Beyond ChatGPT: A conceptual framework and systematic review of speech-recognition chatbots for language learning

Abstract

The diversification of chatbot technology, such as the emergence of large language models and their incorporation into various technologies, necessitates a conceptual framework for a comprehensive understanding of different chatbot types and their possibilities for educational use. However, despite the fact that chatbots with different characteristics can provide learners with different interaction experiences, previous research has drawn on a loose conceptualization of chatbots, ignoring the common or unique design features of different chatbots and the educational affordances that are provided accordingly. In response to this concern, this review aims to further our understanding of different types of speech-recognition chatbots for language learning and the affordances provided by the chatbots. Based on an analysis of 37 empirical studies on uses of chatbots ranging from those with predefined dialogue systems to those utilizing artificial intelligence technology, this review proposes a conceptual framework that comprises three key components of a chatbot system: goal-orientation, embodiment, and multimodality. Using this framework as an analytical tool, eight chatbot types are identified and defined. Additionally, a total of 12 affordances are derived from the presence and absence of each component of the framework. Analysis of the studies through the framework also offers specific insights into how future chatbot research and development should be pursued in terms of goal-orientation, embodiment, and multimodality. Finally, we discuss the potential of the framework as a relevant model for understanding chatbots in adjacent disciplines and types other than speech-recognition chatbots, including ChatGPT and other large language models.