Perceptually Grounded Faithfulness in Harmonic Serialism

1 Statement of the Problem

Steriade (2008) argues that faithfulness constraints in Optimality Theory are perceptually grounded: the faithfulness of a phonological mapping is directly proportional to the perceptual similarity between the input and the output of that mapping. For example, when a phonological process like place assimilation, voice assimilation, or deletion affects a medial consonant cluster /VC₁C₂V/, it usually targets C₁ rather than C₂.¹ The explanation is that changing C₁ is less of a change perceptually than changing C₂, because C₂'s prevocalic position gives it stronger perceptual cues than C₁. Formally, this difference in strength of perceptual cues is reflected in the ranking of faithfulness constraints: processes affecting C₁ violate lower-ranking faithfulness constraints than processes affecting C₂.

A mapping has an input and an output, and faithfulness constraints require an input and an output to compare. In the standard parallel version of Optimality Theory, referred to here as P-OT, the input to every mapping is the underlying representation and the output is the surface representation (Prince and Smolensky 1993/2004).

When this property of P-OT is combined with perceptually grounded faithfulness, a problem arises (Blumenfeld 2006, Flemming 2006, 2008a,b, Gallagher 2006, Jun 2002, McCarthy 2008b, Wilson 2001): underlying representations lack information that is important for perception. For example, the release of a stop consonant contains important perceptual cues for its place of articulation. In most if not all languages, the distribution of release is determined by the grammar, not the lexicon: stops are released prevocalically, but they may be unreleased (depending on the language) preconsonantally. If important perceptual cues like release are not already determined in underlying representation, how can faithfulness be perceptually grounded?²

In Steriade's original proposal, this problem does not arise because the connection between perception and faithfulness is indirect. Faithfulness constraints are sensitive to the contexts in which perceptual cues are found, not to the perceptual cues themselves. C₂'s greater resistance to change is attributed to the universal rankings in (1). A mechanism called the P-Map is responsible for these rankings. It links the perceptual cues with the constraint system.³

1. 1. Perceptually grounded faithfulness rankings (after Steriade 2008)

   1. a. Ident(place)/_____ V ≫ Ident(place)/______ C

   2. b. Ident(voice)/_____ V ≫ Ident(voice)/______ C

   3. c. Max/_____ V ≫ Max/_____ C

So, for example, /amta/ assimilates to become anta and not *ampa because violating Ident(place)/_____ V is always worse than violating Ident(place)/_____ C.

This indirect approach has a problem that was first recognized by Wilson (2001). The problem arises whenever a phonological process changes the conditions controlling the distribution of the relevant perceptual cues. Syncope is an example of such a process. When the medial vowel of /VC₁VC₂V/ deletes, it changes C₁ from prevocalic (strong cues) to preconsonantal (weak cues) position. The effect that syncope will have on constraints like Ident(place)/_____ V depends on whether they check their contexts at underlying or surface representation. I will consider each of these options in turn.

If these constraints check their contexts at underlying representation, clusters derived by syncope can behave differently than otherwise identical clusters that are already present in underlying representation. Although the universal ranking in (1a) tells us that /amta/ assimilates to anta and not *ampa, it does not tell us what to do with /amita/ in a language with i syncope. Both anta and ampa are equally faithful if Ident(place)/_____ V checks its context in underlying representation, because both /m/ and /t/ are prevocalic. Context-checking at underlying representation therefore predicts inconsistency in the direction of assimilation with clusters derived by syncope. This inconsistency could take the form of actual variation (anta ∼ ampa) or it could take the form of dependence on other constraints (e.g., /amita/ → anta vs. /unipe/ → unte if coronals are less marked than labials).

The predicted inconsistency is not observed. The data in (2) show that C...