Emotion Estimation from Sentence Using Relation between Japanese Slangs and Emotion Expressions

Copyright 2012 by Kazuyuki Matsumoto, Kenji Kita, and Fuji Ren 26th Pacific Asia Conference on Language,Information and Computation pages 343–350 Emotion Estimation from Sentence Using Relation between Japanese Slangs and Emotion Expressions Kazuyuki Matsumoto The University of Tokushima Minami-josanjima, Tokushima, 770-8506, Japan {matumoto;kita;ren}@is.tokushima-u.ac.jp Kenji Kita Fuji Ren Abstract Most of Japanese slang words such as Waka- mono Kotoba are analyzed as “unknown word” or segmented wrongly by the morpho- logical analysis system. These problems are causing negative effect on sentiment analysis in text. These words generally have many va- rieties of notations and conjugations, and they lack versatility. As a result, many of them are not registered in the dictionaries, making mor- phological analysis more difficult. In this pa- per, we aimed to decrease such negative ef- fects of Wakamono Kotoba for the accuracy of emotion estimation from sentence and pro- posed a method to increase the accuracy by us- ing a classification method based on machine learning. In this method we used emotional expressions which had high relevance with Wakamono Kotoba as feature. As a result, the proposed method obtained 20% higher accu- racy than the method only using morpheme N- gram as feature. Emotion Corpus, Japanese Slang, Out of Vo- cabulary 1 Introduction The words that are not registered in the dictionaries are called unknown words. In the field of Natural Language Processing unknown words have been tra- ditionally studied. However, many of these studies focused on proper noun, onomatopoeia or emoticon, while a few research targeted slang such as Waka- mono Kotoba. One of the reasons might be that Wakamono Kotoba has been usually treated as “im- proper expression” or “bad word”(Noguchi , 2004). However, considering that these words are getting more and more frequently used on WWW, it is in- evitable to treat Wakamono Kotoba even though they are improper expressions. In Japan, many of the Internet users are people from teens to people in their fourties1. One of the characteristics of Wakamono Kotoba is that they are specialized in expressing how people especially in their younger age feel. By dealing with such Waka- mono Kotoba, we will be able to use effectively the huge amount of documents on WWW as precious resources for language processing. This paper aims to estimate emotion from utter- ances including Wakamono Kotoba. Most of the existing emotion estimation studies from text did not treat the problem of slang such as Wakamono Kotoba. One of the reasons was that there were few text corpora including Wakamono Kotoba. Cur- rently Weblog became very popular and many doc- uments on WWW are written in spoken language. These texts are available as huge corpus. As the re- sult, recently there are active research on new words or unknown words (Murawaki , 2010),(Jiean et al., 2011) and there are also research on emotion esti- mation based on their findings (Matsumoto et al., 2011),(Matsumoto and Ren, 2011). For example, in (Matsumoto et al., 2011), they used the conventional statistic method to estimate emotion of the sentence …

Content truncated to 500 words