研究発表会

宇佐美 裕子氏研究発表会

９月１７日（火）に宇佐美 裕子氏 (PhD at Lancaster University) による研究発表会を専修大学サテライトキャンパスにて開催いたします。どうぞ奮ってご参加ください。院生、学部生にもお知らせください。

講師：宇佐美 裕子氏（武蔵野大学）
演題：“How can corpora be applied to language testing?”
日時：9月17日（火）午後2時30分〜4時
会場：専修大学サテライトキャンパス　スタジオA
（小田急線　向ヶ丘遊園　北口駅前 アトラスタワー２F）
連絡先：池尾玲子（専修大学）
予約は必要ございません。

Lecture by Dr Hiroko Usami (PhD at Lancaster)
Title: How can corpora be applied to language testing?
Date/ Time : Tuesday, 17 September, 2013 (14:30-16:00)
Venue: Senshu University, Satellite Campus, Studio A
(North exit, Mukogaokayuen, Odakyu Line)
Admission: Free
Contact: Reiko Ikeo, Senshu University

Abstract: How can corpora be applied to language testing?
A corpus is a large collection of naturally occurring language data which has been electronically encoded so it can be analysed via computer software. Recently corpora have been utilised in language testing. However, this kind of research is relatively new and a framework for applying corpus techniques to language testing has not yet been fully developed.
In this presentation, I would like to talk about how corpora and corpus linguistic techniques can be used to identify and improve problematic questions in multiple choice grammar questions in Japanese university entrance exams. Three types of problematic questions were identified: 1) questions which had distractors which were not congruent with learners’ actual errors 2) questions with repeatedly used stems that were not frequent in English 3) questions which had more than one possible correct answer. I identified ten representative questions of each type and attempted to improve them by using different corpora and corpus linguistic techniques. I then gave Japanese students both the original questions and the same questions altered using corpus techniques. The tests were analysed both quantitatively and qualitatively (via the use of the think aloud protocol).
An analysis of the test results showed that many (but not all) of the test questions could be considered to have been improved in terms of having a better discrimination index and item facility. Generally, the questions which had distractors replaced with errors from a learner corpus were found to be the most improved, while those which had new stems based on authentic English from the British National Corpus showed the least improvement. The questions which had multiple correct answers removed and replaced with frequent distractors in the original exams were generally easier to answer, although it could be argued that the original questions were unfair and thus invalid. The results thus suggest that corpus techniques can be effective in demonstrating how problematic questions can be identified and improved.
Finally, I would like to introduce the latest research on the application of corpora to language testing and make suggestions for practical use of corpora in language testing.