We present ASCAT (Arabic Scientific Corpus for Advanced Translation), a high-quality English-Arabic parallel benchmark corpus designed for scientific translation evaluation constructed through a systematic multi-engine machine translation and expert post-editing pipeline. Unlike existing Arabic-English corpora that rely on short sentences or single-domain text, ASCAT targets full scientific abstracts averaging 125.3 words (English) and 111.78 words (Arabic), drawn from five scientific domains: physics, mathematics, computer science, quantum mechanics, and artificial intelligence. Each abstract was translated using three complementary architectures generative AI (Gemini), transformer-based models (Hugging Face quickmt-en-ar), and commercial MT APIs (Google Translate, DeepL) and subsequently post-edited by domain experts at the lexical, syntactic, and semantic levels. The resulting corpus contains 67,293 English tokens and 60,026 Arabic tokens, with an Arabic vocabulary of 17,604 unique words reflecting the morphological richness of the language. We benchmark three state-of-the-art LLMs on the corpus GPT-4o-mini (BLEU: 37.07), Gemini-3.0-Flash-Preview (BLEU: 30.44), and Qwen3-235B-A22B (BLEU: 23.68) demonstrating its discriminative power as an evaluation benchmark. ASCAT addresses a critical gap in scientific MT resources for Arabic and is designed to support rigorous evaluation of scientific translation quality and training of domain-specific translation models.

Arabic information retrieval increasingly relies on multi-stage pipelines in which a fast first-stage retriever produces candidate passages and a neural reranker refines relevance. While transformer cross-encoders deliver strong effectiveness through joint query–passage encoding, multilingual rerankers achieve competitive performance on Arabic benchmarks. However, systematic analysis of calibration, robustness, and deployment behavior in Arabic-specific settings remains limited. We present GATE-Reranker, a compact Arabic cross-encoder initialized from an Arabic semantic embedding backbone and fine-tuned on large-scale mMARCO-style Arabic triplets. The model scores each query–passage pair via full self-attention and a lightweight regression head, enabling plug-and-play second-stage reranking for Arabic search and RAG systems. We evaluate on three Arabic benchmarks covering binary relevance discrimination, controlled multi-negative reranking, and large-scale mMARCO evaluation. While remaining competitive with strong multilingual rerankers in ranking effectiveness, GATE-Reranker demonstrates significantly improved calibration and discriminative behavior. These properties translate into more reliable downstream performance in retrieval and RAG pipelines, while maintaining low GPU memory and latency on a Tesla T4.

This paper presents the Fine-Tashkeel system for Task 2 of the KSAA-2026 Shared Task on Automatic Diacritization of Speech Dictation. Diacritization of speech-derived Arabic text poses challenges due to dialectal variation, morphological ambiguity, and the absence of acoustic cues in text-only pipelines. Our approach treats diacritization as a character-level sequence-to-sequence task, mapping undiacritized text directly to its fully diacritized form. We evaluate 18 models spanning text-only, ASR-augmented, and fine-tuned configurations, finding that text-only Seq2Seq approaches outperform off-the-shelf multimodal models—a gap we attribute to task mismatch in generic ASR systems rather than an inherent audio limitation. Our best submission, using zero-shot inference without task-specific training, achieved a Diacritic Error Rate (DER) of 10.56%, Word Error Rate (WER) of 34.47%, and Sentence Error Rate (SER) of 79.88%, ranking 5th out of 7 teams. Per-nationality error analysis reveals significant dialectal variation (Egyptian 3.70% vs. Algerian 13.73% DER), and diagnostic analysis confirms that case endings and vowel ambiguity are the primary bottlenecks. Code and evaluation scripts are publicly available.

Dialectal Arabic remains significantly underrepresented in parallel resources for direct machine translation with English, particularly for regional varieties such as Saudi Najdi Arabic. In this work, we introduce NAJD-MT, a systematically constructed Saudi Najdi-English parallel corpus designed for training bidirectional neural machine translation models. Starting from the Saudi Arabic Dialectal Annotated (SADA) dataset, we generate English translations using GPT-4.1 and subsequently apply cross-lingual embedding-based cosine similarity filtering to improve semantic alignment and reduce translation noise. We analyze the impact of varying semantic similarity thresholds on corpus size and downstream translation performance. Using the constructed datasets, we train and evaluate multiple Transformer-based models, including NLLB-200, OPUS-MT, mBART, and AraT5v2, in both Najdi→English and English→Najdi directions. Experimental results demonstrate that stricter semantic filtering (cosine ≥ 0.7) consistently improves translation quality despite reducing dataset size, highlighting that data purity plays a critical role in dialectal machine translation training. Our findings provide a reproducible framework for constructing high-fidelity dialect English parallel corpora and emphasize the importance of semantic alignment filtering in low-resource dialectal settings.

