All of these situations involve
cognitive challenges that will find their way into language in systematic
ways, depending on the specific features of the situation.
Some aspects concerning human thoughts are directly reflected in content,
and researchers have used this fact for many decades in cognitive science
(Ericsson & Simon, 1993). For instance, when asked to think aloud, people
will say what they are thinking about and consider their task-related observa-
tions explicitly. Verbal protocols can therefore reveal the cognitive steps that
humans take to solve a complex problem. However, language reflects more
than this, beyond the content that speakers express directly. For instance, in an
utterance like ‘Have you still not finished reading?’, the marker still subtly
conveys the speaker’s underlying expectations. Likewise, in ‘I have heard
about it, too,’ the term too reflects the speaker’s knowledge about not being
the only one to which the utterance applies.
Other aspects that have been established through linguistic research may be
less obvious to a non-expert. For instance, the prepositions in and on convey
more than the geometric properties that they denote: they also suggest a sense
of control, as illustrated by ‘the painting is on the wall’, or ‘the flowers are in
the vase’. Although the flowers are not fully contained in the vase, the term in
can be used because the vase controls the flowers’ location; the flowers will
move with the vase when carried to another place in the room.
Subtle markers such as these are well known in linguistics, and they repre-
sent a vast resource of insights about human thought. CODA offers
a comprehensive way of applying these insights through systematic analysis
of human language use. The goal is to illuminate underlying concepts and
thought processes that may go beyond conscious reflection by individual
speakers, and that may not necessarily be directly observable in linguistic
content. This is because speakers are typically not aware of the underlying
network of options that allows for a range of linguistic choices besides their
own. This network only becomes evident by considering a larger data set
collected under controlled circumstances. To identify the conceptual signifi-
cance of these choices, CODA draws on rigorous data collection procedures
and involves systematic analysis measures. Equipped with the necessary tools,
as described in this book, the researcher can reveal systematic aspects of human
thought by a close look at language use in relation to the context of production.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4 Introduction

The book is structured as follows. The first three chapters provide general
background and discuss the approach from an interdisciplinary perspective.
• Chapter 1 sets the stage by outlining the background and scope for CODA:
what does the method build on, what is it intended for, why and where is it
useful, and what are the limits of this approach to cognition?
• Chapter 2 discusses the relationship between language and thought from
a variety of perspectives. This further elaborates on the background for
CODA by relating the methodology to approaches in psycholinguistics and
cognitive psychology.
• Chapter 3 turns to insights that support the analysis from three different
fields: cognitive linguistics, discourse analysis, and functional grammar.
Each of them offers a host of resources for analysts to draw on: practical
insights as well as theoretical findings to relate to in order to make sense of
linguistic patterns found in the data. This concludes the set of general back-
ground chapters.
Chapters 4–7 address specific cognitive aspects that have frequently been
addressed using CODA, and that illustrate the kinds of phenomena that
CODA can identify. The eight aspects discussed in these chapters can be
regarded as possible analysis perspectives that complement each other, and
that serve as examples of ways to analyse language in relation to cognition.
Linguistic data will normally not be analysed with respect to all of these aspects
at once, nor only from a single one. Instead, awareness of the wide range of
possible analysis perspectives will allow the analyst to identify particularly
interesting phenomena in their data that are relevant to their research question.
• Chapter 4 discusses two central aspects of cognitive orientation, namely
attention and perspective. Whenever we describe something, what we say
will reflect what we attend to; aspects that we barely think about will rarely be
reflected in our descriptions. Also, while we often describe things from our
own point of view, other perspectives are possible, such as our interaction
partner’s standpoint. While we don’t often say explicitly which perspective
we’re using, our language will reflect the underlying viewpoint in systematic
ways.
• Chapter 5 turns to cognitive depth, another aspect of our thinking that is
regularly reflected in the way we formulate our thoughts. On the one hand,
we can think things through, and formulate them in language, in much detail
or we can remain on the surface, sticking to the mere essentials. On the other
hand, we can be more or less certain about what we describe, reflecting our
expertise or the uncertainty of the subject matter itself. Both aspects, gran-
ularity and certainty, work together, as experts will often be able to look into
things in much more depth than novices – but on the other hand they don’t
always need to, as they quickly recognise patterns that they have come across
before.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 Introduction 5

• Chapter 6 addresses two ways in which our minds can be constructive: either
by making inferences on the basis of what’s already there – filling in gaps of
information – or by transforming what’s there into something new. This
twofold ability of the human mind to construct is basic to our existence,
enabling us to go beyond what we encounter around us. The way we talk
reflects both inference and transformation processes systematically.
• Chapter 7 turns to a set of aspects around the formulation of thought. On the
one hand, whenever we put our thoughts into words, we do this for a specific
purpose – typically for the benefit of an interaction partner. The first half of
Chapter 7 discusses communication and dialogic interaction. The second half
addresses the idea of cognitive strategies, which are the main target of think-
aloud protocol analysis in traditional problem-solving studies. These are not
intended to be primarily formulated for an addressee but instead represent the
structure of cognitive processes in temporal sequence.
This concludes the set of chapters addressing analysis perspectives. The
remaining two chapters offer more practical insights.
• Chapter 8 provides a detailed account of practical steps and advice for doing
CODA. Starting with preliminary considerations as to the purpose of indivi-
dual CODA studies, it discusses relevant procedures step by step, including
experimental design aspects, data collection, data preparation, practical steps
of data analysis, and considerations pertaining to qualitative and quantitative
analysis methods.
• Chapter 9 looks beyond CODA in two ways. On the one hand, CODA is only
one among many established ways to examine human thought; linguistic
analysis can be easily and fruitfully combined with other methods. On the
other hand, CODA is rarely used purely for its own sake; more often than not,
linguistic analysis is part of a wider goal. Therefore, the second half of
Chapter 9 concludes the book with a discussion of practical uses for the
methodology in academia and beyond, in applied fields such as architecture
and artificial intelligence.
With this overall scope, I am confident this book will be of interest to readers
from various academic fields, and indeed readers beyond academia in areas
where insights about human ways of thinking would be welcome. While this
book was primarily written for researchers in their early career stages onwards,
there is no prerequisite for doing CODA. Anybody who speaks a language well
enough to have clear intuitions about the network of options available in that
language can do a CODA study. The only prerequisite is a readiness to do
systematic analysis, as this will allow for significant insights based on speak-
ers’ choices in situations where thoughts are relevant.
So the book is open to all – but different readers come with different back-
grounds and might be interested in different aspects of the book. My own
background is a combination of linguistics and cognitive science, and it is quite

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6 Introduction

possible that this has some reflection in my writing. Perhaps those with
a similar background will be the ones most interested in reading the book
cover to cover. Many readers will already be familiar with CODA from their
own research, or from having read the shorter introductory article in Tenbrink
(2015); they might wish to access further information about specific aspects of
CODA in the book.
Other readers might be more selective. Academics with a firm background in
linguistics, especially those aspects covered in the book, may quickly skim
through the earlier chapters or jump directly to the various analysis perspec-
tives starting in Chapter 4. In contrast, cognitive scientists with little back-
ground in linguistics might find the first three chapters particularly
enlightening, together with the procedural Chapter 8 and the final Chapter 9
on extensions. They may also develop their own ideas for analysis perspectives
rather than relying solely on those outlined in Chapters 4–7.
Some researchers interested in language and cognition from a more theore-
tical perspective might get inspiration from the first three chapters alone.
Conversely, some may focus entirely on practical aspects, perhaps in order to
gain insights into doing systematic discourse analysis as part of a larger project;
they may find Chapters 8 and 9 to be entirely sufficient. Finally, it is possible to
use the book as a handbook rather than a textbook to read, and simply access it
sporadically to find specific information on some aspect of language in relation
to the mind. The Register of Linguistic Features and the Index provided at the
end of this book should be helpful for this kind of use.
Whichever way you use this book, I hope you find it inspiring and helpful for
your own purposes. It is my firm belief that language is an excellent medium
that represents our thoughts systematically in accessible and analysable ways –
and I cannot imagine anything more fascinating than the human mind, with all
its complexities and diversities.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:53:22, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1 Background and Scope

Language and thought: they belong together, they are intricately related. Can
you speak (or write) entirely without thinking? Can you think without lan-
guage? Perhaps you aren’t sure; perhaps your answer is a hesitant Yes to either
or both of these questions. Even so, it is fairly obvious that there are limits to the
Yes. Producing language without thought would be restricted to mindless
babble or repetition of words. Thinking without language is more interesting:
this might be closer to intuition, emotion, and other functions that the mind
might perform. You might clearly visualise images in your head that you find
difficult to describe in words. Your mind might explore colours, shapes,
sensations, and the like; it could relate them to each other, and relate them to
the real world. Perhaps this depends on who you are, or what your profession is.
For example, artists or engineers might develop special skills in ‘thinking
without language’.
But what does it actually mean, to think – in relation to language? The
analysis of language produced to express thoughts (in CODA) hits upon this
question in several respects. One central aspect will be explored in Section 1.3
in this chapter, namely the limits of verbalisability. This concept reflects the
fact that not all of our thoughts can be put into words, in speaking or writing, in
a straightforward way. Some things going on in our minds lack a precise word
to describe them (such as mixed emotions), and other things may simply
happen too fast or are too complex and evasive to be verbalisable, even in
retrospect. These phenomena naturally limit the scope of what language ana-
lysis can tell us about thought. There will always be more things going on in our
minds, beyond language – and these things may or may not be regarded as
‘thinking’.
Despite the limits of language in the representation of thought, language
undoubtedly remains one of the most important ways – or perhaps the most
important way – for us to express our thoughts. At a very young age, we learn
to think through language. Parents put their understanding of the world into
words for their children, and they often respond by repeating these words. And
then, they reuse the words in situations where they think they can be applied in
the same way.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 8 Background and Scope

Often, children think aloud as they discover the world, playing with the
words they have heard others say in similar contexts. They do this creatively,
and it is fascinating to watch them try out where words may fit. Frequently
children will end up overusing a word, as they tend to overgeneralise categories
(Tomasello, 2003). Then, not just Toby is a dog, but also Kitty (the cat) – and
perhaps even the hare seen on the field. Such effects reveal thought processes
that make logical sense, but do not correspond to the way the words are
normally used.
As children grow up, they learn to use language to communicate thoughts to
others. They learn that parents cannot read their minds, and that they must
express what they are thinking or feeling in order to get a desired reaction. This
remains the most important use of language throughout adult life: the commu-
nication of thoughts and ideas, feelings, and observations. Sure, there are other
ways of communicating: through actions, through non-linguistic gestures,
through facial expressions. But there are many things that can’t be expressed
in those ways – and that’s when we need language. Language becomes even
more crucial when we’re not in the same room while communicating, which is
now rather common and easy in the modern Internet-connected world. Indeed,
the limits of language become apparent in such a situation, and therefore
Internet users creatively develop additions to language in the form of emojis
and emoticons. These are increasingly perceived and integrated as a proper part
of linguistic communication in the modern world (Gülşen, 2016).
Generally, the relationship between language and thought has fascinated
researchers for a long time, and the topic has been looked at from many
different perspectives. This chapter will look at three aspects relevant to
CODA. First, it will provide a brief overview of the study of language in
relation to thought. Then, it will show ways in which CODA can be applied
across different areas in which thought is represented. Finally, it will address
the limitations of linguistic data analysis in relation to thought, outlining the
extent to which we can ‘trust’ that the language we analyse actually does
represent thoughts and thought processes.

1.1 How Language Relates to Thought

Language and thought: how do they relate to each other, and how does
language relate to other cognitive functions or ‘modules’ (Fodor, 1983)? The
debate in the academic literature is as old as the study of language itself; indeed,
linguistics originated as a branch of psychology, reflecting the close link to the
human mind or ‘psyche’.
Human thought is fundamentally linked to its environment. We understand
the world simply because we live in it with our bodies and perceive it with our
senses (Anderson, 2003; Lakoff & Johnson, 1999; Wilson, 2002). The fact that

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.1 How Language Relates to Thought 9

we’re embodied in the world affects the way we think in many ways – including
language. To learn language, to understand how a word refers to an object in the
world, a child needs to interact with the object or perceive it in relation to the
overall context, and understand how the word relates to it. We process linguistic
input on the basis of our physical experience (Fischer & Zwaan, 2008). The
entire language system can be best understood on the basis of the features and
needs we encounter in the real world (Feldman, 2010).
What happens in our minds when we comprehend language – when we listen
to what someone is saying, or read what others have written? Our minds are
good at forming mental images or ‘models’ out of language-based input
(Johnson-Laird, 1983). We hear a story about a character travelling through
a landscape; the landscape is sketched in the story – and before we know it, we
can ‘see’ it emerging in our mind’s eye, without ever having actually been
there. Or we hear somebody moan how their legs are aching from climbing
a steep hill – and we can sympathise and perhaps even feel our own muscles,
just from imagining the action.
It sounds like magic, but our minds can do it because we have experienced the
world physically, and we can easily imagine being in various parts of it, given
only a few sketched hints through language. As a result, we will likely ‘see’ and
‘feel’ a lot more than has actually been said, in our mind’s eye. We fill in the
gaps, use whatever information we can gain – including some specific previous
experiences of our own, or that we’ve heard about, and expectations that we
build up on this basis. A simple word like poor, in the context of a suburban
area, directly makes us think of houses that have seen better times, of ragged
clothes, perhaps beggars in the street. Our mind is full of associations of this
kind, and after a very short time we may not be able to tell whether something
has actually been described directly, or whether we have simply been making
assumptions based on what was said. The existence of cultural stereotypes and
cognitive biases towards certain ways of thinking reinforces these effects.
If we build up mental models and fill in gaps from what we hear and read,
how does this relate to other kinds of input? How do language-based repre-
sentations relate to other information sources? Consider an everyday example:
is it better to listen to a verbal route description, or to look at a map?
Surprisingly, it seems like both kinds of input are equally suitable for wayfind-
ing (Meilinger & Knauff, 2008). Likewise, Tversky and Lee (1999) found that
people put pretty much the same kind of information into verbal route descrip-
tions as they would do when sketching the route on a piece of paper. What we
need for wayfinding seems to be restricted to a few basic pieces of information,
called ‘skeletal’ by Denis (1997) – something like Example 1.1.
Example 1.1 Turn left at the church, carry on for a few minutes, till you see
the goal on the left.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 10 Background and Scope

Further information can be supplied – through language or through a sketch

map – such as the angle of the left turn, or the exact distance in metres. But this
type of additional information is not essential for finding the route. This would
explain why, in this kind of scenario, language can be equal to maps. Some
additional information would be more easily conveyed through language, such
as a reference to past developments or personal experience; other things can
best be shown on the map, such as the relationship to other places in the
vicinity. While such information can support the wayfinding process, it is
typically not part of an everyday route description given by a stranger in the
street, and we can find our way without it.
In general, it seems that human cognition is fairly flexible with respect to
different types of input. The modules in our minds can skilfully use information
from diverse sources and then come up with similar outcomes (more or less)
independent of the representation format. This is not to say that it doesn’t
matter if we haven’t got access to some kind of input. Blind people, for
instance, are hampered by the lack of vision simply because visual information
is incredibly rich and ubiquitous; it is notoriously hard to compensate for this
through other sources.
As long as the same information is equally presented through different
media, the mind can cope with it equally. This may mean improved perfor-
mance in other areas, such as better listening skills with blind people (Röder &
Rösler, 2003). In reality, it is unfortunately rather hard to provide the exact
same information through different channels. Vision manages to capture many
details at once, while auditory or tactile information must necessarily be
presented piecemeal, sequentially, in what can be a laborious and awkward
process.
In addition, there are various elements inherent to language that make
language different from mental concepts (Levinson, 1997). Languages differ
in how they express reality, and they sometimes leave unexpected gaps for
expression. Not all kinds of thoughts (or knowledge) can easily be put into
words. Try to describe a family member’s face, which you know really well and
will be able to recognise in milliseconds. Or try to capture a strongly emotional
experience adequately: you will soon find that the few terms in the lexicon that
are available to express emotions are ridiculously inadequate.
Clearly, language and thoughts aren’t the same. We can think and feel things
for which we have no words, and different input modes facilitate different types
of thoughts. In that sense, language is just one way of evoking, or expressing,
thoughts and thought processes. Arguably however, it is the most complex and
developed one – accessible to everyone on the planet in one form or other, and
used incessantly for the communication of thoughts, and more. While the
limitations of language can’t be denied, no other medium of communication
has proved equally flexible and useful for a multitude of purposes.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.1 How Language Relates to Thought 11

But how do language and thought interrelate? Does a person’s language

perhaps even shape the way the person thinks, as Whorf (1941) suggested?
Whorf’s idea was that language provides us with a set of patterns that determine
the way we think. He suggested that if a language represents the world in
a particular way, this will fundamentally affect how the speakers of this
language think about and understand the world. In a nutshell, Whorf suggested
that our minds and thought processes are constrained by our language, condi-
tioned by its structure.
This general idea has been debated at length in many different ways (e.g.
Nuyts & Pederson, 1997). And many of the insights that inform the interpreta-
tion of discourse in CODA have been inspired by Whorfian ideas in some way
or other (see Chapter 3). CODA, as such, is about how language represents
thought rather than how thought is shaped or constrained by language, but both
aspects belong together: the relationship between thought and language is
complex and bidirectional; neither can be fully understood entirely without
the other.
To take just one example, Lakoff and Johnson (1980) famously showed how
much our thinking is affected by metaphors in our language. In English (and
related languages), we consistently talk about arguments in terms of war, time
in terms of space, love in terms of a journey, and so on – and these concepts
shape the way we conceive of them, the way we organise our lives. Even
prepositions like on and over have certain associations that are carried over to
more abstract domains (Tyler & Evans, 2003), for instance when we say on the
job or over £100. All of these insights feed into the question of how we think in
relation to language, and how the study of linguistic structures and usage
preferences can guide our understanding of human cognition.
By now, much insight has been gained as to how languages differ in
representing thought, and in representing how humans understand the world.
After nearly eighty years of research in the wake of Whorf’s (1941) sugges-
tions, it is now undeniable that language and cognition are intricately connected
and interdependent, yet obviously distinct. The mutual influence is strong and
has been confirmed and specified in many areas. The way humans think
influences the way language has evolved, and the language used in a culture
affects thought in that culture. Effects of these processes can be seen through
increasing evidence that cognitive flexibility is enhanced through bilingualism
(Adesope et al., 2010). Learning and speaking more than one language does not
only increase metalinguistic awareness, but comes with a range of other
benefits for thinking, such as better attention and working memory.
Most research on language and thought deals with linguistic structure in
relation to conceptual structure, that is, with the general ways in which lan-
guage resembles thought, across speakers and typically abstracting from indi-
vidual differences. CODA takes these rich and varied insights as resources that

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 12 Background and Scope

enable a close, well-informed look at specific uses of language: the analysis of

language use in relation to current cognitive phenomena. If previous research
shows that a particular type of linguistic construction reflects a particular way
of thinking in general, then any specific instance of using this construction
(such as talking about love in terms of a journey) reflects that this way of
thinking is currently relevant, currently being activated by the speaker.
Altogether, we can conclude that language and thought are deeply intercon-
nected. Our language may not strictly determine the way we think, but it
certainly matters for our thoughts what kind of repertory our language makes
available in a certain context. The next section will look at the areas where such
considerations can lead to further insights about the nature of human cognition,
using language analysis in specific situations rather than focusing on general-
ised linguistic phenomena.

1.2 Areas of Application for CODA

If language is closely related to thought, it must be useful to look at language in
order to find out about thought – use language, to an extent, as a window to the
mind. This is the basic tenet in cognitive linguistics, which primarily looks at
language as a system – and it is also the starting point for CODA, which looks at
how we use language to express our thoughts. In the following we will start with
general considerations about using language, and then outline the two main areas
for CODA applications – mental representations and complex cognitive processes.

1.2.1 Using Language to Express Our Thoughts

Whenever we use language, we make a choice from the general repertory of the
language we are using, based on the ways in which our language represents our
thought patterns in general. Such a choice will always be meaningful, as
pointed out by linguists across various theoretical approaches (e.g. Brennan
& Clark, 1996; Fontaine, Bartlett, & O’Grady, 2013). In particular, our linguis-
tic choices relate systematically to the ways in which we think about the things
we talk about. Consider the difference in the following two positive answers to
the same question about ‘Johnny’:
Example 1.2 Johnny is a well-educated, friendly person with polite manners
and a lot of knowledge.
Example 1.3 Johnny? Oh, Johnny. He’s my best friend! What would I do
without him?
While the number of words in the two utterances is the same, and both
appraisals are clearly positive, there is a fundamental difference here that any

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 13

speaker of English will immediately notice. Intuitively, it is strikingly obvious –

though it isn’t said directly – that the relationship between the speaker and
Johnny must be a lot closer in Example 1.3 than in Example 1.2. Somehow the
description seems very distant in Example 1.2, despite the emphasis on positive
character traits, and despite the fact that we actually get more information about
Johnny in Example 1.2 than we do in Example 1.3. To gain a clearer idea about
why this intuition jumps at the reader, we need to go beyond word counts
(equal), feature polarity (everything is positive), or content (more information
details are given in the more distant version). What kinds of factors, exactly,
reflect the difference in interpersonal distance that the reader intuitively
perceives?
The answer lies in a close look at combinations of linguistic features that
include syntactic (or mood) structure, lexical choices, reference types, and
modality. Example 1.2 has a relatively complex declarative clause structure
that invokes the idea of formal, written language, whereas Example 1.3 has
a series of brief exclamations, with only one short declarative sentence fol-
lowed by an exclamation mark. Example 1.2 employs a coherent series of
formal lexical items that share the semantic field of positive personal judge-
ment (well-educated, friendly, polite, a lot of knowledge), whereas Example
1.3 does very little to evaluate the person himself, other than as a ‘friend’.
Example 1.2 refers to Johnny just once, whereas Example 1.3 mentions the
name twice and adds two further references through personal pronouns (he/
him), thus emphasising the link to the person linguistically. And, perhaps
unexpectedly, we can spot the additional use of two first-person pronouns
(my/I), despite the fact that the question concerned Johnny, not the speaker:
this supports the conceptual connection. Example 1.2 represents factual infor-
mation in a concise manner, whereas Example 1.3 uses the modal ‘would’ to
construe an alternative world – leaving it to the addressee to infer how miser-
able such a world would be.
Together, these linguistic features convey a crucial fact about the speaker’s
mind that does not get direct expression in language: that the speaker does not
only think positively about Johnny, but actually feels close to him. This is what
speakers of English would intuitively grasp, based on their knowledge of the
linguistic system in English: if the speaker of Example 1.3 felt less close to
Johnny, they would not make these linguistic choices; they would find other
ways of talking positively about him, as done in Example 1.2. How this implicit
communication of thoughts or emotions works can be derived from a close look
at the linguistic choices made. This is facilitated by the contrast between two
distinct examples, because the contrast highlights the system of choices avail-
able in this situation.
This short example illustrates the basic tenet in CODA: the idea that speakers
always make choices whenever they use language. They choose from the

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 14 Background and Scope

network of available options whatever seems most fitting for the situation at
hand. This will also depend on who they are. Johnny’s best friend would only
choose Example 1.2 for an answer in a fairly formal context, whereas someone
who has only spoken to Johnny a couple of times would typically not dream of
using Example 1.3 in any situation.
There are, of course, limits to this logic. We do not, in all situations, simply
represent our thoughts and feelings using the most accurate set of linguistic
choices and features. We sometimes make errors, and there are certain general
limits to our ability and readiness to ‘speak our minds’. The strongest mediating
factor, arguably, can be subsumed in terms of communicative intent. Whatever
we say has a lot to do with whatever we want to communicate to our interaction
partner in a given situation. For this reason, it is entirely possible to use
Example 1.3 in spite of feeling rather distant to Johnny. Example 1.3 could
be used sarcastically, or with the intention of creating a false impression.
Communication affects our linguistic choices at all levels, at all times.
Language use without communication is difficult to imagine; there are not
many occasions where we speak without intending to communicate.
Communicative intent therefore needs to be taken into account in any language
analysis – including CODA, despite the fact that CODA primarily looks at the
cognitive aspects of language use. Nevertheless, this is not a book about
communication; there are numerous very useful publications available on this
topic (e.g. Clark, 1996; Sperber & Wilson, 1986; Wachsmuth et al., 2013, and
many others).
CODA is typically applied for situations in which the communication aspect
is kept at a minor level or controlled, either because communication is not
relevant (see Chapters 7 and 8), or because the parameters of communication
are narrowly defined by the researcher. This is true for the two core application
areas of CODA: mental representation and complex cognitive processes, which
we will deal with shortly. However, CODA can also be used to address the
interplay of cognitive and communicative aspects in dialogue. This is a major
extended application area of CODA that adds another layer to the analyst’s
challenge. We will return to these issues in Chapter 7.
So what exactly is this method good for? Broadly speaking, anything that can
be meaningfully verbalised by speakers can be meaningfully analysed using
systematic linguistic methods. However, two main areas stand out as particu-
larly relevant for CODA: mental representations and complex cognitive pro-
cesses. Mental representation means how we think about a scene, how we
perceive or remember events, and the like. Complex cognitive processes pertain
to more complicated challenges such as problem solving or decision making.
When we express our thoughts while dealing with such a challenge, our
language will reflect what aspects we think about first, what kinds of things
we take seriously, and so on.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 15

Both of these areas have rich traditions in research, going in very different
directions (see Chapter 2); CODA adds another perspective to them. Studies in
both of these areas have often related to language in some way, but traditionally
they rarely include a systematic analysis of linguistic choices.
In sum, we can identify each instance of discourse as a meaningful choice
from a network of options, namely the repertory of the linguistic system in the
speaker’s mind. Choices reflect aspects of thoughts and feelings, along with
communicative intent and relevance in a specific context. This is particularly
relevant when we use language to express mental representations or complex
cognitive processes. We will now look more closely at each of these areas in
turn.

1.2.2 Mental Representations

Understanding mental representations means understanding how we make
sense of what we see or hear, and how we remember things. This is a very
broad field, which opens up a wide range of interesting questions to ask and
things to discover. We can confront speakers with any kind of scene, shown in
the real world or depicted on a piece of paper or on a screen; it can be a static
scene or an event that unfolds gradually, perhaps involving a series of pictures
or a continuous film. We can ask them to describe what they see, to explain it to
somebody else, to remember and later recall, or to suggest changes to what they
see. In each case, features of the perceived situation will systematically affect
the linguistic representation.
However, simply looking at the language used will not lead to any
deeper insights unless there is a theoretical reason for looking at certain
aspects of the language. That is, any study in the area of mental repre-
sentation must be based on a clear research question that can be addressed
by language analysis.
With CODA, a very basic underlying research goal across all application
areas is to find out more about how humans think. In the area of mental
representation, CODA opens up new perspectives by adding to and combining
insights from traditional approaches. For instance, there is little overlap
between the three traditions of psycholinguistics, cognitive linguistics, and
systemic functional linguistics (SFL). Studies that take a psycholinguistic
experimental approach rarely draw from insights in theoretical cognitive lin-
guistics that could illuminate the conceptual implications of certain empirical
results. And because the idea of systemic choice is not as prominent in
psycholinguistics as it is in SFL, psycholinguistic experimental designs often
severely restrict the choices that speakers can make in the first place. This
makes it impossible to explore the significance of meaningful choice patterns
across speakers, beyond those preselected by the researcher.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 16 Background and Scope

Let’s look at an example. If we consider the world around us, we almost

inevitably need to think about space. Our world consists of space filled by objects
and people; their relationships change over time, and they may have a huge
relevance to humans in their everyday life, depending on the contexts and situa-
tions in which they occur. Spatial relationships are everywhere, and they matter to
us. Within cognitive science, the understanding of spatial relationships is central.
Traditionally, research in psycholinguistics and psychology aims to reduce
the natural complexity of spatial relationships to highly abstract scenes and
situations. For instance, Franklin, Henkel, and Zangas (1995) positioned parti-
cipants in a circular room and asked them to point to objects in various
directions, as well as to describe their positions. Such a scenario provides
a controlled environment in which particular factors that influence the out-
comes can be determined. This is highly informative for assessing some aspects
of human concepts of space. However, language production studies in this
tradition are not designed to highlight speakers’ choices out of the entire range
of available options. Instead, the range of what participants are allowed to say is
typically restricted; for instance, in Franklin et al.’s (1995) study, they were not
allowed to use compass or clock directions.
Some studies do allow for language use in a less restricted way, and this
brings them closer to the tenets of CODA. For instance, Ehrich and Koster
(1983) identified systematic patterns in speakers’ descriptions of living room
arrangements, and Carroll and von Stutterheim (1993) presented a cross-
linguistic comparison of speakers’ preference when describing spatial config-
urations. Taking these insights as a starting point, we (Andonova, Tenbrink, &
Coventry, 2010; Tenbrink, Coventry, & Andonova, 2011b) looked further into
spatial relationships in complex situations, as follows.
We were particularly interested in how people describe where objects are
within a complex configuration, and also how they are oriented (for instance
facing the wall). Our scenario involved fifteen pieces of doll’s house furniture
that were arranged on a tray. The tray was placed in front of participants, and they
were asked to describe the arrangement so that another person would be able to
rearrange the pieces in the same way. This task may sound easy, but actually it
involves quite a few challenges, such as structuring the description in a useful
way, and deciding on a perspective that would work for each object. Consider the
following example description, translated from our original German data:
Example 1.4 Okay, so on the left side of the room there is an armchair with
a yellow cushion in the bottom left corner. And in front of this
armchair there is a rack with red rungs. And behind this rack
there is a sofa with a yellow cushion. To the right of this sofa, not
very far away, there is a red chair. And below this red chair there
is, again, another red chair.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 17

Figure 1.1 Furniture item arrangement used in Andonova et al. (2010) and
Tenbrink et al. (2011b).
The arrangement in (a) includes some functionally meaningful groups of
objects, but the one in (b) doesn’t.

And so on. Were you able to follow this description, so far? Would you be
able to arrange the items, and would you be confident that they are placed
correctly? If not, you’re not alone. It is very difficult to create a spatial
description that is entirely clear and unambiguous, and which doesn’t leave
some questions open, such as: how far are the objects apart? Which perspective
is used with the words in front of – is it in front of the armchair’s seat, or do I use
my own view on the scene? What is meant by below – should I be stacking the
objects on top of each other? Just as in other contexts, language doesn’t always
solve these problems in a very direct way; utterances are underspecified and
ambiguous in many ways (Carston, 2002). No matter how you put it, there may
always be other (and possibly better) ways of saying it. Therefore, it is very
interesting to look at the choices that speakers make in spatial description.
What helps us in our everyday life, when we talk about spatial relationships,
is the fact that objects are sometimes functionally related to each other (Carlson
& van der Zee, 2005; Coventry & Garrod, 2004). For instance, if chairs are
arranged around a table, it is very easy to think of the whole group of objects as
a dining-room arrangement. In a description, people would therefore describe
the group as a whole, rather than as individual, unrelated objects. We thought
that this aspect could be very relevant for our study, and so we used two
different configurations of furniture items. In one case (Figure 1.1(a)), some
groups of objects were arranged in a meaningful way, such as a living room
table with several armchairs around it. In the other case (Figure 1.1(b)), no such
functional relationships were included. And indeed, we found that this influ-
enced the descriptions in interesting ways. If objects were arranged in mean-
ingful groups, they were described in a different order than if there were no
functional relationships. Also, the amount of detail was affected, and whether

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 18 Background and Scope

people mentioned in which direction objects were facing (Andonova et al.,

2010).
After we had determined how functional relationships affected descriptions
in Andonova et al. (2010), we then wondered what other kinds of thought
patterns we might find reflected in the linguistic data. In Tenbrink et al. (2011b)
we started by spelling out the whole range of what speakers did in this scenario,
and then showed how often they made specific linguistic and conceptual
choices. What kinds of information did they provide about location and orien-
tation of objects? What kinds of things did they relate specific objects to in
order to describe where they are? For instance, they could refer to another
object, as in the chair is to the left of the table, or they could refer to the room’s
shape or features, as in the chair is on the left side (or wall). What kinds of
perspective did they use – did they stick to their own point of view, or did they
sometimes use an object’s sides, such as the front side of a chair for in front of
the chair?
By looking closely at speakers’ typical choices in these respects, we could
identify frequent strategies of description – and also several cases where
individual speakers did something completely different. For instance, most
speakers very clearly went with their own perspective throughout most of their
descriptions, without changing to any other perspective out of the overall range
of options. This was independent of whether the array included functional
relationships or not.
However, sometimes this unwritten rule was broken locally, based on parti-
cular features within the configuration. If objects had very clear intrinsic front
and back sides, they often prompted speakers to disregard their own perspec-
tive – but only when these objects were facing towards the right or left, from the
speaker’s point of view. Thus, if a cupboard’s doors faced to the speaker’s right,
the speaker might mention a table in front of the cupboard – even though the
table would be on the right of the cupboard from the speaker’s point of view.
And another interesting exception was one single speaker who very rarely used
his own perspective at all, because he actually used compass directions such as
north and south; this was totally unexpected in this table-top scenario. These
were very interesting deviations from an otherwise consistent description
strategy across speakers, which we could not have predicted from earlier
research in the area.
People not only typically used one consistent perspective, but they also had
a range of other systematic preferences. For instance, they usually described
objects in relation to the room and its walls, as well as objects mentioned
directly before. Other, earlier-mentioned objects were rarely used for reference,
and speakers also did not often refer to their own position. Interestingly, this
pattern was reversed when describing where objects were facing. Although
speakers rarely said the chair is in front of me, they did very often say things

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 19

like the chair is facing towards me. Clearly, they were aware that basically all
objects were located more or less in front of them, so this wouldn’t have been
very informative. In contrast, the objects’ orientation could be described in this
way without leading to any obvious ambiguity.
Altogether, these two publications report a wide range of qualitative and
quantitative analyses based on our corpus of doll’s house furniture arrangement
descriptions. This helps us to understand better how people perceive and think
about the spatial relationships in a complex scene, to the extent that this is
reflected in language. Note that these descriptions of doll’s house furniture
arrangements did not involve a dialogue partner; however, we (Tenbrink et al.,
2008, 2017) also carried this line of research further to address how people
describe dolls’ house furniture arrangements to each other (see Chapter 7 for
details).
As we can see in this example, the verbal expression of mental representa-
tions can get quite complex and intricate, especially with increasing complexity
of the situation that is represented. And so far, we have only talked about static
scenes. Observing, memorising, and describing dynamic events can add var-
ious exciting layers to this kind of analysis. For instance, when people report
and describe events, they usually do so by providing causal links – not just
temporal order. If events are described as following each other, they very often
do so for a reason: if Peter sees Mary on the opposite side of the road and then
starts walking across the road, this is probably because he wants to talk to her.
However, this does not always get expressed in language directly. People
might just say things like When Peter saw Mary on the other side of the road, he
walked across to her. This would typically imply that Peter walked across to her
because (not just, coincidentally, when) he saw Mary. Such causal inferences
happen in the minds of humans, they are made very easily and very quickly,
whether consciously or subconsciously. The language people use reflects such
inferences to some extent. In fact, some linguistic items such as after very often
imply causal connections, even though they only represent the temporal aspects
directly (Tenbrink & Schilder, 2003).
Generally, analysing verbal descriptions of perceived or remembered events
can provide a wide range of insights about how such events are represented in
the speakers’ minds. For instance, Zacks, Tversky, and Iyer (2001) looked into
the ways in which people think about the structure of events shown in videos.
They found that the participants structured the events in their minds in
a hierarchical way, guided by the goals that were achieved within a scene.
Since overarching goals can normally be achieved by sequentially completing
some relevant subtasks, a hierarchical structure helps to understand purposes of
actions – or the ‘meaning’ of events and sub-events in context. This hierarch-
ical structure was reflected both by how people segmented the videos and by
how they described them verbally.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 20 Background and Scope

In sum, the analysis of mental representations is a very broad field, which

encompasses static scenes as well as events, and pertains to perception as well
as memory. Across these diverse scenarios, CODA can be used to enhance the
understanding of how we mentally represent the world we perceive and
remember.

1.2.3 Complex Cognitive Processes

So far, we have looked at how people conceptualise and describe things they
perceive or remember. Such mental representations can be complex, yet they do
not involve what psychologists would call complex cognitive processes. This is
because mental representations simply concern how we naturally conceive of
things, without needing to solve a problem or make a decision. In other words,
there is no directly associated challenge or action in the mind. We can observe
a complex scene or an entire film completely passively, without a care in the
world, without trying to get at some solution in any sense. The only task in such
situations is the task given to the participants in our experimental studies,
designed to obtain an external representation of the internal concepts that we
develop in our minds, based on what we perceive. This kind of task does not
normally pose any challenge to experimental participants.
In the case of complex cognitive processes, in contrast, people actively seek
some kind of solution, depending on the task. This involves developing rele-
vant thoughts over time, as they come to mind while accomplishing the task. In
cognitive psychology, such sequential thought processes have been of great
interest for many decades, and researchers have sought to identify relevant
patterns of thought related to a range of different tasks. Mainly, these tasks fall
into the two areas of problem solving (Newell & Simon, 1972) and decision
making (Ranyard, Crozier, & Svenson, 1997). These areas are interrelated, but
not identical: making a decision, such as which company to invest in, can come
with a range of sub-problems to solve. And solving a problem, such as how to
cross a river without a bridge, might involve a range of decisions along the way.
Understanding how people solve problems and make decisions is central to
understanding who we are: it explains how and why we do the things we do,
based on decisions we make, related to the everyday challenges and problems
in life.
When solving complex tasks of problem solving and decision making,
people are normally to a large extent aware of their thought processes, and so
they find it relatively easy to put them into words. (Try it out: how many
windows are there in a house that you know very well, for instance the one you
grew up in? Say your thoughts out loud until you come up with a number.) For
this reason, research in psychology has an extensive tradition of using language
data to access solution paths to complex cognitive tasks. This is done by asking

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 21

people to think aloud during such tasks, or to provide a retrospective report of

what they remember thinking while they solved the problem. The resulting
language data are called verbal protocols; the methodology is described in
depth in Ericsson and Simon (1993) and is based on early work by Newell and
Simon (1972).
It is important to note that the method involves a range of well-considered
procedures, going well beyond simply (informally) asking people how they
solved the task at hand. We will get back to relevant methodological concerns
in Chapter 8. The method spelled out by Ericsson and Simon (1993) has over
the past decades been widely used across many different domains, including
spatial cognition (Gugerty & Rodes, 2007), medical areas (Kuipers,
Moskowitz, & Kassirer, 1988), design studies (Purcell & Gero, 1998), reading
research (Afflerbach & Johnston, 1984), usability (Krahmer & Ummelen,
2004), and many more.
Conventionally, analysing verbal protocols in psychology means focus-
ing (primarily) on content. Such analysis targets precisely those aspects
that the speakers are themselves aware of (or ‘heed’, in the words of
Ericsson & Simon, 1993), such as specific thought processes, strategies,
problem-solving steps, or criteria for making decisions. Chapter 2 will take
a closer look at what these are – what kinds of cognitive processes can be
expected to be found in verbal protocols. Experts in a particular problem
domain who are familiar with theoretical insights and prior findings in the
field can often gain very specific insights just by examining the content of
verbal reports. There are specific methods for doing this systematically
(Krippendorff, 2012 – we will take a closer look in Chapter 8). Thorough
content analysis should always be the first step when analysing verbal
protocols, and this often leads to well-founded specific insights about the
cognitive processes involved (see, for instance, a detailed script analysis in
Kuipers & Kassirer, 1984).
However, content is not all that we can gain from verbal protocols. Just as
with verbal representations of visual scenes and events, how people formulate
their thoughts might be as revealing as what they say. In traditional verbal
protocol analysis, linguistic structure has rarely been taken into account. Since
most researchers in this tradition have been trained as psychologists, not
linguists, there is typically little awareness of the implications of particular
ways of saying things. Here, CODA provides a substantial step forward from
established verbal protocol analysis procedures, and enhances them in several
ways. CODA points to the cognitive significance of specific linguistic features
that are used to express certain contents. These insights can be used to oper-
ationalise, validate, and interpret content categories: if a particular type of
content is expressed regularly in a particular way, this can be used as evidence
that this is indeed a conceptual category of its own.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 22 Background and Scope

The identification of linguistic markers can help to make content category

annotation more objective, less dependent on the individual annotator’s intui-
tions (see Chapter 8 for procedural details). Also, insights about the language
used to express a particular content category can provide valuable insights
about the underlying concepts in the speaker’s minds, adding to the interpreta-
tion of the content category. Apart from supporting content analysis in this way,
CODA further extends the scope of analysis beyond content. Systematic
linguistic choices can reveal cognitive aspects that are not necessarily con-
sciously available to the speakers, and therefore do not get to be verbalised
explicitly. This opens up avenues for addressing research issues that have not
yet been investigated using verbal data so far.
Consider a simple example from Ericsson and Simon (1993:316ff).
With respect to a task of sorting blocks on the basis of given informa-
tion, a participant produced the following in the process of thinking
aloud:
Example 1.5 (. . .) a white block is further down than the black one
and the black one is between the red and the green
white, red, black, green
I’ll say then
but the green one is bottommost but one
then I’ll say
white, green, black, red
instead
then the white one is bottommost (. . .)

Note that the sketchiness of this extract is fairly typical for think-aloud
protocols; participants do not necessarily produce whole sentences (or
readable, coherent text). Ericsson and Simon (1993) present a content-
based analysis of this example, and show how the logic in the partici-
pant’s thinking plays out. In particular, they distinguish the repetition of
premises given in the task from novel inferences made by the speaker.
They note, on the basis of this content, that most inferences start with
then.
From a linguistic point of view, in contrast, it makes sense to start directly
from the use of conjunctions such as then. Such markers are known to serve
certain established purposes in discourse: for instance, and is additive, but is
contrastive, then is consecutive, and so on, among other possibilities (Halliday
& Matthiessen, 2014). If a speaker uses these conjunctions in the context of
thinking aloud while solving a problem-solving task, this reflects how the
speaker relates the verbalised insights to each other. Someone saying then
expresses a consecutive logic between two thoughts.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 23

Conjunctions can therefore be expected to yield relevant insights, and they can
be used as evidence, based on the body of knowledge available in linguistics
about such terms and structures. In this example, looking closely at where exactly
the speaker says then could yield valuable insights concerning how the speaker
conceives of decisions, insights, or actions during problem solving. Such a focus
adopted in linguistic analysis provides cues to places where important cognitive
steps can be expected to occur. This analysis perspective is based not just on
intuition or on the common-sense interpretation of content, but relies on
a straightforward operationalisation, using linguistic markers as cues.
To take another example, consider the following statement, taken from a route
instruction in downtown Freiburg (Hölscher, Tenbrink, & Wiener, 2011). This
was translated from a German description written by a participant; here we will
treat it as though it was produced in English, to facilitate matters in our example,
since it is a plausible part of an English route direction:

Example 1.6 At the concert hall take the Sedanstraße in the direction of the
theatre.
Looking at content, the problem of how to get to the goal is solved by
describing the path, referring to landmarks (concert hall) and streets. Route
directions have been subject to wide-ranging analyses in spatial cognition
research, as they reveal crucial aspects about how we think about space (e.g.
Denis, 1997; Klein, 1979). Such analyses have primarily focused on the types
of elements that are included in a route description, showing what kinds of
information we need in order to find a route.
However, there’s more to be gained from a route description. Take a closer
look at Example 1.6. The statement starts from a reference to the concert
hall – but where is the concert hall? This information is not part of the
statement itself; the speaker must be assuming that the listener knows
where it is – the location is presupposed, as indicated by the definite article
‘the’. There may be several reasons for such a presupposition; for instance,
the location of the concert hall may be part of the earlier discourse, such as the
previous sentence in the route description. More generally, a context can be
assumed which has mentally disposed the speaker to start this sentence from
this particular location, and to take it as given. As in this example, you might
find it revealing to simply focus on articles when doing discourse analysis; the
use of definite articles reveals what speakers take for granted in a particularly
clear-cut and intuitive way.
The next reference in Example 1.6, the Sedanstraße, is the proper name of
a street, and it is anchored within the statement (at the concert hall). Here the
preposition at expresses how the speaker thinks of the relevant spatial

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 24 Background and Scope

relationship. As with many specific linguistic choices, the literature can help
us determine their conceptual associations. Bateman et al. (2010) describe
the preposition at as denoting functionally based proximity. Thus, the route
describer refers to a very close connection between the concert hall and the
Sedanstraße, with a functional association in the sense that the concert hall’s
location relates to a specific street address.
Next, the location of the theatre is again presupposed, which is striking
as it occurs in the later part of the statement, where speakers normally
package new information (Halliday & Matthiessen, 2014). This indicates
that the theatre’s location is firmly anchored in the speaker’s mind, but it is
not made explicit in the linguistic representation. Nevertheless, the refer-
ence to the theatre is used to describe the direction of the next movement:
the addressee is supposed to orient towards the theatre, but is given no
further information about its location. The route giver probably assumes
that the wayfinder, when reaching this particular location, will be able to
identify the correct direction. This is often possible based on visual clues
in the environment: the theatre may be visible once they get there, or there
may be signs pointing to the theatre. It is also possible that there are no
alternatives, such as if the Sedanstraße happened to extend in only one
direction at this location.
If all this is not the case, the wayfinder may really have a problem – but the
route giver is obviously not aware of this. The route is described just in the way
that it comes to mind, and information is added as far as it seems to be relevant
in this particular route-giving situation. There may be all sorts of reasons for
why some facts and factors are deemed relevant in a given context (Sperber &
Wilson, 1986). Clearly, any description will always come with certain limits –
ways in which it may not be entirely complete.
This analysis suggests that presuppositions may be fairly crucial in route
descriptions, as they reflect how the speaker thinks of distant buildings such as
the theatre. A systematic CODA analysis of this aspect would mean identifying
all relevant instances of a particular type of presupposition within the verbal
data of all speakers collected in an experimental context. This could, for
instance, be a definite article used along with references to the theatre. The
use of definite articles in this context may depend on various aspects related to
the scenario design, for instance, familiarity with the city at hand, the intended
addressee, the complexity of the route, the discourse status (i.e. previous
mention) of the theatre, and the like. Such structures can be detected by
systematic annotation of relevant features. And changing these aspects in the
design in another experimental condition will shed even more light on the
particular role of the presupposition.
Further insights can be gained by looking more closely at other factors, such
as the routes chosen for the description, features of the situational context, or

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.2 Areas of Application for CODA 25

features of the referent. For instance, if particular environmental aspects (for

instance well-known public buildings such as the theatre) are systematically
presupposed more often in route descriptions than other aspects (such as
marginal side roads), this would suggest that speakers think differently about
public buildings to how they think about marginal side roads.
In related research, a colleague and I were interested in the strategies that
people use to reduce the complexity of a spatial planning task (Tenbrink &
Seifert, 2011). Our participants were given a map of Crete or, in one version of
the study, a similar island with a different shape, called Cretopia. The map
showed spatial relationships plus some further information about activities that
could be done at particular places. Participants were then asked to plan
a holiday route for a friend, with the aim to include much variation in terms
of getting around and doing various activities. They were allowed to draw the
route on the map, and they wrote a list of targets to be visited in a particular
order. Next, they wrote a retrospective report of the problem-solving process.
The study was carried out in four different versions so as to gain insights about
the impact of traveling mode (car vs. bike) or the shape of the island (Crete vs.
Cretopia), and so on.
The written reports were first examined with respect to content: what
kinds of planning strategies did the participants mention explicitly? Next,
we looked at linguistic indicators for specific planning strategies, inspired
by previous findings in the relevant literature, and also (post hoc) by the
data at hand. Our results showed that participants used a number of
planning heuristics that were very similar to strategies found in the litera-
ture in other spatial planning contexts. For instance, they oriented towards
particular spatial regions and noted where places were close together. In
other cases, they mentally imagined the travel route or the environment as
a whole or in part, and they avoided line crossings and detours. We will
return to this aspect of the study (mentally imagining a route, based on
a map) in Chapter 7.
To sum up what was laid out in this section: CODA is suitable for
analysing what people say in relation to what they think, or how they
conceptualise certain aspects that they perceive, remember, or currently
deal with. The main areas of application are mental representations and
complex cognitive processes such as problem solving and decision making.
Ideally, language data are collected by sets of speakers in a controlled
experimental situation, which makes it easier to identify systematic pat-
terns in the linguistic choices made by speakers (see Chapter 8). However,
there is no direct limit to the kinds of language data that can be analysed
using the principles of CODA. Meaningful patterns of linguistic features
can also be identified in existing corpora such as mountaineering reports
(Egorova, Tenbrink, & Purves, 2015), in the think-aloud data of a single

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 26 Background and Scope

speaker (Tenbrink, 2008), in joint-action dialogues (Tenbrink et al., 2017),

and in any other context that enables systematic analysis in relation to
a conceptual domain or challenge.

1.3 Reliability, Generalisability, and Limits of CODA

To what extent can we trust language data? Can we really trust what people
say? As we all know, people don’t always say what they mean. But even if
you’re willing to cooperate, you might sometimes find it difficult to express
your thoughts. You might feel that you’re not expressing what you’re thinking
or feeling very clearly. Perhaps you don’t really want to, or perhaps you can’t –
perhaps you don’t have the right words to express what you think. Your
language might not have perfectly suitable words in its lexicon, or your
language competence may be limited. As a matter of fact, not even adult native
speakers are ‘perfect’ in their own language – nobody knows all the words,
everybody makes mistakes in writing, and sometimes we get the grammar
wrong, or we have different opinions about what is right (Dąbrowska, 2012).
Also, you might get to the limits of your attention span when you speak along
with a given task, and this might affect how well chosen your words are when
representing your thoughts. And on top of that, there may be aspects to
‘thought’ that you are not aware of – very rapid or subconscious processes in
the brain that you will never be able to report in language.
Because of all this, there are clearly limits to the kinds of insights that we can
gain from systematic language analysis. In this section, we will take a closer
look at the extent to which CODA yields reliable insights, what kinds of
insights can be generalised, and where the analyst must stop in order to avoid
over-interpretation.
As we saw in the previous section, language data have for a long time been
used to gain insights about cognitive processes, both in the area of mental
representations (primarily in psycholinguistics) and in the area of complex
cognitive processes (primarily in cognitive psychology). Trying to capture the
entire rationale for these wide research traditions would take a few more books,
so we’ll focus here on several central aspects that are particularly relevant to
CODA.

1.3.1 What Kinds of Aspects of Language Use Can We Trust?

One of the main motivations for doing CODA is to add weight and reliability to
the analysis of content. CODA means going beyond content analysis in a range
of ways. That is, the primary contribution of CODA does not concern what
people say, but rather, how they formulate their thought processes. If we collect
data in a controlled experimental setting, and several speakers independently

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.3 Reliability, Generalisability, and Limits of CODA 27

express a thought or thought process in a similar way, then there may be an

interesting reason for that. If certain linguistic patterns are associated with
a certain thought category, this means moving away from mere intuition when
looking at what kinds of thoughts people express.
Patterns of linguistic features that are shared across a range of speakers are
typically meaningful in relation to the given situation. This is especially clear if
statistically significant differences can be identified, but a qualitative or
descriptive pattern in itself can be meaningful, too (see Chapter 8 for more
details). The interpretation of such patterns can be substantially supported by
established findings on language, drawing on a wide range of relevant theore-
tical and empirical resources. This approach adds a substantial layer of relia-
bility to the intuitive interpretation of verbalised thoughts.

1.3.2 Why Can We Trust Them?

People reporting their thoughts may not always wish to be entirely truthful or
complete – there may be aspects in their thought processes that they do not wish
to share. In such cases they might focus on other things to report. Crucially
however, they will be unable to report anything that they have not even thought
of. To take an extreme example, imagine a situation in which you are con-
fronted with the problem of interpreting the following formula:
Pt λ2 gt gr σb f 4
Pr ¼
ð4πÞ3 r4

Example 1.7 The Bistatic Radar Equation defines the received power when
radio waves are transmitted from one site and reflected from an
object before reaching the receiver.
Example 1.8 This must be a mathematic formula of some kind, I think it
defines a value called P.
Would you be able to come up with anything resembling the statement in
Example 1.7? Most likely, unless you’re a genuine expert in the field, your
interpretation of the formula would be far more general, perhaps rather like
Example 1.8. Note that Example 1.7 is far more specific and far more confident
in style, containing no linguistic markers of uncertainty like the items must,
some kind, and I think in Example 1.8. While uncertainty can to some extent be
feigned, expertise cannot: the level of specific knowledge that is reflected in
Example 1.7, if accurate, must represent expert knowledge (unless the speaker
had access to a reliable information source). The accuracy of expertise can be
assessed based on the given task; the analyst will be able to determine to what
extent a statement like Example 1.7 is correct.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 28 Background and Scope

Moreover, an expert trying to mislead the audience with respect to the given
task would be more likely to provide different, misleading, content; people
rarely ‘lie’ about linguistic uncertainty markers such as must, some kind, and
I think. Nevertheless, it is advisable to err on the side of caution: if verbalisa-
tions remain general and vague, this does not necessarily mean that the speaker
is unable to provide more detailed and specific responses. There might be other
factors in the situation that keep them from doing that. However, it is still often
enlightening in an empirical scenario to look at levels of specificity across
speakers. If a factor in a particular condition (such as expertise or verbal ability)
systematically leads to more specific and accurate responses, this does reflect
a likely difference in the underlying conceptualisations. We will get back to
issues related to granularity and certainty in Chapter 5.
Note also that experimental participants typically do try to comply with the
researcher’s aims, as experience over many decades of research in empirical
settings shows (e.g. Ericsson & Simon, 1993). In fact, this assumption is the
basis for doing any empirical research in the first place – if we couldn’t trust
participants at all, there would be little a researcher could do. If a person reports
their thoughts during a problem-solving process, it is rather unlikely that they
talk about a strategy that they actually find completely irrelevant and would
never really use, while using a completely different one secretly in their minds,
without mentioning it in any way.
In fact, the well-established central ‘bottleneck’ in attentional processes may
make this virtually impossible (Pashler, 1984), pushing participants towards
honesty. While they may be unable to put all of their thought processes
adequately into words, the strategies that they do talk about must at least play
a role in their problem-solving process – and any major strategy or problem-
solving step that they are aware of will normally find their way into language in
some way. If a participant aimed to consciously mislead the researcher, there
would have to be a reason for that. In some scenarios participants may be more
likely to aim to hide their real thoughts than normally – for instance, if asked
about a sensitive issue that humans are typically ashamed of. Such issues need
to be taken into account in any experimental design.
In addition, CODA means looking at aspects that participants will not be
aware of, such as patterns in the features of linguistic choices. Consider the
travel planning scenario described in Section 1.2.3 (Tenbrink & Seifert, 2011).
A purposefully misleading response would have to include references to con-
ceptual aspects that weren’t actually relevant to the speaker. An utterance like ‘I
have decided to cross Crete on the coast west of Heraklion’ indicates
a cognitive focus on a spatial region (the coast west of Heraklion), among
other aspects. It is hard to imagine that this participant actually used a different
planning strategy in which spatial regions were completely irrelevant. This
would not make sense in this context. This is primarily due to the fact that the

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.3 Reliability, Generalisability, and Limits of CODA 29

content of this utterance (crossing Crete on the coast west of Heraklion) is not
in the main focus of analysis, but rather the underlying concepts that get to be
expressed in language – and these are hard to fake.
To take another example, imagine a researcher interested in the kind of
perspective adopted when describing a spatial scene (similar to Tenbrink
et al., 2011b, as shown in Section 1.2.2). If they directly asked participants
about the perspectives that they adopt, they might get unreliable responses –
people might not know which perspective they use, they might not understand
the concept, or they might not want to say. However, if the researcher instead
asks the participants to describe the scene, participants will automatically adopt
a perspective, probably without thinking about it. The perspective they choose
will be expressed in language, either explicitly (as in The ball is to the right of
the table from my point of view) or (more likely) implicitly (as in The ball is to
the right of the table). Through their knowledge of the scene, the analyst will
then be able to derive the underlying perspective reliably.

1.3.3 To What Extent Can We Generalise CODA-Based Insights?

Whenever patterns of language are found in relation to a specific context, it is
vital to specify the features of this specific context, and to remain aware of them
throughout the interpretation of verbal data. It is very easy to fall into the trap of
overgeneralising insights gained from experimental settings, as they are nor-
mally motivated from a more overarching background – and the setting is
chosen to represent this more general idea. However, linguistic choices are
made on the basis of the exact situation that is perceived by the speaker.
Keeping the experimental context constant – including the exact wording
used to explain the experimental task – is therefore essential wherever data
are collected from more than one person. This is crucial in any empirical
setting, but it might be even more crucial for CODA. Fine-grained systematic
language data analysis across speakers makes most sense with a strict proce-
dure, even though the content of speakers’ verbalisations may roughly remain
the same even if the instruction is worded slightly flexibly.
To take a famous example from the spatial domain, based on one seminal
study by Linde and Labov (1975), it was widely believed for decades that
speakers normally describe spatial scenes from the perspective of an ‘imagin-
ary wanderer’ who walks through the scene. The limits of generalisability were
unknown, and have remained largely unexplored to this day. However, our
study described above (Tenbrink et al., 2011b) strongly suggests that the
imaginary wanderer may at least depend on scale. In our table-top doll’s
house furniture scenario, none of the participants imagined walking through
the scene, although it would have been possible to imagine mentally scaling up
to a real furniture setting.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 30 Background and Scope

Many results gained through CODA are actually of a more qualitative

nature, raising fewer concerns about generalisability. Qualitative analysis
means showing that certain phenomena exist and how they work. For instance,
in our route planning study (Tenbrink & Seifert, 2011) we demonstrated how
participants used and combined a range of conceptual strategies that had so far
only been identified for other kinds of tasks. Based on these insights, it is then
a meaningful next step to ask under what (generalised) kinds of circumstances
speakers regularly use these strategies, and when they don’t; this is likely to
reach far beyond the specific context of planning a holiday trip on Crete for
a friend.

1.3.4 What Are the Limits of Language Analysis Using CODA?

Despite the wide range of things that we can learn through a close analysis
of linguistic patterns beyond content, it is important to be aware of the
kinds of cognitive processes that will not get any expression in language.
This limits the scope of research questions that can be meaningfully
addressed via language analysis. In general, in terms of cognitive neu-
roscience, only non-automatic cognitive processes will get into language in
some way. Automatic cognitive processes are consciously inaccessible to
such an extent that they won’t ever be verbalised. This includes such
things as memory retrieval – how, exactly, do we extract memories from
our minds? Even when we do this consciously – trying to remember the
details of an earlier experience – it is very hard to put into words what we
are doing in order to retrieve these memories.
Or recognition – what are the cognitive processes that enable us to know that
we have seen a person or object before? We may be able to identify features that
we remember, or situations in which we have seen them. But the cognitive
process of recognition itself will not be verbalisable.
Or automated procedures – what goes on in our minds when we ride
a bicycle? If you have ever tried to explain how to ride a bike, you will have
noticed how difficult it is to describe in words what our bodies achieve, or how
exactly they achieve it. We learn to do such things by observation and practice,
and then we just know.
Or sudden insights or realisations – what exactly led us to suddenly ‘see’ or
understand something that we didn’t a moment ago? Some of these cognitive
effects can be reported after the fact, as far as we remember them, that is, to the
extent that they leave a trace in short-term memory (Ericsson & Simon, 1993).
Also, think-aloud data may to some extent reflect cognitively crucial moments
by subtle features such as hesitation markers, pauses, changes in intonation,
and the like. Such linguistic features do not directly express the low-level
cognitive processes themselves, but they can certainly reflect that something

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 1.3 Reliability, Generalisability, and Limits of CODA 31

important has been going on. However, such features are subtle and can have
multiple reasons and purposes, both in terms of cognition and in terms of
communication.
Because it is important to have a clear understanding of what one might
expect to derive from verbal data through CODA, we will return to these issues
in Chapter 2. For now, let’s note that the overall field of cognitive science is
rather wide; much research in the field concerns lower levels of cognitive
processing that will not find any reflection in language, and will therefore
need to be addressed in different ways.
There are also some limits within the realm of higher-level cognitive pro-
cesses, of the kind that can in principle be reflected in language. Some parti-
cipants may not actually be aware of the thought processes and strategies that
they adopt, or they may first adopt them implicitly, and only later realise that
they are doing so – which is when they start reporting them verbally (Siegler &
Stern, 1998). However, this is where CODA can make a valuable contribution,
because patterns of language use often reveal more than the speakers realise
themselves, in particular if choices are related to specific (implicit) aspects of
the task in ways that speakers are unaware of.
For instance, the situation may constrain preferred perspective choices,
guide the cognitive focus of attention, and so on – as in the examples discussed
in this chapter (and there will be many more in the following chapters). None of
these pertain to directly verbalised conceptual strategies on the part of the
speakers, and therefore the speakers do not actually need to be consciously
aware of them, or be able to report them on request. On the other hand, these
factors do not directly reflect mental activities at a neuronal level either.
Implicit aspects such as these represent an intermediate cognitive level that is
reflected in the structure of linguistic representations, and that is accessible to
systematic analysis that takes into account the range of variability available to
speakers.

To sum up, in this chapter we have explored the background for CODA and its
motivation and scope in a range of ways. We have found that language is
intricately related to thought, and that this has been recognised and utilised in
a wide range of ways in the literature. We have taken a closer look at the kinds
of linguistic data that can be meaningfully addressed using CODA; these fall
into the main areas of mental representation and complex cognitive processes.
Finally, we discussed the extent to which linguistic data analysis of this kind
can be reliable in terms of informing us about what goes on in the human mind.

Downloaded from [Link] Columbia University - Law Library, on 24 Jan 2020 at 13:58:19, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2 Language as a Representation of Thought

Chapter 1 laid out the general scope for CODA against the background of
several approaches to the relationship between language and thought in the
literature. As we saw, language analysis is often used as a ‘window’ to the
mind to address a wide variety of research questions, many of which relate to
mental representations or complex cognitive processes. This chapter explores
the associated research and methodologies in more depth, in relation to
CODA.
There are two main ways in which language is seen as a representation
of thought. Either you can think of the system of language in a static
way, or you can think of how the system is used by speakers to express
their thoughts, and what happens in their minds as they do so. Looking at
the system itself, some approaches primarily explore how linguistic
structure represents cognitive structure; specific aspects of the linguistic
system are identified as expressing specific aspects of thought. This leads
to theories about language(s) and linguistic structure(s) in relation to the
mind. For purposes of CODA, two of these theories are particularly
useful, although they have rarely been combined, and they are also
based on very different assumptions: cognitive linguistics and Systemic
Functional Grammar. As these theories represent central resources for
CODA, they will be addressed separately in the next chapter, along with
a third: the broader field of discourse analysis, which includes a wide
range of methods for analysing language in use, for diverse purposes
besides accessing cognition. Doing discourse analysis to access cognition
means doing CODA, bringing these various resources together; this is
what this book is about.
In this chapter, we look at the wider range of approaches that explore
language in use in order to access cognition, that is, that regard discourse as
a representation of thought. We start with an overview of methodologies and
look at some related projects and findings, and then return to the two major
areas of research in which, as pointed out in Chapter 1, the use of language is
regularly used to access aspects of cognition: psycholinguistics and cognitive
psychology.

32

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.1 Research on the Use of Language to Represent Thought 33

2.1 Research on the Use of Language to Represent Thought: A Brief

Overview
This section will first take a brief look at the overall research field, and then
zoom in to one particular domain: spatial language and cognition.

2.1.1 Sketching the Field

Language can be used and investigated in a vast range of ways. This includes
spontaneous natural discourse, in all sorts of variations. Everything can be
varied: how many people are talking? Who is talking to whom? Men to men,
women to women, or a mix? What is their age, their social background, their
culture? What are they talking about? Are they actually talking, or are they
writing – or using digital media to chat or post messages? What is the situation,
the context in which they are communicating? What are they looking at? What
are they thinking about? What is relevant for the current speaker? And so on.
All of these factors will affect how people use language, how they speak or
write, and how they choose their words. This has led to wide-ranging findings
using methods of sociolinguistics, Critical Discourse Analysis, conversation
analysis, pragmatics, and more. Typically, the aim in such work is to find out
how people use language in specific situations or locations, often combined
with a particular analysis perspective: what are the linguistic features of
a particular text type? Do men talk differently from women? How do people
use language to manipulate others? How do they display their power and
superiority – or else their deference or lack of power? And so on.
Apart from looking at patterns in available resources of natural discourse,
there is the option of collecting linguistic data in controlled experimental
contexts, where the variability of language use is restricted in some ways for
a particular research purpose. This is done, for instance, in experimental
pragmatics (Noveck & Sperber, 2006) and in psycholinguistic approaches,
which we will look at in Section 2.2. Clearly, as soon as some restriction is
imposed on the speaker, their choices will be narrowed down; controlled
studies seek this effect, but it doesn’t necessarily correspond to natural lan-
guage use. Also, the presence of a researcher collecting data will inevitably
affect the language produced; this is known as the ‘observer’s paradox’ or
‘Hawthorne effect’. This cannot be entirely avoided, but it can be minimised
(see Chapter 8). Any relevant effects need to be taken into account in the
interpretation of language data collected in empirical studies.
Language plays a central role for many of the techniques used to gain
insights about the human mind (see, e.g., Anderson, 2009), typically without
building on linguistic expertise. For example, language is treated in laboratory
contexts as a medium for accessing memory, mental representation, or other

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 34 Language as a Representation of Thought

cognitive phenomena. This is achieved by restricting participants’ responses to

a set of predefined options, or by extracting decontextualised features of text
that can be handled statistically. Such approaches abstract from the fine detail
of linguistic structure and thus avoid dealing with the complexity of natural
discourse and social implications (Edwards & Potter, 1992). It is for this reason
that grounding in linguistic theory is not necessarily seen as essential for
analysis in this area. Nevertheless, the widespread use of such methods bears
witness to the many insights that can be gained by using language as an external
representation of phenomena that would otherwise remain inaccessible within
the human mind.
Going in the opposite direction, there is the option of largely ignoring
situational factors by drawing on huge and varied data sets. This is done in
methods of corpus linguistics (e.g. Biber, Conrad, & Reppen, 1998). Just like in
controlled experimental studies, quantitative and inferential statistical methods
are widely used in this type of research. Corpus linguistic methods are very
useful for finding out how people speak in general – what is normal for them
across a wide range of situations. This can be narrowed down to specific
dialects or sociological regions, depending on how a corpus was collected.
Knowledge about such findings can be very useful for the interpretation of
data collected in a specific setting. If we know how people speak in general, this
will help us see the significance of their linguistic choices when they express
their thoughts in a particular context. And corpus linguistic methodologies can
also be useful as a tool to analyse a data set for CODA, as the collected data may
actually be treated as a (small) corpus. Then, results from a large corpus can be
compared to a small-group corpus collected in relation to thought (i.e. a CODA
data set). We will return to this idea in Chapter 9.
As we can see, there is a broad range of diversity when it comes to
investigating the use of language. Some of these approaches have their main
focus on other aspects or a more or less purpose-free methodology, where the
relationship between language and cognition is not central. However, there
now is a strong and growing tradition of approaches directly targeting cognition
through language analysis, with major research centres addressing relevant
aspects in depth. For instance, the ways in which cognition and communication
relate to each other have been explored in projects and institutes such as
Cognition, Communication, and Learning1 (e.g. Holsanova, 2008),
Language, Communication and Cognition2 (e.g. Maes, Arts, & Noordman,
2004), and the Cognition and Communication Research Centre at Northumbria
University3 (e.g. Bugmann, Coventry, & Newstead, 2007).
1
[Link]
2
[Link]
3
[Link]/research/research-areas/psychology-psychiatry-and-neuroscience/cog
nition-communication-research-centre

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.1 Research on the Use of Language to Represent Thought 35

This list could be expanded; the interface of cognition and communication

continues to be exciting, and opens up ever more avenues to explore. Here, we
can see how language represents and expresses thoughts or thought processes –
and there is no limit to the kinds of insights to be discovered there. Also, the
range of methods used to explore this interface is not limited as such; all of the
methods for analysing language in use that we have briefly noted above can be
adopted for looking at how thoughts are expressed by language and commu-
nicated to another person.

2.1.2 Research on Language Used to Represent Spatial Cognition

As already hinted in Chapter 1, space is central for human thinking: who we
are, the way we think, and the way we talk – all of this depends on where we are.
If we don’t know where we are, we feel lost – a horrible experience for most of
us. If we need to get somewhere, we need a clear idea of how to do so, to avoid
losing our way. All of the many objects we use in our daily lives are located
somewhere, and if we don’t know where they are, that can be a major problem
as well. Also, we meet people in certain places in order to live and to work, to
interact and communicate. This need becomes somewhat reduced through the
invention of ‘virtual spaces’, social networks that enable us to communicate
without depending on spatial constraints. However, there definitely is some-
thing special about being co-present, being together in one place, within a room
or outdoors in the city or in a beautiful natural setting.
Space matters, and our thinking is to a large extent influenced by this fact.
We will see in Chapter 3 how cognitive linguistics has picked up on this:
a substantial portion of the linguistic repertory is affected by spatial concepts.
Much of our language starts from a spatial relationship and then gets trans-
ferred to something more abstract, as in the following two well-known
examples:
Example 2.1 She’s in high spirits today.
Example 2.2 Christmas is around the corner.
In both cases, spatial terms (high and around the corner) are used to express
something that is definitely not meant in any direct spatial sense. Feelings and
time are not spatial as such, but apparently our knowledge of space helps us to
understand and talk about them. There are many ways in which language is
based on physical experience. Experts debate the question as to whether
cognition is achievable at all without this kind of embodiment (Wilson,
2002) – an interesting point to consider in relation to artificial intelligence.
For instance, research has revealed a range of things that we do when we do
something as simple as finding our way around in everyday life: we choose

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 36 Language as a Representation of Thought

a particular perspective when thinking and talking about a route (Tversky,

1999). We orient ourselves towards landmarks such as churches, well-known
places, or highly visible landscape features (Denis, 1997; Egorova et al., 2015;
Newman et al., 2007). We organise our route concepts sequentially, around
decision points and spatial segments, guided by intersections and chunks of
space that belong together in our minds (Klippel, 2003). All of these, and more,
have been identified for language just as well as for non-linguistic cognition
(e.g. Allen, 2000; Couclelis, 1996). Thus, the language of route descriptions
can reveal very clearly how we think about routes – what is important for us
when we consider how to get somewhere.
Not all descriptions are entirely correct or consistent, especially when they
are given spontaneously by a stranger on the street when we ask them for the
way. Perhaps the person doesn’t know the way and is simply guessing; this
would be quite normal. However, perhaps they would be perfectly able to find
the way to the same location themselves. Even then, describing the route can be
really difficult. This shows just how difficult it is for humans to describe
something adequately that we know, or that we can do ourselves. We might
have the procedural knowledge of getting somewhere, but may not be able to
put it into words. Remembering all relevant aspects and concepts and repre-
senting them in suitable words can be challenging no matter what the context is,
and spatial relationships are notoriously difficult in this respect.
But even with completely correct and consistent route descriptions, there are
differences in quality. Denis (1997) showed that route descriptions are typically
preferred that crisply represent the most essential elements that we use to find
our way – landmarks, decision points, and the like. Other route descriptions
were either judged as containing insufficient information – some elements are
missing and people can get lost – or more information than is really useful.
Getting too much information across in a route description can indeed be
harmful, as the wayfinder will typically need to remember what they were
told; irrelevant information can therefore overload memory and distract from
the goal.
Intuitively, we share this kind of knowledge within a culture, as we will all
have heard several route descriptions, and found our way on this basis, before
we have to provide one ourselves. It is part of the common knowledge that we
share, as we grow up with similar experiences. Our environment is for a large
part shared, our thought processes about it are similar, and the descriptions we
produce on this basis share systematic features.
Nevertheless, there are a number of challenges that come up every single
time we describe a route. Besides the problems of remembering the details of
the route and putting them into words, we need to make certain assumptions
about the specific situation at hand: how much does the wayfinder already
know? Can we, for instance, refer to the popular shop that used to be on that

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.2 Psycholinguists Analyse Language Use in Relation to Cognition 37

corner but is no longer there? Is the listener capable of following a complex

description, or would they find this rather challenging? Do they have high
spatial abilities? How detailed should the description be? How much should the
listener be expected to remember?
Because we can only make best guesses according to what we know about
our interaction partner, route descriptions can differ quite a lot according to
familiarity. A stranger on the street might get very different advice to a friend.
And it really matters whether a listener is actually present or only imagined, for
instance when writing down a route description for a friend’s or stranger’s future
trip. The impact of such considerations and factors has not been addressed in
much detail so far, though a number of authors emphasise their importance
(Filipi & Wales, 2004; Haddington, 2010). Further evidence for the role of this
variability comes from more general research on discourse and dialogue (Clark,
1996; Pickering & Garrod, 2004; Schober & Brennan, 2003; and see Chapter 3).
As we can see, just looking at route descriptions – a seemingly small and
limited research area – opens up a wide range of concerns and questions related
to the ways in which language represents thought in this particular domain. And
altogether, across the board, the range of insights that have been gained from
analysing the use of language as a representation of cognition is continually
expanding; a single book chapter would never be sufficient to outline the scope
of the state of the art adequately in this regard. So this is not the goal here.
Importantly however, many of the individual studies carried out in these rich
traditions fit into the scope of CODA, or are very closely related to the idea of
CODA: discourse analysis (of some kind) is employed to access cognition. This
includes, for instance, countless studies that address how factors of a situation
affect how people use language to express their concepts in that situation. Thus,
CODA is not a new invention, in spite of the fact that a coherent methodology
had not been recognised and spelled out. To do so is the aim of this book: to
bring to light what we are doing when we analyse discourse to access cogni-
tion – and to explore some well-established and systematic ways to do so.
We will now take a closer look at the two main established traditions within
which language use is analysed to explore cognition: psycholinguistic methods
that address mental representation, and verbal protocol analysis of complex
cognitive processes in cognitive psychology. Both of these traditions have
established clearly defined methodologies that are directly relevant to, and in
part overlap with, CODA as a discourse analytic approach.

