[2604.06346] Severity-Aware Weighted Loss for Arabic Medical Text Generation

Computer Science > Computation and Language arXiv:2604.06346 (cs) [Submitted on 7 Apr 2026] Title:Severity-Aware Weighted Loss for Arabic Medical Text Generation Authors:Ahmed Alansary, Molham Mohamed, Ali Hamdi View a PDF of the paper titled Severity-Aware Weighted Loss for Arabic Medical Text Generation, by Ahmed Alansary and 2 other authors View PDF HTML (experimental) Abstract:Large language models have shown strong potential for Arabic medical text generation; however, traditional fine-tuning objectives treat all medical cases uniformly, ignoring differences in clinical severity. This limitation is particularly critical in healthcare settings, where errors in severe cases contain higher clinical risk. In this work, we propose a severity-aware weighted loss for fine-tuning Arabic language models on medical complaint-response data. The method depends on soft severity probabilities to dynamically scale token-level loss contributions during optimization, thereby prioritizing clinically critical interactions without modifying model architectures. Experiments are conducted using the MAQA dataset, which provides Arabic medical complaints and trusted human responses. Severity labels and probabilistic scores are automatically derived using a fine-tuned AraBERT-based classifier and incorporated exclusively at the loss level. The proposed approach is evaluated across ten Arabic large language models of varying architectures and parameter scales. While standard cross-entropy fine...