Abstract
After briefly discussing the heterogeneities inherent to language production and how they influence corpus evidence, we describe a scale for the classification of individual morphological variants by their relative frequencies that has recently been independently proposed in Mluvnice současné češtiny (2010) (A Grammar of Contemporary Czech, hereafter GCCz), of which we are co-authors, and in Bermel & Knittl (2012). Those variants with relative frequency (roughly) within 1% and 10% are classified by the respective authors as “sparse” and “marked”, and those occurring in (roughly) less than 1% cases as “unexpected” and “isolated”.
Another feature of the scale is the “equipollence” of variants of a doublet having relative frequencies within (roughly) 1/3 and 2/3 (for this criterion see also Štícha 2009). The scale in GCCz is heuristically based on Shannon entropy and valid for synchronic functionally equivalent variants.
Recently, R. Čech (2012) has claimed to have revealed “a serious statistical deficiency” in GCCz. We show that this is a misunderstanding stemming from his not distinguishing between the null-hypothesis statistical significance testing and the effect size evaluation.
We end with a brief note on the structure of the resources employed in GCCz.