PRIMT: A Pick-Revise Framework for Interactive Machine Translation

Proceedings of NAACL-HLT 2016, pages 1240–1249, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics PRIMT: A Pick-Revise Framework for Interactive Machine Translation Shanbo Cheng, Shujian Huang, Huadong Chen, Xinyu Dai and Jiajun Chen State Key Laboratory for Novel Software Technology Nanjing University Nanjing 210023, China {chengsb, huangsj, chenhd, daixy, chenjj}@nlp.nju.edu.cn Abstract Interactive machine translation (IMT) is a method which uses human-computer interac- tions to improve the quality of MT. Tradition- al IMT methods employ a left-to-right order for the interactions, which is difficult to di- rectly modify critical errors at the end of the sentence. In this paper, we propose an IMT framework in which the interaction is decom- posed into two simple human actions: pick- ing a critical translation error (Pick) and revis- ing the translation (Revise). The picked phrase could be at any position of the sentence, which improves the efficiency of human computer in- teraction. We also propose automatic sugges- tion models for the two actions to further re- duce the cost of human interaction. Experi- ment results demonstrate that by interactions through either one of the actions, the transla- tion quality could be significantly improved. Greater gains could be achieved by iteratively performing both actions. 1 Introduction To obtain high quality translations, human transla- tors usually have to modify the results generated by a machine translation (MT) system (called post edit- ing, PE). In many cases, PE needs a lot of modifica- tions, which is time-consuming (Plitt and Masselot, 2010). To speed up the process, interactive ma- chine translation (IMT) is proposed which instantly update the translation result after every human ac- tion (Langlais et al., 2000; Foster et al., 2002; Bar- rachina et al., 2009; Koehn, 2009; González-Rubio et al., 2013; Alabau et al., 2014). Because the trans- lation quality could be improved after every update, IMT is expected to generate high quality transla- tions with less human actions (Sanchis-Trilles et al., 2014). Typical IMT systems usually use a left-to- right sentence completing framework pioneered by Langlais et al (2000), in which the users process the translation from the beginning of the sentence and interact with the system at the left-most error. By assuming the translation from the beginning to the modified part (called "prefix") to be correct, the sys- tem generates new translations after the given pre- fix (Koehn, 2009; Barrachina et al., 2009; Ortiz, 2011; Alabau et al., 2014). Despite the success of this left-to-right frame- work, one potential weakness is that it is difficult to modify critical translation errors at the end of a sen- tence. Critical translation errors are those errors that has large impact on the translation of other words or phrases. When a translation ambiguity occurs at the end of a sentence while it causes translation errors at the beginning, modifying this critical errors first may bring great positive effects on previous parts of the translation, which may reduce human efforts in an IMT process. Modifying from left to right will delay the modification of the …

Content truncated to 500 words