{"id":369,"date":"2026-10-31T12:00:00","date_gmt":"2026-10-31T04:00:00","guid":{"rendered":"https:\/\/portraitor.ai\/blog\/?p=369"},"modified":"2026-11-06T12:04:32","modified_gmt":"2026-11-06T04:04:32","slug":"conversational-repair","status":"publish","type":"post","link":"https:\/\/portraitor.ai\/blog\/conversational-repair","title":{"rendered":"Conversational repair: how a chat fixes its own mistakes"},"content":{"rendered":"\n

Conversational repair is the set of ways people deal with trouble in speaking, hearing or understanding while a conversation is under way. Emanuel Schegloff, Gail Jefferson and Harvey Sacks described it in the journal Language<\/em> in 1977 and showed that speakers repair their own trouble far more often than other people repair it for them.<\/p>\n\n\n\n

Every chat is full of small repairs: a word retyped with an asterisk, a message that opens with \u201cwait, I meant\u201d, a \u201cwhat?\u201d from the other side. A 2014 study of Facebook chat found the same machinery at work in typing as in talk, plus one kind of repair that speech does not have. That extra kind is the one an exported chat cannot show you.<\/p>\n\n\n\n

What conversation analysts mean by repair<\/h2>\n\n\n\n

The 1977 paper, \u201cThe Preference for Self-Correction in the Organization of Repair in Conversation\u201d<\/a>, appeared in Language<\/em> volume 53. Its abstract describes an organization of repair that operates in conversation and is addressed to recurrent problems in speaking, hearing and understanding.<\/p>\n\n\n\n

Repair is wider than fixing a mistake. Joanne Meredith and Elizabeth Stokoe, in their 2014 paper in Discourse & Communication<\/em><\/a>, quote the 1977 paper\u2019s own line that repair is \u201cneither contingent upon error, nor limited to replacement\u201d. Correction is the narrower class inside it, the case where there has been an actual error to replace.<\/p>\n\n\n\n

The stretch of talk that gets repaired is called the trouble source. A speaker who stops halfway through a sentence to find a better word is doing repair, even if the first word was not wrong.<\/p>\n\n\n\n

Who spots the trouble and who fixes it<\/h2>\n\n\n\n

The framework asks two separate questions about any repair: who starts it, and who completes it. In the terms Meredith and Stokoe set out, a repair is self-initiated when the current speaker starts it, and it is self-repair when the same speaker both starts and completes it. When the other participant is the one who signals the trouble, the repair is other-initiated.<\/p>\n\n\n\n

Two questions with two answers give four combinations. A speaker can catch and fix their own slip. A listener can flag a problem and leave the speaker to fix it. A speaker can flag their own trouble, such as a word they cannot find, and the listener supplies it. And a listener can both notice the trouble and fix it, which is what most people mean by being corrected.<\/p>\n\n\n\n

Keeping the two questions apart matters because the answers do not always match. The 2013 study discussed below defines its subject exactly this way: repair initiated not by the speaker of the first turn but by the other participant, with the original speaker then producing a turn that aims to solve the trouble.<\/p>\n\n\n\n

Why speakers usually get to their own mistakes first<\/h2>\n\n\n\n

The central finding of the 1977 paper is a skew. In the abstract\u2019s words, self-repair predominates over other-repair, and the authors set out to explain the mechanism that produces that skew and to show a preference for self-repair at work in how repair is organised.<\/p>\n\n\n\n

Meredith and Stokoe summarise the main reason in one sentence: opportunities for self-initiation come before opportunities for other-initiation. The speaker gets the first chance, inside the turn that holds the trouble, before the listener has said anything at all. By the time the other person could step in, the speaker has often already fixed it.<\/p>\n\n\n\n

Huh and what as the other person\u2019s way in<\/h2>\n\n\n\n

When the other person starts a repair, the word they use is a repair initiator. Mark Dingemanse, Francisco Torreira and N. J. Enfield studied the shortest one in a 2013 paper in PLOS ONE<\/em><\/a>: an interjection like \u201chuh?\u201d, used when, for example, someone has not clearly heard what was just said. From 196 instances in recorded informal conversation across 10 languages, they argued that it is a universal word, shaped by the shared environment of other-initiated repair rather than built in at birth.<\/p>\n\n\n\n

They call this case open other-initiated repair, because \u201chuh?\u201d signals that there is a problem but leaves open what the problem is. Alongside the interjection, the languages they studied also had a question word option, comparable to English \u201cwhat?\u201d.<\/p>\n\n\n\n

A 2015 PLOS ONE<\/em> study led by Dingemanse<\/a> widened the sample to 12 languages from 8 language families. In 48.5 hours of informal conversation it found 2,053 cases of other-initiated repair, about one every 1.4 minutes. All the languages shared three types of initiator: an open request such as \u201cHuh?\u201d, a restricted request such as \u201cWho?\u201d, and a restricted offer that proposes what was said and asks for confirmation. People chose the most specific type they could.<\/p>\n\n\n\n

