Response generation reflecting the personality of a chat dialogue system

This paper investigates the relationship between speakers' personalities and their utterances in dialogue, and propose a response generation method that reflects the personality of a chat dialogue system. We construct a corpus of utterances by persons with specific personality features, and calculate percentages of various types of speech act. From the analysis result, features of utterances by various personalities are extracted. With these utterance features, response generation strategies for specific personality features are designed and implemented in a simple chat dialogue system. The experimental result shows that the personality features of "shyness" and "nurturing parent" are appropriately reflected in our dialogue system.