Daniel Harbour-Morphosemantic Number From Kiowa Noun Classes To UG Number Features-Springer (2008) PDF

MORPHOSEMANTIC NUMBER:
FROM KIOWA NOUN CLASSES

TO UG NUMBER FEATURES
Studies in Natural Language and Linguistic Theory
VOLUME 69
Managing Editors
Marcel den Dikken, City University of New York
Liliane Haegeman, University of Lille
Joan Maling, Brandeis University
Editorial Board
Guglielmo Cinque, University of Venice
Carol Georgopoulos, University of Utah
Jane Grimshaw, Rutgers University
Michael Kenstowicz, Massachusetts Institute of Technology
Hilda Koopman, University of California, Los Angeles
Howard Lasnik, University of Maryland
Alec Marantz, New York University
John J. McCarthy, University of Massachusetts, Amherst
Ian Roberts, University of Cambridge
The titles published in this series are listed at the end of this volume.
MORPHOSEMANTIC NUMBER:
FROM KIOWA NOUN CLASSES
TO UG NUMBER FEATURES
by
DANIEL HARBOUR
Library of Congress Control Number: 2008923537
ISBN 978-1-4020-5039-8 (PB)

ISBN 978-1-4020-5037-4 (HB)
ISBN 978-1-4020-5038-1 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved

c 2008 Springer Science+Business Media B.V.

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any
means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written
permission from the Publisher, with the exception of any material supplied specifically for the purpose of
being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Table of Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Abbreviations and notational conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Kiowa phonemes and orthography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
1. FRAMEWORK 1
1.1 Aim: A morphosemantic theory of number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Main claim and overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Conceptual clarification: ‘Morphosemantic’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Overview of Kiowa structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5.1 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5.2 The people . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5.3 The language . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5.4 Orthography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2. KIOWA’S NOUN CLASSES 21
2.1 Overview: Meaning and features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 The noun classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 The SDP class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 The SDI class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 The IDP class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.4 The IDS class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.5 The IDI class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.6 The SDS class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.7 The PPP class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.8 The SSS class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.9 The SII class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4 Semantic coherence of the classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4.1 The animate classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.2 The main vegetal classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.4.3 Symmetric non-constant classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.4 Symmetric constant classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4.5 The default class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.5 Against a tenth class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
v
vi TABLE OF CONTENTS
2.6 Phonological incoherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.6.1 Inverse allophones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.2 Thematic nouns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.6.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3. NUMBER FEATURES 61
3.1 Referential cardinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1.1 Natural classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.1 Classification by cardinality features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.2 ClassP and number on D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2.3 Derivations I: Mnemonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.4 Derivations II: More inverse marking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3 Mnemonic naturalness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4 Spurious s/p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.4.1 Types of grouphood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.4.2 Derivations III: Grouphood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.5 Mass nouns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.5.1 Conjunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.7 Appendix: Missing mnemonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.7.1 Beyond Merrifield’s method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.8 Appendix: Formal demonstrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.8.1 Cardinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.8.2 Grouphood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4. AGREEMENT AND SUPPLETION 117
4.1 Suppletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.1.1 Number-sensitive predicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.1.2 Clarification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2 Analysis of the Basic Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.3 Inversive mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.3.1 Minus-valued classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.3.2 Plus-valued class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
TABLE OF CONTENTS vii
4.4 Group-induced mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

4.4.1 Collectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.4.2 Mass nouns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.5 Harder cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.1 Pluralia-tantum-induced mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.5.2 Animate- and reflexive-induced mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.7 Appendix: Adverbs built on suppletive roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5. THE AGREEMENT PREFIX 157
5.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.1.1 Theoretical assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.2 Reduction of explicanda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.2.1 ‘any’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.2.2 More x deletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.2.3 Animate/reflexive deletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
5.2.4 Impoverishment: Person deletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.2.5 Impoverishment: More person deletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.3 Segmentation I: Subregularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.4 Segmentation II: Segments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.4.1 Ditransitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.4.2 Transitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.4.3 Intransitives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.6 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
6. CONCLUSIONS AND CONSEQUENCES 193
6.1 Noun classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
6.1.1 Gender and declension class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
6.1.2 Gender-number systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
6.2 Parting comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
A A Hunting Story 201
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
Abstract
The current work argues for the necessity of a morphosemantic theory of number, that is, a theory
of number serviceable both to semantics and morphology. The basis for this position, and the
empirical core of the book, is the relationship between semantically based noun classification and
agreement in Kiowa, an indigenous, endangered language of Oklahoma.
The central claim is that Universal Grammar provides three number features, concerned
with unithood, existence of homogeneous subsets, and properties of those subsets.
The features are used to analyze a wide variety of data. Semantic topics include the dif-
ference between granular and non-granular mass nouns, collective, non-collective and distributive
plurals, and cardinality. Syntactic topics include the structure of DP, noun marking, agreement and
suppletion. Morphological topics include the inventory of morphological operations, the featural
basis of complex syncretisms, the difference between agreement and suppletion, and the nature of
the Kiowa/Tanoan inverse.
ix
Acknowledgements
Tempting though it is to dwell on the numerous ways I have become indebted to those whose names
follow, to do so would only trivialize with details how much indeed I owe them. This applies to
none more than to Ken Hale, Vincent (Sun Boy) Bointy and Dorothy (White Feature) Kodaseet,
whose absence I note with fondness and regret.
The ideas presented below have reached their current form only through the generous criti-
cism of many teachers and colleagues. Foremost amongst these are Morris Halle, Irene Heim, Alec
Marantz, and Norvin Richards, who supervised the work during its dissertation phase. Thanks are
also due to David Adger, Susana Béjar, Noam Chomsky, Ken Hale, David Pesetsky, Winfried Lech-
ner, Barry Schein, Laurel Watkins, and two anonymous SNLT reviewers for questions, suggestions,
and, in Watkins’ case, data, that have permitted substantial improvements on earlier formulations.
Of course, none of this would have been possible without the willing participation of many
Kiowa speakers—Vincent Bointy, Dorothy Delaune, Georgia Dupoint, Dorothy Gray, Ellafay
Horse, Dorothy Kodaseet, and Florene Taylor—and the kind support of many members of their
community—Dennis Belindo, Grace Bointy, David Geimausaddle, Bobby and Ann Guoladdle,
Carrie Guoladdle, Carl and Vanessa Jennings, Gus Palmer Jr., Glenda Redbird, and George and
Margie Tahbone. Funding for fieldtrips was generously provided by the Ken Hale Field Fund and
MITWPL.
Finally, I wish to thank my family—parents, brothers, grandparents, aunts, uncles, cousins,
and others—for their support across two oceans, their interest in the face of incomprehensibilities,
and their constant love, the details of which, like my debt to my teachers, I cannot expand upon
without diminishing. Ich widme dieses Buch meinem Mann, Dirk Hannemann.
xi
Abbreviations and Notational Conventions
* of agreement prefixes, causing subsequent verb to have low tone

1 first person
2 second person
3 third person
1–11 noun classes (Kiswahili)
1 referential cardinality 1
a animate plural agreement type
adv adverbial
agr agreement (Kiswahili)
bas basic (not inverse marked)
conj conjunction (intersententially: switch reference)
d agreement/suppletion type, typical of 2
detr detransitive
distr distributive
emph emphatic
ex exclusive
freq frequentative
fut future
hab habitual
hsy hearsay
i agreement type, typical of inverse marked nouns
imp imperative
impf imperfective
in inclusive
inv inverse (marking on nouns)
loc locative
masc masculine
name proper name form of kin terms
neg negative
nom nominal
nom nominative (Sanskrit)
p agreement/suppletion type, typical of 3
pf perfective
xiii
xiv ABBREVIATIONS AND NOTATIONAL CONVENTIONS
pres present
priv privative
q question particle
real nominal suffix denoting prototypicality
recip reciprocal
refl reflexive
rel relative
s agreement/suppletion type, typical of 1
stat stative
th theme
top topic marker (Japanese)
voc vocative
Kiowa Phonemes and Orthography
Consonants
Labial Dental Alveolar Palatal Velar Laryngeal

Voiceless stop p t k (P)
Aspirated stop ph th kh
Ejective stop p! t! k!
Voiced stop b d g
Voiceless affricate x [c]
Ejective affricate x!
Voiceless fricative s sy [C] h
Voiced fricative z
Nasals m n
Liquid l
Glide w y
Vowels
Short Long Diphthong

Front Back Front Back Front Back
High i u ii uu ui
Mid e o ei ou oi
Low a [5,a] O [6] aa OO ai [5e] Oi [6e]
xv
xvi KIOWA PHONEMES AND ORTHOGRAPHY
Vowel diacritics
High tone Falling tone Low tone

Short Long Short Long Short Long
Oral V́ V́V́ V̂ V̂ V V VV
Nasal M
V́ M V́M
V́ M
V̂ M VM
V̂ M
V M VM
V
Chapter 1
FRAMEWORK
1.1. Aim: A morphosemantic theory of number
The current investigation aims to lay the conceptual and basic theoretical
foundations of a morphosemantic theory of number. Until now, morphologi-
cal and semantic theories of number have addressed very different problems
and have produced answers of little mutual relevance (see below). So, the
notion that the study of Universal Grammar requires a theory that covers
both the semantics and the morphology of number constitutes, I believe, a
substantial departure from previous work.
To be specific, morphological theories of number have been primarily con-
cerned with inventories of pronouns and agreement of the world’s languages
(see Corbett 2000, Cysouw 2003 for excellent overview and synthesis). For
instance, the morphologist might wonder whether there exist cardinally ex-
act trial, quadral or quintal number, or whether the forms that permit such
readings are really paucals, and hence cardinally inexact. More rarely, mor-
phologists concern themselves with the relationship between members of such
inventories. For instance, in a system with singular, dual and plural, one can
wonder whether each of these is sui generis, or whether, for instance, dual
is a type of ‘expanded’ singular or ‘restricted’ plural. Even more rarely, obs-
ervations that numbers in some languages are not sui generis have led to
attempts to specify the features that underlie them. An important example
of this work is Noyer (1992), which the current investigation follows both
in spirit and in content. So, morphological theories of number examine lan-
guages’ pronoun inventories and agreement categories, aiming to explain why
only certain ones are attested or how members of the inventory are related.
1
2 CHAPTER 1
Such results are of little interest to semanticists, whose theories of num-

ber are concerned with how to represent the meaning of plurality so as to
capture similarities between mass nouns and plural count nouns, or with the
relationship between plurality of nouns and plurality of events, or with col-
lectivity and distributivity of plurals. This lack of interest is reciprocated:
it is notable that Link’s (1983) use of lattice theory to provide an ontologi-
cal basis for mass∼plurality similarities, though seminal in its area, has had
little, if any impact, on morphological theories of number.
To parody the situation, given the sentence ‘We [you and I] carried a
piano upstairs’, the morphologist would worry about how much like ‘they
(two)’ ‘we [you and I]’ is, whereas the semanticist would worry about how to
determine the quantity of pianos transported, and this while both claiming
to work on number. As measures of the morphologists’ and semanticists’
lack of interest in each other’s work, consider Corbett (2000) and Winter
(2001). In a 350-page survey of number and its morphology, Corbett devotes
barely a page to semantic work on the topic; and Winter’s highly articulated
theory of the semantics of number shows no concern with the difference
between agreement with a head and the head itself, a difference that is basic
to morphology (see his comments on Hungarian).1
The untenability of this mutual uninterest is thrown into relief by Kiowa,
the language that forms the empirical core of this study (see Section 1.5 for
general information). Kiowa displays two typologically noteworthy charac-
teristics. On the one hand, it possesses a complicated rich agreement sys-
tem, that is, an object of interest to morphologists. On the other hand, it
possesses a semantically-based noun class system. As some of these noun
classes pick out sets of nouns of traditional interest to semanticists—such as,
granular versus non-granular mass nouns, nouns that form collective versus
non-collective plurals—Kiowa’s noun classes are of interest to semanticists.
Crucially, however, noun class and agreement interact in Kiowa. Indeed, as
we will see in Chapter 2, it is precisely on the basis of agreement that noun
classes are primarily recognizable. This means that a semantic theory of mass
and plurality or of collectivity is answerable to the morphology: whatever
primitives it posits must be usable in an account of Kiowa’s complex agree-
ment. Conversely, it means that a morphological theory of Kiowa agreement
1
These observations are not intended as real criticisms of Corbett or Winter. Indeed,
I choose their work as examples because, being so good in their respective domains, they
can easily take some mild ribbing. And, of course, we should also note that the division
between morphological and semantic work is not absolute. See, for instance, Ojeda (1998).
FRAMEWORK 3
and syncretisms must be answerable to the semantics: whatever primitives

it posits must be capable of rigorous definition and implementation in a se-
mantic theory of the classification criteria of Kiowa nouns. Consequently,
Kiowa necessitates a unified morphosemantic theory of number capable of
addressing both core morphological and core semantic issues.
Let me spell this argument out a little less abstractly. To do this, I will
show how the classic morphological problem that Kiowa agreement poses
quickly turns into a semantic issue, thereby illustrating the interrelatedness
in Kiowa of the semantic and morphological aspects of number.
One of Kiowa’s most striking characteristics is its rich agreement system.
Unlikely as it may seem, three distinct arguments in (1)—the third person
plural giver, the third person plural recipient, and the third person singular
gift—are in some way encoded by the agreement prefix á∗ (‘∗’ marks that
the prefix lowers the tone of the following verb). This encoding can be
appreciated by changing each of the arguments in turn, to third person dual,
say: each change in argument results in a different agreement prefix (2)−(4).
(1) Á∗- OM OM
3a:3a:3s-give.pf
‘They gave it to them’
(2) M ∗-
É OM OM
3D:3a:3s-give.pf
‘They two gave him to them’
(3) Mé- ÓMÓM
3a:3D:3s-give.pf
‘They gave him to them two’
(4) Et- ÓMÓM
3a:3a:3D-give.pf
‘They gave them two to them’
This system is immediately fascinating. At the most general level, there

is the issue of how so much meaning can get into so little sound. One can
show that prefixes are not unanalyzable wholes; rather each of the arguments
makes a distinct and regular phonological contribution to the prefix (Chapter
5). So, one can ask, more specifically, why each change to an argument has
the precise phonetic effect that it has: why et in (4) and mé in (3), and not
vice versa? Furthermore, one can wonder what (3) and (4) share in virtue
4 CHAPTER 1
of which the verb has high tone, in contrast to the low tone of the earlier
examples. This is, therefore, practically the paradigm case of a morphological
problem: which phonetic features of the prefix realize which syntactic features
of which argument?
Now, the syntactic features that prefixes realize do not start their syn-
tactic existence huddled together beside the verb. Rather, taking a standard
Minimalist view, the features that comprise the agreement cluster are copies
of the feature content of the D0 of each argument DP. In particular, the
number features of each argument will be copied from D0 . Number features
do not begin their syntactic existence there either, though. We will see,
in Chapter 3, that the number content of D0 is determined by two lower,
DP-internal heads. We can sketch the flow of number features through the
syntax, as follows (assuming IP to be the locus of the fully inflected, agreeing
verb).
(5) Syntactic transmission of number information
IP
..
. Inflected Agreeing Verb
DP morphologically significant
number information
D D
6 6
semantically significant
number information
(The account of inverse marking in Chapter 3 shows that the information

does indeed flow through D.)
The content of the DP-internal heads is determined—and this is the cru-
cial point, where things turn semantic—by the noun itself. Kiowa nouns
belong to one of nine classes. I will argue that the agreement behavior of
these classes is readily explicable if we suppose that Kiowa nouns bring with
them into the syntax one or more number features (rather like gender features
of many Indo-European nouns). The semantic connection is this: Kiowa’s
nouns classes are strongly semantically based, in two distinct senses.
FRAMEWORK 5
(6) a. Each class is internally coherent, that is, all classmate nouns share
particular semantic characteristics.
b. The number feature(s) that nouns of a given class bring with
themselves into the syntax is, or are, appropriate to the semantic
property of that class.
So, for instance, if the defining property of a class is the type of plurality its
nouns form, collective versus non-collective, then nouns of that class bring
into the syntax a feature that distinguishes (non-)collectivity of plurals. (The
collective versus non-collective distinction is at stake in class membership for
much vegetation in Kiowa.) Or, if the defining property of a class is that its
nouns generally do not occur in pluralities, then such nouns bring non-plural
number features into the syntax. (This is the case for self-propelling entities,
including animates, in Kiowa.) Given the semantic properties that several
of the classes pick out, they are of immediate and obvious relevance to core
research into the semantics of number: those consisting non-granular mass
nouns, or of granular mass nouns, or of collective plurals, or of nouns that
permit a ‘collection of collections’ reading.
So, we have traced a path from a classical morphological problem, the
constituents of the Kiowa agreement prefix, via agreement relations with D0 ,
through the DP-internal heads that determine the content of D0 , to the rela-
tion between number features and such properties as granularity, masshood,
collectivity, and so on, a classical semantic problem. Consequently, it will not
do to offer a morphological analysis of the prefix that posits features which
cannot serve as the foundation of an analysis of the semantic properties just
listed. Nor can one countenance a semantic analysis with primitives that can
in no way be correlated with the phonetic units that comprise the agreement
prefix. Rather, a unified morphosemantic theory of number must be offered
that does justice to the concerns of morphologists and semanticists alike.
The current investigation aims to present core elements of such a theory.
1.2. Main claim and overview
The core number-theoretic claim of this investigation is that Universal Gram-

mar provides three binary number features: [±singular] concerns unithood;
[±augmented], the existence of (homogeneous) subsets; [±group], properties
of those subsets, should they exist. The investigation concentrates on mo-
tivating these features, on rigorously defining them, and on showing how
6 CHAPTER 1
their manipulation, syntactically and postsyntactically, leads to a tightly

constrained theory with wide and diverse empirical coverage. The topics and
phenomena that motivate and deploy these features are as follows.
Chapter 2 presents the empirical core of the investigation, Kiowa’s noun
class system. First, it shows how Kiowa’s four agreement types can be used
to distinguish nine noun classes. On the basis of this agreement behavior, a
new system of noun class mnemonics is proposed, to facilitate the discussion.
The nine classes are shown to be internally semantically coherent, that is,
that the nouns in each class share semantic characteristics:
(7) Characteristics
First person
Animates and animate-like entities (physically similar to ani-
mates or capable of self-propulsion or determining direction of
motion)
Naturally regarded both as individual and as occurring in col-
lections; permits ‘different types of’ reading with s-agreement
Individuable, non-shape-inductive
Non-granular mass nouns
Pluralia tantum, composite nouns (and granular mass nouns for
some speakers); abstract nouns
Default for vegetation and implements; granular mass nouns for
some speakers
Vegetation forming natural collections in which members are not
readily individuable; implements that act as a group to produce
a single effect
Default; no unifying properties
The chapter also introduces one of Kiowa’s most fascinating phenomena,

inverse marking. Kiowa nouns, like English nouns, if drawn straight from
the lexicon and uttered unaltered can be used to refer only to a restricted
number of tokens. For instance, English table refers to a single table, chair
to a single chair, and so on. To talk of not-one-table or not-one-chair, one
adds something to the noun, the plural -s. In Kiowa, the range of meanings of
nouns drawn straight from the lexicon is wider: some nouns refer to exactly
two (k!Ôn ‘two tomatoes’), others to two or more (áá ‘two or more sticks’),
FRAMEWORK 7
others to one or two (tógúl ‘one young man or two’), and so on. Interestingly,
Kiowa has a morpheme, just as English has the plural, that permits one to
talk of a quantity of nouns other than what noun by itself means. This
is traditionally called the inverse. So, k!Ôˇ ǑdO, ‘tomato’+inv, means ‘one
tomato’ or ‘three or more tomatoes’; áádO, ‘stick’+inv, means ‘one stick’;
tógúúdÓ, ‘young man’+inv, means ‘three or more young men’. Note that
the morpheme is the same dO in all cases.2
Chapter 2 shows that inverse marking is a major indicator of class mem-
bership. For the most part, the chapter is expository and empirical. Some
readers may prefer to skim through it just to gain a feel for the facts and
system as a whole, rereading it more thoroughly at a later point.
Chapter 3 presents the theoretical core of the investigation. It mo-
tivates and defines the unithood feature [±singular], the subset-existence
feature [±augmented], and the subset-property feature [±group]. Positing a
non-novel DP-structure, it claims that these number features occur in two
DP-internal positions both of which are semantically contentful: Number is
determined by cardinality (singular, dual, plural), and Class, by semantic
properties of the noun. A computation over Number and Class determines
the number features on D.
(8) DP
NumberP D
depends on content
of Class and Number
⎡Class: Noun ⎤ Number
(±singular) ±singular
⎣(±augmented)⎦ ±augmented
(±group)
The number-feature content of D is not itself semantically interpreted. How-

ever, it is syntactically active and, sometimes, phonetically realized. Syn-
tactically, it is the head that triggers agreement and so contributes features
to the agreement prefix. Phonetically, it is realized as inverse marking, but
is otherwise generally null. Several more complex DPs, such as adjectivally
modified ones, are examined, to show that inverse marking is located at D.
2
The variation in tone is purely phonological. See Watkins (1984) or Harbour (2002).
The morpheme itself is subject to phonological conditioning (Section 2.6.1).
8 CHAPTER 1
The chapter shows further that the relationship between a class’ mnemo-
nic and the semantic characteristic of the nouns it subsumes is principled, in
contrast to the semi-arbitrary gender classification of Indo-Europe.
The chapter concludes with discussion of what constitutes a possible noun
class of a Kiowa-like language. The problem addressed is one of generative
capacity. Given the description of Chapter 2, one would think that 64 differ-
ent noun classes are possible. This prompts one to wonder why Kiowa should
instantiate just nine. It is shown that the theory just sketched is highly res-
trictive and that Kiowa all but optimally exploits the space of possibilities.
Chapter 4 addresses mismatches between agreement and suppletion,
which readers, depending on their inclinations, are likely to regard either as
delightful or as horrifying. Kiowa has two sets of number-sensitive suppletive
predicates: those distinguishing singular from dual/plural, and those distin-
guishing plural from singular/dual. In the vast majority of cases, agreement
and suppletion match. That is, if one is talking of a single young man, agree-
ment will reflect that singularity and any suppletive predicates will be in
their singular or singular/dual form:
(9) Tógúl ∅- ét
young man 3S-big.S
‘The young man is big’
Similarly, talking of two tomatoes, agreement will reflect that duality, and
suppletive predicates will be in their singular/dual or dual/plural form:
(10) K!Ôn nen- ót
tomato 1s:3D-drop.s/D
‘I dropped two tomatoes’
Likewise, mutatis mutandis, for talk of several sticks.
However, in a variety of cases, agreement and suppletion seem to indi-
cate different numbers. Two examples are (11), where agreement indicates
plurality and suppletion, singularity, and (12), where the reverse holds:
(11) MŹMde t!óMúM gya yáM -
É dôi- et
this shirt :1s:3P-too-big.S
‘This shirt is too big for me’
(12) Tóú ∅- sÓl
house 3S-be set.P
‘There are houses standing’
FRAMEWORK 9
Theoretical arguments are presented that the features that condition sup-
pletion are located on a different head from the features that trigger agree-
ment. D triggers agreement whereas the interpretable features directly under
D, generally Number, condition suppletion. The theory of Chapter 3 claims,
independently of suppletive facts, that divergences can arise between the
content of Class, Number and D. Given such divergences, we predict not
only when agreement∼suppletion mismatches will occur, but precisely which
otherwise unexpected combination of agreement and suppletion will result.
Thus, agreement∼suppletion mismatches support the theory of Chapter 3.
Chapter 5 sketches in detail an analysis of the agreement prefix. This
is crucial to the project of morphosemantic research on number, given the
argument that Kiowa not only presents classical morphological and semantic
number problems, but demands that solutions to one be exportable to treat-
ments of the other. Having motivated three number features primarily on
semantic grounds (though with reference to DP-level morphology and, more
cursorily, to the agreement prefix), it becomes crucial to show that phono-
logical subparts of the prefix are precisely realizations of those features. It is
shown, moreover, that rather complex allomorphic relations can be naturally
stated in terms of these features and the natural classes they define.
Chapter 6 concludes by placing Kiowa’s noun class system in crosslin-
guistic context, arguing that it is different from Indo-European gender but
strongly similar to Bantu gender-number systems.
As the solutions proposed for the morphological and semantic number
problems that Kiowa poses employ the same elements, we are on the way to
a unified morphosemantic theory of number. In presenting a unified analysis
of diverse grammatical phenomena in Kiowa, the investigation attempts to
provide some definite answers. In its theoretical aims, however, it is merely
programmatic, suggesting a line of inquiry that strikes me as important and
interesting and offering some elements of what a fuller morphosemantic the-
ory of number might comprise.
1.3. Conceptual clarification: ‘Morphosemantic’
Now, in light of the foregoing exposition, the term ‘morphosemantic’ itself re-
quires clarification, particularly in comparison to terms such as ‘morphosyn-
tactic’ or ‘syntacticosemantic’. ‘Morphosyntactic’ refers to the interaction
10 CHAPTER 1
between morphology and syntax, ‘syntacticosemantic’ to that between syn-

tax and semantics, and the terms can be used to name interfaces between the
relevant modules of the grammar. In arguing for a morphosemantic theory of
number, I am not suggesting that there is a morphology-semantics interface
involving direct communication between the conceptual and pronunciational
systems. Everything I argue for is compatible with the familiar Y-model.
(13)
Syntax
Morphology Semantics
Phonology Conceptual
Interface
Articulatory
Interface
Syntax builds hierarchical objects and transmits information (via agreement

and movement operations) from one part of the structure to another, as de-
picted in (5). Some parts of the resulting structures and chains are relevant
to the semantics (i.e., are interpreted), others are relevant to the morphology
(i.e., are pronounced). It is syntax that plays the role of an interface between
morphology and semantics, to the extent that there is one. And to the extent
that the phenomena addressed by semantic number theories and morpholog-
ical number theories are fed by the same syntactic features, a single theory
of both sets of phenomena is required. What is interesting about Kiowa is
that it provides a relatively straightforward argument that the overlap, and
hence the need for a single ‘morphosemantic’ theory of number, is great.
1.4. Framework
The syntactic and morphological frameworks I adopt are Minimalism and

Distributed Morphology, respectively. My reason for adopting these theo-
ries, rather than any others, is that they offer a constellation of assumptions
that, in my opinion, make for a particularly interesting research program.
Specifically, Minimalism places heavy emphasis on the importance of inter-
faces and it constrains the variety of features one can posit: every feature
FRAMEWORK 11
must be interpretable, that is, for every feature, [F], there must be a head, X0 ,
such that, when [F] is located on X0 , [F] contributes to the interpretation of
the sentence (what Pesetsky and Torrego 2001 term ‘weak functionalism’).3
And Distributed Morphology claims that ‘words’ have an internal structure
and that the structure is composed, primarily, by the syntax. It claims fur-
thermore that the phonological structures associated with syntactic feature
bundles are inserted postsyntactically, at the terminal nodes. Let me illus-
trate how each of these assumptions is important to the investigation.
My aim is, as stated, to present elements of a unified theory of number
serviceable to morphology and semantics alike. The initial motivation for
this theory—a path from a core morphological problem, via syntax, to a core
semantic one—itself relies on several tacit assumptions about the relationship
between syntax, semantics, and morphology, especially those just outlined.
It is immediately obvious that this an interface project. It is concerned
with what nouns bring with themselves into the syntax in virtue of their
meaning—this is the lexicon-syntax interface. It is concerned with the ex-
act relationship between phonological pieces of agreement prefixes and the
feature content of syntactic heads—this is the morphology-syntax interface.
And most obviously, it is concerned simultaneously with two different mod-
ules of the grammar, morphology and semantics. Consequently, both the
methodology and the aims of the project itself only make sense in a world
where interfaces are central to research. Minimalism offers such an approach.
Second, the motivation of the project began with morphological pieces
and ended up at semantic ones. Conversely, the more detailed chapter sum-
mary began with semantic features in Chapter 3 and ended with morpho-
logical ones in Chapter 5. Now, everyone expects that we will be able to
trace paths from one module of the grammar to another: no module is an
island. However, there is no reason to expect a priori that semantic and
morphological atoms will be one and the same. Sameness of atoms is ex-
pected, indeed forced, if we assume, with Distributed Morphology, that the
pieces of inflection are the phonetic realization of the pieces of syntax, and
if we assume, with Minimalism, that all features have an interpretation.
Third, I argue below, and sketched above, that the content of D is de-
termined by the content of two lower heads, Number and Class, evidence for
the content of which is semantic. The content of D determines agreement
and D’s phonetic realization (as inverse marking). Thus, both agreement and
3
This is not the same as assuming that all instances of all features are interpreted.
12 CHAPTER 1
inverse marking depend on a syntactic computation. This postsyntactic de-

pendence is completely natural if we assume late insertion, with Distributed
Morphology. However, it is quite unnatural if we adopt a morpheme-based
theory, where the presence of each morpheme is justified by its contribution
the meaning of the word. This is because the semantically interpreted heads
are Class and Number. D, where inverse marking is located, is uninterpreted.
In other words, inverse marking does not have a meaning, but occurs when a
constellation of features is distributed over Number and Class. Consequently,
the inverse is a sound without a meaning; it is not a Saussurean sign.
1.5. Overview of Kiowa structure
In the remainder of this chapter, I present the basics of Kiowa grammar and
apply some of the framework assumptions above.
1.5.1. Sources
All Kiowa sentences and word lists cited in the current work are from four
sources: Harrington (1928), Watkins (1984), Watkins (p.c.), and my own
fieldwork (August 2001, July 2002, January and December 2003, November
2004, August and December 2005). Sentences from the first three sources are
always cited as such; sentences without cited sources are from my fieldnotes.
The fieldwork was conducted with speakers, aged 70−85, from Anadarko,
Carnegie, and Mount Scott: Vincent Bointy, Georgia Dupoint, Dorothy Ko-
daseet (the primary consultants for this study); Dorothy Delaune, Dorothy
Gray, Ellafay Horse, and Florene Taylor (who provided supplementary data—
our collaboration has chiefly focused on other linguistic domains). All were
monolingual in Kiowa until commencing school, aged 7−9, and retained flu-
ency into adulthood. Though English is their dominant idiom, all revert to
Kiowa with ease, save for occasional lexical blockage (I have not been able
to record the word for ‘tickle’). Dialectal variation between these speakers,
and those with whom Harrington and Watkins worked, though existent, is
confined, I believe, to minor phonological, morphological, and lexical details,
noted, when relevant, below.
This work is not an ethnographic or historical study of the Kiowas. How-
ever, a number of such sources have, of necessity, been consulted and their
contents, referred to, in passing, below. These are Mooney (1898/1979),
Boyd (1983), Ellis (1996), and Merrill, Hanson, Green, and Reuss (1997).
FRAMEWORK 13
1.5.2. The people
Kiowa is a Tanoan-related language of Oklahoma.4 It is spoken fluently now

only by some 40 people, aged 70−95. Members of the next oldest generation
often have good comprehension of the language, but they rarely have so inti-
mate a grasp of its grammatical, lexical and stylistic subtleties. Knowledge of
the language declines sharply as one moves into the younger generations. The
prognosis for the language is therefore not good. However, comprehensive
documentation is underway and it is likely that future generations of Kiowas
will inherit a substantial record of their language in its cultural context, even
if direct inheritance of the language becomes impossible.
The Kiowa Tribal Complex is located in Carnegie, Oklahoma, and mem-
bers of the tribe live mostly in Caddo, Kiowa and Comanche counties. The
community’s distribution over three counties is the result of deliberate US
Government policy. By 1876, White exploitation of resources had precipi-
tated the collapse of the buffalo population, on which the Kiowas depended
not only for food, but also for clothing, implements, and shelter (in the shape
of hides for teepees). This made the continuation of their traditional lifestyle
impossible. Instead, in return for government rations, they (finally) agreed
to be settled on a joint Kiowa-Comanche-Apache reservation.
The US government nonetheless quickly realized that the reservation pro-
vided the tribes with autonomy and coherence enough for the maintenance
of their identities, cultures and languages. To force assimilation (and to gain
access to the valuable grasslands that comprised the reservation), Congress
passed the General Allotment Act of 1887 and, more controversially, the
Jerome Act of 1901, which, by 1907, had had the effect of transferring own-
ership of 80% of the former reservation into White hands and interspersing
the allotments assigned to Indians, not only amongst each other’s tribes, but
also amongst the Whites. The consequent lack of a geographically coherent
4
The family is called Kiowa-Tanoan. The double barrelled appellation primarily reflects
geographic and cultural differences, rather than linguistic ones: the Kiowa are, geograph-
ically and culturally, of the Northern Plains, in contrast to their linguistic relatives, all
of whom are Pueblo peoples, resident in the South West. Given these differences, the
relationship of Kiowa to Tanoan could not be assumed, but required proof. So it was
that the Kiowa-Tanoan family was posited (e.g., Harrington 1910). Since Hale’s definitive
demonstration of the relationship (Hale 1962) and his subsequent partial reconstruction
of the proto-phonology (Hale 1967), the double barrelled name has become otiose and is
linguistically misleading. However, the nomenclature is entrenched and the best we can
do now is, as above, to avoid the cacophonic phrase ‘Kiowa is a Kiowa-Tanoan language’.
14 CHAPTER 1
Kiowa community appears to have contributed to the decline in the tribe’s

language and certainly militates against efforts to revitalize it.
Kiowas enter Whites’ historical record several centuries ago, in Montana.
Their migration southwards occurred after introduction of the horse, which
became central to their culture; it took over from the dog the designation
xêˇǐ, with dogs becoming known as xêˇǐhǐǐ ‘dog’+real and later as xégun.5
Kiowas’ earlier history is unclear, and, indeed, quite intriguing, given the
geographic and cultural distance between them and their linguistic relatives
in the South West. Kiowa folk memory speaks of a white bear and suggests
in other ways an existence even further north. And earlier still, Kiowas had
lived underground, escaping only with the assistance of Séndé, the creator-
prankster, who pulled them out through an owl hole in a cottonwood tree,
having heard their knocking when he chanced to rest against it.
1.5.3. The language
Basic word order
Kiowa is a rich agreement language with relatively free word order. A basic
(informationally unmarked) order is nonetheless discernible:
(14) Particles Subject Indirect–Object Direct–Object Verb

(15) HÓn PaithalŹŹ P!ÓÓthÓpdek!ii áádO Ó- thêm- OM OM mOO
neg Vincent Daniel stick.inv 3s:3s:3i-break-make.neg
‘Vincent didn’t make Daniel break the stick’
Sentences like (15) are rare for two reasons. First, Kiowa permits pro-drop
of any argument DP, as in (1)−(4) (Watkins 1990), and sentences with three
overt arguments are infrequent. Second, DPs, as well as other constituents,
are frequently dislocated to the left or right edge of the sentence.
(16) HÓn máthOn ∅- xáM áM nÔO

neg girl 3s-arrive.neg
‘The girl didn’t arrive’
(17) MáthOn hÓn ∅- xáM áM nÔO
girl neg 3s-arrive.neg
approximately ‘The girl, she didn’t arrive’
5
Watkins (1984) gives the origin of gun as associated with poles or travois.
FRAMEWORK 15
(18) HÓn ∅- xáM áM nÔO máthOn

neg 3s-arrive.neg girl
approximately ‘She didn’t arrive, the girl.’
These dislocations correlate with information structure, leftward dislocation

for topic and focus, rightward for old information, as the approximate trans-
lations attempt to capture. The rightmost of the particles in (14) can be
regarded as marking the right edge of part of the clause that hosts leftwards
displaced constituents. These dislocations, and their nature, effect and lim-
its, are the topic of ongoing joint research.
Like DPs, the particles in (14) occur in a relatively fixed order. Seman-
tically, they express a variety of aspectual, modal and evidential meanings,
as well as negation. Many obligatorily cooccur with inflection suffixes on the
verb. (See Watkins 1984 for a thorough overview and Adger and Harbour
2007 for discussion of phrase structure implications.)
(19) BéthOO hÓn ám em-dÓMÓM-mÔO- hel

unbeknownst neg 2 2s- be- neg-hsy
‘I didn’t realize it wasn’t you’
(20) Háyátto hÓn ∅- deM Mi- héMŹM-mÔO- t!OO
maybe neg 3s-sleep-die- neg-fut
‘Maybe he won’t fall asleep’
(21) BéthOO an Ó- bôu- hOnx!ou- yii- t!OO- dei
unbeknownst hab :3s:3i-always-come late-impf-fut-hsy
‘I didn’t realize he was going to keep on coming late’
As illustrated by the underlining, particles in selectional relations with in-

flectional suffixes of the verb occur in the opposite order from those suffixes.
A variety of word orders and particles occur in the example sentences
of the chapters that follow. However, word order is never a factor in the
grammatical properties relevant below.
Nouns
Nominal morphology is sparse in Kiowa. There is no case marking either

for DPs or pronouns, and the only marking for number is inverse marking:
Taken straight from the lexicon and uttered unaltered, nouns are limited in
the number of tokens they can refer to. For instance, tógúl means ‘one or two
16 CHAPTER 1
young men’, áá means ‘two or more trees’ and k!Ôn means ‘two tomatoes’.
When this inherent number is not the same as the number of tokens talked
of, the noun is inverse marked. This results in the curious situation that
one and the same suffix attaches to the nouns just given for the plural, the
singular and the non-dual: tógúúdÓ ‘young men’, áádO ‘a tree’ and k!ÔˇǑdO
‘a tomato’ or ‘more than two tomatoes’. The form of inverse marking and
its effect on agreement is discussed at length in Chapter 2.
Given Kiowa’s freedom of word order, it is, perhaps, not overly surprising
to find split constituents. Compare (22), which was spontaneously uttered,
with its split-free paraphrase (23).
(22) Dôiette an pénhaa gya- ôugu xóŹkya

too much hab sugar 1s:3s-pour.impf coffee.loc
‘I’m always putting too much sugar in my coffee’
(23) An xóŹkyá dôiette pénhaa gya-ôugu
hab coffee.loc too much sugar 1s:3s- pour.impf
‘I’m always putting too much sugar in my coffee’
Three more examples are:
(24) PáágO gya- bóMúM ˇ á

k!yá ˇ hẐˇǐ
one 1s:3s-see.pf man
‘I saw the one man’ (Harrington 1928, p. 45)
(25) ˇŹˇde hâtêl k!Ódál á-
É dÓÓ?
that who.q car :3s:3s-be
‘Whose car is that?’
(26) ÉŹgútk!o hâtêl yŹŹ eM - hân?
biscuit who.q two 3s:3d-devour.pf
‘Who ate two biscuits?’
Quantifiers may be pre- or postnominal: éˇŹˇde tóú ∼ tóú éˇŹˇde ‘this house’,
yŹŹ tóú ∼ tóú yŹŹ ‘two houses’, étté tóú ∼ tóú étté ‘many houses’.
Attributive adjectival modification is not an overly common strategy, as
the language generally prefers verbs, particular in headless relatives, to play
this role. However, when adjectives do occur, they are postnominal:
(27) x!óú-êl xégun-syan tou- gúl k!ŹŹ- t!áMŹM- de

rock-big.s dog- small house-red male-white-bas
‘big rock’ ‘little dog’ ‘red house’ ‘White man’
FRAMEWORK 17
DP modifiers
Consistent with its head finality (default verb finality, postnominal modifi-
cation), Kiowa has postpositions, rather than prepositions:
(28) k!ÓÓ- pa Indian Fair-kya álOO- aadO- yOO

bank-against Indian Fair-at apple-tree.inv-from
‘against the bank’ ‘at the Indian Fair’ ‘from the apple tree’
Similarly, focus modifiers are suffixal. (They differ from quantifiers in being
inseparable from the noun.)
(29) nÓÓ-(x!)al Indian Fair-deki háote- kÓÓ

1- too Indian Fair-only several-just
‘me or us too’ ‘only the Indian Fair’ ‘just a few’
Pronouns
Kiowa has only two pronouns: nÓÓ for all first persons, whether singular, dual
or plural, inclusive or exclusive; and ám for second person, whether singular,
dual or plural. There are no third person pronouns, but deictics can be used
instead, éˇŹˇde∼éˇŹˇgO ‘this’∼‘this.inv’ and óŹde∼óŹgO ‘that’∼‘that.inv’.
The complex verb
Verbs consist of two parts, an agreement prefix (next subsection) and a com-
plex verb. The latter, in which only the root is obligatory, consists of:
(30) Incorporates Root Distributive Aspect/Negation Future Evidential
Incorporates can be adverbs:
(31) Gyat- hón-pOl- /t!Om-pÓl- tOO

1s:3p-last- eat-/first- eat-fut
‘I’ll eat last/first’
(32) Gyá- hágyá- ton
:2s:3s-already-fat
‘You’ve already got it fat’ (Watkins 1984, p. 241)
—or verbs, with or without complement nouns:

18 CHAPTER 1
(33) TÓÓdé de- t!ÓÓthâl-dou

long time 1s:3a-listen- hold
‘I listened for a long time’
(34) A- hot- kôm- hap- do yáM - ol- sÓ- mOOgO
1s-travel-distr-freq-because :1s:3p-belongings-lay.p-expert
‘Because I travel so much, I’m expert at packing’
—or nouns alone:
(35) A- pénhaa-ou
2s:3s-sugar- pour.imp
‘Pour the sugar’
(36) P!él- heM Mi hétÓ gya-ol- sÓl
drop.p-priv still 3p- belongings-lie.p
‘The groceries are still lying [in the car] unloaded’
(37) Bé- k!ÔO- saa
2s:3i-knife-cut.imp
‘Cut it with a knife’
Object incorporation is not valence reducing: (35) is overtly transitive.

