Jump to ContentJump to Main Navigation
Localism versus Globalism in Morphology and Phonology$

David Embick

Print publication date: 2010

Print ISBN-13: 9780262014229

Published to MIT Press Scholarship Online: August 2013

DOI: 10.7551/mitpress/9780262014229.001.0001

Show Summary Details
Page of

PRINTED FROM MIT PRESS SCHOLARSHIP ONLINE (www.mitpress.universitypressscholarship.com). (c) Copyright The MIT Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in MITSO for personal use (for details see www.mitpress.universitypressscholarship.com/page/privacy-policy). Subscriber: null; date: 16 January 2019

Potentially Global Interactions Are Resolved Locally

Potentially Global Interactions Are Resolved Locally

(p.155) 6 Potentially Global Interactions Are Resolved Locally
Localism versus Globalism in Morphology and Phonology

David Embick

The MIT Press

Abstract and Keywords

This chapter examines the specific empirical predictions that globalism makes for allomorphic interactions, focusing on possible forms of evidence in favor of global computation of morphology and phonology. It considers two types of “global over local” effects, “unconditioned” allomorphs/phonological effects and (phonologically driven) allomorphic vacillation. It also discusses opacity in phonologically conditioned allomorphy (PCA) in the context of unconditioned allomorphy and allomorphic vacillation. The chapter presents cases of “outward-sensitive” allomorphy showing “nonuniform” paradigms as well as allomorphy conditioned by local, adjacent morphemes rather than phonological properties of the word. In other words, the strong predictions of globalism are not attested, regardless of whether paradigms are uniform or not. The chapter concludes with a discussion of cyclic Optimality Theory and global optimization in Latin.

Keywords:   Optimality Theory, globalism, morphology, phonology, opacity, phonologically conditioned allomorphy, unconditioned allomorphy, allomorphic vacillation, morphemes, Latin

This chapter looks directly at the empirical predictions that distinguish globalist and localist theories; in particular, it is centered on possible forms of evidence in favor of global computation of morphology and phonology. In the abstract, this means an argument showing that the morphological and phonological properties of some structure in some language are computed in a way that cannot be analyzed in a localist theory.

The type of argument that dominates the discussion below is based on scenarios in which global requirements effectively “override” other, more local considerations. Schematically, this type of case is stated in (1) in a way that is tailored to the discussion of phonologically conditioned allomorphy (PCA):

  1. (1) Global instead of local interactions

    These are cases in which local phonological considerations favor one allomorph, whereas global considerations—for example, brought about by the phonological form of “outer” affixes, or the phonology of the entire word—favor another allomorph, where it is the latter that is chosen.

More specifically, (1) refers to cases in which there is more than one allomorph for some morpheme—say, x1, x2, x3—such that (i) the distribution of these allomorphs is phonologically conditioned; and (ii) in a case where “local” conditioning requires x1, and global optimization requires x3, the language shows x3.

When cases of this type are examined in greater detail, it is possible to identify different types of effects that fall under the general heading of (1). The following subcategories are examined in detail below:

  1. (p.156) (2) Possible instantiations of (1)

    1. (a) ”Unconditioned” allomorphs/phonological effects: Theories with (at least some) global interaction between morphology and phonology (global-MP) allow for what look like locally “unconditioned” allomorphs to be inserted, or locally “unconditioned” phonological effects to be found, in cases in which this results in globally optimal outputs.

    2. (b) (Phonologically driven) allomorphic vacillation: Globalist theories predict that there should be cases in which the allomorph chosen for part of the paradigm of some Root differs from the allomorph chosen in another part of the paradigm. In such a case, different allomorphs are inserted for the same Root in a way that depends on the global phonological context. The head showing the different allomorphs can be said to show allomorphic vacillation in this scenario. Crucially, these hypothesized effects could go beyond the local types of outward-sensitive allomorphy predicted by the theory of part I.

The search for (1) and its manifestations in (2) connects with another point. In many cases of PCA, phonological processes make selection opaque by removing from the surface form the phonological factor that determines the choice among competing allomorphs. Questions about opacity are natural in the discussion of globalist versus localist theories, for reasons that dominate discussion in the phonological literature on parallel versus serial rule/constraint interaction (see Idsardi 2000 for one overview).

The introduction of opacity into the discussion of allomorphy highlights the architectural predictions of globalist theories. While it is true that opaque allomorphy presents certain challenges for theories that deny serialism (see, for different perspectives, Vaux 2003; Aranovich et al. 2005; Łubowicz 2005; Paster 2006; Bye 2008), there is a sense in which globalist theories also predict global effects that go far beyond normal opacity. Cases of the latter type are crucial to understanding the strong predictions of globalism. The discussion below therefore advances via a general discussion of opacity in PCA in section 6.1, with the two types of “global over local’’ effects outlined in (1)—unconditioned allomorphy and allomorphic vacillation—at the center of section 6.2 and section 6.3. The main thrust of these sections is that there are situations in which the strong predictions of globalist theories—those in (1) and (2)—could be (p.157) seen, but that the interactions found in actual languages are those expected in a localist framework. In particular, there is no evidence for unconditioned allomorphy, and in cases where a morpheme is expected to vacillate given the shape of outer morphemes, no such alternation is found. Thus, the strong predictions of globalism are not borne out.

In cases where allomorphic vacillation perhaps should occur, but does not, it could be argued that this is the result of constraints that force the same allomorph to be chosen throughout a “paradigm”: PARADIGM UNIFORMITY, in the sense of Kenstowicz 1996 and related work. In section 6.4, this point is addressed with reference to cases of “outward-sensitive” allomorphy. Cases of this type show “nonuniform” paradigms. At the same time, these cases show allomorphy conditioned by local, adjacent morphemes, not by phonological properties of the word. In other words, the strong predictions of globalism are not attested, whether paradigms are uniform or not.

6.1 From Opacity Effects to Global Interactions

Opaque interactions are generally held to be problematic for Optimality Theory (OT), for reasons that have been amply detailed in the phonological literature. Thus, the idea advanced above—namely, that effects that are related to opacity are important for understanding the strong predictions of globalism—requires some unpacking.

Opaque allomorphic selection is in evidence in several of the examples studied in preceding chapters, including Haitian Creole definite allomorphy (section 5.1) and many of the case affixes seen in Djabugay and Yidiɲ (section 5.2 and 5.3). The defining property is that these cases involve (i) PCA conditioned by some element in the host, and (ii) additional phonological processes (often, but not always, deletion) that render the allomorphic conditioning opaque.

In the Haitian Creole definite, for example, -a is inserted after V-final nouns and -la after C-final nouns. In the subset of the V-final nouns that have epenthesis—that is, those nouns that end in [+ATR] vowels—a glide is inserted. Viewed sequentially, this looks as follows for bato ‘boat’:

  1. (3) Sequence




    Vocabulary Insertion


    Glide Insertion

(p.158) In a theory without serial steps, however, there is a prima facie difficulty with batowa. The presence of the glide in the surface form makes it effectively C-final, such that, all other things being equal, the -la allomorph is expected. Put slightly differently, the surface distribution of allomorphs is complicated by this effect: the -a allomorph appears on the surface after both consonants and vowels, whereas -la appears only after consonants.

In a serialist theory with intermediate derivational steps, this kind of interaction is expected. Specifically, in the representation that is accessed for Vocabulary Insertion -a is inserted, since at that stage of the derivation, the definite morpheme is next to a V-final host. In this type of analysis, it can be said that the definite morpheme is in the local conditioning environment for the insertion of -a when Vocabulary Insertion takes place. In serialist theories, then, the fact that the conditioning factor for some change is not “local” to the locus of the change in a surface form is irrelevant; the point is that at an earlier derivational stage where the relevant computation (in this case, Vocabulary Insertion) is executed, the local conditioning environment for the computation is found.

The notion of local conditioning environment is crucial to understanding the predictions of different frameworks. In the domain of phonological interactions, globalist and parallelist theories like OT effectively dispense with the idea that being in a local conditioning environment is what determines that a form changes in a particular way. Instead, whether or not a surface form is changed relative to the input is determined by the globally interacting system of constraints. This makes the notion of local conditioning environment epiphenomenal; to the extent that local interactions take place, they are entirely derivative of the global system of constraint interaction.

