Expressing Implicit Semantic Relations without Supervision

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 313–320, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Expressing Implicit Semantic Relations without Supervision Peter D. Turney Institute for Information Technology National Research Council Canada M-50 Montreal Road Ottawa, Ontario, Canada, K1A 0R6 peter.turney@nrc-cnrc.gc.ca Abstract We present an unsupervised learning al- gorithm that mines large text corpora for patterns that express implicit semantic re- lations. For a given input word pair X :Y with some unspecified semantic relations, the corresponding output list of patterns Pm P , , 1 � is ranked according to how well each pattern iP expresses the relations between X and Y . For exam- ple, given X=ostrich and Y=bird , the two highest ranking output patterns are “X is the largest Y” and “Y such as the X”. The output patterns are intended to be useful for finding further pairs with the same relations, to support the con- struction of lexicons, ontologies, and se- mantic networks. The patterns are sorted by pertinence, where the pertinence of a pattern iP for a word pair X :Y is the expected relational similarity between the given pair and typical pairs for iP . The algorithm is empirically evaluated on two tasks, solving multiple-choice SAT word analogy questions and classifying seman- tic relations in noun-modifier pairs. On both tasks, the algorithm achieves state- of-the-art results, performing signifi- cantly better than several alternative pat- tern ranking algorithms, based on tf-idf. 1 Introduction In a widely cited paper, Hearst (1992) showed that the lexico-syntactic pattern “Y such as the X” can be used to mine large text corpora for word pairs X :Y in which X is a hyponym (type) of Y. For example, if we search in a large corpus using the pattern “Y such as the X” and we find the string “bird such as the ostrich”, then we can infer that “ostrich” is a hyponym of “bird”. Ber- land and Charniak (1999) demonstrated that the patterns “Y’s X” and “X of the Y” can be used to mine corpora for pairs X :Y in which X is a meronym (part) of Y (e.g., “wheel of the car”). Here we consider the inverse of this problem: Given a word pair X :Y with some unspecified semantic relations, can we mine a large text cor- pus for lexico-syntactic patterns that express the implicit relations between X and Y ? For exam- ple, if we are given the pair ostrich:bird, can we discover the pattern “Y such as the X”? We are particularly interested in discovering high quality patterns that are reliable for mining further word pairs with the same semantic relations. In our experiments, we use a corpus of web pages containing about 5×1010 English words (Terra and Clarke, 2003). From co-occurrences of the pair ostrich:bird in this corpus, we can generate 516 patterns of the form “X ... Y” and 452 patterns of the form “Y ... X”. Most of these …

Content truncated to 500 words