The ANLP_UniSo team submitted a system to Subtask 1 (Named Entity Recognition) of KnowledgeGraphEval 2026, which focuses on nested NER in Arabic and evaluates crossdomain and cross-dialect generalization from the Wojood corpus to the multi-domain Konooz benchmark. The system is a combination of an Efficient GlobalPointer decoder with rotary position embeddings and a MARBERT backbone. To reduce the domain gap, first domainadaptive pretraining (DAPT) is performed on unlabeled Konooz text, and then a pseudolabeling self-training stage is applied. A second model based on AraBERT is trained and combined to the other model using a stacking ensemble to improve prediction diversity. As for the Wojood test set, the best ensemble achieves a micro F1 of 0.9288, increasing to 0.9292 when entity-type-specific decision thresholds are used, compared with 0.9211 for the single MARBERT model. On the official blind Konooz test set, however, the submitted system obtains a micro F1 of 0.7106 (macro F1 = 0.6101, weighted F1 = 0.7013, P = 0.8143, R = 0.6304, accuracy = 0.9948). The lower recall suggests that the decoding threshold chosen based on the Wojood validation data was too strict for the domain-shifted Konooz data. the system study the difference and propose a threshold-recalibration strategy to improve the performance under the target-domain distribution.

Recent transformer-based language models rely on subword tokenization and contextual embeddings, raising questions about the continued effectiveness of traditional preprocessing techniques such as stemming. This issue remains underexplored for low-resource Arabic dialects, particularly Tunisian Dialect (TD). In this paper, we investigate the impact of light and aggressive stemming on two main Tunisian NLP tasks: sentiment analysis and domain classification. We compare classical machine learning methods with transformer-based Arabic language models, including TunBERT. To support this study, we release an open Tunisian stemming resource containing 7,730 words annotated with both light and aggressive stems. Our experiments provide a comprehensive analysis of TD processing and show that the effectiveness of stemming depends on both the task and the model type.