Collocation Translation Acquisition Using Monolingual Corpora

Collocation Translation Acquisition Using Monolingual Corpora Yajuan LÜ Microsoft Research Asia 5F Sigma Center, No. 49 Zhichun Road, Haidian District, Beijing, China, 100080 t-yjlv@microsoft.com Ming ZHOU Microsoft Research Asia 5F Sigma Center, No. 49 Zhichun Road, Haidian District, Beijing, China, 100080 mingzhou@microsoft.com Abstract Collocation translation is important for machine translation and many other NLP tasks. Unlike previous methods using bilingual parallel corpora, this paper presents a new method for acquiring collocation translations by making use of monolingual corpora and linguistic knowledge. First, dependency triples are extracted from Chinese and English corpora with dependency parsers. Then, a dependency triple translation model is estimated using the EM algorithm based on a dependency correspondence assumption. The generated triple translation model is used to extract collocation translations from two monolingual corpora. Experiments show that our approach outperforms the existing monolingual corpus based methods in dependency triple translation and achieves promising results in collocation translation extraction. 1 Introduction A collocation is an arbitrary and recurrent word combination (Benson, 1990). Previous work in collocation acquisition varies in the kinds of collocations they detect. These range from two- word to multi-word, with or without syntactic structure (Smadja 1993; Lin, 1998; Pearce, 2001; Seretan et al. 2003). In this paper, a collocation refers to a recurrent word pair linked with a certain syntactic relation. For instance, <solve, verb-object, problem> is a collocation with a syntactic relation verb-object. Translation of collocations is difficult for non- native speakers. Many collocation translations are idiosyncratic in the sense that they are unpredictable by syntactic or semantic features. Consider Chinese to English translation. The translations of “解决” can be “solve” or “resolve”. The translations of “问题” can be “problem” or “issue”. However, translations of the collocation “解决 ~ 问题” as “solve~problem” or “resolve~ issue” is preferred over “solve~issue” or “resolve ~problem”. Automatically acquiring these collocation translations will be very useful for machine translation, cross language information retrieval, second language learning and many other NLP applications. (Smadja et al., 1996; Gao et al., 2002; Wu and Zhou, 2003). Some studies have been done for acquiring collocation translations using parallel corpora (Smadja et al, 1996; Kupiec, 1993; Echizen-ya et al., 2003). These works implicitly assume that a bilingual corpus on a large scale can be obtained easily. However, despite efforts in compiling parallel corpora, sufficient amounts of such corpora are still unavailable. Instead of heavily relying on bilingual corpora, this paper aims to solve the bottleneck in a different way: to mine bilingual knowledge from structured monolingual corpora, which can be more easily obtained in a large volume. Our method is based on the observation that despite the great differences between Chinese and English, the main dependency relations tend to have a strong direct correspondence (Zhou et al., 2001). Based on this assumption, a new translation model based on dependency triples is proposed. The translation probabilities are estimated from two monolingual corpora using the EM algorithm with the help of a bilingual translation dictionary. Experimental results show that the proposed triple translation model outperforms the …

Content truncated to 500 words