Automated metaphor detection in Brazilian Portuguese

Abstract

Automated metaphor detection plays a crucial role in various Natural Language Processing (NLP) applications, including sentiment analysis, machine translation, and text generation. Given the intricate nature of metaphorical language, its automatic processing poses a significant challenge, requiring models to go beyond literal meanings to capture underlying conceptual mappings. While significant research has been conducted on metaphor detection for high-resource languages like English, many other languages remain underexplored. This is particularly true for Brazilian Portuguese, where the scarcity of annotated corpora and limited availability of studies hinder advancements in the field. The present work addresses this gap by constructing an annotated corpus for metaphor detection at token-level in Brazilian Portuguese and using it to evaluate encoder-only and decoder-only approaches for the task. This work presents, to the best current of knowledge, the first resource of its kind for Brazilian Portuguese and establishes an initial benchmark for automated metaphor detection in the language.