Détails Publication
ChatGPT Does Not Understand Moore: On the Challenges of Integrating African Languages in Large Language Models,
Lien de l'article:
Discipline: Informatique et sciences de l'information
Auteur(s): Aminata Sabané, Tegawendé F. Bissyandé
Renseignée par : BISSYANDE T. François D'Assise
Résumé

The advent of Large Language Models (LLMs) likely constitutes a turning point in the history of mankind. LLMs have the potential to revolutionize the way we interact with computers and each other. While current state of the art LLMs mostly support western languages, such as English, it is paramount that we consider the opportunity of integrating African languages into LLMs. First, LLMs can be used to generate text, translate languages, and answer questions in African languages. This can help to make information and resources more accessible to people who speak African languages. Second, LLMs can be used to develop African language keyboards, chatbots, and voice assistants. This can help to make African languages more accessible and user-friendly. Third, LLMs can be used to generate new content in African languages, translate African language texts into other languages, and create educational resources in African languages. This can help to ensure that African languages are used and spoken in the future.

Despite the potential benefits of LLMs for African languages, there are still a number of challenges that need to be addressed before LLMs can be widely deployed for indigenous communities in Africa. In this paper, we will discuss the challenges for training a large language model with African languages in more detail. We will also discuss some of the recent efforts to address these challenges.

Mots-clés

LLM, low-resource languages, natural language processing, Moore

935
Enseignants
7820
Publications
49
Laboratoires
101
Projets