Conversation Analysis in Language Learning
Conversation analysis (CA) is the study of the hidden machinery of ordinary talk: how two or more people build a conversation together, turn by turn, and manage in real time to take turns, understand each other, and put things right when they go wrong. It grew up not inside linguistics but in sociology, when Harvey Sacks, Emanuel Schegloff and Gail Jefferson began, in the 1960s, doing something then radical — tape-recording real conversations and transcribing them in obsessive detail, down to the pauses, overlaps and in-breaths. What they found overturned the intuition that casual talk is loose and disorderly: everyday conversation is one of the most tightly organized systems human beings operate, and it runs on rules its own speakers cannot state.
CA is one branch of the wider field of discourse analysis — the study of language above the sentence. Where discourse analysis also covers written texts, genre and register, CA narrows in on one thing: naturally occurring spoken interaction, studied from recordings of real talk rather than invented examples. That narrowing is its strength. By refusing to make up data and looking only at what people actually do, CA uncovered an order in conversation that no armchair theory had predicted — and, as it turns out, an order a language learner has to master to sound like a competent speaker rather than a walking grammar book.
The machinery of talk: turns, pairs and repair
CA's method is intellectual patience: take a fragment of recorded talk, transcribe every detail, and ask why the participants did exactly what they did, in exactly that order. Its roots are in ethnomethodology — Harold Garfinkel's study of the taken-for-granted reasoning people use to make everyday life orderly. From that starting point CA produced three findings that between them describe the skeleton of any conversation.
The first is turn-taking. In their landmark paper “A Simplest Systematics for the Organization of Turn-Taking for Conversation” (1974), Sacks, Schegloff and Jefferson showed that speakers build turns out of recognizable units, project where a unit will end, and apply an ordered set of rules at each possible completion point — the current speaker may select the next, or a next speaker may self-select, or the current speaker may continue. This machinery is why conversation flows with astonishingly little silence and overlap even though nobody agreed in advance who would speak when. The system is universal in outline but locally calibrated: how long a silence is comfortable, whether an overlap reads as enthusiasm or rudeness, varies from language to language — which is exactly why perfect grammar does not stop a learner from constantly interrupting, or from never managing to get a word in.
The second is the adjacency pair: paired utterance types in which a first part makes a particular second part expectable — question and answer, greeting and greeting, invitation and acceptance-or-refusal. The relationship is one of conditional relevance: once the first part is out, its second is not merely likely but noticeably absent if it fails to appear. And the two possible second parts are not equal. CA calls this preference organization: the preferred response (acceptance, agreement) comes fast and plain, while the dispreferred one (refusal, disagreement) comes delayed, softened and accounted for — “well… I’d love to, but…”. This softening apparatus, invisible to the native speaker who produces it automatically, is a real learning target: a refusal delivered bare, without the hesitation and the reason, is grammatically flawless and socially jarring. Even beginnings and endings are engineered — Schegloff and Sacks showed in “Opening Up Closings” (1973) that conversations are not simply stopped but jointly closed, through pre-closing moves (“well…”, “anyway…”) that offer each party a last chance to raise a topic before the goodbyes.
The third, and for learners the most important, is repair — the organized way talk handles its own troubles of speaking, hearing and understanding. In “The Preference for Self-Correction in the Organization of Repair in Conversation” (1977), Schegloff, Jefferson and Sacks mapped its architecture. A repair has a trouble source, an initiation, and an outcome, and it can be classified by who does what: self-initiated repair (the speaker flags the trouble: “I mean…”) versus other-initiated (a listener flags it: “sorry?”, “you mean the blue one?”), each of which can be completed by self or other. Their central result is a robust preference: speakers overwhelmingly prefer to correct themselves, and conversation is built to give them the first chance to do so. Other-initiation, when it happens, is done gently — an open-class “huh?”, a partial repeat, a candidate understanding — precisely because correcting someone else is interactionally delicate. For a language learner this is not academic: repair is the survival system of real conversation.
Conversation analysis and learning a language: CA-for-SLA
For a long time CA and second-language research barely spoke. Mainstream second language acquisition (SLA) modelled learning as something happening inside an individual mind, measured against native-speaker grammar; CA insisted on studying observable, collaborative behaviour and refused to treat one speaker as a defective version of another. The turning point was a 1997 article by Alan Firth and Johannes Wagner in the Modern Language Journal, which argued that SLA had become lopsidedly cognitive and mentalistic and needed to take the social and interactional dimension of language use seriously. That challenge opened the door through which CA walked.
What emerged, named “CA-for-SLA” by Numa Markee and Gabriele Kasper in a 2004 special issue of the same journal, is a distinctive way of studying language learning. Its boldest move is methodological: rather than infer hidden mental change from test scores, it treats learning as something that can be seen in the interaction itself — a change over time in how a learner participates, in what they can do with and through talk. Markee, in his 2000 book Conversation Analysis, and later work catalogued the observable “learning behaviours” that surface in real L2 interaction: word searches (a learner audibly hunting for a term, inviting help), designedly incomplete utterances in which a teacher leaves a slot for the learner to fill, doing pronunciation, insisting on finishing a turn, and explicit checks of understanding. None of these is a private mental event; all are public, joint, and analysable — the participants themselves display what they are treating as a learnable.
Repair is where this pays off most directly. The other-initiated repair sequences that CA had already described turn out to be, in a second-language conversation, exactly the moments where meaning is negotiated and something new can be picked up: a listener signals trouble, the parties work it out together, and the resolution is jointly built and often reused a moment later. Kasper and Wagner, in their 2011 overview “A conversation-analytic approach to second language acquisition”, framed the whole enterprise around interactional competence — the idea that what a learner is acquiring is not just words and rules but the ability to deploy them in the sequential, turn-by-turn machinery of real talk, and that this competence visibly develops. Learning, on this view, is not the private growth of a grammar; it is becoming a more capable participant in conversation.
