Classifying Spanish se constructions: from bag of words to language models

Resumen

Spanish se constructions are a complex linguistic phenomenon that challenges Natural Language Processing (NLP) tasks such as part-of-speech or dependency relation tagging. Se is a high-frequency word that appears in nine different types of syntactic constructions and adds information of diverse nature depending on the context. Thus, to solve the problem Spanish se constructions poses in an efficient way, this study proposes a tagging system for se applied to a corpus composed of 2,140 sentences. This corpus is used in a classification experiment where 9 classifiers based on machine learning models and a dependency parser are tested. Results show that pre-trained language models based on transformers architecture reach the highest accuracy (0.83) and f-score (0.70) values.