Recipes for enjoy-talking conversational systems
Development of transformer-based conversational systems
We are studying a social dialogue system that satisfies people's desire for dialogue through natural conversation.
We have developed a deep-learning-based Japanese social dialogue system, which is pre-trained with the largest-scale Japanese dialogue data obtained from Twitter and fine-tuned with high-quality dialogue data that NTT has cultivated over many years of research. We also conducted a quantitative analysis of the utterances of the constructed system and identified remaining issues such as contradictions and discontinuous topics.
We believe that the desire to communicate with others is one of our fundamental desires. We aim to realize a social dialogue system as a partner that continuously satisfies this need for dialogue.
 H. Sugiyama, H. Narimatsu, M. Mizukami, T. Arimoto, Y. Chiba, T. Meguro, H. Nakajima, , “Development of conversational system talking about hobby using Transformer-based encoder-decoder model,” in Proc. Special Interest Group on Spoken Language Understanding and Dialogue Processing (SIG-SLUD), Vol. B5, No. 02, pp. 104-109, 2020 (in Japanese).
 H. Sugiyama, T. Meguro, Y. Yoshikawa, J. Yamato, "Improving Dialogue Continuity using Inter-Robot Interaction," in Proc. IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), pp. 105-112, 2018.
Hiroaki Sugiyama / Interaction Research Group, Innovative Communication Laboratory