Conversational repair is the coordinated process by which people notice and resolve problems in speaking, hearing, understanding, or responding to an earlier contribution. It includes correcting one's own words, asking another person to repeat or specify something, offering a candidate interpretation, and confirming a newly clarified meaning. Repair makes communication resilient because participants can interrupt the course of an activity, identify what needs attention, and restore sufficient understanding to continue. Its organization is a central object of conversation analysis, psycholinguistics, pragmatics, and the cognitive science of dialogue. Albert and de Ruiter (2018); Schegloff, Jefferson, and Sacks (1977).
The scientific question extends beyond which words people use to ask for clarification. Researchers investigate the sequence in which trouble becomes visible, the distribution of responsibility for repairing it, the information conveyed by different requests, and the cognitive processes that support monitoring and response. Naturalistic recordings are particularly important because a repair form derives its significance from what happened before it and what participants do next. Comparative corpora and controlled experiments provide complementary evidence about the generality and possible causes of those patterns. Dingemanse et al. (2015); Albert and de Ruiter (2018).
What Is Conversational Repair?
Repair, misunderstanding, and clarification
A misunderstanding may involve an unheard syllable, a misidentified word, an ambiguous pronoun, an unclear instruction, or an unexpected social action. These sources are analytically different. If one person says “Take the green folder,” a recipient may request repetition because the color word was obscured, or ask “Which green folder?” because multiple objects fit the description. A correct solution is therefore contingent on the trouble that becomes relevant. Conversation analysts generally identify repair from participants' displayed conduct rather than inferring a hidden error from the analyst's private interpretation. Albert and de Ruiter (2018); Drew (1997).
Clarification is one important activity within the broader organization of repair. Repair also includes speaker-initiated restarts, substitutions, corrections, insertions, and collaborative word completion. Some revisions occur before anyone else has responded; others are prompted by an observable sign of uncertainty. Distinguishing the format from its function matters because a repetition can acknowledge, challenge, tease, or seek confirmation. Only the larger conversational sequence supports a defensible interpretation. Kendrick (2015a); Schegloff, Jefferson, and Sacks (1977).
The four elements of an exchange
An instructive sequence contains a trouble source, an initiation of repair, a solution, and an outcome. Imagine a speaker saying “Meet me beside the library,” followed by “Which library?” and “The central library on King Street.” The first turn supplies a potentially ambiguous reference, the question identifies it, and the final response specifies it. An acknowledgment or a relevant action can display sufficient resolution. The episode can be very short, or it can involve several successive attempts if the first solution leaves an unresolved ambiguity. Dingemanse, Blythe, and Dirksmeyer (2014); Albert and de Ruiter (2018).
These components are analytical tools rather than mandatory surface sentences. A speaker can self-correct within the trouble-source turn; the recipient can point rather than speak; a visible object movement can demonstrate understanding without an explicit “okay.” Analysts must follow the participants' orientations to the problem. This sequential attention is what differentiates the study of repair from a general list of good communication tips. Albert and de Ruiter (2018); Clark and Schaefer (1989).
Four Configurations of Repair
Self-initiated self-repair
Self-initiated self-repair occurs when a speaker both indicates and fixes the problem. “We arrived Tuesday—no, Wednesday” illustrates a replacement, while “The yellow—uh, the larger envelope” can replace a description. Restarts and insertions also belong here when they address a locally relevant trouble. Many are produced rapidly while a turn is still unfolding. Their presence in fluent conversation shows that ongoing speech production is adaptable and subject to monitoring. Schegloff, Jefferson, and Sacks (1977).
The revised expression may correct an error, clarify a potentially unclear choice, or better fit the recipient's likely understanding. The same pause or restart can have several causes, however, and transcription alone rarely identifies a unique internal mechanism. A researcher should avoid diagnosing a word-finding problem simply because a speaker changes a word. The functional issue is what the revision accomplishes in its conversational location. Albert and de Ruiter (2018); Schegloff, Jefferson, and Sacks (1977).
Self-initiated other-repair
A speaker may display difficulty but allow a partner to provide the solution: “I saw the actor from that film—what's his name?” A recipient supplying the name completes the repair. The format combines searching for an expression with distributing participation in the conversation. Its frequency and design depend on the available common knowledge and on whether the speaker makes another person's assistance relevant. Schegloff, Jefferson, and Sacks (1977); Clark and Brennan (1991).
A word completion may also be playful, supportive, pedagogical, or intrusive, depending on the activity. A seemingly helpful interruption cannot be classified by lexical content alone. The speaker's subsequent acceptance, revision, or resistance provides evidence for how the contribution was treated. This illustrates why turn sequence, participation roles, and the status of shared knowledge belong in the scientific description. Albert and de Ruiter (2018).
Other-initiated self-repair
In other-initiated self-repair, a recipient signals a problem and the original speaker supplies the solution. “She's at the station.” “Which station?” “The north station” is a simple case. It is particularly informative about mutual comprehension because the request publicly identifies uncertainty while giving the initial speaker an opportunity to reformulate. Cross-linguistic research finds recurring structures of this general kind despite extensive differences in grammatical systems. Dingemanse et al. (2015); Dingemanse, Blythe, and Dirksmeyer (2014).
