Entity Disambiguation to Wikipedia for Languages with Different Corpora Volumes

This paper is devoted to Entity Disambiguation to Wikipedia for languages with different corpora volumes. Disambiguation is one of the most crucial part of successful communication. However, in the field of natural language understanding (especially considering semantic interpretation or disambiguation) there is ample room for improvement. At the same time, Entity Disambiguation is able nowadays to drammatically increase the results on related tasks, including but not limited to Named Entity Classification, Coreference Resolution and Relation Extraction regardless of language and corpora sizes. Unfortuntely, Entity Disambiguation is performed nowadays for high-resource languages mostly. Therefore, the aim of the study is to develop a language-independent neural network approach to Entity Disambiguation task. We attain this by linking entity mentions in the text with the appropriate entity in the knowledge-base compiled from Wikipedia. We also compare different approaches for learning distributed representations for tokens. Additionally, this study proves the importance of enriching joint embeddings with information about knowledge base structure.