First Steps Toward Developing a System for Terminology Extraction

The aim of this paper is to describe first steps in developing a system for terminology extraction. First a data sample is built from synopses of doctoral theses at the Faculty of Humanities and Social Sciences, University of Zagreb, accepted in the period from 2004 to 2009 written mostly in Croatian language. Data sample consists of 420 documents and 338, 706 tokens. A small sample was manually tagged for terminology to be used in an initial experiment. The approach for terminology extraction is knowledge-driven and consists of differential analysis of reference and domain-specific corpora. Specific method used is log-likelihood ratio test. Experiment deals with different reference corpora and linguistic pre-processing. First results are promising. Further research guidelines are discussed.