Extracting Causes of Emotions from Text

International Joint Conference on Natural Language Processing, pages 932–936, Nagoya, Japan, 14-18 October 2013. Extracting Causes of Emotions from Text Alena Neviarouskaya Toyohashi University of Technology Toyohashi, Japan alena@kde.cs.tut.ac.jp Masaki Aono Toyohashi University of Technology Toyohashi, Japan aono@kde.cs.tut.ac.jp Abstract This paper focuses on the novel task of auto- matic extraction of phrases related to causes of emotions. The analysis of emotional causes in sentences, where emotions are explicitly indi- cated through emotion keywords can provide the foundation for research on challenging task of recognition of implicit affect from text. We developed a corpus of emotion causes specific for 22 emotions. Based on the analysis of this corpus we introduce a method for the detec- tion of the linguistic relations between an emotion and its cause and the extraction of the phrases describing the emotion causes. The method employs syntactic and dependency parser and rules for the analysis of eight types of the emotion-cause linguistic relations. The results of evaluation showed that our method performed with high level of accuracy (82%). 1 Introduction and Background Emotional reactions to three salient aspects of the world, namely (1) events and their conse- quences, (2) agents and their actions, and (3) ob- jects, are based on the nature of cognitive origins and can be triggered under specific conditions (Ortony et al., 1988). The cognitive model of emotions (OCC model of emotions) arranges 22 emotions in three substantially independent classes according to the aspects of the world that are in focus of evaluation. Recently, the task of automatic recognition of distinct emotions conveyed in text has been gain- ing increased attention of researchers in the areas of natural language processing and computation- al linguistics (Alm, 2008; Aman and Szpakowicz, 2008; Boucouvalas, 2003; Chaumartin, 2007; Katz et al., 2007; Kozareva et al., 2007; Liu et al., 2003; Neviarouskaya et al., 2011; Purver and Battersby, 2012; Strapparava and Mihalcea, 2008; Suttles and Ide, 2013). To understand emotions expressed in written language, it is im- portant to analyse the causes of emotions ("what caused a particular emotion") and eliciting con- ditions ("under what conditions"). The challenge of emotion cause detection in text has been re- cently tackled by Chen et al. (2010), who devel- oped two sets of linguistic pattern-based features (manually generalized patterns and automatically generalized patterns) for extraction of causes for emotions in Chinese. The linguistic-pattern- based methodology described in (Chen et al., 2010) inspired the development of a method for the identification of Italian sentences that contain emotion cause phrases and the retrieval of emo- tion – emotion cause phrase couples (Russo et al., 2011). In their subsequent work, Caselli et al. (2012) semi-automatically assigned polarity val- ues to Italian nouns that potentially represent nominal cause events associated with emotions. In this work, we introduce a novel method for automatic extraction of emotion causes. The main contributions of our work are as follows: (1) development of a corpus of emotion causes and (2) deep analysis of cause events specific for 22 emotions from the OCC …

Content truncated to 500 words