Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.

What Is Optimality Theory?1

John J. McCarthy*
University of Massachusetts Amherst

Abstract
Optimality Theory is a general model of how grammars are structured. This
article surveys the motivations for Optimality Theory, its core principles, and the
basics of analysis. It also addresses some frequently asked questions about this
theory and offers suggestions for further reading.

1. Introduction
In 1991, Alan Prince and Paul Smolensky began presenting their work on
a new approach to language. By 1993, this new approach had a name –
Optimality Theory (OT) – and it became known through their widely
circulated manuscript ‘Optimality Theory: Constraint Interaction in Gen-
erative Grammar’ (Prince and Smolensky 1993/2004). The impact of this
work on the field of phonology was extensive and immediate; since 1993,
it has also stimulated important research in syntax, semantics, sociolinguistics,
historical linguistics, and other areas. Optimality Theory would probably
go on anyone’s list of the top five developments in the history of generative
grammar.
This survey article sketches the motivations for OT, its core principles,
and the basics of analysis. It also addresses some frequently asked questions
about this theory. Because any survey of such a broad topic is necessarily
incomplete, each section concludes with suggestions for further reading.

2. The Motivations for OT

Since the early 1970s, it has been clear that phonological and syntactic
processes are influenced by constraints on the output of the grammar. Two
different kinds of influence can be identified:

(i) Processes can be blocked by output constraints. In Yawelmani Yokuts

(now usually called Yowlumne), syllables are maximally consonant–vowel–
consonant (CVC). A process of final vowel deletion is blocked from
applying when it would produce an unsyllabified consonant (Kisseberth
1970): /taxa:-k%a/ → [ta.xak%] ‘bring!’ vs. /xat-k%a/ → [xat.k%a], *[xat.k%]
© 2007 The Author
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 261

‘eat!’ (the period/full stop marks syllable boundaries. Unsyllabified con-

sonants are in italics). In English, wh movement is blocked when it would
create a that-trace sequence (Chomsky and Lasnik 1977): Who did you say
t left? vs. *Who did you say that t left?

(ii) Processes can be triggered by output constraints. In Yawelmani,

unsyllabifiable consonants in the input are fixed by epenthesizing a vowel:
/%ilk-hin/ → [%i.lik.hin], *[%il.k.hin] ‘sings’. In Spanish, clitics are
moved to satisfy an output constraint requiring that second person
pronouns precede first person, regardless of whether they refer to the
direct object (DO) or indirect object (IO) (Perlmutter 1971): TeIO meDO
presento/TeDO meIO presentas ‘I introduce myself to you’/‘You introduce
yourself to me’.

Although the blocking and triggering relationships between the processes

and the output constraints are easy to grasp at an intuitive level, it is not
obvious how to express these relationships formally in linguistic theory.
In fact, phonological and syntactic theory, which developed along parallel
lines until the 1970s, diverged on exactly this point.
The mainstream syntactic approach to blocking and triggering is
represented by works like Chomsky and Lasnik (1977). All transformations
are optional. An input to the grammar – that is, a deep structure – freely
undergoes any, all, or none of the transformations, and the result is submitted
to the surface-structure constraints, called filters. Any candidate surface
structure that satisfies all of the filters is a well-formed sentence of the
language. A transformation T is in effect blocked whenever a surface filter
rules out the result of applying T. Similarly, T is in effect triggered whenever
a filter rules out the result of not applying T. The net result is that much
of the burden of explaining syntactic patterns is shifted from the theory
of transformations to the theory of filters, which is arguably better
equipped to produce such explanations. In Government-Binding (GB)
theory, the theory of transformations ultimately withered away to just
Move α (Chomsky 1981).
The filters model (hereafter FM) was extremely influential in syntax,
but not in phonology. The main reason, in my view, is that FM required
all filters to be inviolable, and this assumption will not work with typical
phonological data. (It does not work in syntax either, but that limitation
was not so immediately apparent.)
Yawelmani can be used to illustrate this. In FM, as we saw, obligatory
application of a process is obtained by positing an output constraint against
structures where that process has not applied. In Yawelmani, epenthesis is
obligatory before consonants that cannot be syllabified within Yawelmani’s
restricted CVC syllable canon: /%ilk-hin/ → [%ilikhin]. In FM, this means
that the epenthesis transformation is optional, but failure to epenthesize is
marked as ungrammatical by a constraint against unsyllabified consonants
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
262 John J. McCarthy

in the output: *[%il.k.hin]. Final vowel deletion is another obligatory process,

so some output constraint must be marking *[ta.xa:.k%a] as ungrammat-
ical. An obvious candidate for this constraint is a general prohibition
on final vowels, *V#. But this cannot be correct because it also marks as
ungrammatical those well-formed surface structures where deletion has
correctly failed to apply, such as /xat-k%a/ → [xat.k%a]. So the constraint
responsible for the obligatoriness of deletion in /taxa:-k%a/ → [ta.xak%]
has to be something more specific: a prohibition on final vowels that are
preceded by a single consonant, *VCV#.
This seemingly innocent analytic move disguises a fundamental failure
of explanation. The constraint *VCV# stipulates something that should
be accounted for independently: final vowel deletion is blocked in a VCCV#
context because then it would produce an unsyllabified consonant: /xat-
k%a/ → *[xat.k%]. With inviolable constraints there is no way of saying that
*V# is enforced unless enforcing it would produce an unsyllabifiable con-
sonant. There is no way of establishing such priority relationships between
two inviolable constraints; at least one of the constraints (the lower priority
one) has to be violable for priority relationships to make any sense.
Another reason for FM’s lack of influence in phonology is the continuing
force of the phonological research program defined by The Sound Pattern
of English (SPE) (Chomsky and Halle 1968). This program strongly influ-
enced the kinds of explanations that could be entertained. SPE’s central
hypothesis, embodied in the evaluation metric, is that rules statable with
a few features have greater explanatory value than rules requiring more
features, ceteris paribus. The SPE theory supplies abbreviatory conventions
that capture generalizations by allowing certain rules to be stated in a
more compact form. Kisseberth (1970) proposed a theory of blocking
effects along these lines. The idea is that the output constraint against
final clusters can be used to abbreviate the rule of final vowel deletion,
simplifying it from V → Ø /VC___# to just V → Ø /___#. The problem
with this proposal is that it does not extend to the triggering side of
conspiracies. Yawelmani’s epenthesis rule Ø → i/C___CC could not be
abbreviated as just Ø → i, because the theory lacked any sort of economy
mechanism to ensure that Ø → i applies only when it is needed and not
otherwise.
Novel theories of phonological representation, which rose to prominence
beginning in the mid-1970s, led to progress of a different sort on the
blocking and triggering problems. As phonological representations became
increasingly complex, it became possible to imagine an almost rule-less
phonology in which automatic satisfaction of universal constraints on
representations was all that mattered. Goldsmith (1976a,b) and Prince (1983)
developed proposals along these lines for autosegmental and metrical
phonology, respectively. This work ran headlong into another problem,
however: the proposed universal constraints did not hold in every language
all of the time. That is why the subsequent literature on autosegmental
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 263

and metrical phonology, such as Pulleyblank (1986) and Hayes (1995),

places much more reliance on rules than constraints.
This last point brings us to another of the primary motivations for OT:
the problem of constraint universality or the lack of it. In both phonology
and syntax, initially plausible candidates for universal constraints often
become increasingly dubious as they are surrounded with an apparatus of
hedges and parameters to deal with empirical challenges.
The constraint requiring all syllables to have initial consonants [Ito’s
(1989: 222) Onset Principle] supplies a nice example of how a good
constraint can go bad if parametrization is taken seriously. In Arabic or
German, every syllable must have an onset, tout court, and [%] is epenthesized
whenever it is needed to ensure that outcome. In Timugon Murut (Prentice
1971), on the other hand, onsetless syllables are tolerated, so the Onset
Principle needs to be parametrized: [onsets required: yes/no]. In Axininca
Campa (Payne 1981), onsetless syllables are tolerated word initially, but not
word medially, so a further parameter is required: [onsets required medially:
yes/no]. Dutch is similar to German, except that it allows syllables to be
onsetless if they are unstressed (Booij 1995: 65), so yet another parameter
will be necessary. Of course, no analyst has seriously proposed such
nuanced parametrization of this constraint, but the facts would seem to
demand it of anyone with a serious commitment to a parametric theory
of phonological typology.
Even such complex parametrization is doomed to fail, however,
because constraints can still be active in languages that seem to violate
them freely. Timugon Murut, as was just mentioned, allows onsetless
syllables: [am.bi.lu.o] ‘soul’. Onsetless syllables are nevertheless avoided in
two circumstances. A single intervocalic consonant is syllabified as an onset,
such as the [l] of [am.bi.lu.o], not *[am.bil.u.o]. And reduplicated syllables
must have onsets, even at the expense of infixing the reduplicative
morpheme (in boldface) by copying the second syllable instead of the first
one: [a.ba.lan] ‘bathes’ reduplicates as [a.ba.ba.lan], not *[a.a.ba.lan] (with
consonant-initial words, infixation is not necessary to get an onset, so
[bu.lud] ‘hill’ reduplicates as [bu.bu.lud]). Examples like these show that
a constraint can remain ‘on’ in the phonology of a language even when
it would seem to have been turned off parametrically.
By the end of the 1980s, there was certainly a consensus about the
importance of output constraints, but there were also major unresolved
questions about the nature and activity of these constraints. That ‘conceptual
crisis at the center of phonological thought’, as Prince and Smolensky
(1993/2004: 2) refer to it, was not very widely acknowledged at the time,
but in hindsight it is hard to miss. It is a major feature of the intellectual
context in which OT was developed.
Suggestions for further reading. McCarthy (2002: 48–65) is a short
overview of the intellectual context of OT. Some of the main issues that
arose in the field of phonology in the years after the publication of SPE
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
264 John J. McCarthy

are given a historical perspective in Anderson (1985: 328–50). Prince and

Smolensky (1993/2004: 238–57) contrast OT with phonological systems
that attempt to combine rules and constraints. Harmonic Grammar,
another important predecessor of OT, has not been touched on here; see
Prince and Smolensky (1993/2004: 234 – 8) for brief discussion and Leg-
endre, Sorace, and Smolensky (2006) for a much more detailed treatment.

