Natural Language Processing requires flexibility. This statement has considerably influenced Natural Language systems during the last years. Recent studies (3), (5), (15) have suggested that nonmonotonic logic would be an attractive frame. Associated with a bottom-up parse of sentences, it could help to solve ambiguity, to reduce parsing choices, and find corrections when the parse fails. This paper presents a methodology for describing Natural Language grammars in a nonmonotonic first-order logic theory. The general idea is to specify first the most general rules defining the grammatical symbols, then all the exceptional cases. We will expose how this method may help us to solve difficulties, as competence errors. For that purpose, we will use TMS default logic (2) in our examples. Finally, we will argue that this methodology shifts the problems to the definition of the grammar nonmonotonicity allows us to express additional informations that are needed in several cases, even if at the same time other questions are raising.