A Metaphorical Text Classifier to Compare the Use of RoBERTa-Large, RoBERTa-Base and BERT-Base Uncased

This work presents a literal and metaphorical language classifier for the Trofi corpus (Gao G. et al. 2018), through LSTM cells, comparing the results for the use of three pretrained language models RoBERTa-large, RoBERTa-base and BERT-base uncased. Through this article, it is proposed to address three fundamental points of the study of metaphorical language: the different tools for its vectorial representation, the use of LSTM cells to work metaphorical language and the impotence of the central task presented, its classification. Finally the results are compared against the state of the art and that work presents some observations as a conclusion.