Talking your language: UAE unveils AI tool designed to master Arabic dialects

The Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi has unveiled an advanced Arabic AI tool named Jais 3. This innovative platform is designed to understand and respond to various Arabic dialects, including modern standard Arabic and dialects spoken across 19 Arab countries. Built in partnership with Abu Dhabi's Inception42, Jais 3 features Jais Speech, a voice recognition model capable of transcribing conversations, creating documents, and handling diverse language tasks.

Haonan Li, a senior research scientist at MBZUAI's Institute of Foundation Models (IFM), explained that Jais 3 is specifically tailored to the nuances of Arabic language spoken in different regions and across generations. The AI system can handle code-switching, where bilingual individuals switch between languages within a single sentence, such as slipping into English or French mid-sentence and back again.

To achieve this, Jais Speech transcribes Modern Standard Arabic and local dialects spoken across the region. It identifies the dialect being used, allowing it to maintain the original script during transcription. The tool also adds Arabic vowel marks to the transcript, ensuring accurate representation of the spoken language.

Developing Jais 3 posed several challenges, with a focus on creating a balanced representation of the various Arabic dialects. The team had to work extensively on data synthesis and filtering to ensure the quality and relevance of the dataset, especially for Gulf dialects where open models tend to default to Modern Standard Arabic. The same issue arose for Morocco, where available data heavily favored Moroccan Darija, leaving less representation for other dialects.