Dispreferred responses: how a no is built, not said

A dispreferred response is the conversation analysis name for the shape a no comes in: delayed, prefaced, hedged, and followed by a reason, where a yes arrives early and bare. In a typed conversation the delay between messages stays on the page and the pause inside a turn does not.

Anita Pomerantz described the pattern in a 1984 book chapter about how people agree and disagree with each other’s assessments. The point was not that people dislike refusing. It was that a refusal is assembled differently from an acceptance, closely enough to the same recipe each time to be written down.

Thirty one years later two researchers timed 195 responses in recorded telephone calls and found that the most familiar version of the claim does not survive measurement. That correction is the useful part here, because it decides what a chat export can show about a no and what it cannot.

What a dispreferred response is built out of

The University at Albany, where Pomerantz worked, holds the record for the chapter, titled “Agreeing and disagreeing with assessments: some features of preferred/dispreferred turn shapes”. Her summary there says the chapter explicates two kinds of organisation: the relationship between experiencing an event and assessing it, and the relationship between the status of a conversational action and the turn shape in which it is performed.

The inventory of practices that make up that turn shape is set out by Kobin Kendrick and Francisco Torreira in a 2015 paper in Discourse Processes, which compares one speaker’s acceptance and rejection of the same invitation. The rejection uses five things the acceptance does not: prefatory particles and in-breaths; qualification and mitigation; appreciation of the invitation; the rejection itself placed somewhere other than next to the invitation; and an account explaining why the speaker cannot accept. Kendrick and Torreira credit that list to Atkinson and Drew (1979), Levinson (1983) and Heritage (1984).

A response built from practices like those has what they call a dispreferred turn format. A response without them has a preferred turn format. The two terms matter separately, because, as they note after Schegloff, a dispreferred action can be delivered in a preferred format and a preferred action in a dispreferred one.

Counted across their collection, the majority of dispreferred responses, 65.2 percent, included at least one of a turn-initial “well”, a turn-initial “u(h)m” or a turn-initial in-breath, against 23.6 percent of preferred responses. Read the other way round: when a response opened with “well”, the probability that it was a dispreferred action was 0.78; with “u(h)m”, 0.67; with an in-breath, 0.6. When a response opened with none of the three, the probability fell to 0.21.

Preference in this literature is a property of the sequence being answered, not a report of what anyone wanted to say.

The delay belongs to the turn format, not to the rejection

Kendrick and Torreira drew their data from eight corpora of telephone calls. Across 185 calls they identified 238 sequences in which the first turn was recognisable as a request, offer, invitation, proposal or suggestion. Eleven were set aside because overlapping talk made the start of the response impossible to locate, and 33 because the sequence contained an increment or an inserted sequence, either of which would distort a measurement in milliseconds. That left 195.

They then measured the gap three ways: silence only, silence plus preparatory noises such as in-breaths and clicks, and silence plus prefatory particles such as “well” and “u(h)m”.

Under the first measure there is no delay to find. The most frequent dispreferred responses begin about 50 milliseconds earlier than the most frequent preferred ones, because breathing is free to overlap the end of someone else’s turn. Under the second measure both sit at roughly 250 milliseconds, which the authors describe as an ordinary transition between speakers. Only under the third, which counts the hedging words as gap, does a difference of about 300 milliseconds appear: the most common dispreferred response arrives around 600 milliseconds, the most common preferred one around 300.

Their own summary is worth quoting exactly. “While we do not find that long delays are characteristic of dispreferred actions, since short gaps are the most frequent before dispreferreds, we do find that dispreferred actions are characteristic of long delays.” On the third measure, 47.2 percent of dispreferred responses arrived later than 600 milliseconds, against 19.5 percent of preferred ones. The long gaps really are mostly refusals. Most refusals are not in the long gaps.

The reframing they propose is that timing is part of how a turn is built rather than part of which action it performs. Once turn format is counted separately from action, the difference shows up in ordinary cases and not only extreme ones: after only 300 milliseconds the share of dispreferred formats overtakes the share of preferred formats. In overlap or within the first 300 milliseconds, a plain acceptance in a plain format was the response 62.9 percent of the time, against 19.2 percent for a rejection in a hedged format.

So a pause does not predict a no. Past 300 milliseconds, Kendrick and Torreira say, the action to predict is still an acceptance, but one in a shape that qualifies it.

Most rejections never say the word no

Kendrick and Torreira coded their collection with a deliberately mechanical rule. The turn after the first was taken as the response, and if it contained any component that accepted, agreed, confirmed or granted, it counted as a preferred action regardless of where in the turn that component sat. Everything else counted as dispreferred, so that the features under test played no part in the classification.

The rule had to be built this way because, as they report citing Kitzinger and Firth (1999), rejections normally do not contain an explicit rejection component such as “no”. A recognisable withholding of acceptance does the job instead. In one of their examples a woman answers an invitation to come and stay with “Yer a r’l sweetheart”, waits 0.7 seconds, and then says she wishes the other person had a car so she could come over there instead. The invitation is declined and the word no never appears.

The reverse exists too. A rejection delivered with no softening at all they call a flat rejection: the rejection component first, no hedge, no account. An acceptance delivered with softening they call a qualified acceptance. Just over a quarter of the preferred responses in the collection met their criteria for one, which include an acceptance made conditional on a date or on a third party, an acceptance said at low pitch in a compressed pitch range, an acceptance containing “I think” or “maybe”, and an acceptance the other speaker afterwards handled as insufficient.

A research assistant with no previous experience of conversation analysis coded a fifth of the preferred responses against those written criteria and agreed with the authors on 92 percent of them, a Cohen’s kappa of 0.83.

