Phonetic accommodation of Mandarin-speaking children in human-human interaction and human-robot interaction

Abstract

Speech accommodation involves interlocutors adjusting speech features in response to their partner’s speech. Existing literature has this phenomenon in human-human interaction and human-machine interaction, where accommodation is fulfilling affective and cognitive functions. Choosing accommodation strategies and the degree of accommodation demonstrated huge individual differences. However, the developmental trajectory of child speech accommodation, especially individual variabilities in interactions with human and robot interlocutors, is less understood. This thesis investigates the acoustic features of spontaneous speech in Mandarin-speaking children during interactions with peers or a social robot, aiming to enhance our understanding of developmental patterns and individual variabilities in child speech accommodation. The inclusion of Mandarin-speaking children contributes to cross-linguistic evidence in speech accommodation theories. Study I involves a child-child interaction study where two children identify differences between their pictures using only speech. Study II, a child-robot interaction, maintains a similar design with a different partner (the robot). In both studies, children were recorded saying a group of keywords before and after the interaction. The keywords produced during the interaction were also recorded and extracted for analysis. Acoustic analysis in Study I revealed post-interaction convergence in fundamental frequency (f0), vowel duration, and F1, with divergence in vowel space area (VSA) and divergence in f0 during interaction. Divergence in f0 during interaction was also reported. Study II displayed post-interaction convergence in f0, vowel duration, and VSA, and divergence in F1. Convergence in F0 and divergence in F1 during interaction were observed. Personality traits’ role in predicting individual variabilities and speech accommodation degree was explored in both studies. Self-evaluated personalities mainly predicted adjustments, and evaluations of own and partner’s traits guided accommodation. The reported personality of the conversation partner had a minor effect in Study I while in Study II, the addition of social responses increased f0 accommodation with the robot. The suggestion is that certain patterns of speech accommodation are transferable from interactions with humans to interactions with robots. However, there are specific features observed in interactions with robots that do not necessarily derive from interactions with humans. Taken together, children were found to accommodate their speech with both their peers and a social robot, indicating that their development of speech accommodation aligns with their early development of social communication skills. The dynamic interplay model was used to interpret the findings and extended to incorporate the concept that the nature of the conversation determines the dominant function of speech accommodation. This, in turn, dictates the corresponding accommodation strategies, manifested by the respective proportions of convergence and divergence during the interplay. Additionally, this process demonstrated cue-independent features and significant individual variabilities, influenced by personality traits. This study provides new insight in child language acquisition and enriches the theories of speech accommodation.