3. The Fundamentals of OT
How does OT respond to the issues raised in the previous section? That
is, how does OT address the following questions?
(i) How are constraints on the output of the grammar satisfied? What is the
relationship between constraints on output structures and the operations
that transform inputs into outputs? How are triggering and blocking
effects accounted for?
(ii) What is the relationship between the universal and the language particular?
How can constraints differ in their activity from language to language?
The answers to these questions follow directly from the fundamental
properties of OT.
OT sets up a basic dichotomy between the operational component of
the grammar and the constraint component. The operational component,
called Gen, constructs a set of candidate output forms that deviate from
the input in various ways. The constraint component, called Eval, filters
the candidate set, selecting a member of it to be the actual output of the
grammar.
Gen functions something like FM’s optional transformations or GB’s
Move α. Gen applies all linguistic operations freely, optionally, and some-
times repeatedly. This property of Gen, which is known as freedom of
analysis, is assumed for two reasons. First, it is simpler to define Gen
with freedom of analysis than without it. For example, the phonological
Gen can repeatedly epenthesize – that is, Gen(/pa/) includes {pa, paE,
paEE, paEEE, . . . } – because complicating Gen with limits on epenthesis
is unnecessary. Through minimal violation (see below) Eval puts limits
on epenthesis anyway. Second, Gen’s freedom of analysis is necessary
because of the related assumption that Gen is universal. Because Gen is
the same in every language, it must in effect anticipate all of the ways that
any language could transform a given input, so as to be certain that all of
these options are represented in the candidate set. The simplest way to do
this is to supply Gen with certain operations and allow them to apply
freely, thereby over- rather than undergenerating the necessary range of
candidates. Again, overgeneration by Gen does not mean overgeneration
by the grammar, because the output of Gen is filtered by Eval.
The grammar of a language is a constraint hierarchy. The constraint
ranking relation is referred to as domination, written >>. For example,
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 265

the hierarchy [C1 >> C2], read as ‘C1 dominates C2’, is a statement about
the relative priority of constraints C1 and C2 in a particular language.
Eval receives the candidate set from Gen, evaluates it using the lan-
guage’s constraint hierarchy, and selects its most harmonic or optimal
member as the output of the grammar. Eval starts with the constraint that
is ranked highest, C1, and it chooses only those candidates that are most
favored by C1. The candidates that are most favored by C1 (or any other
constraint) have two things in common: they receive the same number of
violation marks from C1, and no other candidate receives fewer violation
marks from C1. There is always at least one candidate that is most favored
by C1, and it is possible for all of the candidates to be most favored by
C1 (if all candidates violate C1 equally).
We now have a subset of the original candidate set consisting of just
those candidates that are most favored by C1. This subset is passed along
to the next constraint in the ranking, C2, and it does the same thing: it
locates the subset of candidates that it most favors. This process continues
until the set has been reduced to just one candidate. It is optimal. It
satisfies the constraints as they are ranked in the language-particular hierarchy
better than any other candidate in the original candidate set.
This brief explanation of Eval illustrates several important points about OT:
(i) Although constraints are violable, violation is always minimal. The
optimal candidate may violate C1, but no other candidate can violate C1
less than the optimal candidate does.
(ii) The preferences expressed by a lower-ranking constraint are important
only insofar as they decide between candidates that tie on all higher-
ranking constraints. In other words, OT constraints are arranged in
strict-domination hierarchies, in which superior performance on
lower-ranking constraints can never overcome inferior performance on
higher-ranking constraints.
(iii) Eval never asks for candidates that obey a constraint; it only asks for
candidates that are favored by a constraint. This is an important distinction,
because it means that there is always some optimal candidate (unless,
absurdly, the initial candidate set is empty). Being favored by a constraint
is not the same as obeying it. One or more candidates are always favored,
but it will sometimes happen that no candidate obeys a given constraint.2
Each language has its own constraint ranking. The strongest hypothesis
is that constraint ranking is the only thing in the grammar that is language
particular: Gen, Eval, and even the constraints themselves are universal.
The universal constraint component is called Con. If the ranking of
universal Con is the only difference between grammars, then all of the
constraints in Con are present in the grammars of all languages. Only the
ranking is different.
Con itself consists of two types of constraints. Markedness constraints
are similar to the surface-structure constraints or filters of the 1970s. The
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
266 John J. McCarthy

inventory of markedness constraints in Con is a substantive theory of

linguistic well-formedness – for example, complex consonant clusters or
that-trace sequences are bad. A significant innovation in OT is the notion
of a faithfulness constraint. Faithfulness constraints are inherently con-
servative, requiring the output of the grammar to resemble its input.
Because markedness constraints favor some linguistic structures over others,
they are often in tension with faithfulness constraints, which resist changes
to input structures. This tension is called constraint conflict, and it is
resolved in OT by ranking. Conflict between two markedness constraints
or between two faithfulness constraints is also possible, of course.
These various aspects of OT are strongly interdependent. For example,
faithfulness constraints make very little sense unless constraints are violable,
because otherwise differences between input and output would be impossible. 3
Likewise, constraint violability and ranking are necessary for the claim that
the constraints are universal, because otherwise languages could not differ.
This brief summary of the most important properties of OT is sufficient
to support some preliminary answers to the questions raised at the outset
of this section.
The first question concerns the interaction between linguistic processes
and the output constraints that seem to trigger or block them. OT’s answer
is that there is no real interaction, as processes and constraints are in
separate grammatical components, Gen and Eval, and information flows
in only one direction, from Gen to Eval. Strictly speaking, processes are
neither triggered nor blocked; instead, the process component Gen
supplies a broad array of possible outputs that reflect the results of applying
various operations. In Yawelmani, for example, unsyllabifiable consonants
are resolved by epenthesis because candidates with epenthesis are among
those supplied by Gen, and because Eval favors less-marked [%i.lik.hin]
over more-faithful *[ %il.k.hin] and differently-unfaithful *[ %il.hin].
Eval is responsible for choosing the winning candidate, not for generating it.
The second question asks how output constraints can differ in activity
from language to language. OT attributes differences in constraint activity
to differences in ranking. The ranking of a markedness constraint relative
to faithfulness constraints and other markedness constraints determines
whether and when it is active. We can illustrate this with some of the
examples mentioned near the end of Section 2. Con includes a marked-
ness constraint, called Onset, that requires syllables to have onsets. If
Onset dominates the antiepenthesis faithfulness constraint (and certain
other ranking requirements are met), then candidates that resolve onsetless
syllables by epenthesis will be preferred to candidates that preserve them:
Arabic /al-walad/ → [%al-walad], *[al-walad]. If in addition Onset is
ranked below a faithfulness constraint specific to word-initial syllables,
then epenthesis will occur medially but not initially, as in Axininca
Campa: /i-n-koma-i/ → [inkomati] ‘he will paddle’. On the other hand,
if Onset is ranked below the antiepenthesis and antideletion faithfulness
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 267

constraints, then we get a language like Timugon Murut that permits

onsetless syllables initially and medially: /ambiluo/ → [am.bi.lu.o].
Even low rank is no guarantee that a constraint is completely inactive,
however. Recall that Timugon Murut respects Onset in two ways:
intervocalic consonants are syllabified as onsets ([am.bi.lu.o], not *[am.bil.u.o]);
and the reduplicative prefix must have an onset ([a.ba.ba.lan], not
*[a.a.ba.lan]). These effects of Onset occur in situations where Onset can
be active without the need for epenthesis or deletion. Situations like this,
where even a low-ranking markedness constraint can be decisive in situations
where faithfulness is not at issue, are known as emergence of the
unmarked (McCarthy and Prince 1994).
As these few examples make clear, constraint ranking gives much more
subtle control over constraint activity than parameters do. These examples
also remind us, as we saw in Section 2, that more subtle control is exactly
what is needed.
Suggestions for further reading. The fundamentals of OT are laid
out primarily in Chapter 5 of Prince and Smolensky (1993/2004). Other
overviews of this topic, in approximate order of increasing length and
depth of coverage, include Archangeli (1997), Tesar, Grimshaw, and
Prince (1999), Kager (1999: 1–51), Prince and Smolensky (1997; 2003),
McCarthy (2002: 3–47; 2003a), and Smolensky, Legendre, and Tesar
(2006). Legendre (2001) is an introduction to OT syntax.

4. Doing Linguistic Analysis in Optimality Theory

As constraint ranking is hypothesized to be the only way in which grammars
may differ, the core of any OT analysis is a collection of constraint
rankings and the justification for them. In general, a ranking is justified
by comparing two output candidates from the same input. One of these
candidates, called the winner, is the actually observed output for that
input in the language under analysis. The other candidate is a loser: it is
derived by Gen from that same input, but it is not the most harmonic
candidate according to Eval. For Eval to select the correct candidate as
winner, some constraint that favors the winner over the loser must dominate
every constraint that favors the loser over the winner. The logic of this
statement follows from the properties of Eval (see Section 3). Constraint
ranking arguments depend on this logic.
There are three essential elements to a valid ranking argument: First,
the constraints to be ranked must conflict; that is, they must disagree in
their assessment of a pair of competing output candidates derived from
the same input. For instance, Yawelmani / %ilk-hin/ has competing
candidate output forms [ %i.lik.hin] and *[%il.k.hin]. The markedness
constraint against unsyllabified consonants (call it *C, recalling the
italicization of unsyllabified consonants) and the antiepenthesis faithfulness
constraint (known as Dep) disagree in their assessment of these forms.
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
268 John J. McCarthy

Second, one member of this pair of competing candidates must be the

actual output form for the given input – the winner, in short. The
constraint that favors the winner, *C, must dominate the constraint that
favors the loser, Dep. Third, the ranking argument is secure only if there
is no third constraint that could also be responsible for this winner beating
this loser. In the Yawelmani example, such a constraint would have to
meet two conditions to be problematic: like *C, it would have to favor
[%i.lik.hin] over *[%il.k.hin]; and it could not be ranked below *C. Sup-
pose for the sake of argument that there were a constraint *lk that is
violated by any [lk] consonant sequence. Because *lk prefers [%i.lik.hin]
over *[%il.k.hin], just as *C does, *lk threatens to undermine the ranking
argument we have constructed, leaving us with a less definitive disjunction:
at least one of *C and *lk dominates Dep. This disjunction can be
resolved by independently showing that Dep dominates *lk, using a
ranking argument based on the winner/loser pair [ %il.kal]/*[%i.li.kal]
‘might sing’.
The elements of a ranking argument are illustrated with a tableau. The
original violation tableau format of Prince and Smolensky (1993/2004)
is illustrated in (1). Each cell (row, column) indicates with asterisks the
number of violations, if any, of constraint column incurred by candidate
row. When a constraint knocks a candidate out of competition, that result
is signaled by an exclamation point. Cells are shaded when they can have
no possible effect on the outcome because higher-ranking constraints are
decisive.
(1) Violation tableau

/%ilk-hin/ *C DEP

a.→ %i.lik.hin *
b. %il.k.hin *!