With regard to the inflectional particles, the main points were mentioned
above in tandem with sentential particles. Note, though, that aspect and
negation are in complementary distribution (though particles that govern
them are not; in fact, hÓn neg, which cooccurs verbal -neg, and án hab,
which cooccurs verbal -impf, frequently cooccur in that order with the mean-
ing ‘never’; the verb is inflected for negation in these cases, not aspect). The
suffixes show allomorphy for a variety of different properties, including agen-
tivity, stativity and transitivity. For instance, the future suffix, fut, has
an agentive form tOO (31) and a non-agentive form t!OO (20). The hearsay
suffix, hsy, too shows variation, appearing as dei after impf-fut in (21), as
êi fused with impf as in tóˇ ú
ˇ nêi ‘say.impf.hsy’, and elsewhere, as in (19),
as hel. For full exposition, see Watkins 1984, Harbour 2004.
The agreement prefix
Along with the root, the prefix is only other obligatory part of the verbal
complex. It registers agreement for up to three DPs: external argument, indi-
rect object and direct object. Kiowa has some 100-to-160 prefixes depending
FRAMEWORK 19
how one counts certain homophones. Harbour (2003a) shows that this prefix
is phonologically independent from the rest of the verb. The only exception
is that some prefixes, notated with a ‘∗’, lower the tone of the subsequent
verb; compare (1)−(4).
Prefixes, discussed at length in Chapter 5, are glossed as follows. In z-,
z is the subject of an unaccusative predicate, as in (20). In x :z -, x is the
agent of a transitive verb and z, the direct object as in (31). In x :y :z -, x is
the agent of a (di)transitive verb and y, the indirect object, and z, the direct
object, as in (1)−(4). Finally, in :y :z -, z is the subject of the unaccusative
(it triggers however agreement identical to that triggered by the z direct
object in (36)), and y is the indirect object, such as the possessor of z or a
benefactor of the event, as in (21). Thus, in prefix glosses, something of the
form :n: is an indirect object; something of the form :n is a direct object;
and something of the form n is a subject/agent.
Phonology
Kiowa phonology is not overly complicated, though it does throw up some

interesting surprises, such as Dental-velar switching, summarized, together
with a variety of alternations in Section 5.1.1. Appendix A contains brief
comments on tonal interactions. Note that glosses always show surface tones,
not underlying tones. (This accounts, amongst other things, for neg’s having
falling tone in (15), mÔO, but low tone in (19), mOO.) The language’s phoneme
inventory is described immediately below together with the orthography.
1.5.4. Orthography
The orthography used here is that of Harbour and Guoladdle (in prep.).
Brief descriptions of the sounds they represent follow. See Watkins (1984)
for greater detail. Phoneme charts are provided on pp. xv–xvi.
The consonants b d g h m n s w y have their IPA (International Pho-
netic Alphabet) values (Lagefoged 1993). The palatal affricate, IPA [ts], is
written x, and the alveolo-palatal fricative, IPA [C], sy. The latter represen-
tation acknowledges that sy is very occasionally realized as s+y, that is, as
IPA [sj]. Kiowa l is generally preceded by a laterally released [d], sometimes
somewhat devoiced; hence it varies between IPA [dl l], [d◦ l l] and [d◦ l ◦l ] . The
degree of devoicing of [d] varies to the extent that it sometimes approaches
20 CHAPTER 1
IPA [l] (particularly when intervocalic and morpheme internal, as in álOO

‘apple’ and bÓlO ‘butter’); the [l] pronunciation appears to be more pervasive
in Anadarko than in Carnegie. Finally, as concerns consonants, stops show a
four-way contrast: voiced, ejective, voiceless unaspirated, voiceless aspirated.
The voiced stops g d b and the voiceless unaspirated k t p have their IPA
values. Aspiration, IPA [h ], is marked by h as in kh th ph. Ejectivity, IPA
[P], is marked by ! as in k! t! p!. X has an ejective counterpart, x!, but no
aspirated one.
Kiowa has six vowels, high, mid and low, front and back: i e a u o O.
These have their IPA values, except for u, IPA [U]; O, IPA [6]; and a, IPA [5],
fronting to [a] after palatalized velars (unless part of the diphthong ai). All
the back vowels and the front low vowel form diphthongs with i; however,
this is pronounced as IPA [e] if the first vowel is low: hence ai [5e] and Oi [6e].
Vowels contrast for length, tone, and nasality (but diphthongs only for tone
and nasality). For high and low vowels, length is indicated by repetition: ii
aa uu OO. For mid vowels, which diphthongize when long, it is indicated by
the appropriate high vowel: ei ou. Nasality is indicated by a Polish hook,
e.g.: ǎ ǎǎ. However, nasality is predictable, and so unmarked, on vowels
cosyllabic with a nasal stop (e.g.: má not má ˇ ). Tonally, vowels are either
high (acute), falling (circumflex), or low (unmarked). Illustrating with the
low front vowel, these are: á â a for short vowels, áá âa aa for long. Observe
that, in long falling vowels, only the first of the pair bears a circumflex (as
the second segment is primarily low). Tonology is discussed at length by
Watkins (1984) and Harbour (2002). Unless pertinent to the discussion, I
abstract away from the causes and effects of tone interactions.
Chapter 2
KIOWA’S NOUN CLASSES
2.1. Overview: Meaning and features
This chapter presents Kiowa’s noun class system, the empirical core of the
current study.
The key insight into noun classification in Kiowa was provided by Won-
derly, Gibson, and Kirk (1954):
(1) Kiowa nouns are classified according to number properties
That is, number plays in Kiowa the role that gender plays in Indo-European
languages; it is the primary classifier of nouns. The substantive difference
between the Kiowa and Indo-European systems lies in their degree of ar-
bitrariness. Gender-based classification is necessarily largely arbitrary, for
most of a language’s nouns, such as implements, most vegetation and places,
lack gender. We will see below and in the next chapter that, in Kiowa, mem-
bers of each noun class share semantic characteristics with a natural nexus
to number concepts.
A second insight was provided by Merrifield (1959a), who refined Won-
derly, Gibson and Kirk’s four-class system. Whereas the latter focused on
number marking on the noun, Merrifield focused on number agreement in
the verb prefix (cf., Harrington 1928).
(2) Kiowa noun classes are revealed through their effect on the verb prefix
In this way, he was able to argue for a seven-class system. Watkins (1984),
the most thorough description of Kiowa grammar to date, upheld Merrifield’s
21
22 CHAPTER 2
classification, and Takahashi (1984) provided insight in the semantic coher-

ence of the seven classes.
The aims of this chapter are:
(3) a. To explain how Kiowa noun classes are distinguished on the basis
of number agreement in verbal prefix.
b. To argue that this methodology reveals nine classes and to
present a new nomenclature for these.
c. To show that membership of the nine classes is semantically co-
herent.
The later sections of the chapter treat two residual issues: whether the verbal
agreement prefixes support a tenth class, as Harrington (1928) believed, and
whether phonology plays a role in Kiowa’s noun class system. The features
that underlie the system, their semantics and their manipulation in the syn-
tax are explored in subsequent chapters. The important issue of how Kiowa
compares to other languages with rich noun class systems, such as those of
Bantu, is left until Chapter 6.
2.2. Preliminaries
The basis of noun classification in Kiowa, and of the mnemonic nomenclature

adopted below, is the correlation between agreement and singularity / duality
/ plurality. As preliminaries, therefore, the relevant properties of agreement
and number are highlighted.
By number, I intend referential cardinality, the cardinality of the set
of things that a noun is used to refer to on a particular occasion. Formally:
(4) Definition: Referential Cardinality

Let U be an utterance containing a noun N and let S be the subset of
the universe of discourse that N denotes. The referential cardinality
of N is defined as | S |, i.e., the cardinality of the set that N denotes.1
So, in ‘I saw two men’, the referential cardinality of the subject is 1 and the
referential cardinality of the object is 2.
1
This definition glosses over certain technicalities, such as whether proper nouns, like
John, should be semantically represented as singleton sets or as individuals (on this issue,
see the Schwarzchild’s 1996 appendix on Quinean set theory) and how to accommodate
the contention that DPs are referential and NPs predicative (Winter 2001).
KIOWA’S NOUN CLASSES 23
Kiowa distinguishes singular, dual, and plural, that is, the language dis-
tinguishes referential cardinalities 1, 2, and 3 or more. In order to keep
morphological and semantic terminology separate, the following notational
convention is adopted:
(5) Notation: Referential cardinality

1 denotes referential cardinality 1 (singular).
2 denotes referential cardinality 2 (dual).
3 denotes referential cardinality ≥ 3 (plural).
Now consider the morphological form of verbal agreement in Kiowa, ab-

stracting away from person. Agreement on the verb covaries with referential
cardinality as well as with morphological marking on the noun. Four types
of agreement are observed:
(6) Notation: Agreement types

s, d, p, i
As a rule of thumb, s, mnemonic for ‘singular’, occurs when referential car-

dinality is 1; d, mnemonic for ‘dual’, occurs when referential cardinality is
2; and p, mnemonic for ‘plural’, occurs when referential cardinality is 3. i,
mnemonic for ‘inverse’, occurs when the verb agrees with an inverse-marked
noun (see Section 1.5.3 and below); i-agreement can occur when referential
cardinality is 1, 2, or 3.
Kiowa nouns do not all trigger the same agreement type for a given ref-
erential cardinality. For instance, for referential cardinality 1, some nouns
trigger s-agreement, but others, i. A learner of Kiowa must learn the agree-
ment behavior of each noun, just as a learner of French, German or Russian
must learn the gender properties of each noun they encounter.
Information about the correlation between referential cardinality and
agreement type is represented below in the form of mnemonics. In each
triliteral mnemonic, the first letter indicates the type of agreement triggered
when referential cardinality is 1; the second letter, the type of agreement
triggered when referential cardinality is 2; and the third letter, the type of
agreement triggered when referential cardinality is 3. So, a noun, N, with
the fictitious mnemonic dip would trigger d-agreement for referential cardi-
nality 1 (‘Here’s an N ’ would show d-agreement); i-agreement for referential
cardinality 2 (‘Here are two N ’s’ would show i-agreement); and p-agreement
for referential cardinality 3 (‘Here are three N ’s’ would show p-agreement).
24 CHAPTER 2
2.3. The noun classes
Merrifield (1959a) observed that Kiowa’s noun classes are revealed through
covariation between referential cardinality and agreement on the verb. I will
now show that this methodology reveals the existence nine noun classes.
(There are, of course, 43 = 64 mnemonically possible classes. Section 3.7
addresses why 55 are unattested.) The nine classes, in mnemonic form, are:
(7) a. sdp
b. sdi
c. sii
d. idp
e. ids
f. idi
g. sds
h. sss
i. ppp
A noun from, say, the sdp class will be referred to as an sdp noun.
To illustrate the mnemonics, consider an sdp noun. For referential car-
dinality 1, we find s-agreement on the verb; for referential cardinality 2,
d-agreement; and for referential cardinality 3, p-agreement. Consider also
an ids noun. For referential cardinality 1, we find i-agreement; for 2, d-
agreement; and for 3, s-agreement.
To justify the existence of nine classes and the appropriateness of their
mnemonics, it is necessary to show that each mnemonic corresponds to one
or more nouns and that each noun is described by one of the mnemonics in
(7), proceeding through each mnemonic in turn. Immediately below, I will
concentrate only on showing that each mnemonic corresponds to one noun.
Fuller lists of members of each noun class are given during the discussion of
the classes’ semantic coherence (Section 2.4), supporting the claim that all
Kiowa nouns are described by one of the listed mnemonics.
As the following sections are rather data rich, it should be pointed out
that the reader is not required to remember any individual fact or form
given below. Empirical specifics will be repeated in the theoretical chapters
as and when they become relevant. The main purpose of the discussion that
follows is to acclimatize the reader to Kiowa, as it were, familiarizing them
with morphological and semantic characteristics of the noun class system
that will be treated in subsequent chapters. A summary of classes and their

characteristics is presented in (77).
2.3.1. The SDP class
Consider tóúdé ‘shoe’. When referential cardinality is 1, the verb shows

s-agreement.2, 3
(8) Tóúdé ∅- dÓÓ

shoe 3S-be
‘It’s a shoe’
When referential cardinality is 2, the verb shows d-agreement.
(9) Tóúdé eM - dÓÓ

shoe 3D-be
‘It’s two shoes or a pair of shoes’
And when referential cardinality is 3, the verb shows p-agreement.
(10) Tóúdé gya-dÓÓ

shoe 3P- be
‘They’re shoes’
By triggering s-agreement when referential cardinality is 1, d-agreement

when referential cardinality is 2 and p-agreement when referential cardinality
is 3, tóúdé ‘shoe’ shows itself to be an sdp noun and so justifies sdp as a
noun class mnemonic.
Subsequent mnemonics are illustrated by running them through the three
sentence frames above: ‘It’s a ’, ‘It’s two s’, ‘They’re s’.
2.3.2. The SDI class
Tógúl ‘young man’ is an sdi noun.

2
It is to be assumed in all of the following examples that the agreement shown is the
only option permissible.
3
It will quickly be noticed that the phonological form of the agreement prefixes bears no
obvious relation to its gloss. The rather opaque relationship between prefixes’ sound and
meaning is a classic problem in Kiowa-Tanoan linguistics. See Chapter 5 for a discussion
of the Kiowa case.
26 CHAPTER 2
(11) Tógúl ∅- dÓÓ

young man 3S-be
‘It’s a young man’
(12) Tógúl eM - dÓÓ
young man 3D-be
‘It’s two young men’
(13) TógúúdÓ e- dÓÓ
young man.inv 3I-be
‘They’re young men’
2.3.3. The IDP class
Kútaa ‘pencil’ is an idp noun.
(14) KútaadO e- dÓÓ

pencil.inv 3I-be
‘It’s a pencil’
(15) Kútaa eM - dÓÓ
pencil 3D-be
‘It’s two pencils’
(16) Kútaa gya-dÓÓ
pencil 3P- be
‘They’re pencils’
2.3.4. The IDS class
Áá ‘tree’ is an ids noun.
(17) ÁádO e- dÓÓ

tree.inv 3I-be
‘It’s a tree’
(18) Áá eM - dÓÓ
tree 3D-be
‘It’s two trees’
(19) Áá ∅- dÓÓ
tree 3S-be
‘They’re trees’
2.3.5. The IDI class
K!Ôn ‘tomato’ is an idi noun.
(20) K!ÔMOM dO e- dÓÓ

tomato.inv 3I-be
‘It’s a tomato’
(21) K!Ôn eM - dÓÓ
tomato 3D-be
‘It’s two tomatoes’
tomato.inv 3I-be
‘They’re tomatoes’
2.3.6. The SDS class
P!ÓÓ ‘river’ is an sds noun.
(23) P!ÓÓ ∅- dÓÓ

river 3S-be
‘It’s a river’
(24) P!ÓÓ eM - dÓÓ
river 3D-be
‘It’s two rivers’
(25) P!ÓÓ ∅- dÓÓ
river 3S-be
‘They’re rivers’
2.3.7. The PPP class
Hólda ‘dress’ is a ppp noun.
(26) Hólda gya-dÓÓ

dress 3P- be
‘It’s a dress’ or ‘It’s two dresses’ or ‘They’re dresses’
28 CHAPTER 2
2.3.8. The SSS class
The appropriateness of the mnemonic sss cannot be illustrated via the sen-
tence frames used for the previous classes. Nouns in this class are all mass
nouns and, so, are, strictly speaking, uncountable. Nonetheless, the mnemo-
nic can be justified, by mass-to-count conversion and conjunction.4 The
significance of employing means different from those used above in order to
justify the current mnemonic is discussed at the end of this section.
As a preliminary to conjunction of sss nouns, observe that an increase in
referential cardinality by conjunction affects agreement the same way implicit
numeral modification above does. Thus compare (27), seen above, with (28).
(27) Tógúl eM - dÓÓ

young man 3D-be
‘It’s two young men’
(28) MŹMde tógúl
É gO óŹde eM - dÓÓ
this young man and that 3D-be
‘It’s this young man and that (one)’
In both cases, referential cardinality is 2, in (27), owing to implicit numeral

modification, in (28), owing to the conjunction of éˇŹˇde tógúl ‘this young
man’ and óŹde ‘that’ (1 + 1 = 2). Similarly, compare (29) and (30).

young man.inv 3I-be
(30) MŹMde tógúl
É gO óŹde yŹŹ e- dÓÓ
this young man and that two 3I-be
‘It’s this young man and those two’
In both cases, referential cardinality is 3, in (29), owing to implicit numeral

modification, in (30), owing to the conjunction of éˇŹˇde tógúl ‘this young
man’ and óŹde yŹŹ ‘those two’ (1 + 2 = 3).
Having illustrated that conjunction and numeral modification are equiva-
lent for the purposes of noun class identification, we can now use conjunction
to justify the sss mnemonic.5
4
Conjunction justifies also the assignment of other mass nouns to the ppp class.
5
Thóˇú
ˇ ‘water’, thóˇú
ˇÓlkhÓŹ ‘whisky’ [lit.: crazy water], xóŹ ‘coffee’ and xóŹgúl ‘tea’ are
(31) ThóMúM ∅- dÓÓ

water 3S-be
‘It’s water’
(32) ThóMúM gO xóŹ ∅- /*eM - dÓÓ
water and coffee 3S-/ 3D-be
‘It’s water and coffee’
(33) ThóMúM gO xóŹ gO xóŹgúl ∅- /*gya-dÓÓ
water and coffee and tea 3S-/ 3P- be
‘It’s water and coffee and tea’
Further motivation for the sss mnemonic comes from mass-to-count con-
version, that is, use of one water to mean ‘one portion / helping of water’.
Though the mechanisms of mass-to-count conversion themselves require clar-
ification, it is sufficient to defer this clarification for the moment and to
observe that mass-to-count converted sss nouns are still sss.
First, observe the typical sss agreement of a noun in this class.
(34) HÓ thóMúMÓlkhÓŹ an a- thónmO

q whisky hab 2s:3S-drink.impf
‘Do you drink whisky?’
Second, observe that, in (35), ‘whisky’ has been converted to a count noun.
Notwithstanding, we find s-agreement; d-agreement is unacceptable.
(35) YŹŹ thóMúMÓlkhÓŹ gya /*nen- thóm

two whisky 1s:3S-/ 1s:3D-drink.pf
‘I drank two whiskies’
Even when the explicit measure phrase k!Oâl ‘dish, cup’ is included, the verb
shows s-agreement.
(36) YŹŹ k!Oâl thóMúMÓlkhÓŹ gya /*nen- thóm

two cup whisky 1s:3S-/ 1s:3D-drink.pf
‘I drank two glasses of whisky’
(37) YŹŹ k!Oâl gya- thóm thóMúMÓlkhÓŹ
two cup 1s:3S-drink.pf whisky
‘I drank two glasses of whisky’
sss nouns in virtue of their semantics: all liquids are in this class, as discussed below.
30 CHAPTER 2
(38) PháM áM o k!Oâl an gya /*gyat- hânmO

three cup hab 1s:3S-/ 1s:3P-devour.impf
‘I used to get through three glasses [of whisky]’
The irrelevance of the measure phrase k!Oâl ‘dish, cup’ in determining

the s-agreement above is further underlined by the following sentence.
(39) PáágO k!Oâl xóŹ gya- thóm

one cup coffee 1s:3S-drink.pf
‘I drank one cup of coffee’
What is surprising here is the absence of inverse marking on k!Oâl ‘dish,

cup’. It is an idp noun:6
(40) K!OáttO bé- ól x!ep

dish.inv 2s:3I-drop.s/d-lay.s/d
‘You dropped the plate’
(41) K!Oâl men- ót
dish 2s:3D-drop.s/d
‘You dropped two plates’
(42) K!Oâl bat- p!ét
dish 2s:3P-drop.p
‘You dropped some plates’
As an idp noun, it is expected to show inverse marking for referential cardi-

nality 1. Yet, in (39), it is not inverse marked. This suggests that nouns used
as measure phrases do not function as normal nouns do for the purposes of
6
The reader may notice that (40) contains two verb roots, ‘drop’ and ‘lay’, whereas
(41) and (42) contain just one, ‘drop’. I have not been able to discover the difference that
‘lay’ makes. However, it should be noted that it is independent of referential cardinality.
(i) HÔndé a- ót?

what 2s:3s-drop.s/d.pf
‘What did you drop?’
(ii) Étté gyat- ól- k!op

many 1s:3p-drop.s/d-lay.p.pf
‘I dropped a lot’
See Section 4.7 for discussion of the surprising combination of incorporated s/d ‘drop’
with main verb p ‘lay’ in (ii).
number marking and agreement. Consequently, it is appropriate to attribute

the s-agreement in the preceding sentences to thó ˇ ú
ˇÓlkhÓŹ ‘whisky’ and xóŹ
‘coffee’ and to conclude that these nouns remain sss whether they are pure
mass nouns or have undergone mass-to-count conversion.
We therefore have two means of justifying the sss mnemonic. However,
one might regard these as means of justifying only an s mnemonic. The
problem is that mnemonics are meant to track agreement under changes of
referential cardinality. As all putative sss nouns are mass nouns, and as mass
nouns lack referential cardinality (one cannot say *‘It’s a water’ or *‘It’s two
waters’, et cetera), the mnemonic seems to track the untrackable.
Though the strength of this objection is real, it reveals at worst a slight
equivocation in the use and interpretation of triliteral mnemonics. Even if
mass noun mnemonics measure something other than referential cardinality,
what they measure is worth measuring, as it reveals that the agreement
behavior of mass nouns is not universal.7 Furthermore, qualms about the
meaning of mass mnemonics could have been sidestepped entirely by using,
for all classes, conjunction-based sentence frames, such as ‘It’s a and
a ’, without mention of referential cardinality per se. The conclusion
to be drawn is that the mnemonics are mere expository, or organizational,
conveniences. It is in this way that they are used here, preparatory to the
theoretical investigation of the next chapter.
2.3.9. The SII class
The sii class is special. Its sole member is the first person and it is the
only class triggering i-agreement for referential cardinality 2. Its existence
is justified on the basis of morphological syncretism. The full data set is
considered in Section 3.2.3. For now, I present the subpart that most imme-
diately motivates an sii class: for agents of (di)transitives and the subjects
7
Compare the non-singular agreement of English (i) with Kiowa (ii). The English shows
the plural agreement typical of any conjunction, whether of mass nouns or not, the Kiowa
shows s-agreement, the agreement triggered by each conjunct alone.
(i) Tea and coffee taste(*s) good
(ii) XóŹgúl gO xóŹ ∅ /*eM - t!ólOM OM

tea and coffee 3s/ 3d-tasty
‘Tea and coffee taste good’
32 CHAPTER 2
of unaccusatives, the first person exclusive dual and first person exclusive plural
trigger i-agreement. This is shown below for unaccusatives.8, 9
(43) NÓÓ a- dÓÓ

1 1S-be
‘It’s me’
(44) NÓÓ e- dÓÓ
1 1I-be
‘It’s me and him or me and her’
(45) NÓÓ e- dÓÓ
1 1I-be
‘It’s me and them’
Compare (44) and (45) with (46) which has i-agreement in virtue of the
inverse-marked tógúúdÓ ‘young men’.

young man.inv 3I-be
The crucial point is the emergence of i-agreement, inverse agreement, for

referential cardinalities 2 and 3, which strongly suggests an sii class.
2.4. Semantic coherence of the classes
We have seen that each mnemonic corresponds to some noun. It must now
be shown that each noun falls under some mnemonic. Ideally, to show this,
we would assign every noun in the language to one class or another. Such
thoroughness belongs, however, to a project of dictionary writing and is
excessive here. Instead, I establish the following claim:
(47) Kiowa nouns classes are internally coherent in that there are semantic
properties common to members of a given noun class.
8
Unergatives exhibit object agreement and are classed with transitives.
9
Kiowa has only two pronouns. NÓÓ, glossed as ‘1’, is used for first person singular, dual
and plural, inclusive and exclusive. Ám, glossed as ‘2’, is used for second person singular,
dual and plural. For third person, deictics are used (Watkins 1984).
The reader will observe as we progress through the classes that there is
unlikely to be a noun that cannot be assigned to any class, especially given
that, e.g., sdi is the default for animates, idp for vegetation and implements,
and sdp is default in general. Consequently, establishing (47) implies that
every noun in Kiowa does in fact fall into some class or other.
Let me now clarify what constitutes success in showing that the noun
classes are semantically coherent, or rather, let me forestall three likely mis-
understandings of what constitutes failure.
Non-uniqueness. The claim that classmates cohere semantically does
not entail the converse, that semantically coherent nouns are classmate.
Falsehood of the converse is not surprising, as one noun can possess charac-
teristics of two classes. For instance, rivers are moving bodies of water. Con-
sequently, p!ÓÓ ‘river’ might reasonably be assigned to one of two classes: to
sdi, which contains many things that move autonomously, or to sds, which
contains many bodies of water. As nouns must be assigned to one class
or another—class membership is not determined ‘on-line’ according to the
property most salient at the moment—‘river’ cannot be classmate with every
noun with which it shares semantic characteristics. Interestingly, Dr McKen-
zie, with whom Watkins worked, assigned ‘river’ to the sdi class, whereas
the speakers with whom I have worked assign it to the sds class.
Arbitrariness. Semantic generalizations over members of a class need
not be exceptionless. Exceptions are typical of noun class systems. For
instance, in Russian, there is a strong implication from real-world gender to
grammatical gender. Thus, nouns denoting females generally end in -a in
the nominative singular, as do ženščina ‘woman’, devuška ‘girl’, tsarina
‘czarina’, Karenina ‘female of the Karenin family’; such nouns end in -u in
the accusative singular. By contrast, nouns denoting males generally end in
-a in the accusative singular, but in a consonant in the nominative, as do
starik ‘old man’, maljčik ‘boy’, tsar j ‘czar’, Karenin ‘male of the Karenin
family’. Despite these robust generalities, mužčina ‘man’, ending in -a in
the nominative singular, patterns with female-denoting nouns. At the other
end of the classificatory spectrum are languages like Arapesh in which class
membership is primarily a matter of phonology (Foley 1986, Aronoff 1994).
Here too nouns are found in classes without exhibiting the class’ phonological
characteristic. Such classificatory residue does not undermine the claim that
noun classification in Russian or Arapesh is systematic. The same will be
true of Kiowa, in a very few cases.
34 CHAPTER 2
Subclasses. A single semantic characteristic need not be common to

all the nouns in a given class. Instead, they can divide into subclasses where
the nouns in each subclass share a single characteristic. We will see examples
of this, for instance, in the ppp class, which subsumes both abstract and
pluralia tantum nouns, and in the idi class, which comprises a hair and a
fruit subclass. Subclasses do not undermine the claimed internal coherence
of the classes for the following reason. Classes are defined by a classifying
number feature. Let P and P be two properties and let [F] be a number
feature. If there are conceptually natural relations both between P and [F],
and between P and [F], then [F] can serve as the classificatory feature for
nouns exhibiting either property. Consequently, the noun class defined by [F]
will comprise two subclasses, one for each property, P and P . Examination
of the internal semantic coherence of noun classes is a means to discover the
semantics of the number features. Having two semantic characteristics that
correlate with a single number feature is beneficial, not detrimental.
To reiterate, then, the aim is to show that nouns in each (sub)class share
a semantic characteristic, subject to the provisos immediately above.
In the next chapter, where we turn to the number features that underlie
the noun classes, I argue that the noun classes are ‘mnemonically’ coher-
ent; that is, idi nouns, say, cohere not merely in virtue of Property X, but,
furthermore, their mnemonic is semantically appropriate to Property X. In
other words, it is not accidental that the semantic property shared by idi
nouns correlates with the idi agreement profile, rather than with, say, the
ppp profile.
2.4.1. The animate classes
The SII class
This class is trivially semantically coherent, having a single member. More-

over, its member, the first person, is clearly semantically distinct from all
other class members.
The SDI class
The sdi class subsumes all animate nouns and several inanimate nouns that
share certain interesting properties with animates, centering on motion (cf.,
‘motility’, Noyer 1992). Let us begin with animacy.
Human animates. Words for humans are all sdi. For example:
(48) Gloss 1/2 3

baby iip!ÓÓgya iip!ÓÓgOt
child ŹŹ yyóŹ
child sân sâM aM dO
boy thalŹŹ thalyóp
girl máthOn máthOM OM dO
young man tógúl tógúúdÓ
young woman yÓkÓŹ yÓkÓŹgú
man k!yáM áM hẐMiM k!yáM áM hyóp (/hyôm/hyôMiM)
woman maayŹM maayóp
old, old man k!yápthOO k!yápthOOgO
maternal uncle têMiM têMiMdO
father’s mother thál tháályóp
wife tháá thêi
husband k!ŹŹ k!yóŹ
women’s sister p!ŹŹ p!yóŹ
man’s sister tháM áM tháM áM dO
Similarly, the sdi class includes designations for humans formed by at-
taching a gender suffix, -k!ii ‘male’ or -maa ‘female’, to a group name, such
as ‘Apache’, ‘Comanche’, ‘Mexican’, ‘White’, or adjective or noun, such as
‘crazy’, ‘chief’, ‘big, old’.
(49) Gloss 1/2 3

Apache man/woman ThOgûik!ii/maa ThOgûi
Black man/woman Paidômdek!ii/maa PaidôMuM bO
Black man/woman KhóMúM gŹMóMúM k!ŹŹ/máá KhóMúM gŹMóMúM
Comanche man/woman Kyâik!ii/maa Kyâigu
Mexican man/woman K!ópt!ÓkhÓŹk!ŹŹ/máá K!ópt!ÓkhÔi
White man/woman T!ÓkhÓŹk!ŹŹ/máá T!ÓkhÔi
chief k!yátáŹk!ii k!yátâi
elder10 êlk!ii/maa bẐMiMdO
madman/madwoman ÓlkhÓŹk!ŹŹ/máá ÓlkhÔi
10
‘Old’, from which comes ‘elder’, is suppletive for number in Kiowa. See Chapter 4.
36 CHAPTER 2
When referential cardinality is 3, the gender suffix is omitted and is replaced

by the inverse (Watkins 1984, p. 196). Alternatively, there are gender specific
inverse forms, -k!yoi for males, -maimO for females.
Non-human animates. Also in the sdi class are animal, bird, reptile
and insect names. McKenzie (n.d.a) lists 135 birds and reptiles and McKenzie
(n.d.b) lists 154 mammals, all but one sdi nouns.11 A sample, and some
insects and similar creatures, is given below (not all from McKenzie’s lists).
(50) Gloss 1/2 3

bird t!eM Miné t!eM Minóp
butterfly khOibatôulé khOibatôulóp
cedar waxwing zont!OOiyátmaa zont!OOiyátmaimO
dog xégun xéguMuM dO
frog khOOlék!yâalé khOOlék!yâalóp
grasshopper k!OOlÓtkÔMOM yŹ k!OOlÓtkÔMOM yóp
horse xêMiM xêMiMgO
kid kááboliiii kááboliiyyoi
quail pêMiMsyan pêMiMsyaM aM dO
rattle snake saM aM néhiMMi saM aM néhyoM Mi
snake saM aM né saM aM nóp
spider k!OM OM nÓÓthOM OM k!OM OM nÓÓthOM OM gO
Inanimates. Although all animates are sdi, not all sdi nouns are an-
imate. One such is hêˇǐii ‘doll’. Its subsumption suggests that animate-
likeness is sufficient for sdi membership. In the case of ‘doll’, the nature of
this likeness is clear, but it leads one to ask what other core properties of
animacy inanimates may share.
Self-propulsion or the ability to determine the course of motion is one
such property. Several sdi inanimates are explained in this way, including
heavenly bodies, machines and implements.12
11
The unique exception is the sdp noun kÓl, for which McKenzie writes:
A cow or cows; a buffalo or buffaloes; usu., in the sense of “beef” or “beeves.”

Sometimes, “herd” or “herds.” (The term has only one form-like English
“sheep.”) (McKenzie n.d.b, p. 3)
The availability of the collective ‘herd(s)’ makes this word different from sdi animals.
12
Watkins gives p!ÓÓ ‘river’ as an sdi noun, which might be explicable on these lines.
For my consultants, it is sds, however.
(51) Gloss 1/2 3

moon p!ÓÓ p!ÓÓgO
star táM áM táM áM gO
sun páŹ páŹgú
car khÔO khÔMOM gO
wheel, wagon, car k!Ódál k!ÓdáttO
awl, (bone) needle xóMúM xóMúM gO
axe hÓMÓMthoM uM hÓMÓMthoM uM gO
comb ÓlsóMúM ÓlsóMúM gO
eyebrow tweezers tááOlxon tááOlxoM uM dO
hoe dOmkûu dOmkûugO
knife k!ÔO k!ÔOgO
scissors t!áMŹMkhÓtháá t!áMŹMkhÓtháágO
However, three terms given by Watkins do not lend themselves to this way
of thinking: k!ÔOsoMǔ ‘whetstone’ [lit.: knife-grind], hôux!o ‘decorative silver
button (worn on head)’, and t!ÓˇÓˇ ‘spoon’. She suggests for the latter that
being made of animal material is relevant—Kiowas frequently made spoons
from horn, an sdi body part.
Several body parts belong to the sdi class. Watkins lists the following:13
(52) Gloss 1/2 3

buttock t!él t!éttO
ear t!ÓÓdé t!ÓÓgÓ
egg t!áMŹM t!áMŹMmÓ
eye tááde táágO
gall bladder t!ÓMÓMdé t!ÓMÓMgÓ
heart thén théMŹMdO
hide khÓŹ khÓŹgO
hip pŹŹthel pŹŹthettO
horn gúM úM dé gúM úM gÓ
kidney t!ÓlthOn t!ÓlthOM OM dO
knee t!elbôMuM t!elbôM uM gO
leg thóMúM dé thóMúM gÓ
liver t!Ólel t!ÓlettO
13
She also includes mOnx!óˇ ‘fingernail’. In my fieldnotes, it is sds, as are Onx!óˇ and
Onk!Ón, both of which mean ‘toenail’.
38 CHAPTER 2
Gloss 1/2 3
neck k!ól k!óttO
toe Onthál OntháttO
tongue dén déMŹMdO
spinal cord pâisen pâiseM MidO
spine gómthoM gómthoM gO
tooth zóMúM zêm
tripe ÓÓbŹMŹM ÓÓbŹMŹMgÓ
Some of these body parts again suggest salient properties of animate beings,
such as motion and movement (‘legs’, ‘spine’, etc.), perception and speech
(‘ear’, ‘eye’, etc.), being a major organ (‘heart’, ‘liver’, etc.). Others (‘horn’,
‘tooth’) are implement-like (in particular, cutter-like), as are some items in
(51).
For the most part then, we see that sdi nouns are animates or inanimates
that share certain salient properties with animates, such as ability to move
or determine direction of motion. There is a residue of cases not readily
explained by such principles. However, their relative rarity suggests that
the principles are broadly correct and that the exceptions may be no more
than the slight arbitrariness expected of any class system. Alternatively,
the correct analogy with animate entities may simply have eluded me; these
elements may be subsumed under a separate subclass in virtue of a different
semantic property. In either case, they do not undermine the claim that
members of the sdi class semantically cohere.
2.4.2. The main vegetal classes
The idp and ids classes are, with sdi, the largest in the language. They
subsume ‘plants and plant material, natural and man-made objects and a
small number of body parts’ (Watkins 1984, p. 85). Examples of each are:14
14
Many thanks to Laurel Watkins for supplying inverse forms missing from my notes.
(53) Gloss 1 2/3

apple, peach, plum tree álOOaadO álOOaa
corn éŹtháttO/éŹthâatO éŹthâl
cottonwood ááhyoM Mi ááhiMMi
cottonwood ááthap ááthap
elm kÓlaadO kÓlaa
elm thOOáádO thOOáá
grape vine t!éttéeibO/gO t!éttéei
grass sóMúM dO són
hackberry ááephep ááephep
Osage orange zépgutk!oaadO zépgutk!oaa
pecan trees OM OM kuáádO OM OM kuáá
persimmon tree ááhyalaadO ááhyalaa
peyote, cactus séMŹMgO(t) séMŹM
pine tree zónaadO zónaa
redbud khẐMiMgulaadO khẐMiMgulaa
skunkberry bush táMŹMpeeibO/gO táMŹMpeei
tree áádO áá
walnut tree phohónaadO phohónaa
watercress thóMúMéŹbÓ/gÓ thóMúMéŹ
weed thépsOyaadO thépsOyaa
willow séáádO séáá
(54) Gloss 1 2/3

bean t!ÓlthOM OM dO t!ÓlthOn
cantaloupe phŹdátkÓt phŹdátkyá
firewood k!ŹŹbÓ/gÓ k!ŹŹ
flower áákhiMMigOt áákhiMMigya
leaf áŹdeM MigO áŹdeM Mi
nut zémk!ÓÓdop zémk!ÓÓde
onion sôutO/sôl15 sôl
pecan nut OM OM kuéŹbÓ/gÓ OM OM kuéŹ
seed éŹbÓ/gÓ éŹ
walnut phohóneibO/gO phohónei
15
Dr McKenzie gives the first form, which is regular given (89), Mrs Kodaseet the second,
which unexpectedly shows zero marking after l.
40 CHAPTER 2
(55) Gloss 1 2/3

arrow zêibOt zêiba
belt tOnphâM aM gO tOnphâM aM
blanket khÓÓdÓ khÓÓ
breech cloth tÔigop tÔide
bucket dOâatO/dOáttO dOál
cradle pháM áM top pháM áM tol
door xát xát
ember phŹŹdÓ phŹŹ
fan sálpákomgOt sálpákomgya
gourd rattle thÔMOM gO thÔMOM
gun hÓMÓMgOt hÓMÓMgya
lamp phŹbóMáádO phŹbóMáá
match tháádóp tháádé
paper mÓx!áMŹMmO mÓx!áMŹM
peg, stake tóp tól
pipe sÓÓtóp sÓÓtól
saw ááthagO áátha
sheet for bed t!áMŹMmÓ t!áMŹM
shield khyôMiM khẐMiM
(56) Gloss 1 2/3

bone thóMúM segO thóMúM se
feather (downy) x!óMúM gOt x!óMúM gya
feather (hard-stem) áágÓt áá
foot Onsôi Onsóú
head Ólthêm ÓlthóMúM
large intestine sékhôMiM sékhóMúM
lymph gland áâatO áál
nose, beak mOOk!Ôn mOOk!Ón
penis sóp sóú
rib gúúdÓ gúú
scalplock kyâip!OM OM dO kyâip!On
sinew téMŹMgÓt téMŹM
udder OOzâi OOzáá
wing x!ôutO x!ól
Given the classes’ size, membership is best defined negatively, as excluding

all plants, natural objects, man-made objects and body parts belonging to
other classes.
Definition by exclusion suggests that these classes are to some extent de-
fault. However, this is not to say that they are incoherent. Watkins observes
that compounds built on áá ‘stick, pole, tree, brush, timber’ provide insight
into the difference between the basis of idp∼ids classification. Trees that
grow in spinneys and groves are naturally grouped and hard to individuate,
so that the collection of trees itself can by thought of as a single unit. Such
trees tend to be ids nouns, as, indeed, is áá ‘tree’ itself.
(57) IDS Gloss 2/3

elm thOOáá
hackberry ááeiphep
pecan OM OM kuáá
pine zónaa
tree áá
This is corroborated by a spontaneous comment of Mrs Dupoint’s: the fol-
lowing sentence would be said when one sees a grove.
(58) TéŹ phohónaa ∅- dÓÓ

all walnut tree 3s-be
‘They’re all black walnuts’
By contrast, idp trees tend to be easily individuable even when there are
several of them. This can be due either to their relative smallness or distance
from other trees of the same kind.
(59) IDP Gloss 2/3

redbud khẐMiMgulaa
weed thépsOyaa
willow séáá
As an idp noun, áá means ‘pole, stick’, which is smaller than a tree and, as an
implement, bound up with human intentions, tends towards individuability.
As there are a few discrepancies between Watkins’ idp/ids classification
of plants and mine, I shall not assign all the words in (53) to one or other
category. The grounds for the discrepancy are unclear. Possible sources are
artifacts of judgment elicitation methodology, regional variation, or difference
42 CHAPTER 2
in speakers’ knowledge of the relevant properties of the trees in question.

Uncertainty with respect to some items does not, however, undermine the
validity of the observation that a concept of grouphood distinguishes ids from
idp nouns, especially as the same concept pertains to the classification of
idi, sds and sss nouns. Indeed, differences between idp and ids implements
plausibly reflect grouphood, too. Two types of implements, primarily, are
ids: those that give off light or heat (Takahashi 1984), such as embers,
lamps, guns, and pipes, and those that act as covers, such as blankets, door
flaps, and shields. All of these are capable of collective action, that is, a
single outcome emerges from the group.
So, the idp and ids classes are semantically coherent in two ways: they
are the main classes for trees and other natural entities, and they are distin-
guished by a property of grouphood.
2.4.3. Symmetric non-constant classes
There are four classes with symmetric mnemonics, i.e., mnemonics of the
form xyx. Two of these, ppp and sss, are constant, whereas as the other
two, idi and sds, are non-constant. We turn to the latter pair now. Section
2.4.4 addresses ppp and sss.
The IDI class
The idi class is very small, with only eight members (four, according to
previous reports), mostly hair and fruit.16
(60) Gloss 2 1/3

eyebrow tááOl tááOOdO
eyelash tááphO tááphOgO
hair (of head) Ól ÓÓdÓ
apple / plum álOO álOOgO/bO
blackberry saM aM néei saM aM néeigO
brain17 k!yagóp k!yagóp
orange thóMt!ólOM OM thóMt!ólOM OM gO
tomato k!Ôn k!ÔMOM dO
16
Póúei ‘strawberry’ may also be idi. On one occasion, Mr Bointy assigned phŹdátkyá
‘cantaloupe’ to this class, though at all other times, he and others treat this as idp.
Fundamental to this class is, I suggest, the observation that the Italian
capelli is a count noun whereas its nearest English counterpart, hair, is a
mass noun (Chierchia 1998). Hence the following contrast, where the singular
is preferred to the plural in English, but vice versa in Italian.
(61) a. My hair is long

b. ??My hairs are long
(62) a. ??Il mio capello è lungo
the.masc.s my.masc.s hair is long.masc.s
b. I miei capelli sono lunghi
the.masc.p my.p hair are long.masc.p
‘My hair is long’
Chierchia observes that ‘hair’ vacillates between a mass noun and count
noun crosslinguistically, in contrast to strongly count nouns, such as ‘man’
or ‘woman’, or strongly mass nouns, such as ‘sand’ or ‘water’.
idi nouns are those that can be easily regarded either as individual or
as part of collections. With regard to hair, this arises because strands of
hair are clearly individual, the minimal parts of bodies of hair. Yet, in a
head of hair, or a fringe of eyelashes, or an eyebrow, the individual parts are
not readily recognizable, and so the body too is a salient individual, albeit a
group-like one.
idi fruits—apples, plums, oranges, blackberries, tomatoes—are like hair
in that they are clearly individuable and yet grow clustered together in col-
lections that are themselves generally more salient than the individuals that
comprise them. However, when gathered or plucked, they, unlike, say, grain,
are salient individually. Thus, like hair, idi fruits can naturally be conceived
of both as individuals and as part of collections.
This line of thought does not readily extend to brains. They are certainly
composed of parts poorly distinguishable in the whole; however, I do not
know whether their typical mode of preparation in Kiowa cookery would bre-
ak them into distinguishable subparts.18 Leaving brains aside (until p. 95),
17
When a noun takes a zero allophone of the inverse marker, as is common for p-
final nouns like k!yagóp ‘brain’, class membership is still detectable through Merrifield’s
method of examining the agreement the noun triggers on the verb for referential cardinal-
ities 1, 2, and 3.
18
Mrs Dupoint said: An k!yagóp dé-pÓttO. T!ěǐnéxeiyot!ǎǐmO gO k!yagóp ét-
khOleitOnmO. E-ÓÓt!olǑǑ ‘I eat brains. They cook cook them together with chicken eggs.
44 CHAPTER 2
idi nouns share two properties: salient individuality, but occurrence in large
collections where the collection itself is salient.
Watkins has observed a third, morphosemantic, property of this class.
The non-inverse noun, with s-agreement, yields a ‘different types of’ reading.
Note, again, the relevance of grouphood in connection to s-agreement.
(63) ÁlOO bâa∗- bOO

apple 2p:1s:3S-bring.imp
‘Bring me apples [of three or more sorts]’ (Watkins 1984, pp. 88–9)
(64) HÓndé Ól bÓ- sÓMÓMmŹŹ
what hair :2p:3S-interesting
‘What interesting [kinds of] hair you all have’
idi nouns, therefore, are semantically coherent in virtue of three properties.
The SDS class
Like the idi class, the sds class is comparatively small. It subsumes:19
(65) Gloss 1/2/3

cloud phán
house (not teepee) tóú
land(holding) dÓm
path, road hóMOn
pond séxó
puddle xenthóMúM
river p!ÓÓ
song dÓÓgya
fingernail mÓnx!óM
ring sôude
toenail Onk!Ón
toenail Onx!óM
Small as it is, the list comprises two subclasses. The last four nouns share
the surprising property of ‘being digit appendages’. Interestingly, several
They’re really delicious.’ The implication was that brains and eggs were scrambled.
19
Watkins gives xóŹ ‘liquid’ as sds. I have this word only as sss with the meaning
‘coffee’. Given that all liquids I have recorded are sss, to find ‘liquid’ in this class would
be surprising, unless it meant ‘body of liquid’, a sense similar to several items in (65).
implements also trigger s-agreement for 3, those capable of collective action,

of working as a group to produce a single outcome or effect (ids, p. 42).
Regarded as implements for grasping and grappling, and for protecting finger-
and toe-tips, nails, interestingly, share this property; though they differ from
the ids nouns in being body parts, rather than inanimate implements.
The other sds nouns are ‘non-shape-inductive’ (cf., Takahashi 1984).
That is, although all are clearly individuable (one can tell one river from
another, one song from another, and so on), the shape of one is a poor in-
dicator of the shape of the next. One cannot extrapolate from the shape of
one to the shape of the next. This is an unexpected criterion of classhood.
However, strong support for its actuality comes from my having discovered
the sds nouns ‘cloud’, ‘puddle’ and ‘song’ by thinking of things that are
individuable but non-shape-inductive.
2.4.4. Symmetric constant classes
Kiowa has two more classes with symmetric mnemonics, ppp and sss. These
are strongly semantically coherent.
The SSS class
The sss class consists exclusively of non-granular mass nouns, such as ‘milk’
and ‘honey’, or mass nouns of dubious granularity, such as ‘snow’ and ‘sleet’,
which, if granular, are so primarily when descending.
(66) Gloss 1/2/3

beer phŹtthóMúM
coffee xóŹ
juice, soda thóMúM t!ólOM OM
milk kÓlzep
soda thóMúM gul
tea xóŹgúl
water thóMúM
whisky thóMúMÓlkhÓŹ
honey t!elséppenhaa
honey, syrup áápenhaa
sugar pénhaa
46 CHAPTER 2
Gloss 1/2/3
ash sOphán
hail, sleet t!én
ice téMŹMgya
rain sép
snow t!ól
eyebutter, sleepy dust táásek!On
snot, mucous sén
All the names for liquids in my fieldnotes are in this class. Some items
closely akin to particular liquids (for instance, semi-liquid or frozen forms of
precipitation), and sugar (which, when Kiowas first encountered it, came in
blocks and not granularly) are included by association. The class’ semantic
coherence is obvious.
The PPP class
The ppp class is more heterogeneous than the sss class, subsuming two or,
depending on the speaker, three subclasses: pluralia tantum and composite
nouns, abstract nouns, and, for some, granular mass nouns.
Pluralia Tantum and Composite Nouns. Objects composed of
several parts are constant plurals in Kiowa. These include items familiar
as pluralia tantum from English, such as ‘trousers’.
(67) Gloss 1/2/3

belongings, stuff ól
book, letter, picture kút
book, letter tóMúM kút
choker k!ólphaM
clothing, shirt t!óMúM gya
hoop game k!ÓdálOOgya
necklace k!ólphaM aM
pack saddle olpháM thOOgya
quilt pál
roach (headdress) k!ókÓŹouphOl
teepee tóú
trousers khÓÓdé
war headdress aat!OhÓŹ
Granular Mass Nouns. Granular mass nouns are distinguished from

non-granular mass nouns in having natural minimal parts. Thus, sand, which
comes in grains, is granular, whereas water, which does not, is not. Some
examples are:
(68) Gloss 1/2/3

flour éŹtOM
pepper xóŹséMoM uM gya
rice éŹyóguei
salt ÓÓtháM t!aM Mi
sand péŹgya
These nouns are classified as ppp on the basis of conjunction, analogously to

sss nouns:
(69) ÓgO gya-kót gO ÓgO gya-phÓÓ dé gya-/*eM - mOsótdOO

rel 3p- expensive and rel 3p- cheap nom 3p- / 3d-be mixed
‘The expensive and the cheap (salt) are mixed up’
Although two quantities of salt are talked of, we find p-agreement, not d.
It is, however, to be noted that these nouns are not ppp for all speak-
ers. Rather, they appear to be idp nouns, with the inverse-marked noun
corresponding to the minimal part:
(70) XóŹséMoM uMgOt mÓÓ e- ét

pepper.inv like 3i-big.s
‘The pepper grain is kind of big’
Such speakers permit a corresponding dual reading for the granular sense.
(71) MŹMde yŹŹ xóŹséMoM uM gya eM - bẐn

É
this two pepper 3d-big.d/p
‘These two grains of pepper are big’
However, speakers for whom granular mass nouns are ppp strongly reject
sentences like (70) and (71).20 Instead, they offer syÓnde xóŹséˇǒǔgya ‘a
little pepper’ as a Kiowa equivalent for ‘a grain of pepper’.
20
Watkins gives no granular mass noun as ppp but assigns them to her Class II, a
superclass of my idp (her IIa) and ids (her IIb). Her exposition of the noun class system
does not dwell on mass nouns and so the crucial sentences are not supplied.
48 CHAPTER 2
ppp classification of granular mass nouns is on a semantic par with the

treatment of pluralia tantum: both trousers and piles of salt are single objects
composed of complex parts. Note, however, that this composition of parts
differs from the group-like idi and sds plurals (and ids below). There, the
plurality consists of naturally salient parts, such as apples or hairs, or clouds
or rivers. In the case of pluralia tantum nouns, the constituent parts are
non-uniform and so there is no minimal part; and, in the case of granular
mass nouns, though there are minimal parts, a grain of sand is not salient to
the same extent as, say, an apple is.
Abstract Nouns. The second ppp subclass common to all speakers
is reserved for abstract nouns. Watkins observes that abstract nouns, such
as ‘zero’, ‘biology’, ‘adverb’ are ppp (though, with the discussion focused
on productivity of noun classification, Kiowa examples are not provided;
Watkins 1984, p. 92). The following from my fieldnotes fit this description.
(72) Gloss 1/2/3

problem mOÓŹde
dance kúngya
word, language tóMúM gya
sleep déMŹMgyá
work sÓÓtégya
She hypothesizes that this is a natural extension of the use of p-agreement

for unidentified or unspecified arguments, exemplified below.
(73) Gya-gŹMŹM- sal

3P- night-hot
‘It’s a hot night’ (Harrington 1928, p. 64)
(74) YáM - kÔm háágyá gya-k!úl
2s:1s:3P-indicate.imp where 3p- be lying.p
‘Show me where they are’ (Watkins 1984, p. 235)
(75) NÓÓ mán- pŹMŹMÓmtÓÓ nO dáal mán∗- pOO
1 1s:2d:3P-cook.fut conj must 2d:3P-eat.imp
‘I will or If I cook for you, you must eat’ (ibid., p. 237)
If so, ppp classification of abstract nouns is independent of pluralia tantum,

composite nouns and granular mass nouns. Consequently, the class comprises
two disjoint subclasses, each semantically coherent.
2.4.5. The default class
Speakers of languages for whom Kiowa’s noun class system is somewhat

foreign are likely to anticipate sdp as the most natural class, given that sdp
agreement transparently reflects referential cardinality. Yet the class is small:
(76) Gloss 1/2/3

arrow smoother zêibat!OM MiOM OM ba
boot tókŹMŹMnŹŹ
earring t!Oáá
key pháM áM xo
moccasin touhẐMiM
nail (for hammering) hÓMÓMt!Okuu
rock, stone x!óú
rope yáŹpÓ
shoe tóúdé
Naturally, it excludes elements that ought to be in other classes, such as
animates, vegetation, mass nouns, and non-shape-inductive items. However,
there is little in the way of positive criteria for this class, beyond a collection
of footwear. This suggests that sdp is the default class, to which nouns are
assigned when they do not meet the membership criteria of other classes.
2.4.6. Summary
The foregoing discussion of the noun classes has shown that they are inter-
nally semantically coherent, that is, that there are properties that members
of each class generally share. These are summarized below.
50 CHAPTER 2
(77) Class Characteristics

sii First person
sdi Animates and animate-like entities (physically similar
to animates or capable of self-propulsion or determin-
ing direction of motion)
idi Naturally regarded both as individual and as occur-
ring in collections; permits ‘different types of’ reading
with s-agreement
sds Individuable, non-shape-inductive
sss Non-granular mass nouns
ppp Pluralia tantum, composite nouns (and granular mass
nouns for some speakers); abstract nouns
idp Default for vegetation and implements; granular mass
nouns for some speakers
ids Vegetation forming natural collections in which mem-
bers are not readily individuable; implements that act
as a group to produce a single effect
sdp Default; no unifying properties
In the coming chapter, we turn to a second way in which the classes are
semantically coherent, namely the connection between the semantic charac-
teristics of class members and the class mnemonic. That is, I will argue that,
given the internal semantic coherence of the classes, there could not be a
language, Kiowa , in which the mnemonics and characteristics are randomly
permuted. Before that, the coming two sections discuss and dismiss the pos-
sibility of a tenth noun class and the notion that class membership is to some
extent phonologically, rather than purely semantically, determined. Readers
of unbridlable enthusiasm may skip these sections.
2.5. Against a tenth class
Noun classes were distinguished above by examining the correlation between

referential cardinality and agreement, following Merrifield (1959a). Harring-
ton (1928), employing a similar method, identified a class not represented
above. On current terms, it would be sda, where a-agreement is animate plu-
ral agreement.21 Watkins explains that speakers typically use a-agreement to
21
Harrington distinguished between ‘animate major’ (sda) and ‘animate minor’ (sdi)
categories; pp. 14, 237ff.
refer to fellow tribal members but i-agreement for members of other tribes, as
illustrated below. Observe that, in both cases, the nouns are inverse marked.
(78) KÓŹgú á- kúúyÓ

