Improving Arabic Clinical Question Quality through Domain-Adaptive Masked Language Modeling

Improving Arabic Clinical Question Quality through Domain-Adaptive Masked Language Modeling | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Improving Arabic Clinical Question Quality through Domain-Adaptive Masked Language Modeling Walid Ounachad, Mohamed Khenchouch, Imad Zeroual, Yousef Farhaoui This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8007820/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 15 You are reading this latest preprint version Abstract Arabic clinical NLP systems often receive short, vague, or incomplete questions, which yields weak downstream answers even with strong encoders. We address this bottleneck by making question quality a first-class, measurable objective. Using domain-adaptive (continued) pretraining with a masked-language objective (DAPT-MLM) on AHQAD (~ 808k Arabic health Q–A pairs), we adapt two widely used backbones—AraBERT and the generator variant of AraELECTRA—to the lexical, syntactic, and discourse patterns of well-formed medical questions. Evaluation is aligned with the learning signal: we report cross-entropy and perplexity only at masked tokens, top-k accuracy restricted to masked spans, and lexical-diversity measures to discourage formulaic phrasing. A length-controlled test design (Short/Long/Very Long) isolates modeling gains from verbosity. Results show consistent intrinsic improvements for the domain-adapted models; AraBERT-MLM is best overall (macro Top-5 = 0.8392, lowest CE/PPL), outperforming AraBERT (orig.) by + 6.0 pp Top-5 and AraELECTRA (orig.) by + 17.2 pp. A 200-item human study (clinician + linguist) corroborates these gains (mean ± 95% CI: Clarity 4.12 ± 0.18, Fluency 3.68 ± 0.22, Semantic Fidelity 3.15 ± 0.25, Usefulness 3.42 ± 0.21; substantial agreement, κ ≈ 0.77) and highlights residual semantic drifts that inform simple, slot-constrained decoding fixes. Overall, the proposed reformulation module produces more natural and clinically relevant Arabic questions and can be plugged into Arabic clinical QA pipelines as a measurable, tunable front-end. Arabic clinical NLP question reformulation domain-adaptive pretraining masked language modeling intrinsic evaluation human validation 1 Introduction Writing well-formed questions in Arabic remains challenging for NLP systems, with persistent gaps between Modern Standard Arabic (MSA) and dialects highlighted by recent surveys and benchmarks [ 1 , 2 ]. In semantic search, virtual assistants, and QA, short, elliptical, or loosely structured queries often produce off-target answers—even with strong encoders such as ARBERT/MARBERT [ 3 ]. Part of the difficulty is linguistic—spelling variation, rich morphology, and the MSA–dialect gap—and part of it is practical: few curated resources spell out what a “good” Arabic question” looks like, particularly in clinical settings. Scaling efforts in large Arabic LLMs also underscore opportunities and constraints relevant to clinical scenarios [ 4 ]. We address this problem at its source through domain-specific adaptive pretraining for Arabic question reformulation. Using AHQAD/AHD, a large corpus of Arabic health question–answer pairs [ 5 ], we continue pretraining two widely used backbones—AraBERT and the generator variant of AraELECTRA—under a masked-language objective so they better capture the structure of well-formed medical questions. To test models under realistic conditions, we build length-balanced test sets and, for each item, create a truncated version that must be reconstructed into a coherent, context-appropriate question. Our evaluation mirrors the training signal: we compute cross-entropy and perplexity at masked token positions and measure top-k accuracy only where masking occurs, following token-aligned scoring for masked language models [ 6 ]. We also conduct a human study with one language specialist and one domain expert who rate clarity, fluency, semantic fidelity, and practical usefulness of the reformulated questions; inter-rater agreement is summarized with modern reliability coefficients [ 7 ]. Taken together, these components provide a reusable path to improving Arabic question formulation: large-scale domain adaptation, measurements consistent with the learning objective and controlled for length, and expert validation that links numerical gains to judged quality. The resulting module can be plugged into Arabic clinical QA pipelines [ 8 ] and treated as a measurable, tunable component rather than an assumption. Key Contributions. Method . A reproducible framework for Arabic question reformulation via domain-adaptive (continued) pretraining on AHQAD with the MLM objective. Evaluation design . Token-level metrics aligned with MLM (cross-entropy, perplexity, top-k on masked positions), length-controlled splits, and lexical-diversity tracking. Human validation . Expert review (linguistics + domain) assessing clarity, fluency, semantic fidelity, and usefulness alongside automatic metrics. Practicality . A reformulation module that integrates cleanly into Arabic clinical QA pipelines, turning question quality into a measurable, tunable target. 2 Related Work 2.1 Arabic pretrained encoders Arabic-specific encoders consistently outperform multilingual baselines across diverse NLU tasks [ 1 , 3 ]. AraBERT adapts BERT’s Masked Language Modeling (MLM) objective to Arabic and established early state-of-the-art performance on common benchmarks. AraELECTRA brings ELECTRA’s replaced-token detection to Arabic, offering a strong efficiency–accuracy trade-off. Larger families such as ARBERT/MARBERT expand coverage to dialectal and social-media text, underscoring the importance of scale and domain breadth in Arabic pretraining. More recent initiatives like AraT5 extend this line to generative, text-to-text formulations [ 3 ], while large-scale surveys consolidate the progress and remaining challenges in Arabic LLMs [ 1 ]. 2.2 Domain-adaptive pretraining (DAPT/TAPT) A second pretraining phase on in-domain text reliably improves downstream performance. [ 9 ] formalized Domain-Adaptive Pretraining (DAPT) and Task-Adaptive Pretraining (TAPT), showing consistent gains across domains and data regimes. Subsequent work confirms that adaptation to task-specific corpora reduces cross-domain drift and enhances generalization [ 6 , 9 ]. We follow this paradigm by adapting Arabic encoders on QA-style health data prior to evaluation, aligning pretraining distribution with target clinical use cases. 2.3 MLM objectives and token-level evaluation BERT-style Masked Language Modeling (MLM) motivates token-aligned evaluation rather than sequence-only scores. [ 6 ] introduced pseudo-log-likelihood (PLL) scoring to obtain token-level probabilities from MLMs, and subsequent studies refined PLL for greater theoretical consistency [ 10 ]. This motivates our use of cross-entropy and perplexity at masked positions, top-k accuracy restricted to masked tokens, and lexical-diversity tracking to avoid formulaic phrasing. 2.4 Arabic QA resources (recent and large-scale) Arabic QA has evolved from early reading comprehension datasets to large, contemporary corpora. ArabicaQA [ 8 ] provides 89 k + MRC questions, an open-domain QA benchmark, and a dense retriever (AraDPR) forming a modern testbed for readers and retrievers [ 11 ]. In the health domain, AHQAD/AHD [ 5 ] aggregate approximately 808 000 Arabic Q–A pairs across nearly 90 specialties, enabling large-scale domain adaptation and probing. Together, these resources make Arabic-specific DAPT feasible and closer to real-world distributions [ 12 , 13 ]). 2.5 Reformulation and human evaluation in Arabic NLP Compared to English, Arabic question reformulation remains underexplored—most prior work targets encoder or reader architectures without isolating the effect of query form. Recent interest in Arabic clinical and general-domain LLMs has emphasized pretraining and reasoning, but systematic human studies assessing clarity, fluency, and semantic fidelity remain scarce [ 1 ]. We complement prior efforts by (i) performing DAPT on QA-style medical data to directly improve question form, and (ii) conducting a two-expert human evaluation aligned with intrinsic MLM-consistent metrics [ 7 , 14 ] . 3 Methodology 3.1 Problem Setting We address the problem of reformulating Arabic clinical questions so that they are complete, clear, and clinically useful[ 8 , 12 ]. Starting from a natural question, we create a “truncated” version by hiding one or more informative spans (e.g., a symptom, a duration, a medication, or discourse cues). The system’s task is to reconstruct a well-formed question that preserves the original intent and improves readability and completeness. This setup mirrors real interactions where user queries are often short, elliptical, or underspecified. 3.2 Data and Pre-processing Corpus. We use AHQAD/AHD, a large Arabic health question–answer collection (approximately 808,000 pairs across ~ 90 specialties)[ 5 ]. The corpus provides realistic clinical phrasing and domain terminology. Cleaning and normalization. We remove duplicates, normalize Unicode, strip optional diacritics, collapse elongation marks and extra spaces, and apply light orthographic normalization (alif/hamza and yaa/maqṣūra variants). Non-Arabic script is discarded. We also apply length filtering to reduce extremes and noise[ 1 , 15 ]. Evaluation set. We build a controlled evaluation set of 600 questions stratified by length into three balanced buckets (Short, Long, Very Long; 200 each). For every fully written “reference” question, we create a truncated counterpart by removing an informative span and inserting a placeholder. Length control. All reporting is provided per length bucket with macro averages to separate perceived quality from verbosity[ 10 ]. 3.3 Domain-Adaptive Pretraining We perform domain-adaptive (continued) pretraining on Arabic clinical text using the masked-language-modeling objective. A fixed proportion of tokens is selected for masking; among those, most are replaced with a mask symbol, a smaller share is replaced with a random token, and the remainder is left unchanged. This standard policy encourages the model to use both sides of the context and to learn domain-specific lexical and structural patterns[ 6 , 9 ]. Backbones. We adapt two widely used Arabic encoders: AraBERT (BERT-style) and the generator variant of AraELECTRA. For each, we compare the original public checkpoint with a domain-adapted variant obtained by continued pretraining on AHQAD[ 3 ]. Training setup. We use AdamW, standard warmup and linear decay, mixed precision when available, early stopping on development performance, and three random seeds. Sequence length and batch size are tuned within practical ranges; we log learning curves and keep the best development checkpoint[ 16 ]. 