0% found this document useful (0 votes)

21 views52 pages

JohnMcCarthy 2007 Chapter2 HiddenGeneralizations

This chapter discusses opacity in phonology, analyzing its implications within rule-based phonology and Optimality Theory (OT). It highlights the differences between underlying and surface representations, the role of derivations, and the interactions between phonological rules, particularly focusing on feeding and bleeding orders. The conclusion suggests that a serial derivation approach may better accommodate the complexities introduced by opacity while retaining essential elements of classic OT.

Uploaded by

Arwa Alsahafi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

21 views52 pages

JohnMcCarthy 2007 Chapter2 HiddenGeneralizations

Uploaded by

Arwa Alsahafi

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

2 Opacity, derivations, and Optimality

Theory

2.1 Overview
All rights reserved. May not be reproduced in any form without permission from the publisher, except fair uses permitted under U.S. or applicable copyright law.

This chapter begins (§2.2) by explaining what opacity is and how it is analyzed
in rule-based phonology. The discussion then turns (§2.3) to a description of
‘classic’ Optimality Theory, the problems that opacity presents for classic OT,
and various ideas about how to modify the classic theory to accommodate it.
The conclusion I draw (§2.4) is that there is something fundamentally correct
about rule-based phonology’s serial derivation, leading to the proposal in §3
for an analogue of the serial derivation in a framework that retains all of classic
OT’s essential elements.

2.2 Opacity and derivations

2.2.1 Levels of representation
The theory of generative phonology recognizes two principal levels of repre-
sentation, underlying and surface. At the underlying level, every morpheme has
a unique representation. For example, the three principal surface alternants of
the English plural suffix — [-], [-], and [-] — are derived by phonological
rules from a single underlying representation, such as /-/. Only suppletive or
allomorphic alternants of morphemes require distinct underlying representa-
tions, such as the plural allomorphs /-/ of children and /--/ of geese.
When a morpheme alternates nonsuppletively, its underlying representa-
tion must be discovered by the analyst and by the learner. In paradigms like
German []/[] ‘multicolored/pl.’ and []/[] ‘federation/pl.’,
distinct underlying representations are required because there are distinct pat-
terns of voicing alternation: // ‘multi-colored’ is voiceless throughout its
paradigm and // ‘federation’ alternates between voiced and voiceless.
In theory and in actual practice, the relationship between the hypothesized
2007. Equinox Publishing Ltd.

EBSCO Publishing: eBook Collection (EBSCOhost) printed on 1/17/2025 6:43:52 AM UTC via PENNSYLVANIA STATE UNIVERSITY - PATERNO LIBRARY 547836; John
McCarthy; Hidden Generalizations: Phonological Opacity in Optimality Theory Account:ehost.
8 Hidden Generalizations

underlying representation and the observed surface paradigm is sometimes

less transparent than this.
Some recent research explores alternatives to positing an underlying level
of representation. These approaches are monostratal in the sense that they
recognize only a single level of representation, the surface form. In Declarative
Phonology (Scobbie, Coleman, and Bird 1996), the work of underlying repre-
sentations is done by constraints that describe morphemes. These descriptions
are crucially incomplete in the case of alternating morphemes: e.g., for German
[]/[], a constraint requires a final alveolar stop in ‘federation’ but
says nothing about its voicing. Another monostratal approach seeks to express
phonological generalizations purely in terms of relations between surface forms
(e.g., Albright 2002, Burzio 2002).
In this context, it is worth reviewing the reasons why generative phonology
posits an underlying level of representation (see Kenstowicz and Kisseberth
1979: chapter 6 for an overview of the evidence). The main argument comes
from paradigms where the relationships among surface forms make sense
only when mediated by an underlying form that is distinct from all of the
surface forms. Schane’s (1974) Palauan example in (2-1) is a well-known
case. Because unstressed vowels reduce to [] and there is only one stress per
word, disyllabic roots like ‘cover an opening’ and ‘pull out’ never show up
with more than one surface nonschwa vowel. The hypothesized underlying
representations // and // record the quality of the vowels as they
appear when stressed in different members of the paradigm. These underlying
representations incorporate all of the unpredictable phonological information
about these morphemes. In generative phonology, the underlying representation
of a root is the nexus of a set of related words, so it must contain sufficient
information to allow the surface forms of all of those words to be derived by
the grammar of the language. (See §4.3.3 for detailed argumentation in support
of underlying representations in a case similar to Palauan.)
(2-1) Palauan Vowel Reduction
Underlying Present Middle Future Participle
// - - ‘cover an opening’
// - -l ‘pull out’
Generative phonology in the tradition of The Sound Pattern of English (SPE
— Chomsky and Halle 1968) also allows for any number of levels intermediate
between the underlying and surface levels. These intermediate levels are the
result of sequential application of phonological rules. If a language has n rules
in its grammar, it has n–1 intermediate representations, each of which is a
potentially distinct way of representing the linguistic form that is being derived.

In Palauan, for example, the SPE theory requires an intermediate level at which
stress has been assigned but vowel reduction has not yet applied: /-/ → stress
[] →reduction []. Indeed, SPE requires rules to apply sequentially
even when simultaneous application would produce the same result.

2.2.2 Derivations
Any mapping from the underlying to the surface level of representation is a
derivation. In this sense, any multistratal theory of phonology is derivational,
including classic OT. The various multistratal theories differ significantly, how-
ever, in the complexity and internal organization of the derivations they posit.
The SPE approach to derivations retains considerable currency because
it is often assumed even in contemporary research that has moved far beyond
SPE’s other hypotheses about rules and representations (see §2.2.6). In SPE,
the grammar consists of an ordered list of rules. The rules are applied in a strict
sequence, with the output of rule i supplying the input to rule i+1. The output
of each rule (except the last) is therefore a level of representation intermediate
between the underlying and surface levels.
An important insight, due originally to Kiparsky (1968), is that rules may
have different functional relationships to one another. In the least interesting
case, a pair of rules may not interact at all — an example would be word-initial
vowel epenthesis and word-final obstruent devoicing. When rules do interact,
however, the functional relationship between them can often be classified as
feeding or bleeding.
Rule A is said to feed rule B if A can create additional inputs to B. If A in
fact precedes B, then A and B are in feeding order. (If B precedes A, then they
are in counterfeeding order, which will be explained in §2.2.3.) An example
of feeding order is the interaction between vowel and consonant epenthesis in
Classical Arabic. Words that begin with consonant clusters receive prothetic
[] (or [], if the next vowel is also []). As the derivation in (2-2) shows,
prothesis of [] is the result of a feeding interaction between [] epenthesis
before word-initial clusters (= rule A) and [] epenthesis before word-initial
vowels (= rule B).
(2-2) Feeding order in Classical Arabic
Underlying // ‘beat (m. sg.)!’
Vowel epenthesis 
[] epenthesis 
Surface []

Rule A is said to bleed rule B if A can eliminate potential inputs to B. If A in

fact precedes B, then A and B are in bleeding order. (If B precedes A, then
they are in counterbleeding order, which will also be explained in §2.2.3.) For
example, in a southern Palestinian variety of Arabic, progressive assimilation
of pharyngealization (= rule B) is blocked by high front segments, among them
[]. When the vowel [] is epenthesized into triconsonantal clusters (= rule A),
it also blocks assimilation, as shown in (2-3) (Davis 1995).
(2-3) Bleeding order in southern Palestinian Arabic
Underlying /-/ ‘her stomach’
Vowel epenthesis 
Progressive assimilation Blocked
Regressive assimilation 
Surface []
Feeding and bleeding orders have something in common: when rules apply in
feeding or bleeding order, those structures that are derived by rules are treated
exactly the same as similar structures that were already present in underlying
representation. For example, the process of []-epenthesis in Classical Arabic
applies to words with an underlying initial vowel, /--/ → []
‘the boy (nominative)’, and also to words with a derived initial vowel, such
as the intermediate representation [] in (2-2). Likewise, epenthetic and
nonepenthetic [] equally block progressive assimilation in Palestinian Arabic,
as shown by (2-3) and // → [], *[] ‘health’. In feeding
and bleeding interactions, what you see is what you get: when derived and
underived structures are identical, they exhibit identical phonological behavior.
This is emphatically not the case with counterfeeding and counterbleeding
interactions.

2.2.3 Opacity in derivations

If rule A feeds rule B and they are applied in the order B precedes A, then these
rules are said to be in counterfeeding order. For example, in a Bedouin Arabic
dialect (see §4.3.3), there are processes raising short /a/ to a high vowel in a
nonfinal open syllable (= rule A) and deleting short high vowels in nonfinal
open syllables (= rule B). These processes are in a feeding relationship, since
raising has the potential to create new inputs to deletion. But their order is
actually counterfeeding, as shown in (2-4). High vowels derived by raising
are treated differently from underlying high vowels; only the underlying high
vowels are subject to deletion. When rules apply in feeding order, derived and
underlying structures behave alike, but when they apply in counterfeeding
order, derived and underlying structures behave differently.

(2-4) Counterfeeding order in Bedouin Arabic1

Underlying a. // ‘he pushed’ b. /-/ ‘she drank’
Deletion — 
Raising  —
Surface [] []

The same is true of counterbleeding order, where rule A bleeds rule B but
they are applied with B preceding A. In this same Arabic dialect, there is also
a process palatalizing velars when they precede front vowels (see §3.3.3).
Deletion (= rule A) bleeds palatalization (= rule B), since deletion can remove
a high front vowel that would condition velar palatalization. But their order
is counterbleeding, as shown in (2-5). High front vowels, even when they are
absent from surface forms, induce adjoining velars to palatalize. Effects like
this are typical with counterbleeding order.
(2-5) Counterbleeding order in Bedouin Arabic
Underlying a. /-/ b. /--/
Palatalization  —
Deletion  
Surface [] []
‘ruling (masculine plural)’ ‘they (feminine) rule’
The result of counterfeeding and counterbleeding interactions is phonological
opacity. Kiparsky’s (1973: 79, 1976: 178–179) definition of opacity appears in
(2-6). Clause (c) of this definition describes all processes of neutralization and
so it is not relevant to our concerns here. We will therefore focus on clauses
(a) and (b).
(2-6) Opacity
A phonological rule P of the form A → B / C___D is opaque if there are
surface structures with any of the following characteristics:
a. instances of A in the environment C___D,
b. instances of B derived by P that occur in environments other than C___D,
or c. instances of B not derived by P that occur in the environment C___D.

In the derivation // → [] in (2-4), the high-vowel deletion rule is

opaque under clause (a) of this definition: [] has [] (= A) in an open
syllable (= C__D). Rules applied in counterfeeding order produce opacity of
the clause (a) type, in which surface forms contain phonological structures that
look like they should have undergone some process but in fact did not. 2
In the derivation /-/ → [] in (2-5), the palatalization rule
is opaque under clause (b) of this definition: [] has [] (= B) derived

by palatalization (= P), but [] is not adjacent to a front vowel (= C__D). Rules
applied in counterbleeding order produce opacity of this type, in which surface
forms contain derived phonological structures without the context that shows
how they were derived.
Counterfeeding and counterbleeding interactions supply the best — arguably,
the only — evidence for language-particular rule ordering. It is not surprising,
then, that skepticism about stipulated, language-particular ordering stimulated
efforts to deny that opaque interactions involve living phonological processes
(cf. §1.1). According to the proponents of Natural Generative Phonology
(NGP), authentic phonological rules must state surface-true generalizations
and they must be unordered (Hooper [Bybee] 1976, 1979, Vennemann 1972,
1974). NGP therefore maintains that opaque processes are merely the lexical-
ized residue of sound changes that are no longer productive — opaque rules
were said to be ‘not psychologically real’. (Recent work advocating similar
views in an OT context includes Green (2004), Mielke, Hume, and Armstrong
(2003), and Sanders (2002, 2003).) In fact, much if not all of the abstractness
controversy of the 1970’s, which dealt with proposed limits on the degree of
disparity between underlying and surface representations (see Kenstowicz and
Kisseberth 1977: Chapter 1, 1979: Chapter 6), was really an argument about
opacity, since abstract underlying forms can influence the output only if opaque
rules apply to them.
Certainly, there have been dubious analyses based on opaque rules and
excessively abstract underlying forms, but outright denial of all opaque interac-
tions is an empirically unsupportable overreaction. The example of Bedouin
Arabic is instructive. (See §4.3.3 for detailed discussion.) Al-Mozainy (1981)
presents several arguments that the opaque processes in this language are alive
and productive. First, they are active in borrowed words. Second, high vowel
deletion, even though it is opaque, applies productively in external sandhi, as
shown in (2-7). If a process applies in external sandhi, it cannot be lexicalized,
since it is impossible to list the infinite number of word collocations that the
syntax provides. 3
(2-7) Phrase-level deletion in Bedouin Arabic (Al-Mozainy 1981: 50–51)
/-/ . ‘writing the letter’
*.
/-/ . ‘you give it to the one
*. from the clan of
Musai`īd’
Third, the most compelling evidence that raising is productive comes from a
kind of play language. Although raising usually affects any short /a/ in a nonfinal
open syllable, there are phonological conditions under which raising regularly

fails to apply: after a guttural consonant ([], [], [], [], [], []), or before a
guttural consonant or coronal sonorant ([], [], []) that is itself followed by [a].
Bedouin Arabic has a secret language that permutes the consonants of the root,
and this will sometimes affect the position of gutturals or coronal sonorants
relative to the potentially raised vowel. When that happens, the vowel raises
or fails to raise in exact conformity with these generalizations, as (2-8) shows.
Other secret language data show that palatalization is also productive, even
though it is opaque (see §3.3.3). In sum, the opaque phonology of Bedouin
Arabic is also its living, productive phonology. (For further examples of proc-
esses that are productive yet opaque, see Donegan and Stampe (1979).)
(2-8) Raising alternations in a secret language
// Underlying representation
 Unpermuted form
 Raising as expected
 No raising before guttural + []
 "
 No raising after guttural
 "
Although this sort of evidence shows that opacity is a fact of phonological
life, certain types of opacity have received and deserve a skeptical reception. A
famous example is SPE’s // → [] right. The point is that a few dubious
analyses are not grounds to reject a theoretical construct, particularly when it
is strongly supported by sound analyses, as it is in Bedouin Arabic.
A type of opacity that received particular attention in the 1970’s is the
Duke-of-York derivation (Hogg 1978, Pullum 1976). Like the eponymous Duke
of the nursery rhyme, 4 underlying /A/ is changed by a rule to intermediate [B],
but a later rule changes [B] back into [A]. Unlike the Duke’s peregrinations, this
activity is not as pointless as it seems: during the temporary [B] stage, erstwhile
/A/ may opaquely escape an A-affecting process or cause a B-triggered one.
More often, though, Duke-of-York derivations are simply an artifact of the
commitment to sequential rule application. We will return to this topic, with
exemplification, in §2.3.2.

