Repair Sequence¶

Resolve any conversational trouble by locating it in a small grid — trouble source, initiation, completion, crossed by who repairs (self or other) and where in the turn-sequence — and read the likely path to resolution off the preference for self-repair.

Core Idea¶

A repair sequence is the orderly local routine by which participants in conversation identify and resolve a problem of hearing, speaking, or understanding — a mishearing, an ambiguous reference, a wrong word, an apparent misunderstanding — without halting the talk. The framework was developed by Emanuel Schegloff, Gail Jefferson, and Harvey Sacks and documented across diverse languages and interaction settings. The sequence has a small inventory of structural positions: the trouble source (the turn or item that becomes problematic), the repair initiation (the turn that flags the trouble, ranging from a naked "huh?" to an explicit "do you mean X or Y?"), and the repair completion (the turn that resolves the trouble). Two parameters cross-classify the organizational space: the agency parameter (whether initiation and completion are performed by the same speaker who produced the trouble source — self — or by another participant — other) and the position parameter (whether repair is initiated in the same turn as the trouble source, in the transition space after it, in the next turn, or in a later third position). The cross-classification yields a typology that empirically covers most conversational repair across languages.

The pattern is a local feedback loop on mutual intelligibility: conversation maintains shared understanding turn-by-turn through this mechanism rather than relying on global error-correction after the fact. Schegloff, Jefferson, and Sacks documented a systematic preference asymmetry — self-initiated self-repair is preferred over other-initiated self-repair, which in turn is preferred over other-initiated other-repair — reflecting conversational organization's design to preserve speaker face and minimize overt correction. The empirical finding holds across English, Tzeltal, Korean, and sign languages, establishing it as a structural regularity of conversational organization rather than a cultural convention. A simple instance: "She said the meeting's moved to three" / "Three today, or three tomorrow?" (other-initiation) / "Three tomorrow" (self-repair completion) — the trouble is a temporal-reference ambiguity, B's next-turn check surfaces it, A resolves it on the third turn, and the same two-or-three-turn structure recurs across sound-trouble, reference-trouble, and projection-trouble.

Structural Signature¶

Sig role-phrases:

the trouble source — the prior-turn item that becomes problematic for hearing, speaking, or understanding
the repair initiation — the turn that flags the trouble, from a bare "huh?" to an explicit "do you mean X or Y?"
the repair completion — the turn that resolves the trouble, closing the loop without halting the talk
the agency parameter — whether initiation and completion are done by self (the trouble-source speaker) or by other, separating who notices from who fixes
the position parameter — whether repair falls in the same turn, the transition space, the next turn, or a later third position
the preference asymmetry — the documented ordering self-initiated self-repair > other-initiated self-repair > other-initiated other-repair, which cells conversation reaches for first
the position-predicts-completion projection — from where trouble is flagged, the likely path to resolution follows, with departures from the ordering marked as doing extra interactional work
the intelligibility-maintenance function — the local turn-by-turn feedback loop that keeps conversation from accumulating uncorrected divergence

What It Is Not¶

Not correction in the loose sense. Not every replacement of one expression by another is a repair, and not every repair alters a proposition: a word-search, a same-turn restart, and a bare "huh?" all count, while many substantive corrections fall outside the organization entirely. The concept ranges over the repair organization — trouble source, initiation, completion — not over error-fixing in general.
Not the same as a clarification question. Clarification is one repair operation, but the organization also covers self-corrections, mid-word restarts, and word-searches that involve no question at all. Treating "repair" as "asking for clarification" mistakes a single move for the whole position-and-agency system.
Not a politeness convention. The preference asymmetry — self-initiated self-repair favored over other-initiated self-repair, and that over other-initiated other-repair — holds across English, Tzeltal, Korean, and sign languages, which marks it as a structural regularity of conversational turn-taking rather than a cultural manner. A population that systematically violated the ordering would be the notable finding, not the norm.
Not engineering error-correction. Hamming codes and checksums are pre-arranged redundancy schemes applied to a signal; a repair sequence is a negotiated, multi-party turn sequence flagged and resolved in real time. The two are not the same mechanism — one is designed redundancy, the other is sequentially organized interactional work.
Not feedback in general. The repair sequence instantiates a local feedback loop on intelligibility (factoring into feedback + error_correction + turn_taking / protocol), but its distinctive cargo — the self/other agency parameter, the same-turn/transition-space/next-turn/third-position grid, the documented preference asymmetry, and its reading as face-preservation — presupposes negotiated, sequentially organized talk. An ATC readback or a chatbot prompt borrows the diagnostic vocabulary while the underlying mechanism is the general feedback-and-error-correction one, not a repair-sequence-specific pattern.

