Beginning Syntax

A coherent introduction to generative syntax by a leader in the field,

this textbook leads students through the theory from the very beginning,
assuming no prior knowledge. Introducing the central concepts in a sys
tematic and engaging way, it covers the goals of generative grammar, tacit
native-speaker knowledge, categories and constituents, phrase structure,
movement, binding, syntax beyond English and the architecture of gram
mar. The theory is built slowly, showing in a step-by-step fashion how
different versions of generative theory relate to one another.
Examples are carefully chosen to be easily understood, and a
comprehensive glos-sary provides clear definitions of all the key terms
introduced. With end of chapter exercises, broader discussion questions,
and annotated further reading lists,

beginning syntax is the ideal resource for instructors and beginning

undergraduate students of syntax alike. Two further textbooks by Ian
Roberts, Continuing Syntax and Comparing Syntax, will take students to
intermediate and advanced level.

IanRoberts_9781316519493_C000.indd 1 10-09-2022 16:54:50

CAMBRIDGE TEXTBOOKS IN LINGUISTICS

General editors: u. ansa ld o, p. aust i n, b. com r i e, r. las s,

d. lightf oot, k. r ice, i. roberts, s. rom a i n e, m. sheehan,
i. tsimpli

Beginning Syntax
An Introduction to Syntactic Analysis
IanRoberts_9781316519493_C000.indd 2 10-09-2022 16:54:50
In this series:

r. cann Formal Semantics

j. laver Principles of Phonetics
f. r. palmer Grammatical Roles and Relations
m. a. jones Foundations of French Syntax
a. radford Syntactic Theory and the Structure of English: A Minimalist
Approach
r. d. van valin jr and r. j. lapolla Syntax: Structure, Meaning and
Function
a. duranti Linguistic Anthropology
a. cruttenden Intonation Second edition
j. k. chambers and p. trudgill Dialectology Second edition
 c. lyons Definiteness
r. kager Optimality Theory
j. a. holm An Introduction to Pidgins and Creoles
g. g. corbett Number
c. j. ewenh. van der hulst The Phonological Structure of Words
f. r. palmer Mood and Modality Second edition
b. j. blake Case Second edition
e. gussman Phonology: Analysis and Theory
m. yip Tone
w. croft Typology and Universals Second edition
f. coulmas Writing Systems: An Introduction to Their Linguistic Analysis
p. j. hopper and e. c. traugott Grammaticalization Second edition
l. white Second Language Acquisition and Universal Grammar
i. plag Word-Formation in English
w. croft and a. cruse Cognitive Linguistics
a. siewierska Person
a. radford Minimalist Syntax: Exploring the Structure of English
d.büring Binding Theory
m. butt Theories of Case
n. hornstein, j. nuñes and k. grohmann Understanding Minimalism
b. c. lust Child Language: Acquisition and Growth
g. g. corbett Agreement
j. c. l. ingram Neurolinguistics: An Introduction to Spoken Language
Processing and Its Disorders
j. clackson Indo-European Linguistics: An Introduction
m. ariel Pragmatics and Grammar
r. cann, r. kempson and e. gregoromichelaki Semantics: An
Introduction to Meaning in Language
y. matras Language Contact
d. biber and s. conrad Register, Genre, and Style
l. jeffries and d. mcintyre Stylistics
r. hudson An Introduction to Word Grammar
m. l. murphy Lexical Meaning
j. m. meisel First and Second Language Acquisition
t. mcenery and a. hardie Corpus Linguistics: Method, Language and
Practice
j. sakel and d. l. everett Linguistic Fieldwork: A Student Guide
a. spencera. luís Clitics: An Introduction
g. corbett Features
a. mcmahon and r. mcmahon Evolutionary Linguistics
b. clark Relevance Theory
b. longpeng Analyzing Sound Patterns
b. dancygier and e. sweetser Figurative Language
j. bybee Language Change
s. g. thomason Endangered Languages: An Introduction
a. radford Analysing English Sentences Second edition
r. clift Conversation Analysis

IanRoberts_9781316519493_C000.indd 3 10-09-2022 16:54:50

r. levine Syntactic Analysis
i. plag Word-Formation in English Second edition
z. g. szabó and r. h. thomason Philosophy of Language
j. pustejovsky and o. batiukova The Lexicon
d. biber and s. conrad Register, Genre, and Style Second edition
f.zúñiga and s. kittilä Grammatical Voice
y. matras Language Contact Second edition
t. hoffmann Construction Grammar
w. croft Morphosyntax: Constructions of the World’s Languages
i. roberts Beginning Syntax

Earlier issues not listed are also available.

IanRoberts_9781316519493_C000.indd 4 10-09-2022 16:54:50

Beginning Syntax
An Introduction to Syntactic Analysis

IAN ROBERTS
Downing College, University of Cambridge
IanRoberts_9781316519493_C000.indd 5 10-09-2022 16:54:50

Shaftesbury Road, Cambridge CB2 8EA, 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
103 Penang Road, #05–06/07, Visioncrest Commercial, Singapore 238467

Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.

[Link]
Information on this title: [Link]/9781316519493
DOI: 10.1017/9781009023849
© Ian Roberts 2023
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 & Assessment.
First published 2023
 Printed in <country> by <printer>
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
ISBN 978-1-316-51949-3 Hardback
ISBN 978-1-009-01058-0 Paperback
Cambridge University Press & Assessment 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.

IanRoberts_9781316519493_C000.indd 6 10-09-2022 16:54:51

Contents

Preface page xi List of Abbreviations xiii

Introduction: What Linguistics Is About and

What Syntax Is About 1 What Is (a) Language? 3 What Is Language? vs What
Is a Language? 3
Natural Language vs Artificial Languages 4
Language and Languages 7 Natural Language 9 Exercises 12 For
Further Study and/or Discussion 13 Further Reading 13

1 Tacit Knowledge (or: Several Things You Didn’t Know

You Knew about English) 15 1.1 Introduction 15 1.2 Some Fish 15 1.3
More Fish and Some Relative Clauses 22 1.4 Really Challenging Fish 25 1.5
Conclusion: What the Fish Have Taught Us 31 Exercises 32 For Further Study
 and/or Discussion 33 Further Reading 34

2 Constituents and Categories 35 2.1 Introduction and Recap 35 2.2

Categories 36
2.2.1 Introduction: Lexical and Functional Categories 36 2.2.2
Morphological Diagnostics for Categories 38 2.2.3 Syntactic
Diagnostics for Categories 39 2.2.4 Phonological Diagnostics for
Categories 42 2.2.5 Semantics and Categories 42 2.2.6 Conclusion 44
2.3 Constituent Structure 44 2.3.1 Phrasal Constituents 44 2.3.2 Hierarchical
Relations among Constituents 45 2.4 Conclusion 47

vii

IanRoberts_9781316519493_C000.indd 7 10-09-2022 16:54:51

viii contents

Exercises 48
Further Reading 49

3 Phrase-Structure Rules and Constituency Tests 51

3.1 Introduction 51
3.2 Phrase-Structure Rules 52
3.3 Recursion 55
3.4 Constituency Tests 59
3.5 Summary and Conclusion 75
Exercises 76
For Further Discussion 77
Further Reading 77

4 X′-theory 78
4.1 Introduction 78
4.2 Possible and Impossible PS-Rules 78
4.3 Introducing X′-theory 80
4.4 X′-theory and Functional Categories: The Structure of the Clause 87
4.5 Adjuncts in X′-theory 95
4.6 Conclusion 96
Exercises 97
For Further Discussion 98
Further Reading 98

5 Movement 99
5.1 Introduction and Recap 99
5.2 Types of Movement Rules 100
5.3 Wh-Movement: The Basics 104
5.4 Subject Questions, Indirect Questions and Long-Distance
Wh-Movement 112
5.5 The Nature of Movement Rules 118
5.6 Limiting Wh-Movement 122
5.7 Conclusion 126
Exercises 127
For Further Discussion 128
Further Reading 128
 6 Binding 130
6.1 Introduction and Recap 130
6.2 Pronouns and Anaphors 130
6.3 Anaphors 132
6.4 Pronouns 142
6.5 R-expressions 143
6.6 The Binding Principles 144
6.7 Variables, Principle C and Movement 146
6.8 Conclusion 153

IanRoberts_9781316519493_C000.indd 8 10-09-2022 16:54:51

Contents ix

Exercises 154
For Discussion 154
Further Reading 155

7 Syntax beyond English 156

7.1 Introduction and Recap 156
7.2 Approaching Universal Grammar 156
7.3 Verb Positions in French and English 157
7.4 Word Order and X′-theory 160
7.5 A Case Study: German 165
7.6 The Wh-Movement Parameter 171
7.7 Conclusion 174
Exercises 176
For Further Discussion 178
Further Reading 178

8 The Architecture of Grammar 179

8.1 Introduction and Recap 179
8.2 A Model of Grammar 179
8.3 The Lexicon 180
8.4 D-structure 187
8.5 S-structure 190
8.6 Phonological Form (PF) 195
8.7 Logical Form (LF) 203
8.8 Conclusion 211
Exercises 212
Further Reading 214

Conclusion 215

Glossary 219
Index 234
IanRoberts_9781316519493_C000.indd 9 10-09-2022 16:54:51
IanRoberts_9781316519493_C000.indd 10 10-09-2022 16:54:51

Preface

The purpose of this book is to present the essential elements of the theory of
syntax, presupposing no prior knowledge of either syntax or linguistics more
generally. The first chapter introduces the thinking behind modern formal
linguistics, defining the core background concepts. The second chapter is an
extended demonstration of linguistic competence, revealing to an English
speaker their tacit, untutored knowledge of the intricacies of the syntax of the
language. The next four chapters introduce the core technical concepts of syn
tax: phrase structure, constituency, movement rules and construal rules. Chapter
7 is a very brief introduction to comparative syntax, illustrating how the theory
developed in the earlier chapters can apply to languages other than English.
Finally, Chapter 8 discusses the overall architecture of grammar, introducing
several levels of representation, including the interface levels of Phonological
Form and Logical Form.
The book is based on a first-year lecture course I have taught at Cambridge
for most of the past ten years, called ‘Structures’. That course has successfully
laid the groundwork for more advanced study; several professional
syntacticians had their first exposure to the subject in that context.
I do not attempt to review either the history or the state of the art in contem
porary generative syntax. Instead, the book is intended to provide a solid basis
from which the student can move on to more advanced topics. In the decades
since its inception, generative theory has yielded a range of core insights and
concepts which can be studied independently of the details of a given
 framework and which form the core of any formal approach to syntax (and core
background to the formal study of other areas of linguistics, in particular
semantics). These are the focus of this introduction, where the theoretical
orientation of the book is most clearly set out. This volume is intended as the
first in a series; future vol
umes will build on the material presented here, as well as both adding an explic
itly comparative and typological dimension and engaging more directly with the
minimalist programme for linguistic theory. The overall aim of the series is to
provide a complete course in syntax, taking the student from beginner level to
being able to engage with contemporary research literature.
This volume is intended for students starting out in syntax and assumes no
prior knowledge. As mentioned above, it corresponds to a first-year course
taught at Cambridge. It can be taught over a single term or semester with
tutorial

xi

IanRoberts_9781316519493_C000.indd 11 10-09-2022 16:54:51

xii preface

support (as I have done at Cambridge) or over an entire year as a lecture course.
The book tries to introduce the main concepts of syntactic theory in an engaging
way with a minimum of technicalities. Each chapter features exercises, some
straightforwardly testing the material covered in the chapter, others raising fur
ther questions for reflection and discussion (e.g. in tutorials), as well as sugges
tions for further reading. Model answers are provided separately for the first type
of exercise. The second type presupposes student participation in discussions.
Chapter 1 is somewhat different from the others in that it introduces some fairly
complex data; the point at this stage of the exposition is to demonstrate tacit
knowledge, i.e. native-speaker competence; the technical issues are, as appropri
ate, revisited and introduced more gradually later. Instructors may wish to skip
over some of this material and/or revisit it later (for example, the Italian data can
supplement the material covered in Chapter 7).
I would like to thank the other people who have taught this course in whole or
in part over the years: Dora Alexopoulou, Theresa Biberauer, Craig Sailor and
Jenneke van der Wal, as well as the many doctoral students, postdocs and others
who have given tutorials for the course. Most of all, I’d like to thank the students
themselves: you ultimately make everything happen. Thanks also to Bob Freidin
and Dalina Kallulli for comments on early drafts. And of course, thanks to the
two Helens at Cambridge University Press: Helen Barton and Helen Shannon,
whose tolerance of missed deadlines fully matches my capacity to bring them
about.
IanRoberts_9781316519493_C000.indd 12 10-09-2022 16:54:51

Abbreviations

AAVE African-American Vernacular English

acc accusative case
Agr agreement
A(P) Adjective (Phrase)
Adv(P) adverb (phrase)
ASL American Sign Language
Aux(P) auxiliary (phrase)
BSL British Sign Language
C(P) Complementiser (Phrase)
CNPC Complex NP Constraint
Comp complementiser
Conj conjugation
D(P) determiner (phrase)
gen genitive case
 LBC Left Branch Condition
LF Logical Form
M mood
Mod modifier
Neg(P) negation (phrase)
nom nominative case
N(P) noun (phrase)
NSL Nicaraguan Sign Language
num number
OV object-verb
OSV object-subject-verb
OVS object-verb-subject
PF phonological form
Pl plural (preceded by an Arabic numeral, denotes the relevant person, e.g.
3Pl = third-person plural)
P(P) preposition(al phrase)
Perf(P) perfect (phrase)
Pres present
Prog(P) progressive (phrase)
pst past
S sentence

xiii

IanRoberts_9781316519493_C000.indd 13 10-09-2022 16:54:51

xiv list of abbreviations

SASL South African Sign Language

Sg singular (preceded by an Arabic numeral, denotes the relevant
person, e.g. 3Sg = third-person singular)
SOV subject-object-verb
SVO subject-verb-object
Spec specifier
top topic (marker)
T(P) tense (phrase)
UG Universal Grammar
VO verb-object
V(P) verb (phrase)
VOS verb-object-subject
VSO verb-subject-object
WALS World Atlas of Language Structures (Haspelmath et al. 2013)
IanRoberts_9781316519493_C000.indd 14 10-09-2022 16:54:51

Introduction: What Linguistics Is About

and What Syntax Is About

This book is an introduction to one part of the general subject of linguistics:

syn tax (words which are boldfaced are either technical terms or names of
languages and language families which may not be familiar; at the end of the
book you can find a glossary where these terms are defined). Syntax is the
study of the struc ture of sentences, how these sentences arise and how speakers
of a language are able to use and understand sentences by having a mental
representation of their structure. In this book, the language we focus on is
English, not because there is anything particularly special about English, but
just because that’s the one language I can be sure you know, since you’re
reading this book.
Syntax is one of the central sub-disciplines of modern linguistics. The other
central areas of linguistics are phonology (the study of sounds and sound sys
tems) and semantics (the study of meaning). There are many other aspects to
linguistics (e.g. morphology, the study of the form of words), but these three
form the core of language: the fundamental thing about language is that it con
nects sound and meaning through sentences. Phonology deals with the sounds,
semantics deals with the meanings and syntax deals with the sentences. Syntax
is the bridge between the sound and the meaning of sentences. We’ll have
plenty more to say about this in the chapters to follow, but first let’s look at
linguistics more broadly.
Modern linguistics is the scientific study of language. The education systems
in many parts of the world make a major distinction between sciences (like biol
ogy, physics and chemistry) and arts or humanities (like history, music or lit
erature). Languages are almost always classified on the arts/humanities side, so
it might seem odd to talk about looking at language and languages scientifi
cally. ‘Science’ conjures up visions of men (mostly) in white coats carrying out
experiments using various kinds of specialised equipment, all of which seems a
far cry from reading Shakespeare, learning French verbs or trying out Spanish
conversation.
But really, looking at something scientifically means two main things. First, it
means observing things as part of nature (rocks, stars or slugs, for exam ple).
Language is so much a part of us that it can be difficult to think of it as a
natural object, the way rocks, stars and slugs obviously are. But imagine you
were a Martian observing planet Earth: you’d see rocks, slugs, plenty of insects
and a featherless biped building machines (including spacecraft), in control of
everything, and making noises with its breathing and eating apparatus. As an

IanRoberts_9781316519493_Introduction.indd 1 05-09-2022 11:17:14

2 introduction

intelligent Martian, you’d soon recognise that there is some connection between
the bipeds’ noises, their ability to build machines and their control over all the
other species. You’d also notice that only the featherless bipeds make these
noises and that the young of the species spontaneously start making the noises
when they’re very small and still dependent on their parents for survival. After
abducting some bipeds and looking inside their brains, you’d realise that there’s
just a couple of areas of the biped brain which seem to control the noise-making
ability. The noises are language (Martians would probably first notice phonol
ogy, but being super-intelligent they’d quickly realise that there was more to
language than just that). By looking at ourselves through Martian ‘eyes’ (which
might of course be infrared heat sensors attached to their feet, but never mind),
we can appreciate language as a natural object.
The second thing about science is that it’s not just about observing things;
it’s also about putting the observations together to make a theory of things. One
of the most famous of all scientific theories is Einstein’s Theory of Relativity.
This is hardly the place to go into that, but the point is that Einstein and other
physicists wanted to construct a general understanding, based on observation, of
certain aspects of nature: the Theory of Relativity is mostly about space, time
and gravity. A good theory, like relativity, makes sense of the observations we
have already made and also predicts new ones that we should be able to make.
So saying linguistics is scientific doesn’t necessarily mean that we have to put
on white coats and start a laboratory. But it does mean that we should try to make
observations about language and put those observations together to make a theory
of language. Since we’re going to focus on syntax, we’re going to make obser
vations about syntax and make a theory of syntax. So the goal of this book is to
present a particular theory of syntax, a scientific theory of what it is that links sound
and meaning (as you can imagine, there are separate but connected theories of pho
nology and semantics). The theory that I will introduce here is called generative
grammar. The reasons for that name will emerge over the next couple of chapters.
Approaching language, or syntax, in this way means that there are aspects
of the more traditional, humanities-style approaches to language that we leave
behind, since they do not contribute to the goal of making observations and
building theories. First, it implies that the goal of linguistics is not to set stand
ards of ‘good’ speaking or writing. This is not to imply that such standards
should not be set, or that linguists, or others with particular expertise in lan
guage, should not be those responsible for setting them. Instead, it means that
setting these standards falls outside the scientific study of language, since doing
this doesn’t involve making observations about what people say, write or under
stand but involves recommending that people should speak or write in a cer
tain way. Zoologists don’t tell slugs how to be slimy; linguists don’t tell people
how to speak or write. Ideas of ‘good’ syntax have no place here. Prescriptive
grammatical statements of a kind once common in the teaching of English in
schools in most of the English-speaking world (see below for some examples)
are similarly irrelevant to our concerns here.

IanRoberts_9781316519493_Introduction.indd 2 05-09-2022 11:17:14

What Is (a) language? 3

A second consequence of leaving behind the traditional humanities-style

approach to the study of language and adopting a scientific approach instead
is that we do not evaluate language aesthetically. This does not mean that the
aesthetic appreciation of language, in particular literary language, is without
interest or value as a study in itself, merely that this represents a different way of
studying (and enjoying) language from the scientific one. Botanists may appre
ciate the beauty of flowers as much as anybody else, and zoologists no doubt
admire the beauty of slugs, but the scientific study of flowers and slugs leaves
aside any aesthetic response to them. Similarly, linguists may well appreciate
the beauty and power of great works of literature, but the goal of modern lin
guistics is not to explicate our reaction to such works; that is for our colleagues
in the literature departments.
There is also a third aspect of a scientific approach to a field of investigation, a
really basic one. That is to begin by defining our terms, so let us start in this way.
If linguistics is the scientific study of language, then we should start by defining
language itself. That, as we shall see, is not as simple as might first appear.

What Is (a) language?

What Is language? vs What Is a Language?

