Clustering emotions in Portuguese

Abstract

In this paper we present some exploratory studies of emotion words based on large annotated corpora of Portuguese. Those corpora were automatically annotated with emotionality, and each emotion word was assigned one or more groups out of 26 emotion groups. Our goal is to evaluate those groups by applying different statistical approaches to our material, namely based on (a) co-occurrence in a sentence as a sign of closeness of meaning, and (b) word embeddings. After looking at the full material, we turn our attention to two specific emotion groups: Amor (‘love’) and Desespero (‘despair’), investigating whether clustering with those underlying techniques can help improve the shape, or redesign, particular emotion groups. In the paper we suggest some novel forms of measuring semantic coherence on word embedding models. Since computational research on emotion words in Portuguese is still rare, our methods and resources will lay the ground for future investigations.

Keywords

• emotions
• corpus linguistics
• word embeddings
• empirical studies
• word vector space models