Modeling a phonotactic approach to segment recovery: The case of Japanese high vowels
James Whang (Seoul National University)
Kakeru Yazawa (University of Tsukuba)
Abstract
Japanese listeners often have difficulties perceiving consonant clusters accurately and
report hearing a vowel between the consonants, despite Japanese speakers producing
numerous consonant clusters that result from a highly productive high vowel devoicing
process. This poses a substantial challenge for phonological learning, as the task is to
learn the strong CVCV preference in Japanese based on surface consonant clusters that
violate this very preference. The current study investigates this learnability issue by
building a computational model that induces Optimality Theoretic (OT) phonotactic
constraints based strictly on overt speech. Two versions of the model are tested: one
with classic OT-style faithfulness constraints that penalize violations, and one with
positive constraints that reward sequences conforming to the constraints. Both have
traditional markedness constraints. The results show that positive constraints are
superior to faithfulness constraints in modeling phonotactic repair.
Keywords
acquisition, computational phonology, Japanese, perceptual repair