Automatic Determination of a Domain Adaptation Method for Word Sense Disambiguation Using Decision Tree Learning

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1107–1115, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Automatic Determination of a Domain Adaptation Method for Word Sense Disambiguation Using Decision Tree Learning Kanako Komiya Tokyo University of Agriculture and Technology 2-24-16 Naka-cho, Koganei Tokyo, 184-8588 Japan kkomiya@cc.tuat.ac.jp Manabu Okumura Tokyo Institute of Technology 4259 Nagatsuta Modori-ku Yokohama 226-8503 Japan oku@pi.titech.ac.jp Abstract Domain adaptation (DA), which involves adapting a classifier developed from source to target data, has been studied intensively in recent years. However, when DA for word sense disambiguation (WSD) was carried out, the optimal DA method varied according to the properties of the source and target data. This pa- per describes how the optimal method for DA was determined depending on these properties using decision tree learning, given a triple of the target word type of WSD, the source data, and the target data, and discusses what properties affected the determination of the best method when Japanese WSD was performed. 1 Introduction Classifiers in standard supervised machine learn- ing have been trained for data in domain A using manually annotated data in domain A, e.g., to train classifiers for newswires using newswires. How- ever, classifiers for data in domain B have some- times been necessary when there have been no or few manually annotated data, and there have only been manually annotated data in domain A, which have been related to domain B. Domain adapta- tion (DA) involves adapting the classifier that have been trained from data in domain A (source do- main) to data in domain B (target domain). This has been studied intensively in recent years. However, the optimal method of DA varied ac- cording to the properties of the data in the source domain (the source data) and the data in the target domain (the target data) when DA for word sense disambiguation (WSD) was carried out. (We will show it in Section 4.) We define a case as a triple of the target word type of WSD, the source data, and the target data. This paper describes how the optimal method for DA was determined depending on these properties using decision tree learning given a case and dis- cusses what properties affected the determination of the best method when Japanese WSD was per- formed. This paper is organized as follows. Section 2 reviews related works on DA and Section 3 ex- plains how a DA method is automatically deter- mined. Section 4 describes the data we used. How to label the data and how to train the classifiers us- ing these are explained in Section 5. We present the results in Section 6 and discuss them in Sec- tion 7. Finally, we conclude the paper in Section 8. 2 Related Work The DA problem can be categorized into three types depending on the information for learning, i.e., supervised, semi-supervised, and unsuper- vised approaches. A classifier in a supervised ap- proach is developed from a large amount of la- beled …

Content truncated to 500 words