This architectural claim of globalist theories makes clear predictions about which factors are potentially visible for the purposes of allomorphic selection. Globalist theories allow for a multitude of nonlocal interactions, of which standard cases of opacity are a subtype. Although surface-oriented theories might have difficulties with ‘‘standard’’ cases of opacity, the other types of allomorphic interaction that are predicted to exist if local conditioning environments are epiphenomenal are crucial for testing the predictions of globalism.

6.1.1 Opacity and Global Interactions

Some initial points about opacity and its relation to global effects can be made concrete with reference to a textbook example of opacity: epenthesis (p.159) in Turkish (see, e.g., Lewis 1967; Kager 1999). The first person singular possessive morpheme has the form -m. It surfaces as such after V-final nouns (4a); after C-final nouns, affixation of -m is accompanied by epenthesis, so that, as seen in (4b), a vowel is inserted between the final consonant of the host and the -m suffix:

  1. (4) Epenthesis in Turkish







    ‘measure’/’my measure’




    ‘hand’/’my hand’

The opacity involving the -m morpheme arises in cases in which the epenthetic vowel appears after a velar consonant. Turkish has a phonological rule of Velar Deletion that deletes such consonants intervocalically. So, for ajak ‘foot’, the 1sg possessive form is ajaim. In a theory with ordered rules, this effect is analyzed with a derivation in which Epenthesis is ordered before Velar Deletion:

  1. (5) Example






    Velar Deletion

Putting to the side various ways in which surface-oriented theories could produce ajaim over (e.g.) *ajam, there is a general point here for the study of global interactions. The effect seen in (5) is one in which an epenthetic vowel appears in an environment in which it is not locally conditioned on the surface. The localist theory accounts for this with ordering: the structural description for Epenthesis is found at an intermediate stage of the representation, so that the epenthetic vowel is, in the terms employed above, locally conditioned.

While this particular type of surface-unconditioned effect is difficult for OT, the broader point is that locally ‘‘unmotivated’’ effects are in principle not a problem for theories that espouse globalism. As stressed above, one of the defining properties of such theories is the ease with which they dispense with the notion of local conditioning environment. Thus, the fact that a ‘‘change’’ occurs in a way that does not seem locally motivated in surface forms is not problematic in general. Rather, the problems in the specific case of Turkish epenthesis (and other cases like this) are the following. First, there are no obvious, phonologically natural factors in the surface form of the word that would produce the actual form (i.e., that would motivate epenthesis).1 Second, by ordering Epenthesis before Velar (p.160) Deletion, the serialist theory provides an obvious answer to the question of why the epenthetic vowel appears in spite of not being between consonants on the surface.

6.1.2 Over-/Underapplication in Allomorphy?

Whatever solution is offered for “standard” opacities of the type discussed immediately above, the crucial point for present purposes is that theories with global-MP predict overapplication and underapplication in allomorphic selection, in the same way that overapplication and underapplication are predicted in the phonology. This can be seen when the intuition behind standard OT treatments of overapplication in reduplication are extended to allomorphic interactions.

Recall that the general idea in globalist theories is that what is relevant for surface phonological form is not whether a particular element is in a configuration that triggers a change. Instead, the change happens when the overall constraint ranking prefers the candidate with the change, even if the local conditioning environment for the change is not found in the surface form. This type of reasoning is illustrated in (6), which shows McCarthy and Prince’s (1995) analysis of overapplication in Tagalog /paN-RED-pu:tul/, which surfaces as pa-mu-mu:tul ‘a cutting in quantity’. In this example, the stem-initial /p/ surfaces as /m/, even though it is not adjacent to the paN- affix that triggers nasalization. The analysis involves the interaction of three constraints: a phonological Constraint that forces “mutation” of /p/ to /m/, a constraint requiring basereduplicant identity (IDENTBR), and the standard input-output faithfulness constraint (FAITH10):

  1. (6) McCarthy and Prince’s (1995) analysisPotentially Global Interactions Are Resolved Locally

The stem-initial /p/ surfaces as /m/ in the winning candidate because base-reduplicant identity outranks the faithfulness constraint that penalizes candidates with changes to the underlying form. Thus, even though the relevant /p/ is not in the local conditioning environment associated (p.161) with the /p/ → /m/ mutation, it surfaces as /m/ because of the identity constraint. In this way, the global constraint ranking enforces a change that is not locally expected given the surface form.

The general effect seen in this type of analysis can be called nonlocal (NL) application:2

  1. (7) NL-application

    An effect is found in a surface form even though the effect is not constrained to its (typical) local conditioning environment, because the constraint system allows global forces to override local ones.

The example from Tagalog analyzed in (6) does not directly involve globalism in the global-MP sense. While it involves apparent ‘‘action at a distance,’’ in the way described in (7), it is not the same kind of allomorph selection that is studied throughout this book. However, the type of interaction that it shows can easily be formulated in a way that implicates global-MP as well, to yield predictions about PCA; this is the topic of the next section.

6.2 Allomorphy and Nonlocal Application

‘‘Standard’’ opacity effects are a subcase of NL-application, namely, the subset in which the serialist theory would have the effect derive from local conditioning by an element at an intermediate stage of a derivation. While standard cases of opacity are congenial to localist/serialist theories, the general type of NL-application allowed by globalist theories—that is, the general principle that global effects can trump local conditioning in ways that do not involve local interaction at intermediate stages—defines a range of cases that cannot be analyzed on a localist approach. Identifying the properties of these cases is a crucial step in understanding the predictions of globalism.

6.2.1 Turkish Third Person Singular Possessive

A case of allomorphic selection that illustrates the possibility of NLapplication is found in the Turkish 3sg possessive morpheme (see Lewis 1967; Carstairs 1987; Kornfilt 1997; Aranovich et al. 2005; Paster 2006). This appears to be a relatively straightforward case of (C)V allomorphy, with -SI after vowels and -I after consonants (vowel harmony also affects the vowel component; examples from Paster 2006)

  1. (p.162) (8) Two allomorphs: -I (after C); -SI (after V)





    ‘its price’


    ‘tis own’


    ‘its tool’



    ‘its attrition’


    ‘its apple’


    ‘its be’

The alternation between -SI and -I interacts with the process of Velar Deletion, described above. Recall that this rule deletes velars intervocalically:3

  1. (9) Velar Deletion

    K→ Ø/V_V

The 3sg possessive allomorph -I is inserted after /k/-final stems. This produces the environment for Velar Deletion, which then applies to yield forms that have hiatus and that are opaque in terms of allomorph selection:

  1. (10) Possessives of velar-final nouns






    hunger’/’its hunger’



    ‘baby’/’its baby’



    ‘necklace’/’its necklace’



    ‘bread’/’its bread’

Assuming a localist theory like the one in part I of this book, and on the further assumption that the -SI/-I alternation involves competition between two distinct allomorphs, Turkish has the VIs in (11):4

  1. (11)

    • [poss] ↔-SI/V_

    • [poss] ↔I/C_

After Vocabulary Insertion, Velar Deletion applies in the phonology.

For OT, these facts present a general challenge, as is typically the case with opacity. Whatever solutions might be proposed for this particular case, the Turkish 3sg allomorphy—and some hypothetical variants of Turkish in particular—illustrate the predictions that globalist frameworks make concerning NL-application.

(p.163) 6.2.2 NL-Application: “Overriding” Local Concerns

Informally, localist theories are incapable of accounting for lookahead conditioning, of the type in (12):

  1. (12) Insert affix x in a particular environment, unless doing so creates an undesirable representation because of interaction with other phonological or morphological processes that occur later in the derivation.

In the terms employed above, a theory with NL-application could easily derive such effects. They would not involve lookahead, obviously, but instead a constraint ranking in which the global system produces a result that looks surprising from the perspective of a theory in which computations are restricted to apply in local conditioning environments.

Schematically, the specific manifestations of NL-application that are expected by globalism can be seen as types of overapplication:

  1. (13)

    1. a Allomorphic overapplication

      A locally ‘‘unconditioned’’ allomorph is inserted instead of the expected one, because when the whole word is taken into account, the net result is better.

