Embedding Based Multilingual Atlas of Semantic Fields

Semantic fields(domains) are an important construct in neuroscience, linguistics, psychology, and natural language processing. However, semantic field resources typically lack scalability and are not based on language usage, but on scientific and commercial taxonomies. The present project aims to create maps of semantic fields for multiple languages constructed from Word Embeddings. The clustering process is systematically described, and preliminary results for the Spanish language are presented, showing similarities and differences compared to current classifications. The present work opens up possibilities for a usage-based word classification of semantic fields and for the generation of language atlases that allow for multilingual comparison and improve the development of the aforementioned disciplines.