Scope of Application¶

The repair sequence lives across the subareas of conversation analysis and the sociolinguistics built on it; its reach is one substrate — same-channel, time-pressured, sequentially organized talk-in-interaction — and the formalized borrowings (ATC readbacks, support scripts, chatbot clarification turns) instantiate the general feedback + error_correction + turn_taking / protocol parents rather than the repair-sequence typology itself.

Ordinary conversation — the home turf, where the trouble-source / initiation / completion skeleton and the agency-by-position grid were first documented and the preference asymmetry established.
Institutional interaction — classroom, medical, courtroom, and service talk, where the same typology describes how participants flag and resolve trouble within asymmetric, task-structured exchanges.
Sign-language interaction — repair organization documented in signed conversation, one of the cross-modality settings that establishes the structure as turn-taking-general rather than speech-specific.
Cross-linguistic and sociolinguistic studies of repair — the comparative work across English, Tzeltal, Korean, and other languages that treats the preference ordering as a structural regularity of conversational organization.

Clarity¶

Naming the repair sequence separates repair from correction in the loose sense, a conflation that otherwise makes conversational trouble look like undifferentiated error-fixing. Not every replacement of one expression by another is a repair, and not every repair alters a proposition: a word-search, a same-turn restart, or a "huh?" all count, while many substantive corrections do not fit the organization at all. By fixing a small inventory of positions — trouble source, initiation, completion — the concept turns a blur of conversational micro-events into a structure with nameable slots, and lets the analyst ask the operative question of any troubled stretch of talk: which position is this, and what does that predict about how the trouble will be resolved?

The sharpest distinction it draws is between two questions ordinary talk fuses: who notices the trouble? and who fixes it? The agency parameter (self vs. other) and the position parameter (same-turn, transition-space, next-turn, third-position) cross-classify these independently, so that "B flags A's ambiguity but A resolves it" becomes a specific, locatable type (other-initiated self-repair) rather than a vague instance of clearing something up. That decomposition is what makes the field's central empirical finding even statable — the preference asymmetry by which self-initiated self-repair is favored over other-initiated self-repair, and that over other-initiated other-repair — because the preference is a claim about which cells of the typology conversation reaches for first. Without the position-and-agency grid, that regularity, and its reading as a face-preserving design feature of conversational organization, could not be seen at all.

Manages Complexity¶

The surface phenomena of conversational trouble are bewilderingly varied: a "huh?", a mid-word restart, a stalled word-search, a "do you mean X or Y?", a "no, I meant tomorrow", a delayed "wait, you're saying we should cancel?" — each looks like its own kind of event, and an analyst cataloging them one by one faces an open-ended heap of micro-behaviors with no obvious order. The repair sequence compresses that heap by resolving every instance into the same three-position skeleton — trouble source, repair initiation, repair completion — and then locating each instance in a two-parameter grid laid over those positions: the agency parameter (is the initiation, and the completion, done by self or by other?) and the position parameter (does repair fall in the same turn as the trouble, the transition space, the next turn, or a later third position?). The cross-classification of agency by position is a small finite set of cells, and it empirically covers most conversational repair across languages. So the analyst stops asking, of each troubled stretch, "what sort of thing is happening here?" and instead reads off two coordinates — who, and where — that fix the repair's type. A "huh?" is other-initiation in the next turn; a mid-word self-correction is self-initiated self-repair in the same turn; "do you mean X?" followed by "yes, X" is other-initiated self-repair. The unbounded variety of micro-events reduces to position in a grid.

