Paradigms in Phonological Theory

Title Pages
University Press Scholarship Online
Oxford Scholarship Online
Paradigms in Phonological Theory

Laura J. Downing, T. Alan Hall, and Renate Raffelsiefen
Print publication date: 2004

Print ISBN-13: 9780199267712
Published to Oxford Scholarship Online: February 2010
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Title Pages
(p.i) Paradigms in Phonological Theory
(p.ii) Oxford Studies in Theoretical Linguistics
(p.iii) Paradigms in Phonological Theory
General editors
David Adger, Queen Mary College London;
Hagit Borer, University of Southern California
Advisory editors
Stephen Anderson, Yale University; Gennaro Chierchia,

University of Milan; Rose-Marie Dechaine, University of
British Columbia; Elan Dresher, University of Toronto; James
Higginbotham, University of Southern California; Pat Keating,
University of California, Los Angeles; Ruth Kempson, School of
Oriental and African Studies, University of London; James
McCloskey, University of California, Santa Cruz; Gillian
Ramchand, University of Oxford; Maria-Luisa Zubizarreta,
University of Southern California
This series provides a forum for cutting-edge work in

theoretical linguistics. Its focus is on the interfaces between
Page 1 of 5
Title Pages
the subcomponents of grammar and between grammar and

other components of the mind.
PUBLISHED
1. The Syntax of Silence
Sluicing, Islands, and the Theory of Ellipsis
by Jason Merchant
2. Questions and Answers in Embedded Contexts
by Utpal Lahiri
3. Phonetics, Phonology, and Cognition
edited by Jacques Durand and Bernard Laks
4. The Syntax–Pragmatics Interface
Concept Formation and Verbal Underspecification in Dynamic

Syntax
by Lutz Marten
5. The Unaccusativity Puzzle
Explorations of the Syntax–Lexicon Interface
edited by Artemis Alexiadou, Elena Anagnostopoulou, and

Martin Everaert
IN PREPARATION
Tense, Mood, and Aspect
edited by Alessandra Giorgi, James Higginbotham, and Fabio

Pianesi
The Ecology of English Noun–Noun Compounding
by Ray Jackendoff
The Oxford Handbook of Linguistic Interfaces
edited by Gillian Ramchand and Charles Reiss
[published in association with the series]
Page 2 of 5
Title Pages
(p.iv)
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the

University of Oxford.
It furthers the University’s objective of excellence
in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dares Salaam Hong Kong

Karachi
Kuala Lumpur Madrid Melbourne Mexico City
Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic

France Greece
Guatemala Hungary Italy Japan South Korea
Poland Portugal
Singapore Switzerland Thailand Turkey Ukraine
Vietnam
Oxford is a registered trade mark of Oxford

University Press
in the UK and in certain other countries
Published in the United States

by Oxford University Press Inc., New York
© Editorial matter and organization

Downing, Hall, and Raffelsiefen
© The several contributors and in this collection,

Laura J. Downing, T. A. Hall, and Renate
Raffelsiefen, 2005
The moral rights of the authors have been

asserted
Database right Oxford University Press (maker)
First published 2005
Page 3 of 5
Title Pages
All rights reserved. No part of this publication

may be reproduced,
stored in a retrieval system, or transmitted, in
any form or by any means,
without the prior permission in writing of Oxford
University Press,
or as expressly permitted by law, or under terms
agreed with the appropriate
reprographics rights organizations. Enquiries
concerning reproduction
outside the scope of the above should be sent to
the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other

binding or cover
and you must impose the same condition on any
acquirer
A catalogue record for this title is available from

the British Library
Library of Congress Cataloging in Publication

Data
Paradigms in Phonological Theory /

edited by Laura J. Downing, T. A. Hall, and Renate
Raffelsiefen.
p. cm. – (Oxford studies in theoretical linguistics;
8)
Includes bibliographical references and index.
I. Grammar, Comparative and general–Verb 2.
Grammar,
Comparative and general–Phonology 3.
Morphology. I. Downing, Laura J.
II. Hall, T. A. III. Raffelsiefen, Renate. IV. Series.
P381.U53 2003 415’.6–dc22 2003060967
ISBN 0–19–926771–5 (pbk)

ISBN 0–19–926770–7 (hbk)
1 3 5 7 9 10 8 6 4 2
Typeset in Adobe Garamond

by Peter Kahrel Ltd., Lancaster
Page 4 of 5
Title Pages
Access brought to you by:
Page 5 of 5
Oxford Studies in Theoretical Linguistics General Preface


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
(p.vii) Oxford Studies in

Theoretical Linguistics General
Preface
The theoretical focus of this series is on the interfaces
between subcomponents of the human grammatical system
and the closely related area of the interfaces between the
different subdisciplines of linguistics. The notion of ‘interface’
has become central in grammatical theory (for instance, in
Chomsky’s recent Minimalist Program) and in linguistic
practice: work on the interfaces between syntax and
semantics, syntax and morphology, phonology and phonetics
etc. has led to a deeper understanding of particular linguistic
phenomena and of the architecture of the linguistic component
of the mind/brain.
The series will cover interfaces between core components of

grammar, including syntax/morphology, syntax/semantics,
syntax/phonology, syntax/pragmatics, morphology/phonology,
phonology/phonetics, phonetics/speech processing, semantics/
pragmatics, intonation/discourse structure as well as issues in
the way that the systems of grammar involving these interface
areas are acquired and deployed in use (including language
acquisition, language dysfunction, and language processing).
It will demonstrate, we hope, that proper understandings of
Page 1 of 2
Oxford Studies in Theoretical Linguistics General Preface
particular linguistic phenomena, languages, language groups,

or inter-language variations all require reference to interfaces.
The series is open to work by linguists of all theoretical

persuasions and schools of thought. A main requirement is
that authors should write so as to be understood by colleagues
in related subfields of linguistics and by scholars in cognate
disciplines.
We are pleased to present the eighth volume in the series,

which is a collection of papers on the notion of paradigm
uniformity—the levelling effect morphological paradigms have
on the phonology of their individual members. This
phenomenon has become important in the development of new
theoretical approaches within Optimality Theory and the
volume illustrates the explanatory potential of these
approaches across a range of languages. The collection aims
to provide a theoretical and empirical grounding for future
work on the interrelationship between paradigms and
phonological representations.
David Adger
Hagit Borer
Page 2 of 2
Abbreviations


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
(p.viii) Abbreviations
ACL
Association for Computational Linguistics
BLS
Proceedings, Berkeley Linguistics Society
CLS
Proceedings, Chicago Linguistics Society
DM
Distributed Morphology
ESCOL
Eastern States Conference on Linguistics
GA
Gradient Attraction
GTT
Generalized Template Theory
MA
Metrical Alignment
MC
Metrical Consistency
MWF
Metrical Well-Formedness
NDEB
Non Derived Environment Blocking
NELS
Proceedings, Northeastern Linguistics Society
Page 1 of 3
Abbreviations
OP
Optimal Paradigms
OT
Optimality Theory
PC
Paradigm Contrast
PES
Principles of English Stress (Burzio 1994a)
PU
Paradigm uniformity
REH
Representational Entailments Hypothesis
ROA
Rutgers Optimality Archive
SPE
Sound Pattern of English (Chomsky and Halle 1968)
SSG
Sonority Sequencing Generalization
SWP
Stress to Weight Principle
TCT
Transderivational Correspondence Theory
UE
Uniform Exponence
UR
Underlying Representation
VOT
Voice onset time
WCCFL
West Coast Conference on Formal Linguistics
WSCLA
Workshop on the Structure and Constituency of the
Languages of the Americas
Page 2 of 3
Abbreviations
Page 3 of 3
Contributors


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
(p.ix) Contributors
Adam Albright, Massachusetts Institute of Technology
Outi Bat-El, Tel-Aviv University
Luigi Burzio, Johns Hopkins University
Stuart Davis, Indiana University
Laura J. Downing, Centre for General Linguistics, Typology,

and Universals Research (ZAS), Berlin
T. A. Hall, Indiana University
Michael Kenstowicz, Massachusetts Institute of Technology
John J. McCarthy, University of Massachusetts Amherst
Renate Raffelsiefen, Freie Universität, Berlin
Péter Rebrus, Research Institute for Linguistics, Hungarian

Academy of Sciences; Theoretical Linguistics Program, Eötvös
Loránd University, Budapest
Miklos Törkenczy, Research Institute for Linguistics,

Hungarian Academy of Sciences; English Linguistics
Department, Eötvös Loránd University, Budapest; Theoretical
Linguistics Program, Eötvös Loránd University, Budapest
Page 1 of 2
Contributors
Suzanne Urbanczyk, University of Victoria (p.x)
Page 2 of 2
Introduction: The Role of Paradigms in Phonological Theory


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Introduction: The Role of Paradigms

in Phonological Theory
Laura J. Downing (Contributor Webpage)
T.A. Hall (Contributor Webpage)
Renate Raffelsiefen (Contributor Webpage)
DOI:10.1093/acprof:oso/9780199267712.003.0001
Abstract and Keywords

This introductory chapter provides a brief historical overview
of the sorts of phonological exceptions that paradigm
uniformity (and contrast) have been proposed to explain.
Section 1.2 discusses the role of paradigms in pre-generative
work. Section 1.3 shows how paradigm uniformity effects are
captured in traditional rule-based approaches, i.e. in terms of
rule ordering and the cycle. Section 1.4 presents a brief
comparison of various approaches to paradigm uniformity (and
contrast) in the Optimality Theory (OT) framework. Since most
of the chapters here are written within that framework, their
contribution to current theories of the role of paradigms is
discussed in Section 1.4 as well.
Keywords: phonological exceptions, paradigm uniformity, contrast, pre-

generative work
1.1 INTRODUCTION
Page 1 of 34
Since the Neogrammarians’ work on sound change in the

nineteenth century, it has been recognized that many
exceptions to regular phonological processes (or sound laws)
can be explained by proposing that paradigms of
morphologically related words influence each other’s
pronunciation. In this introduction we provide a brief
historical overview of the sorts of phonological exceptions that
paradigm uniformity (and contrast) have been proposed to
explain. In s. 1.2 we discuss the role of paradigms in pre-
generative work. Section 1.3 shows how paradigm uniformity
effects are captured in traditional rule-based approaches, i.e.
in terms of rule ordering and the cycle. Section 1.4 presents a
brief comparison of various approaches to paradigm
uniformity (and contrast) in the OT framework. Since most of
the chapters here are written within that framework, their
contribution to current theories of the role of paradigms is
discussed in s. 1.4 as well.
Page 2 of 34
(p.2) 1.2 THE STATUS OF PARADIGMS IN PRE-

GENERATIVE WORK
As is well known, the main purpose of European pre-
Structuralist linguistics was the reconstruction of (unique)
words in the Indo-European ancestor language on the basis of
variation in attested cognates. In the wake of attempting to
develop a rigid scientific method for determining these proto-
forms, the Neogrammarians came to devote considerable
attention to identifying the various Triebe ‘forces’ potentially
causing change in the sound shape of words. On a particularly
constrained view such change had to be determined entirely
by the ‘dualistic’ and inherently conflicting forces of
physiologically motivated sound laws and psychologically
motivated analogical changes, including paradigm uniformity.1
Neogrammarians were keenly aware of the fact that this

proposal had scientific significance only in so far as the
respective notions of sound laws and analogical changes were
independently motivated and restrictive. A view considered to
be too restrictive by many was that sound laws necessarily had
to involve a reduction of articulatory effort, to save the speech
organs time and effort so as to maximize ‘economy’ (cf
Whitney 1867: 28). The most significant advance in this area
was the publication of Sievers’s Grundzüge der
Lautphysiologie in 1876, which established the physiological
underpinnings of the theory of sound change. The view of
sound laws as having a physiological basis fit well with the
doctrine of the exceptionlessness of sound change. This
particular doctrine was fundamental in sparking interest in
analogy, as putative exceptions to sound laws could be
dismissed if they lent themselves to an explanation in terms of
analogy.
As one might expect, opponents of the Neogrammarian

doctrine of exceptionless sound laws were particularly hostile
to the notion of analogy, which was ridiculed as a convenient
fad to deal with obvious counter-examples. To be sure, if
analogy can be invoked whenever a sound shape in some word
a comes to (partially) match the sound shape of some word b,
and the relatedness between a and b is unrestricted, the
empirical explanatory power of analogy is virtually nil. Critics
have insisted that analogical formation is in principle always
possible, but never necessary (Curtius 1885: 39). While the
latter is in fact also true of physiologically motivated sound
laws (cf. Tobler 1879), many agreed that recourse to analogy
Page 3 of 34
should be an ultimum refugium to the linguist, to be invoked

only if all else fails, and that such an analysis should be
‘happily’ abandoned for any alternative explanation, especially
a strictly phonological approach (cf. (p.3) Schmidt 1882;
Curtius 1885). While other researchers felt that granting a
fundamental priority to strictly phonological explanations was
not necessarily justified, they still deplored the lack of rigid
criteria to motivate the legitimate use of analogy (Scherer
1868:177).
Which criteria were then proposed to restrict the relation

between words such that analogy can be plausibly invoked?
Paul (1880: 106) asserts that words attract one another and
form groups in the mind of the individual, and the relevant
groups can be large or small, tightly connected or more
loosely organized, such that the associations may differ from
speaker to speaker. He further claims that there is evidence
for historical shifts such that words which at some point were
associated in a certain manner (for some idealized speaker of
that period) exhibit different types of association or none at
other stages.
Paul’s groups are not mutually exclusive: a larger group may

include several smaller groups and a word can simultaneously
be a member of several groups. He crucially distinguished
between material groups and formal groups. Material groups
include words associated foremost on the basis of a common
element of meaning, typically accompanied by form
correspondences resulting from etymological relatedness (e.g.
{{mother, mothers}, motherly, mothering} but not necessarily
(e.g. {mother, father}, {old, new}, {good, better}). Formal
groups include words exhibiting certain morphological
properties (e.g. all 3rd person singular forms {walks, sings,
writes, does, …}, all plural forms {dogs, cats, children,
women, …}, all comparative forms {older, cheaper, better, …},
where again groups typically include subgroups and a word
may belong to a range of distinct groups. Within material
groups inflectionally related words are claimed to be more
tightly connected than derivationally related words. Within
inflectional groups all verbs sharing a certain feature for tense
are more tightly associated with each other than with other
verbs, with which they may share number or person features.
Page 4 of 34
Regarding the preference for uniform paradigms, Paul

advanced a number of claims, as shown in (1).
(1)
a. The more tightly words are associated with
each other, the stronger the preference for
uniform paradigms (e.g. inflectionally related
words are more likely to be subject to leveling
than derivationally related words) (Paul 1880:
206).
b. The higher the token frequency of individual
members of a paradigm, the higher the
resistance to undergo paradigm leveling (e.g.
lack of paradigm uniformity in paradigms like
{do, does}, {am, are, is}, as opposed to {walk,
walks}, {blame, blames}) (Paul 1880: 208).
c. Alternations within paradigms which come to
be associated with grammatical functions resist
leveling (e.g. umlaut in German, which came to
serve as a plural marker).
d. The more alternations there are within a
paradigm, the higher the resistance to undergo
paradigm leveling.
The claims in (1) can lead to contradictory analyses, a problem

which Paul discussed (p.4) but did not resolve. Consider the
two typical examples for paradigm leveling in the history of
German illustrated in (2):
(2) (partial) preterite paradigm for sterben ‘to die’:

(partial) present tense paradigm for leben ‘to live’:
a. Stage I Stage II
1st sg starb 1st sg starb
1st pl sturben 1st pl sturben
b. Stage I Stage II
1st sg l[e:]be ‘lebe’ 1st sg l[e:]be ‘lebe’
2nd sg l[e]bst ‘lebst’ 2nd sg l[e:]bst ‘lebst’
3rd sg l[e]bt ‘lebt’ 3rd sg l[e:]bt ‘lebt’
Page 5 of 34
The question is how to model these changes. Phrased in

current terms, one approach would be to posit the respective
paradigms at stage I as input forms encountered in acquisition
and to rank the constraint requiring uniform paradigms above
both phonological constraints and faithfulness constraints.
Paul vehemently rejected such an approach, asserting that, for
example, the plural form sturben was not modified but in fact
was not in the picture at all when the change occurred. The
relevant change should be modeled as analogical creation by
learners who simply were not aware of the existence of
sturben. Such creations involve minimally three members of
formal and material groups, such that one word shares
membership with both types of groups, as is illustrated in (3).
(3)
a. Rad [rat] ‘wheel’
Rades [ra:d∂s] ‘wheel (gen. sg.)’
b. Rad [ra:t]
Rades [ra:d∂s]
While this approach makes sense of the generalization in (1b),

in that a word is the more likely to be ‘in the picture’ and block
analogical creations the more frequent it is, it encounters
problems with generalization (1d). Specifically, the fact that
leveling tends to be influenced by the degree of phonological
opacity in the relation between group members is unexpected
if speakers are unaware of the relevant forms. Yet, there is
strong evidence to support generalization (1d). One of the
relevant cases noted by Paul (1880: 203) concerns the vowel
quantity leveling in (2b), which is attested for all inflectional
verbal paradigms with either no qualitative vowel alternation
or with the common ‘a–ä ’ alternation (e.g. gr[a:]be, gr [ä:]bst,
gr [ä:]bt ‘dig’). By contrast, paradigms with other alternations
have often escaped leveling to this day (e.g. n[e:]me, n[i]mmst,
n[i]mmt} ‘take’; g [e:]be, g[i]bst, g[i]bt ‘give’). Clearly, this
generalization is not explained by the sort of rule given in (3).2
(p.5) When paradigmatic associations can shift over time

there is an additional possible explanation for leveling.
Conceivably, the change concerns not the reranking of some
constraint requiring paradigm uniformity with respect to other
constraints but rather the extension of the paradigm. This
Page 6 of 34
approach was suggested by Misteli (1880), who argues that in

the mind of the speakers there existed no past tense paradigm
for the verb sterben at stage I, but only at stage II.3
Apart from the question of determining the extension of

groups there is also the issue of explaining the direction of
leveling within some group. The property mentioned often in
this context is token frequency (cf Paul 1880: 207). Evidence
to support this generalization also supports the view of
paradigm uniformity as analogical creation illustrated in (3),
as more frequent words are presumably acquired first. Yet,
other factors were mentioned as well. For instance, it has
often been observed that the nominative serves as an attractor
when other case forms, especially the accusative, are leveled
(cf. Misteli (1880: 447) and references therein).
While the discussion concerning the question of possible

restrictions on leveling effects largely centered on the
extension of paradigms and the choice of the attractor, other
issues were addressed as well. Schuchardt (1885) claimed that
strictly physiologically motivated alternations such as
coarticulation are not susceptible to leveling. Schuchardt also
drew attention to the need to distinguish conceptual analogy,
based on paradigm membership, from phonetic analogy, which
describes the extension of a sound change to new
environments determined on phonological grounds (for
discussion, see Vennemann 1972). Others drew attention to
how additional Triebe in languages conflict with the two
aforementioned ones, like the preference for preserving the
given (Curtius 1885) and the preference for maintaining or
creating meaningful contrast within paradigms (Paul 1880;
Curtius 1885).
It goes without saying that since paradigms may differ with

respect to ‘tightness’ of association, may lose or gain members
over time, and distinct members of a paradigm can in principle
serve as the attractor determining the direction of leveling,
paradigm uniformity has continued to provide an easy target
for critics, who dismiss the approach as unscientific. It is clear,
however, that if the phenomenon of uniformity based on the
association of words in the mental lexicon exists, then it
cannot be dismissed on the grounds that the contributing
factors are too complicated. These points are developed in the
most important contributions in the post-Neogrammarian era:
Page 7 of 34
Kurylowicz’s (1947) ‘laws’ of analogy and Manczak’s (1958,

1978) ‘tendencies’ of analogical change.
(p.6) 1.3 THE STATUS OF PARADIGMS IN THE POST-

SPE PERIOD
In derivational generative phonology only inputs have
psychological reality. Outputs (i.e. surface forms) have no
formal status, but are just the ‘accidental’ by-product of
derivational rule application. Since paradigm uniformity (PU),
by definition, posits identity relationships among output forms,
relationships between related words in this framework can be
established only in terms of intermediate derivational stages:
rule ordering and the cycle. In this section we discuss how
several examples of PU effects are accounted for in rule-based
generative phonology, i.e. early generative grammar (SPE:
Chomsky and Halle 1968) and Lexical Phonology (Kiparsky
1982a, b).4
The example in (4) below illustrates how PU effects are

captured in the early generative (rule-based) framework of
King ( 1969: 51–4). In stage I of German (illustrated in (4a)
and also above in (2b)), we see that simplex words like Rad
differed from morphologically related ones like Rades in terms
of the length of the stem vowel. In the simplex form the vowel
was short, and in the morphologically complex one it was
long.5 At stage 2 (illustrated in (4b)) we see that the stem
vowel is uniformly long. The explanation for this in the history
of German is that this is a PU effect: the vowel in the simplex
form was lengthened to match the long vowel of the
morphologically related word. (Note that this is only a partial
PU effect, since the stem in Rad and Rades differs in terms of
the voicing of the final consonant.)
(4)
ab [ap] ‘off’
(5)
Page 8 of 34
Stage I Stage II
/rad/ /rad + ∂s/ /rad/ /rad + ∂s/
1. (6 b) rat — 1. (6a) ra:d ra:d + ∂s
2. (6a) — ra:d +∂s 2. (6b) ra:t —
[rat] [ra:d∂s] [ra:t] [ra:d∂s]
Page 9 of 34
That this is a true PU effect is supported by simplex words like

ab in (5), which had no morphologically related forms.
Examples like these had a short vowel at stage 1 which did not
lengthen at stage II.
In a surface PU approach one might say that there was vowel

lengthening at stage 2 in (4b) to make the paradigms more
uniform and that there was no vowel lengthening in ab
because there was no morphologically complex related word
(i.e. a word belonging to the same paradigm) which would
trigger lengthening. In the analysis of the data in (4)–(5)
proposed by King (1969) the PU effects are accounted for in
terms of (p.7) rule ordering. Thus, the rule of Vowel
Lengthening in (6a) applies only before voiced obstruents,
crucially ordered before a rule of Final Devoicing in (6b).
Derivations for stage I and stage II are provided in (7).
(6)
a. Vowel Lengthening: V → V: / __ [-son, +voice]
b. Final Devoicing: [-son] → [-voice] / __ #
(7)
late:/lelt/ latter: /lelt/ matter:/maetr/
Level 1:
Dentalization — — /mætr/
Level 2:
Affixation — le1tr —
[le1t] [le1tr] [mætr]
The derivation for stage 1 indicates that Final Devoicing

precedes and bleeds Vowel Lengthening. The PU effect is
captured by reordering the two rules at stage 2. Here it can be
observed that Vowel Lengthening ‘overapplies’ to examples
like Rad which surface with a final voiceless obstruent. This
example of overapplication is captured in this framework by
ordering Vowel Lengthening before Final Devoicing. In this
approach the nonapplication of Vowel Lengthening in words
like ab is accounted for because the underlying representation
for these words has been restructured to /ap/. Since the
obstruent is underlyingly voiceless vowel lengthening cannot
Page 10 of 34
affect the vowel: Vowel Lengthening only goes into effect

before voiced obstruents.
In Lexical Phonology (Kiparsky 1982 a, b, 1985; Kaisse and

Shaw 1985; Mohanan 1986; Hargus and Kaisse 1993) PU
effects are captured in terms of the interweaving of
phonological and morphological rules. Consider for example
the data in (8) which illustrate dentalization in Belfast English
(from John Harris 1989: 40). The examples in (8a, b) reveal
that the anterior coronal noncontinuant consonants /t d n l/
are realized as dental before /r/.
(8)
a. [t̪]rain, [d̪]rain
b. ma[t̪]er, la[d̪]er, spa[n̪]er, pi[l̪]ar
Harris assumes that both dentals and r are [+distributed] and

captures this assimilation with the spreading rule in (9):
(9) Dentalization:
The data
relevant
for PU
effects are
presented
in (10).
Here we
see that Dentalization does not affect a stem-final /t d n l/ if it
is followed by a class II (derivational) suffix (in 10a) or an
inflectional suffix (in 10b). (Note, however, that the rule does
apply across class I suffixes, e.g. elemen[t̪]-ary.)
(p.8)
(10)
a. shou[t]er, ru[n]er, ki[l]er
b. la[t]er, lou[d]er, fi[n]er
In a surface-oriented PU account one could argue that this is

an example of the ‘underapplication’ of the allophonic rule of
Dentalization (see Benua 1997a). This means that
Dentalization fails to occur in the examples in (10) because
dentalized consonants in the morphologically complex forms
would cause them to differ from the corresponding simplexes
(i.e. la[t]e and la[t]er would not be a uniform paradigm). In
Page 11 of 34
Harris’s Lexical Phonology analysis these PU effects are

captured by ordering the rule of Dentalization at level 1,
before the suffixes at level 2 (i.e. the suffixes in (7)) are added.
This point is illustrated in the derivations in (11) for late, later,
and matter.
(11)
Page 12 of 34
(11) /origin/ NONFINAL ALIGN-R OO-IDENT ALIGN-L ≫
a. o(rígin) *! *
b. (óri)gin *
c. (óri)gin *
d. ? (óri)gin *
Page 13 of 34
In (11) we can see that Dentalization applies to the /t/ in

matter at level 1. The rule does not affect the /t/ in late (since
there is no following /r/) and crucially also does not affect the /
t/ in later (because the suffix is not added until level 2). Note
that Dentalization is correctly predicted to apply across class I
suffixes, e.g. elemen[t̪]-ary, if class I morphology precedes
Dentalization.
A final example of PU effects in rule-based approaches can be

illustrated by considering a well-known example from English
stress that illustrates the cyclic application of phonological
rules. Longer monomorphemic words like lòllapalóoza are
regularly stressed on the initial syllable. However,
heteromorphemic words like orìginálity provide systematic
exceptions to this generalization. The explanation for the
exceptional stress of orìginálity is that it maintains as much as
possible of the stress pattern of the morphological base,
oríginal, which it contains. In a cyclic analysis (see Kiparsky
1979), the base exerts an influence on orìginálity because it is
derived on an earlier cycle. This point can be seen in the
following simplified derivations in which feet are indicated
with parentheses.
Page 14 of 34
(12) origin original origin-al-ity
1: footing (ori)gin o(rigi)n-al o(rigi)n-al
2: morphology — — o(rigi)n-al-ity
3: footing — — o(rigi)(n-al-i)ty
Page 15 of 34
In (12) we can observe that binary footing occurs at the first

cycle, i.e. before -ity is added by a morphological rule. By
contrast, origin-al is assumed to form the input to the first
cycle. On the second cycle (i.e. at step 3) we can see that the
same phonological rule applies once again, namely binary
footing. (Later rules are responsible for the (p.9) demotion of
the primary stress in the first foot in originality to secondary
stress.) The important point here is that the PU effects with
respect to original v. originality are captured in terms of the
cyclic application of footing.
1.4 THE STATUS OF PARADIGMS IN OPTIMALITY

THEORY
Optimality Theory (OT, McCarthy and Prince 1993b ; Prince
and Smolensky 1993) has brought renewed attention to the
role of paradigms in explaining surface exceptions to
phonological generalizations. The reason for this interest lies
in OT’s strong claim of parallelism for phonological analyses:
there are only two levels of representation, the input and the
output, and all constraints relevant to a particular structure
interact in a single constraint hierarchy evaluating just these
two levels of representation.6 Cyclic analyses of paradigm
effects like the ones described above in Section 1.3 are
incompatible with parallelism, as cyclicity is inherently
derivational in proposing intermediate levels of representation
between the input and output.7 Further, cyclicity leads to
surface exceptions to otherwise general phonological
processes, and so it is not obvious how a single constraint
hierarchy can account both for the general phonological
processes and morphologically motivated exceptions. These
problems can be illustrated with the example from English
stress discussed above. In OT, it has been proposed beginning
with work such as Benua (1997a, b), Kenstowicz (1996,1997),
Raffelsiefen (1995), Steriade (1997a), and Kager (2000) that
Correspondence constraints, already motivated to evaluate
similarity between inputs and outputs as well as between
redu-plicants and bases, can be appealed to in order to
optimize similarity between the outputs of morphologically
related forms. A base form like original can influence the
stress pattern of originality, because an Output–Output
correspondence constraint requiring stress identity between
these two morphologically related forms outranks (some of)
the constraints accounting for the regular stress pattern. Two
Page 16 of 34
models of Output–Output correspondence have been

developed within OT, a Base Priority model and a Symmetrical
model. Both are represented in this volume and are briefly
introduced and compared below.
The most explicitly formalized Base Priority model of Output–

Output correspondence is Benua’s (1995,1997a, b)
Transderivational Correspondence Theory (TCT). In TCT, Base
Priority is formalized through recursive tableaux sets, with
each recursion motivated by a morphological operation
applied to some Base. Each recursion contains the same
constraint rankings, but each recursive tableau is outranked
by the preceding recursion (i.e. that of its morphological and
phonological Base). These points (p.10) are illustrated by
Benua’s (1997a : 37–8) analysis of stress in the
morphologically related set, origin, original, originality. The
relevant constraints are given in (13).
(13)
a. NONFINALITY: Do not parse a word final
syllable into a foot.
b. ALIGN-R: AlignR(Foot, Word)
c. ALIGN-L: AlignL(Foot, Word)
d. O-O-IDENT: The foot structure of a derivative
and Base are identical.
Constraints (13a–c), ranked in that order, define regular stress

for nouns and adjectives of English: the final syllable is
optimally not parsed for stress (13a); otherwise the rightmost
foot is at the right edge of the word (13b); and the initial
syllable of the word is stressed (13c). For a Base to influence
its derivative, the Output–Output constraint (13d) must
outrank one of these constraints. Since the effect of a Base
like original is to move the stress away from the initial syllable,
O-O-IDENT must outrank ALIGN-L. The tableau at (14)
exemplifies the analysis (adapted, Benua 1997a: 38).
(14) Recursion A
Page 17 of 34
≫ /origin+al/ NONFINAL ALIGN-R OO-IDENT ALIGN-L ≫
a′. o(rígi)nal * *
b′. (óri)ginal **!
c ′. o(rígi)nal * * *
d′. ? o(rígi)nal * * *
Page 18 of 34
Recursion B
Page 19 of 34
≫ /original+al+ity/ NON FINAL ALIGN-R OO-IDENT ALIGN-L
a″. o(rìgi)(náli)ty * *
b″. (òri)gi(náli)ty *
c″. (òri)gi(náli)ty * **!
d″. ? o(rìgi)(náli)ty * * *
Page 20 of 34
Recursion C
Page 21 of 34
Recursion A Recursion B
/læri/ OO-ID *ær]σ IO-ID ≫ /læri- OO-ID *ær]σ OO-ID

TRUNC/
a. lari *! a′. lar *
b.lari *! b'. lær * *
c.laeri c'. lar *! *
d′. ? lær d'. ? lær *
Page 22 of 34
(p.11) In this recursion tableau, the paradigm consisting of

(14d, d′, d″) is optimal, as it contains the set of outputs which
best satisfy the highest-ranked constraints defining the stress
pattern, NONFINALITY and ALIGN-R. The Output–Output
constraint, OO-IDENT, crucially outranks ALIGN-L, optimizing
stress identity between the Base in Recursion B and its
derivative in Recursion C.
As we can see, TCT retains many features of cyclicity. Each

recursion, like each cycle, is motivated by a morphological
operation, and the input and output of each recursion must be
a phonologically and morphologically well-formed word.
Recursion mimics the inside-out morphological derivation of
the cycle, and ranking each new tableau recursion below that
of its Base accounts for the inside-out phonological influence
of a Base on its derivative. As Benua (1997a) makes clear, the
ranking between tableaux also allows TCT to resolve ranking
paradoxes, just as the cycle provides a solution to rule
ordering paradoxes. This point is illustrated in Benua’s
(1997a) analysis of nickname truncation in some dialects of US
English. In these dialects, the nickname L[æ]rry, for example,
is truncated to L[æ]r, even though this pronunciation violates
an otherwise exceptionless constraint against tautosyllabic
[ær] sequences: *ær]σ.8 The most plausible motivation for this
exception is that the truncation is required to match the vowel
quality of the Base nickname. (A high-ranked Output–Output
constraint, OO-ID, formalizes this generalization.) The tableau
below exemplifies the analysis (Benua 1997a : 34, fig. 16).
(15)
/capital + STRESSCLASH OO-Foot FOOTING

istic/ Base: STRUCTURE CONSTRAINTS
[(cápi[?]al)]
? a. *
(càpi[?]a)
(lístic)
b. (càpi) *!
{tha(lístic)}
c. (càpi) *!
(thà)(lístic)
Page 23 of 34
In this tableau, *ær]σ must outrank IO-ID (the constraint

requiring inputs and outputs to be identical) to satisfy
Richness of the Base (Prince and Smolensky 1993): ær]σis
normally never an optimal output, no matter what the input
might be.9 However, IO-ID in Recursion A must outrank *ær]σ
in Recursion B, otherwise the paradigm /lari, lar/ (15a, a′)
would be optimal. It is not entirely clear, then, whether
recursion is entirely compatible with OT’s parallelism
requirement, as McCarthy (Ch. 8 below) points out.10
(p.12) This is perhaps one reason why other work that

adopts the Base-Priority approach does not adopt TCT’s
tableau recursion formalism to encode the asymmetrical
relationship between a Base and a morphologically related
form. (Indeed, none of the chapters (3, 4, 5, 7) in this volume
that formally implement a Base-Priority approach to account
for paradigm uniformity use TCT.) Instead, the convention is to
list the output Base with the input as a correspondence target
for a morphologically derived form. For example, Davis (Ch. 5
below) argues that capi[c] alistic has a flapped t, instead of the
expected aspirated one (as in regular mili[t h]aristic) in order
to match the footing of its Base, capital. The tableau
exemplifying the analysis is given at (16) (adapted, Davis (this
volume, fig. (26)); parentheses indicate feet and curly brackets
superfeet).
(16)
Page 24 of 34
Input /lem-n-/ *CÀC.CÁC OP-Max-Vfeat-obj *∂́ IO-MAX-VFEATWD
a. l∂́mnámš
l[∂́]mnálxw
l∂́mnám∂ ***** *****
l∂́mnál∂
l∂́mn∂xw
b. lèmnámš
lèmnálxw
lèmnám∂ *!***
lèmnál∂
lémn∂xw
c l∂́mnámš
l∂́mnálxw
l∂́mnám∂ *!*** **** ****
l∂́mnál∂
lémn∂xw
Page 25 of 34
Like other Base-Priority analyses in this volume and elsewhere

(e.g. Kenstowicz 1996, 1997; Steriade 1997a ; Kager 2000)
this one is typical in not formalizing the analysis of the Base
output, and so the formal relationship between the grammars
accounting for the Base and its derivative is also not
discussed. As a result, it is not always clear in these analyses
whether the Base and its derivative are optimized by a uniform
set of constraint rankings. This makes it difficult critically to
evaluate these analyses and compare them with alternative
approaches to cyclicity, like the Symmetrical approaches
adopted by other chapters in this volume (4, 6, 7, 8, 11).11
Another reason why other work in the Base-Priority model

does not adopt TCT’s tableau recursion is that recursion
follows the cyclic model in proposing that words can only
influence each other’s phonology if one is morphologically
derived from the other. However, as Burzio (Ch. 4 below)
argues, following Steriade (1997a), the liaison pronunciation
of many masculine prenominal adjectives in French is
influenced by two Bases: the non-liaison pronunciation of the
masculine and the corresponding feminine form. And in
Hebrew, Bat-El (Ch. 3 below) shows that the imperative form
of verbs is phonologically based on the output of the future
form. Yet, it is implaus- (p.13) ible to propose that the future
is the morphological Base of the imperative, as the imperative
and future are mutually exclusive inflections of the same verb.
As these works argue, it is an important advantage of the
Output–Output approach that it can account for these kinds of
cases, while the cyclic model fails, as it requires a
phonological Base to be also the (unique) morphological Base.
TCT shares in this failing, as its formalism incorporates a
cyclic definition of the Base.
Another advantage of the (non-TCT) Output–Output model of

Base-Priority over the cyclic one is that it can account for
exceptions to surface generalizations motivated by
paradigmatic contrast (or anti-homonymy) as well as
paradigmatic uniformity, as demonstrated by two chapters in
this volume, Chs. 7 and 10. For example, Kenstowicz (Ch. 7)
argues that the regular phonology of Russian would predict
that the 3rd singular and plural present tense forms of ‘smoke’
should be pronounced identically: kúr’-it. Instead, the 3rd
singular form is kúr’-it, while the plural form is replaced by
kúr’-ut. This sort of anti-homonymy would be impossible to
account for in a cyclic model. First, the 3rd singular is not the
Page 26 of 34
morphological Base for the 3rd plural. Rather, both take a verb
root as Base and independently take the appropriate
inflections. Further, the cyclic model can explain only
unexpected similarities among morphologically related words,
not unexpected differences. However, Output–Output anti-
Faithfulness correspondence constraints can account for this
sort of data, as they establish a relationship between outputs
of morphologically related words and penalize identity
between outputs. As in Base-Priority accounts of paradigm
uniformity, in paradigmatic contrast we find an asymmetrical
relationship between members of the paradigm. The
phonological Base (in the Russian case, the 3rd person
singular) conforms to the regular phonology of the language,
while a morphologically related form (the 3rd person plural)
shows exceptional phonology due to a correspondence
relationship with the Base.
In contrast, the other model of Output–Output correspondence

to paradigm uniformity, implemented by Chs. 6, 7, 8, 11, posits
a symmetrical relationship among members of a paradigm,
with all members potentially able to influence the
pronunciation of the others, instead of one member having
priority. As Kenstowicz (1996,1997; Ch. 7) and McCarthy (Ch.
8) discuss in thoughtful detail, the important morphological
motivation for developing the Symmetrical model (called
Uniform Exponence by Kenstowicz and Optimal Paradigms by
McCarthy) is to account for uniformity in inflectional
paradigms. As noted above in discussing Russian, members of
an inflectional paradigm do not stand in a cyclic relationship
to each other. Instead, each member of the paradigm is
morphologically derived from a shared base root, and the
generalization underlying uniformity is that a shared string of
all members of the paradigm has to be as similar as possible
for some phonological property. This generalization is
obviously not amenable to a cyclic analysis. However, it can be
formalized in terms of Output–Output constraints requiring
identity for some phonological property to hold for all
members of a paradigm.
The important distinction between the Symmetrical approach

and the Base-Priority approach to paradigmatic effects, then,
is not in the kinds of constraints involved. (p.14) Both define
paradigm uniformity (or contrast) in terms of Output–Output
correspondence constraints holding among morphologically
related forms. Rather, the difference lies in the
Page 27 of 34
implementation. In the Base-Priority approach, the Base and

each of its derivatives is evaluated in a separate tableau, as
shown in (14)–(16), above, while an entire paradigm is
evaluated in a single tableau in the Symmetrical approach. For
example, Urbanczyk (Ch. 11 below) argues that in the
Hul’qumi’num’ main stressed schwa is generally banned by a
markedness constraint: *[∂́]. A regular exception to this
generalization is found in the ‘limited control’ object
paradigm, where a root contains a stressed schwa—[l∂́mn∂w]
‘see him, her, it, them’—if the root has a schwa in the other
members of the paradigm: [l∂́mnám̓š] ‘see me’, [l∂́náíxw] ‘see
us’, [l∂́mnám∂] ‘see you (sg)’, [l∂́mnál∂] ‘see you (pl)’.
Urbanczyk argues that this is accounted for by ranking an
Output–Output constraint relating members of this paradigm
(OP-MAX-VFEAT-OBJ) above the constraint that penalizes
stressed schwa. This analysis is exemplified in (17)
(Urbanczyk, Ch. 11 (11)).
(17) *CÀC.CÁC, OP-MAX-VFEAT-OBJ ¯ *∂́, IO-MAX-

VFEATWD
Page 28 of 34
/origin/, / NONFINAL ALIGN-R OO-IDENT ALIGN-L IO-IDENT

original/, /
originality/
a. (óri)gin, *
(óri)ginal, **!
(òri)gi(náli)ty *
b. ? (óri)gin, * √
o(rígi)nal, * * *
o(rìgi)(náli)ty * *
c. o(rígin), *! √ *
o(rígi)nal, * *
o(rìgi)(náli)ty * *
d. (óri)gin, * √
o(rígi)nal, * * *
(òri)gi(náli)ty ** *! *
Page 29 of 34
As shown, each candidate set is a complete paradigm.

Paradigm uniformity is not the result of some member(s) of the
paradigm conforming to the output of one other attractor, a
Base, but rather by what McCarthy (Ch. 8 below) calls a
‘majority rules’ effect: the more members of a paradigm share
an output string, the more pressure there is for all members of
the paradigm to share the same output string.
Given two models for implementing Output–Output constraints

—a Base-Priority model and a Symmetrical model—the
question arises of whether there is any principled way of
predicting which sorts of paradigmatic effects are most
appropriately (p.15) analyzed in a particular model.
McCarthy (Ch. 8), for example, suggests that the Symmetrical
model is best suited to analyzing uniformity in inflectional
paradigms, while the Base-Priority model is best suited for
derivational paradigms. However, as noted above,
paradigmatic contrast is found in inflectional paradigms, yet
its analysis depends on an asymmetrical relationship between
a phonologically regular Base and a phono-logically
exceptional other member of the paradigm. As Albright (Ch. 2
below) argues, uniformity in inflectional paradigms in Latin
and Korean are also best explained by proposing that there is
a Base form—the one that most reliably predicts the outputs of
the rest of the paradigm—which exerts pressure on the other
members of the paradigm. On the other hand, as Downing (Ch.
6 below) shows, the Symmetrical model is most appropriate
for analyzing uniformity in causative verb stems in Bantu
languages such Jita, although the causative is considered a
derivational paradigm.
It is also not possible to correlate Base-Priority with cyclic

effects and the Symmetrical model with noncyclic effects. As
noted above, Base-Priority analyses are appealed to in
accounting for uniformity or contrast effects that are not
amenable to cyclic analysis in several chapters in this volume:
Chs. 3, 7, and 10. Further, as Downing (Ch. 6) and Kenstowicz
(1996, 1997) argue, constraint ranking can have the effect of
defining a Base in the Symmetrical approach, often imitating
the inside-out influence of a Base on a morphologically derived
form. This is shown at (18). Symmetrical reanalysis of the
stress pattern of origin, original, originality :
(18)
Page 30 of 34
gruob, Ist gruoben Ist pl starb, Ist X

sg : = sg :
‘dug’ ‘dug’ ‘died’ X starben,

= Ist pl
‘died’
As shown, the Symmetrical analysis chooses the same optimal

paradigm (18b) as the Base-Priority analysis given in (14),
above. The morphologically derived words in the paradigm
cannot influence the stress pattern of their Bases due to high-
ranked mark-edness constraints, as shown in the non-optimal
paradigms (18a) and (18c). Paradigm (p.16) (18d) is non-
optimal, if we adopt Burzio’s (Ch. 4) concept of Gradient
Attraction to refine the proposal that, in a Symmetrical
analysis, every member of the paradigm is a Base for every
other member. Gradient Attraction proposes instead that the
nearest derivational ‘neighbor’ in a paradigm will serve as an
attractor, as it shares more segments with a form immediately
derived from it than a derivationally more distant form does.
(See Steriade 1997a for a similar proposal.) In this case, even
though origin is a potential Base for originality, it does not
influence the stress of originality because original shares more
segments with originality than origin does. An important topic
for future research is to see how successfully other Base-
Priority analyses can be recast in the Symmetrical model.
As Raffelsiefen (Ch. 9) shows, it is also a matter for future

research to determine whether all identity effects holding
among the outputs of morphologically related words must be
analyzed in terms of Output–Output constraints. The chapter
argues persuasively that certain similarities among
morphologically related words which superficially resemble
paradigm uniformity effects result instead from the alignment
of morphological and prosodic boundaries. Clarifying this
issue is particularly relevant for determining the level of
abstractness at which paradigm uniformity constraints
operate. If certain effects arguably result from alignment
rather than from paradigm uniformity, there is no need to
assume that the relevant features are lexicalized. This sort of
evidence is crucial for evaluating Steriade’s (2000) claim that
paradigm uniformity effects can hold for non-contrastive
features.
Page 31 of 34
To sum up, this volume illustrates a range of ways that

paradigms can explain exceptional synchronic phonology in a
variety of unrelated languages. The analyses, whether cast in
Base-Priority or Symmetrical models of OT, present numerous
examples of paradigm effects that cannot be accounted for in
previous models, such as cyclicity As we can see, though, the
contributors to this book also raise many interesting
unresolved issues for future work on paradigms in OT. One is
why some effects require definition of a single Base, while
others do not. For systems that do require a single Base, what
factors help predict which forms are likely to be chosen as
Bases? Why is it that, in a single language, some paradigms
require identity (or contrast) while others have only the forms
predicted by regular phonology? While this volume suggests
some answers to these questions, its most important
contribution will be to provide an empirical and formal basis
for future work on the role of paradigms in phonological
theory.
Notes:
(1) There are a wide variety of terms in use referring to
analogical changes, some indicating the author’s attitude
towards the phenomenon, including falsche Analogie ‘false
analogy’ (cf Whitney’s (1867) mistaken analogy),
Formübertragung ‘transfer of form’, Uniformierung
‘uniforming’, Parallelbildung ‘parallel formation’, Association
‘association’, Angleichung ‘similarity’, Ausgleichung ‘leveling’,
Verschleppung ‘abduction, spreading’, Contamination
‘contamination’, Stumpfsinnige Übertragung ‘mindless
transfer’, Verflechtung ‘interweaving’, Verschmelzung
‘blending’, Kreuzung ‘crossing’, Rückbildung ‘back-
formation’ (Curtius 1885: 36, 37).
(2) Paul (1880: 205) suggests that perhaps this is a hearer-

based phenomenon: a hearer will object to a new form less if it
differs only slightly from a familiar form.
(3) Leveling effects for words belonging to groups based on

semantic relatedness have also been proposed. For instance,
numerals often exhibit sound changes resulting in partial
sameness with the following number (e.g. the f in English four,
which cannot be explained on phonological grounds but
anticipates the f in five (Osthoff and Brugman 1878)). The
Page 32 of 34
relatedness of these effects with paradigm uniformity is

obvious: in both cases the sound structure of some word is
(partially) modified to match the corresponding feature in
some associated word (Jeffers and Lehiste 1979: 57).
(4) It should be noted that paradigms play an important role in

frameworks outside the generative tradition, e.g. Natural
Morphology (Wurzel 1989; Mayerthaler 1981).
(5) We see the effects of Final Devoicing in simplex forms like

Rad.
(6) See, however, work in the OT framework that argues for

more than one level, e.g. Kiparsky (2000); Rubach (1997);
Raffelsiefen (this volume).
(7) But see Orgun (1996,1999) for arguments against the view
that cyclicity is inherently derivational.
(8) As Kager (1999b) notes, while this constraint may look ad

hoc, it actually is well grounded phonetically Since English [r]
is pronounced with a low, back, [-ATR] tongue position, it is
most compatible with low, back, [-ATR] vowels.
(9) Kager’s (1999b) reanalysis of this data assumes that /lari/

is a possible input for [læri], just as either /kar/ or /kær/ should
be a possible input for [kar], with *æ]σ ≫ IO-ID choosing [kar]
as the optimal output in either case. However, as Benua
(1997a) and Kiparsky (2000) argue, a and æ contrast
everywhere except before r, so no plausible constraint ranking
would optimize [læri] as an output for /lari/.
(10) As Kiparsky (2000) points out, the truncation in (3)

illustrates opacity rather than cyclicity, since the
morphological Base for truncation is not contained in its
derivative. Indeed, he argues that truncation represents the
sort of opacity problem that motivates serial OT: distinct stem,
word, and post-lexical levels of phonology, with the output of
each level forming the input for a potentially distinct
constraint ranking at the next level. It is beyond the scope of
this chapter to discuss serial OT further. See Orgun
(1996,1999) for arguments that distinct phonologies
correlating with distinct morphological constructions or levels
are not inherently derivational and so are compatible with a
declarative theory of phonology like OT.
Page 33 of 34
(11) See work like Kiparsky (2000) and Orgun (1996, 1999) for
other critiques of the Base-Priority approach to cyclicity, and
alternative theories of cyclicity within OT
Page 34 of 34
The Morphological Basis of Paradigm Leveling


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
The Morphological Basis of

Paradigm Leveling
Adam Albright
DOI:10.1093/acprof:oso/9780199267712.003.0002

This chapter argues that the Latin honor change was caused
by more than simply a sporadic pressure for paradigm
uniformity or Uniform Exponence constraints to assert
themselves over IO-Faithfulness constraints. It shows that the
spread of [r] to nominative forms did more than just create
uniform paradigms; it also extended a pattern of non-
alternation that was already dominant in the lexicon. Details of
the change, such as its restriction to polysyllabic nouns and
non-neuters reflect the fact that these were especially strong
contexts for [r] stems. Furthermore, the ‘backwards’ direction
of the leveling, with oblique forms influencing the nominative
singular, can be explained by a particular model of Latin noun
paradigms, in which an oblique form served as the base, and
nominative forms were derived from oblique forms by rules
operating on surface forms. More generally, this result
provides evidence for a model of paradigm learning in which
learners choose the base form that is the most informative —
i.e. that preserves the most distinctions between classes of
Page 1 of 53
words, and allows the remainder of the paradigm to be

predicted with the greatest accuracy and confidence.
Keywords: Latin, honos, honor, paradigm uniformity, nominative forms,

paradigm learning
2.1 INTRODUCTION
In older stages of Latin, many nouns exhibited paradigmatic s
~ r alternations created by rhotacism in suffixed forms, as in
(1a). In the period immediately before Classical Latin, these s
~ r alternations were eliminated by extending the r to the
nominative form. At approximately the same time, an
independent change shortened long vowels before word-final
sonorants, resulting in the paradigm in (1b).1
(p.18)
(1) Elimination of s ~ r alternations in pre-Classical

Latin
a. Pre-leveling b. Post-leveling
[honoːs] ⇒ [honor] ‘honor’(nom.sg.)
[honoːris] [honoːris] (gen.sg.)
[honoːri:] [honoːri] (dat.sg.)
[honoːrem] [honoːrem] (acc.sg.)
etc… etc…
As Hock(1991: 179–90) points out, the change from [honoːs] to

[honor] can be described in two different ways. The first is as
a four-part analogy, as in (2), in which [honoːs] was influenced
by words that already had an [r] in the nominative, such as
[soror] ‘sister’.
(2) Extension of r by four-part analogy:

[soroːris] : [soror] : : [honoːris] : X (X = [honor])
Although the four-part analogy notation expresses the change

as the influence of one particular lexical item (in this case,
[soror]), it is generally recognized that such changes are
actually due to the collective influence of many words, such as
[soror], [kruor] ‘blood’, and also the numerous agentive nouns
ending in -or ([o:ra:tor] ‘speaker’, [gladia:tor] ‘gladiator’, etc.).
However, even when we recognize that the four-part analogy
notation is just a shorthand for the influence of a larger
Page 2 of 53
pattern, it is far from an explanation of the change. How many

words does it take for speakers to construct such an analogy?
How similar do they have to be? More importantly, it has often
been noted that four-part analogy cannot tell us why the
influence was not from [honoːs] to [soror] ([honoːris] :
[honoːs] : : [soroːris] : X, X = [soroːs]) (Hock 1991; Barr 1994;
Kiparsky 1997), or why the change was not in the opposite
direction, undoing rhotacism (something like [soror] :
[soroːris] : : [honoːs] : X, X = [honoːsis]). This last question is
especially puzzling, because the actual change in (2) violates
the tendency for analogical change to extend from more
‘basic’ or underived forms to less basic, or derived forms
(Kurylowicz 1947).
The four-part analogy notation can equally well capture the

leveling of alternations ([honoːs] ⇒ [honor]) or the extension of
alternations ([soror] ⇒ [soroːs]). An alternative account of the
spread of [r] to the nominative in [honor] is as paradigm
leveling, with the nominative form changing to match the
remainder of the paradigm ([honoːris], [honoːriː], [honoːrem],
etc.). The pressure to level paradigms has been formalized in
Optimality Theory as constraints on paradigm uniformity or
Uniform Exponence (Kenstowicz 1995; Steriade 2000); for
example, Kenstowicz schematizes the [honoːs] ⇒ [honor]
change as the promotion of a constraint demanding uniformity
in noun paradigms (UE). In the first stage, shown in (3), the
ban on intervocalic [s] (*VsV) outranks Faithfulness for /s/
(FAITH-/S/), yielding rhotacism in suffixed forms. FAITH-/s/ in
turn outranks UE, meaning that rhotacism does not overapply
in the nominative form. The result is a paradigm with s ~ r
alternations, as in (3a). Note that candidate (c) contains a
vowel length alternation ([or] ~ [o:ris]—Kiparsky 1997; Hale,
(p.19) Kissock, and Reiss 1998; Baldi 1999: 323), for which
Kenstowicz does not record a UE violation. It is entirely
possible that there are separate UE constraints for different
alternations, with UE for s ~ r alternations ranked higher than
UE for vowel length alternations. If that is the case, then UE
(S ~ R) is the relevant constraint here, and this is what I will
assume.
(3) Stage 1: *VsV ≫ FAITH-/S/ ≫ UE(s ~ r)
Page 3 of 53
/honoːs/ (nom.), *VsV FAITH-/ UE(S ~

/honoːs-is/ (gen.), s/ R)
/honoːs-em/ (acc.)
a. F √ ** *
[honoːs],
[honoːr-is],
[honoːr-em]
b. [honoːs], *(gen.)! √ √
[honoːs-is], *(acc.)
[honoːs-em]
c. [honor], √ ***! √ (?)

[honoːr-is],
[honoːr-em]
Under this analysis, the change consists of promoting UE over

FAITH-/S/, so that rhotacism overapplies in the nominative, as
in (4). This leaves two possible candidates for the nominative:
[honoːr] (candidate (c)) and [honor] ((d) and (e)). The first of
these violates a high-ranking phonotactic constraint against
final [o:r] in Latin, favoring a paradigm with shortening in the
nominative ((d) or (e)); of these, the paradigm that preserves
long [o:] in the oblique forms is more faithful to the long [o:] of
the input (FAITH-/V:/), and thus candidate (d) is selected.
This analysis avoids many of the problems pointed out by Hale,

Kissock, and Reiss (1998) by limiting UE to s ~ r alternations.
An alternative possibility, however, is that the Uniform
Exponence analysis actually captures the creation of the
intermediate variant [honoːr] (see n. 1), with perfect UE
satisfaction, and that the underlying form of this word had
already been reanalyzed as /honoːr/ (as Hale, Kissock, and
Reiss claim) by the time final vowel shortening occurred, so
Uniform Exponence was irrelevant by that stage.
The Uniform Exponence account formalizes the intuition that

paradigm leveling is due to a pressure for nonalternating
paradigms, and that the resulting paradigm is one which (in
this case) satisfies both paradigmatic constraints and also
general phonotactic constraints of the language (such as *VsV,
and *oːr#). However, it leaves many details unaccounted for.
Page 4 of 53
Why, for example, was the old s ~ r alternation suddenly

intolerable, at the same time that a new o ~ oː alternation was
being created? We might have expected the UE constraint to
move above both the ːoːr# and *VsV constraints, (p.20) since
there was no crucial ranking between them before the change,
yielding a uniform paradigm with overapplication of both
rhotacism and shortening: [honor], [honoris], [honorem], etc.
Furthermore, we might expect an increased drive for
paradigm uniformity to level other alternations in noun
paradigms, but these remained by and large intact (see (5)):
(4) Stage 2: *VsV, UE(s ~ r) ≫ FAITH-/s/, FAITH-/V/
Page 5 of 53
/honoːs/ (nom.), *o:r *VsV UE(S ~ R) FAITH-/s/ FAITH-/V:/

/honoːs-is/ (gen.),
/honoːs-em/ (acc.)
a.… √ √ *(nom.)! ** √
[honoːs],
[honoːr-is],
[honoːr-em]
b. [honoːs], √ *(gen.)! *(acc.) √ √ √

[honoːs-is],
[honoːs-em]
c. [honoːr], *! √ √ *** √
[honoːr-is],
[honoːr-em]
d.… F √ √ √ *** *
[honor],
[honoːr-is],
[honoːr-em]
e. [honor], √ √ √ *** **!*

[honor-is],
[honor-em]
Page 6 of 53
(5) Alternations preserved in Latin noun paradigms
gloss ‘honor’ ‘city’ ‘art’
alternation [o] ~ [oː] [p] ~ [b] Ø ~ [t]
nom. [honor] [urps] [ars]
gen. [honoːris] [urbis] (*[urpis]) [artis]

(*[aris])
dat. [honoːri:] [urbiː] (*[urpiː]) [artiː]

(*[ariː])
Another unexplained mystery is why UE was promoted over

FAITH-/s/ to extend rhotacism, rather than FAITH-/s/ being
promoted over *VsV to eliminate rhotacism. As mentioned
above, the rhotacism constraint (*VsV) plays a crucial role in
Kensto-wicz’s analysis, ensuring that the resulting paradigm
will have uniform [r] and not uniform [s]. However, we could
just as easily have used the same analytical device (historical
reranking of two constraints) to predict the opposite change,
promoting (p.21) the IO FAITH-/S/ constraint to yield uniform
[s]. Thus, the mere existence of an active rhotacism constraint
at one stage in the grammar is not sufficient to explain why it
should continue to be true at the next stage in the grammar.
One possible explanation, proposed by McCarthy (1998), is
that output–output constraints start at the top of the grammar
in the initial state. If this is so, then we need only assume that
learners sometimes fail to demote UE below the relevant
markedness constraints, predicting a tendency for Uniform
Exponence to move up in grammars over time. But why would
one generation of Latin learners suddenly fail to apprehend
the correct ranking of FAITH-/S/ ≫ UE? If Uniform Exponence
is to have any explanatory force in accounting for paradigm
leveling, we would ideally like to be able to predict when such
UE ≪=≪ Faithfulness flips are likely to occur, and which
Faithfulness constraints will be demoted.
A Uniform Exponence analysis also fails to capture various

other details about the [honoːs] ⇒ [honor] change. Most
notably, the spread of [r] to the nominative form was
complicated by the fact that it was restricted primarily to
polysyllabic, non-neuter nouns such as [honoːs] ‘honor’ (masc.)
and [arboːs] ‘tree’ (fem.), shown in (6a) (Hock 1991; Barr
Page 7 of 53
1994; Kiparsky 1997). Monosyllabic nouns, such as [floːs]

‘flower’ (masc.) were not affected (6b), nor were polysyllabic
neuter nouns, such as [korpus] ‘body’ (neut.) (6c).
(6) Leveling restricted to masc. and fem. Polysyllables
Page 8 of 53
a. honoːs ⇒ honor ‘honor’ (masc.)
arboːs ⇒ arbor ‘tree’ (fem.)
odoːs ⇒ odor ‘odor’ (masc.)
(augus) ⇒ augur ‘omen’ (masc.)
b. floːs ⇒ floːs ‘flower’ (masc.) (⇏ flor)
oːs ⇒ oːs ‘mouth’ (neut.) (⇏ or)
moːs ⇒ moːs ‘custom’ (masc.) (⇏ mor)
c. korpus ⇒ korpus ‘body’ (neut.) (⇏ korpor, korpur)
tempus ⇒ tempus ‘time’ (neut.) (⇏ tempor, tempur)
onus ⇒ onus ‘burden’ (neut.) (⇏ oner, onur)
Page 9 of 53
Furthermore, the new [r] forms appear to have replaced the

older [s] forms relatively slowly; in the historical period we
find both [honoːs] and [honor], [odoːs] and [odor], [arboːs] and
[arbor], [lepo:s] and [lepor] ‘charm’, [laboːs] and [labor]
‘labor’ (Leumann 1977:179).2 Both these problems could be
handled by various means—UE could be restricted to non-
neuter polysyllables in some way, for example, and the free
variation could be accomplished by the gradual promotion of
stochastically ranked constraints (Boersma and Hayes 2001).
As above, the real problem is not in finding (p.22) theoretical
machinery that can describe the change; it is in understanding
why the change should have occurred in this direction, and to
these particular words.
In sum, a variety of questions (7) must be answered if our

understanding of the Latin honor analogy is to move beyond
description to actual explanation:
(7)
a. Why was a basic, ‘unmarked’ isolation form
(the nominative) rebuilt on the basis of more
marked suffixed forms, contrary to the usual
direction of analogical change? (Lahiri and
Dresher 1984; Bybee 1985: ch. 3)
b. What role (if any) did similar words, like
[soror], play in the change? Is there a minimum
number of such words necessary to effect such a
change?
c. Why did [honoːs] change to [honor], and not
[soror] to [soroːs]?3
d. Why were monosyllables and neuters
generally not affected?
e. Why did both [oːs] and [or] variants persist for
so long, and why was no variation induced in
etymological [or] words such as agentives?
I propose here that the change of [honoːs] to [honor] was

driven by more than just a phonological change involving
paradigm uniformity constraints. I suggest that it was actually
a morphological effect, resulting from the way that paradigms
are learned. I present a computationally implemented model of
paradigm acquisition that proceeds in two stages: first, it
compares all the available paradigms and selects the base
form that allows the remainder of the paradigm to be
projected as reliably, effectively, and efficiently as possible, in
Page 10 of 53
a way which will be quantified in s. 2.2.1. It then develops a

gradient morphological grammar to project the rest of the
paradigm, using the minimal generalization method of
morphological rule induction, developed by Albright and
Hayes (2002).
When this process is applied to Latin, it emerges that the

preferred base is an oblique form, not the nominative form,
accounting for the unusual direction of the analogical change
(7a). Furthermore, when an oblique form is used to project
nominative forms, the system makes essentially the right
predictions for the honor analogy: [-oːs] nominatives are
strongly favored for monosyllabic and neuter nouns, [-or]
nominatives are preferred for polysyllabic masculine and
feminine nouns with [-oːs] remaining a strong second choice,
and [-or] is strongly favored for agentive nouns.
This analysis is similar in spirit to the four-part analogy

explanation of the change, but the gradient nature of the rules
in this system gives us a quantitative expression of the
influence of other lexical items (7b), and also helps predict
which pattern should win out for each class of words (7c).
Finally, the close competition between [-oːs] and (p.23) [-or]
for many forms suggests an intriguing interpretation for the
persistence of [-oːs] nominatives into the attested period: I
conjecture that perhaps they were not merely a conservative
retention of memorized archaic forms, but that they may also
have been supported to a certain extent by the synchronic
grammar of Classical Latin. In other words, when Latin
speakers heard an archaism like [honoːs], even if it was not
the synchronically preferred form, it may have struck them as
moderately grammatical, allowing [-oːs] forms to persist as an
archaism much longer than some other archaic features.
Clearly more philological work is needed to support this
hypothesis, but it is a good example of how the current model
makes predictions about not only how forms are innovated,
but also how they may be retained or lost.
2.2 A MODEL OF BASE DISCOVERY
Page 11 of 53
The premise of the model proposed here is that language

learners seek the most accurate rules possible to describe
their language. In order to do this, they must select certain
forms to memorize, and then they must construct rules to
derive the remaining forms. The hypothesis that I will pursue
here is that learners initially explore many different
candidates for base status, comparing the ease with which
other parts of the paradigm can be derived from each
potential base. Once a sufficient number of words have been
learned, a base is chosen, and the learner focuses on
constructing a grammar to derive the rest of the paradigm.
The choice of which forms are memorized and which are

derived often has a substantial effect on the ability to
construct accurate rules. Consider, for example, the
hypothetical language in (8), which has a single nominative
ending ([-us]), but two possible genitive endings ([-iː] and [-
oris]).
(8) Neutralization in the nominative
nom. gen.
[gluptus] ~ [gluptiː]
[nokus] ~ [nokiː]
[reptus] ~ [reptoris]
[kortus] ~ [kortoris]
In the case of (8), there are two classes of words: the [-us] ~ [-
iː] words and the [-us] ~ [-oris] words. The two classes are
neutralized in the nominative, so the mapping from the
nominative to the genitive is unpredictable. If a learner were
to memorize just the nominative form, then two rules would be
necessary to project genitives ([-us]→[-iː] and [-us]→[-oris]).
Each of these rules would only have 50 percent accuracy in
the existing lexicon, each covering two out of four words.
Furthermore, for a hypothetical new word [tulpus], there
would be two possible genitives ([tulpiː] and (p.24)
[tulporis]), each with a 50 percent chance of being right,
leaving a 0 percent margin to decide by.
In such a situation, the learner should recognize that the

genitive form simply contains unpredictable information that
cannot be derived by rule, but must rather be memorized. If
Page 12 of 53
we were to derive nominative forms from genitive forms, we

still need two rules ([-oris]→[-us] and [-iː]→[-us]), but each
would have 100 percent accuracy in the existing lexicon.4 The
hypothetical new genitive forms [tulpiː] and [pulkoris] could
only have nominative forms [tulpus] and [pulkus], respectively,
with 100 percent certainty.
Thus, it appears that we can identify the ‘most informative’

base algorithmically, given two abilities: a capacity for
discovering the rules needed to derive one part of the
paradigm from another part, and a set of metrics to estimate
how effective or reliable the rules are in each direction.
2.2.1 The minimal generalization model of rule induction

One system for hypothesizing morphological rules is the
minimal generalization algorithm, outlined by Pinker and
Prince (1988) and developed and implemented by Albright and
Hayes (2002). The minimal generalization procedure starts
with pairs of related forms, such as the ones given above in
(8). It compares the left and right member of each pair, in
order to determine what material is constant across both
forms of the word, and what material changes. The result is a
word-specific rule for each input pair, as in (9).
(9) Word-specific morphological rules for the words in

(8)
a. [us] → [iː] / glupt —— #
b. [us] → [iː] / nok —— #
c. [us] → [oris] / rept —— #
d. [us] → [oris] / kort —— #
If the learner’s only task was to reproduce inflected forms that

had already been heard, then these rules would be sufficient.
However, in real life, speakers must frequently produce new
forms of words—particularly in richly inflected languages.
Therefore, it is necessary to be able to generalize. Under the
minimal generalization approach, this is done by comparing
word-specific rules that share the same change. For example,
in this hypothetical language, [reptus] and [kortus] both
exhibit the change [us] → [oris]; (p.25) these two words are
thus compared further to see whether there is some shared
property that may be conditioning this particular change. Fig.
2.1 shows how two words are compared to construct the
environment for a new, generalized rule.
Page 13 of 53
Such pairwise
comparisons
are iterated
across the
entire lexicon,
yielding a
sizeable set of
(often Fig 2.1. Minimal generalization to
redundant) discover environments
generalized
rules for each
morphological change. When a change occurs in a diverse set
of phonological environments, then the generalization
procedure in Fig.2.1 can yield very general rules; for example,
comparing the words [reptus] ~ [reptoris] and [kraus] ~
[kraoris] would yield the very general rule [us] → [oris] /——#
(assuming that [a] and [t] have no feature values in common).
In addition to hypothesizing rules, we also need a way to keep

track of how accurate or trustworthy the rules are. This is
done by calculating the reliability of each rule, as defined in
(10).
(10) Definition of a rule’s reliability:
For example, the generalized rule created in Fig.2.1 has a

structural description that covers three of the words in this
hypothetical language ([gluptus], [reptus], and [kortus]), but
only two of those words actually take [-oris] in the genitive
([reptoris] and [kortoris]). Thus, the reliability of the rule is ⅔
= .67. These reliability ratios are then adjusted using lower
confidence limit statistics to yield a confidence value, following
a suggestion by Mikheev (1997), so rules that attempt to cover
just a few forms are penalized for being too unambitious. For
example,
when
using a
confidence
level of α = .95, a rule with reliability of 5⁄5 is assigned a
confidence of .825, while a rule with reliability of 1000/1000 is
assigned a confidence of .999. (For details of calculating
confidence limits, see Mikheev 1997.)
Page 14 of 53
This generalization algorithm constructs rules that are quite

traditional in their format, but the resulting grammar is
unconventional in allowing a large number of specific and
often overlapping or redundant rules. When the grammar is
used to derive new forms, it is therefore possible that there is
more than one applicable rule. In this case, the rule with the
highest reliability value is the one that applies. Continuing
with the (p.26) example from the previous paragraph, we can
consider the predictions of the grammar for a novel word
[loptus]. The comparison of [gluptus] and [nokus] would yield
a rule of [us] → [i:] after voiceless stops, yielding an output of
[lopti:] for the novel word. The reliability of the rule used to
derive [lopti:] is 2/4 = .5, adjusted downward to .31, so we can
say that [lopti:] is projected with a confidence of .31. The
comparison of [reptus] and [kortus], on the other hand, yields
a rule of [us] → [oris] specifically after [t], which works for 2/3
of the existing words, for a reliability of .67, adjusted
downward to .40. This rule generates the output [loptoris] for
the novel word, with a confidence of .40. Thus, the [us] → [oris]
rule is more reliable in this environment, and [loptoris] would
be the winning output. The winning margin between the
winning output and the next best competitor in this case
would be .40 − .31, or .09.
The minimal generalization approach is designed to discover

which morphological processes have the highest reliability,
and in which phonological environments. In many cases,
however, the true reliability of a morphological process may be
obscured by phonological changes; consider, for example, the
language in (11).
(11) Phonological neutralization
nom. gen.
[neks] ~ [negis]
[arks] ~ [argis]
[teks] ~ [tekis]
[flurs] ~ [fluris]
In this case, the standard analysis would be to say that all of

the words behave the same morphologically (nominative [-s],
genitive [-is]), but there is a phonological process enforcing
voicing agreement in final obstruent clusters, yielding /negs/ →
Page 15 of 53
[neks], /args/ → [arks]. The procedure outlined above, however,

would analyze [neks] ~ [negis] with the morphological rule
[ks] → [gis] / ne ——#. What we need is a way to discover that
this word could also be analyzed as [s] → [is] / [neg] ——#, with
a further phonological process fixing the illegal word-final
*[gs]. Note that it would not be reasonable to equip the
morphological learner with the necessary phonological rule or
constraint ranking ahead of time, because the evidence for
such a rule would come precisely from alternations in
morphologically related words.
The Albright and Hayes (2002) implementation of minimal

generalization learning uses the following approach to solve
this problem: the learner is assumed to have prior knowledge
of what sequences are phonotactically legal in her language,
but not the specific rules governing alternations.5 When
assessing the reliability of a rule, the (p.27) learner then
checks whether the output of the rule would yield an illegal
sequence. For example, in learning rules for the forms in (11)
in the genitive → nominative direction, the forms [tekis] ~
[teks] and [fluris] ~ [flurs] would generalize to yield a
morphological rule [is] → [s] / ——#. In order to calculate the
reliability of the rule, we can try applying it to all the input
forms ([fluris], [tekis], [argis], [negis]). This yields the correct
result for [flurs] and [teks], but the wrong result for *[args]
and *[negs]. However, given the knowledge that [gs] is
phonotactically ill-formed, the learner posits a phonological
rule changing /gs/ to [ks]. This rule, when applied after the
morphological operation, allows the [is] → [s] rule to yield the
correct outcomes for [arks] and [neks] as well, for a reliability
of 4/4.
The system described in this section provides an automated

method for hypothesizing a set of morphological rules and
estimating the effectiveness of individual rules in explaining
the input data. The examples so far have involved the mapping
between just two slots in the paradigm (e.g. nom. → gen.); I
will refer to the set of rules between just two forms as a
subgrammar. The complete grammar of a language would
require many such subgrammars—e.g. nom. → gen., nom. →
dat., and so on. What remains, then, is to define a way to
estimate the reliability of an entire subgrammar, in order to
decide which parts of the paradigm can be derived relatively
Page 16 of 53
effectively by rule, and which forms should be memorized

instead.
2.2.2 Criteria for base selection

How can we decide that it is easier to project form X from
form Y, rather than the other way around? In other words,
what is the criterion for deciding that the subgrammar of rules
from X → Y is ‘better’ than the subgrammar from Y → X? Given
a subgrammar of rules with reliability values attached to them,
there are various possibilities for how to evaluate the
subgrammar as a whole.
One approach would be to examine inherent properties of the

subgrammars involved. We could, for instance, compare the
number of rules that are hypothesized in mapping X → Y and Y
→ X. Another possibility would be to measure the average
reliability of the rules in the subgrammar.
A second approach is to examine the performance of the

grammar in deriving forms. The word-specific rules in (9)
ensure that the system will always be able to derive the
correct answer for words that were part of the input data.
However, if we deprive the system of these word-specific rules
and then ask it to derive outputs for the training data, we can
measure how well the system is able to capture the data using
higher-level generalizations. One intuitive measure of the
success of a subgrammar is the proportion of times that the
‘correct’ (attested) form is chosen as the winning form.
Another possible measure is the average reliability of the rules
that derive the winning outputs, on the theory that more
ambiguous mappings will yield rules with lower reliabilities. A
third way to measure the performance of a mapping would be
to count the average number of competitors for each output,
since there will always be at least two possible outcomes for
an ambiguous input. Finally, we could measure the margin
between the (p.28) winning output and the next highest
competitor, since truly ambiguous mappings will lead to ‘coin-
toss’ situations where two possible outcomes are equally likely.
A reliable or effective subgrammar, then, should have these

properties: high reliability rules, high accuracy in deriving the
correct form for the training data, high confidence in the
winning output, few competing outputs for each input form,
and a large margin between the winning output and the next
highest competitor. Although these metrics reflect logically
Page 17 of 53
distinct features of grammars, and we could imagine cases in

which they did not agree with one another (for example, a
grammar with high accuracy but low winning margins), in
practice they turn out to be highly correlated with one
another; therefore, rather than selecting a single metric or
coming up with some formula to combine them, I will simply
present the results of all of these metrics side-by-side, for
comparison.
For present purposes, I will assume that a learner faced with

the task of learning Latin noun paradigms is attempting to
identify one single form that will serve as the base for the
entire paradigm. This is not the only possible approach. We
could imagine, for example, a system under which the learner
is allowed to choose multiple bases, upon discovering that
certain mappings (such as between the dative and the
ablative) are nearly 100 percent predictable, while other
mappings (such as between the nominative and the genitive)
are not nearly so predictable. Indeed, a multiple-base
approach of this type would be required to learn something
like the standard dictionary entries for Latin nouns, which list
two forms (the nominative and genitive singular). Confining
the learner to a single base form, at least for the purposes of
learning a ‘smallish’ paradigm of five cases and two numbers,
provides a more restrictive initial hypothesis that makes
stronger predictions about possible paradigmatic changes.
Page 18 of 53
2.2.3 Results for Latin noun paradigms

In the small, hypothetical examples discussed above, the
neutralizations in the nominative made the genitive →
nominative mapping clearly easier than the nominative →
genitive mapping. In real languages, however, the situation is
rarely so clear. Neutralizations typically affect only a subset of
the segments in the language, so the uncertainty they cause
may affect only a small number of words. Furthermore,
neutralizations often affect different parts of the paradigm for
different words. Thus, it is not always easy to intuit whether a
mapping is easier in one direction than the other, or the
magnitude of the asymmetry.
The question of interest here is whether Latin nouns were

easier to project in the oblique → nominative direction than
vice versa. Latin nouns are traditionally divided into five
classes, or declensions, each of which was inflected for five
major cases: the nominative, genitive, dative, accusative, and
ablative. (Two additional cases, the vocative and locative, were
almost always identical to other cases.) A full description of all
of the declensions and their subclasses is clearly beyond the
scope of this chapter—see Leumann (1977), Kühner (1912), or
Allen (1903) for in-depth discussions. What is important here
is that the distinctions between many of these classes were
neutralized or nearly neutralized in various parts of the
paradigm.
(p.29)
(12) Neutralizations in the nominative
Page 19 of 53
a. nom. gen. gloss
[populus] [populiː] ‘people’
v. [manus] [manuːs] ‘hand’
v. [korpus] [korporis] ‘body’
v. [genus] [generis] ‘kind’
b. [ager] [agriː] ‘field’
v. [gener] [generiː] ‘son-in-law’
v. [fraːter] [fraːtris] ‘brother’
v. [karker] [karkeris] ‘prison’
v. [iter] [itineris] ‘journey’
c. [daps] [dapis] ‘feast, banquet’
v. [urps] [urbis] ‘city’
Page 20 of 53
(13) Neutralizations in the genitive
Page 21 of 53
a. nom. gen. gloss
[populus] [populiː] ‘people’
v. [bonum] [boniː] ‘good(ness)’
b. [soror] [soroːris] ‘sister’
v. [honoːs] [honoːris] ‘honor’
v. [korpus] [korporis] ‘body’ (distinguished

only by short [o])
c. [miːles] [miːlitis] ‘soldier’
v. [kaput] [kapitis] ‘head’
v. [ankeps] [ankipitis] ‘danger’
d. [lak] [laktis] ‘milk’
v. [noks] [noktis] night’
Page 22 of 53
Some of these neutralizations affect relatively large numbers

of words. The neutralization between masculine ([-us]) and
neuter ([-um]) second declension nouns in (13a) involves two
very large (and productive) classes of nouns. The
neutralization caused by voicing agreement in final obstruent
clusters in (12c), on the other hand, affected relatively fewer
words. An additional complication is that other factors, such as
grammatical gender, could help the speaker know which suffix
to use in a potentially ambiguous situation—so, for example,
the use of [-us] or [-um] in the nominative of a word with [-iː]
in the genitive is almost completely predictable given the
gender of the word. Thus, if there is an asymmetry in
predictability between nominative and other forms, it would be
because of differences in the ‘severity’ of the neutralizations
involved, and the ability to predict the correct form using
gender. Were the neutralizations in the nominative in fact
more severe than in other cases?
In order to answer this question, I started with a database of

fully inflected classical Latin nouns, prepared in 1997–8 by a
group working under the supervision of Bruce Hayes at UCLA.
This database contained all of the nouns with five or more
tokens in a lemmatized frequency count from classical texts
(Delatte et al. 1981), based on a corpus of approximately
800,000 words (582,000 from prose, 212,000 from poetry).
Nouns beginning with the letters R–Z were omitted from the
current study (p.30) because the database was found to have
incomplete information for many paradigms in this section of
the alphabet. Nominative forms were listed in their forms prior
to the [honoːs] ⇒ [honor] change; in cases of uncertainty,
words were listed with a final [s]. The rationale for this was
that we are interested in seeing if the model will favor [r]
forms in spite of numerous [s] forms in the training data, and
we do not want this to be the result of the influence of
spurious [r] forms.
The model of base selection being tested here is that learners

evaluate the usefulness of prospective bases early in the
learning process. Therefore, the only input data which would
be available to the learner for comparisons would be the most
common words. As an idealization, words with 50 or more
tokens in Delatte et al. (1981) were selected, for a total of 494
input nouns. Six forms were considered as possible bases: the
nominative, genitive, dative, accusative, and ablative singular,
and the nominative plural. For each possible base form,
Page 23 of 53
training data files were then constructed to project each of the

remaining forms (nom. → gen., nom. → dat., nom. → acc., etc.),
yielding 30 (= 6 × 5) training sets in total.6
Nouns in the input files were listed in phonemic transcription.

In order to take into account the effect of phonological
processes, a list of illegal sequences was also prepared,
including final clusters disagreeing in voicing (*bs#, *gs#,
*ds#), final geminates (*ll#, *dd#, *ss#), the clusters *rts,
*lts, and *nts, and a few other illegal sequences whose repair
caused alternations (*oːr#, *kt, *iiː). Because some of these
illegal sequences refer crucially to word boundaries, word
boundaries were also marked explicitly in the input files with
brackets. Since rhotacism was not surface-true in this stage of
Latin (cf. [ka:sus] ‘fall’, [rosa] ‘rose’), intervocalic [s] was not
included as an illegal sequence.7 Finally, each noun was
provided with a numeric code indicating the grammatical
gender and the number of syllables (monosyllabic v.
polysyllabic), since (p.31) the current implementation of the
minimal generalization learner does not have an independent
capacity for considering general prosodic properties of words.
Token frequencies were also included in the input files, but
they were not employed in the simulations reported here.
The input files

were
submitted to
the minimal
generalization
learner,
yielding
subgrammars
of rules with
confidence
values. The
word-specific
rules were
then
eliminated,
Fig. 2.2. Comparison of potential bases
and the
according to the criteria in s. 2.2.3
resulting
subgrammars
were tested
on the input forms. The metrics proposed in s. 2.2.2 were
calculated for each subgrammar, to obtain an estimate of the
Page 24 of 53
usefulness of each slot in the paradigm for predicting the

remainder of the paradigm. The results, given in Appendix A,
show that the predictability between all forms is quite high
(over 80 percent); it is not the case that any form suffers from
neutralizations that affect the majority of forms. Nevertheless,
the oblique forms tend to be substantially better than the
nominative form on almost all the metrics considered. In Fig.
2.2. the candidates for base status are compared in terms of
their mean effectiveness in projecting the five other forms in
the paradigm.
(p.32)
As can be
seen, the
criteria
proposed in s.
2.2.2
generally
agree on the
relative
effectiveness
of the various Fig. 2.3. Average rank of forms as
forms as potential bases
possible
bases. The
only exception is the number of rules in the grammar, which
yields an uninterpretably different ranking. However, we saw
above that the number of rules failed to distinguish even the
simple four-word language in (8), so it is not surprising that it
performs poorly here. The combined results from all five
remaining criteria (excluding number of rules) are shown in
Fig. 2.3.
The nominative is the worst choice of base under all criteria,

and thus receives the lowest rank for all metrics. This reflects
the fact that the nominative suffers from more neutralizations,
affecting both more words and more segments, than the
oblique forms. Interestingly, the accusative also fares
relatively poorly, because it is the same as the nominative for
all neuter nouns, and thus shares many of the same
neutralizations. Among the remaining forms, the ablative
comes out slightly ahead of the genitive and the dative,
because there are a few nouns (the so-called ‘i-stems’) that
Page 25 of 53
have an unpredictable [i] in the ablative singular and genitive

plural, and often in the nominative and accusative plural forms
as well (14).
(14) Neutralization of i-stems in the genitive
gen. abl. gloss
[ignis] [igniː] ‘fire’ (masc.)
[imbris] [imbriː] ‘rain’ (masc.)
v. [patris] [patre] ‘father’ (masc.)
[fiːnis] [fiːne] ‘end’ (masc.)
Although Allen (1903) lists a large number of i-stem nouns, the

database employed here contained only six of them; thus,
there is only a very slim advantage to choosing a base form in
which the i-stems are distinct.8
(p.33) An additional factor that has not been discussed here

is the relative frequency of the different forms in the
paradigm. As an idealization, I have assumed that learners
have access to all forms of all nouns. Clearly this is not true in
real life, however; some cases are more frequent than others,
and the frequency of cases may differ from word to word. As
mentioned in s. 2.2.1, the Albright and Hayes (2002)
implementation of minimal generalization uses confidence
statistics to estimate the effectiveness of rules, so that rules
covering a few forms are penalized more than rules covering
many forms. Therefore, with more realistic input data,
including different amounts of data about different cases,
subgrammars involving less frequent cases would be penalized
because their rules would be based on fewer forms.
A simulation taking this into account would require more

detailed frequency information about Latin noun paradigms
than is currently available to me. Nevertheless, intuitively, it
seems that there are substantial differences in the frequency
of the oblique cases, and this would probably be the decisive
factor in choosing a base from among the oblique forms that
are more or less equivalent by all other criteria. For the
purposes of the [honoːs] ⇒ [honor] change, it is sufficient that
the model proposed here selects something other than the
nominative as the base form; in the discussion that follows, I
Page 26 of 53
will use the genitive singular as the base for deriving the
nominative, but the same result could be achieved using the
ablative or dative singular.9
2.3 PROJECTING NOMINATIVES FROM THE GENITIVE

Choosing an oblique form as the base in Latin noun paradigms
gives us only half the explanation for the [honoːs] ⇒ [honor]
change. In particular, it explains the ‘backwards’ direction of
the change (oblique forms affecting nominatives). This
answers the question in (7a), of why it was the nominative that
changed in Latin. What remains to be shown, then, is that
once an oblique form has been chosen as the base, the model
makes the right predictions for nominative forms: namely, that
polysyllabic non-neuter -oːs nouns changed to -or.
Recall that an assumption of the current model is that bases

are selected early in the learning process, but learners
continue to fine-tune their grammars to derive the remainder
of the paradigm.10 Therefore, in order to test the predictions
of the model (p.34) for nominatives using an oblique form as
the base, the model was trained on the full set of 1,687 nouns
in the gen. → nom. direction. The resulting grammar was then
used to generate possible nominatives for all genitive forms
ending with sequences that could potentially arise from
rhotacism: [-oːris-oːris], [-oris], [-uris], [-eris] (157 in all). The
grammar derived several possible nominatives for each noun,
each with its own confidence value. For example, for [-oːris-
oːris] and [-oris] genitives, the possible nominatives typically
included an [-oːs] nominative, an [-or] nominative, and various
other possibilities, such as [-oːris-oːris] (on the basis of words
like [kanis] ‘dog-nom./gen.sg.’, which were identical in the
nominative and genitive), [-us] (like [korpus] ‘body-nom.sg.’),
etc. For each noun, the best possible -r nominative was
compared against the best possible -s nominative, in order to
gauge the model’s preference for -r nominatives. As expected,
the preference for -r or -s varied substantially from word to
word, with four distinct types of words emerging (Fig. 2.4.)
(note that in the graph, bars indicate the size of a standard
deviation, not the standard error).
Page 27 of 53
As Fig. 2.4.
shows,
agentive
nouns, which
are all
polysyllabic
and masculine
or feminine,
strongly favor
-r. In fact, Fig. 2.4. Preference for -r or -s in four
these words distinct sets of words
contained -r
etymologically, and continued to have -r in the nominative with

no variation or hypercorrections. For other polysyllabic
masculine and feminine nouns, there is a slight tendency to
favor -r, but there is strong competition from -s. This is the
honor class of words, which were etymologically -s but
changed to -r, with some attested variation and occasional
hypercorrections of etymological -r to -s (Neue and Wagener
1902: 265). For the polysyllabic neuter nouns, -s is favored to a
moderate degree; these contained -s etymologically, and
continued to have -s except in some cases of contamination
from masculine doublets, and a few isolated other examples.
Finally, monosyllables, which also contained -s etymologically,
strongly favor -s ; in fact, these did retain -s with no variation.
Why does this pattern emerge? The differing strength of -r and

-s for different words is due to the fact that the system
employs multiple rules, with different confidence values in
different contexts. Among polysyllabic non-neuter nouns,
genitives in [-oːris-oːris] frequently have nominatives in -or.
Thus, the rule of [o:ris] → [or] / (p.35) [X] [polysyl,–neut] ——#
has a relatively high confidence (.727), correctly deriving
words like [soror] and [cruor], and all agentives, but failing for
words like [honoːs]. Among these forms, then, there is a slight
preference for -or in the nominative. It is also possible,
however, to write a more specific rule that covers just the
agentives, since these are not only polysyllabic and non-
neuter, but they also all have a stem-final [s] or [t]: doctor,
audi:tor, ce:nsor, etc.11 Thus, the more specific rule [o:ris] →
[or] / [X {s,t}] [polysyl,–neut] —— # is able to describe the
agentives quite narrowly, and has an extremely high reliability
(.980).
Page 28 of 53
Outside the class of polysyllabic non-neuter nouns, the

reliability of -r is much lower. Among polysyllabic neuters, very
few nominatives end in -r, so the general rule [ris] → [r] / [Xr]
[polysyl,+neut] ——# has a rather low confidence (.196). There
are, however, a few local pockets of -r nominatives among the
neuters, especially among those with -aris and -eris in the
genitive ([kalkar]/[kalka:ris] ‘spur-nom./gen.’, [nektar]/
[nektaris] ‘nectar-nom./gen.’, [aker]/[akeris] ‘maple-nom./
gen.’). This is the reason why the model disfavors innovative -r
nominatives among neuters, and why there is also a good deal
of item-by-item variation among them.12 Finally, among
monosyllables, almost all -ris genitives had nominatives ending
in -s—a notable exception being [fu:r] ‘thief’. Thus, the model
correctly learns that in this environment, the s ~ r alternation
is extremely robust, and final -r cannot compete with it.
It should be reiterated that the slight preference of the model

for -r in words like [honor] emerges in spite of the fact that
they were listed with -s in the training data. In other words,
the grammar produces an output that is different from the
existing form. Therefore, under this analysis, pre-change
forms like [honoːs] would have been considered irregular, and
would have had to have been listed as exceptions to the -ris ~ -
r pattern. Of course, if learners had perfect memories and
access to all forms of all words, then they could perfectly well
have memorized [honoːs] and continued to produce it, and the
language would not have changed. However, in real life this is
not the case, and speakers must sometimes synthesize new
forms. The model is intended to predict what forms a speaker
would produce in such situations, and in this case it correctly
predicts errors, or overregularizations, like honor. It cannot,
however, predict when existing forms would be unavailable,
forcing speakers to use their grammars. What is missing from
this model, then, is a production mechanism which uses both
the lexicon and grammar to produce forms. Even so, the result
in Fig. 2.4. is still significant, because we can assume that
speakers do sometimes make overregularization errors
(Marcus et al. 1992; Pinker 1999), and the errors that the
model makes correctly mirror the attested historical change.
It is worth mentioning that the [honoːs] ⇒ [honor] change was

just one of many that affected nominatives in the history of
Latin. Numerous nouns with highly irregular (p.36)
nominative forms were regularized; for example, the
nominative of [juppiter]/[jowis] ‘Jupiter-nom./gen.’ was
Page 29 of 53
eventually replaced by [jowis] (cf. [kanis]/[kanis] ‘dog-nom./

gen.’), and the nominative of [boːs]/[bowis] ‘cow-nom./gen.’ is
attested as [bowis] (Kieckers 1960: ii. s. I.21.3; Kühner 1912:
s. 63.2). The converse change, of regularizing nouns by fixing
the oblique forms, generally did not occur (*[juppitris] or
*[juppitri:], *[bo:ris], etc.).13
Furthermore, the form of most nouns in modern Romance

languages can be traced back to oblique forms in Latin. For
example, Latin [peːs]/[pedis] ‘foot-nom./gen.’ has yielded
Italian [pjede], instead of the expected *[pe] (cf. Latin [treːs] ⇒
Italian [tre] ‘three’); similarly [ars]/[artis] ‘skill’ ⇒ Ital. [arte],
[floːs]/[floːris] ‘flower’ ⇒ Ital. [ineq], [niks]/[nivis] ‘snow’ ⇒ Ital.
[neve], and so on. A full analysis of all of these changes is
beyond the scope of this study, but the computational model
presented here does prefer many of the attested changes as
regularizations in these cases as well.
2.4 DISCUSSION
Page 30 of 53
This analysis captures two common intuitions about Latin

nouns, and about the [honoːs] to [honor] change. The first is
that oblique forms are more revealing about the declension of
a noun than the nominative—seen, for example, in the common
practice of listing both the nominative and genitive forms in
dictionary entries, as the nominative alone is not considered
informative enough to predict the entire paradigm.14 This
intuition is reflected in the current analysis by the fact that an
oblique form is chosen as the base, and the remainder of the
paradigm is derived from an oblique form.
The second intuition is that the change from [honoːs] to

[honor] involved replacing a small, irregular, morphologized
alternation with the more general default pattern of non-
alternation. This analysis shares with Barr (1994) the idea that
this can be captured with competing rules, but differs with
respect to why different classes of words were treated
differently. In Barr’s system, a single small X s ~ X ris rule
competed with a single large X ~ X is rule; the fact that X s ~
X ris was lost among polysyllabic non- (p.37) neuters and was
retained in other cases is attributed to extra-grammatical
factors, such as salience and ‘degree of allomorphy’, which
made the rule harder to learn or apply in different contexts. In
the current system, on the other hand, the difference between
different genders and word lengths is attributed to the
existence of multiple versions of the rules in question, at
varying levels of generality, and with differing reliability in
different contexts. The use of multiple overlapping rules might
be seen as unwanted redundancy in the model, but in fact
cases such as Latin are taken as evidence that speakers, too,
have detailed knowledge about the reliability of different
processes in different environments. Furthermore, an ability to
assess the reliability of rules in different environments is
required in any event in order for learners to locate the best
rules to describe the patterns of their language.
Although this analysis makes use of several intuitions about

the factors that are thought to drive paradigm leveling, it
ignores certain other factors that have been proposed in the
literature. Some notable factors that do not play a role in this
analysis are the frequency of an allomorph within the
paradigm, the token frequency of various surface forms, or the
semantic naturalness of different nouns in different cases. It is
Page 31 of 53
useful to consider, therefore, the extent to which these other

factors could provide an alternative explanation of the honor
analogy, and whether the current model would benefit from
incorporating any of these factors.
2.4.1 Frequency of occurrence within the paradigm

It is often suggested that the [honoːs] to [honor] change was
encouraged by the fact that every form in the paradigm except
the nominative singular contained [r]; we may refer to this as
the ‘majority rule’ hypothesis (McCarthy, Ch. 8). Under the
model proposed here, each paradigm has a single unique base,
and forms are derived by grammars relating individual pairs of
forms. If the most informative form had turned out to be the
nominative singular, the prediction of this model is that all the
remaining forms could have been rebuilt on the basis of a
single form. Thus, this model has no way to capture the
majority rule intuition. It is not clear to me, however, that
there is evidence that paradigm leveling is truly driven by
majority rule. There are numerous cases in which a single
form seems to have driven a paradigmatic change (see e.g.
Albright 2002 for a case from Yiddish), whereas it is difficult to
prove that a leveling like [honoːs] ⇒ [honor] would not have
happened if [s] had occurred in more slots in the paradigm.
Furthermore, the majority rule hypothesis cannot explain
several aspects of the Latin facts. First, forms with [r]
outnumbered forms with [s] not only in the polysyllabic non-
neuter nouns, but in monosyllabic and neuter nouns as well.
From the point of view of paradigm-internal pressures, there is
no reason that these two classes of words should have
behaved differently.
In addition, there are many other noun paradigms in which the

nominative had a different form from the rest of the paradigm,
but was not leveled (e.g. [iter] ~ [itineris] ‘road-nom./gen.sg.’).
Therefore, frequency of occurrence within the paradigm does
not seem to add anything to the account of the change.
(p.38) 2.4.2 Token frequency of different paradigm members

A natural hypothesis, pursued by Manczak (1958) and others,
is that less frequent forms are often rebuilt on the basis of
more frequent forms within the paradigm. Could it be the case
that the nominative was actually a lot less frequent than the
oblique forms in Latin? This is especially relevant in Latin
because many or most of the words affected by the honor
Page 32 of 53
analogy were inanimate or abstract nouns, which are perhaps

more frequent in oblique forms than in the nominative.
In order to get a rough (and very informal) estimate of the

relative frequency of case forms for different nouns, I
performed some counts on the complete works of Cicero, as
found in the Perseus Digital Library (http://
www.perseus.tufts.edu, accessed 25 Feb. 2004). As (15) shows,
it is true that among singular forms, nominatives rarely
constitute the majority of tokens for any noun. This may help
to explain why nominative forms were open to rebuilding in
Latin—they were, on the whole, not so frequent that they
could always be reliably memorized and retrieved.
(15) Distribution of singular tokens for some Latin

nouns
Page 33 of 53
Noun Total Nom. Gen. Acc. Abl.

Sg. (%) (%) ( %) (%)
Polysyllabic, non-neuter
honoːs / honor ‘honor’ 285 19 25 18 32
laboːs / labor ‘work’ 163 17 21 37 35
Odoːs / odor ‘odor’ 4 50 0 0 50
Monosyllabic
floːs ‘flower’ 16 25 0 56 19
moːs ‘custom’ 146 18 2 17 63
oːs ‘mouth’ 65 29 9 6 55
Neuter
corpus ‘body’ 174 21 47 — 28
onus ‘burden’ 40 45 40 — 15
tempus ‘time’ 935 32 15 — 51
Masculine, agentive
reːx ‘king’ 207 23 21 27 18
homoː ‘man’ 1049 19 23 35 12
senaːtor ‘senator’ 43 33 23 28 14
Page 34 of 53
Note: A problem arises in counting frequencies for neuters, since the nom. and acc. forms are identical. The hypothesis
being tested here is that the frequency of [s] forms in the paradigm determines their susceptibility to leveling, so I have
counted all s forms in the nom. column, to facilitate comparison with the masc. and fem. nouns.
Page 35 of 53
What the counts at (15) cannot explain, however, is why the

change should have been restricted only to the non-neuter
polysyllabic nouns. The nominative does not (p.39) seem to
be less frequent in this class of nouns that in any other class.
Furthermore, there is apparently not even a difference
between masculine agentive nouns like ‘king’, ‘man’, and
‘senator’, and inanimate, abstract nouns like ‘honor’ or
‘custom’. Thus, a frequency-based account can explain only
the direction, but not the details of the [honoːs] ⇒ [honor]
change.
2.4.3 Semantics and local markedness

Another intuition, related to token frequency but logically
distinct from it, is that the semantics of particular lexical items
make them more ‘natural’ in some case forms than in others.
Tiersma (1982), for example, shows that singular forms in
Frisian have been rebuilt on the basis of plural forms, but just
for those nouns that occur more naturally in the plural than in
the singular (such as ‘teeth’ or ‘geese’). He refers to this
phenomenon as local markedness. On the whole, we would
expect local markedness to be reflected in token frequency,
which is much easier to measure (see above). I have no
estimate of the naturalness of the nominative forms which
changed from [-oːs] to [-or], but I see no reason why this would
fare any better than token frequency as an explanation of the
change. Nouns like [onus] ‘burden’, [korpus] ‘body’, and [floːs]
‘flower’ seem to me to be just as non-agentive as [honoːs]
‘duty’, [odoːs] ‘odor’, or [arboːs] ‘tree’. It appears that the
class of nouns that changed is best defined by prosodic and
morphological properties, and adding a sensitivity to
frequency or semantics would not improve the model’s
predictions in this case.
2.4.4 Leveling v. extending alternations

The analysis of paradigm leveling proposed here relies on a
strong pre-existing pattern of non-alternation in the lexicon—
in this case, the non-alternation of [r]. This proposal
immediately raises two related questions: first, if paradigm
uniformity is really just the extension of an existing pattern of
non-alternation, then what happens when the dominant
pattern is alternation? Why does there seem to be a universal
tendency towards leveling?
Page 36 of 53
As an example of a language with a dominant pattern of

alternation, consider a previous stage of Korean (S. E. Martin
1992):
(16) Conservative Korean consonant alternations
Page 37 of 53
/—— # Example /——V ACC [-i ɭ] gloss
[t ̚] [nat ̚] [d] [nadi ɭ] ‘grain’
[t ̚] [pat ̚] [th] [pathi ɭ] ‘field’
[t ̚] [t∫Λt ̚] [t∫] [t∫Λtd3iɭ] ‘milk’
[t ̚] [k’ot ̚] [t∫h] [k’ot∫hɭ] ‘flower’
[t ̚] [ot ̚] [s] [osi ɭ] ‘clothing’
Page 38 of 53
(p.40) As (16) shows, all stem-final coronal obstruents

alternate with [t ̚] word-finally. As with Latin, this alternation
could be expressed as the result of a markedness constraint
against manner and laryngeal specifications in coda position
(favoring [t ̚]) outranking faithfulness constraints (which
preserve underlying contrasts). If there was a universal
pressure for Uniform Exponence constraints to move above
IO-Faithfulness constraints, then we would expect that
paradigmatic changes in Korean should bring Korean closer to
non-alternating paradigms, perhaps as in (17). (The
intervocalic voicing of /t/→[d] is a completely predictable
process in Korean.) Note that although the phonotactics of
Korean rule out a completely non-alternating paradigm ([nat ̚]
~ *[nati ̵], or *[nad] ~ [nadi ̵), we may assume that the
relatively minor, predictable allophonic alternation between [t
̚] and [d] better satisfies Uniform Exponence than a [t ̚] ~ [s] or
[t ̚] ~ [t∫h] alternation, just as the shortening of final /o:r/→ [or]
in Latin is assumed to be a less serious violation of Uniform
Exponence than a [s] ~ [r] alternation is.
(17) Expected Korean paradigm leveling
Page 39 of 53
/——# Example /——V ACC [-i ɭ] gloss
[t ̚] [nat ̚] [d] [nadi ɭ] ‘grain’
[t̚] [pat ̚] [d] [padiɭ] ‘field’
[t ̚] [t∫Λt ̚] [d] [t∫Λdiɭ] ‘milk’
[t ̚] [k’ot ̚] [d] [k’odiɭ] ‘flower’
[t ̚] [ot ̚] [d] [odi ɭ] ‘clothing’
Page 40 of 53
In fact, the attested change in Korean noun paradigms is quite

different. As it turns out, the majority of coronal obstruent-
final stems contained [s] or [t∫h] etymologically (i.e., most
were like [k’ot ̚]/[k’ot∫hi ̵] or [ot ̚]/[os i ̵]), and many Korean
noun paradigms are being rebuilt to contain [s] or [t∫h]:
(18) Actual change in Korean Paradigms (Martin 1992;

Hayes 1995a)
Page 41 of 53
/——# Example /——V ACC [-iɭ] gloss
[t ̚] [nat ̚] [s] [nasi ɭ], ?? [nadi ɭ] ‘grain’
[t ̚] [pat ̚] [th], [t∫h], [s] [pathi ɭ], [pathi ɭ], [pasi ‘field’
ɭ]
[t ̚] [t∫Λt ̚] [s], [t∫] [t∫Λsiɭ], [t∫Λd3iɭ] ‘milk
[t ̚] [k’ot ̚] [t∫h], [s] [k’ot∫hi ɭ], [k’osi ɭ] ‘flower’
[t ̚] [ot ̚] [s] [osi ɭ] ‘clothing’
Page 42 of 53
Although there is a considerable amount of word-by-word and

speaker-by-speaker variation, it is clear that the restructuring
underway in Korean is introducing, not eliminating
alternations. For the most part, the dominant alternations of [t
̚] ~ [s] and [t ̚] ~ [t∫h] are coming to replace other, arguably
less drastic alternations like [t] ~ [d].
(p.41) The explanatory challenge, therefore, is to explain

why in some cases a pattern of alternation is extended (as in
Korean), while in other cases, alternations are eliminated (as
in Latin). The model of paradigm learning advocated here
always extends the strongest pattern, regardless of whether it
is alternating or uniform. The reranking of paradigm
uniformity constraints, on the other hand, can explain only
leveling; the spread of alternations would have to be handled
by other means, such as anti-correspondence constraints
(Hayes 1999), leaving us with no explanation for why
sometimes paradigm uniformity wins out, and sometimes anti-
correspondence wins out.
The contrast between Latin and Korean may also help to shed
light on why there seems to be a typological bias towards
paradigm leveling; in Latin, rhotacism affects just one segment
(/s/), so the only words that exhibit the alternation are those
with stem-final [s]. In Korean, on the other hand, coda
neutralizations affect all obstruents, so almost all lexical items
exhibit alternations. I conjecture that morphophonemic
alternations typically affect only a smallish subset of the
phonemic inventory; rhotacism affects just /s/, umlaut affects
only back vowels, palatalization tends to affect just coronals
(and often just coronal stridents), and so on. The end result is
that alternations tend to be the minority pattern, and there
will be a tendency to generalize non-alternation, or in other
words, to level.
Page 43 of 53
2.5 CONCLUSION
I have argued here that the Latin [hono?s] ⇒ [honor] change
was caused by more than simply a sporadic pressure for
paradigm uniformity or Uniform Exponence constraints to
assert themselves over IO-Faithfulness constraints. I have
shown that the spread of [r] to nominative forms did more
than just create uniform paradigms; it also extended a pattern
of non-alternation that was already dominant in the lexicon.
Details of the change, such as its restriction to polysyllabic
nouns and non-neuters reflect the fact that these were
especially strong contexts for [r] stems. Furthermore, the
‘backwards’ direction of the leveling, with oblique forms
influencing the nominative singular, can be explained by a
particular model of Latin noun paradigms, in which an oblique
form served as the base, and nominative forms were derived
from oblique forms by rules operating on surface forms.
More generally, this result provides evidence for a model of

paradigm learning in which learners choose the base form that
is the most informative—i.e. that preserves the most
distinctions between classes of words, and allows the
remainder of the paradigm to be predicted with the greatest
accuracy and confidence. This echoes a proposal by Lahiri and
Dresher (1984) that certain forms in the paradigm ‘matter
more than others’ to learners when they are determining what
class a word belongs to. The remainder of the paradigm is
then derived using a grammar of gradient morphological rules,
providing a measure of the relative strength of different
patterns, and allowing different patterns to have different
productivity in various contexts. The prediction of (p.42) this
model is that distinctions that are preserved in the base form
will be easily learned and maintained, whereas distinctions
that are neutralized in the base form may be lost by leveling or
regularization.
APPENDIX: METRICS FOR BASE SELECTION

The table lists all the effectiveness measures of each of the six
candidates for base status, based on the 494 most frequent
Latin nouns. Rows indicate the input forms, and columns
indicate the output forms; for example, the average winner
confidence for the nom. → gen. mapping is 0.76, in the upper
left.
(p.43)
Page 44 of 53
↓In/Out→ nom.sg. gen.sg. dat.sg. acc.sg. abl.sg. nom.pl. TOTAL
Av. winner nom.sg. 0.76 0.70 0.86 0.82 0.76 0.78

conf.
Av. margin 0.66 0.59 0.77 0.68 0.65 0.67
Av. 5.68 6.44 3.57 5.76 6.15 5.52

competitors
Percentage 0.82 0.77 0.87 0.85 0.80 0.82

correct
No. of 2341 2618 1987 2340 2308 2319

constraints
Av. grammar 0.35 0.31 0.45 0.33 0.33 0.35

conf.
Av. winner gen.sg. 0.88 0.97 0.94 0.96 0.95 0.94

conf.
Av. margin 0.69 0.92 0.91 0.91 0.79 0.84
Av. 3.16 1.92 1.48 2.89 2.38 2.37

competitors
Percentage 0.93 0.97 0.96 0.96 0.94 0.95

correct
No. of 1877 2781 2253 2609 2618 2428

constraints
Page 45 of 53
Av. grammar 0.45 0.70 0.67 0.69 0.53 0.61

conf.
Av. winner dat.sg. 0.86 0.97 0.94 0.96 0.95 0.94

conf.
Av. margin 0.69 0.93 0.89 0.91 0.81 0.85
Av. 3.68 1.83 2.99 2.69 3.25 2.89

competitors
Percentage 0.92 0.98 0.96 0.97 0.96 0.96

correct
No. of 1920 2418 2538 2805 3377 2612

constraints
Av. grammar 0.43 0.81 0.59 0.65 0.43 0.58

conf.
Av. winner acc.sg. 0.89 0.91 0.90 0.91 0.90 0.90

conf.
Av. margin 0.76 0.82 0.80 0.79 0.77 0.79
Av. 3.29 5.40 7.01 3.79 5.39 4.98

competitors
Percentage 0.95 0.94 0.92 0.93 0.92 0.93

correct
Page 46 of 53
No. of 1504 2045 2460 2467 2465 2188

constraints
Av. grammar 0.55 0.73 0.60 0.60 0.53 0.60

conf.
Av. winner abl.sg. 0.86 0.96 0.96 0.95 0.94 0.93

conf.
Av. margin 0.68 0.93 0.90 0.91 0.85 0.85
Av. 2.76 3.36 1.65 1.29 2.45 2.30

competitors
Percentage 0.92 0.97 0.97 0.97 0.95 0.95

correct
No. of 1829 2377 2514 2087 2398 2241

constraints
Av. grammar 0.45 0.78 0.72 0.72 0.56 0.65

conf.
Av. winner nom.pl 0.84 0.94 0.92 0.92 0.92 0.91

conf.
Av. margin 0.69 0.76 0.81 0.86 0.81 0.79
Av. 2.61 1.77 3.24 2.11 2.95 2.54

competitors
Page 47 of 53
Percentage 0.89 0.94 0.94 0.94 0.94 0.93

correct
No. of 1424 1855 2260 1955 2010 1901

constraints
Av. grammar 0.50 0.69 0.56 0.66 0.62 0.61

conf.
Page 48 of 53
Notes:
(1) There is some evidence for an intermediate stage in the
late pre-Classical stage in which [honoːr] with long [oː] was at
least one possible variant of the nominative form. Early poets
such as Plautus and Enni-us still allowed final long [oːr], using
it to satisfy metrical requirements for heavy syllables, and also
sometimes used [-o(ː)r] instead of [-oːs], e.g. Ennius 545
(Skutsch 1985) Clām[ōː]r ād cǣlūm vōlvēndŭs pěr ǣthěră
vāgît; also Ennius 409, 428. As far as I know, there is no
evidence concerning how ‘clean’ this intermediate stage was—
the shortened variant [honor] may also have been used from
the very beginning of the [s] ⇒ [r] change. The analysis that I
propose here is compatible with the existence of an
intermediate [honoːr] stage, but does not rely on it. It would,
however, make a crucial difference for some other analyses,
such as the Uniform Exponence analysis sketched below.
(2) It is difficult to know in many cases whether the

occurrence of forms like [honoːs] alongside [honor] reflects
free variation, or simply literary archaism. Cicero, for
example, systematically used the form [honoːs] instead of
[honor], in both philosophy and oratory texts (elevated styles),
as well as in letters (potentially less elevated/archaic); at the
same time, he used [labor] instead of [laboːs] in all three
contexts.
(3) A potentially relevant observation is that the paradigm of

[honoːs] already had [r] forms in it, while the paradigm of
[soror] never had [s] forms. Following Steriade (1997a) we
might call this the ‘lexical conservatism analysis, in which
speakers may use or extend only allomorphs that are already
attested. It is possible that a lexical conservatism analysis
could explain this part of the asymmetry ([honor] but not
*[soroːs]), but there are hints that Latin speakers were not
always bound by lexical conservatism—for example, see n. 12
regarding the use of [femus] for [femur], which otherwise
never had an [s] anywhere in the paradigm. The lexical
conservatism analysis would also tell us nothing about (7d),
since monosyllables and neuters also had an available [r]
allomorph which could have been extended to the nominative
singular.
Page 49 of 53
(4) I will leave aside the possibility of storing abstract roots (/

glupt-/, /nok-/, etc.), because such an ability would not help
significantly in this hypothetical language; we could have a
single nominative rule adding [-us] to roots, but the words
with [-oris] genitives would either have roots [rept-], [kort-],
requiring a separate [-oris] rule for the genitive, or they would
have roots [reptor-] and [kortor-], requiring readjustment rules
to delete the [or] in the nominative. A further possibility is to
list roots with diacritics (/glupt-/[i: class], /rept-/[oris class], etc.),
but in this case, too, we must learn what the classes are, and
which forms to look at in order to determine the class that
each root belongs to. Thus, no matter which style of lexical
representation is preferred, the learner must somehow
determine that there is unpredictable information in the
genitive that must be encoded in some fashion in the lexical
Entry.
(5) Recent research on infants has shown that children are

able o distinguish phonotactically illegal sequences from legal
ones at a remarkably early age (Friederici and Wessels 1993;
Jusczyk et al. 1993; Jusczyk, Luce, and Luce 1994); see Hayes
(2004) and Smolensky, Davidson, and Jusczyk (forthcoming)
for overviews, and specific proposals for how phonotactic
distributions could be encoded grammatically at the earliest
stages.
(6) All the training input sets and results files for the
simulations discussed here, as well as the original database of
nouns, can be downloaded from http://ling.ucsc.edu/~albright/
papers/latin.html, accessed 25 Feb. 2004.
Page 50 of 53
(7) One possible objection is that the presence of intervocalic

[s] does not necessarily preclude the possibility that rhotacism
continued to be a synchronically active process in Latin,
perhaps restricted to a particular morphological environment,
such as /V —— +V, where ‘+’ indicates a morpheme boundary,
For the purposes of the current model, the synchronic status
of rhotacism actually makes very little difference. Including a
*Vs+V constraint would improve the reliability of nominative →
oblique grammars slightly, because the model could learn to
apply rhotacism in mappings such as [honoːs] → [honoːris]
(instead of [hono#x002D0;sis]); however, -oːs nominatives
make up only a small fraction of the language as a whole, so
improving the model’s predictions for this subset of the
vocabulary does not make a substantial difference in the
calculations reported below. Note also that including a
rhotacism constraint does not help the model at all in the
oblique → nominative direction, since an oblique form with -
oːris-oːris could come from either underlying /s/ or underlying /
r/. More generally, assuming that rhotacism was
synchronically active in Latin can help to explain why the
paradigm of ‘honor’ was not leveled to [honoːs], [honoːsis], etc.
with underapplication of rhotacism, but it cannot explain other
facts, like why speakers did not assume that forms like
[soro:ris] were also the result of rhotacism (predicting the
incorrect nominative [soroːs]), or why speakers did not simply
tolerate the rhotacism alternation to remain. It could also be
added that many authors have tried to make use of the exact
opposite intuition: by recognizing that rhotacism was no
longer synchronically active in Latin, we can understand why
[honoːs] and [honoːris] could no longer be related to one
another by an automatic phonological process, and why the
alternation was then open to leveling (Klausenburger 1979;
Wetzels 1984; and others).
(8) In fact, the synchronic status of many i-stems seems

doubtful; Allen (1903) writes that ‘The i-declension was
confused even to the Romans themselves, nor was it stable at
all periods of the language, early Latin having i-forms which
afterwards disappeared’ (s.73). For this reason, some of the
nouns listed by Allen as i-stems were not listed as such in
other sources, and were included in the database without the
[i].
Page 51 of 53
(9) A common intuition is that the [honoːs] ⇒ [honor] change

may be due to the collective influence of all the oblique forms
combined, and not the effect of a single oblique form on the
nominative singular. I will discuss this possibility further in s.
2.4.1.
(10) The two-stage nature of this model—first selecting a base

and then honing the grammar—may not be crucial. In the case
of Latin, and in several other cases examined, it turns out that
the asymmetries between forms seen in Fig. 2.2. are clearest
when fewer words have been learned, but do not change
quantitatively when the full data set is considered.
(11) This common phonological trait is not a coincidence:

agentives were formed from the perfect passive participle (4th
stem), which was generally formed by adding a [t], or in some
phonological contexts, by changing a stem consonant to [s]
(e.g. [kad-] → [kaːs-] ‘fall’).
(12) In fact, Kieckers (1960: ii s. 22) points to one, possibly

quite isolated example of a neuter noun with etymological –r
being written with an -s : femus ‘femur’.
(13) There seems to be at least one set of cases in which a

property of the nominative was extended to the remainder of
the paradigm: the paradigm of words like [vo:ks] ‘voice’
originally had a long [oː] and [k] in the nominative, and a short
[o] and [kw] elsewhere ([vo:ks], [vokwis], etc.; see Meiser
(1998: 141) regarding vowel length, Leumann (1977: 148) and
Kieckers (1960: II.13) regarding [kw]). The long [oː] and simple
[k] of the nominative were subsequently extended to all forms:
[vo:ks], [vo:kis]. It would be interesting to compare the
relative chronology of these changes, since these nominative-
driven changes may have occurred at an older stage of the
language in which nominatives suffered from fewer
phonological reductions, while oblique-driven changes
occurred throughout the Classical and Late Latin period.
(14) The relative uninformativeness of the nominative in Latin

is due, in part, to the fact that the nominative suffix for one
large class of nouns lacked a vowel (-s), creating coda clusters
that resulted in phonological simplifications (e.g., *arts ⇒ ars).
The oblique forms always provided a prevocalic context for the
Page 52 of 53
stem, resulting in far fewer neutralizations; rhotacism is a rare

exception.
Page 53 of 53
Competing Principles of Paradigm Uniformity: Evidence from the Hebrew
Imperative Paradigm


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Competing Principles of Paradigm

Uniformity: Evidence from the
Hebrew Imperative Paradigm
Outi Bat-El
DOI:10.1093/acprof:oso/9780199267712.003.0003