      Example (Turkish′) : In the Turkish case above, -SI is inserted after velar-final stems, in order to avoid the hiatus created by Velar Deletion. This would yield, for example, bebek-si. (Viewing this as allomorphic underapplication of the -I allomorph amounts to the same thing.)

    2. b. Allomorph-driven phonological overapplication

      Rather than inserting an ‘‘unexpected’’ allomorph to avoid a problem, it should also be possible to see the surface results of a phonological change, even though its environment for application is not met locally.

      Example (Turkish′) : In the Turkish case above, the velar /k/ is deleted and -SI is inserted to yield bebe-si.

The specific analyses of the patterns in (13) can be sketched in a way that illustrates the basic point. Beginning with Turkish′ in (13a), if a constraint penalizing hiatus, *HIATUS, is ranked higher than the *VKV constraint that enforces velar deletion, then the -SI/-I alternation could be analyzed directly as a case of Phonological Selection, where, exclusively with velars, the ‘‘local’’ effect that selects -I with C-final hosts is (p.164) overridden. This analysis is shown in (14), where (l4i,ii) show the simple cases of allomorphy, and (14iii) shows the *HIATUS-driven allomorphic overapplication effect:5

  1. (14) Turkish′Potentially Global Interactions Are Resolved Locally

In the analysis of Turkish′, the simple cases of allomorphy between -SI and -I emerge from the interaction of the constraints *HIATUS and NOCODA. The constraint MAX(C) prevents deletion of consonants and rules out other conceivable surface forms like *bede-SI. Because of the way the constraint driving Velar Deletion, *VKV, interacts with these constraints, the optimal candidate for velar-final stems is bebek-SI, with -SI instead of -I The net result of this constraint ranking is a version of Turkish in which -SI is optimal for velar-final stems, because this allomorph choice avoids both hiatus and intervocalic velars.6

The Turkish′ example shows the insertion of what is, in effect, a locally unconditioned allomorph, as outlined in (13a). The Turkish′′ (13b) type of case, in which a phonological process overapplies, is easy to formalize as well. In particular, it is also possible to rank the constraints so that bebe -SI is optimal:

  1. (15) Turkish ′′Potentially Global Interactions Are Resolved Locally

(p.165) Naturally, the constraint rankings involved in either of the two hypothetical languages just considered would have to be supported by larger analyses of the language. At the same time, these two possible systems clarify the types of phenomena that would provide evidence for globalism.

A localist theory has some difficulties producing the hypothetical forms. The generalization for Turkish′ is that -I is inserted after nonvelar consonants, and-si elsewhere. It is not clear that VIs could refer to a phonologically unnatural class in this way. The VIs required would have to be these:

  1. (16)

    • [3sg] ↔-I/C[—vel]⏜_

      [3sg] ↔ -SI

Reference to an unnatural phonological environment (nonvelar consonants) might be impossible, depending on how this part of the theory is configured.

The situation with bebe-SI is similar, although slightly more is required of a localist theory. The VIs in (16) could be employed to state the distribution of these exponents. In the case of velar-final stems, an additional (readjustment) rule is required that deletes the stem-final velar in front of the -SI suffix.

6.2.3 A More Extreme (Hypothetical) Case

The examples from hypothetical Turkish might be salvageable on a localist theory, as just indicated. The reason that some potential localist analyses of these effects can be formulated is that allomorph choice can still be made on the basis of something that is locally visible to the 3sg possessive morpheme. But it is also possible to construct examples in which the factors forcing allomorph selection are not adjacent to the morpheme in question. This kind of effect is easy to formulate in a globalist theory, but goes beyond what a localist theory can express.

(p.166) One type of example along these lines has an additional morpheme intervening between two other morphemes that show allomorphy. Consider, for example, a language in which Roots may be followed by three morphemes, -X, -Y, -Z, where these have the allomorphs listed:

  1. (17) Potentially Global Interactions Are Resolved Locally

  2. (18) Root-X-Y-Z

    1. a X:-tak; -ilub

    2. b Y:-o

    3. c Z

      1. I Z1: -bat

      2. ii Z2: tarag

In the simple cases—that is, in examples where -Y and -Z are null or not present—the -X morpheme shows PCA based on the metrical properties of the host:

  1. (19)

    • - tak after odd-syllabled host

    • - ilub after even-syllabled host

Suppose further that the -Z morpheme is not subject to contextual allomorphy at all, so that, for example, -Z1bat and -Z2tarag are associated with different feature combinations.7

With global interaction, it is possible to set things up so that the allomorph of the -X morpheme vacillates depending on what is inserted into the outer and nonadjacent -Z morpheme. Beginning with the simple cases with only -X, it can be hypothesized that a PARSE-σ constraint favoring even-numbered words accounts for the pattern of allomorph selection shown by -X (footing shown):

  1. (20) Root-X cases; Roots = blik, golut

    1. (a) (blik-tak)

      *(blik-i)lub (violates PARSE-σ)

    2. (b) *(golut)-tak (violates PARSE-σ)


In the more complex structures, with the additional -Y and -Z morphemes, what is optimal at -X depends on which morpheme appears in the outer and nonadjacent -Z position. This is illustrated in (21), where foot boundaries are again shown for expository purposes; the two sub (p.167) cases show how -X varies depending on whether -bat or -tarag is inserted at -Z:

  1. (21)

    1. a

      1. I blik-X-o-bat:-tak inserted at -X



      2. ii blik-X-o-tarag:-ilub inserted at -X



    2. b

      1. I golut-X-o-bat:-ilub inserted at -X



      2. ii golut-X-o-tarag:-tak inserted at -X



Clearly, the way these examples work involves global considerations. The superficially “local” requirement that -tak appear after oddand -ilub after even-syllabled hosts is overridden by the global pressure exerted by the phonology of -Z’s exponent. The (output) phonology determines the morphology of allomorph selection, and the properties of the whole word have to be visible simultaneously for this to be done properly.

In a localist theory, this effect cannot be derived. In the theory of chapter 2, there are two reasons for this. The first is that an inner morpheme cannot be sensitive to the phonology of an outer morpheme, by the assumption of cyclic or “outward” Vocabulary Insertion. The second reason is that the -X morpheme is not adjacent to the -Z morpheme and therefore cannot see it for allomorphic purposes. The most that could be stated is the part of the distribution that is seen in the “basic” cases, where -ilub is inserted for -X next to a foot boundary:

  1. (22)

    • [X] ↔ilub/... ]⏜

      [X ] ↔tak

This analysis predicts that -X’s allomorphy should depend only on the metrical properties of what is to its left, whatever form -Z may ultimately take. It is incapable of stating the pattern described above.

There are two points to be made about the kind of example examined here. The first is that such cases would be a clear argument in favor of a globalist theory. The second is that there appears to be no evidence (p.168) that this type of effect is found in any language; in actual cases where something like this hypothetical scenario can be found, the facts are those expected in the localist model; they show no evidence for global computation.

6.2.4 Local and Cyclic Interactions

The literature has to a limited extent addressed predictions of globalist theories along the lines schematized above. In one type of case, the point has been made that a localist or cyclic theory makes the correct predictions. Some shorter cases of this type are reviewed below, followed by some comments on cyclic OT in section 6.2.5. Another type of case involves explicit arguments for surface phonology determining allomorph selection. In section 6.3, I examine in greater detail arguments along these lines from Mester 1994 for global allomorph selection in Latin. I show that when the relevant facts are analyzed in detail, the argument for global interaction collapses. Affix Placement in Huave

An early clarification of the predictions of globalism is made by Noyer (1993), who discusses the behavior of “mobile” affixes in Huave. Some affixes in this language, like -t- pasttense, attach to an element that is analyzed as a theme vowel. The theme vowel is sometimes a prefix, sometimes a suffix, in ways that correlate with transitivity: the theme is a prefix with transitive verbs and a suffix with intransitive verbs. The set of affixes to which -t- belongs attaches to the theme in either case (i.e., whether the theme is a prefix or a suffix):

  1. (23)

    1. a t -a -wit’

      PAST -TH -raise

      ‘he/she raised (it)’

    2. b wit’ -I -t

      raise TH -PAST

      ‘he/she rose up’

Affixes like -t- are ‘‘mobile’’ in the sense that they may occur either as prefixes or as suffixes.8

Noyer explores the possibility that this distribution results from the requirement that Huave words must have final codas. This is a version of Phonological Selection in which output phonology determines not allomorphy, but the placement of morphemes in a word.