2.2.4 Simultaneous application

Discussions of rule ordering often overlook an important alternative to the
sequential derivation: simultaneous application. In many cases, rules could be
applied simultaneously with no loss of generality, and so it is worth exploring
which phenomena are and are not consistent with simultaneous application (for

discussion, see Anderson 1974: 64–67, Donegan and Stampe 1979: 150, Hyman
1993: 204ff., Koutsoudas 1976, Koutsoudas, Sanders, and Noll 1974: 5–8).
Since simultaneous application is a somewhat unfamiliar notion, we should
first get clear on what it means in rule-based phonology. A phonological rule
describes a configuration that must be met in the rule’s input — the rule’s struc-
tural description — and a change that is to be effected in the rule’s output — its
structural change. If two rules are applied simultaneously, then their structural
descriptions are analyzing exactly the same representation. It follows, then, that
neither rule has access to any information that is contributed by the other rule’s
structural change. In sequential application, by contrast, the later rule always
has access to information contributed by the earlier rule’s structural change.
Opaque interactions are often compatible with simultaneous application,
but transparent interactions require sequential application. The counterbleeding
derivation /-/ → [] in (2-5), for example, would also work
if the rules of palatalization and deletion were applied simultaneously. The
structural description of the palatalization rule analyzes an input that contains
[] before [], and so [] is palatalized with complete indifference to the fact
that the deletion rule is analyzing that same input toward the goal of deleting
[]. The important thing in this opaque derivation is that deletion must not
precede palatalization; that desideratum could in principle be fulfilled by order-
ing palatalization before deletion, as in (2-5), or by requiring them to apply
simultaneously.
Similarly, the counterfeeding derivation // → [] in (2-4) is pos-
sible if deletion of high vowels and raising of low vowels apply simultaneously.
The structural description of the high-vowel deletion rule is not met by //,
but the raising rule’s structural description is met, so only raising actually
applies. The important thing in this opaque derivation is that deletion should not
apply to the output of raising; that desideratum could in principle be fulfilled
by ordering raising before deletion, as in (2-4), or by requiring them to apply
simultaneously.
Feeding and bleeding interactions, however, are incompatible with
simultaneous application. In the feeding derivation // → [] (2-2),
for instance, the structural description of [] epenthesis is not met until after
vowel epenthesis has applied, so sequential application is necessary. In the
bleeding derivation /-/ → [] (2-3), simultaneous application
of vowel epenthesis and progressive assimilation would produce the result
*[], in which the epenthetic vowel is neither subject to nor a blocker
of assimilation.
It is interesting that simultaneous application of rules typically produces
opaque interactions but not transparent ones (unless the rules do not interact at

all). Classic OT, though it evaluates candidates in which the effects of several
processes are felt simultaneously, can model transparent interactions but not
opaque ones (see §2.3.3). The reason for this difference is that rules and OT
markedness constraints analyze different levels of representation. The structural
description of a rule is met by the rule’s input, which is sometimes identical to
the underlying representation. The structural description of an OT markedness
constraint is met in the ultimate output, the surface representation. Opacity
requires reference to conditions obtaining in presurface representations,
whereas transparency requires reference to conditions obtaining in surface
representations.

2.2.5 Theories of rule ordering

In SPE, the order in which the rules are applied is extrinsic, which means that it
is imposed on the rules by the language-particular grammar and cannot usually
be predicted from rule form or function. From about 1969 through 1980, a
voluminous literature developed around the question of whether some or even
all aspects of rule ordering could be predicted. (See Anderson (1979: 15–18)
and Iverson (1995) for brief surveys or Anderson (1974) and Kenstowicz and
Kisseberth (1977: chapters 4, 6) for more extensive discussion.)
An SPE-style phonology of Classical Arabic must include a statement to
the effect that vowel epenthesis precedes [] epenthesis to ensure that these
rules apply in the order observed in (2-2). In some revisions of that model (e.g.,
Anderson 1974, Koutsoudas, Sanders, and Noll 1974), this ordering statement
was dismissed as superfluous on the grounds that feeding order is unmarked
or natural. In what sense is feeding order natural? If rules are allowed to apply
freely at any point in the derivation when their structural descriptions are
met, then the result will be the same as (2-2). Feeding orders maximize rule
applicability. As was noted in §2.2.2, feeding orders also help to ensure that
rules enforce true generalizations about surface structure: in Arabic, no word
starts with a vowel because [] epenthesis is ordered after and thereby fed by
vowel epenthesis. In the terminology of Donegan and Stampe (1979: 157),
counterfeeding order of vowel epenthesis and [] epenthesis would act as a
‘constraint’ on the latter, preventing it from acting on derived representations
and so rendering the []-epenthesis generalization not categorically true.
Although feeding order was generally seen as natural and a second natural
order was believed to exist, there is disagreement in the literature of that era
over the question of whether the other natural rule order is bleeding or coun-
terbleeding. In the earliest work on this topic, Kiparsky (1968) argued that
historical change tends to maximize feeding and minimize bleeding orders. On

the assumption the languages are attracted toward natural rule orders, this would
mean that feeding and counterbleeding orders are natural. Anderson (1974)
integrates this idea into his theory of local ordering, according to which feeding
and counterbleeding order constitute a default case that can only be overridden
by language-particular stipulation. Anderson’s evidence includes analyses, none
of them uncontroversial, in which maintaining the natural interaction between
a pair of rules can cause them to apply in different orders within a single lan-
guage. (This is the sense in which ordering is ‘local’: the theory comprehends
ordering as a local relation between a pair of rules rather than a global list of
ordered rules in the SPE fashion. Cf. §3.2.3.) Koutsoudas, Sanders, and Noll
(1974) also argue for the naturalness of counterbleeding order.
Another body of work took the position that bleeding rather than coun-
terbleeding order is natural (Iverson 1974, Kenstowicz and Kisseberth 1971,
Kiparsky 1971). Apart from disagreements about analyses, the dispute is really
one about the principle that determines the natural orders. If feeding and coun-
terbleeding orders are natural, then what makes them natural is a principle that
favors maximizing rule applicability: if A feeds B, then A supplies additional
opportunities for rules to apply; and if A is not allowed to bleed B, then A cannot
steal away some of B’s opportunities to apply. If feeding and bleeding orders
are natural, then what makes them natural is a principle that favors maximizing
rule transparency: counterfeeding and counterbleeding orders produce opacity,
whereas feeding and bleeding orders produce transparency, in which the effects
of phonological generalizations are visible at surface structure.
In the course of research during the 1970’s, these and other ordering princi-
ples were discussed, and there were even proposals about priority relationships
among them (Anderson 1974: 217–218, Iverson 1976). The ultimate goal of
the research program, according to some (e.g., Koutsoudas, Sanders, and Noll
1974), was the elimination of all language-particular ordering statements in
favor of universal principles of applicational precedence. Supposedly prima
facie arguments against this position have been adduced, such as two Canadian
English dialects that differ solely in rule order (Bromberger and Halle 1989,
Joos 1942), but in reality the argument is not that easy to make (Iverson 1995:
612–613). There never was a knock-down argument in support of language-par-
ticular ordering, nor was there general agreement on rule-ordering principles.
Instead, the decade ended with a tacit consensus that research on universals of
rule ordering had gone about as far as it could go.

2.2.6 Later developments

Interest in the topic of rule ordering waned around 1980. (An important excep-
tion is Goldsmith (1993b).) As the focus of phonological research moved else-
where, however, matters of rule ordering and interaction sometimes reemerged
in new contexts.
The development of nonlinear phonology and underspecification theory,
beginning with works like Goldsmith (1976a), Kahn (1976), Liberman (1975),
Liberman and Prince (1977), Clements and Ford (1979), McCarthy (1981),
Prince (1983), and Archangeli (1984), took some of the analytic pressure off of
phonological rules and shifted it to well-formedness constraints on phonologi-
cal representations. In principle, an enriched theory of representations might
lead to a reduction in the need for language-particular rule ordering, but in
actual practice this line of research received little attention.
Satisfaction of representational constraints, however, required a new class
of persistent rules that apply automatically at any point in the derivation
when they are required (Chafe 1968, Myers 1991a). For example, when a
consonant becomes unsyllabified in the course of a derivation, a persistent
rule immediately adjoins it to a nearby syllable: // → []σ []σ []σ
→ syncope []σ  []σ →persistent []σ []σ. An important role of persistent rules,
then, is to repair violations of well-formedness conditions, thereby ensuring
that these conditions are respected not only at the beginning or end of the
derivation but also in the middle. The free (re-)applicability of persistent rules
is, of course, consistent with the principle favoring maximal rule application
that was mentioned at the end of §2.2.5.
Another relevant post-1980 development is the theory of Lexical Phonology
(Kaisse and Hargus 1993b, Kaisse and Shaw 1985, Kiparsky 1982, 1985,
Mohanan 1982, among many others). Lexical Phonology is an extension of
SPE’s theory of cyclic rule application. Certain rules may be designated as
cyclic — in SPE, these are the English stress rules — and this causes them to
apply repeatedly to successively larger morphological or syntactic constituents.
The cycle accounts for transderivational similarities like the following: 5
(i) Monomorphemic words like Kalamazoo and Winnepesaukee exhibit
the normal English stress pattern when three light syllables precede the
main stress. Derived words like accreditation and imagination devi-
ate from this pattern under the influence of accredit and imagine.
(ii) A closed, sonorant-final syllable is normally unstressed in prestress
position: serendipity, gorgonzola, Pennsylvania. But the same kind
of syllable may be stressed in the derived words authenticity and
condemnation under the influence of authentic and condemn.

In SPE, the aberrant stress of derived words is explained by their bracketing

and cyclic application of stress. The stress rules first apply on the inner con-
stituents of [accredit]ation or [authentic]ity and then on the outer constituents.
The primary stress assigned on the first cycle becomes a secondary stress on
the second cycle, when the stress rule reapplies and a new primary stress is
assigned further to the right. Monomorphemic Kalamazoo and serendipity
have no inner cycle, so they show the effects of just a single pass through the
stress rules.
Lexical Phonology departs from SPE in regarding cyclic application as
the norm rather than the exception for certain phonological rules. In addition,
Lexical Phonology imposes further structure on the grammar, dividing the
phonology up into separate components, called strata. At a minimum, there
are two such strata, lexical and postlexical. The input to the lexical stratum is
the underlying representation; the output of the lexical stratum is the input to
the postlexical stratum; and the output of the postlexical stratum is the surface
representation. Each stratum is a separate phonological grammar, though spe-
cific overlap requirements have sometimes been imposed (Borowsky 1986,
Kiparsky 1984: 141–143, Myers 1991b) (see §2.3.4.2). It is usually assumed
that the lexical stratum actually consists of several strata, and at each lexical
stratum a different set of morphological and phonological processes may be
in effect. For example, English suffixes like -ity are affixed in the first lexical
stratum, and that is also where the stress assignment rules apply. Suffixes
like -ness are not attached until the second lexical stratum, at which point the
stress assignment rules are no longer active. That is why suffixes like -ity are
stress-determining and suffixes like -ness are stress-neutral. It is sometimes also
assumed that rules apply cyclically within each lexical stratum, as each affix
of that stratum is added: e.g., /period/ → period → periodic → periodicity,
all within the first lexical stratum.
Lexical Phonology retains SPE’s assumption that the rules within a gram-
mar (= a stratum) are in a strict linear order. Despite this within-grammar strict
ordering, the same rule can be observed to reapply at different points in the
course of an entire derivation. As in SPE, the cycle offers one opportunity: a rule
can reapply in the same stratum as multiple affixes are added. But even without
any affixation at all, a rule can reapply if it is assigned to more than one stratum.
(This situation is not unusual.) If a rule is included in the grammar of more than
one stratum, it will simply reapply when the later stratum is reached. Lexical
Phonology is thereby able to reanalyze some (perhaps all) of the evidence that
had earlier been adduced in favor of a principle of unmarked feeding order or
maximizing rule application. An example is Kiparsky’s (1984) reanalysis of
the interaction of Icelandic u-umlaut and syncope, which had previously been

cited by Anderson (1974) as evidence for local ordering. Instead of allowing

u-umlaut and syncope to apply in either order, whichever produces a feeding
relationship, the Lexical Phonology approach fixes the within-stratum order
as u-umlaut precedes syncope, but then allows u-umlaut to follow syncope by
reapplying in a later stratum.
Assignment of rules to different strata offers a way of imposing extrinsic
ordering on them: if rule A applies only in stratum 1 and rule B applies only
in stratum 2, then A necessarily precedes B. Therefore, assignment of rules to
strata could be used to reproduce some of the effects of SPE-style extrinsic
ordering. This leads to some questions: Is extrinsic ordering within strata
truly necessary? Could all rules in the same stratum apply simultaneously
or in a universally predictable order? These questions were not asked, much
less answered, in mainstream work on Lexical Phonology, though they were
discussed in work that is not usually identified with the Lexical Phonology
research program (Goldsmith 1993a, Lakoff 1993). In any case, the questions
persist to this day, as we will see in §2.3.4.2.
Apart from these developments, the common consensus about rule ordering
and opacity did not change very much in the period after 1980. Most phonolo-
gists, perhaps more from a lack of interest than strong conviction, continued
to assume something like the SPE model of rule interaction.

2.3 Opacity in Optimality Theory

2.3.1 Properties of classic OT
This section is not intended as an introduction to or comprehensive overview
of OT (for the former see Kager (1999a), and for the latter see Prince and
Smolensky (2004) or McCarthy (2002b)). Rather, the goal is to review those
aspects of OT that assume particular significance in the analysis of opacity.
In OT, a grammar of a language is a ranking of constraints. Ranking differs
from language to language, so the ranking relation between any given pair of
constraints is not generally predictable. Because language-particular ranking
offers a way of accounting for language differences, it is reasonable (though not
strictly necessary) to adopt the null hypothesis that the constraints themselves
are universal and so are drawn from a universal constraint component, called
con. If con is indeed universal, as is standardly assumed in classic OT, then it
is fair to say that OT is an inherently typological theory of language. In other
words, OT and a specific hypothesis about con combine to predict all and only
the possible grammars of human languages.

In classic OT, the constraints in con are limited to two types: markedness
constraints evaluate output forms, favoring some over others; and faithful-
ness constraints evaluate input-output mappings, favoring those mappings that
maintain identity. Classic OT, in the sense employed here, also incorporates
the assumption that the faithfulness constraints are formalized in terms of
a correspondence relation between input and output forms (McCarthy and
Prince 1995, 1999). Because the substantive properties of con, particularly
the markedness constraints, are largely unknown, empirical research in OT is
mostly focused on developing a detailed picture of con. Some of this research
has led to proposals for constraint types that are neither markedness nor faithful-
ness, such as antifaithfulness (Alderete 2001a, 2001b) or morpheme realization
(Kurisu 2001), but these ideas go well beyond the limits of what I am calling
classic OT.
OT is inherently comparative. In the simplest case, the evaluative com-
ponent eval applies a language-particular constraint hierarchy to the task of
comparing two possible outputs derived from a common input. Of these two
outputs, called candidates, the more harmonic one is that which performs better
on the highest-ranking constraint on which they differ. 6 The most harmonic
or optimal candidate is the one that is more harmonic, in this sense, than any
of its competitors.
Moreton (2003) has shown that classic OT entails a requirement of harmonic
improvement. Assume that every candidate set contains at least one candidate
that is fully faithful by virtue of obeying all of the faithfulness constraints in
con. 7 If the output of an OT grammar is not this fully faithful candidate, then it
must be a candidate that is less marked than the fully faithful candidate relative
to the language’s constraint hierarchy. Moreton provides a formal proof of
this result, but the intuition behind it is also clear: since a classic OT grammar
has only markedness and faithfulness constraints, the only reason to violate a
faithfulness constraint is satisfaction of a higher-ranking markedness constraint.
Informally, you can stay the same or get better, but you can’t get worse.