The recipient can request the entire message again, isolate one phrase, or present a candidate interpretation for confirmation. Each option conveys a different amount of information about the listener's current understanding. When the response is sufficiently precise, the interaction can move back to the activity that the repair momentarily suspended. If it is not, another request may follow. Kendrick (2015a); Albert and de Ruiter (2018).
Other-initiated other-repair
A recipient may both indicate trouble and supply a correction, as when someone responds “You mean Thursday” to an incorrectly specified date. The original speaker can accept the correction or reject it. This form shows that solutions need not come from the person who produced the initial message. Nevertheless, correcting another person has social implications, and participants organize such actions in relation to local rights, knowledge, and roles. Schegloff, Jefferson, and Sacks (1977).
Descriptions of conversational preference should not be mistaken for moral rules. The empirical preference for self-correction concerns opportunities and tendencies within interaction, including ways of inviting the first speaker to fix a problem. Direct correction can be entirely appropriate in urgent situations, technical work, teaching, or a jointly recognized factual task. The next turns reveal how it was received. Schegloff, Jefferson, and Sacks (1977); Albert and de Ruiter (2018).
The Anatomy of a Repair Sequence
Identifying the trouble source
The trouble source is the earlier contribution, action, or aspect of context that participants treat as requiring repair. It can be a spoken word, a reference, a grammatical construction, an implication, or an unexpected response. The analyst's task is to determine what the participants themselves make accountable. An open “Sorry?” after a sudden topic change may address relevance or social expectation as much as auditory perception. Drew (1997); Albert and de Ruiter (2018).
A source may become evident only after an intervening action. For example, a recipient may follow an instruction incorrectly, prompting the speaker to clarify which object was meant. Such cases require tracing the relation between earlier wording and later uptake. It is possible for an analyst to detect an ambiguity that the participants never orient to; this is a linguistic possibility, but it should not be coded automatically as an observed episode of repair. Clark and Schaefer (1989); Albert and de Ruiter (2018).
Initiation: making uncertainty public
Repair initiation creates a new obligation or opportunity for a response. An open request minimally indicates a problem. A targeted question identifies missing information. A candidate interpretation asks the original speaker to evaluate a possible solution. The initiating turn therefore has a retrospective function, pointing to earlier talk, and a prospective one, structuring what should happen next. Dingemanse, Blythe, and Dirksmeyer (2014); Kendrick (2015a).
Not every apparent repair marker functions solely as a repair request. The same grammatical form can display surprise, disbelief, joking, or disagreement. Researchers consider timing, intonation, sequential location, and how the other person answers. Treating every question mark or every repetition as a uniform instance of misunderstanding would obscure the very organization that the analysis aims to explain. Kendrick (2015a); Drew (1997).
Solution and acceptance
A repair solution may repeat an inaudible stretch, replace a name, identify a referent, expand an elliptical expression, reformulate a proposition, or confirm the recipient's candidate. What counts as sufficient depends on the participants' current project. The answer “The north door” can resolve a navigation instruction while remaining inadequate for a detailed architectural description. Repair is therefore assessed against the practical requirements of the interaction. Clark and Brennan (1991); Albert and de Ruiter (2018).
Closure often appears through an acknowledgment, a relevant answer, or the continuation of a previously suspended activity. A recipient's subsequent action is useful evidence, though an “okay” does not guarantee complete private understanding. Two people may proceed together while retaining different background assumptions. Research on grounding thus emphasizes available evidence for functional sufficiency rather than direct access to identical internal representations. Clark and Schaefer (1989); Clark and Brennan (1991).
Open Requests, Restricted Requests, and Candidate Interpretations
Open-class requests
“What?”, “Sorry?”, “Pardon?”, and “Huh?” can indicate trouble without identifying precisely which component was unclear. Such requests leave considerable latitude for the original speaker to repeat, reformulate, or explain. They can be efficient when little of the preceding message was recovered. Drew's analysis also demonstrated that open repair initiators occur in sequential environments involving topic changes or responses that do not fit expectation. Drew (1997); Kendrick (2015a).
The apparent generality of an open request should not be confused with absence of communicative structure. Its placement after the previous turn and the form of the next reply can establish exactly what participants were doing. A speaker might repeat a word, defend a surprising claim, or recognize that the other person has challenged the relevance of the preceding contribution. Acoustic, semantic, and social troubles can all be implicated. Drew (1997); Albert and de Ruiter (2018).
Restricted requests
“Who?”, “Which meeting?”, and “At what time?” identify the missing component of an otherwise partly intelligible message. The recipient publicly displays what they have recovered and where the uncertainty remains. Because a more specific request places a smaller burden on the speaker, it can shorten the subsequent solution. These forms recur across languages, though their grammatical realization, intonation, and conversational conventions vary. Dingemanse, Blythe, and Dirksmeyer (2014); Dingemanse et al. (2015).