On the assumption that the theme vowel’s status as a prefix or a suffix depends on morphosyntactic factors, the phonology determines the placement (p.169) of the mobile affix. Specifically, candidates like t-a-wit’ and a-wit’-t and t-wit’-a and wit’-a-t are considered for the prefixal and suffixal theme cases, respectively. The constraint system selects the candidates that meet the phonological condition requiring final codas.

Noyer goes on to discuss a further set of examples that implicate the questions about globalism raised above. In some forms, the mobile affixes can occur inside of other affixes; this is seen in (24a) for past tense -t-, and in (24b) for 1sg -n-, which is also mobile:

  1. (24)

    1. a wit’ -i -t -as -on

      raise -TH -PAST -1AUG

      ‘we-INCL rose’

    2. b sa -wit’ -i -n -on

    3. (1) FUT -raise -TH -1 -AUG

      ‘we-EXCL will rise’

As Noyer points out, if the whole word is evaluated in these types of cases, there is no reason for the mobile affixes to appear where they do. As far as the condition on final codas is concerned, the 1sg -n- morpheme in (24b) could be realized as a prefix, as in *sa-n-wit’-i-on. The solution Noyer offers is that evaluation of well-formedness occurs cyclically. In the case of (24b), for example, this means that when the placement of -n- is determined, “outer” suffixes are not present in the computation.

Examples of this type are important: they are cases in which the strong predictions of globalism could conceivably be manifested, but instead what is found is what is expected from a cyclic point of view. A fully global theory predicts that there should be interactions that do not show this kind of cyclic effect—that is, where the full globality discussed in the preceding sections is required. In such a theory, it is possible to model this type of interaction, either indirectly or directly (in the latter case, by assuming cyclic or stratal OT; see section 6.2.5); however, placing restraints on the theory in this way is not an argument for globalism. Saami

Moving past affix placement to allomorph selection, the literature provides additional cases in which global considerations allow for a type of allomorph selection that is not possible in the localist view, and, once again, the globalist theory must be “restrained” to produce the correct results, that is, to exclude other cases that might be expected to arise.

The sensitivity of various allomorphs in Saami (Lappish) to metrical structure is addressed in Dolbey 1997 and Orgun and Dolbey 2007 (also (p.170) see Bergsland 1976; Hargus 1993), which discuss the interaction of cyclic and local factors versus global optimization in Saami verbs. The allomorphy in question appears to be phonologically optimizing in the sense that it yields surface forms that contain an even number of syllables; the examples here are drawn from the person/number system, along with a passive morpheme:

  1. (25) Person-marking/Passive allomorphy

    1. a. Allomorphs by host syllable count












      3pl pret





      b. Examples: jearra ‘ask’, veahkehea ‘help’













      3pl pret






From the perspective of the phonology, this pattern of allomorphy creates even-syllabled forms that can be exhaustively parsed into binary feet.

Dolbey (1997) makes the point that the evaluation that results in this distribution appears to be local rather than global in character. In cases in which more than one of these affixes is added to a host, there is more than one possible outcome that optimizes the syllable count. For example, with a 2du passive form, adding two monosyllabic affixes results in an even syllable count, just as adding two disyllabic affixes does. A localist theory predicts that the disyllabic affix must be inserted in the inner morpheme position, since this is what the local context demands; following this, the local environment forces selection of another disyllabic affix.

The facts show that in the cases in question, two disyllabic affixes are selected:9

  1. (26)

    • je:rro-juvvo-beahti; *je:rru-vvo-hppi

      veahkehu-vvo-beahtti; *veahkehu-juvvo-hppi

Again, this is the type of situation in which the strong predictions of globalism could be manifested. If in the cases where two metrically conditioned (p.171) allomorphs are found, two monosyllables were inserted, the putative phonological “target” of allomorph selection—exhaustively parsable structures—would be achieved; and it would be achieved in a way that could not be stated in the localist theory, where lookahead to outputs is impossible. Instead, however, the interaction appears to be local, in a way that follows naturally from the localist theory. It is, of course, possible to state such a pattern in a globalist theory, but that is not at issue. Rather, the point is that a case in which the strong predictions of globalism could conceivably be found functions in terms that can be analyzed in the more restrictive localist architecture.

6.2.5 Some Comments on Cyclic Optimality Theory

The idea that local concerns trump global ones is, in some sense, the motivation for cyclicity. Some theories have sought to restrain the predictions of a fully globalist architecture by proposing that constraint evaluation is cyclic, in the sense familiar from Lexical Phonology (see, e.g., Kiparsky 2000 and subsequent work).

It is important to note that while cyclic OT is able in principle to account for (at least some of) the cases examined above, it still makes predictions that are very different from those of a localist theory. In particular, a cyclic OT theory is still globalist within any given stratum of affixation. While this type of theory restrains predictions about allomorphy in comparison with a fully globalist model, there appears to be no evidence for this limited amount of global interaction between morphology and phonology.

The specific predictions made by a stratal or cyclic OT model depend on how cycles of affixation are defined. The primary point to be made is that, in any theory that allows three morphemes to have their morphology and phonology computed in the same cycle, NL-application is predicted. This point is schematized in (27):

  1. (27) Root-X-Y-Z

If the heads X, Y, and Z are processed in the same cycle (perhaps in a way that excludes other, outer heads), the theory predicts that allomorph insertion at X could be sensitive to the phonology of Z, or the phonology of the whole object containing Z. These types of effects cannot be stated in the localist theory; but they do not seem to be found. Other effects, such as those involving the phonological form of two morphemes, as in the Saami example above, would also be predicted to show global behavior as long as the two morphemes are in the same stratum. Again, there is no clear evidence that this kind of limited global interaction is attested.

(p.172) Thus, while appealing to stratal or cyclic OT might rule out some of the (unattested) cases predicted by a fully globalist model, it makes predictions about morphology-phonology interactions that are evidently not found.

6.2.6 Interim Assessment

A number of cases, both hypothetical and real, were examined above in order to specify and test the predictions of globalist theories. A basic point where globalism and localism differ is that globalism predicts allomorphic effects in PCA that are locally unconditioned, but that make sense when the global, surface phonology is taken into account.

There appears to be no clear evidence that interactions of this type are found in natural languages. In the cases that have been studied, selection appears to proceed step by step, in a way that is expected from the point of view of a cyclic localist theory.

The same points are made by a more detailed examination of certain patterns of allomorphy in Latin verbs, to which I now turn.

6.3 Case Study: Arguments for Global Optimization in Latin

The predictions of globalist theories can be seen quite clearly in two case studies from Latin, drawn from Mester 1994. Each one involves the distribution of allomorphs in the verbal system: perfect -u versus -s in conjugation II verbs, and theme vowel -ī-versus -i- in the so-called io-verbs. In each case, standard handbooks of Latin allude to metrical patterns that correlate with the allomorphic patterns. Whatever status these claims might have within Latin historical phonology and morphology, Mester goes one step further than this, by arguing that the distribution of allomorphs in the synchronic grammar of Classical Latin requires a globalist framework in which selection of (certain) host-allomorph combinations is computed by generating all of the relevant combinations and letting the phonology determine the winner.

Closer examination of both cases shows that Mester’s arguments for globalism fail to provide any convincing evidence for such a framework. The proposals apply only to a carefully selected set of forms and make incorrect predictions when extended beyond these. The two cases do, however, pave the way for discussion of a further strong prediction of globalism, called allomorphic vacillation above: a “switch” in the selected allomorph for a particular root, based on (phonological) properties of outer morphemes. This strong prediction is not borne out in Latin; nor, to my knowledge, is it manifested elsewhere.