Kiowa.inv 3A-be lying.p.distr
‘Kiowas are camped about’ (Watkins 1984, p. 84)
(79) Kyâygu e- kúúyÓ
Comanche.inv 3I-be lying.p.distr
‘Comanches are camped about’ (ibid.)
Unlike Harrington, Watkins does not treat sda as a separate class however.
Watkins’, I believe, is the correct response.
Nearly any noun, or for some speakers, any noun, capable of triggering
a- is also capable of triggering i-agreement. So, on the assumption that
classes are pairwise disjoint, sda is not a class distinct from sdi. Rather,
nouns triggering a-agreement are sdi nouns with a special property. On the
basis of the following types of complementary examples, I suggest that this
property is ‘empathy’ or ‘degree of identification’ with the sdi noun.
In a hunting story related to me by Mr Bointy in August 2001 (see Ap-
pendix), k!yáˇ á
ˇ hyóp ‘men’ triggers both i- and a-agreement. It relates an
incident in which Mr Bointy was not a participant. The three opening sen-
tences state where the men were and what they were doing—a simple report-
ing of facts, with little scope for empathy. The verbs show i-agreement.
(80) PháM áM o k!yáM áM hyóp e- MŹ

ŹM- aM aM - hel áá- dom. SyÓn- de gya-hóMOn-
three man.inv 3I-hunt-come-hsy tree-loc small-adv 3p- path-
dOM OM -mei dé- em téŹ e- khOléŹ- aM aM - hel. GO e- xánde-
be- impf.hsy nom-loc all 3I-together-come-hsy conj 3I-reach-
hel ÓgO áádO e- bôn- dOM OM -mei dé- eM Mi
hsy rel tree.inv 3i-bend-be- impf.hsy nom-loc
‘Three men were hunting in the woods. They were all going along
a narrow path when they arrived where the path lay under a bent
tree.’
Then follow a few sentences of reported speech, concerning what should be

done about a panther lying in ambush. These relate an experience with
which Mr Bointy, an experienced hunter, could well empathize. Thereafter,
the verbs show a-agreement.
52 CHAPTER 2
(81) NO thÓp á- khûtde- hel. ∅- T!Om-áM áM - dé gôm

conj beyond 3A-pop/flee out-hsy 3s-first- come-nom back
gya- bóMúM - hêl nO héMŹM ∅- dÓMÓM-mêi. Tékhop á∗-
3s:3p-look-hsy conj gone 3s-be- impf.hsy everywhere 3A:3s-
donhel né hÓn á∗- thOM OM -mOO.
seek-hsy but neg 3A:3s-find- neg
‘So they passed [quickly] to the other side. The one at the head
looked back and it was gone. They looked for it all over but didn’t
find it.’
Another pair of examples on similar lines concerns animals. In (82), which

is about horses, the verb bears a-agreement, whereas in (83), which is about
ants, it bears i-agreement. (I assume (82) to have a pro argument; this is
irrelevant to agreement type; Watkins 1990, Adger and Harbour 2007.)
(82) Á- zéMŹMma

3P-move.p.impf
‘They (horses in a pasture) are milling around, moving around’
(Watkins, p.c.)
(83) EmhâM aM mop e- xóMúM - zéMŹMma
ant.inv 3I-crawl-move.p.impf
‘Ants are crawling around’ (Watkins, p.c.)
a-agreement in (82) does not motivate an sda class, as, generally, xêˇǐgO
‘horses’ triggers i-agreement, so that xêˇǐgO is best regarded as the inverse-
marked form of the sdi noun xêˇǐ ‘horse’. A typical example is:
(84) ÔihyOO xêMiMgO Ó- phÓÓhel

then horse.inv :3s:3I-get.pf.hsy
‘That’s when he got the horses’
Once again, I propose that a-agreement in (82) should be explained by appeal

to empathy, given the centrality of horses to traditional Kiowa culture—a
conclusion drawn on the basis of numerous stories and facts told me by Mr
Bointy in the course of our work together—a centrality, or empathy, in which
ants by no stretch of the imagination participate.
Grammaticalized empathy is in part a cultural phenomenon. Indeed,
Watkins observes that i-agreement has a slightly disrespectful connotation
when used for adults. a-agreement is all but obligatory for KÓŹgú ‘Kiowas’:
(85) KÓŹgú né- /??nÓ- xán

Kiowa.inv :1s:3A/ :1s:3I-arrive.pf
‘The Kiowas came to me’
The constant occurrence of KÓŹgú ‘Kiowas’ with a-agreement may be seen

as a default, cultural empathy between any Kiowa speaker and their kin.22
Given that putative sda nouns are just empathized with sdi nouns, it is
incorrect to posit a tenth class (with fluid membership).
2.6. Phonological incoherence
Crosslinguistically, membership of noun classes is frequently phonologically

determined. Phonological classes are of two types: there are Bantu-style
classes (see Chapter 6), where noun and adjective begin with m- if the noun
is in Class 1, but with wa- if it is in Class 2; and there are Indo-European-
style classes, where nouns with a particular ending are all classmate, as are
Latin nouns in -iō. Bantu-style classes are apparent from phonological ele-
ments external to the noun itself; Indo-European-style classes, from elements
internal to the noun. Dramatic instances of the latter are the Arapesh and
Yimas languages of Papua New Guinea (Foley 1986, Aronoff 1994). A natural
question is whether Kiowa’s noun classes are (to some extent) phonologically
defined and whether, then, they are not organized purely semantically.
However, Kiowa noun classes are not phonologically defined. First, the
ten or so allophones of inverse marking distribute freely across all inversive
classes (Section 2.6.1), so classes are not defined Bantu-style, by selecting
particular affixes. Second, thematic nouns (defined in Section 2.6.2), which
could imaginably yield Indo-European-style classes, cross-cut the number
classes and show semantics to be the key determinant of class membership.
22
The notion of ‘default empathy’ may help to explain why a-agreement is systematically
ambiguous between animate plural agreement and reflexive agreement:
(i) De- hól

1s:3a-kill.pf
‘I killed myself or them (people)’
The reflexive ‘oneself’ might be regarded as involving obligatory empathy, rather than as
being in some mysterious sense obligatorily animate plural.
54 CHAPTER 2
2.6.1. Inverse allophones
The only morpheme capable of yielding Bantu-style noun classes in Kiowa

is the inverse, as it is the only nominal affix that triggers agreement. For
instance, in (86), the idp direct object in the relative clause is inverse marked
and triggers inverse marking on the subordinate verb, on the relative clause
itself and on the matrix verb.
(86) [PŹMáádO e- ét ]-gO dé- hÓÓgya

table.INV 3I-big.s -INV 1s:3I-get.pf
‘I bought a table that is big’ (Watkins 1984, p. 231)
Similarly, demonstratives agree for inverse with their noun. This is shown
below for the sdi noun sân ‘child’. In (87), where referential cardinality is
1, there is no inverse marking; however, in (88), where referential cardinality
is 3, there is inverse marking on the demonstrative, the noun and the verb.
(87) MŹMde /óŹde sân ∅- khópdOO

É
This/that child 3s-sick
‘This/that child is sick’ (Watkins 1984, p. 97)
(88) MŹMgO
É /óŹgO sâM aM dO e- khópdOO
This.INV/that.INV child.INV 3I-sick
‘These/those children are sick’ (ibid.)
The inverse marker has some ten distinct forms. So, it is reasonable to
ask whether any are proprietary to particular classes. To show that they are
not, it suffices to show that all are purely phonologically conditioned.
The inverse marker has the forms in (89).23 Thematic nouns (-th) such
as ‘tobacco’ are discussed in Section 2.6.2.
23
This follows Watkins, except that -∅ is my addition. Some phonological comments are
needed. First, note three informally stated processes: (i) VN→V M VM in the context of inv
dO; (ii) l→t in the context of tV; (iii) V→∅ in the context of V V. Second, note
that the tone of the inverse marker need not be specified, contra Watkins’ practice, as it is
predictable (Harbour 2002; -gu resists local tone lowering; hence the final high in KÓŹgú,
in contrast to final lows in KÓŹmaa ‘Kiowa woman’, KÓŹtǒgya ‘Kiowa language’).
(89) Suffix Example

Form Context Basic∼Inverse Gloss
-mO MŹM
V t!áMŹM∼t!áMŹMmO egg
-bO m kóm∼kóMúM bO friend
n k!Ôn∼k!ÔMOM dO tomato
-dO
l tógúl∼tógúúdÓ young man
-tO l tâl∼tâttO skunk
-Ot th tháábá∼tháábÓt tobacco
-gu Vi KÓŹ-∼KÓŹgú Kiowa
K!ópsénáŹ-∼K!ópsénâi Mexican
-̂i Vi
ÓlkhÓŹ∼ÓlkhÔi madman
-oi ii p!ŹŹ∼p!yóŹ (female’s) sister
-op e saM aM né∼saM aM nóp snake
k!óp∼k!óp mountain
-∅ p/t
xát∼xát door
xêMiM∼xêMiMgO horse
-gO elsewhere
thóMúM se∼thóMúM segO bone
(Note that -gu∼-̂i are in competition and must be learned on a case-by-

case basis.) Tone change too may occur, sometimes with vowel change or
suffixation.
(90) Processes Basic∼Inverse Gloss

tone change only mOOk!Ón∼mOOk!Ôn nose
tone + vowel change thâa∼thêi wife
tone change + suffixation zêiba∼zéŹbOt arrow
And there is a unique l∼p alternation in words built on tól ‘peg’ (55).
The point to observe, relating to (89) and (90), is that no form of the
inverse marker is proprietary to any one noun class. Each is found in any class
with a phonologically appropriate noun in it. Hence, we find -tO on sdi tâl
‘skunk’ and idp k!Oâl ‘dish’; -∅ on sdi t!áp ‘deer’, ids k!óp ‘mountain’ and
idi k!yagóp ‘brain’; -dO on idi k!Ôn ‘tomato’, sdi sân ‘child’, and idp són
ˇŹˇ ‘white faced cattle’, idp mÓx!á
‘grass’; -mO on sdi t!óút!á ˇŹˇ ‘paper’, ids t!á
ˇŹˇ
‘sheet’; -op is sdi sǎǎné ‘snake’ and idp mÓnsóúde ‘bracelet’; and so on.
Even the competing -gu∼-̂i do not define classes, as, by lexical coincidence,
56 CHAPTER 2
they are almost entirely confined to animates, which are classmate.24

Particularly striking in this regard are words that serve as the base for
compounds or metaphorical extensions of meaning. Though the phonological
shape that conditions the inverse allophone is identical, the class membership
varies according to semantic principles. So, t!ÓlthOn is sdi as ‘kidney’ but
idp as ‘bean’; thó ˇt!ólǑǑ is sss as ‘soda, fruit juice’ but idi as ‘orange’;
áá is idp as ‘stick, pole’ but ids as ‘tree’, and it forms the idp compound
khẐˇǐgulaa ‘redbud’ and ids zépgutk!oaa ‘Osage orange’; t!á ˇŹˇ ‘white’ forms
sdi t!á ˇŹˇ ‘egg’ and t!óút!á
ˇŹˇ ‘white faced cattle’ and ids t!á
ˇŹˇ ‘sheet’; ppp
hÓˇÓˇgya ‘metal’ forms ids hÓˇÓˇgya ‘gun’ and sss ÓlhÓˇÓˇgya ‘money’.
Similarly, the inverse form of noun+adjective compounds is determined
by the phonology of the adjective, to which the inverse suffix attaches, and
not by the noun: hence, in k!óp ‘mountain’, the inverse suffix is -∅, but in
K!ópéttO ‘Mount Scott’ [lit.: big mountain], it is -tO, attaching to êl ‘big.s’;
compare also the inverse markers of áákyǒǐmO ‘long pole’ and áádO ‘pole’.
Consequently, no form of the inverse marker can be regarded as consti-
tuting the phonological signature of any particular class. This means that
affixal phonology does not define classhood, as it does in other languages.
2.6.2. Thematic nouns
In Indo-European, nouns with the same ending frequently belong to the

same gender class. Many Kiowa nouns end in the ‘themes’ -gya, -de and
-ba, which might be reminiscent of gender-specific noun endings in Romance.
However, a noun’s number properties correlate with the theme it takes, sug-
gesting that phonology is again irrelevant. Moreover, these number prop-
erties are not always the same as the semantic principles that determine
membership of the classes discussed above. Where they conflict, the seman-
tic principles are decisive, not the theme.
Themes
I begin with morphological and semantic properties of the themes.25

24
Note though khÓŹgú ‘cloth.inv’ from khÓŹ; and Onsôi ‘foot.inv’ from Onsóú, OOzâi
‘udder.inv’ from OOzáá, and sékhôˇ ǐ ‘large intestine’ from sékhóˇú
ˇ , by apparent overap-
plication of -̂i.
25
The discussion follows Watkins’ Section 3.131, though aspects of the morphological
decomposition are my own.
Kiowa has a variety of nominal roots that cannot occur independently.

They must be suffixed, for instance, by -gya ‘in’ (91), or incorporated (92).
(91) Táá-gya hÓndé éM- xél

eye- in something :1s:3s-be set.s/d
‘There is something in my eye’ (Watkins 1984, p. 93)
(92) tá- hot- guM uM
eye-travel-distr
‘watch everywhere’ (ibid.)
When without a postposition or when unincorporated, thematic nouns end

in a special -CV(C) suffix, the ‘theme’; táá- ‘eye’ takes -de for referential
cardinalities 1 and 2, and -gO for 3. (Agreement correlates -gO with inverse.)
The themes -de∼-gO occur with:
(93) Gloss de-Form

ear t!ÓÓdé
eye tááde
half zâide
horn (of cow) gúM úM dé
leg thóMúM dé
shoe tóúdé
trousers khÓÓdé
These nouns naturally occur in pairs. Interestingly, e is a frequent exponent

of d-agreement in the verbal prefixes (Chapter 5), and O, of i-agreement:
(94) né = :1s:3d; ménêi∗ = 2d:1s:3d; . . .

nÓ = :1s:3i; ménÔO∗ = 2d:1s:3i; . . .
This suggests that these thematic nouns have e when non-inverse because
they denote inherent pairs. Consequently, the vowel appears to be inde-
pendent from the consonant in the thematic CV suffix and is connected to
number. The consonant may be supposed to be g: g+O trivially yields gO
and g+e yields de by regular Kiowa phonology.26
The other themes support this decomposition. Consider -gya:
26
Dentals become velar before i/y and velars dental before e. See Halle (2005) for a
recent treatment.
58 CHAPTER 2
(95) Gloss gya-Form

beads póMúM gya
cantaloupes phŹdátkyá
downy feathers x!óMúM gya
fabric, dry goods phátkyá
sand péŹgya
seeds dóMúM gya
shirt t!óMúM gya
These form their inverse in -gOt. Decomposing onset and rhyme yields -g-
and -ya / -Ot. Again, the rhymes are recognizable from the verbal prefixes:
(96) gyát = :1d/p:3p; gyát = 3s:2s:3p; . . .

dÓt = :1d/p:3i; gÓt = 3s:2s:3i; . . .
The theme’s non-inverse form correlates with p-agreement morphology and

the nouns in (95) naturally occur in groups or are pluralia tantum.
Similar observations can be made of the handful of nouns that take the
theme -ba. Decomposing, we have the theme -b- and the plural-associated
rhymes -ya / -Ot.27 Again, the relevant nouns plausibly occur in groups.
(97) Gloss ba-Form

arrows zêiba (inv = zéŹbOt)
reed pipes tóMúM bá
tobacco tháábá
The resulting picture of Kiowa thematic nouns is that they consist of a
root followed by a thematic consonant, b or g, followed by a number-related
coda, ya/Ot or e/O.
Themes and classes
The relationship between themes and the meaning of the nouns to which
they attach shows that themes are not indicative of phonological noun clas-
⎡ ⎤⎡ ⎤
+consonantal −consonantal
(i) [−αback] / ⎣−sonorant ⎦ ⎣+sonorant ⎦
[αback] →
αback
27
The combination b+y standardly simplifies to b; for instance compare (96)’s gyát,
:1p:3p, from d+y. . . , with bát, :2p:3p, from b+y. . . .
sification. The ‘pairhood’ that conditions e/O is not a conditioning factor for
noun class membership. Such cases might open the door to phonologically
determined class membership, but instead underline the semantic basis of the
system. In such cases, the semantic principles of each noun class determines
class membership: of the nouns in -de, ‘eyes’, ‘ears’, ‘legs’ and ‘horns’ are
sdi, ‘trousers’ ppp, and ‘shoes’ sdp; of the nouns in -gya, ‘shirts’ and ‘sand’
are ppp, ‘beads’, ‘cantaloupes’ idp, and ‘guns’ ids. Indeed, two words in this
last group for etymological reasons, iip!ÓÓgya ‘baby’ and bélkŹtkya ‘screech
owl’, are animates and, so, sdi. Thus, nouns that share thematic endings fall
into different noun classes and do so on purely semantic grounds.
2.6.3. Conclusion
Kiowa noun classes are not phonologically defined. Rather, the semantic
classification in (77) stands.
This page intentionally blank
Chapter 3
Number Features
The previous chapter showed that nouns in Kiowa are organized into nine
number-related classes, according to semantic properties of their referents.
Such a system poses several questions:
(1) a. What are the features that do the classifying?

b. How do classifying features interact with referential cardinality
to yield inverse marking on the noun?
c. How do the classifying features and referential cardinality inter-
act with the syntactic mechanisms responsible for agreement to
create patterns like sdp, idi, ppp, et cetera?
d. What is the relationship between a class mnemonic and the prop-
erties of the nouns it subsumes? (Is it coincidence that animates
are sdi and hair types idi and not vice versa?)
e. Why are there nine classes and why no did, sip, or psi?
These questions are the focus of this chapter. Answering them consists in
providing an inventory of number features, properly semantically defined,
together with a theory of their distribution and interaction in the syntax and
their treatment in the morphology. To do this, I make two claims.
Claim #1. There are three number features, motivated on seman-
tic grounds: [±singular] and [±augmented] by referential cardinality (Noyer
1992; also Conklin 1962, Matthews 1972, McKay 1978, 1979; cf.,
Krifka 1992) and, later, by properties of noun classes, [±group] purely by
properties of noun classes (cf., Lasersohn 1995, Winter 2001, Krifka 1992).
Thus, all three features determine noun class and the first two additionally
61
62 CHAPTER 3
determine referential cardinality; the first two, further, are phonologically

realized as inverse marking on the noun and agreement on the verb.
Claim #2. DPs have the structure below (cf., Carstens 1991, Ritter
1991, 1993), where Class is the locus of classificatory features, Number the
locus of referential cardinality, and D the locus of the features that trigger
agreement and inverse marking.
(2) DP
NumberP D
Class: Noun Number
Given Claim #1, this means that features are distributed as in (3).
(3) DP
NumberP D
±singular
±augmented
(±group)
The features in (3) are not all of the same status. Those on Class and Num-
ber are intrinsic to the meaning of the head, that is, they are interpretable
in the sense of Chomsky (2000, 2001). The number features on D are not
intrinsic to its meaning. When merged, D bears uninterpretable number,
notated [usingular uaugmented], which must be valued. Valuation (Section
3.2.2 and following) involves matching the feature content of D with that of
Class and Number. In a departure from other work, such as Chomsky (op.
cit.), uninterpretability is argued to involve overspecification, rather than
underspecification, of the feature. That is, [uF] is argued to abbreviate spec-
ification of both [−F] and [+F], which must be matched with interpretable
counterparts or else deleted, rather than specification of a valueless [ F].
This view permits valuation of D to produce the feature conflicts, [−F +F],
on which the analysis of the inverse hinges.
NUMBER FEATURES 63
(4) DP
NumberP D
u singular
u augmented
(±group)
Verbal agreement is with D, once valued; the agreeing heads do not enter into
a direct relationship with Number or Class. The evidence for the content of
Class and of Number is semantic. Evidence for the feature content of D
is morphosyntactic and is discussed below, primarily in relation to inverse
marking on nouns, and in Chapter 5.
I will show that the features’ semantics explain the relationship of class
characteristics to class mnemonics and that Kiowa’s class system makes near
optimal use of the inventory of classificatory features at its disposal.
Claim #1 is particularly important in relation to the current work’s aim
of establishing the core elements of a morphosemantic theory of number.
Specifically, as the references cited on p. 61 show, the features argued for
below have precedents both in morphological and semantic work. However,
what is particularly striking is the precise formulation of the features (the
significance of the formal definitions is discussed below):
(5) Feature definitions x, a free variable

a. [+singular] = atom(x)
b. [+augmented] = λP∃y[P(x) ∧ P(y) ∧ y x]
c. [+group] = λP[+augmented] ∃Q[P(x) ∧ P = Q ∧ Q-atom(x)]
(5a−b), although based on previous morphological work (especially Noyer

1992), are in fact variants of definitions proposed by Krifka (1992) in his
treatment of aspect and Aktionsart. Their formal equivalence is demon-
strated in Harbour (2006b). Furthermore, the notion of Q-atomicity, crucial
to (5c), is also in common with Krifka. This convergence complements, and
concretizes, the purely architectural argument of Chapter 1 by showing that,
not merely is a morphosemantic theory necessary, but that its core elements
have begun to emerge in the seemingly disjoint work of both subfields.
64 CHAPTER 3
Structure of the chapter
This chapter contains the core ideas of the current study: proposals concern-
ing the feature inventory of Universal Grammar, these features’ distribution
and operations affecting them. In consequence, the chapter is rather full.
Readers are, therefore, advised to return to this section if, at any time, they
find the progress of the argument to be obscured by its detail.
The chapter falls into three parts. The first, Sections 3.1−3.3, concerns
the features [±singular] and [±augmented] and their associated classes. The
second, Sections 3.4−3.5, concerns the feature [±group] and its associated
classes. The third, Section 3.7, concerns the totality of possible classes.
The first part of the chapter is structured as follows. Section 3.1 argues
for two features, [±singular] and [±augmented], on the basis of referential
cardinalities and the natural classes that they form. Proofs relating refer-
ential cardinalities to particular feature-value combinations are presented in
Appendix 3.8.
Section 3.2 comprises several subsections. Section 3.2.1 shows that Kiowa
nouns bring with them into the syntax different specifications for the features
[±singular] and [±augmented] and that noun classification consists in such
feature-value specifications. The content of Class is specified, in terms of
[±singular] and [±augmented], for the classes sii, sdi, idp, idi, and sdp
(and, partially, for ids).
Section 3.2.2 proposes a simple principle by which referential cardinality
and nouns’ inherent number specifications interact to yield inverse marking
on the noun. The basis of this claim is that D is valued by a computation
over the features located at Class and Number and that this computation
leads to inverse marking / i-agreement when Class and Number have opposite
specifications for a single feature.
Finally, Sections 3.2.3−3.2.4 present further derivations. Examples of
each of the classes discussed in Section 3.2 are given, and it is shown how
positing an sii class for first person derives syncretic properties of first person
agreement. DPs modified by adjectives, demonstratives, relative clauses and
‘only’ are also derived. (Readers may prefer to skip these technical excursions
on early readings of the chapter.)
Section 3.3 argues that the relationship between a noun class’ mnemonic
and the semantic properties of the nouns it subsumes is non-arbitrary. The
basis of this explanation is the conclusion that mnemonics are not primi-
tive, but result from nouns’ Class specification. Thus, the question is not
NUMBER FEATURES 65
how class characteristics relate to mnemonics, but how they relate to Class.
Since the features at Class have by this stage been rigorously defined, the
relationship between Class and nouns’ semantic properties are readily expli-
cable. Consequently, we are able to say why animates are sdi and hair types
idi, not vice versa. This concludes the first part of the chapter.
The second part of the chapter concerns s-agreement that occurs when
referential cardinality is not 1 and p-agreement that occurs when referential
cardinality is not 3. Section 3.4.1 argues that these are complementary phe-
nomena, to be implemented as different values of a single feature. The feature
[±group] is then rigorously defined and argued to predicate of [+augmented].
The Class specifications for sds, ids, and ppp (and for idi on its ‘different
types of’ reading) are then given.
Section 3.4.2 proposes how [±group] affects the valuation of D. Deriva-
tions of various referential cardinalities of sds, ids, ppp, and idi nouns are
given. The section also explains the mnemonic naturalness of sds, ids, ppp.
Finally, Section 3.5 analyzes the specification of Class for mass nouns,
arguing that these crucially involve the feature [±group]. This section also
returns to the topic of agreement triggered by conjunctions, discussed in
Section 2.3.8, where mass nouns and their mnemonics were introduced. This
concludes the second part of the chapter.
The third part of the chapter asks why Kiowa has only nine of 64 mnemon-
ically possible classes. It argues that Kiowa almost optimally exploits the
space of possible mnemonics generated by the features at its disposal.
3.1. Referential cardinality
The main claim of this section is that 1, 2 and 3, the values of referen-
tial cardinality, are not atomic. Rather, they are composed of two fea-
tures, [±singular] and [±augmented], to be defined below. The evidence
for this claim comes from the distribution of i-agreement and inverse mark-
ing. (This section recapitulates, in part, arguments from Noyer 1992 and
Harbour 2003a. The analysis of cardinality is essentially Noyer’s.)
3.1.1. Natural classes
The values 1, 2 and 3 tempt one to posit three monovalent number features,
[singular], [dual], [plural]. However, such privative features predict that there
66 CHAPTER 3
are no natural classes of number features, that no process of the grammar

refers both to 1 and 2 or to 2 and 3. However, such processes do exist, both
in the morphology (Chapter 5) and in the syntax, as inverse marking shows:
Recall that there are four agreement types in Kiowa: s, d, p, i. i-
agreement always cooccurs with inverse marking on the noun. This is exem-
plified below for the sdi noun tógúl ‘young man’, for the idp noun kútaa
‘pencil’ and the idi noun k!Ôn ‘tomato’.

young man.INV 3I-be
(7) KútaadO e- dÓÓ
pencil.INV 3I-be
‘It’s a pencil’
tomato.INV 3I-be
‘It’s a tomato’ or ‘They’re tomatoes’
As the examples of the previous chapter show, there is no noun marking

concomitant with and proprietary to any of the other agreement types.
A natural question, then, is what inverse marking means. However, class
mnemonics cast doubt on the coherence of this question, as i-agreement oc-
curs with referential cardinalities 1 (idp), 2 (sii), and 3 (sdi).1 Moreover,
if we examine the sets of referential cardinalities for which inverse marking
occurs, the result is not enlightening: {1} (idp/ids), {3} (sdi), {1, 3} (idi),
and {2, 3} (sii). Nothing seems to unite the classes beyond idi’s being the
union of the first two.2
By contrast, a quite neat picture emerges if we ask what non-inverse
marked members from each class mean. This corresponds to the natural
assumption that the unsuffixed form of the noun is basic. The sets of ref-
1
Other agreement types do not pose this problem. d-agreement occurs exclusively with
2. s-agreement occurs primarily with 1, the exception being objects that are group-like
and hence 1-like. And p-agreement occurs primarily with 3, the exception being objects
that are always composed of parts, such as pluralia tantum nouns, which are 3-like.
2
Strictly, one can posit [±F], where [+F] := {1}, and [±G], where [+G] := {2}. The
required classes follow: [−F] = {2, 3}, [−G] = {1, 2}, [−F −G] = {3}. However, when we
come to consider the relationship between the nouns’ semantic properties and their class
mnemonic, it will become apparent that these are not the desired features.
NUMBER FEATURES 67
erential cardinalities that emerge are {1} (sii), {2} (idi), {1, 2} (sdi), and
{2, 3} (idp/ids). These classes follow if referential cardinalities are taken to
be composed as below:3
(9) [±singular] [±augmented]

1 + −
2 − −
3 − +
{1, 2} is the natural class defined by [−augmented]; {2, 3}, the natural class
defined by [−singular]; {1}, [+singular]; and {2}, [−singular −augmented].
For this feature composition to be more than a formal nicety, [±singular]
and [±augmented] must be properly defined and their connection to inverse
marking explained. Answering these questions lays the foundations for ex-
plaining the relationship between nouns’ semantic properties and their class
mnemonic, e.g., why hair types are idi and animates, sdi and not vice versa.
3.1.2. Definitions
I assume a general lattice-theoretic semantics, following Link (1983), accord-

ing to which the denotation of nominal predicate is a set of atoms and the
set of subsets of the atoms. Technically, this constitutes a join-complete
semi-lattice. It is graphically represented as:
. . . . . .
s t u v w x
The bottom stratum represents the atoms. If the lattice corresponds to the
nominal predicate chair(x), then s, t, and so on are individual chairs; if the
predicate is person(x), then they are individual persons; and so on. The
points of intersection of lines emanating from the atoms represent sets of
atoms. So, the point between s and t represents {s, t}, the point directly
3
The combination [+singular +augmented] corresponds to no referential cardinality.
See Appendix 3.8.
68 CHAPTER 3
above t, {s, t, u}.4 More generally, the lowest row of points of intersection
represents dyads, above that are triads, and so on.
The cardinality features [±singular] and [±augmented] apply to the lat-
tice, partitioning it into [−singular] and [+singular], and [−augmented] and
[+augmented], regions. As a convention, I will define only [+F].5 Then:
(10) Feature negation

[−F] = ¬[+F]
The definitions build on previous morphological work (especially Conklin

1962, Matthews 1972, McKay 1978, 1979, Noyer 1992). However, in their
current form, they most closely resemble definitions from Krifka (1992). The
surprising finding that morphologists concerned with pronoun and agreement
categories and semanticists concerned with aspect and Aktionsart have con-
verged only logical equivalent discoveries is reported in Harbour (2006b).
It strikes me as highly significant for an incipient joint theory of morphose-
mantic number.
The feature definitions are:
(11) Definition: [±singular]

[+singular] = atom(x) x, a free variable
Correspondingly, [−singular] = ¬atom(x). So, the feature partitions the

lattice into its atomic and non-atomic regions.
4
The lattice is simplified in that not all such sets (e.g., {s, u}) are represented. The
simplification aids readability.
5
The convention should not be mistaken for a claim that there is a universal correlation
between ‘+’ and markedness. What the marked value of the feature is, is an empirical
matter. For instance, based on p-agreement’s being the default agreement in Kiowa, it
appears that minus is the unmarked value of [±singular] but that plus is the default
value of [±augmented] (Harbour 2003a). Now, one can transform a feature inventory
into one with a uniform representation of markedness as plus, by negating the definition
and switching the sign of any feature with the marked value minus (Harbour 2003b, p.
135). However, this is possible only if marked values are invariant language internally and
crosslinguistically; otherwise ‘the marked value of [±F]’ and ‘[+F]’ are non-equivalent, no
matter how we redefine or switch signs, as the former is variable, the latter not.
NUMBER FEATURES 69
[−singular]
. . . . . .
[+singular]
The definition of [±augmented] is slightly more complex:
(12) Definition: [±augmented]

[+augmented] = λP∃y[P(x) ∧ P(y) ∧ y x] x, a free variable
That is, [+augmented] guarantees that whatever properties hold of x hold

also of some proper subset of x. By contrast, [−augmented] guarantees that
no such subset exists, that x is a minimal element satisfying P. Note that
join-complete semi-lattices lack bottom elements (⊥), the lattice-theoretic
equivalent of the emptyset. So, the emptyset is never a possible value for x
in [−augmented]. I note this now and omit reference to the condition below.
(13) x = ∅
Therefore, for common nouns, such as chair(x), [±augmented], if acting in

isolation, induces the same partition as [±singular].
[+augmented]
. . . . . .
[−augmented]
However, the features are non-equivalent in at least three contexts. The

first, not relevant here, concerns personal pronouns and minimal∼augmented
versus singular∼plural systems (such as Ilocano versus English; see Noyer
1992 or Harbour 2006b). The second, also irrelevant here, concerns their
use in Krifka’s treatment of Aktionsart. And the third, of major importance
here, concerns the interpretation of feature bundles in which both features
occur, as already intimated in (9). When the features cooccur, they are
interpreted by function application:6
6
Composition in the other order is possible, but the learner would never have semantic
evidence for it, as the resulting system is identical to one in which [±augmented] is the
only number feature (Harbour 2006a).
70 CHAPTER 3
(14) Semantic composition

[±singular ±augmented] = [±augmented]([±singular])
We can now justify (9) intuitively, as follows. An atom is clearly singu-

lar and, as a singleton, lacks proper subsets. So, the singular is [+singular
−augmented]. On the other hand, pluralities are clearly non-atomic, and,
so, [−singular]; and any plurality like {s, t, u, v, w} has a proper subset,
{s, t, u, v} for instance, that is also non-atomic, and, so, is [+augmented].
So, the plural is [−singular +augmented]. The limiting case, though, are
dyads. Clearly, {s, t} is non-atomic and, so, [−singular]. However, its proper
subsets, {s} and {t}, are not non-atomic. Thus, dyads are [−augmented],
and, so, the dual is [−singular −augmented]. For more formal demonstra-
tions, see Appendix 3.8. The relevant lattice is provided below.
[−singular +augmented]
[−singular −augmented]
. . . . . . [+singular −augmented]
Summary
The referential cardinalities 1, 2 and 3 have been shown not to be primitive.

Rather they are composed, as shown in (9), by the features [±singular] and
[±augmented], defined in (11) and (12). Adopting the proposal that Number
constitutes a separate projection within the DP (Carstens 1991, Ritter 1991;
see also the suggestive discussion in Dryer 2005), we have arrived at:
(15) DP
NumberP D
Noun Number
±singular
±augmented
NUMBER FEATURES 71
3.2. Class
We are now well placed to answer, in part, several of the questions in (1).
With respect to the classes sii, sdi, idp, idi, and, to some extent, ids, we
can say what the classifying features are. A simple hypothesis about the
nature of inverse marking further enables us to say how the inverse arises.
Furthermore, we can explain the relationship between the mnemonics of the
classes just listed and the semantic properties of the nouns they subsume.
3.2.1. Classification by cardinality features
Let us begin by taking stock of the noun class mnemonics as a whole. There
is a broad correlation between referential cardinality and agreement type: 1
with s-agreement, 2 with d-agreement, 3 with p-agreement. Therefore, we
can extend the correlation between features and referential cardinalities in
(9) to include also agreement types:
(16) Referential Features Agreement

Cardinality [±singular] [±augmented] Type
1 + − s
2 − − d
3 − + p
So, sdp nouns, which exhibit no robust, unifying semantic commonalities,

trigger agreement as expected on the basis of referential cardinality.
Other mnemonics diverge from this correlation in two ways. First, some
have i, where s, d, or p is expected. For instance, replacing p by i in sdp
gives sdi; replacing s gives idp; replacing both gives idi. Second, some, like
ppp, exhibit p where s or d are expected, or, like sds, s where d or p are.
We return to unexpected s/p in Section 3.4 and focus now on i.
An ideal explanation of the mnemonics sii, sdi, idp and idi would make
the occurrence of s, d and p as straightforward as possible and would sub-
sume all occurrences of i under a single principle. This is what we shall now
do, by means of the features [±singular] and [±augmented].
As emphasized in the preceding section, the referential cardinalities where
agreement is as expected form natural classes with respect to [±singular] and
[±augmented]. The table below expresses when s, d and p occur as expected,
given (16); the middle column gives the features that must occur under Num-
72 CHAPTER 3
ber and the right column gives the corresponding agreement type(s).
(17) Class Feature Agreement Type

sii [+singular] s
sdi [−augmented] s/d
idp [−singular] d/p
idi [−singular −augmented] d
For instance, for sii, the feature [+singular] must occur under Number. This
happens only with [+singular −augmented], which, according to (16), yields
s-agreement for referential cardinality 1. Consequently, s appears in the first
mnemonic position. For idp, the feature [−singular] must occur under Num-
ber. This happens with [−singular −augmented] / [−singular +augmented],
which, according to (16), yields d-agreement for 2 / p-agreement for 3. Con-
sequently, d/p appear in the second/third mnemonic positions.
I suggest that we take (17) as representing the classificatory features of
each of the classes shown. By so doing, we can state a simple generaliza-
tion about the expected occurrence of s, d and p and another concerning
occurrence of i, for these classes.
(18) Generalization: S, D, P
s, d, p occur when Class ⊆ Number.
(19) Generalization: I
i occurs when Class ⊆
/ Number.
The first states that s/d/p occur when the features that make up referential
cardinality subsume the class features. The second states that, if a class
feature is not also a referential cardinality feature, then i occurs.
Note that these generalizations apply also to sdp and, partially, to ids.
Consider the classificatory features below:
(20) Class Feature

sdp ∅
ids [−singular . . . ]
If Class is empty for sdp, then it is trivially a subset of Number. Conse-

quently, i never occurs. If ids is [−singular], then i will arise for referential
cardinality in the first mnemonic position. (Neither principle accounts for s
in third position and so Class is at present incomplete; hence, the ‘. . . ’.)
NUMBER FEATURES 73
So far, all generalizations have been stated at the level of the mnemonic:
“A mnemonic has i in such-and-such a position if so-and-so holds”. Such
generalizations are superficial, as the mnemonics are mere expository devices.
We must now ask what mechanisms and structures underlie them.
3.2.2. ClassP and number on D
I now present a syntactic structure and a computational mechanism that

produces s, d, p and i in accord with the generalizations above. The broad
view of syntax adopted is that of Chomsky (2000, 2001), which has been
applied in detail to Kiowa by Adger and Harbour (2007). Other theoretical
assumptions concern the location of class features, to which I now turn.
Following ideas of Marantz (1997), Kihm (2002) and Ouhalla (2005) (see
also Ritter 1993, Kihm 2005), I propose that classificatory features occupy
their own projection, which I label Class. Syntactically, Class serves to make
the featureless root nominal and visible to the computational system. Like
Number, it is a locus of the features [±singular] and [±augmented].
(21) DP
NumberP D
Class: Noun Number

(±augmented) ±augmented
To address questions of agreement, I adopt the framework of Chomsky

(2000, 2001; see Adger and Harbour 2007 for application to Kiowa). Verbal
agreement is the result of an Agree (feature matching) relation between heads
in the extended verbal projection and the D head of the verbal arguments.
As agreement is number sensitive, D must bear number features. However,
semantically, D is the locus of definiteness, to which number is irrelevant.
So, number on D is misplaced, or, technically, uninterpretable.
74 CHAPTER 3
(22) DP
NumberP D
u singular
u augmented
Class: Noun Number
The eventual specification of uninterpretable features depends on those heads

for which the features are interpretable. In the current case, the number
content of D depends on Class and/or Number. Now, if Number alone valued
D, all nouns would be sdp, and if Class alone did, agreement would be
invariant across referential cardinality. So, D crucially depends on both.
This is possible given that Class and Number are in a mutual dominance
relation and, so, equidistant from D. The mechanism of valuation involves
replication of the content of Class and of Number on D.
(23) Valuation of D
Uninterpretable number on D is valued by a computation over Num-
ber and Class. The features of both are replicated on D.
To make this concrete, consider an sdi noun of referential cardinality 1:

[−augmented] is located at Class and [+singular −augmented] at Number.
(24) DP
NumberP D
u singular
u augmented
Class: Noun Number

−augmented +singular
−augmented
Valuation is straightforward, as Class is a subset of Number:

NUMBER FEATURES 75
(25) DP
NumberP D
+singular
−augmented
Class: Noun Number

−augmented +singular
−augmented
Bearing [+singular −augmented], D will trigger s-agreement (16).

Exactly the same mechanisms account for an sdi noun’s d-agreement
when referential cardinality is 2. Class is [−augmented], as before, and Num-
ber is [−singular −augmented]. Prior to valuation, we have:
(26) DP
NumberP D
u singular
u augmented
Class: Noun Number

−augmented −singular
−augmented
The content of Class and Number is replicated on D, yielding:
(27) DP
NumberP D
−singular
−augmented
Class: Noun Number

−augmented
Again, by (16), this is the correct feature content to trigger d-agreement.

Matters are more interesting for sdi nouns of referential cardinality 3,
where, in contrast to the preceding cases, Class is not a subset of Number:
76 CHAPTER 3
(28) DP
NumberP D
u singular
u augmented
Class: Noun Number

+augmented
Copying yields opposing specifications on D: it is both [+augmented] and

[−augmented].
(29) DP
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
Class: Noun Number +augmented

+augmented
Although surprising at first glance, I suggest that such feature conflicts

are not only possible, but are the precise configuration behind the inverse:
(30) Kiowa inverse

[−F +F] ⇔ i/inv
That is, inverse marking on nouns and i-agreement on verbs is the vocabularic
reflex of feature conflict (abstracting away from actual phonological content).
Now, this requires slight revision of the standard notion of uninterpretabil-
ity. In general, [uF] is taken to be [±F] without a value, that is, [ F], an
underspecified version of [±F]. However, if valuation involves simply ‘filling
in the blank’, then it is impossible to value [uF] as [−F +F]. Introduction of
an extra [αF] violates Inclusivity. Instead, I propose that uninterpretability
is overspecification: [uF] means [−F +F]. Valuation involves pairing these
features with interpretable counterparts. Only if paired in this way do the
features survive at the interface, to be interpreted and pronounced. Depend-
ing on one’s view of the interfaces, uninterpretable features that have not
been paired with an interpretable counterpart must either be deleted, or else
NUMBER FEATURES 77
they are not visible and, so, are ignored.7

So, D in (22) bears [−singular +singular −augmented +augmented]. In
(26), the interpretable number features are [−singular] and [−augmented].
So, the plus values are invisible/deleted and only the minus ones survive to
be pronounced as agreement. In (29), all but [+singular] survive, creating a
feature conflict, which, by (30), is pronounced as inverse marking and goes
on to trigger i-agreement. Taking sdi ÓÓpŹˇŹˇ ‘fish’ as the root noun, the plural
DP is pronounced as:
(31) DP
NumberP D

dÓ
Class: fish Number

ÓÓpŹˇŹˇ ∅
The nature of uninterpretability and valuation is not intended to be a

major focus of the current investigation. However, two observations are
worthwhile. First, there is independent, semantic, evidence for the reality of
conflicting feature specifications. Harbour (2006a), a thoroughgoing crosslin-
guistic investigation of number categories and number features, proves that
the most complex categories, such as unit augmented, minor paucal and trial,
result from multiple specification of the feature [±augmented]. For instance,
the trial is [−singular −augmented +augmented], subject to (14) and (32).
(32) [±singular −augmented +augmented]

= [−augmented]([±singular +augmented])
It is only possible for categories so defined to trigger agreement if feature

conflicts are generally possible on agreeing heads. Second, default agreement,
which is problematic for the underspecification approach, is trivial for the
overspecification approach. Default agreement is whatever agreement type
arises when there is nothing to agree with. Given underspecification, failure
to establish an Agree relation means that [uF] receives no value. Rather
than triggering default agreement, this should cause the derivation to crash.
7
The general view is the former, that everything must be legible at the interface;
illegibility causes crashes. However, this seems to me to be an unnecessary assumption: the
interfaces ‘ignores’ what it cannot read, in the way that the human eye ‘ignores’ infrared.
78 CHAPTER 3
But, with overspecification, if no Agree relation is established, both [−F]

and [+F] delete (or are ignored) and only the host head is pronounceable,
yielding featureless, hence default, agreement.
So, the DP structure (22), the overspecification approach to uninter-
pretability and the feature conflict view of the inverse (30) constitute an
account of the mechanisms of inverse marking and i-agreement. Furthermore,
(25), (26) and (28), show that, if Class is [−singular], then s-agreement re-
sults for referential cardinality 1, d-agreement for 2, and i-agreement for
3. That is, they derive the sdi mnemonic from the Class specification
[−augmented]. Given (17) and (20), we can derive several other mnemonics,
as the coming section shows.
3.2.3. Derivations I: Mnemonics
This section applies the mechanisms of Section 3.2.2 to the Class features of
Section 3.2.1 to derive other class mnemonics:
(33) Class Feature

sii [+singular]
sdi [−augmented]
idp [−singular]
idi [−singular −augmented]
sdp ∅
(Details of the derivations for sdp, idp and idi, though not sii, are much
the same as for sdi. Readers satisfied with the latter may wish to skip the
former. The reader may refer back to the example sentences in Section 2.3.)
The SDP class
The sdp class is the simplest case. Only Number bears interpretable number
features. Class bears only categorial features (Kihm 2002). Consequently,
the content of Number is always replicated on D and agreement reflects
referential cardinality. The general case of a fully valued DP is shown below.
NUMBER FEATURES 79
(34) DP
NumberP D
αsingular
βaugmented
Class: Noun Number

∅ αsingular
βaugmented
For s-agreement (1), α is plus and β is minus; for d-agreement (2), α and β
are both minus; for p-agreement (3), α is minus and β is plus.
The IDP class
Here, Class is [−singular]. For referential cardinality 1, Number is [+singular

−augmented]:
(35) DP
NumberP D
u singular
u augmented
Class: Noun Number

−singular +singular
−augmented
Class and Number oppose for [±singular], creating feature conflict on D:
(36) DP
NumberP ⎡ D ⎤
−singular
⎣+singular ⎦
Class: Noun Number −augmented

−augmented
The result is an inverse-marked noun and i-agreement.

80 CHAPTER 3
For 2, no such incompatibility arises. Number, [−singular −augmented],

subsumes Class and no conflict arises on D. The valued DP is:
(37) DP
NumberP D
−singular
−augmented
Class: Noun Number

−singular −singular
−augmented
The noun will not be inverse marked and will trigger d-agreement. (Note
that (36) and (37) also derive agreement and inverse marking for ids nouns;
(38), however, does not represent ids nouns of referential cardinality 3.)
For 3, Number, [−singular +augmented], again, subsumes Class:
(38) DP
NumberP D
−singular
+augmented
Class: Noun Number

+augmented
The noun will not be inverse marked and will trigger p-agreement. This
derives the idp class.
The IDI class
Here, Class is [−singular −augmented]. D, therefore, avoids conflicting spec-

ifications only when Number equals class:
NUMBER FEATURES 81
(39) DP
NumberP D
−singular
−augmented
Class: Noun Number
−augmented −augmented
Such DPs trigger d-agreement, and so only for 2 do nouns of this class trigger
agreement that reflects referential cardinality.
For 1 (40), Class and Number oppose for [±singular], and, for 3 (41),
they oppose for [±augmented].
(40) DP
NumberP ⎡ D ⎤
−singular
⎣+singular ⎦
Class: Noun Number −augmented
(41) DP
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
Class: Noun Number +augmented
−augmented +augmented
Given the conflicts, both nouns will be inverse marked and trigger i-agreement.
This derives the idi class.
The SII class: Justification by syncretism
As mentioned in Section 2.3.9, the sii class is less readily observable than the
others, being bound up with person exponence. However, its sole member,
82 CHAPTER 3
the first person, exhibits four distinct sets of syncretisms and it is in the
explanation of these syncretisms that positing an sii class yields dividends.
Syncretism #1. For agents of (di)transitives and the subjects of un-
accusatives, the first person exclusive dual and first person exclusive plural
trigger i-agreement.
(42) NÓÓ a- dÓÓ

1 1S-be
‘It’s me’
(43) NÓÓ e- dÓÓ
1 1I-be
‘It’s me and him or me and her’
(44) NÓÓ e- dÓÓ
1 1I-be
‘It’s me and them’
Compare (43) and (44) with (45) which has i-agreement in virtue of the
inverse-marked tógúúdÓ ‘young men’.8

young man.inv 3I-be
i-agreement for referential cardinalities 2, 3 immediately suggests an sii class.

