Arabic is not one language in practice. Modern Standard Arabic dominates news and textbooks, but everyday conversation in the UAE happens in Emirati Arabic, a Gulf dialect with its own vocabulary, idioms, and cultural references. The Technology Innovation Institute (TII) has released Falcon-Emirati-7B, a model built on top of its Falcon-H1-Arabic family, to handle this dialect the way a native speaker would.
The base model uses a hybrid architecture that runs Mamba state-space layers and Transformer attention in parallel, which the team says helps with long sequences and a morphologically rich language. For dialect adaptation, the team assembled three data sources: authentic Emirati web content, Modern Standard Arabic material about Emirati culture and heritage, and synthetic data generated under strict glossaries and style rules to avoid generic "Gulf-ish" output.
The team notes that dialect adaptation is hard because Emirati is mostly spoken, meaning is often non-literal, and there is no established recipe for mixing data or choosing training stages. They chose the 7B variant as a compromise: the 34B model might push quality further but would be too costly for a dialect-specialized chat model, while the 3B model lacks the capacity for the required cultural and linguistic depth.