以CLEC語料庫為基準來設計實作一個英語作文錯誤檢查系統

Abstract

隨著英語課程的進入小學，以及英語檢定的普及，學習英語已蔚為風行。因應這樣的發展，學生們已開始在小學接受英語教學，以及相關的測驗，而在這種情形之下，在外補習或是自學的人數也隨之增長；但是，以學習英語作文為例，當學習者在完成一篇作文的時候，只能尋求人工的方式批改，不但耗費時間，評分標準也可能不一。不同於有著固定答案的選擇題或是非題，早就以電腦閱卷的方式取代人工批閱，即便是有少數系統可以批改，正確率也有待提升。 因此，本研究試著去提昇實驗室中的英語作文的錯誤檢查系統之正確率，希冀以CLEC語料庫中分類清晰，且具代表性的錯誤類型為驗證藍本，撰寫相對應的作文錯誤檢查模組。期望以一致的角度，先將作文當中的拼字、文法錯誤，或是文不對題的情形清楚的標示出來，以提供閱卷者在閱卷時評分的參考。 同時，由於英語對我們華人來說，畢竟屬於外來的語言，其中的文法和語態與中文也不盡相同，導致學習上的困難。因為如此，我們希望建立起一套系統，除了可提供學習者做相關的事前練習之外，也可作為閱卷者評分的參考；更重要的是，系統以CLEC語料庫為驗證背景，整併及撰寫新的文法檢查模組，使錯誤類型之分類更為完整，並提高系統之錯誤檢查率。此外，也可讓受試者在不需要他人協助的情況之下，就可以自行操作練習，並不會受到外在環境的條件限制，也充分的運用了資訊設備，真正享受網路所帶給我們的便利性。

As the elementary school starts teaching English and the popularity of GEPT (General English Proficiency Test), many people consider learning English indispensable. Following this trend, many students start to receive English courses and take English tests. Therefore, the number of students who go to English institutions or appeal to self-study has increased as well. However, taking learning English composition as an example, when the learners finish their compositions, they must seek someone to check their works. This is a time-consuming process and, furthermore, the grading standard may vary. Multiple-choice and true and false questions have fixed answers that can be checked through Optical Mark Recognition, replacing the manual grading method. Even though English composition can be graded by a few systems, its accuracy may be varied due to the difference of learning background. As a result, this research intends to improve the error checking system for Chinese learner’s English composition. The explicitly categorized and the representative error types in CLEC will be the basis to write the error checking system for English composition. By using the coordinate approaches, the misspellings, grammatical errors, and the irrelevance sentences can be marked clearly and served as references for the graders. Since English is still foreign to Chinese, the differences among voice and tenses have caused difficulties in learning. Thus, we must design a system not only providing the learners to do relevant practice beforehand but also functioning as references for the graders. More importantly, our system is based upon CLEC and integrates many grammar checking modules to cover the error types as many as we can. For a learner, it can be used to his own English composition conveniently. For a grader, it can be used to uncover all possible errors before they do the final semantic checks. Thus, the grading basis for all testers’ English compositions is the same. Numeric English composition samples (list in appendix) are used to demonstrate the feasibility and applicability of the proposed system.