This paper presents Ketaba-OCR-LoRA, a system developed for the NakbaNLP 2026 Shared Task on Arabic Manuscript Understanding (Subtask 2), which targets the transcription of the historically significant Omar Al-Saleh Memoir Collection written in Ruq’ah and Naskh scripts. We propose a parameter-efficient adaptation of a publicly available pretrained Arabic-English Handwritten Text Recognition (HRT) model, originally trained on handwritten corpora including the Muharaf dataset. Instead of adapting general Vision-Language Models from scratch, we fine-tune the HRT backbone using Low-Rank Adaptation (LoRA) and 4-bit quantization (QLoRA), reducing memory requirements from 40GB to approximately 8GB. Our final submission combines multiple model variants through a novel Linear+Boost weighted ensemble strategy. Our approach achieves a CER of 0.0819 and WER of 0.2588 on the blind test set (per-line evaluation), ranking 1st on per-line evaluation; on the official corpus-wide leaderboard, we rank 3rd (CER 0.0938, WER 0.2996). This work demonstrates that specialized pretrained HRT models substantially outperform general-purpose Vision-Language Models for Arabic manuscript transcription, and that parameter-efficient fine-tuning provides a practical and reproducible approach for low-resource cultural heritage digitization.

Semantic textual similarity (STS) is central to retrieval, clustering, and semantic understanding, yet Arabic remains underserved by high-quality sentence-embedding models and evaluation resources. We introduce GATE, a compact family of Arabic sentence embedding models and a controlled study of Arabic-specific backbones, contrastive Matryoshka Representation Learning, dimensional truncation, and hybrid STS-NLI supervision. We evaluate the resulting models across Arabic semantic similarity, NLI, pair classification, sentence classification, clustering, retrieval, reranking, and dialectal semantic relatedness, against strong multilingual baselines including BGE-M3, Arctic-Embed-v2, and Qwen3-Embedding. Despite their 135M-parameter scale, GATE-family models are competitive with substantially larger multilingual encoders and achieve strong results on several native-Arabic semantic tasks. A three-annotator audit further finds high Arabic surface adequacy and 84.2% majority relation preservation for the translated supervision. The released checkpoints, data resources, and experimental artifacts support reproducible Arabic embedding research.

Modern visual document retrieval (VDR) models remain strongly English-centric, leaving Arabic substantially under-served in this rapidly advancing field. To address this gap, we introduce DEAM, the first open Arabic visual document retrieval dataset explicitly designed for fine-tuning and evaluating multimodal retrievers on Arabic. DEAM is built through a human-in-the-loop construction pipeline with annotators community annotators, combining Arabic document rendering, query creation, validation, and retrieval-oriented hard-negative mining. The dataset comprises18,275 rendered Arabic documents spanning nine source domains and 82,727 contrastive retrieval triplets, each including five mined hard negatives and one easy negative for contrastive fine-tuning. Zero-shot evaluation of 27 retrievers shows that multilingual text retrievers outperform visual retrievers on Arabic by 13.6 nDCG@10, except on layout-rich news pages. DEAM establishes a public foundation for Arabic visual document retrieval.https://huggingface.co/collections/NAMAA-Space/deam-for-arabic-multimodal-retrieval

Stance detection identifies an author’s position toward a specified target, yet Arabic social-media text remains challenging because stance is often implicit and shaped by dialect, sarcasm, informal expression, and culturally specific references. StanceEval-2026 evaluates this problem in two transfer settings: a related target within a seen thematic domain (Track 1) and entirely unseen topics (Track 2). We compare four modeling strategies: single-task Arabic encoder fine-tuning, multi-task learning with sentiment and sarcasm supervision, 4-bit QLoRA adaptation of the ALLaM-7B model, and Retrieval-Augmented In-Context Learning (RAG-ICL) with frozen large language models (LLMs). Our strongest single-backend configuration is a data-centric RAG-ICL pipeline combining label-balanced retrieval, hashtag and negation adaptation, text cleaning, similarity filtering, and target descriptions. A majority-vote ensemble of this configuration across three LLM backends obtains official F_avg2 scores of 0.8719 on Track 1 and 0.9016 on Track 2, ranking 3rd and 7th, respectively. Under the shared-task test conditions, the retrieval-based systems substantially outperform the supervised systems evaluated here, particularly on Track 2.

