A general framework for minimizing translation effort: towards a principled combination of translation technologies in computer-aided translation

A general framework for minimizing translation effort: towards a principled combination of translation technologies in computer-aided translation Mikel L. Forcada Felipe S´anchez-Mart´ınez Dept. de Llenguatges i Sistemes Inform`atics Universitat d’Alacant E-03071 Alacant (Spain) {mlf,fsanchez}@dlsi.ua.es Abstract This paper motivates the need for an ho- mogeneous way of measuring and estimat- ing translation effort (quality) in computer- aided translation. It then defines a general framework for the measurement and esti- mation of translation effort so that trans- lation technologies can be both optimized and combined in a principled manner. In this way, professional translators will bene- fit from the seamless integration of all the technologies at their disposal when work- ing on a translation job. 1 Introduction Imagine that you are a professional translator and you are given a translation job. The text to be trans- lated comes divided in N segments, s1, s2, . . . sN: your job is therefore the ordered set {si}N i=1. You are supposed to turn this into a translation, that is, an ordered set {ti}N i=1 with the translations of each segment, and get paid for that. Yes, this is a sim- plified view of your work: the translation of each sentence is treated as an independent event, which is not always the case. In any case, even in this sim- plified form, the job is already quite challenging. Help coming your way? Of course, you could translate each segment si by hand, i.e. from scratch, into an suitable segment ti in the target language. You are a translator and you know this is usually harder than most people think. However, there are translation technologies out there that are supposed to help you by reducing your translation effort; they c⃝ 2015 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. usually come packaged as computer-aided transla- tion (CAT). Machine translation: You could, for instance, use machine translation (MT) to get a draft of the translation of each segment, MT(si); vendors and experts tell you that you will save effort by post- editing MT(si) into your desired translation ti.1 Machine translation output MT(si) may just be text, but it could come with annotations to help you make the most of it; for instance, words could be color-coded according to how confident the sys- tem is about them (Ueffing and Ney, 2007; Ueffing and Ney, 2005; Blatz et al., 2004), or unknown words that come out untranslated may be marked so that you spot them clearly. Machine-translated segments could even be accompanied by indicators of their estimated quality (Specia and Soricut, 2013; Specia et al., 2010; Blatz et al., 2004) which may be used to ascertain whether the output of the MT system is worth being post-edited or not. If some- one measured your post-editing effort (in time, in number of keystrokes, in number of words changed, in money you would have to pay another translator to do it, etc.), when turning MT(si) into ti, they could call …

Content truncated to 500 words