With about 3 000 different languages and dialects, Africans lose out when it comes to technology that is usually in English and major European languages, but efforts are underway to change that through conversational AI. Stellenbosch University research fellow Johan Steyn explains.
When my eight-year-old son recently demonstrated our Google Home device to his friends, and bragged about how good it was at answering questions, his friend Mandla was excited to ask a question in Zulu. But the gadget failed to understand the language that is spoken widely in South Africa.
No ad to show here.
It made me wonder why the technology is so bad with African languages. Owing to the increasing use of voice-based interfaces such as Amazon’s Alexa, Apple’s Siri, and Google’s Home Assistant, it is estimated that internet searches by voice will far outnumber other search methods in 2022.
Natural language understanding (NLU) is an area of study within artificial intelligence (AI), providing text, speech, or a combination of the two as input to computers. Enabling human-computer interaction, computers can interpret commands without the codified syntax of computer languages because they can understand human languages.
Most of us have interacted with rudimentary NLU technology through chatbots. Often a popup on an internet browser or a mobile phone application, these bots aim to answer our questions – a job that human would otherwise do. Unfortunately interacting with this technology is often a source of frustration. Able to only understand and answer the most basic questions, bots are often no more than glorified FAQ apps.
But increasingly, many internet chats use conversational AI that is able to understand the meaning behind questions, and utilises all available data to provide suitable answers. The technology is a great deal more efficient and user-friendly than its forebears.
The societal benefits of conversational AI are vast. Able to provide contextual and accurate advice on important topics such as financial planning and healthcare, these platforms are destined to become our primary go-to advisors.
Imagine your young child is severely ill with a fever and you desperately need advice fast. Imagine speaking to an AI application on your phone, describing the problem and having the techno-advisor understand you, and provide quick and accurate advice.
Now imagine you are in a similarly perilous situation, but that English is your third or fourth language – or that you do not speak English at all – what are you to do?
You can neither communicate your problem, nor get a satisfactory response from the platform.
With an estimated 3,000 different languages and dialects, Africans are disproportionally disadvantaged in accessing technology that is usually available only in English, and major European languages.
So there is a pressing need to create language data sets unique to our continent. Fortunately, there are already many initiatives like Masakhane, a grassroots organisation which is at the forefront of developing African language data for conversational AI applications.
Zindi, a network of data scientists in Africa, in cooperation with AI4D-Africa – a coalition of AI researchers – are working on support and funding for Natural Language Processing initiatives on the continent.
My hope is that soon the languages, idioms and expressions of my son’s friends will be fully understood by devices that are taking pride of place in our lives.
- Johan Steyn is a research fellow at the school of data science and computational thinking at Stellenbosch University, South Africa.