A descriptive and formal perspective for grammar development

For many linguistic theories, a grammar is a mechanism making it possible to define a natural language. The idea behind generative theories is more precisely to consider grammars as an enumerative process deriving a language. This idea is still very present, even in non-generative approaches, and a grammar is considered as a device used to check whether an input belongs or not to the language. This conception is very restrictive for many reasons. First, it has a lot of consequences on the way of representing linguistic information which is expressed in order to rule out ungrammatical utterances. This is particularly clear in the Optimality Theory (Prince and Smolensky [1993]) in which constraints (considered as universals) are stipulated precisely in this perspective. Second, considering grammar as a way of defining a language relies on a clear distinction between grammatical and ungrammatical productions. However, we know that such a distinction doesn't fit with the reality of language for many reasons. Corpus linguistics has shown for many years how often inputs can be ill-formed, and this characteristics concerns written as well as spoken productions. A linguist as well as an engineer has to deal with such material either to explain or to treat it. This means that we need an approach taking into account such aspects. Moreover the notion of grammaticality requires a total covering of the language : uncertainty, incompleteness, heterogeneity have no place in such approaches