2006 Fiscal Year Final Research Report Summary
The Research and Development of an Automatic Evaluation System for Oral Communication Proficiency
| Project/Area Number |
17520406
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Foreign language education
|
|---|
| Research Institution | Osaka Women's College |
|---|
Principal Investigator |
TEAMAN Brian D. Osaka Women's College, English Department, Professor, 国際・英語学部, 教授 (20294481)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
TAMURA Takeshi Kobe Institute of Computing, Graduate School of Information Technology, Professor, 情報技術研究科, 教授 (60291605)
MCCARTY Steve A. Osaka Women's Junior College, English Department, Professor, 英語科, 教授 (90290585)
|
|---|
| Project Period (FY) |
2005 – 2006
|
| Keywords | English / Foreign Language / Phonetics / Education / CALL / Automatic Speech Recognition |
|---|
| Research Abstract |
The current speaking test system consists of three types of users : students, teachers and administrators. Students can register, take tests and change their profile (including adding themselves to classes and changing their password). The teacher users are the most important part of the system and can perform many functions like the following: 1. Teachers can grade tests which involves selecting a student's tests and listening to it and applying a rating to it. From the test grading module, teachers can easily see a list of all students who have taken a particular test, choose tests to grade and then evaluate each item using a five-level scale that can be applied to any token depending how the teacher wants to grade it. For example the scale could refer to overall comprehensibility or focus on one aspect such as whether the answer is correct or the pronunciation is correct. 2. The teacher can also control the machine-grading system. To do this the teacher must create grammars and dictionaries in order to perform the machine grading. The machine-grader then gives a score from 0 to 100 depending on how accurately the spoken word matches the teacher-created model. 3. The teacher can also manage classes and tests. The administrator can use their account to manage teachers and their accounts.
There are currently 152 different users registered on the MASLE system. These users have taken ten different tests making a total of over 2000 different test items recorded. There are currently four different machine-gradable tests that have been given to 25 native Japanese speakers and five different native English speakers to test the accuracy of the speech recognition system. The results showed that machine recognition is very promising since non-native speakers ranged from about 30 to 80 (out of 100) and native speakers ranged from 80 to 100.
|
Research Products
(2 results)