First of all, it is useful to distinguish language as a general concept
from individual languages. Our definition of linguistics makes reference to this
general concept and, as such, tells us that linguistics is concerned with this gen
eral concept rather than with the study of individual languages (and hence tells
us that linguistics is not about acquiring proficiency in individual languages; this
is another aspect of the traditional humanities approach to language that modern
linguistics leaves aside). It is very difficult to give an adequate general definition
of language, especially in advance of the first steps in constructing any kind of
theory of language or its subparts. For the moment, let us define language as the
communicative and cognitive capacity that, as far as we know, most clearly dis
tinguishes humans from other animals; this is what the Martians saw when they
observed terrestrial featherless bipeds making noises. All human societies and
cultures that have ever been discovered have language, most commonly spoken
language; current estimates are that there are over 7,000 languages spoken in
the world ([Link]), although individual languages are hard to
count, and we will see below that there are reasons to question whether English,
French, etc., are really natural objects. On the other hand, no non-human animal
has a linguistic or communicative capacity comparable to ours. This is not to
deny that communication systems of various kinds are easy to observe in other
species, but none of these systems seem to have the structural complexity we
can observe in human language; as we shall see, the nature of the syntax of
human language may lie behind the gulf between human language and animal
communication. Indeed it could be argued that language is not just unique to our

IanRoberts_9781316519493_Introduction.indd 3 05-09-2022 11:17:14

4 introduction

species, but a defining property of it. The official biological name of our species
(in the Linnean taxonomy) is Homo sapiens, Latin for ‘wise man’. Since a
glance at a history book or a newspaper leads one to question the wisdom of our
species, one could think that Homo loquens (‘talking man’) might have been a
better term to define us.
As just mentioned, the general concept of language, roughly defined above,
should be distinguished from individual languages. An individual language
can be taken to be a specific variant of the uniquely human capacity defined
above, usually part of a given culture: English, Navajo, Warlpiri, Basque, etc.
Concepts such as ‘English’ are thus, at least in part, cultural concepts. The dis
tinction between language (in general) and languages (individual languages as
partly cultural entities) distinguishes two variants of a single word in English.
This is rather like the way we can distinguish the general concept of cheese
from individual cheeses such as Brie, Cheddar, Parmigiano Reggiano and so on.
Language and cheese are mass nouns; they denote general concepts. Individual
cultures in different parts of the world produce their local variants of the gen
eral thing, associated with countable versions of the same nouns, i.e. individual
languages and cheeses. In several European languages, different words are used
to make the distinction between language and languages. French distinguishes
langage (language in general) from langue (individual languages); in Italian lin
guaggio is distinguished from lingua in a similar way, and Spanish distinguishes
lenguaje from lengua.

Natural Language vs Artificial Languages

A very important distinction in the context of defining the object of
study of scientific linguistics is that between natural languages and artificial lan
guages. Natural languages are natural objects (like slugs, etc., as we have seen),
and so they can be the object of scientific study. Natural languages are languages
spoken as native languages (they are almost always somebody’s mother tongue)
and are intrinsically capable of fulfilling a full range of communicative func
tions (making statements, asking questions, giving orders, etc.). Moreover, the
origins of natural languages are obscure in that we are unable to fix a specific
time or place for where they began. For example, handbooks on the history of
the English language typically date the beginning of English to the arrival of the
Angles, Saxons and Jutes in the British Isles starting in the fifth century CE. The
Angles, Saxons and Jutes spoke closely related Germanic dialects which origi
nated where they themselves originated, in what is now Southern Denmark and
Northern Germany. As far as we know, nothing changed about their language
when it was transplanted, along with its speakers, from the mainland to the off
shore islands. Hence, the date attached to the ‘beginning of English’ reflects
a random aspect of external history and almost certainly does not reflect any
linguistic change at all. English is a direct continuation of the Germanic dialects
spoken on the continent before the fifth century by the Angles, Saxons and Jutes.

IanRoberts_9781316519493_Introduction.indd 4 05-09-2022 11:17:14

What Is (a) language? 5

These dialects were in turn a continuation of one dialect of Indo European,

which was originally spoken (as far as we know) three or four millennia earlier
in either the Pontic-Caspian steppe region of what is now Ukraine or in Central
Anatolia (modern Turkey). Again, the beginnings and endings of ‘English’,
‘Germanic’ and ‘Indo European’ are arbitrarily chosen historical events. The
languages themselves simply continue; what came before Indo European is
uncertain, and this further illustrates the point that natural languages have an
essentially unknown origin. All we can be sure of is that our capacity for lan
guage must have evolved at some stage in the history of our species, but this
event is lost in the mists of time many millennia before Indo European or its
precursors.
Artificial languages, on the other hand, are languages designed for some
specific purpose and restricted in terms of their functions. We can usually say
when and by whom they were invented. As such, they are not natural objects.
So, for example, logic, which can thought of as an artificial language, was
invented by Aristotle ca 400 bce, and the various computer-programming lan
guages in use today and in recent years were all invented after roughly 1950,
when modern computers themselves were invented. Semaphore, Morse code
and other signalling systems are also artificial languages of this general kind,
and also highly restricted in function. Artificial languages are of less interest for
our purposes here, since they are not natural objects. So I will leave them aside
in what follows.
In this context, we should mention other kinds of language and languages.
Sign language is of great interest. In fact, sign language is not a single linguistic
entity: there are various different Sign Languages, e.g. British Sign Language
(BSL), American Sign Language (ASL), Nicaraguan Sign Language (NSL),
South African Sign Language (SASL) and many others in many parts of the
world. Sign languages are primarily used by Deaf communities. It is now an
established finding of modern linguistics that sign languages are natural lan
guages in precisely the sense defined above: they are natural objects. Sign lan
guages serve the full range of communicative functions and their origins are
obscure. Most importantly, sign languages are not manual versions of the spo
ken or written language used by the hearing communities around them. For
example, BSL is the creation of the British Deaf community. Its history goes
back to accounts in the fifteenth century describing deaf people using signs. The
first description of those signs appears in the Marriage Register of St Martin’s,
Leicester, in 1576, describing the vows signed by Thomas Tillsye, who ‘was
and is naturally deafe and also dumbe, so that the order of the forme of mar
riage used usually amongst others which can heare and speake could not for his
parte be observed’ ([Link]/british-sign-language-history/beginnings/
marriagecertificate-thomas-tillsye).
The only difference between sign language and familiar languages such
as English is that sign languages use a different medium, the gestural/visual
medium, while familiar languages use the oral/aural medium (and of course

IanRoberts_9781316519493_Introduction.indd 5 05-09-2022 11:17:14

6 introduction

writing). One of the striking features of natural languages, both spoken and
signed, is that their salient structural properties are independent of their medium
of transmission: sign-language syntax, for example, has the same structural
properties as English, French and other spoken languages.
Constructed languages, or conlangs, are also relevant here. Conlangs are of
two main kinds. First, there are languages which were deliberately constructed
as ‘ideal’ languages intended to serve for more efficient communication than
could be afforded by seemingly imperfect already existing languages. Esperanto
is probably the best known, although far from the only, language of this kind.
Esperanto was invented by L. L. Zamenhof in 1887, at a time when there was
no clear international lingua franca in the Western world (since 1945, English
has played this role; at earlier times Latin and French did). The vocabulary of
Esperanto is a mixture of Romance, Germanic and Slavic elements, and the
grammatical system is a highly simplified and artificially regularised. The sys
tem is, however, closely based on various European, principally Romance,
languages. As such it is arguably a natural language, which happens to have a
particular origin and purpose. It is debatable, however, whether Esperanto has
any true native speakers, although it is certainly used as a second language by
up to 2 million people all around the world (compare this with the estimated 2
billion second-language speakers of English). Many other invented languages
(Interlingua, Volapük, etc.) have the same status as Esperanto, although these
days they are hardly spoken at all.
The other principal kind of constructed languages are those invented for
fictional purposes, in order to give linguistic realism to invented worlds and
their denizens. Among the best-known examples are the languages invented by
J. R. R. Tolkien in his extensive mythological writings (The Lord of the Rings,
The Hobbit, The Silmarillion, etc.). Tolkien was a professional philologist, and
so the languages he constructed have an air of authenticity. Like Esperanto,
however, they are largely based on existing, mainly European, languages. They
are thus natural languages which are unusual in their origin and purpose; they
are also barely spoken at all and certainly have no native speakers. Above all for
this last reason, we leave such languages aside here; our interest is in languages
which are acquired naturally.
Two other kinds of ‘language’ should also be mentioned. First, there is ‘body
language’: frowning, shaking or nodding one’s head, crossing one’s arms or
legs, etc. Body language can certainly communicate various attitudes or emo
tional states (friendliness, aggression, etc.), but it is not a language in the sense
that it lacks the structural properties of natural, spoken and signed, languages;
it has no discernible syntax, for example. Second, there is music. Music is often
said to be a language, and indeed it has been shown to have structural, syntac
tic properties which are akin to, perhaps even identical to, natural language.
However, although music may have a profound emotional impact, it lacks a clear
propositional semantics, in that it cannot communicate true or false statements.
It may be that music is a cognitive capacity which shares some, but not all, of

IanRoberts_9781316519493_Introduction.indd 6 05-09-2022 11:17:14

Language and Languages 7

the properties of natural languages. Certainly, music appears to be both univer

sal across human cultures and unique to humans; in these important respects, it
resembles natural language.
 To sum up, our focus here is on natural language. Modern linguistics is really
about natural language in the sense described at the beginning of this section
and, henceforth, I will restrict the discussion along these lines. Linguists thus
believe that it makes sense to study natural language in general and not just
individual languages. If natural language forms a coherent object of study, and
if individual languages are specific variants of language in general, then this
implies that all individual languages must have something in common. Hence, a
major focus of linguistic research, particularly since the middle of the twentieth
century, but with much older historical roots, has been the following question:
what are the common properties of natural languages not shared by other sys
tems of communication?1 One of the central goals of modern linguistic theory
is the attempt to answer this question, and we will address this question as it
applies to syntax in much of what follows.

Language and Languages

Once we define natural language in the way we have done so far,
its study can still be approached from various perspectives. As we mentioned
earlier, one approach, which comes from the traditional humanities, is the pre
scriptive one. This involves defining precepts of ‘good’ English, in principle
independently of how native speakers might actually speak or write the lan
guage. This has led to the formulation of precepts such as ‘Don’t use two neg
atives, since that makes a positive.’ In many varieties of non-standard English
all over the English-speaking world, it is easy to observe people saying things
like I don’t like no-one or Mick can’t get no satisfaction, and so on. Here two
negatives are used, but the result is not a positive. Invoking an artificial kind of
logic, which does not correspond to the facts we can observe about the syntax
and the meanings of these sentences, is not in line with the scientific approach
to language we are adopting here. Such artificial ‘rules’ mostly originate in the
eighteenth and nineteenth centuries and reflect little of substance about the true
nature of English or any other language.2 So I will say no more about ‘rules’ of
this kind (although negative sentences of the kind shown above are very com
mon in many languages of the world, including French; they illustrate a phe
nomenon known as negative concord).

1
See R. Robins (1967), A Short History of Linguistics from Plato to Chomsky, London: Long
man, and V. Law (2003), The History of Linguistics in Europe, Cambridge: Cambridge Univer
sity Press, for discussion.
2
See R. Freidin (2020), Adventures in English Syntax, Cambridge: Cambridge University Press,
for a very illuminating discussion.

IanRoberts_9781316519493_Introduction.indd 7 05-09-2022 11:17:14

8 introduction

Another perspective is a teleological one. This is certainly relevant for the

study of individual languages with the goal of attaining proficiency in speaking
and/or writing them, another important aspect of the traditional way of studying
language and languages. If I say I am learning to speak Catalan, then ‘speaking
Catalan’ here is defined as a goal I am striving towards. It does not correspond
to any currently existing knowledge or ability I have. But this is clearly some
thing distinct from studying Catalan, or any other individual language, with the
goal of understanding its structure; in particular, understanding its structure in
relation to the general question of the possible structural commonalities of all
natural languages, as discussed in the previous section. Hence, the teleological
approach to the study of individual languages, although of course worthwhile,
falls outside of modern linguistics as we understand it here.
Still another perspective on language is the sociopolitical one. One very
important distinction in this connection is that between a dialect and a ‘stand
ard’ language. Despite the social and political importance of this distinction, if
we consider individual languages and dialects to be variants of the more general
notion of language, then we are led to consider all natural-language systems as
equal, whatever their social or political status (or that of the people who speak
them). If we consider language to be a single coherent entity, then we expect to
find the same general structural characteristics in all individual languages and
dialects; this, of course, is the initial hypothesis that we intend to investigate and,
if possible, substantiate.
Moreover, all natural languages and dialects appear to be of roughly the same
level of complexity. Although this point is hard to assess and verify in detail and
in a fully satisfactory way, there does not appear to be any reason to assert that
one or another language or dialect is ‘simpler’ or ‘more basic’ or ‘more primi
tive’ than another. In structural terms, then, there is no reason to privilege one
language or dialect over any other; the fact that one particular variety may have
emerged as a standard is usually just a historical accident. For example, Standard
(British) English emerged in the fifteenth and sixteenth centuries as English
became the main language of commerce and correspondence. Standard English
came to predominate as the written form of the language and came to be taught in
schools as the ‘correct’ or ‘standard’ form. The emergence of Standard English
was a consequence of social and political factors; it had nothing to do with the
structural features of that variety of English, and there is accordingly no reason to
think that that particular variety is in any way intrinsically superior to any other.
So, from the point of view we adopt here, the distinction between dialects and
standard languages is an artificial one, extrinsic to the structural features of the
varieties in question. When we look at language variants as natural objects, we
see that the language vs dialect distinction is spurious; it comes from culture, not
nature. Back in 1945, the American sociolinguist Max Weinreich captured this
point by saying that ‘a language is a dialect with an army and a navy’. Political,
economic and military power confer prestige on standard languages, but from a
purely linguistic point of view, they are not distinct from dialects.

IanRoberts_9781316519493_Introduction.indd 8 05-09-2022 11:17:15

Natural Language 9
 We can take this line of thought further. The notion of a standard language is
really a type of idealisation, one specially constructed by language planners and
educationalists often interested in telling people what to do. In many countries,
one of the goals of teaching the standard language is teleological, in the sense dis
cussed above. In reality, nobody really speaks the ideal standard language; every
body speaks some form of dialect. One very clear result of the scientific study
of the social aspects of language (the field known as sociolinguistics) is that no
language is homogenous. Everybody, including people who consider themselves
speakers of ‘Standard English’, actually speaks a slightly different variant of the
language, approximating an ideal standard in slightly different ways (which, in
practice, one may hardly ever have cause to notice). As pointed out well over a
century ago by the German historical linguist Hermann Paul, ‘we must in reality
distinguish as many languages as there are individuals’. No two speakers of a
given language actually know precisely the same things about that language or
use it in precisely the same way. Alongside dialects, then, we recognise idiolects,
the variety of a language employed by a particular person. We may also want
to recognise sociolects (varieties associated with particular social classes), eth
nolects (varieties associated with particular ethnic groups) and so on.
All of this may leave you feeling perplexed. If all speakers have different idio
lects, is there a notion of an individual language that can really be usefully defined
at all? Atkinson (1992:23) gave the following definition of an English speaker:
The person in question has an internal system of representation … the overt
products of which (utterance production and comprehension, grammatical
ity judgements), in conjunction with other mental capacities, are such that
that person is judged (by those deemed capable of judging) to be a speaker
of English.3

We can see from this that defining ‘a language’ isn’t as simple as we

might have at first thought.
Leaving aside the social, cultural and political dimensions (which are evident
in Atkinson’s definition), we can try to understand what ‘language’ is so that we
can then define ‘a language’ as a specific variant (token) of this more general
entity (type). In this way, we take the everyday terms for individual languages
(English, French, etc.) to be primarily sociocultural. Strictly speaking, these
terms are not, in fact, part of scientific linguistics, since they designate cultural
rather than natural objects.

Natural Language
Leaving aside the sociocultural dimensions, then, we concentrate on
looking at natural language from a scientific perspective. More precisely, we

3
M. Atkinson (1992), Children’s Syntax, Oxford: Blackwell.

IanRoberts_9781316519493_Introduction.indd 9 05-09-2022 11:17:15

10 introduction

will look at language from a cognitive perspective. This approach treats lan
guage as a form of knowledge; the central idea is that our uniquely human lan
guage capacity is part of our cognitive make-up. Language is really a kind of
instinct humans, and no other species, are pre-programmed with (in fact, The
Language Instinct is the title of an influential and important book by Stephen
Pinker in which this approach is explained in detail; see Further Reading at the
end of this Introduction). This is the view that forms the basis of the theory of
generative grammar and so this is the view I will adopt here. This approach, and
most of the concepts introduced in this section, are due to Noam Chomsky (see
Further Reading at the end of the Introduction for more details).
The cognitive approach treats our capacity for language as an aspect of human
psychology which is ultimately rooted in biology. In this way, it clearly treats
language, and our capacity for language, as natural objects. On this view, there
are three factors in language design. The first is contributed by genetics: an
innate aptitude for language that is unique to our species, our ‘language instinct’.
The second factor is experience, particularly in early life. The language we are
exposed to as children represents the crucial linguistic experience; when I was
growing up in England, everyone around me spoke English and so I acquired
English. My great-grandfather was surrounded by Welsh speakers when he
was growing up in nineteenth-century North Wales, so he acquired Welsh.
Whichever language, or languages, we are exposed to as small children, our
innate language-learning capacity is brought to bear on this experience in such a
way as to result in our competence as native speakers of our first language (I will
say more about language acquisition below). Third, general cognitive capac
ities, not specific to language and perhaps not specific to humans, clearly play
a role in shaping our knowledge and use of language, although the exact role
these capacities play and how they interact with (and can be distinguished from)
the language-specific aspects of our linguistic capacities are difficult questions.
Together, these three factors constitute the human language faculty. It can be
extremely difficult to distinguish the specifically linguistic aspects that contrib
ute to forming the language faculty from more general cognitive abilities, but
the distinction can certainly be made in principle and is of course very important
for our general cognitive theory of language.
In these terms, Universal Grammar (UG) is the theory of the first factor
which makes up the language faculty: our innate genetic endowment for lan
guage. As already mentioned, UG is assumed to play a central role in language
acquisition. It is also vital in helping us to understand how we can make sense
of the idea that specific languages are actually variants of a single type of entity,
language. Our goal here, then, is to look at one part of UG: how words combine
to form sentences. In so doing, we will construct the theory of syntax.
We can now make an important distinction. Language can be seen from an
internal, individual perspective, I-language, or it can be seen as external to the
individual, E-language. Here we are going to focus on I-language, which arises
from the interaction of the three factors just introduced. This is a natural approach,
 IanRoberts_9781316519493_Introduction.indd 10 05-09-2022 11:17:15
Natural Language 11

given our interest in how syntactic structures arise in the mind. E-language is
in fact a more complicated notion, involving society, culture, history and so on.
Concepts like ‘English’ and ‘French’ in their everyday senses are E-language
concepts; it is for this reason that they do not, strictly speaking, form part of our
object of study. I-language is a natural object; E-language is not.
Taking I-language as the central notion treats language as part of individ
ual psychology. This is a cognitive theory of language, because it intrinsically
involves the human mind. In fact, this theory of language could be part of an
overall theory of the mind. However, we would really like our theory of language
to be part of an overall theory of the brain; being an obviously physical object,
the brain is a natural object, and it is a clearer notion than ‘mind’ (in fact, philos
ophers have worried for centuries about whether ‘mind’ is something physical,
but we don’t need to). Unfortunately, though, at present our understanding of
how brain tissue supports cognitive processes such as language, thought and
memory is extremely limited, and so we are unable to say very much about the
relation between the physical brain and cognitive processes. That’s why I will
continue to use the older, strictly speaking vaguer, term ‘mind’.
What does it mean to say our theory is a formal theory? A formal approach to
any kind of problem or phenomenon assumes discrete, systematic ways of form
ing complex things out of simple things. For example, the alphabet is formal: its
twenty-six letters can be combined in various different, but more or less system
atic, ways to form a very large number of words. Arithmetic is a formal system
combining numbers of various kinds and functions such as addition, subtraction,
multiplication, etc.
In formal syntax and formal semantics, we combine simple elements (roughly
words and their meanings) to form more complex elements (sentences and their
meanings). The modes of combination must be systematic and precise; as we
will see, this is a major part of the challenge of constructing such a theory.
A further natural question to ask is why we want a formal and cognitive theory.
The answer to this lies in certain general trends of thought both in psychology
and in philosophy of mind (at least in the English-speaking world). It is widely
believed that the best way to understand the mind is to think of it as a kind of
computer. Computers manipulate symbols according to formal instructions, i.e.
algorithms and programs. We can think of I-language in these terms as a piece
of cognitive software, a program run on the hardware of the brain (this raises
the intriguing question, which I will not go into here, of who or what wrote the
program). Therefore, our theory of I-language must be a formal theory. What we
are interested in is how our knowledge of I-language is represented in our mind.
If we can get an idea of this, which I believe we can from studying syntax, then
we gain a very important insight into the human mind; what it is to be human,
what it is to be you.
To conclude this general discussion, we will concentrate here on developing
one aspect of a formal, cognitive theory of I-language: the theory of syntax.
As already mentioned, syntax is concerned with how relatively simple units,

IanRoberts_9781316519493_Introduction.indd 11 05-09-2022 11:17:15

12 introduction

words, are combined to form more complex units, sentences. This is taken to
be a cognitive capacity all humans have, as a reflex of their genetic endowment
which includes UG. UG interacts with linguistic experience in early life, and
with domain-general ‘third factors’ as described above, to give rise to mature
adult competence in one’s native language. This competence manifests itself
in the ability to produce and understand an unlimited number of sentences, and
to make judgements regarding both the syntax and the semantics of those sen
tences. In the next chapter, we will look in detail at a concrete example of this
competence in action, for native speakers of English.
Just a final note: given what I’ve said, strictly speaking I shouldn’t talk about
‘English’, as it is not a scientific term since it does not designate a natural object.
Instead of talking about ‘English speakers’, I should really say something like
‘individuals who identify themselves and are identified in their cultural milieu
as possessing an I-language which corresponds to the E-concept “English”’. For
brevity I will gloss over this more accurate formulation and continue to talk about
‘English speakers’; the same goes for other E-language names (Italian, etc.).
More generally, the term ‘language’ will refer to that aggregate of I-languages
whose speakers recognise themselves and each other as belonging to the same
E-language community.
Now we can move from the rather general matters we have been considering
here to actually starting out on the study of syntax. As we delve more and more
into the detailed and intricate nuts and bolts of the theory of syntax in the chapters
to follow, the background issues we have discussed here should be kept in mind
as they form the overall conceptual underpinning to the theory we’ll develop.
But now for the nuts and bolts. Or, actually, the fish.

