Constraint-Based Language Translation for Approximate Redundancy Resolution

The proposed project aims to improve machine translation (MT) quality (e.g., for translating from Chinese to English) of Statistical Machine Translation (SMT) systems. Redundancies (also known as over-generation) are an important problem for almost current state-of-the-art statistical machine translation systems. One the one hand, they pollute the search space with hypotheses that can lead to identical translations. On the other hand, these redundancies can be used to re-estimate the probabilities of translation candidates and help improve translation quality by reranking the n-best output of a decoder. Current approaches exploit redundancies after decoding and therefore can not influence the generation of the search space. In this project, we aim to exploit redundancies, introduced by a overlapping translation rules, before decoding to steer the search space generation in a way that already takes into account some of the spurious ambiguities that can arise during decoding. The sources of spurious ambiguities are modeled as constraints that can restrict the search space in different ways. This project is breaking new ground by applying approximate constraint solving methods to statistical machine translation. The expected results of the project will significantly enhance our understanding of how redundancies can be exploited to influence the generation of the search space and thereby improve translation quality.