This paper describes the system we designed for our participation in the AraSentEval 2026 shared task on Arabic dialectal sentiment analysis. We propose a transformer-based approach relying on MARBERT combined with a multi-seed ensemble strategy and several optimization techniques. Our system integrates seven independently trained models with different random initializations and applies Stochastic Weight Averaging (SWA) to improve generalization. To address class imbalance, we augment the training data through dialectal synonym replacement, increasing the dataset size by 13.9% while preserving dialect distribution. In addition, we incorporate Test-Time Augmentation (TTA) and investigate the use of pseudo-labeling based on high-confidence predictions. We report our experiments on the official dataset covering Moroccan, Egyptian, Jordanian, and Saudi dialects, and analyze the contribution of each component through ablation experiments. Our system achieved a macro F1-score of 84.62% on the test set, ranking 3rd among 15 participating teams.

Legal text is hard for named entity recognition. New statutes, parties and courts appear continually, and expert annotation is costly. Open-world NER can assign types that were not fixed at training time. These systems are built on general-domain text, though, and legal uses adapt them corpus by corpus and language by language. We take the opposite route, and apply OPENER unchanged, a frugal pipeline of ready-to-use components. Its entity embedder stays generic, and no part of it is ever fitted to a legal schema. We test its two typing heads on four legal corpora, in English, Portuguese and German. One needs no target label at all, the other fits a small linear probe. We compare them against GLiNER, GNER, OWNER and a 4-bit Qwen, on quality, latency and energy at once. The transfer holds in English, and weakens across languages. The label-free head degrades sharply. The supervised probe is barely affected, and scores best overall (35.9 AMI end-to-end, 61.4 on gold mentions). Both stay one to two orders of magnitude cheaper than the generative baselines. Detection, not typing, is what caps every system here. The English detectors recover as little as 10.2% of the German mentions. On the ones they do find, our supervised head assigns the right type 78.7% of the time on average over the four corpora, against 64.6% for the detector alone. Yet typing all of them correctly would still stop at 70 AMI out of 100. Where labels exist, the supervised head is the practical operating point, and in-language detection is where adaptation should go first.

Named Entity Recognition for dialectal Arabic remains severely constrained by data scarcity, and LLM-based augmentation is the natural remedy. We study LLM-based augmentation for the Lebanese and Palestinian sub-corpora of Wojood 2024, using LLaMA-3.1-8B-Instruct as a generation backbone, and compare three strategies (entity-level, context-level, combined) across two k-shot sampling paradigms. The proposed multi-dimensional pipeline yields a per-sentence composite score Q(s) that exposes a sharp dialectal-authenticity gap between entity- and context-level rewriting (DA 0.92 vs. 0.51). Despite lower register fidelity, context augmentation with curriculum training delivers the most consistent gains (Curras F1 +5.70 pp); entity augmentation lifts Lebanese F1 from a zero baseline to 50.00 on Lebanese-only training, a figure the 85-sentence test set makes coarse. Ablation, weight-sensitivity, threshold sweep, a random-selection control, and a second LLM backbone show that the dialectal-authenticity dimension carries most of Q(s)’s protective effect on Lebanese and that generator choice is itself a first-class register-preservation lever.

This paper describes the system we designed for our participation to SemEval2023 Task 12 Track 6 about Algerian dialect sentiment analysis. We propose a transformer language model approach combined with a lexicon mixing terms and emojis which is used in a post-processing filtering stage. The Algerian sentiment lexicons was extracted manually from tweets. We report on our experiments on Algerian dialect, where we compare the performance of marbert to the one of arabicbert and camelbert on the training and development datasets of Task 12. We also analyse the contribution of our post processing lexical filtering for sentiment analysis. Our system obtained a F1 score equal to 70%, ranking 9th among 30 participants.