Statistical Machine Translation with Automatic Identification of Translationese

Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 47–57, Lisboa, Portugal, 17-18 September 2015. c⃝2015 Association for Computational Linguistics. Statistical Machine Translation with Automatic Identification of Translationese Naama Twitto-Shmuel Dept. of Computer Science University of Haifa Israel naama.twitto@gmail.com Noam Ordan Cluster of Excellence, MMCI Universit¨at des Saarlandes Germany noam.ordan@gmail.com Shuly Wintner Dept. of Computer Science University of Haifa Israel shuly@cs.haifa.ac.il Abstract Translated texts (in any language) are so markedly different from original ones that text classification techniques can be used to tease them apart. Previous work has shown that awareness to these differences can significantly improve statistical ma- chine translation. These results, how- ever, required meta-information on the on- tological status of texts (original or trans- lated) which is typically unavailable. In this work we show that the predictions of translationese classifiers are as good as meta-information. First, when a monolin- gual corpus in the target language is given, to be used for constructing a language model, predicting the translated portions of the corpus, and using only them for the language model, is as good as using the entire corpus. Second, identifying the por- tions of a parallel corpus that are translated in the direction of the translation task, and using only them for the translation model, is as good as using the entire corpus. We present results from several language pairs and various data sets, indicating that these results are robust and general. 1 Introduction Research in Translation Studies suggests that translated texts are considerably different from original texts, constituting a sublanguage known as Translationese (Gellerstam, 1986). Awareness to translationese can significantly improve statis- tical machine translation (SMT). Kurokawa et al. (2009) showed that French-to-English SMT sys- tems whose translation models were constructed from human translations from French to English yielded better translation quality than ones cre- ated from translations in the other direction. These results were corroborated by Lembersky et al. (2012a, 2013), who showed that translation mod- els can be adapted to translationese, thereby im- proving the quality of SMT even further. Aware- ness to translationese also benefits the language models used in SMT: Lembersky et al. (2011, 2012b) showed that language models complied from translated texts better fit the reference sets in term of perplexity, and SMT systems constructed from such language models perform much better than those constructed from original texts. To benefit from these results, however, one has to know whether the texts used for training SMT systems are original or translated, and previous work indeed used such meta-information. Unfor- tunately, annotation reflecting the status of texts, or the direction of translation, is typically unavail- able. The research question we investigate in this work is whether the predictions of translationese classifiers can replace manual annotation. In a va- riety of evaluation scenarios, we demonstrate that this is indeed the case. When a monolingual cor- pus in the target language is given for constructing a language model for SMT, we show that automat- ically identifying the translated portions of the cor- pus, and …

Content truncated to 500 words