Paradigms can be non-uniform in different ways. This chapter
shows the differences within languages and across two stages
of a language. The difference between Colloquial and Tiberian
Hebrew is in the ranking of markedness and faithfulness
constraints. In Colloquial Hebrew, faithfulness constraints are
ranked above markedness constraints, and the imperative is
thus identical to the future base (with the exclusion of the
truncated material). That is, Structural Identity is respected.
In Tiberian Hebrew, markedness constraints are ranked above
faithfulness constraints, and the imperative is thus
segmentally (stop v. fricative) and prosodically (epenthesis and
deletion) different from the future base. That is, Structural
Identity is violated. Another effect of the dominance of
faithfulness constraints in Colloquial Hebrew is the violation of
Existence. When the imperative cannot be identical to its base,
there is no imperative and Existence is violated. Such an effect
Page 1 of 60
Imperative Paradigm
is not found in Tiberian Hebrew since faithfulness constraints

are low ranked.
Keywords: Colloquial Hebrew, Tiberian Hebrew, paradigms, Existence,

faithfulness constraints
3.1 INTRODUCTION
Crystal (1985: 220) defines a paradigm as ‘a set of forms
having a common root or stem’, where each form appears ‘in a
certain grammatical environment’. A paradigm thus expresses,
according to van Marle (1985: 225), ‘the ways in which
linguistic entities may be mutually connected’. These
definitions set the basis for the principles of paradigm
uniformity proposed in s. 3.1.2.
In concatenative morphology, the common root/stem is usually

apparent, and segmental alternation follows a general
principle (e.g. Turkish case paradigm: ‘rope’: ip ‘nom. sg.’, ipin
‘gen. sg.’, ipler ‘nom. pl.’, iplerin ‘gen. pl.’; ‘stamp’: pul ‘nom.
sg.’, pulun ‘gen. sg.’, pullar ‘nom. pl.’, pullarn ‘gen. pl.’). I use
here the notion of relation, as discussed in Anderson (1992), to
indicate the formal distinction between the forms in the
paradigm (e.g. Turkish pulun is related to pul by the addition
of the suffix -un).
In Hebrew, the stem is relatively apparent in the person–

number–gender paradigm, as exemplified in (1a) by the
relation between the 3rd ms.sg. and the 2nd ms.sg. forms (p.
45) of the past. In this case, suffixation (-ta) is accompanied
by lowering the vowel in the ultimate stem syllable. In the
tense paradigm, however, exemplified in (lb) by the relation
between the past and future forms of the 2nd ms.sg., there is a
larger variety of alternations beyond the addition of the prefix
(tV-): the vocalic pattern can be distinctive in one or two
vowels (e.g. hit-labaš-ta—tit-labeš, xipas-ta—te-xapes), and the
prosodic structure of the stem can be different as well (e.g.
gadal-ta—ti-gdal). Although these alternations are typical of
Semitic morphology, they are not found in the future-
imperative relation in (1c). First, this relation is not expressed
by affixation, but rather by truncation of CV or V from the
future prefix. Second, the truncated imperative preserves the
vocalic pattern and the prosodic structure of the 2nd-person
stem.
(1)
Page 2 of 60
Competing Principles of Paradigm Uniformity: Evidence from the Hebrew Imperative Paradigm
Hebrew inflectional paradigm (only the relevant categories)

a. b. c.
Past 3ms.sg. Past 2ms.sg. Future 2ms.sg. Imperative ms.sg.
Gadal gadal-ta ti-gdal Gdal ‘to grow’
Xipes Xipas-ta te-xapes txapes ‘to search’
Hitlabeš hit-labaš -ta tit-labeš Tlabeš ‘to get dressed’
Page 3 of 60
Imperative Paradigm
The preservation of the phonological structure of the base may

suggest that the imperative paradigm in Colloquial Hebrew is
uniform.1 However, this is not the case. First, in some cases
the future form does not have an imperative counterpart.
Second, some feminine imperative forms are not derived from
the future by truncation but rather from the truncated
masculine by suffixation. I argue that this non-uniformity
results from the competition among the principles of paradigm
uniformity (presented in s. 3.1.2), and the priority of some
markedness (surface structure) constraints. In order to detect
the competition, it is necessary to admit the distinction
between main paradigm and sub-paradigm (given in s. 3.1.1),
since a principle respected in the main paradigm is violated in
the sub-paradigm, and vice versa. The approach in terms of
competition is drawn from Optimality Theory (Prince and
Smolensky 1993), which is used for the analysis of the data
presented here, from both Colloquial Hebrew (ss. 3.2 and 3.3)
and Tiberian Hebrew (s. 3.4).
3.1.1 Paradigms and sub-paradigms

As suggested in van Marle (1985), a paradigm can be viewed
as a network of words related on two axes, one (vertical) for
the shared grammatical category of the words, and the other
(horizontal) for the shared lexeme (the latter also expresses
the syntagmatic (p.46) relation in additive morphology). A
paradigm often expresses embedded relations, in which case it
consists of a main paradigm and a sub-paradigm (or sub-
system, in Wurzel’s (1989) terms). The paradigm of Colloquial
Hebrew imperatives, presented in (2), is an embedded
paradigm. The main paradigm includes the future and the
imperative, and the sub-paradigm includes the masculine and
the feminine of each category in the main paradigm (stress is
final unless otherwise specified; all future forms are in the 2nd
person, and past forms, which will be relevant later on, are in
the 3rd person).2
(2)
Page 4 of 60
The paradigm of Colloquial Hebrew imperative truncation

Future Imperative MAIN PARADIGM
Masculine Feminine Masculine Feminine SUB-PARADIGM
tiftax tiftexi ftax ftexi ‘to open’
Tegadel tegadli tgadel tgadli ‘to raise’
Page 5 of 60
Imperative Paradigm
Based on the phonological and morphological structure of the

words in a paradigm, a generative morphological analysis will
arrive at one of the following conclusions regarding the
relation exhibited by every future–imperative pair on the
horizontal axis: (i) The related words are derived from some
abstract underlying form, or (ii) one of the words is derived
from the other. The paradigm in (2) is of the latter type, where
the imperative is derived directly from the corresponding
future form of the same gender (see arguments in s. 3.2.1). In
Optimality Theoretic terms, in particular Correspondence
Theory (McCarthy and Prince 1995), such a relation between
surface forms is referred to as output–output correspondence,
where one of the outputs serves as the base for the other (see
also Kenstowicz 1996; Benua 1997a).
As illustrated in (3), the paradigm under consideration

consists of two types of structural relations: (i) future–
imperative for each gender, which is expressed by truncation,
and (ii) masculine future–feminine future, which is expressed
by suffixation (in each pair, the first mentioned is the base).
The masculine and feminine imperative forms are only
indirectly related, but as will be shown in s. 3.3.2, there are
cases where they are directly related.
(3)
Base-output relations in the imperative

paradigm (truncation⇒suffixation→)
Masculine: Future ⇒ Imperative tiftax-ftax
↓ tiftax-tiftexi
Feminine: Future ⇒ Imperative tiftexi-ftexi
(p.47) 3.1.2 Paradigm Uniformity

A uniform paradigm obeys the principles in (4).
(4) Given a paradigm with categories A and B, where B

is derived from A:
a. All forms in B are derived from A—Structural
Relation
b. Every form in B is phonologically identical to
its counterpart in A—Structural Identity.3
Page 6 of 60
Imperative Paradigm
c. Every form in A has a counterpart in B (i.e. no

gaps)—Existence.
Structural Relation (4a) and Existence (4c) express a stronger

version of van Marle’s (1985) view of the paradigm as a
network (see beginning of s. 3.3.1), as they require a perfect
network. Similarly, Structural Identity (4b) requires not only a
common stem, as suggested in Crystal’s (1985) definition of a
paradigm, but also an identical stem.
However, more often than not, there are non-uniform

paradigms, which result from competition among the
principles in (4). The priority of the principles in the
imperative paradigm will turn out to be as shown in (5).
(5) Priority of principles in the imperative paradigm
That is, in
both
paradigms,
Structural
Identity is
respected.
However, in the main paradigm, Existence is violated, while
Structural Relation is respected, and in the sub-paradigm it is
the other way around.
When Structural Identity cannot be obtained due to the

priority of markedness constraints, gaps are encountered in
the paradigm, i.e. a future form lacks an imperative
counterpart, and thus Existence is violated. In the absence of
an imperative form, both Structural Identity and Structural
Relation are vacuously respected.
While gaps are allowed in the main paradigm, they are

prohibited in the sub-paradigm. That is, while a future form
can lack an imperative counterpart (in which case all the
imperative forms of the stem are missing), the presence of a
masculine imperative requires the presence of a feminine
imperative counterpart. Such a situation arises when
markedness constraints rule out the feminine but not the
masculine form. To circumvent the problem, the feminine
imperative is formed from the masculine imperative, rather
than the feminine future. That is, Structural Relation is
Page 7 of 60
Imperative Paradigm
violated, but Existence is respected, and so is Structural

Identity, which is now obtained by the masculine–feminine
pair.
(p.48) The ensuing discussion provides a detailed analysis of

the relevant data. Evidence for the direct relation between the
future and the truncated imperative forms is provided in s.
3.2.1, followed in s. 3.2.2 by an analysis of imperative
truncation within the framework of Optimality Theory. The
paradigmatic gaps are accounted for, in s. 3.3.1, as a result of
constraint interaction, and the gapless sub-paradigm is
presented in s. 3.3.2. The imperative paradigm of Colloquial
Hebrew is compared in s. 3.4 with its parallel in Tiberian
Hebrew, with emphasis on the distinction between Tiberian
Hebrew imperatives without a prefix (s. 3.4.1) and with a
prefix (s. 3.4.2). It is argued that the non-uniformity of the
paradigm in Tiberian Hebrew is due to multiple
correspondences, i.e. that there is reference to both the future
and the past forms (s. 3.4.3). Concluding remarks are given in
s. 3.5.
3.2 IMPERATIVE TRUNCATION

This section provides an analysis of the main features of
Colloquial Hebrew imperative truncation. It begins with the
identity effects, which provide evidence that the future form is
the base of the imperative (s. 3.2.1), and then proceeds with
the analysis of truncation within the constraint-based
approach of Optimality Theory (s. 3.2.2).4
3.2.1 Identity effects as evidence for the base

The examples in (6) below provide evidence that the base of a
Colloquial Hebrew truncated imperative (hereafter TI) is the
corresponding 2nd-person future form.
(6)
Page 8 of 60
Imperative truncation
Masculine Feminine
Base: Output: Base: Output:
Future TI Future TI
a. CV truncation tiftax ftax tiftexi ftexi ‘to open’
titlabeš tlabeš titlabši tlabši ‘to dress’
b. V truncation tevakeš tvakeš tevakši tvakši ‘to request’
tikanes tkanes tikansi tkansi ‘to enter’
Page 9 of 60
Imperative Paradigm
The structural identity between the base and the TI points

unequivocally towards the 2nd-person future form as the base
of the TI. The TI reflects identity effects (Benua 1997a) by
preserving phonological properties of the base, which could
not be (p.49) derived independently by the phonology of the
language or be drawn from another base (Bolozky 1979; Bat-El
2002).
The first argument is based on the presence of an initial

fricative in TIs such as ftax ‘open!’ (6a) and a post-consonantal
fricative in TIs such as tvakeš ‘request!’ (6b); these fricatives
cannot be drawn from the phonology of the language. A
fricative in Hebrew is either phonemic or derived by post-
vocalic spirantization. Post-vocalic spirantization is opaque in
the current stage of the language, but it is still possible to
identify whether or not a fricative is phonemic (Adam 2002). A
phonemic fricative appears throughout the tense paradigm, as
in xipes, mexapes, yexapes ‘to search Past, Participle, Future’.
When the paradigm exhibits stop–fricative alternation, as in
kibes, mexabes, yexabes ‘to launder Past, Participle, Future’,
the phoneme is the stop and the fricative is derived by
spirantization.5 It is clear that the fricatives in ftax and tvakeš
are not phonemic, since there is a stop in their corresponding
past forms patax and bikeš respectively. It also cannot be the
result of post-vocalic spirantization, since the fricative is in
initial position in ftax and in post-consonantal position in
tvakeš. It must then be the case that the fricative in ftax and
tvakeš is drawn directly from the future forms tiftax and
tevakeš respectively, where the post-vocalic environment of
spirantization is met.
The second argument is based on the presence of an initial t in

TIs such as tvakeš (6b); this t is neither a stem consonant nor
an imperative prefix. The claim that it is not a stem consonant
is based, again, on the past form bikeš, which does not include
the t. This t is also very unlikely to be an imperative prefix
since all verbal prefixes in Hebrew are followed by a vowel.
The t in the TIs in (6 b) is thus drawn from the future base,
where it functions as a future prefix, with the exception of
tlabeš, where the t is a derivational prefix, but certainly not
part of the stem (cf the related verb lavaš ). Notice that the
arguments above rely on the accessibility of the entire
paradigm; i.e. the source of the segments (stem or affix) and
Page 10 of 60
Imperative Paradigm
their phonemic quality (fricative or stop) is based not only on

the future–imperative relation, but also on the past–future
relation. This issue is further discussed in s. 3.4.3.
A TI thus differs from its corresponding base in the truncated

segments, which may cause syllabic rearrangement in the first
syllable of the imperative form (e.g. tif. tax–ftax). Otherwise, a
TI is structurally identical to its corresponding base: (i) there
is no segment in the TI that does not appear in the base (i.e.
no epenthesis), (ii) every segment that appears in the TI is
identical to its correspondent in the base (i.e. no segmental
alternation), and (iii) the position of stress in a TI is identical
to that in the corresponding base (i.e. no stress shift).
3.2.2 The analysis of imperative truncation

Both CV truncation (6a) and V truncation (6b) are derived by
the same constraint ranking: ONSET, *[ΣCCC, IMPTRUNC Â%¯
MAXSEG. The constraints are stated in (7).
(p.50)
(7)
a. Markedness constraints:
• ONSET (Prince and Smolensky
1993) A syllable has an onset.
• *[σCCC (a member of *COMPLEX;
Prince and Smolensky 1993) A
triconsonantal onset is prohibited.
b. Morphological constraint:
• IMPERATIVE TRUNCATION
(IMPTRUNC) (anti-MAX; see
discussion below) Not every segment
in the base has a correspondent in
the output.
c. Faithfulness constraint:
• MAX SEGMENT (MAXSEG)
(McCarthy and Prince 1995) Every
segment in the input has a
correspondent in the output.
The most important competition is between the morphological

constraint IMPTRUNC and the faithfulness constraint
MAXSEG. IMPTRUNC does not specify how many segments to
truncate; rather, it states that at least one segment must be
Page 11 of 60
Imperative Paradigm
truncated. MAXSEG penalizes for every truncated segment,

and therefore satisfaction of IMPTRUNC requires at least one
violation of MAXSEG. Thus, in order for truncation to be
surface true, IMPTRUNC must outrank MAXSEG. Notice,
however, that although low ranked, MAXSEG minimizes the
number of the truncated segments, since IMPTRUNC is
equally satisfied when one or two segments are truncated.
The tableaux at (8) demonstrate how the constraint ranking

proposed above selects the optimal candidates (for ease of
exposition, the truncated material is enclosed in angled
brackets).
(8)
Page 12 of 60
CV truncation—in bases with an initial CVCσ]: Future tiftax–TI ftax ‘to open’
ti-ftax IMP ONSET *[σCCC IMP TRUNC MAX SEG
a. tiftax *!
b. tiftax *! *
c. tiftax *! *
d. ### tiftax **
e. tiftax ***!
Page 13 of 60
Imperative Paradigm
The distinction between the two types of truncation, CV vs. V

truncation, is due to the structure of the first syllable in the
base; the bases that undergo CV truncation have an initial
closed syllable (8), and those that undergo V truncation have
an initial open syllable (9). In (9), the optimal candidate (c)
undergoes V truncation, while in (8), the parallel candidate (c)
is ruled out by *[σCCC, as truncation of V would result in an
impermissible triconsonantal cluster. In both cases, ONSET
rules out the onsetless candidates (b), and MAXSEG, as noted
above, minimizes the number of the truncated segments. Of
course, the candidate faithful to the base (9a) does not satisfy
IMPTRUNC and is thus eliminated. However, as will be shown
in s. 3.3.1, there are cases where the faithful candidate is
optimal, in which case there is no TI.
(p.51)
(9)
Page 14 of 60
V truncation—in bases with an initial CVσ]: Future tevakeš–TI tvakeš ‘to request’
te-vakeš IMP ONSET *[σCCC IMP TRUNC MAX SEG
a. tevakeš *!
b. tevakeš *! *
c. #### tevakeš *
d. tevakeš **!
e. tevakeš *! ***
Page 15 of 60
Imperative Paradigm
As argued in Bat-El (2002, 2003), IMPTRUNC is a

morphological constraint, activated by the specification of the
property required of the output (IMP, specified on the base);
that is, there is a matching of the feature IMP(erative),
specified in the base and the constraint. This follows Russell
(1995, 1999) and Hammond’s (1995, 1998) approach to
morphology in Optimality Theory, where all morphological
processes, including affixation, are accounted for by
morphological constraints (see also Yip 1998; Adam and Bat-El
2000). As proposed in Alderete (1998, 2001), morphological
constraints are anti-faithful, since their purpose is to create a
paradigmatic contrast (see also Bat-El 2002, 2003).
IMPTRUNC is thus the anti-faithfulness counterpart of the
faithfulness constraint MAXSEG, and the two compete as
illustrated above.
Although V is truncated when the base begins with an open

syllable (9), there are some verbs, whose stem is monosyllabic,
where a CV is truncated in this same environment.
(10)
Page 16 of 60
CV truncation in bases with an initial CV]σ

Masculine Feminine
Base: Future Output: TI Base: Future Output: TI
taruc ruc *truc tarúci rúci *trúci ‘to run’
tikax kax *tkax (tikxi) (kxi) ‘to take’
Page 17 of 60
Imperative Paradigm
There are no markedness constraints which can rule out the

complex onsets in the ill-formed TIs in (10), as such onsets are
acceptable in other TIs (e.g. tkabel ‘receive!’, traxem ‘have
pity!’).
The apparent exceptions in (10) are due to the position of

stress. Following the generalizations summarized in (11)
below, a stressed syllable in a TI is segmentally identical to its
corresponding stress syllable in the base (boxes A, B, and D).
This structural identity cannot, however, be obtained by
truncating a stem consonant (box C). That is, the stressed
syllables in the base and the TI are identical, unless this
identity has to be obtained by deleting a stem segment.
(p.52)
(11)
Page 18 of 60
Generalization (properties refer to the TIs)

Source of the Position of the complex onset
1st C
In an unstressed σ In a stressed σ
Input: Future Output: TI Input: Future Output: TI
Stem ti-t.fe.rí tfe.rí A ti-t.fór tfór C
ti-k.xí kxí
Prefix tit-.ka.rév tka.rév B ta-.kúm kúm *tkúm D
tit-.kar.ví tkar.ví ta-.kú.mi kú.mi *tkú.mi
Page 19 of 60
Imperative Paradigm
The constraint responsible for the segmental identity of the

corresponding stressed syllables is FAITH STRESSED
SYLLABLE (Steriade’s 1999a LEXσ́), and the one blocking
truncation of a stem consonant is MAX SEGMENT STEM.6
(12)
a. FAITH STRESSED SYLLABLE (FAITH[ineq])
Corresponding stressed syllables are identical
b. MAX SEGMENT STEM (MaxSegS) (Beckman
1997) Every segment in the stem of the base has
a correspondent in the output
The crucial ranking of these two constraints is illustrated at

(13) for FAITH [INEQ]Â¯ MAX-SEG and (14) for MAXSEGSÂ¯
FAITH [INEQ].
(13)
Page 20 of 60
Future tikax—TI kax ‘to take’ (Faithσ́Â¯ MAXSEG)

ti.káx IMP ONSET *[σCCC IMPTRUNC MAXSEGS FAITHΣ́ MAXSEG
a. tit.fór *!
b. <t>it.fór *! *
c. titfór *! *
d. <t>itfór **
e. <tit>fór *! * * ***
Page 21 of 60
Imperative Paradigm
The addition of FAITHσ́ above MAXSEG in (13) allows

selecting the optimal candidate (d) out of the two candidates
(c and d ) that do not violate the high-ranked constraints;
without FAITHσ́, the candidate with the complex onset (c)
would be wrongly selected. In (14), however, the CVC
candidate (e), which preserves the structure of the stressed
syllable in the base, violates MAXSEGS and is thus ruled out,
allowing the candidate with the complex onset (d) to win.
(p.53)
(14)
Page 22 of 60
Future titfor—TI tfor ‘to sew’ (MAXSEGSÂ¯ FAITH FAITHΣ́)

tit.káx IMP ONSET *[σCCC IMPTRUNC MAXSEGS FAITHΣ́ MAXSEG
a. tit.fór *!
b. <t>it.fór *! *
c. ttfór *! * *
d. <ti>tfór * **
e. <tit>fór *! ***
Page 23 of 60
Imperative Paradigm
In addition to FAITHσ́ and MAXSEGS, the structural identity

between the future base and its corresponding TI is due to the
following faithfulness constraints, which are high-ranked and
always satisfied in the future–TI relation. In the terminology
used in Kenstowicz (1996), these are the base-identity
constraints responsible for the uniformity of the paradigm.
(15) Undominated faithfulness constraints

a. DEP (no epenthesis) (McCarthy and Prince
1995) Every segment in the output has a
correspondent in the base.
b. IDENT (no alternation) (McCarthy and Prince
1995) Corresponding segments have identical
value for the feature [F].
c. FAITHHEAD (no stress shift) (see
ANCHORPOS in McCarthy 2000 a) The output
syllable corresponding to the stressed syllable in
the base is stressed.
These constraints are responsible for structural identity, such

that there is no epen-thesis (DEP), no segmental alternation
(IDENT), and no stress shift (FAITHHEAD) in the imperative–
future relation. This is the highest possible degree of
conformity to Structural Identity (recall that there are still
distinctions in syllable structure), given that truncation must
apply (i.e. truncation is not blocked by unfaithful syllable
structure as in tit.for–tfor).
As will be shown in the ensuing section, Structural Identity

overpowers Existence. When an undominated markedness
constraint threatens DEP, the formation of the TI is blocked,
i.e. the faithful non-truncated candidate is selected.
3.3 GAPS IN THE PARADIGM

The strict faithfulness imposed on the future–TI relation by the
dominant faithfulness constraints cannot always be
maintained. When an undominated markedness constraint
does not allow strict faithfulness, a gap is forced on the
paradigm (s. 3.3.1). However, as noted in s. 3.1.2, a gap is not
allowed in the sub-paradigm, and therefore an alternative base
is selected (s. 3.3.2).
(p.54) 3.3.1 Inevitable gaps
Page 24 of 60
Imperative Paradigm
When the base begins with a CV syllable and the stem begins
with a sonorant consonant, the TI should have an initial
sonorant-consonant cluster in the onset (cf. (9)). Such a cluster
violates the Sonority Sequencing Generalization (SSG), which
is an undominated markedness constraint in the language.
Outside the paradigm of imperative truncation a potential

sonorant-consonant onset is simplified by an intervening e. For
example, when the plural suffix is added to nouns or
adjectives, a low vowel in a penultimate open syllable is often
deleted (e.g. gamal-im → gmalim ‘camels’, zaken-im → zkenim
‘old ms.pl.’). However, when the stem initial consonant is a
sonorant, e appears between the two initial consonants, to
rescue the SSG violation (e.g. makom-ot → mekomot ‘places’,
ratuv-im → retuvim ‘wet ms.pl.’).
The sonorant-consonant cluster in the onset of a TI cannot be

simplified by epen-thesis, due to the high ranking of DEP,
which secures Structural Identity. As shown in the examples
below, bases with stem initial sonorants do not have a TI (in
which case the future base is used for imperative); neither the
TIs with the sonorant-consonant onset nor the TIs with the
epenthetic vowel are surface true.
(16)
Page 25 of 60
Blocking truncation in bases with a stem initial sonorant

Masculine Feminine
Base: Future Output: TI Base: Future Output: TI
tirkod *rkod *rekod tirkedi *rkedi *rekedi ‘to dance’
tinšom *nšom *nešom tinšemi *nšemi *nešemi ‘to breathe’
timkor *mkor *mekor timkeri *mkeri *mekeri ‘to sell’
tilmad *lmad *lemad tilmedi *lmedi *lemedi ‘to study’
Page 26 of 60
Imperative Paradigm
(17)
Page 27 of 60
Future tirkod—no TI ‘to dance’ (SSG, DEPÂ¯ IMPTRUNC)

ti-rkod IMP ONSET *[σCCC MAXSEGS SSG DEP IMPTRUNC
a. tirkod *
b. <t>irkod *!
c. tfór *!
d. <ti>kod *
e. <tir>kod *! *!
f <ti>rekod *!
g terkod *!
Page 28 of 60
Imperative Paradigm
(p.55) The blocking effect, as illustrated in (17), is due to the

ranking of SSG and DEP above IMPTRUNC. DEP rules out the
candidates with the epenthetic vowel (f and g), and SSG rules
out the candidates with the initial sonorant-consonant cluster
(c and d). The other truncated candidates are eliminated by
the constraints discussed above.
The paradigm thus ends up with quite a few gaps, as not every
future base has a truncated imperative form (see more gaps in
s. 3.4.2). These gaps, however, are not accidental; they are
predicted to appear for every base beginning with a CVC
syllable, where the second C (the stem initial consonant) is a
sonorant. Bases beginning with a CV syllable truncate the V
and the stem initial sonorant is thus the second member of the
cluster (e.g. telamed, tlamed ‘to teach Future, TI’). However,
as will be shown in the ensuing section, there is one case
where such verbs do have a TI.
3.3.2 Impermissible gaps

In all the examples given above, the feminine and the
masculine bases of the same verb have an identical initial
syllable structure, either CVC (e.g. tif.tax, tif.te.xi ‘will open
ms.sg., fm.sg.’) or CV (e.g. te.ka.bel, te.kab.li ‘will receive
ms.sg., fm.sg.’). Therefore, both the feminine and the
masculine forms are equally affected by the constraint
rankings.
There are, however, verbs where the masculine base begins

with a CV syllable and the feminine with a CVC (e.g. te.šev,
teš.vi ‘will sit ms.sg., fm.sg.’). Such forms truncate the initial
CV (e.g. šev, švi ‘sit! ms.sg., fm.sg.’), due to *σ[CCC in the
feminine form (as in (8a)), and due to FAITHσ́ in the masculine
form (as in (13)).
When the stem of such verbs begins with a sonorant, only the
masculine TI can be derived (e.g. tered, red ‘to descend
Future, TI’); the feminine form would have an impermissible
sonorant-consonant cluster (e.g. terdi, *rdi ), and its formation
should thus be blocked (as in (15)). However, this is not the
case; as noted in s. 3.1.2, there are no gaps in the sub-
paradigm. As can be seen from the examples in (18), the
feminine TIs, when compared to the feminine future forms,
seem to have undergone enthesis (DEPviolation) and stress
shift (FAITHHEAD violation).
Page 29 of 60
Imperative Paradigm
(18)
Page 30 of 60
No strict faithfulness between the feminine future and the TI forms

Masculine Feminine
Input: Future Output: TI Input: Future Output: TI
Telex léx telxí *lxí léxi ‘to go’
Teréd réd terdí *rdí rédi ‘to descend’
Page 31 of 60
Imperative Paradigm
It appears that a verb may lack a TI, as shown in s. 3.3.1

above; but the moment it has one TI, all the TI cells in the sub-
paradigm must be filled (this includes the plural form, léxu and
rédu; see fn. 2). This is not a specific property of the
imperative para- (p.56) digm, but rather a general property of
the verbs system; all verb forms must be inflected for gender
(as well as number for imperatives, and also person for past
and future).
While Structural Identity has been shown in s. 3.3.1 to have

priority over Existence, it seems as if in (18) it is the other way
around. I argue, however, that this is not the case. In order to
conform to Structural Identity as well as Existence in the sub-
paradigm, the base of the feminine TIs in (18) is not the future
form but rather the masculine TI. That is, when Structural
Identity is at risk in the sub-paradigm, an alternative base is
selected, and thus Structural Relation is not respected. Thus,
regardless of the base of the TI, Structural Identity,
represented by DEP, IDENT, and FAITHHEAD, is respected.
Notice that the forms in the masculine–feminine TI pair in (18)

are structurally identical (with the exception of the syllabic
rearrangement imposed by the addition of the suffix), while
most masculine–feminine TI, as shown in (19), are not.
(19)
Page 32 of 60
Masculine Feminine Masculine Feminine
káx kxí *káxi ‘take!’ šév šví *šévi ‘sit!’
gáš gší *gáši ‘approach!’ tén tní *téni ‘give!’
ftáx ftexí *ftáxi ‘open!’ tfór tferí *tfóri ‘sew!’
Page 33 of 60
Imperative Paradigm
This suggests that the masculine TI serves as a base only for

the feminine TIs in (18); all other feminine TIs are derived
from the future forms, where the masculine and feminine
forms are usually not structurally identical (e.g. tešév, tišví
‘will sit ms., fm.’, tiftáx, tiftexí ‘will open ms., fm.’), and
therefore the masculine and feminine TIs are also not
structurally identical.
The derivation of the feminine from the masculine is not

foreign within the verb paradigm. In the future, for example,
the feminine forms are derived from the masculine, usually
without fully conforming to Structural Identity (e.g. V ~ Ø in
gadal, gadla ‘grew ms., fm’, V ~ e in tixtov, tixtevi ‘will write
ms., fm.’). There are a few past form masculine–feminine pairs,
which do conform to Structural Identity (e.g. kám, káma ‘got
up ms., fm.’).
The availability of the masculine form to rescue Structural

Identity in the forms in (18) suggests that all structural
relations in the paradigm are accessible, and can become
active when the appointed one (in this case the feminine
future) cannot be maintained. As I suggest in the discussion on
Tiberian Hebrew, access to the entire paradigm may appear as
a general case, and not only as a rescue device.
3.4 COLLOQUIAL HEBREW V. TIBERIAN HEBREW
Page 34 of 60
Imperative Paradigm
As shown in the discussion above, Structural Identity is fully

respected in the imperative paradigm of Colloquial Hebrew. In
the main paradigm, Existence is violated, (p.57) while
Structural Relation is respected, and in the sub-paradigm,
Structural Relation is violated, and Existence is respected (see
(5)).
The imperative paradigm of Tiberian Hebrew reflects different

priorities: it conforms to Existence in both the main and the
sub-paradigm, but Structural Identity is heavily violated. As a
complete analysis of Tiberian Hebrew imperatives is beyond
the scope of this paper, I confine myself to some descriptive
differences between the two (noncontiguous) stages of the
language, point out the relevant generalizations, and provide a
partial analysis.
As shown in s. 3.2, Colloquial Hebrew imperatives are derived

by CV or V truncation, depending on the initial syllable of the
future base (i.e. whether it is CV or CVC). Crucially, truncation
does not make reference to the morphological class of the verb
(binyan).7 The imperative paradigm of Tiberian Hebrew may
suggest that truncation distinguishes between the
morphological classes, since the imperatives of B2, B3, and B5
have a prefix (20a), while those of B1 and B4 do not (20b).
(20)
Page 35 of 60
Tiberian Hebrew imperatives

Masculine Feminine
Future Imperative Future Imperative
a. Imperatives
without a prefix
B1 ti-vrah bƐrah tivrƐh-ii birh-ii ‘to flee’
B4 tƐ-xabbeð kabbeð tƐxabbƐð-ii kabbƐð-ii ‘to respect’
b. Imperatives
with a prefix
B2 ti-kkaanes hi-kkaanes ti-kkans-ii hi-kkans-ii ‘to enter’
B3 t-albiš h-albeš t-albíiš-ii h-albíiš-ii ‘to dress Caus.’
B5 tit-qaddeš hit-qaddeš tit-qaddƐš-ii hit-qaddƐš-ii ‘to consecrate

Int.’
Page 36 of 60
Imperative Paradigm
I discuss in the next sections the non-prefixed imperatives (s.

3.4.1), the prefixed imperatives (s. 3.4.2), and the extent to
which the formation of the imperative is contingent upon other
categories in the paradigm (s. 3.4.3).
3.4.1 Imperatives without a prefix

Structural Identity in Tiberian Hebrew is heavily violated in
the imperative forms of B1. As the additional examples below
show, a stem initial fricative in the future corresponds to a
stop in the imperative (IDENT violation), and a consonant
cluster in the future is simplified by an epenthetic schwa when
it appears in word initial position in the imperative form (DEP
violation). In the feminine forms, there is also alternation in
the prosodic structure, due to Ø ~ i and ə~ Ø alternations
(DEP and MAXSEG (p.58) violation, where the latter is not
directly triggered by the morphology, as the truncated
segments are).
(21)
Page 37 of 60
Tiberian Hebrew imperative paradigm of B1

Masculine Feminine
ti-vrah bərah ti-vrəh-ii birh-ii ‘to flee’
ti-əmol gəmol ti-əməl-ii giml-ii ‘to compensate’
ti-ðroš dəəroš ti-ðrəš-ii ti- dirš-ii ‘to inquire’
ti-zkor zəxor zkər-ii zixr-ii ‘to remember’
Page 38 of 60
Imperative Paradigm
Following Benua (1997a), the phonological structure of both

the imperative and future forms is determined by markedness
constraints. The constraint triggering spirantization in the
future forms is *VSTOP, which prohibits a stop after a vowel.
In the imperative forms, the first stem consonant is initial in
the prosodic word, and thus not affected by this constraint.
*COMPLEX is responsible for the epenthetic schwa in the
masculine imperative forms, and the ranking *FRICATIVEÂ¯
*STOP for the stop in word initial position. The syllabic
structure of the feminine imperatives is determined by the
constraint * LIGHTLIGHT, which prohibits a sequence of two
light syllables (*bi. rəg.hii). As shown below, the dominance of
the markedness constraints causes the violation of Structural
Identity in the paradigm (the candidate t ivrah is not
considered here; it violates the undominated constraint
ONSET).
(22)
Page 39 of 60
Future tivrah—Imperative bərah ‘to flee’ (Tiberian Hebrew)

MARKEDNESS FAITHFULNES
S
tivrah IMP IMPTRUNC *COMPLEX *FRIC *STOP DEP IDENT
a. tivrah *! *
b. t(i)vrah *! *
c. (ti)vrah *! *
d. tibrah *! * *
e. tivərah *! *
f. ə tibərah * * *
Page 40 of 60
Imperative Paradigm
The ranking *ComplexÂ¯ Dep is responsible for epenthesis (i.e.

bərah ( f ) is better than *brah (d )), and the ranking
*FRICATIVEÂ¯ *STOPÂ¯ IDENT is responsible for the stop–
fricative alternation (i.e. bərah ( f ) is better than *vərah (e)).
The high ranking of IMPTRUNC, in particular above MAXSEG
(which is not mentioned here) forces truncation.
(p.59) The imperative forms are thus unfaithful, i.e. not

phonologically identical to their corresponding future bases
due to the ranking MARKEDNESSÂ¯ FAITHFULNESS. Faithful
relations, as in Colloquial Hebrew, are due to the ranking
FAITHFULNESSÂ¯ MARKEDNESS. The historical change can
be thus viewed as the demotion of markedness constraints
below faithfulness constraints (see Adam 2002 for the changes
affecting spirantization in Hebrew).
The prosodic structure of B4 verbs, which do not have a prefix

in the imperative form as well, is stable throughout the
inflectional paradigm, as the future prefix constitutes an
independent CV syllable and thus its truncation does not result
in a complex onset (təšammer, šammer ‘to preserve Future,
Imperative’). The only unfaithful relation between the future
and the imperative forms in B4 is in the stop–fricative
alternation (təxabbeð, kabbeð ‘to respect Future-Imperative’).
In the future form, the stem initial consonant is preceded by
the prefix vowel and thus subject to post vocalic spirantization,
due to *VSTOP. In the imperative form, this consonant is at the
left edge of the prosodic word, and thus surfaces as a stop,
due to the ranking *FRICATIVEÂ¯ *STOP.
Table (23) compares the imperative paradigms of Tiberian and

Colloquial Hebrew.
(23)
Page 41 of 60
Tiberian Hebrew (TH) vs. Colloquial Hebrew (CH) prefixless imperatives8

Masculine Feminine
B1 TH tivrah bərah tivrəh-ii birh-ii 虏to flee’
CH tivrax vrax tivrex-i vrex-i
B4 TH təxabbeð kabbeð təxabbəð-ii kabbəð-ii 虏to respect’
CH texabed txabed texabd-i txabd-i
Page 42 of 60
Imperative Paradigm
The data in (23) do not provide evidence as to whether the

truncated material in Tiberian Hebrew is morphological (the
future prefix) or phonological (CV). In Colloquial Hebrew, as
argued in the discussion above, the truncated material is
clearly phonological; B1 verbs undergo CV truncation (e.g.
tidroš, droš ‘to require Future, TI’), and B4 verbs undergo V
truncation (e.g. tekabel, tkabel ‘to receive Future, TI’).9 In s.
3.4.3 I argue that also in Tiberian Hebrew the truncated
material is phonological.
Another relevant difference between the two languages is in

the base of the feminine imperative. In Colloquial Hebrew the
base of the feminine imperative is, in most cases, the feminine
future, with the exception of a few feminine imperatives that
cannot be derived from the future and are thus derived from
the masculine imperative (p.60) (see s. 3.3.2). In Tiberian
Hebrew it seems that all feminine imperatives are derived
from the masculine imperatives.
Evidence is drawn from stems where the final consonant is a

fricative derived by spirantization. In such forms the feminine
imperative exhibits a post consonantal fricative (e.g. šixaii ‘lie
down! fm.’), rather than a stop. This is contrary to what is
expected given the high ranking of markedness constraints in
the imperative paradigm of Tiberian Hebrew, in particular the
ranking *FRICATIVE »*STOP. However, if we assume that the
feminine imperative is derived from the masculine (e.g. šəxaa,
šixaii ‘lie down! ms., fm.’), then the presence of a post-
consonantal fricative can be considered as base-output
faithfulness. It should be noted that the same phenomenon
appears in the masculine–feminine relation of the past (e.g.
šaaxav, šaxvaa ‘lay down ms., fm.’). That is, in the masculine–
feminine relations faithfulness is above markedness with
respect to spirantization.10
3.4.2 Imperatives with a Prefix

In the other verbal categories the imperative forms appear
with a prefix that does not appear in the future forms.
(24)
Page 43 of 60
Imperatives forms with a prefix (Tiberian Hebrew)

Masculine Feminine
B2 tikkaanes hikkaanes tikkans-ii hikkans-ii ‘to enter’
tiššaaver hiššaver tiššavr-ii hiššavr-ii ‘to be broken’
B311 taqdiš haqdeš taqdíiš-ii haqdíiš-ii ‘to consecrate

Trans.’
talbiš halbeš talbíiš-ii halbíiš-ii ‘to wear Caus.’
B5 titqaddeš hitqaddeš titqaddəš-ii hitqaddəš-ii ‘to consecrate Int.’
titlabbeš hitlabbeš titlabbəš-ii hitlabbəš-ii ‘to get dressed’
Page 44 of 60
Imperative Paradigm
As it will turn out (see (30)), the truncated material in the

forms in (24) is C. Following traditional studies (e.g. LaSor
1979), I view the h in the imperative as an epenthetic
consonant which serves to rescue the onsetless syllable.
The question is, however, why the formation of imperatives is

not unified throughout the derivational paradigm, i.e. why the
imperatives of B2, B3, and B5 have a prefix, (p.61) while
those in B1 and B4 do not. Assuming low ranked MAX and
DEP, we could expect C truncation and an epenthetic h for all
verb forms (e.g. B1 tivraħ, *hivraħ; B2 tikkaanes, hikkaanes ;
B3 taqdiš, haqdeš ; B4 təxabbeð, *həxabbeð, B5 titqaddeš,
hitqaddeš ). Alternatively, assuming a high ranked ALIGNL
(Stem, PrWd), we could expect CV truncation and
degemination or internal vowel epenthesis (e.g. B1 tivraħ,
bəraħ; B2 tikkaanes, *kaanes, B3 taqdiš ; *qədeš; B4 təxabbeð,
kabbeð; B5 titqaddeš, *təqaddeš ).12 I propose an answer to
this question in s. 3.4.3.
Observe the differences between Colloquial and Tiberian

Hebrew in (25).
(25)
Page 45 of 60
Tiberian Hebrew (TH) vs. Colloquial Hebrew (CH) prefixed imperatives

Masculine Feminine
B2 TH tikkaanes hikkaanes tikkans-ii hikkans-ii ‘to enter’
CH tikanes tkanes tikans-i tkans-i
B3 TH talbiš halbeš talbíiš-ii halbíiš-ii ‘to dress Caus.’
CH talbiš — talbí š-i —
B5 TH titqaddeš hitqaddeš titqaddəš-ii hitqaddə#x016 ‘to consecrate

1;-ii Int.’
CH titkadeš tkadeš titkad š-i tkad š-i
Page 46 of 60
Imperative Paradigm
In Colloquial Hebrew, B2 and B5 verbs undergo V truncation

(e.g. B2 tikanes, tkanes; B5 titlabeš, tlabeš); due to the ranking
DEP » *COMPLEX, there is not epenthesis. As for B3 verbs, it
is argued in s. 3.4.3 that the vowel in the first syllable of the
future base is a stem vowel (rather than a prefix vowel), and
thus cannot be truncated due to the high ranking of
MAXSEGS. Consonant epenthesis, which in Tiberian Hebrew
rescues the onsetless syllable, cannot take place in Colloquial
Hebrew due to the high ranking of DEP. Truncation is thus
blocked, as in the forms discussed in s. 3.3.1 above, and the
optimal candidate is the one identical to the base. That is, B3
verbs do not have a truncated imperative form in Colloquial
Hebrew.
(26)
Page 47 of 60
B3: Future tazkir—no TI ‘to remind’ (Colloquial Hebrew)

t-azkir IMP ONSET MAXSEGS DEP IMPTRUNC MAXSEG
a. ə tazkir *
b. tazkir *! *
c. Tazkir *! (a) *
d. tazkir *! (a) **
e. thazkir *! (h) *
Page 48 of 60
Imperative Paradigm
(p.62) Assuming that the imperatives are derived by

truncation in both Tiberian and Colloquial Hebrew, the
differences are due to the high ranking of DEP in Colloquial
Hebrew, as opposed to its low ranking in Tiberian Hebrew. The
low ranking of faithfulness constraints, in particular below
markedness constraints, allows the non-uniform paradigm on
the horizontal axis. As for the non-uniformity of Tiberian
Hebrew paradigm on the vertical axis (i.e. that some binyanim
have a prefix and others do not), I argue below that it is due to
a faithfulness constraint, which refers to the past forms.
3.4.3 Multi-relational paradigms

Notice in (24) that in the imperatives of B2 and B5 in Tiberian
Hebrew, the h is followed by an epenthetic i, while in B3 it is
followed by an a. The vowel following the prefix in B3 verbs
such as tazkir is a stem vowel and therefore cannot undergo
truncation. As argued in Bat-El (2002) for Colloquial Hebrew,
but holding for Tiberian Hebrew as well, this vowel is
identified as a stem vowel on the basis of reference to the past
form. If the vowel following the future prefix consonant has a
correspondent in the past form, then it is a stem vowel;
otherwise, it is part of the prefix. All the past forms of B3
begin with hi (in a few forms, ho or he), as in hizkir, and the
vowel following the h corresponds to the vowel following the
future t in tazkir. That is, in this case, reference to the past
form is crucial for the imperative paradigm, which seemed
until now to consist only of the future and the imperative
categories. This approach is not new; similar correspondence
relations between one output and two (or more) bases have
been proposed in Burzio (1998) and Steriade (1999a).
In Colloquial Hebrew, reference to the past form is activated

only when the future form consists of two syllables, because
the expected number of syllables in the future form is three.
This is true for all B3 verbs, as well as B1 verbs (see Bat-El
2002 for additional examples from the class of irregular
verbs). The past forms of B1 verbs do not have segmental
material preceding the first consonant of the stem, and
therefore the vowel following the future prefix t is not
considered a stem vowel. In B3 past forms, the first consonant
of the stem is preceded by a vowel in the past form, and
therefore the vowel preceding the first consonant of the stem
Page 49 of 60
Imperative Paradigm
in the future form is considered a stem vowel (the t in the

future form is easily identified as a future prefix).
(27)
The status of the vowel preceding the first stem

consonant
Past B1: gadal B3: higdil
Future tigdal yagdil
Future vowel has a no yes

correspondent in the pastə
A stem vowelə no yes
While in Colloquial Hebrew reference to the past form needs

to be activated (by the number of syllables in the future form),
it seems that in Tiberian Hebrew it is (p.63) constant; that is,
the formation of every imperative form has to refer to the past
form. The relevant observation is that an imperative form is
accompanied by a prefix whenever there is a prefix in the past
form (B2, B3, and B5); verbs which do not have a prefix in the
past form (B1 and B4), do not have a prefix in the imperative.
This generalization can be accounted for with the faithfulness
constraint ANCHORL (McCarthy and Prince 1995), which
requires here correspondence between the stem segments at
the left edge of the past and the imperative.
(28) ANCHOR LEFT (ANCHORL)

A stem segment at the left edge of the past form
corresponds to a stem segment at the left edge of the
imperative form.
Since ANCHORL refers to the past form, its presence in the

constraint ranking of imperative truncation of Tiberian
Hebrew requires a constant reference to the past forms.
ANCHORL blocks epenthesis in B1 and B4 verbs (cand-b in
(29)), since the past forms of these verbs do not have a prefix
and thus begin with a stem consonant.
(29) Past forms without a prefix (Tiberian Hebrew)

a.
Page 50 of 60
a. tiγdal, gəaðal ‘to grow Future, Imperative B1’

Fut. tiədal IMP Past: IMPTRUNC ANCHORL MAXSEG DEP
gaaðal
a. tiədal *! *
b. <t>hiədal *! * *(h)
c. ə <tig>gəðal ** *(ə)
Page 51 of 60
Imperative Paradigm
b.
Page 52 of 60
b. ddel, gaddel ‘to raise Future, Imperative B4’

Fut.: təəaddel IMP Past: IMPTRUNC ANCHORL MAXSEG DEP
giddel
a. təəaddel *! *
b. <t>həəaddel *! * *(h)
c. <t>gaddel **
Page 53 of 60
Imperative Paradigm
Notice in (29a) that ANCHORL selects the candidate with the

internal epenthesis (c), rather than the one with the external
epenthesis (b), and thus preserves the stem segment at the left
edge.
In the prefixed verbs, the candidate with the external

epenthesis is selected. When the past form is accompanied by
a prefix, ANCHORL is vacuously respected since there is no
stem segment at the left edge of the past form. In this case,
the lower ranked constraints select the optimal candidate.
The different results of truncation in the prefixed and

prefixless verbs in Tiberian Hebrew are thus attributed to
ANCHORL, which requires reference to the past forms.
(p.64)
(30) Past form with a prefix (Tiberian Hebrew)

a.
Page 54 of 60
a. titqaddes, hitqaddeš ‘to consecrate Future, Imperative B5’

Fut.: titqaddeš MP Past: IMPTRUNC ANCHORL MAXSEG DEP
hitqaddeš
a. titqaddeš *!
b. <ti>təqaddeš ** *!
c. ə <t>hitqaddeš * *
Page 55 of 60
Imperative Paradigm
b.
Page 56 of 60
b. tikkaanes, hikkaanes ‘to enter Future, Imperative B2’13

Fut.: tikkaanes IMP IMPTRUNC ANCHORL MAXSEG DEP
Past: nixnas
a. tikkaanes *!
b. <ti-k>kkaanes ***!
c. ə <t>thikkaanes * *
Page 57 of 60
Imperative Paradigm
3.5 CONCLUSION
Paradigms can be non-uniform in different ways. In this
chapter, we have seen differences within languages and across
two stages of a language.
The difference between Colloquial and Tiberian Hebrew is in

the ranking of markedness and faithfulness constraints. In
Colloquial Hebrew, faithfulness constraints are ranked above
markedness constraints, and the imperative is thus identical to
the future base (with the exclusion of the truncated material).
That is, Structural Identity is respected. In Tiberian Hebrew,
markedness constraints are ranked above faithfulness
constraints, and the imperative is thus segmentally (stop v.
fricative) and pro-sodically (epenthesis and deletion) different
from the future base. That is, Structural Identity is violated.
Another effect of the dominance of faithfulness constraints in

Colloquial Hebrew is the violation of Existence. When the
imperative cannot be identical to its base, there is no
imperative and Existence is violated. Such an effect is not
found in Tiberian Hebrew since faithfulness constraints are
low ranked.
The way and the degree to which a paradigm is (non-)uniform

is thus a matter of the interaction between markedness and
faithfulness constraints.
Notes:
(1) A paradigm is usually named after the grammatical
category specifying all members in the paradigm (e.g. tense
paradigm, gender paradigm, etc.). Here I am concerned with
two categories only, future and imperative, and thus refer to
the paradigm using the derived form, the imperative.
(2) The third category in the sub-paradigm is the plural (i.e.

the sub-paradigm is actually gender-number). I ignore this
category here since the plural forms are identical to the
feminine ones, where the distinction is only in the quality of
suffix, -i for feminine and -u for plural.
(3) The principle of structural identity is not very different

from the principle of Uniform Exponence, proposed in
Kenstowicz (1996). The latter requires minimal differences in
the realization of lexical items.
Page 58 of 60
Imperative Paradigm
(4) This is a type of non-templatic truncation, similar to what

appears, for example, in Danish imperatives (Anderson 1975)
and Koasati plurals ( J. Martin 1988); see Kurisu (2001) for a
detailed list of references and languages.
(5) The opacity is due to the presence of non-alternating stops

(e.g. kibel, mekabel, yekabel ‘to receive Past, Participle,
Future’), in addition to the non-alternating fricative and the
alternating stop–fricative. See detailed discussion in Adam
(2002).
(6) Notice the distinction between the term ‘base’, which

refers to the future surface form, and the term ‘stem’, which
refers to the future form without the prefix (the suffixes are
not relevant here, as truncation targets the left edge).
(7) Hebrew verbs are classified into classes, called ‘binyanim’,

identified on the basis of the vocalic pattern, the prosodic
structure, and in some classes also the prefix. I refer to these
classes as B1, B2, etc.
(8) The length distinction found in Tiberian Hebrew has been

lost in Colloquial Hebrew. There are also various segmental
changes, where the ones appearing in the data presented here
are, ə > e or Ø, Ä§ > x, q > k, γ > g, and ð > d. There is no
evidence whether the labial fricatives in Tiberian Hebrew
were bilabial or labiodental, and I thus use the less marked
fricative, i.e. the labiodental.
(9) See, however, Bat-El (2002) for variation in B4 verbs,

where one of the variants undergoes CV truncation (e.g.
tekabel, tkabel ∼ kabel ).
(10) Since the stem final consonant is a fricative throughout

the entire tense paradigm, it is possible to consider this issue
within the global paradigmatic approach proposed in
McCarthy (Ch. 8).
Page 59 of 60
Imperative Paradigm
(11) There is also an i ∼ e alternation in the final syllable of B3

verbs. As noted by Ora Schwarzwald (personal
communication), this alternation is rooted in an earlier stage
of the language, where the future form used to end in a vowel,
while the imperative form always ended in a closed syllable
(notice that the alternation does not appear in the feminine
forms, where there is a suffix in both the future and
imperative). This alternation may provide evidence against a
truncation analysis, but I will not pursue this issue here. I
assume, for comparative purposes, that in both Tiberian and
Colloquial Hebrew the imperatives are derived by truncation.
(12) Degemination appears in ti-tten, ten ‘to give Future,

Imperative’ (where ti-tten is derived from /ti-nt-en/ by
assimilation; cf. the past form naaTan). This form is given in
Benua (1997a), following Prince (1975), in support of the
truncation approach.
(13) I consider degemination in (ti-k kaanes as a MAXSEG

violation, though it is actually a MAX MORA violation. Internal
epenthesis (ti-k kaanes) is ruled out by GEMINATE INTEGRITY
(Hayes 1986).
Page 60 of 60
Sources of Paradigm Uniformity


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001

Luigi Burzio
DOI:10.1093/acprof:oso/9780199267712.003.0004
Page 1 of 80
This chapter argues that paradigm uniformity (PU) effects

ultimately show that, while the move towards parallelism
initiated by OT was much on the right track, the degree of
parallelism in the system is in fact more extensive, concerning
not only the internal structure of phonology, but also the
relationship between phonology and morphology. It further
argues that the mental calculations or ‘processing’ required by
phonology and morphology are not only parallel, but also
‘distributed’, indeed as in ‘Parallel Distributed Processing’ —
the formal term for connectionism. The chapter proceeds as
follows. Sections 4.2 and 4.3 argue that representations
influence or ‘attract’ each other to a degree that is inversely
related to their geometrical or global distance. Section 4.4
argues that the role of global distance reveals that mental
representations have fundamental properties of neural nets,
and also that a major factor behind the Optimality Theory (OT)
notion of OO-F is precisely the noted attraction effect. Section
4.5 elaborates further on the proposal of Section 4.4, pointing
to additional consequences of significance. Section 4.6
suggests that a coherent and improved approach to
morphology can also be developed from this perspective, one
that further defines the OO-F relations at work within the
phonology in terms of morphological relations. Section 4.7
reviews the typology of PU effects, showing that it is captured
by the present approach though not by traditional means.
Keywords: paradigm uniformity, parallelism, Optimality Theory, global

distance, geometrical distance, OO-F
4.1 INTRODUCTION
Generative Phonology was only in its teens, when J. W. Harris
(1973) argued on the basis of facts like those in (1) that
Paradigm Uniformity (henceforth ‘PU’) must be recognized as
a principle of grammar.
(1)
love-Imperf. Indic. Latin Spanish
ISG amá:bo amába
2SG amá:bas amábas
3SG amá:bat amába
IPL ama:bá:mus amábamos
Page 2 of 80
love-Imperf. Indic. Latin Spanish
2PL ama:bá:tis amábais
3PL amá:bant Amában
Since stress in the Romance languages is quite generally

faithful to the original stress of Latin, there seems no reason
for the shifts in the first and second plural in (1) unless
‘paradigms’ are under pressure to be uniform. The theoretical
machinery of the day had no way of incorporating PU,
however, and Harris’s proposal was quickly forgotten. Years
later, Roca (1992) will argue that Spanish simply has two
different stress (p.66) systems, one for verbs and one for
nouns. The former is essentially ‘lexical’, associating stress
with a specific morpheme, in (1) the thematic vowel -a-, while
the latter calculates stress from the right edge of individual
words, though not always in the same way: señór ‘gentleman’,
pistóla ‘pistol’, fábrica ‘factory’, etc. While theoretically
troublesome, Harris’s solution would have been superior to
Roca’s, since it required no stipulation: PU obtains only with
verbs because in Spanish only verbs form this kind of
inflectional paradigm.1
In Burzio (1994a) (henceforth PES) and earlier work leading

up to it (1991 et seq.) I proposed an account of the English
stress-transfer effects in (2a′, b′) based on the three blocks of
violable constraints of (2.I, II, III).
(2)
I. MWF II. MC III. MA

(Metrical (Metrical (Metrical
Well- Consistency) Alignment)
Formedness)
a. √ n/a √
an(tágonis)tø
a.′ a(mérica) √ √ *
(nìstø)
b.(ònoma) √ n/a √
(tólogy)
Page 3 of 80
I. MWF II. MC III. MA

(Metrical (Metrical (Metrical
Well- Consistency) Alignment)
Formedness)
b.′. √ √ *
phe(nòme)
(nólogy)
Example (a) shows the preferred right-edge Metrical

Alignment (III) for nouns: a foot boundary coincides with the
end of the last syllable with an overt nucleus, the final
consonant t forming a degenerate syllable with a null nucleus.
Example (a′) shows that this requirement can be violated when
so compelled by ‘Metrical Consistency’ – the competing
requirement that morphologically complex forms stress
consistently with their bases, the base here being américan.
Example (b) shows the preferred Metrical Alignment at the left
edge of a word, with a foot boundary coinciding with the left
edge of the first syllable. This requirement is violated in (b′)
again as compelled by Metrical Consistency, this time to
maintain the stress of phenómenon. In (2), top-ranked MWF
defines the range of possible feet, thus accounting for the
absence of Metrical Consistency effects under foot
degeneracy, as in *ca(tàs)tróphic (cf. catástrophe);
*(cómpensa) tòry (cf. cómpensàte). In the PES analysis these
cases are excluded by the ill-formedness of unary feet and of
ternary feet with a heavy median syllable, respectively.
The PES approach, which independently introduced basic

elements of Prince and Smolensky’s (1993) ‘OT’, directly
accounts for the facts in (1). Latin primary stress reflected a
strictly deterministic system: a penultimate syllable was
always stressed if heavy, and never if light (except in
disyllables). In terms of (2), MWF left no window of
opportunity for MC, whence the lack of PU in Latin. In
contrast, Spanish MWF does have certain degrees of freedom:
a light penultimate can either be stressed—pistóla—or not—
fábrica. With top-ranked MWF equally satisfied, MC will (p.
67) be correctly expected to make the choice, hence
accounting for the uniform Spanish stress in (1)—the illusion
of ‘lexical’ stress. In the Optimality Theoretic framework as
further developed by McCarthy and Prince (1995, 1999), the
‘Metrical Consistency’ of PES and PU more generally can be
Page 4 of 80
formally treated as a specific instance of ‘faithfulness’—the

same type of relation that exists between an output and its
input, or between a reduplicant and its base, as argued in
Benua (1997a). The introduction of such Output–Output
Faithfulness/Consistency into OT raises important questions,
however, notably the ones in (3).
(3)
a. Exactly which other output is the source of
Output–Output Faithfulness for any form under
calculation?
b. How may the existence of Output–Output
Faithfulness (OO-F) constraints affect our view
of morphology? Specifically, what must we make
of the fact that the type of surface-to-surface
resemblances established by traditional
morphological relations are insufficient,
requiring the addition of OO-F?
c. May ‘cyclic’ derivations, once employed to
capture some transfer of phonological properties
(such as the one in (2b′)) still be viable within OT
after all as an alternative to OO-F, should
questions (3a, b) prove troublesome?
In this chapter I address the critical issues in (3). I will argue,

as I also do in Burzio (2002b), that PU effects ultimately show
that, while the move towards parallelism initiated by OT was
much on the right track, the degree of parallelism in the
system is in fact more extensive, concerning not only the
internal structure of phonology, but also the relationship
between phonology and morphology. I will claim further that
the mental calculations or ‘processing’ required by phonology
and morphology are not only parallel, but also ‘distributed’,
indeed as in ‘Parallel Distributed Processing’—the formal term
for connectionism. The specific organization of the chapter is
that in ss. 4.2 and 4.3 I argue that representations influence or
‘attract’ each other to a degree that is inversely related to
their geometrical or global distance. In s. 4.4, I argue that the
role of global distance reveals that mental representations
have fundamental properties of neural nets, and also that a
major factor behind the OT notion of OO-F is precisely the
noted attraction effect. Section 4.5 elaborates further on the
proposal of s. 4.4, pointing to additional consequences of
significance. In s. 4.6 I suggest that a coherent and improved
approach to morphology can also be developed from this
Page 5 of 80
perspective, one that further defines the OO-F relations at

work within the phonology in terms of morphological relations.
In s. 4.7 I review the typology of PU effects, showing that it is
captured by the present approach though not by traditional
means. In s. 4.8 I conclude.
Page 6 of 80
4.2 THE ROLE OF DISTANCE

PES (276 and passim) noted certain clustering effects. For
instance, items like larynx, pharynx, syrinx have two attested
plurals: a regular one like lárynxes, and an irregular (p.68)
one like larýnges. The irregular plural has regular stress, on a
heavy penultimate, while the ‘regular’ one has the stress of
the singular. What one observes here is thus co-variation of
two types of identity: segmental identity and identity in stress.
The regular plural is identical to the singular both ways, while
the irregular one is different both ways. Anderson (1992:193f.)
had noted similar clustering effects with -able adjectives. For
instance, cómparable, which is different in stress from its base
compáre, is also divergent from it semantically, while
compárable, which retains the stress of the verb, is also
strictly faithful to the verb semantically. Anderson suggested
that -able may be ambiguous between being either a Level 1 or
a Level 2 affix within Lexical Phonology (Kiparsky 1982a, b),
given that both re-stressing and semantic divergence are
known properties of Level 1 formations while being both
absent at Level 2. While this view might be extended to the
previous case, (Level 1 v. Level 2 plural), it does not really
solve the clustering problem, but merely subsumes it under a
fairly general stipulation of Lexical Phonology, since—as I note
below—the latter framework is not particularly successful in
accounting for this clustering of properties across the
postulated levels. In addition, this account would not extend to
other clustering cases such as the one in (4)
(4)
Base Derivative Faith (stress) Faith (V-length)
a.rémedyremé:di- * *
able
b.lévy lévi-able √ √
In (4a) the derivative is unfaithful to its base with respect to

both stress and vowel length, while in (4b) faithfulness obtains
both ways. Here, it would not do to appeal to a Level 1/Level 2
distinction, since the difference between re-stressed
remédiable and non re-stressed léviable is fully predictable
from Level 1 properties alone. In the PES analysis, the
metrical structure *(ré.me.di.a), with extrametricality of the
final ‘weak’ syllable and pre-antepenultimate stress is not well
Page 7 of 80
formed, barred by the MWF of (2) which excludes

quadrisyllabic feet, while that of (lé.vi.a) is well formed.
Sameness of stress is then evidently able to induce sameness
of V-length in (4b), blocking the otherwise regular
phenomenon of CiV-lengthening of (4a) and many other cases,
like Caná:dian, Mongó:lia, etc. The Lexical Phonology machine
provides no clue for this clustering. On that approach, CiV-
lengthening would be expected to apply to both (4a, b) equally,
whether or not it is a cyclic process. Hence, while inheritance
of properties can sometimes be handled by ‘cyclicity’, the
inheritance of the short vowel in leviable cannot be.
The cases illustrated in (5), from Steriade (1997a, 1999a), are

also beyond the reach of any traditional means.
In French, when occurring in certain prevocalic contexts,

masculine adjectives ending in a vowel resolve the potential
vowel hiatus by taking on a form similar to that of the feminine
allomorph. In the forms in (5), dialect (a) borrows just the final
consonant [n] from the feminine form, while dialect (b)
borrows both the final consonant (p.69) and the non-
nasalized preceding vowel of the feminine: [εn]. We can see
this variation as the result of two competing clustering effects,
expressed by the enclosed pairs of cells in (5). On the one
hand, the liaison form is masculine, in agreement with the
head noun, and is thus faithful to the masculine citation form
(Base 1) in this respect. On the other hand, by borrowing the
consonant from the feminine allomorph (Base 2), the
phonology places the liaison form in a partial faithfulness
relation with the latter as well. The dialectal variation in (5)
would show that V-quality may cluster either with the final C,
yielding [ε] in conformance with the feminine allomorph (case
(b)), or with gender, yielding [ε̃] in conformance with the
masculine one (case (a)). The ‘clustering’ interpretation is
confirmed by the fact that, where FAITH-2 (C#) is breached,
as in [gʁoz aʁbʁ] gros arbre ‘big-M tree’ compared with
feminine [gʁɔs] grosse ‘big-F’ ([z] v. [s]), V-quality is no longer
influenced by the feminine allomorph, *[gʁɔz aʁbʁ], but only
by the masculine [gʁo] gros ‘big-M’. A special restriction
imposes [z] rather than [s] in such liaison contexts, as
discussed by Steriade.
Page 8 of 80
(5) French Liaison (Steriade 1997a, 1999): prochain

arrêt ‘next-M stop’ Dialect (a): [pʁɔ∫ɛ̃n aʁe] Dialect (b):
[pʁɔ∫εn aʁe]
Page 9 of 80
Base 1(MASC.) Base 2(FEM.) Liaison form/ Faith-1(M) Faith-2(F)

dialect
gender V-quality C# V-quality
[pʁɔ∫ε̃]prochain‘ [pʁɔ∫εn]prochai a. pʁɔ∫εn √ √ √ *

next-M’ ne‘next-F’
b. pʁɔ∫εn √ * √ √
Page 10 of 80
Clustering of faithfulness effects characterizes as well one of

the syndromes referred to as ‘Non Derived Environment
Blocking’ (NDEB), exemplified in (6).
(6) Sanskrit ruki rule (Kiparsky 1982a)

a. No retroflection: sisa ‘lotus’ (non-derived)
b. Retroflection: sis̩-t̩a ‘taught’ (base: sās
‘instruct’)
The assimilatory process that turns [s] to retroflex [s̩] after r,

velars or high vowels (r, u, k, i) applies morpheme internally
only if some other change has applied, as in (6b), where the [ā]
of the base has ablauted to [i], but not as in (6a), where no
other change has occurred. Hence faithfulness for
retroflection of the [s] ([±high] according to Kiparsky) appears
to cluster with faithfulness to other aspects of the
representation of the morpheme.
What emerges from these cases and others (reviewed in

Burzio 2002b) is the descriptive generalization in (7).
(7) Allomorphs attract each other in a way that is

inversely related to their global distance.
(p.70) This generalization describes the noted clustering

effects: when faithfulness is satisfied on some dimension
(shorter global distance), it tends to be satisfied on other
dimensions as well (stronger attraction). The notion of
‘allomorph’ in (7) seems in fact superfluous. Since allomorphs
are by definition structures that are relatively similar in sound
and meaning, the definitional similarity of allomorphs may
simply be subsumed under the notion of global distance of (7),
and the specific term ‘allomorphs’ may be replaced by the
more general ‘representations’.
Summing up, we find that the faithfulness constraints that are

at work in allomorphy defy unique ranking. Their rank appears
to be rather modulated by distance, such that the closer the
target representation is to the trigger of the faithfulness, the
higher is the rank. The above discussion has also shown
indirectly that the faithfulness that controls allomorphy cannot
in general be faithfulness to some ‘Underlying
Representation’ (UR). The reason is that such faithfulness is
sensitive to surface properties that would not be present in
UR. For instance, we have seen that lengthening in
remé:diable is crucially aided by the fact that its stress is
Page 11 of 80
different than that of rémedy, yet the latter stress cannot be in

UR since it is perfectly regular and predictable. Similarly, in
the French case in (5), the liaison allomorph is not in a
faithfulness relation to some UR. For one thing, a unique UR
for both masculine and feminine allomorphs seems unlikely,
the [ε̃] (MASC.)/[εn] (FEM.) alternation being more plausibly
analyzed as just part of the morphology (suppletion) rather
than phonologically induced. Even if there were such a
common UR, say /pʁɔ∫εn/, the dialect (a) liaison form would
not be derivable from it, since vowel nasalization would have
to be contingent on deletion of the nasal to correctly derive
masculine and feminine allomorphs, but this relation would
automatically exclude [pʁɔ∫̃n], where both the nasal and
nasalization coexist. Rather, the faithfulness relation in
question appears to hold on related outputs, implicating no
UR. Finally, we find that, while ‘cyclicity’ may capture some
surface-to-surface faithfulness effects (since each cycle
produces a surface form), it is challenged by the cases at
hand. No form of cyclicity can relate liaison forms to
masculine and feminine allomorphs simultaneously as just
noted, and none can distinguish remé:diable from léviable as
also noted. The conclusion is thus that it is the extent to which
surface representations themselves intersect (global distance)
that is relevant to allomorphy, not the extent to which their
derivational histories do.
4.3 DISTANCE IN PHONOLOGY AND PHONETICS

While the previous section has illustrated the role of distance
in allomorphy, that is distance between morpheme-sized units,
it is well known that distance plays a central role with
segment-sized units as well. This is the essence of the
‘Dispersion Theory’ of Lindblom (1986) and Flemming (1995)
(see also Padgett 1997), embodying the claim that segmental
inventories are shaped by maximal perceptual distance
between their members. That claim is expressible as the same
notion of attraction over distance (p.71) of (7) that was
applicable to allomorphs. Reduced distance between segments
results in greater attraction, forcing weakly contrasting
segments to neutralize, thus bringing about greater distance
between the remaining segments. Steriade (1994, 1997b) has
argued that not only segmental inventories, but segmental
neutralizations also attest to the role of perceptual distance.
For instance, coda devoicing in Dutch ([bεt] ‘bed’, versus
[bεdən] ‘beds’) results in her view from the fact that the
Page 12 of 80
perceptual cues for voicing in the positions that we generally

refer to as ‘codas’ are weaker, thus heightening the similarity/
reducing the distance between members of the voiced/
unvoiced contrasts. In this conception, segmental
neutralizations are the expected inventory reductions in
contexts that effectively shrink the perceptual space.
Wilson (2000) has shown further that a proper OT

characterization of neutralization effects resolves formal
problems that had long remained recalcitrant within OT,
including the ‘opacity’ (counterfeeding/counterbleeding)
effects. While the reader is referred to Wilson (2000, 2001) for
detailed discussion, and Burzio (2000b, 2001) for a synopsis, a
further succinct characterization can be provided for present
purposes by comparison with McCarthy’s relatively well-
known ‘Sympathy’ Theory (McCarthy 1999; Kager 1999b: s.
9.2). McCarthy shows that opacity effects can be produced
within OT by requiring an output to be faithful not only to its
input, but simultaneously also to another specially designated
output candidate, referred to as the ‘sympathetic’ or ‘flower’
candidate. The essence of Wilson’s contribution is in showing
that the needed further source of faithfulness or ‘attractor’ is
independently provided by the theory of segmental
neutralizations. On Wilson’s solution, opacity effects result
from the fact that, while a word or morpheme-sized
representation is being faithful to its own input (which, on the
present approach, is just another output), some segment-sized
substructure within it is simultaneously also in an attraction
relation with a competing representation of that substructure.
In the Dutch example, the degenerate [d] of output candidate
[bεd] is being attracted to the independent well-formed
representation [t], whence [bεt].
Assimilatory phenomena also seem to be driven by distance.

Thus, it seems rather generally the case that segments that
assimilate are independently similar. For example, in Spanish,
voiced stops spirantize post-vocalically, but unvoiced ones do
not: Cu[ß]a ‘Cuba’/co[p]a ‘wine glass’. One can take this to
indicate that continuancy is shared with a preceding vowel
only if voicing is already shared. Similarly, the lack of nasal
place assimilation in standard English i[n]famous can be
attributed to clustering of place and continuancy. If this
clustering obtains, then assimilation will be blocked by the
mark-edness of the resulting nasal (labiodental) fricative *[m]
(although fast speech does tolerate the marked pronunciation,
Page 13 of 80
[imfəməs]). Similarly, in Sudanese Arabic a stop assimilates in

continuancy to a following obstruent: kitaa[b] ‘book’ /
kitaa[f]Fáth̩i ‘Fáth̩i’s book’, but only if trigger and target
already share place: kitaa[p] Samíir (not *kitaa[f])‘Samíirs
book’ (Kenstowicz 1994: 54). Vowel harmony phenomena are
also often ‘parasitic’ on some degree of independent identity.
For instance rounding harmony in Yawelmani depends on
identity in [±high] (Kenstowicz 1994: 108).
It is well known that alongside of assimilatory effects,

dissimilatory (‘OCP’) effects (p.72) also exist. They are also
modulated by distance. For instance in Yucatec Maya (Fuka-
zawa 1998), only sequences of obstruents that are homorganic
in both place and manner (e.g. tt) undergo dissimilatory
repairs. Sequences sharing only one of those features (?t, kt,
st) undergo no repair.
The existence of both assimilation and dissimilation seems

rather mysterious on a standard approach to markedness: if it
is ‘marked’ to agree in features (OCP), how could it also be
marked to disagree (assimilations)? The notion of attraction
sheds light on this mystery in so far as it entails two different
regions of relative stability: identity, where attraction is
satisfied, and greater distance, where attraction is reduced
(see also discussion below). The locality or proximity
conditions on both assimilatory and dissimilatory processes
can themselves be understood in terms of attraction-inducing
similarity by regarding sequential proximity itself as a form of
similarity, as I discuss further below.
Hence attraction between segments plays a central role in

several areas of phonology as currently understood and its
links to perception. It seems also central to the notion of
perception more generally. Our ability to map degenerate
auditory inputs into a set of mental canons can be viewed in
terms of attraction. If attraction over distance is a property of
mental representations in general, then, in order to be
interpreted, a representation formed on a perceptual input
need not match a mental category exactly, but need be merely
within its range of attraction. Mental categories can thus be
viewed as ‘perceptual magnets’, indeed as independently
proposed by Kuhl and Iverson (1995).
Page 14 of 80
Repp (1982) describes as ‘phonetic trading relations’ the fact

that the weakening of one perceptual cue can be offset by the
strengthening of another. If global distance is what matters,
then indeed distances over individual dimensions will be in
trade-off relations with one another. Note that trading
relations are implicit in the clustering effects observed above
in so far as we found no absolute restriction on what
dimensions could cluster together (segmental structure,
prosodic structure, semantics). The ability of some identity
effect IA to cluster together equivalently either with some
other identity IB or yet some other identity IC: IA & IB ≈ IA & IC
constitutes a type of trading relation in the domain of
allomorphy comparable to that described by Repp for
perception. The effect identified in Ganong (1980) reveals
further that the same perceptual mechanisms span over
phoneme and word-size units. The latter effect consists of
(8)
Page 15 of 80
Base 1 (MASC.) Base 2 (FEM.) Liaison form/ Faith-1(M) Faith-2(F)

dialect
gender V-quality C# V-quality
[nuvo] nouveau [nuvεl] a. *nuvol √ ? √ *

‘new-M’ nouvelle‘new-F
b. nuvεl √ * √ √
Page 16 of 80
(p.73) the fact that a phoneme identification task is

facilitated if the stimulus that it is part of is a lexical item. This
is consistent with the present view that attraction over
distance is a property of mental representations of any size.
The Ganong effect can be characterized by saying that a
phoneme-size perceptual unit is a candidate for attraction by a
phoneme-sized mental category while at the same time the
word-sized perceptual unit it is part of may also be attracted
by a member of the word lexicon, and that the two attractions
are cumulative. In fact, a similar conclusion would seem
warranted by the French liaison case in (8) to be compared
with that of (5).
The data in (8) are laid out identically to those in (5), but here
there is no dialectal variation, as the form that takes the final
C from the feminine and the preceding vowel from the
masculine:*nuvol, is not found. To make sense of this
divergence, we must revise the above discussion slightly. We
must assume that there are two simultaneous attraction
effects from the masculine allomorph (Base 1): one at the level
of the whole word, and one at the level of individual segments,
exactly along the lines of the Ganong effect. The word-level
effect is evidently by itself insufficient to impose the vowel of
the masculine, whence (8a) *nuvol, while the segment-level
attraction is absent in (8) because [o] is not a perceptual
neighbor of [ε]. FAITH-1(V-quality) is thus voided, as indicated
in (8) by the question mark, and consequently so is the
clustering of (8a). In contrast, in (5a) above the vowel [ε] of
feminine [pʁɔ∫εn] is a perceptual neighbor of the [ε̃] of the
masculine [pʁɔ∫ε̃n]. The resulting attraction by the masculine
form thus results in satisfaction of FAITH-1(V-quality) and the
clustering of (5a) above. Segment-level attraction cannot be
sufficient to turn [ε] to [ε̃], however, since it is not a general
property of French that [ε]s merge with their nasalized
counterparts. Hence the word-level attraction must therefore
also be at work, drawing the liaison form to the masculine
allomorph under gender identity. Thus it appears that the
purely distributional data of the French liaison effectively
reproduce the properties of the Ganong effect: a segment-
sized representation is influenced simultaneously by two types
of neighbors, segment-level and word-level.
Page 17 of 80
In conclusion, the notion of attraction over distance (7) finds a

central role to play in phonology/phonetics as well as
perception. In phonology/phonetics it is key to the
characterization of segmental inventories, segmental
neutralizations, opacity effects, assimilations, and
dissimilations. In perception, the ‘perceptual magnet’ effect,
the ‘phonetic trading relations’ and the Ganong effect, all
seem to embody such a notion as well.
4.4 HEBB’S RULE

The attraction over distance of (7) is beyond the expressive
power of the symbolic computation of traditional generative
linguistics. The reason is that symbols, such as distinctive
features in phonology, are by definition means to isolate
individual dimensions of the representational space while
abstracting away from others. As such, they (p.74) are
insufficient to capture a notion of distance that is defined over
multiple dimensions simultaneously. This is in contrast to
neural computation, in which interactivity among dimensions
is the natural by-product of the pattern of connectivity among
units. In Burzio (2002b) I propose to capture the role of
distance by means of the hypothesis in (9), with transparent
links to principles of neural computation.
(9) Representational Entailments Hypothesis (REH):

Mental representations of linguistic expressions are
sets of entailments; e.g. a representation consisting of
A and B corresponds to the entailments: A ⇒ B, B ⇒ A.
The REH (9) is a virtual restatement of Hebb’s (1949) rule of

mental learning according to which neurons that are active
simultaneously develop synaptic connections that render their
future co-activation necessary. The latter is parallel to the
claim in (9) that simultaneous instantiation of A and B in some
representation would make their co-instantiation necessary in
other representations. Since the entailments of (9) are
elementary constraints, the REH blurs traditional distinctions
such as the one between representations and grammar and
the one between grammar and lexicon. Under the REH all
representations are effectively part of the grammar.
Conversely, we will see below that formerly grammatical
notions, such as that of the word-formation rule, and OT
notions of both Output–Output faithfulness and Input–Output
faithfulness can be reduced to representational entailments.
Page 18 of 80
The REH yields the attraction over distance effect of (7) in the
way illustrated in (10).
(10) Effect of Entailments over Neighboring

Representations
Page 19 of 80
Entailments violated by ¬D in R2, A⇒D
for R1= A, B, C, D: B⇒D A⇒D
C⇒D B⇒D A⇒D
1 2 3 4
Varying R2: A A A ¬A
B B ¬B ¬B
C ¬C ¬C ¬C
¬D ¬D ¬D ¬D
Page 20 of 80
The diagram in (10) reports the effects of a representation R1

on a second representation R2, as the distance between them
increases. R1 consists of the four components A, B, C, D. At
point 1 in the diagram, R2 differs from R1 by a single
component: ¬D. In this situation, the presence of ¬D in R2
gives rise to violations of three of the entailments generated
by R1, as indicated. The reason for those violations is that R2
has each of A, B, C in common with R1, but not D. The
diagram tracks the entailments violated by the presence of ¬D
at each further step, as distance from R1 increases. At point 2,
the number of entailments violated by the presence of ¬D
decreases to two, because now R2 no longer shares C with R1,
which was the source of one of the violations at point 1.
Comparable decreases obtain at points 3 and 4. The violated
entailments in (10) constitute the pressure on ¬D of R2 to
become identical to its counterpart D of (p.75) R1—what we
referred to above as ‘attraction’. Such attraction is directly
proportional to the number of components that R1 and R2
already share, namely their overall proximity, and hence
inversely proportional to their overall distance. The attraction
over distance effect, which I will also refer to as ‘Gradient
Attraction’ (GA) can be further visualized as in (11).
(11) Gradient Attraction effect from entailments
In (11),
the R1 of
(10) is a
point in
multidimensional space which can be thought of as the center

of a hole with variably sloping walls. The R2 of (10) can be
thought of as an object on the slope, affected by gravity. The
closer it moves to R1, the steeper the slope, and hence the
stronger the force pushing it to the center. The linear function
of (10) (linear decrease in the number of violated entailments)
is the first derivative of the sloping function in (11) (steepness
decreases linearly with distance from R1).
Representational entailments are understood to be elementary

constraints that are violable like standard OT constraints, and
will in fact be taken as the basis for various classes of OT
constraints as noted. At the same time, entailment summation
will be taken as the basis for the ranking of those constraints.
Page 21 of 80
Specifically, the class of Input-Output Faithfulness constraints

(henceforth ‘IO-F’) will be characterized as a form of
attraction in which the attractor, R1 of (11) is the input
representation, while for the Output–Output Faithfulness
constraints (henceforth ‘OO-F’) such an attractor will be an
independently existing output form. At the same time, the
attraction resulting from the entailments can also subsume
various forms of markedness in standard OT, since it is held
responsible for triggering various repairs, as in the
assimilation/dissimilation and neutralization cases.
The clustering effects reviewed in s. 4.2 will now follow in the

manner illustrated for the case of (4) above in (12)–(14) below.
(12)
R1: Attractor verb rémedy R2: Forms of the

/lévy stem
A Basic meaning A A
B and segmentism B B
C Stress C ¬C
D Length of stressed V ¬D ¬D
OO-F OO-F (V-

(V- length)2
length)1
≫
(A, B, (A, B ⇒ D)
C ⇒ D)
In (12), it is assumed that the base verb, either remedy or levy

of (4), plays the role of the attractor R1 in (10) and (11). For
present purposes it will be sufficient to assign to (p.76) such
representations the rough internal composition shown, using
the four components A, B, C, D. We take A, B as stand-ins for
the basic meaning and segmental composition of the verb,
while C and D represent the exact position of the stress and
the length of the stressed vowel, respectively. The stem of the
respective -able adjectives is now R2 in (10), (11), subject to
attraction by R1. Such attraction constitutes a form of OO-F in
OT, except that, unlike standard OT constraints, its rank is
modulated by distance. In (12), I consider OO-F for vowel
length, and the modulation shown is due to the stress. When
Page 22 of 80
R2 has the same stress as R1 (C), a difference in vowel length

(¬D) causes a three-entailment violation, reported in (12) as a
higher ranked OO-F. However, when R2 has a different stress
than R1 (¬C), then a difference in vowel length (¬D) only
causes a two-entailment violation, reported in (12) as a lower
ranked OO-F. The two variants of OO-F (V-LENGTH) can now
be deployed in standard OT fashion to correctly obtain the
optimal candidates, as in (13) and (14) below. Shaded cells
express irrelevance of either OO-F variant.
(13)
Page 23 of 80
rémedy MWF MC OO-F (V-LENGTH) CiV-Length OO-F (V-

1 LENGTH)2
a. rémediable * *
b. remédiable * *
c. ? remé:diable * *
Page 24 of 80
In (13), candidate (a) satisfies the Metrical Consistency of (2)

above, also a form of OO-F, but only by violating top-ranked
Metrical Well-Formedness for the reasons discussed in
connection with (4) above, and is thus not optimal. Candidate
(b) is also not optimal because, once stress is shifted, only OO-
F2 is in force, and that is lower-ranked than the constraint
imposing lengthening in the context ‘__CiV’. Candidate (c) is
thus optimal. Note that, just like OO-F (V-LENGTH), so MC =
OO-F (STRESS) should be given in different variants, because
equally modulated by distance. In particular, V-lengthened
candidate (13c) should be dealt with by a lower-ranked MC.
This would only improve its status, however, and hence not
change the conclusions. Consider now (14).
(14)
Page 25 of 80
lévy MWF MC OO-F (V-LENGTH) CiV-Length OO-F (V-

1 LENGTH)2
a. ????? léviable *
b. lé:viable *
Page 26 of 80
Here, the winning candidate satisfies both MWF and MC,

under the parse (lé. vi.a) of the PES analysis. OO-F1 is thus in
force, and candidate (b) is correctly excluded under the given
ranking.
In sum, the REH (9) directly yields the attraction over distance
effect that we earlier (p.77) found to be pervasively present.
The latter is directly relatable to principles of neural
computation and in fact coincides with the claim that
representations are ‘distributed’. The reason is that taking an
entailment A ⇒ B due to a representation R1 to be relevant
also to some other representation R2 presupposes that the A
and B of R1 and those of R2 are the same entities, and
therefore that representations such as R1 and R2 are
‘distributed’ over a general pool of units that include A and B,
rather than being autonomous or local. The foregoing
discussion has provided a partial answer to question (3a)
above on the sources of OO-F effects: the latter effects obtain
between representations that are independently similar.
However, in the case of allomorphy, the independent similarity
has a specific source that we still need to deal with. For
example, in the case of remediable, the independent similarity
with remedy is partly a semantic fact: the intended meaning is
‘able to be remedied’, not ‘able to be read’, or other. But it is
also a morphological fact. It is a general property of the affix -
able that it attaches to verbs; hence the presence of -able
enforces some identity relation between its stem and some
verb in the lexicon. Our residual question is then how could
this be expressed within the framework of the REH (9), or to
put it differently, what is the role of morphology in a system
that sees words as related to one another ‘surface to surface’
in the ways prescribed by the REH. This is in fact question
(3b) above, and the topic of s. 4.6. Section 4.5 next takes us on
a brief detour to address some of the more pressing questions
for the REH.
Page 27 of 80
4.5 WHY SOME ENTAILMENTS MAY BE MORE EQUAL

THAN OTHERS
One of the questions that arises naturally for the above
discussion is what the dimensions of the relevant space are,
namely what A, B, C, D of (10)–(12) may be allowed to refer to.
The answer is that, at least as a first approximation, the nature
of the dimensions involved is orthogonal to the present claim,
which merely addresses issues of the general architecture of
the system. For present purposes one may well assume
decomposition of sound structure in terms of the usual
distinctive features or other traditional notions, although there
are significant inadequacies to the traditional distinctive
features, stemming from the now established role of
perceptual distance. For instance, the common rounding of
back vowels cannot be adequately accounted for by simply
allowing distance between front and back vowels to compound
over the two standard features of backness and rounding,
since this would predict the same distance if front rather than
back vowels were rounded. The inadequacy is in the almost
purely articu-latory basis of the standard distinctive features
(as noted by Boersma 1998). What predicts that rounding
should go with back vowels is their acoustic and hence
perceptual properties, related to length of the front cavity,
already greater with back vowels and hence suitably enhanced
by rounding (Lindblom 1986). Similarly, the same feature (p.
78) of voicing measures different perceptual distances
depending on whether it pertains to an onset or a coda and is
hence in itself inadequate to capture the role of perception. I
must put these issues concerning the proper decomposition of
the perceptual space aside, along with the corresponding
issues concerning the decomposition of the semantic space.
The present claim is merely that there is a notion of global
distance to which all dimensions can in principle contribute.
A second question is the apparent commitment to uniform

interaction among all components of a representation. For
example, given a word of a certain length, the above
discussion may suggest that the first segment in the word
could equally entail the presence of the segment that
immediately follows it and that of the very last segment in the
word. However, it is not part of the present claim that all
components of a complex representation must necessarily
entail each other equally. For one thing, I will argue below that
the strength of individual entailments can compound over
different representations, so that ultimately relative strength
Page 28 of 80
of entailments will come to reflect the overall structure of

one’s mental lexicon or experience. At the same time, there is
also no doubt that the system of mental representation has
inherent biases, as reflected by language universals, and this
will translate in certain entailment structures being privileged
over others. The present claim is—again—only that general
associative mechanisms must be available, not that they are
sufficient. Mapping out the finer structure of the
representational space will be an additional task for future
work, perhaps different than it might have been previously,
but not completely. There is, however, one general prediction
that the present approach makes, which provides an important
second approximation refinement of the REH (9). This is the
prediction that in general the strength of an entailment
relation between two co-occurring components A and B will
depend on their own degree of similarity, as I discuss next.
Consider first that given any two components of the same

representation A, B, they as well as a resulting entailment may
be analyzed into sub-parts, as in (15).
(15) From entailments to sub-entailments for a

representation A, B
I. II. Sub- III. IV. Sub-

Components components Entailment entailments
A w, x A⇒B w ⇒ y; w ⇒
z
B y, z x ⇒ y; x ⇒ z
In (15), supposing that components A, B of (I) are analyzable

into their sub-components w, x and y, z respectively, as in (II),
then the entailment A ⇒ B of (III) will correspondingly be
analyzable into the four sub-entailments shown in (IV). That is
to say, if A entails B, then presumably each sub-component of
A: w, x will entail each sub-component of B: y, z, as in (IV). Now
suppose further that, beside being part of the same
representation, A and B are themselves not unrelated, but are
rather partially similar by virtue of x = y. Then (16) will hold.
The internal structures of A, B assumed in (15) will give rise to

the respective internal entailments in (16.I) and their mirror
images, which I put aside. However, given the identity of x and
y, this will also result in the entailments labeled ‘transitivity
Page 29 of 80
effect’ in (16.II), namely if w entails x, and if x = y, then w

entails y, etc. Such transitivity effect (p.79) results because A
and B are neighbors, and is in fact a formal analog of what is
referred to as a ‘neighborhood effect’ in psycholinguistics,
describing the fact that representations which are active cause
their neighbors also to be partially active. Within the present
discussion, this follows from their entailment structure.
Neighboring representations share some components by
definition, and hence the entailments produced by one
representation will partially instantiate the other
representation as well. We have noted that, for this reason,
neighboring representations are in an attraction relation,
which may sometimes result in their neutralization, but we are
abstracting away from that here. What we are focusing on is
rather that the sub-entailments from the transitivity/
neighboring effect in (16.II) will now compound with those of
(15.IV) that analyze the more general case, giving rise to the
overall structure of sub-entailments in (16), where the
exponents refer to multiple instantiations of the same
entailment. In other words, while—under the REH—two
components A and B will entail one another by being part of
the same representation, they will also entail one another to
the extent that they are neighbors. The conclusion is then that
not all components of the same representation will entail each
other equally. Rather, they will do so in a way that is further
enhanced by proximity. One can see a number of more specific
consequences in this connection, some of which I develop
next.
(16) Sub-entailments for a representation A, B when A,

B are neighbors
I. Internal II. Transitivity III. Sub-

entailments effect (x = y) entailments,
overall
A (w, x) B (y, z) w ⇒ x; y ⇒ z (w ⇒ y)2; w ⇒ z
w⇒xy⇒z x ⇒ y; (x ⇒ z)2
Consider a representation like (17) where the sequential or

temporal scale is characterized schematically by a system of n
binary features, where n is the number of phoneme-sized time
units minus 1.
Page 30 of 80
(17) Entailments under sequential proximity:
Page 31 of 80
Phonemes i n e p t
Time elapsed 0 1 2 3 4
Time Vector ¬1 1 1 1 1
¬2 ¬2 2 2 2
¬3 ¬3 ¬3 3 3
¬4 ¬4 ¬4 ¬4 4
Page 32 of 80
What the above discussion implies for a representation like

(17) is that, while i co-occurs with both n and t, the entailment i
⇒ n will be stronger than the entailment i ⇒ t. The reason is
that unlike i and t, i and n have maximally similar time
representations or vectors by virtue of being sequential
neighbors, respectively the vectors: (p.80) (¬1, ¬2, ¬3, ¬4)
and (1, ¬2, ¬3, ¬4). These three phonemes i, n, t are
compared in (18), where ‘Time′’ refers to the portion of the
time vector over which i and n are identical, namely the last
three components, and ‘Other’ refers to the rest of each
representation: the first component of the time vector and the
structure of each phoneme in space (let us say in terms of
distinctive features).
(18) Spatio-temporal analysis:
Phonemes i n t
Other O-i O-n O-t
Time′ T′-i: ¬2, ¬3, T′-n: ¬2, ¬3, T′-t: 2, 3,

¬4 ¬4 4
The two entailments i ⇒ n and i ⇒ t due to co-occurrence of

these phonemes in the structure [inept] will now map into sub-
entailments in the manner of (19) under the criteria of (15)
above.
(19) From entailments to sub-entailments
i⇒n O-i ⇒ O-n; (O-i ⇒ T′-n)
(T′-i ⇒ O-n); T′-i ⇒ T′-n
i⇒t O-i ⇒ O-t; O-i ⇒ T′-t
T′-i ⇒ O-t; T′-i ⇒ T′-t
However, the two entailments parenthesized in (19) will be

effectively boosted by the spatio-temporal proximity of i and n,
not paralleled by the proximity of i and t. The reason is the
transitivity effect in (20), an instance of the effect in (16)
above.
(20) Sub-entailments added by representational overlap
Page 33 of 80
Internal entailments Transitivity Effect
i: O-i, T′-i n: T′n, O-n due to: T′-i = T′-n
O-i ⇒ T′-i T′-n ⇒ O-n O-i ⇒ T′-n; T′-i ⇒ O-n
That is, given their internal structures, i and n will generate

the internal entailments given, respectively. However, given
that T′-i is identical to T′-n as shown in (18), those entailments
will translate into the ones in the rightmost column in (20),
respectively, exactly identical to the parenthesized ones in
(19), which are thereby boosted. No such boosting effect will
occur for t since i and t do not share any temporal structure
within the time frame considered. If the frame were extended,
e.g. to instants 5, 6, etc. some sharing would occur, but always
less than occurs between i and n. Hence both components of n,
O-n and T′n, are entailed by i in virtue of the simple co-
occurrence of i and n in the same structure [inept] (the sub-
entailments in (19)), but they are both additionally entailed in
virtue of the sequential proximity of i and n.
What the foregoing means effectively is that segments should

be better able to ‘bind’ together in recurrent patterns when
they are contiguous—exactly the generalization that describes
‘concatenative’ morphology. We are thus shedding light on the
non-existence in languages such as English of discontinuous
morphemes like a hypothetical in-…-y, such that for example,
given accurate, in-accurac-y would just mean (p.81)
‘inaccurate’. A well-known alternative to concatenative
morphology is of course the Semitic-type ‘root and pattern’
morphology (McCarthy 1981). From the present point of view
this would in fact involve an alternative type of binding also
expected, in the way illustrated in (21).
(21) Semitic morphology (Arabic)

katab-a ‘he wrote’
kutib-a ‘it was written’
aktub-a ‘he was writing’
…
Page 34 of 80
Phonemes k a t a b a
Other O-k O-a O-t O-a O-b O-a
Major Class [+ +C -C +C -C +C -C
conson.]
Page 35 of 80
The case in (21) is simply the spatial counterpart to the

temporally based (18), as the segments that bind together
here are spatial rather than sequential neighbors. In
particular, the member of the root segmentism ktb share the
major class feature [+consonantal] and are thus expected in
(21) to tie to each other more strongly than to the vowels by
parity of reasoning with (18)–(20).2 The conditions under
which elements can bind together in morphological systems—
spatio-temporal proximity—are thus predicted to be very much
analogous to those that underlie the phonological attraction
relations manifested in assimilatory and dissimilatory
phenomena from this point of view.3
To recap, while in general clustering effects seem able to

implicate unrelated dimensions, there is also reason to believe
that co-occurring elements that are sequential or spatial
neighbors cluster more strongly. The required refinement of
the REH appears to involve simply expanding components into
sub-components, and correspondingly entailments into sub-
entailments. The sub-entailments yield neighborhood effects
that overlay the basic co-occurrence effect under the same
basic principle of the REH, thus accounting for the stronger
clustering or binding of neighbors.
(p.82) 4.6 WHAT TO DO WITH MORPHOLOGY

4.6.1 How affixes encode OO-Faith relations
We return here to question (3b) above concerning the
relationship between OO-F constraints and morphology. A
fundamental characteristic of the way in which affixes
combine with stems is that they do so violably, as shown in
(22).
(22)
Phonological alternations
I. Morphological and II. Contras III. Allophonic

semantic irregularity tive variation ariation
a. compUlS-ive, der[i]vAT-ive
derivAT-ive
b. regrett-able, adm[i]r-able
INEVIT-able
c. symptomAT-ic paras[i]t-ic parasi[?]-ic
Page 36 of 80
Phonological alternations
d. permi[?]-ing
In (22), the suffixes -ive, -able, -ing all select verbs as their
stems, but rather than the forms in (22), the actual verbs are:
compel (a), der[ay] ve (a), adm[ay] re (b), and permi[t] (d). In
addition, there is no verb (in)evit (b), and regrettable (b) does
not have the expected meaning ‘able to be regretted’, but
rather refers to things that must be regretted instead.
Similarly, -ic selects nouns, but the actual nouns are rather
symptom and paras[ay]te instead of the forms in (22c). Such
distortions can be divided into the three categories of (22).
There are simply irregularities, both of form and meaning, as
in column (I), although irregularities may in turn correspond
to various degrees of sub-regularity, e.g.: symptomAt-ic,
problemAT-ic, axiomAT-ic, etc. Then there are regular
phonological effects, further subdivided into contrastive
(column (II)), and allophonic (column (III)). English vowel
shortening is a contrastive type of variation because it
neutralizes a contrast, that between short vowels and their
long or diphthongal counterparts. These contrast because they
have non-complementary distributions: [bayt]/ [bIt], etc.
Instead, American English flapping is allophonic as the
alveolar flap [?] is in complementary distribution with regular
alveolar stops [t/d]. Each individual suffix appears to control
the general degree to which its stem can fail to match an
independent word in the lexicon. The affixes that were once
classified as ‘Level 1’ in ‘Lexical Phonology’ (Kiparsky 1982a,
b) and exemplified in (22a, b, c) tolerate mismatches rather
well, hence featuring both sheer irregularities and regular
phonological alternations. In contrast, those that were
classified as Level 2, such as -less, -ness, -ful, and -ing of (22d)
tolerate few mismatches, essentially only those coming from
allophonic variation, such as the one shown in (22).
Such violability of the combinatorial properties of affixes

would follow if the latter were types of entailments derivable
from the REH (9), since we have seen that those (p.83) are
generally violable. This view can in fact be upheld
straightforwardly, as we see by considering (23).
(23)
Page 37 of 80
LEXICON: Entailments Entailments Entailments

I II III
preventive ive ⇒ / prevent ⇒ V ive ⇒ /V ___

prevent___
generative ive ⇒ / generate ⇒ ive ⇒ /V ___

generat___ V
Summation: noise ___ ive ⇒ /V ___
Under the REH, a morphologically complex item like prevent-

ive will have, as part of its entailment structure, some
entailments to the effect that, if there is a sequence ive, there
must be a sequence prevent preceding it, and similarly for
generat-ive, as indicated in column (I). By themselves,
entailments such as those of column (I) would be largely
discordant, resulting in no generalization when they are
conjoined. Consider, however, that besides containing the
items preventive and generative, the mental lexicon also
contains prevent and generate, both of which are verbs. The
specific internal structure of prevent will now contain the
entailment of column (II), namely: if there is a sound structure
[prIvent], the associated category is ‘verb’, where lexical
category is presumably part of semantic representation. In
combination with the entailment from column (I), this new
entailment will result in the entailment of column (III) by
transitivity, namely: if ive entails prevent, and if prevent is a
verb, then ive entails a verb. Similarly for generat-ive, despite
the difference between the stem gener[Ə]t- and the verb
gener[ey]te: the two structures are still sufficiently similar to
ensure that most of the entailments abbreviated by the single
conditional of column (I) will be relevant to column (II). The
entailment of column (III) will thus still hold substantially in
the case of generative, though presumably in a somewhat
weakened form. Now, unlike the entailments of column (I),
those of column (III) are concurrent, thus yielding a higher-
order entailment or generalization by summation over the
lexicon. Such higher-order entailments are effectively
subcategorization frames for affixes, expressing their general
selectional or combinatorial properties. I will henceforth refer
to them as ‘SELECT’ entailments. They differ ontologically
from the subcategorization frames or the word-formation rules
of previous approaches to morphology by not being part of an
extrinsically given grammar, but being rather emergent
Page 38 of 80
properties of the lexicon itself—a reflex of the self-organization

resulting from the REH. Furthermore, they differ from the
machinery of those approaches also empirically, by virtue of
being violable—a crucial difference as we see below.
We are now in a position to provide a full analysis of the OT

notion of OO-F, the driving force behind PU. The latter is the
result of the combination of two different types of entailments,
those due to the GA effect, and those due to the SELECT
effect, in turn both reflexes of the REH. The respective
contributions of these two effects are underscored by the fact
that, on the one hand, SELECT effects are not always present.
For instance, in the French cases in (5) and (8), the liaison
form is not related to the feminine allomorph via affixation,
hence no SELECT entailments are expected. The case PU over
affixes, where affixal rather than stem material is kept
invariant, also does not depend on the presence of further
affixes, as we see in the next section. In those (p.84) cases,
GA will go it alone. On the other hand, when SELECT effects
are present, they have the ability to substantially modulate the
resulting OO-F/PU effect. We have noted with regard to (22)
above that with the Level 1 affixes the OO-F imposed on the
stems is generally weak, to the point that there need not be an
independent base at all: inevit-able/*inevit, in contrast to the
Level 2 affixes that impose a strong identity requirement. The
GA effect strictly depends on global distance, and it is thus
only the SELECT entailments that could be responsible for the
difference. There are two different ways in which SELECT
entailments can obtain different strengths. One is numerical. If
such entailments sum up over instances as in (23.III), then the
number of instances in the lexicon (roughly, what in
psycholinguistics is referred to as the ‘type frequency’, i.e. the
lexical frequency of that type of morphological formation) will
be a determinant of strength. The other factor is what we may
call ‘seeded idiosyncrasy’. If a set of morphologically derived
forms comes into the language with a high degree of
idiosyncrasy, as is surely the case for the English Latinate
(Level 1) lexicon in general, then it will give rise to weaker
SELECT entailments. For instance a form like inevit-able will
not contribute—but will in fact contribute negatively—to the
entailment that -able must select a verb. Intermediately
idiosyncratic forms will correspondingly contribute to SELECT
entailments only weakly. On the present approach,
morphological regularity has a self-feeding character since
Page 39 of 80
regular forms recruit others into regularity via the SELECT

entailments they generate. The English Level 1 lexicon could
thus be in the current equilibrium through a combination of
relatively low numbers and historical accident, remaining
below a threshold that may lead to the almost complete
regularity of the Level 2 lexicon once crossed.
While an outgrowth of the approach independently developed

in PES, the present perspective converges substantially with
Bybee’s (1988, 1995). Bybee’s notion that there exists a
network of surface-to-surface connections whose strength
depends on the degree of similarity between items has a direct
counterpart in the present claim that representations
influence each other by means of the entailments, an influence
that is expected to depend on their degree of similarity (GA).
Bybee’s further notion that sets of parallel connections form
‘schemas’ that can generate new items finds an analog in the
SELECT entailments of (23.III), and the contribution of ‘type
frequency’ to the strength of a schema postulated by Bybee is
captured here by the summation of such entailments over the
lexicon.
Returning to (22), given that the SELECT entailments (23.III)

and the resulting OO-F effect are violated with the Level 1
affixes, the question is what exactly compels those violations.
The phonological alternations in (22.II, III), as in adm[I]r-able,
permi[ſ]-ing, etc., are obviously due to markedness constraints
of the phonology. As for the morphological and semantic
irregularities of (22.I), as in compUlS-ive, etc., I have argued
in Burzio (2000a, 2002a) that the OT notion of Input–Output
faithfulness (IO-F) is relevant in this connection. Once
morphological relations cease to be expressed by a common
input in UR and are rather handled by OO-F (= SELECT and
GA), the notion of input can then be utilized with
morphologically complex words to express their
idiosyncrasies, just as with morphologically simple ones.
Morphological (p.85) irregularity is then controlled by the
relative ranking of OO-F and IO-F. Dominant OO-F yields
morphological regularity—the Level 2 syndrome exemplified
by affix -hing of (22d), while dominant IO-F enables
irregularity to exist: the Level 1 syndrome of (22a–c), as shown
in (24).
(24)
Page 40 of 80
a. Morphological irregularity (Level 1 Affixes)
Input: /compUlS-ive/ Base: / IO- OO-

compel/ F F
a. compell-ive *
b. ☞ compuls-ive *
b. Morphological regularity (Level 2 Affixes)
Input: /compUlS-ing/ Base: / OO- IO-

compel/ F F
a. ☞ compell-ing *
b. compuls-ing *
On this approach, morphological irregularity receives an

analysis within OT which parallels that of phonological
markedness. The latter is also a form of ‘irregularity’, arising
when markedness constraints—the source of regularity—are
dominated by IO-F, as shown in (25).
(25)
a. Phonological irregularity (Marked value,
English)
Input: /cri:me/ IO-F (V-length) *V:
a.crime *
b.☞ cri:me *
b. Phonological regularity (Unmarked value,

Italian)
Input: /cri:mine/ *V: IO-F (V-length)
a.cri:mine *
b. ☞ crimine *
On this analysis, IO-F constraints are thus the sole source of

irregularity, both phonological and morphological.
Note that in the present approach, in which there are only

surface forms, the ‘input’ can no longer be thought of as
constituting a separate representation like the traditional UR
and thus needs to be reconceptualized. This consequence
Page 41 of 80
obtains in significant ways in Prince and Smolensky’s (1993)

formulation as well. Under their notion (p.86) of ‘Lexicon
Optimization’, in morphologically simple items like [mérIk]
input and output representations would be identical, a claim
equivalently statable as the present claim that there is only a
surface form and no UR. Under Lexicon Optimization, such a
single representation must be thought of as having the
internal partition depicted in (26).
(26) Surface forms under ‘Lexicon Optimization’ (for

morphologically simple words)
That is to
say, in the
single
surface
form [?
mérIk?], some properties, like the specific choice of segments,
will result from input information. Let us call these ‘PI’. Other
properties on the other hand, like the stress and
syllabification, will result from Mark-edness constraints. Let us
call these properties ‘PM’. In the Lexicon Optimization
approach, the PI/PM distinction is not reflected by respective
distributions over different levels, but rather by constraint
ranking. For any aspect of the representation α, PI α obtains iff
IO-F(α) ≫ Markedness(α), while PM α obtains iff the opposite
ranking holds. Hence, the long vowel of cri:me (25) above is a
type of PI, while the short one of Italian crimine (25b) is a type
of PM. So, while under Lexicon Optimization the input to [?
mérIk?] is the same full-fledged [?mérIk?], only some if this
input—what I have called PI, is effectively ‘active’ in
determining the output. The rest of it is passive, under the
sway of the Markedness constraints—what I have called PM.
Prince and Smolensky do not extend Lexicon Optimization to

morphologically complex words, because they adopt the
traditional approach to allomorphy, in which morphemes have
a unique form in UR. Since these forms cannot be
simultaneously identical to all of their surface reflexes,
Lexicon Optimization becomes inapplicable. The present
approach rejects UR and by doing so can maintain the
arguably more natural claim that Lexicon Optimization is
always true. The only difference between simple and complex
words in the present perspective is that the latter invoke one
Page 42 of 80
additional set of constraints: OO-F constraints, hence turning

the bi-partition of (26) into the tri-partition of (27).
(27) Surface forms in the Output–Output Faithfulness

approach
The new
set of
properties
PO will
represent
the aspects of the representation that are determined by OO-F
constraints by virtue of the constraint ranking. As in the case
of [ƏmérIkƏ], the input to compulsive can be taken to be the
exact same form, but only (p.87) some of it will be effectively
active. In (24) above this part of the input was given in caps.
The rest of the input conforms with OO-F, just as the stress of
[?mérIk?] conforms with Markedness. I return below to the
question of why only a fragment of the input /compUlS/ (the
portions in caps), is active in this fashion (i.e. corresponds to
PI of (27)). Now in (24b) above the same input compUlS- was
given for the form compelling, but in that case it was only as a
hypothetical input, to show that it would have no effect.
In sum, we take surface forms to be evaluated by three classes

of constraints simultaneously. IO-F and Markedness
constraints work much as in standard OT, while OO-F
constraints take over the role of traditional morphology,
resulting in a fully integrated parallel system. This move
makes UR dispensable not only with underived items (as with
Prince and Smolensky’s ‘Lexicon Optimization’), but
altogether. In turn, this makes it possible for IO-F constraints
to be utilized to express morphological irregularity similarly to
the way in which they standardly express phonological
markedness.
4.6.2 Interfacing phonology and morphology

This approach, in which affixes are instrumental in modulating
the rank of OO-F constraints as in (24) above, has the unique
ability to account for the important correlation that exists
between morphological and phonological regularities. As
noted in connection with (22) above and discussed in Burzio
(2002a) those regularities correlate inversely. The sectors of
the lexicon that abound in morphological irregularity, such as
that Level 1 affixes are rich in certain phonological
Page 43 of 80
alternations which are perfectly well-expected and hence

‘regular’, such as re-stressing: párent/paréntal, re-
syllabification: in.ter.ve:.nø /in.ter.ven.tio.nø (continuing to
assume the final null nuclei of the PES analysis), and vowel
shortening: parasitic, criminal, etc. In contrast, items featuring
Level 2 affixes, which are exempt from morphological
irregularity, exhibit few of the expected phonological
alternations, and are in this respect highly irregular
phonologically, e.g. éffortless (exceptional stress), discreetness
(exceptional syllable creet), cri:melessness (no shortening).
The same is true for the much debated English past tense. The
morphologically regular class is systematically irregular
phonologically, by way of otherwise unattested syllables like
that of [biyp.tø] (beeped), while the morphologically irregular
class features straightforward phonological regularities, like
the expected shortening of [kiy.pø]/[kep.tø]. Essentially, the
only kind of phonological alternation that obtains within the
Level 2 lexicon is the allophonic kind, as in (Am. English)
permit /permi[?]ing (22d) above.4
(p.88) The correlation in question follows from considering

the rankings that were invoked to deal with the distribution of
morphological regularity, repeated in (28).
(28) Morphologically motivated rankings:
Page 44 of 80
Morphological Irregularity: (24a)
OO-F2 ≫ IO-F ≫ OO-F1
Morphological Regularity: (24b)
Page 45 of 80
Given these rankings, any Markedness constraint of the

Phonology will always have a greater chance of being
dominated by OO-F2 than by OO-F1, and never the opposite.
Since Markedness constraints are the source of phonological
regularity, it follows that Level 2 items, controlled by OO-F2,
will be phonologically less regular than Level 1 items, whence
the noted correlation.
The rankings in (28) also predict two additional effects related

to the relative ranking of IO-F constraints. One is that Level 2
items (controlled by OO-F2) should tend to be phonologically
more marked than underived items (controlled by IO-F), as is
surely correct: there are no syllables like that of [biyp.tø]
among underived items, or stress patterns like that of
éffortless(ness). The second expected effect is that
phonological processes that may work over Level 1 items
(controlled by OO-F1) may ‘block’ in underived items
(controlled by IO-F). This is exactly the case with vowel
shortening, as shown in (29) in contrast to (25) cri:me above.
(29) Vowel shortening with Level 1 items
Base: cri:me *V: OO-F1 (V-length)
a. cri:minal cri:min-al *
b. ☞ criminal crimin-al *
Note that the case in (29) is not just an instance of the ‘tri-
syllabic’ shortening, which would naturally not be applicable
to cri:me. As argued in Burzio (1993, 2000a) and PES, tri-
syllabic shortening is a purely descriptive category, of no
theoretical significance. Shortening applies to all positions
within Level 1 items in principle (including final syllables:
infinite), the constraint at work being the maximally general
one of (29), which is independently motivated by the
difference between English and Italian of (25) above. The
English Level 1 lexicon is in fact very much like the whole of
Italian, favoring short vowels. Vowels may turn up long under
certain circumstances, however, in stressed open penultimates
in particular, as in desí:rous. This is because of competition
with the Metrical Consistency (MC) of (2) above. Given a
prohibition on stressed light penultimates (the foot *(L σ),
banned by the Metrical Well-Formedness of (2)), shortening
would force a violation of MC, as in desí:re/*désirous. Even this
effect is paralleled by Italian, where stressed open
Page 46 of 80
penultimates undergo lengthening: sapó:ne (p.89) ‘soap’. The

only difference is the source of stress: IO-F in Italian, OO-F in
the English case. However, it turns out that MC does not quite
dominate *V:, but in fact essentially ties with it, hence
permitting variation as in desí:rous (MC satisfied)/
blásphemous (*V: satisfied. Cf. blasphé:me), a point to which I
return below. On the other hand, antepenultimate stress does
not require a heavy syllable (witness ámerica, etc.), whence
the much greater regularity of shortening in the ‘tri-syllabic’
cases—criminal, divinity, natural, etc.—in which both MC and
*V: are satisfied simultaneously. Other environments also yield
regular shortening for comparable reasons, as argued in the
references.
Consider further that, under the ranking in (28), any process

that blocks in morphologically underived items, like vowel
shortening, is guaranteed also to block in Level 2 items for the
obvious reason that any Markedness constraint dominated by
IO-F, will also be dominated by OO-F2 by transitivity. We have
already seen that this prediction is indeed correct for
shortening: cri:melessness, etc.
In sum, the complex distribution of English vowel shortening,

given in (30), follows from the postulated constraint rankings.
(30)
Domain English Vowel Ranking

shortening responsible
a. Underived no: parasi:te; IO-F ≫ *V:

di:nosaur; crime
b Level 1 yes: parasit-ic; *V: ≫ OO-F1

(some crimin-al, divin-ity
contexts)
b′. Level 1 yes/no: blasphem- *V: ≫/≪ MC

(other ous, desi:r-ous
contexts)
c. Level 2 no: cri:me-less-ness OO-F2 ≫ *V:

derived
I will return below to the fact that while OO-F1 (V-length) is

outranked by *V: as in (30b), MC, also a form of OO-F1, only
ties with *V:, as in (30b′).
Page 47 of 80
In contrast to vowel shortening, consider now allophonic

variation, in general attributed in OT to the schema in (31)
(Kirchner 1997, Burzio 2000a) which is applied here to the
case of American English flapping of (22.III) above.
(31) Allophonic variation in OT (e.g. Amer. English

Alveolar Flap):
Markedness, ≫ Markedness, ≫ IO-F

contextual eneral
*t, d/V́ ___V ≫ *Flap ≫ Ident

(flap)
Putting aside the exact featural analysis of the alveolar flap

([+sonorant] according to Kenstowicz 1994: 68), this sound is
in this analysis excluded in general by an a-contextual
markedness constraint that dominates IO-F, but is then
mandated by a contextual markedness constraint that
dominates the rest of the schema in turn, as in (31). The latter
constraint excludes regular coronal stops t, d, between a
stressed vowel and another vowel (further ranking among
faithfulness constraints, not given in (31), (p.90) would also
be required to exclude alternative repairs). Consider now what
our ranking in (28) above predicts in this connection. With
morphologically derived items, OO-F will replace IO-F in (31)
in the present system. With OO-F1, it is clear that the schema
will remain unaffected (since OO-F1 is dominated by IO-F).
This correctly guarantees that allophonic variation will never
distinguish between underived and Level 1-derived items as
vowel shortening does. Hence, the well-recognized fact that
allophonic variation never ‘blocks’ in underived environments
(Kiparsky 1993 and refs.) follows automatically from this
analysis. With OO-F2, however, there is no guarantee that the
allophonic variation of underived items will still obtain, but
there is also no guarantee that it will disappear. All we know is
that the ranking ‘OO-F2 ≫ IO-F’ holds, so that OO-F2 may or
may not end up dominated in (31). Thus the observed fact that
flapping has the same distribution with Level 2 items as with
all other items is consistent with the present analysis. Other
types of allophonic variation do disappear at Level 2, however,
as is also consistent with the analysis. For instance, the
variation between syllabic [l̝] and non-syllabic [l] of cyc [l̝]
(cycle)/cyc[l]ic blocks with Level 2 -ing in some dialects, that
Page 48 of 80
thus feature OO-Faithful cyc[l̝]ing, rather than the regularly

allophonic cyc[l]ing of other dialects (see also Burzio 2002a).
In sum, the relative ranking of IO-F and OO-F constraints

motivated by morphological generalizations automatically
yields significant phonological generalizations as well.
Specifically, it yields an implicational hierarchy ‘Level 2 ⇒
Underived ⇒ Level 1’, such that any process applying to Level
2 items will also apply everywhere else as well, while any
process applying to underived items will also apply to Level 1
items, though not necessarily the other way around in each
case. It also correctly derives the well-established
generalization that if a process is allophonic it will never block
in underived environments.
4.6.3 Finer-grained interactions

The foregoing discussion has argued that OO-F reduces to the
more primitive notion of Representational Entailment. It
appears that the OT notion of IO-F can also be so reduced. On
the present view, an input representation is a bundle of
entailments potentially in competition with other entailments
coming from the rest of the lexicon. In particular, entailments
that are multiply instantiated and thus represent cross-lexical
generalizations will constitute a source of pressure on an input
representation. Such cross-lexical generalizations come in two
basic classes in the present system: phonological markedness
constraints, and OO-F. To address the interaction between
such generalizations and an individual representation, we
must consider that generalizations are by definition always
lower-dimensional than individual representations. Thus
compare a hypothetical generalization ‘A ⇒ B’ with a
representation ‘A, ¬B, C, D’ which violates it, as in (32).
(32)
Representation: A, Generalization:
¬B, C, D
Entailments A ⇒ ¬B; C ⇒ ¬B; D ⇒ A⇒B

