Explore prosodic markers for sociability, engagement and trust in human and robot speech.
Accents and speaking styles have a major influence on how people are perceived. In many circumstances, the information content of speech is relatively peripheral to its role in promoting social bonds between individuals and within larger groups. The goal of this project is to try and understand the prosodic properties of speech that influence our sense of engagement and trust in other people, and to apply them to our interactions with speaking robots.
There is a wealth of evidence that the accent of a speaker influences our belief in what they say. This goes beyond the factors of social valence and stereotyping, as infants have also been shown to be strongly affected by a speaker’s accent. Do accents and speaking styles associated with engagement and trust exhibit distinctive prosodic properties? Infant-directed speech and persuasive oratory in adults are perceived as relatively regular in their timing, more closely resembling music than does typical conversation. Is it possible that the accents associated with positive affect may exhibit such distinctive prosodic properties? If so, could we modulate the prosody of synthetic speech to improve our interactions with speaking robots? It is well established in robot-human interaction that the expression of affective states alongside verbal information will yield a more rewarding interaction, as users perceive the robot to have a personality that resonates with their social instincts. This is evident in the increased neural and behavioural synchrony when using familiar accents in human-machine interfaces. As spoken interactions tend to be marked by convergence between speakers at multiple levels – phonetic, prosodic, gestural – it may be that prosodic properties which promote such convergence are socially favourable. In which case, are there distinctive prosodic styles that will similarly promote engagement in robot-human interaction? To address these questions we aim to identify – using behavioural and electrophysiological techniques – the prosodic properties of speech that promote engagement and trust. These will be implemented in synthetic speech and tested in robot-human interactions, seeking to develop mechanisms for dynamic inter-speaker adaptation and thereby more fulfilling and successful interactions.
Jeremy Goslin, Laurence White, Caroline Floccia, Tony Belpaeme, Angelo Cangelosi (Plymouth University)