Emirati dialect-trained AI models are released by Abu Dhabi's Technology Innovation Institute
The speech recognition model has 1.6 billion parameters, but it was more accurate than a 30-billion-parameter model, according to TII.
Abu Dhabi's Technology Innovation Institute (TII) has unveiled AI models that have been trained on the Emirati Arabic dialect, enhancing translation, transcription, and voice-enabled services. The Institute, a state-run research body, has introduced the Falcon-Emirati model and associated tools for speech recognition, as well as extracting Arabic text from images and documents.
These models are notably compact, with the speech recognition model boasting 1.6 billion parameters, which outperform a more extensive 30-billion-parameter model in terms of accuracy, according to TII.
The development of AI training in Arabic has historically lagged behind that of other languages due to the vast variations in dialects and their often incomprehensibility to those who are only familiar with formal Arabic used in news reports and textbooks. Consequently, machine translations have produced comically inaccurate results, sometimes converting innocuous words into vulgar references in English.
TII aims to rectify these errors by training models on local dialects, idioms, and cultural references, thereby making AI more applicable and beneficial in the region.
Mohammed Sergie, a representative of TII, expressed the Institute's objective of eliminating such inaccuracies by leveraging the unique linguistic nuances of the Emirati Arabic dialect. This effort is expected to elevate the utility of AI technologies in the region and contribute to more precise and culturally relevant translations and communications.
Written by urgent.news from Semafor's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
