Annotated corpus and the empirical evaluation of probability estimates of grammatical forms

The aim of the present study is to demonstrate the usage of an annotated corpus in the field of experimental psycholinguistics. Specifically, we demonstrate how the manually annotated Corpus of Serbian Language (Kostić, Đ. 2001) can be used for probability estimates of grammatical forms, which allow the control of independent variables in psycholinguistic experiments. We address the issue of processing Serbian inflected forms within two subparadigms of feminine nouns. In regression analysis, almost all processing variability of inflected forms has been accounted for by the amount of information (i.e. bits) carried by the presented forms. In spite of the fact that probability distributions of inflected forms for the two paradigms differ, it was shown that the best prediction of processing variability is obtained by the probabilities derived from the predominant subparadigm which encompasses about 80% of feminine nouns. The relevance of annotated corpora in experimental psycholinguistics is discussed more in detail