Syncretism #2. For agents of (di)transitives and the subjects of un-
accusatives, the first person inclusive dual and first person inclusive plural
trigger the same agreement as second person plural. That is, the same pat-
tern as #1, but syncretic with second, not third, person. This agreement is
notated 2i, reflecting the assignment of second person to the animate, sdi
class.
(46) NÓÓ ba-dÓÓ

1 2I- be
‘It’s me and you.s’
(47) NÓÓ ba-dÓÓ
1 2I- be
8
d-agreement is always nasal in Kiowa, unless a structurally higher position is non-
singular (see Harbour 2003a, or Chapter 5, for details). Consequently, the lack of nasality
in (43) and (46) is genuinely noteworthy and stands in need of explanation.
NUMBER FEATURES 83
‘It’s me and you.d/p’
Compare (46) and (47) with second person plural agreement in (48). Note
that (48) does not have a dual reading; the dual would have the prefix ma-.
(48) Ám ba-dÓÓ

2 2I- be
‘It’s you.p’
Syncretism #3. For indirect and direct objects, the first person in-
clusive dual, the first person inclusive plural, the first person exclusive dual,
and the first person inclusive plural all syncretize. Moreover, these forms are
distinct from the inverse and second plural as well as from other persons.9
(49) DÓ- bóM úM

3p:1d/p-see.pf
‘They saw me and you or me and him or me and them’
(50) BáougO dÓt- ÓMÓM
cat.inv 3p:1d/p:3i-give.pf
‘They gave cats to me and you or me and him or me and them’
Syncretism #4. First person agreement never distinguishes dual from

plural. This can be verified by examining Syncretisms #1−3. Nonetheless,
the difference is syntactically represented and is evident in the suppletion
they condition in number-sensitive predicates (Chapter 4).
(51) A- x!óŹgyá
1s-fall.s/d.pf
‘I fell’
(52) E- x!óŹgyá
1i-fall.s/d.pf
‘She and I fell’
9
More specifically, for given values of object agreement (s, d, p, i), some prefixes in this
group are homophonous with others. E.g.: dét means both 3s:1d/p:3d (as in ‘He gave
us two cats’) and 3s:2s:3d (as in ‘He gave you.s two cats’). However, this is phonological
coincidence: second singular and first non-singular indirect object agreement diverge for
other values of object agreement: for instance, 3s:1d/p:3i (as in ‘He gave us many cats’)
is dÓt, whereas 3s:2s:3i (as in ‘He gave you.s two cats’) is gÓt.
84 CHAPTER 3
(53) E- k!úŹgyá
1i-fall.p.pf
‘They and I fell’
Given the mechanisms argued for above, we can account for these syn-
cretisms by positing two rules of postsyntactic deletion (impoverishment).
Now, it is important to note that positing such rules does not in any sense
vitiate the explanatory role that the sii class plays. Assigning Class the value
[+singular] in the current theory causes the syntax to ‘set things up’ in such
a way that the morphological component need do only the basic types of
things that it generally does (e.g., deletion; Bonet 1991, Halle and Marantz
1993) in order to produce the desired syncretisms. That is, the morphological
component need not do formal somersaults to produce second plural agree-
ment from first inclusive dual arguments; the syntax all but gives it this,
and other, syncretisms. To the extent that the morphological component
must do something, similar deletions are required elsewhere in the language
(Chapter 5); they are not mere stipulations about the sii class.
I assume person features [±author], [±hearer] (Hale 1973, Noyer 1992).

These yield first person inclusive and exclusive, and second person thus:10
(54) Person [±author] [±hearer]

first inclusive + +
first exclusive + −
second − +
We begin with the unproblematic first person singular. The structure

before valuation of number on D is:
10
[−author −hearer] corresponds to third person. However, as here, one may also repre-
sent third person as absence of person (Adger and Harbour 2007, references therein).
My own investigations of person systems lead me to posit three features, [±author],
[±participant] and privative [hearer]. Two features are adopted here for simplicity.
NUMBER FEATURES 85
(55) DP
NumberP ⎡ D ⎤
u author
⎢u hearer ⎥
⎢ ⎥
⎡Class: I ⎤ Number ⎣u singular ⎦
+author +singular u augmented
⎣−hearer ⎦ −augmented
+singular
For parity with earlier examples, first person is treated as the root noun,
with [+author −hearer] located underneath, for readability. Also on Class is
the first person’s class feature [+singular]. Number for first person singular
is naturally [+singular −augmented]. Observe that D bears uninterpretable
person as well as number. Without this, person agreement would not arise;
it was irrelevant to, and so omitted from, earlier derivations.
D is valued by matching the uninterpretable features with interpretable

counterparts on Class and Number:
(56) DP
NumberP ⎡ D ⎤
+author
⎢−hearer ⎥
⎢ ⎥
⎡Class: I ⎤ Number ⎣+singular ⎦
+author +singular −augmented
⎣−hearer ⎦ −augmented
+singular
This straightforwardly triggers first person singular agreement.
Next, consider the first person exclusive. Dual and plural are conflated
as [−singular ±augmented]. Class is as before and, so, Class and Number
oppose for [±singular], resulting in feature conflict on D:
86 CHAPTER 3
(57) DP
NumberP ⎡ D ⎤
+author
⎢−hearer ⎥
⎢ ⎥
⎡Class: I ⎤ Number ⎢−singular ⎥
⎢ ⎥
+author −singular ⎣+singular ⎦
⎣−hearer ⎦ ±augmented ±augmented
+singular
Finally, consider first person inclusive. Here, [±hearer] is specified as plus,

and, because second person, like nearly all animates, is sdi, [−augmented] is
also present on Class. 2 and 3 are shown separately, as (58) and (59).
(58) DP
NumberP ⎡ D ⎤
+author
⎢+hearer ⎥
⎢ ⎥
Class: I, you⎤ Number ⎢−singular ⎥
⎡ ⎢ ⎥
+author −singular ⎣+singular ⎦
⎢+hearer ⎥ −augmented −augmented
⎢ ⎥
⎣+singular ⎦
−augmented
(59) DP
NumberP ⎡ D ⎤
+author
⎢+hearer ⎥
⎢ ⎥
Class: I, you⎤ ⎢−singular ⎥
⎡ Number ⎢ ⎥
+author −singular ⎢+singular ⎥
⎢+hearer ⎥ ⎢ ⎥
⎢ ⎥ +augmented ⎣−augmented⎦
⎣+singular ⎦ +augmented
−augmented
Given (57)−(59), we can now move on to the syncretisms.

Agents of (di)transitives and subjects of unaccusatives (without indirect
objects) fall under Syncretisms #1−2; indirect objects and direct objects
fall under Syncretism #3. For simplicity, let us label these two contexts
NUMBER FEATURES 87
Nominative and Non-Nominative (see Adger and Harbour 2007 for an anal-
ysis of Case and Case-related syncretisms in Kiowa).
To derive Syncretisms #1−3, I propose that [±author] and [±hearer]
can be deleted postsyntactically dependent on the Case context; that is, the
bundle containing these features is impoverished.

Nom.
(60) [αauthor] → ∅ /
⎡ ⎤
Non-nom.
(61) ∅ / ⎣ +author ⎦
[αhearer] →
To see how these work, consider the agreement that DPs (57)−(59) trigger.
Agreement involves matching the features on D with the agreeing verbal
head. Suppose that head is in a Nominative context. Then we have:
(62) Nominative(57)
⎡ ⎤ Nominative(58)
⎡ ⎤ Nominative(59)
⎡ ⎤
+author +author +author
⎢−hearer ⎥ ⎢+hearer ⎥ ⎢+hearer ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢−singular ⎥ ⎢−singular ⎥ ⎢−singular ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣+singular ⎦ ⎣+singular ⎦ ⎢+singular ⎥
⎢ ⎥
±augmented −augmented ⎣−augmented⎦
+augmented
Author deletion (60) applies in this context and yields:

(63) Nominative
⎡ (57)
⎤ Nominative
⎡ (58)
⎤ Nominative
⎡ (59)
⎤
−hearer +hearer +hearer
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥
+augmented
There are no specific vocabulary items for [−hearer], so Nominative (57) is

equivalent to [−singular +singular ±augmented], which is simply a personless
inverse [−F +F]. On the other hand, Nominative (58) and Nominative (59)
are equivalent to an inverse with the second person feature [+hearer]. Since
second person, like most animates, is sdi, this is the representation of second
person plural agreement. So, we have derived that, in Nominative contexts,
1ex.d/p syncretize with 3i and 1in.d/p with 2i (Syncretisms #1−2).
Finally, consider first person (in)direct objects, both dual and plural.
88 CHAPTER 3
(64) Non-nominative(57)
⎡ ⎤ Non-nominative(58)
⎡ ⎤ Non-nominative(59)
⎡ ⎤
⎢−hearer ⎥ ⎢+hearer ⎥ ⎢+hearer ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥
+augmented
Hearer deletion (61) applies in this context and yields:

(65) Non-nominative
⎡ ⎤(57) Non-nominative
⎡ ⎤(58) Non-nominative
⎡ ⎤(59)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥
+augmented
With respect to person, all are identical, but are distinct from second and
third. Moreover, by (30), all are identical with respect to number, constitut-
ing a [+author] inverse, but are distinct from the singular. So, 1in.d, 1in.p,
1ex.d and 1ex.p all syncretize, but remain distinct from other persons of
the same number and other numbers of the same person (Syncretism #3).
Syncretism #4 follows as the consequence of the previous three. It should
be noted that, in contrast to other DPs of the language, the pronouns nÓÓ
1 and ám 2 are never inverse marked. I assume that this is a specially
conditioned form of zero marking.
3.2.4. Derivations II: More inverse marking
We have now seen how the mnemonics sdp, sdi, sii, idp, and idi derive
from the interaction between feature conflict exponence (30) and the Class
specifications in (33). How this derives an inverse-marked noun from an
underlying DP was illustrated for sdi ‘fish’ (31); the same processes apply,
mutatis mutandis and modulo sii, to DPs of other classes. We now turn to
derivations of more complex inverse-marked objects: relative clauses, third
person possessed kin terms, and DPs modified by deictics, adjectives, and
‘only’. Essentially, this section concerns the location of uninterpretable num-
ber throughout the clause. To some extent, it serves to justify the assumption
that D is one such locus. Readers more concerned with the class system and
its underlying feature structure may skip this section.
NUMBER FEATURES 89
Let us begin with adjectival modification. Matters differ little from the
DPs above. Consider:
(66) k!yáM áM hŹM-k!oM uM -bO

man- old- inv
‘old men’ (Watkins 1984, p. 99)
Assuming uncontroversially that adjectives are below D, and adjoined, we

have the DP below, once number on D has been valued:
(67) DP
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
NumberP Adjective +augmented
Class: man Number

+augmented
After vocabulary insertion we have:
(68) DP
NumberP D

bO
NumberP Adjective

k!óm
Class: man Number

ˇ á
k!yá ˇ hẐˇǐ ∅
ˇ á
Standard phonology yields k!yá ˇ hŹˇk!ǒǔbO.
Note that adjectival modification serves to emphasize that the inverse
marker and, hence uninterpretable number, is D-level, rather than lower in
the structure. Conversely, we can place an upper bound on the height of
inverse marking by considering other forms of modification.
Let us begin with demonstratives. Kiowa has two deictic roots, proximal,
90 CHAPTER 3
éˇŹˇ, and distal, óŹ. They are found, for instance, in the locatives, éˇŹˇhOO ‘here’
and óŹhyOO ‘there’. Demonstratives comprise a deictic root and a number-
sensitive suffix, -gO if modifying an inverse-marked noun, -de otherwise.
(69) éMŹMde/óŹde thalŹŹ

this /that boy
‘this/that boy’ or ‘these/those (two) boys’
(70) éMŹMgO /óŹgO thalyóp
this.INV/that.INV boy.INV
‘these/those (several) boys’
For thalŹŹ ‘boy’, an sdi noun, the inverse-marked demonstratives occur when
referential cardinality is 3. For an idp noun, say, they occur for 1.
(71) éMŹMgO /óŹgO áádO

this.INV/that.INV stick.INV
‘this/that stick’
(Demonstratives occur before the noun, as above, or after it, discontiguous

from it, or without an overt noun.)
To derive (70), we can posit the following structure. It is simply (31)
embedded under a demonstrative, where the demonstrative, like D, has un-
interpretable number that must be valued.
(72) DemonstrativeP
DP Demonstrative

u singular
u augmented
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
Class: boy Number +augmented

+augmented
As second cycle of probing and matching values the higher number:

NUMBER FEATURES 91
(73) DemonstrativeP
DP Demonstrative
⎡ ⎤
−singular
⎣−augmented⎦
+augmented
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
Class: boy Number +augmented

+augmented
By (89), p. 55, we have (74):
(74) DemonstrativeP
DP Demonstrative

gO
NumberP D

op
Class: boy Number

thalii ∅
Inserting óŹ, the distal deictic, we have thalyóp óŹgO. Variant word orders
are derived by syntactic means irrelevant here. The non-inverse-marked form,
DP+óŹde, follows if -de is a default realization of number on Demonstrative.
The same suffixes, basic -de and inverse -gO, are found with relative
clauses, depending on the class and number properties of the head noun:
(75) [ám gyá- tóM úM - khÔM OM mO] -de

anph 1s:2s:3s-talk-name.impf-BAS
‘who I was speaking to you about’ (Watkins 1984, p. 231)
(76) [pŹMáádO e- ét] -gO
table.inv 3I-big.s-INV
‘a big table’ (Watkins 1984, p. 231)
The simplest way to derive common number marking between relative clauses
92 CHAPTER 3
and deictically modified DPs is to suppose that the two are structurally iden-
tical at the relevant level or that -de/gO are default exponents. Both are
plausible. However, it was suggested, in Section 2.6.2, that -de/gO decom-
pose into a D-related element, g, with allophone d, and pure number e/O, for
bas/inv. If they are morphologically complex, with number conditioned by
D, then -de/gO are not defaults. So, syntactic similarity is a more likely ex-
planation of the common exponence. Specific possibilities are that relatives
are D-embedded CPs (Vergnaud 1974, Kayne 1994) or that demonstratives
are highly reduced relatives; however, I do not pursue the matter here.
Basic -de and inverse -gO feature in yet other constructions. They form
part of Noun+‘only’, though, here, the inverse alternant is optional:11
(77) a. KÓŹgú- de /gO- ki

Kiowa.inv-bas/inv-only
‘only Kiowas’
b. áádO- de /gO- ki
tree.inv-bas/inv-only
‘only a tree’
(78) a. KÓŹ- mathOn-de /*gO- ki
Kiowa-girl- bas/ inv-only
‘only a Kiowa girl’
b. áá- de /*gO- ki
tree-bas/ inv-only
‘only trees’
They also form part of third person kin possession (sometimes with -de
devoiced; see Watkins 1984 for possession in general):
(79) á- ii- te

3poss-son-BAS
‘his/her son or two sons’
(80) á- yyoi- gO
3poss-son.inv-INV
‘his/her sons’
11
After numerals, -kOO is used, as in yŹŹ-kOO ‘only two’ (*yŹŹ-de-ki). After personal pro-
nouns, irrespective of number, only -de-ki is permitted, consistent with the impossibility
of inverse marking on these elements.
NUMBER FEATURES 93
In (80), both the possessed noun and the possessor suffix are inverse marked.
Further modification with ‘only’ leads, optionally, to triple inverse marking:
(81) [á- yyoi- gO] -gO- ki

3poss-son.INV-INV-INV-only
‘only his/her sons’
Such structures warrant two observations. First, the feature clash generated
by the base noun spreads through the syntax in a manner reminiscent of
argument-verb agreement and can clearly be handled by positing further in-
stances of uninterpretable number. Second, if Kiowa is the head-final version
of English phrases such as only the X, then ‘only’ places an upper bound
on the locus on nominal inverse marking, for which D (English ‘the’) is a
clear possibility. (Possessive structures, only partially described above, are
too complicated for treatment here.) This complements the lower bound for
inverse marking, again D, motivated by adjectival constructions.
3.2.5. Summary
The preceding subsections have motivated the following distribution of DP:
(82) DP
NumberP D
u singular
u augmented
Class: Noun Number
The features at Class are determined by the root noun:
(83) Class Feature

sii [+singular]
sdi [−augmented]
idp [−singular]
ids [−singular ...]
idi [−singular −augmented]
sdp ∅
94 CHAPTER 3
D in (82) bears uninterpretable number features, which must be valued and

then go on to trigger verbal agreement (as well as number marking on demon-
stratives, relative clauses, possessives, and ‘only’). Uninterpretable features
involve specification of both values; however, only those matched with same
value on Class or Number are legible at the interface. The result is replica-
tion of Class and Number on D. In some cases, this leads to feature conflict,
[−F +F], on D. Exponence of such conflicts leads to inverse marking and
i-agreement. This system generates several of the mnemonics of the previous
chapter: sdp, sii, sdi, idp and idi.
The economy of this account is worth emphasizing. It relies on a variety
of theoretical assumptions concerning DP structure, feature organization and
agreement. However, beyond slight modification of uninterpretability, these
assumptions are neither novel nor specific to this analysis. The most arbitrary
element so far is the claim that noun N occurs with Class [αF], whereas
N occurs with [α F ]. However, the relationship between a noun and its
classificatory features is, in fact, far from arbitrary, as we will now see.
3.3. Mnemonic naturalness
With the underlying classificatory features of five (and a half) classes uncov-
ered, we can consider the naturalness of their mnemonics given the semantic
properties of the nouns they subsume.
The sdp class represents the minimal case. Absence of inherent number
features explains why no positive properties unite members of this class, an
assortment of implements, footwear, and natural items (2.4.5): lack of posi-
tive properties is the semantic correlate of the lack of classificatory number
features.
The sii mnemonic is entirely natural: it claims that first person, the sole
sii member, is inherently [+singular]. This is not to claim first person dual
or plural ineffable. However, these are not multiple instances of the author,
in the way that chairs denotes multiple instances of chairs. Rather, they
denote groupings of people of which the author is one element, and always a
unique, hence conceptually [+singular], one.
The sdi mnemonic is also appropriate for animates and other nouns that
determine or influence the course of their own motion. Elements that inde-
pendently determine their course of motion do not readily form homogeneous
groups. The feature [+augmented], however, ensures that properties of the
whole, P(x), are properties, P(y), of the subpart, y x. By contrast, then,
NUMBER FEATURES 95
[−augmented] ensures non-homogeneity. So, there is a natural nexus between

sdi nouns and the Class specification [−augmented].
In this light, kÓl, the other exception to the implication from animacy
to sdi membership also appears natural. In addition to ‘cattle’ and ‘buf-
falo’, it means ‘herd(s)’. Herds are crucially collective and so describable as
[+augmented], for properties of herds are properties of subparts of herds. For
nouns that are ambiguously herd-denoting or individual-denoting, the clas-
sification [−augmented], and with it sdi class membership, is inappropriate.
The idp mnemonic results from [−singular] at Class. Now, [+singular]
defines atomicity, which is conceptually close to individuality. So, inherently
[−singular] nouns are those that are not salient, or recognizable, as individu-
als. This is obvious for vegetation, and for other members of the class, such
as implements. The same reasoning applies to i and d of the ids mnemonic.
The idi mnemonic is at first sight mysterious. Given that [−singular
−augmented] is the featural correlate of referential cardinality 2, it seems
that these nouns are conceptualized as inherently dual. However, this makes
scant sense. First, of all idi nouns, only eyebrows come in pairs—apples,
hair, blackberries, brains, eyelashes, plums and tomatoes do not.12 Second,
there are good many things that are inherently paired—the thematic nouns
in -de (Section 2.6.2) for instance—but these are not idi nouns. Third, it
is doubtful that ‘two hairs’, ‘two apples’, et cetera are frequently enough
uttered and heard for idi membership to be acquirable in this way.
The naturalness of the mnemonic becomes apparent, however, if each class
feature is considered separately. Recall, from 2.4.3, Chierchia’s observation
that hair is classified as count in some languages and as mass in others
(this applies also to ‘brain(s)’, which, in English, varies between count and
mass). This crosslinguistic classificatory equivocation can be expressed in
terms of the number features motivated above. The individuability of count
12
One might think that the inherent pairhood of eyebrows or, possibly even, of brains,
owing to their hemispheres, reflects true inherent duality. However, eyebrows are idi
nouns by conceptual parity with other hair types. And the irrelevance of hemispheres
to the classification of brains can be seen in (i), where ‘brain’ triggers i-agreement. The
point is that only one brain is being talked of; if k!yagóp really meant ‘pair of brain
hemispheres’ rather than simply ‘brain’, we would see d-agreement.
(i) K!yagóp hááxo Ó - háı́-dOO!

brain.inv how :3s:3i-q- be
‘What a brain he has!’ (Idiomatic expression of intelligence)
96 CHAPTER 3
nouns implies [−augmented]; however, salience of the collection, and hence

deemphasis the parts, implies [−singular]. Alternatively, thinking of apples,
et cetera, like all vegetation, they are [−singular] because inherently non-
individuable. However, given the clusters in which they grow, the constituent
fruit are clear and evident, like the strands in a head of hair, and, so, the
collection is non-homogeneous [−augmented]. Consequently, [−singular], on
its own, indicates a property of idi nouns, and [−augmented], on its own,
does so too. Of course, copied onto D, they trigger d-agreement; but this
apparent inherent duality is accidental. So, ‘two hairs’, ‘two apples’, et cetera,
of implausible frequency, are not requisite data for acquisition of this class.
This implies that the interpretation of [−singular −augmented] is subtly
different for Class and for Number. For Class, each feature applies to the
root noun directly and, so, separately. For Number, [−augmented] applies to
[−singular] and [−augmented]([−singular]) then applies to the root noun (14)
(see also Lemmas 1−3, p. 113 f.). This is not surprising, however. First, it is
clear that the interpretation of number features in the two cases is somewhat
different: Number has a direct, truth-conditional effect on interpretation,
but Class affects interpretation more marginally (as when one root has two
interpretations class-dependently, e.g., áá is ids ‘tree’ but idp ‘stick, pole’,
kÓl is i ‘heads of cattle’ but p ‘herds of cattle’). Now, if the effect of Class
is to determine the precise meaning of the root, then, naturally, all Class
features apply directly to the root, rather than to each other. On Number,
however, locality forces interpretation of the bundle [−singular −augmented]
before the features can apply to the neighboring head, {Class, Noun}.
The preceding chapter argued that the classes are internally semantically
coherent, in that the nouns each class subsumes share properties. We now see
that these shared properties are naturally expressible in terms of number fea-
tures. Moreover, these number features are responsible, in combination with
the mechanisms for valuing and pronouncing number on D, for the agreement
types s, d, p and i. Consequently, the system of mnemonics accords with
the semantic properties of the classes. As we discuss the remaining classes
and their underlying features and values, the semantic naturalness of those
other mnemonics will also become clear.
This concludes the first part of the chapter. The discussion now
moves away from the features [±singular] and [±augmented] and
the classes SDP, SDI, SII, IDP, and IDI to the feature [±group] and
the classes SDS, IDS, SSS, and PPP.
NUMBER FEATURES 97
3.4. Spurious S/P
The first part of the chapter (3.1−3.3) established and explained the broad
correlation between s-agreement and referential cardinality 1, d-agreement
and 2, and p-agreement and 3. It further explained how i-agreement inter-
feres with this correlation. The classes under consideration in this section,
and in the second part of the chapter generally, also interfere with the cor-
relation. However, they do so by virtue of what might be termed spurious
s and spurious p, occurrences of s-agreement and p-agreement that do not
correlate with their typical referential cardinality values: sds, ids and idi in
its ‘different types of’ reading show s-agreement when referential cardinality
is 3, and ppp nouns show p-agreement for referential cardinalities 1 and 2.
Below, I argue that spurious s and p should be attributed to a feature
[±group]. As before, the feature is given a clear semantics, associated with
the relationship between wholes and parts. The feature has an effect on the
valuation of D that reflects its interpretation.
3.4.1. Types of grouphood
Nouns that trigger spurious s-agreement form pluralities in which the whole
is more salient than the part. For sds nouns, this results because they are
non-shape-inductive. Faced with a single cloud, or a small number of them,
it is generally clear where the boundaries are. However, in a large group,
boundaries of the single clouds are unclear and one is likely to view the whole
as a single mass itself. For ids, similar reasoning holds. A single mountain or
a single tree may be well defined. However, in a mountain range or in a grove
or spinney, the boundaries of individuals are indistinct and the range or grove
is viewed as a whole itself. For idi, Watkins describes the reading they attain
under s-agreement as one of ‘plural sets’: ‘three or more separate collections
of a single type, e.g., varieties of apples in separate piles or bags’, or ‘more
than two sets of hair, i.e., heads of hair belonging to different individuals’
(Watkins 1984, 88−89). The key notion of ‘collection’ or ‘set’ again exhibits
the property of not exhibiting clear or inductive boundaries. So:
(84) Generalization: Spurious S nouns form pluralities in which the

whole is more salient than the part.
98 CHAPTER 3
Pluralia tantum nouns exhibit a property almost the opposite of (84).

Even when singular, trousers, war headdresses, necklaces, and so on are
clearly composed of parts. That is:
(85) Generalization: Spurious P nouns have salient parts even when

non-plural.
Given the complementarity of (84) and (85), we should define a single binary
feature for which the whole∼part relations correspond to the plus and minus
values. The solution that I will shortly offer is that the feature predicates of
[±augmented]. This solution arises naturally from the failure of some more
obvious possibilities, from which we begin.
At first, one might think to define simply a predicate Parts(x):
(86) Parts(x) is true if, and only if, x has salient subparts.
However, this feature is too coarse. We are concerned not with properties of
x per se but with properties of the pluralities that x forms. Dependence on
plurality might be captured by including an implication in the definition:
(87) [+group] if [+augmented] → Parts(x)

[−group] if [−augmented] → ¬Parts(x)
However, (87) violates the definition of feature negation, [−F] = ¬[+F], and
cannot be fixed by choosing some truth functor other than ‘→’ (Appendix
3.8). The technical problem can be avoided by use of α-notation:
(88) [αgroup] = [αaugmented] → αParts(x)
However, this notation merely disguises that one cannot define one value and
deduce the other, and, so, is of a richer type than we have needed until now.
Instead, I propose that [±group] be a ‘partial’ feature. That is, it has a
definition in the standard style but predicates of [+augmented].
(89) [+group] = λP[+augmented] [P(x) ∧ ¬Parts(x)]
Now, it is possible to dispose of the novel notion Parts(x), in favor of more

conventional predicates, as follows. Observe first, that Parts(x) is almost the
converse of atomhood, as atoms lack (salient) parts and things with salient
parts are non-atomic. So, we can think [+group], as defined in (89), as
introducing a new predicate, Q, and asserting that x is atomic with respect
NUMBER FEATURES 99
to Q. Incorporating Q-atomicity into the definition yields:
(90) Definition: [±group]

[+group] = λP[+augmented] ∃Q[P(x) ∧ P = Q ∧ Q-atom(x)]
x, a free variable
Importantly, the notion of Q-atomicity is adopted from Krifka (1992) and so

is not a novel assumption of the current theory. Rather, it is a legitimate
resource to call on in a morphosemantic theory of number.
(91) Definition: Q-atom

Q-atom(x) = Q(x) ∧ ¬∃y[Q(y) ∧ y x]
According to (90), [+augmented] x is [+group] if it lacks salient (distinguish-

able) subparts and so is atomic with respect to some other predicate (namely,
the collective); and it is [−group] if it has salient subparts, and so is non-
atomic with respect to other predicates. Note, crucially, the generality of
the phrase ‘other predicates’, which arises by negating (−) the existential of
[+group]. This captures the fact that the composite parts of pluralia tantum
nouns are non-uniform, that is, there is no single predicate with respect to
which they are atoms.
If [+augmented] is absent, then [αgroup] does not predicate of anything.
Now, this might be thought to cause a presupposition failure. However, as
discussed on p. 96, class features do not affect the truth conditions directly,
but constrain the interpretation of the root noun. So, if [+augmented] is
absent, [±group] does not constrain the noun at all, and so is inert, both for
interpretation and syntax, that is, both conceptually and computationally.
Note an immediate consequence of restricting [±group] according to (90).
Watkins observed that idi nouns, when they trigger s-agreement, refer to
three or more groupings (p. 97 above). The noun has this [+group] reading
when Number is [+augmented]. The reading does not arise for other values of
Number: álOO cannot mean ‘two groups of apples’ when Number is [−singular
−augmented]. This follows from the restriction of [+group] to [+augmented]:
without it, [+group] is inert and the ‘collections’ reading unavailable.
We can now state the classificatory features of several more classes.
100 CHAPTER 3
(92) Class Feature

sds [+group]
ids [−singular +group]
idi [−singular −augmented (+group)]
ppp [+augmented −group]
For idi, [+group] is parenthetic because the group reading is optional. Obs-
erve that ppp’s classificatory features must include [+augmented]. Other-
wise, for referential cardinalities 1 and 2, there would be no [+augmented]
feature and [−group] would be inert, resulting in sdp. Its inclusion is con-
ceptually natural, given, on the one hand, the composite nature (hence, plu-
rality) of pluralia tantum and granular mass nouns, and, on the other, that
[+augmented] is the crucial defining feature of plurality.
3.4.2. Derivations III: Grouphood
Having argued that there is a third feature on Class, we must ask whether
there is a third uninterpretable feature on D. If so, then the valued DP of an
sds noun of referential cardinality 3 would be:
(93) DP
NumberP ⎡ D ⎤
−singular
⎣+augmented⎦
Class: Noun Number +group

+group −singular
+augmented
If uninterpretable number is uniform throughout the language, then [+group]

would propagate from D onto verbs, demonstratives, relative clause, and the
other elements of Section 3.2.4. However, this is problematic, as [±group] is
never itself pronounced. This does not undermine the claim that it exists; its
semantic motivation still stands. Instead, it suggests that [±group] is never
present in the feature bundles comprise the verbal prefix and so on. Rather,
the effect of [±group] is on the valuation of D, triggering s-agreement for
referential cardinality 3 and p for 1/2. Naively, then:
NUMBER FEATURES 101
(94) Computational effect of [±group]

Suppose [±group] occurs (with [+augmented]) on Class/Number.
Then D is valued as [±singular ∓augmented].
Essentially, then, [±group] acts like a decoy for valuing [usingular] and
[uaugmented]: [usingular] acquires the value of [±group], [uaugmented] the
opposite. A summary of the divergence [±group] forces between Number and
D (in those cases where its effect is non-trivial) makes this apparent:
(95) [±group] Number D

+singular −singular
[−group]

[−group]

[+group]
+augmented −augmented
Now, permitting non-matching features to match values is, of course, a

highly questionable complication of the feature matching mechanism. How-
ever, so radical a revision is not necessary if we consider again the semantics
[±group]. Conceptually, groups are atomic pluralities: that is, they are plu-
ral with respect to one predicate, such as tree(x), but atomic with respect
to the corresponding collective, such as grove(x). Conversely, collections
of diverse objects are atomic with respect to a pluralia tantum predicate,
such as trousers(x), but there is no natural corresponding non-collective,
as the parts that comprise trousers are non-uniform. The semantic effect
of [±group], then, is to shift interpretation of the root noun between col-
lective and non-collective. In so doing, it recalibrates the relevant notion of
atomhood, with pluralities becoming atomic and non-pluralities becoming
non-atomic. This is notion of Q-atomicity in the definition (90).
So seen, the computational effect of [±group] is more natural. It shifts
the root noun to or from a collective, inducing a new set of atoms, and it
is with respect to the new atoms that the feature concerned with atomicity,
[usingular], on D, is valued. It is valued as [+singular] if [+group] induces
a set of atomic plurals, and as [−singular] if [−group] induces a set of non-
atomic plurals. (The value of [±augmented] could be incorporated into the
syntactic computation. However, I believe that it is a default in these cases,
102 CHAPTER 3
arising via the mechanisms discussed by Noyer 1998, Harbour 2003a, a di-
gression profitably avoided here. See Section 4.5.1 for crucial evidence that
the representation of [±augmented] in these cases is indeed different from
others.)
A note on mnemonic naturalness
The semantics of the feature [±group] has been motivated directly on the
basis of the semantic properties of the nouns that it classifies. Consequently,
nothing more need be said about the naturalness of the role that [+group]
and [−group] play in classification. Similarly, its effect on valuation of D has
already been discussed. Consequently, nothing more need be said about the
mnemonic naturalness of the classes in (92).
Spurious S-agreement
We now consider derivations for sds and ids nouns.

We begin with referential cardinalities 1 and 2 for the sds class. Class is
[+group] and Number is [±singular −augmented].
(96) DP
NumberP D
u singular
u augmented
Class: Noun Number

+group ±singular
−augmented
As (96) contains no (interpretable) instance of [+augmented], [±group] is

conceptually, and, so, computationally inert: it does not affect interpretation
of the root noun or the valuation of D. So, D simply replicates Number:
NUMBER FEATURES 103
(97) DP
NumberP D
±singular
−augmented
Class: Noun Number

+group ±singular
−augmented
This is all but identical to an sdp DP for referential cardinalities 1/2. s/d-
agreement result, as desired.
For ids nouns of referential cardinality 1/2, Class is [−singular +group]:
(98) DP
NumberP D
u singular
u augmented
Class:
Noun
Number
−singular ±singular
+group −augmented
Again, without interpretable [+augmented], [+group] is conceptually and

computationally inert. So, (98) is in practical terms identical to an idp DP
of referential cardinality 1/2. So, i/d-agreement result, as desired.
Now consider an sds noun of referential cardinality 3:
(99) DP
NumberP D
u singular
u augmented
Class: Noun Number

+group −singular
+augmented
Here, [+group] is rendered active by [+augmented] on Number. So, by (94),

D is valued as [+singular −augmented], triggering s-agreement, as desired.
104 CHAPTER 3
(100) DP
NumberP D
+singular
−augmented
Class: Noun Number

+group −singular
+augmented
The derivation is identical for an ids noun of referential cardinality 3.

The extra feature, [−singular], on Class, has no effect on the computation:
(101) DP
NumberP D
+singular
−augmented
Class:
Noun
Number
+group +augmented
Again, the DP triggers s-agreement.
Spurious P-agreement
Now consider a pluralia tantum ppp noun, for any referential cardinality.
Class is [+augmented −group]; Number is expressed in the general form
[±singular ±augmented].
(102) DP
NumberP D
u singular
u augmented
Class: Noun Number
+augmented ±singular
−group ±augmented
NUMBER FEATURES 105
Irrespective of the value of [±augmented] on Number, the [+augmented] of

Class renders [−group] active. Therefore, D will be valued as [−singular
+augmented], triggering p-agreement, for all referential cardinalities.
(103) DP
NumberP D
−singular
+augmented
Class: Noun Number
−group ±augmented
3.4.3. Summary
We can now expand the list of noun classes and classificatory features.
(104) Class Feature

sii [+singular]
sdi [−augmented]
idp [−singular]
sds [+group]
sdp ∅
The feature [±group] pertains to salience of objects’ subparts. It predicates

of [+augmented] and is conceptually and computationally active only in the
presence of that feature. That is, if [+augmented] is found either on Class
or Number, [±group] affects the computation. Otherwise, it is inert. The
computational effect of [±group] is to value D [±singular ∓augmented]. It
has this effect in virtue of introducing a new predicate, the (non-)collective
counterpart of the root noun, and recalibrating the notion of atomhood in
accord with it. Uninterpretable number on D is valued in accord with the
recalibrated notion of atomhood; hence [±group] and [±singular] share signs.
106 CHAPTER 3
3.5. Mass nouns
We now have a firm enough grasp of [±singular], [±augmented] and [±group]

to move from the semantic properties of mass nouns to their class features.
For both granular and non-granular mass nouns, divisibility (any subpart
of water is water) entails both non-atomicity and augmentation. So, they
are either negatively, or zero, specified for [±singular] and are [+augmented].
Further, granularity means salience of subparts, and so corresponds to
[−group]; conversely, non-granularity corresponds to [+group]. We can,
therefore, ascribe to mass nouns the following classification:
(105) Class Feature

ppp [−singular +augmented]
sss [−singular +augmented +group]
The classification makes two non-crucial assumptions. The first is that mass
nouns do not project NumberP. This is justifiable in terms of the well noted
incompatibility between mass nouns and count expressions. It ensures con-
stant agreement for both granular and non-granular nouns. If we abandoned
the assumption, the combination [+augmented +group] would ensure con-
stant s-agreement for non-granular mass nouns; addition of [−group] to Class
would ensure constant p-agreement for granular mass nouns. The second as-
sumption is that mass nouns have negative, rather than zero, specification
for [−singular]. If zero specified, then [−singular], necessary for p-agreement,
can be assumed to arise postsyntactically, by insertion of unmarked feature
values (Noyer 1998, Harbour 2003a). It is included in (105) for purely expos-
itory reasons, so that valued D resembles that of previous trees. Fully valued
sss/ppp DPs are shown, in accord with both assumptions, in (106)−(107):
(106) DP
⎡Class: Noun⎤ D
⎣+augmented⎦ −augmented
+group
NUMBER FEATURES 107
(107) DP
Class: Noun D
+augmented +augmented
3.5.1. Conjunction
The foregoing is enough to justify constant s/p-agreement for mass nouns.

However, the sss/ppp mnemonics were justified mainly by their agreement
behavior under conjunction. A brief examination of conjunctions of non-mass
nouns establishes several generalizations from which the agreement behavior
of mass noun conjunctions follows directly.13
Conjunctions can have greater referential cardinality than any of their
individual conjuncts. For instance, 1 + 1 = 2 (but 1 + 3 = 3). Recall from
Chapter 2, examples (27)−(30), that, for sdi nouns, increasing in referential
cardinality through conjunction has the same effect on agreement as increase
in referential cardinality by numeral modification. That is, ‘this young man
and that’ triggers d-agreement, ‘this young man and those two’, i-agreement.
By contrast, conjunctions of sdp nouns trigger agreement purely according
to referential cardinality of the conjunction. Therefore, the contents of Class
and Number of each conjunct contribute to the agreement triggered by the
13
Agreement triggered by conjunction is a topic of my ongoing research. It is relevant to
the number theory for several reasons. First, observe that languages have mechanisms for
determining the referential cardinality of a conjunction from the referential cardinalities
of its individual conjuncts. Clearly, some form of computation is involved and it is the
task of a theory of number to account for such mechanisms. Second, intra- and inter-class
conjunction provide an opportunity to test the content proposed for Class and Number
here; the agreement type triggered by a conjunction
n
C= ∧ Ci
i=1
for some integer n > 1, should follow from simple generalizations concerning the featural
content of Class and Number for each conjunct Ci . This is no small task, as the total
number of intra- and inter-class conjunctions, even excluding the sii class, is (8 × 3)2 +(8 ×
3)3 = (8 × 3)2 × 25 = 122 × 100 = 14, 400. The reason that this is a topic of ongoing
research is not that I intend to sift through each one of these cases. Rather, it is that each
modification to one’s view of the feature content of Class, Number and D in turn changes
what are the crucial test-case conjunctions, requiring further trips to the field.
108 CHAPTER 3
conjunction as a whole. I implement this by proposing that the functional

projection that hosts gO ‘and’ cooccurs with Number and D projections.
(108) DP
NumberP D
u singular
u augmented
⎡Class: andP ⎤ Number
(±group)
Number’s features are determined by a computation over the c-commanded

Number heads, i.e., those of the conjuncts under andP. Likewise, those of
Class are determined by a computation over the c-commanded Class heads.
The computation of Number is simply ‘addition’ (recall that 3 is the equiv-
alence class of all cardinalities greater than two). Based on the behavior of
conjunctions of sdi nouns, let us tentatively assume that, when all conjuncts
are classmate, Class in (108) is simply copied from the lower Class nodes.14
Therefore, in a conjunction of sss or ppp nouns, Class in (108) will be
as for the conjuncts. Number in (108) is slightly trickier. One possibil-
ity, is that, in the absence of lower Number heads, Number is valued by a
computation over Class heads. Alternatively, absence of Number on the con-
juncts might entail absence of Number on the conjunction. Assuming that
some such mechanism guarantees either the absence of Number or its iden-
tity with Class, we can disregard Number for conjunctions of mass nouns. In
14
In fact, matters are slightly more complex, as some intra-class conjunctions permit dif-
ferent agreement options. For instance, both s-agreement and p-agreement are acceptable
for conjunctions of ids nouns of referential cardinality 3.
(i) P!ÓÓ gO áá gya /gyat- gúttO

river and tree 1s:3s/1s:3p-draw.impf
‘I’m drawing rivers and trees’
A second type of agreement variation is that, in Mr Bointy’s dialect, ‘this young man
and those two’ triggers either d- or i-agreement. This is not a right-conjunction effect, as,
more generally, d-agreement and i-agreement are both possible for any conjunction of sdi
nouns, where each conjunct has referential cardinality 1 or 2. Other speakers do not have
this variation. On the d∼i alternation in Kiowa generally, see Section 4.5.2.
NUMBER FEATURES 109
consequence, such conjunctions agree just as the unconjoined nouns do.
This concludes the second part of the chapter.
3.6. Conclusion
Noun class systems are often seen as bastions of the arbitrary. However,
Kiowa’s noun class system is highly principled. We saw in the last chapter
that the nouns subsumed under each class mnemonic are semantically coher-
ent. In this chapter, the feature system that underlies the classification has
been examined. The results are that [±singular] and [±augmented], the fea-
tures that compose referential cardinality, are also used for noun classification
along with a third feature, [±group], which qualifies type of augmentation.
(109) Class Feature

sii [+singular]
sdi [−augmented]
idp [−singular]
sds [+group]
sdp ∅
sss [−singular +augmented +group]
ppp [−singular +augmented]
There is a principled relationship between a noun’s semantic properties and

its classificatory feature. Furthermore, class features interact with referential
cardinality in a consistent and straightforward fashion, valuing D and so
generating the class mnemonics. Consequently, the relationship between class
semantics and class mnemonics is non-arbitrary.
The theory above employs well justified and/or standard assumptions
concerning the structure of DPs and agreement. The main points are reiter-
ated below. Number features occur on Class, Number and D:
110 CHAPTER 3
(110) DP
NumberP D
u singular
u augmented
(±group)
Uninterpretability is taken, novelly, to involve overspecification: [uF] means

[−F +F]. Valuation involves matching each uninterpretable [αF] with an in-
terpretable correspondent. Only matched [αF] are visible at the interface,
for pronunciation and interpretation. The loci of interpretable number fea-
tures are Class and Number, which are equidistant from D and so equally
accessible to it. The effect of valuation, then, is to replicate the feature con-
tent of both heads on D. This view of valuation permits conflicting feature
specifications to arise on D. Such conflicts are pronounced on D and related
heads as inverse marking, and on verbal heads as i-agreement.
Valuation of D is also affected by [±group]. The feature predicates of
[+augmented]. When no [+augmented] is present, it modifies an aspect of
the root noun that is irrelevant to the DP and so is conceptually and compu-
tationally inert. When active, it introduces a new predicate and recalibrates
the notion of atomhood. It is with respect to this new set of atoms that
uninterpretable number on D is valued. So, in essence, [±group] acts like a
surrogate for the valuing of [±singular] and, from that, of [±augmented].
3.7. Appendix: Missing mnemonics
Chapters 2−3 have concentrated on the classes that Kiowa has and not on
the classes it hasn’t. Yet, there are 64 mnemonically possible classes (111)
and it is natural to ask why only nine are attested (sdp, sdi, idp, ids, idi,
sds, ppp, sss, sii). Where are the missing 55, such as ddd, pip, pds?
⎧ ⎫⎧ ⎫⎧ ⎫
⎪ s ⎪⎪ s ⎪⎪ s ⎪
⎨ ⎪
⎪ ⎬⎪⎨ ⎪ ⎬⎪⎨ ⎪ ⎬
d d d
(111)
⎪ p ⎪⎪ p ⎪⎪ p ⎪
⎩ ⎪
⎪ ⎭⎪⎩ ⎪ ⎭⎪⎩ ⎪ ⎭
i i i
NUMBER FEATURES 111
In fact, most of these classes are excluded on principled grounds and we

find that Kiowa makes near optimal use of its classificatory resources. (See
Harbour 2006c for further discussion and inter-Tanoan comparison.)
An attempt to explain the scarcity of attested mnemonics might concen-
trate on the inventory of mnemonics itself:
Consider d-agreement. This arises only when D is valued

as [−singular −augmented]. This in turn requires that Num-
ber be specified as [−singular −augmented] (and Class as some
subset thereof, [(−singular) (−augmented)]). The requirement
that Number be specified in this way restricts d-agreement to
the referential cardinality corresponding to the feature combina-
tion [−singular −augmented], i.e., 2. Therefore, d is confined to
the middle position of the mnemonic. The number of possible
mnemonics reduces from 64 to 3 × 4 × 3 = 36.
If the third position of the mnemonic is s, then the class is
[+group]. If the first position is p, it is [−group]. Taking [−group
+group] to be semantically impossible, we exclude p. . . s and
reduce the number of possible mnemonics to 36 − 4 = 32.
If the second position of the mnemonic is i, then the class is
either [+singular] or [+augmented]. If [+augmented], then the
first and second positions are identical, unless the class is spec-
ified as [−group]. But [−group] would block i-agreement when
referential cardinality is 2. So, pis and pip are excluded, and pii.
Likewise, iis is impossible as the last position requires [+group],
the first two [+augmented]; but [+augmented +group] is sss.
The number of possible mnemonics reduces to 28.
One can imagine how such reasoning would continue, but it is rather super-
ficial, emphasizing the mnemonics first and the features that generate them
second. This study focuses on features first and, so, a better test of the
theory developed above is to see how it constrains the space of mnemonics.
The three class features are [±singular], [±augmented], [±group]. For
each class, they are specified as plus, minus, or zero. This yields 27 classes:
⎧ ⎫⎧ ⎫⎧ ⎫
⎨ [+singular] ⎬⎨ [+augmented] ⎬⎨ [+group] ⎬
(112) [−singular] [−augmented] [−group]
⎩ ⎭⎩ ⎭⎩ ⎭
∅ ∅ ∅
112 CHAPTER 3
Therefore, 37 mnemonically possible classes are featurally impossible.