2.3.2 Process interaction in classic OT

Except for digressions in chapter 2 of Prince and Smolensky (2004) and in the
appendix of McCarthy and Prince (1993b), classic OT has usually included an
assumption of parallelism. This means that the candidates under evaluation
can show the effects of several phonological processes simultaneously — that
is, the effects of processes are evaluated in parallel. Classic OT is therefore a
bistratal theory: it recognizes two levels of representation, input and output, but

nothing in between. This is obviously very different from SPE, which has nearly
as many intermediate levels of representation as there are rules (see §2.2.1).
In general, transparent interaction of processes is fully compatible with
parallelism. Consider first a feeding interaction like (2-2), where underlying
// becomes surface [], showing the effects of two processes, vowel
epenthesis and [] epenthesis. Taken separately, each process involves, inter
alia, a basic markedness-dominates-faithfulness ranking, as shown in (2-9)
and (2-10). 8 Faithful syllabification of the initial cluster in // is impossible
because of *comPlex-onset and other markedness constraints. Violation of the
lower-ranking antiepenthesis constraint DeP is the chosen alternative. Faithful
syllabification of a word-initial vowel is a breach of onset, which also ranks
above DeP. The feeding interaction between the two types of epenthesis is
simply the result of satisfying both *comPlex-onset and onset simultaneously.
Among the candidates derived from // is one in which both vowel and []
epenthesis have occurred. This candidate is favored by both of the high-ranking
constraints, as tableau (2-11) illustrates.
(2-9) *comPlex-onset >> DeP

// *comP-ons DeP

→  2

 W1 L

(2-10) onset >> DeP

/--/ onset DeP

→  1

 W1 L

(2-11) Feeding interaction

// onset *comP-ons DeP

→  2

 W1 L

 W1 L1

I noted in §2.2.4 that transparent interactions are incompatible with simul-

taneous application of phonological rules. That is because a rule’s structural
description analyzes that rule’s input, and the fed rule’s structural description
cannot be met until after the feeding rule has applied. In OT, however, the
structural descriptions of markedness constraints analyze outputs, and feeding
interactions are simply a consequence of satisfying such constraints. This is
the sense in which classic OT exhibits parallelism: high-ranking markedness
constraints can favor a candidate that differs from the input by the simultaneous
effects of two or more processes, as in (2-11).
The situation is the same with the other type of transparent interaction,
bleeding. The difference is that bleeding interactions may involve conflict
between markedness constraints. For instance, the mapping /-/ →
[] in (2-3) shows that the markedness constraint responsible for
progressive assimilation of pharyngealization is crucially dominated by two
other markedness constraints, one forbidding pharyngealization of [] and
the other ruling out medial triconsonantal clusters (and thereby demanding
[] epenthesis). In this way, the output [] is favored over alterna-
tives like *[], with pharyngealized [i], and *[], with a
triconsonantal cluster.
Parallel evaluation in classic OT also eliminates the need for certain
kinds of Duke-of-York derivations (see §2.2.3). An example comes from
Nuuchahnulth, formerly known as Nootka (Campbell 1973, Kenstowicz and
Kisseberth 1977: 171ff., McCarthy 2003c, Sapir and Swadesh 1978). 9 This
language has a process that rounds velars and uvulars when they follow round
vowels (2-12), as well as a process that unrounds velars and uvulars at the end
of a syllable (2-13). (Syllable boundaries are shown by a period/full stop.)
These two processes are in a mutual feeding relationship: when a velar or uvular
follows a round vowel, as in (2-14), rounding creates inputs to unrounding and
unrounding creates inputs to rounding. In the SPE tradition, this kind of conflict
can only be resolved by rule ordering, and indeed mutual feeding relationships
presented special challenges to those seeking to predict rule ordering (§2.2.5).
The stipulated ordering is given in (2-14). Because unrounding gets its hands
on the form later in the derivation, it states the surface-true generalization that
syllable-final consonants are unrounded. The truth of the rounding generaliza-
tion consequently suffers: there exist some nonrounded velars and uvulars that
are preceded by a round vowel.
(2-12) Rounding in Nuuchahnulth
Underlying /-/ ‘ten on top’
Rounding 
Surface [] (cf. [] ‘on top’)

(2-13) Unrounding
Underlying /-/ ‘to take pity on’
Unrounding 
Surface [] (cf. [] ‘pitiful’)
(2-14) Duke-of-York derivation
Underlying // ‘throwing off sparks’
Rounding 
Unrounding 
Surface [] (cf. [] ‘phosphorescent’)

In OT, deriving [] from // does not require passing through the
intermediate step []. Rather, this is a matter of conflict between marked-
ness constraints, and it is resolved, as are all constraint conflicts, by ranking the
conflicting constraints. In (2-15), I introduce two ad hoc markedness constraints
and show how the higher-ranking constraint is the one that favors nonround
consonants syllable-finally. Both are ranked above the faithfulness constraint
IDent(round), to account for the predictability of consonant rounding in this
context.
(2-15) *Kw]σ >> *uK >> IDent(round)

// K ]σ uK ID(round)

→ 
1

 W1 L W1

It is useful to compare the SPE-style analysis in (2-14) with the OT analysis

in (2-15). The comparison shows why parallelism is and should be the null
hypothesis for OT. In the SPE model, ordering is a way of establishing priority
relationships among rules, and in a case like Nuuchahnulth it is the last rule that
has priority in the sense that it states a surface-true generalization, even though
the earlier rule does not. In OT, priority relationships among constraints are
established by ranking them, and this example shows that ranking can replace
at least some applications of rule ordering. The null hypothesis, then, is that
OT can dispense with ordering and all of its trappings, including intermediate
derivational steps. In its place, OT has constraint ranking, which is required
independently. This very strong claim is certainly not uncontroversial, and
opacity presents the main challenge.

Before we go on to look at opacity in OT, however, it is appropriate to

examine some conceptual arguments that have been advanced against paral-
lelism and in favor of SPE-style serial derivations. One of these conceptual
arguments holds that sequential rules accurately model a system of mental
computation (Bromberger and Halle 1997). The failure of the Derivational
Theory of Complexity showed that this idea is very far off the mark, at least
in syntax (Fodor, Bever, and Garrett 1974); the same is true in phonology
(Goldsmith 1993b). Indeed, if the goal of generative grammar is to construct
competence models (Chomsky 1965), then it is a category mistake to ask
whether these models faithfully replicate mental computation.
Another argument offered in favor of sequential rule application is that
it makes sense in terms of language history (Bromberger and Halle 1989):
the ordering of synchronic rules matches the chronology of diachronic sound
changes. The principal problem with this view is that it misconceives language
change. If language learners in generation Y innovate a sound change, they do
not simply add a rule onto the end of generation X’s phonological grammar
— they cannot, since generation Y does not have direct access to generation X’s
internalized grammar. Generation Y’s learning is informed exclusively by X’s
actual productions, as filtered through Y’s perceptual system. X’s productions
offer only indirect evidence of X’s grammar, subject to well-known limitations
like the absence of negative evidence. From this perspective, we neither expect
nor do we necessarily observe that grammars change by accreting rules at the
end of the ordering.

2.3.3 Opacity in classic OT 10

Classic OT recognizes just two types of constraints, markedness and faithful-
ness, and just two levels of representation, underlying and surface. Markedness
constraints can refer to only one of those levels of representation, surface
structure. Faithfulness constraints refer to both levels, but they can only do
one thing: require identity. The standard derivational approach to opacity relies
on having intermediate levels of representation (see §2.2.3), but classic OT
has none. Furthermore, the limitation of markedness constraints to evaluating
surface structure has unwelcome consequences for the analysis of counterbleed-
ing opacity.
In counterbleeding opacity, a phonological process occurs even though the
conditioning environment is not present in surface structure. In the Bedouin
Arabic example (2-5), for instance, // palatalizes even though it is not fol-
lowed by a front vowel in the surface form: /-/ → []. In
other words, the // → [] unfaithful mapping is a response to phonological

conditions that are not visible in the output form, though they are visible in the
input. Because markedness constraints are limited to evaluating outputs, the
markedness preference for [] over [] before front vowels cannot be invoked
to explain why // is palatalized before a vowel that is no longer present.
The problem is apparent from tableau (2-16), which shows that [] is
harmonically bounded by *[].
(2-16) Counterbleeding opacity in classic OT11

/-/ iCV max ki ID(back)

→  1 1

a.  1
L

b.  W1 L 1

c.  W1 L W1 L

Row (a) in (2-16) contains no W’s and one L, so the candidate in (a) harmoni-
cally bounds the intended winner. Moreover, since this candidate is more faith-
ful and less marked than the intended winner, no other classic OT faithfulness
or markedness constraint could be introduced to break this harmonic bounding.
(On an alternative analysis with coalescence, see §2.3.4.1. For the OT-CC
analysis of palatalization, see §3.3.3, and for the analysis of syncope — minus
the ad hoc constraint *iCV — see §4.3.3.)
In this and other cases of counterbleeding opacity, an unfaithful mapping
occurs for reasons that cannot be explained with classic OT markedness con-
straints because the conditions that encourage the unfaithful mapping are no
longer apparent in surface structure. Although analyses of particular instances
of counterbleeding opacity (including this one) have been proposed, there is
no general solution that remains within the strictures of classic OT.
In contrast to counterbleeding opacity, counterfeeding opacity can in prin-
ciple be accommodated in classic OT. Consider the Bedouin Arabic example
in (2-4), in which underlying // deletes (/-/ → []) but [] derived
from // does not (// → [], *[]). As above, let *iCV stand for
the constraint that favors [] over faithful []; it dominates max.
Let *aCV stand for the constraint that favors [] over faithful []; it
dominates IDent(+low). Tableau (2-17) shows that the desired output []
is unattainable with just these four constraints. To circumvent this paradox, we
require a constraint that favors the desired winner in (2-17) over the loser in

(a). This constraint, which can be called max-A, forbids the /a/ → Ø mapping.
max-A meets the formal requirements for faithfulness constraints: it requires
identity between underlying and surface structure. Ranked above *iCV, max-A
correctly favors [], as shown in (2-18). Furthermore, max-A does not
interfere with the analysis of high vowel syncope in forms like /-/. (See
§4.3.3 for the full analysis.)
(2-17) Impossibility of [] without max-a

// aCV ID(low) iCV max

→  1 1

a.  L L W1

b.  W1 L L

(2-18) Counterfeeding opacity in classic OT

// max-A aCV iCV ID(low) max

→  1 1

a.  W1 L L W1

b.  W1 L L

In theory, this mode of analysis could be generalized to all instances of coun-

terfeeding opacity, thereby providing classic OT with a ready-made solution to
this half of the opacity problem. In practice, though, that would not be a good
idea. Dealing with the full range of counterfeeding interactions will require
a very rich faithfulness theory, undoubtedly much richer than we want or
would otherwise need. Another counterfeeding interaction in Bedouin Arabic
illustrates. Raising of // to [] in an open syllable is not fed by a process of
epenthesis that breaks up final consonant clusters: // → [], *[]
‘grave’. To analyze this phenomenon in the same manner as (2-18), we would
need a faithfulness constraint with the following definition: ‘Assign a violation
mark for every instance of a surface high vowel that stands in correspondence
with an underlying low vowel, provided that this surface high vowel is fol-
lowed in the next syllable by a vowel that has no underlying correspondent.’
In other words, the counterfeeding interaction in [] requires a version
of IDent(+low) that is applicable only if the vowel in the next syllable is

epenthetic. Constraints like this are necessarily embedded in a faithfulness

theory that makes unattested and implausible typological predictions. Rather
than demonstrate this now, I return to the matter in §2.3.4.1 when I discuss a
theory of faithfulness that countenances such constraints, local conjunction.
Classic OT has an inherent bias toward transparent interactions (§2.3.2).
Counterfeeding opacity requires undesirable enrichment of faithfulness theory,
and counterbleeding opacity is usually intractable. Since opacity appears to be
an authentic property of phonological systems, classic OT needs to be modified.
The question is how.

2.3.4 Previous approaches to opacity in classic OT

More than a few different proposals have been made about how to integrate
the analysis opacity into OT. Some are recent or short-lived; others date back
to the earliest work in the theory. For discussion purposes, they can be grouped
into four broad categories:
i) Changes in substantive properties of phonological representation or
the constraint component con (§2.3.4.1). The goal is to analyze some
or all cases of opacity by enriching representations or creating new
constraints. (The approach discussed at the end of the previous section
is an example.)
ii) Introduction of intermediate derivational stages and something like
rule ordering to OT (§2.3.4.2).
iii) Introduction of an equivalent of intermediate derivational stages, but
without any direct counterpart to rule ordering (§2.3.4.3).
iv) Reinterpretation of opacity as a mechanism for preserving underlying
contrasts (§2.3.4.4).

2.3.4.1 Opacity via novel substantive assumptions

This section describes approaches to opacity that place the main analytic burden
on assumptions about substantive matters. Three lines of attack will be dis-
cussed in turn: representational approaches, which enrich surface structure in
ways that allow opaque processes to be reanalyzed as transparent; reanalysis of
counterbleeding opacity as a type of segmental coalescence; and reanalysis of
counterfeeding opacity as a faithfulness effect using local constraint conjunc-
tion.

Representational approaches to opacity. Opacity issues arose in the very first

work on OT, Prince and Smolensky (2004). The topic comes up in the context
of two analyses, Lardil (pp. 145, 148) and Fula (p. 255).
In Lardil nominative case forms, final vowels are deleted (Hale 1973):
// → [] ‘oyster species (nominative)’ (cf. the nonfuture accusative
[-], with suffix /-/ and no truncation). When this apocope process
exposes a final consonant that is not allowed syllable-finally (Wilkinson 1988),
the consonant deletes as well: // → [] ‘termite’. Apocope
therefore feeds consonant deletion. Crucially, apocope must not be fed by
consonant deletion; if it were, then we would expect to find apocope and
consonant deletion chewing through words until a licit coda is found (as in
*[] from // ‘nullah’) or the bimoraic word minimum is
reached (as in *[] from // ‘tata-spear’). This is an example of
counterfeeding opacity.
Fula has two processes that refer to geminate consonants. One process
shortens a geminate after a long vowel, and the other hardens geminate continu-
ants into stops (Paradis 1988). They interact in counterbleeding fashion, with
an underlying geminate continuant undergoing hardening even if it is also
shortened: // → [] ‘roads’.
The analytic strategy that Prince and Smolensky apply to both of these
cases is closely connected with their implementation of faithfulness constraints.
Faithfulness is essential to OT, since without faithfulness markedness runs amok,
driving every input down to some least marked output like [] (cf. Chomsky
1995: 380fn.). The idea of faithfulness is thus a key insight without which
OT would be a failed enterprise. The implementational details are much less
central, though relevant to the analysis of opacity. The implementation adopted
in Prince and Smolensky (2004) is based on a principle dubbed Containment in
McCarthy and Prince (1993b): all of the phonological material in the underlying
representation must be preserved in every candidate output form.
Containment therefore entails that there are no literal deletion processes.
Instead, the effects of deletion are obtained from the joint action of three
additional assumptions, all with precedents elsewhere:
(i) Underlying representations lack prosodic structure, particularly syl-
labification.
(ii) Phonological material may remain unincorporated into prosodic struc-
ture.
(iii) Unincorporated phonological material receives no phonetic interpreta-
tion.

Thus, a deleted segment like the final // of [] is present in the output
form but syllabically unparsed: []σ []σ []σ , or more compactly [<>].
With a shortening process like // → [], an underlying mora is pre-
served in the output but also syllabically unparsed. Both situations violate
constraints from the Parse family, which require segments, moras, and other
structural elements to be incorporated into prosodic structure. Some theories
of syntactic deletion are a close parallel (e.g., Chomsky 1995).
Containment supplies an analytic strategy for many cases of opacity. For
example, in Prince and Smolensky’s analysis of Lardil, apocope is the result
of satisfying the constraint Free-V ‘Word-final vowels must not be parsed (in
the nominative)’ (p. 123). In [<>], the word-final vowel is unparsed,
as requested, and the preceding [] is unparsed because it is not a licit syl-
lable coda. Nonparsing of the preceding [], however, would violate Parse
for no reason — if ‘word-final’ means ‘rightmost segment, parsed or not’,
then the word-final vowel is [], and [] has no claim to word-final status.
Apocope cannot feed itself, then, because apocope can only affect a vowel
that is word-final in underlying representation. (This also explains why the last
but not word-final vowel of // does not apocopate: [<>]
‘queen-fish’.)
In Fula, we need to explain how a mapping with hardening and degemina-
tion of geminate continuants (// → []) can be more harmonic than a
mapping with degemination alone (// → *[]). Since *[] is not
pronounced with a geminate, it should not lose to the hardened former geminate
in []. Prince and Smolensky’s solution (p. 255) again relies on Containment.
The two skeletal positions linked to geminate // in // can never be literally
deleted; rather, both are present but one is syllabically unparsed in candidates
with degemination like [] and *[]. The markedness constraint against
geminate continuants defines a ‘geminate’ as a consonant linked to two skeletal
positions, regardless of whether the skeletal positions are syllabified. This
markedness constraint, then, is sensitive to the representation and not the pro-
nunciation, and the representation of *[] contains a ‘geminate’ continuant
because it is derived from // under Containment. Once a geminate, always
a geminate, as far as this constraint is concerned.
This theory of opacity requires no changes in OT proper, and that makes it
attractive. It has empirical problems of two types, however: there are observed
opaque interactions that it cannot easily accommodate; and there are transparent
interactions that ought to be opaque if this theory is right. We will examine
each in turn. (For related discussion, also see the critique of Containment in
McCarthy and Prince (1995, 1999).)

A basic prediction of the Containment model is that syllabification always

interacts transparently with processes because syllabification is present only
in the output. With syllabification though not with segmental structure, the
pronunciation and the representation are true to one another. Cases like the
Bedouin Arabic // → [] example (§2.3.3) are therefore problematic:
// raises to [] in an open syllable, and [] has an open syllable. There
is no earlier stage of syllabification to refer to opaquely, in which // is in a
closed syllable. Another example along the same general lines can be found
in Levantine Arabic (see §4.2). When a final cluster is resolved by epenthesis,
stress is assigned to the erstwhile final syllable, in conformity with the general
pattern for words ending in such ‘superheavy’ syllables: /-/ → [] ‘I
wrote’. Forms with the same surface syllable structure but without epenthesis
are stressed differently: /-/ → [] ‘she wrote’. Since syllabification
is necessarily transparent under Containment, this opaque interaction between
stress and syllabification/epenthesis is inexpressible.
The other problem is that many transparent processes, which should
be unremarkable, end up tripping over the unparsed remnants of deletion
(cf. Beckman 1997: 27–31). In Maltese, for example, there is a completely
transparent process of regressive voicing assimilation in obstruent clusters
(Borg 1997). Because it is completely transparent, this process also affects
consonant clusters that are created by syncope: /--/ → [] ‘we
write’. Under Containment, syncope does not affect string-adjacency relations
among segments because no segment is literally deleted. Therefore, voicing
assimilation affects a sequence of noncontiguous consonants: [<>]. This
is a surprising result, since voicing assimilation has never been observed to
traverse a pronounced vowel in any language. This problem could be avoided
by adopting a more sophisticated theory of locality that reckons segments as
adjacent if no parsed segment appears between them, but this move would be
inconsistent with the Containment-based analysis of Lardil, where unparsed
segments do count in determining whether a vowel is final or not.
A usual (though not essential) accompaniment to Containment is the
assumption that epenthesis is not literal segmental insertion but rather prosodic
overparsing (after Broselow 1982, Ito 1986, 1989, Lowenstamm and Kaye
1986, Piggott and Singh 1985, Selkirk 1981b and others). In overparsing, syl-
lables are created with empty structural positions. The phonetic content of these
empty positions is determined extrasystemically — that is, outside the phono-
logical grammar proper. Those positions that are devoid of segmental content
violate faithfulness constraints from the FIll family, which militate against
such mismatches between segmental and prosodic structure. An example: the
phonological output corresponding to Classical Arabic [] is [ON],

where O and N stand for an unfilled onset and nucleus, respectively. The
spell-out of O as [] and N as [] happens in some later module that interprets
the output structures derived by the OT phonological grammar.
Because the phonetic identity of epenthetic segments is supplied
extraphonologically, processes of segmental phonology should treat them
opaquely, as if they were not present, whereas syllable-sensitive processes
should treat epenthesis transparently, for reasons already given. There are indeed
some cases where epenthesis interacts opaquely with segmental phonology.
For example, Herzallah (1990: 109–110) reports for her northern Palestinian
Arabic dialect that the vowel [] causes a preceding pharyngealized // to lose its
pharyngealization: [] ~ [] ‘he classified ~ he classifies’. Epenthetic []
does not have this effect, however: // → [] ‘cutting’. This observation
is consistent with the claim that information about the quality of epenthetic
vowels is determined after the phonological grammar has done its work. On
the other hand, the example in (2-3) shows for a southern Palestinian dialect
that epenthetic [] blocks the spread of pharyngealization, acting just like non-
epenthetic [] in this respect. So epenthesis does not show consistent opaque
interaction with segmental processes. Rather, interaction may be transparent
or opaque on a language-specific basis. This is contrary to the predictions of
the FIll-based model of epenthesis.
The containment theory of faithfulness is, as we have seen, also a theory
of opacity, but not an entirely successful one. Two main problems have been
identified. Under Containment, deleted segments should be consistently visible
to processes that are conditioned purely by segmental adjacency but consist-
ently invisible to processes that are conditioned by syllable structure. This
predicts opaque interactions in the former case and transparent interactions
in the latter, but there are counterexamples to both predictions. Under the
empty-node theory of epenthesis, epenthetic segments should be consistently
invisible to processes that are conditioned by segmental adjacency but consist-
ently visible to processes that are conditioned purely by syllable structure. This
predicts opaque interactions in the former case and transparent interactions in
the latter, but again there are counterexamples to both predictions. The inherent
simplicity and consequent attractiveness of this theory of opacity yields to its
empirical inadequacies.
There is some later work exploring enhancements of this theory of opacity to
grant it greater descriptive power (Goldrick 2000, Goldrick and Smolensky 1998).
The key idea of this approach, called Turbidity, is that the symmetric is associ-
ated with relation between prosodic and segmental structure is divided into two
asymmetric relations: segments project prosodic structure, and prosodic structure
is pronounced as segments. Usually, these two relations operate in tandem, with
segment s projecting prosody p if and only if p is pronounced as s.

The hallmark of opacity in Turbidity theory is a mismatch between the

Project and Pronounce relations. Compensatory lengthening presents a typical
example. 12 Compensatory lengthening is a type of counterbleeding opacity: a
deleted segment projects a mora, but that mora is pronounced with a different
segment, thereby lengthening it. In Turkish, for example, coda // is option-
ally deleted before a continuant or nasal, in which case the preceding vowel
lengthens (Sezer 1985: 230): [] ~ [] ‘coffee’. The representation
of [] is shown in (2-19), with upward and downward arrows standing
for the Project and Pronounce relations, respectively. In this representation,
the segments [] and [] each project a mora (upward arrows), but the mora
projected by [] is pronounced as [] (diagonal downward arrow).
(2-19) Compensatory lengthening in Turbidity theory
μμ μ
↕↑ ↕

In Turbidity theory, markedness constraints are defined in terms of the Project
and Pronounce relations. One constraint requires coda consonants, such as []
in (2-19), to project a mora. This constraint is indifferent to whether the [] is
pronounced with its projected mora. Another markedness constraint requires
that every mora be pronounced with some segment. This constraint is indiffer-
ent to whether the mora is pronounced with the segment that projects it. Though
such mismatches between Project and Pronounce are possible — and (2-19) is
an example — they are marked, violating a constraint called recIProcIty. This,
in outline, is how opaque analyses are constructed in this theory.
What would it take to extend Turbidity theory to deal with the full range
of opaque interactions? Very likely, it will require two coexistent phonological
representations, the pronounced one and the projected one. These two repre-
sentations are folded together in (2-19), but (2-19) is not representative of the
full range of opaque interactions. An instructive example is Bedouin Arabic
// → [], where // is not raised in a derived open syllable. In Turbidity
terms, this means that [] must be represented with two coexistent syllabic
parses, one where [] projects as the coda of the syllable [] and one where
it is pronounced as the onset of the syllable []. Phenomena like Palestinian
Arabic // → [] require extending the Project/Pronounce distinc-
tion to linear order relations among segments. Epenthetic [] is pronounced
as the successor to [] in the segmental string, but [] is projected as []’s
successor. In short, there can be Project/Pronounce mismatches in all of the
ways that phonological elements relate to one another. This means that there
are two complete phonological representations, with two sets of markedness

constraints. recIProcIty maintains a check on divergence between the two

representations, and violation of recIProcIty is the source of opacity.
Looked at in this way, Turbidity has much in common with those theories of
opacity that posit a single additional level of representation besides underlying
and surface structure. It also shares some of the limitations of these theories.
We will examine those limitations in §2.3.4.2.

Segmental coalescence. Segmental coalescence may sound like a peculiar

theory of opacity, but it plays such a role in much of the OT literature. It
has been applied to one rather common form of counterbleeding opacity, in
which a segment is observed to assimilate to another nearby segment that has
deleted. The palatalization/syncope interaction in (2-5) is typical: /-/
→ []. Although this derivation is opaque under the assumption that
palatalization is the result of assimilation, it can be analyzed as transparent
under the assumption that palatalization and syncope are united into a single
process of segmental coalescence. The underlying /12/ sequence fuses into
the single output segment [1,2]. There is no literal deletion, no failure of
input-output correspondence, so max is satisfied. The resulting segment is
palatalized because [1,2] is faithful to the color features of one of its underlying
correspondents, /2/. Another alternative: when // deletes, it leaves behind the
feature specification [–back], which reassociates autosegmentally to the preced-
ing //. In this case, although max is violated, the feature-specific constraint
max(–back) is not. Analyses along these general lines can be found in Causley
(1997), Gnanadesikan (1997, 2004), Lamontagne and Rice (1995), McCarthy
and Prince (1995), and Pater (1996), among others.
These alternatives to opacity have their merits, but they also have their prob-
lems. A parochial concern is that palatalization in Bedouin Arabic is not limited
to deleted //. Overt front vowels also cause palatalization, so palatalization must
not be inextricably linked with deletion of the triggering segment, as both the
coalescence and autosegmental analyses imply. A broader worry is that observed
counterbleeding interactions are not conveniently limited to situations that
can plausibly be regarded as coalescence. For example, Donegan and Stampe
(1979: 153) point out that there is a counterbleeding interaction in English
between intervocalic //-flapping and optional desyllabification of prevocalic
liquids: // →-flapping [] →optional [] shattering. There is no way
of reanalyzing this opaque interaction as a single coalescence process. Another
example, this time from Kenstowicz and Kisseberth (1979: 292–294): in Tunica
(Gulf, Louisiana), a sequence /V1/ is altered by assimilating // to the color
of /V1/ and by deleting /V1/ if it is unstressed. The result is a counterbleeding
interaction in cases like /-/ → [] ‘she dances’, with // assimilat-

ing to the deleted vowel. The deletion + assimilation combination is unlikely to

be reducible to a single process of coalescence for three reasons: (i) Deletion
occurs independently of assimilation in cases like /-/ → []
‘he dances’; (ii) Assimilation occurs independently of deletion in cases like
/-/ → [] ‘she takes’; and (iii) Coalescence of nonadjacent seg-
ments is probably unattested and very likely impossible (see §3.2.4.3).

Local constraint conjunction. As I noted in §2.3.3, counterfeeding opacity

can be accommodated in classic OT if the theory of faithfulness constraints is
sufficiently rich. (There is no comparable way of dealing with counterbleeding
opacity.) It has been proposed that local conjunction of faithfulness constraints
is the proper mechanism for incorporating this richer theory of faithfulness into
con (Ito and Mester 2003c, Kirchner 1996, Moreton and Smolensky 2002).
Local constraint conjunction is proposed by Smolensky (1995) as a theory
of the internal structure of con. Complex constraints are built by conjoining
simpler constraints. (The simpler constraints may be irreducible, or they may
themselves be the product of local conjunction.) The local conjunction of
constraints A and B, [A&B]δ, is defined as a constraint that is violated once for
each instance of the domain δ in which both A and B are violated. Conjunction
of markedness constraints supplies the most persuasive examples. Codas are
marked by the constraint no-coDa and voiced obstruents are marked by the
constraint no-vcD-obst. The local conjunction of these constraints within
the domain of a segment, [no-coDa & no-vcD-obst]Seg, militates against
the combination of these two marked properties, a voiced obstruent in coda
position. In general, local conjunction of markedness constraints forbids the
cooccurrence of marked structures in near proximity to one another.
In counterfeeding opacity, unfaithful mappings cannot occur in close proxim-
ity to one another. For instance, the counterfeeding interaction in Bedouin Arabic
// → [] requires the local conjunction of IDent(low) and DeP in the domain
of adjacent syllables: [IDent(low)&DeP]Adj-σ. By ranking [IDent(low)&DeP]Adj-σ
above the markedness constraint responsible for the open-syllable raising process
(*aCV in (2-20)), we ensure that opaque [], which satisfies this constraint,
is more harmonic than transparent *[], which violates it.
(2-20) Counterfeeding opacity with local conjunction13

// [ID(low)&DeP]Adj-σ comP-coDa aCV DeP ID(low)

→  1 1

a.  W1 L 1
W1

b.  W1 L L

This is a particularly elegant theory of counterfeeding opacity, but it cannot

account for the full range of opacity phenomena, and it predicts a kind of
pseudo-opacity that does not seem to exist (McCarthy 1999: 365–366, 2002a,
2003a, Padgett 2002). The reasons for both of these problems go right to the
core of the local-conjunction theory: real counterfeeding opacity is a matter of
forbidden process interaction, but local conjunction regulates process proxim-
ity. Interaction and proximity are two very different things, and it is a mistake
to confound them.
The Bedouin Arabic example illustrates this mistake. The [] example
shows that raising is blocked when the epenthetic vowel follows the syllable
with the potentially raised vowel. On the other hand, (2-21) shows that raising
is not blocked when the (italicized) epenthetic vowel precedes the (boldface)
raised vowel. The conjoined constraint [IDent(low)&DeP]Adj-σ is unable to make
this distinction, since it forbids raising and epenthesis in adjacent syllables,
regardless of their linear order.
(2-21) Adjacent epenthesis and raising in Bedouin Arabic
/ -/ [] ‘he pursued his sheep’
This is not a mere technical glitch, to be solved with a more sophisticated
theory of the domains of conjunction. Rather, it is a basic failure of principle.
It is not an accident that raising is prohibited before an epenthetic vowel but
allowed after one. When the epenthetic vowel follows, epenthesis interacts with
raising, since following epenthesis puts the potentially raised vowel into an
open syllable. When the epenthetic vowel precedes, however, epenthesis does
not interact with raising, since preceding epenthesis has no effect on whether
the potentially raised vowel is in an open syllable. Local conjunction uses
proximity — the adjacent-syllables domain — as a proxy for interaction, and
interaction is the real basis for opacity. Because phonological processes are
usually locally conditioned, proximity is often successful as a proxy for opacity,
but examples like this one decouple the effects of proximity and interaction,
showing that interaction, not proximity, is what really matters.
Another way of grasping the problem with local conjunction’s proximity
= interaction equation is to look at the effects of locally conjoining faith-
fulness constraints in inappropriate domains. For example, the constraint
[IDent(low)&DeP]Wd — identical to Bedouin Arabic, except that the domain
is larger — will block raising if a vowel has been epenthesized anywhere in
the same word. If Bedouin Arabic were to have such a constraint, underlying
/--/ ‘I heard you (masculine singular)’ would map to [] instead
of the expected []. No known language exhibits this sort of hyperopac-
ity, in which counterfeeding behavior is extended from a local, interacting

context to a distant, noninteracting context. Yet the local conjunction theory of

counterfeeding opacity would seem to predict exactly this, since the domain of
conjunction is stipulated independently of the conjoined constraint (Alderete
1997, Ito and Mester 2003a: 105ff.).
Similar problems arise when inappropriate constraint combinations are
assembled by local conjunction. Imagine a language that is identical to Bedouin
Arabic except that it also has final devoicing of obstruents. The conjoined
constraint [IDent(low)&IDent(voice)]Adj-σ could block raising whenever an
adjoining syllable contains a devoiced obstruent: // → []. This sort of
hyperopacity is never attested — that is, we never find that one process blocks
another if the two processes by their very nature cannot interact.
There have been efforts to impose restrictions on local conjunction to
address some of these problems (Bakovic 1999, Fukazawa and Miglio 1998,
Hewitt and Crowhurst 1996, Ito and Mester 2003a: 102ff., 2003c, Łubowicz
2002, 2006). Typically, these proposals rely on the shared formal properties of
two constraints to determine whether they are conjoinable or, if conjoined, what
their domain is. None of these proposals has been fully successful in addressing
the problems described here and elsewhere (McCarthy 1999, 2002a, 2003a,
Padgett 2002). The reason for this failure is not far to seek: counterfeeding
opacity is a matter of forbidden process interaction, and process interaction is
not something that can be determined solely by looking at the formal properties
of faithfulness constraints. 14 Whether and under what conditions two processes
will interact is something that depends on the circumstances that obtain in a
particular language. We require a theory of opacity that is sensitive to these
circumstances. Rule ordering is an example of such a theory, but others will
be discussed here and in later chapters.

