Automatic conversion of sentence-end expressions for utterance characterization of dialogue systems

摘要

Building characters for dialogue agents is important in making the agents more friendly and human-like. To build such characters, utterances suitable for the designated characters are usually manually prepared. However, it is expensive to do this for a large number of utterances for various types of characters. We propose a method for automatically converting system utterances into those that are characteristic of designated personal attributes, such as gender, age and area of residence, to characterize agents. In particular, we focus on converting sentence-end expressions, which are considered to greatly affect personal attributes in Japanese. Conversion is done by (ⅰ) automatically collecting conversion candidates from various utterances on the Web (e.g., Twitter postings), and (ⅱ) using syntactic and semantic filters to suppress the generation of ill-formed utterances. Experimental results show that our method can convert approximately 95% of utterances into those that are grammatically and semantically acceptable and approximately 90% of utterances into those that are perceived to be acceptable for designated personal attributes.