1/25
Looks like no tags are added yet.
Optimality Theory, introduced in Prince and Smolensky (1993/2004)
is a constraint-based theory that is surface-focused. Interaction of constraints that refer only to output (surface) forms can account for rule conspiracies where a number of rules create the same surface form.
input
UR
G EN
generates an infinite number of outputs for the UR
universal candidate generator, constructs candidate output forms
CON
universal constraint set
H
a constraint hierarchy, language-specific
EVAL
the judge
evaluates output candidates, selects the optimal one
output
winner!
Universality
Universal Grammar supplies a set of constraints.
Violability
Constraints are violable, but violation is minimal
Ranking
Constraints are ranked on a language-particular basis; the notion of minimal violation (or best satisfaction) is defined in terms of this ranking
Parallelism:
do any changes all at once
Best-satisfaction of the constraint hierarchy is computed over the whole hierarchy and the whole candidate set. Candidates may differ from the input by showing the effect of multiple notionally distinct processes simultaneously
Markedness constraints
refers only to the output surface
are formulated as very general statements barring marked or complex structures. For example, *NOCODA is a constraint barring segments from occurring in syllable coda. *VOICEDOBSTRUENT is a constraint barring the occurrence of voiced obstruent’s.
Faithfulness constraints
if we only had markedness constraint, we wouldnt have any hard words
says the input needs to look like the output
require the output (surface form) to resemble the input (underlying form). Faithfulness constraints are formulated in terms of the relationship between input and output representations. For example, IDENT IO- VOICE requires that any value for the feature [voice] that is in the input must be present in the output.
M AX – IO
input segments must be present in the output (don’t delete)
maximizing the input
D EP – IO
output segments must correspond to segments in the input (don’t insert)
says dont insert anything
Constraints
universal, assumed to be present in all languages, focus on the output, specify banned or marked surface structures without linking them to a particular change, can be violated
Rules
language specific, the change demanded by the rule and the structure undergoing the rule are linked, are always obeyed at some level of representation
Derivation vs. Parallelism
: In rule-based theories rules are ordered and apply in sequence to an underlying representation. The output of the first rule is the input to the following rule and so on. There are intermediate levels of representation between the UR and the PR.
are there sequential derivations in OT?
there are no sequential derivations. All constraints apply simultaneously to an input form. The grammar generates an infinite number of potential outputs. Output candidates are evaluated by the constraints. Because constraints are often conflicting constraint ranking is needed in order to determine the output.
this makes sense
How to make and read tableaux:
Constraints:
• one in each column (except the first)
• presented with highest-ranked constraints on the left
Inputs and outputs:
• listed in the first column
• input in the first row
• one output candidate on each following row
Violation marks:
• for each constraint, for each candidate, does the candidate violate the constraint?
if yes, put a * in the cell
it the candidate violates the constraint more than once, put one * in the cell for every violation
if no, leave it empty
Selection of optimal form (winner):
starting with the highest (leftmost) constraint:
does any candidate violate the constraint?
are there any candidates that satisfy the highest ranking constraint?
if there are candidates that satisfy the highest ranking constraint, eliminate any candidate that violates it
indicate fatal violations with !
repeat for each successive lower-ranked constraint until only one candidate is left (has not incurred a fatal (!) violation)
that candidate is the output
examples of constaints:
*VNASAL
*VORALN
IDENTIO- NASAL
– A vowel must not be nasal
– a vowel must not be oral when followed by a nasal in the same syllable.
– Corresponding segments in the input and output must have identical
specifications for the feature [nasal]. (faithfulness)
when do you insert a fatal “!”
when no other candidate is in violation of that category
work left to right
valid ranking argument requires: (3)
a conflict; the constraints being ranked must conflict in their analysis of some output candidate
a winner; one of a pair of competing candidates must be the actual output for a given input (the winner). No ranking arguments can be made on the basis of comparison of losers.
the absence of any additional constraint which could be responsible for the winner being evaluated as better than the loser
can unviolated constraints be ranked to one another?
no
In order to illustrate your analysis with a ranking argument, you will need the following candidates:
the faithful candidate (The faithful candidate should always be included. It may also sometimes fall into one of the following categories. It can be the winner or it could be an informative loser.)
the winner (actual output)
informative losers; in order to establish a constraint ranking CONSTRAINTA >> CONSTRAINTB, start with the winner and eliminate one or more violations CONSTRAINTB and add one or more violations of CONSTRAINTA, without adding violations of any additional constraints unless they are independently known to be ranked below CONSTRAINTA
Method of Mark Eliminability
Look at every violation mark incurred by the winning candidate in your tableau. Think of all the ways that those violations could be eliminated and make sure that all candidates that would eliminate those violations do worse than the winner on some higher ranked constraint.