Both studies recorded spoken conversation. Neither measures how often a typed \u201cwhat?\u201d appears in a chat, so their figures are not a rate to expect in a message thread.<\/p>\n\n\n\n

Three places a typed repair can happen<\/h2>\n\n\n\n

Meredith and Stokoe studied one to one Facebook chat between friends and intimates. They found that repair happens in typing as it does in talk, and that the preference for self-correction holds in Facebook chats as much as in spoken conversation. What changes is where a repair can sit. Typing gives it three positions.<\/p>\n\n\n\n

Before the message is sent.<\/strong> The paper calls this message construction repair: changes made while a message is being typed, which are unavailable to the recipient. The researchers could see it only because one participant recorded her own screen. Of the other person\u2019s typing, neither they nor she had any access to what he wrote, edited or deleted before sending.<\/p>\n\n\n\n

After sending, inside the same message.<\/strong> Telegram\u2019s FAQ<\/a>, describing its groups, says you can edit your messages after posting. That is a repair made in place, to a message the other person may already have read, without adding a new turn to the conversation.<\/p>\n\n\n\n

After sending, in a new message.<\/strong> This is what Meredith and Stokoe call visible repair, seeable by both parties. The asterisk is its best known form. They cite Collister\u2019s 2011 study of chat in the game World of Warcraft, in which a player who typed \u201cot\u201d followed it with \u201cout*\u201d, the asterisk marking the corrected version, and Collister found the same mark used for both self-repair and other-repair. A message that begins \u201cwait, I meant\u201d does the same job in words.<\/p>\n\n\n\n

Meredith and Stokoe say plainly that they cannot know why a writer fixes some typos while composing and leaves others to be fixed after posting.<\/p>\n\n\n\n

What an exported chat keeps of a repair<\/h2>\n\n\n\n

The three positions leave very different traces once a conversation is exported.<\/p>\n\n\n\n

Message construction repair leaves none. It was never sent, so it is not part of the chat history on either side, and no export from either phone can contain it.<\/p>\n\n\n\n

A follow-up correction is a message in its own right. WhatsApp\u2019s help page on exporting<\/a> describes the export as your chat history in a text file, and Telegram\u2019s announcement of its export tool<\/a> says it produces the history as JSON or HTML. A line reading \u201cout*\u201d or \u201cwait, I meant Thursday\u201d sits in that history next to the message it repairs, so the repair can still be read afterwards. The guides to exporting a WhatsApp chat<\/a> and exporting a Telegram chat<\/a> walk through getting that file.<\/p>\n\n\n\n

An edit is the uncertain case. Neither WhatsApp\u2019s export help page nor Telegram\u2019s export announcement says whether an export records that a message was edited, or what it said before. On those pages alone, an export may hold only the final wording, and a reader of the file may have no way to tell that a repair happened.<\/p>\n\n\n\n

That gap has a practical consequence. If one person tends to fix things with a follow-up message and the other tends to edit in place, the export can make it look as though only one of them ever corrected anything.<\/p>\n\n\n\n

Portraitor<\/a> reads an exported chat file. Its home page asks you to pick the one person you want a portrait of, and says their messages are read while everyone else stays context, which is the choice described in analysing your own messages<\/a>. The home page does not say how an edited message, or any edit marker in a file, is handled, and this article does not guess. Which files it accepts is covered in which chat export files can be analysed<\/a>.<\/p>\n\n\n\n

Reading a repair after the fact has a ceiling<\/h2>\n\n\n\n

The research on repair was built from recordings and screen captures, where an analyst can watch a turn being produced. An export is a later record of what was sent. It keeps the messages in order, and it does not keep the deleted draft that Meredith and Stokoe found only by recording a screen.<\/p>\n\n\n\n

The figures in the 2015 study describe informal spoken conversation in 12 languages. They are not a benchmark for how much repair a chat ought to contain, and a thread with few visible repairs does not show that nothing needed fixing. The fixing may simply have happened before the messages were sent.<\/p>\n\n\n\n

The finding that self-correction is preferred in typing comes from one corpus of one to one Facebook chat. It is a careful study of that setting, not a measurement of every app. Repair is also only one way that typed conversations go wrong or right; why text messages get misunderstood<\/a> and grounding in communication<\/a> look at the medium from other directions.<\/p>\n\n\n\n

A refusal is a different shape of trouble from an error, and conversation analysts give it its own machinery. The piece on how a dispreferred response is built<\/a> sets out the pauses, prefaces and accounts that carry a no without the word.<\/p>\n\n\n\n

