The AFRL-MITLL WMT17 Systems: Old, New, Borrowed, BLEU

Proceedings of the Conference on Machine Translation (WMT), Volume 2: Shared Task Papers, pages 303–309 Copenhagen, Denmark, September 711, 2017. c⃝2017 Association for Computational Linguistics The AFRL-MITLL WMT17 Systems: Old, New, Borrowed, BLEU Jeremy Gwinnup, Timothy Anderson, Grant Erdmann, Katherine Young Air Force Research Laboratory {jeremy.gwinnup.1,timothy.anderson.20, grant.erdmann,katherine.young.1.ctr} @us.af.mil Michaeel Kazi, Elizabeth Salesky, Brian Thompson, Jonathan Taylor MIT Lincoln Laboratory {michaeel.kazi,elizabeth.salesky, brian.thompson,jonathan.taylor} @ll.mit.edu Abstract This paper describes the AFRL-MITLL machine translation systems and the im- provements that were developed during the WMT17 evaluation campaign. This year, we explore the continuing prolifera- tion of Neural Machine Translation toolk- its, revisit our previous data-selection ef- forts for use in training systems with these new toolkits and expand our participation to the Russian–English, Turkish–English and Chinese–English translation pairs. 1 Introduction As part of the 2017 Conference on Machine Trans- lation (WMT, 2017) news-translation shared task, the MITLL and AFRL human language technol- ogy teams participated in the Russian–English, English–Russian, Turkish–English and Chinese– English tasks. Our machine translation systems this year are a departure from our previous Moses (Koehn et al., 2007) based systems from WMT16 (Gwin- nup et al., 2016). We employ systems built with the Nematus (Sennrich et al., 2017) toolkit as in our IWSLT2016 (Kazi et al., 2016) systems, the Nematus-compatible Marian training toolkit and AmuNMT decoder (Junczys-Dowmunt et al., 2016) and the OpenNMT (Klein et al., 2017) toolkit. For the Russian–English and Turkish–English language pairs, we submitted an entry comprising the best systems combined using the Jane system combination method (Freitag et al., 2014) and the best-scoring single system for that language pair. Portions of this work are sponsored by the Air Force Re- search Laboratory under Air Force contracts FA-8721-05-C- 0002 and FA-8650-09-D-6939-029. For the Chinese–English and English-Russian lan- guage pairs, we only submitted our single-best sys- tem. 2 Data and Preparation 2.1 Data Used We utilized all available data sources provided for the language pairs we participated in, including the Commoncrawl (Smith et al., 2013), Yandex1, UN v1.0 (Ziemski et al., 2016), SETimes (Tyers and Alperen, 2010) corpora. 2.2 Data Preparation The Russian/English files were cleaned to re- move blank lines, replace carriage returns with line feed characters, remove wrong-language text, and correct mixed alphabet spellings, following techniques outlined in (Young et al., 2016) and (Schwartz et al., 2014). The number of non-parallel blank lines in the Russian/English news commentary files indicated some sentence alignment errors, so these files were re-processed using the NLTK Punkt (Kiss and Strunk, 2006) sentence segmenter and the Cham- pollion sentence aligner (Ma, 2006) before clean- ing. Altogether, 9537 of the original 236,314 newscommentary lines were removed during the clean-up process. The Chinese files were word-segmented with Jieba2 and the Stanford Chinese segmenter (Chang et al., 2008). The Chinese–English parallel data was cleaned to replace carriage returns and to re- move wrong-language text. Lines with URLs in http were also removed, because of a difference in the Chinese and English tokenization. Alto- gether, the clean-up process removed 310,121 of the 21,248,495 lines in the …

Content truncated to 500 words