Publications

Advanced search

Abstract

Francisco Casacuberta, Enrique Vidal. Speech in Machine Translation and Computer-Assisted Translation. The ELRA Newsletter, 2005. Vol. 10 (3-4), pp. 5-8.

Translation services are fundamental in the administrative organization of the European Union (EU). Moreover, some EU members have more than one official language (e.g. Spain, Belgium, etc.). In these cases, the official writings, the Parliament speeches, etc. must be produced in all the official languages. On the other hand, the translation of many court documents becomes crucial to avoid a slow down of the proceedings. Finally, the translation to other non official EU languages is becoming increasingly important due to the existence of an important immigrant community (medical consultation, legal consulting, services, etc.) The need of translation of huge volumes of documents has fostered the development of machine translation (MT) systems. Nowadays, there are many commercial MT systems for text-to-text translation (T2TT) available. But T2TT systems are far to be perfect and high-quality translation is required in many cases. In practice, T2TT systems generally need human post-processing to correct the possible errors incurred by the system. However, this is a time consuming process due to the required human effort (generally, less than the effort needed to translate the document without any tool). An alternative to increase the productivity of the whole translation process (T2TT plus human work) is to incorporate the human correction activities within the translation process itself, in a computer-assisted translation (CAT) system. The idea is to use a T2TT system in an iterative process where human translator activity is included in the loop. In spite of the limitations of the present T2TT systems, they are widely accepted in many private and public organizations. Unfortunately, this is not the situation of real speech-to-speech (or speech-to-text) translation (S2ST), where the reliable systems are laboratory prototypes because the present technology only allows S2ST in very restricted domains. However, speech can play an important role in MT, since the human translator can use speech recognizers in addition to the keyboard and the mouse. He or she can dictate portions of target sentences that the system has produced with errors in order to correct them as well as to give commands to the CAT system.