Skip to content
Conference Open access

Gradient Descenders at DialectSentEval 2026: A Comparative Study of Pretrained Arabic Transformers for Multi-Dialect Sentiment Analysis

2026 · Proceedings of The Fourth Arabic Natural Language Processing Conference: Shared Tasks · 1 citation · 13 references

Abstract

This paper presents the Gradient Descenders system for DialectSentEval Subtask 1 on multi-dialect Arabic sentiment analysis. Because general-purpose models often miss dialectal nuances, we compare three pretrained encoders with distinct linguistic orientations: MAR-BERTv2, DarijaBERT, and AraBERT. We formulate the task as three-class classification, employing a parameter-efficient head-only adaptation strategy that freezes encoder parameters. Our results show MARBERTv2 outperforms the others, achieving a Macro F1 of 80.11% and Accuracy of 81.90%, securing fifth place over-all. These findings emphasize that backbone selection is critical; specifically, broad dialect-oriented pretraining yields the most transferable representations for multi-dialect generalization.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.