A Syntactic Pattern Recognition Approach to Computer Assisted Translation Jorge Civera 1 , Juan M. Vilar 2 , Elsa Cubel 1 , Antonio L. Lagarda 1 , Sergio Barrachina 2 , Francisco Casacuberta 1 , Enrique Vidal 1 , David Pic´ o 1 , and Jorge Gonz´ alez 1 1 Departamento de Sistemas Inform´ aticos y Computaci´ on, Universitat Polit` ecnica de Val` encia Instituto Tecnol´ ogico de Inform´ atica, E-46071 Val` encia, Spain tt2iti@iti.upv.es http://www.iti.upv.es/˜prhlt 2 Departamento de Lenguajes y Sistemas Inform´ aticos Universitat Jaume I, E-12071 Castell´ on de la Plana, Spain Abstract. It is a fact that current methodologies for automatic translation cannot be expected to produce high quality translations. An alternative approach is to use them as an aid to manual translation. We focus on a possible way to help human translators: to interactively provide completions for the parts of the sentences already translated. We explain how finite state transducers can be used for this task and show experiments in which the keystrokes needed to translate printer manuals were reduced to nearly 25% of the original. 1 Introduction It is becoming increasingly clear that current automatic translation methodologies can- not be expected to produce high quality translation in the near future. An alternative way to take advantage of the technologies developed is to use them in order to help human translators. One such approach, proposed by [1], can be explained as follows: the translator begins to type the translation and the system guesses the best completion for the text typed so far. The user can then accept the suggestion of the computer or part of it. This should reduce the amount of work of the translator. This approach has two important aspects: the models need to provide adequate com- pletions and they have to do so efficiently. To fulfill these two requirements, we have decided to use Stochastic Finite State Transducers (SFST) since they have proved in the past to be able to provide adequate translations [2–4] and, as we show in this paper, efficient parsing algorithms can be easily adapted in order to provide completions. The rest of the paper is structured as follows. The following section presents the general setting for machine translation and finite state models. In section 3, the search procedure for an interactive translation is presented. Experimental results are presented in section 4. Finally, some conclusions and future work are explained in section 5. 2 Machine Translation with Finite-State Transducers Given a source sentence s, the goal of MT is to find a target sentence ˆ t that maximizes: ˆ t = argmax t Pr(t | s)= argmax t Pr(t, s) ≈ argmax t Pr T (t, s) (1) A. Fred et al. (Eds.): SSPR&SPR 2004, LNCS 3138, pp. 207–215, 2004. c  Springer-Verlag Berlin Heidelberg 2004