Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance
huggingface ·
The Technology Innovation Institute has released Falcon-Emirati-7B, a large language model specialized in understanding and generating Emirati Arabic dialect. Built upon the Falcon-H1-Arabic architecture, this 7-billion parameter model addresses the challenges of adapting general Arabic models to specific cultural and linguistic nuances. The team utilized a three-part data strategy involving authentic web crawls, MSA content focused on Emirati heritage, and strictly guided synthetic data to ensure accuracy. Evaluation results show that Falcon-Emirati-7B achieves an 84.83% score on the Alyah benchmark, outperforming larger multilingual competitors. Crucially, the model demonstrates superior dialect fidelity, reliably responding in Emirati rather than defaulting to Modern Standard Arabic like other models. It also scored highest on the ArabCulture-Dialogue benchmark for recognizing culturally appropriate responses in UAE-specific scenarios. The release highlights that targeted dialect training is essential for capturing non-literal meanings, poetry, and social norms that generic models often miss.
huggingface