Annotating Modality Expressions and Event Factuality for a Japanese Chess Commentary Corpus

Annotating Modality Expressions and Event Factuality for a Japanese Chess Commentary Corpus Suguru Matsuyoshi1, Hirotaka Kameko2, Yugo Murawaki3, Shinsuke Mori4 1Graduate School of Informatics and Engineering, The University of Electro-Communications 2Graduate School of Engineering, The University of Tokyo 3Graduate School of Informatics, Kyoto University 4Academic Center for Computing and Media Studies, Kyoto University 1Chofugaoka, Chofu, Tokyo, Japan 2Hongo, Bunkyo-ku, Tokyo, Japan 3,4Yoshida-honmachi, Sakyo-ku, Kyoto, Japan 1matuyosi@uec.ac.jp, 2kameko@logos.t.u-tokyo.ac.jp, 3murawaki@i.kyoto-u.ac.jp, 4forest@i.kyoto-u.ac.jp Abstract In recent years, there has been a surge of interest in the natural language processing related to the real world, such as symbol grounding, language generation, and nonlinguistic data search by natural language queries. We argue that shogi (Japanese chess) commentaries, which are accompanied by game states, are an interesting testbed for these tasks. A commentator refers not only to the current board state but to past and future moves, and yet such references can be grounded in the game tree, possibly with the help of modern game-tree search algorithms. For this reason, we previously collected shogi commentaries together with board states and have been developing a game commentary generator. In this paper, we augment the corpus with manual annotation of modality expressions and event factuality. The annotated corpus includes 1,622 modality expressions, 5,014 event class tags and 3,092 factuality tags. It can be used to train a computer to identify words and phrases that signal factuality and to determine events with the said factuality, paving the way for grounding possible and counterfactual states. Keywords: game commentary, modality, factuality, symbol grounding 1. Introduction The field of natural language processing (NLP) has expe- rienced a resurgence of interest in the symbol grounding problem (Harnad, 1990). An increasing number of datasets available align natural language expressions with real world objects in the form of images and videos (Hashimoto et al., 2014; Mori et al., 2014b; Ferraro et al., 2015), and systems built on top of such datasets typically perform image/video- to-text generation (Ushiku et al., 2011; Yang et al., 2011). These systems are, however, more akin to great apes than to humans in the sense that “[t]heir lives are lived entirely in the present” (Donald, 1991). While there is no evi- dence that nonhuman animals communicate past episodes and planned future events, human language is abundant with them (Szagun, 1978). It is even suggested that the faculty of language has a close connection to the ability to image the future (Suddendorf and Corballis, 1997). Here we argue that commentaries for an extensive-form game (e.g., chess) are an ideal testbed for developing truly intelligent systems. Specifically, we focus on shogi (Japanese chess). Unlike typical image captions, human comments on shogi games are full of references to past and future moves as we will see in Subsection 2.2. Yet, thanks to the well-definedness of the world, many of such refer- ences can be grounded in a game tree. Although ambi- guities inherent in natural language remain a challenging problem, modern game-tree search algorithms (Tsuruoka et al., 2002) help distinguish realistic states from unreal- istic …

Content truncated to 500 words