TII launches Emirati language, speech and document AI models

Abu Dhabi's Technology Innovation Institute announced three AI models for Emirati Arabic, multilingual speech transcription and extracting Arabic text from documents. Falcon-Emirati is a 7-billion-parameter language model that the institute says scored 84.83% on Alyah, a native Emirati benchmark, outperforming every Arabic and multilingual open-source model evaluated. Falcon-ASR converts speech in Standard and Emirati Arabic, English, French, Spanish and Portuguese into text and records word timestamps. Falcon-OCR-Arabic extracts text and structured information, including tables and mathematical formulas, from images and documents. Falcon-Emirati will be available through a Falcon Chat platform. Its reported benchmark covers everyday language, figurative expressions, heritage knowledge and poetry.