The remaining classes are tabulated in (113). Of the 27 possibilities,
[(±singular) −augmented ±group] and [+singular +augmented (±group)]
are excluded below: as [±group] predicates of [+augmented], the specifi-
cation [−augmented ±group] is semantically questionable; and [+singular
+augmented] is simply contradictory (Corollary, p. 114).15 ‘0’ represents
absence of a feature; boldfacing marks classes not attested above.
(113) [±singular] [±augmented] [±group] Class

+ − 0 sii
+ 0 + SIS
+ 0 − SIP
+ 0 0 sii
− + + sss
− + − ppp
− + 0 IIP
− − 0 idi
− 0 + ids
− 0 − idp
− 0 0 idp
0 + + sss
0 + − ppp
0 + 0 IIP
0 − 0 sdi
0 0 + sds
0 0 − sdp
0 0 0 sdp
Only three of the resulting mnemonics are new: sip, sis and iip, correspond-
ing to [+singular ±group] and [(−singular) +augmented]. Two observations
are appropriate. First, these classes are attested in related languages, such
as Jemez (Sprott 1992, Yumitani 1998; see Noyer 1992, Harbour 2006c for
crosslinguistic analysis). So, the system does not overgenerate. Second, ob-
serve that all are based on plus specification of [±singular] and [±augmented].
As a general rule, Kiowa avoids such specification: it occurs only for the first
15
Non-granular mass nouns are the nearest approximation to such nouns. However,
[(−singular) +augmented +group] is more satisfying, especially when faced with two liq-
uids at once.
NUMBER FEATURES 113
person singular (sii [+singular]), or in conjunction with [±group]. So, sip,

sis and iip, are naturally absent from Kiowa, avoiding ‘local’ overgeneration.
3.7.1. Beyond Merrifield’s method
Throughout this work, classes have been identified on the basis of Merrifield’s
method (cf., Harrington 1928), examining correlations between referential
cardinality and agreement type. However, in the discussion of subclasses
(p. 34), attention was drawn to the possibility that featurally distinct classes
can be mnemonically indistinguishable, as with ppp, which comprises pluralia
tantum and granular mass subclasses.
Given (113), we can move beyond Merrifield’s method, associating seman-
tic properties of nouns, not with mnemonics, which underdetermine class fea-
tures, but with the class features themselves directly. For instance, footwear
might be sdp not because it is default, but because it is [−group]; and sub-
classes of idp nouns might be distinguished by zero versus minus specification
for [±group] (vegetation versus implements and body parts, say). This still
permits many-to-one correspondences between semantic characteristics and
feature sets. However, the method of detecting such subsets is more direct
than Merrifield’s method permits. I leave this for future research.
3.8. Appendix: Formal demonstrations
3.8.1. Cardinality
I now present four simple proofs justifying (9) and the exclusion of [+singular
+augmented].
Lemma 1. [+singular −augmented] corresponds to 1.
Proof. By definition, atom(x) entails ¬∃y[y x]. So, in particular,

¬∃y[atom(y) ∧ y x]. Now, if P |= Q, then P |= Q ∨ ¬P. So:
atom(x) |= ¬∃y[atom(y) ∧ y x] ∨ ¬atom(x)

≡ ¬∃y[atom(x) ∧ atom(y) ∧ y x]
≡ [¬λP∃y[P(x) ∧ P(y) ∧ y x]](atom(x))
≡ [−augmented]([+singular])
114 CHAPTER 3
By definition, 1 corresponds to atomicity. So, [+singular −augmented] = 1.
Corollary. [+singular +augmented] is contradictory.
Lemma 2. [−singular −augmented] corresponds to 2.
Proof. It suffices to show that only dyadic x satisfies:

[−singular −augmented] ≡ [−augmented]([−singular])
≡ [¬λP∃y[P(x) ∧ P(y) ∧ y x]](¬atom(x))
≡ ¬∃y[¬atom(x) ∧ ¬atom(y) ∧ y x]
≡ ∀y[atom(x) ∨ atom(y) ∨ y x]
This simplifies to ∀y[atom(y) ∨ y x], given the monotonicity of standard

logics (specifically, x such that atom(x) is disregarded as a means of satisfying
[−singular −augmented], as it contradicts the prior [−singular]).
Consider the logically equivalent formula ∀y[y x → atom(y)]. Now, if
x = {a, b}, then {a} and {b} are the only y such that y x; and atom(y).
However, for any larger x, such as x = {a, b, c}, there is non-atomic y, such
as {a, b}, such that y x. So, only dyadic x satisfies the formula.
Lemma 3. [−singular +augmented] corresponds to 3.
Proof. It suffices to show that triadic x, or larger, satisfies:

[−singular +augmented] ≡ [+augmented]([−singular])
≡ [λP∃y[P(x) ∧ P(y) ∧ y x]](¬atom(x))
≡ ∃y[¬atom(x) ∧ ¬atom(y) ∧ y x]
The formula requires that x be non-atomic with a non-atomic subpart. The

smallest value for y, then, is a dyad. So, the smallest value for x is a triad.
Theorem. (9) accurately represents, and exhausts, the combinatorial pos-

sibilities of [±singular] and [±augmented].
Proof. Lemma 1, Lemma 2, Lemma 3, Corollary.
Note. See p. 70 for a graphic interpretation in the shape of a lattice.

NUMBER FEATURES 115
3.8.2. Grouphood
On p. 98, we entertained the [±group] definition:
(114) [+group] if [+augmented] → Parts(x)

[−group] if [−augmented] → ¬Parts(x)
However, this violates the feature negation condition, [−F] = ¬[+F]. I now
prove that this is not fixed by choosing a more complex truth functor than
‘→’. Rather, it is simply impossible to define the feature in this way.
Theorem. There is no feature / truth functor [+group] = ∗(A, B), such

that [−group] = ¬[+group] and:
∗ (A, B) ≡ ¬A → B (3.1)
¬ ∗ (A, B) ≡ A → ¬B (3.2)
Proof. Suppose, for a contradiction, that ∗ exists. By (3.2):
| ¬ ∗ (A, B) | = 0 iff | A → ¬B | = 0
iff | A | = | B | = 1 (3.3)
However, it follows from the definition of negation that:
| ¬ ∗ (A, B) | = 0 iff | ∗(A, B) | = 1
So, by (3.1):
| ¬ ∗ (A, B) | = 0 iff | ¬A → B | = 1 (3.4)
Combining (3.3) and (3.4) produces the desired contradiction:
| A | = | B | = 1 iff | ¬A → B | = 1
Comment. Taking A as [+augmented] and B as Parts(x), the theorem

shows that there is no logical manipulation of [+augmented] and Parts(x)
that permits [±group] to be defined as in (114).
Chapter 4
Agreement and Suppletion
Several predicates in Kiowa are sensitive to the number of their inner argu-
ment; that is, Kiowa has predicates that supplete for number. Consequently,
sensitivity to number is a property of two parts of Kiowa grammar: agreement
and suppletion. Mostly, these two operate in tandem: if referential cardinal-
ity is 1, the verb will occur in its s-suppletive form bearing s-agreement, or if
referential cardinality is 3, the verb will occur in its p-suppletive form bear-
ing p-agreement. However, at times, agreement and suppletion mismatch,
with agreement implying one referential cardinality and suppletion, another.
This chapter is concerned with the mechanisms of suppletion in general
and with agreement∼suppletion mismatches in particular. I argue that these
facts receive a natural analysis given the theory developed in Chapter 3. The
aspect of that theory relevant to mismatches is the divergence permitted
between the feature content of Number and D (and between Class and D). By
providing divergent feature specifications in a single syntactic structure, the
theory permits divergence between number-sensitive phenomena, provided
they depend on different heads. For instance, if agreement depends on D, as
argued in Chapter 3, and if suppletion depends on Number or Class, then
agreement and suppletion will be conditioned by different features whenever
grouphood or inverse marking affect D.
The discussion begins with the introduction of Kiowa’s number-sensitive
predicates and clarification of the notion of suppletion, as opposed to allo-
morphy and phonological readjustment. I then suggest some syntactic and
morphological principles following Sportiche (1997) and Adger, Béjar, and
Harbour (2001), according to which it is natural for agreement and suppletion
to be sensitive to different heads. With these principles in hand, basic cases
117
118 CHAPTER 4
of suppletion without mismatches are derived. These are followed by treat-

ment of inversive mismatches, mass nouns, [±group]-induced mismatches,
and reflexive-induced mismatches. An appendix provides information on
further uses of suppletive pairs and their interesting differences in meaning.
4.1. Suppletion
This section introduces Kiowa’s suppletive predicates and defines suppletion.
4.1.1. Number-sensitive predicates
Kiowa’s number-sensitive predicates fall into two classes. Those in (1) display
an S∼D/P opposition, those in (2), S/D∼P.
(1) Gloss S D/P

big, old, important êl bẐn
small, young syÓn syán
tall, long kyóMŹM kŹMŹMnŹŹ
short xéŹ xáádóú
(2) Gloss S/D P
be sitting áM áM gya k!úl
be lying k!ÓÓ k!úl
be sitting.inan xél sÓl
set, put in xéŹ sÓÓ
lay x!óú k!úú
land, fall against, fall x!óŹgyá k!úŹgyá
drop, fall ól p!él
sever t!ál tháá
wander thóú zéMŹM
In featural terms, the predicates in (1) are sensitive to the value of [±singular],
those in (2), to [±augmented].1 Terminologically, these features are said to
condition suppletion of the respective predicates. The semantic division
is between individual-level (1) and stage-level (2). It is natural to conjecture
as to semantic principles behind this correlation: perhaps because stages
1
Some predicates in (2) share roots: ‘lay’ ∼ ‘land, fall against, fall’; ‘be sitting.inan’
∼ ‘set, put in’. See Watkins (1984) on the semantics of -l, -i and -gyá.
AGREEMENT AND SUPPLETION 119
can be in a subpart relation, they are associated with the subpart feature
[±augmented]; and because permanent states are not in a subpart relation,
they are associated with the atomic feature [±singular]. However, I leave
this aside, as supporting evidence is scant (but see Appendix 4.7.)
Number-conditioned suppletion can be illustrated by combining ‘big’ from
(1) and ‘be lying’ from (2) with the sdp noun tóúdé ‘shoe’. In (3), referential
cardinality is 1 and we find s-agreement and the s-form of the predicate, êl.
(3) Tóúdé ∅- ét

shoe 3s-big.s
‘The shoe is big’
In (4)−(5), for referential cardinalities 2−3, we find the d/p-form of the

predicate, bẐn. The two are distinguished by d∼p-agreement.
(4) Tóúdé eM - bẐn

shoe 3d-big.d/p
‘The two shoes are big’
(5) Tóúdé gya-bẐn
shoe 3p- big.d/p
‘The shoes are big’
Now consider ‘be lying’. In (6)−(7), for referential cardinalities 1−2, we

find s/d-form of the predicate, k!ÓÓ, distinguished by s∼d-agreement.
(6) Tóúdé ∅- k!ÓÓ

shoe 3s-be lying.s/d
‘The shoe is lying’
(7) Tóúdé eM - k!ÓÓ
shoe 3d-be lying.s/d
‘The two shoes are lying’
In (8), for 3, we find, the p-form of the predicate, k!úl.
(8) Tóúdé gya-k!úl

shoe 3p- be lying.p
‘The shoes are lying’
Both patterns can be shown in tandem (cf., Hale 1997 for Hopi). Here,
‘big’ is used attributively; ‘drop’ is sensitive to the number of its internal
120 CHAPTER 4
argument. (Tóúdé ‘shoe’ loses its -de when modified; Section 2.6.2.)
(9) Tóú- êl gya- ót

shoe-big.s 1s:3s-drop.s/d.pf
‘I dropped the big shoe’
(10) Tóú- bẐn nen- ót
shoe-big.d/p 1s:3d-drop.s/d.pf
‘I dropped the two big shoes’
(11) Tóú- bẐn gyat- p!ét
shoe-big.d/p 1s:3p-drop.p.pf
‘I dropped the big shoes’
Lastly, before clarifying some concepts relevant to the analysis of such

sentences, consider some examples of agreement∼suppletion mismatches: i-
agreement with the s/d- and p-form of the predicate; and p-agreement with
the s-form, and s-agreement with the p-form, of the predicate.
(12) a. E- k!ÓÓ
1/3I-be lying.S/D
‘It is lying’ or ‘We two are lying’
b. NÓ- k!úl
:1s:3I-be lying.P
‘Mine are lying’
(13) a. YáM - dôi- et
:1s:3P-too-big.S
‘It is too big for me’
b. ∅- SÓl
3S-be sitting.P
‘It is sitting’
These examples may seem to undermine the classification of the predicates

as s/d/p-forms. However, we will see that the mismatches are principled and
follow naturally from the theory of Chapter 3.
4.1.2. Clarification
There now follows a definition of suppletion and a comparison of suppletion,

readjustment and allomorphy, as well as some description of Kiowa phonology
relevant to distinguishing suppletion from forms that are related but by non-
obvious means. Only the definition of suppletion is central to what follows
and even this may be skipped by readers familiar with the concept.
Suppletion, readjustment and allomorphy
‘Suppletion’ has been used to cover three phenomena that probably deserve
explanation by different theoretical mechanisms. English examples are:
(14) a. Past tense morphemes -ed∼∅

b. Present∼past forms sing∼sang
c. Present∼past forms go∼went
Only the last of these is suppletion in the sense relevant here. The other two
differ from it in ways detailed below.
A pair ψ∼ψ is suppletive if the two forms realize the same root in different
grammatical contexts, but are not related by synchronic phonology.
(15) Definition: Suppletion

An item is said to supplete if and only if:
a. it is a root
b. it has (at least) two phonologically distinct exponents, ψ and
ψ , such that:
(i) ψ is conditioned (by, say, tense, number or animacy)
(ii) no phonological process of the language is capable of gen-
erating ψ from ψ (or vice versa).
Consequently, a suppletive item is a vocabulary item of the form below: a

single root with two or more exponents, conditioned by morphosyntactic
features (with the possible exception of an unconditioned elsewhere form).
√
(16) ⇔ ψ / [F]
..
.
⇔ ψ
On this definition, go∼went is suppletive:2

2
I leave aside the issue of whether went is, in fact, the root wend together with the
past tense morpheme of send∼sent, lend∼lent, et cetera.
122 CHAPTER 4
√
(17) go ⇔ went / [+past]
⇔ go
Cases of suppletion should be distinguished from alternations like -ed∼∅

and sing∼sang for the following reasons. Chomsky
√ and Halle (1968/1991)
reject a suppletive analysis for sing∼sang ( sing ⇔ sang / [+past];
⇔ sing). Rather, simplifying slightly, they posit a phonological rule that is
triggered by [+past]. On such an account, only one form of the root is stored.
√
(18) sing ⇔ sing
Chomsky and Halle observe that the phonological processes that derive sang
from sing can be called on elsewhere in the grammar, as in the superficially
unique āi∼æ alternation of satisfy∼satisfaction (pp. 201−202). See also
Yang (1999, 2002) for more recent discussion of consequences of distinguish-
ing go∼went from sing∼sang alternations.
Now consider -ed∼∅, which I regard as allomorphy:
(19) Definition: Allomorphy

An item is said to exhibit allomorphy if and only if:
a. it is not a root
b. it has (at least) two phonologically distinct exponents, ψ and
ψ , such that:
(i) ψ is conditioned (by, say, tense, number or animacy)
(ii) no phonological process of the language is capable of gen-
erating ψ from ψ (or vice versa).
Terminology aside, allomorphy differs from suppletion in exactly one respect:

suppletion is confined to roots, allomorphy to non-roots. Adger, Béjar, and
Harbour (2001) argue that this terminological difference is warranted be-
cause the apparent locality conditions on suppletion and allomorphy differ.
Suppletion requires sisterhood, whereas allomorphy does not. This reflects
the syntactic difference between roots and affixes. Affixes are feature bun-
dles that can enter Agree relations; these Agree relations are non-adjacent
and can provide the conditioning context that licenses ψ as opposed to ψ.
Roots, by contrast, are primitive, and so do not enter into Agree relations;
consequently, they have no access to non-adjacent features. (This difference
between affixal allomorphy and root suppletion is exportable to other frame-
works making a principled distinction between affixes and roots.)
It is clear that the alternations in (1) and (2) could not be allomorphy,
as they concern roots. Moreover, as is clear from the thorough description of
segmental alternations in Harrington (1928) and, especially, Watkins (1984),
(1) and (2) cannot be the result of readjustment rules. Therefore, I conclude
that these are all cases of suppletion.
Suppletion versus phonology
As just emphasized, what counts as suppletion in a given language depends on

that language’s phonology. A skeptic may wonder whether, say, xéŹ ‘set.s/d’
and sÓÓ ‘set.p’ might not be related by, for instance:
(20) xéŹ → séŹ / Context1

séŹ → sÓÓ / Context2
To make such a claim, one must have evidence for each step. In (20), the
steps are otherwise unattested and the context unspecified. It should be
noted, however, that the demand for evidence is not setting the standard
impossibly high, for, first, the child must likely have such evidence, and,
second, the standard can be met in some quite daunting cases, as I now
illustrate. (This illustration is not crucial to the analysis of suppletion that
follows and readers without a taste for phonological oddments may wish to
proceed directly to subsequent sections.)
Watkins (1984, p. 164) lists four root∼perfective pairs as suppletive. I
suggest that they are actually phonologically regular.
(21) Gloss Root Perfective

die hẐMiM hêm
disappear yŹŹ yâi
exit, carry out khẐi thép
pull kŹMŹM tém
To explain ‘disappear’, observe that -i∼-ǐ is an infrequent form of the

perfective (the nasal form occurs with nasal vowels):

roast tÓMÓM tÓMŹM
sound póMúM póMŹM
speak, say tóMúM tóMŹM
124 CHAPTER 4
Rightwards spreading of high tone onto -i∼-ǐ and shortening of the root vowel
are phonologically regular (Watkins 1984, Harbour 2002). Second, note the
alternation (y)i∼ya:
(23) Gloss i -Form ya-Form

fire; kindling phŹŹ phyáŹsón
God; Jesus dOOk!ŹŹ dOOk!yaŹŹ
uncle.voc; uncle.name segẐi segyâi
Though not entirely understood, we may hypothesize that a sufficient con-

dition for the alternation is that i be short and followed by another i (an
OCP effect, not to be confused with the orthographically identical long ii).
Last, note that anything following the root yŹŹ has low tone (e.g., yŹŹhOO
‘keep
√ on vanishing’, yŹŹt!OO ‘will vanish’). The perfective now follows from
root+pf, yŹŹ-i:
(24) Concatenation yields: yŹŹ-i

Low tone assignment yields: yŹŹ-ı̀
Shortening of the root vowel yields: yŹ-ı̀
yi → ya before i yields: yá-ı̀
By orthographic convention, V́V̀ is written V̂V. So, the desired yâi results.
To derive the other three forms, several of the same mechanisms are
called on. First, -p∼-m is another infrequent form of the perfective; it, too,
conditions vowel shortening (and, again, the nasal form occurs with nasal
vowels):

drink thóMúM thóm
hit gûu góp
lay.p k!úú k!óp
pour óú óp
recall, be aware of hÓÓ hÓp
Observe the lowering of the high vowel uu in ‘hit’ and ‘lay.p’ in (25). Gen-
eralizing lowering to all high vowels, we derive ‘die.pf’ from hẐˇǐ-m:
(26) Concatenation yields: hẐˇǐ-m

Vowel shortening yields: hẐˇ-m
High vowel lowering yields: hêˇ-m (= hêm)
Recall that velars become dental before e (Section 2.6.2). This switching,
together with processes affecting hẐˇǐ ‘die’, apply to kŹˇŹˇ ‘pull’:
(27) Concatenation yields: kŹˇŹˇ-p

Vowel shortening / lowering yields: kéˇ-p
Velar-to-dental switching yields: téˇ-p
Nasalization yields: téˇ-m (= tém)
Finally, observe that falling tones cannot be realized over the sequence
Vp; hence, the tonal simplification of gûu to gó in (25). Given this, the
perfective of khẐi ‘pull’ is derivable:
(28) Concatenation yields: khẐi-p

Vowel shortening / lowering yields: khê-p
Velar-to-dental switching yields: thê-p
Tone simplification yields: thé-p (= thép)
Returning to (1) and (2), the claim is that none of these pairs is explicable
by synchronic phonology in the manner just illustrated and for that reason
all are related by suppletion, rather than readjustments.3
4.2. Analysis of the Basic Cases
We now analyze the basic cases of Section 4.1.1.
4.2.1. Assumptions
The analysis is based on the theory of allomorphy and suppletion of Adger,

Béjar, and Harbour (2001). The conclusions relevant here are:
(29) a. Suppletion requires adjacency of root to conditioner.

b. Uninterpretable features do not condition suppletion.
These two conclusions tightly constrain the syntactic structure in terms of

which suppletion must be explained. There are three heads capable of bearing
3
The emphasis on ‘synchronic’ is important. Some pairs may have been synchronically
related at an earlier stage of the language. For instance, consider x!óú∼k!úú ‘lay’. X!óú
has a homophone x!óú ‘rock’. Interestingly, ‘rock’ in Rio Grande Tewa is k’u:; compare
also Kiowa x!ól ‘wing’ with Rio Grande Tewa k’un, x!éŹ ‘thick’ with k’a:’i’; but k!ól
‘neck’ with k’é: (Kroskrity 1993, Appendix 1, citing Randall and Anna Speirs, p.c.).
126 CHAPTER 4
number features: Class, Number and D. For an sdp noun, Class is empty.
So, suppletion depends on Number or D. The number features on D are
uninterpretable, however, and so cannot condition suppletion (29b). This
leaves Number, the highest head with interpretable number features, as the
possible source of suppletion conditioning.
The adjacency requirement (29a) requires the following structure:
(30) VP
NumberP V
Class: Noun Number
If so, D cannot immediately dominate Number, as assumed in the Chapter 3.

Instead, a position nearer to that of Sportiche (1997) is required, according to
which D and its complement NP do not begin as complements in the syntax.
Rather, NP begins as the complement of the verb and both are lower than
D; the NP moves to D later, creating a DP. For current purposes, it must be
NumberP that is the complement of V and that moves to D:
(31) DP
.. D
.
u singular
VP u augmented
NumberP V
Class: Noun Number
This structure has the requisite adjacency relations for Number to condition
V-suppletion, as we shall now see.4 If Noun is a mass noun, then Number is
absent and, so, the verb is adjacent to Class. So, for mass nouns, suppletion
will reflect the class features.5
4
Sportiche suggests that Number, like D, is higher than V. It seems to me that his
arguments really address the position of D and are neutral on number.
5
Readers disinclined to accept the foregoing assumptions need not despair. Early pre-
dictions below may be viewed as demonstrating what suppletion and agreement depend
on. Several predictions concerning the specifics of mismatches will still be left.
4.2.2. Analysis
Consider first the general case of an sdp noun. Number, represented in the
abstract form [αsingular βaugmented], merges with the verb:
(32) VP
NumberP V
Class: Noun Number

∅ αsingular
βaugmented
The value α will condition suppletion of the verb if it is [±singular]-sensitive,

or β will if it is [±augmented]-sensitive. D is Merged higher in the structure
and attracts NumberP. (Movement can be implemented via an EPP feature
D, in the framework of Chomsky 2000.) Movement and valuation yield:
(33) DP
NumberP D
αsingular
βaugmented
Class: Noun Number

∅ αsingular
βaugmented
D will trigger agreement that reflects referential cardinality, hence sdp.

As a specific example, consider (4), repeated below.
(34) Tóúdé eM - bẐn

shoe 3d-big.d/p
‘The two shoes are big’
For referential cardinality 2, Number is [−singular −augmented]. The pred-

icate ‘big’ is [±singular]-sensitive:6
6
In (35), as in suppletive vocabulary entries below, both forms have conditioning con-
texts, [+singular] for êl, [−singular] for bẐn. This may be excessive, as one could be the
elsewhere form and so be uncontextualized. I avoid this for two reasons. First, there is no
128 CHAPTER 4
√
(35) big ⇔ êl / [+singular]
⇔ bẐn / [−singular]
In (36), the value of [±singular] is minus and bẐn is inserted.
(36) VP
NumberP V

bẐn
Class: shoe Number

∅ −singular
−augmented
Raising NumberP to D and valuing of D’s uninterpretable number yields:
(37) DP
NumberP D
−singular
−augmented
Class: shoe Number

∅ −singular
−augmented
D triggers d-agreement, pronounced, for a simple intransitive, as ě. The

resulting complex verb, then, is ě-bẐn, as desired.
As an example of the other predicate class, consider (8), repeated below.
(38) Tóúdé gya-k!úl

shoe 3p- be lying.p
‘The shoes are lying’
Number is [−singular +augmented]. The predicate ‘be lying’ is [±augmented]-

sensitive, so its plus form (39) is used (40).
√
(39) be lying ⇔ k!ÓÓ / [−augmented]
⇔ k!úl / [+augmented]
clear
√evidence for which√is the elsewhere form in each case. Second, s might be elsewhere
for big, and d/p for small. However, the predicates of each class pattern alike with
respect to suppletion; so, something more systematic than elsewhere forms is required.
(40) VP
NumberP V

k!úl
Class: shoe Number

∅ −singular
+augmented
Raising NumberP to D and valuing of D’s uninterpretable number yields:
(41) DP
NumberP D
−singular
+augmented
Class: shoe Number

∅ −singular
+augmented
D triggers p-agreement, yielding the complex verb gya-k!úl.
Last, consider (10), repeated below, as an example of adjective with verb.
(42) Tóú- bẐn nen- ót

shoe-big.d/p 1s:3d-drop.s/d.pf
‘I dropped the two big shoes’
The adjective ‘big’ is adjoined to NumberP (Section 3.2.4). So, adjective and
verb are adjacent to Number. By (35) and (43), we have (44) (where, for the
sake of brevity, the whole structure is shown with vocabulary items):
√
(43) drop ⇔ ól / [−augmented]
⇔ p!él / [+augmented]
130 CHAPTER 4
(44) VP
NumberP V

ót
NumberP Adjective

bẐn
Class: shoe Number

tóú- −singular
−augmented
NumberP moves to D, yielding the DP tóúbẐn ‘big shoes’. D is valued as

[−singular −augmented] and triggers d-agreement, yielding the complex verb
nen-ót ‘I dropped them two’.
4.2.3. Summary
The system of agreement and suppletion triggering stated and illustrated

in this section, with suppletion dependent on Number but agreement dep-
endent on D, derives the correct agreement and suppletion forms for sdp
cases. However, it may seem overcomplicated, particularly in its reliance on
Adger, Béjar, and Harbour (2001), which forces suppletion and agreement,
two rather similar phenomena, to depend on different parts of the structure.
For sdp nouns, where the feature content of D is always that of Number,
it would indeed be sufficient for both to look at one. Chapter 3, however,
presented a variety of cases in which Number and D differ. The full variety
of agreement∼suppletion mismatches is summarized below:
(45) Cardinality, Class Agr−Adj.S∼D/P Agr−V.S/D∼P

1 idp, ids, idi i−s i−s/d
1 ppp p−s
1 sss s−p
2 sii i−d/p i−s/d
2 sss s−p
3 sdi, idi i−d/p i−p
3 sds, sss, idi s−d/p s−p
3 sdi a−s a−s/d
4.3. Inversive mismatches
i-agreement is opaque to referential cardinality. It can occur when it is 1

(idp, ids, idi), 2 (sii), and 3 (sii, sdi, idi). Since inverse marking is the
reflex of feature conflict, it is overspecified and cannot condition a unique
suppletive form. We predict, however, that verbs with i-agreement ignore
D and supplete instead according to referential cardinality. The subsections
below consider minus-valued, i.e., non-sii, and plus-valued classes separately.
4.3.1. Minus-valued classes
To illustrate the prediction, consider the idi noun ‘hair’ and the s/d∼p
predicate ‘be lying’ (Watkins 1984, p. 89). For referential cardinality 1, the
noun is inverse marked and triggers i-agreement. However, the predicate, in
its s/d-form, reflects referential cardinality.
(46) MŹMgOO ÓÓdÓ

É e- k!ÓÓ
there hair.inv 3I-be lying.S/d
‘There’s a hair lying there’
For 2, the predicate is still in its s/d-form but bears d-agreement and the
noun is not inverse marked.
(47) Ól éMŹMgOO eM - k!ÓÓ

hair there 3D-be lying.s/D
‘There are two hairs lying there’
For 3, the noun is again inverse marked and triggers i-agreement. Again, the
predicate, now in its p-form, still reflects referential cardinality.
(48) ÓÓdÓ hóldap nÓ- k!úl

hair.inv dress.loc :1s:3I-be lying.P
‘I’ve got some hair on my dress’
As a sample derivation, consider (48). The VP with its NumberP com-

plement is (40), and, by (39), [+augmented] conditions k!úl, as in (38).
When NumberP raises to D, Class, [−singular −augmented], and Number,
[−singular +augmented], create feature conflict on D, yielding inverse mark-
ing on the noun and i-agreement on the verb, as desired.
132 CHAPTER 4
As examples of the s∼d/p predicate type, consider sdi (49) and idp (51)
nouns. (2, essentially the same as for sdp nouns, is omitted.)
(49) NÓÓ-baou éM- ét, né ám-baougO gÓ- syân

1- cat :1s:3S-big.S but 2- cat.INV :2s:3I-small.d/P
‘My cat is big, but your cats are small’
For sdi nouns, such as báou ‘cat’, Class is [−augmented]. In the first clause,
where referential cardinality is 1, Number is [+singular −augmented] and
conditions the s-form of ‘big’ (35). D is valued as [+singular −augmented],
triggering s-agreement. The resulting verb, with agreement for the indi-
rect object (possessor), is éˇ-ét. In the second clause, Number is [−singular
+augmented], which conditions the d/p-form of ‘small’:
√
(50) small ⇔ syÓn / [+singular]
⇔ syân / [−singular]
Class and Number have opposite specifications of [±augmented]. So, the
possessed noun, báougO, is inverse marked and triggers i-agreement. The
resulting verb, with agreement for the possessor, is gÓ-syân, which shows
number-conditioned suppletion despite number-neutral i-agreement.
In contrast to (49), consider (51), with an idp noun.
(51) MŹMgO
É pŹMáádO e- ét, né óŹde gya-syân
this.INV table.INV 3I-big.S but that 3P- small.d/P
‘This table is big, but those tables are small’
Here, class is [−singular]. So, in the first clause, where Number is [+singular
−augmented], the s-form of ‘big’ is conditioned (35). However, Number
and Class clash, so that demonstrative, éˇŹˇgO, and noun, pŹˇáádO, are inverse
marked and the verb bears i-agreement. The resulting e-ét again shows
number-conditioned suppletion despite number-neutral i-agreement. In the
second clause, both agreement and suppletion are transparent to Number:
[−singular +augmented] conditions the d/p-form of the predicate and trig-
gers p-agreement via D. These, with (50), yield gya-syân.
Consider, lastly, an appositively used s∼d/p adjective with the idi noun
álOO ‘apple’. Since we are only concerned with the DP, I simplify the discus-
sion by assuming a structure in which NumberP, with its adjectival adjunct,
has raised to D. The forms below are taken from Wonderly, Gibson, and
Kirk (1954, p. 6, with minor corrections to tone). The adjective suppletes
for number as expected and the DP as a whole is inverse marked exactly on

a par with any idi noun.
(52) álOO- et- tO

apple-big.s-inv
‘a big apple’
(53) álOO- bin

apple-big.d/p
‘two big apples’
(54) álOO- biMMi- dO

apple-big.d/p-inv
‘several big apples’
Consider first ‘a big apple’. Class, [−singular −augmented], and Number,

[+singular −augmented], cause feature conflict on D:
(55) DP
NumberP ⎡ D ⎤
−singular
⎣+singular ⎦
NumberP Adjective −augmented
Class: apple Number

Number conditions êl, the s-form of ‘big’ which, in turn, phonologically

conditions the inverse marker tO on D:
134 CHAPTER 4
(56) DP
NumberP D

tO
NumberP Adjective

êl
Class: apple Number

álOO +singular
−augmented
Standard phonology yields the desired form, álOOettO.
Things are slightly simpler for 2. The valued tree is:
(57) DP
NumberP D
−singular
−augmented
NumberP Adjective
Class: apple Number

Because Class and Number are both [−singular −augmented], no conflict

arises on D. There is no inverse marking. Number conditions bẐn, the
[−singular] form of ‘big’.
(58) DP
NumberP D

∅
NumberP Adjective

bẐn
Class: apple Number

álOO +singular
−augmented
Again, standard phonology yields the desired form, álOObin.
Finally, consider referential cardinality 3. The valued DP is:
(59) DP
NumberP ⎡ D ⎤
−singular
⎣−augmented⎦
NumberP Adjective +augmented
Class: apple Number

This combines aspects of the preceding cases: as for 1, there is feature conflict
on D, though this time for [±augmented]; and, as for 2, [+augmented] on
Number conditions bẐn for ‘big’.
136 CHAPTER 4
(60) DP
NumberP D

dO
NumberP Adjective

bẐn
Class: apple Number

álOO +singular
−augmented
Again, standard phonology yields the desired form, álOObǐǐdO.

So, the correct forms are easily derivable for all relevant noun classes
under varying values of Number and in a variety of syntactic constructions.
4.3.2. Plus-valued class
Consider now the [+singular] class sii, the only member of which is the first
person, [+author]. Recall from Chapter 3 that a valued first person DP for
referential cardinality 2 or 3 has the following structure:
(61) DP
NumberP ⎡ D ⎤
+author
⎢−singular ⎥
⎢ ⎥
⎢+singular ⎥
⎡Class: I (you)⎤ Number ⎢ ⎥
+author −singular ⎢±augmented ⎥
⎢ ⎥
⎢+singular ⎥ ±augmented ⎣(+hearer) ⎦
⎢ ⎥
⎣(±hearer) ⎦ (−augmented)
(−augmented)
This corresponds to first person inclusive if [+hearer] is included (and with

it [−augmented]).
The important point to observe in (61) is that Number distinguishes 2
from 3, but agreement does not. This differs from, say, sdi and idp nouns,
where there are distinct agreement types for each referential cardinality. It
also differs from idi, however. Both idi and sii trigger i-agreement for two
values of referential cardinality. However, for idi, these values are 1 and 3,
a non-natural class for suppletive predicates. For sii, they are 2 and 3, the
natural class defined by [−singular]. One might imagine, therefore, that sii
simply lacks specification for [±augmented]. Given the preceding discussion,
we can use suppletion to show that this is not so: [±augmented]-sensitive
predicates distinguish first person dual from first person plural.
For the first person exclusive, dual and plural are distinguished as in (62);
(63) gives the relevant vocabulary items:
(62) a. E- x!óŹgyá
1i-fall.s/d.pf
‘She and I fell’
b. E- k!úŹgyá
1i-fall.p.pf
‘They and I fell’
√
(63) a. lay ⇔ x!óú / [−augmented]
⇔ k!úú / [+augmented]
b. detr ⇔ -Źgyá
Similarly, for first person inclusive:
(64) a. Ba-thóúya
2p- move.s/d.impf
‘You and I are moving around’ (Watkins, p.c.)
b. Ba-zéMŹMma
2p- move.p.impf
‘You, I and (s)he/they are moving around’ (Watkins, p.c.)
√
(65) wander ⇔ thóú / [−augmented]
⇔ zéˇŹˇ / [+augmented]
Summary
The DP structure of Chapter 3, by permitting divergence between the content

of Number and D, provides the correct features to derive the agreement∼sup-
pletion mismatches that arise with cardinally opaque i-agreement.
138 CHAPTER 4
4.4. Group-induced mismatches
Further agreement∼suppletion mismatches are predicted to arise in virtue

of [±group]. This follows because, if Class and Number bear [+augmented]
and [αgroup], then D bears [ᾱaugmented], even if Number bears the oppo-
site value [αaugmented]. We examine first the sds, ids and idi cases, then
sss, then [+group] mass nouns. Although pluralia tantum nouns, ppp, fall
naturally into this discussion, being [−group], they are deferred until Section
4.5, as their behavior is more complicated than [+group] nouns’.
4.4.1. Collectives
For countable [+group] nouns, agreement∼suppletion mismatches are pre-

dicted to occur for referential cardinality 3, where the DP has the structure:7
(66) DP
NumberP D
+singular
−augmented
Class: Noun Number

+group −singular
+augmented
Number and D are oppositely specified for [±singular] and [±augmented].

Therefore, for s∼d/p predicates, like ‘big’, we predict s-agreement on the
d/p-form; and for s/d∼p predicates, like ‘be set’, s-agreement on the p-
form. These combinations are exemplified below.
First, consider the s/d∼p suppletive predicate ‘be set’. This is illustrated
with the sds noun, tóú ‘house’ (Watkins 1984, p. 90).
(67) Tóú ∅- xél

house 3S-be set.S/d
‘There’s a house standing’
(68) Tóú eM - xél
house 3D-be set.s/D
‘There are two houses standing’
7
I abstract away from the possibility of postsyntactic feature insertion (Section 3.4.2).
(69) Tóú ∅- sÓl

house 3S-be set.P
‘There are houses standing’
In the first two sentences, agreement and suppletion both transparently re-
flect referential cardinality. However, in (69), the p-form bears s-agreement.
This is exactly as expected given (66) and the vocabulary items:
√
(70) a. be set ⇔ xéŹ / [−augmented]
⇔ sÓÓ / [+augmented]
b. stat ⇔ -l
(71) VP
NumberP V

sÓl
Class: Noun Number

+group −singular
+augmented
The [+augmented] on Number licenses sÓl, but D, [+singular −augmented],

triggers s-agreement, yielding (70).
The same point is made for ids nouns by (Watkins 1984, p. 87):
(72) ÁádO e- xél

tree.inv 3i-be set.s/d
‘A tree is standing / growing’
(73) Áá eM- xél
tree 3d-be set.s/d
‘Two trees are standing / growing’
(74) Áá ∅- sÓl
tree 3s-be set.p
‘Trees are standing / growing’
As details are as discussed above for sds nouns and for the inverse, these
examples require no further comment.
Consider, now, s/d∼p suppletive predicates. These are illustrated with
idi nouns on their ‘different types of’ reading.
140 CHAPTER 4
(75) HÓndé Ól bÓ- kŹMŹMnŹŹ /*kyóMŹM

what hair :2p:3S-long.d/P/ long.s
‘What long [types of] hair you all have’
(76) HÓndé Ól bÓ- xáádóú /*xéŹ
what hair :2p:3S-short.d/P/ short.s
‘What short [types of] hair you all have’
In each case, ‘hair’ triggers s-agreement. However, the s-form of the predi-
cates is unacceptable and the d/p-form must be used. Again, this is exactly
as expected given (66) and the vocabulary items:
√
(77) long ⇔ kŹˇŹˇnŹŹ / [+singular]
⇔ kyó ˇŹˇ / [−singular]
√
(78) short ⇔ xáádóú / [+singular]
⇔ xéŹ / [−singular]
Thus, nouns that trigger s-agreement when referential cardinality is 3
trigger suppletion as expected.
4.4.2. Mass nouns
Thus far, all mismatches have been cases in which suppletion ‘sees through’
agreement to true referential cardinality. In this regard, non-granular mass
nouns are very interesting. If their uncountability indicates absence of Num-
ber (Section 3.5), then suppletion in such cases cannot depend on Num-
ber. Nonetheless, it can depend on number features, as these are present
on Class. So, for mass nouns, Class-conditioned suppletion, rather than
Number-conditioned suppletion is expected, in virtue of the structure:
(79) VP
Class: Mass V
Given (79), the feature composition of Class is of key importance for mass
nouns, especially sss. Generally, s-agreement arises because D dominates
only [+singular] and [−augmented]. However, I argued that the classification
[+singular −augmented] makes no sense for mass nouns; rather, they are
[−singular +augmented +group]. The correctness of this argument can now
be verified by the predictions it makes with respect to suppletion.
With sss nouns, [±augmented]-sensitive predicates occur their in p-form.
(80) Étté /syÓndé thóMúM ∅- sÓl dOálkya

much/little water 3s-be sitting.P bucket.loc
‘There’s much / little water in the bucket’
(81) SyÓndé thóMúM ∅- p!étkyá /*ótkyá
little water 3s-fall.P.detr.pf/ fall.s/d.detr.pf
‘A little / a drop of water fell’
(82) K!Oâlkya étté pénhaa ∅- sÓl /*xél
dish.loc much sugar 3s-be sitting.P/ be sitting.s/d
‘There’s a lot of sugar in the dish’ (suggestive of a bowl)
(83) K!Oâlkya étté pénhaa ∅- k!úú /*k!ÓÓ
dish.loc much sugar 3s-be lying.P/ be lying.s/d
‘There’s a lot of sugar on the dish’ (suggestive of a plate)
Evidence for how sss mass nouns behave with [±singular]-sensitive pred-
icates is harder to come by, owing to the meanings of these predicates. Sen-
tences such as ‘The sugar is short’ or ‘The whisky is long’ do not make much
sense, and ‘The water is large’ or ‘The snow is small’, to the extent that they
are interpretable, are so in a non-mass sense, such as ‘body of water’ or ‘snow
flake’. The nearest I have come to a suitable example is ÓlhǑǑsyan ‘dime’,
from ‘money’+‘small’. This seems to support the idea that ‘money’, and
so sss nouns, are [−singular]. However, here, the language tricks us, for, of
the four s∼d/p predicates, precisely ‘small’ ceases to be [±singular]-sensitive
when attributive: syân is used in all cases (Watkins 1984, p. 99).
For completeness, observe that ppp mass nouns also occur only with
the p-form of predicates. This is expected given the inherent classification
[−singular +augmented]. (Again, for semantic reasons, I have no examples
of [±singular]-sensitive predicates with these nouns.)
(84) ÓÓtháM t!aM Mi

pŹMááthai gya-sÓl
salt table.loc 3p- be sitting.P
‘There’s salt on the table’
(85) K!ÓMÓMetOM OM étté ÓÓtháM t!aM Mi gya-sÓl
frybread much salt 3p- be sitting.P
‘There’s a lot of salt in the frybread’
142 CHAPTER 4
(86) PéŹgya toudôm yáM - sÓl

sand floor.loc :1s:3p-be sitting.P
‘I’ve gotten sand on the floor’
The examples of mass nouns show that it is irrelevant to suppletion

whether number features are on Number or Class, and that the sss class
features [−singular +augmented +group], motivated on semantic grounds,
correctly predict [+group]-induced suppletion∼agreement mismatches paral-
lel to those of sds, ids and idi nouns.
4.5. Harder cases
The agreement∼suppletion mismatches induced by inverse and grouphood

fall under a single generalization: that the information relevant to supple-
tion is the feature content of Number, or Class, when Number is absent.
This result has been established above for the vast majority of nouns in the
language, sdp, sdi, sii, idp, idi, ids, sds, sss, and ppp mass. In this sec-
tion, we examine two corners of the language where the generalization that
governs the previous cases breaks down: pluralia tantum and reflexives.
4.5.1. Pluralia-tantum-induced mismatches
Pluralia tantum nouns, such as hólda ‘dress’, khÓÓdé ‘trousers’, tóú ‘teepee’,
ˇkút ‘book, letter’, and t!ó
tó ˇ ú
ˇ gya ‘shirt’, are ppp, [+augmented −group].
Their behavior with respect to suppletion only partly conforms to the gener-
alization that suppletion depends on the feature content of Number, or Class,
when Number is absent: [±singular]-sensitive predicates obey the generaliza-
tion, [±augmented]-sensitive ones do not, displaying the p-form throughout.
Consider, first, a [±singular]-sensitive predicate.
(87) MŹMde t!óMúM gya yáM -

É dôi- et
this shirt :1s:3P-too-big.S
‘This shirt is too big for me’
Here, referential cardinality is 1. Number is [+singular −augmented] and, so,

conditions the s-form of ‘big’ (35). However, the class features [+augmented
−group] cause D to be valued as [−singular +augmented], triggering p-
agreement. So, the mismatch is correctly predicted. Now consider:
(88) Hólda yáM - dôi- bin

dress :1s:3P-too-big.D/P
‘My dresses are too big for me’
For referential cardinalities 2−3, Number is [−singular ±augmented]. Con-

sequently, [±singular]-sensitive predicates occur in their d/p-form, matching
p-agreement, as expected. Similar examples are:
(89) MŹMde k!ókÓŹouphOl gya-dôi- kyoM Mi /xei

É
this roach 3P- too-long.S/short.S
‘This roach is too long/short’
(90) TéŹ gya-dôi- kiMMinii /xaadou
all 3P- too-long.D/P/short.D/P
‘They [roaches] are all too long/short’
More complicated are [±augmented]-sensitive predicates, like ‘lay’. The

predicate is in its p-form, irrespective of referential cardinality.
(91) T!óMúM gya gyat- p!ÓŹ-k!op

shirt 1s:3P-mis-lay.P.pf
‘I lost a shirt / some shirts’
Though this is predicted for referential cardinality 3, it is entirely unexpected

for 1 and 2. For these, Number is [±singular −augmented] and so, given (63),
x!óú should occur. However, it is in fact ungrammatical.
(92) *T!óMúM gya gyat- p!ÓŹ-x!ep

shirt 1s:3P-mis-lay.S/D.pf
‘I lost one/two shirt(s)’
Similar examples are:
(93) K!ókÓŹouphOl yáM - p!étkyá /*ótkyá

roach :1s:3P-fall.P.detr.pf/ fall.S/d.detr.pf
‘My roach fell off’
(94) TéŹ k!ókÓŹouphOl gya- p!étkyá
all roach :3a:3P-fall.P.detr.pf
‘Everyone’s roaches fell off’
The constant occurrence of p-forms contradicts the notion of Number-

144 CHAPTER 4
sensitivity. A near solution is possible, however, if we take Number to be

partially defective, deprived of [±augmented], for these nouns.
(95) NumberP
Class: Pluralia Tantum

Number

−group
The features in (95) are the correct ones to condition suppletion: [±singular]-
sensitive predicates will covary with the specification of that feature on
Number, whereas [±augmented]-sensitive predicates will have access only
to [+augmented] on Class. Two issues require comment:
First, there is the issue of the defectiveness of Number. Carstens (1991)
argues the content of Number varies crosslinguistically (see Harbour 2006a for
a detailed proposal): English Number bears just [±singular], as the language
contrasts just singular∼non-singular, rather than singular∼dual∼plural, as
Kiowa does. So, to posit (95) is to claim that crosslinguistic variation can
be replicated language internally. Split ergativity can be considered in this
way: some languages are ergative, others accusative, and others exhibit the
variation internally. More immediately relevant is the total defectiveness of
Number for mass nouns. As some languages have totally defective Number
for all nouns (my 2006a treatment of Pirahã, Everett 1986), Kiowa already
presents a case of crosslinguistic variation language internally. So, I regard
the partial defectiveness in (95) as unproblematic.
Second, there is the issue of sisterhood. The verb, if sister to Number, is
not sister to Class, and so Class cannot condition verb suppletion. However,
as Class and Number are not in a mutual dominance relation, it is possible
to define the notion of sisterhood, in terms of dominance, so that Class can
condition verb suppletion (possibly, just in cases where Number is unspecified
for the relevant feature). Whether this redefinition is insightful or mere
technical opportunism, however, lies beyond the scope of the current work.
It is, I think, fair to conclude that the suppletion conditioned by pluralia
tantum nouns constitutes a slightly harder case than those examined above,
but that it is not beyond analysis.
4.5.2. Animate- and reflexive-induced mismatches
Animates and reflexives constitute a second source of unexpected agreement∼

suppletion mismatches, conditioning suppletion as though of referential cardi-
nality 1, yet agreeing, and being interpreted, as though 2 or, more frequently,
3. However, their variable agreement behavior reveals a feature distribution
(108) consistent with the theory and mechanisms above, even if somewhat
semantically surprising.
Animate plural agreement, ‘a’, is triggered by animate nouns, of referen-
tial cardinality 3, that are subject to ‘empathy’ (Section 2.5); so, it is near
obligatory with KÓŹgú ‘Kiowas’, common with k!yá ˇ á
ˇ hyóp ‘men’, maayóp
‘women’, xêˇǐgO ‘horses’, and rare with children and lower animates.
Semantically, the featural correlate of ‘a’ is expected to be [−singular
+augmented +empathic] (where the last feature is the correlate of empa-
thy). However, animates condition suppletion exactly conversely, as though
[+singular −augmented]. So, [±singular]-sensitive predicates occur in their
s-form (96), and [±augmented]-sensitive ones, in their s/d-form (97):
(96) a. KÓŹ- k!yakoM uM bO á- kyóMŹM

