Chapter 2 Literature Review
2.1 Optimality Theory
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
7
Chapter 2
Literature Review
2.1. Optimality Theory
This section gives a brief introduction of Optimality Theory as well as Cophonology, a version of OT.
2.1.1. Basic Concept
Optimality Theory (henceforth “OT”) was first introduced by Prince and Smolensky around 1990. As an alternative model in generative phonology, OT assumes that there is a Universal Grammar among languages of the world. The universal grammar contains a set of violable constraints that exists in all languages, and the systematic differences between languages are accounted for by the different rankings of such a set of universal constraints.
OT assumes that constraints are violable (Violability), and that the candidate that has the minimal violations of constraints will be the optimal output. Based on this assumption of constraints, if the constraint A is ranked higher than the constraint B, a candidate that violates the constraint A but satisfies the constraint B will not be chosen as the optimal output. On the contrary, a candidate may be chosen as the optimal output if it violates the constraint B but satisfies the constraint A.
In OT, the evaluation of the input and output mapping is done with the parallel application of constraints (Parallelism) rather than with the step-by-step application of rewrite rules. All possible outputs are generated by GEN at the same time and are
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
evaluated by the constraints simultaneously. There are no intermediate levels.
There are two major constraint types in OT: markedness constraints and
faithfulness constraints. The former require an output to conform to the phonological restrictions in the language, and the latter confine the identity between an input and its correspondent output. The two kinds of constraints will interact and compete, forming a ranked constraint set to choose the optimal output of an input.
The operation of OT is presented schematically in (1).
(1) Input
GENERATOR
Candidate A, Candidate B, Candidate C …
EVALUATION
{Constraint 1 >> Constraint 2 >> Constraint 3 …}
Optimal Candidate
In OT, the outputs are not derived by applying phonological rules step by step. Rather, all possible outputs (named as candidates in OT) for an input are produced by GEN (Generator). All candidates will be evaluated by EVAL (Evaluator). EVAL contains a set of universal constraints, and the constraints are ranked with one another. The evaluation of candidates is done in a parallel way, and the one with the minimal numbers of violations will be chosen as the optimal output.
In OT, the interaction of constraints and the violations that are incurred by candidates are displayed in tableaux, as shown in (2).
‧
In the tableau, constraints are placed on the top of the row, candidates are put in the right side of the tableau, and the symbol ☞ next to the candidate indicates that it is the winner, that is, the optimal output in this tableau with such constraint ranking. The dotted line means that constraint 2 and constraint 3 are mutually unranked in the ranking hierarchy. The four listed constraints are ranked as follow: Cons1 >> Cons 2, Cons 3 >> Cons 4. Violations are presented with an asterisk mark ‘*’, and a crucial violation is indicated by ‘!’. In addition, shading appears in the column under a constraint if it is not crucial in determining the optimal output. As shown in tableau (2), candidates (a), (c), and (d) violate Cons3, Cons2, and Cons1, respectively. Since these constraints are ranked higher than Cons4, these candidates are ruled out.
Candidate (b) is chosen as the optimal output, although it violates Cons4. This conforms to the premise of OT that constraints are violable.
2.1.2. Feature Assimilation Constraint
Assimilation is often discussed under theories of autosegmental phonology.
McCarthy (2008b), on the other hand, developed an OT analysis, and proposed a constraint named as SHARE(F).
(3) SHARE(F): Assign one violation mark for every pair of adjacent segments that are not linked to the same token of [F].
Regarding the locution of ‘not linked to the same token of [F]’ in (3), McCarthy
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
generalized the situations where violations occur, which are summarized below. Here
‘S’ stands for some arbitrary segments.
(4) SHARE(F) is violated in condition
if [F] is equipollent if [F] is privative
(a) (b) (c) (d) (e)
[αF1] [-αF1] [αF1] [αF2] [F] [F1] [F2]
S S S S S S S S S S
Any candidate that, depending on whether [F] is equipollent or privative,
conforming to one of the situations above will violate SHARE(F). For features that are equipollent, take the [ATR] feature as an example. The adjacent segments violates SHARE(F) if one is a [+ATR] segment and the other is [-ATR] one, as in (4a), or if one is a [+ATR] segment and the other is a [+round] segment, as in (4b).
As for features that are privative, take the [nasal] feature as an example. The adjacent segments violates SHARE(F) if one is linked to a [nasal] autosegment but the other is not, as in (4c), or if none of the two adjacent segments is linked to any [nasal]
autosegment, as in (4d), or if one is linked to a [nasal] autosegment and the other is linked to a [voice] autosegment, as in (4e).
This constraint is employed to account for the substitution of coda nasals in this thesis, and the shared feature is [back]. According to McCarthy, the [back] feature is one of ‘the best cases against privativity’. However, he also assumed that features involved in the harmony phenomenon are privative. We will follow this assumption and viewed the [back] feature as privative when evaluating candidates.
‧
As a version of OT, Cophonology was proposed by scholars to account for the nature of the diversity, such as registers, dialects, and free variations, within a
language. Orgun (1996), Anttila (1997), and Inkelas and Zoll (2007), etc. claimed that there are different phonological systems that co-exist in a language, and the difference lies in the different rankings of a single set of constraints.
The innovation of cophonology is better represented by a grammar lattice, as in (5). In the grammar lattice, the core grammar of a language L is placed in the
superordinate node, that is, ‘Master Ranking’. Constraints in the core grammar are partially ranked. That is, Cons1 dominates Cons2 and Cons3, but the ranking of Cons2 and Cons3 are not specified. Such specification is determined in the two subgrammars in language L, they are, Corphonlogy A and Corphonology B. In
Corphonlogy A, Cons2 dominates Cons3, while in Corphonology B, Cons3 dominates Cons2. Variations in language L are accounted for by these different rankings.
(5) A grammar lattice of language L (based on Inkelas and Zoll 2007) Master Ranking
Cons1 >> {Cons2, Cons3}
Cophonology A Cophonology B
Cons1 >> Cons2 >> Cons3 Cons1 >> Cons3 >> Cons2
The concept of cophonology is employed in analyzing Japanese loanwords in Taiwanese in this thesis. Variations are found in some of the segmental substitutions.
For example, [as.sa.ɾi] ‘frankly’ can become [a.sa.liʔ] as well as [at.sa.liʔ] in
Taiwanese. The two substitutes are in free variation, and they are properly accounted for by two different rankings of a single constraint set.