We analyzed the relationship between speaker’s personalities and utterance features using dialogue corpus with personality information. We considered two kinds of personality: personality traits obtained from personality tests, and personality impressions inferred by other people from dialogue logs. We considered occurrence frequency of dialogue acts and specific vocabulary categories as utterance features. The results of the correlation analysis showed that while there were weak correlations between personality traits and utterance features in dialogue, many utterance features had moderate correlations with personality impressions. To realize a dialogue system that gives users a specific personality impression, we constructed a response generation model from pre-trained language model T5 by fine-tuning using this dialogue corpus and personality impression data and conducted an experiment on the reproducibility of personality. The experimental results showed that our response generation model can generate responses that reflect the high or low personality impression specified at the time of generation.
Sugimoto et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: