LAR-WordNet: A Machine-Translated, Pan-Hispanic and Regional WordNet for Spanish.

WordNet is one of the most used resources in Natural Language Processing (NLP). However, the only WordNet available for Spanish is mainly representative of Spain and its size is approximately 50 % compared to Princeton's WordNet in English. To address these issues, we automatically translate the Princeton version using lemmas and sentences from all the available corpora annotated with WordNet senses (LAS-WordNet). In addition, we enrich the translated version using lexicons that contain Pan-Hispanic regionalisms extracted from Twitter (LAR-WordNet). The proposed resources were evaluated in the task of Semantic Textual Similarity in Spanish and cross-lingual between Spanish and English. The results showed that LAS-WordNet significantly outperformed the current Spanish WordNet and that the regionalisms added to LAR-WordNet do not hinder its performance. Although the proposed resources are noisier than the current WordNet in Spanish, their size and representativeness make them suitable for many NLP applications.