Robust Arabic speech processing must accommodate dialectal variation, multilingual code-switching, and distribution shift. We present NAMAA Community’s systems for three NADI 2026 speech subtasks: Mixed-Dialect Arabic ASR, Code-Switched Arabic ASR, and 20-way Spoken Arabic Dialect Identification. For Mixed-Dialect ASR, an Arabic-specific Conformer–Transformer achieves 0.45 word error rate (WER) on the blind set, outperforming a development-selected normalization and consensus pipeline at 0.49. For Code-Switched ASR, two-stage LoRA adaptation of the Tunisian Whisper-derived FarukSTT checkpoint, with multilingual tokenization and unconstrained decoding, yields 20.38% WER and 8.06% character error rate (CER) on validation and 16.98% blind-test WER. For dialect identification, a four-model ensemble reaches 0.950 development accuracy but 0.31 on the out-of-domain blind set. Label-free class-spread and inter-model agreement diagnostics identify an outlying component; removing it improves blind accuracy to 0.36, and a refined three-model submission reaches 0.38. Across Subtasks 1.2 and 2, development-selected calibration did not reliably transfer to blind data, although the available comparisons do not isolate the causal effect of individual components.

Multimodal hate speech detection requires interpreting how an image and its accompanying text jointly construct a target, stance, or derogatory implication. In Arabic memes, this relation is often shaped by locally recognizable public figures, political references, visual analogies, and indirect humor, making either modality insufficient on its own. We tackle this problem in Track A of ArGuard 2026, covering binary hateful meme detection (A1) and fine-grained multi-label classification (A2). Our experiments compare jointly trained dual encoders, parameter-efficient vision–language model (VLM) adaptation, and training-free prompting. We fine-tune Qwen3-VL, Llama 3.2 Vision, and Gemma 4 with QLoRA using task-specific binary and multi-label objectives. We also build CLIP–MARBERTv2 and SigLIP–AraBERT-Twitter cross-attention systems and evaluate frozen VLMs with zero-shot prompting and in-context learning. Qwen3-VL achieves our highest A1 score (0.7610), while SigLIP–AraBERT-Twitter leads A2 (0.3851), where label sparsity and multi-label prediction remain challenging.

Hierarchical, definition-guided genre classification places a short Arabic text under a broad genre and, within it, a fine-grained specific genre, using only a natural-language definition of each. We present NAMAA’s submission to AraGenre 2026. Fine-tuned bi-encoder classifiers reached 0.93 hierarchical macro F1 on development via data augmentation and cross-lingual pretraining, but collapsed to as low as 0.22 on the released test set, having specialized to a hand-engineered definition style that missed the test set’s larger, 74-genre taxonomy. We root-caused this gap and switched to zero-shot LLM classification, matching the best fine-tuned ensemble’s test score (0.6899 hierarchical F1) with no training; layering self-consistency sampling and disagreement resolution raised this further to 0.7013, ranking 3rd of 18 teams. Specific Macro F1 (0.5114) lags well behind Broad Macro F1 (0.8913), showing that most remaining errors concentrate in fine-grained genre choices rather than broad-category mistakes. Zero-shot LLM prompting generalizes across this definition and taxonomy shift where fine-tuned, definition-conditioned encoders do not. We release our code and system resources to support reproducibility and further research.[<https://github.com/NAMAA-ORG/Aragenre-Definition-Guided-Genre>].

Fine-grained Arabic readability assessment is challenging due to the language’s rich morphology and the ordered nature of readability labels. We investigate this problem in the sentence-level Strict Track of the BAREC Shared Task 2026, which uses a 19-level readability scale. We evaluate six controlled configurations spanning engineered linguistic and morphological features, TF-IDF/SVD lexical representations, contextual transformer encoders, multi-encoder fusion, output-level ensembling, and ordinal-aware training, followed by a final refinement using a single EmbeddingGemma encoder. Contextual transformer representations provide the largest improvement over classical feature-based baselines, while multi-encoder fusion and voting yield smaller additional gains. In contrast, word-repetition and adjacent-word-swap augmentation degrades performance, indicating that these perturbations do not reliably preserve readability labels. Our strongest configuration combines EmbeddingGemma with Gaussian soft labels, KL-divergence training, expected-value decoding, class weighting, and best-checkpoint selection, achieving 84.0 QWK on the test set and 82.4 QWK on the official blind test set. Overall, the results suggest that explicitly accounting for the ordinal structure of readability labels is a promising alternative to simply increasing architectural or ensemble complexity. We further summarize the resulting design insights into a practical framework for Arabic ordinal readability modeling.

Function calling connects language models to external tools, but Arabic structured tool use remains insufficiently evaluated. We present , a shared task at ArabicNLP 2026 built around 12,220 Arabic requests spanning five dialect groups and eight service domains. Each request has four candidate schemas drawn from a registry of 27 tools, including seven tools that never occur as gold calls; 223 requests require abstention. Systems decide whether to call, select a function, and extract its arguments. Two ranked tracks evaluate core calling and reasoning-trace compliance, with a third diagnostic examining dialect robustness. Twenty-two teams submitted to the blind phase, and 21 produced valid predictions on 1,125 test requests. We analyze 73 submissions and 13 system descriptions. A LoRA-adapted 4B model ranks first on both competition tracks, with scores of 0.9279 and 0.9400, while argument exact match peaks at 0.8816. However, a deterministic rule combining candidate order and released metadata matches the best observed function-name accuracy of 0.9973, limiting what this metric establishes about language understanding. Exact argument extraction provides most of the competitive separation. Maghrebi has the lowest score for every system, although its small test subset and differences in example composition limit interpretation. We provide public task data, a baseline checkpoint, the scorer, and final rankings, and document evaluation shortcuts and remaining reproducibility limitations.[<https://huggingface.co/spaces/TuwaiqAcademy/AISA-ArabicFC-Shared-Task>]

Arabic sentiment swap aims to reverse the polarity of a dialectal sentence while preserving its underlying meaning. We present NAMAA’s system for DialectSentEval 2026 Subtask 2. Fine-tuning AraT5v2 achieves 79.56 development sentiment-style accuracy, while reranked sampling over eight candidates, using sentiment agreement and content-overlap signals, raises this to 93.57. Cross-backbone ensembling further reaches a development score of 96.78. On the hidden test set, however, this ordering does not transfer: a widened rerank pool with a re-tuned content weight achieves our best official score of 0.7597, whereas the most complex all-source ensemble scores 0.7343. We identify two likely contributors to this development-to-test mismatch: the absence of gold source-polarity labels at test time, which requires classifier-based polarity inference, and train–development sentence duplication that may inflate development performance. These results show that additional generation and ensemble complexity can improve development metrics substantially without guaranteeing better generalization under shifted evaluation conditions.

Arabic dialect sentiment analysis is challenging because regional varieties show considerable lexical, morphological, syntactical, and orthographic variation. This paper presents a transformer-based system developed for DialectSentEval 2026 Subtask 1, a three-class sentiment classification task covering Moroccan, Egyptian, Jordanian, and Saudi Arabic. The system independently fine-tunes three pretrained Arabic language models: AraBERTv2, MARBERTv2, and CAMeLBERT, using a common training protocol. The implementation uses an 80/20 stratified train-validation split, a maximum sequence length of 128, a batch size of 16, a learning rate of 2 × 10^-5, and focal loss with 𝛾=2=2. Model selection was conducted based on validation macro F1. Using the published labeled test set, MARBERTv2 achieves the strongest independent result with 94.07% macro F1, followed by CAMeLBERT at 81.96% and AraBERTv2 at 79.77%. The results demonstrate the importance of Arabic pretraining and dialect-oriented representations and establish MARBERTv2 as the strongest standalone backbone in the current system.

Arabic knowledge graph construction is a challenging problem due to the linguistic complexity of Arabic and the need for robust generalization across domains. This paper presents our systems for the KnowledgeGraphEval 2026 shared task on Arabic knowledge graph construction and domain adaptation, covering two subtasks: AdaptNER for cross-domain named entity recognition and relation extraction (RE). For AdaptNER, we employ an entity-span ensemble combining two BIO sequence taggers with constrained conditional random field (CRF) decoding and a GlobalPointer span classifier. Our system achieved a micro-F1 score of 76.00%, ranking first in the official evaluation. For relation extraction, we use a majority-vote ensemble of three LoRA-fine-tuned causal language models. This approach achieved a micro-F1 score of 93.48%, ranking fifth. These results demonstrate the effectiveness of model ensembling and complementary prediction strategies for Arabic named-entity recognition and relation extraction.

