The Syntax-Morphology Interface

This page intentionally left blank
THE SYNTAX–MORPHOLOGY INTERFACE
Syncretism – where a single form serves two or more morphosyntactic

functions – is a persistent problem at the syntax–morphology interface. It
results from a ‘mismatch’ whereby the syntax of a language makes a
particular distinction, but the morphology does not. This pioneering
book provides the first full-length study of inflectional syncretism, pre-
senting a typology of its occurrence across a wide range of languages. The
implications of syncretism for the syntax–morphology interface have
long been recognized: it argues either for an enriched model of feature
structure (thereby preserving a direct link between function and form), or
for the independence of morphological structure from syntactic structure.
This book presents a compelling argument for the autonomy of morph-
ology, and the resulting analysis is illustrated in a series of formal case
studies within Network Morphology. It will be welcomed by all linguists
interested in the relation between words and the larger units of which they
are a part.
M A T T H E W B A E R M A N is Research Fellow at the University of Surrey. He

is author of The evolution of fixed stress in Slavic (1999) and has written on
the subject of syncretism, morphology and prosody for a wide range of
journals including Language, Lingua, Russian Linguistics, and Natural
Language and Linguistic Theory.
D U N S T A N B R O W N is Senior Lecturer in Linguistics at the University of

Surrey and currently Treasurer of the Linguistics Association of Great
Britain, with research interests in computational linguistics, linguistic
typology and morphology. He has written widely on these subjects for
journals such as Language, Lingua, Journal of Slavic Linguistics and
Yearbook of Morphology.
G R E V I L L E G . C O R B E T T is Distinguished Professor of Linguistics at the

University of Surrey, and formerly President of the Linguistics Association
of Great Britain. Among other books he is author of Gender (Cambridge
University Press, 1991), and Number (2000), and co-editor of Heads in
Grammatical Theory (1993).
In this series
66 ANTHONY R. WARNER: English auxiliaries: structure and history
67 P. H. MATTHEWS: Grammatical theory in the United States from
Bloomfield to Chomsky
68 LJILJANA PROGOVAC: Negative and positive polarity: a binding approach
69 R. M. W. DIXON: Ergativity
70 Y A N H U A N G : The syntax and pragmatics of anaphora
71 K N U D L A M B R E C H T : Information structure and sentence form: topic,
focus, and the mental representation of discourse referents
72 L U I G I B U R Z I O : Principles of English stress
73 J O H N A . H A W K I N S : A performance theory of order and constituency
74 A L I C E C . H A R R I S A N D L Y L E C A M P B E L L : Historical syntax in
cross-linguistic perspective
75 L I L I A N E H A E G E M A N : The syntax of negation
76 P A U L G O R R E L : Syntax and parsing
77 G U G L I E L M O C I N Q U E : Italian syntax and universal grammar
78 H E N R Y S M I T H : Restrictiveness in case theory
79 D . R O B E R T L A A D : Intonational morphology
80 A N D R E A M O R O : The raising of predicates: predicative noun phrases and
the theory of clause structure
81 R O G E R L A S S : Historical linguistics and language change
82 J O H N M . A N D E R S O N : A notional theory of syntactic categories
83 B E R N D H E I N E : Possession: cognitive sources, forces and grammaticalization
84 N O M T E R T E S C H I K - S H I R : The dynamics of focus structure
85 J O H N C O L E M A N : Phonological representations: their names, forms and
powers
86 C H R I S T I N A Y . B E T H I N : Slavic prosody: language change and phonologi-
cal theory
87 B A R B A R A D A N C Y G I E R : Conditionals and predication: time, knowledge
and causation in conditional constructions
88 C L A I R E L E F E B V R E : Creole genesis and the acquisition of grammar:
the case of Haitian Creole
89 H E I N Z G I E G E R I C H : Lexical strata in English: morphological causes,
phonological effects
90 K E R E N R I C E : Morpheme order and semantic scope: word formation and
the Athapaskan verb
91 A P R I L M C M A H O N : Lexical phonology and the history of English
92 M A T T H E W Y . C H E N : Tone Sandhi: patterns across Chinese dialects
93 G R E G O R Y T . S T U M P : Inflectional morphology: a theory of paradigm
structure
94 J O A N B Y B E E : Phonology and language use
95 L A U R I E B A U E R : Morphological productivity
96 T H O M A S E R N S T : The syntax of adjuncts
97 ELIZABETH CLOSS TRAUGOTT and RICHARD B. DASHER:
Regularity in semantic change
98 M A Y A H I C K M A N N : Children’s discourse: Person, space and time across
languages
99 D I A N E B L A K E M O R E : Relevance and linguistic meaning: The semantics
and pragmatics of discourse markers
100 I A N R O B E R T S A N D A N N A R O U S S O U : Syntactic change: a minimalist
approach to grammaticalization
101 D O N K A M I N K O V A : Alliteration and sound change in early English
102 M A R K C . B A K E R : Lexical categories: verbs, nouns and adjectives
103 C A R L O T A S . S M I T H : Modes of discourse: the local structure of texts
104 R O C H E L L E L I E B E R : Morphology and lexical semantics
105 H O L G E R D I E S S E L : The acquisition of complex sentences
106 S H A R O N I N K E L A S a n d C H E R Y L Z O L L : Reduplication: doubling in
morphology
107 S U S A N E D W A R D S : Fluent aphasia
108 B A R B A R A D A N C Y G I E R a n d E V E S W E E T S E R : Mental spaces in
grammar: conditional constructions
109 MATTHEW BAERMAN, DUNSTAN BROWN AND GREVILLE
G . C O R B E T T : The syntax–morphology interface: a study of syncretism.
Earlier issues not listed are also available

CAMBRIDGE STUDIES IN LINGUISTICS
General editors: P. AUSTIN, J. BRESNAN, B. COMRIE, S. CRAIN,
W. DRESSLER, C. J. EWEN, R. LASS, D. LIGHTFOOT, K. RICE, I. ROBERTS,
S. ROMAINE, N. V. SMITH
The Syntax–Morphology Interface
A Study of Syncretism
T H E SY N T A X –
MORPHOLOGY
INTERFACE
A STUDY OF SYNCRETISM
MATTHEW BAERMAN
University of Surrey
DUNSTAN BROWN
GREVILLE G. CORBETT
  
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge  , UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521821810
© Matthew Baerman, Dunstan Brown, and Greville G. Corbett 2005
This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2005
- ---- eBook (EBL)

- --- eBook (EBL)
- ---- hardback

- --- hardback
Cambridge University Press has no responsibility for the persistence or accuracy of s
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
To our families
Contents
Preface page xv
List of abbreviations and symbols xvii
1 Introduction 1
1.1 History of the notion 3
1.2 Delimiting the notion of syncretism 4
1.3 Scope of the investigation 7
1.3.1 Typological methodology 7
1.3.2 Selection of forms 8
1.4 Accidental versus systematic homophony 9
1.5 Using this book 10
1.5.1 Supporting materials 10
1.5.2 Glossing conventions 11
1.5.3 Structure of the book 12
2 Characteristics of syncretism 13
2.1 Syncretic paradigms 13
2.1.1 Types 13
2.1.2 Implications 17
2.2 Domains of comparison 17
2.2.1 Morphological classes 17
2.2.2 Feature values 19
2.3 Morphological characteristics 23
2.4 A typology of interpretations 27
2.4.1 Syncretism as neutralization 28
2.4.2 Syncretism as uninflectedness 30
2.4.3 Canonical syncretism 33
2.5 Conclusion 35
3 Cross-linguistic typology of features 37

3.1 Case 38
3.1.1 Introduction 38
3.1.2 Types of case syncretism 40
xi
xii Contents
3.1.3 Conclusion 56
3.2 Person 57
3.2.2 Patterns of syncretism 59
3.2.3 Directional effects 63
3.2.4 Diachrony 70
3.2.5 Summary 75
3.3 Person syncretism in two-place verbs 75
3.3.2 Syncretism of subject person 76
3.3.3 Syncretism of object person 79
3.3.4 Other patterns 80
3.3.5 Summary 81
3.4 Gender 81
3.4.2 Gender and number 82
3.4.2.1 Smaller systems 83
3.4.2.2 Larger systems 86
3.4.3 Syncretism restricted by target 90
3.4.4 Summary 91
3.5 Number 92
3.5.2 Values 93
3.5.3 Directionality 94
3.5.4 Summary 95
3.6 Tense-aspect-mood 95
3.6.2 Affix suppression 96
3.6.3 Syncretic affixes 100
3.6.4 Compound systems 101
3.6.5 Summary 103
3.7 Polarity effects 103
3.7.2 Morphological systematicity 105
3.7.3 Semantic systematicity 108
3.7.4 Summary 111
3.8 The interaction of features 111
3.8.2 Syncretisms and their contexts 112
3.8.3 Nominal feature interactions 113
3.8.3.1 Typologies of interaction 113
3.8.3.2 Exploring interaction 114
3.8.3.3 Constraints on nominal features 118
3.8.4 Verbal feature interactions 119
Contents xiii
3.8.4.1 Verbal paradigms 120

3.8.4.2 Exploring the verb data 121
3.8.5 Interpreting the generalisations 123
3.8.5.1 Number, case and gender on nominals 123
3.8.5.2 TAM and agreement on verbs 123
3.8.6 Summary 124
3.9 Conclusion 124
4 Formal representation 126

4.1 Introduction 126
4.2 Defining sets of values 126
4.2.1 Natural classes 126
4.2.2 Unnatural classes 131
4.2.3 A note on polarity effects 132
4.3 Symmetrical versus directional rules 133
4.3.1 Convergent bidirectional syncretism 136
4.3.2 Divergent bidirectional syncretism 139
4.3.3 Symmetrical versus directional rules: a summary 144
4.3.4 Ranked constraints as an alternative to directional
syncretism 145
4.4 Possible constraints on syncretism 150
4.4.1 Variants on hierarchical structures 150
4.4.2 Carstairs (1987), Carstairs-McCarthy (1998a, b) 151
4.4.3 Impoverishment 160
4.4.4 Stump (2001) and Zwicky (2000) 163
4.5 Summary 166
4.5.1 Predictions and counter-examples 166
4.5.2 Towards a model of syncretism 169
5 Formal framework and case studies 171

5.1 Network Morphology and syncretism 172
5.1.1 Inferential-realizational theories and morphology 175
5.1.2 Default inheritance 177
5.1.3 Underspecification and semantic naturalness 180
5.1.4 Systematicity in Network Morphology 182
5.2 Case study 1: Dhaasanac 183
5.3 Case study 2: The Dalabon verbal system 186
5.3.1 The structure of the Dalabon intransitive paradigm 187
5.3.2 The Dalabon transitive paradigm and the inadequacy of
underspecification 188
5.3.3 The Dalabon transitive paradigm: a generalized referral
analysis 194
5.3.4 The verbal hierarchy 199
5.3.5 The shape of the verbal paradigm 200
xiv Contents
5.3.6 Referral of 1 > 2sG to 3 > 2sG 203

5.3.7 Referral of 2 > 1 to 3 >1 203
5.3.8 Dalabon: summing up 203
5.4 Case study 3: The Russian nominal system 204
5.4.1 Domains of syncretism 206
5.4.1.1 Phonologically determined ‘syncretism’ 206
5.4.1.2 Lexically determined syncretism 206
5.4.1.3 Morphologically determined syncretism 207
5.4.1.4 Candidates for syntactically determined syncretism 213
5.4.2 Orthogonal specification of syncretism 216
5.5 Conclusion 217
6 Conclusion 219
6.1 Taking stock 219
6.2 Results 220
6.3 Consequences 221
Appendix 1: Case syncretism in the World Atlas of Language

Structures sample 223
Appendix 2: Person syncretism in the World Atlas of
Language Structures sample 228
Appendix 3: Syncretism in two-place verbs in the World
Atlas of Language Structures corpus 233
Appendix 4: DATR fragment for Dhasaanac case study 236
Appendix 5: DATR fragment for Dalabon case study 242
Appendix 6: DATR fragment for Russian case study 248
References 254
Author index 271
Language index 275
Subject index 279
Preface
This book has an interesting history of collaboration. It began life in

research done by Greville Corbett and Norman Fraser on the morphology
of Russian, starting in 1990, research which was inspired by the work of
Roger Evans and Gerald Gazdar on DATR. The ESRC and Leverhulme
Trust provided funding, which brought Dunstan Brown and Andrew
Hippisley to Surrey, and the work developed into a more general theor-
etical framework, Network Morphology. We found syncretism of increas-
ing importance in the development of the framework and gave
presentations at the following places: Krems (Austria), University of
Sussex, Linguistics Association of Great Britain (at the University of
Surrey), University of California (Berkeley), Gregynog (Wales), Heinrich-
Heine-Universität (Düsseldorf), University of Edinburgh, University of
Cologne, University of Helsinki, La Trobe University, Norsk Forening for
Språkvitenskap (Oslo), Institutt for Østeuropeiske og Orientalske Studier
(University of Oslo), Moscow University, University of Oxford, Cornell
University, Twelfth International Conference on Historical Linguistics
(University of Manchester), Conference on Lexical Structures (Wuppertal),
British Association for Slavonic and East European Studies (Cambridge),
University of Sheffield, University of Essex, University of Pennsylvania,
Leipzig University, Association for Linguistic Typology (University of
Amsterdam), Second Mediterranean Meeting on Morphology (University
of Malta), Max Planck Institute for Psycholinguistics (Nijmegen), Second
Winter Typological School (Istra, Moscow district), Ninth International
Morphology Meeting (Vienna), University of California (Santa Barbara),
University College London, Second Northwest Conference on Slavic
Linguistics (Berkeley), Stockholm University, 37th Meeting of the Chicago
Linguistic Society, Fakultetets Forsknings Fredage (University of
Copenhagen), University of Melbourne, Scandinavian Slavists’ Summer
School (Kungälv, Sweden), University of Leeds, School of Oriental and
African Studies (London), University of Manchester Institute of Science
xv
xvi Preface
and Technology, Lancaster University, University of Catania, Second

International Seminar ‘Computer Treatment of Slavonic Languages’
(Bratislava), and University of York. We are very grateful for all the com-
ments we have received on these occasions. We applied for funding to work
specifically on syncretism within this framework, which allowed Matthew
Baerman to join the Surrey Morphology Group. He undertook the careful
typological work which led to the Surrey Syncretisms Database, and which is
a basis for the book. Collaboration with Nicholas Evans on Dalabon added
an important impetus to the work. As it became clear that the research on
syncretism had more substance than could fit into a journal article, Baerman’s
role became increasingly important. He is the book’s first author. Brown and
Corbett are together the second author, with Brown’s role being particularly
significant in the development of the formal side of Network Morphology,
while Corbett’s role was most important at the start of the project. Portions of
this book have been adapted from previously published material, specifically
Evans, Brown and Corbett (2001) (Chapter 5: x5.3); Baerman, Brown and
Corbett (2002b) (Chapter 3: x3.1); Corbett, Baerman and Brown (2002)
(Chapter 5: x5.4.1); Baerman (2005) (Chapter 3: x3.2, Chapter 4: x4.4.2 and
x4.5); and Baerman (2004) (Chapter 1: x1.5 and Chapter 4: x4.3). The material
here supercedes the earlier works.
We are very grateful to our friends and colleagues who read the book in
draft and gave us helpful comments from their different perspectives:
Jim Blevins, Jonathan Bobaljik, Andrew Carstairs-McCarthy, Martin
Haspelmath, Andrew Spencer, Greg Stump. The following gave helpful
feedback on specific sections: Helma van den Berg, Michael Cysouw,
Nicholas Evans, Roger Evans, while Lisa Mack substantially improved
the presentation of the draft. For help with the maps we are indebted to
Hans-Jörg Bibiko, and we thank Tom Khabaza and the Clementine soft-
ware for help with the data analysis in x3.8, and Marina Chumakina for
assistance with Russian data. We are very grateful to the ESRC for fund-
ing, under grant R000237939 and partially under grants R000271235 and
RES000230082. The University of Surrey Research Committee also pro-
vided timely support.
Abbreviations and symbols
1 first person
2 second person
3 third person
A transitive subject (where forms may differ from those
of the intransitive subject)
ABESS abessive
ABL ablative
ABS absolutive
ACC accusative
ADIT aditive
ADJ adjective
ALL allative
AN animate
CAR caritative
CAUS causative
CMP comparative
COM comitative
CONT contactive
COORD coordinative
DAT dative
DEF definite
DES designative
DIS disharmonic
DU dual
EL elative
ERG ergative
ESS essive
EXCL exclusive
F feminine
GEN genitive
xvii
xviii Abbreviations and symbols
HARM harmonic
HON honorific
HUM human
ILL illative
IMPRF imperfect
INAN inanimate
INCL inclusive
INDF indefinite
INESS inessive
INS instrumental
INTR intransitive
LOC locative
M masculine
N neuter
N- non- (e.g. NSG for non-singular)
NARR narrative
NOM nominative
OBJ object
PER perlative
PFV perfective
PL plural
PRF perfect
PROL prolative
PRS present
PST past
REL relative
S intransitive subject (where the forms may differ from
those of the transitive subject)
SBJ subject
SBJV subjunctive
SG singular
SUBORD subordinate
SUPERESS superessive
TAM tense-aspect-mood
TR transitive
TRI trial
TRANS translative
VOC vocative
Abbreviations and symbols xix
- used to form compound names for morphosyntactic

values, e.g. NOM-ABS = ‘nominative-absolutive case’
(a single morphosyntactic case in the language in question)
xy form ‘x’ alternates with form ‘y’
x/y feature value ‘x’ is syncretic with feature value ‘y’
x>y in a transitive verb, ‘x’ is the subject and ‘y’ is the object
[x] the feature value ‘x’ has no overt morphological expression
(e.g. English dog would be glossed as ‘dog[SG]’)
(x) the feature value ‘x’ is inherent to the lexeme, and has no
overt expression (e.g. French plage would be glossed as
‘beach(F)’)
1 Introduction
We might expect that a language’s sentence-structure and word-structure

would mesh rather straightforwardly. For instance, if the syntax of a
particular language distinguishes different arguments of the verb, and
the morphology distinguishes different cases, it seems natural to assume
that the two systems will line up. In real languages the situation is often
more complex. One of the most persistent and interesting problems at this
syntax–morphology interface is syncretism. As a first informal character-
ization, syncretism is the situation where the morphology ‘lets down’ the
syntax. To make that more concrete, let us take some Russian examples.
(Normally we shall give detailed glosses but here, so as not to build our
assumptions into the examples, we shall give only the basics.)
(1) Maša čitaet knigu
Masha reads book
‘Masha reads a book.’
(2) Na stole ležit kniga
on table lies book
‘A book lies on the table’, ‘There’s a book on the table.’
These sentences are representative, in that they show what they seem to
show. Russian distinguishes subject from object in its syntax, by a variety of
means (for example, the verb agrees with its subject but not with its object).
This appears to be reflected in the inflectional morphology. In (1) we have
knigu ‘book’, in the accusative case, as opposed to kniga in (2) when it is in
the nominative. (Similarly Masˇa in (1) is in the nominative.) As any reason-
able non-linguist would expect, the two systems work hand in hand to
distinguish subject and object and to facilitate the task of the hearer.
But now compare:
(3) Maša čitaet pis 0 mo
Masha reads letter
‘Masha reads a letter.’
1
2 The Syntax–Morphology Interface
(4) Na stole ležit pis 0 mo

on table lies letter
‘A letter lies on the table’, ‘There’s a letter on the table.’
Here the syntactic structures are as in the earlier examples, but the noun
fails to show the expected morphological distinction. We say that pis 0 mo
‘letter’ in (3) and (4) shows syncretism of nominative and accusative case.
We understand syncretism therefore as a mismatch between syntax and
morphology. We know that Russian syntax requires reference to subject
and object. This is reflected in the morphosyntactic category of case, which
distinguishes nominative and accusative (as in (1) and (2) above). However,
the morphology of pis 0 mo in (3) and (4) fails to make this distinction. The
key components of the definition are:
a. a morphological distinction which is syntactically relevant (i.e. it
is an inflectional distinction)
b. a failure to make this distinction under particular (morpho-
logical) conditions
c. a resulting mismatch between syntax and morphology.
Thus syncretism is the failure to make a morphosyntactically relevant
distinction.
A good way to look at it is to say that examples (1) and (2) set up the
expectation that there will be two forms of pis 0 mo in (3) and (4). Syncretism
is the breaking of that expectation; the nominative singular and the accu-
sative singular of pis 0 mo are identical. As Spencer (1991: 45) puts it ‘a single
inflected form may correspond to more than one morphosyntactic descrip-
tion.’ A similar definition is: ‘Identity in form between two grammatically
different inflections’ (Trask 1997: 215).
There are various questions to be asked about our example pis 0 mo ‘letter’.
For example, is it an odd exception, going against the general trend?
No, there are thousands of nouns in Russian which behave similarly; there
are also many thousands of the kniga ‘book’ type. And pis 0 mo does not fail to
draw certain other inflectional distinctions. It has a distinct locative, as shown
by: v pis 0 me ‘in the letter’. We might think that we could simply divide
Russian nouns into those which distinguish nominative and accusative and
those like pis 0 mo which do not. But here we find that kniga ‘book’, which
marks the distinction in the singular, fails to do so in the plural (both forms
are knigi).
We shall investigate which distinctions can fail to be drawn. We have
seen an instance where case is involved, but there are several other
Introduction 3
inflectional distinctions which may be treated similarly. We shall ask when

this occurs. It may be that our expectation is based on some lexemes (like
kniga) and is not met by others (like pis 0 mo). Or it may be that other
features provide the circumstances (kniga ‘should’ distinguish nominative
and accusative, as the singular shows, but it fails to do so in the plural). It
is not self-evident that all the phenomena which fall under the broad
umbrella of syncretism should be modelled in the same way, and so we
shall look carefully at the arguments for particular ways of treating differ-
ent instances of syncretism.
While we shall be inclusive in our coverage we shall ensure that we are
indeed dealing with inflection. That is, the expectation of a difference in
form must arise from the syntax and morphology of the given language.
Languages may have a distinct locative case, as in Russian, but there is no
language-internal evidence for such a morphological case in languages like
English. We shall not, therefore, treat English book as syncretic between
nominative and locative. Nor shall we be concerned with derivational
morphology. There are, of course, interesting coincidences of form in
derivational morphology, but we do not have the same expections of
consistency and completeness for derivational morphology as for inflec-
tion. Syncretism is also distinct from lexical homonymy, where there is a
single form with distinct meanings (as in bank (of river) and bank (financial
institution); this could be characterised as the lexicon letting down the
semantics.
1.1 History of the notion

The term ‘syncretism’ ultimately descends from the Greek tió&
‘union, federation of Cretan communities’ (Liddell and Scott 1996), refer-
ring to the practice that the continually feuding Cretan communities had of
laying aside their differences and banding together in the face of a common
enemy. In post-classical times Erasmus of Rotterdam reintroduced the
term, using it to designate ‘the coherence of dissenters in spite of their
difference of opinions, especially with reference to theological divisions’
(Herbermann, Pace, Pallen, Shahan and Wynne 1907–18). At some
point, through learned folk etymology, the term was confounded with
"ńni ‘to mix’ and its derivatives (e.g. ´ to& ‘mixed’), so
acquiring a more general meaning of a mixture of originally different
elements or viewpoints, typically with respect to religion (Grimm and
Grimm 2004). In the nineteenth century, use of the term became
fashionable in other realms as well. Pott (1836) is generally credited with

introducing the term into linguistics (Curtius 1863: 160; Wackernagel
1924: 32; Hjelmslev 1935–7: 59),1 where it is used to mean the diachronic
collapse of originally distinct inflectional forms, either through merger of
the forms, or through the merger of their underlying functions. Pott
himself uses the term only in passing (p. 638),2 but it is more generously
applied in the index (p. 792) – written not by Pott but by Heinrich Bindseil –
where it refers the reader to sections discussing the merger (typically
through sound change) both of case marking in nouns and person marking
in verbs. In spite of this early use of the term, it did not enter into general
use until the 1890s (Meiser 1992: 212, fn. 2). In the twentieth century, this
diachronic approach to syncretism was recast in synchronic terms within
the framework of structuralist linguistics (Hjelmslev 1935–7: 60; Jakobson
1936 [1971]: 67). On this approach, syncretism involves the contrast not
between an earlier and a later stage of a language, but between an under-
lying system and its concrete realization. This is how we understand
syncretism in the present work, though of course we shall not ignore the
diachronic processes that may have led to such a state of affairs.
1.2 Delimiting the notion of syncretism

Some authors have suggested that the term ‘syncretism’ should be reserved for
the products of certain diachronic developments. Roughly speaking, the
diachronic merger of forms within an inflectional paradigm can have two
sources, either as the result of blind phonological change, or the result of a
more complex morphosyntactic readjustment. Blind phonological change
can be illustrated by the merger of nominative and accusative singular in
first declension nouns in Vulgar Latin (5), a result of the regular loss of word-
final -m. Note that in other declension classes the two forms remained distinct,
as their difference was not solely due to the presence or absence of -m.
1
Curtius (1863: 160) attributes to Pott the notion of a ‘syncretic case’, i.e. a case historically
descended from two or more cases, as with the Greek dative or genitive (see (6)). However,
the reference he gives (‘Pott Et. Forsch. I1, 22’, namely Pott 1859: 22), although it deals with
this topic, does not include the term ‘syncretic case’ as such.
2
‘Im Lat. sog. Abl. und Dat. Plur. scheinen die Functionen des eig. Abl. [ . . . ], Instrumentalis
[ . . . ], Locativs [ . . . ], und endlich Dativs [ . . . ] vereinigt; dabei wird ebenfalls theilweise
Formen-, theilwiese vielleicht bloßer Begriffs-Synkretismus obgewaltet haben.’ (‘In Latin
the so-called ablative and dative plural appear to have united the functions of the original
ablative, instrumental, locative and dative; here too [as with the Greek dative] we see, in
part, syncretism of the forms, and perhaps also, in part, sheer syncretism of the concepts.’)
Introduction 5
(5) Nominative/accusative singular in Vulgar Latin (Coleman 1976: 50–4)
Classical Latin > Vulgar Latin
first declension NOM SG luna luna

ACC SG lunam luna
second declension NOM SG annus annus
ACC SG annum anno
third declension NOM SG pater pater
ACC SG patrem patre
Merger resulting from morphosyntactic change is illustrated by the devel-

opment of the Proto-Indo-European dative and locative singular, which
were combined in Ancient Greek into the case traditionally known as the
dative (6). By the laws of sound change that applied between Proto-Indo-
European and Ancient Greek, these two forms should be distinct in o-stem
(6) Dative/locative singular in Ancient Greek (Buck 1933: 180–5)
Proto-Indo-European > Greek
o-stem DAT SG -o
i -o
i
LOC SG -oi, -ei
consonant stem DAT SG -ei -i
LOC SG -i
and consonant-stem nouns, so their merger cannot be attributed to phono-

logical change. Nor can it be attributed to the outright loss of one of the cases,
since both the dative and locative have contributed to the syncretic forms:
* In the o-stems, the Greek form descends from the Proto-
Indo-European dative singular. The Proto-Indo-European locative
singular is marginally preserved in some adverbialized forms, such
as oikoi ‘at home’, originally from oikos ‘house’.
* In consonant stems, the Greek form descends from the Proto-
Indo-European locative singular. The original dative form has
been lost completely.3
3
In the a-stems, these two forms were already syncretic in late Proto-Indo-European,
presumably the result of sound change: early Proto-Indo-European dative singular *-
a-ei
and locative singular *-
a-i both developed into *-
ai (Szemerényi 1989: 200).
Thus, the Vulgar Latin example represents a superficial inflectional homo-

phony, the result of blind sound change. The Ancient Greek dative singular
results from a blending of both the forms and the functions of the Proto-Indo-
European dative and locative, the result of some fundamental reanalysis of
the system of morphosyntactic oppositions. Meiser (1992: 190) and Luraghi
(2000) both suggest that the term ‘syncretism’ be restricted to the second kind
of change. Luraghi suggests that the product of phonological change should
be called ‘homophony’ while Meiser (p. 190) proposes reviving the term
‘synemptosis’, used by ancient Greek grammarians for situations where a
morphological expression belonged to several grammatical categories.
However, though this distinction is one of undoubted theoretical signi-
ficance, in practice it is often difficult to draw the line. Consider the merger
of ablative and dative singular in second declension nouns in Latin (7).
This seems to have been the result of two independent sound changes. On
the one hand, final consonants were lost after a long vowel, so that ablative
(7) Ablative and dative singular in Latin (Buck 1933: 176, 181)
Proto-Indo-European/
Italic5 > Latin
second declension (o-stems) ABL SG -o

d -o

DAT SG -o
i -o

first declension (a-stems) ABL SG -a
d -a

DAT SG -a
i -ai (-ae)
singular -o d became -o (Buck 1933: 157). On the other hand, long
diphthongs were monophthongized, losing their second element, so that
dative singular -o
i became -o
(Buck 1933: 90). The first change was quite
regular and is also found in the first declension (thus, ablative singular -
ad
became - a). However, the development of the original long diphthongs in
Latin turns out to have been erratic. Sometimes they were monophthong-
ized, but in other contexts they were shortened; the conditions which
determined which change took place remain obscure (Leumann 1977:
271–2).4 While the dative singular of the second declension underwent
the first change, the dative singular of the first declension underwent the
second change, with - ai becoming -ai (orthographically -ae), though there
4
5
According to Leumann, V is the expected prepausal reflex, -Vi elsewhere.
The Proto-Indo-European a-stem dative singular ending was - ai. The ending -
ad is an
innovation within Italic, formed on analogy with the ablative of the o-stems.
Introduction 7
is evidence from early inscriptions that monophthongization took place

sporadically as well (Buck 1933: 176). Consequently, there is no syncretism
between ablative and dative singular in the first declension.
Although the merger of ablative and dative singular can be portrayed as
the result of sound change, a key element remains unaccounted for, namely
why did the long diphthong develop one way in one declension class and
another way in the other? For some reason syncretism was favoured in the
second declension but not in the first, in a way that does not obviously
follow from phonological developments. Such examples show that there is
not always a clear distinction between phonological and morphological
change, much less a way to classify phenomena whose history remains
unknown. It seems useful, then, to retain ‘syncretism’ as a cover term that
will apply to all instances of inflectional homophony, regardless of their
origin or interpretation; indeed, this is how the term was first used by Pott
(and Bindseil) in 1836.
1.3 Scope of the investigation

1.3.1 Typological methodology
The core of the present book is a cross-linguistic typological investigation
of syncretism, with two complementary goals. On the one hand, we explore
the logical space of syncretism: what features may be involved, and what
sort of patterns do these describe? On the other hand, we have aimed for a
diverse sampling of the world’s languages. In particular, we have brought
the evidence of non-Indo-European languages to bear, since these remain
relatively under-represented, a legacy of the fact that the notion of syncret-
ism was born in comparative Indo-European studies. To ensure genetic
breadth we followed, in part, the selection of languages used for the World
Atlas of Language Structures (Haspelmath, Dryer, Gil and Comrie 2005).
While genetic breadth can be achieved by the application of apriori
criteria, typological breadth – in order to see how many of the logical
possibilities are in fact attested – can be attained only by sifting through
masses of information, unconstrained by prior notions of what one may
find. This is a task which will never and can never be completed, but the
present study represents at least an introduction.
1.3.2 Selection of forms

In this book we focus our attention on syncretism between inflected whole
word forms. In principle, one could speak of syncretism between the
individual components of inflected words, e.g. by comparing prefixes to

prefixes, stems to stems and suffixes to suffixes, regardless of whether there
is homophony at the level of the word form as a whole. For example, in the
Nilo-Saharan language Mursi (8), the verb stem is identical in the first and
second singular of the indefinite aspect (baaio), and in the first and third
singular of the indefinite aspect (baaka). However, in neither aspect is there
syncretism at the level of the whole word, because 1SG is distinguished by
the prefix ka-.
(8) Singular indicative forms of the verb ‘eat’ in Mursi (Turton 1981:
341–2)
indefinite aspect definite aspect
1SG ka- baagio wa- ka- baaka

2SG baagio wa- baaku
3SG baato wa- baaku
Stump (2001: 217) terms this block syncretism, because the pattern obtains
within a given block of rules. So long as a word form is transparently
segmentable, such an approach has the advantage of widening the field of
investigation. But the question of segmentation into components (rule
blocks, morphemes or formants) is not always easy to resolve. For example,
take Pike’s (1965) analysis of the six present tense forms and the infinitive of
the German verb sein ‘to be’ (9). At the level of the whole word, these show
1PL/3PL syncretism alone.
(9) German sein ‘to be’
INF sein
1SG bin
2SG bist
3SG ist
1PL sind
2PL seid
3PL sind
Within these forms, however, eight distinct patterns of identity can be

isolated, as shown in (10): the elements b-, s-, -ei-, -ist-, -in-, -i-, -n- and -t
Introduction 9
(10) Patterns of identity in German sein (adapted from Pike 1965: 198)
3SG i s t
2SG b i s t
1SG b i n
3PL s i n d [t]
1PL s i n d [t]
2PL s e i d [t]
INF s e i n
(note that orthographic d in (10) is indistinguishable from t in word-final

position) each combine different person-number values. Whatever the merits
of such an analysis, it is not one which is compatible with most morpho-
logical models. The drawback to such an approach is that the more a word
is broken up into components, the more the resulting parts are peculiar to a
specific analysis. This is not desirable in the context of a large-scale typo-
logical investigation such as ours. On the other hand, the status of the
whole word, while hardly self-evident, is nevertheless more uncontro-
versial than that of such elements as morphemes or formants. By limiting
the investigation to whole word forms, we aim to keep the typological part
of the investigation theory-neutral.
1.4 Accidental versus systematic homophony

