Annotating Event Appearance for Japanese Chess Commentary Corpus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4302–4308 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4302 Annotating Event Appearance for Japanese Chess Commentary Corpus Hirotaka Kameko, Shinsuke Mori Academic Center for Computing and Media Studies, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, Japan {kameko, forest}@i.kyoto-u.ac.jp Abstract In recent years, there has been a surge of interest in natural language processing related to the real world, such as symbol grounding, language generation, and non-linguistic data search by natural language queries. Researchers usually collect pairs of text and non-text data for research. However, the text and non-text data are not always a “true” pair. We focused on the shogi (Japanese chess) commentaries, which are accompanied by game states as a well-defined “real world”. For analyzing and processing texts accurately, considering only the given states is insufficient, and we must consider the relationship between texts and the real world. In this paper, we propose “Event Appearance” labels that show the relationship between events mentioned in texts and those happening in the real world. Our event appearance label set consists of temporal relation, appearance probability, and evidence of the event. Statistics of the annotated corpus and the experimental result show that there exists temporal relation which skillful annotators realize in common. However, it is hard to predict the relationship only by considering the given states. Keywords: game commentary, modality, symbol grounding 1. Introduction These days, the interest in the symbol grounding problems becomes larger and larger. A large num- ber of datasets are now available in such as image and text (Young et al., 2014; Chen et al., 2015). These corpora are widely used for description gener- ation (Vinyals et al., 2015; Xu et al., 2015), non-text information retrieval (Ushiku et al., 2017), and so on. Some of these corpora consist of pairs of non-text data and human-writing texts. We usually treat the pairs that the text and non-text data are compatible. How- ever, the assumption is not always true if the cor- pora are automatically collected. For example, non- text data sometimes reminds humans of some associ- ated things, and they mention to them instead of the exactly given data. To reproduce intelligent systems which work like a human, we must analyze the gap between the text and non-text data. We focused on commentary for extensive-form games like chess. The rules of these games are well-defined, and we can treat them with computer programs eas- ier than the real world. With this well-defined world, computers, can access to the previous states and search the future states for predicting easily. Human play- ers also consider the previous states and future predic- tions, and they often refer to these states when they comment on the game. For accurate analyzing and processing, we must analyze the relationship between texts and the real world for the first step. In this paper, we propose “Event Appearance” la- bel. Event appearance shows the relationship between events that are mentioned …

Content truncated to 500 words