Why do AI models fail to speak African languages?

Artificial Intelligence (AI) models that speak over a hundred languages but have neglected over 2000 African languages.

Sundar Pichai, Google’s CEO, promised nine years ago to make AI accessible to everyone but for Africans that promise remains distant as AI has failed to bridge linguistic divides and provide equal access to knowledge for all.

According to TRT Global, an estimated 1.2 billion people on the continent are unable to communicate with advanced AI models in their own language.

One of the leading chatbots in the market ChatGPT has over 800 million weekly active users and it can only understand 10-20 per cent of sentences written in Nigerian language, Hausa.

Several other languages including Yoruba, Igbo, Swahili and Somali are all facing the same treatment.

Researchers blame the low-resources such as websites, books and transcripts available in these languages online behind their neglect in AI models.

An AI researcher at University of California who focuses on Large Language Models (LLMs) said, “Our measure for progress and research is based on what works for western languages.”

Hellina Hailu Nigatu said, “African languages are categorized as low resource and are usually excluded,” adding that systems perform poorly on them when they are added.

Digital data is the main source of learning for most of the LLMs, since most of the data available online is in English, so chatbots perform best in English language.