Formative Assessment of Speaking for Young Learners Using Customized Generative AI

This chapter examines the use of generative artificial intelligence (GenAI), specifically ChatGPT, to design formative speaking assessments for young learners. Speaking is inherently interactive and requires interlocutors, which poses practical challenges for classroom instruction and assessment. Traditional approaches to speaking assessment, whether direct or semi-direct, have long struggled with issues of feasibility, cost, and construct underrepresentation. GenAI offers a novel alternative by providing scalable, adaptive, and interactive systems that mimic human-like communication while delivering immediate and individualized feedback. Drawing on established theories of communicative competence and models of feedback in formative assessment, the chapter outlines how GenAI can be leveraged to address linguistic, sociolinguistic, and interactional aspects of speaking proficiency. We describe the process of creating a customized GPT for English practice, emphasizing task specifications, tone, feedback design, and iterative testing. Sample interactions demonstrate how the system delivers feedback across multiple levels—task, process, self-regulation, and self—while maintaining a supportive learning environment. The chapter also discusses pedagogical applications, including classroom integration and opportunities for learners to review their speech performances through transcripts. Limitations of the approach are addressed, including data privacy, age restrictions, lack of multimodal inputs, and concerns about the transparency and reliability of automated scoring. We argue that while such systems cannot replace human interlocutors, they provide a cost-effective and pedagogically sound means of extending speaking practice and formative assessment opportunities. The chapter concludes by highlighting implications for teachers, learners, and researchers, and by calling for continued evaluation of GenAI’s role in language education.