Arabic function calling model scores top in dialect robustness test
QMUL_AK at AISA-ArabicFC: Schema-Constrained Function Calling with ALLaM
Computation and Language
Summary
The paper presents a system that helps computers understand and call functions in Arabic, even when different dialects are spoken. The authors adapted an existing Arabic-English language model to improve how it recognizes function calls and their details. Their system performs very well on official tests, ranking third overall and first in handling dialect differences. The study shows that while it's easy to pick the right function, getting all the details right is still challenging.
What this means in practice
- •For software developers: Integrate robust Arabic function-calling capabilities into AI systems that handle multiple dialects for better user interactions.
- •For customer service platforms: Improve automated Arabic language understanding to correctly interpret user requests in various dialects for function execution.
Authors
Ali Khairallah, Arkaitz Zubiaga
Abstract
We describe the QMUL_AK submission to AISA-ArabicFC 2026. The system adapts the Arabic-English ALLaM-7B-Instruct model to function calling with low-rank adaptation (LoRA). It combines an Arabic schema-aware prompt, dialect and no-call resampling, a token-weighted training objective, and deterministic validation of generated calls against the candidate tool schema. The same output supports Track A function calling and Track B reasoning-augmented calling. Training uses only the official task data. On the development set, the system achieves 0.9982 function accuracy and 0.8920 argument exact match, corresponding to overall A and B scores of 0.9345 and 0.9454. On the blind test set, the system reaches overall A and B scores of 0.9253 and 0.9380, ranking 3rd on both submission tracks and 1st of 21 teams on the Track C dialect-robustness diagnostic with a macro-averaged score of 0.9075 and the smallest dialect gap in the shared task (0.077). The results show that function selection is nearly saturated, while argument extraction remains the main source of error.