Exercises
1. Write a short paragraph (not more than half a page as an absolute
maximum; ten–twelve lines per concept would be ideal) to explain
in your own words what the following notions mean to linguists:
• The language faculty
• Formal approaches to the study of language
• ‘Language’ vs ‘languages’
• English

2. You are Martian Space Cadet λxCloverx231057. Write a report

to Martian Mission Control describing your observations having
abducted a featherless terrestrial biped with the aim of understanding
 more about its noise-making capacities.
3. ‘Children are very good imitators and are able to learn when their
mistakes are corrected by their elders.’ How could you go about
refuting these claims about language acquisition?

IanRoberts_9781316519493_Introduction.indd 12 05-09-2022 11:17:15

Further Reading 13

4. Martian Space Cadet λxCloverx231057 has discovered that the ter

restrial featherless bipeds can communicate by means additional to
the noises produced by their vocal apparatus. Report these discover
ies to Mission Control.

For Further Study and/or Discussion

1. No two people speak the same language.
2. Could we understand an alien language?
3. Are conlangs ‘real’ languages?

Further Reading
Adger, D. 2019. Language Unlimited. Oxford: Oxford University Press.
Pinker, S. 1994. The Language Instinct. New York: Harper Perennial Modern Classics.
Roberts, I. 2017. The Wonders of Language, or How to Make Noises and Influence
People. Cambridge: Cambridge University Press.

These three books all provide general introductions to language and linguistics assum
ing no prior knowledge on the reader’s part. Each book has its own perspective: Adger
concentrates on how syntax is fundamental to human language and cognition, and so is
perhaps most in line with our concerns here. Pinker is a classic; it is a witty and engaging
introduction to linguistics with the emphasis on the relation between language and mind.
Roberts offers a comprehensive introduction to several different subfields of linguistics,
ranging from phonetics to historical linguistics; syntax is covered in Chapter 4.
Larson, R. 2010. Grammar as Science. Cambridge, MA: MIT Press, Unit 1. This book
is an introduction to syntax whose central idea is to present the field as an exercise in the
construction of a scientific theory. In this first unit, the central ideas regarding knowledge
of language and Universal Grammar are introduced in an attractive and original way.
Carnie, A. 2013. Syntax: A Generative Introduction. Oxford: Blackwell, Chapter 1.
Another very sound and well-written introduction to generative grammar. This first
chapter presents the central ideas behind syntactic theory, along with a section on dif
ferent approaches to syntax (something I do not attempt here). Isac, D. & C. Reiss.
2008. I-language. Oxford: Oxford University Press, Chapters 1 and [Link] the title sug
gests, the focus of this book is on the I-language approach to linguistics and syntax.
The first chapter starts with a presentation of linguistic data in order to illustrate the
approach. The third chapter considers different notions of language, similar to what has
been presented here but with a different overall slant; reading that chapter will comple
ment this one nicely.
Freidin, R. 2012. Syntax: Basic Concepts and Applications. Cambridge: Cambridge
University Press, Chapter [Link] chapter contains an excellent introduction to I-language,
the nature of grammaticality, native speakers’ ability to recognise deviant sentences as
an indication of knowledge of language, language acquisition and the argument from
the poverty of the stimulus, and, in particular, language production and comprehension.

IanRoberts_9781316519493_Introduction.indd 13 05-09-2022 11:17:15

14 introduction

Adger, D. 2003. Core Syntax: A Minimalist Approach. Oxford: Oxford University

Press, Chapter [Link] chapter also gives an excellent introduction to the central concepts
behind generative grammar, including the notions of grammaticality and acceptability,
tacit knowledge, recursion, poverty of the stimulus and I-language.

Other good general introductions to the goals and assumptions of generative

grammar include the following:
Anderson, S. & D. Lightfoot. 2002. The Language Organ. Cambridge: Cambridge
University Press, Chapters 1 and 2.
Radford, A. 2016. Analysing English Sentences. Cambridge: Cambridge University
Press, Chapter 1
Smith, N. 2005. Chomsky: Ideas and Ideals. Cambridge: Cambridge University Press,
Chapter 1.
 IanRoberts_9781316519493_Introduction.indd 14 05-09-2022 11:17:15

1 Tacit Knowledge (or: Several Things You

Didn’t Know You Knew about English)

1.1 Introduction
In this chapter, we begin our study of the theory of syntax. As we
will see later, there is no end to syntax. It is also rather difficult to discern
where it begins. Here we will make a start by doing two things. First, you’ll
find out some rather surprising facts about your knowledge of English, things
you didn’t know you knew. This will give you a concrete illustration of your
competence in English. Second, by carrying out a little translation exercise,
we’ll get a first glimpse of what makes languages similar to one another and
what makes them differ.
We begin with very simple sentences (see (1)) and build up to rather strange
and complex ones. It’s not necessary to follow every detail of the discussion
here; all the technical ideas are presented again in later chapters. Our goal here
is to demonstrate the nature of your tacit knowledge of English syntax. We can
do it with just one word: fish. This chapter can also be skipped and returned to
later (e.g. after Chapter 7).

1.2 Some Fish

Let’s start with a single, rather banal-looking English word. We’ll

see that we can build some quite striking and intriguing sentences with it. The
word is fish.1 So our first sentence is:
(1) Fish!

Like quite a few basic words in English, fish is actually ambiguous. It can be
understood either as a verb or as a noun. As a noun, it refers to a class of
aquatic animals; as a verb, it refers to the activity of hunting those aquatic
animals. We indicate this ambiguity in the standard way, by surrounding the
word with square brackets, and writing the ‘category label’ (Noun/Verb) as a
subscript to the left bracket. So, the sentence in (1) actually has two distinct
representations:

1
Thanks to my good friend and colleague Professor Robert Freidin of Princeton University for
these examples. The implications of the fish sentences are discussed and explored in more
detail in Chapter 1 of Freidin (2012), . Freidin’s discussion is more detailed than here,
although it focuses exclusively on English. The exercises given there are also well worth
trying.

15

IanRoberts_9781316519493_C001.indd 15 06-09-2022 11:48:49

16 1 tacit knowledge

(2) a. [Noun Fish ]

b. [Verb Fish ]

The brackets are just a way of saying ‘what is inside here is a Noun/Verb’. A
representation like that in (2) is called a labelled bracketing.
The fact that fish, like many other words including cook, book, police, report,
promise and many others, is ambiguous between a verb and a noun is a conse
quence of the fact that English has very few inflectional endings to mark gram
matical information of various kinds. If we compare English with a more richly
inflected language such as Italian, we see that the two versions of fish correspond
to two differently inflected words:
 (3) a. [Noun fish ] = Italian pesce
b. [Verb fish ] = Italian pescare
It is quite easy to see that the Italian words share the root pesc- and distinct
inflections: -e, indicating a singular noun; and -are, indicating the infinitive of
a first-conjugation verb (the infinitive is the basic form of a verb with no tense
marking; ‘first conjugation’ refers to an arbitrary morphological class of verbs
in Italian: there are four conjugations altogether, as we will see below). To cut
a very long historical story short, English has largely lost its equivalents of -e
and -are, and so we are left with the equivalent of the ambiguous root pesc- (you
may also note a family resemblance between pesc- and fish; this is because both
are ultimately derived from an ancient root in the common ancestor language of
English and Italian, Indo European).
The ambiguous single word in (1) and (2) constitutes an entire sentence on its
own. More precisely, each interpretation of fish constitutes a sentence, so really
there are two distinct sentences here. We can represent them as follows (here,
again following standard conventions, ‘Noun’ is abbreviated as N and ‘Verb’ as
V, and S stands for ‘Sentence’):
(4) a. [S [N Fish ]]
b. [S [V Fish ]]
Interpreted as in (4a), the sentence draws attention to the presence of a single fish or
group of fish (note that fish is one of a relatively small group of English nouns that
is identical in the singular and the plural: the plural form fishes exists, but it denotes
several species of fish, not several individual fish of the same species). Interpreted
as in (4b), it is an imperative, indicating an order given to go fishing. Here there
is an implicit second-person pronoun, since an order is naturally understood as
being addressed to an interlocutor, so we could elaborate (4b) as in (5):
(5) [S [N You ] [V fish ]]
Here the pronoun (a kind of noun, hence N) is ‘understood’; we take that to
mean that it is present in the syntactic and semantic representations of the sen
tence, but not in the phonological representation, what you actually hear or say.
This is our first indication that there is more to syntax than meets the ear.

IanRoberts_9781316519493_C001.indd 16 06-09-2022 11:48:49

1.2 Some Fish 17

Two occurrences of the word fish also make up a grammatical sentence:

(6) Fish fish.

The natural interpretation of this sentence is to treat the first fish as a noun and
the second one as a verb, the combination again forming a sentence. We repre
sent this as in (7):
(7) [S [N fish ] [V fish ]]

In this sentence, the noun fish is the subject, understood as carrying out the
action performed by the verb. The verb fish indicates the action the subject car
ries out. So the sentence means “Fish fish stuff”. As this rough gloss indicates,
there is an implicit direct object here, indicating, somewhat vaguely, what is
being fished for, what undergoes the action of fishing.
Alternatively, we can understand the first fish as a verb, and the second fish
as a noun. This is easier to see if we add an exclamation mark to (6) (Fish fish!),
corresponding to a different intonation pattern in speech. Then the sentence has
the structure in (8):
(8) [S [V fish ] [N fish ]]

Here, as in the verbal interpretation of the single-fish example in (1), the verb
is understood as an imperative and so there is a deleted second-person pronoun
(you) as subject:
(9) [S YouN [V fish ] [N fish ]]

The second fish, the noun, is an explicit direct object, indicating that fish are
what is fished.
In Italian, where the ambiguities of the English roots are clarified by inflec
tional endings, the two interpretations of (6) can be rendered as in (10):
(10) a. I pesci pescano. (= (7))
b. Pesca pesci! (= (8/9))

In addition to disambiguating the English sentence in (6), these translations

show us that there might be more silent material in the English examples than
just the deleted pronoun in (8/9). The first word of (10a), i, is the masculine
plural form of the definite article, i.e. ‘the’. It is obligatory in this context
in Italian. What it shows us is that the subject is more than just a noun, but a
phrasal category of which the noun is the most important member: a Noun
Phrase, or NP. The basic function of the definite article in both English and
Italian is to indicate that the noun refers to some unique known existing entity,
in this case what philosophers would call the ‘natural kind’ fish. Since there
is no fundamental meaning difference between (6) with the structure and inter
pretation in (7) and (10a) (in other words, since (10a) is a reasonable translation
of the English sentence), the noun fish in (7) also denotes the unique, known,
existing, natural kind fish. So we could posit a silent “the” in the subject of (7)

IanRoberts_9781316519493_C001.indd 17 06-09-2022 11:48:49

18 1 tacit knowledge

and, correspondingly, an NP rather than just an N. Then (7) lines up exactly

with (10a):
(11) [S [NP the fish ] [V fish ]]
[S [NP I pesci ] [V pescano ]]
Why go to this trouble to line up English and Italian in this way? There are two
reasons. First, the English and the Italian sentences mean the same thing. Although
we tend in our everyday lives to pay more attention to cultural differences than
to cognitive similarities, it seems reasonable to assume that English speakers and
Italian speakers are cognitively alike (we could perhaps say that English speakers
and Italian speakers are I-alike but E-different, thinking of the distinction made
in the previous section between I-language and E-language). In that case, the
semantic representations of the two sentences ought to be the same. It therefore
simplifies the connection (more technically, the ’mapping’) between the syntax
and the semantics if the respective syntactic representations ‘line up’, as in (11).
Secondly, we are attempting to develop a theory of Universal Grammar (UG, see
the Introduction); we therefore want our syntactic representations of comparable
(i.e. semantically matching, or at least highly similar) sentences in different lan
guages to be as uniform as possible. Of course, it is not always possible to achieve
a parallel as straightforward as that depicted in (11), but we should strive towards
this goal as part of the UG project. Ideally, then, we would want (a) the semantic
representations of sentences which mean the same thing to be identical across
languages, (b) the syntactic representations of such sentences to be as uniform as
possible across languages while, of course, recognising (c) that their phonologi
cal and phonetic shapes, including the linear order of the words in a sentence, are
different. The syntax and semantics are ‘externalised’ with different phonologies
in different languages, but the syntax is (near-)uniform and the semantics entirely
uniform. We will come back to these issues in Chapters 7 and 8.
The Italian sentence in (10a) tells us two more interesting things. First, the
noun pesci is plural: it has the plural ending -i (compare the singular pesce in
(3a)). In the English version, fish is plural too; we have already noted that fish
has no overt plural ending. If we substitute a more regular noun, one which
forms its plural with -s, we can see this:
(12) Boys fish.

Second, the verb pescano consists of the root pesc- and the ending -ano (again
compare the infinitive ending -are in (3b)). This is not the place to go into the
full details of Italian verbal morphology, but this ending contains the infor
mation that the verb is first conjugation, present tense, indicative mood (it
makes a statement of fact) and third-person plural (‘they’). That’s quite a bit
of information in just three overt phonemes. The -a- part of the ending recurs
in the infinitive ending, and this can be seen as the marker of first conjuga
tion. The -no part of the ending shows up in many other tenses and is clearly
marking third-person plural. Present tense and indicative mood are arguably

IanRoberts_9781316519493_C001.indd 18 06-09-2022 11:48:49

1.2 Some Fish 19

just ‘understood’. As we have already seen, saying something is ‘understood’

is a way of saying that it is part of the semantic representation of the sentence,
where there is no pronounced element that represents it. In order to simplify
the syntax–semantics connection, we want the syntactic representation to reflect
the semantic interpretation as fully as possible. Since we have silent syntactic
elements, perhaps we can also have silent morphological elements. So we are
led to think that pescano has the structure in (13):
(13) pesc-a- Indicative - Present -no

Furthermore, we can, and for the sake of consistency must, put labelled brackets
around each part of the structure:
(14) [V [V pesc- ] [Conj -a- ] [M Indicative ] [T Present ] [Agr -no ]]
The labels are fairly straightforward: ‘V’ indicates that both the root and the
whole thing form a verb; ‘Conj’ indicates the conjugation-class marker ‘M’ indi
cates mood (the traditional grammatical category indicating whether a sentence
describes a state of affairs believed to be true or not), ‘T’ indicates tense and
‘Agr’ indicates agreement: the verb is third-person plural because the subject, i
pesci, is third-person plural.
Now, if the Italian verb has a structure like (14), and the Italian sentence in
(10a) is an accurate translation of the English sentence in (7) and our assumption
that English and Italian verbs should be minimally different means that we want
structures to ‘line up’ across languages, our logic leads us to posit something
like (15) as the structure of the verb fish in (7):
(15) [V [V fish- ] [Conj ?? ] [M Indicative ] [T Present ] [Agr 3Pl ]]
Here, Conj, M, T and Agr are all silent. But our logic leads us to conclude that
they are all structurally present. Now we come up against the limits to this lin
ing-up-in-the-name-of-UG approach. It seems reasonable to take the verb fish in
(7) to be indicative in mood, present in tense and third-plural in agreement, but
the conjugation-marking seems to impose an idiosyncracy of Italian morphology
onto English, moreover an idiosyncracy that appears to have no semantic corre
late. So perhaps we should eliminate ‘Conj’ from (15). The other silent endings
are, however, justified by the semantics and our universalising methodology.
Both tense and agreement endings do audibly appear on English verbs (mood
may too, but this is a little trickier and so I’ll leave it aside). If we make the sub
ject of our example in (12) singular, but keep the present tense, we have:
(16) The boy fishes.

Here the verb has the ending -es. If we make the verb past tense, we have:
(17) The boy fished.

Here the verb has the ending -ed. So we could give the verbs in (16) and (17) the
structures in (18a) and (18b) respectively:

IanRoberts_9781316519493_C001.indd 19 06-09-2022 11:48:49

20 1 tacit knowledge

(18) a. [V [V fish- ] [M Indicative ] [T Present ] [Agr -s ]]

b. [V [V fish- ] [M Indicative ] [T -ed ] [Agr 3Sg ]]
Here, we have dropped the Italocentric Conj, for the reasons given above; Mood
remains silent, but T is realised as -ed in the past tense and Agr is realised as -es in
the third-person singular (agreeing with the third-person singular subject the boy)
in the present tense. So there is justification for lining up these aspects of English
verbal inflection with Italian. The fundamental difference between English and
Italian has already been mentioned: in the history of English many markers of
inflection have lost their overt phonological representation, while their Italian
counterparts have retained most of theirs. So English has more silent inflections
than Italian. Semantics and UG together lead us to postulate those inflections
nonetheless.
We can also observe another interesting parallel between English and Italian
brought out in (16). Here, unlike in (12) where boys is plural, an article must
appear with boy. Compare:
(19) *Boy fishes.

The article doesn’t have to be definite. In fact, a range of determiners of various

kinds can appear here:
(20) A/This/That/One/Every/Each/No/Some boy fishes.

The fact that where boys is plural as in (12) no audible determiner is required,
combined with the unique, known, existing ‘natural kind’ interpretation assigned
to boys here, supports the idea that there is a silent determiner. This in turn sup
ports the lining-up of English and Italian seen in (11).
We can now replace (11) with a fuller lined-up, (nearly) uniform English and
Italian structure, in which the determiners, silent and overt, are represented by
the category D:
(21) [S [NP [D the ] [N [N fish-[Num Pl ]]] [V [V fish- ] [M Indicative ] [T Present ]
[Agr 3Pl ]] ]
[S [NP [D I ] [N [N pesc-[Num i ]]] [V [V pesc- ] [Conj -a- ] [M Indicative ] [T Present ]
[Agr -no ]] ]
Here I’ve added the specification of number on the noun (the ending Num);
plural number is silent in the case of fish in English as we have observed. We
could add further specifications: Person on the noun (3rd here), number marking
on the Italian article i (implying, by our now-familiar reasoning, that English
the has a silent plural ending). Furthermore, the Italian articles show gender
marking: article i is masculine. So we should add that specification to the Italian
sentence. Should we add it to the English one? The Italian gender is grammati
cal, not semantic. Pesce is a masculine noun, but not all fish are male (or there
would be no more fish, since they reproduce sexually). In Italian, all nouns have
masculine or feminine gender, and this has no semantic basis at all in many
cases: ‘sun’ is masculine (il sole); ‘moon’ is feminine (la luna), ‘sincerity’ is

IanRoberts_9781316519493_C001.indd 20 06-09-2022 11:48:49

1.2 Some Fish 21

feminine, ‘feminism’ is masculine, ‘violence’ is feminine, ‘socialism’ is mascu

line and so on. Try as one might, it is hard to find the slightest metaphorical or
cultural basis for this: the words are just arbitrarily assigned to two distinct mor
phological classes. Hence gender-marking has no consequences for the seman
tic representation. So perhaps, as with the Conj-ending on verbs, it would be
Italocentric of us to impose Italian genders on English nouns or articles.
One last point on the structure in (21): are M, T and Agr part of the verb V,
as indicated there, or, analogously to D, part of the Verb Phrase, VP? We could
think that V and N each have their ‘satellite categories’ which carry grammatical
information often realised as inflection or as ‘little words’ like the articles the
and i: the satellites of N are D and Num, and the satellites of V are M, T and
Agr. Semantically, though, Mood and Tense arguably modify the sentence, by
indicating whether it makes a true statement and relating the time of the situation
described to the speech time (this is what Tense does). So M and T look like
they belong to S. Agr seems to connect the subject NP and the verb; as such, this
too looks like a sentence-level, rather than a verb-level, property. We leave the
question of the status of M, T and Agr open for now; later we will see that M
and T are indeed parts of S, while Agr may be a relation rather than a category.
Let us now go back to the structure in (9), which I repeat here for convenience:
(9) [S YouN [V fish ] [N fish ]]

We saw the Italian translation of (9) in (10b):

(10b) Pesca pesci!

