From Language to Action in Arabic: Reliable Structured Tool Calling via Data-Centric Fine-Tuning

TL;DR

This paper introduces AISA-AR-FunctionCall, a robust Arabic function-calling framework that significantly improves structural stability and accuracy through systematic data-centric fine-tuning and reasoning enhancements.

Contribution

It presents a new Arabic function-calling system with a comprehensive training pipeline, reducing parse failures from 87% to below 1% and enhancing accuracy across dialects and domains.

Findings

Parse failures reduced from 87% to below 1%.

Function name accuracy increased over eightfold.

Improved argument alignment across dialects and domains.

Abstract

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.