What the grid then lets the analyst read off — the move that turns description into prediction — is the field's central regularity, which is itself a statement about which cells conversation reaches for first. The documented preference asymmetry (self-initiated self-repair preferred over other-initiated self-repair, and that over other-initiated other-repair) is only statable because the agency and position parameters separate "who notices the trouble?" from "who fixes it?"; with those held apart, the analyst predicts from the position of an initiation what kind of completion is likely to follow, and reads the whole ordering as a face-preserving design feature of conversational organization rather than as a pile of unrelated habits. Because the regularity holds across English, Tzeltal, Korean, and sign languages, the analyst treats it as structural rather than re-deriving it per culture. The sprawling question "how do speakers fix all the different things that go wrong in talk?" reduces to: identify the trouble source, read the initiation's agency and position off the grid, and let the preference ordering project the likely path to completion.

Abstract Reasoning¶

The framework's basic move is a two-coordinate classification that fixes a repair's type from two independent reads rather than a holistic judgment of "what is happening." For any troubled stretch of talk, the analyst locates the trouble source and then reads off agency (is the initiation, and the completion, done by self or by other?) and position (same turn, transition space, next turn, or later third position?), and the cross-classification names the type: a "huh?" is other-initiation in the next turn; a mid-word self-correction is self-initiated self-repair in the same turn; "do you mean X?" answered "yes, X" is other-initiated self-repair. The decisive sub-move is holding apart two questions ordinary talk fuses — who notices the trouble? and who fixes it? — because the agency and position parameters cross-classify them independently, and that separation is what makes a vague "they cleared it up" into a specific, locatable cell.

That grid converts description into prediction in two linked ways. First, the position of an initiation predicts the kind of completion likely to follow: from where in the turn-sequence trouble is flagged, the analyst projects the path to resolution, so an other-initiation in the next turn projects a self-repair completion in the following turn (as in the temporal-reference example). Second, the field's central regularity — the preference asymmetry by which self-initiated self-repair is favored over other-initiated self-repair, and that over other-initiated other-repair — is itself a statement about which cells conversation reaches for first, and the analyst uses it as a projection rule: given trouble, predict that participants will route toward the self-initiated self-repair end before the other-initiated other-repair end. This makes the ordering readable as a face-preserving design feature of conversational organization, and it gives a diagnostic edge to deviations — a stretch that goes straight to other-initiated other-repair (overt correction of another's turn) is inferred as marked, doing interactional work (disaffiliation, asymmetry of authority) beyond mere trouble-fixing, precisely because it departs from the preferred ordering.

A boundary move separates repair from correction and fixes what the inferences range over: not every replacement of one expression by another is a repair, and not every repair alters a proposition — word-searches, same-turn restarts, and a bare "huh?" all count, while many substantive corrections fall outside the organization entirely. So the analyst reasons over the repair organization, not over error-fixing in general. A generality move treats the regularity as structural rather than cultural: because the preference asymmetry and the position-agency typology hold across English, Tzeltal, Korean, and sign languages, the analyst applies the grid and the projection rule to a new conversation without re-deriving them per culture — and treats a population that systematically violates the ordering as the notable finding. These inferences presuppose the substrate of human conversational turn-taking — same-channel, time-pressured, sequentially organized talk — so the position/agency reasoning and the preference projection apply to talk-in-interaction and its formalized descendants (readback-hearback routines, designed clarification turns), while what survives stripping that substrate is generic loop-closing rather than the repair-sequence-specific typology.

