Synchrony as a measure of conversation difficulty: Movement coherence increases with background noise level and complexity in dyads and triads

Abstract

When people interact, they fall into synchrony. This synchrony has been demonstrated in a range of contexts, from walking or playing music together to holding a conversation, and has been linked to prosocial outcomes such as development of rapport and efficiency of cooperation. While the basis of synchrony remains unclear, several studies have found synchrony to increase when an interaction is made challenging, potentially providing a means of facilitating interaction. Here we focus on head movement during free conversation. As verbal information is obscured when conversing over background noise, we investigate whether synchrony is greater in high vs low levels of noise, as well as addressing the effect of background noise complexity. Participants held a series of conversations with unfamiliar interlocutors while seated in a lab, and the background noise level changed every 15-30s between 54, 60, 66, 72, and 78 dB. We report measures of head movement synchrony recorded via high-resolution motion tracking at the extreme noise levels (i.e., 54 vs 78 dB) in dyads (n = 15) and triads (n = 11). In both the dyads and the triads, we report increased movement coherence in high compared to low level speech-shaped noise. Furthermore, in triads we compare behaviour in speech-shaped noise vs multi-talker babble, and find greater movement coherence in the more complex babble condition. Key synchrony differences fall in the 0.2-0.5 Hz frequency bands, and are discussed in terms of their correspondence to talkers' average utterance durations. Additional synchrony differences occur at higher frequencies in the triads only (i.e., >5 Hz), which may relate to synchrony of backchannel cues (as multiple individuals were listening and responding to the same talker). Not only do these studies replicate prior work indicating interlocutors' increased reliance on behavioural synchrony as task difficulty increases, but they demonstrate these effects using multiple difficulty manipulations and across different sized interaction groups.

Fig 1. Study setup.

(A) shows the positions of loudspeakers and participant chairs for dyads, and (B) shows positions for triads. (C) shows an example of the noise level changes as a function of time (varying between 54, 60, 66, 72, and 78 dB).

Fig 2. Dyads in speech-shaped noise.

Top plots of (A) and (B) show the average wavelet coherence responses for Real vs Pseudo during 54 dB and 78 dB respectively. The mean coherence over all dyads (with standard error) is shown across a range of frequencies. The lower plots show the corresponding Pseudo-Real effect size (Cohen’s d), with both significance and FDR-adjusted significance (if any). (C) shows the corrected (baseline removed) coherence difference between 78 and 54 dB.

Fig 3. Triads in speech-shaped noise.

Top plots of (A) and (B) show the average wavelet coherence responses for Real vs Pseudo during 54 dB and 78 dB respectively. The mean coherence over all triads (with standard error) is shown across a range of frequencies. The lower plots show the corresponding Pseudo-Real effect size (Cohen’s d), with both significance and FDR-adjusted significance. (C) shows the corrected (baseline removed) coherence difference between 78 and 54 dB.

Fig 4. Triads in babble.

Top plots of (A) and (B) show the average wavelet coherence responses for Real vs Pseudo during 54 dB and 78 dB respectively. The mean coherence over all triads (with standard error) is shown across a range of frequencies. The lower plots show the corresponding Pseudo-Real effect size (Cohen’s d), with both significance and FDR-adjusted significance. (C) shows the corrected (baseline removed) coherence difference between 78 and 54 dB.

Fig 5. Triads in speech-shaped noise vs babble.

Top plots of (A) and (B) show corrected (baseline removed) coherence responses comparing SSN vs Babble conditions for 54 and 78 dB. Cohen’s d effect sizes, with corresponding significance points (if any), are shown in the lower plots.