In this talk I'll introduce the current state of the art in statistical machine translation (SMT) and outline my work at the University of Edinburgh modelling machine translation (MT) as a probabilistic machine learning problem. Although SMT systems have made large gains in translation quality in recent years, most are currently induced using a hand engineered pipeline of disparate models linked by heuristics. Although such techniques are effective for translating between related languages (e.g. English and French), they fail to capture the latent structure necessary to translate between languages which diverge significantly in syntactic structure, such as Chinese and English. I'll present non-parametric Bayesian models for inducing synchronous context free grammars. These models are capable of learning the latent structure of translation equivalence from a corpus of parallel string pairs. I'll discuss the difficult inference problems posed by such models and describe Monte Carlo sampling techniques that can help solve them. Finally I'll present experiments demonstrating competitive results on full scale translation evaluations.
