Discourse analysis: approaches, steps and an example

Discourse analysis is the study of language in use: how talk and texts construct versions of the world, and what those versions do. Instead of reading an interview answer as a window onto what someone thinks, you ask how the answer is built, which ways of talking it draws on and what it achieves. The term covers several traditions with different questions and different demands on your transcripts.

What is discourse analysis?

‘Discourse’ has two main senses. In the narrow sense it means language in use, spoken or written, studied in its context. In the broad sense that goes back to Michel Foucault, a discourse is a historically formed way of talking about something, such as a medical discourse, which shapes what can be said and who can say it. James Paul Gee writes a capital-D ‘Discourse’ for socially recognisable ways of being a certain kind of person, which combine language with actions, values and objects.

The traditions share a social constructionist starting point: language does not simply mirror the world but helps to make it. That is why discourse analysts avoid claims about what participants ‘really’ think, and ground every claim in the talk itself. Marianne Jørgensen and Louise Phillips’s Discourse Analysis as Theory and Method (2002), which compares Laclau and Mouffe’s discourse theory, critical discourse analysis and discursive psychology, is a good first read.

Four approaches to discourse analysis compared

For a dissertation, choose one tradition and follow it consistently. These four are the ones students meet most often.

Gee’s seven building tasks, from An Introduction to Discourse Analysis: Theory and Method, work as questions to ask of any stretch of talk, which makes them easy to apply to interviews. Conversation analysis is sometimes listed as a fifth approach. It studies the sequential organisation of talk, such as turn-taking (Sacks, Schegloff and Jefferson, 1974) and repair, mainly in naturally occurring talk, and needs the most detailed transcripts of all.

Critical discourse analysis: Fairclough, Wodak and van Dijk

Critical discourse analysis (CDA), now often called critical discourse studies, examines how language takes part in power, inequality and ideology, and takes a stance against the injustice it finds. Its three best-known approaches differ in what they treat as context.

• Fairclough’s three-dimensional model. Norman Fairclough treats every communicative event as at once a text, a discursive practice and a social practice. You describe the text (vocabulary, grammar, structure), interpret the discursive practice (how the text is produced and read, and which discourses it mixes) and explain the social practice (how the event relates to institutions and wider social structures). The model first appears in Language and Power (1989), now in its third edition (2015), and fully in Discourse and Social Change (1992); Analysing Discourse (2003) applies it to social research. He later called his approach dialectical-relational.
• Wodak’s discourse-historical approach. Martin Reisigl and Ruth Wodak set it out in Discourse and Discrimination (2001). It places texts in their historical and political context, combines several kinds of data and theory, and analyses discursive strategies such as how social actors are named, which qualities are attributed to them and how claims about them are argued.
• Van Dijk’s sociocognitive approach. Teun van Dijk adds a cognitive link: discourse and society are connected through the shared knowledge and ideologies of social groups. His ideological square describes how texts stress good things about ‘us’ and bad things about ‘them’, and play down the opposite.

Wodak and Meyer’s Methods of Critical Discourse Studies (3rd edition 2016) has a chapter on each approach by its authors. CDA suits policy documents, media coverage and institutional talk. With interviews it works best when your question connects what participants say to the discourses and institutions around them.

Foucauldian discourse analysis and discursive psychology

Foucauldian discourse analysis draws on Foucault’s historical work, notably The Archaeology of Knowledge (1969; English translation 1972), which treats discourses as practices that form the objects they speak of. The analysis asks which discourses make an object such as ‘stress’ or ‘the international student’ thinkable, which subject positions they offer, and how they relate to power and to what people can do and feel. Foucault offered no step-by-step method, so students often follow Carla Willig’s six stages in Introducing Qualitative Research in Psychology (4th edition 2022): discursive constructions, discourses, action orientation, positionings, practice and subjectivity.

Discursive psychology began with Jonathan Potter and Margaret Wetherell’s Discourse and Social Psychology: Beyond Attitudes and Behaviour (1987). They argued that attitudes are not fixed inner states that talk simply reports: people describe the same thing differently depending on what the talk is doing. Their key tool is the interpretative repertoire, a recurrent set of terms and metaphors that speakers draw on to describe actions and events. Derek Edwards and Jonathan Potter’s Discursive Psychology (1992) turned to how speakers handle fact, interest and accountability, for example how an account is made to sound factual rather than self-serving. Later work moved closer to conversation analysis, and Potter and Hepburn (2005) set out the problems with research interviews.

How to do discourse analysis on interview transcripts

The steps below follow discursive psychology; adapt them for CDA. Use the terms of your tradition, and expect to return to the transcripts many times.