The comparative tableau, introduced by Prince (2002), adds annotations

to each loser row indicating whether the constraint favors that loser or
the winner. In the comparative tableau (2), *C favors the winner over the
loser [%il.k.hin], so a W appears in the *C cell of the loser row. Dep favors
the loser, so an L is placed in the Dep cell. In a properly ranked compar-
ative tableau, every loser row contains at least one W with no Ls to its
left (readers may find it helpful to pause at this point and convince
themselves that the previous statement follows from the nature of Eval).
(2) Comparative tableau
/%ilk-hin/ *C DEP
a. → %i.lik.hin *
b. %il.k.hin *W L

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 269

The comparative tableau is best for dealing with the ranking problem,
as Brasoveanu and Prince (2005) call it. In the ranking problem, the
winner is already known and we are trying to figure out which ranking
will produce that winner. The ranking problem arises whenever we are
trying to analyze some data: the winners are known because they are the
data of the language, and it is the ranking that we seek to discover. The
violation tableau format is appropriate for what Brasoveanu and Prince
call the selection problem. In the selection problem, the ranking is
already known but the winner is not. This is a typical situation in the
study of language typology, where we want to check each ranking to
determine which candidate wins.
Every theory presents its own special challenges to the analyst, and OT
is no exception. One rather difficult task is determining the range of
candidates that must be considered. Overlooking a losing candidate that
ties or beats the intended winner ‘invites theoretical disaster, public
embarrassment, and unintended enrichment of other people’s careers’
(McCarthy and Prince 1993: 13), as the overlooked candidate has the
potential to undermine the entire analytic edifice. Systematic exploration
of some space of candidates is usually the best way to avoid this problem. 4
For example, once we have established that Dep must be outranked by
other constraints in Yawelmani, it makes sense to consider all of the ways
in which Gen can alter /%ilk-hin/ by epenthesis. We can quickly determine
that candidates with absurdly repeated epenthesis like *[%i.lik.hi.ni.i.i.i]
are no threat, as they violate Dep more than the winner does and do not
beat it on any known constraint. But a candidate like *[%i.li.ki.hi.ni] has
a more serious claim on our attention. It discloses that Dep must dominate
the markedness constraint No-Coda, which prohibits syllable-final
consonants. Even more important is the candidate *[%il.ki.hin], because it ties
with the intended winner [%i.lik.hin] on both *C and Dep. The analysis
would not be complete without some constraint that decided between
these two candidates.
Another special responsibility of the analyst in OT is the introduction
of new constraints. In theory, Con is universal, so the analyst’s job is just
a matter of finding some ranking of Con that reproduces the data of the
language in question. In practice, however, the details of the constraints
in Con are the topic of ongoing research and discussion. Because Con
embodies all of the substantive properties of human language, our present
inability to deliver a definitive theory of Con should come as no surprise
to anyone willing to concede that there are things about language that are
not yet fully undersood.
What this means for the analyst is that any actual analysis may very well
involve some constraints whose status in universal grammar (i.e., Con) is
not yet settled. For example, while there is probably a consensus among
phonologists that Con includes a faithfulness constraint Max that militates
against deleting elements of the input, there is disagreement about whether
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
270 John J. McCarthy

the antiepenthesis constraint Dep is also needed, as many of its effects can
be obtained from independently necessary markedness constraints [see
Gouskova (2007) and Urbanczyk (2006) for opposing views on this question].
Another example: different ways of rationalizing the ad hoc markedness
constraint *C depend on different theories of the relationship between
syllabification and phonotactics. OT itself says nothing about syllabification
or phonotactics, although it certainly provides a framework in which a theory
of syllabification and phonotactics can be worked out using violable constraints.
OT provides the analyst with ample opportunities to test and refine any
hypothesis about Con. Any newly proposed constraint must, of course,
offer insight into the analysis at hand. But this is not enough; the proposer
is also responsible for exploring the (un)intended consequences of the
new constraint under ranking permutation. The source of this novel
obligation on the analyst is OT’s inherently typological character: every
language-particular hierarchy of n constraints is pregnant with n!–1 other
hierarchies, each of which is predicted to be the grammar of a possible
human language.5 Fortunately, not all constraints conflict, so many different
rankings will produce identical results, making the task of the analyst-cum-
theorizer far less daunting than it might at first seem.
As an example of this sort of reasoning, we can return to a question
raised just above: Is the faithfulness constraint Dep unnecessary because
independently motivated markedness constraints do the same work? Take,
for instance, a language like Timugon Murut that allows onsetless syllables.
In this language, Onset must be crucially dominated to rule out consonant
epenthesis. Either Onset is dominated by Dep or, in the Dep-less theory,
Onset is dominated by a set of markedness constraints that is sufficient
to rule out every imaginable consonant that could be inserted to satisfy
Onset. For convenience, we can refer to this set of constraints as *Cons.
The Dep-less theory requires the existence of *Cons, and therefore it
makes certain predictions about the effects of ranking permutation. For
example, by ranking *Cons above Max, we could get a language that has
no consonants whatsoever, only vowels (Gouskova 2003: 71ff ). No such
language is known, and nearly everyone would agree that we do not want
a theory to predict the existence of such a language. In this case as in so
many others, tests against language typology act as a check and corrective
on proposals about Con (see Section 6.1 for further discussion.)
Suggestions for further reading. The special challenges of analysis
in OT are briefly addressed in McCarthy (2002: 30 –42). McCarthy
(forthcoming) is a book-length treatment of this topic.

5. Illustrations
This section puts some flesh on the theoretical and analytic skeleton
assembled in the previous sections. It does this by illustrating OT at work
with two examples, one phonological and the other syntactic.
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 271

In the Nigerian Niger-Congo language Emai (Casali 1996: 62–8;

Schaefer 1987), Onset is satisfied at V1+V2 juncture by changing V1 into
a glide, deleting V1, or deleting V2. The choice among these three options
depends on whether V1 is a high vowel and on whether the morphemes
that contain V1 and V2 are lexical or functional:6
(3) Synopsis of Emai alternations
At V1+V2 juncture:
a. If V1 and V2 are both contained in lexical morphemes and V1 is a high
vowel [i] or [u], then V1 changes into the homorganic glide [ j] or [w]:
/ku amε/ [kwamε] ‘throwLex waterLex’
/fi cpia/ [fjcpia] ‘throwLex cutlassLex’
Otherwise:
b. If one of the morphemes is lexical and the other is functional,
delete the vowel in the functional morpheme:
/cli ebe/ [clebe] ‘theFunc bookLex’
/ukpode cna/ [ukpodena] ‘roadLex thisFunc’
c. If both morphemes are lexical or both are functional, delete V1:
/kc ema/ [kema] ‘plantLex yamLex’
/fa edi/ [fedi] ‘pluckLex palm-nutLex’
/oa isi ci/ [oasci] ‘houseLex ofFunc hisFunc’
The faithful candidates derived from these inputs violate Onset: *[ku.a.mε],
etc. The main analytic challenge is determining how Onset is satisfied,
and that is primarily a matter of determining the correct ranking of certain
faithfulness constraints. One of them might be called Ident(syllabic); it is
violated when a vowel is replaced by a glide. The others are versions of
the antideletion constraint Max. In Emai, as in various other languages,
the roots of major lexical category items (nouns, verbs, adjectives) are
treated more faithfully than function words or affixes. This evidence shows
that Con must contain a constraint MaxLex that is violated only by deletion
of segments from noun, verb, or adjective roots (McCarthy and Prince
1995). MaxLex is distinct and separately rankable from the general constraint
Max, which is violated by deletion of any segment, regardless of the lexical
status of the morpheme that contains it. Similarly, many languages show
greater faithfulness to segments in initial syllables, motivating the posi-
tional faithfulness constraint MaxInit (Beckman 1997, 1999; Casali 1996).
It too is distinct and separately rankable from other Max constraints.
With these typologically justified constraints in hand, we are now in a
position to establish the ranking responsible for the alternations in (3). We
will begin with the process of glide formation. Changing a vowel to a
glide is a violation of the faithfulness constraint Ident(syllabic). Whenever
an output of the grammar violates a faithfulness constraint, some markedness
constraint must dominate that faithfulness constraint. In this case, the
high-ranking markedness constraint is Onset:
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
272 John J. McCarthy

(4) Onset >> Ident(syllabic)

/kuLex amεLex/ ONSET IDENT

a.→ kwa.mε *
c. ku.a.mε *W L

Deleting a vowel is an obvious alternative to changing it into a glide,

and that possibility must be excluded with another constraint, also ranked
above Ident(syllabic). For reasons that will become clear later, that role
belongs to MaxLex and not to either of the related constraints MaxInit and
plain Max:
(5) MaxLex, Onset >> Ident(syllabic)
/kuLex amεLex/ ONSET MAXLex IDENT

a.→ kwa.mε *

b. ka.mε *W L
c. ku.a.mε *W L

Given just the ranking in (5), we might expect to see other situations
where glide formation is preferred to deletion, such as the following:
(6) e]Lex [Lexa → ja
a]Lex [Lexi → aj
Neither of these alternations occurs, so some constraints must rule
them out. The mapping /e a/ → [ ja] violates an undominated faithfulness
constraint, Ident(high). The output [aj] violates an undominated markedness
constraint against diphthongs with falling sonority, such as [aj] and [aw].
Although a high vowel at the end of a lexical morpheme can become
a glide because of the ranking in (5), a high vowel at the end of a
functional morpheme deletes instead: /cli Func ebe Lex/ → [c.le.be],
*[c.lje.be]. Deleting the /i/ of /cli/ is a violation of general Max, but not
of MaxLex. Therefore, as (7) shows, Ident(syllabic) dominates Max. With
this ranking, deletion is preferable to a vowel→glide mapping, as long as
the contents of lexical morphemes are not threatened (on the winner’s
remaining Onset violation, see Note 6).
(7) Ident(syllabic) >> Max

/cliFunc ebeLex/ ONSET MAXLex IDENT MAX

a. → c.le.be * *
b. c.lje.be * *W L
c. c.li.e.be **W L

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 273

From the ranking arguments in (4), (5), and (7), we can assemble the
constraint hierarchy [MaxLex, Onset >> Ident(syllabic) >> Max]. Accord-
ing to this hierarchy, Onset is satisfied by deleting the vowel of a func-
tional morpheme, or else changing the vowel of a lexical morpheme into
a glide. In this hierarchy, MaxLex and Onset are not yet ranked with
respect to one another. Ranking requires conflict, and the conditions that
will place MaxLex and Onset into conflict are quite precise: both vowels
must be in lexical morphemes, and glide formation must be ruled out for
the reasons given just below (6). An example that meets both of these
criteria is /kcLex emaLex/ → [ke.ma]. It shows that the vowels of lexical
morphemes can delete, so Onset must dominate MaxLex:
(8) Onset >> MaxLex

