Corpus frequency and acceptability judgments: A study of morphosyntactic variants in Czech

Using data from a 100-million-word representative corpus and a large-scale acceptability survey, we have investigated the relationship between corpus data and acceptability judgments. We conclude that the relative proportions of morphosyntactic variants in a corpus are the most significant predictor of a variant's acceptability to native speakers, and that in particular high relative proportions of one variant in a corpus are reliable indicators of high acceptability to native speakers. At the same time we note the limits of this predictability: low-frequency items, as noted elsewhere in the literature, often enjoy high levels of acceptability. Statistical preemption thus appears as a more limited phenomenon than had heretofore been posited.