– an MT system for closely related languages

The demonstration of our system addresses one very important part of the translation business – the localization of texts and programs from one source language into a group of mutually related target languages. It shows step by step a simple method of machine translation between related languages and its incorporation into an existing commercial translation aid using the concept of translation memory. It is quite clear that the localization of the same source into several typologically similar target languages, one language pair after another, is a waste of money and effort. In the translation process it is necessary to solve very similar problems for each source-target language pair. The use of one language from the target group as a pivot and to perform the translation and localization through this language seems to be quite natural solution of these problems. It is of course much easier to translate texts from Czech to Polish or from Russian to Bulgarian than from English or German to any of these languages. Introduction As part of our “pivot” language solution, we are using a combination of an MT system with a commercial machine aided translation (MAT) system. We are using the TRADOS system, although any such system will do. The system uses the concept of translation memory, which contains pairs of previously translated sentences from a source to a target language. When a human translator starts translating a new sentence, the system tries to match (with a degree of similarity set by a user) the source with sentences already stored in the translation memory. If found, the human translator decides whether to use it, to modify it or to reject it. 1 Translation Memory Integration The segmentation of the translation memory (the texts are stored as relevant pairs of source/target language sentences) is the key feature of our method. The translation memory may be exported into a text file and thus allows for an easy manipulation with its content. Let us suppose that we have at our disposal two translation memories – one human made for the source/pivot language pair and the other created by an MT system for the pivot/target language pair. The substitution of segments of a pivot language by the segments of a target language is then only a routine procedure. The human translator translating from the source to the target language then gets a translation memory for the required pair (source/target); there is no trace of the pivot language left. The system of penalties applied in TRADOS Translator’s Workbench guarantees that a previous humanmade translation present in the memory gets higher priority than the automatic translation. This method has at least three advantages: – The use of machine-made translation memory only as a resource supporting the direct human translation from the source to the target language has no negative effect on the quality of translation and from the user’s point of view. – There is no difference (except for the small difference in the quality of translation memories) when our method is used compared to the original process of working with the support of solely human-made translation memories. – The third advantage is the fact that given a sufficient quality of the MT from the pivot to the target language, our method may substantially increase the speed and reduce the costs of the translation from the source to the target languages. 2 The System SÍLKO The system !