Knowledge Transfer¶

Within conversation analysis the repair sequence transfers as full mechanism. The three-position skeleton (trouble source, initiation, completion), the agency-by-position grid, the preference asymmetry, and the position-predicts-completion projection all carry intact across the settings and languages the framework was built on — and the central finding's robustness across English, Tzeltal, Korean, and sign languages is precisely the claim that the machinery is structural to human conversational turn-taking rather than a cultural convention. So a CA analyst applies the grid and the preference ordering to a new conversation without re-deriving them per culture, and treats a population that systematically violates the ordering as the notable result. This holds across the field's subareas — ordinary talk, institutional interaction, sign-language interaction, sociolinguistic studies of repair — because all share the substrate: same-channel, time-pressured, sequentially organized talk.

Beyond that substrate the case is (B): the repair sequence factors cleanly into more general mechanisms that travel, while its own sequence-organizational machinery does not. The structural commitment underneath is a local feedback loop on intelligibility — fix divergence turn-by-turn, close to the trouble, rather than relying on global after-the-fact error-correction — and lifted out of conversation that decomposes into feedback (close the loop fast to prevent divergence), error_correction (local fixes are cheaper than global ones), and turn_taking / protocol (the sequential channel discipline). Those parents really do co-instantiate across domains: air-traffic readback-hearback routines are a formalized feedback loop with a designed same-turn other-initiation; customer-support and incident-response clarification scripts are designed self-initiated other-repair; human-AI dialogue borrows the typology to decide when a chatbot should ask a clarifying question. In each, what recurs at the structural level is the feedback-plus-error-correction-plus-protocol combination, not a distinctive repair-sequence mechanism. The home-bound cargo is everything the typology adds: the self/other agency parameter, the same-turn/transition-space/next-turn/third-position grid, the documented preference asymmetry, and the reading of that asymmetry as face-preservation — all of which presuppose negotiated, multi-party, sequentially organized talk. These borrowings are worth marking carefully: the diagnostic vocabulary (trouble source, initiation, completion) and the design prescription (place an explicit clarification turn close to the trouble source) genuinely do transport as a useful checklist, but what they instantiate is the parent combination, so invoking "a repair sequence" for an ATC readback or a chatbot prompt borrows the vocabulary while the underlying mechanism is the general feedback/error-correction one. The honest cross-domain move is to carry those parents — and to note the adjacent, still-unconfirmed pattern of a local-correction preference (correction applied close in time and space to the trouble being cheaper and more reliable, across distributed systems and organizations as well as talk) — rather than the conversation-specific concept, whose machinery does real work only on talk-in-interaction (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

The defining study is Schegloff, Jefferson, and Sacks's 1977 paper "The Preference for Self-Correction in the Organization of Repair in Conversation" (Language), which established both the position-and-agency typology and the preference asymmetry from recorded ordinary talk. Take a compact illustrative exchange of the kind the paper analyzes. A: "Have you seen Bill lately?" B: "Who?" A: "Bill — from accounting." Here A's turn contains the problematic reference; B's "Who?" is a minimal next-turn flag that something needs fixing; and A's "Bill, from accounting" resolves it. Contrast a same-turn case: A: "So they moved in around, uh— around sixty-three." Here A both produces and immediately mends the word-search within one turn. The paper's central finding is that talk is organized to favor exactly this self-mending: speakers get the chance to correct themselves before others do, and other-initiations like "Who?" overwhelmingly still hand the fixing back to the original speaker.

Mapped back: In the first exchange, A's ambiguous "Bill" is the trouble source, B's "Who?" is the repair initiation, and A's "Bill, from accounting" is the repair completion. Reading who flags versus who fixes is the agency parameter — B initiates (other), A completes (self) — while "Who?" landing in the following turn is the position parameter (next-turn). That B flags but A resolves, rather than B correcting outright, exemplifies the preference asymmetry favoring self-repair.

