Joining forces to resolve lexical ambiguity: East meets West in Barcelona

Joining forces to resolve lexical ambiguity: East meets West in Barcelona Richard WICENTOWSKI*, Grace NGAI†1, Dekai WU‡2 Marine CARPUAT‡, Emily THOMFORDE*, Adrian PACKEL* *Swarthmore College Swarthmore, PA USA † Dept. of Computing † HK Polytechnic University Hong Kong ‡ HKUST, Dept of Computer Science Human Language Technology Center Hong Kong richardw@cs.swarthmore.edu, csgngai@polyu.edu.hk, dekai@cs.ust.hk marine@cs.ust.hk, ethomfo1@cs.swarthmore.edu, packel@cs.swarthmore.edu Abstract This paper describes the component models and combination model built as a joint effort be- tween Swarthmore College, Hong Kong PolyU, and HKUST. Though other models described elsewhere contributed to the final combination model, this pa- per focuses solely on the joint contributions to the ”Swat-HK” effort. 1 Introduction This paper describes the two joint component mod- els of the Swat-HK systems entered into four of the word sense disambiguation lexical sample tasks in Senseval-3: Basque, Catalan, Italian and Roma- nian, as well as a combination model for each lan- guage. The feature engineering (and construction of three other component models which are described in (Wicentowski et al., 2004)) was performed at Swarthmore College, while the Hong Kong team constructed two component models based on well- known machine learning algorithms. The combina- tion model, which was constructed at Swarthmore, uses voting to combine all five models. 2 Experimental Features A full description of the experimental features for all four tasks can be found in the report submitted by the Swarthmore College Senseval team (Wicen- towski et al., 2004). Briefly, the systems used lexi- cal and syntactic features in the context of the target word: • The “bag of words (and lemmas)” in the con- text of the ambiguous word. • Bigrams and trigrams of words (and lemmas, 1The author would like to thank the Hong Kong Polytechnic University for supporting this research in part through research grants A-PE37 and 4-Z03S. 2The author would like to thank the Hong Kong Research Grants Council (RGC) for supporting this research in part through research grants RGC6083/99E, RGC6256/00E, and DAG03/04.EG09. part-of-speech tags, and, for Basque, case in- formation) surrounding the ambiguous word. • The topic (or code) of the document containing the current instance of the word was extracted. (Basque and Catalan only.) These features have been shown to be effective in previous WSD research. Since our systems were all supervised, all the data used was provided by the Senseval organizers; no additional (unlabeled) data was included. 3 Methodology The systems that were constructed by this team in- cluded two component models: a boosting model and a maximum entropy model as well as a com- bination system. The component models were also used in other Senseval-3 tasks: Semantic Role La- beling (Ngai et al., 2004) and the lexical sample tasks for Chinese and English, as well as the Multi- lingual task (Carpuat et al., 2004). To perform parameter tuning for the two compo- nent models, 20% of the samples from the training set were held out into a validation set. Since we did not expect the senses of different words to share any information, the training data …

Content truncated to 500 words