Extracting Japanese Domain and Technical Terms is Relatively Easy

We argue that the important task of extracting domain and technical terms is much easier in Japanese than is commonly believed, and that technical term extraction should and can be more widely used. We present a method which shows that the implementation of such a tool for Japanese can be remarkably simple due to the regularity and rigidity of morphosyntactic characteristics of Japanese technical terms. Our learning algorithm bootstraps from the tagged output of a Japanese tokenizer/tagger to learn syntactic and morphological properties of technical terms. We show that very simple linguistic patterns are reliable enough to yield high precision technical terms from tagged texts, even for terms which occur only once and are usually overlooked by English technical term extractors. In addition, by incorporating a model for unknown words, we are able to extract correct technical terms from words not properly tagged.