Corpus-Based Induction of Lexical Representation and Meaning

The acquisition of linguistic knowledge, i.e., the identication, extraction, and encoding of linguistic information in a corpus, has been one of the main motivations for data-driven approaches to natural language. Methods have been developed for the acquisition of, for instance, parts of speech, noun compounds, collocations, support verbs, subcategorization frames, phrase structure rules, selectional restrictions and sense induction (cf. Armstrong (1993) for an overview). Drawing on this body of research, I am investigating the acquisition of lexical semantic knowledge from corpora, thereby addressing the logical problem of language acquisition, one of the fundamental issues in linguistics and cognitive science. My guiding assumption is that syntactic as well as semantic representations are projected from information in the lexicon, and that a crucial part of the relevant lexical information is the result of language experience, and hence can be induced from corpora. The proposed research includes three main subtasks: (a) induction of di erent types of (\low-level") lexical semantic information (i.e., subcategorization frames, selectional restrictions, semantic classes), using established corpus-based methods; (b) combination of the induced types of lexical semantic information into (\highlevel") semantic representations, based on existing theories of the lexicon; (c) evaluation of the resulting model against human intuitions. By applying corpus-based techniques to lexical semantics, i.e., a classical representational problem in linguistics, I hope to contribute to bridging the gap between current data-driven approaches to language and the knowledge-driven methods of traditional linguistics.