The focus of this book is inflectional morphology: what interests us are
instances of inflectional homophony that might be seen as systematic, that is
somehow represented in morphological structure. However, it is undoubtedly
the case that not all instances of homophony within inflectional paradigms are
morphologically encoded as such. Consider the Russian forms in (11).
(11) a. stem-stress ‘place’ b. end-stress ‘wine’
orthographic phonetic orthographic phonetic
NOM/ACC SG mesto :mje.st e vino vji.:no

GEN SG mesta :mje.st e vina vji.:na
For the noun mesto ‘place’, the genitive singular form is identical to the
nominative/accusative singular, while for vino ‘wine’, the genitive singular is
distinct. However, the difference between the two nouns is easily explained in
phonological terms. Russian has a general rule whereby /a/ and /o/ are distin-
guished only under stress. In this declension class, the nominative/accusative
singular ending is -o and the genitive -a, as reflected in the orthography. For
vino, these endings are stressed, and remain distinct, while for mesto they are
unstressed, and hence homophonous. The collapse of the genitive with the
nominative/accusative is a superficial by-product of phonology and need not
be reflected in a morphological analysis.
In general we have excluded such obvious examples of accidental homo-
phony from consideration. However, it should be borne in mind that the
question is seldom so clear-cut. On the one hand, a pattern of syncretism
may be restricted to a particular phonological environment without there
being any generally applicable phonological rule that would account for it.
On the other hand, there is evidence that originally accidental homophony
may be ascribed by speakers to a morphological rule, a reanalysis which
remains covert until revealed by diachronic processes (see Chapter 4:
x4.5.1). Therefore, if we have erred in our selection of material for presen-
tion, it has been on the side of inclusiveness.
1.5 Using this book

1.5.1 Supporting materials
The text is supported by a range of additional materials which have been
made available to the reader. There is an annotated bibliography of syncret-
ism (Baerman 2002a), which contains details of 100 items. This is freely
available online at: http://www.surrey.ac.uk/LIS/MB/Bibliography.htm.
Since this bibliography is available, we can restrict the references section in
the book to those items which are discussed in the text.
It is important that our account of syncretism is based on a wide range of
languages rather than on the few usual suspects. Our work is grounded on
an investigation of syncretism in a sample of genetically diverse languages.
These data are available in the Surrey Syncretisms Database (Baerman,
Brown and Corbett 2002a), which records all instances of syncretism in
thirty genetically diverse languages, comprising 1,256 separate entries.
This can be searched online at: http://www.smg.surrey.ac.uk. Its rationale
is explained in Brown (2001). The provision of this database means that
the reader can frame hypotheses about syncretism and investigate them
on-line. The essential information for using the database is provided in
readme files at the web address given. There is a second database covering
Introduction 11
a more limited phenomenon over a much broader range of languages,

cataloguing syncretism of subject person in intransitive verbs in 111 lan-
guages. Person marking was chosen for this large-scale survey because, of
all inflectional features, it is the one which is most comparable cross-
linguistically, the same core values being nearly universally present
(Baerman 2002b); this too can be searched over the web at http://
www.smg.surrey.ac.uk/. We participated in the research of the World
Atlas of Language Structure (Haspelmath, Dryer, Gil and Comrie 2005)
and contributed to maps, found in chapter 3 of the book, keyed to data
found in Appendices 1–3.
We wish to stress too that our analyses are not ‘cute’ accounts of
convenient selections of data. We have worked out full accounts of large
and complex morphological systems, within the Network Morphology
framework (see Corbett and Fraser 1993, Evans, Brown and Corbett
2001 and references there). We have implemented these analyses, so that
others can check that the analyses do indeed account for all the relevant
inflectional forms of the languages analysed. Sample fragments are given
in Appendices 4–6; the samples are simplified in parts to help the reader to
see the essentials. Full versions, to enable the reader to check the detail, are
available at http://www.surrey.ac.uk/LIS/SMG/Syncretism.
1.5.2 Glossing conventions

We adopt the Leipzig Glossing Rules (available at http://www.eva.mpg.de/
lingua/files/morpheme.html). In some ways these conventions codify
best practice, but where different possibilities are found they opt for one
alternative. This is helpful: to take a trivial example, there may be more than
one reasonable abbreviation for a feature value and it then makes sense to
standardize on one. Most of the abbreviations we use are from the Leipzig
list, but we have had to add a few (see our full list, p. xvii–xix). Here, for
illustration, is an example similar to those discussed earlier:
(12) Na polk-e lež-it knig-a
on shelf-LOC.SG lie.PRS-3SG book-NOM.SG
‘A book lies on the shelf’, ‘There’s a book on the shelf.’
If we were writing about word order, for example, this might be considered
an adequate gloss. For a book on syncretism, however, it is important to
be clear how potential ambiguities are handled in glosses. In example
sentences, linguists generally give a morphosyntactic gloss. This means
that the syntactic context is taken into account to resolve potential
ambiguities. Given the example John bid eighty pounds at the auction
yesterday, most linguists would gloss bid as past tense, even though out
of context it could be the imperative, among other things. In (12), polke is
glossed as locative singular. In principle it could be dative singular or
locative singular. We might gloss it as locative singular because the pre-
position na ‘on’ takes the locative case (as we know from nouns where
dative and locative are distinct). Where the theoretical ambiguity matters,
as generally it will, we either discuss it explicitly in the text, or we include all
the possibilities: LOC SG/DAT SG. Where forms are discussed in the text, then
the default is to give all the morphological descriptions, since there is no
syntactic context to favour one of them. Note that we are using the slash
for alternative glosses; it is available for this use because the other function
it may fulfil, namely to gloss fused subject and object markers, is carried
out in the Leipzig conventions by ‘>’.
1.5.3 Structure of the book

The book progresses from an empirical to a formal investigation of syn-
cretism. In Chapter 2 we lay out the typological parameters and define
basic concepts. Chapter 3 is a cross-linguistic survey of syncretism. In
xx3.1–3.7 we look at how syncretism affects individual features, and in
x3.8 we look at how different features interact. In Chapter 4 we look at the
implications these observations will have on how a formal model of
morphology accounts for syncretism, investigating some previous proposals.
In Chapter 5 we propose our own formal model, within the framework of
Network Morphology. The appendices contain the detailed data for the
statistical claims made in Chapter 3 and are based on our work for the
World Atlas of Language Structures.
2 Characteristics of syncretism
We shall map out the range of data to be considered. We include problems

familiar from the widely cited languages, but we shall also considerably
extend the scope of the discussion and of the languages investigated. We
first look at the patterns of syncretism and their implications (Chapter 2:
x2.1), and then examine the domains which allow us to compare paradigms
(Chapter 2: x2.2). This permits us to begin considering the types of analysis
available (to be discussed fully in Chapter 4). In Chapter 2: x2.3 the
important issue of directionality is raised. Then we consider the ‘extreme’
interpretations in Chapter 2: x2.4, namely neutralization and uninflected-
ness. These prove to describe the easy instances: the more challenging ones
lie in between and form the subject of the remaining chapters.
2.1 Syncretic paradigms

2.1.1 Types
As we have characterized it, syncretism involves the identity of cells within
an assumed morphosyntactic paradigm. Graphically, we shall represent
this by generating a complete inflectional paradigm, and enclosing the
identical forms within a box.1 Before considering what morphosyntactic
values may form the parameters, let us outline a brief typology of the
ways paradigmatic cells may be united. These are illustrated below with
examples of case syncretism. The different types will turn out to have
important consequences for the representation of feature structure.
In the simplest pattern, which we call simple syncretism, two or more
cells with different values for a feature are merged. For example, in Central
Alaskan Yup’ik (1), the absolutive and relative cases are identical in the
plural and dual.
1
Note that we do not ascribe any significance directly to the geometry. However, the graphic
representation gives a clear indication of the morphosyntactic description of the cells.
13
(1) Simple syncretism in Central Alaskan Yup’ik (Jacobson 1995: 469–71)
singular dual plural
ABS nuna nunak nunat ‘land’

REL nunam nunak nunat
LOC nunami nunagni nunani
ABL nunamek nunagnek nunanek
ALL nunamun nunagnun nunanun
PER nunakun nunagnegun nunatgun
CMP nunatun nunagtun nunacetun
If simple syncretism is compounded across different environments, the

result is nested syncretism. In the West Slavonic language Upper Sorbian
(2), a-stem nouns have syncretism of the dative and locative in the singular,
while all nominals have syncretism of the dative, locative and instrumental
in the dual. Thus the syncretic pattern of the singular can be said to be
nested within the larger syncretic pattern of the dual.
(2) Nested syncretism in Upper Sorbian (Šołćina and Wornar 2002: xx 4, 10, 12)
plural singular dual
NOM žony žona žonje ‘wife’

ACC žony žonu žonje
GEN žonow žony žonow
DAT žonam žonje žonomaj
LOC žonach žonje žonomaj
INS žonami žonu žonomaj
Multiple patterns of syncretism are also found in the paradigms of Nuer,

a Nilo-Saharan language (3). However, they are not nested within each
other as in Upper Sorbian. Each case may be syncretic with each other
case, without there being an implicational hierarchy. Thus the genitive is
syncretic with the locative in ‘egret’, but with the nominative in ‘girl’, while
the locative is syncretic with the nominative in ‘bug’.2 Following Williams
2
The default singular pattern is represented by ‘egret’, while the default plural pattern
involves no case distinction, e.g. bo— o— Nn— i ‘egrets.NOM/GEN/LOC.PL’. The other patterns are
restricted to small sets of nouns. It should be noted, though, that the default pattern
accounts for a relatively small portion of the Nuer noun lexicon. Frank (1999) notes that
Characteristics of syncretism 15
(1981), we term this contrary syncretism, because the pairings in each

paradigm are mutually exclusive.
(3) Nuer (Frank 1999: 84–6)
‘dog’ ‘egret’ ‘girl’ ‘bug’
NOM SG jiök bööN nyal baan

GEN SG jio— k bööNka— nyal baankä
LOC SG jio— o— k bööNka— nyaal baan
The juxtaposition of contrary syncretic patterns may imply the existence

of a morphosyntactic distinction which itself has no unique expression. For
example, in the Dardic language Phalura (4), the distribution of inflected
forms warrants the assumption of five cases: absolutive (or nominative),
accusative, dative-locative, ergative and genitive-ablative. However, in the
material described by Morgenstierne (1941), no paradigm displays more
than four distinct case forms. Neither the accusative nor the dative-locative
has a distinct form in any paradigm. The accusative is syncretic with the
absolutive in nouns, with the dative locative in plural pronouns. The dative-
locative is syncretic with the ergative in nouns and with the accusative in
pronouns. (Singular pronouns display only two forms, e.g. first person
absolutive/accusative/dative-locative ma, ergative/genitive-ablative mi .)
Following Zaliznjak (1973 [2002]: 629) we can speak of the accusative and
dative-locative as non-autonomous values. (Similarly, as pointed out by Jim
Blevins (personal communication), we might speak of non-autonomous
inflection classes, defined by their patterns of syncretism – compare ‘egret’
and ‘bug’ in the Nuer example in (3) above.)
The identification of non-autonomous values as syncretism is not
always certain, since in some instances this surface effect may be attributed
to a syntactic asymmetry. For example, most modern descriptions of
Latvian assume seven cases (nominative, accusative, genitive, dative, loca-
tive, vocative). However, there are a few prepositions which govern the
accusative or genitive case in the singular, but the dative in the plural
(Mathiassen 1997: 184–5). By distributional criteria one might wish to
postulate two additional cases to account for this. However, the fact is
only 22 out of his corpus of 264 nouns are fully regular; in order to fully account for the
nouns in the corpus, one would need 207 distinct inflectional classes.
(4) Non-autonomous cases in Phalura (Morgenstierne 1941: 16–18)
noun pronoun
‘man’ ‘we’
ABS mi:š be
ACC mi:š asa:m
DAT-LOC mi:ša asa:m
ERG mi:ša asim
GEN-ABL mi:ši asi:
that all prepositions govern the dative case in the plural, so it may be just as
well to define the case government of individual prepositions only in the
singular, alongside a global stipulation about their case government in the
plural (see Fennell 1975 for a discussion).3
Finally, it is also possible to identify syncretism on the basis of an unusual
distribution of forms within the paradigm of an individual lexeme. Consider
the Old Irish paradigm in (5), representative of masculine o-stem nouns.
What concerns us here is the behaviour of the form fir, which appears in
both the singular and plural. Within the case paradigm for each of these
numbers, the form is not syncretic: in the singular it serves as the genitive
and in the plural it serves as the nominative. But all the same, by our criteria
fir is syncretic, since it fills two cells in the morphosyntactic paradigm of the
lexeme. Nor can these two cells easily be collapsed, representing as they do
distinct values for both case and number. Adapting a term from Cushitic
studies, we call such patterns polarity effects (see Chapter 3, x3.7).
(5) Polarity effect in Old Irish (Pokorny 1923: 77)
singular plural dual
NOM fer fir fer ‘man’

ACC fer firu fer
GEN fir fer fer
DAT fiur feraib feraib
3
Traditionally, grammatical descriptions of Latvian have concerned themselves only with the
behaviour of the preposition ar ‘with’. Historically, this governed the instrumental case, since
fallen together with the accusative in the singular. Its persistence as a separate case in the
Latvian grammatical tradition is thus due to a mixture of diachronic and distributional criteria.
2.1.2 Implications
Each of these patterns has different implications for the possible relation-
ship between the feature values involved. Simple syncretism suggests a link
between two values, as in the Central Alaskan Yup’ik example, repre-
sented in (6a). Nested syncretism suggests a branching hierarchy of links,
as in the Upper Sorbian example, represented in (6b). Contrary syncretism
suggests multiple independent links, as in Nuer, represented in (6c).
(6) a. b. c.
ABS REL DAT LOC INS NOM GEN LOC
Polarity effects cannot readily be portrayed as a relationship between

values of a feature. The ramifications that these patterns have for
morphological models are discussed in depth in Chapter 4.
2.2 Domains of comparison

In order to identify syncretism, we typically compare different paradigms
involving the same feature values and note a discrepancy between them in
the distribution of forms. In the previous section we looked at the different
sorts of discrepancy that are possible. In this section we look at the domains
of comparison, that is what kinds of paradigms can be compared with each
other? We illustrate the possibilities with syncretism of the feature person.
2.2.1 Morphological classes

Syncretism may be concomitant with variation in the inflectional
morphology. That is, some affixes may be syncretic, and others not,
as in Abipon (7), an extinct Guaicuruan language formerly spoken in
Argentina. There are seven conjugation classes, each associated with diff-
erent person-marking prefixes; the second person suffix -i remains con-
stant throughout. In class IV, the first person and third person prefixes are
syncretic. (In all conjugation classes the form of the stem remains
constant.)
(7) Verbal affixes in Abipon (Najlis 1966: 32–3)4
I II III IV V VI VII
1 ei
n ha- ri- m- eh- ah- li-
2 n- -i Ø- -i gr- -i m- -i e- -i Ø- -i l- -i
3 n- r- n- m- ei- ai- l-
Alternatively, the inflectional affixes may be the same for all verbs but
interact differently with different stem classes. For example, in the Carib
language Macushi (8), the distinction between 2SG a- and 3SG i- is lost when
these are prefixed to vowel-initial stems.5
(8) Verbal prefixes in Macushi (Abbot 1991: 101)
C initial stem V initial stem
initial heavy syllable (VC, V: or VV) initial light syllable (V)
1SG u- Ø- Ø-
2SG a- aw- (stem-initial V:)
3SG i- aw- (stem-initial V:)
A related possibility is where a single set of affixes is used, but occupy

different positions in different inflection classes, and so interact differently
with the stem. For example, in Somali (9), person-number markers are
prefixed to a few verb stems, and otherwise are suffixed. When prefixed,
1SG Ø- is distinct from 3SG M y-, but when suffixed, this distinction is lost.
(2SG and 3SG F are syncretic in either case.)
4
Some morphological variation is not shown in the table: (i) before vowel-initial stems, the
prefix-final vowels of I and II are deleted, and second person prefix of II is h-, (ii) the first
person prefix of III is gr- when the stem takes a plural suffix, and (iii) the third person prefix
in II and III may be i-, under conditions which are not explained by Najlis (1966).
5
Although the stem classes can be characterized in phonological terms, and the syncretic
patterns appear to originate in the phonological interaction of prefix and stem, the syncret-
ism itself cannot entirely be reduced to a synchronic phonological rule. Thus, while the
structure of the prefix aw- seems to be due to constraints on syllable structure (V + heavy V
yields two syllables, and non-initial syllables must be C-initial), there is no automatic rule
whereby i fi a before w (e.g. i-wa0 ka-ri 3-axe-POSSESSION ‘his axe’ (Abbot 1991: 85)).
(9) Simple past in Somali (Saeed 1999: 86, 98)
prefix conjugation ‘say’ suffix conjugation ‘wait’
1SG Ø-idhi sug-Ø-ay

2SG t-idhi sug-t-ay
3SG F t-idhi sug-t-ay
3SG M y-idhi sug-Ø-ay
Finally, the full range of inflectional distinctions may depend on the

interaction of a stem alternation with affixation, where both exhibit block
syncretism (see above, x1.4.2). If there is a class of stems lacking the stem
alternation, this leads to syncretism. For example, in German (10), the
ending -t marks both 3SG and 2PL. Some stems have a vowel alternation in
the 2SG and 3SG, which disambiguates the forms in -t (10a). Other stems
lack this alternation, yielding 3SG/2PL syncretism (10b).
(10) Present tense in German
a. ‘give’ b. ‘live’
1SG geb-en leb-en

2SG gib-st leb-st
3SG gib-t leb-t
1PL geb-en leb-en
2PL geb-t leb-t
PL geb-en leb-en
2.2.2 Feature values

Person features occur on nominals, typically to mark possession, and on
verbs to refer to arguments. Given its distribution across different word
classes, person occurs in the presence of a wide variety of other morpho-
syntactic features (including case, number, gender, tense, voice, negation),
thereby allowing us to see how the paradigm of one feature may be
influenced in a variety of contexts. The presence of particular values of
these accompanying features can affect the differentiation made in the
person paradigm.
Case
In languages which mark case and possession on nouns, where possession
marking realizes the person (and number) of the possessor, we can observe
the interaction of case and person. In the Samoyedic language Selkup (11)
a variety of cases use the same form for the second person singular
possessor and third person singular possessor.
(11) Syncretism in the context of case in Selkup (Helimski 1998: 558–89)6
possessor
1SG 2SG 3SG
NOM nommı̈ nomlı̈ nomtı̈ ‘God, heaven’

GEN nomnı̈ nomtı̈ nomtı̈
ACC nommı̈ nomtı̈ nomtı̈
INS nomnı̈sä nomtı̈sä nomtı̈sä
CAR nomnı̈kåålı̈k nomtı̈kåålı̈k nomtı̈kåålı̈k
TRANS nomno(qo) nomtoo(qo) nomtoo(qo)
COORD nomnı̈šj ak nomtiš j ak nomtiš j ak
DAT-ALL nomnı̈nı̈k nomtı̈nı̈k nomtı̈nı̈k
ILL, LOC, EL nopqäk nopqäntı̈ nopqı̈ntı̈
PROL nommäk nommäntı̈ nommı̈ntı̈
Person
Syncretism of person in the context of person may occur in verbs that mark
more than one argument. In the Papuan language Yimas (12), of the Sepik-
Ramu family, second and third person singular subject values in transitive
verbs are syncretic when the object person is third singular.
(12) Syncretism in the context of person in Yimas (Foley 1991: 217)
object
1SG 2SG 3SG
1SG — kampantayłcut nakatayłcut

subject
2SG maNatayłcut — nantayłcut

3SG naNatayłcut nanantayłcut nantayłcut
6
Some of the forms with 1SG possessor are optionally based on the stem nuu-; these have been
omitted.
Gender
Similar to case, gender is a feature which is prone to syncretism (see
Chapter 3, x3.1, x3.4 and x3.8.3), but gender may still provide feature
values which determine the form of other features, such as person. In the
Indo-Aryan language Sindhi (13) the gender value ‘feminine’ in the plural
determines syncretism of second and third person.
(13) Syncretism in the context of gender in Sindhi (Khubchandani 2003: 647)
masculine feminine
singular plural singular plural
1 lIkh-«nd-UsI lIkh-«nd-ası̃ lIkh-«nd-«sı̃ 1Ikh-«nd-yũsı̃ ‘write’

2 lIkh-«nd-ẽ lIkh-«nd- lIkh-«nd-ı̃«˜ lIkh-«nd-yũ
3 lIkh-«nd-o lIkh-«nd-a lIkh-«nd-i lIkh-«nd-yũ
Number
In the Papuan language Kobon (of the Trans-New Guinea phylum), the value
‘dual’ of the number paradigm accompanies a loss of distinction between the
second and third person for all verbs (here illustrated with a past tense form).
(14) Syncretism in the context of number in Kobon (Davies 1981: 166)
1 arnö arno arlo ‘went’

2 arna arbe arlö
3 ara arla arlö
Tense-aspect-mood
Person syncretism can also occur in the presence of tense-aspect-mood. In
Aymara (spoken primarily in Bolivia), verbs may have syncretism between
first and second person in the future perfect but have separate forms for
this in the present tense (15).
Voice
The verbal feature of voice can also be a context for person syncretism. In
Gothic (16), the distinction between first and third person singular is lost in
(15) Syncretism in the context of tense-aspect in Aymara (Deza Galindo

1992: 103–5)
present future perfect
1SG muntua munchiyäta

2SG muntahua munchiyäta
3SG munihua munchı̈na
(16) Syncretism in the context of voice in Gothic (Wright 1930: 135–6)
indicative subjunctive
active passive active passive
1SG nima nimada nimau nimaidau ‘take’

2SG nimis nimaza nimais nimaizau
3SG nimiþ nimada nimai nimaidau
1PL nimam nimanda nimaima nimaindau
2PL nimiþ nimanda nimaiþ nimaindau
3PL nimand nimanda nimaina nimaindau
the passive voice. That this is systematic is suggested by the fact that
this syncretism also occurs in the subjunctive mood, but with different
forms. The passive combined with the plural provides a context for
complete loss of person distinctions in both the indicative and the
subjunctive.
Negation
Where negation is realized in the morphology it is possible to find person
interacting with negation. In the positive paradigms of the Nilo-Saharan
language Dongola Nubian (17), here represented by the present indicative,
four person-number forms are distinguished, while in the negative, only
two forms are distinguished, one for 3PL and one for all other values.
Some observers have seen the relationship between feature values and
syncretism in terms of markedness. That is, the more complex ‘marked’
feature value is associated with diminished variability (Moravcsik and
Wirth 1986: 1–3). One explanation put forward for loss of distinctions in
the presence of a ‘marked’ value is that the system avoids an excessive
number of ‘marked’ values (for instance, Cairns 1986: 18, McCarthy 2002: 81).
(17) Syncretism in the context of negation in Dongola Nubian

(Armbruster 1960: 195, 205)
present indicative ‘drink’
positive negative
1SG niri nimunun

2SG nin nimunun
3SG nin nimunun
1PL niru nimunun
2PL niru nimunun
3PL niran nimunan
As Cairns (1986: 18) puts it, ‘in a marked member of any given opposition,
fewer other oppositions are likely to occur, because, for each further
opposition, there would have to be a marked member.’ Some of the
examples do seem to be compatible with this approach, e.g. syncretism
under dual number in Kobon, the passive in Gothic, or negation in
Dongola Nubian. On the other hand, the predictive value of markedness
should not be exaggerated, as the quirky distribution of forms in Selkup
shows. In addition, one should note that in all the examples adduced above –
indeed, in most examples – differences in feature values are concomitant
with differences in the inflectional morphology. The interpretation of such
examples thus confronts an inherent ambiguity: is the syncretism a property of
the conditioning feature value, or of the morphology that realizes that
paradigm?
2.3 Morphological characteristics

The morphological realization of a syncretic pattern may in itself influence
its interpretation. Below we consider three characteristics, regularity,
directionality and morphological unmarkedness.
By regularity we mean the repretition of a syncretic pattern across multiple
exponents. Consider the syncretism of dative and ablative in Latin (18). In
the plural, these cases are syncretic for all declension classes, realized in some
by the ending -is and in others by the ending -ibus. Not only do these endings
look different, they have distinct diachronic origins (Buck 1933: 182, 186).
In Kashmiri (19), ergative and ablative plural are also syncretic for all
declension classes. However, this ending, -av, is the same in all declension
classes, even though other endings might differ.
(18) Latin
second delension ‘star’ third declension ‘chief’
NOM stella stellae princeps principes

ACC stellam stellas principem principes
GEN stellae stellarum principis principium
DAT stellae stellis principi principibus
ABL stella stellis principe principibus
(19) Kashmiri (Kachru 1969: 112–16)
first declension: ‘child’ second declension ‘tree’
ABS gobur gobar kul kul’

ERG gobran gobrav kul’ kul’av
ABL gobri gobrav kuli kul’av
DAT gobur gobran kulis kul’an
The regularity of the syncretic pattern in Latin over two distinct endings
suggests that this pattern is systematic. Otherwise, we would have to
assume two instances of accidental homophony which happen to coincide.
The Kashmiri example is ambiguous; since only one ending is involved, the
notion that this is accidental is more plausibly entertained than in the case
of Latin.
The second characteristic, directionality, concerns the possible morpho-
logical affiliation of the syncretic form to one of its component values.
Consider the imperfective paradigms from the Dagestanian language Lak,
shown in (20). The syncretic 1SG/2SG past tense ending is not morphologic-
ally related to any of the endings in the non-syncretic present tense para-
digm. Now consider the paradigms from the Tungusic language Udihe,
likewise illustrated in (20). The syncretic 1SG/2SG ending of the future tense
is identical to the 2SG ending as found in the past tense.
The situations in Lak and Udihe present different descriptive tasks.
In the case of Lak, there is a static relationship between the endings and
the morphosyntactic values they express; we need only state that the
(20) Lak (Xajdakov 2001: 354) and Udihe (Nikolaeva and Tolskaya 2001:
212–13)
Lak imperfective ‘stand up’ Udihe ‘sing’
present past past future
1SG izan-na izajssij-av jexe:-mi jexezeNe-i

2SG izan-ssara izajssij-av jexe:-i jexezeNe-i
3SG izan-ssar izajssij-a jexe:-ni jexezeNe-ni
ending -av expresses both 1SG and 2SG. In the case of Udihe, the relationship
between the ending and the morphosyntactic values varies. We must some-
how account for the fact that the ending -i sometimes functions solely as 2SG,
and sometimes as both 1SG and 2SG. We term this a ‘directional effect’,
because it looks as if the form for one value is the source of the form for the
other value; e.g. in Udihe, it looks as if the 1SG future form is based on the 2SG.
The third characteristic, unmarkedness, concerns the possible relationship
within a paradigm between a morphologically unmarked form (the bare stem,
lacking affixes or evidence of other morphological operations) and syncretism.
For example, in the Francisco Leon dialect of Zoque (a Mixe-Zoquean lan-
guage), the bare stem is used for 1SG and 3SG. This is illustrated in (21) using the
preterite form; the person-number markers do not vary with tense. (Ny-
indicates nasalization of the initial consonant of the stem, followed by -y-.)
(21) Francisco Leon Zoque (Engel, Allhiser de Engel and Mateo Alvarez
1987: 378)
poyu ‘ran’ underlying forms of affixes
1 poyu potyamu Ø Ø -tam

2 mbyoyu mbyotyamu Ny- Ny- -tam
3 poyu poyaju Ø Ø -yaj
Here, syncretism coincides with the morphologically unmarked form,

which may suggest a causal connection between the lack of inflectional
marking and syncretism: Francisco Leon Zoque lacks 1SG and 3SG
morphology and uses the bare stem to fill in this gap.7 In this case the use of
the bare stem form also coincides with what are commonly seen as the
default morphosyntactic values for person and number, namely third
singular. But one also finds instances where morphological and morpho-
syntactic unmarkedness do not line up. The present tense forms of English
are a familiar example (walks versus walk). A similar example comes from
the far past conjugation of the Papuan language Orokaiva (Trans-New
Guinea phylum) (22), where only the 3SG and 2PL have overt endings, while
the bare stem is used elsewhere (the final -a is the indicative mood marker).
On the other hand, syncretism may fail to coincide with zero morphology –
in the other tense-aspect-mood paradigms (roughly two dozen), Orokaiva
uses a different set of person-number endings, where 1PL and 3PL are
syncretic, but it is the 2SG form which lacks an overt person marker. This
is shown by the mid-past B form in (22).
(22) Orokaiva ‘walk’ (Healey, Isoroembo and Chittleborough 1969: 40, 59, 62)
far past mid-past B
1 hembu -a hembu -a hembu -ha -n -a hembu -ha -r -a

2 hembu -a hembu -w -a hembu -ha -a hembu -ha -w -a
3 hembu -n -a hembu -a hembu -ha -j -a hembu -ha -r -a
In this case of the mid-past B forms in (22), the syncretic pattern cannot
so readily be attributed to the extension of a morphological default. Such
patterns may also be found where the morphological default seems to
coincide with the morphosyntactic default, as in the Niger-Congo lan-
guage Dogon, where the bare stem is used for the 3SG, but syncretism
affects the 1PL/2PL. (The 3PL ending varies in form with aspect-mood, but
otherwise there is a single set of person-number endings for all aspect-
mood paradigms.)
7
Engel, de Engel and Alvarez (1987: 379) report that the enclitic first person pronoun -0 ijtzi
is
used to distinguish between 1SG and 3SG; judging by the examples in the text, its use is not
obligatory
(23) Dogon, habitual form of ‘stay’ (Plungian 1995: 30–1)
singular plural
1 wada-de-m wada-de-y
2 wada-de-w wada-de-y
3 wada-de wada-d-iN
2.4 A typology of interpretations

It has been shown in Chapter 2: x2.1.1 that the comparison of multiple
paradigms allows us to determine the presence in a language of particular
feature values and determine where the distinctions represented by those
values may be collapsed in the morphology. In other words, it can be
determined that particular morphosyntactic distinctions are present
underlyingly, even if absent morphologically. However, if we mechanic-
ally combine all the logically possible morphosyntactic distinctions
which have an expression in form, without considering the relationship
between syntax and morphology, this will produce an abundance of
information, not all of which is of equal interest. Some of the maximal
paradigms predicted by comparing multiple paradigms will turn out not
to be relevant for syntax. This is neutralization, which is discussed in
Chapter 2: x2.4.1. At the other end of the spectrum, morphology may fail
to distinguish any values of a syntactically relevant feature. This is
uninflectedness, which is discussed in Chapter 2: x2.4.2. Neutralization
and uninflectedness are violations of two properties of what we may call
‘canonical inflection’: (a) that feature values found in a language com-
bine mechanically into fully distinguished combinations, and (b) that
morphology always distinguishes syntactically relevant feature values.
Neutralization violates property (a), and uninflectedness is a straight-
forward violation of property (b). There is a more challenging violation
of property (b), which is canonical syncretism. This is discussed in
Chapter 2: x2.4.3. While uninflectedness involves a total failure to distinguish
syntactically relevant values, canonical syncretism involves partial fail-
ure to do so. We therefore need to distinguish between these three related
ways of understanding syncretism. In order to do this we will, in examples
(24)–(31), have two separate glosses for the feature values differentiated
by morphology and by syntax.
2.4.1 Syncretism as neutralization

Across the spectrum of possibilities for syncretism, certain instances can be
accounted for easily. For example, Russian pronouns, adjectives and verbs
all distinguish three genders (masculine, feminine and neuter) and two
numbers (singular and plural). As gender is an agreement category for
these three word classes, it is therefore syntactically relevant. However, in
Russian, gender is only distinguished in the singular; there is no context in
which gender is distinguished in the plural.8
(24) knig-a interesn-a

morphology: book(F)-SG.NOM interesting-F.SG
syntax: book(F)-SG.NOM interesting-F.SG
‘the book is interesting’
(25) rasskaz interesen

morphology: tale(M)[SG.NOM/ACC] interesting.M.SG
syntax: tale(M)[SG.NOM] interesting.M.SG
‘the tale is interesting’
(26) predloženi-e interesn-o

morphology: proposal(N)-SG.NOM/ACC interesting-N.SG
syntax: proposal(N)-SG.NOM interesting-N.SG
‘the proposal is interesting’
In examples (24)–(26) gender is syntactically relevant in that this infor-

mation must be available to the predicate adjective in order for it to agree.
Seen from another perspective, the presence of a particular agreement
marker on the adjective constrains the choice of noun which the adjective
may occur with, and this also disambiguates the case syncretism repre-
sented by the morphological glossing for the nouns in (25) and (26),
because the nouns in (25) and (26) are required by the syntax to be in the
nominative case so as to control predicate agreement. We can contrast the
role of gender in the singular with the situation in the plural.
(27) knig-i interesn-y

morphology: book(F)-PL.NOM/ACC interesting-PL
syntax: book(F)-PL.NOM interesting-PL
‘the books are interesting’
8
The one possible exception to this is the quantifier ‘both’ oba obe which, in the prescribed
written standard at least, has oblique forms which distinguish gender.
(28) rasskaz-y interesn-y

morphology: tale(M)-PL.NOM/ACC interesting-PL
syntax: tale(M)-PL.NOM interesting-PL
‘the tales are interesting’
(29) predloženi-ja9 interesn-y

morphology: proposal(N)-PL.NOM/ACC interesting-PL
syntax: proposal(N)-PL.NOM interesting-PL
‘the proposals are interesting’
In (27)–(29) the morphological gloss of the form interesny does not

contain any gender values. As an alternative, we could have added the
choice of values F/M/N to the morphological gloss for interesny. Because
this choice would involve all possible values for gender, its presence is
syntactically no more constraining than its absence. This demonstrates
that, although the noun lexemes have inherent lexical gender, syntax does
not need to make reference to gender in plural contexts.10 It can be
concluded that, for Russian, gender in plural contexts is syntactically
irrelevant. This can therefore be easily accounted for as an instance of
syntactic neutralization, which, because the form reflects the irrelevance of
the feature for syntax, it is not the role of inflectional morphology to
account for.11
9
We have transliterated, rather than transcribed, the Russian examples. This affects the
segmentation of morphological glossing, because certain of the so-called ‘soft’ vowel letters
may actually represent the last element of a root or stem as part of the vowel of the ending.
This is the case with the form predlozˇenija, for example, where /j/ is actually the last
phoneme of the stem but is represented as combined with the final vowel /a/ by the letter
~, which represents combinations of /j/ plus /a/ or indicates that the consonant before /a/ is
palatal. As we have transliterated the examples, we have accordingly made the morpheme
divisions, as if they corresponded to the written words. Similar comments can be made
about other Russian examples in this chapter, such as (33), but this issue is of no material
importance for the points being made here.
10
This generalization includes the agreement of the morphologically plural, but semantically
singular, honorific vy ‘you’ with long-form predicate adjective in the singular.
Vy molčaliv-aja
You silent-F.SG.NOM
‘You are silent.’ (addressed to a woman) (Corbett 1983: 53)
The challenge for syntactic theory here is to account for the singular agreement. Given the
singular form of the adjective, we expect gender agreement.
11
It is of interest for typology, as shown by Greenberg and many authors following him.
Absolute neutralization is covered by typological claims of the type ‘A language never has
more gender categories in nonsingular numbers than in the singular’ (Greenberg 1963: 95).
The total absence of distinct values of a particular feature, in a given

context, is necessary, but not sufficient to define neutralization. Because
neutralization involves lack of syntactic relevance, total absence of dis-
tinctions for one set of syntactic objects in a given syntactic context is not
sufficient, as other syntactic objects may still realize that set of distinctions.
Hence, syntactic relevance means absence of distinctions within the mor-
phology of all word classes.
Neutralization
Neutralization is defined as follows:
i. In the presence of a particular combination of values of one or
more other features (the context), there is a general loss of all
values of a particular feature F found elsewhere in the language.
ii. No syntactic objects distinguish any values of feature F in the
given context, and feature F is therefore syntactically irrelevant in
that context.
Given the examples in (27)–(29), neutralization represents the most
straightforward explanation: the lack of formal distinction merely reflects
the irrelevance of the feature in question for syntax.
2.4.2 Syncretism as uninflectedness