(B/¬B): ¬B
(p.91) In (32), there is one entailment favoring ¬B due to the

generalization, but there are three entailments favoring B
from the representation itself, a clear advantage. Higher
dimensionality is thus the edge that enables input
Page 49 of 80
representations to output faithfully, violating existing

generalizations. The generalization in (32) could be, for
example, the SELECT entailment ‘IVE ⇒ /V ___’ of (23) above.
We know that this entailment is shorthand for a larger set of
entailments, since ive is a complex representation. However, a
still larger set is represented by the competing entailment ‘ive
⇒ /compuls ___’(effectively: ‘when combining with the meaning
‘compel’, -ive is preceded by compuls). This is because the
structure compuls with its associated meaning is more
complex than just ‘V’. At the same time, some of the
entailments corresponding to compulsive also benefit from
other similar representations, like repulsive, impulsive, etc.,
further assisting the irregular outcome, and accounting for the
noted fact that lexical irregularities are often subregularities.
While lower-dimensional, generalizations have their own

numerical advantage, however, in the fact just noted they are
multiply instantiated. Hence, in principle, the competition
between the general and the specific can have either outcome,
the former benefiting from its multiple instantiations, while
the latter benefits from its higher dimensionality which results
in a larger number of entailments validating each component
of the representation. Individual outcomes will depend on the
actual numbers that enter into the entailment calculus. While
we are not in a position to deal with such actual numbers at
present, we have reviewed some of the statistical reasons that
would lead to different outcomes in the Level 1 and the Level 2
domains, with individual idiosyncrasy prevailing in the former,
while regularity prevails in the latter.
Our discussion has thus far not accounted for the fact that
morphological idiosyncrasy within the Level 1 lexicon is not
totally unrestrained. If the dominance relation ‘IO-F ≫ OO-F’
postulated above were categorical, with an input
representation always prevailing over the demands of
morphology, then OO-F should have no effect at all, and any
resemblance between affixed stems and independent lexical
items should arise only by sheer accident, which is surely
incorrect. The remedy to this is in the fact that the basis to
OO-F constraints is not only in the SELECT entailments, which
may have a fixed rank, but also in the GA effect, which
modulates the overall OO-F effect in accordance with
representational distance. The revised prediction is, then, that
for regions that are very close to an attractor, OO-F may well
dominate IO-F even in the Level 1 domain, while for more
Page 50 of 80
remote regions the opposite ranking would hold as assumed

above. While we are also not in a position to verify this
prediction with a high level of precision, several facts seem
consistent with it. One is the pervasive phenomenon of bound
stems, e.g. arbore-al. In such cases there is no attractor,
except semantically (‘arbore’ ≈ ‘tree’), and OO-F is predicted
to be at its minimum, hence interfering minimally with this
type of form. In contrast to this and other idiosyncrasies, there
are virtually no exceptions to the metrical consistency effects
of (2a′, b′) above in the analysis of PES. This would follow from
the fact that metrical consistency obtains under substantial
segmental identity almost by definition. Specifically, cases of
bound stems are irrelevant to metrical consistency, precisely
because of the absence of (p.92) a base with which to be
consistent. Hence, metrical consistency occurs under
conditions of maximal attraction. Recall that where segmental
identity is breached, metrical consistency tends to fail as well
as expected: lárynx/larýnges. This then accounts for the
relatively higher rank given to MC in (30) compared with the
more general OO-F1. The distribution of vowel shortening is of
further relevance in this connection. In the analysis of PES, the
shortening constraint *V: of (30) is expected to be always
satisfied in ‘tri-syllabic’ environments like divínity (30b)
because it does not interfere with MC, but it is expected to be
satisfied only variably in penultimate syllables as in (30b′),
where it does interfere. However, the tri-syllabic shortening
does have exceptions, like obé:sity. In Burzio (2000a), I argued
that these exceptions arise in the same way as the
morphological irregularity of compulsive, namely by input
specification of vowel length. With IO-F dominating *V:, the
long vowel of obé:sity would be on a par with that of underived
obe:se. However, exceptions to trisyllabic shortening appear to
have particularly transparent semantic relations with their
bases, so that an alternative might be to postulate a higher-
ranked OO-F relation due to the tight semantic connection.
Similarly, the variability of the shortening of penultimate
syllables was handled in Burzio (2000a) by IO-F. With MC and
*V: in a tie, IO-F would be the natural tie breaker. However,
once again, non-shortening cases like desi:rous seem more
palpably related to desi:re than shortening blasphemous is to
blasphe:me, making an account in terms of variably-ranked
OO-F1, also a possibility. Ultimately, this issue may be of the
‘chicken-and-egg’ kind, since we take semantic non-
transparency to be a form of IO-F, like morphological non-
Page 51 of 80
transparency.5 In conclusion, the ranking ‘IO-F ≫ OO-F’ does

not seem to obtain categorically within the Level 1 domain. In
this regard the earlier discussion was thus simplified, but
remains none the less valid to the extent that the latter
ranking finds substantial opportunity within Level 1, and none
within Level 2.
Summing up this section, the REH (9) enables us to construe

the combinatorial principles of morphology as types of
entailments and, as such, as violable. Violability is established
by two classes of violations—phonologically driven allomorphy
and morphological irregularity—both resulting in disuniform
paradigms. Paradigm uniformity results when the
combinatorial principles (the SELECT entailments of (23. III))
are high-ranked, yielding high-ranked OO-F. Approaches that
take morphology to consist of inviolable rules or principles
need to deal with the violations in terms of sequential ordering
of operations. On those approaches, morphology would be
enforced first, yielding an ‘underlying’ level of representation,
and then other machinery would apply next, distorting the
morphology’s outcome. This traditional conception is falsified
by two considerations. First, there is never any good reason
from this point of view why morphology should ever be
irregular. The various ‘readjustment (p.93) rules’, ‘filters’,
and other devices proposed in this connection merely state the
problem, in so far as they fail to reduce irregularity to
independent properties of the system. This contrasts with the
present approach, which seeks to reduce irregularity to the
interplay of IO-F and OO-F, both independent components of
the system. The second consideration is that there will never
be any good reason why the two types of phenomena that
occur downstream from the morphology—irregularity and
phonologically based allomorphy—should have substantially
the same distribution: both are present with Level 1 affixation,
and both are absent with Level 2 affixation. However one
chooses to capture morphological irregularity in that
perspective, the relevant subsystem will end up being
independent of the one responsible for the phonology. It is only
in the present approach, in which morphological
generalizations, i.e. OO-F relations, can either be weak or
strong, that those two subsystems can have something in
common: they are both competing with the morphology. When
morphology is weak (low-ranked OO-F), competitors are let in.
When it is strong, they are shut out, together. By contrast, in
Page 52 of 80
Kiparsky’s (1982a, b) ‘Lexical Phonology’, for instance, there

is a partial account of morphological irregularity in the
assumption that the output of each level must be a lexical
item. So, while one can have Level 1 ARBORE-al, one cannot
have Level 2 *ARBORE-less (but only tree-less). The reason is
that the input to Level 2 must be the output of Level 1 and
hence necessarily a lexical item, excluding a bound stem like
ARBORE-. Aside from being stipulatory this assumption is
unrelated to the one needed to ensure that most phonological
alternations are excluded at Level 2, which is handled by
assuming that Level 2 phonology has fewer rules. Recasting
matters into the contemporary OT framework will not help.
Level 2 phonology would now presumably be characterized by
higher-ranked IO-Faithfulness, but that will remain unrelated
to the fact that its input cannot be a bound stem. Note, too,
that it is not faithfulness in general that is high-ranked at
Level 2, but rather just the faithfulness that concerns stems.
Thus, contrastive variation is banned quite generally within
Level 2 stems, but not within Level 2 affixes, witness cat[s]/
dog[z], bribe[d]/peep[t], etc. The Level 2 generalization does
therefore not concern phonology per se, but rather stems, and
consists of the requirement that they must equal words, as
argued in PES (274) to which the present discussion is a
sequel.
Phonology is also intractable within the ‘dual route’ approach

to morphology of Pinker (1991) and others. That approach
maintains that regular morphology consists of a system of
symbolic rules, while irregular morphology reflects an
analogical/asso-ciational system. There is no reason from that
point of view why the phonology should be unrestricted within
the analogical route, while within the symbolic route it is
highly restricted if it applies to stems, but unrestricted again if
it applies to affixes. There seems no principled way to
characterize the curious distribution of phonological regularity
from that perspective. The notion that there is one ‘route’
specifically dedicated to regularity only predicts that regular
phonology (e.g. regular stress, regular syllabification) should
coexist with regular morphology. As we have seen, the actual
facts are exactly the opposite (see also Burzio 2002a).
(p.94) 4.7 TYPES OF PU

4.7.1 Uniformity of stems
Page 53 of 80
Beside GA, which is triggered by overall similarity, we have

seen that the selectional properties of affixes are also an
important source of PU. The general schema in (33) is the one
instantiated by (23).
(33) Uniformity with a bare base
Entailments I Entailments Entailments

II III
X-A A ⇒ /X ___ X⇒V A ⇒ /X
where X=V
preventive ⇒ / prevent ⇒ V

ive prevent ___
In (33), X is a stem like prevent and A an affix like -ive. As we

saw above, the entailment (I)will result in the ‘SELECT’
entailment (III) when combined with the entailment (II)
contributed by the lexical item prevent itself. This result
follows from the transitivity of the entailments. The SELECT
entailment is given here in a slightly different format than in
(23) for expository reasons. From this perspective, a
comparable SELECT effect will also be expected when the
‘base’ is not unaffixed as in (33), but is rather itself an affixed
member of the same paradigm, as in the Spanish case in (1)
above. This is shown in (34).
(34) Uniformity with an affixed base
Entailments Entailments Entailments

I II III
X-A1 A1 ⇒ /X ___ A1 ⇒ /X X ___
X-A2 X ⇒ /___ A2
where X= /___
A2
am á-ba ba ⇒ /am á ___
am am á ⇒ / ___bamos
x00E1;-
bamos
Page 54 of 80
In (34), two affixes A1 and A2 share the same stem X. Each

affix then entails the presence of the stem next to it just as in
(33) and as indicated in column (34.I) for A1. Entailments
generated by co-occurrence are symmetrical, however, so that
entailments in the other direction will also obtain, as indicated
in column (II) for A2. The combination of entailment (I) and
entailment (II) will yield entailment (III) by transitivity just as
before. That is, the X that A1 is associated with is also
systematically associated with A2, so that A1 will come to
select as its stem a general object whose property it is that A2
also occurs attached to it (column III). This contextual
property of X in (34) is parallel to the property of X in (33) of
being a verb when it occurs by itself. The attained strength or
rank of the SELECT entailments in both (33) and (34) is
expected to depend on class size, that is on how many times
entailments (I) and (II) are simultaneously instantiated
through the lexicon (roughly related to the notion of ‘type (p.
95) frequency’). This sheds light on the fact that in the
overwhelmingly larger Romance first conjugation (It. -are; Fr. -
er; Sp. -ar), PU is much more prevalent than in the other,
smaller conjugations. For instance the Italian -ere conjugation
exhibits palatalization alternations like vo[l]-ete/
vo[λcorrespondingly only hasλ]-iamo ‘want 2Pl/1pl’, but the -
are conjugation correspondingly finds only non-alternating
vo[l]-ate/vo[l]-iamo ‘fly 2Pl/ipl’. All other segmental
alternations of Italian are correspondingly also missing from
the -are conjugation. Similarly, in French, the -ir conjugation
has alternations like je[paʁ]/nous [paʁt-õ] ‘I leave/we leave’ in
which the stem final consonant appears only in a ‘linking’
capacity, similarly to the liaison cases of (5) and (8) above. The
-er conjugation, however, correspondingly only has je[pɔʁt]/
nous [pɔʁt-õ] ‘I bring/we bring’ where the consonant is
uniformly present (see Burzio 1997, forthcoming, for
additional Italian material; Tranel 1996 for further discussion
of French linking consonants; Bullock 2004 for discussion of
the role of PU in the history of French).
The situation characterized in (34) is obviously symmetrical,

A1 and A2 being interchangeable in the discussion.6 But if (34)
is parallel to (33) as argued, we may expect the situation in
(33) to then be symmetrical as well, and therefore that bare
stems should also be influenced by their derivatives, not just
the other way around. We consider this in (35).
Page 55 of 80
(35) Uniformity with one’s derivative (hypothetical)
Entailments I Entailments II Entailments III
X-A V⇒X V⇒X
X=V X ⇒ / ___ A where X= / ___ A
The tabulation in (35) is like the one in (34) except that,

instead of alternating between different affixal contexts, X
alternates between one affixal and one categorial context in
which it takes on the category V. For instance, for the verb
prevent, entailment (I) would read: ‘if something is a verb, it
has the sound structure [prIvent]’ For the adjective preventive,
entailment (II) would then read: ‘if something has the sound
structure [prIvent], it must occur followed by -ive’. The
combination of (I) and (II), and summation over instances of X
(prevent, generate, assert, etc.) yields (III): ‘if something is a
verb, it has the same sound structure as also occurs before -
ive’. This would predict that morphologically derived forms
should be able to influence their bases just as well as the
opposite. I have argued in PES (245), Burzio (1994b) that this
is true, the exceptional stress pattern of verbs, e.g. prevént,
being reducible in this manner to the regular one of their
affixed forms, like prevént-ing, regularly stressed on a heavy
penultimate. Similarly, PES (246) also takes the exceptional
stress pattern of -ic adjectives to reduce to the regular one of
their -ical counterparts, e.g. académic/académical. This ‘back-
copying’ does not seem to hold more generally, though (as also
confirmed by Kurylowicz’s (1949) ‘second law of analogy’,
which privileges simpler over more (p.96) complex forms).
For instance, the stress of párent is regular rather than being
like that of parént-al, despite the fact that the stress pattern of
*parént, while a bit rare, is attested for nouns: cemént, etc.
The factors that militate against this kind of backcopying can
in fact be traced to the entailments (II) of (35). With
unproductive affixes, including -ive, such entailments are often
false. For instance there is a verb presént (=X of (35)), but it is
not the case that it occurs in the context of -ive : *presént-ive.
We may presume that such falsifications tend to cancel the
force of the entailment (II) in (35), and hence the force of the
entailment (III) that depends on it. This is in contrast with the
entailment (II) of (33), which is not comparably falsified. That
is, it is much more generally true that if a sequence precedes -
ive, it occurs as a verb.7 Hence, at least with unproductive
Page 56 of 80
morphology, the SELECT effect and hence the pressure for

uniformity is asymmetrical, prevalently from base to
derivative.8 The case of affixed bases in (34) also presupposes
productivity for the same reasons, namely the need for
entailments (II) to be true with consistency—a criterion that is
met by the Spanish example. Note that with Level 1
morphology, even the entailments (II) of (33) incur occasional
falsifications, e.g. compUlS-ive, FURT-ive, INEVIT-able, where
the base is either non-existent or partially different. But that
was in fact the reason given for the relative weakness of PU
effects at Level 1 altogether, where virtually the only such
effects are the ‘metrical’ ones in (2) above, parasitic on
segmental identity as argued in the previous section.9
With productive morphology, however, uniformity of a base

with its derivatives will be expected. For instance, it is always
the case that if something is a verb it will occur with -ing.
Hence, entailments (II) in (35) are not falsified in that case,
and hence it remains plausible to suppose that bare verbs copy
the stress of their -ing counterparts. In addition to productivity,
it seems also likely that the distinction between inflection and
derivation may modulate the transitivity effect that yields
entailments (III). The reason relates to the identity of the two
Xs in each of (33)–(36). Such identity is—strictly speaking—not
true. For example, the ‘prevent’ of [prevent]v and that of
prevent-ive are not completely identical. The first is part of an
entailment structure (p.97) that includes the property of
being a verb, while the second is part of an entailment
structure that includes the presence of -ive and its associated
semantics.10 This misidentity, which is bound to attenuate the
force of the entailment (III) in each of (33)– (36) depends on
the magnitude of the contribution made by the affix, both
positive, in adding to meaning and sound structure, and
negative, in suppressing the former lexical category. Since the
affixes generally referred to as ‘inflectional’ are the ones that
contribute more modestly, leaving most of the properties of
their bases unaffected, most notably their argument structure,
one would expect stronger PU effects with inflection than with
derivation. In this respect, the -ic/-ical cases will classify with
inflection, since -al makes virtually no contribution and there is
a good degree of consistency with which -ic adjectives have -
ical counterparts. The plausibility that these cases also
constitute a true instance of PU thus remains as well.
Page 57 of 80
In sum, besides reflecting a general ‘Gradient Attraction’

effect, PU effects on stems are modulated by the selectional
properties of affixes. We have seen that a bare stem is
expected to affect its derivatives. The opposite effect, however,
is expected only for productive and—we presume—inflectional
morphology, the same restrictions holding for stems
alternating between different affixes.
Page 58 of 80
4.7.2 Uniformity of affixes

To the extent that all representations are taken to yield
entailments, the REH predicts no fundamental difference in
PU between stems and affixes. This seems correct, as I argue
below. In PES I contended that stems and affixes are often in
competition for uniform expression of their metrical
properties, yielding for instance the two different outcomes of
(36).
(36)
a. -ist: Stem consistency prevails a(mérica)(n-
ìstø); (pròpa)(gánd-is)tø
b. -ic: ffix consistency prevails ti(tán-icø); *(títan-
i)cø
In the PES analysis, the suffix of (36a) exploits the special

status of the final ‘weak’ syllable, which may or may not
remain extrametrical, and treats that syllable in accordance
with the stem stress. In (36b), however, the suffix does not
avail itself of the same opportunity, but rather manifests a
unique choice as regards the final syllable, thus forcing the
stem to re-stress (cf. títan). The choice made by -ic in parsing
the weak syllable is one that ensures the uniformity of -ic /-ical
paradigms as noted above, as in aca(dém-icø)/aca(dém-ica)lø,
but here there is no *?titán-ical to be consistent with and that
could thus not be the immediate reason. Rather, one must
conclude that in (p.98) (36b) -ic is simply consistent as an
affix, parsing in ti(tán-icø) just as it does in aca(dém-icø), and
winning the consistency competition with the stem, unlike -ist
of (36a). It is argued in PES that the basis for such competition
is essentially statistical. Those affixes that are not able to
accommodate the stem stress in all cases choose to satisfy
their own uniformity instead. This is the case with -ic. While
the stress of títan could in principle be accommodated as
indicated in (36), with items like al(lérg-icø)/ al(lérg-i)cø,
neither option for the final ‘weak’ syllable enables the stress to
duplicate that of (áller)gy (whose final syllable is also weak:
PES, sect. 3.6). This is in contrast with (áller) (g-ìs.tø), where -
istø has enough material to form its own separate foot, unlike
*(áller)(g-ì.cø) where the final foot headed by a light syllable is
ill formed. We can characterize this situation by postulating
that the uniformity of affixes prevails over that of stems in the
more general case as in (36b), but that matters reverse when
the uniformity of stems can be achieved consistently over all
stems, as in (36a). The reversal can be understood in terms of
Page 59 of 80
entailment summation. With affixes like -ist, the SELECT

entailment ‘IST ⇒ /N __’ can always be satisfied with respect to
stress (recall that a SELECT entailment is an abbreviation for
a cluster of entailments that pertain to specific pieces of the
representation), and it can thus undergo summation over the
whole class of stems, resulting in a stronger requirement of
metrical consistency for its stems: ‘IST ⇒ /N(stress) __’. In the
case of -ic, however, the class of cases like allérg-ic constitute
inevitable falsifications of the corresponding SELECT
entailment, resulting in an ultimately weaker requirement that
the stem be metrically consistent: ‘IC ⇒ /N(stress) __’. This
weaker requirement is evidently outranked by the consistency
of the affix itself. It would appear then that affixes are
endowed with greater inherent stability than stems, only
subject to the reversal just described. The vowel shortening
facts in (37) confirm this conclusion.
(37) Vowel shortening in affixes:
a. satir-i:ze, organ- i:ze a′.organ-iz-ation
b. gener-a:te b′. gener-at-ive
c. altern-a:te c′. altern-ateN
As argued above, vowel shortening is a rather general

phenomenon of Level 1 formations. It occurs for example in
the stem satir- of (37a) (cf sati:re). Yet affixes are generally
spared as shown in (37a, b, c), except under the conditions of
(37a′, b′, c′). We interpret these facts as follows. Each
occurrence of an affix is in a Gradient Attraction relation to
other occurrences of the same affix through the lexicon. Hence
the -i:ze of satir-i:ze in (37a) is attracted to that of organ-i:ze
and vice versa, the -a:te of (37b) is attracted to that of (37c)
and vice versa, etc. More accurately, each occurrence of an
affix is conditioned by the set of entailments collectively
generated by all other occurrences—just the consequence of
the REH. We then take lexical category to be a major
component of semantic representation, consistently with some
of the previous discussion. When an affix bears categorial
information, as in (37a, b, c), attraction is maximal because
their representation is maximally similar to that of their
attractor. The usual considerations of limited productivity of
Level 1 morphology enable us to iden- (p.99) tify as the main
attractor the long-vowel allomorph, rather than the short
Page 60 of 80
vowel ones of (37a′, b′, c′)—the same considerations that

enabled us to dismiss a Level 1 derivative as a significant
attractor to its base (discussion of (35)). In contrast, stems
such as satir-of (37a) are in a weaker relation to their
attractors precisely because they have lost their lexical
category. We know that the affix in (37a) acts as an enhancer
of the stem’s attraction, by way of its SELECT entailment, but
we also know that, with such Level 1 affixes, the enhancement
effect is relatively weak. Hence we take the stronger
attraction affecting the affixes in (37a, b, c) as sufficient to
prevail over the vowel shortening constraint, while the weaker
attraction of Level 1 stems will itself be prevailed upon. When
affixes themselves lose their canonical lexical category in
Level 1 processes, however, their behavior will be expected to
change to the more general one of Level 1 stems, undergoing
shortening, indeed as in (37a′, b′, c′). The case in (c′) shows
that it is not further affixation per se that is necessary, but
rather just the switch away from the category of the main
attractor.
The above interpretation is further confirmed by the Italian

facts in (38), which closely parallel the English ones in (37).
(38) Italian (Burzio 1998)
a. v énd-ere ‘sell-INF.’ a′. v ínc-ere ‘win-INF.’
b. vend- út-o ‘sold- b′. v ín-t-o ‘won-PART.’

PART.’
c. énd-ut-o/-it-o c′. ínc-ut-o/-it-o

*v *v
dv ènd-it-óre‘seller’ d′ v ìnc-it-óre‘winner’
d. v ènd-it-a ‘sale’ e′ v x00EC;nc-it-a ‘win-

NOM.’
The participle in (b) shows that the participial suffix -út-

imposes its own stress, overriding the metrical consistency of
the stem vénd- of the infinitive in (a) (adjacent stresses being
excluded). The final -o/-a in (38) are merely inflectional
endings that will not affect the discussion. De-stressing of the
participial suffix, as in (c) or similarly (c′), appears not to be an
option, and this is the general predominance of affixes noted
above as in (36b). The case in (38b′) shows further, however,
that the Metrical Consistency/PU of the stem can be attained
Page 61 of 80
here by resorting to a syncopated form of the participial affix,

which, lacking vowels, will thus lack stress. The syncopated
affix, which is either -t- as in (38b′), or -s- as in pre-s-o ‘taken’
depending on the verb, is obviously disuniform from its
canonical counterpart -út-. In Burzio (1998) I suggested that
the metrical consistency for the affix dominates its segmental
consistency, syncope violating only the latter, while the forms
in (c, c′) would violate the former. In the present context, we
can reinterpret this account in terms of Gradient Attraction,
however. Segmentally faithful -ut- is excluded when unstressed
because too similar to the main attractor -út-, with which it
would thus neutralize. This is the outcome in (38b). The
participial allomorphs -t- and -s- are then to be thought of as
suppletive forms—a type of morphological irregularity similar
to that of (22a) above compulsive. As such, they are due to
some ‘input’ provided to the surface form of the participle
directly. This accounts for the unpredictability of the forms
themselves, in particular the -s- allomorph, as well as the
general unpredictability of which verbs utilize which (p.100)
form. As with other irregularities, these are shared by multiple
items, however, resulting in certain level of entailment
satisfaction across each class. The segmental divergence
between primary allomorph -út- and suppletives -t- and -s- can
now be taken to weaken the GA effect sufficiently to permit
the latter’s existence, which will thus result in satisfaction of
the stem’s metrical consistency as in (38b′). The oscillation of
(38b, b′) can be viewed as a roughly even tension between the
attraction by main allomorph -út- (weakened for the
suppletives), which prevails in (38b), and the stem’s metrical
consistency, which prevails in (38b′). While this account is
unable to predict which verbs within this conjugation will have
a syncopated participle and which will not (as this seems truly
unpredictable (Burzio 1998)), it does correctly predict that
only verbs of this conjugation will have syncopated participles.
The reason is that other conjugations have affix-stressed
infinitives, and hence stems with unstressed final syllables,
e.g. sap-ére ‘know’. Since the claim is that participial syncope
(a violation of affixal PU) is needed to maintain the stress of
the stem-final syllable (stem PU), no syncope will be expected
when there is no such stress to maintain, correctly, as there is
no case like sap-ére/ *sát-t-o, ‘know/known’ but only sap-út-o
‘known’ (see also Burzio 2003).
Page 62 of 80
Turning to the nominalizations in (38d, d′, e, e′), they mirror

the English cases in (37a′, b′, c′): when an affix is no longer
associated with the same lexical category as its attractor, it is
free to undergo further changes (weaker attraction). Hence
the unstressed and yet un-syncopated participial allomorph -it-
which was excluded in (b, b′) now becomes possible and is
utilized to satisfy other types of stress uniformity—both that of
the stem, and that of the nominal suffix -óre—avoiding a stress
clash between them by means of the intervening syllable. An
interesting further twist lies in the fact that the unstressed
allomorph is in fact -it-, rather than the expected -ut-. The
latter is borrowed from another conjugation, as shown by
(39).11
(39)
(39) I. Conjugation/ II. Participle III.

infinitive (canonical) Departicipial
Nominal
a- ere: v énd-ere -út-: vend-út- -it-: v ènd-it-óre

‘sell’ o
b. - ére: conten- -út-: cònten- it-: cònten-it-

ére‘contain’ út-o- óre
c. - íre: sped- -ít-: sped-íto -it-: spèd-it-óre

íre‘send’
What (39) shows is that the -út-/ít- contrast of the participles

disappears when the stress is removed, as in the nominals. We
can deal with this by thinking of the distance between -ut- and
-it- as being perceptually enhanced by the presence of stress—
the same assumption needed to account for the structure of
vowel inventories under the Dispersion Theory of s. 4.3 above.
When stress is removed, distance is reduced, promoting
neutralization under GA. Hence this is another OO-F effect.
The choice of uniform -it- over uniform -ut- may perhaps be due
to the less marked status of i com- (p.101) pared with u. Note
that alongside of the syncretism of column (39.III), there is a
partial one in column (II) as well. By analogy with the
infinitive, the participle in (39a) should be unstressed rather
than stressed -út-. We may thus assume as already suggested
that the segmental identity of the stressed and unstressed
variants will bring about their neutralization to -út- (yet
another OO-F effect). The asymmetry between participles and
Page 63 of 80
their infinitives in (39a, b), which do not undergo a

comparable neutralization, is likely to reduce to a
generalization to be reviewed in the next section, to the effect
that the more ‘basic’ categories are able to maintain the most
distinctions. The facts in (39) also shed light on the fact that in
the competition among -út-, -t-, and -s- for the expression of
participial morphology in the conjugation of (39a) (shown in
(38b, b′)), it is -út- that acts as the main competitor or
attractor. The reason for this is that -út- is more regularly
instantiated through the lexicon, being present not only in the
conjugation of (39a), but in that of (39b) as well..
In sum, English derivational affixes exhibit PU effects in their

resistance to the otherwise general phenomenon of V
shortening. This is paralleled by the fact that in Italian affixes
do not undergo stress allomorphy. The superior stability of
affixes appears tied to their role as exponents of the word’s
lexical category. When such role is revoked, affixes undergo
allomorphy just like stems—a typical clustering effect in the
present analysis. Such PU properties of affixes cannot be
expressed by more traditional means. Without surface-to-
surface relations, affixes are only expected to be uniform by
virtue of a common UR, an expectation that is falsified by the
critical role played by surface properties. Thus, the difference
between -ic and -ist in (36) would require associating -ic but
not -ist with a fixed metrical structure in UR. However, this
would miss the fact that the different behaviors are ultimately
predictable from the ability of -ist, but not of -ic, to
accommodate the stem stress over the whole class of stems, a
global property over surface structures. Correspondingly, the
disuniformity of the paradigm -út-/-t-/-s- in the expression of
the participial affix in the -ere conjugation would have to result
from different URs, given both the segmental unpredictability
of the -s- allomorph and the unpredictable distribution of
allomorphs within that conjugation. But this would miss the
fact that such allomorphy is none the less motivated by the
metrical consistency of the stem—again a surface property.
Finally, the uniform -it- in (39.III) would require either
underlying -it- rather than -ut- in (39a), or a rule turning -ut- to
-it- in unstressed position. The first account would miss the
fact that the -ut- to -it- switch is controlled by surface
properties—absence of stress—while the second would miss
the fact that the resulting -it- exists independently in another
conjugation (a natural attractor in the present account). The
Page 64 of 80
stem metrical consistency in the -ere conjugation, which drives

the participial syncope, is itself also inexpressible
derivationally. In order to transfer the stress from the infinitive
to the participle, a derivation would rather implausibly have
first to form the infinitive and calculate its metrical structure,
then remove the infinitival affix (the one responsible for the
stem stress) and finally form the participle from the same—
now stressed—stem (see Burzio 2003) for more detailed
discussion). The present conclusion is thus that the patterns of
allo-morphy reviewed above do not reflect the contingencies of
derivations, but rather the (p.102) general character of the
lexicon as being ‘conservative’ in the sense of Steriade (1997a,
1999a), namely in making maximal use of independently
existing material, a property that we have reduced to the REH
and Gradient Attraction. When the forces of allo-morphy,
either phonology or idiosyncrasy (= IO-F) move a
representation away from its main attractor, other neighboring
attractors stand at the ready.
4.7.3 Morphological leveling

Another well-known instance of PU is provided by
morphological syncretism, sometimes also referred to as
‘morphological leveling’ (Hock 1986: s. 9.1.4.). The Old
English (OE) paradigms in (40) provide one illustration.
(40) Old English weak conjugation: trymman

‘strengthen’
Page 65 of 80
Present Indic. Subjunct. Past Indic. Subjunct.
1sg trymme trymede
2sg trymest trymme trymedest trymede
3sg trymeð trymede
1pl
2pl trymmað trymmen trymedon trymeden
2pl
Page 66 of 80
In OE, the inflectional paradigm of the subjunctive was

considerably more leveled than that of the indicative. That of
the past indicative was also more leveled than that of the
present (the past first and second person singular have
identical inflections), and in the indicative the plural was more
leveled than the singular. The same three asymmetries are in
fact also detectable in Modern English, as the -s of the third
person singular indicative has no counterpart either in the
past, the subjunctive, or the plural. These tendencies as well
as the further ones in (41) appear to hold cross-linguistically,
pointing to the generalization in (42). The difference noted for
Italian (39.I, II) above suggests that ‘infinitive’ is also a less
marked category than ‘participle’.
The framework of Distributed Morphology (DM: Halle and

Marantz 1993, 1994; Noyer 1992; Embick and Noyer 2001;
Bonet 1991) provides a way to capture such effects by means
of special rules of ‘Impoverishment’. For example, the case of
contemporary English subjunctive, It is necessary that he
walk, could be handled in the manner of (43).
In both examples in (43), the inflection node in the syntactic

structure would initially bear the feature [3sg], reflecting its
agreement with the subject. In the indicative case, vocabulary
insertion would then apply directly, choosing vocabulary item -
s over (p.103) -ø (null), because it best matches the features
of its insertion node. In the subjunctive, however, DM would
postulate the Impoverishment rule shown, applying prior to
vocabulary insertion. With the featural content of the insertion
node thus ‘impoverished’, vocabulary insertion will now find
the null affix as the best-fitting candidate.
(41)
Page 67 of 80
Sample More Indicative Active Present Singular Masculine Nominative

languages contrasts
Fewer Subjunctive
contrasts
English, OE,
Romance
Passive
Gothic
English, OE Past
English verbs, Feminine

OE adjectives
Russian
adjectives,
OE nouns,
adjectives
Obliques
OE pronouns,
demonstrativ
es, nouns,
adjectives
Page 68 of 80
(42) Syncretism Generalization: More marked

categories exhibit fewer internal contrasts (more
syncretism/leveling) than less marked ones.
(43)
Example Vocabulary Impoverishment Voc.

rule Insertion
He walk- [3sg] [3sg] x21D4; [-s]

s x21D4; [-s] x00F8; /governed
by
x21D4; -x00F8; Subjunct.
…that he x00F8;
walk-
x00F8;
While descriptively adequate, impoverishment rules provide

little explanation since they add complexity to the formalism
for no reason that the formalism itself makes evident. In
contrast, morphological syncretism is directly expected under
the REH, modulo the assumption in (44).
(44) The unmarked categories—Indicative, Present,

Singular, Masculine, Nominative—are unspecified.
According to (44), for example, ‘Indicative’ would not consist

of a specific value for the feature ‘Mood’, but rather of the
absence of such a feature altogether. This assumption (p.104)
seems natural and has been independently advanced by work
such as Jensen (1990) and Noyer (1993). Of particular
relevance is the fact that it would also be required by the DM
approach. If unmarked categories are unspecified, then
impoverishment rules simply cannot refer to them. Hence, for
instance, the rule in (43) could not refer to the indicative,
incorrectly yielding leveling in the indicative rather than the
subjunctive, because ‘indicative’ does not exist—the mere
absence of a feature.12
Returning to the REH, if (44) is correct, then marked

categories always add a dimension or component to the
representation. This will have the effect of enhancing the
leveling pressure among the members of the paradigm, in the
way illustrated in (45).
Page 69 of 80
(45) Morphological leveling
Page 70 of 80
Indicative Subjunctive
a. walk, basic sound A A A A

and meaning
B B B B
b. Added component: C C
[Subjunctive]
c. Alternating S S S S
component:
[3sg] ⇔ [-s] / other ⇔- ø
d Entailments A⇒S A ⇒¬S A⇒S A⇒¬S
B⇒S B ⇒ ¬S B⇒S B ⇒¬S
C⇒S C ⇒¬S
e. Leveling pressure 2 3
(contradictory
entailments)
Page 71 of 80
In (45), we make again the simplifying assumption that the

basic sound and meaning of a verb like walk can be given a
simple two-component representation A, B, as in (a). Then we
take the two affixes -s and -ø to instantiate entailments from
sound to meaning and vice versa as in (45c), consistently with
both the DM view in (43) and the REH. In the table, we refer
to these two complementary choices of affixes simply as S (for
affix -s) and S. Unlike in DM, the entailments provided by the
REH will not only run internally to each morpheme, however,
but across different morphemes as well. For instance, in the
representation walks, the subpart walk entails-s, as would be
expressed by the two entailments A ⇒ S, B ⇒ S. But the
representation walk, as in ‘I walk’ will generate contradictory
entailments from walk to absence of -s, namely: A ⇒ S, B ⇒ S.
Hence the Indicative alternation is disharmonic as indicated in
(45d, e) by way of two violated entailments, which thus
measure the extant pressure towards (p.105) morphological
leveling. Recall that in the present approach all morphological
alternations are disharmonic (as in (23.I) above, for example)
and hence occur only as compelled. Specifically, failing to
combine walk with -s as in ‘He walks’ would violate the
entailment ‘[3sg] ⇒ [-s]’of (45c), which is evidently able to
stem the leveling pressure over the indicative paradigm. But
when we move to the subjunctive, the leveling pressure will be
greater due to the added component—the abstract semantic
feature [subjunctive]: C of (45b). The new pressure consists of
three entailments, as indicated in (45d, e), and this is evidently
able to turn things around, now overriding the entailment
‘[3sg] ⇒ [-s]’of (45c) and resulting in the fully uniform
paradigm of the subjunctive. This discussion has so far not
provided a way to choose between -s and -ø as the leveling
outcome. We presume that the choice of -ø is due to the fact
that the latter is independently predominant in the paradigm,
so that extending it rather than -s constitutes a more minimal
change.
In sum, while it would be hard to predict exactly when

syncretism will occur, we are correctly predicting the general
circumstances under which it may occur: those of (41), (42).
As with other types of PU, more traditional resources shed
little light on the phenomenon of syncretism, by requiring
additional machinery to express it. In contrast, on the present
approach, the latter is reducible to the general neutralization
effect predicted by the REH: GA. The reason that more marked
Page 72 of 80
morpho-semantic categories are more highly syncretic is that

they correspond to representations that are independently
more similar by sharing more features. As usual, similarity
promotes further similarity, hence the leveling. Obviously,
syncretism is yet another type of PU that the notion of cyclicity
has no way of capturing. The fact that the subjunctive
inflection of ‘ … that he walk-ø ’ is the same as that ‘of … that I
walk-ø’ is plainly not because they are related by a cyclic
derivation.
Page 73 of 80
4.8 CONCLUSION
Paradigm Uniformity can be dealt with in terms of the OT
notion of Output–Output faithfulness: members of a paradigm
are required to be faithful to other members of the same
paradigm. Yet this view raises a number of fundamental
questions, listed in (3) above, which the present chapter has
sought to answer.
One question (3a) was what criteria identify the sources of

OO-F for any form under calculation. If being members of the
same paradigm is at least a necessary condition, then the
question is what defines the notion of paradigm. Here, I have
argued first that the problem of allomorphy demands a theory
able to express the notion of attraction over distance, a
respect in which this problem intersects significantly with
problems in both phonology proper and perception, suggesting
a single overarching property of mental representations. I
have then argued that such a theory needs to incorporate the
Representational Entailments Hypothesis, under which
representations become analyzable as sets of mutual
entailments or dependencies among their components, (p.
106) similarly to the distributed representations of neural
nets. In part, OO-F would then consist of attraction by
neighboring representations, and ‘paradigms’ would be
definable as neighborhoods. I have argued further, however,
that regular morphological relations modulate the attraction
effects substantially. For instance, the attraction of parént by
paréntal is insignificant, the underlying reason being that the
relation of nouns to -al adjectives is not sufficiently regular or
general, witness *studental, while attraction in the opposite
direction is significant, as the relation of -al adjectives to
nouns is fairly regular. Even this regularity is moderate for
Level 1 formations, however: arboreal /*arbore, in contrast to
the much greater regularity of Level 2 formations. In further
contrast, in cases of fully productive morphology, such as
prevént/prevénting or the Spanish paradigm in (1), pairwise
relations are regular in both directions, and attraction is thus
expected—at least in principle—to be symmetrical. From this
point of view, the notion of ‘paradigm’ would then also require
reference to the degree of regularity of morphological
relations. It is ill-suited to express the asymmetries, however
(e.g. the one of parént/paréntal, since there is no sense in
which one item could form a paradigm with the other but not
vice versa) and hence the term ‘paradigm’ remains of limited
Page 74 of 80
theoretical relevance. In the present analysis, the descriptive

notion of PU is thus replaced by the OT notion of OO-F, in turn
replaced by the combination of GA entailments, expressing
attraction over distance, together with SELECT entailments,
expressing the role of morphological relations.
Another question (3b) concerned the relation between

morphology and OO-F and the potential redundancy between
them in generating surface-similarity patterns. In answer to
this question, I have argued that morphology is OO-F, in the
form of the SELECT entailments, and hence that there is no
redundancy. For instance the entailment ‘IST ⇒ /N __’
simultaneously validates both the segmental structure of
améri-canìst (2a′) above, formerly the result of word-formation
rules of the morphology, and at the same time its accentual
relation to américan, a type of OO-F. Even aside from their
redundancy with OO-F, traditional approaches to morphology
have been argued to be inadequate because of their reliance
on inviolable formalisms, a characteristic that makes it
impossible to integrate the morphology with the phonology, in
particular to express the inverse distributional relation of
morphological and phonological regularities of (22) above.
Finally, a third question (3c) was whether there was any

residual validity to the traditional notion of the cycle, the
answer to which was negative. Cyclicity would at best be
applicable to only a small portion of the PU phenomena
reviewed: those where an affixed derivative is uniform with an
unaffixed base, like (2a′) and possibly (2b′). It is inapplicable to
the cases where there is no affixation, like the French liaison
cases in (5); it is also inapplicable to cases where the base has
its own affix, such as the Spanish cases in (1) and the Italian
ones in (38a′, b′) (i.e. to the general schema in (34)); it is
inapplicable as well to the uniformity of affixes of s. 4.7.2, and
to the morphological leveling/syncretism cases of s. 4.7.3.
The present discussion has thus shown that no legitimate

derivational basis remains to either phonology or morphology.
Notes:
(1) To the limited extent that nouns also form inflectional
paradigms, these are also uniform, as Harris had noted: señór/
señóres ‘gentleman/gentlemen’, etc.
Page 75 of 80
(2) It also seems less than an accident from this point of view
that the only sequentially discontinuous morpheme in English
is constituted of segmentally identical parts as in en-light-en,
etc.
(3) The binding together of elements under spatio-temporal

proximity thus predicted by their entailment structure is also
likely to cover effects formerly addressed by the ‘Feature
Geometry’ program (Clements 1985; Sagey 1986). While
features tend to cluster in assimilatory processes for the
reasons discussed earlier, features sharing certain subspaces,
like the features [±voice], [±spread glottis], both pertaining to
the same articulator: the glottis, appear to cluster more
systematically. This would follow from a representation along
the lines of (i) for an aspirated [t].
((i))
Page 76 of 80
Featural Analysis, Features [−voice] [+spread gl.] [−anterior] [−contin.]

[th]
Feature Class Glottal Glottal Coronal
Page 77 of 80
The shared feature class [Glottal] predicts stronger binding of

the elements involved just as in the previous cases of shared
elements, thus accounting for such complex assimilations as /b
th/ → [phth] (Ancient Greek, Kenstowicz 1994: 156f. and refs.).
Note as well that, because entailments are violable, we predict
that clustering by feature class should be violable, just as
independently argued in Padgett (2002), and contrary to the
rigid clustering predicted by feature trees.
(4) Note that there is sense of ‘regularity’ in which Level 2

items may seem the more regular, contrary to the text claim.
That sense is conveyed by the observation that the
phonological processes that do apply to the Level 2 lexicon,
like the allophonic one just noted are exceptionless and/or
‘automatic’, in contrast to Level 1 processes like shortening,
which have ‘lexical’ exceptions, like obe:sity. I will return to
this fact below.
(5) With regard to the issue of lexical exceptions of n. 4, we

are correctly expecting their exclusion with allophonic
processes applying to Level 2 items. On the IO-F account, such
exceptions will be banned by the fact that IO-F is totally
dominated in the allophonic schema (31). On the alternative
OO-F account, lexical variation will be excluded by the
uniformly high semantic correspondence between Level 2
items and their bases.
(6) This abstracts away from the possibility that either affixed
form in (34) may be more prominent than the other perhaps
for semantic reasons, or for reasons related to frequency of
use (Bybee 1995).
(7) This backcopying would be different from the phenomenon

known as back-formation, like the formation of construct from
construction in the history of English. The latter leads to
satisfaction of the conditional: ‘if there is a form X-ion, then X
exists as a verb’. This is essentially the SELECT entailment for
-ion, and an OO-F effect. In the case of backcopying, on the
other hand, the conditional is in the opposite direction: ‘if
there is a bare verb, it must have the same form (stress) as
when it appears affixed with -ing’. On the one hand, the text
analysis can treat back-formation simply as a case of OO-F. On
the other, it is designed freely to permit bound stems (i.e.
absence of back-formation) with Level 1 affixes. Further
Page 78 of 80
refinements would therefore be required to deal with back-

formation.
(8) This account of the asymmetry supersedes the one

attempted in Burzio (2002b), based on the notion that a bare
base always violates some of the entailments produced by the
derivative, like ‘parent → /___al’ due to parental but violated by
parent. This consideration was invoked to argue that a bare
base will always be subject to a weaker attraction by a
derivative, but it is challenged by the fact that, at least with
category-changing morphology, there is a comparable
entailment violation by the derivative; e.g. the entailment
‘parent → N’, due to the base is violated by the derivative
parental, as noted in the text just below.
(9) Segmental PU effects are rare with Level 1 morphology.

They include levy/levia-ble of (4) above, where they are
parasitic on stress identity as argued. Another rare case is
apha:sia (regular CiV-lengthen-ing)/apha:s-ic.
(10) It would then be technically incorrect to say that -ive

‘subcategorizes’ for a verb, since its stem is not an actual one.
One could argue that in prevent-ive, the stem is still a verb, the
adjective meaning roughly: ‘able to prevent’. While obviously
some of the verbal meaning is still present, the text reasoning
is that major properties of the verb are lost in the adjective,
which is true at least for the argument structure.
(11) Beside the three conjugations in (39), Italian has the

already mentioned -áre conjugation, which is totally regular
and hence also immune from the syncretisms of (39): port-áre/
port-át-o/pòrt-at-óre ‘carry/ carried-PART./carrier’.
(12) The unspecification of (44) is not to be confused with the

underspecification approach to syncretism (see especially
Wunderlich 1995). In the latter approach, for example the
English past tense morpheme -ed would lack specifications for
person or number (a kind of underlying impoverishment). The
difference is clear in that in (44) it is the non-syncretic
categories (e.g. present) that are unspecified. The general
effect that needs to be captured is that there appears to be a
cap on featural specifications: if a form is specified for tense
or mood it is less likely to be also specified for person. One can
well imagine the underspecification approach postulating such
Page 79 of 80
a condition. The present approach is the only one, however, in

which the cap follows from the inherent properties of the
system, as shown below.
Page 80 of 80
Capitalistic v. Militaristic: The Paradigm Uniformity Effect Reconsidered


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Capitalistic v. Militaristic: The

Paradigm Uniformity Effect
Reconsidered
Stuart Davis
DOI:10.1093/acprof:oso/9780199267712.003.0005

This chapter argues for a different view than Steriade's of the
paradigm uniformity effect with respect to the distinction
between capitalistic v. militaristic. It contends that the
nonflapping of the /t/ in the third syllable of mìlitarìstic reflects
a general pattern in American English and need not be a
paradigm uniformity effect. This is evidenced by the metrically
comparable monomorphemic words such as Mèditerránean
and Návratilóva both of which also have a nonflapped /t/ at the
beginning of the third syllable. Rather, what needs to be
explained is the flapped /t/ in the third syllable of càpitalístic.
The chapter argues that the flap can be explained by paradigm
uniformity with cápital. However, the uniformity effect involves
uniformity of foot structure with cápital. Consequently, the
paradigm uniformity effect with cápital does not specifically
involve a noncontrastive phonetic feature as Steriade has
argued.
Keywords: paradigm uniformity, capitalistic, militaristic, flap, Steriade
Page 1 of 23
5.1 INTRODUCTION
In an influential paper, Steriade (2000) argues that
noncontrastive details are relevant for the evaluation of
paradigm uniformity, thus showing the role of noncontrastive
phonetic properties on phonological patterning. As Steriade
(2000: 314) says, her argument is part of ‘a larger agenda’
that claims that ‘the distinction between phonetics and
phonological features is not conducive to progress and cannot
be coherently enforced’. This is reflective of a functional
approach to phonology developed in the works of the UCLA
group (Flemming 1995; Jun 1995; Kirchner 1997; Silverman
1997; Steriade 1999b; Zhang 2000; etc.) which maintains that
phonology is reducible to general properties of human motor
behavior and perception, and this entails explic- (p.108) itly
incorporating low-level phonetic machinery into the
synchronic phonological grammar. Such a position can be
contrasted with the one espoused by Ladefoged (1983) who
specifically emphasizes the importance of abstract units in
phonology such as the phoneme commenting that ‘… some of
the really interesting linguistic phenomena are precisely those
for which there are no general phonetic explanation’ (ibid. 8).
Earlier versions of this chapter were presented as papers

at the seventh Midcontinental Workshop in Phonology at
the University of Iowa in October 2001 and at the second
North American Phonology Conference at Concordia
University in Montreal in April 2002 and at the
University of Potsdam in October 2003. I thank the
participants at those meetings for useful discussions of
the papers. I also wish to thank the editors of this volume
for their helpful comments.
Steriade (2000) draws on English and French data in arguing

for the role of non-contrastive phonetic properties on
phonological patterning. Here, I focus on Steriade’s main
example from English for the role of noncontrastive phonetic
properties on phonological patterning. (See Raffelsiefen (this
volume) for further critical discussion of Steriade 2000.) This
example comes from the difference in American English
between the flapped /t/ at the beginning of the third syllable of
càpitalístic v. the unflapped /t/ at the beginning of the third
syllable of mìlitarístic. The judgement is robust. As originally
pointed out by Withgott (1982), the /t/ seems obligatorily
flapped in càpitalístic whereas in mìlitarístic the /t/ not only
seems unflapped, but aspirated as well. Given that the
Page 2 of 23
metrical patterns of these two words appear the same (i.e. an

initial dactylic sequence followed by a syllable with primary
stress), why would there be a difference in the nature of the /t/
at the beginning of the third syllable of these two words? The
nonflapped /t/ of ‘militaristic’ is the more problematic case
under the view held by Steriade (and going back to Kahn
(1976)) that the /t/ should be flapped before a stressless vowel
(as in wríting or cápital). Steriade’s (2000: 323) explanation for
the unflapped /t/ in mìlitarístic revolves around paradigm
uniformity with respect to durational equivalence: ‘The
principle at work is paradigm uniformity: the paradigm of
{military, militar (-istic)} becomes less variable phonologically
if the stop t [as opposed to the flap t] is generalized to the
stressless syllable … nonflapping maintains to a greater extent
the durational equivalence [of the third syllable in mìlitarístic
with the third syllable of mílitàry]’. Durational equivalence in
this context is specifically argued by Steriade to be a
(noncontrastive) phonetic property. That is, in the example of
mìlitarístic, the paradigm uniformity effect aims for the /t/ of
the third syllable to have the same (closure) duration as the /t/
in mílitàry. Duration is defined by Steriade as duration in
precise milliseconds and so would not be considered a
standard phonological feature. Consequently, according to
Steriade, the paradigm uniformity effect of mìlitarístic with
mílitàry involves a noncontrastive phonetic property.
In this chapter I argue for a different view than Steriade’s of

the paradigm uniformity effect with respect to the distinction
between capitalistic v. militaristic. I will contend that the
nonflapping of the /t/ in the third syllable of mìlitarístic reflects
a general pattern in American English and need not be a
paradigm uniformity effect. This is evidenced by the metrically
comparable monomorphemic words such as Mèditerránean
and Nàvratilóva both of which also have a nonflapped /t/ at the
beginning of the third syllable. Rather, what needs to be
explained is the flapped /t/ in the third syllable of càpitalístic.
Here I maintain that the flap can be explained by paradigm
uniformity with cápital. However, I argue that the uniformity
effect involves uniformity of foot structure with cápital.
Consequently, the paradigm uniformity effect with cápital does
not specifically involve a noncontrastive phonetic feature as
Steriade has argued.
Page 3 of 23
(p.109) 5.2 THE THIRD SYLLABLE OF NONFINAL

DACTYLIC SEQUENCES
Steriade (2000) maintains that a paradigm uniformity effect
accounts for the non-flapped (and aspirated) nature of the /t/
closure of the third syllable of the word mìlitarístic. According
to her, the /t/ is not flapped in order to approximate more
closely the closure duration of the corresponding nonflapped /
t/ in mílitàry. Steriade assumes that a paradigm uniformity
effect is at issue with respect to the first /t/ closure in
mìlitarístic since she is assuming that without paradigm
uniformity the /t/ should be flapped with a very short closure
duration. It is important to keep in mind that paradigm
uniformity can be at issue only when the occurrence of an
allomorph in a paradigm is phonologically unexpected or
unjustified. Thus, Steriade assumes that the unflapped /t/ in
mìlitarístic is phonologically unexpected. Consequently, if we
are to be convinced that mìlitarístic is a case of paradigm
uniformity in ‘durational equivalence’ with the nonflapped /t/
of mílitàry, then non-derived words of the same stress pattern
should have the /t/ flapped in this environment. In this section,
however, I will argue that the unflapped /t/ in the third syllable
of mìlitarístic reflects a general pattern in American English
and thus does not reflect a paradigm uniformity effect.
To see why a paradigm uniformity effect may not be at issue

with respect to the unflapped (and aspirated) /t/ in the third
syllable of mìlitarístic, consider the mono-morphemic words
with an initial dactylic sequence preceding the primary stress
given in (1) and where the transcribed stop is underlined.
(1) Monomorphemic words with a dactylic preceding

the primary stress
a. [t Mèditerránean Nàvratilóva
h
]
b. [ph] lollapalooza pèripatétic Wìnnepesáukee

Wìnnepegósis
c. [kh] abracadabra Nèbuchadnézzar
The data in (1) have not been typically considered in earlier

discussions on English stop allophony. They are discussed
neither by Kahn (1976) nor by Selkirk (1982). On the other
hand, both Jensen (2000: 209) and Pater (2000: 270) have
recently discussed the pattern in (1a–c) and specifically refer
Page 4 of 23
to these voiceless stops as being not only unflapped in the

case of (1a), but as being aspirated. Such data suggest that
the unflapped (and aspirated) nature of the /t/ of the third
syllable of mìlitarístic is phonologically expected and thus
paradigm uniformity would not be at issue.
While Steriade (2000) does not seem to be aware of the range

of data in (1), in a footnote to her article (p. 334) she indeed
observes (based on Withgott 1982) that the non-derived word
Mèditerránean with an unflapped /t/ poses a problem for her
analysis of the paradigm uniformity effect of mìlitarístic with
mílitàry. In order to preserve the paradigm uniformity analysis
of mìlitarístic Steriade (2000) maintains that the word
Mèditerránean is simply an exception. She says (p. 334) ‘I
attribute the unflapped [t] in this unique form [Mediterranean]
to the orthographic gemin- (p.110) ate “rr”, which is
interpreted by speakers as an indication of secondary stress
on the preceding vowel.’ However, Mèditerránean does not
seem to be an exception as evidenced by the range of data in
(1). Moreover, it is odd to claim that the vowel of the third
syllable of Mèditerránean is interpreted as having a secondary
stress since it is clearly reduced.
Steriade views Mèditerránean as an exception since she

follows Kahn (1976: 57) as having /t/ being flapped in the
environment before a stressless vowel when the preceding
vowel is also stressless. However, Kahn only considered words
like cápital and obésity where the /t/ is flapped in the last
syllable of the word. He did not consider words like those in
(1) where an initial dactylic sequence precedes the syllable
with primary stress. The two cases are not identical. The
footing possibilities for the third syllable of the dactylic
sequence in words like in (1) are different from the third
syllable in a dactylic sequence in words like cápital and obésity
where that syllable is word-final and so can be incorporated
only into the preceding foot. In other words, the /t/ at the
beginning of the third syllable of the dactylic sequences in
cápital and obésity is in a foot-internal environment, but the /t/
at the beginning of the third syllable of a word like
Mèditerránean need not be foot-internal.
Furthermore, preliminary phonetic work reported in van Dam

and Weaver (2001) shows a clear difference in aspiration
between the bolded voiceless stops in the words in (2) and
those in (3). (We cite this preliminary study because we know
Page 5 of 23
of no other acoustic work that examines the stops in words

like those in (1). Pater (2000) and Jensen (2000) who cite
words like in (1) as having an aspirated stop at the beginning
of the third syllable do not actually cite any acoustic study in
support, but base it on impressionistic data. Steriade (2000),
too, does not cite any independent phonetic study for her view
that those same stops would not be aspirated.)
(2) Wìnnepegósis, Mèditerránean, Nèbuchadnézzar

(3) múppets, móccasins, bútter
The stops in (3) are unaspirated, having short voice onset

times (VOT) in van Dam and Weaver’s study (less than 20
milliseconds on average for the noncoronal stops; the coronal
flap is briefer). On the other hand, the stops in (2) were shown
by van Dam and Weaver to have an average VOT of over 50
milliseconds which made them almost as aspirated as the
bolded stops in (4) which occur immediately before a stressed
vowel, an environment well known to favor aspiration (e.g.
Selkirk 1982).
(4) appéals, attáck, Chicágo
Further, van Dam and Weaver report that the bolded stops in
(2) are also more aspirated than the bolded stops in (5) where
the stop is at the beginning of the last syllable of a word-final
dactyl.
(5) Oédipus, Américan, Samáritan
Essentially, the preliminary study by van Dam and Weaver

supports the impressionistic data from Pater (2000) and
Jensen (2000) in (1) and also points to a difference between a
voiceless stop at the beginning of the last syllable of a word-
final dactyl as in (p.111) (5) and one at the beginning of a
nonfinal dactyl as in (2). While the preliminary studyneeds to
be replicated, it does suggest that Pater (2000) and Jensen
(2000) are correctin considering the beginning of the third
syllable of a (non-final) dactylic sequenceas an environment
for aspiration. Consequently, the unflapped (and aspirated) /t/
ofmìlitarístic reflects the expected regular pattern in (1) and
need not be a paradigm uniformity effect with mílitàry.
An important issue that arises from the aspirated nature of the

stops in (1) and (2) is why those stops would be aspirated at
all. The beginning of a non-initial stressless syllable, as in the
Page 6 of 23
third syllable of Mèditerránean, is not one of the environments

mentioned in the traditional generative works discussing
English aspiration (e.g. Kahn 1976; Selkirk 1982). Aspiration
normally occurs with a voiceless stop, either at the beginning
of a word or at the beginning of a stressed syllable (or foot-
initially as in Kiparsky 1979; Nespor and Vogel 1986; Iverson
and Salmons 1995). Consider the typical cases of aspiration
given in (6).
(6) Environments of aspirated stops in American

English
Page 7 of 23
Pony [ph] Pacífic [ph] appear [ph] Chésapèake [ph]
terrible [th] tomáto [th] atomic [th] Atàscadéro [th]
candy [kh] canáry [kh] acúte [kh] cucumber [kh]
Page 8 of 23
The first two columns in (6) show that voiceless stops are
aspirated word-initially whether at the beginning of a stressed
syllable or a stressless one. The last two columns in (6) show
that a voiceless stop is aspirated at the beginning of a non-
initial stressed syllable, be it a syllable with primary stress or
secondary stress. The voiceless stop which begins the third
syllable of a word like Mèditerránean in (1) does not seem to
match any of the environments in (6), since it is neither word-
initial nor at the onset of a stressed syllable. In fact, it appears
to be at the onset of a syllable with a reduced vowel.
In order to understand why the voiceless stops in the onsets of

the third syllables in (1) are aspirated we must consider the
foot structure of words with dactylic sequences. There are
different proposals regarding such words in the early metrical
phonology literature as well as in the more recent literature.
Consider the foot structure of the initial dactylic sequence
shown in (7) for the word Wìnnepesáukee, a structure that I
will argue against.
(7) Wìnnepesáukee (Fs = superfoot, F = foot, σ s =

stress syllable, σ w = stressless syllable; a final syllable
is shown incorporated into foot structure, the word
treeis not shown)
(p.112)
In (7), the
initial
dactylic
sequence
comprises
a
superfoot. This reflects the analysis of English found in Hayes
(1981) and McCarthy (1982) searchers assumed a similarity in
which a superfoot consists of a binary foot followed by an
adjoined single syllable. These researchers assumed a
similarity of environments between a word-final dactylic
sequence as exemplified by the trisyllabic words cápital and
párity and nonfinal dactylic sequences as in the first three
syllables of Wìnnepesáukee. The last syllable of a word-final
dactylic sequence behaves as foot-internal under the view that
flapping occurs in a foot-internal environment, as opposed to
aspiration, which is foot-initial (cf. Kiparksy 1979). The fact
that the /t/ in cápital and párity is flapped constitutes evidence
of a foot-internal environment. Researchers such as Hayes
Page 9 of 23
(1981) who incorporated the superfoot structure in (7) for

nonfinal dactylic sequences essentially assumed that they
were parallel with word-final dactylic sequences. However, the
problem with the foot structure shown in (7) is that there is no
clear reason why the stop would be aspirated at the beginning
of the third syllable since that syllable is not foot-initial. Hayes
(1981) and McCarthy (1982) did not consider aspiration in
their proposed footing of words like Wìnnepesáukee with
nonfinal dactylic sequences.
An alternative foot structure that I will argue for and that does
account for the aspiration in the third syllable of
Wìnnepesáukee is shown in (8). This involves a superfoot as
well, but different from that in (7). Here, the third syllable of
the initial dactylic sequence is adjoined as the first syllable of
a superfoot and does not group with the preceding syllables.
That is, the third syllable of the initial (nonfinal) dactylic
sequence forms a constituent with what follows rather than
with what precedes as in (7).
(8)
The foot
structure
in (8) for
nonfinal
dactylic
sequences
was
originally proposed by Withgott (1982) but was also suggested
in a footnote by McCarthy (1982). More recently, this foot
structure has been adopted by Jensen (2000) and Davis and
Cho (2003), and is suggested by Pater (2000). One immediate
advantage of this footing is that it makes for a unified
statement regarding American English aspiration (as noted by
Jensen 2000): voiceless stops are aspirated in foot-initial
position. This is what unifies the aspiration environments in
(6) with that shown in (1). We can illustrate this with the
convenient word pèripatétic in (9), where each foot-initial
position consists of a voiceless stop.
In (9) we see that all foot-initial stops are aspirated whether

they begin a binary foot like the first and fourth syllables of
Page 10 of 23
pèripatétic or the superfoot like the third syllable of

pèripatétic.
(p.113)
(9) Peripatetic
Further
evidence
for the
superfoot
structure
comes
from
words like potáto and other words such as those in the second
column of (6) which begin with an initial stressless syllable.
When such syllables have an initial voiceless stop they surface
as aspirated. This can be accounted for by the foot structure
shown in (10) where the initial syllable forms the first syllable
of a superfoot.
(10) potáto
The superfoot in (10) has the same structure as the superfoot

shown in (8) and (9). Thus, the superfoot is the same whether
it is word-initial or not.
An additional piece of evidence for the foot structure proposed

in (8) comes from expletive infixation. McCarthy (1982)
observed that words like Wìnnepesáukee (and the others in
(1)) show variation with respect to expletive infixation as seen
in (11).
(11)
Expletive infixation of Wìnnepesáukee

a. Winne-fucking-pesaukee
b. Winnepe-fucking-saukee
McCarthy’s observation regarding the variant forms of

expletive infixation in (11) seems correct and robust. Such
variation is applicable to all the words shown in (1). Given the
foot structure in (8), one can make a clear generalization
Page 11 of 23
accounting for the variant forms of expletive infixation

illustrated in (11), namely, that the expletive occurs before a
foot boundary. This can either be before a superfoot as in (11a)
or the binary foot as in (11b). Moreover, the generalization
must be that the expletive occurs before a foot boundary and
not after. This follows from McCarthy’s (1982) discussion of
expletive infixation in words having the prosodic pattern of
potáto illustrated in (10). In such words the expletive occurs
between the first and second syllables (e.g. po-fucking-tato,Ne-
fucking-braska). Given the foot structure for these words as
exemplified in (10), the generalization regarding expletive
infixation for both (10) and (11) is (p.114) that the expletive
occurs before the foot boundary. Thus, the foot structure
proposed in (8) accounts for variation in expletive infixation as
well as allowing for a unified generalization of English
aspiration as foot-initial.
Consequently, in considering the word mìlitarístic, paradigm

uniformity has no necessary role. Like Wìnnepesáukee, it has
the foot structure of (8), shown at (12).
(12)
The /t/ at
the
beginning
of the
third
syllable is
aspirated,
as expected, because it is foot-initial (that is, at the beginning
of the superfoot). Moreover, as indicated in (13), expletive
infixation shows variation with this word.
(13) Expletive infixation of militaristic

a. mili-fucking-taristic
b. milita-fucking-ristic
Notice that the variation of expletive infixation is not a fact

about the word military where the expletive can occur before
the relevant third syllable (mili-fucking-tary) but not after it
(*milita-fucking-ry). Consequently, we maintain that, contrary
to Steriade (2000), mìlitarístic does not display a paradigm
uniformity effect with mílitàry involving the noncontrastive
phonetic property of closure duration. Rather, mìlitarístic
Page 12 of 23
displays the same phonological patterning of the similarly

shaped monomorphemic words in (1). Te nonflapped nature of
the /t/ at the beginning of the third syllable is what is
expected, and it does not crucially reflect paradigm uniformity.
5.3 THE ROLE OF PARADIGM UNIFORMITY

In the previous section I have argued against the view of
Steriade (2000) that paradigm uniformity is at issue in
explaining the unflapped /t/ in the third syllable of militaristic.
Instead I have argued that an unflapped (and aspirated) /t/ is
expected. However, what is unexpected on the analysis
presented so far is the apparent obligatory flapping of the /t/
in the third syllable of càpitalístic. My contention, following
Pater (2000: 270), is that there is a paradigm uniformity effect
in càpitalístic with cápital. However, unlike Pater (2000), I do
not view the paradigm uniformity effect here as involving a
flap (or the closure duration of a flap), rather I contend,
following Burzio (this volume) that the paradigm uniformity
effect involves uniformity with respect to foot structure. That
is, càpitalístic preserves the foot structure of cápital. Under
this view, the flapping (p.115) of the /t/ in càpitalístic is a
consequence of the uniformity of foot structure with cápital.
Since the /t/ in cápital is foot-internal and intervocalic (i.e. a
salient environment for flapping) it will surface as a flap.
Hence, if càpitalístic preserves the foot structure of cápital,
the /t/ at the beginning of the third syllable of càpitalístic is
flapped because of its foot-internal (and intervocalic)
environment.
First, let us consider the foot structure in a word like cápital.

Our contention is that the footing of a third syllable of a
dactylic sequence that is word-final is different from such a
syllable when nonfinal as in (1). Words like those in (1) have
the foot structure shown in (8) where the third syllable of the
dactylic sequence forms a superfoot with the syllables that
follow. This is not an option if the third syllable of the dactylic
sequence is word-final since there is no following foot to which
it can adjoin. As a result, such a syllable is incorporated into
the preceding foot as shown in (14).
(14) cápital
Page 13 of 23
This foot
structure
readily
accounts
for the
flapping of
the /t/ on the view that flapping occurs foot internally (in an
intervocalic environment). Even if one maintains that final
syllables in English nouns are extrametrical, the foot structure
in (14) would reflect the surface foot structure of the word
(and not necessarily the initial foot structure under a
derivational view of phonology). As mentioned earlier, an
alternative to the foot structure in (14), found in Hayes (1981),
is to assume that the final syllable in (14) is adjoined to the
end of a superfoot in a way identical to what is shown for the
first three syllables in (7). While such a structure would still
allow for the expression of the flapping environment to be
foot-internal, the foot structure would not be independently
motivated since there do not seem to be other patterns that
would require an adjoined syllable at the end of a superfoot.
This should be contrasted with the superfoot type proposed in
this chapter in which the adjoined syllable is at the beginning
of the superfoot as seen in (8) and (9). As argued, the same
superfoot structure occurs in words like that in (10).
Consequently, I would maintain that the only superfoot
structure allowed in English is one that begins with an
adjoined syllable as illustrated in (8)–(10).
Now given the structure in (14), the form càpitalístic would

have the structure in (15) in which the dactylic foot from
cápital is preserved.
(15) cápitalistic
Page 14 of 23
(p.116)
The
flapping of
the /t/ in
the third
syllable of
càpitalístic is a consequence of the uniformity of foot structure
with cápital. As we have seen from data like those in (1), the
beginning of the third syllable of càpitalístic is not an
environment where one would otherwise expect flapping.
Evidence that the paradigm uniformity effect involves foot

structure and not, for example, the phonetic closure duration
of the coronal stop in terms of milliseconds, comes from
expletive infixation. As shown by the comparison of the two
possibilities in (16), the expletive can only occur after the third
syllable (at the foot boundary) as in (16a) and not before it
(foot-internally) as in (16b). (Infixation before the third syllable
might be possible if the coronal stop is pronounced as a
nonflap.)
(16)
a. capi[?]a-fucking-listic
b. *capi-fucking-[?]alistic
This should be compared with the optionality found with

expletive infixation with mìlitarístic that was shown in (13).
That the difference between (16) and (13) seems to be robust
can be taken as independent evidence that the paradigm
uniformity effect involves foot structure between cápital and
càpitalístic and not the noncontrastive phonetic property of
closure duration. It is worth mentioning that Steriade (2000)
reports experimentally that a paradigm uniformity effect was
induced in such nonce pairs as primitive–primitivistic in that
speakers who flapped the /t/ in primitive would also flap the
same /t/ in primitivistic. I would interpret this paradigm
uniformity effect as likewise involving foot structure in a way
similar to cápital and càpitalístic.
Page 15 of 23
5.4 OPTIMALITY-THEORETIC ANALYSIS

In the previous two sections of this chapter I have argued that
the unflapped (and aspirated) /t/ in the third syllable of
mìlitarístic reflects a regular pattern as exemplified by the
monomorphemic words in (1) while the flapped /t/ in the third
syllable of càpitalístic reflects a paradigm uniformity effect
involving foot structure. The initial ternary foot of càpitalístic
reflects the footing of cápital. The /t/ of the third syllable is
flapped because it is foot-internal and intervocalic which is the
salient environment for flapping. The aspirated nature of the
stop of the third syllable of the initial dactylic sequence in
mìlitarístic and the other forms in (1) constitute a minor issue
for a comprehensive analysis of aspiration in American
English. In this section I will briefly sketch an optimality
theoretic analysis based on Davis and Cho (2003) that
accounts for the environments in which aspirated stops appear
and then will formalize the paradigm uniformity effect
involved in càpitalístic by reference to Output–Output
constraints along the lines of Benua (1997 b) and Pater (2000).
The analysis of American English aspiration developed in

Davis and Cho (2003) and sketched briefly in modified form
here is an alignment approach to English aspiration within
optimality theory. Te analysis assumes (noncrucially) that
English voiceless (p.117) stops are not underlyingly specified
for the aspiration feature [spread glottis]. It follows Kiparsky
(1979); Nespor and Vogel (1986); Iverson and Salmons (1995);
and Jensen (2000) in viewing aspiration as occurring on
voiceless stops in foot-initial position. As discussed earlier,
foot-initial position is what connects the various environments
of aspiration shown in (6) and in (1). (See Davis and Cho for
discussion on forms that are potentially counterexamples to
the foot-initial view of aspiration.) The gist of the analysis
centers on the constraints given in (17) with the critical
ranking shown in (18).
(17) Constraints (sg = spread glottis)

a. AlignL(Ft, [sg]): Align the left edge of the foot
with the feature [spreadglottis].
b. *[sg, + voice] : The feature [sg] cannot be
realized on sounds that are[+voice].
c. Dep-[sg]: A feature [sg] in the output
corresponds to a feature [sg] in theinput.
Page 16 of 23
(18) *[sg, + voice] ≫ AlignL (Ft, [sg]) ≫ Dep-[sg]
The constraint in (17a) requires every foot to begin with a

segment having the feature [sg]. However, the force of this
constraint is mitigated by the higher ranked feature
cooccurrence constraint in (17b) that disallows the feature [sg]
from being realized on a voiced sound. (I leave open the
question whether [sg] is realized on voiceless fricatives in
American English; if it is not, then a constraint *[sg,
+continuant] would be high ranked. See Vaux (1998) for
relevant discussion.) The constraint in (17c) militates against
insertion of the feature [sg] and is the lowest ranked of the
three constraints in (17). The constraints are ranked as in (18).
This ranking can be justified by the tableaux shown in (19) and
(20).
(19) ‘pie’ /pay/—[phay]
/pay/ *[SG, ALIGNL (FT, DEP-

+VOICE] [SG]) [SG]
a. pay *!
b. ? *
h
p ay
(20) ‘buy’ /bay/—[bay]
/bay/ *[SG,+VOICE] ALIGNL (FT, [SG]) DEP-[SG]
a. *! *
h
b ay
b. ? *
bay
The tableau in (19) provides an argument for the ranking of

ALIGNL(FT, [SG]) over Dep-[sg] since the reverse ranking
would wrongly select (19 a) with an unaspirated (p.118) [p]
as the output. And the tableau in (20) provides an argument
for the ranking of *[SG, +VOICE] over ALIGNL(FOOT, [SG])
since the reverse ranking would incorrectly choose (20a) with
aspirated [bh] as the output. Under this analysis, the feature
[sg] is added only in foot initial position. Tis can be seen by the
tableau in (21) where the voiceless stop is not in foot-initial
position.
Page 17 of 23
(21) ‘lapse’ /læps/—[læps]
/læps/ *[SG, ALIGNL (FT, DEP-

+VOICE] [SG]) [SG]
a. * *!
h
læp s
b. ? *
læps
c. *! *
h
l æps
The key comparison is between candidates (21a) and (21b).

Candidate (21a) loses out because of its superfluous violation
of DEP-[SG]. That is, the insertion of [sg] on the /p/ in (21a) is
unmotivated because it is not in foot-initial position. The
analysis sketched here readily accounts for the aspiration in
the third syllable of data like those in (1) or (2). This is shown
in (22) where we compare two candidates, one with the third
syllable aspirated and one without the aspiration. The
candidate without aspiration, (22b), is eliminated because of
its fatal violation of the higher-ranked alignment constraint.
(22) ‘peripatetic’ /pær1p∂tεt1k/—[phaèr∂ph∂thέ?1k]

(brackets = superfoot; parentheses = foot)
/pær1p∂tεt1k/ *[SG, ALIGNL (FT, Dep-

+VOICE] [SG]) [sg]
a. ? (phàe.r1){ ph∂ ***

(thέ. ?1k)}
b. (phàe.r1){ p∂ *! **
(thέ. ?1k)}
The tableaux in (19)–(22) illustrate the alignment approach to

aspiration, the full details of which can be found in Davis and
Cho (2003). It is of some interest that Jensen (2000: 215)
specifically argues against an alignment analysis of English
aspiration. He says, ‘We cannot simply state an alignment
constraint that requires all aspirated stops to appear initially
in a foot, since this would be equally unviolated by aspirated
and unaspirated stops in that position. Conversely, we cannot
require that all feet begin with an aspirated stop, since other
segment types must be permitted in that position.’ However,
Page 18 of 23
this criticism does not hold. First, Jensen’s statement that ‘we
cannot require that all feet begin with an aspirated stop, since
other segment types must be permitted in that position’
ignores the fact that the constraint requiring all feet to begin
with the feature [sg] can be dominated. This is shown by the
ranking of *[SG, +VOICE] over ALIGNL(FOOT, [SG]), as
exemplified in the tableau in (20). With this ranking a voiced
sound cannot be aspirated even if it is in foot-initial position.
(p.119) Moreover, Jensen’s second criticism about an

alignment constraint not being able to distinguish an aspirated
from an unaspirated stop applies only if the alignment
constraint were stated as in (23) where the feature [sg] is
aligned to the beginning of the foot.
(23) ALIGNL([SG], FT): Align the feature [spread

glottis] with the left edge of thefoot.
If we consider the tableau with this constraint in (24) we see

that Jensen is right.
(24) ‘pie’ /pay/—[phay]
/pay/ ALIGNL ([SG], FT)
a. pay
b. phay
The constraint as stated in (23) is unviolated by unaspirated

stops. The candidate with the unaspirated stop in (24a)
vacuously satisfies the constraint since there is no feature [sg]
present. Crucial for the analysis, then, is that the constraint
must be stated as in (17a), ALIGNL(FT, [SG]), whereby the foot
is aligned with the feature [sg]. Stated this way, the alignment
constraint distinguishes between aspirated and unaspirated
stops as seen by tableaux like those in (19) and (22).
Consequently, Jensen’s criticisms of an alignment analysis do
not hold. Te alignment analysis posited in (17)–(18) and
illustrated by the tableaux in (19)–(22) accounts for aspiration
in American English. (See Davis and Cho 2003 for a more
thorough and detailed analysis.)
The remaining matter to be discussed is the optimality

theoretic analysis of càpitalístic. I argued in s. 5.3 that the
flapping of the /t/ in the third syllable of càpitalístic is a result
Page 19 of 23
of a paradigm uniformity effect with respect to the foot

structure of cápital. That is, càpitalístic preserves the ternary
foot of cápital. Flapping occurs because the foot-internal
environment for flapping is met if the foot structure is
preserved. The question that arises is why the foot structure of
cápital can be maintained in càpitalístic whereas the foot
structure of mílitàry is not maintained in mìlitarístic. The
answer is quite straightforward. As noted by Chomsky and
Halle (1968) and Myers (1987 b), the suffix -ic requires that
the syllable immediately preceding it be stressed. From Prince
(1990) we can understand this as a requirement that -ic be a
weak member of a trochaic foot, as is evidenced by its
triggering vowel shortening in such pairs as cone–conic and
Semite–Semitic. Thus, in both mìlitarístic and càpitalístic the
final two syllables comprise a trochaic foot. Given this, if the
foot structure of mílitàry were preserved in mìlitarístic there
would be a stress clash in that there would be a stress on both
the third and fourth syllables: * mìlitàrístic. Following Pater
(2000), we view the constraint against stress clash as
outranking a constraint on stress identity, which in our context
would be viewed as a general Output–Output constraint on
foot structure. (We label this OO-foot structure in the tableaux
at (25).) Because of the stress clash constraint, (p.120) the
third syllable of mìlitarístic cannot surface with stress and
consequently would be footed like the monomorphemic words
in (1) which reflect the regular footing constraints for English
(indicated in (25) by FOOTING CONSTRAINTS, which is
shorthand for the set of constraints responsible for English
foot structure).
The tableau in (25) shows the evaluation for mìlitarístic.

(Orthographic forms are used for convenience.)
(25) mìlitarístic (parentheses indicate a foot, braces

indicate a superfoot)
/military STRESS OO-FOOT FOOTING

+ istic/ CLASH STRUCTURE CONSTRAINTS
(míli)
(tàry)
a. (mìli) *! *
h
(t à)
(rístic)
Page 20 of 23
/military STRESS OO-FOOT FOOTING

+ istic/ CLASH STRUCTURE CONSTRAINTS
(míli)
(tàry)
b. ? (mìli) *
h
{t a(rístic)}
c. (mìlita) * *!
(rístic)
We see from candidate (25a) that maintaining a foot structure

in mìlitarístic that reflects mílitàry where the third syllable is
stressed (i.e. the head of a foot) would result in a stress clash
with the following stressed syllable. Candidate (25a), though,
respects OO-FOOT STRUCTURE since the foot heads in the
word mílitàry are also foot heads in mìlitarístic. Nonetheless, it
is the higher ranked stress clash constraint that prevents (25a)
from being the winning candidate. Consequently, the output of
mìlitarístic cannot respect the OO-foot structure constraint.
The choice then is between (25b) and (25c); neither violates
STRESS CLASH, but both violate OO-FOOT STRUCTURE in
that the head of the second foot in the base word mílitàry is
not a foot head in either (25b) or (25c). The winning
candidate, (25b), is preferred to (25c) since (25b) results from
the normal footing of a dactylic sequence that is not word-
final, obeying FOOTING CONSTRAINTS. Thus, the footing of
mìlitarístic is like that of the monomorphemic words in (1) as
shown in (8).
On the other hand, the foot structure of cápital can be

completely maintained when the final two syllables of
càpitalístic comprise a trochaic foot; there is no stress clash in
càpitalístic. Since there is no stress clash, OO-FOOT
STRUCTURE can be respected; there is no overriding
constraint that would compel the third syllable of càpitalístic
to be refooted with the syllables that follow in violation of OO-
FOOT STRUCTURE. The tableau in (26) for càpitalístic
contrasts with mìlitarístic in (25).
(26) cápitalistic
Page 21 of 23
/capital + STRESS OO-FOOT FOOTING

istic/ CLASH STRUCTURE CONSTRAINTS
(cápital)
a. ? *
(càpita)
(lístic)
b. (càpi) *!
h
{t a(lístic)}
c. (càpi) *! * *
h
(t à)
(lístic)
(p.121) As seen by the winning candidate in (26a), the foot

structure of càpitalístic can parallel that of cápital without
violating STRESS CLASH. While this candidate violates the
expected foot structure constraints by having the third syllable
as part of the initial foot, it is none the less preferred to (26b)
given the ranking of OO-FOOT STRUCTURE over the
FOOTING CONSTRAINTS. The flapping of the /t/ in the third
syllable of càpitalístic is just a consequence of its foot-internal
environment reflecting the preserved foot structure from
cápital.
5.5 SUMMARY AND CONCLUSION

In this chapter I have argued against the view of Steriade
(2000) that the unflapped /t/ of the third syllable of mìlitarístic
reflects a paradigm uniformity effect involving the
noncontrastive phonetic property of closure duration in
milliseconds. In s. 5.2 I showed that the unflapped (and
aspirated) /t/ is expected in mìlitarístic and so paradigm
uniformity is not at issue. Instead, I argued in s. 5.3 that the
flapped /t/ in the third syllable of càpitalístic is phonologically
unexpected and can be explained by paradigm uniformity. But,
here, paradigm uniformity involves uniformity with respect to
the foot structure of cápital and not to a noncontrastive
phonetic property of duration in milliseconds with respect to
the flap. A major claim of my analysis is that there is a
difference between a word-final dactylic sequence as in the
word cápital and one that is not word-final as in
Mèditerránean. The third syllable of a word-final dactylic
sequence forms a foot with the preceding syllables. That is, a
word-final dactylic sequence surfaces as a single ternary foot.
Page 22 of 23
The /t/ in cápital is flapped because it is foot-internal and

intervocalic, the salient environment for flapping. This is
contrasted with the third syllable of a dactylic sequence that is
not word-final as in Mèditerránean or the other words in (1).
The third syllable here forms the beginning of a superfoot with
what follows. The voiceless stop in such an environment would
be aspirated because it is foot-initial. The flapped /t/ of the
third syllable of càpitalístic mirrors the foot structure of cápital
where the /t/ is flapped because it is foot-internal. The
optimality-theoretic analysis of s. 5.4 frames the aspiration
issue more generally in terms of a foot-alignment approach.
What the analysis captures with respect to the difference
between càpitalístic and mìlitarístic is that the foot structure of
cápital can be maintained in càpitalístic without violating a
high-ranked stress clash constraint whereas the foot structure
of mílitàry with a stressed third syllable cannot be maintained
in mìlitarístic since that would result in a violation of the stress
clash constraint (*mìlitàrístic) where both the third and fourth
syllables would have stress. In conclusion, while there may be
cases of paradigm uniformity involving noncontrastive
phonetic properties as Steriade contends, the càpitalístic-
mìlitarístic distinction is not a convincing case.
Page 23 of 23
Jita Causative Doubling Provides Optimal Paradigms


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Jita Causative Doubling Provides

Optimal Paradigms
Laura J. Downing (Contributor Webpage)
DOI:10.1093/acprof:oso/9780199267712.003.0006

This chapter argues that the cyclic causative doubling found in
Bantu languages such as Jita (spoken in Tanzania) and Bemba
(spoken in Zambia) provides evidence in favor of the latter
approach. The chapter proceeds as follows. Section 6.2
presents the Jita causative doubling data. Then, following
Hyman's account of similar Bemba data, it is shown that
causative doubling is clearly morphologically conditioned but
is also clearly problematic for a traditional cyclic account.
Section 6.3 argues for an Optimal Paradigms (OP) analysis of
causative doubling. An important result of the analysis is to
show that only the causative, and not other derivational stem
suffixes, defines a paradigm distinct from its morphological
base stem because the causative is the only derivational suffix
which induces phonological alternations in stems to which it
attaches.
Keywords: Bantu languages, Jita, cyclic causative doubling, Bemba, Optimal

Paradigms
6.1 Introduction
Page 1 of 46
A body of recent work within Optimality Theory (OT) has

argued that cyclic phonological effects can be accounted for in
a non-derivational way through Output–Output
correspondence. Derivationally related words are grouped into
paradigms, and their outputs are compared for identity or
uniformity of a shared base string. An important issue in the
various OT approaches to Paradigm Uniformity is how to
define the shared base string. Work such as Benua (1997a, b)
and Kager (1999a, 2000), following the derivational cyclic
tradition (see Cole 1995 for an overview), has proposed (p.
123) that bases must be morphologically well-formed
subconstituents of the derivationally related words on which
Paradigm Uniformity is imposed. They argue further that
entire words (not smaller lexical constituents) are compared in
evaluating Paradigm Uniformity. In contrast, work such as
Buckley (1999); Burzio (1994a, 2000b, Ch. 4 above);
Kenstowicz (1996, 1997, Ch. 7 below), McCarthy (Ch. 8 below)
and Steriade (1997a, 1999a) proposes that Output–Output
constraints enforcing Paradigm Uniformity can focus on the
shared string of a constellation of related forms whether that
string is a word or not. The shared string also need not be a
well-formed morphological subconstituent of the words in the
paradigm.
In this chapter, I argue that the cyclic causative doubling

found in Bantu languages such as Jita (spoken in Tanzania)
and Bemba (spoken in Zambia) provides evidence in favor of
the latter approach. Te argument is structured thus: in s. 6.2, I
present the Jita causative doubling data. Then, following
Hyman’s (1994) account of similar Bemba data, I show that
causative doubling is clearly morphologically conditioned but
is also clearly problematic for a traditional cyclic account. In s.
6.3, I argue for an Optimal Paradigms (OP) (McCarthy, Ch. 8)
analysis of causative doubling. An important result of the
analysis is to show that only the causative, and not other
derivational stem suffixes, defines a paradigm distinct from its
morphological base stem because the causative is the only
derivational suffix which induces phonological alternations in
stems to which it attaches.
6.2 Why Causative Doubling is Morphologically

Conditioned
6.2.1 The Jita data
Page 2 of 46
Let us begin by looking at what we mean by causative

doubling in Jita. As shown by the data in (1), the causative
morpheme in Jita, as in numerous Bantu languages (Bastin
1986; Hyman 1993, 1994; Poletto 1998) is the palatal glide, /-
y-/. Notice that in Jita, as in many Bantu languages, the
causative /-y-/ triggers spirantization of some consonants. In
Jita, only ‘r’ is productively spirantized. (See Bastin 1986;
Downing 2001; Hyman 1994, 2003 and references therein for
discussion.)
Page 3 of 46
(1) Jita Causative (Downing 1996, 2001, and unpublished field notes; causative is bolded; ‘=’ indicates the
stem boundary)
Infinitive Gloss Causative of Base

Infinitive
okw=aik-a ‘talk’ okw=aik-y-a
oku=βilim-a ‘run’ oku=βilim-y-a
okw=I:gurur-ir-a ‘open for’ okw=i:gurur-is-y-a (‘cause [to open for]’)
oku=gur-a ‘to buy’ oku=gus-y-a (‘to sell’)
Page 4 of 46
(p.124) As shown by the data in (2), when other suffixes are

added to a causative stem, the /-y-/ is repeated. It occurs not
only in its semantically motivated position, but also after the
final consonant of every subsequent morpheme in the stem:1
Page 5 of 46
(2) Jita Causative Doubling (Downing 1996 and unpublished notes)
Infinitive Gloss Base+Caus. Base+Caus.

+Recipr.
a. oku=kór-a ‘to do’ oku=kós-y-a oku=kos-y-á:n-y-a
b. okw=óig-a ‘to bathe’ okw=ó:s-y-a okw=o:s-y-á:n-y-a
Infinitive Gloss Base+Caus. Base+Caus. Base+C+Ap+Rec

+Appl.
c. oku=gur-a ‘to buy’ oku=gus-y-a oku=gus-i:s-y-a oku=gus-i:s-y-

a:n-y-a
d. oku=βón-a ‘to see; get’ oku=βón-y-a oku=βon-y-és-y-a
Infinitive Gloss Base+Caus. Base+Caus.

+Perf.
e. okw=aik-a ‘to talk’ okw=aik-y-a =aik-i:s-y-e /=aik-y-ire/
f. oku=βilim-a ‘to run’ oku=βilim-y-a =βilim-i:s-y-e /=βilim-y-ire/
(cf. =aik-ire ‘talk’, perfective; =βilim-ire ‘run’, perfective)
Page 6 of 46
For example, in the last word in (2c) [oku=gus-i:s-y-an-y-a] ‘to

sell to each other’, we find three instances of the causative
(bolded): one in the semantically motivated position following
the first CVC and one after each of the subsequent
derivational morphemes, /-ir-/ and /-an-/. It is this repetition of
the causative that is referred to as causative doubling.
As Hyman (1994, 2003) shows, the causative glide (or the

spirantization triggered by the causative glide) is doubled in
many Bantu languages. For example, in Bemba, spoken in
Zambia, the causative suffix /-y-/ triggers spirantization of a
preceding consonant, as shown in (3a). When additional
suffixes are added to the causative stem, as shown in (3b),
both the consonant preceding the causative glide and the base
stem-final consonant are spirantized even though the base
stem is not in the context for spirantization.2
Causative doubling can be understood as a variant on the

prosodic displacement of the causative /-y-/ Hyman (1993) and
Hyman and Mchombo (1992) have shown is (p.125)
commonly found in Bantu languages. The Kinande data in (4)
provide a clear example of this prosodic displacement. Notice
in these data that the causative /-y-/ occurs on the surface
following the rightmost consonant of the stem, even though its
semantically motivated position may be several morphemes to
the left:3
Page 7 of 46
(3) Bemba causative doubling (Hyman 1994, figs. (9–10), pp. 85–6)
Derivational Stem Gloss Causative Gloss
a. -leep- ‘be long’ -leef-y- ‘lengthen’
-up- ‘marry’ -uf-y- ‘marry off’
-fiit- ‘be dark’ -fiis-y- ‘darken’
-ónd- ‘be slim’ -óns-y- ‘make slim’
-buuk- ‘get up’ [intr.] -buus-y- ‘get someone up’
Applicative Stem Gloss Caus.+Applic. Gloss
b. –leep-el- ‘be long for/at’ -leef-es-y- ‘lengthen for/at’
-up-il- ‘marry for/at’ -uf-is-y- ‘marry off for/at’
-fiit-il- ‘be dark for/at’ -fiis-is-y- ‘darken for/at’
-ónd-el- ‘be slim for/at’ -óns-es-y- ‘make slim for/at’
-buuk-il- ‘get up for/at’ -buus-is-y- ‘get someone up for/

at’
Page 8 of 46
(4) Kinande causative displacement (Mutaka and Hyman 1990; Mutaka 1994)
Infinitive Gloss Base+Causative Base+Caus+Appl Base+C+A+Reci

ic p
a. eri=gul-a ‘to buy’ eri=gul-y- eri=gul-ir-y-a eri=gul-ir-an-y-a
b. *eri-bul-a ‘to ask’ erí=bul-y-a erí=bul-ir-y-a erí=bul-ir-an-y-a
Infinitive Gloss Base+Causative Base+Caus+Perf

ective (-ire)
c. erí=kohol-a ‘to cough’ erí=kohol-y-a /=kohol-ir-y-e/
d. eri=huma ‘to beat’ eri=hum-is-y-a /=hum-is-ir-y-e/
Page 9 of 46
What makes the doubling found in Jita and Bemba different

from the prosodic displacement found in Kinande is that the
causative /-y-/ (or associated spirantization) occurs not only in
the prosodically displaced position following the stem-final
consonant but also in its semantically motivated position and
following all morphemes in between the two. (This can be
clearly seen by comparing the sets of cognates in (2c) and
(4a).)
This is the problem to be accounted for, then: how does the

causative morpheme come to occur (and/or trigger
spirantization) following every morpheme added to the
causative stem? In the next section, I extend Hyman’s (1994)
analysis of causative doubling in Bemba to the Jita data and
conclude, following Hyman, that causative doubling is
morphologically conditioned: the pronunciation of a causative
stem is determined by the pronunciation of its morphological
subconstituents.
(p.126) 6.2.2 Hyman’s (1994) analysis of causative doubling

Hyman (1994) analyzes the Bemba causative doubling in (3)
and demonstrates clearly that a morphology-blind account of
causative doubling is unworkable. (Hyman’s arguments for
Bemba are extended straightforwardly to the Jita data in this
section.) A purely phonological account of causative doubling
would be to propose that it results from a process of
palatalization harmony, applying either leftward from the
semantically motivated position of the causative suffix (as in
(5a)) or rightward from the prosodically displaced position (as
in (5b)):
(5) Doubling as palatalization harmony

a.
OR
b.
Page 10 of 46
rightward spread rule like that in (5 a) predicts that any

palatal glide should trigger palatal harmony on coronal
sonorants to the right. However, as shown by the Jita data in
(6), that prediction is incorrect.
(6) Failure of rightward ‘palatal

harmony’ from non-causative
palatal glides (failed harmony
targets bolded)
a. oku=u:ny-ír- to smell out; (cf. oku=ú:ny-a ‘to

isy-a ‘ sniff’ smell’)
b. oku=sy-e:r-a ‘to grind for’ (cf. oku=sy-a ‘to

grind’)
Root final palatal glides in verbs like those in (6a, b) do not

trigger palatal harmony on coronal sonorants to their right.
Only the causative palatal glide does. (The rightmost palatal
glide in (6a) is part of the ‘long causative’ suffix /-isy-/, and so
is in its semantically motivated position.)
A morphology-blind rule of leftward palatal harmony like that

in (5b) also makes incorrect predictions. By this rule, we
would expect that a palatal glide should trigger palatal
harmony on (at least) all coronal sonorants to its left. However,
as shown by the example in (6a) and the additional data in (7),
the semantically motivated position of the causative glide
defines the left edge of ‘palatal harmony’. Coronal sonorants
preceding this morphological position do not palatalize. Note,
too, that in all of these examples, the palatalized consonants
are found either stem-initially or at the right edge of
morphemes, never morpheme internally.
As Hyman (1994) argues, all of these morphological conditions

fall out in a cyclic account which requires the derivational
interleaving of phonology and morphology (Brame 1974;
Kiparsky 1982a; etc.). In fact, doubling represents the sort of
morphologically conditioned rule overapplication that is one of
the main motivations for the (p.127) theory of cyclic
phonology. The morphological rule displacing (and copying)
Page 11 of 46
the causative /y/ and the phonological rule of spirantization

triggered by this morpheme reapply every time another suffix
is added instead of only once. As shown in (8), if the causative
is copied and displaced rightwards (by a rule requiring the
causative to occur both in its input position and following the
stem-final consonant) after the addition of each subsequent
derivational suffix to the causative stem, doubling is the
expected result:
(7) Failure of leftward ‘palatal harmony’ to coronals

preceding the semantically
motivated position of the causative glide (failed harmony
targets bolded)
Base Gloss (Intensive) Gloss

Stem Causative
a. =ig- ‘be open for’ =ig-urur- ‘cause to be

urur-ir-a is-y-a open for’
b. =ig- ‘be open’ =ig-urus- ‘cause to

urur-a y-a open’
c. =mar-a ‘finish’ (intrans.) =mar-is-y- ‘finish

a up’ (trans.)
d. =twa:r-a ‘get married’ =twa:r-is- ‘give in

y-a marriage to’
e. =fúm-a ‘insult’ (intrans.) =fum-íris- ‘curse

y-a at’ (intensive,
trans.)
f. = βey- ‘lie to, about’ =βey-er- ‘cause to lie

er-a es-y-a about;
slander’
g. =som- ‘appoint’ (intrans) =som- ‘point to;

ek-a eres-y-a designate’
h. =nyurur- ‘pull’ =nyurus- ‘cause to

a y-a pull’
(8) Cyclic derivation of (2c) [=gus-i:s-y-a:n-y-a] ‘to sell to

each other’
Cycle 1
Page 12 of 46
morphology gur-i
phonology
displacement/copy N/A
spirantization + gus-i
Cycle 2
morphology gus-i-ir
phonology
displacement/copy gus-i-ir-i
spirantization + gus-i:s-i
Cycle 3
morphology gus-i:s-i-an-
phonology
displacement/copy gus-i:s-i-an-i
spirantization + gus-i:s-i-an-i-
gliding gus-i:s-y-a:n-y-
(p.128) The semantically motivated position of the causative

necessarily defines the left edge of ‘palatal harmony’ if the
‘harmonic’ glides result from doubling which cycles rightward
from the input position of the causative glide. Only the
causative would trigger ‘palatal harmony’ if this is a cyclic rule
introduced with the causative glide. In fact, causative doubling
is a classic case of the sort of surface exceptions to
morphological or phonological processes cyclicity was
developed to explain. The overabundance of causative glides
(and of spirantized/palatalized consonants) is accounted for by
preserving the results of processes applying at an
intermediate stage of the morphological derivation in surface
forms. And it is a hallmark of cyclic rules that they should not
apply morpheme internally.
6.2.3 Problems for a cyclic analysis

Even though causative doubling provides convincing evidence
for cyclic rule application, doubling also presents some
important problems for classic cyclicity theory, as Hyman
(1994) points out. First, we expect morphemes to show a one-
to-one correspondence between form and meaning. Even
though, as we have seen, doubling has a morphological, not a
Page 13 of 46
phonological, explanation, it repeats the form of the causative

morpheme without adding repeated causative meaning. It is
unexpected for cyclic word derivation to lead to a cyclic
repetition of the form of a morpheme with no related change
in meaning. (As shown in (6), since only causative /y/ doubles,
not other root-final /y/s, doubling is clearly a process of
morpheme repetition, not a phonological process like palatal
harmony.)
This leads to a second, related, problem, namely, that cyclic

theory requires the output of each cycle to be a well-formed
morphological subconstituent of the input to the next cycle
(Cole 1995, etc.). For example, oríginal is uncontroversially a
well-formed subconstituent of orìginálity. As a result, it is
consistent with a cyclic analysis to propose that the
unexpected stress on the second syllable of orìginálity is
preserved from regular stress assignment to oríginal on the
preceding cycle. However, as shown in (9b), the doubled
causative leads to a morpho-syntactically ill-formed
representation since the position of the second causative is
inconsistent with the semantic scope of the causative
morpheme in this stem.4 Further, as shown in (9a), there is not
always an obvious morpho-syntactic position for the copy of
the causative to adjoin to. As shown by the data in (3 c, d), the
phonological requirement that the causative /-y-/ must occur
following the final consonant of the stem forces it to interrupt
the perfective suffix, /-ire/. This leads to a doubly ill-formed
morphological result, since the perfective, as an inflectional
suffix (IFS), should be outside the domain of derivational
affixation. (The causative is a derivational stem (DStem)
suffix.)
A final problem is that there is no other phonological or

morphological evidence for the cycle in Jita (or Bemba). In
fact, even the other glide morpheme, the passive (p.129) /-
w-/, does not double, as shown by the Jita data in (10), even
though it is subject to the same positional requirement as the
causative glide: it follows the final consonant of the stem and
so also interrupts the perfective suffix, /-ire/.
(9) Ill-formed morpho-syntactic representations derived

from Causative Doubling
a.
Page 14 of 46
b.
(10) Passive in Jita, showing displacement without

doubling
Stem Gloss Passive Passive+Perfective
a. =u:mbak- ‘build =u:mbak-ir-

ir-a for’ w-a
b. =sakir-a ‘help’ =sakir-w-a
c. =βum-a ‘hit’ =βum-w-a =βum-ir-w-e
d. =si:k-a ‘bury’ =si:k-w-a =si:k-ir-w-e
e. =te:k-er-a ‘cook =te:k-er-w-a

for’
Notice that displacing the passive morpheme in the applicative

forms in (10a, e), makes them ambiguous. The pronunciation is
identical whether the passive morpheme has semantic scope
over the base verb (e.g. ‘build’ or ‘cook’) or the applied form
(e.g. ‘build for’ or ‘cook for’). Even though doubling the
passive glide would have a clear semantic motivation because
it could serve to disambiguate forms like these, it does not
occur. The OT reanalysis of causative doubling presented in
the next section addresses these problems.
6.3 Causative Doubling as an Optimal Paradigms Effect
Page 15 of 46
Since the cyclic analysis of causative doubling is inherently

derivational, requiring a level of representation corresponding
to the output of each morphological operation, it would appear
to pose a challenge for versions of Optimality Theory (OT)
which posit only two levels of representation, the input and
output. However, work such as Benua (1997a), Buckley (1999),
Burzio (1994a, 2000b), etc., has shown that many cyclic
phenomena can be analyzed within two-level OT without
reference to intermediate stages of a morphological
derivation. (See, however, Kiparsky 2000 for a thoughtful
critique of some OT analyses of cyclic phenomena.) The
generalization underlying (p.130) the cyclic analysis of
doubling is that a phonological output (spirantization)
motivated in one form of the stem is carried over to a
morphologically derived form of the same stem. As a result,
related causative stems are pronounced more similarly to each
other if doubling occurs than if no doubling is found. This can
be seen by comparing the underlined strings in the Jita verb
stems in (11b) with the equivalent strings in the hypothetical
*Jita forms without doubling given in (11c). (Recall that ‘i’ and
‘y’ are featurally identical).
(11) Phonological motivation for doubling (/-gur-a/ ‘buy’; /-

gur-y-a/ ‘sell’)
a. Input b. Causative Doubling c. No Doubling

(Jita) (*Jita)
/=gur-y- =gus-y-a =gus-y-a

a/
/=gur-y- =gus-i:s-y-a =gur-is-y-a

ir-a/
/=gur-y- =gus-y-an-y-a =gur-an-y-a

an-a/
/=gur-y- =gus-i:s-y-a:n-y-a =gur-ir-an-y-a

ir-an-a/
The requirement that related stems have similar pronunciation

can be captured within OT using Correspondence constraints
which compare the outputs of morphologically related words.
As Benua (1997a) suggests in her analysis of the Bemba data
in (3), this general approach can easily account for the sort of
cyclicity that motivates causative doubling. In the next
Page 16 of 46
sections, I develop an Output–Output correspondence analysis

of causative doubling within Optimal Paradigms (OP) theory
(McCarthy Ch. 8).
6.3.1 Doubling as Paradigm Optimization

The generalizations any analysis of causative doubling must
account for are that, first, the causative glide ‘y’ productively
spirantizes a preceding ‘r’. We can account for this
generalization by proposing that a Markedness constraint *ry
(12a) penalizes this output sequence. As argued in Downing
(2001), ‘r’ is a [–cont] tap in Jita and ‘s’ is a [+cont] fricative,
so spirantization crucially involves a [cont] alternation, most
plausibly motivated by the narrow stricture of the following
palatal glide and by the articulatory incompatibility between
‘r’ and the palatal glide.5 As shown in (12c), ranking *ry (12a)
above the Faithfulness constraint MAX/DEP [CONT] (12 b),
optimizes spirantizing ‘r’ to avoid marked ‘ry’ sequences.
As noted in s. 6.2.1, above, doubling is related to the prosodic

displacement of the causative glide: it occurs following the
stem-final consonant, no matter where its semantically
motivated position in the stem might be. As Downing (2001)
argues, the causative occurs in this position because there is
ample cross-Bantu evidence that the stem-final consonant
defines the right edge of a morpho-prosodic constituent,
PStem, defined as the string from the stem-initial consonant
through the stem-final consonant. (p.131) 6 The Alignment
constraint in (13) formalizes the requirement that the
causative glide occur at the right edge of PStem, in effect
Chomsky-adjoining the causative glide to the right edge of
PStem.
(12) Constraints and ranking optimizing causative

spirantization
a. *ry: [ry] sequences do not occur OUTRANKS
b. MAX/DEP [CONT]: For corresponding input
and output segments, each [cont] value of the
input should occur in the output (MAX) and each
[cont] value of the output should occur in the
input (DEP).
c.
/=gur-y- a/ *ry MAX/DEP [CONT]
Page 17 of 46
a. F =gus-]y]-a *
b. =gur-]y]-a *!
(13) ALIGN /y/: Align R(PStem, Caus /y/)

Align right edge of (Causative ) PStem with the
right edge of Causative /y/
The simultaneous morphological and prosodic parse of a

causative verb, showing the mismatch between morphological
and phonological structure resulting from ALIGN /y/, is given
in (14).
(14) Morphological v. Morpho-prosodic structure of

causative stems
a. Morphological structure (given
hierarchically):
b.
Morpho-prosodic structure (indicated by square

brackets in a)
As shown in (14), the MStem dominates the root and suffixes

in their semantically motivated positions. In contrast, the
Causative PStem contains the string from the root-initial
consonant through the phonologically determined surface
position of the causative. (Notice that the displaced causative
‘y’ has no morphological affiliation, only a morpho-prosodic
one.)
(p.132) Since the effect of ALIGN /y/ (13) is to optimize a

different output position for the causative glide than its input
position, it must outrank CONTIGUITY (15a). Te Faithfulness
constraint DEP /y/ (15b), which penalizes /y/s in the output
which are not present in the input, is the constraint violated
when doubling occurs.
(15)
Page 18 of 46
a. CONTIGUITY: Segments that are adjacent in

the input must be adjacent in the output.7
b. DEP /y/: Every /y/ in the output must have a
correspondent in the input.
The tableau in (16) shows that the constraints and rankings so

far define the *Jita causative pattern illustrated in (11c), where
the causative displaces but does not copy. (‘]’ marks the right
edge of the Causative PStem (see (14b).)
(16) *Jita
Page 19 of 46
/=gur-y-an-a/ *ry ALIGN /y/ MAX/DEP [CONT] DEP /y/ CONTIGUITY
a. F =gur-an]y]-a *
b. =gus-y-an]y]-a * *! *
c. =gus-y-an]a *! *
d. =gur-y-an]y]-a *! * *
Crucial rankings:
*ry >> MAX/DEP [CONT] optimizes spirantization to satisfy markedness
ALIGN /y/ >> CONTIGUITY optimizes causative /y/ displacement
Page 20 of 46
Candidate (16a) is optimal as it satisfies the constraints

motivating spirantization (*ry) and displacement (ALIGN /y/).
The causative doubling candidate in (16b) is non-optimal as
the doubled /y/ violates DEP /y/, while the doubling candidate
in (16d) also violates *ry. Candidate (16c) is non-optimal as the
causative glide is not at the right edge of PStem, in violation of
high-ranked ALIGN /y/.
To optimize doubling, some constraint must outrank DEP /y/.

As noted in discussing (11), Output–Output correspondence
straightforwardly motivates doubling as this process makes
the outputs of related Causative Stems more similar. In
Optimal Paradigms (OP) formalism (McCarthy, Ch. 8), which I
adopt for the analysis (this choice will be justified in the next
sections), all derivationally related Causative Stems form a
paradigm and are subject to the following OP faithfulness
constraint (17).
(p.133)
(17) OP-CS: MAX/DEP-OP Causative Stem

The Causative Stem common to all paradigm
members must contain all the segments (MAX) and
only the segments (DEP) found in the
corresponding position in the other paradigm
members.
The constraint in (17), when ranked above DEP /y/ (15a),

optimizes doubling, as the effect of doubling is for all
segments in a Causative Stem to occur in the same position in
all related stems.
The tableau in (18) shows how these constraints and rankings

optimize causative doubling in the partial paradigm taken
from (11b). (In this tableau, the Causative Stems being
compared for OP-CS are underlined;‘]’ marks the right edge of
the Causative PStem (see (14b).)
(18)
Page 21 of 46
/oku=gur-y-a/, / *ry ALIGN /y/ OP-CS MAX/DEP DEP/y CONTIGUITY

oku=gur-y-an-a/ [CONT]
a. F *
oku=gus]y]-a,
oku=gus-y- * * *
an]y]-a
b. *
oku=gus]y]-a,
oku=gusy-an]-a *! *
c. oku=gur] *!
y]-a,
oku=gury-an]y]- * * *
a
d. *! *
oku=gus]y]-a,
oku=gur-an]y]-a *
Crucial constraint rankings, same as (16), plus:
OP-CS >> DEP /y/ optimizes doubling to satisfy paradigm uniformity
Page 22 of 46
In this tableau, paradigm (18a) is optimal because it satisfies

OP-CS (17), as the shared causative stem (underlined) is
identical throughout the paradigm. It also satisfies the highly
ranked constraints motivating spirantization (*ry) and
requiring displacement of the causative /y/ (ALIGN /y/).8
Although paradigms (18b) and (18c) satisfy OP-CS, (18b) is
non-optimal because it fails to satisfy ALIGN /y/, and (18c) is
non-optimal because it fails to satisfy *ry. This illustrates
McCarthy’s (Ch. 8) point that optimal paradigms tend to
generalize outputs that overapply the high-ranked markedness
or prosodic well-formedness constraints of the language.
Paradigm (18d), the one which is optimal in *Jita (see (16)), is
non-optimal in Jita as it violates high-ranked OP-CS. (p.134)
Comparing (16a) with (18d) clearly shows the crucial role of
OP in distinguishing doubling languages such as Jita from non-
doubling languages that otherwise have the same grammar for
the causative stems.
To sum up this analysis, doubling can be considered a

paradigm uniformity effect optimized by an OP constraint (17)
requiring the Causative Stem of morphologically related verbs
to be identical. As shown in (18), candidate paradigms consist
of words, since only words are complete outputs. However, OP
constraints compare a specified string within the related
words (the shared Causative Stem, in this case) for identity.
This specified string need not be a word nor need it be a
morphological subconstituent of the words in the paradigm. As
McCarthy (Ch. 8 below) notes, these properties of the OP
approach to paradigm uniformity make it similar to the
theories proposed by Burzio (1994a, 1996, 2000b, 2002b, Ch. 4
above) and Kenstowicz (1996, 1997, Ch. 7 below). Paradigms
are defined as ‘constellations’ of morphologically related
words, and the paradigm base is a mainly phonological notion
(the string in the word or words influencing the pronunciation
of the others in the constellation). This contrasts with the
theories of Benua (1997a) and Kager (1999a, 2000), who adopt
from the theory of cyclicity (Brame 1974; Kiparsky 1982a;
Inkelas 1989; etc.) the requirement that the paradigm base
must be a ‘fully prosodized, independently occurring word,
which is also compositionally related to the output’ (Kager
1999a;2000). In the next sections, I show causative doubling
supports the OP approach to paradigm uniformity by
comparing the OP analysis first with Hyman’s (1994)
Page 23 of 46
derivational cyclic analysis of Bemba and then with the

Transderivational Correspondence Theory (TCT) analysis of
Bemba sketched in Benua (1997a).
Page 24 of 46
6.3.2 How OP improves on a derivational cyclic analysis

As shown in s. 6.2, above, Hyman’s (1994) derivational cyclic
analysis of causative doubling provides an elegant account of
the morphological conditions on this process. However, it also
faces two important problems. First, as highlighted in the
representations in (9) and (14), above, doubling leads to ill-
formed morpho-syntactic representations. Notably, the
displacement process which is linked to doubling sacrifices
morpho-syntactic well-formedness to the prosodic well-
formedness condition that the causative glide follow the stem-
final consonant, whatever its morphological affiliation. Tis is a
problem for any theory of the morphology–phonology interface
such as cyclicity which requires the inputs and outputs of
phonological processes to be morphologically well-formed, as
prosodic displacement creates a morphologically ill-formed
context for causative spirantization.
However, there is a body of work demonstrating that

morphological and phonological well-formedness conditions on
word formation do not always agree. It is familiar from
infixation and cliticization, for example, that phonological
factors can conflict with morpho-syntactic conditions on
morpheme placement. As Booij and Lieber (1993), among
others, suggest, this conflict can be resolved by proposing that
certain affixes subcategorize for both a morpho-syntactic and
a prosodic base. The surface (p.135) position of the affix will
be the one that best satisfies both. This is also the approach to
infixation motivated in the OT literature (McCarthy and Prince
1993a, b, etc.). Infixation is optimal when constraints on
phonological well-formedness outrank constraints on
morphological well-formedness. For example, /um/ infixation in
Tagalog can be accounted for by ranking Onset, No Coda
above the constraint defining /um/ as a prefix, aligned with the
left edge of the stem, so that prosodically well-formed [grum-
adweit] is optimal, not morphologically well-formed [*um-
gradweit].
Likewise, causative displacement is accounted for formally by

ranking Align /y/ (13), which requires the causative /y/ to occur
at the right edge of a PStem, above Contiguity (15b), the
constraint requiring the causative to occur in its semantically
motivated position. (The resulting mismatch between
morphological and phonological structure is illustrated in (14),
above.) In OT, then, there is no need to insist that the Base for
Paradigm Uniformity be morphologically well-formed, as only
Page 25 of 46
output strings and output structure (including prosodic

structure) are compared for identity by Output–Output
constraints. This approach emphasizes that paradigm
uniformity serves to reinforce the recognition of semantic
similarity among related words by increasing their
phonological similarity. The role of morphological structure in
defining related-ness is down-played, serving only to
approximately specify the string that is evaluated for
uniformity.
Another problem for a derivational cyclic account is that

causative doubling provides the only evidence for cyclicity in
the grammar of Jita and Bemba. No explanation for why the
causative alone should introduce this grammatical
complication is available in this theory. It is also true in the OP
analysis that the causative is the only morphological process
clearly motivating paradigm uniformity. However, the theory of
Optimal Paradigms does provide a way of formalizing why the
causative is special. As shown by the data in (19), other
derivational stem suffixes, including the phonologically similar
passive glide /-w-/ (see (10), above), trigger no phonological
alternations in the stem to which they are attached:
(19) Phonologically and semantically compositional stem

suffixes
Base Gloss Derived Gloss

infinitive form
a. oku=gur-a ‘to oku=gur-ir-a ‘to buy for’

buy’
oku=gur-ir- ‘to buy for

an-a each other’
b. oku=sír-a ‘to oku=sir-ír-a ‘to send for’

send’
oku=sír-w-a ‘to be sent’
c. oku=tér-a ‘to put’ oku=ter-ér-a ‘to put for’
d. okw=á:t-a ‘to oku=a:t-ík-a ‘to be

break’ breakable’
That is, other derivational stem suffixes form a uniform

paradigm with their morphological base stem, satisfying the
general OP constraint in (20).
Page 26 of 46
(20) OP: MAX/DEP DStem

The DStem in each paradigm member must contain
all the segments (MAX) and only the segments
(DEP) found in the corresponding position in the
other paradigm members.
(p.136) The causative, however, cannot satisfy this more

general OP constraint, due to high-ranking *ry (12a), the
markedness constraint motivating spirantization of the final ‘r’
of the stem to which the causative attaches. This is illustrated
in the tableau in (21), essentially identical to (18) except the
general OP constraint in (21) has replaced the Causative Stem
specific OP-CS (17). The string evaluated by OP (20) is
underlined in each candidate set.
(21)
Page 27 of 46
/oku=gur-a/, / *ry ALIGN /y/ op MAX/DEP DEP/y/ CONTIGUITY

oku=gur-y-a/ / [CONT]
oku=gur-y-an-a/
a. oku=gur-a, *
oku=gus]y]-a, *
oku=gus-y-an] *! * *
y] -a
b. oku=gur-a,
oku=gur]y]-a, *
oku=gur-y-an] - *! *
a
c. oku=gur-a,
oku=gur]y]-a, *!
oku=gur-an]y]-a *
d. F oku=gur-
a,
oku=gus]y]-a, * *
oku=gur-an]y]-a *
Page 28 of 46
In (21) we can see that doubling (21a) is non-optimal, as

doubling makes each Causative Stem less similar to the Base
Stem evaluated by OP (20). However, not doubling, as in the
optimal candidate (21d), also fails to satisfy OP. Indeed, the
only candidates which do satisfy OP are non-optimal as they
necessarily violate high-ranked *ry and ALIGN /y/.
Even though causatives cannot consistently fit into a uniform

paradigm with their non-causative base due to spirantization,
as shown in (18), above, doubling allows related causative
stems to be uniform within their own distinct paradigm. This
analysis predicts, then, that there is a close connection
between spirantization and doubling. Spirantization makes the
causative stem distinct from the non-causative stems derived
from the same base. As doubling enhances the distinctiveness
of the causative (p.137) paradigm, we would expect to find
doubling only in languages where the causative is already
distinct due to spirantization. This prediction remains to be
tested. We would also expect that once spirantization defines a
distinct paradigm for the causative, the causative might
become more distinct from the base stem in other ways. This
prediction is borne out by the Jita data. While only the [r]
productively spirantizes, there are some exceptional verbs
ending in other consonants that also spirantize before the
causative, as shown in (22b), making spirantization a partly
unpredictable process (22).
(22) a. Regular spirantization of ‘r’ before the Causative

(oku- is the infinitive prefix)
Base infinitive Gloss Causative
oku=ku:rír-a ‘to weed rice’ oku=ku:rís-y-a
okw=é:r-a ‘to be clean’ okw=é:s-y-a
oku=lir-a ‘to cry’ oku=lis-y-a
oku=kor-a ‘to do’ oku=kos-y-a
oku=βúr-a ‘to get lost’ oku=βús-y-a
oku=sír-a ‘to send’ oku=sís-y-a
oku=te:m–ér- ‘to cut for’ oku=te:m–és-y-a

a
oku=kum–ír-a ‘to heap for’ oku=kum–ís-y-a
Page 29 of 46
b. Irregular spirantization of other consonants
Base infinitive Gloss Causative
okw=á:t-a ‘to break’ okw=á:s-y-a
oku=βat-a ‘to be sticky’ oku=βas-y-a
okw=ó:g-a ‘to okw=ó:s-y-a

bathe’ (intrans)
BUT
oku=lé:t-a ‘to bring’ oku=lé:t-y-a ~

oku=lé:s-y-a
oku=yag-a ‘to scratch’ oku=yag-y-a
oku=βu:t-a ‘to be high (of oku=βu:t-y-a ~

meat)’ oku=βu:s-y-a
oku=fúg-a ‘to paddle’ oku=fúg-y–a
In contrast, there is no irregular phonology associated with

the other derivational suffixes of the language.9
Other researchers have pointed to the importance of

phonological alternation, especially irregular phonological
alternation, in defining particular forms as potential bases for
paradigm uniformity. Burzio (1996, 2000b, 2002b, Ch. 4 above)
and Steriade (1999a) propose that the degree of phonological
uniformity imposed on a paradigm depends on the
transparency of the derived words’ relation to each other. The
more phonologically similar morphologically related words
are, the more likely they are to impose paradigm uniformity.
The less similar they are, and the less predictable the
differences are, the less likely related words are to attract
each other into paradigms. As Burzio (2000b, Ch. 4 above)
points out, following Flemming (1995), this is the (p.138)
morphological equivalent of the principle, familiar from
phonetic studies of perceptual contrast, that entities that are
already similar tend to become more similar, while entities
that are already distinct tend to become maximally distinct.
Recent work outside the OT framework, like Bybee (1985,
1996) also argues that it is autonomous forms that are the
most likely bases for paradigms. One of the factors she argues
determines autonomy is morpho-phonemic irregularity. If the
output of a morphological process is irregular, it has to be
stored as a lexical item. This is what makes it likely to
Page 30 of 46
influence the pronunciation of related words in the paradigm.

