Machine Translation with Binary Feedback: a Large-Margin Approach
Avneesh Saluja, Ian Richard Lane, Ying Zhang · 2012
Viewing machine translation as a structured classification problem has provided a gateway for a host of structured prediction techniques to enter the field. In particular, large-margin structured prediction methods for discrimina-tive training of feature weights, such as the structured perceptron or MIRA, have started to match or exceed the performance of exist-ing methods such as MERT. One issue with structured problems in general is the diffi-culty in obtaining fully structured labels, e.g., in machine translation, obtaining reference translations or parallel sentence corpora for ar-bitrary language pairs. Another issue, more specific to the translation domain, is the diffi-culty in online training of machine translation systems, since existing methods often require bilingual knowledge to correct translation out-put online. We propose a solution to these two problems, by demonstrating a way to incorpo-rate binary-labeled feedback (i.e., feedback on whether a translation hypothesis is a “good ” or understandable one or not), a form of super-vision that can be easily integrated in an on-line manner, into a machine translation frame-work. Experimental results show marked im-provement by incorporating binary feedback on unseen test data, with gains exceeding 5.5 BLEU points. 1