NUS-PT: Exploiting Parallel Texts for Word Sense Disambiguation in the English All-Words Tasks

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 253–256, Prague, June 2007. c⃝2007 Association for Computational Linguistics NUS-PT: Exploiting Parallel Texts for Word Sense Disambiguation in the English All-Words Tasks Yee Seng Chan and Hwee Tou Ng and Zhi Zhong Department of Computer Science, National University of Singapore 3 Science Drive 2, Singapore 117543 {chanys, nght, zhongzhi}@comp.nus.edu.sg Abstract We participated in the SemEval-2007 coarse-grained English all-words task and fine-grained English all-words task. We used a supervised learning approach with SVM as the learning algorithm. The knowledge sources used include local col- locations, parts-of-speech, and surrounding words. We gathered training examples from English-Chinese parallel corpora, SEMCOR, and DSO corpus. While the fine-grained sense inventory of WordNet was used to train our system employed for the fine-grained English all-words task, our system employed for the coarse-grained English all-words task was trained with the coarse-grained sense inventory released by the task organizers. Our scores (for both recall and precision) are 0.825 and 0.587 for the coarse-grained English all-words task and fine-grained English all-words task respectively. These scores put our systems in the first place for the coarse-grained English all-words task1 and the second place for the fine-grained English all-words task. 1 Introduction In this paper, we describe the systems we devel- oped for the coarse-grained English all-words task 1A system developed by one of the task organizers of the coarse-grained English all-words task gave the highest over- all score for the coarse-grained English all-words task, but this score is not considered part of the offi cial scores. and fine-grained English all-words task of SemEval- 2007. In the coarse-grained English all-words task, systems have to perform word sense disambiguation (WSD) of all content words (noun, adjective, verb, and adverb) occurring in five documents, using a coarse-grained version of the WordNet sense inven- tory. In the fine-grained English all-words task, sys- tems have to predict the correct sense of verbs and head nouns of the verb arguments occurring in three documents, according to the fine-grained sense in- ventory of WordNet. Results from previous SENSEVAL English all- words task have shown that supervised learning gives the best performance. Further, the best per- forming system in SENSEVAL-3 English all-words task (Decadt et al., 2004) used training data gathered from multiple sources, highlighting the importance of having a large amount of training data. Hence, besides gathering examples from the widely used SEMCOR corpus, we also gathered training exam- ples from 6 English-Chinese parallel corpora and the DSO corpus (Ng and Lee, 1996). We developed 2 separate systems; one for each task. For both systems, we performed supervised word sense disambiguation based on the approach of (Lee and Ng, 2002) and using Support Vector Machines (SVM) as our learning algorithm. The knowledge sources used include local collocations, parts-of-speech (POS), and surrounding words. Our system employed for the coarse-grained English all- words task was trained with the coarse-grained sense inventory released by the task organizers, while our system employed for the fine-grained English all- words task was …

Content truncated to 500 words