2.2 How Psycholinguists Analyse Language Use in Relation

to Cognition
The wider field of psycholinguistics is understood to include areas such as child
language acquisition, the study of how words are organised and accessed in the

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 38 Language as a Representation of Thought

brain, how we manage to understand written and spoken language, and how we
produce language (Aitchison, 1983; Altmann, 2001; Klabunde & von
Stutterheim, 1999). Of these, psycholinguistic studies of language production
are most closely related to CODA studies because CODA addresses how we
use language. Like CODA, this kind of research also builds on the assumption
that aspects of language production systematically reflect cognitive features.
However, the purposes and analysis goals are rather different. While CODA
looks at the significance of linguistic choices a speaker makes, psycholinguistic
studies aim primarily to find out how words and sentences are processed in the
brain. Because of this, they typically use a far more constrained approach to
data collection than what is typically used in CODA studies. But nonetheless
there are considerable links. Many findings from psycholinguistic studies are
directly relevant to CODA research, and CODA studies may be directly
informed by them, including methodological aspects.
A typical psycholinguistic experimental design involves a specific task
setting that will highlight how language is produced under certain circum-
stances, related to the mental representation of a concept. Participants are given
a clearly defined linguistic task, which is typically quick and easy and does not
allow for many choices. They might be asked to choose an appropriate expres-
sion or phrase from a set of available options, or to fill a gap in a nearly
complete sentence. Frequently, such studies do not only look at linguistic
choices, but additionally involve related methods such as the analysis of eye
movements, reaction times, acceptability judgements, and more. For instance,
Corley and Scheepers (2002) asked speakers to complete sentences such as The
bank manager gave the cheque . . . They were interested in the grammatical
choices that speakers made, and how they were influenced (primed) by what
they did immediately before. In addition, they looked at response times, that is,
how long it took the participants to start typing their answers.
Priming effects such as those addressed by Corley and Scheepers (2002) are
widely researched in psycholinguistics, and this has led to profound insights as
to how speakers access words and concepts in the brain, guided by what they
have just accessed before. Pickering and Garrod (2004) even suggested that
much of what happens in dialogue is actually quite automatic, triggered by the
kinds of things that our minds have just connected to. As we listen to each other
and produce language in response, we align to each other’s use of language, and
even align with the underlying concepts in the mind.
All of this is basically guided by the familiar effects of priming – one concept
primes another, leading on to further thoughts that again prime related
thoughts. This explains why we are so often able to complete each other’s
sentences, as shown very nicely by Clark and Wilkes-Gibbs (1986) and others –
corresponding to our everyday life experience: ‘Are you going . . . ’ – ‘ . . . out?
Yes, planning to!’ This kind of thing happens all the time – as if we could read

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.2 Psycholinguists Analyse Language Use in Relation to Cognition 39

each other’s minds. In Pickering and Garrod’s (2004) theory, dialogue is the
most fundamental way of using language, since language was developed for
communicating. Therefore, the production and comprehension of language in
our minds can to a large extent be explained by such processes of interactive
alignment, drawing on various priming mechanisms.
Priming is very relevant for CODA, even though CODA does not target the
mechanisms of language production and comprehension in the mind.
However, it is important to recognise that priming mechanisms are at work
whenever people produce language. This may affect the analysis and inter-
pretation of linguistic choices, as speakers may be primed by what they heard
or read or did before. Choosing from the linguistic network of options is not
necessarily an unbiased or completely open process – many factors may
influence this choice, and priming can be an extremely strong factor in this
regard. This is relevant in the context of task instructions in an experimental
design – including giving examples, answering clarification questions, and so
on (see Chapter 8) – and it is also centrally important in the context of
communication (see Chapter 7).
For instance, Clark and Wilkes-Gibbs (1986) looked at how speakers work
together in dialogue in order to agree on an object that they’re talking about.
They found that references to objects systematically change over time: they
become shorter and shorter. When their participants referred to a tangram
figure, they said the following in a sequence of different trials of the same
experiment (Clark & Wilkes-Gibbs, 1986:12):
Example 2.3
1. All right, the next one looks like a person who’s ice skating, except they’re
sticking two arms out in front.
2. Um, the next one’s the person ice skating that has two arms?
3. The fourth one is the person ice skating, with two arms.
4. The next one’s the ice skater.
5. The fourth one’s the ice skater.
6. The ice skater.
This study, like many others in the psycholinguistic tradition of analys-
ing dialogue, used a task design that is widely known as the referential
communication task. In this task paradigm, two people are asked to solve
a task together, and one person has more information than the other. In the
study by Clark and Wilkes-Gibbs (1986), the partners had to arrange cards
showing tangram figures in a particular order, and this task was repeated
over a sequence of six trials. This setup led to the above finding: that
references to the same object by the same set of speakers can have
a certain life course.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 40 Language as a Representation of Thought

Referential communication tasks resemble face-to-face conversations in

many respects (Clark, 1996), and researchers widely agree that they provide
a relatively naturalistic setting for dialogue – at least compared to monologic
psycholinguistic task settings in which linguistic choices are restricted to
a minimum. Nevertheless, studies in this paradigm vary rather widely with
respect to the actual freedom given to the participants. In many such studies
(e.g. Hanna, Tanenhaus, & Trueswell, 2003), one of the dialogue participants
is a ‘confederate’ – a person who was secretly instructed by the experiment
designers to act according to a script, enabling tight control of the dialogue. In
contrast, in the study by Clark and Wilkes-Gibbs (1986), both dialogue
participants were untrained; both were actually participating in the experi-
ment, and all of their language was analysed afterwards. The data collected in
studies of this kind is inevitably far more complicated and complex to
analyse, as free dialogue can develop in complex and unforeseen ways. It
certainly takes considerable effort and a good eye to identify meaningful
overarching patterns in free-production dialogue data.
Understandably, many researchers shy away from this kind of challenge.
One of the aims of this book is to make unwieldy linguistic data sets
a little more accessible, by demonstrating feasible analysis perspectives
that can be adopted meaningfully across a range of possible scenarios
(Chapters 4–7), and by offering some practical tips for efficient analysis
procedures (Chapter 8). Equipped with such tools, referential communica-
tion tasks can be very well suited as scenarios for CODA purposes. In
fact, some of the analyses done in the psycholinguistic tradition are
actually very close to CODA, in that they focus more on the linguistic
choices made naturally by speakers than on the processing of language in
the brain.
And some previous CODA studies are based on referential communication
tasks. In a dialogue-based doll’s house study that will be further explored in
Chapter 7, Tenbrink, Andonova, and Coventry (2008) and Tenbrink et al.
(2017) used a classic referential communication task design without confeder-
ates, in which one dialogue partner instructed the other to furnish a doll’s house
in order to match it to their own. The results gathered relate closely to those
gained from earlier psycholinguistic studies, even though the motivation under-
lying data collection was somewhat different.
Altogether, there is no sharp dividing line between CODA and psycholin-
guistic approaches. Where the methods overlap, the difference is reduced to
a subtle shift in focus: psycholinguistics is typically more concerned with
accessing linguistic items in the mind (how much time does it take? what
concepts are more readily available?), whereas CODA looks at the linguistic
choices themselves (what words and constructions do speakers choose?),
with respect to how they represent concepts in the mind. Sometimes, both

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 41

aspects are informed by an analysis of speakers’ linguistic preferences in

a particular situation. However, there may be differences in the theoretical
(motivating) background, which might also affect how results are interpreted
and discussed in light of the literature.

2.3 How Cognitive Psychologists Analyse Language Use in Relation

to Cognition
As with CODA, the primary aim of research in cognitive psychology is to
understand how our minds work. This includes what is often referred to in
terms of higher-level cognitive processes: namely, the kinds of processes that
are based on knowledge and used to solve problems or make decisions, and the
like. As mentioned in Chapter 1, these would often be processes that we
recognise as our active thoughts, and that we can put into words in some
way. Insights and methods in cognitive psychology are therefore highly rele-
vant for CODA.
Verbal data can provide a range of insights that would not otherwise be
accessible to the analyst. For instance, consider a situation where a participant
makes a mistake – a behavioural result that is clearly visible to the analyst.
However, it may be less easy to see what caused the mistake. Failures may
occur on different levels, and all of these lead to less than optimal results.
According to Reason (1990), errors can be ascribed to specific subprocesses of
human planning actions. The intended plan itself may be faulty, or the execu-
tion of the plan is other than intended. Verbal data typically provide some
insights concerning the participant’s underlying intentions and thus contribute
to understanding at which level the errors occurred.
The tradition of using verbal data to access cognitive processes, plans,
intentions, problem-solving strategies, and the like is extremely widespread.
The range of problems and cognitive challenges that can be addressed using
verbal data is unlimited – just as our minds are not in any simple way limited,
considering the vast range of things we can think about. In the following, we
will look at various aspects related to this kind of research, with a view to how
CODA can inform this tradition.

2.3.1 What Kinds of Complex Thought Processes Are Interesting

for Cognitive Psychology Research?
Problems are abundant in our everyday life; they engross our thoughts about
the big challenges just as well as the small ones, from a very young age. How
can I reach the bowl of biscuits on the top shelf? How should I tie my shoes?
How will I find my dream partner? How will I become happy for the rest of my
life? How can I solve the riddle in the newspaper’s entertainment section? How

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 42 Language as a Representation of Thought

am I to obtain my dream profession? Not all of these problems are directly

solvable, of course; accordingly, they don’t lend themselves to research in any
straightforward way. Research in cognitive psychology focuses on types of
problems that fulfil some essential criteria.
First of all, the problem needs to be manageable within a limited amount of
time, so that participants can be observed while solving it, and all relevant
thoughts can be captured without missing out on decisive aspects. Deciding on
a profession, for instance, is a long-term process that involves a host of
thoughts and experiences and reasoning processes; the researcher would need
to observe and interview the participant over a very long time to ensure that all
relevant aspects are included. Doing this with a sufficient number of partici-
pants in order to identify systematic quantitative patterns is then a major
challenge, and will typically not be feasible. In contrast, the problem of tying
shoes is solvable fairly quickly, although it is cognitively challenging for
a child.
Similarly, the problem needs to be bounded, that is, limited to clearly
discernible factors. The aim is to reduce all distracting contextual factors that
might come into play. For this reason, research tends to focus on a task that is
clearly defined within the experimental context, drawing on as little prior
background knowledge as possible. In this light, the problem of tying shoes
turns into a question of observing how children learn to tie shoes. In order to
capture this process in its entirety, the child’s very first exposure to a shoe-tying
process would need to be recorded and analysed, and this again makes this kind
of problem unwieldy and impractical. In contrast, if participants are given
a riddle, perhaps of the kind that appears in the newspaper’s entertainment
pages, the researcher will only need to make sure that they haven’t seen or
solved it before. If not, they can be reasonably sure that the problem is confined
to the experimental setting.
However, this does not mean that the participant will not draw on any prior
knowledge at all. On the contrary – it is one of the basic assumptions in
cognitive psychology that participants can solve problems precisely because
they have solved problems of some kind before. One of the goals in research on
problem solving, then, is to identify common principles. Consider the follow-
ing riddle (found at [Link]):
Example 2.4 All the electricity was out in town and none of the street lights or
traffic signals had power. A dark limousine was cruising down
the newly paved blacktop, with its headlights off. A young boy
dressed totally in black (with no reflectors) stepped out to cross
the street. The moon wasn’t out and the boy had no flashlight,
yet the driver stopped to let the boy cross the street. How did the
driver see the boy?

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 43

Confronted with this riddle, you might start thinking about what the solution
might be: did the driver have supernatural powers, did they perhaps sense the
boy in some other way – or did the description fail to mention another type of
artificial light source? Then, you might remember how riddles normally work:
perhaps there is some trick involved, some obvious aspect that you’re somehow
expected to forget? And this memory of previous riddles might lead you to the
solution: the driver saw the boy because it all happened in broad daylight.
In this example, a common underlying principle is the idea of a trick, or the
expectation of a typical inference of the kind that humans make during reading.
Inferences happen all the time, as language never conveys all the information
about the situation described to the last detail. Crucially, we assume that the
things that are mentioned in a text are there because they are relevant (Sperber
& Wilson, 1986 – see also Chapter 3). In Example 2.4, there are lots of
references to the lack of artificial light sources; a very natural assumption is
that these light sources would be required in order to see properly. Therefore,
we automatically assume that the story must be set during night-time. We might
not even remember having made this assumption – it simply happens as we
read.
The riddle draws on this very basic and common reading principle; it
encourages the assumption that the artificial light sources are relevant – but
in fact they’re not. It also fails to mention that it was actually daytime. As
a consequence, it is safe to assume that most readers will be led towards
believing that it was dark. The riddle solver’s cognitive task is to break this
principle, or to realise consciously that there might have been a trick there, an
unspoken obvious inference that they must resist in order to solve the riddle.
Such a realisation is most likely to be possible if the riddle solver has seen
similar principles at work before.
Once a riddle of this kind has been solved, the same person will be less likely
to be tricked, and will find it easier to solve similar riddles. The feeling of
satisfaction after having solved such a riddle comes from the subtle awareness
that normally inferences are made automatically, but the successful problem
solver was clever enough to resist this principle. If inferences weren’t normally
at work, there would be no challenge and no satisfaction. A reader of the above
riddle would not even see a problem to solve – we don’t normally (during
daytime) have problems seeing people cross the street, even without artificial
light sources.
Generally, in cognitive psychology researchers are not simply interested in
how participants solve the specific problem at hand. It is far more interesting to
try to find common underlying principles. So when observing how people solve
a specific problem, researchers will aim to relate it to something more general.
In this example, this concerns both the underlying reading principle (as
a feature of the riddle), and the processes through which this is detected.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 44 Language as a Representation of Thought

Some riddle solvers might refer to previously solved riddles, and some might
be consciously aware of inference processes. Others might actually think of the
sun as a possible light source directly, and not make the night-time inference in
the first place. This leads to different problem-solving procedures, which again
might be similar to problem-solving procedures in other domains.
So this is another requirement: beyond being manageable and bounded,
problems that are targeted in cognitive psychology should also be generalisa-
ble in some sense. The aim is to study cognitive processes on a general scale,
with specific problems as examples. By combining and comparing solution
paths to different kinds of specific problems, researchers can distil the cognitive
processes involved in problem solving in a more abstract way, that is, on a more
general scale. Accordingly, the specific problems used in an experimental
setting in cognitive psychology are not picked randomly. Instead, they are
often chosen in order to support or falsify a certain existing theory about
problem solving, and involve specific hypotheses about the underlying cogni-
tive processes and strategies. Newell and Simon (1972) provided a thorough
overview of basic problem-solving principles identified at that time; most
current research on the details of such principles ultimately relates back to
their seminal work.
Importantly, the thought process at hand also needs to be measurable in some
way. There are many ways of accessing cognitive processes; these include
reaction times, behavioural results, task responses, brain measures, and so on.
The crucial point is that the researcher needs to identify a measure that
adequately reflects the cognitive process at hand. Or in other words, the kind
of cognitive process that is being targeted needs to be accessible by the measure
that the researcher is using. Here, we are concerned with verbal data analysis
for purposes of accessing cognitive processes. Accordingly, we are only con-
cerned with those kinds of thought processes that are actually verbalisable in
some way or other.
As we have already seen, there are limitations to the kinds of information that
can be derived on the basis of language. Generally, only processes that require
a minimum of conscious thought are verbalisable; this excludes memory
retrieval, activation and recognition, automated procedures, sudden insights
or realisations, and the like. Some of these can be reported after the fact, if
people remember them (Ericsson & Simon, 1993). Higher-level cognitive
processes and strategies may be either unconscious or conscious, or we may
first be unaware of them before they reach our consciousness sufficiently that
we can report them (Siegler & Stern, 1998).
As a simple example, assume that John has decided to search for a particular
object, say, his keys. Then he could verbalise things like ‘I have decided to look
for the keys now,’ and ‘I am still searching for the keys,’ and so on. These
verbalisations aptly reflect the conscious decision and the action of searching.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 45

However, in terms of cognitive processes, there’s much more going on. John’s
decision to start the search process activated a range of perception processes,
including visual search patterns (eye movements) that John will not normally
be aware of. Likewise, when John finds his keys, he’ll be able to shout, ‘Found
them!’ – but he will not be able to say how exactly he managed to recognise the
keys. Both perceptual processes and recognition are low-level, largely auto-
matic cognitive processes. They are associated to the verbalisable cognitive
processes of deciding and acting, as well as to the realisation that the keys have
been found – but as such, they will not enter any verbal protocol.
That said, CODA as a tool is nonetheless suitable for – and indeed developed
for – detecting a range of things in language that the speaker might not actually
be aware of. The structure of language represents some aspects of the speaker’s
thought that will enter the verbalisation, independent of whether or not the
speaker intends them to come through. For instance, if John exclaims, ‘Found
them!’, this particular choice of words and intonation signals some element of
surprise and delight, even though he doesn’t actually say that he’s surprised and
delighted. Other aspects conveyed implicitly and subtly through linguistic
choices include the perspective used, the level of granularity, the kinds of
things that the speaker takes for granted or represents as new, and so on.
However, looking for such aspects will still depend on verbal data being
available. If John hadn’t said anything when finding his keys, only the beha-
vioural action of holding the keys in his hand would be available for the
researcher to analyse and to conclude that John had found them. In such
a situation, if the researcher wants to address subconscious cognitive processes
that do not get reflected in any kind of verbalisation, CODA will not provide
any insights – there simply won’t be any relevant verbal data to analyse.
In sum, complex thought processes can meaningfully be analysed within
cognitive science – and using CODA – to the extent that they are not random,
but instead represent generalisable processes that relate to previous research in
some way. Analysis of language data will be most successful if the task at hand
is manageable, bounded, specific, measurable, and verbalisable; the clearer the
task can be represented along these lines, the more meaningful conclusions can
be drawn from the resulting patterns in the linguistic data.

2.3.2 How Are Verbal Data Collected to Address Complex

Thought Processes in Cognitive Psychology?
In the history of cognitive psychology, one of the earliest methods of investi-
gating problem-solving processes was introspection. The analysts simply
reported the various problem-solving steps from their own memory. That is,
they put into words what they remembered doing themselves as they were
solving the problem at hand. Although these researchers did not actually

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 46 Language as a Representation of Thought

analyse language as a representation of thought, they certainly used language to

represent their thought processes. Systematic and detailed representations of
this kind were regarded as good analyses of the problem-solving process. This
methodology has not completely died out, although it is now primarily used for
illustration of some well-known and easily communicated phenomena, rather
than being regarded as analysis in itself.
Later, researchers started asking other people about their problem-solving
processes, and this led to increased recognition of the fact that different
individuals may well solve the same problem in very different ways. Again,
verbal representations of thought were taken as evidence for actual thought
processes. Both of these approaches – introspection and asking participants
about their problem-solving steps – came to be attacked; they were widely
criticised as being too subjective and biased to the analyst’s own views. Indeed,
there is little control in such approaches, as the analyst may very easily
influence what they report from their own thoughts, or how they ask the
questions to the participants, guiding them towards the kinds of answers that
they’d like to hear.
Nevertheless, the idea that complex thought processes can be expressed by
talking about them has not been dropped from cognitive science, and in fact
there is little choice: language simply is the most preferred medium of com-
munication; this is the way in which we are used to telling others what we think.
The challenge, then, is to make this ‘window’ to the human mind more
systematic, more reliable as evidence, and less susceptible to the analyst’s
preferences and biases.
One way to do this is to ask more than one person: to collect data from
a larger group of participants who individually verbalise their thoughts in
a given task, under controlled circumstances. This tradition was developed
in the second half of the twentieth century, including a range of very
clearly motivated design considerations for experimental studies. The
methodology was laid out in much detail by Ericsson and Simon (1993),
who demonstrated that controlled verbal data elicitation can take various
different forms, serving diverse purposes. The most prominent forms are
think-aloud protocols produced during task performance, retrospective
reports that are collected subsequently, and post-performance interviews,
which can involve a range of further questions asked of the participants
once they have found a solution.
All of these different verbalisations can highlight cognitive strategies that the
participants use, but they do so in different ways. This has to do with memory
constraints. When people remember solving a problem, they might access
information from different sources: directly from short-term memory, or from
long-term memory, depending on what kinds of information were stored in
what stages of the problem-solving process. Accessing long-term memory in

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 47

particular involves the danger of integrating information that stems from other
sources than the task just solved. For instance, people easily confuse the
memory of what they just did with what they usually do. Therefore, it is useful
to employ different methods for different purposes, and it is necessary to be
aware of the various implications of accessing memory.
Ericsson and Simon (1993) proposed that the most effective way of obtain-
ing verbal data that directly concerns the task at hand, rather than information
from long-term memory, is to ask participants to think aloud during the task. If
participants simply say aloud what they are thinking as they are solving the
task, they will neither have time nor any need to access information from long-
term memory, except for the type of information that they do actually need to
solve the task (if any). Therefore, think-aloud verbalisations are likely to reflect
what participants really attend to for the problem-solving or decision-making
process.
However, importantly, this is only the case if participants are conscientiously
encouraged to say out loud what they are thinking, nothing else. Researchers
who are keen to find reasons for thoughts, or to find out further details about
a particular aspect of the problem-solving process, might be tempted to ask
questions that go beyond simply thinking aloud – and then the participants will
provide verbalisations that might not actually represent their own thought
processes closely. Instead, they would be guided and influenced by the
researcher. Therefore, the instruction for a think-aloud protocol needs to be
as neutral as possible.
More recent articles (e.g. Clark, Li, & Shepherd, 2018; Leow et al.,
2014) demonstrate that the general procedure advocated by Ericsson and
Simon (1993) is still regarded as valid and useful. Van Someren, Barnard,
and Sandberg (1994) offered a complementary, more pragmatically
oriented account of think-aloud data elicitation practices. And there are
a number of related methods. Spiers and Maguire (2008) recorded partici-
pants’ behavioural decisions during a virtual-reality wayfinding task.
Following task completion, the participants were shown the recording and
asked what they were thinking at particular places. While one might
suspect that participants may have been reporting ideas that they did not
actually consider during task performance, results turned out to be fairly
systematic, consistent with the behavioural data, and extremely informative
with respect to the research questions at hand. Thus, this type of data
provides yet another perspective on the cognitive processes that may be
relevant for problem-solving tasks.
For research on the usability of software applications, Boren and Ramey
(2000) suggested extending Ericsson and Simon’s approach to a communication-
based one. In their approach, the experimenter is allowed to communicate in
a fairly natural way with the participant in order to elicit more information, and to

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 48 Language as a Representation of Thought

support the user in exploring the usability of the application being tested.
Krahmer and Ummelen (2004) compared this approach directly with Ericsson
and Simon’s. They found that dialogic interaction during performance appeared
to have an influence on task success but not necessarily on the contents of the
comments being produced (think-aloud vs. dialogue).
Generally, we can conclude that verbal data can be (and have been) collected
in a range of different ways in cognitive science. The more systematic methods
are more likely to lead to meaningful results than randomly remembered
thoughts or informal interviews. Chapter 8 will provide a more general outline
of suitable methods to elicit verbalisations of complex cognitive processes for
CODA purposes.

2.3.3 How Directly Are Thoughts Expressed in Verbal Protocols?

Some thought processes are straightforward enough that the procedural steps
are clear and can easily be verbalised. Others may be a bit more complicated,
and they might involve more subconscious processes or aspects that do not
quite make it to conscious awareness. In such a case, it can become a rather
challenging task to speak out loud what you’re thinking, as you’re not actually
very clear about what you’re thinking. In some cases, thinking aloud might not
be feasible at all.
As a result, think-aloud protocols can, under certain circumstances, be
somewhat cryptic. Consider the following extract from a study that involved
participants travelling through a kind of virtual tunnel, shown on the computer
screen (Tenbrink & Salwiczek, 2016). The numbers denote hours:minutes:
seconds–tens of seconds in the recording.
Example 2.5
uh 00:08:32–3
so – left. 00:08:33–9
alright now – left turn 00:08:42–3
um 00:08:44–3
pretty sharp left turn 00:08:48–6
uh 00:08:50–0
yeah. 00:08:53–3
This participant did not say much, although they clearly adhered to the
instruction – they did verbalise what they experienced during the tunnel travel.
The information given in this think-aloud protocol is very sparse. It is basically
a verbalisation of the tunnel’s turns as shown on the screen, supplemented by
a few discourse markers such as uh, so, alright, and the like which are typical in
spoken language. Analysing verbal material of this kind can be very

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 49

challenging, and the researcher definitely needs a clear analysis perspective to

derive meaningful insights.
For these reasons, in order to avoid any influence of verbal data collection on
the actual thought processes, and in order to avoid the unintelligibility of think-
aloud protocols in some settings, it is sometimes worth considering other
methods of data collection. Ericsson and Simon (1993) suggested that under
certain circumstances retrospective reports can be nearly as adequate in repre-
senting thoughts as think-aloud protocols. If participants are asked directly
after the task what they remember thinking about during the task, there is a high
chance that the actual thought processes are still available in short-term mem-
ory. Only if this is the case can participants be expected to actually report their
thoughts? Otherwise, they would inevitably resort to thinking about and report-
ing some selected thoughts from long-term memory – such as how they would
usually solve a task like this, or what they think is the best way of doing it, or
what the reasons are for doing things in a particular way.
In one case study (Tenbrink, 2008), I compared individual verbal protocols
of cognitive processes related to a problem-solving process: think-aloud data,
a written retrospective report, and a more common or familiar text type: writing
an instruction for a friend. While Ericsson and Simon (1993) were primarily
concerned with the veridicality of verbal protocol data – that is, how accurately
they could be assumed to reflect thoughts – I was interested in the different
kinds of information and textual features of the three types of data. One goal
was to identify to what extent moving away from the immediate, on-line
production of a verbal protocol distorted the data. Would information be
added that did not seem to be directly relevant to the actual problem-solving
process? Would participants be led into further thoughts that they did not really
employ while doing the task? Along with this, another goal was to identify to
what extent the verbalisations were useful for the researcher in terms of
intelligibility.
Just like in Tenbrink and Salwiczek (2016) (see our earlier Example 2.5) and
in many studies reported in Ericsson and Simon (1993) and elsewhere, the
think-aloud protocol analysed in Tenbrink (2008) was rather sparse and
required a fair amount of interpretation and inferencing on the part of the
analyst. In contrast, the written reports were coherent and more like normal
texts; accordingly, their content was clearer, more explicit, and easier to
analyse.
In terms of content, there were indeed fundamental differences – each of
the verbalisations provided a different perspective on the same thought
process. Although sparse, the think-aloud data provided insights about
incremental learning processes over a period of time. Also, they reflected
more direct, lower-level cognitive processes that would not make their way
into a retrospective report, which requires more conscious reflection. In

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 50 Language as a Representation of Thought

contrast, the retrospective report focused more on general processes, and

provided an adequate, concise, coherent summary of what was going on.
Finally, the instruction written for a friend included further information –
added thoughts that might be useful for a person confronted with the same
task in the future. Just as Ericsson and Simon (1993) suggested, such
additional thoughts were not necessarily part of the participant’s own cog-
nitive processes during the problem-solving task. They must therefore be
analysed at a different level, taking into account the more fictitious nature of
this kind of verbalisation.
Altogether, different kinds of verbalisations can lead to different kinds of
insights. The researcher collecting data in this way needs to be aware of the
variability of cognitive perspectives, and should keep in mind how important it
is to use a method of collecting data that adequately suits the research goals in
this light.

2.3.4 What about Metacognition?

One problem that has often been pointed out by critics of verbal data analysis
in cognitive psychology is that the participants might not report their thoughts
directly, but rather report what they think about their own thoughts. That’s
called metacognition, and the critics are right: it is not the same as reporting
one’s thoughts. In fact, it is quite similar to the following common mistake
made (for instance) by inexperienced researchers, or by non-experts who are
simply interested in something. Imagine you want to find out how frequently
people use swear words in different everyday contexts: in the family, at
school, with friends, during work. The easiest way, it seems, is to simply
ask them. You could use one of the many questionnaire or survey tools that are
now available on the web for free, and collect data without any effort
whatsoever. So you design your questionnaire, and just ask all the questions
you’re interested in: how often do you use swear words in the family? How
often at school? How often with friends? And so on. People can then choose
and click their answers: Never. Once a day. Once every 100 words. Every five
minutes.
However, there is a problem with this. How can they be sure? How accurate
is the information they provide? Quite certainly, people won’t record their
everyday language use for you and do an exact frequency count before they
participate in your survey. They will just make a guess. Perhaps it’s a very
informed guess, or perhaps it is a wild one – you have no way of knowing! It
would probably make no difference if you asked them explicitly: how often do
you think you use swear words in the family? at school? And so on. People’s
answers will be the same – they will reflect what they think they do, rather than
what they actually do.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 51

That is, unless you explicitly get them to verify their intuitions. In order to
find out how often they really use swear words, you could, for instance, ask
them to note down every single instance of a swear word, whenever they use
them, along with the context: in the family, at school, and so on. This should get
you a bit closer to the truth, in terms of actual usage. However, these results
would reflect how often people remembered to note down their use of a swear
word – and that’s still different from their actual usage.
But – does this make the use of questionnaires entirely worthless? That
would be going too far. In fact, questionnaires and surveys are frequently used
in research across many disciplines; they are generally regarded as a very
valuable (and valid) research tool (see, e.g., Rasinger, 2010, for linguistics).
However, the question is what exactly is taken from them. If you ask people
how often they use swear words, the answer will definitely show you some-
thing: namely, how often people think they use swear words. And this can be
a quite interesting kind of result, particularly in relation to other results – such
as actual usage, identified in other ways. Then, it could be a target of research to
find out how aware people are of their use of swear words. The degree of
correspondence between people’s guesses and their actual use of swear words
could depend on certain factors. Perhaps they have a strategy for using lan-
guage, or they have been told explicitly to avoid swear words at home. Whether
this affects the accuracy of guesses about actual habits could be quite interest-
ing to investigate.
So, back to our question about metacognition. If we ask people what they’re
thinking, won’t they just tell us what they think they are thinking? Experimental
participants will not want to mislead us – in fact, they’ll be doing their best to
find out what they’re thinking, so that they can report it to us. Simply asking
somebody to report their thoughts can make them aware of their own thought
processes, and this can change them: it can bring things to consciousness that
weren’t there before – and it can mean that they think they have been thinking in
a particular way, when in fact they haven’t.
For instance, you might be caught dreaming or intensely thinking about
something, and absent-mindedly doing something that was fully unintended:
let’s say you have been scribbling drawings or words onto a piece of paper in
front of you, which might happen to be an important document. Your friend’s
exclamation of ‘What do you think you’re doing?!’ makes you ‘wake up’, and
you realise, with a start, what you are doing. Perhaps your answer to the friend’s
question will be something like ‘Oh – oops – nothing, sorry!’ However, it’s also
entirely possible that you instead come up with an explanation of your thoughts
and actions – you would rationalise your actions post hoc, so to speak, and you
might even believe that you’re telling the truth. Saving face is a natural human
aim, which might lead to some confusion of thought where face is felt to be
threatened.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 52 Language as a Representation of Thought

Similarly, people who are asked to verbalise their thoughts in an experi-

mental task such as problem solving might start thinking about (and
verbalising) how they think about their task, related to what they think
is expected of them. This might change how they actually or naturally
think about it, or it may trigger new kinds of conscious thoughts
(Smagorinsky, 1998). To address this problem, Ericsson and Simon
(1998) specified the methodological circumstances under which cognitive
processes can fairly safely be assumed to be simply externalised without
alteration.
Most crucially, it clearly matters in what way procedural instructions are
given to the participants (see Chapter 8 for concrete procedural suggestions).
Participants should be clear about the expectation to simply say out loud what
they’re thinking anyway, no matter what it is, without any specific value
attached to any particular direction of thought. Ideally, they should be trained
to do this in a warm-up task. In contrast, Ericsson and Simon (1998) proposed
that explaining or describing thoughts (as in social speech) represent different
types of discourse that involve further considerations with respect to the
communicative situation.
As long as appropriate care is taken in the procedural design, there are good
indications that, under the right circumstances, verbalisations will indeed express
thought fairly directly. Problem-solving tasks concern high-level cognitive opera-
tions, such as decisions and plans for future actions. These are the kinds of
thoughts that typically involve conscious awareness anyway, independent of
being asked to think aloud. In Ericsson and Simon’s (1993) terms, expertly
collected think-aloud protocols (and, under adequate circumstances, retrospective
reports) elicit just those aspects that are attended to (or ‘heeded’) by participants
anyway.
If performance is not affected by the requirement to think aloud, then it is
likely that the thought processes are not affected either. However, it is not easy
to predict whether or not this will be the case in any specific study. Having to
formulate thoughts out loud might either hamper the subconscious problem-
solving processes, or it might help them, as the need to think aloud can favour
some kinds of processes that are beneficial to the task at hand. Since there is no
way of eliminating all potential issues coming up along these lines, a control
group without the requirement to think aloud may be the only way to ensure
that there is no direct effect of thinking aloud on the actual problem-solving
process and performance. Russo, Johnson, and Stephens (1989) came to this
conclusion nearly thirty years ago, and a host of subsequent research has not
changed this: there are probably as many studies showing some effects of
thinking aloud during the task as there are studies showing no effects whatso-
ever (Gralla, 2013). This depends on the nature of the task and many other
factors playing together.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 53

2.3.5 What Can Linguistic Analysis (CODA) Add to Traditional Verbal

Protocol Analysis in Cognitive Psychology?
The main contribution of CODA to verbal protocol analysis lies in the sys-
tematic analysis of linguistic structures, based on insights from cognitive
linguistics and other resources that will be explored in subsequent chapters.
In contrast, Ericsson and Simon (1993) suggested identifying aspects of inter-
est and extracting them from the original data set for further exploration.
Typically, this would be conceptual aspects that the speakers are themselves
aware of, such as particular thought processes or strategies. To achieve this, the
content of verbal data is in focus, ignoring to a large extent the linguistic
structures used to express this content. In fact, Ericsson and Simon (1993)
explicitly recommended that the analyst should transform the original input
into a categorisation scheme that is directly related to the theoretical motivation
for the study at hand – moving away from the actual words used. With this kind
of analysis, the linguistic structures used by participants get lost in the process
and then cannot be accounted for in subsequent analysis steps.
Another suggestion by Ericsson and Simon (1993) was that annotation and
analysis should be as context-free as possible, and therefore the analyst should
not be biased by the preceding or subsequent discourse. But language use is
heavily influenced by context – it is sometimes nearly impossible to interpret
the speaker’s meaning correctly if the preceding discourse is entirely ignored.
At the very least, a range of potential implications and concepts will get lost in
this way. Accordingly, this recommendation has subsequently been criticised
and questioned, for example, by Yang (2003).
Despite these suggestions, linguistic structure hasn’t been entirely ignored in
this research tradition. Ericsson and Simon (1993) themselves give some
examples for linguistic patterns that turned out to be meaningful in their data.
And other researchers have explored the potential contribution of linguistic
features in various ways. Bartl and Dörner (1998) and Roth (1985) both went
beyond established content-based protocol analysis by looking at the linguistic
structure of think-aloud protocols. They were particularly interested in the
difference in thought processes between successful and unsuccessful problem
solvers. Success could be reflected in the way hypotheses were formed and
expressed, or in differences in the discourse structure or other features detected
by systematic analysis.
Roth (1985) focused on possible differences in ‘thinking style’ that were
detectable on the linguistic surface, drawing upon a wide range of linguistic
features motivated by the earlier literature. For instance, according to his
analysis, unsuccessful problem solvers produced more words of a sort that
Roth associated with dogmatic writing and thinking, such as all, always, with-
out any doubt. Also, they used adversative conjunctions (but, besides, either/

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 54 Language as a Representation of Thought

or) and negations (none, no, never, nothing) more frequently than successful
problem solvers; this could be traced back to a reduced clarity of understanding
during the task. Such linguistic markers may be indicative of relevant indivi-
dual differences in how people deal with challenges, such as when important
information is missing at the beginning of a task.
Linguistic analysis can also enhance the understanding of content. Content-
based categories tend to depend heavily on the analyst’s intuitions, in spite of
genuine attempts to do this systematically, for example, following the very
detailed procedures suggested by Krippendorff (2012). Content analysis gen-
erally means that the analyst needs to do a lot of interpretation and abduction –
inferences based on what they know about the discourse context. Content
analysts use their ability to comprehend a text intuitively, based on their
knowledge of language together with context information, typically related to
an analytic framework (Krippendorff, 2012).
Instead of ignoring linguistic structures in this process, a closer look at the
linguistic features of the text can decisively support this analysis step, based on
insights about meaningful linguistic categories – for instance by providing
criteria for content categories. For example, the analyst might find that
a specific aspect of a problem-solving procedure is regularly expressed using
specific types of expressions or grammatical constructions. Such an association
between a particular cognitive aspect and a linguistic feature can be very
interesting, particularly if it can be quantified. It might be interesting to point
out that one participant said ‘Heureka!’ upon finding a solution to a problem –
but it would be more interesting to see if there are some kinds of exclamations
that are regularly and systematically associated with finding a solution to
a problem, for instance ‘ah!’ If the exclamation ah! is recognised as a marker
indicating (sudden) insight in one study, this finding can guide subsequent
studies in specifying its role (and that of further linguistic indicators) within the
problem-solving procedure. In Chapters 4–7, and in the Register of Linguistic
Features, we will take a closer look at various linguistic markers that have been
identified as being associated with particular cognitive phenomena.
Some indicators will be specific for a certain kind of task and will need to be
identified directly for the study at hand. However, others are fairly generic and
informative across scenarios. For instance, an ongoing search process may
typically be accompanied in think-aloud data by verbs that directly express
search or perception more generally, such as look (for), search, see, find,
perhaps along with locative questions or statements using where.
While such expressions represent the general cognitive process itself, other
indicators externalise thought in other (conventional) ways. The exclamation
ah! is an example of this: it expresses the problem solver’s insight indirectly,
not by talking about the insight, but by a spontaneous reaction that signals it.
Similarly, the discourse marker okay is often used to mark the beginning of

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 2.3 How Cognitive Psychologists Analyse Language 55

a new step in a problem-solving process. Verbalisations of this type would

typically reflect cognitive processes that the speakers themselves are (or can be
made) aware of. If asked why they exclaimed ah!, the speaker will be likely to
say something along the lines of ‘Oh, I just realised/understood that . . . ,’
explaining the nature of the insight.
Linguistic analysis often also highlights systematic patterns that the partici-
pants themselves take for granted, and that are not directly visible just by
reading the words carefully. Any sentence, uttered in a specific context, carries
a host of features that reflect certain cognitive structures. This includes the
underlying perspective of the speaker, the level of granularity chosen, the kinds
of things they take as given information, what they currently attend to, and the
like: all of these specific conceptual aspects are reflected in language system-
atically – without necessarily requiring the speaker’s conscious awareness. We
will take a closer look at these aspects in Chapters 4–6.
One previous systematic approach that used linguistic methods for the
investigation of verbal protocols was reported in a collection by Caron-
Pargue and Gillis (1996). They described the initiation of a project called
PROVERB, following related studies (Caron-Pargue & Caron, 1989). The
project leaders proposed that ‘a careful study of the linguistic form of subjects’
verbal reports during a cognitive task enables a characterization of the cogni-
tive processes involved . . . [L]inguistic markers and their organization in the
developing discourse play a crucial role in discovering the intricacies of the
underlying cognitive processes’ (Caron-Pargue & Gillis, 1996:6).
The various articles in the collection showcased diverse applications of this
approach. For instance, certain modal particles and other lexical choices were
shown to indicate central elements, hierarchical structures, and particular
operations in a problem-solving process. Also, certain changes of reference
types systematically reflected a cognitive change, rather than serving any
communicative purposes. These ideas and initial findings are well within the
scope of CODA. Unfortunately, PROVERB had to be terminated shortly after
the publication (for reasons unrelated to the project itself). Only specific
aspects of the original project were pursued further, such as the interpretation
of discourse markers with respect to their relation to cognitive functions and
decision processes (Bégoin-Augereau & Caron-Pargue, 2009; Caron-Pargue &
Caron, 2000).
There is one further important layer that a linguistic perspective adds to
traditional protocol analysis in cognitive psychology. As noted above, verbal
protocols of cognitive processes can take various different forms, such as think-
aloud protocols, retrospective reports, post-performance interviews, and
a range of further methods such as task dialogues or instructions to future
problem solvers. From a CODA perspective, these considerations are relevant
in terms of text types. Different forms of text have long been known in linguistic

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 56 Language as a Representation of Thought

research as being systematically different in decisive respects, since they serve

different kinds of purposes.
Relevant insights come from areas of discourse analysis and text linguistics
(an earlier research field aiming at identifying different types of text, now
mostly covered by discourse analytic approaches), and they will be taken up
in more detail in Chapter 3. For instance, an interview situation differs from
a think-aloud situation in that a direct addressee is involved; the speaker will
formulate their thoughts in a way they think is suitable for the interviewer. In
this respect, an interview is similar to a task dialogue in joint action, or to an
instruction to a future problem solver. In all these cases, speakers will think
about their partner’s knowledge and expertise, and adjust the way they talk
accordingly.
Additionally, in dialogue, speakers often take up some formulations that they
have used or heard before – we discussed this in Section 2.2 in terms of priming
and interactive alignment. All of these aspects are more related to communica-
tion than to cognition, but they influence the features of verbal protocols
considerably. Again, conforming to the conclusions of Ericsson and Simon
(1993), think-aloud protocols are the most neutral version, as the influence of
the addressee is kept to a minimum. Nevertheless, a present experimenter may
well be the ‘addressee’ for a think-aloud protocol in the participant’s mind –
even if they tell them not to explain anything to them but simply express their
spontaneous thoughts. This effect cannot be ruled out entirely, and
a linguistically aware analyst should keep an eye out for indications and effects
of communicative considerations in verbal protocols.

In summary, CODA adds a layer of linguistic insights to the traditional analysis

of verbal protocols of complex cognitive processes. This additional layer does
not have to be based on in-depth linguistic expertise; it simply means looking at
the medium of the message conveyed. Beyond categorising content as is
typically done in cognitive psychology, the analyst examines patterns in the
way people express their thoughts in language. This can lead to a range of
insights about cognition that go beyond content, as discussed extensively in
Chapters 4–7.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 12:38:38, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 3 Resources

When doing CODA, researchers can draw from a wide range of insights
across approaches and disciplines – we have already seen some of this in
previous chapters, and we will explore the idea further here. The range of
findings in the literature that can highlight the significance of systematic
patterns in language data is vast; your creativity for bringing in relevant
insights when using CODA is as valuable as anyone’s. CODA itself is
theory neutral; the goal when doing CODA is to investigate thoughts and
thought processes on the basis of language data. This goal will guide the
analyst when selecting resources to interpret language data meaningfully in
relation to a specific research question.
This chapter will highlight three main research areas that have previously
informed CODA projects, so as to exemplify a range of relevant resources, and
hopefully inspire you to explore more of the available rich resources in
linguistics and beyond. The three areas that we will outline here are cognitive
linguistics, discourse analysis, and functional grammar – each of them far too
wide and diverse to be represented adequately. Nevertheless, the relevance of
established insights and methods in these fields should become clear in this
chapter, and there will be plenty of pointers to the literature to follow up.

3.1 Cognitive Linguistics

Cognitive linguistics is the study of language in relation to the mind. Typically,
in this field, this concerns structures in the linguistic system: cognitive linguists
are interested in how linguistic systems reflect structures in the mind. They look
at syntactic as well as semantic structures in many different ways, develop
theoretical accounts of how structures in language relate to thought, discuss
how the physical world and the human body relate to thought and language,
compare different languages with respect to how they represent the world and
their speakers’ thoughts, and much else. The field is very varied and consists of
more than a single homogeneous approach, where the term cognitive linguistics
serves as an umbrella term for research that addresses the relationship between
language and thought.

57

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 58 Resources

Historically, cognitive linguistics developed out of a need to counteract

a prominent belief commonly held early in the second half of the twentieth
century, namely that the concepts expressed in language can be adequately
described by abstract formalisms and definitions. Formal approaches to lin-
guistic structure were mainly inspired by Chomsky’s generative grammar
theories (e.g. Chomsky, 1972), which reinforced and developed earlier
approaches (e.g. Montague, 1970). In cognitive linguistics, there is general
agreement that such formal approaches are not sufficient to capture linguistic
concepts, relations, or structures in a cognitively adequate way; there is more to
the relationship between language and thought, and this must be addressed in
different ways. Cognitive linguists aim at accounting for the structures in
natural language by focusing on meaning, and they consider how fundamental
aspects of cognition are reflected in language. The following sections will take
a closer look at a range of central insights.

3.1.1 Theoretical Basis: Conceptual Meaning Does Not Directly Reside in

Words
How do words relate to meanings? Can a word really express what we have in
our minds? What is there in a word such as car that evokes a mental image in
our head? Does the word itself somehow carry this image, or what else is
happening? Common sense tells us that it can’t really be the word itself – the
sound of the word car does not sound like a car, the written version does not
look like a car, and other languages have quite different words for the same
thing anyway. Is a car more like a car than an Auto (German) or a bil
(Norwegian) or a ‫‘( ﺳﻴﺎﺭﺓ‬sayyāra’, Arabic)? There is no reason to believe that
this is the case. The words are recognised as direct translations; as such, the
form of the word is irrelevant, both in terms of its sound and in terms of its
written shape.
There are some interesting exceptions. In languages with a character-based
writing system like Chinese, one may often find that a character represents part
of its meaning – see examples in Table 3.1. In Chinese, some numbers or
quantities are represented in a transparent analogical manner. The character for
person looks like a simplified drawing of a person, the character for big
resembles a person opening their arms to show something big, and the character
for door looks like a door. However, these correspondences are limited and
fairly abstract. Without prior knowledge, it is not easy to infer the meaning
from the characters even in these cases, and many characters in modern
Chinese do not have any obvious relation to their meaning at all.
Alphabet-based writing systems like the one used in English do not provide
any such clues at all. Systematic correspondences exist between sounds and
letters, but not between letters and meanings. Correspondences between sounds

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 59

Table 3.1 Examples of Chinese characters that

represent their meaning to a degree

1 一
2 二
3 三
tree 木
many trees 林
forest 森
fire 火
burn 炎
flame / spark 焱
person 人
big 大
door 门

and meanings do exist, but they are rare and exceptional – as when we pretend
that the cat says miaow and the dog says bow-wow or woof-woof. However, this
idea is limited by the fact that several versions of this kind of onomatopoetic
language exist even within English, and that other languages have their own
versions of it. So in general, we may conclude that words aren’t the same as
meanings. But nevertheless words mean, without a doubt. How do they do this?
Many theories and approaches are available that aim to solve this puzzle. Here,
we will look at Evans’ (2009a) LCCM theory as one example that represents
a well-developed framework which is based on widely shared assumptions in
cognitive linguistics. LCCM distinguishes linguistic content (LC) from concep-
tual models (CM). The idea is that concepts, as such, are too rich and complex to
be encoded directly in language. The linguistic level expresses a limited amount
of content, but the more crucial meaning aspect is represented by something
much more complex: the conceptual model that the lexical item provides access
to. This view is also sometimes referred to, quite aptly, as ‘access semantics’. So
instead of representing concepts with their full meaning, language provides
a more abstract and schematic access point to conceptual knowledge.
For instance, imagine you hear the word red. This word evokes a certain
colour in your mind; it provides access to visualisations of this colour. Your
mind draws on rich and varied experiences for this, which you can use flexibly.
Which shade or quality of the colour will you think of when hearing the word?
Without context, this is really hard to predict, even though there are some
prominent structures in our minds (as we will see in Section 3.1.2). But if you
add a noun to the adjective, you will immediately get a more specific, more
restricted idea of the colour. Think of red ink – that’s more concrete than simply
hearing the word red. The work ink narrows down the colour range to include

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 60 Resources

only those shades of red that you would expect to see in ink. In contrast, think of
red hair – now the visualisation in your mind will probably be quite different
from the mental image you had when considering red ink (Evans, 2009b). It
will be a different red.
Thus, adjectives of colour allow for a wide range of possible associations.
But this is not only true for adjectives. Nouns such as ink and hair also both
provide access to complex concepts of particular types of entities, which are
schematically represented. Ink can be black or blue or red, and so on, and also
vary in other respects; hair also comes in different styles and colours. On
hearing the nouns, you may have a vague idea of this variability, or you may
not be entirely aware of it. As soon as context comes into play, the variability
becomes less flexible. A combination with the adjective red allows you to
narrow down the range of meanings associated with ink and hair – just as ink
and hair narrow down the range of possible visualisations of the colour red.
As shown in this example, words can provide access to a wide range of ideas.
These are referred to in cognitive linguistics as encyclopaedic knowledge. It is
almost as though we had an encyclopaedia in our heads, which we open as soon
as we come across a word. Our minds instantly provide the knowledge that we
need as we’re processing the word. This is not to say that the whole encyclo-
paedic entry is activated, with all its subtleties. Since we almost always get the
context immediately along with the words, we can safely ignore other kinds of
knowledge that we might have access to in our minds – everything that isn’t
relevant in this particular context of usage.
When we produce language, the principle is the same: we use words as
access points to our minds, referring to the vast resource of encyclopaedic
knowledge that resides there. Words may not adequately express what we
mean, but they can certainly serve as pointers. We use words in context to
mean much more than what we can actually say, as the words themselves are
rather limited. And then we tend to take it for granted that our listeners will
have similar encyclopaedic knowledge, activated by those same access points.
More often than not, this is not exactly the case – and this explains why people
often misunderstand each other. Even if they use the same words in the same
context, they may not share precisely the same ideas.
With respect to CODA, the lesson is that words mean in a complicated way,
that they may mean different things to different people, and also that they are
highly relevant access points to the speakers’ minds when representing con-
cepts. When analysing linguistic data, we need to be aware of the flexibility of
these access points, and of the crucial importance of context when interpreting
them. To understand more systematically what can be conveyed, by the use of
words in context, we need a clearer idea of how meanings are structured in
words – and in our minds. To this we turn next.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 61

3.1.2 Semantic Structure Reflects Conceptual Structures

Some of the most well-known and valuable insights in cognitive linguistics
concern how conceptual meanings are organised in the mind. Humans have for
a long time been interested in the categories that are expressed by words. For
instance, what is a chair? Intuitively, when asked about the nature of the
category chair, most people would explain what kinds of features a chair
has – typically four legs, a surface, a backrest, and so on – and possibly also
what it is used for – a chair is something to sit on.
Digging deeper along these lines points the hobby linguist in the same
direction as earlier approaches, going back all the way to Aristotle (Taylor,
1989): namely the challenge of finding the precise set of necessary and
sufficient features that correspond to our intuitive understanding of the cate-
gory chair. In Aristotelian terms, a category is defined by whatever features
constitute its essence. The essence of a man would be something like male two-
legged animal; other features that are associated with specific men would be
accidental – such as being tall and slim (or not). This kind of view has
dominated many disciplines (including linguistics) for a long time. It is repre-
sented in dictionaries and encyclopaedias that provide definitions for technical
terminology, and it continues to serve a range of very useful and legitimate
purposes – such as providing scientific definitions for categories that need
a precise treatment.
However, while expert definitions have their rightful place, this way of
handling categories might not correspond to what people normally have in
mind, or how they conceptualise a category. Sometimes expert definitions clash
with folk concepts. A well-known example is the treatment of mammals such as
whales as fish, even though most people are aware that this is not ‘really’
(scientifically) the case, as the definition of fish excludes mammals. Most
clashes between expert and everyday categorisation are probably more subtle,
and less fundamental. But above all, the classical feature-based approach
simply does not do justice to the categorisation processes in people’s minds.
There is more to human categorisation than degree of expertise. Human
categories are highly structured and meaningful, not just limited by lack of
knowledge about specific facts.
Wittgenstein (1953) was one of the first, and the most widely heard, to
challenge the classical feature-based view. He pointed out that it is perfectly
possible for a category to have members that do not share a single feature. Such
categories might get multiple entries in an encyclopaedia, or a complex defini-
tion with many alternatives. For instance, consider the word game. What do all
games have in common? In Wittgenstein’s argumentation, there is
a counterexample for any feature that one might think of: games might be
played alone or with others, they might be competitive or not, they might be for

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 62 Resources

fun or serious (as with the Olympic Games), and so on. Nevertheless, the mere
existence of the umbrella term game seems to suggest some commonality,
some joint basis that warrants placing all of those different items into a single
category.
Wittgenstein’s solution for this apparent dilemma was to suggest a different
approach altogether: he introduced the notion of family resemblance. In
a family, not all members need to have anything in particular in common –
yet they might still be recognised by others as belonging to a single family.
Some share red hair, others share a particular form of nose, some might have
both features, and yet others might share something else entirely. As long as
there are some connections throughout, the family resemblance is recognisable,
providing a conceptual motivation for the overarching category. Families, as
such, are of course genetically related, so this is only a metaphor for explaining
what might be going on in other categories. Not all games have something
specific in common, but games around the world share some out of a wide
range of possible features that games might have. Most games will be for fun,
but those that aren’t share other features also found in other games.
These ideas triggered an astonishing range of empirical studies concerned
with the nature of categories in the human mind. Eleanor Rosch (e.g. 1978), for
instance, carried out a battery of studies that identified crucial structures of
categories, well beyond the earlier assumption that categories can be exhaus-
tively understood based on features defined by experts. From this line of work,
two main findings stand out that complement Wittgenstein’s ideas: first, that
categories normally exhibit a prototype structure, and second, that some cate-
gories are more basic than others.
Prototype structure means that some members of a category are regarded as
more representative than others. Consider the category bird. What kind of bird
comes to your mind first when you hear the word? Most likely, it will be one of
the most common birds encountered in your spatial environment: a robin
perhaps, or a blackbird. While the bird type seen as prototypical is culturally
dependent, the phenomenon as such is not: all cultures would regard some birds
as more typical members of the category bird than others. A rarely encountered
bird which exhibits unusual features (or lacks the usual ones), such as
a penguin, would seldom be regarded as a very good example of the category.
The other, equally important insight that came into focus in the 1970s
concerns a deeper understanding of the hierarchical structures of categories.
We all know that some categories are subtypes of other categories – for
instance, an armchair is a type of chair, and a chair is a type of furniture. Not
all of these category types are conceptually equal, or play the same role in
linguistic communication. Brown (1958) suggested that there is a level of
maximum utility, a word that is most frequently used rather than other options –
and this is what parents use when speaking to their children, and accordingly it

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 63

will be the word they learn. Thus, children will learn to use the word fish to refer
to a broad range of animals in the water, and differentiate on land between dogs
and cats.
Rosch et al. (1976) took these ideas further by examining in detail what
in particular characterises such a basic level of category usage. They found
that basic-level categories tend to share the highest number of features,
while being most discriminative from other categories. For instance, chairs
are structurally and functionally relatively similar, but clearly distinguished
from other types of furniture such as tables. The more specific notion of
armchair adds some specific features, but armchairs are still similar to
other kinds of chairs.
Both aspects of categorisation, prototypes and basic-level categories, are
cognitively very central and play a major role in the use of words. The under-
standing of conceptual semantics has benefited vastly from the discovery of
these structures in the way humans categorise. With respect to CODA, these
insights can be very useful for interpreting language data in the context of
a particular situation. For instance, if an experimental task triggered an unu-
sually high number of references to quadrupeds rather than dogs, cats, horses,
and the like, this would point to a major shift in categorisation (away from the
basic level) that is likely to be significant in the empirical context.
Also, it can normally be assumed that people refer to prototypical members
of a category when using a word; deviating from this will often trigger the
usage of a more specific term. Thus, when encountering a robin (in an environ-
ment where this bird is common), people might stick to the basic-level term
bird, but they might refer to a penguin as penguin because of its non-
prototypicality within the bird category. Such choices of words are meaningful,
and need to be interpreted in light of what is known about human
categorisation.
Categorisation is central for understanding the nature and structure of mean-
ing in the human mind, but it is not the only cognitive principle at work. The
most thorough and systematic approach to cognitive semantics to date was
provided by Talmy (2000). Talmy identified a range of organising principles
that explain how meaning is structured and expressed in language. First of all,
linguistic meaning is schematic rather than highly specific. This concerns all
sorts of words. To return to our example of a chair, the concept would invoke in
speakers’ minds a schematic image of a chair, rather than a specific one.
Speakers are usually able to sketch such schemas on paper from an early age,
demonstrating that they know how to abstract from specific features of parti-
cular examples of objects. But not only objects are schematically conceptua-
lised in this sense – other words are, too: even a (small and seemingly
insignificant) preposition like on exhibits a regular schematic conceptual
structure.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 64 Resources

Another organising principle discussed by Talmy (2000) relates to the

flexibility of language use. There is always more than one way of expres-
sing thoughts and facts. Very often, how people choose to say things
depends on what is important or in the foreground for the speaker, and
which kinds of things are seen in contrast to this. Talmy’s notions of Figure
and Ground are a central case in point. In the sentence The bike is next to
the church, the word bike is the Figure, and church is the Ground. To some
extent, the notion of Ground is similar to ‘background’. The Figure is in
focus, it is what is being talked about, against the background of something
that is (typically) already known. The church, in the example, would be
a prominent landmark that people know about, and which can be used to
describe the current position of the object in question (the bike). Turning
the sentence around would therefore be very odd – The church is next to
the bike. From an objective, spatial point of view, there would be no reason
why this shouldn’t be exactly the same spatial relationship. Speakers,
however, make their linguistic choices on the basis of underlying principles
such as foregrounding and backgrounding. Talmy (2000) suggests a wide
range of further principles in this regard. Many of these can be extremely
helpful for interpreting linguistic choices when doing CODA.
Altogether, cognitive linguistic research recognises a range of ways in which
meanings are organised in the human mind. These structures highlight the close
relationship between language and thought in general, representing a rich
resource for the interpretation of language used in specific situations.

3.1.3 Syntactic Structure Reflects Conceptual Structures

Similar to insights in semantics and in part overlapping in scope, cognitive
linguists have also looked at syntax from a conceptual perspective, and sug-
gested that structural aspects in grammar relate systematically to structural
aspects of conceptualisation. This view departs fundamentally from
approaches that address syntactic structures primarily to identify systematic
rules of representation (e.g. Chomsky, 1980), or to relate them to functions of
language use (as addressed in Section 3.2). One of the most prominent propo-
nents of cognitive grammar is Langacker (1986). Langacker’s approach is
parallel to Talmy’s approach to cognitive semantics in many ways. Like
Talmy, Langacker points to systematic abstract meaning structures across
domains in both semantics and syntax.
Langacker’s cognitive grammar is unconventional in many ways. For
instance, it makes heavy use of drawings to describe schematic meanings in
syntactic structures, such as Figure 3.1. This kind of metalanguage expresses
the view that grammatical structures, including word categories such as nouns
and verbs, are symbolic in nature, and not merely autonomous and arbitrary

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 65

(a) tr (b) tr (c) tr

lm lm lm

t t

GO AWAY GONE

Figure 3.1 Profiled meanings in Langacker (2002:6).

In (a), go refers to the interconnections between entities where a trajectory (tr)
moves away from a landmark (lm). In (b), the relationship of being away is
demonstrated by the distance between trajectory and landmark. In (c), gone
indicates a trace of the movement leading up to the state of being away.

structures of language. In other words, grammatical units express conceptual

meaning just as lexical items do, as discussed in the previous section.
Langacker (1986) spells out how schematic meanings that are associated
with individual lexical items combine with other schematic meanings in
a sentence structure. For instance, in drop the cup, the lexical item drop
schematically assumes two participants – one that drops and the other that is
dropped. Langacker calls these trajector and landmark (similar to Figure and
Ground in Talmy, 2000). Trajectors and landmarks combine in sentences by
profiling particular aspects of the described situation, that is, making them more
prominent than other aspects. In drop the cup, only cup (the landmark) is
explicit – but not who or what is doing the dropping (the trajector). This
activates a specific substructure of drop by providing information about one
of its participants. In a similar vein, Figure 3.1 contrasts the profiled meanings
of go, away, and gone.
The notion of profiling highlights that the same situation can be expressed
from different perspectives: compare The hill falls gently to the bank of the
river to The hill rises gently from the bank of the river. The distinct perspectives
induce a sense of directionality even though the described situation is static –
there is no motion involved in a landscape description. Langacker explains
such phenomena by introducing a distinction between sequential and summary
scanning. In sequential scanning, we experience events or situations as tempo-
rally extended, as when we follow a ball’s trajectory flying through the air. In
summary scanning, our experience is accessed as a whole; the ball’s trajectory
can be described as a curve, a single schema rather than a sequence of
successive states. Speakers’ choices when using language highlight the way
they conceive of the situations, and how they wish to profile them. The choices

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 66 Resources

are not inevitable, there are always different ways of profiling the same state of
affairs.
Altogether, much like the structures in meaning, language also reflects
structures on a grammatical level that systematically relate to structures of
thought. Analysing the syntactic structures in discourse data can therefore
highlight interesting aspects of thought, reflected in the constructions chosen
by speakers.

3.1.4 Usage Patterns Reflect Conceptual Structures

While much research in cognitive linguistics focuses on the structures of
language and what they are capable of expressing, some strands more directly
address the use of language. Apart from CODA, there are two main ways in
which language use is prominent in cognitive linguistics. On the one hand,
examining how language is used allows for the identification of generalised
patterns across contexts. This highlights how speakers tend to conceptualise
certain phenomena. On the other hand, generalised language usage patterns
explain decisive aspects of language acquisition.
Concerning the first idea, Levinson (2003) showed how speakers in different
cultures use their languages differently to describe spatial relationships. First of
all, he developed a theoretical account of spatial reference systems, and identi-
fied a limited number of options which he called intrinsic, relative, and abso-
lute. All reference systems involve a locatum and a relatum. The locatum is an
object whose location is to be described (e.g. where is the ball?) The relatum is
another object (or person) that speakers refer to in order to describe where the
locatum is. For instance, in The ball is in front of Anita, Anita plays the role of
relatum. In an intrinsic reference system, only the locatum and relatum are
needed, but the relatum must have some kind of intrinsic directionality. The ball
can’t be in front of the balloon because balloons don’t (normally!) have front
sides. But it can certainly be in front of the car – a car is a suitable relatum for an
intrinsic reference system, as it has a clear front (just like Anita).
If a relatum is used that doesn’t have any clear sides of its own (like
a balloon), speakers will normally need something else in addition – some
kind of perspective, or point of view. This is what Levinson (2003) calls
a relative reference system. The ball is in front of the balloon makes sense if
you use your own perspective on the scene, or your partner’s – you could say,
From your point of view, the ball is in front of the balloon.
The third type of reference systems in Levinson’s classification is called
absolute: a system for directions that is always present in a culture, independent
of perspectives or intrinsic sides. In Western cultures, compass directions are
the most typical examples for this. Thus, using an absolute reference system,
you could say, The ball is north of the balloon.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 67

This sentence might sound a tiny bit unusual – and indeed it is, for many
speakers. This was precisely the focus of Levinson’s extensive research pro-
gramme. He addressed the situations and contexts in which speakers regularly
used spatial reference systems across different cultures. Absolute reference
systems are widely known in Western cultures through the compass, but their
use seems to be restricted to certain contexts: compass directions are primarily
used to describe spatial relationships on a geographical scale. Whether people
otherwise prefer intrinsic or relative reference systems is difficult to say – this
seems to depend to a large extent on the usage context.
Across cultures, there are two remarkable patterns with regard to reference
system usage. First, not all languages provide a set of options that corresponds
to those found in English. While intrinsic reference systems seem to be pretty
much universal, relative reference systems are not available in all cultures – and
absolute reference systems can differ widely in their nature. Often, there are
specific environmental features that cultures draw on to establish absolute
reference systems. In the language of Tzeltal, for instance, people refer to
objects as being uphill or downhill from each other – in a very similar way as
one might refer to north or south in a Western society. This is because their
landscape involves a prominent structure that appears to feature very promi-
nently in the minds of Tzeltal speakers, even in situations where the hills
themselves are not visible (Brown & Levinson, 1993).
And this is the second remarkable pattern: that speakers make use of the
available reference systems in very different, culture-dependent ways.
Absolute reference systems, for instance, may not be part of many people’s
everyday language when describing object locations in English, but other
cultures use them all the time. They know at all times in what direction things
and locations are, and how they are actually related to each other in space.
Speakers of Yucatec Maya, for instance, typically show this actual relationship
in gestures, rather than relating gestures to their own current or imagined
perspective (Le Guen, 2011). Based on abundant empirical evidence,
Levinson (2003) demonstrated that this is not just a matter of linguistic
preferences, but instead reflects different patterns of thought. And this makes
sense. To constantly be able to describe object locations in terms of their true
relationship in space, equivalent to The ball is north of the balloon, people must
think about space in a very specific way – a way that is unavailable to speakers
in most Western cultures. Try it out: which compass direction are you currently
facing? If you know the compass direction without having to look at the sun or
calculate, you’re a bit like the Yucatec Maya. Chances are, however, that you
can’t tell – not without clues to support your thinking.
Research of this kind investigates the use of certain aspects of a linguistic
repertory to find out about patterns of thought, as with CODA; this includes
studies that elicit unconstrained natural language data in controlled task

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 68 Resources

settings (e.g. Senft, 1994). The account in Levinson (2003) focuses more on the
general patterns within a culture, rather than addressing specific challenges for
thought processes in empirical settings. The insights gained on that level are
valuable resources for a deeper understanding of the significance of linguistic
choices in more specific settings. For instance, if the general pattern in Western
cultures is that speakers rarely make use of absolute reference systems in indoor
settings, but a certain empirical scenario is found to trigger a preference for
compass directions, this will be very meaningful for the analyst.
Apart from studies that directly target usage patterns along these lines, the
overall use of language is understood as being the central foundation for
findings in cognitive linguistics across the board. Patterns in linguistic structure
such as those described in Section 3.1.3 are deeply motivated by the ways in
which speakers use language (Langacker, 2000). The more frequently people
use a specific linguistic structure, the more it will be entrenched in people’s
minds and in the linguistic system – and this tells us something about the
importance of this structure for the specific language and cultural thought.
Consider Langacker’s distinction of trajectors and landmarks, described in
Section 3.1.3. A great many verbs profile precisely two participants, mapping
onto these roles. For instance:
Example 3.1 Anne is reading a book.
Jake took the hint.
Grandma is starting to cook dinner.
The grammatical subjects of these sentences are trajectors (Anne, Jake,
Grandma), with the grammatical objects as their landmarks (a book, the hint,
dinner). While trajectors and landmarks play out differently for other sentence
structures and are therefore broader notions than the traditional subject and
object categories, the simple structure demonstrated here is nevertheless one of
the most cognitively salient, most frequent and entrenched, most central for
human thinking as expressed in language. It comes as no surprise, then, that the
same phenomenon – the ability of verbs to point to two (or less, or more)
participants with different roles – has received much attention across various
grammatical approaches. We will look at another one in Section 3.3. There we
will see how central aspects in Systemic Functional Grammar are designed
entirely around verbs and their decisive functional role in a clause.
Since frequency and usage are so central for explaining cognitively salient
structures, they are also important factors for child language acquisition
(Tomasello, 2003) – the second main idea mentioned in the beginning of
this section. Children often do not acquire individual words but rather pick
up frequently heard constructions, which they learn and first repeat as
a whole, and then expand creatively. They might only be able to repeat

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 69

single words to begin with, but a short word or phrase may stand for an
entire utterance – which would be more complex in adult language. Thus,
a young child might say ball to mean Where’s the ball?, or That! to mean
I want that! (Tomasello, 2003:65). Phrases like Lemme see or Gimme that
are likely to be understood by the child, at first, as wholes, and only later
parsed into their constituent elements. At some point, they learn to replace
individual elements out of phrases such as Where’s the X?, I wanna X, More
X, It’s a X, and so on. Such constructions are central and frequent in
children’s lives; they represent usage patterns that directly link up with the
way they are starting to understand the world and interact with other human
beings. In this way, a child’s grammatical repertory gradually grows and
becomes more complex, as a greater range of usage patterns become
relevant.
This also explains why it is very different to learn a second language: usage
patterns are already far developed and highly varied. It takes a long time to fill
these pre-existing conceptual ‘slots’ in a new way, with a different language.
This is even more challenging if the languages are built around different usage
patterns, or patterns of thought. If you grew up using intrinsic and relative
reference systems to describe where objects are, you won’t find it very easy to
learn a language in which absolute reference systems are the norm. Indeed, you
might learn the words as such easily enough, but using them appropriately
across a range of contexts might be quite a different matter.
Altogether, patterns of usage play a major role in cognitive linguistic
research. Usage patterns explain certain structures in linguistic systems as
well as in how they are acquired. As usage patterns differ in relevant ways
between languages and cultures, research into how language is used across
cultures, and in speakers of more than one language, sheds light on questions
concerning the interrelationship between language and cognition.

3.1.5 Conceptual Structures Start from Embodiment and Can

Be Transferred
Due to the fact that usage is central (as just outlined), language has developed in
close relation to our lives, our ways of interacting with the world. This point
leads us to the final central insight in cognitive linguistics that will be sketched
here: meaning itself transfers from concrete physical action events towards
more abstract settings. This insight is central for one of the most widespread
and well-known areas in cognitive linguistics – the investigation of transferred
meaning, often called metaphors. With their best-selling book Metaphors We
Live By, Lakoff and Johnson (1980) triggered a keen interest in abstract uses of
language, shared by a multitude of researchers throughout the world who aim to
discover how and why metaphors characterise our thought and language.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 70 Resources

For instance, much of our thinking has to do with value, either good or bad.
Value, as such, is not physically accessible; it is more abstract than actions and
things. But frequently good is related to more in the physical world, and more is
related to up: more coins are good, and they add up to a higher pile! Such
associations make the idea of value less abstract, more accessible in the real
world. Over time, in English, good has become generally linked with up, as we
can see in many expressions, as well as its counterpart, bad, being associated
with down:
Example 3.2 Fran is in high spirits today.
Matt scored lower than his friends.
After a period of feeling downhearted, they now became
upbeat about the plans.
The party was highly successful.
Lakoff and Johnson (1980) demonstrated that systematic patterns like these
can be found in language in many ways that we aren’t normally aware of. To
take another example, relationships are frequently associated with journeys, as
seen in expressions like the following:
Example 3.3 They started as a couple in 1982.
Anita and John had a fine friendship, without many hassles
along the way.
Why did our love have to end?
The link is established through the joint notion of time: relationships develop
over time, and journeys unfold in space as well as time. Space is physical, but
time isn’t – so space is brought into play in order to express the more abstract
notion of a relationship. There are many metaphors that relate to space in some
way. Indeed, time itself is very often expressed through the language of space:
Example 3.4 It was a long day.
Christmas is around the corner.
Good times are ahead.
In many of the expressions that invoke a transfer from a physical domain
(such as space) to an abstract one (such as time), prepositions play a major role.
By examining these prepositions in detail, Tyler and Evans (2003) spelled out
how transfers of meaning can develop. For instance, the preposition over has
a complex network of interrelated meanings, as demonstrated by the following:

Example 3.5 The cloth is over the table.

He jumped over the fence.
Like a bridge over troubled water . . .

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.1 Cognitive Linguistics 71

The spatial relationships expressed in these examples are all somewhat dif-
ferent. The fact that various spatial meanings can be conveyed by the preposition
over is the basis for the emergence of diverse non-spatial meanings, as in:
Example 3.6 She took a long time to get over the disappointment.
Over time, things became better.
I will think it over.
Tyler and Evans (2003) show how abstract meanings such as these relate to
spatial meanings of over in a complex and interconnected system where spatial
relationships are transferred to abstract domains. This is contrary to the wide-
spread idea that prepositions express a random set of unrelated meanings.
Goldberg (1995) applied the idea of transferred meanings to the level of
grammar (related to the ideas discussed in Section 3.1.3). For instance, consider
verbs that involve three elements in a sentence. One of the most frequent
examples of this type of verb is give. The grammatical pattern associated
with the verb give reflects direct practical experience where things change
hands: somebody gives something to somebody else. The three elements can be
expressed in two different ways:
Example 3.7 Anna gave Paul a dog.
Example 3.8 Anna gave a dog to Paul.
Both constructions are common, but they are associated with somewhat
different usage patterns – each construction can seem odd or just right, depend-
ing on various contextual factors that won’t be explored further here (though
the topic is interesting to pursue in its own right). There is also the possibility of
leaving out the giver:
Example 3.9 Paul was given a dog.
Now, the same notion of transfer can be expressed without actually using the
verb give:
Example 3.10 Joe painted Sally a picture. (Goldberg, 1995:143)
The use of this particular structure, which resembles the grammatical struc-
ture associated with give, automatically invokes the idea that Joe gave the
picture to Sally when he was finished. Moreover, Goldberg (1995) points out
that Joe must have intended for Sally to have that picture while he was painting
it; it is not possible that he painted it for someone else and then later changed his
mind. This is how strongly the giving notion is invoked by grammar – simply
by the use of a construction that involves three elements and represents them in
this particular way.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 72 Resources

Altogether, many patterns in linguistic systems can be traced back to the fact
that we live and talk in a physical world. Frequently experienced embodied
patterns, such as transferring goods from one person to another, or stacking up
things in a growing pile, are reflected in structures of meaning and grammar in
language, allowing for a transfer into more abstract domains of thought.
This concludes my brief introduction to aspects of cognitive linguistics that
may be useful when doing CODA. There is a lot more to be found in this wide
field, and my selection of insights is not intended to be representative.
However, I hope that my short introduction might provide a useful starting
point for understanding the language people use to express their thoughts.