Reliable Arabic hallucination detection requires not only strong ranking performance but also robustness to distribution shift, calibrated probabilities, and useful uncertainty for selective prediction. We present the NAMAA Community’s submission to HalluScoring 2026 Task ~1.1 and a set of post-submission diagnostic experiments. The scored, template-compliant system is a CAMeLBERT cross-encoder over the question and generated answer; it achieves 0.7596 ROC-AUC on the hidden test set, ranking third among seven teams. We compare this template-defined input path with reference-aware answer-pair models that use the gold answer. A CAMeLBERT input-framing ablation raises the unseen-question ROC-AUC from 0.8713 to 0.9272, while similar gains appear across ARBERT and MARBERTv2. Reference-free distillation recovers only a small portion of this advantage, whereas LoRA-InfoNCE substantially improves direct cosine discrimination. On a separately trained CAMeLBERT reliability replicate, temperature scaling reduces expected calibration error from 0.0993 to 0.0562, and MC-Dropout variance reduces retained-set error from 14.08% to 8.35% when the 30% most uncertain cases are deferred. These results motivate reporting distributional generalization, calibration, and selective risk alongside ROC-AUC. Code and checkpoints for the scored submission are publicly available. [<https://github.com/NAMAA-ORG/HalluScoring-Arabic-Hallucination-Detection>]

Large language models may generate inaccurate Islamic citations when answering religious questions. We present the Namaa Community system for Subtask 2 of IslamicEval 2026, which identifies hallucinations in Qur’anic verses (Ayah), hadith texts (matn), chains of narration (isnad), and claimed sources. We formulate the task as retrieval-grounded verification: citation spans are normalized and matched against the Qur’an and the six major hadith collections using character n-gram retrieval and edit-distance re-ranking. Type-specific verifiers assess Ayah and matn similarity, while claimed sources and isnads are checked against their parent citations. The deterministic pipeline requires no language model or GPU, and every decision is traceable to a canonical record. It achieves a macro accuracy of 0.846 on the development set, 17.9 percentage points above a per-type majority-class baseline, and 0.668 on the official blind test; we trace most of that shortfall to a submission-time configuration error and decompose the remainder by segment type. Our code and submissions are publicly available at https://github.com/NAMAA-ORG/Namaa-Community-at-IslamicEval-2026.

We present NAMAA Community’s systems for the four closed-track Arabic sentence-segmentation subtasks of AraSeg 2026. Heterogeneous ensembles combining LoRA-adapted bidirectional language models, XLM-R variants, a character-level SaT model, and punctuation-aware models outperform their development-selected single-model baselines. Large weights assigned to weaker but architecturally distinct members are consistent with complementary errors, although they do not establish causality. A training-only logistic MEMM improves F1 by +0.9 and +0.6 points on the punctuation-free subtasks but degrades both punctuated ones; the same directional pattern appears on the blind split. Aligning punctuated and punctuation-free inputs reveals F1 points of oracle headroom against the contemporaneous NoPnx locks, while only 52.5% of removed tokens are covered by the restoration label inventory. Genre-proxy analysis localizes the largest degradation to scriptural text. On the blind split, in the closed track, our systems achieved F1 scores of 94.4/91.3/89.9/87.0 on PA, NP, NoPnx-PA, and NoPnx-NP, respectively, ranking 6th/3rd/4th/3rd. The same systems, entered unchanged in the open track, rank 4th/2nd/3rd/3rd. Implementation and experimental resources are publicly available.[<https://github.com/NAMAA-ORG/NAMAA-Community-AraSeg-2026>]

Arabic encompasses diverse spoken varieties with limited parallel data, making both translation and automatic evaluation difficult beyond Modern Standard Arabic. We present the NAMAA Community submissions to all three AlexandriaX-2026 subtasks: English-to-dialect dialogue translation, cross-dialect translation, and dialectal MT error detection. Our systems combine few-shot prompting, fine-tuned encoder-decoder and decoder-only models, and encoder- and LLM-based span detection. Held-out ablations show that minimum-Bayes-risk consensus can suffer from model-family imbalance; we therefore use per-dialect routing for translation and 2-of-3 character-level agreement for error-span detection. Our systems rank fourth and fifth of six in the unconstrained and constrained Subtask 1 tracks, third of four on Subtask 2, and second of six on Subtask 3. The Subtask 3 system achieves 0.5448 overlap-F1, exceeding the overall winner on span localization, with category assignment remaining the main weakness. Code and resources are publicly available.[<https://huggingface.co/collections/NAMAA-Space/namaa-at-alexandriax-2026>]

