本研究提出一種方法，將英文學習者寫作中出現的冗餘表達，替換為更精準的替代詞。此方法將輸入句子轉換為向量查詢，分別從片語層級與句子層級的索引資料庫中檢索出語意接近的替換候選詞。透過此流程設計，能進一步實現自動學習學習者寫作中的冗詞用法、自動生成具語意保留的改寫建議，並同時確保語句的流暢性與符合使用者指定的CEFR 語言等級。我們實作了一個名為Wordy2Tidy 的雛形系統，應用此方法於英文學習者的句子編輯上。實驗結果顯示，在語句精煉度方面優於單純使用GPT 提示生成的結果，並且在消除冗贅程度上優於現有的人工校訂語料。我們的方法透過精準替換與等級控制的優化，幫助學習者在不改變原意的情況下提升表達的清晰度。

We introduce a method for replacing redundant expressions in a given learner-written English sentence with precise alternatives. In our approach, sentences are transformed into embedding-based queries that retrieve accurate candidates from both phrase-level and sentence-level indexes. The method involves automatically learning redundancy patterns in learner writing, automatically prompting rewrites that preserve meaning and automatically ensuring both fluency and CEFR-level appropriateness. We present a prototype paraphrasing engine, Wordy2Tidy, that applies the method to learner-written English editing. Evaluation on a set of learner-written sentences shows that the method outperforms GPT prompting in clarity and outperforms human-edited corpora in reducing redundancy. Our methodology supports learner rewriting through precise substitution and level-aware refinement, resulting in improved clarity without loss of meaning.