Conversation Analysis — Studying Talk-in-Interaction
- Introduced by
- Harvey Sacks, Emanuel Schegloff, and Gail Jefferson
- Year
- Late 1960s–1974 (foundational lectures 1964–1972; landmark article 1974)
- Type
- Qualitative research method
- Family
- Discourse Analysis
What does it do?
Conversation Analysis (CA) is a qualitative research method that examines the fine-grained sequential structure of naturally occurring talk and social interaction.
Developed by sociologists Harvey Sacks, Emanuel Schegloff, and Gail Jefferson in the 1960s and 1970s, CA investigates how participants in a conversation accomplish social actions — such as invitations, refusals, or diagnoses — through the precise moment-by-moment organisation of their talk, including turn-taking, sequence structure, repair, and recipient design.
The intuition behind the method
When two people talk, an enormous amount of social work happens in fractions of a second: one person knows when to stop and let the other speak, an awkward silence signals trouble, a hesitation before 'yes' can mean 'no'. CA treats conversation not as a neutral vehicle for transmitting information but as the fundamental site of social life. Like a musician analysing a score, the CA researcher uses detailed transcripts — capturing overlaps, pauses, intonation, and restarts — to reveal how speakers systematically construct actions and make sense of each other's behaviour turn by turn, in real time.
History
Conversation Analysis emerged from the sociology of Harold Garfinkel's ethnomethodology in the mid-1960s, when Harvey Sacks — initially working with Schegloff and later Jefferson — began analysing recordings of calls to a suicide prevention hotline. Sacks's lectures at the University of California (1964–1972) developed the core concepts of turn-taking, adjacency pairs, and sequence organisation. The landmark 1974 article by Sacks, Schegloff, and Jefferson in the journal Language established the 'simplest systematics' of turn-taking and introduced the Jefferson transcription conventions that remain standard. Since the 1980s, CA has expanded into institutional interaction across medicine, law, education, and technology-mediated communication.
When to use it
CA is the method of choice when the research question concerns how social actions, identities, or institutional tasks are accomplished through the sequential organisation of talk. It is especially well suited to institutional settings — medicine, law, education, psychotherapy, service encounters — where the structure of interaction has practical consequences. CA requires audio or video recordings of real interactions; it cannot be conducted on interview or survey data. It is not appropriate when the goal is to measure attitudes, build substantive theory about a social process, or understand subjective lived experience — grounded theory, phenomenology, or mixed-methods designs are better fits for those goals.
Limitations and considerations
- Requires access to genuine recordings; retrospective accounts, interviews, or reconstructed notes are insufficient for CA.
- Jefferson transcription is time-intensive — one hour of recorded interaction can take 10–20 hours to transcribe at the level of detail CA demands.
- Findings describe interactional structure, not subjective meaning or social causation; CA does not explain why participants act as they do in terms of motives, attitudes, or social forces.
- The emphasis on sequential detail can make findings seem narrow or inaccessible to audiences outside linguistics, sociology, or communication studies.
- Ethical and logistical challenges in obtaining genuine recordings of sensitive interactions (e.g., medical consultations, legal proceedings) can be substantial.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Grounds analysis in naturally occurring data, avoiding the artificiality of researcher-imposed tasks or interview questions.
- The next-turn proof procedure provides a built-in, participant-derived check on analytic interpretations.
- Reveals the micro-level sequential machinery of talk that other methods treat as background noise.
- Highly applicable to institutional settings where the organisation of interaction has direct practical and policy implications.
- Does not require a large pre-defined sample; even a handful of episodes can yield replicable structural descriptions when systematically collected.
Applications
- Examining how physicians and patients negotiate diagnoses, treatment decisions, and information delivery in clinical consultations.
- Analysing how teachers manage classroom participation, correct errors, and scaffold learning through interaction.
- Investigating how customer service agents handle complaints, requests, and account terminations in call centres.
- Studying how courtroom questioning constructs facts and undermines witness credibility through sequential structure.
- Exploring how news interviewers maintain neutrality and how politicians resist answering questions directly.
- Analysing how users interact with voice-based AI assistants and where breakdowns in human-machine conversation occur.
How it works
How it works
The researcher collects audio or video recordings of naturally occurring interactions — medical consultations, courtroom hearings, everyday phone calls, classroom exchanges, or workplace meetings. The defining requirement is that the interaction be genuine (not staged or elicited by interview), because CA is committed to studying talk as participants actually produce and encounter it in real social settings.
