Gradient and categorical constraints: evidence from Cantonese

Classical generative phonology draws a distinction between possible words (accidental gaps) and impossible words (systematic gaps) [4]. This model predicts that speakers should categorically judge accidental gaps to be well-formed and systematic gaps to be ill-formed, and thus rate accidental gaps more highly than systematic gaps in a wordlikeness judgment task. Such structure models attribute the listener’s differing judgments to abstract phonological generalizations, such as a ban on e.g. [coronal][coronal] sequences [5]. However, recent research into well-formedness suggests speaker judgments of both words and nonwords to be gradient in nature, and influenced by lexical as well as phonotactic factors [1] [2] [3] [6] [7]. This class of models, sometimes known as unit models, attribute gradient judgments to differences in experience, such as unit frequency or lexical density. However, most nonword research has focused on accidental gaps. We designed a wordlikeness judgment task to explore judgments of both systematic and accidental gaps in Cantonese.