This paper presents a statistical approach to machine translation, focusing on the translation from French to English. The authors describe a system that uses language models and translation models to estimate the probability of source sentences given a target sentence, aiming to find the most probable source sentence. The system employs an n-gram model for language modeling and a translation model that considers word alignments and distortions. The parameters of these models are estimated using a large database of aligned sentence pairs from the Canadian parliament proceedings (Hansards). The authors report on two pilot experiments, one for parameter estimation and another for translating French to English, achieving 48% success rate in the latter. They discuss future improvements, including handling dependencies between source words, improving language modeling, and integrating morphological and syntactic information.This paper presents a statistical approach to machine translation, focusing on the translation from French to English. The authors describe a system that uses language models and translation models to estimate the probability of source sentences given a target sentence, aiming to find the most probable source sentence. The system employs an n-gram model for language modeling and a translation model that considers word alignments and distortions. The parameters of these models are estimated using a large database of aligned sentence pairs from the Canadian parliament proceedings (Hansards). The authors report on two pilot experiments, one for parameter estimation and another for translating French to English, achieving 48% success rate in the latter. They discuss future improvements, including handling dependencies between source words, improving language modeling, and integrating morphological and syntactic information.