This paper investigates why two classes of French words, both of which contain words that in isolation are phonetically realized as vowel-initial, vary with respect to elision when pronounced right-adjacent to a definite article: one class does not trigger the deletion of the expendable vowel, while the other class does. It is argued that these two classes differ in that one contains underlyingly vowel-initial words, while the other class consists of words whose underlying representations contain an underspecified consonant segment (termed a ghost consonant; Kiparsky, 2003). The paper also addresses how a learner could posit a ghost consonant in the underlying representation of this second class of words given that the ghost consonant is phonetically null. Through Inconsistency Detection (Tesar, 2004) and Presence Feature setting (Nyman & Tesar, 2019) the learner is able to posit a phonetically null segment, resulting in the resolution of a ranking paradox that would otherwise obtain.
"Learnability of a Phonetically Null Segment,"
University of Pennsylvania Working Papers in Linguistics: Vol. 27
, Article 21.
Available at: https://repository.upenn.edu/pwpl/vol27/iss1/21