Semantic pattern learning through maximum entropy-based WSD technique

Semantic Pattern Learning Through Maximum Entropy-based WSD technique Maximiliano Saiz-Noeda Depto. de Lenguajes y Sistemas Inform´aticos Universidad de Alicante Alicante, Spain max@dlsi.ua.es Armando Su´arez Depto. de Lenguajes y Sistemas Inform´aticos Universidad de Alicante Alicante, Spain armando@dlsi.ua.es Manuel Palomar Depto. de Lenguajes y Sistemas Inform´aticos Universidad de Alicante Alicante, Spain mpalomar@dlsi.ua.es Abstract This paper describes a Natural Lan- guage Learning method that extracts knowledge in the form of semantic pat- terns with ontology elements associated to syntactic components in the text. The method combines the use of EuroWord- Net’s ontological concepts and the cor- rect sense of each word assigned by a Word Sense Disambiguation(WSD) module to extract three sets of pat- terns: subject-verb, verb-direct object and verb-indirect object. These sets de- fine the semantic behaviour of the main textual elements based on their syntac- tic role. On the one hand, it is shown that Maximum Entropy models applied to WSD tasks provide good results. The evaluation of the WSD module has re- vealed a accuracy rate of 64% in a pre- liminary test. On the other hand, we ex- plain how an adequate set of semantic or ontological patterns can improve the success rate of NLP tasks such us pro- noun resolution. We have implemented both modules in C++ and although the evaluation has been performed for En- glish, their general features allow the treatment of other languages like Span- ish. This paper has been partially supported by the Spanish Government (CICYT) project number TIC2000-0664-C02- 02. 1 Introduction Semantic patterns, as defined in this method, con- figure a system to add a new information source to Natural Language Processing (NLP) tasks. To obtain these semantic patterns, it is necessary to count on different tools. On the one hand, a full parser must make a syntactic analysis of the text. This parsing will allow the selection of the differ- ent syntactic functional elements such as subject, direct object (DObj) and indirect object (IObj). On the other hand, a WSD tool must provide the correct sense in order to ensure the appropriate selection of the ontological concept associated to each word. Finally, with the parsing and the cor- rect sense of each word, the pattern extraction method will form and store ontological pairs that define the semantic behaviour of each sentence. 2 Full parsing The analyzer used for this work is the Conexor’s FDG Parser (Pasi Tapanainen and Timo J¨arvinen, 1997). This parser tries to provide a build depen- dency tree from the sentence. When this is not possible, the parser tries to build partial trees that often result from unresolved ambiguity. One vi- sual example of this dependency trees is shown in Figure 1 where the parsing tree of sentence (1) is illustrated. (1) The minister gave explanations to the Government. As seen in Figure 2, the analyzer assigns to each word a text token (second column), a base form (third column) and functional link 10 The the det:>2 @DN> DET SG/PL 2 minister minister subj:>3 @SUBJ N NOM SG 3 …

Content truncated to 500 words