A constraint satisfaction approach to machine translation

Constraint satisfaction inference is presented as a generic, theory-neutral inference engine for machine translation. The approach enables the integration of many different solutions to aspects of the output space, including classification-based translation models that take source-side context into account, as well as stochastic components such as target language models. The approach is contrasted with a word-based SMT system using the same decoding algorithm, but optimising a different objective function. The incorporation of sourceside context models in our model filters out many irrelevant candidate translations, leading to superior translation scores.