We introduce Definition-Guided Learning (DGL), a contrastive training framework that uses dictionary definitions as semantic queries and lemmas as retrieval targets, instantiated here for Arabic reverse dictionary retrieval. We construct a root-stratified Arabic lexical benchmark that limits morphological root overlap between training and test partitions. Zero-shot evaluation suggests a notable gap, with the best pre-trained model achieving only 11.16% Top@1 accuracy on this benchmark. Fine-tuning with DGL yields 41.13% Top@1 and NDCG@10 of 49.32%, an improvement of roughly 3.7 times over the strongest zero-shot baseline we tested. Our results suggest that in-batch negative sampling tends to outperform Triplet Loss in this setting, and that definition length and lexical difficulty appear to trade off in importance across retrieval depth. These findings offer preliminary evidence that lexicographic resources can serve as a useful supervisory signal for Arabic embedding learning, at least within the scope of the benchmark studied here.

Academic paraphrasing in Modern Standard Arabic must change expression without distorting scientific content. We study this balance through AraScholar, a corpus of 20,002 manually written source–paraphrase pairs with 522 raw domain labels and item-level provenance. On 2,000 held-out sentences, fine-tuned AraT5v2 favors lexical reuse under beam search, while frontier models produce more extensive rewrites. With corrected Arabic punctuation normalization, diverse decoding and preservation-constrained reranking attain 74.25% coverage of a fixed novelty–preservation band, compared with 52.60% for zero-shot GPT-4o. Gold coverage is only 46.65%, showing that this band measures constraint compliance rather than human quality. Post-hoc target matching substantially improves a calibrated GPT-4o control and narrows the gap, although candidate budgets differ. Crucially, an independent 200-instance audit favors frontier models on meaning and paraphrase quality. We document the corpus, decoding controls, negative distillation result, and metric sensitivity, including test-derived calibration and checkpoint-selection limitations. The findings establish operating-point control as a useful capability while demonstrating why calibration must be evaluated alongside independent assessments of writing quality.

Arabic machine translation benchmarks provide limited coverage of variation within Saudi Arabic. We introduce SAFIR, a Saudi-centric benchmark of 610 unique sentence pairs organized into 12 creator specific evaluation splits across five regional groups: Najdi, Hijazi, Southern, Northern, and Eastern. Native speakers authored the Arabic sentences and their English translations; 235 pairs received independent double review for dialect authenticity, translation adequacy, and fluency. We evaluate twelve systems, comprising losed-weight LLMs, open-weight LLMs, and dedicated MT models, in the zero-shot Saudi→English setting using reference-based metrics and GPT-4o judgments. Closed-weight LLMs obtain the strongest aggregate results. Gemini 2.5 Pro achieves the highest average BLEU (34.41), while GPT-5 achieves the highest BERTScore (83.60). Rankings vary across splits, and Southern varieties present particular challenges in judged adequacy. Differences between reference overlap, adequacy, and fluency support reporting performance by sampled variety and evaluation dimension. SAFIR and its leaderboard support more detailed assessment of Saudi Arabic translation.

Arabic is a highly diglossic language where most daily communication occurs in regional dialects rather than Modern Standard Arabic (MSA). Despite this, machine translation (MT) systems often generalize poorly to dialectal input, limiting their utility for millions of speakers. We introduce Alexandria, a large-scale, community-driven, human-translated dataset designed to bridge this gap. Alexandria covers 13 Arab countries and 11 high-impact domains, including health, education, and agriculture. Unlike previous resources, Alexandria provides unprecedented granularity by associating contributions with city-of-origin metadata, capturing authentic local varieties beyond coarse regional labels. The dataset consists of parallel English-Dialectal Arabic multi-turn conversational scenarios annotated with speaker-addressee gender configurations, enabling the study of gender-conditioned variation in dialectal use. Comprising 107K total turns, Alexandria serves as both a training resource and as a rigorous benchmark for evaluating MT and Large Language Models (LLMs). Our automatic and human evaluation benchmarks the current capabilities of Arabic-aware LLMs in translating across diverse Arabic dialects and sub-dialects while exposing significant persistent challenges.The Alexandria dataset, the creation prompts, the translation and revision guidelines, and the evaluation code are publicly available in the following repository: https://github.com/UBC-NLP/Alexandria

