The 7th Conference Terminologie etIntelligence Artificielle (TIA'2007), October 8-9, 2007, Sophia Antipolis, France
term extraction; distribution similarity; interactivity
In the context of an automatic term extraction system, our study is on the impact of using the choices of correct terms, as indicated by a user, to reorder a list of candidate terms. To establish an interterm relationship between the chosen terms and the proposed candidates, we are exploring the distributional similarity that allows for the expression of the tendency of lexical units to appear together in a corpus. Distributional similarity can serve first to direct the terminologist's attention to subthemes, but, above all, it has a more global impact by increasing the precision of the candidates at the top of the list of candidate terms. We demonstrate this by means of an evaluation in the field of machine translation using a gold standard, as established by ten experts in the field. In this experimentation, the precision of candidate subsets at the top of the list increases by 3% to 11%, depending on the size of these subsamples.
Proceedings of the 7th Conference Terminologie etIntelligence Artificielle (TIA'2007).