Abstrakt
An initial probe of Czech conversational data reveals a very strong negative correlation between the identifiability of words (extracted from complete utterances, which are readily comprehensible as a whole) and the word-reduction rate. Our data show a typical hearer to understand the entire utterance without difficulties but to identify only between 37-40% of words out of their sentential context.
We work with the hypothesis that there will be detectable patterning in the interplay between reduction rates, syntactic organization, and various types of chunking, the latter leading to the notion of 'lexicalized reductions'. The results of a pilot experiment suggest that utterance comprehension might involve a combination of syntactic linearization (in Czech based on information structure encoding), semantic/functional predictability of words in particular positions due to syntactic bootstrapping (e.g. syntagmatic predictability based on complement structure), as well as inherently salient phonetic features independently of grammatical structure.