Encoder-only transformer models remain widely used for discriminative NLP tasks, yet recent architectural advances have largely focused on English. In this work, we present AraModernBERT, an adaptation of the ModernBERT encoder architecture to Arabic, and study the impact of transtokenized embedding initialization and native long-context modeling up to 8,192 tokens. We show that transtokenization is essential for Arabic language modeling, yielding dramatic improvements in masked language modeling performance compared to non-transtokenized initialization. We further demonstrate that AraModernBERT supports stable and effective long-context modeling, achieving improved intrinsic language modeling performance at extended sequence lengths. Downstream evaluations on Arabic natural language understanding tasks, including inference, offensive language detection, question-question similarity, and named entity recognition, confirm strong transfer to discriminative and sequence labeling settings. Our results highlight practical considerations for adapting modern encoder architectures to Arabic and other languages written in Arabic-derived scripts.

Arabic Large Language Models are usually evaluated using Western-centric benchmarks that overlook essential cultural contexts, making them less effective and culturally misaligned for Arabic-speaking communities. This study addresses this gap by evaluating the Arabic Massive Multitask Language Understanding (MMLU) Benchmark to assess its cultural alignment and relevance for Arabic Large Language Models (LLMs) across culturally sensitive topics. A team of eleven experts annotated over 2,500 questions, evaluating them based on fluency, adequacy, cultural appropriateness, bias detection, religious sensitivity, and adherence to social norms. Through human assessment, the study highlights significant cultural misalignments and biases, particularly in sensitive areas like religion and morality. In response to these findings, we propose annotation guidelines and integrate culturally enriched data sources to enhance the benchmark’s reliability and relevance. The research highlights the importance of cultural sensitivity in evaluating inclusive Arabic LLMs, fostering more widely accepted LLMs for Arabic-speaking communities.

Mainstream large vision-language models (LVLMs) inherently encode cultural biases, highlighting the need for diverse multimodal datasets. To address this gap, we introduce PEARL, a large-scale Arabic multimodal dataset and benchmark explicitly designed for cultural understanding. Constructed through advanced agentic workflows and extensive human-in-the-loop annotations by 37 annotators from across the Arab world, PEARL comprises over 309K multimodal examples spanning ten culturally significant domains covering all Arab countries. We further provide two robust evaluation benchmarks (PEARL and PEARL-LITE) along with a specialized subset (PEARL-X) explicitly developed to assess nuanced cultural variations. Comprehensive evaluations on state-of-the-art open and proprietary LVLMs demonstrate that reasoning-centric instruction alignment substantially improves models’ cultural grounding compared to conventional scaling methods. PEARL establishes a foundational resource for advancing culturally-informed multimodal modeling research. All datasets and benchmarks are publicly available.

Enhancing the linguistic capabilities of Large Language Models (LLMs) to include low-resource languages is a critical research area. Current research directions predominantly rely on synthetic data generated by translating English corpora, which, while demonstrating promising linguistic understanding and translation abilities, often results in models aligned with source language culture. These models frequently fail to represent the cultural heritage and values of local communities. This work proposes a methodology to create both synthetic and retrieval-based pre-training data tailored to a specific community, considering its (i) language, (ii) cultural heritage, and (iii) cultural values. We demonstrate our methodology using Egyptian and Moroccan dialects as testbeds, chosen for their linguistic and cultural richness and current underrepresentation in LLMs. As a proof-of-concept, we develop NileChat, a 3B parameter Egyptian and Moroccan Arabic LLM adapted for Egyptian and Moroccan communities, incorporating their language, cultural heritage, and values. Our results on various understanding, translation, and cultural and values alignment benchmarks show that NileChat outperforms existing Arabic-aware LLMs of similar size and performs on par with larger models. This work addresses Arabic dialect in LLMs with a focus on cultural and values alignment via controlled synthetic data generation and retrieval-augmented pre-training for Moroccan Darija and Egyptian Arabic, including Arabizi variants, advancing Arabic NLP for low-resource communities.We share our methods, data, and models with the community to promote the inclusion and coverage of more diverse communities in cultural LLM development: https://github.com/UBC-NLP/nilechat.

