Conversation analysis (CA) is an empirical approach to studying the sequential organization of talk-in-interaction, originating in the ethnomethodological tradition of Sacks, Schegloff, and Jefferson. Talk is organized through a turn-taking system in which speakers project transition-relevance places and negotiate who speaks next. Adjacency pairs — paired actions like question-answer, greeting-greeting, or offer-acceptance — form the basic structural unit of conversational sequence. When trouble arises in speaking, hearing, or understanding, participants deploy systematic repair mechanisms. Preference organization describes how certain responses (e.g., acceptances) are structurally favored and delivered without delay, while dispreferred responses (e.g., rejections) are typically delayed, hedged, and accompanied by accounts.
Transcribe a five-minute natural conversation using CA conventions (marking pauses, overlaps, and intonation). Identify each turn-constructional unit and transition-relevance place. Map out the adjacency pairs and locate any repair sequences — note whether the repair is self-initiated or other-initiated, and whether it is self- or other-completed.
Conversation analysis (CA) begins with a deceptively simple observation: ordinary talk is not chaotic. When you study discourse analysis and speech act theory, you learn to categorize what utterances *do* — requests, assertions, promises. CA takes the next step and asks *how* those acts are sequentially organized: in what order do speakers produce them, how do they signal readiness to yield the floor, and how do they repair when something goes wrong? CA treats conversation as a social institution with discoverable rules, and those rules are built up moment by moment through participant action.
The centerpiece of CA is the turn-taking system. Conversation is fundamentally a sequenced exchange, and speakers must solve the coordination problem of who talks when. The solution rests on the concept of the transition-relevance place (TRP): a point in talk where the current unit of speech (a turn-constructional unit, or TCU) could possibly be complete. These are projectable — listeners can anticipate where the current turn will end based on syntax, prosody, and pragmatics. At a TRP, current speaker can select next speaker, any speaker can self-select, or the current speaker can continue. Most natural conversation produces remarkably few gaps and overlaps because participants are continuously projecting upcoming TRPs and timing their entry accordingly. Overlaps that do occur typically cluster precisely at TRPs — evidence of finely calibrated coordination, not disorder.
Adjacency pairs are the second building block. A first pair-part (an invitation, a question, a greeting) creates a normative expectation for a fitted second pair-part (acceptance, answer, return greeting). Absence of the second is not neutral — it is noticeably absent, and participants treat it as such. This conditional relevance is what gives adjacency pairs their social force. Within an adjacency pair, preference organization describes how certain second pair-parts are structurally favored. Acceptances, agreements, and grants tend to be delivered immediately and without qualification; rejections, disagreements, and refusals tend to be delayed, hedged, and accompanied by accounts. This asymmetry is observable in the transcript regardless of what speakers privately want — it is a feature of the conversation's surface organization.
Repair is the mechanism by which participants address trouble in speaking, hearing, or understanding. CA distinguishes who initiates repair (self or other) from who completes it (self or other), producing four combinations. Self-initiated, self-completed repair is most common and least disruptive; other-initiated, other-completed is rarest and most face-threatening. Systematic preferences for self-correction over other-correction reflect deep norms about interactional politeness and autonomy. Understanding repair sequences explains phenomena like the clarification request, the self-correction mid-sentence, and the reformulation — all are instances of the repair system at work.
What distinguishes CA from other discourse approaches is its insistence on data-first, bottom-up analysis. Rather than applying a theoretical framework to conversation, CA researchers transcribe recordings in exquisite detail — marking pauses to the tenth of a second, noting overlaps, pitch rises, and cut-offs — and then look for patterns in the sequential organization of the talk itself. Theoretical claims are grounded in what participants demonstrably orient to, not what analysts impute from outside the conversation. This empirical rigor makes CA findings unusually robust, and its analytical concepts — TRPs, adjacency pairs, repair, preference — are tools for describing what is structurally present in conversation, not interpretive glosses layered on top.