Japanese discourse parsing matches English in dialogue structure analysis
Japanese Discourse Structure Analysis Based on Segmented Discourse Representation Theory Using LLMs
Computation and Language
Summary
Understanding how parts of a conversation relate to each other is important in language technology. The authors show it is possible to analyze Japanese dialogue structure as well as English by translating data and using large AI language models. They found that the models rely more on the flow of the conversation than specific Japanese markers, which might explain why the approach works across very different languages.
What this means in practice
- •For natural language processing teams: Improve Japanese dialogue systems by integrating discourse parsing models comparable to English ones for better conversational understanding.
- •For machine translation developers: Develop cross-lingual discourse-aware translation tools that maintain structural dialogue coherence between English and Japanese.
Authors
Satomu Tomisaki, Yosuke Mikami, Amane Watahiki, Hitomi Yanaka
Abstract
Discourse parsing based on Segmented Discourse Representation Theory (SDRT) is crucial for capturing the structural dynamics of dialogues. However, most prior research focuses on English, and its effectiveness for Japanese remains unclear owing to the lack of Japanese SDRT corpora. To address this issue, we construct Japanese SDRT corpora by translating existing English SDRT corpora using an LLM. Through a comparative evaluation of prompt-based and fine-tuned LLMs, we show that the models achieve comparable performance between English and Japanese, suggesting the cross-lingual applicability of SDRT-based discourse parsing. In order to investigate why the English and Japanese parsers achieve similar performance, we conduct ablation analyses on discourse markers and contextual inputs. Our results reveal that most of the Japanese discourse markers we examined have little impact, and the models' predictions depend heavily on contextual discourse relations rather than on the content of the context clauses. This structural reliance might explain the comparable performance despite the substantial typological differences between English and Japanese.