English-Chinese Knowledge Base Translation with Neural Network

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2935–2944, Osaka, Japan, December 11-17 2016. English-Chinese Knowledge Base Translation with Neural Network Xiaocheng Feng, Duyu Tang, Bing Qin, Ting Liu Harbin Institute of Technology, Harbin, China {xcfeng, dytang, qinb, tliu}@ir.hit.edu.cn Abstract Knowledge base (KB) such as Freebase plays an important role for many natural language pro- cessing tasks. English knowledge base is obviously larger and of higher quality than low re- source language like Chinese. To expand Chinese KB by leveraging English KB resources, an effective way is to translate English KB (source) into Chinese (target). In this direction, two major challenges are to model triple semantics and to build a robust KB translator. We address these challenges by presenting a neural network approach, which learns continuous triple repre- sentation with a gated neural network. Accordingly, source triples and target triples are mapped in the same semantic vector space. We build a new dataset for English-Chinese KB translation from Freebase, and compare with several baselines on it. Experimental results show that the proposed method improves translation accuracy compared with baseline methods. We show that adaptive composition model improves standard solution such as neural tensor network in terms of translation accuracy. 1 Introduction Knowledge base (KB) like Freebase1 and Yago2 has attracted a lot of attention in both research and industry communities. Knowledge Base contains massive triples (entries), each of which is a fact con- sisting of two arguments and one predicate, such as (Una White, profession, Nurse). A large, high-quality KB is valuable and can be applied to many natural language processing and information retrieval tasks (Graupmann et al., 2005; Hotho et al., 2006; Ferr´andez et al., 2009; Bouma et al., 2009; Shi et al., 2016; Feng et al., 2016). Let us take Freebase as an example, English KB contains 2.7 billion entries and the accuracy is higher than 80%. This is obviously larger and better than a KB with less amount of entries such as Chinese. < Una_White , profession , Nurse > English Triple 尤纳·白色(one color) 尤纳·蛋白(albumen) 尤纳·怀特(the name) 职业 护士 (a health worker) 保姆(a baby-sitter) Chinese Translation Candidates Translator Figure 1: An example of translation ambiguity in English-Chinese KB translation. A straightforward way to enrich Chinese KB is to directly translate English KB (source) to Chinese (target) based on the surface texts of a triple with existing machine translation system. However, we find that they suffer from the problem of ambiguity. An example is given in Figure 1. The argument “nurse” has two translation candidates namely “护士” (a person taking care of sick people in hospital) and “保 姆” (baby sitter). “Una White” have three translation candidates, so that there are totally six ambiguous 1http://en.wikipedia.org/wiki/Freebase 2http://en.wikipedia.org/wiki/YAGO (database) 2935 Knowledge Base Word Vector Space Compositionality Triple Vector Space Relation : Profession Chinese candidates Entity 1 Entity 2 Una_White Nurse 尤纳·白色 (one color) 尤纳·蛋白 (albumen) 尤纳 ·怀特 (the name) 职业 护士 (a healthy worker) 保姆 (a baby-sitter) English Space Chinese Space Nurse Una_White 保姆 …

Content truncated to 500 words