This study proposes the design of an AI chatbot-based pragmatic feedback application to support the teaching of business Korean request-email writing to intermediate and advanced learners of Korean as a Foreign Language (KFL). In business emails, ...
This study proposes the design of an AI chatbot-based pragmatic feedback application to support the teaching of business Korean request-email writing to intermediate and advanced learners of Korean as a Foreign Language (KFL). In business emails, a request is a speech act that can impose a burden on the recipient, since it asks the recipient to carry out a particular action. Writing such an email appropriately therefore requires not only grammatical accuracy but also the adjustment of discourse structure, request strategies, mitigation, supportive moves, and honorifics and style in light of the recipient relationship, the degree of imposition, and the formality of the email genre. In KFL settings, however, learners have limited opportunities to receive individualized, repeated pragmatic feedback on their own drafts. Accordingly, this study explores the design of an AI chatbot-based feedback tool that helps learners examine and revise their own drafts against pragmatic criteria. This study is a design-oriented study that proposes feedback categories, a response structure, a learning flow, and a prototype, rather than an empirical study that verifies the app's instructional effectiveness.
To this end, the study first reviewed prior research on business Korean email-writing instruction and public-email discourse, on request speech acts and the CCSARP framework, on the politeness and pragmatic appropriateness of request emails, and on generative AI-based writing feedback and chatbot-supported learning. It then analyzed how request emails are presented in fifty request-related materials (forty-eight full emails and thirty-seven core materials) drawn from four business Korean and email-writing textbooks. The analysis found that, although the textbooks present the structure and expressions of request emails to some extent, they largely embed pragmatic elements within completed model examples, offering little support for learners to check individually whether their own drafts are appropriate to a given request situation. On the basis of this theoretical discussion and textbook analysis, the study derived five pragmatic feedback categories: discourse structure; request strategy and directness; internal mitigation; external supportive moves; and honorifics and style, including over-politeness.
The proposed application is organized into a six-stage learning flow: setting the request situation, reviewing a pragmatic writing guide, composing a first draft, receiving AI-based pragmatic feedback, revising the draft, and conducting a final check. In the situation-setting stage, the learner specifies the recipient relationship, the degree of imposition, the purpose of the request, and the reply deadline, and the AI pragmatic feedback for each category is delivered in the structure of a statement of the problem, an explanation of its reason, an example of an alternative expression, and a reflective question. The design follows three principles—pragmatic-appropriateness-based feedback, process-oriented writing support, and teacher-complementary AI use—and adopts response-restriction principles, such as restrictions on composing the email on the learner's behalf, on categorical evaluation, and on inducing over-politeness, so that the AI neither writes the email for the learner nor judges errors on the basis of the learner's background alone. The design was implemented as a web-based prototype using HTML, CSS, and JavaScript. It was implemented in a version connected to a generative AI API to produce feedback, while a rule-based demonstration reproducing the same review structure was provided for the expert review.
The educational appropriateness and usability of the design were examined through a preliminary review by four experts in Korean discourse and pragmatics, Korean writing education, and educational technology. An analysis of the Likert-scale and open-ended responses showed that the feedback categories, the process-oriented response structure, and the staged learning flow were generally evaluated positively, while the conciseness of the recommended expressions, the learner-friendly conversion of academic terms, the adjustment of on-screen information load, and the verification of response stability after actual AI integration were identified as points for improvement. The study revised the design to reflect these findings.
This study is significant in that it links the theoretical discussion of request speech acts, textbook analysis, pragmatic feedback categories, and the design of an AI chatbot-based learning tool into a single design rationale. In particular, it positions AI feedback not as a replacement for teacher feedback but as a teacher-complementary tool that supports the learner's first-stage self-checking and revision. This study is limited, however, in that it did not include the analysis of actual learner output, verification of live generative AI integration, or usability and effectiveness testing with learners. Future research should therefore conduct usability testing with KFL learners, examine the response stability and accuracy of the API-connected version, and extend the design principles of this study to other business email speech acts such as apology, refusal, and suggestion.