These proposals predict that the causative, as the only
derivational suffix that changes the pronunciation of the base
stem, is not likely to form a paradigm with the base stem.
Instead, it is likely to become even more distinct from the base
by imposing uniformity on related causative stems. As all the
other derivational stem affixes, including the phonologically
similar passive /-w-/, never change the pronunciation of the
stem to which they are affixed, they are not predicted to define
their own paradigms distinct from the base stem.
The distinction between the paradigmatic status of the

causative compared to the other derivational stem suffixes can
be seen as similar to the distinction proposed for English
between Class 1 affixes (like -ity) and Class 2 affixes (like -
ness). As argued in Anshen and Aronoff (1988), Aronoff (1976),
and Burzio (1996), words formed with Class 1 affixes have
fewer lexical connections to their base stems due to their
frequently irregular and non-compositional phonological and
semantic relationship with the base. Class 2 affixes have a
stronger lexical connection, as they are phonologically and
semantically transparently related to their base stems. As a
result, Class 1 affixes, like causative stems, often form
separate paradigms from their base stems for processes like
stress assignment, while Class 2 affixes never do. For example,
when Class 2 affixes are added to démonstrate—e.g.
démonstrator, démonstrating—there is no change in the stress
of the base stem. However, when a Class 1 affix like -ive is
added, the stress changes—e.g. demónstrative—and creates
paradigmatic pressure influencing stress assignment to
related words like demònstrabílity. (See Burzio 1996, Ch. 4
above, and Steriade 1999a for discussion.)
To sum up this section, the OP analysis solves the main

problems raised by the derivational cyclic analysis of causative
doubling. The requirement that the Base be a phonologically
well-formed substring of all members of the paradigm
correctly downplays the role of shared morpho-syntactic
constituency in defining paradigms. This is important as
causative stems are morphologically ill-formed due to a high-
ranking phonological constraint ((13) ALIGN /y/) on the
position of the causative glide. The OP analysis also highlights
what is distinctive about the Causative Stem that leads it to be
the only derivational suffix associated with a paradigm distinct
from that of the Base stem. As the causative morpheme is
Page 31 of 46
associated with a phonological process, spirantization, which

alters the output pronunciation of its base, Causative Stems
cannot have a uniform pronunciation with the other
derivational stems associated with that Base. The distinctness
of Causative Stems from the Base Stem paradigm is then
enhanced by doubling, as it enforces uniformity within the
causative paradigm at the expense of uniformity with the base
stem paradigm.
Page 32 of 46
(p.139) 6.3.3 How OP improves on a TCT analysis of doubling

Although we have seen that OP provides a straightforward
analysis of causative doubling, McCarthy (Ch. 8 below)
suggests that OP is only appropriate for inflectional
paradigms. The priority of the base effects associated with
derivational paradigms such as causative doubling are best
accounted for within Transderivational Correspondence
Theory (TCT; Benua 1997a). In this section, I show that OP can
easily account for the priority of the base in causative
doubling. I show further that the TCT analysis of Bemba
causative doubling sketched in Benua (1997a) encounters the
same sorts of problems that, as McCarthy (Ch. 8) argues,
make TCT unable to account for inflectional paradigms.
In TCT, paradigm sets are not conceived of as constellations of

related words as they are in OP. Instead, each member of a
paradigm set must be the morphological base of the adjacent
member of the set, differing from it by a single morphological
operation. The morphological base of a paradigm is also the
phonological base. In TCT, uniformity is evaluated for
morphologically adjacent pairs of words in a paradigm using
recursive tableaux schemas, rather than over the entire
paradigm in a single tableaux as in OP and other paradigm
constellation theories (e.g. Burzio 1996, et seq.; Kenstowicz
1996, 1997). For example, in the paradigm set—órigin,
oríginal, orìginálity—a paradigmatic correspondence can be
established between the stress of orìginálity and that of
oríginal because only one morphological operation (suffixation
of -ity) distinguishes the two words. However, the stress of
orìginálity cannot be in correspondence with that of órigin
because two morphological operations distinguish those two
words. Benua (1997a) argues that an advantage of this
hierarchical conception of paradigms is that it formalizes the
‘priority of the [morphological] base’ in conditioning the
pronunciation of related, morphologically complex words. It is
the morphological base which conditions the pronunciation of
words derived from it. Derived words do not condition the
pronunciation of their base. For example, the stress of oríginal
influences stress placement in orìginálity, but we would not
expect the reverse direction of influence.
However, as Kenstowicz (1996, 1997) has shown, priority of

the base can also be optimized in a constellation paradigm
theory by ranking the relevant OP constraint below the
markedness and/or faithfulness constraints defining the
Page 33 of 46
process generalized to give Paradigm Uniformity. Since the

priority of the base falls out from the constraint rankings in
the OP analysis, there is no need to adopt an alternative
formalism, TCT, to explicitly encode this aspect of doubling.
This is illustrated in (23). (Note that both OP-CS (17) and OP
(20) are included in this tableau. The string evaluated by OP is
capitalized; the string evaluated by OP-CS is underlined.) In
(23) OP-CS (17) outranks OP (20). As a result, candidate
paradigm (23a), where the Causative Stems are uniform with
respect to each other but not uniform with respect to the
underived stem, remains optimal. While candidate (23b)
satisfies both OP-CS and OP, it is non-optimal as it violates the
high-ranked markedness constraint, *ry.
Comparing candidate paradigms (23c) and (23d) shows how

this constraint ranking (p.140) derives priority of the base.
Candidate (23d) satisfies OP but gratuitously violates MAX/
DEP [CONT], because uniformity comes at the expense of
spirantizing an ‘r’ that is not in the marked ‘ry’ context. The
ranking MAX/DEP [CONT] >> OP ensures that spirantization
from a derived form will not be generalized to a base that does
not meet the context for spirantization. Candidate (23c) is non-
optimal as it violates OP-CS. This candidate shows crucially
that ranking OP-CS above the faithfulness constraints MAX/
DEP [CONT] and DEp/y/ ensures that alternations like
spirantization (and doubling) will ‘cycle’ outwards from the
innermost string providing the context for the constraints
motivating the alternation generalized throughout the
paradigm.
(23)
Page 34 of 46
/oku=gur-a/, / *ry ALIGN /y/ OP-CS MAX/DEP DEP/y/ OP CONTIGUITY

oku=gur-y- [CONT]
a/ /oku=gur-
y-an-a/
a. F oku=gur- *
a,
oku=GUS]y]- *
a,
oku= GUS-y- * * *
an]y]-a
b.
oku=GUR-
a,
oku=GUR]y]- *!
a,
oku= GUR-y- * * *
an]y]-a
c.
oku=GUR-
a,
oku=GUS]y]- * *
a,
Page 35 of 46
oku= GUR- *! *
an]y]-a
d. *!
oku=GUS-
a,
oku=GUS]y]- *
a,
oku= GUS-y- * * *
an]y]-a
Page 36 of 46
Crucial constraint
rankings:
*ry, OP-CS » OP optimizes uniformity within the

Causative Stem to satisfy *ry at the
expense of non-uniformity with the
base stem
OP-CS » MAX/ optimizes outward cycling of

DEP [CONT], alternations
DEP/y/
MAX/DEP penalizes changes in [cont] to satisfy

[CONT], >> OP uniformity
(p.141) Another difference between OP’s definition of a

paradigm and TCT’s definition is that in constellation
paradigm theories like OP the base can be a phonological
substring shared by words in the paradigm. The shared base
string need not be a word itself (though, of course, since
paradigms consist of outputs, each member of the paradigm is
a word). In contrast, in TCT each shared base string must
itself be a word. This predicts correctly for languages such as
English that bound roots like -ceive which are not the domain
for any phonological processes also cannot be the Base for
Output–Output correspondence. (See Brame 1974 and
Kenstowicz 1996 for arguments that the word vs non-word
distinction is crucial in other languages, too.) However, the
insistence on the primacy of the entire word as the Base for
correspondence in TCT leads to a paradox in accounting for
paradigm uniformity in highly inflected languages for the
following reason. On the one hand, the two members of a sub-
paradigm in correspondence must be compositionally related:
the phonological base must also be the morphological base,
with Output–Output correspondence introduced by the
morphological operation that links the two forms. On the
other, the two members must be words, each fully inflected,
even though one fully inflected word often cannot be the
morphological base for another. Instead, each word in an
inflectional paradigm is derived by affixing the appropriate
inflection to a shared stem. (See McCarthy, this volume, for
further discussion of this point.)
Page 37 of 46
This problem is clearly illustrated by Jita and Bemba causative

doubling. In all Bantu languages, verb derivational stems (root
plus derivational suffixes) must occur with an inflectional final
suffix and inflectional prefixes indicating, minimally, subject
and tense/aspect/mood.10 Since there are around twenty
subject prefixes and at least a dozen tense/aspect/mood
prefixes (plus a negative prefix and twenty object prefixes) and
three inflectional final suffixes, each verb stem can occur in
literally hundreds of inflected words. As shown in a selection
of these forms in (24), causative doubling is completely
insensitive to the inflectional morphology of the word in which
the Causative PStem occurs. Uniformity is imposed only within
the Causative PStem (underlined).
(24) (Some) inflected forms of the Jita causative stem, /

=βon-y-a/ ‘show’ (tone is omitted)
a. Verb lexeme /=βon-y-a/ ‘show’
oku=βon-y-a ‘to show’
a:ma=βon-y-a ‘s/he showed

(yesterday) ’
ga=βon-y-e! ‘show it (cl.6)!’
a-ta:=βon-i:sye ‘s/he did not show (long

ago)’
ca:=βon-i:sye ‘we showed (long ago)’
mu-li-βi=βon-y-a ‘you pl. will show them

(cl.8)’
b. Verb lexeme /=βon-y- ‘show to’

es-y-a/
oku=βon-y-es-y-a ‘to show to’
βa=βon-y-es-y-e ‘show to them!’
ca:ma-mu=βon-y-es-y-a ‘we showed you pl.

(yesterday)’
eni=βon-y-es-y-a ‘I am showing’
a-ta-ku=βon-y-es-y-a ‘s/he is not showing’
(p.142) The choice of one inflected form as the base for

doubling would be completely arbitrary, as the inflections play
no role in determining the pronunciation which is generalized
Page 38 of 46
by doubling. However, to be consistent with the TCT approach,

some particular inflected form must be chosen as base. If,
though, oku=βon-y-a (the infinitive) is chosen as the Base for
the causative forms in (24b) which show doubling, then more
than one morphological operation must be performed to relate
that base to most of words in (24b) which would be part of the
same paradigm: one to add the derivational suffix, and at least
one other to add the appropriate inflectional affixes. This
conflicts with the TCT definition of a Base as being separated
by only one morphological operation from the other member of
the subparadigm. In fact, the extensive inflection occurring on
Bantu verb words creates an insurmountable problem for any
theory of doubling which requires the phonological base of
uniformity to be an inflected word that is in its entirety a
morphological subconstituent of the other member(s) of the
paradigm. OP and other ‘constellation’ paradigm theories
avoid this problem because the constraints enforcing
uniformity define a shared substring of the words in the
paradigm as the phonological base. This is an advantage for
inflectional languages such as Bantu languages (and Arabic
(McCarthy, Ch. 8)), since it is obvious that a compositional
relationship holds only among related stems and not among
the fully inflected words containing these stems.
It is also not possible to account for doubling within TCT by

proposing that the causative stem is also a word, and
modifying TCT so that bases are required to be either
phonological words or morphological words. This is because,
in most Bantu languages, stems are demonstrably neither
morphological words (MWords) nor phonological words
(PWords; see e.g. Downing 1999a; Myers 1987a). For example,
in Jita we find productive processes sensitive to word
boundaries: long vowels are shortened PWord finally and
adjacent input high tones (such as on a monosyllabic verb and
a following high-toned enclitic or prepositional phrase) are
tolerated across MWord boundaries (indicated by ‘-’). This is
illustrated in (25).11
Page 39 of 46
(25) Jita word edge alternations
a. No long vowels PWord-finally
Infinitive Gloss Infinitive+enclitic Gloss
oku=gwa ‘to fall’ oku=gwa:-mó ‘to fall in’
oku=gusya ‘to sell’ oku=gusya:-kó ‘to sell some’
b. Sequences of input High tones tolerated across MWord boundaries
oku=tá ‘to put’ oku=tá-mó ‘to put in’
oku=tá ‘to put’ oku=tá nánye ‘to put by/with me’
Page 40 of 46
(p.143) However, as shown in (26a), long vowels regularly

occur preceding the stem, and sequences of adjacent high
tones are not tolerated across the prefix=stem boundary, as
shown in (26b).
Page 41 of 46
(26) Jita stem edge alternations
a. Long vowels pre-stem
/mu-a=βóna/ → mwai= βóna ‘you pl. saw today’

(*mwa=βóna)
/u-a=lyá/ → wa:=lyâ (*wa=lyá) ‘you ate today’
/a-tá-a=gwa/ → a-tá:-gwa (*a-tá=gwa) ‘s/he did not fall today’
b. Sequences of high tones not tolerated across stem boundary
/oku-mú= βóna/ → oku=mú-βona (*oku- ‘to see him’

mú=βóna)
/u-amá=yá:na/ → wa:-ma=yá:na (*wa:- ‘you gave yesterday’

má=yá:na)
/a-amá=lyá/ → a:-ma=lyá (*a:- ‘s/he ate yesterday’

má=lyá)
Page 42 of 46
This contrast between word edge and stem edge phonology

falls out if the stem is not a separate phonological or
morphological word from the prefixes, but rather a
subconstituent of the word containing the prefixes. It is totally
unexpected if the stem is a separate phonological word. As a
result, I conclude the causative stem, like other stems, is not a
phonological word but a subordinate morpho-prosodic domain,
the PStem. This is not a surprising result. In fact, there is a
well-established body of literature (Czaykowska-Higgins 1996,
1998; Downing 1999a, 2000, 2001; Kiparsky 1982a, 2000;
Mutaka 1994; Myers 1987a, etc.) demonstrating that, cross-
linguistically, we need to identify both the stem and the word
as possible phonological domains. If an important criterion
identifying possible bases for paradigm uniformity is that they
must independently be phonological domains, then causative
stems clearly meet this criterion even though they are not
words.
6.4 Conclusion
To sum up, I have shown that Causative Doubling is best
accounted for as a form of Output–Output correspondence
enforcing Optimal Paradigms (McCarthy, ch. 8 below). The
analysis supports the constellation approach to paradigms that
defines bases as the phonological strings shared by a set of
morphologically related words. As we have seen, the shared
string for doubling is not a well-formed morphological
subconstituent of the words in the paradigm, instead it is a
morpho-prosodic constituent, the Causative Stem. It is also not
an independent word, but instead a bound stem. Even though
McCarthy (Ch. 8 below) has suggested that TCT, rather than
OP, (p.144) better accounts for derivational paradigms, these
two properties make doubling, a derivational process,
unformalizable in the TCT theory of Output–Output
correspondence. As I have shown, though, there are good
reasons also to adopt OP instead of TCT to account for
derivational paradigms. The priority of the base requirement
characteristic of derivational paradigms can fall out from
appropriate constraint rankings in OP. The TCT formalism is
not needed to express this restriction. Further, when
derivational paradigms are embedded in a highly inflectional
system, like that found in Bantu languages, TCT cannot
maintain its claim that each paradigm member is both a fully
inflected word and separated by only one morphological
operation from its base. Finally, I have shown that OP provides
Page 43 of 46
a way of formalizing why the causative, and not other derived

stems, forms a distinct paradigm from the base stem. As we
have seen, the causative cannot consistently satisfy uniformity
within the base stem paradigm due to spirantization. Doubling
ensures uniformity within a separate Causative paradigm,
generalizing spirantization and enhancing the distinctness of
Causative stems.
Notes:
(1) Notice that the causative is realized as half a long vowel
before an [i]-initial morpheme like the applicative /-ir-/ and as
a glide (triggering palatalization or spirantization of the
preceding consonant) before other vowels. This falls out from
the standard assumption that [i] and [y] have identical
features, and the realization as a vowel or a glide is
determined by syllable structure. Note, too, that vowels are
regularly (compensatorily) lengthened following glides in Jita,
except word-finally. The orthographic sequence ‘ny’ in the Jita
data is realized as a palatal nasal segment, and compensatory
lengthening is not found following this nasal. See Downing
(1996) for discussion and analysis.
(2) There are some differences between causative doubling in

Jita and in Bemba. First, all preceding consonants, not just ‘r’,
spirantize in Bemba. Also while the reflex of the causative
generally remains as a palatal off-glide in Jita, in Bemba the
causative glide is absorbed by all the preceding fricatives
except the rightmost. Finally, Bemba has some exceptions to
doubling which are not found in my Jita data. Discussing these
differences is beyond the scope of this chapter. The interested
reader is referred to Hyman (1994) for further details.
(3) Notice that in Kinande, as in Jita (2ef), the /-y-/ occurs

following the stem-final consonant even though this involves
interrupting the perfective suffix /-ire/. See Hyman (1993,
1994, 2003); Hyman and Mchombo (1992); and Poletto (1998)
for discussion of similar data in other Bantu languages.
(4) See Hyman and Mchombo (1992) for detailed discussion of

scope problems raised by morpheme doubling and other
processes that lead morphemes to occur in positions that are
incompatible with semantic compositionality
Page 44 of 46
(5) See Downing (2001) for more detailed discussion of the

phonetic motivation for ‘*ry’ and of the morphological
conditions on causative spirantization in Jita. Hall (2000)
provides ample evidence for the cross-linguistic markedness of
‘ry’ sequences.
(6) Space does not permit reiterating all the arguments for
PStem as an important domain for phonological processes and
phonotactic generalizations here. The interested reader should
consult Downing (2001), Hyman (1998), and Hyman and
Inkelas (1997) for discussion.
(7) As an anonymous reviewer points out, each displaced ‘y’

incurs more Contiguity violations than the single one indicated
in the tableaux that follow. In (16a), for example, there are,
strictly speaking, three violations—‘ri’, ‘sy’, ‘ya’—and in (16b,
d) there are two—‘s/ry’ and ‘ya’. Since Contiguity is too low-
ranked for these extra violations to play a role in the analysis,
I indicate only one violation per displaced ‘y’ to save space.
(8) I am crucially assuming that ALIGN /y/ is satisfied if there

is a causative /y/ at the right edge of the PStem and violated
only if there is no causative /y/ at the right edge of the PStem.
(9) See Hyman (2001, 2003) for discussion of other Bantu

languages where spirantization has distanced the causative
stem from its Base in sometimes extremely striking ways.
(10) Note, though, that no segmental inflectional prefixes

occur in the imperative in most Bantu languages. As in many
languages, no overt subject is required in the imperative, and
in Jita, for example, only tone indicates the imperative mood.
(11) The Jita tone patterns cited in (25) and (26) abstract away
from processes of final contouring and tone shift for ease of
exposition. See Downing (1996) for detailed discussion of Jita
tone. Enclitics like /-mó/ are not included in MWord. For this
reason, High tone sequences are tolerated over the MWord-
Enclitic boundary. However, they are included in PWord and
count as the final element in PWord.
Page 45 of 46
Page 46 of 46
Paradigmatic Uniformity and Contrast


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Paradigmatic Uniformity and

Contrast
Michael Kenstowicz
DOI:10.1093/acprof:oso/9780199267712.003.0007

This chapter reviews several examples from the contemporary
and traditional literature that argue for systematic
phonological relations between members of a paradigm that
do not satisfy the containment relation of the cycle.
Specifically, the output form assigned to a given word depends
crucially on a related word in the paradigm that cannot
plausibly be treated as an intermediate stage in the derivation.
Phenomena of this form are expected if Output–Output
faithfulness relations are present in the grammar. But they are
surprising from a cyclic point of view in which one word is
derived essentially in isolation from its kin.
Keywords: grammar, pairs, paradigmatic uniformity, paradigmatic contrast,

Spanish diminutive allomorphy
This chapter reviews several cases where either the grammar

strives to maintain the same output shape for pairs of inflected
words that the regular phonology should otherwise drive apart
(paradigmatic uniformity) or where the grammar strives to
Page 1 of 45
keep apart pairs of inflected words that the regular phonology

threatens to merge (paradigmatic contrast).
Page 2 of 45
7.1 INTRODUCTION
The general research question that this chapter addresses is
the proper treatment of cases of opacity in which the
triggering or blocking context for a phonological process is
found in a paradigmatically related word. Chomsky and Halle’s
(1968) discussion of the minimal pair comp[∂]nsation v.
cond[ε]nsation is a classic example of the problem. In general,
the contrast between a full vowel v. schwa is predictable in
Eng- (p.146) lish as a function of stress; but comp[∂]nsation
v. cond[ε]nsation have the same σ̀σσ̀σ stress contour and thus
raise the question whether English schwa is phonemic after
all. SPE’s insight was that the morphological bases from which
these words are derived provide a solution to the problem:
comp[∂]nsate has a schwa while cond[ε]nse has a full stressed
vowel. Chomsky and Halle’s suggestion is that such
paradigmatic relations among words can be described by
embedding the derivation of one inside the derivation of the
other. In other words, the derivation of complex words
proceeds cyclically with the morphological base forming an
intermediate stage /kómpensé:t/ v. /kóndéns/ over which stress
and vowel reduction are computed and relies on the model’s
serial architecture for subsequent destressing processes to
obscure the context for vowel reduction. The cycle became the
basic (and essentially only) tool for describing paradigmatic
phonological relations in the generative framework. It
received considerable theoretical development in the Lexical
Phonology model Kiparsky (1982a, 1985). While many
languages have been successfully described in cyclic terms
(see Cole 1995 for a recent review), cases were discovered
where the cycle does not provide the proper coverage or
elucidation (e.g. S. Chung 1983; see also Crosswhite (1997).
For constraint-based models of phonology that eschew

structure building and structure changing rules in favor of
static well-formedness conditions evaluating output forms,
such ‘cyclic’ phenomena present a real challenge. Within
Optimality Theory there have been two contrasting
approaches to the problem. The first (pursued by
Kiparsky2000; cf. also Kenstowicz 1995; Booij 1996; Rubach
2000) retains Lexical Phonology’s modular conception of the
grammar but trades in rules for constraints and a one-step
Input–Output mapping within each module. The hypothesis is
that a minimal number of intermediate levels (ideally
motivated independently by the morphology of the language)—
Page 3 of 45
root, stem, word—will provide just enough leverage to handle

all types of opacity. (See Goldsmith1993for a similar proposal.)
The second approach has been to extend OT faithfulness
constraints from Input–Output and Base–Reduplicant relations
to related words in a paradigm in the form of so-called Output–
Output constraints. This approach has been pursued by Benua
(1997a); Bur-zio (1996); Crosswhite(1998); Flemming (1995);
Itô and Mester (1997); Kager (1999b); Kenstowicz (1996);
Kraska-Szlenk (1995); McCarthy (8 below); Raffelsiefen
(1995); Steriade (2000); and others. The general idea is that
words sharing the same morphological base form a network of
possible phonological influences. The research program is to
document these influences, to characterize their nature, and
to develop appropriate formalisms to express them. Of
particular interest is whether cases can be shown to exist that
do not meet the containment condition of the classical SPE
cycle in which the derivation of one word is embedded in the
derivation of the other. Of course, analogical relations of this
form are well-known from traditional historical linguistics.
Genuine cases of grammar change based on the idea that
certain allomorphs have a privileged status certainly exist: see
Lahiri and Dresher (1984) for discussion. The question is
whether such paradigmatic relations play an active role in the
synchronic grammar in computing Input–Output mappings or
instead are more properly viewed as the product of the
learning module for constructing underlying/base forms with
par- (p.147) ticular members of the paradigm having a
privileged status because of their frequency, their perceptual
salience, their relative markedness, etc.
This chapter presents some additional examples of

paradigmatic relations that are arguably part of the
synchronic system rather than simply the residue of a
diachronic restructuring process. Once a critical number of
such cases have been documented and described, we can
begin to provide a theoretical account of the phenomenon. In
s. 7.2 we examine a hitherto overlooked aspect of the
distribution of the allomorphs of the Spanish diminutive suffix
that exhibit a striking example of paradigmatic uniformity. We
then document cases of paradigmatic contrast from Russian,
Bulgarian, Ch-Mwi:ni, and Arabic in which the phonology
conspires to ensure that underlyingly distinct members of the
inflectional paradigm remain phonetically distinct. The
Page 4 of 45
chapter concludes with a summary of the findings and tasks

for future research.
7.2 SPANISH DIMINUTIVE ALLOMORPHY

It is well known that the Spanish diminutive suffix has two
allomorphs -sit-o/a and -it-o/a (see Crowhurst 1992; J. W. Harris
1992;1994; and Elordieta and Carreira 1996 for recent
discussion). Our remarks here are based on the observations
of Aguero-Bautista (1998;). Scrutiny of the data in (1) indicates
that the -sit allomorph is chosen when the nondiminutive base
word ends in [r] or [n]. -it-o/a is chosen when the base word
ends in a vowel. This distribution does not conform to the
widespread tendency for C≈V allomorphy to be resolved by
syllable markedness constraints optimizing CV sequences at
the juncture between the base and the affix (see Mascaró 1996
for discussion).
(1)
a. amor amor-sit-o ‘love’
balkon balkon-sit-o ‘balcony’
limon limon-sit-o ‘lemon’
b. koron-a koron-it-a ‘crown’
libr-o libr-it-o ‘book’
bark-o bark-it-o ‘ship’
cokolat-e cokolat-it-o ‘chocolate’
lava-dor- lava-dor-it-a ‘washing machine’

a
seca-dor- seca-dor-it-a ‘dryer’

a
Aguero-Bautista suggests that the allomorphs are distributed

so as to maintain the syllabic profile of the nondiminutive
source: -sit nudges the preceding consonant into the coda so
that [a.mor.si.to] matches [a.mor] while -it draws the final
consonant of the stem into the onset so that [ko.ro.ni.ta]
matches [ko.ro.na]. The tableaux in (2) illustrate the idea with
a constraint demanding corresponding syllabic roles: if x and y
are corresponding segments then x and y have the same
syllabic analysis (onset, nucleus, coda).
Page 5 of 45
(p.148)
(2)
a. /amor-sit≈-it-o/
CORR-Σ-ROLE
a. ?a.mor.-si.t-o
b.a.mo.r-i.t-o *!
cf. a.mor
b. /koron-sit≈-it-a/
CORR-Σ-ROLE
a.koron-sit-a *!
b. ?koron-it-a
cf. ko.ro.n-a
The data in (3) show cases where the syllable-matching

generalization unexpectedly breaks down: -sit occurs even
though the nondiminutive source ends in a vowel and hence
the base-final consonants have divergent Σ-roles: e.g. the [n]
in rat-on-a is an onset but occupies the coda in rat-on-sit-a.
CORR-Σ-ROLE demands rat-on-ita.
(3)
rat-on-a rat-on-sit-a ‘mouse’
ladr-on-a ladr-on-sit-a ‘thief’
yoron-a yoron-sit-a ‘cry baby’
mandon-a mandon-sit-a ‘bossy’ f.
trabaja-dor-a trabaja-dor-sit-a ‘worker’ f.
Aguero-Bautista’s explanation for the difference between (1b)

and (3) is that the latter enter into a paradigmatic relation
with the corresponding masculine forms in (4). The masculines
take -sit by Σ-ROLE CORR.
(4)
rat-on rat-on-sit-o ‘mouse’ m.
Page 6 of 45
ladr-on ladr-on-sit-o ‘thief’ m.
yoron yoron-sit-o ‘cry-baby’ m.
mandon mandon-sit-o ‘bossy’ m.
trabaja-dor trabaja-dor-sit-o ‘worker’ m.
The reality of the phenomenon is demonstrated by the fact

that the diminutive of lavador-a changes from lavador-it-a
‘washing machine’ to lavador-sit-a when it has the meaning of
‘washer-woman’ (paired with lavador-sit-o ‘washer-man’). The
generalization is thus that the syllable matching allomorphy is
overridden in favor of a fixed shape for the stem in the
subparadigm formed by the paired masculine and feminine
inflection. A comparable example occurs in Polish diminutives
(which level the o~u alternation that regularly appears in the
inflection of the base word; for discussion see Kenstowicz
1996based on Kraska-Szlenk (1995).
(p.149) The Spanish example is interesting because there

are two conflicting paradigm uniformity constraints at play.
The first requires the diminutive stem to parallel the
syllabification of the source and hence demands different
allomorphs for the diminutive suffix when it is added to bases
with zero inflection in the masculine. But a countervailing
second force imposes a uniform stem shape for the paired
masculine/feminine inflection. The latter constraint wins out
here.1 Finally, uniformity in the inflectional paradigm could be
achieved by generalizing either the -sit or the -it allomorph.
Leveling of -sit could reflect its status as the default allomorph.
Alternatively, the choice might be resolved on the basis of the
unmarked status of the masculine—comparable to the
privileged status of singular (v. plural), nominative (v. oblique),
present (v. past), etc. as models for analogical generalization.
The overall analysis is sketched in (5). Following Flemming
(1995); Raffelsiefen (1995); and others, the PARADIGM
UNIFORMITY constraint evaluates alternative paradigms.
(5)
PARADIGM CORR-Σ- *-IT

UNIFORMITY ROLE »*-
SIT
Page 7 of 45
a. ? {ra.ton.- *
si.t-o, ra.ton.-
si.t-a}
b. {ra.to.n-i.t- * *!*
o, ra.to.n-i.t-
a}
c. {ra.ton.- *!
si.t-o, ra.to.n-
i.t-a}
It should be clear that the classical cycle will have serious

problems with this example. Since it derives each word in
isolation from any other one (except to the extent that one is a
substring of the other) it cannot express the contrast between
(1b) and (3) in a natural way.2 More generally, the cycle’s
success reflects some of the idio- (p.150) syncrasies of
English morphology, which is poor in inflection and where it
does have inflection, the base is typically an independent word
with no (overt) affixation and hence can be treated as a
substring of the other form(s) of the paradigm.
7.3 PARADIGMATIC CONTRAST
Page 8 of 45
In the remainder of this study we examine several cases in

which the phonology conspires to ensure that two
phonologically distinct members of the paradigm remain
phonetically distinct. Contrast constraints of various types
have been discussed in the recent generative literature. Let us
briefly mention some of them here in order to situate better
our use of the notion. Syntagmatic contrast (formalized in
terms of the OCP) is a well known determinant of phonological
form. Côté (2000) discusses various examples of consonant
cluster phonotactics that depend on the nature and degree of
phonetic contrast among the elements composing the cluster.
Research into under-specification (see Steriade 1995 for a
review) showed that a feature’s status as marked v. contrastive
can be decisive in determining its phonological behavior (cf.
Calabrese1995). Flemming (1995) points to cases in which the
markedness of a segment cannot be determined independently
from the system in which it occurs, observing for instance that
a high central vowel is typically marked in systems that
contrast front and back vowels but it is the unmarked,
expected vowel in columnar systems such as Marshallese that
contrast vowels for height but not for backness. Yip (1995),
Soh (1996), and Kelepir(1997) document cases in which
corresponding segments in a Base–Reduplicant structure are
required to differ along some phonological dimension such as
vowel height or consonantal place. Alderete (1998) argues for
parallel cases of ‘anti-faithfulness’ along the Input–Output
dimension to mark a morphological contrast: e.g. a switch in
vowel length in Dinka to express the singular v. plural
distinction. Various types of ‘push chains’ have also been noted
(e.g. Kirchner’s (1996) discussion of vowel raising and deletion
in Bedouin Arabic). Finally, Crosswhite (1997) documents
cases of paradigmatic homophony avoidance in Bulgarian and
Russian that are directly relevant to us. In the next section we
recapitulate one of her examples and add several of our own to
demonstrate the reality of the phenomenon.
7.3.1. Russian vowel reduction

Our first example comes from Russian. Most dialects of
Russian including the standard one reduce the underlying five
vowel system /i, e, u, o, a/ to three /i, u, a/ in unstressed
position. When not in the immediate pretonic syllable,
nonhigh /a/ is reduced fur- (p.151) ther to schwa. The
mapping of vowels is such that unstressed nonhigh vowels
Page 9 of 45
appear as high and unround after ‘soft’ (palatalized or palatal)

consonants (so-called ikan’e) and as nonhigh and unround /a/
elsewhere (so-called akan’e).
(6)
Page 10 of 45
k’ij l‘úd’i l’és n’ós m’ása
k’ijá l‘ud’éj l’isá n’islá m’isnój
‘cue’ ‘people’ ‘forest’ ‘carried’ ‘meat’
mí∫ zúp stól ribák
mi∫ám zubám stalá ribaká
‘mouse’ ‘tooth’ ‘table’ ‘fisherman’
Page 11 of 45
Reduction is regular in roots and stems where it can lead to

neutralization of distinct lexical items; but reduction may be
curtailed in inflectional endings. One of the most interesting
cases involves the realization of the 3 pl. ending of second
conjugation verbs according to the Old Moscow norm of
pronunciation. In this variety (described in Jakobson 1948) the
normal, etymologically expected inflection -at is replaced by -
ut when it is unstressed instead by -it in accordance with the
normal ikan’e reduction process. The sample paradigm in (7)
(based on Avanesov 1964) illustrates the phenomenon. Verbs
such as govorít’‘to speak’ have stress on the inflection in the
present tense and hence take -át in the 3 pl. But for verbs such
as kurít’‘to smoke’, where stress remains primarily on the root
when the verb is inflected, the expected 3 pl. form kúr'-it/kúr'-
at/ is replaced by kúr’-ut. Other examples: díf-ut ‘breathe’, útf’-
ut ‘teach’, v’íd’-ut‘see’, lóv’-ut‘catch’
(7)
Page 12 of 45
Singular plural singular plural
1 st g∂var’-ú g∂var’-ím kur’-ú kúr’-im
2nd g∂var’-í∫ g∂var’-ít’∂ kúr’-i∫ kúr’-it’∂
3rd g∂ar’-ít g∂var’-át kúr’-it *kúr’-itkúr’-at/
‘speak’ ‘smoke’ kúr’-ut
Page 13 of 45
Why is just the 3 pl. ending changed and why is it changed

only when it is unstressed? An answer is suggested by looking
at the rest of the paradigm of which it is a member. One of the
hallmarks of the second conjugation in Russian is that the final
consonant of the stem is palatalized or palatal throughout the
inflection and hence will be the site of ikan’e reduction when
the inflectional ending is unstressed. Also, the 3 sg. ending is –
it : kúr’-it ‘smokes’, dí∫-it ‘breathes’, út∫’-it ‘teaches’, v’íd’-it
‘sees’, lóv’-it ‘catches’. It is now easy to see that vowel
reduction (ikan’e), if given full sway, would realize
unstressed /-at/ with a high unround vowel since it follows a
palatalized or palatal consonant. The consequence would be a
merger of the distinction between the singular and plural
forms of the third person for any verb and thus a violation of
PARADIGM CONTRAST (see Rebrus and Törkenczy, 10 below,
for a useful definition). The [u] that we find instead could be
computed by the phonology substituting the only other
unstressed vowel that is permitted after a soft consonant
(violating faithfulness for [round]). Alternatively, the -ut might
be recruited from the first conjugation where the regular 3 pl.
ending is in fact -ut (this is essentially the analysis of Jakobson
1948).
(p.152) Under either analysis the substitution of -ut for -at

must take account of the stress properties of the stem and
hence must be part of the Input–Output computation. Russian
stress is a dominant-recessive system with one stress per word
(see Halle 1997 for recent discussion). Both stems and suffixes
must be marked for one of three categories: stressed, post-
stressed, unstressed. Depending on the particular combination
of morphemes composing a word, all but one stress is deleted;
if the input contains no stress then one is inserted (usually on
the initial syllable). Unless one is willing to build the stress
calculation into the -at ≈-ut allomorphy, the grammar must
wait until the proper stress location for the verb has been
determined and then resolve the choice between the
allomorphs. The tableau in (8) sketches the intended analysis.
The Stress constraint abbreviates a battery of constraints that
delete and insert stresses to achieve a single stress per word
(see Alderete 1998 for a recent OT analysis).
(8)
Page 14 of 45
/kur’-at/ STRESS IKAN’E PARADIGM FAITH [ROUND]

CONTRAST
a.kúr’at *!
b. kur’át *!
c ?kúr’ut *
d.kúr’it *!
Page 15 of 45
cf. kúr’it ‘he smokes’
7.3.2 Bulgarian vowel reduction

Like Russian, Bulgarian has a mobile stress that shuttles
between the stem and the desinence in a lexically determined
class of stems. Bulgarian also has vowel reduction. Crosswhite
(1997) calls attention to Stojkov’s (1963) discussion of the
Trigrad dialect that is particularly relevant here. In this dialect
unstressed /o/ reduces to [a]: dó3d, da3dóm ‘rain’. The
reduction is largely automatic in stems but is inhibited to
various degrees in inflectional endings. One of the more
striking instances of the phenomenon involves neuter nouns.
According to Stojkov (1963: 16–17), ‘In the case of neuter
nouns, it has to be noted that reduction to “a” is found only
when there is no homophony in the forms for singular and
plural… when there is homophony the reduction to “a” is not
found’.3 Stojkov illustrates with the examples in (9). In (9a) the
neuter singular suffix /-o/ is reduced to [a] while reduction is
blocked in (9b).
(9)
a sg. /-o/ pl. /-a/
kapít-a kapit-á ‘hoe’
kláb-a klab-á ‘ball of thread’
pér-a per-á ‘feather’
rébr-a rebr-á ‘rib’
(p.153)
b. zórn-o zórn-a ‘grain, seed’
pétal-o pétal-a ‘horseshoe’
blág-o blág-a ‘blessing’
cigaríl-o cigaríl-a ‘cigarette’
The difference systematically correlates with the mobility of

stress. Nouns that reduce the -o also shift the stress to the
ending in the plural (9a); nouns that block reduction maintain
fixed stress on the stem in both the singular and the plural
(9b). The generalization is thus that reduction of -o is blocked
by Paradigm Contrast (PC) if it would lead to phonological
Page 16 of 45
merger of the singular and the plural forms. Reduction

(penalizing unstressed [o]) is possible in (9a) because the
singular and plural are distinguished by the shift of stress.
(10)
a./zórn-o/ PARADIGM CONTRAST REDUCTION
? zórn-o *
zórn-a *!
cf. zórn-a (pl.)
b./kapít-o/ PARADIGM CONTRAST REDUCTION
kapít-o *!
? kapít-a
cf. kapit-á (pl.)
7.3.3 Chi-Mwi:ni perfect

Our next example of paradigmatic contrast comes from
Kisseberth and Abasheikh’s (1974) discussion of the
phenomenon in the Bantu language Chi-Mwi:ni. It concerns
the perfect and applied forms of the verb. The perfect is
marked by a suffix /-i:l/ while the applied (benefactive) is
marked by /-il/. The vowel of the perfect suffix originates
historically from a super-high i that triggered various changes
in the stem before merging with the simple high vowel of the
applied form. We follow Kisseberth and Abasheikh in assuming
that the processes have been morphologized. The high vowels
of these suffixes lower to mid when the root contains a mid
vowel by a process found in many other Bantu languages. The
flapped lateral of the applied suffix (transcribed here as l)
changes to plain before the perfect suffix whose lateral in turn
harmonizes with that of the preceding applied suffix: /stem-rl -
i:l -e/⇾ /stem-il-i:l -e/⇾/stem-il-i:l-e/.
(11)
infinitive perfect
base x-so:m-a som-e:l-e ‘read’
appl. x-som-el -a som-el-e:l-e
base x-kom-a kom-e:l-e ‘arrive’
Page 17 of 45
appl. x-kom-el -a kom-el-e:l-e
base x-ko:d-a kod-e:l-e ‘talk’
appl. x-ko:d-el-a kod-el-e:l-e
(p.154) We invoke two ad hoc constraints for these

morphophonemic changes: * i:l /__perf bars the flapped lateral
before the perfect suffix while Lateral Harmony requires
successive laterals to agree in apicality. If these constraints
dominate faithfulness for [distributed] then the appropriate
changes can be implemented.
(12)
/-il- *i:L/ LATERAL FAITH-

i:l/ perf HARMONY [DISTR]
ll *!
ll *! *
ll *! * *
? ll **
With this background, we can now turn to the matters of

interest. Several morphophonemic changes are relevant to the
discussion. First, the perfect suffix -i:l regularly ‘mutates’ the
final consonant of the stem with concomitant shortening of the
suffixal vowel: labials and velars are replaced with palatals
and coronal stops spirantize.
(13)
infinitive perfect
base x-su:k-a su:sh-il-e ‘plait’
appl. x-su:k-il -a suk-il-i:l-e
base x-kokot-a kokos-el -e ‘drag on ground’
appl. x-kokot-el -a kokot-el-e:l-e
base x-pa:nd-a pa:nz-il-e ‘climb’
appl. x-pa:nd-il -a pand-il-i:l-e
Page 18 of 45
Thus in the perfect form kokos-el-e from /kokot-i:l -e/ the /t/
has mutated to /s/ and the vowel of the perfect suffix /-i:l/ is
shortened and lowered to mid and thus appears as /-el/. The
perfect form is still distinct from the applied because the latter
does not mutate the final consonant of the stem; also the final
vowels are distinct. In the applied perfect /kokot-il-i:l-e/ kokot-
el-e:l-e the vowel length of the perfect suffix is retained since it
does not immediately follow a mutation site. Both suffixes
lower their vowels as a function of the mid vowel of the root.
Finally, the lateral of the applied suffix switches to plain /l/
before the perfect suffix whose lateral in turn harmonizes with
the preceding base.4
The language has another morphophonemic process

Kisseberth and Abasheikh term ‘Ablaut’, which avoids three
successive laterals by deleting the middle one and (p.155)
contracting the resultant vowel sequence. The process is
restricted to certain morphological contexts and reflects the
imbrication phenomenon discussed by Bastin (1983) and
Hyman (1994). The perfect verbs in (14) illustrate.
(14)
infinitive perfect
x-pulul-a pul -i:l-e ‘shell corn’ pulul-i:l-e
ku-la:l-a le:l-e ‘sleep’
la:l-il-e
We can now turn to the items of interest. Stems ending in the

flapped lateral /l/ mutate to /z/ in perfect; they also avoid three
successive liquids in the applied perfect, as expected (15).5
(15)
infinitive perfect
base ku-mo:l-a mo:z-el-e ‘shave’
appl. ku-mo:l-el-a mol-e:l-e mol-el-e:l-e
base x-pe:l-a pe:z-el-e ‘sweep’
appl. x-pe:l-el-a pel-e:l-e pel-el-e:l-e
Page 19 of 45
Thus, /mo:l-i:l-e/ surfaces as mo:z-el-e by mutation and

shortening of the perfect suffix’s vowel; the suffixal vowel also
agrees in height with the root vowel. But stems ending in
plain /l/ do not mutate in perfect, as shown in (16). We have
sul-i:l-e not suz-il-e.
(16)
infinitive perfect
base x-su:l-a sul-i:l-e ‘want’
appl. x-su:l-il-a sul-il-i:l-e *sul-i:l-e
Base x-kal-a kal-i:l-e ‘press’
appl. x-kal-il-a kal-il-i:l-e *kal-i:l-e
Base x-komel-a komel-e:l-e ‘look’
appl. x-komel-el-a komel-el-e:l-e *komel-

e:l-e
The descriptive generalization of interest is that in the perfect

of the applied form verbs ending in plain /−l/ systematically
fail to ‘ablaut’ and instead surface with three laterals in a row.
Thus, we have sul-il-i:l-e ‘want’ instead of *sul-i:l-e.Why is the
otherwise regular process of liquid deletion (ablaut)
suspended here? The answer, as Kisseberth and Abasheikh
observe, is that if these verbs did ablaut the resultant form
would be the same as the basic perfect. This is evident from
examination of the paradigms in (16).
The generalization is thus that mutation of the flapped /l/ to z

in the verbs of (15) such as mo:z-el-e renders the simple
perfect form sufficiently distinct so that lateral (p.156)
deletion (ablaut) can apply in the applied perfect to produce a
form that would be identical to the corresponding simple
perfect if the latter had not undergone mutation. If the
constraint against paradigm merger (PARCONT) outranks
lateral deletion (ABLAUT) then the correct output is derived.
(17)
a. /sul-il-i:l-e/ PARADIGM CONTRAST ABLAUT
? sulili:le *
Page 20 of 45
suli:le *!
cf. suli:le (perfect)
b. /mol-il-i:l-e/ PARADIGM CONTRAST ABLAUT
molele:le *!
? mole:le
cf. mo:zele (perfect)
The validity of this analysis is confirmed by stems (many of

them Arabic loans) that end in the flapped lateral /l/ but
exceptionally fail to mutate in the perfect (18). Consider the
following paradigms. (‘T’ represents a voiceless emphatic
(velarized) dental stop.)
(18)
Infinitive perfect
base x-sajil-a sajil-i:le ‘record’
appl. x-sajil-il-a sajil-il-i:l-e *sajil-i:l-e
base x-qa:Til-a qaTil-i:l-e ‘kill’
appl. x-qaTil-il-a qaTil-il-i:l-e *qaTil-i:l-e
They also fail to ablaut in the applied perfect since if they did
the result would be identical to the basic perfect. Thus, while
the regular verbs in (15) such as ku-mo:l-a mutate in the
perfect (mo:z-el-e) and ablaut in the applied perfect (mol-e:l-e/
mol-el-e:l-e/), the irregular verbs of (18) such as x-sajil-a are
exceptions to mutation (cf. perfect sajil-i:l-e); the latter fact
explains the corresponding absence of ablaut in the applied
perfect form sajil-il-i:l-e(*sajil-i:l-e).
In sum, a regular morphophonemic process(avoid three

successive laterals in the applied perfect) is blocked just in
case the output would be identical to the output of the
corresponding simple perfect. Like the Russian and Bulgarian
cases in which the suspension of vowel reduction depends on
the stress properties of the root, the decision whether or not
to suspend ablaut in Chi-Mwi:ni depends on the application of
another process(‘mutation’). It would be difficult to maintain
that the phenomenon (p.157) is merely the residue of some
Page 21 of 45
earlier historical change and not part of the active synchronic

phonology.
Page 22 of 45
7.3.4 The Arabic perfect

As observed by Mitchell (1993), the 3 sg. feminine of the
perfect inflection of the verb in many modern Arabic dialects
is the site of several phonological quirks. The various
allomorphic substitutions, stress shifts, and gemination can be
understood as maneuvers the language makes to block the
syncope of the suffixal vowel which otherwise would merge
this form of the paradigm with the 1 sg. and 2 sg. masculine.
(See Heath 1987: 289, 293 for similar observations for the
Moroccan dialect.) In what follows we review the evidence
cited by Mitchell as well as data from some other dialects. See
also McCarthy (8 below) for an approach to Arabic in which
paradigms play an important role.
Consider first the paradigms in (19) for canonical CaCCaC and

CeCeC stems /?allam/ ‘teach’ and /seme?/ ‘hear’ in the
Damascus dialect.6
(19)
1 sg. ?allám-t smé?-t
2 sg. m. ?allám-t smé?-t
2sg. f. ?allám-ti smé?-ti
3sg. m. ?állam séme?
3 sg. f. ?állam-et sém?-et
1pl. ?allám-na smé?-na
2pl. ?allám-tu smé?-tu
3pl. ?állam-u sém?-u
These paradigms exemplify two of the major processes that

are the linchpin for the phonological structure of most
colloquial Arabic dialects: stress and syncope. First, stress is
assigned within a three-syllable window at the right edge of
the word (essentially following the Latin Stress Rule): stress
appears on the rightmost heavy syllable (long vowel or closed
with final CVC counting as light) and otherwise on the
(ante)penult. Second, schwa (transcribed here as /e/) is
deleted from an unstressed open syllable. Thus when the suffix
begins with a consonant, stress is drawn to the closed penult
and the initial schwa of the CeCeC stem syncopates. For
example, /seme?-na/ appears as smé?-na. But when the suffix
Page 23 of 45
starts with a vowel, stress appears on the first syllable and the
second stem vowel is elided instead: /seme?-u/ → sém?-u. Both
schwas surface in the 3 sg. masc. /seme?/ → séme?:, the first
because it is stressed and the second because the syllable is
closed.
The constraints in (20) will be active in the analysis.
(20)
a. FOOT-FORM: metrical feet are bimoraic
trochees (Hayes 1985)
b. NON-FINALITY: penalize a foot aligned with
the right edge of the Prosodic Word (Prince and
Smolensky 1993)
(p.158)
c. LAPSE: penalize a sequence of two unparsed
syllables (Selkirk 1984)
d. *e. : penalize an unstressed syllable-final
schwa
The LAPSE constraint is demoted below NON-FINALITY. This

retracts the bimoaric foot from the right edge of the word.
(21)
Page 24 of 45
a./?allam-u/ FOOT FORM *e NON-FINALITY LAPSE
? (?ál)lamu *
(?álla)mu *!
?al(lámu) *!
b. /seme?-u/ FOOT FORM *e. NON-FINALITY LAPSE
(séme)?u *!
? (sém)?u
se(mé?u) *! *
S(mé?u) *!
c. /seme?-na/ FOOT FORM *e. NON-FINALITY LAPSE
se(mé?)na *!
? s(mé?)na
Page 25 of 45
The paradigms in (22) show the 3 sg. bases ?állam ‘he taught’
and ?állam-et ‘she taught’ and the 1 sg. ?allám-t‘I taught’
amplified with the suffixes marking object inflection.
(22)
Object ?állam ?állam-et ?allám-t
1sg. ?allám#ni ?allam-ét#ni ?allám-t#ni
2sg. ?állam#ak ?allam-ét#ak ?allám-t#ak

m.
2sg. f. ?állam#ek ?allam-ét#ek ?allám-t#ek
3sg. ?állam#o ?allam-ét#o ?allám-t#o

m.
3sg. f. ?allám#(h)a ?allam-ét#(h)a ?allám-t#(h)a
1pl. ?allám#na ?allam-ét#na ?allám-t#na
2pl. ?allám#kom ?allam- ?allám-

ét#kom t#kom
3pl. ? ?allam- ?allám-

allám#(h)om ét#(h)om t#(h)om
‘he taught ‘she taught ‘I taught pro’

pro’ pro’
The object suffixes are incorporated into the three-syllable

stress window and hence shift the accent on the base in
accord with the stress rule. What is of interest here is the 3 sg.
fem. base /?allam-et/ ‘she taught’. When the object suffix starts
with a consonant it closes the final syllable of the base to
create a heavy syllable that attracts the stress (e.g. ?allam-
ét#ni ‘she taught me’). What is anomalous is the shift of stress
when the suffix begins with a vowel (e.g. ?allam-ét#o ‘she
taught him’). A HLLL sequence (H (p.159) = heavy, L = light)
should produce antepenultimate stress with syncope of the
suffixal vowel: /?allam-et#o/ → /?allám-et#o/ → *?allám-t#o.
But such an Input–Output mapping would merge this form of
the paradigm with the 1 sg./2 sg. masc. /?allam-t#o/ → ?allám-
t#o. Hence, the otherwise unmotivated stress shift. If the
constraint of PAR-CONT is ranked above NON-FINALITY, then
syncope of the suffixal vowel is blocked and the suffixed 3 sg.
Page 26 of 45
fem. form remains distinct from the 1 sg. form at the cost of an
otherwise anomalous violation of antepenultimate stress.
(23)
/?allam- PARADIGM *e NON-

et#o/ CONTRAST FINALITY
?al(lámet)o *!
?al(lám)to *!
?? *
alla(méto)
cf. ?allámto ‘I taught him’
How strong is the evidence that paradigmatic contrast is the

best explanation for the anomalous stress of ?allam-ét#o? Let
us consider some alternatives. The first relevant observation is
that stress always precedes the object suffix when ?állam-et is
the base. This might lead one to suspect a boundary accent
along the lines of Latin li:mina ‘thresholds’ but li:miná-que ‘and
the thresholds’, as in the analysis of Mester (1994). But then
we expect uniform stress on the /?allam/ base. However, this is
not found, as shown by the first paradigm in (22) (cf. the
Banias dialect where this development has apparently
occurred (Ali-Adra 1999)). We might preserve the idea of
columnar stress but restrict it to subject suffixes. In fact, the
other, vowel-final suffixes lengthen their vowels before any
object suffix and so they attract stress via the regular rules: /?
allam-na#ha/ → ?allam-náa#ha ‘we taught her’. But other data
indicate that this is not the right analysis for the data in (22).
First, Measure-I verbs of the shape CaCaC and Hollow verbs of

the shape CaaC have the expected phonology with stress
retained on the verb stem and syncope of the suffixal vowel
before a vowel-initial object inflection. They thus behave
differently from CaCCaC stems. To show this we first consult
the paradigms of the simple perfect bases in (24). (‘D’
represents a voiced emphatic (velarized) dental stop.)
(24)
1 sg. Daráb-t ∫úf-t
Page 27 of 45
2 sg. m. Daráb-t ∫úf-t
2sg. f. Daráb-ti ∫úf-ti
3sg. m. Dárab ∫áaf
3 sg. f Dárb-et ∫áaf-et
1pl. Daráb-na ∫úf-na
2pl. Daráb-tu ∫úf-tu
3pl. Dárab-u ∫áaf-u
‘hit’ ‘see’
(p.160) The Damascus dialect in general preserves

underlying /a/ in an unstressed open syllable (the defining trait
of a so-called ‘differential’ dialect (Cantineau 1939)). The one
context where a drops out by a special rule is in the 3 sg. fem:
cf. Dárb-et (/Darab-et /)‘she hit’ v. Dárab-u ‘they hit’. In Hollow
verbs such as ∫áaf ‘see’, the root vowel changes to high in the
first and second persons. The upshot is that the 3 sg. fem. and
1 sg. forms are distinct in Measure-1 and Hollow verbs: Dárb-
et v. Daráb-t and ∫áaf-et v. ∫úf-t. When we turn to the behavior
of Dárb-et and ∫áaf-et under object inflection (25), we find that
the regular phonology returns. (The raised schwa in Dár e b-
t#ak is an epenthetic vowel inserted to break up clusters of
three successive consonants.)
(25)
Object Dárb-et ∫áaf-et
1sg. Darb-ét#ni ∫aaf-ét#ni
2sg. m. Dáreb-t#ak ∫áaf-t#ak
2sg. f. Dáreb-t#ek ∫áaf-t#ek
3sg. m. Dáreb-t#o ∫áaf-t#o
3sg. f. Darb-ét#(h)a ∫aaf-ét#(h)a
1pl. Darb-ét#na ∫aaf-ét#na
2pl. Darb-ét#kon ∫aaf-ét#kon
3pl. Darb-ét#(h)on ∫aaf-ét#(h)on
Page 28 of 45
As expected, stress shifts to the -et suffix before a consonant-

initial object inflection that closes the stem-final syllable. But
stress remains on the verbal root before vowel-initial
inflections accompanied by syncope of the suffixal vowel: /
Darab-et#o/ → Dár e b-t#o and /∫aaf-et#o/ → ∫áaf-t#o. This
regular phonology stands in sharp contrast to the quirky stress
of ?allam-ét#o. Such contrasting behavior is explained by
paradigm structure. Regular antepenultimate stress and
syncope will merge the 3 sg. fern.
(26)
a /Darab- PARADIGM *e NON-

(Dár)beto *!
? (Dáreb)to
Dar(béto) *!
cf. Darábto ‘I hit him’
b. /∫aaf- PARADIGM *e. NON-

(∫áa)feto *!
? (∫áaf)to
∫aa(féto) *!
cf. ∫úfto ‘I saw him’
(p.161) and 1/2 sg. masc. forms of CaCCaC verbs such as ?

allam before a vowel-initial object suffix. Quirky stress blocks
this merger. Quirky stress does not arise in the Measure 1
CaCaC and the CaaC Hollow verbs because the normal
phonology does not lead to paradigm merger with these bases.
Another relevant point, also observed by Mitchell (1993),

concerns the behavior of feminine nominals, including
participles. Participles inflect for gender and number but not
for person agreement with the subject: Dáareb ‘hitting’ m. sg.,
Dáarb-a ‘hitting’ f. sg., and Daarb-íin pl. The paradigms in (27)
show the participles combined with object suffixes.
(27)
Page 29 of 45
Object Dáareb Dáarb-a Daarb-íin
1g. Daaréb#ni Daarb-ét#ni Daarb-íin#ni
2g. m. áarb#ak Dáareb-t#ak Daarb-íin#ak
2g. f. áarb#ek Dáareb-t#ek Daarb-íin#ek
3g. m. áarb#o Dáareb-t#o Daarb-íin#o
3g. f. Daaréb#(h)a Daarb- Daarb-

ét#(h)a íin#(h)a
1pl. Daaréb#na Daarb-ét#na Daarb-íin#na
2pl. Daaréb#kon Daarb- Daarb-

ét#kon íin#kon
3pl. Daaréb#(h)on Daarb- Daarb-

ét#(h)on íin#(h)on
The singular feminine suffix /-a/ takes the allomorph /-et/ when
followed by a pronominal suffix or a governed NP—the so-
called iDafa or Construct State construction: Dáarb-a ‘her
hitting’ but Daarb-ét#ni ‘her hitting me’, Dáarb-it Faríid ‘her
hitting Fariid’. This construct suffix is equivalent in underlying
shape to the 3 sg. feminine suffix of the perfect: /-et/. Both
share the formal features [+fem, –pl]. But the phonological
behavior of the construct suffix is quite different—it is
essentially regular. Before consonant-initial object suffixes the
construct state suffix is stressed while before the vowel-initial
ones it syncopates—regardless of the shape of the preceding
stem: cf. /Daareb-et#o/ → Dáar e b-t#o ‘her hitting him’ v. /?
allam-et#o/ → ?allam-ét#o ‘she taught him’.7 Why this
difference in the behavior of the same suffix in essentially the
same phonological context? The notion of paradigmatic
contrast provides an attractive explanation. There is no other
member of the participial paradigm competing for the same
phonetic output. Hence the regular phonology can have its
way.
Dialect variation provides a couple of further arguments in

favor of this explanation. First, the 3 sg. feminine augmented
with vowel-initial object suffixes is the site of various
consonant geminations and vowel substitutions across the
Arabic dialects. We mention a couple of cases here. In the
Damouri dialect of Lebanon (Haddad 1983) the /t/ of the
feminine perfect suffix is geminated before a vowel: sakar-
Page 30 of 45
ít#ni ‘she intoxicated me’ v. sakar-ítt#ak ‘she intoxicated you

m. sg.’. And in the Syrian coastal dialect of Banias (Ali-Adra
Ali-Adra 1999) the vowel is lengthened: /ba?at-it#ni/ → ba?t-
ít#ni ‘she sent me’ but /ba?at-it#ik/ → ba?t-íit#ik ‘she sent you
fem.’. In neither Damouri nor (p.162) Banias are these
lengthening processes extended to the feminine construct
suffix in nominals.8 More generally, they are different ways of
achieving the same end—paradigmatic contrast.
Another point worth making is that the 3 sg. fem. allomorphy

looks to its right. Sensitivity to a vocalic versus consonantal
distinction in the immediate context is perhaps the most
common factor distributing allomorphs (cf. English a ≈ an,
French le ≈ l’, Korean -i ≈ -ka). But the rule typically looks
inward towards the stem (Carstairs 1987) so that prefixal
allomorphy is determined by the right-hand context and
suffixal allomorphy by the left-hand context. The Arabic case is
different. The shape of the 3 sg. fem. suffix depends on the
following context (as does the -a v. -et of the feminine nominal
suffix). Hence, it is not a typical case of affixal allomorphy that
falls under Carstairs’s Generalization. Rather, it is controlled
by PARCONT.
Finally, Mitchell (1993) observes that the phonologically

unmotivated stress shift found in the 3 sg. feminine perfect is
conspicuously absent from the Palestinian dialects where the
suffix retains the low-vowel of Classical Arabic: fahhám-at#ak
‘she explained to you’, halhál-at#uh ‘she loosened it’. Since
there is no general syncope of /a/ in Palestinian, there is no
threat of paradigmatic merger and hence no reason for quirky
stress or gemination.
7.3.5 North African Arabic dialects
Page 31 of 45
In this section we present the results of an informal survey of

several modern colloquial Arabic dialects of North Africa to
see to what extent they reflect and/or preserve the quirky
phonology motivated by the preservation of paradigm contrast
as discussed in s. 7.2.4.9 All the dialects studied here enforce a
ban against short high vowels or schwa in unstressed open
syllables (chiefly by syncope). They also all encode the 3 sg.
feminine perfect suffix with a syncopatable vowel (schwa or /
i/). The major result of the survey is that every dialect
manifests traces of the quirky phonology seen in the Damascus
dialect. The dialects fall into three main types. The first—
represented by Oujda Morocco—shows the Damascene pattern
with quirky phonology blocking syncope in Measure 1 and
Measure 2 verbs while regular phonology asserts itself in
Hollow verbs. The other two dialect types generalize the
quirky phonology in different ways.
7.3.5.1 Oujda
The perception of stress in Moroccan Arabic is quite subtle
and largely remains a question for future research.
Fortunately, the location of accent does not bear on the gen-
(p.163) eral points we wish to make and so we do not
attempt to transcribe it here. In our transcriptions schwa is
indicated by e; we abstract away from the coloring induced by
neighboring consonants. In (28) we see the perfect tense
paradigm for the reflex of a Measure 1 verb. An exceptionless
ban on open-syllable schwa accounts for the realization of
CeCeC as CCeC. This constraint is also responsible for the
syllable reshuffling found in the stem when the following suffix
starts with a vowel.
(28)
1 sg. rfed-t ‘carry’
2 sg. m. rfed-t
2sg. f. rfed-ti
3sg. m. rfed
3 sg. f. refd-et
1pl. rfed-na
2pl. rfed-tu
Page 32 of 45
3pl. refd-u
The paradigms in (29) show the 3 sg. masculine, 3 sg. feminine

and 1 sg. bases amplified by the object suffixes.
(29)
base rfed refd-et rfed-t
1sg. rfed#ni refd-et#ni rfed-t#ni
2sg. refd#ek refd-at#ek rfed-t#ek
3sg. m. refd#u refd-at#u rfed-t#u
3sg. f. rfed#ha refd-et#ha rfed-t#ha
1pl. rfed#na refd-et#na rfed-t#na
2pl. rfed#kum refd-et#kum rfed-t#kum
3pl. rfed#hum refd-et#hum rfed-t#hum
‘he ‘she carried pro’ ‘I carried pro’

carried
pro’
The first relevant observation is that the CCeC stem appears

as CeCC before the vowel-initial object suffixes #ek and #u.
The object suffixes thus syncopate the base just as the subject
suffixes do. Second, the 3 sg. feminine suffix -et is changed to -
at when the following object suffix begins with a vowel: refd-
at#ek ‘she carried you’, refd-at#u ‘she carried him’. We can
understand this allomorphy as a maneuver the grammar
makes to block the reshuffling of the stem’s syllable structure
that would otherwise obtain in order to adhere to the ban
against open syllable schwa. Such an output would be
identical to the 1 sg. In other words, the normal phonology
should transform /refed-et#u/ into rfed-t#u. But this shape is
identical to the 1 sg. form rfed-t#u ‘I carried him’. Replacing -
et by -at blocks syncope and keeps the 3 sg. fem. and the 1 sg.
forms phonologically distinct.
The same analysis carries over to the Measure 2 paradigm

represented by beddel ‘change’ in (30). Once again -et
becomes -at before a vowel: beddl-at#u ‘she changed him’. If
Page 33 of 45
this change did not occur then /beddel-et#u/ would be realized

as beddel-t#u and merge with the 1 sg.
(p.164)
(30)
base beddel beddl-et beddel-t
3 sg. m. beddl#u beddl-at#u beddel-t#u
3 sg. f. beddel#ha beddl-et#ha beddel-t#ha
‘he ‘she changed ‘I changed

changed pro’ pro’
pro’
Crucially, in Hollow verbs such as /∫af/ ‘see’ (31) the root

vowel ablauts to schwa in the first- and second-person forms.
Consequently, the verb stem takes a different shape from that
of the third person. When a vowel-initial object clitic is
appended to /∫af-et/ ‘she saw’, the schwa of the suffix -et
syncopates: ∫af-t#u ‘she saw him’. No allomorphic change in
the suffix is needed to keep ∫af-t#u ‘she saw him’ distinct from
∫ef-t#u ’I saw him’.
(31)
base ∫af ∫af-et ∫ef-t
3 sg. m. ∫af#u ∫af-t#u ∫ef-t#u
3 sg. f. ∫af#ha ∫af-et#ha ∫ef-t#ha
‘he saw pro’ ‘she saw pro’ ‘I saw pro’
The tableaux in (32) summarize the proposed analysis. The

constraint of Paradigm Contrast rises above faithfulness in
vowel height (Ident-[low]) allowing an otherwise unmotivated
lowering of the schwa in the suffix.
(32)
a /refed- *e. PARADIGM IDENT-

et#u/ CONTRAST [LOW]
rfedetu *!*
Page 34 of 45
refdetu *!
rfedtu *!
? rfedatu *
cf. rfedtu ‘I carried him’
b./∫af- *e. PARADIGM IDENT-

et#u/ CONTRAST [LOW]
∫afetu *!
? ∫aftu
∫afatu *!
cf. ∫ef-t#u ‘I saw him’
Essentially the same patterns of data are found in the Algiers

and Oran dialects of Algeria. In these dialects -et is replaced
by -at before a vowel but this process is blocked in the Hollow
verbs.10
(p.165) 7.3.5.2 Other dialects

If paradigmatic contrast is not invoked to explain the quirky
phonology seen in the Oujda and Damascus dialects then one
is faced with a rather complex pattern of allomorphy—one that
depends on both the preceding and the following contexts. In
effect, the -et suffix is modified if a vowel follows but this
change is overridden if the preceding root is Hollow. What we
find in other North African dialects is a redistribution of the
allomorphs for the 3 sg. feminine suffix. The patterns of
redistribution can be described in terms of whether reference
to the preceding or to the following context is dropped.
First we consider simplification on the left. Here reference to

the Hollow verb is dropped with the result that the quirky
phonology is generalized to all types of verb stem. But
reference to the following context is still required, at least in
those dialects where the repair chosen allows a distinction to
be made. The Djerba dialect of Tunisia falls into this class (33).
Here the repair is the same allomorphic replacement of -et by -
at found in Oujda. But it is extended to the Hollow verbs: ∫af-
at#u ‘she saw him’. However, it still only applies before the
vowel-initial suffixes: ∫af-et#ha ‘she saw her’.
(33)
Page 35 of 45
1 sg. ðreb-t ‘hit’
2. sg. f. ðreb-ti
3 sg. m. ðreb
3 sg. f. ðerb-et
1 pl. ðreb-na
2 pl. ðreb-tu
3 pl. ðerb-u
base ðreb ðerb-et
3 sg. m. ðerb#u ðerb-at#u
3 sg. f. ðreb#ha ðerb-et#ha
‘he hit pro’ ‘she hit pro’
3 sg. m. ?ellem ‘teach’
3 sg. f. ?ellm-et
1 sg. ?ellem-t
base ?ellem ?elhn-et
3 sg. m. ?ellm#u ?ellm-at#u
3 sg. f. ?ellem#ha ?ellm-et#ha
‘he taught pro’ ‘she taught pro’
3 sg. m. ∫af ‘see’
3 sg. f. ∫af-et
1 sg. ∫ef-t
base ∫af ∫af-et
3 sg. m. ∫af#u ∫af-at#u
3 sg. f. ∫af#ha ∫af-et#ha
‘he saw pro’ ‘she saw pro’
(p.166) The second and probably better-known exemplar of

this general pattern is Cairene Arabic. In this dialect the
repair is to shift the stress, as in Damascus. In Cairene Arabic
unstressed high vowels are barred from a VC__CV context via
Page 36 of 45
syncope. Thus a Measure 1 verb such as fíhim ‘understand’

loses its medial vowel before a vowel-initial suffix.
(34)
1 sg. fihím-t ‘understand’
2 sg. m. fihím-t
2sg. f. fihím-ti
3sg. m. fíhim
3sg. f. fíhm-it
1pl. fihím-na
2pl. fihím-tu
3pl. fíhm-u
The 3 sg. feminine perfect suffix -it is thus expected to

syncopate its vowel before another vowel. But this change is
blocked by shifting the stress to the suffix. In Cairene Arabic
this stress shift has been generalized to all verbs and gives
rise to the irregular stress pattern noted by Mitchell (1960),
Broselow (1976), and Welden (1980), among others. In (35) we
cite a few forms from Welden (1980).
(35)
Dárab-it ‘she hit’
Darab-ít#ik ‘she hit you f.’
Darab-ít#ha ‘she hit her’
?ista?bíl-it ‘she received’
?ista?bil-ít#ak ‘she received you m.’
?ista?bil-ít#ha ‘she received her’
∫áaf-it ‘she saw’
∫aaf-ít#ak ‘she saw you m.’
∫aaf-ít#ni ‘she saw me’
ráma ‘she threw’
ram-ít#u ‘she threw it m.’
Page 37 of 45
ram-ít#ha ‘she threw it f.’
The second type of simplification of the 3 sg. feminine

allomorphy process drops reference to the following context.
The Casablanca dialect of Morocco exemplifies this pattern
(36).
(36)
base rfed refd-at rfed-t
3 sg. m. refd#u refd-at#u rfed-t#u
3 sg. f. rfed#ha refd-at#ha rfed-t#ha
‘he carried ‘she carried ‘I carried

pro’ pro’ pro’
base ?ellem ?ellm-at ?ellem-t
3 sg. m. ?ellm#u ?ellm-at#u ?ellem-

t#u
3 sg. f. ?ellem#ha ?ellm-at#ha ?ellem-

t#ha
‘he taught ‘she taught ‘I taught

pro’ pro’ pro’
base ∫af ∫af-t ∫ef-t
3 sg. m. ∫af#u ∫af-t#u ∫ef-t#u
3 sg. f. ∫af#ha ∫af-t#ha ∫ef-t#ha
‘he saw pro’ ‘she saw ‘I saw

pro’ pro’
(p.167) Casablanca continues to draw a distinction between

Hollow verbs like ∫af and the rest; but it no longer cares about
what follows. The result is that the quirky allomorph -at has
been extended to the bare form (as well as to consonant-initial
object clitics). The Hollow verbs have simplified -et to -t.
In sum, the simplifications in the Djerba, Cairene, and

Casablanca dialects suggest that Paradigmatic Contrast has
been lost (or demoted) as the determining factor in the
allomorphy in these dialects. In no case is the quirky
allomorph extended to the nominal participles, however,
Page 38 of 45
indicating that the analogical spread is restricted to the verbal

inflectional paradigm.
Perhaps the best-known appeal to homonym avoidance is the

example due to Gilliéron discussed in Bloomfield (1933) from
certain southern French dialects where the word for ‘cock’ /
gal/ mysteriously drops out. Gilliéron noted that this happens
precisely in dialects that changed final /l/ to /t/. If the sound
change applied to this lexeme, it would merge with /gat/ ‘cat’.
To avoid this homophony, the lexical item is dropped from the
language.
The examples discussed in this chapter are quite different.

First, they are susceptible to a purely formal and precise
definition. PARADIGM CONTRAST penalizes an Input– Output
mapping that converges on another derivation within the same
inflectional paradigm: words sharing the same base and
differing in formal (inflectional) features. Second, it is not
plausible to attribute the cases reviewed here to factors of
language use. The latter might make sense for certain kinds of
paradigm gaps. To take another example from Arabic, in the
Makan dialect dative clitics consisting of the preposition -l plus
a pronoun may be attached to the verb, where they shorten a
final CVVC syllable: ∫aaf ‘he saw’, ∫af#l-u ‘he saw for him’.11
Participles may also host these clitics, with the results
indicated in (37).
(37)
sg. masc. maktuub maktub#l-u
sg. fem. maktuub-a maktuub-at#l-u
pl. masc. maktuub-iin maktuub-in#l-u
pl. fem. maktuub-aat ?
‘written’ ‘written to pro’
The feminine /-a/ takes the construct form /-at/ under

suffixation. While the dative #l-u attaches easily to the
masculine plural maktuub-iin with shortening of the vowel to
give maktuub-in#l-u, speakers resist constructing the
corresponding feminine plural (p.168) form. Why? Closed-
syllable shortening would produce a form identical to the
singular maktuub-at#lu. In this case one might say that a
pragmatic maxim (‘Be Clear’) explains the speaker’s
Page 39 of 45
reluctance to complete the attachment and hence that the

phenomenon is a matter of language use rather than language
structure. But this kind of explanation is not plausible for the
Arabic feminine perfects discussed in this section because the
grammar intervenes to calculate a minimally different output
that avoids the ambiguity but is otherwise not available.
Page 40 of 45
7.4 SUMMARY AND CONCLUSION

In this chapter we have reviewed several examples from the
contemporary and traditional literature that argue for
systematic phonological relations between members of a
paradigm that do not satisfy the containment relation of the
cycle. Specifically, the output form assigned to a given word
depends crucially on a related word in the paradigm that
cannot plausibly be treated as an intermediate stage in the
derivation. Phenomena of this form are expected if Output–
Output faithfulness relations are present in the grammar. But
they are surprising from a cyclic point of view in which one
word is derived essentially in isolation from its kin.
Although the number of examples in our survey is small, let us

see what generalizations can be drawn. First, the phonological
processes that are avoided can be either morphophonemic
(Chi-Mwi:ni) or phonological (Russian, Bulgarian, Arabic).
Second, the repair strategies are also varied and include
allomorphy (Russian), blocking (Bulgarian, Chi-Mwi:ni), and
recourse to auxiliary repair processes (Arabic gemination and
stress shift). Third, our corpus supports the traditional notion
of inflectional paradigm: words sharing the same stem and
differing in the exponence of the formal features of gender,
number, person, and tense/aspect. Avoidance of homophony
for words sharing the same inflection but different stems
appears to be much more sporadic and less systematic when it
does occur. In both Russian and Bulgarian vowel reduction in
stems is automatic and may lead to homophony: cf. Russian
m’át[ineq], m’it∫-í‘ball’ and m’étf, m’it∫-í‘sword’ or l’és, l’is-á
‘forest’ and l’ís, l’is-á ‘fox’. For Chi-Mwi:ni, Kisseberth and
Abasheikh report that mutation in the perfect may create
homophonous forms. The process is apparently blocked for a
few lexical items. For example, x-tu:nd-a ‘pick off’ and x-tu:Ng-
a ‘compose’ share the mutated perfect tu:nz-il-e while x-pa:mb-
a ‘decorate’ unexpectedly fails to mutate in the perfect,
evidently because the expected mutated form pa:nz-il-e
belongs exclusively to x-pa:nd-a ‘climb’. Fourth, words
displaying the uniformity and contrast effects are neighbors in
paradigmatic space and differ by just a single morphological
(formal) feature: Spanish (gender class), Russian and
Bulgarian (singular v. plural), Arabic (third v. first person), and
Chi-Mwi:ni (benefactive v. nonbenefactive). This suggests that
the search space over which paradigmatic contrast is defined
is sharply circumscribed and well defined. Fifth, only the Chi-
Page 41 of 45
Mwi:ni example conforms to ‘Base Priority’ (Benua 1997a) in

the (p.169) sense that it is the morphologically derived form
(the applied perfect) that accommodates to the more basic
source (the simple perfect). In the Russian, Bulgarian, and
Arabic examples the site of quirky phonology has the same
stem+affix structure as its paradigmatic neighbor and thus
neither one is contained in the other. Furthermore, in the
Bulgarian and Arabic examples it is the semantically
unmarked singular or third-person form that changes. Finally,
even in Chi-Mwi:ni Base Priority is of no real use. We must
wait until the Word Level cycle in which the perfect inflection
has been added in order to perform lateral deletion (ablaut).
Ablaut takes place in pel-e:l-e from /pe:l-il-i:le/ because the
simple perfect /pe:l-i:l-e/ mutates to pe:z-el-e. But the latter
computation is not a stage in the derivation of the former
comparable to condense as a stage in the derivation of
condensation. Once again the generalization is that the
phonology strives to maintain paradigmatic contrast. Sixth, in
all our examples it is the member of the paradigm that the
phonology threatens with merger that is also the site of repair.
Contrast could be maintained by letting the phonology have its
way but alter the paradigmatically related form—a type of
chain shift. It remains to be seen if cases of this nature can be
documented (see Gussenhoven 2000: 230 for mention of a
possible example from the Central Hessian dialect of German).
Also, in each case the phonology strives to keep two
exponentially distinct input forms from merging. It remains to
be seen if the phonology ever conspires to drive a part two
phonologically identical input forms in order to express a
paradigmatic distinction (e.g. different treatments of the
plural and possessive suffixes in /-s/ in English or the 1 sg. v. 2
sg. m. /-t/ of Arabic).
A crucial task for future research is to develop appropriate

analytic tools and grammatical formalisms that allow
phenomena such as paradigmatic uniformity and contrast to
be productively studied and the resultant generalizations to be
perspicuously expressed. Generative grammar has focused on
the study of individual words and sentences isolated from the
rest of the lexicon and the phonetic and semantic/pragmatic
modules. This idealization has been tremendously productive.
But there is growing evidence that it is also too severe.
Page 42 of 45
Aspects of grammatical form are sensitive to contextual

factors. Exactly how and where is a task for future research.
Notes:
(1) There are also cases where the competition is resolved in
favor of syllable matching: cf. aleman, aleman-a ‘German’ and
the corresponding diminutives aleman-sit-o, aleman-it-a; also
the personal names Ramon, Ramon-a and Ramon-sit-o, Ramon-
it-a. These appear to be in the minority.
(2) A reviewer suggests an alternative that does not require

reference to paradigmatic relations but otherwise provides a
‘clean analysis’ of the Spanish data. According to this solution
the diminutive suffix is -(s)itV whose theme vowel V is
determined by the gender/number of the diminutive stem. The
analysis has the following ingredients. First, the rule of
allomorphy states that -sitV is chosen after root-final r and n
and -itV is chosen elsewhere. Second, the diminutive suffix -
(s)itV is affixed to a lexeme’s root rather than to its stem.
Crucially, the final vowels of libro ‘book’ and korona ‘crown’
are exponents of the root while the final a of ratona is a suffix.
Thus, /raton-a/ has internal structure while /libro/, /korona/ do
not. Third, /libro-(s)ito/, /korona-(s)ita/ require a rule (ordered
after the allomorphy rule deleting s) to truncate the final vowel
of the root. While this analysis ‘works’, its key features seem
liabilities rather than assets. First, the rule distributing the
allomorphs of the -(s) it suffix refers to an arbitrary set of
segments. The fact that the paradigmatically related bases
amor, limon end in a consonant and thus share a property in
common is just an accident. Second, there is no independent
reason to assign the final vowels of libro, korona, etc. to the
root but retain a suffixal analysis for the final vowel of ratona
(i.e. /raton-a/). Indeed, it turns out to be a coincidence that the
Spanish lexicon has so many feminine ‘roots’ that terminate in
the vowel a and that exactly the same vowel is assigned as a
suffix in feminine ratona. Finally, the vowel truncation rule has
no independent motivation in the phonology of Spanish. Under
the analysis suggested in the text the allomorphs of -(s)it are
distributed in a phonologically natural fashion and the
traditional morphological analysis of [root-suffix-gender/
number theme] is uniformly assigned for all words under
discussion.
Page 43 of 45
(3) Thanks to Roumyana Izvorski and Olga Vaysman for help

with the translation.
(4) Chi-Mwi:ni also shortens long vowels that fall outside of a

three-syllable window computed at the right edge of the
phrase by the Latin Stress Rule (Selkirk 1986). This accounts
for the shortening of the root vowels of /so:m-/, /ko:d-/, /su:k-/,
and /pa:nd-/.
(5) Kisseberth and Abasheikh posit a rule that turns the

flapped lateral to plain before the applied suffix; the lateral of
the suffix then harmonizes with the lateral of the stem.
(6) Thanks to Jean-Michel Tarrier for confirming these data.
(7) See McCarthy (1980) for complications cited in Cowell

(1964).
(8) In Banias it appears that lengthening of /-it/ to /-iit/ before

vowel-initial suffixes has been generalized to all verbs
regardless of the shape of the preceding stem—presumably an
effect of paradigmatic uniformity.
(9) The survey was conducted in the spring of 2001 while the
author was a Visiting Scholar at the Institut de Linguistique et
Phonétique Générales et Appliquées in Paris. I wish to thank
Fatiha Abdulaziz, Nassira Betar, Samira Boumelassa,
Mohamed Elmedlaoui, Abdi Fassi-Fehri, Lilia Ould-Younes,
Karim Shoul, Abderrazzak Tourabi, Mohamed Yeou, and
Chakir Zeroual.
(10) Two other Moroccan dialects in the survey also block

syncope of -et except in the Hollow verbs; but they employ
different blocking mechanisms. In Fez the strategy is to
geminate the consonant (?ellem, ?ellm#u; ?ellm-et, ?ellm-
ett#u; ∫af-et,∫af-t#u) while in Taroudent there is no repair: the
schwa simply remains in an open syllable (?ellem, ?ellm#u; ?
ellm-et, ?ellm-et#u; ∫af-et, ∫af-t#u).
(11) I thank Rawiah Kabrah for these data. Shortening is not

triggered by a direct object clitic even though a phonologically
parallel VVCCV cluster is created: ∫aaf#ni ‘he saw me’.
Evidently, there is rebracketing of verb#l-pro to verb-l#pro to
create the CVVC-C# sequence that triggers shortening. See
Page 44 of 45
Lahiri (2000) for discussion of comparable rebracketings in

Bengali and Germanic.
Page 45 of 45
Optimal Paradigms


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Optimal Paradigms
John J. McCarthy
DOI:10.1093/acprof:oso/9780199267712.003.0008
Page 1 of 71
Optimal Paradigms
This chapter introduces a novel formalization of surface

resemblance through shared paradigm membership, couched
within Optimality Theory (OT) and correspondence theory. In
this Optimal Paradigms model (OP), an OT constraint
hierarchy evaluates candidates consisting of entire paradigms.
Within a paradigm qua candidate, every output realization of a
lexeme stands in correspondence with every other output
realization of that lexeme. Faithfulness constraints on this
intraparadigmatic Output–Output correspondence relation
resist alternation within the paradigm. Through interaction
with markedness and standard faithfulness constraints, they
account for surface resemblance and failure of resemblance
among members of a paradigm. The OP model is illustrated
and supported by a type of data that has not figured in
previous discussions of the surface-resemblance problem. The
chapter argues that certain properties of Classical Arabic root-
and-pattern morphology cannot be satisfactorily explained
without the OP model. This argument also shows how OP
supports the minimalist goals of Generalized Template Theory
(GTT), which seeks to eliminate templates and similar
stipulations from linguistic theory, replacing them with
independently motivated constraints. The analysis of Arabic
templates sets in clear contrast the differences between OP
and serial-derivational models, including proposals for melding
OT with Lexical Phonology.
Keywords: Optimality Theory, correspondence theory, Optimal Paradigms

model, surface resemblance, Generalized Template Theory, Arabic templates
8.1 Introduction
Traditional ideas of analogy, SPE’s phonological cycle, and
recent proposals about Output–Output faithfulness all have the
same goal: to account for surface resemblances among
morphologically related words. For instance, the trisyllabic
participial form lightening has a syllabic n because of its
relationship to the verb lighten—a relationship not shared with
the disyllabic noun lightning. Phenomena like this have figured
prominently in discussions of analogy and its successors.
In one thread of recent research (see s. 8.2), words have a

surface resemblance because of shared membership in a
paradigm. A paradigm is a set of inflected forms based on a
common lexeme or stem: <lighten, lightens, lightened,
lightening> is an example. The idea is that members of a
Page 2 of 71
Optimal Paradigms
paradigm should have similar surface phonology, such as the

syllabic n that appears consistently throughout the lighten
paradigm.
(p.171) In this chapter, I will introduce a novel formalization

of surface resemblance through shared paradigm membership,
couched within Optimality Theory (OT: Prince and Smolensky
1993) and correspondence theory (Benua 1997a; McCarthy
and Prince 1995, 1999). In this Optimal Paradigms model (OP),
an OT constraint hierarchy evaluates candidates consisting of
entire paradigms. Within a paradigm qua candidate, every
output realization of a lexeme stands in correspondence with
every other output realization of that lexeme. Faithfulness
constraints on this intraparadigmatic Output–Output
correspondence relation resist alternation within the
paradigm. Through interaction with markedness and standard
faithfulness constraints, they account for surface resemblance
and failure of resemblance among members of a paradigm.
The OP model is illustrated and supported by a type of data

that has not figured in previous discussions of the surface-
resemblance problem. Certain properties of Classical Arabic
root-and-pattern morphology, I will argue, cannot be
satisfactorily explained without the OP model. This argument
also shows how OP supports the minimalist goals of
Generalized Template Theory (GTT), which seeks to eliminate
templates and similar stipulations from linguistic theory,
replacing them with independently motivated constraints. The
analysis of Arabic templates sets in clear contrast the
differences between OP and serial-derivational models,
including proposals for melding OT with Lexical Phonology
(see s. 8.4).
8.2 Background to and Overview of the Op Model

This section provides an informal introduction to the Optimal
Paradigms model and to the analysis of Arabic templates.
Later sections fill in the details of the theory and the analysis.
Benua’s (1997a) Transderivational Correspondence Theory

(TCT) says that a morphologically derived surface form stands
in a relation of Output–Output (O–O) correspondence with its
morphologically simplex counterpart, called the base. For
example, lightening is in O–O correspondence with lighten.
Harmonic evaluation proceeds recursively, from the base
outward, by a principle of Base Priority. The surface phonology
Page 3 of 71
Optimal Paradigms
of the base lighten is carried over, by obedience to O–O

faithfulness constraints, to the derived form lightening. Base
Priority disallows influences running in the other direction, so
derived lightening can never influence the base lighten via O–O
correspondence. Because of Base Priority, TCT is inherently
asymmetric: if form A influences the phonology of a
morphologically related form B, then form B can never
influence the phonology of form A.
Another approach to the surface-resemblance problem is

inherently symmetric, requiring that morphologically related
words (or even individual morphemes) resemble one another
without assigning priority.
(p.172)
(1)
a. UNIFORM EXPONENCE (Kenstowicz 1996)
Minimize the differences in the realization of a lexical
item (morpheme, stem, affix, word).
b. METRICAL CONSISTENCY (Burzio 1994a: 228)
Every morpheme must be as metrically consistent as
possible.
In later work, Burzio (1996, 2002b) generalizes METRICAL

CONSISTENCY TO A PRINCIPLE OF ANTI-ALLOMORPHY,
which requires consistent realization of morphemes in all their
phonological properties, not just stress. With this
generalization, these principles become equivalent:
morphemes, stems, or words should not vary in their
phonological realizations. Unlike TCT, this theory is
symmetric: it is possible for form A to influence the phonology
of form B, while at the same time form B influences the
phonology of form A.
These two theories of surface resemblance among related

words are more complementary than competing.
Transderivational Correspondence Theory deals with the
classic cases of cyclic or stratal behavior, such as English
condénse/còndènsátion (Chomsky and Halle 1968; Pater 2000):
ranked faithfulness constraints on surface forms, formalized
under correspondence theory, enforce similarity, but by virtue
of Base Priority they effectively enforce it only on the derived
form, not the basic form. UNIFORM EXPONENCE (UE) has
been applied to paradigm-uniformity phenomena inter alia,
Page 4 of 71
Optimal Paradigms
preventing alternations among forms that are related

inflectionally, typically where no morphologically simplex base
is identifiable.
Transderivational Correspondence Theory is not applicable to

inflectional paradigms because it is an asymmetric, base-
prioritizing theory (pace Benua 1997a: Ch. 6). In TCT, the base
is the first step in the recursive evaluation. The derived form,
which is the next step in the recursive evaluation, is obtained
from the base by applying a morphological operation, such as
affixation. Inflectional paradigms have no base in this sense:
Latin amat ‘he loves’ is not derived from amo ‘I love’ or vice
versa; rather, both are derived from the lexeme /am-/. This
deficiency in TCT might be remedied by invoking
morphosyntactic unmarkedness, rather than morphological
underivedness, as a way of identifying the base form in an
inflectional paradigm (cf. Kenstowicz 1996; Kiparsky 1998: 31–
2; Kraska-Szlenk 1995; Kurylowicz 1949; Selkirk 2001). If this
criterion is applied, then Latin amat is presumably the base of
the /am-/ verbal paradigm. But this method of base
identification will not work for the Arabic example discussed
here (see s. 8.4.3).
Just as TCT has difficulties with inflectional morphology, so UE

runs into problems with derivational morphology. If applied to
derivation, UE overpredicts surface-resemblance effects. With
UE, it is possible for the derived form to influence the
phonology of the base or for a bound morpheme in one word to
influence the phonology of that morpheme in another word.
Solid synchronic examples of these predictions do not seem to
exist.1 Furthermore, UE represents more of an intuition (p.
173) than a usable phonological principle. In OT, a constraint
is a function from a linguistic expression to zero or more
violation-marks. Injunctions like ‘minimize the differences’ or
‘be as metrically consistent as possible’ are not well-defined
constraints. Moreover, expressions like ‘minimize X’ and ‘as X
as possible’ incorporate into themselves part of the definition
of EVAL, and so they should not appear in constraint
definitions (McCarthy 2002: 40). (See van de Weijer 1999 for
an improved formalization of UE and Buckley 1999 for related
discussion.)
The Optimal Paradigms (OP) model proposed here synthesizes

the best elements of TCT and UE. From TCT it draws the idea
of using correspondence theory as a foundation.
Page 5 of 71
Optimal Paradigms
Correspondence theory supplies a range of well-defined,

rankable constraints enforcing resemblance between forms.
From UE, OP incorporates the idea of evaluating surface
resemblance symmetrically across inflectionally related forms.
The central premises of the OP model are therefore as
described in (2).
(2) OP in Outline
a. Candidates consist of entire inflectional
paradigms, where an inflectional paradigm
contains all and only the words based on a single
lexeme (for similar ideas, see Bonet and Lloret
2001; Kenstowicz 1996: 385; McCarthy 1998;
Raffelsiefen 1995, 1999c ; Tesar and Smolensky
2000).2
b. Markedness and Input–Output faithfulness
constraints evaluate all members of the
candidate paradigm. The violation-marks
incurred by each paradigm member are added
to those incurred by all the others.3
c. The stem (output form of the shared lexeme)
in each paradigm member is in a
correspondence relation ℜop with the stem in
every other paradigm member. (That is, for
every candidate paradigm P there is a relation
ℜop on P×P.) There is no distinctive base—rather,
every member of a paradigm is a base of sorts
with respect to every other member.4
(p.174)
d. There is a set of Output–Output faithfulness
constraints on the ℜop correspondence relation.
For example, suppose we have a language with no suffix in the

singular and the suffix -i in the plural. Suppose this language
also has coronal palatalization before i. From the lexeme /mat/,
GEN will produce such candidate paradigms as <mat, mati>,
<mat, mat∫i>, and <mat∫, mat∫i>. Each candidate paradigm
brings with it a correspondence relation ℜop that relates the
stems in each paradigm member: mat ℜop mat∫iand,
symmetrically, mat∫i ℜop mat. (The portions standing in OP
correspondence are underlined.) The candidate <mat, mat∫i>
violates the constraint OP-IDENT(high) (or whatever feature
distinguishes t from t∫).5
Page 6 of 71
Optimal Paradigms
The OP model presupposes a distinction between inflectional

morphology, which is organized into paradigms, and
derivational morphology, which is organized hierarchically by
the relation ‘is derived from’. (See Spencer 1991: ch. 6 for a
review of the issues surrounding this assumption.)
Derivational morphology, I assume, is analyzed within TCT, as
before. But inflectional paradigms are different from
derivational hierarchies; in paradigms, all members are co-
equal in their potential to influence the surface phonology of
other members of the paradigm. This is formalized by ℜop
correspondence, which gives every paradigm member a
chance to affect any other member. Whether it does or does
not depends on the ranking.6
As we will see in greater detail in ss. 8.4 and 8.5, this model
predicts certain interactional patterns that set it apart from
other approaches, particularly TCT. One pattern is
overapplication-only. Given a language with a general process
of coronal palatalization, there are two ways to level
alternations within the paradigm of /mat/: <mat∫, mat∫i> or
<mat, mati>. The first of these paradigms shows
overapplication of the palatalization process—there is
palatalization of /t/ even in the unsuffixed form, where the
conditioning i is absent. The second paradigm shows
underapplication of palatalization: the process is blocked in
the suffixed form mati because there is no palatalization in the
unsuffixed form mat. OP-IDENT(high) is satisfied either way,
but the paradigm with underapplication cannot be obtained in
the OP model. (Some care is required in defining what over-
and underapplication mean in the context of a (p.175)
constraint-based theory like OT, so this statement should not
be applied indiscriminately. See s. 8.5.2.)
The problem with underapplication is that it competes with

overapplication. Over-application satisfies the high-ranking
markedness constraint that is responsible for the basic
palatalization process, but underapplication does not.
Underapplication does better on IO faithfulness, but that is
irrelevant, because the assumed existence of the process in
the language as a whole shows that IO-IDENT(high) is ranked
below the responsible markedness constraint. This means that
there is only one way for underapplication to win: some other
constraint must block overapplication. For an example of
overapplication-only, see s. 8.4.1. For examples where
overapplication is blocked and underapplication happens
Page 7 of 71
Optimal Paradigms
instead, see ss. 8.4.2 and 8.5.2. For a general evaluation of the
overapplication-only hypothesis, see s. 8.5.2. And for the
reduplicative parallel, upon which this argument is based, see
McCarthy and Prince (1995, 1999).
A related prediction of OP is attraction to the unmarked. I will

call a paradigm member an attractor if other members of its
paradigm are forced to resemble it by visibly active OP
faithfulness constraints. For example, in the leveled paradigm
<mat∫, mat∫i> from /mat/, the form mat∫i is the attractor, with
mat∫ forced to resemble it by OP-IDENT(high). Now suppose
we have a situation where there are two different ways to
satisfy an OP constraint—two different ways to level a
paradigm—depending on which member is acting as the
attractor. That is, there are candidates <A1, B1> and <A2,
B2> that equally satisfy the high-ranking OP constraint, but
differ in which paradigm member is doing the attracting: in
the first paradigm, A1 is the attractor, but in the second
paradigm, B2 is the attractor. Unless IO faithfulness is
decisive, the winning paradigm will be determined by
markedness, according to the logic in (i)–(iv).
(i) Identify the highest-ranking markedness constraint

that favors A1 over A2. Call it M(A1 > A2).
(ii) Identify the highest-ranking markedness constraint
that favors B2 over B1. Call it M(B2 > B1).
(iii) If M(A1 > A2) dominates M(B2 > B1), then A1 is
the superior attractor and so the paradigm <A1, B1>
wins.
(iv) But if M(B2 > B1) dominates M(A1 > A2), then B2
is the superior attractor and so the paradigm <A2, B2>
wins.
In other words, the markedness of the attractor is what

matters.
Attraction to the unmarked follows directly from the theory: in

OP, the markedness violations of a candidate paradigm are the
summed markedness violations of its individual members. The
markedness violations incurred by <A1, B1> are those
incurred by A1 or B1, so if the A1-favoring markedness
constraint dominates the B2-favoring one, the paradigm that
contains A1 is optimal.
Page 8 of 71
Optimal Paradigms
Here are some hypothetical examples to illustrate this

prediction; for reallife cases, see s. 8.4.2; Downing (this
volume); and Raffelsiefen (1995, 2000, this volume). Over-
application v. underapplication in <mat∫, mat∫i> v. <mat, mati>
is perhaps the simplest (p.176) example that can be
constructed; indeed, overapplication-only is a special case of
attraction to the unmarked. In <mat∫, mat∫i>, the suffixed
form mat∫i is the attractor, while in <mat, mati> unsuffixed
mat is the attractor. Which paradigm wins depends on which
markedness constraint is higher ranked: M(mat∫i > mati) or
M(mat > mat∫). Under the assumption that this language has
a general process of coronal palatalization, M(mat∫i > mati) is
top-ranked, so overapplication wins. A more complex example
can be constructed from a language with a -u suffix in the
singular and -i in the plural, with Japanese-style phonology of t
before these vowels: affrication to ts before u and
palatalization to t∫ before i. Then there is competition between
two different ways to overapply, <matsu, matsi> v. <matsu,
mat∫i>. By attraction to the unmarked, the choice between
them comes down to this question: is M(matsu > mat∫u)
ranked higher or lower than M(mat∫i > matsi)? The answer
could go either way; in fact, this might be the only situation
where these two constraints can be brought into conflict.
The OP model also predicts the possibility of majority-rules

effects, where the pattern that is most common in a paradigm
acts as an attractor for others. Majority-rules effects are not a
routine matter in the OP approach; the empirical
circumstances and constraint rankings that will produce them
are highly specific, as we will see in s. 8.5.1. But when
conditions are propitious, we expect to see results like the
following. Stems followed by a consonant-initial suffix
alternate one way, in accordance with undominated
markedness constraints. Stems followed by a vowel-initial
suffix alternate another way, also in accordance with those
undominated constraints. If markedness does not decide how
stems with no suffix will alternate, then they will be attracted
to the pattern that is more common in the rest of the
paradigm, which depends on whether consonant-initial or
vowel-initial suffixes happen to be more frequent. This result
follows from minimization of OP faithfulness violations—
though some OP faithfulness violation is unavoidable because
markedness forces differences between the two suffixed
Page 9 of 71
Optimal Paradigms
conditions, fewer violations of OP faithfulness are incurred if

the unsuffixed forms conform to the more common of the two
suffixed patterns.
It is important to realize that attraction to the unmarked,

overapplication-only, and majority-rules effects are not special
stipulations or auxiliary principles. Rather, they are
consequences of the OP model that devolve from its basic
architecture. It is also important to realize that OP, as a theory
of paradigms, asserts these claims only about inflectional
morphology, not derivational. If inflectional morphology turns
out to conform to these predictions, then the OP theory
receives strong support. If the predictions turn out to be
wrong, then the problem is profound and there is no easy way
to patch around it because the predictions are so deeply
connected to the tenets of the theory. Needless to say, whether
they are right or wrong, our theories should always make such
strong, falsifiable claims.
Much more detail and full exemplification will be provided in

ss. 8.4 and 8.5. But first we need to look at the phenomenon to
be analyzed, the template of the Arabic verb.
(p.177) 8.3 Background to and Overview of the Empirical

Problem
The goal of the theory of prosodic morphology is ‘to explain
the character of morphology/phonology dependencies
(templatic morphology, shape canons, circumscription, for
example) in independent, general terms, calling on universal
and language-particular principles’ (McCarthy and Prince
1994b: A1). This theory is successful to the extent that it
avoids positing its own special rules, constraints, or principles
that are invoked to analyze templatic or reduplicative
morphology but not applicable elsewhere.
Over the years, there has been gradual progress toward this
goal. Work started with the CV-template, which was applied to
root-and-pattern morphology (McCarthy 1981) and to
reduplication (Marantz 1982). This was later generalized to
incorporate syllabic information (Levin 1983) and prosodic
structure generally (McCarthy and Prince 1986/1996), leading
to the hypotheses in (3).
(3) Premises of the Theory of Prosodic Morphology
Page 10 of 71
Optimal Paradigms
a. The Prosodic Morphology Hypothesis

Templates, circumscriptional domains, and
canonical word-forms are defined in terms of the
fundamental units of prosody: moras, syllables,
feet, and prosodic words.
b. Template Satisfaction Condition
Satisfaction of templates is obligatory and
determined by universal and language-particular
requirements on the units they refer to.
These hypotheses shift much of the analytic burden from the

theory of prosodic morphology itself onto the theory of
prosody generally. The goal of independent, general
explanation is advanced because analyses are lifted out of the
domain of some specific phenomenon, such as reduplication,
and embedded into the overall prosodic phonology of the
language under investigation as well as the universal
principles of prosodic structure.
Work on prosodic morphology within Optimality Theory

(McCarthy and Prince 1993b) has taken these goals still
further. The Template Satisfaction Condition is not a special
stipulation, but rather an instantiation of constraint
satisfaction generally: constraint interaction, which is the
central element of OT, ensures that templates are satisfied
within ‘the universal and language-particular requirements on
the units they refer to’. Templates themselves are also seen as
consequences of interaction, with no special independent
status. Markedness constraints supplied by Universal
Grammar, ranked in ways that allow their effects to emerge in,
say, reduplication (McCarthy and Prince 1994a), are arguably
responsible for all phenomena that had in the past been
attributed to templates.
(p.178) The research program just described is called

Generalized Template Theory (GTT). In conformity with the
overall goals of the theory of prosodic morphology, GTT
proposes to eliminate even the vestigial prosodic-morphology-
specific principles in (3), relying on emergence of
independently motivated markedness constraints and
interaction with faithfulness to produce all apparent templatic
effects.7
Page 11 of 71
Optimal Paradigms
Like syntactic Minimalism (Chomsky 1995), which it abstractly

resembles, GTT must bear a heavy analytic burden if it is to
address the various phenomena previously analyzed with
richer theories of templates. The templatic system of the
Arabic verb presents obvious challenges.
Word formation in Arabic and other Semitic languages is the

premier example of prosodic morphology: words come in
certain fixed shapes that mark various morphological
distinctions, such as Classical Arabic kataba/kattaba ‘he
wrote’/‘he caused to write’ or kita;bun/kutubun ‘a book
(nom.)’/‘(some) books (nom.)’. These morphologically governed
variations in word-shape have in the past been attributed to
CV templates (McCarthy 1981), syllable-and-mora templates
(McCarthy and Prince 1986/1996), foot-based templates
(McCarthy and Prince 1990b), and the combination of a single
prosodic template with various affixes (McCarthy 1993;
Ussishkin 2000). This earlier work has mostly focused on one
important aspect of the problem: how are the different word-
shapes specified? That is, how does the grammar encode the
fact that causative verbs look like kattaba or some plural
nouns look like kutubun?
Here, I will look at a different aspect of the problem: what are

the shared properties of Arabic verbal templates? The
Classical Arabic verb comes in as many as fifteen different
derivational classes (see Appendix A for the full list),
traditionally called conjugations (by Orientalists), ʔawza:n (in
Arabic, singular wazn), or binjanim (in Hebrew, singular
binjan). The ‘template of templates’ in McCarthy (1981)
generalizes over the templates of all the conjugations, showing
that they have a great deal in common (4).
(4) Template of templates for Classical Arabic verb
Why are the

verb’s
templates
limited to the
expansions of
this schema?
How are the many stipulations inherent in (4) to be reconciled
with the minimalist goals of GTT? Can they be said to emerge
from independently motivated constraints?
Page 12 of 71
Optimal Paradigms
The nominal morphology of Arabic supplies a clue. The

template of templates says that verb stems must end in CVC].8
There are, then, verb stems like faʕal, faʕʕal, and (p.179)
daħra3.9 But noun stems are not so restricted (see Appendix B).
Nouns can have stems ending in CVC, CV:C, and CVCC: faʕal,
faʕa:l, faʕl, etc. Verb templates differ from noun templates in
this respect.
The template of templates also says that verb stems can begin
with [CV or [CCV: faʕal, faʕʕal v. ftaʕal, stafʕal, etc. But the
stems of Arabic nouns (except for obviously deverbal nouns)
always begin with a single consonant: faʕal, faʕi:l, faʕl, etc. In
this case, it is the nouns, rather than the verbs, that are
subject to the more stringent requirement.
The templates of verbs and nouns are different in these two

respects. This observation suggests that some independent
difference between nouns and verbs plays a role in
determining their phonological shapes. Just one independent
difference has the potential to do that: verbs and nouns inflect
differently. The inflectional system of Arabic nouns is quite
limited. There are no inflectional prefixes, and the inflectional
suffixes are all vowel initial (5).
(5) Inflectional suffixes of Classical Arabic noun
Singular -u ‘nominative’
-i ‘genitive’
-a ‘accusative’
Dual -a: ‘nom.’
-aj ‘gen./acc.’
Plural -u: ‘nom. masculine’
-i: ‘gen./acc. masc.’
The singular suffixes are followed by -n if the noun is

indefinite; the dual and plural suffixes are followed by -ni and -
na, respectively, if the noun is not in the construct state. There
is in addition a feminine plural suffix -a:t, which is followed by
the singular desinences in (5). Clitic pronouns that mark
possession come after these case- and number-marking
suffixes: baqar-a:t-u-hu ‘cow-pl.-nom.-his = his cows’.
Page 13 of 71
Optimal Paradigms
The paradigm of the Classical Arabic verb is much larger,

exceeding 150 members if some less common distinctions of
mood are included. The verbal paradigm includes inflectional
prefixes, all of the form CV-, and inflectional suffixes, both V-
initial and C-initial. The list in (6) is limited to those
inflectional affixes that attach directly to the verb stem.
(6) Stem-affixing Inflections in the Classical Arabic Verb
a. Perfective
C-initial suffixes
-tu ‘1st singular common’
-ta ‘2nd sg. masc’
-ti ‘2nd sg. feminine’
- ‘2nd dual com.’

tuma:
-na: ‘1st plural com.’
-tum ‘2nd pl. masc.’
- ‘2nd pl. fem.’

tunna
V-initial suffixes
-a ‘3rd sg. masc’
-at ‘3rd sg. Fem.’
-a: ‘3rd du. masc’
-ata: ‘3rd du. Fem.’
-u: ‘3rd pl. masc’
b. Imperfective indicative10
CV prefixes
ʔa- ‘1st sg. Com.’
ta- ‘2nd com., 3rd sg. and du. fem.’
ja- ‘3rd masc, 3rd pl. fem.’
na- ‘3rd pl. com.’
C-initial suffix
-na ‘2nd and 3rd pl. fem.’
Page 14 of 71
Optimal Paradigms
V-initial suffixes
-u ‘1st and 3rd sg. com., 2nd sg. Masc., 1st pl.
com.’
-i:na ‘2nd sg. fem.’
-a:ni ‘du.’
-u:na ‘2nd and 3rd pl. masc.’
(p.180) The inflectional affixes of the verb are obviously

much more diverse than those of the noun. Nouns have
suffixes only, but verbs also have CV prefixes in the
imperfective. The shape of noun suffixes is always V-initial, but
verbs have both V-initial and C-initial suffixes. From this
difference in inflections, it is possible to make sense
phonologically of the templatic differences between nouns and
verbs. I begin here with an informal sketch of how to analyze
one of the two main problems: verb stems must end in CVC],
but noun stems are not so restricted. The rest of the analysis
will be found in s. 8.4.
Suppose, in conformity with OT’s thesis of richness of the base

(McCarthy 2002: 68–82; Prince and Smolensky 1993), that the
lexicon supplies verb stems that are as diverse as noun stems.
This means that alongside the actual verb stem faʕal there are
also hypothetical verb stems faʕa:l and faʕalk. We seek to
explain why these other verb stems are not merely
hypothetical but impossible.
The starting point is to consider some candidate paradigms

derived from one of these hypothetical verb stems, faʕa:l (7).
In candidate (7a), the whole paradigm is faithful to the input
verb stem /faʕa:l/, preserving the long vowel throughout the
paradigm. This is fatal, however, because medial superheavy
syllables like ʕa:l are ruled out (p.181) for markedness
reasons.11 In the terminology of rule-based phonology, closed
syllable shortening has underapplied in (7a).
(7) Candidate Remarks

Paradigm
a. <faʕa:la, The form faʕa:ltu is phonotactically

faʕa:ltu, …> bad because of the medial superheavy
syllable ʕa:l.
Page 15 of 71
Optimal Paradigms
b. <faʕa:la, This paradigm has closed-syllable

faʕaltu, …> shortening. It is phonotactically OK,
but vowel length alternates within the
paradigm.
c. <faʕala, This paradigm has no vowel-length

faʕaltu, …> alternations and no phonotactic
problems. But it is indistinguishable
from the paradigm of faʕal.
Candidate (7b) is the most interesting one: this candidate

preserves the underlying long vowel when it is phonotactically
permitted, before V-initial suffixes, but shortens it when the
phonotactics demand, before C-initial suffixes. Candidate (7b)
is non-optimal, however. The alternation between a: and a
within the paradigm is detected by faithfulness constraints on
ℜop, the intraparadigmatic correspondence relation. By the
central hypothesis of OP, there are correspondence relations
between the stems in every pair of paradigm members: faʕa:la
ℜop faʕaltu, symmetrically faʕaltu ℜop faʕa:la, and so on. The
faithfulness constraint OP-IDENT-WT (cf. Urbanczyk 1996) is
breached whenever vowel length alternates within a
paradigm. If OP-IDENT-WT is ranked above its Input–Output
faithfulness counterpart IO-IDENT-WT, then (7b) is ruled out
because it tolerates intraparadigmatic alternation that is
avoidable by shortening throughout the paradigm, as in (7c).
Candidate (7c) wins. It is completely unfaithful to faʕa:l’s

underlying long vowel; no trace of that vowel’s length can be
found anywhere in the paradigm—in rule-based terms, closed
syllable shortening overapplies. This paradigm wins precisely
because of the ranking just described, which has an OP weight
faithfulness constraint ranked above an IO weight faithfulness
constraint. And because (7c) wins, an input verb stem like
faʕa:l is pointless, since it everywhere neutralizes to faʕal. This
is what Prince and Smolensky (1993) call ‘Stampean
occultation’, in tribute to Stampe (1973a, b). Though the
underlying form faʕa:l is in principle possible under richness of
the base, learners will never be moved to set it up as an actual
lexical item because it is hidden or ‘occulted’ by the actually
occurring form faʕal, with which it always neutralizes (for a
previous application of Stampean occultation to paradigms,
see McCarthy 1998).12
Page 16 of 71
Optimal Paradigms
This section began with the problem of explaining why Arabic

verbal templates must end with CVC], but nominal templates
can also end with CV:C] and VCC].
(p.182) This problem emerges from Generalized Template

Theory, which demands explanations in terms of independently
motivated constraints, abjuring mere stipulations like (4). The
analysis just sketched is a first installment on this explanation.
The crucial constraints—the markedness of superheavy
syllables and the faithfulness constraint OP-IDENT-WT—are,
respectively, an uncontroversial element of markedness theory
and a basic entailment of the OP model and correspondence
theory. The role of OP in this explanation is clear: it supplies a
way of precisely controlling alternations within paradigms
using correspondence theory. Subsequent sections fill in the
details of this analysis and show how this and other results are
obtained from OP.
8.4 Optimal Paradigms Theory and Arabic Templates

The preceding section identified two main problems in the
analysis of Arabic templates. Verbal templates always end in
CVC], but nominal templates can also end in CV:C] and VCC].
Verbal templates can begin with [CV or [CCV, but nominal
templates can only begin with [CV. I sketched a solution to the
first problem that relies on the OP model and the observation
that verbs have more diverse suffixing inflection than nouns
do. The formal details of that solution are supplied in s. 8.4.1,
and s. 8.4.2 extends the solution to the second problem. In s.
8.4.3, serial approaches to the same phenomena are compared
with OP and found lacking. Finally, s. 8.4.4 describes some of
the conditions where OP faithfulness constraints are violated
in Arabic, resulting in paradigms that are not completely
leveled. This is, of course, fully expected in OT: any constraint,
including OP faithfulness, is violable.
8.4.1 Suffixing inflection and the right edge of the template

Arabic verbs inflect with suffixes that are both V-initial and C-
initial, but Arabic nouns only inflect with V-initial suffixes.
With the OP model and some independently motivated syllabic
phonology of Arabic, these templatic restrictions on the right
stem-edge can be explained.
The story begins with syllable phonotactics. In Classical

Arabic, sequences like [C1V:C2C3V] or [C1VC2C3C4V] are never
found (though see n. 11). Under richness of the base, we
Page 17 of 71
Optimal Paradigms
cannot assume that they are conveniently absent from inputs;

rather, their ill-formedness must be derived from constraint
interaction. Markedness constraints that rule out the various
ways of parsing these sequences must dominate some relevant
faithfulness constraint, so that any instances of these
sequences that occur in the input are treated unfaithfully in
the output: e.g. /C1V:C2C3V/ → [C1VC2C3V]. Among these
markedness constraints are *µµµ]σ, which prohibits trimoraic
syllables, (p.183) and *APP-σ which prohibits linking a coda
consonant directly to the σ node as an appendix (see Sherer
1994 and references there). There are other ways of faithfully
parsing [C1V:C2C3V] that must also be excluded, such as
syllabifying C2 as a nucleus or having it share a mora with the
preceding vowel, or parsing C2C3 as a complex onset. Here I
will focus on just *µµµ]σ and *APP-σ, with the understanding
that constraints against these other configurations are ranked
similarly.
As was just noted, *µµµ]σ and *APP-σ must dominate some

relevant faithfulness constraint(s) if they are to succeed in
ruling out the forbidden sequences. Alternations that occur in
external sandhi tell us what those faithfulness constraints are.
Sequences with a long vowel are resolved by closed-syllable
shortening (8a), and sequences with a triconsonantal cluster
lead to epenthesis (8b).
(8) a. Closed-syllable Shortening
/fi: l-na:s-i/ /abu: l-wazi:r-i/
fin.na:.si a.bul.wa.zi:.ri
‘among the people’ ‘the vizier’s father’
b. Epenthesis
/qa:l-at sma⊕/ /muhammad-un l-nabijju/
qa:.la.tis.mat⊕ mu.ham.ma.du.nin.na.bij.ju
‘she said “listen!”’ ‘Mohamed the prophet’
Closed-syllable shortening supplies an argument that the

markedness constraints *µµµ]σ and *APP-σ dominate the
Input–Output faithfulness constraint IO-IDENT-WT (9).
(9) *µµµ]σ and *APP-σ IO-IDENT-WT
Page 18 of 71
Optimal Paradigms
/abu: l-wazi:r-i/ *µµµ] σ *APP-σ IO-ID-WT
a. ☞ a.bul.wa.zi:.ri *
b. a.builµ.wa.zi:.ri *!
c. a.bu:lσ.wa.zi:.ri *!
The notation l µ betokens l’s status as a moraic coda to a

superheavy syllable, and l σ indicates an appendix to a heavy
syllable. Neither is a possible analysis because the respective
markedness constraints against them are top-ranked. Instead,
vowel shortening results, violating low-ranked IO-IDENT-WT.
In principle, /C1V:C2C3V/ sequences could be resolved by

epenthesizing a vowel or deleting a consonant instead. That
this does not occur shows that other Input–Output faithfulness
constraints, IO-DEP-V and IO-MAX-C, dominate IO-IDENT-WT
(10). IO-MAX-C is unviolated in Classical Arabic, but IO-DEP-V
is violated with triconsonantal clusters, where vowel
shortening is simply not an option (11).
(p.184)
(10) IO-DEP-V, IO-MAX-C IO-IDENT-WT
/abu: l-wazi:r-i/ IO-DEP-V IO-MAX-C IO-ID-WT
a. ☞ a.bul.wa.zi:.ri *
b. a.bu:.li.wa.zi:.ri *!
b. a.bu:.wa.zi:.ri *!
(11) *µµµ]σ,*APP-σ, IO-MAX-C ≫ IO-DEP-V( ≫ IO-IDENT-

WT)
Page 19 of 71
Optimal Paradigms
/qa:l-at smaʕ/ *µµµ]σ *APP-σ ; IO-MAX-C IO-DEP-V
a. ☞ qa:.la.tis.matf *
b. qa:.latsµ.matf *!
c. qa:.latsσ.matf *!
d. qa:.lat.matf *!
Page 20 of 71
Optimal Paradigms
To summarize, superheavy syllables or equivalent

configurations are avoided by vowel shortening or, when
shortening is not possible, by vowel epenthesis (cf. Yawelmani
in Kisseberth 1970). The top-ranked constraints in (11) rule
out superheavy structures and consonant deletion; the lower-
ranking constraints express the preference for shortening over
epenthesis.
The constraint ranking given in (11) is sufficient background

for analyzing the phonology of the right edge of the verb stem
template. As I will now show, the same markedness constraints
that are active in (11), *µµµ]σ and *APP-σ, also affect the right
edge of verb stems. Verbal suffixes are both V-initial and C-
initial. When a suffix is C-initial, then *µµµ]σ and *APP-σ force
unfaithful analysis of any putative verb stem ending in CV:C]
or CVCC]. The constraints of the OP model transmit that
unfaithful analysis throughout the paradigm, even to forms
with V-initial suffixes.13 Nouns, though, are effectively immune
from this leveling process because nouns do not have C-initial
inflections.
In the verb, the crucial conflict is between the paradigm

constraints OP-DEP-V/ OP-IDENT-Wt and their IO counterparts.
With the OP constraints ranked above the IO constraints,
uniformity within the paradigm takes precedence over
faithfulness to the input. This can be seen with the candidate
paradigms of faʕa:l, which appeared previously in (7).
(p.185)
(12) OP-IDENT-WT IO-IDENT-WT
Page 21 of 71
Optimal Paradigms
/faʕa:l/ + {a, tu,…} *µµµ]σ *APP-σ OP-ID-WT IO-ID-WT
a. ☞ <faʕala, faʕaltu, **
…>
b. <faʕa:la, faʕa:lσtu, *!
…>
c. <faʕa:la, faʕa:lµtu, *!
…>
d. <faʕa:la, faʕaltu, … *! *
>
Page 22 of 71
Optimal Paradigms
The candidates, in order of appearance, include a paradigm

(l2a) where the input long vowel has been shortened
throughout, two paradigms (l2b, c) where the input long vowel
has been preserved throughout at the cost of greater
markedness, and an alternating paradigm (l2d), where the
input long vowel is preserved before V-initial suffixes and
shortened before C-initial suffixes.14
In the OP model, a markedness constraint assigns marks to a

whole paradigm by summing over the marks assigned to each
of its members. Candidates (l2b) and (l2 c) are shown with one
mark each from the constraints *APP-σ and *µµµ]σ,
respectively. In fact, there are many more such marks, once
the whole paradigm is considered. In the perfective and
imperfective indicative, there are ten forms with C-initial
suffixes, so a paradigm that is faithful to input /faʕa:l/ will have
ten violations of *APP-σ or *µµµ]σ. Whether one or 10, these
marks are of course fatal.
Candidate (12d) is the important one. In the OP model, every

candidate brings with it a correspondence relation among all
the stems within the paradigm. In candidate (12d), the relation
is faʕa:la ℜop faʕaltu, placing long a: in correspondence with
short a.15 But with OP-IDENT-WT ranked above IO-IDENT-WT,
intraparadigmatic length alternations are avoided by
shortening the vowel throughout the paradigm, even before V-
initial suffixes. Hence, candidate (12a) emerges as the winner.
It has no fatal markedness violations and no vowel-length
alternations—at the cost of obliterating every trace of the
underlying long vowel of /faʕa:l/. Because it shortens the vowel
throughout the paradigm, it incurs as many marks from IO-
IDENT-WT as there are forms in the paradigm, but that does
not matter because IO-IDENT-WT is ranked at the bottom.
The tableau at (13) makes the same point for the matched pair
of faithfulness constraints OP-DEP-V and IO-DEP-V. Candidates
(13b) and (13c) have the same markedness problems that
afflict (12b) and (12c). In candidate (13d), there is vowel
epenthesis to relieve the forbidden triconsonantal cluster. But
this leads to an intraparadigmatic (p.186) vowel/zero
alternation: faʕla ℜop faʕiltu. This alternation violates OP-DEP-
V (or, symmetrically, OP-MAX-V). In (13a), epenthesis
metastasizes throughout the paradigm, even in forms where it
Page 23 of 71
Optimal Paradigms
is not required for markedness reasons. This candidate is

optimal because OP-DEP-V dominates IO-DEP-V.
(13) OP-DEP-V IO-DEP-V
Page 24 of 71
Optimal Paradigms
/faʕl/ + {a, tu, …} *µµµ]σ *APP-σ OP-DEP-V IO-DEP-V
a. ☞ <faʕila, faʕiltu, … **
>
b. <faʕla, faʕlσtu, …> *!
c. <faʕla, faʕlµtu, … > *!
d. <faʕla, faʕiltu, … > *! *
Page 25 of 71
Optimal Paradigms
We now have all the elements of an explanation for the fact

that Arabic verb stem templates never end in CV:C] or VCC].
In OT, an output structure [X] is absolutely ill formed in a
language L if the grammar of L maps all inputs to outputs
other than [X] (see McCarthy 2002: 68–82, 195–200 and
references there). Tableaux (12) and (13) show that the
grammar of Classical Arabic maps the inputs /faʕa:l/ and /faʕl/
onto paradigms that do not preserve the stem-final CVC or
CVCC. Before C-initial suffixes, these inputs must be changed
by shortening or epenthesis, and this change carries over to
paradigm members that have V-initial suffixes because of the
high-ranking OP constraints OP-DEP-V and OP-IDENT-WT.
To complete this part of the argument, it is necessary to show

that no input will map to paradigms that preserve stem-final
CV:C or CVCC. The inputs /faʕa:l/ and /faʕl/ are merely the
most likely suspects; there are other inputs that could
conceivably be mapped onto one of the forbidden paradigms.
We can quickly reason through these possibilities. Clearly,
having more long vowels or more consonants in the input, or
combining the two (/fa:ʕl/), presents no danger, since the
interactions in (12) and (13) cover these situations too. Inputs
without long vowels or clusters, such as /faʕal/ or /faʕ/, are not
a problem either, because Classical Arabic has no phonological
processes that could create long vowels or consonant clusters.
In sum, given the rankings in (12) and (13), literally no input
will map to a verbal paradigm with surface stem-final CV:C or
CVCC.
Noun stems are different. Because nouns only have V-initial

suffixes, the markedness constraints *APP-σ and *µµµ]σ are
satisfied without further ado. The noun stems faʕa:l and faʕl
remain unchanged throughout the nominal paradigm: faʕa:l-u,
faʕa:l-a, etc. Because noun stems never have to accommodate
to C-initial suffixes, the OP constraints have no real work to do
in the noun.
This analysis has shown that the observed restriction on the

right edge of the verb-stem template and the absence of this
restriction in the noun can be derived from independently
motivated constraints of markedness theory and the OP model.
No special template of templates like (4) is needed. More
generally, there is no need for (p.187) an apparatus of rules,
representations, or constraints that are designed specifically
Page 26 of 71
Optimal Paradigms
for prosodic morphology. What we have, then, is exactly the

kind of explanation required by Generalized Template Theory.
The same kind of reasoning can be applied to another

templatic generalization about the right stem-edge: verb and
noun stems never end in a vowel.16 Imagine a vowel-final stem
like *faʕa. Since both verbs and nouns have vowel-initial
suffixes, there will always be at least some paradigm members
where combining this stem with a suffix threatens to yield
hiatus: *faʕa.a, *faʕa.at, *faʕa.u:, *jafʕa.u for verbs; *faʕa.u,
*faʕa.i, *faʕa.a for nouns. Hiatus is intolerable, however,
because ONSET is undominated in Arabic. Hiatus is resolved
by epenthesizing ʔ, so an input like /faʕa-at/ will surface as
faʕaʔat. From there, it is clear how to proceed: the epenthetic
ʔ, which is forced before vowel-initial suffixes by ONSET,
metastasizes to forms with consonant-initial suffixes because
OP-DEP-C dominates IO-DEP-C. The ranking argument has the
same basic structure as (13), mutatis mutandis. Readers can
work out the details for themselves.
The analysis of Classical Arabic in this section illustrates one

of the OP model’s consequences described in s. 8.2,
overapplication-only. In the paradigm <faʕala, faʕaltu, …>
from input /faʕa:l/, the process of closed-syllable shortening is
observed to overapply, since the vowel has been shortened in
forms like faʕala where the syllable is not closed. In the
competing paradigm *<faʕa:ʕa, faʕa:ltu, …> shortening
notionally underapplies: the form *faʕa:ltu has no shortening,
thereby preserving resemblance with its faithful fellow
paradigm member *faʕa:la. Tableau (12) reveals why under-
application cannot win. The candidates with underapplication,
(12b) and (12c), violate the top-ranked markedness constraints
*µµµ]σ and *APP-σ. The only constraint that unambiguously
favors these candidates, the IO faithfulness constraint IO-
IDENT-WT, must be ranked below *µµµ]σ and *APP-σ because
the language as a whole has an active process of closed-
syllable shortening. (If IO-IDENT-WT were ranked above
*µµµ]σ and *APP-σ, then there would simply be no closed-
syllable shortening anywhere, and this is not what is meant by
the term ‘underapplication’.) The only way to redeem (12b, c)
would be for some additional constraint, ranked above the
markedness constraints, to tip the balance in favor of
underapplication (see s. 8.4.2 for an example). In short,
Page 27 of 71
Optimal Paradigms
although both underapplication and overapplication satisfy OP

faithfulness constraints, underapplication cannot win because
it loses to overapplication (cf. McCarthy and Prince 1995,
1999).
The account of why there are no vowel-final stems also

exemplifies overapplication-only. The competing level
paradigms are *<faʕa.a, faʕatu, …>, with underapplication of
ʔ epenthesis, and <faʕaʕa, faʕaʔtu, …>, with overapplication.
Because the paradigm with underapplication has as many
ONSET violations as there are vowel-initial suffixes, and
ONSET is an undominated constraint, underapplication is a
sure loser. The only way to level a paradigm in OP is by
overapplication (unless it is blocked—see ss. 8.2 and 8.4.2).
(p.188) Overapplication-only distinguishes the OP model

from TCT (see s. 8.2). Because TCT has a principle of Base
Priority, there can be underapplication of a process in a
derived form in order to maintain similarity with the base.
English examples like còndènsátion are typical; the process of
sonorant destressing underapplies in the syllable den in order
to maintain similarity with the main-stressed syllable of the
base condénse. Underapplication does seem to be an authentic
characteristic of derivational morphology, where the base can
be identified morphologically.17 But the OP model, which is
limited to inflectional morphology, treats all members of a
paradigm equally; there is no special base form, so there is no
base priority, and hence true underapplication is impossible.
The empirical question of whether true underapplication ever
actually occurs in inflectional morphology is revisited in s.
8.5.2. The circumstances where underapplication is possible
are discussed in the next section.
8.4.2 Prefixing inflection and the left edge of the template

At the left edge of the verb stem, the permitted structures are
richer than in the noun. Verb stem templates can begin with
[CV or [CCV sequences, but noun stems (except for
nominalized verbs) can only begin with [CV. As I will show, this
difference follows from the fact that verbs have CV-
inflectional prefixes, but nouns do not. The idea is that the
presence of a CV- prefix in the imperfective verb forces an
underlying /CCV …/ stem to surface faithfully, and this cluster
carries over to the prefixless perfective through the agency of
OP correspondence.
Page 28 of 71
Optimal Paradigms
The analysis starts with the restriction on nouns—a restriction

that verbs violate. The non-existence of [CCV nouns entails
that any input of this shape receives an unfaithful analysis.
Since we know from (8b) that there is vowel epenthesis in
Arabic, a hypothetical noun stem like /fʕa:l/ must be mapped
onto the paradigm <fiʕa:lu, fiʕa:la, …>. To ensure this result,
some markedness constraint violated by faithful *<fʕa:lu,
fʕa:la, …> must be ranked above IO-DEP-V.
This markedness constraint comes from the ALIGN family. We

also know from (8) that Arabic has syllabification across word
boundaries. Though nouns never have prefixes, a putative
[CCV noun like *fʕa:lu would show up in all phrasal contexts
with the f parsed as a coda:
(14) Syllabification of impossible [CCV noun *f ʕa:lu
a. After pause
ʔif.ʕa:.lu
b. After C-final word
…Cif.ʕa:lu
c. After V-final word
…Vf.ʕa:lu
(p.189) In short, the stem-initial f of *fʕa:lu is never syllable-

initial because of the way that syllabification and epenthesis
work in Arabic. The markedness constraint responsible for the
ill-formedness of *<fʕa:lu, fʕala, …> is therefore ALIGN-
L(Stem, σ), which requires that stem-initial segments also be
syllable-initial.18 Input /fʕa:l/ cannot map faithfully to *<fʕa:lu,
fʕa:la, …> because stem-initial f is never syllable-initial in any
context. The tableau in (15) certifies this ranking argument.
(15) ALIGN-L(Stem, σ) IO-DEP-V
ALIGN-L(Stem, σ) IO-DEP-
V
a. ☞ <.fiʕa:lu, .fiʕa:la, **
…>
Page 29 of 71
Optimal Paradigms
ALIGN-L(Stem, σ) IO-DEP-
V
b. <f .ʕa:lu, f .ʕa:la, **!

