Conversation Analysis - The Deep Order of Ordinary Talk

Garfinkel's student Harvey Sacks took the same commitment — order at all points, evidence in the data — and applied it to the one activity everybody performs all day and nobody examines.

The result is the most technically rigorous branch of micro-sociology, and it produces findings you can verify yourself before finishing this lesson.

"Are you free on Saturday?"

Version one.

A: Are you free on Saturday? B: Yeah, sure.

Version two.

A: Are you free on Saturday? B: (0.8 second pause) Uh… Saturday. Um, I'd love to, but I think we might have my sister's thing? Let me — can I check and get back to you?

You knew the answer to version two before any word arrived that meant no. In fact you knew it during the pause.

Now notice how much information was carried by the machinery rather than the content. The delay. The filler. The repeat of the question word. The appreciation before the refusal. The account given without being requested. The hedged commitment at the end. And notice that B never actually said no — and that you would be baffled if A responded as though B had said yes.

Every competent speaker of every language does this, produces it accurately, reads it instantly, and cannot state the rules.

That is a system. It has structure, it is learned, it is followed with great precision, and it is invisible to the people using it.

The obvious question — how would you study something like that? — has an obvious-in-retrospect answer that nobody had implemented before the 1960s: record real talk and look at it very closely.

Not interviews about conversation. Not experiments. Not intuitions about what people would say. Recordings of actual talk — telephone calls, family dinners, courtrooms, doctors' consultations — transcribed to a level of detail that includes pauses measured in tenths of a second, in-breaths, overlaps, and rising or falling intonation.

That decision, made by Sacks with Emanuel Schegloff and Gail Jefferson, produced conversation analysis. Jefferson's transcription system is still the standard, and looking at a page of it is startling: the amount of order in five seconds of ordinary talk is more than anyone expected.

The idea, in plain words

The turn-taking system

The founding paper — Sacks, Schegloff and Jefferson (1974) — set out to explain a puzzle: how do speakers manage to talk one at a time, with very little overlap and very little silence, without any external coordination?

The answer is a system built from two components.

Turn-constructional units. A turn is built from units — a word, a phrase, a clause, a sentence — whose possible completion is projectable by the hearer before it arrives. This projectability is the key: listeners are continuously forecasting where the current unit will end.

Transition-relevance places. At the possible completion of a unit, a transition becomes relevant, and a set of rules operates in order: the current speaker may select the next speaker (by naming, by gaze, by asking a question); if not, another may self-select; if not, the current speaker may continue.

The consequences are remarkable and measurable. Cross-linguistic research — including a well-known study by Stivers and colleagues across ten languages from five continents — found the modal gap between turns clustering near around 200 milliseconds, with strikingly similar distributions across languages as different as Danish, Japanese, ǂĀkhoe Haiǁom and Tzeltal. Two hundred milliseconds is faster than the time required to plan an utterance, which means responders must be preparing their turn while still listening, using projection.

That is an extraordinary fact about human beings, and it was discovered by people transcribing telephone calls.

Adjacency pairs and preference

Adjacency pairs. Certain utterance types come in ordered pairs: question–answer, greeting–greeting, invitation–acceptance/decline, offer–accept/reject, request–grant/refuse, accusation–denial/admission.

The first part makes the second conditionally relevant: its absence is not merely nothing happening, it is a noticeable absence. Ask a question and receive silence, and the silence is the person not answering — an event with consequences. This is why the machinery has moral force: a first pair part creates an obligation.

Preference organisation. The concept that explains the Saturday example, and it is one of the most useful in the field.

Second pair parts are not equal. One alternative is preferred — meaning structurally favoured, not psychologically desired — and the other dispreferred. Preferred responses (acceptance, agreement, granting) are typically immediate, brief and unmitigated. Dispreferred responses (refusal, disagreement, rejection) are systematically marked by:

Delay — a pause, often several tenths of a second. A preface — "well," "uh," "I mean." An appreciation — "I'd love to," "that's very kind." An account — a reason, offered without being asked. Mitigation and hedging — "I think," "might," "possibly." And frequently, no explicit refusal at all — the refusal is conveyed and never stated.

This is why you knew at 0.8 seconds. The delay itself is a signal, produced systematically and read reliably, and its function is to allow the first speaker to withdraw the request before it is formally refused — which is Goffman's face-work (2.4.2) running inside the grammar of conversation.