3.4 Reformulation Inference At inference time, we do not free-generate entire questions. Instead, we replace each placeholder in the truncated question by one or a short span of mask tokens and fill only those masked slots with the domain-adapted model. Decoding is greedy (Top-1) or Top-k and is restricted strictly to the masked positions[ 6 , 10 ]. We then detokenize, normalize, and fix whitespace and punctuation. This constrained procedure keeps inference consistent with the training signal and limits semantic drift. 3.5 Evaluation Protocol Intrinsic, token-aligned metrics. We evaluate only where masking occurs, to stay faithful to the learning objective. We report cross-entropy and perplexity at masked positions (lower is better), Top-k accuracy at masked positions (higher is better), and lexical diversity through the type–token ratio, complemented by a moving-average variant to reduce length bias. Results are reported per length bucket and as macro averages[ 6 , 17 ]. Human evaluation. Two experts—a linguist and a clinician—independently rate each reconstructed question for clarity, fluency, semantic fidelity, and practical usefulness on five-point scales. We report per-criterion means with 95% confidence intervals and inter-rater agreement (weighted kappa). Items are randomized and raters are blinded to model identity[ 7 , 14 ]. Statistical testing. Within each backbone, we run paired, non-parametric tests comparing original and domain-adapted variants across metrics and buckets, with false-discovery-rate correction for multiple comparisons. We also report an effect size for the primary endpoint [ 18 , 19 ]. 3.6 Implementation and Reproducibility Platform. Experiments are conducted on Google Colab Pro + with A100/V100/T4 GPUs depending on the session, at least 25 GB RAM, and sufficient ephemeral storage. We checkpoint frequently to mitigate session resets. Software. Python 3.10; PyTorch; Hugging Face Transformers, Datasets, and Accelerate; PEFT when needed. Mixed precision uses bf16 on A100 and fp16 on V100/T4. Tokenizer parallelism is disabled to reduce non-determinism. Randomness control. We fix three seeds and enable deterministic settings when supported. Released artifacts. We provide normalization and truncation scripts, train/dev/test splits with cryptographic hashes, evaluation masks, training logs, and both final and best-development checkpoints. Ethics and intended use. The module reformulates questions; it does not produce diagnoses or medical advice. Data are public and anonymized. We recommend human-in-the-loop use in clinical settings. 4 Results 4.1 Intrinsic results by model All scores are computed exclusively at masked positions and reported by length (Short / Long / Very Long) plus a macro average. Notation: ↓ (CE, PPL) = lower is better; ↑ (Top-k, TTR) = higher is better. Table 1 AraBERT (original) Length PPL (↓) CE (↓) Top-1 (↑) Top-3 (↑) Top-5 (↑) TTR (↑) Short 7.695 2.040 0.564 0.763 0.801 0.145 Long 8.182 2.101 0.616 0.735 0.784 0.030 Very Long 7.050 1.953 0.587 0.738 0.789 0.032 Macro Avg 7.642 2.031 0.589 0.745 0.791 0.069 Table 2 AraELECTRA (original) Length PPL (↓) CE (↓) Top-1 (↑) Top-3 (↑) Top-5 (↑) TTR (↑) Short 120.511 4.791 0.521 0.650 0.682 0.145 Long 184.893 5.219 0.498 0.621 0.670 0.030 Very Long 434.491 6.074 0.461 0.597 0.647 0.032 Macro Avg 246.631 5.361 0493 0.623 0.667 0.069 Table 3 AraELECTRA-MLM (DAPT) Length PPL (↓) CE (↓) Top-1 (↑) Top-3 (↑) Top-5 (↑) TTR (↑) Short 6.026 1.796 0.597 0.801 0.849 0.145 Long 7.45 2.008 0.574 0.741 0.779 0.030 Very Long 10.911 2.389 0.529 0.685 0.734 0.032 Macro Avg 8.129 3.064 0567 0.742 0.742 0.069 Table 4 AraBERT-MLM (DAPT) Length PPL (↓) CE (↓) Top-1 (↑) Top-3 (↑) Top-5 (↑) TTR (↑) Short 4.488 1.501 0.645 0.806 0.860 0.145 Long 4.177 1.429 0.712 0.797 0.859 0.030 Very Long 6.138 1.814 0.629 0.759 0.797 0.032 Macro Avg 4.934 1.581 0.662 0.787 0.839 0.069 Table 5 Macro-average comparison Model CE (↓) PPL (↓) Top-1 (↑) Top-3 (↑) Top-5 (↑) TTR (↑) AraBERT (orig.) 2.031 7.642 0.589 0.745 0.791 0.069 AraELECTRA (orig.) 5.361 246.63 0.493 0.623 0.667 0.069 AraELECTRA-MLM 2.064 8.129 0.567 0.742 0.787 0.069 AraBERT-MLM 1.581 4.934 0.662 0.787 0.839 0.069 Key takeaway. Domain-adaptive pretraining (DAPT) yields consistent gains; AraBERT-MLM is best overall (highest macro-Top-5 0.839, lowest CE/PPL), with the strongest advantage on short items. 4.2 Qualitative analysis (before → output → reference) To complement the intrinsic metrics, we report representative triplets that illustrate constrained fill-ins at masked spans. Each triplet comprises the truncated prompt (before), the model’s reformulation confined to the masked slots (output), and the fully specified reference question. We label cases High-Gain (HG) when the output preserves the clinical intent with only superficial differences, Borderline (BL) when the intent is mostly preserved but discourse scaffolding or phrasing is imperfect, and Failure (FL) when key semantics drift—for example, substituting a diagnosis for a symptom (or vice versa). Table 6 Qualitative examples: before → model output → reference. Q_before Q-generated (model) Q_after(reference) Verdict ما العلاقة بين ___ في المعدة وصعوبة في البلع؟ ما العلاقة بين ألم في المعدة وصعوبة في البلع؟ ما العلاقة بين حرقة شديدة في المعدة وصعوبة في البلع؟ BL ما هو البروتوكول العلاجي المناسب ___ مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ ما هو البروتوكول العلاجي المناسب لزيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ ما هو البروتوكول العلاجي المناسب في حالة زيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ HG ما هي الأسباب المحتملة ___ مثل طفح جلدي منتشر مصحوب بحكة شديدة؟ ما هي الأسباب المحتملة لظهور مثل طفح جلدي منتشر مصحوب بحكة شديدة؟ ما هي الأسباب المحتملة لظهور أعراض مستمرة مثل طفح جلدي منتشر مصحوب بحكة شديدة؟ BL ما هو البروتوكول العلاجي ___ تورم القدمين وضعف في الشهية وخمول عام؟ ما هو البروتوكول العلاجي لعلاج تورم القدمين وضعف في الشهية وخمول عام؟ ما هو البروتوكول العلاجي المناسب في حالة تورم القدمين وضعف في الشهية وخمول عام؟ HG ما هي الأسباب المحتملة لظهور ___ حرقة شديدة في المعدة وصعوبة في البلع؟ ما هي الأسباب المحتملة لظهور أعراض حرقة شديدة في المعدة وصعوبة في البلع؟ ما هي الأسباب المحتملة لظهور أعراض مستمرة مثل حرقة شديدة في المعدة وصعوبة في البلع؟ BL ما مدى خطورة استمرار ___ والتعب المزمن وألم في الصدر؟ ما مدى خطورة استمرار الصداع والتعب المزمن وألم في الصدر؟ ما مدى خطورة استمرار ظهور ضيق التنفس والتعب المزمن وألم في الصدر؟ FL ما الإجراءات الطبية المطلوبة ___ مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ ما الإجراءات الطبية المطلوبة لزيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ ما الإجراءات الطبية المطلوبة عند ملاحظة زيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟ HG كيف يمكن تشخيص حالة ___ ضيق التنفس والتعب المزمن وألم في الصدر؟ كيف يمكن تشخيص حالة من ضيق التنفس والتعب المزمن وألم في الصدر؟ كيف يمكن تشخيص حالة مريض يعاني من ضيق التنفس والتعب المزمن وألم في الصدر؟ HG Observation. HG rows show faithful restorations with natural Arabic surface form (often differing only by a preposition or a clinical template such as “في حالة … / عند ملاحظة …”). BL rows primarily miss required scaffolding (“أعراض مستمرة”, “مريض يعاني من …”) or have slightly awkward linking. FL rows exhibit semantic drift, typically substituting an incorrect clinical entity. 4.3 Human Validation To gauge perceived quality, two independent experts—a clinician and a linguist—rated 200 reformulated questions on four five-point scales: Clarity, Fluency, Semantic Fidelity, and Practical Usefulness (1 = poor, 5 = excellent). The raters worked independently; we then averaged their scores. Inter-rater agreement was estimated with weighted Cohen’s κ, with 95% confidence intervals reported for each criterion. Differences between models were tested using the paired Wilcoxon signed-rank test with Benjamini–Hochberg false-discovery-rate correction.[ 20 – 22 ]. Table 7 Human evaluation on 200 items by two independent raters Criterion Mean ± 95% CI Weighted κ (95% CI) Interpretation Clarity 4.12 ± 0.18 0.81 (0.77–0.85) Clear and neatly structured rewrites. Fluency 3.68 ± 0.22 0.78 (0.73–0.82) Grammatically clean, with subtle lexical noise. Semantic Fidelity 3.15 ± 0.25 0.74 (0.70–0.79) Occasional semantic confusion between symptom and diagnosis. Practical Usefulness 3.42 ± 0.21 0.76 (0.72–0.81) Clinically appropriate and generally adoptable. Across the 200 reformulated questions, human ratings indicate that the domain-adapted AraBERT-MLM produces clear, fluent outputs while maintaining acceptable semantic fidelity and practical relevance. Inter-rater agreement was substantial (overall κ ≈ 0.77), reflecting consistent judgments across reviewers. The remaining weaknesses are mostly semantic—occasional swaps between related clinical entities or missing discourse scaffolding. Taken together, these results validate the intrinsic gains and identify AraBERT-MLM as the most reliable option for Arabic clinical question reformulation.. 5 Discussion Domain-adaptive pretraining (DAPT) delivers consistent gains in the linguistic coherence and clinical relevance of Arabic question reformulations. The adapted AraBERT-MLM outperforms both its original baseline and the AraELECTRA models, showing lower perplexity and higher Top-k accuracy across all length-controlled splits. Human reviewers reached the same conclusion: the reformulated questions were judged clearer, more fluent, and closer to the intended clinical meaning. This alignment between automatic metrics and expert judgment indicates that continued masked-language pretraining effectively tunes models to Arabic medical discourse. The remaining shortcomings—minor semantic drift and occasional missing discourse scaffolding—point to combining DAPT with explicit clinical templates or multi-task objectives to further improve contextual precision and naturalness. 