The specificity of a question is informative but should not be overinterpreted. It indicates a publicly available candidate for the trouble rather than revealing the whole underlying mental representation. A question can be strategically narrow to accelerate a task, and a broader formulation may be used despite partial comprehension. Context and subsequent uptake still determine the action performed. Albert and de Ruiter (2018); Kendrick (2015a).
Candidate understandings
“The downtown office?” or “You mean her sister?” offers a possible interpretation for confirmation. Such an initiation can be highly economical: a correct candidate requires only confirmation; an incorrect one helps the first speaker identify exactly what to repair. Candidate offers connect incremental comprehension, referential inference, and collaborative verification. They are especially useful in studies of partial understanding because the listener supplies a visible hypothesis about intended meaning. Dingemanse et al. (2015); Dingemanse, Blythe, and Dirksmeyer (2014).
A candidate can also perform social work. Repeating an unexpected statement with questioning intonation may invite confirmation and display astonishment at the same time. There is no simple one-to-one mapping between a surface grammatical form and one private cognitive state. The sequential response shows whether the candidate was treated as a question, a challenge, a correction, or a combination. Kendrick (2015a); Drew (1997).
Why specificity is scientifically important
A comparative study of twelve languages belonging to eight language families found broadly similar functional classes of other-initiated repair. The authors reported that participants often selected the most specific available format, consistent with minimizing shared repair effort. The evidence supports an interactional economy principle across the sampled communities; it does not by itself establish a single mental optimization algorithm or neural architecture. Dingemanse et al. (2015).
This distinction between behavioral organization and cognitive mechanism is central. A pattern in recordings can reflect available grammatical resources, local expectations, attentional state, partial recognition, or cost distribution. Identifying which factors cause a particular choice requires designs that vary those factors or compare conditions under carefully controlled circumstances. Observational regularity and causal explanation are related scientific achievements with different evidential requirements. Albert and de Ruiter (2018); Dingemanse et al. (2015).
How People Notice That Something Went Wrong
Auditory and lexical monitoring
A listener must extract a speech signal under changing noise, accent, tempo, and pronunciation. Inadequate access to a syllable can lead to a broad request for repetition; an uncertain name can produce a more targeted question. However, a recognizable “What?” in a transcript does not disclose whether the trouble arose in auditory encoding, lexical recognition, meaning construction, or the relation of an utterance to its context. Distinguishing these sources demands suitable experimental measures. Albert and de Ruiter (2018); Drew (1997).
A speaker may also detect a formulation problem during production and revise it before the interlocutor responds. The resulting restart demonstrates public monitoring and adaptation, but an explanation in terms of a particular language-processing stage remains a theoretical inference. The temporal organization of self-repair provides valuable evidence to be combined with other methods, not a complete psychological localization on its own. Schegloff, Jefferson, and Sacks (1977); Albert and de Ruiter (2018).
Referent resolution
Recognizing the words “the blue one” does not determine which object is intended when several candidates are present. A recipient can ask for a distinctive attribute, an object location, or a person’s identity. These requests make uncertainty about reference explicitly discussable. Repeated interactions can also establish a shared label that becomes easier to use in subsequent turns, reducing the need to describe the referent from scratch. Brennan and Clark (1996); Clark and Brennan (1991).
Brennan and Clark's experiments on conceptual pacts are especially relevant. They found that speakers' later descriptions of objects were influenced by previous naming interactions with particular partners. This supports the importance of interaction history in reference selection. It does not settle every question about how conceptual agreements are stored or how much deliberate reasoning each use involves. Brennan and Clark (1996).
Pragmatic and sequential expectations
Even completely intelligible words can yield puzzlement when an utterance performs an unexpected action. A statement might be heard as a request, invitation, objection, joke, or correction. A clarification can therefore ask what someone intended to do rather than what they literally said. A response that seems irrelevant may provoke repair because it does not fit the projected sequence of actions. Drew (1997); Albert and de Ruiter (2018).
This aspect connects repair with pragmatics without collapsing the subjects into one. Pragmatic interpretation concerns how context helps determine intended and inferred meaning; repair concerns what participants do when some feature of their ongoing interaction is made problematic. A repair request can expose a pragmatic misunderstanding, but pragmatic meaning is negotiated in many interactions where no explicit repair occurs. Clark and Brennan (1991); Drew (1997).
Grounding: How Shared Understanding Becomes Sufficient
A collaborative process
Grounding is the collaborative establishment of evidence that a contribution has been understood adequately for present purposes. It does not require that participants possess identical comprehensive knowledge. A brief instruction to move an object might be sufficiently grounded when the correct object is moved, while an explanation of legal consent requires different evidence. Clark and Brennan developed the collaborative account of grounding, and Clark and Schaefer analyzed how contributions become accepted within discourse. Clark and Brennan (1991); Clark and Schaefer (1989).