The full representations of both examples, with silent endings indicated, are
given in (22):
(22) a. [S [NP you ] [VP [V [V fish- ] [M Imperative ] [T Present ] [Agr 2 ]] [[D some ]
[N [N fish-[Num Pl ] ]
b. [S [NP tu ] [V [V pesc- ] [Conj -a- ] [M Imperative ] [T Present ] [Agr 2Sg ]] ]
[NP [D i ] [N [N pesc-[Num i ]]]

Both examples have an understood second-person subject. English you can be

either singular or plural, and so (8/9) can be interpreted as giving an order to a
single person or to a group of people. The Italian subject is singular, as the ver
bal ending shows (the plural is pescate), and the understood subject is the singu
lar pronoun tu. M is specified as Imperative: these sentences do not make factual
statements but give orders. Imperative is the ‘ordering Mood’, while, as we saw,
Indicative is the ‘stating Mood’. The object NPs are similar to the subject NPs
in (21), except for the Ds. The Italian example does not have a definite article
and the English silent D is ‘some’. Both of these points are related to the fact
that the object here has a subtly different interpretation from the subject in (21).
In (21), the subject names a unique, known, existing ‘natural kind’; this inter
pretation is known as the generic reading of the NP. The object does not denote
a natural kind; it merely denotes some arbitrary, not necessarily known and not

IanRoberts_9781316519493_C001.indd 21 06-09-2022 11:48:49

22 1 tacit knowledge

necessarily existing group of fish (the fishing expedition could be completely

unsuccessful with no fish at all actually being caught). This is an indefinite
interpretation of the NP. Hence the definite D, silent in English, overt in Italian,
does not appear. I have glossed the silent indefinite D as ‘some’ in English.
From the above, we can see that our simple two-fish sentence in (6) has a
great deal more to it than meets the eye, or ear. The Italian translations are very
revealing, especially when we try to line them up as closely as we can to their
English counterparts. We see that both verbs and nouns have satellite cate
gories which may be realised as inflectional endings or as articles such as the
and i, and are frequently silent. We also see that not all satellite categories are
universal: Italian has grammatical gender on nouns and Conj-marking on verbs,
but there is no reason to posit counterparts to these in English. Of course, a very
natural, and, it turns out, very difficult question is whether the other categories
such as D, M, T and Agr are universal. We are in no position to answer this
question at the moment, but we can observe that it seems reasonable to take
these categories to be shared by Indo-European languages such as Italian and
English.
So we see that simple sentences like English (6) and Italian (10) can tell us
a lot. In the next section, we’ll take this line of thought a bit further, with some
more complicated sentences (but still using the same English word).

1.3 More Fish and Some Relative Clauses

In this section, we’ll look at three-, four- and five-fish sentences.
These sentences are simultaneously silly and serious; they’re silly because they
describe rather unlikely situations (which you can nonetheless imagine), and
they’re serious because they can reveal a lot about our linguistic competence.
So let’s look at what happens when we combine three occurrences of fish:
(23) Fish fish fish.

The Italian translation of (23) is (24):

(24) I pesci pescano pesci.

Both of these are full sentences where the first ‘fish’ is the subject, the second the
verb and the third the direct object. Clearly, what is added to the two-fish exam
ples discussed above is an explicit direct object, compared to (7), and an explicit
subject, compared to (9). Traditional grammars tell us that a complete sentence
consists of a subject, what the sentence is about, and a predicate, which says
something about the subject. The predicate contains the main verb and the direct
object. Thus we could represent (23) as follows (leaving aside for the moment
the representation of the silent inflections):
(25) [S [N Fish ] [Predicate [V fish ] [N fish ]]]

IanRoberts_9781316519493_C001.indd 22 06-09-2022 11:48:49

1.3 More Fish and Some Relative Clauses 23
Since the verb is the most important part of the predicate, we can give a more
strictly categorial representation of (25) as follows:
(26) [S [N Fish ] [VP [V fish ] [N fish ]]]
Here VP, the Verb Phrase, is the category which functions as the predicate.2
The ‘full’ representations for (23) and (24), complete with specification of
silent inflections and other satellite categories, are shown in (27):
(27) a. [S [NP [D the ] [N [N fish-[Num Pl ]]] [VP [V [V fish - ] [M Indicative ] [T Present ]
[Agr 3Pl ]] [NP [D some ] [N [N fish-[Num Pl ]]]]
b. [S [NP [D I ] [N [N pesc-[Num i ]]] [V [V pesc- ] [Conj -a- ] [M Indicative ] [T Present ]
[Agr -no ]] [NP [D i ] [N [N pesc-[ Num i ]]]]
These representations are almost exactly the expected combination of (21) with
an explicit direct object as in (22). Again, the subject and object have differ
ent structures and interpretations, with the subject being generic and the object
indefinite.
Now let’s try four fish:
(28) Fish fish fish fish.

This sentence is both grammatical and interpretable, but, unlike the two-fish and
three-fish examples it requires a little thought. The intonation with which it is to
be read is important: it should be ‘ish FISH fish [short pause] FISH’, with stress
on the second fish and the strongest stress on the final fish. The structure here
is significantly more complex than our earlier examples. For this reason, from
now on I will mainly give simplified representations of these more complex
examples, leaving aside the silent inflectional material (although we should not
forget that it is there).
What does (28) mean and what is its structure? Again, the Italian translation
can give us an important clue:
(29) I pesci che i pesci pescano pescano.

We have seen that pesci is unambiguously a noun and pescano is unambigu

ously a verb. At the absolute minimum, then, (29) has the structure N che N V
V. The English example in (28) lines up with this, except that there is no coun
terpart to che. The Italian word che usually translates as that. We can add that
to (28), to get:
(30) Fish that fish fish fish.

The markers that and che introduce a relative clause, a clause which modifies a
noun; for our purposes, the only real difference between that and che is that che
appears obligatorily in this context, while that can be ‘dropped’ (i.e. silent). So

2
We will look in detail at the relation between grammatical categories like NP and VP and gram
matical functions like subject and predicate in Chapter 2 of Volume II.

IanRoberts_9781316519493_C001.indd 23 06-09-2022 11:48:49

24 1 tacit knowledge

the second two fish, the sequence noun-verb, constitute a relative clause mod
ifying the first noun fish (known as the head of the relative). The fourth fish is
the verb of the main clause, with an implicit direct object just as in the two-fish
example with the structure in (7). We can thus make a first pass at a structure for
(28) as follows:
(31) [S [N fish ] [Relative clause fish fish ] [VP [V fish ]]]
The relative clause is a kind of subordinate clause. In fact, we can tell that the first
fish inside the relative clause is the subject and the second fish is the verb of the
predicate. So we have an NP-VP structure here (remember that the presence of the
overt definite article in Italian shows us that the subject is an NP, not just a noun):
(32) [S [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ]]] [VP [V fish ]]]
The main clause, i.e. the independent clause containing the relative clause, indi
cated as S here, must also have a subject NP, not just N. This NP includes the
relative clause, since the relative clause modifies the noun, so we have (33):
(33) [S [NP [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ]]]] [VP [V fish ]]]
Both the main clause and the relative have a subject and a predicate. Identifying
predicates with VP again, (33) tells us that each VP contains just the verb fish. In
the case of the main clause, this may be correct, since the object here is implicit
and has a rather vague interpretation (roughly ‘stuff’). But the object of the rel
ative has a completely different, and very precise, interpretation: ‘fish’. Here
we see an important characteristic of relative clauses: there is always something
missing, often referred to as a ‘gap’, which corresponds semantically to the head
of the relative (which we could then call the ‘filler’). Since we know that syntac
tic representations contain silent elements (actually, we are beginning to see that
they mostly contain silent elements) and we want to keep the syntax-semantics
mapping as straightforward as possible, the best way to account for the interpreta
tion of the gap is that it is a silent copy of the filler. So we replace (33) with (34):
(34) [S [NP [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N fish ]]]]] [VP [V fish ]]]
We said above that English that, unlike its Italian counterpart che, can be
‘dropped’. This implies that sentences with and without that are synonymous: that
and che are just meaningless syntactic markers (saying ‘what comes next is a rel
ative clause’ in our examples). However, (30) has another interpretation that (28)
doesn’t have. In (28) the gap inside the relative clause is the direct object, as (34)
shows; it means ‘fish that get fished fish stuff’. But (30) allows a further interpre
tation, where the gap can be interpreted as the subject of the relative clause: ‘fish
that fish other fish fish stuff’. This example has the following structure:
(35) [S [NP [N fish ] that [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N fish ]]]]] [VP [V fish ]]]
Here it is the subject of the relative that is the silent copy of the head, and
the marker that is obligatory. The Italian translation of this English sentence is
unambiguous and should be compared to the one in (29):
 IanRoberts_9781316519493_C001.indd 24 06-09-2022 11:48:49
1.4 Really Challenging Fish 25

(36) I pesci che pescano pesci pescano.

It is now easy to interpret a five-fish example:

(37) Fish fish fish fish fish.

As we saw, the four-fish example in (28) had an implicit direct object in the main
clause (‘Fish fish fish fish stuff’). The five-fish example in (37) has an explicit
direct object in the main clause. We can therefore give the structure as in (38),
and the Italian translation in (39):
(38) [S [NP [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N fish ]]]]] [VP [V fish ]
[NP [N fish]]]]

(39) I pesci che i pesci pescano pescano pesci.

By now we may begin to wonder just how many repetitions of the word fish (in
the phonology, the subpart of the syntactic structure that is actually pronounced)
can give rise to a grammatical, interpretable sentence. It is clear that the step
from three fish to four fish, involving as it did the introduction of a relative
clause into the structure, places a burden on comprehension. But this burden can
be readily overcome and the sentence is fully comprehensible; in fact we can
recognise that it describes a rather odd state of affairs, the very silliness of the
sentence proves that we can understand it. We see, then, that the comprehension
burden does not increase linearly as we add extra words; the structural complex
ity of the four-fish sentence is significantly greater than the three-fish one, and
this is reflected in our initial hesitation in understanding it.
In the next section, we’ll see some really challenging examples as we increase
the number of fish still further.

1.4 Really Challenging Fish

Here we’re going to look at six- and seven-fish sentences. These are
really hard to understand, especially the six-fish one (it’s harder than seven, as
we’ll see). You don’t have to follow every detail of the analysis of these sen
tences here, as long as you get the main point, which comes after (54). Skip ahead
to that discussion if the fish tire you out. Or, fish-like, just go with the flow.
The non-linear increase in comprehension difficulty becomes very sharp
indeed when we move from five fish to six fish. Here is the six-fish example:
(40) Fish fish fish fish fish fish.

Most people are completely stumped by this sentence at first sight; it appears to
be simply a random repetition of the word fish with no meaning (and so, perhaps,
no structure) at all. But in fact this isn’t true, as we’ll see below. What is striking
though is that seven fish is easier to interpret than six:
(41) Fish fish fish fish fish fish fish.

IanRoberts_9781316519493_C001.indd 25 06-09-2022 11:48:49

26 1 tacit knowledge

We know from (28) that a sequence of three fish can be a relative clause with
an object gap, ‘fish that are fished by other fish’. In (28), the relative clause is in
the subject position. But relative clauses can be in object position too (in fact,
(28) can be interpreted that way too, a point that was left aside above). So (41)
has the approximate structure in (42a), given in fuller detail in (42b) (where RC
stands for relative clause, and the structure is presented as a tree diagram, as this
makes it easier to see the relations among the fish; we’ll look at tree diagrams
more systematically in the next chapter):
(42) a. [ [ Fish fish fish ] [ fish [ fish fish fish ]] ]

b.
Pronounced with the right intonation (‘Fish FISH fish [pause] fish fish FISH
fish’), this example is fairly easy to understand, especially in the light of the
four- and five-fish examples. But none of this makes (40), with just six fish, any
easier to understand.
In order to see what the structure, and hence the interpretation of (40) is, we
must go back to our earlier, simpler examples. Here, once again, is the two-fish
example with noun-verb interpretation, repeated from (7):
(7) [S [N fish ] [V fish ]]

Of course, we now know that N and V here are contained in their respective
NPs and VPs. Relative clauses are really a kind of sentence, as we pointed out
above. So let us label them as S from now on. We go from two fish to four fish
by putting the structure in (7) inside the subject NP:
(43) [S [NP [N fish ] [S [N fish ] [V fish ]] ] [VP [V fish ]]]

Actually it’s not quite accurate to say we put (7) inside the subject NP; really, we
put the three-fish sentence, with the structure seen in (26), inside the subject NP
 and make the object a silent version of the head of the relative:

IanRoberts_9781316519493_C001.indd 26 06-09-2022 11:48:50

1.4 Really Challenging Fish 27

(44) [S [NP [N fish ] [S [NP [N fish ]] [VP [V fish ] [NP [N fish ]]]]] [VP [V fish ]]]

On the basis of (44), we can add two more pronounced fish by inserting a relative
clause inside the subject NP of the relative clause, i.e. after the second [N fish ].
This gives the string of six fish in (41), and shows us what the structure is:
(45) [S [NP [N fish ] [S [NP [N fish [S [NP [N fish ] [VP [V fish ] [NP [N fish ]]]]]]] [VP [V fish
] [NP [N fish ]]]]] [VP [V fish ]]]
If we add some relative markers, preferably which (another relative marker in
English alongside that), we can just about start to make sense of (45):
(46) Fish which fish which fish fish fish fish.

So (41) is, despite initial appearances, a grammatical and interpretable sentence:

the string is N N N V V V. It might be easier to see this if we use some different,
and more varied, vocabulary. Let’s start with (47):
(47) The mouse the cat chased died.

By now, we know a relative clause when we see one. The essential structure of
(47) has a relative clause inside the subject containing an object gap:

(48) [S [NP the mouse [S [NP the cat [VP chased [ the mouse ]]]]] [VP died ]]

(48) is quite easy to understand. The analogue to (41), but with different words, is (49):

(49) [S [NP the mouse [S [NP the cat [S [NP the dog [VP bit [NP the cat ]]]]]]] [VP chased
[NP the mouse ]]]]] [VP died ]]

Again, the jump from (48) to (49), caused by simply adding another relative
clause modifying the cat, poses severe difficulties of comprehension. Adding
some relative markers (that again) makes (49) considerably easier to understand:
(50) The mouse that the cat that the dog bit chased died.

This phenomenon is known as centre-embedding; it seems that embedding

more than one relative clause inside another without an overt relative marker
creates real comprehension difficulties. Nonetheless, such sentences are in fact
grammatical and interpretable. The rules of English syntax allow us to embed
relatives inside one another and, under the right conditions, to ‘drop’ relative
markers. The fact that embedding more than one relative clause inside another
causes comprehension difficulties is a matter of performance; our short-term
memory seems unable to ‘keep track of’ the structure. But since these sentences
are generated by the rules of English syntax they are grammatical; so we desig
nate them as grammatical, but, because of the comprehension load they impose,
unacceptable. Grammaticality is a competence notion, while acceptability is a
performance notion. Most of the time, grammaticality and acceptability coin
cide, but in cases like centre-embedding we can see the distinction. Another
famous example where the two notions come apart is the sentence Colourless

IanRoberts_9781316519493_C001.indd 27 06-09-2022 11:48:50

28 1 tacit knowledge

green ideas sleep furiously; most people find this sentence unacceptable for the
perfectly good reason that it doesn’t make any sense. It is, however, syntac
tically well-formed, as its exact formal counterpart Revolutionary new ideas
spread quickly shows. Both sentences are grammatical in that they conform to
the rules of English syntax; the first one is semantically anomalous (in fact, it
is self-contradictory) and so unacceptable. This contrast also shows that syntax
and semantics are distinct in that a sentence can be syntactically well-formed but
semantically ill-formed; we will see more examples of this type in Section 8.3.
Of course, examples with centre-embedding like (50) become still more
difficult to understand when the same phonological word occurs six times over,
as in (40). There is, in principle, no difference in grammaticality between (40)
and (50) but a real difference in acceptability (they just go from bad to worse).
But now we can line (40) up with the variant of (50) without the relative mark
ers, and we can see how (40) works, and see that it is in fact grammatical:
(40) Fish fish fish fish fish fish.

(51) The mouse the cat the dog bit chased died.

Finally, here are the easier versions complete with the relative pronouns which
and that inserted in the appropriate places:

(46) Fish which fish which fish fish fish fish.

(50) The mouse that the cat that the dog bit chased died.

These examples have the structures shown in (49) and (45) respectively.
Relative clauses can be put one inside another without limit. They can just go
on and on. Consider (52):
(52) This is [NP1 the guy [ S1 who loved [NP2 the girl [S2 who befriended [NP3 the boy
[S3 who lived ]]]]]].
For those familiar with the Harry Potter stories, each NP here describes one
of the central protagonists by means of modification by a relative clause, each
relative clause is inside the NP it modifies, and each NP – except NP1 – is
embedded in the next relative clause. NP3 describes Harry Potter himself, the
boy who lived. NP2 describes Hermione Granger, the girl who befriended the
boy who lived, and NP1 describes Ron Weasley, the guy who loved the girl who
befriended the boy who lived. Relative clauses give language a great deal of its
expressive power, by making it possible to modify nouns using modifiers that
can be as complicated as we want: in principle, there is no upper limit to the
number of times one relative clause can be embedded inside another. This is
reflected in nursery rhymes like ‘The House that Jack Built’: the dog who bit
the cat who ate the rat who lived in the house that Jack built. Centre-embedded
relatives can also, in principle, be constructed without limit. The problem is
that, in practice, as we have seen, they quickly become very hard to understand.

IanRoberts_9781316519493_C001.indd 28 06-09-2022 11:48:50

1.4 Really Challenging Fish 29

Compare the following centre-embedded sequence with the excerpt from ‘The
House that Jack Built’ just given:
(53) [ The rat [ the cat [ the dog [ bit ]] [ ate ]] [ lived in the house that Jack built ]

All of this leads us to three questions:

(54) a. What is the highest number of repetitions of the word fish alone that
constitutes a grammatical English sentence?
b. Have you seen these sentences before?
c. Where did you learn these facts about English?

