Automatic Construction of Machine Translation Knowledge Using Translation Literalness

Automatic Construction of Machine Translation Knowledge Using Translation Literalness Kenji Imamura, Eiichiro Sumita ATR Spoken Language Translation Research Laboratories Seika-cho, Soraku-gun, Kyoto, Japan {kenji.imamura,eiichiro.sumita}@atrcojp Yuji Matsumoto Nara Institute of Science and Technology Ikoma-shi, Nara, Japan matsu@is.aist-nara.acjp Abstract When machine translation (MT) knowl- edge is automatically constructed from bilingual corpora, redundant rules are acquired due to translation variety. These rules increase ambiguity or cause incorrect MT results. To overcome this problem, we constrain the sentences used for knowledge extraction to "the appropriate bilingual sentences for the MT." In this paper, we propose a method using translation literalness to select ap- propriate sentences or phrases. The translation correspondence rate (TCR) is defined as the literalness measure. Based on the TCR, two automatic con- struction methods are tested. One is to filter the corpus before rule acquisition. The other is to split the acquisition pro- cess into two phases, where a bilingual sentence is divided into literal parts and the other parts before different gener- alizations are applied. The effects are evaluated by the MT quality, and about 4.9% of MT results were improved by the latter method. 1 Introduction Along with the efforts made to accumulate bilin- gual corpora for many language pairs, quite a few machine translation (MT) systems that automati- cally construct their knowledge from corpora have been proposed (Brown et al., 1993; Menezes and Richardson, 2001; Imamura, 2002). However, if we use corpora without any restriction, redun- dant rules are acquired due to translation varieties. Such rules increase ambiguity and may cause in- appropriate MT results. Translation variety increases with corpus size. For instance, large corpora usually contain mul- tiple translations of the same source sentences. Moreover, peculiar translations that depend on context or situation proliferate in large corpora. Our targets are corpora that contain over one hun- dred thousand sentences. To reduce the influence of translation vari- ety, we attempt to control the bilingual sentences that are appropriate for machine translation (here called "controlled translation"). Among the mea- sures that can be used for controlled translation, we focus on translation literalness in this pa- per. By restricting bilingual sentences during MT knowledge construction, the MT quality will be improved. The remainder of this paper is organized as fol- lows. Section 2 describes the problems caused by translation varieties. Section 3 discusses the kinds of translations that are appropriate for MTs. Sec- tion 4 introduces the concept of translation literal- ness and how to measure it. Section 5 describes construction methods using literalness, and Sec- tion 6 evaluates the construction methods. 2 Problems Caused by Translation Variety First, we describe the problems inherent in bilin- gual corpora when we automatically construct MT knowledge. 2.1 Context/Situation-dependent Translation Some bilingual sentences in corpora depend on the context or situation, and these are not always cor- rect in different contexts. 155 For instance, the English determiner 'the' is not generally translated into Japanese. However, when a human translator cannot semantically identify the following noun, a determinant modifier such as `watashi-no (my)' or …

Content truncated to 500 words