There’s no Accounting for Taste – Except in Dictionaries …

This paper discusses essential aspects of a new methodology for a corpus based lexicographic approach as developed within the project on Descriptive Lexical Specifications (DELIS, LRE 61.034). The main topics treated here are (1) description and analysis of electronic corpus data in terms of semantic frames and (2) interpretation of the analysis results towards formal lexical representations to be used in computational lexicography. The frame semantics approach provides the theoretical basis for lexical specification of corpus data. The descriptive model (exemplified by perception verbs) interrelates semantic, syntactic and morphological levels. 0. Introduction DELIS is a broadly formulated project covering 7 work packages with different working tasks. Research and development are carried out in co-operation between a number of institutes and language groups including Danish. The main objective of the project is innovative in itself: to integrate corpus exploration, lexical modeling and tool development. Key tasks are • lexical description on the basis of observable facts in corpus material • contrastive lexicology across languages within commonly selected semantic domains and on the basis of a common theoretical hypothesis • use of a common descriptive model for lexical specification • tool development for corpus search, formal modeling and dictionary encoding. For each task the key notion is reusability. For a more detailed description, see Heid (1994) elsewhere in these proceedings. This contribution mainly deals with items related to the three first mentioned tasks. For Danish there does not yet exist any dictionary or comprehensive research report containing systematic semantic descriptions for verb classes based on detailed analysis of corpus data. Electronic corpora are widely used in the ongoing work on the dictionary of contemporary Danish (Den Danske Ordbog) similar to corpus based dictionary work for English (e.g. in COBUILD).