To answer (54a), we could keep on trying longer and longer fish sentences.
There is certainly no reason to take seven, the most we have seen here (see (41)
and the structures in (42)), to be an upper limit. If we look at the examples with
even numbers of fish, (28) and (40), we see that they have the general form in
(55):
(55) a. N1 + N2 + V2 + V1 (four fish, (28))
b. N1 + N2 + N3 + V3 + V2 + V1 (six fish, (40))
(This is not their exact structure, as we have seen, but these simplified representa
tions make our point here.) Each pair Nn + Vn, for n = n forms a relative clause
modifying Nn-1. Hence, as Freidin (2012:13) points out ‘any sentence containing
an even number of fish from four onwards will have at least one grammatical
representation’. Furthermore, although each relative clause (i.e. each pair Nn +
Vn in (55) for n > 1) has an object gap, as we saw above (see (44) and (45)), V1
does not have an object. But of course it could have, and this object could be fish.
This is what got us from four fish to five fish and from six fish to seven fish. To
quote Freidin (2012:13) again: ‘As this procedure [adding an object to V1, IR]
can be applied to any sentence with an even number (n) of fish, there will be a
corresponding sentence with an odd number (n+1) of fish … Hence any number
of fish will correspond to a sentence of English.’ In principle, then, a sentence
containing any number of repetitions of the word fish is grammatical. Our lin
guistic competence allows us to understand such sentences, although more than
five iterations, with the exception of seven, give rise to sentences that are difficult
to understand in practice, i.e. sentences which may be unacceptable to varying
degrees. We are able to assign syntactic representations, more technically known
as structural descriptions, to the sentences, and from there recover the meaning.
The full representations of these sentences, as we saw in our discussion of the
Italian counterparts of the two- and three-fish examples, are quite complex. The
full representation of the seven-fish example in English would be (56):
(56) [S [NP [D the ] [N [N fish-[Num Pl ]]] [S [NP [D the ] [N [N fish-[Num Pl ]]] [VP [V [V
fish- ] [M Indicative ] [T Present ] [Agr 3Pl ]] [NP [D the ] [N [N fish-[Num Pl ]]]]]]
[VP [V [V fish- ] [M Indicative ] [T Present ] [Agr 3Pl ]] [NP [D some] [N [N fish-
[Num Pl ]]] [S [NP [D the ] [N [N fish-[Num Pl ]]] [VP [V [V fish- ] [M Indicative ]
[T Present ] [Agr 3Pl ]] [NP [D the ] [N [N fish-[Num Pl ]]]]]]

IanRoberts_9781316519493_C001.indd 29 06-09-2022 11:48:50

30 1 tacit knowledge

You don’t need to pick your way through all the details of this representation in
order to see two things. First, most of the structure is silent. Second, sentences
of this kind are highly complex objects. But remember, and this is the really
important thing, we quite unconsciously compute such complex sentence struc
tures all the time.
Concerning question (54b), it is highly unlikely that anyone would come
across sentences of this kind (outside of a linguistics class). But, as we have
seen, they can be fairly readily understood. It is extremely difficult to see how
any account of linguistic knowledge based on simply imitating, learning and
generalising routines (perhaps through processes of stimulus, response and rein
forcement, as in behaviourist psychology) could account for our general capac
ity to produce and understand, mostly instantaneously, sentences we have never
heard before and which may never have been uttered before.
Question (54c) is the question that goes to the heart of the matter. As native
speakers of English, we have very rich tacit knowledge, of even the silliest and
most awkward sentences, which we were almost certainly never ‘taught’ in any
meaningful sense of the word. Our experience in early childhood, combined with
our cognitive capacities (both language-specific and not), has made it possible
for us to perform the kinds of mental computations required to extract structure
and meaning from what appear to be merely strings of repetitions of the same
word. Of course, this ability is not restricted to native speakers of English: native
speakers of Italian can just as easily make sense of sentences like (24), (29), (36)
and (39), and the same exercise can be repeated, in principle, with any speaker
of any language.
The fish sentences clearly illustrate the distinction between competence
and performance. As we have mentioned, competence refers to a state of
knowledge, knowledge of an I-language, that a native speaker of a given lan
guage (in roughly the ‘E’-sense as discussed in the Introduction) possesses.
Performance refers to the application of I-language in a given situation: com
bined with other cognitive and social capacities, performance makes normal
speaking and comprehension possible. Freidin’s conclusion that ‘any number
of fish will correspond to a sentence of English’ clearly concerns competence;
the rules of English syntax are such that sentences of this kind can exist, but
of course, owing to performance limitations, no actual speaker could produce
or understand a sentence consisting of an infinite number of occurrences of
the word fish and nothing else. Since placing an upper bound on the number
of times one relative clause could be embedded in another would be arbitrary,
we regard the rules of English (and other languages) as allowing for this in
principle. Of course, the same applies to numbers: no individual can ever write
out an infinite number, although our mathematical competence tells us, and
this can of course be proved, that the series of natural numbers is infinite. In
Chapters 3 and 4, we will see the precise structure-building mechanisms that
achieve this result.

IanRoberts_9781316519493_C001.indd 30 06-09-2022 11:48:50

1.5 Conclusion: What the Fish Have Taught Us 31

1.5 Conclusion: What the Fish Have Taught Us

In this chapter we have tried to outline and illustrate, with the con
crete help of the fish sentences, what linguistic theory investigates. The central
goal is to elucidate what it means to be a competent speaker of a language. In
other words, linguistic theory is primarily about knowledge: what is it that a
person described as a native speaker of English (for example) knows? What
are the properties of individual I-languages and what are the properties of UG
that underlie them? Clearly this must involve the operations that are capable
of producing the kinds of structures we saw in relation to the fish examples. A
further central question is where this knowledge comes from: how do children
acquire their first language? Related to this is the question of how similar and
different languages are. We saw that we can ‘line up’ English and Italian quite
well, especially if we assume that English has a number of silent inflectional
‘satellite’ categories many of whose Italian counterparts are overtly pronounced.
In language-acquisition terms, this suggests that children are looking for evi
dence – semantic, syntactic, morphological or phonological – for the presence
of these categories. Many complex issues arise as we try to pursue these ideas,
but we saw in connection with the Italian fish examples that the combination
of semantic evidence and UG predispositions of various kinds can lead us to
significantly limit the differences between English and Italian. As we said there:
semantics is universal, syntax is near-universal and languages differ primarily in
their morphophonologies. We will return to these ideas repeatedly in the chap
ters to follow.
This is a formal, cognitive theory of linguistic knowledge. Each individual
over the age of five or so has a specific I-language which makes possible their
competence in their native language. This I-language corresponds more or less to
a sociocultural E-language construct such as English, Italian and so on. We also
saw that English and Italian speakers are I-alike but E-different: the I-languages
may turn out to be quite similar to each other, while many aspects of the cultures
borne by the corresponding E-languages are strikingly different (in ways that
people generally find fascinating). Given how little we understand about the
brain, as well as ethical restrictions on what kinds of experiments we can do on
healthy children or adults, we can only find out about what is inside the mind,
how an individual’s grammatical system works, from the observed outputs of
the I-language. This includes, quite simply, the things we say and hear (language
production and comprehension) along with judgements about grammaticality,
interpretation and so on.
We saw in the Introduction that there are three factors that make an I-language.
These are, first, the genetic endowment. Here we take this to be UG, containing
the various kinds of rules and systems of rules that build and interpret syntactic,
semantic and phonological representations. The details of what constitutes UG
will occupy us for much of what follows. The second factor is experience, in

IanRoberts_9781316519493_C001.indd 31 06-09-2022 11:48:50

32 1 tacit knowledge

particular linguistic experience: exposure to a language as a small child. This

happens to almost everyone, and the nature of the early linguistic experience
will determine which E-language group an individual belongs to, in part deter
mined by the I-language which develops from the interaction of the linguistic
experience, UG and the third factors. The third factors cover a range of different
things, but among the most important are general cognitive abilities, not specific
to language. As we mentioned in the Introduction, it can be very difficult in cer
tain cases to distinguish general cognitive abilities from language-specific abili
ties (which we would attribute to UG unless they clearly stem from experience).
All the above remarks apply to linguistic theory, conceived as here, in gen
eral: to semantics, syntax, morphology and phonology. The goals of syntactic
theory in particular are above all to investigate the nature of the adult’s tacit
competence in the native language, i.e. to investigate I-language. This is done
by discovering and describing grammatical and ungrammatical structures
(which, as we saw, mostly but not always correspond to acceptable and unac
ceptable sentences according to native-speaker judgements); these observa
tions are systematized as a model for that knowledge. In the chapters to follow,
we will see how this is done. This model is then tested and adapted against
further language data; if we are looking at UG, we may look at data from a
range of languages. The rule systems are grounded in the three factors dis
cussed above. So three really fundamental, and in this approach, intertwined
questions are:
(57) a. how does a given I-language relate to UG?
b. how does adult I-language result from acquisition?
c. how is I-language related to general cognition and the third factors?

Each of these questions relates I-language to one of the three factors in language
design that underlies it. In this connection, our fish sentences are very important.
They show us that syntactic structures are potentially infinite and yet built out
of very simple elements. In particular, the structures are formed by repeating
the same operations again and again. The next few chapters, as well as much of
Volume II, are devoted to demonstrating this in full detail. The fish have given
us a first inkling of the nature of the syntactic component of I-language, but now
it is time to look more systematically into the details.

Exercises
1. Consider again the four-fish example in (28), repeated here as (i):
(i) Fish fish fish fish.
We assigned the structure in (34), here (ii), to this sentence:
(ii) [S [NP [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N fish
]]]]] [VP [V fish ]]].

IanRoberts_9781316519493_C001.indd 32 06-09-2022 11:48:50

1.5 For Further Study and/or Discussion 33

We also observed that inserting the relative marker that between the
first two fish made possible a different interpretation, which gave in
(35) (= (iii)):
(iii) [S [NP [N fish ] that [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N
fish ]]]]] [VP [V fish ]]].
In all of (i–iii) there is an implicit main-clause object, meaning
something vague like ‘stuff’. It is, however, possible to interpret (i)
as having an overt object. Try to see that interpretation, indicate its
structure (in a rough way, along the lines of (ii) and (iii)), its interpre
tation and, if you can, its approximate intonation. (HINT: look again
at the ambiguity of the two-fish example discussed in (7–9)).
2. Now look again at the five-fish example from (37) (= (i)):
(i) Fish fish fish fish fish.
We said that (i) is the four-fish example plus an overt direct object
for the main-clause verb, giving the structure in (38) (= (ii)):
(ii) [S [NP [N fish ] [Relative clause [NP [N fish ]] [VP [V fish ] [NP [N fish
]]]]] [VP [V fish ] [NP [N fish]]]]
But in fact (i) has a further interpretation, where the main-clause
object is a relative clause. Give the structure for this interpretation
of (i).
Furthermore, the relative clauses in both interpretations can be
marked with that, and in each case this gives rise to a further ambi
guity inside the relative clause. Explain the ambiguity and give the
relevant structures.
3. Give the structure for the six-fish example in (40) (= (i)):
(i) Fish fish fish fish fish fish.
Now try to add a further level of embedding to (40), using any lexical
 items you like (except perhaps fish). This will correspond to an eight
fish sentence; give the structure for this one.

For Further Study and/or Discussion

1. If you speak a language other than English or Italian, translate the fish
sentences into your language as best you can; if you don’t speak any
other language, ask a friend who does (preferably a patient friend).
Try to assign structures along the lines of (27) to your sentences.
What difficulties arise in the translations? How similar to or different
from English or Italian are your translations? Which of the satellite
categories we identified do you need? Do you need other ones?

IanRoberts_9781316519493_C001.indd 33 06-09-2022 11:48:50

34 1 tacit knowledge

2. Google Translate gives (i) as the Italian translation of the three-fish

English sentence:
(i) Pesce pesce pesce.
This is not a grammatical sentence but rather obviously just the
Noun pesce repeated three times. As we saw in (24), the correct Ital
ian translation of the three-fish English sentence is (ii):
(ii) I pesci pescano pesci.
What does an Italian speaker have that Google Translate lacks?

Further Reading
R. Freidin. 2012. Syntax: Basic Concepts and Applications. Cambridge: Cambridge
University Press.
 IanRoberts_9781316519493_C001.indd 34 06-09-2022 11:48:50

2 Constituents and Categories

2.1 Introduction and Recap

In the last chapter we saw, using the fish sentences, that syntactic
structures are potentially infinite and are built out of very simple elements by
repeating the same operations on structure again and again. We also defined the
goals of linguistic theory, construed as a formal, cognitive theory, as being pri
marily to discover what it is that people know when they know a language, i.e.
what is I-language competence? A further central question is how this knowl
edge is acquired.
Our focus henceforth will be on elucidating the nature of I-language in rela
tion to syntax. As we have seen, syntax is a central aspect of language, in that it
relates sound and meaning over an infinite domain, so clearly any theory of
syntax will be a central part of UG. We continue to take UG to be universal
innate knowledge, one of the three factors making up the language faculty. We
will therefore develop the theory of I-language syntax, taking this, like the rest
of the language faculty, to arise from the three factors in language design: UG,
exposure to language in early life and domain-general third factors (as
described in the Introduction).
The fundamental notion in syntax is constituent structure, i.e. the way in
which words group together into intermediate units (or phrases) of various cat
egories, ultimately forming whole sentences. Some groupings of words, which
we hear as different ordered strings of words, are ill-formed, or ungrammatical;
others are well-formed, or grammatical. We saw in Chapter 1 that grammatical
ity and acceptability generally coincide: almost all of the examples we consider
from now on are of this kind. So when I use the terms ‘grammatical/ungram
matical’ (or ‘well-formed/ill-formed’), these will refer to examples which are
correspondingly acceptable or unacceptable, i.e. native-speaker judgements
consistently correspond to what is generated, or not, by the rules of grammar.
Consider, for example, the following fairly simple five-word English sentence:
(1) Goona hopes that Vurdy will be ready for his dinner.

There are ten! = 3,628,800 possible orders for this ten-word sentence, the vast
majority of which are ungrammatical, such as the following:
(2) a. *Hopes Goona that Vurdy will be ready for his dinner.
b. *Hopes that Goona Vurdy will be ready for his dinner.

35

IanRoberts_9781316519493_C002.indd 35 03-09-2022 06:53:57

36 2 constituents and categories

Our theory must make the distinction between grammatical and ungrammat
ical strings and structures; it must tell us how and why (1) is grammatical
while the examples in (2) are ungrammatical. Furthermore it should tell us
when two sentences consisting of different words (and therefore meaning
different things) have the same structure. This is true for (1) and the exam
ples in (3):
(3) a. Ron said that Hermione would be happy with her homework.
b. Boris thinks that Emile must be angry with his behaviour.
c. Alex believes that Eric should be worthy of a medal.

The basic notions we use to do this are those of categories and constituents,
which we now look at in turn.

2.2 Categories
 2.2.1 Introduction: Lexical and Functional Categories
Like many approaches to morphology and syntax, including tradi
tional grammar, we make a first distinction between two main types of catego
ries, which we will call lexical categories and functional categories.
The principal lexical categories correspond to some of the ‘parts of speech’
of traditional grammar: they are noun, verb, adjective, adverb and preposition,
abbreviated standardly as N, V, Adj, Adv and P (we saw nouns and verbs in
our fish examples in Chapter 1). These lexical categories are open: people can
and do invent nouns and verbs all the time, and even the most up-to-date online
dictionaries struggle to keep up. Words belonging to these categories have clear
non-linguistic semantic content: as we saw, fish, as a noun, denotes a certain
class of aquatic animals, while fish, as a verb, denotes the activity of hunting
these animals. New adjectives and adverbs are less readily coined, but they do
arise, e.g. downloadable, googlable, etc. Prepositions are the one lexical class it
is difficult to add to; in this and certain other respects, prepositions are more sim
ilar to functional categories. We will return below to the possibility of defining
syntactic categories semantically.
Furthermore, lexical categories do not vary greatly across languages. The dis
tinction between nouns and verbs, in particular, seems to recur in all known
languages (although there is some debate about this in certain cases). Adjectives
are not found in every language; some indigenous languages of North America
may lack them, for example. In other languages, e.g. the West African languages
Hausa and Igbo, they form a small, closed class. In languages where the class
of adjectives is small or non-existent, the semantic work of adjectives, roughly
describing qualities, is usually done by relative clauses (a bus which reds). It
is probable that prepositions are not universal: the World Atlas of Language
Structures (WALS) lists thirty languages lacking adpositions (a cover term
for pre- and postpositions), including several indigenous languages of North

IanRoberts_9781316519493_C002.indd 36 03-09-2022 06:53:57

2.2 Categories 37

America such as Blackfoot, Cree and Kiowa.1 English is comparatively rich

in prepositions (although they do not really constitute an open class, unlike
the other lexical categories; to some degree, prepositions straddle the border
between lexical and functional categories).
Functional categories include the ‘satellite words’ we introduced in the discus
sion of the Italian fish sentences in Chapter 1; some of these, e.g. determiner (or
article), correspond to traditional parts of speech, but many of the others would be
treated in traditional grammar as parts of words rather than as syntactic categories
(a consequence of the fact that the traditional parts of speech were designed for
the description of Greek and Latin, both highly inflected languages). There we
saw D, Num, Mood, Tense and Agr(eement). Functional categories tend to con
stitute small classes of, typically, rather small words; they may also be realised
as inflectional endings or, indeed, have no overt realisation at all, as we saw for
several cases in English in Chapter 1. English auxiliaries (e.g. must, be, do, have,
etc.) represent several functional categories, notably Mood and Tense. English
is particularly rich in auxiliaries and they play a central role in many aspects of
English syntax, as we will see. Further functional categories of English include
determiners (D: the, a, this, my, etc.) and Number (represented by the plural
inflection -s). There are other functional categories, e.g. complementisers, or C,
the words which introduce subordinate clauses of various kinds: if, that, for, etc.
(these are known in traditional grammar as subordinating conjunctions).
Functional categories are closed classes: it is seemingly impossible to con
sciously invent new ones. Furthermore, their semantic content is either highly
abstract, as in the case of Mood, Tense and determiners, or perhaps non-ex
istent: for example, what is the meaning of the complementiser that in (1)? It
seems that it can be left out without affecting the meaning of the sentence at
all. Furthermore, functional categories appear to vary quite a lot from language
to language. Here we have given examples from English. Even in languages
fairly closely related to English, the auxiliary system is less rich, and various
semantic notions indicated by auxiliaries in English (e.g. future tense will) may
be indicated by inflections instead, as in several of the Romance languages for
example. In Chapter 1 we saw that Italian is much more richly inflected than
English: Mood, Tense and Agreement (M, T and Agr) are mostly realised by
verbal inflections in Italian and the other Romance languages but often not real
ised at all or realised by auxiliaries in English (although English does have a lit
tle bit of tense and agreement inflection, as we saw). In a radically non-inflecting
language such as Chinese or Thai, Mood and Tense are indicated by invariant
(non-inflecting) ‘sentence particles’ and agreement is entirely absent. One of
the major issues in developing the theory of UG is understanding the extent to
which functional categories can vary, both in their realisation (as inflections,
separate words, or silence) and in their existence (recall our discussion of Conj
marking in English and Italian in Chapter 1). We will gradually develop this
1
WALS Map/Feature 85A, [Link]

IanRoberts_9781316519493_C002.indd 37 03-09-2022 06:53:57

38 2 constituents and categories

theme over the coming chapters, and return to it armed with more empirical and
theoretical knowledge in Volume III.
How can we distinguish the various categories? This is quite a tricky matter,
and there are no absolute hard-and-fast diagnostic tests. For lexical categories in
particular, we can make use of four kinds of properties, relating to the main areas
of linguistic structure. Hence we can distinguish categories on the basis of their
morphology, their syntax, their phonology and aspects of their meaning. Let us
look at each of these in turn.

2.2.2 Morphological Diagnostics for Categories

The morphological criteria are arguably the easiest to see. We can
distinguish lexical categories in terms of the typical inflections and other end
ings they can have. So, for example, regular count nouns in English can have
the plural marker -s (as in boys but not fish, as we saw in Chapter 1). There is
a handful of nouns with irregular plurals: children, mice, men, feet and a few
more. Mass nouns such as milk, water, air, bread denote uniform quantities of
stuff rather than countable individuals, and as such do not have plurals. Many of
them can, however, be ‘coerced’ into acting like count nouns and thereby form
ing plurals: for example, beer is a mass noun, but one can say I drank two beers,
meaning two units (glasses, pints or types) of beer.
English verbs inflect for past tense and third-person singular (3Sg) agree
ment in the present, as we saw in Chapter 1. The regular form of the past tense
is -ed, although there are about 150 irregular verbs in Modern English: sang,
brought, drove, took, etc. All verbs (as opposed to auxiliaries) take the regular
3Sg present-tense ending: the only irregularities are says, which is pronounced
‘sez’ (IPA /sɛz/) rather than ‘says’ by most people and does, pronounced /dʌz/
not /du:z/ whether used as a main verb or an auxiliary.
As we saw in Chapter 1, languages with richer inflections offer more morpholog
ical ways of distinguishing categories. We saw that in Italian fish the noun is pesce,
while fish the verb is pescare. This distinction reflects the prevalent pattern in Italian,
making verbs and nouns very easy to distinguish in terms of their inflections (and all
but eliminating fully ambiguous forms comparable to English fish). Nouns mainly
fall into one of three classes, which can be distinguished by their singular endings:
one class ends in -a, forms the plural in -e and is feminine in gender (e.g. mela/mele
‘apple(s)’); a second class ends in -o, forms the plural in -i and is masculine in gen
der (e.g. gatto/gatti ‘cat(s)’) and the third class, illustrated by pesce, ends in -e in the
singular and forms its plural in -i; this last class may be either masculine or feminine.
Verbs largely fall into one of four conjugation classes marked by a characteristic
vowel immediately following the root. The first conjugation is marked by -a-, as in
pescare (this is the truly open, productive conjugation; for example, a fairly recent
innovation is googlare, ‘to google’); the second conjugation is marked by unstressed
-e-, as in leggere (‘to read’), the third by stressed -e- as in vedere (‘to see’) and the
fourth by -i- as in dormire (‘to sleep’). In all cases, the infinitive ending is -re. Italian
verbs have a large number of tense and agreement endings, far too many to list here.

