Exploring the Efficacy of a Local-Language Interfaced ChatGPT as A Self-Learning Support Tool for Junior High School Students in Rural Communities in Africa

Abstract

Large language models have been found to be useful in teaching and learning. Unfortunately, these models are not usually built on low-resource languages like many of the local languages spoken in Ghana and other West African countries. Using the Twi language, spoken in many parts of Ghana, as a case study, this research sought to answer the question, "Does a local language interface on ChatGPT make it a good learning support tool for junior high school students in rural communities in Africa?" The main objectives were to develop a Twi-interfaced ChatGPT that facilitates interaction in Twi and to assess students’ acceptance of various interaction modalities, namely typing and speech. Students from a junior high school in a rural community of Ghana interacted with the standard ChatGPT chatbot, ChatGPT with an English speech interface, the proposed system with a Twi speech interface, and a simulated system in which a human played the role of a Twi interface. The case study illustrates that the concept of a local language interface to ChatGPT as a learning tool is promising: although the performance of the system currently falls short of what is needed, there is user acceptance of the concept, which could work well in the near future as speech recognition and text-to-speech systems for low-resource languages approach human-level accuracy.

Ähnliche Arbeiten