By Johan Steyn, 21 September 2021
The language barrier prevents full advantage of the internet as major companies focus on the English capabilities of their products.
I was on a Zoom call but I could hear the little voices laughing in the next room. My son was proudly demonstrating to his friends his mastery of our Google Home smart speaker. I could hear the gasps as the artificial intelligence (AI) voice answered questions on the weather forecast and the time of the next full moon. A friend shouted, “Let’s try isiZulu.”
“Sawubona, unjani?” I could hear the device was taking longer than normal to respond. Finally a strange sounding accent announced, “Lutho oluningi”. By now the laughter was even more pronounced. “How do you mean ‘nothing much’?” One boy asked. “Ngangisiza na? Can I help you?”
Smart speakers are being used by an increasing number of people to shop, set reminders, and get answers to simple but critical questions on issues such as health care. This is due to the widespread use of voice-based interfaces such as Amazon’s Alexa, Apple’s Siri, and Google’s Home Assistant. According to research, by 2020, voice searches had accounted for half of all internet queries.
One manifestation of such an expectation is the widespread use of language as a means of communicating with machines. We now communicate with virtual assistant systems, listen to their responses, and base our decisions on their suggestions or reactions.
Natural language understanding has to do with the ability of the machine to interpret human language. We are attempting to train the machine to be able to recognise and capture intelligence through the interpretation of language.
When internet connectivity in Africa is discussed, the discourse usually centres on people who have access to the internet and those who do not — the digital divide. However, there is also a linguistic barrier, which has been growing in recent years, particularly when it comes to speech recognition technologies.
Rather than working on languages from developing countries in Africa and Asia, developers are concentrating their efforts on expanding the English language capabilities of their products. With major platforms such as Twitter and Google AdSense not supporting many African languages, they are already at a competitive disadvantage on the internet.
Low-income groups and languages that are not widely spoken in the West are being pushed to the margins of society. Therefore, not only will there be a gap between those who can use this form of AI for communications and those who cannot, this disparity will also make it difficult for users to take advantage of these applications for development interventions such as health care, education and finance.
My son is already using our smart speaker to help him with homework. He has all the information that has ever been created available to him. Imagine what this technology can do for all the children in our country. Imagine what it can do in health care for people in need of health advice but who are either living too far from a doctor or who cannot afford medical care.
When his friends left, my son told me that he thinks there is a Zulu in our smart speaker. I had to smile, but it also made me wonder if we are doing enough as a collective to develop the technological application of native African languages.
To accommodate the needs of African language users, we must use a combination of locally designed machine comprehension algorithms and locally collected training data. We should support initiatives such as Masakhane.io whose goal is for Africans to develop and own technology through a multidisciplinary approach.