Overrated gaps: Inter-speaker gaps provide limited information about the timing of turns in conversation
Abstract
Corpus analyses have shown that turn-taking in conversation is much faster than laboratory studies of speech planning would predict. To explain fast turn-taking, Levinson and Torreira (2015) proposed that speakers are highly proactive: They begin to plan a response to their interlocutor's turn as soon as they have understood its gist, and launch this planned response when the turn-end is imminent. Thus, fast turn-taking is possible because speakers use the time while their partner is talking to plan their own utterance. In the present study, we asked how much time upcoming speakers actually have to plan their utterances. Following earlier psycholinguistic work, we used transcripts of spoken conversations in Dutch, German, and English. These transcripts consisted of segments, which are continuous stretches of speech by one speaker. In the psycholinguistic and phonetic literature, such segments have often been used as proxies for turns. We found that in all three corpora, large proportions of the segments comprised of only one or two words, which on our estimate does not give the next speaker enough time to fully plan a response. Further analyses showed that speakers indeed often did not respond to the immediately preceding segment of their partner, but continued an earlier segment of their own. More generally, our findings suggest that speech segments derived from transcribed corpora do not necessarily correspond to turns, and the gaps between speech segments therefore only provide limited information about the planning and timing of turns.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the German corpus analyzed by Corps et al., speakers' utterances seamlessly followed each other, with mean and median gap durations close to zero.
“The analyses confirmed that the speakers’ utterances seamlessly followed each other, with mean and median gap durations close to zero.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, the distributions of utterances were highly skewed, with short utterances far more common than long ones; the median was one second, or three words, and the mode was just one word.
“However, the distributions of the utterances were highly skewed, with short utterances being far more common than long ones. The median was one second, or three words, and the mode (the most common utterance length) was just one word.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, 24% of the segments were self-continuations.
“Corps and colleagues found that 24% of the segments were self-continuations.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, 23% of the segments were backchannels.
“In the German corpus analyzed by Corps and colleagues, 23% of the segments were backchannels.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
Further analyses of the corpus showed that many utterances automatically labelled as turns were not complete turns, but only parts of turns.
“Further analyses of the corpus showed that many of the utterances that were automatically labelled as turns were not complete turns, but only parts of turns.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, parallel talk happened often.
“In the corpus discussed here, this happened often”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
When self-continuations were combined with the preceding segment by the same speaker, the average turn duration rose to 6.0 seconds and the median to 3.4 seconds.
“When this was done, the average turn duration rose to 6.0 seconds, and the median to 3.4 seconds.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
After combining self-continuations, the gap between turns remained close to zero, with a mean of -.09 seconds and a median of -0.02 seconds.
“The gap between turns remained close to zero, with a mean of –.09 seconds and a median of –0.02 seconds.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
The analyses by Corps and colleagues showed that speakers often did not use all of the time afforded by the partner's utterance to plan their own turn and launch it shortly before the end of the partner's turn; instead, they often began to speak much earlier.
“The analyses carried out by Corps and colleagues also showed that the speakers often did not use all of the time afforded by the partner’s utterance to plan their own turn and launch it shortly before the end of the partner’s turn. Instead, they often began to speak much earlier.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
This situation most commonly arose when the speakers talked at the same time.
“This situation most commonly arose when the speakers talked at the same time”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, the average duration of utterances was two seconds, corresponding to seven words.
“The average duration of the utterances was two seconds, corresponding to seven words.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
In the German corpus, 24% of the segments stemmed from episodes of parallel talk, and 9% of the turns were fully embedded in longer turns.
“As noted already, 24% of the segments stemmed from episodes of parallel talk, and 9% of the turns were fully embedded in longer turns, i.e. began after and ended before the end of a partner’s turn.”
From Timing in Conversation (2023) · cites this paper as Corps et al., 2022
The citation lists haven’t been fetched yet
References 92 · Cited by 42
- Authors
- Corps, R. E., Knudsen, B., & Meyer, A. S.
- Affiliations
- Max Planck Institute for Psycholinguistics (NL)
- Radboud University Nijmegen (NL)
- Corresponding author
- Corps, Ruth E.
- Venue
- Cognition
- Publisher
- Elsevier BV
- Published
- 2022-06
- Volume
- 223
- Pages
- 105037
- ISSN
- 0010-0277 (ISSN-L)
- OpenAlex
- W4210599218
- Cited by
- 42
- References
- 92
- Citations by year
- 2026: 22 · 2025: 5 · 2024: 4 · 2023: 9 · 2022: 2
- FWCI
- 18.82
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- Cognition (publishedVersion)
- PubMed (publishedVersion)
- White Rose Research Online (University of Leeds, The University of Sheffield, University of York) (acceptedVersion)
- Radboud Repository (Radboud University) (submittedVersion)
- Topics
- Social Sciences › Arts and Humanities › Language and Linguistics › Language, Discourse, Communication Strategies
- Social Sciences › Psychology › Experimental and Cognitive Psychology › Phonetics and Phonology Research
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Keywords
- turn-taking, speech segments, speech preparation, conversational speech, Dutch language, German, English language
- MeSH terms
- Communication, Humans, Language, Phonetics, Psycholinguistics, Speech
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