3.2 Discourse Analysis

We have already discussed diverse relevant methods and approaches, but
undoubtedly the most common label typically associated with analysing the
use of language is discourse analysis. This is because ‘discourse’ simply means
language being put to use in actual situations, to communicate and to express
thoughts and ideas, in spoken or written language. The label ‘discourse analy-
sis’ encompasses a wide range of methods that generally aim to gain a better
understanding of discourse. Not all of these methods are primarily linguistic.
Typically, instances of discourse are analysed in relation to the communicative
purposes they serve in a particular context. Especially in linguistics, the term
‘discourse’ is often seen in contrast to examples of individual sentences that are
examined with respect to their structural and semantic features out of context.
Discourse analysis means taking into account aspects on a larger scale than
words or sentences – aspects that pertain to the wider context of language use.
Texts have structure just as sentences do, although the structures might be
more varied. Many written texts start with some kind of introduction, followed
by a main body (which may also have some internal structure) and
a conclusion. Other discourse types have a very different structure; a phone
call would begin and end with a greeting, and perhaps queries and wishes about
well-being, rather than an introduction and a conclusion. Analysing common
and systematic structures in different discourse types can lead to deeper
insights about the nature of discourse in general. And specific aspects can be
helpful in educational or professional training contexts. Knowing how
a newspaper report is written to achieve the intended effects on the readership
can be very useful when attempting a career in journalism.
Extending the scope on the level of language to include a piece of discourse
that is larger than a sentence is one way in which context (or co-text) plays a role.
Most approaches in discourse analysis go further than this, beyond the discourse
as such. They also look at various factors of the context in which discourse takes
place outside language. This would include situational factors such as the time

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.2 Discourse Analysis 73

and place of the discourse, who is speaking to whom, and the like. Many
approaches focus on the significance of a specific instance of discourse within
society – such as a public speech given by an influential political leader at
a certain moment in history. This provides a perspective that motivates much
work on discourse. Especially in the strand known as Critical Discourse Analysis
(Fairclough, 1995; van Dijk, 2008), the main aim is to understand how language
is used in society to influence, persuade, or manipulate other people.
CODA, with its focus on language used to express thought (rather than to
manipulate), is not primarily critical in general. Nevertheless, the insight that
society and situational context play a major role in the interpretation of
language is certainly relevant. In fact, there are a number of connections
between cognitive linguistics and Critical Discourse Analysis (Hart & Lukeš,
2009). For instance, the use of particular metaphors and construals in political
or journalistic discourse not only reveals, to some extent, how the writer thinks
about the matters at hand, but may also have certain effects on readers.
Teun van Dijk (2000), in a non-academic online article, offered a number of
relevant thoughts concerning the notion of Cognitive Discourse Analysis; these
included aspects of discourse processing as well as features of text that point to
particular cognitive structures or operations, such as presuppositions and
implications, lexical connotations, local coherence, and the like. Rather than
developing this particular notion further in his work or applying it to specifi-
cally collected instances of language use in relation to thought processes, van
Dijk has subsequently focused on the interplay between cognitive aspects
(intentions, strategies, knowledge) and sociocultural context models (van
Dijk, 2008). However, his ideas are indeed very relevant to what is presented
in this book: for instance, we will take a closer look at the notion of presupposi-
tion in Section 3.2.2.
Apart from these general frameworks, there are many individual approaches
and contributions that aim to identify systematic patterns in language usage – and
many insights gained in this regard can be very relevant for CODA studies. In
the following, we will first look at methodological aspects in discourse analysis
(Section 3.2.1). Then we will outline a number of specific aspects that concern
discourse rather than individual sentences, all of which may be relevant for
CODA: information structure and presupposition (3.2.2), coherence devices and
discourse markers (3.2.3), and lexical choices and specificity (3.2.4).

3.2.1 Methods in Discourse Analysis

As we have seen at the start of Section 3.2, different types of discourse analysis
can be used to uncover various (social, cognitive, etc.) phenomena reflected in
discourse. It is important to recognise that there is no way of analysing any
instance of discourse completely neutrally, or in its entirety. Discourse analysis

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 74 Resources

normally happens against the background of a research project or goal, and this
determines the perspective adopted to analyse the data at hand. It affects the
range of linguistic, social, cultural, or other phenomena that are taken into
account in the analysis, and often also the type of data that are collected in the
first place. Sometimes it is the data as such that drive the analysis, such as in
corpus-driven discourse studies or conversation analysis; in such cases,
researchers aim to avoid any preconceived expectations or frameworks.
We will look at specific procedures for CODA in Chapter 8. Here, the aim is
to put the methodology into context and highlight the benefits of different, but
related approaches. There is nothing self-evident or inevitable about a choice of
data and analysis approach. Methodological choices of any kind are mean-
ingful, and they affect what kinds of data are collected, what kinds of analysis
procedures are used, and what kinds of insights are expected. Along the way,
there is also the step of transcribing language data – and this, as well, consti-
tutes a choice that can significantly affect the outcome.
A distinction is sometimes made between Discourse (with a capital D) and
discourse (with a small d) (Gee, 2013). The first refers to the Discourse that
happens in society at a larger scale, embedded in other social practices within
a community (e.g. customs, traditions, shared concepts). The second refers to
language in use, that is, specific instances of discourse that are part of a larger
Discourse. Analysing discourse (instances) therefore necessitates taking into
account the kinds of Discourses in society that are associated with this parti-
cular instance. It means understanding a particular interaction as an example or
instantiation of what happens at a larger scale. This perspective will affect what
aspects are looked at, and how they are interpreted.
The idea of experimental control is alien to this approach, because controlled
experimentation seeks to eliminate the societal aspects that are the very target of
discourse analysis of this type. Instead of collecting data under controlled
conditions, therefore, samples of discourse are taken directly from the situations
in which they happen, and analysed in light of the various contextual factors that
come into play: the relationships between interactants, the social and spatiotem-
poral settings, and anything else that might be relevant for the unfolding of
discourse. If a fully authentic sampling of data is impossible, semi-structured
interviews may be a useful way to collect data. A flexible interview setting allows
the participants to talk relatively freely, in a non-prescribed way, and this in turn
allows the interviewer to gain insights from the analysis that are not directly
influenced by any predetermined aspects of the interview itself.
Any instance of spoken language contains a wealth of information that may
or may not end up in a transcript. The language data can be transcribed simply
to represent content – and this is the normal level of transcription that one might
find on the Internet, when looking for written versions of a public speech, court
proceedings, and the like. Such content-based transcripts are typically regarded

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.2 Discourse Analysis 75

as sufficient to represent spoken discourse outside the realm of discourse

analysis. Very often, the versions that are publicly available are edited; the
rough edges of spoken discourse are smoothed out, certain formulations are
improved, interruptions are ignored, and so on. Such editing makes the tran-
scripts more readable and therefore serves their intended purpose: to make the
content of a public event accessible in written format.
However, any deviation from the actual spoken event renders the transcript
less reliable and therefore less suited for the purposes of discourse analysis.
Researchers wishing to make use of existing transcripts should normally
double-check the written version, in order to align it as closely as possible
with the original discourse event.
Beyond deviation, editorial improvements, and inaccurate transcription, a
transcript can be a true record but still inadequate for the purposes at hand. This
is because transcribing always requires decisions as to what to include and what
to ignore. It can be challenging to obtain all of the information that would be
necessary for a complete analysis that addresses the current research goals.
Contextual factors that pertain to the spatiotemporal, social, and cultural setting
might be decisive. Gestures and mimics, body posture, and volume of voice
may be recorded and added to the transcript, alongside language. The language
itself may be transcribed at various levels of detail. At perhaps the most
developed end of the scale of precision – opposite to publicly available non-
academic general-purpose transcripts – one finds the well-established metho-
dology of conversation analysis (Sacks, 1992). This method builds on the
insight that an extremely detailed transcription of a speech event can highlight
subtle mechanisms in conversation that one might otherwise miss. Compare the
following (taken from an insightful tutorial1 on conversation analysis provided
by one of its proponents, Charles Antaki):
Example 3.11 Zoe: mum
Lyn: hello I’m here
Zoe: okay
Example 3.12 Zoe Mum?
Lyn hello (pause)
Lyn I’m here (pause)
Zoe okay- (pause)
Lyn ((coughs/clears throat)) (pause)
Example 3.13 Lyn ((at table with papers))
((door?)) ((faintly, off camera))
(pause)

1
[Link]

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 76 Resources

Lyn [[looks up and over her shoulder towards door; holds gaze
while scratching cheek; looks down again]]
Zoe ((off camera)) Mum?
Lyn hello [[gaze stays down]]
(3 sec)
[[at end of which she
orientates upper body
towards door]]
Lyn I’m here
(brief pause)
Zoe okay
Lyn ((coughs/clears throat))
[[off camera: three ?
crockery bangs]]
(pause)
((door handle opening))
Example 3.14 Zoe ((off camera)) Mum?
Lyn .pt h↑ello
(3 sec)
Lyn I’m here
(.8)
Zoe (okay)-
(1.0)
Lyn ((coughs/clears throat))
((clatter))
(3.0)
While Example 3.11 simply provides the words that were spoken in the
transcribed scene, Example 3.12 represents pauses and a non-linguistic human
noise (coughing) that might seem relevant. However, this is not very detailed.
Example 3.13 includes various non-linguistic aspects as well as body postures.
Example 3.14, finally, exemplifies a transcript that conversation analysts would
typically use. It does not include the non-speech details provided in Example 3.13,
but is instead very informative concerning the speech events themselves, including
length of pauses and aspects that pertain to speech volume and intonation contour.
These are represented by arrows, underlining, dashes, and the like.
As these examples demonstrate, transcripts can be veridical and accurate, but
still vary considerably in what they represent. The analyst decides which aspects to
include, which are relevant for their analysis goals. This may go well beyond the
aspects relevant to speech that are transcribed in conversation analysis, and also
beyond other non-speech events observed directly in the scene. Note that none of

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.2 Discourse Analysis 77

the examples (3.11–3.14) reveal anything about the speakers’ social background,
the time of day, the previous and following discourse, and a variety of other factors
that might explain what is happening in this short conversation. Discourse analysis
means making decisions in this regard, but not random ones; they need to be well
motivated in light of the research goals at hand.
Similarly, the methods undertaken when analysing the data can be rather
varied, as they include systematic scrutiny of linguistic, metalinguistic, and
contextual aspects as deemed appropriate by the analyst. This kind of choice
also needs to be well motivated and appropriate for the purpose of analysis.
Antaki et al. (2003) point to a number of things that one might confuse with
discourse analysis or that may go wrong in attempts at doing discourse analy-
sis – such as simply providing summaries, taking sides, picking out individual
quotations, or spotting random individual features.
As I hope to have shown in this section, existing discourse analytic
approaches have a lot to offer in terms of systematic methodology. Various
excellent introductory accounts, such as Gee (2010) and Johnstone (2008), cover
the field more deeply than can be achieved here. We will now look at a number
of linguistic aspects that are frequently encountered in discourse analysis, and
that may be useful to consider from a cognitive perspective as well.

3.2.2 Information Structure and Presupposition

Compare the following two sentences:
Example 3.15 Sam gave Rosie a book that was good to read.
Example 3.16 The book that Rosie was given by Sam was good to read.
Both sentences express the same information: that Sam gave Rosie a book and
that the book was good to read. However, both sentences make different assump-
tions. In both cases, proper names are used to refer to people: Sam and Rosie. We
either know them already, or we assume that we will soon learn more about them
(unless they appear without context, as happens to be the case here!). In novels
and stories, characters are very often introduced by name only, and the reader
gradually becomes acquainted with them through further reading (or listening).
In real life, upon hearing this kind of reference in a conversation, we might ask,
‘who is Sam?’ – unless we do already know who the speaker is referring to.
The two sentences are equivalent in this respect, but there is a third element in the
event described, namely a book – and this element is not treated in the same way.
Note, first, that Example 3.15 uses an indefinite article (a book), whereas Example
3.16 uses a definite one (the book). The definite construction signals to the reader or
listener that the book should already be known. But in fact, the entire event of giving
the book to Rosie is described as new or previously unknown in Example 3.15,

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 78 Resources

along with the fact that the book was good to read. As a consequence, we could
respond to Example 3.15 by either saying, ‘Oh, did he?’ or ‘Oh, was it?’ In contrast,
in Example 3.16 the only information that is treated as new is the fact that the book
was good to read. So a suitable reaction could be ‘Oh, was it?’ – but not ‘Oh, did
he?’ If we don’t remember that Sam gave a book to Rosie, responding to that part of
the message becomes more complex: ‘Oh, did he give a book to Rosie?’
Many insights can be gained by looking at the ways in which information is
presented. The term information structure refers to the sequential order of
information in a sentence, along with other ways in which information can be
represented as new or already known across different languages (Steube,
2004). And the term presupposition refers to what is assumed to be already
known, as signalled by definite articles and other linguistic devices (van
Deemter & Kibble, 2002). Both work together in discourse, organising what
is known and what is communicated as new. This contributes to the economy of
language use and communication: new thoughts can be interpreted in the
context of existing ones, finding their rightful place in the listener’s mind,
without having to build up an entirely new knowledge base from scratch.
Apart from definite articles, we find the following constructions that pre-
suppose various facts (Spenader, 2002):
Example 3.17 That John stole the money matters.
Example 3,18 Trizia started sailing when she was eighteen.
Example 3.19 It was the bachelor who got madly drunk that night.
In Example 3.17, it is presupposed that John stole the money – there is no
question of that. The only new information is that this fact actually matters. In
Example 3.18, the presupposition triggered by the verb start is that Trizia did
not sail before she was eighteen years old. In Example 3.19, it is already known
that someone got madly drunk that night – the only question is who it was
(namely, the bachelor). Note how the presupposition triggers relate to sentence
structure. New information frequently comes at the end (as in Example 3.17),
but not always (Halliday & Matthiessen, 2014). The specific structure in
Example 3.19 (called cleft structure) enables the sentence to put a different
element into focus (the bachelor), even though it is not presented at the end.
There is a lot more to discover about information structure and presupposition; it
is a fascinating subject that has been taken up in many discourse analysis
approaches. With its many implications for the ways in which knowledge is
organised in our minds, this analysis perspective is undoubtedly relevant for
CODA. We looked at some examples already in Chapter 1, and will return to
these issues in Chapter 6 when considering how CODA can be used to examine
inference phenomena.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.2 Discourse Analysis 79

3.2.3 Coherence Devices and Discourse Markers

How do sentences hang together, and how are utterances and ideas related to
each other? How are such relationships expressed in language? Consider the
following:
Example 3.20 Mary went to the bank. She needed to prepare dinner for fifteen
people the next day.
When reading two sentences in direct succession, we expect that they relate
to each other. In Example 3.20, there is only one direct indication that the two
sentences are related: the reference to Mary through the pronoun she in
the second sentence. One might also say that the word next implies
a temporal relationship between the two sentences, where the event described
in the second sentence takes place a day later than the event described in the
first sentence. Such effects fall under the notion of cohesion as examined in
Systemic Functional Grammar (Halliday & Matthiessen, 2014).
However, the most important aspect remains entirely implicit: namely the
question of why the two events are mentioned together at all. This aspect is
normally examined in terms of coherence. Coherence is somewhat more compli-
cated than cohesion, as it concerns how we interpret sentences and how we
understand their interrelationships on the level of (pragmatic) meaning. To under-
stand how two sentences add up to a coherent whole, we need to make another
inference – similar to the kinds of inferences that we discussed in the previous
section, but on a different level. This particular kind of inference happens on the
level of discourse relations. We assume that there is a reason for putting two
sentences together, and therefore we infer a relationship between them.
In the above example, the inference will typically be causal: Mary went to
the bank because she needed to prepare dinner for fifteen people the next day.
There are a number of associations going on in our minds when we draw this
conclusion, such as:

Example 3.21 Fifteen people is a large number – most people wouldn’t do this
every day. If it was normal for Mary to cook for fifteen people, it
wouldn’t have been mentioned.
If Mary does not cook for fifteen people every day, she might not
have all the ingredients in the house that are needed to cook for
such a large number.
If she doesn’t have the ingredients in the house, she’ll need to go
shopping in advance before she can get started.
The dinner is only one day away. She needs to get started soon as
this is a big task.
She will need to go shopping now. For this, she will need money.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 80 Resources

She is going to the bank because she needs money for the
shopping.

The list of associations could be continued, spelled out in more detail, or

made shorter. For some people, there may be more direct links between going
to the bank and preparing dinner. Sperber and Wilson (1986) spelled out how
such logical inference processes may take place in the mind, enabling us to
determine how sentences can be understood as relevant to each other.
To avoid the effort of making these complicated inferences and assuming
that the relationship is causal, the conceptual link between the two sentences in
Example 3.20 could also have been made explicit by using discourse markers
that signal coherence (Sanders, 1997; Schiffrin, 1987). Both of the following
would contribute to this, in different ways:
Example 3.22 Mary went to the bank because/as she needed to prepare dinner
for fifteen people the next day.
Example 3.23 Mary went to the bank. She needed to prepare dinner for fifteen
people the next day, you see.
Example 3.22 directly states a causal relationship using a discourse con-
nective (because/as). This is the most straightforward way of signalling
a discourse relation. Example 3.23 remains implicit with respect to the actual
discourse relation, but the discourse marker you see appears to invite the
listener to establish a meaningful connection, supporting the inference of
a causal relationship.
There is abundant literature on discourse relations and on the ways in which
discourse markers contribute to them (and do other things in addition).
Independent of the precise goals of discourse analysis, researchers frequently
take these phenomena into account in order to gain a deeper understanding of
the analysed discourse. Especially with implicit coherence, however, there is
always an element of uncertainty as to whether the inferences made by the
listener (or by the analyst) correspond to the intentions of the speaker.

3.2.4 Lexical Choices and Specificity

As a final example of the kinds of things that are frequently observed in
discourse analysis, let’s look at the level of words. Compare the following
two possible utterances:
Example 3.24 There’s a man and a dog on the beach.
Example 3.25 I can see a young athlete, perhaps twenty-five years old, and his
pet, walking along the beach.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.2 Discourse Analysis 81

Obviously, Example 3.24 is much shorter than Example 3.25. Despite this,
the speakers may both be describing the same situation; for instance, they may
be looking at the same picture. One description is simply a bit more specific
than the other. How exactly does this work? Clearly, Example 3.25 is not
simply an expansion of the same – this would have been the case in
a sentence like this:
Example 3.26 I can see that there’s a young man, an athlete, perhaps twenty-
five years old, and a dog, his pet, walking along on the beach.
This example combines both sentences – so now we have a true expansion,
of both original sentences. To get a clearer idea of how Example 3.25 is more
specific than Example 3.24, we need to do more detailed analysis.
While the reference to the beach is identical, the other two participants are
referred to in different ways. In Example 3.24, both of them – the man and the
dog – are referred to on the basic level of categorisation (see Section 3.1.2).
This is the most neutral way of referring, which can be used for a relatively
generic description. In Example 3.25, the case is different. Instead of a generic
reference to a man, this time we learn that the man looks like an athlete, and
that he is young; we even get an estimate of the age. In this context, athlete is
subordinate to the basic-level category man: an athlete is a specific kind of man.
The reference to the dog is rather the opposite, as a dog is a specific kind of
pet. This, as well, is a deviation from the generic basic level; it points to the
specific relationship between the man and the dog in a way that is not accom-
plished in Example 3.24. This is reinforced by the personal pronoun his, which
is missing in Example 3.24. We also get a sense of motion and action, expressed
by the verb walking in Example 3.26; this is completely absent in Example
3.24, rendering the description far more static – which is supported by using the
existential phrase there is.
Lexical choices are meaningful. They represent the way in which the speaker
wishes to talk about what they see or think. A generic level of representation,
using the basic level of categorisation, may not tell us much about the speaker
beyond their desire to remain relatively neutral. Anything beyond this shows
a bit more; in the above example, it reveals an eye for details such as the man
looking athletic and young, as well as the relationship between man and dog.
The interpretation of such choices is most successfully achieved through
contrast and comparison of different samples or instances of discourse, ana-
lysed in light of theories about how lexical concepts are organised and accessed
in the mind (see Section 3.1).
Since the same real-world objects and individuals can be described in many
different ways, on different levels of specificity and with different associations,
some lexical choices may sometimes seem to be more appropriate than others.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 82 Resources

Indeed, speakers sometimes struggle to find a good way of describing things

that is suitable for the context. In some situations, it may be possible to identify
an ideal level of reference; this provides an opportunity for researchers to
examine how good speakers are at identifying it. Consider the following
random list of numbers:
Example 3.27 7, 5, 6, 3, 79, 288, 389, 230, 12, 35
Now, if a speaker wanted to refer to one of these numbers without actually
mentioning it (this could be the rule of a game), they could say:
Example 3.28 It’s the even one-digit number that is in third position in the
sequence, from left to right.
Clearly, the speaker is referring to the number 6. But they could have done
this in a simpler way: the information given in Example 3.28 is overspecified. It
would have sufficed to just refer to the even one-digit number, since only
number 6 falls into this category. And likewise, it would have sufficed to
only refer to the third position of the sequence. Speakers do often include
redundancy of this kind in their descriptions; overspecification is actually quite
frequent in language use (Arts, 2004). In contrast, Example 3.29 would have
been underspecified because there are several one-digit numbers:
Example 3.29 It’s the one-digit number.
Further ways in which utterances can be ambiguous and underspecified can
be found in a collection of papers edited by van Deemter and Peters (1996),
which provides a valuable resource for discourse analysis. Relatedly,
Bierwisch and Schreuder (1992) made the point that single lexical items can
serve various purposes, depending on the syntactic and discourse context. The
associated concepts can normally be disambiguated and identified, as in the
following (Bierwisch & Schreuder, 1992:31):
Example 3.30 He left the institute an hour ago.
Example 3.31 He left the institute a year ago.
The same word institute denotes a building in Example 3.30, and in Example
3.31 an institution or affiliation. It is interesting to see how the time frame
suggested at the end of the sentences – an hour as opposed to a year – gives not
only the word institute a different interpretation, but also the word left. Within
a short time frame, one might automatically assume that the described event is
on a small temporal scale, as in the physical event of leaving a building. Within
a long time frame such as a year, such an interpretation would normally seem
inadequate; therefore, the assumption is that the leaving event is much more

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.3 Functional Grammar 83

consequential and concerns the entire life situation. All of these assumptions
are (again!) inferences that the reader or listener derives from world knowl-
edge that is filled in upon the mention of individual lexical items, such as
temporal references. Discourse analysis often aims at revealing insights of
this kind, and the interpretation of linguistic choices in relation to cognition is
no exception.
Generally, it often makes sense to look closely at the lexical choices
a speaker makes, for any number of reasons. This is a frequent method across
discourse linguistic approaches. With CODA, an analysis focus on lexical
choices means identifying lexical items that reveal certain relevant aspects of
cognition (such as focus of attention, see Chapter 4, or level of granularity,
Chapter 5). In other approaches in discourse analysis, lexical choices may
reveal further interesting patterns, depending on the specific type of discourse
at hand. Altogether, as we have seen in this section as a whole, many of the
principles used in traditional discourse analysis are equally valid for CODA.

3.3 Functional Grammar

Systemic Functional Grammar (SFG) (Halliday & Matthiessen, 2014) is
a widely recognised grammatical framework that has a lot to offer for
CODA research, both from a theoretical point of view and with respect to
a number of specific analysis perspectives. We have already come across this
framework several times in this book, and will encounter further details later
on; this section offers a concise introduction for reference. We will start with
an overview of the theoretical basis (3.3.1), which enables us to look at the
three main levels of analysis that are central to SFG: the interpersonal level
(3.3.2), the ideational level (3.3.3), and the textual level (3.3.4).

3.3.1 Theoretical Foundations of Functional Grammar

Doing CODA means looking at linguistic choices: identifying what speakers
do with language, out of the many things that they could have done.
Understanding the significance of what speakers choose to do, therefore, has
a lot to do with understanding what else they could have done. SFG provides
a framework for this kind of understanding. It accounts for the various func-
tions of language by showing what the overall system of linguistic choices
looks like and what it offers to speakers. In fact, every aspect of grammar can be
explained using the notion of choice, as part of the overall system. Take, for
instance, the grammatical notion of mood. The same content can be expressed
in various ways depending on mood:
Example 3.32 Luke will travel on Thursday.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 84 Resources

Example 3.33 Will Luke travel on Thursday?

Example 3.34 Luke, travel on Thursday!
The choices that are available to express this same content represent the mood
system in English: namely, declarative (Example 3.32), interrogative (Example
3.33), and imperative (Example 3.34) mood. These are the options on a grammatical
level that a speaker has at their disposal. The speaker’s choice in a situation
expresses their relation to it: they either express the content as a fact, or they
question whether this same content is a fact, or they command it to become a fact.
Mood therefore has to do with what goes on at the interpersonal level – the
level where grammar shows the relationship between speaker and hearer (or
writer and reader), which we will explore in Section 3.3.2. The content, as such,
is expressed by grammatical choices on the ideational level, as we will see in
Section 3.3.3. In addition, the information in the text is organised by a number
of devices that show what a speaker takes as relevant and wishes to point out.
This relates to the notion of information structure that we briefly looked at from
a theory-neutral point of view in Section 3.2.2. In SFG, this is addressed in
terms of the textual level of analysis (Section 3.3.4). At all three levels, speak-
ers always make a choice whenever they use language. The grammatical
system of the language they use provides them with options in many different
ways, and all of these choices are meaningful: after all, it is a big difference
whether somebody simply notes that Luke will travel on Thursday, or wonders
if that’s the case, or commands him to do it.
With this notion of choice as a central concept, SFG provides a thorough
toolbox for interpreting every instance of discourse in relation to its occurrence.
Awareness of the overall repertory, and of the distinctions between the options,
enables a valid assessment and interpretation of specific linguistic choices in
relation to the situational context. While any account of grammar in linguistics
provides relevant insights with respect to a language at hand, SFG provides
a particularly useful resource as it is designed up front in terms of a network of
options. This network allows for identifying the meaning relationships and
metafunctions that constitute the structure of lexicogrammar.
Knowing how the network operates in general helps in identifying aspects of
language that are specifically meaningful for a certain analysis perspective. For
CODA purposes, the distinction between different levels of meaning and
functions of language highlights those linguistic features of a text that relate
to cognitive aspects, alongside other features that pertain to interpersonal or
textual functions, or simply represent what’s happening in the real world. Here,
knowing how the network works in general helps to understand what
a particular choice shows us about the way the speaker represents their
thoughts.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.3 Functional Grammar 85

Other approaches use SFG for a range of other purposes and are less focused
on cognitive aspects. In fact, the main founders and current proponents of SFG
are primarily oriented towards understanding language in relation to society.
Text and discourse are seen as meaning-making instances of language; they are
best interpreted in relation to the overall system, taking into account the context
of where they are produced. Moreover, structures of linguistic features in texts
are explained by reference to the purposes for which the texts are produced.
Different kinds of situational contexts call for different ways of using the
network of options, and SFG provides a precise basis for explaining how this
is done.
SFG was developed for the English language (Halliday & Matthiessen,
2014), but its approach and framework are not limited to it, and have in fact
been transferred to a broad range of other languages (Butler & Gonzálvez-
García, 2014). Halliday himself has worked extensively on Chinese and has
included various discussions of it in his work. Schlobinski (1996) used insights
from functional grammar as a basis for empirical linguistic analysis methods
for German. The list could be continued. Altogether, a large and vibrant
community works with and expands SFG today, both as part of and beyond
the International Systemic Functional Linguistics Association.
After these general observations, we will now turn to the specific analysis
levels that are prominent in SFG, starting with the interpersonal level in the
next section.

3.3.2 The Interpersonal Level of Analysis in Functional Grammar

At the interpersonal level of language, grammatical choices tell us how
a speaker relates to a certain idea, and how they wish to convey that idea to
the listener. One might say that, at this level, the relationship between speaker
and hearer is represented in grammar. We have already encountered the notion
of mood, which tells us whether the speaker represents something as a fact
(using the declarative mood), or questions it (using the interrogative mood), or
commands it to become a fact (using the imperative mood). This aspect belongs
to the interpersonal level of analysis.
There is an interesting grammatical observation here. In English, mood
choice affects only two elements of a clause – the Subject and the Finite.
Returning to Examples 3.32–3.34, if we say Luke will [travel on Thursday],
the Finite will follows the Subject Luke, and this indicates a declarative clause.
If we turn the two around, we get Will Luke [travel on Thursday] – and as
a speaker of English we immediately recognise the interrogative. That’s all it
takes. The imperative is special in that the Subject can be left out – though it
may be included for clarity: (Luke), travel [on Thursday]! Therefore, analysis
on the interpersonal level in English strongly focuses on Subjects and Finites –

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 86 Resources

which are not always easy to identify, and I will leave discussion of this to the
grammar books (e.g. Halliday & Matthiessen, 2014).
But there’s more to observe. For one thing, speakers aren’t always sure
about facts, or there may be other ways in which the message is less than
straightforward. For another, it is not very polite to express direct com-
mands, and sometimes speakers may want to avoid asking direct questions
using the interrogative mood. As a result, the function of an utterance is
not always directly reflected in its grammatical form. Nevertheless, the
grammatical system does reflect the underlying interpersonal relationships
in a range of ways.
Let’s look at modality first. Modality means that the message is somehow
modified with respect to the speaker’s stance towards it, or the extent to which
something can be argued. Not every declarative message directly reflects
a clear and indisputable element of truth; facts can be represented as true or
not true, or somewhere in between these two poles. For instance, the speaker
might want to modify the definiteness of Luke’s travel plans (in Example 3.32)
a little, and change various aspects around – all related to the Finite:
Example 3.35 Luke may/might/could travel on Thursday.
Example 3.36 Luke must/should travel on Thursday.
Example 3.37 Luke will certainly/definitely travel on Thursday.
In Example 3.35 and 3.36, the Finite itself is changed from will (expressing
a high degree of determination) to something other, indicating levels of cer-
tainty, possibility, or obligation. In Example 3.37, similar modulation is
expressed outside the Finite, namely by a modal Adjunct (certainly or defi-
nitely) that is associated with the Finite. The Adjunct fulfils very similar
purposes to the modal Finite, but it does so in a different way – and there are
many of them. SFG provides the entire network of options in English, which
can be accessed in Halliday and Matthiessen (2014).
Second, consider the relation between function and form. Instead of using
a direct command in the imperative form, an English speaker is more likely to
express a wish by using modality, for instance a version of Example 3.35 or
3.36. In such a case, we might still understand the utterance to be some kind of
command or order, but it is expressed in a socially acceptable way: as
a declarative with appropriate modality, rather than in imperative form.
Although the imperative is socially less acceptable in English in many situa-
tions, it does occur in certain social contexts, such as in emergencies: HELP!!,
towards children: Sit down now and be quiet!, and in recipes: Heat oil in a pan.
Add the garlic . . . etc. This is an example where the choice of grammatical
patterns reflects social contexts in which they are appropriate.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.3 Functional Grammar 87

Following this brief introduction to the interpersonal level of analysis in

SFG, we will now turn to the analysis of how the world is represented – the
ideational level.

3.3.3 The Ideational Level of Analysis in Functional Grammar

The ideational level concerns how ideas are expressed in language. At this
level, perhaps a little surprisingly, the analyst looks first to the main verb. As we
have already noted above, verbs come with different kinds of expectations, so
to speak: they demand either one or more elements (called ‘participants’ in
SFG) to go with them. Importantly, these participants frequently need to be of
a particular nature. Consider our previous example, repeated here:
Example 3.38 Luke will travel on Thursday.
The main verb in this example is travel. If you think about travelling as an
activity, it is sufficient to have just one participant that does this activity: the
Actor, in this case Luke. Therefore, it is possible to end the sentence after the
word travel – the bit about Thursday is merely a circumstance; it is not essential
to the sentence. You can’t add another participant such as *Luke will travel
Joanne . . . If we want to include Joanne, we can either add her to the Actor bit,
as in Luke and Joanne will travel – or we can add her in a similar way to how we
have added the information about Thursday: in a prepositional phrase, as
a circumstance, as in Luke will travel with Joanne / to Joanne / without
Joanne . . . etc.
Many verbs are like travel in this respect, in that they require only an Actor:
jump, walk, read – there is no limit to the lexicon here. However, there are also
plenty of verbs that do not work that way. For instance, we can’t just say Mary
gives. To make a clause with a verb like give complete, we need as many as
three participants: an Actor (e.g. Mary), a Goal (the book), and a Recipient
(Paul).
Example 3.39 Mary gives Paul the book.
This is not to say that every clause needs to express all three participants
explicitly – languages are rather too flexible for that. Passive constructions are
excellent resources to ‘hide’ the Actor:
Example 3.40 Paul was given the book.
Nevertheless, the third participant is understood as somehow present,
although we don’t learn who (or what) it is in a passive construction like
Example 3.40. The verb give normally demands three participants, and

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 88 Resources

therefore the implication can be readily inferred that there must be a third
participant, even though it is not expressed.
Importantly, not all participants are persons – as we have seen by calling the
book a participant. The notion of participant is one of function, not form, and
therefore there can also be other kinds of participants. In fact, some verbs
regularly call for participants that are often as complex as a whole clause:
Example 3.41 Timothy said that we should go home immediately.
In Example 3.41, we can’t simply end the sentence after said – we would
immediately wonder what Timothy said. This is called the Verbiage. Verbs like
say expect a Verbiage as one of their participants, along with another partici-
pant called the Sayer (Timothy). Instead of that we should go home immedi-
ately we might, however, hear any of the following:
Example 3.42 Timothy said a word.
Example 3.43 Timothy said nothing.
Example 3.44 Timothy said more than anybody else in the room.
All of these different linguistic forms can serve the function of Verbiage. We
can’t do without a Verbiage with a verb like say, but we can express it in many
different forms.
Because verbs behave very differently with respect to the kinds of partici-
pants that they require, they are categorised into various different classes. Verbs
requiring an Actor (and potentially other participants) fall into the material
category, and verbs requiring a Sayer and Verbiage into the verbal one.
A further category that is particularly interesting from a CODA perspective is
the mental one, which comprises verbs that have to do with thinking and
feeling:
Example 3.45 Jean wondered about the question.
Example 3.46 The grandparents felt that the children were safe.
Verbs such as wonder and felt expect a Senser and a Phenomenon – someone
who senses, namely Jean in Example 3.45 and The grandparents in Example 3.46,
and something that is sensed, namely the question and that the children were safe,
respectively. Halliday and Matthiessen (2014) pointed out that while the range of
possible participants that can be Sensers is fairly restricted (only someone or
something with sensorial ability can sense!), the range of things that can serve
as Phenomenon is very wide – we can think and feel quite a lot of different kinds of
phenomena, and we can express them in many different grammatical ways.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.3 Functional Grammar 89

The category of relational processes concerns clauses that are constructed

around the verb be (and some related options), as in:
Example 3.47 The world is your oyster.
Example 3.48 The statue was extremely tall.
Example 3.47 relates the world to a different concept, your oyster – and thus
identifies the world as something that can be expressed in these terms also.
Example 3.48, in contrast, relates the statue to an attribute (tall). Again,
the second participant in these kinds of clauses can take on many different
forms, any of which serves the function of a participant in this kind of clause.
Altogether, Halliday and Matthiessen (2014) highlighted the importance of
verbs in the organisation of clauses. Suggesting that verbs are used to express
different kinds of ‘worlds’, they differentiated the world of abstract relations
(as expressed by relational processes) from the world of consciousness (think-
ing and feeling) and the physical world (acting and behaving). This kind of
fundamental distinction provides a useful resource for analysing verbalisations
of mental representations. Identifying types of verbs can be a very useful
starting point for identifying systematic patterns in the speakers’ thoughts,
looking at whether relations or physical actions are expressed, and to what
extent the speaker refers to thought processes and feelings directly.
The next section will conclude our brief introduction to SFG analysis levels
by looking at the third level, which concerns textual structure.

3.3.4 The Textual Level of Analysis in Functional Grammar

The third level of analysis in SFG concerns the organisation of text as text: for
example, what kinds of ideas are presented first or last, and how they are
structured in relation to each other. In SFG terminology, the Theme is the part
of a clause that concerns what the text is about, whereas the remainder (called
the Rheme) is used to say whatever there is to say about it. In English, the
Theme is represented as the first element in the clause. So in the previous
Example 3.47, the world is what we’re talking about, and what we’re saying
about it is that it is your oyster. Themes are typically connected in some way to
the context: there is generally a reason for why the Theme is chosen the way it
is, that is, why we choose to talk about this particular Theme. This becomes
particularly interesting when talking about Theme development across clauses.
For instance, consider the following short text:
Example 3.49 John went to the store. It was closed. Opening hours seemed to
have changed.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 90 Resources

The text starts with John, which is a reference to a person that we might or
might not know – in any case, it is clear right away that the text is about John.
Within the remainder of the first clause, in the Rheme, we find a reference to the
store. This is picked up as the Theme in the second clause – namely by the use
of the pronoun it. In the Rheme of the second clause, we find the information
that it was closed. This topic is picked up as the Theme in the third clause,
through the reference to Opening hours. Thus, we can see that, twice, the
Theme of a new clause relates to the Rheme of the previous clause.
Sometimes such a relation is established by pronoun reference (such as the
store – it), and sometimes by semantic links (such as closed – opening hours)
that the reader or listener can identify easily.
Going from Rheme to Theme, as in the previous example, is a relatively
typical thematic development in coherent texts. Another frequent option is to
simply go from Theme to Theme. In Example 3.50, the Theme John in the first
clause is picked up by the pronoun He in the two subsequent clauses.
Example 3.50 John went to the store. He felt tired today. He hadn’t slept
well . . .
Looking at thematic developments across clauses and establishing the relation-
ships between Themes helps to identify the ways in which speakers relate topics
to each other in discourse. This varies considerably across types of discourse, but
the general fact remains that how sentences are structured is important – it matters
how they begin or end. This is closely related, but not identical to the notion of
information structure discussed in Section 3.2.2. Information structure refers to
what is presented as given and what is new – and this is not quite the same as what
a clause is about. The given part of a sentence is very often in Theme position,
making the new part equivalent to the Rheme. But this need not be the case.
Especially in spoken language, speakers have other means of making the new part
prominent in discourse, namely by contrastive stress:
Example 3.51 My brother went to jail, not my sister!
To conclude, discourse shows systematic patterns and features on three
levels. A focus on the interpersonal level highlights how the relationship
between speaker (or writer) and hearer (or reader) is represented in the dis-
course. A focus on the ideational level reveals the representation of the (real or
imagined) world, and a focus on the textual level draws attention to the way
discourse is structured, related to the medium in which it occurs.
At all three levels, analysis becomes most revealing when looking beyond
the elements and features of a single clause, towards the patterns and structures
of entire texts and instances of discourse. This is, in fact, what SFG is designed
and used for: the interpretation of text beyond single clause structures, in light

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 3.3 Functional Grammar 91

of the various functions fulfilled by linguistic choices. We have seen this most
clearly at the textual level, because thematic development is best understood
across clauses. However, discourse analysis based on SFG works similarly at
the other levels as well. Looking at the interpersonal level is most interesting
when tracking in detail throughout a text how speakers put ideas across to
recipients, using various forms of modality. And looking at the ideational level
can be highly revealing when identifying patterns of verbs and participants,
which may vary at different places in a text.

Throughout this chapter, we have identified a range of linguistic phenomena

that cross over theoretical approaches in cognitive linguistics, discourse ana-
lysis, and functional grammar, all of which are relevant for analysing the
language speakers use to express their thoughts. The fact that similar notions
and insights appear as part of different approaches underscores the idea that
CODA, as such, is theory neutral. Indeed, some of the notions or tenets in
different approaches (such as functional grammar and cognitive linguistics)
may be felt as incompatible, or at least not easily reconcilable. Rather than
aiming to combine or contrastively review theories, my goal in this chapter was
to point to relevant insights within and across theories that can be drawn upon
to understand speakers’ use of language in collected data in relation to
cognition.
In this vein, linguistic resources of any kind can be helpful for identifying the
significance of linguistic choices. It may be more important to recognise that
linguistic choices are significant – relative to the situation of language produc-
tion in light of the overall network of options – than to subscribe to a specific
theoretical approach in linguistics.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 10:22:53, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 4 Identifying Cognitive Orientation

Now that we have covered the general resources and theories underlying
CODA, it is time to get more specific: what kinds of conceptual aspects can
be explored using CODA? Let me stress again that there is no actual limit here.
The language we use is rich with phenomena and conceptually interesting
structures, many of which may yet be detected in the future. Some insights
stand out as points of connection between linguistics and cognitive science –
where linguistic research highlights conceptual aspects that the linguistic
system makes prominent, precisely in areas that are frequently targeted in
cognitive science research. The next four chapters will explore a selection of
these. Chapters 4 and 5 deal with conceptual aspects that are represented in
every instance of language use, and Chapters 6 and 7 concern further phenom-
ena of relevance for many settings targeted in linguistic and cognitive science
research. Each phenomenon will be discussed along with an example from
CODA research.
In this chapter, we will look at two conceptual aspects that both pertain to
cognitive orientation – the way we perceive the world and think about it:
attention (Section 4.1) and perspective (Section 4.2). Attention concerns
what we orient to, in terms of cognitive focus, and perspective concerns how
we orient to it, that is, from which point of view.

4.1 Attention
The role of attention for cognitive processes can hardly be overestimated. We
can’t think consciously without paying attention – this would be a contradiction
in terms. Conscious thinking about something means (among other things) that
we are attending to it. Therefore, knowing where someone’s attention is
focused at any given time is central to understanding the cognitive processes
involved in a given task. This is true for any type of cognitive activity: attention
is just as central for viewing a picture as it is for solving a tricky problem. In
both cases, we might pay attention to some things but completely miss others.
In addition to our conscious attention focus, thinking involves some aspects
that we aren’t aware of, and aren’t paying attention to. Our subconscious might

92

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.1 Attention 93

guide our attention in ways we don’t notice. It might lead us to sudden insights
or solutions to a problem that we have been trying hard to solve (e.g. Schooler,
Ohlsson, & Brooks, 1993). We might be unaware of something we see from the
corner of our eye. All these things are possible, and they are exciting areas in
cognitive science research. However, the things that we are paying attention to
are the things that are most prominent in our minds. Much research in cognitive
science therefore aims at gaining insights about this central factor (see, e.g.,
Anderson, 2009), and about the interplay of conscious attention focus with
other, less prominent processes that guide thinking.
Not surprisingly, attention likewise guides the use of language. Focus of
attention constrains what is said, as well as how it is said. In a sense, everything
we say reveals where attention is currently focused. Remember when you were
a child, at school, and were daydreaming – just not paying attention to what the
teacher said? You wouldn’t have been able to answer a direct question, leading to
embarrassing silence or making something up instead of a meaningful response.
And perhaps you have been caught asking a question concerning something that
you should already know – had you only been paying attention . . .
Spouses tend to know exactly when their partner pays attention to what
they’re saying and when they don’t; they can identify the little signs in the
loved one’s reactions that reveal whether they’re actually with them – or rather
far away in their thoughts. And sometimes there is something in a formulation
that strikes you as odd, as not quite in sync with what you’ve just said. Perhaps
you couldn’t quite put your finger on it, but you do get the feeling that your
conversation partner isn’t entirely on the same page. And this may well be the
case because they weren’t paying attention – or because their focus of attention
was on something slightly different. Your attention focus differs from theirs,
and this directly affects communication, or how well you feel understood.
Based on these everyday experiences, we all have intuitions about the close
relationships between attention, thought, and language use. Such intuitions can
be important when doing CODA, as they can provide a suitable starting point
for deeper scrutiny. Doing analysis may mean turning intuitive observations
into something far more systematic, something that can be used as evidence for
a particular phenomenon that the analyst is pursuing. In the following, we will
explore the conceptual factor of attention in more detail first from a cognitive
science perspective (Section 4.1.1), second from a linguistic one (4.1.2), and
then look at connections between the two in previous research (4.1.3). An
example from CODA research will conclude this section (4.1.4).

4.1.1 Attention from a Cognitive Science Perspective

Nearly half a century ago, Neisser and Becklen (1975) conducted an ingenious
experiment in which participants were shown short video films that were

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 94 Identifying Cognitive Orientation

overlaid visually, such that they were watching two films at once. They were
asked to pay attention to only one of these two films: they had to react to certain
things that happened in it, whereas the other one could be ignored, as it was not
needed for the task. It turned out that the participants were incredibly good at
doing this. Their performance in reacting to the films dropped only very slightly
in comparison to a condition in which they were only watching a single film.
Moreover, they barely noticed what was going on in the other film. The
experimenters had inserted some unexpected scenes that would have been
noticed by anyone who paid attention to the films; however, participants did
not recall any of those if they were asked to attend to the other one of the films.
This experiment powerfully demonstrates how good we are at consciously
directing our attention to whatever is relevant in a given situation. But it also
demonstrates how we can totally fail to notice some things if we happen to be
distracted.
In common parlance, we recognise this phenomenon as selective attention –
and we are familiar with it from a range of everyday situations. In a room full of
people talking, such as a party, we can focus our attention on the conversation
we are currently involved in. It might be challenging and tiring, but it does
work. If we catch something from another conversation nearby, our attention
might switch – and as a direct effect we will miss something from the con-
versation that we were supposed to be following. This is essentially the same
effect as the visual one demonstrated by Neisser and Becklen (1975), only in
the listening domain. If we make an effort to pay attention to certain aspects, we
perceive relevant things much more clearly and intensively than if we don’t
(Kahneman, 1973). There are certainly individual differences; some people get
more easily distracted than others, and physical impairments such as hearing
difficulties might make it much more difficult to focus. On the whole, though,
we know intuitively that we are capable of directing attention to certain things,
at the expense of other aspects of a situation. We cannot attend, at the same
level, to everything at once – this is clearly impossible.
Likewise, we know that it is very difficult to remember things that we
did not pay attention to. It has long been known that memory is directly
affected by attention focus (e.g. Moray, 1959) – and this opens up inter-
esting questions about the exact nature of the dependency between the two.
For instance, although attention guides what we perceive, there are always
aspects that do not quite reach consciousness but that are nevertheless
processed to some extent (Corteen & Wood, 1972). If that’s the case, to
what extent can we learn (and then perhaps subconsciously remember)
things that we don’t actually pay attention to? Nissen and Bullemer (1987)
observed that awareness during learning need not be entirely the same as
paying attention during learning. And indeed, in a series of experiments
they found that in order to learn, one need not always be aware of certain

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.1 Attention 95

aspects. Some information may just be carried by a task, and people don’t
need to notice it in order to learn it. However, they do need to pay
attention to the task as a whole, in general. Thus, it appears that attention
is central to cognition in many respects, and while it is intricately related
to conscious awareness it is not identical to it.
Another way in which attention plays a role in relation to cognitive task
performance concerns the opposite effect of learning: the better we get at a task,
the less attention we need to pay to it in order to get it done (Norman & Shallice,
1980). You might remember how much attention you initially had to pay to every
single movement or action when you learned how to drive (or to ride a bicycle, or
any other acquired skill). The more experience you get, the less you need to think
about it. Your responses to different traffic situations become more and more
automatic, and you may often not be aware of what you are doing, especially if the
situation is not particularly complex. This enables you to engage in deep con-
versations, or do other things while driving. Naturally, this also involves a danger
(Canas et al., 2003): since you’re not really paying attention to the current task of
driving, some things may escape you. And this might lead to tricky situations,
where experienced drivers can make errors that learners, whose attention is fully
activated anyway, may not necessarily do. It is therefore important to be flexible
enough to switch back to enhanced attentional control, out of the ‘autopilot’ mode
that enables experienced people to do a task without paying much attention to it.
So, from a cognitive science perspective, we can see that attention is one of the
most central factors guiding cognitive processes, and it never ceases to be of
interest because of its complex relationships to other factors such as awareness and
automated processes. Attention also guides performance in cognitively challen-
ging tasks, because solutions sometimes can’t be found if the problem solver’s
attention focuses on something else. And it provides a bottleneck for simulta-
neously executing certain motor actions – we all know how difficult it is to do one
type of motion with one hand and a different one with the other. It can be learned,
by paying attention to both motions sequentially and automatising the processes –
but the bottleneck is there. There are many further exciting findings in this area.
Anderson (2009), for instance, provides a useful review of earlier literature, and
more can be discovered by reading up on more recent developments in cognitive
science.

4.1.2 Attention from a Linguistic Perspective

If attention is central to our thinking, we can expect it to have systematic
reflection in language. Language, after all, is the prime way to express our
thinking. And indeed: language never represents facts or perceptions neutrally,
but rather serves to spotlight and privilege certain aspects while neglecting
others. Within cognitive linguistics, all of the main grammatical approaches

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 96 Identifying Cognitive Orientation

highlight in some way how attention focus is represented in language. We have

already considered some of them in Chapter 3. Now that we’re looking directly
at the cognitive factor of attention, it is easy to see how linguistic structures
systematically account for the fact that some things are more prominent, more
in focus, than others. For instance, Langacker (1986) shows in numerous
examples how language can be used to profile specific aspects of a scene
while backgrounding others. Even small and seemingly negligible changes in
grammatical structures can be significant in this respect. Consider the differ-
ence between the following sentences (Langacker, 1986:15):
Example 4.1 I sent a walrus to Antarctica.
Example 4.2 ? I sent Antarctica a walrus.
Example 4.3 I sent the zoo a walrus.
Something is odd about Example 4.2, as Langacker suggests (indicated by
the question mark). In some grammatical approaches that are not cognitively
motivated, Examples 4.1 and 4.2 are regarded as entirely equivalent, and
therefore they should not differ with respect to acceptability. Also, grammati-
cally, both Antarctica and the zoo are noun phrases – so there’s no grammatical
reason for Example 4.3 to be better than Example 4.2.
Langacker’s explanation for this puzzle is that there is a slight difference in
profiling, which is illustrated in Figure 4.1. Using the preposition to profiles
a trajectory from an unknown source to the destination indicated in the sen-
tence – to Antarctica. This profiles the path, much more than Examples 4.2 and
4.3 do. Without the preposition, the recipient of the transfer is profiled:
Antarctica in Example 4.2, the zoo in Example 4.3. Now, it intuitively makes
sense to donate a walrus to a zoo, and this justifies directing attention (linguis-
tically) to the recipient. In contrast, there is something odd about donating
anything at all to a continent. The focus on the recipient therefore does not seem
to make sense. Shifting focus to the trajectory of the transfer (ever so slightly,
by using the preposition to as in Example 4.1) makes the sentence acceptable.

I A I A

W W W
W

I sent a walrus to Antarctica. I sent Antarctica a walrus.

Figure 4.1 The different profiles of semantically similar sentences (following

Langacker, 1986).

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.1 Attention 97

Although the notion of profiling in cognitive grammar clearly accounts for

the different ways in which attention can be focused and expressed in
language, Langacker (1986) did not himself discuss the relation between
profiling and attention as a cognitive phenomenon. Langacker’s approach
can nevertheless be useful for explaining specific linguistic choices in terms
of cognitive focus, in cases where systematic differences in profiling choices
are found across sets of language data, related to different kinds of cognitive
tasks or task aspects.
As indicated in Chapter 3, Talmy (2000) offered a different approach to
explaining linguistic structure from a cognitive perspective, with a stronger
focus on semantics. Talmy’s distinction between Figure and Ground in
linguistic constructions is entirely compatible with Langacker’s notion of
profiling and the associated notions of Trajector and Landmark. This
explains differences such as those between the bike next to the church
and the church next to the bike. Some entities are more likely to be profiled
as the item to be localised – such as the bike – than others. A church is not
movable and normally does not have many competing items looking
similar in the nearby environment; therefore, it is fairly difficult to think
of a context where it would be natural to identify the church’s location as
being next to the bike.
It is probably safe to say that every single grammatical distinction somehow
reflects a difference in conceptual representation – and thus, a slight difference in
just how attention is allocated. Taking these ideas further, Talmy (2007) offered
a comprehensive account of more than fifty ways in which linguistic structure
highlights attention. Each of these factors introduces a difference in linguistic
structure that draws attention to some aspect while decreasing attention to
another. For instance, even though aunt means parent’s sister, both parent and
sister are much more in focus when saying parent’s sister than when simply
saying aunt. This is because the word aunt conflates the notions of parent and
sister, leaving them implicit – and this is different to actually mentioning these
terms directly. Similarly, consider the following ways of expressing the fact that
some event happens in the past or in the future (Talmy, 2007):
Example 4.4 a. When he arrived, . . . b. When he arrives/will arrive, . . .
Example 4.5 a. On his previous arrival, . . . b. On his upcoming arrival, . . .
On a semantic level, there does not seem to be much difference between
these two examples: both of them express an arriving event at a time located
before or after the time of speaking. However, while Example 4.4 expresses the
notion of past or future simply by using grammatical morphemes (arrive-d,
arrive-s, will arrive), Example 4.5 uses content adjectives (previous/upcoming)
to do the same. In addition, the action of arriving is expressed by a verb in

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 98 Identifying Cognitive Orientation

Example 4.4, but by a noun in Example 4.5. These two factors work together to
create the overall impression of a much stronger focus on the event in the future
or in the past in Example 4.5.
To illustrate another factor, consider the choice of grammatical roles in the
following examples, which are also arguably semantically equivalent:
Example 4.6 The landlord rented the apartment to the tenant.
Example 4.7 The tenant rented the apartment from the landlord.
Although renting could be justifiably regarded as a balanced act, with
complementary actions carried out by the people involved, the two examples
clearly guide attention to the part played by the landlord in Example 4.6, but to
the part played by the tenant in Example 4.7. In each case, the grammatical
subject (landlord vs. tenant) is perceived to be more active, more in charge,
more accountable for the decisions made.
Further factors include the following, just to cite some inspiring examples
from Talmy (2007):
• Lexical choices that highlight different aspects of related terms: compare the
different associations of dirt, soil, earth. All of these can occur almost
interchangeably in some contexts but not others.
• Order of elements in a sentence, similar to thematic structure in SFG
(Halliday & Matthiessen, 2014), as discussed in Chapter 3.
• Linguistic ways of drawing attention to terminology, as in this gadget is
called a ‘pie segmenter’.
• Linguistic ways of drawing attention to the delivery style of an utterance, as
in So then I’m like: Wow, I don’t believe this! – uttered by a young person
while telling a story.
And many more. Obviously we cannot and need not reiterate all of the many
ways in which linguistic features reflect differences in attention focus identi-
fied by Talmy (2007). For current purposes, it suffices to note that language is
extremely well equipped and powerful in expressing attention focus and
guiding an interaction partner’s attention to certain aspects of a situation
(including language itself) rather than others.
With respect to CODA, it can be wise (in the absence of more direct predictors
from previous literature) to think in the opposite direction: starting from linguistic
data collected in an experimental study, rather than from the differences available
in the linguistic system. If one data set primarily represents a certain state of affairs
in one way, and another data set (perhaps in a different condition, or a different
time throughout an experimental procedure) favours a different way, then this may
very well have to do with distinctions contributed by the attentional system of
language. Talmy’s (2007) work, among others, may then serve as a valuable
resource to provide a theoretical explanation for the differences found in the data.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.1 Attention 99

Altogether, we can conclude that language offers a wide range of ways in

which attention focus can be reflected and communicated. As users of language
we normally just accept these features in the language we use and encounter,
without giving it much thought. As analysts, it can be highly revealing to
examine such features systematically in a linguistic data set.

4.1.3 Attention: Links between Language and Cognition

Attention is central to cognition, as we have seen – and language has multiple
ways of reflecting attention shifts. But how much evidence is there to show that
language and cognition are actually linked through the attentional system?
Various directions of research indicate such a link, and we will look at three
examples here: eye movements, child language acquisition, and discourse
structure. Although we won’t have space for more, this list could be expanded
in many ways, showing the intricate relationship between attention and lan-
guage use.
When we pay attention to something, we are also typically looking at it. If
you want to direct a partner’s attention to something you see, you might say,
Look at this! – and point in the relevant direction. This concrete experience also
extends into non-visible aspects, as when we say things like Look, what I mean
is . . . in a more figurative context. And when parents have something very
important to say to their children, they will often ask them to Look at me! –
hoping to make them pay close attention to what they have to say.
So, looking is closely related to paying attention, and this has been utilised
for research on attention in cognitive science. Unlike attention itself, eye
movements can be observed, recorded using eye-tracking devices, and mea-
sured; therefore, they serve as an external representation of the internal phe-
nomenon of attention. In this respect, eye movements are like language:
utterances can also be recorded and analysed in detail. Once we have these
two measures that both relate to cognition, we can relate them to each other to
identify patterns of attention focus.
In an inspiring collection of articles that report research of this kind,
Henderson and Ferreira (2004) demonstrate various ways in which eye move-
ments are systematically related to language processing. For instance, attention
processes guide reading. We don’t read individual letters or even words, but our
eyes jump across chunks of text, comprehending meanings rather than visual
shapes as we go. Similar findings can be noted for comprehending the real
world: as we look at scenes, we process them on the basis of meaningful
concepts and relevance, rather than by taking in a lot of details. Indeed, the
details may easily get lost as we understand scenes based on what we see.
Sometimes we completely miss unexpected aspects if they are not connected
to anything that is relevant to our current situation. This in turn is reflected in

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 100 Identifying Cognitive Orientation

the way humans describe scenes through language. What we look at, or have
just looked at, gets represented in language. Moreover, the order in which we
look at things in a scene is systematically related to the order in which they
appear in language. Attention focus, as measured by eye movements, also
guides comprehension and processing of language – and sometimes we can
even predict what people will say, or how they will interpret an utterance, by
observing what they look at.
Looking at something and talking about it starts early in life. Carers fre-
quently name objects when speaking to young children, typically accompanied
by a pointing gesture: Look at this shiny red train, it’s nice isn’t it? Expanding
on this everyday experience, Tomasello (2003) claimed that child language
acquisition, on a very general scale, is guided by joint attention. Children learn
language precisely because of their ability to direct their attention to exactly the
same thing as their carers do. After all, the world is full of objects and sounds.
Without joint attention, how would children be able to understand that the
sound of saying train has anything to do with a particular shape of object, out of
the many objects visible at the time? The early experience of following a carer’s
attention gradually expands to being able to understand communicative inten-
tions, to follow a train of thought, and to derive meanings from abstract
contexts – and along with this, vocabulary and conceptual understanding
grows.
As we can see from looking at child language acquisition, joint attention is
a central factor for communication: understanding each other’s intentions
through language works because we can guide each other’s attention focus
through what we say, and how we say it. Some theories of language and
communication use this idea as a fundamental element in spelling out how
discourse works. For instance, Grosz and Sidner (1986) introduced
a computational theory that models features of discourse structure, capturing
elements such as conjunctions, referring expressions, and the like that organise
the discourse. According to this theory, discourse is constrained by attention in
decisive ways.
For instance, agreeing on a specific referent in dialogue works because the
interactants’ attention is jointly focused on this particular referent; this enables
the speakers to use simple terms like pronouns to refer back to them. This is not
entirely foolproof: pronouns like he, she, it, and they can be ambiguous in some
contexts – particularly if the speakers’ attention is not entirely aligned, which
means that they may be thinking of a different referent. Thus, pronouns are
a frequent source of misunderstandings.
Furthermore, cue phrases such as conjunctions (but, and, first, moreover,
etc.) serve to guide attention throughout the structure of verbal interaction. For
instance, if you hear the phrase but anyway, . . . you can assume that the
attention focus in the discourse has somehow drifted away from the original

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.1 Attention 101

train of thought, and the cue phrase helps to bring it back. This normally works
even though the phrase is very unspecific: it doesn’t provide any details as to
where exactly attention should be focused instead. The discourse partner will
need to listen to the remainder of the utterance and make the connection in their
mind, in order to link back to the previous focus of attention.
Altogether, it is safe to say that attention and language are clearly interlinked.
This is supported by research on eye movements. In child language acquisition
just as well as in everyday language use, joint attention is needed to achieve
communication successfully. Specific cues in language can signal clashes of
attention in dialogue.

4.1.4 Attention: A CODA Example

Since attention is so central to language use and clearly linked to cognitive
phenomena, this aspect of cognition is a natural target in CODA projects.
Independent of what kind of task is given to participants, the focus of their
attention throughout the task will determine what they say and how they
formulate it. If you describe a picture, you will focus on the kinds of things
that you pay attention to – it will be almost impossible not to notice where your
focus lies when considering how you describe it. Likewise, if you think aloud
while solving a problem, your words will reveal your current attention focus in
various ways. To illustrate this idea, we will now look at one example from
CODA research in more detail.
Imagine you’re given a famous picture by a classic artist, and are asked to
describe it. What would you pay attention to, what would you focus on?
Perhaps your description will reveal something about the extent to which
you’re interested in paintings of this kind, and perhaps you will linger on
aesthetic aspects or those that you find particularly convincing or impressive.
Now, what would be different, do you think, if you were a professional painter
rather than a layperson admiring a famous picture? As such, describing
a picture is not a particularly demanding task – you don’t need years of
professional training to do that. However, the way in which you do it may
very well depend on who you are, and how you think about paintings. Expertise
may guide attention, and thereby influence the way in which descriptions are
presented in language.
In Cialone, Tenbrink, and Spiers (2018), we took this train of thought a step
further and hypothesised that professionals with some kind of spatial expert
background would look at pictures showing spatial environments in a different
way to how other people do. This idea is related to the Whorfian theory (see
Chapter 1), which suggests that societal and cultural background determines
structures in language, reflecting structures of thought. Most research in the
Whorfian tradition addresses differences in cultural thinking as evidenced in

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 102 Identifying Cognitive Orientation

the differences between entire languages. Following a slightly different tack, in

Cialone et al. (2018) we set out to test whether thinking, as reflected by
language use, may be influenced by particular aspects of a speaker’s profes-
sional background.
We designed a study with three groups of people who had been working in
a spatial profession for at least eight years – sixteen painters, sixteen sculptors,
and sixteen architects – plus a control group containing sixteen people with
a similar range of experience in non-spatial professions. All participants in the
study were interviewed individually, and they were given as much time to
respond to the questions as they liked. We gave them three different pictures to
look at: a Google Street View screen shot, a famous painting, and a surreal
landscape produced by a digital artist. For each picture, participants were asked
the same set of three questions:
Question 1 ‘Could you please describe the environment that you see in this
picture?’
Question 2 ‘How would you explore the space in this image, where would
you go?’
Question 3 ‘If you were given the chance, how would you change the envir-
onment in this image?’
These questions were intended to encourage the participants to look at the
pictures in some depth, and to think about them in various different ways, none
of which required any particular professional expertise. And indeed, the
answers did not directly represent professional training – none of the partici-
pants tried to educate the experimenter by explaining expert facts about paint-
ings or architecture or anything of that sort. And none of the ‘non-spatial’
participants had difficulty dealing with the questions. But nevertheless, there
were striking differences between the groups: they were indeed looking at the
pictures in different ways, depending on their particular expertise. They paid
attention to different things, thought about them differently, used different
concepts.
Most prominently, it was clear to see that painters, sculptors, and
architects produced more elaborate descriptions than the group without
any particular spatial expert background. Their descriptions revealed
more attention to detail, more willingness to focus on this task for
a longer period of time. But then there were also differences between the
groups of professionals, revealed by a close CODA analysis that involved
the identification of linguistic markers for various conceptual categories
throughout the collected language data.
Results of this analysis showed that painters were more likely to describe the
depicted space as a two-dimensional (2D) image; they said things like ‘It’s
obvious the image wants you to follow the boat off onto the horizon.’ More than
others, painters switched between describing the scene as a three-dimensional

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 103

(3D) space and as a 2D image. This reveals that they frequently shifted attention
between the picture as a picture, and the environment depicted within it. By
contrast, architects focused their attention more on barriers and boundaries of
space, and this was revealed in utterances such as ‘There are voids within walls
which become spaces in their own right.’ Sculptors’ responses were between the
two; they were somewhat like architects except for one measure: with respect to
describing bounded space, they appeared more like painters.
Altogether, our study showed that even within the same culture, people of
different professions differ in how they appreciate the world – in this case, how
they focus their attention when looking at pictures that show environmental
scenes. The findings also raise the possibility that people who are already
inclined to see the world as a 2D image, or who focus on the borders of
a space, may be more inclined to pursue painting or architecture. This also
makes sense – perhaps we develop our thinking in a particular way, for
whatever reasons, and this paves our way towards a particular profession.

4.2 Perspective
As human beings, we live in the ‘here and now’ – we perceive the world from
a certain vantage point in space and time. However, we can adopt a different
point of view in our minds. For instance, we can imagine seeing the world from
the perspective of our interaction partner, or any other position that we can
think of. This is also true in terms of concepts, in various ways (Schober, 1998).
Perspective taking means considering different worldviews, beliefs, concepts,
physical viewpoints, and more. Our ability to do this means that we can take
various perspectives into account when making judgements, or when talking
about complex facts or ideas.
Like the speaker’s focus of attention, the underlying conceptual perspective
is represented systematically in linguistic structure, and it can be highlighted by
the analyst in relation to the discourse and task context. In this section, we will
start by looking at spatial perspectives and how people choose and switch
between them (Section 4.2.1), followed by discussion of a CODA study that
addresses this (4.2.2). Section 4.2.3 discusses other types of perspective taking,
and this will again be followed by an example study that involves various
conceptual perspectives (4.2.4).

4.2.1 Spatial Perspectives

We perceive the world from wherever we are, through our own eyes, or ears, or
other sensory organs that might play a role. This simple fact has a direct effect
on the ways in which we use language. Clearly, talking about the world is
easiest when we can just follow our own perceptions. If I see a dog in front of

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 104 Identifying Cognitive Orientation

me, I can simply say, There is a dog in front of me. If I happen to be walking
along with a friend, this just as simply translates to There is a dog in front of us.
I can include my friend in this spatial description, relating the dog’s position to
both of us, as I am aware that we are both roughly sharing the same perception,
or perspective, relative to the dog’s location.
However, things become a bit more complicated if my friend is not walking
beside me, but has a different perspective on the same scene. If my friend is
coming towards me, still further away than the dog, I might want to switch from
the use of in front of to a different spatial term, and might perhaps say, There is
a dog between you and me. But if they’re located at a certain angle, this won’t
work: the term between requires a more or less straight line connecting two
locations, with the described object somewhere in the middle. In such a situa-
tion, I will have to choose: do I use my own perspective, or my friend’s? Would
I say that the dog is in front of me – or that the dog is in front of you?
Incidentally, I do not only choose between perspectives, but also relate the
dog to a different person in each case: either to myself, or to my friend. In
technical terminology, what I choose is a relatum. In our example so far, the
relatum (me, or my friend) has their own view of the scene, which provides
a certain kind of direction (a view direction). Levinson (2003) calls this kind of
situation an intrinsic reference system. Any entity (object, person, animal, etc.)
with some kind of intrinsic direction can be a relatum in an intrinsic reference
system, as in Examples 4.8–4.11. This doesn’t always have to be based on
perception, as we can see in Examples 4.10 and 4.11:
Example 4.8 The dog is in front of me (from my point of view).
Example 4.9 The dog is in front of you (from your point of view).
Example 4.10 The dog is in front of the car (i.e. in front of the headlights and
direction of movement).
Example 4.11 The dog is in front of the cupboard (in front of the doors).
Example 4.12 ? The dog is in front of the ball.
Example 4.12 is trickier: the ball doesn’t have an intrinsic front, as such.
However, it is possible to imagine situations where the sentence makes sense.
For instance, if the ball is moving, it establishes a kind of directionality that
might provide some kind of perspective (Tenbrink, 2011):
Example 4.13 The dog is in front of the ball that is rolling down the hill.
Also, it might be possible to use the dog’s perspective, in this particular
example, to make sense of the scene:
Example 4.14 The dog is in front of the ball, examining it.
However, such an interpretation seems to be only possible with in front, not
with other terms like behind or to the left/right of:

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 105

Example 4.15 ? The dog is behind the ball, touching it with its tail.
Clearly, intrinsic reference frames aren’t always available. If the relatum
doesn’t have an intrinsic perspective, it is difficult to identify a front, left, right,
or back direction without any further clue. However, there are other options.
Levinson (2003) shows that there is another very common type of reference
frame, which he calls relative. With a relative reference frame, the perspective
is not associated with the relatum, but rather with another entity, called the
origin. This means that there are three entities involved in the spatial config-
uration: the locatum (object to be located), the relatum, and the origin. With this
option, we can say things like:
Example 4.16 The dog is behind the ball from my perspective.
Example 4.17 The box is to the right of the table as you walk into the room.
In Example 4.16, the origin is the speaker, and the perspective is explicitly
associated with them. In Example 4.17, the origin is the person walking into the
room; they will then use their own perspective to see where the box is, related to
the table. There may be further options and interpretations, as evidenced by
countless everyday misunderstandings even in cases of fully specified spatial
descriptions.
As we can see from these examples, humans have a very flexible range of
options for thinking and talking about spatial relationships. We can use our own
point of view, but we don’t need to do so. It is not too difficult to consider
someone else’s perspective, or imagine looking at the scene from a different
vantage point.
When do we use which perspective, what aspects influence our preferences
when we talk about spatial relationships? A lot of research over the past few
decades has addressed this question. Getting a clear idea about systematic
patterns of usage is, however, hampered by the fact that the literature has
used a wide range of different terminologies to describe this same phenomenon
(Levinson, 2003). In psychology, for instance, it is far more common to talk
about egocentric as opposed to allocentric reference frames.
As Levinson (2003) demonstrated, this dichotomy can be translated to some
extent to his classification of intrinsic and relative reference frames (plus
absolute ones, as when we use compass directions). However, it is
a complicated translation because the reference frame classifications don’t
allow for a direct match, based on the very different definitions they use.
Using an egocentric reference frame in the egocentric–allocentric dichotomy
means using someone’s own point of view, both for intrinsic and for relative
reference frames. That is, the person in question would either relate an object to
themselves (which would be an intrinsic reference frame in Levinson’s classi-
fication), or to something else from their own point of view (which would mean

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 106 Identifying Cognitive Orientation

a relative reference frame). In contrast, any perspective other than one’s own
would be allocentric.
Egocentric reference frames are easier and more intuitive, as they corre-
spond to our own point of view. Some people seem to be readier than others to
rely on allocentric reference frames, that is, to locate things independent of
their own point of view. This seems to be related to different abilities in sense of
direction, at least in navigational contexts (Wen, Ishikawa, & Sato, 2013).
People with a good sense of direction find it easier to understand where things
are and how the various spatial locations relate to each other, whereas people
with a poor sense of direction can struggle to get a clear idea of the overall
environment structure independent of their own location (Sholl, 1988).
In navigational contexts, relying on your own perspective facilitates the
acquisition of route knowledge: you know how to get to a certain place because
you have walked that route several times, and you can remember what the
various streets, buildings, and intersections looked like and where you had to
turn. In contrast to this, survey knowledge is more flexible: you know more
generally how places are related to each other, and would be able to find them
without necessarily following the same route every time. These two different
types of knowledge (Siegel & White, 1975) are related to two different types of
perspective as expressed in language, called route and survey perspective
(Examples 4.18 and 4.19 taken from Taylor & Tversky, 1996:379):
Example 4.18 As you come in, turn right. To your right will be the ‘personal
computers’ room. Continue until you’re forced to make a left.
The ‘Stereo components’ room will be in front of you as you
turn left.
Example 4.19 North of town are the White Mtns. and east of town is the White
River, which flows south from the White Mtns. The main road
by town runs in the east-west directions and crosses the White
River. The stables are on the south side of this road.
These two ways of describing the memory of a spatial scene clearly demon-
strate the different types of knowledge related to different perspectives: route
knowledge is based on someone’s own (egocentric) perspective and serves as
the basis for describing a tour through the memorised environment, as in
Example 4.18. People who adopted a route perspective in Taylor and
Tversky’s study often used active verbs such as enter, walk, head, find, turn,
go, and continue, and descriptions that related locations to the viewer, using
terms like left, right, front, and back. In contrast, survey knowledge is not based
on a particular perspective, but relies on compass terms and static terms (such
as forms of be) to describe how locations are related to each other, as in
Example 4.19. People have different skills and preferences for acquiring and

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 107

using each of these, and they also often mix perspectives (Taylor & Tversky,
1996).
What people will do in any specific situation, what perspective they will use,
is not entirely determined by skills or random preferences – it also depends on
context factors of various kinds. In Taylor and Tversky’s (1996) study, speakers
preferred particular types of perspectives for different types of environment.
Some environments apparently lend themselves well to an imagined tour,
whereas others may be more easily described by using a survey perspective,
as if seeing the scene from above. This corresponds to findings that speakers
may sometimes choose a perspective that allows them to use a relatively simple
spatial description over another that would be more complex (Tenbrink, 2007).
As the literature shows, the language that people use to express reference
frames depends on the scenario. Whereas Taylor and Tversky (1996) identified
linguistic markers for survey and route perspectives, other types of reference
system may be reflected in discourse in different ways. Utterances may also be
genuinely ambiguous, as in the ball to the right of the table: if the perspective is
not specified, it is not clear where the ball is. However, an observer hearing the
utterance and seeing the scene will normally be able to identify the intended
perspective from context. Also, many people will interpret a sentence like the
ball is to your right as using the addressee’s perspective, whereas the ball is to
the right of you might be relying on the speaker’s perspective (Olloqui
Redondo, Tenbrink, & Foltz, 2019).
Another important factor for choice of perspective concerns the interaction
partner. Tversky and Hard (2009) found that the mere presence of a person
induced many people to take that person’s perspective, especially when action
was involved and the person was ready to act. Speakers are also known to align
with interaction partners in dialogue; they agree on a particular way of describ-
ing a scene (Clark & Wilkes-Gibbs, 1986). This can either mean sticking to the
same perspective throughout, that is, either one of the interaction partners’
perspectives, or it could mean that each speaker always uses their own per-
spective, or that each speaker always uses the partner’s perspective. All of these
options can be implicitly agreed in dialogue, facilitating interpretation. Schober
(1995) found that such decisions are sensitive to task features, in particular the
difficulty of adopting a different perspective. Under certain circumstances,
speakers choose to avoid using perspective-dependent spatial terms altogether.
Thus, again, ease of reference seems to play an important role for the choice of
perspective in spatial description.
In sum, perspective plays a major role when we talk about spatial relation-
ships. Most descriptions of space need to be understood from a particular kind
of perspective. This is not always clear, as perspective is often not mentioned
explicitly, and it may not be shared between interaction partners. Also, the
language used to represent underlying perspectives can vary across scenarios.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 108 Identifying Cognitive Orientation

Identifying patterns of speakers’ perspectives in discourse reveals an important

aspect of how they conceive of the spatial situation in question.