Repair is one way to obtain missing evidence. A request for specification displays that the initial contribution is not yet adequate for some part of the task. A proposed solution and its subsequent acceptance can establish a renewed basis for action. People continually adjust what counts as enough evidence in relation to stakes, communication channel, shared history, and the cost of an error. Clark and Brennan (1991); Albert and de Ruiter (2018).
Positive and negative evidence
A relevant answer, accurate action, or appropriate continuation may provide positive evidence that a message has been understood. A clarification question indicates that something requires additional work. These signs are context dependent. An acknowledgment such as “yes” may show careful understanding or simply encourage a storyteller to continue. The behavior following the acknowledgment can therefore be more informative than the acknowledgment alone. Clark and Schaefer (1989); Bavelas, Coates, and Johnson (2000).
An absence of overt repair is not proof of perfect understanding. People can carry an incorrect interpretation forward without publicly detecting a problem, or continue because the discrepancy is unimportant at that point. Conversely, a repair request can serve a rhetorical purpose despite adequate comprehension. Empirical research must distinguish visible displays of trouble from a complete inventory of unspoken interpretive differences. Albert and de Ruiter (2018).
The principle of collaborative economy
A communicative problem can be resolved by repeating a large stretch of discourse or by targeting a specific missing detail. The latter often reduces joint effort because it preserves the part of the message already understood. This idea is consistent with grounding theory and cross-linguistic evidence about specificity. The efficiency of a format nevertheless depends on how much the recipient actually knows and what kind of response the format makes available. Clark and Brennan (1991); Dingemanse et al. (2015).
Efficiency is not the only conversational consideration. Participants also orient to social relationships, rights to correct, attention, and the task's urgency. A highly direct correction may be expedient in one setting and socially disruptive in another. Strong explanations evaluate these influences in relation to particular events, rather than assuming every speaker mechanically selects the shortest possible utterance. Schegloff, Jefferson, and Sacks (1977); Albert and de Ruiter (2018).
Timing, Pauses, and the Relationship to Turn-Taking
The next opportunity to respond
Repair often appears at the earliest workable point after a trouble source. This placement limits how far the main activity proceeds under uncertainty. A recipient may start a question as the prior speaker completes a turn, or delay briefly while deciding whether a clarification is needed. Kendrick's work investigates how the timing of repair initiation relates to ordinary turn-taking and the opportunities for the first speaker to add or correct information. Kendrick (2015b).
A clarification request takes its own turn and may create an insertion sequence. The ongoing topic or practical task pauses while the uncertainty is handled. Repair therefore reshapes the local distribution of speaking opportunities. The timing of a response can be consequential, but a fixed duration does not carry one universal meaning in all activities, communities, or communication media. Kendrick (2015b); Albert and de Ruiter (2018).
Delays and hesitation
A gap can allow the first speaker to revise a statement spontaneously, or give the recipient time to prepare a more specific repair initiation. It can also arise from searching memory, distraction, reluctance, or difficulty processing the signal. The action underway determines which interpretation is plausible. Researchers avoid turning a measure of silence into a psychological diagnosis. They inspect the preceding and following turns and, when available, prosody and visible behavior. Kendrick (2015b); Schegloff, Jefferson, and Sacks (1977).
Delayed repair can also return to an earlier phrase after a later response exposes a misunderstanding. Suppose someone moves toward the wrong entrance; the speaker may then say “I meant the entrance on the east side.” The visible action supplies evidence about the earlier reference. Analyzing such cases requires following dependencies across more than one neighboring pair of turns. Clark and Schaefer (1989); Albert and de Ruiter (2018).
Prosody, Repetition, and Reformulation
Intonation shapes a request
A rising “Tomorrow?” can propose a candidate day, display surprise, or seek confirmation. Stress on one word may isolate the troublesome component. Tone and sequence jointly determine what the recipient's contribution is doing. Written transcripts that remove pauses, overlap, and prosodic information can make distinct actions appear identical. Audio analysis is essential when the scientific claim depends on the actual sound of a repair initiation. Kendrick (2015a); Albert and de Ruiter (2018).
Repeating all or part of a turn
Repeating the full prior turn may invite the speaker to restate or explain it, while a partial repetition can focus the uncertainty on a single expression. Repetition can also be a form of acknowledgment or affiliation. Its mere presence is not a valid stand-alone indicator of misunderstanding. Analysts examine whether the subsequent contribution provides a solution and whether the interaction resumes the prior activity. Dingemanse, Blythe, and Dirksmeyer (2014); Kendrick (2015a).
Reformulation supplies new information
A speaker can replace a vague description with a more discriminating one, transform an unfamiliar phrase into simpler vocabulary, or clarify the relation between people and events. Reformulation is especially important when repetition leaves the problem intact. An unclear instruction repeated loudly remains unclear if its problem is ambiguity rather than audibility. The listener's uptake provides evidence about the adequacy of the revised message. Albert and de Ruiter (2018); Clark and Schaefer (1989).
