Recognising a word in the Tunisian dialect such as “3asselama” and translating it easily and correctly on Google, in English or other languages could now be possible!
According to the TAP, This is one of the objectives of the “dataset” project called “Tunizi”, and developed by Tunisian start-up “iCompass”, specialising in artificial intelligence (AI), and more particularly in natural language processing (NLP).
It is a database containing 9000 sentences written in Tunisian dialect and includes the first large set of sentiment analysis data used on social media.
The purpose of the project and other similar ones under development in Africa and the Arab world is to create annotated databases and to develop and submit data sets of any underserved and under-represented dialect for use in future NLP applications such as machine translation, conversational agents (Chatbot) or speech recognition.
For artificial intelligence, this is necessary because annotated datasets now condition the success of the algorithms used as a basis for learning.
Concretely, “iCompass” proceeded to collect data from comments on Youtube.
The videos selected included sports, politics, comedy, TV shows, TV series, arts and Tunisian music videos so that the dataset was representative and contained different types of age, education, writing, etc. The videos were then analyzed by the “iCompass” team.
Ahmed Nouisser, Co-founder and CEO of iCompass said “the general principle is to offer the Tunisian, Maghreb and African ecosystem more annotated datasets, called “datasets”, to use them in the development of research (social, scientific, human and other …)
(TAP)