人工智慧主導下的郵件寫作？應用文寫作課程中二語寫作者與聊天機器人的合作案例研究

Composing Emails in an AI-dominated mode? A Case Study of L2 English Writers’Collaboration with Chatbots in a Professional Writing Course

應用文寫作（Professional Writing）課程以電子郵件、個人履歷、信函等體裁為講授內容，是高校英語教學體系中經常設置的課程，然而，人工智慧（AI）生成工具的迅速發展，為寫作課的課堂教學設計與習作評價帶來了複雜的挑戰，並已引起了教師及研究者的關注（Cardon et al., 2023；Ibrahim & Kirkpatrick , 2024）。為了解英語二語寫作者如何處理AI生成的文本，研究者聚焦於電子郵件寫作任務，要求學生向教授撰寫一封英文郵件，並諮詢研究生入學錄取事宜。在閱讀ChatGPT生成的郵件文本後，學生需要根據自身理解，結合具體語境對郵件進行修改。同時，研究者邀請了三位高校教師，對學生郵件的完成質量進行評估。基於對郵件文本的分析，研究者發現，AI工具生成的郵件篇幅往往較長，但學生作者卻容易忽視表達中的冗餘問題。此外，讀者表示，儘管詞彙誤用和語法錯誤得到了有效控制，格式問題依然大量存在。在郵件修改的過程中，寫作者在合理採用語用策略、適當增補個人信息方面也遇到了一定困難。本研究為寫作教師提供了與人機互動相關的教學啟示，有助於教師更新寫作課堂中的教學設計，並制定適用於AI輔助寫作的評估標準。

The rapid introduction of text-generating software programs, which has brought complicated challenges into writing classrooms, has been discussed with concerns by teachers and researchers in writing studies (Cardon et al., 2023; Ibrahim, 2023). As a course offered in most of the higher education institutions, Professional Writing (PW) usually features explanations of writing genres such as emails, resumes, and letters. To observe how L2 English writers process texts generated by AI, this classroom-based study focuses on the design of a professional email writing assignment in a PW course. L2 English writers were asked to draft an email to a professor and consult graduate school admission possibilities. After being provided with email texts produced by ChatGPT, students needed to revise the email based on their own understanding of the situation. Professors with both L1 and L2 English speaking backgrounds were invited to evaluate the email. Content analysis of students’submissions and reader interviews show that AI chatbot tends to generate email texts that are lengthy, while student writers would often dismiss the redundancy issue. Email drafts submitted by students also demonstrated difficulties with personalizing the text. In addition, formatting glitches become much prominent to professors as readers. Although vocabulary misuse and grammatical inaccuracies have mostly been kept at a minimum, formatting problems have been reported by readers as strong indicators of low writing proficiency.
This study offers writing instructors insights of human-AI interaction, and will benefit the design of writing pedagogy in future classrooms. Interviews with both readers would also clarify the evaluation criteria of AI-assisted writing, which are expected to provide more guidance for both teachers and students.