Whereas it is possible on the one hand to account for lack of formal
differentiation as reflecting syntactic irrelevance, it is also possible to
interpret it as reflecting a lack of response by morphology to syntactically
relevant distinctions. This is uninflectedness. For example, the Russian
noun pal 0 to ‘coat’ is indeclinable and so has no distinct number or case
forms. However, a typical Russian noun does inflect for case and number,
as does any adjective which agrees with pal 0 to. Nouns like pal 0 to therefore
constitute a particular morphological class which is insensitive to distinc-
tions which are still syntactically relevant.
With the Russian noun pal 0 to, there is no syntactic feature value which is
the context for its uninflectedness. This contrasts with neutralization,
where at least one feature value is present as a context, because otherwise,
if all syntactic objects failed to make distinctions for a given feature when
there is no context, there would be no evidence for the existence of that
feature. Uninflectedness may still occur in a particular morphosyntactic
context, however. In Polish, nouns such as muzeum ‘museum’ do not
inflect in the singular, but do in the plural (Kotyczka 1980: 95, 105–6;
Tokarski 1993: 257). Importantly, the uninflected singular forms can still
be used in different syntactic contexts.
(30) now-ego muzeum
morphology: new-M.SG.GEN museum(M)[SG]
syntax: new-M.SG.GEN museum(M)[SG.GEN]
‘of (the/a) new museum’
(31) now-ych muze-ów

morphology: new-PL.GEN museum(M)-PL.GEN
syntax: new-PL.GEN museum(M)-PL.GEN
‘of (the) new museums’
In (30) the same form can be used in the genitive singular context as the
nominative singular, even though there is no genitive singular inflection as
such, whereas there is a specific genitive plural inflection in (31).12 This
cannot be treated as neutralization, because a typical Polish noun will inflect
for case in the singular, and hence this feature is syntactically relevant.
Therefore while the value ‘genitive’ is absent in the morphological gloss of
muzeum in (30), the syntax treats the form as though it were genitive (as with
similar forms of the Russian pal 0 to ‘coat’).
Occasionally it may be more difficult to distinguish between instances of
uninflectedness and neutralization. The verbal system of Russian is an example
of this. Russian verbs have two synthetic paradigms, past and non-past.13 Past
tense forms mark gender and number (32) but do not mark person.
(32) Russian verb ‘played’
singular plural
M igral igrali
F igrala igrali
N igralo igrali
In contrast, non-past forms mark person and number (33).
12
That this contrast between uninflectedness in the singular and inflectedness in the plural is
reflected in use is easily checked by searching Polish language websites for the different
number and case combinations. For instance, we can find example sentences containing
the noun phrases nowego muzeum and nowych muzeów.
13
Simple non-past forms have either a future interpretation, if a verb is perfective, or a
present tense interpretation, if a verb is imperfective.
(33) Russian verb ‘play’
singular plural
1 igraju igraem
2 igraeš 0 igraete
3 igraet igrajut
However, in contrast with the examples (27)–(29), where we saw that no

agreement targets in Russian mark gender in the plural, we cannot say that
gender is neutralized in the non-past, because it may be relevant for other
agreement targets.
(34) Esli vopros okazyva-et-sja glup-ym . . .

if question(M) turns.out-3.SG-REFL stupid-M.SG.INS
‘If the question turns out to be stupid . . . ’
In the past tense equivalent of (34) the verb would also require the
correct gender form to agree with the subject, but the non-past form in
(34) never marks gender. In (34) the non-past form of the verb okazyvat 0 sja
‘to turn out’, which has a reflexive marker added to the non-past person
and number marking, occurs with a predicate adjective which marks
gender. Hence, in this example, although the verb does not distinguish
any gender features in the non-past, this differs from neutralization, as
gender features may still be syntactically relevant. Therefore, we say that
gender is present syntactically, but that the non-past verb forms are unin-
flected for gender. Of course, syntactic relevance or presence will depend
on the view one has of syntactic structure, and so there are examples of
uninflectedness which are harder to distinguish from neutralization, as is
the case for the Russian verb.
Uninflectedness, in common with neutralization, involves total absence
of distinctions for a given feature. In contrast with neutralization, this total
absence of distinctions may be limited to a particular morphological or
syntactic class. As there are other classes which maintain the appropriate
distinctions in the same context, this feature is still relevant for syntax.
Hence, whereas neutralization is about syntactic irrelevance as reflected in
morphology, uninflectedness is about morphology being unresponsive to a
feature that is syntactically relevant.
Uninflectedness is defined as follows:
i. There is, in certain lexemes only, a loss of all values of a particular

feature F found elsewhere in the language. This loss may depend
on the presence of a particular combination of values of one or
more other features (the context).
ii. Other syntactic objects distinguish values of feature F, either
generally or in the given context, and feature F is therefore
syntactically relevant.
Uninflectedness and neutralization deal with the interaction of morpho-
syntactic features. They are both important theoretically, because they
suggest possible ways of treating syncretism. They are also the most
straightforward parts of the spectrum of morphology-syntax interaction
which involve loss of feature distinctions, as they can be accounted for
by constraints on feature combination. The more problematic part
of the spectrum, canonical syncretism, cannot be treated in terms of
uninflectedness or neutralization, unless the feature set is modified in
some way.
2.4.3 Canonical syncretism

Uninflectedness and neutralization cannot readily account for the chal-
lenge of syncretism involving the partial collapse of syntactically relevant
feature distinctions. For uninflectedness, where the feature involved is
syntactically relevant, the problem is accounting for the fact that the
morphology is only partially inert, as opposed to totally inert. The prob-
lem for neutralization, in contrast, is that the feature is still syntactically
relevant.
The partial loss of distinctions can be illustrated by Slovene, where the
l-participle, as other agreement targets, marks masculine, feminine and
neuter genders in the plural and singular (35).
(35) Slovene l-participle of the verb ‘write’
M pisal pisala pisali

N pisalo pisali pisale
F pisala pisali pisala
In the dual, the distinction between neuter and feminine is collapsed,

while there is still a separate masculine form. This is therefore an instance
of gender syncretism, which cannot be treated in terms of gender neutral-
ization or uninflectedness, as we have seen for Russian in the plural and
past tense respectively.
Canonical syncretism is defined as follows:
i. There is, in certain contexts, a loss of distinctions between some
but not all values of a particular feature F. This loss may depend
on the presence of a particular combination of values of one or
more other features (the context).
ii. Other syntactic objects distinguish those values of feature F, and
they are therefore syntactically relevant.
The Slovene example (35) involves syncretism of gender in the dual, where
the gender distinctions are only partly lost, which naturally entails that
gender is still syntactically relevant in the dual. The fact that the loss of
feature distinctions is not total, and therefore still involves values of the
feature in question, means that this part of the spectrum of morphology–
syntax interaction is the most challenging. Unless the inventory of feature
values is further augmented, it cannot be claimed that the loss of distinc-
tion reflects lack of relevance for syntax (as would be the case with
neutralization), nor is it just a fact about morphology that it is featurally
inert in dual contexts (as would be the case with uninflectedness).
It may be tempting to extend the notions of uninflectedness and neu-
tralization to instances of canonical syncretism. For instance, the Russian
long-form adjective collapses distinctions between masculine and neuter
genders in the oblique cases (36), but retains a partial contrast between
masculine and neuter on the one hand and feminine on the other.
(36) Russian ‘new’ in the singular
M SG N SG F SG
NOM novyj novoe novaja

ACC novyj novogo novoe novuju
GEN novogo novogo novoj
DAT novomu novomu novoj
LOC novom novom novoj
INS novym novym novoj(u)
Gender is neutralized in the plural in Russian, which suggests that there

is a constraint on feature interaction whereby number is a context for loss
of gender. If case is added as a potential context, then the masculine/neuter
syncretism can also be treated as an example of feature interaction (under-
specification of gender in the presence of particular case and number
values). The problem that arises here is that case itself is not fully distin-
guished in the presence of the feminine gender. So the genitive, dative,
locative and, if the archaic form novoju is not used, the instrumental, all
share the syncretic form novoj. This is syncretism, because these cases have
distinct forms for the masculine and neuter genders. If novoj is treated as
underspecified for case (F.SG), and novomu as underspecified for gender
(SG.DAT), there is a problem determining which form to use for the feminine
dative singular, as both specifications are compatible with this. This indi-
cates that, on its own, the most natural formal construct associated with
uninflectedness and neutralization, namely underspecification, may well
be insufficient to deal with the more challenging examples of canonical
syncretism.
2.5 Conclusion
Syncretism exhibits different patterns: the simple and nested syncretism
patterns might suggest a hierarchical structure for the features involved,
contrary syncretism and polarity effects demonstrate that such structures
are insufficient for describing all patterns of syncretism. We have also seen
that these patterns may depend on the domains of comparison. A syncretic
pattern may be associated with a particular morphological class; it may
arise because of the effects of different stems on the same affix (as in the
Macushi example in (8)), it could be the result of different orderings of the
same affix, or it could arise in some classes and not others, because of the
presence or absence of alternation to the stem.
Finally, we have identified a spectrum of morphology–syntax inter-
action involving loss of feature distinctions, which could broadly be
described using the term ‘syncretism’. We will naturally be concentrating
on that part of the spectrum which is left over when we have removed the
obvious explanations for the absence of a morphological distinction.
These obvious explanations are irrelevance for syntax (neutraliza-
tion) and morphological inertness (uninflectedness). They can both be
explained by the constraints on the ordering of features, and both neu-
tralization and uninflectedness, because of what they tell us about the
ordering of features, give us a hint about how to deal with some examples
of syncretism. At other times it is not self-evident how to analyse examples
of syncretism. Under particular models of grammar it is possible to take
some or all instances of syncretism and analyse them in terms of neutral-
ization or uninflectedness, by creating further structure within a feature.
But if one chooses to analyse what we have termed canonical syncretism
in this way, additional assumptions are required which pose problems
and need justification.
3 Cross-linguistic typology
of features
Although the theoretical interest implicit in inflectional syncretism has been

recognized for some time, the range of languages that have contributed to its
study has remained limited. The notion was first applied to Indo-European
languages, and it is from these that the bulk of examples has been drawn.
Languages from other families have been brought to bear in theoretically or
typologically oriented works, but to a lesser extent, e.g. Afro-Asiatic
(Carstairs 1987, Fradkin 1991, Johnston 1997, Noyer 1997), Uralic
(Bátori 1990, McCreight and Chvany 1991, Kiparsky 2001, Blevins 2003),
Tibeto-Burman (Gvozdanović 1991), Altaic (Carstairs 1987), and more
broadly defined, languages of the Caucasus (Hjelmslev 1935–7, Boeder
1976, Carstairs 1987, Carmack 1997, Kibrik 1997, Helmbrecht 1999), of
Australia (Zaliznjak 1973, Goddard 1982, Bavin and Shopen 1987, Heath
1991, Noyer 1997, Evans, Brown and Corbett 2001), of the Americas
(Boeder 1976, Hewson 1989, Heath 1998, Lakämper and Wunderlich
1998, Harbour 2003), of Siberia (Kibrik 1997, Spencer 2000) and New
Guinea (Kibrik 1997, Noyer 1998, Wunderlich 2001b). However, these
works address at most several languages. Cross-linguistic studies relevant
to syncretism are rare: Cysouw (2003) is a comprehensive treatment of the
morphological and lexical expression of pronominal features (person and
number), while Arkadiev (2003) discusses case syncretism.
Thus, there remains a real gap in our empirical understanding of syncret-
ism, which the following chapter should begin to redress. We have gathered
examples from a wide range of languages. The sample is based in part on that
used in the World Atlas of Language Structures (see Baerman and Brown,
2005 a and b) and partly on our own databases (Baerman, Brown and
Corbett 2002a, Baerman 2002b), and has been supplemented by additional
languages chosen to illustrate special points. In xx3.1–3.7 we look at syncret-
ism with respect to major morphosyntactic features (case, person, gender,
number, tense-aspect-mood). The selection of features was determined
partly by their cross-linguistic frequency in inflectional systems – thus, one
37
can imagine syncretism with respect to degrees of honorificity, but it is

neither sufficiently frequent as an inflectional feature, nor are the values
sufficiently comparable across languages, to make any confident cross-
linguistic generalizations. Also relevant was the availability of more than
two distinct values for the feature, which follows from the first part of our
definition of ‘canonical’ syncretism (Chapter 2: x2.4.3.); this eliminates from
consideration features with inherently binary distinctions such as definite-
ness. In x3.7 we consider syncretism which cannot be construed as involving
the simple collapse of feature values. In x3.8 we look at the role of feature
values as conditioning factors for the syncretism of other features.
3.1 Case
3.1.1 Introduction
Inflectional syncretism is often implicitly understood as case syncretism;
e.g. the entry under ‘Syncretism’ in the recent Morphology: an international
handbook on inflection and word-formation (Luraghi 2000) deals exclusively
with case. For its own part, case syncretism is seen mostly in terms of Indo-
European languages. Since Indo-European patterns of case syncretism
have, rightly or wrongly, contributed so much to our understanding of
syncretism as a whole, it makes sense to take them as a starting point.
Case syncretism in Indo-European languages is nearly ubiquitous: it was
already present in Proto-Indo-European and carried through to the
daughter languages that retained case. Among contemporary languages,
the Slavonic family achieves the peak of variety and complexity. In
Russian, one can enumerate at least six regular patterns, as shown in (1),
correlated with word class, inflection class, number and animacy. In (1a),
nominative and accusative are syncretic, in (1b) accusative and genitive, in
(1c) genitive and locative, in (1d) locative and dative, in (1e) genitive,
locative and dative, and in (1f) genitive, locative, dative and instrumental.
(Further combinations of these patterns are possible as well.)
The challenge of accounting for such an array of patterns led to
Jakobson’s (1936 [1971]) study of Russian case semantics, which remains,
directly or indirectly, one of the most influential works on case syncretism.
The six primary cases are broken down into three semantic primitives,
expressing directionality, scope and peripherality (see (2)). The values for
individual cases are construed as composites of these primitives. For exam-
ple, the genitive expresses scope, the locative expresses scope and peripher-
ality, the dative expresses directionality and peripherality, and so on. Case
Cross-linguistic typology of features 39
(1) Patterns of syncretism in Russian nominals
a.‘table’ b.‘student (M)’ c.‘new.PL’ d.‘book’ e.‘mother’ f.‘forty’

0
NOM stol student novye kniga mat sorok
ACC stol studenta novye knigu mat0 sorok
GEN stola studenta novyx knigi materi soroka
LOC stole studente novyx knige materi soroka
DAT stolu studentu novym knige materi soroka
INS stolom studentom novymi knigoj materju soroka
syncretism results from suppressing the expression of the semantic primitives.

For example, nominative/accusative results from suppressing the expression
of directionality, and genitive/locative results from suppressing the expres-
sion of peripherality. On this view, case syncretism is a reflection of the
underlying network of semantic values which make up the case system.
(2) directionality scope peripherality
NOM
ACC +
GEN +
LOC + +
DAT + +
INS +
The particular schema in (2) was revised by Jakobson himself (1958) and
further by Neidle (1988). Although the particular features proposed by
Jakobson have found only limited application beyond Russian – e.g. to
Polish by Schenker (1964) and Slovene by Miller (1990) – the basic notion
remains widely accepted, namely that individual morphosyntactic cases
are the reflection of some underlying semantic network, and that case
syncretism reveals otherwise covert aspects of this network (e.g.
Bierwisch 1967, Wiese 1996, 2003, Halle 1997, Calabrese 1998, Ivanov
2001 and Kiparsky 2001).
Now let us consider non-Indo-European languages. Although not as
ubiquitous as in Indo-European, syncretism is still common. In a con-
trolled sample, described in Appendix 1, out of sixty-four non-Indo-
European languages which mark case inflectionally, thirty evince case
syncretism. However, the sorts of patterns seen in (1) are not equally well
represented. The distribution of attested patterns is clearest when we
describe them in terms of core and peripheral cases. On this basis we
distinguish three types of case syncretism:
* Type 1: syncretism of the core grammatical cases, as in (1a).
* Type 2: syncretism of a core case (typically the accusative or
ergative) with a peripheral case, as in (1b).
* Type 3: syncretism of peripheral cases, as in (1c–f).
In Appendix 1, languages where case syncretism is restricted to type 1 are
distinguished from those where types 2 and 3 are found. Their geographic
distribution is illustrated in Map 1. All three types are well represented in
Indo-European. Non-Indo-European languages, on the other hand,
favour type 1. Out of the seventeen languages in the sample where case
syncretism is restricted to type 1, sixteen are non-Indo-European.
3.1.2 Types of case syncretism

Type 1: syncretism of the core grammatical cases
Syncretism of the cases representing the core grammatical functions of
subject and object is the most common type in all languages.
Conventionally, two primary systems of core case distinction are recog-
nized, nominative accusative and ergative absolutive. In a nomin-
ative accusative system, there is a subject case (nominative) and object
case (accusative). In an ergative absolutive system, there is a case for
transitive subjects (ergative) and a case for intransitive subjects as well as
objects (absolutive).1 Syncretism occurs under both systems.
Nominative/accusative syncretism is represented in nearly every Indo-
European language. In Proto-Indo-European this pattern characterized
neuter nominals, as reflected in Ancient Greek adjectives (3a), and was
also innovated independently in the daughter languages (see Ringe 1995),
as in Latvian, where it is found in feminine plurals (3b).
Nominative/accusative syncretism is also widespread in non-Indo-
European languages, e.g. in plural pronouns in the Algic language Yurok
(4a), and in inanimate nouns in the Dravidian language Telugu (4b).2
1
Additionally, one may note active stative systems, where the treatment of intransitive
subjects is variable.
2
Accusative may be optionally marked by the ending -ni in inanimates, but the preferred
option is not to do so.
Inflectional marking of case is absent
Syncretism of core cases only
Syncretism involving both core and non-core cases
Inflectional marking of case is not syncretic
Map 1. Case syncretism

(3) a. Greek adjective b. Latvian nouns (Mathiassen 1997: 43, 46)
‘wise’ ‘sister’ ‘father’

(neuter) (masculine) (feminine) (masculine)
NOM SG soph-on soph-os NOM PL masas tevi

ACC SG soph-on soph-on ACC PL masas tevus
GEN SG soph-ou soph-ou GEN PL masu tevu
DAT SG soph-oi soph-oi DAT PL masam teviem
LOC PL masas tevos
(4) a. Yurok pronouns b. Telugu (Krishnamurti

(Robins 1958: 20–1) and Gwynn 1985: 88–9)
plural singular inanimate animate

‘we’ ‘I’ ‘houses’ ‘dogs’
NOM nekah nek NOM il::lu kukkalu

ACC nekah nekac ACC il::lu kukkalani
COM neka:noØ neka?aØ GEN il::la kukkala
LOC (?)neya:?ik (?)neya:?ik DAT il::laki kukkalaki
Absolutive/ergative syncretism is rather better represented in non-Indo-

European languages than in Indo-European, though this is presumably
due to the limited scope of ergative case marking in Indo-European
languages (it is an innovation within Indo-Iranian languages, though
here it is widespread). For example, both the non-Indo-European
Dagestanian language Tsakhur (see (5a)) and Indo-European Domaaki
(see (5b)) display absolutive/ergative syncretism in pronouns (in the case of
Domaaki, only 2SG and 3SG pronouns).
Some languages juxtapose nominative accusative and ergative
absolutive systems and represent a special instance of core case syncretism
familiar under the name ‘split ergativity’. For example, in the Pama-
Nyungan language Guugu Yimidhirr, pronouns appear to display a nomi-
native accusative system, and nouns an absolutiveergative system (see
(6)). However, as argued by Zaliznjak (1973), Goddard (1982) and Comrie
(1991), we can treat these as different morphological realizations of a single
underlying system which distinguishes three core cases: accusative for
(5) a. Tsakhur (Kibrik 1999: 130, 201) b. Domaaki (Lorimer 1939: 44, 76)
pronoun noun pronoun noun

‘I’ ‘idiot’ ‘you.SG’ ‘man’
ABS zi baIčar ABS tu mniš«

ERG zi baIčare ERG tu mniš«an
DAT zas baIčaris GEN te mniš«ei
ACC tUs mniš«(ek)
DAT tUšu mniš«šo
objects, ergative for transitive subjects and another for intransitive sub-
jects, which we shall give the non-committal label ‘nominative-absolutive’.
On this interpretation, split ergativity can be treated as the result of two
different patterns of case syncretism: the nominative-absolutive is a non-
autonomous case, syncretic with the ergative in pronouns and with the
accusative in nouns (see (7)).
(6) Guugu Yimidhirr (Haviland 1979: 47–51, 66–7)
‘I’ ‘girl’
NOM ngayu ABS gabiir

ACC nganhi ERG gabiirrngun
(7) Guugu Yimidhirr (alternative representation)
‘I’ ‘girl’
ERG ngayu gabiirrngun

NOM-ABS ngayu gabiir
ACC nganhi gabiir
This interpretation is strengthened by the fact that some languages have

a system similar to Guugu Yimidhirr’s, but with a set of nominals where all
three core case distinctions are morphologically overt. For example, in the
Pama-Nyungan language Wagaya, the nominative-absolutive is syncretic
with the ergative for first and second person pronouns and with the
accusative for demonstratives and nouns, while for the third person pro-
nouns it is distinct (see (8)). Morphologically, the Wagaya paradigms
illustrate an important feature of many languages that display split erga-
tivity: the accusative and ergative are morphologically marked (accusative
-iny and ergative -l:), while the nominative-absolutive is morphologically
unmarked. The three paradigms in (8) can then be described in terms of
differential marking of the accusative and ergative. First and second
person pronouns mark a distinct accusative, while third person pronouns,
demonstratives and nouns mark a distinct ergative.
(8) Wagaya (Breen 1976: 591)
I II III
‘you.PL’ ‘he’ ‘that.M’
ERG ir yuw«l: bul«l:

NOM-ABS ir yuwu bulu
ACC iriny yuwiny bulu
I first and second person pronouns

II third person pronouns
III demonstratives, nouns
The split between accusative and ergative marking is typically correlated

with a hierarchy of properties first proposed for Australian languages
(Silverstein 1976), and now commonly known as the animacy hierarchy.
This name is somewhat misleading, in as much as animacy is only one
component, and it is perhaps better seen as a hierarchy of ‘inherent
referential content’ (Goddard 1982) or ‘individuation’ (Timberlake
1975). At one extreme, specific, animate entities have a distinct accusative
form, and at the other extreme, non-specific, inanimate entities have a
distinct ergative form. A representative example comes from the Pilbara
languages, a group of Pama-Nyungan languages spoken in Western
Australia, where split ergativity shows different cut-off points in the dif-
ferent languages of the group, permitting the extrapolation of the hierar-
chy in (9):
(9) 1SG > 2SG > 1INCL DU > 1INCL PL > 2DU > 2PL > 3 >this > that >
indefinite > animate > meat, vegetable > other inanimate
(Dench 2001: 122)
One can see this as the conjunction of several component hierarchies, for
example literal animacy (animate > inanimate), person (1 > 2 > 3),
number (singular > non-singular), word class (pronouns > demonstratives
> nouns) and definiteness (definite > indefinite). Because the interpreta-
tion of split ergativity seems to be relatively clear, it is tempting to extend it
to the analysis of languages which have syncretism of the core cases but do
not display split ergativity as such, as in examples (6)–(8). If valid, this
would manifest itself in two respects.
* Morphologically, the accusative and ergative should be marked,
while the nominative and absolutive should be the unmarked
default. Core case syncretism should then manifest itself as the
extension of this default from the nominative-absolutive to both
core cases.
* The division between syncretic and non-syncretic paradigms
should be sensitive to the animacy hierarchy. That is, given a
hierarchy such as in (9), all instances of nominative/accusative
syncretism should fall towards the right-hand side, and all
instances of absolutive/ergative syncretism should fall towards
the left-hand side.
It turns out that these expectations are only partly met.
The morphological correlation appears often to be true, with syncretic
nominative/accusative being realized by a default form otherwise used
solely for the nominative, and syncretic absolutive/ergative realized by a
default form otherwise used solely for the absolutive. For example, this is
the case in Russian when one compares student ‘student’ and stol ‘table’,
shown in (1a) above: nominative/accusative stol is morphologically the
bare stem, and this corresponds to the distinct nominative singular form
student, likewise the bare stem. The identification of bare stems as the
morphological default here might seem unobjectionable.3 However,
neither is it automatic. Consider the plural forms of Russian sapog ‘boot’
and soldat ‘soldier’ in (10), which represent a class of Russian nouns whose
genitive plural ending is zero. The syncretic nominative/accusative sapogi
corresponds to the distinct nominative soldaty (the alternation i y is
conditioned by the stem-final consonant), but it is rather the genitive
plural (or accusative/genitive in the case of soldat) which represents the
3
An assumption often made within recent works within the framework of Optimality
Theory, e.g. Kiparsky (2001) and Wunderlich (2004).
bare stem. While we may still want to equate the nominative plural with the
morphological default, we must also admit that there is no independent
evidence for doing so.
(10) Russian masculines with zero genitive plural
inanimate ‘boots’ animate ‘soldiers’
NOM PL sapog-i soldat-y

ACC PL sapog-i soldat
GEN PL sapog soldat
LOC PL sapog-ax soldat-ax
DAT PL sapog-am soldat-am
INS PL sapog-ami soldat-ami
Further, there are examples which show the opposite pattern, where
nominative/accusative syncretism appears to suggest that it is the nomi-
native which is marked with respect to the accusative. This is familiar from
Indo-European, for example in Latvian, where a plural ending of the shape
-Vs is used for the accusative in the masculine declension classes and for the
nominative/accusative in the feminine declension classes (see (3b)).
Sometimes, both patterns occur in a single language. Thus, in the second
declension of Latin, the syncretic neuter ending -um corresponds to the
accusative of masculines, while in some third declension nouns, the syn-
cretic neuter form is the bare stem, corresponding to the nominative of
masculines (see (11)).
(11) Latin
second declension third declension
neuter masculine neuter masculine

‘war’ ‘slave’ ‘sea’ ‘conqueror’
NOM SG bell-um serv-us NOM PL aequor victor

ACC SG bell-um serv-um ACC PL aequor vict
or-em
GEN SG bell-i serv-i GEN PL aequor-is vict
or-is
DAT SG bell-
o serv-
o DAT PL aequor-i vict
or-i
ABL SG bell-
o serv-
o LOC PL aequor-e vict
or-e
The predictions suggested by the animacy hierarchy about the distribu-

tion of syncretic paradigms are met only in certain respects. We shall look
at the three components of the hierarchy which are best represented in our
corpus: (i) literal animacy, (ii) word class and (iii) number.
There is evidence of a positive correlation between literal animacy and
the distinction of nominative and accusative, for example in Telugu (4b)
and Eastern Armenian (Minassian 1980). Russian has a similar correla-
tion, but it is embedded within the system of infectional classes. Thus for
certain classes, nominative and accusative are distinct only for animates
(as in student ‘student(M)’ in (1b)), while for other classes animacy is
irrelevant (as in the singular paradigm of mat0 ‘mother’ in (1e)). The
syncretism displayed by neuter nouns in Indo-European (see the Greek
example in (3a)) may represent a largely morphologized version of the
same correlation, in that these nouns typically have inanimate referents.4
However, there is no direct correlation between semantic animacy and
gender: while the neuter gender is composed of inanimates, many inanimates
are found in the masculine and feminine gender as well, and so may have
distinct nominative and accusative forms.
Curiously, in languages with exclusively ergative absolutive marking,
the analogous correspondence does not seem to obtain. That is, we do not
find languages (at least in our sample) where absolutive/ergative syncret-
ism is directly correlated with inanimacy. This suggests that sensitivity to
inanimacy is contingent on sensitivity to animacy: a language can have
case marking determined by animacy, or both by animacy and inanimacy,
but not by inanimacy alone. A correlation between inanimacy and abso-
lutive/ergative syncretism can occur only in the context of split ergativity.
The word class hierarchy predicts, broadly speaking, an opposition
between pronouns (possibly restricted to first and second person) and
other nominals, with core case syncretism being found in one group but
not the other. This seems to be largely true. For example, Indo-European
languages regularly have nominative/accusative syncretism in nouns but
may lack it in pronouns. Non-Indo-European languages may also show a
similar distribution. For example, in Finnish, where nouns and pronouns
take essentially the same case endings, the accusative is always syncretic
with some other case for nouns, while pronouns have the distinct accusa-
tive ending -t (Fromm 1982). In Nenets, nominative and accusative are
4
Exceptions include Sanskrit mitra ‘friend’ and vr:tra ‘foe’ (Gregory Stump (personal com-
munication), citing Whitney 1889: x1185c).
frequently syncretic for nouns (namely in all dual paradigms, all plurals
with a possessive suffix, and singulars with a first person possessive suffix),
but distinct for all personal pronouns in all numbers (Salminen 1997). In
Yurok, nominative and accusative are syncretic for the 3SG pronoun but
distinct for 1SG and 2SG pronouns (Robins 1958). Correspondingly, abso-
lutive/ergative syncretism affects pronouns more than nouns. Thus in
Tsakhur, pronouns have it and nouns do not (see (5a)). In Yup’ik, 1SG
and 2SG pronouns have relative/absolutive syncretism and the 3SG does not
(Jacobson 1995); the relative case in Eskimoan languages combines the
functions of ergative and genitive. Of course, the correspondence between
word class and core case syncretism is not always perfect. For example, in
the Papuan language Suena (of the Trans-New Guinea phylum), there is a
distinct accusative form for interrogative pronouns and demonstratives,
but not for personal pronouns (Wilson 1974), which would seem to be the
reverse of what the hierarchy predicts. Nevertheless, one can extract a
rough implicational hierarchy: if syncretism of nominative and accusative
is found somewhere among pronouns, it is found somewhere among case-
marked nouns too; and if syncretism of absolutive and ergative is found
somewhere among nouns, it is found somewhere among case-marked
pronouns as well.
The one element of the animacy hierarchy which clearly does not apply
to languages without split ergativity is the number correlation. We should
expect non-singular numbers to favour nominative/accusative syncretism,
and the singular to favour absolutive/ergative. Instead, we find simply that
both patterns of syncretism are more likely in the non-singular. Thus
nominative/accusative syncretism in the dual and plural is quite general
in Indo-European. The same is common in non-Indo-European lan-
guages, as seen in the languages described in the sample in Appendix 1
(see references there): in Finnish (nouns), in the Australian language
Mangarayi (demonstratives), in the Nilo-Saharan language Murle (in
one class of nouns), in the Samoyedic language Nenets (plural nouns
with a possessive suffix, and all dual nouns), in the Uto-Aztecan language
Yaqui and in Yurok (first and second person pronouns). While this
corresponds to the expectation, by the same token, we find absolutive/
ergative in the non-singular alone in Basque (proximate nouns, demon-
stratives and the second person pronoun), Domaaki (nouns; see (5b)) and
Yup’ik (nouns). Thus the correlation seems to be a more general one
between non-singular number and syncretism as such (familiar since
Hjelmslev 1935–7; see also Chapter 2: x2.2.2.), and nothing to do with
the relationship between core case marking and the animacy hierarchy.
Even this correlation does not hold without exception, for example in
Elliot’s (1999) description of Rincón Luiseno (Uto-Aztecan; cited in
Kathol 2002), nominative and accusative are syncretic for nouns in the
singular but not in the plural (12); note that number is not distinguished by
the non-core cases.
(12) Rincón Luiseno ‘sling’: Elliot (1999: 21)
singular plural
NOM pı́ı́vanla-sh pı́ı́vanla-chum

ACC pı́ı́vanla-sh pı́ı́vanla-shmi
ILL pı́ı́vanla-yk
ABL pı́ı́vanla-Nay
LOC pı́ı́vanla-Na
INS pı́ı́vanla-tal
Type 2: syncretism of a core case with a non-core case