…>
To aid in determining alignment violations, the stem-initial

consonant is italicized and nearby syllable boundaries are
indicated by a period/full-stop. As tableau (15) shows,
epenthesis is forced by left stem-edge alignment, which is also
known to block prothesis or resyllabification in other
languages (McCarthy and Prince 1993a, b).
By virtue of this ranking, the paradigm resulting from the

input /fiʕa:l/ is indistinguishable from the paradigm derived
faithfully from the input /fiʕa:l/. By the logic of Stampean
occultation, there are no [CCV noun templates in Arabic
because the grammar always maps them onto surface forms
with [CV templates, so there is no reason for learners to set up
underlying /CCV…/ nouns.
But verbs do have [CCV templates. In a verbal paradigm like

<s.tafʕala, jas.tafʕilu, …> (conjugation X in the traditional
Western nomenclature), some greater imperative overrides
ALIGN-L(Stem, σ). To identify that imperative, we need to look
at the competition (i) and (ii).
(i) In *<.sitaf ʕala, ja.sitaf ʕilu, …>, underlying /stafʕal/

undergoes epenthesis everywhere, thereby satisfying
ALIGN-L(Stem, σ) perfectly. The problem, which will be
explained shortly, is that prefixed *ja.sitafʕilu has
marked prosodic structure that the winner jas.tafʕilu
does not.
(ii) In *<.sitafʕala, jas.tafʕilu, …>, underlying /stafʕal/
undergoes epenthesis only when unprefixed, just like
the noun in (15). So the unprefixed members of the
verbal paradigm (the perfectives) satisfy ALIGN-
L(Stem, σ)—not perfectly, but better than the winner
<s.tafʕala, jas.tafʕilu, …>. Furthermore, with no
epenthesis in the prefixed form, there is no problem
with marked prosodic structure. Nonetheless,
*<.sitafʕala, jas.tafʕilu, …> fails because it exhibits
intraparadigmatic vowel/zero alternations, a breach of
OP faithfulness.
Page 30 of 71
Optimal Paradigms
(p.190) I will now fill in the details of this analysis, beginning

with the candidate in (ii).
The failed candidate in (ii) shows that vowel/zero alternations

within the paradigm are avoided at the cost of poor alignment.
This is a straightforward generalization of the results in s.
8.4.1, where the OP faithfulness constraints prohibiting vowel/
zero alternations were also important. In the required ranking,
OP-DEP-V (or OP-MAX-V) is ranked above ALIGN-L(Stem, σ),
as (16) shows.
(16) OP-DEP-V ALIGN-L(Stem, σ)
{ja, …} + /stafʕal/ + {a, OP-DEP- ALIGN-L(Stem,

u, …} V σ)
a. ☞<s.taf.ʕa.la, **
jas.tafʕilu, … >
b. <.sitafʕala, jas.taf *! *
ʕilu, … >
The paradigm in (16b) includes corresponding pairs like

sitafʕala ℜop ja stafʕil u, which violate OP-DEP-V. The winning
candidate avoids this violation by leveling, even though it
means that ALIGN-L(Stem, σ) is violated throughout the
paradigm. Compare this tableau with (13), which likewise
shows the paradigm-leveling effect of OP-DEP-V.
The failed candidate in (i), *<si.tafʕala, ja.si.tafʕilu, …>, has a

level paradigm with no OP faithfulness violations.
Furthermore, it satisfies ALIGN-L(Stem, σ) perfectly.
Nonetheless, it loses for prosodic reasons involving the
interaction of stress and syllable weight—reasons that turn out
to be irrelevant in nouns because they lack prefixes. To see
this, we first require some background about Arabic prosody
specifically and prosodic theory generally.19
There is no direct testimony about Classical Arabic stress from

the native grammatical tradition. Still, some inferences can be
drawn from internal evidence like versification and from
consistencies among the stress patterns of the modern Arabic
dialects. Classical Arabic stress was without doubt quantity-
sensitive, treating heavy (CVC and CV:) syllables differently
than light (CV) syllables. It surely also had extrametricality of
final syllables. All modern dialects have bounded stress
Page 31 of 71
Optimal Paradigms
systems (that is, binary feet); Bedouin dialects are often

iambic, and sedentary dialects are always trochaic. Most of the
trochaic dialects have right-to-left foot assignment, but
Egyptian goes the other way. The iambic dialects all have left-
to-right footing, as expected since right-to-left iambic stress is
probably universally impossible (Hayes 1995b: 262 ff.; Kager
1993; McCarthy and Prince 1993b). The analysis I present
below is worked out under the assumption that Classical
Arabic stress is right-to-left trochaic with final syllable
extrametricality, but the results are the same if stress is left-to-
right iambic or trochaic. (I do not present the details of the
responsible stress constraints since they can be easily gleaned
from any of the standard texts, such as Kager (1999b).)
On the theoretical side, Gouskova (2003) argues that the

constraint called SWP (p.191) (for Stress to Weight Principle
(cf. Prince 1990)) is responsible for compelling syncope
processes in many languages. SWP assigns a violation-mark to
any stressed light syllable. Inter alia, it favors feet consisting
of a single stressed heavy syllable—(′H) feet—over feet
consisting of two light syllables, one of which is stressed—(′LL)
or (L′L) feet. When it is ranked above MAX-V, SWP can force
one of the vowels to delete in a sequence of light syllables: /
fiʕal-aw/ (′fiʕ)law in Iraqi Arabic.
In Classical Arabic, however, the ranking of MAX-V and SWP is

just the opposite: there is no general syncope process, as
shown by the tableau at (17).20
(17) IO-MAX-V SWP
IO-MAX-V SWP
a. ☞ <(fáʕa)lu, (fáʕa)la, …> **
b. <(fáʕ)lu, (fáʕ)la, …> **!
In (17a), the feet consist of two light syllables, so SWP is

violated. The candidate paradigm (17b) corrects these
violations, but at the expense of deleting input vowels. With
this ranking and with constraints against vowel or consonant
lengthening equally high ranked, SWP cannot compel
unfaithfulness to the input.
Page 32 of 71
Optimal Paradigms
Though it cannot force syncope in Classical Arabic, SWP

blocks satisfaction of ALIGN-L(Stem, σ) by epenthesis when it
would create additional (LL) feet. This effect can be seen by
comparing the winning paradigm <s.tafʕala, jas.tafʕilu, …>
and its better-aligned competitor *<.sitafʕala, ja.sitafʕilu, …>.
Prefixed (jàs)(táf)ʕilu has only (H) feet, while *(jàsi)(táf)ʕilu has
one (LL) foot, thereby violating SWP. From this, we can
conclude that SWP dominates ALIGN-L(Stem, σ) (18).
(18) SWP Ȧ ALIGN-L(Stem, σ)
{ja, ta, …} + /stafʕal/ + {a, tu, SWP ALIGN-

u, na, …} L(Stem, σ)
a. .☞ <s.(táf)ʕala, s.(tàf)(ʕál)tu, ****

(jàs).(táf)ʕilu, (jàs).(tàf)(ʕíl)na, …
>
b. <.si(táf )ʕala, .si(tàf )(ʕál)tu, **

(jà.si) (táf )ʕilu, (jà.si)(tàf )
(ʕíl)na, …>
Because suffixes can affect foot parsing, I have included

representative paradigm members with both V-initial and C-
initial suffixes. Tableau (18) shows that, despite being ranked
below IO-MAX-V, SWP is active in Classical Arabic, blocking
epenthesis in (p.192) prefixed [CCV verb stems even at the
expense of inferior alignment. One might think of this as a
kind of anti-syncope: though it has no syncope process, the
language is blocked from creating configurations of the type
that are known to undergo syncope in other languages. This is
an expected result of OT’s inherently typological nature and
constraint violability. Like SWP in Arabic, a constraint can be
active even when crucially dominated.21
This proposed ranking for SWP must be checked against three

conditions. First, it must account for all extant [CCV verb
stems in Arabic. Second, it must not permit [CCV noun stems,
and this should be related to the absence of CV- prefixes in the
noun. Third, it must not interfere with the results about the
right stem-edge in s. 8.4.1, since constraint interactions must
be consistent within the language. I address each of these
tests of the analysis in turn.
Page 33 of 71
Optimal Paradigms
The existing Arabic [CCV verb templates include conjugation

VII /nfa’al/, conjugation VIII /ftaʕal/, conjugation IX /fʕalal/,
conjugation X /stafʕal/, the rare conjugations XI–XV, and the
rare third quadriliteral conjugation (see Appendix A). The rare
conjugations all have the same prosodic structure as /stafʕal/:
an initial consonant followed by a heavy syllable. In all
relevant respects, they will behave exactly like the candidates
in (18) and need not detain us further.
Conjugations VII, VIII, and IX also have the same prosodic

structure as one another, so an analysis that is valid for one
can be readily extended to the others. With, say, VIII /fta’al/ as
the input, the candidates of current interest are the winner <f.
(táʕa)la, (jàf).(táʕi)lu, …> and the perfectly aligned loser
*<.fi(táʕa)la, (jà.fi)(táʕi)lu, …>. Winner and loser both violate
SWP, but the loser does worse (19).
(19)
{ja, ta, …} + /ftaʕal/ + {a, tu, u, na, SWP ALIGN-

…} L
a. ☞ <f .(táʕa)la, f .ta(ʕál)tu, (jà f). ** ****

(táʕi)lu, (jà f).ta(ʕíl)na, …>
b. <.fi(táʕa)la, .(fìta)(ʕál)tu, (jà.fi) *****!

(táʕi)lu, ja.(fìta)(ʕíl)na, …>
Perfect satisfaction of SWP could be achieved by syncope, but

as I have already shown in (17), IO-MAX-V’s high rank
excludes that possibility. Though perfection is not possible,
SWP is still doing its job, blocking epenthesis in the prefixed
forms *(jà.fi)(táʕi)lu and *ja.( fìta)(ʕíl)na.
The second test of the analysis is whether SWP interferes with

epenthesis in noun stems. If SWP were to block epenthesis in
nouns, then it would undermine the results of the ranking
argument in (15). It turns out that SWP is not decisive in
nouns because the relevant candidates tie in their
performance, leaving the choice up to ALIGN-L (p.193)
(Stem, σ), exactly as in (15). As we will now see, the reason
why they tie is that nouns lack prefixes and so SWP never
comes into play.
Page 34 of 71
Optimal Paradigms
Consider the putative cluster-initial noun stem /fʕal/. Since

observed noun templates never begin with clusters, this input
must be mapped unfaithfully onto the winning paradigm
<.fiʕalu, .fiʕala, …>, much the same as (15). The interesting
competitor is *<f.ʕalu, f.ʕala, …>—interesting because it is
faithful but impossible for a noun. The winner better satisfies
ALIGN-L(Stem, σ), but since SWP dominates ALIGN-L(Stem,
σ), it is important to check that SWP does not favor the loser.
And in fact it doesn’t, as the tableau at (20) shows.22
(20)
/fʕal/ + {u, a, i} SWP ALIGN-L(Stem, σ)
a. ☞ <.(fíʕa)lu, .(fíʕa)la, … **
>
b. <f.(ʕálu), f .(ʕála), … > ** **!
Both candidates have (LL) feet throughout, so both violate

SWP equally. This leaves the decision up to ALIGN-L(Stem, σ),
which favors the candidate without an initial cluster, exactly as
in (15). The same argument can be made for other
hypothetical cluster-initial noun templates like /fʕa:l/, /dħara3/
and /dħar3/.
Because verbs have CV- prefixes, epenthesis into the stem-

initial cluster that immediately follows a prefix creates an
immediate danger of violating SWP, as examples like *(jà.fi)
(táʕi)lu show. But nouns lack inflectional prefixes and, as (20)
indicates, this means that nouns are not big enough for SWP
to be decisive. It is possible, however, to imagine a noun
template that is big enough to allow SWP to block epenthesis.
For example, the invented stem /fʕalakt/ yields the candidate
noun paradigms <f.ʕa(lák)tu, f.ʕa(lák)ta, …> and <.(fìʕa)
(lák)tu.(fìʕa)(lák)ta, …>. SWP favors the former even though it
has an initial cluster. But noun stem templates like /f’alakt/ are
ruled out for an entirely different reason. Arabic templatic
nouns, like the verbs, are built on roots of two, three, or at
most four consonants. Three-consonant roots can be extended
by derivational affixes and still be templatic (e.g. mifta:ħ ‘key’,
from /ftħ/ ‘open’), but four-consonant roots with the same
affixes are non-templatic by independent criteria (see
Appendix B). Nouns like hypothesized /fʕalakt/ with five
Page 35 of 71
Optimal Paradigms
consonants or more are non-templatic, lying outside the basic

root-and-pattern morphological system of the language. They
are therefore irrelevant to the analytic goals of this chapter.
Earlier, I noted that there are three tests of the analysis

presented in this section. Two have already been addressed;
the third is a test for consistency: do the constraint
interactions of this section fit with those of s. 8.4.1? The first
step is to assemble all the ranking results into a single
diagram, with the highest-ranked constraints at the top (21).
(p.194)
(21)
The lines
indicate
proven
constraint
domination; the numbers are those of the examples where the

ranking argument is presented.
The diagram (21) is useful first as a check for incompatible

ranking results; there are none. The diagram also suggests
where to look for further ranking arguments. For example, the
undominated markedness constraints *µµµ]σ and *APP-σ might
be brought into conflict with some of the higher-ranked
constraints in the main chain along the left, and the same
might be done with the OP faithfulness constraints.
The result of these lucubrations is an argument that *µµµ]σ,

*APP-σ, OP-DEP-V, and OP-IDENT-WT all dominate SWP. The
argument is based on the candidates in (12) and (13), where
vowel shortening or vowel epenthesis occur at the right stem-
edge before C-initial suffixes and thence are transmitted to the
rest of the paradigm. Because shortening and epenthesis
create light syllables, they can potentially introduce violations
Page 36 of 71
Optimal Paradigms
of SWP. Those violations are tolerated because SWP is

dominated by the responsible constraints. The tableaux (22)–
(25) clarify the details of these arguments.
In the winner (22a), the paradigm members with V-initial

suffixes all violate SWP.
(22) *µµµ]σ *APP-σ, OP-DEP-V SWP (cf (13))
Page 37 of 71
Optimal Paradigms
/faʕl/ + {a, tu, …} *µµµ]σ *APP-σ OP-DEP-V SWP
a. Ⅾ <(fáʕi)la, fa(ʕìl)tu, *
…>
b. <(fáʕ)la, (fáʕlσ)tu, *!
…>
c. <(fáʕ)la, (fáʕlµ)tu, *!
…>
d. <(fáʕ)la, fa(ʕíl)tu, *!
…>
Page 38 of 71
Optimal Paradigms
(p.195) This violation is compelled by the joint action of the

two markedness constraints and OP-DEP-V, which forces
epenthesis before a V-initial suffix to match the phonotactically
driven epenthesis before a C-initial suffix.
Likewise, (23a) violates SWP, but this is unavoidable because

of the high-ranking markedness and OP faithfulness
constraints.
(23) *µµµ]σ, *APP-σ, OP-IDENT-WT SWP (cf (12))
Page 39 of 71
Optimal Paradigms
/faʕa:l/ + {a, tu, …} *µµµ]σ *APP-σ OP-ID-WT SWP
a. ☞ <(fáʕa)la, *
fa(ʕál)tu, … >
b. <fa(ʕái)la, *!
fa(ʕá:lσ)tu,…>
c. <fa(ʕá:)la, *!
fa(ʕá:lµ)tu,…>
d. <fa(ʕá:)la, *!
fa(ʕál)tu,…>
Page 40 of 71
Optimal Paradigms
Now that all the elements of the analysis are in hand, we are
in position to bring them together and see how it works, with
our eyes on the goal of reconciling the facts of Arabic with the
tenets of Generalized Template Theory. A [CCV noun stem
like /fʕal/ undergoes epenthesis to improve alignment, and
epenthesis is not blocked by SWP (24).
(24)
/fʕal/ + {u, a, i} SWP ALIGN-L(Stem, IO-DEP-

σ) V
a. ☞ <.(fíʕa)lu, . ** **
(fíʕa)la, … >
b. <f .(ʕálu), f . ** **
(ʕála), …>
A [CCV verb stem like /ftaʕal/ is misaligned, however, because

SWP disfavors epen-thesis in the prefixed form, and this lack
of epenthesis is carried over to the rest of the paradigm by OP
faithfulness (25).
(25)
{ja, ta, …} + /ftaʕal/ + OP- SWP ALIGN-L

{a, tu, u, na, …} DEP-V (Stem, σ)
a. ☞ <f.(táʕa)la, (jà f). ** **

(táʕi)lu, …>
b. <.fi(táʕa)la, (jà.fi) ***!

(táʕi)lu, … >
c. <.fi(táʕa)la, (jà f). *! ** *

(táʕi)lu, … >
But in nouns, which lack prefixes, SWP is not decisive, so the

well-aligned candidate wins. That is why nouns do not have
[CCV templates. Like the right stem-edge, the observed
templatic conditions on the left stem-edge follow from a
combination of markedness requirements (SWP, alignment)
and OP faithfulness.
Here is a less formal summary. Verbs and nouns differ in the

shape of inflectional suffixes and the availability of inflectional
prefixes. This difference has consequences (p.196) for their
form. At the right edge of the stem, verbs are less diverse than
Page 41 of 71
Optimal Paradigms
nouns because only verbs must deal with both -V and -CV
suffixes. At the left side of the stem, verbs are more diverse
than nouns because only verbs take prefixes. These
differences in inflectional morphology, combined with
independently motivated markedness and OP faithfulness
constraints, explain the templatic differences between nouns
and verbs. There is no need for a template per se nor for
special templatic constraints or similar mechanisms. Deriving
templatic effects from independently motivated constraints, as
in this analysis, is in accordance with the reductionist goals of
Generalized Template Theory.
In s. 8.2, I described several consequences that can be

deduced from the OP model. One of them is attraction to the
unmarked, referring to the special role that markedness
constraints have in determining which members of the
paradigm will influence others via OP faithfulness. The
analysis presented in this section illustrates attraction to the
unmarked.
As shown in (25), there are two ways to satisfy OP faithfulness,

the winner <f.(táʕa)la, (jàf).(táʕi)lu, …>, which has no
epenthesis anywhere in the paradigm, and the loser
*<.fi(táʕa)la, (jà.fi)(tàʕi)lu, …>, which has epenthesis in every
member of the paradigm. In the winner, the prefixed form
jaftaʕilu is acting as the attractor, forcing its stem-initial
cluster on the unprefixed form ftaʕala in spite of ALIGN-
L(Stem, σ). In the loser, it is unprefixed *fitaʕala that is acting
as the attractor, with its epenthetic vowel spreading
throughout the paradigm. In situations like this, where the OP
constraint is satisfied either way and where the relevant IO
faithfulness constraints are ranked too low to make a
difference, the winning paradigm is the one whose attractor is
least marked relative to its counterparts in competing
paradigms.
Refer again to tableau (25). Candidate (25c) has an

intraparadigmatic vowel/zero alternation that is fatal, given
OP-DEP-V’s high rank. Candidates (25a) and (25b) differ in
which form is doing the attracting, as I’ve already noted. Since
(25a) and (25b) satisfy the OP constraint equally well and
since IO faithfulness is ranked too low to matter, the choice
between them is made by the highest-ranking markedness
constraint on which they differ. That is, the markedness of the
attractor is what distinguishes between (25a) and (25b). The
Page 42 of 71
Optimal Paradigms
tableau in (26) limits the comparison to just these candidates

and the two markedness constraints, with shared violation-
marks cancelled (McCarthy 2002: 6, Prince and Smolensky
1993).
(26)
SWP ALIGN-L(Stem, σ)
a. ☞ <f .(táʕa)la, (jà f). **

(táʕi)lu, … >
b. <. fi(táʕa)la, (jà. fi) *!

(táʕi)lu, … >
These two candidates differ exactly in the markedness of the

attractor. In (26b), the attractor is *fi(táʕa)la, and it is less
marked according to ALIGN-L than f (táʕa)la in (26a). In (26a),
the attractor (jàf )(táʕi)lu is less marked according to SWP
than *(jàfi)(táʕi)lu in (26b). Paradigm (26b) is non-optimal for
this reason: its attractor performs (p.197) better on a lower
ranking markedness constraint than the optimal candidate’s
attractor. This is attraction to the unmarked: the attractor is
optimized relative to the highest-ranking markedness
constraint.
The analysis in this section also illustrates another

consequence of the OP model, overapplication-only (cf. s.
8.4.1), though in a somewhat backhanded way. To a cursory
inspection, the <ftaʕala, jaftaʕilu, …> paradigm looks like
underapplication of epenthesis, with the losing paradigm
*<fitaʕala, jafitaʕilu, …> being an example of overapplication
of epenthesis. In one sense, this interpretation is correct:
underapplication can only win when overapplication is blocked
by a high-ranking constraint, and here SWP is blocking
overapplication. In another sense, though, even this case of
underapplication is really overapplication. It is overapplication
of the blocking constraint itself. (Speaking very loosely, a
markedness constraint overapplies if its effects are
transmitted through the paradigm via OP faithfulness
constraints.) SWP blocks epenthesis in the prefixed form and,
via OP correspondence, it indirectly blocks epenthesis in the
unprefixed form. True underapplication is predicted never to
occur for the reasons given in ss. 8.2 and 8.4.1.
8.4.3 Comparison with other models
Page 43 of 71
Optimal Paradigms
This is a good opportunity to compare OP with standard

derivational approaches to resemblance among related words.
These approaches include not only the cycles of Chomsky and
Halle (1968) but also the strata of the theory of Lexical
Phonology (Borowsky 1986; Hargus 1985; Hargus and Kaisse
1993; Kiparsky 1982a, b; Mohanan 1986; Rubach 1993;
Strauss 1982) and Stratal OT, which organizes several OT
grammars into a series of strata.23 Since these various
theories are more similar to each other than to OP, I will lump
them together and refer to them as LP.
LP analyzes all phonological resemblances between related

forms with a serial derivation. To return to an example in s.
8.1, the word lightening has a syllabic n because, earlier in the
derivation, it was lighten. Base Priority is an automatic and
unavoidable consequence of this theory; the base has priority
because, in the temporal metaphor of derivational phonology,
it existed prior to the derived form.
The results and predictions of OP are not duplicated in LP. For

LP, the base of some form X is identified morphologically: it is
just X minus the results of the last morphological operation,
usually affixation. Though this may be appropriate for
derivational morphology, there is no general, non-arbitrary
way to identify a base in this sense in an inflectional paradigm
—inflected forms are separately derived from the shared
lexeme, not from each other. So the LP model is a poor fit to
the morphological structure of paradigms (see s. 8.2).
(p.198) Because LP is committed to identifying the base by

its derivational priority, it cannot explain or even describe
attraction to the unmarked. In the previous section, I showed
how the OP model identifies the attractor in a paradigm by its
phonology, using the markedness constraints as ranked in the
language as a whole. I also argued that this is the right way to
understand Arabic. But in LP, the attractor is just the
derivational predecessor—attractors, then, can only be
identified on morphological grounds, so attraction to a
phonologically unmarked form is inexplicable, unless the
derivational predecessor also just happens to be unmarked.
A further point about OP—and a corollary to attraction to the

unmarked—is that different paradigm members may act as
attractors with respect to different phonological properties. In
Classical Arabic, the phonology of the right stem-edge is
Page 44 of 71
Optimal Paradigms
determined by those paradigm members that have C-initial

suffixes; the phonology of the left stem-edge is determined by
those paradigm members that have CV- prefixes. Tableau (27)
shows how these different attraction effects are negotiated
within a single, consistent constraint ranking.
(27)
Page 45 of 71
Optimal Paradigms
{ja, ta, …} + / OP-DEP-V OP-ID-WT SWP ALIGN-L IO-DEP-V IO-ID-WT

stafʕa:l/ + {a,
tu, u, na, …}
a. ☞ <s.
(táf )ʕala, s.
(tàf )(ʕál)tu,
(jàs).(táf )ʕilu, 4 4
(jàs).(tàf )
(ʕíl)na, …>
b.
<.si(táf )ʕala, .si
(tàf )(ʕál)tu,
(jà.si)(táf )ʕilu, 2! 4 4
(jà.si)(tàf )
(ʕíl)na, …>
c. <s.(tàf )
(ʕá:)la, s.(tàf )
(ʕál)tu,
(jàs).(tàf )(ʕí:)lu, 4! 4 2
(jàs).(tàf )
(ʕíl)na, …>
Page 46 of 71
Optimal Paradigms
{ja, ta, …} + / OP-DEP-V OP-ID-WT SWP ALIGN-L IO-DEP-V IO-ID-WT

stafʕa:l/ + {a,
tu, u, na, …}
d.
<.si(táf )ʕala, .si
(tàf )(ʕál)tu,
(jàs).(táf )ʕilu, 4! 2 2 4
(jàs).(tàf )
(ʕíl)na, …>
e. <.si(tàf )
(ʕá:)la, .si(tàf )
(ʕál)tu,
(jàs).(tàf )(ʕí:)lu, 4! 4! 2 2 2
(jàs).(tàf )
(ʕíl)na, …>
Page 47 of 71
Optimal Paradigms
To keep the tableau reasonably sized, the actual count of

violation-marks is reported, and the only candidates
considered are those that obey the undominated marked-ness
and IO faithfulness constraints (e.g. *µµµ]σ and IO-MAX-V).
The winner, (27a) (=(18a)), has a level paradigm at the
expense of poor alignment and unfaithfulness to the input’s
long vowel. Its first competitor, (27b) (=(18b)), improves
alignment by epenthesizing a vowel after the stem-initial
consonant. Though epenthesis is in general (p.199) possible
because of IO-DEP-V’s low rank, it is not permitted here
because it introduces violations of SWP. The remaining
candidates, (27c–e), present various ways of achieving better
alignment, satisfaction of SWP, and greater faithfulness to the
input’s long vowel. None survives, however, because all incur
violations of one or both OP faithfulness constraints.
The winning candidate (27a) illustrates the main point of this

discussion: different paradigm members can act as attractors
with respect to different phonological properties. The prefixed
imperfectives are acting as attractors with respect to the
phonology of the left edge, blocking epenthesis in the
perfective via OP-DEP-V. The forms with C-initial suffixes, both
perfective and imperfective, are acting as attractors with
respect to the phonology of the right edge, forcing vowel
shortening via OP-IDENT-WT. There is no identifiable base to
charge with the responsibility of accounting for both these
attraction effects.
The diehard supporter of LP would be forced to scour the

paradigm looking for a suitable base form that has both a CV-
prefix and a C-initial suffix and then derive all other forms
from that. It is possible to find such a form—the 2nd and 3rd
feminine plural tastafʕilna and jastafʕilna—but LP can offer no
principled explanation as to why this form is chosen as the
base. It certainly seems far-fetched to claim, as the LP
supporter must, that feminine plural verbs are morphologically
less marked than any other members of the paradigm.
This is not to say that description of the Arabic facts is beyond

the power of LP. For example, Bobaljik (2002) proposes to
analyze the right-edge stem restriction with cyclic closed-
syllable shortening: /faʕa:l/ shortens to faʕal on the first cycle,
and then the final syllable is opened up by the addition of a V-
initial suffix on the second cycle, yielding faʕala. This analysis
requires the stipulation that verb stems but not noun stems
Page 48 of 71
Optimal Paradigms
are cyclic domains, since there is no shortening in faʕa:l-u.

Mere description is possible, then, but true explanation
remains elusive: the difference between nouns and verbs is
what we seek to explain, not to stipulate. In the OP account,
this difference is derived from the lexical fact of what the
affixes are, which any analysis must specify. The array of
results and predictions obtained from the OP model shows
what kinds of explanations it is capable of.
8.4.4 Domination of OP Faithfulness

The fundamental thesis of OT is that constraints are violable.
Violation is never gratuitous, but a constraint must be violated
if all compliant candidates have been ruled out by higher-
ranking constraints. Any proposed revision to the OT
constraint set, then, must be examined through the lens of
violability.
The OP faithfulness constraints in (21) are unviolated in

examples seen thus far. Through ranking permutation, there
are languages where these same constraints are crucially
dominated and not visibly active. For example, the modern
Arabic dialects permit intraparadigmatic vowel/zero
alternations, which violate OP-DEP-V. More importantly,
ranking permutation also predicts a middle ground, where OP
faithfulness (p.200) constraints are visibly active in some
circumstances but not others. In fact, this is the situation in
Classical Arabic.
The analysis of Arabic thus far has focused on what are

traditionally called sound verbs (‘sound’ in the sense of
healthy). As we have seen, sound verbs resist various
intraparadigmatic alternations. The so-called weak verbs have
complex alternations, however, including some that sound
verbs avoid. The weak verbs are identifiable on phonological
grounds and fall into two classes: geminate verbs, whose last
two root consonants are identical (e.g. /smm/ ‘poison’); and
verbs with a high glide w or j as one of their underlying root
consonants (e.g. /wld/ ‘bear a child’, /qwm/ ‘rise’, /rmj/
‘throw’). The analysis of these verbs, especially those with a
glide, presents many difficult questions (see Rosenthall 2002
for recent discussion). Here, I will focus on just one
alternation that involves the geminate verbs.
Page 49 of 71
Optimal Paradigms
As we have seen, OP-DEP-V is an active, high-ranking

constraint in the grammar of Classical Arabic, but it is not
necessarily nor in fact unviolated. The paradigms of geminate
verbs exhibit vowel/zero alternations, indicating that OP-DEP-
V is dominated (as is OP-MAX-V). (For another case where a
paradigm-leveling constraint is crucially dominated, see
Raffelsiefen (2000: 153–5).) In geminate verbs, the identical
consonants are fused into an actual geminate unless a C-initial
suffix follows.
(28) Vowel/Zero Alternations in the Verb
a. Biliteral roots (McCarthy 1981)
/samam/
samamtu ‘I poisoned’
samma ‘he poisoned’
/ja-smum/
jasmumna ‘they (f.) will poison’
jasummu ‘he will poison’
b. Ninth and eleventh conjugations
/ħmarar/
ħmarartu ‘I reddened’
jaħmarirna ‘they (f.) will redden’
ħmarra ‘he reddened’
jaħmarru ‘he will redden’
Stems of the form … VCiVjCi show deletion of Vj before vowel-

initial suffixes, while stems of the form … CCiVjCi metathesize
Vj with the preceding consonant in the same context.
The markedness constraint responsible for this alternation is

something of a vexed question (see Gafos 2001 and Rose 2000
for proposals). The configuration being ruled out is .CiV.CiV,
with identical consonants in the onsets of successive syllables.
Presumably, this is connected with dissimilatory processes in
other languages, but the details of this connection are
obscure. Having nothing more to offer at this time, I will
simply invoke the ad hoc constraint *.CiV.CiV.
Page 50 of 71
Optimal Paradigms
(p.201) The constraint *.CiV.CiV must dominate the Input–

Output faithfulness constraint IO-MAX-V to allow for unfaithful
mappings like /hmarar-a/ → hmarra. It must also dominate the
intraparadigmatic faithfulness constraint OP-MAX-V for the
same reason. Tableau (29) supplies the details of these ranking
arguments.
(29) *.CiV.CiV Ȧ IO-MAX-V, OP-MAX-V
{ja, ta, …} + /ℏmarar/ + {a, *.CiV.CIO-

iV OP-
tu, u, na, …} MAX-V MAX-V
a. ☞ <ℏmarra, ℏmarartu, ** ****

jaℏmarru, jaℏmarirna, …>
b. <ℏmarara, ℏmarartu, **!

jaℏmariru, jaℏmarirna, …>
In the losing candidate (29b), the underlying stem shape

remains intact throughout the paradigm, so both IO and OP
faithfulness constraints are obeyed. But the price is fatal
violation of the markedness constraint *.CiV.CiV in all
paradigm members with V-initial suffixes. The winner (29a)
avoids the marked structure by deleting a vowel despite the
resulting imperfect uniformity of the paradigm. A parallel
argument can be made for metathetic forms like /jasmum-u/ →
jasummu.
In the parlance of rule-based phonology, the alternation in (29)

would be called ‘normal’ application—there is neither
underapplication nor overapplication of vowel deletion
between identical onset consonants (cf. s. 8.2).
Underapplication is out because it violates top-ranked
*.CiV.CiV. Paradigm uniformity could in principle be achieved
by overapplication, however, yielding paradigms like
*<ℏmarra, ℏmarrtu, jaℏmarru, jaℏmarrna, …> with the vowel
deleted between the identical consonants before both V-initial
and C-initial suffixes. Overapplication is, however, ruled out by
markedness considerations: tautosyllabic geminates violate
the constraints introduced in s. 8.4.1, *µµµ]σ and *APP-σ.
This example shows that perfect paradigmatic uniformity is

not always achieved, even in languages where the OP
faithfulness constraints are visibly active. Whether, where, and
Page 51 of 71
Optimal Paradigms
when there is paradigm uniformity is a matter of constraint

interaction, as always in OT.
8.5 Further Consequences

The OP model makes typological predictions that happen not
to be exemplified in the analysis of Classical Arabic. This
section discusses two of them, both of which were introduced
in s. 8.2: the potential for majority-rules effects and the claim
that (p.202) underapplication effects in inflectional
morphology always involve blocked overapplication.
8.5.1 The majority rules

In OT, violations of a markedness constraint are summed over
all instances of a marked structure in a form. For example, if a
word contains five syllables, three of which lack onsets, then it
will receive three violation-marks from the constraint ONSET.
In OP, constraint violations are also summed over all the forms
in a paradigm. For example, if a paradigm has three members,
one with a single onsetless syllable and another with two
onsetless syllables, then it receives three violation-marks from
ONSET.
This calculus of violation means that it should be possible to

see majority-rules effects. In a majority-rules effect, the
pattern that is most common in a paradigm acts as an
attractor to other paradigm members. Under certain rather
special ranking conditions, majority-rules effects are predicted
to occur by the OP model. Here I argue that this aspect of OP
can solve a long-standing problem in the phonology of
Moroccan Arabic (Benhallam 1990; Boudlal 2001; Harrell
1962).
The distribution of ǝ in Moroccan Arabic is almost fully

predictable. Two undominated markedness constraints
establish the milieu: ǝ is banned from open syllables (*ǝ]σ) and
clusters of three consonants are prohibited (*CCC).24 When a
word contains three consonants and no other vowels, there are
in principle two ways to satisfy these constraints: CǝCC and
CCǝC. In these words, the choice between CǝCC and CCǝC is
automatic, but the conditions are different for nouns and
verbs.
In nouns, with few exceptions, the choice between CǝCC and

CCǝC is determined by sonority conditions:
Page 52 of 71
Optimal Paradigms
(30) Moroccan Arabic CCC

nouns
a. C1ǝC2C3 if C2 > C3 in sonority or C2C3 is

a geminate
kǝlb ‘dog’
bǝrd ‘wind’
dǝnb ‘sin’
∫ǝm∫ ‘sun’
lǝʕb ‘game’
mwǝxx ‘brain’
b. C1C2ǝC3 if C2 ≤ C3 in
sonority
r3ǝl ‘leg’
ktǝf ‘shoulder’
ħbǝl ‘rope’
bƔǝl ‘mule’
wtǝd ‘peg’
kfǝn ‘shroud’
(p.203) To avoid a digression, I will defer detailed analysis

and call the constraint(s) responsible for this pattern
SonCon.25
In verbs, however, only the pattern CCǝC is possible,

regardless of sonority: ktǝb ‘he wrote’, ∫rǝb ‘he drank’, kbǝr
‘he grew’, r3ǝʕ ‘he returned’, lʕǝb ‘he played’, rbǝtʁ ‘he tied’.
This can lead to noun/verb minimal pairs when C2 is more
sonorous than C3: ∫ǝrb ‘drinking; love of alcoholic drink’ v.
∫rǝb ‘he drank’. So SONCON is crucially dominated by some
constraint that only affects verbs. This difference between
nouns and verbs is a classic puzzle in the study of this
language.
The explanation for this difference comes from differences in

noun and verb inflection. The modern Arabic dialects,
including Moroccan, lost the case-marking inflection of the
classical language. Clitic pronouns are suffixed to the noun,
but clitics are outside the inflectional paradigm. Nouns like
Page 53 of 71
Optimal Paradigms
those in (30) do not form plurals by suffixation. In short, there

are no inflectional suffixes on the nouns of interest, so their
paradigms contain only a single member, the noun stem itself.
In that situation, the OP faithfulness constraints are vacuously
satisfied, so they can have no effect on the outcome. The
constraints subsumed by SONCON are the sole determinants
of the distribution of ǝ in nouns.
The Moroccan verbal paradigm, though, retains some of the

richness seen in Classical Arabic. In (31), the full paradigm of
the perfective verb is shown, organized according to the
position of ǝ in the stem.
Page 54 of 71
Optimal Paradigms
(31) CCǝC CǝCC
∫rǝb 3 m. sg. pf.
∫rǝb-t 1 c. sg. pf. ∫ǝrb -u 3 c. pl. pf.
∫rǝb-na 1 c. pl. pf. ∫ǝrb-ǝt 3 f. sg. pf.
∫rǝb-ti 2 c. sg. pf.
∫rǝb-tu 2 c. pl. pf.
Page 55 of 71
Optimal Paradigms
Except for unaffixed ∫rǝb, the undominated markedness

constraints *ǝ]σ and *CCC fully determine the distribution of
schwa in verb stems throughout the paradigm. The CǝCC stem
occurs before V-initial suffixes, where CCǝC cannot appear
because ǝ is banned from open syllables: *∫rǝbu. Before a C-
initial suffix, CCǝC is required and CǝCC is impossible, since
triconsonantal clusters are prohibited: *∫ǝrbna.
Given this basically phonological distribution, why does the

unaffixed 3rd masculine singular perfective verb consistently
have CCǝC shape, instead of accommodating (p.204) to
sonority conditions as otherwise identical nouns do? The
answer is that verbs, unlike nouns, have non-trivial paradigms,
so OP faithfulness is potentially active, and the relevant OP
constraint is ranked above the sonority constraints that are
determinative in nouns. The tableau in (32) presents the
overall framework of the analysis.26
(32)
Page 56 of 71
Optimal Paradigms
/∫ǝrb/ + {t, na, *ǝ]σ *CCC OP-MAX-V SON CON IO-MAX-V IO-DEP-V
ti, tu, u, ǝt}
a. ǝ 20 *’s * 5 *’s 5*’s

<∫rǝb,∫rǝbt,∫rǝ
bna, ∫rǝbti,
∫rǝbtu, ∫ǝrbu,
∫ǝrbǝt>
b. 24 *’s! 4 *’s 4 *’s

<∫ǝrb,∫rǝbt,∫rǝ
bna,
∫rǝbti,∫rǝbtu,∫ǝ
rbu, ∫ǝrbǝt>
c. <∫rǝb, **! * 7 *’s 7 *’s

∫rǝbt, ∫rǝbna,
∫rǝbti, ∫rǝbtu,
∫rǝbu, ∫rǝbǝt>
d. <∫ǝrb, ****!
∫ǝrbt, ∫ǝrbna,
∫ǝrbti, ∫ǝrbtu,
∫ǝrbu, ∫ǝrbǝt>
Page 57 of 71
Optimal Paradigms
The input is here taken to be /∫ǝrb/, but /∫rǝb/, /∫ǝrǝb/, or

even /∫rb/ would do just as well, because the IO faithfulness
constraints are ranked at the bottom. Candidates (32c, d) have
leveled the paradigm to avoid all ǝ/zero alternations. Neither
is satisfactory, because both contain forms like *∫ rǝbu and *∫
ǝrbti that violate undominated markedness constraints against
ǝ in open syllables and triconsonantal clusters.
The phonotactically viable candidates, then, are (32a, b),

which differ only in whether the 3rd person masculine singular
verb is ∫rǝb or *∫ǝrb. OP-MAX-V, the next constraint in the
ranking, favors ∫rǝb because the CCǝC stem pattern is better
represented in the rest of the paradigm. The calculus of
violations proceeds like this. In (32a), there are five forms with
the stem ∫rǝb- and two forms with the stem ∫ǝrb-. This makes
for a total of twenty (5 × 2 × 2) ordered pairs like ( ∫rǝb,
∫ǝrbu) and ( ∫ǝrbu, ∫rǝb) where there is an intraparadigmatic
vowel/zero alternation and hence a violation of OP-MAX-V. In
(32b), on the other hand, there are four forms with the stem
∫rǝb- and three forms with the stem ∫ǝrb-. This makes for a
total of twenty-four (4 × 3 × 2) ordered pairs with a vowel/
zero alternation. Since OP constraints disfavor alternation
within (p.205) the paradigm as a whole, they can detect even
this modest advantage that comes from assigning unaffixed
∫rǝb to the more populous class of stem shapes.27 (See
Kenstowicz (this volume) for an alternative analysis of this
problem—Eds.)
We know that SONCON favors ∫ǝrb over ∫rǝb because ∫ǝrb is

the noun derived from this root, and sonority is decisive for
stem-shape in nouns. Because OP-MAX-V is successful in
favoring ∫rǝb as the verb form, it must dominate SONCON.
Nouns, however, have paradigms with only a single member,
so OP-MAX-V is vacuously satisfied. This leaves the choice of
noun stems up to the low-ranking SONCON, which emerges to
favor a sonority-based distribution of ǝ in nouns. To
paraphrase Thoreau, nouns constitute a majority of one.
This result about majority-rules effects, though it has some

intuitive appeal, is rather surprising, since it might seem to
imply a vote-counting approach to phonology. As usual in OT,
however, it is not counting but comparison that is crucial: the
paradigm in (32a) is better than the one in (32b) according to
Page 58 of 71
Optimal Paradigms
OP-MAX-V; the absolute number of violations is not given any

interpretation by the theory.
Majority-rules effects seem to be unusual, and this may be

because they are permitted by the OP model only when three
special conditions are met simultaneously:
(i) The competing attractors must not differ in

markedness. The competing paradigms in (32a, b)
differ in which stem-form is acting as the attractor,
CCǝC or CǝCC. As I have argued (ss. 8.2, 8.4.2), the OP
model entails attraction to the unmarked: the winning
attractor better satisfies the markedness constraints, as
ranked in the language in question, than its
competitors. A majority-rules effect is possible, then,
only when the markedness constraints ranked above OP
faithfulness do not favor one attractor or the other.
That is the case in (32a, b), since these two paradigms
equally satisfy the top-ranked markedness constraints
*ǝ]σ and *CCC.
(ii) Total leveling of the paradigm must be ruled out by
constraints ranked above OP faithfulness. If there are
viable candidates with no intraparadigmatic
alternations, then OP faithfulness is fully satisfied and
the majority becomes unanimity. In Moroccan Arabic,
the candidates with level paradigms (32c, d) violate
undominated markedness constraints so they are non-
viable.
(iii) Because a majority-rules effect involves
performance on a single OP constraint, it follows that
the competing attractors must have the same kind of
alternation, so their competition on that specific
constraint is decisive. Moroccan Arabic meets this
condition because the competing attractors CCǝC and
CǝCC involve the same alternation of ǝ with zero. A
majority-rules effect is not predicted when the
competing attractors exhibit different alternations,
such as CCǝC and CC, the latter with consonant
deletion.
(p.206) Classical Arabic, as analyzed in s. 8.4, does not

exhibit majority-rules effects. That is because it does not meet
either of the first two conditions. The first condition for
majority-rules says that the competing attractors must not
differ in markedness. But in Classical Arabic, as shown in (25),
Page 59 of 71
Optimal Paradigms
the high-ranking markedness constraint SWP favors one

attractor over the other. Attraction to the unmarked invariably
trumps majority-rules because of the way OT works:
satisfaction of a higher-ranking constraint always takes
precedence over minimizing violation of a lower-ranking
constraint. The second condition for majority-rules says that
the competing paradigms must not be level, but in Classical
Arabic the competition is between paradigms that have been
leveled in the relevant dimension. Therefore, majority-rules
effects are neither expected nor observed in Classical Arabic.
8.5.2 Underapplication in inflectional paradigms

In ss. 8.2 and 8.4, I showed that the OP model produces
overapplication effects, limiting underapplication to situations
where overapplication is blocked by some high-ranking
constraint. This is a strong claim, though to grasp it fully it is
necessary to be clear about what over- and underapplication
mean in the context of a constraint-based theory like OT.
The over- and underapplication terminology is inherited from

rule-based phonology, specifically from Wilbur’s (1974) work
on reduplication/phonology interactions. A rule is said to
overapply if its structural description is met in only one
reduplicative copy but it applies in both: a process of coronal
palatalization overapplies in hypothetical /RED-pat-i/→pat∫-
pat∫i. A rule is said to underapply if its structural description
is met in only one copy but it applies in neither: if a language
has an otherwise general process of coronal palatalization, /
RED-pat-i/→pat-pati is a case of underapplication.
This terminology was transposed to the study of reduplication

in OT by McCarthy and Prince (1995, 1999) and further to the
study of Output–Output faithfulness by Benua (1995, 1997a).
OT’s nearest analogue to a process is a hierarchy where some
markedness constraint M is crucially ranked above an
antagonistic faithfulness constraint F.28 M can then force an
unfaithful, F-violating mapping. A process, in this sense,
overapplies if the same mapping occurs in another member of
the reduplicative or Output–Output pair where there is no
danger of violating M: in pat∭-pat∭i, the markedness
constraint against ti accounts for the second t∭, and this
effect carries over to the first t∭ even though it is not
followed by i. A process underapplies if M is violated in one
member of the pair because it is vacuously satisfied in the
Page 60 of 71
Optimal Paradigms
other: in pat-pati, the markedness constraint against ti is

breached because the first t is not followed by i.
The theory of reduplication in McCarthy and Prince’s work

and the OP model presented here have a common
characteristic: true underapplication is predicted not to occur.
The reason is that underapplication always competes with
overapplication, (p.207) since both achieve perfect identity
between the reduplicative copies or the paradigm members.
And overapplication normally wins this competition because it
satisfies the markedness constraint responsible for the process
but underapplication does not. For instance, pat∭-pat∭i and
pat-pati perform equally on base-reduplicant identity
constraints, but the first is more harmonic because it also
satisfies the markedness constraint against ti. OP has the same
basic logic as based-reduplicant identity, so it similarly
predicts that underapplication is only possible in inflectional
paradigms when overapplication is ruled out by some high-
ranking constraint.29 See ss. 8.4.1 and 8.4.2 for
exemplification.
Underapplication certainly occurs in derivational morphology,

where it is predicted by TCT and other theories that have a
notion like the base. But OP is a theory of the phonology of
inflection, not derivation, which leads to a typological
question: does true underapplication ever occur in inflectional
paradigms, contrary to this prediction of OP? A possible case
comes from Tiberian Hebrew. Benua (1997a: ch. 4) argues that
vowel epenthesis in Hebrew, though it applies normally in
nouns, underapplies in verbs in situations where it threatens
paradigm uniformity. Here, I will sketch an analysis that is
consistent with the principles of OP: epenthesis underapplies
in verbs because overapplication is blocked by a higher-
ranking constraint.30
In general, Tiberian Hebrew prohibits word-final consonant

clusters, resolving them by vowel epenthesis: /malk/→ ′mεlε k
‘king’; /dammaɕq/ → dam′mεɕεq ‘Damascus’.31 (For details, see
Coetzee 1999a, b; Garr 1989; McCarthy 1979; Malone 1993;
Prince 1975 ; or Gesenius 1910.) Verbs, however, can end in a
cluster under certain conditions. Certain verbs—those with
roots ending in w or j—have vowel-final stems on the surface: ji
b′kε ‘he will cry’. In the inflectional categories known as the
jussive and wa:w-consecutive,32 the final vowel of the
Page 61 of 71
Optimal Paradigms
imperfective is truncated, leaving a word-final cluster in its

wake: ′je: bk ‘let him cry’. (For further details, see Benua
1997a; Prince 1975; Speiser 1926; or the handbooks.)
This looks like underapplication. An otherwise general process

of epenthesis is blocked in words like ′je: bk in order to
maintain similarity with its paradigmatic comrade ji b′kε. The
candidate where epenthesis has applied normally, *′ji bεk, is
ruled out because it has a vowel between b and k that has no
correspondent elsewhere in the paradigm.
(p.208) The basic principles of OP entail that

underapplication is possible only when over-application is
ruled out by some higher-ranking constraint. In other words, if
OP is right, there must be some constraint that rules out the
paradigm *<ji bǝ′kε, ′ji bεk, …> in favor of <ji b′kε ′je: bk, …
>.33 And in fact there is: ji b ǝ′ kε is out because it has ǝ in a
VC—CV context, which is generally impossible in Tiberian
Hebrew. In other words, the constraint against schwa in this
context crucially dominates the markedness constraint against
final clusters. The tableau at (33) is intended only to show the
logic of the argument, using ad hoc markedness constraints.
(33)
*VCǝCV OP-MAX-V *CC#
a. ☞ <jib′kε, ′je:bk, …> * *
b. <jibǝ′kε, ′jibεk, … > *! *
c. <jib′kε, ′jibεk, …> **!
All three candidates violate OP-MAX-V at least once because

the jussive is related to the imperfective indicative by
truncation (cf. Horwood 1999). But (33c) incurs an additional
violation of this constraint because it has a vowel/zero
alternation in the b — k context. Paradigm (33b) shows
overapplication of epenthesis, but this is ruled out because it
requires ǝ in an impermissible context. That leaves the
candidate with underapplication, (33a), as the winner despite
its final cluster.34
There is no evidence of true underapplication here. Rather,

this is a case of underapplication as an alternative to blocked
overapplication, much like the Arabic example of s. 8.4.2.
Other cases of inflectional underapplication may very well
Page 62 of 71
Optimal Paradigms
exist. The remarks here certainly do not address them all, but
rather they suggest the overall approach that can be taken
within the strictures of OP.
8.6 Conclusion
In this chapter, I have introduced the Optimal Paradigms
model of the interaction of phonology with inflectional
morphology. Candidates in OP consist of entire inflectional
paradigms. Within each candidate paradigm, there is a
correspondence relation from every paradigm member to
every other paradigm member. Faithfulness constraints on this
intraparadigmatic correspondence relation resist alternation
within the paradigm.
(p.209) This model was illustrated and supported with a type

of evidence that has not figured in previous discussions, the
templatic structure of the Classical Arabic verb. A goal was to
show that certain restrictions on Arabic templates could be
derived from independently motivated constraints, as required
by Generalized Template Theory.
Some of the questions for future research are suggested by

the preliminary results reported in s. 8.5. OP predicts the
possibility of majority-rules effects and it denies the possibility
of true underapplication within paradigms. It will be
interesting to see whether these predictions are fully borne
out.
APPENDIX A: THE CLASSICAL ARABIC CONJUGATIONS
Conjugation Perfective (3 sg. m. Imperfective (3 sg. m.

active) indic. active)
1 faʕala jafʕalu
2, 5 faʕʕala, tafaʕʕala jufaʕʕilu, jatafaʕʕalu
3, 6 fa:ʕala, tafa:ʕala jufa:ʕilu, jatafa:ʕalu
4 ʕafʕala jufʕilu (from /juʕafʕilu/)
7 nfaʕala janfaʕilu
8 ftaʕala jaftaʕilu
9 fʕalla (from /fʕalala/ jafʕallu (from /jafʕalilu/

—cf. fʕalaltu (1 c. — cf. jafʕalilna (3 pl.
sg.)) f.))
Page 63 of 71
Optimal Paradigms
Conjugation Perfective (3 sg. m. Imperfective (3 sg. m.

active) indic. active)
10 stafʕala jastafʕilu
11 (rare) fʕa:lla (from /f jafʕa:llu (from /

ʕa:lala/—cf. jafʕa:lilu/— cf.
fʕa:laltu (1 c. sg.)) jafʕa:lilna (3 pl. f.))
12 (rare) fʕawʕala jafʕawʕilu
13 (rare) fʕawwala jafʕawwilu
14 (rare) fʕanlala jafʕanlilu
15 (rare) fʕanla: jafʕanli:
Quadriliteral dahra3a, tadahra3a judahri3u, jatadahra3u

1, 2
Quadriliteral dhanra3a jadhanri3u

3 (rare)
Quadriliteral dhara33a (from / jadhari33u (from /

4 (rare) dhar3a3a/— cf. jadhar3i3u/ —cf.
dhar3a3tu (1 c. jadhar3i3na (3 pl. f.))
sg.))
(Sources: McCarthy 1981; Wright 1971)
APPENDIX B: THE CLASSICAL ARABIC NOUN TEMPLATES
Triliteral: faʔfl-u, faʔal-u, fa:ʔal-u, faʔa:l-u, fa:ʔa:l-u
Quadriliteral: daħra3-u, daħra:3-u
Note: The vowel a is just a stand-in for any of the three vowels
a, i, and u. So /faʔl/, /fiʔl/, and /fuʔl/ are all licit noun stems.
Arabic also has non-templatic nouns. Templatic nouns are by
far the majority and include not only native words but also
many (p.210) loans. Non-templatic nouns are rare and are
nearly all loans. There is an independent criterion for
determining whether a noun is templatic: with few exceptions,
all and only templatic nouns form their plural by internal
change (‘broken’ plurals— McCarthy and Prince 1990a).
(Sources: McCarthy and Prince 1990a, b; Wright 1971)
Notes:
Page 64 of 71
Optimal Paradigms
(1) For example, Kenstowicz (1996) proposes that the leveling

of the prefix des- as deh- in aspirating dialects of Spanish is a
consequence of UE acting on all instances of this bound
morpheme. But there is no evidence that this leveling occurs
in the synchronic grammar: since deh- shows up in all
contexts, there is no justification for positing underlying /des-/.
Another example: Burzio (1994a: 201) proposes that Eng lish
words with the suffix -ic have penult stress to maintain
uniformity with the same words ending in -ical: e.g.
académical affects académic (cf. Chomsky and Halle 1968: 88).
One problem with this analysis is that many words in -ic have
no related form in -ical (sulfuric, Ethiopic, Olympic, Byronic) or
they have a related form that is found in dictionaries but not
widely known, such as academical, taxonomic (al), semantic (al),
prosodic (al), and genetic (al)). Another problem is that other
suffixes, such as -id and -ish, have exactly the same stress
behavior but no variants -id-al or -ish-al.
(2) Here, I assume that paradigms are ‘flat’, consisting of a list

of all paradigm members. It is conceivable, however, as John
Alderete and Diamandis Gafos point out, that paradigms have
internal hierarchical structure. For example, Latin noun
paradigms might decompose into separate subparadigms for
singular and plural: <<pater, patris, …>sg, <patre:s, patrum,
…>pl.>, (glosses: <<father (nom.), father (gen.), …>, <fathers
(nom.), fathers (gen.), …>>). It is a straightforward matter to
adapt OP to these structured paradigms: either OP faith is
violated once for every (sub)paradigm that hosts an
alternation, or there are distinct correspondence relations
(and distinct OP faithfulness constraints) within and between
subparadigms. This possibility, though certainly intriguing, will
not be pursued here because the evidence under discussion
does not require it.
(3) The violation profile of a form is a vector representing all

its constraint violations in rank order, such as <*, **, Ø, ***>
(Samek-Lodovici and Prince 1999). The violation profile of an
entire paradigm is the vector sum of the violation profiles of
all members of that paradigm: e.g. <*, **, Ø, ***> + <Ø, *, **,
Ø> = <*, ***, **, ***>.
Page 65 of 71
Optimal Paradigms
(4) Limitation of the correspondence relation to the shared

lexeme recalls Alderete’s (1998) notion of stem-to-stem
correspondence, which is required in his accounts of pre-
accentuation in Cupeño and accent shift in Japanese. Since
ℜop is a relation on P×P, every member of a paradigm is also in
correspondence with itself. This is harmless, since self-
correspondence can never lead to faithfulness violations.
(5) Technically, the paradigm <mat, mat∫i> receives two

marks from OP-IDENT(high), one for the matℜop mat∫i
correspondence relation and the other for its symmetric
counterpart. The paradigm <ma, mat∫i> incurs one violation
of OP-MAX for the ma ℜop mat∫i relation and one violation of
OP-DEP for the mat∫i ℜop ma relation.
(6) Jeroen van de Weijer and the members of the Leiden

Phonology Group raise an objection: the phonological effects
of derivational and inflectional morphology are sometimes the
same. For example, English stress-neutral suffixes can be both
derivational (-ness) and inflectional (-ing). This is exactly as
the OP model predicts: under ranking permutation, we expect
to find cases where OP faithfulness constraints, which pertain
to inflection, and OO faithfulness constraints, which pertain to
derivation, are ranked similarly with respect to markedness.
The model also predicts, however, that inflection and
derivation can act differently: only derivation can show true
underapplication effects, and only inflection can show
phonological influences from different paradigm members
simultaneously (see s. 8.4.3).
(7) Works discussing Generalized Template Theory and

kindred notions include: Alderete et al. (1999); Carlson (1998);
C. W. Chung (1999); Downing (1999b); Gafos (1998);
Hendricks (1999); Ito, Kitagawa, and Mester (1996); McCarthy
and Prince (1994a, b, 1995, 1999); Spaelti (1997); Struijke
(1998, 2000a, b); Urbanczyk (1996, 1999); Ussishkin (1999,
2000); and Walker (2000).
(8) This is properly true only for ‘sound’ verbs, those without
glides or double consonants in the root. See s. 8.4.4.
(9) Troughout, I will give examples as they appear in standard

handbooks, using the citation triliteral root √fʕl ‘do’ and the
citation quadriliteral root √dħr3 ‘roll’.
Page 66 of 71
Optimal Paradigms
(10) The vowel of the imperfective prefix is u in the passive

voice. It is also u in conjugations II, III, and IV and the first
quadriliteral conjugation. The suffix -u is replaced by -a in the
subjunctive.
(11) Superheavy syllables can occur in absolute phrase-final

position (‘in pause’). [CV:C]σ syllables can also occur when the
coda C is the first half of a geminate: masa:mmi: ‘porous’. I
will disregard these complications here, since they do not bear
on the main point.
(12) After an earlier version of this chapter was circulated,

Diamandis Gafos provided me with a copy of a manuscript
(Gafos 2001) in which a similar argument is presented. This
convergence of independent lines of research is perhaps an
indication that this analysis is on the right track.
(13) This analysis, then, uses the form of the inflectional

morphemes to predict properties of the stem templates. Why
should the explanation go this way? That is, why stipulate the
form of the inflectional morphemes and then use that to
explain the stem templates, instead of stipulating the stem
templates and using them to explain the inflectional
morphemes? The inflectional morphemes are a closed class
and they must be listed in any case, but the stems are an open
class. The grammar, then, is responsible for explaining which
stem shapes are and are not permitted, but it is not
responsible for explaining why the handful of noun inflections
are all vowel-initial; this is just an accident. (Thanks to Linda
Lombardi for raising this point.)
(14) Vowel length alternations are observed in the paradigms

of verbs like jaqu:mu/jaqumna ‘he arises’/ ‘they (f.) arise’. See
s. 8.4.4.
(15) When the whole paradigm is considered, long a: stands in

correspondence with short a many, many times. The perfective
and imperfective indicative paradigm has ten forms with C-
initial suffixes and fourteen forms with V-initial suffixes. There
are, then, 280 ordered pairs where a: stands in
correspondence with a (280 = 10 × 14 × 2, because the
correspondence relation is fully symmetric).
(16) This statement does not hold for words whose final root
consonant is a high glide. See s. 8.4.4.
Page 67 of 71
Optimal Paradigms
(17) Similarly, the UE model of Kenstowicz (1996) allows a

particular member of the paradigm to have priority. This is
invoked in Kenstowicz’s analysis of over- and underapplication
of a vowel-raising process in Polish diminutive paradigms. The
problem with this example is that the raising process itself is
exceedingly irregular and unproductive (Buckley 2000;
Gussmann 1980: ch. 4; Sanders 2002), so it does not supply
convincing evidence for base priority within inflectional
paradigms.
(18) The responsible alignment constraint may actually be the

more general ALIGN-L(Stem, PrWd), as argued in McCarthy
and Prince (1993a, 1994b). This matter, though relevant to
foundational issues in Generalized Template Theory, is
tangential to the point here.
(19) I am grateful to Jonathan Bobaljik, Yoonjung Kang, and

the editors of this volume for their questions about a previous
analysis, which prompted some significant revisions.
(20) The first conjugation of the Arabic verb has a vowel/zero

alternation faʕala/jafʕalu that has sometimes been taken as
evidence for an active syncope process (Brame 1970;
McCarthy 1981). If this is indeed a syncope process, then it is
completely isolated, since there are no other such alternations
in the classical tongue. It seems more plausible to regard the
alternation as allomorphic. Allomorphy is most often observed
in high-frequency, underived forms, such as the English strong
verbs. The Arabic first conjugation is similar: it is the most
common conjugation and it is the one that is unmarked
morphologically.
(21) The argument in (18) does not depend on knowing

whether Classical Arabic had trochaic or iambic feet. The foot
(jàsi) violates SWP either way. This is a welcome result, since
the evidence is inconsistent about whether Classical Arabic
feet were iambic or trochaic.
(22) A possible variation on candidate (20b) is *<f.(ʕá)lu, f.

(ʕá)la, …>, depending on how NON-FINALITY is ranked with
respect to FT-BIN. Either way, SWP is violated.
Page 68 of 71
Optimal Paradigms
(23) Versions of stratal or cyclic OT can be found in the

following works, among others: Bermúdez-Otero (1999); Black
(1993); Cohn and McCarthy (1994/1998); Hale and Kissock
(1998); Hale, Kissock, and Reiss (1998); Ito and Mester
(2002); Kenstowicz (1995); Kiparsky (1998); McCarthy
(2000b); McCarthy and Prince (1993b); Potter (1994); Rubach
(2000); and many of the contributions to Hermans and van
Oost-endorp (1999) and Roca (1997).
(24) Though I use the constraint *ǝ]σ for simplicity, the

limitation of ǝ to open syllables should perhaps be derived
from constraint interaction, as Diamandis Gafos points out. If
*COMPLEX-ONSET dominates NO-CODA, then e.g. kǝtbu is
more harmonic than *ktǝbu. See Gafos (2002) for more about
the phonetics and phonology of ǝ in Moroccan Arabic.
(25) Maria Gouskova suggests the following analysis of the

nominal pattern. Assume that the consonant immediately
following ǝ is a mora-bearing coda. Cross-linguistic evidence
shows that there is a constraint favoring mora-bearers of
higher sonority (Zec 1995), and this constraint will prefer
C1ǝC2C3 just in case C2 is more sonorous than C3. Then
C1C2ǝC3 is favored in the equal-sonority condition if
*COMPLEX-CODA dominates *COMPLEX-ONSET.
(26) For simplicity, I present the analysis here using only the
perfective verb. The full paradigm includes the imperfective as
well, as I have argued for Classical Arabic. The result still goes
through when the imper-fective is considered, though some
additional analysis is required to account for prefixed forms
like nǝ∫rǝb ‘I drink’.
(27) Bobaljik (2002) notes a related prediction: words with

defective paradigms can reverse the usual majority and
thereby exhibit a different phonological pattern. Since
majority-rules effects are rare for reasons given in the text,
and defective paradigms are also quite unusual (and usually
principled (Hetzron 1975), which can affect this prediction),
crucial examples will not be easy to find.
(28) This is a necessary but not a sufficient condition for a

process. See McCarthy (2002: 67–8) for the full story.
Page 69 of 71
Optimal Paradigms
(29) Allophonic processes can pose a trap for the unwary by

creating the illusion of underapplication (McCarthy and Prince
1995: 355–9,1999: 285–9). For example, Tokyo Japanese has
an alternation between g initially and ŋ medially: gai-koku
‘foreign country’ v. koku-ŋai ‘abroad’. In reduplicated
mimetics, there is g initially and medially: gara-gara ‘rattle’. At
first glance, this looks like underapplication of a process
changing medial g to ŋ. A better alternative, though, is to see g
nasalization as a general process that is blocked by a
constraint against initial ŋ. In effect, it is the constraint
against initial ŋ that is overapplying. See McCarthy and Prince
(1995) and Ito and Mester (1997).
(30) Benua (1997a: Ch. 4) also discusses a case in Tiberian

Hebrew where processes underapply in the affix of the 2nd
feminine singular perfective verb allegedly to maintain
similarity with the affix of the 3rd feminine singular perfective
verb. This analysis is incompatible with OP and, as far as I can
tell, with TCT as well. Polish has also been claimed to exhibit
paradigmatic underapplication (Kenstowicz 1996). See n. 17.
(31) In Hebrew examples, underlining indicates post-vocalic

spirantization.
(32) The meaning of the jussive is hortatory. The wa:w-

consecutive is a narrative tense, always preceded by the
conjunction ‘and’, which is spelled with the letter wa:w.
(33) The vowel ε is the regular realization of ǝ in a closed

syllable (Coetzee 1999b ; Garr 1989; Prince 1975). That is why
these vowels are paired in <ji bǝkε,.′ji bεk, …>.
(34) When the final cluster would contain a sonority reversal

or a coda guttural, then candidates like (33c) win anyway:
jεgεl, *′je:gl ‘let him uncover’; ′jaʕaɕ, *′jaʕɕ ‘let him make’. This
shows that OP-MAX-V is crucially dominated by other
markedness constraints (cf. s. 8.4.4).
Page 70 of 71
Optimal Paradigms
Page 71 of 71
Paradigm Uniformity Effects Versus Boundary Effects


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Paradigm Uniformity Effects Versus

Boundary Effects
Renate Raffelsiefen (Contributor Webpage)
DOI:10.1093/acprof:oso/9780199267712.003.0009

This chapter argues for the necessity of properly
distinguishing paradigm uniformity (PU) effects from domain
effects and have proposed various criteria for doing so. Proper
distinction of the relevant phenomena is a crucial prerequisite
for all further inquiry. PU effects, once they have been
properly delimited, shed light on many aspects of lexical
structure of major concern to linguists, including the
organization of paradigms in the mental lexicon and the
question of which phonetic features are represented in the
lexicon. Mostly false conclusions will be drawn on the basis of
misclassified PU effects.
Keywords: paradigm uniformity, linguistics, paradigms, lexicon, PU effects
9.1 Introduction
Morphologically complex structures sometimes exhibit sound
properties that differ from those occurring in simplexes. The
relevant properties concern both distinctive features as in (1a)
and arguably non-distinctive features as in (1b). In (1a) the
occurrence of the diphthong before a stressless syllable closed
Page 1 of 114
by a noncoronal obstruent violates the so-called ‘Arab Rule’ (cf.

Fidelholtz 1967; Myers 1987b).1 In (lb) the phonetic contrasts
concern several properties: shyness and minus differ both in
the length of the diphthong (marked by the superscripts
‘+l’ (+long) versus ‘−l’ (–long) in the narrow transcription in
(1b)) and in the amplitude of n. Umeda and Coker (1974: 5)
characterize the phonetics of suffix-initial n as ‘more
consonantal’ as opposed to the medial n in words like minus,
which is characterized as ‘more vocalic’. The relevant contrast
is represented by ‘+f’ (‘fortition’) versus ‘−f’ (‘lenition’) in (1b).
Because of those phonetic (p.212) contrasts the words in (1b)
do not rhyme perfectly (cf. Z. D. Harris 1951; D. Jones 1956;
Wells 1990a, 1990 b).
(1)
a. /nó℧mIk/ vs. / b. /∫[á+l1n+∫ ]əs/ vs. /

kɑmIk/ m[á−l1n−f ]əs/
‘gnomic’ ‘cómic’ ‘shyness’ ‘minus’
How can the special phonology of the suffixed words in (1) be

described? Two traditional explanations invoke a preference
for uniform paradigms, that is, a preference for ‘sameness’ in
the sound structure of semantically related units of speech (cf.
Humboldt 1836) and ‘boundary signals’, which function to
mark the edges of morphological constituents (cf. Trubetzkoy
1936, 1939: 241–61).
To begin with paradigm uniformity (PU), the relevant

preference was traditionally conceived to be
‘psychological’ (i.e. ‘one meaning—one form’), inherently
conflicting with ‘physiological’ (i.e. strictly phonologically
motivated) preferences, where the conflict concerns surface
forms. As is well known, this sort of approach was entirely
dismissed in Generative Phonology (for insightful discussion,
see Vennemann 1972), but has been revived in Optimality
Theory (OT). Specifically, violations of ‘regular’ phonology due
to PU are analyzed in terms of the domination of phonological
mark-edness constraints by PU constraints or output–output
correspondence constraints (cf. Benua 1995, 1997a, b; Burzio
1996; Kenstowicz 1996; Kraska-Szlenk 1995; Raffelsiefen
1995, 2000). The effect in (1a) could accordingly be described
as in (2), where the constraint PU[NUC], which requires
Page 2 of 114
identical nuclei for words belonging to a single paradigm,

dominates ARAB, the phonological markedness constraint that
prohibits heavy nuclei in certain positions.2
(2)
Base: nó℧m+1k PU[NUC] ARAB
nɑ́m1k *!
ünó℧mik *
The ranking in (2) explains why the Arab Rule is violated in

words like b[e]sic ‘basic’ (cf. b[e]se ‘base’) and an[i:]mic
‘anemic’ (cf. an[i:]mia ‘anemia’), but not in words lacking a
relevant base such as traffic, panic, endemic, and polemic.
A perhaps novel proposal is the extension of PU to account for

the low-level phonetics illustrated in (1b) (cf. the works
discussed in s. 9.5). The idea would be to describe the length
of the diphthong in shyness (vis-à-vis the diphthong in minus)
also as a PU effect, based on vowel length in the free-standing
word shy as in (3a).
(p.213)
Steriade
(2000)
emphasizes
the
theoretical implications of a PU-based description for the
phonology–phonetics interface. To be affected by PU
constraints, phonetic features must be represented in the
lexicon. If universally non-distinctive features could be shown
to be affected by PU the distinction between phonology and
linguistic phonetics would break down and the notion of
distinctiveness would lose its relevance.
While the conclusion in the preceding sentence follows from

the premise3 the question of whether the premise applies and
a certain ‘identity effect’ indeed qualifies as a PU effect must
be carefully examined. On the second approach to
morphophonology mentioned above, that is, descriptions in
terms of ‘boundary signals’, there is no need for lexical
representation of the low-level phonetic contrasts illustrated in
(1b). Tat is, on Trubetzkoy’s view the rhymes of words like
shyness and minus are identical at the phonemic level, and the
Page 3 of 114
relevant contrasts would be analyzed as boundary signals

representing the delimitative, rather than the distinctive,
function of phonology.4 In OT, this analysis can be formalized
in terms of alignment constraints, which require the edges of
morphological constituents to align with phonological
constituents (cf McCarthy and Prince 1993a, a generalization
of Selkirk’s 1986 edge-based theory of the syntax–prosody
interface). The relevant constraints are discussed in more
detail below. For now it suffices to assume that shyness and
minus are lexically represented as in (4a), that words are
systematically aligned with phonological word boundaries
such that suffixes are excluded as in (4b), and that feet and
syllables must be properly contained within phonological
words (henceforth pwords) as in (4c) (‘ω’ = pword, ‘Σ’ = foot):
(p.214)
Given the
representations in (4c) the length contrasts of the diphthongs

as well as the contrasts in the amplitude of the n in (1b) result
from a contrast in prosodic organization. That is, the fact that
the diphthong in shyness exhibits the phonetic properties of
word-final (or foot-final) diphthongs is conditioned by the fact
that this diphthong does indeed occur in pword-final (and foot-
final) position, as opposed to the diphthong in minus. Similarly,
the fact that the n in shyness exhibits ‘initial’ phonetics (cf
Umeda and Stoker 1974: 5) results from the syllabification of
this n as an exclusively syllable-initial consonant, again as
opposed to the n in minus, which also closes the preceding
stressed syllable.5 One important difference between the two
descriptions is then that in (4) all phonetic properties of both
shyness and minus are canonical in that they are determined
by the position of speech sounds within prosodic constituents,
whereas on the PU analysis in (3a) the phonetics of shyness
involves a genuine violation of canonical phonology. A second
difference is that the description in (4) accounts for both the
Page 4 of 114
difference in the length of the diphthong and the difference in

the amplitude of n as two reflections of a single phenomenon
(i.e. the location of a specific boundary) whereas the
correlation between these two effects is in principle accidental
in terms of PU constraints. Moreover, it is debatable whether
the ‘special phonology’ of the n in the suffix -ness can be
described as a PU effect because the suffix does not relate to a
free-standing form.
It appears then that while often exhibiting converging results,

that is, ‘surface’6 resemblance among related words, PU
effects and boundary effects are fundamentally distinct. PU
effects presuppose the recognition of paradigmatic relations in
the lexicon together with a constraint that prohibits
phonological variation (at some specific level of abstractness)
among paradigm members. In terms of learnability PU effects
raise the questions of how learners know the extension of
some paradigm (i.e. how they distinguish members from non-
members) and how they identify the member which functions
as the ‘attractor’ and hence determines the direction of
leveling (cf. Albright, Ch. 2 above, McCarthy, Ch. 8 above). By
contrast, boundary effects, which I will also refer to as
‘domain effects’, presuppose knowledge of syntagmatic
structure. In terms of learnability boundary effects raise the
questions of how learners recognize morphosyntactic
structure and how they learn rules for aligning specific
morphological constituent edges with prosodic boundaries.
The question of whether and how PU effects and boundary

effects can be distinguished on empirical grounds is vital not
only for a proper understanding of PU-related issues (e.g.
Under what conditions can PU effects arise? Which members
of a paradigm can function as attractor?) but also is of crucial
significance for the phonology–phonetics distinction. Given the
apparent assumption in some recent work that any sort of
deviation from simplex phonology observed in morphologically
complex words should be analyzed in terms of PU (e.g. Hayes
2000; Steriade 2000) the (p.215) main concern of this
chapter is to develop criteria for recognizing boundary effects
(cf s. 9.2). The application of the relevant criteria to English in
s.9.3 is the basis for the discussion of major empirical
differences between PU effects and boundary effects in s. 9.4.
In s. 9.5 I discuss some descriptions that I consider marred by
Page 5 of 114
the failure to properly distinguish PU effects from boundary

effects, focusing on poorly motivated use of PU constraints. In
s. 9.6 I conclude.
9.2 GENERAL CRITERIA FOR RECOGNIZING BOUNDARY

EFFECTS
A brief review of Dixon’s (1977) work on pwords in Yidiɲ in s.
9.2.1 will serve as a basis for developing criteria for
recognizing boundary phenomena in s. 9.2.2. In s. 9.2.3 it is
shown that these criteria are not met in the work of Aronoff
and Sridhar (1983), Szpyra (1989) and Hammond (1999),
arguably as a result of failing to distinguish boundary effects
from PU effects. The question of which phonological
diagnostics are relevant for recognizing boundary effects in
English is discussed in s. 9.2.4.
Page 6 of 114
9.2.1 The pword in Yidiη

It is a commonplace observation that morphologically complex
words may violate constraints which are satisfied in simplexes.
Dixon (1977) cites the Yidiɲ words in (5):
(5)
Constraint violated:
wáηal#gimbáːl#du * LAPSE
‘boomerang— Adjacent unstressed syllables

without-ERG’ are prohibited
bigúːn#gímbal * CLASH
‘shield—without- Adjacent stressed syllables are

ABS’ prohibited
Dixon argues that stress patterns in (5) are explained by the

prosodic structure in (6), in that the relevant constraints are
satisfied within each pword.
(6) (wá ηa)ω (gimbáːldu)ω (bigúːn)ω (gímbal)ω
Dixon shows that the pwords in (6) account for clusters of

correlating prosodic properties besides stress. Moreover, he
discovers that for affixes the number of syllables determines
prosodic integration into the stem: monosyllabic affixes are
‘coherent’, that is, they form a single pword together with the
stem, whereas disyllabic affixes are ‘non-coherent’, that is,
they form separate pwords. By contrast, postinflectional
affixes, which syntactically function as modifiers (e.g. -di ‘self,
-la ‘now’, -bud y un ‘still’), always form separate pwords,
regardless of the number of syllables. Those affixes are
referred to as ‘MOD’ for ‘modifiers’ in table (7), where Dixon’s
rule is illustrated with some abstract examples.
(p.216)
(7)
a. [σσσσ]WORD → (σσσσ)ω
b. [[σσσ]STEM[σ]SUFFIX]WORD → (σσσσ)ω
c. [[σσ]STEM[σσ]SUFFIX]WORD → (σσ)ω(σσ)ω
Page 7 of 114
d. → (σσ)ω(σσσ)ω
[[σσ]STEM[σσ]SUFFIX[σ]SUFFIX]WORD
e. [[σσ]STEM[σσ]SUFFIX[σ]MOD]WORD → (σσ)ω(σσ)ω(σ)ω
The term WORD in (7) refers to the ‘grammatical word’. The

examples in (7a, b) illustrate a certain type of neutralization
resulting from cohesion, where distinct morphological
structures are mapped into identical prosodic domains.
9.2.2 Restrictions on syntax to prosody mapping

Based on Dixon’s description of Yidiη I will propose certain
criteria for restricting boundary analyses where the
restrictions concern the input, the output, and the mapping.
9.2.2.1 Restrictions on the input

The structure of the input forms must be independently
motivated. In (7) this condition is satisfied because both the
distinction between affixes and modifiers and the distinction
between monosyllabic and disyllabic affixes can be established
independent of the aspects of surface prosody to be explained
(e.g. foot structure, vowel lengthening).
9.2.2.2 Restrictions on the output

Within the theory of prosodic phonology, prosodic constituents
are defined in terms of their relation to other constituents
within the prosodic hierarchy. The relevant part of the
prosodic hierarchy is given in (8).
The prosodic
hierarchy is
subject to the
Strict Layer
Hypothesis (cf
Selkirk 1981),
which
consists of two constraints, here referred to as HEADEDNESS
(cf. Selkirk 1995) and CONTAINMENT. The definitions given
below are adopted from Nespor and Vogel (1986: 7)7
(9)
a. HEADEDNESS
Page 8 of 114
A given non-terminal unit is composed of one or

more units of the immediately lower category.
(e.g. a pword dominates at least one foot)
(p.217)
b. CONTAINMENT
A unit of a given level is exhaustively contained
in the superordinate unit of which it is a part.
(e.g. feet are properly contained within pwords)
In this study both HEADEDNESS and CONTAINMENT are

assumed to be inviolable constraints.8 Deviations from simplex
phonology caused directly by the satisfaction of either of these
constraints are referred to as boundary effects (cf s. 9.5). For
instance, the occurrence of stress on adjacent monosyllabic
modifiers in Yidiη results from the satisfaction of
HEADEDNESS and qualifies therefore as a boundary effect.
Page 9 of 114
9.2.2.3 Restrictions on the mapping

The mapping from syntactic to prosodic structure is stated in
terms of alignment constraints, where the left or right
boundary of some morphosyntactic category (‘GCAT’ in (10)) is
required to align with the left or right boundary of some
prosodic category (‘PCAT’ in (10)). ‘E’ stands for left or right
edge (cf. McCarthy and Prince 1993a).9
(10)
a. ALIGN (GCat, E; PCat, E)
b. ALIGN (PCat, E; GCat, E)
The generality of the constraint schemata in (10) conflicts with

Nespor and Vogel, who define the pword as ‘the lowest
constituent of the prosodic hierarchy which is constructed on
the basis of mapping rules that make substantial use of
nonphonological notions’ (1986: 107). I assume here that the
pword is the lowest constituent of the prosodic hierarchy
which necessarily aligns with some grammatical boundary,
which means that the constraint in (11) is also taken to be
inviolable.10
(11) ALIGN (ω, E; GCat, E)
The violability of the ‘mirror image’ constraint ALIGN (GCat,

E; ω, E) is demonstrated by the existence of cohering affixes
as in Yidiη. Assuming that constraints are violated only to
satisfy higher-ranking constraints the question arises which
constraints crucially dominate alignment constraints thereby
causing cohesion. For Yidiη, Dixon’s generalization that only
monosyllabic suffixes cohere indicates the relevance of the
phonological markedness constraint FTBIN, which requires
binary feet. Specifically, assuming that both HEADEDNESS
and CONTAINMENT are inviolable (cf. s. 9.2.2.2) the ranking
in (12) has the desired effect of integrating monosyllabic, but
not disyllabic suffixes, into the pword of the stem.
(p.218)
(12) FTBIN ≫ ALIGN (Suffix, L, ω, L), ALIGN (Suffix, L,

ω, R)
Recall that monosyllabic morphemes are not necessarily

integrated into the preceding pword but only if they are
suffixes. The regular occurrence of monosyllabic feet in
Page 10 of 114
modifiers is described by ranking the constraints ALIGN (Mod,

L, ω, L) and ALIGN (Mod, R, ω, R) above FTBIN.
9.2.2.4 Summary
It is proposed here to recognize boundary effects if and only if
all of the following criteria are satisfied:
(13)
a. Both the morphosyntactic and the
phonological structure of the input forms are
independently motivated.
b. The relation between the morphosyntactic
structure and the prosodic structure is
describable in terms of alignment constraints,
such that ‘cohesion’follows from constraint
domination.
c. Inviolability of the constraint ALIGN (ω,E;
GCat,E): pword boundariesnecessarily align with
morphosyntactic boundaries.
d. Inviolability of the constraints HEADEDNESS
and CONTAINMENT.
In s. 9.2.3 I briefly discuss some previous work on English

morphophonology in terms of pwords which fail to satisfy most
of the criteria in (13). In s. 9.3 I present an alternative
description of prosodic domains in English, where all criteria
are satisfied. Since noncohesion yields effects that
superficially resemble PU effects criterion (13b) will be
especially relevant for the empirical distinction between the
two types of effects. The working hypothesis to be motivated
in this chapter can then be stated as follows:
(14) If a morphologically complex word exhibits a sound

property absent from comparable simplexes, where this
property is necessarily entailed (in the
respectivelanguage) by a domain description satisfying
all criteria in (13), then that property should be
analyzed as a boundary effect, not a PU effect.
For instance, because of the prosodic domains described by

the constraint ranking for Yidiɲ above, a word with a
monosyllabic suffix, but not words containing disyllabic
suffixes or modifiers, may exhibit stress ‘shift’. Stress
Page 11 of 114
‘neutrality’ in the latter cases is accordingly a boundary effect,

not a PU effect.
9.2.3 Previous pword-based analyses of English morphophonology

The core idea of the descriptions of English to be reviewed
here is to use pwords to account for so-called stress neutrality.
Stress neutrality is often associated with specific affixes and
can be illustrated with the examples in (15a) versus (15b):
(15)
a. ‘stress-shifting’ b. ‘stress-neutral’ affixes

affixes
rádical+ity → rádical+ness →
ràdicálity rádicalness
phónème+ic → devélop+ment →
phonémic devélopment
sólid+ifý → solídify hóspital+íze →

hóspitalìze
Sudán+ése → seléct+ée → selèctée

Sùdanése
(p.219) All words in (15 b) exhibit some type of irregular

stress pattern, including three adjacent stressless syllables
(e.g. rádicalness), no stress on a closed penultimate syllable
(e.g. devélopment), internal sequences of unstressed syllables
in verbs (e.g. hóspitalìze), and *CLASH-violations (e.g.
selèctée). No such patterns are seen in comparable simplexes
(e.g. the verb elíminàte, not * éliminàte, guàrantée, not *
guaràntée). Aronoff and Srid-har (1983) propose to describe
stress-neutrality by excluding ‘stress-neutral’ affixes from the
pword of the stem. This proposal, which is adopted by Szpyra
(1989) and Hammond (1999), is illustrated in (16).
(16)
a. ‘stress-shifting’ affixes b. ‘stress-neutral’ affixes
(ràdicálity)ω (rádical)ωness
(phonémic)ω (devélop)ωment
(solídify)ω (hóspital)ωìze
Page 12 of 114
(Sùdanése)ω (selèct)ωée
The prosodic structures in (16) account for stress violations

and also satisfy criterion (13c) in that all pword boundaries
align with morphosyntactic boundaries. However, the other
criteria in (13) are not satisfied. Criterion (13a) is not met
because the distinction between the category ‘STEM’ and
‘WORD’ as in (17), which determines prosodic integration in
Aronoff and Sridhar’s description, is not motivated on
syntactic grounds but merely encodes the distinct
phonological effects of affixation:11
(17)
a. [radical]STEM+ity b. [radical]WORD+ness
[phoneme]STEM+ic [develop]WORD+ment
[solid]STEM+ify [hospital]WORD+ize
[Sudan]STEM+ese [select]WORD+ee
The same criticism applies to Szpyra (1989), who proposes an

algorithm for mapping the ‘syntactic’ structures in (18) into
prosodic structures. Again, the syntactic representation of the
suffix -ity with no left bracket, as opposed to the
representation of the suffixes -ness or -ize with a left bracket,
serves no other purpose than to account for stress neutrality.
(18)
a. [[radical] ity] → (radicality) ω
b. [[radical][+ness] → (radical)ω(ness)ω
c. [[alcohol][+ize] → (alcohol)ω(ize)ω
Hammond (1999) does not attempt to analyze the difference

between stress-neutral and stress-shifting suffixes in quasi-
syntactic terms but formulates the constraint (p.220)
NEUTRALITY, which says that ‘Certain affixes cannot be in the
prosodic word’ (1999: 327).12 The wording ‘certain affixes’,
which reveals the entirely diacritic character of the rule,
stands for the suffixes -ist, -ly, -ance, -ence, -ant, -ent, -er, -age,
-al, -ive, -ment, -less, -y, -able, -ish, -ism, -ed, -en, -s, -ous, -ure,
-dom, -ness, -ace, -ian, -ward, -or, -t, -ton, -ate, -ite, -ine, -hood,
-ize, -itude, -ful, -eme, -ship, -ette, -ee, -ée, -eer, -ade, -ine, -
esque (Hammond 1999: 325–7).13
Page 13 of 114
Violation of criterion (13a) correlates with the violations of

criteria (13 b, d ). Specifically, the structure in (19), which is
assumed in all descriptions under review, violates the
constraint CONTAINMENT, in that syllables are not properly
contained within pwords.
In the
descriptions
reviewed here
CONTAINMENT is violated whenever a consonant-final stem

precedes a stress-neutral, vowel-initial suffix. In addition,
some of Szpyra’s structures also violate the constraint
HEADEDNESS, in that pwords have no stress (e.g. (ness)ωin
(18b)). Finally, all three analyses fail to obey criterion (13b), in
that the question of why cohesion occurs is not addressed, let
alone explained.
The joint failure of these descriptions to satisfy criteria (13a,

b, d) results from a single cause, namely the idea that stress-
behavior requires a unified treatment. For the related
language Dutch, Booij (1977: 85) notes that ‘The decisive
criterion for which a boundary must be associated with an
affix is not its stress behavior but its influence on the
syllabification patterns of the complex word in which it
occurs.’ In his subsequent work he identifies the relevant
boundaries as pword boundaries (cf. Booij 1983, 1995). In s.
9.3 it is shown that once syllable structure is identified as the
main diagnostic for recognizing pwords a description for
which all criteria in (13) are satisfied becomes feasible. On
that description there are two distinct sources of stress-
neutrality (which in many cases overlap and yield identical
results): in words with noncohesive suffixes stress-neutrality is
analyzable as a boundary effect (i.e. to satisfy
CONTAINMENT) whereas in words with cohesive suffixes it
can only be a PU effect (cf. s. 9.4). Before reviewing evidence
for this proposal the precise aspects of syllable structure
serving as diagnostics for pword structure are discussed in s.
9.2.4.
Page 14 of 114
(p.221) 9.2.4 Syllable-structure Related Diagnostics for

Recognizing Pword Structure
The specific contexts for fortition versus lenition in English
simplexes are well-known indicators of syllabification
principles (cf Kahn 1976). Te relevant phonetic contrasts can
be observed for all consonants (cf. Umeda and Coker 1974)
but are most conspicuous for t. In (20a, b) aspiration, a type of
fortition, indicates onset syllabification whereas glottalization
in (20c) or flapping in (20d, e) are types of lenition, which
indicate coda association. The data in (20) accordingly suggest
that a consonant followed by a stressed vowel is syllabified
exclusively as an onset. Another consonant may intervene
provided that the cluster exhibits a sonority increase and
occurs word-initially (e.g. tr, but not tl, st). Other postvocalic
consonants are associated with coda position, where a
following vowel may give rise to ambisyllabicity manifested in
flapping (cf. Kahn 1976).
Violations of
the patterns
in (20)
abound.
Specifically,
lenition in
(21a, b) indicates coda association, although exclusive onset
syllabification is expected. Similarly, fortition in (21c) indicates
exclusive onset syllabification, even though (ambisyllabic)
coda association would be expected.
The choice of
syllable-
structure
related
properties
such as
fortition or
lenition as decisive diagnostics for recognizing pword
structure is supported by the generalization in (22), which is
free of exceptions.
(22) Unexpected lenition indicates that a grammatical

boundary follows immediately (e.g. là[r]#éx ‘late ex’,
ní[t ¬]#ràte ‘night rate’). Unexpected fortition
Page 15 of 114
indicates that a grammatical boundary precedes

immediately (e.g. prò#[th]ibétan ‘pro-Tibetan’).
Apart from satisfying criterion (13c) (i.e. necessary alignment

of pword boundaries with grammatical boundaries) the pwords
recognized on the basis of ‘unexpected’ (p.222) lenition or
fortition (as opposed to stress-related diagnostics) also
invariably satisfy CONTAINMENT. Satisfaction of
CONTAINMENT can in fact be identified as the very cause of
the ‘unexpected’ syllabification patterns illustrated in (21).
Compare for instance the structures in (23a, b), where
‘expected’ onset syllabification of the boldfaced t in night rate
would violate CONTAINMENT:
Similarly, the
t in pro-
Tibetan is
prevented
from closing
the preceding
stressed
syllable
because of
undominated
CONTAINMENT. The occurrence of lenition in late ex is less
clear because flapping possibly indicates ambisyllabicity. Here
I follow Kahn who separates the obligatory syllabification of
word-final consonants as codas (his rule II) from the ‘later’ and
optional rule of associating such consonants with following
vowel-initial words (his rule V). That is, coda association of
prevocalic t in (24b) is necessary to satisfy CONTAINMENT
and the relevant structure is preserved in registers where
perhaps ONSET is satisfied as well.
The analysis
of the
syllabification
patterns in
(23b), (24b)
as boundary
effects
correlates
with the
satisfaction of
HEADEDNESS, manifested by the stability of the weak foot in
Page 16 of 114
night rate, as opposed to nitrate (cf the variant nitr[ə] te).

Similarly, the stability of the boldfaced pretonic foot in
subeditor shown in (25a) correlates with satisfaction of
CONTAINMENT (i.e. ‘unexpected’ coda syllabification of b
before a stressed vowel). Regular phonology of pretonic
syllables within single pwords is shown in (25b), where the
(historically related) prefix is prosodically fused with the stem
(cf. also bsérvient, ‘subservient’, [səb] sídiary ‘subsidiary’).
(p.223)
The claim that

satisfaction of
CONTAINMENT correlates systematically with satisfaction of

HEADEDNESS does not entail that every monosyllabic
morpheme forms a separate pword and therefore must be
stressed. Consider again the example shyness in (26), where
fortition of n is a boundary effect which indicates that n does
not (ambisyl-labically) close the preceding syllable (as opposed
to the n in minus). Whether shyness involves recursive pword
structure as in (26a), or not, as in (26b), both CONTAINMENT
and HEADEDNESS are satisfied. Note that in (26a) the
topmost pword satisfies HEADEDNESS because it dominates a
foot (though not immediately).
While
accounting
for
‘unexpected’
lenition in
(23b), (24b),
the
syllabification
of the pword-
final
consonant as a coda raises the question of how to describe the
exceptional rhyme structure in those words. The fact that
‘superheavy’ rhymes such as [a1t] occur word-finally, but not
word-internally (cf [æ´t.] las, but not *[á1t.] las) has led some
Page 17 of 114
pho-nologists to analyze word-final consonants in English as

onsets, rather than codas (cf. Giegerich 1985; Kaye,
Lowenstamm, and Vergnaud 1990; John Harris 1994; Piggott
1999).14 In (27) it is shown that both generalizations could be
upheld if one were to distinguish between a phonological
(lexical) and a phonetic syllable, such that rhyme (p.224)
restrictions are evaluated in the lexicon. The observation that
word-final consonants pattern structurally with onsets, but
phonetically with codas, is henceforth referred to as ‘Final-C
effect’.
Given the
representations in (27), t is ‘resyllabified’ in goat, but not in

suffixed goatee. While the approach to the special status of
word-final consonants in (27a) is by no means cogent15 it is
adopted here because of the striking parallelism in the
distribution of word-final consonants and word-internal onsets
(rather than word-internal coda consonants).16 Implications of
the representation in (27) for possible PU effects are discussed
in s. 9.5.2. For now it suffices to state that syllable-based
diagnostics for pword structure satisfy criteria (13c) and
(13d). Satisfaction of criteria (13a) and (13b) is the topic of s.
9.3.
9.3 Prosodic Boundaries in English Morphology

In this section I present morphosyntactic representations of
English words, including compounds (cf s. 9.3.1), prefixed
words (cf. s. 9.3.2), suffixed words (cf s. 9.3.3) and some clitic
structures (cf. s. 9.3.4), on the basis of which prosodic
boundaries can be described in terms of interacting alignment
and markedness constraints. As a result a significant part of
English morphophonology is described without invoking PU
(or lexical strata). For more details I refer the reader to
Raffelsiefen (2004a), a further development and partial
revision of my earlier work on the pword in English (cf.
Page 18 of 114
Raffelsiefen 1993,1999a, b). Similar structures are assumed

by Hall (2001, 2002).
(p.225) 9.3.1 The prosody of compounds

9.3.1.1 Morphosyntactic structure and alignment
For compounds I assume the morphosyntactic representations
in (28a), which are regularly mapped into the prosodic
structures illustrated in (28b). The relevant alignment
constraints are stated in (28c).
(28)
a. [[la1m]WORD[la1t]WORD]WORD → b. (la1m)ω
(la1t)ω
c. ALIGN (Word,L; ω,L), ALIGN (Word,R;ω,R)
The mapping in (28) is supported by the regular occurrence of

boundary effects, including the HEADEDNESS effect (i.e. the
stability of the final foot in (29a)) and the Final-C-effect (i.e.
[a1m.l]).
Unmistakable
CONTAINMENT effects are seen in compounds like

nigh[t¬]rate ‘night rate’ or nigh[r]owl ‘night owl’.
The rule in (28) accounts for all productive compounding in

English and suffices for the purpose of this chapter. Closer
scrutiny of the data reveals a systematic asymmetry between
compound members. If the righthand member of the
compound becomes obsolete thereby losing its word status,
fusion results necessarily (e.g. nose-†thyrl— (nóstril)ω, know-
†leche—(knówledge)ω). By contrast, if the lefthand member
becomes obsolete boundary effects may persist (e.g. †Tues-day
—(Túes)ω(dày)ω).
9.3.1.2 Cohesion
Page 19 of 114
Compounds do not exhibit systematic phonologically

conditioned fusion. There are sporadic cases of historical
fusion of formerly separate prosodic constituents as in (30),
which I will refer to as ‘High Frequency Fusion’ (HFF).
(30) (vine)ω(yard)ω > (vinyard)ω (sheep)ω(herd)ω >

(shepherd)ω (cup)ω(board)ω > (cupboard)ω
(neck)ω(lace)ω > (necklace)ω
HFF differs from other types of fusion discussed below in that

it affects the prosodic (p.226) structure of individual words
and is insensitive to syntactic structure.17 HFF is instructive in
that it reveals the relatedness of ‘Final-C effects,
CONTAINMENT effects and HEADEDNESS effects (e.g.
v[án]yàrd > v[´n]y[ə]rd ‘vineyard’).
9.3.2 The prosody of prefixed words

Regarding their prosodic structure, historically prefixed words
in English fall into three categories illustrated by the phonetic
transcriptions of the verbs seduce, reduce, berate and rerate
in (31) (cf Raffelsiefen 1999a).18 Prosodic structures
unattested in sim-plexes are boldfaced.
(31)
Phonetic Prosodic
transcriptions representations:
(Wells 1990a):
I [s1djúːs], [sədjúːs] (sedúce)ω ‘seduce’
II [r1djúːs], [rədjúːs], ((re)Σ(dúce)Σ)ω ‘berate’

[riːdjús]
[b1ré1t], [bəré1t], ((be)Σ(ráte)Σ)ω ‘reduce’

[biːré1t]
III [rìːré1t]19 (re)ω(rate)ω ‘rerate’
Phonetically, the (historical) prefix in seduce is

indistinguishable from comparable strings in simplexes like
sedan, which is transcribed as [sidǽn], [sidǽn] in Wells
(1990a). Since they exhibit no morphophonological effects,
category I words are represented as a single pword in (31).
Page 20 of 114
The claim that prefixes can form separate pwords as in

category III is consistent with all diagnostics (cf. Booij and
Rubach 1984). The ungrammaticality of reducing the pretonic
syllable (cf. *r[ə] ráte) follows from the inviolability of
HEADEDNESS. Correlations with CONTAINMENT effects can
be observed for consonant-final prefixes (cf. the discussion of
subeditor in (25a)). Additional correlations with ‘Final-C
effects are seen in cases like [à℧t.r]ún ‘outrun’, [pòυst.g]
ráduate ‘postgraduate’. Prefixes which form separate pwords
differ prosodically from first members of compounds only with
respect to relative prominence:
(32)
Page 21 of 114
a. ωW ωS b. ωS ωW
([post]PREFIX) ([graduate]WORD) ([post]WORD) ([office]WORD)
Page 22 of 114
The prosodic motivation for category II in (31) concerns the

fact that for many speakers the prefix in reduce is prosodically
distinct from both seduce, which exhibits no effect, and rerate,
where the prefix forms a separate pword. Specifically,
category II words are typically represented with a third
pronunciation in Wells (1990a), which is (p.227) unattested
in simplexes but unlike the prefixes in category III is not
transcribed with a separate stress mark (cf. Wells 1990a,
Webster’s 1984). To account for the relevant prosodic
contrasts I tentatively propose the representation in (31),
where the variation characteristic for category II results from
the instability of a pretonic foot not forming a separate pword.
I will now present evidence that the three distinct prosodic

categories motivated on phonetic grounds in (31) reflect three
distinct syntactic properties of prefixes and are therefore
analyzable as boundary effects. Category I includes words for
which the (historical) prefix is simply not recognized. The
prefixes in category II determine the category of the derived
word and function accordingly as heads while those in
category III do not affect the category of the derived word and
therefore function as modifiers. Assuming that head affixes by
definition combine with stems whereas modifier affixes
combine with words the prosody of prefixation can be derived
from morphosyntactic structures in terms of the alignment
constraints in (28c) and (33b).20
(33)
a.
I [sedúce]WORD → sedúce)ω
II [[re]HEAD → ((re)Σ
PREF[dúce]STEM]WORD dúce)ω
[[be]HEAD → ((be)Σ
PREF[ráte]STEM]WORD ráte)ω
III [[re] MOD → (re)ω)

PREF[ráte]WORD]WORD (ráte)ω
b. ALIGN (Mod Prefix, L; ω, L), ALIGN (Mod

Prefix, R; ω, R) ALIGN (Head Prefix, L; Σ, L),
ALIGN (Head Prefix, R; Σ, R)
Page 23 of 114
According to the description in (33) ‘stem recurrence’ is

irrelevant to the prosody of prefixed words. In fact, se-
prefixation never exhibits boundary effects, regardless of stem
recurrence as is shown in (34a). By contrast, be-, re-, and de-
prefixation consistently exhibits boundary effects, even if the
stem does not recur as is shown in (34b).
(34)
a. s[I]dúce, ‘seduce’ reduce, induce,

s[ə]dúce, produce, deduce,
*s[iː]dúce adduce
s[I]léct, s[ə]léct, ‘select’ elect, prelect,

*s[iː]léct collect
s[1]clúde, ‘seclude’ exclude, include,

s[ə]clúde, preclude, conclude
*s[iː]clúde
b. b[1]gín, b[ə]gín, ‘begin’ —

b[iː]gín
r[1]lént, r[ə]lént, ‘relent’ —

r[iː]lént
d[1]síre, d[ə]síre, ‘desire’ —

d[iː]síre
Why is a prefix recognized in (34b), but not in (34a)? The

prefixes in (34b) relate to a category III-prefix (e.g. re-, de-,
pre-) or occur in combination with words. For instance, on the
basis of the type of words in (35a, b) learners infer that be- is a
head prefix. This (p.228) knowledge is then applied to all
verbs or adverbs where the prefix be- can be recognized with
the result that the remaining part of the word is categorized as
‘STEM’:
(35)
a. [[be]
[little]ADJ]VERB
[[be][witch] If [beX]VERB, then [[be]HEAD

NOUN]VERB PREF[X]STEM]WORD
Page 24 of 114
[[be]
[devil]VERB]VERB
b. [[be]
[low]ADJ]ADVERB
[[be] If [beX]ADVERB, then

[hind]ADJ]ADVERB
[[be] [[be] HEAD PREF[X] STEM]

[fore]ADJ]ADVERB WORD
The relevance of the syntactic category of the derived words in

(35) is indicated by the fact that no prefix is recognized in the
adjective benign or the noun regatta (cf the ungrammatical
pronunciations *b[iː]nígn, *r[iː]gátta). Similarly, the
pronunciation *s[iː]dúce is unacceptable because there are no
verbs which allow for a prefix se- to be recognized.21
Whereas the prosody of category II words depends on the

recognition of prefixes (rather than recurrent stems) the
prosody of category III words depends on the syntactic
relation between the prefixed word and its base. If they
commute the prefix is classified as an optional modifier and
mapped into a separate pword (the symbol ‘↔’ indicates
commutation in (36)):22
(36) If [[X]PREF [Y]]WORD ↔ [Y]WORD, then [[X]MOD

PREF[Y]WORD]WORDExample: Mary [rewrote ↔ wrote] the
letter.
If the syntactic condition in (36) is satisfied even non-recurring

prefixes like ab- regularly form a separate pword (cf. also
(ant)ω(arctic)ω antarctic, (arch)ω(angelic)ω arch-angelic):
(37) [abnormal ↔ normal] (behavior) therefore [[ab]MOD

PREF[normal]WORD]WORD → (ab)ω(normal)ω
Similarly, all prefixes in (38a), but none of the (historical)

prefixes in (38b), function as modifiers and consequently form
separate pwords:
(38)
Page 25 of 114
a. [[mis]MOD → (mìs)ω(allíance)ω b. (mísanthròpe)ω

PREF[allíance]WORD]W (mìs)ω(allíance)ω
ORD
[[im]MOD → (ìm)ω(móbile)ω (immédiate)ω

PREF[móbile]WORD]W
ORD
[[in]MOD → (ìn)ω(élegant)ω (insípid)ω

PREF[élegant]WOR
D]WORD
[[il]MOD → (ìl)ω(légal)ω (illúsive)ω

PREF[légal]WORD]W
ORD
Page 26 of 114
For phonetic evidence that even assimilating prefixes can form

separate pwords I refer the reader to Wells (1990a) (cf
Webster’s 1984), where prefixes functioning as modifiers are
distinct with respect to syllabification and stress from
(etymologically) related (p.229) prefixes which do not
function as modifiers (for detailed discussion, cf Raffelsiefen
1993,1999a). The phonetic facts accordingly argue against
descriptions where putative ‘level-1-properties’ such as
assimilation are used as evidence to motivate integration of
the prefix iN- into the pword of the stem (cf. Szpyra 1989).
The syntactic determination of prosodic form observed in

prefixation is part of the larger generalization that modifier
function words as in (39a) form separate pwords whereas head
function words illustrated in (39b) do not. Again, the syntactic
classification is based on optionality as in (39a) versus
obligatoriness as in (39b):
(39)
a.ðen, *ðən ‘then’ And (then) she left.
mΛt∫, *mət∫‘much She is (much) older now.
jet, *jət ‘yet’ It has (yet) to be decided.
tu:, *tə ‘too’ Mary (too) likes fish.
ɔ:l, *əl ‘all’ She was (all) skin and bones.
b. bΛt, √bət ‘but’ There’s no one here *(but) me.
ðæn, √ðən ‘than’ She’s older *(than) Beth.
kæn, √kən ‘can’ She *(can) do it.
ænd, √ənd ‘and’ Mary *(and) Beth are here.
tui, √tə ‘to’ Mary wants *(to) go.
æz, √əz ‘as’ Mary is happy *(as) a lark.
The seemingly mysterious absence of the weak form *nət,

perhaps first noted by Zwicky (1977), is characteristic for
modifiers. Given the mapping in (40a) this absence follows
from the necessary satisfaction of HEADEDNESS and is hence
a boundary effect. The generalization that function words do
not form separate pwords (cf. Selkirk 1995) should
accordingly be restricted to head function words for English:
Page 27 of 114
(40)
a. [not]MOD FW b. [but]HEAD FW →(not)ω →(but)Σ
Given that affixes are also function words the prosody of

prefixation follows the pattern illustrated in (40). Note that the
generalizations in question cannot be mimicked in terms of PU
constraints.
9.3.2.2 Cohesion
Category I prefixes cohere by definition whereas category III
prefixes exhibit only HFF illustrated in (41a) (compared to
regular prosody in (41b)):
(41)
a. (disintegrate)ω b. (dis)ω(organize)ω
(mistake)ω (mis)ω(print)ω
(extraordinary)ω (extra)ω(marital)ω
Fusion in (41a) is not a synchronic process but is a fact about

individual words, comparable to the fused compounds in (30).
Contracted n’t forms mentioned above also exhibit this sort of
fusion (e.g. (won’t)ω (can’t)ω versus may (not)ω).
(p.230) Unlike modifier prefixes, head prefixes exhibit

regular phonologically conditioned cohesion. In (42a) a
monosyllabic prefix is followed by a stem with an initial
stressless syllable. A Ci-final prefix is followed by a vowel-
initial stem in (42b) and by a Cj-initial stem in (42c), where Ci
and Cj are consonants which may differ only in voicing.
(42)
a. [[re]HEAD PREF → (réconcìle)ω

[concíle]STEM]WORD
[[im]HEAD PREF → (ímbecìle)ω

[becíle]STEM]WORD
[[de]HEAD PREF → (dérogàte) ω

[rogáte]STEM]WORD
Page 28 of 114
b. [[dis]HEAD PREF [áster] → (disáster)ω

STEM]WORD
[[en]HEAD PREF → (enámor)ω

[ámor]STEM]WORD
[[sub] HEAD PREF [órdinate] → (subórdinate)ω

STEM]WORD
c. [[sub]HEAD PREF → (su[p]oéna)ω

[poéna]STEM]WORD ‘subpoena’
[[en] HEAD PREF [nóble] → (e[n]óble)ω

STEM]WORD ‘ennoble’
[[im] HEAD PREF [mérse]STEM] → (i[m]érse)ω

WORD ‘immerse’
Cohesion in (42a) can be described in terms of domination of

alignment constraints by phonological markedness
constraints. The observation that a head prefix followed by a
stem with an initial stressless syllable never exhibits boundary
effects is described by ranking FTBIN above ALIGN (Head
Prefix, E; Σ, E) (cf. the similar effects in Yidiɲ). This is because
the initial stressless stem syllable is integrated into the
syllable headed by the prefix to satisfy FTBIN. This specific
effect is ruled out for stems with initial stress with the result
that the regular boundary effects in the verbs in (43a) are
absent in the respective cognate nouns in (43b), where the
prefixes regularly cohere. The fused words in (43b) are
prosodically indistinguishable from simplexes (cf. restrictions
on vowel quantity as in d [è]rivátion, [é]lefant, not *d
[ì:]rivátion, *[í:]lefant).23
(43)
Page 29 of 114
a. [[de] HEAD → ((d[i: ])Σríve)ω ‘derive’

PREF[ríve]STEMWORD
[[re] HEAD PREF [fúte] → ((r[i: ])Σfúte)ω ‘refute’

STEM] WORD
b. [[de] HEAD PREF → (d[è]rivátion)ω ‘derivation’

24
[rivátion]STEM]WORD
[[re] HEAD PREF → (r[è]futátion)ω ‘refutation’

[futáion] STEM]WORD
Page 30 of 114
The relevant constraint causing cohesion in (42b) is the

constraint ONSET, which requires every syllable to have an
onset. In (42c) it is the constraint *CiCi, which prohibits
adjacent consonants with (near) identical features. The
ranking is given in (44):
(44) FTBIN, ONSET, * CiCi » ALIGN (Head Prefix, E; Σ,

E)
The morphosyntactic structures in (42) are not necessarily

meant to represent speaker knowledge. Indeed, the absence of
boundary signals in the fused pwords in (42) is likely to
prompt (re)analysis of these words as simplexes. The point is
that for head prefixes the sort of phonological conditions
illustrated in (42a, b, c) necessarily leads (p.231) to prosodic
fusion. There is thus a real, syntactically conditioned
difference between the words in the ‘A-rows’ in (45), which
contain modifier prefixes and yield systematic boundary
effects, and the corresponding words in the ‘B-rows’, which
contain head prefixes and do not exhibit boundary effect.25
(45)
Page 31 of 114
A [[re]MOD PREF → (r[ì: ])ω(combíne)ω ‘recombine’

[combíne]WORD]WORD
B [[re]HEAD PREF → (r[é]concìle)ω ‘reconcile’

[concíle]STEM]WORD
A [[un]MOD PREF [áble] → (ùn)ω(áble)ω ‘unable’

WORD]WORD
B [[en]HEAD PREF [áble] → (enáble)ω ‘enable’

STEM]WORD
A [[im]MOD PREF [mórtal] → (ì[m])ω([m]órtal) ‘immortal’

WORD]WORD
B [[im]HEAD PREF → (i[m]íngle)ω ‘immingle’

[míngle]STEM]WORD
Page 32 of 114
The absence of phonologically conditioned cohesion for

modifier prefixes follows from the ranking of the constraint
ALIGN (Word,L; ω,L) above the phonological marked-ness
constraints in (44), which also accounts for the absence of the
relevant cohesion effects in compounds.
9.3.3 Suffixation
Unlike prefixes, suffixes never function as modifiers but
always as heads.26 Analogous to the representation of head
prefixes, suffixed words are consequently uniquely
represented as [[X]STEM[Y]HEAD SUFFIX] WORD. The
morphosyntactic and corresponding pro-sodic structures are
illustrated in (46).
(46)
[[hope] STEM [less] HEAD SUFFIX] WORD → (hoʊp)ωləs
[[whole] STEM [some] HEAD SUFFIX] → (hoʊl)ωsəm

WORD
[[cool]STEM[ness]HEAD SUFFIX]WORD → (ku: l)ωnəs
[[taste] STEM [ful] HEAD SUFFIX] WORD → (teist)ωfəl
[[duke] STEM [dom] HEAD suffIx] WORD → (du:

k)ωdəm
[[ail] STEM [ment] HEAD SUFFIX] WORD → (e1l)ωmənt
(p.232) The right pword boundary in (46) could be described

by positing the constraint ALIGN (Stem, R; ω, R) or,
alternatively, the constraint ALIGN, (Head Suffix, L; ω, R). The
mapping in (46) is supported by all relevant diagnostics.
Apparent violations of rhyme structure are explained if the
stem-final consonant is pword-final and consequently
associated with an onset in the lexicon. Correlating ‘Final-C’
and CONTAINMENT effects are seen in (47a), where the stem-
final consonant has exclusive coda properties (i.e.
glottalization), thereby contrasting in a subtle manner with
corresponding word-internal stops as in (47b). Related effects
were observed in shyness versus minus.
(47)
Page 33 of 114
a. hó[p̚l]ess ‘hopeless’ b. dú[pl]icate ‘duplicate’
lú[k̚l]ess ‘luckless’ dè[kl]arátion ‘declaration’
The analysis of the special phonology reviewed here as

boundary effects raises the question of how learners arrive at
the underlying morphosyntactic structures in (46). Analogous
to the procedure specified for head prefixation I assume that
given the occurrence of some suffixations based on free stems
as in (48a) learners abstract the affix-based syntactic parsing
rule specified in (48b).
(48)
Page 34 of 114
a [[goal]NOUN[less]]A b.
DJ
[[time]NOUN[less]]A f [[X][less]]ADJ,
DJ
[[tooth]NOUN[less]]A then
DJ [[X]STEM[less]HEAD
SUFFIX]WORD
Page 35 of 114
The result of applying the parsing rule in (48b) is illustrated in

(49). Significantly, syntactic properties of the stem in isolation
(i.e. free versus bound) are not represented.27
(49) [[fruit] stem[less] head suffix]WORD

[[gorm] stem [less] head suffix] word [[ruth] stem [ls] head
suffix] word
To rule out the application of the rule in (48b) to adjectives

such as metícu[ləs] ‘meticulous’ frivo[ləs] ‘frivolous’ additional
criteria have to be invoked, including the presence of a
privative meaning and, for literate learners, perhaps also a
certain spelling. Importantly, the data from head suffixation
confirm the generalization established earlier: if there exist
words on the basis of which a head affix can be learned, the
words containing that affix exhibit stable boundary effects,
whether or not the stem recurs.28 Here a significant contrast
with PU effects emerges, to be discussed in s. 9.4.3.
(p.233) 9.3.3.2 Cohesion

Pwords included in suffixed words are unexceptional in that
they are sporadically affected by historical HFF, with
concomitant loss of boundary effects. Examples are given in
(50).
(50)
(wise) ωdom > (w1z.dəm) ω ‘wisdom’
(worth)ωship > (wεr.∫əp)ω ‘worship’
(hand)ωsom > (hæn.səm)ω ‘handsome’
(friend)ωship > (fren.∫əp)ω ‘friendship’
As for systematic phonologically conditioned cohesion,

suffixation exhibits interesting parallels to other head
affixation discussed in s. 9.3.2.2. Clear evidence for the
domination of ALIGN (Stem, R; ω, R) by ONSET is seen in (51),
where aspiration of the stem-final consonant indicates the
integration of the vowel-initial suffix.
(51)
Page 36 of 114
[[sánit]STEM[íze]HEAD → (sáni[th]ìze)ω
SUFFIX]WORD
[[púppet] STEM[éer] HEAD → (pùppe[th]éer)ω

SUFFIX]WORD
[[límit]STEM[átion]HEAD → (lìmi[th]átion)ω
SUFFIX]WORD
[[góat] →(gòa[th]ée)ω
STEM[ée]HEAD
SUFFIX]
WORD
Recall that pword-final consonants are never aspirated, even if

a stressed vowel follows. The overall prosodic contrast
resulting from distinctions in morphosyntactic structures is
illustrated in (52), where t is associated with coda position in
(52a), but not in (52b).29
(52)
Page 37 of 114
a. [[párrot]WORD[éyes]WO → (párro[ɾ]) (párro[?])(èyes) ω

RD]WORD (èyes)ω‘parrot eyes’ ‘parrot eyes’
b [[párrot]STEM[íze] HEAD → (párro[th]ìze)ω ‘parrotize’

SUFFIX]WORD ‘parrotize’
Page 38 of 114
Recall that in head prefixation, cohesion also occurs to satisfy

the constraint *CiCi. A similar effect is seen in suffixation.
(53)
Page 39 of 114
[[X]STEM[ty]HEAD a. (frail)ωty ‘frailty’ b . (modes[t]y)ω

SUFFIX]WORD
[[X] STEM [ly]HEAD (night)ωly ‘nightly’ly’ (woo[l]y)ω ‘wooly

SUFFIX]WORD
[[X]STEM[let]HEAD (pl[e+l1])ω[l+f ]et (ow[l]et)ω ‘owlet’

SUFFIX]WORD
‘playlet’
Page 40 of 114
It should be noted that ‘degemination' in (53b) is a systematic

property of the suffixed words since the pronunciations
modes[t.t]y, ow[l.l] et, woo[l.l]y are clearly ungrammat- (p.
234) ical. Yet, the examples in (54), which include geminates,
show that cohesion in (53b) cannot be described by simply
ranking *CiCi above the relevant alignment constraint.
(54)
[[goal]STEM [less ] HEAD (goa[l])ω[l]ess

SUFFIX]WORD ‘goalless’
[[keen]STEM[ness ]HEAD (kee[n])ω[n]ess

SUFFIX]WORD ‘keenness’
[[frail] STEM [ly ] HEAD (frai[l])ω[l]y ‘frailly’

SUFFIX]WORD
[[embalm] STEM [ment] HEAD (embal[m])ω[m]ent

SUFFIX] WORD ‘embalmment’
One of the main claims defended here is that cohesion, unlike

PU behavior, is not an idiosyncratic property of individual
affixes. There is indeed an independent property which
distinguishes the suffixes in (53) and (54). For each suffix in
(53) a corresponding suffix exists, which has identical
morphosyntactic properties and differs only in the absence of
the initial consonant.
(55)
a.-[ti]N ˜ -[i]N b. [[frail]Aty]N - [[jealous]Ay]N
-[li]A ˜ -[i]A [[earth]Nly]A - [[earth]Ny]A
-[lət]n ˜ -[ət]n [[pig]Nlet]N - [[midge]Net]N
The generalization is then that suffix pairs CiX, X which have

identical morphosyntactic properties neutralize after stem-
final Ci: only X appears.30 As a result, the constraint *CiCi is
always satisfied for the suffixes in (53). If no relevant suffix
exists, fusion does not occur resulting in surface geminates as
in (54).31
A third phonological condition for cohesion is illustrated in

(56). Again, the suffixes form a single domain of syllabification
together with the stem.
Page 41 of 114
(56)
[[fly] STEM[t]HEAD SUFFIX]WORD → (fly+t)ω ‘flight’
[[ten] STEM[th]HEAD SUFFIX]WORD → (ten+th)ω

‘tenth’
[[girl] STEM[s]HEAD SUFFIX]WORD → (girl+s)ω ‘girls’
Assuming that a syllable must have a nucleus (arguably a

HEADEDNESS effect), that obstruents cannot function as a
syllable nucleus in English, and that all segments must be
parsed into syllables, the integration of the suffixes in (56) into
the pword of the stem can be described in terms of the
domination of alignment constraints by phonological
markedness constraints.
9.3.4 Word plus clitic

The evidence from boundary effects consistently shows that
function words are not regularly integrated into the pword of
the stem.
(p.235)
(57) [áim]WORD [to]HEAD FW → (áim)ωto ‘aim to’

[léave]WORD [me]HEAD FW → (léave)ωme ‘leave me’
[pút]WORD [it]HEAD FW → (pú[ɾ])ωit ‘put it’
The last example is in fact inconclusive because all consonants

weaken before a stressless vowel, whether or not a pword
boundary intervenes. Evidence for the non-integration of
function words, as opposed to suffixes, into the pword of the
stem comes from the behavior of the final consonant under
stress. When a suffix as in (58a) is stressed because of further
suffixation, aspiration applies, which indicates prosodic
integration. By contrast, the flap remains before vowel-initial
function words, regardless of (contrastive) stress (indicated by
capitals in (b)), which indicates non-integration:
(58)
a. [[fát]STEM[al]HEAD → (fá[ɾ]al)ω ‘fatal’cf.

