我們提出了一種方法，用於生成符合特定CEFR熟練度等級的句子結構。在我們
的方法中，輸入的學習者句子會透過檢索增強生成（RAG）提示進行轉換，生
成語法增強的句子，該提示設計為納入分級語法模式。該方法透過關鍵詞自動
檢索英語語法概覽（EGP）中的相關語法模式，構建提示來引導生成式AI 模型
（如ChatGPT），以生成符合目標CEFR 等級的句子。在學習者句子的評估中，
該方法能夠有效實現目標級別的語法結構，並顯著優於未結合語法模式檢索的
傳統方法。我們的方法幫助學習者的句子達到語法準確性的提升，顯著改善語
言學習應用效果。此外，該方法還支持靈活增強各種語法模式，使學習者能夠
提升各種文法結構的等級。

We introduce a method for generating sentences with grammatical structures aligned to a specified CEFR proficiency level. In our approach, an input learner sentence is transformed into a grammatically enhanced sentence using a Retrieval-Augmented Generation (RAG) prompt designed to incorporate graded grammar patterns. The method involves automatically retrieving relevant grammar patterns from the English Grammar Profile(EGP) based on keywords, constructing prompts to guide a generative AI model such as ChatGPT, and generating sentences consistent with the target CEFR level. Evaluation on learner sentences shows that the method effectively achieves target-level grammatical structures, significantly outperforming baseline approaches that do not leverage grammar pattern retrieval. Our methodology helps learners' sentences achieve enhanced grammatical accuracy, providing significant improvements in language learning applications. It also supports flexible augmentation with varied grammar patterns, enabling learners to improve their proficiency in various grammatical structures.