A paraphrase is not automatically equivalent to the first expression. It can add assumptions, narrow a reference, or change an intended meaning. Participants may therefore initiate another repair or explicitly reject the new formulation. Good scientific description tracks what becomes mutually established rather than assuming linguistic resemblance guarantees shared understanding. Clark and Brennan (1991); Albert and de Ruiter (2018).
Cross-Linguistic Findings
Twelve languages and recurring formats
The cross-linguistic study by Dingemanse and colleagues analyzed 48.5 hours of sampled conversation in twelve languages belonging to eight language families. It identified 2,053 other-repair initiations, averaging approximately one every 1.4 minutes in the sampled conversations. Across languages the researchers found open requests, restricted requests, and restricted offers as broad recurring categories. These are findings about a systematically assembled observational corpus, not a census of all languages or activities. Dingemanse et al. (2015).
Their study also found functional regularities in selection and sequencing that suggest a common interactional infrastructure. The precise words and constructions differ substantially among languages. The scientific significance lies in the similarity of the problem being solved and the broad organization of the solution, alongside considerable cultural and linguistic diversity in realization. Dingemanse, Blythe, and Dirksmeyer (2014); Dingemanse et al. (2015).
The “Huh?” question
Dingemanse, Torreira, and Enfield investigated similarities among brief interjections used for open repair. They argued that recurrent conversational demands can lead languages toward functionally comparable learned expressions. Their account involves convergent cultural evolution rather than assuming one identical innate lexical item. The same basic repair task can favor short, recognizable forms while local language conventions shape their sound and use. Dingemanse, Torreira, and Enfield (2013).
Short repair interjections may be prominent because they can efficiently suspend progress without specifying the problem. Nonetheless, a superficially similar sound across languages cannot establish identical social meaning or usage frequency. The appropriate comparison considers conversational function, phonological form, learning, and actual behavior within the community. Dingemanse, Torreira, and Enfield (2013); Dingemanse et al. (2015).
What a universal claim can establish
A systematic sample from multiple language families provides stronger evidence than isolated anecdotes. Yet sampling, speaker selection, situational diversity, transcription, translation, coding reliability, and statistical dependence constrain generalization. The findings justify a conclusion about broad shared organization across the investigated languages. Strong claims about every language, population, or sensory modality require further representative evidence. Dingemanse et al. (2015); Dingemanse, Blythe, and Dirksmeyer (2014).
Repair Can Express Surprise, Disbelief, or Challenge
Repair-looking forms have multiple uses
A participant may repeat a claim with marked intonation because it seems astonishing. An open “What?” may query an unexpected topic shift rather than signal that the audio was missed. Kendrick's corpus-based study showed that forms associated with other-initiation of repair may also perform joking, teasing, or displays of disbelief. A coding method that treats all such forms as pure failures of comprehension would erase meaningful variation in social action. Kendrick (2015a).
Drew's analysis of open-class initiators is similarly important for recognizing sequentially generated trouble. A response can be puzzling because it does not fit the social trajectory of the conversation. Misunderstanding may concern what an utterance is doing in the interaction and how the other person is positioned to respond. The analysis must extend beyond words and propositions to the organization of action. Drew (1997).
Understanding is distinct from agreement
A person can understand a statement perfectly and disagree with it. They may ask for repetition to express surprise, seek evidence, or create an opportunity for the first speaker to revise a claim. Conversely, an apparent agreement can mask a different interpretation. Scientific analysis differentiates clarification, correction, alignment, assent, and disagreement according to their sequential consequences. Applied conflict-management strategies belong to a separate research purpose. Albert and de Ruiter (2018); Drew (1997).
Nonverbal and Multimodal Repair
Gaze, pointing, and action
Face-to-face communication provides visible information about the topic and the listener's response. A point toward an object may repair an ambiguous demonstrative, and moving an object can show what was understood. Eye gaze, timing, posture, and gesture may help participants identify the referent or see whether the instruction was followed. Such resources should be analyzed in context: visible behavior does not always have one interpretation or serve as a reliable index of private mental states. Albert and de Ruiter (2018).
Experiments on narrative interaction show why recipient feedback matters. Bavelas, Coates, and Johnson studied pairs telling stories and manipulated listeners' opportunity to attend. They found changes in specific listener responses and features of the narration when listeners were distracted. This provides evidence for the collaborative character of real-time discourse, while not serving as a direct estimate of the prevalence of clarification sequences. Bavelas, Coates, and Johnson (2000).
Repair beyond auditory spoken language
Interactional repair is relevant to signed and tactile communication as well as spoken conversation. Research on tactile Auslan has analyzed misunderstandings and repair in a communication system whose perceptual resources differ from face-to-face speech. Such studies challenge the assumption that all repair must begin from hearing speech. Explanations must account for visual or tactile access, turn allocation, attention, and context specific to the modality studied. Willoughby et al. (2014); Albert and de Ruiter (2018).
Findings about spoken repair cannot simply be copied to every signing community. The functions of requesting clarification may be comparable while the bodily resources for signaling and resolving trouble differ. Cross-modal claims should therefore distinguish the general organization of joint meaning-making from modality-specific linguistic and sensory mechanisms. Willoughby et al. (2014).
