Parallelism of constraint satisfaction has been argued to be a central, if not crucial, property of Optimality Theory (Prince & Smollensky 1993, McCarthy & Prince 1993, etc.). Here, however, I show that the adoption of parallelism in Optimality Theory (OT) requires the use of multiplanar representations. Specifically, this work examines affixes in Afar whose position in a string is variable. An example of this is shown with the second person marker in (1) which appears as a prefix on some verbs and as a suffix on others.
You (sg.) won.She drank milk. In this thesis I suggest that these affixes surface alternately as prefixes or suffixes in order to satisfy syllabification constraints in the language. Multiplanar representations account for the differing order of morphemes in forms such as those in (1) and also to account for contrasts such as y-uktub-e-n (3-read-aspect-plural [B123]) vs. nak-n-e ((drink milk-plural-aspect) [B125]). In particular, note that aspect (e) precedes plural (n) in the former verb but follows it in the latter. The analysis of these is problematic in Optimality Theory. The constraint hierarchy for y-uktub-e-n must require that plural (n) be the rightmost morpheme whereas the constraint hierarchy for nak-n-e must require that aspect (e) be the rightmost morpheme. This phenomenon can be accounted for in a nonlinear Multiplanar Model where a set of affixes occupies its own plane. The idea is that the affixes are aligned by morphological constraints in the order shown in (2).
Higher ranked onset, however, overrides the order of aspect and plural and forces plural leftmost to provide an onset just in case the affix plane would otherwise not have an onset.
(3) Affix order with onset ranked above morphological constraints a. person - aspect - plural: [y-e-n]
b. plural - aspect: [n-e]
The multiplanar analysis allows plural to precede aspect just in case an onset is needed on the affix plane. For y-uktub-e-n, a representation like that in (4) is proposed, where the person marker, [y], serves as the onset to both the affix plane and the word plane, and the order of the relevant affixes is that aspect precedes plural.
In [rab-n-e], the plural, [n], surfaces preceding aspect in order to fill the required onset position on the affix plane. In other words, the order is forced by a higher-ranked phonological constraint to be [n-e] instead of [e-n].
We drank milk. [B125]
This analysis is important because it is the first to show that OT requires three-dimensional representations. The emphasis in OT has been to move away from the use of input representations to account for linguistic generalizations (McCarthy 1995, Hammond 1995, etc.), e.g., the move away from CV tiers in nonconcatenative languages (Gafos 1995). Here I propose that the necessary representations are indeed found in the output, not in the input.
The layout of this thesis is as follows. In Chapter 1 I provide the empirical and theoretical frameworks. First I introduce Afar and its phonology and morphology. I then introduce Optimality Theory and its basic tenets, along with a discussion of Correspondence Theory and Generalized Alignment.
In Chapter 2 I analyze two prosodic phonological phenomena which play a role in the analysis of the variable-position affixes. First, as I argue that the variable-position affixes occur in various positions due to syllabification constraints, specifically onset and constraints which require CVX syllable structure, I motivate the constraints needed to account for syllabification. I argue that the location of variable-position affixes is due not only to morphological constraints, but to the fact that these affixes can provide onsets for syllables that would otherwise violate onset. I then analyze a phenomenon where [y]s do not occur after consonants. This is important to the analysis of the variable-position affixes because one of these affixes is a /y/ and its distribution is different from that of the other variable-position affixes.
In Chapter 3 I turn to an analysis of the variable-position affixes themselves. I argue that an OT model assuming parallelism of constraint satisfaction requires that Afar have at least three planes: one for roots, one for affixes and one for words.
In Chapter 4 I show the alternative within OT is to posit a model that consists of at least two levels. I then show the Multiplanar Model is preferable. First, it makes strong predictions about where the effects of phonological constraints are observed. These predictions are consistent with the data from Afar. Second, it is able to capture the fact that consonant co–occurrence constraints hold not only within roots but across affixes as well. Finally, it does not suffer from theoretical defects found in the Levels Model: the planes are morphologically defined and therefore could maximally have one level per morpheme. The Levels Model is not defined morphologically, however, and there is no a priori reason there could not be a different level for each constraint, making it a serial model. Finally, the Levels Model requires additional unmotivated constraints not required in the planar model.