As large language models (LLMs) become increasingly integrated into daily life, ensuring their cultural sensitivity and inclusivity is paramount. We introduce PALM, a year-long community-driven project covering all 22 Arab countries. The dataset contains instruction–response pairs in both Modern Standard Arabic (MSA) and dialectal Arabic (DA), spanning 20 diverse topics. Built by a team of 44 researchers across the Arab world—each an author of this paper—PALM offers a broad, inclusive perspective. We use PALM to evaluate the cultural and dialectal capabilities of several frontier LLMs, revealing notable limitations: while closed-source LLMs generally perform strongly, they still exhibit flaws, and smaller open-source models face greater challenges. Furthermore, certain countries (e.g., Egypt, the UAE) appear better represented than others (e.g., Iraq, Mauritania, Yemen). Our annotation guidelines, code, and data are publicly available for reproducibility. More information about PALM is available on our project page: https://github.com/UBC-NLP/palm.

The translation between Modern Standard Arabic (MSA) and the various Arabic dialects presents unique challenges due to the significant linguistic, cultural, and contextual variations across the regions where Arabic is spoken. This paper presents a system description of our participation in the OSACT 2024 Dialect to MSA Translation Shared Task. We explain our comprehensive approach which combines data augmentation techniques using generative pre-trained transformer models (GPT-3.5 and GPT-4) with fine-tuning of AraT5 V2, a model specifically designed for Arabic translation tasks. Our methodology has significantly expanded the training dataset, thus improving the model’s performance across five major Arabic dialects, namely Gulf, Egyptian, Levantine, Iraqi, and Maghrebi. We have rigorously evaluated our approach, using BLEU score, to ensure translation accuracy, fluency, and the preservation of meaning. Our results showcase the effectiveness of our refined models in addressing the challenges posed by diverse Arabic dialects and Modern Standard Arabic (MSA), achieving a BLEU score of 80% on the validation test set and 22.25% on the blind test set. However, it’s important to note that while utilizing a larger dataset, such as Madar + Dev, resulted in significantly higher evaluation BLEU scores, the performance on the blind test set was relatively lower. This observation underscores the importance of dataset size in model training, revealing potential limitations in generalization to unseen data due to variations in data distribution and domain mismatches.

This study undertakes a comprehensive investigation of transformer-based models to advance Arabic language processing, focusing on two pivotal aspects: the estimation of Arabic Level of Dialectness and dialectal sentence-level machine translation into Modern Standard Arabic. We conducted various evaluations of different sentence transformers across a proposed regression model, showing that the MARBERT transformer-based proposed regression model achieved the best root mean square error of 0.1403 for Arabic Level of Dialectness estimation. In parallel, we developed bi-directional translation models between Modern Standard Arabic and four specific Arabic dialects—Egyptian, Emirati, Jordanian, and Palestinian—by fine-tuning and evaluating different sequence-to-sequence transformers. This approach significantly improved translation quality, achieving a BLEU score of 0.1713. We also enhanced our evaluation capabilities by integrating MSA predictions from the machine translation model into our Arabic Level of Dialectness estimation framework, forming a comprehensive pipeline that not only demonstrates the effectiveness of our methodologies but also establishes a new benchmark in the deployment of advanced Arabic NLP technologies.

Semantic search tasks have grown extremely fast following the advancements in large language models, including the Reverse Dictionary and Word Sense Disambiguation in Arabic. This paper describes our participation in the Contemporary Arabic Dictionary Shared Task. We propose two models that achieved first place in both tasks. We conducted comprehensive experiments on the latest five multilingual sentence transformers and the Arabic BERT model for semantic embedding extraction. We achieved a ranking score of 0.06 for the reverse dictionary task, which is double than last year’s winner. We had an accuracy score of 0.268 for the Word Sense Disambiguation task.

This paper describes our participation in the ArAIEval Shared Task 2024, focusing on Task 2C, which challenges participants to detect propagandistic elements in multimodal Arabic memes. The challenge involves analyzing both the textual and visual components of memes to identify underlying propagandistic messages. Our approach integrates the capabilities of MARBERT and ResNet50, top-performing pre-trained models for text and image processing, respectively. Our system architecture combines these models through a fusion layer that integrates and processes the extracted features, creating a comprehensive representation that is more effective in detecting nuanced propaganda. Our proposed system achieved significant success, placing second with an F1 score of 0.7987.