Chinese Verb Sense Discrimination Using an EM Clustering Model with Rich Linguistic Features

Chinese Verb Sense Discrimination Using an EM Clustering Model with Rich Linguistic Features Jinying Chen, Martha Palmer Department of Computer and Information Science University of Pennsylvania Philadelphia, PA, 19104 {jinying,mpalmer}@linc.cis.upenn.edu Abstract This paper discusses the application of the Expectation-Maximization (EM) clustering algorithm to the task of Chinese verb sense discrimination. The model utilized rich linguistic features that capture predicate- argument structure information of the target verbs. A semantic taxonomy for Chinese nouns, which was built semi-automatically based on two electronic Chinese semantic dictionaries, was used to provide semantic features for the model. Purity and normalized mutual information were used to evaluate the clustering performance on 12 Chinese verbs. The experimental results show that the EM clustering model can learn sense or sense group distinctions for most of the verbs successfully. We further enhanced the model with certain fine-grained semantic categories called lexical sets. Our results indicate that these lexical sets improve the model’s performance for the three most challenging verbs chosen from the first set of experiments. 1 Introduction Highly ambiguous words may lead to irrelevant document retrieval and inaccurate lexical choice in machine translation (Palmer et al., 2000), which suggests that word sense disambiguation (WSD) is beneficial and sometimes even necessary in such NLP tasks. This paper addresses WSD in Chinese through developing an Expectation-Maximization (EM) clustering model to learn Chinese verb sense distinctions. The major goal is to do sense discrimination rather than sense labeling, similar to (Schütze, 1998). The basic idea is to divide instances of a word into several clusters that have no sense labels. The instances in the same cluster are regarded as having the same meaning. Word sense discrimination can be applied to document retrieval and similar tasks in information access, and to facilitating the building of large annotated corpora. In addition, since the clustering model can be trained on large unannotated corpora and evaluated on a relatively small sense-tagged corpus, it can be used to find indicative features for sense distinctions through exploring huge amount of available unannotated text data. The EM clustering algorithm (Hofmann and Puzicha, 1998) used here is an unsupervised machine learning algorithm that has been applied in many NLP tasks, such as inducing a semantically labeled lexicon and determining lexical choice in machine translation (Rooth et al., 1998), automatic acquisition of verb semantic classes (Schulte im Walde, 2000) and automatic semantic labeling (Gildea and Jurafsky, 2002). In our task, we equipped the EM clustering model with rich linguistic features that capture the predicate-argument structure information of verbs and restricted the feature set for each verb using knowledge from dictionaries. We also semi- automatically built a semantic taxonomy for Chinese nouns based on two Chinese electronic semantic dictionaries, the Hownet dictionary1 and the Rocling dictionary.2 The 7 top-level categories of this taxonomy were used as semantic features for the model. Since external knowledge is used to obtain the semantic features and guide feature selection, the model is not completely unsupervised from this perspective; however, it does not make use of …

Content truncated to 500 words