Domain adaptation has recently gained interest in statistical machine translation to cope with the performance drop observed when testing conditions deviate from training conditions. The basic idea is that in-domain training data can be exploited to adapt all components of an already developed system. Previous work showed small performance gains by adapting from limited in-domain bilingual data. Here, we aim instead at significant performance gains by exploiting large but cheap monolingual in-domain data, either in the source or in the target language. We propose to synthesize a bilingual corpus by translating the monolingual adaptation data into the counterpart language. Investigations were conducted on a stateof-the-art phrase-based system trained on the Spanish–English part of the UN corpus, and adapted on the corresponding Europarl data. Translation, re-ordering, and language models were estimated after translating in-domain texts with the baseline. By optimizing the interpolation of these models on a development set the BLEU score was improved from 22.60% to 28.10 % on a test set. 1

Statistical machine translation systems are usually trained on large amounts of

Statistical machine translation (SMT) systems have made great strides in translation quality. However, high quality translation output is dependent on the availability of massive amounts of parallel text in the source and target language. There are a large number of languages that are considered “low-density”, either because the population speaking the language is not very large, or even if millions of people speak the language, insufficient online resources are available in that language. This tutorial covers machine learning approaches for dealing with such situations in statistical machine translation where the amount of available bilingual data is limited. A statistical translation system can be improved and/or adapted by incorporating new training data in the form of parallel text. The problem of learning from insufficient labeled training data has been dealt with in machine learning community under two general frameworks: (i) Semi-supervised Learning, and (ii) Active Learning. The goal of semi-supervised learning is to take advantage of abundant and cheap unlabeled data, together with labeled data, to build a high quality mapping from examples (the input space) to labels (the output space). On the other hand, the goal of active learning is to reduce the amount of labeled data required to learn a high

Abstract. Statistical Machine Translation (SMT) is currently used in real-time and commercial settings to quickly produce initial translations for a document which can later be edited by a human. The SMT models specialized for one domain often perform poorly when applied to other domains. The typical assumption that both training and testing data are drawn from the same distribution no longer applies. This paper evaluates domain adaptation techniques for SMT systems in the context of end-user feedback in a real world application. We present our experiments using two adaptive techniques, one relying on log-linear models and the other using mixture models. We describe our experimental results on legal and government data, and present the human evaluation effort for post-editing in addition to traditional automated scoring techniques (BLEU scores). The human effort is based primarily on the amount of time and number of edits required by a professional post-editor to improve the quality of machine-generated translations to meet industry standards. The experimental results in this paper show that the domain adaptation techniques can yield a significant increase in BLEU score (up to four points) and a significant reduction in post-editing time of about one second per word.