Reply speed: what ten languages say about taking turns
A 2009 study of ten languages found the same turn-taking pattern everywhere, with the whole cross-language spread inside 250 milliseconds. Text does not work that way at all.
Stivers and colleagues compared turn-taking in ten languages in 2009 and reported striking universals: every language they tested showed a general avoidance of overlapping talk and a minimisation of silence between turns, with the differences in average gap falling within a range of 250 milliseconds from the cross-language mean.
The anthropological literature had long reported that some cultures leave long silences between turns and others talk over each other, and that these are deep differences in conversational style.
A team at the Max Planck Institute for Psycholinguistics set out to test that claim, and reported something close to the opposite.
What Stivers and colleagues measured across ten languages in 2009
The paper is Universals and cultural variation in turn-taking in conversation, published in the Proceedings of the National Academy of Sciences on 30 June 2009 by Tanya Stivers, N. J. Enfield, Penelope Brown, Christina Englert, Makoto Hayashi, Trine Heinemann, Gertie Hoymann, Federico Rossano, Jan Peter de Ruiter, Kyung-Eun Yoon and Stephen C. Levinson.
They describe their material as a worldwide sample of ten languages, drawn from traditional indigenous communities through to major world languages. The question they set was whether the cultural differences reported in the anthropological literature show up when response latency is actually measured.
Every language avoided overlap and minimised the silence
The finding, in the authors’ own framing, is that there are striking universals in the underlying pattern of response latency. All of the languages tested provided clear evidence for a general avoidance of overlapping talk and a minimisation of silence between conversational turns.
They also report that all of the languages show the same factors explaining the variation in speed of response within each language. That second point is the less quoted one and it is the stronger claim: not only is the target the same everywhere, the things that push a speaker above or below it are the same everywhere too.
The whole spread across languages fits inside 250 milliseconds
There were differences. The paper reports differences across the languages in the average gap between turns, within a range of 250 milliseconds from the cross-language mean.
A quarter of a second is a real difference to sit inside, and the authors suggest that a natural sensitivity to these tempo differences is what produces the subjective perception of dramatic or even fundamental differences recorded in ethnographic reports. Their conclusion is that the local variations are quantitative only, pointing to a single shared infrastructure for language use.
One scoping note that matters for anything you might want to do with this. The published abstract gives the spread around the mean and does not give the mean itself, so this page will not state a single universal figure for how long a gap is. The claim that holds is about how little the languages differ, not about the absolute number.
Text removes the pressure the system was built around
Spoken turn-taking is under load. The floor is occupied, silence is audible to everyone, and a delay has to be paid for immediately. That is the setting the study measured, and it is the setting the shared infrastructure appears to be tuned for.
A written conversation removes every part of that. Nobody holds the floor, silence costs nothing in the moment, and a reply can arrive in two seconds or two days without interrupting anyone. The pressure that keeps spoken gaps inside a narrow band simply is not acting.
Which is why reply delay in a chat reads as meaning rather than as mechanics. In speech a long gap is marked because the system minimises silence; in text there is no comparable baseline, so a reader supplies one, usually from how they themselves would have replied. The research on how badly senders judge their own messages is the closest thing to a correction for that, and accommodation between two writers is the other half of how a pair settles into a pattern.
What a timestamp in an export can carry, and what it cannot
A chat export does hold the times. Portraitor’s home page offers to process all messages or the last twelve, six or three months, the last month, or a custom range you drag, which is only possible because the dates in the file are read.
What those times will not give you is a turn-taking measurement in the sense this study means. A gap in a thread mixes attention, availability, sleep and intent, and none of those is separable from the others after the fact. The study measured conversations where both people were present and the clock was running for both; a chat is not that.
There is a second limit that is specific to how the product works, and it is worth being plain about. The home page says you pick the one person you want a portrait of, that their messages are read, and that everyone else stays context. A reading built that way describes one person’s own pattern, not the exchange between two people, so it is not the right instrument for a question about turn-taking even in principle. Choosing the window and what the reading actually covers both describe that boundary, and the home page states it before you upload anything.
Why early messages carry so many questions is the subject of uncertainty reduction theory, read in a message thread.
The move that takes a turn away from the other person and back to the speaker has a name and a literature of its own, set out in conversational narcissism and the shift response.
Frequently asked questions
Frequently asked questions
How long is a normal gap between turns in conversation?
The published abstract of the 2009 study reports the spread rather than the mean: differences across the ten languages in the average gap, within a range of 250 milliseconds from the cross-language mean. It does not state the mean itself, so no single universal figure is quoted here.
Did any language turn out to be slower than the rest?
There were differences in average gap between the languages, but the authors describe them as quantitative only and place the whole spread within 250 milliseconds of the cross-language mean.
Does this research apply to text messages?
Not directly. The study measured spoken conversation, where silence is audible to both people and the floor is occupied. A written exchange removes that pressure, which is why the finding is interesting for text as a contrast rather than as a rule.
Can a chat analysis measure how fast I reply?
A file holds the times, but a reply gap in a thread mixes attention, availability and intent, and Portraitor reads one chosen person’s messages rather than the exchange between two people. It is not an instrument for turn-taking.
Expectations about how a conversation should run are the subject of their own literature. Expectancy violations theory and the missing expectation sets out how a departure from an expectation is judged by its direction and by who made it, and what a written thread keeps of that.
Ten languages agreed on how quickly to answer, within a quarter of a second. Text is the first form of conversation that lets that agreement lapse entirely.
Your conversations, read the way you would read someone else’s.
Portraitor takes a chat export you already have, anonymises it in your browser, and returns a structured reading of how you come across. Identifying details are removed in your browser before anything is sent.
See how it works →- Stivers, T., Enfield, N. J., Brown, P., Englert, C., Hayashi, M., Heinemann, T., Hoymann, G., Rossano, F., de Ruiter, J. P., Yoon, K.-E., & Levinson, S. C. (2009). Universals and cultural variation in turn-taking in conversation. Proceedings of the National Academy of Sciences, 106(26), 10587to 10592.
- Portraitor, home page.