Automatically assessing oral narratives of Afrikaans and isiXhosa children

  • 2025-07-17 15:15:43
  • R. Louw, E. Sharratt, F. de Wet, C. Jacobs, A. Smith, H. Kamper
  • 0

Abstract

Developing narrative and comprehension skills in early childhood is criticalfor later literacy. However, teachers in large preschool classrooms struggle toaccurately identify students who require intervention. We present a system forautomatically assessing oral narratives of preschool children in Afrikaans andisiXhosa. The system uses automatic speech recognition followed by a machinelearning scoring model to predict narrative and comprehension scores. Forscoring predicted transcripts, we compare a linear model to a large languagemodel (LLM). The LLM-based system outperforms the linear model in most cases,but the linear system is competitive despite its simplicity. The LLM-basedsystem is comparable to a human expert in flagging children who requireintervention. We lay the foundation for automatic oral assessments inclassrooms, giving teachers extra capacity to focus on personalised support forchildren's learning.

 

Quick Read (beta)

loading the full paper ...