Maastricht University at AMIYA: Adapting LLMs for Dialectal Arabic using Fine-tuning and MBR Decoding

TL;DR

This paper enhances large language models for dialectal Arabic by fine-tuning with LoRA, adapter merging, and MBR decoding, significantly improving dialectal fidelity and translation accuracy.

Contribution

It introduces a novel combination of LoRA fine-tuning, adapter merging, and MBR decoding specifically for dialectal Arabic adaptation.

Findings

Merging adapters improves dialectal fidelity.

MBR decoding enhances translation accuracy.

Framework is compact and effective for dialectal Arabic generation.

Abstract

Large Language Models (LLMs) are becoming increasingly multilingual, supporting hundreds of languages, especially high resource ones. Unfortunately, Dialect variations are still underrepresented due to limited data and linguistic variation. In this work, we adapt a pre-trained LLM to improve dialectal performance. Specifically, we use Low Rank Adaptation (LoRA) fine-tuning on monolingual and English Dialect parallel data, adapter merging and dialect-aware MBR decoding to improve dialectal fidelity generation and translation. Experiments on Syrian, Moroccan, and Saudi Arabic show that merging and MBR improve dialectal fidelity while preserving semantic accuracy. This combination provides a compact and effective framework for robust dialectal Arabic generation.