2.3.4.2 Analogues to serial derivations and rule ordering

Since rule-based phonology uses serial derivations to account for opacity, it
is natural to ask whether derivations and the effects of rule-ordering can be
reconstructed in OT, which is a theory without rules. A multi-step serial deriva-
tion can be obtained simply by assuming that the output of an OT grammar is
not the surface form but instead is the input to another OT grammar. Actual
implementations differ in whether or not the second grammar is the same as
the first one. The approaches to be discussed are: single-grammar serial OT,
which is known as harmonic serialism; multi-grammar serial OT, which is
sometimes known as Stratal OT; and output-output faithfulness, often referred
to as OO correspondence.

Harmonic serialism. In harmonic serialism, the output of an OT grammar is

returned as the input to that same grammar (McCarthy 2000a, 2002b: 159–163,
2007a, Prince and Smolensky 2004: 6–7, 94–95). This process continues until
‘convergence’, when the output of a pass through the grammar is identical to
the output of the previous pass. (Convergence in a finite number of passes is
guaranteed for reasons discussed by Moreton (2003).)
Harmonic serialism in its simplest form turns out to be surprisingly inef-
fective in dealing with opacity. It is no better off than classic OT in dealing
with counterbleeding opacity. In (2-16), we saw that classic OT stumbles on a
case of counterbleeding opacity like /-/ → [] because there
is no visible motive in surface structure for palatalization of the //. Harmonic
serialism does no better. On the first pass through the grammar, there is noth-
ing to prevent the transparent mapping /-/ → *[], just like
(2-16). In general, wherever classic OT has a problem with counterbleeding
opacity, harmonic serialism will too, since harmonic serialism is just classic
OT, iterated.
Harmonic serialism actually does worse than classic OT on some kinds of
counterfeeding opacity. Recall from (2-18) that classic OT can accommodate
counterfeeding interactions by positing the right faithfulness constraints. In
Bedouin Arabic, because // deletes in the same environment where // changes
to [], what is needed is a constraint that specifically militates against delet-
ing //, max-A. But max-A is useless under the harmonic serialism regime.
The first pass through the grammar, which is shown in tableau (2-22), maps
// to []. The output of the first pass becomes the input to the second
pass, shown in tableau (2-23), and [] qua pass-two input is mapped to
*[]. On the third pass, [] qua pass-three input maps to itself, and there
is convergence — on the wrong output.
(2-22) Harmonic serialism: first pass through grammar

// max-A aCV iCV ID(low) max

→  1 1

a.  W1 L L W1

b.  W1 L L

(2-23) Harmonic serialism: second pass through grammar

] max-A aCV iCV ID(low) max

→  1

a.  W1 L

b.  W1 W1 L

The problem in (2-23) is this: with [] as the input, max-a no longer
protects the vowel in the first syllable from deletion. When the second and
subsequent passes through the grammar come around, information about the
original input is no longer available to eval. For this reason, counterfeeding
opacity in general cannot be analyzed using harmonic serialism. (See Norton
2003: 247ff. for related discussion.)
These failures of harmonic serialism show that a single-grammar imple-
mentation of serial OT is of no value in analyzing opacity. We will see in
§3.2.3, however, that harmonic serialism has some significant connections
with OT-CC.

Multi-grammar serial OT. The principal thesis of the theory of Lexical

Phonology is that the phonological system of a language consists of a series
of separate modules, called levels or strata, each of which is an SPE grammar in
its own right (see §2.2.6). Strata are usually associated with different morpho-
logical subsystems in the lexicon or with the difference between word-internal
and phrasal phonology. There is an ordering among the strata, and the output
of one stratum is the input to the next. The output of the last or postlexical
stratum is the actual surface form.
It seems like a small step to go from assuming that strata are SPE grammars
to assuming that they are OT grammars, and so this move has been advocated
almost since the beginning of OT. Although implementational details differ,
this idea of linking OT grammars serially is common to all of the approaches
that go under names like LP/OT, Derivational OT, or Stratal OT. Throughout,
I will use the name Stratal OT to refer to any theory that incorporates these
basic assumptions. 15
Stratal OT uses the ordering of strata to reproduce the effects of opaque
ordering in rule-based phonology. If rule A precedes rule B in counterfeeding
order in a rule-based analysis, then the Stratal OT reanalysis posits two strata.
The grammar of the first stratum effects mappings equivalent to rule A, and
the grammar of the second stratum effects mappings equivalent to rule B. The

output of the first stratum is the input to the second stratum just as the output
of rule A is the input to rule B. Because of the assumed correlation between
strata and morphological subsystems, the Stratal OT hypothesis about opacity
is somewhat stronger than the rule-based hypothesis, which establishes no
linkage between morphology and opacity.
Each stratum is an OT grammar, so within-stratum interactions are neces-
sarily transparent just as they are in classic OT. The different strata are moreover
different OT grammars from one another — that is, they are different permuta-
tions of the universal constraint set con. This assumption is essential to Stratal
OT’s theory of opacity. Without it, Stratal OT would be another version of
harmonic serialism, and we have already seen that harmonic serialism is a
failed theory of opacity.
Stratal OT’s central analytic strategy for opacity, then, is to isolate the
opaquely interacting processes into different strata, with the ordering of the
strata supplying the counterbleeding or counterfeeding order of the processes.
For instance, the counterbleeding order between Bedouin Arabic palatalization
and syncope in /-/ →palatalization [-] →syncope [] shows
that the stratum where palatalization occurs must be ordered before the stratum
where syncope occurs. Because strata correlate with morphological subsystems
or the lexical/postlexical distinction, it will sometimes be possible to use other
evidence to determine exactly which strata are involved. Since syncope occurs
in phrases as well as words (see (2-7)), it must occur in the postlexical stratum.
For palatalization to precede syncope in counterfeeding order, palatalization
must occur in some earlier, therefore lexical stratum. The OT grammar of the
lexical stratum, shown in (2-24), takes the input /-/ and maps it to
[], with palatalization but no syncope. The grammar of the postlexical
stratum in (2-25) then takes [] as input and maps it to [], with
syncope. Observe that the two strata are inconsistent in how they rank max
and *iCV; they are, in every sense, different OT grammars.
(2-24) Lexical stratum

/-/ max ki iCV ID(back)

→  1 1

a.  W1 L L

b.  W1 L 1

c.  W1 1
L

(2-25) Postlexical stratum

// iCV ki max ID(back)

→  1

a.  1
W1

b.  W1 L

c.  W1 W1 L W1

There are two main problems with Stratal OT as a theory of opacity. First,
Stratal OT is not powerful enough to deal with the full range of observed
opaque interactions. Second, Stratal OT is also too powerful, since it massively
overpredicts phonological systems that are never observed and seem impos-
sible. There is, then, a two-way mismatch between the predictions of Stratal
OT and the typology of known opaque interactions.
The argument that Stratal OT has insufficient power was foreshadowed at
the end of §2.2.6. Like Stratal OT, rule-based Lexical Phonology allows for
the possibility of between-stratum opaque orderings. But since each rule-based
Lexical Phonology stratum is an SPE grammar, within-stratum opaque ordering
is also possible. In general, the Lexical Phonology research program never
sought to eliminate within-stratum rule ordering, including opaque ordering. In
light of the extensive pre-Lexical Phonology literature arguing for the elimina-
tion of extrinsic ordering, this failure to pursue an obvious hypothesis might
seem surprising, at least until one realizes the reason for it: 16 the hypothesis
was self-evidently wrong. That is, research on rule-based Lexical Phonology
never progressed in the direction of eliminating within-stratum opaque ordering
because there was no shortage of Lexical Phonology analyses that crucially
relied on such ordering, such as Kiparsky’s (1984) analysis of Icelandic or
Kiparsky’s (1985) analyses of Catalan and Russian.
In Catalan, for example, there is a counterbleeding relationship between
nasal place assimilation and final cluster simplification. According to Kiparsky,
cluster simplification must be assigned to the lexical stratum because clusters
cannot be rescued by postlexical resyllabification before vowel-initial words:
pont antic ‘old bridge’ is pronounced as [] and not *[].
Since nasal place assimilation precedes cluster simplification, as shown in
(2–26), nasal place assimilation must also apply in the lexical stratum. The
result in this case, as in so many other Lexical Phonology analyses, is a within-
stratum opaque (counterbleeding) order.

(2-26) Counterbleeding order in Catalan (Kiparsky 1985:96–97)

Underlying /n-/
Place assimilation []
Cluster simplification []
Surface []
‘I sell’

Bedouin Arabic supplies another example of the insufficiency of between-

stratum ordering as a theory of opaque rule ordering. (See §4.3.3 for details.)
In a rule-based analysis, deletion of high vowels must precede raising of low
vowels in counterfeeding order: // →deletion DNA17 →raising [] (see (2-4)).
We know from (2-7) that deletion is a process of the phrasal phonology, so it
must occur as late as the postlexical stratum. Raising, on the other hand, is not
a phrasal process — it only applies within words and never when its open-
syllable context arises by resyllabification across word juncture. Therefore,
raising cannot occur later than the last lexical stratum. Since raising is lexical
and deletion is postlexical, these processes are intrinsically ordered by virtue
of their stratal assignments, and so raising must precede deletion. But this is
exactly the wrong conclusion, since it puts them in feeding order rather than
counterfeeding order. Because raising is lexical, the lexical stratum maps //
to [], and because deletion is postlexical, the postlexical stratum goes on
to map [] to *[] (just as it maps // ‘was pushed’ to []).
There is more to be said about this example. As I showed in (2-18), classic
OT can analyze this counterfeeding interaction if it has a constraint max-
A. This constraint prevents deletion of any underlying //, even if its surface
realization is something other than []. max-A is of no help in the stratal
account, however. The problem is that the lexical stratum output [] is the
postlexical stratum input, and so the postlexical phonology sees an input [] in
the first syllable of this word. max-A does not protect input []s from deletion.
In Stratal OT, faithfulness constraints are local to each stratum: they require
identity between that stratum’s input and its output, and they have no way of
accessing the original underlying representation //. The information that
the first vowel of [] is an erstwhile // has been lost irretrievably by the
time the postlexical stratum comes along, and so neither max-A nor any other
constraint can account for the counterfeeding interaction between these two
processes whose stratal assignments place them in feeding order. With respect
to this example, then, Stratal OT is actually worse off than classic OT.
The Catalan and Arabic examples reveal some general properties of theories
that seek to reduce opaque interactions to between-stratum orderings. If rule A
precedes rule B in counterbleeding order, then A must apply on some stratum
that is earlier than the stratum where B first applies. A may continue to apply

on later strata, but A’s earliest application must precede B’s earliest application.
If rule A precedes rule B in counterfeeding order, then A must apply on some
stratum that is earlier than the stratum where B first applies, and A must not
apply on B’s earliest stratum or any subsequent stratum. These entailments of
Stratal OT tell us what situations would constitute prima facie counterexamples
to this theory of opacity, such as counterbleeding order with B in the earliest
stratum or counterfeeding order with A in the last stratum. A specific prediction:
postlexical processes like syncope in Bedouin Arabic are never opaque. See
§4.3 for various demonstrations that this process is indeed opaque.
From the examples discussed, it appears that Stratal OT’s premises are
insufficient to account for the full range of observed opaque interactions (see
also Noyer 1997: 515, Paradis 1997: 542, Roca 1997b: 14ff., Rubach 1997:
578 for similar remarks). The literature in support of Stratal OT and its variants
has mostly focused on exhibiting between-stratum opaque interactions and
arguing against other approaches to opacity in OT, such as sympathy theory
(§2.3.4.3). I am not aware of comparable work arguing that Stratal OT is
sufficient to account for the full range of observed opaque interactions. The
evidence described here and in the Lexical Phonology literature challenges
this claim.
Stratal OT is also an overly powerful theory because it imposes no limits
on differences among strata within a single language. There is a profound but
mostly unacknowledged difference between rule-based Lexical Phonology
and Stratal OT on exactly this point. Each Lexical Phonology stratum is an
SPE grammar and each Stratal OT stratum is an OT grammar. This seeming
parallelism is misleading, however, because the literature on rule-based Lexical
Phonology was highly attentive to the problem of constraining between-stratum
differences. There are serious and well-argued (though not uncontroversial) pro-
posals about how to do this. The earliest proposals took the form of principles
for separating lexical and postlexical processes (e.g., Kaisse and Hargus 1993a:
16–17, Kiparsky 1983, Mohanan 1982): lexical rules are structure-preserving
(i.e., neutralizing or nonallophonic); lexical rules are word-bounded; lexical
rules apply only in derived environments; lexical rules apply only to the lexical
categories noun, verb, and adjective; only lexical rules may have exceptions;
only lexical rules are sensitive to word-internal morphological structure; and
lexical rules are categorical, never gradient. This body of work culminated
in the Strong Domain Hypothesis (Borowsky 1986, Kiparsky 1984, Myers
1991b, Selkirk 1982b): all strata, lexical and postlexical, share a single SPE-
type grammar. The observed differences between strata are obtained from a
combination of universal principles like structure preservation, which can
prevent some rules from applying in lexical strata, and language-particular

stipulations about when certain rules stop applying. An approach like this is
clearly far more restrictive than the original Lexical Phonology thesis that each
stratum is a separate SPE grammar.
This restrictive version of Lexical Phonology cannot be reconstructed in
Stratal OT, however. Structure preservation, for example, is the cornerstone of
the Strong Domain Hypothesis, but there is no hope of developing an analogue
to structure preservation in Stratal OT. The principle of structure preservation
says that rule application in lexical strata cannot create segments or structures
that are not already present in underlying representations. In other words, the
well-formedness conditions on underlying representations persist as conditions
on rule application throughout the lexical strata, although they may be relaxed
or turned off in the postlexical stratum.
Structure preservation has no OT analogue for two reasons:
First, the hypothesis that grammars differ only in constraint ranking entails
that there can be no language-particular conditions on underlying repre-
sentation (McCarthy 2002b: 70–71, Prince and Smolensky 2004). This
requirement is called richness of the base (ROTB) (see also §3.5.2). Under
ROTB, the grammar itself, unaided by restrictions on its inputs, is respon-
sible for observed phonotactic patterns. Since there are no restrictions on
inputs, it would make no sense to speak of such restrictions persisting in
their effects through the lexical strata.
Second, OT offers no way of reconstructing rule-based Lexical Phonology’s
notion that some lexical constraints are turned off in later strata. The
naïve supposition is that turning-off effects can be simulated by demoting
markedness constraints or promoting faithfulness constraints. In reality,
though, OT offers no simple equivalence between demotion or promotion
and deactivation. Even low-ranking markedness constraints may be active
in situations where the faithfulness constraints ranked above them are not
relevant. Thus, the specific effects of markedness demotion or faithfulness
promotion cannot be predicted without meticulous examination of the
entire constraint hierarchy and array of inputs. Prince and Smolensky
(2004: 27ff.) emphasize this point for faithfulness constraints; reduplica-
tive emergence of the unmarked illustrates the same point for markedness
constraints (Alderete et al. 1999, McCarthy and Prince 1994). Known
conditions of literal deactivation of a constraint, such as Panini’s Theorem
(Prince and Smolensky 2004: 97–99), have such specific conditions that
they are of little value in characterizing permitted differences between
strata.