In type 2, the core cases are kept distinct from each other, but one of the
core cases is itself syncretic with one of the peripheral cases. Typically it is
the marked core case which is affected, that is the accusative in a nomina-
tive accusative system and the ergative in an ergative absolutive
system. For example, in Finnish, the ending -n, which is used solely for
the genitive in pronouns, is used for both genitive and accusative in
singular nouns (see lukko in 13a). In the Chukotko-Kamchatkan language
Koryak the locative ending found in the first declension (see ‘(a) grand-
mother’ in 13b) is used for both locative and ergative-instrumental in the
second declension (see ‘grandmother’ in 13b). (The second declension
pattern is used for specific human individuals, the first declension pattern
is used for all other nouns; some nouns with human referents may follow
either pattern.)
A possible explanation of the motivation for the type 2 pattern comes
from the Slavonic languages, such as Russian (see Klenin 1983 for a
detailed discussion). In general, it seems to have been used to restore the
distinction between nominative and accusative that had been lost in a
number of nominal paradigms. The earliest records show that nominative
and accusative were syncretic in most nominal paradigms, while the inter-
rogative/relative pronoun had type 2 syncretism of the genitive/accusative.
(13) a. Finnish core cases

(Fromm 1982: 89; b. Koryak core cases
foldout 1) (Žukova 1972: 99)
noun pronoun second declension first declension

‘lock’ ‘I’ ‘grandmother’ ‘(a) grandmother’
NOM lukko minä ABS anja anja

ACC luko-n minu-t ERG-INS anja-na-k anja-ta
GEN luko-n minu-n LOC anja-na-k anja-k
This type 2 pattern was extended to paradigms with nominative/accusative

syncretism, following a word-class hierarchy (pronouns > nouns) and,
within nouns, a hierarchy of animacy or individuation (singular > plural,
male > female, human > animal), where it has remained restricted to
animate arguments.5 Eastern Armenian underwent a similar development,
extending the accusative/dative syncretism of pronouns to animate nouns,
which otherwise have nominative/accusative syncretism.
The diachronic layering of type 1 and type 2 syncretism produced the
synchronic alternation of the two patterns we still see in Russian (see (1a,
b)) and similarly in Eastern Armenian, where nominative/accusative alter-
nates with accusative/dative. The same sort of alternation, involving the
ergative case, is found in Lak, where the ergative is syncretic with the
genitive in nouns and the absolutive in pronouns (see (14); note this is just a
portion of the unusually extensive case paradigm).
(14) Lak (Žirkov 1955: 36, 64–6)
noun pronoun
‘house’ ‘I’
ABS k’atta na
ERG k’atlu-l na
GEN k’atlu-l ttu-l
DAT k’atlu-n ttu-n
SUPERESS k’atlu-j ttu-j
COM k’atlu-šša ttu-šša
5
In pronouns, the extension of genitive/accusative syncretism was extended even to the point
of replacing a distinct accusative ending, as happened with the 3SG feminine pronoun. In
nouns, however, the presence of a distinct accusative ending seems to have blocked the
spread of this pattern.
More unusually, we may find an alternation between two type 2 pat-

terns, as in the Mongolic language Bonan, where the accusative is syncretic
with the genitive in nouns and with the dative-locative (a single case in
Bonan) in pronouns (see (15)). Note that for both languages the case
endings are independent of number.
(15) Bonan (Todaeva 1997: 35)
noun pronoun
‘foliage’ ‘he’
NOM labčoN ndžaN

GEN labčoN-ne ndžaN-ne
ACC labčoN-ne ndžaN-de
DAT-LOC labčoN-de ndžaN-de
ABL labčoN-se ndžaN-se
INS-COM labčoN-Gale ndžaN-Gale
However, there are also examples of type 2 syncretism where this func-
tional explanation – the restoration of core case distinctions – clearly
does not apply. In the Pama-Nyungan language Ngiyambaa, the distinct
ergative-instrumental ending has been replaced by the dative in the speech
of younger speakers. Since the core cases were already distinct in the older
system, this has no obvious functional motivation. Rather, it seems to be
an instance of morphophonological simplification. In the older system, the
dative ending was -gu for all stem types, while the ergative-instrumental
had a series of allomorphs, depending on stem type: -dhu for stems ending
in -y or -yN, -du for stems ending in -n, -u for stems ending in -r or -l, and
-gu (identical to the dative) elsewhere. What the younger speakers appear
to have done is to generalize the default allomorph -gu, eliminating the
morphophonological alternation (Donaldson 1980: 84–5; Austin 1986).
In defining type 2 syncretism we have spoken simply of syncretism with
a peripheral case, but the choice of peripheral case does appear to be
limited. With the accusative, we typically find either the genitive (as in
the Slavonic languages, Finnish6 and the Pama-Nyungan language
6
Presumably owing to a phonological accident: the original accusative *-m and genitive *-n
fell together as a result of the merger of these sounds (Anttila 1972: 103), a process common
to all of Baltic Finnic.
Martuthunira (see Appendix 1)) or the dative (as in Eastern Armenian,

Ngiyambaa, and Phalura) or both (Bonan).7 It is generally held that these
two cases, along with the core cases, together constitute the structural or
syntactic cases, which express basic syntactic functions. These are opposed
to the semantic cases, such as directional cases or the instrumental, which
typically serve as adjuncts. Thus, the choice of case form to stand in for the
marked core case under type 2 syncretism is constrained by this division.
Within that, though, it is difficult to find a motivation for the choice of
genitive versus dative. This is especially striking when one compares
Slavonic with Eastern Armenian. Both have essentially the same inventory
of cases and employ type 2 syncretism to the same end (marking of animate
accusatives), but one uses the genitive and the other the dative.
With the ergative, type 2 syncretism in our sample most typically joins it
with the genitive, as in the Tacanan language Araona, the isolate
Burushaski, Lak and the Tibeto-Burman language Limbu (see Appendix
1). It is likely that this is not a random choice, in that there are languages
which have cases which inherently combine the functions of ergative and
genitive (e.g. the relative case of the Eskimoan languages). Such construc-
tions may have their origin in nominalizations, with the agent expressed by
the genitive. However, although diachronic explanations may be found, it
is unlikely that a direct, synchronic motivation can be demonstrated for
most type 2 patterns.
Type 3: Peripheral case syncretism

Whereas both Indo-European and non-Indo-European languages display
similar behaviour with respect to syncretism involving the core cases (types
1 and 2), they show significant differences when it comes to syncretism of
peripheral cases. Many Indo-European languages display multiple con-
flicting patterns of peripheral case syncretism, such as we have seen in
Russian (1). It is just these sorts of patterns which have led to the idea that
case values can be decomposed into an intersecting network of component
values, as in Jakobson’s (1936 [1971]) analysis shown in (2). However,
evidence for such networks is exceedingly sparse outside of Indo-
European. The sorts of patterns which are found in non-Indo-European
are themselves fairly marginal within Indo-European languages.
7
One also finds languages where both these case functions are combined in a single case, e.g.
the dative-accusative of Pengo.
One pattern involves the collapse of all peripheral cases and is found in
Ingush (16) and in Georgian (see (20) below). In both instances the rich
declensional paradigm is reduced to two forms, one for the absolutive form
and one for everything else. This presupposes type 2 syncretism, in addi-
tion to the collapse of peripheral cases, since the ergative patterns with the
peripheral cases.
(16) Ingush (Nichols 1994: 99)
‘big village’
ABS joqqa jurt

ERG joqqača jurtuo
GEN joqqača jurta
DAT joqqača jurtaa
Similar patterns have arisen occasionally in individual branches of Indo-

European (no comparable pattern was present in Proto-Indo-European),
where it likewise co-occurs with core case syncretism (either type 2, as in
Ingush or Georgian, or type 1). For example, in Russian, the numerals ‘40’,
‘90’ and ‘100’ have only two forms, one for the core cases nominative and
accusative, and one for the peripheral cases (see above, 1f). In the middle
Iranian language Sogdian, the plural paradigm of light stem nouns shows
essentially the same pattern, collapsing the nominative and accusative on the
one hand, and the genitive, locative and ablative on the other (Sims-Williams
1982). In German, weak adjectives, exemplified by böse ‘bad’, likewise dis-
play only two forms: nominative singular versus a form which serves for the
genitive/dative singular as well as all plural cases (17). The behaviour of the
accusative varies with the gender of the modified noun. If neuter or feminine,
(17) German
‘the bad child’ (neuter) ‘the bad man’ (masculine)
NOM das böse Kind der böse Mann

ACC das böse Kind den bösen Mann
GEN des bösen Kindes des bösen Mannes
DAT dem bösen Kind(e) dem bösen Mann(e)
it follows the type 1 pattern and so is identical to the nominative. If mascu-

line, it follows the type 2 pattern and is identical to the peripheral form.
The second type of peripheral case syncretism which is found relatively
often in non-Indo-European languages involves syncretism of a direc-
tional case (e.g. illative or allative) with the dative or locative, which may
well have a semantic motivation. Thus, in the Finnic language Erzja
Mordvin, the dative ending is used for the illative in singular definite
nouns (and optionally in plural definite nouns).
(18) Erzja Mordvin (Feoktistov 1966: 180)
‘the house’ ‘(a) house’
NOM kudos’ kudo

GEN kudont’ kudon’
DAT kudonten’ kudonen’
ILL kudonten’ kudos
ABL kudodont’ kudodo
INESS kudosont’ kudoso
EL kudostont’ kudosto
PROL kudovant’ kudova
CMP kudoškant’ kudoška
ABESS kudovtomont’ kudovtomo
In the Pama-Nyungan language Diyari, the allative case is non-autono-

mous: with singular nouns and male personal names it is syncretic with
the dative, and otherwise it is syncretic with the locative; cf. ‘boy’ versus
‘people’ in (19).
(19) Diyari (Austin 1981: 51)
‘boy’ ‘people’
ERG kankuli kan:awar:ali

ABS kanku kan:awar:a
ACC kanku kan:awar:ana
;
LOC kankun5i kan:awar:aNu

ALL kankuya kan:awar:aNu
DAT kankuya kan:awar:an:i
ABL kankundu kan:awar:aNundu
There is some cross-linguistic evidence that the use of the dative or

locative case form in a directional function may have some semantic
motivation. For example, in Turkish, the dative is used for goal of motion
as well as for indirect objects (cf. Blake 1994: 145), and the alternation of
location and direction is familiar from Indo-European prepositions such
as Latin in ‘in, to’ or in Russian v ‘in, to’, where the different functions are
correlated with different patterns of case government. Nevertheless, the
examples of peripheral case syncretism from non-Indo-European lan-
guages have no direct parallel in Indo-European languages, as the latter
typically lack a distinct directional case.
Probably the closest thing to an Indo-European style system of multiple
syncretic patterns is found in Georgian, which has two patterns. The first is
found in the so-called Old Georgian plural paradigm, inherited from Old
Georgian but little used in the contemporary language. (The normal system
of case marking in the plural is identical to the singular.) It involves the
collapse of all peripheral cases, as seen in buzni ‘flies’ in (20), though note
that the vocative does not pattern with the peripheral cases. This paradigm is
also defective, in that it is not used for the instrumental or adverbial case.
The second syncretic pattern is found with attributive adjectives and
involves syncretism of absolutive/genitive/instrumental and dative/adver-
bial. This combination of cases does not correspond to any pattern found in
an Indo-European language (or any other non-Indo-European language).
(20) Georgian (Aronson 1991: 219, 232–3)
‘flies’ ‘old book’
ABS buzni dzveli c’igni

INS — dzveli c’igni-t
GEN buz-t dzveli c’igni-s
ERG buz-t dzvelma c’ignma
DAT buz-t dzvel c’ign-s
ADV — dzvel c’ign-ad
VOC buz-o dzvelo c’igno
Morphologically, the syncretism in these paradigms reflects a sort of agglu-

tinative structure quite unlike what is found in Indo-European. Both
paradigms involve partial morphological truncation. The syncretic form
of the Old Georgian plural lacks case endings altogether, substituting the
(typically verbal) plural marker -t. The syncretic forms of the attributive
adjectives result from deletion of the final elements -s ‘GEN’, -t ‘INS’, -s ‘DAT’
and -ad ‘ADV’ (Aronson 1991: 236; Carstairs-McCarthy 1994: 778–9), leav-
ing a consonant-final base (dzvel) or a vowel-final base (dzveli).
3.1.3 Conclusion
In Indo-European languages, case syncretism displays a wide variety of
patterns, as illustrated in (1), which has led to the proposal that there is a
network of semantic primitives that underlie case: individual cases are
composites of these primitives, and case syncretism within a language repre-
sents their different combinatorial possibilities. Cross-linguistically, how-
ever, most instances of case syncretism can be described in terms of the
behaviour of the marked core cases (accusative and ergative), which are
either identical to the unmarked core case (type 1), or to one of the peripheral
cases (type 2). Less frequently, peripheral case distinctions are effaced. These
patterns show a sensitivity to basic syntactic relations but are not striking
evidence for a network of semantic relations. That said, the semantic value of
cases probably does play a direct role in some instances of case syncretism (as
in (18) or (19)), but sporadically enough that no extensive, cross-linguistically
viable system of semantic primitives can be derived from them.
We have speculated that the motivation for types 1 and 2 may lie in the
differential marking of objects or agents, such as is found in languages with
split ergativity, where accusative marking may be omitted low in the
animacy hierarchy, and ergative marking omitted high in the hierarchy.
Overall, there is a correlation, but it is quite inexact. For example, in the
variety of Luiseño described by Malécot (1963) (21), if one were to look
solely at the contrast in accusative case marking between the animate noun
ja?ásˇ and the inanimate noun sˇamUt, one might say simply say that
animates mark the accusative and inanimates do not. But the agreeing
adjective alaxwIsˇ marks the accusative in both instances. Thus, there is
differential object marking here, though it is correlated with semantic
properties and still requires a morphological account.
(21) Luiseño (Malécot 1963: 288)
animate: man + bad = ‘bad man’ inanimate: grass + bad = ‘bad grass’
NOM ja?áš alaxwIš šamUt alaxwIš

ACC ja?átšI alaxwItšI šamUt alaxwItšI
As often as not, patterns of case syncretism are thoroughly morpholo-

gized and yield only dim echoes of this putative original motivation. At any
one synchronic stage multiple versions of the same pattern may be found.
For example, in Russian, nominative/accusative has at least three different
sources, depending on nominal class,8 and further, may or may not be
sensitive to animacy, also dependent on nominal class. We interpret the
cross-linguistically dominant patterns primarily as the unmotivated, mor-
phologized residue of diachronically recurring, motivated patterns.
3.2 Person
3.2.1 Introduction
More so than any other feature, the values of person are relatively constant
cross-linguistically. Thus, the values first person (possibly distinguishing
inclusive and exclusive), second person and third person are largely suffi-
cient to characterize any language, with further values involving distinc-
tions subordinate to these (e.g. honorific versus familiar, proximate versus
obviative, and same subject versus different subject). This constrained
range of values allows us to make a fairly direct comparison between
syncretic patterns in different languages. This is of particular interest
because it can be used to assess claims that patterns of syncretism are
correlated with underlying morphosyntactic or semantic relationships,
since these relationships are presumably common across languages.
The most common domain for person marking is as the subject marker
on verbs, and it is on this that we will focus in this section. (The interaction
with object marking involves a distinct set of conditions and is treated
in x3.3.) Syncretism of subject person is common: in the language sample
from the World Atlas of Language Structures, illustrated in Appendix 2,
out of 141 languages with inflectional marking of subject person, 61 dis-
play syncretism in some portion of the verbal paradigm (see Map 2). In this
section we shall ask two questions: (i) do patterns of person syncretism
reflect morphosyntactic or semantic natural classes, and (ii) do directional
effects reflect markedness relationships between person values?
8
Thus, looking at nouns alone, nominative/accusative was inherited from Proto-Indo-
European in neuters (fourth declension), innovated in Common Slavonic as a result of
sound change in singular o-stem masculines (first declension), singular i-stems (third
declension), and the plural of all but o-stem masculines, and then further innovated in the
Old Russian period through morphological change in the plural of o-stem masculines.
Inflectional marking of person is absent
Inflectional marking of person is syncretic
Inflectional marking of person is not syncretic
Map 2. Syncretism in the marking of subject person on verbs

3.2.2 Patterns of syncretism

It will be useful to distinguish two contexts. First, a given set of values may
be syncretic for all paradigms, which we will term ‘complete’ syncretism.
Second, a set of values may be syncretic only in some paradigms, restricted,
for example, to a particular tense or conjugation class. This we term
‘partial’ syncretism. Note that nothing further is implied by this distinction;
we do not assume that two different interpretations are warranted. Nor is
the distinction always meaningful; for example, if a language has only one
set of person markers, it achieves complete syncretism vacuously.
Nevertheless, it is a useful criterion, allowing some distinct tendencies to
emerge.
Where syncretism is complete, there is a sharp contrast between the
behaviour of person in the singular versus non-singular. Complete syncret-
ism of person solely in the singular is uncommon. In the sample described
in Baerman (2002b; sources listed there), out of twenty-nine examples of
complete person syncretism in the sample (from twenty-seven languages),
only six are restricted to the singular (see (22)). Of these, two come from
languages where person is not distinguished in the plural, so that it is not so
much the case that the syncretic pattern is restricted to the singular, but
that person marking itself is restricted to the singular. In the non-singular,
1/2 and 2/3 both occur in roughly equal measures, while 1/3 is less com-
mon. A similar distribution is found where number is irrelevant, with
examples of 1/2 and of 2/3 predominating.
(22) Examples of complete syncretism, by language
singular non-singular number-neutral
1/3 Koiari*, Zoque Aleut, German, Hindi
2/3 Atakapa, Hindi, Amele, Kapau, Kewa, Chitimacha, Guambiano,

Nivkh*, Nubian Kobon, Slovene Kiwai, Wambon
1/2 Burarra, Dogon, Hunzib, Ingush, Nez Perce,

Manchad, Nubian, Sango, Waskia
Prinmi, Tetun
*
Person distinguished in singular only.
Thus, cross-linguistically, there seems to be a preference for syncretism

of first with second person, and of second person with third, in both cases
restricted to non-singular or number-neutral contexts. Interestingly, this
corresponds well with the sorts of patterns found in free pronouns, as

shown in (23).
(23) Compound person values in free pronouns
singular non-singular number-neutral
1/3 Dakar Wolof
2/3 Amele* , Kalam* , Kawesqar

Kamoro, Kobon, Korafe* ,
Meyah, Mansim,
Nganasan, Nez Perce* ,
Sango* , Warekena* ,
Wolof
1/2 Awa* , Fongbe, Slave* , Winnebago

Yimas
*Examples from Cysouw (2003). Other sources: Dakar Wolof from Nussbaum,
Gage and Varre (1970), Fongbe from Lefebvre and Brousseau (2002), Kamoro
from Voorhoeve (1975), Kawesqar in Clairis (1985), Mansim and Meyah from
Reesink (2002), Winnebago from Lipkind (1945), Yimas from Foley (1991).
As with inflectional marking, nearly all the examples of combined

person values in free pronouns involve 1/2 and 2/3 in the non-singular;
none involve syncretism of singular person values alone. (Note that the
evidence from number-neutral pronouns is exceedingly thin: we are aware
of only two examples, the 1/2 emphatic personal pronoun of the Souian
language Winnebago (Lipkind 1945: 29) and the 2/3 pronoun of the
Patagonian language Kawesqar (Clairis 1985: 465); the latter is only
partial, in that there are distinct possessive forms for these two persons.)
The preference for certain patterns across languages of different families
and with different morphological structure, and the parallel between these
patterns and the evidence of free pronouns, suggests a connection. Since it
is reasonable to suppose that free pronouns, as discrete lexical items, are
semantically coherent, this connection may lie in the morphosyntactic or
semantic structure of person values. However, we know of no model of
person features that will account precisely for both the rarity of 1/3
syncretism and the rarity of syncretism which is restricted to the singular.
For example, Harley and Ritter (2002), in a model designed to account for
the person values of pronominal forms, predict that 1/2 pronouns should
be found, but not 2/3.9 However, as we have seen, not only are 2/3 pronouns
found, they appear to be more common than 1/2 pronouns. They further
predict that if a language does display a 2/3 or 1/2 pronoun, this is a result
of accidental homophony and will be disambiguated by verb agreement
(p. 513, note 42). This prediction is contradicted by Amele, Kobon (dual
only) and Nez Perce. Even if this prediction were true, it is curious that
accidental homophony would be more frequent than systematic syncretism.
Further, their model predicts that number values should have no influence
on patterns of person syncretism (because number is construed as a separate
node in their model of feature structure and does not interact with person).
If we choose to ascribe a semantic rationale to these patterns, it is
probably significant that non-singular numbers favour syncretism, since
this is precisely the context where there may be referential overlap, and
hence ambiguity. Thus, in a language without an inclusive exclusive
distinction, first person plural may or may not include the addressee, so
blurring the distinction between first and second person. Likewise, second
person plural may or may not be construed as including some non-
addressees, so blurring the distinction between second and third person.
Nevertheless, there are reasons to be cautious about overemphasizing the
semantic naturalness of such syncretic combinations, which are especially
apparent when one considers the behaviour of the first person inclusive.
Since semantically it overlaps with first and second person, we should
expect syncretism with those two persons. While this does occur, instances
of 1INCL/1EXCL syncretism are far more frequent than 1INCL/2, which is not
readily accounted for if semantic overlap is deemed to license the syncret-
ism; signficantly, 1INCl/2 syncretism is no more frequent than 1INCL/3,
which cannot be accounted for by the notion of semantic overlap (Cysouw
forthcoming). An example of such an ‘unnatural’ pattern comes from the
Austronesian language Kwamera (24), where the first inclusive is syncretic
with the third person in the dual, precisely the pattern one would expect
not to find.10
9
Harley and Ritter (2002) model person features as a binary branching tree structure. The
base node (‘Referring expression’) is interpreted by default as third person, the node
dependent on that (‘Participant’) is interpreted by default as first person, and the node in
turn dependent on that (‘Addressee’) is interpreted as second person. Thus, the only
syncretisms that can be expressed by this model are 1/2 (through underspecification for
the Addressee node) and 1/2/3 (through complete underspecification).
10
Note though that Kwamera has an impersonal marker k-, and a morphological relation-
ship between impersonals and the first plural is known from other languages. Thus, in
some Athapaskan languages, the unspecified person marker is used for the first plural
(24) Kwamera verbal prefixes (Lindstrom and Lynch 1994: 10)
1 iak- iak-rou- iak-ha-

2 ik- ik-rou- ik-ha-
3 r- k-rou- ha-
1INCL — k-rou- sa-ha-
Conversely, in the Tibeto-Burman language Kinnauri (25), the first person

inclusive has a distinct form (dual and plural), while the first person exclu-
sive and second person non-singular are syncretic (1INCL PL is optionally
syncretic with this form).
(25) Kinnauri verbal suffixes (Sharma 1988: 132–3)11
1INCL – -ič -e -č -e

1 -k -č -č
2 -n -ñ (HON) -č -č
3 -Ø -t -d -š (HON) -š -š
Since 1EXCL and second person are mutually exclusive, there can be no
question of a semantic or functional overlap between the values of the
syncretic form (see Noyer 1997 for discussion of a similar pattern in
Mam).12
Partial syncretism does not lend itself to the same generalizations as
complete syncretism. Most strikingly, instances of syncretism affecting
solely singular person values predominate. Clearly, no strong case can be
made for the sort of semantic motivation discussed above. In part, the
cause must lie in the morphology of the forms. One banal observation is
(Rice 2000: 201), in Ngiti, the third person indefinite form is used for first person inclusive
when preceded by a free pronoun (Kutsch Lojenga 1994), in colloquial Finnish, imperso-
nal forms may substitute for first plural (Karlson 1999: x97), and, of course, the use of the
French impersonal on for first plural is well known.
11
Second and third person singular have honorific forms (in the case of third person, the
plural is used). Sharma (1988) does not explain what conditions the other allomorphs.
12
Other examples of 1EXCL/2 syncretism alongside a distinct 1INCL come from Udihe (see (37)
below), Burarra (see (36) below), Aymara, Nunggubuyu and Tiwi (Baerman 2002b).
that many of the instances of partial syncretism involve affixes which are
simply phonologically less robust than the non-syncretic affixes, for exam-
ple they are shorter, or vocalic as opposed to consonantal, so making them
more susceptible to phonological decay or merger. Of course, the incom-
pleteness of a syncretic pattern does not preclude its having a semantic
motivation, any more than completeness is a proof of semantic motiva-
tion. The lack of clear patterning under partial syncretism is at best an
indication of multiple causation.
3.2.3 Directional effects

Directional effects indicate a morphologically asymmetric relationship
between two values. It has often been suggested that this asymmetry
obtains at the level of content too. For example, the default status of the
third person has long been recognized; e.g. Watkins’ Law (Watkins 1962)
holds that 3SG serves as the base for analogical remodelling of conjuga-
tional paradigms. More recently, under Impoverishment Theory (Noyer
1998), it is held that third person is unmarked with respect to other
persons, and that directional effects result from the replacement of first
or second person forms by third person forms. Thus, under this frame-
work, Bobaljik (2002) argues that the syncretic 2SG/3SG aorist and imper-
fect forms in Macedonian (26) result from the replacement of second
person by third person. (The example is taken from Stump (1993) and is
identified as being directional because (i) the person marker is -Ø, char-
acteristic of the third person in the present, and (ii) it lacks the element -v-,
whose absence is characteristic of the third person (singular and plural) in
both the aorist and imperfect.)
(26) Macedonian
present aorist imperfect
1SG -am -v -ev

2SG -š -Ø -eše
3SG -Ø -Ø -eše
1PL -me -vme -evme
2PL -te -vte -evte
3PL -at -a -ea
Under such an approach, it is assumed that feature values are arranged

in a hierarchy of markedness, and that synchronic directional effects are
prima facie evidence of this hierarchy. Thus, where first or second

person are syncretic with third, the resulting syncretic form should
somehow be identifiable as third person rather than first or second
(provided directional effects are observable at all). In the case of 1/2
syncretism, opinions vary about the markedness hierarchy that obtains;
for example, Noyer (1997: 114) and Helmbrecht (2003: 6) argue that first
person is marked, Harley and Ritter (2002: 486) argue the opposite.
However, when we look at a larger set of examples, we find little
evidence that directional effects reflect any consistent hierarchy. Below
we review what we consider to be the reasonably convincing examples of
directionality that we have come across; the corpus could be expanded
or shrunk depending on what one considers to be a convincing example
of directionality. We have limited ourselves to examples where the
syncretism is realized by an overt morphological marker, rather than
by a bare stem. In so doing we eliminate instances which would be
transparently interpretable as underspecification under practically any
formal model.
First person / third person

There are not many convincing examples of directionality involving these
values. The Papuan language Koiari of the Trans-New Guinea phylum (27)
seems to confirm the expectation that the third person form should prevail,
in as much as the 1SG obligatory mood may optionally be identical to the 3SG
(elsewhere, first and third person are always identical). Livonian (see (42)
below) presents a diachronic example where the third person form prevailed.
(27) Koiari (Dutton 2003: 345, 351)
obligatory mood
option option 2 imperfect perfect
1SG -ahina -ahima -ma -nu

2SG -ihama -ihama -a -nua
3SG
PL
-ahima
-ihava
! -ahima
-ihava
-ma
-a
-nu
-nua
But there are also examples where the first person form seems to prevail. In
the Nilo-Saharan language Murle (28), first person (inclusive) and third
person are syncretic (in both numbers) in the subjunctive but not the perfect.
In the perfect, the first person is characterized by prefixed k-, while the third
person has no prefix. The syncretic form in the subjunctive has a prefixed k-,
just as the distinct first person of the perfect. Thus, it appears as if an overtly
first person element is serving for third person as well.13
(28) Murle (Arensen 1971: 83)
perfect subjunctive
1SG k- -a ! k-
2SG -u Ø
3SG (-un) k-
1INCL PL
1PL
k- -it
k- -da
! k--it
k--da
2PL -tu -it
3PL -it k--it
Second person / third person

Here there are rather more clear examples of directionality than with 1/3
syncretism, though the results are mixed. In some examples the third
person form prevails, e.g. in Macedonian, as discussed in (26). Likewise,
in the Nilo-Saharan language Nobiin (29), the 2SG appears to take the form
of the 3SG in both tenses (present and past). In Dutch (30), the syncretic
pattern is correlated with the position of the subject pronoun. When
subject pronouns are preposed, 2SG takes the ending -t, identical to that
of the 3SG. When the subject pronoun is postposed (as occurs in questions
and in subordinate clauses), only 3SG takes -t; thus jij kom-t kom je
‘you’re coming are you coming?’ versus zij kom-t kom-t zij ‘she’s
coming is she coming?’ (see Ackema and Neeleman (2003) for a recent
discussion of the conditions).15
13
Note though that the related language Mursi (also a member of the Surmic branch of Nilo-
Saharan) has a similar pattern, in which the 3SG may have a stem alternant distinct from the
others (Turton 1981: 344).
14
The suffix -un is found optionally in the perfect.
15
This pattern has some exceptions, as there is one verb which displays a stem alternation
which disambiguates the 2SG and 3SG (jij heb-t ‘you have’ versus zij heef-t ‘she has’), and
another verb where 2SG has -t even though 3SG does not ( jij ben-t ‘you are’ versus zij is
‘she is’).
(29) Nobiin present past

(Werner 1987: 147–9, 153–4) (30) Dutch ‘come’
verb + pronoun +
indicative interrogative pronoun verb
1SG -ir -is -re 1SG -Ø -Ø

2SG -nam -onam -i -o 2SG -Ø -t
3SG -i -o ! -i -o 3SG -t ! -t
1PL -ir -is -ro -so 1PL -en -en
2PL -rokom -sokom -ro -so 2PL -en -en
3PL -inna -sa -inna -sa 3PL -en -en
Although there are examples where the third person form prevails, there
are even more which favour second person. For example, in Callahuaya, a
Quechua-based language, the original second person marker -nki is some-
times found with third person; contrariwise, third person -n is not used for
second person (Muysken 1997: 437–8). In the Bantu language Kongo (31),
indicative and subjunctive prefixes are distinguished in the 3SG only,
allowing for differences in syllabicity.16 The syncretic 3SG prefix o- of
the indicative can be identified with the distinct 2SG w- of the subjunctive.
In Old Icelandic (32), the syncretic 2SG/3SG ending -er of the present
(31) Kongo (Carter and Makoondekwa (32) Old Icelandic weak verb (Noreen
1979: 6–11, 19–21) 1923: 353–4)
subjunctive, present
subjunctive indicative preterite indicative
1SG y- i- 1SG -a -a
2SG
3SG
w-
k-
! o-
o-
2SG
3SG
-er
-e
! -ar
-ar
1PL tw- tu- 1PL -em -om
2PL nw- nu- 2PL -eþ -eþ
3PL b- be- 3PL -e -a
16
The alternation between syllabic indicative prefixes and asyllabic subjunctive prefixes is
phonologically regular: indicative prefixes attach to the (consonant-initial) verb stem,
while the subjunctive prefixes precede the subjunctive marker a-, which is prefixed to the
verb stem.
indicative is the same as the distinct 2SG ending found in the other tense-
mood paradigms. (The indicative subjunctive contrast is marked by a
vowel alternation in the ending for all but 1SG and 2PL.)
The Papuan language Dani (Trans-New Guinea phylum) (33) shows a
syncretic 2PL/3 ending -ep in the hypothetical mood which matches the
distinct 2PL ending -ip of the past. (The hypothetical mood is also char-
acterized by a lowering of the vowel of the ending.) In Carib (Kalihna)
(34), the interrogative form of the copula distinguishes 2PL and 3PL, while
elsewhere they are syncretic, displaying the form of the 2PL. The Papuan
language Suena (Trans-New Guinea phylum) presents a particularly strik-
ing example (35): the syncretic 2DU/3DU and 2PL/3PL of the remote tense
have the same element -w- that is found in the 2DU and 2PL endings found in
(33) Dani (Bromley 1981: 192) (34) Carib copula (Hoff 1968: 212)
past hypothetical interrogative present
1SG -i -e 1SG waN wa

2SG -in -en 2SG maN ma:na
3SG -e -ep 3SG naN maN, na
1PL -u -o 1PL kı̈ta:toN kı̈ta:toN
2PL
3PL
-ip
-a
! -ep
-ep
2PL
3PL
mandoN
nandoN
! mandoN
mandoN
the other tenses. (The forms in the first column are used with the future,
present, today’s past, yesterday’s past and past tenses. The symbol -V-
denotes the variable mood marker.)
(35) Suena (Wilson 1974: 59)
default remote
1SG -n-V -n-V

2SG -s-V -s-V
3SG -i-V -nu-V
1INCL DU -n-V-ge -n-V-ge
1EXCL DU -n-V-to -n-V-to
2DU
3DU
-w-V-to
-r-V-to
! -w-V-to
-w-V-to
1INCL PL -n-V-kai -n-V-kai
1EXCL PL -n-V-kare -n-V-kare
2PL
3PL
-w-V
-r-V
! -w-V
-w-V
First person / second person

In the light of the failure of 1/3 and 2/3 syncretism to reflect any consistent
morphological hierarchy between these values, we should not be surprised
to find the evidence of 1/2 syncretism to be equally inconclusive. In some
examples the second person form prevails. Thus, in the non-Pama-
Nyungan Australian language Burarra (36), 1/2 augmented (plural) and
unit-augmented (dual) are marked by the same prefix nyi- which serves
for second person in the singular.
(36) Burarra (Glasgow 1984, cited in Cysouw 2003)
minimal augmented unit augmented
1INCL arr- ngu-burr- a-rri-

1(EXCL) ngu- nyi-burr- nyi-rri-
2
3
nyi-
(a-)
! nyi-burr-
a-burr-
nyi-rri-
(a)birri-
In the Tungusic language Udihe (37), first and second person are syncretic in
both the singular and the plural in various paradigms, and the form corres-
ponds to the distinct second person form as found in other paradigms.17
(37) Udihe (Nikolaeva and Tolskaya 2001: 212–13)
future,
permis- converbs,
past, sive, present
past subjunct- perfect, and future
participle present ive conditional participles
1SG -mi -mi -mi -i -i

