A linguistic analysis of emotion and cause

Abstract

Previous research on emotion analysis often neglects the interactions between emotions and events, regardless of the integral role event plays in emotion studies. In this paper, emotion construction is deemed to be composed of three sequentially structured subevents, namely cause event (i.e. emotion triggering event), emotion state, and elicited event (i.e. event that is induced by the emotion). We blend the insights from emotion studies in the fields of linguistics, computer science, psychology, etc. and investigate the interactions between emotions and cause events on that basis. In general, 71% of the instances in the emotion-sentence corpus express emotions, with sentences expressing happiness accounting for the most, and sadness the least. For each emotion type, approximately 81% of the instances, on average, contain a cause event. It indicates that an emotion frequently co-occurred with its cause event in text. Moreover, it is suggested that cause events tend to be expressed by means of verbal events, and they often occur at the position to the left of the emotion keyword. Therefore, an in-depth analysis of cause event will have great implications for emotion processing in text. Following Chang et al. 's (2000) dichotomical classification of emotion verbs, emotion verbs are classified into change-of-state emotion verbs and homogeneous state emotion verbs. We investigate the interrelations between emotions and cause events in terms of transitivity and epistemicity. As for the degree of transitivity of cause events, we examine the affectedness of experiencers of the emotions by focusing on three parameters, namely agentivity, kinesis, and participation. The analysis establishes the relationship between emotion types and cause events. The change-of-state and homogeneous emotion verbs share certain cause event features, except for the fear emotion. The cause events of happiness have the highest degree of transitivity, the cause events of sadness have the lowest. The cause events of anger and surprise share similar features. Unlike other emotions, the cause event features of fear are different for change-of-state and homogenous emotion verbs in terms of the kinesis of events, i.e. motion. As for epistemicity, we introduce epistemic markers which are verbs that indicate the experiencer's cognitive awareness of the emotion causes. While cause events of homogeneous emotion verbs are rarely found to be introduced by epistemic markers, cause events of change-of-state emotion verbs are often preceded by epistemic markers. It is observed that the cause events of a positive emotion such as happiness tend to be more frequently marked by epistemic markers than that of a negative emotion, such as sadness or anger. It can be explained by the motivation of the experiencer. That is, the higher motivation the experiencer has to express the certainty of the cause events, the more explicit epistemic marking of cause events is allowed. This paper considers an emotion as a pivot event that links the cause events and the elicited events, and presents the interactions between emotions and cause events using a corpus-based approach. The annotated corpus will provide useful and valuable resources for linguistic analysis as well as natural language processing. Besides, the linguistic account of the emotion-cause interactions will not only contribute to the construction of linguistic model of emotions, but also lay the groundwork for the computational implementations for automatic emotion detection and classification. 本文以事件结构为出发点研究情绪，解释文本中情绪与其肇因事件之间的互动。文章指出，在涉及情绪文本分析的研究中，情绪和事件之间的关系通常被研究者所忽略，而事件往往是情绪触发的最重要因素之一。本文基于事件的情绪分析方法，通过构建情绪肇因事件语料库的方式探索不同情绪与事件之间的关系。研究分析表明情绪类别与肇因事件特征存在着紧密的联系，而认知标记也可以显性地表达状态变化情绪动词的肇因事件。