/kcLex emaLex/ ONSET MAXLex

a. → ke.ma *
b. kc.e.ma *W L

The /kcLex emaLex/ → [ke.ma] mapping in (8) and other examples raise
another question: when glide formation is not an option, which vowel is
deleted? The facts are as follows. If one vowel is hosted by a lexical
morpheme and the other is hosted by a functional morpheme, then
the vowel of the lexical morpheme is preserved because of MaxLex:
/ukpodeLex cnaFunc/ → [ukpodena]; /cliFunc ebeLex/ → [c.le.be]. If both
vowels are hosted by lexical morphemes or both are hosted by functional
morphemes, then Max Init favors preservation of the vowel that is
morpheme initial: /kcLex emaLex/ → [ke.ma].
Interestingly, although MaxInit is crucial for the analysis of /kcLex
emaLex/ → [ke.ma], this example does not tell us how MaxInit is ranked.
As tableau (9) shows, MaxInit has the role of a tie-breaker: no other
constraint favors one of the candidates. This means that MaxInit does not
conflict with any of the other constraints over this pair of candidates
[Observe that there is no L annotation in row (b)]. As conflict is the basis
of ranking arguments, tableau (9) offers no evidence about how MaxInit
is ranked.
(9) MaxInit as tie-breaker
/kcLex emaLex/ ONSET MAXLex IDENT MAX MAXInit

a. → ke.ma * *
b. kc.ma * * *W

MaxInit is rankable on the basis of other data, however. To get a conflict

and therefore a ranking argument, we need an example where the choice
is between deleting a Lex vowel that is noninitial and deleting a Func
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
274 John J. McCarthy

vowel that is initial. Examples like /ukpodeLex cnaFunc/ → [ukpodena] are

just what is required. As the Func-initial vowel is deleted and the Lex-
final vowel is not, MaxLex must dominate MaxInit:
(10) MaxLex >> MaxInit
/ukpodeLex cnaFunc/ MAXLex MAXInit

a. → u.kpo.de.na *
b. u.kpo.dc.na *W L

The diagram in (11) illustrates the rankings that have been established. In such
a diagram, if there is a strictly downward path from A to B, then A dominates
B by direct argument or by transitivity of the domination relation.
(11)

In the parlance of 1970s phonological theory, Emai would be said to have

a conspiracy (a term originally due to Haj Ross). In a conspiracy, several
distinct rules act in concert to achieve the same output result (Kisseberth
1970). Emai would require at least three rules – glide formation, deletion
of the first vowel, and deletion of the second vowel – all in service of
eliminating onsetless syllables. A conspiracy is really just a description of
something that rule-based phonology is unable to explain. In OT, on the
other hand, there are no rules and hence no conspiracy. The output
constraint Onset favors candidates without onsetless syllables, and it dom-
inates all four faithfulness constraints in (11). Ranking relations among the
various faithfulness constraints determine which unfaithful candidates are
the winners. The output constraint Onset is obviously central to the
analysis and not just a post hoc rationalization of why this particular
language should have these three particular processes.
Grimshaw’s (1997) account of do-support in English illustrates a very
different kind of analysis in OT. The goal is to explain the economy of
unstressed do: why does it appear only when it is needed? Unstressed do
is required in inversion and negation constructions but forbidden in simple
declaratives:
(12) a. Did Robin leave?
b. When did Robin leave?
c. Robin didn’t leave.
d. *Robin did leave.
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 275

Do’s only-when-needed distribution means that there must be a markedness

constraint that militates against it, to force its absence from surface structure
except when some higher ranking constraint compels its presence. For
Grimshaw, the constraint with anti-do effect is Full Interpretation
(Full-Int), which is violated by any lexical item in surface structure that
does not contribute to the interpretation of that structure.
Full Interpretation is sometimes in conflict with the constraint
Obligatory Heads (Ob-Hd), which is violated by any projection that
lacks a head. When a projection would be headless, the choice is between
supplying a head that violates Full Interpretation or tolerating the
breach of Obligatory Heads. In English, do-support is the result of a
constraint hierarchy in which Obligatory Heads ranks higher:
(13) Obligatory Heads >> Full Interpretation

OB-HD FULL-INT
a. → [CP When didi [IP Robin ei [VP leave t]]] *
b. [CP When e [IP Robin e [VP leave t]]] **W L
c. [CP When e [IP Robin did [VP leave t]]] *W *

Although it is ranked below Obligatory Heads and does not affect the
outcome in (13), Full Interpretation is not inactive. There are situations,
such as simple declaratives, where every projection has a head without further
ado, so to speak. In that case, Full Interpretation emerges to ban do
from the output:
(14) Economy of do

OB-HD FULL-INT

a. → [VP Robin left]

b. [IP Robin did [VP leave ] *W

For the same reason, do is impossible in combination with another

auxiliary: *When will Robin do leave?
Exactly parallel situations are found in phonology (Gouskova 2003): in
Lillooet, for instance, the vowel [e] appears only when it is needed for
markedness reasons, in words that would otherwise be vowelless and in
clusters that would otherwise violate sonority-sequencing requirements.
Grimshaw and Gouskova show, for syntax and phonology respectively,
how such economy effects follow from intrinsic properties of OT rather
than stipulated economy principles. With English do-support, and likewise
with Lillooet [e], the source of the economy effect is minimal violation
of constraints like Full-Int. Minimal constraint violation is a sine qua non
of OT; it is inherent to OT’s core component, Eval (see also Section 6.2).
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
276 John J. McCarthy

6. Some Questions and Answers about OT

Many of the most commonly asked questions about OT touch on fun-

damental aspects of the theory. These questions therefore serve as an
excellent framework for further exploration of OT’s concepts and
consequences [for other lists of frequently asked questions about OT, see
McCarthy (2002: 239–45), Smolensky, Legendre, and Tesar (2006: 523–
32), and Prince and Smolensky (1993/2004: 232)].

6.1 language universals

Because all constraints are violable, how does OT account for universals
of human language? In general, what sorts of predictions does OT make
about possible and impossible languages?
In answering this question, it is helpful to begin with a distinction due
to Prince (1997a). Some predictions follow from just OT’s basic properties,
while other predictions follow from OT’s basic properties under some
specific assumptions about the contents of Con. Universals of the first
type can be called ‘formal’, and those of the latter type can be referred to
as ‘substantive’.
The simplest example of a formal universal in OT is harmonic
improvement. As Con contains only markedness and faithfulness con-
straints, a necessary condition for an unfaithful candidate to win is that it
be less marked than the faithful candidate – in other words, an unfaithful
winner must improve harmonically over a faithful loser. This property follows
from the basic structure of OT. If the input is /bi/, then no faithfulness
constraint disfavors faithful [bi] as the output. For unfaithful [be] to win,
then, some higher-ranking markedness constraint M must favor [be] over
[bi]. OT’s basic assumptions entail that unfaithfulness is possible only
when it improves markedness, where improvement is measured relative to
the universal markedness constraints as they are ranked in the language in
question.
Moreton (2003) develops a formal proof of this result and explores its
empirical consequences. One is that no OT grammar can analyze a language
with a system of circular mappings like /bi/ → [be] and /be/ → [bi]. The
reason is that no single constraint ranking can simultaneously assert that
[be] is less marked than [bi] (to get the /bi/ → [be] mapping) and [bi] is less
marked than [be] (to get the /be/ → [bi] mapping).7 Another consequence
is that no OT grammar can analyze a system of unbounded growth, like
/bi/ → [bi%i], /bi%i/ → [bi%i%i], . . . The reason is that a point will always
be reached where further addition of syllables is not harmonically
improving.
Substantive universals are mainly concerned with language typology.
Everything in OT is invariant across languages – Gen, Eval, Con, and
even the inputs (see Section 6.3) – except for the constraint ranking.
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 277

Therefore, the various ways of ranking Con define the permitted range
of between-language variation. This is factorial typology: the claim that
every permutation of Con is a possible grammar of a human language,
and every actual human language has a grammar that is a permutation of
Con. Claims about typology are substantive universals because they
depend in an obvious way on what constraints Con does and does not
contain.
Consider a simple substantive universal like the fact that no language
satisfies Onset by epenthesizing a consonant cluster, mapping, for example,
/apati/ to [trapati]. If [trapati] is in /apati/’s candidate set, as is standardly
assumed (McCarthy and Prince 1995, 1999), then the explanation for this
universal requires that [trapati] not be the optimal member of /apati/’s
candidate set under any permutation of Con. This mapping is correctly
predicted to be universally impossible if, say, Con does not include a
markedness constraint that favors initial consonant clusters over single
initial consonants. This same claim about Con makes other typological
predictions as well. For instance, there can be no language that requires
all of its syllables to begin with a cluster, nor can there be a language that
limits reduplication to cluster-initial words.
Because substantive universals depend on specific assumptions about
Con, disagreement among analysts is expected. In fact, typology is an
essential check on all hypothesized changes to the make-up of Con. No
newly proposed constraint is secure until its consequences under ranking
permutation have been checked for typological plausibility (see Section 4).
Suggestions for further reading. For further discussion of formal
universals in OT, see Prince (1997a) and McCarthy (2002: 109–11).
Almost any of the suggested readings at the end of Section 3 is a good
place to learn more about factorial typology and substantive universals.
The use of factorial typology to draw or confirm inferences about Con
is (or at least should be) ubiquitous in the OT literature.

6.2 constraints versus parameters

How is constraint ranking different from constraint parameterization? Isn’t
a low-ranking constraint effectively turned off, just like a parameter?
This issue was already touched on at the end of Section 2, but it is
important enough to bear repeating. Crucial domination of a constraint
is no guarantee of inactivity. There are some very special circumstances
where inactivity can be guaranteed (Prince 1997b: 3; Prince and Smolensky
1993/2004: 97–9, 130–2, 264–5), but otherwise a low-ranking constraint
is always potentially active. A low-ranking constraint will be active in fact
when it decides among candidates that tie on all higher-ranking constraints
and perform better than all other candidates on those constraints. This is
decidedly unlike the behavior of parameterized constraints: when such a
constraint is turned off, it is never active under any circumstances.
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
278 John J. McCarthy

Constraints that are low-ranking but active are particularly important

in OT’s explanation for economy effects. Faithfulness constraints ensure a
kind of economy of derivation. Even when a faithfulness constraint is
crucially dominated, such as Dep in Yawelmani, violation is always mini-
mal, and so gratuitous epenthesis is ruled out: /%ilk-hin/ → [%i.lik.hin],
*[%i.li.ki.hi.ni]. Certain markedness constraints effect a kind of economy
of representation in both syntax and phonology (see Section 5).
There are other, more subtle differences between ranking and para-
metrization. To cite a phonological example that also has syntactic ana-
logs, consider structures of the form [...(...)foot...]word. There are constraints
that require every foot to be aligned at the left or right edge of the word,
Align-L(foot, word) and Align-R(foot, word). Align-L(foot, word)
favors [(|pata)kama] over [pa(|taka)ma] and [pata(|kama)]. Align-
R(foot, word) favors [pata(|kama)] over [(|pata)kama] and [pa(|taka)ma]. It
might seem, then, that [(|pata)kama] or [pata(|kama)] are the only possible
winners, depending on whether Align-L(foot, word) or Align-R(foot,
word) is ranked higher. But this quasi-parametric conception of ranking
overlooks candidates like [(|pata)], which manage to satisfy both of the
alignment constraints at the expense of eliminating some input material.
For a real-life example, reduplication in Diyari, see McCarthy and Prince
(1994).
Cases like this are particularly relevant to the comparison of OT with
parametric theories, as Samek-Lodovici (1996: 216) argues. In a parametric
theory of stress like Hayes (1995), the direction of foot assignment is a
parameter with two values, left-to-right and right-to-left. Nothing further
follows from this. In OT, however, the constraints Align-L(foot, word)
and Align-R(foot, word) together entail a further possibility: satisfaction
of both constraints in words consisting of exactly one foot. Here, as in many
other cases, OT’s nonparametric approach to between-language variation
has economy of structure as an unavoidable – and welcome – side effect.
Suggestions for further reading. The difference between ranking
and parameterization is a particular focus of Grimshaw (2002), McCarthy
(2002: 11–2, 109–10, 127–08 et passim), McCarthy and Prince (1994),
Prince and Smolensky (1993/2004: 164 –7), and Samek-Lodovici (1996),
among others.