It follows, then, that the restrictive theory of differences between strata that was
developed in rule-based Lexical Phonology does not and presumably cannot
inform our understanding of such differences in Stratal OT. This problem is
not unknown to proponents of Stratal OT, and they have attempted to develop
simple principles for relating the constraint hierarchies of different strata
within a language. An example: Kiparsky (1997: 17) proposes that the ranking
of markedness constraints is constant across all of the strata of a language,
so between-stratum differences are limited to promotions and demotions of
faithfulness constraints. Another example: Koontz-Garboden (2003), citing a
personal communication from Kiparsky, proposes that between-stratum rerank-
ing is limited to promoting constraints to undominated status in later strata
(that is, stratum n+1 is identical to stratum n except that some lower-ranking
constraint(s) in n are undominated in n+1). It is not hard to find counterexamples
to these hypotheses in the Stratal OT literature. For example, Ito and Mester
(2001: 274–276) argue that the lexical and postlexical strata in German differ
in markedness ranking, contrary to the first hypothesis. The second hypothesis
is inconsistent with Kiparsky’s (2003) analysis of syncope in colloquial Arabic.
In that analysis, syncope is the result of satisfying a constraint against light
syllables. This constraint is promoted to a higher rank in the word stratum than
in the earlier, stem stratum. But this promotion cannot be to undominated status;
the constraint against light syllables must be dominated since the language has
some light syllables that escape the effects of syncope. If it were undominated,
then the language could have no light syllables whatsoever.
In summary, Stratal OT has not and probably cannot recapture Lexical
Phonology’s restrictive theory of between-stratum differences, nor does it yet
have a workable substitute. Absent such restrictions, Stratal OT allows the
strata of a single language to differ by as much as one language differs from
another. A stratum is just a ranking of con, with no obligations to the rankings
of con in other strata of the same language. From the perspective of language
typology and learnability, this is an unwelcome conclusion.
Output-output faithfulness. The theory of output-output correspondence
posits faithfulness relations among morphologically related output forms
(Benua 1997, Kenstowicz 1996a, Pater 2000, and many others). Information
flows via faithfulness constraints from an output form called the ‘base’ to
other forms derived from it by affixation. OO correspondence can be applied
to phonological opacity, as in Kager (1999b). If the base transparently under-
goes or fails to undergo the potentially opaque process, then OO faithful-
ness constraints can transmit this information by compelling related forms to
resemble the base. For example, the opaquely palatalized velar in Bedouin
Arabic /-/ → [] could be explained with reference to the
unaffixed singular base form [], where palatalization is transparent.

This is accomplished formally by deploying the output-output faithfulness

constraint OO-IDent(back), ranking it higher than its input-output counterpart
IO-IDent(back) (which is referred to as just IDent(back) in tableau (2-16)). In
this way, the phonologically unremarkable velar palatalization in [] is
transmitted to the rest of the paradigm, even to forms where the triggering front
vowel is absent from surface structure.
OO faithfulness does not suffice as a theory of opacity, however. There
are three main arguments against it (also see Benua 1997, Booij 1996, 1997,
Ito and Mester 1997b, Karvonen and Sherman [Ussishkin] 1998, McCarthy
1999: 385–387, Noyer 1997, Paradis 1997, Rubach 1997).
First, it is impossible to use OO faithfulness as a comprehensive theory
of opacity and also have a principled theory of what can be the base of an OO
correspondence relation. The [] example is attractive because the
morphologically basic form is the one where the process is transparent and the
forms derived from it are the ones where the process is opaque. In the Bedouin
Arabic counterfeeding case // → [], however, there is no word that is
morphologically more basic than [] ‘grave’. The only paradigm members
where the lack of raising can be explained transparently are the derived forms,
such as [] ‘my grave’. Clearly, it is unreasonable to insist that []
‘grave’ is derived from [] ‘my grave’, but that is exactly what would be
required to account for []’s unraised vowel using OO faithfulness.
Another argument against OO faithfulness as a theory of opacity is the
existence of cases where the (non-)application of a process is transparent in
no member of the paradigm. The analysis of Tiberian Hebrew epenthesis in
McCarthy (1999) is an example. Alternations like those in (2-27) show that
surface [] is derived from an underlying form with a final glottal stop,
//. In a rule-based analysis (Malone 1993: 59–60, 93–94, Prince 1975:
37ff.), this mapping is the result of three processes applied in the opaque order
shown in (2-28). 18 Epenthesis renders stress opaque, and deletion of [] renders
epenthesis opaque.
(2-27) Alternations of underlying //
//  ‘a wonder’ (Exodus 15, verse 11)
/-/  ‘your wonder’ (Psalms 89, verse 6)
/-/  ‘wonders’ (Lamentations 1, verse 9)
(2-28) Tiberian Hebrew // → [] derivation
Underlying //
Stress final closed syllable 
Epenthesis in final cluster 
Deletion of final [] 
Surface []

To analyze opaque stress and epenthesis using OO faithfulness, at a minimum

we would need to find paradigm members where stress on [] and epenthesis
between [] and [] are occurring transparently. There are none. The rest of
the paradigm consists of words with vowel-initial suffixes. Because of these
suffixes, stress is never retracted as far as [] and epenthesis is unnecessary.
Therefore, neither of these opaque phenomena can be obtained with OO cor-
respondence constraints.
The same problem for OO faithfulness — transparency nowhere in the
paradigm — arises whenever an underlying phonological contrast undergoes
absolute neutralization. For example, the underlying pharyngeal // in Maltese
appears to condition a number of phonological processes, though it is always
deleted at the surface (Borg 1997, Brame 1972). One such process lowers vowels
next to pharyngeal consonants: // → [] ‘I wipe’. This process is
conditioned opaquely by the deleted //: // → [] ‘I hear’. Nowhere
in the paradigm of // or, indeed, any other word of standard Maltese is the
// preserved on the surface, to condition lowering transparently.
The Hebrew and Maltese critiques of OO faithfulness as a theory of
opacity apply with equal force to approaches based on paradigm uniformity
(see Downing, Hall, and Raffelsiefen (eds), (2005)). Paradigm uniformity
allows information to flow in any direction among paradigm members, so a
morphologically complex form can affect a simple form. This greater freedom
is useless, however, in analyzing opacity when no member of the paradigm
meets the transparency requirement.
A final argument against OO faithfulness as a theory of opacity is the
existence of cases where OO faithfulness overpredicts opaque behavior. In
Levantine Arabic (see §4.2), stress and epenthesis interact opaquely, leading
to surface contrasts like [] ‘she wrote’ (from /-/) vs. [] ‘I
wrote’ (from /-/). Stress is assigned transparently in /-/ → [],
but stress is assigned opaquely in /-/ → [], as if the epenthetic
vowel were not present. To account for the opaque stress of [] ‘I wrote’
in OO faithfulness terms, we would need to explain why this form is taking
its cues, stress-wise, from paradigm members like [] ‘we wrote’ and
not from [] ‘she wrote’ or [] ‘he wrote’. Furthermore, we would
need to explain why a high-ranking OO faithfulness or paradigm uniform-
ity constraint affects only [], the form that just happens to contain an
epenthetic vowel. Any OO faithfulness constraint that would affect []
would surely resist all stress alternations throughout the paradigm, so we would
expect consistent stress on the second syllable: *[] for ‘she wrote’,
*[] for ‘he wrote’, and so on. An OO faithfulness analysis of these facts
seems quite hopeless.

OO faithfulness’s inadequacy as a theory of opacity is not entirely unex-

pected. OO faithfulness is a reasonable theory of phonological similarity among
morphologically related forms, but this is a far cry from opacity’s hidden
generalizations.

2.3.4.3 Analogues to intermediate derivational forms

Sympathy (McCarthy 1999, 2003c), targeted constraints (Wilson 2000), enriched
inputs (Sprouse 1997, 1998), and comparative markedness (McCarthy 2003a,
2003d) are four theories of opacity in OT that share a commitment to using a
third form, neither input nor output, in candidate evaluation. Since sympathy
theory has been examined more extensively than the other approaches, the
discussion here will focus on it exclusively. 19
In sympathy theory, the third form is called the sympathetic candidate. The
sympathetic candidate is just that, a candidate, so it is a kind of output form,
though different from the actual output. The sympathetic form differs from the
actual output by virtue of satisfying some faithfulness constraint that the actual
output violates. This faithfulness constraint is called the selector. Apart from
obeying the selector, the sympathetic candidate is as harmonic as possible; it
is, in short, the most harmonic candidate among those that obey the selector.
For example, in the Bedouin Arabic palatalization/syncope interaction (2-5),
the selector constraint is max, so the sympathetic candidate does not have
syncope, though the actual output does. But because the sympathetic candidate
is maximally harmonic in all other respects, it shows the effects of all of the
other (transparent) phonology of the language. Therefore, the sympathetic
candidate from input /-/ is [], without syncope but with
palatalization, since palatalization of velars is required before front vowels.
The sympathetic candidate influences the choice of the actual output form
by way of sympathy constraints. Sympathy constraints look like faithfulness
constraints, but they evaluate resemblance to the sympathetic candidate rather
than resemblance to the input. With the right ranking, as shown in (2-29), the
sympathy constraint favors opaque [], whose palatalization matches
sympathetic [], over the transparent form [], which has no
palatalization. Two candidates in (2-29) obey the selector constraint max, (b)
and (c). Of these, (b) is more harmonic by virtue of satisfying *ki, so it is the
sympathetic candidate. (To avoid circularity, the sympathy constraint itself
must be ignored in determining the sympathetic candidate.) The sympathy
constraint IDent(back)Sym is satisfied by candidates that match the palatalization
in the sympathetic candidate, thereby ruling out (a). Recall from (2-16) that
(a) harmonically bounds the intended winner in classic OT. The sympathy
constraint breaks this harmonic bounding.

(2-29) Counterbleeding opacity with sympathy

ID(back)Sym max
/-/ *iCV *ki ID(back)
(sympathy) (selector)

→  1 1

a.  W1 1
L


b. (sympathetic W1 L 1
cand.)

c.  W1 W1 W1 L L

The sympathetic candidate [] is identical with the intermediate stage

of the serial derivation (2-5). This is no accident. The changes that sympathy
theory requires in classic OT are not so different from the changes that are
required in a derivational approach like Stratal OT. The selection of the sym-
pathetic candidate requires a separate harmonic evaluation in which a single
faithfulness constraint, the selector, is promoted to undominated status. Except
for the more limited reranking possibilities, this is not unlike the grammars of
different strata in Stratal OT. Furthermore, there is a fundamental asymmetry
between the sympathetic candidate and real output candidates: the sympathetic
candidate influences the choice of the actual output, but the actual output is not
allowed to influence the choice of the sympathetic candidate. This asymmetry
is a necessary property of serial derivations: the early stages of the derivation
influence the later stages, and not vice-versa.
Various objections have been raised against sympathy theory (Bye 2001,
Idsardi 1997, Ito and Mester 2001, Kiparsky 2000, McMahon 2000), but the
biggest problem may be the analysis of multiple interacting opaque proc-
esses. The most famous example of multiple interaction is Yawelmani Yokuts.
(References on this language include Archangeli 1985, Archangeli and Suzuki
1996, 1997, Cole and Kisseberth 1995, Dell 1973, Goldsmith 1993a, Hockett
1973, Kenstowicz and Kisseberth 1977, 1979, Kisseberth 1969, Kuroda 1967,
Lakoff 1993, Newman 1944 (the original source), Noske 1984, Prince 1987,
Steriade 1986, Wheeler and Touretzky 1993, Zoll 1993.)
Yawelmani has three processes that interact opaquely:
a) Height-stratified rounding harmony: a suffix vowel takes on the round-
ing of the preceding vowel if they agree in height. E.g., the high-vow-
eled nonfuture suffix /-/ alternates as follows: [] ‘lead by
the hand’ vs. [] ‘find’; cf. [] ‘eat’, [] ‘tangle’. In

contrast, the nonhigh-voweled dubitative suffix /-/ alternates like

this: [] ‘throw’ vs. [] ‘recognize’; cf. [] ‘procure’,
[] ‘touch’.
b) Long-vowel lowering: underlying long high vowels become mid. E.g.,
/-/ → [] ‘fan’, /-/ → [] ‘urinate’.
c) Closed syllable shortening: long vowels are shortened in medial and
final closed syllables. E.g., /-/ → [] ‘burn’, /-/ →
[] ‘arrive’.

In a rule-based analysis or its Stratal OT counterpart (Kiparsky 2001), these

three processes must apply in the order shown in (2-30). (For a different analysis
of Yawelmani, see §3.3.5.) No other order of application will do. If the order
of rounding harmony and lowering were reversed, then the high suffix vowel
would not harmonize with an originally high root vowel that has been low-
ered because of its length: /-/ →lowering [-] →harmony DNA →shortening
*[]. And if the order of lowering and closed syllable shortening were
reversed, then the derived short vowel would fail to lower: /-/ →harmony
[] →shortening [] →lowering DNA →surface *[].
(2-30) Rule-based or Stratal OT derivation for Yawelmani
Underlying /-/ ‘steal’
Rounding harmony 
Long-vowel lowering 
Closed syllable shortening 
Surface []
Unsurprisingly, Yawelmani’s multiple opaque interactions cannot be analyzed
with a single selector constraint choosing a single sympathetic candidate. Two
selector constraints, two sympathetic candidates, and two sympathy constraints
are required. One of the selectors is the antishortening constraint max-μ, and
from the input /-/ it favors the sympathetic candidate []. The
other selector is IDent(high), and it favors the sympathetic candidate [].
Each sympathetic candidate has its own sympathy constraint. The role of
the sympathetic candidate [] is to determine the height of the actual
output []’s root vowel. It does this by way of a sympathy constraint
called IDent(high)max-μ. (The max-μ subscript is there to index this sympathy
constraint to the selector of its sympathetic candidate.) The other sympathetic
candidate, [], determines the rounding of []’s suffix vowel. It
does this by way of the sympathy constraint IDent(round)IDent(high). In sum, the
output’s vowel height is taken from the sympathetic candidate selected by

max-μ, whereas the output’s rounding comes from the sympathetic candidate
selected by IDent(high).
Once sympathy theory is provided with these additional analytic resources
to handle multiple opacity, however, it is in serious danger of overgeneration.
Kiparsky (2001) presents a simple but striking example. Assume that con has
only the markedness constraint no-coDa and consonant- and vowel-specific
versions of the faithfulness constraints DeP and max. As shown in (2-31),
DeP-v and max-c can each act as a selector. Given the input // and a
ranking where no-coDa dominates both DeP-v and max-c, DeP-v selects the
sympathetic candidate [], which deals with the potential coda by deletion, and
max-c selects the sympathetic candidate [], which deals with the potential
coda by epenthesis. The sympathy constraints favor outputs that resemble
these two sympathetic candidates in specific ways. The sympathy constraint
DeP-cDeP-v favors any candidate that has no consonants that are not present
in the DeP-v-selected sympathetic candidate []. This means that DeP-cDeP-v
favors forms that replicate []’s consonant deletion. Similarly, the sympathy
constraint max-vmax-c favors any candidate that has all of the vowels that are
present in the max-c-selected sympathetic candidate []. This means that
max-vmax-c favors forms that replicate []’s vowel epenthesis. The net
result is that the winner is [], a form that reproduces both []’s consonant
deletion and []’s vowel epenthesis.
(2-31) An unwelcome result of sympathy
DeP-cDeP-v max-vmax-c DeP-v max-c
// no-coDa
(sympathy) (sympathy) (selector) (selector)

→  1 1

a.  W1 W1 W1 L L


b. (sympathetic W1 1
L
via max-c)


c. (sympathetic W1 L 1
via DeP-v)

The problem with (2-31) is that this sort of opaque interaction is unattested and
no doubt impossible. The sympathetic candidates reflect two different ways of
satisfying no-coDa, epenthesis and deletion. The sympathy constraints force
the winner to reproduce the effects of both ways of satisfying no-coDa, both

epenthesizing and deleting when either one alone would be enough. Obviously,
additional constraints could be introduced to rule out [], but this sort of local
fix misses the broader point. This is simply not an attested opaque interaction;
it is gratuitous unfaithfulness, a kind of hyperopacity. It would seem, then,
that giving sympathy the power to deal with multiple opacity also gives it the
power to produce such unlikely results as (2-31).
A final remark about sympathy. Bye (2001, 2003), Jun (1999), and Odden
(1997) introduce variations on sympathy theory that make the sympathetic
form part of the candidate that is evaluated. One way of implementing this
idea is to assume that a candidate is not a single form but rather an ordered
pair: (sympathetic-form, output-form). In Bedouin Arabic, for example, the
winning candidate would be ([], []), and it competes against
alternatives like those listed in (2-32). There are various ways of constructing
a system of constraints for evaluating candidates like these; the comments at
the right in (2-32) give a sense of what the constraints will need to be sensitive
to. As we will see in §3, candidate chains are a somewhat similar idea.
(2-32) Candidates as (sympathetic-form, output-form)
([], []) Winner.
([], []) Palatalization mismatch.
([], []) No palatalization in the sympathetic form.
([], []) Syncope in the sympathetic form.

