Metaphoricity detection in adjective-noun pairs

Resumen

In this paper we propose a neural network approach to detect the metaphoricity of Adjective-Noun pairs using pre-trained word embeddings and word similarity using dot product. We found that metaphorical word pairs tend to have a lower dot product score while literal pairs a higher score. On this basis, we compared seven optimizers and two activation functions, from which the best performing pairs obtained an accuracy score of 97.69% and 97.74%, which represents an improvement of 6% over other current approaches.