LINKING MODAL AND AMODAL REPRESENTATIONS THROUGH LANGUAGE COMPUTATIONAL MODELS

Abstract

"Language computational models such as Latent Semantic Analysis (LSA) has been criticized for not having direct contact with the real world. However, recent findings have shown the ability of the LSA to capture embodied features such as words’ emotional content. In the present study we tested whether LSA can predict the emotions contained in short written texts such as tweets. It was found that a multiple logistic regression model receiving as input LSA information classified correctly 73,9% of the tweets analyzed according to the emotional content. These results provide additional evidence underlying the representative power of abstract symbols and showing the link between modal representations (emotional) and amodal representations (abstract symbols) through the LSA."