When agreeing is the dispreferred thing to do

Pomerantz’s summary names three environments, and they do not behave alike. When the thing being assessed has nothing to do with either speaker, respondents orient to agreeing as preferred. When the previous assessment was a self-deprecation, respondents orient to agreement as dispreferred. When the previous assessment was a compliment, recipients orient to multiple constraints at once.

The middle case is the one that breaks the folk version of the idea. “I was useless today” makes agreement the response that gets delayed, prefaced and hedged, and makes contradiction the fast bare one. The preference belongs to the sequence in front of you, not to yes.

The compliment case is the one with no single answer, and Pomerantz records it as multiple constraints rather than resolving it. Politeness theory, and why please can read as rude takes up the neighbouring question of how one wording comes to land two opposite ways.

In texting the boundary moves from milliseconds to a minute

The same organisation has been looked for in typed conversation. Johanna Rendle-Short published “Dispreferred responses when texting: Delaying that no response” in Discourse and Communication in 2015. The Australian National University’s record of the paper, which carries the published abstract, gives the result: across 329 texting interactions, a response to a request or an invitation that was delayed by more than one minute was much more likely to be a no than a yes, at p less than 0.001. Preferred responses were sent quickly and dispreferred ones were delayed. The abstract describes this as preference organisation similarities between spoken interaction and texting, with texters orienting to social norms concerning delayed responses.

Notice what carried over and what changed. The practices are the same ones: the abstract names silence, prefacing the turn with tokens such as “well”, “uhm” and “uh”, and accounts for why the request cannot be accepted. The unit is not the same. In speech the decisive boundary sat in the low hundreds of milliseconds. In texting the boundary this study reports is a minute, roughly two hundred times longer.

A minute is a thing a chat file can hold. Every message carries a time, which is why the third step of the flow on the Portraitor home page asks which time period you want to analyse and lets you pick a preset or drag a custom range. The second step asks which person in the conversation you want a portrait of, and says their messages are read while everyone else stays context, which is what analysing your own messages rather than someone else’s means in practice. What that page says is looked for is patterns in tone, attachment, communication style and emotional cues; it does not say the interval between two messages is one of them, so neither does this article. Reply speed, and what ten languages say about taking turns sets out what a timestamp in an export will and will not support.

The in-breath is the part no file records

Close to half of the dispreferred responses in the telephone collection, 44.5 percent, began with a pre-beginning component: an in-breath or a click before any word. Only 17.1 percent of preferred responses did. Because breathing is free to overlap the other person’s talk, the audible preparation for a refusal can start before the invitation has finished, which is why the strictest measure of silence makes refusals look slightly early rather than late.

None of that reaches a typed thread. Nor does pitch, and pitch was doing real work in the study: one of the four criteria for a qualified acceptance was a phonetic realisation displaying a negative affective stance, such as low pitch or a compressed pitch range. A written yes has no pitch to compress.

What does reach the thread is every practice made of words. A preface, a hedge, an appreciation and an account are all characters in a message. Portraitor states on its home page that the chat is masked on your device before it is sent, and that what is stripped there is names, numbers, emails and links. On that description the identity inside a sentence goes and the shape of the sentence stays, because “well”, “maybe”, “I think” and the reason someone gives are none of those four things. What masking removes from a chat, and what it misses sets out the limits of that.

The placement practice survives as well, in a form speech does not have: when the account arrives in its own message after a gap, the record shows two messages with two times on them. Conversational repair, and how a chat fixes its own mistakes and Grice’s maxims in texting, and what a short reply implies both work the same seam, where an inference that was carried by a voice has to be carried by a message boundary instead.

Four reasons one long gap proves nothing

First, short gaps are the most frequent setting for a refusal as well as for an acceptance. Kendrick and Torreira put the consequence plainly: the analysis of a response as a dispreferred action cannot rest on the observation of a short delay in production.

Second, what a gap beyond 300 milliseconds made more likely in their data was a dispreferred format, not a dispreferred action. Just over a quarter of the acceptances in the collection were qualified ones, and a qualified acceptance is still an acceptance.

Third, the interval between two messages is not the thing the spoken studies measured. Their measure is the silence between the end of one turn and the first component of the next, which only exists because both people are present and listening. A messaging app requires neither. Rendle-Short’s finding is an association across 329 interactions, and that same abstract says further research is needed to understand in what contexts a person might choose one communicative medium over another.

Fourth, the term is less settled than it reads. Pomerantz’s own note on the University at Albany record says the concept of preference has been used in numerous ways, sometimes to describe disparate phenomena, and that each new study must explicate which phenomenon it is investigating rather than use the concept as if it were adequately explanatory. What a chat analysis cannot tell you about someone is the longer version of that limit.

The honest use of all this is as a question to ask of a thread, never as a verdict about the person who typed it.

Frequently asked questions

Frequently asked questions

The hedges that soften a no are still in your thread. The half second before it never was.

Sources

1. Pomerantz, A. (1984). Agreeing and disagreeing with assessments: some features of preferred/dispreferred turn shapes. Book chapter record, University at Albany Scholars Archive.
2. Kendrick, K. H. and Torreira, F. (2015). The Timing and Construction of Preference: A Quantitative Study. Discourse Processes, pages 255 to 289. Record at White Rose Research Online.
3. Kendrick, K. H. and Torreira, F. (2015). The Timing and Construction of Preference: A Quantitative Study. Author produced full text at White Rose Research Online.
4. Rendle-Short, J. (2015). Dispreferred responses when texting: Delaying that no response. Discourse and Communication 9(6), pages 643 to 661. Record at the Australian National University, carrying the published abstract.
5. Portraitor home page, visible copy read on 1 November 2026.