[ENG] Unlike most of traditional approaches to grammar, today's formal and computational models make use of the lexicon as an essential part of the whole system. Differences between current theoretical and computational models are more and more insignificant albeit the latter are perhaps still characterized by yielding full priority to minimalist solutions over more abstract considerations such as exhaustivity or formal elegance. The paper introduces some strategies in linguistic computation for natural language processing (NLP) according to a typology based on functional complexity and describing in every case the scope and the role performed by the dictionary: pattern matching, semàntic grammars, syntactic parsers, augmented transition networks, unification formalisms, case frame grammars, etc. It ends with an exploration into lexicographically oriented procedures of conceptual dependency, which are found beyond the NLP, deeply within the domain of artificial intelligence.