Earticle

초록

영어

Generative AI may support autonomous language learning, but its feedback can extend beyond the learner’s intended revision target. This study examines the characteristics and limitations of Gemini-generated feedback on Japanese dialogues written by early-intermediate university students in Korea and considers its potential for autonomous learning. The data consisted of 99 dialogues produced by 32 students. After an exploratory examination of prompt conditions intended to reduce input burden and excessive rewriting, one prompt was selected and applied to all dialogues. The results showed that, even when instructed to revise only grammatical and lexical errors while preserving the original text, Gemini also modified expressions related to conversational naturalness and pragmatic appropriateness. It also produced missed errors and inappropriate revisions. These findings suggest that Gemini can provide feedback beyond linguistic accuracy, but learners need to evaluate whether revisions preserve their original meaning and communicative intentions.

목차

Abstract
1. はじめに
2. 先行研究
3. 研究方法
3.1 研究対象と会話データ
3.2 会話作成課題
3.3 使用した生成AIと実施条件
3.4 プロンプト検討の手順
3.5 分析対象と分析方法
4. プロンプトの検討
5. Geminiのフィードバックの特徴
5.1 待遇表現・丁寧さの調整
5.2 会話らしさの調整
5.3 対人配慮・発話機能の調整
5.4 場面に応じた語彙・定型表現への変更
5.5 表記の変更
6. 確認が必要な修正の傾向
6.1 必須ではない修正
6.2 修正の未検出
6.3 誤修正
6.4 入力条件による出力の相違
7. 考察
8. まとめと今後の課題
参考文献

키워드

저자

• 伊藤江美 [ Ito, Emi | 済州大学校招聘教員 ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  언어학연구 [Journal of Linguistic Studies]

• 간기

  연3회

• pISSN

  1226-9859

• 수록기간

  1996~2026

• 등재여부

  KCI 등재

• 십진분류

  KDC 700 DDC 400