Full-coverage Identification of English Light Verb Constructions

International Joint Conference on Natural Language Processing, pages 329–337, Nagoya, Japan, 14-18 October 2013. Full-coverage Identification of English Light Verb Constructions Istv´an Nagy T.1, Veronika Vincze1,2 and Rich´ard Farkas1 1Department of Informatics, University of Szeged {nistvan,rfarkas}@inf.u-szeged.hu 2Hungarian Academy of Sciences, Research Group on Artificial Intelligence vinczev@inf.u-szeged.hu Abstract The identification of light verb construc- tions (LVC) is an important task for sev- eral applications. Previous studies focused on some limited set of light verb construc- tions. Here, we address the full coverage of LVCs. We investigate the performance of different candidate extraction methods on two English full-coverage LVC anno- tated corpora, where we found that less se- vere candidate extraction methods should be applied. Then we follow a machine learning approach that makes use of an ex- tended and rich feature set to select LVCs among extracted candidates. 1 Introduction A multiword expression (MWE) is a lexical unit that consists of more than one orthographical word, i.e. a lexical unit that contains spaces and displays lexical, syntactic, semantic, pragmatic and/or statistical idiosyncrasy (Sag et al., 2002; Calzolari et al., 2002). Light verb constructions (LVCs) (e.g. to take a decision, to take sg into consideration) form a subtype of MWEs, namely, they consist of a nominal and a verbal compo- nent where the verb functions as the syntactic head (the whole construction fulfills the role of a verb in the clause), but the semantic head is the noun (i.e. the noun is used in one of its original senses). The verbal component (also called a light verb) usually loses its original sense to some extent.1 The meaning of LVCs can only partially be com- puted on the basis of the meanings of their parts and the way they are related to each other (semi- compositionality). Thus, the result of translating their parts literally can hardly be considered as 1Light verbs may also be defined as semantically empty support verbs, which share their arguments with a noun (see the NomBank project (Meyers et al., 2004)), that is, the term support verb is a hypernym of light verb. the proper translation of the original expression. Moreover, the same syntactic pattern may belong to a LVC (e.g. make a mistake), a literal verb + noun combination (e.g. make a cake) or an idiom (e.g. make a meal (of something)), which suggests that their identification cannot be based on solely syntactic patterns. Since the syntactic and the se- mantic head of the construction are not the same, they require special treatment when parsing. On the other hand, the same construction may func- tion as an LVC in certain contexts while it is just a productive construction in other ones, compare He gave her a ring made of gold (non-LVC) and He gave her a ring because he wanted to hear her voice (LVC). In several natural language processing (NLP) applications like information extraction and re- trieval, terminology extraction and machine trans- lation, it is important to identify LVCs in con- text. For example, in machine translation …

Content truncated to 500 words