Repair Through Telephone, Messaging, and Digital Tools
Telephone communication
Audio-only conversation removes shared pointing and many visible cues, potentially increasing the burden of verbal reference specification. Participants can still use repetition, targeted questions, candidate understandings, and sequential acknowledgment. The actual distribution of repair formats varies with noise, familiarity, accessibility, task requirements, and communication norms. The existence of fewer visual cues alone does not establish an exact causal effect on repair frequency. Clark and Brennan (1991); Drew (1997).
Text messages and asynchronous exchange
In written messaging, quoting a sentence, replying to a particular message, or editing a previous contribution can help localize ambiguity. The durable record may make earlier expressions easier to revisit. Response timing, however, can reflect availability rather than immediate comprehension. The classic research reviewed here addresses mostly spoken conversation, so the extension to messaging is a conceptual application rather than a verified quantitative claim about all platforms. Albert and de Ruiter (2018); Clark and Brennan (1991).
Humans and language-generating systems
A user and an automated system may exchange clarification requests that resemble human conversational repair at the level of dialogue structure. Whether a system responds appropriately depends on its technical design, training, context window, interface, and error handling. Similar surface behavior does not establish identical underlying cognitive mechanisms. Questions concerning a system's subjective experience or philosophical status require separate evidence and argument; the psycholinguistics of conversational repair does not settle them. Albert and de Ruiter (2018).
Learning and Development
Children's participation in repair
Children encounter requests for repetition, clarification, and better reference in interactions with caregivers and peers. These episodes create opportunities to learn that a listener may need information beyond what the speaker initially supplied. Development can involve learning how to localize trouble, respond to a targeted question, and use the partner's next action as feedback. Age, language experience, task demands, and caregiver behavior matter when interpreting a child's response. Clark and Brennan (1991); Albert and de Ruiter (2018).
The adult interaction studies that define the core repair typology should not be turned into a developmental timetable. They do not establish universal ages of mastery for every format. Nor does a child who fails to clarify a single instruction thereby meet criteria for a clinical disorder. Developmental conclusions require age-appropriate empirical designs and, for clinical questions, separately validated measures. Albert and de Ruiter (2018).
Language learning and multilingual contexts
Second-language and multilingual conversations can involve unfamiliar words, pronunciation variation, culturally specific expressions, or different assumptions about shared context. A recipient may request an explanation or propose a candidate meaning. Repaired turns can supply additional linguistic input, but lasting language learning depends on memory, attention, repetition, and later use. A repair episode itself is not proof that a new expression has been retained. Dingemanse, Blythe, and Dirksmeyer (2014); Albert and de Ruiter (2018).
Repair is also common among highly proficient speakers of the same language. A high rate in one interaction can reflect complex content, background noise, task difficulty, or limited shared knowledge. It cannot be used independently to assign a language deficit to one participant or to infer a lower level of competence from accent or multilingual status. Dingemanse et al. (2015).
Research Methods and Measurement
Naturalistic conversation analysis
Conversation analysts examine recordings of actual exchanges, preserving details such as pauses, overlaps, timing, prosody, and sometimes visible movement. They compare cases to discover recurrent relations between a trouble source, a particular form of initiation, and the following solution. This approach explains how people publicly organize their interaction. It is less suited, without further evidence, to identifying a unique private cognitive representation or the neural implementation of a repair. Albert and de Ruiter (2018); Schegloff, Jefferson, and Sacks (1977).
Cross-linguistic corpus studies
Comparative studies develop coding criteria that permit researchers to classify functional formats across languages. Sampling from multiple speakers and activities is essential to prevent one conversation from dominating conclusions. Quantification requires a defined denominator: number of initiations, sequences, conversational minutes, or speakers. These measures answer different questions and are sensitive to coding decisions about repeated requests or ambiguous social actions. Dingemanse et al. (2015); Dingemanse, Blythe, and Dirksmeyer (2014).
Controlled experimentation
Experiments can manipulate referential ambiguity, prior partner experience, visible access, attention, or shared task knowledge and measure how participants respond. Brennan and Clark's conceptual-pact experiments show the value of manipulating naming history, while Bavelas and colleagues' work illustrates how listener participation can influence an unfolding story. Such causal evidence complements sequential descriptions, although artificial tasks may change participants' behavior relative to everyday conversation. Brennan and Clark (1996); Bavelas, Coates, and Johnson (2000).
Methodological limits
A request for clarification is observable; the exact moment at which a person privately noticed a problem may not be. A fluent answer shows that the interaction advanced but cannot prove that both people hold identical mental representations. Laboratory outcomes need ecological validation, and naturalistic corpora need precise operational definitions. The strongest research programs combine methods while keeping their inferential limits visible. Albert and de Ruiter (2018).
