Finite-state models for computer assisted translation
Elsa Cubel, Jorge Civera, Juan Miguel Vilar, Antonio L. Lagarda, Francisco Casacuberta, Enrique Vidal, David Picó, Jorge González, Luis Javier Rodríguez-Fuentes · 2004
Abstract. Current methodologies for automatic translation cannot be expected to produce high quality translations. However, some techniques based on these methodologies can increase the produc-tivity of human translators. The basis of one of these methodolo-gies are finite-state transducers, which are adequate models for com-puter assisted translation. These models have proved its efficiency in many pattern recognition and artificial intelligence tasks such as speech recognition, handwriting recognition and machine translation for specific domains. These finite-state models present some advantages. On the one hand, finite-state models can be learnt from bilingual corpus to in-fer transducers. On the other hand, there are well-known and effi-cient algorithms to perform the parse of the best translation accor-ding to these models (e.g. Viterbi search). In this paper, the concept of interactive search will be introduced along with some efficient techniques that solve the problem of producing a translation given a sentence in the source language and a prefix (from the output sen-tence) typed by the user. Needless to say that this system must run under real-time constraints to be useful for human translators. This approach has been tested on a corpus of printer manuals and the first results reflect that human translators would only need to type the 25 % of the characters of the whole translated text, increasing in this way their throughput and reducing their effort. 1