African Next Voices: Bridging the AI Language Gap with Massive African Dataset

The Language Gap in AI

Artificial Intelligence (AI) tools like ChatGPT, DeepSeek, Siri, and Google Assistant have predominantly been developed by the global north and trained in English, Chinese, or European languages. This has left a significant gap in the representation of African languages in AI systems 1

1

.

The African Next Voices project, a collaborative effort involving African computer scientists, linguists, and language specialists, has been working for two years to address this critical issue 2

2

.

The Importance of Language in AI

Language plays a crucial role in AI development and usage. It serves as the primary medium for human-AI interaction, enabling us to communicate our needs and assess the AI's understanding. The project team emphasizes that language is not just a tool for communication but also a carrier of culture, values, and local wisdom 1

1

. By excluding African languages from AI models, we risk overlooking a vast wealth of human knowledge and cultural diversity.

Challenges Facing African Languages in AI

The underrepresentation of African languages in AI stems from historical factors, including colonialism and policy choices that have prioritized colonial languages in education, media, and government. This has resulted in a scarcity of high-quality, digitized text and speech data necessary for training robust AI models 1

1

.

Moreover, the development of AI for African languages faces additional challenges, such as the lack of basic language tools like dictionaries, keyboards, fonts, and spell-checkers. The rich dialect diversity and variations in orthography further complicate the process of building comprehensive datasets 2

2

.

The African Next Voices Project

To address these challenges, the African Next Voices project has embarked on an ambitious initiative to collect speech data for Automatic Speech Recognition (ASR) in various African languages. The project, primarily funded by the Gates Foundation with additional support from Meta, involves a network of African universities and organizations 2

2

.

Data Collection and Methodology

The project's approach is characterized by its commitment to diversity and ethical data collection practices:

1. In Kenya, the Maseno Centre for Applied AI is collecting voice data for five languages, representing three main language groups: Nilotic (Dholuo, Maasai, and Kalenjin), Cushitic (Somali), and Bantu (Kikuyu) 1

   1

   .

2. Data Science Nigeria is focusing on five widely spoken languages: Bambara, Hausa, Igbo, Nigerian Pidgin, and Yoruba 2

   2

   .

3. In South Africa, the Data Science for Social Impact lab and its collaborators are recording seven languages: isiZulu, isiXhosa, Sesotho, Sepedi, Setswana, isiNdebele, and Tshivenda 1

   1

   .

The data collection process ensures diversity in age, gender, and educational background of participants. It covers various domains, including everyday conversations, healthcare, financial inclusion, and agriculture. Importantly, all recordings are obtained with informed consent, fair compensation, and clear data-rights terms .

Implications and Future Prospects

The African Next Voices project represents a significant step towards making AI more inclusive and accessible for millions of African language speakers. By addressing the language gap in AI, the project aims to unlock the potential of AI applications in various sectors, including education, healthcare, and agriculture, for African communities 1

1

.