Inferring grammar from lexis: machine-readable dictionaries as sources of wholesale syntactic and semantic information

The main idea behind research into machine-readable dictionaries (MRDs) is that the re-use of such resources enables the NLP community to build up large-scale lexicons with a realistic level of coverage instead of the customary purpose-built 'toy' lexicons containing just a few sample items. Most of the earlier work in this area was concerned with the systematic extraction of easily identifiable types of information: pronunciation etc. More recently the focus has shifted to less tractable aspects: subcategorization, semantic typing, taxonomic-thesauric relations etc. The research performed by the author's group at Amsterdam University reflects this. The ASCOT project was restricted to the extraction of syntactic information from the MRD version of the Longman Dictionary of Contemporary English (LDOCE-P. Procter, 1978). In the LINKS project attention was turned to the semantic interconnections in the same dictionary, and in ACQUILEX this line of approach has been extended, on the basis of both LDOCE and two Van Dale dictionaries (P.J.G. van Sterkenburg & W.J.J. Pijnenburg, 1984; and W. Martin & G.A.J. Tops, 1986) to the construction of semantic taxonomies and lexical knowledge bases.<
>