6 Conclusion We asked a practical question: can Arabic encoders produce cleaner, more clinically useful questions when exposed to health-domain text? Using the AHQAD corpus for continued masked-language pretraining, the domain-adapted AraBERT-MLM delivered the most reliable gains—lower perplexity (≈ 4.93) and higher macroTop-5 accuracy (≈ 0.839)—and two independent raters reached substantial agreement (κ ≈ 0.77) that its outputs were clearer and more fluent. The remaining issues were limited and concrete (occasional semantic drift and missing discourse cues), suggesting that lexical and stylistic alignment to medical Arabic—rather than architectural changes—drives most of the improvement. In practice, this makes reformulation a measurable, tunable front end for Arabic clinical QA. Next steps are to verify impact in retrieval-based and generative settings, and to test light template guidance or multi-task objectives to stabilize meaning in longer queries and across specialties. Declarations Author Contribution W.O. (Walid Ounachad) conceived the study, designed the methodology, and performed the main experiments and analysis.M.K. (Mohamed Khenchouch) contributed to data preprocessing, experimental setup, and result validation.I.Z. (Imad Zeroual) provided supervision, conceptual guidance, and critical review of the manuscript.Y.F. (Yousef Farhaoui) contributed to the interpretation of results and the refinement of the research framework.W.O. wrote the main manuscript text.All authors discussed the results, contributed to the final version of the manuscript, and approved it for submission. Acknowledgement The authors would like to thank the IMIA Laboratory, Faculty of Sciences and Techniques of Errachidia, Moulay Ismail University of Meknes, for providing the research environment and computational resources that supported this work.The authors also thank all colleagues who provided valuable comments and discussions that improved the quality of the manuscript. Data Availability All data and materials used in this study are publicly available. The AHQAD/AHD datasets are accessible from their original publication, and all code used for preprocessing, model training, and evaluation is available on request from the corresponding author. References Mashaabi, M., Al-Khalifa, S., & Al-Khalifa, H. (2025). A Survey of Large Language Models for Arabic Language and its Dialects. http://arxiv.org/abs/2410.20238 , https://doi.org/10.48550/arXiv.2410.20238 Koto, F., Li, H., Shatnawi, S., Doughman, J., Sadallah, A., Alraeesi, A., Almubarak, K., Alyafeai, Z., Sengupta, N., Shehata, S., Habash, N., Nakov, P., & Baldwin, T. (2024). ArabicMMLU: Assessing Massive Multitask Language Understanding in Arabic. In L. W. Ku, A. Martins, & V. Srikumar (Eds.), Findings of the Association for Computational Linguistics: ACL 2024 (pp. 5622–5640). Association for Computational Linguistics. https://doi.org/10.18653/v1/2024.findings-acl.334 Abdul-Mageed, M., Elmadany, A., & Nagoudi, E. M. B. (2021). ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic. In: Zong, C., Xia, F., Li, W., and Navigli, R. (Eds.) Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). pp. 7088–7105. Association for Computational Linguistics, Online https://doi.org/10.18653/v1/2021.acl-long.551 Lakim, I., Almazrouei, E., Abualhaol, I., Debbah, M., & Launay, J. (2022). A Holistic Assessment of the Carbon Footprint of Noor, a Very Large Arabic Language Model. In: Fan, A., Ilic, S., Wolf, T., and Gallé, M. (Eds.) Proceedings of BigScience Episode #5 – Workshop on Challenges & Perspectives in Creating Large Language Models. pp. 84–94. Association for Computational Linguistics, virtual + Dublin https://doi.org/10.18653/v1/2022.bigscience-1.8 Al-Majmar, N. A., Gawbah, H., & Alsubari, A. (2024). AHD: Arabic healthcare dataset. Data in Brief 56, 110855 https://doi.org/10.1016/j.dib.2024.110855 . Salazar, J., Liang, D., Nguyen, T. Q., & Kirchhoff, K. (2020). Masked Language Model Scoring. In: Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (Eds.) Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 2699–2712. Association for Computational Linguistics, Online https://doi.org/10.18653/v1/2020.acl-main.240 Krippendorff, K. (2019). Content Analysis: An Introduction to Its Methodology . SAGE Publications, Inc. https://doi.org/10.4135/9781071878781 Abdallah, A., Kasem, M., Abdalla, M., Mahmoud, M., Elkasaby, M., Elbendary, Y., & Jatowt, A. (2024). ArabicaQA: A Comprehensive Dataset for Arabic Question Answering. In: Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. pp. 2049–2059. Association for Computing Machinery, New York, NY, USA https://doi.org/10.1145/3626772.3657889 Gururangan, S., Marasović, A., Swayamdipta, S., Lo, K., Beltagy, I., Downey, D., & Smith, N. A. (2020). Don’t Stop Pretraining: Adapt Language Models to Domains and Tasks. In: Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (Eds.) Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 8342–8360. Association for Computational Linguistics, Online https://doi.org/10.18653/v1/2020.acl-main.740 Kauf, C., & Ivanova, A. (2023). A Better Way to Do Masked Language Model Scoring. http://arxiv.org/abs/2305.10588 , https://doi.org/10.48550/arXiv.2305.10588 Face, A. D. P. R. H. https://huggingface.co/abdoelsayed/AraDPR , last accessed 2025/10/15. Alhuzali, H., Alasmari, A., & Alsaleh, H. (2024). MentalQA: An Annotated Arabic Corpus for Questions and Answers of Mental Healthcare. http://arxiv.org/abs/2405.12619 , https://doi.org/10.48550/arXiv.2405.12619 Alasmari, A., Alhumoud, S., & Alshammari, W. (2024). AraMed: Arabic Medical Question Answering using Pretrained Transformer Language Models. In: Al-Khalifa, H., Darwish, K., Mubarak, H., Ali, M., and Elsayed, T. (Eds.) Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024. pp. 50–56. ELRA and ICCL, Torino, Italia. Gwet, K. L. (2014). Handbook of inter-rater reliability: the definitive guide to measuring the extent of agreement among raters . Advances Analytics, LLC. Hegazi, M. O., Al-Dossari, Y., Al-Yahy, A., Al-Sumari, A., & Hilal, A. (2021). Preprocessing Arabic text on social media. Heliyon , 7 , e06191. https://doi.org/10.1016/j.heliyon.2021.e06191 Loshchilov, I., & Hutter, F. (2019). Decoupled Weight Decay Regularization. http://arxiv.org/abs/1711.05101 , https://doi.org/10.48550/arXiv.1711.05101 Zhao, C., Wang, X., Huang, Y., Lu, J., & Liu, Z. (2025). TASE: Token Awareness and Structured Evaluation for Multilingual Language Models. http://arxiv.org/abs/2508.05468 , https://doi.org/10.48550/arXiv.2508.05468 Benjamini, Y., & Yekutieli, D. (2001). The control of the false discovery rate in multiple testing under dependency. The Annals of Statistics , 29 , 1165–1188. https://doi.org/10.1214/aos/1013699998 Storey, J. D. (2002). A direct approach to false discovery rates. Journal of the Royal Statistical Society: Series B (Statistical Methodology) , 64 , 479–498. https://doi.org/10.1111/1467-9868.00346 Tam, T. Y. C., Sivarajkumar, S., Kapoor, S., Stolyar, A. V., Polanska, K., McCarthy, K. R., Osterhoudt, H., Wu, X., Visweswaran, S., Fu, S., Mathur, P., Cacciamani, G. E., Sun, C., Peng, Y., & Wang, Y. (2024). A framework for human evaluation of large language models in healthcare derived from literature review. npj Digit Med , 7 , 258. https://doi.org/10.1038/s41746-024-01258-7 van der Lee, C., Gatt, A., van Miltenburg, E., & Krahmer, E. (2021). Human evaluation of automatically generated text: Current trends and best practice guidelines. Computer Speech & Language , 67 , 101151. https://doi.org/10.1016/j.csl.2020.101151 Thomson, C., Reiter, E., & Belz, A. (2024). Common Flaws in Running Human Evaluation Experiments in NLP. Computational Linguistics , 50 , 795–805. https://doi.org/10.1162/coli_a_00508 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 15 Jan, 2026 Reviews received at journal 04 Dec, 2025 Reviews received at journal 01 Dec, 2025 Reviews received at journal 28 Nov, 2025 Reviews received at journal 16 Nov, 2025 Reviewers agreed at journal 14 Nov, 2025 Reviewers agreed at journal 11 Nov, 2025 Reviewers agreed at journal 10 Nov, 2025 Reviewers agreed at journal 10 Nov, 2025 Reviewers agreed at journal 10 Nov, 2025 Reviewers agreed at journal 10 Nov, 2025 Reviewers invited by journal 10 Nov, 2025 Editor assigned by journal 10 Nov, 2025 Submission checks completed at journal 07 Nov, 2025 First submitted to journal 01 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8007820","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":545886916,"identity":"16c3f6bc-4cba-45f5-9d61-1af9c5e33e70","order_by":0,"name":"Walid Ounachad","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9klEQVRIiWNgGAWjYBACgwMQOoGBIfkAAw+IyUy8lrQEkrXkGEC0EHYY8zHpghq7PP72nI8P3rYxJK5tZ2D88INhmzwuLWYH2NKkZxxLLpY483az4Vyglm2HGZglexhuGzbg1MJjJs3DdiCx4UbuNmleiBYGaQaG24z4tfw7kDj/Rs7z31AtzL+BWuxxabEHaeFtO5C44UYOGzNUCxvIlkRcWiwPsyVb8/YlJ24888xYcs45CeNthxnbLHsMbifj0mJwvPngbZ5vdonzjic//PCmzEZ22/nDh2/8qLhti0sLesRJADHI4wa41I+CUTAKRsEoIAYAAM0lWrM46E0ZAAAAAElFTkSuQmCC","orcid":"","institution":"Faculty of Sciences And Techniques of Errachidia, Moulay Ismail University of Meknes","correspondingAuthor":true,"prefix":"","firstName":"Walid","middleName":"","lastName":"Ounachad","suffix":""},{"id":545886917,"identity":"a1028502-1757-433d-8fa0-258da30a6307","order_by":1,"name":"Mohamed Khenchouch","email":"","orcid":"","institution":"Faculty of Sciences And Techniques of Errachidia, Moulay Ismail University of Meknes","correspondingAuthor":false,"prefix":"","firstName":"Mohamed","middleName":"","lastName":"Khenchouch","suffix":""},{"id":545886918,"identity":"0b8ba980-acb1-4a05-aefd-73b1ca077e18","order_by":2,"name":"Imad Zeroual","email":"","orcid":"","institution":"Faculty of Sciences And Techniques of Errachidia, Moulay Ismail University of Meknes","correspondingAuthor":false,"prefix":"","firstName":"Imad","middleName":"","lastName":"Zeroual","suffix":""},{"id":545886919,"identity":"cf2965c5-3bc2-47e8-b669-b3a1d226115e","order_by":3,"name":"Yousef Farhaoui","email":"","orcid":"","institution":"Faculty of Sciences And Techniques of Errachidia, Moulay Ismail University of Meknes","correspondingAuthor":false,"prefix":"","firstName":"Yousef","middleName":"","lastName":"Farhaoui","suffix":""}],"badges":[],"createdAt":"2025-11-01 20:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8007820/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8007820/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96320039,"identity":"4eb4fbdd-e04f-4635-a317-06a681ab68b1","added_by":"auto","created_at":"2025-11-19 18:40:00","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82166,"visible":true,"origin":"","legend":"","description":"","filename":"ImprovingArabicClinicalQuestionQualitythroughDomainAdaptiveMaskedLanguageModeling.docx","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/d49aef20426ec9ae74554170.docx"},{"id":96320037,"identity":"7b7e6335-7813-464f-a1ae-2ac6808a81c3","added_by":"auto","created_at":"2025-11-19 18:40:00","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6888,"visible":true,"origin":"","legend":"","description":"","filename":"deeb8e15283747168a81f3728bf578da.json","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/64a0585e8e4e9bf4e731c7de.json"},{"id":96320038,"identity":"15f2364d-f4dd-482e-ac61-d9b0ac5be636","added_by":"auto","created_at":"2025-11-19 18:40:00","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96303,"visible":true,"origin":"","legend":"","description":"","filename":"deeb8e15283747168a81f3728bf578da1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/875dc61477c6ced9c06276e3.xml"},{"id":96320040,"identity":"c36d7c61-c489-4a61-8b54-d8153146a414","added_by":"auto","created_at":"2025-11-19 18:40:01","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":93304,"visible":true,"origin":"","legend":"","description":"","filename":"deeb8e15283747168a81f3728bf578da1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/01c356c806b1289e91f83c73.xml"},{"id":96366768,"identity":"bc39de2a-7b9e-4388-84bb-f773ac9a3da2","added_by":"auto","created_at":"2025-11-20 10:11:53","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":93804,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/2f4d77936cb5ed43b8045050.html"},{"id":96453511,"identity":"6c5f7df3-ace3-4e40-8e0c-6c1e0a69c447","added_by":"auto","created_at":"2025-11-21 10:00:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":908455,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8007820/v1/0aed37d5-4b00-4f33-8fcc-60f81d4838cf.