Václav Cvrček

Automatic Term Recognition Based on Data-Mining Techniques

We present a new method for automatic term extraction which is based on training datasets created to build inductive models for term identi¿cation. Existing approaches employ simple statistical and linguistic rules designed merely ad-hoc and are unable to utilize complex relations of linguistic units. In contrast to those approaches, our method does not require such manually ascribed rules of extraction. The data for our research is taken from the Czech National Corpus which is lemmatised and morphologically tagged. Statistical information (frequency, distribution etc.) is generated automatically and thus the only expert contribution needed is to label terms in the training dataset.The data mining software creates models that perform the extraction without any further human input. Additionally, feature ranking can serve as valuable aid for understanding of the extraction process and its future development and in terminology research.