(p.173) Part of this section is thus devoted to a negative demonstration. In addition, though, these cases show important patterns to attend to—in perfect formation in particular—and these patterns show the kind of locality effect discussed in section 6.2.

6.3.1 Latin Perfect Allomorphy in the Second Conjugation

Mester’s (1994) influential discussion of the perfect forms of (some) Latin verbs is often cited as an example in which global prosodic considerations play a decisive role in allomorph selection.

Conjugation II Latin verbs show the theme vowel -ē- in the present tense system: thus, we find infinitives like mon-ē-re ‘to warn’ and aug-ē-re ‘to increase, enlarge’. The argument that Mester makes for phonology determining allomorphy is based on the perfect forms of (some) verbs from this conjugation.

Throughout the Latin verbal system, the perfect tenses show a great deal of allomorphy. I will assume here that, in addition to stem changes, what is at issue is the allomorphy of the Aspectual head Asp[perf ]; recall the discussion of section 3.1.1, where the structure in (28) is assumed:

  1. (28) StructurePotentially Global Interactions Are Resolved Locally

The Asp[perf] head in finite forms shows different allomorphs, including the vowel -u (often written -v), -s, and -i (1sg citation form employed here):10

  1. (29) Perfect allomorphs in Latin

























(p.174) The case that Mester concentrates on is in conjugation II, where the distribution of -u and -s in the (29b,c) types is, according to the traditional literature (see Meiser 1998, 2003 for overviews), correlated with metrical factors: light stems take u, and heavy stems -s.11

Whether or not the prosodic correlations connected with this pattern of allomorphy are descriptively accurate, the interesting point for present purposes is how Mester accounts for this effect in terms of competition in the synchronic grammar of Classical Latin. Mester’s primary focus is on sets of effects correlated with trapping configurations—more precisely, instances of medial trapping, where an unparsed light syllable appears after footed material:12

  1. (30) Medial trapping


This sort of configuration arises in a moraic theory where trochees are both minimally and maximally bimoraic.13 The essential idea behind Mester’s proposal is that it is the avoidance of medial trapping that determines the choice between -u and -s. This means that for any given verb of the type under consideration (i.e., conjugation II with -u or -s perfect), the input, consisting of a Root and a perfect morpheme, is associated with candidates with different allomorphs; thus, for monēre, mon-u-ī is competing with (among other things) mon-s-ī. The constraint or constraints that disfavor medial trapping or its equivalent (i.e., a trimoraic trochee) do the rest, effectively selecting one allomorph and rejecting the other:

  1. (31)

    1. a [monu]〈ī〉

    2. b *[au]gu〈ī〉 (trapping)


Mester does not formalize the competition, but explains the intuition guiding his analysis by remarking that “a lexical selection process ...is driven by a prosodic criterion choosing the best among several alternatives’’ (1994, 46). The u-perfect is given “default” status, appearing where selection plays no role; for Mester, this is the case with verbs that he classifies as denominal, which appear with the u-perfect without regard to Root phonology.

To this point, the proposal looks exactly like many of the cases discussed above. When we consider entire sets of inflected perfects, moreover, it is possible to see the strongest predictions of the globalist view. Consider, to begin with, the two types of verbs analyzed above, inflected (p.175) for the perfect indicative; for reference, the metrical structure of the output is presented:

  1. (32) Types: Perfect indicative active


    Light Root

    Metrical structure

    Heavy Root

    Metrical structure









































The crucial form to consider in (32) is the 1pl of the heavy verb augēre.14 This appears with the s-perfect, which, in combination with the 1pl agreement morpheme, results in the configuration with medial trapping [σ¯]σσ¯. This point is crucial because of the competition logic that underlies the optimization approach to allomorphic selection. Within this kind of theory, the medial trapping perfect with -s is generated and compared with other possible perfects. The u-perfect for augeēre, auguimus has the metrical structure σ¯σσσ¯ and, according to Mester’s assumptions, can be exhaustively parsed: [σ¯][σσ]σ¯. Thus, if prosodic well-formedness is really the driving factor in selecting allomorphs, *auguimus should be grammatical, contrary to fact.

It would always be possible to appeal to the force of other constraints to account for the presence of -s, by invoking (e.g.) UNIFORM EXPONENCE constraints, as in Kenstowicz 1996 and related work. Such constraints enforce identical allomorphy across the different forms. However an analysis in these terms might be implemented, this type of solution subverts the strongest predictions of the globalist approach. Since the globalist theory allows the entire word’s phonological properties to be taken into account in determining the winner of the competition, it—unlike the localist theory—predicts that there should be cases of suppletive allomorphy that show allomorphic vacillation, where the chosen allomorph depends on outer, global properties. There is no vacillation, and the pattern found with Latin augēre is clearly compatible with the localist theory; it can, of course, be made compatible with the globalist theory, but it provides no arguments for that (more expressive) view.15

(p.176) The failure of the strong prediction is not restricted to 1pl perfects. The same point can be made for the pluperfect indicative, with a lot more force, it appears:

  1. (33) Pluperfect indicative of augēre



    Metrical structure

    -u form

    Metrical structure































In the perfect indicatives in (32), selection of the -s allomorph avoids medial trapping for heavy verbs like augēre, except in the 1pl. In the pluperfect, the selection of the -s allomorph creates trapping configurations in the entire paradigm of inflected forms. Crucially, these trapping configurations are not created by the u-perfect, where the light syllable with -u can be footed across the board. If all the different host-allomorph combinations were generated, with the phonology selecting the winner on the basis of metrical felicity—that is, if the strong predictions of Mester’s theory were correct—this pattern would not be found; pluperfects with these stems should show -u. However, the -s forms are the grammatical ones.

In sum, something must be added to a globalist theory in order to make prosodic optimization the determining factor for allomorphy in only some contexts. It should be clear by this point that such an addition would not compromise such a theory directly. As noted above, it would be possible to posit additional constraints to ensure that augsimus wins the competition; for example, a constraint requiring that allomorphy be held constant for a particular root could be ranked above the constraints that enforce prosodic well-formedness.16 ı® This would penalize *auguimus for taking a different allomorph from the rest of the paradigm, so that the prosodically worse augsimus would then win. The fact that the globalist theory can be altered in this way or other ways to yield the correct output is not really what is at issue, however. If there were cases in which something like auguimus did surface because of global prosodic considerations, then it would be a clear argument in favor of a globalist theory. As with other examples shown above, it is clear exactly what sort of effect (p.177) would be a strong argument for globalism in the Latin case, and we do not see such effects.

6.3.2 Generalizations about Latin Perfect Allomorphy

While the prospects for a globalist approach to Latin perfect allomorphy look quite unpromising, there are important generalizations about this system that relate directly to themes developed throughout this book.17 As a first step, consider the classification of Latin verbs in (34), which divides the verbal system into conjugation classes and shows the theme vowel that is found in each class:18

  1. (34) Conjugations and theme vowels



    Theme vowel



















It will be assumed here that the theme vowel is the Spell-Out of a head Th, attached to the v head in the PF component (Oltra-Massuet 1999). The reason for approaching the perfect in terms of conjugation is that there are basic associations between conjugation class and what happens in the perfect. Putting aside various readjustments that apply to the stem, there are two pieces of information that are central to these patterns: first, whether or not there is a theme vowel in the perfect form; and second, what allomorph of the head Asp[perf] appears, -vi, -si, or -i (henceforth, these forms are used rather than -u and -s; recall the discussion of chapter 3). The basic associations between conjugation and perfect type are as follows (here and below, I use orthographic -v- in the exponent of Asp[perf] that has both vowel and glide surface forms):

  1. (35) Perfect type by conjugation: Basic associations


    Perfect type


    thematic with -vi


    athematic with -vi






    thematic with -vi

(p.178) The associations are “basic” in the sense that most verbs in the relevant conjugations behave accordingly. At the same time, there are departures from these norms. The following chart summarizes attested patterns:19

  1. (36) Perfect types by conjugation







































































































































Despite the large number of filled cells in this chart, which suggests a highly disorderly pattern, the formation of the perfect is, by and large, determined systematically by conjugation class.20 My analysis of these patterns builds directly on the idea that aspects of perfect formation, in particular whether or not a Root is thematic or athematic in the perfect, are correlated directly with conjugation class features. In particular, all verbs of conjugations II, III, and III(i) are athematic in the perfect, along with a handful of verbs from conjugations I and IV (list in (37)). For concreteness, I assume that there is a rule that deletes (or does not assign) the Th node to such verbs in the perfect:

  1. (37) Athematic perfect rule

    ν is athematic / {[II][III][III(i)]LIST}Asp[perf]
    LIST = {Crep,Ven,}

(p.179) Simply listing the conjugation features in this manner might seem arbitrary, but it is more or less necessary. There is no overarching generalization that unites the verbs of conjugations II, III, and III(i). There is, moreover, no generalization that unifies conjugations I and IV, those conjugations that are by default thematic in the perfect. This means that the information regarding the presence or absence of a theme in the perfect must be stated in terms of processes that refer to the conjugation features [II], [III], and [III(i)], along with the additional Roots from the other conjugations.

