Conversation Turn-Taking (2D)

A timing model only. Not a test of anyone, not a diagnosis, not advice.

Switch to 3D

Presets

Partner A (gold) and partner B (blue)

Clock

Independent numerical checks

The science

This is a small timing model of two people taking turns. Each turn lasts at least 1 s and on average D seconds (a shifted exponential: D = 1 + Exp(mean D - 1)). The next speaker starts after a gap g ~ Normal(mu, sigma) measured from the end of the previous turn: positive g is a pause, negative g is a start slightly before the other has finished (an overlap). Conversation analysts report that, across many languages, typical gaps between turns are very short, on the order of a few hundred milliseconds (Stivers et al., 2009, report averages roughly between 0 and 500 ms depending on the language; check the paper before quoting a figure).

cycle = D_A + D_B + mu_A + mu_B, P(overlap) = Phi(-mu/sigma), E[overlap] = sigma phi(mu/sigma) - mu Phi(-mu/sigma), E[pause] = sigma phi(mu/sigma) + mu Phi(mu/sigma), and the time budget A only + B only + both + nobody = 1.

The 2D view

The swimlane timeline shows each partner's talk as a bar and the overlaps in red; the budget bar splits time into four states; the map shows the probability that a reply starts early for every typical gap and variability (the contours are the closed form Phi(-mu/sigma)). Dots mark the current partners.

What this is not

It is not a test of any person and it does not diagnose, treat or give advice about any condition. There is no "right" setting: people, cultures, languages, topics and moods differ widely, many people need more time to think or to find words, and longer pauses are often perfectly comfortable. Real conversation also involves eye contact, gesture, tone, shared history and meaning, none of which are modelled. If a conversation or social situation worries you or someone you know, talking to a trusted person or a qualified professional is the right step, not a simulation. Model limits: independent gaps, a fixed turn-length law, alternate turns only (no three-way talk or back-channel), a 0.3 s minimum advance rule that is almost never triggered at the defaults.

Gap distribution: measured vs normal model
Talk share of A: running value vs prediction
Turn lengths: measured vs shifted exponential
Made with 🟠 Blender