Interactive Machine Translation Based on Partial Statistical Phrase-based Alignments

International Conference RANLP 2009 - Borovets, Bulgaria, pages 330–336 Interactive machine translation based on partial statistical phrase-based alignments Daniel Ortiz-Mart´ınez Dpto. de Sist. Inf. y Comp. Univ. Polit´ecnica de Valencia 46071 Valencia, Spain dortiz@dsic.upv.es Ismael Garc´ıa-Varea Dpto. de Sist. Inf. Univ. de Castilla-La Mancha 02071 Albacete, Spain ivarea@info-ab.uclm.es Francisco Casacuberta Dpto. de Sist. Inf. y Comp. Univ. Polit´ecnica de Valencia 46071 Valencia, Spain fcn@dsic.upv.es Abstract State-of-the-art Machine Translation (MT) sys- tems are still far from being perfect. An alterna- tive is the so-called Interactive Machine Transla- tion (IMT) framework. In this framework, the knowledge of a human translator is combined with a MT system. We present a new tech- nique for IMT which is based on the generation of partial alignments at phrase-level. The pro- posed technique partially aligns the source sen- tence with the user prefix and then translates the unaligned portion of the source sentence. The generation of such partial alignments is driven by statistical phrase-based models. Our tech- nique relies on the application of smoothing tech- niques over the phrase models to appropriately assign probabilities to unseen events. We report experiments investigating the impact of the dif- ferent smoothing techniques in the accuracy of our system. In addition, we compare the results obtained by our system with those obtained by other well-known IMT systems. Keywords Statistical machine translation, interactive machine translation, phrase-based translation, phrase-based alignments, smoothing. 1 Introduction Information technology advances in modern society have led to the need of more efficient methods of trans- lation. It is worth mentioning that current MT sys- tems are not able to produce ready-to-use texts. In- deed, MT systems usually require human post-editing in order to achieve high-quality translations. One way of taking advantage of MT systems is to combine them with the knowledge of a human transla- tor, constituting the Interactive Machine Translation (IMT) paradigm. This IMT paradigm can be consid- ered a special type of the so-called Computer-Assisted Translation (CAT) paradigm. An important contribution to IMT technology was carried out within the TransType (TT) project [11, 7, 5]. This project entailed a focus shift in which inter- action directly aimed at the production of the target text, rather than at the disambiguation of the source text, as in former interactive systems. The idea pro- posed in that work was to embed data driven MT tech- niques within the interactive translation environment. Following these TT ideas, [1] proposed a new ap- proach to IMT. In this approach, fully-fledged statisti- cal MT (SMT) systems are used to produce full target sentence hypotheses, or portions thereof, which can be partially or completely accepted and amended by a human translator. Each partial correct text segment is then used by the SMT system as additional infor- mation to achieve further, hopefully improved sugges- tions. Figure 1 illustrates a typical IMT session. In this paper, we also focus on the IMT approach to CAT. Specifically, we propose a new IMT engine based on the generation of partial alignments at phrase-level. The proposed technique partially …

Content truncated to 500 words