Co-training applied in automatic term extraction

This paper discusses the use of a setting similar to co-training in automatic terminology processing. Two aspects of terms (internal aspect, i.e linguistic, and statistical properties; and external aspect, i.e. contexts) will be used interchangeably in a bootstrapping manner, in order to extract more and more terms and context patterns. The results show that, using only a small set of seed terms, the method can extract terms, with higher success rates than those of other methods. Further more, this method can also discover interesting context patterns, which can be used in other terminology processing applications.