In this paper, we describe a new language model based on dependent word sequences organized in multi-level hierarchy. We call this model MCnv, where n is the maximum number of words in a sequence and $\nu$ is the maximum number of levels. The originality of this model is its capability to take into account dependent variable-length sequences for very large vocabulary. In order to discover the variable-length sequences and to build the hierarchy, we use a set of 233 syntactic classes extracted from the eight French elementary grammatical classes. The MCnv model learns hierarchical word patterns and uses them to reevaluate and filter the n-best utterance hypotheses outputed by our speech recognizer MAUD. The model have been trained on a corpus (LeM) of 43 million of words extracted from ``Le Monde'' a French newspapers and uses a vocabulary of 20000 words. Tests have been conducted on 300 sentences. Results achieved 17% decrease in perplexity compared to an interpolated class trigram model. Rescoring the original n-best hypotheses results also in an improvement of 5% in accuracy.