AraDiCE: Benchmarks for Dialectal and Cultural Capabilities in LLMs

TL;DR

This paper introduces AraDiCE, a comprehensive benchmark with synthetic dialect datasets and cultural evaluation for Arabic in LLMs, highlighting performance gaps and the need for tailored models to handle dialectal and cultural nuances.

Contribution

It presents the first dialectal and cultural benchmark for Arabic LLMs, including synthetic datasets, evaluation metrics, and analysis of model performance across dialects and regions.

Findings

Arabic-specific models outperform multilingual models on dialect tasks

Significant challenges remain in dialect identification and translation

Tailored training improves dialect and cultural understanding in LLMs

Abstract

Arabic, with its rich diversity of dialects, remains significantly underrepresented in Large Language Models, particularly in dialectal variations. We address this gap by introducing seven synthetic datasets in dialects alongside Modern Standard Arabic (MSA), created using Machine Translation (MT) combined with human post-editing. We present AraDiCE, a benchmark for Arabic Dialect and Cultural Evaluation. We evaluate LLMs on dialect comprehension and generation, focusing specifically on low-resource Arabic dialects. Additionally, we introduce the first-ever fine-grained benchmark designed to evaluate cultural awareness across the Gulf, Egypt, and Levant regions, providing a novel dimension to LLM evaluation. Our findings demonstrate that while Arabic-specific models like Jais and AceGPT outperform multilingual models on dialectal tasks, significant challenges persist in dialect identification, generation, and translation. This work contributes $\approx$45K post-edited samples, a cultural benchmark, and highlights the importance of tailored training to improve LLM performance in capturing the nuances of diverse Arabic dialects and cultural contexts. We have released the dialectal translation models and benchmarks developed in this study (https://huggingface.co/datasets/QCRI/AraDiCE).