Online Automatic Post-editing for MT in a Multi-Domain Translation Environment

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 525–535, Valencia, Spain, April 3-7, 2017. c⃝2017 Association for Computational Linguistics Online Automatic Post-editing for MT in a Multi-Domain Translation Environment Rajen Chatterjee(1,2), Gebremedhen Gebremelak(1,2), Matteo Negri(2), Marco Turchi(2) (1)University of Trento (2)Fondazione Bruno Kessler {chatterjee,gebremelak,negri,turchi}@fbk.eu Abstract Automatic post-editing (APE) for machine translation (MT) aims to fix recurrent er- rors made by the MT decoder by learn- ing from correction examples. In con- trolled evaluation scenarios, the represen- tativeness of the training set with respect to the test data is a key factor to achieve good performance. Real-life scenarios, however, do not guarantee such favorable learning conditions. Ideally, to be inte- grated in a real professional translation workflow (e.g. to play a role in computer- assisted translation framework), APE tools should be flexible enough to cope with continuous streams of diverse data coming from different domains/genres. To cope with this problem, we propose an online APE framework that is: i) robust to data diversity (i.e. capable to learn and apply correction rules in the right contexts) and ii) able to evolve over time (by contin- uously extending and refining its knowl- edge). In a comparative evaluation, with English-German test data coming in ran- dom order from two different domains, we show the effectiveness of our approach, which outperforms a strong batch system and the state of the art in online APE. 1 Introduction Automatic post-editing (APE) systems for ma- chine translation (MT) aim to correct the errors present in a machine-translated text before show- ing it to the user, thereby reducing human work- load and eventually increase translation produc- tivity. The choice of having such post-processing systems is well motivated in (Bojar et al., 2015) and becomes a must when the MT engine used to translate is not directly accessible for retrain- ing or for more radical internal modifications (e.g. when working with a third party MT system). As pointed out by (Parton et al., 2012; Chatterjee et al., 2015b), from the application point of view an APE system can help to: i) Improve MT output by exploiting information unavailable to the de- coder, or by performing deeper text analysis that is too expensive at the decoding stage; ii) Pro- vide professional translators with improved MT output quality to reduce (human) post-editing ef- fort and iii) Adapt the output of a general-purpose MT system to the lexicon/style requested in a spe- cific application domain. Similar to what is usu- ally done in MT, APE components learn post- editing rules from “parallel” corpora consisting of machine-translated text (mt, optionally with its corresponding source text – src) and its post-edits (pe) provided by human post-editors. The effec- tiveness of learning from relatively small amounts of post-edited data is evident from the impressive outcomes of the recently held APE shared task at WMT 2016 (Bojar et al., 2016). Different APE paradigms, like neural (Junczys-Dowmunt and Grundkiewicz, 2016), hybrid (Chatterjee et al., 2016), and …

Content truncated to 500 words