Falcon Emirati, Falcon ASR and Falcon OCR Arabic show how Abu Dhabi is building AI around the language, speech and documents used across the region. 

[For more news, click here]

An AI system can understand Arabic without necessarily understanding how people in the UAE speak it. Emirati Arabic has its own vocabulary, expressions and cultural references, while systems trained primarily on Modern Standard Arabic can respond correctly without sounding natural to local users. As artificial intelligence moves into customer service, government services, transcription and document processing, the quality of that interaction becomes more important for organizations deploying these systems. 

Abu Dhabi's Technology Innovation Institute, the applied research arm of the Advanced Technology Research Council, has released three Falcon models aimed at these different uses. Falcon Emirati is designed specifically for Emirati Arabic, Falcon ASR handles speech across Emirati and Modern Standard Arabic as well as English, French, Spanish and Portuguese, while Falcon OCR Arabic is designed to extract Arabic text and document structure from images. TII announced the models on October 6 as part of its continuing work on Arabic AI.  

A model built for the way Emiratis speak 

Falcon Emirati builds on TII's Falcon H1 Arabic work and was developed using native Emirati content alongside material covering Emirati culture and heritage. TII says it performs strongly on Alyah, its benchmark for native Emirati Arabic, which covers everyday language, figurative expressions, heritage knowledge and poetry. The model is also being made available through Falcon Chat.  

TII's evaluation also measures whether models maintain the Emirati dialect rather than switching to Modern Standard Arabic. Falcon Emirati recorded a higher dialect fidelity score than the other Arabic models included in the comparison. The results come from TII's own testing and are not an independent industry ranking, so they provide an indication of what the model was built to improve rather than a definitive measure of its performance across the wider AI market. 

The model arrives as several Gulf countries and technology companies invest in Arabic language systems. Abu Dhabi based Inception and Microsoft have developed Seraj for enterprise applications involving Arabic, while Saudi Arabia has continued developing ALLAM. For companies operating across the region, these efforts are creating more alternatives to global general purpose models when local language performance is a priority. 

Speech and documents move into the same picture 

Falcon ASR focuses on spoken information. It converts speech into text across six languages and provides word level timestamps so transcripts can be matched to their original recordings. TII says the model was developed for conditions including background noise, overlapping speakers, reverberation and telephone audio. Those capabilities can support meeting transcription, subtitles, searchable recordings and customer interactions where spoken conversations need to be converted into text.  

Falcon OCR Arabic works with information contained in documents and images. It extracts Arabic text and structured information, including tables and other document elements, and is intended for uses such as document digitization, searchable archives and automated information extraction. TII's evaluation covered a range of documents including official records, forms, newspapers, receipts, labels, business cards and invoices.  

For organizations handling large volumes of Arabic paperwork, converting scanned material into machine readable information can remove a layer of manual work. Government records, invoices, forms and archived documents can be searched, classified and passed into other software once their contents have been extracted. The usefulness of the model will depend on how it performs with the documents and workflows of individual organizations, but document processing is one of the areas where specialized AI can be applied directly to an existing business process. 

As Gulf businesses begin integrating AI into routine operations, Arabic language capability will increasingly sit alongside factors such as cost, deployment requirements, data handling and performance. The growth of locally developed models gives organizations in the region more choice, while also creating pressure on their developers to prove that these systems work reliably beyond controlled evaluations and in everyday use. 

Related Articles 

Inception42 and Microsoft Launch Seraj, an Enterprise AI Model Built for Arabic Fluency 

Microsoft to Bring Saudi Arabia's Arabic AI Model ALLAM to Its Global Platform 

Exclusive: Arabic AI Is Becoming a Global Battleground for Technology, Culture, and Power