Kiowa-people.inv 3A-tall.S
‘Kiowas are tall’
b. KÓŹ- k!yakoM uM bO á- khóú- xei
Kiowa-people.inv 3A-body-short.S
‘Kiowas are short’
(97) a. Á- péŹ- x!óŹgyá
3A-dead-land.S/d.pf
‘They fell down dead’
b. KÓŹgú de- deM Mi- x!ép
Kiowa.inv 1s:3A-sleep-lay.S/d.pf
‘I laid the Kiowas down to sleep’
Animate plurals share this suppletive pattern with reflexives. For in-
stance, there are three expressions built on suppletive predicates and de-
manding reflexive agreement: hóút!al ‘part ways’ (travel-sever.s/d), hóúol
‘turn off’ (travel-drop.s/d), and hónx!oigya ‘come late’ (last-land.s/d):8
8
Hónx!oigya ‘come late’ is a dative reflexive, with the reflexive triggering, generally,
i-agreement. See pp. 146ff.
146 CHAPTER 4
(98) ∅- SÓOM OM dehel gO . . . téphOi em- tóúdO gO hegÓ

3s-angry.hsy and everyone 3s:3a-gather.pf and then
ém- hóú- t!alhel
3a:3A-travel-sever.S/d.pf.hsy
‘He became angry and . . . he gathered everyone [his followers] to-
gether they parted ways [from the others]’ (Harrington 1928, p. 252)
(99) Ét- hóú- ot gO e- p!Ôi
1i:3A-travel-drop.S/d.pf and 1i-lose.pf
‘We turned off and got lost’
(100) HÓn dÓ- sÓŹ- Om-then- dOM OM mOO-do dÓt- hón-x!oigya
neg :1p:3s-fast-do- heart-be.neg- because :1p:3A-last- fall.S/d.pf
‘We arrived late because we didn’t want to drive fast’
In all of the preceding, the subjects are plural (‘everyone’, ‘we’). However,
the predicates all appear in their s/d-forms. Indeed, [+augmented]-forms,
like *hóúthaa and *hónk!uigya, are explicitly rejected. Consistently, if the
subject is singular, the predicate appears in s/d-form, as in:
(101) Hágyá- xo de- hóú- ot

somewhere-instead 1s:3A-travel-drop.S/d.pf
‘I turned off somewhere’
It is to some extent unsurprising reflexives and animate plurals should act

as a class for suppletion, as they form a class for other purposes, too. Most
basically, reflexive agreement simply is animate plural agreement, resulting
in the systematic ambiguity recorded by Harrington (1928); for example:
(102) De- hól

1s:3A-kill.pf
‘I killed myself or them (people)’
The one agreement type is ambiguous, for all person-number combinations

between acting reflexively and acting on a group of empathic animates.9
9
They can be disambiguated with ÓˇÓˇgO ‘self’ (i); (ii) and (iii) show other uses of ÓˇÓˇ gO,
emphatic and adverbial.
(i) ÓMÓMgO de- hól

self 1s:3a-kill.pf
‘I killed myself / *them’
Also of importance here is the behavior of a-agreement as the direct ob-

ject of a ditransitive, as it gives significant insight into the category’s feature
composition. In this case, there is no special exponent of a-agreement, in
contrast to (in)transitives. Instead, such cases exhibit a d∼i alternation.
Consider, first, animate plurals. For the combination of agent and bene-
factive below, two agreement prefixes are possible, né (103a) and nÓ (104a).
The former is also used when the direct object is dual (103b) and, so, is d-
agreement; and the latter is i-agreement, as it is used when the direct object
is inverse marked (104b).
(103) a. Né- hól

2s:1s:3A-kill.pf
‘You killed them (people) for me’
b. YŹŹ né- hól
two 2s:1s:3D-kill.pf
‘You killed two for me’
(104) a. TéŹ k!yáM áM hyóp nÓ- háŹgyádOO
all man.inv :1s:3A-know
‘I know all the men’
b. ZéŹbOt nÓ- p!Ôi
arrow.inv :1s:3I-lose.pf
‘I lost my arrow.’
Reflexives, too, show d∼i variation in the presence of indirect object

agreement. Consider the class of dative reflexives (105): (106) shows the
d-agreement of (103), (107) the i-agreement of (104) (though, for the most
part, each predicate occurs either with d or with i).
(ii) ÓMÓMgO a- xán

self 1s-arrive.pf
‘I arrived’
(iii) ÓMÓMgO em- tóM úM - hÔn

self 3s:3a-talk-exhaust.pf
‘He became silent of his own accord’ (Harrington 1928, p. 29)
148 CHAPTER 4
(105) Gloss Root

awake án
be fittingly dressed dólbé
come late hónx!oide
have one’s heart skip khútdOO
look nice dótkyai
need ÓtkhyádOO
trip OnyŹŹ
(106) HÓ né- dólbé?

q :2s:3A-be fittingly dressed
‘Am I fittingly dressed?’
(107) NÓ- hónx!oigya

:1s:3I-come late.pf
‘I came late’
So, again, animate plurals and reflexives behave as a natural class with
respect to a morphosyntactic process, the d∼i-object-agreement alternation
conditioned by indirect object agreement.
The d∼i-alternation gives a strong clue as to the feature composition of
a-agreement. Suppose that Number is [+singular −augmented], as supple-
tion suggests, and, additionally, that Class is [−singular +empathic], where
[±empathic] is the feature that distinguishes sda from sdi. Valuation of D
involves replication of all features:
(108) DP
NumberP ⎡ D ⎤
−singular
⎢+singular ⎥
⎢ ⎥
Class: Animate ⎣−augmented⎦
Number
−singular +singular +empathic
+empathic −augmented
Immediately, this creates a feature conflict on D, which causes inverse mark-

ing on the noun, as in (96)−(97), and on related nominal heads (see Section
3.2.4), such as relative clauses (109) and demonstratives (110):
(109) [háote á- dÓÓ]-gO

several 3A-be -INV
‘the few of them’
(110) ÓŹhyOO- gO háyá ém- hóú- OM OM zonhel
that one-INV somewhere 3A:3a-travel-start off.pf.hsy
‘Those ones traveled off somewhere’ (Harrington 1928, p. 252)
I suggest that D in (108) triggers a/d/i-agreement as follows. For all

subjects, and for objects of transitives, there are particular vocabulary items
for feature bundles containing [+empathic]. These are what is labeled a-
agreement. For objects of ditransitives, there are no special vocabulary items.
So, the next most highly specified, those for [−F +F], are used, resulting in
i-agreement. However, as Watkins observes, these forms have slightly disre-
spectful for connotations when used for adults. If [+singular] is impoverished
here, as a grammaticalized means to avoid disrespect, the remaining features
[−singular −augmented +empathic] result in d-agreement. (One can, per-
haps, rationalize impoverishment of [+singular] over [−singular] by observing
that, if the latter cooccurs with [+empathic] on Class, then the former is the
more marked value in the context of [+empathic]; so, impoverishment of
[+singular] is impoverishment of the marked.)
So, the surprising s∼s/d suppletion pattern of animate plurals and reflex-
ives is amenable to treatment within the current framework. This requires
a slightly enlarged feature inventory; however, all accounts will need some
parallel posit. The only point requiring clarification, then, is the effect of
[+empathic] on Class and Number. Given that the feature only ever quali-
fies pluralities, its occurrence with [−singular] on Class is natural. (In fact,
one could go further and suppose that Class is fully plural in these cases,
bearing [−singular +augmented +empathic]. The treatment above requires
only one change, that impoverishment affect [+singular +augmented], rather
than just [+singular].) The effect of [+empathic] on Number strikes me as
genuinely mysterious, and, at this stage, I leave the matter open.
4.6. Conclusion
Agreement∼suppletion mismatches are a challenge for any theory of number.

Nonetheless, I believe that the theory of Chapter 3 meets this challenge more
than satisfactorily. It distributes number features throughout three locations,
Class, Number and D, and claims that each can diverge from the others in
150 CHAPTER 4
systematic ways. Independent theories of agreement and of suppletion ensure

that agreement is sensitive to the content of D and suppletion, to Number
or Class. This correctly predicts nearly all mismatches. Of the remainder,
pluralia tantum nouns with [±augmented]-sensitive predicates are explicable
if Number is partial defective, and animate plurals are accommodable given a
natural specification for Class and a somewhat surprising one for Number.10
4.7. Appendix: Adverbs built on suppletive roots
When the two classes of suppletive predicates were introduced, it was noted
that the [±singular]-sensitive ones are individual level and the [±augmented]-
sensitive ones, stage level. It is interesting to wonder whether this is sem-
antically principled.11 At present, I am not prepared to argue to matter
either way. However, I wish to note some properties of [±singular]-sensitive
predicates, which are interesting both in themselves, and in relation to the
possibility of a semantically principled connection between predication level
and suppletive sensitivity. The facts concern semantic differences between
adverbs built on [±singular]-sensitive roots.
Kiowa forms adverbs by affixation of -de∼-te (Watkins 1984, p. 185) and
all roots in (111) form adverbs:
(111) Root Gloss Adverb

big.s a lot étté
big.d/p a lot, much bẐnde
long.s a long time kyóMŹMde
long.d/p a long time kŹMŹMnŹŹte
short.s a short time xéŹde
short.d/p a short time xáádóúte
small.s a little syÓnde
small.d/p a little syânde
The glosses suggest that the meanings of each s∼d/p adverb pair are iden-
tical. However, this is not so.
10
Agreement∼suppletion mismatches are not confined to Kiowa. Hale (1975) discusses
the phenomenon in Navajo. Straightforward application there of the theory developed
here is hampered by agreement and suppletion’s interaction with ‘conjunct movement’, a
process that itself requires clarification. A number of others are detailed in Corbett 2000.
11
It would be a gross abuse of terminology to think that [±singular]-sensitivity and
individual level predication are connected because individuals are inherently singular.
To be sure, there are cases where the members of each pair are inter-
changeable. (Dôi ‘too’ in (113) is irrelevant; compare (123)−(124).)
(112) KyóMŹMdé /kŹMŹMnŹŹte bé- tháá

long.s.adv/long.d/p.adv 2s:3i-sever.p.imp
‘Cut long [pieces]’
(113) KŹMŹMnŹŹte /dôi- kyoM Mide an án- tóMzánma
long.d/p.adv/too-long.s.adv hab :3s:3p-talk.impf
‘She speaks (too) long’
(114) ThaM aM te étté /bẐnde gya-
maternal grandmother.name big.s.adv/big.d/p.adv 3s:3p-
éŹ- k!óp
seed-lay.p.pf
‘Grandma planted a lot’
Such interchangeability is independent of the referential cardinality of the

nouns involved: one can crosscut s∼p ‘short’ with 1∼3 ‘tree’:
(115) ÁádO xéŹde bé- t!âl

tree.inv short.S.adv 2s:3i-sever.S/D.imp
‘Cut the tree short’
(116) ÓŹde áá xéŹde a- tháá
that tree short.S.adv 2s:3s-sever.P.imp
‘Cut those trees short’
(117) ÓŹgO áádO xáádóúte bé- t!âl
that.inv tree.inv short.D/P.adv 2s:3i-sever.S/D.imp
‘Cut that tree short’
(118) ÓŹde áá xáádóúte a- tháá
that tree short.D/P.adv 2s:3s-sever.P.imp
‘Cut those trees short’
However, interchangeability is not generally the case. Consider first (119),

where both forms of ‘big’ are simultaneously attested. Discussion of similar
examples with two speakers made it clear that the s-form étté has quan-
tificational force, ‘many’, and the d/p-form bẐnde, secondary predicational
force, describing the pies. Compare also the two examples that follow.
152 CHAPTER 4
(119) ÉŹt!áppây bẐnde étté gyat- ÓMÓMméŹ

pie big.d/p.adv big.s.adv 1s:3p-make.pf
‘I made lots of pies big’
(120) BẐnde dé- tháttetOO
big.d/p.adv 1s:3i-break up.fut
‘I’m going to cut it into big pieces’
(121) Étté dé- tháttetOO
big.s.adv 1s:3i-break up.fut
‘I’m going to cut it into many pieces’
An initial conclusion here might be that bẐnde simply is not quantificational.

However, this is incorrect. It has at least two quantificational senses.
First, Watkins (personal communication) observes that, in her exam-
ples, bẐnde is quantificational with respect to food, as in (122). The same
holds true for examples in my fieldnotes. Representative examples are food-
oriented (123), where bẐnde is acceptable, versus non-food-oriented (124),
where it is not. (On the limits of food-orientation, see (114).)
(122) XêMiM hégÓ gyá- hágyá- ton k!Ot dôi- binde a-

horse already :2s:3s-already-fat conj too-big.d/p.adv 2s:3s-
máágop
feed.impf
‘You already have the horse fat and yet you feed it too much’
(Watkins 1984, p. 241)
(123) Háote an gya- pÓttO?
how much hab 3s:3p-eat.impf
Étté /bẐnde /dôiette /dôibinde
big.s.adv/big.d/p.adv/too.big.s.adv/too.big.d/p.adv
‘How much does he usually eat?’ ‘Too much’
(124) Háot an gya- sÓÓtétOO?
how much hab 3s:3p-work.impf
Étté /*bẐnde /dôiette /*dôibinde
big.s.adv/ big.d/p.adv/too.big.s.adv/ too.big.d/p.adv
‘How much does he usually eat?’ ‘Too much’
Second, bẐnde is used for some types of event quantification. BẐnde in

(125) emphasizes repeated gain of lots of money; I surmise that étté (126)
is more stative, implying constant wealth.
(125) BẐnde ÓlhÓMÓMgya gya- dôu

big.D/P.adv money 1s:3s-hold
‘I have a lot of money [on different occasions]’
(126) Étté ÓlhÓMÓMgya gya- dôu
big.S.adv money 1s:3s-hold
‘I have a lot of money [generally]’
A similar difference arises for other pairs in (111). For instance, for
‘small’, the [−singular] adverb syânde means ‘a little at a time’, but the
[+singular] adverb syÓnde simply ‘a little’.
(127) MŹMde syânde

É éM- máágop gO a- sÓOM OM detOO
this small.D/P.adv 3s:1s-feed.impf and 1s-anger
‘She’s serving me a little at a time and I’m getting mad’
(128) SyÓnde éM- ôu
small.S.adv 2s:1s:3s-pour.imp
‘Pour me a litte’
Similarly, for ‘long’, the d/p-form is associated with iterativity, the s-form
with a simple event (which can only be interpreted iteratively owing to the
habitual particle an).
(129) KŹMŹMnŹŹte an em- sôugu

long.D/P.adv hab 3s:3a-sew.impf
‘She sews with great distance between stitches’
(130) KyóMŹMde an em- sôugu
long.S.adv hab 3s:3a-sew.impf
‘She sews for a long time’
The generalization to emerge from these examples is that adverbs built

on [±singular]-sensitive predicates are semantically equivalent only in some
cases. Where they differ, the [+singular] variant is nearer to object quantifi-
cation and the [−singular] variant, to event quantification. However, the ex-
act domains of quantification will have to be defined with care, as [+singular]
adverbs sometimes do not quantify over individuals (130), nor [−singular]
ones over events (114). Nonetheless, the generalizations are suggestive and
154 CHAPTER 4
may well impact on any attempt to make suppletive sensitivity a matter of

predicate semantics. If such account is possible, it will provide an interesting
source of insight into the meaning and uses of number features.
PRO
In the more informal context of this appendix, I wish to register one final
speculation: examples such as (115)−(118) may indicate that PRO can
have any number specification in Kiowa (Norvin Richards, p.c.). If sec-
ondary predicates, then they have a more articulated internal structure than
simple adverbs. Now, Watkins (1984, p. 203) observes that adverbial ‘-dé
looks suspiciously like nominal dé’. As this is also the suffix for relative
clauses (Section 3.2.4) adverbs might be highly reduced, agreementless rel-
ative clauses (cf., the same suggestion for demonstratives in Section 3.2.4).
Their meaning would then be something like ‘being big’, and, being agree-
mentless, they would license only PRO. Interchangeability of [±singular]-
sensitive predicates would then indicate that PRO can be specified with
either value of the feature.
Verb incorporation provides similar evidence for [±augmented]-sensitive
predicates. Recall the examples of interchangeability and crosscutting from
p. 30 (the slightly different translations for (131)−(132) and (133)−(134) are
artifacts of the elicitation sessions):
(131) HÔndé a- ól- x!ep?

what 2s:3s-drop.S/D-set.S/D.pf
‘What did you drop?’
(132) Étté gyat- ól- k!op?
much 1s:3p-drop.S/D-set.P.pf
‘I dropped a lot’
(133) p!él- x!ep
drop.P-set.S/D
‘knock off [e.g.: from a table]’
(134) p!él- k!op
drop.P-set.P
‘knock off [e.g.: from a table]’
The incorporated predicates show the same indifference to number as

the adverbials in (115)−(118). Again, if incorporation is analogous to an
infinitival or manner complement in English, then they contain PRO, which
is, again, apparently arbitrarily variable in number.
Such facts lie well beyond the scope of this study and so I merely note
them without offering further analysis.
Chapter 5
The Agreement Prefix
This chapter addresses one of the central problems of Kiowa (and Tanoan)
linguistics—the structure and content of the verbal agreement prefix—in
terms of the theory developed in the preceding chapters.
The classic problem is simply put. In (1), the form of the agreement pre-
fix depends on three arguments: the third person animate plural sender, the
third person animate plural recipient, and the third person singular sendee.
The reality of this dependence can be shown by changing any of the argu-
ments, say, to third person dual, and observing that the prefix covaries with
each change (2)−(4). (‘∗’ indicates that the prefix lowers the following verb’s
tone.) With the agreement prefixes encoding information about three par-
ticipants in as little as a single vowel, the question is how so much meaning
gets into so little sound.
(1) Á∗- tot

3a:3a:3s-send.pf
‘They sent him to them (people)’
(2) M ∗-
É tot
3D:3a:3s-send.pf
‘They two sent him to them (people)’
(3) Mé- tót
3a:3D:3s-send.pf
‘They sent him to them two’
(4) Et- tót
3a:3a:3D-send.pf
157
158 CHAPTER 5
‘They sent them two to them (people)’
Related to this is the size of the inventory of agreement prefixes. Given,

first, that Kiowa distinguishes thirteen person-number combinations (four
persons, three numbers, inverse, animate plural), second, that these may be
external arguments, indirect objects or direct objects, and, third, that verbs
may be ditransitive, transitive, intransitive, and intransitive with dative, the
total number of agreement combinations is naively expected to be 2, 548.
(5) Total number of argument combinations

= 3n=1 total number of n-argument combinations
n)
= 3n=1 (3+(−1)
2
× 13n
= 2, 548
Clearly, some of these 2, 548 are excluded by person-case restrictions (*‘I

will bring you to him’, Adger and Harbour 2007) and binding conditions
(*1s:1d). However, these considerations are insufficient to bring the number
of prefixes down to the mere 160 or so that actually exist. (The exact number
depends on how one counts certain homophones). So, the question is how so
few agreement prefixes encode so many argument combinations.
These questions are relevant to the current investigation because they
permit a justification of the feature inventory and mechanisms that value D
argued for in Chapter 3. That is, if agreement replicates D, then all properties
of the agreement prefix must be explicable in terms of the feature content of D
argued for above: [±singular], [±augmented], and person/empathy features.
Structure of the analysis
Answering to these questions consists in providing a syntax-to-phonology

mapping that transforms bundles of syntactic features into the phonologically
familiar agreement prefixes. That is, we must specify the mechanisms that,
in (1) for instance, produce á∗ from the feature specification of 3a:3a:3s.
⎧ ⎫
⎪
⎪ ⎡ Agent ⎤ ⎡ Ind. Obj. ⎤ Dir. Obj. ⎪ ⎪
⎪
⎪ ⎪
⎪
⎨ −singular −singular ⎬
⎢ ⎥ ⎢ +singular ⎥ +singular −→ á∗
(6) ⎢+singular ⎥ ⎢ ⎥
⎪
⎪
⎪ ⎣−augmented⎦ ⎣−augmented⎦ −augmented ⎪ ⎪
⎪
⎪
⎩ ⎪
⎭
+empathic +empathic
THE AGREEMENT PREFIX 159
I regard an explanation of this sort as consisting in specifying:
(7) a. A structure in which the agreement features of each argument

are located.
b. A list of morphological processes that affect this structure and
the features it contains.
c. A list of correspondences between feature bundles and phonolog-
ical strings (vocabulary items).
d. A list of phonological processes that yield the surface forms.
My analysis builds very much on past ones: Merrifield (1959b), Trager (1960),
Watkins (1984) and later manuscript revisions, Takahashi (1984), Harbour
(2003a). Watkins’ inventory of prefixes is assumed (differences from others’ are
minor), but their distribution is slightly altered to reflect recent discoveries.
A full analysis of the prefix system proceeds in several stages. First, the
number of explicanda is reduced by appeal to morphological operations. On
the one hand, these explain why the inventory of prefixes is substantially
smaller than the number of possible argument combinations. On the other,
they allow one to exclude from active analysis prefixes that are phonologically
predictable from others. Second, the prefixes are decomposed into segments
that are correlated with particular arguments or sets of arguments. Finally, a
list of correlations between phonological strings and morphosyntactic features
(vocabulary items) is given. This accounts for the segmentation and the
regular relationship between sets of prefixes.
Detailed though it is, this chapter is still preliminary to a full treatment: I
leave for future work the synthesis of all exponents below into a single vocab-
ulary list and explicit derivation of the entire inventory (see also Appendix
5.6). Such delimitation is reasonable within the confines of the current inves-
tigation as the Kiowa prefix system is at least as complicated as Catalan’s
clitic clusters, which have been the topics of books in their own right. The
crucial point below is that all generalizations are best stated in terms of the
features proposed above, not in terms of the traditional categories ‘singular’,
‘dual’, and so on. So, the analysis, though preliminary, supports the theory
of the preceding chapters.
Structure of chapter
The analysis below is divided as follows.

160 CHAPTER 5
Section 5.1 presents the full inventory of prefixes and explains the slightly
modified prefix notation used throughout the chapter. In addition, it summa-
rizes various theoretical assumptions, such as prefix structure, and empirical
generalizations, such as phonological alternations.
Section 5.2 reduces the number of prefixes to be derived by appeal to
morphological rules. First, by deletion of whole agreement nodes, the section
shows why the inventory of prefixes is significantly smaller than number of
possible argument combinations. Second, by deletion of features (rather than
of entire nodes), it shows how various syncretisms arise.
Section 5.3 focuses on sets of prefixes that are phonologically predictable
from others. These are important in two respects. Methodologically, pre-
dictable prefixes can be excluded form active consideration. Theoretically,
they show that combinations of morphosyntactic features from Chapter 3
have constant phonological realizations. Together, Sections 5.2−5.3 reduce
the number of prefixes to be derived from 160 to 64.
Section 5.4 presents the segmentation of the prefixes, concentrating on
syntactically natural classes, such as ditransitive prefixes, transitive prefixes
with non-singular agents, with singular agents, and so on. The methodolog-
ical and theoretical points of the previous paragraph apply again here.
The broader view: Morphosemantic number
Micromorphology is not to everyone’s taste. However, it must be emphasized

that this chapter does more than decompose and segment. It completes the
architectural argument which opened Chapter 1: that the theory of Univer-
sal Grammar requires a unified morphosemantic theory of number, rather
than disjoint morphological and semantic theories. It achieves this by solv-
ing a classic morphological problem in terms of features that originate in
the semantic treatment of noun classification. It shows, therefore, that the
semantic primitives of such core oppositions as count∼mass, collective∼non-
collective, granular∼non-granular, and the morphological primitives of expo-
nence and complex syncretisms are one and the same.
5.1. Preliminaries
Before proceeding to the analysis, I give a full list of Kiowa’s prefixes and
outline some theoretical assumptions.
(8) Kiowa Agreement Prefixes
x :y : z
∅ 3s 3d 3p 3i 3a
1s:(3a:) a gya nen gyat dé de
1ex:(3a:) e é∗ et ét∗ ét ét
1in:(3a:) ba bá∗ bet bát∗ bét bé
2s:(3a:) em a men bat bé be
2d:(3a:) ma má∗ mén mán∗ mén∗ mé
2i:(3a:) ba bá∗ bet bát∗ bét bé
3s:(3a:) ∅ ∅ eM gya é em
3d:(3a:) eM éM∗ én én∗ én én
3i:(3a:) e é∗ et ét∗ ét ét
3a:(3a:) á á∗ et gyá∗ et ém
∅/2s/3s:1s: éM éM né yáM nÓ né
2d:1s: mâa∗ mâa∗ ménêi∗ mánẐi∗ mÓnÔO∗ d∼i
2i:1s: bâa∗ bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
3d:1s: êM Mi∗ êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ d∼i
3i:1s: êi∗ êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
3a:1s: âa∗ âa∗ dêi∗ gyâa∗ dÔO∗ d∼i
any:1d/p: dÓ dÓ dét gyát dÓt d∼i
∅/1s:2s: em gyá nén yán gÓ d∼i
other :2s: gO gÓ dét gyát gÓt d∼i
any:2d: mÓ mÓ mén mán mÓn d∼i
any:2i: bÓ bÓ bét bát bÓt d∼i
1s:3s: gyá nén yán gÓ d∼i
1ex:3s: êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
1in:3s: bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
∅/2s/3s:3s á én án Ó d∼i
any:3d: mé mén mén mén d∼i
any:3i: bé bét bét bét d∼i
2d:3s: mâa∗ ménêi∗ mánẐi∗ mÓnÔO∗ d∼i
2i:3s: bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
3d:3s: êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ d∼i
3i:3s: êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
3a:3s: âa∗ dêi∗ gyâa∗ dÔO∗ d∼i
162 CHAPTER 5
The table requires the following comments: (a) Cells representing impos-
sible agreement combinations are blank. (b) ‘3a’ represents animate plural /
reflexive agreement discussed in Sections 2.5 and 4.5.2. (c) ‘any’ includes ∅,
unless z =∅—the argument combination ∅:x:∅ is impossible in Kiowa. ‘any’
is also constrained by binding theory; for instance, in ‘any:2d:z ’, ‘any’ can-
not be second person. (d) Agreement prefixes for transitive sentences with
first / second person objects are represented as agent:object:∅. E.g.: 2i:1s:∅
would be used in ‘You all (2i) saw me (1s)’. The reason for this notation is
that first / second person objects behave morphologically as indirect objects.
See Adger and Harbour (2007) for discussion. (e) Intransitive agreement is
represented as argument:∅. E.g., 1s:∅ would be used in ‘I arrived’. (Note
that these cannot be termed either ‘agents’ or ‘external arguments’. Con-
sequently, they are labelled simply ‘subjects’ in (8). Note also that if the
z -agreement is ∅ then a must be absent in such prefixes as 1s:(3a:)z. I.e., a
is the prefix only for 1s:∅, not for 1s:3a:∅, which is, in any event, an impossi-
ble argument combination in Kiowa.) (f) Third plural inanimate intransitive
agreement, 3p:∅, absent from the table, is gya.
Throughout this chapter, I will refer to prefix positions by letter: x, y, z.
(9) x :y:z x = subject of ditransitive

y = indirect object of ditransitive
z = direct object of ditransitive
x :z x = subject of transitive
z = direct object of transitive
:y:z y = possessor
z = possessee
x :∅ x = subject of unaccusative
For example, in (2), we would write x =3d, y=3a, z =3s. Referring to prefix
positions by letter has two advantages. First, it avoids cumbersome phrases
like ‘subject of (di)transitive’. (Compare the readability of, for instance, ‘if
x =3d’ with ‘if the subject of a (di)transitive is 3d’.) Second, it permits us to
abstract away from the sometimes complicated syntactic and featural reality
behind the prefix positions. (See Adger and Harbour 2007 for analysis.)
5.1.1. Theoretical assumptions
Prefix structure
Following Harbour (2003a), I assume that the prefix contains as many heads
as there are agreeing arguments (i.e., one, two, or three) and that these
heads form a cluster dependent on the verb. (The prefix forms a phonological
domain separate from the verb.)
(10) V
Prefix V
x y z
Consequently, the feature structure of the prefix in (1) is:
(11)
Prefix
⎡ x ⎤ ⎡ y ⎤ z
−singular −singular +singular
⎢+singular ⎥ ⎢+singular ⎥ −augmented
⎢ ⎥ ⎢ ⎥
⎣−augmented⎦ ⎣−augmented⎦
+empathic +empathic
The challenge posed by Kiowa agreement prefixes can now be clarified.

It is to explain how (11) yields á∗, or how general instantiations of (10) yield
(8). The answer offered relies in part on the morphological and phonological
rules that follow.
Phonology
The following phonological processes, informally described, are assumed (Har-

rington 1928, Merrifield 1959b, Watkins 1984; see Harbour 2002 on tone):1
1
I abstract away from rule ordering. Indeed, there is evidence of cyclic application,
in the shape of an ordering paradox. On the one hand, for 2i:1/3s:3p, Nasalization must
precede Dental-velar switching to derive mánẐi from /b+ia+d+ia+[+nasal]/; the reverse
ordering yields *máNẐi→*máẐi. On the other hand, for ∅/1s:2s:3d, Dental-velar switching
must precede Nasalization to derive nén from /g+e+d+[+nasal]/; the reverse ordering
yields *Nén→*én. The new cycle might be associated with (re)syllabification.
164 CHAPTER 5
(12) Dental-velar switching

g / k(!/h) → d / t(!/h) / i/y
d / t(!/h) → g / k(!/h) / e
(13) Glide insertion a → ya / g/k(!/h)
(14) Glide formation i → y / g/k(!/h) a
(15) Nasalization [+nasal] spreads to all phonemes of the prefix
(16) Engma-deletion N → ∅
(Note: engma is not a phoneme of the language.)
(17) (y)i∼ya alternation (y)i ↔ ya
(Note: the conditions for this alternation are not entirely under-
stood.)
(18) Final devoicing d/b → t/p / [σ V ]
(Note: there are no syllable-final velars, except by assimilation.)
(19) Vowels in hiatus V → ∅ / V
(Note: the proper statement of the rule does not affect diphthongs.)
Morphology
The following morphological processes are assumed (Bonet 1991, Noyer 1998,
Harbour 2003a; see the last two for discussion of how a morphological theory
with these operations succeeds in restricting the class of possible grammars).
Feature deletion (impoverishment) removes features from a terminal
node. Such rules, which leave the node intact, are of the form: [αF] → ∅.
Applied to [αF βG], this yields [βG].
Node deletion removes a terminal node and all features located at it.
Applying | → ∅ to yields

[αF] αF βG γH βG γH
Feature insertion inserts the (contextually) unmarked value of a fea-
ture. Such rules are of the form ∅ → [αF]. Taking minus to be the unmarked
value of [±singular] in the context of [+augmented], the rule ∅ → [−singular]
would apply to [+augmented] to yield [−singular +augmented].
5.2. Reduction of explicanda
In this section, morphological operations are appealed to to reduce the ar-

gument combinations and prefixes that must be accounted for. Specifi-
cally, deletion of syntactic features or of whole nodes neutralizes differences
between argument combinations, inducing many-to-one correspondences be-
tween argument combinations and prefixes. A variety of such deletions suc-
ceed in reducing the number of prefixes to be derived from 160 to 116.
5.2.1. ‘any’
What is the featural reality behind ‘any’ ? Concretely, consider any:2d:3s.

There are no restrictions on the external argument beyond binding theory
(e.g., *2s:2d:3s). So, it could be zero, as with an experiencer predicate (‘You
two know him’, ∅:2d:3s):
⎡ y ⎤ z
−author +singular
⎢+hearer ⎥ −augmented
⎢ ⎥
⎣−singular ⎦
−augmented
Or the agent could be first person exclusive, any number, or third person,
any number:
⎡ x ⎤ ⎡ y ⎤ z
(+author) −author +singular
⎢(−hearer) ⎥ ⎢+hearer ⎥ −augmented
⎢ ⎥ ⎢ ⎥
⎣ ±singular ⎦ ⎣−singular ⎦
±augmented −augmented
How are these seven argument combinations realized as a single form?

I suggest that the two structures are rendered identical by deletion of the x
node. When vocabulary insertion occurs, this identity results in syncretism.
In the current case, deletion is:
|
(20) x → ∅ / [y 2d ] [z 3s ]
166 CHAPTER 5
However, deletion of the x node should apply not just when z =3s, but also for
z =3d, z =3p, and so on. Furthermore, it should apply not just for y=2d, but
for y=1i, y=2i, y=3d/i too. These correspond to the natural class defined
by [−singular]. So, we can generalize (20) to:
|
(21) x → ∅ / [y −singular]
Note a welcome consequence of (21). A cursory glance at the top part of (8),
specifically, at the simple transitive part, shows that the featural composition
of the x node has a phonological effect on the agreement prefix—scanning
down any column, i.e., keeping the direct object constant, we find that the
cells vary as the external argument changes. In the ‘any’ cases, we require
a means of preventing the external argument from having any phonological
effect on the prefix. That is, we must derive their syncretism with absence
of an external argument, ∅. Deleting the x node achieves just that.
5.2.2. More x deletion
The syncretism ∅:2s:z ≡ 1s:2s:z is also captured by node deletion:

|
⎡ y ⎤
⎡ x ⎤ −author
(22) +author
⎢−hearer ⎥
→ ∅ / ⎢+hearer
⎢
⎥
⎥
⎢ ⎥ ⎣+singular ⎦
⎣+singular ⎦
−augmented
−augmented
For ease of illustration, we can paraphrase this as 1s → ∅ / 2s, the effect

of which is:
(23)
1s 2s z → 2s z
What then is the featural reality behind ‘other ’ in other :2s:z ? Again,
we find the prefixes phonologically invariant despite variation of the external
argument. So, x features must be deleted. However, deletion of the entire
node incorrectly forces syncretism with ∅:2s:z . I follow Harbour (2003a) in
regarding the relevant operation as deletion of features but not of the x node
itself. Consequently, (24) applies to (25) to yield (26).
(+author) ⎡ y ⎤
(−hearer) −author
(24) ± singular
→ ∅ / [ ] ⎢+hearer
⎢
⎥
⎥
⎣+singular ⎦
± augmented
−augmented
(25)
⎡ x ⎤ ⎡ y ⎤ z
(+author) −author ±singular
⎢(−hearer) ⎥ ⎢+hearer ⎥ ±augmented
⎢ ⎥ ⎢ ⎥
⎣ ±singular ⎦ ⎣+singular ⎦
±augmented −augmented
(26)
x ⎡ y ⎤ z

−author ±singular
⎢+hearer ⎥ ±augmented
⎢ ⎥
⎣+singular ⎦
−augmented
This empty node is automatically targeted for insertion of the unmarked

feature [−singular]. As this feature conditions allomorphy in z -exponence
(59), it prevents syncretism of (26)/(27) with ∅:2s:z .
(27)
x ⎡ y ⎤ z

−singular −author ±singular
⎢+hearer ⎥ ±augmented
⎢ ⎥
⎣+singular ⎦
−augmented
5.2.3. Animate/reflexive deletion
Recall (Sections 2.5 and 4.5.2) that transitive agreement prefixes are system-
atically ambiguous; that is, x :3a:z ≡ x :z. Gya, for instance, means 1s:3s
and 1s:3a:3s, so that one and the same verb form is ambiguous between ‘I
killed it’, ‘I killed it for them’, and ‘I killed it for myself’. Formally:
168 CHAPTER 5
|
⎡ y ⎤
(28) −singular → ∅
⎢+singular ⎥
⎢ ⎥
⎣−augmented ⎦
+empathetic
5.2.4. Impoverishment: Person deletion
In the treatment of first person syncretisms and the sii class, two rules of
person deletion were introduced. These accounted for the syncretisms of
first exclusive with inverse and first inclusive with second plural. In current
notation, they are:
(29) [±author] → ∅ / [x ]
(30) [±hearer] → ∅ / [y,z +author ]
5.2.5. Impoverishment: More person deletion
The effect of (29), syncretism of first person with third person, arises else-
where and again requires person deletion, though under a different condi-
tioning environment: specifically, x :1s:z ≡ x :3s:z, for non-zero x , z :2
2
The discrepancy with respect to the ∅ column is as explained following (8). 2d:1s:∅, for
example, is used in, say, ‘You two saw me’, where 1s is a direct object. It is represented
in the prefix notation as an indirect object, ‘:1s:’, because first / second direct objects
share certain syntactic, and, consequently, morphological properties with indirect objects
(Adger and Harbour 2007). Third person direct objects do not display these syntactic
properties and third person z -agreement is distinct from third person y-agreement. As a
result, the agreement prefix for ‘You two saw her’ is 2d:3s, not *2d:3s:∅, which is not a
possible argument combination in Kiowa. Thus, it is not that the correlation ‘x :1s:z =
x :3s:z, for all non-singular x ’ breaks down for z =∅. Rather, when one bears in mind the
syntactic reality behind this generally useful morphological notation, one realizes that the
empty cells are impossible, and so trivially exempt from such correlations.
(31) x :y : z
∅ 3s 3d 3p 3i 3a
2d:1s: mâa∗ mâa∗ ménêi∗ mánẐi∗ mÓnÔO∗ d∼i
2d:3s: mâa∗ ménêi∗ mánẐi∗ mÓnÔO∗ d∼i
2i:1s: bâa∗ bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
2i:3s: bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
3d:1s: êMiM∗ êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ d∼i
3d:3s: êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ d∼i
i:1s: êi∗ êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
i:3s: êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
a:1s: âa∗ âa∗ dêi∗ gyâa∗ dÔO∗ d∼i
a:3s: âa∗ dêi∗ gyâa∗ dÔO∗ d∼i
Now, the cause of this correlation cannot be that y=1s and y=3s are
realized by the same vocabulary items. If they were, then, for instance,
∅:1s:3s (éˇ) and ∅:3s:3s (á) would be identical.
Instead, I suggest that first person singular [y +author −hearer +singular
−augmented] is rendered identical to third person singular [y +singular
−augmented] by deletion of the person features [+author −hearer]. In order
to prevent syncretism of ∅/2s/3s:1s:z and ∅/2s/3s:3s:z , it is necessary to
contextualize this deletion to non-singular x .3
3
The deletion (32) competes with (20): in 2i:1p:3s, for example, either could apply. The
correct results are achieved if (20) precedes (32). It is not clear to me that this ordering
is intrinsically given, for instance, by Pān.ini’s Principle. And, though it is possible to
avoid this problem by recontextualizing (32), to [x −singular] [y +singular] say, this
is not overly satisfying. A more exciting alternative, arising from discussion with Elena
Anagnostopoulou—I do not pursue it here as it lies well beyond the scope of this inquiry—
departs from the observation that external argument and indirect object in Kiowa almost
never both agree for person. Indeed, if one is [−singular], then it is never the case that both
agree fully. In this chapter, this is treated as a purely morphological fact: whenever such
argument combinations enter the morphology, some of their features are deleted. However,
it is possible that the syntax itself never gives rise to structures of the relevant form, so that
these morphological rules are superfluous, the true explanation stemming from a deeper
syntactic fact. Interestingly, similar restrictions hold in two other rich agreement systems
that I am familiar with, namely, Yimas (Foley 1991) and Georgian. To the extent that
Yimas differs from Kiowa in this respect, it does so by use of portmanteau 1+2 affixes,
the syntactic nature of which is not entirely clear. In all three languages, there is a case
to be made that, when the (in)direct object is second person and the external argument
is first, second person, rather than first, agrees. See Béjar (2007) and references therein
for discussion, and Heath (1998) for an overview of several other cases.
170 CHAPTER 5

+author
(32) → ∅ / [x −singular] [y ]
−hearer
With this in hand, the list of prefixes to be derived diminishes, as we can
conflate two sets of prefixes in (31) under the label x :1/3s:z . Combining all
of the above, we have the reduced table (33).
Summary
(33) Kiowa Agreement Prefixes Simplification 1
x :y : z
∅ 3s 3d 3p 3i 3a
1s:(a:) a gya nen gyat dé de
2s:(a:) em a men bat bé be
2d:(a:) ma má∗ mén mán∗ mén∗ mé
2i:(a:) ba bá∗ bet bát∗ bét bé
3s:(a:) ∅ ∅ eM gya é em
3d:(a:) eM éM∗ én én∗ én én
i:(a:) e é∗ et ét∗ ét ét
a:(a:) á á∗ et gyá∗ et ém
2d:1/3s: mâa∗ mâa∗ ménêi∗ mánẐi∗ mÓnÔO∗ d∼i
2i:1/3s: bâa∗ bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗ d∼i
3d:1/3s: êMiM∗ êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ d∼i
i:1/3s: êi∗ êi∗ édêi∗ égẐi∗ édÔO∗ d∼i
a:1/3s: âa∗ âa∗ dêi∗ gyâa∗ dÔO∗ d∼i
any:2d: mÓ mÓ mén mán mÓn d∼i
∅/2s/3s:3s á én án Ó d∼i
any:3d: mé mén mén mén d∼i
any:i: bé bét bét bét d∼i
The effect of the morphological deletion and impoverishment operations

simultaneously derives the desired syncretisms and helps to explain why the
number of prefixes is significantly smaller than the number of possible argu-
ment combinations. Counting 1d and 1p as separate argument combinations,
as they were when 2, 548 was calculated, the reader may easily verify that
several hundred morphological distinctions have been neutralized. (I omit
an exact count as the figure of 2, 548 was itself only approximate, neglecting
person-case and binding-theoretic restrictions.) More specifically, we have
reduced the number of prefixes under active consideration from 160 to 116.
5.3. Segmentation I: Subregularities
By considering identity relations between prefixes corresponding to different

argument combinations, it has proved possible to reduce the total number of
different forms that must be derived. The discovery of identity relations is
not, however, the only way to reduce the explanatory burden. Another is to
find correspondences between classes of prefixes. For instance, suppose that
the prefix x 1 :z is ϕ1 and that the prefix x 2 :z is ϕ1 +ϕ2 . That is, to say x 2 :z ,
you say x 1 :z (ϕ1 ) plus something extra (ϕ2 ). Then we want ϕ1 to realize
features common to x 1 and x 2 (to derive the overlap), and we want ϕ2 to
realize features of x 2 that x 1 lacks (to derive the additional element). Such
correlations induce requirements on the vocabulary list and are generally
phrased as such below. By the end of this section, the number of prefixes to be
derived will have been reduced from 116 to 66, at which point, decomposition
of individual prefixes can proceed quite easily.
Duals
Prefixes where x /y=3d are predictable from prefixes where x /y=3i, by nasal-
ization of the entire prefix. Informally: 3d ≡ 3i+nasalization.
For instance, consider the ditransitive prefixes any:3d:z and any:3i:z .
Every cell in the top row is the nasalization of the cell directly below it:
(34) x :y : z
3s 3d 3p 3i 3a
any:3d: mé mén mén mén mén
any:3i: bé bét bét bét bét
172 CHAPTER 5
For the prefixes 3d:1/3s:z and 3i:1/3s:z , the same holds (by orthographic
convention, nasality is not marked on vowels cosyllabic with nasal stops).
(35) x :y : z
∅ 3s 3d 3p 3i 3a
3d:1s: êMiM∗ êMiM∗ éMnêi∗ éMnẐi∗ éMnÔO∗ éMnêi∗
3i:1s: êi∗ êi∗ édêi∗ égẐi∗ édÔO∗ édêi∗
Only z =3p requires further comment. Straightforward nasalization yields

*éˇNı̇i, not éˇnẐi. However, if the underlying form is /é-d-Ẑi/, then éˇnẐi arises
by nasalization and égẐi by Dental-velar switching (Section 5.1.1).
For transitives too, the correlation holds. The only difficulty here is with
z =3d. Nasalization of 3i:3d, et, yields en, rather than 3d:3d én, which has
high tone. We return to this below (p. 181).
(36) x: z
∅ 3s 3d 3p 3i 3a
3d: eM éM∗ én én∗ én én
3i: e é∗ et ét∗ ét ét
The correspondence extends to the second person dual. For example:
(37) x :y : z
∅ 3s 3d 3p 3i 3a
any:2d: mÓ mÓ mén mán mÓn mén
any:2i: bÓ bÓ bét bát bÓt bét
The correlation 2d ≡ 2i+nasalization faces two exceptions, again, from tran-

sitive prefixes, and again, with respect to tone. 2d:3d is mén, but the result
of nasalizing 2i:3d, bet, is men. And 2d:3i is mén∗, but the result of
nasalizing 2i:3i, bét, is mén.
These results prompt (38), subject to the provisos in (39):
(38) a. If [x,y i ] ⇔ ϕ, then [x,y d ] ⇔ ϕ

b. [x,y d ] ⇔ [+nasal].
(39) Tone
2d:3d has high tone although 2i:3d has not.
2d:3i is ∗ although 2i:3i is not.
3d:3d has high tone although 3i:3d has not.
We will see that much the same realizations hold for z =d (59), but that
there is an extra allomorphy condition. First person ‘duals’ are excluded
from (38), as they trigger inverse agreement, being sii.
x:1/3S:z
We consider z =∅ and z =∅ separately, as x :1s:∅ is predictable from intransi-

tive x :∅, and x :1/3s:z, z =∅, from x :z. Recall that, by (32), x is non-singular
in these prefixes.
Prefixes of the form x :1s:∅ are identical to x :1/3s:3s.4
(40) x :y : z
∅ 3s
2i:1/3s: bâa∗ bâa∗
1/3i:1/3s: êi∗ êi∗
3a:1/3s: âa∗ âa∗
There are two ways to capture this identity relation. Either z =3s is realized
as zero in the context of y=1/3s; or whatever realizes z =3s in the context of
y=1/3s is non-zero but phonologically redundant. Anticipating that tonal
evidence from other :2s:∅ favors the latter solution, we can write:
(41) If [z 3s ] ⇔ ϕ / [y 1/3s ] , then x /y ⇔ ϕ.

(If y=1/3s, then z =3s’s realization is redundant, given the realiza-
tions of x and y.)
For z =∅, x :1/3s:z is predictable from x :z. To illustrate this, consider

3i:1/3s:z and 3i:z .
(42) x :y : z
3s 3d 3p 3i
3i: é∗ ét ét∗ ét
3i:1/3s: êi∗ édêi∗ égẐi∗ édÔO∗
The second row can be derived from first as follows. Lengthening of the
3i:3s vowel into a falling tone plus ∗ yields 3i:1/3s:3s.5 Addition of êi to
4
For some speakers, x :3s prefixes are used here. This follows immediately from (32).
5
The specification ‘into a falling tone’ is redundant, given that 3i:3s lowers tones that
follow it. A falling tone is simply a vowel that lowers itself in the middle. See Harbour
174 CHAPTER 5
3i:3d yields 3i:1/3s:3d. (Note also the difference in voicing of the alveolar.
I follow Watkins in assuming that the same vocabulary item is involved in
both and that it is /d/, which sometimes nasalizes to n and at other times
devoices to t.) Addition of ı̇i to 3i:3p yields 3i:1/3s:3p, once again voicing
the alveolar, and switching it to velar before i/y. Lastly, addition of ÔO to
3i:3i and voicing of the alveolar yields 3i:1/3s:3i.
Now, the identity of the added vowels, êi, Ẑi and ÔO, is not at all mys-
terious. E is a standard component of prefixes with z =3d, as can be easily
verified by scanning down the 3d column of (8). Similarly, i is a standard
component prefixes with z =3p, generally in its ya∼a allophones. And O is a
component of prefixes with z =3i, when there is y-agreement. Consequently:
(43) If z ⇔ -V-, then [y 1/3s ] ⇔ -V̂V

(If y=1/3s, then the vowel that realizes z occurs twice.)
(Nasalization yields 3d:1/3s:z from 3i:1/3s:z .)

