Minimally Supervised Learning of Affective Events Using Discourse Relations

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 5758–5765, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 5758 Minimally Supervised Learning of Affective Events Using Discourse Relations Jun Saito Yugo Murawaki Graduate School of Informatics, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501, Japan {saito, murawaki, kuro}@nlp.ist.i.kyoto-u.ac.jp Sadao Kurohashi Abstract Recognizing affective events that trigger pos- itive or negative sentiment has a wide range of natural language processing applications but remains a challenging problem mainly be- cause the polarity of an event is not neces- sarily predictable from its constituent words. In this paper, we propose to propagate affec- tive polarity using discourse relations. Our method is simple and only requires a very small seed lexicon and a large raw corpus. Our experiments using Japanese data show that our method learns affective events effec- tively without manually labeled data. It also improves supervised learning results when la- beled data are small. 1 Introduction Affective events (Ding and Riloff, 2018) are events that typically affect people in positive or negative ways. For example, getting money and playing sports are usually positive to the experi- encers; catching cold and losing one’s wallet are negative. Understanding affective events is impor- tant to various natural language processing (NLP) applications such as dialogue systems (Shi and Yu, 2018), question-answering systems (Oh et al., 2012), and humor recognition (Liu et al., 2018). In this paper, we work on recognizing the polarity of an affective event that is represented by a score ranging from −1 (negative) to 1 (positive). Learning affective events is challenging be- cause, as the examples above suggest, the po- larity of an event is not necessarily predictable from its constituent words. Combined with the unbounded combinatorial nature of language, the non-compositionality of affective polarity entails the need for large amounts of world knowledge, which can hardly be learned from small annotated data. In this paper, we propose a simple and effective method for learning affective events that only re- quires a very small seed lexicon and a large raw corpus. As illustrated in Figure 1, our key idea is that we can exploit discourse relations (Prasad et al., 2008) to efficiently propagate polarity from seed predicates that directly report one’s emotions (e.g., “to be glad” is positive). Suppose that events x1 are x2 are in the discourse relation of CAUSE (i.e., x1 causes x2). If the seed lexicon suggests x2 is positive, x1 is also likely to be positive because it triggers the positive emotion. The fact that x2 is known to be negative indicates the negative po- larity of x1. Similarly, if x1 and x2 are in the dis- course relation of CONCESSION (i.e., x2 in spite of x1), the reverse of x2’s polarity can be prop- agated to x1. Even if x2’s polarity is not known in advance, we can exploit the tendency of x1 and x2 to be of the same polarity (for CAUSE) …

Content truncated to 500 words