Templatic Morphology and Indexed Markedness Constraints

The existence of morpheme-specific markedness constraints has been debated in Optimality Theory work on both templatic morphology and exceptions to phonological generalizations. Generalized template theory (GTT; McCarthy and Prince 1995) argues that morphemes that impose prosodic templates on outputs should be accounted for by the interaction of morpheme-specific faithfulness constraints and general markedness constraints, rather than by template-enforcing markedness constraints indexed to specific morphemes (e.g., RED=FOOT), as had previously been assumed. GTT therefore claims that no markedness constraints are morpheme-specific. More recently, the indexed-constraint theory of lexical exceptions has revived the possibility of morpheme-specific markedness constraints. This theory argues that constraints can be indexed to exceptional lexical items (Fukuzawa 1999, Ito and Mester 1999, 2001, Pater 2000, 2007); some proposals claim that indexed (i.e., morpheme-specific) markedness constraints are needed to account for certain exceptional patterns (Ota 2004, Pater 2000, 2007). Such proposals inherently conflict with GTT's claim that only faithfulness constraints may be morpheme-specific.

This squib provides a new source of evidence for morpheme-specific markedness constraints in Dinka (Nilotic, Sudan; Andersen 1995), where an exceptional prosodic template imposes a fixed mora count on vowels. An indexed-constraint analysis of this phenomenon cannot use indexed faithfulness constraints alone; rather, it crucially depends on indexed markedness constraints. This contradicts the GTT prediction that all morphological templates can be accounted for using general, rather than morpheme-specific, markedness constraints.

1 Morphological Vowel Length and Indexed Markedness Constraints in Dinka

Dinka verb forms are nearly all monosyllabic; verb roots are subject to a rich system of nonconcatenative morphology that mutates root vowels in various ways (data from Andersen 1995). Vowels contrast in voice quality (breathy or creaky ; no vowels have modal voice), tone (high, low, or falling), and length (verb roots may have one or two moras; morphologically complex verbs may have one, two, or three).

The following discussion will focus on vowel length and the ways it can be manipulated by morphology. Dinka morphemes can affect the length of verb root vowels in two possible ways: they can add moras or impose moraic templates. When mora-adding morphemes interact with morphological moraic templates, a template imposed by one morpheme can block lengthening that should be induced by another. Indexed markedness constraints will be crucial in explaining this blocking interaction.

1.1 Typical Patterns of Morphological Vowel Lengthening

Morphologically complex Dinka verbs may surface with trimoraic nuclei, despite the crosslinguistic rarity of such structures (Ladefoged and Maddieson 1996:320). Verb roots themselves, however, have maximally bimoraic nuclei. This inventory restriction can be accounted for if a markedness constraint against trimoraic vowels, *V_μμμ, is ranked above MAX-μ. This is shown in the hypothetical richness-of-the-base tableau in (3).

(1) *V_μμμ
A vowel cannot be linked to three moras.

(2) MAX-μ
Each mora in the input must have an output correspondent.

(3)

A number of morphemes add a mora to both monomoraic and bimoraic verb roots. Third person singular agreement (3SG) and centrifugal (CF, 'movement away from') are two such lengthening morphemes; of the 16 verbal morphemes described by Andersen (1995), 11 add moras to some or all roots.

(4) →    'kick.3SG'
→     'dust.3SG'
→     'roll.3SG'
→    'pull.3SG'

(5) →    'kick.CF'
→     'dust.CF'
→     'roll.CF'
→    'pull.CF'

Trimoraic vowels can have any combination of vowel quality, voice quality, and tone; a single verbal morpheme can trigger both lengthening and alternations in vowel quality, voice quality, and tone. CF, for example, can change a root vowel's tone in addition to inducing lengthening, as in (5). Vowel properties other than length will be largely ignored here, though see section 2 for their relevance to a possible analysis in terms of MORPHREAL, which demands that morphemes have some nonnull surface realization (Kurisu 2001).

When a lengthening morpheme applies to a bimoraic root, the resulting morphologically complex form has a trimoraic vowel. The input-output faithfulness constraint MAX-μ cannot force these marked trimoraic vowels to surface in morphologically complex words, as it was shown in (3) that *V...