What the Evidence Establishes
Established: orderly, recurrent repair practices
The classic research on repair configurations, detailed studies of English other-initiation, and comparative cross-language evidence converge on a strong conclusion: interaction contains systematic opportunities for speakers and recipients to manage trouble and restore progress. Self-correction has a prominent organizational role; targeted requests and candidate understandings supply information about the recipient's partial understanding. These patterns are robust within the studied settings. Schegloff, Jefferson, and Sacks (1977); Kendrick (2015a); Dingemanse et al. (2015).
Well supported, but bounded: cross-cultural commonalities
The twelve-language study supports broad similarities among three classes of other-initiation and evidence of collaborative economy. It is a valuable comparative finding, not evidence that every community uses identical words or that all interactions contain one repair every fixed number of minutes. Situation, speakers, recording conditions, and coding standards affect the scope of any numerical generalization. Dingemanse et al. (2015).
Theoretical: cognitive alignment and grounding
Grounding models characterize the collaborative achievement of sufficient understanding, while conversation analysis examines publicly available practices of repairing interactional trouble. These perspectives can be integrated in specific studies, but a recorded sequence cannot fully resolve debates about how meanings are mentally represented. Theoretical interpretation is valuable when it generates distinguishable predictions and connects the observable sequence with cognitive evidence. Clark and Brennan (1991); Albert and de Ruiter (2018).
Limited: unique neural explanations
The research that establishes the organization of repair does not demonstrate a dedicated single neural center for clarification. Auditory perception, lexical access, attention, referential inference, and production planning may contribute in different situations. Neural explanations require direct and well-controlled studies of the relevant processes. Claims that every repair sequence has the same neural cause exceed the evidence reviewed here. Albert and de Ruiter (2018).
Contested: a single reason for choosing a form
The greater informativeness of candidate interpretations is consistent with minimizing collaborative effort, but format choice can depend on availability of hypotheses, conversational constraints, expectations, and social action. Observation of a preference does not settle the causal weight of these factors. Competing accounts are best evaluated through carefully designed contrasts that distinguish efficiency, comprehension, and interactional norms. Dingemanse et al. (2015); Kendrick (2015b).
Practical Interpretation of Misunderstandings
Identify what requires clarification
When the recipient already understands part of an expression, a question about the missing component can be more informative than an open request. “Which building?” offers the original speaker a more precise task than “What?” An open request is still useful when very little of the prior message was recovered. This is a functional consequence of the different repair formats, not a universal interpersonal rule. Dingemanse et al. (2015).
Look for subsequent uptake
A response's effectiveness becomes visible in what participants do next: selecting an object, answering the repaired question, or resuming an interrupted activity. An “okay” may sometimes supply sufficient evidence, while higher-stakes tasks call for more specific confirmation. Comprehension is assessed relative to current purposes, and subsequent action is one potentially informative source of evidence. Clark and Schaefer (1989); Clark and Brennan (1991).
Allow revision without stigmatizing ordinary trouble
Self-correction, hesitation, targeted requests, and reformulation belong to competent everyday conversation. Their occurrence alone cannot establish poor listening, low intelligence, dishonesty, anxiety, or a clinical diagnosis. Assessments of aphasia, developmental language disorder, hearing difficulties, or other conditions have different criteria and require independent clinical evidence. The scientific study of repair should preserve this distinction between a normal communicative mechanism and a diagnostic judgment. Schegloff, Jefferson, and Sacks (1977); Albert and de Ruiter (2018).
Frequently Asked Questions
What is an example of conversational repair?
One person says, “Use the folder near the printer.” The recipient asks, “The blue one?” and the first person replies, “Yes, the blue folder.” The request proposes a candidate referent and the answer confirms it. Another example is a speaker saying, “The appointment is Friday—sorry, Thursday,” correcting a word without a request from anyone else. These illustrate different configurations of the same broader organization. Schegloff, Jefferson, and Sacks (1977); Dingemanse et al. (2015).
Is clarification the same as repair?
Clarification is a prominent way of initiating or supplying repair, but the category of repair is broader. It also includes self-corrections, restarts, substitutions, repetitions, and cooperative completions. Researchers distinguish who identifies a problem, who supplies the solution, and where the relevant actions occur in the conversation. The distinction matters when comparing studies or interpreting interactional transcripts. Albert and de Ruiter (2018).
Why do people ask “What?” when they heard the words?
An open repair form can reflect uncertain meaning, a puzzling conversational action, surprise, disbelief, or an unexpected topic change. The preceding and following turns distinguish these functions better than the word alone. An audible sentence can still be unclear in reference or intention. Asking for clarification therefore cannot be equated automatically with failure of hearing. Drew (1997); Kendrick (2015a).
Does repair happen in every language?
Comparative studies document major similarities across a diverse sample of languages, including recurrent classes of open requests, restricted questions, and candidate offers. The strongest cited study examined twelve languages and does not literally sample every language on Earth. The evidence supports broadly recurring interactional infrastructure with local linguistic variation, not a claim of complete uniformity across all possible settings. Dingemanse et al. (2015).
Is self-repair preferable to correcting another person?
