Automatic construction of CALL system from TV news program with captions

Many language learning materials have been published in Japan. However, they are limited in their scope and content. In addition, we doubt whether the speech sounds found there are natural in various situations. These days, some TV news programs (by NHK, CNN, ABC[1], etc.) have closed/open captions corresponding to the speech of the announcer. We have developed a system that automatically makes CALL (Computer Assisted Language Learning) materials from such captioned newscasts. Materials compiled by this system have the following functions: repetition listening, consulting an electronic dictionary, and automatic construction of a dictation test. The materials have the following advantages: polite and natural speech sound, various and timely topics, and abundant materials. In this paper, we describe the organization of our new system.