2001 Volume 25 Issue 2 Pages 85-94
In Computer-Assisted Language Learning System, the student makes various mistakes. Following constructivism, the system has to make him aware of his mistakes by himself, before pointing them out. One way to achieve the purpose is to allow the student compare his sentence with correct ones belonging to the same or different teaching items by retrieving and presenting such sentences. For that purpose, we are studying automated retrieval of sentences from corpora of Japanese text in terms of teaching items using the arts of NLP (Natural Language Processing) . The retrieval is carried out using teaching item extracting rules consisting of conditional parts coming from morphemic and syntactic informants and conclusion parts of teaching items. We have developed a repertoire of teaching item extracting rules. By using these rules, other than realizing automated retrieval, we can index the sentences by teaching items, thus speeding up the later retrieval or making them undergo scrutiny by the teacher before providing them to the students as assistance material for reading and writing. The performance of the prototypical automatic indexing system was examined by inspection. The accuracy was 87% and the recall rate was 89%. The prototype retrieval system was evaluated by professionals and volunteers engaged in teaching Japanese as second language. Their opinions are being implemented in the revised version.