Recordings are transcribed using the Jefferson notation system, which captures not only the words spoken but also pauses (timed in tenths of a second), overlapping speech, intonation, volume, speech rate, and other prosodic details. This granular transcript is the primary analytic object; its precision is what allows the researcher to observe the sequential machinery of talk that ordinary note-taking would miss.
The analyst reads through the transcript looking for a recurring interactional phenomenon of interest — for example, how speakers manage turn transitions, how they initiate and resolve misunderstandings (repair), or how a particular action type (invitation, refusal, request) is structured across many instances. CA works inductively: the phenomenon drives the analysis rather than a pre-specified coding scheme.
Each selected instance of the phenomenon is examined in detail for its sequential context: what came immediately before (the prior turn), what the current turn does as a social action, and how the recipient responds (the next turn). Because the next turn reveals the recipient's own understanding of the prior turn, it serves as a naturally occurring check on the analyst's interpretation. Collections of instances are compared to identify the systematic patterns and variant forms.
After building a collection of instances and describing the sequential pattern, the analyst formulates a structural description of the phenomenon — for example, the preference organisation of a particular action sequence or the resources speakers use to project turn completion. This description can be applied to new data, refined through further cases, and connected to broader claims about social interaction or institutional practice.
Common pitfalls
- Analysing interview data as if it were natural interaction — interview talk is shaped by the researcher's agenda and cannot be treated as ordinary conversation; CA must be applied to data the participants would have produced regardless of the researcher's presence.
- Skipping the collection phase — analysing single instances without building a collection of comparable cases undermines the method's claim to identify systematic patterns rather than idiosyncratic behaviour.
- Imposing external categories before analysis — coding for 'politeness', 'power', or 'empathy' before examining the sequential structure imposes the analyst's framework on data that should reveal participants' own categories.
- Under-transcribing — omitting pauses, overlaps, and prosody strips away the very features that CA shows to be interactionally consequential; a standard orthographic transcript is not adequate for CA.
- Conflating CA with discourse analysis or content analysis — CA has distinct theoretical commitments (sequential organisation, participant orientation, next-turn proof) that are not shared by thematic or critical discourse approaches.
Frequently asked questions
Can I use CA on interview data I have already collected?
No. CA requires recordings of naturally occurring interaction — talk that participants would have produced regardless of the researcher's presence. Research interview data are shaped by the interviewer's questions and the interviewee's awareness of being studied; they cannot be analysed as ordinary or institutional conversation. If your data are interviews, discourse analysis or thematic analysis is more appropriate.
How many interactions or recordings do I need?
There is no fixed minimum. CA typically works with collections of 20–50 or more instances of a specific phenomenon drawn from a corpus of recordings. Some studies build collections from hundreds of instances across many hours of data; others achieve analytic saturation with fewer recordings when the phenomenon is clearly delineated. The key is building a collection large enough to identify systematic patterns and their variant forms, not meeting a participant count.
Do I need to learn the Jefferson transcription system?
Yes, if you are conducting CA yourself. Jefferson notation — covering timed pauses, overlap onset and offset, intonation contours, volume, and speech rate — is not optional decoration; it records the very features that CA demonstrates to be interactionally consequential. Many analyses hinge on the exact placement of a 0.3-second pause or the onset of overlap. Commercial transcription services and speech-to-text tools produce orthographic transcripts that are inadequate for CA without substantial specialist revision.
What is the difference between CA and critical discourse analysis (CDA)?
CA and CDA share an interest in talk and text but differ fundamentally in their theoretical commitments. CA suspends assumptions about power, ideology, or social structure, working only from what participants demonstrably orient to in the interaction itself. CDA begins from a critical theoretical framework and uses discourse data as evidence of broader social inequalities or ideological positions. CA practitioners are generally cautious about importing external explanatory frameworks because doing so risks overriding what participants themselves are doing in the data.
Is CA suitable for online and text-based communication?
CA has been extended to text-based interaction (chat, SMS, social media) and multimodal interaction (video calls, gesture, gaze). These adaptations require modified transcription conventions that capture the medium's specific features — for example, message delivery timing and platform affordances in chat. The core sequential logic of CA applies, but analysts must adapt transcription practices to the modality.
Related methods
Sources
Sacks, H., Schegloff, E. A., & Jefferson, G. (1974). A simplest systematics for the organization of turn-taking for conversation. Language, 50(4), 696–735.
Open publication referenceCopied.Sidnell, J. (2010). Conversation Analysis: An Introduction. Wiley-Blackwell.
ISBN: 978-1405159098Copied.