For a closer look at what deixis looks like in a chat export, see what deixis looks like in a chat export<\/a>.<\/p>\n\n\n\n

Frequently asked questions<\/h2>\n\n\n\n
What is conversational repair?<\/summary>\n

Conversational repair is how people handle trouble in speaking, hearing or understanding while a conversation is happening. The term comes from the 1977 paper by Schegloff, Jefferson and Sacks in Language<\/em>, and it covers more than correcting errors: a speaker searching for a better word is doing repair too.<\/p>\n<\/details>\n\n\n\n

What is the difference between self-initiated and other-initiated repair?<\/summary>\n

A repair is self-initiated when the person who produced the trouble starts fixing it, and other-initiated when someone else signals the problem first. Who completes the fix is a separate question, which is why the literature distinguishes self-repair from other-repair as well.<\/p>\n<\/details>\n\n\n\n

Why is \u201chuh?\u201d called a repair initiator?<\/summary>\n

Because it starts a repair without saying what is wrong. Dingemanse, Torreira and Enfield describe it as open other-initiated repair: the listener signals a problem with what was just said and leaves the speaker to work out what it was and fix it.<\/p>\n<\/details>\n\n\n\n

Is repair the same as correction?<\/summary>\n

No. Correction is one kind of repair, the kind that replaces an actual error. The 1977 paper describes repair as neither contingent upon error nor limited to replacement, so a restart or a search for a word counts as repair even when nothing was wrong.<\/p>\n<\/details>\n\n\n\n

Does an exported chat show which messages were edited?<\/summary>\n

The official pages checked for this article do not say. WhatsApp\u2019s export help page and Telegram\u2019s export announcement both describe what format the history comes in, and neither mentions edits, so they give no basis for expecting an edit marker or the earlier wording in the file.<\/p>\n<\/details>\n\n\n\n

In a typed conversation, the repairs you can read afterwards are the ones somebody chose to send.<\/p>\n","protected":false},"excerpt":{"rendered":"

Conversational repair is how people fix trouble in talk. Schegloff, Jefferson and Sacks found that speakers usually fix their own, and typing adds a kind of repair no export keeps.<\/p>\n","protected":false},"author":6,"featured_media":368,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"pp_title_highlight":"","pp_meta_title":"","pp_meta_description":"Conversational repair explained: who fixes trouble in talk, why self-correction wins, and which typed repairs an exported chat keeps and which it loses.","pp_sidebar_cta":"","pp_sources":"[{\"html\": \"Schegloff, E. A., Jefferson, G., & Sacks, H. (1977). The preference for self-correction in the organization of repair in conversation. Language<\/em>, 53(2), 361 to 382.\", \"href\": \"https:\/\/doi.org\/10.2307\/413107\"}, {\"html\": \"Meredith, J., & Stokoe, E. (2014). Repair: Comparing Facebook ‘chat’ with spoken interaction. Discourse & Communication<\/em>, 8(2), 181 to 207. Loughborough University repository copy.\", \"href\": \"https:\/\/ndownloader.figshare.com\/files\/17098574\"}, {\"html\": \"Dingemanse, M., Torreira, F., & Enfield, N. J. (2013). Is “Huh?” a universal word? PLOS ONE<\/em>, 8(11).\", \"href\": \"https:\/\/journals.plos.org\/plosone\/article?id=10.1371\/journal.pone.0078273\"}, {\"html\": \"Dingemanse, M., Roberts, S. G., et al. (2015). Universal principles in the repair of communication problems. PLOS ONE<\/em>, 10(9).\", \"href\": \"https:\/\/journals.plos.org\/plosone\/article?id=10.1371\/journal.pone.0136100\"}, {\"html\": \"Telegram, FAQ<\/em>.\", \"href\": \"https:\/\/telegram.org\/faq\"}, {\"html\": \"Telegram, Chat Export Tool, Better Notifications and More<\/em>.\", \"href\": \"https:\/\/telegram.org\/blog\/export-and-more\"}, {\"html\": \"WhatsApp Help Center, How to export your chat history<\/em> (Android).\", \"href\": \"https:\/\/faq.whatsapp.com\/1180414079177245\/?cms_platform=android\"}, {\"html\": \"Portraitor, home page.\", \"href\": \"https:\/\/portraitor.ai\/\"}]","pp_endcard":"","pp_howto":"","footnotes":""},"categories":[3],"tags":[],"class_list":["post-369","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-understanding-yourself"],"_links":{"self":[{"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/posts\/369","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/comments?post=369"}],"version-history":[{"count":3,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/posts\/369\/revisions"}],"predecessor-version":[{"id":570,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/posts\/369\/revisions\/570"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/media\/368"}],"wp:attachment":[{"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/media?parent=369"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/categories?post=369"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/portraitor.ai\/blog\/wp-json\/wp\/v2\/tags?post=369"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}