NUS-IDS at AMIYA/VarDial 2026: Improving Arabic Dialectness in LLMs with Reinforcement Learning
Year: 2026
Type: conference-paper
Abstract: In this paper, we describe models developed by our team, NUS-IDS, for the Closed Data track at the Arabic Modeling In Your Accent (AMIYA) shared task at VarDial 2026.The core idea behind our solution involves data augmentation enabled by a dialect classifier trained on AMIYA data.We effectively combine various translation, summarization, and question answering prompts with the training data to form dialectal prompts for use with state-ofthe-art Large Language Models (LLMs).Next, dialect predictions on outputs from these LLMs are used to compile preference data for Reinforcement Learning (RL).We report model performance on dialectal Arabic from Egypt, Morocco, Palestine, Saudi Arabia, and Syria using FLORES+, a benchmark dataset for multilingual machine translation.Our experiments illustrate that though our RL models show significant performance gains on dialectness scores, they underperform on translation metrics compared to base LLMs.
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7140085810
FWCI:
Cites:
Cited by:
Related to: