A Bilingual Attention Network for Code-switched Emotion Prediction

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1624–1634, Osaka, Japan, December 11-17 2016. A Bilingual Attention Network for Code-switched Emotion Prediction Zhongqing Wang†, Yue Zhang†, Sophia Yat Mei Lee‡, Shoushan Li§ and Guodong Zhou§ † Singapore University of Technology and Design, Singapore ‡Department of Chinese and Bilingual Studies, The Hong Kong Polytechnic University § Natural Language Processing Lab, Soochow University, China wangzq.antony@gmail.com, yue zhang@sutd.edu.sg ym.lee@polyu.edu.hk, {lishoushan, gdzhou}@suda.edu.cn Abstract Emotions in code-switching text can be expressed in either monolingual or bilingual forms. How- ever, relatively little research has placed emphasis on code-switching text. The challenges of this task include the exploration both monolingual and bilingual information of each post and cap- turing the informative words from the code-switching context. To address these challenges, we propose a Bilingual Attention Network (BAN) model to aggregate the monolingual and bilingual informative words to form vectors from the document representation, and integrate the attention vectors to predict the emotion. The experiments show the effectiveness of the proposed model. Visualization of the attention layers illustrates that the model selects informative words qualita- tively. 1 Introduction Microblogs such as Twitter and Facebook have gained tremendous popularity in the past decade, they often contain extremely current, even breaking, information about world events. However, the writing style of microblogs tends to be quite colloquial and nonstandard, unlike the style found in more tra- ditional, edited genres (Li et al., 2015; Vo and Zhang, 2015). In addition, authors from multi-lingual communities tend to write code-switching posts frequently (Ling et al., 2013; Wang et al., 2015). These pose challenges for automatic emotion prediction tasks. There has been some previous research focusing on both emotion analysis (Pang et al., 2002; Lee et al., 2014) and code-switching text analysis (Solorio and Liu, 2008; Ling et al., 2013; Jamatia et al., 2015). However, little research has focused on predicting emotion in code-switching text. Different from monolingual emotion prediction, the emotion in code-switching posts can be expressed in either monolingual or bilingual forms. In this study, we focus on Chinese and English mixed code-switching text from Chinese social media. Although Chinese is the major language, it has been shown that English words are critical for emotion expression (20.1% posts express emotion through English text). E1 - E4 show four examples of code-switching posts that contain both Chinese and English words. E 1. ��˜eÌÓwso happy œ (I went rollerblading the whole afternoon, so happy!) E 2. mƱ5§2C�œX"ØS-�%�"(¢TgCu?˜e�"""sigh ˜ (I have been grumpy and emotional since the first day of school, unstable mindset too. It’s really time to self-evaluate ... sigh. ) E 3. þ�˜U�‘§ÓfholdØ Ø Ø4 4 4�B" (I have been teaching the whole day, my throat can’t take it anymore.) E 4. @å†�Ö˧Õõþ•£�§£�˜�úLÕ§õohappy�˜U" (I drank too much in the morning. I got drunk and went back to school by bus, and I missed my stop. Such a happy day.) This work is licenced under a Creative Commons Attribution 4.0 International License. License details: http: //creativecommons.org/licenses/by/4.0/ 1624 开学以来，浮躁的情绪。不安稳的心态。确实该自己检讨一下了。。。sigh ~ …

Content truncated to 500 words