4.2.2 Spatial Perspectives: A CODA Example

As we have seen, we use our own as well as others’ perspectives all the time.
Some people are more skilled than others in using perspectives other than their
own. For instance, for some of us it is quite easy to think of street networks in
terms of a map, knowing how the roads are related to each other as if from
above, using an allocentric survey-based reference system. Researchers in
cognitive psychology have been interested in preferences for egocentric or
allocentric reference systems for a long time; one major goal is to find indica-
tors towards classifying people according to whether they intuitively adopt an
allocentric or egocentric perspective when learning environments.
One mechanism in particular had been suggested as a possible instrument for
such a classification (Gramann et al., 2005; Schönebeck, Thanhäuser, & Debus,
2001): the so-called tunnel task. This task involves a short travel through
a simple virtual tunnel. The tunnel has almost no features – there are no
landmarks or road signs to orient to, no traffic to distract from the journey.
The only change that the observer can perceive is that the tunnel curves, once or
twice, until it ends. The task given to experimental participants is then to
indicate where the entrance was by choosing an arrow (see Figure 4.2).
According to Gramann et al. (2005), there are always two possible solutions to
this task, one of which reflects an egocentric reference system, and the other an
allocentric one. However, alternative explanations for the two different kinds of
answers have been suggested. For instance, in a similar task, Riecke and Wiener

Figure 4.2 Screenshot of a tunnel during the virtual journey in the tunnel task
used in Schönebeck et al. (2001), followed by a set of arrows shown when
emerging from the tunnel.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 109

(2007) identified further systematic responses that were not consistent with any
reference frame, but consistent with a failure to update turn changes.
We (Tenbrink & Salwiczek, 2016) were intrigued by this scenario, and set
out to collect verbal data associated with the tunnel task. As outlined above,
previous literature had identified several linguistic differences that correspond
to certain underlying perspectives and reference frames. Accordingly, we
reasoned that if the tunnel task led to two distinct reference frames, then the
language that people use to talk about it should reflect this in some way. This
might not be the case directly – people wouldn’t be saying things like I am
using the allocentric reference frame, or even I’m relying on my own point of
view. Actually, reference frame use may not be a matter of conscious decision;
speakers may not be aware of the perspective they’re using. Nevertheless,
people are easily able to use language to express spatial relations in various
ways, using different reference frames, and the language they use does typi-
cally reveal the underlying frame systematically – as long as it is clear what
they are referring to. Therefore, we expected to be able to collect meaningful
and informative language data using the tunnel task.
Although the tunnel task scenario had not been previously associated with any
specific linguistic terms, we expected that some kind of systematic difference
should emerge if people were consistently using distinct (allocentric vs. egocentric)
reference systems. We (Tenbrink & Salwiczek, 2016) used the linguistic distinc-
tions suggested by Taylor and Tversky (1996) as a starting point, since their survey
perspective corresponded to an allocentric reference frame, and their route per-
spective to an egocentric one. However, since the scenario was very different in our
study, we also did a close qualitative post hoc analysis of the linguistic data we
collected in order to allow for the emergence of different patterns from the data.
Apart from recording the responses that our participants gave during the
tunnel task, we collected two types of verbal data. First, participants were asked
to think aloud while travelling through the tunnel, and also when they gave their
response indicating where the tunnel entrance was. Second, after completing
twenty short virtual tunnel trips, they were asked the following retrospective
question: ‘Please describe to me in a few sentences what you think was
happening and how you made your decision at the end of each tunnel journey.’
This was intended to trigger further thoughts that may have run through the
participants’ minds, without them necessarily having had time to speak them
out loud before. Also, combining think-aloud with retrospective reports
allowed us to look at two different kinds of verbalisations, which could be
expected to be quite different in nature (see also Chapter 2).
Our results showed that just like in the studies before us that used the tunnel task,
most participants consistently chose one of the two possible solutions, associated
with the allocentric and the egocentric reference frame by Gramann et al. (2005) and
others. However, quite in contrast to this expected behavioural finding, almost all of

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 110 Identifying Cognitive Orientation

the linguistic markers of reference systems found in the verbal data fell into the
category of route perspective indicators, pointing to an egocentric reference system.
This was independent of the behavioural choices. Less than two per cent of the
linguistic markers, forty-three in number, fell into the survey category, which
contained survey markers found previously in the literature as well as some
additional terms identified post hoc in this data set, such as map and compass.
These seemed to be associated with the allocentric response to the tunnel task more
often than with the egocentric one.
Nevertheless, the fact that almost all markers indicated a route perspective left
us puzzled. Apparently, our participants did not consistently verbalise a survey
perspective even when they produced the kind of response that seemed consistent
with a survey perspective. What did they actually do? To address this, we
conducted a qualitative analysis of several data sets, in terms of individual case
studies. This turned out to be particularly illuminating in this study.
The results showed that participants developed highly idiosyncratic stra-
tegies to deal with the tunnel task. Only a few participants showed the
expected behaviour in a consistent way – conceptualising the tunnel from
either a route or a survey perspective, and representing this in the way they
talked about it. The others came up with an intriguing set of creative
strategies that helped them to cope with the considerable cognitive chal-
lenge that the tunnel task apparently poses. This, in itself, was a surprise, as
the tunnel task had been proposed as a simple test that could easily be used
to identify basic cognitive mechanisms and preferences for choosing
between them. Altogether, our data did not confirm that participants were
actually using the kinds of (egocentric and allocentric) reference systems
that are widely discussed in the literature and known from other contexts.
Instead, they seemed to be adopting a range of strategies that were more or
less consciously accessible. This suggests that using consistent spatial
reference systems in virtual reality settings may be more challenging than
previously assumed.
In sum, this example illustrates a study in which perspective was examined
systematically on the basis of language data along with a spatial situation in
which perspective was decisive. Since perspective was not articulated expli-
citly, it took various different analysis approaches to draw conclusions about
the underlying perspectives. Results were interesting in that they revealed
different concepts to those previously expected.

4.2.3 Conceptual Perspectives

Spatial perspectives are intriguing for linguistic research, because it is possible
to relate the spatial situation to the linguistic expression chosen by a speaker
and thereby to identify the underlying perspective. However, perspective

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 111

taking can mean much more than this. We take other people’s perspectives in
many different ways, not just spatially.
In a sense, most misunderstandings mean some kind of clash in perspectives.
If your understanding of an utterance or a situation differs from mine, you
clearly have a different perspective on it, and we might miscommunicate.
Consider the following fictive example:

Example 4.20 Jill: This window really needs cleaning before our guests arrive,
and dinner will already be late.
Pete: It’s okay for me to eat a bit later!
If, in this example, Jill hoped to get some help from Pete for the household
chores, she will be very disappointed with his answer. From Pete’s perspective,
it’s up to Jill to do the window cleaning and also to prepare dinner; it does not
occur to him that she might be indirectly asking for some help. From Jill’s
perspective, there is no such thing as a given task allocation; her aim in
mentioning both urgent tasks was to divide the labour, so that both can be
accomplished at the same time. Clearly, the perspectives clash and as a result
the speakers miscommunicate, leaving Jill disappointed.
What constitutes a speaker’s perspective, beyond the spatial perspective already
discussed in Sections 4.2.1 and 4.2.2? First, perspective is determined by the here
and now – by the speaker’s time, place, and identity (Schober, 1998). These
aspects can be more or less decisive in a speech situation, or may come into
play in different ways. As I am writing this book, I write with you in mind as
readers – but I have no idea where and when you will read it, or even who you are.
Likewise, you will have no idea what time of day or season in the year it was when
I wrote it, or what my desk looked like. You might to some extent be interested in
the identity of me as the author, or the name might mean nothing to you.
Thus, in this particular instance of communication, the speaker’s (or writer’s)
time, place, and identity seem of little relevance. Books are written to be understood
independently, in a relatively context-free way; different readers should ideally be
able to understand them at a different time and across different places – though there
are limits to this. In contrast, if you talk to your friend about the argument you had
yesterday with your partner, it will typically matter quite a lot who you are, when the
argument took place, and also perhaps where you talk to your friend.
Language represents such aspects through the class of so-called deictic
terms: I, you, here, now, then, there, and the like. These terms immediately
convey a sense of context dependency; you need to know who I is, and where
here is, when hearing a simple exclamation like I am here! Apart from that, it is,
of course, also possible to quote someone else saying these words: Anne
shouted, ‘I am here!’ Then, the context makes it clear that the word I refers
to Anne, not to the speaker who quotes Anne.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 112 Identifying Cognitive Orientation

In a literary context, the perspective from which a story is told is one of the most
basic features that stand out, affecting the atmosphere created by the narrative.
Literary stories can be told from a first-person perspective, and in the most extreme
case the reader may never learn the narrator’s name at all. As everything is
recounted from the narrator’s point of view, they might never have any reason to
state their own name. This represents the most direct way of expressing someone’s
perspective from the inside, so to speak. The story will be restricted to the kinds of
things that the first-person narrator knows, and we might learn something about
their thoughts and feelings that nobody else could know.
Although the first-person pronoun I provides a direct mechanism to convey
such an inside perspective, very similar effects can be achieved (perhaps less
obviously) using third-person pronouns (he or she). A story told from a third-
person perspective, as in He went to the store, feeling hungry . . ., also allows
the reader to adopt this person’s perspective, and to identify, to some extent,
with the main character. This will depend on whether the reader learns about the
character’s thoughts and feelings, and whether the story is told in such a way as
to restrict the character’s knowledge in a realistic way. In contrast, an ‘omnis-
cient’ narrator telling the story would not be part of the story as such; they
would know things that none of the characters could possibly know, and might
be able to report thoughts and feelings of various people.
In literary genres, authors often experiment with these ways of taking
perspective. Some stories are narrated, chapter by chapter, by different people.
Perspectives can switch in the middle of a story. First-person and third-person
pronouns can switch even while using the same person’s perspective. And
sometimes the reader can be deceived by an untruthful narrator, to be revealed
towards the end of a story. There is a lot of room for creativity here, and the
distinction in terms of pronoun choice (first- or third-person perspectives, or
even second-person where the author pretends to address the reader) only
affects the surface, not the scope for perspective taking as such. Readers may
identify with a first-person narrator more readily than with a third-person one
(Brunyé et al., 2009), but perspective taking takes place all the same.
Just as we are able to take other people’s perspectives and imagine their
identity when perceiving the world, we can imagine being at other places (there
rather than here) and at other times (then rather than now). Considering what
the world was like during the Middle Ages immediately evokes a range of
images of dirt, poverty, horses, perhaps castles – possibly supported by films
we may have seen, or books we may have read. We will surely be aware of the
technological limitations at that time, allowing us to develop a sense for the
perspective of a person in the past who could not travel by car, not commu-
nicate by online chat, and not gather information through the Internet. Our
minds are incredibly flexible in this regard; as long as we have some informa-
tion to go by, we can imagine and elaborate on other kinds of perspectives.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 113

Schober (1998) further highlights how our perspectives are revealed by the
ways in which we think of objects and people. Any person or object can be
referred to in many different ways: we can talk about Dr Anthony Myers, or
Uncle Tony, or Dad, or the Golf Club president, or neighbour Anton, or my
stepson – this depends on who we are and how we relate to this person. Likewise,
we could mention the school, or the workplace, or that horrible building over
there, or the meeting place – all of these may be relevant in different contexts for
a single location, depending on the perspective of the speaker.
In this way, the use of a single noun phrase reveals the underlying conceptual
perspective by contrast with the many other ways in which the same person or
thing could have been referred to. In this regard, it can be very interesting to
focus specifically on phenomena highlighted by cognitive linguists (see
Chapter 3): for instance, if a person consistently talks about how anger boiled
up or the argument became heated, this person clearly adopts the metaphorical
concept of ANGER IS HEAT – a common concept with many implications
(Lakoff & Johnson, 1980). Then, the person’s underlying perspective and
understanding is revealed in their use of lexical choices, beyond noun phrases
used to refer to objects and people.
Schober (1998) furthermore suggests that a speaker’s conversational agenda
also corresponds to some kind of perspective. To illustrate this idea let’s
consider the following exchange:

Example 4.21 Mary: I had such fun at the party yesterday.

Colin: No need to rub it in!
Here, Mary may have wanted to share the memory of the party experience
with Colin, without any further implications. Colin, however, has a different
perspective; for some reason, he feels offended by the fact that Mary had fun at
the party. Perhaps he feels that he should have been invited; perhaps he is jealous
for some other reason. Clearly, Mary’s innocent communicative intention is not
received very well. She failed to take Colin’s perspective, and therefore her
conversational goal of sharing a positive experience did not work out.
To really understand what is going on at this level in any given instance, we
need to know details about the people involved and their background, their
communicative intentions, their relationships to each other, and so on. Failing
to understand another person’s perspective at the level of an unspoken con-
versational agenda concerns misunderstandings of a kind that may be fairly
hard to clarify. The (semantic) meaning of the utterance itself is clear enough,
so there’s nothing to clarify there. But explaining communicative intentions (or
aspects pertaining to the pragmatic level of language) is not necessarily some-
thing that people are used to doing, or willing to do. It takes a lot of perspective
taking to really understand a speaker’s underlying meaning at this level.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 114 Identifying Cognitive Orientation

In sum, perspectives are not only decisive in space, but actually underlie all
communication in a range of ways. Identifying the conceptual perspectives in
a given discourse is not easy, but can be central for making sense of the
discourse, for understanding it properly.

4.2.4 Conceptual Perspectives: A CODA Example

We have seen, in the previous section, that a person’s perspective has a lot to do
with their identity, with their personal background, the time and place they’re
speaking, their conversational agenda, and so on. In a theoretical essay,
Tenbrink et al. (2014) took a closer look at a communication situation where
various different perspectives come together, which frequently leads to con-
ceptual clashes and communicative challenges. This is the case in architectural
design, especially when designing new complex public buildings such as large
railway stations or hospitals.
In our essay, we considered the perspectives of architects, their clients, and
building users. Architects construe a building from their own professional
perspective, which involves aspects such as aesthetics, functionality, innova-
tion, and more, combined in a creative way. Their clients are the people tasking
them with building construction, and their perspective on this task can be vastly
different, and more economically oriented. But the people who will finally end
up using the building may not be concerned with any of those things; users
perceive a building in a different way than architects do; they may or may not
identify with its structure, but first, they will need to find their way around.
Appreciation of a creative and innovative building has its limits if the building
users frequently get lost in it (Carlson et al., 2010). Communication between
architects, clients, and users therefore involves a high amount of flexibility in
perspective taking – a significant cognitive challenge.
Related to this line of work, we (Tenbrink, Brösamle, & Hölscher, 2012)
used a qualitative CODA approach to investigate to what extent trained archi-
tects switch perspectives, and in particular how they manage to take a building
user’s perspective during the architectural design process. We used a set of
interviews conducted previously with architects to learn about their conceptua-
lisations during design (Brösamle & Hölscher, 2012).
The interviews included three ways of thinking about buildings. First, the
architects were asked to describe and comment on a schematised plan of a basic
building structure; this encouraged thoughts and ideas about the building as
such. Second, they were invited to design the basic building structure further by
introducing a circulation system, such as staircases in appropriate places; this
induced a design perspective. The third question concerned building users;
architects were asked to imagine a person trying to find their way to an office
within the building. Architects’ answers were generally very elaborate, and as

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 4.2 Perspective 115

such provided plenty of evidence for the ability to take different kinds of
perspectives related to architectural design.
Our analysis in Tenbrink et al. (2012) focused on the ways in which
different perspectives were expressed in language. We speculated that it
might be possible to associate pronoun use with different perspectives: it
for the building, I for the design (as in, I’d change this aspect), and you for the
user (as in to find the office, you’d have to go this way). However, in actual
fact the situation turned out to be rather more complicated. While the different
underlying perspectives did find systematic expression in language, all three
pronouns could be used with all three perspectives, as shown in the following
examples:

Example 4.22 I mean, I don’t know what the scale of the drawing is but
I suppose . . .
Do you have rooms for patients towards the internal corridors?
Maybe it’s some kind of lecture hall or something.

Example 4.23 I’d keep it fairly classical.

You might put a stair round the back of it.
It might make sense kind of placing them here.

Example 4.24 I naturally would walk to the third floor.

If you are meeting someone on the third floor that’s not the
staircase for you.
The person will try to find a different staircase in order to go up.
In Example 4.22, the building is described in various ways: from the
observer’s point of view (first person), from a metaphorical owner’s point of
view (second person), and with respect to the buildings features as such (third
person). In Example 4.23, the first-person perspective describes the possible
design changes from the architect’s point of view as expected. However,
the second- and third-person forms are also available for a similar purpose,
where the pronoun you represents a less personal way of expressing the
possibilities, and the third person is used for impersonal constructions such
as passive voice and references to the envisioned state of the building. In
Example 4.24, the user’s perspective could be expressed in either one of the
three pronoun types, corresponding to the different ways in which stories can
be told as outlined in the previous section.
It can be seen from these brief examples that, although pronoun use as such
was not an indicator for perspective, the ways in which pronouns were used in
context revealed the different perspectives adopted in each case, with different
degrees of personalisation. Interestingly, the second-person form you was only

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 116 Identifying Cognitive Orientation

rarely used in its core sense, namely to address the interviewer. This shows how
immersed the architects were in the task of discussing various perspectives on
the architectural design process.
Another interesting finding in Tenbrink et al. (2012) was that while the
questions targeted the three different perspectives one by one, architects actually
moved freely between them in relation to all three questions. Therefore, we could
observe frequent shifts of perspective in a single response, as in this example
where the speaker shifts back and forth between the building and the user:

Example 4.25 . . . if you come from there [USER]

because they don’t seem to have entries.
It seems to be the only one or maybe this is a disabled entry.
I don’t know. [BUILDING]
Or you could come from here [USER]
if that’s a connection. It only seems to be a connection on the
first floor.
So I think this is the entry. [BUILDING]
So you come in and you see this stair and that goes to nowhere.
So that’s the first problem.
So you don’t – you probably go to – you want to go here. [USER]
This example illustrates a recurring phenomenon that we identified in
our data. When the architects shifted perspectives between building,
design, and user, they frequently employed causal or conditional conjunc-
tions such as so, so that, if, in order to, because, and then. Also, we noted
that shifts between perspectives were frequently necessitated by the nature
of the task question given to the architects in the interviews. In Example
4.25, to discuss the challenges a wayfinder faces in the building, the
architect’s thought process goes back and forth between the wayfinder’s
actions and the features of the building that the wayfinder encounters
along the way. This demonstrates the flexibility with which architects
conceptualised and reframed the various scenarios posed to them in the
course of the interviews.

This chapter has looked at two main ways in which human thoughts are
oriented, as reflected in language use: attention and perspective. Attention
means the focus of thoughts on a particular aspect of a situation or idea,
whereas perspective is the angle from which such a focus happens. Together,
these two cognitive phenomena represent a powerful mechanism for concep-
tualising the world in an idiosyncratic way, generating the intriguing diversity
in both thought and language that cognitive scientists encounter in their quest to
understand the human mind.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:31:20, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 5 Identifying Cognitive Depth

While the previous chapter looked at the ways in which we cognitively orient
towards specific aspects in the world or in our minds, this chapter turns to the
depth in which we do this. Once more, there are two aspects to this: on the one
hand, we can think about things in more or less detail; this is captured by the
notion of granularity (Section 5.1). On the other hand, our depth of cognition
can be limited in terms of understanding and knowledge; here we look at the
certainty of our thoughts and thought processes (Section 5.2). Together, these
two notions complement the insights on cognitive orientation in terms of
attention and perspective discussed in Chapter 4. As with the previous chapter,
we will find strong links between cognitive science and linguistics, adding to
the kinds of insights that we can gain about cognition through a close look at
how people speak in a controlled scenario.

5.1 Granularity
Imagine you are in a museum with your friend, looking at a famous picture that
was painted centuries ago. You ask your friend whether she likes it, and she
replies in either one of the following ways:
Example 5.1 It’s a nice painting. I like the colours and the scenery in it.
Example 5.2 It is fantastic, truly intriguing, an inspiring picture. Just look at
how the depth of blue shifts in the sky in various places, and how
the leaves on the trees show maybe twenty different versions of
green, depending on the light intensity, the type of tree, and even
the freshness of the leaves at springtime!
Both of these possible answers are perfectly normal, and they could be
given by any of us. The answer given in Example 5.2 is perhaps more
likely to come from a person who likes art very much, or has some
professional background in the area, whereas Example 5.1 sounds some-
what less informed. But in general, there is nothing in the answers that
actually requires or directly reveals professionality or expertise. Moreover,

117

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 118 Identifying Cognitive Depth

considering the analysis perspectives discussed in the previous chapter, we

don’t see many differences in attention or perspective. Both mention
positive evaluation, colours, and scenery, and look at the picture as
a whole. But they use a different amount of words for the description,
and this gives a first indication of the main difference here: namely, that
the two examples are very different in how much detail they offer.
Example 5.1 is restricted to fairly general expressions: nice, colours, scen-
ery. These terms express three different aspects that your friend has thought
about: her general opinion of the painting, and then (somewhat more specifi-
cally), the colours and scenery depicted in it. However, she does not go into any
more detail, and does not say what she likes in general, or what she thinks about
the colours and scenery.
In contrast, Example 5.2 elaborates on each of these three aspects. The
general evaluation is more specific in that it adds the adjectives intriguing
and inspiring: she has clearly engaged with the painting intellectually, and
that’s why she thinks it is fantastic (not just nice). Then, she provides some
details as to how colours are depicted in the scenery. It turns out that both of
these aspects actually work together for her. The speaker elaborately points to
the ways in which the scenery is represented in colour variations and nuances,
going well beyond what might be seen as a blue sky and green leaves. This
reveals a close look at the picture. At first glance, even from a distance, one
might discover that the picture is good, and has appealing colours and a fine
scenery. However, to be able to describe these aspects in as much detail as your
friend does in Example 5.2, one will need to ‘zoom in’ to the picture – really
take a close look at it, discover in detail how the colours shift and vary along
with elements of the scenery.
Variability of granularity, as shown in this example, can be understood as
cognitive zoom. The more you zoom in, the more you will discover, at a more
fine-grained level. If you look at a picture from a distance, you might see what it
shows in general and perhaps whether you like it. In contrast, a microscope
would allow you to see fine-grained details in very small spaces.
However, this comes at a cost: the use of a microscope might prevent you
from seeing the picture as a whole, from appreciating its general impression.
Therefore, a given level of granularity is not necessarily good or bad. It simply
describes the level or scale from which things are considered, and this can be
equal to or more or less appropriate than any other level. A coarse description at
a general overview level may be useful for some purposes, whereas detailed
elaborations of fine-grained insights and ideas serve different ones. When
asking your friend about a picture in a museum, you may be entirely happy
with an answer like Example 5.1. You may not be interested in dwelling on the
topic for longer, or in looking so closely at specific aspects of colour variety.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.1 Granularity 119

Too many details can be inappropriate in some situations, or simply boring for
the communication partner. And too few details may not suffice.
In the following, we will take a closer look at two notions that are associated
with granularity: scale (Section 5.1.1) and elaboration (Section 5.1.2).
A CODA example on variable granularity in route descriptions will conclude
this section (Section 5.1.3).

5.1.1 Scale
As just outlined, granularity means mentally zooming in, and this very often
corresponds to providing more details on a finer scale. What exactly do we
associate with the notion of scale? Scales are well known in human society, in
many ways, not only as an idea but also as concrete objects. Scales are tools for
measuring weight, distance, volume, wind force, and other things, depending
on their nature. What they have in common is that they can be used at different
levels of granularity. Weight, for instance, can be measured using scales at the
levels of tons or micrograms; distances can be miles or millimetres, or mea-
sured at an even smaller scale.
Scales can be introduced for all sorts of structures, along various dimensions
(Hobbs, 1995). We can effectively carry out comparisons based on the scales
we use, and we can use them in a composite way. For instance, if a car is
damaged, we can assess the damage on the scale of associated costs, as well as
on the scale of loss of functionality. These are two different dimensions, and we
may be able to weigh them against each other, or account for their
interdependency.
Through multiple experiences in everyday life, we are used to these effects.
However, our minds don’t offer any precise metrical scales for us to work with. We
might have a relatively good idea of what it feels like to hold a 100 gram bar of
chocolate in our hands (or perhaps 70 gram, depending on our cultural habits!), but
to precisely determine 100 gram of anything, we will need a tool.
Accordingly, the conceptual notion of scale that we are considering for
current purposes is not based on exact metric measures. Instead, it is primarily
a qualitative notion. Qualitative scales are reflected in language, in the way we
speak. Imagine Corinne trying out a new running route without consulting any
measuring instruments, and then saying the following to Mike:
Example 5.3 This route was 2.135 miles less of a challenge than my usual
route.
Example 5.4 This felt short. My usual route must be longer!
Mike would probably be astonished about Corinne’s ability to measure
distances without tools if she said something like Example 5.3, whereas

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 120 Identifying Cognitive Depth

Example 5.4 would seem perfectly normal. We don’t feel (or otherwise know)
precisely how much shorter a particular route is, but we may feel that it is short
as compared to something else. Importantly, attributes such as short are
inherently scalar. They inherently classify something as less than expected,
compared to something else from our experience, or some cultural or individual
norm. How long is a short route? We can’t tell unless we have some notion of
a distance to compare it against. This will depend on context; what seems long
for a walking route may be very short compared to our habitual running
distance.
To take a different example, we might have an idea of the approximate size
range of a large mouse or a small elephant; in each case, we intuitively compare
the animal in question with an average exemplar of its category. Large mice are
larger than the average mouse, and small elephants are smaller than average
elephants. However, it would be rather a stretch of imagination to consider even
an extremely large mouse to be larger than an extremely small elephant – at
least in the real world. Fiction can provide different scales, offering new
opportunities for comparison. The fact remains, though, that scalar attributes
are relative; without context or associated world knowledge, they do not have
any concrete value of their own.
Granularity variations help to identify the underlying cognitive scale that is
used implicitly for establishing the contrast that is at stake in a given context
(Freksa, 1981). Compare the following with Example 5.4, in the same
situation:
Example 5.5 This felt fairly short. My usual route must be longer!
Example 5.6 This felt incredibly short. My usual route must be longer!
If Corinne utters the statement in Example 5.4 (this felt short), we might infer
that the underlying scale has fairly coarse distinctions between short, normal,
and long. In Examples 5.5 and 5.6, the words fairly and incredibly serve as so-
called amplifiers (Martin & Rose, 2003); they add further levels of detail to the
scale. In Example 5.5, we might assume that the scale not only distinguishes
between short, normal, and long, but further introduces fairly short and
possibly very short as additional options. Example 5.6 seems to add even
further detail, resulting in a possible underlying fine-grained scale such as
incredibly short–very short–short–fairly short–somewhat short–normal (and
so on).
We may not consciously think of all of those possible attributes, but we
do understand intuitively that incredibly short is further away from normal
than a simple short would be, and also than very short would be. As
a flexible and simple term, very could be expected to serve the purpose of
highlighting that something is quite far away from normal; the fact that

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.1 Granularity 121

even this term is not considered sufficient tells us that the distance must be
extreme indeed. Alternatively, the speaker might be habitually exaggerat-
ing, speaking in extreme terms when a moderate term might have
been fine.
If we happen to know the metric information in Example 5.3 to be true, we
can infer from Example 5.5 that Corinne typically runs long distances where
a difference of 2.135 miles does not matter very much, whereas in Example 5.6,
the subtraction of 2.135 miles seems to have reduced the running route to
a minimum and consequently, her normal routes may not be longer than three
miles. This inference is nowhere to be found in the language itself, and may
well be unfounded. Nevertheless, the choice of amplifiers (or their absence)
does provide important insight into the conceptualised underlying scale on the
part of the speaker. This may allow for hypotheses concerning the reasons for
the chosen scale – especially if we have access to metric information, com-
plementing the qualitative measures used by the speaker.
The fact that context-dependent attributes allow for inferences has also been
discussed in the context of the Gricean maxim of quantity (Grice, 1975). Grice
suggested that speakers generally aim to provide as much information as they
perceive to be required in a given context. Therefore, if Corinne utters
a statement like Example 5.4, Mike (as the listener) can assume that Corinne
does not know exactly, or does not deem it necessary to share, how much
shorter the route was – because she would have uttered Example 5.3 if this
amount of information was available and relevant. On top of that, if she had felt
that the route was not just shorter than normal but extremely so (e.g. because
she normally doesn’t run more than three miles), she would have uttered
Example 5.6. Because of this, Mike can assume that Corinne (uttering
Example 5.4) felt the difference but did not perceive it as extreme.
This kind of phenomenon is called scalar implicature, and it applies well
beyond the use of context-dependent attributes. Carston (1998:179) discusses
the following examples:

Example 5.7 a. Bill has got some of Chomsky’s papers.

b. The speaker believes that Bill hasn’t got all of Chomsky’s
papers.
Example 5.8 a. There will be five of us for dinner tonight.
b. There won’t be more than five of us for dinner tonight.
In both cases, the first part (a) includes one word (some and five, respec-
tively) that licenses the implicature given in (b). If Bill possesses some papers,
it might be logically possible for him to possess all papers – but considering the
underlying qualitative scale (e.g. some–most–all), it is normally assumed that

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 122 Identifying Cognitive Depth

he does not possess all of the papers, or even most of them. Likewise, if we use
a number like five, the normal understanding is that the actual number will not
be higher, even though it may be logically true that five people is a subset of six,
and therefore a dinner party of six does include five people, strictly speaking.
However, like all implicatures, this inference can be cancelled depending on
context. For instance, if the speaker in Example 5.8a needs at least five people
for some reason, the relevant information would be that the target number has
been reached. In such a case, it may be irrelevant to add the exact number, as in
Example 5.9:
Example 5.9 There will be five of us for dinner tonight – we’ve made it!
Actually we will have four guests, so we’ll be six people, more
than enough.
Scales are also relevant in the context of time frames. Consider the
following:
Example 5.10 Fanny, what will you do after lunch today?
Example 5.11 Fanny, what will you do after your holidays?
The use of after in cases such as these suggests the notion of immediate
succession (Tenbrink & Schilder, 2003): the question pertains to the next
action that Fanny has in mind, immediately after lunch – and immediately
after the holidays. The difference between the two types of action lies in
the underlying conceptual scale. The concept lunch in Example 5.10
suggests a time frame of hours in a day; Fanny might say that she will
go shopping, fitting to the time frame. She might not say that she will do
the dishes and put her shoes on, and the like: these actions are more fine-
grained, and do not interfere with the idea of ‘next action’. In contrast, the
use of holidays in Example 5.11 invokes a temporal scale of days or
weeks; it would be very odd if Fanny answered that she will go shopping.
Both questions implicitly ask for the next activity on the same level of
granularity, without making the scale explicit. Explicit scales are not
necessary, due to our intuitive understanding of the granularity level
invoked by concepts such as lunch and holidays.
For the case of space, Montello (1993) suggested a classification that is now
widely accepted and used in the spatial cognition literature. He argued that
from a psychological perspective, different scales of space are regularly and
systematically important across the contexts in which we live. Also, previous
literature on how we understand space had pointed to a number of effects that
occur at a particular scale, in contrast to another, as follows.
On the most fine-grained level of granularity, figural space concerns every-
thing that is smaller, or is perceived as smaller, than the human body. This

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.1 Granularity 123

includes pictures representing space of any kind; maps (independent of what

they show) belong to figural space. This reflects the idea that we think of spatial
objects and pictures differently if they appear smaller than ourselves. We can
handle them differently, manipulate them with our hands, and so on. Not every
figural-space object can or needs to be touched directly in order to be fully
understood, but the small scale of space still matters conceptually.
On the next level, vista space concerns everything that we can perceive from
a current vantage point. This will naturally depend on the environment that we
happen to be in; indoor vista space is naturally restricted by walls, but if you’re
standing on the seashore or looking at the mountains, your vista space may
include fairly large distances and objects.
Next, environmental space includes space surrounding us that is too large (and
partially occluded) to be perceived at once, but it can be apprehended and learned
over an extended length of time. Knowledge about environmental space includes
knowledge of areas like cities and the interior of complex buildings; it must be
gradually accumulated, but is still based on personal experience.
Finally, geographical space is a scale of space that is too large to be
apprehended by a human being’s experience. Our knowledge of geographical
relationships between entire countries, or relations between oceans and land
masses, is never based entirely on our own experience. Instead, such knowl-
edge is acquired through maps or models that help us understand what we
cannot perceive directly. As Montello (1993) pointed out, such maps or models
only represent geographical space, but they are themselves objects in figural
space. Thus, when we consider knowledge at this scale, it is useful to keep in
mind that this kind of knowledge has been acquired through figural space rather
than through the actual large-scale space of geographical relationships.
Altogether, the notion of scale is a powerful mechanism that provides multi-
ple ways of conceptualising the world at different levels of granularity. The use
of a particular linguistic term to describe a concept or attribute frequently
reveals the underlying conceptual level of granularity, based on the scale that
is associated with it. In the case of space, references to diverse kinds of spatial
objects or relationships reveal the underlying scale just as reliably. A person
referring to a map relies on the figural space that a map provides, whereas
a person who recounts the relative location of not currently perceivable places
from memory would rely on experientially acquired knowledge of environ-
mental space. In practice, these different types of knowledge at different scales
often interact and are combined in useful ways.

5.1.2 Elaboration
We can think about and describe the world at different scale levels, as just
described – but this is not the only way in which cognitive detail can be revealed.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 124 Identifying Cognitive Depth

Even at the same scale level, our thoughts can expand to varying degrees, leading
to more or less detail in thoughts and descriptions. Further to the examples given
above, consider the following two ways of expressing very similar ideas:
Example 5.12 This felt short. My usual route must be longer!
Example 5.13 I really had the impression that this route was short, it felt
shorter than usual. The route that I normally run must be
longer, it makes me more tired, today I did not really feel
exhausted after this route.
Intuitively, there is not much difference in content of these two utterances –
they both make the point that the route was shorter than usual. Example 5.13 is
simply more elaborate than Example 5.12; it provides more details, spells out in
more words what exactly was felt.
Providing further elaboration is a common discourse feature that combines
textual elements (utterances, clauses, sentences) in sequence. In Example 5.12,
the main point is already conveyed in the first three-word sentence – ‘this felt
short’: if the route felt short, the implication is that there must be a longer route
that feels normal. The scalar nature of the adjective suggests this; short is
understood in relation to an implicit norm on this scale, as described above.
The second sentence in the same utterance does not introduce any substantial
changes or new elements; rather, it serves to provide further details on the same
idea. In short, it elaborates the first sentence. In Example 5.13, all follow-up
clauses elaborate the first one, and this results in the overall impression of
a fairly elaborate representation of the experience.
Elaboration plays a central role in expressing granularity. It is a discourse
feature that allows the speaker to go into more depth without necessarily chan-
ging scale (in the sense of the previous section). This simply means that the
speaker provides further details, further evidence of the same point, perhaps
different perspectives and aspects that play into it (as in Example 5.13). Halliday
and Matthiessen (2014:461) offer the following very useful definition (annota-
tion and symbols removed from the examples they give; my numbering):
In elaboration, one clause elaborates on the meaning of another by further specifying or
describing it; for example:

Example 5.14 Is there any way of disputing when a priest, any priest, says, ‘This
is my conviction, this is what the god I serve thinks about this.’
Example 5.15 Moo, however, and the novel I’m writing now, which is
a racehorse novel, are comic. They are set in a more stable time, where
things aren’t crushed and lost, they simply go on.

The secondary clause does not introduce a new element into the picture but rather
provides a further characterization of one that is already there, restating it, clarifying it,

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.1 Granularity 125

refining it, or adding a descriptive attribute or comment. Thus in the first example above,
This is my conviction is elaborated through restatement by this is what the god I serve
thinks about this. The thing that is elaborated may be the primary clause as a whole, or it
may be just some part of it – one or more of its constituents. Thus which is a racehorse
novel elaborates part of the clause Moo, however, and the novel I’m writing now are
comic; it elaborates the novel I’m writing now and, being a hypotactic elaboration, it
follows its domain of elaboration directly and is as a result included within the dominant
clause (the traditional category of ‘non-defining relative clause’).
Elaboration is very common in everyday discourse; we frequently elaborate
what we’ve just said. We do not want to say the same thing twice, but there may
be more details to be added, more to be said on the subject. And we usually
want to make sure that we have been understood; elaboration often helps to
clarify what has been said. Teachers frequently ask students to elaborate on
a statement they have made, testing the depth of their knowledge – or the extent
to which they are simply repeating something they have heard. If students
cannot elaborate, the chances are that they haven’t properly understood
a concept or phenomenon. Elaboration doesn’t necessarily add anything new,
and may therefore sometimes be considered redundant, but in fact it is quite
essential in order to communicate successfully. We need a certain amount of
cognitive depth in order to exchange our thoughts effectively, and elaboration
is an important discourse tool to achieve this.
In the verbalisation of thought, elaboration would not necessarily be under-
stood primarily as a tool for better communication. After all, CODA means
looking at cognition, not communication. A CODA study design that involves
representing experience through language is not primarily aimed towards
successful communication, but towards the expression of cognition. In this
light, a speaker who further elaborates something that has been said before may
simply be expressing increasing depth of understanding. They may start out
thinking about something in a tentative way, and then go on to discover and
understand more and more aspects of it that allow them to elaborate their
thoughts. This would then be reflected in their language in terms of increasing
elaboration, revealing increasing cognitive depth.

5.1.3 A CODA Example: Variable Granularity in Route Descriptions

How much cognitive depth do we need, or employ, when describing routes?
Previous literature on wayfinding suggested that route descriptions are char-
acterised by a set of essential elements, such as landmarks, direction changes,
segments, and the like (Denis et al., 1999). These are particularly pronounced at
critical decision points. In effect, this means that there may be differences in
terms of granularity. In Tenbrink and Winter (2009) we specifically pursued
this point in a number of ways. We explored the notion of granularity in this

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 126 Identifying Cognitive Depth

context, and proposed a model that enabled us to explore the phenomenon

systematically. The model distinguishes between granularity at one-
dimensional (1D) and two-dimensional (2D) levels as well as elaboration.
The following examples are cases in point.
Example 5.16 At the first intersection, turn right. Then at the next intersection,
walk straight on. Turn right at the next intersection after that.
Example 5.17 Turn right at the first intersection and then again at the second
intersection after that.
Example 5.18 I live in the city centre, close to the cathedral, in a small
building next to a little shop.
Example 5.19 You will need to turn right at a nearly 120-degree angle.
Examples 5.16 and 5.17 describe the same situation, but they do so in
different ways. Example 5.16 describes each individual segment of the route
separately, from decision point to decision point, even if it means just walking
straight on. Example 5.17 groups two intersections together by referring to
the second intersection. This effect is known as chunking, and it leads to
coarser 1D granularity, that is, less detail with respect to the linear progression
of the route.
Example 5.18 shows a shift in 2D granularity within the same utterance. The
speaker describes the location of a building, their home, by narrowing down to
it in a gradual fashion. They start out with a reference to a relatively large area,
the city centre, then zoom in to a smaller area relative to a large landmark, the
cathedral. The last step is to refer directly to the target building itself, locating it
next to a different, smaller landmark, the little shop.
Example 5.19 illustrates elaboration of a specific route element,
a direction change. It is fairly typical for route directions to leave the
angles of turns unmentioned; commonly, one would hear turn right without
any further specification. In this example, the speaker chooses to elaborate
the turn information by noting the angle, possibly because this particular
turn is unusually sharp. Because a prototypical right turn would have
something like a 90-degree angle, 120 degrees is sufficiently different to
be mentioned. This would support the listener in identifying the correct
road into which they need to turn.
In Tenbrink and Winter (2009), we further specified and operationalised this
classification of three types of granularity changes – 1D, 2D, and Elaboration –
to enable rigorous annotation of natural language data. We were interested in
patterns of granularity changes in route descriptions of two very different
kinds: one automatically generated by a web-based system (provided by
German railway at [Link]), and another produced by young adults
who were asked to write them down on a piece of paper. For this purpose, we

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.1 Granularity 127

chose a route that contained various modes of travel (walking, train, and tram)
as well as more and less familiar sections for the traveller. The web-based
system produced a description that used the same route as the young adults in
our study, enabling direct comparison.
Our results showed some interesting patterns. First, we noted that the
essential information required to find the way was provided across the board,
both by the system and by the young adults. This means that each leg of the
route was reliably identified in some way.
Next, both types of route directions exhibited a hierarchical structure. In the
web-based system, this was realised by the option of learning further details by
clicking into particular elements of the route. In the human route directions, the
hierarchy was realised by route elements at different levels of importance. Our
analysis started from the identification of spatial units, which we defined as
abstract, spatially based categories that reflected the basic level of 1D granu-
larity without elaboration. Some of these spatial units were crucial; they were
mentioned by everyone rather than being subsumed in some kind of chunking
or left out altogether. Others were integrated by shifts to coarser levels of 1D
granularity, marking them as less crucial. Yet other segments were described in
more detail than the basic information needed for a spatial unit, showing finer
granularity in the 1D, 2D, or Elaboration categories.
Both the web-based system and the young adults thus provided further
details, offering aspects of the route that were not strictly crucial for successful
travel. However, they did so in very different ways. The web-based system
primarily offered alternatives in terms of time and travel mode, whereas the
handwritten route descriptions almost completely ignored such details, and
only referred to the fact that the train left at a particular time every hour.
Most importantly, however, the young adults clearly accounted for the
information needs of the traveller in a much more flexible way than the system
did. While the system generated the same information independent of travel
direction, the young adults recognised details that depended on the direction of
travel: for instance, finding the exit of a large railway station when arriving by
train is easier than finding the correct track on departure. The young adults’
descriptions offered detailed information precisely for the more difficult parts
of the route, showing that they recognised a higher need for elaboration. This
was the case, for instance, when changing mode of travel, or when identifying
the correct turn after a lengthy footpath. The automatic system, in contrast, was
much more rigid and incapable of accounting for different degrees of informa-
tion needs; elaboration (wherever it was offered) did little to clarify a complex
or challenging situation. Instead, the automatically generated information was
sometimes redundant in the way it was presented, creating information clutter
rather than providing useful support at the level of granularity required for
travel.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 128 Identifying Cognitive Depth

In conclusion, CODA as applied in Tenbrink and Winter (2009) supported

the identification of specific features of human route descriptions by contrast-
ing them with the features of a system that was developed to support human
wayfinding. The results highlighted various ways in which the system did not
match how humans think and talk about routes. For humans, different aspects
of a route come with different challenge levels, and this is reflected by variable
granularity in route descriptions, going into further cognitive depth wherever
and whenever required. Clearly, a substantial amount of everyday background
knowledge is needed to develop adequate intuitions about flexible information
needs. Our young adults had no trouble doing this, in clearly identifiable
systematic and meaningful ways. In spite of accessing a rich database of spatial
information, the web-based system turned out to be far less adaptable.
More generally, with this analysis perspective we can see how much detail is
provided by a speaker in a given situation. More detail does not necessarily
mean more knowledge, but it does mean that the speaker dwells on the thought
in more depth. The represented details may not be altogether correct or lead to
more insight (this is the topic of the next section), but they do reflect the fact
that speaker is cognitively ‘zooming in’. Note that some speakers may still
‘zoom in’ but not represent this fact in language – this is always a possibility.
However, we can only analyse the data that are available to us, and identify
significant patterns across speakers in relation to specific situations.

5.2 Certainty
How do we know how much another person knows? One way is to ask them;
most people will be able to assess their own wisdom to some extent. However,
knowledge is hard to quantify. If asked how much you know about tap dancing,
you might say, ‘nothing’ or ‘a little’, or you might start explaining the things
you know. If you don’t know much, your words will soon start to reveal
uncertainty about the subject – unless you’re really good at making things up
as you go.
Even though it is difficult to quantify knowledge and certainty, we still tend
to have a good sense of our own, or other people’s, level of expertise on
a particular topic. If someone starts talking about a subject of general interest
or skill, it may not take us long to assess whether the person is knowledgeable
or a newcomer to the subject. This is especially true if we are experts ourselves;
newcomers reveal themselves through simplifications and slight misrepresen-
tations of facts. If we’re not experts, we can often recognise their lack of
knowledge by the way they’re formulating things. Alternatively, we might be
impressed by the level of certainty and confidence in the speaker, but there’s
always the chance that this is ‘put on’ – a fraud; the speaker may be pretending
to know more than they actually do, hiding their uncertainty.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 129

But how do we do this in the first place? How do we assess another

person’s level of knowledge, or degree of certainty, concerning the con-
cepts they are talking about? What factors help us when we make these
judgements? More specifically, what is there in the language that people
use that might serve as clues to identify certainty and depth of insight? In
CODA, our everyday intuitions are taken a step further, as with all other
levels of analysis. Rather than aiming for a general estimate about the
speaker’s expertise, CODA seeks to systematically identify linguistic mar-
kers that reflect cognitive aspects pertaining to the speaker’s knowledge
and certainty. To get a better idea of how this can be achieved reliably, in
Section 5.2.1 we will look at relevant insights from cognitive psychology,
turn to linguistic markers of certainty in Section 5.2.2, and finally discuss
a CODA study in Section 5.2.3.

5.2.1 Expertise and Certainty in Cognitive Psychology

Generally, we would expect that experts can speak with confidence about
a topic, or use a skill more accurately and efficiently than others, based on
their extensive knowledge of it. Experts have spent years in the realm of their
expertise, have talked many times about it, have used their skills in multiple
ways, know all the details, and may have trained others to learn it. However, not
all of the seemingly trivial conclusions that one might draw from this general
understanding of expertise are actually valid. In some domains, expertise does
not necessarily lead to better performance of a skill.
In a study by McManus et al. (2011), for instance, people varied in how good
they were at cropping photographs, according to independent judgements of
the croppings. Surprisingly, the skilled people weren’t necessarily experts, and
experts’ croppings weren’t judged to be any better than non-experts’ croppings.
Experts mostly differed from non-experts in that their croppings were system-
atically different in some ways, and they took longer to do it, taking more
factors into account. Clearly, their performance was affected by expertise, and
they exhibited greater cognitive depth in approaching the task. Nevertheless,
this did not lead to a preference for their croppings by either expert or non-
expert viewers of the cropped photographs.
Another expectation that could be associated with expertise, but does not
necessarily hold, concerns the kinds of strategies used for doing
a challenging task. One might expect that non-experts are more naïve,
less knowledgeable, and therefore might not be aware of some expert
strategies for solving the task. However, Chi, Glaser, and Rees (1982)
reviewed previous work in which this is demonstrably not the case. For
instance, the strategies that expert chess players employ are not substan-
tially different from those of novices, and they do not look any further

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 130 Identifying Cognitive Depth

ahead either. Their expertise is reflected in other ways, such as a better

ability than novices to handle and memorise specific chess positions. The
reason for this is that they have seen certain kinds of configurations so often
that they can think of them as wholes, or chunks of information, rather than
having to conceptualise and memorise the details. This makes it much
easier to recall the structure later, and to work with it during the game play.
Cognitive chunking processes are also known from other domains. As it
happens, the use of language is a case in point. Compare the following two
utterances, heard in an everyday context:
Example 5.20 Better to have loved and lost than never to have loved at all!
Example 5.21 Paul asked Nina first, but Adrian threw in many stupid remarks.
Can you remember both of these sentences with your eyes shut? Chances are
that you have no problem recalling the first one, even verbatim, but not
the second – or at least not the details. This is because the first is a well-
known quote that you are likely to have encountered many times before. If
you’re familiar with it, you’ll have glanced at it, recognised it, and stored it in
short-term memory in an instant, without any processing effort. The second
one, Example 5.21, is a novel combination of words; to memorise it you’d need
to process every detail and possibly make a conscious effort of remembering
the exact words.
Experts, therefore, require less processing effort for familiar situations or
configurations within their area of expertise. They can assess the significance of
a particular detail or constellation because they are trained, explicitly or by
experience, to recognise it. Moreover, they are able to structure aspects of
a situation in a hierarchical way, based on the chunks they are familiar with (Chi
et al., 1982). This is remarkably similar to the observations on hierarchical
thought processes observed in Section 5.1. Experts find it easier to structure
a domain into different layers, and handle each challenge by conceptualising
appropriate levels of detail. Clearly, cognitive depth, whether situational or
expertise based, enables us to establish several layers of thought, to be drawn
upon and referred to when we need it. In contrast, lack of depth or expertise
leaves all aspects of a situation at the same level of importance, due to an
inability to impose a useful structure or identify priorities.
We can expect knowledge of entire chunks, or thinking at this level, to be
reflected in verbal protocols in systematic ways. In problem-solving studies in
the domain of physics (Chi et al., 1982), experts used the same formula as
novices to derive the solution. While experts simply stated the result in a quick
one-step solution, novices evoked the formula, spelled it out, and gradually
came to a solution on this basis. This reflects how problem-solving speed is
enhanced by expertise, and it also shows that consciously thinking about fine

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 131

details is no longer required in the case of experts – they simply know. Our
everyday experience is likely to be similar. If Emily knows what is required to
bake an apple cake, she may simply be asked to buy the ingredients – no further
details are needed. But someone who has never baked a cake, or even not gone
shopping before, will need a lot of details before they can carry out this
particular task.
Again, therefore, we see how granularity is related to cognition.
Whereas a more detailed verbal description reflects enhanced cognitive
depth as discussed in Section 5.1, a short and crisp, seemingly superficial
description that goes into no details may simply be a matter of expertise.
Someone who is very familiar with all the details may not see the need to
spell them out, or may not even be aware of the details any more because
they are simply too normal, too common to be noticed. Nevertheless, the
expert will be able to speak with confidence, and to elaborate if needed in
a particular situation.
Indeed, people may be more or less verbose, but this does not necessarily
reflect how much knowledge they have on a subject. Experts are able to restrict
their attention to the highest level of relevant chunks in the hierarchy, in
contrast to novices who would not be able to recognise them. Thus, a shorter
verbal protocol by a novice may remain on the surface of aspects in a situation
just like the expert’s, but these may not be the same kind of aspects that an
expert would identify in the same amount of words. Chi et al. (1982) confirmed
that quantitative measures (such as word count) do not reflect expertise – it is
what people say, not how much they say, that reveals their knowledge level.
Novices may leave out decisive aspects, overlook crucial relationships, or
misrepresent facts, whereas experts would be able to crisply mention every-
thing essential, including details where required.
Experts can confidently verbalise their thoughts within the domain of their
expertise, but they are also more aware of the limits of their knowledge, or of
the subtle aspects that might sometimes complicate things – more than novices
may be aware of. For this reason, a novice may sometimes sound overly
confident or certain of what they are doing, simply because of a lack of
recognition of subtler aspects. For instance, Mercier and Sperber (2011)
noted that people who are not aware of counterarguments to a particular
position tend to be overconfident; their lack of knowledge is exhibited precisely
by the confidence with which they present a problematic position, one that can
be challenged given a bit more knowledge. Experts may be more cautious and
conservative in such cases. This is especially true when expressing estimates
about future developments. For instance, in the domain of weather forecasts,
uncertainty is pervasive and inevitable (Hoffman, Trafton, & Roebber, 2005).
Long-term forecasts are therefore purposely vague and coarse, rather than

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 132 Identifying Cognitive Depth

attempting to provide precise details for specific time spans or small spatial
areas.
However, even experts can be overconfident within their own area (Chi,
2006). This seems to become most apparent when they are asked to rate their
own performance or assess their degree of comprehension. In such a situation,
novices may be more aware of their own limitations, enabling them to judge
their own level of knowledge more accurately.
In conclusion, we can observe that expertise can be manifest in a range of
ways, and is sometimes – but not always – reflected in degree of confidence.
Lack of certainty expressed in language can therefore have various different
reasons. Expertise is often a strong reason for confidence, but clearly not the
only one – and even experts may become overly confident. Novices may
pretend they know more than they do, or they may simply lack awareness of
the associated complications. And some expert domains are characterised by
a high degree of uncertainty, which may be represented in the language used by
experts when they speak about a situation in which not all factors are available,
such as future developments. In such a case, uncertainty is more a feature of the
situation than a feature of expertise.

5.2.2 Linguistic Markers of Certainty

As outlined in Section 5.2.1, we cannot equate expertise or knowledge with the
certainty expressed in language. However, if the level of knowledge is assessed
independently (e.g. based on previous experience or training), linguistic indi-
cations of certainty can be highly revealing with respect to the cognitive
processes in a novice or expert. For instance, some aspects of a situation may
be described with more confidence than others. Such patterns may differ
between experts and novices. Experts would be more confident about central
aspects that are clear enough in the situation at hand, and their uncertainty
markers would be mainly related to aspects that are truly underspecified by the
situational parameters. Novices, in contrast, may be uncertain about any aspect
of the situation, or they may confidently misrepresent certain aspects due to
lack of knowledge about the potential complications and subtleties. This sec-
tion reviews the kinds of linguistic features that may be relevant in this regard –
those aspects of language that reflect confidence and certainty in a speaker.
In Chapter 3, we looked at Systemic Functional Grammar (SFG) as
a framework that distinguishes between different levels of analysis: the idea-
tional level (for content), the interpersonal level (for aspects pertaining to the
speaker and hearer), and the textual level (for discourse structure) (Halliday &
Matthiessen, 2014). For the purposes of current concerns, it is the interpersonal
level that is most relevant, as this is where we see the speaker’s ‘intrusion’ (in
Hallidayan terms) into the situation. This level of analysis therefore highlights

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 133

various ways in which a speaker can assess their own certainty about the
represented facts.
Neutral and undisputed poles of certainty are represented as clear positive
and negative statements, such as those in Examples 5.22 and 5.23. The speaker
is confident about the future, and can make a clear prediction about the rain the
following Tuesday. In SFG terminology, will and won’t are temporal operators
indicating the future, with either positive (Example 5.22) or negative (Example
5.23) polarity. The following examples then add a degree of indeterminacy,
ranging from low confidence (Example 5.24 positive, and 5.25 negative) via
median (Examples 5.26 and 5.27) to high (Examples 5.28 and 5.29):

Example 5.22 It will rain on Tuesday.

Example 5.23 It won’t rain on Tuesday.
Example 5.24 It can rain on Tuesday.
Example 5.25 It needn’t rain on Tuesday.
Example 5.26 It should rain on Tuesday.
Example 5.27 It shouldn’t rain on Tuesday.
Example 5.28 It ought to rain on Tuesday.
Example 5.29 It can’t rain on Tuesday.
All of these variations tell us something about the speaker’s certainty that
this event will happen in the future. Events in the past and the present can be
expressed likewise, with different reasons for the represented level of con-
fidence. Whereas nobody can be entirely certain about the future (because it
simply hasn’t happened yet), lack of confidence about something in the present
or past would be related to lack of specific knowledge. Examples 5.30–5.33
show some of the options for the present tense. Note that the modal will can
either express clear certainty (positive polarity) for the future (Example 5.22),
or high probability for the present (Example 5.31). Context is needed to
determine which of these options is intended by the speaker.

Example 5.30 This is/is not the right solution.

Example 5.31 This must be/will be/may be/should be/can’t be the right
solution.
Example 5.32 This is possibly/probably/certainly/probably not the right
solution.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 134 Identifying Cognitive Depth

Example 5.33 This will probably be the right solution.

If we don’t know for certain whether something is, was, or will be true, we
can express the level of certainty by using suitable modal operators. We might
use variations of the modal Finite as in Examples 5.22–5.29 and Example 5.31,
or modal adjuncts as in Example 5.32. Example 5.33 demonstrates that these
two distinct grammatical options can be combined, making the element of
indeterminacy more transparent.
There are further ways of expressing the speaker’s stance, beyond these
options, that are directly related to the interpersonal level of analysis in SFG.
Certainty can also be expressed as an attribute of the speaker as in Example
5.34, as an expression of mental processes as in Examples 5.35 and 5.36, or
through nominalisations of such processes as in Example 5.37. With options
such as these, the English lexicon contains a wide range of opportunities to
express degree of determinacy.
Example 5.34 I am sure/certain/convinced that this is the right solution.
Example 5.35 I expect/believe/assume that this is the right solution.
Example 5.36 I wonder/doubt that this is the right solution.
Example 5.37 My expectation/conviction/hunch is that this is the right
solution.
Analysis of certainty in CODA should therefore address a range of ways of
expressing indeterminacy: modal Finites and modal adjuncts as the main
grammatical indicators, plus further lexical items and constructions that
express related notions.
It is worth noting that the same grammatical options can also express other
concepts, such as obligation or inclination, in addition to probability. Consider
the following:
Example 5.38 Should I write a book?
Example 5.39 We ought to be on our way.
Example 5.40 Mary will help.
As Halliday and Matthiessen (2014) point out, Example 5.40 is on the
surface a statement of positive polarity about an event in the future – but if
Mary is listening, it will now be very difficult for her to refuse to help. In this
sense, the statement can express a degree of obligation rather than mere
certainty. In spite of their clear modal function, modal verbs (like any other
linguistic feature) therefore need to be interpreted in light of their pragmatic
meaning in context.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 135

Uncertainty can furthermore be conveyed through hesitation markers,

hedges, and other tell-tale aspects that may make speakers seem not entirely
sure of themselves or of what they are saying. Hesitation markers include non-
lexical fillers such as uh, uhm, um, er, and ah, as well as lexical devices such as
like, okay, right, and you know, which are commonly perceived as semantically
empty when they’re used as fillers. Hedges are similar in that they can also fill
gaps of speech, but they can also indicate vagueness, for instance sort of, kind
of, technically, somewhat, I think, though, well, and various others (e.g.
Culpeper & Kytö, 2010).
Together with pauses in the speech flow and delays in starting an utterance,
all of these features indicate fluency interruptions of some sort, and are there-
fore perceived as indicating some kind of hesitation. There may be various
reasons for this. Lack of confidence about the topic at hand or about the
appropriateness of the choice of words are major possibilities, which is why
it makes sense to consider these features as potential indicators of uncertainty.
However, there are plenty of other possible reasons for their occurrence. For
instance, people may simply be considerate of their communication partner’s
concerns; if they are talking about a sensitive issue, they may want to package
their message carefully, not wanting to sound overwhelming or over-assertive.
Consider the following examples, which can be interpreted as indicating
hesitation for different reasons:
Example 5.41 Um, I don’t really know what to do next, uh, (3) sort of
stuck now.
Example 5.42 Um, did you, uh, realise that you (3) sort of messed this up?
Example 5.41 could be part of a think-aloud protocol in which the speaker
verbalises a problem-solving process. Since there is no communicative task
here, there is no reason for social hedging. The hesitation markers in this
example (um, really, uh, sort of, 3-second pause) match well with the lexical
indications of uncertainty (don’t know what to do, stuck). Together, the indica-
tions for uncertainty in this particular phase of the problem-solving process are
therefore very strong and fairly unambiguous.
Example 5.42, in contrast, would more likely be heard in a communicative
context; the speaker is addressing someone else who has in their opinion done
something wrong. Alerting the person to the problem is socially problematic,
and could lead to a loss of face. As a result, the speaker hedges and tries to skirt
around the problem as much as possible; the pragmatic markers (um, uh, sort of,
3-second pause) accompany the explicit goal of making the person realise (did
you realise) that something went wrong. The use of colloquial language
(messed this up) further supports the speaker’s solidarity with the listener, in
the hope of avoiding this coming across as a formal reproach.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 136 Identifying Cognitive Depth

Further frequent reasons for displaying hesitation is cognitive load as well as

dispersion. Someone who is confronted with a complex task or with several
tasks at once will speak less fluently than otherwise. Lindsey et al. (1995)
designed a study to identify reasons for temporal non-fluency (pauses and
delayed speech onset). In particular, they looked at social reasons for non-
fluency as opposed to cognitive load. Their results showed that while people do
sometimes employ non-fluency markers for social purposes, cognitive load
plays a major role in challenging situations, for instance if the speaker needs to
convey several communicative goals at once. Non-fluency markers were con-
siderably reduced when people got a chance to prepare their messages in
advance, indicating that they weren’t simply strategic. With respect to
Example 5.42, it is therefore possible that the various hesitation markers reflect
the cognitive challenge of getting a communicatively problematic thought
across, along with the possibility of strategic hedging in order to avoid over-
whelming the listener.
Generally, we may note that hesitation markers, hedges, pauses, and the like
are typically indicators of some kind of problem. Something is there that slows
the speaker down, or makes them formulate their messages with caution, in
a less than assertive style. Further aspects of the linguistic data, or of the
context, typically provide indications as to the nature of the problem. This
can be uncertainty about the task at hand, or about the associated thoughts, or
about suitable words, as well as cognitive load and communicative concerns
about the relationship to the interaction partner.
Finally, the tone of voice can reveal the speaker’s degree of certainty in
various ways, ranging from individual variability in assertiveness to sys-
tematic differences in tone patterns. Because speakers differ a lot in the
habitual degree of forcefulness in their voice, it will often be difficult to
determine a specific degree of certainty in a particular situation. Differences
can often be subtle, and will be a matter of degree, relative to other situa-
tions. However, Halliday and Matthiessen (2014) suggest that patterns of
falling and rising tone can reveal degrees of certainty across individuals,
with a falling tone indicating more certainty than a rising tone. For instance,
questions are typically associated with a rising tone, and this would reflect
the highest degree of uncertainty: an explicit request to learn about some
facts which the speaker is not certain about (except in rhetorical or educa-
tional questions). Combinations of rising and falling tones, according to
Halliday and Matthiessen (2014), indicate various intermediate degrees of
certainty, such as ‘seems to be certain but isn’t’ with a falling–rising pattern,
and ‘seems not to be certain but is’ with a rising–falling pattern.
However, even though these patterns are presented in SFG as part of the
grammar, there are huge individual and sociolectal variations in this regard.
In certain dialects or sociolects, it is common to end statements with

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 137

a rising tone, simply as a matter of convention, rather than an expression of

uncertainty (Ladd, 1996). This is known as uptalk, and is seen as associated
particularly with female speakers in North America, Australia, and New
Zealand (although the actual picture can be a bit more complex than this;
see, e.g., Warren, 2005). It stands to reason that a rising tone in conversa-
tion invites the listener to agree with a statement and thus enhances
interaction; this would then be a dialogue function rather than a marker
of uncertainty.
To conclude, language is rich with resources for expressing levels of
certainty in many ways. Expressions of uncertainty are often very prominent
in language, and may intuitively seem worth focusing on in analysis.
However, it is critical to distinguish between different levels and sources
for uncertainty, not all of which may be directly relevant to the goals of
a research study. We will now turn to a CODA study in which certainty was
targeted directly.

5.2.3 Certainty and Expertise: A CODA Study

Since expertise is an important topic in cognitive science and language offers
many different ways of conveying certainty, the interplay of these two factors is
a fruitful target of analysis in CODA studies. To exemplify this, we will now
look at a study on the strategies people use to find their way in a complex
building, relative to how well they know the building (Tenbrink, Bergmann, &
Konieczny, 2011a).
Imagine it’s your first day at the university, and you’ve just managed to find
the room where you’re expected, somewhere on the first floor in the confusing
building complex. Now someone asks you to take them to the cafeteria. Or
something happens and you need to find a way out – quickly. Since you don’t
know your way around in the building, you may not be sure about either one of
these goals. However, chances are that you wouldn’t just give up. Somehow
you would manage to find the cafeteria, or an exit within reasonable distance –
even though you might not get there on the shortest possible route. But how do
you do this? What are the strategies that you might use to find your way around
in an unfamiliar building?
In our study, we confronted two different groups of participants with
just this kind of task, namely novices and experts. Novices were students
during their first week at university, and experts were employees with
offices in the building. We expected that these two groups would solve the
task in very different ways, and that the differences in expertise should be
reflected in their language use. All participants were asked the following
two questions, in balanced order (translated from the German original):

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 138 Identifying Cognitive Depth

• Imagine you have to leave this building as quickly as possible. Please explain
to me as well as you can how you would do that.
• Imagine you want to go to the cafeteria in this building. How do you get there
using the shortest route?
Following this verbal task, participants were asked if they were willing to do
more, and those who consented received the following instruction:
• Please walk to [the goal they described last: cafeteria/nearest exit] using the
shortest route. It is not important to us whether you walk the same route that
you just described or take a route that is even shorter. However, it is important
to us that you think aloud while you are walking. Please say everything that
crosses your mind.
This short think-aloud description, without a chance to practise, was used
to keep things simple. Participants had not been given any prior warning
about the study and took part spontaneously on an entirely voluntary basis;
therefore, we did not wish to use more of their time than absolutely
necessary. With a more complex study design where participants are pre-
pared to spend some time in the lab, it is useful to elaborate on the think-
aloud procedure and let participants practise before doing the actual task
(see Chapter 8). For the purposes of this wayfinding study, we hoped that
the task would be simple enough for the participants to verbalise without
further elaboration. And indeed, the verbalisations were adequate for the
purposes of our study.
In this way, we collected two types of verbal data (two planning descriptions per
participant, plus an optional think-aloud protocol) along with associated beha-
vioural data for each data set, that is, the route that participants chose in each case.
The route choices revealed participants’ performance in the sense of route effi-
ciency, whereas the verbal data reflected their certainty in relation to the task.
Results showed, unsurprisingly, that experts found shorter routes than
novices did (see Figure 5.1), and they also assessed their own knowledge to
be higher than novices did. Also, novices sometimes did not verbalise any
specific route at all, but referred to generic wayfinding strategies instead. These
results confirmed that linguistic features in the language data sets could indeed
be traced back to differences in certainty about the routes involved, because
novices were unable to identify the shortest route.
The linguistic analysis was qualitative as well as quantitative, and distin-
guished between the two types of data sets (planning and think-aloud) as well
as between experts and novices. For the quantitative analysis, we counted a range
of operationalised features in the language data, which were motivated in part by
the previous literature and in part by inspection of the data (as it is not always
possible to predict the kinds of features speakers will employ in free language
production). Our operationalised features included uncertainty markers of
a generic kind, as well as more scenario-specific indicators of different degrees

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 MAIN B1050 B1060

B1070
B1071
ENTRANCE
MAIN
STAIRCASE

B1320
& B1170 B1

B1160

B1171
B1330

B1420

B1215 B1400
B1210 B1430
&
B1202

B1440

A1120 A1110 A1101

B1450

A1100
MINOR
B1212 B1410
ENTRANCE & B1216 B1460
B1211
A1160
&

A1150
A1161

B1470
Turm E

A
A
A1180 B1480
B1820

terms of use, available at [Link] [Link]

& B1810

A1250
A1240
A1240
A1230
A1220
B1760 &
START B1821 B1490
& B1811
B1822 B1500
&
B1510

Figure 5.1 Example for paths chosen by novices and experts (Tenbrink et al., 2011a).
Experts described a quick route to the nearest exit, while the novice’s path leads to the main entrance via the main staircase.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
 140 Identifying Cognitive Depth

of spatial knowledge, namely concrete spatial elements (such as specific refer-

ences to landmarks, start and endpoint, distances, changes of direction, etc.) as
well as markers of orientation (such as public orientation aids including signage,
verbs of orientation including look out for and search, etc.). Generic uncertainty
markers, in this study, fell into the following categories:
• vagueness: words that include the morpheme some: somehow, something,
somewhere, sometime, etc.
• restarts: incomplete words or sentences followed by a fresh start
• referential problems (indicated by specific markers in German)
• modal verbs: could, may, might, should, would, etc.
• hedges: in a sense, probably, possibly, maybe, etc.
• mental processes/stance: believe, think, know, etc.
The quantitative results revealed very clear patterns, indicating how exper-
tise is reflected in language in this scenario. Experts produced more
concrete spatial elements, but fewer orientation and uncertainty markers,
than novices. This confirms that the experts had more specific spatial
knowledge that they could rely on during wayfinding, that they had less
need for orientation, and less reason to express uncertainty. Furthermore,
planning protocols included more concrete spatial elements than think-
aloud data, associated with more uncertainty. However, there was no
difference with respect to orientation markers. Thus, when planning
ahead, there was more need to verbalise specific spatial knowledge than
when actually navigating in the environment, where concrete spatial ele-
ments were directly visible. The higher amount of uncertainty in the
planning process reflects the fact that the environment itself provided
clues that reduced uncertainty and supported wayfinding.
In addition, our qualitative analysis of the protocols’ content revealed
a range of different strategies that included the above markers to different
degrees. People with very little knowledge of the building typically relied on
basic generic strategies that could be used anywhere, such as looking for signs
or asking people. Such descriptions do not contain any concrete spatial ele-
ments, but could include orientation and uncertainty markers. People with
limited knowledge of the building referred to environment-specific elements,
such as the signs above doors or the nearest staircase. People with somewhat
more knowledge would refer to central elements of the building such as the
main staircase, adding a limited number of concrete spatial elements to the
description. This central point–oriented strategy is known from previous lit-
erature to be common in wayfinding tasks in unfamiliar environments: if you
don’t know much, orient towards landmarks of high saliency that you can
always come back to. With even more specific knowledge of the building, our
participants were finally able to employ turn-by-turn directions of the kind one
would expect in familiar environments: turn right at the next corner, take the

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 5.2 Certainty 141

staircase down, it will be on your left. Such descriptions were based on fairly
elaborate knowledge of the building, and accordingly contained plenty of
concrete spatial elements, and few uncertainty or orientation markers.
Naturally, they were more frequently produced by experts than by novices.
In sum, our study in Tenbrink et al. (2011a) identified various cognitive
strategies that participants employed in a situation of potential uncertainty,
along with a range of verbal devices that revealed the degree of confidence. The
speakers’ language use reflected aspects of their minds (their expertise) in
systematic but subtle ways, not by explicit statements declaring lack of knowl-
edge, but by signalling uncertainty and vagueness through the choice of words
and use of suitable spatial strategies.

This chapter has highlighted two major ways in which cognitive depth is
manifest in the way we speak. On the one hand, differences in granularity
levels reveal the extent to which a speaker conceptualises elaborate details and
fine-grained aspects of a situation, as opposed to employing a more abstract and
general perspective. On the other hand, cognitive depth can mean depth of
knowledge based on expertise. Lack of confidence, related to extensive knowl-
edge or other factors including indeterminacy of the scenario itself, is regularly
expressed in verbalisations by various markers of uncertainty.
These two aspects of cognitive depth, granularity and certainty, are not
directly interdependent, but interrelate systematically. Experts who are certain
of a particular fact are normally able to describe this fact either in much detail,
or in abstract terms (e.g. as a ‘chunk’, a familiar pattern expressed as a whole),
depending on context. Similarly, uncertainty can be expressed either in relation
to general ideas, or in relation to fine details as the speaker zooms in to the issue
at stake. In this sense, looking at the two aspects of cognitive depth jointly may
be highly revealing to the analyst.

Downloaded from [Link] Nottingham Trent University, on 25 Jan 2020 at 06:23:55, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 6 Identifying Cognitive Constructiveness

This chapter will address inference and transformation – two processes of

human cognition that are often involved in problem-solving tasks, and which
may also be part of mental representations to some extent. Both of them are
ways in which the human mind creatively constructs reality, rather than simply
observing and representing objective facts in the brain. Inference involves taking
observable facts and combining them with further knowledge or assumptions, in
order to come to new insights or conclusions that aren’t directly observable.
Transformation, on the other hand, involves taking observable facts or objects
and turning them into something different, something that isn’t yet there but that
can be accomplished using available tools and operators. The following sections
will look at each of these processes of cognitive constructiveness in turn.

6.1 Inference
Imagine you read the following beginning of a story:
Example 6.1 Helena lived in a farmhouse with her parents. She didn’t have
many friends, and was often bored.
A simple and easy-to-read text such as this one immediately invites you to
make a range of assumptions. Readers may differ in the extent to which they
jump to conclusions, but at least the following three inferences are fairly likely:
• This story will be about Helena as the main character.
• Helena is a child.
• The farmhouse is located in a rural area, definitely not in a city.
How do we know all this? None of it is directly mentioned in the two sentences of
Example 6.1. In fact, all of these assumptions could be wrong. The story could be
about somebody else, with Helena introduced as a minor character for reasons that
will become apparent later. Helena could be grown up and still living with her
parents in the farmhouse. And there’s nothing to prevent farmhouses being located
in cities, even though it does not seem likely – the city might have grown around the
house over centuries. So, all of the inferences that the reader makes may be
unfounded and proven wrong as the story unfolds.

