There was a text dump made a few years ago, I think by utunnels (aka magil).
http://cc.ffsky.cn/scripts/jp/The numbers at the beginning of the file names are the numbers of the dumped files they were extracted from in the Japanese version.
It's technically considered incomplete because the Japanese version uses its accent system on character-unique dialogue. The web interface tries to automatically detect the speaking character from the tag at the beginning of a text, but there are some cases where the tag that says exactly which character is not used. As a result, there is no way to automatically create a perfect script dump from just the original text data.