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Improving Arabic Clinical Question Quality through Domain-Adaptive Masked Language Modeling","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eWriting well-formed questions in Arabic remains challenging for NLP systems, with persistent gaps between Modern Standard Arabic (MSA) and dialects highlighted by recent surveys and benchmarks [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. In semantic search, virtual assistants, and QA, short, elliptical, or loosely structured queries often produce off-target answers\u0026mdash;even with strong encoders such as ARBERT/MARBERT [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Part of the difficulty is linguistic\u0026mdash;spelling variation, rich morphology, and the MSA\u0026ndash;dialect gap\u0026mdash;and part of it is practical: few curated resources spell out what a \u0026ldquo;good\u0026rdquo; Arabic question\u0026rdquo; looks like, particularly in clinical settings. Scaling efforts in large Arabic LLMs also underscore opportunities and constraints relevant to clinical scenarios [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eWe address this problem at its source through domain-specific adaptive pretraining for Arabic question reformulation. Using AHQAD/AHD, a large corpus of Arabic health question\u0026ndash;answer pairs [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], we continue pretraining two widely used backbones\u0026mdash;AraBERT and the generator variant of AraELECTRA\u0026mdash;under a masked-language objective so they better capture the structure of well-formed medical questions. To test models under realistic conditions, we build length-balanced test sets and, for each item, create a truncated version that must be reconstructed into a coherent, context-appropriate question.\u003c/p\u003e\u003cp\u003eOur evaluation mirrors the training signal: we compute cross-entropy and perplexity at masked token positions and measure top-k accuracy only where masking occurs, following token-aligned scoring for masked language models [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. We also conduct a human study with one language specialist and one domain expert who rate clarity, fluency, semantic fidelity, and practical usefulness of the reformulated questions; inter-rater agreement is summarized with modern reliability coefficients [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTaken together, these components provide a reusable path to improving Arabic question formulation: large-scale domain adaptation, measurements consistent with the learning objective and controlled for length, and expert validation that links numerical gains to judged quality. The resulting module can be plugged into Arabic clinical QA pipelines [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] and treated as a measurable, tunable component rather than an assumption.\u003c/p\u003e\u003cp\u003e\u003cb\u003eKey Contributions.\u003c/b\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMethod\u003c/b\u003e. A reproducible framework for Arabic question reformulation via domain-adaptive (continued) pretraining on AHQAD with the MLM objective.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eEvaluation design\u003c/b\u003e. Token-level metrics aligned with MLM (cross-entropy, perplexity, top-k on masked positions), length-controlled splits, and lexical-diversity tracking.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eHuman validation\u003c/b\u003e. Expert review (linguistics\u0026thinsp;+\u0026thinsp;domain) assessing clarity, fluency, semantic fidelity, and usefulness alongside automatic metrics.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003ePracticality\u003c/b\u003e. A reformulation module that integrates cleanly into Arabic clinical QA pipelines, turning question quality into a measurable, tunable target.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e"},{"header":"2 Related Work","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Arabic pretrained encoders\u003c/h2\u003e\u003cp\u003eArabic-specific encoders consistently outperform multilingual baselines across diverse NLU tasks [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. AraBERT adapts BERT\u0026rsquo;s Masked Language Modeling (MLM) objective to Arabic and established early state-of-the-art performance on common benchmarks. AraELECTRA brings ELECTRA\u0026rsquo;s replaced-token detection to Arabic, offering a strong efficiency\u0026ndash;accuracy trade-off. Larger families such as ARBERT/MARBERT expand coverage to dialectal and social-media text, underscoring the importance of scale and domain breadth in Arabic pretraining. More recent initiatives like AraT5 extend this line to generative, text-to-text formulations [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], while large-scale surveys consolidate the progress and remaining challenges in Arabic LLMs [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Domain-adaptive pretraining (DAPT/TAPT)\u003c/h2\u003e\u003cp\u003eA second pretraining phase on in-domain text reliably improves downstream performance. [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] formalized Domain-Adaptive Pretraining (DAPT) and Task-Adaptive Pretraining (TAPT), showing consistent gains across domains and data regimes. Subsequent work confirms that adaptation to task-specific corpora reduces cross-domain drift and enhances generalization [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. We follow this paradigm by adapting Arabic encoders on QA-style health data prior to evaluation, aligning pretraining distribution with target clinical use cases.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 MLM objectives and token-level evaluation\u003c/h2\u003e\u003cp\u003eBERT-style Masked Language Modeling (MLM) motivates token-aligned evaluation rather than sequence-only scores. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] introduced pseudo-log-likelihood (PLL) scoring to obtain token-level probabilities from MLMs, and subsequent studies refined PLL for greater theoretical consistency [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. This motivates our use of cross-entropy and perplexity at masked positions, top-k accuracy restricted to masked tokens, and lexical-diversity tracking to avoid formulaic phrasing.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Arabic QA resources (recent and large-scale)\u003c/h2\u003e\u003cp\u003eArabic QA has evolved from early reading comprehension datasets to large, contemporary corpora. ArabicaQA [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] provides 89 k\u0026thinsp;+\u0026thinsp;MRC questions, an open-domain QA benchmark, and a dense retriever (AraDPR) forming a modern testbed for readers and retrievers [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. In the health domain, AHQAD/AHD [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] aggregate approximately 808 000 Arabic Q\u0026ndash;A pairs across nearly 90 specialties, enabling large-scale domain adaptation and probing. Together, these resources make Arabic-specific DAPT feasible and closer to real-world distributions [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Reformulation and human evaluation in Arabic NLP\u003c/h2\u003e\u003cp\u003eCompared to English, Arabic question reformulation remains underexplored\u0026mdash;most prior work targets encoder or reader architectures without isolating the effect of query form. Recent interest in Arabic clinical and general-domain LLMs has emphasized pretraining and reasoning, but systematic human studies assessing clarity, fluency, and semantic fidelity remain scarce [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. We complement prior efforts by (i) performing DAPT on QA-style medical data to directly improve question form, and (ii) conducting a two-expert human evaluation aligned with intrinsic MLM-consistent metrics [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] .\u003c/p\u003e\u003c/div\u003e"},{"header":"3 Methodology","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Problem Setting\u003c/h2\u003e\u003cp\u003eWe address the problem of reformulating Arabic clinical questions so that they are complete, clear, and clinically useful[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Starting from a natural question, we create a \u0026ldquo;truncated\u0026rdquo; version by hiding one or more informative spans (e.g., a symptom, a duration, a medication, or discourse cues). The system\u0026rsquo;s task is to reconstruct a well-formed question that preserves the original intent and improves readability and completeness. This setup mirrors real interactions where user queries are often short, elliptical, or underspecified.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Data and Pre-processing\u003c/h2\u003e\u003cp\u003eCorpus. We use AHQAD/AHD, a large Arabic health question\u0026ndash;answer collection (approximately 808,000 pairs across ~\u0026thinsp;90 specialties)[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The corpus provides realistic clinical phrasing and domain terminology.