Az ember, a korpusz és a számítógép: Magyar nyelvű szóhasonlósági mérések humán és disztribúciós szemantikai kísérletben

The paper reports on the results of two word similarity experiments. The first experiment is a subjective human test: similarity values for 31 pairs of Hungarian words have been collected from 28 subjects. The test method comes from Rubenstein & Goodenough (1965) and it reflects the intuition that word similarity is a continuum from clear cases of synonymy to the complete lack of apparent similarity. The Hungarian results correlate very well with the data collected by Rubenstein and Goodenough (Spearman r=0,959, p<0,01) and also with the English replica experiments (Miller & Charles 1991 and Resnik 1995). In the second experiment presented here, a computer program collected similarity data for the same words, based on the context in which they typically occur. The correlation between the subjective and the corpus-based data series is r=0,591 (p<0,01).