Classical conversation analysis reports a preference for self-correction in the organization of ordinary repair. The finding describes sequential opportunities and observed tendencies, not a moral prohibition on correction. The demands of a task and participants' roles can make direct correction appropriate. The scientific issue is how the selected practice fits the unfolding interaction and how others respond. Schegloff, Jefferson, and Sacks (1977).
Do clarification requests reveal conscious mental states?
A request shows that a participant publicly orients to a problem, but it does not provide complete access to private cognition. People can intentionally request information, revise speech rapidly, or use repair-looking forms for social purposes. Determining conscious awareness requires additional experimental and conceptual work. The transcript alone supports narrower conclusions about observable action. Albert and de Ruiter (2018).
What is the difference between turn-taking and repair?
Turn-taking organizes opportunities to speak and respond, while repair identifies and resolves trouble. They intersect because a repair initiation occupies a turn and may temporarily interrupt the projected sequence. Timing affects opportunities for the initial speaker to correct their own words and for the recipient to request clarification. Each topic retains distinct explanatory questions. Kendrick (2015b).
Can repair be studied in experiments?
Yes. Researchers can vary shared history, ambiguity, visibility, and attention, then examine how speakers describe objects and how listeners respond. Such methods help test causes, while naturalistic conversation analysis clarifies whether the measured behavior corresponds to actual repair in everyday talk. Both methods contribute to a fuller scientific account when their inferential limits are respected. Brennan and Clark (1996); Bavelas, Coates, and Johnson (2000); Albert and de Ruiter (2018).
Related Articles
Conversation Turn-Taking: How Speakers Coordinate Dialogue — Timing, turn transitions, and coordination.
Pragmatics in Psycholinguistics: Meaning Beyond Literal Words — Context and interpretation beyond literal wording.
Discourse Comprehension: How We Understand Connected Language — Comprehension and inference across connected speech.
Common Ground in Communication: How People Establish Shared Meaning — How interlocutors negotiate shared knowledge, reference, and grounding.
Gesture and Speech: How Multimodal Language Is Understood — Co-speech gesture and multimodal linguistic interpretation.
References
Albert, S., and de Ruiter, J. P. (2018). Repair: The interface between interaction and cognition. Topics in Cognitive Science, 10(2), 279–313. https://doi.org/10.1111/tops.12339
Bavelas, J. B., Coates, L., and Johnson, T. (2000). Listeners as co-narrators. Journal of Personality and Social Psychology, 79(6), 941–952. https://doi.org/10.1037/0022-3514.79.6.941
Brennan, S. E., and Clark, H. H. (1996). Conceptual pacts and lexical choice in conversation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22(6), 1482–1493. https://doi.org/10.1037/0278-7393.22.6.1482
Clark, H. H., and Brennan, S. E. (1991). Grounding in communication. In L. B. Resnick, J. M. Levine, and S. D. Teasley (Eds.), Perspectives on Socially Shared Cognition (pp. 127–149). American Psychological Association. https://doi.org/10.1037/10096-006
Clark, H. H., and Schaefer, E. F. (1989). Contributing to discourse. Cognitive Science, 13(2), 259–294. https://doi.org/10.1016/0364-0213(89)90008-6
Dingemanse, M., Blythe, J., and Dirksmeyer, T. (2014). Formats for other-initiation of repair across languages: An exercise in pragmatic typology. Studies in Language, 38(1), 5–43. https://doi.org/10.1075/sl.38.1.01din
Dingemanse, M., Roberts, S. G., Baranova, J., Blythe, J., Drew, P., Floyd, S., et al. (2015). Universal principles in the repair of communication problems. PLOS ONE, 10(9), e0136100. https://doi.org/10.1371/journal.pone.0136100
Dingemanse, M., Torreira, F., and Enfield, N. J. (2013). Is “Huh?” a universal word? Conversational infrastructure and the convergent evolution of linguistic items. PLOS ONE, 8(11), e78273. https://doi.org/10.1371/journal.pone.0078273
Drew, P. (1997). “Open” class repair initiators in response to sequential sources of troubles in conversation. Journal of Pragmatics, 28(1), 69–101. https://doi.org/10.1016/S0378-2166(97)89759-7
Kendrick, K. H. (2015a). Other-initiated repair in English. Open Linguistics, 1(1), 164–190. https://doi.org/10.2478/opli-2014-0009
Kendrick, K. H. (2015b). The intersection of turn-taking and repair: The timing of other-initiations of repair in conversation. Frontiers in Psychology, 6, 250. https://doi.org/10.3389/fpsyg.2015.00250
Schegloff, E. A., Jefferson, G., and Sacks, H. (1977). The preference for self-correction in the organization of repair in conversation. Language, 53(2), 361–382. https://doi.org/10.2307/413107
Willoughby, L., Manns, H., Iwasaki, S., and Bartlett, M. (2014). Misunderstanding and repair in tactile Auslan. Sign Language Studies, 14(4), 419–443. https://doi.org/10.1353/sls.2014.0014