\u003c/p\u003e\u003cp\u003eCleaning and normalization. We remove duplicates, normalize Unicode, strip optional diacritics, collapse elongation marks and extra spaces, and apply light orthographic normalization (alif/hamza and yaa/maqṣūra variants). Non-Arabic script is discarded. We also apply length filtering to reduce extremes and noise[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eEvaluation set. We build a controlled evaluation set of 600 questions stratified by length into three balanced buckets (Short, Long, Very Long; 200 each). For every fully written \u0026ldquo;reference\u0026rdquo; question, we create a truncated counterpart by removing an informative span and inserting a placeholder.\u003c/p\u003e\u003cp\u003eLength control. All reporting is provided per length bucket with macro averages to separate perceived quality from verbosity[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Domain-Adaptive Pretraining\u003c/h2\u003e\u003cp\u003eWe perform domain-adaptive (continued) pretraining on Arabic clinical text using the masked-language-modeling objective. A fixed proportion of tokens is selected for masking; among those, most are replaced with a mask symbol, a smaller share is replaced with a random token, and the remainder is left unchanged. This standard policy encourages the model to use both sides of the context and to learn domain-specific lexical and structural patterns[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eBackbones. We adapt two widely used Arabic encoders: AraBERT (BERT-style) and the generator variant of AraELECTRA. For each, we compare the original public checkpoint with a domain-adapted variant obtained by continued pretraining on AHQAD[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eTraining setup. We use AdamW, standard warmup and linear decay, mixed precision when available, early stopping on development performance, and three random seeds. Sequence length and batch size are tuned within practical ranges; we log learning curves and keep the best development checkpoint[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003e3.4 Reformulation Inference\u003c/h2\u003e\u003cp\u003eAt inference time, we do not free-generate entire questions. Instead, we replace each placeholder in the truncated question by one or a short span of mask tokens and fill only those masked slots with the domain-adapted model. Decoding is greedy (Top-1) or Top-k and is restricted strictly to the masked positions[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. We then detokenize, normalize, and fix whitespace and punctuation. This constrained procedure keeps inference consistent with the training signal and limits semantic drift.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.5 Evaluation Protocol\u003c/h2\u003e\u003cp\u003eIntrinsic, token-aligned metrics. We evaluate only where masking occurs, to stay faithful to the learning objective. We report cross-entropy and perplexity at masked positions (lower is better), Top-k accuracy at masked positions (higher is better), and lexical diversity through the type\u0026ndash;token ratio, complemented by a moving-average variant to reduce length bias. Results are reported per length bucket and as macro averages[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eHuman evaluation. Two experts\u0026mdash;a linguist and a clinician\u0026mdash;independently rate each reconstructed question for clarity, fluency, semantic fidelity, and practical usefulness on five-point scales. We report per-criterion means with 95% confidence intervals and inter-rater agreement (weighted kappa). Items are randomized and raters are blinded to model identity[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eStatistical testing. Within each backbone, we run paired, non-parametric tests comparing original and domain-adapted variants across metrics and buckets, with false-discovery-rate correction for multiple comparisons. We also report an effect size for the primary endpoint [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e3.6 Implementation and Reproducibility\u003c/h2\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003ePlatform. Experiments are conducted on Google Colab Pro\u0026thinsp;+\u0026thinsp;with A100/V100/T4 GPUs depending on the session, at least 25 GB RAM, and sufficient ephemeral storage. We checkpoint frequently to mitigate session resets.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSoftware. Python 3.10; PyTorch; Hugging Face Transformers, Datasets, and Accelerate; PEFT when needed. Mixed precision uses bf16 on A100 and fp16 on V100/T4. Tokenizer parallelism is disabled to reduce non-determinism.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eRandomness control. We fix three seeds and enable deterministic settings when supported.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eReleased artifacts. We provide normalization and truncation scripts, train/dev/test splits with cryptographic hashes, evaluation masks, training logs, and both final and best-development checkpoints.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eEthics and intended use. The module reformulates questions; it does not produce diagnoses or medical advice. Data are public and anonymized. We recommend human-in-the-loop use in clinical settings.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4 Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e4.1 Intrinsic results by model\u003c/h2\u003e\u003cp\u003eAll scores are computed exclusively at masked positions and reported by length (Short / Long / Very Long) plus a macro average. Notation: \u0026darr; (CE, PPL)\u0026thinsp;=\u0026thinsp;lower is better; \u0026uarr; (Top-k, TTR)\u0026thinsp;=\u0026thinsp;higher is better.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAraBERT (original)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLength\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPL (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCE (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTop-1 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTop-3 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTop-5 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTTR (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eShort\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7.695\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.040\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.564\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.763\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.801\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.145\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLong\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8.182\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.101\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.616\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.735\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.784\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.030\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVery Long\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7.050\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.953\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.587\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.738\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.789\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.032\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMacro Avg\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7.642\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.031\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.589\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.745\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.791\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAraELECTRA (original)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLength\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPL (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCE (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTop-1 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTop-3 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTop-5 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTTR (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eShort\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e120.511\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4.791\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.521\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.650\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.682\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.145\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLong\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e184.893\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e5.219\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.498\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.621\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.670\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.030\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVery Long\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e434.491\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e6.074\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.461\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.597\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.647\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.032\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMacro Avg\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e246.631\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e5.361\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0493\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.623\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.667\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAraELECTRA-MLM (DAPT)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLength\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPL (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCE (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTop-1 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTop-3 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTop-5 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTTR (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eShort\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6.026\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.796\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.597\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.801\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.849\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.145\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLong\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e7.45\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.008\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.574\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.741\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.779\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.030\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVery Long\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e10.911\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e2.389\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.529\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.685\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.734\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.032\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMacro Avg\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e8.129\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e3.064\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0567\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.742\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.742\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eAraBERT-MLM (DAPT)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLength\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePPL (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eCE (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTop-1 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTop-3 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTop-5 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTTR (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eShort\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e4.488\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.501\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.645\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.806\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.860\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.145\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLong\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e4.177\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.429\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.712\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.797\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.859\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.030\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVery Long\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e6.138\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.814\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.629\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.759\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.797\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.032\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMacro Avg\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e4.934\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.581\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.662\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.787\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.839\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMacro-average comparison\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCE (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePPL (\u0026darr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eTop-1 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTop-3 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eTop-5 (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eTTR (\u0026uarr;)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAraBERT (orig.)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2.031\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e7.642\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.589\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.745\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.791\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAraELECTRA (orig.)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e5.361\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e246.63\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.493\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.623\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.667\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAraELECTRA-MLM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2.064\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e8.129\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.567\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.742\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.787\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAraBERT-MLM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.581\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e4.934\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.662\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.787\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e\u003cp\u003e0.839\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e\u003cp\u003e0.069\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eKey takeaway.\u003c/b\u003e Domain-adaptive pretraining (DAPT) yields consistent gains; AraBERT-MLM is best overall (highest macro-Top-5 0.839, lowest CE/PPL), with the strongest advantage on short items.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e4.2 Qualitative analysis (before \u0026rarr; output \u0026rarr; reference)\u003c/h2\u003e\u003cp\u003eTo complement the intrinsic metrics, we report representative triplets that illustrate constrained fill-ins at masked spans. Each triplet comprises the truncated prompt (before), the model\u0026rsquo;s reformulation confined to the masked slots (output), and the fully specified reference question. We label cases High-Gain (HG) when the output preserves the clinical intent with only superficial differences, Borderline (BL) when the intent is mostly preserved but discourse scaffolding or phrasing is imperfect, and Failure (FL) when key semantics drift\u0026mdash;for example, substituting a diagnosis for a symptom (or vice versa).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eQualitative examples: before \u0026rarr; model output \u0026rarr; reference.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eQ_before\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eQ-generated (model)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e\u003cp\u003eQ_after(reference)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eVerdict\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما العلاقة بين ___ في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما العلاقة بين ألم في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما العلاقة بين حرقة شديدة في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eBL\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما هو البروتوكول العلاجي المناسب ___ مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما هو البروتوكول العلاجي المناسب لزيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما هو البروتوكول العلاجي المناسب في حالة زيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eHG\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما هي الأسباب المحتملة ___ مثل طفح جلدي منتشر مصحوب بحكة شديدة؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما هي الأسباب المحتملة لظهور مثل طفح جلدي منتشر مصحوب بحكة شديدة؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما هي الأسباب المحتملة لظهور أعراض مستمرة مثل طفح جلدي منتشر مصحوب بحكة شديدة؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eBL\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما هو البروتوكول العلاجي ___ تورم القدمين وضعف في الشهية وخمول عام؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما هو البروتوكول العلاجي لعلاج تورم القدمين وضعف في الشهية وخمول عام؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما هو البروتوكول العلاجي المناسب في حالة تورم القدمين وضعف في الشهية وخمول عام؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eHG\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما هي الأسباب المحتملة لظهور ___ حرقة شديدة في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما هي الأسباب المحتملة لظهور أعراض حرقة شديدة في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما هي الأسباب المحتملة لظهور أعراض مستمرة مثل حرقة شديدة في المعدة وصعوبة في البلع؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eBL\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما مدى خطورة استمرار ___ والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما مدى خطورة استمرار الصداع والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما مدى خطورة استمرار ظهور ضيق التنفس والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eFL\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eما الإجراءات الطبية المطلوبة ___ مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eما الإجراءات الطبية المطلوبة لزيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eما الإجراءات الطبية المطلوبة عند ملاحظة زيادة مفاجئة في الوزن واضطرابات في الدورة الشهرية؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eHG\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eكيف يمكن تشخيص حالة ___ ضيق التنفس والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eكيف يمكن تشخيص حالة من ضيق التنفس والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eكيف يمكن تشخيص حالة مريض يعاني من ضيق التنفس والتعب المزمن وألم في الصدر؟\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eHG\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eObservation.\u003c/b\u003e HG rows show faithful restorations with natural Arabic surface form (often differing only by a preposition or a clinical template such as \u0026ldquo;في حالة \u0026hellip; / عند ملاحظة \u0026hellip;\u0026rdquo;). BL rows primarily miss required scaffolding (\u0026ldquo;أعراض مستمرة\u0026rdquo;, \u0026ldquo;مريض يعاني من \u0026hellip;\u0026rdquo;) or have slightly awkward linking. FL rows exhibit semantic drift, typically substituting an incorrect clinical entity.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e4.3 Human Validation\u003c/h2\u003e\u003cp\u003eTo gauge perceived quality, two independent experts\u0026mdash;a clinician and a linguist\u0026mdash;rated 200 reformulated questions on four five-point scales: Clarity, Fluency, Semantic Fidelity, and Practical Usefulness (1\u0026thinsp;=\u0026thinsp;poor, 5\u0026thinsp;=\u0026thinsp;excellent). The raters worked independently; we then averaged their scores. Inter-rater agreement was estimated with weighted Cohen\u0026rsquo;s κ, with 95% confidence intervals reported for each criterion. Differences between models were tested using the paired Wilcoxon signed-rank test with Benjamini\u0026ndash;Hochberg false-discovery-rate correction.