The correspondences just discussed apply, mutatis mutandis, to 2i:1/3s:z
and 2i:z , with one minor difference.
(44) x :y : z
3s 3d 3p 3i
2i: bá∗ bet bát∗ bét
2i:1/3s: bâa∗ bédêi∗ bágẐi∗ bÓdÔO∗
The z =3s form follows from lengthening of the vowel, and the 3d, 3p and 3i
forms from respective addition of êi, Ẑi and ÔO. The minor difference concerns
the first vowel in 2i:1/3s:3i, which is Ó, not é, as in 2i:3i. This difference is
discussed below.6 (Again, the 2d forms are predictable by nasalization.)
Finally, consider:
(2002) for detailed discussion of such cases.

6
In Mr Bointy’s Mount Scott dialect (possibly shared by some Carnegie speakers),
there is no difference in the vowel, both forms beginning bé. . . . An additional difference
between his speech and that of Dr McKenzie, Watkins’ primary consultant, is that the
alveolar does neither voices intervocalically nor velarizes, as in Dr McKenzie’s bágẐi. For
Mr Bointy, this was bát.Ẑi, where the ‘.’ indicates the syllable boundary. Mrs White Horse
Taylor, who appears in several respects to speak as Dr McKenzie did, says bá.gẐi. For
convenience, I give the relevant variants below:
(45) x :y : z
3s 3d 3p 3i
3a: á∗ et gyá∗ et
3a:1/3s: âa∗ dêi∗ gyâa∗ dÔO∗
Again, for the z =3s form, the vowel lengthens into a falling tone. For the
other three, there is once again addition of the expected long falling vowel,
though for z =3p, it is in its ya allophone. However, the 3d and 3i forms
present a minor wrinkle. Straightforward addition of the expected vowels
yields the wrong forms, respectively, *édêi and *édÔO. In these cases, the
initial vowel is not inserted, or, if inserted, does not survive the phonology.
Summarizing these patterns, we have:
(46) a. If x :z ⇔ ϕ, then x :1/3s:z ⇔ ϕ z =∅

b. If x :z ⇔ V, then x :1/3s:z ⇔ V̂V z =∅
Also, the following minor deviations should be explained:
(47) a. Absence of initial vowels from 3a:1/3s:z prefixes.

b. Change from e to O of the medial vowel in 2i:1/3s:3i.
Third singular agent
There is an intriguing correlation between transitive 3s:x prefixes and intran-

sitive x :∅ prefixes.7
(48) 3s 3d 3p 3i 3a
:∅ ∅ eM gya e á
3s: ∅ eM gya é em
For z =3s/3d/3p, x =3s makes no difference at all to the prefix. Given this,
it is tempting to imagine that [x 3s ] is simply deleted. Certainly, this would
(i) x :y : z
3d 3p 3i
2i:1/3s: bét.êi∗ bát.Ẑi∗ bét.ÔO∗
1/3i:1/3s: ét.êi∗ ét.Ẑi∗ ét.ÔO∗
7
Just in this subsection, the same letter, x, is used as a variable over two different prefix
positions. This is to emphasize that we are dealing with essentially the same feature bundle
in the two positions.
176 CHAPTER 5
work well for 3s:1s:z (since 3s:1s:z ≡ ∅:1s:z ) and for 3s:3s:z (3s:3s:z ≡
∅:3s:z ). However, it would not work for 3s:2s:z , which is non-syncretic with
∅:2s:z . Nor would it work for z =3i/3a in (48): intransitive i-agreement,
3i:∅, is e, but 3s:3i, é, has high tone (however, observe that, on p. 185, this
high tone attributed, for independent reasons, to allomorphy of y=3i in the
context of x ); and intransitive 3a-agreement, 3a:∅, is á, but 3s:3a is em. I
am unsure how far to pursue this resemblance.
Summary
By applying a series of morphological operations (Section 5.2) and then by

noting systematic phonological correspondences between sets of prefixes (Sec-
tion 5.3), we have reduced both the number of argument combinations to be
accounted for and the number of prefixes to be explained, replacing these with
a series of vocabulary items and phonological and morphological processes.
This explains why the inventory of attested prefixes, at 160, is substan-
tially smaller than the inventory of logically possible argument combinations,
2, 548, and has reduced the inventory of prefixes to be accounted for from
160 in (8) to 66:
(49) Kiowa Agreement Prefixes Simplification 2
x :y : z
∅ 3s 3d 3p 3i 3a
1s:(a:) a gya nen gyat dé de
2s:(a:) em a men bat bé be
2i:(a:) ba bá∗ bet bát∗ bét bé
3s:(a:) ∅ ∅ eM gya é em
i:(a:) e é∗ et ét∗ ét ét
a:(a:) á á∗ et gyá∗ et ém
∅/2s/3s:3s á én án Ó d∼i
5.4. Segmentation II: Segments
We now turn to the segmentation proper of prefixes. We begin with the

ditransitive prefixes, where, with tone almost exceptionlessly high, segmen-
tation is somewhat simpler. (‘Ditransitive’ means prefixes where y =3a, as
morphological rules induce x :3a:z ≡ x :z , which are transitive prefixes.)
5.4.1. Ditransitives
We are now concerned with the following prefixes.
(50) Kiowa Ditransitive Agreement Prefixes
x :y : z
∅ 3s 3d 3p 3i 3a
∅/2s/3s:3s á én án Ó d∼i
Let us first consider y=singular prefixes (from which other :2s:z is ex-
cluded on principled grounds, as explained below).
Singular y
We now concentrate on the following prefixes.
(51) x :y : z
∅ 3s 3d 3p 3i
∅/2s/3s:1s: éM éM né yáM nÓ
∅/1s:2s: em gyá nén yán gÓ
1s:3s: gyá nén yán gÓ
∅/2s/3s:3s: á én án Ó
178 CHAPTER 5
Let us begin with the bottom row, ∅/2s/3s:3s:z . Observe that x =2s/3s
makes no phonological contribution beyond that of x =∅. Consequently, I
suggest that the phonological components of these prefixes are vocabulary
items realizing (some subset of) y=3s and the relevant object, z .
(52) a. á ⇔ [y 3s ] [z 3s ]
b. én ⇔ [y 3s ] [z 3d ]
c. án ⇔ [y 3s ] [z 3p ]
d. Ó ⇔ [y 3s ] [z 3i ]
Now, 1s:3s:z and ∅/1s:2s:z rows are identical (modulo em) and both
are derivable from the last row of (51) by standard phonology following the
addition of g- to (52): g+á → gyá by Glide insertion; g+én → nén by
Dental-velar switching and Nasalization; g+án → yán by Glide insertion,
Nasalization, Engma-deletion; and g+Ó → gÓ transparently.
The question is whether the g- is the same in both cases. If it is, then
the features that g- realizes are those common to 1s:3s:z and ∅/1s:2s:z . So:
⎧ ⎫ ⎧ ⎫
⎪
⎪ ⎡ x ⎤ y ⎪
⎪ ⎪ ⎪
⎪⎡ y ⎤⎪⎪
⎪
⎪ ⎪
⎪ ⎪ ⎪
⎪
⎨ +author ⎬ ⎨ −author ⎬
g- ⇔ ⎢ −hearer ⎥ +singular ∩ ⎢+hearer ⎥
⎪ ⎢ ⎥ ⎢ ⎥
⎪
⎪ ⎣+singular ⎦ −augmented ⎪ ⎪
⎪
⎪
⎪
⎪⎣+singular ⎦⎪ ⎪
⎪
⎪
⎩ ⎪
⎭ ⎩ ⎪ ⎪
⎭
Equivalently, g- ⇔ [y +singular −augmented ] = [y 3s ].

Given (52), this incorrectly forces the occurrence of g- in the last row of
(51). Unless we wish to invoke further morphological operations, we should
claim that 1s:3s:z and ∅/1s:2s:z are realized by different, accidentally ho-
mophonous, g-. Since 1s:3s:z ≡ x =1s + ∅:3s:z , I attribute one g- to x =1s.
(53) g- ⇔ [x 1s ]
And since a zero agent, being featureless, is not realized by any vocabulary
items, I attribute the other g-, of ∅:2s:z and 1s:2s:z , to y=2s.
(54) g- ⇔ [y 2s ]
When we turn to other :2s:z and to 1s:(3a:)z , we will see that this accidental
homophony is not theoretically problematic. On the contrary, it correctly
predicts two other sets of prefixes.
For the top row, matters are more complicated. Unlike the other rows,
∅/2s/3s:1s:z cannot be derived by adding phoneme(s) to ∅/2s/3s:1s:z . How-
ever, several properties of these prefixes are predictable if we claim that two
processes, metathesis8 and unmarked feature insertion, required elsewhere in
the agreement system, also apply here. De-metathesis yields:9
(55) Onset∼coda De-Metathesis

x :y : z
∅ 3s 3d 3p 3i
∅/2s/3s:1s: éM éM én án Ón
The distribution of codas here is different from that in the other rows of
(51) in that Ón (z =3i) has one. However, in (57), the distribution of codas is
identical to (55): codas if and only if z =3d/3p/3i. This distribution of codas
is sensitive to non-singular x /y (Harbour 2003a), but, here, y is singular.
The mismatch can be fixed if we invoke feature insertion. That is, we
delete [+singular] in 1s and then insert what Harbour (2003a) argues to be
the default value, [−singular]. This yields:
(56) [y +author −hearer −singular −augmented]
Being non-singular, the distribution of codas will now be as in (57). More-

over, a second property of ∅/2s/3s:1s:z prefixes is derived. Number in (56)
is dual and so, by (38), causes the prefixes to be nasal. (Such spurious duals
are not confined to the current case: recall the dual-like realization of 3a in
the presence of indirect objects; Section 4.5.2.)
8
Watkins (1984) relies on metathesis to derive 3p:∅ gya from /ia+d/, for instance:
Metathesis yields /dia/, Dental-velar switching yields /gia/, Glide formation yields /gya/.
Like Watkins, I am unsure of the conditions that trigger metathesis. Note, interestingly,
that Halle and Vaux (1998) face the same problem with respect to Latin nominal mor-
phology. They decompose the genitive plural and dative plural endings (fifth declension),
-rum and -bus, into r+um and bu+s. Undoing the effect of the Latin rhotacism rule, we
have s+um and bu+s. Halle and Vaux observe: ‘The affix order [in bu+s] is the reverse
of that in the Genitive Plural [s+um]; there is at present no explanation for this fact.’
One might appeal to well-formedness of syllables in this case (in preference to *sbu and
*ums), though this is not clearly exportable to the Kiowa cases.
9
Abstracting away from tone, yáˇ derives from /d+ia+[+nasal]/. To derive the form
in (55), observe that we have /ia+d+[+nasal]/, prior to Metathesis; Nasalization yields
/ian/; Vowel-in-hiatus deletion yields /an/, the desired form, modulo tone.
180 CHAPTER 5
Non-singular y
We now concentrate on the following prefixes.
(57) x :y : z
∅ 3s 3d 3p 3i
any:1d/p: dÓ dÓ dét gyát dÓt
other :2s: gO gÓ dét gyát dÓt
any:2i: bÓ bÓ bét bát bÓt
any:i: bé bét bét bét
Again, as in the cases considered above, the onsets are constant across
each row: d- for any:1d/p:z (with Dental-velar switching for 3p); g- for
other :2s:z (with Dental-velar switching for 3d); b- for the other two rows.
Similarly, the rhymes are also nearly constant throughout the columns, with
two exceptions. Working right to left, in the 3i column, the rhyme is -Ót
(except that the vowel for any:3i:3i is e); in the 3p column, the rhyme is -át
(except that the vowel for any:3i:3p is e again); in the 3d column, the rhyme
is -ét; in the 3s column, the rhyme is -Ó (except that the vowel for any:3i:3p is
e again); in the ∅ column, the rhyme is -Ó (except that the tone for other :2s:∅
is low; any:3i:∅ is, recall, an impossible argument combination). What is the
relationship between these regularities and those discussed for singular y?
First, let us compare the rhymes of two groups (ignoring the more involved
cases any:3i:z and ∅/2s/3s:1s:z ).
(58) z
3s 3d 3p 3i
singular y á én án Ó
non-singular y Ó ét át Ót
There are three differences between the two rows. The 3d rhymes and the
3p rhymes are nasal∼non-nasal counterparts of one another. The 3i rhyme
has a -t coda in the non-singular case, though none in the singular. Third,
the 3s rhymes are just different. We will see, when we turn to the transitive
prefixes, that the first two differences hold there too, except that singularity
of x , rather than y, is the crucial factor. The third difference does not hold
of transitive prefixes for non-interesting reasons: the vowel O is confined to y
prefixes. I suggest, therefore, that all three differences depend on the value
of [±singular] on y or x —that is, [±singular] the higher argument conditions
allomorphy of z -agreement.
(59) z -allomorphy [±singular]-conditioned

a. á ⇔ [z 3s ] / [+singular ]
Ó ⇔ [z 3s ] / [−singular ]
b. ∅ ⇔ [z 3i ] / [y +singular ]
-t ⇔ [z 3i ] / [y −singular ]
c. [+nasal] ⇔ [z 3d ] / [+singular ]
[−nasal] ⇔ [z 3d ] / [−singular ]
A potential problem for these allomorphy relations is that other :2s:z con-
ditions all the [−singular] allomorphs though 2s itself is [+singular]. Nor are
all values of x compatible with other themselves [−singular], as other covers
3s. However, recall that the featural reality behind other is an x node from
which all features have been deleted (24). Following Noyer’s (1998) argument
that Universal Grammar permits the insertion of unmarked feature values,
Harbour (2003a) argues that Kiowa too permits such insertion and that, in
the case of number, the unmarked feature and value inserted is [−singular].
Having deleted all but the x head itself, which I take to be a root of a ϕ-
structure (Chapter 3), the unmarked [−singular] is inserted, as shown above
in (27). Consequently, other :2s:z patterns with others that have non-singular
x /y with respect to the allomorphs of z -agreement that they condition.
Consider now the two rhyme exceptions. First, e in any:3i:z . Observe, in
(49), that occurrence of e irrespective of the value of z is also a characteristic
of 3i:(3a:)z . Consequently, I assume that this is a property of x /y=3i.
(60) Irrespective of z , x /y=3i is realized as e.
A number of technical means are available to satisfy (60): there could be a

specific vocabulary item e that realizes i and d and that deletes the vow-
els that realize z (via the Vowel-in-hiatus rule); e could be a default; or e
could result from phonological readjustment (cf., Chomsky and Halle’s 1968
treatment of sing∼sang, tell∼told, and similar). At this point, it is not
necessary to decide between these options.
The second rhyme exception is the low tone of gO other :2s:∅. Interest-
ingly, the only other low tone prefix in this part system is em 1s:2s:∅. Given
(32), the only features realized by em are [y 2s ]. So, one might think low
tone a systematic property of [y 2s ]. However, the intransitive 2s:∅ of ‘You
182 CHAPTER 5
arrived’ is also em. This suggests that em realizes the features common to
intransitive 2s-agreement and indirect object 2s-agreement. Consequently,
the low tone of em is a fact about that vocabulary item per se, not about the
pair other :2s:∅ / 1s:2s:∅ generally. Thus, the low tone of gO other :2s:∅ does
not appear to correlate with anything else and may represent an idiosyncracy
of this argument combination.
Notice, however, that (41) (in dative prefixes, no vocabulary items realize
z =3s beyond those realizing x and y) extends to nearly all prefixes of the
form x :1/2s/d/i:∅, i.e., those used when the direct object is first or second
person, as in ‘x saw me/us/you’.
(61) x :y : z
∅ 3s
2i:1/3s: bâa∗ bâa∗
i:1/3s: êi∗ êi∗
a:1/3s: âa∗ âa∗
any:1d/p: dÓ dÓ
any:2i: bÓ bÓ
∅/2s/3s:1s: éM éM
1s:2s: em gyá
other :2s: gO gÓ
The two ‘exceptions’, ruled off from the others, are 1s:2s:z and other :2s:z .
The first may be disregarded for the reason given in the previous paragraph,
namely, that 1s:2s:∅ is some form of default agreement and so irrelevant
to generalizations concerning realizations of y. However, (41) suggests an
explanation for the second exception, other :2s:∅. If high tone realizes z =3s
in the context of y, then every prefix of the form x :y:3s will have high tone.
And if every y except 2s has high tone, then every prefix of the form x :y:∅
will have high tone, except x :2s:∅. Now, of course, there is considerable
redundancy in the determining of tone here. However, as this affects only a
single form, I am content to leave the exact solution open, noting only that
a solution is possible.
So, despite initial appearances to the contrary, all exceptions to rhyme
regularities can be captured through generalizations that emerge from more
careful examination of the agreement system.
Now let us consider the onsets. The onset g- for y=2s is expected given
the analysis of the singular y-prefixes; the same correlation was found there.
The onset d- for y=1d/p is straightforwardly accounted for:10
(62) d- ⇔ [y 1d/p ]
Finally, there is the onset b-. Given that this occurs for both 2i and 3i, it is
either a default or an accidental homophone. Amongst prefixes with second
person x , b- is frequent; however, it does not occur in prefixes with x =3i (or
other third persons). Thus, the approximate distribution is:
(63) 2 3i
x b- ∅-
y b- b-
Given (63), some accidental homophony may be unavoidable. Either b- gen-

erally realizes second person, in which case it is too highly specified to realize
y=3i; instead, a second, homophonous b- must be provided. Or it is the else-
where form for y-agreement, in which case it has the wrong specification to
realize x =2; again, a second, homophonous b- must be provided. The only
homophony-free possibility is to make b- a default for x and y, with the
zero in (63) representing a more highly specified vocabulary item. Surprising
though this situation would be, it is not impossible. However, it should be
noted that it does not obviate the need for accidental homophones, as we
will need more than one zero, to deal with 3s:∅ and with 2s:3s. So, little is
gained by regarding any:2i:z and any:3i:z as being realized by a single b-.
(64) b- ⇔ [y 3i ]
b- ⇔ [y 2i ]
10
The identification of the underlying form as dental necessitates a slight revision to
earlier statements. Specifically, the rhyme for z =3p cannot be simply át. Rather, it must
begin underlyingly in a segment that causes dentals to become velar. The obvious options
are ia. . . and ya. . . . We can rule out ya. . . owing to words like yal ‘hopefully’, yáŹpÓ
‘rope’, yátkyá ‘eight’: if the underlying form were ya. . . , as it is in yál ‘hopefully’, et
cetera, then ∅:3s:3s would be yán. However, if it is ia. . . , then there is no need for a
specific rule to delete i word-initially: the Vowel-in-hiatus rule will do this automatically.
(Clearly, then, the Vowel-in-hiatus rule must be ordered after the Glide formation. Note,
though, that i also deletes before b, as in 2s:3p bat, though not because by is an impossible
onset in Kiowa, witness paabyói ‘brother.inv’.) (1s:3s:3p and ∅/2s/3s:2s:3p yán have
initial y-, in contrast to ∅/2s/3s:3s:3p án, owing to opaque phonological processes. Yán
derives from g+ián by Glide formation, gyán, Nasalization, Nyán, and Engma-deletion,
yán. By contrast, in ∅/2s/3s:3s:3p án, there is never a consonant before ián, so that
Vowel-in-hiatus deletion applies, yielding án.)
184 CHAPTER 5
Summary
This concludes the discussion of ditransitive prefixes. The important point

to observe, in relation to the goals of this chapter and its context in the
investigation of morphosemantic number as a whole, is that Kiowa’s verb
agreement prefixes decompose, not just into traditional categories such as
‘singular’, ‘dual’, ‘first’, ‘second’, but into more finegrained ones. For in-
stance, observe that allomorphy relations and deletion operations are stated
with respect to, e.g., [−singular], the specific features argued for in Chapter
3. We will now see that the transitive prefixes support the same conclusion,
with, indeed, several of the segments identified above, identifiable there.
5.4.2. Transitives
We are now concerned with the following prefixes. (The notation ‘(:3a):’ is
dispensed with, in virtue (28).)
(65) Kiowa Transitive Agreement Prefixes
x: z
3s 3d 3p 3i 3a
1s: gya nen gyat dé de
2s: a men bat bé be
2i: bá∗ bet bát∗ bét bé
3i: é∗ et ét∗ ét ét
3a: á∗ et gyá∗ et ém
Singular x
3s:z prefixes were discussed on p. 175. Therefore, here, attention is confined

to 1s:z and 2s:z . These two are identical with respect to rhymes and are
strongly reminiscent of the y-rhymes of (58), which are included below:
(66) z
3s 3d 3p 3i 3a
singular x a en at é e
singular y á én án Ó
non-singular y Ó ét át Ót
The points of difference are tone, nasality for z =3p, the vowel for z =3i, and
the existence of distinct z =3a forms. However, none poses major difficulties.
First, with regard to tone, observe that the singular x -rhymes have low
tone throughout, except for z =3i. Here, instead of O, which one might expect,
extrapolating from the ditransitive rhymes, we have é. As both vowel and
tone are unpredictable, they can be taken as idiosyncratic properties of the
vocabulary item (as (67) applies only when there is no y, other statements
about [z 3i ] must be contextualized to y, as (59) was):
(67) é ⇔ [z 3i ]
Second, with regard to existence of distinct z =3a forms, recall that this
agreement category is subject to morphological deletion operations (Section
4.5.2). Therefore, I assume that the distinctive realization of a-agreement
depends on these operations not having occurred.
(68) e ⇔ [z 3a ]
Finally, with regard to nasality, I take nasality for z =3d and non-nasality
for z =3p as the norm. The former follows immediately from (38), with the
latter forming the complement case. Examples are 1s:3d nen and 2s:3d
men, as against 1s:3p gyat and 2s:3p bat. The total lack of nasality in the
non-singular y row of (66) follows from (59). The nasality of z =3p forms in
the middle row of (20) requires additional explanation, however.
(69) [+nasal] ⇔ [z 3p ] / [x −singular] [y 3s ]
The more general condition on nasality, given (38), is:
(70) [+nasal] ⇔ [ d ]
The details of all of these allomorphic variations should not obscure the
main point, namely, that many of the regularities and patterns from the
ditransitive prefixes carry over in large measure to the transitive system.
So saying, let us consider the onsets.
186 CHAPTER 5
(71) x: z
3s 3d 3p 3i 3a
1s: gy-a n-en gy-at d-é d-e
2s: ∅-a m-en b-at b-é b-e
The second person onsets are reminiscent of the ditransitive system (see
the discussion preceding (64)), and so one is led to:
(72) b- ⇔ [x 2s ]
(The m- in 2s:3d arises standardly by nasalization.) However, observe that

2s:3s has no b-. It, a, consists of the expected rhyme without an onset. So,
either there is a zero vocabulary item that realizes x =2s in the context of
z =3s, or the features in (72) ordinarily realized by b- have been deleted by a
rule, contextualized by z =3s, prior to vocabulary insertion. I am not aware of
a general principle to decide between the allomorphy- and the deletion-based
solutions, but for concreteness adopt the latter.11
|
⎡ x ⎤
(73) −author → ∅ / [s]
⎢+hearer ⎥
⎢ ⎥
⎣+singular ⎦
−augmented
Observe that there is no requirement that z , rather than y, be s. So under-

specified, (73) correctly predicts absence of the onset b- from 2s:3s:z and
2s:1s:z , respectively the top and bottom rows of (51).
First person onsets are simpler; indeed, they are predictable. Adding g-
(54) to the rhymes in (66) yields the correct results by standard phonology:
gya(t) from g+ia(t) by Glide formation, d(é) from g+(é) by Dental-velar
11
That allomorphy and deletion can achieve the same surface result may point to a
deficiency in the system: how, one might imagine, is the child to decide between them?
The only attempt I am aware of to address this redundancy is Trommer’s (1999) for-
malization of Distributed Morphology, in which all morphological operations are reduced
to vocabulary insertion. Two morphological operations deployed above—feature insertion
and node (as opposed to mere feature) deletion—are not obviously replicable as vocabulary
insertion; nor are non-structure-preserving morphological operations, which may establish
or disestablish the adjacency relations taken to be central to suppletion in Chapter 4.
Given the inherent interest of Trommer’s proposal and of the issue that motivates it, these
criticisms should prompt refinement, rather than abandonment, of the work.
switching, nen from g+en by Dental-velar switching and Nasalization.

Thus, many of the generalizations over ditransitive prefixes apply to sin-
gular x transitive prefixes, and some new generalizations apply equally well
to the ditransitive ones.
Non-singular x
We are now concerned with the following prefixes.
(74) x: z
3s 3d 3p 3i 3a
2i: bá∗ bet bát∗ bét bé
3i: é∗ et ét∗ ét ét
Little beyond what has already been said is required here.
The most striking fact here is the distribution of ∗. It is restricted to
z =s/p, which form a curious natural class, being oppositely specified for
[±singular] and [±augmented]. Superficially, we can write:
(75) ∗ ⇔ [x −singular] [z ±singular ∓augmented]
However, this has the problem that z =3i is predicted to be tone lowering,
as i is of the general form [−singular −augmented +singular/augmented].
To avoid this, it is necessary to look beyond the notational device ‘∗’ to the
underlying tonal mechanisms. Following my own analysis (Harbour 2002),
the tonal/metrical constituency of tone lowering ∗-prefixes is ϕ)V, that is, a
separate foot in the same tonal/metrical domain as the verb. So, (75) really
describes the conditions under which a prefix contains a right parenthesis
(forms its own foot). For simple (non-∗) high prefixes, the tonal/metrical
constituency is ϕ|V or ϕ)|V, that is, a domain separate from the verb. So,
if z =3i is realized both by the foot demarcator ‘)’ and additionally by the
domain demarcator ‘|’, then it is consistent with (75) but does not tone lower.
Moving on to non-tonal segments, consider 3i:z . Recall that non-singular
higher arguments block nasalization by d and that they condition a coda
when z =3i (59). Here, we see this again, with the 3i:3d, et, being non-nasal
and with 3i:3i, ét, having a coda. Recall also (60), according to which, if a
higher argument is i, then we find the vowel e, instead of the usual z vowel.
This applies here too. (We return to 3i:3a, momentarily.)
188 CHAPTER 5
Consider 2i:z . The onset is predictably b-, given (64) and the restriction
of (24) to singular. Compare the rhymes of 2i:z and 2s:z :
(76) x: z
3s 3d 3p 3i 3a
2s: a en at é e
2i: á∗ et át∗ ét é
Again, effects of (59) are evident in the non-nasality of 2i:3d and the coda
on 2i:3i. Further changes concern the tone of 2i:3s/p, which have high tone
and ∗, whereas 2s:3s/p simply have low tone. This is predicted by (75).
For z =3a, observe that 2i:3a, bé, is simply 2s:3a, be, with high tone.
Interestingly, a similar relation holds form 3a:3a. It, ém, is simply 3s:3a,
em, with high tone. This suggests:
(77) a. If ϕ ⇔ [x +singular F] [z 3a ], then ϕ ⇔ [x −singular F] [z 3a ]

b. ´ ⇔ [x −singular F] [z 3a ]
The prefix 3i:3a requires particular comment. It is the only z =3a pre-
fix with a t-coda. Others, by contrast, derive from a regular onset plus a
predictable vowel plus predictable tone. For instance, 1s:3a is g+e; 2s:3a is
b+e; 2i:3a is b+e+´; for 3s:3a, the form is unexpected, but given it, em,
3a:3a is predictable as em+´. With its coda, 3i:3a ét looks quite different.
In fact, its form is predictable, if we apply the mechanisms that account
for a-agreement’s y-conditioned d∼i alternation. Applied exceptionally to
3i:3a, the result is 3i:3d/i, which, correctly, is ét.
(78) The mechanisms of the d∼i alternation apply to 3i:3a.
Observe that, in the preceding paragraph, 3s was treated as the singular

correspondent of 3a, for the purposes of applying (77). The z =3p prefixes
reinforce this correlation. Consider the relationship between 2s:3p and 2i:3p,
bat and bát∗, which vary only with respect to tone. Exactly the same
relationship holds between 3s:3p and 3a:3p, gya and gyá∗. However, for
other prefixes, any correlation is less obvious:
(79) x: z
3s 3d 3p 3i 3a
If the onset for x =3a is ∅, then 3a:3d is as expected: ∅+et is et. (Note that
3a, being [−singular], conditions the non-nasal allomorph of z =3d agree-
ment.) Given this, 3a:3i has the expected onset, vowel and coda ∅+e+t,
and is surprising only in lacking high tone.
Thus, the analysis of the ditransitive prefixes extends naturally to the
transitives with minor modifications. Note, again, the crucial reference to
features of earlier chapters, for instance in creating a natural, if marked,
class of s and p (75) and the analysis of 3i:3a (78).
5.4.3. Intransitives
We now turn to the last prefixes in the system, namely, the intransitive.
(80) x: z =∅
1s: a
2s: em
2i: ba
3s: ∅
3i: e
3a: á
Given the preceding discussion, there is little to surprise us here. 2s:∅, em, is
a special realization of the features common to 2s intransitive and 2s indirect
object agreement (p. 181). The correlation between intransitive and indirect
object agreement is suggestive also for 1s:∅ and 3s:∅, neither of which was
realized by an onset when the values of y. Here, again, we find them onsetless.
This correlation cannot be extended, however, to non-singular intransitives.
If it could, 3i:∅ would have onset b-, as any:3i:z prefixes have. Nonetheless,
the non-singular prefixes are reminiscent of other prefixes. Specifically, for
non-singular x, x :∅ is predictable from x :3s.
(81) x: z
∅ 3s
2i: ba bá∗
3i: e é∗
3a: á á∗
If the x :∅ forms are taken as basic, then addition of ∗ yields x :3s. (Recall
that the metrical reality behind ∗ entails that, if a prefix is ∗, then it has
190 CHAPTER 5
high tone, p. 187.) This suggests that the x :∅ forms are, for non-singular
prefixes, the basic vocabulary items for these persons.
(82) a. ba ⇔ [ 2i ]
b. e ⇔ [ 3i ]
c. á ⇔ [ 3a ]
If this reasoning is more or less correct, then the only additional expla-
nation required for intransitive prefixes is the occurrence of a for 1s:∅.
(83) a ⇔ [ 1s ]
5.5. Conclusion
The point of the foregoing micromorphological argumentation is that it places

us in a position to give a vocabulary list with allomorphic variants which
map prefix structures that have been acted on by morphological rules into
phonological segments that yield familiar prefixes by regular phonology.
At this point, though we are clearly nearing a solution, I lay the matter to
rest. As emphasized above, the analysis of the agreement prefixes is relevant
to the current investigation as a means of justifying the existence of the
features and values argued for herein. By showing that complex allomorphy
conditions and morphological operations are statable precisely in these terms,
that aim has been achieved above. I leave a complete and final analysis, down
to the level of a full and fully compositional vocabulary list, for future work.
5.6. Appendix
Although the Kiowa prefix system has been the object of extensive investiga-
tion for nearly a century now, one aspect of it has escaped previous attention:
the syncretic behavior of ∅:3y:3a. Ordinarily, y-agreement triggers the d∼i
alternation for z =3a. For ∅:3y:3a, however, the alternation does not occur.
Furthermore, this is (I tentatively claim) the only argument combination
where it is significant whether z =3a arises from an animate plural or from
a reflexive. Let us distinguish these cases as ∅:3y:3a and ∅:3y:refl. For
animate plurals, we apparently have the syncretism ∅:3y:3a ≡ 3y:3a; that
is, for third persons, action on and possession of an animate plural results
in identical agreement. For ∅:3y:refl, matters are more complex: note the
odd switching of s for p and p for s.
(84) x: :z y
3s 3d 3p 3i
∅: :refl 3p:3a ∅:3d:3i 3s:3a ∅:3i:3i
As these facts have only newly come to light, they require further inves-
tigation. However, in the context of the current chapter, which is merely a
detailed sketch, not a full solution, this open-ended aspect is not detrimental.
Chapter 6
Conclusions and Consequences
Despite its crosslinguistic frequency, noun classification is commonly con-

ceived of as an arcane and arbitrary feature of language, something possessed
of little less arbitrariness than the Saussurean sign itself. The preceding chap-
ters, investigating one such system, that of Kiowa, have shown that, far from
being an isolated fact about individual lexical items, the mechanisms of noun
classification are deeply embedded in the grammar, exhibiting effects in the
semantics, syntax, and morphology. In this final chapter, I wish to consider
what Kiowa suggests for the treatment noun classification more generally.
6.1. Noun classification
At first glance, the Kiowa-Tanoan inverse is an exotic morpheme that switches

or toggles number between opposing values. However, we have now seen that
the inverse does no such thing as toggle. Rather, it is simply a response to
feature conflicts that arise on D from the interaction of Class and Number.
With the veil of exoticism lifted, it is useful to ask to what extent Kiowa’s
noun class system resembles those of other languages, such as the gender
and declension class systems of Indo-European, or the gender and number
systems of Bantu. Kiowa, it emerges, is like Bantu, but distinct from Indo-
European. To show this, I consider an example of each system, Sanskrit for
Indo-European, Kiswahili for Bantu, and show that it is possible to combine
the core characteristics of Kiowa and Sanskrit (Section 6.1.1), but that it is
impossible to do likewise for Kiowa and Kiswahili (Section 6.1.2). This shows
that Kiowa and Kiswahili represent manifestations of the same system.
193
194 CHAPTER 6
6.1.1. Gender and declension class
An important distinction inherited by generative grammar from traditional

grammar is that between gender and declension class (see, e.g., Aronoff
1994). Though both are (mostly) arbitrary properties of individual lexi-
cal items, they are crucially different. Gender ‘propagates’ throughout the
syntax, whereas declension class is confined to the individual noun itself. For
instance, consider three Sanskrit noun phrases (based on Coulson 1992):
(1) rājā kāntah.

king.nom.s beloved.masc.nom.s
‘beloved king’ -an class
(2) ātmā kāntam
self.nom.s beloved.neut.nom.s
‘beloved self’ -an class
(3) suhr.t kāntah.
friend.nom.s beloved.masc.nom.s
‘beloved friend’ consonant-final class
The notion of gender is illustrated by the adjectives, that of declension class

by the nouns. The nouns rājan ‘king’ and ātman ‘soul’ belong both to
the -an declension class. Accordingly, in the nominative singular, they end
in ā. By contrast, suhr.d ‘friend’ belongs to the consonant-final declension
class, and so, in the nominative singular, ends in zero, with the consonant
devoicing. Thus, from the point of view of case endings, ‘king’ and ‘soul’
form a natural class to the exclusion of ‘friend’.
However, when we turn to the adjectives, we find that ‘king’ and ‘friend’
form a natural class to the exclusion of ‘self’. For the former, ‘beloved’ ends
in -ah., but, for the latter, it ends in -am. According to the traditional
designations, ‘king’ and ‘friend’ are ‘masculine’, ‘self’ ‘neuter’. Clearly, these
genders are independent of declension class, as the two classificatory schemes
crosscut.1 Declension class determines only the form of the morpheme that
1
The fourth possibility, neuter gender, consonant-final declension, is manas ‘mind’.
Nominative singular on the noun is zero and on the adjective -am; by regular phonology:
(i) manah. kāntam

mind.nom.s beloved.neut.nom.s
‘beloved mind’ consonant-final class
CONCLUSIONS AND CONSEQUENCES 195
realizes nominative singular, and other case-number combinations, on the

noun itself, such as ā for the -an declension class, zero for the consonant-
final declension class. Gender determines the form of the morphemes that
realize nominative singular, and other case-number combinations, on agreeing
elements beyond the noun, such as -ah. for masculine, -am for neuter.2
With this distinction in mind, we can now ask:
Does Kiowa have a gender or declension class system?
Kiowa does not have a gender or declension class system. The easiest way to
see this is by sketching languages, Kiowa and Kiowa , that combine Kiowa’s
noun class system with a declension class and a gender system, respectively.
To sketch Kiowa , let us first simplify Kiowa by removing all allomorphy of
the inverse marker (Section 2.6.1), replacing them with two allomorphs -inv
and -erse. We now define two declension classes, the -inv declension class
and the -erse declension class, as follows. Take all the nouns in Harrington
(1928) and assign them to the -inv class. All other nouns are assigned to
the -erse class. This information is represented on the vocabulary entries.
For example, Harrington gives the Kiowa for ‘stick’ but not ‘walking stick’.
Consequently, ‘stick’ belongs to the -inv and ‘walking stick’ to the -erse
declension class in Kiowa . The vocabulary items are:
√
(4) stick ⇔ áá-inv class
√
(5) walking stick ⇔ t!optéMaa-erse class
The superscripts are ‘class diacritics’ (cf., Harris 1991) that condition the
realization of feature conflict on D, that is, the inverse marker:
(6) [−F +F] ⇔ -inv / [ N-inv class ] [ D ]

⇔ -erse / [ N-erse class ] [ D ]
Note that the assignment of nouns to declension classes affects only the form
of inv the inverse suffix on nouns. It does not affect the noun classes them-
2
As an example of other things the endings of which are determined by gender, consider
a past participle, such as pratibuddha ‘awoke’. In the masculine singular nominative,
pratibuddhah. may be combined with the masculine nouns above to yield Rājā/Suhr.t
kāntah. pratibuddhah. ‘The beloved king/friend awoke’; and in the neuter singular nom-
inative, pratibuddham may be combined with the neuter nouns to yield Ātmā/Manah.
kāntam pratibuddham ‘The beloved soul/mind awoke’.
196 CHAPTER 6
selves, such as sdi, idp; nor does it affect the form of i-agreement on the
verb. The possibility of combining Kiowa’s noun class system with declension
classes that crosscut them, Kiowa , shows that Kiowa noun classification is
distinct from declension classes.3
By way of illustration, consider the idp nouns ‘stick’ and ‘walking stick’ in
the sentence frame ‘It’s a ’. Kiowa (7) and (8) are analogous to Sanskrit
(1) and (3), the -an and consonant-final masculine nouns.
(7) Áá- inv e- dÓÓ

stick-inv 3i-be
‘It’s a stick’ Kiowa , -inv class
(8) T!optéMaa- erse e- dÓÓ
walking stick-inv 3i-be
‘It’s a walking stick’ Kiowa , -erse class
To sketch Kiowa , we define two genders, inv and erse, by assigning all
nouns in Harrington (1928) the gender inv and all others erse, where the
phonological reflexes of these genders are, say, creaky voice and lip rounding,
respectively. That is, if N, a Kiowa noun, has gender inv in Kiowa , then,
wherever N triggers the inverse marking or agreement form ψ in Kiowa, it
will trigger ψ + creaky voice, ψ , in Kiowa . For example:
(9) [ÁádO e- ét] -gO dé- hÓÓgya

stick.inv 3i.inv -big.s-inv.inv 1s:3i.inv -get.pf
‘I got a stick that is big’ Kiowa
And if N has gender erse in Kiowa , then, wherever N triggers the inverse
marking or agreement form ψ in Kiowa, it will trigger ψ + rounding, ψ, in
Kiowa . For example:
(10) [T!optéMaadO e- ét] -gO dé- hÓÓgya

walking stick.inv 3i.inv -big.s-inv.inv 1s:3i.inv -get.pf
‘I got a walking stick that is big’ Kiowa
Observe that in both (9) and (10), the inverse marking on the noun is exactly
as it would be in Kiowa and, moreover, that this marking is unrevealing of
3
Note that the claim here is that Kiowa is a logically possible language. I make no
claim as to whether Kiowa , or any other of the grammatical hybrids below, are acquirable.
gender, as in (1) and (2).4 Combining Kiowa’s noun class system with a
gender system that crosscuts these classes, Kiowa shows that Kiowa noun
classification is distinct from gender classification.
6.1.2. Gender-number systems
A system more similar to Kiowa’s than those considered above is that of

Bantu, or, more precisely, Kiswahili on Carstens’ (1991) analysis.5 It is
typical of Bantu languages in possessing a large number of noun classes. Each
is associated with a class-particular nominal prefix and with ‘a distinctive
pattern of agreement borne by modifiers and arguments of the noun, and on
auxiliaries and predicates in the relevant syntactic relations to it’ (pp. 2–3).
Some classes and examples are provided below (cf., Carstens 1991, p. 3).
(11) Class Example Gloss

1 m-tu person
2 wa-tu people
3 m-ti tree
4 mi-ti trees
5 gari car
6 ma-gari cars
7 ki-atu shoe
8 vi-atu shoes
9 n-yumba house
10 n-yumba houses
11 u-bao board
.. .. ..
. . .
4
The syntactic mechanisms that give rise to the different forms of agreement and inverse
marking in Kiowa are, I assume, the same as in Sanskrit, and other languages, in which
adjectives, verbs and so on agree for gender and number. This would be implemented by
having D bear number and gender features simultaneously, rather as it bore person and
number features in the treatment of sii syncretisms. Thus, in the preceding examples, D
would bear [−F +F inv -gender] or [−F +F erse-gender]. Other gender-number combina-
tions are easily representable: for example, for an sdi noun of gender inv and referential
cardinality 2, D would be [−singular −augmented inv -gender].
5
The language is more commonly called ‘Swahili’ in English. I follow Carstens’ termi-
nology; see her footnote 1, p. 1.
198 CHAPTER 6
(12) Mtoto huyu wangu mzuri a- me-anguka

1.child 1.this 1.my 1.good 1.agr-pf- fall
‘This my good child has fallen down’
(13) Watoto hawa wangu wazuri wa- me-anguka
2.child 2.this 2.my 2.good 2.agr-pf- fall
‘These my good children have fallen down’
(14) Mti huu wangu mzuri u- me-anguka
3.tree 3.this 3.my 3.good 3.agr-pf- fall
‘This my good tree has fallen down’
(15) Miti hii yangu mizuri i- me-anguka
4.tree 4.this 4.my 4.good 4.agr-pf- fall
‘These my good trees have fallen down’
The traditional views in Bantu and Kiowa linguistics make the two lan-
guages’ noun classification systems seem quite dissimilar. Bantuists have
considered ‘Class, an amalgam of number and gender, . . . to be a lexical
property of Bantu nouns and/or their prefixes’ (p. 6). Generalizations about
Classes are then of the form ‘If a noun is singular in 1, it is plural in Class 2’, or
(cf., p. 28) ‘If a Class 11 noun triggers Class 3 agreement in the singular, it
triggers Class 10 agreement in the plural’. Traditional Kiowa descriptions,
such as Wonderly, Gibson, and Kirk (1954), view referential cardinality as
part of the inherent meaning of the noun; for instance, tógúl means ‘one or
two young men’, with the inverse suffix giving the inverse number. (Such
approaches, of necessity, deemphasize mass nouns and pluralia tantum.)
Carstens, however, gives an analysis of the morphology and syntax of
Kiswahili DPs that solves several problems in traditional Bantu linguistics
and at the same time makes Kiswahili more similar to Kiowa, on the analysis
given above. Specifically, she claims that there is no primitive notion of
Class that amalgamates gender and number. Rather, these belong to distinct
projections in the syntax. Modifying her structure slightly, to emphasize
commonalities between our analyses, mtu ‘a person’ has the structure:6
6
The differences are (a) that Carstens writes ‘Group A’ where I have ‘Gender A’—the
extra term does not benefit the discussion here; (b) that the projections in Carstens’ tree
are labelled category-neutrally—the functional structure of the DP is dealt with later in
her dissertation; (c) that the content of Number is not [+singular −augmented]. The last,
for uniformity with Kiowa, is the only controversial alteration. I believe that the content
of Number varies crosslinguistically; in Kiswahili, it is probably just [±singular].
(16) DP
⎡ D ⎤ NumberP
+singular
⎣−augmented⎦
Gender A Number Class: person

+singular Gender A
m −augmented
tu
On Carstens’ analysis, prefixes like m- are the realization of gender and

singular∼plural alternations such as m-∼wa- are contextual allomorphy:7
(17) [Gender A] ⇔ m- / [D +singular ]

⇔ wa- / [D −singular ]
The point of contact between Carstens’ analysis of Bantu and the current
analysis of Kiowa is that, in both cases, what has been traditionally treated
as class-specific morphology is analyzed instead as the realization of features
on D. Moreover, these features come to be on D by copying from lower
heads, Class and Number. Kiowa and Kiswahili differ, however, in that
the classifying features of Kiswahili are not number features. Consequently,
copying Class and Number onto D cannot result in feature conflict.
A further point of contact between Kiowa and Kiswahili is that both
are compatible with a declension class system. This was shown above for
Kiowa and similar facts are found in Kiswahili. Animate nouns provide one
example of this. All animates trigger, on their modifiers, arguments and
predicates, 1-agreement forms in the singular and 2-agreement forms in the
plural. However, animate nouns themselves do not always bear the 1∼2
prefixes m-∼wa-. For instance, vifaru ‘rhinos’ displays the Class 8 prefix
vi-. However, it triggers Class 2 agreement on its modifiers and predicates,
as in:
(18) Vifaru wawili wa- na- pigana

8.rhino 2.two 2.agr-pres-fight.recip
‘Two rhinos are fighting’
7
Alternatively, m-/wa- could realize [±singular Gender A] on D.
200 CHAPTER 6
These facts are reminiscent of the Sanskrit cases treated above, where mark-
ing on the noun is not a reliable predictor of agreement forms elsewhere.
Carstens’ analysis is that all animates have Gender A (i.e., are Class 1 when
singular, Class 2 when plural), but that some, such as ‘rhino’, are lexically
marked as conditioning prefixes from different classes on themselves. So, the
noun imposes a declension class on its own prefix, above and beyond any
morphological forms that it imposes elsewhere in the sentence.
Note further that the Kiowa and Kiswahili systems are not wholly mutu-
ally compatible, as expected if they are different manifestations of the same
mechanisms. Given that D in Kiowa does not always have number features
that reflect the content of Number, there can be no language Kiowa in which
nouns bear inverse morphology for certain referential cardinalities and where
agreement forms of nouns’ modifiers, arguments and predicates consistently
reflect referential cardinality. Thus, Kiowa’s noun classification system is
akin to that of Kiswahili and so is a gender-number system.
6.2. Parting comments
This investigation has had both empirical and theoretical aims.