SUFFIX]WORD (fà[t h]álity)ω
Page 42 of 114
b. [pút]WORD [it]HEAD FW → (pú[ɾ])ω it ‘put

it’cf. (pú[ɾ])ωÍT
The patterns in (58) indicate that ALIGN (Word,R; ω,R), as

opposed to ALIGN (Stem, R; ω, R), is not dominated by
ONSET. Assuming that the morphosyntactic structures in (58)
are correct this observation entails that words with
nonintegrated suffixes like shyness must be represented
prosodically as ((shy)ness)ω. This structure violates the
constraint NONRECURSIVITY, which prohibits the domination
of pwords by pwords (cf. Selkirk 1995: 443). This conclusion
can perhaps be avoided by some alternative analysis but will
be accepted for now.
The observation that function words are not integrated into

the pword of the preceding word to satisfy ONSET is part of
the larger generalization that there is no phono-logically
conditioned cohesion across syntactic word (as opposed to
stem) boundaries. (Historical) HFF as in (59) is fairly common.
(59)
(give)ω me > cf. (*simme) ‘[sieve me] some

(gimme) ω flour’
(want)ω to > cf. (*planna) ‘[plan to] go’

(wanna) ω
(let)ωus > (let’s) ω cf. (*get’s) ‘[get us] some tea’
9.3.5 Summary
In this section I have presented evidence that the pword
structure of English words can be described on the basis of
independently motivated morphosyntactic structure with no
need for diacritic marking of affixes. Some representative
examples are repeated in (60):
(60)
a. [not]MOD FW → (not)ω ‘not’
b. [[un]MOD PREFIX [áble] → (ùn)ω(áble)ω

WORD]WORD ‘unable’
Page 43 of 114
[[im]MOD PREFIX [mórtal] → (ìm)ω(mórtal)ω

WORD]WORD ‘immortal’
[but]HEAD FW → (but)Σ ‘but’
[[re]HEAD PREFIX → ((re)E néw)w

[new]STEM]WORD ‘renew’
c. [[en]HEAD PREFIX [áble] → (enáble)ω

STEM]WORD ‘enable’
[[góal]STEM [less]HEAD → ((góal)ωless)ω

SUFFIX]WORD ‘goalless’
[[góal]STEM[íe]HEAD → (góalie)ω‘goalie’
SUFFIX]WORD
[pút]WORD[it]HEADFW → (pú[]ωit ‘put it’
(p.236) Systematic cases of affix cohesion, boldfaced above,

are described in terms of the domination of alignment
constraints by phonological markedness constraints. All pro-
sodic units satisfy constraints on prosodic domination
including HEADEDNESS and CONTAINMENT.
The distinction between cohering and noncohering affixation

illustrated in (60) plays a crucial role in English
morphophonology. Specifically, in native word formation only
cohesive affixation exhibits stem-allomorphy (cf (61a)), stem
selection (cf (61b)), and phonologically conditioned gaps (cf.
(61c)).32
(61)
a. Vietnám+ése → ballóon+líke →
(Vìetnamése)ω (ballóon)ω(lìke)ω
b. émphas-is+íze → cúri-ous+ness →
(émphasìze)ω ((cúrious)ωness)ω
ámput-àte+ée → shóv-el+ful →
(àmputée)ω ((shovelful)ω
c. revéal+al → Ø goal+less →
((goal)ωless)ω
invólve+al → Ø leaf+let → ((leaf )ωlet)ω
cóllapse+al → Ø law+ful → ((law)ωful)ω
Page 44 of 114
Stem-allomorphy results when phonological markedness

constraints are satisfied at the expense of violating PU
constraints. ‘Stress shift’ in (61a) satisfies *CLASH but
violates PU. No such allomorphy exists in noncohesive
affixation.33
Similarly, (boldfaced) stem selection in (61b) satisfies

phonological markedness constraints (i.e. OCP-constraints) at
the expense of violating the constraint SUBCATWORD, which
requires affixes to combine with free stems (cf. Raffelsiefen
(2004a, b)). SUBCATWORD is never violated in noncohesive
word formation.34
Finally, phonologically conditioned gaps as in (61c) result

when an affix is required to satisfy both phonological
markedness constraints and PU constraints. The gap in (61c)
results from the fact that noun-forming -al-suffixation satisfies
the constraint LiLi, which prohibits the co-ocurrence of
identical liquids within the same pword. No such effect occurs
in noncohesive affixation. The morphophonological differences
between cohesive and noncohesive affixation illustrated in (61)
result directly from the rules for aligning morphosyntactic
with prosodic boundaries. This is because the prosodic
boundaries delimit the domains within which phonological
markedness constraints are evaluated.
In addition to explaining restrictions on morphophonological

phenomena as in (61) the contrast between cohesive versus
noncohesive affixation is also crucial to explaining restrictions
on affix ordering. The generalization is that noncohesive der-
(p.237) ivational suffixes cannot be followed by cohesive
ones but the reverse order is possible (e.g.
*((kind)ωness)ω+ish versus √((monk+ishω+ness)ω). Similarly,
noncohesive prefixes cannot be preceded by cohesive ones but
the reverse order is possible.
Independent motivation for the domain analysis illustrated in

(60) is important in that a significant share of ‘identity effects’,
including the cases in (62), results from noncohesion and
should therefore not be cited as evidence for PU:
(62) vowel length in both (sh[á+11])ω ‘shy’ and

((sh[á+11])ωness)ω ‘shyness’ aspiration in both ([k
h
]ombine)ω ‘combine’ and (re)ω([khombine)ω ‘recom-
Page 45 of 114
bine’ cluster simplification in both (to[m(b)])ω ‘tomb’

and (to[m(b)])ωless)ω ‘tomb-less’ cluster simplification
in both ([(k)n]it)ω ‘knit’ and ((be)Σ [(k)n]it)ω ‘beknit’
Apart from the fact that it is superfluous to invoke PU to

account for the phenomena in (62) there are arguments
against doing so. Those are reviewed in the next section.
9.4 Empirical Differences between Boundary Effects and

Pu Effects
9.4.1 Real versus apparent violation of canonical phonology
The distinct conditions pertaining to boundary effects versus
PU effects to be reviewed below result from the property
noted earlier that only PU effects involve genuine violations of
regular phonology. A few examples will illustrate this point
further. English shares with many other languages a window
restriction on stress, which must not fall to the left of the
antepenultimate syllable (e.g. hippopótamus, not *
hippópotamus, mahógany not *máhogany). Assuming the
mapping of the morphosyntactic to prosod-ic structures in
(63), the relevant constraint is violated in (63a) but not in
(63b). This is because the relevant point from which to
compute the window restriction on stress is the (innermost)
right pword boundary.
(63)
a. [[vínegar] STEM [ish] HEAD → (vínegarish)ω

SUFFIX]WORD
[[cándidat] STEM [ure] HEAD → (cándidature)ω

SUFFIX] WORD
[président] STEM [y] HEAD → (présidency)ω

SUFFIX] WORD
[dífficult] STEM [y] HEAD SUFFIX] → (dífficulty)ω

WORD
b. [[rádical] STEM [ness] HEAD → ((rádical)ωness)ω

SUFFIX]WORD
[[cítizen] STEM [ship] HEAD → ((cítizen)ωship)ω

SUFFIX] WORD
[[mónitor] WORD [it] FW → (mónitor)ω it
Page 46 of 114
The examples in (63a) versus (63b) illustrate two distinct

sources of ‘stress neutrality’. In (63a), regular restrictions on
stress placement are violated to satisfy PU(STRESS), a
constraint which requires the stress in the derived word to
correspond to the stress (p.238) in the base. In (63 b), stress
placement in the complex structures is determined by the
relevant domain boundaries. The correspondence to the stress
in the respective base could be explained in terms of identical
domain structures without invoking PU.
The second example illustrating the claim that only PU effects,

but not boundary effects, violate regular phonology concerns
the distribution of the velar nasal and (historical) [g]-loss in
English. After [ŋ], [g] is regularly deleted as in (64a) unless it
precedes a segment more sonorous than [ŋ] (i.e. vowel, liquid,
glide) as in (64b) (cf. Kahn 1976: 130 ff.).
(64)
a. ri[Ŋg] > ri[Ŋ] ‘ring’ b. Mo[Ŋ.góŊ]lia

‘Mongolia’
tu[Ŋgs]ten > tu[Ŋ.s]ten fi[Ŋ.g3]r (*fi[Ŋ]er)

‘tungsten’
a[Ŋgz]iety > a[Ŋ.z]iety ga[Ŋ.gl]ing

‘anxiety’ (*ga[Ŋl]ing)
I[ngv]aeonic > hu[Ŋ.gr]y

I[n.v]aeonic ‘Ingvaeonic’ (*hu[Ŋr]y)
A[Ŋgm]ering > li[Ŋ.gw]al

A[Ŋ.m]ering ‘Angmering’ (*li[Ŋw]al)
Assuming that there are constraints which force onset

syllabification of consonants followed by segments with higher
sonority the stability of [g] in (64b) indicates the constraint
ranking in (65), where ONSET: *ŋ prohibits the occurrence of
[ŋ] in onset position.35
(65) ONSET: * ŋ ≫ *ŋg
Given the prosodic structures in (66) ONSET: *[ŋ] is violated

in (66a), but not in (66b). This is because phonetically, all
pword-final consonants are necessarily associated with coda
position to satisfy CONTAINMENT.
Page 47 of 114
(66)
Page 48 of 114
a. [[síng]STEM[er] HEAD → (sí[ŋ]er)ω ‘singer’

SUFFIX] WORD
[[síng]STEM[ing] HEAD → (sí[ŋ]ing)ω ‘singing’

SUFFIX] WORD
[[síng] STEM [able] HEAD → (sí[ŋ]able)ω ‘singable’

SUFFIX] WORD
[[lóng] STEM[ish] HEAD → (ló[n]ish) ‘longish

SUFFIX] WORD
b. [[ríng]WORD → (rí[ŋ])ω(lèader)ω ‘ring leader’

[léader]WORD]WORD
[[ríng]WORD [líke]HEAD → (rí[ŋ])ω(lìke)ω ‘ringlike’

SUFFIX]WORD
[[ríng] STEM[less] HEAD → ((rí[ŋ])ωless)ω ‘ringless’

SUFFIX]WORD
Page 49 of 114
9.4.2 Generality of boundary effects versus affix-specificity of PU

effects
It has been demonstrated in s. 9.3 that pwords identified on
the basis of syllable-related diagnostics can be described
without referring to individual affixes. Assuming (p.239) the
inviolability of CONTAINMENT it follows that non-integrated
affixes are necessarily stress-neutral. This is because feet are
necessarily contained within pwords and cannot incorporate
pword-external syllables after affixation. Stress-neutrality is
indeed a property of all consonant-initial suffixes in English.36
(67)
[[rádical]STEM[ness]HEAD → ((rádical)ωness)ω
SUFFIX]WORD
[[devélop] STEM[ment] HEAD → ((devélop)ωment)ω

SUFFIX]WORD
[[méaning] STEM[less]HEAD → ((méaning)ωless)ω

SUFFIX]WORD
[[púrpose] STEM [ful]HEAD → ((púrpose)ful)ω

SUFFIX] WORD
[[offícial] STEM[dom]HEAD → ((offícial)ωdom)ω

SUFFIX]WORD
[[frólic]STEM[some]HEAD → ((frólic)ωsome)ω
SUFFIX]WORD
[[párrot] STEM[let]HEAD → ((párrot)ωlet)ω

SUFFIX]WORD
[[ínstant] STEM [ly]HEAD → ((ínstant)ωly)ω

SUFFIX]WORD
[[síngle]STEM [ton] HEAD SUFFIX] → ((síngle)ωton)ω

WORD
[[súbtle] STEM [ty]HEAD SUFFIX] → ((súbtle)ty)ω

WORD
[[únder] STEM [ling]HEAD → ((únderωling)ω

SUFFIX] WORD
[[héaven] STEM [wd]HEAD → ((héaven)ωward)ω

SUFFIX]WORD
Page 50 of 114
[[nèighr]STEM [hood] HEAD → ((néighbor)ωhood37)ω

SUFFIX]WORD
By contrast, integrated affixes may or may not be stress-

neutral (cf. Raffelsiefen 1996, 1999b). Compare the vowel-
initial nominal suffixes -ity and -acy, both of which are
integrated into the pword of stem to satisfy ONSET. ‘Stress
shift’ in (68a) to satisfy *LAPSE defined in (5), versus ‘stress-
neutrality’ in (68b, c) to satisfy PU(STRESS) with respect to
the corresponding words in the righthand column, are
idiosyncratic properties of the respective suffixes.38
(68)
Page 51 of 114
a. [[rádical] STEM[ity]HEAD → (ràdicálity)ω

SUFFIX]WORD
[úrban]STEM[ity]HEAD → (urbánity)ω
SUFFIX]WORD
[[be] STEM[ity]HEAD → (bèllicósity)ω

SUFFIX]WORD
B [[áccur]STEM [acy]HEAD → (áccuracy)ω áccurate

SUFFIX]WORD
[[ádequ] STEM[acy]head → (ádequacy)ω ádequate

suffIx]word
[[móder] → (móderacy)ω móderate

STEM[acy]HEAD
SUFFIX]WORD
[dop] STEM[able] HEAD → (devélopable)ω devélop

SUFFIX]WORD
[[límit] STEM [able]HEAD → (límitable)ω límit

SUFFIX]WORD
[[ségreg]STEM[able]HEA → (ségregable)ω ségregàte

D SUFFIX]WORD
Page 52 of 114
Similarly, the cohesive adjectival suffixes – ic and -ish differ in

that FTBIN dominates PU(STRESS) for – ic (cf. 69 a) whereas
the opposite ranking holds for -ish (cf. 69b).
(p.240)
(69)
Page 53 of 114
a. [[vámpìre]STEM[ic]HEAD → (vampíric)ω
SUFFIX]WORD
[[móròn] STEM[ic]HEAD → (morónic)ω

SUFFIX]WORD
[[próphet] STEM[i c] → (prophétic)ω

HEAD SUFFIX]WORD
b [[vámpìre]STEM[ish]HEA → (vámpìrish)ω vámpìre

D SUFFIX]WARD
[[vínegar] STEM [ish] → (vínegarish)ω vínegar

HEAD SUFFIX]WORD
[[gíant]STEM[ ish]HEAD → (gíantish)ω gíant

SUFFIX]WORD
Page 54 of 114
‘Rightward stress shift’ in cohesive suffixation is possible

because the formation of feet that span both stem and suffix
syllables does not violate CONTAINMENT (e.g.
((ràdi)Σ(cálity)Σ)ω). This possibility is excluded for noncohesive
affixes. The relevant generalization is missed by Benua
(1997a), where all affixes are diacritically assigned to some
class to account for stress behavior.
TCT [Transderivational Correspondence Theory RR] accounts

for affix class behavior without representational or serial
assumptions. Both kinds of affixes can be prosodically
integrated into the stem, and all words can be derived in
parallel, without intermediate stages. The differences between
class 1 and class 2 words follow from the rank of identity
constraints on two distinct OO [output–output RR]-
correspondence relations. Benua (1997a:169)
The description of morphophonology based on affix-class

advocated by Benua not only misses the generalization that
‘neutrality’ can in many cases be predicted on the basis of the
phonological form of the suffix but also falsely implies that all
types of ‘non-neutral’ morphophonology can be subsumed
under some unique constraint ranking.39
Consider next the evidence from the velar nasal, which on the
domain analysis cannot be followed by g before non-cohering
suffixes. The relevant words include suffixes with initial liquids
or glides (cf the data in (64b)), which in fact, are never
preceded by [Ŋg]:
(70)
[[fáng] STEM [less] HEAD → ((fá[Ŋ)ωless)ω‘fangless’

SUFFIX] WORD
[[kíng]STEM [ly] HEAD → ((ki[Ŋ])ωly)ω ‘kingly’

SUFFIX] WORD
[[yóung]STEM [ling] HEAD → ((you[Ŋ])ωling)ω

SUFFIX]WORD ‘youngling’
[[ríng]STEM[let]HEAD → ((ri[Ŋ])ωlet)ω
SUFFIX]WORD ‘ringlet’
By contrast, the extension of coda [Ŋ] from a base into words

with cohesive suffixes to satisfy PU (while violating *ONSET:
*ŋ) is somewhat idiosyncratic. For agentive -er, PU dominates
Page 55 of 114
*ONSET: *n whereas for comparative -er, PU effects are

unsystematic. Other cohesive suffixes exhibit similarly
unpredictable behavior, where variation among speakers is
common (cf. Wells 1990a ; OED).40
(p.241)
(71)
a. No PU (stró[Ŋg]er)ω ‘stronger’
effect
(diphthó[Ŋg]al)ω ‘díphthongal’
(díphtho[Ŋg]ìze)ω, ‘díphthongìze’
b. PU effects (sí[Ŋ]er)ω (cf. sí[Ŋ]) ‘singer’
(yóu[Ŋ]ish)ω (cf. yóu[Ŋ]) ‘youngish’
(Quísli[Ŋ]ism)ω (cf. Quísli[Ŋ])

‘Quislingism’
(hà[Ŋ]ée)ω (cf. ha[Ŋ]) ‘hangée’
(Pèki[Ŋ]ése)ω (cf. Péki[Ŋ])

‘Pekingese’
(díphtho[Ŋ]ìze)ω (cf. díphtho[Ŋ])

‘díphthongíze’
Page 56 of 114
9.4.3 The relevance of the base

Assuming the criteria for recognizing morphosyntactic
structure specified in s. 9.3 it follows that for head affixation,
the syntactic status of the stem (i.e. free or bound) is
irrelevant for prosodic structure. Recall that head prefixed
words as in (72) are indeed prosodically indistinguishable,
whether or not a base exists.
(72)
[[be]head prefix ((be)Σ gét)ω

[gét]STEM]WORD ‘beget’
[[be]head prefix ((be)Σ gín)ω

[gín]STEM]WORD ‘begin’
In general it is expected that for head affixation boundary

effects persist, whether or not the stem continues to
correspond to an independently occurring word. This is
because that sort of information is not reflected in the
syntactic representation on which the prosodic structures are
based.41 For instance, the syntactic representation of the -ful-
suffixations in (73) does not encode the fact that dole became
obsolete whereas soul continues to exist as an independent
word. As a result of their identical morpho-syntactic structures
the words in (73) also have the same prosodic properties. In
fact, the ‘Final-C effect’ in doleful has persisted after dole
became obsolete.
(73)
[[sóul] STEM [ful] HEAD SUFFIX] ((s[oʊl])ωful)ω

WORD ‘soulful’
[[dóle] STEM [ful] HEAD SUFFIX] ((d[oʊl])ωful)ω

WORD ‘doleful’
Recall that word-internal ‘Final-C’ effects are systematically

eliminated as a result of ‘HFF’, and do not occur in
simplexes.42 The systematic persistence of this boundary
effect in noncohesive suffixation based on ‘bound stems’
therefore strongly supports the description of syntax to
prosody mapping spelled out in s. 9.3. In general, suffixed
Page 57 of 114
words based on bound stems as in (74a) and those based on

free stems as in (74b) are prosodically indistinguishable.
(74)
Page 58 of 114
a. [oʊn.l]y ‘only’ (on)ωly b. l[oʊn.l]y (lone)ωly
g[ɔrm.l]ess (gorm)ωless h[arm.l]ess (harm)ωless

‘gormless’
[ɔ1nt.m]ent (oint)ωment app[ɔ1nt.m]ent (appoint)ωment

‘ointment’
gr[e1t.f ]ul (grate)ωful h[e1t.f]ul (hate)ωful

‘grateful’
w[1st.f ]ul ‘wistful’ (wist)ωful f[1st.f]ul (fist)ωful
r[u:θ.l]ess (ruthωless t[u:θ.l]ess (tooth)ωless

‘ruthless’
v[est.m]ent (vest)ωment inv[est.m]ent (inves)ωtment

‘vestment’
g[æst.l]y ‘ghastly’ (ghast)ωly v[æst.l]y (vast)ωly
l[1st.l]ess ‘listless’ (list)ωless r[est.l]less (rest)ωless
Page 59 of 114
(p.242) Similarly, CONTAINMENT effects persist regardless

of the occurrence of the stem as an independent word as is
demonstrated in (75). Clearly, reckless, which is based on a
bound stem, patterns prosodically with checkless, rather than
with comparable sim-plexes.
(75)
Page 60 of 114
a. re[k ̚less ‘reckless’ (reck)ωless b. che[k ̚ l]ess (check)ωless
fe[k ̚.l]ess (feck)ωless spe[k ̚ l]ess (speck)ωless

‘reckless’
ha[p ̚.l]ess (hap)ωless ma[p ̚ l]ess (map)ωless’

‘hapless’
Page 61 of 114
In general, noncohesive affixation based on bound stems is

rare in English because noncohesive affixes are typically
native, which means that the occurrence of bound stems
presupposes simultaneous loss of some word Wi and survival
of a word historically derived from Wi.43 Yet, the conclusion is
clear: the question of whether the stem is ‘free’ or ‘bound’ is
irrelevant for the occurrence of boundary effects in
noncohesive suf-fixation. This is not to deny the crucial
relevance of free stems for learning a head affix in the first
place which was noted above. Recall that the prefix se- plays
no role in English morpho(phono)logy because it is exclusively
based on bound stems and therefore it is not learnable.
Similarly, the historical diminutive head suffix -kin in English is
no longer learnable, which accounts for the parsing of
pumpkin as a single pword (76).
(76) [púmpkin]WORD → (púmpkin)ω
As a result of the parsing in (76) FINAL-C effects are prone to

disappear. Wells (1990a) lists the pronunciations pu[mpk]in,
pu[nk]in, where the former is probably a spelling
pronunciation.44 No corresponding reduced variants are
attested for any of the words exhibiting internal FINAL-C
effects in (74). Crucially, these words differ from pumpkin in
that they include an affix that also occurs with free stems and
therefore is learnable.
(p.243) Turning now to PU effects in cohesive affixation, the

occurrence of the relevant sound property in a free-standing
word is crucial. Recall the violations of the ARAB rule in -ic-
suffixation discussed in s. 9.1 which are necessarily licensed
by an independent word. Additional examples are given in
(77).
(77) (ph[óʊ]bic)ω—ph[óʊ]bia ‘phobia’ (sc[í: ]nic)ω—

sc[i: ]ne ‘scene’ (phon[í: ]mic)ω—phón[ì: ]me
‘phoneme’ (chr[óʊ]mic)ω—chr[óʊ]me
‘chrome’ (am[í: ]bic)ω—am[í: ]ba ‘amoeba’
The relevance of a freestanding word for licensing PU effects

can also be illustrated with stress. The (historically
conditioned) stress on the preantepenultimate syllable in -able
-suffixation tends to regularize as in (78a) unless it is
Page 62 of 114
‘licensed’ by corresponding stress in an independent word as

in (78b):
(78)
Page 63 of 114
a. (hóspitable)ω > b. (cháritable)ω chárity

(hospítable)ω
(déspicable)ω > (inhéritable)ω inhérit

(despícable)ω
(fórmidable)ω > (médicable )ω médicàte

(formídable)ω
(ápplicable)ω > (prácticable)ω práctice, práctical

applícable)ω
Page 64 of 114
The sensitivity to independent base words observed in (78)

contrasts with nonco-hesive suffixation as in envíronment,
where the ‘unexpected’ stress (the skipping of the closed
penult) is stable even for speakers who are unfamiliar with
envíron as a freestanding word.
Additional support for the generalization established here

comes from the phonology of the velar nasal. Recall the
occurrence of [ŋ] in cohesive suffixation as in si[ŋ] er ‘singer’,
belo[ŋ]er ‘belonger’, which violates *ONSET: *ŋ to satisfy PU.
The dependence of this effect on licensing by an independent
word is shown by the noun (war) mo[ŋg] er ‘monger’, whose
historical base (i.e. Old English mangian ‘to barter’) became
obsolete. There is good evidence that the agentive suffix -er
continues to be recognized in this noun (cf. the acceptability of
modification with continual as in continual liar, continual war
monger, as opposed to *continual thief). As has been
demonstrated above, for noncohesive affixation recognition of
the suffix suffices to secure boundary effects.45 By contrast,
PU effects require licensing of the relevant feature by an
independent word to be stable.
The data considered here indicate that the mere existence of a

related word with some specific sound feature is not enough to
license a PU effect. In addition, the words have to have a close
semantic relation, which plausibly determines recognition of
paradigm membership. This point can be illustrated further
with the distribution of schwa in the words in (79). In the base
verbs the constraint *SCHWA, which prohibits the occurrence
of schwa, is violated to satisfy a constraint on sonority
sequencing.46 By (p.244) contrast, in the suffixed forms
violation of *SCHWA is not motivated on phonological grounds
but is necessary to satisfy PU. The occurrence of the relevant
PU effect in the gerunds in (79a), but not in the (historical)
gerunds in (79b), is determined by semantic relatedness.
Specifically, the fact that the nouns in (79b) are unaffected by
the relevant PU constraints results from the fact that their
paradigm membership is not recognized due to semantic
disassociation. Those cases are similar to cases like inkling,
where schwa cannot be analogically extended because the
base word has become obsolete ( ink[ə]l ‘to hint’).
(79)
Page 65 of 114
a. (chuck([ə])ling)ω ‘the act of chuck[ə]l

chuckling’ ‘chuckle’
(tick([ə])ling)ω ‘the act of tick[ə]l ‘tickle’

tickling’
(jugg([ə])ling)ω ‘the act of jugg[ə]l

juggling’ ‘juggle’
b. (crackling)ω ‘crisp skin of roast crack[ə]l

pork’ ‘crackle’
(twinkling)ω ‘an instant’ twink[ə]l

‘twinkle’
(kindling)ω ‘dry sticks of wood kind[ə]l

used to start a fire’ ‘kindle’
(lightning)ω ‘natural electric light[ə]n

discharge in the ‘lighten’
atmosphere’
Similarly, semantic drift in (80a) rules out the possibility of

analogical leveling. By contrast PU effects are possible
whenever there is a close semantic relationship to the base as
is shown by the historical replacement of the lax vowels (cf.
historical ob[e]sity) by tense vowels or diphthongs in (80b):
(80)
a. (extr[e]mities)ω ‘hands extr[i:]me ‘extending

or feet’ far beyond the
norm’
(gr[æ]vity)ω ‘force of gr[e1]ve ‘very serious’

gravitation’
(m[1]serable)ω ‘very m[a1]ser ‘an avaricious

bad’ person’
(f[æ]bulous)ω ‘extremely f[e1]ble ‘type of

pleasing’ narrative’
b. (ob[i:]sity)ω ‘quality of ob[i:]se ‘obese’

being obese’
Page 66 of 114
(cons[oʊ]latory)ω cons[oʊ]le ‘console’

‘affording an
instance of consoling’
([eI]pical)ω ‘pertaining [eI]pex ‘apex’

to the apex’
The irrelevance of semantic affinity for boundary effects in

noncohesive affixation follows from the irrelevance of the
status of the stem.47
9.4.4 Prosodic consistency

Analyzing the occurrences of stressed [iː] in phon[íː]mic, d
[íː]cency, or ob[íː]sity as PU effects has no implication for other
sound properties in the relevant words. For (p.245) instance,
PU(STRESS) may or may not be satisfied as well (e.g.
PU(STRESS) satisfaction correlates in décency—décent, but
not in phonémic—phónème). By contrast, the analysis of some
sound property as a boundary effect necessarily implies
correlating effects. Recall the discussion of the example
shyness, where the analysis of vowel length as a pword-final
boundary effect necessarily implies an ‘initial’ boundary effect
for the following consonant (i.e. ‘unexpected’ fortition) and
vice versa. This sort of mutual implication is referred to as
‘prosodic consistency’ here and will be used as follows to
distinguish boundary effects from PU effects.
(81) The Prosodic Consistency Criterion If in a

morphologically complex string a boundary analysis
captures clusters of systematically correlating
properties non-occurring in simplex words, each of
these properties is to be analyzed as a boundary effect.
To illustrate the proposal in (81) consider aspiration in the

prefixed word in (82b), which could be analyzed as a PU effect
licensed by word-initial aspiration in the base. Note that a stop
between a stressed and an unstressed vowel is normally
unaspirated in English as in (82a) (cf Kahn 1976).
Page 67 of 114
Given a
domain
analysis as
described in
9.2.2 (based
on [[re]HEAD
FW[cogníze]STEM]WORD and [[re]MOD FW[combíne]WORD]WORD)
a PU analysis of aspiration would be ruled out by the Prosodic
Consistency Criterion in (81). This is because aspiration
correlates systematically with other sound properties entailed
by the pword structure in (83).
The lax vowel

in the
cohering CV-
prefix in (83a)
satisfies
Trisyllabic
Laxing and
conforms
entirely to the phonology found in corresponding position in
simplex words (e.g. [é]lefant, p[é]lican, t[é]lephone). The non-
cohering CV-prefix in (83b) satisfies the constraints on vowel
quality and quantity determined by its pword status (cf. CV-
content words such as (síː)ω ‘sea’, (tíː)ω ‘tea’) to the effect that
such vowels are non-occurring in the corresponding position in
polysyllabic simplexes. The non-cohering prefix implies
exclusive onset syllabification for the following stop, which
causes aspiration in (83b).The analysis of aspiration as a
boundary effect, and not a PU effect, in accord- (p.246) ance
with the Prosodic Consistency Criterion, entails and is
supported by additional stress correlations. The strong–weak
pattern in récognìze conforms to the rule that in verbs with
two feet, the final foot is weak if it does not branch
(círcumcìse, réconcìle, álternàte).48 The weak–strong pattern
in rècombíne conforms to the rule that prefixes forming a
separate pword in verbs and adjectives are always weak (cf
(32a)). The 1–3 pattern in (82 a), as opposed to the 2–1 pattern
in (82b), follows from the distinct effects yielded by feet
contained in a single pword (i.e. stronger contrast in relative
prominence) as opposed to feet belonging to separate
pwords.The observation that these correlating stress patterns
Page 68 of 114
could not be mimicked in terms of PU confirms the irrelevance

of PU for the description of morphophonology in (82b).
9.4.5 Sensitivity to speech register

Whereas PU constraints by definition require identity of sound
structure in related words, the sort of identity caused by
satisfied alignment results from identical positions of speech
sounds with respect to some domain boundary. For instance,
there is no need to assume that glottalization of t in nigh[t̚]
rate or doub[t̚] less motivates a PU constraint requiring
phonetic identity to the corresponding stops in night and
doubt, respectively. Rather, any t in syllable-final position will
exhibit the relevant features.
The claim that identity effects resulting from noncohesive

morphology differ in essence from identity effects resulting
from PU constraints is supported by evidence from casual
speech. Specifically, ‘identity’ resulting from noncohesion in
some cases disappears, while boundary effects persist, a
phenomenon which cannot be observed for PU effects.
Compare the boundary effect in the slow and careful
pronunciation nigh[t̚] owl (i.e. glottalization before a vowel),
which could be mistaken for a PU effect licensed by
independent nigh[t̚], with the boundary effect in the fast (or
perhaps normal) pronunciation ‘nigh[r]owl’ (i.e. flapping
before a stressed vowel). Persistence of the boundary effect
results from the persistence of prosodic structure itself: the t
is in both pronunciations followed by a pword boundary and
must therefore associate with the coda to satisfy
CONTAINMENT. By contrast, while (certain) PU effects,
including the schwa in chuck([ə])ling discussed in (79), may
disappear in fast speech, such a change results invariably in
conformity to simplex phonology.
(84) The Casual Speech Criterion If the fast or casual

pronunciation of a morphologically complex string
exhibits a sound property absent from comparable
simplexes and distinct from the careful pronunciation of
that string, that property is to be analyzed as a
boundary effect.
(p.247) To illustrate the Casual Speech Criterion with an

additional example, compare the careful and the casual
pronunciations of the German clitic structures in (85):
Page 69 of 114
(85)
a. careful speech: b. casual Gloss

speech:
[έsʔÍç] ‘ess ich’ [έzÍç] ‘do I eat’
[kÍfʔç] ‘kiff ich’ [kÍvÍç] ‘do I smoke

marijuana’
[∫tɔˊpʔÍç] ‘stopp [∫tbÍç] ‘do I stop’

ich’
Significantly, the casual pronunciations violate a constraint

against voiced obstruents following short stressed vowels in
German, arguably a constraint prohibiting voiced ambisyllabic
obstruents (cf. Becker 1998). There are a few lexical
exceptions for stops (e.g. [έbə] ‘Ebbe’ ‘low tide’) but none for
fricatives. As a result of satisfying both conditions in (84) the
voiced fricatives in (85b) must be analyzed as boundary
effects. Obstruent voicing is indeed most plausibly analyzed in
terms of persisting prosodic boundaries as is shown by the
mapping in (86). In German, too, vowel-initial suffixes, but not
clitics, cohere:49
(86)
Page 70 of 114
a. [έs]WORD [Íç]function → (έs)ω(´ç)ω ‘ess ich’ ‘do I eat’

word
b. [[∫trέs]STEM → (∫trέ sIç)ω ‘stressig’ ‘stressful’

[Íç]HEAD
SUFFIX]WORD
c. [έsIç]WORD → (#x03AD;sIç)ω ‘Essig’ ‘vinegar’
Page 71 of 114
Arguably, the constraint against voiced ambisyllabic

consonants, which accounts for the ungrammaticality—also in
casual speech—of suffixed *[∫trέzIç] (cf. (87b)) and simplex
*[έziç] (cf. (87c)), is dominated by a constraint prohibiting
voiceless ambisyllabic consonants in pword-final position (cf.
(87a)).
(87)
The
German
data show
that
boundary
signals
can be
more
salient in casual speech than in careful speech. That is, [έsiç]
‘ess ich, a more careful pronunciation, is at least near-
homophonous to the simplex [έsiç] ‘Essig’, whereas the more
casual pronunciation [Ezέç] ‘ess ich differs markedly.
(p.248) 9.5 Case Studies: Pu-Effects Versus Boundary

Effects
In s. 9.5.1 I investigate various morphophonological effects
discussed in the literature on æ-tensing, to illustrate the
application of some of the criteria in s. 9.4. The focus of ss.
9.5.2 and 9.5.3 is on poorly motivated use of PU constraints.
9.5.1 The case of æ-tensing

In Philadelphia English the vowel [æ] and its tense
counterpart [E] are (mostly) in complementary distribution.
[E] appears before (phonetically) exclusively tautosyl-labic
voiceless fricatives or nasals other than [ŋ] as in (88a), [æ]
appears elsewhere (cf Ferguson (1972)). In (88b), the
segmental condition for æ-TENSING is not satisfied, in (88c)
the syllabic condition is not satisfied.
(88)
Page 72 of 114
a.b[E]n ‘ban’ b. r[æŋ]k ‘rank c.b[æ]nner ‘banner’
st[E]ff ‘staff’ b[æt] ‘bat’ tr[æ]ffic ‘traffic
m[E]s[$]ter ‘master’ bl[æŋ]ket ‘blanket’ t[æ]ssel ‘tassel’
Page 73 of 114
Benua (1995:10) describes this distribution in terms of the

constraint ranking in (89):
(89) æ-TENSING Â» *TENSE-LOW
The constraint *TENSE-LOW, which prohibits tense low vowels

and accounts for vowel laxness in the ‘elsewhere
environment’ (cf. (88b, c)) ranks below æ-TENSING, a
constraint prohibiting lax [æ] before specific tautosyllabic
consonants. Given this distribution, the occurrences of [E]
qualify as PU effects in (90a), but not in (90b).
(90)
a. b[É]nner‘banner (one who b. har[É]ssment

bans)’ ‘harassment’
c[É]nnable‘cannable (can c[É]n òpener ‘can

be canned)’ opener’
p[È]ssée ‘Passee(one who c[È]n it ‘can it’

is passed)’
Specifically, given the prosodic structures in (91) the

occurrence of tense [E] satisfi æ-tensing in (91b). This is
because the consonant following æ is syllabified in coda po
ition to satisfy CONTAINMENT.
(91)
Page 74 of 114
a [[bán]STEM[er]HEAD → (b[É]nner)ω cf. (b[æ´]nner)ω

SUFFIX]WORD
[[cán] STEM [able] HEAD → (c[É]nnable)ω cf. (c[æ´]nnibal)ω

SUFFIX] WORD
[páss] STEM [ée] HEAD → (p[È]ssée)ω cf. (p[æ` ]ssé)ω

SUFFIX] WORD
b [harass]STEM [mt] HEAD → ((har[É]ss)ωment)ω cf. ([æ´ ]sthma)ω50

SUFFIX] WORD
[[cán]WORD[ópener]WO → (c[É]n)ω(òpener)ω cf. (H[æ´]nnòver)ω

RD]WORD
[[cán]WORD[it]FW → (c[É]n)ωit cf. (J[æ´]nit)ω
Page 75 of 114
(p.249) The phonologically unmotivated occurrence of [E] in

(91a) motivates a PU analysis:
(92)
Page 76 of 114
Base: b[É]n+er PU[NUC] æ-TENSING *TENSE-LOW
(b[æ´]nner)ω *!
✓(b[É]nner)ω *
Page 77 of 114
Comparison of the prosodic structures of the complex words in

(91a) with the corresponding simplexes in the right-hand
column shows that PU effects, but not boundary effects, result
in contrastive distribution of the vowels (i.e. true minimal
pairs).
Turning now to the predictability criterion, we find that tense

[E] indeed occurs regularly in noncohesive contexts (cf
((m[E]ss)ωless)ω ‘massless’). By contrast, in cohesive contexts
the distribution of [æ] versus [E] is more complicated.
Ferguson notes that ‘the inflectional endings -es and -ing
invariably preserve the identity of the stem; agentive -er
generally does so, and there are occasional examples of other
derivational suffixes, although most do not’ (1972: 263). Some
examples are given in (93).
(93)
a. No PU (cl[æ]ssic)ω ‘classic’
effect
(cl[æ]ssify)ω ‘classify’
(m[æ]ssive)ω ‘massive’
(p[æ]ssage)ω ‘passage’
b. PU (cr[E]sser)ω ‘crasser’ (comparative)

effects (cf. cr[E]ss)
(cl[E]sses)ω ‘classes’
(cf. cl[E]ss)
(m[E]ssing)ω ‘massing’
(cf. m[E]ss)
Potential PU effects in (93a) are probably inhibited by

semantic drift. While unfortunately there are no relevant test
cases there is a clear prediction that tense [E] in noncohesive
harassment would persist, even if the word developed an
idiosyncratic meaning or harass became obsolete. By contrast,
in cohesive suffixation like b[É] nner ‘one who bans’, the effect
is dependent on the independent existence of the semantically
closely related base form b[E] n ‘ban’, which licenses [E].
In addition to the cases in (90a), where æ-TENSING ‘over-

applies’, there are also several cases of ‘under-application’.
Ferguson (1972: 264) notes that there are ‘four common verbs
Page 78 of 114
and two archaic forms of a common verb’ which exceptionally

fail to undergo æ-TENSING: am, can ‘be able’, ran, began,
hast, hath. However, ‘commonness’ is not a relevant property
here. Failure of æ-TENSING in am, can ‘be able’ is a domain
effect, resulting from the non-alignment of pword boundaries
with head function words. Additional cases of this type of
‘under-application’ are listed in (94a), where æ-TENSING
never applies even under (contrastive) stress. The claim is
then that there is an additional prosodic condition on æ-
TENSING: the rule applies only if the relevant syllable is
dominated by a pword.51 Failure of æ-TENSING in (94a)
correlates then with (p.250) stress instability (e.g. (æm)Σ ~
(əm)σ), as opposed to the stable feet in (94b), which satisfy
undominated HEADEDNESS.52
(94)
Page 79 of 114
a. [am]HF WORD → ([æ]m)Σ‘am’ b. [ham]WORD → (h[E]m)ω

‘ham’
[can]HF WORD → (c[æ]n)Σ ‘can’ [can]WORD → (c[æ]n)ω ‘can’
(auxiliary)
[and]HF WORD → ([æ]nd)Σ ‘and’ [land]WORD → (l[E]nd)ω

‘land’
[an]HF WORD → ([æ]n)Σ ‘an’ [ban]WORD → (b[E]n)ω ‘ban’
[than]HFWORD → (th[æ]n)Σ‘tha [man]WORD → (m[E]n)ω

n’ ‘man’
Page 80 of 114
Turning now to Ferguson’s remaining cases of under-

application, failure of æ-TENSING in ran and began is a PU
effect. Tis is because æ-TENSING fails precisely for derived
forms based on words with some lax vowel other than [æ],
which are not affected by tensing.53
(95)
a. (sw[æ]m) ‘swam’—(sw[1]m) b. h[E]m

‘swim’ ‘ham’
(r[æ]n) ‘ran’—(r[Λ]n) ‘run’ sc[E]n ‘scan’
(beg[æ]n) ‘began’—(beg[1]n) ‘begin’ p[E]n ‘pan’
‘Under-application’ of æ-TENSING in (95a) is described by the

ranking in (96). The candidate sw[1]m]PAST, which perfectly
satisfies PU[NUC] (cf tableau (2)), is eliminated because it
violates the constraint CONTR: TENSE, which requires past
tense forms to differ from present tense forms (for related
effects, cf. Alderete 2001; Kenstowicz, Ch. 7 above, Rebrus and
Törkenczy, Ch. 10 below). The candidate sw[E]m]PAST, which
satisfies æ-TENSING, is eliminated because it violates
PU[NUC] more than the optimal candidate sw[æ]m]PAST,
which violates PU[NUC] only minimally to satisfy CONTR:
TENSE:
(96)
Page 81 of 114
Base: CONTR: TENSE PU[NUC] æ-TENSING *TENSE-LOW

sw[1]m+PAST
sw[E]m]PAST **!
✓ sw[æ]m]PAST * *
sw[1]m]PAST *!
Page 82 of 114
clusters (e.g. cild > cild ‘child’, climban > cliimban ‘climb’, cf.
Luick (19641 242 ff ). Luick (1964: 243) notes that the rule did
not apply to function words such as and, under, should, would.
(p.251) The archaic forms of have are amenable to either the

domain analysis, since they (perhaps) could be analyzed as
head function words, or to the PU analysis, since their base
have has a lax vowel not affected by æ-TENSING.54
A third case of under-application also discussed by Ferguson is

illustrated in (97) (cf. Kahn 1976; Benua 1995).
(97)
P[æ]mela ‘Pamela’ P[æ]m ‘Pam’
J[æ]nice ‘Janice’ J[æ]n ‘Jan’
M[æ]ssachuse ‘Massachuset’ M[æ]ss ‘Mass’
Benua (1995:12) analyzes the vowel quality in the truncated

forms as a violation of regular æ-TENSING to satisfy the
higher-ranking constraint IDENT-BT [TENSE], which requires
the feature [tense] in the truncated form to be identical to the
corresponding feature in the base. The analysis is repeated in
(98) (Tableau (20) in Benua 1995).
(98)
Base: [pæmələ] IDENT-BT [TENSE] æ-TENSING
a. ✓ pæm *
b. pEm *!
While I consider the analysis basically correct I disagree with

Benua’s claim that the correspondence constraint refers to
phonetic output forms. Instead, I suggest that PU[NUC] is at
work again, which requires identity of nuclei in lexical
representations. I assume then that the phonetic contrast
between [æ] and [E] is encoded in the lexicon as a result of
Lexicon Optimization (cf. Prince and Smolensky 1993), despite
the largely complementary distribution. In general, I assume
that all phonetic surface features are lexicalized as long as
they are universally contrastive and are consistent with lexical
syllable structure. The last condition is meant to account for
the observation that the phonetics of pword-final consonants
Page 83 of 114
appears to be inaccessible to PU effects. Consider the lack of

correspondence in the truncations in (99), noted by Hale,
Kissock, and Reiss (1998), which is problematic for a strictly
phonetic-based approach to PU.
(99) Pe[ɾ]er ‘Peter’ Pe[t ̚]e Ju[ɾ]ith ‘Judith’ Ju[d]e
It appears that the ungrammaticality of paradigmatic leveling

in (99) (i.e. both Pe[ɾ]er —*Pe[ɾ] and *Pe[t ⌝] er—Pe[t ⌝] e) is not
a specific fact about English. Assuming that PU constraints
apply at the lexical level and given the assumptions regarding
lexical versus phonetic syllabification stated in s. 9.2.4, the
possibility of leveling in (99) can perhaps be ruled out in
principle. That is, whether we assume that Peter or Judith are
represented with an ambisyllabic intervocalic consonant in the
lexicon or not and whether we assume that ambisyllabicity is
extended to the truncated form to satisfy PU or not, inviolable
CONTAINMENT will result in an exclusively coda-associated
pword-final (p.252) consonant in the phonetic representation
of Pete and Jude.55 The corresponding allo-phony (e.g.
glottalization for t) is an automatic consequence of the
phonetic syllable structure.
Consider next
the question
of why pword-
final coda
phonetics (as
opposed to
the phonetics
of the
nucleus)
seems to be immune to undergoing paradigmatic leveling (e.g.
Pe[t⌝]e—Pe[ɾ]er, not *Pe[t⌝]er, wri[t⌝]e ‘write’—wri[ɾ]er, not
*wri[t⌝]er ‘writer’). A possible explanation here is that the
lexicalization of coda-specific phonetic features is blocked if
the relevant consonant is associated with onset position in the
lexicon. The relevant hypothesis is stated in (101).
(101) The lexicalization of syllable-conditioned

phonetics presupposes non-distinct syllable structure in
the lexicon.
Page 84 of 114
According to (101) it is expected that for languages in which

pword-final consonants have onset profile (e.g. English,
German) the corresponding coda-conditioned phonetics cannot
be lexicalized and therefore cannot give rise to PU effects. By
contrast, syllable-conditioned allophonic variation affecting
the nucleus is highly prone to lexicalization and, as a result,
susceptible to PU constraints (cf. the data discussed in Harris
1994).
The second type of data cited by Hale, Kissock, and Reiss

(1998) to refute the analysis in (98) concerns the formations in
(102), which show ‘that truncated forms may maintain vowel
contrasts that are neutralized in full forms’ (1998: 229).
(102)
P[ə]tricia ‘Patricia P[æ]t (εf. *P[ə]t)
G[ə]rard ‘Gerard’ G[]r (cf. *G[ə]r)
L[ə]rraine ‘Lorraine’ L[o]ri (cf. *L[´]ri)
(p.253) The assertion by Hale, Kissock, and Reiss that in

(102) truncated forms are faithful to alleged abstract
underlying forms raises the question of how those forms are
learned. On the analysis proposed here phonology of
truncation in (102) is not an identity effect of any sort but
rather follows directly from undominated HEADEDNESS and
is hence a domain effect. This is because truncated words, too,
are mapped into pwords, which necessarily dominate a foot
with the result that a full vowel (rather than a schwa) surfaces.
The precise choice of the full vowel in (102) is apparently
determined by the spelling, which means that the coinage of
the truncated forms necessarily presupposes knowledge of the
relevant written forms. The complete consistency of the
patterns in (102) with those in (97) is illustrated in (103) (‘LR’
means ‘lexical representation).
(103)
Page 85 of 114
a. LR: /pətrI∫ə/ HEADEDNESS PU-NUC æ-TENSING

‘Patricia’
⌗(pæt)ω *
(pət)ω *!
b. LR: /pæmələ/
‘Pamela’
⌗(pæm)ω *
(pEm)ω *!
Page 86 of 114
To summarize, Hale, Kissock, and Reiss (1998) fail to refute

the relevance of PU constraints in truncation but merely
present evidence that such constraints are violable (in
contrast to HEADEDNESS and CONTAINMENT) and do not
apply to phonetic surface forms. Their conclusion that all
truncated forms should be treated as ‘lexicalized’ on a par
with cases like Ted (from Edward) or Bob (from Robert) misses
important generalizations, including the observation that the
vowel in Pam is a genuine PU effect.
9.5.2 Alleged PU effects for l-velarization in English

In various dialects of English the pronunciation of /l/ depends
on syllable position: in exclusive onset position, so-called
‘light’ [l] occurs, which implies tongue-blade contact, whereas
in exclusive coda position so-called dark [l] occurs, for which
tongue-blade contact can be lost (i.e. a type of lenition). In
addition, [l] implies specific allophony of the preceding vowel
illustrated in (104) (excerpt from the table in Hayes 2000: 93).
(104)
Default Example Pre-[l] Example

quality allophone
[i:I] tree [t⌋i] [iə] feel [fbəl]
[I] fit [fIt] [iə] fill [fIəl]
[eI] pay [peI] [eə] pail

[peəl]
[aI] tie [taI] [aIə] tile [taIəl]
[aʊ] cow [ʊə] cowl

[kaʊ] [kaʊəl]
(p.254) Hayes (2000) conducts an experiment where he

reads a list of seventeen words both with [l] and with [1],
where the two pronunciations systematically correlate with
the corresponding vowel allophony illustrated in (104). Ten
informants, who were told the meanings of the words, were
asked to rate the pronunciations from 1 designating ‘just
right’, to 7 designating ‘awful’. The results of this study for
simplexes are given in (105), where the numbers are the mean
of the judgement of all ten consultants. The characterization of
syllable position is mine (based on Kahn 1976).
Page 87 of 114
(105)
Page 88 of 114
with [l] with[ł]
a. Exclusive Onset light [l]ight—[ł]ight 1.30 6.10
Louánne [l]ouánne— 1.10 5.55

[ł]ouánne
b. Exclusive Coda bell [bɛl]—[bɛəł:] 6.60 1.20
help [hɛlp]—[hɛəłp] 6.60 1.05
c. Ambisyllabic (Norman) Mailer 2.00 2.00

[meIlɚ]—[meɛłɚ
Hayley (Mills) [heIli]— 1.55 3.05

[heɛłi]
Page 89 of 114
The ratings for the two candidates in both exclusive onset and
exclusive coda position are fairly clear. For the ambisyllabic
context the ratings are interpreted as indicative of ‘free
variation’ with a ‘slight preference for [l]’ (Hayes 2000: 95,
96).56
The (for the purpose of this chapter) significant part of the

study concerns the ratings of l in morphologically complex
words. Three contexts can be distinguished in structural terms
which correlate with three rating patterns:
(106)
Page 90 of 114
with [l] with[ł]
a. Word + clitic mail it [méIlIt]— 4.40 1.10

[méəłIt]
b. Stem + consonant- gray-ling [gréIlŋ]— 1.39 3.17

[gréəłIŋ]
initial suffix gay-ly [géIlI]—[géəłi] 1.45 3.65
free-ly [fríːli]—[fríəłi] 1.85 3.20
c. Stem+ vowel-initial mail-er [méIlɚ]— 2.79 2.00

[méəłɚ
suffix hail-y [héIli]—[héəłi] 4.00 1.56
(touchy-) feel-y [fíːli]— 2.00 220

[fíəłi]
Page 91 of 114
In the approach to English word prosody outlined in s. 9.3 the

three distinct rating patterns in (106) follow from the three
distinct types of prosodic structures in (107). That is, strong
preference for dark [ł] in mail it and (somewhat weaker)
preference for light [l] in grayling are boundary effects, which
follow from the position of the laterals within the respective
syllabification domains. The l in mail it exhibits coda properties
(i.e. it is dark) because it is syllabified as a coda to satisfy
CONTAINMENT. Similarly, the l in grayling exhibits onset
properties (i.e. it is light) because it is syllabified as an onset
to satisfy CONTAINMENT. The stronger preference for [l] in
(105a) compared to (106b) can also be explained based on
prosodic context: l is pword-initial in (105a), but not in (106b).
(p.255)
(107)
a. [mail]WORD [it]HEADFW → (mail)ωit ‘mail it’
b. [[gray] STEM[ling] HEAD → ((gray)ω ling)ω

SUFFIX]WORD ‘grayling’
c. [[mail]STEM[er]HEAD → (mailer)ω
SUFFIX]WORD
Assuming the prosodic representation in (107) the phonetics

of l in mailer is not a potential boundary effect. However, it is
far from clear whether there is a PU effect. The ratings for
dark [l] in simplex Mailer and suffixed mailer are identical
according to (105c) and (106c). It is unclear what accounts for
the significant difference between the ratings for mailer,
touchy-feely, which barely, if at all, deviate from simplex
patterns, and haily, galy, which do deviate. Could the
preferences expressed by the consultants be influenced by
their desire to (artificially) express links to semantic base
forms for words that are utterly obscure? Moreover, are there
PU effects in cases where l appears in pretonic (i.e. onset)
position such as pòllée ‘one who is questioned in a poll’,
expèllée ‘one who has been expelled’ and if not, why not? And
if true PU effects were demonstrated in (106c), could it be that
the relevant effect is limited to the phonetics of the vowel
without affecting the pronunciation of the consonant (cf the
discussion of hypothesis (101) in s. 9.5.1)? Detailed
Page 92 of 114
experimental studies are needed to answer those questions.

For now, I conclude that there are two clear boundary effects
in (107a) and (107b) and an unclear PU effect in (107c), where
PU possibly affects only the vowel.
Consider now Hayes’s (2000) description of the data in (106)

exclusively in terms of PU. To account for the distinct ratings
between mail it and mailer two separate PU constraints are
introduced, PU (PHRASAL) for phrasal contexts versus PU
(MORPHOLOGICAL) for morphological contexts, ‘with the
former ranked within UG as necessarily stricter than the
latter’ (Hayes 2000:102). This proposal raises many questions.
What exactly is the extension of the paradigm in each case?
How is the attractor determined in each case? Which of the
two constraints is responsible for the (presumably strongly
preferred) dark [l] in mail order, or is there a third constraint,
PU (COMPOUND)?
Both cases in (107b) and (107c) are described by

PU(MORPHOLOGICAL), which raises the problem that
‘paradoxically, these constraints cannot be stated on /l/ per se.
This is because in grayling and similar forms, the crucial light
[l] does not actually occur in the base form gray (Boersma and
Hayes 2001). This paradox is solved by basing PU constraints
on vowels and by specifying undominated constraints to
secure the appropriate match-up of [l] and [l] with the
corresponding vowel allophones (cf. Hayes (2000: 101)). The
proposal is illustrated in (108), where the arrow ‘ ⇔ ’ indicates
the required match-up between vowel allophones and l-
allophone illustrated in (104):
The parallel
treatment of
the cases in
(108) raises
the question of why the preference for light [l] in grayling is
stronger than the preference for dark [ł] in mailer. This
contrast (p.256) could not be caused by the pronunciation of
the alleged attractors: grey is clearly pronounced with [éI] as
mail is clearly pronounced with [éə]. On my view this contrast
indicates that two separate phenomena are involved. Light [l]
in (108a) is a boundary effect and the alleged ‘match-up’ with
the preceding vowel is simply a matter of ‘pro-sodic
consistency’ (cf. s. 9.4.4). Internal boundary effects are
Page 93 of 114
excluded for mailer but a PU effect, especially one affecting

the stressed nucleus, is plausible (cf. s. 9.5.1).
Additional evidence in support of the analysis of light [l] in

grayling as a boundary effect comes from semantics. The sort
of semantic dissociation in (109) is unproblem-atic for a
boundary analysis, but not for a PU analysis. Recall the data in
(79), where comparable semantic dissociation inhibited any
sort of paradigmatic influences from etymological base forms:
(109) grayling ‘a fresh-water fish (genus Tymallus)’

versus grey ‘colour term’ eyelet ‘a small hole through
which a rope etc. is passed’ versus eye ‘organ of sight’
The distribution of l-allophony in affixation based on a bound

stem would be an even better testing ground. Unfortunately,
there is only a single relevant example, the somewhat archaic
adjective thowless defined as ‘devoid of energy or spirit’ in the
OED. There is no reason to doubt that this adjective is like
reckless, hapless in that once the suffix is recognized, perhaps
based on the privative meaning of the adjective or perhaps
because of its spelling, its prosodic structure is
indistinguishable from words based on an independently
existing stem such as cowless.
(110)
(110) [[thów] STEM ((thow)ωless)

‘thowless’
[less]HEAD SUFFIX]WORD
[[ców]STEM[less]HEAD ((cow)ωless)‘cowless’
SUFFIX]WORD
Whereas on a boundary approach the suffix-initial l would be

light, precisely as in cow-less ‘destitute of cows’,57 there is a
clear, but incorrect, prediction on Hayes’s approach that this
word must be treated like a simplex, with free variation
between [θaʊləs] and [θaʊəłəs].
To summarize, Hayes’s attempt to mimic the boundary effects

in (106a, b) in terms of PU constraints comes not only at the
expense of introducing a questionable separation of PU
constraints in phrasal versus morphological constraints. In
addition, the approach fails empirically in that certain
boundary effects (i.e. semantically dissociated words or words
Page 94 of 114
based on bound stems) simply cannot be analyzed as PU

effects.
Page 95 of 114
9.5.3 Alleged durational PU effects in French

According to Rialland (1986) schwa loss in casual speech in
(111a) does not lead to homophony with (111b). Rather the
subtle phonetic contrasts described in (111c) exist.
(p.257)
(111)
a. [padəʁ iol] > [padʁ iol] b. [padʁ jol]
‘pas de/d’rôle’ ‘no role’ ‘pas drôle’ ‘not

funny’
[baʁ iətʁuve] > [baʁ itʁuve] [baʁ ktʁuve]
‘bas re/r’trouvé’ ‘stocking bar trouvé ‘bar

found again’ found’
c ʁ exhibits typical onset properties (e.g. greater

i acoustic energy longer duration no
lengthening effect on preceding vowel)
ʁ exhibits typical properties of a syllable

j internal onset (e.g. decreased energy in the
higher formants shorter duration)
ʁ exhibits typical coda properties (e.g.

k decreased energy shorter duration
lengthening effect on preceding vowel)
To account for the distribution of the ʁ-allophones in (111)

Rialland suggests that schwa loss does not imply syllable loss
but results in the reassociation of the nucleus with the
following consonant as in (112b). The phonetic contrasts
between (111a) and (111b) follow then from the distinct
syllable structures in (112b) versus (112c).
Steriade
(2000) rejects
Rialland’s
analysis
because both
d’rôle and
drôle count as monosyllabic in metrical scansion. The
association of [ʁ] with both nucleus and onset position in
(112b), presumably a novel type of syllable structure resulting
Page 96 of 114
exclusively from schwa loss, seems indeed unmotivated on

phonetic grounds and furthermore fails to express identity to
the corresponding sound in (112a). A serious empirical flaw
concerns the fact that schwa loss in d(e)rôle is treated on a par
with schwa loss in the word f(e)rai, which however for most
speakers results in true homophony with (historically schwa-
less) frais. The ‘Ø’ in (113) marks the site of the historical
schwa.58
(113) a. pas d[Ø] rôle ≠ pas drôle b. f[Ø]rai = frais
The last criticism relates to the question of acquisition. Can

learners be expected to interpret the phonetic contrast
perceived in pairs like drôle versus drôle in terms of fossilized
syllable structures as in (112)? The criticisms of Rialland’s
analysis are summarized in (114).
(114)
a. incorrect predictions regarding syllable count
(cf Steriade 2000: 328);
b. questionable type of ambisyllabicity in (112b)/
introduction of novel,non-structure preserving
syllable structure;
c. inadequate representation of phonetic
correspondence; cf. (112a) vs (112b)
(p.258)
d. lack of discrimination between distinct types
of schwa loss (cf. (113));
e. the question of acquisition.
Steriade explicitly dismisses the idea of linking the allophonic

distribution described in (111) to syllable structure. Instead
the phonetics of r in (111) is claimed to satisfy PU (Left:
Duration), presupposing no contrast in syllable structure
between d’rôle and drôle.
(115) PU (Left: Duration) (Steriade 2000):If two

consonants, C and C', stand in correspondence and C is
morphemeinitial in the careful pronunciation of the
relevant morpheme, C' is durationally equivalent to C.
The analysis is illustrated in (116), where identity expressed

by ‘=’ is required to satisfy PU (Left: Duration), resulting in
deviation from simplex phonetics expressed by ‘≠’.
Page 97 of 114
Invoking PU
to account for
the effect
observed in
(111) is
questionable
since non-
distinct word
forms
belonging to a single paradigm but distinct registers are
related.59 Before discussing Steriade’s analysis further I will
demonstrate that reference to PU constraints is unnecessary
in that none of the criticisms in (114) apply to a boundary
analysis based on the syntactic distinctions specified in (117).
The crucial and, I believe, uncontroversial distinction is that
r(e) trouvé, but not d(e)rôle, is a word. Note the use of a single
phoneme /ʁ/ in the left column and the predictability of the
three allophones in the right column on the basis of position
within the parenthesized syllabification domains (i.e. [ʁi]:
initial, [ʁi: prenuclear, non-initial, [ʁk]: coda).
(117)
a. [pa]MOD[d]HEAD FW → (pa)(də)(ʁiol) pas

[ʁol]WORD d’rôle ‘no role’
b. [pa]MOD [d]HEAD FW → (pad)(ʁiol) pas d’rôle

[ʁ;ol]WORD ‘no role’
c. [pa]MOD [dʁol]WORD → (pa)(dʁ j ol) pas drôle

‘not funny’
d. → (ba)( ʁiə)(tʁuve) ‘bas

[ba]WORD[[ʁə]MOD[tʁuve]WORD ]WORD
retrouvé’
e. → (ba)( ʁitʁuve) ‘bas

[ba]WORD[[ʁ]MOD[tʁuve]WORDr’trouvé’
]WORD
f. → baʁ k)(tʁuve) ‘bar

[baʁ]WORD[tʁuve]WORD trouvé’
(p.259) The constraints relevant for the mapping in (117) are

stated in (118), where NUC: VOC is the phonological
markedness constraint enforcing cohesion of the consonantal
morphemes in (117b, e). Other relevant consonants which I
assume are undominated include PARSE (‘every segment must
be parsed into syllable structure’) and SYLL: NUC (‘every
Page 98 of 114
syllable must dominate a nucleus’), where the latter can

arguably be subsumed under HEADEDNESS. Reference to
pwords is avoided because in French syllabification domains
fail to coincide with stress domains.
(118)
a. NUC: VO C (‘a nucleus must be filled by a
vowel’)
b. ALIGN (Word, L; σ,L), ALIGN (Word, R; σ,R)
c. ALIGN (Mod, L;σ,L), ALIGN (Mod, R; σ,R)
d. ALIGN (Head FW, L; σ,L), ALIGN (Head FW,
R; σ,R)
NUC: VOC forces the integration of consonantal morphemes

into an adjacent syllable. Ranking ALIGN (Word, L; σ,L) above
ALIGN (Word, R; σ,R) causes leftward integration of the head
function word in (117b) and rightward integration of the prefix
in (117e). Alternatively, one could also ‘derive’ the schwa-less
forms by ranking the constraint *SCHWA above NUC: VOC to
describe casual speech. The first candidates listed in tableau
(n9 a) and (119 b), respectively, are optimal in careful speech,
where *SCHWA (p.260) ranks lower than the constraints in
(118). For both registers the alignment constraints in (118c, d)
rank below the constraint in (118a).
(119)
Page 99 of 114
a. [pa]MOD *SCHWA SYLL: NUC ALIGN (Word,L; ALIGN (Word,R;

[d]FUNCWORD[ol]WO σ,L) σ,R)
RD
(pa)σ (da)σ (Kol)σ *!
(pa)σ(d)σ( Kol)σ *!
(pa)σ (dirol)σ *!
√ (pad)σ (Kol)σ
b. [ba]WORD[[ʁə]MOD[
tʁuve]WORD]WORD
(ba)σ (ʁə)σ(tʁuve)σ *!
(ba)σ (ʁ)σ(tʁuve)σ *!
(baʁ)σ(tʁuve)σ * *!
√ (ba)σ (ʁtʁuve)σ *
c. [fəʁε]WORD
(fəʁε)σ *!
√ (fʁε)σ
Page 100 of 114

The description in (119) agrees with Rialland’s analysis (and

differs from Steriade’s PU analysis) in that the subphonemic
contrasts are interpreted in terms of distinct syllable
structures as shown in (120) but none of the criticisms in (114)
apply. Both pas drôle and pas d’rôle are bisyllabic (cf.(114a)),
the description is entirely ‘structure-preserving’ as no new
type of syllable structure is introduced (cf. (114b)) and
phonetic correspondence is related to corresponding syllable
structures (cf. (114c)). The phonetic contrast between schwa
loss in d(e)rôle compared to f(e)rais is explained (cf. (114d)).
Finally, the description causes no problems for acquisition,
assuming that children recognize the difference between
words and function words (cf. (114e)).
Unlike
Rialland’s
analysis, the
domain
analysis
relates the
distribution of
r-allophones in (111) to subphonemic syllabification effects
which do not involve distinct registers. Consider now the
boldfaced contrasts for (121a, a'), (121b, b') and (121c, c')
respectively, which are determined by contrasts in
morphosyntactic structure.
(121)
Page 101 of 114

a. [reste]WORD → (ʁ[ɛs.t]e) rester ‘remain’
a'. → (d[e])σ(stabilize)σ déstabilizer

[[de]MOD[stabilize]WOR
D]WORD
‘destabilize’
b. [syblime]WORD → (sy[.bl]ime) sublimer ‘sublimate’
b'. → (sy[b])σ(lέgɥal)σ ‘sublingual’

[[syb]MOD[έgɥal]WORD]W
ORD
c. [diatɔnik]WORD → (d[j]atɔnik)σ ‘diatonic’

‘diatonique’
c'. → (d[i])σ(atomik)σ ‘diatomique’

[[di]MOD[atɔmik]WORD]W
ORD
Page 102 of 114

Word-internally, the cluster st is heterosyllabic in French. As a

result, lax [ε] as in (121a), but not tense [e], may precede st
because of a rule known as ‘Closed Syllable Adjustment’ (cf.
Schane 1968; Lowenstamm 1981). A (noncoronal) stop is
syllabified as onset before [l] as in (121b). Finally, to satisfy
the constraint *[V.V], which prohibits hiatus, [i] loses its
syllabicity before [a] as in (121c) (cf. Gougenheim 1935;
Johnson 1987). The specific types of violations of these regular
syllable patterns boldfaced in (121a', b', c') follow immediately
from the satisfaction of the alignment constraints in (118). The
(erroneous) impression that in French the phrase, rather than
the word, constitutes the domain for syllabification is
presumably due to undomin-ated ONSET, which induces
cohesion for all consonant-final morphemes followed by (p.
261) vowel-initial morphemes thereby causing prosodic
neutralization as in (122).60
(122)
a. [inisjal]WORD → ([i.ni]sjal) initial ‘initial’
b. [[in]MOD[imitabl]WORD]WORD → ([i.ni]mitabl)
inimitable‘inimitable’
Unlike ONSET, the second phonological constraint triggering

cohesion, the above mentioned constraint *[V.V], does not
dominate all alignment constraints. Rather, cohesion results
across stem boundaries as in (122 a), but not across word
boundaries as in (122 b).
(123) [[kɔlɔni]STEM[al]HEAD SUFFIX]WORD → (kɔlɔn[j]al)σ

‘colonial’ [[səmi]MOD[aʁid]WORD]WORD →
(səm[i])σ(aʁid)σ‘semi-arid’
Returning to Steriade’s analysis it appears that the constraint

PU: LEFT DURATION mimics certain aspects of ALIGN (Word,
L; σ,L) and associated CONTAINMENT effects. Given that the
problem with Rialland's analysis concerning syllable count, as
well as the remaining problems in (114), can be solved within
a domain analysis in which a single representation of /ʁ/ is
assumed in the lexicon (cf (119)) it is unnecessary to assume
the lexical representation of various (universally)
nondistinctive allophones implied by the PU analysis. The
French data therefore do not refute the notion of (universal)
distinctiveness as an essential property of lexical phonological
structure.
Page 103 of 114

Given the immense theoretical significance of the question of

whether the PU analysis of subphonemic contrast in cases like
(111) is valid it is perhaps important to go beyond
demonstrating that reference to PU constraints is superfluous
and to present evidence for empirical inadequacy. The PU
analysis fails to account for prosodic consistency in that only
domain-initial, but not the corresponding domain-final,
boundary effects are expressed. Steriade mentions some
relevant data in her informal discussion: ‘It is the relative
duration of a in [ba] and the syllable-initial properties of [ʁ] in
[ʁətʁuve] that are preserved in the corresponding schwa-less
phrase [baʁtʁuve].’ While the relevant data could conceivably
be described by positing additional PU constraints to describe
‘morpheme-final’ phonetics, the systematic clustering of the
relevant effects would not be captured (cf. the Prosodic
Consistency Criterion in (81)). The Casual Speech Criterion in
(84), if valid, would also rule out a PU analysis as for instance
casual [pad. ʁol] ‘pas d’rôle’ exhibits a sound property (i.e. a
heterosyllabic cluster [d. ʁ]), which occurs neither in
corresponding careful pronunciation (i.e. [pa. dəʁol] nor in the
casual pronunciation of simplexes.
Coda syllabification of morpheme-initial d in [pad. Kol] ‘pas

d’rôle’, which violates PU: (LEFT: DURATION), brings to light
the major challenge for a PU based analysis, that is, to account
for cohesion. Why do function words consisting of a single
consonant fail PU if the following word starts with a
consonant? (Because PU: NUC forces (p.262) leftward
cohesion.) Why is Steriade’s constraint PU: (LEFT:
DURATION) in (115) limited to consonants? (Because high-
ranking ONSET causes cohesion for many vowel-initial
morphemes.) Why is there a PU effect for final [i] in prefixes,
but not in stems? (Because ALIGN (Word, L; σ,L) dominates
*[V.V], causing cohesion of suffixes.) None of the relevant
phenomena can be explained in terms of a PU analysis but
they strongly motivate a boundary approach.61
Page 104 of 114

9.6 Conclusions
In this chapter 1 have argued for the necessity of properly
distinguishing PU effects from domain effects and have
proposed various criteria for doing so. Proper distinction of
the relevant phenomena is a crucial prerequisite for all further
inquiry. PU effects, once they have been properly delimited,
shed light on many aspects of lexical structure of major
concern to linguists, including the organization of paradigms
in the mental lexicon and the question of which phonetic
features are represented in the lexicon. Mostly false
conclusions will be drawn on the basis of misclassified PU
effects.
Notes:
(1) The rule has exceptions, especially among proper nouns
(cf. Jacob, Josef).
(2) I assume with McCarthy (Ch. 8 above) that PU effects are

restricted by ‘Base Priority’ (cf. Benua 1997) unless the
concept of a base does not apply as in inflectional morphology
(especially in agreement morphology, cf Raffelsiefen 1995).
(3) The conclusion follows from the premise provided that the
relevant features are indeed affected by PUconstraints directly
(cf Davis (Ch. 5 above), see also n. 61).
(4) Trubetzkoy does not discuss this particular example but

comparable cases like we learn versus willearn, where the
phonetic differences observed for the l are analyzed as
boundary effects derived from identical phonemic
representations.
(5) According to Wells (1990b) the n in minus is simply a coda

consonant whereas the n in shyness forms an onset.
(6) The citation marks are used here because the precise
meaning of ‘surface’ is the bone of contention here.
(7) Peperkamp (1997: 37) describes the second clause of the

original Strict Layer Hypothesis by a constraint ‘PROPER
NESTING’, defined in terms of alignment.
(8) These assumptions are not accepted by all linguists. Szpyra

(1989) and Hall (1998) propose descriptions in which
Page 105 of 114

HEADEDNESS is violated. In Hannahs’s (1995) description of

French, CONTAINMENT is violated (cf. s. 9.6.3)
(9) The constraint in (10a) says: ‘For any GCat in the

representation, align its (right or left) edge withthe (right or
left) edge of some PCat.’
(10) Peperkamp (1997: 30) argues on the basis of Italian data

that pword boundaries do not necessarilycoincide with
syntactic boundaries. However, the evidence cited in support
of this claim is inconclusivebecause the relevant effects could
be PU effects instead.
(11) Aronoff and Sridhar’s rule for constructing pwords says

‘Assign a pword boundary to the edge of any element which is
a member of a major lexical category unless that edge is
adjacent to a stem affix’ (1983: 15). ‘Stem affixes’ differ from
‘word affixes’ with respect to ‘the category of elements to
which they attach’ (ibid. 13)).
(12) The term ‘prosodic word’ is a synonym for ‘phonological

word’.
(13) The omission of -let, -ling, -ing is presumably an oversight.

The list also includes a few questionableitems (e.g. stress in -
ous-suffixation never falls to the left of the antepenultimate
syllable as in gelátinousbased on gélatine).
(14) According to Piggott (1999), word-final consonants should

be analyzed as codas in languages where they have a coda
profile and as onsets in languages where they have an onset
profile. English clearly represents the latter type.
(15) cf. Hall (2001, 2002) for an alternative analysis, where the
special status of pword-final consonants isdescribed by a
constraint which requires trimoraic rhymes to align with the
right edge of a pword.
(16) Some independent evidence for distinguishing

phonological from phonetic principles of syllabification is
discussed in Raffelsiefen (2004a).
(17) HFF may be partially sensitive to phonological properties,

specifically the occurrence of perceptually weak boundary
signals.
Page 106 of 114

(18) While category I prefixations are typically Romance

loanwords and category III prefixations are typically native
coinages, category II prefixations include both loanwords (e.g.
reduce) and native coinages(e.g. berate).
(19) The verb rerate is not listed in Wells but the prefix re- in
many comparable words including rerun,rewrite, redo is
transcribed with a stress mark and vowel length.
(20) It might be objected that the distinct morphosyntactic

representations of seduce versus reduce in (33a) fail to
account for the shared ‘duce—duct‘ alternation observed for
various verbs ending in -duce (e.g. reduce—reduction, induce—
induction, seduce—seduction). However, such alternations can
be shown to ‘survive’ in words which synchronically are
clearly analyzed as simplexes such as German fressen ‘eat (of
an animal)’, past tense: fr aβ (where initial [fr] is the fused
reflex of the prefix ver-) which continues to pattern with the
historical base essen ‘eat (of a person)’, past tense: aβ.
(21) This description differs from previous ‘stem-oriented’

approaches, where reduce and seduce are represented on a
par (e.g. re=duce, se=duce in Chomsky and Halle (1968)).
(22) Occasional apparent counter-examples such as the

intransitive use of reconsider, where the prefixforms a
separate pword, (√Mary reconsidered versus *Mary
considered) are best understood as elliptical.
(23) The dependency of boundary effects on initial stem stress

is also seen in (i), where the initial stem syllable can be
stressed as a result of being closed.
((i))
[[re]HEAD → ((r[iː])Σ láx) ‘relax’

PREFIX[láx]STEM]WORD át
[[re] HEAD → ((r[iː])Σ ‘relaxation’

PREFIX[láxàtion] láxàtion)ω
STEM]WORD
(24) The suffix -ation coheres and can therefore be ignored

here (cf. s. 9.3.3.2).
Page 107 of 114

(25) In Webster’s (1984), but not in Wells (1990a), immortal

and other words including the modifier iN-(e.g. illegal,
irregular, immoral) are represented with an optional geminate
sonorant, as opposed to immingle and other words including a
head prefix, which never include a geminate. Wells represents
a contrast insyllabification between words like inapt (i.e. inapt),
where iN- functions as modifier, and words where itdoes not
function as a modifier (e.g. i$nert ‘inert’). Both dictionaries
represent a contrast in stress betweenmodifier prefixes,
including iN-, which have secondary stress, as opposed to head
prefixes. The contrast insyllabification between enable and
unable is represented in MacCarthy (1945), but neither in
Webster’s norWells. MacCarthy also distinguishes modifying
iN-, always transcribed with secondary stress and with
[n]before velars (e.g. ì[n]cápable, ì[n] cómpetent) from head
prefixes, which he never marks for secondary stressand
transcribes with [ŋ] before velar stops (e.g. i[ŋ]clóse ‘inclose’,
i[ŋ]cúr ‘incur’).
(26) This claim may seem to be contradicted by the fact that

the derived form commutes with the basein cases like piglet—
pig, waitress—waiter. However, in related languages where
grammatical gender distinctions are retained it is seen that
the gender of the suffix, rather than the base, invariably
determinesthe gender of the derived word (e.g. German
diminutive -chen is neuter, therefore der StuhlMASC but das
NEUT).
(27) I disagree here with Hall, who posits ‘underlying’ pword

structure for affixations based on boundstems such as
ruthless, as opposed to derived pword structure in affixations
based on free stems such as sunless. Hall’s claim that in cases
like ruthless ‘the morphological boundaries were lost but the
prosodic structure was retained’ (Hall 2001: 432) conflicts
with the fact that native speakers do recognize a suffix in
thisword and raises the question of why putative loss of
morphological structure invariably entails retentionof prosodic
structure as long as an affix can be recognized, but not
otherwise (cf the cases nostril, seduce s.9.3.2.1, and pumpkin,
s. 9.4.3).
Page 108 of 114

(28) This claim contradicts the common assertion that the

category of the stem is crucial to the prosodicorganization of
suffixed words: ‘To describe the types of junction
characteristic of a suffix, it is necessaryto distinguish two
types of theme, independent and dependent. The theme is
independent when it maystand alone, without the suffix, as a
complete word … The dependent theme is one which cannot
standalone’ (Newman 1948). The opinion expressed in this
quote is based on a misinterpretation of the fact that
noncohesive affixation overwhelmingly occurs with free stems.
For detailed discussion of this point, see Raffelsiefen (2004a,
Chapter 7).
(29) The integration of vowel-initial suffixes into the pword of

the stem is also observed in German and Dutch. It is
sometimes claimed that Dutch -achtig, German -artig are
exceptions to this generalization in that they are not
integrated into the pword of the stem (Booij 1983; van der
Hulst 1984; Peperkamp 1997). At least German -artig is not
exceptional in that it is not a suffix, but rather a (probably
even complex [[árt]STEM[ig]HEAD SUFFIX]WORD) word. The
distinct prosodic effect of English -átion in (51) versus German
-ártig results accordingly from the distinct morphosyntactic
structure (i.e. [[gút ]WORD[ártig]WORD]WORD → (gút)ω(àrtig)ω)
(30) The same generalization holds for German. For instance,

the highly productive and etymologicallyunrelated adjectival
suffixes [l1ç] -lich and [1ç] -ig neutralize after [l]: only [1ç] -ig
appears (e.g. Öl ‘oil’ +[l1ç]~ [19] → ö [lhç] (*öl.[l1ç]) ‘ölig’
‘oily’).
(31) Adverbial -ly does, however, regularly cohere after stem-

final l preceded by schwa (e.g. simp[əl]+ ly →(sim[pl]y)), which
distinguishes this suffix from the other suffixes in (54) (e.g.
o[pən.n]ess ‘(open)ωness’).Conceivably, the relevant property
here is the CV-shape of -ly. For Swedish there is clear evidence
thatCV-suffixes like -ma, -ga (where C is noncoronal) regularly
cohere whereas CVC-suffixes like -bar do notcohere.
(32) For detailed discussion of these cases see Raffelsiefen

(1999b, 2004a, b).
(33) The sort of stress shift resulting from the ‘rhythm rule’
across pword boundaries in cases like thìrteen mén differs
Page 109 of 114

from the pword-internal shift in (61a) in that stressless

syllables never reduce to schwa(√Vìetn[ə]mése, but *thìrt[ə]n
mén)
(34) This claim is consistent with the fact that bound stems
can arise historically if the base of an established affixed word
becomes obsolete as in ruthless, †ruth, doleful, †dole.
(35) Kahn’s (1976) description is misleading when he notes

that liquids and glides tolerate preceding [ŋg]. Instead these
segments require [ŋg] (rather than [ŋ]). The pronunciation
E[ŋ]lish instead of canonical E[ŋg]lish is a PU effect based on
E[ŋ]land, where [g]-loss is a boundary effect (cf.
[[Eng]STEM[land]HEAD SUFFIX]WORD → ((E[ŋ)])ωland)ω. The
variant (E[ŋg]land)ω results from HFF (cf (Lo[ŋg]Island)ω
‘Long Island’).
(36) I assume that the stress in insíghtful is due to HFF (i.e.

(insíghtful)ω).
(37) The stability of the foot on the suffix in neighborhood (as

opposed to the other suffixes in (67)) is nota HEADEDNESS
effect but rather is necessary to satisfy a constraint which
requires h to align with a left footor pword boundary (cf.
Raffelsiefen 1999b, 2004a).
(38) It is irrelevant for the point under discussion whether the

suffix is analyzed as -acy, which selects(truncated) stems, or
just -y, which causes spirantization of the stem-final
consonant). Importantly, áccuracy differs from urbánity in that
it ends in three unstressed syllables, a
(39) For instance the suffixes -ity and -acy would presumably
both be assigned to Class I based on theirassociation with
stem-allomorphy, which raises the question of how to account
for their distinct stressbehavior. In fact, there is not a single
pair of cohesive suffixes In English whose systematic
morphophono-logical properties could be described
exhaustively by some unique ranking of constraints (cf
Raffelsiefen1999b, 2004).
(40) This is not to deny the existence of pervasive

generalizations concerning the ranking of PU constraints
including the fact that PU constraints are generally
undominated for inflectional affixes in English.
Page 110 of 114

(41) Recall that head affixes by definition combine with

‘stems’.
(42) The same sort of boundary effect can be observed in

German (e.g. Kan[iːnç]en (Kanin)ω chen ‘rabbit’,where Kanin is
a bound stem and [iːnç] could not occur in simplexes).
(43) By contrast, for cohesive affixation, which is often non-

native, there are additional sources for boundstems: (a) often
a word is borrowed without borrowing its etymological base as
well (e.g. Old Frenchhaïneus, mollifier, baptiser were
borrowed into English (cf heinous, mollify, baptize), but their
respectivebases were not (i.e. Old French haïne ‘hate’, Latin
mollis ‘soft’, Greek baptein ‘to dip’)); (b) cohesive, butnot
noncohesive, suffixes can select ‘truncated’ stems in native
word formation (cf. (61b)).
(44) Similarly, historical prosodic fusion of the words in (i) in

German results from the fact that the suffix-der (probably
related to English tree) occurred only with bound stems and
was therefore no longer learnable:
((i))
Older stages Modern German
[[hólun]STEM[ter]HEAD → (hólun)ter [Holunder]WORD

SUFFIX]WORD → (Holúnder)ω ‘elder’
[[wáchsol] STEM[tr] → (wáchsol)ω → [Wr] WORD

HEAD SUFFIX]WORD (W)ω ‘juniper’
(45) The claim is that for instance fa[ŋ] less will remain fa[ŋ]
less, even if fang becomes obsolete.
(46) For evidence that schwa deletes unless it is necessary to

satisfy constraints see Raffelsiefen (1995,2000). Hooper
(1976) observes that in American English post-tonic schwa
deletes if the preceding sound is less sonorous than the
following sound (e.g. év(e)ry, féd(e)ral, also the examples in
(79)), but not vice versa (e.g. thérapy, cólony, also gerunds like
quarreling), which indicates that the constraint prohibiting
schwa is dominated by NOCODA.
Page 111 of 114

(47) Consider also the existence of boundary effects in head

prefixation in cases like reduce, prevent, deduct, etc. where
semantic affinity plays no role.
(48) The fact that the final foot is strong if it ends in a cluster
(còndescé[nd], cìrcumvé[nt], rèprehé[nd]) indicates that stress
is based on syllable structure. The last syllable in réconcìle is
open (recall that the pword-final consonant is syllabified as an
onset in the lexicon) and therefore has a non-branching rhyme.
The last syllable in còndescénd is closed, that is, with a
branching rhyme, which attracts main stress.
(49) For a different analysis, see Hall (1999:118 ff.)
(50) The example is relevant only for the pronunciation with

voiceless [s].
(51) Sensitivity of the tensing rules to the pword is not an

isolated phenomenon. Many cases of exceptional trimoraic
internal rhymes result from an Old English rule of vowel
lengthening before certain consonant
(52) The contrast in (94) thus supports the mapping of words

(and modifiers), but not head functionwords to separate
pwords. The data argue against both Selkirk’s claim that
‘pronounced in isolation,function words appear in strong form
and are indistinguishable stress-wise and vowel quality-wise
frommonosyllabic lexical category items’ (Selkirk 1995: 446)
and McCarthy’s claim that ‘the normal or defaultcondition for
a function word is to be a clitic. Alignment constraints can
impel a function word into PWD[‘pword’ R.R.] status but only
under duress’ (McCarthy 2002:138). Rather, it appears that
pword boundaries are always determined solely on the basis of
lexical properties of input forms.
(53) Ferguson mentions the verb swam only in a footnote,

perhaps, because it does not fit his characterization of the
exceptions to æ-TENSING as ‘common verbs’. Kahn’s (1976)
analysis of these verbs by way ofextrinsically ordering ablaut
is also uninsightful.
(54) Recall that æ-TENSING applies only before voiceless

fricatives.
Page 112 of 114

(55) The question of how learners know the identity of that

consonant is a separate matter. If Hale, Kissock, and Reiss
(1998: 229) were correct in claiming that the contrast
between /t/ and /d/ is neutralized intervocalically then spelling
would be the only source. However, there is some evidence
that flapping is not a neutralizing rule. Hooper (1976) notes
that flapping of t blocks schwa deletion (e.g. li[ɾər]al ‘literal’
not *li[ɾr]al) because of the insufficient sonority decrease
between [ɾ] and [r]. By contrast, d in the corresponding
environment does not block schwa deletion (e.g. federal
fe[dr]al), which indicates that the sonority structure is
different: phonetic d is still a stop. Conceivably, learners need
no recourse to written representation to know that Peter is
represented with a /t/ whereas Judith is represented with a /d/
in the lexicon.
(56) The examples are perhaps not well-chosen as they are

strongly reminiscent of suffixed words. The ratings for mellow,
which is unproblematic in this regard, are not given.
(57) cf. the related phonetics of suffix-initial n in shyness in (1).
(58) This particular difference between schwa loss in (113a)

and (113b) is confirmed by Caroline Féry (p.c.). Rialland (p.c.)
also acknowledges that the cases in (113a) and (113b) differ
and expresses doubt that there is any phonetic contrast in
(113b).
(59) Steriade (2000: 331) notes that her analysis agrees with
Rialland’s in that a characteristic property of the citation or
careful form is inherited by the schwa-less variant.
(60) Pulgram (1970: 86) characterizes French as a language

‘in which words lose both their segmental and their
suprasegmental identity within the cursus’. However, the
examples cited in support of this claim involve cohesion to
satisfy ONSET. Regular boundary effects are observed in other
contexts (cf the distinct syllabification of trois petites roues
versus trois petits trous discussed by D. Jones (1931: 60)).
Page 113 of 114

(61) Unlike the French data, Steriade’s second example, the

occurrence of a flap in capi[ɾ] alistic, but not in mili [t]aristic,
observed by Withgott (1982), is not a boundary effect (both
words are single pwords: [[capital]STEM[istic] SUFFIX]WORD →
(capitalistic)ω, [[militar]STEM[istic] SUFFIX]WORD →
(militaristic)ω Instead flapping in capitalistic is a PU effect,
which, as Pater (2000: 270) suggested and Davis (Ch. 5 above)
shows in detail, affects not low-level phonetics but (lexical)
foot structure.
Page 114 of 114

Uniformity and Contrast in the Hungarian Verbal Paradigm


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
Uniformity and Contrast in the

Hungarian Verbal Paradigm
Péter Rebrus
Miklós Törkenczy
DOI:10.1093/acprof:oso/9780199267712.003.0010

This chapter analyzes lexical allomorphy in the Hungarian
verbal paradigm, specifically the variety that is
paradigmatically motivated. The traditional strategy to
account for allomorphy with reference to a common
underlying form cannot succeed in the case of lexical
allomorphy because in lexical allomorphy the allomorphs do
not have phonological properties that permit the postulation of
a single underlying source. Therefore lexical allomorphy is
inaccessible to derivational analysis or an analysis based on
the comparison of underlying and surface forms. By contrast,
the current analysis relies on Output–Output constraints. It
shows that lexical allomorphy is due to an interaction of
paradigmatic uniformity and contrast constraints which are
relativized to morphosyntactic dimensions. Hungarian verbal
inflectional morphology is basically agglutinative, but crucially
the correspondence between morphosyntactic values and
distinct suffixes is not always one-to-one: in some forms more
than one morphosyntactic category can be expressed by one
Page 1 of 76
suffix. The chapter develops an alternative interpretation of

PAR constraints to handle this situation. This interpretation
explains an interesting phenomenon (of which the Hungarian
paradigm is an example), namely the case when lexical
allomorphy is conservative, i.e., when the lexical allomorph
must be one which is attested at some other point(s) in the
paradigm.
Keywords: paradigm uniformity, Hungarian verbal paradigm, definiteness

neutralization, anti-harmony, lexical allomorphy
Page 2 of 76
10.1 Introduction: Output–Output Constrains

In the recent phonological literature (e.g. Steriade 2000, as
well as the other studies in the present volume) paradigm has
become an important concept in expressing the phonological
relationship between words. Specifically, it has come to be
assumed that the surface shape of a word w 1 in a given
paradigm x can influence/determine the surface shape of
another word w 2 even if w 2 is not derived from w 1 in a cyclic
sense (i.e. w 1 is not embedded in w 2). Tese noncontainment-
based paradigmatic! relationships are well known from
traditional grammar and historical linguistics (e.g. Bynon 1977;
Jeffers and Lehiste 1979; Paul 1920),1 but are inaccessible for
Chomsky and Halle (1968; henceforth SPE) and derivational
frameworks originating from SPE (such as cyclic phonology or
Lexical Phonology) and also for most other models which refer
to (p.264) morphological domains (e.g. Government
Phonology, cf. Kaye 1995). These nonderi-vational
paradigmatic relationships have recently received
formalization as Output–Output correspondences in Optimality
Theory (cf. Kager 1999 b). Thus, inasmuch as the phenomenon
itself is a legitimate object of phonological research, its
analysis is a major success of (the Correspondence variant of)
Optimality Theory (McCarthy and Prince 1995) since there are
no comparable insightful formal analyses of the same
phenomenon. At the same time, it has to be stressed that
nonderivational paradigmatic relationships are not necessarily
in principle optimality theoretic: any other theory that has a
formal means of comparing surface forms may access these
relationships.
In the various treatments of the phenomenon it is assumed

that there are two competing ‘forces’ active within a paradigm
which may determine the phonological/phonetic properties of
a paradigm-member: one requiring the identity of certain
properties of stems/affixes (expressed in the form of paradigm
uniformity constraints)2 and another requiring contrast (non-
identity) between members of a paradigm (expressed in the
form of paradigmatic contrast constraints). Note that this
presupposes (contra previous cyclic analyses of the
phonological relatedness of morphological related forms) that
the notions ‘paradigm’, ‘paradigm member’, and ‘cell’ (a point
in the paradigm where the values of all its dimensions are
specified) have theoretical status.
Page 3 of 76
Paradigm uniformity can be stated in a general form as (1)

(based on Steriade 2000: 313).
(1) PAR:
All surface realizations of µ, where µ is the morpheme
shared by members of paradigm x, must be identical.
PAR constraints are calculated over a substring of a string

realizing a member of the paradigm, i.e. they require that
paradigm members should be partially identical. The general
constraint requiring paradigmatic contrast can be stated as in
(2).
(2) CON:
The surface realizations of morphologically distinct
members m i … m n of a paradigm x must be
phonetically distinct.
CON requires that different members of a paradigm must be

phonetically different. As opposed to PAR, it is calculated over
the whole string realizing the member of the paradigm. PAR
and CON constraints are assumed to be valid within a ‘space’
defined by the paradigm and taken to be inoperative outside
that space. For example, homophony is typically not avoided
between affixed forms of different stems. For instance, in
Hungarian, tense/mood and Person /number homophony can
occur when different stems are involved: ér ‘s/he touches sg.
(3Sg. indef. Pres. Indic.)’ contrasts with ért ‘s/he understands
sg. (3Sg.indef.Pres.Indic.)’, but ér-tek ‘you (Pl.) touch sg.
(2Pl.indef.Pres. Indic.)’ is homophonous with ért-ek ‘I
understand sg. (1Sg.indef.Pres.Indic.)’. Note, however, that
the exact definition of paradigm as the space within which
these constraints (p.265) hold is far from being obvious. In
the current literature, working definitions range from the
narrow (‘words sharing the same stem’, Kenstowicz, Ch.7
above) to the rather inclusive (‘A paradigm is a set of words
sharing a morpheme (e.g.{bomb, bomb-ing, bomb-ard, …}) or
a set of phrases sharing a word (e.g. {bomb, the bomb, …}).’
Steriade 2000). The definition of the space within which
correspondence relations hold is a central problem in the
research of paradigm uniformity and contrast. It is possible
that this space is different in different languages/dialects. The
problem of how paradigm or paradigmatic space can be
defined (in OT) is beyond the scope of this chapter;
‘operationally’, we shall use it in the following sense: paradigm
Page 4 of 76
x is a set of words sharing (i) the same value in some

morphosyntactic dimension, or (ii) the same stem—this is close
to the definition which refers to the set of words sharing a
morpheme.
10.2 Relativized Paradigmatic Uniformity and Contrast in

the Hungarian Verbal Paradigm
Constraints of paradigm uniformity and paradigmatic contrast
are usually invoked to account for some cases of allomorphy
when one allomorph occurs unexpectedly instead of the
allomorph that phonological and/or morphological regularities
would require. In this chapter we shall discuss two types taken
from Hungarian verbal inflection, each of which exemplifies a
violation of a type of regularity. In one case a morphological
regularity is violated, in the other, a phonological one.
In the first case, in the 1Sg.indef.Past and the 1Pl.def.Cond, a

Person /number suffix occurs which is the ‘wrong’ one because
it is the same as the suffix with the opposite value for
definiteness. The morphologically expected form is avoided
and the definite-ness contrast is neutralized. This is shown in
(3) where (3a) and (3c) list 1Sg.Pres.Indic and 1Pl.Pres.Indic
forms, respectively, for the sake of contrast, and (3b) and (3d)
list the forms in which the unexpected suffix alternants occur
(the Person /number suffixes are emboldened and the
unexpected suffix alternants are underlined):3
Page 5 of 76
(3) a. 1Sg.Pres.Indic b. 1Sg.Past
Stem Gloss indefinite definite indef. def
kap ‘get’ kapok kapom kaptam kaptam
nyom ‘push’ nyomok nyomom nyomtam nyomtam
rúg ‘kick’ rúgok rúgom rúgtam rúgtam
csíp ‘pinch’ csípek csípem csípte m csíptem
néz ‘look’ nézek nézem nézte m néztem
tör ‘break’ török töröm törte m törtem
c. IPl.Pres.Indic d. IPl.Cond
Stem Gloss indef. def. indef. def.
kap ‘get’ kapunk kapjuk kapnánk kapná nk
nyom ‘push’ nyomunk nyomjuk nyomnánk nyomná nk
rúg ‘kick’ rúgunk rúgjuk rúgnánk rúgná nk
csíp ‘pinch’ csípünk csípjük csípnénk csípné nk
néz ‘look’ nézünk nézzük néznénk nézné nk
tör ‘break’ törünk törjük törnénk törné nk
Page 6 of 76
(p.266) In the second case we shall examine, in

1Sg.indef.Cond, the conditional suffix, which normally
alternates as a result of backness harmony, shows up in its
phonologically unexpected, ‘wrong’ front alternant after back-
vowel stems. This is shown in (4) where (4a) lists 1Sg.def.Cond
forms for the sake of contrast, and (4b) lists the forms in which
the unexpected suffix alternants occur (the conditional suffixes
are emboldened and the unexpected suffix alternants are
underlined).
Page 7 of 76
(4) a. ISg.def.Cond b. ISg.indef.Cond.
front stem back stem front stem back stem
csípném rúgnám csípnék rúg né k
törném nyomnám törnék nyom né k
Page 8 of 76
We will show that the unexpected behavior of the suffixes

illustrated in (3) and (4) above can be explained with reference
to the interplay between paradigmatic uniformity and contrast
(a) if we interpret PAR and CON as constraints that can be
relativized to the various morphosyntactic dimensions of the
paradigm, and (b) if these constraints that are relativized to
different dimensions are ranked with respect to one another.
We define relativized PAR (PAR(D )) and CON (CON(D )) as in
(5) and (6).
(5) PAR relativized to the morphosyntactic dimension

PAR(D)
All surface realizations of µ (where µ is the morpheme
shared by members of paradigm x) that have the same
value within the morphosyntactic dimension D, must be
identical.
(6) CON relativized to the morphosyntactic dimension

CON(D)
A form realizing some value of a morphosyntactic
dimension D of paradigm x must be phonetically distinct
from forms realizing other values of D.
As pointed out above, the domain within which PAR

constraints hold are substrings. In the case of PAR(D), the
substring whose identity is examined is uniquely identified by
the setting for the morphosyntactic dimension, i.e. if a
morphosyntactic dimension is specified, it follows that its
exponent (the affix) is the substring whose identity (p.267) is
required.4 For CON(D) (as for all CON constraints) the
selection of the domain of application is not an issue: it is the
whole string realizing a member of the paradigm, since the
essence of a CON constraint is the avoidance of complete
identity between different members of a paradigm.
We consider paradigmatic uniformity and contrast as

essentially co-determining the shape of a paradigm in crucial
ways: it is never just one or the other that is involved. They are
also constraints of the same kind, which is expressed in OT
terms by their formulation as Output–Output constraints. PAR
and CON constraints may be in conflict, as is to be expected
between the two conflicting ‘forces’ within the paradigm. In
addition to PAR constraints being in conflict with CON
constraints, PAR constraints may be in conflict with other PAR
constraints and CON constraints may be in conflict with other
Page 9 of 76
CON constraints. In this chapter we shall show that all three

types of interaction are involved in the Hungarian verbal
paradigm, specifically, that the conflict between CON
constraints relativized to different morphosyntactic
dimensions is constrained by their interaction with relativized
PAR constraints. We also propose a new PAR constraint
WEAKPAR, which militates against lexical allomorphy within a
specific cell of the paradigm, and is crucially involved in
choosing between different strategies of maintaining
paradigmatic contrast. We shall develop a hierarchy of these
PAR and CON constraints (and some phonological constraints)
that accounts for the characteristics of the Hungarian verbal
paradigm focusing on the properties illustrated in (3) and (4).
The rest of the chapter is organized as follows. In s. 10.3, we

describe some characteristics of the verbal inflectional
paradigm in Hungarian which provide the background
necessary to understand the phenomena we focus on. We also
make a distinction between two kinds of allomorphy:
‘phonological’ and ‘lexical’. Phonological allomorphy is
allomorphy required by some general phonological regularity
in a language (in OT terms it is the state of affairs when a
phonological constraint dominates a paradigmatic uniformity
constraint). The main topic of the chapter is the other kind of
allomorphy, which cannot be described with constraint
interaction of this kind (i.e. ‘lexical allomorphy’). In ss. 10.4
and 10.5, we give an OT analysis of the two cases of lexical
allomorphy we illustrated above with reference to paradigm
uniformity/contrast: Definiteness Neutralization (s. 10.4) and
Anti-Harmony (s. 10.5). In s. 10.6 we draw some conclusions
and identify some problems for further research. This is
followed by an Appendix showing the complete inflectional
paradigm of the Hungarian verb. Throughout the chapter we
shall use the stems akar ‘want’ and teker ‘turn’ as typical
examples for the behavior of back-vowel stems and front-vowel
stems, respectively. Naturally, the phenomena we shall discuss
are characteristic of the verbal paradigm and are not limited
to particular stems.
(p.268) 10.3 Hungarian Data
Page 10 of 76
In this section we summarize the relevant data concerning the

verbal inflectional paradigm in Hungarian. The following
description is not exhaustive: here only those categories and
generalizations are introduced that are important for our
purposes. For more detailed descriptions the reader is
referred to Vago (1980), Kenesei, Vago, and Fenyvesi (1998),
Siptár and Törkenczy (2000), Rebrus (2000a, b) among others.
In s. 10.3.1 those regularities are dealt with that modify the
‘ideal’ agglutinative pattern: namely the fusional and templatic
properties of the verbal inflectional paradigm. In s. 10.3.2 we
briefly introduce data which are exceptions to general
phonological regularities, namely those types of allomorphy in
which the allomorphs are phonologically unrelated (which we
call lexical allomorphy).
10.3.1 Verbal inflectional categories

Hungarian is basically an agglutinative language with
concatenative morphology. There are, however, a number of
deviations from this basic pattern: in a single verb-form the
values of the morphosyntactic categories are not always
expressed by the concatenation of distinct affixes that are
associated to their categories in a one-to-one fashion. In
present-day Hungarian the person and number of the subject
are expressed by fusional morphemes: they present indivisible
suffixes (for the actual forms of the suffixes, see the complete
verbal paradigm in the Appendix). In other words, there is a
paradigmatic dimension called PERSON /NUMBER with six
values (cf. (9)).
Tense and mood are expressed by mutually exclusive suffixes.