2SG
3SG
-i
-ni
-i
-ini, -ili
-i
Ø
! -i
Ø
-i
-ni
1INCL PL -fi -fi -fi -ti -fi
1PL -mu -u -u -u -u
2PL
3PL
-u
-ti
! -u
-iti, -du-
-u
-du-
-u
-du-
-u
-ti
17
Etymologically, the second person had the form -sV, with the -s- lenited to -h- in some
varieties of Udihe (Sunik 1997: 238) and to Ø in others.
In the Omotic language Shinassha (38), the 1PL subjunctive appears to

be based on the 2PL form, in both the prefix conjugation and the suffix
conjugation.
(38) Shinassha (Lamberti 1993: 149–53, 163–5)
(prefix conjugation suffix conjugation)
default subjunctive
1SG tı̀- -è ni- -ee

2SG nı́- -ı́ ni- -ii
3SG M bı́- -é ni- -ee
3SG F bı̀- -à bi- -aane
1PL nò- -ò it- -«te
2PL
3PL
ı́t- -«t
bó- -nóó
! it- -«te
bo- -noo
On the other hand, there are languages where first person form seems to
prevail over second. Thus, Nobiin shows this pattern in the plural inter-
rogative; in (39) we reproduce the paradigm shown above in (29); note that
there, the second person is involved in a different directional effect in the
singular. In literary Kannada (40), 1SG is distinguished from 2SG by the
addition of the element -nu in the future. In the past, the -nu element is
extended to 2SG.
(40) Literary Kannada (Sridhar

(39) Nobiin 1989: 221–2)
(present past)
indicative interrogative future past
1SG
2SG
-ir -is
-nam -onam
-re
-i -o
1SG
2SG
-enu
-e
! -enu
-enu
3SG -i -o -i -o 3SG M -anu -anu
1PL
2PL
-ir -is
-rokom -sokom
! -ro -so
-ro -so
3SG F
3SG N
-alu
-uu_
-alu
_
-itu
3PL -inna -sa -inna -sa
First / second / third person

In the future tense in Gujarati (41), 2SG has no distinct form: it is either
identical to the 1SG or to the third person, the two forms being in free
variation (Cardona 1964: 142). Note that Dutch displays a similar alter-
nation, at least superficially (see above, 30); we do not treat this example as
parallel to Gujarati because the 1SG/2SG pattern is transparently a default
form.
(41) Gujarati future (Cardona and Suthar 2003: 682, 684)
option 1 option 2
1SG -iš -iš

2SG -iš -še
3
1PL
-še
-š(i)ũ
! -še
-š(i)ũ
2PL -šo -šo
3.2.4 Diachrony
Some compelling instances of directionality can be found in diachrony,
where the expected form for one value is replaced by the form for another
value. A familiar example, adduced in KuryØowicz’s famous article on
analogy (1949 [1960]), involves Old Icelandic, cited above in (32). These
paradigms are interpreted as the result of the extension of the 2SG ending to
the 3SG in place of the expected *-þ (Haugen 1982: 129). The syncretic
pattern itself is attributed to analogy with verbs in -l and -n, where it was
the result of a general phonological development (syncope of the theme
vowel and assimilation of the ending into the stem-final consonant;
KuryØowicz 1949 [1960: 81]). We can ask the same questions about this
diachronic scenario as we have asked above about synchronic patterns:
(i) where does the syncretic pattern come from?, and (ii) why does one form
prevail over the other?18 Let us look at a few examples with these two
questions in mind.
18
Of course, we can also ask ‘Why was the syncretic pattern extended?’ This is not a question
of syncretism per se, so much as one of a general theory of analogy. In all of the examples
known to us, an obvious functional explanation (such as that offered for the genitive/
accusative in Slavonic; see x3.1.2) is lacking. Since avoidance of homophony is invoked at
least as often as extension of homophony in order to account for unusual morphological
Some examples involve substituting a third person form for the original
form, which would follow from the assumption that third person serves as a
default of some sort. For example, in Livonian (42), the 3SG present tense
ending -b is found in place of the expected 1SG ending *-n ! -Ø (Viitso
1998: 112). Compare the paradigm from the closely related Estonian, where
the original 1SG ending is found.19 Note that in monosyllabic stems in West
Livonian dialects, vacillation was recorded between the original 1SG form and
the innovative one, thus the verb ‘to be’ has the singular forms uo or uob ‘1SG’,
uod ‘2SG’, uob ‘3SG’ (Kettunen 1938: lx). Kettunen (1938: lx–lxii) attributes this
to analogy with the preterite paradigm, where 1SG and 3SG fell together as the
result of regular sound change. Thus, as with Old Icelandic, the syncretic
pattern was already established in the language by regular sound change.
(Note that the 2PL/3PL pattern of the preterite was not extended.)
(42)
present preterite
Livonian Estonian Livonian Estonian
1SG lugub loen lugiz lugesin

2SG lugud loed lugist lugesid
3SG lugub loeb lugiz luges
1PL lu’gg«m loeme lugizm« lugesime
2PL lu’gg«t loete lugist(«) lugesite
3PL lu’gg«b«d loevad lugist(«) lugesid
German likewise shows what appears to be the extension of third person

forms. In Middle High German (MHG), all persons were distinct in the
present tense plural; by the Early New High German (ENHG) period, four
different syncretic patterns were found, varying according to dialect (43).
In type I, the third person form has been extended to second person. In
type II, it has been extended to all three persons. In type III (ultimately
established as the literary norm), 1PL and 3PL are syncretic. Though the form
looks like the original 1PL, the source is more likely to have been the -n found
in the 3PL preterite. In IV, this -n is extended to all plural persons.
innovations (cf. Bentley and Eythórsson 2001), we prefer not to address this question. Note
that KuryØowicz (1949) does not address it either. Although this example appears in an
article on analogy, it is used to illustrate quite a different point; the analogical extension of
the syncretic pattern is itself taken for granted.
19
Though note that the expected reflex of final -n should be -Ø in Estonian as well. Its
retention in 1SG forms is anomalous (Anttila 1972: 79).
(43) Present tense endings in Middle and Early New High German
(Wegera 2000: 1546; Grosse 2000: 1333)
ENHG
MHG I II III IV
1PL -n -n -nt -n -n
2PL -t -nt -nt -t -n
3PL -nt -nt -nt -n -n
The third person also prevails in the Kumta dialect of Kannada (44), where,
in place of the historically expected 1PL ending -vV (found in other dialects, as
well as the literary language), the 3PL ending -ru is found. As in the German
examples above (43), this same ending may replace the 2PL ending -ri as well.
(44) Kannada ‘kept’ (Upadhyaya 1976: 130–2)
Bellary Kumta
1PL ittive ittru

2PL it_t_ri it_t_ru ittri
3PL it_t_ru (M/F) it_t_ru __
it_t_uvu (N) __
__
A similar development has been reconstructed for the plural passive of

Gothic: 1PL and 3PL fell together by regular sound change, and the 2PL was
analogically altered to match them, leading to the attested 1PL/2PL/3PL
ending -anda (Szemerényi 1989: 255).
A particularly striking example of the extension of third person forms
comes from the Oceanic language Anejom (45). In the nineteenth century,
the auxiliary had distinct forms for first person (inclusive and exclusive),
second and third in the dual, trial and plural. Between the nineteenth and
twentieth centuries the language underwent catastrophic change (due in
part to population loss). The result is a system with considerable variation,
if not to say confusion. Lynch (2000) hypothesizes that it is moving in the
direction of generalizing the third person plural for all non-singular
persons.
(45) Anejom auxiliary (Lynch 2000: 91–5)
(aorist, typical for other TAM paradigms)
19th century > 20th century
1SG ek ek
2SG na na
3SG et et
1INCL DU intau
1DU ecrau
2DU ekau
3DU erau
1INCL TRI intaj
1TRI ektaj, ektij era
2TRI ahtaj
3TRI ehtaj
1INCL PL inta
1PL ecra
2PL eka
3PL era !
However, not all such examples favour the third person. Thus, Romani
shows evidence of multiple directionality (46). In the present tense, syncret-
ism of 2PL/3PL is common to all the dialects. (The origin of this pattern is
unclear, so we take it as a given.)20 What is of interest to us is the extension of
this pattern into the perfect, which occurs in a few dialects. Significantly, it is
sometimes the 2PL ending which prevails (Sinti), and sometimes the 3PL
ending (Northeastern Romani, and partly in Welsh Romani).
(46) Romani (Matras 2002: 145)
perfect
Central, Finnish,
present reconstructed Balkan, Vlax Sinti Welsh Northeast
1PL -as *-am -am -am -am

..................
-am
.
.
. .
2PL -en *-an -an, -en -an .
.
.
.
-e, -an .
.
.
.
-e
.
. .
.
. .
3PL -en *-e -e -an .
.
.-e
................
.
.
. -e
20
The syncretic 2PL/3PL corresponds to what would be expected for the 3PL. This is unex-
plained, but possibly the resemblance between the 2PL perfect ending -an and the 3PL
present ending -en had something to do with it.
Thus, while there is some diachronic evidence that third person forms can
replace others diachronically, it is hardly an exceptionless generalisation.
Especially striking are the examples that appear to show second person
replacing third, for which no explanation readily comes. Nevertheless,
there is one area where the default status of third person is more firmly
established, namely the wholesale loss of person marking, as seen above in
German, the Kumta dialect of Kannada and in Gothic; in all these examples
it is the etymological third person form which prevails. A seeming counter-
example to this observation comes from the present tense in Modern East
Scandinavian languages (Swedish, Danish and Norwegian), where person
distinctions have been eliminated, and there is but a single form in -r, which,
as discussed above, was originally second person. However, it was only after
it was extended to third person singular that it ultimately predominated (47);
that is, the form which was extended to all person values served as a third
person form (among other things) immediately prior to its extension.
(47) Eastern Scandinavian (present indicative)
proto-ES > Old ES > Modern ES
1SG *-(V) -(V) -(V)r

2SG *-(V)r ! -(V)r -(V)r
3SG *-(V)þ -(V)r ! -(V)r
This suggests that two types of operation were at work: (i) an idiosyncratic,
morphologically specified extension of the second person to the third, and
(ii) a cross-linguistically unexceptional extension of the 3SG form.
In the examples above, a form originally associated with one value was
extended to other values. Such developments, where incomplete, may
account for some of the instances of synchronic directionality effects that
we have seen. However, in some cases we may be seeing the result of the
reverse development, as suggested by the historical development from Old
Nubian (first attested in the seventh century; Browne 2002: 1) to its direct
descendant, Nobiin, examples from which are cited above ((29) and (39)).
Old Nubian had regular 2SG/3SG and 1PL/2PL syncretism in all its verbal
paradigms (48); the origins of this pattern are unknown. In Nobiin, 2SG
and 2PL suffixes were innovated and added to the indicative and negative
paradigms, but not to the interrogative or conditional. Thus, the syncretic
(48) Old Nubian present indicative (Bechhaus-Gerst 1996: 237)
SG PL
1 -ire -iro
2 -ina, -ena -iro
3 -ina, -ena -iran, -eran
paradigms represent the older state of affairs, and the directional effect
that results is a reflection of diachronic layering.
3.2.5 Summary
Seen in terms of the syncretic values, some common patterns emerge,
which are most easily expressed negatively: syncretism of first person and
third is rare, and syncretism which is confined to the singular is rare.
Directional effects, from which we might deduce hierarchical relationships
between feature values, do not lend themselves to any generalization,
except for the observation that, diachronically, when person distinctions
are effaced completely, the form which prevails is likely to have originated
as the third person form.
3.3 Person syncretism in two-place verbs

3.3.1 Introduction
Verbs which mark both subject and object (two-place verbs) are a case apart
from verbs which mark only a single argument (one-place verbs). The inter-
action of subject and object person appears to encourage person syncretism,
especially that of the subject. Significantly, syncretism in two-place verbs is
independent of syncretism in one-place verbs. Thus, in the languages of the
World Atlas of Language Structures sample, illustrated in Appendix 3, twenty
languages evince person syncretism in two-place verbs (out of sixty-nine
languages which mark both subject and object person), and of these, thirteen
have person syncretism only in two place verbs. Note that in the remaining
seven, the syncretic pattern in two-place verbs is distinct from the one found
in one-place verbs; instances where a syncretism found in one-place verbs is
simply carried over to two-place verbs have not been recorded. Below we
distinguish three patterns, arranged according to their frequency: syncretism
of subject person, syncretism of object person, and other patterns.
3.3.2 Syncretism of subject person

Syncretism of subject person is the most common pattern affecting two-
place verbs, found in thirteen out of the twenty languages in the World
Atlas of Language Structures sample. In many instances first or second
person objects are involved. It is known that first and second person
objects trigger exceptional morphological marking (see Heath 1991,
1998, Kibrik 1997, Noyer 1997). Kibrik (1997) accounts for this in terms
of the animacy hierarchy and the notion of prototypical agents and
patients: arguments of high animacy or prominence (such as first and
second person) are prototypical agents, and arguments of low animacy
or prominence (such as third person) are prototypical patients. When the
former are in the patient role, this may lead to some special morphological
treatment, the most familiar example being the inverse marking of
Algonquian languages. Syncretism in this context implies the complete
neutralization of subject–person distinctions, since only two subject-person
values are possible (second or third person with a first person object, and
first or third person with a second person object). Various means are
available for neutralizing subject person.
Subject person not marked

In some cases it is evident that subject marking is simply absent. For
example, in Guarani (49), all values of subject person are syncretic when
the object is first person. Note that the marker which is used is identical to
the so-called ‘personal reference’ form for first person. (The personal
reference form is used for the subject of stative verbs and for possessors.)
Thus, in the context of a first person object, subject person is not marked
at all.
(49) Person prefixes in Guarani (Gregores and Suárez 1967: 131–2)
object
‘personal
reference’ 1SG 1PL 2SG 2PL Ø or 3
1SG še- — — ro- po- a-

1PL ore- — — ro- po- ro-
subject
2SG ne- še- ore- — — re-

2PL pene- še- ore- — — pe-
3 i- še- ore- ne- pene- o-
Nevertheless, these forms still seem to be sensitive to the presence of subject

person, if not its person value, in as much as we find the ‘personal reference’
prefixes sˇe- ‘1SG’ and ore- ‘1PL’, and not simply the intransitive subject prefixes
a- ‘1SG’ and ro- ‘1PL’. In other languages we may find a system which allows for
distinctions in the value of subject person even where the subject is not overtly
marked, as in the Chukotko-Kamachatkan language Koryak. In (50) and
(51) the person-number markers of Koryak are shown. (Dual is shown in
preference to plural because it is morphologically basic in Koryak; their
behaviour is the same with respect to subject and object marking.)
First consider the suffixes shown in (50). Where the object is 1DU or
second person, subject person is not distinguished; where the object is third
person, first person, 2SG and 3DU subject person are not distinguished from
each other. We can interpret these forms as marking the object person
only, unspecified for subject.
(50) Koryak verbal suffixes (Žukova 1972: 232, 252–4)
object
Ø 1SG 1DU 2SG 2DU 3SG 3DU
1SG -k — — -gi -t«k -n -net

1DU — — -gi -t«k -n -net
subject
2SG -j -j -m«k — — -n -net

2DU -t«k -t«k -m«k — — -tk« -tk«
3SG -j -j -m«k -gi -t«k -nin -nin
3DU -g«hi -g«m -m«k -gi -t«k -n -net
Now consider the distribution of the prefix ne- in (51).
(51) Koryak verbal prefixes (Žukova 1972: 232, 252–4)
object
Ø 1SG 1DU 2SG 2DU 3SG 3DU
1SG t«- — — t«- t«- t«- t«-

1DU m«t- — — m«t- m«t- m«t- m«t-
subject
2SG ine- ne- — —

2DU ine- ne- — —
3SG ine- ne ne- ne-
3DU ne- ne- ne- ne- ne- ne-
Except where the object is 1DU, it serves to distinguish subject person,

though it is clear from its distribution that it does not overtly mark subject
person as such. Following Comrie (1980), we can interpret them as a kind
of inverse marking, indicating that the relationship of subject to object
runs counter to the person-number hierarchy 1 > 2 > 3SG > 3DU/PL. The
prefix ne- is found when the subject is lower on the hierarchy (i.e. to
the right) than the object, or when both are at the bottom of the hierarchy.
The forms with the prefix ine- likewise lack overt marking of the subject
person but have a different structure from those in ne-. Morphologically,
the ine- forms are intransitives (Žukova 1972: 254–5), apparently based on
the antipassive paradigm (Spencer 2000: 210). Nevertheless, because these
forms are used only with a 1SG subject, subject person is unambiguously
implied.
Extension of subject markers

In some cases the range of the overt subject marker is quite extensive.
The Gunwinyguan (Australian, non-Pama-Nyungan) language Dalabon
employs this approach when singular subjects are combined with a first
person object. In that case the third person transitive subject form kah- is
used for both second and third person subjects (52). Note further that kah-
is the form used when the object is lower in animacy than the subject.
We discuss this in our case study of Dalabon in Chapter 5: x5.3. (Where
the object is second person, only the object person is marked, as discussed
above.)
(52) Dalabon singular transitive verbal prefixes (Evans, Brown and

Corbett 2001: 199)
object
1SG 2SG 3SG
1SG — djah- ngah-

subject
2SG kah- — dah-

3SG kah- djah- kah- (bvkah-)
The Tibeto-Burman language Trung likewise employs substitution of

subject person when the object is first person (53). In this case, though,
the directionality is reversed: the second person marker -n« is used for third
person subjects as well.
(53) Trung verbal markers (Watters 2003: 396)
object
1 2 3 or ø
1 — -N -N
subject
2 n«- -N — n«-
3 n«- -N n«- Ø
3.3.3 Syncretism of object person

Syncretism of the object person is less common than syncretism of subject
person. In the World Atlas of Language Structures sample it is found in
seven out of the twenty languages displaying syncretism in two-place
verbs. In contrast with subject-person syncretism, the interaction of sub-
ject and object person does not necessarily play a role in syncretism of
object person. This is due to the existence of systems where both arguments
are overtly marked, with distinct subject and object forms. Thus transitive
verbs provide an environment for a distinct series of person-marking
forms, which in turn may be syncretic. For example, in Barbareño
Chumash (54), an extinct language of California, subject markers are
prefixed and object markers are suffixed, and each series is distinct.
Subject prefixes distinguish three persons (and three numbers), while
object suffixes have syncretism of first and second person (as well as dual
and plural number). Thus, syncretism of the object marker is inherent in its
form and not conditioned by the value of subject person.
(54) Subject and object markers in Barbareño Chumash

(Beeler 1976: 255)
subject object
1SG k- -it
2SG p- -in
3SG s- -us (or Ø)
1DU kiš- -iyuw
2DU piš- -iyuw
3DU siš- -wun
1PL kiy- -iyuw
2PL piy- -iyuw
3PL šiy- -wun
Otherwise, the examples are not numerous enough to make any mean-
ingful generalizations.
3.3.4 Other patterns

Other patterns may be found besides the simple syncretism of subject or
object person, either by transposing subject and object roles, or by com-
bining incompatible subject and object values. Mojave illustrates both
possibilities (55). First, the combination of second person subject and
third person object is not distinguished from a third person subject and
second person object, both being marked with m-. Given that m- otherwise
functions as a second person prefix, it is clear that where both second and
third person are involved, only the second person argument is marked.
Subject or object status plays no role. With a first person argument, the
situation is more complex. The prefix ny- marks any first person argument
involved in a transitive construction, except where the object is third
person. Where the subject is second person, this too is marked, yielding
the sequence ny-m-. As a result, the prefix ny- used by itself marks the values
‘first person > second person’ and ‘third person > first person’. Thus,
although both the functions of ny- involve first person, in one it acts as the
subject, in the other as the object.
(55) Verbal prefixes in Mojave (Munro 1976: 12–13)
object
1 2 Ø or 3
1 — ny- ?-
subject
2 ny-m- — m-
3 ny- m- Ø
The Nilo-Saharan language Turkana goes even one step further in

combining incompatible values (56). The prefix kI- marks (i) a first plural
argument with any other argument, (ii) a second person subject with a first
person object, and (ii) a third person subject with a second person object.
In the case of Mojave, ny- was identifiable as some kind of a first person
marker. Here, the prefix kI- is not uniquely associated with any one person
(56) Transitive verbal prefixes in Turkana (Dimmendaal 1983: 282)
object
1SG 1PL 2 3
1SG — — ka- a-
subject
1PL — — kI- kI-

2 kI- kI- — I-
3 ka- kI- kI- e-
value. At best it can be defined disjunctively: it involves first and/or second

person.21
3.3.5 Summary
Most examples of syncretism in two-place verbs involve subject-person
distinctions when the object is first or second person, a phenomenon which
some have seen as involving the animacy hierarchy: high animacy objects
disfavour the explicit marking of the subject. To a much more limited
extent we find syncretism of object person, without any obvious connec-
tion to the animacy hierarchy. Other syncretic patterns may involve under-
specification for the distinction between subject and object, or remain
difficult to characterize.
3.4 Gender
3.4.1 Introduction
Gender categories differ greatly across languages. At the core of all sys-
tems is some distinction of sex or animacy, but beyond that, the principles
21
In fact, Dimmendaal (1983) treats it as a case of accidental homophony, attributing to it
three separate underlying morphological representations, predicated on the following
assumptions: (a) where neither argument is third person, the subject is marked in pre-
ference to the object; (b) first and second person are marked in preference to third; (c) k- is a
sort of inverse marker (not termed so by Dimmendaal, though), used whenever first or
second person are objects:
kI! kI1 1PL (1PL > 3)
k + kI! kI2 1PL ‘inverse’ (1PL object or 1PL > 2)
k + I! kI3 second person ‘inverse’ (2 > 1SG object or 3 > 2 object)
However, this analysis crucially depends on the existence of the distinct inverse prefix k-,
for which there is no independent evidence, so the analysis remains speculative. Heine
(1981: 116) likewise treats a similar pattern in the related language Camus as a result of
accidental homophony.
by which nouns are assigned to gender may vary wildly, determined by

language-specific semantic and morphological principles. Nevertheless,
allowing for a certain degree of abstraction, we can make some valid
cross-linguistic observations. We will look at gender syncretism in two
domains: across numbers, and across different targets.22
3.4.2 Gender and number

Gender shows a special affinity for syncretism across number. It is prob-
ably no exaggeration to say that, where gender marking is morphologically
distinct between singular and non-singular, gender syncretism is more the
rule rather than the exception.23 Corbett (1991: 155–6) describes two
patterns associated with gender syncretism across number, convergent
and crossed, based on the Africanist practice of drawing lines to connect
singular gender forms with plural gender forms. In a convergent system,
there is a many-to-one relationship between singular and plural forms. The
Dagestanian language Karata (57a) displays this configuration, where
masculine and feminine have distinct forms in the singular but are asso-
ciated with a single form in the plural. In this case, there is syncretism in the
plural. A crossed system is one where there is a many-to-many relationship
between singular and plural forms (Heine 1982: 197). Thus, in Albanian
(57) a. Karata ‘that.ABS’ b. Albanian ‘this.NOM’

(Magometbekova 1967: 329) (Newmark 1982: 263)
SG PL SG PL
M hadiw M ky N
këta
F hadij hadibaj F kjo këto
N hadib hadiraj
(57b) there are two distinct forms in each number; the masculine singular
form may be associated with either of the plural forms, while the feminine
plural form may be associated with either of the singular forms. This is the
result of the existence of a large and productive class of nouns which seem
to switch gender across number, taking masculine forms in the singular
and the feminine forms in the plural (Newmark 1982: 133). Although the
22
In as much as the present book concerns syncretism in inflection, we look only at gender
agreement marking, and not the overt marking of gender on controller nouns.
23
And even where gender marking is not morphologically distinct, as in the Andi dialects
discussed below.
Albanian grammatical tradition does not sanction this label, by distribu-

tional criteria this class justifies being seen as a distinct gender, which we
can call neuter, which is syncretic with the masculine in the singular and
with the feminine in the plural (as is done with Romanian, which has
essentially the same system; see (62) below).
These schemata are useful to keep in mind, especially as there is a large
body of literature on Niger-Congo languages (particularly rich in gender,
and gender syncretism) in which gender marking is presented precisely in
this way. However, for the remainder of this section we will present such
paradigms in the terms outlined in Chapter 2: x2.1.1, where masculine and
feminine plural in Karata are an example of simple syncretism (58a), and
the neuter in Albanian is an example of a non-autonomous value (58b).
(58) a. Karata (= 57a) Albanian (= 57b)
SG PL SG PL
M hadiw hadibaj M ky këta

F hadij hadibaj F kjo këto
N hadib hadiraj N ky këto
3.4.2.1 Smaller systems

In order to isolate cross-linguistically valid generalizations, it will be useful
if we first look at simple systems with only three genders, as in the examples
above. The first observation we can make is a paraphrase of Greenberg’s
familiar universal 37 (‘A language never has more gender categories in
non-singular numbers than in singular’; 1963: 112): convergence typically
goes in the direction of singular to plural, and not the reverse, with the
result that we may find more distinct gender forms in the singular than
the plural, but not more forms in the plural than the singular. As we shall
see below (in Kisi and Ju/ 0 hoan), there are convincing counterexamples to
this claim, but it does describe a strong tendency. For example, while com-
plete neutralization of gender distinctions in the plural is not uncommon
(e.g. Russian or German, with three gender distinctions in the singular and
none in the plural), the reverse is all but unknown (the Nilo-Saharan
language Fur, with a distinction of human versus non-human gender in
the plural but not the singular, is one of the rare examples; Jakobi 1990:
102–3).
Gender convergence in the plural itself seems to be under certain con-

straints. Consider the examples in (59). Both are languages with distinct
masculine, feminine and neuter forms in the singular, and with syncretism
of two genders in the plural. In Laal (59a), an unclassified language of
Chad, masculine and feminine plural are regularly syncretic, illustrated
here with the possessive pronoun. In the Iroquoian language Seneca (59b),
feminine and neuter dual and plural are regularly syncretic.24 Both these
are common patterns; less frequently found are patterns such as that found
in the Nilo-Saharan language Turkana (60), where gender convergence
results in syncretism of the masculine and neuter in the plural.
(59) a. Laal possessive pronoun

(Boyeldieu 1982: 14) b. Seneca (Chafe 1997: 563)
singular plural singular dual plural
M da:r dè:rı́ M ha- hni- hati-, he˛n-

F dò:g dè:rı́ F ye- kni- wati-, we˛n-
N dà:ná duà:ná N ka-, w-, y- kni- wati-, we˛n-
(60) Turkana (Dimmendaal 1983: 217)
(restrictive non-restrictive)25
singular plural
M e- lo- Ni- lu-
F a- na- Na- na-
N i- ni- Ni- lu-
As further examples below will illustrate, the plural appears to accord a

privileged status to such semantic categories such as male personal, mascu-
line, human or animate. In this respect, masculine and feminine form a
natural grouping (‘human’) as do feminine and neuter (as the remainder
after masculine has been isolated), but masculine and neuter do not. Of
course, different languages have different means for assigning gender, and
24
Where multiple forms are listed for Seneca, these are allomorphs conditioned by the initial
segment of the following stem.
25
Restrictiveness is a notion allied with, but not exactly equivalent to, definiteness
(Dimmendaal 1983: 218).
these are not always semantic but may be based on phonology, morpho-
logy, or outright lexical stipulation for some items. Nevertheless, there is
typically a semantic core which justifies gender designations such as ‘mas-
culine’ or ‘human’, and the generalization above still applies to these core
genders, regardless of the contribution of non-semantic principles to gen-
der assignment.
The two patterns represented by Laal and Seneca may themselves be
combined with any pattern in the singular. For example, the Dravidian
language Telugu (61a) behaves like Laal in the plural, with syncretism of
masculine and feminine, but it also has regular syncretism of feminine and
neuter in the singular. The system found in the Kru (Niger-Congo) lan-
guage Grebo (61b) can be seen as the mirror-image. It has three genders:
one for humans, one for large, valuable things (‘non-human 1’), and one
for all other things (‘non-human 2’). There is clearly a scale of prominence,
with humans as the most prominent and the second non-human gender as
the least. In the plural, the two less prominent genders are syncretic, while
in the singular the two most prominent genders are syncretic.
(61) a. Telugu copular suffix

(Krishnamurti and Gwynn b. Grebo third person pronoun
1985: 138) (Innes 1966: 52–3)

M -d:u -ru HUM O o
F -di -ru NHUM‘1’ O e
N -di -y(i) NHUM ‘2’ e e
Most significantly, masculine/neuter syncretism may occur in the singular.

Romanian, illustrated in (62), is a familiar example, where it is combined
with feminine/neuter syncretism in the plural (this pattern is consistent on
all targets). Romanian can serve as a good example of the contrast between
(62) Romanian third person pronoun,

nominative case (Rothe 1957: 78)
singular plural
M el ei
F ea ele
N el ele
gender syncretism in the singular and the plural, if we look briefly at its
history (the following account is based on Windisch 1973). Roughly speak-
ing, the gender affiliation of nouns in contemporary Romanian continues
that of their Latin forebears, masculine, feminine and neuter. There is no
good evidence that the syncretic patterns as such had any but a phono-
logical or morphological basis. In Latin, masculine and neuter were only
weakly distinguished from each other in the first place, while in the plural,
phonological merger has been proposed, though there is no consensus
(Windisch 1973: 157–69). Thus, we can take as a point of departure a
system with two semantically arbitrary patterns of gender syncretism, one
in the singular and one in the plural. However, over time there has been a
continual migration of inanimate nouns from the masculine to the neuter,
so that the masculine gender has become increasingly restricted to male
animates (Windisch 1973: 117, 196). Because masculine and neuter are
syncretic in the singular, morphologically, this has consequences only in
the plural: the switch from masculine to neuter involves using the ‘femi-
nine’ agreement form rather than the masculine. As a result, the distribu-
tion of forms in the plural has a much greater degree of semantic
predictability than in the singular. A similar system has also developed in
Albanian (see (58b) above), and was also characteristic of Tokharian
(Windisch 1973: 18–20).
3.4.2.2 Larger systems

When we move to systems with a larger inventory of genders, we largely see
the same themes repeated, namely the consolidation and realignment in
the plural of gender distinctions into groups with a more-or-less transpar-
ent semantic justification. A particularly clear illustration of the ‘semanti-
cization’ of gender forms in the plural comes from dialects of the
Dagestanian language Andi, discussed by Corbett (1991: 198–9). The
dialects termed types A, B and C all display four distinct gender forms in
the singular and plural. The conservative type A dialect has the simplest
system (63a), distinguishing human males (gender I, with the marker w),
human females (gender II, with the marker j), most non-human animates
(gender III, with the marker b) and the remainder (gender IV, with the
marker r). These gender markers are the same for singular and plural. In
the more innovative type B dialect (63b), animate members of gender III
pattern with female nouns in the plural, taking the marker j. In the type C
dialect (63c), animate members of gender IV do the same. As a result, in the
type C dialect, even though the repertory of gender markers is the same in
the singular and plural, there is a greater degree of semantic predictability

in the plural, since all non-male animates take the same marker j, whereas
in the singular they are distributed across the three markers j, b and r.
(63) Andi (Xajdakov 1980, cited in Corbett 1991: 198–9)
a. Type A b. Type B c. Type C
SG PL SG PL SG PL
I w w I w w I w w
II j j II j j II j j
III b b III AN b j III AN b j
IV r r III INAN b b III INAN b b
IV r r IV AN r j
IV INAN r r
The Maasinankore dialect of the Atlantic (Niger-Congo) language Fula,

described by Breedveld (1995), represents a particularly extreme example
of the consolidation of gender forms in the plural. The basic system is given
in (64); minor genders consisting of one member are omitted. There are
twelve distinct gender markers in the singular, but only four in the plural:
(64) Fula (Breedveld 1995: 457)
singular plural
'O BE human
'O 'DI
NGE DI
NDU 'DI largely animate
NGOL 'DI
NDI 'DE
NDE 'DE
NGO 'DE
BA 'DE inanimate
KI 'DE
KA 'DE
NGAL 'DE
NGEL KOY augmentatives
most genders take the markers 0 DI or 0 DE in the plural. The paradigm in (64)
in fact represents an oversimplification. All of the singular gender markers
which are associated with 0 DE may take 0 DI as well (with the exception of the
augmentative gender NGEL). 0 DE is used strictly for inanimates (though
including ‘nasty animals that bite or sting’; p. 446). 0 DI is used for animates,
but also appears to function as the default, in that even inanimate nouns
which used to take 0 DI are now showing a tendency to use it.
To the extent that the reorganization of gender groupings in the plural
involves such consolidation, the predictions of Greenberg’s universal 37
are maintained. However, reorganization in the plural can also involve
massive splitting, to the extent that there are more distinct gender forms in
the plural than in the singular. This is the case with the Atlantic (Niger-
Congo) language Kisi, which we will look at in some detail, based on
Childs (1995: 162–70). It has seven genders, realized by three forms in
the singular and five forms in the plural. The approximate semantic
composition of the genders is given in (65), listed in order of the number
of nouns they contain (out of a total sample of 910).
(65) Kisi (Childs 1995: 162–8)
SG PL semantics
o la no semantic core; default class for
borrowed inanimates not denoting liquids 43.4%
o a virtually all animates 27.3%
i N inanimate, but little semantic cohesiveness
(possible semantic core: small and round
objects) 15.4%
le i long and thin, string-like objects 4.3%
le la inanimates; productive for deverbal or
denominal abstract nouns 3.2%
le ma 1. liquids (productive for borrowings) 3.1%
2. pointed objects
o i trees and tree-like plants 3%
The i/N gender is a semantically diffuse class of inanimates, which is distinct

from other genders in both the singular and plural. Of the remaining genders,
the plural forms lend themselves much more readily to semantic character-
ization than the singular ones, with three of the forms having a relatively
clear semantic basis, and the remaining one functioning as a default:
* a is for animates
* i is for long objects
* ma is for (i) liquids and juicy plants, and (ii) sharp, pointed objects
* la is the default for inanimates (except those belonging to the i/N
gender)
Neither of the singular forms can be so characterized. The marker le is used

exclusively for inanimate genders but otherwise can be paired with any of the
plural markers, while o can only be defined negatively, being used for any-
thing but long, thin, string-like objects. The most salient distinction between
the two appears rather to be morphological, that between a productive and
unproductive marker. The o class is large (74 per cent of nouns) and is
productive for borrowings except for words denoting liquids, while the le is
small (11 per cent) and is unproductive except for nouns denoting liquids.
In Kisi, the syncretic gender forms in the singular mark agreement with a
heterogeneous group of nouns, while in the syncretic forms of the plural
there is a strong semantic component to the grouping. It thus conforms to
the generalization that plural forms show a more semantically coherent
grouping of gender than the singular ones. However, even this is not
always the case, as demonstrated by the Khoisan language Ju/ 0 hoan,
which violates both this generalization and Greenberg’s universal 37.
Ju/ 0 hoan distinguishes five genders. The first three are relatively easy to
characterize: I is for humans, II for animals, as well as the names of
(non-Ju/ 0 hoan) nations and ethnicities, and III is for (most) plants and
foodstuffs. The other two genders are rather harder to characterize on a
semantic basis. Both are exclusively inanimate; IV contains many long
objects, but otherwise shows ‘little semantic coherence’ (Dickens 1992: 16).
Gender V is likewise heterogeneous, being used, among other things, for
deverbal nouns, for reference to clauses, and for body parts. Gender is
syncretic in both the singular and the plural, whereby free pronouns have
more distinct gender forms in the plural than in the singular (four versus
three). What is more, the syncretic gender form in the singular seems
semantically more coherent than that in the plural. In the singular, genders
I, II and III are syncretic, thus embracing all animates, as well as living, or
at any rate growing, inanimates. In the plural, genders II and IV are
syncretic; this grouping defies any obvious semantic characterization.
Güldemann and Vossen (2000: 112) account for this unusual distribution
of forms by positing that the forms ha and hı` perform distinct functions
depending on the animacy of the noun. For inanimate nouns they mark
(66) Ju/0 hoan (Dickens 1992: 12–16)
pronoun
singular plural
I ha sı̀la
II ha hı̀
III ha ha
IV hı̀ hı̀
V ká ká
gender without reference to number:26 ha for plants and hı` for long objects.
For animate nouns they mark number: ha for singular and hı` for plural.
(The other two forms, ká and sı`la, are monofunctional.) On this view, what
we see in Ju/ 0 hoan is not so much gender syncretism as syncretism between
the features of gender and number; in either case, it is an unusual system.
3.4.3 Syncretism restricted by target

