|Title:||Cues or Contexts in Feature Licensing Constraints|
|Abstract:||This paper explores a problem raised by current research into the phonetic
bases of sound patterns. Steriade (1993, 1995, 1996) has proposed that the
distribution of phonological contrasts is explained, not by a principle of
Prosodic Licensing (Ito 1986, Goldsmith 1990) (licensing of features only
in certain prosodic positions, e.g. onset position), but rather by the
presence of sufficiently audible phonetic cues in the relevant contexts
that signal these contrasts (these can be expressed as feature/cue
cooccurrence constraints). The problem, as Steriade observes, is that the
presence or absence of a cue in a given token of some utterance is often
not dispositive of the distribution of a licensed feature specification.
Rather, it is whether the licensing cue is TYPICALLY present or absent.
My proposed solution involves consideration of the distinct effects of
feature cooccurrence constraints (and surface well-formedness constraints
generally) on speech recognition and production. I show that surface
well-formedness constraints play no direct role in the recognition task:
only input/output faithfulness constraints are active. Consequently,
hearers can correctly recognize a contrastive distinction produced by
someone else, although it would be neutralized in the output of their own
grammar, due to the absence of the licensing cue in that token (cf.
Smolensky 1996, contra Reiss and Hale 1996). As an aside, I consider how
OT grammars can be applied to certain problems of speech recognition,
including interspeaker variation and the role of top-down knowledge.
As for production, the stability of the feature, notwithstanding variation
of in the presence of the licensing cue in particular tokens, can be
understood in terms of the effect of the feature cooccurrence constraint on
underlying representations, a species of Stampean Occultation (Prince and
Smolensky 1993, ch. 9).