On the Use of Confidence Measures within an Interactive-predictive Machine Translation System

On the Use of Confidence Measures within an Interactive-predictive Machine Translation System Jes´us Gonz´alez-Rubio Inst. Tec. de Inform´atica Univ. Polit´ec. de Valencia 46021 Valencia, Spain jegonzalez@iti.upv.es Daniel Ortiz-Mart´ınez Dpto. de Sist. Inf. y Comp. Univ. Polit´ec. de Valencia 46021 Valencia, Spain dortiz@dsic.upv.es Francisco Casacuberta Dpto. de Sist. Inf. y Comp. Univ. Polit´ec. de Valencia 46021 Valencia, Spain fcn@dsic.upv.es Abstract In this work, we address the question of how to integrate confidence measures into a interactive-predictive machine transla- tion system and reduce user effort. Specif- ically, we propose to use word confidence measures to aid the user in validating cor- rect prefixes from the outputs given by the system. Experimental results obtained on a corpus of the Bulletin of the European Union show that confidence information can help to reduce user effort. 1 Introduction The research in the field of machine translation (MT) aims to develop computer systems which are able to translate text or speech without human in- tervention. However, present translation technol- ogy has not been able to deliver fully automated high-quality translations (Kay, 1997; Hutchins, 1999; Arnold, 2003). Typical solutions to improve the quality of the translations supplied by an MT system require manual post-editing. This serial process prevents the MT system from taking ad- vantage of the knowledge of the human translator and the human translator can not take advantage of the adapting ability of the MT system. An alternative way to take advantage of the ex- isting MT technologies is to use them in collabo- ration with human translators within a computer- assisted translation (CAT) or interactive frame- work (Isabelle and Church, 1997). Interactivity in CAT has been explored for a long time. Sys- tems have been designed to interact with human translators in order to solve ambiguities or update c⃝ 2010 European Association for Machine Translation. user dictionaries (Slocum, 1985; Whitelock et al., 1986). An important contribution to CAT technology was pioneered by the TransType project (Foster et al., 1997; Langlais and Lapalme, 2002; Foster et al., 2002). It 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 proposed in that work was to embed data driven MT techniques within the interactive translation environment. Following the TransType ideas, Bar- rachina et al. (2009) proposed, in the TransType- 2 project, the use of fully-fledged statistical MT (SMT) systems to produce full target sentences hypotheses, or portions thereof, which can be ac- cepted or amended by a human translator. Each correct text segment is then used by the MT sys- tem as additional information to achieve improved suggestions. More specifically, in each iteration, a prefix1 of the target sentence is fixed by the hu- man translator and, in the next iteration, the system predicts a best (or N-best) translation suffix(es)1 to complete this prefix. This process is known as Interactive-predictive Machine Translation (IMT). In this paper, we also focus on the IMT approach …

Content truncated to 500 words