Decision Lists for English and Basque

Decision Lists for English and Basque David Martinez IXA NLP Group University of the Basque Country 649 pk. 20.080 Donostia. Spain. jibmaird @si.ehu.es Abstract In this paper we describe the systems we developed for the English (lexical and all- words) and Basque tasks. They were all supervised systems based on Yarowsky's Decision Lists. We used Semcor for training in the English all-words task. We defined different feature sets for each language. For Basque, in order to extract all the information from the text, we defined features that have not been used before in the literature, using a morphological analyzer. We also implemented systems that selected automatically good features and were able to obtain a prefixed precision (85%) at the cost of coverage. The systems that used all the features were identified as BCU-ehu-dlist-all and the systems that selected some features as BCU-ehu-dlist- best. 1 Introduction Our group took part in three tasks in Senseval- 200 1: all-words and lexical sample for English, and lexical sample for Basque. We applied the same algorithm in all tasks, but using different feature sets. The method we used was based on Yarowsky's Decision Lists (Yarowsky, 1994). We have to mention that different motivations were pursued when working for English or for Basque. In the last years, our work for English has focused on studying the contribution of different kinds of features to WSD (Agirre and Martinez, 2000; Agirre and Martinez, 200la) and on analyzing different knowledge types on a common setting (Agirre and Martinez, 200lb). The English tasks gave us the chance to compare the performance of our method with state-of-the-art systems. Unfortunately, due to time constrains, we could not train the system with syntactic and semantic features, as was our Eneko Agirre IXA NLP Group University of the Basque Country 649 pk. 20.080 Donostia. Spain. eneko@si.ehu.es goal. The systems we used for the English tasks were trained with topical and local features already mentioned in the literature (Yarowsky, 1994). In the English lexical sample, we presented two systems: one trained with all the features (BCU-ehu-dlist-all) and another that selected automatically good features and was able to obtain a prefixed precision (85%) at the cost of coverage (BCU-ehu-dlist-best). In the all-words exercise, we presented only one system, which used all the features and was trained using Semcor (previously we mapped automatically WN 1.6 senses with the WN 1.7 Pre-release). The features and the systems will be described in Section 3. The Basque task presented interesting challenges for us. Previous work in WSD was performed using MRDs, but this was our first approach to the problem using Decision Lists. The Basque language has some particularities that make the selection of features a difficult task. First of all, Basque is an agglutinative language, and some syntactic information is given by inflectional suffixes. Therefore, it is necessary a powerful morphologic analysis of the text in order to identify the lemma and the different parts of each word. Also, phenomena like noun ellipsis have to be taken into …

Content truncated to 500 words