Phonetic features guide phonotactic generalizations in perception.

Abstract

Although features are useful theoretical constructs (they allow parsimonious descriptions of phonological inventories, patterns, and alternations), their etology remains debated. Three hypotheses have been advanced: (1) Phonological patterns involving phonetically natural classes arise from historical changes affecting similar sounds, but features have no psychological reality; (2) features emerge in the listeners’ phonology on the basis of phonetic and phonological experience (including exposure to patterns resulting from historical changes); (3) features are abstract symbols provided by universal grammar (UG), and they are independent of phonetic implementation. To test these alternatives, French and English listeners were exposed to a constraint on obstruent voicing, and tested implicitly on their generalization of the constraint to untrained obstruents. If the historical hypothesis is true, no generalization would occur. Since voiced stops and fricatives form a phonetic natural class in French but not in English, emergent features would favor generalization only in this language. In contrast, voiced stops and fricatives are natural classes in terms of abstract phonological features in both languages; therefore, the UG hypothesis predicts both groups will generalize. Current results support the emergent hypothesis [French generalized: t(11) = 2.5, p 0.1], but additional data are being collected.