IanRoberts_9781316519493_C002.indd 38 03-09-2022 06:53:57

2.2 Categories 39

All of this makes it very easy to identify verbs in terms of their morphology. What
we see in Italian is typical of the Romance languages, and, with variations, com
mon across the Indo-European languages as a whole. Some languages show case
marking on nouns, which indicates the function of the noun (or really of the NP) in
the clause: nominative case marking typically marks subjects, accusative direct
objects, dative indirect objects, etc. So in Latin dominus (‘master’) has the -us
ending for nominative singular (in this class of nouns, known as the second declen
sion), -um for accusative singular, -o for dative singular and in fact three more cases.
In languages of this type, nouns can be identified by case marking (adjectives agree
with the nouns they modify, so they can too).
 In Modern English, common adjectives can be identified by their ability to
form the comparative with the ending -er (tall – taller) and the superlative in -est
(tallest). This applies only to adjectives of two syllables or less; hence there is no
comparative beautifuller or superlative beautifullest. This constraint applies to
the root form of adjectives, so unhappier exists, since the root is the two-syllable
happy. As usual, there is a handful of irregular adjectives, e.g. good – better –
best, bad – worse – worst.
Regular adverbs are formed from the corresponding adjective by adding -ly,
e.g. beautifully, happily, badly. There are some irregular adverbs, e.g. well, and
some which simply do not add -ly, e.g. fast. Conversely, there are some adjec
tives which end in -ly, e.g. friendly. Prepositions are invariant: they do not inflect
at all in English.
So morphological criteria can identify lexical categories in English. In most
of the Romance and the other Germanic languages, where there is typically more
inflection than in English, these criteria are correspondingly more useful and
reliable. As we saw with our fish sentences in Chapter 1, English allows many
apparently inflectionless, and hence morphologically ambiguous, forms (we
suggested that this may be due to a large amount of silent inflection in English).

2.2.3 Syntactic Diagnostics for Categories

Given the paucity of overt inflection in English, syntactic criteria
may provide better diagnostics for categories, especially functional categories
(which tend either to show irregular inflection or to be invariant). Typical syn
tactic tests rely on the distribution of categories, identifying positions in the
sentence which are reserved for words of a given category; each category thus
has its characteristic distribution. For example, only auxiliaries invert in direct
yes/no questions, so alongside the declarative in (4a) we have the yes/no inter
rogative (so-called because it invites the answer ‘yes’ or ‘no’) in (4b):
(4) a. Cambridge will flood as a consequence of global warming.
b. Will Cambridge flood as a consequence of global warming?

In (4a), the declarative has the order Subject – Auxiliary – Verb … . In (4b),
the interrogative shows the ‘inverted’ order of subject and auxiliary. Here the
auxiliary is will, indicating (roughly) future tense. If there is no auxiliary in the

IanRoberts_9781316519493_C002.indd 39 03-09-2022 06:53:57

40 2 constituents and categories

declarative sentence, the seemingly meaningless, ‘dummy’ auxiliary do appears

in the ‘inverted’, pre-subject position in the interrogative. This can be seen in (5):
(5) a. Cambridge flooded after all the heavy rain last spring.
b. Did Cambridge flood after all the heavy rain last spring?

In the declarative (5a), there is no auxiliary and past tense is marked by the -ed
ending on the verb. In (5b), the auxiliary do appears, marked for past tense as
did, and precedes the subject, while the verb has no ending. Placing the inflected
verb in front of the subject is ungrammatical (indicated by the asterisk * preced
ing the sentence):
(6) *Flooded Cambridge after all the heavy rain last spring?

This way of forming interrogatives is characteristic of Modern English. Until

roughly the seventeenth century, verbs could also ‘invert’ with the subject. So
we find examples like the following in Shakespeare:
(7) Looks it not like the king? (Hamlet Act One, Scene I)

For this reason, readers familiar with Shakespeare and other writers from the
seventeenth century or earlier may find that examples like (6) have a certain
Shakespearean ring to them, but they are not really part of Modern English.
Following traditional grammar, we have said that clauses are divided into a
subject and a predicate. The predicate is usually a Verb Phrase, VP. Other cate
gories are able to function as predicates, though, providing we include the aux
iliary be. We can see this in examples like the following, where the predicative
category is labelled in the way we saw for the fish sentences in Chapter 1 (AP
stands for Adjective Phrase, PP for Prepositional Phrase):
(8) a. John is [AP nice ].
b. John is [AP interesting ].
c. John is [PP in a bad mood ].
d. John is [NP a nice person ].
e. John is [VP sleeping ].

In (8e), the verb sleep appears in its ‘progressive’ form marked with -ing, indi
cating an ongoing situation.
In addition to following be, predicative categories can also follow certain
verbs, such as seem.
An important distributional property of VPs is that, unlike predicative APs,
PPs or NPs, they are unable to directly follow verbs like seem, as the ungram
maticality of (9e) shows:
(9) a. John seems [AP nice ].
b. John seems [AP interesting ].
c. John seems [PP in a bad mood ].
d. John seems [NP a nice person ].
e. *John seems [VP sleeping ].

IanRoberts_9781316519493_C002.indd 40 03-09-2022 06:53:57

2.2 Categories 41

So this test picks out VPs, since only VPs are ungrammatical following seem.
One important distributional test which picks out NPs is that only NPs can be
subjects. Thus only an NP (which may consist of just a single noun or pronoun)
can appear in the blank in (10):
(10) ___ can be a pain in the neck.

From (10), we can form the sentences in (11a) by inserting a single noun or pro
noun (representing a whole NP), or those in (11b) where we see more complex
NPs, but not those in (11c), where the words inserted are not nouns:
(11) a. You/kids/injections/syntax/Dave can be a pain in the neck.
b. Professors of Linguistics/other people’s kids/injections which go wrong/
fish fish fish can be a pain in the neck.
c. *Walk/tall/in can be a pain in the neck.

Similarly, only VPs (which can be just a single verb) can appear between an
auxiliary and a manner adverb, i.e. in the slot in (12):
(12) Students can ___ quickly.

In (13a), we have single-verb VPs in that position and in (13b) more com
plex VPs. In (13c) the words inserted are not VPs and so the sentences are
ungrammatical:
(13) a. Students can talk/write/learn/understand quickly.
b. Students can dissolve in sulphuric acid/get married/conclude that you’re not
worth listening to/fish fish quickly.
c. Students can *Olly/*kids/*injections/*syntax/*tall/*in quickly.

The distributional tests we have seen so far allow us to identify NPs, VPs and
auxiliaries.
Here is a distributional test for APs. Only (gradable) APs can appear in the
blank slot in (14):
(14) The students are very ____ .

In (15a) we see simple, one-adjective APs in that slot, in (15b) we have complex
APs of various kinds, while (15c) shows that other categories cannot appear there:
(15) a. The students are very intelligent/diligent/nice/eager.
b. The students are very much more intelligent than I expected/diligent in
handing in their essays/nice to talk to/eager to please.
c. The students are very *from/*walk/*Olly.

Finally, a test for PPs is that they can be identified by their characteristic intensifiers
such as straight and right. Thus only PPs can appear in the blank slot in (16):
(16) John walked straight ____ .

In (17), we have the same pattern as in (11), (13) and (15). Example (17a) gives
simple PPs (intransitive PPs, containing just a P); (17b) illustrates PPs of various

IanRoberts_9781316519493_C002.indd 41 03-09-2022 06:53:57

42 2 constituents and categories

kinds, almost always with an NP object, and (17c) shows that APs, NPs and VPs
cannot appear in this slot:
(17) a. John walked straight out/on/up/in.
b. John walked straight out of the room/on to his destiny/up the hill/into the
pub.
 c. John walked straight *red/*talk/*Olly.

We see that syntactic, distributional tests can isolate the main lexical categories,
NPs, PPs, VPs and APs. In an inflectionally poor language like English, these
are probably the most reliable category diagnostics available.

2.2.4 Phonological Diagnostics for Categories

Phonological tests are rather limited in scope in English. However,
there are some. For example, there is a small class of disyllabic words in English
which, like fish, are ambiguously nouns or verbs in their written form. However,
the categorial distinction is marked by stress, as the following examples
illustrate:
(18) a. Apple wants to increase its profits.
b. Apple wants an increase in its profits.

In (18a), increase is a verb and has stress on the second syllable (shown by the
bold letters). In (18b), increase is a noun and is stressed on the initial syllable.
Another case where phonology indicates an aspect of syntactic structure concerns
the difference between word stress and phrasal stress. Hence blackbird, with
stress on the first syllable, is a word, in fact a compound noun. This noun denotes
a particular species of bird. On the other hand, black bird, with stress on bird, is an
NP (or part of an NP; strictly speaking there should be a determiner or plural mark
ing). This NP denotes any bird which happens to be black, following the usual rules
for attributive modification in English (so, red bus denotes any bus which happens
to be red, and so on). Thus one can say That black bird is not a blackbird without
self-contradiction, which would be impossible if the different stress patterns did not
indicate different structures (NP vs N) and therefore different meanings.

2.2.5 Semantics and Categories

Turning now to semantic criteria, these are typically invoked in
traditional definitions of syntactic categories (often called ‘parts of speech’).
Nouns are defined as naming a person, place or thing, while verbs name actions
and adjectives qualities. Here are some typical definitions of verbs: ‘A verb
describes what a person or thing does or what happens’ ([Link]
[Link]/grammar/word-classes/verbs); ‘A verb is simply a doing or action
word’ ([Link]/bitesize/articles/zpxhdxs); ‘Verbs are words that show
an action (sing), occurrence (develop), or state of being (exist)’ ([Link]
[Link]/dictionary/verb).

IanRoberts_9781316519493_C002.indd 42 03-09-2022 06:53:57

2.2 Categories 43

There is something to these definitions, although they are rather vague and lim
ited. On their own, they scarcely suffice to identify categories. Take, for example,
‘what someone does’ as part of the first definition just given, or ‘action’ in the
second and third definitions (after all, actions are what people do). The word
action clearly denotes ‘action’, but it’s a noun. Similarly for occurrence, also a
noun, or ‘state of being’ (existence is a noun). Or consider the following example:
(19) The economy worries John.

Here worries is a verb, but it’s neither an action nor a ‘state of being’ (either
of the economy, or of John). Is it an occurrence? That seems a rather difficult
question to answer, but it’s not clear that answering either way would shed much
light on things. Similarly, nothing is a noun, but it doesn’t denote a thing, and it
certainly doesn’t denote a place or a person.
Despite the vagueness of these definitions, we can discern some semantic
component in category distinctions. This can be seen relatively clearly in English
where, as we have repeatedly noted by now, words can belong to different cate
gories without changing form. For example, consider the different meanings of
round in the following examples:
(20) a. the round church
b. Round the rugged rock the ragged rascal ran.
c. These cars round corners very nicely.
d. Time for another round.

In (20a), round is an adjective, making up an AP on its own, and, much in line

with the traditional semantic definition of adjectives, it denotes a property of
the noun it modifies, its shape. In (20b), round is a preposition. It denotes a path
in relation to the NP following it; again, prepositions are typically defined as
denoting locations or paths. In (20c), round is a verb. Here again, it denotes a
path in relation to its object, corners, traversed by its subject, these cars; so we
can roughly think of it as denoting an action or a characteristic situation involv
ing the subject and the object. In (20d), we have round as a noun. Here it denotes
a particular cultural convention (characteristic of English pubs) metaphorically
based on a circular shape. In each case, there is the same core of meaning in
round, involving a circular shape or path; how that notion is interpreted in rela
tion to the other elements in the sentence depends on the word’s syntactic cate
gory and, as a consequence of that, its syntactic relations with the other words.
So the traditional idea that adjectives denote qualities, prepositions locations or
paths, verbs actions or situations and nouns “things” has some purchase here. We
saw the same with fish: as a noun, it denotes a category of animals (a ‘thing’), as
a verb it denotes the action of hunting the thing in question. The adjective fishy,
formed by adding the characteristic adjectival suffix -y, denotes the quality of
being ‘fish-like’ (and has various metaphorical extensions which are perhaps
more commonly used than the basic sense). There is no prepositional variant of
fish, and it is rather difficult to imagine what that might be.

IanRoberts_9781316519493_C002.indd 43 03-09-2022 06:53:57

44 2 constituents and categories

2.2.6 Conclusion
We have now seen several different ways of identifying syntactic
categories. None of them is absolutely foolproof, but taken together, they are
fairly reliable. Arguably, the distributional syntactic criteria are the most reliable
for English. In languages with richer inflectional morphology, such as Italian,
morphology can be more informative than in English.
There is little doubt that the diagnostics for categories vary from language
to language. But do the categories themselves vary? Does every language have
the same category inventory? In discussing this question at the beginning of
this section, we pointed out that the noun–verb distinction may be universal,
while adjectives and prepositions probably are not. Moreover, there appears to
be variation in the inventory of functional categories a language can have. We
might want to include Gender as a functional category, a ‘satellite’ of N, in
Italian given the presence of grammatical gender in that language, but we would
probably not include it in English, since there is no grammatical gender. This
reasoning might lead us to exclude quite a few familiar functional elements,
determiners for example, from our grammar of Chinese. These questions remain
open and, as we said above, we will revisit them in Volume III, when we have
more facts and more theory to bring to bear on them. For now, we will assume
that English has the lexical categories N, V, Adj, Adv and P and at least the func
tional categories Aux(iliary), D(eterminer) and C(omplementiser). Furthermore,
our conception of the nature and inventory of the functional categories will be
successively revised as we go on.

2.3 Constituent Structure

2.3.1 Phrasal Constituents
Now that we have an idea of the inventory of categories (at least for
English) and how to find them using various morphological, syntactic, semantic
or phonological tests, we can begin to look at how categories combine to form
constituents and constituent structure. This is really the core notion of syntax;
everything else flows from our conception of the structure of complex syntactic
units such as phrases and sentences, and how these are made up of words (or
smaller units) of various categories.
Let us start with some very simple two-word sentences:
(21) a. Clover slept.
b. Vurdy grew.
c. Charlie laughed.
d. Fish fish.

Taking our cue from how we analysed (21d) in (7) of Chapter 1, we can take
each of these sentences to consist of at the very minimum a noun and a verb, and
assign them the labelled bracketings in (22):

IanRoberts_9781316519493_C002.indd 44 03-09-2022 06:53:57

2.3 Constituent Structure 45

(22) a. [Noun Clover ] [Verb slept ]

 b. [Noun Vurdy ] [Verb grew ]
c. [Noun Charlie ] [Verb laughed ]
d. [Noun Fish ] [Verb fish ]
But we have seen that nouns belong in NPs. So we can substitute more complex
NPs for the simple nouns in (22):
(23) a. [NP The fat cat ] [Verb slept ]
b. [NP The boy who lived ] [Verb smiled ]
c. [NP The owner of the famous dog] [Verb laughed]
d. [NP Fish fish fish ] [Verb fish ]
These more complex categories are Noun Phrases, NPs. NPs consist of at least
a noun, along with other words and phrases that depend on and/or modify that
noun (adjectives/APs, PPs, relative clauses, etc.) as well as satellite functional
categories such as Num and D, as we saw in Chapter 1.
Similarly, we have seen that verbs belong in VPs, so we can substitute more
complex VPs for the simple verbs in (22) and (23):
(24) a. [NP Clover ] [VP ate a mouse ].
b. [NP Kate] [VP hopes that chocolate will not get more expensive ].
c. [NP Beth ] [VP ate her dinner with a nice Chianti ].
d. [NP Fish ] [VP fish fish fish fish ].
So here we see another kind of complex category: the Verb Phrase, VP. VPs
consist of a verb and other words and phrases that depend on/modify that verb
(objects, adverbs, adverbial phrases, subordinate clauses, PPs, etc.). We will
come back to the question of the status of satellite functional categories con
nected to VP such as T, M and Agr in Chapter 4.

2.3.2 Hierarchical Relations among Constituents

As we saw in Chapter 1 with the fish examples, we can represent syn
tactic structure generally with labelled bracketings, such as the following for

(24a): (25) [S [NP [N Clover ]] [VP [V ate ] [NP [D a ] [N mouse ]]]]

In technical terms, the labelled bracketing provides a structural description of
the sentence. Another way to give the structural description of a sentence is
with a tree diagram (or phrase marker):
(26) S
VP NP2
NP1 N1
V
Clover ate N2

D
a mouse

IanRoberts_9781316519493_C002.indd 45 03-09-2022 06:53:58

46 2 constituents and categories

(Here NP1, N1, NP2 and N2 are numbered simply to keep them distinct; the num
bers have no theoretical significance.) It is important to see that tree diagrams
and labelled bracketings present exactly the same information in typograph
ically different ways. The choice between them is a matter of convenience.
However, most people find trees easier to work with, since the relations among
the elements of the tree are more readily visible than in labelled bracketings.
But this is just a matter of personal preference; nothing theoretical depends on
it. In what follows, I will mostly use trees to represent structural descriptions.
We can now introduce some terminology that relates to the parts of the tree
diagram. S, NP, VP, etc. are the nodes of the tree; the nodes are linked by
branches. Branches never cross and all emanate from S, which is often referred
to as the ‘root’ node (in this sense the tree is in fact upside down in relation to
botanical trees, since the root is depicted as being at the top; again this is purely a
matter of convention). The words are terminal nodes; the category symbols (S,
NP, VP, etc) are non-terminal nodes. Pursuing the tree comparison, one could
think of the terminal nodes as the leaves.
The most important relations in the tree are the vertical, hierarchical ones.
The two fundamental relations are dominance and constituency. A category A
dominates another category B just where A is both higher up in the tree than B
and connected to B. We can put this more precisely as follows:
(27) A node A dominates another node B just where there is a continuous path of
branches going down the tree from node A to node B.

In terms of (27), we can define immediate dominance:

(28) Node A immediately dominates node B just where A dominates B and no
node C intervenes on the downward path from A to B (i.e. there is no node
C such that C dominates B and C does not dominate A).

Constituency is the reverse of dominance. A given category, call it B, is a con

stituent of another category A just where B is both lower down in the tree than
A and connected to A by a continuous path of branches going up the tree from
node B to node A. In parallel with immediate dominance, we now define imme
diate constituency as follows:
(29) B is an immediate constituent of A just where B is a constituent of A and no
node C intervenes on the upward path from B to A (i.e. there is no node C
such that B is a constituent of C and A is not a constituent of C).

It is easy to see that (immediate) dominance and (immediate) constituency are

inverse relations: dominance ‘looks downward’ in the tree, while constituency
‘looks upward’. More generally, we can say:
(30) A node A (immediately) dominates a distinct node B if and only if B is an
(immediate) constituent of A.

Let us now apply these definitions to the tree diagram in (26), which I repeat
here for convenience:

IanRoberts_9781316519493_C002.indd 46 03-09-2022 06:53:58

2.4 Conclusion 47
(26) S
VP NP2
NP1 N1
V
Clover ate N2

D
a mouse

We can observe the following dominance relations in (26):

(31) a. NP1 and VP are immediately dominated by S.
b. V and NP2 are immediately dominated by VP.
c. V, NP2, D, N2 and N1 are dominated, but not immediately dominated, by S.
d. D and N2 are immediately dominated by NP2.
e. D and N2 are dominated, but not immediately dominated, by VP.
f. N1 is immediately dominated by NP1.
g. NP1 does not dominate VP or anything VP dominates.
h. VP does not dominate NP1 or anything NP1 dominates.
From (31a) and (31c), and the fact that immediate dominance entails dominance,
we see that S dominates all the nodes in the tree. In fact, this is a good definition of
the root node: the root node dominates all the other nodes in a tree; all the nodes in a
tree are constituents of the root. Similarly, all non-terminal nodes dominate at least
one distinct non-terminal or terminal node, and terminal nodes dominate nothing.
The dominance statements in (31) are exactly equivalent to the constituency
statements in (32):
(32) a. NP1 and VP are immediate constituents of S.
b. V and NP2 are immediate constituents of VP.
c. V, NP2, D, N2 and N1 are constituents, but not immediate constituents, of S.
d. D and N2 are immediate constituents of NP2.
e. D and N2 are constituents, but not immediate constituents, of VP.
f. N1 is an immediate constituent of NP1.
g. NP1 is not a constituent of VP or of any constituent of VP.
h. VP is not a constituent of NP1 or of any constituent of NP1.
The relations of constituency and dominance are the most fundamental syntactic
relations.