The syncretic patterns described above are restricted to a particular mor-
phosyntactic context, namely to a particular number value. Gender syn-
cretism may also be restricted on the basis of the target; that is, some
lexemes display the syncretism and others do not. For example, in Latin,
there are some third declension adjectives where masculine and feminine
are syncretic; compare the nominative singular forms fortis ‘strong.M/F’
versus acer ‘sharp.M’, acris ‘sharp.F’. Given that distinctions between
target types involve language-specific lexical and morphological features,
the expectation of finding any valuable cross-linguistic generalizations is
slim. We will therefore content ourselves with indicating the variety of
patterns that can result.
A good illustration (and a particularly complex one) of the sort of
variation that can be found comes from the non-Pama-Nyungan
Australian language Gaagudju. The language has four genders: I for
human males, II for human females, III for plants and IV for remaining
inanimates. The basic set of gender-marking prefixes distinguishes all four;
26
Note that, although the pronouns for genders III–V do not distinguish between singular
and plural, it is not because number is irrelevant to nouns of these genders; e.g. utò ~ utòsı`
‘car ~ cars’ (gender III), n!àmà ~ n!àmàsı` ‘road ~ roads’ (gender IV) and !àihn ~ !àihnsı` ‘trees’
(gender V).
it is used for declension 1 adjectives, and a similar set is found in demon-

stratives, and in most verb paradigms. Otherwise, various patterns of
syncretism are involved. Second declension adjectives display syncretism
of genders III and IV. Free pronouns, as well as indirect object enclitics,
display syncretism of genders II, III and IV. Third person subjects of
intransitive verbs in the realis present tense display syncretism of genders
II and IV, while in transitive realis verbs with a first person subject, third
person objects show syncretism of genders I and IV. While the nominal
patterns may be explicable semantically (second declension adjectives
show syncretism of the non-human genders, and pronouns oppose male
humans to everything else), the verbal patterns are quirkier. Why the
patterns should be distributed across the targets in this way remains
unclear.
(67) Gaagudju (Harvey 2002: 144, 157, 224–5)
realis verbs
adjectives present and

present unmarked tense
first second (intransitive object with first
declension declension pronoun subject) person subject
I (male) Ø na- naawu Ø arra-

II (female) njiN- njiN- ngaayu nj-dja- nji-rra-
III (plant) ma- naN- ngaayu ma-ya- ma-ra-
IV (residue) gu- naN- ngaayu nj-dja- arra-
3.4.4 Summary
Cross-linguistically, gender syncretism shows some strong tendencies
when considered in the context of number. The typical pattern is the
consolidation of genders in the non-singular into syncretic classes with a
more-or-less transparent semantic basis (or at any rate, more transparent
than in the singular). However, this is only a tendency, and languages such
as Ju/ 0 hoan show that this need not be so. Gender syncretism that is
restricted to particular targets is, by its nature, language-specific, and
even within a language (such as Gaagudju), shows a mixture of semantic
and morphological motivation.
3.5 Number
3.5.1 Introduction
Syncretism within the feature number is a surprisingly straightforward
affair, typically dictated by the semantically natural grouping of non-
singular number values (see Corbett 2000: 39–42). Number syncretism is
often correlated with the animacy hierarchy (Corbett 2000: 55–66), with
items lower on the hierarchy showing a smaller range of non-singular
values. For example, in Gothic, dual and plural are distinct only for first
and second person, reflected both in free pronouns and in person-number
agreement on verbs (68). All other nominals (pronouns, nouns and adjec-
tives) distinguish only singular and non-singular. (See Cysouw 2003 for
further cross-linguistic data.)
(68) Number values in Gothic (Wright 1930: 91, 120, 135)
pronoun (nominative) noun (nominative) verb (‘take’, present)

‘I’ ‘he’ ‘guest’ 1 2
SG ik is gast nima nimiþ

DU wit eis gasteis nimo
s nimand
PL weis eis gasteis nimam nimand
In the Gunwinyguan (Australian, non-Pama-Nyungan) language Bininj

Gun-Wok, the collapse of number values is correlated not with the value of
person, but with the morphological context. Bound pronouns (illustrated
here by intransitive subject markers) distinguish three number values for
each person (dual, trial and plural in the case of the first inclusive). Free
pronouns distinguish only singular and plural; the distinction between first
person exclusive and inclusive is also neutralized.27
The Slovene example cited below in (63) in Chapter 4: x4.4.4 and (9) in
Chapter 5: x5.1.1 represents a collapse of dual/plural which is even more
deeply embedded in language-specific morphology.
Given the regularity with which the semantically natural pattern is
manifested cross-linguistically, it is all the more interesting to consider
examples which deviate from the expected pattern, either in terms of the
27
This also obtains for part of the bound pronominal system, namely in transitive verbs
which mark both subject and object.
(69) Person and number in Bininj Gun-Wok (Evans 2003: 263)
subject (past tense) free pronoun

1SG nga- ngaye
1DU ngani-
1PL ngarri-
1INCL DU ngarr- ngad
1INCL TRI gani-
1INCL PL garri-
2SG yi- ngudda
2DU nguni- wudda
2PL ngurri-
3SG ba- nungga (M), ngaleng (F)
3DU bani- bedda
3PL barri-
values which are combined, or in terms of the apparent markedness

relationship between number values.
3.5.2 Values
If we limit ourselves to languages with singular, dual and plural, we should
expect that dual and plural should fall together. The unexpected patterns
then involve the dual falling together with the singular, or the singular and
plural falling together, leaving the dual distinct. The first pattern is found
in the distal habitual paradigm of the East Papuan language Yele (70).
In Yele, person, number and tense-aspect-mood are marked on the auxiliary
(which Henderson (1995) calls a ‘prenuclear element’). The distal habitual
auxiliary is the same as the near past with the addition of the suffix -mo
(though note that 1DU nyi- corresponds to the immediate past form). The
form dpıˆ is used for 2DU/3DU in the near past paradigm, while in the distal
habitual it is extended to the 3SG as well.
Regular syncretism of singular and plural (but not dual) does not occur
in our material, except alongside other patterns of number syncretism.28
A particularly striking example comes from the Kiowa-Tanoan language
Kiowa (discussed in greater detail in x3.7), which illustrates all three logical
possibilities. Number syncretism is dependent on noun class. Class I
28
A large part of the possessed noun paradigm in Yup’ik displays singular/plural syncretism,
in the absence of any other patterns of number syncretism. However, it appears to have a
fairly transparent phonological or morphophonological basis.
(70) Auxiliary forms for continuous events in Yele (Henderson 1995: 36)
immediate near distal

past past habitual
1SG nı̂ nı̂ nı̂-mo

1DU nyi ny:oo nyi-mo
1PL nmı̂ nmı̂ nmı̂-mo
2SG nyi nyi nyi-mo
2DU dpı̂ dpı̂ dpı̂-mo
2PL nmyi nmyi nmyi-mo
3SG Ø dı̂ dpı̂-mo
3SG Ø dpı̂ dpı̂-mo
3PL Ø dnyi dnyi-mo
(largely animates) has singular/dual syncretism, class II (inanimates) has

‘normal’ dual/plural syncretism, and class III has singular/plural syncret-
ism. The coexistence of these patterns, and particularly the unusual class
III pattern (this class consists of only four nouns), is bound up with the
system of inverse number marking (see x3.7).
(71) Kiowa nouns (Watkins 1984: 82–9)
I ‘horse’ II ‘tree’ III ‘apple’
SG ce˛ˆ : á:-dÒ álÒ:-bÒ

DU ce˛ˆ : á: álÒ:
PL ce˛ˆ :-gÒ á: álÒ:-bÒ
3.5.3 Directionality
Common assumptions about the markedness relations between number
values would lead us to expect that dual forms should be based on plural
forms. However, there are examples where the syncretic dual/plural form
looks like the dual instead. One such example comes from Koryak, where
the plural is morphologically derived from the dual (see (50–1) in x3.3.2).
For first and second person arguments, the plural is formed from the dual
plus the suffix -la. A careful reading of the sources suggests that -la is
unstable in some contexts (most often where the plural subject acts on a
third person object) and liable to be omitted.29 In such cases dual and
plural are syncretic, but it is the form otherwise associated with the dual
which prevails.
3.5.4 Summary
Most instances of the collapse of number values can be characterized as the
consolidation of non-singular number values, and this is largely the same
across different languages. Since this is in keeping with the number hier-
archy, most examples can be treated as the result of a language employing
different cut-off points on the hierarchy for different items. However, as
the system of plural marking in Koryak shows, this semantic hierarchy is
not necessarily mirrored by the morphological hierarchy. Nor is the num-
ber hierarchy an absolute constraint on number syncretism, but it comes
close. In this respect the behaviour of number is far more predictable than
that of other features.
3.6 Tense-aspect-mood
3.6.1 Introduction
In this section we consider syncretism of values of tense, aspect and mood
features, which we shall consolidate under the single rubric of ‘tense-
aspect-mood’ (TAM). The principles behind TAM systems are particu-
larly heterogeneous from language to language, and we are unlikely to be
able to make many cross-linguistic generalizations by looking at the values
of individual features. However, it turns out that there are some distinct
tendencies in morphological realization of TAM syncretism, which we
shall focus on.
Consider English weak verbs with the ending -ed, as in walked. This
realizes the syncretic combination of preterite and past participle, other-
wise distinct in strong verbs, as in drove versus driven. In a small number of
verbs (monosyllabic roots ending in a dental consonant, with a lax high
root vowel), the ending -ed, though possible, is typically absent, resulting
in syncretism of all three possible distinctions.
29
The variation is partly determined by the source and, within a particular source, the tense-
mood paradigm (where the system of person-number marking should not vary, in princi-
ple); within Žukova (1972) the variation also depends on the chapter where the material is
presented. See the entry under Koryak in Baerman, Brown and Corbett (2002a) for details.
(72) PRESENT drive walk quit quit

PRETERITE drove walked quitted quit
PAST PARTICIPLE driven walked quitted quit
Thus English illustrates two contrasting morphological patterns. In the

first case there is an affix which directly encodes the syncretic values: -ed
typically marks the syncretic combination of preterite and past participle.
Where these two are distinct, their morphological realization is by other
means (ablaut, and possibly the ending -en). In the second case, the
syncretism is realized by not applying a rule (see Zwicky 1991: 117). The
syncretic form is the morphological base to which nothing has been done.
It is our impression that most examples of TAM syncretism that we are
familiar with are of this second type
3.6.2 Affix suppression

Some of the clearest examples of the suppression of what appears to be a
regular TAM affix come from Oceanic languages. For example, both
Gapapaiwa and the related language Tawala have a marker -na-, identified
as a future tense marker in Gapapaiwa and as a potential mood marker in
Tawala. It is regularly inserted after the person-number markers, except
that in Gapapaiwa it is not found with the 1INCL PL (73a), and in Tawala it
is not found in conjunction with any of the first person markers (73b). This
leads to syncretism with the morphologically simplest TAM form, the
subjunctive in Gapapaiwa and the simple aspect in Tawala.
(73) a. Gapapaiwa (McGuckin

2002: 308) b. Tawala (Ezard 1997: 117)
subjunctive future simple aspect potential mood
1SG a- a-na- 1SG a- a-

2SG ku- ku-na- 2SG u- u-na-
3SG i- i-na- 3SG i- i-na-
1INCL PL ta- ta- 1INCL PL ta- ta-
1PL ka- ka-na- 1PL to- to-
2PL ko- ko-na- 2PL o- o-na-
3PL i- i-na- 3PL hi- hi-na-
Note that, at least in the case of Tawala, the absence of -na- cannot be
attributed to any inherent incompatibility between first person and the
potential mood, as there are verbs such as ‘meet together’ (74) whose stems
alternate for all persons in the potential mood, a result of what appears
originally to have been incorporation of the -na- marker into the stem.
(74) Tawala ‘meet together’ (Ezard 1997: 117)
simple aspect potential mood
1SG a-‘m-boina a-n‘om-boina

2SG u-‘m-boina u-n‘om-boina
3SG i-‘m-boina i-n‘om-boina
1INCL PL ta-‘m-boina ta-n‘om-boina
1PL to-‘m-boina to-n‘om-boina
2PL o-‘m-boina o-n‘om-boina
3PL hi-‘m-boina hi-n‘om-boina
In this respect it is interesting to note that Loniu, another Oceanic lan-

guage, shows a similar effect with its potential mood marker, in this case
affecting the second person. The potential mood differs from the present/
past by the addition of the prefixed element kV- (in the plural, the vowel
depending on conjugation class) or k-. But the 2SG lacks this k-, resulting in
regular syncretism of the potential and present/past.
(75) Loniu ‘come’ (Hamel 1994: 73)
present/past potential
1SG* i-me, u-me ki-me, ku-me

2SG e-me e-me
3SG i-me ki-me
PL me ke-me
*free variation in the vowel of the 1SG prefix
Another Oceanic language, Ura (76), displays tense syncretism correl-

ated with number. Ura has three basic tense-mood paradigms which are
distinguished solely by their person-number endings: recent past, distant
past and optative. The recent past and distant past are identical for all
plural persons. These syncretic forms resemble, morphologically, the
recent past: while 1SG and 2SG distant past differ from the recent past by the
addition of -mi-, this element is lacking in the plural forms (the -mi- of the
1PL exclusive belongs to the pronominal element). The third person form in
the plural has the initial consonant c- characteristic of the recent past of the
third person form in the singular. There is no obvious semantic reason why
these two tenses should be undistinguished in the plural; note that in the
closely related language Sye, they are distinct (Crowley 1998: 90, 95).
(76) Ura (Crowley 1999: 157–61)
recent past distant past optative
1SG yau- yaumi- yaupi-

2SG ki- kami- kapi-
3SG (c)i- (y)i- pi-
1INCL PL (q)ur- (q)ur- qispir-
1PL qimir- qimir- ——*
2PL qir- qir- qipir-
3PL (c)ir- (c)ir- pir-
*not attested
An analogue to affix suppression can also be found involving stem

alternations (see (10) in Chapter 2: x2.2.1). For example, in the Nilo-
Saharan language Fur (77), this occurs with an accentual alternation.
Verbs distinguish three TAM paradigms (subjunctive, perfect and pres-
ent), marking person-number in all three. Within each TAM paradigm, in
addition to the person-number prefixes, there are two stem suffixes, one
for the 3PL and one for all other person-number values. These suffixes are
always distinct for the present tense, but, for most verb classes, the non-3PL
suffix is the same for the subjunctive and perfect. In class I verbs, sub-
junctive and perfect for the non-3PL forms are distinguished by a tonal
alternation realized on the suffix vowel (low tone is indicated by a grave
accent, high tone is unmarked). In class II verbs, there is no tonal alterna-
tion, so that subjunctive and perfect are syncretic for all but the 3PL.30
Similarly, stem vowel ablaut may be involved. In Latvian (78) there is an
interesting interplay between stem allomorphy and affix allomorphy
30
This does not seem to have an obvious phonlogical explanation, as -i can receive low tone,
e.g. the 1/2 person stem of class II,2 verbs has the shape HH-ı`, where H=high tone syllable
(Jakobi 1990: 111).
(77) Fur (Jakobi 1990: 104, 109)
‘descend’ ‘show’
subjunctive perfective subjunctive perfective
1SG ùfuNo ùfuNò àuli àuli

2SG jùfuNo jùfuNò jàuli jàuli
3SG buuNo buuNò faali faali
1PL kùfuNo kùfuNò kàuli kàuli
2PL bùfuNo bùfuNò bàuli bàuli
3PL NHUM buuNòl buuNùl faaliàl faaliè
3PL HUM kùfuNòl kùfuNùl kàuliàl kaulieè
across tense. Latvian has three synthetic tenses: past, present and future.
The future has a distinct set of endings and so is never syncretic with the
others. The present and past are partly distinguished by their endings, but
the first singular ending -u is the same for both tenses. Thus distinction of
the two tenses in the first singular is dependent on a stem alternation. First
conjugation verbs may undergo root vowel ablaut, or an alternation of the
stem-final consonant, while third conjugation verbs have a vowel exten-
sion in the past tense, as in ‘wish’ in (78). Second conjugation verbs, such as
‘wash’ in (78), and many first conjugation verbs undergo no stem alterna-
tion, and so have syncretism of present and past in the first singular.
(78) Latvian (Mathiassen 1997: 104–5)
third conjugation second conjugation

‘wish’ ‘wash’
present past present past
1SG gribu gribeju mazgaju mazgaju

2SG gribi gribeji mazga mazgaji
3SG/PL grib gribeja mazga mazgaja
1PL gribam gribej
am mazgajam mazgajam
2PL gribat gribej
at mazgajat mazgajat
There are two phenomena to note here. First, it has been claimed (Kariņš
1994: 113, citing Graudiņa 1969: 16–20), that some speakers reanalyse
second conjugation verbs as third conjugation verbs, precisely in order to
circumvent this homophony. Second, observe the second singular ending.
In the past it is -i, while in the present there are two allomorphs: -i, identical
to the past tense ending, and -Ø, as in the third person. The choice of
allomorph is partly determined by the presence or absence of a present
past stem alternation. Stems lacking an alternation have -Ø, as in ‘wash’,
thus maintaining the contrast between the two tenses: present mazg a
versus past mazg aji (-j- is inserted intervocalically by a general phonologi-
cal rule). Since -i is the historically expected ending, and its loss does not
follow from regular sound change, it has been argued that -i ! Ø was a
morphologically driven change, motivated by avoidance of syncretism
(Kariņš 1994: 113, citing Endzelins 1922). Note, however, that the range
of the new Ø ending extends in the first conjugation even to verbs which do
have a present past stem alternation, so the correlation is not complete.
The syncretism with the third person form which resulted in many cases (in
some verbs they have distinct stem alternants) was, evidently, more
tolerable.
3.6.3 Syncretic affixes

Now let us turn to examples such as English -ed, where an overtly syncretic
affix is used. English -ed is in fact fairly exceptional, in that it marks verbal
features without any indication of person.31 Most examples involve mark-
ers with cumulative marking of TAM and person. For example, in the
Otomanguean language Chichimeco, TAM and person are marked cumu-
latively by prefixes (which may differ according to conjugation class), and,
for some verbs, by a stem alternation as well (number is separately
marked). Example (79) illustrates a verb of the default conjugation class
I, showing only singular forms. There are eight TAM values; in the first
and in the third persons all eight TAM values have distinct prefixes. In the
second person the same prefix ki- is used for the anterior past, recent past,
future and present. The present may be distinguished by a stem alterna-
tion32 (as in this example, where the stem has three alternants: ’e, te´, nde´),
but the first three TAM values are always syncretic in this conjugation
class.33
31
The modern English forms were historically the base forms on which further inflection was
added (so that in Old English, past tense forms included person-number marking following
the -d).
32
The prosodic alternation manifest on the prefix is associated with the verb stem as well.
33
One class of verbs has su- for the second person present in place of ki- (p. 171), another has
su- for both the present and anterior past (p. 172).
(79) Chichimeco ‘give’ (de Angulo 1932: 165)
I II III IV V VI VII VIII
1SG tú-’e gá-’è kú-’e é-’è ú-’è nù-’é rá-’è sú-’è

2SG kı̀-té kı̀-té kı̀-té kı́-’è ı̀-té mı̀-té gı̀-té sı̀-té
3SG ù-ndé gà-ndé kù-ndé é-’è zú-’è mù-’é rú-’è sú-’è
I anterior past
II future
III recent past
IV present
V immediate past
VI potential
VII contemporaneous
VIII negative
3.6.4 Compound systems

Both of the morphological patterns described above – affix suppression
and the use of syncretic affixes – may occur within the same language, as is
the case with English. A particularly striking example comes from the
Papuan language Kiwai of the Trans-New Guinea phylum (Table 1),
with ten finite TAM values distinguished both by TAM affixes and by
the person prefixes (number is marked separately).
In the non-singular numbers, the TAM affixes by themselves suffice to
distinguish all the values, but some of these affixes are not found in the
singular, namely -ru- (definite past and present habitual), -du- (present
habitual) and -duru- (present and near past habitual). The person prefixes
in turn distinguish TAM, but only partly. First person is invariant n-, while
2/3 person marking is distributed as follows:
r- present, remote future, present habitual
g- definite past, past habitual, near past habitual
w- near past, immediate future, indefinite future, future habitual
Omission of -du-, -ru- and -duru- in the singular interacts with TAM
syncretism in the person markers. With the first person prefix n-, which
does not distinguish TAM, the near past, definite past and present are
syncretic, likewise the present habitual, past habitual and near past habit-
ual. Tense marking on the 2/3 person prefix largely disambiguates these
Table 1: Island Kiwai (Wurm 1975: 338–9)
near definite immediate indefinite remote present past near past future
past past present future future future habitual habitual habitual habitual
1SG n-S n-S n-S n-S-ri ni-mi-S-ri

ni-do-S-ri n-a-S-go n-a-S-go n-a-S-go ni-d-a-S
2/3SG w-S g-S r-S w-S-ri wi-do-S-ri
ri-mi-S-ri r-a-S-go g-a-S-go g-a-S-go wi-d-a-S
1DU n-S-do n-S-ru-do n-S-duru-do ni-do-S-ri ni-mi-du-
ni-du-do-S-ri n-a-du-do- n-a-ru-do- n-a-duru- ni-du-d-a-S
do-S-ri S-go S-go do-S-go
2/3DU w-S-do g-S-ru-do r-S-duru-do wi-do-S-ri wi-du-do-S-ri ri-mi-du- r-a-du-do- g-a-ru-do- g-a-duru- wi-du-d-a-S
do-S-ri S-go S-go do-S-go
1PL n-S-mo n-S-ru-mo n-S-duru-mo ni-mo-S-ri ni-du-mo-S-ri ni-mi-du- n-a-du-mo- n-a-ru-mo- n-a-duru- ni-du-m-a-S
mo-S-ri S-go S-go mo-S-go
2/3PL w-S-mo g-S-ru-mo r-S-duru-mo wi-mo-S-ri wi-du-mo-S-ri ri-mi-du- r-a-du-mo- g-a-ru-mo- g-a-duru- wi-du-m-a-S
mo-S-ri S-go S-go mo-S-go
Note:
* ‘S’ here stands for any verb stem
* trial number (not shown) is distinguished from the plural by the additional affix -bi-
values, except for the past habitual and near past habitual, which share the
prefix g- and so are syncretic for the 2/3 person as well.
3.6.5 Summary
Although we have concentrated on the morphological manifestation of
TAM syncretism, it is worth considering what relation these patterns
might have to the values which are syncretic. Affix suppression reflects a
morphological hierarchy: the syncretic form is basic, the non-syncretic
forms are morphologically derivative. Whether this can also be seen as
reflecting a hierarchy of feature values depends on whether one believes
that morphological structure reflects feature structure. Seen in terms of the
TAM values, the examples shown above are compatible with that assump-
tion, for example in Gapapaiwa (80a), Tawala (80b) and Kiwai (Table 1) it
is the future which is morphologically derived, which we can reasonably
construe as a non-basic tense value as well. In Ura (76) the distant past is
derived from the recent past, in Latvian (78) the past from the present.
Whether or not these reflect genuine tense hierarchies is uncertain, but it is
not improbable. However, the environment in which these patterns occur
is harder to account for on this view, e.g. why only the singular should be
affected in one case, the plural in another, first person in another, and so
on. In none of these cases does there seem to be a natural relationship
between, on the one hand, the feature value which conditions the suppres-
sion of the affix and, on the other hand, the value expressed by the affix.
Likewise, the syncretism encoded by syncretic affixes, such as English -ed,
does not necessarily express a natural class of values. Indeed, the identity
of the past tense and the participle in English is a well-known example of a
morphosyntactically unnatural class, used by Aronoff (1994: 23–5) to
argue for the existence of morphosyntactically arbitrary morphological
forms (‘morphomes’). Nor do classes of related values seem to be involved
in the Kiwai example (see the distribution of r-, g- and w- outlined above)
or the Chichimeco example (recent past, anterior past and future – and
only for second person, at that).
3.7 Polarity effects

3.7.1 Introduction
So far we have looked at examples where syncretism can be described as
the collapsing of distinct values of a single feature, which has allowed us to
consider whether this is a reflection of some closer relationship between the
given feature values. But examples which show polarity effects (see
Chapter 2: x2.1.1) defy description in these terms, because the syncretism
seems to operate over a paradigm rather than features. Somali provides a
well-known example. Determiners, which are suffixed to nouns, have two
distinct forms: broadly speaking, those in k- attach to nouns which are
masculine singular or feminine plural, while those in t- attach to nouns
which are feminine singular or masculine plural. (Note that there are
exceptions: class 3 and 4 nouns take k- for both numbers.)
(80) Somali definite article (Saeed 1999: 112)
singular plural
F -ta -ka
M -ka -ta
In its canonical formulation (e.g. Hetzron 1967: 184) polarity involves just
this sort of perfect mirror-image identity of non-contiguous paradigmatic
cells. As such, it is a rare phenomenon (the well-known examples are from
Semitic and Cushitic, but similar effects may be found in Oceanic; see
Corbett 2000: 162–4). However we might profitably think in looser terms
of generalized polarity effects, whose shared property is that their mor-
phosyntactic description requires reference to more than one feature. This
is what Pike (1965: 219) terms ‘second degree neutralization’, defined by
the fact that the syncretic forms can be made adjacent to each other only
through a linear arrangement, but not through a grid with two axes.
We can distinguish three types of polarity: (i) full polarity, (ii) partial
polarity, and (iii) mediated polarity. Full polarity is represented by Somali
above. Partial polarity is simply one half of full polarity, that is a syncretic
form which occupies non-adjacent cells,34 without a corresponding mirror-
image form. This is found in subject person marking on verbs in Siuslaw
(81), an extinct language of Oregon, where 2SG and 3PL are regularly
syncretic.
The third type, mediated polarity, is a relaxed version of partial polarity:
the form can be represented as adjacent cells, but a full description still
34
We stress that it is not the geometry of the paradigmatic representation per se which is
important, but the fact that reference must be made to different values of two (or more)
features.
(81) Siuslaw (Frachtenberg 1922: 468)
1INCL — -nØ -ns

1 -n -nxan -auxûn, -axûn,
2 -nx -tcı̂ -ts
3 Ø -nx -aux
requires reference to multiple conflicting feature values. For example, in

the Papuan language (82) Wojokeso, of the Trans-New Guinea phylum,
same-subject non-future medial verbs have a single form which serves for
first and second person in the singular, and second and third person in the
dual.
(82) Wojokeso (West 1973: 10)
1 -onji -ontae -ontone

2 -onji -onji -ontØfi
3 -i -onji -ontØfi
The existence of polarity effects is undeniable. However, as we shall see

in Chapter 4: x4.2.3, many morphological models exclude them outright
and so assume that, where they occur, they must be accidental. Therefore,
it is worth asking whether there is any morphological or semantic evidence
to suggest that they are ever systematic.
3.7.2 Morphological systematicity

As with other syncretic patterns, there are two ways of demonstrating
morphological systematicity: (i) the pattern is regular, that is, it is repeated
across distinct inflectional exponents (see Chapter 2: x2.3), and (ii) there is
diachronic evidence that the pattern was extended analogically.
Examples of the first type are hard to come by. For example, the system
in Somali, pervasive as it may be, involves a single set of inflectional
markers (t- and k-). This is typical of most examples – indeed, as we have
observed, this is typical of most examples of syncretism in general.
Nevertheless, there is at least one striking example from another Cushitic
language, Dhaasanac (data from Tosco 2001: 123–206). Subject-person

marking on verbs (see (83)) is found in the perfect and imperfect positive,
the dependent positive and the short past, and is realized as a stem alter-
nation: one form (labelled ‘A’) is found in the first person singular, third
person masculine, first person inclusive plural and the third plural; the
other form (labelled ‘B’) is found in the second person, third singular
feminine and the first exclusive plural.
(83) Dhaasanac stem alternations
singular plural
1INCL — A
1 A B
2 B B
3F B A
3M A A
The stem alternation takes a number of different shapes, illustrated in (84),

involving changes in the stem-final consonant, (e.g. kufi kuyyi) stem-
initial consonant (yes ces), as well as vowel alternations (seð sieti).
(84) Dhaasanac stem forms (Tosco 2001: 123–206)
form A form B
leeði leeti ‘fall down.PRF’

kufi kuyyi ‘die. PRF’
guurma guuranna ‘migrate. IMPRF’
?uufumi ?uufeeni ‘cough. PRF’
seð sieti ‘walk. PRF’
yes ces ‘kill. PRF’
The systematicity of the ‘A’ ‘B’ stem alternation is undeniable, as is the

heterogeneity of the syncretic values. Even if ‘A’ is treated as an elsewhere
form (Tosco 2001 considers it the morphological base form), ‘B’ remains an
example of mediated polarity. We analyse Dhaasanac in Chapter 5: x 5.2.
Another, perhaps less striking, example comes from Pashto, where the
distinct feminine and masculine forms show regular syncretism of the obli-
que singular and direct plural cases, here illustrated with adjectival inflection:
(85) Pashto ‘ripe’ (Grjunberg and Edel’man 1987: 15)
masculine feminine
singular plural
direct pox paxa pa:x« paxe

oblique pa:x« paxe paxo
Although evidence of synchronic morphological systematicity is hard to

find, there are some plausible diachronic examples. For example, the
pattern in Dhaasanac, as suggested by Sasse (1976), originally arose by
regular sound change. Verbs fall into two classes, one where person-
number markers were suffixed, and a much smaller class where they are
prefixed. The related language Arbore represents something like the older
stage of Dhaasanac (86). Regular sound change led to the assimilation of a
stem-final resonant (n, m, or l) with a following -t or -n suffix, thus yielding
syncretism of the first plural with the second person and third singular
feminine in resonant-final stems. This syncretic pattern was then extended
by analogy to all the other stem classes, as well as the small class of verbs
with person-marking prefixes, as in yes ces in example (84) above (Sasse
1976: 219–20).
(86) Arbore (Hayward 1984: 254 )
prefix conjugation suffix conjugation

singular plural
1 ?- -Ø n- -n
2 t- -t t- -t
3F t- -t y- -Ø
3M y- -Ø y- -Ø
A similar example comes from Sami declension. In the North Sami

dialects, the comitative singular and the locative plural are syncretic in
nouns. In the Western Finnmark dialects, which preserve the more archaic
pattern, these two forms remain distinct in pronouns. However, in the
Eastern Finnmark dialects, this syncretic pattern has been extended to
pronouns as well (Hansson 1996: 15–16).
(87) North Sami nominal declension, ‘eye’
singular plural
NOM čalbmi čalmmi-t

ACC/GEN čalmmi čalmmi-id
ILL čalbmá-i čalmmi-de
LOC čalmmi-s čalmmi-in
COM čalmmi-in čalmmi-iguin
ESS čalbmi-n čalbmi-n
(88) North Sami pronominal declension, ‘who’
Western Finnmark Eastern Finnmark