Applied / In Practice¶

Dingemanse, Torreira, and Enfield's cross-linguistic field research (published 2013–2015) deployed the repair framework on natural conversation recorded across a genetically and geographically diverse set of languages — including field sites such as Cha'palaa in Ecuador, Murrinh-Patha in northern Australia, Siwu in Ghana, and Lao, alongside Icelandic, Italian, and others. Coding troubled sequences with the same trouble-source/initiation/completion apparatus, they found that every language possessed a short, low-effort, questioning "huh?"-like interjection used to open other-initiated repair in the next turn, and that the broader system of other-initiated repair recurred with the same structural options everywhere sampled. The work turned the preference organization from an English-based claim into a documented candidate universal of human interaction, doing real comparative-linguistic work on the design of conversation itself.

Mapped back: The "huh?"-type interjection each language supplies is a the repair initiation in its purest minimal form, and its consistent placement is the position parameter set to next-turn. It is other-launched, so it fixes the agency parameter at other-initiation, and the finding that it hands resolution back to the original speaker upholds the preference asymmetry. Establishing all this across unrelated languages is the framework's generality claim made empirical.

Structural Tensions¶

T1: Preference ordering as default versus deviation as meaning (the regularity whose breach is the point). The preference asymmetry — self-initiated self-repair over other-initiated self-repair over other-initiated other-repair — is the framework's central regularity, and it doubles as a projection rule: given trouble, expect conversation to route toward the self-repair end. But much of the framework's interpretive power comes from departures: a stretch that goes straight to other-initiated other-repair (overt correction of another's turn) is read as marked, doing interactional work — disaffiliation, an assertion of authority — precisely because it violates the preferred ordering. The tension is that the ordering functions simultaneously as a baseline to predict from and as a norm whose breach carries the richest meaning, so an analyst who treats it only as a statistical regularity misses the marked cases, while one who reads every deviation as meaningful risks over-interpreting ordinary variation. Diagnostic: Is this departure from the preferred ordering doing interactional work (disaffiliation, authority asymmetry), or is it an unremarkable repair the projection rule should simply absorb?

T2: Grid coverage versus residual trouble (a typology that covers most, not all). The agency-by-position grid is a small finite set of cells that "empirically covers most conversational repair across languages" — that near-exhaustive coverage is what turns a heap of micro-events into a readable classification. But "most" is not "all," and the framework itself insists that not every replacement is a repair: substantive corrections fall outside the organization, and multi-party overlap, embodied repair in face-to-face signing, and gesture-borne trouble strain the four-position, self/other scheme. The tension is that the grid's clean coverage is exactly what tempts an analyst to force a residual case into a cell rather than recognize it as outside the repair organization, so the typology's power to classify is also a pressure to over-classify. Diagnostic: Does this troubled stretch genuinely fit the position-agency grid, or is it a correction or embodied move outside the repair organization being shoehorned into a cell?

T3: Structural universality versus institutional redistribution (the machinery is constant, the cells are not). The robustness of the preference ordering across English, Tzeltal, Korean, and sign languages establishes it as structural to human turn-taking, which is what licenses applying the grid to a new conversation without re-deriving it per culture — a genuine economy. But institutional and authority-asymmetric settings (courtroom, classroom, clinic) systematically shift which cells get used: a teacher other-corrects a pupil, an interpreter's role redistributes who may repair whom, and the face-preserving default is overridden by the setting's power structure. The tension is that the machinery's universality can be mistaken for uniformity of use, when the same grid is populated very differently once authority enters — so treating the ordering as fully context-free obscures how setting redistributes the preferences. Diagnostic: Is the preference ordering itself operating universally here, or is an institutional authority asymmetry redistributing which cells are used while the grid stays constant?