2.4 Conclusion
The central goal of the theory of syntax is to account for how words
are grouped into larger units, phrases and sentences. Categories and constituents
are the building blocks for analyses. In this chapter, we have seen how to test for
and isolate various categories of English. None of these tests is foolproof, espe
cially when taken alone, but taken together they do allow us to identify and dis
tinguish categories. The second thing we have seen here are the ways in which
we can present the structural description of a sentence: labelled bracketings and

IanRoberts_9781316519493_C002.indd 47 03-09-2022 06:53:58

48 2 constituents and categories
tree diagrams (or phrase markers). In these terms (using a tree diagram because
it is more convenient), we defined the fundamental relations of dominance and
immediate dominance, constituency and immediate constituency.
We take the notions of category and constituent to be part of UG (they are prob
ably not primitives, but defined in terms of more abstract aspects of UG; nonethe
less, their existence and nature are determined, perhaps indirectly, by UG). Other
languages may have categories English lacks or may lack categories English has
(although we have mentioned that the verb–noun distinction seems to hold up pretty
well everywhere). But the notion that the words of a language fall into distinct cat
egories, isolable by some battery of tests (which, as we saw, may relate to various
other parts of the structure of language: phonology, morphology, semantics), is a
hypothesis about the nature of UG which has certainly not been disproved to date.
Similarly, other languages have constituent structures that differ from those
of English; not all tree diagrams represent universal structures. But, again, the
fundamental ideas of (immediate) dominance and constituency are a hypothe
sis about the nature of UG, as are the proposals that branches never cross, all
branches emanate from the root node and words are terminal nodes. As hypoth
eses about UG, these are ultimately hypotheses about genetically given aspects
of human cognition, as we mentioned in Chapter 1.
In the next chapter, we will see ways to justify the particular constituent struc
tures we have proposed here.

Exercises

A. Categories
1. Assign the following words to a syntactic category (noun,
verb, adjective, preposition), using the tests discussed in
this chapter along with any others you know of. In your
answer, show not only the category, but also how you
apply the tests:
cat, moon, sing, fish, to, possibly, well, dispute, round
2. Now do the same with the following words: noble, enno
ble, near, asleep and around
B. Constituents
3. Give the labelled bracketings and tree diagrams for the
following examples:
a. Boris snorted.
b. The onx splooed the blarrg.
c. Fish fish fish fish.
d. Vurdy saw Goona.
 IanRoberts_9781316519493_C002.indd 48 03-09-2022 06:53:58
Exercises 49

e. Clover ate the mouse.

f. Harry met Hermione.
4. Prepositions (invariant words typically denoting spatial or
temporal relations such as on, about, with, to, behind,
etc.) form Prepositional Phrases, or PPs, by combining
with an NP, as follows:
(i) [PP [P on ] [NP the bus ]]
Give the tree structure and the labelled bracketing for the following
sentences, all of which contain PPs with a structure like (i):
a. Boris talked about himself.
b. Boris talked to the reporter about himself.
c. Boris sent some poison to Emile.
What challenges do you encounter in the last two examples?
C. For further discussion:
5. What do you think might be the structure of NPs like the
following? Present your hypothesis as a tree diagram.
a. fat cat
b. the invisible visible stars
6. Specify all the (immediate) dominance relations in the following
tree:
S
NP N PP
V
VP
Ron thought
D D N the problem

about

NP

6. How can we define terminal and non-terminal nodes in

terms of constituency?
7. S is the root of the tree. How can we define the root in
terms of constituency?

Further Reading
All of the references below discuss the ways in which we can distinguish categories,
using tests broadly similar to those discussed here and, as here, concentrating largely (but
 IanRoberts_9781316519493_C002.indd 49 03-09-2022 06:53:59
50 2 constituents and categories

not exclusively) on English. All of them therefore provide useful backup, and plentiful
further examples, to the ideas introduced here.

Carnie, A. 2011. Modern Syntax: A Coursebook. Cambridge: Cambridge University

Press, Chapters 2 and 3.
Carnie, A. 2013. Syntax: A Generative Introduction, Oxford: Blackwell, Chapters 2, 3
and 4.
Haegeman, L. & J. Guéron. 1999. English Grammar: A Generative Perspective.
Oxford: Blackwell, Sections 1.2.2 and 1.2.3.
Koeneman, O. & H. Zeijlstra. 2017. Introducing Syntax. Cambridge: Cambridge
University Press, Chapter 1.
Larson, R. 2010. Grammar as Science. Cambridge, MA: MIT Press, Unit 9.
Radford, A. 2016. Analysing English Sentences. Cambridge: Cambridge University
Press, Chapter 2.
Tallerman, M. 1998. Understanding Syntax. London: Routledge, Chapter 2.
 IanRoberts_9781316519493_C002.indd 50 03-09-2022 06:53:59

3 Phrase-Structure Rules and Constituency

Tests

3.1 Introduction
In the previous chapter, we saw the basics of the structural descrip
tions of sentences, the essential features of syntactic representations. Structural
descriptions can be presented as labelled bracketings, as in (1a), or as tree dia
grams, as in (1b):
(1) a. [S [NP [N Clover ]] [VP [V ate ] [NP [D a ] [N mouse ]]]].
b. S
VP NP
NP N
V
Clover ate N

D
a mouse

We also introduced basic tree terminology: node, root node, terminal node,
non-terminal node, branch, etc., and the very important mutually definable rela
tions of (immediate) dominance and (immediate) constituency. In addition to
constituent structures, as represented in trees and labelled bracketings (i.e. struc
tural descriptions), categories are the other fundamental notion. We illustrated
the main categories of English, as well as the very important distinction
between lexical and functional categories, and showed how categories can be
fairly relia bly isolated by a battery of syntactic, semantic, morphological and
phonological tests (which vary somewhat from language to language).
The goals of this chapter are to start to describe (and explain) the difference
between grammatical and ungrammatical sequences. In order to do this, we
introduce the mechanisms that generate structural descriptions, i.e. the rules
that build trees and labelled bracketings. These are the Phrase-Structure
rules. The exact Phrase-Structure rules and the structural descriptions they
generate can be justified by independent tests designed to isolate and
distinguish the constituents in a tree or labelled bracketing; these are the
constituency tests. We look first at the Phrase-Structure rules and then at some
constituency tests which, for the most part, work quite well in English.

51

IanRoberts_9781316519493_C003.indd 51 05-09-2022 19:49:18

52 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s

3.2 Phrase-Structure Rules

Here are some examples of ungrammatical sentences of English (in
the sense that they use English words and are just about intelligible to English
speakers; strictly speaking, these sentences are part of neither I-language English
or E-language English):
(2) a. *Spoke John.
b. *Hopes Goona that Vurdy will be on form.
c. *Loves the Leader of the Opposition his wife.

The question is: how do we, as speakers of English, know that these sen
tences are ungrammatical? Clearly we don’t merely store this information
for each individual sentence or we wouldn’t have grammaticality judgements
about novel sentences such as the fish sentences we saw in Chapter 1. What
is needed is some general schema determining which are the grammatical
sentences of English (that is, of the I-language of the typical native speaker
of English).
To put it more technically, we need a mechanism to generate the well
formed structural descriptions of English sentences. The mechanism in ques
tion is the set of Phrase-Structure rules, PS-rules for short. PS-rules are the
formal devices which generate constituent structure, by specifying all and
only the possible ways in which categories can combine. An example of a
basic PS-rule of English (and probably many other languages) is given in
(1), where S stands for Sentence;

(3) i. S → NP VP

This is read as ‘Rewrite the symbol S as the sequence NP VP in that order.’

More generally, the rules are instructions to replace the symbol on the left of
the arrow with the symbol or symbols on the right of the arrow in the linear
order given. In (4) we see some more PS-rules of English:
(4) ii. VP → V (NP) (PP)
iii. NP → (D) N (PP)
iv. PP → P NP

The brackets indicate optional categories. Strictly speaking (4ii), for exam
ple, abbreviates VP → V, VP → V NP, VP → V PP and VP → V NP PP, but
we collapse these rules under (4ii) using the bracket notation. These rules can
generate the structural description for the sentence Clover ate a mouse, given
as a tree diagram in (1a) and a labelled bracketing in (1b), and similar ones.
Here is another example of a tree diagram (which we saw in the Exercises
sections in the previous chapter):

IanRoberts_9781316519493_C003.indd 52 05-09-2022 19:49:18

(5) S NP
VP
3.2 Phrase-Structure Rules 53
N V PP
Ron thought
P D N the problem

about

NP

This tree can be generated by the PS-rules in (6), which combines the rules in
(3) and (4):
(6) i. S → NP VP
ii. VP → V (NP) (PP)
 iii. NP → (D) N (PP)
iv. PP → P NP

Each rule generates a small piece of tree, typically (but not always) a branch
ing node and the constituents which that branching node immediately domi
nates. As such, each rule specifies a set of immediate dominance/constituency
relations, such that the category to the left of the arrow is the label of the node
which immediately dominates the constituent(s) to the right of the arrow, and
the category or categories to the right of the arrow label the node(s) which
is/are immediate constituent(s) of the category to the left of the arrow. So,
PS-rule (6i) generates the structure in (7):
(7) NP VP
S

PS-rule (6ii) has two categories in brackets on the right of the arrow. The
brackets indicate that the categories in question are optionally immediate con
stituents of VP (another way to say this is to say that they are optionally part
of the ‘expansion of VP’, given that PS-rules generally expand the category
to the left of the arrow by specifying more than one category on the right).
Strictly speaking, (6ii) really collapses the four distinct PS-rules seen in (8):
(8) a. VP → V
b. VP → V NP
c. VP → V PP
d. VP → V NP PP

Option (8c) of rule (6ii) generates the piece of structure in (9):

IanRoberts_9781316519493_C003.indd 53 05-09-2022 19:49:19

54 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s
VP PP
(9)
V

Rule (6iv) generates the structure in (10):

(10) P NP
PP

Like rule (6ii), rule (6iii) collapses four rules using the bracket notation. These
are given in (11):
(11) a. NP → N
b. NP → D N
c. NP → N PP
d. NP → D N PP

Rule (6a) generates (12a) and rule (6b) generates (12b):

 (12) a. b. N

NP

NP DN

The pieces of structure in (7), (9), (10) and (12a,b) combine to form the tree in
(5). In (13), (5) is repeated with each part of the structure annotated giving the
PS-rule which generates it:
(13) NP rule (8c) (one
rule VP subcase of (6ii))
S
rule (6i)
(11a) V
N PP

Ron thought P
rule (11b) (one subcase
NP
of (6iii))
rule (6iv)
D N the problem
about
This illustrates how the structural description of the sentence is generated by
the PS-rules. The tree diagram in (13) gives the structural description of the
grammatical English sentence in (14):

(14) Ron thought about the problem.

Let us now return to the ungrammatical sentences, repeated here:

(2) a. *Spoke John.
b. *Hopes Goona that Vurdy will be on form.
c. *Loves the Leader of the Opposition his wife.

IanRoberts_9781316519493_C003.indd 54 05-09-2022 19:49:21

3.3 Recursion 55

In (2a) the verb and the subject are in the wrong order for English. PS-rule (6i),
S → NP VP states “Rewrite the symbol S as the sequence NP VP in that order.’
Whatever expansion of VP we choose from the options in (6ii) (see (8)), V is
always the first immediate constituent of VP. It follows from the combination of
rules (6i) and (6ii) that the subject NP (the NP in (6i)) will precede the verb. Hence
the sentences in (2) cannot be generated by our PS-rules where the immediately
postverbal NP is the subject; if it is interpreted as the object, then the subject is
missing and again the rules in (6) cannot generate the sentence. Our PS-rules are
examples of the PS-rules of English, specifying all and only the well-formed struc
tural descriptions of English sentences, and so the sentences in (2) are ill-formed.
It is important to see that the notion ‘ill-formed’ here means ‘not generated by
the syntactic rules of English’. Our aim is to make that notion coincide as far as
possible with native speakers’ intuitive judgements regarding the grammaticality
of English sentences, which we take to reflect their I-language competence. Here
it is important to remember what we saw in our discussion of centre-embedding in
Chapter 1: native-speaker judgements reflect performance, i.e. acceptability, rather
than competence, i.e. grammaticality. As we mentioned there, though, most of the
time grammaticality and acceptability coincide, and so most of the time matching
native-speaker judgements is a good benchmark for how well our theory is doing.
As we can see from our account of the ungrammaticality of (2), PS-rules
give information about the linear (left-to-right) order of nodes, including the
words at the terminal nodes. They also give information about hierarchical
structure, in that they specify immediate dominance and constituency rela
tions. Finally, since the symbols they use are category symbols (S, N, V, P,
etc.), they give information about the category labels of nodes. So PS-rules
specify three kinds of information about structural descriptions simultane
ously. We will see in Volume II that these three rather distinct kinds of
information can be teased apart.

3.3 Recursion
One of the most important things our discussion of the fish sentences
in Chapter 1 showed us was that the length of natural-language sentences is in
principle unbounded. If infinitely long sentences are grammatical (but maybe
not acceptable, as they can never be ‘performed’), they should be well-formed.
In other words, they should be generated by the PS-rules. Let us see how.
In fact, the rules we already have can generate infinite structures. Consider
again rules (6iii) and (6iv):

(6) iii. NP → (D) N (PP)

iv. PP → P NP

It is easy to see that NP appears on the left of the arrow in (6iii) and on the right
of the arrow in (6iv). Given the way PS-rules work, this means NPs can appear
inside other NPs, i.e. NPs can be constituents of other NPs, as illustrated in (15):

IanRoberts_9781316519493_C003.indd 55 05-09-2022 19:49:21

56 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s

(15) D P N1 P P generated by (6iv)

NP1 NP2

generated by (6iii)

(Again, the subscript numbers on the Ns and NPs serve merely to keep the two
occurrences of this category distinct; they have no theoretical significance.)
Since nothing requires us to apply rules (6iii) and (6iv) in that order, after
applying both rules so as to generate the structure in (15), we can go back and
, thereby intro
apply rule (6iii) again so as to expand NP2 identically to NP1 ducing a second PP,
which we can expand by rule (6iv) to give a third NP,
which we can expand by rule (6iii) to give a fourth NP, which we expand by
rule (6iv) to give a third PP, and so on. In principle, there is no limit to how
many times we can keep applying rules (6iii) and (6iv) to their own output.
The result of iterated application of these rules is complex recursive NPs of
the following kind:
(16) [NP [D the] height [PP of [NP [D the] lettering [PP on [NP [D the ] covers [PP of [NP [D the
] manuals [PP on [NP [D the ] table [PP in [NP [D the ] corner [PP of [NP [D the ] room …

The ability of rules to apply to their own output is known as recursion, a con
cept that originates in mathematics and logic. Rules (6iii) and (6iv) are not indi
vidually recursive, but together they form a recursive rule system. Recursion is
an extremely important concept, as it gives rise to the possibility of sentences
of unlimited length and underlies the fact that human languages are able to
make ‘infinite use of finite means’. As we have just seen, we can construct an
infinitely long NP using just rules (6iii) and (6iv). If our minds contain a recur
sive rule system of this kind for generating sentences in our native I-language,
then our finite brains have infinite capacity. This is clearly a very important
and interesting claim about human cognition, one which justifies looking at
language from the formal and cognitive perspective adopted here.
In fact, we already saw recursion in action in the fish sentences in Chapter 1.
Consider the structural description we gave there for the seven-fish example:
(17) [S [NP [N fish ] [RC [NP [N fish]] [VP [V fish] [NP [N fish ]]]]] [VP [V fish ] [NP [N fish ] [RC
[NP [N fish]] [VP [V fish] [NP [N fish ]]]]]]]

The relative clauses introduce NPs inside other NPs and the object relative
clause introduces a VP inside the main-clause VP. Since, as we said just after
introducing the representation in (17), relative clauses are really sentences,
i.e. of category S, we should substitute S for the RC labels in (17). Then
we see that relative clauses introduce Ss inside Ss. So the reason we can
have infinite fish sentences is that the PS-rules generating relative clauses
are recursive. They feature occurrences of the symbols S, NP and VP on both
sides of the arrow, just as rules (6iii) and (6iv) do for NP and PP.

IanRoberts_9781316519493_C003.indd 56 05-09-2022 19:49:22

3.3 Recursion 57

Relative clauses are one kind of subordinate clause. I won’t give the PS-rules
that generate them here as certain details of their structure remain uncertain and
controversial. Another kind of subordinate clause whose structure appears to be
much more straightforward, however, are complement clauses. A simple com
plement clause is illustrated in (18):
(18) Goona hopes [ that Vurdy arrived].

Here we see a finite clause functioning as the complement, essentially a

clausal direct object, of the verb hope, introduced by the complementiser
(subordinating conjunction) that. It is easy to observe that what follows that
in (18), Vurdy arrived, is a complete sentence on its own. We introduce the
PS-rules in (19) to generate the complement-clause structure of (18):

(19) i. VP → V S′
 ii. S′ → Comp S

Rule (19i) adds to the various possible expansions of VP seen in (8). We could add
‘(S′)’ to rule (6ii); this would subsume (19i) and predict that the sequence V NP S′
is possible (it is, as in persuade Mary that John arrived), the sequence V PP S′ is
possible (it is, as in say to Mary that John arrived) and the sequence V NP PP S′
is possible (this may not be correct, but I will leave this complication aside here).
In (19ii) ‘Comp’ abbreviates ‘complementiser’. Rule (19ii) states that Comp and S
are immediate constituents of the subordinate clause S′. S′ and S are not the same
category: S is a clause and constitutes the root node of a tree, while S′ is the label
of a subordinate clause introduced by rule (19i). Thanks to rule (19ii), we have a
further case of S-recursion: S appears on the left of the arrow in rule (6i) and to
the right of the arrow in rule (19ii). These rules will together generate structural
descriptions with S inside another occurrence of S.
We can expand S introduced by rule (19ii) as NP VP (in fact, this is the only
expansion of S our PS-rules allow as we have formulated them so far). We can
then expand VP using rule (19i) and introduce S again by rule (19ii) and expand
S again as NP VP, reapply rule (6i), reapply rules (19i) and (19ii), and so on
without limit. Once again we see rules applying to their own output, the basic
property of recursion. Recursive application of these rules in this way gives rise
to unlimited sequences of subordinate clauses embedded inside one another.
This is observed in English examples like (20):
(20) Mary hopes that John expects that Pete thinks that Dave said that …

As far as competence is concerned, just as with the fish sentences and the
complex NP in (16), there is no limit to the length, or to the depth of embed
ding, of sentences like (20). The usual performance restrictions mean that
infinite sentences cannot be uttered or written down, but three PS-rules ((6i),
(19i) and (19ii)) suffice to generate them.
The structural description of (18) in tree format is given in (21):

IanRoberts_9781316519493_C003.indd 57 05-09-2022 19:49:22

58 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s
rule (19i)
(21) NP
VP
S
generated by
N V S′
Goona hopes
Comp
that
generated by rule (19ii)
S
V
NP VP arrived
Vurdy
There are other complementisers in English, which introduce different kinds
of subordinate clauses. As we can see in (18) and (21), that introduces a
finite, declarative subordinate clause. In (22a), whether introduces an indi
rect question, and in (22b) for introduces an infinitival clause:

(22) a. Goona wonders [S′ whether Vurdy will arrive on time ].

b. Goona arranged [S′ for Vurdy to arrive on time ].

