The Relationship between Semantic Similarity and Subcategorization Frames in English: A Stochastic Test Using ICE-GB and WordNet

The Relationship between Semantic Similarity and Subcategorization Frames in English: A Stochastic Test Using ICE-GB and WordNet* Sanghoun Song, Jae-Woong Choe Dept. of Linguistics, Korea University, Anam-dong, Sungbuk-gu, Seoul, KOREA {yooseon21, jchoe}@korea.ac.kr Abstract. In this paper, we test a working hypothesis that there is a significant relationship between semantic similarity and subcategorization frames in English. This paper is under the assumption that if a group of verbs form a cluster sharing a similar meaning, they tend to share subcategorization frames. In the process, we propose a statistical method to test this assumption, making use of two language resources, namely, ICE-GB and WordNet. We come to the conclusion that the proposed hypothesis holds true to the degrees of 42~73%, based on our stochastic analysis. Keywords: semantic similarity, subcategorization frames, ICE-GB, WordNet, statistical method, clustering, dendrogram, selectional preference strength 1. Introduction The following examples taken from Lasnik and Uriagereka (2005) bring out an issue concerning the relationship between semantic similarity of verbs and the corresponding subcategorization frames. (1) a. She asked what time it was. b. She asked the time. c. She wondered what time it was. b. *She wondered the time. Sentences in (1) show that verbs like ask and wonder, which are similar in meaning, show different ways of argument realization. Although it is true that arbitrariness is at the core of the form-meaning mapping in any language, there are other aspects of language where formal or grammatical realization seems to be largely determined by meaning. Levin (1993) provides many * Acknowledgments: We owe special thanks to Jieun Jeon♡ for her assistance and partial participation in the preparation of this paper, especially for extracting relevant data from ICE-GB. We also would like to express our thanks to Prof. Seok-Hoon You for his support for our project. Finally, comments from three anonymous reviewers for PACLIC 22 are much appreciated. Copyright 2008 by Sanghoun Song, Jae-Woong Choe 358 22nd Pacific Asia Conference on Language, Information and Computation, pages 358–367 examples of semantic classes of English verbs that lead to the same or similar pattern of argument realizations. For example, the cut verbs (e.g. chip, cut, scrape, snip, etc.), a semantically identifiable group of words in the sense that they all involve notions of motion, contact, and effect, do not allow causative alternations. That is, their subcategorization frames are affected by their meaning. (2) a. Carol cut the bread. b. *The bread cut. The question we raise in this paper is if the meaning restriction applies in a significant way to the realization of argument structure, if it does, to what extent it holds. In other words, how much is a syntactic pattern tied up with the semantic properties of the English verbs? In order to test this research question in a rather comprehensive way, we make use of large sets of language resources, in particular, ICE-GB and WordNet. In the process, we propose a more articulated methodology to take advantage of the language resources statistically and computationally. Also introduced in this paper …

Content truncated to 500 words