Advertisement

This scientist is integrating AI technology with over 500 African languages

Ife Adebara is working to make AI technology more accessible for speakers of various African languages, to help avoid the loss of those languages. (Ife Adebara - image credit)
Ife Adebara is working to make AI technology more accessible for speakers of various African languages, to help avoid the loss of those languages. (Ife Adebara - image credit)

Ife Adebara is on a mission to make technology accessible to all people in the African continent — by making their language available on the technology they use.

"Someone who speaks a minority language has to put their language aside in order to be able to get technology in English for example," said Adebara, a programmer and scholar at the University of British Columbia's linguistics department.

"Over time their usage of the language begins to drop and that can have long-term consequences of language endangerment … we need to mitigate that."

Adebara says AI technology is moving fast but leaving behind people who speak languages other than English.

Her project, called Afrocentric Natural Language Processing, works to create awareness, tools and programs that is accessible to members of the public who speak an African language, including Swahili and Zulu.

So far, she says, the team has released two language identification programs online called SERENGETI and AfroLID.

Ahead of her interview on The Early Edition, Adebara spoke with CBC's Ali Pitargue to share more about her work, and the need to make technology inclusive.

The following interview has been edited for clarity and length.

What is the AI and African languages initiative?

The idea is trying to get technology to African people in their indigenous languages.

This involves developing these assets, building artificial intelligence models and deploying it to the African people, so that they can interact with technology in the languages that they feel most comfortable with, which is usually their indigenous languages.

Which languages are you working on?

There are 2,000-plus languages in Africa. So right now I've been working on about 517 of them, which are spoken from 50 out of 54 countries in Africa.

That number is growing because I'm getting more languages added to the list I'm working on. The goal is to be able to get to as many languages as possible within the African continent.

These languages are often referred to as low-resource languages. Can you elaborate more on that and the challenges it comes with? 

Low-resource languages are languages that do not have adequate data to build classical language models for AI.

So the challenge with these types of languages is the performance is usually lower than languages that have high resources like English, French and some other Indo-European languages.

One approach that I've been using to mitigate this challenge is putting together many languages in the same model. This way the model learns from multiple languages and the data set becomes bigger than what it was before and the performance is a little better.

But it still needs to be solved, because if we're going to get to near-human accuracy, then we need more data to be able to achieve that.

Why is it important to make sure African languages don't get left behind in the development of these technologies?

There are two reasons why this is important. The first is that more than one billion people in Africa, which is about 17 per cent of the world's population, are excluded from global conversations in their indigenous languages.

They're not hearing what other people are saying and we are not hearing what they are saying.

The other thing is that for a lot of African languages, their grammatical features are very diverse and sometimes unique just to the African continent.

If we build language technology and exclude African languages, the models and technologies are not learning certain language features. Which is also not good, because they're not versatile across different grammatical features that exist in human language.

What do you hope this project achieves overall?

I hope the technologies become available and accessible to the average African. This would definitely have long-term effects on education.

They can have access to information on the web in their language, either translated to their language or translated from another language. I see people being able to have access to health information in their language or being able to use Google Maps.

For more stories about the experiences of Black Canadians — from anti-Black racism to success stories within the Black community — check out Being Black in Canada, a CBC project Black Canadians can be proud of. You can read more stories here.

(CBC)