T4: Face-preservation reading versus information-access reading (two explanations of one asymmetry). The entry reads the preference for self-repair as a face-preserving design feature of conversational organization — speakers get to mend their own trouble before others do. But the same asymmetry admits an efficiency explanation that is not about face at all: the trouble-source speaker has privileged access to what they meant, so self-repair is simply the fastest and most reliable route to resolution regardless of any face concern. The two readings usually coincide, which is why the regularity underdetermines them, but they diverge in cases where face and information access point different ways — where the other participant plainly knows the correct form and the speaker does not. The tension is that a single robust regularity is made to bear an interpretive commitment (face) that an equally good functional account (information access) would not require. Diagnostic: When face-preservation and repair-efficiency point in different directions, does conversation follow the face-preserving ordering or route repair to whoever has the information?

T5: Autonomy versus reduction (a conversation-analytic typology or a local feedback loop on intelligibility). The repair sequence is a named CA framework with heavy talk-specific cargo — the self/other agency parameter, the four-position grid, the documented preference asymmetry, its reading as face-preservation — all presupposing negotiated, multi-party, sequentially organized talk, and within conversation analysis it travels as full mechanism across languages and settings. But the structure underneath is a local feedback loop on intelligibility that factors cleanly into feedback, error_correction, and turn_taking/protocol, and those parents co-instantiate in air-traffic readback-hearback, support clarification scripts, and human-AI dialogue. Those borrowings take the diagnostic vocabulary (trouble source, initiation, completion) and the design prescription (place a clarification turn close to the trouble) while the underlying mechanism is the general feedback-and-error-correction one — invoking "a repair sequence" for an ATC readback borrows the vocabulary, not the typology. Diagnostic: Resolve toward feedback/error_correction/turn_taking when the substrate is any loop-closing channel; toward the repair sequence only for sequentially organized talk-in-interaction with its self/other agency and turn-position grid.

Structural–Framed Character¶

The repair sequence sits toward the structural end of the spectrum — best read as mixed-structural: a genuine, evaluatively neutral regularity of human interaction wearing heavy conversation-analytic vocabulary, with one nuance that keeps it a hair nearer the middle than a purely physical mechanism like isostasy. On evaluative_weight its credentials are clean and structural: the typology describes, it does not convict — classifying a stretch of talk as "other-initiated self-repair in the next turn" passes no verdict, praises and blames no one, and even the marked cases (a bare other-correction reading as disaffiliation) are read as doing interactional work, not as errors to be indicted. On human_practice_bound the reading is more delicate but still leans structural in the way that matters. The concept is substrate-bound to human talk-in-interaction — remove conversation and there is no trouble source to repair — but crucially the regularity itself runs observer-free within that substrate: people flag and mend conversational trouble, and route toward self-repair, whether or not any analyst is coding them, and the entry insists (and the cross-linguistic work confirms) that this is a structural regularity of turn-taking, not a cultural convention. Unlike a fallacy, which is constituted by a normative practice and dissolves into a mere social act once the evaluative frame is removed, the repair sequence is a discovered behavioral regularity that would persist unchanged if the discipline studying it vanished. That is the mark of a real mechanism, not a practice-constituted verdict — though the substrate being human behavior rather than lithospheres or bodies is what tethers it more tightly to a "practice" than isostasy or Rensch's rule.

On institutional_origin it splits in a revealing way: the object — the preference asymmetry, the "huh?"-type other-initiation attested from Cha'palaa to Icelandic — is a candidate universal of human interaction, a fact discovered rather than invented, which is a strong structural credential; the apparatus (the three-position skeleton, the agency-by-position grid) is a scholarly coding construct, which is the constructed part. On vocab_travels it points framed: the operative vocabulary — trouble source, repair initiation, repair completion, self/other agency, transition-space and third position, the preference asymmetry — is pinned to talk-in-interaction and does not float free; off the substrate only the general feedback/error-correction terms survive, which is why an ATC readback keeps the checklist but loses the typology. And import_vs_recognize is bimodal: within conversation analysis (ordinary talk, institutional interaction, signed conversation, cross-linguistic studies) reuse is genuine mechanism-recognition across the whole human-talk substrate, whereas off-substrate reuse is import — readback-hearback routines and chatbot clarification turns instantiate the parent primes, borrowing the diagnostic vocabulary rather than recognizing a repair-sequence-specific mechanism.

