Maximum Entropy Modeling: A Suitable Framework to Learn Context-Dependent Lexicon Models for Statistical Machine Translation

摘要: Current statistical machine translation systems are mainly based on word lexicons. However, these models usually context-independent, therefore, the disambiguation of a source must be carried out using other probabilistic distributions (distortion and language models). One efficient way to add contextual information lexicons is maximum entropy modeling. In that framework, context introduced through feature functions allow us automatically learn context-dependent lexicon models. In first approach, modeling after process learning standard (alignment lexicon). second integrated in expectation-maximization models. Experimental results were obtained for two well-known tasks, French--English Canadian Parliament Hansards task German--English Verbmobil task. These proved use both approaches, can help improve performance systems.