Islamic inheritance is a highly structured and rule-intensive domain that requires precise reasoning. The QIAS 2026 Shared Task introduces a benchmark for evaluating generative artificial intelligence on end-to-end inheritance problem solving. In this paper, we present our team Silah’s participation in the QIAS 2026 shared task, where we compare three approaches: (1) a multi-stage retrieval-augmented, rule-guided pipeline, (2) supervised fine-tuning of generative large language models, and (3) a retrieval-augmented fine-tuning approach. We evaluate several open-source models, including Qwen2.5, Llama, DeepSeek, and Fanar. Our results show that supervised fine-tuning consistently outperforms retrieval-based approaches, with the fine-tuned Fanar-1-9B-Instruct model achieving the best performance (MIR-E = 0.83) and ranking sixth overall in the shared task. These findings suggest that learning implicit reasoning patterns through fine-tuning is more effective than explicit rule injection under current retrieval setups, and emphasize the need for more accurate and minimal rule selection mechanisms in future retrieval-augmented approaches.

This paper presents our submission to the AdabEval 2026 shared task on Arabic politeness classification and pragmatic category prediction. We explored a range of Arabic-specific and multilingual transformer models and integrated their outputs through an ensemble strategy. Our approach achieved state-of-the-art performance in the shared task, ranking first in both subtasks with a macro-F1 score of 0.89 and an accuracy of 0.93 on subtask A, and a macro-F1 score of 0.58 on subtask B. Although our approach delivered high performance on overall politeness classification, pragmatic category prediction remains more challenging. Despite achieving the top ranking in this subtask, the comparatively lower macro-F1 score suggests that modelling fine-grained pragmatic functions requires further methodological refinement and experimentation.

We describe the GHADAI approach to the AraGenre 2026 shared task, which requires classifying Arabic text into broad and specific genre categories, including specific genres unseen during training. Our approach is a two-stage pipeline: a fine-tuned multilingual bi-encoder scores all 27,972 test sentences against the 74 candidate definitions, and a large language model reranks the top candidates per sentence using cyclic option rotation to mitigate positional bias. On the hidden test set, the bi-encoder achieves a hierarchical macro-F1 of 0.4213, which increases to 0.5006 with Fanar-9B reranking and to 0.6050 with Qwen-32B reranking over the top 15 candidates using five rotations. The final pipeline ranked fourth on the official leaderboard. These results demonstrate the effectiveness of combining definition-based retrieval with LLM reranking for generalization to unseen Arabic text genres.

This paper describes the author’s participation in the Arabic track of the AbjadAuthorID shared task which focuses on multiclass authorship attribution using transformer-based models. The task involves identifying the author of a given text excerpt drawn from diverse genres and historical periods, posing significant challenges due to stylistic variation and linguistic richness. Experimental results demonstrate strong performance, with an ensemble of MAR BERTv2 and ARBERTv2 achieving achieving an accuracy of 92% and a macro-averaged F1 score of 89%, ranking second on the leader board, and highlighting the effectiveness of the proposed approach for Arabic authorship identification.