| Project/Area Number |
09680303
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Japanese language education
|
|---|
| Research Institution | SCIENCE UNIVERSITY OF TOKYO |
|---|
Principal Investigator |
ITOH Kohji Science University of Tokyo,Department of Applied Electronics Professor, 基礎工学部, 教授 (20013683)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
伊丹 誠 東京理科大学, 基礎工学部, 助教授 (70212983)
|
|---|
| Project Period (FY) |
1997 – 1998
|
| Project Status |
Completed (Fiscal Year 1998)
|
| Budget Amount *help |
¥3,500,000 (Direct Cost: ¥3,500,000)
Fiscal Year 1998: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 1997: ¥2,900,000 (Direct Cost: ¥2,900,000)
|
|---|
| Keywords | Learning Assistance System / Second Language Learning / Japanese / Diagnostic Processing / Learning Item / LTAG / Sample Text Data Base / Indexing / 日本語教育 / 誤り診断 / 形態素解析 / 概念辞書 / 機械学習 / 自然言語処理 / 知的CAI / 機能インデックス / 状況依存 / 診断 / 対話データベース / TAG |
|---|
| Research Abstract |
The research aims at fusion of grammatical and communicative approaches developing an intelligent assistance system for the learners of Japanese as second language. The system provides the learners with a text database with indices of learning items designating discriminatory use of expressions. Assisting the learners in comprehension of the contexts by way of linking with electronic dictionaries, the system asks the learners to fill in the blanks made in the sample texts by composing sentences selecting from the specified vocabulary and inflecting if necessary. The system diagnoses the composition and, if the erroneous sentences are not far from feasible ones, the system retrieves corresponding sample texts by the learning item indices, presenting them to help the learners recognize their errors by themselves.
We developed a diagnostic parsing algorithm making use of Lexcalized Tree Adjoining Grammar and 2 stacks which proceeds by unification and linear back tracking. Provided the word
… More
list and the semantic modification relationship are shared with the learners, the algorithm successfully could detect excessive modifications, lacking of modifiers missing or missuse of collectives, erroneous inflections, obstacles against modification, crossed modification and inappropriateness viewed from the situations. In order to help the authors index the sample text, we developed a set of rules for mapping from the surface characteristics of the sample sentences to learning items. Use of only the morpheme patterns resulted in 90% of reproducibility and 84% of precision.
Use of syntactic analysis focusing only single stages of modification resulted in 96% of reproducibility and 80% of precision. This lowering of precision may have been avoided by taking into account multiple morpheme candidates, 2 or more stages of modification and by prohibiting partial matching. It was confirmed that introduction of conceptual dictionary would greatly improve precision provided certain level of word sense disambiguation were achieved. Less
|