Empirically, it has sought to provide an analysis of a variety of seemingly
unrelated phenomena in Kiowa grammar: the distribution of agreement types
and inverse marking, the semantic characteristics of noun classes, the struc-
ture of adjectivally or deictically modified DPs, the inventory of noun classes,
the motivation and nature of agreement∼suppletion mismatches, the nature
of Kiowa’s agreement prefixes. Theoretically however, it has suggested a line
of inquiry—research into a unified morphosemantic theory of number—that
strikes me as important and interesting, and has offered some elements of
what such a theory of number might comprise. If this program of research
is accepted by others, then the immediate goal is to find other empirical
domains, such as the nature of distributivity, or associative plurals, or plu-
ractionality, in which semanticists and morphologists have equal stakes.
By thus widening the domain of inquiry of both morphologists and seman-
ticists, a broader range of data is brought to bear on core problems than has
until now been considered relevant. Affording the total theory of Universal
Grammar one theory of number, the morphosemantic, rather than two, the
morphological and the semantic, we make it ontologically more constrained,
descriptively more adequate and explanatorily more powerful. This, I believe,
is what linguistics should be about.
Appendix A
A Hunting Story
This appendix presents the short text quoted in Section 2.5. It was recorded
from Mr Vincent Bointy in August 2001, at Anadarko, Oklahoma and tran-
scribed (with some amendments) under his supervision. It illustrates several
phenomena discussed in the preceding chapters, including the use of number-
conditioned suppletive predicates. The accuracy of the glossing owes much
to Laurel Watkins’ comments on an earlier draft.
The following points explain aspects of the glossing, such as policy deci-
sions and tricky cases. For discussion of individual grammatical points, see
Watkins (1984) or the more specific references cited below.
1. Kiowa employs a switch reference system for clause linking (Watkins

1993). The switch reference markers are simply glossed as conj. As a
rule of thumb, gO (and its derivative forms, such as gigÓ) indicates con-
tinuity of reference, whereas nO (and its derivative forms, such as negÓ)
indicates discontinuity of reference. (One might, on these grounds, have
expected nO instead of gO in the sentence after Kyáp á-ŹˇŹˇbaahel ‘They
went on hunting in the other direction’. If this expectation is correct, it
is possible that the process of transcription, with its frequent interrup-
tions and repetitions, disrupted the contintuity of the discourse and,
so, of reference tracking.)
2. Morphemes’ surface tones depend on their position in the word and

their wordmate morphemes (Harbour 2002). Glosses below reflect sur-
face, not underlying tones. Hence the tonal variation between dÓˇÓˇ∼dǑǑ
and mêi∼mei in:
201
202 APPENDIX A
(a) gya-tháp-dÓˇÓˇ-mêi (b) gya-hóMOn-dǑǑ-mei

3p- dry- be- impf.hsy 3p- path-be- impf.hsy
‘They were dry’ ‘There was a path’
3. As Mr Bointy did not participate in the events described, hearsay forms
are used throughout the narrative (though not in recreation of partici-
pants’ speech).
4. Two distinct morphemes are glossed as fut: tOO and t!OO, used for
verbs with external arguments and verbs without, respectively. E.g.:
(a) éM- thÓtté-tOO (b) ba- thóúgyâi-t!OO
3s:1s-shoot- fut 1p.in-pass.pf- fut
‘He will shoot me’ ‘We will pass’
5. The agreement prefix in En-thÓttétOO ‘He will shoot them [two]’ is
unexpected, as 3s:3d is ě, not en. Watkins (p.c.), who has also en-
countered such forms, suggests that n is a phonologically conditioned
variant of the pure nasal.
6. Agreement in Gya-gúúdÓÓÔˇ ǑmO ‘They were singing hunting songs’ is
puzzling. Gya is either 3s:3p or 3p. However, 3s:3p fails to capture
the plurality of the agent, and 3p, which is intransitive, fails to match
the transitivity of the verb root. Possibly, I misheard the verb, which
should be detransitive, Gya-gúúdÓÓÔˇ Ǒma; the prefix 3p then makes
sense and an impersonal translation is appropriate ‘There was hunt
song singing going on’ (cf., German ‘Es wurde(n Jagdlieder) gesun-
gen’). Alternatively, singular agents, as in 3s:3p, may be a grammatical
default (cf., the paragraph beginning ‘They got ready to sleep’); this
remains to be investigated.
7. Square brackets around parts of the translation indicate elements not
strictly part of the Kiowa original but which are implied by it and aid
in reading the English.
8. ÚúúŹ is a narrative element seeming to indicate (traversal of a) dis-
tance. It may be related to the distal óŹ, as in the demonstratives óŹde
‘that’, óŹgO ‘that.inv’. It is not glossed below.
9. Several different morphemes—em, ěǐ, pa—are glossed as loc. Their
differences are discussed by Watkins (1984).
APPENDIX A 203
ÍˇŹˇhěǐtegya
Phá ˇ á
ˇ o k!yá
ˇ á
ˇ hyóp e-ŹˇŹˇǎǎhel áádom. SyÓndé gya-hó ˇOndǑǑmei déem
téŹ e-khOléŹǎǎhel. GO e-xándehel ÓgO áádO e-bôndǑǑmei déěǐ gO
ˇOn dómgyá ÓphOO gya-xóúdêi. NO ’gO t!ópá
hó ˇ á
ˇ hêlde tó
ˇ ú
ˇ nêi: ‘Xáá!
ÓgO áádO e-bôndOOde dómgyá ba-thóúgyâit!OO. Bát-mOOhol, hÓˇÓˇ
bá-salOm. ÓÓkO, áábǒǔdOdethai khó ˇ ú
ˇ gya t!ámde mOOkyáǐǐk!OO. Bát-
mOOkholdOOpe. Dómbéba-khûtdet!OO. Háyátto tháŹde khyâigunt!OO.’
A Hunting Story
Three men were hunting in the woods. They were all going along a narrow
path when they arrived where the path lay under a bent tree. The one in
front said: ‘Friends! We’re going to go under that bent tree. Let’s get ready
and load the guns. There, up in the bent tree, there’s a black panther lying
in wait. Let’s be ready: we’re going to pass underneath, and maybe he’ll
pounce on [us].’
PháM áM o k!yáM áM hyóp e- MŹ

ŹM- aM aM - hel áá- dom. SyÓn- dé gya-hóMOn-dOM OM -
three man.inv 3i-hunt-come-hsy tree-loc small-adv 3p- path- be-
mei dé- em téŹ e- khOléŹ- aM aM - hel. GO e- xánde-hel ÓgO
impf.hsy nom-loc all 3i-together-come-hsy conj 3i-reach- hsy rel
áádO e- bôn- dOM OM -mei dé- eM Mi gO hóMOn dómgyá ÓphOO gya-
tree.inv 3i-bend-be- impf.hsy nom-loc conj path under there 3p-
xóú-dêi. NO ’gO ∅- t!óp-áM áM - hêl- de ∅- tóMúM - nêi: ‘Xáá!
lay- impf.hsy conj rel 3s-first-come-hsy-nom 3s-say-impf.hsy friends
ÓgOáádO e- bôn- dOO-de dómgyá ba- thóúgyâi-t!OO. Bát-
rel tree.inv 3i-bend-be- nom under 1p.in-pass.pf- fut 1p.in:3p-
mOOhol, hÓMÓM bá∗- salOm. ÓÓkO, áá- boM uM dO- de- thai khóMúM gya
ready.imp gun 1p.in:3s-load.imp there tree-bend.inv-nom-above black
t!ámde ∅- mOOkyá-MiMi- k!OO. Bát∗- mOOkhol- dOOpe. Dómbé ba-
panther 3s-ready- hunt-lie 1p.in:3p-ready.inc-be.imp under 1p.in-
khûtde- t!OO. Háyátto tháŹ- de ∅- khyâi- gun- t!OO.’
pop/flee out.detr-fut maybe on top-adv 3s-stretch-jump-fut
204 APPENDIX A
NO thÓp á-khûtdehel. T!Omá ˇ á

ˇ dé gôm gya-bó ˇ ú
ˇ hêl nO héˇŹˇ dÓˇÓˇmêi.
Tékhop á-donhel né hÓn á-thǑǑmOO. Ém-xáŹhel: ‘Hagya t!ámde
óbÓŹk!OO?’
Kyáp á-ŹˇŹˇbaahel gO páá pêˇǐ hólhel. ÚúúŹ ÓgO thó ˇ ú
ˇ OtkhÓngyá
déem, thó ˇ ú
ˇ k!ÓÓpa, ÓŹhyOO ém-k!úúhel. K!ŹŹêlxeihel ÓŹhyOO dépeidou
hÓn hÓndé hÔigO Ómdét!OOdei. ÓŹhyOO pêˇǐ á-pǐǐOmhel gO á-hanhel.
So they passed [quickly] to the other side. The one at the head looked
back and it was gone. They looked for it all over but didn’t find it. ‘Was it
really there?’ they asked themselves.
They went on hunting in the other direction and one of them killed a
turkey. Well, way over where the water came to an end, against a bank,
that’s where they camped. They built a big fire so that nothing would come
near and there they cooked the turkey and ate it.
NO thÓp á- khûtde- hel. ∅- T!Om-áM áM - dé gôm gya- bóMúM - hêl
conj beyond 3a-pop/flee out-hsy 3s-first- come-nom back 3s:3p-look-hsy
nO héMŹM ∅- dÓM ÓM-mêi. Tékhop á∗- donhel né hÓn á∗-
conj gone 3s-be- impf.hsy everywhere 3a:3s-seek-hsy but neg 3a:3s-
thOM OM -mOO. Ém- xáŹ- hel: ‘Hagya t!ámde ∅- óbÓŹ- k!OO?’
find- neg 3a:3a-ask-hsy possibly panther 3s-really-lie
Kyáp á- MŹ

ŹM- baahel gO páá pêMiM ∅- hól-hel. ÚúúŹ ÓgO
other way 3a-hunt-come-hsy conj some turkey 3s:3s-kill-hsy . . . rel
thóMúM ∅- OtkhÓn-gyá dé- em, thóMúM - k!ÓÓ- pa, ÓŹhyOO ém- k!úú-
water 3s-end- detr nom-loc water-bank-loc there 3a:3a-camp.s/d-
hel. ∅- K!ŹŹ- êl- xei- hel ÓŹhyOO dé- peido hÓn hÓndé
hsy 3s-wood-big.s-lay.s/d-hsy there nom-because neg something
hÔigO ∅- Óm-dé- t!OO- dei. Ó ŹhyOO pêMiM á∗- piMMi- Om- hel gO
near 3s-do- detr-fut-hsy there turkey 3a:3s-food-make-hsy conj
á∗- hanhel.
3a:3s-eat up-hsy
APPENDIX A 205
GO ’igO ém-hêˇǐtesOOhel phŹŹÓÓk!obe. NegÓ páá tó ˇ ú

ˇ nêi: ‘SÓt a-
péŹgyaxan, úúúŹ, gya-gúúdÓÓÔˇǑmO gO éˇŹˇgO dó
ˇ ú
ˇ gya a-héŹba gO éˇŹˇhOde-
dOOgya gya-hâapO: “HOyO hOO, . . . ” ’
Án-pháttéhelxěǐ, mááde hÓndé án-xátháádehel gO pâagO-dOOgya
hâapehel. ÚúŹde án-dÓÓpó ˇ ú
ˇ á
ˇ á
ˇ hêl. Mámdexo mÓÓ em-dÓÓtÓÓyẐi.
Khodêide mám khûtdehel. . .
Gya-děǐmOOhólhel. K!ÓÓpa gya-hŹnhel. Ó ˇÓˇgO gya-thélhel, ÓˇÓˇgO
héŹbet!OO déotte gya-hŹnhel.
Then they sat down around the fire to tell stories. One said: ‘I just
thought of [this time], way back, they were singing hunting songs and I went
down into the center and I started to sing this song: “HOyO hOO, ...” ’
When he finished, something upstream gave a shout and started to sing
the same song. And the sound of singing got closer. It was like the singing
was coming from above. Suddenly, it flew right over them. . .
They got ready to sleep. They dug into the bank. Each one opened up
[a hole], each one dug just enough to get into.
GO [he]gÓ ém- hêMiMte- sOO-hel phŹŹ-ÓÓk!obe. NegÓ páá

conj then 3a:3a-tell stories-sit- hsy fire- around conj.then some
∅- tóMúM - nêi: ‘SÓt a- péŹgya- xan, úúúŹ, gya-gúú- dÓÓ-
3s-say-impf.hsy just 1s-thought-arrive.pf ... 3p- good shot-song-
ÔMOM mO gO éMŹMgO dóMúM gya a- héŹb- a gO éMŹMhO-de- dOOgya
do.impf conj there down 1s-enter-intr.pf conj this- nom-song
gya- hâap-O: “HOyO hOO, . . . ” ’ Án- phátté-hel- xeM Mi, mááde
1s:3s-raise- tr.pf ...... :3s:3p-finish- hsy-when upstream
hÓndé án- xát- hááde-hel gO pâagO-dOOgya ∅- hâape-hel.
something :3s:3p-war cry-shout- hsy conj same- song 3s:3s-raise- hsy
ÚúŹde án- dÓÓ- póMúM - áM áM - hêl. Mám- de- xo mÓÓ em- dÓÓ- tÓÓ-
there :3s:3p-sing-sound-come-hsy above-nom-instead like 3s:3a-sing-act-
yẐi. Khodêide mám ∅- khûtde- hel. . . Gya-deM Mi- mOOhól-hel.
impf.hsy suddenly above 3s-pop/flee out-hsy 3p- sleep-ready- hsy
K!ÓÓ- pa gya- hŹn-hel. ÓM ÓM gO gya- thél- hel, ÓM ÓM gO ∅- héŹbe-t!OO
bank-loc 3s:3p-dig- hsy self 3s:3p-open up-hsy self 3s-enter- fut
dé- otte gya- hŹn-hel.
nom-only 3s:3p-dig- hsy
206 APPENDIX A
ˇ á
SÓt ém-děǐsÓÓgyáá gO hÓndé án-xóúá ˇ pó
ˇ ú
ˇ hêl. ÁŹdě gya-thápdÓˇÓˇ-
mêi gO gya-phótkhÓtkÓdǑǑmei. PahŹˇŹˇ thÔOxo án-xóúá ˇ á
ˇ pó
ˇ ú
ˇ nêi.
PhÓˇÓˇhel gO, klk klk klk, hÓˇÓˇgOt é-sálÓmhêl. TéŹ pááxokO á-Ónhêl:
‘Hagya nÓÓ éˇ-thÓttétOO, hagya óŹde en-thÓttétOO?’
Khodêide án-xátÓnhel gO mám gya-áŹhel gO gya-phŹŹthá ˇ á
ˇ holhel.
NO sOphól á-bǒǔ. BéthOO sOphól bôudǑǑmei. MÓn x!ól ě-khúŹbOOdo
ˇ á
mÓÓ án-xóúá ˇ pó
ˇ ú
ˇ nêi.
They had just sat down to sleep when they heard something crawling.
The leaves were dry and had become all crackly. Clearly, the crawling sound
was coming from over there. It stopped and, click click click, loaded a gun.
They all thought the same thing: ‘Is he going to shoot at me or the other
two?’
Suddenly, he gave a shout, flew up and showered embers down on them.
And [in the light] they saw an owl. It had been an owl all along. I guess
he’d been dragging his wings and that’s why it had sounded like someone
crawling.
SÓt ém- deM Mi- sÓÓgyáá gO hÓndé án- xóú- áM áM - póMúM - hêl. ÁŹdeM
just 3a:3a-sleep-sit.pf conj something :3s:3p-crawl-come-sound-hsy leaf
gya-tháp-dÓMÓM-mêi gO gya-phótkhÓtkÓ-dOM OM -mei. PahŹMŹM thÔO-
3p- dry- be- impf.hsy conj 3p- crackly- be- impf.hsy clearly beyond-
xo án- xóú- áM áM - póMúM - nêi. ∅- PhÓMÓM-hel gO, klk klk klk,
instead :3s:3p-crawl-come-sound-impf.hsy 3s-stop- hsy conj [clicks]
hÓMÓMgOt é- sálÓm-hêl. TéŹ pááxokO á- Ón- hêl: ‘Hagya nÓÓ éM-
gun.inv 3s:3i-load- hsy all alike 3a-think-hsy possibly 1 3s:1s-
thÓtté-tOO, hagya óŹde en- thÓtté-tOO?’ Khodêide án- xát- Ón-
shoot- fut possibly that 3s:3d-shoot- fut suddenly :3s:3p-war cry-sound-
hel gO mám gya- áŹ- hel gO gya- phŹŹ-tháM áM -hol-hel. NO sOphól
hsy conj above 3s:3p-go off-hsy conj 3s:3p-fire- beat-kill-hsy conj owl
á∗- boM uM . BéthOO sOphól ∅- bôu- dOM OM -mei. MÓn x!ól
3a:3s-see.pf unbeknownst owl 3s-long time-be- impf.hsy perhaps wing
eM - khúŹ- bOO- do mÓÓ án- xóú- áM áM - póMúM - nêi.
3s:3d-drag-bring-because like :3s:3p-crawl-come-sound-impf.hsy
Bibliography
Adger, David and Harbour, Daniel. 2007. Syntax and Syncretisms of the
Person-Case Constraint. Syntax 10:1ff.
Adger, David, Béjar, Susana, and Harbour, Daniel. 2001. Allomorphy:
Adjacency and Agree, paper presented at the 24th GLOW colloqium,
Braga.
Aronoff, Mark. 1994. Morphology by Itself: Stems and Inflectional
Classes. Cambridge, MA: MIT Press.
Béjar, Susana. 2007. Conditions on Phi Agree. In Daniel Harbour, David
Adger and Susana Béjar, eds., Phi Theory: Phi Features Across In-
terfaces and Modules. Oxford: Oxford University Press.
Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in
Romance. Ph.D. thesis, MIT, Cambridge, MA.
Boyd, Maurice. 1983. Kiowa Voices: Myths, Legends and Folktales. Fort
Worth, TX: Texas Christian University Press.
Carstens, Vicki May. 1991. The morphology and syntax of determiner
phrases in Kiswahili. Ph.D. thesis, University of California, Los An-
geles.
Chierchia, Gennaro. 1998. Plurality of mass nouns and the notion of
‘semantic parameter’. In Susan Rothstein, ed., Events and Grammar ,
53–103, Dordrecht: Kluwer.
Chomsky, Noam and Halle, Morris. 1968/1991. The Sound Pattern of
English. Cambridge: MIT Press, paperback edition.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Roger
Martin, David Michaels, and Juan Uriagereka, eds., Step by Step:
Essays on Minimalist Syntax in Honour of Howard Lasnik , 89–115,
Cambridge, MA: MIT Press.
207
208 BIBLIOGRAPHY
Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz, ed.,

Ken Hale: A Life in Language, 1–52, Cambridge, MA: MIT Press.
Conklin, Harold. 1962. Lexicographic treatment of folk taxonomies. In
F. W. Householder and S. Saporta, eds., Problems in Lexicography,
Publication of the Indiana University Research Center in Anthropol-
ogy, Folklore, and Linguistics 21, 119–141, Bloomington, IN.
Corbett, Greville. 2000. Number . Cambridge: Cambridge University
Press.
Coulson, Michael. 1992. Sanskrit: An Introduction to the Classical Lan-
guage. London: Hodder and Stoughton, second edition, revised by
Richard Gombrich and James Benson.
Cysouw, Michael. 2003. The Paradigmatic Structure of Person Marking.
Oxford: Oxford University Press.
Dryer, Matthew S. 2005. Coding of nominal plurality. In M. Haspelmath,
M. Dryer, D. Gil, and B. Comrie, eds., World Atlas of Language
Structures, 138–139, Oxford: Oxford University Press.
Ellis, Clyde. 1996. To Change Them Forever: Indian Education at the
Rainy Mountain Boarding School, 1893–1920 . Norman and London:
University of Oklahoma Press.
Everett, Daniel. 1986. Pirahã. In D. C. Derbyshire and G. K. Pullum,
eds., Handbook of Amazonian Languages I , 200–325, Berlin: Mouton
de Gruyter.
Foley, William. 1986. The Papuan Languages of New Guinea. Cam-
bridge: Cambridge University Press.
Foley, William. 1991. The Yimas Language of New Guinea. Stanford,
CA: CSLI Publications.
Hale, Kenneth. 1962. Jemez and Kiowa correspondences in reference to
Kiowa-Tanoan. International Journal of American Linguistics 28:1–
8.
Hale, Kenneth. 1967. Toward a reconstruction of Kiowa-Tanoan phonol-
ogy. International Journal of American Linguistics 33:112–120.
Hale, Kenneth. 1973. Person marking in Walbiri. In S. R. Anderson
and P. Kiparsky, eds., A Festschrift for Morris Halle, 308–344, New
York: Holt, Rinehart, and Winston, Inc.
BIBLIOGRAPHY 209
Hale, Kenneth. 1975. Navajo linguistics: Part III, ms. MIT.

Hale, Kenneth. 1997. Some observations on the contribution of local
languages to linguistic science. Lingua 100:71–89.
Halle, Morris and Marantz, Alec. 1993. Distributed Morphology and the
pieces of inflection. In Ken Hale and Samuel Jay Keyser, eds., The
View from Building 20 , 111–176, Cambridge, MA: MIT Press.
Halle, Morris and Vaux, Bert. 1998. Theoretical aspects of Indo-
European morphology: the nominal declensions of Latin and Ar-
menian. In Craig Melchert, Jay Jasanoff, and Lisi Oliver, eds., Mr
Curad: Studies in Honor of Calvert Watkins, Innsbruck: Institut für
Sprachwissenschaft der Universität Innsbruck.
Halle, Morris. 2005. Palatalization/velar softening: What it is and what
it tells us about the nature of language. Linguistic Inquiry 36:23–42.
Harbour, Daniel. 2002. On the metrical nature of Kiowa tone, ms. MIT.
Harbour, Daniel. 2003a. The Kiowa case for feature insertion. Natural
Language and Linguistic Theory 21:543–578.
Harbour, Daniel. 2003b. Some outstanding problems of Yimas. Trans-
actions of the Philological Society 101:125–136.
Harbour, Daniel. 2004. Argument structure and morphological richness.
Lecture notes, Center for Advanced Studies in Theoretical Linguis-
tics, Universitätet i Tromsø.
Harbour, Daniel. 2006a. Number: The morphological use of semantic
atoms. Queen Mary’s OPAL #5. (Google: QMOPALs).
Harbour, Daniel. 2006b. On the unity of ‘number’ in morphology and
semantics. In Ryo Otoguro, Gergana Popova, and Andrew Spencer,
eds., Essex Research Reports in Linguistics 47 (Proceedings of the
York–Essex Morphology Meeting 2), 21–30, University of Essex: De-
partment of Language and Linguistics.
Harbour, Daniel. 2006c. Valence and atomic number. Queen Mary’s,
OPAL #8. (Google: QMOPALs).
Harbour, Daniel and Guoladdle, Carrie. In preparation. Carrie’s Kiowa
correspondence course, ms. QMUL and Riverside Indian School.
Harrington, John P. 1910. On phonetic and lexic resemblances between
Kiowan and Tanoan. American Anthropologist 12:119–123.
210 BIBLIOGRAPHY
Harrington, John P. 1928. Vocabulary of the Kiowa Language. Washing-

ton, DC: Government Printing Office.
Harris, Jim. 1991. The exponence of gender in Spanish. Linguistic In-
quiry 22:27–62.
Heath, Jeffrey. 1998. Pragmatic skewing in 1↔2 pronominal combina-
tions in Native American languages. International Journal of Amer-
ican Linguistics 64:83–104.
Kayne, Richard. 1994. The Antisymmetry of Syntax . Cambridge, MA:
MIT Press.
Kihm, Alain. 2002. What’s in a noun: Noun classes, gender, and noun-
ness. www.llf.cnrs.fr/fr/kihm.
Kihm, Alain. 2005. Noun class, gender, and the Lexicon–Syntax–
Morphology interfaces: A comparative study of Niger-Congo and
Romance languages. In G. Cinque and R. S. Kayne, eds., The Oxford
Handbook of Comparative Syntax , 459–512, Oxford: Oxford Univer-
sity Press.
Krifka, M. 1992. Thematic relations as links between nominal reference
and temporal constitution. In Ivan Sag and Anna Szabolcsi, eds.,
Lexical matters, 29–53, Stanford, CA: CSLI.
Kroskrity, Paul V. 1993. Language, History, and Identity: Ethnolinguis-
tic Studies of the Arizona Tewa. Tucson and London: University of
Arizona Press.
Lagefoged, Peter. 1993. A Course in Phonetics. Fort Worth: Harcourt
Brace College Publishers, third edition.
Lasersohn, Peter. 1995. Plurality, Conjunction and Events. Dordrecht:
Kluwer.
Link, Godehard. 1983. The logical analysis of plurals and mass terms: A
lattice-theoretical approach. In R. Bäuerle, C. Schwarze, and A. von
Stechow, eds., Meaning, Use and Interpretation of Language, 302–
323, Berlin: Walter de Gruyter.
Marantz, Alec. 1997. No escape from syntax: don’t try morphologi-
cal analysis in the privacy of your own lexicon. In A. Dimitriadis,
L. Siegel, C. Surek-Clark, and A. Williams, eds., PWPL 4.2, Pro-
ceedings of the 21st Annual Penn Linguistics Colloquium, 201–225,
BIBLIOGRAPHY 211
Pennsylvania: University of Pennsylvania Working Papers in Lin-

guistics.
Matthews, Peter. 1972. Huave verb morphology: Some comments from
a non-tagmemic viewpoint. International Journal of American Lin-
guistics 38:96–118.
McGinnis, Martha J. 2003. On the universal asymmetry between first
and second person, ms. University of Calgary.
McKay, Graham. 1978. Pronominal person and number categories in
Rembarrnga and Djeebbana. Oceanic Linguistics 17:27–37.
McKay, Graham. 1979. Gender and the category unit augmented.
Oceanic Linguistics 18:203–210.
McKenzie, Parker. n.d.a. Kiowa terms for birds and reptiles, ms. Okla-
homa Historical Society.
McKenzie, Parker. n.d.b. Kiowa terms for mammals, ms. Oklahoma His-
torical Society.
Merrifield, William R. 1959a. Classification of Kiowa nouns. Interna-
tional Journal of American Linguistics 25:269–271.
Merrifield, William R. 1959b. The Kiowa verb prefix. International Jour-
nal of American Linguistics 25:168–176.
Merrill, William L., Hanson, Marian Kaulaity, Green, Candace S., and
Reuss, Frederick J. 1997. A Guide to the Kiowa Collections at the
Smithsonian Institution. Washington, D.C.: Smithsonian Institution
Press.
Mooney, James. 1898/1979. Calendar History of the Kiowa Indians.
Washington, D.C.: Smithsonian Institution Press.
Noyer, Rolf. 1992. Features, Positions and Affixes in Autonomous Mor-
phological Structure. Cambridge, MA: MITWPL.
Noyer, Rolf. 1998. Impoverishment theory and morphosyntactic marked-
ness. In D. Brentari, S. Lapointe and P. Farrell, eds., Morphology and
Its Relation to Phonology and Syntax , 264–285, Palo Alto, CA: CSLI.
Ojeda, Almerindo E. 1998. The semantics of collectives and distributives
in Papago. Natural Language Semantics 6:245–270.
212 BIBLIOGRAPHY
Ouhalla, Jamal. 2005. Agreement features, agreement and antiagree-

ment. Natural Language and Linguistic Inquiry 23:655–686.
Pesetsky, David and Torrego, Esther. 2001. T to C movement: causes
and consequences. In Michael Kenstowicz, ed., Ken Hale: a Life in
Language, 355–426, Cambridge, MA: MIT Press.
Ritter, Elizabeth. 1991. Two functional categories in noun phrases: ev-
idence from Modern Hebrew. In Susan Rothstein, ed., Syntax and
Semantics 26 , 37–62, New York: Academic Press.
Ritter, Elizabeth. 1993. Where’s gender? Linguistic Inquiry 24:795–803.
Schwarzschild, Roger. 1996. Pluralities. Dordrecht: Kluwer.
Sportiche, Dominique. 1997. Reconstruction and constituent structure,
paper presented at MIT; available at http://www.linguistics.ucla.
edu/people/sportich/papers/mittalk97.pdf.
Sprott, Robert W. 1992. Jemez syntax. Ph.D. thesis, University of
Chicago.
Takahashi, Junichi. 1984. Case marking in Kiowa: A study in the orga-
nization of meaning. Ph.D. thesis, City University of New York.
Trager, Edith Crowell. 1960. The Kiowa language: A grammatical study.
Ph.D. thesis, University of Pennsylvania.
Trommer, Jochen. 1999. Morphology consuming syntax’ resources: Gen-
eration and parsing in a minimalist version of distributed morphol-
ogy. In Proceedings of the ESSLI Workshop on Resource Logic and
Minimalist Grammars, Utrecht.
Vergnaud, Jean-Roger. 1974. French relative clauses. Ph.D. thesis, MIT.
Watkins, Laurel (with the assistance of Parker McKenzie). 1984. A
Grammar of Kiowa. Lincoln: University of Nebraska Press.
Watkins, Laurel. 1990. Noun phrase versus zero in Kiowa discourse.
International Journal of American Linguistics 56:410–426.
Watkins, Laurel. 1993. The discourse functions of Kiowa switch-
reference. International Journal of American Linguistics 59:137–164.
Winter, Yoad. 2001. Flexibility Principles in Boolean Semantics: Coor-
dination, Plurality, and Scope in Natural Language. Cambridge MA:
MIT Press.
BIBLIOGRAPHY 213
Wonderly, William, Gibson, Lorna F. and Kirk, Paul L. 1954. Number in

Kiowa: Nouns, demonstratives and adjectives. International Journal
of American Linguistics 20:1–7.
Yang, Charles. 1999. Words, rules, and competitions, ms. MIT.
Yang, Charles. 2002. Knowledge and Learning in Natural Language.
Oxford: Oxford University Press.
Yumitani, Yukihiro. 1998. A phonology and morphology of Jemez Towa.
Ph.D. thesis, University of Kansas.
Index
[±augmented], 5, 61, 67, 101, 110 ClassP, 73, 93, 99, 105, 106, 109
definition, 63, 69 interpretation, 96, 99
[±group], 5, 61, 100–102, 110, 115 collections of collections, 44, 99, 139
and suppletion, 138–144 collectivity, 36, 41–44, 50, 95, 97,
definition, 63, 98–99 101, 138–140
[±singular], 5, 61, 67, 70, 101, 110 of action, 42, 45
definition, 63, 68–70 conjunction, 28–29, 107–109, 201
abstract nouns, 48 d∼i alternation, 108, 146, 188, 190

adjacency, 122, 125, 126, 186 default agreement, 77, 179
adjectives, 16, 89, 129, 132–136, 141 deletion, 164–167, 175, 186
adverbs, 17, 150–154 demonstratives, 16, 89–92, 148, 202
agreement prefixes, 9, 18–19, 157– Distributed Morphology, 10–12, 186
191, 202 DP structure, 62, 100, 109, 126,
notation, 19, 162 144
table of, 161 dual, see referential cardinality
agreement types, 23, 71, 72
Aktionart empathy, 51–53, 145–149
semantics of, 63, 68, 99 English, 6, 12, 31, 36, 43, 46, 69,
Aktionsart 93, 95, 121, 144, 155
semantics of, 69
allomorphy, definition, 122 feature insertion, 106, 164, 167, 179,
animacy, 34–38, 50, 94, 145–149, 181, 186
162, 167, 190 features
Arapesh, 33, 63 conflicting, 62, 76, 77
lattice semantics, 67–70
Bantu, 193 negation, 68
body parts, 37, 41–45, 95, 97 semantic composition, 70
first person, 31–32, 34, 50, 81–88,
case marking, 15 94, 136–137, 165, 168–169,
Catalan, 159 182
214
INDEX 215
footwear, 49, 94, 113 Minimalism, 10–12, 73–74

mnemonics, 24, 78–88, 109
gender, 4, 56, 193–200 naturalness, 8, 94–96, 102
Georgian, 169 paucity, 8, 110–113
Hopi, 119 morphosemantic, 9, 99, 160, 200
motion, independent, 34, 36, 94
idi, 27, 42–44, 80–136
idp, 26, 38–42, 79–80, 95, 130, 132 noun class
ids, 26, 38–42, 100, 103–104, 108, mnemonics, 24
130 noun classification
Ilocano, 69 arbitrariness, 33
implements, 36, 39, 42, 45, 50, 95 default, 33, 49, 50, 94
impoverishment, 87, 88, 164, 168– features, 72
170, 179, 181 non-uniqueness, 33
incorporation, 17–19, 154 semantic coherence, 4, 32–50
individual-level predication, 118, 150 subclasses, 34, 44, 113
inverse number features
allophony, 54–56 distribution, 7, 62, 73, 88–93,
and suppletion, 131–137 144
feature specification, 76 syntactic flow, 4
marking, 6, 15 only, 92–93
syntactic locus, 62, 73, 89, 93
Italian, 43 Person-Case Constraint, 158, 171
person features, 84
Jemez, 112 phonology, 19–20, 54–58, 123–125,
Kiowa-Tanoan, 13, 111, 193 163–164, 172, 178–181, 183,
Kiswahili, 193 187, 201–202
plural, see referential cardinality
Latin, 179 pluralia tantum, 46, 50, 98, 99, 101,
liquids, 45–46 104, 142–144
bodies of, 45 possession, 92
ppp, 27, 46–48, 100, 104, 106, 130
markedness, 68 PRO, 154
mass nouns, 28–31, 43 pronouns, 15, 17, 32, 68, 69, 88, 92
and suppletion, 140–142
granular, 47–48, 50, 100, 106 referential cardinality
non-granular, 45–46, 50, 106 and mass nouns, 31
micromorphology, 160, 190 definition, 22
216 INDEX
feature composition, 67, 70, 71, uninterpretability, 62, 73, 76–78,

96, 113–114 110, 126
notation, 23 valuation, 74
reflexives, 53, 145–149, 162, 167,
190 vegetation, 38–42, 50, 95, 97
relative clauses, 16, 54, 91, 148, 154 word order, 14–16
Rio Grande Tewa, 155
rule ordering, 163, 169, 183 Y-model, 10
Russian, 33 Yimas, 53, 169
Sanskrit, 193–195
Saussurean sign, 12, 193
sdi, 25, 34–38, 50–53, 74–78, 94,
107, 130, 132, 136–137
sdp, 25, 78–77, 81, 94, 127–130
sds, 27, 100, 102–104, 130, 138
second person, 86, 168, 172, 182,
183, 186
shape inductivity, 45, 50
sii, 31, 34, 81–88, 94, 168
singular, see referential cardinality
sisterhood, 122, 144
sss, 28–31, 45–46, 106, 130
stage-level predication, 118, 150
suppletion, 8–9, 117–155
definition, 121
mismatches, 130
Swahili, see Kiswahili
switch reference, 201
Tewa, 125
thematic nouns, 56–59, 120
toggle, 193
tone(s), 3, 4, 7, 19–20, 54–55,
124–125, 132, 157, 163,
172–173, 175–177, 179–182,
185, 187–190, 201
trial, 77
STUDIES IN NATURAL LANGUAGE AND LINGUISTIC THEORY
Managing Editors
Marcel den Dikken, City University of New York

Liliane Haegeman, University of Lille
Joan Maling, Brandeis University
Publications
1. L. Burzio: Italian Syntax. A Government-binding Approach. 1986

ISBN Hb 90-277-2014-2; Pb 90-277-2015-0
2. W.D. Davies: Choctaw Verb Agreement and Universal Grammar. 1986
ISBN Hb 90-277-2065-7; Pb 90-277-2142-4
3. K. É. Kiss: Configurationality in Hungarian. 1987
ISBN Hb 90-277-1907-1; Pb 90-277-2456-3
4. D. Pulleyblank: Tone in Lexical Phonology. 1986
ISBN Hb 90-277-2123-8; Pb 90-277-2124-6
5. L. Hellan and K. K. Christensen: Topics in Scandinavian Syntax. 1986
ISBN Hb 90-277-2166-1; Pb 90-277-2167-X
6. K. P. Mohanan: The Theory of Lexical Phonology. 1986
ISBN Hb 90-277-2226-9; Pb 90-277-2227-7
7. J. L. Aissen: Tzotzil Clause Structure. 1987
ISBN Hb 90-277-2365-6; Pb 90-277-2441-5
8. T. Gunji: Japanese Phrase Structure Grammar. A Unification-based Approach 1987.
ISBN1-55608-020-4
9. W. U. Wurzel: Inflectional Morphology and Naturalness. 1989
ISBN Hb 1-55608-025-5; Pb 1-55608-026-3
10. C. Neidle: The Role of Case in Russian Syntax. 1988 ISBN1-55608-042-5
11. C. Lefebvre and P. Muysken: Mixed Categories. Nominalizations in Quechua 1988
ISBN Hb 1-55608-050-6; Pb 1-55608-051-4
12. K. Michelson: A Comparative Study of Lake-Iroquoian Accent. 1988
ISBN1-55608-054-9
13. K. Zagona: Verb Phrase Syntax. A Parametric Study of English and Spanish 1988
ISBN Hb 1-55608-064-6; Pb 1-55608-065-4
14. R. Hendrick: Anaphora in Celtic and Universal Grammar. 1988 ISBN1-55608-066-2
15. O. Jaeggli and K.J. Safir (eds.): The Null Subject Parameter. 1989
ISBN Hb 1-55608-086-7; Pb 1-55608-087-5
16. H. Lasnik: Essays on Anaphora. 1989 ISBN Hb 1-55608-090-5; Pb 1-55608-091-3

17. S. Steele: Agreement and Anti-Agreement. A Syntax of Luiseño. 1990
ISBN 0-7923-0260-5
18. E. Pearce: Parameters in Old French Syntax. Infinitival Complements. 1990
ISBN Hb 0-7923-0432-2; Pb 0-7923-0433-0
19. Y. A. Li: Order and Constituency in Mandarin Chinese. 1990 ISBN 0-7923-0500-0
20. H. Lasnik: Essays on Restrictiveness and Learnability. 1990
ISBN0-7923-0628-7; Pb 0-7923-0629-5
21. M. J. Speas: Phrase Structure in Natural Language. 1990
ISBN0-7923-0755-0; Pb 0-7923-0866-2
22. H. Haider and K. Netter (eds.): Representation and Derivation in the Theory of Grammar.
1991 ISBN 0-7923-1150-7
23. J. Simpson: Warlpiri Morpho-Syntax. A Lexicalist Approach. 1991 ISBN 0-7923-1292-9
24. C. Georgopoulos: Syntactic Variables. Resumptive Pronouns and A’ Binding in Palauan.
1991 ISBN 0-7923-1293-7
25. K. Leffel and D. Bouchard (eds.): Views on Phrase Structure. 1991 ISBN 0-7923-1295-3
26. C. Tellier: Licensing Theory and French Parasitic Gaps. 1991
ISBN0-7923-1311-9; Pb 0-7923-1323-2
27. S.-Y. Kuroda: Japanese Syntax and Semantics. Collected Papers. 1992
ISBN0-7923-1390-9; Pb 0-7923-1391-7
28. I. Roberts: Verbs and Diachronic Syntax. A Comparative History of English and French.
1992 ISBN 0-7923-1705-X
29. A. Fassi Fehri: Issues in the Structure of Arabic Clauses and Words. 1993
ISBN 0-7923-2082-4
30. M. Bittner: Case, Scope, and Binding. 1994 ISBN 0-7923-2649-0
31. H. Haider, S. Olsen and S. Vikner (eds.): Studies in Comparative Germanic Syntax. 1995
ISBN 0-7923-3280-6
32. N. Duffield: Particles and Projections in Irish Syntax. 1995
ISBN0-7923-3550-3; Pb 0-7923-3674-7
33. J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. 1996
ISBN 0-7923-3745-X
34. J. Bayer: Directionality and Logical Form. On the Scope of Focusing Particles and Wh-
in-situ. 1996 ISBN 0-7923-3752-2
35. R. Freidin (ed.): Current Issues in Comparative Grammar. 1996
ISBN 0-7923-3778-6; Pb 0-7923-3779-4
36. C.-T. J. Huang and Y.-H. A. Li (eds.): New Horizons in Chinese Linguistics 1996
ISBN 0-7923-3867-7; Pb 0-7923-3868-5
37. A. Watanabe: Case Absorption and WH-Agreement. 1996 ISBN 0-7923-4203-8
38. H. Thráinsson, S.D. Epstein and S. Peter (eds.): Studies in Comparative Germanic Syntax.
Volume II. 1996 ISBN 0-7923-4215-1
39. C. J. W. Zwart: Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of
Dutch. 1997 ISBN0-7923-4263-1; Pb 0-7923-4264-X
40. T. Siloni: Noun Phrases and Nominalizations. The Syntax of DPs. 1997
ISBN 0-7923-4608-4
41. B. S. Vance: Syntactic Change in Medieval French. 1997 ISBN0-7923-4669-6
42. G. Müller: Incomplete Category Fronting. A Derivational Approach to Remnant Move-
ment in German. 1998 ISBN 0-7923-4837-0
Studies in Natural Language and Linguistic Theory
43. A. Alexiadou, G. Horrocks and M. Stavrou (eds.): Studies in Greek Syntax 1998
ISBN 0-7923-5290-4
44. R. Sybesma: The Mandarin VP. 1999 ISBN 0-7923-5462-1
45. K. Johnson and I. Roberts (eds.): Beyond Principles and Parameters. Essays in Memory
of Osvaldo Jaeggli. 1999 ISBN 0-7923-5501-6
46. R. M. Bhatt: Verb Movement and the Syntax of Kashmiri. 1999 ISBN 0-7923-6033-8
47. A. Neeleman and F. Weerman: Flexible Syntax. A Theory of Case and Arguments 1999
ISBN 0-7923-6058-3
48. C. Gerfen: Phonology and Phonetics in Coatzospan Mixtec. 1999 ISBN 0-7923-6034-6
49. I. Paul, V. Phillips and L. Travis (eds.): Formal Issues in Austronesian Linguistics. 2000
ISBN 0-7923-6068-0
50. M. Frascarelli: The Syntax-Phonology Interface in Focus and Topic Constructions in Ital-
ian. 2000 ISBN 0-7923-6240-3
51. I. Landau: Elements of Control. Structure and Meaning in Infinitival Constructions. 2000
ISBN 0-7923-6620-4
52. W. D. Davies and S. Dubinsky (eds.): Objects and other Subjects. Grammatical Functions,
Functional Categories and Configurationality. 2001
ISBN 1-4020-0064-2; Pb 1-4020-0065-0
53. J. Ouhalla and U. Shlonsky (eds.): Themes in Arabic and Hebrew Syntax. 2002
ISBN 1-4020-0536-9; Pb 1-4020-0537-7
54. E. Haeberli: Features, Categories and the Syntax of A-Positions. Cross-Linguistic Varia-
tion in the Germanic Languages. 2002 ISBN 1-4020-0854-6; Pb 1-4020-0855-4
55. J. McDonough: The Navajo Sound System. 2003 ISBN 1-4020-1351-5; Pb 1-4020-1352-3
56. D.E. Holt (ed.): Optimality Theory and Language Change. 2003
ISBN 1-4020-1469-4; Pb 1-4020-1470-8
57. J. Camacho: The Structure of Coordination. Conjunction and Agreement Phenomena in
Spanish and Other Languages. 2003 ISBN 1-4020-1510-0; Pb 1-4020-1511-9
58. I. Toivonen: Non-Projecting Words. A Case Study of Swedish Particles. 2003
ISBN 1-4020-1531-3; Pb 1-4020-1532-1
59. D. Adger, C. de Cat and G. Tsoulas (eds). Peripheries. Syntactic Edges and their Effects.
2004 ISBN 1-4020-1908-4; Pb 1-4020-1909-2
60. C. Goria: Subject Clitics in the Northern Italian Dialects. A Comparative Study Based on
the Minimalist Program and Optimality Theory. 2004
ISBN 1-4020-2736-2; Pb 1-4020-2737-0
61. V. Dayal and A. Mahajan (eds.): Clause Structure in South Asian Languages 2004
ISBN 1-4020-2717-6; Pb 1-4020-2718-4
62. P. Kempchinsky and R. Slabakova (eds.): Aspectual Inquiries. 2005
ISBN 1-4020-3032-0; Pb 1-4020-3035-5
63. M. Arad: Roots and Patterns. Hebrew Morpho-syntax. 2005
ISBN 1-4020-3243-9; Pb 1-4020-3245-5
64. P. Štekauer and R. Lieber (eds.): Handbook of Word-Formation. 2005
ISBN 1-4020-3595-0; Pb 1-4020-3597-7
65. A. Johns, D. Massam and J. Ndayiragije (eds.): Ergativity. Emerging Issues 2006
ISBN Hb 1-4020-4186-1; Pb 1-4020-4187-X
66. F. Lee: Remnant Raising and VSO Clausal Architecture. A Case Study from San Lucas
Quiavini Zapotec. 2006 ISBN1-4020-4300-7
67. O. Mišeska Tomić: The Balkan Sprachbund Morpho-syntactic Features. 2006
ISBN1-4020-4487-9
68. K. É. Kiss, Event Structure and the Left Periphery: Studies on Hungarian. 2006
ISBN 1-4020-4753-3
69. D. Harbour, Morphosemantic Number: From Kiowa Noun Classes to UG Number
Features. 2007/2008 ISBN Hb 978-1-4020-5037-4; Pb 978-1-4020-5039-8
springer.com

Daniel Harbour-Morphosemantic Number From Kiowa Noun Classes To UG Number Features-Springer (2008) PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Daniel Harbour-Morphosemantic Number From Kiowa Noun Classes To UG Number Features-Springer (2008) PDF

Transféré par

Droits d'auteur :

Formats disponibles

MORPHOSEMANTIC NUMBER:

FROM KIOWA NOUN CLASSES

ISBN 978-1-4020-5039-8 (PB)

Printed on acid-free paper

All Rights Reserved

2.6 Phonological incoherence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.4 Group-induced mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

* of agreement preﬁxes, causing subsequent verb to have low tone

Labial Dental Alveolar Palatal Velar Laryngeal

Short Long Diphthong

High tone Falling tone Low tone

1.1. Aim: A morphosemantic theory of number

Such results are of little interest to semanticists, whose theories of num-

and syncretisms must be answerable to the semantics: whatever primitives

This system is immediately fascinating. At the most general level, there

(5) Syntactic transmission of number information

(The account of inverse marking in Chapter 3 shows that the information

1.2. Main claim and overview

The core number-theoretic claim of this investigation is that Universal Gram-

their manipulation, syntactically and postsyntactically, leads to a tightly

The chapter also introduces one of Kiowa’s most fascinating phenomena,

The number-feature content of D is not itself semantically interpreted. How-

1.3. Conceptual clariﬁcation: ‘Morphosemantic’

between morphology and syntax, ‘syntacticosemantic’ to that between syn-

Syntax builds hierarchical objects and transmits information (via agreement

The syntactic and morphological frameworks I adopt are Minimalism and

inverse marking depend on a syntactic computation. This postsyntactic de-

1.5. Overview of Kiowa structure

1.5.2. The people

Kiowa is a Tanoan-related language of Oklahoma.4 It is spoken ﬂuently now

Kiowa community appears to have contributed to the decline in the tribe’s

1.5.3. The language

Basic word order

(14) Particles Subject Indirect–Object Direct–Object Verb

(16) HÓn máthOn ∅- xáM áM nÔO

(18) HÓn ∅- xáM áM nÔO máthOn

These dislocations correlate with information structure, leftward dislocation

(19) BéthOO hÓn ám em-dÓMÓM-mÔO- hel

As illustrated by the underlining, particles in selectional relations with in-

Nominal morphology is sparse in Kiowa. There is no case marking either

(22) Dôiette an pénhaa gya- ôugu xóŹkya

Three more examples are:

(24) PáágO gya- bóMúM ˇ á

(27) x!óú-êl xégun-syan tou- gúl k!ŹŹ- t!áMŹM- de

(28) k!ÓÓ- pa Indian Fair-kya álOO- aadO- yOO

(29) nÓÓ-(x!)al Indian Fair-deki háote- kÓÓ

The complex verb

(30) Incorporates Root Distributive Aspect/Negation Future Evidential

Incorporates can be adverbs:

(31) Gyat- hón-pOl- /t!Om-pÓl- tOO

—or verbs, with or without complement nouns:

(33) TÓÓdé de- t!ÓÓthâl-dou

—or nouns alone:

Object incorporation is not valence reducing: (35) is overtly transitive.

The agreement preﬁx

Kiowa phonology is not overly complicated, though it does throw up some

IPA [l] (particularly when intervocalic and morpheme internal, as in álOO

KIOWA’S NOUN CLASSES

2.1. Overview: Meaning and features

(1) Kiowa nouns are classiﬁed according to number properties

classiﬁcation, and Takahashi (1984) provided insight in the semantic coher-

The basis of noun classiﬁcation in Kiowa, and of the mnemonic nomenclature

(4) Deﬁnition: Referential Cardinality