Search Results

This chapter proposes algorithms for semisupervised learning. It translates sentences from the source language and then uses them to retrain the statistical machine translation (SMT) system in the hopes of getting a better translation system. It presents detailed experimental evaluations using French-English and Chinese-English data. The French-English translation task used bilingual data from the Europarl corpus, and monolingual data from the same domain as the test set which is drawn from the Canadian Hansard corpus. The Chinese-English task used bilingual data from the NIST large-data track and monolingual data from the Chinese Gigaword corpus.

Statistical machine translation (SMT) systems, which are trained on parallel corpora of bilingual text (e.g., French and English), typically work as follows: for each sentence to be translated, they generate a plethora of possible translations, from which they keep a smaller n-best list of the most likely translations. Even though the typical n-best list contains mostly high-quality candidates, the actual ranking is far from accurate. This chapter presents a novel approach to reranking the n-best list produced by an SMT system. It uses an ensemble of perceptrons that are trained in parallel, each of them on just a fraction of the available data. Experiments were performed on two large-scale commercial systems: a Chinese-to-English system trained on 80 million words and a French-to-English system trained on 1.1 billion words. The reranker obtained statistically significant improvements of about 0.5 and 0.2 BLEU points on the Chinese-to-English and the French-to-English system, respectively.

This chapter explores the application of discriminative learning to the problem of phrase selection in statistical machine translation (SMT). The chapter is organized as follows. Section 11.2 describes previous and current approaches to dedicated word selection. Section 11.3 describes the approach to discriminative phrase translation (DPT). It presents experimental results on the application of DPT models to the Spanish-to-English translation of European Parliament proceedings. In Section 11.4, prior to considering the full translation task, it measures the local accuracy of DPT classifiers at the isolated phrase translation task in which the goal is not to translate the whole sentence but only individual phrases without having to integrate their translations in the context of the target sentence. Section 11.5 tackles the full translation task while Section 11.6 summarizes the main conclusions.

This chapter presents a method for training all the parameters of a syntax-aware statistical machine translation (MT) system in a discriminative manner. The system outperforms a generative syntax-aware baseline. Although all the standard information sources necessary for a state-of-the-art MT system have yet to be added, but the scalability of the system suggests that the main obstacle to doing so has been overcome. The next step is to generalize the tree transducer into a bitree transducer, so that it can modify the target side of the bitree after it is inferred from the source side.

This chapter introduces a method for taking advantage of background linguistic resources in statistical machine translation. It starts with a brief introduction to word-sequence kernels, followed by a description of the notion of factored representation and details of the kernel formulation. The next section validates the kernel construction on an artificial discrimination task reproducing some of the conditions encountered in translation. The chapter concludes with a discussion of related and future work.