This work is a study that is part of the creation of the lexicon of a categorial dependency grammars (CDG) for French and also part of a mixed stochasticdeterministic analysis for large-scale dependency grammars. In particular we develop algorithms for CDG to improve the existing lexicon of the French CDG.We solve several problems for the analysis of these grammars for example, the absence of analysis proposed by the parser for some sentences. We present an algorithm proto-déverb which allows to complete the lexicon of the French CDG by using the sub-categorisation frame for deverbals. The second problem we consider is the fact that the CDG parser currently provides all the compatible solutions for a CDG. We propose a filtering approach to improve dependency parsing. We show that using a morpho-syntactic tagger that chooses the most probable grammatical classes for each lexical unit, we can significantly reduce the rate of ambiguities of the French CDG. Our study concluded that the adequacy of these solutions is mainly based on the compatibility between the lexical units defined by the taggers and the dependency grammar.
