Translation knowledge recycling for related languages

Translation Knowledge Recycling for Related Languages Michael Paul ATR Spoken Language Translation Research Laboratories 2-2-2 Hikaridai, Seika-cho, Soraku-gun, 619-0288 Kyoto Japan paul@slt.atr.co.jp Abstract An increasing interest in multi-lingual translation systems demands a reconsideration of the development costs of machine translation engines for language pairs. This paper proposes an approach that reuses the existing translation knowledge resources of high-quality translation engines for translation into different, but related languages. The lexical information of the target representation is utilized to generate the corresponding translation in the related language by using a transfer dictionary for the mapping of words and a set of heuristic rules for the mapping of structural information. Experiments using a Japanese-English translation engine for the generation of German translations show a minor decrease of up to 5% in the acceptability of the German output compared with the English translation of unseen Japanese input. Keywords reuse of knowledge resources, multi-lingual extensions, related languages 1 Introduction One of the biggest problems for the development of high- quality, multi-lingual translation engines is the high cost of adapting the underlying translation algorithm to mul- tiple language pairs. In particular, the lack of resources (dictionaries, bilingual corpora, etc.) for "uncommon" language pairs forms a bottleneck for multi-lingual extensions. The basic idea of this paper, as described in Section 2, is to devote efforts to the development of translation engines between the main linguistically different languages and to reuse the translation knowledge of these systems for translation into languages closely related to the target language. These languages must have similar grammatical characteristics so that the linguistic infor- mation contained in the target representation, e.g. a parse tree, can be mapped to a corresponding representation for the related language and so that this information can be used to generate the translation output. Our approach does not depend on any specific language or translation engine, but simply requires an internal target representation containing structural and word information. Section 3 describes the translation engine and the know- ledge resources used for the evaluation of our approach. The results of generating German output based on the translation of Japanese spoken-language utterances into English are summarized in Section 4. In this approach, the English parse tree is mapped to a corresponding German one by substituting word phrases according to an English-to-German transfer dictionary and applying heuristic rules defining grammatical equivalen- ces in both languages. Some comments on the feasibility of our approach and future perspectives are given in Section 5. 2 Translation Knowledge Recycling Our aim is to find an inexpensive way to provide multi- lingual extensions of existing translation engines. A simple way to achieve this goal is the concatenation of the respective engines by using a text-based interface as illustrated in Figure 1. In this system, the source language (SL) is translated into an intermediate language (IL) by the first translation engine (TE1), and this translated text becomes the input of the second engine (TE2) that generates the output in the target language (TL). Figure 1: Text-to-text interface The …

Content truncated to 500 words