The portable structural skeleton is a local feedback loop on intelligibility — close divergence turn-by-turn, near the trouble, rather than by global after-the-fact error-correction. That skeleton genuinely travels — to air-traffic control, incident-response scripts, human-AI dialogue — but it does not make "the repair sequence" itself portable, because it is exactly what the entry instantiates from its parent primes, feedback, error_correction, and turn_taking/protocol, not what makes the CA typology travel: the cross-domain reach belongs to that parent combination, while the repair sequence's own distinctive cargo — the self/other agency parameter, the four-position grid, the documented preference asymmetry, the face-preservation reading — stays home in sequentially organized talk. Its character: structural in skeleton — an evaluatively neutral, cross-linguistically documented, recognized-not-imported regularity of human turn-taking that runs whether or not it is studied — but stated in a conversation-analytic vocabulary so substrate-bound that off talk-in-interaction only its feedback-and-error-correction parents survive, leaving it mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section decides why the repair sequence is a domain-specific abstraction and not a prime, and it also carries the case for why it is domain-specific — so it is worth separating the loop-closing skeleton underneath from the conversation-analytic typology that names it.

What is skeletal (could lift toward a cross-domain prime). Strip the talk and a thin relational structure survives: a local feedback loop on intelligibility — divergence is flagged and fixed turn-by-turn, close to the trouble, rather than by global after-the-fact error-correction over a sequential channel. The portable pieces are abstract — a channel that can carry trouble, a fast local flag, a fix that closes the loop before divergence accumulates, and a sequential discipline governing who acts when. That skeleton genuinely travels, which is exactly why the entry factors cleanly into feedback (close the loop fast to prevent divergence), error_correction (local fixes are cheaper than global ones), and turn_taking/protocol (the sequential channel discipline) — parents that co-instantiate in air-traffic readback-hearback, incident-response scripts, and human-AI dialogue. But it is the skeleton the sequence shares with those parents, not what makes the repair sequence distinctive.

What is domain-bound. Everything that makes the concept the repair sequence in particular is conversation-analytic furniture presupposing negotiated, multi-party, sequentially organized talk: the three-position skeleton (trouble source, repair initiation, repair completion); the self/other agency parameter separating who notices from who fixes; the position parameter (same turn, transition space, next turn, third position); the documented preference asymmetry (self-initiated self-repair > other-initiated self-repair > other-initiated other-repair); and its reading as a face-preserving design feature, with departures read as interactional work (disaffiliation, authority asymmetry). These are the worked vocabulary, instruments, and empirical cases (the Schegloff–Jefferson–Sacks typology, the cross-linguistic "huh?" studies). The decisive test: remove human conversational turn-taking and there is no trouble source to repair, no agency-by-position grid, no face to preserve; off the substrate only the general feedback-and-error-correction terms survive, which is why an ATC readback keeps the checklist (place a clarification turn close to the trouble source) but loses the typology.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose transfer is recognition of the same mechanism, not analogy. The repair sequence's transfer is bimodal. Within conversation analysis it travels as full mechanism across ordinary talk, institutional interaction, signed conversation, and cross-linguistic studies — the grid and the preference ordering apply to any new conversation without re-derivation, and their robustness across English, Tzeltal, Korean, and sign languages is the claim that the machinery is structural to human turn-taking. Beyond talk-in-interaction — readback-hearback routines, support clarification scripts, chatbot clarifying turns — the reuse is import: those settings instantiate the parent primes, borrowing the diagnostic vocabulary (trouble source, initiation, completion) while the underlying mechanism is the general feedback/error-correction one, not a repair-sequence-specific pattern. So the cross-domain reach belongs to feedback, error_correction, and turn_taking/protocol; the repair sequence clears the domain-specific bar for conversation analysis, while its portable content is already carried, in more general form, by the parent combination it instantiates.