Repair. How trouble in speaking, hearing or understanding is dealt with. CA's robust finding is a preference for self-initiated self-repair: speakers are given the opportunity to correct themselves, and others' corrections are delayed, softened and often framed as candidate understandings ("you mean the Tuesday one?"). Again: face-work, embedded in machinery.

CA's most productive application is the comparison of ordinary conversation with institutional talk — and the method is beautifully simple. Take the turn-taking system of ordinary conversation as the baseline, and describe how a setting modifies it. The modifications are the institution.

The courtroom. Turn types are pre-allocated: counsel asks, witness answers. The witness may not ask questions. The witness may not initiate topics. Counsel may pursue an answer that ordinary conversation would let pass. The result is that one party controls the agenda, the pacing and what counts as a complete answer — and CA researchers have shown in detail how this asymmetry does work, including in cross-examination of complainants, where question design constrains what a witness can say and how it will be heard.

The classroom. The classic finding is the IRE sequence: teacher Initiates, pupil Responds, teacher Evaluates. Notice that the third turn is what makes it not a conversation: in ordinary talk, answering a question does not entitle the questioner to assess your answer. The IRE sequence is a machine for producing and evaluating knowledge displays, and children learn it early (2.3.5's hidden curriculum, visible in a transcript).

The medical consultation. Extensively studied. Findings include the asymmetry of question rights, the effect of how a doctor's opening question is designed on what patients report, and the systematic difficulty patients have introducing a concern that does not fit the sequence already underway.

News interviews. Interviewers do not respond to answers as a conversationalist would — no assessments, no "really?" — which maintains a display of neutrality while the questions do considerable work.

The general point is the payoff: institutions are, in significant part, systematic modifications of the turn-taking machinery — and those modifications are precisely where the distribution of speaking rights, and therefore of power, becomes observable. This is 1.3.2's macro appearing in the micro, in a form you can transcribe and count.

Go deeper

Membership categorisation

Sacks' other major contribution, from an analysis of a child's story: "The baby cried. The mommy picked it up."

Every hearer understands that the mommy is the baby's mother and that she picked it up because it cried. Nothing in the sentences says either. Sacks' analysis: we hear "baby" and "mommy" as belonging to a single membership categorisation device — the family — and we hear the actions as category-bound activities appropriate to those categories.

The concept matters far beyond children's stories, because categorising a person is an action with consequences. Describing someone as a "youth," a "migrant," a "patient" or a "housewife" imports a whole set of expected activities and obligations, without anything being argued. This is 2.1.5's naming and 2.4.1's typifications, made analysable in actual talk — and it is a precise tool for studying how categories do political work in media, court and everyday conversation.

What CA gets right

Its evidential discipline is exemplary. Claims must be demonstrable in the data, and — the crucial move — participants' own displayed understandings are the primary evidence. If a speaker treats a prior turn as a refusal, that is evidence it was hearably a refusal. Not the analyst's intuition: the next turn is the proof.

It found order where nobody expected it. Ordinary talk looks messy and turns out to be organised at a resolution of tenths of a second.

And it is cumulative. Unusually for sociology, findings replicate, build and travel across languages.

The criticisms

It brackets context and power. The standard objection, inherited from ethnomethodology (2.4.3): CA declines to invoke class, gender or caste unless participants demonstrably orient to them in the data. Critics — including feminist and critical-discourse scholars — argue this makes structural inequality invisible by methodological fiat.

The reply is serious: importing context lets analysts find whatever they expected, and the discipline of showing relevance in the data is what makes CA's findings trustworthy. Both positions have integrity, and the productive middle exists — the institutional-talk literature above is the study of power, conducted with CA's evidential standards intact.

It can be narrow. Extraordinary precision about a small object. A fair charge, and the institutional applications are the answer.

And its transcripts are laborious — an hour of talk can take many hours to transcribe — which limits sample sizes and makes CA a craft as much as a method.

The practical payoff

Unusually for micro-sociology, CA has direct applications.

Communication training in medicine, based on findings about how question design shapes what patients disclose. Emergency call analysis, where the sequential organisation of a call has life-and-death consequences and CA research has informed dispatcher training. The design of conversational technology, where systems fail in ways CA predicts — they cannot project turn completion, so they interrupt or leave gaps that feel wrong by hundreds of milliseconds. Interpreting and mediation, where turn allocation determines who is heard.

