Conversational Turn-Taking: The 200-Millisecond Problem
Ask a question and the reply usually starts before you would think anyone had time to think of one. That gap is one of the most tightly timed things people do, and the timing research treats it as a cousin of keeping a beat.
Short answer: across a wide range of languages, the typical gap between one speaker finishing a turn and the next speaker starting clusters close to 200 milliseconds. That is far too fast for the several hundred milliseconds that planning a spoken reply from scratch normally takes, which means listeners are not waiting for silence and then thinking of an answer. They are predicting when the current turn will end and preparing a response in advance, timing that looks less like a reaction and more like the anticipation behind tapping right on a musical beat.
A universal, fast, tightly timed gap
Stivers and an international team of collaborators set out to test whether the folk idea of cultures having very different conversational tempos held up under careful measurement. They analyzed response timing in ordinary conversation across ten languages, drawn from a genuinely wide sample, from small-scale indigenous communities to some of the most widely spoken languages in the world, and published the results in the Proceedings of the National Academy of Sciences in 2009.
Two things stood out. First, every language sampled showed a strong preference for avoiding overlapping speech, one person finishing before the next clearly begins. Second, response offsets to questions clustered in a similar, fast range across the whole sample, on the order of 200 milliseconds. There was real cultural variation in exactly how fast that typical gap was, cultures were not identical, but the underlying pattern, tight timing and overlap avoidance, showed up everywhere the team looked. That combination, a shared structure with cultural texture layered on top, is the paper's title in a sentence: universals and cultural variation, together.
The planning-speed problem
Two hundred milliseconds is a genuinely awkward number once you know how long spoken language production normally takes. Retrieving a word and preparing to say it typically eats up several hundred milliseconds on its own, well before a whole sentence gets assembled. If a listener only started planning a reply after the other person fell silent, the observed gaps should be considerably longer than what Stivers and colleagues measured.
Levinson and Torreira laid out an influential resolution to that puzzle, publishing their model in Frontiers in Psychology in 2015. Their proposal is that listeners begin formulating a response while the current speaker is still talking, as soon as enough of the turn has been understood to know roughly what needs saying. That plan is then held ready, buffered, until the listener can predict the turn is about to end, at which point the prepared response is released. The 200-millisecond gap, on this account, is mostly the time needed to detect the end of the turn and launch an already-prepared utterance, not the time needed to invent one.
Not a reaction, a prediction
A direct comparison makes the point sharply. Torreira and Bögels measured how fast people could produce a simple vocal response to a plain, non-linguistic cue, a tone stopping, the kind of bare reaction-time task used across timing research generally. That plain reaction time turned out to be slower than the gaps typically observed in real conversation.
Read plainly, that result rules out the simplest explanation for fast conversational turn-taking. People are not simply reacting quickly to silence. Whatever produces a 200-millisecond turnaround has to involve anticipating the moment before it arrives, because waiting for the moment and then reacting to it is measurably too slow to explain what conversation actually does. Our guide to anticipation versus reaction covers the same logical structure in a musical setting: taps that land on or slightly ahead of a beat are evidence of prediction, because a purely reactive tap would always land late. Speech and music are different systems studied by different research communities, but both point to the same underlying lesson: tight, reliable timing in a live exchange is very hard to produce by reacting, and much easier to produce by anticipating.
What makes the gap stretch or shrink
Not every reply comes in at 200 milliseconds, and the variation is informative. Roberts, Torreira and Levinson analyzed a corpus of naturally occurring conversation for Frontiers in Psychology, also in 2015, and related the timing of responses to how demanding they were to produce and how the surrounding talk was organized. Responses that were harder to plan, or that ran against what the previous turn seemed to be angling for, disagreements, hedged answers, dispreferred responses, tended to come with longer gaps or with disfluency standing in for the gap.
