[3Xin4] Poster session 1

[3Xin4-46] Improvement of Incomplete Utterance Rewriting with Multi-stage Transfer Learning

〇Ziqi YIN¹, Daisuke Kawahara¹ (1.WASEDA University)

Keywords:NLP, Incomplete Utterance Rewriting

Because of the frequent use of pronouns in everyday human dialogue, many utterances have incomplete information. Humans can understand by complementing incomplete utterances based on context. Dialogue systems, however, find such a task difficult, which leads to the requirement of construction and training of large datasets. This paper proposes a method for incomplete utterance rewriting in Japanese by constructing only a small dataset and applying multi-stage transfer learning. Our experiments confirmed the effectiveness of the proposed method.