6.3 the lexicon

What is the role of the lexicon in OT?
To a great extent in syntax, and to a lesser extent in phonology, the
lexicon has been seen as the proper repository of all that is language
particular. OT turns that assumption on its head.
One of OT’s basic premises is called richness of the base (Prince and
Smolensky 1993/2004: 205, 225). The ‘base’ is the set of inputs to the
grammar [an allusion to the base component of Chomsky (1957)]. It is
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 279

‘rich’ because, by hypothesis, it is not subject to any language-particular

restrictions. In phonology, richness of the base rules out the use of morpheme
structure constraints, stipulated underspecification, or similar devices that
preemptively remove forms from the set of possible inputs to the grammar.
In syntax, richness of the base means that systematic differences between
languages cannot be attributed to systematic differences in the contents of
their lexicons. Because of richness of the base, all aspects of well-formedness
are under the control of Eval and the constraint hierarchy, and all
systematic differences between languages must be obtained from differences
in constraint ranking. Richness of the base is also central to OT’s resolution of
the so-called duplication problem in phonology (McCarthy, 2002: 71–6).
Richness of the base has been a source of some confusion in criticisms
of OT. It does not mean, absurdly, that all languages have the same
vocabulary, nor does it mean, equally absurdly, that learners posit something
like /^Gkæt/ the underlying form for [kæt] cat. Rather, it simply means
that the lexicon as a system is not subject to any language-particular require-
ments. Furthermore, it means that explanations for linguistic phenomena
cannot involve carefully contrived limits on the inputs to the grammar.
An example from phonology: between-language differences in the behav-
ior of laryngeal features must derive from differences in ranking rather
than differences in lexical specification of [voice] and [spread glottis] (Beck-
man and Ringen 2004). An example from syntax: between-language dif-
ferences in fronting of wh words must derive from differences in ranking
rather than differences in whether wh words are lexically specified with a
strong or weak feature (Smolensky, Legendre and Tesar 2006: 529).
Suggestions for further reading. As richness of the base is ubiquitous
in the OT phonology literature, I focus here on syntax. Richness of the
base in syntax receives special attention from Aissen (1999), Bakovic and
Keer (2001), Bresnan (2001a,b), Grimshaw (1997, 2001), Grimshaw and
Samek-Lodovici (1995, 1998), Legendre, Smolensky, and Wilson (1998),
and Samek-Lodovici (1996: 219).

6.4 ungrammaticality
In a theory where all constraints are violable, how can any linguistic
representation be absolutely ill formed? For example, how is it possible
for English phonology to rule out *[bnık] as phonotactically impossible, and
how is it possible for English syntax to rule out *Who did he say that left?
as ungrammatical?
Although other theories use inviolable constraints and crashing derivations
to account for ungrammaticality, these analytic techniques are not indis-
pensable. In OT, ungrammaticality is a consequence of inferiority to
other candidates rather than an apparatus of inviolable constraints. The
linguistic representation *[A] is ungrammatical in some language if and
only if [A] is not the optimal candidate for any input, given that language’s
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
280 John J. McCarthy

constraint hierarchy. That is, for any input /X/ that has [A] in its candidate
set, there is some output candidate that is more harmonic than [A]
according to the grammar of the language in question.
An input that merits particular attention is /A/ itself, such as /bnık/. As
every faithfulness constraint favors the mapping /bnık/ → [bnık], some
higher-ranking markedness constraint is necessary to rule it out in Eng-
lish. This constraint is presumably a prohibition on onset clusters of two
stops, and if it dominates Dep, then [benık] will be more harmonic than
*[bnık] ([benık] is not a real word of English, but it is phonotactically
possible, and that is the point of the example). This is not quite enough
to secure the absolute ill-formedness of *[bnık], however; that requires
showing that no input will map most harmonically to *[bnık]. That
demonstration ultimately requires using methods of proof rather than
exhaustive search, as there are infinitely many possible inputs.
Suggestions for further reading. Absolute ill-formedness is the topic
of an anthology, Rice (forthcoming). Earlier work that focuses specifically
on this topic includes Bakovic and Keer (2001), Legendre, Smolensky, and
Wilson (1998: 257), McCarthy (2002: 76, 176, 194 –9, 239), Orgun and
Sprouse (1999), Pesetsky (1997), Prince and Smolensky (1993/2004:
57ff ), Raffelsiefen (2004), Rice (2003, 2005), and Smolensky, Legendre,
and Tesar (2006: 528–30).

6.5 infinity
If the number of candidates is infinite or even very large, then how do
speakers find the optimal candidate in a reasonable length of time? For
instance, the phonological Gen described in Section 3 can iterate epenthesis
forever, so the candidate set is unbounded in size.
This question is based on two tacit assumptions. One of them is that
some other theory of language is more plausible psychologically than OT.
Presumably, we are to imagine a speaker mentally applying a sequence of
rules before producing an utterance and to compare this with the image
of a speaker hopelessly trying to evaluate an infinite set of candidate utterances.
The problem is that the image of the speaker applying a sequence of rules
has just as little claim to psychological plausibility. The failure of the
Derivational Theory of Complexity shows this, at least in syntax (Fodor,
Bever and Garrett 1974); the same seems to be true in phonology as well
(Goldsmith 1993).
The other assumption underlying this question is an implicit denial of
the competence/performance distinction. In other words, the question
presupposes that a generative grammar is a model of a speaker’s actual
processing behavior as well as a speaker’s linguistic knowledge. Yet this is
not the stated goal of generative grammar. For example, Chomsky (1968:
117) writes that ‘. . . although we may describe the grammar G as a system
of processes and rules that apply in a certain order to relate sound and
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 281

meaning, we are not entitled to take this as a description of the successive acts
of a performance model . . . – in fact, it would be quite absurd to do so.
. . . If these simple distinctions are overlooked, great confusion must result.’
Issues of efficient generation and parsing are of course important, but
they do not bear directly on the theory of competence. They are properly
addressed using the ideas and methods of fields like computational linguis-
tics. See the suggested readings for some examples.
Suggestions for further reading. There has been a great deal of research
and progress on computational modeling of OT and determining its
computational complexity. Tesar (1995) is probably the earliest work on
these topics, and there is much more (Ellison 1994; Eisner 1997; Ham-
mond 1997; Frank and Satta 1998; Gibson and Broihier 1998; Karttunen
1998; Wareham 1998; Fanselow et al. 1999; Gerdemann and van Noord
2000; Riggle 2004; Becker 2006; and others).

6.6 learning
The number of permutations of n constraints is n!. How then is it possible
for learners to find the right ranking for their language out of the gigantic
space of possible rankings?
The ranking is actually the easiest part of language learning in OT.
Given the right kind of data, the constraint demotion learning algorithm
rapidly and efficiently finds a ranking that will correctly reproduce the
data (Tesar and Smolensky 1998, 2000). The basic idea is this: every
constraint that favors a loser over the intended winner must be dominated
by some constraint that favors the winner over that loser (this is easy to
visualize in a comparative tableau: every L must have some W to its left).
This requirement follows from the nature of Eval, and it is the basis of
ranking arguments like those in Section 4. Learning proceeds by demoting
every loser-favoring constraint below some winner-favoring constraint,
until all loser-favoring constraints are crucially dominated. The losers that
drive this learning algorithm can be obtained from the grammar itself:
they are outputs produced by the incorrectly ranked grammar before
learning is complete.
The data needed for the constraint demotion learning algorithm is
more than just the phonetic signal that the learner hears; it must also
contain information about hidden structure. In both syntax and phonology,
learners may have multiple options for parsing expressions of the ambient
language, and getting the parse right is crucial to getting the ranking
right. In phonology, the ranking also depends on the accurate recovery
of underlying representations from paradigmatic alternations. There has
been significant progress on the hidden structure problem in OT; the
main idea is that learners use their nascent grammar to produce hypotheses
about hidden structure, and further learning reveals whether these hypotheses
are correct (Tesar 1998, 1999; Tesar and Prince 2004).
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
282 John J. McCarthy

A particularly welcome consequence of OT is its success in making

connections among phonological theory, learnability, and empirical
research on language acquisition. With the exception of Natural Phonology
(Stampe 1973; Donegan and Stampe 1979), pre-OT generative phonology
was confounded by the facts of language acquisition: children’s reduced
pronunciations required that child phonology have many rules for which
there is no evidence in the adult language. In OT, children’s reduction
processes are a result of satisfying high-ranking universal markedness
constraints. The very same markedness constraints that, through ranking,
characterize differences between languages are also responsible for differences
between children and adults within a single language.
Suggestions for further reading. There are two anthologies that are
relevant to this topic: Kager, Pater, and Zonneveld (2004) and Dinnsen
and Gierut (forthcoming). The former includes work on learnability
as well as acquisition. The latter deals with disordered as well as
normal acquisition of phonology. McCarthy (2002: 232) has a nearly
exhaustive list, compiled by Joe Pater, of the OT literature on acquisition
before 2002.

