本研究提出一種提升非母語學習者英文寫作品質的方法，著重於在維持語句流暢與連貫性的前提下，改善詞彙使用的精確性與多樣性。該方法針對語法正確但詞彙表達較為初級或不自然的句子進行改寫，使其更加高階且具表達力。訓練資料由學習者原始句子及其由生成式人工智慧模型產生的升級版本組成，升級句子在保留原意的同時提升詞彙選擇、詞彙多樣性及整體流暢度。資料建構上，本研究使用現有學習者語料庫，自動產生符合歐洲共同語言參考架構（CEFR）詞彙水平與語言品質指標的進階句子。系統運作時，學習者文章首先經由語法錯誤修正模型（GEC）處理，以確保語法正確，隨後由詞彙提升模型（WLU）進行語句優化。我們開發了一個原型系統LexiLift，應用本方法於學術寫作輔助場景。根據自動與人工評估結果顯示，系統能有效提升寫作品質，特別是在流暢度與詞彙豐富度方面，同時維持語法正確性。此結果顯示本系統具作為學生寫作輔助工具之潛力。

We introduce a method for enhancing the English writing of non-native learners by improving the lexical quality of their essays while maintaining fluency and coherence. In our approach, we focus on revising sentences that are already grammatically correct but contain basic or unnatural word choices, aiming to make them more advanced and expressive. The method involves training on sentence pairs consisting of original learner-written sentences and their upgraded versions generated by a generative AI model. These upgraded sentences are designed to retain the original meaning while improving aspects such as word choice, lexical diversity, and overall fluency. To build the training data, we utilize existing learner corpora and automatically generate enhanced versions of each sentence, targeting higher levels of writing quality as defined by criteria such as lexical richness and Common European Framework of Reference for Languages (CEFR) level appropriateness. At run-time, learner essays are first processed by a grammar error correction (GEC) model to ensure grammatical accuracy before being passed to a word level upgrading (WLU) model for lexical enhancement. We present a prototype, LexiLift, that applies the method to support writing improvement in academic contexts. Evaluation based on both automatic metrics and human judgment demonstrates that the system significantly improves writing quality, particularly in terms of fluency and lexical richness, while preserving grammatical correctness. These results highlight the potential of the system as an effective tool to support learner writing development.