142

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 143

But how do we get to these inferences in the first place? Why do we make
assumptions such as these, and what are the principles behind this? For this
particular example, we can note that the reader builds on a number of conven-
tions as well as background knowledge. Since the style of the two sentences is
relatively simple, readers will naturally assume that the story follows
a conventional pattern in which the main character is introduced up front.
The easy-to-read style also supports the assumption that Helena is a child, as
this could be the start of a story for children. This is reinforced by the mention
of parents, friends, and bored – familiar elements in a child’s world, even
though they are by no means restricted to children. Since children’s stories are
conventionally told around such elements, the semantic field of a child’s life is
evoked, and supports the idea of a child’s story in the reader. Finally, farm-
houses are typically (though not inevitably) located in rural areas, and the
mention of boredom fits very well with the lack of busy urban life.
Thus, there is an abundance of associations that jointly support the three
assumptions above. Not every reader may have exactly the same associations,
but as long as the cultural background is shared, it is likely that inferences are
made, to some extent, in fairly similar ways.
Inferences such as these happen all the time, and for the most part we’re not
aware of this. We may read a story and build up a mental image of the
represented world without ever realising that half of our insights are mere
assumptions. With stories, this is usually not a problem since we’re only
a passive recipient, we aren’t supposed to act on what we read, and any
erroneous inferences will be corrected as the story unfolds. However, in other
scenarios, inferences can be quite serious and affect future actions in non-trivial
ways.
Consider the two main areas that we’re concerned with in this book –
problem-solving tasks and mental representations. When we solve problems,
inferences are often crucial because not all facts may be explicitly represented.
Indeed, if all the facts were clearly described, there may not be much of
a problem left to solve! So problem solving is very much about having suitable
thoughts about the information at hand, so that solutions can be generated. And
concerning mental representations, even when we merely describe what we
observe, some aspects may well be based on inferences rather than objectively
observable facts. For instance, if part of an object is hidden, we may just guess
what the hidden part looks like – and we will often be right, since we’re relying
on common sense and relevant world knowledge.
Clearly, then, inferences are frequent in our lives in the way we think.
Focusing on inferences when using CODA can provide a host of insights
relevant for the research question at hand. In the following, we will first explore
cognitive processes of inference (Section 6.1.1), and then look at inference
processes specifically in the area of problem solving (Section 6.1.2). Section

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 144 Identifying Cognitive Constructiveness

6.1.3 addresses the role of inferences in language use, and Section 6.1.4 gives an
example of a CODA study that addresses inferences through linguistic analysis.

6.1.1 Cognitive Processes of Inference

In our introductory Example 6.1, one of the assumptions involved expansion
from previous experience or background knowledge, namely that farmhouses
are normally found in rural areas. Generalisations such as these are very
common, and they regularly lead to inferences that may or may not be true.
Here, the generalisation that farmhouses are normally found in rural areas
suggests to the reader that the location of this particular farmhouse is in
a rural area, too. Readers may quickly forget that this is only an inference
from a generalisation, rather than explicit information from the text itself.
Typically, after little more than a minute, readers are unable to distinguish
between the exact words they read and a paraphrase (Sachs, 1974). Based on
rapid comprehension processes, inferences are quickly absorbed and taken for
granted.
Overgeneralisations happen in other contexts as well; they are frequent
processes in everyday thinking (Sternberg & Ben-Zeev, 2001). When con-
fronted with a new challenge, people often recall previous experiences of
a similar nature, and then attempt to deal with the new challenge in a similar
way. In this process, one might go too far and fail to realise that the new
situation might need a different approach. Or one might become confused, if
the new challenge comes with unfamiliar aspects. Consider the following
sample of a think-aloud protocol recorded in an origami paper-folding study
(Tenbrink & Taylor, 2015):
Example 6.2 I usually look at pictures only but sometimes because this is
a more tradition- a more modern way of teaching origami as
opposed to a normal diagram, I took a look, at first I was
disoriented once again so what I ended up doing was I looked
back at the original thing to see that we had to fold it back to the
original position.
Here, the (male) participant refers to his previous experience with origami.
According to this protocol he is used to doing origami folding tasks on the basis
of pictures alone, without the help of accompanying verbal descriptions. What
he saw in this task, however, was a different kind of picture, which did not
contain all the information he needed to solve the task; therefore, reading the
instructions was essential – contrary to his previous experience. This explains
the disorientation in this instance. Relying on previous experience was not
sufficient; this task had to be solved differently.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 145

Why do we rely on previous experience so much – why do we base our

actions and decisions on inferences gained from other contexts? Gentner
(2010) suggests that this ability is one of the basic prerequisites for thinking
and learning throughout our lives. We experience something specific and are
then immediately able to transfer the experience to other situations by analogy.
All we need is some kind of parallel with the new situation, some common
structure that we can recognise, so that we can do the transfer. After identifying
similarities and differences between situations, we try to map their structures
by aligning them as far as possible. As part of this process, we may fill some
gaps by transferring some aspects of one situation to the other, without requir-
ing all the details to be specified explicitly. And this can mean that we make
inferences that we’re not aware of, since much of this process is habitual and
automatic.
For instance, Gentner et al. (2009) showed children two models of
a skyscraper. One of them was stabilised by a diagonal brace, whereas the
other only had a horizontal brace that provided no stability. Once the children
had understood the difference between the models, they were able to transfer
the insight to a different model building. This transfer was facilitated if the two
models shown to the children were otherwise similar. If they were different in
further respects (other than the brace), the children were less able to identify the
decisive difference between the two models that led to the increase in stability.
The transfer to a new building requires abstraction and generalisation abilities,
as well as inferences as to how the brace might work with the new building.
Remarkably, even young children were able to achieve this transfer following
very brief training, involving just one pair of models for demonstration. And
they could transfer the insight to a different-looking building, assuming impli-
citly that the principle would still work – as it did.
Transferring insights from previous experience is thus crucial for learn-
ing, and for understanding the world. The ability leads to generalisations of
individual situations, and likewise to specific solutions for new kinds of
challenges – as long as they are similar enough, in some structural way, to
a familiar kind of challenge. We do this all the time, it is central to how we
think. The world would be far too complex for us to cope with if we were
unable to categorise, generalise, transfer, and abstract from insignificant
differences.
In fact, the same principle is at work for acquiring concepts. Children start by
learning the names of categories such as chair or dog, often based on a single
instance. They might use the new word right away when they see another one
that seems similar. And the more exemplars of a category they see, the more
accurate they will be in transferring the concept to a new exemplar. In the
meantime, children often overgeneralise – they may call a dog a cat, and a sofa
a bed. This reflects the thinking behind concept acquisition: cats look similar

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 146 Identifying Cognitive Constructiveness

enough to dogs that they may be classified in the same category at a particular
stage of development (Tomasello, 2003). And sofas can be used for sitting and
lying on, just as beds. However, as the child sees and hears more about the
world, concept classifications will become more stable.
As adults, the same principle is still at work. We learn new words and
concepts all the time, for instance from context when we read or listen to the
radio, or as new technology introduces novelties. For instance, we learned the
meanings of iPad and virtual reality along with their functions as technology
developed. In some instances, just seeing and understanding one exemplar will
be sufficient to acquire the new concept, whereas other concepts may be too
abstract and complicated to grasp, and we may be unsure of them for a long
time. In any case, the underlying process involves understanding which aspects
of a particular instance are relevant and can be transferred to another instance
by analogy, allowing for the inference that the next item that looks like an iPad
is, in fact, also an iPad, with similar features.
Although central to human cognition in many respects, the ability to make
inferences from past experience comes with a few problems. Generalising too
much, filling in too many gaps by analogy, means jumping to conclusions. We
could easily expect the features of an iPad in another item that is in fact an iPod,
resulting in some unwelcome surprises. Or, to take another example: if three of
our four best friends have long black hair, and two people who we don’t like
have short hair, we may start to assume that people with long black hair are
generally nicer. Logically, this can only make sense if there is some character
trait or cultural preference directly associated with this hairstyle. Even if there
was, taking the inference further to the point of assuming that short-haired
people are all awful would definitely be disastrous. There are so many other
factors to take into account, and a stereotype of this sort could have all sorts of
negative consequences.
Also, over-reliance on analogy may under certain circumstances reduce
creative thinking and cloud judgement. For instance, Sternberg and Ben-Zeev
(2001) reported that teachers often encourage children to establish a pattern for
solving a mathematical task in a particular way. With word problems such as
‘Tom has 5 apples, Jerry takes away 3. How many apples are left?’ (Sternberg
& Ben-Zeev, 2001:163), they may be taught to use the word left as a cue to
perform a subtraction, as necessary for this task. However, if a child takes this
analogy-based strategy too far, they might assume that subtractions are also
required for other tasks involving the word left – including ‘Tom sits to the left
of Jerry.’ Then, the established task-solving procedure actually keeps the child
from processing a new task properly, and hampers identifying suitable solution
paths for it.
We can conclude that inferences happen frequently in our everyday thinking,
allowing us to think and act quickly based on our life experience. The ability to

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 147

transfer from something known to something (slightly) new is useful, and it

works well as long as there are sufficient parallels to allow for a suitable
analogy. The more inconsistencies there are between situations, or the less
complete information is available, the more we will have to rely on inferences
to fill in the gaps. And then, inevitably, overgeneralisations and erroneous
assumptions become more frequent.

6.1.2 Inferences in Problem-Solving Tasks

In problem-solving situations, or when we are confronted with a complex
decision to be made, one of the challenges that people are typically confronted
with concerns the lack of a clear strategy or basis for acting. The available
information is often either incomplete in some way or too complex, and
previous experience may not allow for a direct and complete transfer without
considerable gaps. This is an essential aspect of why there’s a problem or why
making a decision can be really hard. How do we cope with this kind of
situation? Would we try to analyse all aspects and details of the situation before
proceeding, or what else can we do?
Famously, Simon (1956) suggested that we use cognitive ‘shortcuts’ to
accomplish our goals in an efficient way when confronted with complex
challenges. Our rational minds are limited, and we cannot account for all
relevant aspects of complicated situations. Simon (1956) refers to this fact as
bounded rationality; our minds have very clear boundaries, and they do not
work quite as fast as modern computers. What makes us superior to computers,
however, is our efficient way of dealing with information gaps or overload,
which enables us to function well in an infinitely complex world. That is, we
use a range of useful heuristics and strategies that we establish throughout our
lives and then have at our disposal for new situations.
For instance, when we find our way to a new location in a familiar complex
urban environment, we do not need to plan the entire route to perfection in our
head before setting off. The spatial information that we have in our heads about
streets and landmarks in the city is inevitably incomplete (as we never have
access to all the facts), and it may at the same time be too complex to account
for every aspect if we know the urban environment well. Instead of attempting
to optimise the route in advance, therefore, we just estimate the rough direction
of the goal location and start moving towards it, using suitable roads and turns
as we go (Hölscher et al., 2011). This heuristic works really well for most
wayfinding situations in familiar environments, even though there may be
occasional challenges such as blocked roads.
Such heuristics and strategies are employed in more or less conscious ways.
In challenging situations, we typically need to consider a high number of
factors, but we may not be aware of the kinds of cognitive shortcuts that we

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 148 Identifying Cognitive Constructiveness

employ when we get to a solution. Due to our bounded rationality, we cannot

take everything into account. But we do consider a relevant subset of the
aspects involved, such as our knowledge of the general goal direction, and
use this as the basis for a fast and efficient strategy.
These procedures often imply a range of generalised underlying assump-
tions. To return to the example cited in the previous section, the children
trained in Gentner’s (2009) study generalised from the single example they
saw that diagonal braces stabilise buildings. From this experience, they
could apply a diagonal brace to a different building, based on the (correct)
assumption that the spatial arrangement of the brace mattered. There
clearly wasn’t enough evidence available to show that this was indeed
the case across many kinds of buildings; the children did not have access
to any static facts or calculations that would enable them to work out the
optimal positioning and size required of the brace in a new building.
However, the cognitive shortcut of focusing on the diagonal arrangement
worked well enough for the purposes at hand.
In this way, cognitive shortcuts mean that we attach our assumptions from
previous experience to the currently available information, in order to reach
a conclusion that will typically correspond to the actual state of affairs closely
enough to lead to the desired effect. Following Toulmin’s (1958) terminology
of human reasoning processes, we use the available information as data, and
connect it with some kinds of warrants (principles or rules – including cogni-
tive heuristics) in order to come up with claims (inferences or conclusions). In
Toulmin’s theory this then leads to the formulation of arguments in order to
persuade other people of the claims made.
Even without considering arguments, the notions of data, warrants, and
claims are very useful as cognitive components that can be identified in
language data in the realm of CODA studies. They reflect the cognitive process
of reasoning on the basis of background knowledge and cognitive heuristics,
and they can be applied across many different kinds of scenarios.
Verbalisations are often very useful resources to identify corresponding evi-
dence as part of inference processes.
For instance, in one study, expert geologists and undergraduate students
were asked to identify plausible geological structures from artificial rock out-
crops (Kastens, Agrawal, & Liben, 2009). The students combined the available
evidence with their background knowledge to form hypotheses that could be
valid or not. This study showed that well-founded reasoning processes, as
represented in language, were associated with valid claims, and also that
experts produced far more evidence-supported claims than non-experts. In
contrast, students produced more unsupported claims, that is, hypotheses that
were not based on data and warrants. Further, some of the valid inferences
(claims) required geological knowledge as warrants during the interpretation

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 149

process of the evidence (data); as predicted, students were rarely able to refer to
this kind of warrant. Altogether, the study provided extensive evidence con-
cerning the relevance of inferences for reasoning processes in a complex task
situation involving incomplete information.
Across many kinds of scenarios, it can be very enlightening to focus directly
on the claims and inferences humans make based on the data available.
However, perhaps the most interesting part of a reasoning process is the
intermediate step: augmenting the given data using some kind of warrant.
Such warrants may often be related to mental imagery (Johnson-Laird, 1983).
For instance, if we think of our spatial surroundings, our mind quickly fills in
any gaps concerning aspects or areas that we haven’t actually observed, or
don’t remember in detail. We only have the most essential information avail-
able, but we may well be able to visualise some aspects that we don’t have
direct evidence for.
This kind of filled-in information comes from our previous experience,
which we transfer to our mental image of the spatial environment.
Identifying such processes through linguistic data analysis can be highly
enlightening, as it can reveal where and how inferences are warranted or
based on overgeneralisation of previous experience. Then, these inferences
can be evaluated against actual facts, as they may be correct or incorrect. Also,
they may be more or less detailed, revealing different degrees of imagination
involved in the thought process.
In sum, problem-solving processes are based on inferences of various kinds.
Making the right inferences from the available information in combination with
prior knowledge may often be the main challenge when solving a problem.
Therefore, a focus on inferences in analysis can be highly revealing.

6.1.3 The Role of Inferences in Language Use

So far in this chapter, we have looked at various ways in which we draw on
incomplete information in order to learn language, to act, to make decisions, to
solve problems, or generally to understand the world. The world is far too
complex for us to understand in full detail – but we can manage quite well based
on generalised and vague knowledge as well as some practical heuristics.
The same is true for language use. Communication is never perfect and
spelled out to the last detail. In dialogue, speakers don’t ever reach
complete understanding, but they are satisfied if they can understand
each other ‘well enough for current purposes’ (Clark, 1996:226). For
instance, if Anne encourages Greg to sit down she might be alluding to
a specific chair to sit on, but she won’t need to say this explicitly unless
it’s really important that he doesn’t sit down somewhere else. Indeed, the
more important it is to choose the intended place to sit, the more explicitly

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 150 Identifying Cognitive Constructiveness

her intention will be communicated. If it doesn’t matter at all, Anne will

just say sit down. If she intends Greg to sit in a specific area, a vague
gesture might accompany the verbal encouragement. If she wants him to
sit on a specific chair, she might touch the chair and point him directly to
it. And if she wants to make her point in a decisive way, she will
accompany this non-verbal gesture by a clearly pronounced . . . here.
In general, instead of telling each other everything explicitly, we draw
on a number of communicative principles such as those formulated
famously by Grice (1975). Grice suggested that we normally aim to be
truthful, orderly, informative enough for the topic at hand, and relevant to
the context of the situation. However, we might sometimes break those
‘maxims’ for a specific purpose, or it might seem that we do. If this is the
case, this communicates to the listener that there is some additional
meaning that needs to be inferred.

Example 6.3 George: What time is the meeting?

Jenny: You’ll have to hurry up after class.
In Example 6.3, George asks for a specific clock time, but Jenny responds in
a way that might seem to be irrelevant to the question at first sight. She does not
provide a specific time as asked, but instead informs him about a certain con-
sequence for George’s future action. In a way, Jenny goes beyond the question.
Rather than merely sticking to the conventional question–answer pattern, she
makes some inferences about George’s plans: he’ll want to attend the meeting;
but he is in class directly before; in order to get there in time he will have to hurry.
Then she informs him about the outcome of this train of thought.
George, on the other hand, will have to retrace Jenny’s thoughts in order to
derive the answer to his question. He needs to infer that Jenny is aware of his
class, and that he actually intends to go to the meeting; based on this inference,
George can finally infer the answer to his question: that the meeting takes place
at a time shortly after his class ends. All this happens quickly and normally
without the speakers’ awareness. Successful communication can be hard cog-
nitive work, but we’re rarely bothered by it because we are very much used to
the principles of conversation that are normal for our culture.
One reason why we communicate in this way is because we aim to be
economic, or to follow a ‘principle of least communicative effort’ (Clark,
1996). If speakers share a lot of background knowledge, they do not need to
invest much effort into spelling everything out. Allusions and minimal pointers
can be enough to understand each other easily. What guides understanding in
the absence of explicit communication is our expectation of relevance. We can
understand minimal contributions mostly because they happen in a certain
context and are relevant to it.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 151

George can make the correct inferences on hearing Jenny’s utterance in

Example 6.3 because he has just asked about the meeting. In a different context,
the utterance you’ll have to hurry up after class would not tell anybody the time
of the meeting. Thus, somebody who does not share the knowledge about
George’s class time would learn nothing from this statement.
Understanding utterances out of a specific context is much harder and leads
to misinterpretations and miscommunication far more easily. Compare
Examples 6.4 and 6.5. In each case, the initial statement provides a context
that we can use to understand the following (identical) evaluation of a film’s
story.
Example 6.4 Kate: This film had a deep message. This guy had cheated
everyone, and he was surrounded by friends who accepted all of
that. How nice to be forgiven!
Example 6.5 Kate: I really didn’t like this film. This guy had cheated everyone,
and he was surrounded by friends who accepted all of that. How
nice to be forgiven!
Example 6.6 Tom: I agree with Kate – forgiveness is really important.
In Example 6.4, Kate liked the film and then elaborates the kind of deep
message she saw in it – namely how it is nice to be forgiven. In Example
6.5, the same elaboration is used to explain why Kate did not like the film.
The concluding statement, How nice to be forgiven, can now be under-
stood as sarcastic: the speaker does not believe that forgiveness can or
should happen this way, after cheating everyone. In Example 6.6, Tom
might have overheard the last sentence, without its context, so he infers
that Kate made a statement about the importance of forgiveness. While
this may be true in the context of Example 6.4, it certainly is a wrong
inference in Example 6.5.
Considering Tom’s and Kate’s utterances together, then, allows the
analyst to identify the source for misinterpretation, based on Tom’s infer-
ence. Tom lacks the context of how Kate actually judges the film, and
therefore cannot know whether Kate’s concluding statement is sincere or
sarcastic. He does not know, in effect, what the statement is about. He
infers that it must be about forgiveness, but in Example 6.5 at least it is
actually a statement about the film – namely, how it (in Kate’s opinion)
oversimplifies forgiveness.
As noted in Chapter 3, Sperber and Wilson (1986) suggested that commu-
nication is altogether dependent on relevance, and they built an entire theory
around this notion (relevance theory). They claimed that speakers always aim
towards maximal relevance, and that without this principle we would be
unable to interpret each other’s utterances at all. In effect, no utterance is

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 152 Identifying Cognitive Constructiveness

entirely context-free: everything we see or read is embedded in some context,

which enables us to make sense of it. And this sense-making process involves
inferences: both on the side of the speaker, whose contributions reveal what
they have inferred from the context, and on the side of the addressee, who
interprets the contribution based on their own inferences according to
context.
These insights are quite decisive when interpreting language in a discourse
analysis context, such as CODA. Analysts need to be aware of their own
inference processes and treat interpretations with caution. However, they can
also gain valuable insights about speakers’ inference processes by adopting
this kind of analysis perspective. As in the study by Kastens et al. (2009),
cited in the previous section, verbal protocols often reflect participants’
thought processes in problem-solving studies in such a way that inference
processes become apparent. Sometimes speakers will make their inferences
explicit in some way. In Example 6.7 the speaker explicitly derives a specific
inference (this bottle is heavy) based on general knowledge (green bottles are
heavy).
Example 6.7 Here is a green bottle. I know green bottles are heavier than
white bottles; therefore I expect this bottle to be quite heavy.
Example 6.8 Ah, here is a heavy bottle.
At other times, the inferences may be entirely implicit, as in Example 6.8: the
speaker does not spell out how they know that this bottle is heavy – it is simply
referred to as such. If the analyst knows that the speaker has not touched the
bottle, it is clear that this judgement must be based on an inference, such as the
one made explicit in Example 6.7.
Since speakers aim towards minimal effort and maximal relevance, it is
actually rather likely that they will only utter something minimal like Example
6.8 in an experimental setting, at least after some experience. When asked to
think aloud while solving a problem, speakers are not required or expected to
explain their inference processes to an addressee, and therefore will only say
what is relevant to themselves as they solve the problem. In Example 6.8 there
is no need for the speaker to say why they think the bottle is heavy. But the idea
that it is heavy is relevant enough for the speaker to say out loud, as a proper
part of their conscious thought process.
Along these lines, speakers will linguistically represent those aspects of
a perceived scene or problem-solving process that they judge to be important.
They may neglect things that they are unsure about unless these become
relevant for a given goal. As a result, linguistic descriptions reflect what is
prominent and relevant for the speaker while (or after) processing available and
inferred information. It is then up to the analyst to identify which parts of the

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.1 Inference 153

utterances are based on directly observable information, and which are inferred
(as in Example 6.8 if the speaker hasn’t actually touched the bottle). If inferred,
they may be either accurate or inaccurate, which can lead to interesting analysis
results. We will now look at a CODA study in which inferences played a major
role.

6.1.4 Inferences and Linguistic Analysis: A CODA Study

Imagine the following scenario. You’re a rescue worker, trained to find and help
people in need. A large building collapses, and you’re called on the scene to
search for survivors trapped underneath the debris. You can’t enter the scene
yourself, as the debris may collapse further, causing further harm. Instead, you
use a certain type of camera that has been developed for this kind of situation.
The camera works like an endoscope: you can guide it into the debris and it
sends pictures to your screen.
What you see on the screen is unfortunately very hard to understand and
interpret. The camera twists and turns, the scene is cluttered, things are upside
down or partially obscured. Detecting people in this challenging scenario can
be extremely difficult, and success will depend crucially on your ability to
make the right kinds of inferences. Your eyes will not tell you all the details –
your brain will have to fill in some gaps, make sense of what you see.
In a recent study (Tenbrink et al., 2016), we addressed this urban search-and-
rescue scenario to find out how much untrained people understand what they’re
seeing in such a scenario, and what this might depend on: is it just a matter of
having enough time for the search, or is it rather a matter of skill? We designed
a scenario that matched the described emergency in crucial respects but did not
require any technological knowledge or prior expertise. Rather than hiding
people under real debris, we hid small objects in a cluttered Styrofoam con-
struction that was made to look very similar to a (small-scale) collapsed room
(see Figure 6.1). This mock-up construction was filmed with a camera of the
kind used in rescue scenarios, moving slowly through the scene. Participants
were shown the film and asked to find all eight objects hidden in the debris, and
to memorise where they found them so that they could describe their position
later. Afterwards, they were asked what they remembered, and prompted to
describe where and how the objects were positioned. Also, they were asked to
describe the shape of the room if they could, and to draw a sketch of it together
with the objects that they remembered.
To measure the effects of time constraints, we used one condition in
which participants were under time pressure, and another where they had
as much time as they needed. To measure the effects of skill, our parti-
cipants did a spatial ability test that was matched to the kind of challenge
given in the study. To measure participants’ success in identifying objects,

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 154 Identifying Cognitive Constructiveness

Figure 6.1 Screenshot from the film in Tenbrink et al. (2016), showing a tilted
view with artificial horizon. Note the lamp on the right.

we counted the objects that participants detected during the film and
described later.
Further, we aimed to assess participants’ success in making sense of the
cluttered scene they saw. To operationalise this idea, we did three things. First,
we addressed the extent to which descriptions of the scene were coherent, in
that they represented the cluttered room as one consistent whole with
a structure of some kind, rather than individual disconnected bits and pieces.
Here, we found a wide range of variety. Some descriptions consistently started
from one particular object, and then related the following objects to its position.
Others related their object locations to the room area as a whole. In contrast,
some descriptions did not follow any consistent description strategy at all, but
rather reflected the confusion that participants clearly felt concerning this
complicated task.
Second, we looked at how specific descriptions were. A description like at
the wall could apply anywhere and did not count as specific, in contrast to more
precise spatial relations such as to the left of the tennis ball, or on the right side
of the room. And some objects were simply mentioned, without any further
specification. Third, we rated the coherence of room drawings, which

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 155

sometimes looked rather chaotic; this reflected the high degree of challenge in
making sense of the scene.
Comparing these various measures and relating them to each other led to
a number of insights. Most interestingly, people with higher spatial skills were
more able to describe the specific location of objects in the scene, and to draw
coherent sketches of the cluttered room and its contents. Unexpectedly, the time
constraint factor did not matter at all; the quality of descriptions and drawings
was not affected by introducing time pressure to the task.
It appears that spatial ability is a good skill to have when confronted with
a complex scene requiring a lot of inferences about how things add up to
a coherent whole. If you’re not good at this kind of spatial understanding,
you might stare at the pictures for a long time and still not be able to make sense
of it. You might detect more objects with time – which can be very useful – but
you might not be able to explain where they are, since you don’t understand the
overall configuration. This doesn’t mean that these skills can’t be acquired. All
our participants were inexperienced and untrained; they only differed with
respect to general spatial skills, not specific experience. In fact, it is well
known that specific cognitive abilities can be trained very effectively, and
there is no reason to assume that this wouldn’t also be the case for the skills
needed for search and rescue scenarios.
It can be concluded from this example study (Tenbrink et al., 2016) that the
analysis of inference in empirical research can be challenging, because people
won’t directly mark a thought as being inferred. However, there are ways of
operationalising analysis in such a way as to identify underlying inferences,
based on a targeted comparison between available information and conclusions
drawn by the participants.

6.2 Transformation
Many real-world tasks involve transformation of some kind: a child learns to tie
her shoes by transforming two ends of a string into a knot; a piece of paper can
be transformed into a flower or bird using origami folding procedures; pieces of
wood can be transformed into furniture items. Transformation can be action, as
part of a task or problem-solving process. But transformation can also happen
in the mind, as part of a mental process. For instance, if you look at a picture and
try to imagine what it would be like to be there and what you would like to do
once you’re there, you’re already going beyond the picture itself – your mind
takes you further towards a transformed alternative version of what you
actually see.
Transformation is thus another way in which our minds construct reality
beyond input. Inference, as explored in Section 6.1, means to ‘read
between the lines’, fill in the gaps of available or perceivable information

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 156 Identifying Cognitive Constructiveness

through background knowledge and logic. Transformation, on the other

hand, means to ‘go beyond the lines’ – to take the available or perceivable
information creatively towards something that could be there, but isn’t yet,
or isn’t there as such. Our minds are incredibly flexible in this regard, and
the world wouldn’t be the way it is today if they weren’t. Ultimately, the
ability to creatively transform existing things and ideas into new ones lies
at the heart of invention and improvement, or any kind of advancement
from the status quo.
Transformation can be entirely creative; this is the case whenever something
truly novel is created without a prior example. However, most transformation
processes are guided in some way, either by prior experience or training, or by
direct instruction. Transformation, in the sense considered here, is therefore not
identical to creativity, although an element of creativity certainly supports
transformative processes effectively.
What is happening in our minds during transformation, and how can we find
out? Depending on what is being transformed, the process will often include
a range of complex cognitive procedures. If we are confronted with a set of
pieces that we wish to transform into one larger item, we will need to find out
how specific parts can be assigned specific functions. In this way, a wooden
board might become an outer wall in a doll’s house (Gralla & Tenbrink, 2013).
We will also need to think about the goal state, to decide (or clarify) what the
result of the transformation should be. And we may need to anticipate manual
actions that affect objects in the desired way.
Language often plays a crucial role for transformation. Frequently, we will
be able to follow explicit instructions that tell us, more or less exactly, how the
transformation can be carried out and what steps will need to be taken. Also, we
may be able to describe the actions before we do them, or after having done
them. Language is a useful medium to represent our thoughts about the
transformation process, and it may sometimes be the main way to specify
them. People may think aloud while doing transformations, or report how
they did them, or simply represent the mental transformation itself in words –
for example, describing what they would do if they were inside the scene shown
in a picture (Cialone et al., 2018). Verbalised thought can represent a range of
crucial cognitive aspects that are involved in transformation. For this reason,
a focus on transformation in linguistic analysis can lead to exciting insights,
highlighting how the flexibility of human thought affects the real world and
interacts with it.
In the following, we will take a closer look at transformation processes from
a cognitive science point of view (Section 6.2.1), and then turn to two CODA
studies that explored different aspects of transformation: assigning functions to
objects when assembling a doll’s house without instructions (Gralla &
Tenbrink, 2013; Section 6.2.2), and the challenges for transformation when

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 157

following instructions in an origami paper-folding task (Tenbrink & Taylor,

2015; Section 6.2.3).

6.2.1 Insights on Transformation from Cognitive Science

Evidence for human ability to transform given input is abundant – it
happens at all levels of cognition, at all stages of development, with all
types of input. Transformation skills can be measured to some extent, using
tasks that specify precisely what type of transformation is expected. For
instance, to test the ability to mentally transform some visual input, people
may be shown a picture of an object to compare with another picture – for
instance a rotated version of the first (Smith & Dror, 2001). To successfully
compare the two, they need to mentally rotate the first object to see if it
matches the second one.
Mental rotation is one way to transform given input; the visual information
is changed towards a different orientation in order to accomplish another
cognitive task (the comparison with another object). Mental rotation tasks
have been used extensively to test various aspects associated with this type of
transformation. For instance, Shepard and Metzler (1971) found that the more
the angle of the rotated object departs from the original, the more time it takes
to mentally perform the transformation. And Mast and Kosslyn (2002) dis-
covered that mental rotation can help identify different interpretations of the
same picture. Because of insights such as these, the task designed by Shepard
and Metzler (1971) is frequently used to measure participants’ ability to do
mental rotations in their head, in order to compare this skill with performance
in some other task.
The study by Smith and Dror (2001) specifically looked at what happens
when we mentally rotate an object, and what this might depend on. Previous
literature had shown that mental rotation can be either holistic or piecemeal:
either we conceptualise the entire shape as a whole, or we look at specific parts
of it to perform the mental rotation. The authors speculated that this might
depend on whether the objects to be rotated were meaningful or not, so they
showed participants sketches of either meaningful objects (e.g. a ship or
a helicopter) or meaningless objects (similar size and style of line drawing,
but depicting no specific object). Their results showed that meaningless objects
seemed to be mentally rotated as a whole, in holistic fashion, whereas mean-
ingful objects were considered in terms of their specific parts. This reflects the
flexibility of mental transformation processes, which can be adapted to the
purposes at hand.
Mental rotation can be the basis of more complex imaginative processes, as
shown by an intriguing experiment by Finke, Pinker, and Farah (1989). Rather
than simply asking people to mentally rotate objects, they added further

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 158 Identifying Cognitive Constructiveness

instructions that could lead to new meanings, as in Example 6.9 (Finke et al.,
1989:62). The result of this particular mental transformation is the letter ‘T’.
All of these transformations happened in the mind only, but participants had
little trouble identifying the new meanings created in this way, reflecting the
versatility of human imagination.
Example 6.9 Imagine the number ‘7’. Make the diagonal line vertical. Move
the horizontal line down to the middle of the vertical line. Now
rotate the figure 90 degrees to the left.
While mental rotation is a typical example of transformation that has been
explored in much depth, our minds are capable of doing more complex
transformations – and we frequently need to do so. Taking the idea of mental
rotation to another level, Atit, Shipley, and Tikoff (2013) observed that tradi-
tional mental rotation tasks only involve rigid transformations, unlike many
tasks in the real world that involve non-rigid objects that can bend or break. In
their study, they showed participants three-dimensional pictures of bent sheets,
and asked them what the sheets would look like if they were unbent. Results
showed that performance varied according to how exactly the sheet was bent,
with some ways of bending harder than others. Also, interestingly, performance
in this particular task did not correlate with performance in a rigid mental
rotation task. This shows that mentally transforming non-rigid objects requires
different skills to mentally rotating rigid objects – and it indicates that focusing
research on rotation alone, as a prime example of mental transformation of
visual images, is not sufficient to account for human transformation abilities.
Outside the visual domain, transformation can take a wide range of different
forms. Any mathematical equation is a transformation that changes a sequence
of numbers to a different one that is mathematically related to the first. Many
transformation outcomes in the realm of numbers can be visualised as graphs or
diagrams. Interpreting such visualisations is then a matter of making sense of
the transformation.
In fact, while diagrams are useful ways to represent aspects of the world by
transforming them into a visual scheme, observers may often fail to identify
crucial information that the diagram is meant to display. Also, they may
misinterpret accidental features of a diagram as representing specific facts
about the displayed numbers. For instance, if some parts of a diagram use
salient bright colours while others are displayed in grey, observers may think
that the coloured ones are more important or decisive for their purposes. But
both of these are merely transformations of a particular aspect of reality into the
abstract schematic space of a diagram. Relevant aspects of the represented facts
will be carried over, such as quantity of a particular measure: if a number is
higher than others, the diagram will display it in some way, for instance by

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 159

a larger bar. However, other aspects will be neglected or may change – such as
colours.
As can be seen from this example, transformation is a matter of preserving
some aspects while changing others. The challenge, then, is to find out what
changes and what doesn’t. In a series of experiments that have become famous,
Piaget (1952) observed how children gradually learn about object permanence –
the fact that objects remain in the world even if you can’t see them. At first,
children immediately lose interest as soon as an object is hidden. Later, they
start looking for it, and thereby demonstrate that they know it still exists. But
they may still not be able to keep track of it if it changes location or undergoes
other transformations. Even later, the ability to keep track of quantity emerges.
For instance, if the same number of objects are placed with equal distance
between them in each of two rows, children may recognise that the number of
objects is the same in each row. But if the experimenter transforms the situation
slightly, and pushes the objects in one row closer together, the child (at a certain
stage of development) would think that the number has been reduced. The
transformation of the situation has led to confusion: the space covered by the
objects has been reduced by the transformation, but not the number of objects.
Similar to misinterpreting colours in a diagram, it is clearly not always easy to
see which aspects of reality change with transformation, and which remain
constant.
Most actions we undertake in the world are associated with some kind
of change or transformation. Whenever we do something with objects,
even just turn the page of a book, we have changed the shape of the
book ever so slightly – not necessarily permanently, but definitely relative
to how it looked before. When we turn around to look in the opposite
direction, we have transformed nothing else but our own view direction –
but as a result, the visual input we receive has changed completely.
Changes can be permanent or temporary, and transformational opportu-
nities are abundant. In this section, we have only touched the surface of
the wide range of aspects involved in these phenomena. Rather than
aiming to cover the field more exhaustively, we will now turn to two
specific case studies that exemplify how CODA can be used to address
research questions related to transformation.

6.2.2 Transforming Separate Objects into a Coherent Whole: Assembling

a Doll’s House without Instructions
Imagine you’ve just bought a dissembled set of wooden pieces on the flea
market, which you were told fit together to make a cupboard. Unfortunately,
there is no manual – you’ll have to figure it out yourself. How do you go about
this task? This kind of situation challenges you to a kind of transformation that

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 160 Identifying Cognitive Constructiveness

is considerably more complex than the ones discussed in Section 6.2.1. Rather
than changing just one object in a limited way, you’re tasked with creating new
meaning by assembling an entire set of objects together in a particular way, so
as to serve a coherent function. You’ll probably keep the overall function of the
goal object in mind as you go along, making sure that what emerges from your
transformation will actually be useful as a cupboard. One of the main chal-
lenges, however, is to assign specific functions to individual parts. Which of the
pieces could serve as a shelf, which one goes on top, which one could be
a door? How do you know, and how certain can you be in your decisions about
which part will serve which function?
As described in Gralla and Tenbrink (2013), we created a similar scenario by
giving participants a set of objects that together constituted a doll’s house and
asking them to assemble them. We were particularly interested in the effects of
how much information people had in advance. Assigning functions to object
parts should work best if you have a fairly good idea what your goal object
should look like. You might even be able to name individual functions speci-
fically if you know where the object parts will end up in the overall structure. In
contrast, if you have no information other than the dissembled object parts as
you see them, you will have to think creatively about the possible ways of
fitting them together, with a possible overall function.
We tested this assumption by using three different conditions (Gralla &
Tenbrink, 2013), with sixteen or seventeen participants each. In one condition
called ‘verbal & visual goal’, participants were shown a photograph of the
assembled doll’s house and asked to assemble the pictured doll’s house. In
another condition, called ‘verbal goal’, they were not shown a picture but asked
to assemble a two-storey doll’s house. In the ‘underspecified goal’ condition,
they were merely asked to assemble a meaningful object. On this basis, even
though the transformation goal as such remained the same, the challenges for
achieving it were very different in each of the conditions. Participants were
asked to think aloud as they worked, so as to represent their thoughts during the
process.
In addition, we were interested in effects of communication – how do people
instruct others to accomplish the same task? For this purpose, we asked another
sixteen participants to assemble the doll’s house as in the ‘verbal & visual goal’
condition, but without thinking aloud. After successful assembly, their main
task was to instruct another person to assemble the same doll’s house. This
person was able to watch and listen through a one-way video connection that
did not allow for any responses.
We were interested to see if linguistic data would reflect differences in how
people assigned functions to object parts under these various circumstances.
Accordingly, our analysis focused on the matching process of object parts to
functions. Since this particular aspect had not been addressed as such before,

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 161

our analysis started on a qualitative basis, identifying what kinds of linguistic

expressions were used to describe the function assignments, and annotating
them throughout the data. This systematic procedure enabled us, in a next step,
to identify quantitative frequencies in the three conditions.
Results showed that mapping phrases often consisted of a set of three
elements: two noun phrases or pronouns that were related in some way to
each other by a mapping term. Consider the following examples:
Example 6.10 This is a wall.
Example 6.11 This could also be storeys.
Example 6.12 The red building part looks like the roof of a house.
Example 6.13 (. . .) each with a small piece uhm a wall element.
While Examples 6.10 and 6.11 start with a demonstrative pronoun (this),
Examples 6.12 and 6.13 use a more elaborate description of a specific object
(the red building part; a small piece). The object function is expressed in terms
of a noun (wall; storeys) in Examples 6.10 and 6.11, and a more complex noun
phrase (the roof of a house; a wall element) in Examples 6.12 and 6.13.
Whereas terms like roof and wall or wall element point to specific functional
parts of the house, references to the entire structure or larger parts of it, such as
storey, refer to structural aspects of the house that consist of multiple objects.
To relate the first reference term to the second, Example 6.10 uses a simple
relational verb (is), whereas Example 6.11 and 6.12 reveal a more tentative
assignment (could be, looks like). However, Example 6.10 and 6.11 both use
forms of be, as opposed to look like in Example 6.12, which expresses
a representational mapping rather than a direct relation. In contrast, Example
6.13 does not explicitly express any mapping at all – there is no mapping term.
Yet, just as in the other examples the same object is referred to twice: first as an
undefined small piece, then in terms of its function (a wall element). This is
a case of implicitly reframing an object within the same utterance.
Our quantitative analysis accounted for each of these differences system-
atically. For objects, we looked at occurrences of demonstratives (e.g. this,
that) as opposed to nominal references, and also looked at how specific the
nominal references were. Mapping verbs were annotated with respect to
whether they directly mapped objects to functions (which mostly happened
using the verb be) or treated the objects as representing a function (as in look
like or use as). Further, we looked for occurrences of modal verbs (such as
could, must, and should), which express certainty to different degrees, and for
cases of reframing without explicit mapping verbs.
Taken together, our analysis revealed three different mapping strategies as
represented in language, which were used under different circumstances. First,
the strategy shown in Examples 6.10 and 6.11 reveals direct mapping of an

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 162 Identifying Cognitive Constructiveness

object to its function. The certainty with which this strategy is applied is
indicated by the choice or lack of a modal verb such as could. Second,
Example 6.12 illustrates representational mapping, which was less frequent
than direct mapping; here, uncertainty is expressed by the choice of verb, rather
than by modals.
Across these two strategies, higher certainty was expressed most fre-
quently through the modal verb must where participants had most informa-
tion – in the verbal and visual goal condition. In the underspecified goal
condition, where participants had least information, they referred to struc-
tures (house, storey) more often than to specific object parts (wall, roof,
floor).
Third, the reframing strategy shown in Example 6.13 only occurred with
instructions, rather than during assembly. In instructions, all occurrences of
direct mapping expressed high certainty without any modals. Furthermore,
the representational mapping strategy was used more frequently than in the
assembly conditions, and instructors used more nominal references (as
opposed to demonstratives) than assemblers. Thus, the verbalisation of
object-to-function mappings differed markedly between think-aloud and
communicative tasks.
What can we learn from this study about transforming separate objects
into a coherent whole without instructions? A number of insights highlight
the cognitive processes that characterise this task. First, mapping objects
to functions is one of the clearly represented conceptual processes in this
kind of task; its regular presence in the verbal protocols as well as
instructions reflects its importance during the unaided (instruction-less)
assembly process. Second, mapping is preferably conceptualised directly
as a relation (this is, or could be) that is asserted with either more or less
certainty depending on circumstances, but it can also take the form of
a representation (this is used as, or looks like). Third, when giving
instructions, the conceptual challenge of assigning functions as such has
disappeared; instructors primarily communicate the result of the mapping
process to their interaction partner.
One of the strategies to communicate mapping is by reframing. Rather than
expecting that the addressee knows which object functions as a wall, speakers
frequently used both expressions, as in Example 6.13. Even though explicit
mapping was not deemed necessary, the addressee still needed to be
informed. It appears that hesitation markers such as uhm in Example 6.13
serve to replace the mapping term: since simply placing both terms together
yields an ungrammatical sentence, the hesitation marker indicates a restart,
following conventions of spoken language. It would be interesting to see if
this particular strategy would simply disappear in written language or take
a different form.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 163

In sum, our study showed how participants verbalised the transformation

from a set of objects to an assembled doll’s house, based on the assignment of
a specific function to each individual object. Obviously not every single object
was explicitly marked in this way, but the verbalised data revealed a range of
systematic ways in which function assignment was accomplished.

6.2.3 Transforming a Piece of Paper into a Complex Object: Following

Origami Instructions
Transformation can take many forms, and the cognitive processes involved in
this process will depend on what exactly is accomplished. While our previous
example demonstrated the importance of assigning functions to object parts
when transforming separate objects into a coherent whole, this section illus-
trates a different kind of challenge. This time, we are looking at a situation
where only one object (a piece of paper, without specific meaning or function)
is transformed into a different, now meaningful, object.
If you have ever tried origami paper folding, you will be familiar with the
difficulties that may arise while following the instructions. You know that you
are supposed to transform your piece of paper into something nice, such as
a flower, and you are given a range of steps that tell you exactly how to do that,
through pictures and words. Nevertheless, you may very quickly get to a point
where your own product does not look quite the same as the one in the pictures.
You might then try out various ways of improving your object: you could reread
instructions, reinterpret them, refold the paper, try again. In the extreme case,
which is not uncommon, you may be forced to give up, frustrated: this problem
has become too hard for you to solve, you just cannot figure it out.
Most research on problem solving targets situations in which only the goal
state is known (such as an origami flower), and the problem solver faces a range
of constraints as to what can or cannot be done (e.g. in origami, the paper is not
to be cut or torn into pieces). Research then typically addresses the kinds of
intermediate steps that the problem solver undertakes to reach that goal, given
the constraints (Ericsson & Simon, 1993). These intermediate steps could then
be turned into step-by-step instructions that anyone could follow easily, with-
out being confronted with the problem of figuring them out themselves.
However, origami paper folding is hard; it would be unrealistic to assume
that anyone could figure out how to transform a piece of paper into a flower
without step-by-step instructions. In fact, we (Tenbrink & Taylor, 2015) argued
that even just following the instructions is hard enough to constitute
a considerable conceptual challenge. In contrast to most previous research in
the area, we therefore treated the process of following step-by-step instructions
as a problem, while otherwise following traditions of problem-solving research
by analysing think-aloud protocols (Ericsson & Simon, 1993). We were

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 164 Identifying Cognitive Constructiveness

interested in the kinds of thought processes that were associated with the
various steps in the problem-solving process.
Our twenty-four participants were trained to think aloud and then asked to do
so while folding a relatively simple origami object, a tulip, following verbal
and visual instructions shown on a computer screen. They did not receive any
help from the experimenter except brief hints when they admitted to being
stuck, which happened to less than half of the participants at some point during
the task. In this study (Tenbrink & Taylor, 2015), we looked only at the paper
stem of the tulip, as this was sufficiently complicated to reveal a range of
interesting cognitive phenomena.
On average, participants folded the stem within five minutes, but the range of
variety was impressive: some needed twice as much, and some managed the
task in only three minutes. Similarly, they differed very much in the extent to
which they verbalised their thoughts, producing between 113 and 1,738 words
each. Altogether this resulted in a rich corpus of unconstrained language
production data, allowing us to examine participants’ verbalised thoughts
both on the level of content and with respect to linguistic features and patterns.
First of all, we were interested in how much the speakers were actually
verbalising their own thoughts, as opposed to merely reading the task instruc-
tions aloud – which they did quite frequently, as a natural part of the problem-
solving process. To determine this, we identified content categories and anno-
tated each language data unit according to whether:
• participants read the task instructions aloud
• participants reformulated the task instructions without adding their own
thoughts in any way
• participants verbalised additional ideas about a paper-folding step
• participants evaluated their own work with respect to success of paper
folding
• participants referred to background knowledge, such as experience with
origami paper folding
• participants expressed problems with the current problem-solving step
• participants commented on what they were currently doing, describing their
own actions.
We also included an ‘Other’ category, to capture further types of content that
did not fit into these categories but did not reoccur frequently enough to warrant
a further content category. While all of the identified types of content occurred
frequently in the data, ‘reading’ was the most frequent category, followed by
‘comments on current actions’. However, ‘additional ideas about the paper-
folding step’ also occurred quite frequently, and this was a category we found
highly intriguing. We wished to know more about what these additional ideas
were, and what could be learned from them about the participants’ problem-
solving processes. It is in this category that the conceptual challenge of

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 165

transforming a piece of paper into a meaningful object is represented most

directly, as it highlights the creative process of following the instructions.
Based on what we found in the language data, we subcategorised the
instances of ‘additional ideas’ further according to their content. We discovered
many comments on quality (e.g. trying to get a nice result, making sharp and
nicely aligning folds, or matching one’s own results to the instruction), and
various observations associating some meaning with the object, as in This looks
like a crane. Aside from these, we took a closer look at those comments that
observed some sort of structure: either the structure of the folded paper, or the
structure of the folding process. These included the following categories:
• spatial description: patterns in the current status of the object (e.g. it’s in
a diamond shape; or each corner goes to the opposite corner)
• orientation of the folding paper (e.g. that means it’s still on this side that we
have to fold)
• within-step repetition: doing the same action twice (e.g. for left and right
sides) within a folding step (e.g. I gotta do it with the other side)
• across-step repetition: recognising that an action repeats a previous step (e.g.
I have to do that same thing again).
The presence of these various recurring types of comment is an interesting
qualitative result as it highlights the kinds of conceptual patterns that
participants attended to during the paper-folding process. Structural think-
ing clearly mattered; for many participants, the recognition of some sort of
pattern was an integral part of the transformational process of paper
folding.
To gain further insight into the role of spatial thinking within this kind of
problem-solving process, we next took a closer look at the specific spatial
language employed by the participants. Although this analysis was inspired by
the category of ‘spatial description’, we decided to analyse the spatial language
across all content categories, as spatial thinking might influence other aspects
of verbalised content as well.
Again, the analysis required a clear differentiation between what people
simply read from the instructions, as opposed to what they came up with
themselves as part of the creative thought process. We identified all spatial
terms in the data and determined whether they were part of the instruction for
the current paper-folding step. If they weren’t, we counted them as ‘new spatial
terms’ that highlighted the spatial thinking in this transformational task. It
turned out that some terms (particularly prepositions such as in, on, down,
but also side) were used extremely frequently in this way, while others (such as
bisect, overlap, parallelogram) occurred only once or twice. Moreover, parti-
cipants differed very much as to how many new spatial terms they used. Some
were very creative, with one participant using as many as fifty-two different
spatial terms – whereas another participant did not produce a single new spatial

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 166 Identifying Cognitive Constructiveness

term that wasn’t part of the instruction in the current step. This reflects an
astonishing range of variety in how the participants approached the task in
a spatial sense.
Next, we were interested in the order in which the various kinds of content
were produced while thinking about a paper folding step. To assess this, we
identified the sequence of content categories for each participant within each
folding step. Overall, a pattern emerged that suggested the following structure
of verbalised thoughts, namely:
reading–reformulating–reconceptualising–evaluating
In other words, people started out, in each step, by reading the instruction. Next,
they might reformulate the wording by adapting it to their current situation
without adding content (as in turn the paper towards ‘me’, instead of ‘you’ in
the instruction). This was followed by adding their own ideas to the instruction,
that is, reconceptualising them, for instance by identifying spatial structures or
relationships to previous steps. Finally, they might evaluate the results, judging
whether or not their folding step was successful in this instance. Participants
did not verbalise each category for every step. It was also possible for them to
encounter problems that meant they had to go back to reading. Therefore, the
pattern is by no means strict or entirely predictable. Nevertheless, statistical
tests verified that this pattern did indeed reflect the order of content categories
as a general structure, which was adhered to significantly more often than not.
Altogether, our origami paper-folding research (Tenbrink & Taylor, 2015)
highlighted the fact that following instructions can indeed pose a problem,
involving substantial high-level thought processes that can be verbalised. This
is contrary to the traditional idea that think-aloud protocols are best suited to
identify separate steps in a problem-solving process. Furthermore, it showed
the wide range of individual variability in this regard, along with a consistent
basic pattern of conceptualisation: namely, starting with a close look at the
given information and gradually moving away from it and towards cognitive
flexibility, leading up to a result worth evaluating.
Incidentally, moving away from the original instruction towards one’s own
formulation is closely related to the cognitive effect discussed in the first half of
this chapter – inference processes. If simply following the instructions had been
sufficient, no further conceptual flexibility would have been required. Instead,
they frequently verbalised their own thoughts based on what they read, reveal-
ing a range of inferences made while applying the written instruction to the
transformation task at hand. In this sense, cognitive constructiveness works
both ways: we use it to interpret the information we are given, and we use it also
to transform the current status to something else. Our origami study revealed
how these two processes worked together by making inferences from instruc-
tions while transforming a piece of paper into a meaningful object.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 6.2 Transformation 167

In summary, this chapter has dealt with two ways in which people frequently
add to and change reality, so to speak: on the one hand, we add to reality by
making inferences from what we actually perceive or know; we fill in the gaps
in the available information based on what we have experienced previously.
More often than not, such inferences are useful and valid, allowing us to think
and act naturally and efficiently even in the absence of complete information.
On the other hand, we change reality by transforming an existing situation into
something different – we assemble objects to build a larger one, or manipulate
one object to turn it into something else. Transformation depends on the
cognitive ability to identify meaningful changes, to see the potential of certain
manipulations.
A sculptor, as the myth goes, doesn’t create a statue out of a brick – they
simply ‘see’ it underneath, and liberate it from the excess material around it.
This process, to some extent, is the basis of all transformation, and our language
reflects to a high degree just how we accomplish this. Human ability to change
reality in these two respects is remarkable, and it is certainly worth focusing on
as a target in CODA analysis.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 08:42:04, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 7 Using Language to Convey Thoughts

This chapter returns to a more general level as compared to previous chapters.

Here, we will consider the purposes and contexts in which humans use
language to convey thought. Whenever we put our thoughts into words, we
do this for a specific purpose. More often than not, the expression of thoughts
is part of everyday communication: we talk about our thoughts with our
partners and friends. But other contexts are possible, as we will see in this
chapter.
Section 7.1 discusses communication and considers how we interact
with others to exchange our thoughts through dialogue. When we do
that, we normally take into account who we talk to, what is relevant in
the context, and much else. All of this goes beyond the mere expression of
thought, because it pertains to the communicative function of language.
When we communicate, we express our thoughts for the purpose of
making them accessible to others – getting our thoughts across to our
interaction partner.
Section 7.2 turns to situations in which thoughts are expressed aloud
without aiming for communication, namely when thinking aloud. Even
though this is not an established social convention, it does happen some-
times – especially in children, who don’t always need a conversation
partner to chat aloud. In cognitive science, thinking aloud has been estab-
lished as a technique; in empirical studies, participants are encouraged to
simply say out loud what they are thinking as they solve a problem.
Therefore, they don’t need to worry about formulating their thoughts for
an interaction partner. The resulting language then reveals the problem-
solving steps that emerge in the mind of the participant, in a particular
sequence.
Similar to the previous chapters, both areas covered here – communica-
tion and cognitive strategies – relate to distinct analysis perspectives, and
they are associated with extensive research communities and traditions.
This chapter lays out some of the main associated insights, and thereby
highlights two distinct ways in which language is used to convey thought.

168

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 169

7.1 Communication
Language is central to our existence. In most people’s lives, not a single day
passes without using language for some purpose or other: we read, we listen,
we chat, we write, we talk – to one person or more at a time. Language serves
multiple purposes in our everyday lives, and it is hard to imagine what human
life would be like without it. Without doubt, the most prominent goal of using
language is to communicate. Communication is the primary function of lan-
guage. It helps us connect with other people, to exchange information,
thoughts, and ideas, and to share our insights about the world.
Communication can take many forms, and language is only one of them:
we can also communicate by gestures, mimics, sketches, pictures, signs,
diagrams, music, and more. There is no limit to the forms that commu-
nication can take, and each mode of communication has its own merit. But
notably, language typically comes along with these other communication
modes, or is an essential aspect of them. Moreover, language is the most
developed and complex mode of communication. It is therefore often more
suited to convey insights than other modes – especially if these insights are
abstract, complicated, or theoretical. Sign language is a full-fledged lan-
guage in this regard, while spontaneous gestures are more limited in what
they can express.
If communication is the primary purpose for language use, it is clear that
using language will mean more than simply producing verbal representations
of thought. In everyday use of language, we do not only want to externalise our
inner thoughts, but we want to do this for a particular addressee – for an
interaction partner or a larger audience, related to a particular context. This
will have effects on how we use language, as we will see in several examples
below. Many theories of language and communication focus on these effects.
Since the area cannot be covered exhaustively here, we will focus on some
major insights that are particularly relevant for CODA studies.
To provide just one example of a theoretical framework, Halliday’s Systemic
Functional Grammar (SFG) (Halliday & Matthiessen, 2014; see Chapter 3 and
elsewhere in this book) was developed precisely for the purpose of capturing
how the different elements of language serve different functions in commu-
nication, across diverse social contexts. Halliday suggested that every piece of
discourse is characterised by three situational parameters – field, tenor, and
mode – and each of these have a direct reflection in language. Field is what the
discourse is about (the general topic, what portion of the world is being
represented in the discourse). Tenor concerns the relationship between speaker
and hearer, or writer and reader. Mode means the basic distinction between
spoken and written, plus finer distinctions of intermediate modes and subtypes
(such as the special features of Internet chat, or of lectures as opposed to face-to

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 170 Using Language to Convey Thoughts

-face dialogue). The mode systematically affects how the discourse is struc-
tured and organised.
The fact that Halliday’s SFG is geared towards the communicative purposes
of language makes it useful for understanding what language does when it
represents thought. This is one of the main reasons why SFG theory serves as
one of the main resources for CODA, in spite of the fact that it makes no claims,
as a theory, about the relationship between language and thought. Instead, it
helps us recognise that certain linguistic features pertain to certain commu-
nicative functions, whereas others are predominantly means of representing the
world, or thoughts about the world.
In this section, we will now turn to other insights and theories beyond SFG,
in order to explore a range of effects that have been identified as crucial for
language use during interactions with others, where the purpose of commu-
nication is prominent. Section 7.1.1 will start by looking at how it matters who
we talk to, that is, effects of the intended addressee on the verbalisation of
thought. Then we will turn to aspects pertaining to face-to-face dialogue more
specifically. In Section 7.1.2, we will consider how speakers adapt their
language to each other’s way of speaking (and, arguably, thinking), and in
Section 7.1.3 we discuss the structure of dialogues, that is, regular patterns of
exchanging thoughts with an interaction partner. Section 7.1.4 reports an
empirical study designed to collect dialogue data in a situation involving spatial
thinking.

7.1.1 How Interaction Partners Matter

It matters who we talk to – we know this from everyday experience. Compare
the following two ways of getting someone to shut the door:
Example 7.1 Shut the door, love, will you.
Example 7.2 Please, might it be possible for you to close the door?
As we’re all familiar with social conventions of politeness, we intuitively
know that the relationship between speaker and addressee must be very differ-
ent in these two examples. In Example 7.1, a family relationship is likely; the
term of endearment ‘love’ suggests this just as much as the absence of any
politeness markers. The close relationship between the interaction partners
allows for the use of a direct command in imperative form (‘shut’). In
Example 7.2, in contrast, the speaker carefully seeks to avoid imposing an
obligation on the addressee. This is indicated by the modal ‘might’, by the
indirect mood (a question about possibility rather than a command), and by the
politeness marker ‘please’. It seems likely that the addressee is socially super-
ior, perhaps the speaker’s boss. The intended effect of the utterance is the same,

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 171

but the utterance is very different depending on who the speaker talks to.
Outside of suitable contexts, the same formulations may be perceived as rude
(in the case of Example 7.1) or as sarcastic (if Example 7.2 is used in a familiar
context).
If there hadn’t been any addressee at all, the speaker might have formulated
their thoughts aloud, for example by saying either of these:
Example 7.3 I wish someone would shut the door.
Example 7.4 If only the door was shut.
In this instance, the mere fact that there is no addressee turns the direct
command of Example 7.1, or the indirect question of Example 7.2, into an
expression of a desire. This desire can, again, be expressed in different ways, as
shown in the differences between Examples 7.3 and 7.4. Crucially, the linguis-
tic features distinguishing Examples 7.3 and 7.4 will not depend on the nature
of the addressee, but rather on the nature of the desire – that is, the cognitive
side of the utterance, rather than the communicative side. For instance,
Example 7.3 expresses a wish that involves the idea of a person (‘someone’)
acting, whereas Example 7.4 represents a desired state of affairs. (Note that this
is the basic idea in think-aloud protocols: to avoid, as much as possible,
situations in which the differences between Examples 7.1 and 7.2 matter, so
as to be able to focus on differences such as those between Examples 7.3 and
7.4. We will get back to this idea in Section 7.2.)
A focus on communication as an analysis perspective can be extremely
valuable when using CODA. There are many ways in which the intended
addressee can affect linguistic choices, and this fact has motivated many
empirical studies over the past decades. For instance, we (Tenbrink et al.,
2010) looked at how speakers verbalise spatial concepts when they talk to
a computer, as opposed to a human interaction partner. We used a dialogue
system that was designed to follow spatial instructions in a setup resembling
a typed chat situation. When chatting with another human being, our speakers
used a fairly wide range of communicative strategies, switching perspectives
and levels of granularity freely. This resulted in very efficient dialogues. In
contrast, when chatting with an automatic dialogue system, the speakers’
linguistic choices became much more limited. Speakers now formulated very
brief and simple instructions, and employed only a small subset of the linguistic
variety that they used when talking to a human. This was simply triggered by
their beliefs about the computers’ abilities, not by their direct experience of
what the system might have been capable of – they didn’t try!
Addressees affect linguistic choices by virtue of who they are, as we have
seen, but also by what they do. Imagine you are telling your friend a story on the
phone. After a while you notice that there has not been any response for a while.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 172 Using Language to Convey Thoughts

Inevitably you will ask, Are you still there? It would be very surprising if you
heard back, Sure, I am – but you’re the one telling the story so I have nothing to
say! While this statement is correct in terms of describing the situation objec-
tively, it may bother you. Addressees normally show their interest by reacting
in various ways to the speaker’s utterances. They would say Really? or Okay, or
use simple back channels such as m-hm or uh-huh, or just grunt or nod (which
you wouldn’t hear on the phone) – anything to show they are with you as you
tell your story. Typically, these back-channelling reactions come in a very
subtle and time-specific way (Bavelas, Gerwing, & Healing, 2017). So if you
don’t hear anything at all in response to your story on the phone, this is a clear
sign that something is wrong. Maybe your addressee is distracted and not
paying much attention, or your story has become too long-winded, and lack
of responsiveness is used to subtly suggest this.
Speakers don’t just expect addressees to react in those conventional ways to
what they’re saying, but they also notice how exactly they do so. It may be hard
to pinpoint, but you may have experienced this yourself: although your partner
used back-channelling signals in a seemingly normal way, you were sure that
they weren’t really listening. As a child, I could tell from the way my father
used m-hm whether he was paying attention or not; there was a slight but
noticeable change in the tone or intonation pattern. In contrast, addressees who
listen closely might enact in their reactions the mood of the topic appropriately
(RE-ally?!?!), showing cooperation on the level of intonation. Also, they
frequently contribute more than just generic back channels, by asking questions
and rephrasing or even extending the speaker’s utterances.
Clark and Krych (2004) showed that monitoring addressees is a crucial
part of dialogue, in particular in an instruction situation. The more access
people had to their interaction partner’s reactions in their study, the more
precisely they could react to them and adapt their speech accordingly. The
best case was if the partners could see each other and their workspace;
then, the dialogues were characterised by precisely timed pointing ges-
tures, eye gazes, head nods, head shakes, and the like. Sometimes speakers
even altered their utterances mid-course in reaction to their partner’s
signals of understanding. Other scenarios, such as giving instructions by
audiotape without the partner present, or being unable to see the partner’s
workspace, led to less communicative success.
Speakers’ awareness of their interaction partner’s understanding affects their
communication in many ways. When referring to objects, people can use many
different kinds of expressions, ranging from pronouns like it or demonstratives
like this one to complex explicit descriptions such as the hat with curly feathers
on the window sill. What speakers say to refer to an object very much depends
on who they talk to, and what the situation is (Arnold, 2008). Typically, when
we refer to an object, we want to enable our interaction partner to identify what

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 173

we are talking about. This will guide the expression we choose. Therefore, the
speakers’ attention to their addressee’s needs explains many of the linguistic
choices they make in object reference.
However, there are also some limits to this effect. Under certain circum-
stances, speakers fail to adjust their choices as precisely to their addressee’s
needs as they could. For example, they may choose a complex description in
a situation where a simple one would have been sufficient. Arnold (2008)
concluded from her literature review that speakers often make decisions for
the benefit of their addressee, but not all of the time – not all of their choices are
precisely tailored to their interaction partner. This makes sense if we consider
that communication is only one aspect of using language, alongside cognitive
aspects. When we speak, we first of all represent our thoughts. To some extent
we do this for our interaction partner, but we also express what we need to say,
not just what our addressees need to hear. Within limits, we can assume that the
addressee knows that our perspective may be different from their own.
In sum, there are many ways in which communicative aspects affect the way
we speak. Our beliefs about the interaction partner and their needs are central at
many levels of interaction: the words we choose, the details we provide, the
level of politeness we adopt, and so on. Interaction partners readily signal back
what they have understood, and we, as speakers, readily react to these signals.
In this sense, communication is a complex collaborative process (Clark &
Wilkes-Gibbs, 1986).

7.1.2 Mutual Adaptation in Dialogue

As we have just explored, speakers are aware of who they talk to, and they
adjust their linguistic choices according to their addressee’s needs and reac-
tions. But there are further ways in which speakers influence each other in
dialogue. Talking to somebody else is an opportunity to express one’s own
thoughts for the benefit of another, but not only that. It is also an opportunity to
exchange ideas, to gain insights into the interaction partner’s views, and to
negotiate thoughts where they differ.
Dialogue therefore means mutual influence and adaptation, and this happens
on several levels (Pickering & Garrod, 2004). Speakers adapt to each other’s
contributions on the level of sound, by making their words sound similar to the
way the dialogue partner produced them, or by placing the stress in similar
ways. Also, they typically use the same words to refer to things – sometimes so
much so that the particular reference term chosen is specific to a particular
dialogue (Clark & Wilkes-Gibbs, 1986). Thus, speakers would agree to use
a specific term during the course of an exchange, but with a different dialogue
partner they would use other words. Furthermore, the syntax of sentences often
matches previous ones in cases where similar content is expressed. For

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 174 Using Language to Convey Thoughts

instance, Cleland and Pickering (2003) found that when participants heard
a description like The door that’s red, they were more likely to refer to
a picture of a red sheep as The sheep that’s red than to say The red sheep.
Further, there is also evidence that speakers align their understanding of
specific meanings, or of the entire topic being talked about, based on their
exchange of concepts (Pickering & Garrod, 2004). If you have particular
associations about a concept, say for instance cat, after a while of talking to
you I might begin to share some of these associations – or at least understand
that you have them, and account for that in the way I talk about it. If cats, for
you, are dangerous creatures, my utterances to you might incorporate that idea,
and perhaps refer to claws and teeth. Or if you describe your home to me, my
mind may start creating a mental image that to some extent resembles the
mental image that you have in your mind.
The fact that mutual influence is so pervasive in dialogue has led to various
theories and debates about the nature of this influence, and about its precise
effects on speaking. Pickering and Garrod (2004) advocate a mechanistic view
on dialogue according to which alignment in dialogue happens automatically,
without the speakers’ conscious control. In their view, we align to each other
because it is a simple and helpful mechanism: alignment makes it easy to
communicate, as it reduces cognitive load when processing language (Garrod
& Pickering, 2004) – less work to do for our minds.
In social science, very similar effects have traditionally been examined
under the heading of accommodation (e.g. Giles & Coupland, 1991):
speakers align to each other in social contexts – not only on the level of
language – because they wish to show solidarity with each other. Crucially,
according to this theory, they can also fail to do so, and this can be taken
to signal quite the opposite, namely dissociation from the interaction
partner. Imagine that you sit opposite someone you like, and you see
that they cross their legs and sit back. You might then do the same, and
show by your posture that you agree with their relaxed attitude. In con-
trast, if you don’t like someone, you’ll be less likely to mimic any of their
gestures or linguistic choices. If they use a strong American accent you
might prefer to speak in your best standard British English, and so on. Not
all of these effects may at all times be entirely conscious, but they are
under the speakers’ control and can be adjusted at will, with the primary
purpose of showing distance or association with others.
In Clark’s (1996) view, the notion of negotiating common ground is central.
When speakers start talking, they normally already share a certain amount of
common ground. They often belong to the same or a similar culture, and
perhaps share further aspects of everyday life that they might have in common.
The better we know each other, the more we share common ground. We are to
some extent aware of this, and we formulate our utterances accordingly.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 175

Therefore, I would not refer to our mutual friend Lydia as the curly-haired
woman over there – this would be decidedly odd if my dialogue partner knows
her as well as I do.
As dialogue develops, speakers constantly extend the common ground that
they already share further. Every utterance that is heard and understood adds to
the common ground. Sometimes this process isn’t straightforward and requires
some negotiation, such as repairing and reformulating some expressions,
clarification questions, and the like. Speakers like to achieve some kind of
closure at the end of these processes – some kind of evidence that their message
has come across, and information has been added to the common ground.
That’s why the following dialogue sequence may seem somewhat incomplete:
Example 7.5 Heidi: Let’s meet at 12 o’clock.
Jerry: Okay, in the town centre.
Heidi: Bye!
It could be expected that Heidi should confirm the location of the meeting
before closing the conversation, similar to the way Jerry confirms the time of
the meeting by saying okay. Likewise, one might expect Jerry to respond to
Heidi’s suggestion of closing the conversation (Bye) by responding likewise
(Bye). Every aspect of dialogue thus implies some amount of negotiation.
Without the expected response to an initiation (e.g. a suggestion to close the
conversation, like Heidi’s Bye), there is no sense of closure. In the case of
Example 7.5, this may either mean that communication has failed, or that Heidi
is in a hurry to go and simply assumes that common ground has been reached
sufficiently, without signalling this explicitly. This view of dialogue suggests
that speakers are to a high extent in charge of the relevant negotiation processes
(Clark, 1996). Rather than following entirely mechanistic and automatic pro-
cesses, speakers actively monitor and adapt their reactions to each other in
discourse. As a result, the words and concepts they choose when expressing
meaning very often align with those of their interaction partner.
Altogether, we can conclude that interaction partners mutually adapt to each
other in dialogue in many ways. While theories in the literature do not always
agree on how exactly this happens, there is little doubt that mutual influence is
a very strong factor that drives and affects interpersonal communication, more
than we are usually aware of.

7.1.3 Dialogue Structure

Example 7.5, in the previous section, indicates another important fact about
dialogue: that it is not random, but follows certain structural principles.
Halliday and Matthiessen (2014) offered a ‘metaphor of symbolic exchange’
to describe this idea. In their view, each grammatical mood type suggests

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 176 Using Language to Convey Thoughts

a specific type of exchange between speakers: the interrogative mood is used

for (direct) questions, the declarative mood for statements and offers, and the
imperative mood for commands. Questions and statements indicate an
exchange of information: questions are used to demand information, and
statements are used to give information. Similarly, offers and commands are
used to give or demand goods and services.
Each of these four types of exchange (giving or demanding information, and
giving or demanding goods and services) can also be expressed indirectly (or
incongruently, as Halliday and Matthiessen, 2014, put it). For instance, it is
possible to demand information in the declarative mood, or to command
someone using interrogatives. Particularly in the case of commands, indirect
forms are considered to be far more polite. This is illustrated by two examples
that we saw earlier in this chapter – let’s take another look:
Example 7.6 Shut the door, love, will you.
Example 7.7 Please, might it be possible for you to close the door?
Example 7.6 uses the direct imperative form for commands, whereas
Example 7.7 uses the polite (indirect) interrogative form. However, the func-
tion of the utterance remains the same: in both cases, the speaker wants (or
‘commands’) the addressee to shut the door.
Developing these examples further, each utterance comes with an expected
response. In Example 7.6, the most likely response will be the wordless action of
shutting the door. No politeness is required in this familiar context, and a verbal
response is not necessary as long as the service is given as demanded. In Example
7.7, politeness will probably lead the addressee to do both: shut the door and say
something like Of course, no problem. If these are preferred responses, the
alternative is clear: in both cases, the addressee would refuse to do the requested
action. They could do this by saying No or using a more polite formula, perhaps
with a reason for the refusal – or they could simply fail to do as requested.
Failure to respond at all is perhaps the most surprising or disconcerting
reaction. And this is not only true for commands. Recall the situation described
earlier in this chapter: if you tell a story, you will expect your interaction partner
to at least provide some minimal responses, even if there are no questions and
no way in which your partner could actively contribute to the story. This
expectation reveals just how much we expect dialogue to be structured. Each
utterance comes with a certain expected reaction; these may be as varied as
minimal responses (back-channelling, nods, mimics), actions following
a command, or answers to explicit questions. Moreover, conventions and
expectations vary substantially across different contexts and modes of com-
munication. In remote-chat conversations (text or instant messaging, Skype,
etc.), it is not uncommon for reactions to come with significant delay, or not at

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 177

A: evaluate
B: reject request
B: reject request
2 promise (B, A)

request (A, B) inform (B, A) A: evaluate

1 3 4 5
offer (B, A)

2′ accept (A, B)
A: reject offer
A: reject offer

Figure 7.1 An example for a dialogue situation modelled by COR (Shi et al.,
2010).

all, without causing much concern. As communication technologies change

and develop, people demonstrate considerable flexibility and creativity for
making use of them in everyday life.
In the literature, these effects have been studied in a range of ways, for
various purposes. One such purpose is to implement dialogue models in
artificial communication systems, such as automatic telephone services or
assistance robots. In the context of information-seeking dialogues, Sitter and
Stein (1996) developed a model that is exemplary for various subsequent
approaches. Their ‘COnversational Roles’ (COR) model structures dialogues
in the form of recursive networks. Let’s look at a couple of examples that the
COR model can represent (see Figure 7.1).
Example 7.8 Customer: Can you tell me how to get to the airport, please?
Assistant: I certainly can. Tram number 5 will take you there.
Customer: That’s excellent, thank you very much.
Example 7.9 Assistant: How can I help you?
Customer: I’d like to know how to get to the airport.
Assistant: Tram number 5 will take you there.
Customer: Okay. But how do I find tram number 5?
Assistant: Sorry, I can’t help you there.
In Example 7.8, the customer initiates the dialogue by an information
request. The request is followed up by the assistant’s promise to provide the
information, followed by the information itself. The customer evaluates the
information, and as they are satisfied, the dialogue ends there. In contrast,
the dialogue in Example 7.9 starts with information being offered by the
assistant. This offer is accepted by the customer, who specifies what kind of
information is required. The assistant provides the information, and the
customer evaluates it. However, the dialogue does not end there due to
a need for further information, which the customer requests from the

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 178 Using Language to Convey Thoughts

assistant. Instead of providing the information, this request is rejected, and the
dialogue terminates.
The COR model (as sketched in Figure 7.1) represents these sequences and
allows for several variations of them, as well as for iteration based on the
information seeker’s evaluation of the information provided. The dialogue will
last until they are satisfied, or until the information giver rejects a request for
information. It is also possible that further information is offered but rejected,
which also causes the dialogue to end.
It is clear to see that a basic model like this one may capture dialogue
structures in clearly defined situational contexts quite well. The precise for-
mulation of a dialogue contribution does not matter; rather, the model captures
the function of each contribution within the information-seeking context.
Automatic dialogue systems use this kind of basic model to fill in various pre-
formulated utterances (e.g. ‘How can I help you?’) at an appropriate time in the
dialogue.
As information exchange is one of the most basic functions of language
(Halliday & Matthiessen, 2014), such implementations can be quite useful
across a number of contexts. By now, most people living in digitally advanced
cultures will have had an opportunity to talk to some variant of a dialogue
system. After decades of telephone services at various levels of quality, auto-
matic dialogue systems have now become sophisticated enough to be allowed
in our households, in the form of smartphone apps or specifically developed
devices.
However, natural dialogues between humans have a much wider range of
functions, and the structure of everyday conversations may not be as predict-
able as the structure of information-seeking dialogues. For instance, the COR
model (Sitter & Stein, 1996) suggests that a dialogue starts either by the
information seeker requesting information, or by the information giver provid-
ing information. But what happens if both start speaking at the same time?
Obviously, there must be mechanisms that allow for negotiation of such
possible conflicts. Fan and Heeman (2010) took a closer look at such situations
and found that they are actually quite rare; they concluded that whenever they
do happen, this must be unintentional and exceptional. Overlap in initiating
utterances only occurs if speakers actually start speaking at pretty much exactly
the same time. That makes sense: if we notice that our interaction partner has
already started speaking, we will wait to hear what they have to say before
starting our own utterance. Notice, however, that such principles are culture
dependent; overlap is much more normal in some cultures than it is in others.
Even British and American English differ in this regard, with overlap being
more acceptable in American than in British English.
Once overlap in initiating utterances happens, Fan and Heeman (2010) found
that it is normally resolved fairly quickly – overlap typically isn’t allowed to

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.1 Communication 179

last much longer than two syllables. The person who ‘wins’ is normally the
person speaking more loudly: volume is used as a device to organise the
sequence of turn taking in situations of conflict. Also, the overlapped part is
often repeated, to make sure the interaction partner hears what was said while
they were also speaking.
One of the most challenging problems for automatic systems is that most
discourse is context dependent. Face-to-face conversations between humans
take place in a spatial environment, and this is not easy to transfer to speech
technology. A lot of work has gone into ways of capturing dialogic interaction
that is situated in spatial contexts. How can a computer, which hasn’t lived and
grown up in a spatial environment as all humans have, even begin to understand
and talk about the world in a similar way to humans? Only if we understand
human spatial (or contextual) thinking thoroughly can we implement suitable
representations in computers. With this complex challenge in mind, CODA is
a useful tool to gain insights about how humans do actually understand the
world, and how they represent it in language. In Shi et al. (2010) we did just
that: we used results from a CODA study to inform a dialogue model following
Sitter and Stein’s (1996) method. This CODA study will be outlined in the next
section.
In sum, far from being random, dialogue is structured to a high degree. We
expect certain kinds of reactions at certain times, and feel that communication
fails when this does not happen. However, these structures depend on culture
and experience in complex ways. This makes the implementation in automatic
systems extremely difficult.

