In this paper, we describe our submission to the WMT26 Multilingual Instruction Shared Task (MIST), Sub-task 1 (context-based question answering). We fine-tune Qwen2.5-3B-Instruct with a LoRA adapter on 3,112 extractive-QA examples drawn from the answerable subset of TyDi QA as released in the wmt26-mist-sample data. We submit outputs for the monolingual Arabic (question and context both in Arabic) instances of the official qa-context test set. Our system reaches a mean token accuracy of 0.9643 and produces accurate extractive spans. On the held-out split, fine-tuning roughly doubles both exact match (23.81 → 53.97) and token-level F1 (37.00 → 68.52) relative to the base model. The adapter corrects over-refusal tendency in addition to improving span precision. The base model emitted the prescribed no-answer string in 60% of passages with answers, versus 0% for the fine-tuned model. We report error analysis identifying recurring failures, and discuss the model’s weaker behaviour on cross-lingual context instances, which make up the large majority of the official test set, as a direct consequence of training exclusively on monolingual data.

Dialectal Arabic continues to represent a persistent challenge for contemporary large language models, which are predominantly trained and optimized for Modern Standard Arabic (MSA) and therefore exhibit limited capability when processing colloquial varieties. In this study, a dedicated system developed for participation in the AMIYA shared task focusing on Syrian Arabic is presented. The proposed solution is based on the integration of parameter-efficient fine-tuning through Low-Rank Adaptation (LoRA) with prompt-guided inference, aiming to enhance dialectal adequacy and linguistic naturalness. Rather than emphasizing strict factual precision, the system is deliberately designed to prioritize fluent and authentic Syrian Arabic generation, in accordance with the evaluation principles adopted by the AL-QASIDA benchmark. This design choice reflects a focus on human-perceived language quality and dialectal fidelity, which are central to effective dialect-aware language modeling.

We describe our submission to Subtask 1b of ImageEval 2026. The task consists of selecting the single visually valid statement from three candidates paired with an image, and is in both English and Modern Standard Arabic (MSA). Our system uses a pre-trained 7B vision–language model in a zero-shot setting, without task-specific training or fine-tuning. We formulate the task as a constrained selection problem and choose the answer using next-token log-probabilities. To reduce answer-position bias, we apply a Latin-square rotation to the candidate statements. Our system achieves CI scores of 0.0510 on the English test set and 0.0960 on the MSA test set, reducing CI relative to the official 3B baselines reported on devtest by 83.7% and 80.4%, respectively.

This paper describes team Alkhder’s submission to Subtask 1 of the AlexandriaX-2026 shared task, which asks systems to translate English dialogue turns into thirteen Arabic dialects. We fine-tune NileChat-3B-Base with QLoRA on the Alexandria parallel dialogue corpus, training one set of adapter weights for all dialects. Each turn is translated conditioned on the target country and dialect, the domain, the participant roles, the speaker–addressee gender direction, and the preceding dialogue context. At inference, conversations are translated autoregressively and the model is fed its own previous outputs back as context without using any gold target-side test information. Under the constrained-track, using only Alexandria data, it trains in about one hour and translates the private test set in under two hours. The system reaches a country-macro-averaged spBLEU of 20.36 and chrF++ of 36.64, on both tracks.