This gives us the TENSE /MOOD paradigmatic dimension with
four values (cf. (9)). A more ‘exotic’ category is Definiteness,
which is sensitive to the definiteness of the object. This
dimension has two values: definite and indefinite.
The expression of DEFINITENESS is highly complex in

Hungarian. Basically, there are three ways of marking the
definiteness of the object on the verb, as shown in (7). It can
be marked by a Person /number marker only, or by a
definiteness marker only, or by a definiteness marker and a
Person /number suffix: (i) in 1Sg and in 2Sg the Person/
number (p.269) suffix is different in the definite and the
indefinite; (ii) in 3Sg and in 2Pl defin-iteness is expressed by
the definiteness marker -ja/-já; (iii) in 1Pl and in 3Pl the
Page 11 of 76
definite–indefinite opposition is expressed by both the

different Person /number suffix and the definiteness marker (-j
or -já).
Page 12 of 76
(7) Definiteness (in present indicative) is expressed with
INDEFINITE DEFINITE
i. the Person /number akar-ok akar-om ‘I want sg./it’

suffix, e.g.
akar-sz akar-od you (Sg.) want Sg./

it’
ii. a definite marker, akar akar-ja ‘s/he wants sg./it’

e.g.
akar-tok akar-já-tok ‘ you (Pl.) want sg./

it’
iii. both, e.g. akar-unk akar-j-uk ‘we want sg./it’
akar-nak akar-já-k ‘they want sg./it’
Page 13 of 76
There is a futher constraint which in certain verb-forms

disrupts the agglutinative pattern. The definiteness marker -j/-
ja/-já cannot occur in a verb-form marked by a non-zero tense/
mood morpheme. Two simple cases of this templatic constraint
are exemplified in (8). The definiteness marker does not
appear in definite past, definite imperative/subjunctive and
definite conditional forms.
Page 14 of 76
(8) Template: in one word-form tense/mood and definiteness markers are mutually exclusive, e.g. (relevant values
emboldened)
STEM TENSE/MOOD DEFINITENESS PERSON/

NUMBER
a. 1Pl.indef.Ind akar – – unk
1Pl.indef.Past akar t – unk
1Pl.def.Ind akar – j uk
1Pl.def.Past akar t – uk
b. 2/3Pl.indef.Ind akar – – tok/nak
2/3Pl.indef.Cond akar ná – tok/nak
2/3Pl.def.Ind akar – já tok/k
2/3Pl.def.Cond akar ná – tok/k
Page 15 of 76
To sum up, the generalizations constraining the inflectional

morphology of the verb are the following. There are three
paradigmatic dimensions: TENSE/MOOD, DEFINITENESS and
PERSON /NUMBER. Each dimension has several values; a
finite verb-form must take a value in each dimension. The
dimensions and their values are listed in (9). Verb-forms show
morphological contrasts when they have different values in
some dimension.
(9) Three paradigmatic dimensions and their values
dimensions: values:
PERSON/ 1SG, 2SG, 3SG, 1PL, 2PL, 3PL

NUMBER:
TENSE/MOOD: PRESENT/INDICATIVE, PAST/IND,

PRES/SUBJUNCTIVE, PRES/
CONDITIONAL
DEFINITENESS: INDEFINITE, DEFINITE
Suffixes expressing a value of a dimension do not combine

freely, but make selection requirements on other suffixes, and
one verb-form cannot contain more than two inflectional
suffixes. This latter templatic property bans the co-occurrence
of the TENSE/MOOD and the DEFINITENESS markers in the
same verb-form, see (8). Thus, the complete structure of an
inflected verb contains two template positions for inflectional
suffixes: the first position is for TENSE/MOOD and
DEFINITENESS, the second position is for PERSON /
NUMBER, as shown in (10).
(p.270)
(10) The morphological structure of an inflected verb:

STEM + TENSE/MOOD/DEFINITENESS + PERSON /
NUMBER MARKERS
10.3.2 Lexical alternations

The alternation of morphemes in Hungarian is usually driven
by phonological regularities pervasive in the whole system (cf.
Siptár and Törkenczy 2000). However, there are some cases of
what we shall call ‘lexical allomorphy’ which we informally
define as (11).
Page 16 of 76
(11) Lexical allomorphy

The allomorphs are phonologically ‘unrelated’, i.e. the
relationship between the allomorphs is not a regular/
general phonological relationship in the language; the
allomorphs have no generalizable common underlying
form; the alternation may or may not be conditioned by
the phonological environment.
The Hungarian verbal paradigm (see the Appendix) shows the

following types of lexical allomorphy relevant to our topic.
(a) Phonologically motivated lexical allomorphy. In

2Sg.indef.Ind the suffix is -sz [s] after a stem-final nonsibilant,
and -ol/el/öl after a sibilant consonant (e.g. akar-sz ‘you(Sg)
want sg.’, hoz-ol ‘you(Sg) bring sg.’). It is clear that the sz ~ Vl
alternation is not phonological; but at the same time it is
connected with the phonological environment. The result is
that sibilant clusters are avoided at inflectional morpheme
boundaries in the language. Note that sibilant clusters do not
exist monomorphemically either.
(b) Paradigmatically motivated lexical allomorphy. In this case

an irregular affix alternant occurs instead of the regular
(paradigmatically uniform) one. This is due to a process which
tries to avoid homophony between forms that have different
morphosyn-tactic values. The way in which homophony is
avoided is systematic. Here we introduce two subtypes which
we discuss in detail later.
i. The first one is Definiteness Neutralization. In this case the

Person /number marker is irregularly identical in indefinite
and definite forms (e.g. υág-om/υág-j-am ‘I cut it (Indicative/
Subjunctive)’ ≠ υág-ok/υág-j-ak ‘I cut something (Ind./Subj.)’,
but υág-t-am ‘I cut it (Past)’ = υág-t-am ‘I cut something
(Past)’). This makes the relevant verb-forms identical in
definiteness.5 Two suffix alternations are involved, which we
shall discuss in s. 10.4.
ii. In the second type of paradigmatically motivated lexical

allomorphy, the allomorphs are phonologically related, but in
an irregular way. They are irregular harmonic alternants;
vowel harmony is irregularly blocked in one allomorph (e.g.
lep-ne ‘s/he would surprise sb.’ and lep-né-k ‘I would surprise
sb.’, but lop-na ‘s/he would steal’ and (p.271) lop-né-k ‘I
Page 17 of 76
would surprise sb.’. Anti-Harmony,where the irregular

alternation affects only the harmonizing vowel, will be
discussed in s. 10.5.6
Though these lexical allomorphies are limited to isolated forms

in the paradigm, the claim of this chapter is that they are not
completely arbitrary. In fact, all the alternations discussed
above are motivated to some extent. Type (a) is phonologically
motivated and is unrelated to paradigm uniformity/contrast
(but will be crucially relevant to the discussion in s. 10.5.3).
Type (b) however, does have a motivation rooted in paradigm
uniformity/contrast. Specifically, the motivation in all these
cases (i), (ii), is that the relevant forms are irregular in order
to avoid homophony with other forms. In the following sections
we show an optimality theoretic analysis for the last two types
of lexical allomorphy: Definiteness Neutralization (s. 10.4) and
Anti-Harmony (s. 10.5).
10.4 Definiteness Neutralization

10.4.1 Contrast and allomorphy
The first case of lexical allomorphy we shall deal with is when
a definite and an indefinite form are neutralized. This is shown
in (l2 a) for iSg forms where there is an indefinite–definite
contrast which is lost in the Past. Regularly, the Person /
number marker is -k in the indefinite and -m in the definite
(see the first three columns in (12 a)), but in the Past
indefinite, an -m allomorph is unexpectedly substituted for the
Person / number suffix7 (see the last column in (12), where the
lexical allomorph is underlined). Similarly, as can be seen in
(12b) for 1Pl cases, there is a definite–indefinite contrast
expressed by the suffixes -uk and -unk, respectively, which is
lost in the conditional (see the third column). Note that the
loss of the high vowel and the lengthening of the low vowel in
1Pl.Cond forms are due to general phonological processes:
akar+na+unk → akarnánk.
Page 18 of 76
(12) >Lexical allomorphy in 1Sg.indef.Past and in 1Pl.def.Cond
akar
‘want’ INDICATIVE SUBJUNCTIVE CONDITIONAL PAST
a. 1Sg INDEFINITE akar-ok akar-j-ak akar-né-k akar-t-a m
DEFINITE akar-om akar-j-am akar-ná-m akar-t-am
b. 1Pl INDEFINITE akar-unk akar-j-unk akar-ná-nk akar-t-unk
DEFINITE akar-j-uk akar-j-uk akar-ná- nk akar-t-uk
Page 19 of 76
Table (13) shows that the nonexisting regular 1Sg indefinite

past form *akar-t-ak would coincide with another form in the
paradigm, namely the 3Pl. We claim that the unexpected
occurrence of the allomorph -m in the indefinite form akar-t-
am is related (p.272) to this coincidence. It is important to
note that the irregular 1Sg indefinite form akar-t-am also
coincides with another form in the paradigm, the regular 1Sg
definite form. The conditional case (13b) is very similar to this.
The nonexisting regular 1Pl.def.Cond form *akar-ná-k would
be the same as another existing form in the paradigm, the 3Pl
definite form akar-ná-k.8 This homophony is avoided by the
existing form akar-ná-nk in 1Pl definite, but this form is the
same as the indefinite form. This is shown in table (13), where
the Person /number markers are emboldened, the irregular/
lexical alternants are underlined, paradigmatic contrasts are
shown with arrows, and homophony is indicated with vertical
equation marks.
Page 20 of 76
(13) Contrast in Person/number and homophony in definiteness
a.PAST b. CONDITIONAL
1Sg 3Pl 1Pl 3Pl
*akar-t-ak
indef akar-t-am ↔ akar-t-ak akar-ná-nk
‖‖ ‖‖
def akar-t-am akar-ná-nk ↔ akar-ná-k
*akar-ná-k
Page 21 of 76
The main properties of Definiteness Neutralization are: (i) in

certain Person /number and tense/mood values, the definite
and indefinite Person /number markers are irregularly the
same, the result of which is that definiteness is neutralized in
these forms; (ii) a homophony in the paradigm is avoided by
producing another instance of homophony, i.e. the definiteness
contrast is suspended in order to avoid the identity of different
Person /number forms; (iii) the direction of the neutralization
is the opposite in the two cases: the neutralization form in the
first case is the definite, while in the second case it is the
indefinite. This means that this process is not a simple case
when the target of the neutralization is the morphologically
unmarked value (we take it to be the indefinite, which is
supported by the fact that indefiniteness is never expressed by
a distinct marker in the entire paradigm). These properties are
summarized in (14) below.
(14) Properties of Definiteness Neutralization

i. definiteness neutralization caused by lexical
allomorphy
ii. homophony is avoided in one dimension, but
produced in another
iii. the direction of the process is independent of
morphological markedness
(14iii) implies that the motivation of this case of lexical

allomorphy is not the markedness of values within a
dimension. We claim, however, that Definiteness
Neutralization is due to the hierarchy of paradigmatic
dimensions. This is given an OT interpretation in the next
subsection.
(p.273) 10.4.2 Optimality analysis of Definiteness Neutralization

The state of affairs described above, i.e. when homophony is
avoided at one point in the paradigm while the form selected
actually remains homophonous with a form at another point in
the same paradigm, may be given an OT interpretation using
CON constraints that are relativized to two different
morphosyntactic dimensions.
(15) CON(PERSON /NUMBER)

(16) CON(DEFINITENESS)
Page 22 of 76
In addition to these CON constraints, relativized PAR

constraints are involved too. The crucial PAR constraint is
relativized to the morphosyntactic dimension PERSON /
NUMBER, see (5).
(17) PAR(PERSON / NUMBER)
Constraint (17) requires that all surface realizations of the

affix shared by all the members of the same value within the
morphosyntactic dimension PERSON /NUMBER should be
identical.
Before we discuss the interaction of these constraints, we have

to point out a crucial problem in the interpretation of
relativized PAR constraints. Given the definition of PAR(D) as
in (5), (17) means that if the Person /number value is kept
constant while the values of the other dimensions of the
paradigm change, the resulting forms should share an
invariant affix. The problem is that for certain Person /number
values in the Hungarian verbal paradigm this requirement is
not met even if we disregard lexical allomorphy: there is no
invariant affix for some Person /number values. The 1Sg forms
are a case in point—consider (12 a). Given (17) we expect that
the 1Sg forms should share an affix that is the same in all the
forms having different values for tense/ mood and
definiteness. Constraint (12a) shows clearly that there is no
such shared affix (sometimes the 1Sg suffix is -k, sometimes it
is -m). The reason why PAR(D) as defined in (5) fails is that the
Hungarian paradigm is not purely agglutinative. As we have
pointed out in s. 10.3.1, in some Person /number forms,
definiteness is not expressed by a separate suffix, but by the
choice of the Person /number suffix, see (7). Therefore there is
no invariant suffix for some Person /number values. The
description of this ‘portmanteau’ behavior requires a slightly
modified version of PAR, specifically PAR(PERSON /NUMBER),
which does not enforce strict identity (invariance), but only
membership of a restricted set. The modified constraint can be
stated as at (5′).
(5′) PAR relativized to a morphosyntactic dimension

(revised version)
PAR( D)
All surface realizations of µ, where µ is the morpheme
shared by members of paradigm x that have the same
Page 23 of 76
value within the morphosyntactic dimension D, must be

a member of a restricted set assigned to this value.
Constraint (5′) is violated if, for any value of a morphosyntactic

dimension, the allo-morph in a particular form of the paradigm
is not a member of the set assigned to (p.274) that value. In
the PERSON /NUMBER dimension, the relevant P AR(PERSON
/NUMBER) constraint requires that the Person /number suffix
should be one which is a member of a set of suffixes assigned
to a Person /number value (1Sg, 2Sg, etc.). In the particular
values relevant to the discussion (1Sg and 1Pl), P AR(PERSON
/NUMBER) is satisfied if (a) in a 1Sg form a Person /number
suffix is a member of the set {-k, -m}, (b) in a 1Pl form a
Person /number suffix is a member of the set {-unk, -uk}.
Otherwise, a (candidate) form violates PAR(PERSON /
NUMBER): e.g. akar-od (an existing form: akarod ‘want 2Sg.
def.Ind’) and akar-op (a non-existing form) both violate both
constraints.
The definition of uniformity as choice from a restricted set has

an additional advantage. It can explain the ‘conservative’
strategy in the avoidance of homophony, i.e. when the
‘avoidance form’ (i.e. the form that is used to avoid
homophony and maintain contrast within the paradigm at a
specific point) is actually an existing form in the paradigm
(which is the Hungarian state of affairs). If PAR(D) is ranked
sufficiently high and is defined as adherence to a specific set
of allomorphs assigned to a value of D, then truly innovative
candidates, i.e. ones that are not homophonous with any
member of the paradigm (and thus do not contain any of the
allomorphs from the set), would fare worse than the
conservative ones. This will be shown for the 1Sg.indef. Past
and the 1Pl.def.Cond in the tableaux later.
According to (5′), PAR( D) limits 1Sg suffixes to a two-member

set, the doublet {-k, -m}, but permits arbitrary choice between
the two members. However, the choice may be non-arbitrary.
Thus, in Hungarian, the Person /number marker is
systematically one member of the set in the definite, and the
other in the indefinite, i.e. the different members of the set
systematically belong to different specific values of another
dimension. This cannot be expressed with a constraint
relativized to a single dimension. Therefore, it is necessary to
generalize PAR( D) to refer to the direct (Cartesian) product of
more than one dimension. Specifically in Hungarian, we need
Page 24 of 76
a constraint that refers to Person /number and definiteness

together. PAR(PERSON / NUMBER × DEFINITENESS)
requires that all the forms that have a specific value for
Person /number and a specific value for definiteness should
share an affix chosen from a restricted set. In our case the
relevant sets are singletons: e.g. for the pair of values <(1Sg,
indefinite> the set is {-k} and for the pair of values <1Sg,
definite> it is {-m}. In essence, what the constraint says in
these particular cases is simply that the 1Sg definite marker is
-m and the 1Sg indefinite marker is -k, i.e. that in these cases
(in)definiteness has no marker which is independent of the
Person /number affix (see table (7)).
The effect of the two constraints mentioned above is shown in

(18) for a regular 1Sg.indef.Ind form. P AR(PERSON /
NUMBER) penalizes only the totally innovative form akar-op
(which contains a hypothetical Person /number marker -p, see
(18c)). The regular (and grammatical) indefinite form akar-ok
(18a) wins due to the constraint PAR(PERSON /NUMBER ×
DEFINITENESS discussed above, because both (18a) and
(18b) satisfy PAR(PERSON /NUMBER) as defined in (5′).
Returning to Definiteness Neutralization, we assume that (15)

is crucially higher up in the constraint hierarchy than (16)
since in the Hungarian verbal paradigm it is more important to
maintain the PERSON /NUMBER contrast than the
DEFINITENESS (p.275) contrast.10 This is shown for the
1Sg.indef.Past forms in tableau (19) and for the 1Pl. def.Cond
in tableau (22).
(18) Choice of the Person /number suffix in the regular

1Sg.indef.Ind9
akar + PAR(PERSON PAR(PERSON /

1Sg.indef.Ind ‘I /NUMBER) NUMBER × ×
want sg.’ DEFINITENESS)
a. ☞ akar-ok
b.akar-om !*
(1Sg.indef: -ok, -j-
ak, -né-k)
c.akar-op !* *
Page 25 of 76
(1Sg.indef: - (1Sg.indef: -ok, -j-

ok, -j-ak, -né- ak, -né-k)
k 1Sg.def: -
om, -j-am, -
ná-m)
(19) Definiteness Neutralization in 1Sg.indef.Past (cf

(13a))
Page 26 of 76
akar+1Sg.indef.Past CON(PERSON/ PAR(PERSON CON (DEFINITENESS) PAR (PERSON

‘I wanted sg.’ NUMBER) NUMBER) NUMBER ×
× DEFINITENESS)
a. akar-t-ak !*
(=3Pl ‘they wanted
sg.’)
b. ☞ akar-t-am * *
(=def ‘I wanted it‘) (1Sg.indef:
-ok,-j-ak, -né-k)
c.akar-t-ap !* *
(1Sg.indef: 1Sg.indef:
-ok,-j-ak, -né-k 1Sg.def: -ok, -j-ak, -né-k)
-om,-j-am, -ná-m)
Page 27 of 76
(p.276) Tableau (19) shows that the proper ranking of the

constraints gives the optimal output form. Form (19 a) is the
expected ‘regular’ form, but it is homophonous with the 3Pl
form, therefore violates the dominant CON(PERSON /
NUMBER). Form (19b) contains the definite Person /number
marker, so it is homophonous with the definite form, violating
CON(DEFINITENESS). It is also violates P AR(PERSON /
NUMBER × DEFINITENESS) because the set assigned to the
pair of values 1Sg, indefinite>is the singleton {-k}. Note that
neither form (19 a) nor form (19 b) violates the relativized P
AR(PERSON /NUMBER) constraint since they both contain
suffixes that are members of the set assigned to 1Sg value of
Person/ number. If the form is suffixed with a hypothetical non-
existing Person /number marker -p (19c), neither of the two
CON constraints is violated, but the innovative form (c) does
violate P AR(PERSON /NUMBER) and P AR(PERSON /
NUMBER × DEFINITENESS), because the marker -p never
occurs in the set of 1Sg forms. Hence the constraint P
AR(PERSON /NUMBER) cannot be lower in the hierarchy than
CON(DEFINITENESS) since otherwise definiteness
neutralization would not be possible (and the innovative
candidate (c) would win). CON(PERSON /NUMBER) may be
unranked with respect to P AR(PERSON /NUMBER). Similarly,
if a candidate were suffixed with a hypothetical tense/mood
marker -p instead of the regular -t (*akar-p-ak), the form would
satisfy all the constraints in (19), but would violate
PAR(TENSE/MOOD), which is assumed to be higher than
CON(DEFINITENESS). Thus the following ranking of
constraints has to be assumed:
(20) PAR(TENSE/MOOD)
CON (PERSON / NUMBER) » CON

(DEFINITENESS)
PAR (PERSON / NUMBER) PAR (PERSON/UMBER

× DEFINITENESS)
The situation is very similar in the 1Pl case. The regular 1Pl
forms can be analyzed as the same way as the 1Sg forms in
(18).
(21) Choice of the Person /number suffix in 1Pl.def.Past
Page 28 of 76
akar + PAR(PERSON/PAR(PERSON /
1Pl.def.Past NUMBER) NUMBER × ×
‘We wanted it’ DEFINITENESS)
a. ☞ akar-t-uk
b. akar-t-unk !*
(1Pl.def: -j-uk, -j-uk, -
ná-nk)
c. akar-t-up !* !*
(1Pl.indef: - (1Pl.def: -j-uk, -j-uk, -
unk, -j-unk, - ná-nk)
ná-nk
1Pl.def: -j-
uk, -j-uk, -
ná-nk)
(p.277) Definiteness Neutralization in 1Pl indefinite

Conditional is shown in (22). Form (22 a) is homophonous with
the 3Pl.def.Cond form, form (22 b) is homophonous with the
1Pl.def.Cond form. The hypothetical non-existing individual
allomorph -p (form (22c)) is ruled out by P AR(PERSON/
NUMBER) in the same way as in (18), (19), and (21).
(22) Definiteness Neutralization in 1Pl.def.Cond (cf.

(13b))
Page 29 of 76
akar+1Pl.def. Cond CON(PERSON / PAR(PERSON / CON(DEFINITE - PAR(PERSON /

‘we would want it’ NUMBER) NUMBER) NESS) NUMBER × ×
DEFINITENESS)
a. akar-ná-k !*
(=3Pl ‘they would
want it’)
b.☞akar-ná-nk * *
(=indef: ‘we would (1Pl.def: see (21))
want sg.’)
c. akar-ná-p !* ! * (1Pl.def: see (21))

(1Pl.indef see (21))
1Pl.def see (21))
Page 30 of 76
Note that the forms (22 a) and (22b) are not completely
concatenative: in both cases the initial high vowel of the
Person /number suffix is deleted for phonological reasons
(akar+na+unk→ akarnánk and akar+na+uk → akarnák). It is
clear that this deletion is due to a higher-ranking phonological
constraint, which bans this type of hiatus at morphological
boundaries (i.e. *akar-ná-uk, *akar-ná-unk) and thus enforces
phonological allomorphy. (Note that the form *akar-ná-uk
satisfies all the paradigmatic constraints in (22), i.e. it does
not show homophony and is paradigmatically uniform.
Therefore the phonological constraint must be higher up than
CON(Definiteness).)11
Another possibility would be if the 1Pl.def.Cond form

contained the definiteness marker -j (*akar-ná-j_-uk), as it does
in the Present Indicative (akar-j_-uk; for the morphological
structure see (8)). In this case, there would be neither
homophony, nor (p.278) hiatus, but the actual form would
violate a morphological regularity which bans the occurrence
of an overt tense/mood and definiteness marker in one form
(for this tem-platic property, see (8)). We attribute this to a
higher-ranked morphological constraint which is never
violated in the language. To sum up, it is plausible to assume
that the constraints that are responsible for morphological and
phonological regularities dominate paradigm uniformity/
contrast constraints. This is shown in (23) for the relevant
constraint families.
(23) MORPH, PHON » PAR(D), CON(D)
10.5 Anti-Harmony
The second type of lexical allomorphy we discuss is a case
when a phonological regularity is in interaction with
paradigmatic uniformity/contrast. The phonological regularity
involved is backness harmony, which we briefly describe in the
following section.
10.5.1 Backness harmony and neutral vowels

Hungarian vowel harmony is well known in the literature (see
Siptár and Törkenczy 2000 and references cited therein); we
shall not go into details and discuss only as much as is
necessary for the analysis of lexical allomorphy.
The Hungarian vowel inventory and the vowel letters used in

spelling are set out at (24).
Page 31 of 76
(24) Vowel inventory
Page 32 of 76
FRONT BACK
UNROUNDED ROUNDED UNROUNDED ROUNDED
[i] [y]<ü > [u] SHORT
[iː]<í> [yː]<ű> [uː]<ú> LONG
[ø]<ö> [o]<o > SHORT
[eː]<é> [øː]<ő> [oː]<ó> LONG
[ε]<e> [Ↄ]<a > SHORT
[aː]<á> LONG
Page 33 of 76
Note that the accent stands for length without quality

difference except in the case a–á, e–é; the difference between
the latter pair will be important in the analysis.
Backness harmony means that the suffix vowel has to agree

with the stem vowels in backness. Naturally, the phenomenon
is more complex, but here we shall abstract away from stem-
internal harmony, suffix harmony after mixed stems, and
irregularly harmonizing stem-classes. Normally, suffixes have
more than one variant and harmonize in backness. This is
shown in (25) for some verbal suffixes.
(p.279)
Page 34 of 76
(25) Backness harmony in the suffix
back vowel front vowel
high u~ü akar-unk ‘want’ teker-ünk ‘turn’ iPl.indef.Ind
mid o ~ e (~ ö) akar-tok teker-tek 2Pl.indef.Ind
mid-low a~e akar-na teker-ne 3Sg.indef.Cond
low ~ mid á~é akar-ná-d teker-né-d 2Sg.def.Cond
Page 35 of 76
Most of the vowels in the inventory are harmonic, but some

are neutral. Neutral vowels are transparent to backness
harmony. Te neutral vowels are the front non-low unrounded
vowels i, í, and é. Only neutral vowels can occur in suffixes
that are non-alternating (i.e. have just one form).12 It is
important to note that all verbal inflectional suffixes are
alternating.
Page 36 of 76
(26) Non-alternating inflectional suffixes have neutral vowels
ALTERNATING v. NON-ALTERNATING suffixes
-ban/-ben ‘in’ -ért ‘for’ bor-ért ‘wine + for’
-nál/-nél ‘at’ -ig ‘up to’ Ház-ig ‘house + up to’
-tól /-től ‘from’ -i ‘plural’ láb-a-i ‘his feet’
-on/-en/-ön ‘on’ -ít ‘denom. verb vastag-ít ‘thick+en’

forming’
etc.
Page 37 of 76
Hungarian backness harmony can be and has been described

with OT constraints (see e.g. Ringen and Vago 1998).
Naturally, here we shall not present an analysis, but simply
give two informal constraints which will be sufficient for our
purposes. The first one (27 a) governs suffix harmony of non-
neutral vowels, and the second one (27b) handles suffixes
containing neutral vowels. As long–short pairs behave in the
same way except for e–é, we have only indicated length
difference for e–é in (27) (since e is non-neutral, but é is
neutral).
(27) Constraints for backness harmony

a. HAR:
If the stem-vowel(s) is/are front, then the affix-
vowel is front, i.e. *…e/ö/ü/é/i … + … a/o/u …
If the stem-vowel(s) is/are back, then the affix
vowel is back or neutral, i.e. *… a/o/u … + … e/ö/
ü…
b. *NEUTR:
If the stem-vowel(s) is/are back, then the affix
vowel is back, i.e. *… a/o/u … + … e/ö/ü/i/é …
According to HAR (27a) a suffix has to agree in backness with

the stem vowel unless the suffix vowel is neutral. Thus (27a)
permits a neutral-vowel suffix to follow a back-vowel stem.
*NEUTR (27b) on the other hand bans neutral-vowel suffixes
after back-vowel (p.280) stems. *NEUTR is only violated by
surface forms suffixed with non-alternating suffixes (which
contain a neutral vowel; see (26)). In the verbal inflectional
paradigm this does not normally happen since, as we pointed
out above, all verbal inflections harmonize. The only counter
example is Anti-Harmony, the case of lexical allomorphy that
we focus on below.
10.5.2 Partial contrast

Backness harmony breaks down in 1Sg.indef.Cond, where the
normally alternating conditional marker -nal-ne, -nál-né only
occurs in its front variant.13 This can be seen in the last
column of (28) where the conditional marker is emboldened
and the unexpectedly non-harmonizing vowel is underlined.
Page 38 of 76
(28) Backness harmony suspended in 1Sg.indef.Cond.
3Sg. 2 Sg. 1Pl. 2Pl. 3Pl. 1Sg.
akar-na akar-ná-l akar-ná-nk akar-ná-tok akar-ná-nak akar-ná-nak akar-

né -k
teker-ne teker-né-l teker-né-nk teker-né-tek teker-né-nek teker-né-k
Page 39 of 76
The suspension of backness harmony is a strategy to avoid

homophony between 1Sg. indef.Cond and 3Pl.def.Cond. The
expected 1Sg.indef.Cond form would be *akar-ná-k ‘I would
want sg.’, which is homophonous with the attested
3Pl.def.Cond. form akar-ná-k ‘they would want it’. It is
interesting that while backness harmony is suspended in the
back-vowel stems to maintain the contrast, the same strategy
does not apply to the front-vowel stems where homophony
seems to be tolerated: teker-né-k ‘I would turn sg.’ is
homophonous with teker-né-k ‘they would turn it’. This can be
seen in (29).
Page 40 of 76
(29) Partial contrast in Person/number in Conditional forms:
1Sg.indef 3Pl.def
*akar-ná-k
a. back V in the stem akar-né-k ↔ akar-ná-k
b. front V in the stem teker-né-k = teker-né-k
Page 41 of 76
The main properties of Anti-Harmony are: (i) there is partial

contrast in Person / number, i.e. there is contrast for back-
vowel stems, but there is no contrast for front-vowel stems; (ii)
as opposed to Definiteness Neutralization, where a
morphological contrast is maintained to the detriment of
another morphological contrast, here a regular phonological
process is suspended to maintain a morphological contrast;
(iii) the way in which the phonological process is suspended,
i.e. the fact that it is only suspended in back-vowel stems, is
phonologically constrained. It can be shown to be (p.281)
related to a phonological constraint that allows neutral vowels
such as é after back-vowel stems (akar-né-k), but disallows a
back-vowel suffix after a front-vowel stem (*teker-ná-k). This is
summarized in (30).
(30) Properties of Anti-Harmony

i. partial contrast in Person /number
ii. avoiding homophony by suspending a
phonological process
iii. the phonological strategy for avoiding
homophony is phonologically constrained
10.5.3 Optimality Analysis of Anti-Harmony

Anti-Harmony can be analyzed with reference to the constraint
hierarchy used in s. 10.4.2 plus the phonological constraints
HAR and *NEUTR. Most of the paradigmatic uniformity/
contrast constraints of s. 10.4.2 are ranked between HAR and
*NEUTR of which the former is higher ranking. Tableau (31)
shows the analysis of back-vowel stems in 1Sg.indef.Cond. For
the sake of simplicity we disregard P AR(PERSON /NUMBER),
PAR(TENSE /MOOD) constraints and candidates violating
them. These PAR constraints ranked as in (20) would not
change the analysis.
(31) Anti-Harmony for back-vowel stems in

1Sg.indef.Cond
Page 42 of 76
akar+1Sg.indef.Co HAR CON (PERSON / PAR(PERSON / *NEUTR CON

nd ‘I would want NUMBER) NUMBER × × (DEFINITENESS)
sg.’ DEFINITENESS)
a. akar-ná-k ! * (=3Pl: ‘they * (=def: ‘they

would want it’) would want it‘)
b. Ⅾ akar-né-k *…a…+…é…
c. akar-ná-m ! * (iPl.def: see * (=def: ‘I would

(19)) want it’)
d. akar-né-m ! * (1Pl.def: see *… a … + … é …

(19))
Page 43 of 76
(p.282) The paradigmatically uniform (expected) candidate

*akar-ná-k (31a) is homophonous with the 3Pl.def.Cond form
and therefore violates CON(PERSON /NUMBER) and
CON(DEFINITENESS), but it satisfies both harmony
constraints. Candidate (31b) akar-né-k is almost identical with
(31a) but minimally violates harmony. It does not conform to
*NEUTR but satisfies HAR, as the suffix has the neutral vowel
é. Candidate (31b), however, is not homophonous with any
other form in the paradigm. So it does not violate the
CON(PERSON /NUMBER) constraint which is ranked higher
than *NEUTR and therefore bests (31a). It can be seen in (31)
that in this case Anti-Harmony is a better strategy for the
avoidance of Person /number homophony than Definiteness
Neutralization. Candidate (31c) *akar-ná-m has the definite
version of the ISg Person /number marker -m (so it violates P
AR(PERSON /NUMBER × DEFINITENESS)) and, as it
neutralizes definiteness, it violates CON (DEFINITENESS) too.
In order for (31c) to fail, at least one of these constraints must
be ranked higher than *NEUTR. There are several logical
possible rankings, but we assume in tableau (31) that it is only
P AR(PERSON/NUMBER × DEFINITENESS) which is ranked
higher than *NEUTR, and CON(DEFINITENESS) is ranked
lower than *NEUTR. This ranking will be crucial when we
compare Definiteness Neutralization and Anti-Harmony as
competing avoidance strategies in s. 10.5.4.
A mixture of the two strategies (Definiteness Neutralization

and Anti-Harmony) is not a successful strategy. The candidate
*akar-né-m (31d), a form that contains the definite version of
the Person /number marker with a neutral vowel, cannot win
because, although it violates/satisfies the constraints in (31) in
the same way as the optimal candidate akar-né-k, it also
violates P AR(PERSON /NUMBER × DEFINITENESS), which
requires that an indefinite form should bear the indefinite
version of the Person /number marker.
Tableau (32) shows why there is no Anti-Harmony for front-

vowel stems. Although (32b) teker-né-k, a 1Sg form, is
homophonous with a 3Pl. form in the paradigm, it is (p.283)
more harmonic than (32 a) *teker-ná-k which maintains
Person /number contrast, but fatally violates the dominant
HAR constraint.
Page 44 of 76
(32) No Anti-Harmony for front-vowel stem in

1Sg.indef.Cond
Page 45 of 76
teker+1Sg.indef.C HAR CON (PERSON / PAR(PERSON / *NEUTR CON

ond ‘I would turn NUMBER) NUMBER × × (DEFINITENESS)
a. teker-ná-k !*…e…+…á…
b. Ⅾ teker-né-k * (=3Pl: ‘they * (=def: ‘they

would turn it’) would turn it’)
Page 46 of 76
The result of this state of affairs is partial contrast in the

Person /number dimension, i.e. homophony is tolerated for
front-vowel stems, but avoided for back-vowel stems in
1Sg.indef.Cond. The constraint ranking assumed so far is as
(33).
(33) Constraint ranking

PAR(TENSE/MOOD)
HAR » CON(PERSON /NUMBER) » PAR(PERSON /
NUMBER × DEFINITENESS) » » *NEUTR » CON(DEF.)
PAR(PERSON / NUMBER)
It can be seen in (33) that, excepting CON(DEFINITENESS),

PAR and CON constraints are wedged in between phonological
constraints. HAR, like the other phonological constraints
responsible for phonological allomorphy, cannot be violated for
the sake of paradigmatic uniformity and contrast (see (23)).
However, there is another phonological constraint *NEUTR
which—as it ranks lower than most PAR and CON constraints
—can be violated to maintain paradigmatic contrast (though
see s. 10.4.4). This gives Anti-Harmony its special character:
while Definiteness Neutralization is the result of the
interaction between paradigmatic constraints ranked with
respect to one another, Anti-Harmony results from the
interaction of a phonological constraint (*NEUTR) with higher-
ranking paradigmatic constraints. We find Anti-Harmony
where the recessive phonological constraint (*NEUTR) is
violated, but even this is phonologically constrained because
the dominant phonological constraint (HAR) has to be
satisfied.
Page 47 of 76
10.5.4 Weak paradigmatic uniformity

Here we want to point out a residual problem about the
analysis of 1Sg.indef.Cond forms which highlights the crucial
role paradigmatic uniformity plays in the analysis and
suggests that a novel paradigmatic uniformity constraint is
involved too. We showed in (31) why Definiteness
Neutralization is less optimal than Anti-Harmony as a strategy
to avoid homophony for 1Sg.indef.Cond back-vowel stems. We
have also shown why Anti-Harmony does not result in an
optimal form for comparable front-vowel stems (cf. (32)).
However, we have not examined Definiteness Neutralization as
an avoidance strategy for front-vowel stems. It can be seen in
(34) that, according to the constraint hierarchy developed (cf.
(33)), the ungrammatical, definiteness-neutralizing candidate
*teker-né-m ‘1Sg.indef.Cond.’ (34c) would come out as the
most harmonic form instead of the grammatical form teker-né-
k (34b). Thus, the constraint hierarchy assumed in (31) and
(34) yields the correct output for back-vowel stems (e.g. akar-
né-k, (31b)), but an ungrammatical one for front-vowel stems
(e.g. *teker-ném, (34c)).
Note that if the most harmonic forms in (31) and (34) were the
actual grammatical forms, then we would get lexical
allomorphy in 1Sg.indef.Cond. This lexical allomorphy of the
Person /number marker k~m in akar-né-k ~ *teker-né-m would
be of a (p.284) completely new kind. Up to this point we saw
cases of ‘lexical allomorphy’ where the allomorphs occur at
different points of the paradigm. If we think of a paradigm as a
matrix where a row or a column lists forms that belong to a
given value of a dimension, then in the cases we have
examined so far, lexical allomorphy holds between forms in
different cells14 of the matrix, i.e. lexical allomorph α occurs in
some cell or cells, while allomorph a occurs in another cell or
other cells. Figure (35) shows this state of affairs for
Definiteness Neutralization.
(34) Ungrammatical output for front-vowel stems in

1Sg.indef.Cond
Page 48 of 76
teker+1Sg.indef.C HAR CON(PERSON/ PAR(PERSON / *NEUTR CON(DEF.)

ond ‘I would turn NUMBER) NUMBER × ×
a. teker-ná-k !*…e…+…á…
b. teker-né-k ! * (=3Pl: ‘they

would turn it’)
c. ☞teker-né-m * (1Sg.indef: - ok, - *

t-ak, -j-ak)
Page 49 of 76
(35) Lexical allomorphy of 1Sg.indef Person /number

marker k ~ m in different tenses/ moods
The PAR
constraints discussed were designed to penalize lexical

allomorphy of this kind. Note that if the lexical allomorphy
predicted by the constraint hierarchy (p.285) occurred, i.e.
the 1.Sg.indef.Cond of front stems were stem+né+m
(*tekerném, cf (34)) while the 1.Sg.indef.Cond back stems are
stem+né+k (akarnék, cf. (31)), then this allomorphy of the
Person /number marker would be of a new kind, because it
would involve allomorphy within a single specific cell of the
paradigm matrix. This hypothetical state of affairs is shown in
(36).
(36) Lexical allomorphy of 1Sg.indef Person /number

marker k~m within a cell in the Conditional
Page 50 of 76
PAST/ INDICATIVE PRESENT/ SUBJUNCTIVE-IMP. PRESENT/

INDICATIVE CONDITIONAL
… … …
1Sg. indef
… … …
Page 51 of 76
Furthermore, this putative k ~ m allomorphy would be

phonologically unmotivated (unless we wanted to claim—
implausibly—that the choice of -k v. -m depends on the
backness/frontness of the stem).
This kind of allomorphy cannot be penalized with the help of

the PAR constraints discussed so far, since they require that
the exponents of a morpheme assigned to a given value of a
dimension should be chosen from a restricted set. No violation
of PAR(D) would occur if the allomorphs within the specific cell
examined are members of the restricted set assigned to
dimension D. The putative k~m allomorphy in the
1Sg.indef.Cond does not violate P AR(PERSON /NUMBER)
since the restricted set assigned to the 1Sg value of the
paradigm happens to be {-k, -m}. A new constraint is needed
which is stated at (37) as WEAKPAR.
(37) WEAKPAR: weak paradigmatic uniformity No

lexical allomorphy within a specific cell of a given
paradigm.
In essence WEAKPAR says that the set assigned to the PAR

constraint relativized to the direct product of all the
dimensions of the paradigm is a singleton (one-member set).
The relevant constraint is PAR(PERSON/NUMBER ×
DEFINITENESS × TENSE/MOOD) in Hungarian.
Lexical allomorphy within a specific cell of a paradigm seems

to be attested only if it is phonologically motivated (recall the -
ѕz ~ - Vl allomorphy in 2Sg.indef.Ind in Hungarian discussed in
10.3.2). This suggests that WEAKPAR is dominated by
phonological constraints (like other paradigmatic constraints).
If WEAKPAR dominates the paradigmatic constraints in

Hungarian, then the problem in (34) is resolved: the
grammatical form (teker-né-k) is the most harmonic form
under this ranking, because *teker-né-m fails on WEAKPAR
(38).
(38) WEAKPAR » PAR(D), CON(D)
(p.286)
(39) No lexical allomorphy within a cell in order to

maintain Person /number contrast
Page 52 of 76
teker+iSg.indef. WEAKPAR HAR CON (PERSON / PAR(PERSON / *NEUTR CON

Cond ‘I would NUMBER) NUMBER × × (DEFINITE-
turn sg.’ DEFINITENESS) NESS)
a. teker-ná-k ! * … e… + … á
…
b. ☞ teker-né-k * (= 3Pl ‘they * (= def ‘they

would turn it’) would turn it’)
c. teker-né-m !* * (1Sg.indef: - * (= def ‘I

ok, -t-ak, -j-ak) would turn it’)
Page 53 of 76
Note that lexical allomorphy in a given dimension is not a

WEAKPAR violation if the allomorphs do not co-occur in a
specific cell specified in all the dimensions of the paradigm.
Thus, for instance, the fact that there is lexical allomorphy in
the first person singular in the Hungarian verbal paradigm
(k~m, cf. (12)) does not cause a violation of WEAKPAR since
this allomorphy is never found within a specific iSg cell of a
particular Definiteness and a particular Tense/mood value.
This means that the dominant WEAKPAR constraint would not
change the state of affairs shown in the previous tableaux
(such as (19) and (21), for instance).
The ranking in (38) seems to be the unmarked ranking of

WEAKPAR with respect to the other paradigmatic constraints.
The intuition behind this is that if more than one affix belongs
to a given cell and the difference between the affixes is not
phonologically motivated, then the forms containing the affixes
belong to different paradigm-types (form classes, e.g.
declination classes as in Latin; conjugations of closed stem
classes as in some irregular verb types in English).
10.5.5 Anti-Harmony v. Definiteness Neutralization

If the Anti-harmony strategy is used to maintain Person /
number contrast in 1Sg.indef. Cond, the question naturally
arises why it is not used for the same purpose in 1Sg.
indef.Past and 1Pl.def.Cond, where Definiteness Neutralization
is the actual strategy (p.287) (cf. s. 10.4.2), i.e. why not use
anti-harmonic forms (back-vowel suffix alternants after front-
vowel stems, and front-vowel suffix alternants after back-vowel
stems) to avoid homophony?
In 1Sg.indef.Past, the anti-harmonic avoidance forms would

contain the suffix -ek after back-vowel stems (e.g. *akar-t-ek)
and -ak after front-vowel stems (*teker-t-ak). The reason why a
definiteness-neutralizing avoidance form is more harmonic
than an anti-harmonic avoidance form is phonological. The
anti-harmonic avoidance forms violate the highly ranked
constraint HAR. For front-vowel stems, this works in exactly
the same way as in 1Sg.indef.Cond (*teker-t-ak is
ungrammatical for the same reason—i.e. violates and satisfies
the constraints in the same way—as *teker-ná-k, cf. (32)). This
is shown in (40).
Page 54 of 76
(40) No Anti-Harmony but Definiteness Neutralization

in 1Sg.indef.Past for front-vowel stems (cf. (19) and
(32))
Page 55 of 76
teker+1Sg.indef.P HAR CON (PERSON/ PAR(PERSON / *NEUTR CON(DEF.)

ast ‘I turned sg.’ NUMBER) NUMBER × ×
DEFINITENESS)
a. teker-t-ak !*…e…+…a…
b. teker-t-ek ! * (=3Pl ‘they

turned sg.’)
c. ☞teker-t-em * (1Sg.indef -ok,-j- *

ak,-né-k)
Page 56 of 76
In the case of back stems, however, there is an interesting

difference which is due to an asymmetry between the vowel é
and the vowel e with respect to neutrality to back-ness
harmony. As we have seen in s. 10.5.1, é is neutral but e is not.
Consequently, a suffix with é after a back-vowel stem does not
violate HAR, but a suffix with e does. Since in 1Sg.indef.Cond,
the grammatical, non-neutralizing but anti-harmonic
avoidance form (akar-né-k) has the suffix vowel é, it does not
violate HAR—so it can best the definiteness-neutralizing, but
harmonic candidate (*akar-ná-m), cf. (31). By contrast, in
1Sg.indef.Past, the non-neutralizing anti-harmonic avoidance
form (*akar-t-ek) has the suffix vowel e, so it fatally violates
the highly ranked HAR and loses to the harmonic definiteness-
neutralizing candidate (akar-t-am). This is shown in (41).
Let us now turn to the other case when Definiteness

Neutralization occurs: 1Pl. Cond. This case is different from
1Sg.Past discussed above because the vowel of the (p.288)
conditional suffix -ná /-né is long, and thus its front variant is
the neutral vowel é (and not the short non-neutral e we saw in
1Sg.Past). The behavior of front-vowel stems is
straightforward. As we have already seen in (32) for
1Sg.indef.Cond. front-vowel stems, Anti-Harmony is not a
viable strategy to avoid Person /number homophony. The same
is true in this case as well. The paradigmatically uniform 1Pl
definite form *teker-né-k (42b) incurs a fatal CON(PERSON /
NUMBER) violation because it is homophonous with the
1Sg.indef and 3Pl.def forms. The anti-harmonic
paradigmatically uniform *teker-ná-k (42 a) fails on the
dominant phonological constraint HAR (similarly to the
comparable form in (32)). The grammatical form teker-né-nk
(42c) contains the ‘wrong’ (indefinite) Person /number suffix
(thus violates P AR(PERSON /NUMBER × DEFINITENESS))
and is homophonous with the 1Pl. indefinite form (thus
violates CON(DEFINITENESS)). Nevertheless, it wins because
these constraints are ranked lower than HAR. The anti-
harmonic and paradigmatically non-uniform candidate *teker-
ná-nk (42d)—although it is not homophonous with any member
of the paradigm—fails because it violates HAR (in the same
way as (42 a)).
Page 57 of 76
(41) No Anti-Harmony but Definiteness Neutralization

in 1Sg.indef.Past for back-vowel stems (cf. (19), (31)
and (32))
Page 58 of 76
teker+1Sg.indef.P HAR CON (PERSON/ PAR(PERSON / *NEUTR CON(DEF.)

ast ‘I wanted sg.’ NUMBER) NUMBER × ×
DEFINITENESS)
a. akar-t-ek !*…e…+…a…
b. akar-t-ak ! * (=3Pl ‘they

wanted sg.’)
c. ☞akar-t-am * (1Sg.indef: -ok,-j- *

ak,-né-k)
Page 59 of 76
The selection of candidates for back-vowel stems is more

complex. The 1Pl.def Cond, anti-harmonic avoidance
candidates would contain -né-k or -né-nk after back-vowel
stems (e.g. *akar-né-k (43b), *akar-né-nk (43d)). As the suffix
vowel here is neutral é, HAR is not violated. These candidates,
however, fail for another reason. Constraint (43 b) fatally
violates CON(PERSON /NUMBER) because it is homophonous
with the 1Sg.indef.Cond form akar-né-k ‘I would want sg.’.
Constraint (43d) violates PAR(PERSON /NUMBER ×
DEFINITENESS) in the same way as the grammatical form
akar-ná-nk (p.289) (43c). However, the latter form does not
violate the lower-ranked *NEUTR, and therefore bests (43d).
Note that this can happen only because CON(DEFINITENESS)
ranks lower than *NEUTR. Thus, the reason why the Anti-
Harmony strategy is not used here instead of the Definiteness
Neutralization strategy to maintain Person /number contrast is
that (a) it would actually create Person /number homophony
(neutralization) at another point in the paradigm (see (43b));
(b) if no Person /number homophony is created and the
candidates are equally paradigmatically non-uniform, then the
ranking *NEUTR » CON(DEFINITENESS) decides.
(42) No Anti-Harmony in 1Pl.def.Cond (cf. (22) and (31))
Page 60 of 76
akar+1Pl.def.Cond HAR CON(PERSON / PAR(PERSON / *NEUTR CON

‘we would want it’ NUMBER) NUMBER × × (DEFINITENESS)
DEFINITENESS)
a. akar-ná-k !*
(= 3Pl
‘they would
want it’)
b. akar-né-k !* * *
(= 1Sg …a…+ (=indef
‘I would …é… ‘I would
want sg.’) want sg.’)
c.☞ akar-ná-nk * *
(1Pl.indef: (=indef
-j-uk,-t-uk, ‘we would
-j-uk) want sg.’)
d. akar-né-nk * !*
(1Pl.indef: …a…+
-j-uk,-t-uk, …é…
-j-uk)
Page 61 of 76
Note that the constraint hierarchy would decide in the same

way about candidates that satisfy PAR(PERSON NUMBER ×
DEFINITENESS) (since in both cases the candidates would
draw on the constraint, and the lower-ranking constraint
would decide). This explains why in the 2Pl.def.Cond
definiteness homophony is not avoided (e.g. akar-ná-tok
‘you(Pl.) would want it’ = akar-ná-tok ‘you(Pl.) would want
sg.’). This hom-ophony is due to the fact that there is no
distinct 2Pl. Person /number marker in the definite v. the
indefinite (see (7ii)) and the templatic morphological property
mentioned in (8) prevents the expression of definiteness by an
overt definite marker in the Conditional (see (8b)). Tableau
(44) below shows that the anti-harmonic candidate (44b)
necessarily loses to the harmonic but definiteness neutralizing
candidate (44 a).
(p.290)
Page 62 of 76
akar+1Pl.def.cond HAR CON (PERSON/ PAR(PERSON / *NEUTR CON(DEFINITENE

‘ we would turn it.’ NUMBER) NUMBER × × SS)
DEFINITENESS)
a. teker-ná-k !*
b. teker-né-k ! ** (=1Sg ‘I would *(=indef ‘I would

want sg’.) (=3Pl want sg’.)
‘they would want
it’)
c. ☞teker-né-nk * (1Pl.indef: -juk,-t- *(=indef ‘we would

uk,-j- uk) want sg’.)
d. teker-ná-nk !* * (1Pl.indef: -juk,-t-

uk,-j- uk)
Page 63 of 76
Page 64 of 76
akar+2Pl.def.Cond HAR CON(PERSON/ PAR(PERSON/ *NEUTR CON

you(Pl) would want NUMBER) NUMBER × (DEFINITENESS)
it’ XDEFINITENESS)
a.☞akar-ná-tok * (=indef: ‘you(Pl.)

would want sg.’)
b. akar-né-tok *…a…+…é…
Page 65 of 76
(p.291) 10.5.6 Target cells v. reference cells

An important point has to be made here. We have examined
different strategies for the avoidance of homophony of a form
in some specific cell (the ‘target cell’) of the paradigm with a
form in another specific cell (the ‘reference cell’). The
constraint hierarchy developed can explain why a given
strategy and not some other strategy is employed to avoid the
neutralization of paradigmatic contrast, but it does not explain
why the strategy applies in a given cell and not some other
(why optimality is computed for one cell and not for the other),
i.e. it does not distinguish between target cells and reference
cells.
The target–reference relations for the conditional forms

examined are shown in (45), where arrows connect reference
cells and target cells, with the head of the arrow pointing at
the target. The paradigmatically non-uniform suffixes are
underlined. The starred forms in a target cell are the
paradigmatically uniform candidates that are avoided to
maintain paradigmatic contrast. The relevant cells of the
Conditional matrix are shaded.
(45) Reference cells and target cells in the Conditional
It can be
seen in
(45) that
some
avoidance
strategy
(specifically, Definiteness Neutralization or Anti-Harmony)

applies in the target cells to avoid homophony with a form in
one of the reference cells. For instance, the constraint
interaction is computed for the 1Pl.def.Cond form akar-ná-nk
as a target with reference to (to avoid homophony with) the
form akar-ná-k in the 3Pl.def.Cond (reference) cell. The point is
that there is no reason why it should not be the other way
round, i.e. in principle 3Pl.def Cond could be the target cell
and 1Pl.def.Cond the reference cell (i.e. we could be
Page 66 of 76
calculating how the 3Pl.def.Cond should be realized in order to

avoid homophony with the 1PL.def.Cond). This is not what
Hungarian does, but the tableaux do not reflect this
property.15 It can also be seen in (43) that the target–reference
relations may be (p.292) quite complex: the same cell can be
a target cell with respect to one cell and a reference cell with
respect to another one in the paradigm: e.g. 1Sg.indef.Cond. is
a target cell with respect to 3Pl.def.Cond, but a reference cell
with respect to 1Pl.def.Cond.
The issue underlying this state of affairs is that identity

between form X and form Y is a symmetrical relationship,
while the relationship between target and reference is an
asymmetrical one. In some cases this is not a problem since
the phenomenon that might cause homophony in the paradigm
can be such that it automatically establishes the asymmetry.
Kenstowicz (Ch. 7 above) discusses various phonological
processes from several languages that are suspended when
they should result in homophony in the paradigm. He points
out that ‘in all our examples it is the member of the paradigm
that the phonology threatens with merger that is also the site
of repair’. The reason is that the phonological processes he
discusses are asymmetrical themselves in the sense that they
could apply only at one point in the paradigm (the target), but
not at the other (the referent) thereby identifying which is
which.16 It is obvious that the Hungarian morphological
regularities that threaten to produce homophony are not
inherently asymmetrical. Target–referent asymmetry has to be
imposed on the paradigm in some way.
A detailed analysis of this question is beyond the scope of this

chapter. We suggest that the intuition that may help here is
that there are inherent semantic/pragmatic asymmetries
between the values of a dimension active in a paradigm. Some
values seem to be semantically/pragmatically unmarked with
respect to other values. Irregularities like lexical allomorphy
seem to occur at the unmarked point(s) of these asymmetries
(cf. Burzio 2002a; Kenstowicz, Ch. 7 above). This can be seen
in the Person /number dimension, where for example in the
Hungarian conditional cases the ‘greatest degree’ of
irregularity (Anti-Harmony) can be found in semantically/
pragmatically unmarked 1Sg. The 1Sg.indef.Cond form akar-
né-k (45) contains a suffix allomorph -né which is otherwise
absent from the paradigm of the stem. In the iPl.def Cond form
akar-ná-nk, we find a suffix allomorph which is ‘less’ irregular,
Page 67 of 76
it is attested in the paradigm of the stem (Definiteness

Neutralization). By contrast, the semantically/pragmatically
marked 3Pl. form akar-ná-k does not show lexical allomorphy.
We do not have a solution to this problem, but want to
emphasize its importance and leave it for further research.17
Page 68 of 76
(p.293) 10.6 Conclusions

In this chapter we have analyzed lexical allomorphy in the
Hungarian verbal paradigm, specifically the variety that is
paradigmatically motivated. The traditional strategy to
account for allomorphy with reference to a common
underlying form cannot succeed in the case of lexical
allomorphy because in lexical allomorphy the allomorphs do
not have phonological properties that permit the postulation of
a single underlying source. Therefore lexical allomorphy is
inaccessible to derivational analysis or an analysis based on
the comparison of underlying and surface forms. By contrast,
our analysis relies on Output–Output constraints. We have
seen that lexical allomorphy is due to an interaction of
paradigmatic uniformity and contrast constraints which are
relativized to morphosyntactic dimensions. Hungarian verbal
inflectional morphology is basically agglutinative, but crucially
the correspondence between morphosyntactic values and
distinct suffixes is not always one-to-one: in some forms more
than one morphosyntactic category can be expressed by one
suffix. We have developed an alternative interpretation of PAR
constraints to handle this situation. This interpretation
explains an interesting phenomenon (of which the Hungarian
paradigm is an example), namely the case when lexical
allomorphy is conservative, i.e. when the lexical allomorph
must be one which is attested at some other point(s) in the
paradigm.
Potential homophony within a paradigm may have two kinds of

sources: (1) the phonological/morphological regularities of the
language may be such that they give identical surface forms
when applied to words in the paradigm containing different
suffixes; (2) suffixes assigned to different morphosyntactic
values may be identical. We have seen that the requirement to
maintain morphological contrast within the paradigm is
relative: it may be more important to have contrast in one
morphosyntactic dimension than in another. Full Person /
number homophony (homophony for all stems) is absent from
the Hungarian verbal paradigm while various instances of
definiteness homophony are attested. Some irregularities of
the paradigm (lexical allomorphy) are related to this. It is
more important to maintain Person /number contrast than
definiteness contrast (Definiteness Neutralization). This can be
interpreted as the dominance of a CON constraint relativized
to Person /number over a CON constraint relativized to
Page 69 of 76
definiteness. This relationship may be non-language-specific.

Note that the morphosyntactic category of definiteness rarely
occurs cross-linguistically and is marked in this sense (see
Bybee 1985 for the markedness of verbal inflectional
categories).18
(p.294) The way in which homophony is avoided within the

paradigm is not only due to the dominance relationships
between relativized CON constraints. It also depends on (1)
the relationship of these constraints to PAR constraints
(PAR(PERSON /NUMBER), PAR(PERSON /NUMBER ×
DEFINITENESS), PAR(TENSE/MOOD), and WEAKPAR, and (2)
the relationship of CON constraints to phonological
constraints (HAR and *NEUTR) and morphological constraints.
The phonological, morphological constraints and the PAR
constraints limit the strategies in which contrast is
maintained. There may be no way to avoid homophony within
these limits. In front-vowel stems in 1Sg.indef.Cond (teker-né-
k), there is Person /number homophony (with the 3Pl.def.Cond
form) for precisely this reason. The stem vowels determine the
presence/absence of contrast: there is no contrast for front-
vowel stems, but there is for back-vowel stems (partial
contrast).
To sum up, we have found the types of paradigmatic

constraints at (47) active in shaping the Hungarian paradigm.
(47) a. paradigmatic contrast Con(Person /number)

constraints:
Con(Definiteness)
b. paradigm uniformity Par(person /number)

constraints:
PAr(PeRSON /number x
Definiteness)
Par(Tense/mood)
WeakPar
Naturally, these constraints do not uniquely determine the

paradigm by themselves (there are other constraints involved
that we have not discussed19), but they crucially determine
lexical allomorphy within the paradigm.
Page 70 of 76
Appendix: The Verbal Paradigm in Present-Day Colloquial

Hungarian
Note: The table contains the complete inflectional paradigm of
the stem akar ‘want’. Other stems are included only if they are
relevant: e.g. when the harmonic alternant of the suffix is not
predictable by vowel harmony (as in the case of the front-
vowel stem teker ‘turn’), or when phonologically motivated
lexical allomorphy sz ~ Vl occurs (as in the case of the sibilant-
final stem hoz ‘bring’). The Person /number markers are
indicated in bold-face, the tense/mood and definite markers
are set in italics. The parts of the suffixes that irregularly
undergo lexical allomorphy are underlined, and the relevant
cells are shaded. The last row and column of the table
summarize the ‘regular’ forms of the tense/mood and Person /
number affixes, respectively Small capital characters indicate
the vowel alternants according to backness and rounding
harmony (A=a /e, Á=á / é, U=u /ü, O=o /e / ö). Expected forms:
a
*akartak
b
*akarnák
c
*akarál /*akarol, *hozál
d
*tekerje, *tekerjétek, *tekerjék
e
*akarnák
f
*akartanak
(accented vowels are long, sz=[s], s=[∫])
(p.295)
Page 71 of 76
PRESENTSUBJUNCTI
VE
PRESENT PAST INDICATIVE IMPERATI PRESENT Person/number ↓

INDICATIVE VE CONDITIONAL
INDEF. DEFINITE INDEF. DEFINITE INDEF. DEFINITE INDEF. DEFINITE INDEF. DEF.
1Sg. akarok akarom akartam a akartom akarjak akarjam akarné k b akarná m −k −m

tekerné k
2Sg. akarsz akarod akartál akartad akarj ál, akarjad, akarná l akarnád -Á1 −d
c akarj akard
hozol
3Sg. akar akarja akart akarta akarjon akarja akarna akarná ∅ -(j)−a
d tekerte tekerje tekerné
tekeri
1Pl. akarunk akarj uk akart unk akart uk akarj unk akarj uk akarná nk akarná nk −unk -(j)−uk
tekerj ük tekerj ük
2Pl. akartok akarjátok akart atok akartátok akarjatok akarjátok akarná akarná −tok -(j)Á-tok
tekeri tek tok tok
d
3PI. akarnak akarják akartak f akarták akarjanak akarják akarná akarná k −nAk −(j)Á-k
tekerikd nak
1Sg. obj:2 akarlak akarta lak akarjalak akaraná −lAk

lak
∅ -t- -j- - nA- ← tense/mood
Page 72 of 76
Notes:
(1) See also Kiparsky (1972, 1978) for a discussion of related
issues in the framework of classical generative phonology
(2) Some authors use another name, e.g. ‘Uniform

Exponence’ (Kenstowicz 1996), ‘Anti-Allomorphy’ (Burzio
1996).
(3) Throughout this chapter we use normal Hungarian

orthography unless it is crucially important to give a phonetic
transcription, Consonant letters typically have the transparent
values, although there are some notable exceptions: e.g.
sz=[s], s=[∫], cs=[tȢ], ny=[ɲ]. Note that the single or double
acute accent above a vowel letter means length and not stress.
For the phonetic values of vowel letters see (24).
(4) There is a special case of paradigmatic uniformity that

requires that affixed form should be (partially) identical with
the base. This is known as ‘base identity’ (cf. Kager 1999b) and
may be relativized with respect to the phonological property
the base and the affixed form have to share (e.g. stress:
Steriade 1997a; syllabification: Kenstowitz, Ch.7). Base
identity can be interpreted as a PAR constraint relativized to
the stem. As here we concentrate on affix-identity (PAR(D)
constraints), we do not discuss base-identity effects.
(5) This is only true in forms where there is no other strategy

for marking definiteness. Sometimes the definiteness contrast
is expressed in another way, for instance in 3Pl. Past forms,
the indefinite Person /number marker is irregularly identical to
the definite one, but a morphological process prevents
neutralization, e.g. akar-t-ak ‘they want sg.’, but akar-t-á-k
‘they want it’. We shall disregard this here—see Rebrus
(2000a).
(6) For a more radical case that involves consonants too, see
Rebrus (2000b).
(7) Recall that the Person /number suffix is used to express

(in)definiteness in this case, see (7i).
(8) We assume here that two regular phonological processes,

High Vowel Deletion (which deletes high vowels after a stem-
final vowel) and Low Vowel Lengthening (which lengthens a/e
into á/é before suffixes) would apply in the paradigmatically
Page 73 of 76
uniform hypothetical 1Pl.def Cond (akar+na+uk → *akarnák)

similarly to the attested 1Pl.indef Cond (akar+na+unk →
akarnánk).
(10) It is an interesting question for further research how

much of and to what extent the hierarchy ofmorphosyntactic
dimensions is derivable from universal constraints (see s.
10.6).
(9) In this tableau and the following ones we have included the
‘bases’ of the violations in parentheses forease of reference,
i.e. we list the specifications and glosses of the forms that the
candidate examined wouldbe identical with in the case of CON
constraints, and the (sub)strings that the candidate (sub)string
wouldbe different from in the case of PAR constraints. The
crucially relevant parts of the strings and glosses
areemboldened.
(11) It is interesting that there is an obsolete/dialectal

alternative form for 1Pl.def Cond in which the low and high
vowels are merged into a long mid vowel ó [o:] or ő [ø:]:
akar+na+uk → % akarnók, teker+ne+ük →% tekernők ‘we
would turn it’. In this case there is no homophony with the
3Pl.def Cond form: akarnák/ tekernék (‘want/turn 3Pl.def
Cond’) ≠ akarnók/tekernõk (‘want/turn 1Sg.def Cond’), but the
two inflectional suffixes show fusion.
(12) There are a few exceptional non-alternating suffixes that

contain back (non-neutral) vowels. We disregard them here,
but note that they are exceptional in other respects as well.
(13) The difference between the length of the vowel of -na/-ne

on the one hand, and -ná /-né on the other is due to a regular
phonological process (Low Vowel Lengthening). Note that Low
Vowel Lengthening changes the relationship of the vowel to
backness harmony since the lengthening of the non-neutral e
results in neutral é.
(14) A cell is a point of the paradigm where the values of all

the dimensions are specified.
(15) Similarly, in the case of the past tense forms discussed in

s. 10.3.1, constraint interaction is computed for the cell
1Sg.indef Past (target), cf (19). Constraint (19 a) fails on
CON(PERSON /NUMBER) because it is identical with a form in
Page 74 of 76
the 3Pl.indef.Past cell (the reference cell). Again, target and

reference could in principle be reversed.
(16) A case in point is the Bulgarian reduction of /o/ to [a] in

unstressed syllables which fails to apply in case it should
produce homophony between singular and plural forms: /kapít-
o/ → [kapít-a] ‘hoe, Sg.neuter’ (compare /kapit-á/ → [kapit-á]
‘hoe, Pl.neuter’), but /blág-o/ → [blág-o] ‘blessing,
Sg.neuter’ (compare /blág-a/ → [blág-a] ‘blessing, Pl.neuter’).
Vowel reduction could not apply in the plural forms.
(17) Viktor Trón (personal communication) suggests that this

asymmetry may be introduced into the constraint hierarchy by
a ‘decomposition’ of PAR(D) constraints, in the form of PAR
constraints which are relativized not only to a dimension, but a
specific value of a dimension. Such PAR(D) constraints could
apply only to a specific value of a morphosyntactic dimension.
If ranked sufficiently high, this constraint could identify the
specific value in question as the referent (and exclude lexical
allomorphy in the cells belonging to this value). If it is low and
can interact with the other constraints, the specific value in
question may be a target, and thus lexical allomorphy is
possible in its cell(s).
(18) Interestingly, the -ja ~ -i lexical allomorphy of the

definiteness marker (e.g. 1Sg.def Ind: akar-ja ~ teker-i)
provides a further example for the dominance relations
between the morphosyntactic dimensions. This allomorphy is
due to the need to create (partial) contrast in the tense/mood
dimension, but the allomorphy (and the partial contrast) is
suspended in 1Pl.def.Ind in order to avoid person /number
homophony: akar-j-uk ~ teker-j-ük (see the Appendix). This
implies the dominance of the Person /number contrast over the
tense/mood contrast, which is also compatible with Bybee’s
results (see the detailed analysis in Rebrus 2000b).
(19) For instance, there must be constraints that determine

the order of suffixes.
Page 75 of 76
Page 76 of 76
A Note on Paradigm Uniformity and Priority of the Root


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
A Note on Paradigm Uniformity and

Priority of the Root
Suzanne Urbanczyk
DOI:10.1093/acprof:oso/9780199267712.003.0011

This chapter examines a number of alternations, showing that
paradigm uniformity effects obtain for the object agreement
paradigm. The emergence of the root vowel with perfective
stems is shown to be a case where paradigm uniformity does
not hold within the aspectual paradigm. The chapter begins
with an analysis of the reduction of the full vowel to schwa.
This is shown to be an instance of overapplication of vowel
reduction, which occurs within the object agreement
paradigm. The full vowel of the root changes to schwa when it
is followed by a stressed full vowel.
Keywords: paradigm uniformity, object agreement, root vowel, alternations,

schwa
Page 1 of 33
11.1 Introduction
The role of the paradigm in explaining phonological
(ir)regularities has recently reemerged in phonological and
morphological theory. Generally speaking there is a tendency
toward phonological uniformity within a paradigm, but under
some conditions uniformity is not maintained. In order to
understand the role of the paradigm in the grammar it is
important to learn more about when identity is and is not
maintained. Approaches to paradigm uniformity include cyclic
application of phonological rules (Kiparsky 1982 b) or positing
a relationship between fully formed words (see Benua (1997a);
Burzio (1994a); Flemming and Kenstowicz (1995) and other
chapters in this volume). Two issues that have emerged
recently are determining the base for word-formation and
understanding the conditions under which uniformity does or
does not (p.297) occur. What unites these two issues is a
central goal to understand which words stand in a
paradigmatic relation and how phonological processes interact
with morphological structure. In investigating these issues, a
common assumption is that the base is the stem, minus the
formative, which is being added. McCarthy (Ch. 8 above)
refers to this perspective as illustrating base-priority. In this
chapter base-priority refers explicitly to the output word to
which some morphological process applies. It is interesting to
examine cases where base-priority seems to be violated. For
example, in the formation of imperfective stems in Upriver
Halkomelem (Central Salish), the base would be the stems in
(1a.i) and (1b.i).1
Page 2 of 33
(1) Upriver Halkomelem (Galloway 1993)
a. i. [q’ə́lmət] ‘believe someone

(perfective)’
ii. [q’éq’əlmət] ‘believing someone

(imperfective)’
b. i. [łə́qw’ cəsəm] ‘clap one’s hands (once)

(perfective)’
ii. [łáłəqw’ cəsəm] ‘clapping one’s hands

(imperfective)’
Page 3 of 33
What is puzzling in these cases is that the quality of the vowel

in the reduplicated portion (1a.ii and 1b.ii) differs from the
perfective. However, further knowledge of the word-family
reveals that the vowel is in fact a copy of the root vowel, as
can be seen in (2).
(2) Corresponding roots/stems (Galloway 1993)
a. [q’é:l] ‘believe’
b. [łáqw’ət] ‘slap someone or something’
The non-uniformity found in the paradigm is an instance of the

root vowel emerging in reduplication; this shows priority of
the root vowel. As one can see, emergence of the root vowel
cannot be derived by reference to the output perfective
because it has schwa, rather than the root vowel. This is
therefore an instance in which base-priority does not hold,
while root priority prevails.
This chapter will examine a number of alternations related to

the data presented in (1) and (2) above, showing that
paradigm uniformity effects obtain for the object agreement
paradigm. The emergence of the root vowel with perfective
stems will be shown to be a case where paradigm uniformity
does not hold within the aspectual paradigm. The chapter
begins with an analysis of the reduction of the full vowel to
schwa (the process illustrated in (1a.i) and (1b.i)). This will be
shown to be an instance of overapplication of vowel reduction,
which occurs within the object agreement paradigm. The basic
pattern is illustrated at (3). As one can see, the full vowel of
the root (p.298) changes to schwa when it is followed by a
stressed full vowel, as in (3a–d). No such adjacent stress is
present with the form in (3e), and so reduction is claimed to
over-apply, resulting in a uniform object agreement paradigm.
Page 4 of 33
(3) Hul’q’umi’num’ (Leslie 1979:31)
a. /lem-n-am̓š/ lə̀mnám̓š ‘see me’
b. /lem-n-al̓xw/ lə̀mnál̓xw ‘see us’
c. /lem-n-amə / lə̀mnámə ‘see you (sg)’
d. /lem-n-alə/ lə̀mnálə ‘see you (pl)’
e. /lem-n-əxw / lə́mnəxw ‘see him, her, it, them’
Page 5 of 33
The analysis is set within Optimality Theory, and makes use of

McCarthy’s Optimal Paradigms model (see Ch. 8 above).
Section 11.3 then examines the failure of the paradigm
uniformity effect with imperfective aspect, as illustrated in (1)
above. This chapter builds on the work of Downing (1999c),
which shows that there are paradigmatic relations between
reduplicated words. As it turns out, separate paradigmatic
relations must hold between those words inflected for object
agreement from those inflected for aspect. This is formally
accomplished by having different correspondence relations
hold between inflected words.
11.2 Halkomelem Vowel Reduction

Halkomelem is an endangered Central Salish language spoken
in southwestern British Columbia. There are three main
dialects: Downriver Halkomelem (hən̓q̓əmin̓əm̓), spoken
around Vancouver; Upriver Halkomelem (Halq’eméylem),
spoken further along the Fraser River to around Yale
(Galloway 1993), and Island Halkomelem (Hul’q’umi’num’)
spoken on the east coast of Vancouver Island, from around
Duncan to just north of Nanaimo. Each of these dialects is
comprised of a number of smaller dialects. When referring to
properties of the language shared by all dialects, the term
Halkomelem will be used. Otherwise the more specific terms
will be used. The analysis is based on data gathered from
fieldwork on Halq’eméylem and Hul’q’umi’num’, reference to
Galloway (1993), Leslie (1979), and extensive dictionary
searches of Hukari and Peter’s (1995) Hul’qumi’num
Dictionary.
11.2.1 Schwa reduction

A number of phonological and morphological operations in
Salish languages are sensitive to the distinction between roots
or stems with full vowels and schwa (or zero vowel/consonant-
only roots). Schwa is a vowel segment lacking a place
specification, which picks up its vowel quality from the
adjacent consonants. It is represented in the data below as /ə/.
The distribution of schwa is mostly predictable, resulting in a
debate on whether schwa is in the underlying form (Kinkade
1998), hence the terms consonant-only (p.299) roots and
schwa-voweled roots are used to refer to the same types of
roots. None of the claims or observations here depends on
whether or not schwa is in the input or is part of the phonemic
inventory.
Page 6 of 33
One important difference between the two types of vowels is

how stress is assigned. Stress prefers to fall on full vowels,
and stressed schwa is generally avoided (Bianco 1996;
Urbanczyk 1996; Shaw et al. 1999). The following words
illustrate the preference to stress full vowels, and the
avoidance of stressing schwa in Hul’q’umi’num’. Simplifying
Bianco’s (1996) observations about Hul’q’umi’num’ somewhat,
stress falls on the first full vowel, as in (4a) and (4b). If there
are no full vowels, then stress falls on the first schwa. There
are some exceptions to this pattern, but the basic point that
schwa resists stress holds throughout the Salish language
family.
(4) Hul’q’umi’num’ (Bianco 1996)
a. [kw’áyəkw ] ‘fish (verb)’
b. [šəmén] ‘enemy’
c. [txWə́məcən] ‘September’
It is therefore quite striking that a stressed schwa is created.
The process of reducing an underlying full root vowel to a

stressed schwa has been widely observed for Halkomelem
(Galloway 1993; Suttles forthcoming; Shaw et al. 1999).
However, the exact conditions have not been investigated fully.
While it is considered to be a morphophonemic process in the
Musqueam dialect by Shaw et al. (1999), Hukari (p.c.) points
out that in Hul’q’umi’num’, schwa reduction tends to occur
when CVC shaped roots are followed by a -CVC shaped suffix.
The Halq’eméylem data at (5) illustrate the pattern.
(5) Halq’eméylem
a. [qwé:l] ‘talk’2
[qwə́lmət] ‘bawl someone out’
b. [tás] ‘bump (verb)’
[tə́sləxw] ‘bump someone (by accident)’
c. [łí:kw] ‘hook (verb)’
[łə́:kwləxw] ‘hook something (by accident)’
[łə́kwtəl] ‘fishhook [Galloway 1993]
Page 7 of 33
[łə́kwxyəl] ‘trip, stumble on something’ [Galloway

1993]
d. [łíc’] ‘cut’ [Galloway 1993]
[łə́c’cəs] ‘cut on the hand’ [Galloway 1993]
[łə́c’xyəl ‘cut on the foot’ [Galloway 1993]
The reduction is remarkable, because it creates a segment

that is actively avoided in many Salish languages: a stressed
schwa.
Consideration of other root- and suffix-shapes supports the

observation that reduction to schwa is restricted to CVC
shaped roots. Notice that in the following roots, (p.300)
which are greater than CVC, there is no reduction of the root
vowel to schwa, even though they are affixed with the same
reduction-inducing suffixes presented above. An extensive
study of all root shapes (CV, CVC, CCVC, CVCCC, etc.) has yet
to be undertaken to understand the conditions fully.
(6) Halq’eméylem
a. [mélqləxw] ‘forget something/someone’ [Galloway

1993]
b. [tθ′íwə́lmət] ‘be fed up with something/

someone’ [Galloway 1993]3
In addition, suffixes which are larger or smaller than -CVC do

not trigger the vowel change.4
(7) Halq’eméylem
a. /-stəxw/ CAUSATIVE
[s-?ákw’stəxw] ‘to carry it on one’s arm’5
[łélstəxw tə slə́xw əł ‘pull the canoe ashore’
b. /-ət/ ‘control transitive’
[yíx̌wət] ‘take something apart’
[yákwət] ‘break it’
In order fully to understand the pattern, it is necessary to have

a detailed analysis of the metrical structure of morphologically
complex words. However, space limitations preclude a detailed
Page 8 of 33
analysis. One possible approach will be outlined below, but

this has some complications not addressed in the analysis
presented here.
In terms of understanding why reduction occurs, it is

necessary to examine the object agreement paradigms of
words with the different suffixes. The following data illustrate
the object paradigms of different lexemes derived from the
root /lem/ ‘look’, and are from Leslie (1979: 31). (The
underlying representations in (8) are those posited by the
author.) To understand the alternations, a few notes about the
morphological structure are needed. Roots in Salish languages
are generally intransitive (Davis 1996). Transitives are derived
by the addition of a variety of different transitivizing suffixes,
which are then followed by object person markers. The first
lexeme uses a vowel-initial suffix to derive ‘control’ transitive
stems (those that have an agent in control of the action); the
second lexeme has a consonant-initial ‘limited control’ suffix
(indicating that the agent does not have full control over the
activity). There are a few morphologically conditioned
allomorphs in the data at (8) as well.
(p.301)
Page 9 of 33
(8) Hul’qumi’num’ pronominal paradigms, based on root /lem/ ‘look’
a. ‘control’
/lem-ət-am̓š [lèməθám̓š] ‘look at me’
/lem-ət-al̓xw/ [lèmətál̓xw] ‘look at us’
/lem-ət-amə/ [lèməθámə] ‘look at you (sg)’
/lem-ət-alə/ [lèmətálə] ‘look at you (pl)’
/lem-ət-w/ [lémət] ‘look at her, him, it,

them’
b. ‘limited control’
/lem/-n-am̓š/ [lèmnám̓š] ‘see me’
/lem-n-al̓xw/ [lèmnál̓xw] ‘see us’
/lem-n-amə/ [lèmnámə] ‘see you (sg)’
/lem-n-alə/ [lèmnálə] ‘see you (pl)’
/lem-n-əxw/ [lèmnəxw] ‘see him, her,it, them’
Page 10 of 33
The important point is that in (8a), stresses are non-adjacent,

while in (8b) they are adjacent. Schwa reduction occurs with
the latter, and it is proposed that stress adjacency is the
primary trigger. Reducing the full vowel to schwa decreases
the sonority and prominence of the secondary-stressed
syllable, and may be a type of clash avoidance.
Of particular interest to the claims of this chapter is that the

last form in (8b) has primary stress on the root vowel, but no
stress on the third person object suffix /-əxw/. (See Wiltschko
(forthcoming) for evidence that /-əxw/ is third person object
agreement that only occurs with -n ‘limited control’ and -st
‘causative’.) This is a situation where the trigger for reduction
is not present in the word, and so it looks like schwa reduction
overapplies.
In order to account for this, McCarthy’s (Ch. 8 above) Optimal

Paradigms model is adopted. McCarthy’s framework is set
within Prince and Smolensky’s (1993) Optimality Theory, in
which constraints evaluate linguistic forms. There are
generally two types of constraints, which conflict with each
other: markedness constraints, which evaluate the well-
formedness of candidates, and faithfulness constraints, which
evaluate the identity of the candidate forms with some
morphologically related string (the input, another output
word, etc.). McCarthy’s innovation is to propose that all words
in a paradigm are related to each other via a number of
Correspondence relations. (See also Gafos 2003 and other
authors in this volume.) Paradigm uniformity effects are
achieved because those are the most faithful pairings within
the paradigm. McCarthy lists four key components of the
Optimal Paradigms model. First, candidates are not single
words, but rather are sets of words—candidate paradigms.
Second, markedness and regular IO faithfulness constraints
evaluate each paradigm member. The third and fourth points
are related to how correspondence is defined within a
paradigm and are quoted below from McCarthy (Ch. 8 above).
a. The stem ‘in each paradigm member is in a

correspondence relation OP with the stem in every other
paradigm member. (That is, for every candidate
paradigm P there is a relation OP on PHP.)’
(p.302)
Page 11 of 33
b. ‘There is a set of Output–Output faithfulness

constraints on the OP correspondence relation.’
That is, in addition to IO-Faithfulness constraints, there are

also OP-Faithfulness constraints.
The example at (9) illustrates two candidate paradigms of

interest to the analysis. Candidate (9a) is the actual output in
which all forms are as identical as they can be to each other.
Candidate (9b) is the paradigm in which vowel reduction
occurs only in the words that have the trigger; the result is a
non-uniform paradigm. The vertical lines illustrate segments
that stand in correspondence with each other. However, these
representations are simplified, because they do not show how
each form is in correspondence with each other form.
(9) Candidate object agreement paradigms /lem/ ‘look’ + /-n/

‘limited control’
While
candidate (9a)
illustrates
perfect
faithfulness to
the segmental
identity of the
segments in
correspondence, candidate (9b) incurs four violations, because

the 3rd-person object form is different from all four of the
other words. The 1st- and 2nd-person words are identical with
each other and hence do not incur any violations. Because
candidate paradigm (9a) is the actual output, we know that it
is more important to have identity within the paradigm, than it
is to have the root vowel appear in the word.
The analysis requires four constraints: two markedness

constraints and two faithfulness constraints. One markedness
constraint penalizes adjacent stresses with two full vowels.
This is most likely related to a clash avoidance mechanism.
Further work on the metrical structure of Halkomelem would
result in a more detailed and principled (p.303) formalization
of this prohibition; for now it is schematized as in (10a). As is
Page 12 of 33
common in the Salish literature, the symbol ‘A’ is used to refer

to any full vowel, thus excluding schwa from the constraint.
The other markedness constraint penalizes stressed schwa.
(10) Constraints
a. *CÀC.CÁC: adjacent stressed full vowels are not

permitted
b. *ə́: do not have stressed schwa
c. IO-MAX-VFEATWD:
A vowel feature in the input must be present in the

output word
d. OP-MAX-VFEAT-OBJ:
Output vowel features in a correspondence

relation are present in each member of the object
paradigm
The faithfulness constraints require a more detailed

explanation. Following Struijke’s (2000a) model of Word
Faithfulness, IO-MAX-VFEATWD requires all vowel features
present in the input to be present in the output word. Each
individual word comes with an IO-correspondence relation.
Further discussion of this approach will be presented below
when examining the structure of reduplicated words. Because
the input root is /lem/, each word with schwa in the output
would result in a violation of this IO-Faith constraint. Schwa is
considered to be featureless, as assumed by many researchers
(van Oostendorp 1995; Urbanczyk 1996; Shaw et al. 1999).
Finally, OP-MAX-VFEAT-OBJ requires all the vowel features to
be parsed in each of the inflected forms of the object
agreement paradigm. Evidence for why only the object
paradigm is subject to uniformity will be provided later in the
chapter.
The tableau at (11) illustrates the ranking of constraints. The

two highest ranked constraints are the clash avoidance
constraint and the OP-Faith constraint.Candidate (11a) is the
actual output, with schwa reduction applying to all words in
the paradigm. This comes at a cost: there are stressed schwas,
as well as unparsed input vowel features, in all five forms. In
candidate (11b) vowel reduction fails to apply to any of the
words, resulting in fatal violations of the anti-clash constraint.
In candidate (11c) reduction applies to only those words that
Page 13 of 33
have the trigger. It crucially does not apply to the 3rd-person

object form (the last word), and this results in four violations
of OP-Faith constraint, because it differs from all four other
words. This pressure to conform comes from the other
members of the paradigm and is an instance of overapplication
of vowel reduction. McCarthy (Ch. 8 above) refers to this as a
majority-rules effect: ‘the pattern that is most common in a
paradigm serves as an attractor for the others’. The pressure
for one word to conform to the majority of words in a
paradigm occurs because one deviation results in four
violations above (the one word that differs from the four other
members of the paradigm).
Having illustrated how paradigm uniformity occurs to occlude

the root vowel, we now turn our attention to the situation in
which the root vowel emerges unexpectedly.
(p.304)
Page 14 of 33
(11) *CÀC.CÁC, OP-MAX-VFEAT-OBJ>> *ə́, IO-MAX-VFEATWD
Input/lem-n/ *CÀC.CÁC OP-Max-Vfeat-obj *ə́ IO-MAX-VFEATWD
a. ☞ lə̀mnám̓š, ***** *****

w
lə̀mnál̓x
lə̀mnámə
lə̀mnálə
lə́mnəxw
b. lèmnám̓š *!***
w
lèmnál̓x
lèmnàmə
lèmnálə
lémnəxw
c. lə̀mnám̓š *!*** **** ****

w
lèmnál̓x
lèmnámə
lèmnálə
lèmnəxw
Page 15 of 33
11.3 Emergence of the Root Vowel

When words containing a derived stressed schwa are
reduplicated to indicate imperfective aspect, the root vowel
emerges in the reduplicant. Te following examples illustrate
the pattern. Resonant glottalization also accompanies
imperfective reduplication.
Page 16 of 33
(12) Imperfective (Hukari and Peter 1995)
a. /lem-ət/ lémət ‘look at’
/CV-lem-ət/ lél̓əm̓ət ‘looking at; looking after’
b. /lem-n-əxw/ lə́mnəxw ‘see’
/CV-lem-n-əxw/ lél̓əm̓nəxw ‘seeing it’
Page 17 of 33
If an output word serves as the base for reduplication, the

emergence of the root vowel is unexpected. The CV- allomorph
differs from the usual allomorph that occurs with schwa-
voweled stems when forming imperfectives in Halkomelem. As
the data at (13) illustrate, with schwa-voweled stems,
reduplication of schwa, ablaut, or vowel insertion is used as an
exponent of imperfective meaning.
Page 18 of 33
(13) Hul’q’umi’num’ (Hukari and Peter 1995)
a. /cələw̓-ət/ cə́ləw̓t ‘turn it’
/CV-cələw̓-ət/ cə́cələw̓t ‘turning it over’
b. /c̓q̓wə́/ c̓ə́q̓w ‘pierced, shot’
/CV-c̓qw
̓ / c̓əc̓ə́q̓w ‘getting pierced, shot’
c. /c̓q̓w-n-əxw/ c̓əq̓wnexw ‘shoot, pierce (manage

to)’
/ABLAUT- c̓qw
̓ -n-əxw/ c̓aq̓wnəxw ‘shooting it, piercing it
(manage to)’
d. /c̓ət̓q̓w-ətə́/ c̓ə́t̓qw
̓ t ‘grind, pulverize, smash
it’
/ABLAUT-c̓ət̓qw
̓ ət/ c̓at̓q̓wt ‘grinding, pulverizing,
smashing it’
e. /c̓ət̓q̓w-els/ c̓ət̓qw
̓ els ‘grind up’
/ABLAUT-c̓ət̓qw
̓ els/ c̓at̓q̓wəl̓s ‘grinding something up’
f. /cłaqw-ət/ cłaqwt ‘put it’
/EPENTH-cłaqw-ət/ cxłəqwt ‘putting it through’
Page 19 of 33
(p.305) Reduplication of the full vowel is used only with roots

that contain a full vowel (as in (12), above). The exact
conditions for the very rich allomorphy associated with
imperfective meaning are quite complex and a full analysis
cannot be provided here. Suffice it to say that stem shape and
the segmental quality of roots are important predictors of the
allomorphy. (See M. Jones 1976; Leslie 1979; Galloway 1993;
Suttles forthcoming; and Urbanczyk 2000 for descriptions of
the conditions, and Urbanczyk 1999 for an OT analysis of the
main patterns of allomorphy.)
If stem shape alone were the sole indicator of the allomorphy,

we would expect ablaut to occur with all schwa-voweled
words. However, when there is an underlying root full vowel,
CV- prefixing reduplication occurs. This is a clear case where
base priority does not hold, and we have an instance of root
priority.
Because the focus of this chapter is on paradigm uniformity,

where identity between words in a paradigm is at issue, it is
important to consider whether or not words in the aspectual
paradigm illustrate uniformity. McCarthy (Ch. 8 above)
proposes that only words that have inflectional morphology
are in correspondence with each other. (See other chapters in
this volume where it is argued that derivationally related
words stand in a paradigmatic relation.) Imperfective aspect is
assumed to be inflectional in Halkomelem, and so perfective
and imperfective words are in a paradigmatic relation.
The following examples, repeated from above, illustrate the

words that are in a paradigmatic relation. Segments of the
base that are clearly in correspondence are indicated with a
vertical line. The question arises as to which examples
illustrate uniformity in the paradigm. An examination of the
actual segments in correspondence in the following
representation suggests the following: (14a) is non-identity
preserving; there is a vowel feature in the perfective that is
not parsed in the imperfective base. On the other hand, (14b)
shows identity between all the vowels.
(14) Standard OO-Correspondence for reduplicated words
Page 20 of 33
(p.306)
There is a
pervasive
process of
unstressed
vowel
reduction that
drives the
reduction of
the root vowel
to schwa. In
order to
achieve this, a
markedness
constraint
that penalizes
unstressed full vowels is required, represented in (15a) below.
This constraint must be ranked above the OP-Faith constraint
in (10d), because there is non-uniformity among
corresponding vowels. In order to arrive at the correct output,
it must be the case that the paradigmatic faithfulness
constraint evaluates only words in the aspectual paradigm.
This constraint is represented in (15b) below with the
additional -ASP ending.
(15) a. v˘:
Unstressed full vowels are not permitted
b. OP-MAX-VFEAT-ASP:
Output vowel features in a correspondence

relation are present in each member of the
aspectual paradigm
If the same OP-Faith constraint were used for both object

agreement and the aspectual paradigms, then the optimal
paradigm would be as in (16).
(16) Overapplication of vowel reduction
Page 21 of 33
This paradigm
would be
optimal
because IO-
MAX-VFEAT is
ranked below
the OP-Faith
constraint. The ranking established thus far is repeated below.
Only those candidates with unstressed vowel reduction are
included. The symbol indicates that the incorrect candidate is
chosen as optimal.
(17)
Unstressed
vowel
reduction
occurs in
each of the
candidates. If it is assumed that there is only one OP-Faith
constraint evaluating all the words in a paradigm, then
candidate (17a) is ruled out because it violates it. Candidate
(17b), in which vowel reduction (p.307) overapplies, would
be incorrectly selected as optimal. Candidate paradigms
containing words like * [ləl̓ém̓ət]are not included because all
imperfectives have initial stress (Urbanczyk 2000). The
constraint which demands initial stress on imperfective words
would be violated fatally by all such candidates.
Work by Spaelti (1997) and Struijke (1998, 2000) on

reduplication provides evidence that input segments are in
correspondence with both the segments of the base and the
reduplicant. Thus in (17a), the input /e/ is in correspondence
with both reduplicant /é/ and the base /ə/. This is known as
BROAD-IO FAITH; it accounts for why candidate(17a) obeys
IO-MAX-VFEATWD (as long as the input root vowel /e/ is parsed
somewhere in the word, it satisfies IO-MAX-VFEATWD) while
candidate (17b) violates it (the input /e/ has no correspondent
in base OR reduplicant of the perfective word). One possible
approach to rule out the paradigm in (17b) would be to extend
BROAD-FAITH to the Output–Output domain. Such an
assumption would entail that OP-MAX-VFEAT be obeyed by
both candidates.
Page 22 of 33
However, even if this Broad-Faith approach were extended to

the Optimal Paradigms model, the identity problem would
arise in the words with the derived schwa. The tableau (18)
illustrates that in the optimal candidate (18a), a lack of
identity holds between the root vowel of the perfective and the
reduplicant vowel of the imperfective. However, if one were to
assume the standard approach to Output–Output
correspondence (in which only segments of the base are in
correspondence), the correct candidate is selected.
(18)
If Broad-Faith
were assumed
for the
candidates
above,
candidate
(18b) would
be optimal.
Candidate
(18a) would
incur a
violation because there is a vowel feature in the imperfective
that is not parsed in the perfective. (I am considering only
candidates with schwa in the perfective words, because that
can be derived by pressure from the majority-rules effect of
the other members of the object paradigm. The other members
of the paradigm are excluded to save space.)
(p.308) This pattern, in which identity is obeyed in the

paradigm of one lexeme, while it is violated in another, results
in a ranking paradox with OP-MAX-VFEAT-OBJ. The paradox is
independent of the assumptions regarding which segments
stand in correspondence in a reduplicative paradigm.6 This is
precisely the type of evidence that is used to motivate
separate Faithfulness constraints for different morphologically
related strings (as in Beckman 1997; Urbanczyk 1996).
Therefore, it is expected that separate OP-Faith constraint can
hold for different morphologically related strings: for example,
in the aspectual paradigm v. the object agreement paradigm.7
Because the aspectual paradigm is non-identical and the root

vowel emerges, it must be the case that OP-MAX-VFEAT-ASP is
ranked below IO-MAX-VFEATWD. Tableau (19) illustrates the
Page 23 of 33
final ranking of constraints, incorporating the new markedness

and faithfulness constraints into the hierarchy, and contains
the same candidates as tableau (17). Because the paradigm is
for aspectual features, the OP-Faith constraint for object
agreement is not relevant.
(19)
Candidate
(19a) is
optimal, as it
does not
violate any of
the
markedness
constraints
and the only
faithfulness
constraint it
violates is one on paradigm uniformity for aspect. Candidate
(19b) is entirely faithful, but it contains an unstressed full
vowel. Candidate (19c), which illustrates identical vowels in
the correspondent bases, is ruled out (p.309) because it has
a stressed schwa. Therefore, the constraint ranking correctly
eliminates ill-formed paradigms.
(20)
Tableau (20)
illustrates
that this same
ranking will
derive the
correct
results for the
paradigm in
which
stressed
schwa is
derived. In this case, all candidates have at least one stressed
schwa. Recall that this is compelled by high-ranking *CÀC.CÁC
in the object agreement paradigm. In the interests of saving
(possibly three-dimensional) space, I leave aside how the
entire system works. All candidates in (20) have the stressed
schwa derived by the ranking established above and so each
Page 24 of 33
violate IO-MAX-VFEATWD at least once. Candidate (20a) is

selected as optimal; it has only one stressed schwa. It also
expresses an input vowel in one word and it maintains vocalic
identity within the aspectual paradigm. Candidate (20b) is
ruled out because it has an unstressed full vowel; this is not
permitted anywhere in the language. Finally, candidate (20c)
is ruled out because it has two stressed schwas. Notice that it
also fails to parse the input vowel in both words. It can be
ruled out by either one of these constraints.
Root-priority satisfies two key constraints better than having a

uniform paradigm. It minimizes the number of stressed schwas
in the paradigm, and expresses the root vowel. The final
section of this chapter provides a brief discussion of how a
derivational theory would account for the facts.
11.4 Comparison with Derivational Approach

The first point to make is that the overapplication of vowel
reduction (creating a stressed schwa) with 3rd-person object
agreement is straightforwardly captured with the Optimal
Paradigms framework. It could not be achieved in a
derivational model (p.310) without a morpheme-specific rule.
While the phonological context for vowel reduction occurs
with the object agreement markers (adjacent stressed full
vowels), such a context does not arise with the limited.control
+ 3OBJ combination /-n-əxw/. (21) illustrates that the first
stage would be to assign stress. The second would be to apply
the reduction of the full vowel (presumably to avoid clash).
This only applies to those forms with adjacent stressed vowels,
as with the middle form at (21).
Page 25 of 33
(21) /lem-ət-alə/ /lem-n-alə/ /lem-n-əxw/
stress lèmətálə lèmnálə lémnəxw
À→ə̀ DNA lə̀mnálə DNA
Á→ə́ DNA DNA lə́mnəxw
lèmətálə lə̀mnálə lə̀mnəxw
‘look at you (pl.)’ ‘see you (pl.)’ ‘see him/her/it’
Page 26 of 33
A rule reducing the full vowel to schwa would then apply to all
3rd-person object forms. The fact that the reduction applies to
[lə́mnəxw] is an accident, and is not related to the observation
that it forms part of the object agreement paradigm. One
might just as easily expect it not to occur, or to occur in the
absence of any other reduction rule (such as that found for the
other object agreement).
One potential argument in favor of the derivational approach

is that schwa reduction is also triggered by lexical suffixes
that have the shape -CVC. The data at (22) illustrate that the
suffix for ‘hand’ triggers vowel reduction, as can be seen in
(22d). The derivational approach could adopt a more general
rule in which all -CVC shaped suffixes trigger vowel reduction.
(22) Hul’q’umi’num’
a. łic̓əθám̓š ‘cut me’
b. łə̀c̓nám̓š ‘accidentally cut me’
c. łə́c̓nəxw ‘accidentally cut him/her/it’
d. łə́c̓cəs ‘cut hand’
However, the situation is not as straightforward as it seems.

One would still need two separate reduction rules because the
2nd-person object agreement suffixes are vowel-final;
alternatively the rule could be extended to the context of a
following -CVCV shape. Second, lexical suffixes function as
3rd-person objects. As such, they may enter into some form of
paradigmatic relation with the 1st- and 2nd-person object
agreement. The difference between the 3rd-person object
marker and a lexical suffix is that the referent is more specific
than the latter, often referring to a body part. Observe also
that the prosodic shapes of (21c, d) are identical. All words
with a 3rd-person object referent have a primary stressed
schwa, followed by an unstressed schwa: the shared
morphosyntactic feature of 3rd person has a shared word-
shape. Finally, the 3rd-person object agreement and some of
the lexical suffixes (but not ‘hand’) surface with full vowels
when attached to some schwa-voweled roots, as can be seen
below. The pattern is also associated with stress on the second
syllable (although all the forms have yet to be confirmed). Not
all schwa-voweled roots permit the full (p.311) vowel to
Page 27 of 33
appear with the suffix, so it is more complex than represented

here.
(23) Hul’q’umi’num’ (Hukari and Peter 1995; stress marks

added by Bianco)
a. c̓əq̓wnéxw ‘shoot, pierce’
k̓wəłnéxw ‘spill it (liquid)’
k̓wəšnèxw ‘count it (manage to)’
ləkwnéxw ‘break it (stick) accidentally, manage

to’
b. sc̓əq̓wšén ‘sliver (what has pierced the foot)’
ləkwšén ‘break foot or leg’
ləmxwšénəm‘stomp feet on ground or floor’
θəq̓šén ‘poked or stabbed in the foot’
Clearly there is more work needed to sort out the patterns of

vowel reduction. The benefit of the paradigmatic approach is
that it claims that there is one general explanation for why
reduction would extend to other forms. Clash avoidance
causes vowel reduction; the word sets that do not have the
context for clash share the morphosyntactic feature of 3rd-
person object agreement. The creation of a marked segment in
Halkomelem has a trigger in several paradigmatically related
words sharing object agreement; it is not unexpected that this
be extended to other paradigmatically related words. The
derivational approach makes a different claim. Residual cases
of vowel reduction do not have anything in common other than
being consonant-initial. Represented as two separate
processes, we might expect a language to exist that has one
process but not the other. For example, it is predicted that
there could be a language that reduces a full vowel to schwa
when followed by -CVC shaped suffixes, but allows adjacent
stressed full vowels.
The second point to make is that while it is possible to derive

the emergence of the root vowel utilizing a derivational, step-
by-step approach, it does not offer the explanatory power that
the OT approach does. In a derivational approach, the rules for
vowel reduction must be ordered after reduplication. Table
(24) illustrates the order of rule application for 3rd-person
Page 28 of 33
object forms. The last rule to apply is unstressed vowel

reduction.
Page 29 of 33
(24) /lem-ət/ /CV-lem-ət/ /lem-n-əxw/ /CV-lem-n-əxw/
reduplication DNA lelemət DNA lelemnəxw
Stress lémət lélemət lémnəxw lélemnəxw
À→ə̀ DNA DNA DNA DNA
Á→ə́ DNA DNA lə́mnəxw DNA
v̆→ə DNA léləmət DNA léləmnəxw
lémət léləmət lə́mnəxw léləmnəxw
‘look at him’ ‘looking at him’ ‘see him’ ‘seeing him/her/ it’
Page 30 of 33
The derivational approach entails that reduplication occurs

before any other rules. If the order were reversed, with
reduplication after the reduction rules, then the incorrect
result would obtain.
(p.312) If we compare this with the OT approach, there is no

empirical difference. However, there is a conceptual difference
in that the paradigm does not play a role in derivational
models of morphology and phonology. Any morphological or
phonological rules that apply do so in a purely random, ad hoc
fashion. Root priority emerges only because reduplication
applies first. It is an epiphenomenon. In the OT approach, the
root vowel emerges under the demands of two high-ranking
constraints: to avoid stressed schwa, and to be faithful to the
input specifications of the root. In the OP model, the paradigm
plays a key role in determining when processes overapply.
11.5 Conclusion
One of the implications of this short study is that base-priority
does not straightforwardly result in predicting the correct
surface form. The analysis presented here offers support to
McCarthy (Ch. 8 above) who proposes that there is no base
priority in inflectional paradigms. It is not surprising
therefore, that root-priority can emerge from time to time,
supplanting derived properties in paradigmatically related
word forms. Root priority takes precedence over paradigmatic
uniformity, precisely when certain phonotactic considerations
are not relevant in the expression of some inflectional
morpheme. Reference to the input root is clearly a requisite in
deriving the correct imperfective stems. This follows naturally
from OP, where IO-Faith and OP-Faith interact. When OP-Faith
fails, IO-Faith emerges, thus deriving priority of the root.
Furthermore, it was shown that paradigmatic identity is
sensitive to the morphosyntactic information that is being
conveyed. When object agreement and aspectual marking are
morphologically distinct, there are two distinct OP-Faith
constraints: one demanding identity in the object agreement
paradigm, the other forgoing identity in the aspectual
paradigm.
Notes:
(1) I have regularized the transcriptions of data from various
sources to be consistent. They are presented phonemically
using the APA, in which the symbols [c, ƛ, č, š, x̌, y] correspond
Page 31 of 33
to IPA [ts, tɬ, t∫, ∫, χ, j]. Other symbols and diacritics are the
same in both systems. The surface forms abstract away from
the differing qualities of schwa, representing them as [ə].
(2) Halq’eméylem has a pitch accent system with mid-level

pitch being represented by [‵].
(3) This form has two main stresses, as shown here, in

Galloway’s grammar. It is unclear to me why, so I am quoting
the form as given in the source.
(4) The Hul’q’umi’num’ Dictionary lists many causatives with

reduced root vowels, so there is some interesting dialectal
variation occurring. The variation also extends to the idiolectal
level, as Tom Hukari (p.c.) has pointed out that some speakers
of Hul’q’umi’num’ vary in reducing the vowel with the
causative morpheme.
(5) The initial /s/ is a prefix, and is not considered part of the
root. Even if it is considered part of the stem, the result is an
initial cluster and may not be problematic.
(6) As such the pattern here does not provide evidence for or
against extending Broad-Faith to Output-Output
correspondence.
(7) The option of having separate OP-Faith constraints is also

raised by McCarthy (Ch. 8 above). It should be pointed out that
the data that McCarthy examines does not require
differentiating paradigmatic identity between the different
inflectional meanings. That is to say, person, number, gender,
and aspect are expressed cumulatively in the Arabic affixes; in
Halkomelem, they are entirely distinct. It is interesting to note
that Gafos’s (2003) analysis of Arabic paradigms assumes a
separate correspondence relation for the perfect-imperfect
paradigms.
Page 32 of 33
Page 33 of 33
References


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
(p.313) References
Bibliography references:
ADAM, G. (2002). ‘From Variable to Optimal Grammar:

Evidence from Language Acquisition and Language Change’.
Ph.D. dissertation, Tel-Aviv University. [Available at ROA
567-120.]
—— and BAT-EL, O. (2000). ‘Morphological Knowledge without

Morphological Structure: The Emergence of Inflectional
Suffixes in the Acquisition of Hebrew Verbs’, Paper presented
in a workshop on Phonology and Prosodic Morphology of
Semitic Languages.Tel-Aviv University.
AGUERO-BAUTISTA, C. (1998). ‘Cyclic and Identity Effects in

Spanish Diminutives and Augmentatives’, unpublished MS.
ALBRIGHT, A. (2002). ‘Base Selection in Analogical Change: A

German/Yiddish Comparison’, BLS 28: 1–13.
—— and HAYES, B.(2002) ‘Modeling English Past Tense

Intuitions with Minimal Generalization’,in M.Maxwell (ed.),
Proceedings of the 6th Meeting of the ACL Special Interest
Group in Computational Phonology, Philadelphia, July 2002.
ACL, 58–69.
Page 1 of 33
References
ALDERETE, J. D. (1999). ‘Morphologically Governed Accent in

Optimality Theory’. Ph.D. dissertation, University of
Massachusetts at Amherst. [Published 2001 New York:
Routledge.]
—— (2001). ‘Dominance Effects as Transderivational Anti-

Faithfulness’, Phonology 18: 201–53.
——BECKMAN, J. BENUA, L. GNANADESIKAN, A.

MCCARTHY, J. J. and URBANCZYK, S. (1999) ‘Reduplication
with Fixed Segmentism’, Linguistic Inquiry 30: 327–64. [Also
available at ROA-226.]
ALI ADRA, M. (1999). ‘Identity Effects and Opacity in Syrian

Arabic: An Optimality Theory Analysis’. Ph.D. dissertation,
University of Illinois at Urbana.
ALLEN, J. H. (1903). Allen and Greenough’s New Latin

Grammar for Schools and Colleges. Boston: Ginn & Co.
ANDERSON, S. R. (1975). ‘On the Interaction of Phonological

Rules of Various Types’, Journal of Linguistics 11: 39–62.
—— (1992). A-morphous Morphology. Cambridge: Cambridge

University Press.
ANSHEN, F., and ARONOFF, M. (1988). ‘Producing

Morphologically Complex Words’, Linguistics 26: 641–55.
ARONOFF, M. (1976). Word Formation in Generative

Grammar. Cambridge, Mass.: MIT Press
—— and SRIDHAR, S. N. (1983). ‘Morphological Levels in

English or Kannada; or, Atarizing Reagan’, CLS 19:3–16.
AVANESOV, R. I. (1964). Proiznoshenie russkogo literaturnogo

jazyka. Moscow: Prosveshchenie.
(p.314) BALDI, P. (1999). The Foundation of Latin. Berlin:

Mouton de Gruyter.
BARR, R.(1994) ‘A Lexical Model of Morphological Change

Ph.D. dissertation, Harvard University.
BASTIN, Y. (1983). La Finale -ide et l’imbrication en Bantou.

Tervuren: Musée Royale de l’Afrique Centrale.
Page 2 of 33
References
—— (1986). ‘Les Suffixes causatifs dans les langues bantoues’,

Africana Linguistica 10: 56–145.
BAT-EL, O. (2002). ‘True Truncation in Colloquial Hebrew

Imperatives’, Language 78: 1–33.
—— (2003). ‘Anti-Faithfulness: An Inherent Morphological

Property’, in J. Lecarme (ed.) Research in Afroasiatic
Grammar. Amsterdam: Benjamins. 21–34.
BECKER, T. (1998). Das Vokalsystem der deutschen

Standardsprache. Frankfurt: Lang.
BECKMAN, J. N. (1997). ‘Positional Faithfulness’. Ph.D.

dissertation, University of Massachusetts at Amherst.
[Published 2000. New York: Garland.]
(1990). ‘Moroccan Arabic Syllable Structure’, Langues et

littératures 8:177–91.
BENUA, L. (1995). ‘Identity Effects in Morphological

Truncation’, University of Massachusetts Occasional Papers in
Linguistics 18: 77–136.
—— (1997a). ‘Transderivational Identity: Phonological

Relations Between Words’. Ph.D. dissertation, University of
Massachusetts at Amherst. [Published 2000 as Phonological
Relations Between Words. New York: Garland. Also available
at ROA- 259.
—— (1997b).‘Affixes Classes are Derived by Faithfulness’,

University of Maryland Working Papers in Linguistics 5:1–26.
BERMÚDEZ-OTERO, R. (1999). ‘Constraint Interaction in

Language Change: Quantity in English and Germanic’. Ph.D.
dissertation, University of Manchester.
BIANCO, V. (1996). ‘The Role of Sonority in the Prosody of

Cowichan’. MA thesis, University of Victoria.
BLACK, H. A. (1993). ‘Constraint-Ranked Derivation: A Serial

Approach to Optimization’. Ph.D. dissertation, University of
California, Santa Cruz.
BLOOMFIELD, L. (1933). Language. New York: Holt.
Page 3 of 33
References
BOBALJIK, J. (2002). ‘Paradigmaticity without Paradigms: Rich

Agreement and Paradigm Uniformity’. Handout. Rutgers
University, New Brunswick, New Jersey.
BOERSMA, P. (1998). Functional Phonology. The Hague:

Holland Academic Graphics.
—— and HAYES, B. (2001). ‘Empirical Tests of the Gradual

Learning Algorithm’, Linguistic Inquiry 32.1: 45–86.
BOLOZKY, S. (1979). ‘On the New Imperative in Colloquial

Hebrew’, Hebrew Annual Review 3: 17–24.
BONET, E. (1991). ‘Morphology After Syntax’. Ph.D.

dissertation, MIT.
—— and LLORET, M.-R. (2001). ‘OCP Effects in Catalan

Cliticization’. Barcelona: Grup de Gramàtica Teòrica,
Departament de Filologia Catalana,Universitat Autònoma de
Barcelona.
BOOIJ, G. (1977).Dutch Morphology. A Study of Word

Formation in Generative Grammar. Dordrecht: Foris:
—— (1983). ‘Principles and Parameters in Prosodic

Phonology’, Linguistics 21:249 –80.
—— (1995). The Phonology of Dutch. Oxford: Clarendon Press.
—— (1996). ‘Lexical Phonology and the Derivational Residue’

in J. Durand and B. Laks (eds.),(1996), 69–96.
—— and LIEBER, R. (1993). ‘On the Simultaneity of

Morphological and Prosodic Structure’, in S. Hargus and E. M.
Kaisse (eds.), Studies in Lexical Phonology. Phonetics and
Phonology, iv. San Diego: Academic Press, 23–44.
(p.315) —— and RUBACH, J. (1984). ‘Morphological and

Prosodic Domains in Lexical Phonology’, Phonology Yearbook
1: 1–27.
BOROWSKY, T. (1986). ‘Topics in the Lexical Phonology of

English’. Ph.D. dissertation, University of Massachusetts at
Amherst.
—— (1989). ‘Structure Preservation and the Syllable Coda in

English’, Natural Language and Linguistic Theory 7: 145–66.
Page 4 of 33
References
BOUDLAL, A. (2001). ‘Constraint Interaction in the Phonology

and Morphology of Casablanca Moroccan Arabic’. Ph.D.
dissertation, Mohammed V University, Rabat, Morocco.
BRAME, M. (1970).‘ Arabic Phonology: Implications for

Phonological Theory and General Semitic’. Ph.D. dissertation,
MIT.
—— (1974). ‘The Cycle in Phonology: Stress in Palestinian,

Maltese and Spanish’, Linguistic Inquiry 5: 39–60.
BROSELOW, E. (1976). ‘The Phonology of Egyptian Arabic’.

Ph.D. dissertation, University of Massachusetts at Amherst.
BUCKLEY, E. (1999). ‘Uniformity in Extended Paradigms’ in, B.

Hermans and M. van Oosten-dorp (eds.), (1999), 81–104.
—— (2000). ‘What Should Phonology Explain?’ Handout. State

University of New York, Buffalo.
BULLOCK, B. (2004). ‘The Phonological Mediation of

Phonological Complexity: Verb Stem Leveling in the History of
French’, Probus 16: 1–19.
BURZIO, L.(1991). ‘On The Metrical Unity of Latinate Affixes’,

ESCOL 8: 1–22.
[Reprinted in Rivista di Grammatica Generativa 16: 1–27.

Revised version in H. Campos and P. M. Kemp-chinsky (eds.)
(1995),
Evolution and Revolution in Linguistic Theory: Essays in

Honor of Carlos Otero.Washington, DC: Georgetown
University Press.]
—— (1992) ‘Principles in Phonology’,in E. Fava (ed.),

Proceedings of the XVII Meeting on Generative Grammar
(Trieste, February 22–4, 1991). Turin: Rosenberg and Sellier,
97–119.
—— (1993). ‘English Stress, Vowel Length and Modularity’,

Journal of Linguistics 29.2: 359–418.
—— (1994a). Principles of English Stress. Cambridge:

Cambridge University Press.
Page 5 of 33
References
—— (1994b). ‘Metrical Consistency’,in E. S. Ristad (ed.),

Language Computations. Providence, RI: American
Mathematical Society, 95–125.
—— (1996). ‘Surface Constraints versus Underlying

Representation’,in J. Durand and B. Laks (eds.), (1996), 123–
41.
—— (1997) ‘Strength in Numbers’, University of Maryland

Working Papers in Linguistics 5: 27– 52.
—— (1998). ‘Multiple Correspondence’, Lingua 103: 79–109.
—— (2000a). ‘Cycles, Non-Derived-Environment Blocking, and

Correspondence’,in J. Dekkers, van der Leeuw and J.van de
Weijer (eds.), Optimality Theory: Syntax, Phonology, and
Acquisition. Oxford: Oxford University Press, 47–87.
—— (2000b). ‘Segmental Contrast Meets Output-to-Output

Faithfulness’, The Linguistic Review 17.2–4: 368–84.
—— (2001). ‘Zero Derivations’, Linguistic Inquiry 32.4: 658–

77.
—— (2002a). ‘Missing Players: Phonology and the Past-tense

Debate’, Lingua 112: 157–99.
—— (2002b). ‘Surface-to-Surface Morphology: When your

Representations Turn into Constraints’, in P. Boucher in (ed.),
Many Morphologies. Somerville, Mass.: Cascadilla Press, 142–
77. [Available at ROA-341.]
—— (2003). ‘Output-to-Output Faithfulness: The Italian

Connection’, Lingue e Linguaggio 1: 69–104.
(p.316) BURZIO, L. (in press). ‘Paradigmatic and

Syntagmatic Relations in Italian Verbal Inflection’, in J. Auger,
J. C. Clements, and B. Vance (eds.), Selected Proceedings of
the 33rd Linguistic Symposium on Romance Languages.
Amsterdam: Benjamins.
BYBEE, J. (1985). Morphology: A Study of the Relation

between Meaning and Form. Amsterdam: Benjamins.
—— (1988). ‘Morphology as Lexical Organization’,in M.

Hammond and M. Noonan (eds.), Theoretical Morphology, San
Diego: Academic Press, 119–41.
Page 6 of 33
References
—— (1995). ‘Regular Morphology and the Lexicon’, Language

and Cognitive Processes, 10.5: 425–55.
—— (1996). ‘Productivity, Regularity and Fusion: How

Language Use Affects the Lexicon’,in R. Singh (ed.),
Trubetzkoy’s Orphan. Amsterdam Studies in the Theory and
History of Linguistic Science 144. Amsterdam: Benjamins,
247–69.
BYNON, T. (1977). Historical Linguistics. Cambridge:

CALABRESE, A. (1995). ‘A Constraint-Based Theory of

Phonological Markedness and Simplification Procedures’,
Linguistic Inquiry 26: 373–464.
CANTINEAU, J. (1939). ‘Remarques sur les parlers de

sédentaires syro-libano-palestiniens’, Bulletin de la Sociéte de
Linguistique de Paris 40: 80–8.
CARLSON, K. (1998). ‘Reduplication and Sonority in Nakanai

and Nuxalk’, in J. Austin and A. Lawson (eds.), Proceedings of
the Fourteenth Eastern States Conference on Linguistics ’97,
Ithaca, NY: Cornell Linguistics Circle,23 –33. [Also available at
ROA-230.]
CARSTAIRS, A. (1987). Allomorphy in Inflection. London:

Croom Helm.
CHOMSKY, N. (1995). The Minimalist Program. Cambridge,

Mass.: MIT Press.
—— and HALLE, M. (1968). The Sound Pattern of English.

New York: Harper and Row.
CHUNG, C. W. (1999). ‘Reduplication in Korean’. Ph.D.

dissertation, Indiana University.
CHUNG, S. (1983). ‘Transderivational Relations in Chamorro

Phonology’, Language 59: 35–66.
CLEMENTS, G. N. (1985). ‘The Geometry of Phonological

Features’, Phonology Yearbook 2: 223–52.
COETZEE, A. (1999a). ‘Hebrew and Aramaic Segholation and

the Generality and Ordering of Phonological Rules’, Journal of
Semitic Studies 44: 215–25.
Page 7 of 33
References
—— (1999b). Tiberian Hebrew Phonology: Focusing on

Consonant Clusters. Assen: Van Gorcum.
COHN, A., and MCCARTHY, J. J. (1994/1998). ‘Alignment and

Parallelism in Indonesian Phonology’, Working Papers of the
Cornell Phonetics Laboratory 12: 53–137. [Also available at
ROA-25.]
COLE, J. (1995). ‘The Cycle in Phonology’, in J. Goldsmith(ed.),

The Handbook of Phonological Theory. Cambridge, Mass.:
Blackwell, 70–113.
CÔTÉM.-H (2000). ‘Consonant Cluster Phonotactics: A

Perceptual Approach’. Ph.D. dissertation, MIT.
COWELL, M. (1964). A Reference Grammar of Syrian Arabic.

Washington, DC: Georgetown University Press.
CROSSWHITE, K. (1997). ‘Intra-Paradigmatic Homophony

Avoidance in Two Dialects of Slavic’, unpublished MS, UCLA.
—— (1998). ‘Segmental vs. Prosodic Correspondence in

Chamorro’, Phonology 15: 281–316.
CROWHURST, M. J. (1992). ‘Diminutives and Augmentatives in

Mexican Spanish: A Prosodic Analysis’, Phonology 9: 221–53.
CRYSTAL, D. (1985). A Dictionary of Linguistics and Phonetics.

Oxford: Blackwell.
CURTIUS, G. (1885). Zur Kritik der neuesten Sprachforschung.

Leipzig: Hirzel.
CZAYKOWSKA-HIGGINS, E. (1996). What’s in a Word? Word

structure in Moses-Columbia Salish (Nxaʔamxcín). Winnipeg:
Voices of Rupert’s Land.
(p.317) —— (1998). ‘The Morphological and Phonological

Constituent Structure of Words in Moses-Columbia Salish
(Nxaʔamxcín)‘,in E. Czaykowska-Higgins and M. Dale Kinkade
(eds.), Salish Languages and Linguistics: Theoretical and
Descriptive Perspectives, Berlin: Mouton, 153–95.
DAVIS, H. (1996). ‘Deep unaccusativity and zero syntax in

St’át’imcets’. Papers from the 31st International Conference
on Salish and Neighbouring Languages, 39–58.
Page 8 of 33
References
DAVIS, S., and Cho, M.-H.(2003). ‘The Distribution of

Aspirated Stops and /h/ in American English and Korean: An
Alignment Approach with Typological Implications’, Linguistics
41:607 –52.
DELATTE, L., EVRARD, E., GOVAERTS, S. and DENOOZ, J.

(1981). Dictionnaire fréquentiel et Index inverse de la langue
latine. Université de Liège: Laboratoire d’analyse statistique
des langues anciennes.
DIXON, R. (1977). ‘Some Phonological Rules in Yidiny’,

Linguistic Inquiry 8.1:1–34.
DOWNING, L. J. (1996). The Tonal Phonology of Jita. lincom

Studies in African Linguistics 5. Munich: lincom europa.
—— (1999a). ‘Prosodic Stem ≠ Prosodic Word in Bantu’, in T.

