Papers from the Mókus Conference [antikvár]

GRADIENT PHONOTACTIC ACCEPTABILITY A case study from SlovakZsuzsanna Bárkányi Research Institute for Linguistics, HAS, Budapest bzsuSnytud.huAbstractPhonotactic well-formedness judgments are usually gradient, the theoretical interprétation of which is controversial in the phonological literature. In this study we present experimental evidence from Slovak that speakers do have intuitions about unattested grammatical forms as well as attestai ungrammatical or marginal ones and these intuitions can be modeled fairly closely by gradient phonotactic grammars like, for instance, the HayesWilson Phonotactic Learner. Our results suggest similarly to Albright's (2007) findings that gradient phonotactic judgments re-flect knowledge of the relative probability of various combinations of natural classes, rather than knowledge of words in the lexicon directly. Our data was contrasted with Bailey and Hahn's Generalized Neighborhood Model, a lexicon-only approach, as well. We pay special attention to sonority reversai clusters in Slovak and claim that these sequences, although attested in the language, are ungrammatical and thus prone to change.1 IntroductionWestern Slavic languages, among them Slovak, are well-known for allowing "exotic" consonant clusters word-initially that are absent in other Indo-European languages, and are rare cross-linguistically, üke #tk-, #lp-, #mdl- or #pstr-,Among (Western) Slavic languages the study of Polish word-initial clusters h as been in the focus of phonological interest for over 50 years (see Kurylowicz, 1952; Rubach and Booij, 1990; Cyran and Gussmann, 1999; Rowicka, 1999; Scheer, 2004; etc.). Phonologists generally view the patterning of word-initial consonant clusters in a language as the resuit of a categorical grammar, and aim to account for existing and non-existing consonant sequences in terms of natural classes and rules or constraints that regulate them. However, any analysis of Slavic especially Polish or Czech word-initial consonant clusters to date has failed to fully identify existing and non-existing clusters as natural classes. As Scheer (2006) puts it: "the picture appears to be entirely anarchie, disobeying any possible organizing principle." In Scheer (2007) he concludes that there are and can be only two types of word-initial onset grammars: "TR-only" languages like English, where sonority increases in all word-initial clusters, and "anything-goes" languages where all possible combinations of T's