The presence or absence of a theme vowel interacts with the second aspect of perfect formation, the allomorphy of the head Asp[perf]. Here the generalizations are as follows:

  1. 38 Generalizations about perfect formation

    1. (a) Perfects with -si- are always athematic.

    2. (b) Perfects with -i- are always athematic.

    3. (c) If there is a theme vowel in the perfect, it is

      1. always long (i.e.,-ā- or -ī-);

        always followed by the -vi exponent of Asp[perf].

These generalizations are accounted for with the following VIs:

  1. (39)


  2. Asp-i/{SedVertCapVen,}__

  3. Asp ↔ -vi

In these VIs, -si and -i require particular Roots to be inserted. Significantly, the rules for inserting these exponents only apply when the Asp[perf] node is linearly adjacent to the Root. In this way, the insertion of these exponents can only take place in athematic forms. Beyond this, the system defaults to the insertion of -vi. This VI does not have a list associated with it. It will be inserted in environments in which (i) the Root is adjacent, but not on the list for either -si or -i (athematic formation), or (ii) the Root is followed by a theme vowel—either -ā- or -ī-);

In short, there are important generalizations about allomorphy in the Latin perfect: generalizations that take into account local relations, in the way predicted by the theory of part I.

6.3.3 Latin Verbs of Conjugations III/III(i)/IV

A long-standing question in Latin morphology and phonology concerns the behavior of two classes of verbs in the language that, because they (p.180) have 1sg forms that end in -, are often simply referred to as -io verbs. The notable property of these verbs is that they fall into two types, which can be seen in other verb forms: those with a short theme vowel, like capĭmus ‘take, etc.’, and those with long -ī-, like audīmus ‘hear’; I use 1pl forms here because some other forms involve morphophonological rules that obscure this basic pattern. The capimus class—henceforth, conjugation III(i)—is quite small, consisting of fewer than twenty verbs, while the -ī-class—conjugation IV—is very large.

The traditional literature has faced in many forms the question of how these classes are related to one another, since there are clear diachronic connections. The typical approach is to try to derive (in the historical sense) the verbs of conjugation III(i) from what were earlier conjugation IV verbs—that is, to account for theme vowel shortening with a subset of conjugation IV verbs, in a way that eventually became ‘‘morphologized.’’

A point often discussed in such accounts is that there is a phonological subregularity unifying the verbs of III(i): their stems are light. This correlation is potentially enlightening, and many traditional works have sought to derive -ī- shortening as a metrical effect, with varying degrees of success.21

The traditional accounts mentioned above are interested in the historical relationship between these classes of verbs. Mester’s (1994) analysis goes beyond the historical and pushes the quantity differences in the theme vowel into the synchronic grammar; his position is that the III(i) and IV groups show ‘‘underlying unity,’’ because ‘‘for primary verbs the quantity of the theme vowel is to a large extent predictable from the prosodic pattern of the root’’ (1994, 24). The unified approach is implemented with a ‘‘single’’ theme vowel at a morphological level; this single morphological object has two allomorphs (1994, 26):

  1. (40) Theme vowel /i/

    1. (a) Primary allomorph: -ī-

    2. (b) Secondary allomorph: -ī-

For verbs that belong to either conjugation III(i) or conjugation IV (minus certain exceptions, such as those that are denominal), selection is determined by prosody: ‘‘the secondary allomorph -ĭ- is chosen.. .in situations where short quantity results in more optimal prosodic organization’’ (1994, 26–27). Mester illustrates the effects of this selection along the lines of (41), for audīre ‘hear’, aperīre ‘uncover’, and capere ‘take’:

  1. (41) Host-allomorph selection by phonology























(p.181) Mester seeks additional evidence for prosodic selection elsewhere in the verbal system—in particular, in effects found with unprefixed and prefixed verbs, where, in the cases he discusses, there appears to be an alternation in theme vowel length (1994, 27–28). These cases are important in light of the discussion above, since, if one adopts the spirit of the proposal under consideration, changes in a theme vowel’s quantity driven by the addition of a prefix could constitute an instance of allomorphic vacillation:

  1. (42) Prefixation

    No prefix

    Light prefix

    Heavy prefix















Taking the proposal as a whole, there are different ways to approach its predictions. One of the basic tenets of the theory is that the -ī- theme of conjugation IV verbs appears because of metrical optimization. Thus, the theory predicts allomorphic vacillation in other forms of the same verb, in a way that depends on the phonology of outer morphemes (e.g., Tense and Agreement). Here there are fewer cases than there were with conjugation II perfect allomorphy, but there is at least one case where a prediction is made: the 2pl passive, where the agreement ending is disyllabic. This is shown with long and short i in (43):

  1. (43) 2pl present indicative passives

    1. a. audī-minī [σ¯][σ¯]σσ

    2. b. audĭ-minī  [σ¯][σσ]σ

Clearly, the (43b) form should be selected, because the (43a) form traps a light syllable. Again, though, this is not what is found; as with perfect allomorphy, there is no vacillation.

(p.182) An extension of this type of prediction is behind Mester’s take on the prefixed verbs in (42). The idea is that the verb forms differ only in the quantity of the prefix, and, when this can result in suboptimal footing as in (42b), the -ī- allomorph wins out over the expected -ī- theme. The general prediction here is as follows:22

  1. (44) Optimization prediction

  2. Verbs of conjugation III(i) when prefixed by a single light syllable should switch to the -ī- theme.

  3. Rationale: [σσ][σ¯]σ is better than σ[σσ]σ.

While this prediction is supposed to account for pairs like those in (42), it does not generalize. For many of the conjugation III(i) verbs, there are examples with the light prefix re- employed in (42); none of these show the predicted change in theme vowel:

  1. (45) re-prefixed verbs with -ī-


    ‘take, etc.’




    ‘make, etc.’


    ‘make again’




    ‘dig again’


    ‘step, walk’


    ‘go/come back’

In these verbs, the theme vowel is the same in the unprefixed and the prefixed forms. While many things are going on in Latin prefixed verbs, the prediction in (44) is not borne out.

What, then, can be said about the cases adduced by Mester in (42)? The triplet sap-ĭ-mus, re-sip-ĭ-mus, dē-sip-ĭ-mus from (42) is taken by Mester to be ‘‘particularly telling,’’ since the same Root is involved (historically, in any case). Here the facts are simply unclear, for the re-prefixed form in particular. Lewis and Short’s (1969) Latin dictionary shows an infinitive in -ere, which means that it is treated as a verb of conjugation III(i). For the - theme that his argument is based on, Mester cites Niedermann 1908. Niedermann includes the form in a footnote, attributing it to the post-Classical grammarian Charisius; there is, moreover, apparently a text by Charisius in which the vowel is short. As far as I know, there is no other evidence than this for a long-vowel form.

This leaves the verb reperīre ‘find, discern’, which is (at a minimum, historically) related to conjugation III(i) parere and shows the -ī- theme expected on Mester’s account. Given the facts adduced above, this single form is certainly not evidence in favor of the globalist theory. The putative base of this prefixed form, parere, means ‘bear, beget’. It is possible that despite the historical connection, speakers did not analyze these (p.183) forms as possessing the same Root.23 Whatever there is to say about this single case, the point that it provides no argument for the globalist theory is clear.

