Investigation of Social Factor in Conversational Entrainments

Social aspects such as social roles and status play a crucial role in human-to-human conversations, where the interlocutors adapt to each other to achieve conversational entrainment and gain approval. Due to their complexity, these aspects are challenging to quantify in current dialogue systems and human-machine interaction systems. To simplify this problem, we assume that social aspects have a consistent effect on the same style of conversational scenarios. Therefore, we define the concept of “social factor” to measure the quantitative influence of social aspects on conversational entrainments, and propose a method to extract the social factor in different conversation styles. To do so, we designed a Chinese corpus with four conversation scenarios, arguing, comforting, convincing, and sharing happiness to investigate the importance of the social factor. We also employed an existing English corpus to predict the trajectory of conversational entrainment using the social factor. The importance of the social factors was evaluated by comparing the proposed method with a conventional method using speech features. The accuracy for classifying conversation scenarios in Chinese corpus was 52.9% by using the proposed social factor, and 51.7% by using the conventional speech features. For the English corpus, the accuracy of predicting the trajectory of conversational entrainment was 48.8% by using social factor, and 49.0% by conventional speech features. These results indicate that the social factor plays the same importance as that of the speech features. When combining speech features and social factor, the accuracy increases 6.1% in the Chinese corpus, and 2.0% in the English corpus. This suggests that social factor and speech features have distinguishable information that can compensate for each other. This study demonstrated that the social factor may be important in quantifying the pragmatic information involved in the conversations.