As indicated here, both subordinate clauses are S′s, with whether and for in C.
What follows whether in (22a) can stand alone as a complete sentence (Vurdy will
arrive on time), and so there is no difficulty with applying rule (19ii) here. The
difference with the that-complement in (18) is that the whole subordinate S′ stands
for the direct interrogative Will Vurdy arrive on time? What we observe in indirect
interrogatives is the absence of subject-aux inversion (briefly discussed in Chapter
2, see examples (3) and (4) there): in the S′ in (22a) Vurdy precedes will just as in
a declarative. We also observe that the presence of whether contributes what we
could call ‘interrogative force’, i.e. it makes the declarative into an interrogative.
In (22b), what follows the complementiser for cannot stand alone as a com
plete sentence: *Vurdy to arrive on time. We clearly have a subject (Vurdy) and
a predicate (arrive on time) as in (18) and (22), but what is missing is tense: there
is no finite verb or auxiliary. ‘Complete’ clauses, those able to stand alone, must
have a finite verb or auxiliary to mark tense (although the tense marker may be
silent in English, as in the following complete sentences: I/you/we/they/the boys
arrive). The particle to marks non-finiteness; it shows up in almost all infinitives
in English. It looks as though we cannot generate *Vurdy to arrive on time with
rule (6i), since there is no place for to. But it seems clear Vurdy is the subject NP
and arrive on time the VP. Assuming that to is not part of the subject NP, which
seems highly unlikely, there are two possible structural descriptions for *Vurdy
to arrive on time, which are shown in the labelled bracketings in (23):
(23) a. [S [NP Vurdy] [? to ] [VP arrive on time ]]
b. [S [NP Vurdy ] [VP to arrive on time ]]

Rule (6i) cannot generate (23a), but it could generate the NP VP structure in
(23b). But now rule (6ii) (with its various expansions) cannot generate the VP
there, which has to as its first constituent. In order to accommodate infiniti
val subordinate clauses, one of the two basic PS-rules in (6) will have to be

IanRoberts_9781316519493_C003.indd 58 05-09-2022 19:49:23

3.4 Constituency Tests 59

modified. I will leave this question open here, but we will come back to it in
Chapter 4 (see Section 4.4).
Combining (6) and (19), we now have the following set of PS-rules:

(24) i. S′ → Comp S
ii. S → NP VP
iii. VP → V (NP) (PP) (S′)
iv. NP → (D) N (PP)
v. PP → P NP

These rules can generate a large number, in fact, given their recursive nature,
an infinite number of English sentences. However, they do not generate all
the grammatical sentences English: we have seen that there is no place for the
infinitive to here, so we cannot generate (22b).
As we said at the end of the previous section, PS-rules give us three kinds of infor
mation: (i) information about hierarchical structure (‘vertical’ information, in terms
of tree diagrams), information about linear precedence (‘horizontal’ information)
and information about the category labels of nodes. These rules are very powerful
formal devices which are capable of generating structural descriptions in a precise
way. But how do we know that structural descriptions of the kind we have been
looking at in this chapter are the right ones for English? The descriptions make very
clear claims about constituent structure, but are they correct? We have also seen at
least one example where it is not clear where to place a constituent in a structure: the
question of where to put to in infinitival clauses like (22b). Which of the structures in
(23) is the correct one, and why is the possibility that to is a constituent of subject NP
‘highly unlikely’ as I said above? In the next section we will turn to questions of this
kind, by showing how there are various ways of testing constituent structure, just as
there are various diagnostics for categories as we saw in Section 2.2.

3.4 Constituency Tests

Let us look once more at PS-rule (6i):

(6i) S → NP VP

This rule applies to give us the constituent structure (25b), rather than (25c),
for (25a):
(25) a. John ate the cake.
b. [S [NP John ] [VP ate the cake ]]
c. [S [VP John ate ] [NP the cake ]]

But why do we write the rule like this? What would be wrong with writing the
rule as (6i′), which would give the constituent structure we see in (25c) for (25a)?

(6i′) *S → VP NP

IanRoberts_9781316519493_C003.indd 59 05-09-2022 19:49:23

60 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s

One obvious answer comes from the traditional idea that clauses consist of a
subject and a predicate. But this venerable idea does not really tell us anything
about phrase structure (despite what we said in Chapter 1): phrase structure rep
resents hierarchy, order and categories, as we have seen. It does not represent
grammatical functions or relations such as subject and predicate. From what we
have seen up to now, these notions have no place in our theory; we may wish to
build them in somehow, and we will do this in Section 5.2 (since these notions
seem to be so useful for informal discussion we might want to have a theoretical
way to understand them, but the fact remains that we have not actually said any
thing about this so far). We will look at the question of the theoretical status of
grammatical functions in full detail in Chapter 1 of Volume II.
So, coming back to choosing between (6i) and (6i′), what this really amounts
to is choosing between the structural description in (25b) and that in (25c) for
(25a). In (25b), which is the analysis we have been assuming up to now, the
verb and the direct object form a constituent that does not include the subject,
while in (25c) the subject and the verb form a constituent that does not include
the object. The question is: why should we prefer (25b) over (25c)? To put the
question another way, what is the evidence that ate the cake is a constituent and
John ate is not a constituent? The evidence is not directly audible on the basis
of what we hear as (25a); this is because constituent structure, like most of syn
tax, is silent. The linear order in (25a) is clearly compatible with either (25b)
or (25c). So we must find a way to determine what the hierarchical structure is.
The same question arises in relation to the NP P sequence in (26a) or the N A
sequence in (26b):
(26) a. They gave [NP the book ] [P to ] Mary.
b. They gave [NP Mary ] [AP sweet ] cookies.

We would naturally treat to in (26a) as a Preposition forming a PP with Mary.

PS-rule (6iv) will generate this structure (and one version of (6ii) can generate
the sequence V NP PP seen in gave the book to Mary). We could justify this on
the grounds of the tight semantic connection between to and Mary, or on the
grounds that to marks the indirect-object function of Mary here. But these are
arguments based on semantics and grammatical functions, and this is not the
kind of information the structural descriptions generated by PS–rules carry (at
least not on the basis of anything we have seen so far). Similar considerations
apply to (26b): semantic and functional considerations, i.e. the fact that sweet
must be interpreted as modifying cookies and cannot be interpreted as modifying
Mary, would lead us to group the AP sweet with the noun cookies to form an
NP (which has the direct-object function). But can we find constituency-based
arguments for this conclusion? If we can, then not only do we establish the
constituency and have clear indications as to which set of PS-rules is the right
one, we also confirm the intuition (or hunch, really) that constituent structure is
closely connected to semantics, and that grammatical functions are also some
how linked to constituency relations (see Section 5.2).

IanRoberts_9781316519493_C003.indd 60 05-09-2022 19:49:23

3.4 Constituency Tests 61

Constituency tests of various kinds can show us to a large extent what the cor
rect constituent structures are. These tests are manipulations of sentences which
are sensitive to phrasal categories such as NP, VP, etc. There are several kinds
of constituency tests. These involve: (a) manipulating the order of elements in
such a way as to show that certain sequences of words must be manipulated
together, i.e. that they constitute a phrase (these are clefting, wh-movement
and fronting); (b) substituting a sequence of words with ‘pro-forms’ of various
kinds; (c) ellipsis, deleting a sequence of words in such a way that its interpreta
tion is recoverable from the linguistic context; (d) coordination, conjoining two
phrases of the same category with and, and (e) fragments, whether a sequence
can stand alone and be in an intuitive sense ‘complete’, even if it is not a com
plete sentence. We will now look at each kind of test in turn. These operations
are cases of a class of rules distinct from PS-rules, transformational rules, one
type of which (wh-movement) we will focus on in Chapter 5.
Let us begin with wh-movement. Fronting a phrase containing a wh-word (i.e.
the interrogative pronouns and determiners who, what, which, etc.; see Chapter
5) also targets constituents. This operation, known as wh-movement, is a very
important one for syntactic theory, and we will introduce it fully in Chapter 5.
Again, there is a gap in the sentence where the questioned phrase was, which we
mark with a ‘t’ (and which can also be taken as a silent copy):
(27) a. Which friends does Mary hope that John will like t ?
b. What did the Party Chairman send t to John?
c. Who did the Party Chairman send a book to t ?
d. To whom did the Party Chairman send a book t ?

These examples show us that the direct objects in (32a,b) are constituents, that
the NP following the P to in (32c) is a constituent, and that the PP to whom is a
constituent.
Non-constituents, i.e. strings of words that do not form an independent, unique
phrase, bolded in (28), cannot be fronted:
(28) a. *What to did the Party Chairman send t John?
b. *What to John did the Party Chairman send t ?

Strictly speaking the ungrammaticality of (28) does not give us information about
constituency; only the successful cases of wh-movement do this. Wh-movement
does not apply to VP: the various wh-phrases correspond to different grammatical
categories: who, what are NPs (respectively animate and inanimate), which is a
D, why an AdvP or PP (‘for what reason’), when a temporal NP or PP, where a
PP, how an AP or AdvP and how (many) a measure expression. But there is no
wh-word which questions a VP. Hence, we cannot use wh-movement as a diagnos
tic for a VP constituent. It does not follow from this that there is no VP constituent.
A further permutation we can apply to sentences in order to isolate constitu
ents is fronting. This operation ‘highlights’ phrasal constituents by placing them
at the beginning of the sentence. The fronted phrase often functions as a topic

IanRoberts_9781316519493_C003.indd 61 05-09-2022 19:49:23

62 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s

which ‘is commented on’ by the rest of the sentence. So, from the neutral sen
tence (29a) we can derive (29b), where the direct object the new car is fronted:
(29) a. Mary hopes that John will like the new car.
b. The new car, Mary hopes that John will like t.

Again, we see a trace in the position where the direct object would normally be
 in (29b). Fronting can apply to CP and VP, as in (30):
(30) a. That John will like her friends, Mary hopes t.
b. (Mary hoped that John would like her friends) … and [VP like her friends ]
he did t.

For (30b) to sound natural, it helps to give some context, as shown here.
Sentence (30a) may sound a little stilted, but it certainly seems acceptable. The
examples in (30) should be contrasted with (31) (where again non-constituents
are bolded):
(31) a. *A present to, the Party Chairman sent t John.
b. *A present to John, the Party Chairman sent t.

We still have no evidence that the sequences a present to or a present to

John are constituents, but now we have some evidence that VP and CP are
constituents.
A further permutation-based constituency test is clefting. Clefting is the
permutation of the order of elements illustrated in (32):
(32) a. The Party Chairman sent a book to John. →
b. It was to John that the Party Chairman sent a book t.
c. It was John that the Party Chairman sent a book to t.

The sentence in (32a) ‘neutral’. In (32b), the sequence to John has been clefted;
in (32c) just the NP has been clefted, ‘stranding’ the preposition to. The general
schema for clefting is given in (33):
(33) S → It was XP that S.

(33) is not a PS-rule. It represents a different kind of rule, a transformational

rule, that we will introduce properly in Chapter 5. The S on the left of the arrow
represents the neutral sentence, e.g. (32a); what is on the right of the arrow is
the clefted version of that sentence, e.g. (32b). The S on the right of the arrow
contains a gap (similar to the gaps in relative clauses which we observed in the
fish sentences in Chapter 1). Here it is marked as ‘t’ (standing for trace); this
indicates where the XP would normally be in the non-clefted version of the
sentence. We can think of ‘t’ as a silent copy of the XP. It stands for a single
constituent, so in (32b) we do not have two traces following book, one for P and
one for NP, but just one, standing for PP.
For our purposes here, the most important thing about clefting as schematised
in (33) is that XP has to be a constituent. In (32b), the clefted XP is the PP to
John. The sentences in (34) are two further cases of clefting:

IanRoberts_9781316519493_C003.indd 62 05-09-2022 19:49:23

3.4 Constituency Tests 63

(34) a. It was the Party Chairman that t sent a book to John.

b. It was a book that the Party Chairman sent t to John.

In (34a), the NP the Party Chairman is clefted, while in (34b) it is the NP the
present. So the clefting test has isolated three constituents for us: [PP to John ],
[NP the Party Chairman ] and [NP a book ].
If we try to cleft non-constituents, bolded in (35), the result is ungrammatical:
(35) a. *It was the Party Chairman sent that t a book to John.
b. *It was a book to that the Party Chairman sent t John.

In (35a) the sequence the Party Chairman sent is clefted, and in (35b) a book
to. In both cases the result is ungrammatical. Since there could be independent
reasons why clefting some constituents is not good, the failure of a constituency
test does not really tell us anything. For example, clefting VP yields a rather odd
result (although probably not as bad as (35)):
(36) ??It was send a book to John that the Party Chairman did.

Despite the oddity of (36), we do not immediately conclude that VP is not a

constituent in (32a) and, as we have seen, other tests can successfully pick out
VP for us.
So, the success of the test tells us that the fronted sequence is a constituent,
but the failure of the test, strictly speaking, doesn’t tell us anything. Unlike the
VP, however, no constituency test will isolate the strings the Party Chairman
sent or a book to as in (35). These strings are equivalent to the putative VP in
(25c) and the possible constituents discussed in relation to (26).
The fourth type of constituency test involves substitution of pro-forms. Pronouns
(I/me, he/him, she/her, etc.) are noun-like elements that have little semantic content
of their own but stand in for something else. This ‘something else’, known as the
antecedent, determines what the pronoun refers to. So, in an example like (37),
John can be naturally interpreted as the antecedent of he (although it doesn’t have
to be; he could refer to any contextually salient male individual):
(37) John hopes that he will win.

So (37) is interpreted to mean ‘John hopes that he, John, will win.’ Since John is
an NP, we should really call pronouns NPs too, since they stand for whole NPs
rather than just nouns. We can see this from the two cases of pronoun substitu
tion in (38b) and (38c):
(38) a. [NP The man who wears glasses ] hopes that he will win.
b. * The he who wears glasses hopes that he will win.
c. He hopes that he will win.

Treating he as a noun in (38b) leads to ungrammaticality, while treating he as

occupying the same position as the whole relative clause gives the grammatical
(38c). The ungrammaticality of (38b) doesn’t necessarily imply that the position
of he is not a constituent; in fact, the noun man clearly is a consitituent. It shows
us that pronouns are NPs (and not Ns).

IanRoberts_9781316519493_C003.indd 63 05-09-2022 19:49:23

64 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s

Other categories have pro-forms too. Most VPs can be replaced by (do) so,
as in (39):
(39) John has promised Mary a book, and Bill has done so too.

Here we see the VP promised Mary a book in the first conjunct, i.e. the verb
along with both the direct and the indirect object, are replaced in the second
conjunct by done so. As with pronoun (i.e. pro-NP) substitution in (38), do so
stands for the whole VP:
(40) a. * … , and Bill has done so Mary too.
b. *... , and Bill has done so a book too.
c. *... , and Bill has done so Mary a book too.

In (40a), just the verb and the direct object (promise a present) have been sub
stituted; in (40b), just the verb and the indirect object (promise Mary), and (40c)
just the verb promise. Again, the fact that these examples are all ungrammatical
doesn’t necessarily imply that the substituted categories are not constituents (in
fact, the verb on its own clearly is a consitituent); what it shows is that do so
substitutes an entire VP (and not, for example, V).
In the light of the conclusion that do so substitutes for a whole VP, consider
the following examples:
(41) a. John put his car in the garage on Tuesday, and Peter did so too.
b. *John put his car in the garage on Tuesday, and Peter did so on the driveway
on Wednesday.
c. *John put his car in the garage on Tuesday, and Peter did so his bike on
Wednesday.
d. John put his car in the garage on Tuesday, and Peter did so on Wednesday.

In (41a), which is grammatical, do so in the second conjunct appears to be replacing

the sequence put his car in the garage on Tuesday, i.e. the verb put, the direct object
his car, the locative PP in the garage and the temporal PP on Tuesday. We there
fore conclude that all of these elements are in the VP. This conclusion seems to be
confirmed by the ungrammaticality of both (41b) and (41c): in (41b), the locative
PP on the drive is not substituted and ungrammaticality results, while in (41c) the
object NP his bike is not substituted, again leading to ungrammaticality.
But (41d) behaves differently. Here the temporal PP on Wednesday is not
substituted by do so in the second conjunct, but the sentence is grammatical.
The difference between the temporal PPs on Tuesday/on Wednesday in these
examples and the direct object and the locative PPs is that the former are not
obligatory with the verb put. The direct object and the locative are, though, as
the ungrammaticality of sentences like (42b–d), where one or other (or both) of
them is left out, shows in contrast with the grammatical (42a):
(42) a. John put the car in the garage.
b. *John put the car.
c. *John put in the garage.
d. *John put.

IanRoberts_9781316519493_C003.indd 64 05-09-2022 19:49:23

3.4 Constituency Tests 65

The grammaticality of (42a) also shows us that the temporal PPs are not required
for grammaticality. Such PPs are optional modifiers, or adjuncts, giving ‘extra’
information about the time the event took place, but their presence is not required
in order for the sentence to be intuitively ‘complete’. On the other hand, (42b–d)
feel ‘incomplete’: required information about what is being put (or) where is not
given. This is because the direct object and the locative PP are arguments of the
verb put; they must appear in the VP when put is V. More technically, put cat
egorially selects for a direct-object NP and a locative PP. But it does not select
for an adjunct temporal PP.
What the ungrammaticality of (41b,c) tells us then is that selected arguments
of V must form part of the VP with the verb, and hence must undergo do so
replacement. Then (41d) can be taken to indicate that adjuncts are outside the
VP, and so do not correspond to do so replacement. But what about (41a)? The
fact that the substituted VP is interpreted as put the car in the garage on Tuesday
indicates that the adjunct PP is part of the VP. So (41d) appears to be telling us
that the adjunct PP is outside VP, and (41a) appears to be telling us it is inside
VP. We conclude for now that adjuncts (of this kind, at least) are optional con
stituents of VP, while selected arguments are obligatory constituents of VP. This
conclusion accounts for the pattern seen in (41) as well as the examples in (42)
and is consistent with the idea that do so corresponds to VP. We will see a more
sophisticated treatment of adjuncts in the next chapter.
The do so test and the fronting test allow us to find VPs. We can now apply
these tests to our two structures for John ate the cake in (25), repeated here as (43):

(43) a. [S [NP John ] [VP ate the cake ]]

b. [S [VP John ate ] [NP the cake ]]

Fronting gives the following results:

(44) (I thought John would eat the cake and)

a. .. eat the cake he did t!
b. *… John eat did t the cake!

With a slight tweak to the tense of the verb to create a natural context for
VP-fronting, (44a) is perfectly grammatical. Example (44b), on the other hand,
is strongly ungrammatical (bordering on what is sometimes called ‘word salad’,
a sequence of words so unintelligible that it is almost impossible to impose any
kind of structure or meaning on it).
Do so substitution, seen in (45), gives rise to a similar contrast:
(45) a. John ate the cake, and Mary did so too.
b. *John ate the cake, and did so the cake too.

Example (45b) is not word salad, but it is not acceptable. We can also notice that
Mary has completely disappeared from this example. This is a reflection of a
very general fact about substitution and ellipsis operations: substituted and elided
material is subject to a recoverability condition, i.e. the operations of substitu
tion and ellipsis (the latter of which we will look at in depth in Volume II) must

IanRoberts_9781316519493_C003.indd 65 05-09-2022 19:49:23

66 3 p h r a s e-s t r u c t u r e r u l e s a n d c o n s t i t u e n c y t e s t s
be such that the semantic representation can find, i.e. recover, what has been sub
stituted or elided. In (45b), Mary is not recoverable. The contrast in (45) clearly
supports the structure in (43a) over that in (43b). This is a good result because, as
we noted above, it seems that the constituents line up with the functions of sub
ject and predicate (although our PS-rules cannot and do not state this).
As a final point on substitution, consider the following examples:
(46) a. %The Party Chairman sent it John.
b. The Party Chairman sent it.

Example (46a) is ungrammatical for most English speakers, although there are
varieties in the north-west of England, around Liverpool and Manchester, where
it is accepted (the ‘%’ is used to indicate a form which is acceptable in one vari
ety of a language but not another; of course, this kind of variation simply reflects
the fact that the sociocultural E-language concept ‘English’ does not correspond
exactly to aggregates of I-languages – people from the relevant area of England
have slightly different I-languages from those from elsewhere, a fact reflected
in their differing grammaticality judgements of examples like (46a). However,
in these varieties it is interpreted as the direct object (a/the book); it cannot be
interpreted as substituting for a present to. Similarly, (46b) is grammatical, but it
can only be the direct object. An indirect object (i.e. to John) cannot be recovered
here. Again, then, a book to and a book to John fail the constituency test.
Ellipsis is the next kind of constituency test. Ellipsis elides, or deletes, mate
rial, subject to the recoverability requirement we have already seen in relation to
substitution. VP-ellipsis is quite natural in English, as in:
(47) a. John can speak Mandarin and Mary can speak Mandarin too.
b. John will leave tonight and Mary will leave tonight too.
c. John has passed the exam and Mary has passed the exam too.
d. John is writing a book and Mary is writing a book too.

(48) John ate a cake and Mary did eat a cake too.

This example illustrates an aspect of the English auxiliary system that will play
a major role in our analysis of clauses from the next chapter on. The struck-out
sequence in the second conjunct here consists of the verb and its direct object,
and so we can quite reasonably analyse this as VP. This would be consistent with
our other results for constituency tests for VP: fronting and do so substitution (see
(44a) and (45a)). Example (48) differs from (47) in that there is no auxiliary in the
first conjunct and the auxiliary do appears in the second one. There is a difference
between the two conjuncts here: in the first one, the verb bears the tense marking:
we have ate, not eat. In the second conjunct, the past-tense marking is carried by
the auxiliary do, in the form of did, and so the struck-out verb form lacks tense
marking (hence eat, not ate). We see from (47) and (48) that, generally speaking,
auxiliaries do not have to be deleted under VP-ellipsis. This suggests that finite
auxiliaries are not part of VP. VP-fronting confirms this:
IanRoberts_9781316519493_C003.indd 66 05-09-2022 19:49:23