A. Hall and U. Kleinhenz (eds.), Studies on the Phonological
Word. Amsterdam: Benjamins, 73–98.
—— (1999b). ‘Verbal Reduplication in Three Bantu

Languages’,in R. Kager, H. van der Hulst, and W. Zonneveld
(eds.), The Prosody–Morphology Interface. Cambridge:
Cambridge University Press, 62–89.
—— (1999c). ‘Morphological Constraints on Bantu

Reduplication’, Linguistic Analysis 29: 6– 46.
—— (2000). ‘Satisfying Minimality in Ndebele’, ZAS Working

Papers in Linguistics 19: 23–39.
—— (2001). ‘Liquid Spirantisation in Jita’, Malilime: Malawian

Journal of Linguistics 2:1–27.
DURAND, J., and LAKS, B. (eds.) (1996). Current Trends in

Phonology: Models and Methods. Manchester: European
Studies Research Institute, University of Salford.
ELORDIETA, G., and CARREIRA, M. (1996). ‘An Optimality

Theoretic Analysis of Spanish Diminutives’, CLS 32: 49–60.
EMBICK, D., and NOYER, R. (2001). ‘Movement Operations

after Syntax’, Linguistic Inquiry 32.4: 555–95.
FERGUSON, C. A. (1972). ‘Short a’ in Philadelphia English’,in

M. E. Smith (ed.), Studies in Linguistics in Honour of George
L. Trager. The Hague: Mouton.
Page 9 of 33
References
FIDELHOLTZ, J. (1967). ‘English Vowel Reduction’,

unpublished MS, MIT.
FLEMMING, E. (1995). ‘Auditory Representations in

Phonology’. Ph.D. dissertation, UCLA.
—— and KENSTOWICZ, M. (1995). Base-Identity and Uniform

Exponence: Alternative to Cyclicity. MS, MIT.
FRIEDERICI, A. D., and WESSELS, J. E. (1993). ‘Phonotactic

Knowledge of Word Boundaries and its Use in Infant Speech
Perception’, Perception and Psychophysics 54: 287–95.
FUKAZAWA, H. (1998). ‘Theoretical Implications of OCP

Effects on Features in Optimality Theory’. Ph.D. dissertation,
University of Maryland at College Park.
GAFOS, A. (1998). ‘A-Templatic Reduplication’, Linguistic

Inquiry 29: 515–27.
—— (2001). ‘The Initial State and Verbal Stems in Arabic’,

unpublished MS, New York University.
—— (2002). ‘A Grammar of Gestural Coordination’, Natural

Language and Linguistic Theory 20: 269–337.
—— (2003). ‘Greenberg’s Asymmetry in Arabic: A

Consequence of Stems in Paradigms’, Language 79: 317–55.
(p.318) GALLOWAY, B. (1993). A Grammar of Upriver

Halkomelem. University Publications in Linguistics 96.
Berkeley: University of California Press.
GANONG, W. F. (1980). ‘Phonetic Categorization in Auditory

Word Perception’, Journal of Experimental Psychology: Human
Perception and Performance 6: 110–25.
GARR, W. R. (1989). ‘The Seghol and Segholation in Hebrew’,

Journal of Near Eastern Studies 48: 109–16.
GESENIUS, W. (1910). Gesenius’ Hebrew grammar, as edited

and enlarged by the late E. Kautzsche. With a facsimile of the
Siloam inscription by J. Euting. and a table of alphabets by M.
Lidzbarski. Oxford: Clarendon Press. [2nd Eng. edn. revised in
accordance with the 28th German edn. (1909) by A. E. Cowley.
Repr. 1974 from corrected sheets of the 2nd edn.]
Page 10 of 33
References
GIEGERICH, H. (1985). Metrical Phonology and Phonological

Structure: German and English. Cambridge: Cambridge
University Press.
—— (1999). Lexical Strata in English: Morphological Causes,

Phonological Effects. Cambridge: Cambridge University Press.
GOLDSMITH, J. (1993). ‘Harmonic Phonology’,in J. Goldsmith

(ed.), The Last Phonological Rule. Chicago: University of
Chicago Press, 21–60.
GOUGENHEIM, G. (1935). Elements de phonologie française.

Paris: Les Belles Lettres.
GOUSKOVA, M. (2003). ‘Economy of Representation in

Optimality Theory’. Ph.D. dissertation, University of
Massachusetts at Amherst.
GUSSENHOVEN, C. (2000). ‘The Origin and Development of

the Central Franconian Tone Contrast’,in A. Lahiri (ed.),
Analogy, Levelling, and Markedness. Berlin: Mouton, 215–60.
GUSSMANN, E. (1980). Studies in Abstract Phonology.

Cambridge, Mass.: MIT Press.
HADDAD, G. (1983). ‘Problems and Issues in the Phonology of

Lebanese Arabic’. Ph.D. dissertation, University of Illinois at
Urbana.
HALE, M., and KISSOCK, M. (1998). ‘The Phonology–Syntax

Interface in Rotuman’, UCLA Occasional Papers in Linguistics
21:115–28.
—— —— and REISS, C. (1998). ‘Output–Output

Correspondence in Optimality Theory’, WCCFL 16: 223–36.
[Also available at ROA-202.]
HALL, T. A. (1998). ‘A Note on Secondary Stress in German

Prosodic Morphology’, Linguistische Berichte 174: 414–24.
—— (1999). ‘Phonotactics and the Prosodic Structure of

German Function Words’,in T. A. Hall and U. Kleinhenz (eds.),
Studies on the Phonological Word. Amsterdam: Benjamins, 99–
131.
—— (2000). ‘Typological Generalizations Concerning

Secondary Palatalization’, Lingua 110: 1–25.
Page 11 of 33
References
—— (2001). ‘The Distribution of Superheavy Syllables in

Modern English’, Folia Linguistica 35: 399–422.
—— (2002). ‘Against Extrasyllabic Consonants in German and

English’, Phonology 19: 33–75.
HALLE, M. (1997). ‘On Stress and Accent in Indo-European’,

Language 73: 275–313.
—— and MARANTZ, A. (1993). ‘Distributed Morphology’, in K.

Hale and S. J. Keyser (eds.), The View from Building 20: Essays
in Linguistics in Honor of Sylvain Bromberger. Cambridge,
Mass.: MIT Press, 111–76.
—— —— (1994). ‘Some Key Features of Distributed

Morphology’, MIT Working Papers in Linguistics 21: 275–88.
HAMMOND, M. (1995). ‘There is no Lexicon!’ [Available at

ROA 43-0195.]
—— (1998). ‘Lexical Frequency and Rhythm’, in M. Darnell, E.

Moravcsik, F. Newmeyer, M. Noonan, and K. Wheatley (eds.),
Functionalism and Formalism in Linguistics. Amsterdam:
Benjamins, 331–58.
(p.319) —— (1999). The Phonology of English. A Prosodic

Optimality-Theoretic Approach. Oxford: Oxford University
Press.
HANNAHS, S. J. (1995). Prosodic Structure and French

Morphophonology. Tübingen: Niemeyer.
HARGUS, S. (1985). ‘The Lexical Phonology of Sekani’. Ph.D.

dissertation, UCLA.
—— and KAISSE, E. M. (eds.) (1993). Studies in Lexical

Phonology. San Diego: Academic Press.
HARRELL, R. S. (1962). A Short Reference Grammar of

Moroccan Arabic. Washington, DC: Georgetown University
Press.
HARRIS, J. W. (1973). ‘On the Ordering of Certain Phonological

Rules in Spanish’, in S. R. Anderson and P. Kiparsky (eds.), A
Festschrift for Morris Halle. New York: Holt Rinehart &
Winston, 59–76.
Page 12 of 33
References
—— (1994). ‘The OCP, Prosodic Morphology and Sonoran

Spanish Diminutives: A Reply to Crowhurst’, Phonology 11:
179–90.
HARRIS, JOHN (1989). ‘Towards a Lexical Analysis of Sound

Change in Progress’, Journal of Linguistics 25: 35–56.
—— (1990). ‘Derived Phonological Contrasts’, in S. Ramsaran

(ed.), Studies in the Pronunciation of English: A
Commemorative Volume in Honour of A. C. Gimson. London:
Routledge, 87–105.
—— (1994). English Sound Structure. Oxford: Blackwell.
HARRIS, Z. D. (1951). Structural Linguistics. Chicago:

University of Chicago Press.
HAYES, B. (1981). ‘A Metrical Theory of Stress Rules’. Ph.D.

dissertation, MIT. [Distributed by the Indiana University
Linguistics Club, Bloomington, Indiana.]
—— (1985). ‘Iambic and Trochaic Rhythm in Stress Rules’,

BLS 13: 429–46.
—— (1986). ‘Inalterability in CV Phonology’, Language 62:

321–51.
—— (1995a). ‘On What to Teach the Undergraduates: Some

Changing Orthodoxies in Phonological Theory’, in I.–H. Lee
(ed.), Linguistics in the Morning Calm 3. Seoul: Hanshin, 59–
77.
—— (1995b). Metrical Stress Theory: Principles and Case

Studies. Chicago: University of Chicago Press.
—— (1999). ‘Phonological Restructuring in Yidiny and its

Theoretical Consequences’, in B. Hermans and M. Oostendorp
(eds.), (1999), 175–205.
—— (2000). ‘Gradient Well-Formedness in Optimality Theory’,

in J. Dekkers, F. van der Leeuw, and J. van de Weijer (eds.),
Optimality Theory. Phonology, Syntax, and Acquisition. Oxford:
Oxford University Press, 88–120.
—— (2004). ‘Phonological Acquisition in Optimality Theory:

The Early Stages’, in R. Kager, J. Pater, and W. Zonneveld
Page 13 of 33
References
(eds.), Constraints in Phonological Acquisition. Cambridge:

HEATH, J. (1987). Ablaut and Ambiguity: Phonology of a

Moroccan Arabic Dialect. Albany: State University of New York
Press.
HEBB, D. O. (1949). The Organization of Behavior: A

Neuropsychological Theory. New York: John Wiley & Sons.
HENDRICKS, S. (1999). ‘Reduplication Without Templates: A

Study of Bare-Consonant Reduplication’. Ph.D. dissertation,
University of Arizona.
HERMANS, B., and VAN OOSTENDORP, M., (eds.) (1999). The

Derivational Residue in Phonological Optimality Theory.
Amsterdam: Benjamins.
HETZRON, R. (1975). ‘Where the Grammar Fails’, Language

51: 859–72.
HOCK, H. H. (1986). Principles of Historical Linguistics.

Berlin: Mouton.
—— (1991). Principles of Historical Linguistics (2nd edn.).

Berlin: Mouton.
(p.320) HOOPER, J. B. (1976). ‘Constraints on Schwa-

Deletion in American English’, in J. Fisiak (ed.), Recent
Developments in Historical Phonology. The Hague: Mouton,
183–207.
HORWOOD, G. (1999). ‘Anti-Faithfulness and Subtractive

Morphology’, unpublished MS, Rutgers University. [Available
at ROA-466.]
HUKARI, T., and PETER, R. (1995). Hul’qumi’num’ Dictionary.

The Cowichan Dictionary of the Hul’q’umi’num Dialect of the
Coast Salish People. Duncan, Canada: Cowichan Tribes.
HUMBOLDT, W. VON (1971). Linguistic Variability and

Intellectual Development [trans. of Über die Verschiedenheit
des menschlichen Sprachbaues und ihren Einfluß auf die
geistige Entwicklung des Menschengeschlechts, 1836], Miami
Linguistic Series, 9. Coral Gables, Flo.: University of Miami
Press.
Page 14 of 33
References
HYMAN, L. M. (1993). ‘Conceptual Issues in the Comparative

Study of the Bantu Verb Stem’, in S. S. Mufwene and L. Moshi
(eds.), Topics in African Linguistics. Amsterdam: Benjamins, 3–
34.
—— (1994). ‘Cyclic Phonology and Morphology in Cibemba’, in

J. Cole and C. Kisseberth (eds.), Perspectives in Phonology.
Stanford, Calif.: CSLI, 81–112.
—— (1998). ‘Positional Prominence and the “Prosodic Trough”

in Yaka’, Phonology 15: 41–75.
—— (2003). ‘Sound Change, Misanalysis, and Analogy in the

Bantu Causative’, Journal of African Languages and Linguistics
24: 55–90.
—— and INKELAS, S. (1997). ‘Emergent Templates: The

Unusual Case of Tiene’, University of Maryland Working
Papers in Linguistics 5: 92–116.
—— and MCHOMBO, S. (1992). ‘Morphotactic Constraints in

the Chichewa Verb Stem’, BLS 18: 350–64.
INKELAS, S. (1989). ‘Prosodic Constituency in the Lexicon’.

Ph.D. dissertation, Stanford University.
ITÔ J., and MESTER, A. (1997). ‘Correspondence and

Compositionality: The ga-gyo Variation in Japanese Phonology’,
in I. Roca (ed.), 419–62. [Also available at ROA-145.]
—— (2002). ‘Lexical and Postlexical Phonology in Optimality

Theory’, in G. Fanselow and C. Féry (eds.), Resolving Conflicts
in Grammars: Optimality Theory in Syntax, Morphology, and
Phonology. Hamburg: Helmut Buske, 183–207 [Linguistische
Berichte Sonderheft (supplement) 11.]
—— KITAGAWA, Y., and MESTER, A. (1996). ‘Prosodic

Faithfulness and Correspondence: Evidence from a Japanese
Argot’, Journal of East Asian Linguistics 5: 217–94. [Also
available at ROA-146.]
IVERSON, G., and SALMONS, J. (1995). ‘Aspiration and

Laryngeal Representation in Germanic’, Phonology 12: 369–96.
JAKOBSON, R. (1948). ‘Russian Conjugation’, Word 4: 155–67.
Page 15 of 33
References
JEFFERS, R. J., and LEHISTE, I. (1979). Principles and

Methods of Historical Linguistics. Cambridge, Mass.: MIT
Press.
JENSEN, J. T. (1990). Morphology: Word Structure in

Generative Grammar. Amsterdam: Benjamins.
—— (2000). ‘Against Ambisyllabicity’, Phonology 17:187–235.
JOHNSON, W. (1987). ‘Lexical Levels in French Phonology’,

JONES, D. (1931). ‘The “Word” as a Phonetic Entity’, Le Maître

Phonétique 36: 60–5.
—— (1956). ‘The Hyphen as a Phonetic Sign’, Zeitschrift für

Phonetik und Allgemeine Sprachwis-senschaft 9.2: 99–107.
JONES, M. (1976). ‘Morphophonemic Properties of Cowichan

Actual Aspect’. MA thesis, University of Victoria.
(p.321) JUN, J. (1995). ‘Perceptual and Articulatory Factors

in Place Assimilation: An Optimality Theoretic Approach’.
Ph.D. dissertation, UCLA.
JUSCZYK, P. W., LUCE, P. A., and LUCE, J. C. (1994). ‘Infants’

Sensitivity to Phonotactic Patterns in the Native Language’,
Journal of Memory and Language 33: 630–45.
——, FRIEDERICI, A., WESSELS, J. M., SVENKERUD, V. Y., and

JUSCZYK, A. M. (1993). ‘Infants’ Sensitivity to the Sound
Patterns of Native Language Words’, Journal of Memory and
Language 32: 402–20.
KAGER, R. (1993). ‘Alternatives to the Iambic-Trochaic Law’,

Natural Language and Linguistic Theory 11: 381–432.
—— (1999a). ‘Surface Opacity of Metrical Structure in

Optimality Theory’, in B. Hermans and M. van Oostendorp
(eds.) (1999), 207–45.
—— (1999b). Optimality Theory. Cambridge: Cambridge

University Press.
—— (2000). ‘Stem Stress and Peak Correspondence in Dutch’,

in J. Dekkers, F. van der Leeuw and J. van de Weijer (eds.),
Page 16 of 33
References
Optimality Theory: Phonology, Syntax, and Acquisition. Oxford:

KAHN, D. (1976). ‘Syllable-Based Generalizations in English

Phonology’. Ph.D dissertation, MIT. [Distributed by the Indiana
University Linguistics Club, Bloomington, Ind.]
KAISSE, E. M. and SHAW, P. (1985). ‘On the Theory of Lexical

Phonology’, Phonology Yearbook 2:1–30.
KAYE, J. (1995). ‘Derivations and Interfaces’, in J. Durand and

F. Katamba (eds.), Frontiers of Phonology: Atoms, Structures,
Derivations. London: Longman, 289–332.
—, LOWENSTAMM, J., and Vergnaud, J.-R. (1990). ‘Constituent

Structure and Government in Phonology’, Phonology 7:193–
231.
KELEPIR, M. (1997). ‘To Be or Not to Be Faithful’, unpublished

MS, MIT.
KENESEI, I., VAGO, R. M., and FENYVESI, A. (1998).

Hungarian. London: Routledge.
KENSTOWICZ, M. (1994). Phonology in Generative Grammar.

Cambridge, Mass.: Blackwell.
— (1995). ‘Cyclic vs. Non-Cyclic Constraint Evaluation’,

Phonology 12: 397–436.
— (1996). ‘Base-Identity and Uniform Exponence: Alternatives

to Cyclicity’, in J. Durand and B. Laks (eds.), (1996), 363–93.
— (1997). ‘Uniform Exponence: Exemplification and

Extension’, University of Maryland Working Papers in
KIECKERS, E. (1960). Historische Lateinische Grammatik, ii.

Formenlehre. Munich: Max Hueber Verlag. (Originally
published 1931.)
KING, R. (1969). Historial Linguistics and Generative

Grammar. Englewood Cliffs, NJ: Prentice-Hall.
KINKADE, M. D. (1998). ‘How Much does a Schwa Weigh?’, in

E. Czaykowska-Higgins and M. D. Kinkade (eds.), Salish
Page 17 of 33
References
Languages and Linguistics: Theoretical and Descriptive

Perspectives. Berlin: Mouton, 197–216.
KIPARSKY, P. (1972). ‘Explanation in Phonology’, in P. Kiparsky

(1982), Explanation in Phonology, Dordrecht: Foris, 81–118.
— (1978). ‘Analogical Change as a Problem for Linguistic

Theory’, in P. Kiparsky (1982), Explanation in Phonology,
Dordrecht: Foris, 217–36.
— (1979). ‘Metrical Structure Assignment is Cyclic’, Linguistic

Inquiry 10: 421–41.
— (1982a). Lexical Phonology and Morphology, Linguistics in

the Morning Calm. Seoul: Han-shin, 3–91.
— (1982b). ‘From Cyclic Phonology to Lexical Phonology’. In

H. van der Hulst and N. Smith (eds.), The Structure of
Phonological Representations (part 1), Dordrecht: Foris, 131–
75.
(p.322) KIPARSKY, P. (1985). ‘Some Consequences of Lexical

Phonology’, Phonology Yearbook 2: 83–138.
— (1993). ‘Blocking in Non-Derived Environments’, in S.

Hargus and E. Kaisse (eds.), Phonetics and Phonology, iv.
Studies in Lexical Phonology. San Diego: Academic Press,
277–313.
— (1997). ‘Covert Generalization’, unpublished MS, Stanford

University.
— (1998). ‘Paradigm Effects and Opacity’, unpublished MS,

Stanford University.
— (2000). ‘Opacity and Cyclicity’, The Linguistic Review 17:

351–65.
KIRCHNER, R. (1996). ‘Synchronic Chain Shifts in Optimality

Theory’, Linguistic Inquiry 27: 341–50.
— (1997). ‘Contrastiveness and Faithfulness’, Phonology 14:

83–113.
KISSEBERTH, C. (1970). ‘On the Functional Unity of

Phonological Rules’, Linguistic Inquiry 1: 291–306.
Page 18 of 33
References
— and ABASHEIKH, M. I. (1974). ‘A Case of Systematic

Avoidance of Homonyms’, Studies in the Linguistic Sciences
4.1: 107–24.
KLAUSENBURGER, J. (1979). Morphologization: Studies in

Latin and Romance Morphophonology. Tübingen: Niemeyer.
KRASKA-SZLENK, I. (1995). ‘The phonology of stress in

Polish’. Ph.D. dissertation, University of Illinois at Urbana.
KUHL, P., and IVERSON, P. (1995) ‘Linguistic Experience and

the “Perceptual Magnet Effect” ’, in W. Strange (ed.), Speech
Perception in Linguistic Experience: Theoretical and
Methodological Issues in Cross-language Speech Research.
Baltimore: York, 121–54.
KÜHNER, R. (1912). Ausführliche Grammatik der Lateinischen

Sprache (2nd edn.). Hannover: Verlag Hahnsche
Buchhandlung.
KURISU, K. (2001). ‘The Phonology of Morpheme Realization’.

Ph.D. dissertation, University of California at Santa Cruz.
KURYŁOWICZ, J. (1947). ‘The Nature of the So-Called

Analogical Processes’, Diachronica 12.1: 113–45.
— (1947). ‘La Nature des procès dit “analogiques” ’, Acta

Linguistica 5: 15–34. Reprinted in Eric P. Hamp, F. W.
Householder, and Robert Austerlitz (eds.) (1966), Readings in
Linguistics. Chicago: University of Chicago Press, 158–74.
LADEFOGED, P. (1983). ‘The Limits of Biological Explanation

in Phonetics’, UCLA Working Papers in Phonetics 57: 1–10.
LAHIRI, A. (2000). ‘Hierarchical Restructuring in the Creation

of Verbal Morphology in Bengali and Germanic: Evidence from
Phonology’, in A. Lahiri (ed.), Analogy, Levelling, Markedness.
Berlin: Mouton, 71–124.
— and DRESHER, B. E. (1984). ‘Diachronic and Synchronic

Implications of Declension Shifts’, The Linguistic Review 3:
141–63.
LASOR, W. S. (1979). Handbook of Biblical Hebrew. Grand

Rapids, Mich.: W. B. Eerdmans.
Page 19 of 33
References
LESLIE, A. (1979). ‘A Grammar of the Cowichan Dialect of

Halkomelem Salish’. Ph.D. dissertation, University of Victoria.
LEUMANN, M. (1977). Lateinische Laut- und Formenlehre.

Munich: C. H. Beck (Oscar Beck).
LEVIN [BLEVINS], J. (1983). ‘Reduplication and Prosodic

Structure’, unpublished MS, MIT.
LINDBLOM, B. (1986) ‘Phonetic Universals in Vowel Systems’,

in J. J. Ohala and J. J. Jaeger (eds.), Experimental Phonology.
New York: Academic, 13–44.
LOWENSTAMM, J. (1981). ‘On the Maximal Cluster Approach

to Syllable Structure’, Linguistic Inquiry 12.4: 575–604.
LUICK, K. (1964). Historische Grammatik der Englischen

Sprache. Oxford: Basil Blackwell. [Originally published
between 1914 and 1940 by Bernhard Tauchnitz Verlag
Stuttgart.]
(p.323) MACCARTHY, P. A. D. (1945). An English

Pronouncing Vocabulary. Cambridge: W. Heffer & Sons.
MCCARTHY, J. J. (1979). ‘Formal Problems in Semitic

Phonology and Morphology’. Ph.D. dissertation, MIT.
[Published 1985: New York: Garland.]
— (1980). ‘A Note on the Accentuation of Damascene Arabic’,

Studies in the Linguistic Sciences 10.2: 77–98.
— (1981). ‘A Prosodic Theory of Nonconcatenative

Morphology’, Linguistic Inquiry 12: 373– 418.
— (1982). ‘Prosodic Structure and Expletive Infixation’,

Language 58: 574–90.
— (1993). ‘Template Form in Prosodic Morphology’, in L.

Smith Stvan (ed.), Papers from the Third Annual Formal
Linguistics Society of Midamerica Conference. Bloomington,
Ind.: IULC, 187–218.
— (1998). ‘Morpheme Structure Constraints and Paradigm

Occultation’, CLS 32 (Part 2): 123–50.
— (1999). ‘Sympathy and Phonological Opacity’, Phonology 16:

331–99.
Page 20 of 33
References
— (2000a). ‘Faithfulness and Prosodic Circumscription’, in J.

Dekkers, F. van der Leeuw, and J. van de Weijer (eds.),
Optimality Theory: Phonology, Syntax, and Acquisition. Oxford:
— (2000b). ‘The Prosody of Phrase in Rotuman’, Natural

Language and Linguistic Theory 18: 147–97.
— (2002). A Thematic Guide to Optimality Theory. Cambridge:

— and PRINCE, A. (1986/1996). ‘Prosodic Morphology’, New

Brunswick, NJ: Rutgers University Center for Cognitive
Science. [Available at ROA-482-1201.]
— — (1990a). ‘Foot and Word in Prosodic Morphology: The

Arabic Broken Plural’, Natural Language and Linguistic
Theory 8: 209–83.
— — (1990b). ‘Prosodic Morphology and Templatic

Morphology’, in M. Eid and J. J. McCarthy (eds.), Perspectives
on Arabic Linguistics II: Papers from the Second Annual
Symposium on Arabic Linguistics. Amsterdam: Benjamins, 1–
54.
— — (1993a). ‘Generalized Alignment’, in G. Booij and J. van

Marle (eds.), Yearbook of Morphology. Dordrecht: Kluwer, 79–
153. [Also available at ROA-7.]
— — (1993b). ‘Prosodic Morphology: Constraint Interaction

and Satisfaction’, New Brunswick, NJ: Rutgers University
Center for Cognitive Science. [Available at ROA-482.]
— — (1994a). ‘The Emergence of the Unmarked: Optimality in

Prosodic Morphology’, NELS 24: 333–79.
— — (1994b). ‘Two Lectures on Prosodic Morphology (Utrecht,

1994). Part I: Template Form in Prosodic Morphology. Part II:
Faithfulness and Reduplicative Identity’, unpublished MS,
University of Massachusetts at Amherst and Rutgers
University. [Available at ROA-59.]
— — (1995). ‘Faithfulness and Reduplicative Identity’,

University of Massachusetts Occasional Papers in Linguistics
18: 249–384.
Page 21 of 33
References
— — (1999). ‘Faithfulness and Identity in Prosodic

Morphology’, in R. Kager, H. van der Hulst, and W. Zonneveld
(eds.), The Prosody–Morphology Interface. Cambridge:
Cambridge University Press, 218–309.
MALONE, J. L. (1993). Tiberian Hebrew Phonology. Winona

Lake, Indiana: Eisenbrauns.
MAŃCZAK, W. (1958). ‘Tendences générales des changements

analogiques’, Lingua 7: 298–325 and 387–420.
— (1978). ‘Les lois du développement analogique’, Linguistics

205: 53–60.
(p.324) MARANTZ, A. (1982). ‘Re Reduplication’, Linguistic

Inquiry 13: 483–545.
MARCUS, G., PINKER, S., ULLMAN, M., HOLLANDER, M.,

ROSEN, T., and XU, F. (1992). ‘Over-regularization in
Language Acquisition’, Monographs of the Society for
Research in Child Development. Ann Arbor: University of
Michigan.
MARTIN, J. (1988). ‘Subtractive Morphology as Dissociation’,

WCCFL 7: 229–40.
MARTIN, S. E. (1992). A Reference Grammar of Korean. Tokyo:

Charles E. Tuttle.
MASCARÓ, J. (1996). ‘External Allomorphy as Emergence of

the Unmarked’, in J. Durand B. Laks (eds.), (1996), 473–84.
MASCARÓ, J. (1996). ‘External Allomorphy as Emergence of

the Unmarked’, in J. Durand B. Laks (eds.), (1996), 473–84.
MAYERTHALER, W. (1981). Morphologische Natürlichkeit.

Wiesbaden: Akademische Verlags-gesellschaft Athenaion.
MEISER, G. (1998). Historische Laut- und Formenlehre der

lateinischen Sprache. Darmstadt: Wis-senschaftliche
Buchgesellschaft.
MESTER, R. A. (1994). ‘The Quantitative Trochee in Latin’,

Natural Language and Linguistic Theory 12.1: 1–61.
MIKHEEV, A. (1997). ‘Automatic Rule Induction for Unknown-

Word Guessing’, Computational Linguistics 23.3: 405–23.
Page 22 of 33
References
MISTELI, F. (1880). ‘Lautgesetz und Analogie’, Zeitschrift für

Völkerpsychologie und Sprachwissen-schaft 11: 365–475;
12:1–26.
MITCHELL, T. F. (1960). ‘Prominence and syllabification in

Arabic’, Bulletin of the School of Oriental and African Studies
23: 369–89.
— (1993). Pronouncing Arabic. Oxford: Clarendon Press, ii.
MOHANAN, K. P. (1986). The Theory of Lexical Phonology.

Dordrecht: Reidel.
MUTAKA, N. M. (1994). The Lexical Tonology of Kinande.

lincom Studies in African Linguistics 01. Munich: lincom
europa.
— and HYMAN, L. M. (1990). ‘Syllables and Morpheme

Integrity in Kinande Reduplication’, Phonology 7: 73–119.
MYERS, S. (1987a). ‘Tone and the Structure of Words in

Shona’. Ph.D. dissertation, University of Massachusetts at
Amherst.
— (1987b). ‘Vowel Shortening in English’, Natural Language

and Linguistic Theory 5: 485–518.
NESPOR, M., and VOGEL, I. (1986). Prosodic Phonology.

Dordrecht: Foris.
NEUE, C. F., and WAGENER, C. (1902). Formenlehre der

lateinischen Sprache (3rd edn., vol. i (Das Substantivum)).
Leipzig: O. R. Reisland.
NEWMAN, S. S. (1948). ‘English Suffixation: A Descriptive

Approach’, Word 4: 24–36.
NOYER, R. (1992). ‘Features, Positions and Affixes in

Autonomous Morphological Structure’. Ph.D. dissertation, MIT.
— (1993). ‘Optimal Words: Towards a Declarative Theory of

Word-formation’, unpublished MS, Princeton University.
OED (1994). The Oxford English Dictionary, 2nd edn., on

Compact Disc. Oxford: Oxford University Press.
Page 23 of 33
References
ORGUN, C. O. (1996). ‘Sign-Based Morphology and

Phonology’. Ph.D. dissertation, UC-Berkeley
— (1999). ‘Sign-Based Morphology: A Declarative Theory of

Phonology-Morphology Interleaving’, in B. Hermans and M.
van Oostendorp (eds.) (1999), 247–67.
OSTHOFF, H., and BRUGMAN, K. (1878). Morphologische

Untersuchungen auf dem Gebiete der indogermanischen
Sprachen, Part 1. Leipzig: Hirzel.
PADGETT, J. (1997). ‘Perceptual Distance of Contrast: Vowel

Height and Nasality’, Phonology at Santa Cruz 5: 63–78.
(p.325) — (2002). ‘Feature Classes in Phonology’, Language

78. 2: 81–110.
PATER, J. (2000). ‘Non-Uniformity in English Secondary Stress:

The Role of Ranked and Lexically-Specific Constraints’,
PAUL, J.‘(1880). Prinzipien der Sprachgeschichte. Tübingen:

Niemeyer. (English trans. 1891: Principles of the History of
Language. London: Longmans, Green & Co.)
— (1920) Principien der Sprachgeschichte, 5th edn. Halle:

Niemeyer.
PEPERKAMP, S. (1997). Prosodic Words. HIL Dissertations 34.

The Hague: Holland Academic Graphics.
Graphics. PIGGOTT, G. L. (1999). ‘At the Right Edge of Words’,

The Linguistic Review 16.2: 143–85.
PINKER, S. (1991). ‘Rules of Language’, Science 253 (5019):

530–5.
— (1999). Words and Rules: The Ingredients of Language. New

York: Basic Books.
— and PRINCE, A. (1988). ‘On Language and Connectionism:

Analysis of a Parallel Distributed Processing Model of
Language Acquisition’, Cognition 28: 73–193.
POLETTO, R. (1998). ‘Topics in Runyankore Phonology’. Ph.D.

dissertation, The Ohio State University.
Page 24 of 33
References
POTTER, B. (1994). ‘Serial Optimality in Mohawk Prosody’,

CLS 30.1: 347–61.
PRINCE, A. (1975). ‘The Phonology and Morphology of

Tiberian Hebrew’. Ph.D. dissertation, MIT.
— (1990). ‘Quantitative Consequences of Rhythmic

Organization’, CLS 26.2: 355–98.
— and SMOLENSKY, P. (1993). ‘Optimality Theory: Constraint

Interaction in Generative Grammar’, RuCCS Technical Report
2. New Brunswick, NJ: Rutgers Center for Cognitive Science.
PULGRAM, E. (1970). Syllable, Word, Nexus, Cursus. The

Hague: Mouton.
RAFFELSIEFEN, R. (1993). ‘Relating Words: A Model of Base

Recognition’, Linguistic Analysis 23: 2–163.
— (1995). Conditions for Stability: The Case of Schwa in

German, Arbeiten des Sonderfor-schungsbereichs 282.
Düsseldorf: Heinrich Heine Universität.
— (1996). ‘Gaps in Word Formation’, in U. Kleinhenz (ed.),

Interfaces in Phonology. Studia Grammatica 41. Berlin:
Akademie Verlag, 194–209.
— (1999a). ‘Diagnostics for Prosodic Words Revisited: The

Case of Historically Prefixed Words in English’, in T. A. Hall U.
Kleinhenz (eds.), Studies on the Phonological Word.
Amsterdam: Benjamins, 133–201.
— (1999b). ‘Phonological Constraints on English Word

Formation’, in G. Booij and J. van Marle (eds.), Yearbook of
Morphology 1998. Dordrecht: Kluwer, 225–87.
— (2000) (2003, 2nd corrected edn.). ‘Constraints on Schwa

Apocope in Middle High German’, in A. Lahiri (ed.), Analogy,
Leveling, Markedness, Berlin: Mouton de Gruyter, 125–70.
— (2004a). ‘Phonological Effects in Word Formation’.

Habilitationsschrift. Berlin: Freie Universität.
— (2004b). ‘Absolute Ill-Formedness and Other

Morphophonological Effects’, Phonology 21: 91–142.
Page 25 of 33
References
REBRUS, P. (2000a). ‘Morfofonológiai ielenségek’, in F. Kiefer

(ed.), Strukturális magyar nyelvtan, iii. Morfológia. Budapest:
Akadémiai Kiadó, 763–948.
— (2000b). ‘A magyar igei paradigma kivételeirõl’, in M.

Bakró-Nagy, Z. Bántréti, and ÉK. Kiss (eds.), Újabb
tanulmányok a strukturális magyar nyelvtan és nyelvtörténet
körébõl Budapest: Osiris, 249–73.
(p.326) REPP, B. H. (1982). ‘Phonetic Trading Relations and

Context Effects: New Experimental Evidence for a Speech
Mode of Perception’. Psychological Bulletin 92: 81–110.
RIALLAND, A. (1986). ‘Schwa et syllabes en Français’, in L.

Wetzels E. Sezer (eds.), Studies in Compensatory Lengthening.
Dordrecht: Foris.
RINGEN, C. O., and VAGO, R. M. (1998). ‘Hungarian Vowel

Harmony in Optimality Theory’, Phonology 15: 393–416.
ROCA, I. (1992). ‘Constraining Extrametricality’, in W.

Dressler, H. Lüschutzky, O. Pfeiffer, J. Rennison (eds.),
Phonologica 1988. Cambridge: Cambridge University Press,
239–48.
— (ed.), (1997) Derivations and Constraints in Phonology.

Oxford: Oxford University Press.
ROSE, S. (2000). ‘Rethinking Geminates, Long-Distance

Geminates, and the OCP’, Linguistic Inquiry 31: 85–122.
ROSENTHALL, S. (2002). ‘Weak Verb Stems in Classical

Arabic’ unpublished MS, Oakland University, Rochester, Mich.
RUBACH, J. (1993). The Lexical Phonology of Slovak. Oxford:

Oxford University Press.
— (1997). ‘Extrasyllabic Consonants in Polish: Derivational

Optimality Theory’, in I. Roca (1997: 551–81).
— (2000). ‘Glide and Glottal Stop Insertion in Slavic

Languages: A DOT Analysis’, Linguistic Inquiry 3: 271–317.
RUSSELL, K. (1995). ‘Morphemes and Candidates in

Optimality Theory’. [Available at ROA- 44-0195.]
Page 26 of 33
References
— (1999). ‘MOT Sketch of an OT Approach to Morphology’.

[Available at ROA-352- 1099.]
SAGEY, E. (1986). ‘The Representation of Features and

Relations in Non-Linear Phonology’. Ph.D. dissertation, MIT.
[Published 1990 New York: Garland.]
SAMEK-LODOVICI, V., and PRINCE, A. (1999). ‘Optima’,

unpublished MS, University of London and Rutgers University.
[Available at ROA-363.]
SANDERS, N. (2002). ‘Preserving Synchronic Parallelism:

Diachrony and Opacity in Polish’, CLS 37.1: 501–16.
SCHANE, S. (1968). French Phonology and Morphology.

Cambridge, Mass.: MIT Press.
SCHERER, J. (1868). Zur Geschichte der deutschen Sprache.

Berlin: Weidmann.
SCHMIDT, J. (1882). ‘Das Suffix des participium perfecti

activi’, Kuhns Zeitschrift für vergleichen- de Sprachforschung
auf dem Gebiete der indogermanischen Sprachen 27: 329–77.
SCHUCHARDT, H. (1885). Über die Lautgesetze: Gegen die

Junggrammatiker. Berlin: Oppenheim.
SELKIRK, E. O. (1981). ‘On the Nature of Phonological

Representations’, in J. Anderson, J. Laver, T. Meyers (eds.), The
Cognitive Representation of Speech. Amsterdam: North
Holland.
— (1982). ‘The Syllable’, in H. van der Hulst N. Smith (eds.),

The Structure of Phonological Representations, Part II.
Dordrecht: Foris, 337–83.
— (1984). Phonology and Syntax. Cambridge, Mass.: MIT

Press.
— (1986). ‘On Derived Domains in Sentence Phonology’,

Phonology Yearbook 3: 371–405.
— (1995). ‘The Prosodic Structure of Function Words’,

University of Massachusetts Occasional Papers in Linguistics
18: 439–69.
Page 27 of 33
References
— (2001). ‘Morphologically Grounded O-O Correspondence in

Noncyclic OT’, handout. University of Massachusetts at
Amherst.
SHAW, P., BLAKE, S., CAMPBELL, J., and SHEPHERD, C.

(1999). ‘Stress in hәnqәminәm (Musqueam) Salish’, WSCLA 4:
131–63.
SHERER, T. (1994). ‘Prosodic Phonotactics’. Ph.D. dissertation,

University of Massachusetts at Amherst.
(p.327) SIEVERS, E. (1876). Grundzüge der Lautphysiologie.

Leipzig: Breitkopf und Hartel.
SILVERMAN, D. (1997). Phasing and Recoverability. New York:

Garland.
SIPTÁR, P., and TÖRKENCZY, M. (2000). The Phonology of

Hungarian. Oxford: Oxford University Press.
SKUTSCH, O. (1985). The Annals of Q. Ennius. Oxford:

Clarendon Press.
SMOLENSKY, P. (1993). ‘Harmony, Markedness, and

Phonological Activity’. [Available at ROA-86.]
— DAVIDSON L. and JUSCZYK, P. (2005). ‘Optimality in

Language Acquisition, I: Phonology’, in P. SMOLENSKY G.
Legendre (eds.), The Harmonic Mind: From Neural
Computation to Optimality Theoretic Grammars. Cambridge:
Blackwell.
SOH, H.-L. (1996). ‘Vowel-Changing Reduplication in Malay’,

unpublished MS, MIT.
SPAELTI, P. (1997). ‘Dimensions of Variation in Multi-Pattern

Reduplication’. Ph.D. dissertation, University of California,
Santa Cruz. [Available at ROA-311.]
SPEISER, E. A. (1926). ‘Secondary Developments in Semitic

Phonology: An Application of the Principle of Sonority’, The
American Journal of Semitic Languages and Literatures
42:145–69.
SPENCER, A. (1991). Morphological Theory. Oxford:

Blackwell.
Page 28 of 33
References
STAMPE, D. (1973a). ‘A Dissertation on Natural Phonology’.

Ph.D. dissertation, University of Chicago. [Published 1979,
Garland, New York.]
— (1973b). ‘On Chapter Nine’, in M. J. Kenstowicz and C. W.

Kisseberth (eds.), Issues in Phonological Theory. The Hague:
Mouton, 44–52.
STERIADE, D. (1994). ‘Positional Neutralization and the

Expression of Contrast’, unpublished MS, UCLA.
— (1995). ‘Underspecification and Markedness’, in J.

Goldsmith (ed.), The Handbook of Phonological Theory.
Cambridge, Mass.: Blackwell, 114–74.
— (1997a.) ‘Lexical Conservatism,’ Linguistics in the Morning

Calm. Seoul: Hanshin, 157–79.
— (1997b). ‘Phonetics in Phonology: The Case of Laryngeal

Neutralization’, unpublished MS, UCLA.
— (1999a). ‘Lexical Conservatism in French and Adjectival

Liaison’, in M. Authier, B. Bullock, L. Reed (eds.), Proceedings
of the 25th Linguistic Colloquium on Romance Languages.
— (1999b). ‘Phonetics in Phonology: The Case of Laryngeal

Neutralization’, UCLA Working Papers in Linguistics: Papers in
— (2000). ‘Paradigm Uniformity and the Phonetics–Phonology

Boundary’, in M. Broe J. Pierrehumbert (eds.), Papers in
Laboratory Phonology, V. Acquisition and the Lexicon,
Cambridge: Cambridge University Press, 313–34.
STOJKOV, S. (1963). ‘Akane v govora na selo trigrad,

devinksko’, Bulgarski Ezik 13.1: 8–21.
STRAUSS, S. (1982). Lexicalist Phonology of English and

German. Dordrecht: Foris.
STRUIJKE, C. (1998). ‘Reduplicant and Output TETU in

Kwakwala’, University of Maryland Working Papers 150–78.
[Also available at ROA-261.]
Page 29 of 33
References
— (2000a). ‘Reduplication, Feature Displacement and

Existential Faithfulness’. Ph.D. dissertation, University of
Maryland at College Park.
— (2000b). ‘Why Constraint Conflict can Disappear in

Reduplication’, NELS 30: 613–26. [Also available at ROA-373.]
SUTTLES, W. (forthcoming). Musqueam: A Reference

Grammar of hnqәminәm Salish. First Nations Languages and
Linguistics Series. Vancouver: University of British Columbia
Press.
SZPYRA, J. (1989). The Morphology –Phonology Interface:

Cycles, Levels and Words. London and New York: Routledge.
(p.328) TESAR, B., and SMOLENSKY, P. (2000). Learnability

in Optimality Theory. Cambridge, Mass.: MIT Press.
TIERSMA, P. (1982). ‘Local and General Markedness’,

Language 58.4: 832–49.
TOBLER, L. (1879). ‘Über die Anwendung des Begriffes von

Gesetzen auf die Sprache’, Viertel-jahresschrift für
wissenschaftliche Philosophie 3: 30–52.
TRANEL, B. (1996). ‘French Liaison and Elision Revisited: A

Unified Account within Optimality Theory’, in C. Parodi, C.
Quicoli, M. Saltarelli, and M. L. Zubizarreta (eds.), Romance
Linguistics in Los Angeles: Selected Papers from the 24th
Linguistic Symposium on Romance Languages. Georgetown,
Wash.: Georgetown University Press, 433–55.
TRUBETZKOY, N. (1936). ‘Die phonologischen Grenzsignale’,

Proceedings of the 2nd International Congress of the Phonetic
Sciences. Cambridge: Congress of Phonetic Sciences, 45–9.
— (1939). Grundzüge der Phonologie. Travaux du Cercle

Linguistique de Prague 7. Prague: Cercle Linguistique.
UMEDA, N., and Coker C. H. (1974). ‘Allophonic Variation in

American English’, Journal of Phonetics 2:1–5.
URBANCZYK, S. (1996). ‘Patterns of Reduplication in

Lushootseed’. Ph.D. dissertation, University of Massachusetts
at Amherst.
Page 30 of 33
References
— (1999). ‘A-Templatic Reduplication in Halq’eméylem’,

WCCFL 17: 655–69.
— (2000). ‘Upriver Halq’eméylem Continuative Stems: A

Working Paper’, MS, University of Calgary, University of
Victoria.
USSISHKIN, A. (1999). ‘The Inadequacy of the Consonantal

Root: Modern Hebrew Denominal Verbs and Output–Output
Correspondence’, Phonology 16: 401–42.
— (2000). ‘The Emergence of Fixed Prosody’. Ph.D.

dissertation. Santa Cruz, Calif: University of California, Santa
Cruz.
VAGO, R. M. (1980). The Sound Pattern of Hungarian.

Washington, DC: Georgetown University Press.
VAN DAM, M., and WEAVER, P. (2001). Aspiration of stressless

intervocalic voiceless stops in English’, unpublished MS,
Indiana University, Bloomington.
VAN DE WEIJER, J. (1999). ‘Analogical Change in Optimality

Theory’, in On’in Kenkyuu 2 [Phonological Studies], Nihon
On’inron Gakkai [The Phonological Society of Japan.] Tokyo:
Kai-takusha, 145–52.
VAN DER HULST, H. (1984). Syllable Structure and Stress in

Dutch. Dordrecht: Foris.
VAN MARLE, J. (1985). On the Paradigmatic Dimension of

Morphological Creativity. Dordrecht: Foris.
VAN OOSTENDORP, M. (1995). Vowel Quality and Syllable

Projection. Tilburg Dissertations in Language Studies. Tilburg:
University of Tilburg.
VAUX, B. (1998). ‘The Laryngeal Specifications of Fricatives’,

Linguistic Inquiry 29: 497–511.
VENNEMANN, T. (1972). ‘Phonetic Analogy and Conceptual

Analogy’, in T. Vennemann and T. H. Wilbur (eds.), Schuchardt,
the Neogrammarians, and the Transformational Theory of
Phonological Change. Frankfurt am Main: Athenäum Verlag,
181–204.
Page 31 of 33
References
WALKER, R. (2000). ‘Nasal Reduplication in Mbe Affixation’,

WEBSTER’S Ninth New Collegiate Dictionary (1984).

Springfield, Mass.: Merriam-Webster.
WELDEN, A. (1980). ‘Stress in Cairo Arabic’, Studies in the

Linguistic Sciences 10.2: 99–120.
WELLS, J. C. (1990a). Longman Pronunciation Dictionary.

Harlow: Addison Wesley Longman.
— (1990b). ‘Syllabification and Allophony’, in S. Ramsaran

(ed.), Studies in the Pronunciation of English: A
Commemorative Volume in Honour of A. C. Gimson, London:
Routledge, 78–86.
WETZELS, L. (1984). ‘Paradigm Leveling in Latin and Old

French: A Critical View of Rule Inver-sion’, in P. Baldi (ed.),
Papers from the XIIth Linguistic Symposium on Romance
Languages. Current Issues in Linguistic Theory 26.
(p.329) WHITNEY, W. D. (1870). Language and the Study of

Language. New York: C. Scribner & Sons.
WILBUR, R. (1974). ‘The Phonology of Reduplication’. Ph.D.

dissertation, University of Illinois.
WILSON, C. (2000). ‘Targeted Constraints: An Approach to

Contextual Neutralization in Optimality Theory’. Ph.D.
dissertation, Johns Hopkins University.
— (2001). ‘Consonant Cluster Neutralization and Targeted

Constraints’, Phonology 18.1: 147–97.
WILTSCHKO, M. (forthcoming). ‘-exw as 3rd Person Object

Agreement in Halkomelem’, International Journal of American
Linguistics 69.1: 76–91.
WITHGOTT, M. (1982). ‘Segmental Evidence for Phonological

Constituents’. Ph.D. dissertation, University of Texas.
WRIGHT, W (1971). A Grammar of the Arabic Language.

Cambridge: Cambridge University Press. [Third edn. revised
by W. R. Smith and M. J. de Goeje, originally published 1896.]
Page 32 of 33
References
WUNDERLICH, D. (1995). ‘Minimalist Morphology: The Role

of Paradigms’, in G. Booij and J. van Marle (eds.), Yearbook of
Morphology 1995. Dordrecht: Kluwer, 93–114.
WURZEL, W U. (1989). Inflectional Morphology and

Naturalness. Dordrecht: Kluwer Academic.
YIP, M. (1995). ‘Repetition and its Avoidance: The Case of

Javanese’, in K. Suzuki D. Elzin-ga (eds.), Proceedings of South
Western Optimality Theory Workshop; 1995 Arizona Phonology
Conference, 5. University of Arizona, Department of
Linguistics Coyote Papers. Tucson: University of Arizona, 238–
62.
— (1998). ‘Identity Avoidance in Phonology and Morphology’,

in S. Lapointe, D. Brentari, P. Farrell (eds.), Morphology and its
Relation to Phonology and Syntax. Stanford, Calif: CSLI, 216–
46.
ZEC, D. (1995). ‘Sonority Constraints on Syllable Structure’,

ZHANG, J. (2000). ‘Non-Contrastive Features and Categorical

Patterning in Chinese Diminutive Suffixation: max[F] or
Ident[F]?’, Phonology 17: 427–78.
ZWICKY, A. (1977). ‘On Clitics’, in W. U. Dressler and O. E.

Pfeiffer (eds.), Phonologica 1976. Innsbruck: Innsbrücker
Beiträge zur Sprachwissenschaft, 29–40. (p.330)
Page 33 of 33
Index


Print ISBN-13: 9780199267712
DOI: 10.1093/acprof:oso/9780199267712.001.0001
(p.331) Index
Figures and notes are indexed as ‘f’ and ‘n’ in bold.
æ-tensing 248, 250n, 251, 251n

‘ablaut’ 154–6, 169, 250n
in Salish languages 304, 305
abstract:
roots 24n
units and phonology 107
acquisition, of languages 257–8, 260
Adam, G. 49, 49n
O. Bat-El 51
adjectives 12, 234n, 256
affixations 197, 219, 236, 239, 243; see also head, affixations
affixes:
see also under Hungarian; Jita; stress neutrality; paradigm
uniformity (PU); suffixes
cohesion of 217, 236
disyllabic 215, 216
English 101
Italian 101
monosyllables 215, 216
noncohesion in 215, 233n
parsing rule for 232
postinflectional 215
Spanish 147
agentive 22, 249
in Latin 35
Page 1 of 43
Index
nouns see nouns, agentive

suffixes 243; see also suffixes
agglutinative languages 268, 269, 273; see also Hungarian,
languages
agreement morphology 212n; see also morphology
Aguero-Bautista, C. 147, 148
Albright, A. 15, 35, 214
B. Hayes 22, 24, 26, 33
Alderete, J. D. 51, 150, 152, 173n, 250 et al. 178n
Algiers dialect 164
Ali Adra, M. 159, 161
align 216n, 217, 218, 221, 224, 225, 234, 235, 261, 262
alignment constraints 118–19, 190, 192, 195, 198, 226, 230, 236,
246, 250n, 260, 261
see also constraints
violations of 189, 189n
Allen, J. H. 28, 32, 32n, 33n
allomorphy 22n, 37, 70, 72, 77, 86, 92, 100, 101–2, 105, 109, 191n,
236, 265, 267 see also under anti-allomorphy; distance and; French;
Hungarian, lexical allomorphy in;
morphemes; paradigm uniformity (PU); Salish languages
and distance 69–71
degree of, in Latin 37
grammar and 146
in affixes 101
phonologically based 93
Spanish 147
stress in 101
vowels in 98–9
allophony 207n, 252, 255, 258, 260, 261
distribution in 258
rules in 8; see also dentalization
syllable-conditioned 252
variation of 89
alternations 3, 5, 140;
see also under Hebrew; Hungarian; Korean; Latin; vowel/zero
in morphology 105
in paradigms 182
optimal and 191n, 201, 203, 208
in phonology 84
intraparadigmatic vowel/zero 186, 190
leveling of 18
versus extending of 39–41
physiologically motivated 5
(p.332) ambisyllabicity 251, 254, 257
flapping in 221
obstruents, voiced 247
American English 82, 87, 89, 109, 243n
Page 2 of 43
Index
aspiration 112, 116, 119

dialects in 11
flapping in 108
fricatives, voiceless 117
analogy 2, 3, 5, 170
changes in 2, 2n
psychologically motivated 2
four-part notation 18
leveling in 244
notation, in Latin 18
Anderson, S. R. 44, 48n, 68
Anshen, F. and M. Aronoff 138
anti-allomorphy 172, 264n
anti-harmony, in Hungarian see Hungarian, anti-harmony in
anti-homonymy see paradigms, contrasts
anti-syncopes 192; see also syncopes
Arab Rule 211, 212
in suffixation 243
Arabic 142, 147, 156, 157–62, 168, 169, 208;
see also Classical Arabic; individual dialects
affixes 308n
dialects 190, 199, 203
morphology of 81, 171
paradigms in 157, 308n
optimal 172, 178–9, 182, 187–9, 190, 192, 198–200
Sudanese 71
templates 171, 189, 208–9
and optimal paradigm theory 182–8;
nominal 182;
nouns 193,
verbs 176, 178, 181n, 182, 192, 191n, 193
verbs in 186
vowels 209
Aronoff, M. 138
S. N. Sridhar 215, 219, 219n
aspectual paradigms see also Salish languages, paradigms
and aspectual markings 312
aspiration 107–21, 221, 233, 235, 245
see also English, aspiration; foot structure; fortition
American English 112, 116, 119
in prefixed words 245
of stem-final consonants 233
stops 107–21
unaspirated 119;
voiceless 110–12, 117
assimilation 71–2, 81, 229; see also dissimilation
and distance 72
attraction in 72
Page 3 of 43
Index
association of words see words, associations

associations, ‘tightness’ of see paradigms, associations
asymmetry 171, 172
attested cognates see cognates, attested
attraction: see also markedness
in morphology 81
over distance see distance, attraction over
to the marked 175
to the unmarked 196, 197, 198, 205, 206
attractors 5, 16, 196, 197, 198, 199, 202, 214, 255
markedness of 175, 196, 205, 206
pronunciation of 256
Avanesov, R. I. 151
back-copying 95–6, 96n
backness harmony 266
in Hungarian 279, 280–1
Baldi, P. 19
Banias dialect 159, 161, 162, 162n
Bantu languages 124, 125, 125n, 130, 141n, 142, 153; see also
Bemba, Jita
causative:
consonants 123;
doubling 123;
morphemes 123;
spirantization in 137n
verb stems, uniformity in 15, 141
Barr, R. 18, 21, 36
base 16
and stems 297
comparison of potential 31f
discovery 23–33
for word-formation 296
identity 267n–8n
in Latin 30, 32
base-priority 9, 12, 13, 15, 168–9, 171, 172, 88, 188n, 197, 212n,
297, 312
analyses of 12, 16
and surface forms 312
in cyclicity 12n, 15
in inflectional paradigms 312
in Optimality Theory (OT) 12n
base-reduplicant:
identity constraints 207
structure 150
Bastin, Y. 123, 155
Bat-El, O. 12, 49, 59n, 62
Beckman, J. N. 52, 308
Bedouin Arabic 150
Page 4 of 43
Index
dialects 190
Belfast English 7, 8
consonants in 8
dentalization in 7
Bemba 124n, 128, 134
causative doubling 123, 125, 139
suffixes 124
Benua, L. 8, 9, 10, 11, 11n, 46, 58, 61n, 67, 116, 122, 129, 130, 134,
139, 146, 168, 169, 171, 172, 206, 207, 207n, 212, 240, 251, 296
Bermúdez-Otero, R. 197n
Bianco, V. 299, 311
(p.333) binary foot see foot structure
Black, H. A. 197n
Bloomfield, L. 167
Bobaljik, J. 199, 205n
Boersma, P. 77
B. Hayes 21, 255
Bolozky, S. 49
Bonet, E. 102
M.-R. Lloret 173
Booij, G. 146, 220, 233n
R. Lieber 134, 135
J. Rubach 226
Borowsky, T. 197
bound stems 232 n, 236n, 241, 242, 242n, 256;
see also free stems; stems
in English 242
boundary 256
analysis 258, 262
effects:
noncohesion, affixation in 244;
noncohesion,
suffixation in 241, 242;
versus paradigm uniformity effects 211–62
phenomena 215
signals 213, 226n, 230
Brame, M. 126, 134, 191n
broad-IO Faith see constraints, broad-IO Faith
Broselow, E. 166
Buckley, E. 123, 129, 173, 188n
Bulgarian 147, 150, 152–3, 156, 168, 169
homophony in 292n
reduction in 292n
Bullock, B. 95
Burzio, L. 12, 16, 62, 66, 67, 69, 71, 73, 84, 87, 88, 89, 92, 93, 95,
96n, 100, 101, 114, 123, 129, 134, 137, 138, 139, 146, 172, 172–3n,
264n, 292, 296
Bybee, J. 22, 84, 95n, 138, 293, 293n
Page 5 of 43
Index
Bynon, T. 263
Cairene Arabic 166, 167
Calabrese, A. 150
canonical phonology 214; see also phonology
violations of 237
Cantineau, J. 160
careful speech see speech, careful
Carlson, K. 178n
Carstairs, A. 162
Casablanca dialect 166–7
casual speech see speech, casual
Casual Speech Criterion 246, 261
causative 124n, 128, 131, 135, 137, 138
doubling 121–44
Bantu languages 123;
Bemba 123, 126, 135, 139, 141;
cyclicity, analysis of 134–9;
inflectional paradigms in 139–43;
Jita 121–44;
paradigm optimization 130–4;
spirantization in 144
glides 127, 130, 131, 132, 134
languages 134, 136
morphemes 123
palatal (causative) glides 123, 124
optimal paradigms (OP), analysis of 123
stems 124, 127, 132–4, 136, 138, 139–40, 142, 143
non-causative 136;
of verbs 134;
suffixes 126
verbs 131
cells 264, 267; see also Hungarian, cells; paradigms
Central Hessian dialect 169
children and languages 26n
Chi-Mwi:ni 147, 153–7, 168
Chomsky, N. 131, 178
M. Halle 6, 119, 145, 146, 172, 173n, 197, 228n, 263
Chung, S. 146, 178n
Cicero 21n, 38
clash avoidance constraints see Salish languages,
constraints
Classical Arabic 162
see also Arabic; various dialects
conjugations 209
grammar of 200
majority rules in 206
noun templates in 209–10
Page 6 of 43
Index
paradigms, optimal in 178, 179, 182–3, 186, 187, 190, 191,

192n, 198, 200, 201, 203, 204n, 206
Classical Latin 17, 23, 36n, 190; see also Latin
Clements, G. N. 81n
cliticization 134–5
clitics 167, 179, 203, 247, 250n
in English 224
pronunciation, in German 247
words and 234–5
‘Closed Syllable Adjustment’ 260
codas 71, 222, 223, 223n, 224, 232, 233, 240, 246, 248, 253, 254,
258, 261
association 221, 222, 238, 251–2
consonants 183, 203n, 214n
mora-bearing 203n
neutralizations, in Korean 41; see also Korean, neutralizations
Coetzee, A. 207, 208n
cohesion 216, 218, 220, 225, 229–31, 233, 234, 261, 261n, 262
affixes in 236, 243
in compounds 231
in suffixes 220, 233–4, 237, 240, 240n, 242n, 243
of prefixes 237
of consonantal morphemes 259
phonologically-conditioned 230, 231, 233, 23
tense in 249
Cohn, A. and J. J. McCarthy 197n
Cole, J. 122, 128, 146
Colloquial Hebrew 45, 48, 56–9, 59n, 60, 60n, 61–4; (p.334)
see also Hebrew; Tiberian Hebrew
consonants 54
imperative paradigms 55, 56–7
imperative truncation in 45, 46, 48–54, 58, 59
compounds 225, 226
in English 224, 225
con constraints see Hungarian, con constraints
‘concatenative’ morphology 80–1
conceptual analogy see analogy, conceptual
consonant-final:
prefixes 226
stems 220
consonantal morphemes
cohesion of 259
consonants 6n, 178n, 179, 183, 184, 189, 191, 192, 198, 200,
201, 202, 203, 203n, 221, 230, 234, 235, 238, 245, 252, 252n,
257–62
see also Hungarian, consonants in
ambisyllabic, voiceless, in pwords 247
clusters of 186, 207
Page 7 of 43
Index
dentalized 8
in Belfast English 7
intervocalic 251
paradigm uniformity (PU) and 134, 150, 151, 160, 161, 163
phonetic contrasts of 221
pword-final 223, 224n, 233
spirantization of 124
stem-final 232,
aspiration of 233
stressless vowels and 235
syllable-initial 214
tautosyllabic 248
voiced constellation theories 139, 142, 143
constraints 9, 41, 117–18, 140, 157, 205, 206, 208, 238, 243n,
247, 251, 255, 259, 260, 262; see also under alignment;
Hebrew; Hungarian; I-O Faithfuness; markedness; Salish
languages;
against voiced obstruents 247
broad-IO Faith 307
correspondence 9, 130, 173, 173n, 174, 181, 197, 251
relations 301, 303;
theory 171, 172, 182
domination 218
exceptionless 11
hierarchy 9, 281, 289
in morphology 51, 87
in phonology 87
in rhotacism 21
paradigms:
leveling of 200;
uniformity (PU) 18, 22, 53, 215, 217, 218, 232, 234,
phonotactic 19
ranking 1, 5, 9, 11n, 15, 19–20, 21, 139–40, 173, 184, 194, 199,
200, 207, 208, 218, 240
containment, paradigm uniformity (PU) and 216, 217n, 218,
220, 222–3, 225, 226, 236, 238, 239, 240, 242, 246, 248, 251,
253, 254, 261
superordinate units in 217
coronal:
obstruents see obstruents, coronal
palatalization 174, 176, 206
sonorants 126
stops 89–90, 154
correspondence constraints 9, 130, 173, 173n, 174, 181, 197, 251
relations 301
theory 171, 172, 182
Côté, M.-H. 150
Cowell, M. 161
Page 8 of 43
Index
Crosswhite, K. 146, 150, 152

Crowhurst, M. J.
Crystal, D. 44, 47
Curtius 2, 2n, 3, 5
cycle 106, 149–50, 169, 197
and rule-ordering 1, 6
in morphology 11
phonology 146
cyclic OT 197n
cyclicity 10, 68, 70, 105, 127, 134, 146, 172, 263; see also
noncyclicity
analyses of 8, 13, 134
causative doubling 128–30, 134–42;
optimal paradigms (OP) in 134–43;
Optimal Theory (OT) 129;
paradigm effects 9
and English stress 8
of footing 9
base-priority in 12n, 15
derivations in 67
models 13
morphemes and 128
phonology and 122, 126, 263, 296
recursion and 12
representations in 9
symmetrical approaches in 12
Czaykowska-Higgins, E. 143
dactyls:
and non-final 110–11
and word-final 110–11, 121
flapping in 112
sequences in 109–14, 115
binary foot 112;
superfoot 112
Damascus dialect 157, 160, 162, 165, 166
Damouri dialect (Lebanon) 161
Danish 48n
dative forms, in Latin see Latin, forms
Davis, H. 213n, 262n, 300
M.-H. Cho 12, 116, 117, 118, 119
definiteness neutralization see Hungarian, definiteness
neutralization,
Delatte, L. et al. 29, 30
dentalization 8
in Belfast English 7
(p.335)
derivational:
morphology 172, 174, 207
Page 9 of 43
Index
theory see Salish languages

diacritics 220, 235, 240
and abstract roots 24n
dialects 161, 265
‘differential’ dialect 160
Dinka 150
diphthongs 211, 212, 214, 244
discontinued morphemes 80–1; see also morphemes
‘Dispersion Theory’ 70, 100
dissimilation (‘OCP’) 71–2, 81, 200; see also assimilation
and distance 72
attraction in 72
distance 70–7, 74
and allomorphy 69–71
and segment-sized units 70
attraction over 70, 73, 75, 76–7, 105
global 70, 72, 78, 84
perceptual 71, 77, 78
phonetics 70–3
phonology 70–3
role of 67–70
stress 67–8
distributed morphology (DM) 102–3, 104; see also morphology
disyllabic suffixes 217, 218
Dixon, R. 215
Djerba dialect 165, 167
domain effects 262
doubling see causative, doubling
Downing, L.J. 15, 122n, 123, 124, 124n, 130, 130n, 131n, 142,
142n, 143, 175, 178n, 298
Dutch 71, 220, 233n
Egyptian 190
Elordieta, G. and M. Carreira 147
Embick, D. and R. Noyer 102
English 88, 96n, 99, 100, 102, 112, 141, 162, 286;
see also American English
affixes in 101
allophony, stop in 108
aspiration 118
in Optimality Theory (OT) 116–21
compounds in 225
constraints in 10
dialects 253–6
fricatives, nasal in 71
Latinate 84
morphemes in 81n, 104n
morphology in 149
and prosodic boundaries 224–37
Page 10 of 43
Index
morphophonology in 218–20, 236, 242

constraint ranking in 240n;
pwords in 180
nouns 115
paradigm uniformity effects in 172, 172n–3n, 174n, 188, 191n,
215, 218, 223, 223n, 233n, 234, 237, 238, 239, 240n, 245
phonology, irregular 85
pwords in 224, 226, 229, 235
simplexes in 221, 226
stress in 8, 9, 66, 89, 145, 237
cyclic application in 8
voiceless stops 117–18
vowels in 82, 88, 145
affixes in 98;
shortening of 89, 92
words:
morphosyntactic representations of 224;
prosody of 254
Ennius, Q. 17n
entailments: see also gradient attraction (GA), entailments
and generalizations 91
paradigm uniformity (PU) and 78–9, 80, 81, 81n, 83–4, 92, 94,
96, 96n, 97, 98, 100, 104, 105–6;
representations in 90–1, 97
transitivity of 78–9, 80, 83, 94, 96
epenthesis 183, 184, 186, 187, 189, 191, 192, 195, 196, 197, 198–9,
207, 208
vowels in 185, 188, 194, 207
etymology 22
of Latin 34
words in 3
base of 242n
European pre-Structuralist linguistics 2
exceptionless constraints 11; see also constraints
extension of alternations 18; see also alternations
feet see foot structure
Ferguson, C. A. 248, 250n, 251
Fidelholtz, J. 211
‘filters’ 93
‘Final-C’ effects 226, 241, 242
Final Devoicing 6n, 7
flapping 108–10, 112, 114–16, 119, 121, 153, 154, 155, 155n, 156,
221, 235, 252n, 262n; see also lenition
ambisyllabicity in 221, 222
before stressed vowels 246
syllables 121
Flemming, E. 70, 107, 137, 146, 149, 150
M. Kenstowicz 296
Page 11 of 43
Index
foot-initial stops 112–13

foot parsing 191
foot structure 9, 112–16, 118, 119, 120, 191, 193, 216, 217, 222,
223, 225, 246, 246n, 250, 253, 262n
see also constraints, Output–Output; foot structure
aspiration 112
binary 8, 112, 190, 217
iambic 190, 192n
(p.336)
pretonic 227
superfoot 112–14, 115
trochaic 119, 190, 192n
formal groups see groups, formal
forms 5, 35, 99–100, 173; see also under Hebrew; Latin
alternations among 172
and grammar 35, 37
as potential bases 32f
bases of 32
derived 18
in paradigms 38–9
inflected 170
marked structure in 202
nominative 18
related, pairs of 24
underived 18
fortition 211–12, 221–2, 245
four-part notation 18
free stems 241, 242; see also bound stems; stems
French 73, 108, 162, 217n, 261, 261n, 262n
adjectives 12, 68–9
allomorphy in 68–9, 70, 73
conjugation in 95
dialects 167
durational PU effects in 256–62
syllabification in 259, 261n
frequencies, token see token frequencies
fricatives:
in Hebrew 49
in Jita 130
nasal, in English 71
voiced 247
voiceless, in American English 117, 251n
tautosyllabic 248
Friederici, A. D. and J. E. Wessels 26n
Frisian 39
Fukazawa, H. 72
function words 229, 234, 235, 250n, 260, 261;
see also words
Page 12 of 43
Index
head 249
fusion 225, 229, 234
Gafos, A. 178n, 181n, 200, 301, 308n
Galloway, B. 297, 298, 299, 300n, 304
Ganong, W. F. 72, 73
Garr, W. R. 207, 208n
geminates 231n, 234
verbs 200; see also verbs
gemination 157, 162, 168
generalizations 4, 5, 69–70, 80, 90, 162; see also Hebrew,
generalizations; phonology
‘concatenative’ morphology of 80
entailments and 91
faithfulness 70
minimal 33
learners of 31;
morphology, rule-induction of 22
Sonority Sequencing Generalization (SSG) see Hebrew
surface 13
Generalized Template Theory (GTT) 171, 178, 182, 187, 189n, 195,
209
generative 150
grammar 169
linguistics 73
phonology 65, 212; see also phonology
classical 263n
genitive 23–4, 26, 27
German 3, 6, 227n, 231n, 233n, 234n, 242n, 247
boundary effects in 241n
paradigm leveling in 4
vowels in 6, 247
lengthening of 6
words, morphology of 6
gerunds 244, 244n
Gesenius, W. 207
Giegerich, H. 223
glides 123, 124, 128, 130, 178n, 187n, 238n, 240
verbs with high 200; see also verbs
glottalization 221, 246, 252; see also lenition
in Salish languages 304
Goldsmith, J. 146
Gougenheim, G. 260
Gouskova, M. 190
government phonology 264; see also phonology
gradient attraction (GA) 16, 74, 83–4, 91, 94, 97, 99, 100, 102, 105
affixes 98
entailments in 106
grammar 22
Page 13 of 43
Index
morphology, rules of 4
neutralizations and 100
grammar 25–6, 30n, 33, 74, 146, 152, 172n, 178, 186, 263
allomorphy and 146
and forms 35, 37
and lexicon 35
constraints 146
rankings in 12
early generative 6
functions in 3
gender in 231n
paradigms 296
uniformity (PU) and 64
phonology and 145–69
rules 146
in Latin 32
Greek 242n
groups 3, 5
formal 3, 4
leveling 5
material 3, 4
(p.337)
words in 5n
Grundzüge der Lautphysiologie (Sievers) 2
Gussenhoven, C. 169
Gussmann, E. 188n
Haddad, G. 161
Hale, M., and M. Kissock 197n
C. Reiss 18, 19, 197n, 251, 252, 252n, 253
Halkomelem see Salish languages
Hall, T. A. 130n, 217n, 224, 224n, 232n, 247n
Halle, M. and A. Marantz 102
Halq’eméylem see Salish languages
Hammond, M. 51, 215, 219
Hannah, S. J. 217n
Hargus, S. 197
E. M. Kaisse 7, 197
Harris, J. 7, 224, 252
Harris, J. W. 65, 66, 66n, 147
Harris, Z. D. 212
Hayes, B. 26n, 40, 41, 64n, 112, 115, 157, 190, 214, 253, 254, 255,
256
head:
affixations 233, 241n
boundary effects in 241
function words 249, 250n, 251, 259
prefixes 230, 231, 231n, 232, 233, 241, 244n
suffixation 232, 242
Page 14 of 43
Index
headedness 216, 217n, 218, 222–3, 225, 226, 229, 236, 250, 253,
259
Heath, J. 157
Hebb, D. O. 73
Hebb’s rules 73–7
Hebrew 12, 207, 207n;
see also Colloquial Hebrew; Tiberian Hebrew
alternations 45, 49
base selections 54–6
constraints 50, 52, 53, 55, 58, 62, 64
correspondences, multiple 48
forms:
future 58;
imperative 58;
syllables in 62
fricatives 49, 49n, 59, 60, 60n
phonemes 49;
spirantization 49, 60
generalizations in 52
imperative paradigms 44–5, 45n, 46–64
future in 48;
inflectional 45;
person– number–gender 44–5;
sub-paradigm 45–8;
suffixes in 45, 46n;
tense 45;
words 46
markedness constraints 45, 51, 53, 54, 58, 62
Optimality Theory (OT) 45, 46, 48
constraint-based 48;
Correspondence Theory 46
paradigm uniformity (PU) in 19, 39, 41, 44–64,
Sonority Sequencing Generalization (SSG) in 54
structure 57, 57n
and Existence 53, 54;
identity of 53, 54, 56, 58;
relations 57
verbs 55–6, 57n, 62
with prefixes 60–2
vowels 62
Hendricks, S. 178n
heterosyllabic clusters, in French 260, 261
Hetzron, R. 205n
hiatus 187, 260
‘High Frequency Fusion’ (HFF) 225, 226n, 233, 238n, 239n, 241
Hock, H. H. 17, 21, 102
hollow verbs 159, 160, 161, 162, 164, 164n, 165, 167;
see also verbs
Page 15 of 43
Index
homoganic obstruents see obstruents, homorganic

homophony 152, 168, 256, 257, 264, 270, 271;
see also Hungarian, homophony in
Hooper, J. B. 243n, 252n
Horwood, G. 208
Hukari, T. 299, 300n
R. Peter 298, 304, 311
Hul’q’umi’num’ see Salish languages
Hul’q’umi’num dictionary (Hukari and Peter) 298
Hungarian 168, 169, 264, 265n, 268, 273
affixes in 266, 267n, 268, 273, 285
agglutinative language 268, 269, 273, 293
allomorphy in 236, 267, 273, 285
lexical 270, 271–2, 280, 283–6, 292, 293n, 294, 294n;
suffixes in 292
alternations in 227n, 271
anti-harmony 267, 271, 280, 280n, 282, 283, 286–92
optimality analysis of 281–3
backness harmony in 279, 280–1, 287
cells 291
reference, in 291–2, 292n;
target, in 291– 192, 292n
Colloquial 294n
verbal paradigms in 294–5 ap
con constraints 264, 266–7, 273, 275n, 276, 277, 282–4, 288,
289, 291n, 293–4
consonants in 265n, 271n
constraints 274–6, 279, 281, 282, 283, 285, 286, 291, 291n, 293
in morphology 294;
in phonology 294;
definiteness neutralization in 268–9, 272, 273–7, 280n, 282,
283, 284, 286–9, 291, 293
Optimality Theory of 272, 273
homophony in 271, 272, 276, 277, 277n, 280, 281, 282, 283,
288–9, 291–2
morphemes 270–1, 285
fusional 268
morphology of 278, 289
regularity in 265
morphosyntactics in 268, 273, 275, 293
par constraints 264, 266, 273, 274, 275n, 276, 278, 283–5, 288,
294
paradigm uniformity (PU) of 283–6, 293
(p.338)
paradigms in 265, 266, 284, 284n, 292, 294
dimensions in 269, 272, 292;
phonology of 266, 272n;
phonology of 292, 293
Page 16 of 43
Index
stems in 267, 285, 287, 294

suffixes 268, 269, 271n, 273, 274–7, 279–80, 282, 287–8, 291,
293
alternations of 270;
backness harmony in 279;
constraints in 294n
syllabification 267n
verbs 269, 279
inflection in 265, 267–70, 279;
paradigms in 263–95
vowels 279, 279n, 280, 281, 283, 284, 287–8, 294
Hyman, L. M. 123, 124, 124n, 125, 125n, 126, 128,
131n, 134, 137n, 155
iambic see foot structure
identity 68, 246, 266
constraints 240
effects 237
of the stem 249
‘Impoverishment’ in Distributed Morphology ((DM) 102, 104
Indo-European languages 2
infixation 134–5
inflectional morphemes 184, 184n
inflectional morphology 172, 174, 176, 188, 195, 202, 212n
with optimal paradigms 170–210
words 305
related 3
inflectional paradigms 188n, 207, 170–210, 312; see also paradigms
in causative doubling 139
overapplication v. underapplication in 206
inflectional prefixes 188, 193
inflectional suffixes 195–6, 203
inflections 182
endings with 249
of prefixes 188–97
Inkelas, S. 134
Input–Output (IO) 152, 184, 301, 302
inputs 6, 9, 11, 12
intraparadigmatic 181, 185–6, 196, 199, 200, 204, 205, 208
faithfulness constraint in 201
IO-Faithfulness constraints 40, 41, 53, 121, 146, 152, 173, 175, 301;
see also under constraints; Salish languages
allomorphy in 70 anti-faithfulness 150
clustering of 69 mapping 69 146, 159, 167
optimal paradigms (OP) and 181, 183, 184, 187, 198, 200, 201,
204, 208;
paradigm uniformity (PU) and 67, 71, 72, 84–5, 87–92, 92n, 93
i-stem nouns, in Latin see Latin, nouns
Iraqi Arabic 191; see also Arabic
Page 17 of 43
Index
Italian 88–9, 99, 102, 217n

affixes in 101
conjugation in 95, 100n
stress in 89, 100, 100n
Itô, J.
Y. Kitagawa and A. Mester 178n
A. Mester 146, 197n, 207n
Iverson, G. and J. Salmons 117
Jakobson R. 151, 152
Japanese 176
Jeffers, R. J. and I. Lehiste 5n, 263
Jensen, J. T. 104, 109, 110, 111, 112, 117, 118
Jita 124n, 125n, 130, 130n, 133, 137, 141n
affixes in 134–5, 138
causative doubling in 121–44
consonants 123, 137;
morphemes 123
morphology of 123, 126, 127, 128, 129, 134, 138, 141
paradigms in 134, 138, 139, 140
phonetics 130
phonology in 127, 128, 130, 134–5, 142
pronunciation of 129, 142, 142n
suffixes in 135, 138
words in 134, 139–40, 142, 145
Johnson, W. 260
Jones, D. 212, 261n
Jones, M. 3045
Jun, J. 107
Jusczyk, P W.
P A. Luce and J. C. Luce 26n
et al. 26n
Kager, R. 9, nn, 12, 71, 122, 134, 146, 190, 267n
Kahn, D. 108, 109, 110, 221, 222, 238, 238n, 245, 250n, 251, 254
Kaisse, E. M. and P. A. Shaw 7
Kaye, J. 264
J. Lowenstamm and J.-R. Vergnaud 223
Kelepir, M. 150
Kenesei, I., R. M. Vago and A. Fenyvesi 268
Kenstowicz, M. 9, 12, 13, 15, 18, 19, 46, 47n, 53, 71, 81n, 89, 123,
134, 139, 146, 148, 171, 172, 172n, 173, 188n, 197n, 205, 207n,
212, 250, 264n, 265, 292
Kieckers, E. 35n, 36, 36n
Kinande 125, 125n
King, R. 6
Kinkade, M. D. 298
Kiparsky, P 6, 7, 8, 9n, 11n, 12n, 18, 21, 68, 69, 82, 90, 93, 112, 117,
126, 129, 134, 143, 146, 172, 197, 197n, 263n, 296
Kirchner, R. 89, 107, 150
Page 18 of 43
Index
Kisseberth, C. 184
M. I. Abasheikh 153, 154, 155n, 168, 184
Klausenburger, J. 30n
(p.339) Korean 39, 41, 162
alternations in 40–1
base forms in 15
markedness constraint 40
coronal obstruents in 40
faithfulness constraint 40
obstruents in 41
paradigms:
inflectional in 15;
leveling of 40;
nouns 40
phonotactics of 40
Kosati, plurals in 48n
Kraska-Szlenk, I. 146, 148, 172, 212
Kuhl, P. and P. Iverson 72
Kühner, R. 28, 36
Kurisu, K. 48n
Kurylowicz, J. 5, 18, 95, 172
labiodental fricatives see fricatives, nasal
Ladefoged, P. 108
Lahiri, A. 167n
B. E. Dresher 22, 41, 146
languages 174, 176, 177, 186, 189, 192, 199, 200, 205, 206, 223n,
231n, 237, 252, 265
learners 23, 26, 27, 30, 33–4, 37, 232, 252
minimal generalization 31;
of Latin nouns 28, 33
phonology of 49
Triebe 5
universals of 78
LaSor, W. S. 60
Latin 22n, 33n, 41, 45, 65, 159, 172, 242n, 286
alternations in 17, 30, 35, 41
o-o 19; s-r 17–18, 35
analogy 22
notation in 18
forms:
ablative 28, 30, 32, 33;
base 15, 32;
dative 28, 30, 32, 33;
genitive 28, 30, 32–6;
nominative 29, 31, 32, 33, 33n, 38, 34–6, 41;
oblique 32, 33, 33n, 34–6, 36n, 38, 41
genders and word lengths 37
grammar 2, 34, 35
Page 19 of 43
Index
learners of 33, 35
non-alternations 36
nouns 28–33, 30n, 38, 41
abstract 38–9; agentives 18, 34, 34f, 35, 39; analysis of
36–41; base status 42–3; declensions of 28, 36; i-stem
32, 32n, 33, 33n; monosyllables 35, 37; neuter 37;
nominative 30, 36, 36n, 37– 9; paradigms of 28–3, 33,
36, 173n; polysyllables, neuters 35; polysyllables, non-
neuters 21, 33, 35–7, 38–9; preference category 34f;
token frequencies in 38
paradigms, inflectional in 15
rhotacism in 17, 30, 30n, 36n, 41
in grammar 34; nominative case 19
stress rule 65–6, 157
subgrammars in 31
voicing in 30
vowels, long 17
word-final sonorants 17
lax vowels 244, 245, 251; see also vowels
learners, of languages 23, 26, 27, 30, 33–4, 37, 232, 252n, 257
minimal generalization 31
lenition 211, 221–3, 253
Leslie, A. 298, 300, 305
Leumann, M. 21, 28, 36n
leveling:
of alternations 18
of paradigms 3, 5, 18, 21, 37, 41, 214, 252
allomorphy in 37;
in German 4, 5;
majority rules in 37;
morphology of 17–43
Levin, J. 177
lexemes see stems
lexical allomorphy 267, 268, 271–2; see also Hungarian
lexical items 151, 167, 181, 251
lexical phonology 197, 199, 261, 263
Lexical phonology and morphology (Kiparsky) 6, 7, 68, 82, 93, 146
lexical representation 213, 224, 252, 253, 261
lexical strata see paradigm uniformity (PU)
lexical suffixes, in Salish languages 310
‘Lexicon Optimization’ 85, 87, 251
lexicon 72, 84, 87, 90, 94–5, 98, 101, 102, 180, 224, 232, 246n, 251,
252, 252n, 261
and grammar 35
conservatism, analysis of 22n
items in 138
non-alternation in the 39
paradigmatic relations in the 214
Page 20 of 43
Index
constraints 214
phonetics 213
Spanish 149
stress 67
‘limited control’ object paradigm 14
Lindblom, B. 70, 77
linguistics 44, 147, 171, 173, 262, 263, 301; see also generative
linguistics
European pre-Structuralist 2
pre-generative 1–5
liquids 238n, 240; see also glides
Lowenstamm, J. 260
Luick, K. 250n
l-velarization, in English 253–6
McCarthy, J. J. 11, 13, 14, 15, 21, 37, 53, 60n, 71, 81, 112, 113, 123,
130, 132, 133, 134, 139, 141, 142, 143, 146, 157, 161, 173, 177, 178,
180, 181, 186, 191n, 196, 197n, 206, 206n, 207, 209, 212n, 214,
231n, 250n, 298, 301, 303, 305, 308n, 312 (p.340)
A. Prince 9, 46, 53, 63, 135, 175, 177, 178, 178n, 187, 189,
189n, 190, 197n, 207n, 210, 213, 264
‘majority rules’ hypothesis 4, 37, 176, 201, 202, 205, 205n, 206,
209, 307
Makan 167
Malone, J. L. 207
Manczak, W. 5, 38
mapping, from syntactic to prosodic structure 217
Marantz, A. 177
Marcus, G. et al. 35
markedness constraints 14, 15, 72, 86, 87, 89–90, 130, 139, 224,
231, 234, 236, 259, 301
and alignment constraints 230
local, and semantics 39
of attractors 175
of superheavy syllables 182
of verbal inflections 293
optimal paradigms (OP) and 177, 178, 181, 182, 183, 184, 185,
186, 187, 188, 189, 194, 195, 196, 197, 198, 200–8, 208n
paradigm uniformity (PU) and 150, 173, 176,
Marshallese 150
Martin, J. 48n
Martin, S. E. 39, 40
Mascaró, J. 147
material groups see groups, material
Mayerthaler, W. 6n
Meiser, G. 36n
mental:
learning 15
lexicon 5, 78, 83
Page 21 of 43
Index
paradigms in the 262

representations 78
Mester, R. A. 159
metathetic forms 201
metrical:
consistency 66, 76, 88, 91–2, 98, 99, 100, 172, 173
of stems 101
scansion 257
well-formedness 76, 88–9
Mikheev, A. 25
minimal:
confidence statistics and 33
generalizations and environments 25f
pairs 145
prefixes 230
Misteli, F. 5
Mitchell, T. F. 157, 161, 162, 166
modifiers 215, 216, 217, 227, 228–9, 231, 250n
monosyllabic feet in 218
prefixes 231
Mohanan, K. P. 7, 197
monomorphemes 8, 109, 114, 116, 120; see also morphemes
monosyllables 22, 22n; see also Latin, nouns; nouns; syllables
in metrical scansion 257
in nouns 21
morphemes with 223
suffixes with 217, 218
Moroccan:
Arabic 162–3, 202, 202n, 203, 205
dialects 157, 164n
verbal paradigms 203
morphemes 68, 80, 86, 104, 126, 128, 171, 172, 258, 260, 261, 264,
265 see also allomorphy
and cyclicity 128
causative:
in Bantu languages 123; in Jita 123; palatal glide 123
cohesion of consonantal 259
discontinued 80–1
doubling 128n
in English 81n
in Kinande 125
initials of 261
monosyllables in 218, 223
sized units 70, 71
vowel-initial 262
morphologically-complex words 215, 254
metrical structure of 300
morphology 150, 169, 211, 216, 232n, 245, 255, 263
Page 22 of 43
Index
see also distributed morphology; inflectional morphology;

morphophonology additive 46
alternations in 105
and phonology 67, 87–90
Arabic 81
base priority in 9, 12, 12n
concatenative 44, 80–1
cohesion in 246
constituents in 213, 214
constraints in 51, 82–7
cycle in 11
English and 149
environments in 26
forms, related in 12, 84
irregularity in 92–3, 99–100, 138
markedness constraints 87
of German words 6
optimal paradigms in 171, 172, 177, 188, 197, 199
paradigms, leveling 17–43, 102–5
processes 25, 51
regularity in 87, 88, 93, 106
‘root and pattern’ 81
rule inductions 27
minimal generalization algorithm 22, 24
stress in 66
words, related 13, 84–5
morphophonemes 41, 154, 168; see also allomorphy; phonemes
(p.341) morphophonology 236, 246, 248; see also phonology
affixes in 240
non-neutral 240
of English 224–37
pwords in 218
structure and alignment 226, 231
morphosyntactics 214, 217, 218, 227n, 236, 265, 266, 267, 270;
see also Hungarian, morphosyntactics in boundaries 219
properties 234
structures in 227, 230, 232, 233, 235, 241, 260
prosody 237
Mutaka, N. M. 125
L. M. Hyman 125
mwords 142, 142n, 143
stems, in Jita 143
Myers, S. 119, 142, 211
native 242
coinages 226n
word formation 236, 242n
Neogrammarians 1, 2
Nespor, M. and I. Vogel 117, 216, 217
Page 23 of 43
Index
Neue, C. F. and C. Wagener 34

neural computation 74, 77
neutrality 220, 240
neutralizations 28, 31, 36n, 105, 216
codas, in Korean 41
in the nominative 29; see also nominative
in the genitive 29; see also genitive
nouns 29; see also nouns
of Optimality Theory (OT) 71
‘opacity’ of 71
segments 71
under gradient attraction (GA) 100
Newman, S. S. 232n–3n
nominals 203n
feminine 161
suffixes in 162; see also suffixes
templates 181–2; see also templates
nominative 26
case 5
endings 17, 23–4
and genitive endings 24
forms in Latin see Latin, forms
rules 24n
singular 22n, 23
non-agentive nouns 39; see also agentive, nouns;
nouns
non-alternations, in Latin 39; see also alternations;
Latin
noncoronal obstruents 211; see also obstruents
noncohesion 218
affixation in 233n, 240, 242
of prefixes 237
in morphology 246
in suffixes 236–7, 240
in word formation 236
tense in 249
non-optimal paradigms 15–16, 196–7; see also paradigms
nonrecursivity 235
non-trivial paradigms 204
North African Arabic dialects 162–4, 165
nouns 153, 179, 184, 188, 192, 193, 195, 202, 203, 204, 205, 207,
243, 244
agentives see Latin, nouns
changes in 39
monosyllables 21
neuter 35n, 152
non-agentive 39
polysyllables 21
Page 24 of 43
Index
leveling of 21
paradigms 18
in Latin 31
stems of 179, 186, 192
and vowels 187
templates 179, 189, 193
Noyer, R. 102, 104
nucleus 259
position 257
numerals, sound changes in 5n
object agreement paradigms 300, 303, 307
and aspectual markings 312
oblique forms, in Latin see Latin, forms obstruents:
clusters in 26, 29
homorganic 72
Korean 41
coronal 40
voiced 7
ambisyllabic 246
Yucatec Maya 72
Old English 102, 243, 249n–50n; see also English Old French 242n;
see also French
onset 224, 232, 235, 239, 240, 243, 252, 254, 254, 257, 260, 261,
261n, 262, 312
position 255, 257
syllabification of 221, 222, 223, 223n, 230, 238, 245
o–o alternations, in Latin 19
OP-Faith constraints 312
opacity 4, 11, 12n, 71, 145–69
optimal paradigms (OP) 9–16, 18, 123, 143, 170–3, 173n, 174–207,
207n, 208–10 see also under morphology; prefixes; Salish
languages; suffixes; syllables; verbs
causative:
doubling 123, 129–30, 133, 135–6, 143–4; languages
134, 136
constraints 171, 173, 175, 177, 182, 184, 187, 191, 192, 193–4,
200, 201, 202, 203,
cyclicity analysis in 134–43
(p.342)
faithfulness constraints 1174n, 178, 181, 182, 183, 184, 185,
187, 188, 189, 190, 194, 195, 196, 197, 199, 201, 204, 205
formalism 132
phonology with inflectional morphology 170–210
Optimality Theory (OT) 1, 16, 67, 122, 138, 146, 152, 22, see also
under Hebrew; Hungarian; Salish languages
alignment constraints in 213
analyses 116–21
Page 25 of 43
Index
of cyclicity 129
constraints in 87, 199, 171, 279
cyclicity in 12n
faithfulness constraints in 71, 132, 146
in morphology 51
Input–Output faithfulness (IO-F) in 84, 90, 146
markedness constraints in 75
neutralizations in 71
‘opacity’ in 71
optimal paradigms (OP) and 173, 175, 180, 199–201, 205, 206
Output–Output Faithfulness (OO-F) 9, 12, 83–4, 90, 105
correspondence 263
paradigm uniformity (PU) and 66, 67, 83–4, 85, 89, 93
reduplication in 206
serial 12n
Oran dialect 164
Orgun, C. O. 9n, 12n
Osthoff, H. and K. Brugman 5n
Oujda 162–4, 165
Output-Output Faithfulness (OO-F): 9–11, 11n, 12–14, 16, 21,
46, 85, 86–92, 92n, 93, 96n, 100–1, 106, 116, 168, 170, 171,
174, 206, 302, 302
see also constraints; Optimality Theory (OT)
and morphology 67
anti-faithfulness correspondence 13
base-priority approach to 13–15
constraints 123, 135, 212, 263–5, 267
correspondence 122, 140, 143, 144, 171
in optimality theory 264;
of causative doubling 130, 132;
relations 240
foot structure (OO-foot structure) 119–21
symmetrical model 14–15
outputs 6, 9, 11, 12
identity relationships 6, 13
of words in morphology 16
optimal 11
strings 14
overapplication v. underapplication:
in inflectional paradigms 206
of vowel reduction 309–12
optimal paradigms (OP) and 174, 174n, 175–6, 181,
187, 188, 188n, 197, 201, 202, 206–7, 207n, 208, 209
paradigm uniformity (PU) in 249, 250, 251
Padgett, J. 70, 81n
palatal harmony 126, 127, 128, 151
Palestinian dialects 162
par constraints see Hungarian, par constraints
Page 26 of 43
Index
paradigm uniformity (PU) 65–96, 96n, 97–106, 122–3, 137, 168–9,

172, 184, 201, 212, 212n, 224, 232, 234, 236, 237, 239, 240, 243,
245–6, 250, 251, 255, 258, 264, 282, 296–7;
see also under containment; Hebrew; Hungarian; prefixes;
‘Representational Entailments Hypothesis’ (REH); Salish
languages; schwa; simplexes; verbs; words
affixes in 82–4, 87, 93, 101, 103, 104, 106, 94–5, 95n, 96–100,
215, 216, 218–20, 227, 229, 232, 234, 235, 236, 239, 240
alignment in 16
allomorphy and 150n, 152, 157, 161, 162, 163, 165, 168
analysis in 251, 256, 260, 261, 262
and root priority 296–312
constraints 16, 18, 22, 41, 213, 213n, 214, 215, 240n, 246, 248, 251,
252, 255, 258, 261, 265, 267, 281
in truncations 253;
inflectional 13, 15,
97; non-constrastive features 16; on
vowels 255;
doubling in 134
effects 107–21, 114–16, 121, 122, 214, 215, 218, 220, 238,
238n, 241, 243–6, 248, 252, 253, 255, 256, 258, 262, 262n, 300
durational, in French 256–62; in
gerunds 244; in English dialects 253; in
vowels 255; prefixes 262
in cohesion, affixation 243
optimal paradigms (OP) in 134
Optimality Theory (OT) and 83–4 phonology,
theory and 1, 2, 5, 5n, 6–8, 13–14, 16
cyclic application of 296
noncontrastive details in 107
stress 245
versus boundary effects 211–62
paradigmatic relationships 263–4, 268–9, 297,
paradigmatic uniformity and contrasts 145–69, 266, 267
in allomorphy 265
paradigms 23–33, 44–6, 92, 105, 106, 109; see also Hebrew;
nouns; Hungarian, paradigms in; Jita, paradigms in; leveling, of
paradigms; optimal paradigms; Optimality Theory (OT)
alternations in 182
aspectual see Salish languages, paradigms base-priority
approach to 13–14
derivational 15
constraints, rankings of 139–40
(p.343)
contrasts in 1, 5, 13–14, 16, 151, 153, 162–4, 165–8, 264
inflectional 15
correspondence relation 301, 301n
forms 37, 38–9
Page 27 of 43
Index
in pre-generative linguistics 1–5

members 38–9, 207, 264, 267
morphology of 296
non-uniform 47, 64
object agreement 300, 303, 307
optimal paradigms (OP) 170–1, 173n, 176, 177, 180–1, 184,
185, 185n, 188n, 189, 204, 205, 208
phonology of 1, 4, 5–9, 15, 16, 296
optimization 130
rule-induction 24–7
metrics of 24; minimal generalization model 24–7
space in 265
structural:
identity 47, 47n;
relations 47;
Uniform Exponence 47
sub-paradigm 46, 47
symmetrical approach 13–14, 16
‘Parallel Distributed Processing’ 67
parallelism 9, 11, 67, 224
parsing 242, 259
affix-based 232
participles 101, 161
passive glide 135; see also glides
Pater, J. 109, 110, 111, 112, 114, 116, 119, 172, 262n
Paul, H. 3, 4, 4n, 5, 263
Peperkamp, S. 216n, 217n, 233n
Philadelphia English 248; see also American
English
phonemes 72–3, 79, 80, 108, 258
transcriptions of 30
phonemic representations 213n, 224, 252
phonetics 31, 107, 110, 116, 121, 212, 213n, 228, 238
analogy see analogy, phonetic
and distance 70–3
codas, in 252
contrasts in 213, 256, 257, 257n
correspondence in 257, 260
identity in 246
‘initial’ 214
noncontrastive details and 107
of perceptual contrast 137
of suffix-initials 21, 256n, 254, 257, 258, 262, 262n
of vowels 255
paradigm uniformity (PU) in 251, 252
syllable-conditioned 252
phonology:
Page 28 of 43
Index
see also cyclicity; Hungarian, phonology of; Jita, phonology of;

Russian
agentives in 35n
alternations in 84, 93, 137
and distance 70–3
attractions in 81
base-priority in 9, 12, 13
constituents of 213
constraints 146, 261, 267, 283
cycle and 146
grammar and 145–69
markedness constraints 84, 85, 212
morphology in 25, 26, 30, 35n, 67, 87–90
structure of 297
of suffixes 212, 240
optimal paradigms (OP) in 150–7, 172, 173, 175, 179, 182, 184,
186, 197, 198, 199, 205, 205n
paradigm uniformity (PU) and 72, 82, 93, 105–7, 109, 114, 115,
214, 218, 226, 232, 236, 237-8, 244, 245, 253,
paradigms in 1–12, 12n, 13–16
cyclic application in 296
processes 297
quirky 162, 165–8
regular 87, 88, 93, 106
violations of 212
stems 12n, 130
surface exceptions in 9
symbols in 73–4
verb stem templates in 184
words 12n, 220n
phonology-phonetics interfaces 213, 214
phonotactics 159, 181, 204
constraints 19
distributions 26n
of Korean 40
Piggott, G. L. 223, 223n
Pinker, S. 93
A. Prince 24
Plautus, T. M. 17n
Poletto, R. 123, 125n
Polish 188n, 207n
syllables in 148
polysyllables 21–2; see also Latin, nouns; nouns; syllables
post-Neogrammarian era 5
postvocalic consonants 221; see also consonants
Potter, B. 197
pre-generative linguistics see linguistics, pre-generative
prefixation 227, 229
Page 29 of 43
Index
prefixed words:
aspiration in 245
prosody of 226, 228
prefixes:
in English 224
inflections in 188–97
monosyllables, in 230
optimal paradigms (OP) of 172n, 179, 188, 193, 199
(p.344)
paradigm uniformity (PU) and 226, 227, 228, 242, 245, 259
syntactic properties of 227
pretonic foot 227; see also foot structure
pretonic position see onset, position
Prince, A. 61n, 191, 207, 208n
P. Smolensky 9, 11, 66, 85, 86, 87, 157, 171, 180, 181, 196, 251,
301
Principles of English stress (PES) (Burzio) 66, 67, 68, 76, 84, 87, 88,
91, 92, 93, 95, 97, 98
privative meaning 232
pronunciation 1, 13, 138;
see also speech
careful 261
casual 247
in US English 11
of attractors 255
of consonants 255
of English 11n, 242
dialects 253
paradigm uniformity (PU) and 226–7, 228, 233–4, 238n, 246,
248 n, 250n, 254
Russian 13, 151
‘proper nesting’ constraint 216n
prosodic categories 217
phonetic grounds of 227
Prosodic Consistency Criterion 261
prosodic contrasts 227, 233
prosodic domination 236
prosodic fusion 231, 242n
prosodic hierarchy 216, 217
prosodic properties 215, 241
prosodic representations 226, 255
prosodic structures 177, 189, 192, 215, 218, 219, 225, 231, 232n,
238, 241, 246, 248, 249, 254, 256
and syntactic structures 219
mapping in 217
of words 225–6
stems in 241
prosody 190, 228, 229, 236, 240
Page 30 of 43
Index
boundaries in 236, 247

and morphology 16, 214; constituents in 216
in English morphology 224–37
consistency in 244–6, 256, 261
domains in 216
English 218
morphology in 177, 178, 187
neutralization in 261
of compounds 225
organization in 214
words in 220, 220n
of English 254; prefixed 226, 227, 229
prothesis 189
pstems 131, 131n, 132, 133, 135, 141, 143
psycholinguistics, neighbourhood effects in 79
psychologically motivated analogical changes see
analogical
Pulgram, E. 261n
‘push chains’ 150 changes, psychologically
motivated
pwords 142, 142n, 214–18, 219n, 223, 227, 228, 259, 262n
and syntactic boundaries 217n, 218, 219, 221
boundaries 235, 236n, 246, 249
external syllables in 239
feet and 239
final:
boundary effects 245;
codas, phonetics of 252;
consonants 223, 224n, 233, 238, 251–2
fusion in 230
in adjectives 246
in English 224
in verbs 246
of the stems 219, 234, 235
prefixes in 226
pretonic syllables in 222; see also syllables
stress in 220
structures of 221, 224, 235, 245
affixations 232n
suffixes in 233
syllables in 220
voiceless ambisyllabic consonants in 247
Raffelsiefen, R. 9, 16, 108, 146, 149, 173, 175, 200, 212, 212n, 224,
224n, 226, 229, 236, 236n, 239, 239n, 243n
‘readjustment rules’ 92–3
Rebrus, P 268, 270n, 271n, 293n
M. Törkenczy 151, 250
recursion formalism 12
Page 31 of 43
Index
reduction 150–2, 153, 156, 300;

see also Salish languages, vowels
reduplicants in 9, 307
reduplication 177, 207, 307
in Optimality Theory (OT) 206
in schwa see Salish languages, schwa in
morphology 177
of words 298, 303
parallels 175
phonology interactions 206
Repp, B. H. 72
‘Representational Entailments Hypothesis’ (REH):
constraints in 75
paradigm uniformity (PU) and 74–7, 78, 79, 81, 82, 83, 92, 98,
102, 104, 105–6
representations 81n, 83, 84, 86, 92, 105, 106, 186, 214, 224, 227,
240
entailments in 90–1, 97
in semantics 98
morpho-syntactic 129, 134
neighboring 79
space in 78
(p.345)
time 79–80
uniform interaction in 78
rhotacism:
constraints in 21
Latin 30n, 36n
in Latin 17, 19, 30, 30n
nominative case 19
in suffixes 18
rhymes 212, 246n
of words 213
structure 223, 232
superheavy 223
trimoraic 224n, 249n
‘rhythm rule’ 236n
Rialland, A. 256, 257, 257n, 258n, 260, 261
richness of the base 11, 180, 181, 182
Roca, J. 65, 66, 197n
Romance languages 226n
first conjugation in 95
nouns in 36
stress in 65–6
root-and-pattern morphology 81, 177, 131
roots 300
priority of 312
and paradigm uniformity 296–312
Page 32 of 43
Index
Rose, S. 200
Rosenthall, S. 200
Rubach, J. 9, 146, 197, 197n
rule 187
based phonolgy 1, 8, 206
generative phonology see phonology, rule-based
generative
in Latin grammar 32, 34
induction 24–7
ordering 1, 6, 7
and the cycle 1, 6
phonology 201
Russell, K. 51
Russian 13, 147, 150–2, 156, 168, 169
cyclic model in 13
phonology of 13
pronunciation 13
verbs in 13
Sagey, E. 81n
Salish languages (BC, Canada) 264, 298–312
ablaut in 304, 305
allomorphy in 300, 304, 305
alternations in 297–8
constraints 303, 307, 308
clash avoidance 303, 311; rankings 309
derivational theory 309–12
faithfulness constraints 302–3, 306, 308
glottalization in 304
Halkomelem 299, 302–4, 308n, 311
Upriver 297–8
Halq’eméylem 298–300, 305
Hul’q’umi’num’ 298, 299, 300n, 304, 310–n
schwa reduction 301; stress in 301
markedness constraints 302–3, 306, 308
optimal paradigms in 309–12
OP-Faith 303, 306, 308, 308n Optimality Theory (OT) in 311–12
Output–Output correspondence 308n
paradigm uniformity (PU) in 303–5, 309
paradigms 302–3, 305, 307, 308, 309, 311–12
aspectual 308; object agreement 300, 302, 307, 308,
310
reduplication in 304–5, 307–8, 311
roots/stems in 298–302, 304, 305, 312
root priority 309
schwa in 298–306, 310
reduction 310;
stress 304, 309, 310
suffixes in 300, 311
Page 33 of 43
Index
lexical 310
vowels 304–6, 309–12
reduction 303, 306, 310, 311;
root 308;
stress in 299, 310
words in 304–5
Sanders, N. 188n
Sanskrit 69
Schane, S. 260
Scherer, J.
Schmidt, J. 3
Schuchardt, H. 5
schwa 14, 151, 157, 158, 160, 162, 163
see also under Salish languages in English 145–6
loss 256, 257, 257n, 258, 258n, 259, 260, 261
paradigm uniformity (PU) and 203, 208, 236n, 243n, 244n,
246, 252n, 256, 257, 257n, 259–60
violations of 244
‘second law of analogy’ (Kurylowicz) 95–6
segmental neutralizations see neutralizations,
segmental
Selkirk, E. O. 109, 158, 172, 213, 216, 229, 235, 250n
semantics 5n, 84, 128n, 243, 244, 256
affinity in 244, 244n
and local markedness 39
dissociation in 256
drifts in 244, 249
naturalness in 37
shortening, in Latin 19
Semitic languages 81, 178
morphology in 45, 81
sequences, illegal 26, 26n
in nouns 30
word boundaries in 30
Shaw, P. et al. 299, 303
Shere, T. 183
Sievers, E. 2
Silverman, D. 107
(p.346) simplexes 8, 171
paradigm uniformity (PU) and 214, 215, 217, 218, 221, 227,
227n, 230, 241, 241n, 242, 245, 246, 247, 248, 255, 261
phonetics of 258
polysyllabic 245
Siptár, P. and M. Törkenczy 268
Skutsch, O. 17n
Smolensky, P, L. Davidson and P Jusczyk 26n
Soh, H.-L. 150
Page 34 of 43
Index
sonorants; see also obstruents

coronal 126
de-stressing 188
Hebrew 54
Latin 17
sonority 202–3, 203n, 204, 205, 208n, 221, 252n Sonority
Sequencing Generalization (SSG) see Hebrew
sound:
changes, theory of 1, 2, 5
exceptionless of 2; numerals 5n
features 243
in verbs 178n, 200
laws 2
properties 211, 218, 243, 244, 245, 261
shape of words 2
structures 246
Sound Pattern of English, The (Chomsky and Halle) 6
Spaelti, P. 178n, 307
Spanish 65–6, 71, 95, 106, 149n, 150n, 172n
affixes in 147
allomorphy in 147
lexicon 149n
nouns 65–6
paradigm uniformity in 149
stress 65–7
syllables in 147–8
verbs 65–6
spatio-temporal proximity 81, 81n
SPE theory 6, 146, 170, 263
speech 37
articulation in 2
careful 247, 259, 258n
casual 246, 247, 256
fast 246
register, sensitivity to 246–7
sounds 214
pronunciations of 247
Speiser, E. A. 207
spelling 232, 252n, 253, 256
Spencer, A. 174
spirantization 130–3, 136, 137, 140, 154, 207n
see Hebrew, fricatives
causative doubling 143–4; see also causative doubling
in Bantu languages 137n
of consonants 123, 137
stem-final 239n
s ~ r alternations, in Latin 17–18
Page 35 of 43
Index
Stampe, D. 181
‘Stampean occultation’ 181, 189
stem-final consonants see consonants, stem-final
stems 7, 95, 97n, 98, 99, 100, 138
see also under bound stems; free stems; Hungarian; pstems;
verbs Salish languages, roots/stems; words
base and 297
bound 91–2, 93
causative see causative, stems
metrical consistency of 101
morphology of 130
mwords and 143
of nouns 179, 180
of verbs 180
of words 242, 265
optimal paradigms (OP) in 172, 173, 173n, 174, 176, 178, 200,
203, 205, 207
paradigm uniformity (PU) 228, 232, 233 n, 236, 240, 241n, 244,
261
stress 101, 230n
templates 184n
uniformity of 94
Steriade, D. 9, 12, 16, 18, 22n, 52, 62, 68, 69, 71, 102, 107, 108, 109,
110, 114, 116, 121, 123, 137, 147, 150, 213, 214, 257, 258, 258n,
260, 261–2, 262n, 263, 264, 265, 267n
Stojkove, S. 152
stops 25, 107–21, 245, 246, 247, 260
aspiration in 110–12110–11, 121
and non-final 110–11; and word-final dactyls 110–11, 121
foot-initial 112–13
unaspirated 119
voiceless 110–12, 117
Strata OT 197, 197n
Strauss, S. 197
stress 68, 70, 100, 153, 157, 159, 161–2, 166, 168, 173n, 227–8,
237–8, 239n, 246, 259, 267n
see also English, stress in; Italian, stress in
constraint 121, 190
in allomorphy 101
analysis of 10
identity 11
in suffixation 243
neutrality 218
of verbs 95
patterns 190, 215
schwa 14
in English 145–6
shift 218
Page 36 of 43
Index
stems 101
violations 219
(p.347)
stress-behavior 220, 240
stress-neutrality 219, 237, 239
vowel-initial suffixes in 220
stress-related diagnostics 222
stress shift 236, 236n, 239 –40
Stress to Weight Principle (SWP) 190–2, 192n, 193, 193n, 194, 195,
196, 197, 199
stressed vowels 233, 246; see also vowels
stressless:
syllables 211, 236n, 239n; see also syllables
stems of 230
vowels 234
Strict Layer Hypothesis 216, 216n
structure:
identity of, in Hebrew see Hebrew, structure, identity of
Struijke, C. 178n, 303, 307
sub-paradigm see paradigm, sub
subgrammars 27–8, 31, 33
forms in 27
in Latin 31
rules with confidence values 31
rules in 27–8
metrics of 2
subphonemes 260; see also phonemes
Sudanese Arabic 71
suffixation 231–2, 233, 236,
stress in 243
suffixed words 212, 231, 247, 254n
suffixes 8, 82, 97, 100, 127, 300 see also under affixes;
dentalization; Hungarian; Jita; prefixes; Salish languages
causative 126
stems 124
class 8
derivational 7, 123
English 224
forms, rhotacism in Latin 17, 18
Hungarian 266, 268
Latin 29
nouns 33
inflectional 7
junction characteristics of 232n
optimal paradigms (OP) in 176, 179, 180n, 181, 182, 184, 185,
187, 191, 195, 199, 201, 203, 217, 218, 230n, 233, 234, 234n,
235, 237, 239, 239n, 173n, 174, 174n, 240, 240n, 242n, 243–4,
249, 256, 265, 279n
Page 37 of 43
Index
paradigm uniformity (PU) 153, 154, 155, 158, 214, 160, 161,
162, 163, 166
phonology of 240
in Jita 137
Spanish 147
stress of 99
syntactic representation of 219
suffixing inflection 182
superfoot see dactyls, sequences in
superheavy syllables 180–1, 181n, 183, 184; see also syllables
markedness of 182
surface:
exceptions 9
forms 85, 87, 172, 212, 264
base-priority and 312;
outputs 5
prosody 216
resemblance 170–2, 173, 187, 197–9, 214
shapes, of words 263
stem-finals 186
Suttles, W. 299, 305
Swedish 234n
syllabification 189, 267n
domains of 254, 258
in French 259, 261n
lexical versus phonetic 251
onset, of consonants 238
paradigm uniformity (PU) and 220, 221, 222, 224n, 228-22,
231n, 234, 260
syllable-conditioned:
allophony 252
phonetics 252
syllable-structures 257, 258, 259, 260
diagnostics 221
syllables 8, 111, 157, 215, 220, 220n, 223, 224, 234, 248, 257, 259
count of 257
faith-stressed 52–3
flapping in 121
in Latin 17n
intitial consonants 214
loss of 257
onset 221
optimal paradigms (OP) and 160, 162, 168, 170, 177, 181–3,
187, 188–92, 194, 197, 199, 200, 202, 202n, 204, 261
positions, final 246
Spanish 147–8
stressed 119, 221
structure of 246n
Page 38 of 43
Index
symbols, in phonology 73–4

symmetry 171
approach 13–15
‘Sympathy Theory (McCarthy) 71
synchronic phonology 16, 107–8, 172
syncopes 100, 157, 159, 160–1, 162, 164n, 166, 191, 191n, 192
syncretism 101, 104n, 105
in morphology; see morphology, leveling in
syntactic structures and prosodic structures 219
syntax:
classification of 229
distinctions in 258
minimalism 178
structures in 226
(p.348)
words 235
syntax–prosody:
interfaces 213
mapping 216, 241
Szpyra, J. 215, 217n, 219, 220, 229
Tagalog 135
tautosyllables 11, 201
templates 177–8, 179, 188–97
foot-based 178
stems 184n
syllable-and-mora 178
of templates 186
satisfaction, condition of 117
verb stem 184
tensing rule 249, 249n
vowels 244; see also vowels
Tesar, B. and P. Smolensky 173
Tiberian Hebrew 45, 48, 56–64, 59, 59n, 60, 60n, 61–4, 207, 207n,
208;
see also Colloquial Hebrew; Hebrew
fricatives in 57, 59n
imperative paradigms 57–64
jussive in 207–8
prefixes in 48 schwa 57
structure, identity of 57
truncation in 59, 63
Tibetan 221, 222
Tiersma, P. 39
‘tightness’ of associations see paradigms,
associations
time representations see representations, time
Tobler, L 2
Page 39 of 43
Index
token frequencies 31, 37, 39

of paradigm members 38–9
Tokyo Japanese 207n
Tranel, B. 95
Transderivational correspondence theory (TCT) (Benua) 9, 11–
13, 134, 139, 141–4, 171, 172, 172, 173, 174, 188, 207, 207n,
240
transitivity effects see entailments, transitivity
effects in
Triebe :
‘forces’ 2
in languages 5
triconsonantal clusters 185, 203, 204
Trigrad dialect 152
trochaic foot see foot structure
Trubetzkoy, N. 213, 213n
truncations 12n, 252, 253; see also Hebrew,
Colloquial
correspondence in 251
cyclicity in 11n
opacity in 11n, 12n
phonology of 253
words 253
Turkish paradigm 44
Umeda, N. and C. H. Coker 211, 214, 221
underapplication see overapplication v. underapplication
‘Underlying Representation’ (UR) 84, 85–6, 87, 101,
and allomorphy 70
unflapping see flapping
Uniform Exponence (UE) 13, 21, 47, 172, 172n, 173, 188n, 264n
constraints 41; see also constraints
Korean 40
Latin 17n, 19–20, 21
violation 19
uniform paradigms see paradigm uniformity (PU) universal:
distinctiveness 261
grammar 177
Upriver Halkomelem see Salish languages
Urbanczyk, S. 14, 178n, 181, 299, 303, 305, 307, 308
U.S.Salish languages English see American English
Ussishkin, A. 178, 178n
Vago, R. M. 268
van Dam, M. and P. Weaver 110
van de Weijer, J. 173
van der Hulst, H. 233n
van Marle, J. 44, 45, 47
van Oostendorp, M. 197n, 303
Vaux, B. 117
Page 40 of 43
Index
vectors see representations, time

velar nasal 240
phonology of 243
Vennemann, T. 5, 212
verbal paradigms 186
inflections, in Hungarian 268–170
verbs 3, 13
see also under Hebrew; Hungarian; inflectional prefixes; sound;
Spanish
base 75–6, 104
causative stems of 134
clusters of 207
morphology in 13
optimal paradigms (OT) in 170, 179, 193, 195, 200, 202, 203,
204, 204n, 207, 228, 243, 247
paradigm uniformity (PU) of 95, 96, 97, 97n, 99–100, 151, 153,
154, 162, 164, stems 180, 188, 192, 203
and vowels 187; templates 184
stress of 95
tenses 3
future 13; imperative 12–13
unstressed syllables in 2
Vietnamese 236
vocabulary 102–3
(p.349) voiced ambisyllabic consonants 247; see also consonants
voiced fricatives 247; see also fricatives
voiced obstruents see obstruents, voiced
voiceless:
ambisyllabic consonants 247
fricatives, in American English 117
stops 25, 109, 112, 121; see also English, voiceless
stops
aspiration in 110–12; and word-final dactyls 110–11,
121;
and non-final 110–11
voicing 230
agreements in 26
in Latin 30
measures 78
obstruents in 7; see also obstruents
vowel-initial:
morphemes 261, 262
suffixes 233
in German 247; nominal 239
words 222
vowel/zero alternations 186, 189, 190, 191n, 196, 199, 200,
204, 208
in verbs 200
Page 41 of 43
Index
vowels 4, 77–8, 89–90, 150, 150–1, 154–7, 159, 161, 166, 176, 267,
271
see also under English; epenthesis; Hungarian; lax; Salish
languages
allomorphy in 98–9
allophony in 254
Bulgarian reduction in 152–3
epenthesis in 185
German 6, 247
lengthening of 7
Hebrew 62
Italian 88
lengthening of 181, 185n, 216, 245, 249n, 280n long 186, 198–
9
optimal paradigms (OP) 179, 180n, 183, 185, 191, 201, 202,
207, 208n, 209
paradigm uniformity (PU) 220, 235, 245, 248, 253, 255, 256,
257
shortening of 184
stems 187
stressed 221, 246
voiceless 7
Walker, R. 178n
Webster’s ninth new collegiate dictionary
(Webster) 227, 228, 231n
Welden, A. 166
Wells, J. C. 212, 214n, 226, 226n, 227, 228, 231n, 240, 242n
Wetzels, L. 30n
Whitney, W. D. 2, 2n
Wilbur, R. 206
Wilson, C. 71
Wiltschko, M. 301
Withgott, M. 108, 109, 112, 262n
Word Faithfulness (Struijke) 303
word-final:
consonants 222, 223n, 224
sonorants see sonorants, word-final
word-formation, base for see base, for word-formation
word-sized units 72–3
words 1, 3, 45, 113–14, 146
see also under English; function; head, function; Jita; Latin,
nouns; morphology; mwords; phonology; pwords; Salish
languages
affixes 219n
analogy of 3
and clitics 234–5
embedding 146
in groups 5n
Page 42 of 43
Index
lengths and genders 37

morphology, paradigm leveling in 24, 25, 31, 31, 39
optimal paradigms (OP) 170, 172, 173, 173n, 202
paradigm uniformity (PU) effects on 212, 213, 226n, 227, 229,
230, 231n, 240, 241, 242, 243, 254, 256, 260, 261, 261n, 263
prosodic properties of 31, 39, 113
reduplication of 303
related 13, 197–9
rule induction for 23–7
stems of 242, 265
stress patterns of 219
surface shapes of 263
truncations of 253
with defective paradigms 205n
Wright, W 209, 210
Wunderlich, D. 104
Wurzel, W. U. 6n
Yawelmani 71, 184
Yiddish 37
Yidiji 215, 218, 230
modifiers in 217
Yip, M. 51, 150
Yucatec Maya:
obstruents in 72
Zec, D. 203n
Zhang, J. 107
Zwicky, A. 229
Page 43 of 43

Paradigms in Phonological Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Paradigms in Phonological Theory

Uploaded by

Copyright:

Available Formats

Title Pages

University Press Scholarship Online

Oxford Scholarship Online