6.7 derivations
In OT, inputs are mapped to outputs without any intermediate steps. Aren’t there
linguistic phenomena, in both phonology and syntax, that require deriva-
tions with intermediate steps (Chomsky 1995: 223 –5, 380 and many others)?
It is quite correct that the mapping from the input to the output of an
OT grammar does not involve any intermediate steps. For example, the
mapping from underlying /ktub/ to surface [%uktub] ‘write!’ in Arabic involves
a two-step derivation in rule-based phonology (15), with vowel epenthesis
creating the context that necessitates [%] epenthesis. In OT, on the other hand,
the grammar compares candidates that may show the simultaneous effects of
two or more epenthesis operations, and [%uktub] is among them (16).
(15) Arabic /ktub/ → [%uktub] with rules
Underlying /ktub/
Vowel epenthesis uktub
[%] epenthesis %uktub
Surface [%uktub]
(16) Arabic /ktub/ → [%uktub] in OT

/ktub/ ONSET *#CC DEP

a.→ %uktub **
b. ktub *W L
c. uktub *W *L

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 283

Why does OT have flat derivations? There are empirical arguments (many
of which are summarized in McCarthy 2002: 138–63), but the main
reason is theoretical parsimony. Nonflat derivations are often a way of
establishing priority relationships among linguistic requirements, and OT
already has a way of setting priorities, ranking. For example, in the
Wakashan language Nuuchahnulth (formerly called Nootka), there is a
process that labializes certain consonants after a round vowel, and there is
another process that delabializes them at the end of a syllable (Campbell
1973). When a consonant meets both of these conditions, what does it
do? In derivational approaches, the rule that applies last gets to decide, as shown
in (17). In OT, priority is determined by constraint ranking [see (18)].
(17) Nuuchahnulth /m%o:q/ → [m%o:q] with rules
Underlying /m%o:q/
w
q→q /o___ m%o:qw
q →q /___.
w
m%o:q
Surface [m%o:q]
(18) Nuuchahnulth /m%o:q/ → [m%o:q] in OT

/m%o:q/ *qw. *oq Ident(round)

a. → m%o:q *
% w
b. m o:q *W L *W

Although standard approaches to OT have nothing like the derivations in

(15) and (17), there are proposals to link several different OT grammars
serially, with the output of one becoming the input to the next. This idea
has obvious relevance in any situation where we are dealing with separate
modules of the grammar, such as the syntax and the phonology. It has also
been recruited in phonology to account for phenomena such as within-
language differences between word and sentence phonology (McCarthy and
Prince 1993: Appendix; Potter 1994; Kenstowicz 1995; Orgun 1996; Cohn
and McCarthy 1994/1998; Hale Kissock and Reiss 1998; Bermúdez-Otero
1999; McCarthy 2000; Rubach 2000; Ito and Mester 2001, 2003a,b;
Kiparsky 2003).
Suggestions for further reading. McCarthy (2007) includes an
overview and bibliography of most work related to derivations in OT
phonology. Many of the contributions to Hermans and van Oostendorp
(1999) and Roca (1997) are also relevant.

6.8 gen and faithfulness beyond phonology

In phonology, it’s reasonably clear what Gen and the faithfulness constraints
should look like. But what about syntax? In general, how do we go about
establishing the nature of Gen and faithfulness in domains other than phonology?
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
284 John J. McCarthy

There is another way of putting this question: what do OT’s basic

principles tell us about the necessary properties of Gen and the faithfulness
part of Con in any empirical domain? One important idea is that the
input and Gen together define the space of candidates that compete to
be the surface realization of that input. This places lower and upper bounds
on Gen’s freedom of analysis, its capacity to create output candidates that
are different from the input (see Section 3). If a hypothesized Gen is not
free enough, then some inputs might have only observationally ungram-
matical expressions among their output candidates. No adequate theory
of Gen should ever do this, because every candidate set must contain
some winner (Section 6.4). If a hypothesized Gen is too free, on the other
hand, then some observationally grammatical realization of an input might
lose to another, more harmonic candidate.8 Clearly, assumptions about the
nature of the input as well as assumptions about the nature of Gen will
affect how these criteria work out in practice.
Another important idea is that any property of the input that Gen can
alter – any property that can vary among the output candidates for a given
input – must be protected by some faithfulness constraint if it is to affect
the output of the grammar. The reason: apart from possible restrictions on
Gen’s freedom of analysis, faithfulness constraints are the only mechanism
in OT for transmitting information from the input to the output.
From these statements, both of which derive from basic assumptions of
the theory, it follows that a hypothesis about any one of the input, Gen,
the candidate set, and the faithfulness constraints will go a long way
toward determining all of the rest. Legendre, Smolensky, and Wilson
(1998) show how this can be done in a real-life example, wh movement.
Suggestions for further reading. Blaho, Bye, and Krämer (forth-
coming) is an anthology of work on Gen and freedom of analysis.

7. Where to Go Next
Readers whose interest in OT has been piqued by this brief overview
should plan on reading Prince and Smolensky (1993/2004), especially parts I
and III if their interests are not primarily phonological. This exercise is
perhaps best preceded or accompanied by reading a more didactic treatment,
such as Kager (1999), McCarthy (2002), or one of the other relatively
accessible works cited at the end of Section 3.
The next step after that depends on the individual reader’s interests. If
they tend toward phonology, then the articles collected in McCarthy (2003b)
are probably the best place to start. Two other useful anthologies, Lombardi
(2001) and Féry and van de Vijver (2003), are focused on segmental and
syllabic phonology, respectively. Readers of a syntactic bent could not do
better than to consult two relatively recent anthologies of articles on OT
syntax, Legendre, Grimshaw, and Vikner (2001) and Sells et al. (2001). In
addition, there are now several anthologies on OT semantics and pragmatics
© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 285

(de Hoop and de Swart 1999; Blutner and Zeevat 2004; Blutner, de
Hoop and Hendricks 2005), and one on historical linguistics (Holt 2003).
Finally, readers should be aware of two remarkable compendia of research
on OT. One is the Rutgers Optimality Archive (ROA, http://roa.rutgers.edu).
ROA, which was created by Alan Prince in 1993, is an electronic repos-
itory of ‘work in, on, or about OT’. By the time this article is published,
ROA will probably have its 1000th entry. The other notable compendium
is the two-volume set The Harmonic Mind, edited by Paul Smolensky and
Géraldine Legendre (2006). In almost 1200 pages, it collects the work of
the editors and their collaborators on topics in language, human cognition,
the mind, and the brain, much of it centered on Optimality Theory.

Short Biography
John J. McCarthy’s research deals with various topics in phonology, prosodic
morphology, and Optimality Theory. He is the author of A Thematic Guide
to Optimality Theory (Cambridge, 2002) and Hidden Generalizations: Phonological
Opacity in Optimality Theory (Equinox, 2007), and he edited Optimality
Theory in Phonology: A Reader (Blackwell, 2004). His textbook Doing
Optimality Theory will be published by Blackwell in 2008. His articles have
appeared in the journals Phonology, Natural Language and Linguistic Theory,
Language, and Linguistic Inquiry. He has taught courses on Optimality
Theory at two Linguistic Society of America Summer Institutes (Cornell
1997 and MIT 2005). He received a Guggenheim Fellowship in 1992 and
was elected to the American Academy of Arts and Sciences in 2005.

Notes
* Correspondence address: John J. McCarthy, Department of Linguistics, University of
Massachusetts, 226 South College, 150 Hicks Way, Amherst, MA 01003-9274, USA. Email:
[email protected].
1
I am very grateful for comments and suggestions received from Kathryn Flack, Karen Jesney,
Cathie Ringen, an anonymous reviewer, and the University of California San Diego review
team (Eric Bakovic, Kathryn Davidson, Gabriel Doyle, J. Grant Loomis, Bozena Pajak, and
Mohammad Salihie).
2
In theory, Eval could run out of constraints before the candidate set has been reduced to a
single member. This can only happen if two or more candidates receive exactly the same
number of violation marks from all of the constraints. In actual practice, that is unlikely to
occur.
3
Chomsky (1995: 380) criticizes faithfulness constraints on the grounds that identity between
input and output is ‘a principle that is virtually never satisfied’.
4
See Karttunen (2006) for some discussion of this point.
5
n!, pronounced ‘n factorial’, is the product n * n–1 * n–2 * . . . * 2.
6
Examples like /fi cpia/ → [fjcpia] show that Emai tolerates vowel sequences word internally.
Although Schaefer (1987) does not say how these sequences are syllabified, presumably at least
some are heterosyllabic: [fjc.pi.a]. Onsetless syllables are therefore tolerated in word-medial
position, although they are eliminated word intially. This is an indication that the high-ranking
markedness constraint in Emai is specific to word-initial syllables. Flack (2007) has identified a
number of languages that require onsets word initially but not medially, and this leads her to

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
286 John J. McCarthy

distinguish between OnsetWord and OnsetSyllable constraints. Emai’s tolerance for onsetless sylla-
bles phrase initially (e.g., [clebe]) has a similar etiology.
7
The Southern Min or Taiwanese tone circle has been cited as a potential counterexample to
this claim. There is, however, a body of work arguing ‘that Taiwanese tone sandhi is better
viewed as a set of essentially arbitrary alternations between stored allomorphs’ (Myers and Tsay
2002: 4). See Myers and Tsay for extensive references.
8
This criterion must be used with care, because of the possibility of variable constraint ranking
(Anttila 1997; Boersma and Hayes 2001 and others).

Works Cited
Aissen, Judith. 1999. Markedness and subject choice in Optimality Theory. Natural Language
and Linguistic Theory 17.673–711.
Anderson, Stephen R. 1985. Phonology in the twentieth century: theories of rules and theories
of representations. Chicago, IL: University of Chicago Press.
Anttila, Arto. 1997. Variation in Finnish phonology and morphology. Stanford University,
Doctoral dissertation.
Archangeli, Diana. 1997. Optimality Theory: an introduction to linguistics in the 1990s.
Optimality Theory: an overview, ed. by Diana Archangeli and D. Terence Langendoen, 1–
32. Oxford, UK: Blackwell.
Bakovic, Eric and Edward Keer. 2001. Optionality and ineffability. Optimality-Theoretic syntax,
ed. by Géraldine Legendre, Jane Grimshaw and Sten Vikner, 97–112. Cambridge, MA: MIT
Press. (available at http://roa.rutgers.edu/view.php3?start=384.)
Becker, Michael. 2006. CCamelOT – An implementation of OT-CC’s GEN and EVAL in Perl.
Albuquerque, NM: Linguistic Society of America, 80th Annual Meeting.
Beckman, Jill. 1997. Positional faithfulness, positional neutralization, and Shona vowel harmony.
Phonology 14.1– 46.
——. 1999. Positional faithfulness: an Optimality Theoretic treatment of phonological asymmetries.
New York: Garland. (1998 Doctoral dissertation, Amherst, MA: University of Massachusetts;
available at http://roa.rutgers.edu/view.php3?start=234.)
Beckman, Jill N. and Catherine O. Ringen. 2004. Contrast and redundancy in OT. WCCFL
23 Proceedings, ed. by B. Schmeiser, V. Chand, A. Kelleher and A. Rodriguez, 101–14.
Somerville, MA: Cascadilla Press.
Bermúdez-Otero, Ricardo. 1999. Constraint interaction in language change: quantity in English
and Germanic, University of Manchester, Doctoral dissertation.
Blaho, Sylvia, Patrik Bye, and Martin Krämer. forthcoming 2007. Freedom of analysis? Berlin,
Germany: Mouton de Gruyter.
Blutner, Reinhard, and Henk Zeevat. (eds.) 2004. Optimality Theory and pragmatics.
Basingstoke, Hampshire, UK: Palgrave Macmillan.
Blutner, Reinhard, Helen de Hoop, and Petra Hendricks. (eds.) 2005. Optimal communication.
Stanford, CA: CSLI Publications.
Boersma, Paul, and Bruce Hayes. 2001. Empirical tests of the gradual learning algorithm.
Linguistic Inquiry 32.45–86. (available at http://roa.rutgers.edu/view.php3?start=348.)
Booij, Geert. 1995. The phonology of Dutch. Oxford, UK: Clarendon Press.
Brasoveanu, Adrian and Alan Prince. 2005. Ranking and necessity. New Brunswick, NJ:
Rutgers University. (available at http://roa.rutgers.edu/view.php3?start=794.)
Bresnan, Joan. 2001a. Explaining morphosyntactic competition. The handbook of contemporary
syntactic theory, ed. by Mark Baltin and Chris Collins, 11– 44. Oxford, UK: Blackwell.
(available at http://roa.rutgers.edu/view.php3?start=299.)
——. 2001b. The emergence of the unmarked pronoun. Optimality-Theoretic syntax, ed. by
Géraldine Legendre, Jane Grimshaw and Sten Vikner, 113– 42. Cambridge, MA: MIT Press.
(available at http://roa.rutgers.edu/view.php3?start=179.)
Campbell, Lyle. 1973. Extrinsic order lives. Bloomington, IN: Indiana University Linguistics
Club Publications.
Casali, Roderic F. 1996. Resolving Hiatus. UCLA, Doctoral dissertation. (available at
http://roa.rutgers.edu/view.php3?start=215.)

