How should turn-taking be measured so that human and animal communication timing can be compared?

Why the meaningful moment of a signal is the right anchor

A turn is not a single instant. A sign, like a spoken word, unfolds over time: the chimpanzee study broke each turn into preparation, stroke, hold, and retraction, with the stroke carrying the lexical meaning [2]. If you time from the beginning of the whole movement, you mix in wind-up and follow-through that carry no message, and two species with different movement styles will look artificially different. Anchoring on the stroke lets you ask the same question of both partners: how long after one meaningful signal does the next meaningful signal begin?

The same logic applies to overlap. The chimpanzee study predicted and found that overlap was less likely during the stroke and more likely during other phases, and both chimpanzees significantly avoided signing while the human partner was mid-stroke [2]. That is a structural measure—where in the signal the overlap falls—not just a total count. It tells you whether partners are protecting the meaningful part of each other's turns, which is a comparison you can make across species even when the signals themselves differ.

Report latency and overlap rate together

The chimpanzee study reported two complementary numbers. The mean stroke-to-stroke latency was about 623 ms for the chimpanzees and about 1,135 ms for the humans, a difference that was not statistically significant [2]. Separately, the chimpanzees overlapped 28.75% of the time versus 18.46% for humans [2]. Reporting both matters because they answer different questions: latency describes the typical gap between turns, while overlap rate describes how often turns collide. A species could match on average latency yet differ in how often it interrupts.

The study also shows why averages alone can mislead. The two chimpanzees differed sharply: Tatu averaged about 423 ms with a very wide response range, while Moja averaged about 1,165 ms with a narrower range [2]. A single species-level average would have hidden that. Reporting individual ranges alongside the mean keeps the comparison honest, especially with small numbers of participants.

Decide what counts as a turn before you time it

Timing only becomes comparable once you have defined the units. In the chimpanzee work, turns were annotated in ELAN software from an archival video database, covering 256 ASL turn transitions between two cross-fostered chimpanzees and their human caregivers [2]. The annotation scheme—preparation, stroke, hold, retraction—was applied to both species, which is what makes the timing numbers directly comparable rather than two separate measurements stitched together.

Human-robot work approaches the same problem from the other direction. One system used a simple LED display to signal turn-taking timing to a person, then measured response timing to see how that cue affected the exchange [1]. That is a measurement of response latency to a deliberately controlled timing signal, which is useful for pinning down how sensitive turn-taking is to timing cues—but it measures a person reacting to a robot's display, not a natural conversation between peers. The comparison to animal timing is therefore looser: the robot studies calibrate how humans respond to timing cues, while the chimpanzee study measures spontaneous timing in real signed exchanges [1][2].

What the timing patterns may reflect, and what they do not settle

The chimpanzee findings are consistent with a one-at-a-time principle: partners tend to avoid simultaneous meaningful signals, and holds serve as a space where overlap can occur, much as in human signed conversation [2]. That is a description of structural timing, not proof of a shared underlying mechanism. A review of turn-taking across mammals argues that the neuroscience of human turn-taking and duetting could inform animal work, and points to the short time scales involved as a central feature [4]. That is a proposal about where to look, not a demonstrated cause.

Computational work offers another angle on anticipation. In simulated full-duplex dialogues between two instances of a speech model, internal states carried information that supported predicting turn-taking ahead of time, and representational synchronization peaked near zero lag under no-noise conditions and degraded as noise increased [5]. These are model-to-model simulations, not human or animal conversations, so they speak to how anticipatory timing could be measured in a controlled system rather than to how chimpanzees or people actually time their turns. Used carefully, such methods suggest a third measurable quantity—how far ahead a partner's internal state anticipates the next turn—that could in principle be probed in biological systems too.

Sources used in this answer

Related research questions