Mining Name Translations from Entity Graph Mapping

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 430–439, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Mining Name Translations from Entity Graph Mapping∗ Gae-won You† Seung-won Hwang† Young-In Song‡ Long Jiang‡ Zaiqing Nie‡ †Pohang University of Science and Technology, Pohang, Republic of Korea {gwyou,swhwang}@postech.ac.kr ‡Microsoft Research Asia, Beijing, China {yosong,longj,znie}@microsoft.com Abstract This paper studies the problem of mining en- tity translation, specifically, mining English and Chinese name pairs. Existing efforts can be categorized into (a) a transliteration- based approach leveraging phonetic similar- ity and (b) a corpus-based approach exploiting bilingual co-occurrences, each of which suf- fers from inaccuracy and scarcity respectively. In clear contrast, we use unleveraged re- sources of monolingual entity co-occurrences, crawled from entity search engines, repre- sented as two entity-relationship graphs ex- tracted from two language corpora respec- tively. Our problem is then abstracted as find- ing correct mappings across two graphs. To achieve this goal, we propose a holistic ap- proach, of exploiting both transliteration sim- ilarity and monolingual co-occurrences. This approach, building upon monolingual corpora, complements existing corpus-based work, re- quiring scarce resources of parallel or compa- rable corpus, while significantly boosting the accuracy of transliteration-based work. We validate our proposed system using real-life datasets. 1 Introduction Entity translation aims at mapping the entity names (e.g., people, locations, and organizations) in source language into their corresponding names in target language. While high quality entity translation is es- sential in cross-lingual information access and trans- ∗This work was done when the first two authors visited Mi- crosoft Research Asia. lation, it is non-trivial to achieve, due to the chal- lenge that entity translation, though typically bear- ing pronunciation similarity, can also be arbitrary, e.g., Jackie Chan and � � (pronounced Cheng Long). Existing efforts to address these challenges can be categorized into transliteration- and corpus- based approaches. Transliteration-based approaches (Wan and Verspoor, 1998; Knight and Graehl, 1998) identify translations based on pronunciation similar- ity, while corpus-based approaches mine bilingual co-occurrences of translation pairs obtained from parallel (Kupiec, 1993; Feng et al., 2004) or compa- rable (Fung and Yee, 1998) corpora, or alternatively mined from bilingual sentences (Lin et al., 2008; Jiang et al., 2009). These two approaches have com- plementary strength– transliteration-based similar- ity can be computed for any name pair but cannot mine translations of little (or none) phonetic simi- larity. Corpus-based similarity can support arbitrary translations, but require highly scarce resources of bilingual co-occurrences, obtained from parallel or comparable bilingual corpora. In this paper, we propose a holistic approach, leveraging both transliteration- and corpus-based similarity. Our key contribution is to replace the use of scarce resources of bilingual co-occurrences with the use of untapped and significantly larger resources of monolingual co-occurrences for trans- lation. In particular, we extract monolingual co- occurrences of entities from English and Chinese Web corpora, which are readily available from en- tity search engines such as PeopleEntityCube1, de- ployed by Microsoft Research Asia. Such engine 1http://people.entitycube.com 430 automatically extracts people names …

Content truncated to 500 words