A benchmark dataset of narrative student essays with multi-competency grades for automatic essay scoring in Brazilian Portuguese

This paper describes the development of a new database comprising 1235 narrative essays written in Portuguese by 5th-grade students in Brazil. The corpus construction process involved three main steps: acquiring and transcribing photos of the essays, annotating them based on a real pre-defined correction rubric by experts considering four key writing competencies (formal language use, textual typology, thematic coherence, and textual cohesion), and resolving disagreements between the annotators. Two human experts manually evaluated each essay using a five-point scale (Level I: Complete lack of domain - Level V: Excellent mastery) aligned with the correction rubric. In cases of disagreement between the initial evaluators, a third expert facilitated the divergences resolution. To the best of our knowledge, this is the first publicly available dataset of elementary school essays in Brazilian Portuguese that features narrative writing samples with corresponding grades across multiple competencies commonly used in writing assessment. We believe this resource can contribute to developing automatic essay scoring systems tailored for evaluating narrative texts written in Brazilian Portuguese.