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 287

Chomsky, Noam. 1957. Syntactic structures. The Hague, The Netherlands: Mouton.
——. 1968. Language and mind. New York: Harcourt Brace Jovanovich.
——. 1981. Lectures on government and binding. Dordrecht, The Netherlands: Foris.
——. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam and Morris Halle. 1968. The sound pattern of English. New York: Harper
& Row.
Chomsky, Noam and Howard Lasnik. 1977. Filters and control. Linguistic Inquiry 8.425–
504.
Cohn, Abigail and John J. McCarthy. 1994/1998. Alignment and parallelism in Indonesian
phonology. Working papers of the Cornell phonetics laboratory 12.53 –137. (available at
http://roa.rutgers.edu/view.php3?start=25.)
de Hoop, Helen and Henriette de Swart. (eds.) 1999. Papers on optimality theoretic semantics.
Utrecht, The Netherlands: Utrecht Institute of Linguistics/Onderzoeksinstituut voor Taal en Spraak.
Dinnsen, Daniel A. and Judith A. Gierut. (eds.) forthcoming 2007. Optimality Theory: pho-
nological acquisition and disorders. London: Equinox Publishing.
Donegan, Patricia J. and David Stampe. 1979. The study of natural phonology. Current
approaches to Phonological Theory, ed. by Daniel A. Dinnsen, 126–73. Bloomington, IN:
Indiana University Press.
Eisner, Jason. 1997. Efficient generation in Primitive Optimality Theory. Philadelphia, PA:
University of Pennsylvania. (available at http://roa.rutgers.edu/view.php3?start=206.)
Ellison, T. Mark. 1994. Phonological derivation in Optimality Theory. Proceedings of the 15th
International Conference on Computational Linguistics (COLING), 1007–13. Kyoto.
(available at http://roa.rutgers.edu/view.php3?start=75.)
Fanselow, Gisbert, Matthias Schlesewsky, Damir Cavar, and Reinhold Kliegl. 1999. Optimal
Parsing, syntactic parsing preferences, and Optimality Theory. Potsdam, Germany: University
of Potsdam. (available at http://roa.rutgers.edu/view.php3?start=367.)
Féry, Caroline and Ruben van de Vijver. 2003. The syllable in Optimality Theory. Cambridge,
UK: Cambridge University Press.
Flack, Kathryn. 2007. The sources of phonological markedness. Amherst, MA: University of
Massachusetts Amherst Doctoral dissertation.
Fodor, Jerry A., Thomas G. Bever, and Merrill Garrett. 1974. The psychology of language.
New York: McGraw Hill.
Frank, Robert and Giorgio Satta. 1998. Optimality Theory and the computational com-
plexity of constraint violability. Computational Linguistics 24.307–15. (available at
http://roa.rutgers.edu/view.php3?start=228.)
Gerdemann, Dale and Gertjan van Noord. 2000. Approximation and exactness in finite state
Optimality Theory. Tübingen and Groningen: University of Tübingen and University of
Groningen. (available at http://arxiv.org/abs/cs/0006038v1).
Gibson, Edward and Kevin Broihier. 1998. Optimality Theory and human sentence processing.
Is the best good enough? Optimality and competition in syntax, ed. by Pilar Barbosa, Danny
Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky, 157– 91. Cambridge, MA: MIT Press.
Goldsmith, John. 1976a. Autosegmental phonology. MIT, Doctoral dissertation (Published by
Garland Press, New York, 1979).
——. 1976b. An overview of autosegmental phonology. Linguistic Analysis 2.23 – 68.
——. 1993. Introduction. The last phonological rule: reflections on constraints and derivations,
ed. by John Goldsmith, 1–20. Chicago, IL: University of Chicago Press.
Gouskova, Maria. 2003. Deriving economy: syncope in Optimality Theory. Amherst, MA: University
of Massachusetts, Doctoral dissertation. (available at http://roa.rutgers.edu/view.php3?start=610.)
——. 2007. DEP: beyond epenthesis. Linguistic Inquiry 38.4.
Grimshaw, Jane. 1997. Projection, heads, and optimality. Linguistic Inquiry 28.373 – 422.
(available at http://roa.rutgers.edu/view.php3?start=68.)
——. 2001. Optimal clitic positions and the lexicon in Romance clitic systems. Optimality-
Theoretic syntax, ed. by Géraldine Legendre, Jane Grimshaw and Sten Vikner, 205–40.
Cambridge, MA: MIT Press. (available at http://roa.rutgers.edu/view.php3?start=374.)
——. 2002. Economy of structure in OT. University of Massachusetts Occasional Papers in Lin-
guistics 26: Papers in Optimality Theory II, ed. by Angela Carpenter, Andries Coetzee and Paul
de Lacy, 81–120. Amherst, MA: GLSA. (available at http://roa.rutgers.edu/view.php3?start=434.)

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
288 John J. McCarthy

Grimshaw, Jane and Vieri Samek-Lodovici. 1995. Optimal subjects. Papers in Optimality
Theory, ed. by Jill Beckman, Laura Walsh Dickey and Suzanne Urbanczyk, 589 – 606.
Amherst, MA: GLSA Publications.
——. 1998. Optimal subjects and subject universals. Is the best good enough? Optimality and
competition in syntax, ed. by Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis
and David Pesetsky, 193–219. Cambridge, MA: MIT Press.
Hale, Mark, Madelyn Kissock, and Charles Reiss. 1998. Output-output correspondence in
Optimality Theory. The Proceedings of the West Coast Conference on Formal Linguistics
16, ed. by Emily Curtis, James Lyle and Gabriel Webster, 223–36. Stanford, CA: CSLI
Publications. (available at http://roa.rutgers.edu/view.php3?start=202.)
Hammond, Michael. 1997. Parsing syllables: modeling OT computationally. Tucson, AZ:
University of Arizona. (available at http://roa.rutgers.edu/view.php3?start=222.)
Hayes, Bruce. 1995. Metrical stress theory: principles and case studies. Chicago, IL: The
University of Chicago Press.
Hermans, Ben and Marc van Oostendorp. (eds.) 1999. The derivational residue in phonological
Optimality Theory. Amsterdam, The Netherlands: John Benjamins.
Holt, D. Eric (ed.) 2003. Optimality Theory and language change. Dordrecht, The Netherlands:
Kluwer Academic.
Ito, Junko. 1989. A prosodic theory of epenthesis. Natural Language and Linguistic Theory 7.217–59.
Ito, Junko and Armin Mester. 2001. Structure preservation and stratal opacity in German.
Segmental phonology in Optimality Theory, ed. by Linda Lombardi, 261–95. Cambridge,
UK: Cambridge University Press.
——. 2003a. On the sources of opacity in OT: Coda processes in German. The syllable in
Optimality Theory, ed. by Caroline Féry and Ruben van de Vijver, 271–303. Cambridge,
UK: Cambridge University Press. (available at http://roa.rutgers.edu/view.php3?start=347.)
——. 2003b. Lexical and postlexical phonology in Optimality Theory: evidence from Japanese.
Resolving conflicts in grammars: Optimality Theory in syntax, morphology, and phonology.
Linguistische Berichte Sonderheft 11, ed. by Gisbert Fanselow and Caroline Féry, 183 –207.
(available at http://people.ucsc.edu/~mester/papers/2003_ito_mester_lexpostlex.pdf ).
Kager, René. 1999. Optimality Theory. Cambridge, UK: Cambridge University Press.
Kager, René, Joe Pater, and Wim Zonneveld. (eds.) 2004. Constraints in phonological acqui-
sition. Cambridge, UK: Cambridge University Press.
Karttunen, Lauri. 1998. The proper treatment of optimality in computational phonology.
FSMNLP ‘98: Proceedings of the International Workshop on Finite State Methods in
Natural Language Processing, 1–12. Ankara, Turkey: Bilkent University. (available at
http://roa.rutgers.edu/view.php3?start=258.)
——. 2006. The insufficiency of paper-and-pencil linguistics: the case of Finnish prosody.
Intelligent linguistic architectures: variations on themes by Ronald M. Kaplan, ed. by Miriam
Butt, Mary Dalrymple and Tracy Holloway King, 287–300. Stanford, CA: CSLI Publications.
(available at http://roa.rutgers.edu/view.php3?start=818.)
Kenstowicz, Michael. 1995. Cyclic vs. non-cyclic constraint evaluation. Phonology 12.397–436.
(available at http://roa.rutgers.edu/view.php3?start=31.)
Kiparsky, Paul. 2003. Syllables and moras in Arabic. The syllable in Optimality Theory, ed. by
Caroline Féry and Ruben van de Vijver, 147–82. Cambridge, UK: Cambridge University
Press (available at http://www.stanford.edu/~kiparsky/Papers/syll.pdf ).
Kisseberth, Charles. 1970. On the functional unity of phonological rules. Linguistic Inquiry
1.291–306.
Legendre, Géraldine. 2001. An introduction to Optimality Theory in syntax. Optimality-
Theoretic syntax, ed. by Géraldine Legendre, Jane Grimshaw and Sten Vikner, 1–28. Cam-
bridge, MA: MIT Press.
Legendre, Géraldine, Jane Grimshaw, and Sten Vikner. (eds.) 2001. Optimality-Theoretic
syntax. Cambridge, MA: MIT Press.
Legendre, Géraldine, Paul Smolensky, and Colin Wilson. 1998. When is less more? Faithfulness
and minimal links in wh-chains. Is the best good enough? Optimality and competition in
syntax, ed. by Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky,
249–89. Cambridge, MA: MIT Press. (available at http://roa.rutgers.edu/view.php3?start=117.)

