Experiments in Syntactic and Semantic Classification and Disambiguation using Bootstrapping

EXPERIMENTS IN SYNTACTIC AND SEMANTIC CLASSIFICATION AND DISAMBIGUATION USING BOOTSTRAPPING* Robert P. Futrelle and Susan Gauch Biological Knowledge Laboratory College of Computer Science Northeastern University Boston, MA 02115 {futrelle, sgauch}@ccs.neu.edu ABSTRACT Bootstrap methods (unsupervised classification) that generate word classes without requiring pretagging have had notable success in the last few years. The methods described here strengthen these approaches and produce excellent word classes from a 200,000 word corpus. The method uses mutual information measures plus positional information from the words in the immediate context of a target word to compute similarities. Using the similarities, classes are built using hierarchical agglomerative clustering. At the leaves of the classification tree, words are grouped by syntactic and semantic similarity. Further up the tree, the classes are primarily syntactic. Once the initial classes are found, they can be used to classify ambiguous words, i.e., part-of-speech tagging. This is done by expanding each context word of a target instance into a tightly defined class of similar words, a simsct. The use of simsets is shown to increase the tagging accuracy from 83% to 92% for the forms "cloned" and "deduced". INTRODUCTION The identification of the syntactic class and the discovery of semantic information for words not contained in any on-line dictionary or thesaurus is an important and challenging * This material is based upon work supported by the National Science Foundation under Grant No. DIR-8814522. problem. Excellent methods have been developed for part-of-speech (POS) tagging using stochastic models trained on partially tagged corpora (Church, 1988; Cutting, Kupiec, Pedersen & Sibun, 1992). Semantic issues have been addressed, particularly for sense disambiguation, by using large contexts, e.g., 50 nearby words (Gale, Church & Yarowsky, 1992) or by reference to on-line dictionaries (Krovetz, 1991; Lesk, 1986; Liddy & Paik, 1992; Zernik, 1991). More recently, methods to work with entirely untagged corpora have been developed which show great promise (Brill & Marcus, 1992; Finch & Chater, 1992; Myaeng & Li, 1992; Schutze, 1992). They are particularly useful for text with specialized vocabularies and word use. These methods of unsupervised classification typically have clustering algorithms at their heart (Jain & Dubes, 1988). They use similarity of contexts (the distribution principle) as a measure of distance in the space of words and then cluster similar words into classes. This paper demonstrates a particular approach to these classification techniques. In our approach, we take into account both the relative positions of the nearby context words as well as the mutual information (Church & Hanks, 1990) associated with the occurrence of a particular context word. The similarities computed from these measures of the context contain information about both syntactic and semantic relations. For example, high similarity values are obtained for the two semantically similar nouns, 117 "diameter" and "length", as well as for the two adjectives "nonmotile" and "nonchemotactic". We demonstrate the technique on three problems, all using a 200,000 word corpus composed of 1700 abstracts from a specialized field of biology: #1: Generating the full classification tree for the 1,000 most frequent …

Content truncated to 500 words