Bootstrapping Entity Translation on Weakly Comparable Corpora

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 631–640, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Bootstrapping Entity Translation on Weakly Comparable Corpora Taesung Lee and Seung-won Hwang Department of Computer Science and Engineering Pohang University of Science and Technology (POSTECH) Pohang, Republic of Korea {elca4u, swhwang}@postech.edu Abstract This paper studies the problem of mining named entity translations from compara- ble corpora with some “asymmetry”. Un- like the previous approaches relying on the “symmetry” found in parallel corpora, the proposed method is tolerant to asymme- try often found in comparable corpora, by distinguishing different semantics of rela- tions of entity pairs to selectively prop- agate seed entity translations on weakly comparable corpora. Our experimental results on English-Chinese corpora show that our selective propagation approach outperforms the previous approaches in named entity translation in terms of the mean reciprocal rank by up to 0.16 for or- ganization names, and 0.14 in a low com- parability case. 1 Introduction Identifying and understanding entities is a cru- cial step in understanding text. This task is more challenging in the presence of multilingual text, because translating named entities (NEs), such as persons, locations, or organizations, is a non-trivial task. Early research on NE trans- lation used phonetic similarities, for example, to mine the translation ‘Mandelson’→‘曼 德 尔 森’[ManDeErSen] with similar sounds. However, not all NE translations are based on translitera- tions, as shown in Table 1—Some translations, especially the names of most organizations, are based on semantic equivalences. Furthermore, names can be abbreviated in one or both lan- guages, e.g., the ‘World Trade Organization’ (世 界贸易组织) can be called the ‘WTO’ (世贸组 织). Another challenging example is that, a trans- lation can be arbitrary, e.g., ‘Jackie Chan’ → ‘成 龙’ [ChengLong]. There are many approaches English Chinese World Trade Organization 世界贸易组织 [ShiJieMaoYiZuZhi] WTO 世贸组织 [ShiMaoZuZhi] Jackie Chan 成龙 [ChengLong] Table 1: Examples of non-phonetic translations. that deal with some of these challenges (Lam et al., 2007; Yang et al., 2009), e.g., by combin- ing phonetic similarity and a dictionary. How- ever, arbitrary translations still cannot be handled by examining the NE pair itself. Corpus-based ap- proaches (Kupiec, 1993; Feng, 2004), by mining external signals from a large corpus, such as par- enthetical translation “成龙 (Jackie Chan)”, com- plement the problem of transliteration-based ap- proaches, but the coverage of this approach is lim- ited to popular entities with such evidence. The most effective known approach to NE translation has been a holistic framework (You et al., 2010; Kim et al., 2011; You et al., 2012) com- bining transliteration- and corpus-based methods. In these approaches, both 1) arbitrary translations and 2) lesser-known entities can be handled, by propagating the translation scores of known enti- ties to lesser-known entities if they co-occur fre- quently in both corpora. For example, a lesser- known entity Tom Watson can be translated if Mandelson and Tom Watson co-occur frequently in an English corpus, and their Chinese transla- tions also co-occur frequently in a …

Content truncated to 500 words