Are corpus-based predictions mirrored in the preferential choices and ratings of native speakers? Predicting the alternation between the Estonian adessive case and the adposition peal 'on'

Abstract

Recent work in usage-based linguistics stresses the importance of combining corpus-based analyses with experimental studies. A number of studies have compared the performance of a corpus-based statistical model against the behaviour of native speakers in a linguistic experiment. The present paper takes this line of analysis further by combining corpus-based work with two sources of experimental data. A mixedeffects logistic regression model is fitted to the corpus data of the Estonian adessive case and the adposition peal 'on' in present-day written Estonian. In order to evaluate the goodness of the corpus-based model, its performance is compared to the behaviour of native speakers in a forced choice task and a rating task.