1. 1

   Frame a question about language

   Ask how something is constructed or what talk accomplishes, not what participants believe. ‘How do international students account for staying quiet in seminars?’ suits discourse analysis; ‘Why are international students quiet?’ calls for a different method.
2. 2

   Treat the interview as interaction

   Your questions are part of the data. Keep them in the transcript and analyse each answer as a response to them.
3. 3

   Transcribe at the level your approach needs

   Decide which features count: words only, or also pauses, overlaps and emphasis (see the next section).
4. 4

   Select extracts

   Collect every passage relevant to your question. Potter and Wetherell (1987) call this coding, but its purpose is to gather material for analysis, not to count instances or sort them into themes.
5. 5

   Analyse construction, function and variation

   For each extract, ask what the talk does, how it is built to do it and how it varies within and across interviews. When one event is described in two ways, ask what each version does.
6. 6

   Check your claims

   Potter and Wetherell suggest four checks: coherence (does the analysis explain how the talk fits together?), participants’ orientation (do speakers treat your distinctions as relevant?), new problems (does it show the problems a way of talking creates for speakers?) and fruitfulness. Every claim should point to lines in an extract.

How much transcription detail you need

The right level of detail depends on the tradition. For CDA and Foucauldian analysis, an accurate verbatim transcript is usually enough: every word, including repetitions and self-corrections, with clear speaker labels. Discursive psychology and conversation analysis need more, because pauses, overlaps and stress are part of what the talk does. They use Gail Jefferson’s notation, set out in her Glossary of transcript symbols with an introduction (2004). Common symbols include:

• (0.6) for a pause timed in tenths of a second, and (.) for a micropause;
• [ and ] for the start and end of overlapping talk;
• = for talk that follows with no gap (latching);
• °degree signs° around talk that is quieter than the talk around it;
• colons for a stretched sound, as in ‘we::ll’, and underlining for stress;
• ((double brackets)) for the transcriber’s descriptions, such as ((laughs)).

Jefferson notation takes far longer than a verbatim transcript, so many students transcribe everything verbatim and add the notation only to the extracts they analyse closely. Ask your supervisor what your programme expects, and put your transcription key in an appendix.

An automatic transcript can give you a first draft, but no more. Spirelight produces a draft with speaker labels and a timestamp on each turn, and for English recordings its word-for-word style keeps filler words such as ‘um’. It does not produce Jefferson notation: pauses are not timed, overlaps and intonation are not marked, and every word needs checking against the audio. The timestamps take you to the right moment when you add the notation yourself.

Check your university’s rules before you upload recordings anywhere: a voice identifies the speaker, so a recording is personal data even without names. Spirelight offers no data processing agreement; your university, not you, would sign one. Transcription runs on EU servers, files are stored with Google Firebase in the US, and AI features run on OpenAI in the US, which your ethics application and participant information sheet must state.

Where your university allows it, you can transcribe your interviews with timestamps, correct the draft while you listen and export it to Word before you add your notation.

Invented example: a discursive psychology reading

The study, extract and analysis below are invented. The question is how international master’s students account for staying quiet in seminars. The extract uses simplified Jefferson notation; INT is the interviewer and ANA a participant.

Only the wider dataset can show whether these are repertoires: a full analysis would ask whether other participants draw on the same two, and whether Ana uses a different one elsewhere.

Discourse analysis vs thematic analysis

Thematic analysis looks for patterns of meaning across a dataset and usually reads what participants say as an account of their experiences or views. Discourse analysis treats the same words as action: it asks how an account is constructed and what it achieves. Braun and Clarke say their reflexive thematic analysis suits most questions about experiences and views, but not questions about language practice.

The unit of analysis differs too. Thematic analysis builds themes across the dataset; discourse analysis works through extracts line by line, often across fewer interviews in more depth, so agree the size of your dataset with your supervisor. A constructionist thematic analysis can look at how something is talked about, but it is still thematic analysis, so say clearly which method you use.

Writing up discourse analysis in your dissertation

In the methodology chapter, name the tradition and the texts you followed, explain why it fits your question, and describe your transcription conventions and how you selected extracts. Say how your position may have shaped your reading; a positionality statement is one way to do it. In the analysis chapter:

• organise the analysis by repertoire, discourse or strategy, not by interview question;
• number the lines of each extract, refer to them and let your analysis run longer than the extract;
• include contrasting cases, not only the clearest examples;
• avoid claims about what participants really believe or feel.

Label speakers with pseudonyms, and remember that a pseudonymised extract is not anonymous. Check every extract against the recording before you submit your dissertation or thesis.