7.1.4 The Exchange of Spatial Thinking in Dialogue: An Empirical Study

To conclude this section, let’s take a brief look at natural dialogue between
humans, in a situation involving exchange of spatial thinking. Our research
team collected a large corpus of language data in a situation where two people
talked about furnishing a doll’s house (Tenbrink et al., 2008, 2017). One of
the two was assigned the role of ‘director’, and the other one the role of
‘matcher’. Each director instructed their interaction partner, the matcher, to
furnish a doll’s house in a specific way. Like previous examples discussed in
this book, this scenario requires the use of spatial expressions, which are often
highly ambiguous. While the director saw a fully furnished doll’s house, the
matcher looked at the partially furnished one that they were trying to com-
plete. Therefore, their relevant common ground was incomplete and needed
to be negotiated.
In our early paper (Tenbrink et al., 2008) we aimed to find out how the
negotiation of spatial relationships works in this kind of situation, and in
particular in what ways the matcher contributed to this process. One might

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 180 Using Language to Convey Thoughts

expect that they would only listen to the director’s instructions and carry them
out, perhaps giving minimal responses or asking the occasional clarification
question. However, we found they were actually quite active. Matchers con-
tributed to the spatial communication in a range of interesting ways. For
instance, they offered specific suggestions as to where objects should be placed,
based on but extending the instructors’ descriptions. For this, they used their
own conceptual perspective, and they often described different object relations.
Thus, if the director described an object in relation to the wall, the matcher
might find that this wasn’t precise enough or was perhaps ambiguous, and they
might ask how it related to another object like the sink – as in the following
example.
Example 7.10 Matcher: (. . .) and this one is put directly beside it.
Director: Yes, at the wall.
Matcher: That is, at the back left of the sink?
Director: Yes, but at the wall (. . .)
Thus, the matchers used their own understanding of the visual scene in front
of them to come up with meaningful suggestions. They made inferences on the
basis of what they saw, and in particular on the basis of how objects normally
relate to each other (Tenbrink et al., 2017). This enabled them to participate
actively in negotiation processes that ultimately resulted in fairly accurate
object placements, in spite of the fact that the spatial descriptions given by
the director were frequently underspecified.
In Tenbrink et al. (2017) we further explored the effects of how objects relate
to each other to serve a particular function. In the recorded dialogue scenarios,
we had used two different versions. In one version, the rooms in the doll’s
house were conventionally furnished: as a bathroom, a kitchen, a living room,
and a bedroom. In the other version, the furniture was placed randomly in the
rooms – bed next to shower, and so on. Our expectation was that this would
make a difference for the way people solved the task of furnishing the doll’s
house. If objects are meaningfully related to each other, we can use our every-
day knowledge to fill in some information gaps. If they aren’t, this might mean
that directors need to provide more explicit information as to where and how
objects should be placed, or it may lead to communication failure and lower
levels of task success. And in fact, our analysis showed that success rates were
higher in the functional version of the task. Also, the random arrangements led
to more elaborate descriptions, giving more information that helped cope with
the unexpected configurations (Tenbrink et al., 2017).
Altogether, the situated dialogue data represent an excellent resource for the
effects of communication on how spatial relationships are perceived and
described. Both the speakers’ spatial concepts and the dialogic negotiation
structures can be traced directly across a large set of linguistic data. CODA

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 181

helps to do this systematically, in order to identify dialogic and cognitive

principles that are reflected in the discourse.

7.2 Cognitive Strategies

CODA is certainly useful for analysing thoughts communicated through lan-
guage across many scenarios, including dialogue, as we have seen in the first
part of this chapter. However, the development of the methodology was
inspired from a primarily non-communicative approach to verbal data in
cognitive psychology. Ericsson and Simon (1993) outlined a paradigm for
using verbal protocols as data in the context of studies on problem solving
and decision making (see Chapter 8 for details on this method). These cognitive
challenges have been of interest to researchers for a very long time; after all, the
ability to solve problems and make decisions is central to being human. With
Ericsson and Simon’s method, people are asked to say out loud what they are or
were thinking, without engaging in dialogue with anyone else.
The paradigm has proved very useful, and it has been employed fre-
quently in cognitive science to address people’s strategies across various
kinds of complex cognitive tasks. Such tasks have in common that they
involve time and require several steps in the development of thought. If
problems are solved on the spot, or decisions are made in a flash, there is
not much to verbalise beyond the instant result itself. Verbal protocols are
useful for analysing more complex, sequential thought processes that
undergo several stages until completion.
Traditionally, verbal protocol analysis is done by cognitive psychologists
who do not normally draw on specific linguistic expertise. Analysis of verbal
protocols typically concerns the content of what people are saying, rather than
any patterns of linguistic features. In fact, Ericsson and Simon (1993) recom-
mended summarising people’s utterances into a limited number of content
categories and then discarding the original utterances in order to simplify the
analysis process. The basic tenet in CODA, in contrast, is that it matters how
people formulate things; we have seen this in many examples throughout this
book. Just as with any other type of language data, verbal protocol analysis can
therefore clearly benefit from a closer look at the linguistic features of people’s
utterances.
In the remainder of this chapter, we will look at a range of ways in which
CODA can be used to examine the verbalisation of cognitive strategies. Section
7.2.1 starts by examining the types of content that are frequently expressed in
verbal protocols, and the role of language in them. Section 7.2.2 addresses how
linguistic indicators can be used for the operationalisation of coding categories.
Section 7.2.3 turns to conceptual aspects that are typically expressed through
discourse markers, such as temporal expressions representing sequence, or

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 182 Using Language to Convey Thoughts

causal markers expressing a sense of interdependency. Section 7.2.4 looks at

mental switches between task domains and other variations in attention focus.
Concluding this chapter, in Section 7.2.5 we will examine the interplay of
cognitive strategies and communication in a wayfinding study (Hölscher et al.,
2011).

7.2.1 Content and Language Use in Verbal Protocols

Perhaps the most challenging aspect of verbal protocols is their complexity.
Confronted with a complex task, people verbalise all sorts of things as they
think aloud – and not all of it can be easily interpreted by a researcher. Many
thoughts are incomplete and context dependent, and they may not make sense
for any other person than the speakers themselves. It is easy for a researcher to
get lost in this kind of data, or to spend a lot of time trying to make sense of
confusing language. I have met countless researchers who invested major
effort, time, and money into collecting and transcribing verbal data in relation
to cognitively challenging tasks – and then they gave up trying to gather any
meaningful insights from them, as they realised just how complex the verba-
lisations were. This is clearly a trap that is all too easy to fall into. For some
researchers, the only alternative appears to be to refrain from using verbal data
altogether – despite the limitations of other methods, and despite the avail-
ability of Ericsson and Simon’s (1993) useful paradigm.
However, if verbal data were generally too complex and challenging to be
useful at all, I wouldn’t be writing this book. From a CODA perspective, the
key to meaningful and feasible analysis is to establish a focus that sheds light on
the targeted research question. Since language is always complex and can be
analysed in multiple ways, it is one of the most central goals for a discourse
analyst to identify a suitable analysis perspective, and to identify those parts of
the data that are meaningfully addressed using this perspective.
For Ericsson and Simon (1993), the main focus of analysis was typically the
identification of cognitive steps. Verbalisations of problem-solving procedures
reveal what cognitive operations people perform, and in what order. This
perspective guides the entire analysis in their paradigm. With such an analysis
focus, any verbalisations that do not relate to specific cognitive operations can
be ignored. Also, the specific form of a verbalisation is not crucial with this
approach, as only the operation itself and its place in the overall process would
be of interest to the researcher. CODA, in contrast, focuses on linguistic
features. Nevertheless, it is important to keep in mind that complex cognitive
processes develop over time, which means that the content of what people
express can vary across different stages during their thought process. Thus, it is
not always useful to treat the entire data set as a homogeneous resource for
linguistic features (see Chapter 8).

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 183

Beyond cognitive operations (i.e. what is done in what order during problem
solving), people often verbalise relevant aspects of their thought processes such
as conscious strategies and decisions, or plans for possible future actions. For
instance, when confronted with a search task as part of a problem-solving
process, people may decide to search in a particular area first, possibly for
a specific reason that they may also verbalise. How such decisions and strate-
gies are expressed in language can vary considerably, depending on the nature
of the task. For the analysis, it can be useful to highlight the fact that such
strategies are expressed, along with a brief characterisation of the main lin-
guistic features involved. Searching in a particular area could involve
a combination of perception verbs (expressing the search) with locational
terms (expressing the area). A high frequency of such a combination would
therefore indicate a cognitive focus on this particular aspect of the process.
In Tenbrink and Wiener (2009), for instance, we were interested in the
cognitive strategies involved in a particular type of spatial task called the
travelling salesperson problem. A travelling salesperson must visit many
different places to achieve their daily goals. In order to avoid wasting travel
time, they will try to find the shortest path that connects all of the places
before returning home. From a cognitive perspective, this poses a major
challenge. As there are many ways in which places can be connected, it is
by no means easy to find the shortest path. In our study, we used a version of
this task in small-scale space. In our think-aloud data, we could identify
various cognitive operations and processes that were verbalised and repre-
sented by recurring verbs such as searching (for a suitable path trajectory),
planning, imagining, path building, ticking off (places), making choices, and
so on. Beyond this, most participants reported visualising the entire trajectory
of their chosen path in the field. This observation was crucial, as it indicated
a specific problem-solving approach in this kind of task that had been
previously debated in the literature.
Naturally, what types of content can be meaningfully verbalised, and which
ones are of interest to research, will depend directly on the task at hand. In some
cases, participants may spend time considering and describing certain states of
affairs. This would then resemble simpler scenarios in which people verbalise
mental representations, for instance their perception of a picture (rather than
solving a problem). However, in contrast to a simpler task, such a description
would merely represent one certain stage in the process. States of affairs might
change throughout a problem-solving process. As a result, various verbalisa-
tions of mental representations at different stages can be found in verbal
protocols.
In other cases, people may consider how different situations relate to each
other, for instance if one was caused by the other (see Section 7.2.3). People
also sometimes reconsider the same situation in a different way, possibly as

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 184 Using Language to Convey Thoughts

a result of a change in the scenario or due to a sudden insight. Such effects are
frequently characterised by systematic linguistic features that can be traced in
the verbal protocols (see Section 7.2.4).
In sum, the content of language data is often highly revealing when analysing
verbal protocols of complex cognitive processes. Many insights in cognitive
science have been gained on this basis. A systematic analysis of relevant
content, and its expression in language, can allow the researcher to trace the
participants’ cognitive steps as they solve their task.

7.2.2 Linguistic Indicators and Coding Categories

Coding (or annotation) is a central part of any systematic discourse or
content analysis, and verbal protocols are no exception (Ericsson & Simon,
1993). Coding means assigning codes to relevant parts of the data (see
Chapter 8 for practical details), in other words, annotating the data accord-
ing to predefined features. For instance, when annotating content cate-
gories, the analyst might identify all utterances that pertain to a certain
step in the procedure, such as ‘searching’, and assign a label, such as
‘search’, to each of these utterances. The choice of annotation features or
categories can be based on content or on linguistic aspects; they can be
systematically determined on the basis of previous literature, or identified
post hoc, following inspection of the data. These decisions are important,
and they need to be made in light of the research question at hand as well
as the language data available. More often than not, the final set of
annotation criteria will be a combination of predetermined features with
new ones that arose from data inspection.
One of the most crucial aspects of this process is the definition of cate-
gories. If categories are poorly defined, the annotation will be weak and
unreliable; different coders will come to different conclusions, and it will
be difficult to convincingly report and publish the results. Because of this, the
intuitive element of most coding procedures is frequently attacked by oppo-
nents of verbal protocol analysis. Their argument is that scientific rigour
suffers if the analyst’s intuitions are required to interpret the data. And they
have a point. The less intuition and subjective interpretation is required, the
more reliable the analysis will be. Ideally, definitions should be found that
enable the analyst to identify each instance clearly and unambiguously,
leading to consistently annotated data.
Linguistic analysis can support this process. In some cases, it is possible to
identify and define categories entirely on the basis of objectively identifiable
linguistic features, rather than the analyst’s intuitions. For instance, when
identifying utterances that pertain to a specific cognitive process, it may
become apparent that this particular cognitive process is typically expressed

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 185

in the same way in language. Then, the range of linguistic terms can be
specified that indicate this particular category – such as linguistic indicators
for a search process (e.g. look for).
Other possible analysis categories move away from the idea of content types
altogether, and focus entirely on the occurrence and distribution of linguistic
features. For instance, you might find that perception verbs (such as see, look,
hear, listen, smell) occur in different contexts throughout your data. If you wish
to follow this up by systematic analysis, you will need a clear definition of
perception verbs, informed by the relevant literature and further supported by
the types of verbs you find in your data. Crucially, you will need to make clear
decisions as to which verbs to include, and you will need to justify these
decisions in a publication. Once these decisions have been made, the analysis
itself is fairly straightforward, and can often be done automatically (or semi-
automatically) using suitable tools. This removes intuitive interpretations and
subjectivity almost entirely from the coding process.
Intuitive content categories (which are highly subjective) and automatic
search for linguistic features (which is objective) represent two ends of
a scale. Mostly, you will need to define your categories on a precise but
relatively intuitive basis. Linguistic features support this process but do not
necessarily remove subjectivity altogether. For instance, when we analysed
cognitive processes in the travelling salesperson problem (Tenbrink &
Wiener, 2009), we identified a range of linguistic terms that could be
associated with certain cognitive strategies rather than others. Thus,
nouns like path, route, connection, detour, circle, and pattern referred to
the path as trajectory, and they were associated with verbs like searching,
finding, determining, planning, and so on. However, these word lists were
not exhaustive. In verbal protocols, it is often the case that participants
creatively use different terminology to describe similar things. Linguistic
analysis must allow for this possibility, and remain flexible to account for
variations in verbalisations appropriately.
In sum, suitable annotation categories are essential for making sense of
complex linguistic data collected in relation to complex cognitive processes.
The linguistic annotation of verbal protocols depends crucially on suitable
category definitions, identified and specified on the basis of the state of the
art along with preliminary data inspection. Final annotation definitions, as
reported in a publication, are often the result of an iterative process, aimed to
capture relevant phenomena as systematically and objectively as possible.

7.2.3 Discourse Markers as Indicators of Complex Cognitive Processes

While any linguistic markers can be revealing in the analysis of verbal proto-
cols (as in any CODA study, depending on the research question), complex

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 186 Using Language to Convey Thoughts

cognitive processes crucially unfold over time, and this leads to meaningful
sequences of thoughts. Therefore, it is often useful to look out specifically for
linguistic indicators marking the connections between thoughts and actions.
This helps with understanding the structure of the verbalisations in relation to
behavioural data.
Caron-Pargue and Caron (1995) were among the first authors to exam-
ine this idea systematically in relation to a problem-solving study. Their
participants solved the Tower of Hanoi task, which involves placing disks
on pegs following certain constraints. The think-aloud protocols clearly
reflected the fact that the speakers structured their actions sequentially.
Distinct linguistic markers signalled various kinds of connections between
the actions, and thus reflected a range of cognitively significant states and
transitions.
To highlight a specific example, in the same line of research, Bégoin-
Augerau and Caron-Pargue (2003) examined the role of various kinds of
connective markers (such as and, then, next, because, since, but). A speaker
connecting two units with so that clearly has a subsequent effect in mind by
pursuing a certain action. Example 7.11 is taken from Bégoin-Augerau and
Caron-Pargue (2003:83), who stress the fact that the actions are clearly inte-
grated by using the complex connective so that after (pour qu’apres in French,
the language of their study).
Example 7.11 the yellow disk I put it on the B so that after the black one I
leave it where it is (le disque jaune je le mets sur le B pour
qu’après le noir je le laisse ou il est)
Apart from connectives, similar effects of demarcating units can be achieved
by another class of linguistic markers, namely interjections such as oh, well,
okay, m-hm, yeah, and so on. These small and seemingly negligible items
appear to have little linguistic meaning, but their occurrence in context can
be surprisingly meaningful. This is true both for communicative contexts,
where interjections bring across substantial interpersonal meaning (e.g. as
minimal responses within a dialogue, as we have seen above), and for verbal
protocols. Caron-Pargue and Caron (1995) looked specifically at various mean-
ings of the French interjection bon in think-aloud data, and established how its
use during the problem-solving process related to different stages of insight and
evaluation of the situation. Often, the occurrence of an interjection marked the
participant’s surprise, or an opening or closing of a situation (Bégoin-Augerau
& Caron-Pargue, 2003).
In Tenbrink (2008), I compared different types of linguistic data to see how
the representations of thought varied (see also Chapter 2). The task given to
participants was a version of the travelling salesperson problem (see Section

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 187

7.2.1), and they were asked to think aloud while solving the problem, to write
a retrospective report after having solved it, and finally to write an instruction
for a friend to explain how best to solve it. It turned out that while the three
discourse types were compatible and together represented thought processes
accurately (as far as this could be determined in relation to behavioural data),
there were a number of systematic differences in the representations.
One major way in which the three text types differed most markedly was the
use of connectives and interjections. While all text types contained multiple
temporal markers, their meaning differed on the level of granularity. When
thinking aloud, temporal connectives (such as the German als erstes, zuerst,
danach, dann, jetzt, which are diverse equivalents of the English first, then,
now) marked immediate individual actions. However, when describing the
procedure as a whole in the two other tasks, temporal markers referred to
generalised subtasks and procedures, never to individual actions. The instruc-
tion to a friend, in particular, contained a clear temporal sequence, a general
action plan to carry out when solving the task.
Connectives of a more complex kind, that is, those marking semantic rather
than purely temporal relationships, were less evenly distributed across the three
text types. In the instruction task, markers of conceptual complexity (such as
the German jedoch, allerdings, obwohl – equivalents of the English however,
while, nevertheless) were most frequent, but there were no markers for caus-
ality – no reasons for actions were provided to the friend. In contrast, thinking
aloud was characterised most markedly by interjections structuring the pro-
blem-solving process, such as okay and so, plus the German discourse marker
ja (which indicates that something is already known). None of these appeared
in the other two tasks.
In sum, discourse markers can be valuable indicators of certain aspects of
cognitive processes, making it worthwhile to focus on them in the course of
a CODA analysis. Seemingly insignificant words like so, with its various
discourse meanings, can be highly revealing with respect to the participants’
conceptual state or progress.

7.2.4 Conceptual Switches

Another aspect that is specific to verbal protocol analysis is the fact that
speakers sometimes shift between different levels of meaning, in ways unlikely
to be found with less complex representations of thought. As the task proceeds,
people will often adopt different kinds of perspectives on the situation, and
regard it in more than one way over time. This leads to variations in attention
focus, as well as shifts between various semantic domains. Verbal protocols of
complex cognitive processes are far more varied in this respect than

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 188 Using Language to Convey Thoughts

representations of concepts and perceived scenes, allowing for rich and multi-
layered analysis.
To gain insights about how their participants’ focus of attention unfolds,
analysts may aim to identify linguistic representations of key concepts within
the data, and trace their distribution over time. Different stages of a problem-
solving process may require paying attention to different aspects and features
of the situation. Also, shifts of attention may lead to insights and decisions that
would not otherwise have been possible. For instance, people may start out by
focusing on one specific visual aspect – say, the colour of objects, which tends
to be cognitively prominent independent of its significance for the task. Then at
some point, they may realise that some other feature, such as shape or weight or
texture, has a significance for the task that they had not been aware of. This
might then lead to further insights or decisions. Such shifts in perceptual
attention would be reflected in language, by references to colour being replaced
by references to other object features.
Attention shifts can go beyond the level of objects, and concern entire
domains. To examine this, let’s return to a study introduced in Chapter 1. In
Tenbrink and Seifert (2011), we were interested in how people plan a holiday
route based on a map, when the actual environmental situation is remote and
unknown to the traveller during the planning phase. As part of the analysis, we
looked at how people switch between the spatial domain of the map in front of
them, and the large-scale domain of the real-world environment envisioned for
the holiday trip. Such a domain discrepancy is an integral part of any kind of in-
advance travel planning (see Pick et al., 1995, for a verbal protocol study of this
challenge).
Planning is what happens ‘here’ and ‘now’ – it’s what the planner is doing at
present: sitting at a table, with a map in front of them, considering options on
the basis of whatever insights the map offers. The information on the map is
typically limited to a few abstract symbols that the planner can neither feel nor
hear nor smell – they can only imagine what the represented things will be like
in reality. Travelling happens in the future; it is something that the planner may
or may not end up doing. All going well, objects and places shown in the map
will then become real-world objects and locations that one can sense in multi-
ple ways. A representation of a beach on a map is not relaxing, but the beach
itself may well be.
In this sense, a travel planning scenario combines two very different
‘worlds’. Successful planning might relate to the planner’s ability to imagine
the travelling world suitably, avoiding any travel situations that might spoil the
actual trip in the future. While it was unlikely that participants would directly
comment on this challenge, we expected that the required conceptual shifts
should be represented in the language used to describe the problem-solving
procedure in some way.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 189

To operationalise this idea, we first segmented the language data into infor-
mational units. Each unit contained a proposition like Examples 7.12 and 7.13
(here translated from the German original).
Example 7.12 The map was helpful because important things were plotted
there.
Example 7.13 In order to keep the trip from getting boring, some breaks of up
to 2 days are taken.
Then, we identified linguistic markers. Example 7.12 shows an informa-
tional unit that belongs entirely in the planning domain, as indicated by the
terms map, helpful, important, and plotted. In contrast, Example 7.13 indicates
the travelling domain by the use of trip, boring, breaks, and 2 days. It is
important to note that the verbal data also contained many instances that
were not clearly attributable to either one of the domains. Consider
Example 7.14:
Example 7.14 One resort at the south coast and one at the north coast. Both
have larger towns nearby. [Einen Badeort an der Südküste und
einen an der Nordküste. Beide haben in der Nähe größere
Ortschaften.]
While the words coast and towns arguably represent locations in the real
world, they are also clearly identifiable from the map. In fact, there is no simple
alternative way of referring to these places in reference to the map. For this
reason, references to places were not counted as indicating either of the
domains. Example 7.14 can be understood as relevant for travelling just as
well as for planning, as the description of spatial relationships is true for both.
While the examples so far could be attributed to either the travelling or
planning domain, or neither, some of the informational units contained markers
of both domains. These turned out to be particularly interesting, as they marked
conceptual shifts between the domains. Here are a few examples:
Example 7.15 I wanted to let the couple travel once round the western part of
the island [ich wollte das Paar einmal um den westlichen
Inselteil reisen lassen]
Example 7.16 After that I looked for a route that could be travelled [Danach
habe ich geschaut, wo man langfahren könnte]
Example 7.17 in order to design the travel so that it is as diversified as
possible [um die Fahrt möglichst abwechslungsreich zu
gestalten]
Example 7.15 contains references to two different agents: I and the couple.
They are linked through the verb let, which semantically indicates the effects of

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 190 Using Language to Convey Thoughts

planning on future actions in the real world. Other typical indicators of con-
ceptual shifts are modal verbs such as could in Example 7.16, and final
connectors such as in order to in Example 7.17. These linguistic devices
mark the fact that planning is not a conceptual activity in and of itself, but
rather involves a constant awareness of the real-world situation that is invoked
by the planning. Even though our study was hypothetical in the sense that no
actual travelling ever happened, the verbal reports clearly reflected the plan-
ner’s recognition of this second conceptual layer.
To summarise, conceptual switches appear to be a frequent phenomenon in
complex cognitive tasks. They are systematically reflected in language, and
indicate how the participants change back and forth between distinct levels or
aspects of the task. A close analysis of linguistic indicators of conceptual
switches can therefore provide valuable insights into multilayered cognitive
processes.

7.2.5 Cognitive Strategies and Communication in Wayfinding

So far in this chapter, we have treated the verbalisation of cognitive strategies
as distinct from communication as the normal function of language. According
to the paradigm established by Ericsson and Simon (1993), communicative
purposes are reduced to a minimum when thinking aloud or formulating
a retrospective report. However, some of the verbal data we have reported in
this chapter do involve some communicative aspects, such as writing instruc-
tions for a friend. More generally, it is important to keep in mind that commu-
nicative purposes can never be entirely ruled out, even though they may not be
in the foreground of attention. Even well-trained participants in an empirical
study will ultimately think aloud for the purposes of the experiment, and
accordingly communicate, to a degree, with the experimenter. There is prob-
ably no way around that. It is not necessarily a bad thing, but it is a fact that
shouldn’t be ignored or denied.
With this in mind, it is useful to consider the relationship between cognitive
strategies and communication more directly. How does the verbalisation of
thought interact with communicative purposes? There are two ways of thinking
about this. On the one hand, communication can affect problem solving. We
have all experienced situations where we couldn’t see the solution to a problem
on our own, but once we started talking about it to a friend, the situation
gradually became clearer until it was resolved. On the other hand, the verba-
lisation of the problem-solving process or the problem’s solution can change
depending on who we talk to, in light of the goals of communication. Both
effects can be significant and are worth taking into account, as we will see in the
following.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 191

The first effect (how communication affects problem-solving) was addressed

by Trognon, Batt, and Laux (2011), who asked why problems can sometimes be
solved more effectively in dialogue than individually. They gave participants
a logical puzzle to solve, which required some thought before the solution came
to mind. Just like in earlier related research, participants who worked with
a partner did better than those who solved the puzzle on their own. By looking
more closely at the strategies represented in language, Trognon et al. (2011)
found that they truly emerged through the dialogic interaction itself. Thus,
rather than getting help and knowledgeable advice from the interaction partner,
the cognitive strategies emerged jointly based on the exchange of thought,
focused on the logical problem that they were both trying to solve. Based on
this study, we may speculate that the active and partner-oriented formulation of
thought, bouncing ideas off somebody else, triggers a range of thought pro-
cesses that may otherwise not emerge – or at least not as efficiently.
In Hölscher et al. (2011) we addressed the second effect (how verbalisations
of thought are affected by the communicative situation), and asked how
cognitive strategies interacted with purposes of communication. Imagine for
a moment that you are in a town you are familiar with, and you need to get to the
train station. There are multiple paths, as it is a town with many small and
labyrinthine streets, but there are also some major roads that you could follow.
Knowing your way around town, which path would you choose?
Now imagine a stranger asks you for the way to the train station from that
same supermarket. Again, which path would you choose to help the person?
And how would you explain it to them? In our study, we reasoned that
wayfinding and communicative strategies may both be at stake in such
a situation. We would want the stranger to find their way, just as we’d like to
find our way. But in addition to this, we’d also need to communicate the
navigation information to the stranger, which isn’t necessary for ourselves. In
fact, we may not even want or need to make complete plans before we set out
towards the train station at all. This might lead to different planning and
verbalisation outcomes.
In order to disentangle these various factors, we conducted two studies. In
the first one, we tested the effects of planning in advance as opposed to
navigating directly to the goal. Taking the example just used further, imagine
you would first describe the path to the train station from where you are, and
then you would walk there, trying to find the shortest path. Do you think you
would use the same route as the one you just described? Perhaps you’d find
a better one as you’re walking, seeing as the town has so many different
options. And indeed, none of our participants actually used the route they
described in advance. All of them found shorter routes when they walked to
the goal. This result clearly demonstrates that wayfinding happens to a high
degree on the spot – we change and adjust our plans as we go along.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 192 Using Language to Convey Thoughts

In the second study, we added the communicative factor as follows. Each of

our participants was confronted with three different navigation tasks to differ-
ent goals. In one of these tasks, they walked directly to the goal. In another, they
were asked to write down how they would walk to the goal. And in the third,
they were asked to write down a route description for a stranger, as if they had
been asked on the street how to get there. Conditions were randomised in a way
that allowed us to directly compare the routes and descriptions, depending on
the task given in each case.
Results showed an interesting pattern. The paths people chose were most
efficient when they walked directly to the goal. As with the first experiment,
this showed that people were best able to find the shortest route when they had
a chance to actually see where they were going. But there was also something
else. In both of the written tasks, but particularly when writing the route for
a stranger, people relied more heavily on the major roads, as opposed to the
labyrinthine street network. In addition, descriptions were more detailed and
elaborate for the stranger than for the in-advance versions they wrote for
themselves. However, they basically contained the same types of information,
that is, a similar distribution of references to landmarks, paths, actions, and
so on.
Combining the evidence across various types of data gathered in this study,
a novel insight emerged, as follows. Communicative purposes mean that
people try to facilitate the task for somebody else. This has some effects on
the way things are formulated as well as to some extent on the route chosen,
which may be somewhat simpler than the route one would choose to plan for
oneself. However, the main difference is between planning in advance and
walking directly to the goal. These are actually two very different tasks, and
involve different conceptual strategies. When planning in advance, we retrieve
the basic facts about the street network from memory, and sketch the path
mentally towards the goal on that basis – with a high degree of orientation
towards major, more mentally accessible roads. This leads to the kinds of turn-
by-turn directions that we typically give to strangers: ‘Follow the main road
here, then turn right when you get to the church.’
In contrast, when we are free to just walk to our goal, we do not plan these
details in advance, and we do not rely on turn-by-turn information. Instead,
we simply orient ourselves to where we know the goal to be. Based on our
spatial knowledge of the town’s layout, we have a certain sense of direction –
we can point roughly to the train station from where we are. This enables us to
walk there on the shortest possible path, as close as possible to the ‘beeline’.
However, we may not be able to describe this same path in advance to
a stranger, or even to ourselves. That is because we don’t have all the
information readily accessible in our heads, no matter how well we know
the street network. Directional knowledge of this kind may be much more

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 7.2 Cognitive Strategies 193

central to our navigation strategies than we realise. In effect, it may be more

useful than we think to simply provide the stranger on the street with this
basic information: just point to where you know the goal to be, rather than
trying to explain the exact route. As long as the town has a sufficiently dense
network of smaller roads (and few dead ends), the stranger should be able to
find their way.

To summarise this chapter, cognitive and communicative aspects work together

whenever we formulate our thoughts. On the one hand, problem solving some-
times works better when we talk to another person about it. On the other hand,
how we approach a problem is influenced by the communicative situation, and
this will affect verbalisation in various ways. It is important to keep these
effects in mind when doing CODA.
Empirical studies can be designed in such a way that either cognition or
communication is in focus, or the effects of both are systematically compared.
Laboratory studies that involve language are often designed to minimise the
communicative aspect. However, it is unrealistic to assume that this aspect
could ever entirely disappear, since language evolved to communicate. Instead
of denying its presence, it may be more useful to recognise its function, and, if
possible, to identify the effects of communicative purposes on a given task.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 08:54:18, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8 CODA Procedures

So far, we have looked at motivation, background, theory, and a range of

possible analysis perspectives. Along with this we have seen multiple examples
from previous CODA studies that illustrate how it works and what kinds of
insight you might gain. This chapter is devoted to the more practical aspects of
doing CODA.
Let me remind you, to start with, that CODA simply means doing discourse
analysis with a focus on concepts and thought processes. CODA researchers
can draw on extensive practical information available in the literature – there
are many handbooks, textbooks, and research reports describing discourse
analytic methods for a rich variety of purposes, and any of them may be highly
useful and informative for CODA research (see also Chapter 3). In addition,
there are (and will increasingly be, given rapid technological advancement)
sophisticated tools available to support transcription, annotation, and corpus
analysis. These are not hard to find, and researchers can flexibly make use of
any of them to support the specific work in a project, once they have identified
their research goals concerning linguistic data.
For this reason, the main purpose of this book is to highlight the wide range
of concepts and cognition-related aspects that are reflected in language,
encouraging researchers to seek relevant insights through linguistic analysis.
Anyone doing that will be doing ‘cognitive discourse analysis’, that is, CODA,
even though the label doesn’t have to be there, and the exact methods for doing
CODA might differ.
In this chapter, I will not attempt to reintroduce existing methods or provide
extensive software tool reviews; nor do I aim to reproduce insights that are
better described elsewhere (which may include the more succinct CODA
introduction published in Tenbrink, 2015). Instead, the aim of this chapter is
to highlight some important principles that should be kept in mind no matter
which practical approach is eventually chosen. Along the way I will share some
practice-based insights that have accumulated over many years of doing
CODA.
Perhaps these practical tips will be of use to you, and perhaps you will prefer
to do discourse analysis in a completely different way: this is entirely up to you.

194

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.1 Preliminary Considerations 195

This chapter is primarily intended to convey some essential methodological

considerations, and it may perhaps convince you of the simplicity and acces-
sibility of the approach: you will not need to acquire highly priced equipment,
nor will you need extensive training beyond basic understanding of linguistic
principles, a suitable theoretical basis for your study, and systematic thinking.
Section 8.1 will start by looking at preliminary considerations, followed by
issues pertaining to experimental design and data collection in Section 8.2.
Section 8.3 will take a closer look at how to elicit verbal data, Section 8.4 will
discuss data preparation, and Section 8.5 describes practical steps of doing
language data analysis. The two remaining sections will briefly discuss quali-
tative and quantitative analysis insights, respectively, in relation to CODA.

8.1 Preliminary Considerations

Before collecting data for a CODA study, it is useful to have a clear idea about
initial thoughts and motivations. Why do you think linguistic analysis might be
relevant for your research purposes and goals? What kinds of insights do you
expect? Perhaps you have already collected language data, and now you’re
wondering what to do with it? Or perhaps you’re about to start a research study,
and are wondering how to design it so as to make the most of it? Or perhaps you
wish to use publicly available language data? There’s a lot out there: the
Internet itself can be regarded as a corpus of language, and there are many
more specific resources, collected and put together for all sorts of purposes.
Don’t leave this to chance – your research goals must guide all decisions
concerning the data you collect or use.
Any discourse resource can be suitable for doing CODA, as all language
produced by a speaker or writer reveals something about their thoughts. The
language producers’ concepts and way of thinking will always be represented
in some way in the linguistic choices they make. For this reason, you may be
able to conduct a meaningful CODA study on any data set, whatever your
starting point may be. However, different starting points mean different con-
straints on the kinds of insights that can be expected from a CODA study.
For instance, if you look at the language use of one person in one situation,
you will have nothing to compare it with. There may be many reasons why this
person makes the linguistic choices the way they do. You may have trouble
explaining what you find, and in the worst case you may not gain any mean-
ingful insights from the analysis. If you wish to look at the effects of a particular
factor, you will need two situations in which language is produced: one that
contains this particular factor, and one that doesn’t contain it or that changes it
in some way. Also, your conclusions will be much stronger if linguistic data is
produced by more than just one person confronted with either one of these two
situations.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 196 CODA Procedures

I personally think it’s best to start from scratch, with a burning research
question that pertains to language. Only then will you be in a position to
collect just the kind of linguistic data that is best suited for your goals. If
you have already collected data that you now wish to analyse using CODA,
some design decisions that you made earlier might have been practical and
useful at the time, but they may now constrain your CODA analysis in
unwelcome ways. For instance, in some research contexts it may not matter
much how exactly you formulate instructions to participants. So you might
have explained procedures informally to them – a bit differently for each
individual. However, slight changes in the instruction may lead to slight
but systematic (and potentially crucial) changes in the language partici-
pants produce in a study.
If you use available data that you haven’t elicited yourself, you may not have
access to some important information such as speakers’ gender and age, prior
knowledge, motivations for producing language, situational context, discourse
history, and so on, all of which may be decisive for interpretation of language in
relation to thought. Again, I don’t mean to suggest that CODA can only be done
in studies involving specifically collected data. In fact, CODA relates closely to
corpus linguistic approaches (see Chapter 9), and using existing corpora can
lead to a range of fundamental insights (e.g. Egorova, Tenbrink, & Purves,
2015, 2018). However, do keep in mind that not controlling the data yourself
introduces limits and considerations that you’ll have to take into account during
analysis.
Also, don’t forget that language is not the only way in which thoughts
and cognitive processes can be represented. Many insights about the
human mind in the literature are not based on language; there are a
broad range of methodologies in cognitive science that are suitable for
addressing many different features of the human mind. This book focuses
on insights that can be gained through linguistic analysis, beyond inter-
preting the content that is directly represented in language. The analysis
perspectives discussed in this book are by no means exhaustive, but it is
also clear that there are limits to what language can express (see also
Chapter 1). Again, your chosen research methodology must be suitable to
address your research goals.
Finally, research must always be ethical. Universities typically have an
ethical board or panel that provides ethical approval for a planned study,
usually guided by or similar to the APA (American Psychological
Association) Ethics Code. When using publicly available data from the
Internet or elsewhere, there are normally no ethical problems and no written
approval will be required; however, it is fair to eliminate all personal references
and names unless they are relevant to the study (as when analysing public
political speeches). If you collect data yourself, make sure to adhere to ethical

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.2 Experimental Design 197

and data protection (GDPR)1 principles, such as: (1) tell your participants that
their participation is voluntary and they can withdraw at any time, (2) preserve
participants’ anonymity by using codes rather than names, (3) avoid anything
that may potentially cause harm to participants, (4) treat all information con-
fidentially, and (5) keep the data in a secure place until deletion. If you cannot
find an ethically unproblematic way of eliciting data from participants, you
may need to reconsider whether you really need to do the study. However, most
CODA research is simple enough and straightforward in this respect – it
usually doesn’t harm participants to speak or write, which may be all they’re
asked to do, in relation to some kind of perception or thought process.

8.2 Experimental Design

If you collect data yourself, there are a range of things to consider. The
experimental design is crucial to any empirical research, and controlled lan-
guage data elicitation is no exception. CODA studies can be as varied as
language use in real life: there are no limits to the range of potentially inter-
esting designs. However, not every design will be suitable or useful for your
research question and purpose. For this reason, it makes sense to carefully think
about each step of the procedure in advance, plan all the details systematically,
and document them in writing to ensure they can be consulted and reported
later.
Let me just reinforce again that procedures need to be strictly consistent. All
participants in an experimental group must do exactly the same task. It is
unfortunately very easy to introduce small variations into the experimental
task procedure, which may jeopardise the study’s validity. Experimental varia-
tion must be planned and purposeful, and it should pertain to groups of
participants rather than individuals. There is no value in variation that is
introduced accidentally by sloppiness in the design or during data collection.
So, think everything through before you start – from the moment your partici-
pants enter the room until the moment they leave it again.
Let’s say you wish to find out how people describe a picture. This idea is
a perfectly typical goal for a CODA study: people don’t perceive pictures
neutrally, as a computer might do. They ascribe meaning to what they see
in the pictures, they use their experience and knowledge to interpret certain
aspects, and they are influenced by their motives and professional back-
ground. With this in mind, you decide to give people pictures to describe.
Because you want as much variety as possible, you ask all your friends, no
matter what age or native language or whatever else, and you give each
person a different picture (or several) to describe. This is legitimate, and

1
[Link]

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 198 CODA Procedures

you might get a wide range of interesting insights out of this. But let me
warn you. The data that you collect in this way will be extremely varied,
and it will be very hard for you to see any patterns and draw meaningful
conclusions.
The goal of most empirical research is to gain generalisable insights, not
simply momentary observations. If one person describes a picture in one way,
and a different person with a different background describes a different picture
in a different way, all you can say is that there are differences. You can point to
the differences in detail, and provide an in-depth qualitative analysis of what
they are. This can be useful at a certain stage of a research process, or for certain
purposes. The more that is known about an area, the more detailed a qualitative
analysis would have to be to provide any new insights. At a certain stage, it is
far more useful to look for more general patterns, and for possible reasons
behind the identified differences. This is what experimental variation, using
different controlled conditions, is for.
Sometimes you might want to use just two different conditions, and some-
times there will be more. Perhaps you wish to test two different kinds of
participants (such as men and women) in two different tasks, in which case
you have two independent variables (or factors) that may interact with each
other. Then, both of your participant groups (men and women) will need to do
both tasks so that you can compare the effects of task differences as well as the
effects of gender.
You can do this in either one of two ways. In what’s called a within-
participants design, each person does both tasks. This enables you to compare
the effects of task differences for each individual person. This is relevant for
statistics even if you don’t actually compare data on a person-by-person basis.
With a design like that, you need to be aware that doing one task may affect
what people do on their second task. Unless this effect of task order is desirable
for some reason, it is normally advisable to balance the order of tasks: half of
the participants do one task first, and the other half do the other task first. This
allows you to check for any effects of task order.
The alternative is called a between-participants design: one group does
one task, and another group does the other. If you also wish to compare
men and women, you’d need two groups of men and two groups of
women, and they’d all do only one task. Since groups shouldn’t be too
small, it’s clear that studies using a between-participants design will need
to involve more people than studies using a within-participants design.
However, the advantage is that you don’t have to worry about effects of
task order if there is only one task for each participant to do.
The one thing to avoid is to have all the men do one task and all the women
do a different task. This is called conflation – two factors (gender and task) are
influencing your results at the same time, and you have no way of knowing

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.2 Experimental Design 199

which one causes the patterns in your results. It makes sense to avoid having
more than two factors, as every factor introduces further possible combina-
tions; this makes it more difficult to see patterns and do a meaningful statistical
analysis.
Sometimes it’s possible to predict specific patterns in language before you
even start on your study, based on the previous literature. This allows for the
formulation of hypotheses in advance, and it can be very decisive in determin-
ing experimental design. However, whatever your design and motivation is,
make sure you consider in advance what kinds of linguistic patterns you expect
in the data that will shed light on the research question at hand.
For instance, after having read up on the topic, you might hypothesise that
men use more expressive language than women, but only in informal contexts.
To test this, you use two experimental groups, one consisting of men and the
other of women (and of course you might use more groups, allowing you to
include other gender types). Every participant is individually given one task
with a formal context, and another task with an informal context. Half of the
participants start with the formal one, and the other half with the informal one.
This within-participants design will allow you to compare the distribution of
your dependent variable (expressive language) in the collected data, in relation
to the independent variables formality and gender.
Let’s take a closer look at the formulation of instructions to participants. You
can actually test this in a CODA study. Let’s say you wish to see if people
perceive and describe a picture differently depending on how the instruction is
formulated. One of your experimental groups therefore gets the instruction in
Example 8.1, and another gets the instruction in Example 8.2:

Example 8.1 Please could you describe what you see in this picture?
Example 8.2 Tell me what this picture shows.
Both instructions seem very similar, and it is conceivable that intuitively you
might have used these formulations interchangeably, along with further varia-
tions, without being aware of possible effects of this variability. After all, you
always mean the same – surely the participants will always understand the
same, and simply provide a picture description as requested.
However, this may not be the case. At a closer look, the two instructions
differ in rather decisive ways. Example 8.1 is formulated in a polite and indirect
way (using please and a question format), whereas Example 8.2 is a simple
direct command. Example 8.1 uses the verbs describe and see, as opposed to
the verbs tell and show in Example 8.2. Describing is different from telling
someone – it may possibly be perceived as more elaborate, and more process-
oriented, whereas telling is a communicative process, which foregrounds the
researcher as an interaction partner. And seeing is more subjective than

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 200 CODA Procedures

showing, as it pertains to the speaker, whereas showing pertains to the picture.

These semantic differences aren’t as subtle as they may seem at first. Whether
people perceive the instruction as the same on a pragmatic level, and provide a
picture description in just the same way, is an empirical question that a CODA
study would be well suited to address.
In CODA studies, participants should normally be native speakers of the
language used. If you’re interested in conceptual aspects reflected in speakers’
choices, it makes sense for the speakers to have the full network of options at
their disposal. Learners of a language have an ever-increasing range of choices,
but until they are fully fluent they will have to use whatever words they already
know – and this is less than the entire network of options in the language. If you
are interested in how this works, you might want to design a study with two
experimental groups to test this: one with native speakers, and one with non-
native speakers who are learners at a similar stage and with a similar linguistic
background. Both groups would receive the same task, and this would allow
you to study how speakers in each group use language to express their thoughts.
Keep in mind, however, that knowledge of a language is always a decisive
factor. In a study looking closely at linguistic choices, it really doesn’t make
sense to mix native and non-native speakers in one experimental group.
In short, whenever you introduce a difference into your design that may be
decisive (as most differences are, to different degrees), use different experi-
mental groups, if at all possible. In most cases it is best to avoid differences
(such as different wording or linguistic background), unless they are the target
of the study – and this would have to be motivated from a theoretical perspec-
tive, based on previous research in the area.
If you cannot avoid variation, aim for a random or equal distribution.
For instance, if you’re not particularly interested in gender differences, try
to get a balanced gender mix in each experimental group. And if you
don’t expect social background to matter in your study, use a random
sample from the population you’re interested in. In many cases the
population will be undergraduate students, for convenience – these are
typically easiest to find for an empirical study in academia. If that’s the
case, make sure you don’t add one or two people who are not under-
graduate students, as their educational background would introduce a
variable that only applies to a small proportion of the group.
Depending on your research question, you might be interested in collecting
further types of data or evidence beyond language. For instance, in a study
collecting picture descriptions, you might want to record the time that partici-
pants take to describe a picture. If the temporal factor is particularly decisive for
some reason, you might time the exact milliseconds that they need before they
start talking after having been given the picture, as well as the pauses they make
in the process, and so on. Alternatively (or in addition) you might be interested

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.3 Verbal Data Elicitation 201

in the gestures they produce during the description, or in physiological effects

that you’d need to measure using specific instruments.
As indicated by these examples, there are many different kinds of data you
can collect, in addition to language. In a CODA study, language is central, but
other kinds of data might add valuable information that can be used for
triangulation. Most published CODA studies have used various types of evi-
dence, addressing the research question at hand in more than one way. We will
get back to the topic of triangulation in Chapter 9 and outline more broadly how
CODA can be meaningfully combined with other data.
When collecting other types of data, as with language, every detail needs to
be considered carefully in advance, and procedures must be clear and consis-
tent. You will also need to make sure that you will be able to align the language
with other data as required. For instance, if you ask people to do something, to
perform a task in addition to producing language, you will need to keep track of
what they are doing at each point while they are speaking. In most cases this
will require video recording the participants’ actions. As this can make analysis
quite complicated, it may be worthwhile to consider in advance which aspects
of the language are most interesting in relation to which kinds of action. Such
considerations can influence the experimental design from the start, saving
considerable effort later.
Importantly, as you decide on your experimental design and collect your
data, always keep the participants’ well-being in mind. As already mentioned in
the previous section, you need to adhere to ethical principles, and obtain ethical
consent prior to collecting data. This will ensure that you won’t induce any
harm to participants. In addition, it is particularly important in a CODA study
that participants feel at ease, and produce language in a natural way. If they’re
not sure what they are doing, or feel that they are confronted with an awkward
task, this will be reflected in their language. Odd and uncertain situations lead
to odd and uncertain language – and this is unlikely to be your research goal.
With this in mind, in the next section we will take a closer look at ways to elicit
verbal data for a CODA study.

8.3 Verbal Data Elicitation

People use language in many different ways, and throughout this book we have
described many examples of studies where language data was collected in
different ways and formats. Let’s take a look at various ways in which you
might want to elicit verbal data from your experimental participants. Clearly
this is an important decision to make. How you ask your participants to produce
language will affect your results very strongly and constrain the analyses you
can do on the data, and the conclusions you can finally draw. It therefore makes
sense to spend some time thinking through the implications of different ways of

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 202 CODA Procedures

eliciting verbal data. We saw some of this when discussing Examples 8.1 and
8.2. To take another example, consider the difference between the following
instructions:
Example 8.3 Why did you solve the problem in the way you did?
Example 8.4 What were your thoughts during the task?
Example 8.5 Try to recall what you were thinking while solving the problem.
Step by step, what were your thoughts as you solved the
problem?
Imagine what people might say in response to each of these questions.
Example 8.3 is a why question, and as such encourages the participant to
look for reasons for doing what they did. These reasons may or may not
correspond to the participants’ actual thoughts during problem solving.
People sometimes do things for a conscious reason, and sometimes things
just happen, and the problem gets solved. At the very least, the response will
only represent a selection of actual thoughts – people will only report those
ideas that provide a reason for solving the problem.
Example 8.4 directly asks for thoughts. This might work in the way it is
formulated – it doesn’t create a bias towards reasons or anything else. However,
the question is general and asks about any thoughts that come to mind, in any
order. Example 8.5, in contrast, encourages the person to think back to the
procedure of problem solving. Thoughts occur in a certain sequence during a
complex cognitive task like problem solving, and you might be interested in
this aspect. In support of this, Example 8.5 encourages participants to recall
their problem-solving steps in their sequence, whereas Example 8.4 will
probably lead to shorter, less orderly representations of remembered thoughts.
However, CODA is not only good for problem-solving tasks. Let’s take a
closer look at what happens during a description task. Descriptions are reflec-
tions of the speaker’s thoughts with respect to what they describe, and can
therefore be analysed in terms of mental representations. People can describe
many things – there is no limit to the range of concepts and thoughts that can be
looked at on this basis. This field is very flexible, and many CODA studies have
focused on descriptions of various kinds (e.g. Cialone et al., 2018; Egorova et
al., 2015, 2018; Hölscher et al., 2011; Tenbrink et al., 2017, and many more).
Let’s say you are interested in people’s thoughts on XYZ. Then you might
ask participants to describe any of the following: XYZ itself; what they see with
respect to XYZ; what they remember about XYZ; how they would deal with
XYZ; how they feel about XYZ; or anything else along those lines. All of these
questions trigger a specific mindset on the part of the speaker, encouraging
them to think about XYZ in a particular way. Again, what exactly you ask will
be decisive, as it will push the verbalisations and your subsequent analysis in a

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.3 Verbal Data Elicitation 203

particular direction. ‘Describe XYZ’ is the most neutral version, but this may
not always work, or you may wish to elicit more than one description.
Obviously, verbalisations of feelings about XYZ are very different from simply
remembering XYZ. Nevertheless, all of these questions trigger descriptions
and may be of interest for a CODA study.
But what is XYZ in this instance – what could it be? Again, this opens up a
wide range of possibilities. It could be anything that is somehow represented in
the speaker’s mind, or that the speaker currently perceives or remembers
perceiving earlier: an abstract concept, a picture, an event, a scene, a path to
a destination – and many more. Descriptions can be very complex, and extend
over time, as in describing how a story unfolds, perhaps in a film or novel that
speakers recall. On the other hand, a description may be as simple as a single
sentence, if XYZ – the described item – is very simple. For instance, in Vorwerg
and Tenbrink (2007) we showed participants abstract depictions of a square and
one or more circles, nothing else. One of the circles was marked by an ‘x’. We
asked them one of two questions with respect to each picture:
Example 8.6 Where is the element that is marked by an ‘x’?
Example 8.7 Which element in the picture is marked by an ‘x’?
We were interested in the different responses that would be triggered by the
questions. In particular, we reasoned that the question in Example 8.6 would
encourage speakers to focus on the spatial location of the objects relative to
each other, triggering fairly detailed spatial descriptions. In contrast, we
expected that the question in Example 8.7 would be read as an identification
question; responses would have to be sufficient to identify the marked element
in contrast to others in the picture, but the spatial location would not be
important as such. Since spatial location is frequently used to identify objects,
as in ‘the one on the right’, the responses might still be very similar.
Following a CODA analysis of the fine details in the linguistic choices
speakers made, our results showed that where questions (Example 8.6) did
indeed trigger more precise and complex responses. If there was only one circle
present in the picture, the which question (Example 8.7) could be simply
answered by a phrase like ‘the circle’; this was not possible in the case of a
where question. More complex pictures frequently triggered spatial descrip-
tions for which questions just as well as for where questions, but they tended to
be less elaborate.
Descriptions can be very interesting to analyse, and they are simple to elicit.
An everyday question like ‘Please describe the picture’ or ‘What do you see?’
may be fully sufficient for the purposes of a study, requiring no further
explanation. To a high degree, descriptions represent the speaker’s concepts,
mediated by how the question is framed and how willing the participant is to

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 204 CODA Procedures

represent their thoughts, and so they offer a varied and unlimited resource for
accessing thought.
While descriptions express mental representations, they do not directly
reflect thought processes – such as those that happen during problem solving.
If you ask somebody to describe what they did when they solved a problem,
they will tell you how they remember doing it, and what they think they did. As
a mental representation of the problem-solving process, this is fine and can be
very interesting. However, it is important to distinguish between a mental
representation of the process (as reflected in a description), and the thoughts
during the process. To access the actual thought processes during problem
solving, we need to move away from description and look for a different way of
eliciting verbal data.
As a matter of fact, historically the study of problem solving started out by
relying on descriptions, often those of individuals. Early studies relied on the
researchers’ own memory of how they solved problems themselves; they aimed
to identify general principles from these introspective considerations. Later,
they began to ask other people, individual experimental participants, how they
solved the problem. However, insights gained from these studies were increas-
ingly criticised as being too subjective, too anecdotal. People might misre-
member or confabulate; they might report thought processes that never actually
happened. From a current point of view, this is not surprising: after all, research
then focused on descriptions of problem-solving processes, rather than repre-
sentations of the actual thoughts during problem solving.
After experimenting with these concerns for several decades, Ericsson and
Simon (1993) famously established their paradigm for eliciting reliable verbal
data on thought processes that is still widely used in cognitive science (see
Chapter 7). They suggested two main ways of eliciting data, neither of which is
a description; both aim at triggering a more accurate representation of the
actual thought processes, rather than a representation of what the speakers
think they thought. The two methods differ only in one way: when the data are
elicited – either during or after the actual problem-solving process.
Collecting verbal data during problem solving means that participants are
asked to solve a problem, and to verbalise their thoughts as they do so. Again, it
is important to note that they are not asked to ‘describe what they are doing’ or
anything like that – we are not looking for a description of thoughts or actions
in this case, but for a representation of thoughts and thought processes.
Ericsson and Simon (1993) suggested a formulation similar to this one:
In our study we are interested in what you think as you perform a task that we give you.
In order to do this I am going to ask you to THINK ALOUD during the whole procedure
of the task. That is, I want you to say EVERYTHING you are thinking from start to
finish of the task. I would like you to talk aloud CONSTANTLY. Don’t try to plan out
what you say and don’t talk to ME. Just act as if you were speaking to yourself. It is most

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.3 Verbal Data Elicitation 205

important that you keep talking, even though you won’t get any response or feedback.
Do you understand what I want you to do?
This general instruction is usually followed by a couple of warm-up practice
tasks, such as:

Example 8.8 Please multiply two numbers in your head and say out loud what
you are thinking as you get an answer. What is the result of
multiplying 24 × 36?
Example 8.9 How many windows are there in a house you know very well,
for instance the one you grew up in?
Practice tasks should be as far away from the study goals as possible. If you
wish to study how people solve complex mathematical problems, Example 8.8
is not suitable as it also concerns a mathematical problem. Any kind of simple
everyday problem that is not too taxing but nevertheless involves a short
problem-solving process will do. Based on the instruction and practice tasks
of this kind, participants normally understand what is required of them, and do
not struggle with the idea of thinking aloud. Ericsson and Simon (1993) suggest
that thinking aloud is a natural activity, something that we intuitively do, and
that we might have done frequently as a child; it easily comes back to us in a
context like this. While this may not be true for everyone, most participants are
quite happy with it, and soon don’t see it as an additional task. They just say out
loud what is on their minds anyway.
The instruction above may sound a bit complex, but the underlying idea is
really quite simple. Under circumstances that do not allow for long instructions
and training, it may also work to simply ask people to say out loud what they
are thinking. This is a shortcut to the long instruction, but it does express the
entire idea. It is simple, and it avoids any bias towards description or explana-
tion or anything else that would encourage people to express anything else than
their actual thoughts.
Even though thinking aloud, if encouraged in this way, is simple and
straightforward, there are a number of potential problems. There is abundant
evidence in the cognitive science literature that the actual thought process may
be influenced by the task of thinking aloud, in a range of ways. In some cases,
verbalisation is helpful for the task. In Chapter 7, we discussed the frequent
finding that some problems can be solved more easily when talking to some-
body else about it (e.g. Trognon et al., 2011). In the case of thinking aloud, there
is no interaction partner, but the process of verbalising itself may sometimes
lead to a different organisation of thoughts, and this can support problem
solving.
In other cases, verbalisation may do the opposite – it may hamper the thought
process. For instance, if the task involves language in some way – perhaps

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 206 CODA Procedures

solving a linguistic problem – there may be some interference between using

language for verbalisation and considering language as part of the task. Also, it
is possible that people are too busy thinking to express their thoughts aloud at
the same time; this would then lead to silence during a decisive part of the
problem-solving process. And finally, some thoughts cannot be verbalised at
all. Perhaps people are not sufficiently aware of these thoughts to verbalise
them; some kinds of thoughts are too subtle or low-level to be verbalised. Or
perhaps people cannot put these particular kinds of thoughts into words, as not
all thoughts are easily verbalisable; it might be easier to draw something or
express the thoughts in a different way. Clearly, language has its limits as a
representation of thought (see also Chapters 1 and 2).
These limitations don’t mean that collecting think-aloud data is pointless or
shouldn’t be done. They simply need to be considered for the particular case at
hand; the researcher needs to be aware of potential problems and interferences,
and take them into account in the analysis. Also, it is always a good idea to
include a control condition that does not involve thinking aloud but is otherwise
identical to the think-aloud condition. This allows for a comparison between
the results of a problem-solving task done while thinking aloud and without
thinking aloud. For instance, in the case of a mathematical problem like
multiplying 24 by 36 (as in our warm-up task in Example 8.8), you could
compare the percentages of correct results in think-aloud and silent versions, as
well as the time needed to get to the answer, and so on.
In spite of these limitations, thinking aloud during problem solving is
clearly the most direct way of accessing thought processes as they occur.
Any other kinds of verbal representations are necessarily less direct.
However, Ericsson and Simon (1993) suggested that if collected in a con-
trolled and conscientious way, retrospective reports are also very useful and
fairly close to the actual thought process. This brings us back to the examples
in the beginning of this section. If we want people to report their thoughts
retrospectively, there are a number of things to consider in order to avoid
meta-descriptions of what people think about their own thoughts, explana-
tions, confabulations, and other verbalisations that lead away from the actual
thought processes. Here, Ericsson and Simon (1993:378) suggested the fol-
lowing instruction to participants:

Now I want you to see how much you can remember about what you were thinking from
the time you read the question until you gave the answer. We are interested in what you
actually can REMEMBER rather than what you think you must have thought. If possible
I would like you to tell about your memories in the sequence in which they occurred
while working on the question. Please tell me if you are uncertain about any of your
memories. I don’t want you to work on solving the problem again, just report all that you
can remember thinking about when answering the question. Now tell me what you
remember.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.3 Verbal Data Elicitation 207

It is certainly a good idea to adjust this instruction according to the needs of the
current task. Also, eliminating all references to ‘me’ as the experimenter would
make the task more similar to a think-aloud task in which there is no designated
addressee. As in the think-aloud procedure, Ericsson and Simon (1993) recom-
mended practising retrospective reports, perhaps combined with the think-
aloud practice if participants are asked to do both. The practice tasks in
Examples 8.8 and 8.9 work well for this.
Following Ericsson and Simon (1993), think-aloud protocols and retro-
spective reports are two ways of eliciting verbal data to represent thought
processes as closely as possible. Descriptions, as we discussed above, are
ways of accessing mental representations. Alongside these frequent and
well-established methods of eliciting verbal data, there are many further
options that can be useful for CODA studies. Many of these are discussed
at great length in other contexts. For instance, consider interviews, focus
groups, and questionnaires. All of these methods are useful for a range of
purposes in social science and beyond, and all of them involve collecting
verbal data of different kinds. All verbalisations are expressions of
thought, mediated by the context in which they are elicited and by the
way in which this happens.
Beyond the content of what people say in an interview, focus group, or
questionnaire, CODA adds a new layer of analysis by looking at how they say
it. In Tenbrink et al. (2012), for instance, we used CODA on a set of transcripts
from previously conducted interviews with architects. The specific purpose of
the analysis was to identify, on a linguistic level, how the architects expressed
various perspectives during the interview. Perspective taking in architectural
design had been previously identified as a relevant topic, and CODA added
valuable information on the linguistic options used to express the various
perspectives involved (see Chapter 4 for details on this study). This analysis
was neither based on direct mental representations (as expressed in description
tasks) nor on representations of complex cognitive processes (as expressed in
think-aloud protocols or retrospective reports). Instead, the architects
expressed their thoughts and ideas in answer to the questions given to them
in the interview.
Along these lines, any kinds of verbal data may be used for CODA analysis,
and may prove to be interesting resources. As long as the limitations and
features of each type of verbal data are considered in the analysis, any verba-
lisation data can be relevant. However, some ways of eliciting data may be
more useful and suitable than others in light of your research goals. If you’re
looking for direct representations of current thought processes, use think-aloud
protocols if possible. If you wish to elicit mental representations, aim for the
type of description that will suit your purposes. In all cases, the instruction
should be formulated with care, and the data must be collected conscientiously

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 208 CODA Procedures

and in a controlled way, making sure that all participants receive the instruction
consistently in exactly the way they are supposed to.

8.4 Data Preparation

After collecting your data, you are ready to get started on your analysis. To
facilitate this, you may want to transfer everything into a digital format that you
can work with. If you asked your participants to type their responses into a
computer (which is sometimes done in CODA studies), there is not much to do
at this stage. Just make sure you have stored the data safely and consistently,
and separate from any personal information such as names. Depending on the
format in which the data are collected, you may need to transfer them into a
different software – the choice is yours. If you collected handwritten responses,
you will probably need to type them up, or at least the relevant parts; you will
quickly get to the limits of what you can do on paper.
Many CODA studies involve spoken language, however, and this is when
you need to make a few decisions. Transcription can be exceedingly time-
consuming; can you find software that might help you, or delegate this step to
assistants? As technology is changing rapidly, I will refrain from recommend-
ing any specific software here. To date, automatic transcription results are most
reliable with trained systems, which specialise on one particular voice. This is
unfortunately not feasible for most CODA studies, as they involve many
different speakers. It helps if your data contains no interaction or dialogue,
because transcription (both automatic and manual) is much easier to accom-
plish then. In any case, manual work will be needed to verify any automatic
transcription result. If you are involved with the transcription process yourself,
be prepared to spend quite some time on it. In reward, by the end of this process
you will have very good knowledge of your language data, and this takes you a
long way towards meaningful analysis.
Manual transcription can be assisted by useful tools that help with practical
matters. Without a tool, you’d need to switch back and forth between text
processing software, and the audio recording file, which can be very tedious.
There are various tools available to avoid this kind of hassle. They may enable
you to type what you hear directly in to the file, to scroll back automatically
when you stop and start again, to insert time stamps at useful places, to slow
down or accelerate the speed, and so on. Especially with lengthy transcriptions,
it is advisable to spend a moment looking for a suitable transcription support
tool.
Further, you may want to decide whether you actually need to transcribe
everything that was recorded. This will naturally depend on your research
question and empirical procedures. Some of the produced language may be
directly relevant for your purposes, while other parts may have been recorded

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.4 Data Preparation 209

primarily as a backup. For instance, the instruction to participants must be

identical in all cases, but it may still be recorded, and this provides a chance to
verify that the instruction was given as intended. Such parts may not need to be
transcribed. In some cases, the study may include more than one part, and it
may make sense to get started by transcribing only the first one – again saving
time before analysis can start.
Once you have decided what to transcribe and how to do it, the next major
decision concerns the level of detail in the transcription. The range of options is
rather wide here, and again the choice depends on your purposes. Journalists
may tend to transcribe just the gist or the main content of what people are
saying; they may not care about the exact wording at all times – and that is
perfectly fine for their purposes. For CODA, the exact wording is decisive, so a
rough representation or summary of content won’t do. However, do you also
need information on actions, postures, gestures, mimics, intonation contour,
pauses, hesitations, overlaps in dialogue, and so on? This is for you to decide.
There is a trap here that is very easy to fall into. As a conscientious
researcher, you may be convinced that ‘everything is relevant’ and you want
to ‘do it properly’, in case it may be needed later. However, an extremely
narrow transcription, which contains all of these aspects and potentially more,
will be very time-consuming. In the extreme, it may take up all the time you
have, leaving none for analysis. If producing a well-transcribed corpus is not
the final goal of your research, it therefore makes sense to only include the
kinds of information in the transcript that will actually be used in the analysis.
In most CODA studies, this means transcribing exactly what the participants
said, and possibly including information on pauses that last longer than a
second or two, but typically not gestures, intonation, mimics, and the like.
Further, you may want to decide on a set of transcription conventions that
you use on all your transcripts. This will avoid having to make spontaneous
decisions or include lengthy ad hoc explanations. For instance, imagine your
participant saying something that you can’t hear properly on the recording.
This actually happens quite frequently, and it’s useful to have a short conven-
tion for this. Compare the following solutions in a transcript:

Example 8.10 How to get to, ahhhh, this may mean Highlander Street?
errrrm maybe take a no idea what she says!
Example 8.11 How to get to er (inaudible) . . . er maybe take a (inaudible)
Example 8.12 How to get to, uh (Highlander Street)? (2) Uhm maybe take
a (. . .).
In Example 8.10, no conventions seem to be available, and the transcriber
explains what’s going on as in ‘this may mean Highlander Street?’ and ‘no idea
what she says’. The hesitation sounds are transcribed intuitively, based on what

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 210 CODA Procedures

they sound like (‘ahhhh’ and ‘errrrm’). It is not clear what punctuation markers
mean in this example; the transcriber probably did not intend to suggest that the
utterance was an exclamation. The transcriber’s comments are distinguishable
from the actual transcript by the decision to use italics, which may be fine – but
italics may easily get lost in data transfer to a different file, or overlooked in the
analysis process.
In Example 8.11, everything that is not clear enough to be transcribed is
marked by ‘(inaudible)’, and the convention was adopted to consistently use
‘er’ for hesitation. There is no punctuation except for three dots ‘ . . . ’, which
may denote a pause after the inaudible part. These simple conventions can be
easily adopted and used consistently throughout transcription, but the informa-
tion included is minimal.
In Example 8.12, conventions are a bit more differentiated, providing more
information to the analyst without adding much complexity or confusion. The
hesitation markers ‘uh’ and ‘uhm’ are distinguished by the nasal sound at the
end, which may or may not be relevant for the research purposes. Both of the
inaudible parts are marked by using brackets; in the first case, the transcriber’s
interpretation (‘Highlander Street’) is included, and in the second case, the
convention ‘(. . .)’ indicates that there is speech there for which there is no
attempt at an interpretation. The pause after the first inaudible part is indicated
by ‘(2)’, meaning two seconds. Punctuation is used according to normal writing
conventions, where a comma indicates a slight break in the intonation contour
followed by continuation, a question mark indicates an intonation pattern that
makes it likely that a question was intended here, and a full stop indicates the
end of a sentence (or utterance unit, in this case).
As the examples indicate, transcriptions can look very different, depending
on the conventions adopted. As this will affect later analysis and even inter-
pretation, it makes sense to consider carefully what kinds of conventions might
be most useful for the purposes at hand. With a short transcript, ad hoc
conventions like Example 8.10 may be fine, but if lengthy passages are
transcribed, you will probably want a cleaner-looking version like Example
8.11 or 8.12.

8.5 Practical Steps of Data Analysis

So you have collected and transcribed your data, and everything is ready in a
suitable digital format. What next? Again, the most important consideration is
to be systematic in what you’re doing. You may start from initial intuitions,
and you may be inspired by spontaneous discoveries in the language you
collected – all of this is legitimate and useful. However, the main goal of
CODA is to do systematic linguistic analysis, moving from observations and
intuitive insights to the identification of meaningful patterns. In effect, the

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.5 Practical Steps of Data Analysis 211

procedure often moves from qualitative to quantitative analysis: you expect or

discover the existence of specific phenomena in the data, and then analyse
systematically how often and under what circumstances they occur.
As you start your analysis, you may already have a fairly good idea of what
to look for in your data. Several chapters in this book describe specific kinds of
phenomena that a close linguistic analysis may unveil, and this may guide your
analysis. Keep in mind, however, that the described phenomena are meant to be
exemplary rather than exhaustive. The range of cognitive aspects that are
reflected in language use is extensive, and it is not the intention of this book
to capture all of them. As relevant reflections of concepts or thoughts may
happen on different levels of language, you may need to be flexible in the way
you approach your analysis. Also, you may prefer to adopt analysis procedures
that you are familiar with, or have read about, which may be perfectly suitable
for CODA. Here are a few practical considerations that you might find useful,
depending on your background and research goals.

8.5.1 Segmentation
Before you start analysis, you may want to think about the size of each chunk of
language that you analyse at a time. From a theoretical standpoint, there are no
constraints: units of analysis can be as large as millions of words (in computer-
assisted corpus analysis), or even smaller than a word at a time. However, not
all unit sizes will be useful for you.
Sometimes your data will come in a pre-segmented way – in useful chunks
that you can analyse as they are. If you give participants ten images and ask
them to describe each of them briefly in a single sentence, then each image
description may be a feasible and meaningful unit of analysis. And for some
types of analysis, the size of the data unit to analyse does not matter. If you
look for the occurrence of individual words or phrases that you can easily
search for in a data set of any size, then you won’t need to segment the data
at all.
However, for most CODA studies, segmentation into smaller units will
facilitate analysis. There are two main reasons why you may want to segment
your data: for practical purposes, and for purposes of reporting patterns in the
results. If segmentation serves practical purposes only, you can do this intui-
tively, based on what works best for you in light of the research question at
hand. If you aim to report distribution patterns of segments in a publication, or
if you do quantitative calculations based on the number of units, then you will
need to operationalise the segmentation process to make it meaningful and
reliable (see Krippendorff, 2012 for very useful discussions of this).
Very often, the idea of a sentence or independent clause (or clause complex)
is a good starting point for purposes of CODA. Their size means they normally

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 212 CODA Procedures

contain a limited number of features that may be interesting from a linguistic

point of view. For instance, if you’re looking at the kind of perspective used in a
particular spatial task, you will normally be able to determine the underlying
perspective used in each single sentence or clause. And if people change
perspective mid-sentence, this is interesting and worth considering more
closely.
In written language, sentences can easily be identified by looking at punc-
tuation. However, it is often more useful to choose a solution that is anchored in
grammar, rather than how writers happen to place their full stops. Compare the
following:
Example 8.13 The hat was dirty, and the gloves he was wearing were torn.
Example 8.14 The hat was dirty. And the gloves he was wearing were torn.
Both of these are identical in their semantic content, and they may sound the
same in spoken language. If this is part of a transcript, it is very much up to the
transcriber to choose where to put a comma or end a sentence. From a
grammatical point of view, they both contain two independent clauses. Thus,
with a segmentation that is based on independent clauses, both examples would
yield two separate units, whereas a sentence-based segmentation has the entire
Example 8.13 in one unit.
Notice that the second independent clause has an embedded clause within it,
‘he was wearing’. This clause cannot easily be separated from the main clause,
as it depends directly on it. For this and similar reasons, clauses may not be a
good basis for analysis; this is why I suggested focusing on independent clauses
only. However, there may be reasons for looking at the clause level, using both
dependent and independent clauses as separate units. Again, the decision is
yours, but do aim to do this consistently.
Especially in spoken language, it is not always straightforward to
identify sentences or clauses. Spoken language typically contains unfin-
ished sentences, repetitions, hesitation markers, false starts, and other
complications. In such cases, it may make sense to use the notion of a
‘possible sentence’ (Selting, 2000). This can be intuitive to some extent,
but intuitions will be informed by syntactic, semantic, pragmatic, and
intonation-based aspects. Syntactically complete sentences can always be
treated as separate units.
Once you have determined your units, it makes sense to put them into
separate lines or rows in the software you will be using for analysis, for instance
in an Excel file. I will in the following refer to Excel as a simple example for
how standard software can be used for systematic analysis. The principles are
transferable to other tools. If you use Excel, put each unit of analysis into a
separate row, one after the other, leaving no empty rows. If you then number the
rows (which Excel can do semi-automatically), you will have a separate ID for

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.5 Practical Steps of Data Analysis 213

each unit of analysis, as well as being able to see the total number of units that
you will be working with.

