Active learning for interactive machine translation

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 245–254, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Active learning for interactive machine translation Jes´us Gonz´alez-Rubio and Daniel Ortiz-Mart´ınez and Francisco Casacuberta D. de Sistemas Inform´aticos y Computaci´on U. Polit`ecnica de Val`encia C. de Vera s/n, 46022 Valencia, Spain {jegonzalez,dortiz,fcn}@dsic.upv.es Abstract Translation needs have greatly increased during the last years. In many situa- tions, text to be translated constitutes an unbounded stream of data that grows con- tinually with time. An effective approach to translate text documents is to follow an interactive-predictive paradigm in which both the system is guided by the user and the user is assisted by the system to generate error-free translations. Unfortu- nately, when processing such unbounded data streams even this approach requires an overwhelming amount of manpower. Is in this scenario where the use of active learn- ing techniques is compelling. In this work, we propose different active learning tech- niques for interactive machine translation. Results show that for a given translation quality the use of active learning allows us to greatly reduce the human effort required to translate the sentences in the stream. 1 Introduction Translation needs have greatly increased during the last years due to phenomena such as global- ization and technologic development. For exam- ple, the European Parliament1 translates its pro- ceedings to 22 languages in a regular basis or Project Syndicate2 that translates editorials into different languages. In these and many other ex- amples, data can be viewed as an incoming un- bounded stream since it grows continually with time (Levenberg et al., 2010). Manual translation of such streams of data is extremely expensive given the huge volume of translation required, 1http://www.europarl.europa.eu 2http://project-syndicate.org therefore various automatic machine translation methods have been proposed. However, automatic statistical machine trans- lation (SMT) systems are far from generating error-free translations and their outputs usually require human post-editing in order to achieve high-quality translations. One way of taking ad- vantage of SMT systems is to combine them with the knowledge of a human translator in the interactive-predictive machine translation (IMT) framework (Foster et al., 1998; Langlais and La- palme, 2002; Barrachina et al., 2009), which is a particular case of the computer-assisted trans- lation paradigm (Isabelle and Church, 1997). In the IMT framework, a state-of-the-art SMT model and a human translator collaborate to obtain high- quality translations while minimizing required human effort. Unfortunately, the application of either post- editing or IMT to data streams with massive data volumes is still too expensive, simply because manual supervision of all instances requires huge amounts of manpower. For such massive data streams the need of employing active learning (AL) is compelling. AL techniques for IMT se- lectively ask an oracle (e.g. a human transla- tor) to supervise a small portion of the incoming sentences. Sentences are selected so that SMT models estimated from them translate new sen- tences as accurately as possible. There are three challenges when applying …

Content truncated to 500 words