A Text-driven Rule-based System for Emotion Cause Detection

Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 45–53, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics A Text-driven Rule-based System for Emotion Cause Detection Sophia Yat Mei Lee† Ying Chen†* Chu-Ren Huang†‡ †Department of Chinese and Bilingual Studies The Hong Kong Polytechnic University *Department of Computer Engineering China Agriculture University ‡Institute of Linguistics Academia Sinica {sophiaym, chenying3176, churenhuang}@gmail.com Abstract Emotion cause detection is a new research area in emotion processing even though most theo- ries of emotion treat recognition of a triggering cause event as an integral part of emotion. As a first step towards fully automatic inference of cause-emotion correlation, we propose a text- driven, rule-based approach to emotion cause detection in this paper. First of all, a Chinese emotion cause annotated corpus is constructed based on our proposed annotation scheme. By analyzing the corpus data, we identify seven groups of linguistic cues and generalize two sets of linguistic rules for detection of emotion causes. With the linguistic rules, we then de- velop a rule-based system for emotion cause detection. In addition, we propose an evaluation scheme with two phases for performance as- sessment. Experiments show that our system achieves a promising performance for cause oc- currence detection as well as cause event detec- tion. The current study should lay the ground for future research on the inferences of implicit information and the discovery of new informa- tion based on cause-event relation. 1 Introduction Text-based emotion processing has attracted plenty of attention in NLP. Most research has focused on the emotion detection and classification by identifying the emotion types, for instances happiness and sadness, for a given sentence or document (Alm 2005, Mihalcea and Liu 2006, Tokuhisa et al. 2008). However, on top of this surface level information, deeper level information regarding emotions, such as the experiencer, cause, and result of an emotion, needs to be extracted and analyzed for real world applications (Alm 2009). In this paper, we aim at mining one of the crucial deep level types of information, i.e. emotion cause, which provides useful information for applications ranging from economic forecasting, public opinion mining, to product design. Emotion cause detection is a new research area in emotion processing. In emotion processing, the cause event and emotion correlation is a fertile ground for extraction and entailment of new information. As a first step towards fully automatic inference of cause- emotion correlation, we propose a text-driven, rule-based approach to emotion cause detection in this paper. Most theories of emotion treat recognition of a triggering cause event as an integral part of emotional experience (Descartes 1649, James 1884, Plutchik 1962, Wierzbicka 1999). In this study, cause events refer to the explicitly expressed arguments or events that evoke the presence of the corresponding emotions. They are usually expressed by means of propositions, nominalizations, and nominals. For example, “they like it” is the cause event of the emotion happiness in the sentence “I was very happy that …

Content truncated to 500 words