Across the Atlantic: Distinguishing Between European and Brazilian Portuguese Dialects

Abstract

Dialect Identification is the task of determining the regional or social variety of a spoken or written language. While specific languages have received considerable attention in this regard, others, such as Portuguese, remain largely unexplored. Furthermore, previous works on the Portuguese language are often outdated in the rapidly evolving landscape of NLP, and many suffer from methodological flaws. We revisit the task of differentiating between European and Brazilian variants of Portuguese, addressing and rectifying the mistakes found in prior research. For that, we carefully select a parallel corpus and explore both feature-based traditional classifiers and state-of-the-art neural approaches. Our findings1 demonstrate that whereas Transformer-based models provide solutions that are robust to out-of-distribution data, traditional NLP techniques are still competitive in this task. (c) 2024 The Association for Computational Linguistics.