QCRI’s Machine Translation Systems for IWSLT’16

QCRI’s Machine Translation Systems for IWSLT’16 Nadir Durrani, Fahim Dalvi, Hassan Sajjad, Stephan Vogel Qatar Computing Research Institute – HBKU {ndurrani,faimaduddin,hsajjad,svogel}@qf.org.qa Abstract This paper describes QCRI’s machine translation sys- tems for the IWSLT 2016 evaluation campaign. We partic- ipated in the Arabic→English and English→Arabic tracks. We built both Phrase-based and Neural machine translation models, in an effort to probe whether the newly emerged NMT framework surpasses the traditional phrase-based sys- tems in Arabic-English language pairs. We trained a very strong phrase-based system including, a big language model, the Operation Sequence Model, Neural Network Joint Model and Class-based models along with different domain adap- tation techniques such as MML filtering, mixture modeling and using fine tuning over NNJM model. However, a Neural MT system, trained by stacking data from different genres through fine-tuning, and applying ensemble over 8 models, beat our very strong phrase-based system by a significant 2 BLEU points margin in Arabic→English direction. We did not obtain similar gains in the other direction but were still able to outperform the phrase-based system. We also applied system combination on phrase-based and NMT outputs. 1. Introduction We describe QCRI’s phrase-based and Neural MT systems. We participated in the Arabic-to-English and English-to- Arabic MT tracks. Our translation engines have been his- torically based on the phrase-based system trained using the Moses toolkit [1], but during the course of this evaluation, we made a transition towards the newly emerged Neural MT framework [2], using Nematus, a toolkit used by the top per- forming team [3], during the recent WMT campaign. An interesting challenge associated with the IWSLT cam- paign is the problem of domain adaptation. The in-domain data based on TED talks is available in very little quan- tity compared to the out-domain UN corpus [4], which has been found to be harmful previously when simply concate- nated to the training [5]. In this year’s IWSLT, two addi- tional data resources Opus subtitles [6] and the QED corpus [7] were introduced. The latter was also used as an official test-set. Therefore apart from exploring phrase-based versus Neural MT, we geared ourselves towards adapting our sys- tem towards TED and QED talks in this multi-domain sce- nario. With these goals in mind we re-explored both model weighting and data filtering techniques, in these new data set- tings. Below we itemize the most successful attributes of our phrase-based system: • We applied MML-based data selection [8] to the UN and Open Sub-title data, with the goals of filtering out harmful data. • We trained OSM models [9] on separate corpora, and interpolated them [10] by optimizing perplexity on the tuning-set. We also tried this on the OSM models trained on the word classes [11]. • We tried the fine-tuning method of training the NNJM model on the out-domain data and fine-tuning with the in-domain TED data [12]. • We trained big language models using all the English mono data available from the WMT campaign and giga word corpus for Arabic. We trained our Neural MT …

Content truncated to 500 words