Jade Abbott, co-founder and CTO of Lelapa AI, is spearheading a major push to build artificial intelligence models that speak African languages and bridge the communication gap in technology. Her project is focused on creating tools that work well for people who use isiZulu, Sesotho, Afrikaans, and other local languages—especially where existing technologies fail.
One of the flagship tools from Lelapa AI is Vulavula, currently in beta, which transcribes voice to text and can detect names of people and places in written text across four South African languages. The aim is to expand to more languages over time. Abbott and her team are also building models and data-sets in collaboration with communities on the ground to ensure cultural relevance and correctness—gathering, curating, and annotating language data rather than just relying on large internet-scraped data.
Abbott’s background includes her work with Masakhane, a grassroots NLP (Natural Language Processing) collective she co-founded in 2018 that has released hundreds of open-source language models and datasets for African languages. The experience there shaped her belief that inclusion in AI starts with representation—allowing African languages to actually function in AI systems, not just be token mentions.
The stakes are high. Majoring tech platforms often misinterpret or misrepresent translations in African languages—output can be inaccurate, tone lost, names mistranslated—and these errors erode trust and usefulness. For inclusive services (education, healthcare, public service info), accurate language support isn’t just nice to have, it’s essential. Lelapa AI’s tools promise to make services more accessible—for people who speak their mother tongue, especially in rural areas or places underserved by technology.
Challenges remain: limited funding, scarcity of high-quality data in many African languages, the technical complexity of handling oral languages, tone, dialects, and accents. But Abbott and the Lelapa team are navigating these by community partnerships (linguists, local speakers), efficient model design (small and efficient models rather than huge black-box ones), and a strong open-source/grassroots orientation.
0 تعليقات