Previous work on Egyptian Arabic in NLP has focused largely on the prestigious Cairene Egyptian Arabic (CEA) dialect, resulting in a lack of representation for the less prestigious Sa'idi Egyptian Ara...
The study investigates whether Modern Standard Arabic (MSA) dominates internal representations in large language models (LLMs) when generating Arabic, even if prompted with dialectal forms. By extending a language-dominance framework to 26 Arabic varieties, the authors find no evidence that MSA serves as a dominant internal representation; instead, dialect representations are densely overlapping, with mutual information dropping sharply compared to more distinct languages. The strongest separability effects vary across layers and architectures, indicating that generation bias toward MSA does not necessarily reflect internal representational dominance.
arXiv:2609.22796v1 Announce Type: new Abstract: Dialectal Arabic machine translation (MT) remains challenging despite recent progress in Arabic language technologies, particularly because effective t...
This study evaluates large language models (LLMs) for English‑to‑dialectal Arabic machine translation using the MADAR dataset across 16 Arabic dialects, covering both translation directions. It compares specialized Arabic models (Jais, Nile), multilingual models (Gemma, Command‑R, Mistral, Aya), and commercial APIs (GPT‑4.1) with multiple metrics—BLEU, CHRF, COMET, GEMBA—and a small manual assessment. The authors also examine LLMs’ ability to classify dialects and discuss challenges in automatic MT evaluation for Arabic dialects, noting potential training data contamination.
The paper investigates how Arabic dialects are represented in large language models and whether they can be steered at inference time. By analyzing neuron-level sparsity and vector steering, the authors find that only a small fraction of neurons encode dialect-specific features, while distributed activation directions are more effective for steering. Vector steering can induce dialectal output from both dialectal and MSA prompts, whereas neuron steering works only when the prompt is already dialectal.
arXiv:2611.06517v1 Announce Type: new Abstract: The massive permeability of intelligent systems in every sphere of human activities have greatly enhanced clarity, ease of work, and increased general...
EDRAC is the first large‑scale benchmark for dialectal Arabic machine reading comprehension and generative question answering, covering five major dialects—Egyptian, Moroccan, Emirati, Syrian, and Saudi. It contains 499 passages from naturally spoken interactions and 4,977 QA pairs produced via a human–LLM collaborative pipeline. The benchmark evaluates Arabic‑centric and multilingual large language models, revealing gaps between semantic answer quality and dialectal fidelity and underscoring limitations of current evaluation metrics for dialectal Arabic generation.
The paper introduces a rubric-based benchmark to evaluate Saudi Arabic dialect and cultural competence in large language models. It comprises 31 expert-authored prompts covering idiomatic, pragmatic, lexical, and culturally embedded aspects, each paired with an expert-established ground truth. Four state-of-the-art models were scored, revealing that none exceeded 55% accuracy and that ambiguous framing was the most common error type.
arXiv:2605. 02608v2 Announce Type: replace-cross Abstract: Transformer-based models achieve state-of-the-art dependency parsing for high-resource languages, yet their advantage over simpler architectures in low-resource settings remains poorly understood.
arXiv:2606. 16753v1 Announce Type: cross Abstract: As Large Language Models (LLMs) become embedded in everyday communication, capturing regional linguistic variation is essential for reliable and equitable language use.
The paper introduces MADARTB, the first large-scale parallel Arabic dependency treebank covering five city dialects—Beirut, Cairo, Doha, Rabat, and Tunis—alongside Modern Standard Arabic. It contains 5,000 sentences annotated with the CATiB dependency scheme and shows that while overall construction profiles are similar across varieties, local syntactic structures often diverge, indicating significant structural variation. A supervised parser achieves 87.5% LAS, outperforming the best LLM setup at 85.7% LAS across all six varieties, and the authors release the annotations and models publicly.
The study investigates why language models exhibit systematic performance gaps across English dialects, a phenomenon termed the "dialect tax." Using parallel dialect corpora that preserve meaning while altering surface form, the authors confirm that models treat Standard American English and dialectal texts as semantically equivalent, yet find representational disparities that persist through tokenization, pre‑training, post‑training, and inference. Even a character‑level tokenizer does not eliminate input/output asymmetries or accuracy gaps, and dialect pairs produce more divergent gradient updates than unrelated Standard texts, indicating that dialectal content is harder for models to learn.