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 289

Legendre, Géraldine, Antonella Sorace, and Paul Smolensky. 2006. The Optimality Theory-
Harmonic grammar connection. The harmonic mind: from neural computation to Optimality-
Theoretic grammar, vol. 2: linguistic and philosophical implications, ed. by Paul Smolensky
and Géraldine Legendre, 339–402. Cambridge, MA: MIT Press/Bradford Books.
Lombardi, Linda (ed.) 2001. Segmental phonology in Optimality Theory: constraints and
representations. Cambridge, UK: Cambridge University Press.
McCarthy, John J. 2000. The prosody of phase in Rotuman. Natural Language and Linguistic
Theory 18.147–97.
——. 2002. A thematic guide to Optimality Theory. Cambridge, UK: Cambridge University Press.
——. 2003a. Optimality Theory: an overview. Oxford international encyclopedia of linguistics,
2nd edn., ed. by William Frawley. Oxford, UK: Oxford University Press.
——. (ed.) 2003b. Optimality Theory in phonology: A reader. Malden, MA: Blackwell.
——. 2007. Hidden generalizations: phonological opacity in Optimality Theory. London: Equinox
Publishing.
——. forthcoming 2008. Doing Optimality Theory. Malden, MA: Blackwell.
McCarthy, John J., and Alan Prince. 1993. Prosodic Morphology: Constraint Interaction and
Satisfaction. New Brunswick, NJ: Rutgers University Center for Cognitive Science. (avail-
able at http://roa.rutgers.edu/view.php3?start=482.)
——. 1994. The emergence of the unmarked: optimality in prosodic morphology. Proceedings
of the North East Linguistic Society 24, ed. by Mercè Gonzàlez, 333–79. Amherst, MA:
GLSA Publications [available at http://roa.rutgers.edu/view.php3?start=13; excerpted in
Optimality Theory in Phonology: A Reader, ed. by John J. McCarthy, Malden, MA and
Oxford, Blackwell (2004)].
——. 1995. Faithfulness and Reduplicative Identity. University of Massachusetts Occasional
Papers in Linguistics 18, ed. by Jill Beckman, Laura Walsh Dickey and Suzanne Urbanczyk,
249–384. Amherst, MA: GLSA Publications.
——. 1999. Faithfulness and identity in Prosodic Morphology. The Prosody-Morphology
Interface, ed. by René Kager, Harry van der Hulst and Wim Zonneveld, 218–309. Cam-
bridge: Cambridge University Press [excerpted in Optimality Theory in Phonology: A
Reader, ed. by John J. McCarthy, Malden, MA and Oxford, Blackwell (2004)].
Moreton, Elliott. 2003. Non-computable functions in Optimality Theory. Optimality Theory
in phonology: A Reader, ed. by John J. McCarthy, 141–63. Malden, MA: Blackwell. (avail-
able at http://roa.rutgers.edu/view.php3?start=364.)
Myers, James, and Jane Tsay. 2002. Neutralization in Taiwanese tone sandhi. Proceedings of
the Berkeley Linguistics Society 22. 394–405. Taiwan: National Chung Cheng University
(available at http://www.ccunix.ccu.edu.tw/~lngproc/MyersTsayTTSneut.pdf ).
Orgun, C. Orhan. 1996. Sign-based morphology and phonology, with special attention to
Optimality Theory. Berkeley, MA : University of California, Doctoral dissertation. (available
at http://roa.rutgers.edu/view.php3?start=171.)
Orgun, C. Orhan, and Ronald Sprouse. 1999. From MParse to control: deriving ungrammat-
icality. Phonology 16.191–220. (available at http://roa.rutgers.edu/view.php3?start=224.)
Payne, David L. 1981. The phonology and morphology of Axininca Campa. Summer Institute
of Linguistics Publications in Linguistics, 66. Dallas, TX: Summer Institute of Linguistics and
University of Texas at Arlington.
Perlmutter, David. 1971. Deep and surface structure constraints in syntax. New York: Holt,
Rinehart, & Winston.
Pesetsky, David. 1997. Optimality Theory and syntax: movement and pronunciation. Optimal-
ity Theory: an overview, ed. by Diana Archangeli and D. Terence Langendoen, 134–70.
Oxford, UK: Blackwell.
Potter, Brian. 1994. Serial optimality in Mohawk prosody. Proceedings of the Thirtieth Annual
Regional Meeting of the Chicago Linguistics Society, ed. by Katharine Beals et al., 347–61.
Chicago, IL: Chicago Linguistics Society.
Prentice, D. J. 1971. The Murut languages of Sabah. Canberra, Australia: Australian National
University.
Prince, Alan. 1983. Relating to the grid. Linguistic Inquiry 14.19–100.
——. 1997a. Endogenous constraints on Optimality Theory. Lecture notes for course presented

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
290 John J. McCarthy

at Summer Institute of the Linguistic Society of America, Cornell University, Ithaca, NY

(available at http://ling.rutgers.edu/gamma/talks/insthdt1.pdf ).
——. 1997b. Stringency and anti-Paninian hierarchies. Lecture notes for course presented at
Summer Institute of the Linguistic Society of America, Cornell University, Ithaca, NY
(available at http://ling.rutgers.edu/gamma/talks/insthdt2.pdf ).
——. 2002. Arguing optimality. Papers in Optimality Theory II (= University of Massachusetts
Occasional Papers 26), ed. by Angela Carpenter, Andries Coetzee and Paul de Lacy, 269–
304. Amherst, MA: GLSA. (available at http://roa.rutgers.edu/view.php3?start=562.)
Prince, Alan and Paul Smolensky. 1993/2004. Optimality Theory: constraint interaction in generative
grammar. Malden, MA, and Oxford, UK: Blackwell (revision of 1993 technical report, Rutgers
University Center for Cognitive Science; available at http://roa.rutgers.edu/view.php3?start=537.)
——. 1997. Optimality: from neural networks to universal grammar. Science 275.1604–10.
——. 2003. Optimality Theory in phonology. Oxford international encyclopedia of linguis-
tics, ed. by William Frawley, 212–222. Oxford, UK: Oxford University Press (available at
http://ling.rutgers.edu/gamma/oiel.pdf ).
Pulleyblank, Douglas. 1986. Tone in lexical phonology. Dordrecht, The Netherlands:
D. Reidel.
Rice, Curt. 2003. Syllabic well-formedness in Norwegian imperatives. Nordlyd 31.372–384.
(available at http://roa.rutgers.edu/view.php3?start=642 under title ‘Dialectal variation in
Norwegian imperatives’).
——. 2005. Optimal gaps in optimal paradigms. Catalan Journal of Linguistics 4.155–70 (special
issue on phonology in morphology edited by Maria-Rosa Lloret and Jesús Jiménez. Available
at LingBuzz, http://ling.auf.net/buzzdocs/).
——. (ed.) forthcoming 2008. Modeling ungrammaticality in Optimality Theory. London:
Equinox Publishing.
Raffelsiefen, Renate. 2004. Absolute ill-formedness and other morphophonological effects.
Phonology 21.91–142.
Riggle, Jason. 2004. Generation, recognition, and learning in finite state optimality theory.
UCLA, Doctoral dissertation (available at http://hum.uchicago.edu/~jriggle/riggle04_2.pdf.).
Roca, Iggy (ed.) 1997. Derivations and constraints in phonology. Oxford, UK: Oxford University
Press.
Rubach, Jerzy. 2000. Glide and glottal stop insertion in Slavic languages: a DOT analysis.
Linguistic Inquiry 31.271–317.
Samek-Lodovici, Vieri. 1996. Constraints on subjects: an Optimality Theoretic analysis. Rutgers
University, Doctoral dissertation. (available at http://roa.rutgers.edu/view.php3?start=148.)
Schaefer, Ronald P. 1987. An initial orthography and lexicon for Emai. Bloomington, IN:
Indiana University Linguistics Club Publications.
Sells, Peter, Joan Bresnan, Miriam Butt, and Tracy Holloway King. 2001. Formal and empirical
issues in Optimality Theoretic syntax. Stanford, CA: CSLI Publications.
Smolensky, Paul and Géraldine Legendre. 2006. The harmonic mind: from neural computation
to Optimality-Theoretic grammar. Cambridge, MA: MIT Press.
Smolensky, Paul, Géraldine Legendre, and Bruce Tesar. 2006. Optimality Theory: the structure,
use, and acquisition of grammatical knowledge. The harmonic mind: from neural computation
to Optimality-Theoretic grammar, vol. 1: Cognitive architecture, ed. by Paul Smolensky and
Géraldine Legendre, 453–535. Cambridge, MA: MIT Press/Bradford Books.
Stampe, David. 1973. A dissertation on natural phonology. University of Chicago, Doctoral
dissertation. (published by Garland, New York, 1979).
Tesar, Bruce. 1995. Computational Optimality Theory. University of Colorado, Doctoral dissertation.
(available at http://roa.rutgers.edu/view.php3?start=90.)
——. 1998. Using the mutual inconsistency of structural descriptions to overcome ambiguity
in language learning. Proceedings of the North East Linguistic Society 28, ed. by Pius N.
Tamanji and Kiyomi Kusumoto, 469–83. Amherst, MA: GLSA Publications. (available at
http://roa.rutgers.edu/view.php3?start=426.)
——. 1999. Robust interpretive parsing in metrical stress theory. The Proceedings of the West Coast
Conference on Formal Linguistics 17, ed. by Kimary N. Shahin, Susan J. Blake and Eun-Sook
Kim. Stanford, CA: CSLI Publications. (available at http://roa.rutgers.edu/view.php3?start=262.)

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd
 1749818x, 2007, 4, Downloaded from https://compass.onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2007.00018.x by Jawaharlal Nehru University Central Library, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
What Is Optimality Theory? 291

Tesar, Bruce and Alan Prince,. 2004. Using phonotactics to learn phonological alterna-
tions. CLS 39, Part II: The panels. Chicago, IL: Chicago Linguistic Society. (available at
http://roa.rutgers.edu/view.php3?start=620.)
Tesar, Bruce and Paul Smolensky. 1998. Learnability in Optimality Theory. Linguistic Inquiry
29.229 – 68 [available at http://roa.rutgers.edu/view.php3?start=155; reprinted in Optimality
Theory in Phonology: A Reader, ed. by John J. McCarthy, Malden, MA and Oxford,
Blackwell (2004)].
——. 2000. Learnability in Optimality Theory. Cambridge, MA: MIT Press.
Tesar, Bruce, Jane Grimshaw, and Alan Prince. 1999. Linguistic and cognitive explanation in
Optimality Theory. What is Cognitive Science? ed. by Ernest Lepore and Zenon Pylyshyn,
295–326. Oxford, UK: Blackwell.
Urbanczyk, Suzanne. 2006. Reduplicative form and the root-affix asymmetry. Natural
Language and Linguistic Theory 24.179–240.
Wareham, Harold T. 1998. Systematic parametrized complexity analysis in computational phonology.
University of Victoria, PhD dissertation. (available at http://roa.rutgers.edu/view.php3?start=318.)

© 2007 The Author Language and Linguistics Compass 1/4 (2007): 260– 291, 10.1111/j.1749-818X.2007.00018.x
Journal Compilation © 2007 Blackwell Publishing Ltd