(conservative) (innovative)
NOM gii gea-t gii gea-t

ACC/GEN gea-n gea-id gea-n gea-id
ILL gea-sa gea-idda gea-sa gea-idda
LOC gea-s gea-in gea-s gea-inna
COM gea-inna gea-iguin gea-inna gea-iguin
ESS gea-nin gea-nin gea-nin gea-nin
Such patterns may be surprisingly tenacious over time, in spite of their

apparent ‘unnaturalness’ (Maiden 1992).
3.7.3 Semantic systematicity

We have singled out polarity effects as a distinct phenomenon because they
cannot be attributed to natural classes in any obvious sense. Nevertheless,
there are some examples where it is not implausible to seek a coherent
semantic motivation. Singulative number in the Nilo-Saharan language
Camus is an illustration of this principle, though not strictly speaking
syncretic under our definition (examples from Serzisko 1982: 197, citing
Heine 1981). The singulative endings, -i, -ni and -o mark the singular on
nouns whose default value is plural, e.g. l-tóto ‘grinding stones’ l-tóto-i
‘grinding stone’ (the choice among singulative endings is lexically
determined by the noun). These same endings are also used to mark the
plural of nouns whose default value is singular, e.g. l-me´rekec ‘ram’
l-me´rekec-i ‘rams’. Broadly speaking, there is a coherent semantic distinc-
tion between the two types of nouns: those that take singulative marking
describe things that occur in groups or are otherwise unindividuated
(‘tears’, ‘flies’, ‘stars’), while those that take plural marking are individu-
ated, e.g. animates (Heine 1981: 107–8). Thus, the endings -i -ni and -o
could be said to mark ‘unexpected’ number, whose specific value is con-
tingent on the semantics of the noun. However, for purposes of number
agreement (as occurs on verbs), the two types are not distinguished (Heine
1981: 106), so the semantic principle underlying the lexical distribution of
these endings is not active in the syntax of Camus.
In Kiowa, the category of ‘unexpected’ number is syntactically active,
involving a three-way number opposition with a somewhat unusual
semantic basis. Kiowa nouns are divided into a number of different classes
on the basis of how they mark number (both on themselves and in terms of
agreement). Watkins (1984: 92) distinguishes four major classes. By
default (as judged by productivity), class I is for animates, class II for
inanimate count nouns and class IV for inanimate non-count nouns. Class
III is a closed class, consisting of the words for ‘plum/apple’, ‘orange’,
‘tomato’ and ‘hair’. What concerns us is the category known as ‘inverse’
number, whose value depends on the class of the noun it is associated with.
See Corbett (2000: 159–62) for references on the origin of the term and
details of related languages with inverse number, as well as discussion of
Kiowa as marking the unexpected number value. Its value is plural for
class I, singular for class II and both singular and plural for class III. Class
IV does not take inverse number marking. The distribution of inverse
marking, as well as more conventional marking of singular, dual and
plural, is shown in third person intransitive subject verb prefixes illustrated
in (89), where e- is the inverse prefix, e˛- the dual prefix and gya- the plural.
As Watkins describes them, the noun classes can be characterised as
having an inherent number value. Thus, class I nouns are inherently
singular, class II inherently non-singular and class III nouns inherently
dual. Inverse marking occurs where the referential value of number does
not correspond to the inherent value. Broadly speaking, the inherent
number values of the individual classes is not too surprising: individuated
entities (animates) are preferentially singular, unindividuated entities
(inanimates) are plural, and non-count nouns do not mark number. The
class III pattern is unusual, in that it marks singular and plural together as
(89) Kiowa, intransitive third person subject marking (Watkins 1984: 82–9)
I Ø e- e˛-
IIa e- gya- e˛-
IIb e- Ø e˛-
III e- e- e˛-
IVa Ø gya- e˛-
IVb Ø Ø e˛-
IVc gya- gya- gya-
inverse; perhaps significantly, this is an unproductive class. Nevertheless,

one should perhaps avoid overemphasizing the semantic naturalness of
this system, in that the assignment of nouns to one class or another may be
idiosyncratic in some cases, e.g. ‘leg’ belongs to class I but ‘foot’ to class II.
Thus, there is some evidence (though not entirely conclusive) that inflec-
tion can mark the unexpected or marked value of a particular feature, the
specific value being determined in conjunction with the lexeme itself.
Similar accounts in terms of markedness have been proposed for classic
polarity, i.e. that what is being marked is relative markedness (Serzisko
1982: 198, citing Smith 1979). In Somali, for example, k- would indicate the
expected relationship of gender and number (masculines are singular and
feminines are plural), while t- would indicate the unexpected relationship
(namely the reverse). However, the naturalness of these relationships is less
obvious than in Kiowa, and still less than in Camus. The more this analogy
is extended the more tenuous it becomes, and it is doubtful whether this is a
useful perspective to take on patterns such as we find in Sami ((87)–(88)).
There is one more area where a semantic explanation has been suggested
for polarity effects, namely person-number marking. Thus, Foley notes the
frequency of 2SG/1NONSG syncretism in Papuan languages (found even in
free pronouns) and speculates that there may be a semantic connection
between the two values (Foley 1986: 72–3). He claims that this pattern does
not occur in languages with a distinct first inclusive and proposes that in
such instances a singular addressee potentially enters into the semantic
range of first non-singular, hence motivating the syncretism. However,
there are two possible objections to this interpretation. First, based on our
sample (Baerman 2002b ), there does not seem to be any noticeable cross-
linguistic preference for this particular polarity pattern. However, the
examples are too few to make any serious statistical claims. Second, as
we have noted (Chapter 3, x3.2.2), the presence of a distinct first inclusive is

not a barrier to syncretism of first and second person, at least outside of
New Guinea.
3.7.4 Summary
In spite of some tantalizing examples involving number marking (Camus,
Kiowa), there is little reason to believe that polarity effects on the whole can
be analysed as representing any sort of semantic natural class. On the other
hand, there is some good evidence that such patterns may be morphologic-
ally encoded. What is more, the best examples of morphological system-
aticity are probably among the worst examples of semantic systematicity.
This makes polarity effects – where their systematicity can be demonstrated –
compelling evidence that syncretism may be morphologically encoded.
3.8 The interaction of features

3.8.1 Introduction
In xx3.1–3.7 we looked at individual features. However, we also saw,
specifically for person in x3.2 and gender in x3.4 in relation to number,
that there may be interactions with other features. Rather than look at the
distinctions made within a feature (the ‘within-sets’ approach) we specific-
ally concentrate in this section on the interactions between features (the
‘between-sets’ view). In contrast to the within-sets view, it is also possible
to look at the environments in which syncretism occurs, as pointed out in
Chapter 2: x2.2.2. For example, Carstairs (1984) talks of ‘conditioning’
and ‘neutralised’ properties (features). As an example he cites the syncret-
ism between the dative and ablative plural in Latin, treating the distinction
between dative and ablative as ‘neutralised’ and number as the condition-
ing property. The work by Aikhenvald and Dixon (1998), which among
other things uses examples of syncretism to study dependencies between
grammatical features, also belongs to this tradition. Rather than consider-
ing specific values belonging to a feature, we distinguish here between
features serving as a context for a syncretism, similar to Carstairs’ ‘con-
ditioning’ property, and features which serve as the content of the syncret-
ism itself. In this way we abstract away from the issue of the particular
values involved and look at behaviour of the features to which they belong.
In order to do this we will look at a small sample, namely the thirty
languages from the Surrey Syncretisms Database (Baerman, Brown and
Corbett 2002a).
3.8.2 Syncretisms and their contexts

Distinctions in form for one feature may be distributed unevenly in the
presence of values of another feature. Hence, in Russian (90), singular
number can be seen as a context for the syncretism of dative and locative
cases (and dative, locative and genitive for class III).
(90) Case syncretism in Russian
class I class II class III class IV

‘table’ ‘room’ ‘bone’ ‘window’
GEN SG stola komnaty kosti okna

DAT SG stolu komnate kosti oknu
LOC SG stole komnate kosti okne
GEN PL stolov komnat kostej okon
DAT PL stolam komnatam kostjam oknam
LOC PL stolax komnatax kostjax oknax
This contrasts with the plural in Russian where the dative, locative and
genitive have distinct forms for all nouns. This is why number can be
viewed as a context for case syncretism. From the between-sets perspec-
tive, it does not matter which particular value provides the context for a
syncretism. It matters only that there is a difference in the pattern of
syncretism across different values of the feature which is the context.
As we saw for gender in x3.4, there is an interaction with number. There
can also be interaction with case. Gender is an inherent feature for nouns
but is an inflectional feature which is manifested on agreement targets,
such as verbs and adjectives. In the West Slavonic language Lower
Sorbian, as in other Slavonic languages, there is syncretism of masculine
and neuter gender in the oblique cases in the singular, as illustrated in (91).
This is the same pattern as for the Russian example we saw above in
Chapter 2 ((36) in Chapter 2: x2.4.3). As the masculine and neuter are
formally distinguished in the nominative case, they are viable grammatical
distinctions for Lower Sorbian. Consequently, failure to distinguish them
in the genitive, dative, instrumental and locative can be taken as instances
of syncretisms, as was true for the comparable Russian example we saw in
Chapter 2: x2.4.3. Furthermore, case here appears to be a context for
gender syncretism. (For the time being we leave the question of the
feminine singular locative and dative and return to it in x3.8.3.3.)
(91) The adjective dobry ‘good’ in Lower Sorbian (Stone 1993: 630)
masculine neuter feminine
NOM SG dobry dobre dobra

ACC SG dobry dobrego dobre dobru
GEN SG dobrego dobrego dobreje
DAT SG dobremu dobremu dobrej
INST SG dobrym dobrym dobreju
LOC SG dobrem dobrem dobrej
3.8.3 Nominal feature interactions

3.8.3.1 Typologies of interaction
Within Network Morphology, Brown (1998b) discusses the relationship
between number and case in terms of a feature dependency of case on
number. Specifically, the idea is that, ‘where case and number occur
together, a number feature may determine the number of case distinctions,
but not the other way round’ (Brown 1998b: 190). Of course, the statement
of a dependency relation between the two may be interpreted as implying
that number is causal in influencing the distinctions available.
The type of interactions seen in the previous section might lead one to
propose a hierarchical ordering of features.
(92) A possible hierarchy
gender syncretism > case syncretism > number syncretism
In other words, if a language has number syncretism, it must have case

syncretism and gender syncretism; or if it has case syncretism, then it must
have gender syncretism. But this hierarchy needs further refinement. First, if
a language has gender morphology, but not case morphology, for example,
then this hierarchy does not apply. In other words, it does not mean that if a
language exhibits gender syncretism, it must have case morphology, and
there must be syncretism somewhere within that. A second refinement
would be to restrict the hierarchy in (92) to purely inflectional features, so
that, for example, the generalisation about gender syncretism would not
apply to nouns. So it also has to be clear which morphological domain the
hierarchy in (92) applies to. Another problem with (92) is that there are
counter-examples to the implications from which it is constructed.
One set of counter-examples to the relation between case and number in

(92) can be found in the Konstanz universals archive. Universal 115 is
given as follows: ‘It is common for certain case distinctions to be neutra-
lized for a given number, while the opposite phenomenon perhaps never
occurs.’ This is attributed to Uspensky’s (1965: 210) reading of Greenberg
(1963: 103). Counter-examples were suggested by Hjelmslev (1935–7).
Koryak represents a counter-example to (92), because the singular and
plural are conflated in all cases except the absolutive in one set of nouns,
but not in all nouns (93). Furthermore, as there is still a number distinction
in the other cases in the second declension example in (93), this cannot
readily be treated as an example of number neutralization, as there is a set
of syntactic objects which distinguish number in these contexts. (Though
note that these do conflate dual and plural.)
There are a number of possible reactions to such counter-examples. One
of them is to argue that feature ranking plays a role but may differ from
language to language. Stump (2001: 239) develops a Feature Ranking
Principle whereby the specification of the ranking of features is language-
specific; this is discussed in Chapter 4: x4.4.4. As we will attempt to show,
there is something about the relation between the features of case, gender
and number in terms of syncretism which may not be just language-specific.
3.8.3.2 Exploring interaction

In this section we specifically look at cross-linguistic data on interaction
between features, analysing the thirty languages from the Surrey
Syncretisms Database. The sample in the database was not chosen ran-
domly; instead the languages were selected to explore the logical space of
possibilities, including extremes, and so the conclusions are indicative
rather than absolute. The method used involves abstracting away from the
particular feature values. Consider the partial Slovene paradigm in (94).
Dative and locative cases are distinct in the plural in Slovene, and so it
can therefore be considered to have these case values. The syncretism in
(94) is an instance of a more abstract relationship between case and
number, namely the relationship where number is the context and case is
syncretic. We have seen this pattern in other examples, of course. This can
be represented as in (95).
For Slovene any instance of number as context with case as syncretic can
be represented as in (95), thereby yielding one instance of this type of
syncretism for that language. We abstracted information of this type from
the Surrey Syncretisms Database (Baerman, Brown and Corbett 2002a).
(93) Koryak noun declensions
first declension ‘father’ second declension ‘papa’
singular dual plural singular dual plural
ABS en’pič en’piči-t en’piči-w appa appa-nte appa-w

ERG-INS en’piči-k en’piči-k appa-na-k appa-j«k
LOC en’piči-te en’piči-te appa-na-k appa-j«k
ABL an’peče-Nqo an’peče-Nqo appa-na-Nqo appa-j«ka-Nqo
TRANS an’peče-jp«N an’peče-jp«N appa-na-jp«N appa-j«ka-jp«N
DAT an’peče-N an’peče-N appa-na-N appa-j«k-«N
ADIT an’peče-jt«N an’peče-jt«N appa-na-jt«N appa-j«ka-jt«N
DES en’piči-nu en’piči-nu appa-na-no appa-j«čge-no
NARR-CAUS en’piči-kjit en’piči-kjit appa-na-kjet appa-j«-kjet
CONT en’piči-jite en’piči-jite appa-jeta appa-j«ka-jeta
(94) Locative/dative syncretism in a-stem

nouns Slovene
‘grove’
NOM SG dobrava
ACC SG dobravo
GEN SG dobrave
LOC SG dobravi
DAT SG dobravi
INS SG dobravo
(95) Feature interaction (number and case)
NUMBER CASE
Context Syncretic
As there are thirty languages in the database, the potential maximum

number of occurrences of this type of syncretism is thirty. But this does
not mean that the potential maximum number of occurrences in the
database of number as context or case as syncretic, is only thirty, as
these features may occur as contexts or syncretic in the presence of other
features, such as gender, for example.
If the syncretisms represented in the database are abstracted in this way
for all languages and word classes, the number of records in the database is
reduced to about one-quarter of the original size, indicating that this is a
powerful means for making generalizations. If we further select from these
records those which apply only to nominals, that is adjectives, nouns,
pronouns and numerals, then this figure is further reduced. We can visu-
alize the occurrences of the nominal features in syncretisms as a web of
binary pairs of connections between the context values and the syncretic
values for a given pair of features. We then rate these links between the
pairs of values as strong if they co-occur more than thirty-five times, as
medium if they co-occur more than fifteen times, and as weak if they
co-occur less than fifteen times.
In Figure 1 the dotted lines are the weak connections, the medium
connections are thin solid lines and the one strong connection is a thick
solid line. An instance of a weak connection is that between context gender
Figure 1. Web of nominal features and syncretism
and syncretic definiteness, for example. This means that there are very few
examples where gender values serve as a context for syncretism of definite-
ness values. It should also be noted that there may be multiple connections
where context is involved more than once, or where syncretic values are
connected more than once. For instance, where the dative plural of gender
III has the same form as the dative singular of gender III in Tsakhur, this
involves both gender and case as contexts for number syncretism. That is
why there are connections between context gender and context case, for
example. Equally, syncretism between masculine singular and feminine
plural in Somali involves syncretic number and syncretic gender
simultaneously.
The commonest tendency observed for nominal features in the data-
base is represented by the one strong connection in Figure 1: number is
the context and case values are syncretic. This suggests that case being
syncretic in the context of number may be commonly found within
languages which have syncretism. It should be noted that person and
definiteness have only weak links associated with them as either context
or syncretic. This is because these features do not occur with sufficiently
high frequency in nominals for any conclusions to be drawn. Gender can
only be an inflectional feature for nominals which are not nouns (i.e.
adjectives, pronouns and numerals), unless involved in possession con-
structions, and it is not a feature for all the languages in the database.
Consequently it does not have any strong links. It does have medium
strength links, and these all involve gender as syncretic, typically with
case and number as contexts, but also alongside number as a syncretic
feature. This fits with the observation in x3.4 that gender syncretism is
more the rule than the exception. The other medium strength links fill out
the rest of the logical space of connections between case and number as
syncretic or context, so that case can serve with number as a context for
syncretism, or itself have syncretism in the context of number, indicating
that the common tendency for case to be syncretic in the presence of
number is indeed just a tendency.
3.8.3.3 Constraints on nominal features

The hierarchy introduced in (92) can be interpreted as indicating the like-
lihood of a feature to show syncretism, with the left-most feature being
most likely to do so. Equally, the rightmost feature in (92), namely num-
ber, may have a greater tendency to act as the context for syncretism. This
tendency is reflected in the analysis of the languages in the Surrey
Syncretisms Database, as represented in Figure 1. The Russian example
(90) fits this hierarchy, with case distinctions (here locative and dative) lost
in the presence of singular number. In the Sorbian example in (91) the
combinations of singular number with genitive, dative, instrumental and
locative cases are contexts for the loss of distinction between masculine
and neuter genders. Such an approach allows for a hierarchical ordering of
the features involved. However, we find that in the Sorbian data in (91) it is
clear that there is also dative-locative syncretism in the feminine. In this
instance gender is the context and case is syncretic, the opposite way round
from what is predicted by (92). So, if we are to argue that ranking of
features alone is tenable, this leaves us with an apparent paradox. If we
accept that this is not possible, then are there any possible generalizations
to be made which do not have counter-examples ?
We shall propose that, yes, there are indeed two generalisations about
syncretism of case and gender which do appear to hold for the languages in
the database. Furthermore, it should be relatively straightforward to find
counter-examples, if they exist.
The notions of syncretic and context can be defined as follows: the
syncretic values of a syncretism S are the two feature values V1 and V2 of
the same feature, whereby V1 6¼ V2. In contrast, the context values of a
syncretism S are the two feature values V1 and V2 of the same feature
whereby V1 ¼ V2. Of course, a syncretism does not necessarily require a
context, whereas by definition it requires a content. Given these two

definitions we are now placed to make our generalizations.
(96) a. For a word-class W in language L

If values from the feature gender serve as context for a syncretism
in W, then values from the feature gender must be syncretic else-
where in W.
b. For a word-class W in language L
If values from the feature case serve as context for a syncretism in W,
then values from the feature case must be syncretic elsewhere in W.
A word class can range from an individual lexeme and its paradigm
through inflectional classes to nouns, nominals (a combination of nouns,
pronouns and adjectives) and verbs. In the database of thirty syncretism
languages there are no instances which violate the generalizations in (96).
It should be noted that these generalizations do not rank case and gender
relative to each other. However, it does follow from (96b) that, if case is a
context for gender syncretism, as in (91), then it must also be syncretic
somewhere within the same word class. Although this fact about case must
follow from (96b), in principle it has nothing to do with gender being
syncretic. Of course, if gender and number are the only other two features
in W, then in order to satisfy (96b) we will find in a language which
cumulates gender, case and number that case must necessarily be syncretic
in the presence of these, and in the Sorbian example in (91) we find that this
occurs in the feminine singular. Hence, we find that the apparent paradox
of (91) is not at all a paradox. It is the result of superimposing a require-
ment for the co-occurrence of certain features onto the generalizations in
(96) which are logically independent of this requirement. We should also
note from (96a) that the case syncretism in the feminine singular, where
gender is serving as (part of) the context, then logically requires gender to
be syncretic somewhere. And this is satisfied by the masculine/neuter
syncretisms of that paradigm, where case serves as context.
Even Koryak obeys the constraint (96b), as the word for ‘papa’ in (93)
has case syncretism between locative and the ergative-instrumental case,
which means that case is itself also syncretic somewhere within the noun
domain, as well as being the context for number syncretism.
3.8.4 Verbal feature interactions

The number of potential features involved in verbal inflection is greater
than for nominal inflection and this should mean that the space of logical
possibilities is greater. As we shall see in x3.8.4.2, there do appear to be a

number of constraints on which verbal features can be context and which
syncretic. We shall look at tense, aspect and mood (TAM) and agreement.
3.8.4.1 Verbal paradigms

In the classification of Booij (1996: 2), tense, aspect and mood are features
which are inherent to the verb. In contrast, person, nominal number and
gender are agreement features which encode information about the argu-
ments of the verb. Voice and negation (or affirmation) must also be
considered relevant to the verb as they involve the number of arguments
of the verb and the truth conditions of the predication.
(97) A verbal hierarchy
Agreement Syncretism > TAM Syncretism
Where verbal systems have both TAM marking and agreement marking
on the verb this appears to be a reasonably strong generalisation, as long as
it is understood that syncretism within tense, aspect and mood need not be
accompanied by agreement syncretism within the same word. For
instance, in the Tibeto-Burman language Limbu past and non-past are
syncretic in the presence of a second person singular subject and third
person object of either number.
(98) Tense syncretism in Limbu (van Driem 1987 : 370)
‘teach’
non-past past
2SG > 3SG kehu?ru kehu?ru

2SG > 3PL kehu?rusi kehu?rusi
In (98) we see that for the particular combinations involved there is no

syncretism of the agreement features, but there is syncretism of tense.
However, within the detailed study of thirty languages in the database
there is no example of a language which has some TAM syncretism in part
of its verbal paradigm and does not have syncretism of some agreement
feature somewhere within the verbal morphology. In Limbu, for example,
there is syncretism of the 2 > 1PL with 2PL > 1SG, which cross-cuts the
Figure 2. Web of verbal features and syncretism
TAM paradigm. Consequently, (97) may be true of the verbal domain

within languages but is not true at the word level. Furthermore, we cannot
easily make equivalent claims for the verbal domain as those in (96). More
concretely, although (97) appears to hold, we cannot say that TAM will be
a context, if it is syncretic.
3.8.4.2 Exploring the verb data

Figure 2 is a web of connections for the verbal domain and its features.
Again, we count connections which occur more than thirty-five times as
strong (bold) links between pairs of syncretic or context values. Medium
links occur less than thirty-five times, but more than fifteen, and weak links
(dotted lines) occur less than fifteen times.
Features which occur frequently may have a number of strong links for
both syncretic and context values. What is important are the differences
between the strengths of their links. If we consider person, it has three
strong links as syncretic: person often occurs as syncretic with TAM as a
context, with number as a context, and along with number when the
person feature’s values are syncretic. Person also often occurs as a context
when number is syncretic.
Number behaves in almost exactly the same way as person, having the
same strength of connections as person. The pattern of involvement of
gender on verbs is both expected in some respects and surprising in others.
As gender does not occur as much as person and number, it does not have
any strong links, and, as expected, its strongest links are medium ones
where its values are syncretic. In particular, one of its strongest links is with
TAM as a context. While we might expect gender to be a syncretic feature
par excellence, it contrasts with the other agreement features person and
number in never co-conspiring with TAM, voice or negation in a syncret-
ism pact (i.e. where two or more features are syncretic, as we see with
polarity effects). This can be summed up as in (99).
(99) TAM is not syncretic simultaneously with gender
We could assume that (99) is the result of lack of co-occurrence of

gender and TAM marking together. But in the Surrey Syncretisms
Database (Baerman, Brown and Corbett 2002a), ten of the languages
(one-third) have gender marking on verbs. Of these, there are four lan-
guages – Burushaski, Classical Arabic, Kashmiri and Somali – where
gender and TAM are realized together in forms which involve syncretism
of some feature. In Burushaski we can find examples where there is gender
syncretism in the presence of a particular tense, and this may also be
accompanied by person or number syncretism (i.e. more than one feature
being syncretic). In Classical Arabic the distinction between imperfect
jussive, indicative and subjunctive is neutralized in the second and third
person plural, while there is still a gender distinction. This gender distinc-
tion could be argued to be syncretic, because of the difference between
gender marking in the plural and singular, but even so, the mood distinc-
tions are neutralized entirely rather than syncretic. Also, in the second
person singular feminine the jussive and subjunctive are syncretic. Here,
then, we have an example of TAM syncretism in the presence of gender as
one of the contexts, although the interactions are complex. The overall
picture is that gender syncretism can occur in the presence of TAM (as
context), and that, more rarely, TAM syncretism can occur in the presence
of gender (as context), but TAM syncretism and gender syncretism do not
appear to co-occur. Finally, it could be argued that the apparent tendency
for TAM and gender on the verb not to syncretize together could be
attributed to the fact that gender marking on verbs is typically associated
with participles, diachronically or synchronically. However, it should be
noted that there are examples of gender marking in both synthetic and
periphrastic verbal constructions, suggesting that we cannot attribute this
tendency solely to this fact. It is also a fact which does not contradict (97),
which is a statement about the verbal system as a whole.
3.8.5 Interpreting the generalizations

In this section we consider why the generalizations in (96) and (97) appear
to hold.
3.8.5.1 Number, case and gender on nominals

One possible explanation for the difference between case and gender on the
one hand, and number on the other is the cardinality of the respective sets.
One interpretation which can be put forward is that those sets which show
a tendency for syncretism between values are ones which are high cardinality
(i.e. which typically are many-valued). While this may be an explanation
for case, it is less plausible for gender, which can easily occur as a two-
valued system. This is indicated by Corbett’s (2005) survey where 45 per
cent (50/112) of those languages with gender systems have two-valued
systems. Of course, we cannot tell whether there is syncretism once dis-
tinctions in a two-valued system are lost. Gender is also associated with a
wide variety of possible values which can be heterogeneous. Heterogeneity
of features can be manifested cross-linguistically in languages with two-
valued systems. For instance, two-valued gender systems could differ in
terms of the two different values involved.
In features which have a high cardinality there is also a possibility for
greater variation, that is heterogeneity, of feature values. One explanation
for the generalizations in (96) might be that the wide variety of possible
values for gender and case mean that there just is a greater probability of
syncretism occurring among these features. But there does seem to be
something in the nature of these features which lends them to syncretism,
irrespective of the overall heterogeneity of the features cross-linguistically.
Another possible explanation is grounded in the overall syntactic function
of the features. Gender, which in our treatment is an agreement feature, is
non-inherent and is imposed on agreement targets by a noun head or noun
phrase. Equally, case is imposed on nominal words by syntactic dependency
or constituency. In other words, the rules of syntax must clearly make
reference to the features of gender and case if they are present in a language.
Hence, a property of both these features is that they are contextual inflec-
tion, rather than inherent inflection in Booij’s (1996: 2) terms, which may
indicate that this is an important factor for syncretism.
3.8.5.2 TAM and agreement on verbs

As indicated, the generalization in (97) holds for the verbal domain but not
within the same word form. Verbal morphology may be a combination of
inherent TAM and marking of the agreement features for person, number
and gender, among other things. As we have seen, there is an expectation
that syncretism within the TAM system somewhere in the verbal domain
will entail syncretism within the agreement system. Again, it is the con-
textual inflection which is most likely to be affected, but the fact that
gender appears not to be syncretic simultaneously with TAM – on the
rarer occasions when TAM is syncretic – suggests another important
dimension to the interaction of feature sets on the verb, namely that gender
is the best representative of agreement features and is even less likely to
behave in a similar way to the non-agreement features than person or
number.
3.8.6 Summary
Looking at syncretism in terms of the interaction of the feature sets
involved proves particularly fruitful. Our examination of the database
shows that the use of hierarchies of the feature sets is insufficient at the
word level. However, the nominal features of case and gender are much
more prone to syncretism than number, and the agreement features on the
verb are also more susceptible to syncretism than the TAM features, if one
considers this question at the level of the word domain. This aspect of
syncretism suggests a particular paradigmatic design, which can be
obscured in particular instances, but the general tendencies are clear.
3.9 Conclusion
The preceding sections have illustrated something of the morphological
variety and cross-linguistic scope of inflectional syncretism. In particular,
we have considered the role played by semantics or feature structure:
patterns repeated across unconnected languages are likely to have a basis
in semantic properties of inflectional features that are shared across lan-
guages. In this respect different features yield different results. On the one
hand, patterns of number syncretism are quite similar across languages
(collapse of non-singular number distinctions), suggesting a basis in the
semantics of number itself (x3.5). Gender values, while themselves
language- specific, show a general tendency to fall together in non-singular
numbers (x3.4). On the other hand, patterns of tense-aspect-mood syncret-
ism are typically bound to the particulars of verbal morphology and seem
not to have a direct connection with meaning (x3.6). Other features show a
mixture of the two; for example, patterns of core case syncretism show
many similarities cross-linguistically, but non-core case syncretism shows

language-specific peculiarities (x3.1). It will be important to keep these
considerations in mind in the following chapter, as we consider different
formal models of syncretism, which often take the semantic structure of
features as a point of departure.
4 Formal representation
4.1 Introduction
It is a generally shared assumption that any adequate formal model of
morphology ought to take some account of inflectional syncretism. Most
investigators who have addressed syncretism overtly have taken this assump-
tion one step further: it is not enough to describe syncretism, one should also
constrain it. This goal is motivated by two factors. First, syncretism is
something of an aberration: by default we assume a one-to-one relationship
between morphosyntactic function and form, and syncretism is a violation of
this assumption (Carstairs 1987). Second, syncretism displays preferred
patterns, as we have seen throughout Chapter 3.
Ultimately, the morphological description of a particular syncretism
must contain two elements: (i) a list of the set of values which are syncretic,
and (ii) a way of associating this set with a form. To a large extent,
constraints on syncretism are a product of how these elements are treated.
For example, the syncretic set may be a natural class of values or simply a
stipulated disjunction, while the form itself may be defined over the whole
set, or defined in terms of one of the constituent members. In x4.2 below we
examine the inherent properties of different rule types, and in x4.3 we see
how these have been employed in particular accounts of syncretism.
4.2 Defining sets of values

4.2.1 Natural classes
Probably the most common approach to syncretic sets of values is to treat
them as a reflection of underlying feature structure. That is, feature values
are themselves composed of more basic sub-values, which can be grouped
into natural classes (a view first formalized by Bierwisch 1967). Given a
particular model of feature structure, one can predict what will be the possible
and impossible patterns of syncretism. Three basic types of feature structure
126
Formal representation 127
are possible: (i) flat, (ii) hierarchical and (iii) cross-classifying; the last term is
taken from Johnston (1997).
While cross-classifying feature structure is, in principle, unlimited in what
it can describe, flat and hierarchical feature structures have some inherent
restrictions. Therefore it will be revealing to contrast what can and cannot
be described by these two models. As an illustration we use two examples of
gender syncretism, from the Atlantic (Niger-Congo) language Noon and the
non-Pama-Nyungan Australian language Gaagudju, which reveal both the
power and the limitations of the two models of feature structure.
Noon distinguishes seven genders (see (1)). They are all distinct in the
singular, and show a convergent pattern in the plural (see Chapter 3:
x3.4.2): genders 1, 2 and 3 are syncretic, and genders 4, 5 and the diminu-
tive gender are syncretic.
(1) Noon gender (Soukka 2000: 66)
singular modifier plural modifier
1 w- c-
2 f- c-
3 m- c-
4 k- t-
5 p- t-
diminutive j- t-
animate y- ”-
The Noon paradigm can be easily described using hierarchical feature

structure such as in (2), made up of binary branching nodes, which have
been given the arbitrary labels A–H. Syncretism of genders 4, 5 and
the diminutive can be accounted for by a rule which associates the prefix
t- with node D, assuming that the animate form is itself overtly defined.
Syncretism of genders 1, 2 and 3 follows if the prefix c- is associated with
node A; that is, it is the general default. (Other orderings of the genders in
this hierarchical structure are also possible.)
Flat feature structure is illustrated in (3). This cannot be used to model
gender syncretism in Noon. The only locus for rules other than the daugh-
ter nodes is the single mother node (‘gender’), which will allow us to
describe one instance of syncretism (as an ‘elsewhere’ form), but not two.
(2) Hierarchical model of feature structure in Noon

A
H
1 2 3 4 5 6 diminutive animate
(3) Flat model of feature structure in Noon

gender
1 2 3 4 5 dimin. animate
The reverse situation is presented by the Gaagudju paradigms in (4),

repeated here from Chapter 3 (x3.4.3), where a flat feature structure will
work, but not a hierachical one.
(4) Gaagudju gender (Harvey 2002: 144, 157, 224–5)
realis verbs
present and
adjectives pronoun unmarked
tense object
present with first
first second (intransitive person
declension declension subject) subject
I (male) Ø na- naawu Ø arra-

