A translation model for languages of accessing countries

90 A Translation Model For Languages of Acceding Countries Petr Homola, Vladislav Kuboň Inst.of Formal and Applied Linguistics, Charles University, Prague {homola|vk}@ufal.mff.cuni.cz Abstract. . The paper proposes a model for translation between syntactically similar languages of acceding countries. This model is based on the presupposition that the translation of related languages should exploit the relatedness by using as simple methods and tools as possible. In the first part the paper discusses the properties of some “new” languages, the second part describes a simple translation model which has already been tested on several pairs of syntactically similar languages.. 1. Introduction The historical event of EU enlargement scheduled for the 1st May of this year brings many new challenges. Apart from political and economical ones there are also linguistic challenges. The enlargement introduces ten new languages, increasing thus the number of official languages of EU to 21 and the number of language pairs to 210. It is quite clear that such a sudden huge increase of the number of language pairs requires solutions exploiting all possible advantages which might make the enormous translation task a little bit easier. In this paper we would like to propose a model for translation between those languages of acceding countries which are more or less related. This model is based on the presupposition that the translation of related languages should exploit the relatedness by using as simple methods and tools as possible. It also advocates the idea that although for every acceding country it is extremely important to translate from “big” EU languages (English, German, French), it would be useful to shift the research focus to a machine translation among related “new” languages. It might quickly solve at least some problems of translation inside enlarged EU and to help to bridge a gap until full- fledged MT systems are developed. 2. The classification of relatedness Natural languages are grouped into language families. Usually, languages from the same language family are more similar than languages from different language families. Although the division of languages into language families is not a perfect criterion for classification, it provides a raw hierarchy. Let us distinguish the following levels of proximity of languages with examples of • language pairs from the Central and East Europe: Variants of one `underlying' language (e.g., Serbian and Croatian, Upper and Lower • Sorbian). Very closely related languages (e.g., Czech and Slovak or Upper Sorbian) are very similar in morphology, syntax and lexics. Semantic ambiguities are rare. We suppose that no syntactic parser is needed in MT systems for • such languages. Closely related languages (e.g., Czech and Polish or Russian) are similar in morphology and lexics, although some semantic ambiguities occur. Syntactic constructions are not fully compatible, e.g., counterparts of analytic constructions are synthetic and vice versa or different lexems are used, cf. Czech byl jsem [I was] with Pol. byłem, Czech byl zničen [has been destroyed] (aux. verb být) with Polish został zniszczony (aux. verb zostać instead of być). Partial transfer is needed to …

Content truncated to 500 words