This is worth flagging clearly as a different kind of evidence than the reaction-time comparison above. It comes from a corpus of talk that happened on its own, not a manipulated experiment, so it describes a correlation between response complexity and response timing rather than proving that complexity directly causes the delay. It is a strong, consistent correlational pattern, and a reasonable one given the planning-speed logic above, but it sits in a different evidential category than a controlled comparison.
| Measurement | What it shows | Design |
|---|---|---|
| Cross-linguistic response offsets, 10 languages | Universally fast, with real cultural variation in the exact typical gap | Cross-linguistic corpus comparison |
| Word production time in isolation | Several hundred milliseconds, longer than the typical turn-taking gap | Established psycholinguistic estimate |
| Plain vocal reaction time to a non-speech cue | Slower than real conversational gaps | Controlled within-subject comparison |
| Gap length versus response complexity | Harder, dispreferred responses associate with longer gaps | Corpus-observational, correlational |
What this has in common with keeping a beat
None of this research claims that talking and tapping share the same brain machinery in any strong sense; that is a separate and much harder question. What the two literatures share is a logical move. In both cases, researchers noticed that a purely reactive account is too slow to explain the observed timing, and in both cases the resolution was to show that people are running ahead of the signal rather than behind it. Our guide to the metronome effect in speech covers a related but different place where rhythm and talking intersect, external pacing changing fluency, which is a separate finding from the turn-taking timing covered here.
A pulse plays, then goes quiet, and you have to pick up right where it left off. Chronecho asks you to hold a tempo through a silent gap and resume it from memory. It is a fair miniature of what happens the instant someone else stops talking and it becomes your turn: you were already tracking where things were headed before the silence arrived.
Frequently asked questions
How fast do people actually respond in conversation?
Stivers and colleagues measured response offsets across a worldwide sample of ten languages and found a striking cross-linguistic pattern: replies to questions cluster close to 200 milliseconds after the previous turn ends, with all of the languages sampled showing a strong preference against overlapping speech. There is real cultural variation in the exact typical gap, but the general pattern, fast responses and overlap avoidance, showed up everywhere they looked.
Is 200 milliseconds enough time to plan a reply?
Not if planning started after the other person stopped talking. Simple word production alone typically takes several hundred milliseconds, well beyond the observed gap. Levinson and Torreira's model resolves the puzzle by proposing that listeners start planning their response while the other person is still talking, then hold that plan ready and release it once they can predict the current turn is about to end.
So is answering someone more like predicting than reacting?
That is the core claim, and it mirrors a pattern found in a completely different domain, tapping to a musical beat, where taps land on or slightly before the beat because the tapper is anticipating it rather than reacting to it after the fact. Our guide to anticipation versus reaction covers that musical case. Turn-taking research reaches a similar conclusion about speech through its own separate evidence, cross-linguistic timing data and comparisons with plain reaction time, not by borrowing the music finding directly.
Does the gap change depending on what is being said?
Yes. Roberts, Torreira and Levinson analyzed a corpus of natural conversation and found that response timing varies with how demanding the response is to plan and with how the exchange is organized, harder or more dispreferred responses, disagreements, awkward answers, tend to come with longer gaps or more disfluency. This is a correlational, corpus-based finding: it describes a real pattern in naturally occurring talk rather than a manipulated experiment.
Is a 200-millisecond reply just a fast reaction time?
No, and this is one of the more surprising findings in the area. Torreira and Bögels measured how quickly people could produce a vocal response to a simple, non-linguistic cue, a sound stopping, and found that plain reaction time to that kind of cue was actually slower than typical conversational turn-taking gaps. Real conversation is not winning a reaction-time race. It is arriving prepared.
Why does an awkward silence in conversation feel so noticeable?
Because it violates a timing expectation that appears to be almost universal. If listeners are routinely predicting a turn's end and readying a response within roughly 200 milliseconds, a longer-than-expected pause is a detectable break from that pattern, which is very likely why delayed responses in conversation are so reliably read as hesitation, trouble, or an unwelcome answer on the way.
Keep reading
- Anticipation versus reaction: the musical version of predicting a moment instead of waiting for it.
- The metronome effect in speech: a different place where external timing changes how people talk.
- Tempo in everyday life: how speech rate itself varies between languages and settings.
Sources: Stivers, T., Enfield, N. J., Brown, P., Englert, C., Hayashi, M., Heinemann, T., Hoymann, G., Rossano, F., de Ruiter, J. P., Yoon, K.-E. and Levinson, S. C. (2009), Universals and cultural variation in turn-taking in conversation, Proceedings of the National Academy of Sciences; Levinson, S. C. and Torreira, F. (2015), Timing in turn-taking and its implications for processing models of language, Frontiers in Psychology; Roberts, S. G., Torreira, F. and Levinson, S. C. (2015), The effects of processing and sequence organization on the timing of turn taking: a corpus study, Frontiers in Psychology; Torreira, F. and Bögels, S. (2022), Vocal reaction times to speech offsets: implications for processing models of conversational turn-taking, Journal of Phonetics.