Overall, the facts adduced by Mester that would support the predictions of the globalist theory are at best isolated and sporadic. The clear predictions of the theory—those that would support the globalist view and show that the localist view is problematic—are not found.

6.4 Paradigm (Non)uniformity: Outward-Sensitivity Redux

It was noted above that the absence of allomorphic vacillation is something that globalist theories have no trouble in modeling. Thus, the argument is not that globalist theories are incapable of accounting for the attested facts. Rather, the point is that the strongest predictions of such theories do not appear to be borne out; moreover, the attested patterns of allomorphy are accounted for in a localist theory like that of part I.

One way of sharpening the line of argument from the last section is by considering what kinds of factors could rule out allomorphic vacillation in a globalist framework. The most obvious way of doing this would be in terms of PARADIGM UNIFORMITY: a constraint (or set of constraints) that ensures that a Root shows consistent allomorphy throughout its set of surface forms. It could be argued, for example, that the reason that -s-allomorph-taking augēre does not switch from -s to -u in the 1pl is that the constraints enforcing uniform realization of Asp[perf] outrank the constraints responsible for driving allomorph selection prosodically. Assuming that something like PARADIGM UNIFORMITY could produce the correct results, there are two points to be made.

The first point was stressed above: it might be possible for a globalist theory to appeal to PARADIGM UNIFORMITY, but what the paradigmatic constraints do, in effect, is rule out the cases in which the strongest predictions of a globalist system can be seen. Thus, while the resulting theory might make globalist assumptions, it is certainly not an argument for those assumptions. In the absence of any other arguments for global interaction, there is no reason to have a theory that is global but restrained by PARADIGM UNIFORMITY in the first place. However, if paradigms always are uniform, then the arguments of the last section about allomorphic vacillation might lose some of their force. If this type of vacillation is universally ruled out, then the absence of vacillation cannot argue against globalism.

(p.184) These considerations lead up to the second point, which connects the predictions of the ℂ1-LIN theory with this discussion. The idea sketched above, appealing to PARADIGM UNIFORMITY, can be taken to the limit: if the uniformity constraints always dominate the constraints that would force a change of allomorphs for phonological reasons, there should never be ‘‘outward-looking’’ paradigmatic vacillation. At this point, it is important to recall that there is outward-sensitive allomorphy. This was seen in chapter 2 in cases like the Hungarian plural/possessive interaction repeated in (46), and in chapter 3 with the suppletive Latin verb esse ‘be’, repeated in a condensed form in (47):

  1. (46) Hungarian plural/possessive (Carstairs 1987, 165)


    Singular-1sg Poss


    Plural-lsg Poss

















  2. (47) Allomorphy of Latin esse ‘be’




























The Hungarian plural morpheme and the Latin vbe head that is ‘be’ each show outward-sensitive contextual allomorphy and thus ‘‘nonuniform’’ paradigms. As discussed in chapters 2 and 3, these effects are conditioned by adjacent nodes, not by phonological properties of entire words.

The fact that nonuniform paradigms are found, but in a way that shows sensitivity to local factors, is important. Many well-known cases of suppletion have the same general properties seen with Latin esse. If all cases of suppletion (and outward-sensitive allomorphy in general) are conditioned locally, as seems to be the case, then PARADIGM UNIFORMITY cannot be invoked to rescue the globalist theory.

There is at least one case in which it has been claimed that ‘‘outer’’ or surface phonology conditions stem suppletion. The example is the verb andare ‘go’ in Italian. Carstairs (1988) and others have followed traditional discussions of Italian by describing the alternation between (p.185) and- and va(d) -in phonological terms. The pattern is that the stem is va(d)- when under stress, and and- otherwise24

  1. (48) Forms of andare


    Present indicative

    Present subjunctive


















For Burzio (1998), among others, the correlation between stress and suppletion in (48) implies causation in the synchronic grammar: Burzio argues that these facts support a globalist view, with surface phonological properties determining the choice between va(d)- and and-. While the description in terms of stress is correct, on the face of it, this cannot play a role in the analysis in the theory presented in part I, since the output phonology cannot determine earlier Vocabulary Insertion.

Since the suppletion can also be characterized in morphosyntactic terms (the default and- appears in 1pl and 2pl present indicative, subjunctive, and imperatives instead of va(d)-) , an analysis in which φ-features trigger suppletion can be given. As a result, the basic distributional pattern seen in (48) can be stated in either type of theory.25 An important point is that there appears to be no way to look at predictions of the stress-based account beyond the facts in (48): in Standard Italian, there is no way to shift the stress in these forms to create forms in which allomorphic vacillation is predicted to occur.26 As a result, there is no possibility of really testing the hypothesis that the surface position of stress drives stem choice; any claim to the effect that surface stress must be referred to in deriving the allomorphic pattern can be based only on conceptual arguments. Thus, this case is clearly analyzable with globalist assumptions, but it provides no arguments for a framework of that type.27

The conclusions to be drawn from this review of outward sensitivity and nonuniform paradigms are significant. There are cases in which constraints like PARADIGM UNIFORMITY do not apply; that is, changing allomorphs is not ruled out across the board. In cases where allomorphs do change, the strong predictions of globalist theories, with nonlocal factors determining allomorphic selection, should therefore be seen. Critically, (p.186) though, the attested cases of outward-sensitive allomorphy show sensitivity to local nodes, in the way predicted by the ℂ1-LIN theory. When allomorphic vacillation does occur, the vacillation is not triggered by the global phonological context. The strong predictions of global-MP are not found; appealing to PARADIGM UNIFORMITY does not help.

Overall, then, the point is not that outward-sensitive allomorphy does not occur; it does. However, the conditions under which it occurs are not those predicted by a globalist theory. Another way of putting this is that paradigmatic vacillation does exist. However, it operates in ways that reflect the cyclic and linear restrictions of the theory developed in part I: it is driven by local morphemes, not by the phonology of outer morphemes, nor by the phonology of the whole (output) word.

6.5 Summary

The empirical predictions of globalist theories are straightforward. If such theories are correct, there should be cases in which allomorph selection is determined by global phonological properties, in a way that cannot be stated in a localist theory.

As a general point, theories with even limited amounts of global interaction between morphology and phonology predict over and underapplication in allomorphy. In empirical test-cases like the Latin perfect and -io verbs, the theory that surface phonology drives allomorphy predicts allomorphic vacillation with certain ‘‘outer’’ morphemes. This is not found. One possible response to this would be to attribute the nonvacillation to paradigm uniformity effects. However, in cases where stem suppletion or outward-sensitive allomorphy is found, (i) paradigm uniformity does not hold, but (ii) there is still no evidence for the predictions of globalism over localism.

The conclusion that must be drawn from these arguments is that there is no evidence for the strong predictions of the globalist framework. In the cases that have been studied in the literature, the patterns that are found are those expected from a localist, cyclic point of view. It is significant to note that these cases are not arguments for ‘‘hybrid’’ theories like cyclic or stratal OT; rather, such theories predict that global interaction should occur within a given stratum, and there is no evidence that this is correct.


(1.) It is for this reason that in this and other cases, many OT analyses have moved toward paradigmatic resemblance with other morphological forms. That is, if there are no phonological reasons why a form should be as it is, then the reasons must be morphological in nature. As noted in chapter 1, such theories are clearly incompatible with the localist theory of morphosyntax argued for in part I. See also Bobaljik 2002, 2008.

(2.) This is one way of putting it; it would also be possible to say that the notion of local conditioning environment is immaterial, or derivative, or epiphenomenal, or the like, in such theories.

(3.) The rule is simplified, in that it actually applies only with certain affixes; see Inkelas and Orgun 1995 for discussion.

(4.) An alternative is to posit a single VI with the exponent -SI, and some additional rules to delete the consonant under specific circumstances; see below.

(5.) Here and below, matters related to the vowel component of -(S)I are ignored.

(6.) In Turkish’, the effects of Velar Deletion would be seen only with morphemes that have no C-initial allomorphs.