What the learner takes from it: sequences, repair, authenticity
Conversation analysis earns its place in an encyclopedia of language learning because its findings convert unusually well into things a learner can actually practise. Three conversions matter most.
First, conversational sequences are teachable units. Openings and closings, the pre-closing moves that wind a chat down, the softening apparatus of a polite refusal, the little formulas that hold a turn while you plan (“let me think…”, “how should I put it…”) — these are not grammar to be assembled on the fly but recurrent, semi-fixed routines. Most are stored and retrieved whole, exactly as described in the entry on formulaic sequences, which means a learner can collect and rehearse them as ready-made items rather than reinventing them mid-conversation.
Second, repair is a core skill, not an admission of failure. Every conversation breaks down; competent speakers simply have the tools to mend it. A learner who commands a handful of repair moves — a clarification request (“sorry, could you say that again?”), a candidate understanding (“you mean…?”), a self-repair formula (“what I meant was…”) — can keep a conversation alive through gaps that would otherwise end it. CA reframes the very thing beginners fear, the moment of not-understanding, as a normal, manageable, and above all learnable part of talking.
Third, and most consequential for how courses are built, CA supplies the strongest argument for authentic materials. When conversation analysts turned their transcripts on the invented dialogues of language textbooks, the gap was glaring. Jean Wong's studies of telephone talk (2002) found that the opening and closing sequences of textbook phone calls were “absent, incomplete or problematic” when set beside recordings of real calls. Manufactured dialogue tends to be machinery-free: no overlap, no hesitation, no repair, no discourse markers, everyone speaking in complete and orderly sentences and taking exactly one clean turn each. A learner fed only on such talk has never met the thing they will actually hear on the street — which helps explain the classic puzzle of the student who is fluent in the classroom and mute the moment a real conversation, with its interruptions and its trouble, begins.
What this means for language learning
The central lesson of conversation analysis for a learner is that talk is a coordinated, sequential skill, and the unit worth learning is bigger than the word — usually a whole utterance that does a job in a sequence. Knowing a language means being able to take a turn at the right moment, open and close an exchange, refuse without giving offence, and repair a breakdown before it becomes a silence; none of that appears in a list of isolated nouns. The practical consequence is to learn full utterances that carry conversational work — a hedged refusal, a clarification request, a topic change, a closing formula — in context. Items like these are ideally suited to active recall and spaced repetition, which is why the Taalhammer method has learners practise complete sentences rather than words: the sentence is where turn design, preference and repair actually live. For the wider study of language above the sentence, see the entry on discourse analysis; for the classroom philosophy that made real interaction the goal of teaching, the entry on communicative language teaching; and for the memory science of learning conversational routines whole, the entry on formulaic sequences.
Frequently asked questions
What is the difference between conversation analysis and discourse analysis?
Conversation analysis is one branch of the broader field of discourse analysis. Discourse analysis covers everything above the sentence, written and spoken — cohesion in texts, genre structure, register, classroom talk, political rhetoric. Conversation analysis, founded by the sociologists Sacks, Schegloff and Jefferson, studies specifically the mechanics of naturally occurring talk — turn-taking, adjacency pairs, openings, closings and repair — and does so only from recordings of real conversations, never from invented examples.
How can conversation analysis study language learning if learning happens in the mind?
This is exactly the question CA-for-SLA answers differently from mainstream second-language research. Rather than infer hidden mental change from test scores, it treats learning as something visible in interaction: a change over time in how a learner participates. Markee and Kasper described observable “learning behaviours” — word searches, checks of understanding, repair sequences in which meaning is jointly negotiated — that surface in real talk. On this view, learning is not the private growth of a grammar but becoming a more capable participant in conversation.
Why does repair matter so much for a language learner?
Because every real conversation runs into trouble, and competent speakers simply have the tools to fix it. Schegloff, Jefferson and Sacks showed that repair is a highly organized system with a strong preference for speakers correcting themselves. For a learner, a handful of repair moves — a clarification request, a candidate understanding, a self-repair formula — turns the frightening moment of not-understanding into a normal, manageable part of talking, and keeps a conversation alive through gaps that would otherwise end it.
Sources
- Harvey Sacks, Emanuel A. Schegloff, Gail Jefferson, “A Simplest Systematics for the Organization of Turn-Taking for Conversation”, Language 50(4), 1974.
- Emanuel A. Schegloff, Harvey Sacks, “Opening Up Closings”, Semiotica 8(4), 1973.
- Emanuel A. Schegloff, Gail Jefferson, Harvey Sacks, “The Preference for Self-Correction in the Organization of Repair in Conversation”, Language 53(2), 1977.
- Harvey Sacks, Lectures on Conversation (ed. Gail Jefferson), Blackwell, 1992.
- Alan Firth, Johannes Wagner, “On Discourse, Communication, and (Some) Fundamental Concepts in SLA Research”, The Modern Language Journal 81(3), 1997.
- Numa Markee, Conversation Analysis, Lawrence Erlbaum, 2000.
- Numa Markee, Gabriele Kasper, “Classroom Talks: An Introduction”, The Modern Language Journal 88(4), 2004.
- Gabriele Kasper, Johannes Wagner, “A Conversation-Analytic Approach to Second Language Acquisition”, in D. Atkinson (ed.), Alternative Approaches to Second Language Acquisition, Routledge, 2011.
- Jean Wong, “Applying Conversation Analysis in Applied Linguistics: Evaluating Dialogue in English as a Second Language Textbooks”, IRAL 40(1), 2002.