Machine understanding of text is an extremely challenging task for intelligent systems; machines need to understand the meaning behind words and reason about the text and existing knowledge. A highly effective method to obtain this understanding is connecting unstructured text to semi-structured information in the knowledge graphs (e.g., Wikipedia and Wikidata). In this process, referred to as Entity Linking, entities such as people and locations are identified and mapped to their corresponding entries in a knowledge graph. Entity linking is a language-dependent, compute, and data hungry process. How can we make this technology usable for multilingual, formal, and informal texts, requiring only limited computational power?
In this project, we introduce REL 2.0, a publicly available entity linking toolkit that can operate on texts in a variety of languages and forms (e.g., long documents, queries, and conversations), in a reasonable time and using commonly available hardware.