(7.) So if Z were an Agr node, Z1 and Z2 would represent different combinations of person/number features, for example.

(8.) (p.202) Fulmer (1997) reports that certain affixes behave similarly in Afar.

(9.) There are other combinations about which the same type of point could be made. For example, consider an odd-syllabled host that takes both the passive affix and one of the alternating agreement morphemes: Root-PASS-AGR. An even-syllabled output could be derived by inserting the (locally unconditioned) disyllabic passive affix, then a monosyllabic outer affix. As far as I am aware, this does not occur.

(10.) In this section, I refer to the perfect allomorphs with -u and -s, as in Mester’s discussion. In section 6.3.2, I refer to them as -vi and -si, as in chapter 3.

I am putting aside Reduplication, along with various stem-changing processes that apply in the perfect. These cases can be treated as taking the -i exponent of Asp[perf ], in which there is, in addition, action in the form of readjustment rules.

(11.) Mester (1994, 47) excludes from consideration verbs that he classifies as ‘‘denominal,’’ which do not take the allomorph expected on prosodic grounds alone. For example, albēre ‘be white’ has a heavy stem, but shows the –u- perfect alb-u-ī.

(12.) Here, I put aside initial trapping—representations in which the initial syllable is unfooted—although see below.

(13.) Mester also considers ‘‘marked’’ trochees where σ¯σ is footed as [σ¯σ] instead of as [σ¯]σ. In the latter type of approach, the trimoraic trochee is what is avoided when possible.

(14.) The forms in (32) include three distinct entries for 3pl because three different agreement endings—namely, -ērunt, -ēre, and -ērunt—were in variation in this context. In principle, something about optimization could be learned from -erunt; compare monuerunt σ[σσ]σ¯ with augserunt [σ¯]σσ¯. See Sommer 1914, 579, for correlations between perfect allomorph and 3pl agreement endings that might be worth looking into in the context of Latin historical phonology.

(15.) Mester uses an argument based on putative allomorphic vacillation in his second case study from Latin -io verbs; see below in the text. The failure of perfect allomorphs to vacillate is not addressed.

(16.) For this type of ‘‘uniform exponence’’ approach to work properly in this particular case, the allomorph found in the perfect would have to be preferred to the one favored on metrical grounds in the pluperfect, presumably something that could be accomplished in terms of making the former less marked.

(17.) This analysis draws in part on joint work with Morris Halle, most of which (with the exception of Embick and Halle 2005) has not been published.

(18.) With the exception of the theme vowel in conjugation III, given here as /-ɨ-/, this is more or less uncontroversial (recall the comments in chapter 3). Verbs in this conjugation show an –i- theme vowel in certain person/number forms (e.g., dūc-i-t ‘he/she leads’), but, unlike with the conjugation III(i) type verbs, this vowel does not appear in 1sg forms such as dūc-ō. There are other options for the vowel here that have been explored in the literature (e.g., Lieber 1980). Since this particular assumption does not play a role in the analysis of the perfect, I will not say anything more about it.

(19.) (p.203) For conjugation I, the verb iuvāre ‘help’ has the perfect iuvī, which looks like an –i-perfect. The stem-final /v/ in iuvāre makes this case, an apparent instance in which a verb of conjugation I takes an –i- perfect, questionable at best. Conjugation II has some apparently thematic perfects: for example, flēri, flēvī ‘weep’. Aronoff (1994) argues (as does Ernout (1952/1989) that these verbs are not actually in conjugation II. Rather, they happen to end in /-ē-/. The argument is based on the fact that Roots are minimally CV. The suggestion is attractive in that it allows for a cleaner statement of the rules concerning the presence or absence of themes in these verbs in perfect and participial forms, which are then always athematic (though there are some /i/ vowels in participles; e.g., mon-i-tus for monēre).

Another pattern I am not taking into account here involves apparent ‘‘conjugation change.’’ For instance, pēto ‘seek’, with infinitive pet-e-re, seems on the basis of these two forms to be conjugation III, like dūcō; likewise for conjugation III(i) cupiō ‘desire’, with perfect cup-ī-v-ī. However, the perfect form is pet-ī-v-ī, evidently with the -ī- theme vowel that characterizes conjugation IV verbs like aud -ī-re. There are a handful of verbs that behave this way, all showing conjugation III or III(i) behavior in the present system and the –ī- of conjugation IV in the perfect.

Finally, in line with the exclusion of stem-changing processes, I have not included Reduplication as a separate class here, on the assumption that the reduplicated perfects are a subcase of the -i- affixed perfects.

(20.) See Aronoff 1994 for discussion of the fact that the perfect shows many systematic patterns, regarding in particular Lieber’s (1980) claims about the irregularity of the system of perfect formation.

(21.) There are some exceptions to the light-stem pattern. Light-stem verbs that end in liquids are in conjugation IV, not III(i).

(22.) Another prediction is that verbs like venīre that have a long theme despite having a light root syllable should show a short theme when these verbs have a heavy prefix. There are two problems here. The first is that Mester offers no explanation for why these verbs should ever surface with –ī- instead of -i- in the unprefixed forms in the first place. The second is that this additional prediction is not verified.

(23.) It is true that there are many cases in Latin where theme vowels differ in prefixed and unprefixed verbs: examples are pellere ‘push’, compellāre ‘summon’; spernere ‘remove’, aspernārī ‘reject’; capere ‘take’, occupāre ‘seize’; specere ‘look at’, suspicārī ‘mistrust’ (Sommer 1914, 507ff.). These differ in themes and deponency, and they raise questions about when two forms may be said to contain the same Root, as well as other questions about morphophonology. But whatever there is to say about such cases, they offer no support for a globalist theory of morphology and phonology.

(24.) Carstairs (1987, 179ff.) looks at some additional cases of allomorphy that are putatively ‘‘outward-sensitive’’ to phonological properties. These cases do not appear to be fully suppletive; that is, it looks like the majority involve morphophonological rules, not competition for insertion, and thus are not directly relevant to the issue at hand.

(p.204) The central cases (Carstairs 1987, 185ff.) come from Fula and are based on work by Arnott (1970) and McIntosh (1984). One case involves affixes that differ between ‘‘short’’ and ‘‘long’’ forms: anterior –noo/-no, relative past passive -aal-a, and relative past middle -iil-i. The factor conditioning the alternation is phonological, and the alternation itself is clearly not suppletive. The other case is found with the habitual imperative singular suffix, which is typically -atay. In the first person singular, this morpheme surfaces as -at. While there are some phonological correlates of this (the 1sg affix follows the habitual imperative morpheme and is the only vowel-initial agreement morpheme), the alternation is not necessarily suppletive; moreover, it can be stated with reference to the 1sg features, so that the phonological effect is incidental.

(25.) The same pattern of features is required elsewhere in the language’s verbal system. As Carstairs (1988, 1990) discusses, the morphosyntactic pattern seen in (48) is found with other verbs, where it conditions, for example, insertion of the ‘‘infix’’ –isc- (e.g., the verb ‘finish’ has 1sg fin-ίsc-o but 1pl fin-ίamo): on the face of it, -isc- does not appear with stressed affixes.

Maiden (2004) presents a detailed study of such stem alternations in Romance, concentrating on whether particular patterns of paradigmatic distribution of ‘‘stems’’ call for a morphological (versus, say, phonological) treatment. Looking at patterns like that seen in Italian, he presents arguments (2004, 159ff.) against the view that surface placement of stress must be referred to in these patterns of stem allomorphy. See also Corbett 2007, 22, for related discussion.

(26.) I thank Andrea Calabrese for discussion of this and related points.

(27.) The need to look for vacillation in this system is touched on by Kiparsky (1996, 25), who cites comments by Wolfgang Dressler in a discussion period (see also Maiden 2004, 161). It appears that the word ándirivieni ‘coming and going’, where secondary stress appears on and-, does not conform to an analysis in which stress drives allomorphy. However, this form might not be probative, since it is not clear what its synchronic relationship to va(d)-land -is.

In some dialects of Italian, stress shift can be induced by encliticization. As far as I am aware, there is no evidence that stem suppletion with ‘go’ vacillates in such cases.