2.3.4.4 Opacity as a mechanism for preserving contrasts

Donegan and Stampe (1979), Kaye (1974, 1975), Kisseberth (1976), and
Gussmann (1976) propose that opaque rule ordering has a functional explana-
tion. The general idea is that opacity preserves phonemic contrasts, avoiding
neutralizations that would occur if the rules applied in transparent order.
Kaye (1974) looks at counterbleeding orders like the one in (2-33), which
comes from Ojibwa (Algonquian, US and Canada). This is a counterbleeding
order because, if the rules were applied in the opposite order, cluster simpli-
fication would deprive place assimilation of an opportunity to apply, yielding
*[] instead (cf. Catalan). Kaye observes that the opaque order makes
sense functionally: to a listener hearing the surface form, ‘… it is immediately
apparent that the underlying representation ends in k, given the rules cited
above and the fact that  is not part of the inventory of underlying segments.
… The only possible source of  is as a result of the assimilation of a nasal to
a following velar stop’ (Kaye 1974: 144). Kaye uses the term ‘recoverability’
to describe this functional motivation for opacity.

(2-33) Counterbleeding order in Ojibwa (Kaye 1974: 140)

Underlying /-/
Place assimilation 
Cluster simplification 
Surface [] ‘(if) he arrives’

Donegan and Stampe (1979: 145–151) make a somewhat similar point about
counterfeeding order. They see phonology as the result of conflict between
phonetic (articulatory) and phonological (perceptual) aims. Transparent interac-
tion of processes is phonetically motivated, since it presumably maximizes
articulatory ease. Opaque interactions of the non-surface-true variety are
phonologically motivated, in their sense, because opacity ‘bring[s] speech
closer to its phonological intentions’ (p. 147). As an example, Donegan and
Stampe cite nasal deletion and intervocalic flapping in English plant it. For
some speakers, they interact opaquely, as shown in (2-34). (Compare [)]
with [)) )], which other speakers produce from transparent interaction of the
same processes.) In Donegan and Stampe’s view, this and other instances of
counterfeeding order ‘prevent the merger of phonologically distinct representa-
tions’ (p. 147), such as plant it and plan it. The desire to avoid merger must be
weighed against the cost of opaque [)]’s greater articulatory difficulty in
comparison with transparent [)))].
(2-34) Counterfeeding order in English plant it
Underlying //
[nasal] assimilation )
[t] flapping Inapplicable because [t] is not intervocalic
Nasal deletion )
Surface [)]
Łubowicz (2003) develops an Optimality-Theoretic system, called PC theory
(for ‘preserve contrast’), in which these ideas about opacity’s functional moti-
vation are given a formal basis. With Flemming (1995), Padgett (2003), and
others, she assumes that the objects of phonological evaluation are systems
of contrasts, which she calls scenarios, rather than individual forms. (E.g.,
a scenario for German // ‘federation’ might include all of its logically
possible minimal pairs, including //.) This move allows con to include
constraints against neutralization, which can favor opaque interactions precisely
when they help to preserve a contrast that would otherwise be lost.
Łubowicz applies PC theory to, among other things, counterfeeding opacity.
Take an example like Bedouin Arabic // → [] (see (2-20)). Because
the output contains an unraised vowel in an open syllable, it offers a hint that
the openness of the syllable is not original. PC theory expresses this intuition

formally by introducing constraints on scenarios, among which is one called

PCIN(V/Ø). This constraint is violated by any output scenario that neutral-
izes a contrast between a vowel and zero that obtains in the input scenario.
Importantly, PCIN(V/Ø) does not say how the contrast is to be preserved; it can
be preserved as-is, or it can be transferred to some other segment, depending
on interaction with other constraints. Thus, if // and // both map to [],
PCIN(V/Ø) is violated, but it is not violated if // maps to [] and //
maps to [].
In the Arabic case, PCIN(V/Ø) evaluates scenarios like those in (2-35).
(Underlying // is hypothetical; the scenarios deal with possible rather
than actual words.) To block raising and thereby favor the opaque scenario,
PCIN(V/Ø) must crucially dominate the markedness constraint that favors rais-
ing, *aCV. The effect of the ranking PCIN(V/Ø) >> *aCV in (2-36) is that the
underlying contrast between a vowel and zero, which epenthesis threatens to
neutralize, is transferred to the quality of the vowel of the preceding syllable.
This brief analysis glosses over some important issues, but it is sufficient to get
a sense of how PC theory works and how it can account for opacity.
(2-35) Arabic scenarios in PC theory
a. Transparent
// → []
// → []
b. Opaque
// → []
// → []
(2-36) Counterfeeding opacity in PC theory

// PCIN(V/Ø) comP-coDa aCV DeP ID(low)

//

→ // → [] 1 1 1
// → []

a. // → [] W1 L W2
1
// → []

b. // → [] W1 L L 1
// → []

Łubowicz does not discuss counterbleeding opacity, but PC theory also seems
applicable to cases like Kaye’s Ojibwa example. Two of the relevant scenarios are
given in (2-37). The transparent scenario is clearly less marked and more faithful,
yet the opaque one wins. For this to happen, PCIN(C/Ø) must be ranked above
the markedness and faithfulness constraints that favor the transparent scenario.
This ranking argument is shown in (2-38). The idea is that the contrast between
/k/ and Ø is preserved by being transferred onto the preceding consonant.
(2-37) Ojibwa scenarios in PC theory
a. Transparent
/…/ → […]
/…/ → […]
b. Opaque
/…/ → […]
/…/ → […]
(2-38) Counterbleeding opacity in PC theory

/…/ PCIN(C/Ø) comP-coDa ŋ ID(coronal)

/…/

→ /…/ → […] 1 1
/…/ → […]

a. /…/ → […] W1 L L
/…/ → […]

b. /…/ → […] W1 L L
/…/ → […]

PC theory is by far the most original theory of opacity among those we have seen,
since it offers a radical restatement of the entire rationale for opaque interactions.
But it is not without its problems, the most serious of which can be illustrated
with the Ojibwa example in (2-38). The constraint PCIN(C/Ø) chooses the opaque
winner over its transparent competitor in (a). That may seem unexceptionable,
but in a way the PC constraint works too well. The victory of the opaque scenario
in (2-38) is unrelated to the fact that Ojibwa independently has a process of
nasal place assimilation (e.g., [] ‘(if) he arrived then’). In (2-38),
the winner manages to beat (a) without reference to any markedness constraint
that favors [] over []. The analysis in (2-38) therefore makes no connection
between how contrast is preserved and the independent existence of a nasal
place assimilation process in the language.

This means that, without significant modifications, PC theory allows con-

trasts to be preserved in ways that are fundamentally unnatural, since they do
not depend on markedness constraints for their motivation (Łubowicz 2003:
148–153). For example, the contrast between /k/ and Ø in Ojibwa could in
principle be preserved by rounding the preceding vowel: /…/ → […],
/…/ → […]. This outcome is made possible by ranking PCIN(C/Ø) above
*y and IDent(round), even though neither Ojibwa nor any other language is
likely to have a process that transparently maps /…/ to […].
Attested cases of opacity are not like this. The opaque process in Ojibwa
is not different in kind from the transparent processes in other languages. The
opaque processes that we actually find in languages are natural. In OT terms,
this means that those processes should devolve from standard markedness-
over-faithfulness rankings, since such rankings, combined with the universality
of con, are the only means within this theory for explaining phonological
naturalness. This result establishes a strong precondition for the adequacy of
any theory of opacity in OT.

2.4 What have we learned?

Perhaps the most striking result of this review of previous work on opacity is
the central role played by structure that is not present in either underlying or
surface representations. With the exception of contrast preservation (§2.3.4.4),
all reasonably complete theories of opacity make crucial reference, via rules or
constraints, to some nonunderlying, nonsurface representation. In SPE (§2.2.3),
it is the intermediate step of a serial derivation with ordered rules. In Stratal OT
(§2.3.4.2), it is the output of one stratum that is also the input to the next stratum.
In sympathy, targeted constraints, and comparative markedness (§2.3.4.3), it
is a more faithful candidate than the actual output form. And in Containment
or Turbidity (§2.3.4.1), the counterpart to the intermediate derivational step is
not exactly a level of representation, but it is nonetheless present as coexistent,
unpronounced structure in the output form.
The discussion of local conjunction (§2.3.4.1) and contrast preservation
(§2.3.4.4) emphasizes another point: opacity is a result of process interaction.
Our understanding of opacity cannot be separated from our understanding of
what ‘process’ means in OT, nor can it be separated from our understanding
of how different processes may make inconsistent demands on phonological
mappings. Opacity is deeply connected with the phonology of a language, and
any adequate theory of opacity must recognize this.

From all this, I conclude that there is some fundamental truth to the deri-
vational view of opacity. But OT denies the existence of rules and therefore
of rule ordering, for some very good reasons (McCarthy 2002b: chapter 3,
Prince and Smolensky 2004: chapters 2–5, 9). We have also seen that previous
attempts to meld OT with serial derivations or their analogues have not been
fully successful (§2.3.4.1–§2.3.4.3). The challenge, then, is to make use of the
derivational insight without losing hold of OT’s essential properties and basic
results. The next chapter presents a proposal intended to do exactly that.

Notes
1 The derivation in (2-4) also includes a bleeding interaction: deletion bleeds rais-
ing in the derivation of [].
2 Lexical exceptions will also look like they should have undergone a process
but did not. Kiparsky’s definition of opacity is limited to the effects of process
interaction and not exceptionality. See Laferriere (1975) for a useful distinction
between ‘internal’ opacity (exceptionality) and ‘external’ opacity (2-6).
3 Hayes’s (1990) notion of ‘precompilation’ provides a way of lexically listing
morphologized alternations that occur in external sandhi. Precompilation is
therefore intended for phenomena like morphological mutation. It is by intent
and by design inappropriate for examples like high-vowel deletion in Bedouin
Arabic, where there are no morphological or grammatical conditions.
4 Oh, the grand old Duke of York, / He had ten thousand men; / He marched them
up the hill, / And he marched them down again.
5 See Pater (2000) for an analysis of these data in OT using output-output cor-
respondence.
6 This formulation is due to Jane Grimshaw.
7 The qualification ‘at least’ implies that there may be more than one fully faith-
ful candidate from a given input. This is possible if there are dimensions along
which candidates may differ that are not protected by faithfulness constraints.
Syllabification is the standard example. For further discussion, see McCarthy
(2002a, 2003c) and §3.2.4.1.
8 Throughout, I follow Prince (2002) in using comparative tableaux. The winning
candidate appears to the right of the arrow, and losers are in the rows below it.
Subscripted integers stand for the number of violation marks incurred by a can-
didate, replacing the familiar strings of asterisks. In loser rows, the effects of the
constraints are indicated by W and L, W if the constraint favors the winner and
L if it favors the loser. These annotations are much more useful and perspicu-
ous than the exclamation point and shading that they replace. For example, the
sufficiency of any tableau can be easily checked: every L must be outranked by
(≈ to the right of) some W in the same row, and every loser row must contain at
least one W.
9 I am grateful to Adam Werle for help with the Nuuchahnulth data.

10 There is a great deal of previous literature discussing the challenges that opacity
presents to classic OT, including Archangeli and Suzuki (1996, 1997), Black
(1993), Booij (1997), Cho (1995), Clements (1997), Chomsky (1995), Goldsmith
(1996), Halle and Idsardi (1997), Idsardi (1998), Jensen (1995), Kager (1997,
1999b), McCarthy (1996, 1999), McCarthy and Prince (1993b), Noyer (1997),
Paradis (1997), Prince and Smolensky (2004), Roca (1997b), Rubach (1997), and
various contributions to Hermans and van Oostendorp (eds) (1999).
11 The double vertical line in the middle of tableau (2-16) is used to separate blocks
of constraints that cannot be ranked on the basis of the information provided.
That is, no ranking relations are asserted across this double line.
12 On compensatory lengthening, see among others Hayes (1989) and Wetzels and
Sezer (eds) (1986).
13 The constraint *comPlex-coDa (*comP-coDa) is violated by any tautosyllabic
syllable-final cluster.
14 The problems of constraint conjoinability and domains arise regardless of whether
we regard conjoined constraints as literally present in universal con or merely
immanent in it (see Ito and Mester 2003a: 24 on this distinction). Either way, lin-
guistic theory is obliged to explain why certain logical possibilities do not occur.
15 Modular, serial implementations of OT along the lines of the theory of Lexi-
cal Phonology have been proposed or discussed in the following works, among
others: Bermúdez-Otero (1999, forthcoming), Cohn and McCarthy (1994/1998),
Hale and Kissock (1998), Hale, Kissock, and Reiss (1998), Ito and Mester (2001,
2003b, 2003c), Kenstowicz (1995), Kiparsky (2003, to appear), McCarthy
(2000b), McCarthy and Prince (1993b), Orgun (1996b), Potter (1994), Rubach
(2000), and many of the contributions to Hermans and van Oostendorp (eds)
(1999) and Roca (ed.) (1997a).
16 The hypothesis that strata can eliminate the need for extrinsic ordering is
pursued in a somewhat different theoretical context by Goldsmith (1993a) and
Lakoff (1993). Their theories are distinct from both LP, because they do not allow
within-stratum ordering, and from Stratal OT, because they employ ‘two-level’
rules that can refer to input environments. (Stratal OT’s markedness constraints,
like classic OT markedness constraints, can only refer to output environments.)
17 DNA stands for ‘does not apply’.
18 The processes in (2-28), though opaque, are nearly exceptionless. I know of no
lexical exceptions to stressing of final closed syllables or deletion of final []. The
only lexical exceptions to epenthesis in final clusters are the words [] ‘nard’
(Canticles 4, verse 14), a borrowing from Persian, and [] ‘truth’ (Proverbs
22, verse 21), a hapax legomenon.
19 There is a fairly extensive literature on sympathy theory and its applications
(Bakovic 2000, Davis 1997a, 1997b, de Lacy 1998, Dinnsen et al. 1998, Fuka-
zawa 1999, Harrikari 1999, Ito and Mester 1997a, 1997b, 1998, Jun 1999, Kar-
vonen and Sherman [Ussishkin] 1997, 1998, Katayama 1998, Kikuchi 1999, Lee
1999, McGarrity 1999, Merchant 1997, Odden 1997, Parker 1998, Sanders 1997,
Walker 1998, 2003, Wilbur 1998).

