In this paper, we examine the "Dictionnaire Explicatif et Combinatoire" developed by I. Mel’cuk's team in the perspective of NLP. We deal here with collocations (e.g. 'heavy smoker', 'to take a walk') which are considered as restricted cooccurrences. We first present a few characteristics of these lexical associations and the way they are encoded in the DEC by means of syntagmatic lexical functions. We then examine syntactic properties hypothesizing that these properties are not fully idiosyncratic but that a large number of generalizations can be performed within a collocative entry. Lastly, focusing on a subset of emotion nouns, we investigate the feasibility of this proposal by proposing an encoding accounting for the generalizations.

Natural Language Processing systems (e.g. machine translation, information retrieval or generation systems) require large lexical resources which prove extremely costly to build. The automatic or semi-automatic construction of electronic dictionaries is now recognized as a sine qua non in any NLP project and a lot of attention is being paid to methods and tools for acquiring, coding and formalizing the linguistic properties of lexical items appearing in the texts to be processed. This paper deals with the formalization of lexical-semantic relations in computerized dictionaries, focusing more particularly on a few research projects funded by the European Union.