Please use this identifier to cite or link to this item:
Title: 統計式與規則式文法校正系統之發展及成效評估
The Development and Evaluation of Statistics-Based and Rule-Based English Grammar Checkers
Authors: 國立臺灣師範大學英語學系
Issue Date: 2008
Abstract: 第二語言寫作研究中,回饋一直是重要的議題。學生需要適度的回饋以增進 寫作能力。隨著電腦科技的進步,商業化的自動文章評分軟體逐漸被使用來針對 學生的寫作提供回饋。然而,這些軟體所提供的回饋品質常受到許多老師與學生 的質疑,特別是針對文法方面的回饋。因此,研發功能更強大的英文文法校正系 統是有必要的。對於如何增進文法校正系統的功能這議題,研究者已提出許多方 法,而其中最有潛力的莫過於統計式與規則式文法校正系統。雖然這兩種文法校 正系統已得到一些正面的研究結果支持,但仍存有許多增進的空間。 統計式文法校正系統的改進可藉由增加語料庫的量來達成。舉例來說,美國 教育測驗服務社(ETS)的Criterion 系統所使用的語料庫約三千萬字。如有辦法 使用大規模的語料庫將可以進一步增進文法校正系統的效能。除此之外,現行的 統計式系統(例如ETS 的Criterion) 嘗試處理來自不同語言背景之英語學習者 的中介語(Interlanguage)。這確實是個野心的創舉。然而,不同母語背景的學 習者所產生的語言對於系統會造成不少困擾, 因為一般性的偵錯機制無法對於 個別的學習者提供適當的幫助,尤其當學習者來自不同的語言背景。因此,對於 以中文為母語的英語學習者來說,其中一項突破方法是使用以中文為母語之英語 學習者的語料庫來增進現有的文法校正系統的偵測能力。藉由這特定語料庫所發 掘出的資訊,我們可以針對中文為母語之英語學習者的需要來設計文法校正系 統。 規則式文法校正系統能夠處理一些統計式系統無法涵蓋的問題,但研發這種 系統卻很耗時。首先,文法規則要先轉換成電腦代碼,如此一來文法校正系統才 能利用既有的規則來分析學習者的語法。幸運地,研究者現在已有較多編輯工 具可使用,譬如開放原始碼的工具與商業軟體。藉由這些工具的輔助,我們將能 更容易的將新的規則加入文法校正系統,並測試規則的準確性。除此之外,現存 的中文英語學習者語料庫能幫助研究者有系統的發現錯誤型態並將之加入文法 校正系統之內. 上列所提之統計式與規則式文法校正系統的研發是既實際又有學術研究價 值的。在完成文法校正系統系統雛型之後,我們將廣邀老師與學生們參與系統測 試,並根據他們所提供的回饋與意見來改善我們的系統。本計畫希望能研發出適 合於中文為母語之英語學習者的英語文法校正系統,並能提供中學生與大學生免 費使用。
Feedback has always been a very important research issue in second language writing. Students are expected to use feedback to improve their writing performances. As computers become more affordable and powerful, new commercial automated essay scoring systems are used to provide feedback to students. However, the quality of feedback on grammar of these systems is widely questioned by ESL teachers and learners. The need for a better and more robust English grammar checker is very strong. Researchers have proposed different solutions to develop better grammar checkers. Two promising approaches are statistics-based and rule-based grammar checkers. Although some positive results are available, these two types of grammar checkers can be further improved. For the statistics-based checkers, we can further improve its training corpus size. The corpus used by the ETS Criterion system is only about 30-million. We can use larger corpus and make the coverage of the grammar checker more comprehensive. In addition, the existing statistics-based system (e.g., ETS Criterion) tries to process the interlanguage produced by ESL learners from various language backgrounds. Indeed, it is ambitious for a system to handle English writings by learners from different L1 backgrounds. However, the diverse learner output also causes problems for the system since the very general errors detection mechanism cannot offer adequate help for learners from a specific language background. For the needs of Chinese ESL learners, the grammar checker can be improved based on the useful information from Chinese English learner corpora. Based on the valuable information extracted from Chinese ESL learner corpora, we can target at the needs of Chinese ESL learners and thus improve the quality of grammar feedback. For the rule-based system, it can address some problems which can not be covered by the statistics-based system, but it is often time-consuming to develop a system. Much work needed to be done to convert rules into the computer codes so the grammar checker can use these rules to parse various deviant input. Fortunately, authoring tools are now available for researchers. There are open source tools provided by researchers and commercial grammar tools offered by companies. With the help of these tools, it is much easier to add rules into a grammar checker and to test the accuracy of new rules. Moreover, the availability of Chinese learner English corpora helps researchers uncover error patterns more systematically. These patterns then can be added into the grammar checker. These two different approaches to grammar checker development are both practical and worth exploring. Upon the completion of the prototypes, we will invite more teachers and students to try these grammar checkers we develop. Based on user feedback, the systems can be tuned to improve their performance. It is expected that useful English grammar checkers for Chinese students can be developed from this project. High school and college students will have free access to these writing tools.
Other Identifiers: ntnulib_tp_B0212_04_012
Appears in Collections:教師著作

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.