Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2603.16901 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908894689755136 |
|---|---|
| author | Nacar, Omer Alquffari, Deema Alsharideh, Saleh AlOtaibi, Adeem Alabdulkarim, Abdulaziz Alhazmi, Leen Alomar, Nada Alzubaidi, Wareef Alsultan, Nada Alrabghi, Ahmed Alhoshan, Demah Alsayyari, Rana Alruwaili, Hamed Jaafar, Albaraa Alusmani, Khaled Alsohimy, Abdulaziz Alsubaie, Munirah Aldukhayil, Shahd Alali, Arwa BinShihah, Yazeed Alsulaymi, Razan Alhumaid, Nourah Abdulsalam, Razan Alamoudi, Reem Alkhalifa, Mohammed |
| author_facet | Nacar, Omer Alquffari, Deema Alsharideh, Saleh AlOtaibi, Adeem Alabdulkarim, Abdulaziz Alhazmi, Leen Alomar, Nada Alzubaidi, Wareef Alsultan, Nada Alrabghi, Ahmed Alhoshan, Demah Alsayyari, Rana Alruwaili, Hamed Jaafar, Albaraa Alusmani, Khaled Alsohimy, Abdulaziz Alsubaie, Munirah Aldukhayil, Shahd Alali, Arwa BinShihah, Yazeed Alsulaymi, Razan Alhumaid, Nourah Abdulsalam, Razan Alamoudi, Reem Alkhalifa, Mohammed |
| contents | Function-calling language models are essential for agentic AI systems that translate natural language into executable structured actions, yet existing models exhibit severe structural instability when applied to Arabic. We present AISA-AR-FunctionCall, a production-oriented Arabic function-calling framework built on a 270M-parameter FunctionGemma backbone and trained through systematic dataset auditing, schema repair, tool-aware prompt restructuring, and full-parameter supervised fine-tuning. On a held-out test set, fine-tuning reduces parse failures from 87\% to below 1\%, improves function name accuracy by more than eightfold, and substantially enhances argument alignment across dialects and domains. Error analysis reveals a transition from structural collapse to semantic misalignment, suggesting that serialization stability and decision-level reasoning are separable challenges. We further explore a reasoning-augmented LoRA variant that introduces explicit intermediate reasoning prior to tool invocation. All datasets and models are publicly released under the AISA framework. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_16901 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | From Language to Action in Arabic: Reliable Structured Tool Calling via Data-Centric Fine-Tuning Nacar, Omer Alquffari, Deema Alsharideh, Saleh AlOtaibi, Adeem Alabdulkarim, Abdulaziz Alhazmi, Leen Alomar, Nada Alzubaidi, Wareef Alsultan, Nada Alrabghi, Ahmed Alhoshan, Demah Alsayyari, Rana Alruwaili, Hamed Jaafar, Albaraa Alusmani, Khaled Alsohimy, Abdulaziz Alsubaie, Munirah Aldukhayil, Shahd Alali, Arwa BinShihah, Yazeed Alsulaymi, Razan Alhumaid, Nourah Abdulsalam, Razan Alamoudi, Reem Alkhalifa, Mohammed Machine Learning Artificial Intelligence Function-calling language models are essential for agentic AI systems that translate natural language into executable structured actions, yet existing models exhibit severe structural instability when applied to Arabic. We present AISA-AR-FunctionCall, a production-oriented Arabic function-calling framework built on a 270M-parameter FunctionGemma backbone and trained through systematic dataset auditing, schema repair, tool-aware prompt restructuring, and full-parameter supervised fine-tuning. On a held-out test set, fine-tuning reduces parse failures from 87\% to below 1\%, improves function name accuracy by more than eightfold, and substantially enhances argument alignment across dialects and domains. Error analysis reveals a transition from structural collapse to semantic misalignment, suggesting that serialization stability and decision-level reasoning are separable challenges. We further explore a reasoning-augmented LoRA variant that introduces explicit intermediate reasoning prior to tool invocation. All datasets and models are publicly released under the AISA framework. |
| title | From Language to Action in Arabic: Reliable Structured Tool Calling via Data-Centric Fine-Tuning |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2603.16901 |