EBSCOhost: eBook Collection (EBSCOhost) printed on 1/17/2025 6:43:52 AM UTC via PENNSYLVANIA STATE UNIVERSITY - PATERNO LIBRARY. All use subject to
https://www.ebsco.com/terms-of-use.
EBSCOhost: eBook Collection (EBSCOhost) printed on 1/17/2025 6:43:52 AM UTC via PENNSYLVANIA STATE UNIVERSITY - PATERNO LIBRARY. All use subject to
https://www.ebsco.com/terms-of-use.

Derivational Residue in Phonological Optimality Th... - (Surface Opacity of Metrical Structure in Optimality Theory)
No ratings yet
Derivational Residue in Phonological Optimality Th... - (Surface Opacity of Metrical Structure in Optimality Theory)
40 pages
Sympathy and Phonological Opacity
No ratings yet
Sympathy and Phonological Opacity
71 pages
Generative Phonology
No ratings yet
Generative Phonology
20 pages
Generative Phonology and SPE
No ratings yet
Generative Phonology and SPE
14 pages
Generative Phonology
75% (4)
Generative Phonology
13 pages
Generative Phonology Overview
No ratings yet
Generative Phonology Overview
8 pages
Generative Phonology: Presented By: IRMA MAMING (2020203879102001) FADHILAH SANTRI (2020203879102002)
No ratings yet
Generative Phonology: Presented By: IRMA MAMING (2020203879102001) FADHILAH SANTRI (2020203879102002)
9 pages
The Organization of The Grammar
No ratings yet
The Organization of The Grammar
37 pages
Syntax Phonology Interface PDF
No ratings yet
Syntax Phonology Interface PDF
26 pages
Nonlinear Phonology Notes
No ratings yet
Nonlinear Phonology Notes
3 pages
Rule Ordering in Generative Phonology
No ratings yet
Rule Ordering in Generative Phonology
26 pages
Chapter 1 - From Derived Phonology To Optimality
No ratings yet
Chapter 1 - From Derived Phonology To Optimality
23 pages
Eng 781 Assignment On Rule Formalisation and Rule Ordering Group 13
No ratings yet
Eng 781 Assignment On Rule Formalisation and Rule Ordering Group 13
30 pages
Redundancy Rules in Phonology
No ratings yet
Redundancy Rules in Phonology
45 pages
PDF
No ratings yet
PDF
9 pages
Overview of Phonological Theory
No ratings yet
Overview of Phonological Theory
16 pages
Zhang 2019
No ratings yet
Zhang 2019
20 pages
The Handbook of Phonological Theory
100% (1)
The Handbook of Phonological Theory
25 pages
Usino Peace
No ratings yet
Usino Peace
4 pages
Week 6
No ratings yet
Week 6
4 pages
Topic 13 Generative Phonology
No ratings yet
Topic 13 Generative Phonology
4 pages
14 Cole&Iskarous 2001
No ratings yet
14 Cole&Iskarous 2001
14 pages
3 Article 2 (2) 2020
No ratings yet
3 Article 2 (2) 2020
4 pages
History of Phonology
100% (2)
History of Phonology
19 pages
Vowel Epenthesis in Bengali: An Optimality Theory Analysis: Ph.D. Candidate, University of Victoria Linguistics
No ratings yet
Vowel Epenthesis in Bengali: An Optimality Theory Analysis: Ph.D. Candidate, University of Victoria Linguistics
11 pages
Connecting Underlying and Surface Representations
No ratings yet
Connecting Underlying and Surface Representations
14 pages
Phonology's Unique Structure Explained
No ratings yet
Phonology's Unique Structure Explained
20 pages
Introduction To Phonology
No ratings yet
Introduction To Phonology
12 pages
Introduction To Phonology
No ratings yet
Introduction To Phonology
12 pages
Pullum1976 1
No ratings yet
Pullum1976 1
42 pages
Modeling Phonological Competence Coursework Notes 7: Dangers and Deficiencies in The Model
No ratings yet
Modeling Phonological Competence Coursework Notes 7: Dangers and Deficiencies in The Model
4 pages
Underspecification and Markedness
100% (1)
Underspecification and Markedness
44 pages
Phonological Rules and Processes: An Introduction
No ratings yet
Phonological Rules and Processes: An Introduction
14 pages
What Is Optimality Theory
No ratings yet
What Is Optimality Theory
29 pages
Language and Linguist Compass - 2007 - McCarthy - What Is Optimality Theory 1
No ratings yet
Language and Linguist Compass - 2007 - McCarthy - What Is Optimality Theory 1
32 pages
Derivational Residue in Phonological Optimality Th... - (Introduction Optimality Theory and Derivational Effects)
No ratings yet
Derivational Residue in Phonological Optimality Th... - (Introduction Optimality Theory and Derivational Effects)
28 pages
8100
0% (1)
8100
101 pages
Phonological Rules
No ratings yet
Phonological Rules
41 pages
Chong Accepted Language
No ratings yet
Chong Accepted Language
53 pages
Neutralization
100% (3)
Neutralization
17 pages
The Phonetics-Phonology Interface
No ratings yet
The Phonetics-Phonology Interface
90 pages
Optimal It y Theory Phonology
No ratings yet
Optimal It y Theory Phonology
17 pages
Paper3371 PH
No ratings yet
Paper3371 PH
13 pages
Kiparsky, 2000
No ratings yet
Kiparsky, 2000
16 pages
Explanatory Adequacy in Linguistics
No ratings yet
Explanatory Adequacy in Linguistics
16 pages
Phonology in Perception Phonology Amp Phonetics 3110219220 9783110219227 - Compress
No ratings yet
Phonology in Perception Phonology Amp Phonetics 3110219220 9783110219227 - Compress
324 pages
Phonetics & Phonology II Guide
No ratings yet
Phonetics & Phonology II Guide
16 pages
Broselow on Mandarin Phonology Errors
No ratings yet
Broselow on Mandarin Phonology Errors
20 pages
Cohn2007WP16 Phonology
No ratings yet
Cohn2007WP16 Phonology
31 pages
Understanding Underlying Representation
No ratings yet
Understanding Underlying Representation
50 pages
Underlying Representation Slidesx1 PDF
No ratings yet
Underlying Representation Slidesx1 PDF
50 pages
2011 Generative Phonology
No ratings yet
2011 Generative Phonology
147 pages
Theories of Phonological Development
No ratings yet
Theories of Phonological Development
34 pages
17 Cole LgSci 2009
No ratings yet
17 Cole LgSci 2009
18 pages
Output Driven Phonology Theory and Learning
No ratings yet
Output Driven Phonology Theory and Learning
438 pages
Recognition of Design Symbols From Midship Drawings
100% (1)
Recognition of Design Symbols From Midship Drawings
14 pages
Cfa Imc
No ratings yet
Cfa Imc
4 pages
Analyzing Organization Cultures Bruce Fortado Instant Download
No ratings yet
Analyzing Organization Cultures Bruce Fortado Instant Download
77 pages
Impact of e Commerce On Business Performance Content File PDF
No ratings yet
Impact of e Commerce On Business Performance Content File PDF
8 pages
HPSW230-1-Jan-Jun2025-Course Outline-CB-V.1-05022025
No ratings yet
HPSW230-1-Jan-Jun2025-Course Outline-CB-V.1-05022025
159 pages
AE6312 Course Outline
No ratings yet
AE6312 Course Outline
6 pages
Sparse Optimization Lecture: Basic Sparse Optimization Models
No ratings yet
Sparse Optimization Lecture: Basic Sparse Optimization Models
33 pages
WBR PDF 1.0 Pages
100% (4)
WBR PDF 1.0 Pages
32 pages
DIGITAL MARKETING ANALYTICS in Theory and in Practice (KEVIN HARTMAN)
100% (3)
DIGITAL MARKETING ANALYTICS in Theory and in Practice (KEVIN HARTMAN)
274 pages
Group Name List
No ratings yet
Group Name List
12 pages
Measurement Techniques in Physics
No ratings yet
Measurement Techniques in Physics
5 pages
Insurance Textbook Chapter 1 Answers - 3rdedition
No ratings yet
Insurance Textbook Chapter 1 Answers - 3rdedition
3 pages
University of The Poonch Rawalakot: Terminal: Paper: QTB Marks 50 Time: 2Hrs
No ratings yet
University of The Poonch Rawalakot: Terminal: Paper: QTB Marks 50 Time: 2Hrs
3 pages
MMW Syllabus Final BAPOS
No ratings yet
MMW Syllabus Final BAPOS
21 pages
Stored-Product Pest Monitoring Methods: Armed Forces Pest Management Board Technical Guide No. 27
0% (1)
Stored-Product Pest Monitoring Methods: Armed Forces Pest Management Board Technical Guide No. 27
23 pages
ML Unit 5
No ratings yet
ML Unit 5
30 pages
Service Crew Performance in Convenience Stores
No ratings yet
Service Crew Performance in Convenience Stores
57 pages
Analysis of VArience
No ratings yet
Analysis of VArience
18 pages
Misconceptions and Nature of Math
No ratings yet
Misconceptions and Nature of Math
22 pages
Sociology
No ratings yet
Sociology
40 pages
Capital Structure Decisions Evaluating Risk and Uncertainty 1st Edition Agarwal Instant Download
No ratings yet
Capital Structure Decisions Evaluating Risk and Uncertainty 1st Edition Agarwal Instant Download
60 pages
Decision-Making For Autonomous Vehicles On Highway: Deep Reinforcement Learning With Continuous Action Horizon
No ratings yet
Decision-Making For Autonomous Vehicles On Highway: Deep Reinforcement Learning With Continuous Action Horizon
9 pages
mgmt8022 Session 02 2024
No ratings yet
mgmt8022 Session 02 2024
58 pages
Failure Tube Boiler
100% (5)
Failure Tube Boiler
8 pages
9 Instituties
No ratings yet
9 Instituties
21 pages
Writing An Essay 2
No ratings yet
Writing An Essay 2
2 pages
Sustainability Reporting in Electric Utilities
No ratings yet
Sustainability Reporting in Electric Utilities
9 pages
DM - RDL 1 - Module 3
100% (1)
DM - RDL 1 - Module 3
17 pages
Impact of Media Advertising On Consumer Purchasing Behavior in The Food and Beverage Industry in Malaysia
No ratings yet
Impact of Media Advertising On Consumer Purchasing Behavior in The Food and Beverage Industry in Malaysia
8 pages
Conflict Mistakes: Avoid & Improve
No ratings yet
Conflict Mistakes: Avoid & Improve
3 pages

JohnMcCarthy 2007 Chapter2 HiddenGeneralizations

Uploaded by

JohnMcCarthy 2007 Chapter2 HiddenGeneralizations

Uploaded by

2 Opacity, derivations, and Optimality

2.2 Opacity and derivations

underlying representation and the observed surface paradigm is sometimes

Rule A is said to bleed rule B if A can eliminate potential inputs to B. If A in

2.2.3 Opacity in derivations

(2-4) Counterfeeding order in Bedouin Arabic1

In the derivation // → [] in (2-4), the high-vowel deletion rule is

2.2.4 Simultaneous application

2.2.5 Theories of rule ordering

2.2.6 Later developments

In SPE, the aberrant stress of derived words is explained by their bracketing

cited by Anderson (1974) as evidence for local ordering. Instead of allowing

2.3 Opacity in Optimality Theory

2.3.2 Process interaction in classic OT

// *comP-ons DeP

(2-10) onset >> DeP

/--/ onset DeP

(2-11) Feeding interaction

// onset *comP-ons DeP

I noted in §2.2.4 that transparent interactions are incompatible with simul-

// *K ]σ *uK ID(round)

It is useful to compare the SPE-style analysis in (2-14) with the OT analysis

Before we go on to look at opacity in OT, however, it is appropriate to

2.3.3 Opacity in classic OT 10

/-/ *iCV max *ki ID(back)

// *aCV ID(low) *iCV max

(2-18) Counterfeeding opacity in classic OT

// max-A *aCV *iCV ID(low) max

In theory, this mode of analysis could be generalized to all instances of coun-

epenthetic. Constraints like this are necessarily embedded in a faithfulness

2.3.4 Previous approaches to opacity in classic OT

2.3.4.1 Opacity via novel substantive assumptions

Representational approaches to opacity. Opacity issues arose in the very first

A basic prediction of the Containment model is that syllabification always

The hallmark of opacity in Turbidity theory is a mismatch between the

constraints. recIProcIty maintains a check on divergence between the two

Segmental coalescence. Segmental coalescence may sound like a peculiar

ing to the deleted vowel. The deletion + assimilation combination is unlikely to

Local constraint conjunction. As I noted in §2.3.3, counterfeeding opacity

// [ID(low)&DeP]Adj-σ *comP-coDa *aCV DeP ID(low)

This is a particularly elegant theory of counterfeeding opacity, but it cannot

context to a distant, noninteracting context. Yet the local conjunction theory of

2.3.4.2 Analogues to serial derivations and rule ordering

Harmonic serialism. In harmonic serialism, the output of an OT grammar is

// max-A *aCV *iCV ID(low) max

(2-23) Harmonic serialism: second pass through grammar

] max-A *aCV *iCV ID(low) max

Multi-grammar serial OT. The principal thesis of the theory of Lexical

/-/ max *ki *iCV ID(back)

(2-25) Postlexical stratum

// *iCV *ki max ID(back)

(2-26) Counterbleeding order in Catalan (Kiparsky 1985:96–97)

Bedouin Arabic supplies another example of the insufficiency of between-

This is accomplished formally by deploying the output-output faithfulness

To analyze opaque stress and epenthesis using OO faithfulness, at a minimum

OO faithfulness’s inadequacy as a theory of opacity is not entirely unex-

2.3.4.3 Analogues to intermediate derivational forms

(2-29) Counterbleeding opacity with sympathy

The sympathetic candidate [] is identical with the intermediate stage

contrast, the nonhigh-voweled dubitative suffix /-/ alternates like

In a rule-based analysis or its Stratal OT counterpart (Kiparsky 2001), these

2.3.4.4 Opacity as a mechanism for preserving contrasts

(2-33) Counterbleeding order in Ojibwa (Kaye 1974: 140)

formally by introducing constraints on scenarios, among which is one called

// PCIN(V/Ø) *comP-coDa *aCV DeP ID(low)

/…/ PCIN(C/Ø) *comP-coDa *ŋ ID(coronal)

This means that, without significant modifications, PC theory allows con-

2.4 What have we learned?

You might also like

// K ]σ uK ID(round)

/-/ iCV max ki ID(back)

// aCV ID(low) iCV max

// max-A aCV iCV ID(low) max

// [ID(low)&DeP]Adj-σ comP-coDa aCV DeP ID(low)

// max-A aCV iCV ID(low) max

] max-A aCV iCV ID(low) max

/-/ max ki iCV ID(back)

// iCV ki max ID(back)

// PCIN(V/Ø) comP-coDa aCV DeP ID(low)

/…/ PCIN(C/Ø) comP-coDa ŋ ID(coronal)