Relationships to Other Abstractions¶

Current abstraction Repair Sequence Domain-specific

Parents (4) — more general patterns this builds on

Repair Sequence is a kind of Communication Repair Prime

A repair sequence is the conversation-analysis species of communication repair, specialized by an agency-by-position grid and a preference for self-repair.
Repair Sequence is part of Asymmetry Prime

Repair sequence contains an agency asymmetry because self- and other-repair positions are not interchangeable and are reached with a stable preference ordering.
Repair Sequence is part of Decomposition Prime

Repair sequence contains a decomposition that separates trouble source, initiation, and completion and cross-classifies them by agency and sequential position.
Repair Sequence presupposes Turn Taking Prime

Repair sequence presupposes turn-taking because its positions and agency classes are defined relative to turns, transition spaces, and who holds the conversational floor.

Hierarchy paths (15) — routes to 9 parentless roots

Repair Sequence → Communication Repair → Coordination → Concurrency

Show alternative paths (14)

Not to Be Confused With¶

Other-initiated repair. One cell of the grid — the case where a different participant flags the trouble ("Who?", "huh?"). It is a part of the repair organization, not the whole: the framework also covers self-initiated repair, same-turn restarts, and word-searches, and the whole point of the agency parameter is to hold "who notices" apart from "who fixes." Treating other-initiation as the concept mistakes one coordinate for the two-parameter system. Tell: is only the flag-by-another move in view (other-initiated repair), or the full trouble-source/initiation/completion routine cross-classified by agency and position (repair sequence)?
Turn-taking. The more general CA machinery that allocates who speaks when — the sequential channel discipline itself. The repair sequence runs on that discipline (the position parameter is read off turn structure) but is a specific routine keyed to intelligibility trouble, not the allocation system. Tell: is the question who gets the floor and when (turn-taking), or how a flagged trouble source gets located and mended (repair sequence)?
Adjacency pair. The paired-turn unit of CA (greeting/greeting, question/answer) in which a first turn makes a specific second normatively due. Repair sequences often expand or interrupt adjacency pairs but are not defined by pairing; their organizing structure is the three positions plus the agency-by-position grid, not a first-and-second slot. Tell: is the unit a normatively paired first/second turn (adjacency pair), or a trouble-source/initiation/completion routine triggered by trouble (repair sequence)?
Grounding / common ground. The broader account (as in Clark's contribution model) of how participants accumulate mutual belief and confirm understanding over a conversation. Repair is one mechanism grounding can use, but grounding is the wider bookkeeping of shared knowledge, whereas the repair sequence is the local routine for locating and fixing a specific trouble source. Tell: is the focus the running accumulation of mutual understanding (grounding), or the position-and-agency handling of a single flagged trouble (repair sequence)?
Corrective feedback / recast (language pedagogy). The instructional practice of correcting a learner's error, e.g. a teacher's recast of a malformed utterance. This is a normatively loaded teaching move aimed at learning, whereas a repair sequence is an evaluatively neutral, mutual routine for maintaining intelligibility in real time — and much repair (word-searches, "huh?") corrects no error at all. Tell: is the move a teacher improving a learner's form (corrective feedback), or participants preserving shared understanding turn-by-turn regardless of who is "right" (repair sequence)?
The loop-closing primes it instantiates (feedback, error_correction, turn_taking/protocol). The substrate-free skeleton — a local feedback loop on intelligibility, fixing divergence near the trouble over a sequential channel — that the sequence borrows from its parents. These travel to ATC readbacks, incident-response scripts, and chatbot clarification turns; the repair sequence is the talk-pinned specialization, treated more fully under those umbrellas. Tell: strip away conversational turn-taking, the self/other agency parameter, and the face-preservation reading — if a generic loop-closing structure remains, that is the parent combination, not the repair sequence.

Neighborhood in Abstraction Space¶

Repair Sequence sits in a sparse region of the domain-specific corpus (94^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12