Detecting polarization in online discourse is important for understanding social fragmentation , yet it remains difficult for Arabic due to dialect variation, informal writing, and implicit framing. In this paper, we study Arabic polarization modeling in the SemEval-2026 Task 9 (POLAR) setting, focusing on polarization detection (ST1) and polarization type classification (ST2). We compare three approaches: encoder fine-tuning, zero-shot prompting, and retrieval-augmented in-context learning (RAG-ICL), across six Arabic encoders and different LLMs. For ST1, RAG-ICL with Gemma-3-27b-it achieves the best result (test macro F1 = 0.83), while remaining competitive with the best fine-tuned encoder (0.82), and substantially outperforming zero-shot prompting. For ST2, a pipeline that first applies the best ST1 encoder as a hard filter and then performs RAG-ICL achieves a macro F1 = 0.62. Prompt-language effects are model-and task-dependent, with some settings doing better with English prompts and others with Arabic prompts. Chain-of-thought, self-refinement, and contrastive prompting do not outperform standard RAG-ICL.

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.

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.

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.