[\u003cspan additionalcitationids=\"CR21\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab7\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 7\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eHuman evaluation on 200 items by two independent raters\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCriterion\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMean\u0026thinsp;\u0026plusmn;\u0026thinsp;95% CI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eWeighted κ (95% CI)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eInterpretation\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eClarity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e4.12\u0026thinsp;\u0026plusmn;\u0026thinsp;0.18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.81 (0.77\u0026ndash;0.85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eClear and neatly structured rewrites.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFluency\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.22\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.78 (0.73\u0026ndash;0.82)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eGrammatically clean, with subtle lexical noise.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSemantic Fidelity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.15\u0026thinsp;\u0026plusmn;\u0026thinsp;0.25\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.74 (0.70\u0026ndash;0.79)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eOccasional semantic confusion between symptom and diagnosis.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePractical Usefulness\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.42\u0026thinsp;\u0026plusmn;\u0026thinsp;0.21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.76 (0.72\u0026ndash;0.81)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eClinically appropriate and generally adoptable.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAcross the 200 reformulated questions, human ratings indicate that the domain-adapted AraBERT-MLM produces clear, fluent outputs while maintaining acceptable semantic fidelity and practical relevance. Inter-rater agreement was substantial (overall κ\u0026thinsp;\u0026asymp;\u0026thinsp;0.77), reflecting consistent judgments across reviewers. The remaining weaknesses are mostly semantic\u0026mdash;occasional swaps between related clinical entities or missing discourse scaffolding. Taken together, these results validate the intrinsic gains and identify AraBERT-MLM as the most reliable option for Arabic clinical question reformulation..\u003c/p\u003e\u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cp\u003eDomain-adaptive pretraining (DAPT) delivers consistent gains in the linguistic coherence and clinical relevance of Arabic question reformulations. The adapted AraBERT-MLM outperforms both its original baseline and the AraELECTRA models, showing lower perplexity and higher Top-k accuracy across all length-controlled splits. Human reviewers reached the same conclusion: the reformulated questions were judged clearer, more fluent, and closer to the intended clinical meaning. This alignment between automatic metrics and expert judgment indicates that continued masked-language pretraining effectively tunes models to Arabic medical discourse. The remaining shortcomings\u0026mdash;minor semantic drift and occasional missing discourse scaffolding\u0026mdash;point to combining DAPT with explicit clinical templates or multi-task objectives to further improve contextual precision and naturalness.\u003c/p\u003e"},{"header":"6 Conclusion","content":"\u003cp\u003eWe asked a practical question: can Arabic encoders produce cleaner, more clinically useful questions when exposed to health-domain text? Using the AHQAD corpus for continued masked-language pretraining, the domain-adapted AraBERT-MLM delivered the most reliable gains\u0026mdash;lower perplexity (\u0026asymp;\u0026thinsp;4.93) and higher macroTop-5 accuracy (\u0026asymp;\u0026thinsp;0.839)\u0026mdash;and two independent raters reached substantial agreement (κ\u0026thinsp;\u0026asymp;\u0026thinsp;0.77) that its outputs were clearer and more fluent. The remaining issues were limited and concrete (occasional semantic drift and missing discourse cues), suggesting that lexical and stylistic alignment to medical Arabic\u0026mdash;rather than architectural changes\u0026mdash;drives most of the improvement. In practice, this makes reformulation a measurable, tunable front end for Arabic clinical QA. Next steps are to verify impact in retrieval-based and generative settings, and to test light template guidance or multi-task objectives to stabilize meaning in longer queries and across specialties.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eW.O. (Walid Ounachad) conceived the study, designed the methodology, and performed the main experiments and analysis.M.K. (Mohamed Khenchouch) contributed to data preprocessing, experimental setup, and result validation.I.Z. (Imad Zeroual) provided supervision, conceptual guidance, and critical review of the manuscript.Y.F. (Yousef Farhaoui) contributed to the interpretation of results and the refinement of the research framework.W.O. wrote the main manuscript text.All authors discussed the results, contributed to the final version of the manuscript, and approved it for submission.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors would like to thank the IMIA Laboratory, Faculty of Sciences and Techniques of Errachidia, Moulay Ismail University of Meknes, for providing the research environment and computational resources that supported this work.The authors also thank all colleagues who provided valuable comments and discussions that improved the quality of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data and materials used in this study are publicly available. The AHQAD/AHD datasets are accessible from their original publication, and all code used for preprocessing, model training, and evaluation is available on request from the corresponding author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMashaabi, M., Al-Khalifa, S., \u0026amp; Al-Khalifa, H. (2025). A Survey of Large Language Models for Arabic Language and its Dialects. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2410.20238\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2410.20238\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2410.20238\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2410.20238\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKoto, F., Li, H., Shatnawi, S., Doughman, J., Sadallah, A., Alraeesi, A., Almubarak, K., Alyafeai, Z., Sengupta, N., Shehata, S., Habash, N., Nakov, P., \u0026amp; Baldwin, T. (2024). ArabicMMLU: Assessing Massive Multitask Language Understanding in Arabic. In L. W. Ku, A. Martins, \u0026amp; V. Srikumar (Eds.), \u003cem\u003eFindings of the Association for Computational Linguistics: ACL 2024\u003c/em\u003e (pp. 5622\u0026ndash;5640). Association for Computational Linguistics. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2024.findings-acl.334\u003c/span\u003e\u003cspan address=\"10.18653/v1/2024.findings-acl.334\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbdul-Mageed, M., Elmadany, A., \u0026amp; Nagoudi, E. M. B. (2021). ARBERT \u0026amp; MARBERT: Deep Bidirectional Transformers for Arabic. In: Zong, C., Xia, F., Li, W., and Navigli, R. (Eds.) Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). pp. 7088\u0026ndash;7105. Association for Computational Linguistics, Online \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2021.acl-long.551\u003c/span\u003e\u003cspan address=\"10.18653/v1/2021.acl-long.551\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLakim, I., Almazrouei, E., Abualhaol, I., Debbah, M., \u0026amp; Launay, J. (2022). A Holistic Assessment of the Carbon Footprint of Noor, a Very Large Arabic Language Model. In: Fan, A., Ilic, S., Wolf, T., and Gall\u0026eacute;, M. (Eds.) Proceedings of BigScience Episode #5 \u0026ndash; Workshop on Challenges \u0026amp; Perspectives in Creating Large Language Models. pp. 84\u0026ndash;94. Association for Computational Linguistics, virtual\u0026thinsp;+\u0026thinsp;Dublin \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2022.bigscience-1.8\u003c/span\u003e\u003cspan address=\"10.18653/v1/2022.bigscience-1.8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAl-Majmar, N. A., Gawbah, H., \u0026amp; Alsubari, A. (2024). AHD: Arabic healthcare dataset. \u003cem\u003eData in Brief\u003c/em\u003e 56, 110855 \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.dib.2024.110855\u003c/span\u003e\u003cspan address=\"10.1016/j.dib.2024.110855\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalazar, J., Liang, D., Nguyen, T. Q., \u0026amp; Kirchhoff, K. (2020). Masked Language Model Scoring. In: Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (Eds.) Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 2699\u0026ndash;2712. Association for Computational Linguistics, Online \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2020.acl-main.240\u003c/span\u003e\u003cspan address=\"10.18653/v1/2020.acl-main.240\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKrippendorff, K. (2019). \u003cem\u003eContent Analysis: An Introduction to Its Methodology\u003c/em\u003e. SAGE Publications, Inc. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4135/9781071878781\u003c/span\u003e\u003cspan address=\"10.4135/9781071878781\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAbdallah, A., Kasem, M., Abdalla, M., Mahmoud, M., Elkasaby, M., Elbendary, Y., \u0026amp; Jatowt, A. (2024). ArabicaQA: A Comprehensive Dataset for Arabic Question Answering. In: Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. pp. 2049\u0026ndash;2059. Association for Computing Machinery, New York, NY, USA \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/3626772.3657889\u003c/span\u003e\u003cspan address=\"10.1145/3626772.3657889\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGururangan, S., Marasović, A., Swayamdipta, S., Lo, K., Beltagy, I., Downey, D., \u0026amp; Smith, N. A. (2020). Don\u0026rsquo;t Stop Pretraining: Adapt Language Models to Domains and Tasks. In: Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (Eds.) Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. pp. 8342\u0026ndash;8360. Association for Computational Linguistics, Online \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18653/v1/2020.acl-main.740\u003c/span\u003e\u003cspan address=\"10.18653/v1/2020.acl-main.740\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKauf, C., \u0026amp; Ivanova, A. (2023). A Better Way to Do Masked Language Model Scoring. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2305.10588\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2305.10588\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2305.10588\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2305.10588\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFace, A. D. P. R. H. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://huggingface.co/abdoelsayed/AraDPR\u003c/span\u003e\u003cspan address=\"https://huggingface.co/abdoelsayed/AraDPR\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, last accessed 2025/10/15.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlhuzali, H., Alasmari, A., \u0026amp; Alsaleh, H. (2024). MentalQA: An Annotated Arabic Corpus for Questions and Answers of Mental Healthcare. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2405.12619\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2405.12619\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2405.12619\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2405.12619\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlasmari, A., Alhumoud, S., \u0026amp; Alshammari, W. (2024). AraMed: Arabic Medical Question Answering using Pretrained Transformer Language Models. In: Al-Khalifa, H., Darwish, K., Mubarak, H., Ali, M., and Elsayed, T. (Eds.) Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024. pp. 50\u0026ndash;56. ELRA and ICCL, Torino, Italia.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGwet, K. L. (2014). \u003cem\u003eHandbook of inter-rater reliability: the definitive guide to measuring the extent of agreement among raters\u003c/em\u003e. Advances Analytics, LLC.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHegazi, M. O., Al-Dossari, Y., Al-Yahy, A., Al-Sumari, A., \u0026amp; Hilal, A. (2021). Preprocessing Arabic text on social media. \u003cem\u003eHeliyon\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e, e06191. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.heliyon.2021.e06191\u003c/span\u003e\u003cspan address=\"10.1016/j.heliyon.2021.e06191\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLoshchilov, I., \u0026amp; Hutter, F. (2019). Decoupled Weight Decay Regularization. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/1711.05101\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/1711.05101\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.1711.05101\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1711.05101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhao, C., Wang, X., Huang, Y., Lu, J., \u0026amp; Liu, Z. (2025). TASE: Token Awareness and Structured Evaluation for Multilingual Language Models. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://arxiv.org/abs/2508.05468\u003c/span\u003e\u003cspan address=\"http://arxiv.org/abs/2508.05468\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2508.05468\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2508.05468\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBenjamini, Y., \u0026amp; Yekutieli, D. (2001). The control of the false discovery rate in multiple testing under dependency. \u003cem\u003eThe Annals of Statistics\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e, 1165\u0026ndash;1188. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1214/aos/1013699998\u003c/span\u003e\u003cspan address=\"10.1214/aos/1013699998\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStorey, J. D. (2002). A direct approach to false discovery rates. \u003cem\u003eJournal of the Royal Statistical Society: Series B (Statistical Methodology)\u003c/em\u003e, \u003cem\u003e64\u003c/em\u003e, 479\u0026ndash;498. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/1467-9868.00346\u003c/span\u003e\u003cspan address=\"10.1111/1467-9868.00346\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTam, T. Y. C., Sivarajkumar, S., Kapoor, S., Stolyar, A. V., Polanska, K., McCarthy, K. R., Osterhoudt, H., Wu, X., Visweswaran, S., Fu, S., Mathur, P., Cacciamani, G. E., Sun, C., Peng, Y., \u0026amp; Wang, Y. (2024). A framework for human evaluation of large language models in healthcare derived from literature review. \u003cem\u003enpj Digit Med\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e, 258. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41746-024-01258-7\u003c/span\u003e\u003cspan address=\"10.1038/s41746-024-01258-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan der Lee, C., Gatt, A., van Miltenburg, E., \u0026amp; Krahmer, E. (2021). Human evaluation of automatically generated text: Current trends and best practice guidelines. \u003cem\u003eComputer Speech \u0026amp; Language\u003c/em\u003e, \u003cem\u003e67\u003c/em\u003e, 101151. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.csl.2020.101151\u003c/span\u003e\u003cspan address=\"10.1016/j.csl.2020.101151\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThomson, C., Reiter, E., \u0026amp; Belz, A. (2024). Common Flaws in Running Human Evaluation Experiments in NLP. \u003cem\u003eComputational Linguistics\u003c/em\u003e, \u003cem\u003e50\u003c/em\u003e, 795\u0026ndash;805. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1162/coli_a_00508\u003c/span\u003e\u003cspan address=\"10.1162/coli_a_00508\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"language-resources-and-evaluation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"lrev","sideBox":"Learn more about [Language Resources and Evaluation](http://link.springer.com/journal/10579)","snPcode":"10579","submissionUrl":"https://submission.nature.com/new-submission/10579/3","title":"Language Resources and Evaluation","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Arabic clinical NLP, question reformulation, domain-adaptive pretraining, masked language modeling, intrinsic evaluation, human validation","lastPublishedDoi":"10.21203/rs.3.rs-8007820/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8007820/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArabic clinical NLP systems often receive short, vague, or incomplete questions, which yields weak downstream answers even with strong encoders. We address this bottleneck by making question quality a first-class, measurable objective. Using domain-adaptive (continued) pretraining with a masked-language objective (DAPT-MLM) on AHQAD (~\u0026thinsp;808k Arabic health Q\u0026ndash;A pairs), we adapt two widely used backbones\u0026mdash;AraBERT and the generator variant of AraELECTRA\u0026mdash;to the lexical, syntactic, and discourse patterns of well-formed medical questions. Evaluation is aligned with the learning signal: we report cross-entropy and perplexity only at masked tokens, top-k accuracy restricted to masked spans, and lexical-diversity measures to discourage formulaic phrasing. A length-controlled test design (Short/Long/Very Long) isolates modeling gains from verbosity. Results show consistent intrinsic improvements for the domain-adapted models; AraBERT-MLM is best overall (macro Top-5\u0026thinsp;=\u0026thinsp;0.8392, lowest CE/PPL), outperforming AraBERT (orig.) by +\u0026thinsp;6.0 pp Top-5 and AraELECTRA (orig.) by +\u0026thinsp;17.2 pp. A 200-item human study (clinician\u0026thinsp;+\u0026thinsp;linguist) corroborates these gains (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;95% CI: Clarity 4.12\u0026thinsp;\u0026plusmn;\u0026thinsp;0.18, Fluency 3.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.22, Semantic Fidelity 3.15\u0026thinsp;\u0026plusmn;\u0026thinsp;0.25, Usefulness 3.42\u0026thinsp;\u0026plusmn;\u0026thinsp;0.21; substantial agreement, κ\u0026thinsp;\u0026asymp;\u0026thinsp;0.77) and highlights residual semantic drifts that inform simple, slot-constrained decoding fixes. Overall, the proposed reformulation module produces more natural and clinically relevant Arabic questions and can be plugged into Arabic clinical QA pipelines as a measurable, tunable front-end.\u003c/p\u003e","manuscriptTitle":"Improving Arabic Clinical Question Quality through Domain-Adaptive Masked Language Modeling","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-19 18:39:56","doi":"10.21203/rs.3.rs-8007820/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-15T09:49:20+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-04T16:50:02+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-02T04:18:07+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-28T22:17:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-16T21:06:52+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"243873019313933696142025859339645217511","date":"2025-11-14T09:27:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"28451055070372273350773813417737072636","date":"2025-11-12T02:45:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"315046199452537012422311292672594731062","date":"2025-11-10T22:40:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"92742600654000235169349485422473111171","date":"2025-11-10T17:19:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"115584854841451927584469168423060775475","date":"2025-11-10T16:04:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"82223820318188250586989031272010303147","date":"2025-11-10T15:46:49+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-10T15:41:45+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-10T13:43:57+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-07T11:28:08+00:00","index":"","fulltext":""},{"type":"submitted","content":"Language Resources and Evaluation","date":"2025-11-01T20:31:18+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"language-resources-and-evaluation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"lrev","sideBox":"Learn more about [Language Resources and Evaluation](http://link.springer.com/journal/10579)","snPcode":"10579","submissionUrl":"https://submission.nature.com/new-submission/10579/3","title":"Language Resources and Evaluation","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"f00ce755-c7c9-4fb9-ad56-f2c8d2e60b3c","owner":[],"postedDate":"November 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-09T16:10:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-19 18:39:56","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8007820","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8007820","identity":"rs-8007820","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}