Algorithme de recherche d'un rang de prédiction. Application à l'évaluation de modèles de langage
Pierre Alain, Olivier Boeffard.

Within a predictive framework for language model evaluation, Shannon uses a rank distribution in order to bound the entropy of printed English. Taking into account of higher dimensions (prediction of symbols in a raw) and predicting a k-word sequence given a N-word vocabulary is a NP-hard computational task. To achieve this goal, we propose some acceptable and effective search heuristics for an A* algorithm.