Cross-lingual Wikification Using Multilingual Embeddings

Proceedings of NAACL-HLT 2016, pages 589–598, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Cross-lingual Wikification Using Multilingual Embeddings Chen-Tse Tsai and Dan Roth University of Illinois at Urbana-Champaign 201 N. Goodwin, Urbana, Illinois, 61801 {ctsai12, danr}@illinois.edu Abstract Cross-lingual Wikification is the task of grounding mentions written in non-English documents to entries in the English Wikipedia. This task involves the problem of comparing textual clues across languages, which requires developing a notion of similarity between text snippets across languages. In this paper, we address this problem by jointly training multi- lingual embeddings for words and Wikipedia titles. The proposed method can be applied to all languages represented in Wikipedia, in- cluding those for which no machine trans- lation technology is available. We create a challenging dataset in 12 languages and show that our proposed approach outperforms var- ious baselines. Moreover, our model com- pares favorably with the best systems on the TAC KBP2015 Entity Linking task including those that relied on the availability of transla- tion from the target language to English. 1 Introduction Wikipedia has become an indispensable resource in knowledge acquisition and text understanding for both human beings and computers. The task of Wikification or Entity Linking aims at disambiguat- ing mentions (sub-strings) in text to the correspond- ing titles (entries) in Wikipedia or other Knowledge Bases, such as FreeBase. For English text, this problem has been studied extensively (Bunescu and Pasca, 2006; Cucerzan, 2007; Mihalcea and Csomai, 2007; Ratinov et al., 2011; Cheng and Roth, 2013). It also has been shown to be a valuable component of several natural language processing and information extraction tasks across different domains. Recently, there has also been interest in the cross- lingual setting of Wikification: given a mention from a document written in a foreign language, the goal is to find the corresponding title in the English Wikipedia. This task is driven partly by the fact that a lot of information around the world may be written in a foreign language for which there are limited lin- guistic resources and, specifically, no English trans- lation technology. Instead of translating the whole document to English, grounding the important entity mentions in the English Wikipedia may be a good solution that could better capture the key message of the text, especially if it can be reliably achieved with fewer resources than those needed to develop a translation system. This task is mainly driven by the Text Analysis Conference (TAC) Knowledge Base Population (KBP) Entity Linking Tracks (Ji et al., 2012; Ji et al., 2015; Ji et al., 2016), where the target languages are Spanish and Chinese. In this paper, we develop a general technique which can be applied to all languages in Wikipedia even when no machine translation technology is available for them. The challenges in Wikification are due both to ambiguity and variability in expressing entities and concepts: a given mention in text, e.g., Chicago, may refer to different titles in Wikipedia (Chicago Bulls, the City, Chicago Bears, the …

Content truncated to 500 words