Grammars or corpora? Who should we trust? Empirical analysis of morphological doubletism in Croatian.

Morphological doubletism refers to a phenomenon when two (or more) variants inhabit a single cell of an inflectional paradigm (e.g. forms dove and dived can both occur in the cell for the past tense of the English verb to dive). In this paper we analyse this phenomenon in Croatian, a South Slavonic language, by juxtaposing several sources of linguistic data. First we look at how they are treated in grammar books of Croatian. This is followed by a corpus analysis of some examples of doublets in present-day language. More often than not, one can find opposing information in these two sources. Finally, we present the results of a native speakers’ intuitions study. The results show that speakers are highly sensitive to the relative proportions of the two forms as found in the corpus; however, in some instances these frequency effects are cancelled out by analogical effects – phonologically similar forms are given similar ratings even when their corpus distributions are completely inverse.