8.5.2 Content Analysis

Before starting to take a closer look at the features of the individual units of
analysis, it is essential to gain a clear understanding of their content. What
people say is as important as how they say it, in terms of linguistic and
conceptual features. Content (or discourse) analysis is sometimes all that is
done on language data, and this can yield essential insights relevant to the
purposes of the study. As Krippendorff (2012) and others offer very good
insights into relevant procedures and concepts, I will focus here on practical
and CODA-related aspects only.
Content analysis means taking a close look at what people say, in each unit of
analysis. For most purposes, it is useful to identify a limited range of content
categories, defined in light of their relevance to the research purpose. This is
useful for three reasons. First, labelling different units with a short character-
isation of the content provides qualitative insights into verbalisation content
produced in this kind of task. This can be a valuable result in itself that may be
worth reporting in a publication. Second, this analysis gives a good overview of
the structure of the data, in terms of what content is produced at what time, from
a sequential perspective. Again, a systematic sequential analysis of content can
lead to extremely valuable results. In fact, most traditional approaches to the
analysis of verbal problem-solving data target just this aspect (see, for instance,
Ericsson & Simon, 1993).
However, there is a third aspect that is directly relevant to CODA. Content
analysis provides a basis for specifically targeted analysis of linguistic features
pertaining to certain types of content only. To start with, not all content
categories may be targeted in the analysis or suitable for further linguistic
scrutiny. For instance, some units may be suitably labelled as ‘introduction’ or
‘reflections on the task’ or ‘off-talk’, indicating that these verbalisations are not
representations of the task given to participants, but rather reflect some
thoughts around the task, in a sense. Even if relevant they do not reflect
thoughts on the actual task, and so it makes sense to separate them out of the
analysis, and treat them differently in subsequent analysis steps.
Further, particularly in a problem-solving task, people think of different
things at different times, which is reflected in different content categories.
Some linguistic analyses are best done on specific content categories only,
whereas others may be meaningful for the entire data set. This should be
considered carefully, in light of analysis purposes and types. If specific types
of content are linguistically expressed in a limited range of ways, it may be
feasible to characterise these categories on a linguistic level. This can serve to

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 214 CODA Procedures

define and support the content categories, or it can be used to establish

correlations.
For instance, our CODA analysis in Tenbrink and Taylor (2015) was heavily
based on content analysis, supported by a range of more detailed linguistic
analyses. As reported in Chapter 6, we were interested in what people said, and
in what order, while solving an origami paper-folding task, and further in how
their creativity was expressed in their use of spatial language. Particularly the
content category ‘Reformulating description without new thoughts’ was based
on a close look at the formulation, as we identified utterances that were almost
identical to the written instruction given to the participants, but differed from it
in superficial aspects such as changing ‘you’ to ‘I’.
Other content categories were not characterised further on a linguistic level,
but there was one category that was clearly associated with spatial language:
‘spatial description’, a subcategory of ‘Additional ideas about this step’ during
paper folding. As expected, we found a high correlation between the frequen-
cies of this content category and the frequency of producing new spatial terms.
The spatial terms analysis was done on the entire data set, yielding a range of
interesting patterns as reported in Tenbrink and Taylor (2015), and it also
supported our content category ‘spatial description’ through consistent correla-
tion patterns.
In Excel, content category annotation can be done simply by inserting a
column next to the utterance in each row, which allows you to enter a label in
the cell next to the utterance. If required, further columns next to the basic
content analysis can be introduced to identify content subcategories, to the
extent that this is relevant for the research purpose at hand.

8.5.3 Tool Support: Filter, Word Count, and Automatic Search for Words
and Phrases
At this point, let’s consider a few features that Excel, as well as many other
software tools, can offer in support of systematic analysis. To start with, I
mentioned in Section 8.5.2 that not all content categories may be treated in the
same way. To support focusing analysis on specific categories only, a Filter
function can be very helpful and practical, as it allows the researcher to view
and work on only a subset of data at a time, filtered according to their current
analysis focus.
Another useful function is to count words. Excel can count words in a cell,
using a specific formula – as these things change, it is best to search the web for
this function. Word counts are important measures for quantitative analysis,
both in terms of the entire data set and in terms of words used for individual
content (or other) categories. Therefore, it is often useful to allow one column
for word count of the cell containing the verbal data unit.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.5 Practical Steps of Data Analysis 215

Figure 8.1 Screenshot of Excel-based analysis done in Tenbrink and Taylor

(2015).

Figure 8.1 gives an example taken from our origami study (Tenbrink &
Taylor, 2015). Column A gives the number of the unit of analysis, and column
B the participant code (13). Column C captures who is speaking the utterance
that is given in Column D. Since there was no interaction, this is almost always
the participant (P), but there may be clarification questions answered by the
experimenter (E). The utterance unit in column D contains a time stamp,
contributed by the transcription software used in this study (called F4). This
information is transferred to the next column, E (fold time), enabling the
analyst to identify the relevant times at a glance. Column F automatically
counts the number of words in column D; since this includes the time stamp,
the actual number of words is that number minus 1 (this was taken into account
later). Column G identifies the current task given to the participant: folding the
flower’s stem following the origami instructions. Since the instructions came in
steps, we could also specify the utterances in relation to what the participant
was working on, captured in column H, ‘step’.
In column I, content analysis starts with the content category ‘Reading
task description’. Unit no. 2 ‘put the paper in front of you’ was directly
read from the instruction, therefore this content is copied over into cell I3.
Column J automatically counts the number of words read as represented in
column I, using a formula in Excel. Unit no. 3 ‘corner pointing towards
me’ was classed as ‘Reformulating task description’ (column K). Further
word counts, content categories, and linguistic analyses were added to the
right of these columns.
The use of the filter can be seen in the second row of this example. The small
arrows underneath the column header appeared when clicking ‘Filter’.
Clicking on the arrows allows us to see only those rows within each column
that correspond to a certain criterion. For instance, it is possible to show only
the rows that pertain to step 1 in column H, or that contain no words read (0 in
column J). This filter function can be very helpful when focusing on a subset of
the data.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 216 CODA Procedures

Many CODA studies rely heavily on the identification of key words or

phrases that signify specific aspects of the conceptualisation or thought process
at hand. It often makes sense to start from manual analysis in order to identify
such key terms. But once key terms have been established, the task of identify-
ing them in the data can be delegated to the tool, saving much valuable time.
Excel, for example, can not only count words in a cell, but also the number of
occurrences of a specific word in a cell (using another formula that can be found
on the Internet). This can be done multiple times, in different columns. So if
you want to see how the pronouns ‘you’ and ‘I’ are distributed in the data, let
Excel do the search for ‘you’ in one column and for ‘I’ in another. Your analysis
file can contain an unlimited number of columns with automatic word searches
within the verbal data unit. This takes almost no time, and it allows the analyst
to view at a glance how certain key words are distributed within the data.
If the distribution of a particular word is less important than the distribution
of a concept that may be expressed by several different words, it is also possible
to let the software identify and count more than one word at once, within just
one column. This takes up less space, and allows the researcher to update
calculations more quickly if a word is added to the search within a column.
Finally, software tools are obviously very suitable for calculations of any
sort. This is a clear advantage over paper-based colour coding or other manual
analysis that does not involve specific software. Once the analysis is accom-
plished in digital format, it can easily be turned into numbers, which can be
processed further in a range of ways including statistical analysis. For example,
you can sum up the number of words, or of occurrences of a specific word, or of
any other identified feature produced by a participant, or produced within a
content category. Such summary data are best represented in a separate file or
tab, with a line for each participant so that all individual results can be accessed
quickly (see Section 8.7).

8.5.4 Operationalisation of Linguistic Categories

Not all linguistic features are characterised by specific key words that can be
identified (more or less) automatically. For example, you might be interested in
verbalised thoughts that the speakers take for granted – see our discussion of
information structure and presuppositions (implicit assumptions) in Chapter 3.
If no automatic procedures (parsers or software tools) are available that can
identify a target structure for you, you will need to do manual coding. Be
warned: this can be quite time-consuming, and it requires inter-coder reliability
tests (see Krippendorff, 2012, Tenbrink, 2015, and many other resources for
details of this aspect).
Annotation of linguistic features is quite the opposite of intuitive, example-
based analysis. Rather than picking out particularly striking instances of a

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.5 Practical Steps of Data Analysis 217

particular feature, the aim in CODA is to look for patterns. How frequently, and
under what circumstances, does the feature occur? Consider the example of
presuppositions. Perhaps your study is directly relevant for gaining insights on
presupposition use from the outset, motivated from the previous literature or
your own research. Or you might discover, while gaining a deep understanding
of your data during transcription and content analysis (or while listening and
reading through the materials), that some people take specific aspects for
granted, while others don’t.
To start with, this is just an impression, not a fact that you can report – except
for quoting examples. It may be a good idea to refine this impression towards a
more specific hypothesis, aiming to get crystal clear what exactly your data
may be telling you. Do you think there’s a particular moment in the task
procedure where some people take specific things for granted? What distin-
guishes the moment from others? What distinguishes people who do this from
those who don’t? What kinds of things are taken for granted, and which aren’t?
There are endless possibilities here, and only one solution: you need to examine
your data systematically, so you get a clearer idea of what’s going on. In other
words, you need to go through your data file line by line, and identify the
information structure for each single unit.
This will come with a number of challenges. It is worth thinking carefully
about what you wish to do, and why. Once you start annotating, you are likely
to spend quite a lot of time thinking about individual cases. Make sure you
really need each of the annotations and codings that you do, and make sure that
you know exactly what you are doing. For this purpose, you will need a very
clearly formulated operationalisation of each analysis feature. Keeping a sepa-
rate sheet or document in which you clarify or define what exactly happens in
each column in your analysis sheet may be helpful. Some of these will be as
easy as ‘Number of words in a unit’, but once you get started on manual case-
by-case annotation, you will need a more elaborate annotation definition.
To continue our example, the concepts of presupposition and information
structure may be clear enough from a theoretical point of view, but not all of
your linguistic knowledge will be easy to turn into a clear-cut annotation
process. The fact that some thoughts are taken for granted can be expressed
in more than one way on a linguistic level, and you may need to decide which of
them are useful for you to annotate. One very simple linguistic feature indicat-
ing presupposition is the definite article, as it indicates that something is already
known in the discourse context, for whatever reason. In a verbal protocol or
description, the definite article therefore indicates that the speaker is not
introducing this particular element as new, as in the following examples:
Example 8.15 The red flag is partially hidden in the picture.
Example 8.16 I cannot see the red flag very well.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 218 CODA Procedures

Whatever the context of these verbalisations may be, the speaker clearly
indicates that the existence of a red flag is presupposed. If it wasn’t, it would
have been introduced by a phrase such as a red flag. The definite article is
therefore uniquely clear in indicating presuppositions, and can be easily anno-
tated. In the operationalisation, it makes sense to start by identifying all definite
articles (which can be done automatically as explained above), and then, in a
separate step (or column), indicate the phrase that is presupposed (such as ‘the
red flag’). This means considering each and every instance of a definite article
in terms of what it signifies. Cases of doubt may then lead to further specifica-
tions to be entered into the written operationalisation of this analysis feature.
The goal of operationalisation is to enable annotators to do exactly the same
analysis consistently throughout the entire data set. Even if you are doing the
entire analysis on your own, you might find that spelling out exactly what you
are doing will be very helpful both during the process and for future reference.
With a very clear operationalisation, you should also be able to delegate the
annotation task to someone else who would get to the same results as you. This
is called inter-coder reliability, and it can be evaluated by measures such as
Krippendorff’s Alpha (Hayes & Krippendorff, 2007). Most journals expect a
high inter-coder reliability value for manually annotated data in a publication.
Once you have broken down your targeted analysis feature into operationa-
lisable, identifiable aspects, it should be relatively straightforward to work
through the entire data set. An analysis procedure that relies mostly on intui-
tions and ad hoc decisions is awkward and time-consuming, and will almost
certainly not be reliable in the end. In effect, if you don’t invest the time
required to operationalise your analysis meaningfully, you will invest much
more time into doing a somewhat speculative analysis that you may not be able
to publish, as it lacks systematicity. It is much better to identify a suitable
operationalisation that can be clearly outlined and explained to future readers.
Ideally, procedures and decisions should be grounded in the literature suffi-
ciently to be meaningful in the context of the research undertaken. For instance,
the literature on presuppositions incorporates extensive discussions of the
contribution and role of the definite article. This body of research should be
considered in the operationalisation. On this basis, you should be able to
subcategorise instances of the definite article on the basis of what they refer
to: the previous discourse (if the red flag was mentioned before), the situation
context (the red flag is directly visible), uniqueness (as when talking about the
moon), and so on.
Other cases may not be as clear-cut, and it may require some research to
identify relevant resources. Sometimes creative solutions for operationalisa-
tions may be the only way, as each study will open up some kind of new area of
research. Be prepared to be innovative, but always do your analysis system-
atically, and be clear about what you are doing.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.5 Practical Steps of Data Analysis 219

To take another example, you may want to analyse the types of spatial terms
that speakers use in your study. A range of resources that classify spatial terms
are available to guide you (e.g. Bateman et al., 2010), but you will still need
your own definition as to how to operationalise the categorisation of spatial
terms in your study. One way to do this is to specify the range of terms that are,
for the purposes of your study, classified as one of the following:
(1) Projective: Terms (typically prepositions) whose ‘interpretation relies on
the additional step of fixing an underlying reference system . . ., defined
relative to the speaker or hearer, or with respect to some object’s intrinsic or
imposed orientation properties (‘my left’ vs. ‘your left’, ‘the front of the
church’, or ‘in front of the rolling ball’, etc.).’ (Bateman et al., 2010:1030).
Examples: front, back, right, left, up, down, above, below, . . .
(2) Topological: . . .
(3) Path-related: . . .
(4) Compass based: . . .
(5) Other
This example demonstrates three essential aspects of linguistic annotation.
First, it is useful (though not always feasible) to have a clear list of options.
If you annotate the spatial terms in your study, you should be able to choose
from a limited range of options. So, upon encountering the phrase to the left in
your data, you can choose the feature ‘projective’ in the annotation file. Second,
define each option on the basis of the literature as far as possible, as well as
providing examples, as done here for projective terms in (1). Third, to allow for
further cases without requiring new decisions with every new word, an ‘Other’
category (as in (5)) is very useful. Likewise, the examples provided in (1) are
not exhaustive, allowing for annotating further words in the same way if they fit
the definition.
Along with this, your annotation definition file should be explicit as to
whether annotation variants are mutually exclusive, or whether it is possible
to attach more than one feature at a time to a word or unit. If the above list is
used to annotate all spatial terms within a unit of analysis such as a sentence,
then it is unlikely that the features are mutually exclusive, as spatial terms may
be combined in a phrase or sentence.
Mutually exclusive features are typically preferable, as they facilitate quan-
titative analysis. It may take some consideration to prepare your annotation file
in such a way as to enable mutually exclusive annotation. For instance, the
above list of spatial terms could be broken up into different columns (if done in
Excel). This allows for the number of spatial terms of a particular category
(projective, topological, etc.) to be counted separately for each unit of analysis.
Alternatively, you may extract individual spatial terms from your language
data, and annotate them one by one, choosing from the above list. Both methods
render your list of spatial terms mutually exclusive.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 220 CODA Procedures

In sum, following a good grasp (and possibly categorisation) of the content

of your language data, you will typically decide on a set of linguistic categories
that may either be semi-automatically identifiable or require manual annota-
tion. Wherever manual annotation is involved, it is most decisive to develop a
clear and consistent operationalisation. This can start from intuitions, but
should be iteratively developed until it reaches a state where it can be easily
communicated to and adopted by other analysts.

8.6 Qualitative Insights

Most CODA studies combine qualitative and quantitative insights. As
described in Section 8.5, analysts often move from the observation of a
particular feature via a hypothesis as to its distribution to the identification of
patterns in the data, based on systematic annotation. However, my emphasis of
systematic analysis doesn’t mean that qualitative insights aren’t valuable, or
(taking these thoughts further) that only statistically significant results should
be considered. Let’s take a closer look at the range of qualitative insights that
can be gained from a close analysis of language data in relation to cognition.
First, the mere existence of a particular feature in the data can be a valuable
result if it is unexpected or previously unreported. In Hölscher et al. (2011) (see
Chapter 7), one of our research questions concerned whether people plan a
route ahead if they know the environment well. Previous literature had been
unclear on this point, and so it was regarded as a real possibility that the entire
route plan is laid out in people’s heads in advance, requiring no further planning
effort during navigation. Alternatively, they might not bother to plan ahead
very much at all, and simply head off in a direction, specifying the details of
their route as they go.
Supporting the second alternative, we identified multiple indications of
ongoing planning processes in the think-aloud protocols during navigation.
All participants referred to intentions, decisions, and current planning pro-
cesses as they walked to their goal. Among the relevant linguistic markers for
planning processes we found modal verbs such as could, should, might, verbs
of thinking such as consider, think, decide, and discourse markers such as
whether, or, if, better, and maybe. These linguistic choices cannot be explained
if all of the planning is completed beforehand. Since there was no comparison
to be made in this regard, the existence of these markers was an important
qualitative result in our study, and the analysis could have stopped here.
However, we added detailed information on the quantitative distribution of
these markers in the data, in order to be as specific as possible.
Second, it can be very interesting to scrutinise individual participants’
conceptualisations and cognitive processes. Case studies of conceptualisations
or thought processes can serve as illuminative examples; they demonstrate the

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.6 Qualitative Insights 221

existence or possibility of a particular way of thinking, and show in detail how

this might work. In Tenbrink and Salwiczek (2016) (see Chapter 4), we set out
with the sole intention to identify quantitative patterns in the linguistic features
of the think-aloud data collected in our study. Much of our report does indeed
represent statistical distributions, but there is also a sense of certain limitations
of this approach. Some of the results simply did not fit into the overall picture,
and the overall picture itself did not correspond clearly to any of the previously
identified hypotheses.
To gain clarity on these issues, we needed to take a closer look at individual
cases. It turned out that our participants used a wide range of cognitive
strategies to solve the task given to them. The different verbalisations were
highly revealing about the range of possibilities in this respect. Presenting and
discussing individual case studies in our report therefore allowed us to account
for the unexpected quantitative patterns in our results much more clearly.
Finally, sometimes the distribution of patterns in the data as such can be
reported in (more or less) qualitative terms. The idea here is to provide a
general impression or a first understanding of what is happening in the scenario
at hand. Not everything can be expressed in terms of statistical significance,
and not everything needs to be. Unconstrained language production data, as
collected in CODA studies, are notoriously difficult as a basis for statistical
analysis; there are limits to what a researcher can realistically expect in this
regard. Nevertheless, a qualitative description of a certain balance in the data
can be interesting and revealing, and it is likely to lead to the formulation of
more specific hypotheses that can be tested in a targeted way. On this basis,
reaching statistical significance in targeted follow-up studies is far more
realistic.
In spite of the clear benefits of reporting qualitative findings, you might find
that they are more difficult to publish, and some reviewers may advise against
including observations that cannot be ‘proven’ statistically. There is a certain
danger in assuming that a p-value can actually decide about the extent to which
a particular finding is relevant for the academic audience of a publication.
Statistical significance should not be confused with relevance; both are impor-
tant, and have, or should have, their rightful place in academic reports.
To make a particular qualitative result relevant for a readership, it is perhaps
even more important to observe the principles of rigorous analysis. It is
absolutely necessary to be as specific as possible, providing details gained
through in-depth analysis. It is equally essential to relate observations to the
previous literature, contributing insightfully to a theoretical discussion rather
than speculating about the possible relevance for it.
Perhaps the most important trap to avoid concerns unwarranted general-
isations. Ad hoc individual examples picked out from a data set do not
necessarily add up to a pattern. Without further analysis or a clear relation to

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 222 CODA Procedures

the literature, they do not even represent a qualitative result. The mere fact that
a phenomenon is found in your data does not necessarily mean much at all, if
this fact doesn’t answer a question, clarify a controversy, or represent a new
discovery. The fact that one person uses language in a particular way does not
prove that these linguistic features are predominant in your data set, or that
other people in the same situation would do the same. Your impression that
something happens more than something else is not enough – you will need to
show this systematically, through clearly operationalised analysis that looks at
the entire data set in an unbiased way. Typically, quantitative formulations like
‘more’ or ‘rarely’ in a report will invoke the natural expectation that numbers
should be provided, and a statistical test should follow.

8.7 Quantitative Patterns

Moving on from qualitative insights, you may wish to identify the ways in
which the observed phenomena are spread across participants. Most CODA
studies involve different experimental conditions, designed to identify the
influence of a specific factor on the linguistic choices that speakers make.
Comparing the results in relation to conditions is therefore a natural next step
once annotation is completed.
If you followed the procedures in Section 8.5, using a software tool like
Excel, you will now be looking at an extensive analysis sheet with many
annotations and automatically generated numbers. To get a clearer idea about
patterns, you will need to transfer summary results into another sheet. Here, I
would recommend allowing one row per participant, with separate columns for
participant number, demographic data like gender, age, and native language,
overall word count, and then the more specific analysis results: for example,
word count in a particular subtask, overall count of each linguistic feature in the
participant’s data set, count of linguistic features in a particular subtask, and the
like. There can be an unlimited number of columns, showing different sum-
mary results in overview fashion.
Figure 8.2 gives an example for an analysis spreadsheet, again (like
Figure 8.1 in Section 8.5.2) from our origami study (Tenbrink & Taylor,
2015). Columns A–C give participant number and demographic data, and
there are a number of further columns (not visible here) with further
information pertaining to individual participants. Column S gives the
total number of analysis units produced by each participant, and column
T the number of words. Columns U and V reflect numbers of occurrences
of individual content categories that we identified in the data (additional
ideas about object quality and spatial description, respectively); for
instance, participant number 1 mentioned spatial descriptions five times,
but never commented on object quality.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.7 Quantitative Patterns 223

Figure 8.2 Spreadsheet for the analysis in Tenbrink and Taylor (2015).

On this basis, you can then begin to do some easy calculations. To start with,
it often makes sense to look at relative frequencies, rather than absolute
numbers. Participants will vary a lot in the amount of language they produce
in a task. Imagine that person A produces 150 words in a task, and person B
produces 40 words. Now, if you’re investigating a linguistic feature such as
presuppositions, operationalised on the basis of definite articles, you might find
that person A used ‘the’ ten times, and person B used ‘the’ four times. Does this
mean that person A relied more on presuppositions than person B?
With this simple example, you’ll probably see right away that this can’t be
true, as Person A used more words overall. To calculate this effect, we can simply
divide the sum of definite articles by the overall number of words, and we’ll find
that person A used ‘the’ with a frequency of 10/150=0.067, as opposed to person
B with 4/40=0.1 – so the frequency at which ‘the’ is used is higher for person B.
If persons A and B were part of different conditions, we could then look at
whether there is an overall pattern there: perhaps person B’s condition generally
leads to the use of fewer words but more presuppositions, for whatever reason.
Calculating relative frequencies on the basis of the overall word count is the
simplest way of looking for patterns. However, this is not always meaningful;
sometimes you may want to look at units rather than words. For instance, in
Vorwerg and Tenbrink (2007) (see Section 8.3) we asked participants to
describe simple pictures showing abstract object configurations, one at a
time. Answers varied greatly in length, such as the following (taken from our
German data set):
Example 8.17 über dem Quadrat
[above the square]

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 224 CODA Procedures

Example 8.18 das mit einem x markierte Quadrat ist oberhalb etwas mehr
links von dem anderen Quadrat
[the square marked by an x is higher than and somewhat
more to the left of the other square]
While word count mattered in many ways, it was also important for us to
look at the linguistic features within a unit, which in this study corresponded to
a picture description, regardless of length or syntactical construction.
Therefore, many of the calculations in our analysis were based on units rather
than relative to the overall word count. Note that this decision depended on the
fact that units were very clearly defined. In this study, this was easy because
picture descriptions were short enough to lead to feasible units of analysis.
Overall word count and units are often useful as a basis for calculating
relative frequencies, but they may not be the only options. You will need to
consider carefully what kind of percentage makes sense in your case. For
instance, you might wish to calculate the frequency of a particular type of
linguistic feature relative to all words produced in a particular content category
only, rather than all the words in the data set. Similarly, you might wish to look
at the relative frequency of a certain word within all occurrences of a particular
linguistic feature.
Calculating meaningful relative frequencies and comparing them will
give you a first indication of possible patterns in the data. For some
purposes, presenting such patterns in a descriptive way may be sufficient.
However, in most cases inferential statistics will be required so as to test
for significance, moving beyond a mere representation of linguistic patterns
in the data. This book is not the place for introducing statistical methods –
there are plenty of very useful resources for this. Once you have identified
possible patterns, you should aim to find suitable statistical procedures to
test them. In many cases, ANOVA, t-tests, chi-square, or regression tests
will be good choices in CODA studies, but sometimes more complicated
procedures may be required.
Statistical tests are most meaningful if they are based on well-founded
hypotheses. With the procedures described here, you might find multiple
interesting patterns in your data that may or may not be statistically significant.
It is important to avoid getting lost in the data or in the wealth of possible
comparisons. Typically, the best starting point for statistical analysis is the one
that motivated your study. Start with the patterns and comparisons you aimed to
make in the first place. Compare results between conditions, as they were
designed to induce differences. Keep your research question in mind: not all
patterns are meaningful from a theoretical point of view. Everything may be
related to everything else, but it still doesn’t make sense to look at every single
possible interrelationship.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 8.7 Quantitative Patterns 225

Instead, it is advisable to keep statistical tests within reasonable limits. If you

do multiple calculations on a complex data set, you are likely to find some
statistically significant differences – but they are not necessarily interesting.
There is a non-trivial chance that some statistical tests will turn out to be
randomly significant. For this reason, some journals require that all statistical
tests that were carried out need to be reported, regardless of whether they turned
out to be significant and further pursued. So it is wise to be careful. With a
theoretically well-motivated study, you should be able to gain relevant statis-
tically significant results on the basis of only a few (ideally pre-planned)
calculations; and if you don’t, this might be an interesting result in itself.
These considerations also relate to the idea of randomisation in your design,
as mentioned in Section 8.2. Factors like sociodemographic background,
gender, order of tasks, and the like will almost certainly affect your results in
some way, but you cannot concentrate on all of them at the same time. If you
aim for statistical comparisons on any one of those factors, make sure to design
the study in such a way that this factor is a central one, implemented as an
independent variable in the conditions of your study. Otherwise, if not central,
keep them random. Randomised factors will normally not be targeted in your
statistical analysis, other than ruling out possible distorting effects. If in doubt,
do consult a statistics expert on these considerations.
Most importantly, your quantitative analysis must be motivated from a
theoretical perspective, and this will be the basis for the ‘story’ to tell in a
published report of your study. Reporting numbers as such, including poten-
tially random statistically significant results, is rarely meaningful. However, set
against a strong theoretical background, even non-significant or marginally
significant patterns can make a compelling point. In that case, statistical sig-
nificance can only make the effect stronger.

This chapter has laid out some of the main procedural considerations that have
guided CODA studies in the past. It is worth keeping in mind that the basic idea
for CODA is to analyse language use in relation to thought; the precise way in
which this is done is up to the researcher and will depend on the research
question at hand. In this light, this chapter may provide practical guidance for
readers new to research of this kind, and it illustrates how procedures may be
carried out without requiring specialist software or any other more sophisti-
cated tools. Experienced researchers may decide to carry out a CODA study
using a very different approach. Any systematic study of language use that
sheds light on cognition certainly falls within the scope of ‘cognitive discourse
analysis’, CODA.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 12:48:04, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9 Beyond CODA

Identifying thoughts and concepts through language analysis is fun and can be
intensely rewarding and enlightening in itself. New discoveries, answers to
questions you’ve had for a long time, support for an assumption that you
couldn’t prove before, or evidence that you’ve been completely wrong about
something: all of these things can happen as you set out to identify patterns in
discourse, making your research relevant to yourself and others.
But CODA is not necessarily a tool to be used in isolation. More often than
not, CODA will be part of a larger endeavour, one analysis method alongside
others. My own involvement with the method originated in the necessity to
access linguistic evidence systematically in interdisciplinary projects working
towards goals that were frequently practical. For instance, how can we enable
a dialogue system to interact naturally with users (Cuayáhuitl et al., 2010)?
How do we make a robot understand intuitive commands (Moratz & Tenbrink,
2006)? How can we make sense of people’s responses to a seemingly simple
behavioural task (Tenbrink & Salwiczek, 2016)? Or simply, how do we under-
stand how people actually plan routes (Hölscher et al., 2011)?
As language is often just one way in which task-related thought processes
and concepts are represented, CODA can easily be combined with other
methods. Likewise, analysis results can be used for a multitude of applications
and purposes, including implementation in automatic systems that support
human everyday needs. This chapter will explore these kinds of extensions to
CODA, using examples from earlier work to illustrate the scope. Section 9.1
looks at triangulation with other methods, and Section 9.2 outlines a range of
applied purposes. Just like most insights shared in this book, there is no
intention to be exhaustive; this chapter is meant for inspiration, highlighting
pathways towards fruitful triangulation and innovative applications.

9.1 Triangulation: Combinations with Other Methods

CODA as a method emerged from an interest in the human mind, and as such
can be regarded as one of many methods in cognitive science. At the same time,
it is informed by insights and experiences from linguistics and discourse

226

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 227

analysis, as explored in Chapter 3. In these and other ways, CODA naturally

links up with various disciplines. A linguist may combine a CODA study of
natural discourse with a wider discussion of grammatical phenomena, set
against theories of form and function, or inspired by insights in cognitive
linguistics (see Carroll & von Stutterheim, 1993, and Ferrari, 2007 for related
approaches). A psycholinguist may identify patterns in linguistic choices that
shed light on language production processes in the brain (Habel & Tappe, 1999;
Vorwerg, 2009). A critical discourse analyst may find substantive evidence
from a cognitive perspective (Hart & Lukeš, 2009; Knapton & Rundblad,
2014). And a sociologist may relate demographic patterns identified through
questionnaires to a CODA analysis of natural discourse, collected through
descriptions, interviews, or focus groups (Kristiansen & Dirven, 2008; Pütz,
Robinson, & Reif, 2014). CODA, then, contributes one decisive part of the
evidence, allowing for an in-depth discussion of the bigger picture emerging
from the entire data set. Such combinations can bridge important gaps, such as
those between interactional and cognitive approaches to language.
Within cognitive science, methods to access the human mind are abundant.
In the following sections we will take a closer look at a few of them, to see how
each plays out in combination with CODA. Section 9.1.1 will start with directly
observable data such as behavioural performance, Section 9.1.2 will look at
subtler cues such as reaction times, eye tracking, and neuroimaging data, and
Section 9.1.3 addresses cognitive modelling. Beyond cognitive science,
Section 9.1.4 will look at CODA in relation to corpus linguistics.

9.1.1 Performance and Other Observable Data

Many tasks that involve collecting language data in relation to thought also
come with some kind of challenge. That challenge can typically be measured in
a number of ways. People may be more or less successful in doing the task, they
may solve it in more than one way, and they may be more or less efficient. In
fact, this variability is frequently one of the main reasons for doing the study in
the first place: what kinds of pathways do people use for solving the task, how
exactly do they do it, and what is the most promising or successful pathway?
Obviously, what exactly can be measured along these lines, apart from
language data, depends on the study. Having clear success criteria and defining
suitable ways to assess solution pathways are among the most important design
criteria for a study. Not all studies involve an element of success; this is not
a necessary prerequisite for research on human thought. For instance, if people
are asked to describe a film they saw or a scene they remember, there may be no
clear success criteria. One may measure the number of words or of details
described, or look at the time people take, or analyse the diversity of observed
phenomena; all of this may shed light on the research question behind the study.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 228 Beyond CODA

Here, I will illustrate triangulation with performance data by using another

CODA study as an example. Our collaborative work in Gralla et al. (2012) was
motivated by previous research that had identified two distinct ways of solving
problems on the basis of analogy. Analogy is a powerful way of solving
problems by comparison with other problems that one has seen before. For
instance, if you are trying to solve a riddle, it will often help you to think of
other riddles of a similar kind. Many problems relate to other problems in some
way; the more experienced you are, the easier it will be for you to solve new
problems based on analogy with known ones.
The two ways of doing this are as follows. On the one hand, the problem
solver might find that the problems are similar enough to enable a simple
transfer of the solution from one problem to another, with only slight adapta-
tions to match the new problem. This is called transformational problem
solving because the solution is transformed as needed. On the other hand, in
derivational problem solving, the analogy lies in the problem-solving process
itself. The new problem is solved by going through the solution steps one by
one, solving the new problem using the solution pathway from the known one.
It seems clear from this distinction that derivational problem solving
involves a deeper understanding of the problem and its solution. A simple
transfer of the solution might mean that the new problem wasn’t considered in
depth at all. Therefore, we were interested in finding out more about when
people choose which strategy, and how they could be guided towards using the
derivational problem-solving strategy for problems with clear analogies.
We (Gralla et al., 2012) adopted a study design that combined the collection
of language data with other methods that had previously been used to look into
strategy choice. Participants were first encouraged to think of analogies
through a story where one problem was solved on the basis of analogy to
another problem. After reading the story, they were shown an example problem
that involved identifying a complex path trajectory, given a set of constraints.
The solution was explained in detail, supported by a line graph with letters
indicating places (a so-called Eulerian trail). Next, they were presented with
another problem that could be solved using a similar strategy, so they should be
able to use analogy.
There were two conditions that were intended to support either the transfor-
mational or the derivational strategy. In the high-guidance condition, the
initials of target words used in the second problem were identical to those in
the example problem, which meant that the same letters could be used to find
the solution graph. This would enable a direct transfer of the solution from one
problem to the other. In the low-guidance condition, this was not the case,
necessitating a more in-depth problem-solving process using the derivational
strategy. Furthermore, we varied conditions with respect to the collection of
language data. Whereas one-third of participants did not produce any language

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 229

apart from responses to a questionnaire, two-thirds of participants were asked

to do two written tasks each. In particular, altogether we collected the following
data:
• Demographic data.
• Correctness of solutions.
• Overall time needed.
• Time needed specifically to map the letters onto the new problem.
• A third of the participants were asked to write planning protocols describing
how they would solve the task before they actually solved it.
• After they solved the problem, these same participants were asked to write
evaluations of their previous plans.
• Another third of the participants were asked to write reports on the problem-
solving process after they solved it.
• Then, participants in this group were also asked to write an instruction for
a friend to solve the problem.
• Finally, a questionnaire was given to all participants asking how they solved
the task, containing a range of questions pertaining to transformational or
derivational strategies.
These different linguistic and non-linguistic data gave us a solid basis for
insights into the cognitive processes and strategies used in analogical problem
solving. As expected, participants in the low-guidance condition used the
derivational strategy, whereas participants in the high-guidance condition
used the transformational strategy. Transformation was quicker than deriva-
tion, and questionnaires as well as the diverse types of written language data
confirmed participants’ strategy choice, spelling out in detail how this was
achieved. In addition, the verbal data revealed that some people may have
recognised the analogy but then applied the solution path in a wrong way when
using the derivational strategy. As a result, solutions based on transformation
tended to be better.
We can conclude from this example study that looking at other data beyond
language in CODA studies does not have to be difficult, and it can be very
revealing. Most CODA studies to date have used more than one type of data.
Correctness and time needed plus further behavioural aspects are typical data to
collect in problem-solving studies. But triangulation can also mean eliciting
more than one type of linguistic data, such as the planning protocols, evalua-
tions, reports, and instructions in our study.

9.1.2 Subtle Cues

Readily observable data such as those described in the previous section are easy
to collect; accuracy of the solution, questionnaires, and time can usually be
assessed without specialist equipment. However, such measures only offer part

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 230 Beyond CODA

of the picture, and (like the analysis of language) are most suited to address
higher-level cognitive processes. Cognitive scientists have developed a wide
range of further, more subtle methods that are intended to go deeper, to provide
insights about cognitive activities that remain below the threshold of partici-
pants’ consciousness.
There is a lot to be gained by looking at the amount of time needed for a task,
and this can often be measured with a simple everyday device. However, in
some kinds of tasks this is not sufficient. In psycholinguistics, for instance, it is
common to measure reaction times on the level of milliseconds. As outlined in
Chapter 2, psycholinguists are interested in the cognitive processing of lan-
guage, and this happens very quickly. Participants in psycholinguistic studies
can be asked to read something and react to it in a particular way, or to look at
a cue and produce language in relation to it. The answers will be provided
within a second or two, and precise measurements are needed to identify any
significant differences between conditions.
In traditional psycholinguistic experimentation, participants typically have
to choose between two possible answers. This amount of experimental control
means that reaction times can be directly compared. In CODA, in contrast,
speakers are encouraged to make their own linguistic choices when producing
language. Both approaches can be combined: subtle differences in reaction
times can amount to meaningful patterns even in scenarios involving uncon-
strained language use. However, this kind of triangulation requires a careful
design. It would have to be a very limited scenario that naturally constrains the
range of things that people will spontaneously say (or write).
For example, imagine an experimental pragmaticist who is interested in
examining the cognitive effort it takes to express a wish for fresh air in
a room. This kind of scenario is exemplary for raising concerns about polite-
ness and levels of formality, tapping into the communication side of language
use (see Chapter 7). Participants could read an introductory scenario that
describes the need for fresh air in a specific communicative context. While
one may ask for a window to be opened in many different ways, it is likely that
specific phrases would be preferred over others, limiting the range of things
people will say. Also, specific aspects of the situation would be represented in
the participants’ linguistic choices, highlighting how they conceive of the
situation. Measuring precise reaction times (i.e. the time needed to formulate
a response after reading the scenario description) would then add to the
evidence, as some linguistic choices may readily come to mind, whereas others
require more time to work out.
In scenarios where participants are asked to react to visual input, it can also
make sense to analyse when and how they are looking at the visuals. Eye-
tracking methods are useful for the analysis of attention focus, and as such can
be meaningfully combined with CODA (see Chapter 4). While speakers’

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 231

linguistic choices reveal their focus of attention as expressed in language, eye-

tracking data are much more fine-grained and subtle. They can therefore reveal
implicit patterns of attention of which participants are rarely aware (Grant &
Spivey, 2003).
Eye-tracking technology has developed rapidly in recent years and is now
available at a much lower cost than previously. Also, the equipment has become
less obtrusive, now causing only a little disruption to participants during
experiments. Given this improved situation, research increasingly incorporates
eye tracking alongside other methods. Elling, Lentz, and De Jong (2012)
specifically revisited the combination of think-aloud protocols with eye track-
ing. They found that this kind of triangulation can be particularly useful as it
means collecting two highly compatible types of evidence that can easily be
matched.
In particular, Elling et al. (2012) looked at types of verbalised content that
correlated with specific eye movements, as well as times when people did not
speak but still looked at the visual input. While some of the categories repre-
sented in people’s verbalisations could also be inferred from the eye-tracking
data, further insights were gained by analysing verbalised observations and
explanations. It seems only natural that the content expressed in language
provides more meaningful elaboration than the information that can be gath-
ered from what people look at. At the same time, the eye-tracking data proved
particularly useful for filling the gaps when people did not speak. It turned out
that silences were times when people did either one of three things: reading,
scanning, or fixating. This means that they were busy taking in information
from the visual input, rather than actively thinking and representing their
thoughts in language.
While measuring reaction times and eye tracking involve specialist equip-
ment and knowledge for analysis, they are still relatively accessible and wide-
spread. Such methods are suitable for analysing subtle cues that indirectly point
to cognitive processes, but they do not address the actual activities in the brain.
For this, more sophisticated tools are required. The most well-known techni-
ques are electroencephalograms (EEG), which measure electrical activity in the
brain, and functional magnetic resonance imaging (fMRI), which records
changes associated with blood flow. Both of these techniques provide insights
on the activation of particular parts of the brain. Based on increasing knowl-
edge about how brain activities are distributed, this kind of evidence can be
used to gain insights about particular brain functions related to given tasks.
In a recent study, fMRI was used to address the extent to which thinking
aloud corresponds to actual thought (Durning et al., 2013). Because jaw move-
ments would disturb the brain imaging data, the authors had to be creative in
their experimental design. Participants were trained to think aloud and then had
to answer a range of multiple-choice questions in the area of medicine. Then,

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 232 Beyond CODA

a phase began which the authors called ‘thinking aloud’ – but this was not
literally true. They asked participants to formulate in their minds how they
arrived at the solution, as if speaking their thoughts out loud but without
actually doing so. Then, after coming out of the fMRI scanner, participants
produced formal verbal reports. Thus, in effect, the authors elicited retrospec-
tive reports rather than think-aloud protocols, and they did this in two steps, the
first of which yielded fMRI data related to a silent thought process.
The results of this intriguing study show various interesting patterns. There
were fMRI differences between the question-answering phase and the silent
think-aloud phase, but these were only gradual, rather than revealing activities
in different parts of the brain. The authors account for each of these differences
in detail, showing that they can be traced back to the different activities
involved in question-answering and thinking aloud. Altogether, they conclude
that thinking aloud can be regarded as true evidence of thought processes,
which is an encouraging result given that collecting verbal data is much
cheaper and easier to achieve than brain imaging.
In another intriguing study, Lutz et al. (2002) addressed the variability of
EEG data by gathering additional evidence through verbal reports. Variability
can be a major problem for EEG analysis, because the same electrical brain
activity is hard to reproduce even with identical tasks. This is attributed to the
fact that participants’ thoughts will vary each time they do the task, which
cannot be avoided. Lutz et al. (2002) reasoned that if people remember their
thoughts each time they do a task, some generalised effects may be identified
that help deal with the variability in EEG data. Their results showed, among
other things, that discrepancies in the EEG patterns could be related to the
extent to which participants were ready to perceive the image they were shown.
This was mostly the case; being mentally ready is typically expected as part of
an experimental study. However, sometimes people can be distracted, and this
has direct effects on the EEG data. Participants’ reports on this and other effects
were systematically related to EEG variability. Again, this illustrates the value
of triangulation and also supports the validity of using verbal reports as
evidence for cognitive processes.
In sum, there are a range of methods that provide evidence concerning
cognitive processes that are too low level to be reflected in language. Such
types of data require more equipment and expertise to collect, but once avail-
able they can be extremely valuable in combination with CODA.

9.1.3 Cognitive Modelling

As we have seen in earlier chapters, one frequent motivation for doing CODA
is to identify how people solve problems or make complex decisions. Results
can be simply reported as such, as they provide useful insights to the field in

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 233

general. However, studies of complex thought are often part of a wider endea-
vour that aims to capture cognitive processes on a fine-grained basis, adding up
to entire theories about how the human mind works. This is called cognitive
modelling, and it is done in a range of different ways, including diagrammatic
and informal as well as formal ways of representing cognitive processes. Most
prominently, in artificial intelligence, cognitive modelling means implement-
ing procedures into simulations of the human mind. The resulting computa-
tional systems incorporate established insights about human cognitive
processes combined with a range of mechanisms that can only be hypothesised
at the current state of research.
As outlined in Chapter 2, language can represent many cognitive pro-
cesses, but certainly not all of them. Much of what goes on in our minds
happens at a very low level and does not reach consciousness, and there-
fore doesn’t get to be verbalised. Cognitive modelling typically aims to
capture just those low-level cognitive processes, which means that gaps in
the existing evidence need to be filled based on theory. As such, cognitive
models cover the full range of cognitive processes from conscious human
decisions and strategies through to neural activities. The result of
a computationally implemented cognitive model can then be compared
with observable insights from various sources, including language data,
feeding back into improved models. Thus, new insights gained on the basis
of linguistic analysis can be very useful for filling in gaps or correcting
erroneous assumptions in existing cognitive models.
Verbal reports as data provide the general structure of a thought process to be
modelled. If speakers differ in their verbalised strategies, joint patterns can be
extracted as far as possible with the aim of developing a consistent cognitive
model. Alternatively, several different solution paths can be modelled and their
results compared, such as thought processes in novices as opposed to experts.
Either way, the more detailed cognitive processes are typically still under-
determined and could be implemented in various different ways. At the same
time, the overall cognitive model can serve to explain differences in the
verbalisations, as people may have different strategies for verbalising their
thoughts. In that sense, triangulation works both ways, with each method
complementing and informing the other.
Aspects that will not be verbalised include motor actions, visual processing,
knowledge retrieval, and the like. We can report or talk about what we do, but
not exactly how we do it. We can say I am lifting my arm and I see a car and
I remember doing this before – but the neural activities that result in the arm
being lifted aren’t accessible to us, and we can’t explain how we recognise the
car as a car, or what our eyes are doing, or how we activated the memory that we
refer to. Here cognitive modelling can help make predictions about the non-
verbalised steps and cognitive processes, since there may be more than one

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 234 Beyond CODA

possibility, but the number of computational models that generate similar

results as in humans will be far more limited.
Cognitive models can represent the different ways in which problems can be
solved in terms of cognitive shortcuts or solution paths. For instance, imagine
a child trying to solve an addition like 3 + 4. The child won’t have a ready
solution in their head, and so will have to put various types of knowledge
together to reach the answer. In contrast, adults can retrieve the solution
directly by simply accessing knowledge built up from previous experience.
Along these lines, conscious cognitive processes become automatised implicit
processes over time.
One of the most established and long-standing cognitive modelling archi-
tectures that is suitable for capturing these and many other phenomena is called
ACT-R (Adaptive Control of Thought – Rational; Anderson & Lebiere, 1998).
ACT-R was initiated by John Anderson at Carnegie Mellon University and has
been continuously developed since the 1970s. The architecture covers mechan-
isms and procedures across all cognitive levels and modules, ranging from low-
level processes far beyond a person’s awareness to high-level decisions that can
be easily reported in language. More recent approaches may increasingly use
other techniques such as Hidden Markov or Bayesian models, but still often
relate to insights gained through the more traditional ACT-R architecture (e.g.
Zhang, Walsh, & Anderson, 2018).
Some ACT-R models are based directly on verbal reports. For example,
Gugerty and Rodes (2007) presented an ACT-R model of the strategies and
cognitive processes involved in cardinal direction judgements. They gave
participants a quite demanding task where they looked at a scene from above,
as if from an airplane, and had to match the scene with a north-aligned map.
Then, they were asked to state the compass direction between two places that
they saw from above, such as a parking lot located to the west of a building.
Verbal protocols revealed that participants used diverse kinds of strategies to
solve this problem. Gugerty and Rodes (2007) combined this evidence with
further insights in the literature about human direction concepts and other
processes that were not directly reported by participants. On this basis, they
developed a cognitive model in ACT-R that plausibly demonstrates how
humans get to a solution for this kind of problem.
Other approaches focus on the sequential steps to reach a target directly,
without dealing with the lower-level cognitive processes. Such so-called pro-
cess models can be expressed in semi-formal representations such as the one in
Figure 9.1, which may or may not be implemented computationally. Verbal
protocols are very informative in this regard, and content analysis is typically
sufficient to account for the sequential steps very well. However, think-aloud
protocols often contain a range of hypotheses that turn out to be false, and other
trains of thought that are not pursued any further. Process models may or may

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 235

TASK ENVIRONMENT

THE RHETORICAL
PROBLEM TEXT
Topic PRODUCED
Audience SO FAR
Exigency

WRITING PROCESSES
THE WRITER’S LONG- PLANNING TRANSLATING REVIEWING
TERM MEMORY
GENERATING
Knowledge of Topic, ORGANIZING EVALUATING
Audience,
and Writing GOAL REVISING
Plans SETTING

MONITOR

Figure 9.1 A process model of writing (Flower & Hayes, 1981).

not include such false leads, depending on how authentic they aim to be.
Linguistic analysis can provide very clear accounts of thought sequences,
particularly if the temporal order is considered rather than treating the data as
an undifferentiated corpus (see Chapter 8). For instance, false starts and
hesitations at certain decisive moments may be indicative of specific cognitive
steps.
Cognitive modelling researchers are typically experts on the modelling
side of things, rather than on the analysis of verbal protocols. As a result,
models may be built on somewhat limited evidence, such as insights
gained from a single person rather than on a controlled language elicitation
design that would be more typical for CODA studies. An in-depth under-
standing of one person’s cognitive processes as a case study can be highly
informative, but it is unclear to what extent such insights can be general-
ised. Also, most existing approaches are informed by the directly obser-
vable content of verbal protocols only, extracting whatever seems relevant
for the modelling purposes at hand.
Combination with CODA can enhance this approach, ranging from refining
the experimental design via more systematic content analysis to the identifica-
tion of in-depth linguistic aspects. For instance, linguistic analysis can associ-
ate cognitive steps with linguistic structure, and confirm content-based
evidence by linguistic theory. Such additional evidence provides a more solid
basis for analysis, allowing for enhanced confidence in the annotation proce-
dure as well as well-informed data interpretation.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 236 Beyond CODA

As with any CODA studies, it will be important to narrow the scope of

linguistic feature analysis down to those aspects that are particularly decisive
for the purposes at hand, in this case cognitive modelling. The analyst might
focus on particularly interesting steps in a cognitive process and identify the
variability of linguistic markers for specific aspects rather than aiming at
representing the entire process, which would not realistically be feasible in
most cases. Possible analysis foci include investigating the level of granularity
with which the speakers conceive of a problem-solving step, how certain they
are, or which perspective on a situation they adopt (see previous chapters).
Also, it can be worthwhile to take a closer look at ‘non-sequiturs’, that
is, conclusions that aren’t based on previously verbalised thoughts, or other
kinds of jumps in the verbal report’s logic. People verbalising their
thoughts are asked to simply express what they are thinking, so they
don’t necessarily make sure that their line of argument is solid and
obvious. However, the way they formulate any ‘jumps’ in the argument
might reveal how they got from one thought to the other. There may be
instances of presupposition or specific discourse markers that shed some
light on the underlying train of thought that didn’t get to be verbalised
completely (see Chapter 3). Also, there may be traces in the language that
indicate the sources of new insights. The analyst may identify the speak-
er’s information status and attention focus at a decisive point in time,
leading to a possible explanation of non-sequential conceptual aspects.
Computational procedures have become increasingly sophisticated and
accessible to non-computer experts. As a result, the implementation of insights
on complex cognitive processes in computational modelling approaches offers
a growing resource for a wide range of applications and purposes. CODA-
based insights can contribute decisively in this regard, as they build a bridge
between directly verbalised content and more intricate conceptual aspects that
are reflected in language only implicitly.

9.1.4 Corpus Linguistics

In terms of a final example for triangulation, we will have a brief look at corpus
linguistics. This well-established approach has many applications, some of
which are related to cognitive aspects. As noted in Chapter 2, corpus linguistic
studies tend to be on the other side of the scale with respect to experimental
control, compared to typical CODA studies. In corpus-based studies, language
data are collected across scenarios and speaker categories. The resulting
corpora are analysed with respect to general patterns in language use, for
instance in a language (like English) as a whole or within a specific dialect or
sociolect (like Scouse or Cockney). However, more specific applications are
also possible, as corpus linguistic methods can be rather varied.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 237

Corpus linguistic research draws on a range of sophisticated software tools,

such as WordSmith or the UaM corpus tool. Such tools can identify word
frequencies, collocations, grammatical distributions, and the like, and they can
be used to compare different corpora, or different parts of the same corpus, to
each other. The Internet offers ample information on these and other tools, as
well as rich resources of linguistic corpora that are readily available for analysis
(the LINGUIST List website is an excellent starting point: [Link]
.org). One well-known example is the British National Corpus (BNC), which
contains 100 million words from a rich diversity of sources and has been
regarded as representative for British spoken and written language for a long
time. Unfortunately, as the corpus is getting older, it is no longer unanimously
recognised as up to date. For American English, a rich and regularly updated
freely available resource is the COCA corpus, with more than 560 million
words that fall into various categories.
Any corpus linguistic study will need to investigate carefully what kind of
resource is most suitable for current research purposes. Corpora vary not only
with respect to their age and size, but each of them comes with further features.
Some of the available corpora were collected for very specific purposes, or
even within a controlled scenario (see Danino, 2018). For instance, our
‘Dolldialogue’ study on furnishing a doll’s house (Tenbrink et al., 2017)
resulted in a small corpus of about 85,000 words, which is now available online
([Link] along with photographs of the furnished houses and
anonymised participant information. Naturally it is always possible to collect
or compile a new corpus for the purposes of a study that requires a specific type
of linguistic data.
This is where CODA and corpus linguistics meet. Like traditional corpus
linguistic approaches, CODA examines the features of language data in lin-
guistic data sets. It does so with respect to the mind, based on controlled
scenarios, and it uses a range of analysis methods beyond those available
through corpus software. Despite these differences to traditional corpus
approaches, some aspects of CODA can definitely benefit from corpus tools.
For instance, if the occurrence of specific words or phrases has been identified
as relevant in the collected data, their distribution and collocations can be
investigated meaningfully using corpus tools.
Also, it can be very enlightening to compare their occurrence in a small
CODA corpus with a larger reference corpus. If people use a specific
phrase frequently in the context in which the data for the CODA study
were collected, this may either mean that the phrase is decisive for this
particular scenario, or it may simply be a frequently used phrase in the
language the speakers use in general. A systematic comparison with
a reference corpus (ideally with millions of words) provides evidence in
either direction.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 238 Beyond CODA

Using corpus tools for CODA analysis means treating the data as a corpus –
and this typically means using the resource like ‘a bag of words’, without
attending to discourse structure or temporal development. In most discourse
analytic approaches, in contrast, it is particularly important to identify how the
discourse unfolds over time. CODA can benefit from both perspectives,
depending on current purposes. Treating a think-aloud protocol as a corpus
may obscure the main features of gradually developing trains of thought in
a certain sequence. However, in other cases the order in which thoughts or
concepts are mentioned, or the specific moment at which they are mentioned,
may not be as decisive.
In some cases, collecting data in a controlled scenario is not feasible at all.
Danino (2014) was interested in how people talk about an event they do not
fully understand, as was the case in the 9/11 attack on the World Trade Center in
New York City. She compiled a corpus from CNN’s live broadcast on the day in
order to analyse diverse speakers’ productions in the face of shocking but
barely comprehensible perceptions. Looking at the discourse produced during
this unique, deeply troubling event enabled her to trace how language reflected
the speakers’ attempts to put into words what they were seeing. This work
exemplifies the value of going beyond everyday situations, addressing dis-
course used under extraordinary circumstances. In this case, no controlled
study could have achieved the same level of authenticity.
Also focusing on language outside most people’s everyday life, though less
troubling, we (Egorova et al., 2015) set out to analyse the language used in
mountaineering. Much previous research had been invested on route descrip-
tions in urban contexts, yielding a wide range of valuable insights about the
ways in which we conceptualise space. Mountains provide a very different
setting, in which only a few of the environmental features that we use in urban
contexts are available. If you explain to your friend how to reach the summit of
Mount Everest, it seems unlikely that you use any of the following stereotypical
route description phrases:
Example 9.1 Turn right at the second intersection.
Example 9.2 Follow the road for 300 metres until you reach the supermarket,
then turn left.
Example 9.3 Carry on for a few minutes, then your destination is the huge
building on the right.
What might be the equivalent for intersections, roads, supermarkets, and other
typical route direction elements on mountains? Would it make sense to refer to
a temporal estimation like ‘a few minutes’, and how easy would it be to rely on
chunks such as ‘the second intersection’ or ‘300 metres’? Intuitively, the
language of mountaineering descriptions will be very different to this, and it

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.1 Triangulation: Combinations with Other Methods 239

will therefore reveal very different concepts of space. But how can we collect
route descriptions in mountains? Even the concept of a route description, as
such, does not seem to apply very well, as people aren’t likely to stop strangers
on a mountain to ask for the way to the summit. Also, not everybody has
climbed mountains. Whereas route description research in street networks can
be conducted easily with various kinds of settings and speakers, participant
recruitment on mountains comes with considerable challenges.
To work around these obstacles, Egorova (2018; Egorova et al., 2015)
decided to take a different approach, namely to use existing language data
to investigate the features of mountaineering discourse. She compiled
a small corpus of nineteen texts of about 400 words each from an online
resource that provided route descriptions in alpine contexts. In contrast to
more typical CODA approaches, this method did not allow for controlling
parameters such as which mountain was described, what exactly the
purpose of the description was, and so on. Egorova couldn’t formulate
a well-designed instruction to the participants, but instead had to use what
was available. However, the website allowed her to restrict the selection in
some ways, such as the estimated difficulty of the route.
Since the corpus was very small and restricted to the specific context of
mountaineering, we (Egorova et al, 2015) were able to adopt CODA principles
in many other respects. Crucially, we achieved an in-depth analysis of linguistic
features that are prominent in this kind of setting, which we could then compare
with what is known from urban route instruction contexts. For instance, we
found that the locomotion itself is far more prominent in alpine contexts than in
a city. This is reflected in a broader variety of verbs such as climb and ascend in
addition to the more common go or walk, and in frequent references to safety
measures such as climbing shoes, helmet, boots, ice axe, and the like.
Landmarks are prominent in mountains just as well as in cities, but they can
look very different. Instead of buildings, one might refer to where the snow
begins – in a city, such a reference would hardly be regarded as permanent
enough to make it into a route description.
Taking this in-depth linguistic analysis as a starting point, Egorova
(2018) went on to explore automatic methods towards a corpus linguistic
approach. She aimed at developing rules for identifying and extracting
spatial concepts from larger text resources in alpine contexts. Her focus
was on concepts of ‘fictive motion’ – descriptions such as the ridge
descends for two miles. The motion verb descends normally applies to
people who are moving in a downward direction, but here it is used to
describe a ridge’s slope; in this sense, the motion is fictive. Egorova (2018)
specified precisely how cases such as these can be extracted automatically,
using a natural language processing engine, from a large text corpus that is
tagged for parts of speech (verbs, nouns, etc.). In this way, she successfully

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 240 Beyond CODA

transferred insights from CODA-based analysis into the efficient handling

of much larger quantities of language data.
Altogether, corpus linguistics overlaps with CODA in a range of ways, and
expertise in corpus analysis methods can be very useful in connection with
CODA studies. Corpus tools are most useful for quantitative aspects in the data,
based on automatic analysis and extraction processes. Since quantitative
aspects are important for CODA (see Chapter 8), the approaches are highly
compatible.

9.2 Practical Purposes within and outside Academia

The way we speak reveals crucial aspects about the way we think, and this
insight motivates a broad range of CODA studies serving many different
purposes. Verbal protocols and linguistic descriptions are pervasive in many
areas of research, basic as well as applied, reaching as far as the marketing-
oriented research procedures employed in commerce. Very often, the aim in
such studies is to understand human intuitions better in order to support their
goals in everyday life. A close, well-informed look at the language used in such
data, and the significance of its features, opens up many practical opportunities
with potential impact on society.
In this section, we will have a look at some of these avenues to exemplify and
illustrate the scope of collaborative, purpose-oriented research. Section 9.2.1
starts by looking at artificial intelligence and interactions between humans and
automatic systems. Section 9.2.2 outlines CODA in relation to architectural
design, and Section 9.2.3 discusses visualisation tools. Section 9.2.4 concludes
the book by opening up further visions beyond research already done or
underway.

9.2.1 CODA and AI: Understanding Humans in Order to Develop Intuitive

Systems
Artificial intelligence (AI) is an exciting field of research with enormous impact
on the world around us and the society we live in. The idea of manufacturing
artificial agents that resemble humans is as controversial as it is fascinating.
Progress over recent decades has opened up many opportunities for developing
smart and adaptive machines, for designing dialogue systems that can interact
with humans, for constructing artificial brains, and much more. Applications of
AI have long since reached our everyday life, despite a growing worry that we
might be creating something that we could end up being unable to control. The
main reason for this worry is that one of the main current AI methods, machine
learning, enables automatic systems to develop beyond the programmers’
direct steering. However, even with this method, systems can only use the

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.2 Practical Purposes within and outside Academia 241

input they receive, and they won’t be able to creatively work towards sinister
ulterior motives. It therefore remains, as ever, the responsibility of the pro-
grammer to equip (or ‘feed’, if you will) the system with the kinds of informa-
tion that will be beneficial, rather than harmful, for the purposes at hand.
Despite best efforts and significant breakthroughs in AI research, automatic
systems remain very unlikely to ever be able to truly understand humans, or to
simulate the ways in which humans draw on intuitions and common sense. This
kind of understanding can only be gained from humans themselves. The better
we understand how humans think and talk, the more specific insights in this
regard can be implemented in a system. With intelligent programming that is
informed by relevant insights on human cognition, artificial systems become
increasingly intuitive and can play to human strengths and intelligence in the
most supportive way.
The idea of supporting the users’ strengths in an intuitive way is especially
central for assistance systems that are designed to help humans achieve their
goals, rather than doing entire tasks for them. Imagine you are planning to go on
holiday with your friends, and you wish to develop a route plan in advance.
What kind of system would help you with this? You will certainly require
a wide range of information, which a system could provide. Then, you might
appreciate support with the complexity of the task, given that there are a range
of constraints to consider: not to travel too far on any single day, to avoid
a succession of boring days, to have places to stay overnight, and much else.
Nevertheless, you probably wouldn’t want an automatic system to plan the
entire route for you – you want to stay in control, to make the final decisions
about every part of the route.
As outlined in Chapter 7, we (Tenbrink & Seifert, 2011) examined how
humans plan complex holiday routes in advance. Our CODA study
revealed a range of cognitive strategies – ways of thinking about space
that humans intuitively use during planning. Some of these insights were
used for implementation in a travel planning assistance system (Seifert,
2008). For instance, one of the strategies that humans use to reduce task
complexity is to think of the travel environment in terms of regions.
Dividing the space up into smaller areas, in a non-rigid way, helps with
focusing on subtasks rather than having to consider everything and every-
where at once. Thus, thinking of regions supports solving the spatial
planning task in a hierarchical way. While the overall trajectory connects
the regions at a high level of granularity, detailed planning is carried out
one region at a time. Seifert (2008) designed a travel assistance system that
supports this cognitive strategy while enabling the user to make all relevant
decisions in an informed way.
Scenarios that involve actions in the real world impose further chal-
lenges. In robotics, traditional approaches have typically been motivated

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 242 Beyond CODA

from an engineering rather than a cognitive perspective. Industry robots can

relatively easily be programmed to carry out very specific, predefined tasks;
they will do the exact movements and actions determined by their specifi-
cation. This may involve interaction with tools and objects, and sometimes
with humans. The main issue here is that without humanlike concepts,
robots will remain very inflexible – everything needs to be programmed
beforehand, there is little room for diversity. Along with this, there is little
room for verbal human instructions that change depending on the features
of a specific scenario.
Addressing these challenges requires cognition-related research in two
major, interrelated respects. On the one hand, we need to understand how
humans think in order to equip robots with similar flexibility. On the other
hand, we need to understand how humans talk in order to enable robots to talk
naturally with humans – to use language in a flexible and context-dependent
way. Although there are plenty of automatic dialogue systems on the market,
and in use even within our households, their flexibility is still fairly limited. The
main reason for this lies in the very link between thought and language. As long
as automatic systems cannot think like a human, they will have difficulty
talking to a human, especially about constantly changing aspects in the
immediate environment. These need to be perceived and referred to adequately
in order to establish a meaningful dialogue that is appropriate for the situation
at hand.
Research on these issues has made significant progress over the past
decades, but there are still significant gaps. Here, CODA can be applied to
gain decisive insights about human thought and language use. In our
research, a robot was enabled to understand instructions like Go to the
object on the right (Moratz & Tenbrink, 2006); this involved implement-
ing a cognitive model that allowed for an adequate interpretation of
human spatial concepts, as well as a dialogue system that dealt with the
language itself. We also applied CODA to identify specific features of
human–computer (Tenbrink et al., 2010) and human–robot interaction
(Winterboer, Tenbrink, & Moratz, 2013), and to support the development
of an intuitive dialogue system for indoor wayfinding (Cuayáhuitl et al.,
2010).
In all of these areas, it is important to identify what people say
intuitively with respect to different kinds of scenarios, and also how
these intuitions change when their interaction partner is an automatic
system. We don’t talk to a system in the same way as to a human, because
we don’t expect the system to be able to understand our thoughts in the
same way. Although this general assumption is of course correct, our
intuitions as to how to talk to the system may not always be accurate.
A smart dialogue system therefore needs to be designed in such a way as

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.2 Practical Purposes within and outside Academia 243

to account for human expectations about dialogue systems as well as their

intuitive ways of speaking and thinking.

9.2.2 CODA and Architectural Design: Towards Intuitive

Building Design
As seen in various examples throughout this book, CODA has been used in many
studies that involved various tasks and research questions related to space. One of
the main driving questions of spatial cognition research concerns what can be done
to improve wayfinding in people’s everyday life, both in indoor and outdoor
settings. For outdoor navigation areas such as street networks, there are now
plenty of navigation aids available, both for planning a route in advance, and for
navigating en route based on ubiquitous GPS signals. In contrast, navigation
within complex buildings remains a challenge in many ways. The GPS signal
isn’t reliable indoors, and far fewer maps or automatic systems are normally
available to guide people to their destination within buildings.
Complex public buildings are often designed to impress and to fascinate, but
not necessarily to navigate easily. Architectural design is a multilayered pro-
cess that involves various challenges in terms of cognition and communication
(Tenbrink et al., 2014; see Chapter 4). Architects have to consider the build-
ing’s functionality and aesthetic aspects against constraints imposed by the
client, by financial limitations, and by the available space. Ease of navigation is
therefore just one aspect among many, and it is certainly not always a priority.
However, it does not have to be difficult to improve architectural design so that
buildings are more intuitive and facilitate wayfinding. Sometimes simple con-
siderations and adjustments are sufficient to make a considerable difference for
the users of a building. To achieve this, it can be useful to think, for a moment,
like a person trying to find their way inside the building. What would be a likely
purpose of their visit, what will they need, what intuitions or expectations will
they bring in, and what would confuse them as they look for their destination?
Research in this area has led to several decisive insights in this regard. For
instance, in terms of intuitions and expectations, people may be open-minded
when they enter a new building, but they aren’t entirely naïve; they have been in
public buildings before and can draw on this previous experience in a new
situation. If they are looking for a place within the building that is likely to be
frequently visited by many people, it is unlikely that they would choose narrow
or semi-dark hallways if they are unsure of the route. Typically, hallways with
long lines of sight are preferred choices. Likewise, attractive objects serve as
navigation aids at intersections, and they can influence decisions as to which
direction to choose (Frankenstein et al., 2010).
CODA informs this line of research in various ways, for instance by identi-
fying features that consistently appear in verbalisations, or by highlighting the

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 244 Beyond CODA

diversity of navigation strategies reflected in language. The study by Tenbrink

et al. (2011a), reported in Chapter 5, is a case in point. When we asked
participants who were new to a building how they might find their way to the
nearest exit, they reported a range of highly suitable navigation strategies.
CODA enabled us to establish the relationship between these strategies and
more standard turn-by-turn route directions. Also, the verbalisations revealed
the ways in which users related to the building’s features; for instance, they
used prominent places such as the main staircase and certain staircase towers as
decisive orientation aids. These findings are in line with previous studies within
complex buildings in which people were asked to verbalise their navigation
strategies (Hölscher et al., 2006).
Research in this area is ongoing, and there are various questions still to be
asked. For instance, how do wayfinding strategies change with age; how would
a home for the elderly need to be designed to be helpful? O’Malley et al. (2018)
found that elderly people who suffer from memory loss have significant
difficulties finding their way around. Moving into a communal retirement
development is meant to be a way to support them in their everyday needs;
however, getting a clear idea about where places are in this new environment
can be a major challenge. If a building is not designed to be helpful, the result
can be a persistent feeling of disorientation for new tenants. This can easily be
the case if all hallways look the same, or if any pictures or other objects
intended to give the building a fresh look are nondescript and non-distinctive.
All people, not only those suffering from memory loss, require salient land-
marks in the environment for orientation. Keeping a residence for the elderly
neutral and indistinctive clearly imposes a major navigation challenge on the
very people it is designed for.
Independent of the users’ age or memory capacities, it is clear from previous
research that complex buildings must strike a balance between distinctiveness
and simplicity. The overall concept of the building should be clearly under-
standable, which is helped by long lines of sight. At the same time, there must
be sufficiently salient features to enable the users to orient themselves, to know
where they are. These features must be distinct and ideally unique, rather than
being repeated in several locations. Getting this balance right may be tricky, but
it is certainly achievable. Research involving a combination of navigation tasks
and verbalisations, such as interviews and think-aloud data that can be system-
atically analysed using CODA, can inform these endeavours decisively.

9.2.3 CODA and Visualisation Tools: Making Insights Intuitively

Accessible
As we saw at the start of this book, language is just one way of representing
thoughts and insights. Another well-established way is through the use of

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.2 Practical Purposes within and outside Academia 245

diagrams and other visualisations (such as Figure 9.1 earlier in this chapter).
Some concepts and relationships can be expressed better in a visuospatial way
than through words (Larkin & Simon, 1987). One main reason for this is that
language is sequential – you read or hear a message in a certain temporal order.
This is not always ideal; complex relationships that involve more than one
dimension may be best expressed through a representation medium that uses
more than one dimension. Standard diagrams or pictures are two-dimensional,
and you can focus your attention on certain aspects in any order you prefer,
whatever makes most sense to you. Some modern displays use three-
dimensional designs, and dynamic visualisations can even represent a fourth
dimension, showing developments over time.
Another advantage is that visualisations often use figurative and analogous
elements, as opposed to the abstract nature of language. The word dog doesn’t
look like a dog, but a visualisation can incorporate a schematic representation
of a typical dog. Especially in spatial contexts, this aspect can become quite
crucial. We have only a limited range of words in our everyday language to
express spatial directions, and usually happily content ourselves with vague or
qualitative concepts like ‘to the right’, even in cases where the angle isn’t
exactly ninety degrees. In contrast, a schematic representation can represent the
angle more accurately, allowing for more information to be conveyed without
adding complexity to the message.
Taken together, these main advantages allow for representing relevant
aspects of the depicted concepts or situations in an efficient and relatively
simple way. Facilitated by ever-improving software tools, visualisations are
fairly ubiquitous in modern times. They not only serve to illustrate facts and
enhance the content of accompanying text, but further guide the observer
towards certain conclusions or decisions based on what the visualisation
suggests. But what exactly happens in this process – what do we take from
a visualisation, and how do we know what other people would understand
from it?
When looking at a picture or diagram, humans do not perceive all features
and elements equally or objectively, as a computer might do. Instead, the
perception of visual information is guided by a combination of relevance and
salience principles. The perceiver’s attention is drawn towards aspects that
visually stand out and are therefore particularly salient (Fine & Minnery, 2009)
just as well as towards elements that relate directly to current task purposes
(Henderson, Malcolm, & Schandl, 2009). As a consequence, some parts of
a visualisation may remain entirely outside the perceiver’s consciousness. It is
very easy, and very human, to miss seemingly irrelevant aspects if they don’t
stand out in any particular way.
Understanding these principles is vital for designing visualisations in such
a way that they work well in supporting humans in the way they are supposed

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 246 Beyond CODA

to. Simply including relevant information in and of itself is not sufficient if

perceivers don’t notice and understand it easily. They might not only miss vital
information but could also easily misinterpret the message intended by the
designer. For instance, they might confuse accidental features of a diagram as
representing actual states or relationships in the real world. If a map represents
a forest in a dark brown colour, the map user may infer that the forest has
predominantly dark brown trees. This may not be a conscious inference –
people may not even notice that they have come to this kind of conclusion.
Nevertheless, they will work on the basis of this false assumption until they
realise their mistake.
In this respect, conventions and expertise play a major role. We under-
stand language because we use it all the time in line with the conventions
in our communities, which involve certain rules of grammar as well as
accepted usage at more pragmatic levels. Likewise, we understand dia-
grams better if they follow certain conventional principles, and regular
users of a particular type of visualisation will be less likely to be mis-
guided than first-time observers. However, visualisations are far less con-
ventionalised or standardised than language is, and therefore the risk for
miscommunication is generally stronger. The intuitive and pictorial nature
of visualisations goes a long way towards alleviating this problem, but
there will always be abstract or schematic elements or features that can be
misinterpreted.
In expert domains, this issue can become quite crucial. Visualisations are
often generated as part of an effort to enable experts to quickly extract relevant
information for a purpose. The ability to do this efficiently strongly depends on
experience (Jee et al., 2009). To a degree, perceivers of displayed information
are biased by their background as well as by how they judge aspects of
relevance in a given context. This will affect the conclusions and decisions
that they will make on the basis of visualised information, leading to either
desired or undesired outcomes.
These effects are complex, and they are based on cognitive processes on the
part of the people perceiving visualisations. Something goes on in the minds of
the people perceiving visualisations, and it is vital to understand how this
works, as visualisations are becoming increasingly widespread. In recent
years, various endeavours have been undertaken to address these issues, for
example through the analysis of diagram users’ eye movements (Çöltekin,
Fabrikant, & Lacayo, 2010), or by testing the usability of different versions
in the field (Sarjakoski & Nivala, 2005).
Another way to learn about users’ understanding of visualisations is through
language. When we perceive a diagram, we will to a certain extent be able to
express our understanding of the diagram in words. Although much can be
learned by interviewing people concerning their thoughts, simply asking

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.2 Practical Purposes within and outside Academia 247

questions may not be sufficient. Often, information gathered this way does not
make its way into scientific reports, even if it serves to inform researchers on an
informal basis.
Here, as with other research on cognitive processes, CODA is a way to
systematise the insights that can be gained on the basis of language (Tenbrink,
2014). CODA can highlight users’ understanding of visualisations in every
respect outlined in this book. Perhaps the most decisive aspect concerns
attention (Section 4.1); what are the kinds of things people attend to as they
try to interpret a visualisation? It is part of the design process for a visualisation
to consider how users’ attention can be drawn towards the most relevant
aspects. A CODA analysis of attention patterns can verify the extent to
which this is successful, and identify undesired shifts of attention towards non-
central aspects of a visualisation. This can be combined with eye-tracking data,
as discussed earlier in this chapter.
Furthermore, users can perceive visualisations from different conceptual
perspectives (Section 4.2), depending on the situation or task at hand.
Imagine you’re an expert in ocean sciences, and you’re looking at
a diagram showing flooded areas across Britain as part of a research
article. How you look at it, and what you take from it, might be very
different depending on your purpose in reading the article. Are you aiming
to increase your knowledge in the field? Are you looking for something to
relate your own research to? Do you wish to draw on this information to
alert the people in Britain to imminent dangers, and to start a campaign for
building better flood defences? Or is the visualisation something you might
use for your own teaching purposes, in which case you’d have to consider
its suitability for students?
Each of these aims would lead to different perspectives on the visualisation,
and further perspectives will be employed by people with yet different aims.
Non-experts will perceive it differently from experts, and other people might
have very different purposes – such as seeking evidence against flood protec-
tion, or against the notion of global warming. A CODA study of verbalised
perceptions in different situations and by different people can highlight this
diversity and identify the effects it has on people’s understanding of the
visualisation.
Similar considerations apply for each of the other levels of analysis sug-
gested in earlier chapters. People may understand visualisations at different
levels of granularity (Section 5.1), and they may be more or less certain about
the meanings and insights conveyed (Section 5.2). They may make different
kinds of inferences (Section 6.1), depending on how the visualisation triggers
their thoughts. Visualisations may – and often do – enable users to transform
things in the real world (Section 6.2), as is the case with illustrations of steps in
a manual. People often communicate about visualisations (Section 7.1), for

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 248 Beyond CODA

instance to discuss the implications of the insights conveyed. Also, under-

standing the message of a visualisation can be a problem, or it can be part of
a problem-solving task; in this case, the problem-solving process can be
analysed on a sequential basis (Section 7.2). Apart from these central layers
of analysis as outlined in this book, further analysis perspectives can be adopted
in the spirit of CODA, all of which lead to a better understanding of how people
understand visualisations.