And a personal one: knowing preference organisation makes you a better listener. If you can hear a dispreferred second beginning, you can withdraw a request before anyone has to refuse it — which is exactly what the machinery was built to allow.

"Preferred means what the speaker wants." It means structurally favoured — brief, immediate, unmarked. A person may be delighted to refuse and will still produce the dispreferred format.

"Conversation analysis studies what people say." It studies the organisation of talk: sequence, timing, turn design, repair. Content is secondary.

"Silence means nothing." A silence after a first pair part is a noticeable absence with consequences, and its length is meaningful at the scale of tenths of a second.

"The 200-millisecond finding means everyone is the same." Distributions are strikingly similar across languages; there is real variation by language, culture, setting and question type. The similarity is the surprise, not the uniformity.

"CA ignores power." It declines to assume it, and studies it through demonstrable modifications of the machinery — which is what the institutional-talk literature does.

"It's just linguistics." Its object is social organisation, not language structure. The unit is the interaction.

Why it matters

It proves Garfinkel's claim with hard evidence. Order at all points, produced locally by participants, demonstrable in the data. If you doubted that micro-sociology could be rigorous, this is the answer.

It shows where face-work lives. Preference organisation and repair preference are Goffman's cooperative dignity-maintenance, built into the grammar of conversation itself. That is a striking convergence between two traditions that look quite different.

It makes institutional power observable and countable. Who may ask, who may only answer, who may evaluate, who may initiate a topic, who is interrupted and by whom. These are measurable, and they are the same asymmetries Part 8 studies at population scale.

And it gives you something you can check today. Listen for the pause before a refusal. Watch a teacher take the third turn. Notice a doctor's opening question. Every claim in this lesson is verifiable by anyone with functioning ears and an afternoon.

Conversation analysis studies the organisation of talk-in-interaction using recordings and fine-grained transcripts, on Sacks' principle that there is order at all points.

The turn-taking system (Sacks, Schegloff and Jefferson): turns are built from turn-constructional units whose completion is projectable, creating transition-relevance places governed by ordered allocation rules. Cross-linguistic research finds inter-turn gaps clustering near 200 milliseconds — faster than utterance planning, so responses are prepared through projection.

Adjacency pairs make second parts conditionally relevant, so absence is noticeable. Preference organisation: preferred seconds are immediate and brief; dispreferred seconds are delayed, prefaced, appreciative, accounted for, hedged, and often never explicitly stated — which is why a pause conveys the answer.

Repair shows a preference for self-initiated self-repair. Both preference and repair are face-work built into the machinery.

Institutional talk is analysed as systematic modification of ordinary turn-taking: courtroom pre-allocation, the classroom IRE sequence, medical consultations, news interviews — making the distribution of speaking rights observable.

Membership categorisation devices and category-bound activities show how categorising a person imports expectations without argument.

Criticisms: brackets context and power by methodological choice, is narrow in object, and is laborious — with the institutional literature as the productive answer.

Conversation analysis — the study of the sequential organisation of talk-in-interaction using naturally occurring recordings.

Order at all points — Sacks' principle that no detail of interaction is too small to be systematically organised.

Turn-constructional unit — a unit of talk whose possible completion is projectable by hearers.

Transition-relevance place — the point at which speaker change becomes relevant.

Adjacency pair — a paired sequence in which a first part makes a particular second part conditionally relevant.

Conditional relevance — the property whereby the absence of an expected second part is itself noticeable and consequential.

Preference organisation — the structural (not psychological) favouring of one second-pair alternative, marked by contrasting turn designs.

Dispreferred turn shape — delay, preface, appreciation, account, mitigation, and frequently no explicit refusal.

Repair — the organised handling of trouble in speaking, hearing or understanding, with a preference for self-repair.

Institutional talk — talk in institutional settings, analysed as systematic modification of ordinary turn-taking.

IRE sequence — the classroom pattern of teacher Initiation, pupil Response, teacher Evaluation.

Membership categorisation device — a collection of categories heard as belonging together, carrying category-bound activities.

Jefferson transcription — the standard notation recording pauses, overlaps, intonation and breathing.