NRC Russian-English Machine Translation System for WMT 2016

Abstract

We describe the statistical machine translation system developed at the National Research Council of Canada (NRC) for the Russian-English news translation task of the First Conference on Machine Translation (WMT 2016). Our submission is a phrase-based SMT system that tackles the morphological complexity of Russian through comprehensive use of lemmatization. The core of our lemmatization strategy is to use different views of Russian for different SMT components: word alignment and bilingual neural network language models use lemmas, while sparse features and reordering models use fully inflected forms. Some components, such as the phrase table, use both views of the source. Russian words that remain out-ofvocabulary (OOV) after lemmatization are transliterated into English using a statistical model trained on examples mined from the parallel training corpus. The NRC Russian-English MT system achieved the highest uncased BLEU and the lowest TER scores among the eight participants in WMT 2016.

Findings of the 2016 Conference on Machine Translation

“…University of Munich / Charles University (Tamchyna et al, 2016) METAMIND Salesforce Metamind (Bradbury and Socher, 2016) NRC National Research Council Canada (Lo et al, 2016) NYU-MONTERAL New York University / University of Montréal (Chung et al, 2016) PARFDA Ergun Bicici (Bicici, 2016a) PJATK Polish-Japanese Academy of Inf. Technology (Wołk and Marasek, 2016) PROMT PROMT Automated Translation Solutions (Molchanov and Bykov, 2016) Four online statistical machine translation systems Table 2: Participants in the shared translation task.…”

NRC Machine Translation System for WMT 2017

“…The core of the NRC WMT16 MT system (Lo et al, 2016) is Portage (Larkin et al, 2010). Portage is a conventional log-linear phrase-based SMT system.…”

“…The final output was truecased and detokenized in the same way as described in Lo et al (2016). Table 1 shows the results of selected models from our development experiments.…”

“…The five NNJM rescoring features include two Russian-word NNJMs and three Russian-lemma ones. Following Devlin et al (2014) The final output was truecased and detokenized in the same way as described in Lo et al (2016).…”

“…We present NRC's submission to the Russian-English and Chinese-English news translation task of WMT 2017. In contrast to last year, when we participated in the Russian-English task only, with our well-developed phrase-based statistical machine translation system (Lo et al, 2016;Larkin et al, 2010;Foster et al, 2009), this year we built large-scale state-of-the-art neural machine translation (NMT) systems for these two language pairs to facilitate further understanding and discussion of NMT.…”