Conversation analysis.

What is it and when

should I use it?
Conversation Analysis is used to understand the meanings of real language and how people
really speak.

Cathy Gibbons

Nov 8, 2021 • 7 min read

We use 'ums' and 'ers', verbal fillers, (‘y’know’, ‘so’, and ‘like’), and encouragers (‘uhuh’,
‘mmm’). We also interrupt, talk over the other, fail to finish sentences, finish a sentence for
the other, pause and use intonation in our speech to communicate. It can seem random and
disorderly. But the fact that we can largely comprehend each other suggests that there is an
order to this seeming randomness.

When I first came across Conversation Analysis (CA), in John Heritage’s Garfinkel and
Ethnomethodology1, its detail and specificity appealed to my scientist brain. And as a
qualitative researcher the appeal of grasping the detail of individual words and phrases also
appealed, along with apparent script-like talk (being influenced by the likes of Erving
Goffman). But I wasn’t entirely grasping the point of what Conversation Analysis is for.

So, what is Conversation Analysis?

According to Paul ten Have (2007)2 in Conversation analysis what an utterance means (or the
job that it is doing) depends on its order (or sequential position). Not just in the speech of an
individual, but also the speech of others in an interaction. Fundamentally CA is about
analysing turn-taking in speech.

Doesn’t sound very exciting or insightful, does it? And isn’t it just dialectics, or social
scripting, or even discourse? Yes, but it goes beyond turn-taking in speech or analysis of
sequential phrasing. It is also the analysis of sequential positioning of sentences, and the
sequential positioning of individual words across that interaction.

The development of Conversation Analysis by Harvey Sack2 in the 1960's was prompted by
recordings of telephone calls to a suicide prevention centre. There is a fascinating (and a little
chilling) modern example used by Prof Elizabeth Stokoe, in a YouTube vlog, of a call to the
emergency services, where a woman’s spoken words indicate that she is ordering a pizza, but
her meaning is very different.

What’s amazing about that call is that it doesn’t take long – a couple of turns – for the 911
dispatcher to realise that ‘I’d like a pizza for delivery’, means ‘Please send the police because
someone is threatening my life in my home’ .

If Data Could Talk: What is Conversation Analysis and Why is it Important? 18:22-18373

There is a lot to say about this example, especially around the increasing use of AI in
seemingly scripted human interactions, and its limitations. However to keep within the scope
of the blog, Conversation Analysis would be used to study the sequential positioning of words
and their meaning with and across an interaction. The caller has used her knowledge of two
seemingly very different scripts to convey necessary information (her location) and her
meaning (I need assistance). The emergency services dispatcher has used their knowledge of
an alternative script (ordering pizza) to enable the caller to use specific individual words
(yes/no) that do not disrupt her Pizza ordering script, whilst retaining the meanings required
of an emergency call script.

What’s Conversation Analysis good for? When should I

choose to use it?
Crucially, Conversation Analysis is used to understand the meanings of real language and
how people really speak. Whilst CA has usage across many disciplines including
anthropology, linguistics and semiotics, organisational psychology, even business and law,
CA has its roots in understanding sociological levels of communication. Talk-interaction
within organisational - I might stretch as far as saying ‘institutional’- and procedural settings
continues to be a focus of CA, although it is not limited to organised settings. Examples of
interaction include Board meetings, classroom or instructional talk, court proceedings, and
patient-practitioner consultations, or even training scenarios.
These examples illustrate another common focus of CA studies; the examination of power
within interactions. For example, in the board room you might call it the power-play of
individuals attempting to assert dominance or display co-operation, regardless of the quality
of the substance of the ideas and suggestions under discussion. In classroom settings, control,
dominance, and resistance might be examined in relation to gender. In health, the purpose can
be to examine normative power structures, such as the expectation of compliance of the non-
powerful, with the wishes of the powerful e.g. that a patient is expected to take their
medication in the way that the pharmacist has told them to, or they will comply with a weight-
loss programme because the GP has outlined why it is important and a way to achieve it4. CA
can help uncover or develop facilitative interaction patterns which can be applied in the
training of health professionals, and potentially address perceptions of power imbalances, or
at least not make matters worse5.

The position of the researcher in the gathering of data is at a more objective position than
much qualitative research. You are highly unlikely to be gathering the data through
interviewing people yourself. You will be examining audio recordings of the interactions of
others in real settings. You may be an observer, or be given access to recordings made by
others in real encounters. Any encounter that you examine will not have been set-up for the
purpose of research; it will be an end in itself, with no account taken of the research needs.
E.g. a meeting that you are allow access to and permission to record for research purposes, or
recordings that you have been given access and permission to analyse. You are objectively
positioned in the data. That said, CA maintains its qualitative heritage, as the reflexive
examination of subjective positioning of the researcher in the analysis of the data is expected.

Things to bear in mind

Transcript notation.

CA has very particular transcript notation to keep track of all the utterances, intonation,
pauses, emphases and drawing of breath: learning to even read a CA transcript can be difficult
at first. I remember reading a transcript in Paul ten Have’s work (2007)2 and thinking that it
looked like a foreign language to me. Actually, some of the transcripts were in a language that
is foreign to me, (Dutch) and fact that I didn’t spot this for a long time illustrates just how
strange CA can look at first.

This makes transcription time-consuming (and tedious). I recall advice from my own days of
training that 1 minute of talk will take about 10 minutes to create an edited transcription (after
practice), whereas 1 minute for CA transcript will take about 1 hour, even with experience. If
you are part of a team, the decision to use transcription services may be taken out of your
hands. If you are doctoral student, do talk to your supervisor. A doctorate is a key time for
developing your skills as a researcher, and my personal view is that learning the skills of
transcription is time spent with your data, so it's never wasted. However, you will need to
allow time to develop your skills and fluency at both transcribing and reading CA, if you
decide it suits your research purpose.

Gail Jefferson developed the eponymous transcription system for four decades ago and such is
its strength that it remains highly regarded and in widespread use in sequential analyses,
especially CA. You can see an example of Jefferson transcript notation here6, with its under-
linings, arrows, timed pauses and other forms of notation. Although it was developed for CA,
this style of detailed transcription system is used where-ever specific detail of timing, pauses,
emphasis and intonation is required e.g. in some forms of Discourse Analysis. It can also be
adapted to create a convention or protocol, where the full range of notation is not necessarily
required.

The units of analysis and the volume of data required.

The units of analysis in CA are small, with considerable attention to detail required. Paul ten
Have gives an example from the work of R.M. Frankel (ten Have, 2007, P.4.2). It is just 3
lines of text and 18 words of an interaction in which a doctor confirms a patient’s supposition
that their cancer treatment will leave them infertile. The narrative to analyse the conversation
is four-times as long. This should indicate to you the need for careful decision making around
the kinds of interactions that you wish to gather or get access to (and you must have an audio
record). Otherwise, you’ll be left with burdensome and unrealistic amounts of data to
transcribe and analyse, or insufficient data to form an argument. I don’t always recommend a
distinct or short pilot phase for data gathering, but I do for CA. It gives you the chance to
effectively assess what you need access to, and realistically estimate how long your analysis
will take. Don’t be the student who gathers loads of data but leaves themselves no time to
analyse it. Less is more.