在本論文中,我們介紹一個新的方法,對於學習者提供正確的介係詞錯誤改正回饋。在我們的方法裡,我們從大型錯誤標記語料庫中抽取平行樣式文法。此外,我們也從一般英語語料庫中抽取樣式文法。這方法包含在訓練資料中自動辨識句子結構性關係,藉由字頻率自動抽取獨特的結構性關係,自動產生並過濾錯誤更正樣式文法。在執行階段,給定一個句子將被轉換成樣式文法,並應用訓練階段取得的樣式文法做篩選、排名介係詞錯誤更正建議回饋。我們實作了一個雛形系統WriteAhead,可自動抽取並展示鼠標周圍的相關的錯誤更正樣式文法和例句。我們初步的實驗和評估結果在一個公開的資料集中,顯示我們的方法在介係詞錯誤更正的效能上是合理的。
In this paper, we introduce a new method for providing corrective feedback for preposition errors in learners' writing. In our approach, we extract Synchronous Grammar Patterns (SGP) for grammatical error correction (GEC) from a large error-annotated corpus. In addition, we also extract grammar patterns (GP) from a general English corpus to validate and supplement GEC patterns. The method involves automatically identifying syntactic patterns in the training data, automatically extracting distinct patterns with counts, and filtering and generating GEC patterns. At run-time, we identify grammar patterns in a given sentence. We apply our acquired GEC patterns to match the patterns in the given sentence and rank GEC patterns by frequency. We present extit{WriteAhead}, a prototype system, that automatically extracts and displays relevant grammar error correction patterns with examples to prompt the user as they type or mouse around a draft. Preliminary experiments and evaluation results on a publicly available dataset, show our method works reasonably well for preposition errors.