Design and evaluate a chatbot for hesitant language learners' conversational practice

Thesis Abstract

Effective conversational practice is essential for language acquisition; however, hesitant language learners often experience limited opportunities for authentic interaction due to anxiety, lack of confidence, and scarcity of communicative environments. This study addresses the challenge of enhancing spoken language proficiency among hesitant learners by designing, implementing, and evaluating an AI-powered chatbot tailored to support autonomous conversational practice. The primary aim is to develop a user-friendly, contextually responsive chatbot that can identify and adapt to learners’ hesitations, thereby fostering increased engagement, fluency, and confidence in oral communication. The study articulates specific objectives (1) to analyze the conversational needs and specific hesitation markers of language learners, (2) to design a chatbot utilizing natural language processing (NLP) and machine learning techniques, (3) to implement the chatbot within a controlled learning environment, and (4) to evaluate its effectiveness in reducing learners’ anxiety and improving conversational competence. The research adopts a mixed-methods approach, integrating qualitative and quantitative data collection and analysis. A purposive sample of 120 intermediate-level English as a Second Language (ESL) learners from three language institutes will participate in the study. The quantitative component involves a pre-test/post-test experimental design, with participants randomly assigned to either the chatbot-based practice group or a control group engaging in traditional classroom activities. Data collection instruments include standardized speaking proficiency tests, anxiety scales, and engagement questionnaires. The qualitative component employs semi-structured interviews and focus groups with participants and language instructors to explore perceptions and contextual factors influencing chatbot usability and efficacy. Data analysis will involve paired t-tests and ANCOVA to assess differences in oral proficiency and anxiety reduction between groups, as well as thematic analysis for qualitative data. To enhance the interpretability of quantitative outcomes, structural equation modeling (SEM) will be employed to explore relationships among engagement, hesitation reduction, and speaking improvement, while model fit will be evaluated using indices such as CFI and RMSEA. Expected findings anticipate that learners engaging with the chatbot will demonstrate statistically significant improvements in oral fluency, reduced speech-related anxiety, and higher engagement levels compared to the control group. It is also projected that the adaptive features of the chatbot—such as hesitation detection and personalized prompts—will contribute to increased learner confidence and willingness to participate in spontaneous conversation. These outcomes are expected to affirm the significant role of tailored AI-based interventions in addressing affective barriers in language learning and provide evidence of the chatbot’s capacity to serve as an effective supplementary tool for autonomous practice. The study contributes to the growing body of knowledge by integrating theories of communicative competence and technology acceptance models, specifically applying Vygotsky’s social constructivism and the Technology Acceptance Model (TAM) to explain how AI-driven tools influence language learning processes. It develops a conceptual framework linking chatbot design features, learner engagement, affective factors, and language proficiency outcomes, which can inform future pedagogical innovations and technological developments. The main conclusion underscores that a well-designed, context-aware chatbot can effectively reduce learners’ hesitation and promote active participation in conversational practice. Recommendations include integrating such chatbot tools into language curricula, training instructors to facilitate technology-enhanced learning environments, and exploring the scalability and adaptability of similar systems across diverse linguistic and cultural contexts. Finally, the study suggests avenues for future research, such as longitudinal investigations on sustained engagement and the integration of multimodal feedback systems to further support language development.