9.3 Conclusion
How much can we learn from the way we talk, and what can we do with this
knowledge? As I hope I have shown in this final chapter, the scope for
combinations with other methods as well as for applications within and outside
academia is extremely wide. The reason for this is very simple: language is the
primary medium of communication and representation of thought, and it is hard
to imagine a situation in which it plays no role at all. As long as there is
something to put into words in relation to a question at hand, you can get
participants to speak, or you can add an element of verbalisation to a study
you’re already planning, and you’ll be able to gain systematic insights from the
discourse through CODA.
This sounds simple, and in a sense it is. The idea for CODA as well as
the basic procedures are actually very straightforward, as long as a few
basic parameters are considered (see Chapter 8). It is all about taking
language seriously as a representation of thought, and analysing it system-
atically, with a clear awareness of the underlying concepts and states of
mind that lead the speaker to make their linguistic choices in the way they
do. It matters whether people say the car is green or here is a Mercedes
with an odd grass-like colour. There is a reason for the diverse differences
here, and also for the many indications of doubt in an utterance like Oh,
I really cannot say for sure, but, uh, perhaps I could place the, um, teacup
on the, well, on the table?
Differences and markers such as these are obvious once you start looking at
language closely. You don’t need to be an expert on language to understand
their significance. All it takes is an analyst’s mindset. The linguistic features
may seem subtle or irrelevant to start with, and indeed they are easily over-
looked in everyday conversation. We do not normally pay conscious attention
to the exact ways in which we formulate things, neither in speaking nor in
listening or reading. We understand the content and take things further from
there: act on what we’ve heard, respond to it, and so on. However, just because
we don’t normally pay attention to the specifics of how things are said doesn’t
mean that we cannot do it. This is, in a nutshell, the idea for CODA. You don’t
need to be an expert in linguistics to recognise significant linguistic features,

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 9.3 Conclusion 249

but you do need to adopt the mindset and systematicity of an analyst to work
them out properly.
There are a range of areas that still await further exploration. For instance,
consider relations between input and output. How does the production of
language interact with the comprehension of language? How are linguistic
features in the produced language affected by the features of the language
presented to participants? Empirical studies that involve the production of
language, such as descriptions of some observed phenomenon, or verbal pro-
tocols of a problem-solving process, typically also involve instructing the
participants through language. Using language is almost inevitable in academic
research. But clearly this also affects participants, and it is important to be
aware of this. Most of all, as emphasised in Chapter 8, it is essential to keep the
instruction constant, and not vary it randomly across participants.
As shown in Chapter 7, it is clear from previous research that input affects
output in many important ways. We align our utterances to those of our
interaction partner, and we can be primed by certain constructions or word
choices towards using similar ones. Inevitably, the linguistic input that partici-
pants receive in an academic study will influence the language they produce.
This fact can be addressed as a specific target in CODA research. For instance,
different participants could get differently worded instructions with a similar
content, enabling the analyst to focus on the effect of the instruction’s
formulation.
In Tenbrink and Taylor (2015; see Chapter 6) we were interested in how
people follow written origami instructions, and to what extent their thoughts
diverged from those instructions. To identify the divergence, we closely
attended to the instructions’ wording, and compared it to the wording used
by the participants as they verbalised their thoughts. This input–output com-
parison enabled us to identify a range of cognitive strategies that people follow
in relation to written information.
However, there is much more to be learned about the relationship between
language input and subsequent linguistic output. In a current collaboration with
Alice Bell at Sheffield Hallam University, we are addressing reader responses
to a novel. This line of research relates to research in literature studies, with its
focus on the effects of literature on the reader. Typically, in reader response
research, readers’ reactions are addressed through questionnaires containing
Likert scales and open questions. These responses are not normally investi-
gated from a linguistic point of view. However, we found indications that the
novel’s features may have had an effect on the linguistic features in the readers’
responses, especially concerning aspects relating to their thoughts. This opens
up exciting avenues for future research in literature studies and beyond, looking
at the effects of what we read or hear on our thoughts and the ways in which we
express those thoughts through language.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 250 Beyond CODA

Taking this train of thought further, consider users’ understanding of man-

uals that contain visual or textual instructions, or both. Manuals play an
important role in human everyday life, and misunderstanding them can have
severe consequences, depending on circumstances. Our research in Tenbrink
and Maas (2015) revealed that the link between visual and textual elements is
particularly tricky, as it is not always clear how the text relates to the accom-
panying visuals. If visuals are only poorly understood in relation to text, it must
be even more difficult to transfer the visualised information to the object in the
real world that the manual is aiming to explain. Research in this direction could
aim to develop a set of guidelines for improving user manuals, making them
more intuitive and avoiding ambiguities.
As a long-term vision, this strand of research may help to improve the
communication of complex insights in general. Understanding how people
express their thoughts, using CODA, can help to facilitate the two-way com-
munication between humans. Getting thoughts across from one mind to another
is one important function of discourse, independent of the mode of commu-
nication. CODA can be used to address the success of this aspect, to identify the
most suitable ways of formulating thoughts in such a way that the addressee can
easily follow them. Our dialogue work in Tenbrink et al. (2017) is a case in
point, but there is still much more to discover.
Research has brought us a long way towards understanding how language
works as a system, and how we use it in discourse. As technology develops, it
increasingly opens up further ways to incorporate this knowledge – through
new communication media, in automatic dialogue systems and intelligent
agents, and much more. CODA, as an idea or general approach, is simple and
flexible enough to remain available despite major changes in the medium it
addresses. This book was written with the aim of making the method easily
accessible to researchers across disciplines; it is my hope that this will spark
further ideas about how discourse can help us understand human thought.

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 23:47:15, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 Register of Linguistic Features

This section lists key linguistic features that have been mentioned throughout
the book, and which may be of interest to researchers using CODA. While there
is no intent to be exhaustive, the features listed here have all been identified in
the literature as being indicative of specific cognitive aspects in some way, and
so they can serve as starting points for CODA studies. As in this book as
a whole, English is assumed as a starting point but the linguistic features should
usually be represented in other languages as well.
In the following, →arrows tell you that you can access more information
about this concept in this list.
Absence of content or features. Introduction; Sections 5.1.1; 7.1.1. If some
specific content or linguistic feature is not represented in a linguistic data
set, this may have many reasons. Some of these reasons may be uncovered
by analysing patterns of linguistic features in relation to the conditions in
which language was produced. If a feature is frequent in one condition but
absent in another, there must be a reason for this pattern. Otherwise, if an
expected cognitive aspect is completely absent, it may mean that this aspect
is not or cannot be represented in language, even though it may be present
in the mind.
Adjectives. Sections Sections 3.1.1; 4.1.2; 5.1.2; 7.2.4. Like →Nouns and (to
a lesser extent) →Verbs, adjectives can be key indicators of semantic fields
and conceptual domains. Further, they are particularly powerful in expres-
sing perceived object features (as in the blue car), as well as attitudes and
evaluations in different degrees of intensity (such as good or dreadful).
Certainty markers. Sections 1.3.2; 5.2. Language can express levels of
certainty through →Modality, →Pauses and hesitations, and →Verbs
(such as know) that explicitly indicate certainty, through hedges (such
as in a sense and probably), and in many other ways.
Conjunctions. Sections 1.2.4; 2.3.6. Conjunctions (also called discourse mar-
kers) such as because or after explicitly indicate →Discourse relations, or
relations between elements within a clause.
Discourse relations. Sections 1.2.3; 3.2.3; 3.3.5; 5.1.2; 7.2.3. Clauses often
relate to each other, in various ways, such as causally or temporally.

251

Downloaded from [Link] Columbia University Libraries, on 24 Jan 2020 at 15:51:26, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 252 Register of Linguistic Features

Such relations are frequently but not necessarily made explicit by

→Conjunctions. This aspect concerns how the speaker conceives of the
interconnections between the elements described.
Discourse structure. Sections 1.2.3; 2.3.6; 7.1.3. The way a particular piece of
discourse is structured can be revealing when analysed relative to the situa-
tion context. For instance, in think-aloud protocols, discourse structure
reflects the temporal order of thoughts; in dialogue, communication partners
react to each other’s contributions in structured ways.
Grammatical case and function. Sections 1.1.2; 3.3.3; 4.1.2. The form and
function of a →Noun, or noun phrase, within a clause indicate what role the
speaker assigns to them – for instance whether a person is an agent or
a beneficiary in a transaction.
Granularity. Sections 3.2.4; 5.1; 5.2.1. Descriptions of processes, states, or
objects can be more or less elaborate and reflect different levels of detail.
This is related (but not identical) to →Precision and can be traced, among
other aspects, by analysing the features of →Nouns and →Verbs.
Inferences. Sections 2.3.1; 3.2.3; 6.1. Inferences are not directly expressed on
the linguistic surface; however, a close look at what was said against the
information available to the speaker can reveal any inferences that the
speaker has made.
Information structure. Sections 3.2.2; 3.3.4; 4.1.2; 8.5.4. The way a sentence
is structured reflects to some extent which aspects are assumed to be already
known and which are represented as new. Typically, in English, clauses start
with given information and present new information at the end. However,
aspects of information structure such as contrast and focus can also be
expressed by other means, such as tonal stress and intonation contour.
Metaphors and transferred meanings. Sections 2.1.2; 3.1.5. Many linguistic
expressions are not meant in their literal sense, but reflect some kind of
meaning transfer. For instance, good times ahead uses the spatial domain to
describe a temporal phenomenon.
Modality. Sections 1.2.1; 3.3.2; 5.2.2; 6.2.2; 8.6. Modal verbs such as could,
should, would, and might indicate possibilities and alternative worlds that the
speaker conceives of in various ways, related (but not restricted) to
→Certainty.
Mood. Sections 3.3.1; 3.3.2. Mood denotes whether a clause represents
a question, command, or statement, reflecting whether speakers convey
conceptualised facts (as in The window is open) or aim to achieve a certain
discourse goal (as in Could you open the window?).
Nouns. Sections 1.1.2; 1.2.1; 3.1.2; 3.2.4; 3.3.3; 6.1; 6.2.2; 7.2.2; 7.2.4. Every
object or person can be referred to or categorised in more than one way. The
term we use for reference in a given situation indicates the conceptual level of
detail and associated features – for instance, whether we say shoes or high

Downloaded from [Link] Columbia University Libraries, on 24 Jan 2020 at 15:51:26, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 Register of Linguistic Features 253

heels. The lexical choices for nouns also indicate semantic fields and con-
ceptual domains; e.g. shoes belong to the domain of clothing and violin to the
domain of music. Other lexical choices in the co-text (including other forms
such as →Verbs) further contribute to building up a semantic field.
Pauses and hesitations. Sections 1.3.3; 3.2.1; 5.2.2; 8.4. Spoken discourse
typically contains many instances of pauses and hesitation markers such as
uhm, uh, mh. There may be many reasons underlying these delays in speech
fluency, including lack of →Certainty, high cognitive load (i.e. too busy
thinking to talk), or problems finding the right words. Because it can be very
time-consuming to transcribe pauses accurately, it makes sense to identify
beforehand to what extent these aspects can be analysed meaningfully.
Perspective. Sections 1.2.3; 1.3.2; 4.2; 7.1.1; 7.1.4. All utterances reflect the
underlying conceptual perspective in some ways; this is often, but need not
be, the speaker’s own stance. Spatial perspective determines the interpreta-
tion of descriptions like to the right of the chair (see also →Spatial reference
systems).
Precision. Sections 1.3.2; 3.2.4; 5.2.3; 8.3. Descriptions can be more or less
precise or vague, which can reflect degrees of knowledge or →Certainty, or
the speaker’s current focus of attention.
Prepositions. Sections 1.2.4; 3.1.2; 4.1.2; 6.2.3; 8.5.4. Although seemingly
small and irrelevant, prepositions are powerful representations of underlying
concepts. For instance, rather than simply representing a vertical direction
(as in on the table), the preposition on indicates an underlying notion of
location control (as exemplified by on the wall). Prepositions are frequently
used in a transferred sense, as part of a conventionalised →Metaphor; here,
as well, there are subtle but meaningful differences between prepositional
choices (compare on time and in time).
Presuppositions. Sections 1.2.4; 3.2.2; 8.5.4; 9.1.3. Many utterances implicitly
point to underlying assumptions that are presupposed by the speaker. For
instance, reference to the white horse presupposes that there is a white horse
that is somehow accessible to the speaker (and hearer where applicable).
Pronouns. Sections 1.2.1; 4.2.3; 8.5.3. Pronouns are used to represent recurring
elements in a discourse that are known and do not need to be reintroduced by
name. It can be revealing to identify patterns in the use of first- or second-
person pronouns, or to trace the referents of third-person pronouns.
Spatial reference systems. Sections 1.1.2; 3.1.4; 4.2.2. We can refer to space in
several different ways; spatial reference systems are the concepts underlying
our choices – for instance, north of me as opposed to on the table’s right.
Speakers’ assignments of the roles of locatum and relatum as well as the
spatial relationship expressed (typically) by a →Preposition can be highly
revealing of their spatial conceptualisations.

Downloaded from [Link] Columbia University Libraries, on 24 Jan 2020 at 15:51:26, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 254 Register of Linguistic Features

Syntax. Sections 3.1.3; 3.1.4; 4.1.2. The syntax of a clause reflects a range of
abstract meaning structures as well as attention focus through →Information
structure, →Grammatical case, and more.
Verbs. Sections 2.3.6; 3.2.4; 3.3.3; 7.2.1; 7.2.2; 7.2.4; 9.1.4. Similar to lexical
choices for →Nouns, the verbs used to express a particular action or process
highlight the underlying conceptualisation. Some verbs are fairly direct
indicators of processes related to cognition, such as perceptual verbs (see,
feel, hear) or verbs expressing conceptual states or goals (e.g. search, know,
focus on).

Downloaded from [Link] Columbia University Libraries, on 24 Jan 2020 at 15:51:26, subject to the Cambridge Core
terms of use, available at [Link] [Link]
 References

Adesope, O. O., Lavin, T., Thompson, T., & Ungerleider, C. (2010). A systematic
review and meta-analysis of the cognitive correlates of bilingualism. Review of
Educational Research 80(2), 207–45.
Afflerbach, P. & Johnston, P. (1984). On the use of verbal reports in reading research.
Journal of Reading Behavior 16, 307–22.
Aitchison, J. (1983). The Articulate Mammal: An Introduction to Psycholinguistics (2nd
ed.). London: Hutchison.
Allen, G. L. (2000). Principles and practices for communicating route knowledge.
Applied Cognitive Psychology 14, 333–59.
Altmann, G. T. (2001). The language machine: Psycholinguistics in review. British
Journal of Psychology 92, 129–70.
Anderson, J. R. (2009). Cognitive Psychology and Its Implications (7th ed.). New York:
Worth Publishers.
Anderson, J. R. & Lebiere, C. (1998). The Atomic Components of Thought. Mahwah,
NJ: Lawrence Erlbaum Associates.
Anderson, M. L. (2003). Embodied cognition: A field guide. Artificial Intelligence 149
(1), 91–130.
Andonova, E., Tenbrink, T., & Coventry, K. R. (2010). Function and context affect
spatial information packaging at multiple levels. Psychonomic Bulletin and Review
17, 575–80.
Antaki, C., Billig, M. G., Edwards, D., and Potter, J. A. (2003). Discourse analysis
means doing analysis: A critique of six analytic shortcomings. Discourse Analysis
Online 1(1). [Link]
-[Link]
Arnold, J. E. (2008). Reference production: Product-internal and addressee-oriented
processes. Language and Cognitive Processes 23(4), 495–527.
Arts, A. (2004). Overspecification in Instructive Texts. Nijmegen: Wolf Publishers.
Atit, K., Shipley, T. F., & Tikoff, B. (2013). Cognitive Processing 14, 163–73.
Bartl, C. & Dörner, D. (1998). Sprachlos beim Denken – Zum Einfluß von Sprache auf
die Problemlöse- und Gedächtnisleistung bei der Bearbeitung eines nichtsprachlichen
Problems. Sprache und Kognition 17, 224–38.
Bateman, J., Hois, J., Ross, R. R., & Tenbrink, T. (2010). A linguistic ontology of space
for natural language processing. Artificial Intelligence 174, 1027–71.
Bavelas, J., Gerwing, J., & Healing, S. (2017). Doing mutual understanding.
Calibrating with micro-sequences in face-to-face dialogue. Journal of
Pragmatics 121, 91–112.

255

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 256 References

Bégoin-Augereau, S. & Caron-Pargue, J. (2003). Linguistic criteria for demarcation and

hierarchical organization of episodes in a problem solving task. In F. H. van Eemeren,
J. A. Blair, C. A. Willard, & F. Snoeck Henkemans (eds.), Proceedings of the Fifth
Conference of the International Society for the Study of Argumentation, pp. 81–7.
Amsterdam: Sic Sat.
Bégoin-Augereau, S. & Caron-Pargue, J. (2009). Linguistic markers of decision pro-
cesses in a problem solving task. Cognitive Systems Research 10, 102–23.
Biber, D., Conrad, S., & Reppen, R. (1998). Corpus Linguistics: Investigating
Language Structure and Use. Cambridge, UK: Cambridge University Press.
Bierwisch, M. & Schreuder, R. (1992). From concepts to lexical items. Cognition 42,
23–60.
Boren, M. & Ramey, J. (2000). Thinking aloud: Reconciling theory and practice. IEEE
Transactions on Professional Communication 43(3), 261–78.
Brennan, S. E. & Clark, H. H. (1996). Conceptual pacts and lexical choice in conversation.
Journal of Experimental Psychology: Learning, Memory and Cognition 22(6), 1482–93.
Brösamle, M. & Hölscher, C. (2012). Architectural gestures: Conserving traces of
design processes. In C. Hölscher & M. Bhatt (eds.), Proceedings of SCAD Spatial
Cognition for Architectural Design. Report No. 029–08/2012 Report Series of the
Transregional Collaborative Research Center SFB/TR 8 Spatial Cognition,
pp. 215–23. Bremen: Bremen University.
Brown, P. and Levinson, S. C. (1993). ‘Uphill’ and ‘downhill’ in Tzeltal. Journal of
Linguistic Anthropology 3, 46–74.
Brown, R. (1958). How shall a thing be called? Psychological Review 65(1), 14–21.
Brunyé, T. T., Ditman, T., Mahoney, C. R., Augustyn, J. S., & Taylor, H. A. (2009).
When you and I share perspectives: Pronouns modulate perspective taking during
narrative comprehension. Psychological Science 20(1), 27–32.
Bugmann, D., Coventry, K. R., & Newstead, S. E. (2007). Contextual cues and the
retrieval of information from cognitive maps. Memory and Cognition 35(3),
381–92.
Butler, C. S. & Gonzálvez-García, F. (2014). Exploring Functional-Cognitive Space
(Vol. 157). Amsterdam: John Benjamins Publishing Company.
Canas, J., Quesada, J., Antolí, A., & Fajardo, I. (2003). Cognitive flexibility and
adaptability to environmental changes in dynamic complex problem-solving tasks.
Ergonomics 46(5), 482–501.
Carlson, L. A., Hölscher, C., Shipley, T. F., & Dalton, R. C. (2010). Getting Lost in
Buildings. Current Directions in Psychological Science 19(5), 284–9.
Carlson, L. A. & van der Zee, E. (eds.) (2005). Functional Features in Language and
Space: Insights from Perception, Categorization and Development. Oxford: Oxford
University Press.
Caron-Pargue, J. & Caron, J. (1989). Processus psycholinguistiques et analyse des
verbalisations dans une tâche cognitive. Archives de Psychologie 57, 3–32.
Caron-Pargue, J. & Caron, J. (1995). La fonction cognitive des interjections. Faits de
Langues 6, 111–20.
Caron-Pargue, J. & Caron, J. (2000). Les interjections comme marqueurs du fonction-
nement cognitif. Cahiers de praxématique 34, 51–76.
Caron-Pargue, J. & Gillis, S. (eds.) (1996). Verbal Production and Problem Solving.
Antwerp Papers in Linguistics 85, University of Antwerp.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 257

Carroll, M. & von Stutterheim, C. (1993). The representation of spatial configurations in

English and German and the grammatical structure of locative and anaphoric
expressions. Linguistics 31, 1011–41.
Carston, R. (1998). Informativeness, relevance and scalar implicature. In R. Carston &
S. Uchida (eds.), Relevance Theory: Applications and Implications, pp. 179–236.
Amsterdam: John Benjamins.
Carston, R. (2002). Thoughts and Utterances: The Pragmatics of Explicit
Communication. Oxford: Blackwell.
Chi, M. T. H. (2006). Two approaches to the study of experts’ characteristics. In
K. A. Ericsson, N. Charness, P. J. Feltovich, & R. R. Hoffman (eds.), The
Cambridge Handbook of Expertise and Expert Performance, pp. 21–30.
Cambridge, UK: Cambridge University Press.
Chi, M. T. H., Glaser, R., & Rees, E. (1982). Expertise in problem solving. In
R. Sternberg (ed.), Advances in the Psychology of Human Intelligence, pp. 7–75.
Hillsdale, NJ: Erlbaum.
Chomsky, N. (1972). Studies on Semantics in Generative Grammar. The Hague:
Mouton Publishers.
Chomsky, N. (1980). Rules and Representations. New York: Columbia University
Press.
Cialone, C., Tenbrink, T., & Spiers, H. J. (2018). Sculptors, architects, and painters
conceive of depicted spaces differently. Cognitive Science 42, 524–53.
Clark, D. R., Li, D., & Shepherd, D. A. (2018). Country familiarity in the initial
stage of foreign market selection. Journal of International Business Studies
49(4), 442–72.
Clark, H. H. (1996). Using Language. Cambridge, UK: Cambridge University Press.
Clark, H. H. & Krych, M. (2004). Speaking while monitoring addressees for
understanding. Journal of Memory and Language 50, 62–81.
Clark, H. H. & Wilkes-Gibbs, D. (1986). Referring as a collaborative process. Cognition
22, 1–39.
Cleland, A. A. & Pickering, M. J. (2003). The use of lexical and syntactic information in
language production: Evidence from the priming of noun phrase structure. Journal of
Memory and Language 49, 214–30.
Çöltekin, A., Fabrikant, S. I., & Lacayo, M. (2010). Exploring the efficiency of users’
visual analytics strategies based on sequence analysis of eye movement recordings.
International Journal of Geographical Information Science 24(10), 1559–75.
Corley, M. & Scheepers, C. (2002). Syntactic priming in English sentence production:
Categorical and latency evidence from an internet-based study. Psychonomic Bulletin
and Review 9(1), 126–31.
Corteen, R. S. & Wood, B. (1972). Autonomic responses to shock-associated words in
an unattended channel. Journal of Experimental Psychology 94(3), 308.
Couclelis, H. (1996). Verbal directions for way-finding: Space, cognition, and language.
In J. Portugali (ed.), The Construction of Cognitive Maps, pp. 133–53. Dordrecht:
Kluwer Academic Publishers.
Coventry, K. R. & Garrod, S. C. (2004). Saying, Seeing and Acting: The Psychological
Semantics of Spatial Prepositions. London: Psychology Press.
Cuayáhuitl, H., Dethlefs, N., Richter, K.-F., Tenbrink, T., & Bateman, J. (2010).
A dialogue system for indoor wayfinding using text-based natural language.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 258 References

International Journal of Computational Linguistics and Applications 1(1–2),

285–304.
Culpeper, J. & Kytö, M. (2010). Early Modern English Dialogues: Spoken Interaction
as Writing. Cambridge, UK: Cambridge University Press.
Dąbrowska, E. (2012). Different speakers, different grammars: Individual differ-
ences in native language attainment. Linguistic Approaches to Bilingualism 2(3),
219–53.
Danino, C. (2014). Language production and meaning construction mechanisms in the
discourse on an ongoing event: the case study of CNN’s live broadcast on 9/11. PhD
dissertation, University of Poitiers, France.
Danino, C. (2018). Introduction: Les Petits Corpus. Corpus 18. [Link]
.[Link]/corpus/3099
Denis, M. (1997). The description of routes: A cognitive approach to the production of
spatial discourse. Current Psychology of Cognition 16, 409–58.
Denis, M., Pazzaglia, F., Cornoldi, C., & Bertolo, L. (1999). Spatial discourse and
navigation: An analysis of route directions in the city of Venice. Applied Cognitive
Psychology 13(2), 145–74.
Durning, S. J., Artino, A. R. Jr, Beckman, T. J., et al. (2013) Does the think-aloud
protocol reflect thinking? Exploring functional neuroimaging differences with think-
ing (answering multiple choice questions) versus thinking aloud. Medical Teacher
35(9), 720–6.
Edwards, D. & Potter, J. (1992). Discursive Psychology. London: Sage.
Egorova, E. (2018). From text to space: Spatial discourse in alpine route directions and
narratives. PhD dissertation, University of Zurich.
Egorova, E., Tenbrink, T., & Purves, R. S. (2015). Where snow is a landmark: Route
direction elements in alpine contexts. In S. I. Fabrikant, M. Raubal, M. Bertolotto, C.,
et al. (eds.), Spatial Information Theory, pp. 175–95. Berlin: Springer.
Egorova, E., Tenbrink, T., & Purves, R. S. (2018). Fictive motion in the context of
mountaineering. Spatial Cognition and Computation 18(4), 259–84.
Ehrich, V. & Koster, C. (1983). Discourse organization and sentence form: The structure
of room descriptions in Dutch. Discourse Processes 6, 169–95.
Elling, S., Lentz, L., & De Jong, M. (2012). Combining concurrent think-aloud proto-
cols and eye-tracking observations: An analysis of verbalizations and silences. IEEE
Transactions on Professional Communication 55(3), 206–20.
Ericsson, K. A. & Simon, H. A. (1993). Protocol Analysis: Verbal Reports as Data (2nd
ed.). Cambridge, MA: Bradford Books/MIT Press.
Ericsson, K. A. & Simon, H. A. (1998). How to study thinking in everyday life:
Contrasting think-aloud protocols with descriptions and explanations of thinking.
Mind, Culture, and Activity 5(3), 178–86.
Evans, V. (2009a). How Words Mean: Lexical Concepts, Cognitive Models and
Meaning Construction. Oxford: Oxford University Press.
Evans, V. (2009b). Semantic representation in LCCM Theory. In V. Evans and
S. Pourcel (eds.), New Directions in Cognitive Linguistics, pp. 27–55. Amsterdam:
John Benjamins.
Fan, Y. & Heeman, P. A. (2010). Initiative conflicts in task-oriented dialogue. Computer
Speech and Language 24, 175–89.
Fairclough, N. (1995). Critical Discourse Analysis. London: Longman.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 259

Feldman, J. (2010). Embodied language, best-fit analysis, and formal compositionality.

Physics of Life Reviews 7(4), 385–410.
Ferrari, F. (2007). Metaphor at work in the analysis of political discourse:
Investigating a ‘preventive war’ persuasion strategy. Discourse and Society
18(5), 603–25.
Filipi, A. & Wales, R. (2004). Perspective-taking and perspective-shifting as socially
situated and collaborative actions. Journal of Pragmatics 36(10), 1851–84.
Fine, M. S. & Minnery, B. S. (2009). Visual salience affects performance in a working
memory task. Journal of Neuroscience 29(25), 8016–21.
Finke, R. A., Pinker, S., & Farah, M. J. (1989). Reinterpreting visual patterns in mental
imagery. Cognitive Science 13(1), 51–78.
Fischer, M. H. & Zwaan, R. A. (2008). Embodied language: A review of the role of the
motor system in language comprehension. Quarterly Journal of Experimental
Psychology 61(6), 825–50.
Flower, L. & Hayes, J. R. (1981). A cognitive process theory of writing. College
Composition and Communication 32(4), 365–87.
Fodor, J. A. (1983). The Modularity of Mind. Cambridge, MA: Bradford Books.
Fontaine, L., Bartlett, T., and O’Grady, G. (eds.) (2013). Systemic Functional
Linguistics: Exploring Choice. Cambridge, UK: Cambridge University Press.
Frankenstein, J., Büchner, S., Tenbrink, T., & Hölscher, C. (2010). Influence of geo-
metry and objects on local route choices during wayfinding. In C. Hölscher,
T. Shipley, M. Olivetti Belardinelli, J. Bateman, & N. Newcombe (eds.), Spatial
Cognition VII: International Conference, Spatial Cognition 2010, Mt. Hood/
Portland, OR, USA, August 15–19, 2010, pp. 41–53. Berlin/: Springer.
Franklin, N., Henkel, L. A., & Zangas, T. (1995). Parsing surrounding space into
regions. Memory and Cognition 23, 397–407.
Freksa, C. (1981). Linguistic pattern characterization and analysis. Dissertation,
University of California, Berkeley.
Garrod, S. & Pickering, M. J. (2004). Why is conversation so easy? Trends in Cognitive
Sciences 8(1), 8–11.
Gee, J. P. (2010). An Introduction to Discourse Analysis: Theory and method. London:
Routledge.
Gee, J. P. (2013). Discourse versus discourse. In C. A. Chapelle (ed.), The Encyclopedia
of Applied Linguistics. Oxford: Blackwell. [Link]/pdfs/Big%20
D%[Link]?i=1
Gentner, D. (2010). Bootstrapping the mind: Analogical processes and symbol systems.
Cognitive Science 34(5), 752–75.
Gentner, D., Levine, S., Dhillon, S., & Poltermann, A. (2009). Using structural align-
ment to facilitate learning of spatial concepts in an informal setting. In B. Kokinor,
K. Holyoak, & D. Gentner (eds.), Proceedings of the Second Analogy Conference,
pp. 175–82. Sofia, Bulgaria: NBU Press.
Giles, H. & Coupland, N. (1991). Language: Contexts and Consequences. Milton
Keynes: Open University Press.
Goldberg, A. (1995). Constructions: A Construction Grammar Approach to Argument
Structure. University of Chicago Press.
Gralla, L. (2013). Linguistic representation of problem solving processes in unaided
object assembly. PhD dissertation, Bremen University, Germany.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 260 References

Gralla, L. & Tenbrink, T. (2013). ‘This is a wall’ – assigning function to objects. In

M. Knauff, M. Pauen, N. Sebanz, & I. Wachsmuth (eds.), Proceedings of the 35th
Annual Meeting of the Cognitive Science Society: CogSci 2013, pp. 513–18. Austin,
TX: Cognitive Science Society.
Gralla, L., Tenbrink, T., Siebers, M., & Schmid, U. (2012). Analogical problem solving:
Insights from verbal reports. In N. Miyake, D. Peebles, & R. P. Cooper (eds.),
Proceedings of the 34th Annual Conference of the Cognitive Science Society,
pp. 396–401. Austin, TX: Cognitive Science Society.
Gramann, K., Müller, H. J., Eick, E. M., & Schönebeck, B. (2005). Evidence of
separable spatial representations in a virtual navigation task. Journal of
Experimental Psychology: Human Perception and Performance 31(6), 1199–223.
Grant, E. R. & Spivey, M. J. (2003). Eye movements and problem solving: Guiding
attention guides thought. Psychological Science 14, 462–6.
Grice, H. P. (1975). Logic and conversation. In P. Cole, & J. Morgan (eds.), Syntax and
semantics (Vol 3), pp. 41–58). New York: Academic Press.
Grosz, B. J. & Sidner, C. L. (1986). Attention, intentions and the structure of discourse.
Computational Linguistics 12(3), 175–204.
Gugerty, L. & Rodes, W. (2007). A cognitive model of strategies for cardinal direction
judgments. Spatial Cognition and Computation 7(2), 179–212.
Gülşen, T. T. (2016). You tell me in emojis. In T. Ogata & T. Akimoto (eds.),
Computational and Cognitive Approaches to Narratology, pp. 354–75. Hershey,
PA: IGI Global.
Habel, C. & Tappe, H. (1999). Processes of segmentation and linearization in describing
events. In R. Klabunde & C. von Stutterheim (eds.), Representations and Processes in
Language Production, pp. 117–53. Wiesbaden: Springer.
Haddington, P. (2010). Turn-taking for turntaking: Mobility, time and action in the
sequential organization of junction-negotiations in cars. Research on Language and
Social Interaction 43(4), 372–400.
Halliday, M. A. K. & Matthiessen, C. M. I. M. (2014). Halliday’s Introduction to
Functional Grammar (4th ed.). London: Routledge.
Hanna, J. E., Tanenhaus, M. K., & Trueswell, J. C. (2003). The effects of common
ground and perspective on domains of referential interpretation. Journal of Memory
and Language 49(1), 43–61.
Hart, C. & Lukeš, D. (eds.) (2009). Cognitive Linguistics in Critical Discourse Analysis:
Application and Theory. Newcastle upon Tyne: Cambridge Scholars Publishing.
Hayes, A. F. & Krippendorff, K. (2007). Answering the call for a standard reliability
measure for coding data. Communication Methods and Measures 1, 77–89.
Henderson, J. M. & Ferreira, F. (eds.) (2004). The Interface of Language, Vision, and
Action: Eye Movements and the Visual World. New York: Psychology Press.
Henderson, J. M., Malcolm, G. L., & Schandl, C. (2009). Searching in the dark:
Cognitive relevance drives attention in real-world scenes. Psychonomic Bulletin
and Review 16(5), 850–6.
Hobbs, J. R. (1995). Sketch of an ontology underlying the way we talk about the world.
International Journal of Human–Computer Studies 43(5/6), 819–30.
Hoffman, R. R., Trafton, G., & Roebber, P. (2005). Minding the Weather: How Expert
Forecasters Think. Cambridge, MA: MIT Press.
Holsanova, J. (2008). Discourse, Vision, and Cognition. Amsterdam: Benjamins.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 261

Hölscher, C., Meilinger, T., Vrachliotis, G., Brösamle, M., & Knauff, M. (2006). Up the
down staircase: Wayfinding strategies and multi-level buildings. Journal of
Environmental Psychology 26(4), 284–99.
Hölscher, C., Tenbrink, T., & Wiener, J. (2011). Would you follow your own route
description? Cognition 121, 228–47.
Jee, B. D., Gentner, D., Forbus, K., Sageman, B., & Uttal, D. H. (2009). Drawing on
experience. In N. A. Taatgen & H. van Rijn (eds.), Proceedings of the 31st Annual
Meeting of the Cognitive Science Society, pp. 2499–504. Austin, TX: Cognitive
Science Society.
Johnson-Laird, P. N. (1983). Mental Models. Cambridge, MA: Harvard University Press.
Johnstone, B. (2008). Discourse Analysis (2nd ed.). Hoboken, NJ: Blackwell Publishers.
Kahneman, D. (1973). Attention and Effort. Englewood Cliffs, NJ: Prentice-Hall.
Kastens, K. A., Agrawal, S., & Liben, L. S. (2009). How students and field geologists
reason in integrating spatial observations from outcrops to visualize a 3-D geological
structure. International Journal of Science Education 31(3), 365–93.
Klabunde, R. & von Stutterheim, C. (eds.) (1999). Representations and Processes in
Language Production. Wiesbaden: Springer.
Klein, W. (1979). Wegauskünfte (Route descriptions). Zeitschrift für Literaturwissenschaft
und Linguistik (LiLi) 33, 9–57.
Klippel, A. (2003). Wayfinding choremes. In W. Kuhn, M. Worboys, & S. Timpf (eds.),
Spatial Information Theory: Foundations of Geographic Information Science,
pp. 320–34. Berlin: Springer.
Knapton, O. & Rundblad, G. (2014). Public health in the UK media: Cognitive
discourse analysis and its application to a drinking water emergency. In C. Hart &
P. Cap (eds.), Contemporary Critical Discourse Studies, pp. 559–82. London:
Bloomsbury.
Krahmer, E. & Ummelen, N. (2004). Thinking about thinking aloud: A comparison of
two verbal protocols for usability testing. IEEE Transactions on Professional
Communication 47(2), 105–17.
Krippendorff, K. (2012). Content Analysis: An Introduction to Its Methodology (3rd
ed.). London: Sage.
Kristiansen, G. & Dirven, R. (eds.) (2008). Cognitive Sociolinguistics: Language
Variation, Cultural Models, Social Systems (Vol. 39). Berlin: Walter de Gruyter.
Kuipers, B. J. & Kassirer, J. P. (1984). Causal reasoning in medicine: Analysis of a
protocol. Cognitive Science 8, 363–85.
Kuipers, B. J., Moskowitz, A. J., & Kassirer, J. P. (1988). Critical decisions under
uncertainty: Representation and structure. Cognitive Science 12, 177–210.
Ladd, R. D. (1996). Intonational Phonology. Cambridge, UK: Cambridge University
Press.
Lakoff, G. & Johnson, M. (1980). Metaphors We Live By. University of Chicago Press.
Lakoff, G. & Johnson, M. (1999). Philosophy in the Flesh: The Embodied Mind and Its
Challenge to Western Thought. New York: Basic Books.
Langacker, R. W. (1986). An introduction to cognitive grammar. Cognitive Science
10, 1–40.
Langacker, R. W. (2000). A dynamic usage-based model. In M. Barlow & S. Kemmer
(eds.), Usage-Based Models of Language, pp. 1–64. Stanford, CA: CSLI
Publications.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 262 References

Langacker, R. W. (2002). Concept, Image, and Symbol: the Cognitive Basis of

Grammar. Berlin: Mouton de Gruyter.
Larkin, J. H. & Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand
words. Cognitive Science 11, 65–99.
Le Guen, O. (2011). Speech and gesture in spatial language and cognition among the
Yucatec Mayas. Cognitive Science 35(5), 905–38.
Leow, R. P., Grey, S., Marijuan, S., & Moorman, C. (2014). Concurrent data elicitation
procedures, processes, and the early stages of L2 learning: A critical overview.
Second Language Research 30(2), 111–27.
Levinson, S. C. (1997). From outer to inner space: Linguistic categories and non-
linguistic thinking. In J. Nuyts & E. Pederson (eds.), Language and
Conceptualization, pp. 13–45. Cambridge, UK: Cambridge University Press.
Levinson, S. C. (2003). Space in Language and Cognition. Cambridge, UK: Cambridge
University Press.
Linde, C. & Labov, W. (1975). Spatial networks as a site for the study of language and
thought. Language 51, 924–39.
Lindsey, A. E., Greene, J. O., Parker, R. G., & Sassi, M. (1995). Effects of advance
message formulation on message encoding: Evidence of cognitively based hesitation in
the production of multiple-goal messages. Communication Quarterly 43(3), 320–31.
Lutz, A., Lachaux, J. P., Martinerie, J., & Varela, F. J. (2002). Guiding the study of brain
dynamics by using first-person data: Synchrony patterns correlate with ongoing
conscious states during a simple visual task. Proceedings of the National Academy
of Sciences 99(3), 1586–91.
Maes, A., Arts, A., & Noordman, L. (2004). Reference management in instructive
discourse. Discourse Processes 37(2), 117–44.
Martin, J. R. & Rose, D. (2003). Working with Discourse: Meaning beyond the Clause.
London: Continuum International Publishing Group Ltd.
Mast, F. W. & Kosslyn, S. M. (2002). Visual mental images can be ambiguous: Insights
from individual differences in spatial transformation abilities. Cognition 86(1),
57–70.
McManus, I. C., Zhou, F. A., l’Anson, S., et al. (2011). The psychometrics of photo-
graphic cropping: The influence of colour, meaning, and expertise. Perception 40(3),
332–57.
Meilinger, T. & Knauff, M. (2008). Ask for your way or use a map: A field experiment
on spatial orientation and wayfinding in an urban environment. Spatial Science 53(2),
13–24.
Mercier, H. & Sperber, D. (2011). Why do humans reason? Arguments for an argumen-
tative theory. Behavioral and Brain Sciences 34(2), 57–74.
Montague, R. (1970). Universal grammar. Theoria 36(3), 373–98.
Montello, D. R. (1993). Scale and multiple psychologies of space. In A. Frank &
I. Campari (eds.), Spatial Information Theory: A Theoretical Basis for GIS,
pp. 312–21. Berlin: Springer.
Moratz, R. & Tenbrink, T. (2006). Spatial reference in linguistic human-robot interac-
tion: Iterative, empirically supported development of a model of projective relations.
Spatial Cognition and Computation 6(1), 63–106.
Moray, N. (1959). Attention in dichotic listening: Affective cues and the influence of
instructions. Quarterly Journal of Experimental Psychology 11(1), 56–60.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 263

Neisser, U. & Becklen, R. (1975). Selective looking: Attending to visually specified

events. Cognitive Psychology 7, 480–94.
Nissen, M. J. & Bullemer, P. (1987). Attentional requirement of learning: Evidence from
performance measures. Cognitive Psychology 19, 1–32.
Newell, A. & Simon, H. A. (1972). Human Problem Solving. Englewood Cliffs, NJ:
Prentice-Hall.
Newman, E. L., Caplan, J. B., Kirschen, M. P., et al. (2007). Learning your way around
town: How virtual taxicab drivers learn to use both layout and landmark information.
Cognition 104(2), 231–53.
Norman, D. A. & Shallice, T. (1980). Attention to action: Willed and automatic control
of behavior. In R. J. Davidson, G. E. Schwartz, & D. Shapiro (eds.), Consciousness
and Self-Regulation, pp. 1–15. New York: Plenum.
Noveck, I. A. & Sperber, D. (2006). Experimental Pragmatics. London: Palgrave
Macmillan.
Nuyts, J. & Pederson, E. (eds.) (1997). Language and Conceptualization. Cambridge,
UK: Cambridge University Press.
Olloqui Redondo, J., Tenbrink, T., & Foltz, A. (2019.) Effects of animacy and linguistic
construction on the interpretation of spatial descriptions in English and Spanish.
Language and Cognition 11(2), 256–84.
O’Malley, M., Innes, A., Muir, S., & Wiener, J. M. (2018). ‘All the corridors are the
same’: A qualitative study of the orientation experiences and design preferences of
UK older adults living in a communal retirement development. Ageing and Society 38
(9), 1791–816.
Pashler, H. (1984). Processing stages in overlapping tasks: Evidence for a central
bottleneck. Journal of Experimental Psychology: Human Perception and
Performance 10(3), 358–77.
Piaget, J. (1952). The Origins of Intelligence in Children. New York: International
Universities Press.
Pick, H. L., Heinrichs, M. R., Montello, D. R., Smith, K., & Sullivan, C. N. (1995).
Topographic map reading. In P. A. Hancock, J. M. Flach, J. Caird, & K. J. Vicente
(eds.), Local Applications of the Ecological Approach to Human-Machine Systems
(Vol. 2), pp. 255–84. Hillsdale, NJ: Lawrence Erlbaum Associates.
Pickering, M. & Garrod, S. (2004). Toward a mechanistic psychology of dialogue.
Behavioral and Brain Sciences 27, 169–226.
Purcell, T. & Gero, J. S. (1998). Drawings and the design process: A review of protocol
studies in design and other disciplines and related research in cognitive psychology.
Design Studies 19(4), 389–430.
Pütz, M., Robinson, J. A., & Reif, M. (eds.) (2014). Cognitive Sociolinguistics: Social
and Cultural Variation in Cognition and Language Use (Vol. 59). Chicago: John
Benjamins.
Ranyard, R., Crozier, W. R., & Svenson, O. (eds.) (1997). Decision Making: Cognitive
Models and Explanations. London: Routledge.
Rasinger, S. (2010). Research questions in linguistics. In L. Litosseliti (ed.), Research
Methods in Linguistics, pp. 49–67. London: Continuum.
Reason, J. (1990). Human Error. Cambridge, UK: Cambridge University Press.
Riecke, B. E. & Wiener, J. M. (2007). Can people not tell left from right in VR? Point-to-
origin studies revealed qualitative errors in visual path integration. In W. R. Sherman,

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 264 References

M. C. Lin, & A. Steed (eds.), Proceedings of IEEE Virtual Reality 2007, pp. 3–10.
Washington: IEEE Computer Society.
Röder, B. & Rösler, F. (2003). Memory for environmental sounds in sighted, congeni-
tally blind and late blind adults: Evidence for cross-modal compensation.
International Journal of Psychophysiology 50(1), 27–39.
Rosch, E. (1978). Principles of categorization. In E. Rosch & B. B. Lloyd (eds.),
Cognition and Categorization, pp. 27–48 Hillsdale, NJ: Lawrence Erlbaum
Associates.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., Boyes-Braem, P. (1976). Basic
objects in natural categories. Cognitive Psychology 8, 382–439.
Roth, T. (1985). Sprachstatistisch objektivierbare Denkstilunterschiede zwischen
‘guten’ und ‘schlechten’ Bearbeitern komplexer Probleme. Sprache und Kognition
4, 178–91.
Russo, J. E., Johnson, E. J., & Stephens, D. L. (1989). The validity of verbal protocols.
Memory and Cognition 17, 759–69.
Sachs, J. S. (1974). Memory in reading and listening to discourse. Memory and
Cognition 2(1), 95–100.
Sacks, H. (1992). Lectures on Conversation. Oxford: Blackwell.
Sacks, O. (2010). The Mind’s Eye. New York: Knopf.
Sanders, T. (1997). Semantic and pragmatic sources of coherence: On the categorization
of coherence relations in context. Discourse Processes 24, 119–47.
Sarjakoski, L. & Nivala, A. M. (2005). Adaptation to context: A way to improve the
usability of mobile maps. In L. Meng, A. Zipf, & T. Reichenbacher (eds.), Map-Based
Mobile Services, Theories, Methods and Implementations, pp. 107–23. Berlin:
Springer.
Schiffrin, D. (1987). Discourse Markers. Cambridge, UK: Cambridge University Press.
Schlobinski, P. (1996). Empirische Sprachwissenschaft. Berlin: Springer.
Schober, M. F. (1995). Speakers, addressees, and frames of reference: Whose effort
is minimized in conversations about location? Discourse Processes 20(2),
219–47.
Schober, M. F. (1998). Different kinds of conversational perspective-taking. In
S. R. Fussell & R. J. Kreuz (eds.), Social and Cognitive Psychological
Approaches to Interpersonal Communication, pp. 145–74. Mahwah, NJ:
Lawrence Erlbaum.
Schober, M. F. & Brennan, S. E. (2003). Processes of interactive spoken discourse: The
role of the partner. In A. C. Graesser, M. A. Gernsbacher, & S. R. Goldman (eds.),
Handbook of Discourse Processes, pp. 123–64. Mahwah, NJ: Lawrence Erlbaum
Associates.
Schönebeck, B., Thanhäuser, J., & Debus, G. (2001). Die Tunnelaufgabe: Eine Methode
zur Untersuchung kognitiver Teilprozesse räumlicher Orientierungsleistungen (The
tunnel task: A method for the investigation of cognitive subprocesses of spatial
orientation performance). Zeitschrift für Experimentelle Psychologie 4, 339–64.
Schooler, J. W., Ohlsson, S., & Brooks, K. (1993). Thoughts beyond words: When
language overshadows insight. Journal of Experimental Psychology: General 122(2),
166–83.
Seifert, I. (2008). Spatial planning assistance: A cooperative approach. Dissertation,
Bremen University, Germany.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 265

Selting, M. (2000). The construction of units in conversational talk. Language in Society

29, 477–517.
Senft, G. (1994). Spatial reference in Kilivila: The Tinkertoy Matching Games – A case
study. Language and Linguistics in Melanesia 25, 55–93.
Shepard, R. N. & Metzler, J. (1971). Mental rotation of three-dimensional objects.
Science 171(3972), 701–3.
Shi, H., Ross, R. J., Tenbrink, T., & Bateman, J. (2010). Modelling illocutionary
structure: Combining empirical studies with formal model analysis. In A. Gelbukh
(ed.), Computational Linguistics and Intelligent Text Processing, pp. 340–53. LNCS
6008. Berlin: Springer.
Sholl, M. J. (1988). The relationship between sense of direction and mental geographic
updating. Intelligence 12(3), 299–314.
Siegel, A. W. & White, S. H. (1975). The development of spatial representations of
large-scale environments. In H. W. Reese (ed.), Advances in Child Development and
Behavior, pp. 9–55. New York: Academic Press.
Siegler, R. S. & Stern, E. (1998). Conscious and unconscious strategy discoveries:
A microgenetic analysis. Journal of Experimental Psychology: General 127(4),
377–97.
Simon, H. A. (1956). Rational choice and the structure of the environment.
Psychological Review 63(2), 129–38.
Sitter, S. & Stein, A. (1996). Modeling information-seeking dialogues: The conversa-
tional roles model. Review of Information Science 1(1), 165–80.
Smagorinsky, P. (1998). Thinking and speech and protocol analysis. Mind, Culture, and
Activity 5, 157–77.
Smith, W. & Dror, I. E. (2001). Psychonomic Bulletin and Review 8, 732–41.
Spenader, J. (2002). Presupposed propositions in a corpus of dialogue. In K. van
Deemter & R. Kibble (eds.), Information Sharing: Reference and Presupposition in
Language Generation and Interpretation. Stanford, CA: CSLI Publications.
Sperber, D. & Wilson, D. (1986). Relevance: Communication and Cognition.
Cambridge, MA: Harvard University Press.
Spiers, H. J. & Maguire, E. A. (2008). The dynamic nature of cognition during
wayfinding. Journal of Environmental Psychology 28(3), 232–49.
Sternberg, R. J. & Ben-Zeev, T. (2001). Complex Cognition: The Psychology of Human
Thought. Oxford: Oxford University Press.
Steube, A. (ed.) (2004). Information Structure: Theoretical and Empirical Aspects.
Berlin: Mouton de Gruyter.
Talmy, L. (2000). Toward a Cognitive Semantics. Cambridge, MA: MIT Press.
Talmy, L. (2007). Attention phenomena. In D. Geeraerts & H. Cuyckens (eds.),
Handbook of Cognitive Linguistics, pp. 264–93. Oxford: Oxford University
Press.
Taylor, J. (1989). Linguistic Categorization. Oxford: Clarendon.
Taylor, H. A. & Tversky, B. (1996). Perspective in spatial descriptions. Journal of
Memory and Language 35, 371–91.
Tenbrink, T. (2007). Space, Time, and the Use of Language: An Investigation of
Relationships. Berlin: Mouton de Gruyter.
Tenbrink, T. (2008). The verbalization of cognitive processes: Thinking-aloud data and
retrospective reports. In W. Ramm & C. Fabricius-Hansen (eds.), Linearisation and

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 266 References

Segmentation in Discourse. Multidisciplinary Approaches to Discourse 2008 (MAD

08), pp. 125–35. Oslo: Oslo University Press.
Tenbrink, T. (2011). Reference frames of space and time in language. Journal of
Pragmatics 43(3), 704–22.
Tenbrink, T. (2014). Cognitive discourse analysis for cognitively supportive visualisa-
tions. Proceedings of DECISIVe 2014, Open Access Repository, University of
Konstanz.
Tenbrink, T. (2015). Cognitive discourse analysis: Accessing cognitive representa-
tions and processes through language data. Language and Cognition 7(1),
98–137.
Tenbrink, T., Andonova, E., & Coventry, K. R. (2008). Negotiating spatial relationships
in dialogue: The role of the addressee. In J. Ginzburg, P. Healey, & Y. Sato (eds.),
Proceedings of LONDIAL, pp. 201–8. London: Queen Mary University.
Tenbrink, T., Andonova, E., Schole, G., & Coventry, K. R. (2017). Communicative
success in spatial dialogue: The impact of functional features and dialogic strategies.
Language and Speech 60(2), 318–29.
Tenbrink, T., Bergmann, E., Hertzberg, C., & Gondorf, C. (2016). Time will not help
unskilled observers to understand a cluttered spatial scene. Spatial Cognition and
Computation 16(3), 192–219.
Tenbrink, T., Bergmann, E., & Konieczny, L. (2011a). Wayfinding and description
strategies in an unfamiliar complex building. In C. H. Carlson & T. F. Shipley
(eds.), Proceedings of the 33rd Annual Conference of the Cognitive Science
Society, pp. 1262–7. Austin, TX: Cognitive Science Society.
Tenbrink, T., Brösamle, M., & Hölscher, C. (2012). Flexibility of perspectives in
architects’ thinking. In C. Hölscher & M. Bhatt (eds.), Proceedings of SCAD
Spatial Cognition for Architectural Design. Report No. 029–08/2012 Report Series
of the Transregional Collaborative Research Center SFB/TR 8 Spatial Cognition,
pp. 215–23. Bremen: Bremen University.
Tenbrink, T., Coventry, K. R., & Andonova, E. (2011b). Spatial strategies in the
description of complex configurations. Discourse Processes 48, 237–66.
Tenbrink, T., Hölscher, C., Tsigaridi, D., & Conroy Dalton, R. (2014). Cognition and
communication in architectural design. In D. R. Montello, K. E. Grossner, &
D. G. Janelle (eds.), Space in Mind: Concepts for Spatial Learning and Education,
pp. 263–80. Cambridge, MA: MIT Press.
Tenbrink, T. & Maas, A. (2015). Efficiently connecting textual and visual information in
operating instructions. IEEE Transactions on Professional Communication 58(4),
346–66.
Tenbrink, T., Ross, R. R., Thomas, K. E., Dethlefs, N., & Andonova, E. (2010). Route
instructions in map-based human-human and human-computer dialogue:
A comparative analysis. Journal of Visual Languages and Computing 21(5),
292–309.
Tenbrink, T. & Salwiczek, L. (2016). Orientation and metacognition in virtual space.
Journal of Experimental Psychology: Human Perception and Performance 42(5),
683–705.
Tenbrink, T. & Schilder, F. (2003). (Non)temporal concepts conveyed by before, after,
and then in dialogue. In P. Kühnlein, H. Rieser, & H. Zeevat (eds.), Perspectives on
Dialogue in the New Millennium, pp. 353–80. Amsterdam: Benjamins.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 References 267

Tenbrink, T. & Seifert, I. (2011). Conceptual layers and strategies in tour planning.
Cognitive Processing 12(1), 109–25.
Tenbrink, T. & Taylor, H. A. (2015). Conceptual transformation and cognitive processes
in Origami paper folding. Journal of Problem Solving 8(1), 2–22.
Tenbrink, T. & Wiener, J. (2009). The verbalization of multiple strategies in a variant of
the traveling salesperson problem. Cognitive Processing 10(2), 143–61.
Tenbrink, T. & Winter, S. (2009). Variable granularity in route directions. Spatial
Cognition and Computation 9, 64–93.
Tomasello, M. (2003). Constructing a Language: A Usage-Based Theory of Language
Acquisition. Cambridge, MA: Harvard University Press.
Toulmin, S. (1958). The Uses of Argument. Cambridge, UK: Cambridge University
Press.
Trognon, A., Batt, M., & Laux, J. (2011). Why is dialogical solving of a logical problem
more effective than individual solving? A formal and experimental study of an
abstract version of Wason’s task. Language and Dialogue 1(1), 44–78.
Tversky, B. (1999). Spatial perspective in descriptions. In P. Bloom, M. A. Peterson,
L. Nadel, & M. F. Garrett (eds.), Language and Space, pp. 109–69. Cambridge, MA:
MIT Press.
Tversky, B. & Hard, B. M. (2009). Embodied and disembodied cognition: Spatial
perspective-taking. Cognition 110, 124–9.
Tversky, B. & Lee, P. (1999). Pictorial and verbal tools for conveying routes. In
C. Freksa & D. M. Mark (eds.), Spatial Information Theory: Cognitive and
Computational Foundations of Geographic Information Science (COSIT ’99),
pp. 51–64. Berlin: Springer.
Tyler, A. & Evans, V. (2003). The Semantics of English Prepositions: Spatial
Sciences, Embodied Meaning, and Cognition. Cambridge, UK: Cambridge
University Press.
van Deemter, K. & Kibble, R. (eds.) (2002). Information Sharing: Reference and
Presupposition in Language Generation and Interpretation. Stanford, CA: CSLI
Publications.
van Deemter, K. & Peters, S. (eds.) (1996). Semantic Ambiguity and Underspecification.
Stanford, CA: CSLI Publications.
van Dijk, T. A. (2000). Cognitive discourse analysis: An introduction. www
.[Link]/unpublished%20articles/[Link]
van Dijk, T. A. (2008). Discourse and Power. Houndsmills: Palgrave-MacMillan.
van Someren, M. W., Barnard, Y. F., & Sandberg, J. (1994). The Think Aloud Method:
A Practical Guide to Modelling Cognitive Processes. London: Academic Press.
Vorwerg, C. (2009). Consistency in successive spatial utterances. In K. Coventry,
T. Tenbrink, & J. Bateman (eds.), Spatial Language and Dialogue, pp. 40–55.
Oxford: Oxford University Press.
Vorwerg, C. & Tenbrink, T. (2007). Discourse factors influencing spatial descrip-
tions in English and German. In T. Barkowsky, M. Knauff, G. Ligozat, &
D. Montello (eds.), Spatial Cognition V: Reasoning, Action, Interaction,
pp. 470–88. Berlin: Springer.
Wachsmuth, I., de Ruiter, J., Jaecks, P., & Kopp, S. (eds.) (2013). Alignment in
Communication: Towards a New Theory of Communication. Amsterdam: John
Benjamins.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 268 References

Warren, P. (2005). Patterns of late rising in New Zealand English: Intonational variation
or intonational change? Language Variation and Change 17(2), 209–30.
Wen, W., Ishikawa, T., & Sato, T. (2013). Individual differences in the encoding
processes of egocentric and allocentric survey knowledge. Cognitive Science 37,
176–92.
Whorf, B. L. (1941). The relation of habitual thought and behavior to language. In
L. Spier (ed.), Language, Culture, and Personality: Essays in Memory of Edward
Sapir, pp. 75–93. Menasha, WI: Sapir Memorial Publication Fund.
Wilson, M. (2002). Six views of embodied cognition. Psychonomic Bulletin and Review
9(4), 625–36.
Winterboer, A., Tenbrink, T., & Moratz, R. (2013). Spatial directionals for robot
navigation. In M. Dimitrova-Vulchanova & E. van der Zee (eds.), Motion Encoding
in Spatial Language, pp. 84–101. Oxford: Oxford University Press.
Wittgenstein, L. (1953). Philosophical Investigations. New York: Macmillan.
Yang, S. C. (2003). Reconceptualizing think-aloud methodology: Refining the encoding
and categorizing techniques via contextualized perspectives. Computers in Human
Behavior 19, 95–115.
Zacks, J. M., Tversky, B., & Iyer, G. (2001). Perceiving, remembering, and commu-
nicating structure in events. Journal of Experimental Psychology: General 130,
29–58.
Zhang, Q., Walsh, M. M., & Anderson, J. R. (2018). The impact of inserting an
additional mental process. Computational Brain and Behavior 1(1), 22–35.

Downloaded from [Link] Access paid by the UCSB Libraries, on 25 Jan 2020 at 10:57:58, subject to the Cambridge
Core terms of use, available at [Link] [Link]
 Index

accommodation, 174 common sense, 143, 241

ACT-R, 234 communication, 8, 10, 14, 31, 34, 46, 56, 78,
adaptation, 173 93, 100, 114, 125, 135, 150, 151, 160, 169,
addressee, 24, 56, 152, 169, 170, 207, 250 171, 173, 176, 190, 230, 250
adjective, 97, 118, 124 concept, 10, 11, 22, 36, 38, 58, 61, 63, 81, 82,
analysis 99, 103, 113, 122, 123, 145, 174, 188, 194,
qualitative, 19, 27, 30, 109, 110, 114, 119, 211, 216, 226, 245
121, 138, 140, 161, 165, 198, 211, 213, conceptual aspect. See analysis perspective
220–2, 245 conceptual perspective, 64, 103, 110–16,
quantitative, 19, 34, 42, 131, 138, 140, 161, 180, 247
211, 214, 219, 220, 222–5, 240 conceptual strategy, 30, 31, 192
analysis perspective, 23, 33, 40, 49, 83, 84, 152, conjunction, 22–3, 100, 116
171, 236 content, 3, 21, 22, 23, 49, 53, 54, 74, 84, 124,
Anderson, John Robert, 33, 93, 95, 234 166, 181, 183, 184, 196, 213, 220, 235
annotation, 22, 24, 53, 124, 126, 184, 194, 214, content analysis, 21, 22, 26, 54, 184, 213–14,
216–17, 218, 219–20, 222, 235 215, 217, 234, 235
applications, 47, 226, 248 content category, 21, 22, 54, 164, 165, 166,
article 181, 184, 185, 213–14, 215, 216, 222, 224
definite, 23, 24, 78, 217, 218, 223 conversation analysis, 77
indefinite, 77 Coventry, Kenny R., 16, 17, 19, 26, 29, 40, 179,
assembly, 160 202, 237
attention, 2, 31, 83, 92, 116, 182, 187, 190,
230, 236 Dąbrowska, Ewa, 26
awareness, 11, 43, 48, 52, 94, 132, 150, data collection, 38, 49, 197
172, 234 decision making, 14, 20, 25
demonstrative, 161, 162, 172
back-channelling, 172, 176 detail, 13, 17, 28, 43, 76, 81, 99, 117, 118, 119,
behavioural data, 47, 138, 186, 187 123, 130, 141, 149, 173, 227
dialogue, 37, 39, 48, 55, 100, 107, 149, 168,
certainty, 86, 128, 132, 137, 161 171, 172, 173, 175, 178, 179, 180, 191,
choice, 3, 12, 13, 15, 18, 28, 34, 38, 40, 64, 65, 208, 237, 250
81, 83, 84, 97, 98, 113, 171, 195, 200, dialogue system, 171, 178, 226, 240, 242,
222, 230 250
Chomsky, Noam, 58, 64 discourse
clause, 13, 68, 85, 87, 88, 89, 90, 124, 211–13 relations, 79, 80
cognition structure, 53, 99, 100, 132, 238
complex, 14, 20, 26, 156, 181, 182, 186, 187, discourse analysis, 32, 56, 72, 77, 78, 80, 83,
190, 202, 207, 236 91, 152, 194, 227
human, 2, 10, 11, 142, 146, 241 distribution pattern, 211, 221
spatial, 16, 23, 35, 122, 243
cognitive load, 136, 174 EEG, 231, 232
cognitive strategy, 46, 185, 190–3, 241 elaboration, 123, 125, 126, 151

269

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 11:21:46, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 270 Index

empirical, 28, 29, 33, 67, 155, 168, 171, Likert scale, 249
197, 200 limit, 30, 31, 35, 42, 59, 60, 92, 117, 131, 147,
Ericsson, K.A., 21, 22, 28, 30, 44, 46, 47, 48, 169, 173, 196, 197, 202, 206, 207,
49, 50, 52, 53, 56, 163, 181, 182, 184, 190, 213, 230
204, 205, 206, 207 linguistics, cognitive, 15, 32, 57–72, 95,
ethics, 196 113, 227
Evans, Vyvyan, 11, 59, 60, 70, 71 linguistics, corpus, 34, 179, 196, 211, 227, 240
experience, 9, 35, 36, 38, 59, 65, 72, 93, 95, linguistics, systemic functional, 15
100, 119, 120, 123, 125, 131, 144, 145,
146, 149, 156, 179, 197, 234, 246 machine learning, 240
experimental design, 15, 28, 38, 39, 40, 46, marker
197–201, 231, 235 certainty, 129, 132–7
expertise, 27, 56, 61, 102, 128, 129, 132, 137, discourse, 22, 54, 55, 73, 79–80, 181,
141, 153, 246 220, 236
explicit, 22, 25, 49, 80, 122, 143, 144, 150, 152 hesitation, 30, 135, 136, 162
extensions, 226 orientation, 140, 141
politeness, 170
fMRI, 231, 232 uncertainty, 27, 28, 137, 138, 140, 141
focus group, 207, 227 mental rotation, 157, 158
metaphor, 11, 69–72, 73, 113, 115
generalisability, 30 modality, 13, 55, 86, 133, 161, 162, 170,
Goldberg, Adele, 71 190, 220
grammar model
cognitive, 64–5, 97 cognitive, 227, 232–6
functional, 32, 68, 79, 83–91, 132, 133, 134, mood, 13, 83–4, 85–6, 170, 175–6
136, 169
granularity, 117, 117–28, 236, 241, 247 neuroscience, cognitive, 30
Grice, Paul, 121, 150 non-sequitur, 236
noun, 64, 98, 113, 161, 239
Halliday, Michael, 22, 24, 78, 79, 83, 85, 86,
88, 89, 98, 124, 132, 134, 136, 169, 175, observable, 142, 227
176, 178 operationalisation, 21, 126, 138, 154, 181, 189,
hesitation. See pause 211, 217, 222
orientation, 18, 19, 92–116, 157, 165, 192,
implicit, 1, 29, 79, 80, 97, 107, 120, 122, 124, 219, 244
145, 152, 161, 216, 231, 234, 236
indicator, 2, 25, 54, 108, 134, 181, 185, 190 pattern, usage, 69, 71
inference, 19, 43–4, 79–80, 83, 121, 122, 142, pause, 30, 135, 136, 209, 235
142–55, 166 performance, 10, 46, 48, 52, 95, 129, 157,
information structure, 73, 77–8, 90, 217 158, 228
instruction, 29, 39, 47, 52, 156, 163, 187, 196, perspective, 16, 17, 18, 29, 31, 36, 65, 66, 103,
200, 202, 205, 207, 209, 249 103–16, 118, 207, 212
intelligence, artificial, 240–3 Piaget, Jean, 159
interaction, 14, 48, 98, 103, 107, 137, 162, 169, Pinker, Steven, 157
170, 174, 179, 191, 208, 242 preposition, 70–1, 165, 219
interview, 42, 46, 55, 56, 74, 102, 114, 116, presupposition, 24–5, 73, 78, 216, 217, 218,
207, 227, 244, 246 223, 236
intonation contour, 76, 209, 210 principles, 2, 43, 63, 83, 143, 148, 150, 175,
178, 181, 194, 197, 204, 212, 221, 245
Johnson, Mark, 8, 11, 69, 70 problem solving, 14, 20, 21, 23, 42, 44, 47, 52,
53–6, 130, 163, 202, 204–7, 213–14,
Lakoff, George, 8, 11, 69, 70, 113 227–9, 236, 248, 249
Langacker, Ronald, 64, 65, 68, 96–7 procedures, 21, 54, 74, 148, 184, 197, 208, 211,
LCCM, 59–60 213, 216, 222, 224, 236, 248
Levinson, Stephen, 10, 66, 67, 68, 104, 105 pronoun, 13, 79, 90, 115, 161, 172

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 11:21:46, subject to the Cambridge Core terms of
use, available at [Link] [Link]
 Index 271

protocol, verbal, 20–1, 22, 45, 46, 49, 50, 52, 179–81, 183, 186, 187, 188–90, 191–3,
53–6, 131, 152, 168, 171, 181–93, 201–8, 203, 207, 228–9, 231–2, 239–40
217, 232, 233, 234, 235, 249 syntax, 13, 64–6, 82, 173, 224
prototype theory, 62–3 system, spatial reference, 16, 17–19, 35, 66–8,
psycholinguistics, 15, 16, 32, 37–41, 227, 230 103–10, 219
psychology, cognitive, 20, 32, 50, 53–6, 108, systematic, 2, 5, 12, 14, 21, 24, 29, 36, 46, 48,
129–32, 181 55, 57, 63, 77, 129, 130, 184, 210,
218, 249
questionnaire, 50, 51, 207, 227, 229, 249
Talmy, Leonard, 63, 64, 65, 97, 98
reaction times, 38, 44, 230 Taylor, Holly A., 106, 107, 109, 144, 157, 163,
referential communication, 40 164, 166, 214, 215, 222, 249
reliability, 26, 216 Tenbrink, Thora, 16, 18, 19, 23, 24, 25, 28, 29,
report, retrospective, 21, 25, 46, 49, 52, 55, 109, 30, 40, 48, 49, 101, 104, 107, 109, 114,
187, 190, 206, 207, 232 115, 116, 122, 125, 126, 128, 137, 141,
representation, 10, 245 144, 147, 153, 155, 156, 160, 163, 164,
linguistic, 7, 15, 24, 31, 32, 46, 90, 124, 169, 166, 171, 179, 180, 183, 185, 186, 188,
188, 202, 204 194, 196, 202, 203, 207, 214, 215, 216,
mental, 14, 15, 19, 33, 38, 89, 97, 142, 183, 219, 221, 222, 223, 226, 228, 237, 238,
202, 204 239, 241, 242, 243, 244, 247, 249, 250
research think-aloud, 22, 25, 30, 46, 47, 48, 49, 53, 109,
goal, 15, 195, 196, 201, 207, 211 135, 138, 162, 163, 166, 186, 206, 207,
purpose, 33, 195, 210, 213, 214, 237 220, 231, 232, 238, 244
robot, 177, 226, 241, 242 Tomasello, Michael, 8, 68, 69, 100
Rosch, Eleanor, 62, 63 transcription, 74–7, 182, 194, 208–10
route description. See wayfinding transferred meaning. See metaphor
transformation, 155, 155–67, 228, 229
scale, 29, 67, 75, 119, 124 triangulation, 201, 226–40
science Tversky, Barbara, 9, 19, 36, 106, 107, 109
cognitive, 3, 31, 45, 48, 92, 93, 95, 99, 117, Tyler, Andrea, 11, 70, 71
137, 168, 181, 184, 196, 204, 226, 227
social, 174, 207 validation, 21
segmentation, 211 variability, 31, 37, 60, 136, 227, 232, 236
semantics, cognitive, 63–4 verb, 54, 64, 68, 71, 87–9, 97, 134, 161, 162,
signal, linguistic, 45, 54, 77, 78, 80, 101, 141, 183, 185, 189, 220, 239
172, 174 verbalisability, 7, 30, 44, 206
Simon, Herbert A., 20, 21, 22, 28, 30, 44, 46, verbosity, 131
47, 48, 49, 50, 52, 53, 56, 147, 163, 181,
182, 184, 190, 204, 205, 206, 207 wayfinding, 9, 10, 36, 37, 47, 125, 128, 138,
statistics, 27, 34, 198, 199, 216, 221, 224 140, 147, 190–3, 238, 242, 243, 244
study, 16–19, 22, 25, 39, 53–4, 101–3, 114–16, Whorf, Benjamin, 11, 101
137–41, 148–9, 153–5, 159–67, 171, Wittgenstein, Ludwig, 61–2

Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 11:21:46, subject to the Cambridge Core terms of
use, available at [Link] [Link]
Downloaded from [Link] University of Exeter, on 25 Jan 2020 at 11:21:46, subject to the Cambridge Core terms of
use, available at [Link] [Link]