Cross Lingual Mention and Entity Embeddings for Cross-Lingual Entity Disambiguation

Cross-lingual Entity Discovery and Linking (EDL) task involves discovering query mentions in crosslingual documents and linking them to their referent entities in an English Knowledge Base (KB). Traditional entity Linking models heavily rely on engineering manual and often language dependent features. Recently, deep learning based models have emerged as compelling solutions that alleviate the problem of feature engineering. In this paper we propose a deep learning based model in which the cross-lingual contextual information are encoded into mention and entity embeddings and then hard-wired into a linker that can optionally leverages them along with other lexical features in its disambiguation process. The experimental results show that the embeddings can efficiently enhance the performance of the linker while they are learned without relying on specific language features.