II (female) njiN- njiN- ngaayu nj-dja- nji-rra-
III (plant) ma- naN- ngaayu ma-ya- ma-ra-
IV (residue) gu- naN- ngaayu nj-dja arra-
By assuming the flat feature structure in (5), all the patterns can be
described by simple underspecification. That is, the non-syncretic forms
are fully specified for gender, while the syncretic forms naN-, ngaayu,
nj-dja- and arra- are ‘elsewhere’ forms, unspecified for gender, and so are
used wherever there is no more specific form.
(5) Flat model of feature structure in Gaagudju
gender
I II III IV
Now let us see how this might be represented with a hierarchical feature
structure, consisting of binary branching nodes as suggested above for
Noon. Example (6) represents what seems to be a plausible arrangement of
values; for convenience, we can label the node which joins III and IV
‘inanimate’, and the node which joins II, III and IV as ‘non-male’.
(6) Hierarchical model of feature structure in Gaagudju
gender
non-male
inanimate
plant
I II IV III
By simply underspecifying for lower nodes, we can describe the patterns

found in the nominal word classes: III/IV syncretism results from a rule
specifying an ‘inanimate’ form, while II/III/IV syncretism results from a
rule specifying a ‘non-male’ form. This accounts for the nested pattern that
these syncretisms display, since each higher node subsumes the ones under-
neath. This structure can also be used to describe one of the patterns found
with verbs, namely the syncretism of genders II and IV found in the present
tense intransitive subject forms. This will result if the inanimate node is left
unspecified, as illustrated in (7).
(7)
node form
gender Ø
non-male nj-dja
plant ma-ya-
However, this accounts for only one of the two syncretic patterns found
with verbs; I/IV syncretism cannot be accommodated with this hierarchy.
Only values which are adjacent on the hierarchy can be described as
syncretic, since any unspecified value will default down the hierarchy to
the next most specific value. This translates into a graphic rule of thumb:
there is a single linear order of values, and only values which are adjacent
can be described as syncretic (Chvany 1986, Johnston 1997). In this case,
no single linear order will bring all the syncretic values adjacent to each
other; the order in (8), which accommodates most of the patterns, still
leaves genders I and IV separated from each other:
(8)
I II III IV
Thus, we have seen the constraints inherent in the two models. If feature
structure is construed as flat, there can be only one syncretic form per
paradigm, namely the elsewhere form. Hierarchical feature structure allows
multiple syncretisms within a single paradigm but restricts the number of
ways that feature values may be combined in different paradigms. Of course,
hybrid structures are possible as well. For example, one could recast the
model of Noon gender along the lines of (9), where nodes A, B and C/D
constitute a hierarchy, while the structure beneath nodes C and D is flat.
Such a hybrid model simply combines the properties of hierarchical and flat
feature structure and does not add any new descriptive potential.
Cross-classifying feature structure, by contrast, is potentially uncon-
strained. For example for Gaagudju we might posit a system of four feature
values: ‘non-male’, ‘inanimate’, and two additional values, which we can
provisionally term ‘value x’ and ‘value y’. The individual gender values can
thus be portrayed as composites of these four sub-values, as shown in (10).
(9) Hybrid model of feature structure in Noon
C D
animate 1 2 3 4 5 diminutive
The individual syncretic patterns result from rules which make reference
to these component values. The number of syncretic patterns that can be
described is constrained only by the number of values employed, which is
not subject to any inherent limit.
(10)
non-male inanimate value x value y
I þ
II þ þ
III þ þ
IV þ þ þ þ
4.2.2 Unnatural classes

The other approach to defining syncretic classes of values is simply to
stipulate them. The resulting set of values thus represents an unnatural
class, in that its members share no common feature other than subjection
to that stipulation; a recent work employing this approach within Lexical-
Functional Grammar is Dalrymple and Kaplan (2000). Obviously, such a
technique can describe any possible set of values. In practice, the full power
of stipulation is seldom employed; most investigators impose some con-
straints, whether overt or covert. For example, Brown (1998b: 87) requires
that the values be of the same feature or features, e.g. {nominative,
accusative} or {second person singular, first person plural}. Such a restric-
tion seems to be tacitly in place for some other approaches which employ
unnatural classes, e.g. Stump (2001) or Zwicky (2000).
A somewhat less direct way of stipulating sets has emerged in Optimality

Theory. For example, in Kiparsky (2001), the syncretism of genitive and
accusative in singular nouns in Finnish is accounted for by morphological
markedness constraints which block the use of the expected case endings for
the realization of an underlying accusative. Instead, an underlying accusa-
tive is forced to use the genitive ending. Thus, the constraints and their
ranking conspire to define a set consisting of the accusative and genitive.
4.2.3 A note on polarity effects

In the preceding sections we have concentrated on syncretic patterns
confined to values of a single feature. Polarity effects represent a special
problem: since they involve combinations of values which are not directly
related in terms of feature structure, it would appear at first glance that
they could only be described by overt stipulation. However, some investi-
gators have proposed that inflectional rules can make reference to abstract
properties of feature structure, which allow some polarity effects to be
systematically represented in featural terms. The crucial notion here is
markedness: as Serzisko notes, some polarity effects can be described in
terms of markedness (1982: 194–6). For example, the default system of
gender-number marking in Somali (11a) can be viewed in terms of mark-
edness coordination (11b): k- is used where the markedness values of both
features is the same, and t- is used where the values differ, on the assump-
tion that feminine and plural are the marked values.
(11)
a. SG PL b. unmarked marked
M k- t- unmarked U/ U U/M
F t- k- marked M/U M/M
(based on Serzisko 1982: 185, 195)
If the features are construed as being hierarchically structured, the values

break down as in (12a); if cross-classifying, they break down as in (12b).
Both models now lend themselves to a more abstract representation
using variables. With hierarchical features, the feature itself is made a
variable (13a), with cross-classifying features it is the value which is
made a variable (13b).
(12)
a. hierarchical b. cross-classifying
feature structure feature structure
M SG M SG F PL
F SG þF F SG þ F PL
M PL þPL M PL F þPL
F PL þF þPL F PL þ F þPL
(13)
a. hierarchical b. cross-classifying
abstract feature abstract feature

structure structure
M SG M SG aF aPL
F SG þa F SG aF aPL
M PL þa M PL aF aPL
F PL þa þa F PL aF aPL
The structure in (13a) now allows us to describe t- as the form which

represents a single degree of markedness (þa), while k- is an elsewhere
form (such an argument is advanced by Béjar and Hall 1999). The struc-
ture in (13b) allows both forms to be explicitly described: k- is the form
used where the value for both features matches (aF aPL), and t- is used
where they do not match (aF aPL).
Thus, if the use of variables is allowed, some instances of polarity can be
attributed to feature structure. However, as pointed out in Chapter 3
(x3.7), it is doubtful whether many examples can be profitably seen in
terms of neat markedness relationships.
4.3 Symmetrical versus directional rules

As we have defined it, syncretism involves a set of values associated with a
single form. Given this definition, the question of how this association is
effected remains open. In principle, there are two options: (i) the form is
associated with the set as a whole, as in the schematic rule in (14a), or
(ii) the form is associated with one of the component values and is ‘bor-
rowed’ by the other members of the set, as in the schematic rule in (14b).
(14) a. there is a set of values {a, b} which have the form x
b. there is a set of values {a, b} which have the same form; the
form of a is x
It will be useful (borrowing the terminology of Stump 2001) to distinguish
these as ‘symmetric’ rules (14a) versus ‘directional’ (14b) rules. On the face of
it, symmetric rules are conceptually simpler, since they have the same struc-
ture as an ‘ordinary’ morphological rule: there is an element in morphosyn-
tax which is directly associated with a form. Directional rules, by contrast,
entail a two-step process. The most familiar breed of directional rule is the
‘rule of referral’ (Zwicky 1985). In describing (14b) above, a rule of referral
would state that the form associated with the value b is the same as the form
associated with the value a (that is, b refers to a for its form). One of the main
reasons for assuming directional rules is to account for directional effects,
that is those contexts where the syncretic form looks as if it is ‘belongs’ to one
of the component values, numerous examples of which have been seen in the
previous sections (especially involving case and person).
However, it has been argued that directional effects can be described
without directional rules, so that, for reasons of formal economy, they
should be rejected. We can illustrate a directional analysis, as well as a non-
directional counterproposal, with a familiar example from Latin, shown
in (15). Nominative and accusative are syncretic in neuter nouns and distinct
in masculine nouns (at least in the singular). In the singular of the second
declension, the form taken by the syncretic nominative/accusative of the
neuters, -um, is the same as that of the distinct accusative case of the
masculines.
(15) Latin second declension
neuter ‘war’ masculine ‘slave’
NOM SG bell-um serv-us

ACC SG bell-um serv-um
GEN SG bell-i serv-i
DAT SG bell-o serv-o
ABL SG bell-o serv-o
Thus, it looks as if the neuter nominative singular has taken the form of the
accusative. Using a directional rule, this apparent ‘borrowing’ can be
incorporated directly in the analysis. First, one can assume two general
rules that apply to all second declension nouns: the nominative singular
ending is -us (16a), and the accusative singular ending is -um (16b). For
neuters, there is a rule stating that the nominative singular takes the form
of the accusative (16c).
(16) a. NOM SG ¼ stem þ -us
b. ACC SG ¼ stem þ -um
c. NOM SG in neuter ¼ ACC SG
Because the domain of the rule in (16c) is more specific than the domain of
the rule in (16a), in that it additionally contains reference to gender, the
rule in (16c) takes precedence by Panini’s principle (otherwise known as
the ‘elsewhere condition’, ‘blocking’ and the ‘Subset principle’ – we will use
the name ‘Panini’s principle’).
The morphological description above contains two rule types, one
which associates a morphosyntactic value with a form (16a, b), and one
which associates one morphosyntactic value with another (16c). In the
terminology introduced by Zwicky (1985), the former type is a ‘rule of
exponence’, the latter a ‘rule of referral’. Rules of exponence state the
relationship between a morphosyntactic value and a form, while rules of
referral state the relationship between morphosyntactic values.
Some linguists have suggested that rules of referral are unnecessary and
should be dispensed with for reasons of formal economy. Thus, Zwicky
(2000) suggests that all syncretism can be described as symmetrical syn-
cretism, with directional effects derived as a by-product of underspecifica-
tion. We can illustrate his proposal with the same Latin data from (15).
First, declare that nominative and accusative singular form a class, which
is assigned the index ‘X’ (17a). This class is associated with the ending -um
(17b). In (17c) a more specific rule is introduced, stating that the nomina-
tive singular of the masculines has the ending -us, which takes precedence
over the rule in (17b) by Panini’s principle.
(17) a. {NOM SG ¨ ACC SG} ¼ X
b. X ¼ stem þ -um
c. NOM SG in masculine ¼ stem þ -us
Thus, under a symmetrical analysis, the specifically nominative singular
masculine ending -us is opposed to the default nominative/accusative
ending -um.1 What looks like the spread of the accusative form to the
nominative is simply the emergence of the unmarked form.
This alternative analysis will work for examples of what Stump (2001)
calls unidirectional syncretism, that is directional syncretism where the
directional effect seems to move in only one direction (accusative !
nominative in the above example). However, Stump (2001) identifies
another type, bidirectional syncretism, where the directional effect seems
to move in two directions. This does not readily lend itself to the same
analysis. It will be useful here to distinguish between two types, which we
will call convergent bidirectional syncretism and divergent bidirectional
syncretism, to be defined below. Each one presents a distinct problem for
Zwicky’s model.
4.3.1 Convergent bidirectional syncretism

Under convergent bidirectional syncretism, there is a feature value x which
takes the form associated with feature value y in some contexts, and in
other contexts takes the form associated with feature value z. Some
examples have already appeared in Chapter 3, involving the accusative in
Russian and in Bonan, and the ergative in Lak (all in x3.1.2) and the
second person singular in Gujarati (x3.2.3). Bonan presents an especially
clear and straightforward example, and this is what we analyse below.
Recall that in Bonan (the paradigms are given again in (18)), the accusative
case does not have a distinct form: in nouns it is syncretic with the genitive,
while in pronouns it is syncretic with the dative-locative (a single case in
Bonan).
(18) Bonan (Todaeva 1997: 35)
noun ‘foliage’ pronoun ‘he’
NOM labčoN-Ø ndžaN-Ø

GEN
ACC
labčoN-ne
labčoN-ne
ndžaN-ne
ndžaN-de
DAT-LOC
ABL
labčoN-de
labčoN-se
! ndžaN-de
ndžaN-se
INS-COM labčoN-Gale ndžaN-Gale
1
One could construe the arbitrary class ‘X’ as some kind of natural class, e.g. ‘direct case’, but
this does not affect the argument.
Seen in directional terms, it looks as if the accusative takes the form of the
genitive in nouns, while in pronouns it takes the form of the dative-
locative. The following represents a possible description using directional
rules. There are two rules of referral, shown in (19). In nouns, the accusa-
tive takes the form of the genitive (19a). With pronouns, the accusative
takes the form of the dative-locative (19b).
(19) a. ACC in nouns ¼ GEN
b. ACC in pronouns ¼ DAT-LOC
The relevant rules of exponence are shown in (20); note that no accusative
forms are defined, because these are derived by the rules of referral in (19)
(20) a. NOM ¼ stem þ -Ø
b. GEN ¼ stem þ -ne
c. DAT-LOC ¼ stem þ -de
d. ABL ¼ stem þ -se
e. INS-COM ¼ stem þ -Gale
The interaction of the rules of referral with the rules of exponence is
graphically represented in (21). In nouns, the rule of referral in (19a) causes
the form of the genitive to be extended to the accusative, while in pro-
nouns, the rule of referral in (19b) causes the form of the dative-locative to
be extended to the accusative.
(21)
noun ‘foliage’ pronoun ‘he’

NOM ¼ labčoN-Ø NOM ¼ ndžaN-Ø
GEN ¼ labčoN-ne GEN ¼ ndžaN-ne
ACC ¼ ACC ¼
DAT-LOC ¼ labčoN-de DAT-LOC ¼ ndžaN-de
ABL ¼ labčoN-se ABL ¼ ndžaN-se
INS-COM ¼ labčoN-Gale INS-COM ¼ ndžaN-Gale
Now let us see how we might describe the same phenomenon without
directional rules. The syncretisms are represented in (22) as the classes ‘X’
and ‘Y’. In (23), these symmetrical rules define the syncretic forms.
(22) a. {ACC ¨ GEN} ¼ X

b. {ACC ¨ DAT-LOC} ¼ Y
(23) a. X ¼ stem þ -ne

b. Y ¼ stem þ -de
Obviously, this is an incomplete description, because nothing yet has been

said about the difference between nouns and pronouns. As it stands, the
rules for X and Y conflict for the definition of the accusative form. This
conflict is not resolved by Panini’s principle, because neither rule is more
specific.
Two options are available for clarifying the distribution of X and Y. One
would be to incorporate rule ordering, with nouns and pronouns exhibit-
ing different ordering for the rules involving X and Y, as sketched in (24):
the rule for X precedes the rule for Y in nouns, while the rule for Y precedes
the rule for X in pronouns.
(24)
rule ordering for nouns rule ordering for pronouns

i. X ¼ stem þ -ne i. Y ¼ stem þ -de
ii. Y ¼ stem þ -de ii. X ¼ stem þ -ne
In each case, the rule conflict is resolved in favour of the prior rule.
An obvious objection to this approach is that it substitutes one formal
device (rule ordering) for another (rules of referral). Whether this is a
serious problem depends on the status one gives to rule ordering; for
example, in approaches such as Stump’s (2001) Paradigm Function
Morphology and Corbett and Fraser’s Network Morphology (1993),
rule ordering is absent.
The other possibility would be to incorporate information about the
lexical class within the set of syncretic values (thanks to Jonathan Bobaljik
for pointing out this option). In (25), the accusative is specified as nominal
or pronominal.2
2
Of course, one could leave one of these rules underspecified for lexical class; the argument
remains the same, though, since at least one rule will have to incorporate reference to lexical
class.
(25) revision of (22)
a. {ACC noun ¨ GEN} ¼ X

b. {ACC pronoun ¨ DAT-LOC} ¼ Y
Technically, this would yield the correct output, but such rules seem to
hold to the letter but not the spirit of a symmetrical analysis, in that they
encode an obvious asymmetry. For example, (25a) states that form X
serves by default for the genitive, and for one class of lexemes it is also
used for the accusative. In its own fashion, this states that form X is
primarily a genitive form, and secondarily an accusative form.
In summary, the existence of convergent bidirectional syncretism makes
certain demands on a model of morphology: they require rule ordering or
directional rules (or something very much like them).
4.3.2 Divergent bidirectional syncretism

As we have seen above, symmetrical rules can be used to describe
convergent directional syncretism, provided other elements in the mor-
phological model comply. As for divergent bidirectional syncretism, sym-
metrical rules simply cannot be used to describe it in any systematic
fashion. Under divergent bidirectional syncretism, there is a feature
value x which takes the form associated with feature value y in some
contexts, while in other contexts y takes the form associated with x. For
an illustration, we can return to the Latin example from (15), adding
further data as shown in (26).
In addition to the default masculine and neuter types of the second
declension, there are a few nouns (the most prominent being vulgus
‘crowd’, virus, ‘poison’ and pelagus ‘sea’) which display an accusative in -us
(Neue and Wagener 1902: 972).3 If we include these nouns in the picture,
3
The origin of this type is mixed. Pelagus is borrowed from the Greek s-stem pelagos, where
-os is the final part of the stem, and not an inflectional ending. Clearly, however, when
borrowed into Latin, it was interpreted as an ending, since pelagus was assigned to the
second declension (with the ending -us), and not to the third declension s-stem type, where
-us is instead the stem-final element (e.g. genus ‘nation, race’ and corpus ‘body’, where stem-
final -s is realized as -r when followed by endings, as in the genitive singular forms gener-is,
corpor-is). This justifies our treatment of the -us in pelagus and the -us in servus as
representing the same ending. Other borrowings from Greek s-stems followed this pattern
on occasion, such as c etus ‘large sea animal, whale’ and chaus ‘chaos’ (Neue and Wagener
1902: 502–4). Vulgus and virus are native Latin items, and the origin of their exceptional
declension is not known.
(26) Latin second declension
default neuter default masculine accusative in -us

‘war’ ‘slave’ ‘crowd’
NOM SG bell-um serv-us ! vulg-us

ACC SG
GEN SG
bell-um
bell-i
serv-um
serv-i
vulg-us
vulg-i
DAT SG bell-o serv-o vulg-o
ABL SG bell-o serv-o vulg-o
then the Latin second declension is not simply an example of unidirectional

syncretism, but rather has two mirror-image patterns: in the default neuter
type, the nominative looks like the accusative, and in the small class repre-
sented by vulgus, the accusative looks like the nominative.
Using directional rules, this sort of pattern does not differ in principle
from unidirectional syncretism. To the rules from (16) we simply add a
fourth rule (27d), stipulating that the accusative takes the form of the
nominative for this small class.
(27) a. NOM SG ¼ stem þ -us
b. ACC SG ¼ stem þ -um
c. NOM SG in neuter ¼ ACC SG
d. ACC SG in ‘vulgus’ type ¼ NOM SG
For symmetrical rules, divergent bidirectional syncretism imposes more
constraints than does convergent bidirectional syncretism. Rule ordering
is no longer an option, because the same values are syncretic both in the
neuters and in the ‘vulgus’ type. For example, in (28) we represent the
This type is also defective, lacking plural forms, except for the occasional pelag e, whose
ending is transparently Greek, unincorporated into the Latin declensional system. All of
these nouns show a tendency to be reinterpreted as masculine nouns of the servus type, with
nominative -us and accusative -um. One interesting variation occurs in Late Latin (sixth
century), where pelagus is reinterpreted as masculine without altering its declension pattern:
the example furentem pelagus ‘raging sea’ (from the Variae of Cassiodorus, cited in Neue
and Wagener (1902: 503)) has the agreeing participle with the non-syncretic masculine
accusative ending -em. This shows that nominative/accusative syncretism in the singular
was not necessarily tied to neuter gender.
syncretic nominative/accusative forms, using the index ‘X’ to represent

{NOM SG ¨ ACC SG}.

b. X in neuter ¼ stem þ -um
c. X in ‘vulgus’ type ¼ stem þ -us
But what about masculines, which require both -us and -um endings? The
description of a masculine noun would need some version of both rule
(28b) and (28c). In (29) we show what these rules would look like with
reference to the masculine declension added.
(29) Revision of (28)
a. {NOM SG ¨ ACC SG} ¼ X
b. X in {neuter ¨ masculine} ¼ stem þ -um
c. X in {‘vulgus’ type ¨ masculine} ¼ stem þ -us
But this is unworkable: rules (29b) and (29c) conflict within the masculine
declension. Rule ordering cannot resolve this, because either rule would
bleed the other completely.
The only way a symmetrical analysis can represent this pattern is to treat
some aspect of it as accidental. One possibility would be to treat the
syncretism of nominative and accusative as systematic, but to treat the
identity of -us in the masculines and in the ‘vulgus’ type as accidental, by
assuming two distinct but homophonous endings -us1 and -us2, as in (30).

b. X in neuter ¼ stem þ -um
c. NOM SG in masculine ¼ stem þ -us1
d. X in ‘vulgus’ type ¼ stem þ -us2
This analysis treats the paradigms as an example of unidirectional syncre-

tism (between instances of -um forms), while the apparent identity of the
endings in servus and vulgus is treated as an accident. In effect, this
amounts to a denial that there is such a thing as divergent bidirectional
syncretism.
Alternatively, one might treat the identity of endings across different
declension classes as systematic but treat the different instances of the same
case syncretism as accidental. That would be the consequence of recapitu-
lating the analysis sketched above in (25), where the syncretic set of values
includes reference to the lexical class. This is shown in (31), where the class
X describes the distribution of all instances of ‘stem þ -um’, and class Y

describes all instances of ‘stem þ -us’.
(31) a. {NOM SG neuter ¨ ACC SG} ¼ X

b. {NOM SG masculine ¨ ACC SG ‘vulgus’ type} ¼ Y
c. X ¼ stem þ -um
d. Y ¼ stem þ -us
On this analysis, the fact that both X and Y involve syncretism of nomi-
native and accusative is treated as accidental.4
Because of what they imply for the modelling of syncretism, examples of
divergent bidirectional syncretism are particularly important, so it is worth
considering some additional data. So far, the only examples which seem to
have been adduced in the literature are those found in Stump (1993, 2001),
namely the accusative/dative singular in Old Icelandic, the genitive singu-
lar/nominative plural in Russian and the first singular/third plural in
Romanian. Some of these examples have been subject to criticism by
Feldstein (2003) and Wunderlich (2004), who argue that these patterns
are accidental and hence not represented in the morphology. However, the
Latin example above cannot be too readily dismissed, and below we
adduce some further examples which appear to be good examples of
divergent bidirectional syncretism.
Classical Arabic has three cases: nominative, genitive and accusative
(see (32)). In the default pattern (singulars and the broken plurals, which
are formed by an alternation in the vocalic/syllabic pattern) all three cases
have distinct endings: -u, -i and -a. In the so-called sound plurals (formed by
suffixation), genitive and accusative are syncretic, marked by the ending -i:,
which corresponds to the distinct genitive of the default type.5 The
so-called diptotic declension pattern (displayed by certain adjectival
stems, some broken plurals and some personal names; the name reflects
the fact that there are two rather than three case forms) likewise has a
syncretic genitive/accusative, but the ending is -a, corresponding to the
distinct accusative of the default type. The diptotic pattern is found only
with indefinites; the definite form and the construct form (head of an
adnominal construction) revert to the default pattern.
4
This is unfortunate, in as much as there is an important generalization that applies to all
nouns with nominative/accusative syncretism in the singular, namely that they belong to the
neuter gender, taking neuter agreement (see the preceding footnote).
5
The dual likewise displays genitive/accusative syncretism; e.g. nominative dual, construct
state mu’min-a:, genitive/accusative mu’ min-ay ‘believers.’
(32) Classical Arabic noun declension (Fischer 1997: 196)
triptotic (default) pattern

plural diptotic
‘believers.PL’ ‘believer’ ‘black one’ ‘black one’
(construct) (construct) (construct) (indefinite)
NOM mu’min-u: mu’min-u ’aswad-u ’aswad-u

GEN mu’min-i: mu’min-i ’aswad-i ’aswad-a
ACC mu’min-i: mu’min-a ’aswad-a ! ’aswad-a
The plural endings -u: and -i: are assumed to derive from lengthening of
the corresponding singular endings (Kienast 2001: 143). The origin of the
diptotic pattern is unclear; Kienast (2001: 142) cites Brockelmann’s
(1908–13) theory that it started among personal names: some names
ended in -u (e.g. Iazi:du) and some in -a (e.g. Sˇammara), and these were
reinterpreted as fragments of a case paradigm, which was then fleshed out.
This proposal assumes that the diptotic endings -u and -a are etymologi-
cally distinct from the corresponding case endings -u and -a, but that at
some later point the two sets were equated with each other.
As with the Latin example in (26), the analytical problem here is that, if
we say that both -i and -a are genitive/accusative, how can the two endings
be combined in a single paradigm? On the other hand, using directional
rules, one could say that the accusative takes the form of the genitive in
sound plurals, and that the genitive takes the form of the accusative in
diptotic nouns.
A further example is found in the Pama-Nyungan language Diyari. Dual
and plural nouns and pronouns, and female personal names, have an
accusative ending -n5a which is distinct from the absolutive (see (33)). In
male personal names the ending -n5a is found in both the absolutive and the
accusative. A zero ending (the bare stem) is found in the ergative/absolu-
tive of non-singular pronouns, the ergative of non-singular nouns, and the
absolutive/accusative of singular nouns.
As with the previous examples, the problem faced by symmetrical rules
is posed by the overlapping range of the inflectional forms. The zero
ending ranges across all three core cases (ergative, absolutive, accusative),
while -n5a is found in both the absolutive and the accusative. Using symme-
trical rules, we would associate {ERG, ABS, ACC} with -Ø and {ABS, ACC} with -n5a.
Since the two rules overlap in the absolutive and accusative, they
cannot coexist in the same paradigm. Again, directional rules are the
(33) Diyari noun declension (Austin 1981: 47–50, 61)
I II III IV V
ERG -(ya)li -li -Ø -ndu -li

ABS -Ø -Ø ! -Ø -ni -n5a
ACC
LOC
-Ø
-n5i
-n5a
-Nu
-n5a
-Nu
-n5a
-n5aNu
! -n5a
-Nu
ALL -ya -Nu -Nu -n5aNu -Nu
DAT -ya -n?i -n?i -n5aNka -n?i
ABL -ndu -Nundu -Nundu -Nundu -Nundu
I singular nouns
II non-singular nouns, non-singular third person pronouns, singular pronouns
III non-singular first and second person pronouns
IV female personal names, singular pronouns6
V male personal names
only device that can give -Ø and -n5a a unified representation across all of
the paradigms. One possible analysis is to identify -Ø as the absolutive
ending and -n5a as the accusative ending. In type I, the accusative takes the
form of the absolutive, in type III, the ergative takes the form of the
absolutive, and in type V, the absolutive takes the form of the accusative.7
(In type IV a distinct absolutive ending is found.)
4.3.3 Symmetrical versus directional rules: a summary.

Zwicky (2000) has argued that symmetrical rules are sufficient to generate
the surface effect of directionality, thus making directional rules unneces-
sary. However, we have shown that the rejection of directional rules brings
real consequences in the range of facts that can be described (further argu-
ments may be found in Evans, Brown and Corbett 2001, using somewhat
6
Singular pronouns, besides the third person non-feminine, display this pattern in the
ergative, absolutive and accusative, while in the remaining cases they behave like non-
singular pronouns, with the element -ka- preceding the endings. The third person non-
feminine behaves essentially like a non-singular noun, with ergative n5ulu, absolutive n5awu,
accusative n5in5a, and remaining cases based on the stem n5u ka- (Austin 1981: 61).
7
A plausible alternative in this case would be to combine symmetrical and directional rules
by viewing -Ø as the default core case ending. Types I–IV, then, follow a familiar split
ergative pattern, with a distinct ergative in some paradigms and a distinct accusative in
others. In this case, only type V requires a directional rule. In either case, though, directional
rules are required.
different evidence). Although symmetrical rules can describe unidirectional

syncretism, convergent directional syncretism can be described only by
substituting another formal device (rule ordering), or by combining lexical
and morphosyntactic information within a single syncretic value, which is
undesirable on at least some approaches. Divergent bidirectional syncretism
cannot be systematically described with symmetrical rules, requiring that
some element of the pattern be treated as accidental.
4.3.4 Ranked constraints as an alternative to directional syncretism

Alongside Zwicky’s (2000) discussion, the argument that directional
effects can be derived without using directional rules has come from the
perspective of Optimality Theory (OT). The issue is addressed most
directly by Wunderlich (2004), who offers an OT alternative to directional
rules. Most relevant is his treatment of what we have termed convergent
syncretism, for which he analyses the nominative/accusative genitive/
accusative alternation of Russian, described in x3.1.1 and x3.1.2. To
recapitulate: for most declension classes, the accusative case does not
have a distinct form. In most classes, in inanimate nouns it appears to
take the form of the nominative, and in animate nouns it appears to take
the form of the genitive. For the present purposes, it will suffice to consider
the singular of o-stems (which show the animacy-based alternation) and
i-stems (which do not), as shown in (35).
(34) Nominative/accusative genitive/accusative

alternation in Russian
a.‘table’ b.‘student(M)’ c.‘mother’
NOM stol student mat0

ACC stol studenta mat0
GEN stola studenta materi
The feature values that Wunderlich proposes for the morphosyntactic

cases are given in (35). The nominative is simply unmarked. The other
cases are rendered in terms of semantic roles: the accusative is [(þhr)v]
(‘there is a higher role (verbal)’), which expresses its subordinate role
within a transitive verbal clause, while the genitive is [(þhr)N] (‘there is a
higher role (nominal)’), which expresses its subordinate role within a
nominal clause.
(35)
case name feature value
nominative []
accusative [(þhr)v] i.e. ‘there is a higher role (verbal)’
genitive [(þhr)N] i.e. ‘there is a higher role (nominal)’
The relevant case endings are given in (36). The i-stem genitive -i is fully
specified in terms of the underlying morphosyntactic features. The o-stem
ending -a is underspecified: the value [(þhr)], which omits reference to
nominal or verbal contexts, represents genitive/accusative together as a
natural class. The default ending, -Ø, has no feature specifications.
(36)
o-stem i-stem
ending feature value ending feature value

-Ø [] -Ø []
-a [(þhr)] -i [(þhr)N]
The link between the underlying morphosyntactic features and the endings
is effected by three ranked constraints:
* Compatibility: the categorical specification of input and output
must match. That is, genitive case (with a nominal specification)
cannot be used where the input requires accusative case (with a
verbal specification), and vice versa.
* *(þhr)/V inanimate: do not mark accusative case for inanimates.
* Max (þhr): if the feature (þhr), common to the accusative and
genitive, is present in the input, it must be realized in the output.
The interaction of the above elements is shown in the tableaux in (37)–(39). In
animate o-stems, the constraint Max (þhr) blocks the ending -Ø, because it is
not specified for (þhr), thus causing the genitive/accusative to be selected. In
the inanimate o-stems, however, the genitive/accusative is blocked by the
constraint *(þhr)/V inanimate, so the unmarked nominative is selected, in
spite of its violation of Max (þhr). In the animate i-stems, Compatibility
blocks the use of the overtly genitive ending -i for the input accusative, so the
nominative is selected; the issue of animacy does not even arise.
(37) accusative of animate o-stem
input: [(þhr)v], animate Compatibility *(þhr)/V inanimate Max (þhr)
student-Ø [ ] *!
F student-a [(þhr)]
(38) accusative of inanimate o-stem
input: [(þhr)v], inanimate Compatibility *(þhr)/V inanimate Max (þhr)
F stol [ ] *
stol-a [(þhr)] *!
(39) accusative of animate i-stem
input: [(þhr)v], animate Compatibility *(þhr)/V inanimate Max (þhr)
F mat0 [ ] *
mater-i [(þhr)N] *!
To the extent that the factual coverage is the same, the choice between
this analysis and one which employs directional rules (e.g. Corbett and
Fraser 1993) has significance only within the context of the theoretical and
descriptive programme of the individual investigator. Wunderlich’s (2004)
analysis of Russian is part of a larger model of the differential marking of
objects and agents, whereby arguments high on the animacy/prominence
scale tend to mark the accusative, and arguments low on this scale tend to
mark the ergative. Whatever the cross-linguistic insights of this approach, it
fails to capture some fairly striking generalizations that obtain within the
languages actually under analysis. In the Russian example, the genitive/
accusative syncretism displayed by the animate accusative is attributed to
underspecification of the ending, so it is, in effect, a lexical idiosyncrasy. But
it is not just one ending which needs to be underspecified, but rather at least
seven (the four which are treated by Wunderlich, plus three distinct adjectival
and pronominal endings). This is why the animate accusative in Russian and
other Slavic languages has long been used as an example of systematic
syncretism, and particularly of directional rules (starting with Perlmutter
and Orešnik 1973). The alternative as offered by Wunderlich is to treat the
sevenfold repetition of this pattern as purely accidental.8
However, Wunderlich’s analysis differs from a directional approach not
only in its theoretical goal, but in its factual coverage as well, at least in the
form in which it is presented. In effect, the analysis involves symmetrical
rules plus rule ordering, as sketched above in (24); in this case, constraint
ranking substitutes for rule ordering. The combination of lexical specifica-
tions and constraints state, in effect, that a rule for genitive/accusative
precedes a rule for nominative/accusative. Under certain conditions the
genitive/accusative rule is blocked, allowing the nominative/accusative
rule to be first. Naturally, such a model can easily describe unidirectional
syncretism as well. But, as we suggested above, it cannot describe divergent
bidirectional syncretism. Wunderlich raises the issue but does not propose
an analysis, instead rejecting Stump’s (2001) purported example of diver-
gent bidirectional syncretism. However, other examples could be offered in
its place (as was done above in Chapter 2: x2.4 and x2.5). Thus, the
empirical problem created by divergent bidirectional syncretism remains.
However, the difficulty caused by rejecting directional rules is not really a
problem for Wunderlich’s model, since it in fact contains them, even though
they are not directly exploited. To demonstrate this, let us review some of the
key points in the above analysis. The core of the morphological model
consists of three elements: the input, the output and the Max constraint
8
A similar atomization results when we consider another of his proposals, namely that the
constraint *(þhr)/V inanimate can generally be used to account for the nominative/accu-
sative syncretism typical of Indo-European languages. Recall that this constraint blocks the
use of the accusative ending for inanimates ( neuters), which instead use the default form
(informally, the nominative). But consider Latin once again. It is true that this analysis is
possible for the singular of some nouns of the third declension, examples of which were
given in Chapter 3 (x3.1.2): victor ‘conqueror’ versus aequor ‘sea’). The nominative singular
of masculines is the bare stem, and the accusative is the stem plus -em. In neuters, the
nominative/accusative is the bare stem, which could easily be described as the result of
the omission of the accusative ending. But such an analysis cannot be applied to the second
declension nouns described above in (26). The neuter is distinguished from the masculine by
the lack of the nominative singular ending -us. Undoubtedly, this could be accounted for by
some constraint, but, equally undoubtedly, this is not the same constraint *(þhr)/V inan-
imate found in the third declension, which blocks the accusative. Thus, the observation that
neuter nouns in Latin have nominative/accusative syncretism is reduced to the concatena-
tion of two apparently unrelated facts.
which mediates between them. The input is fully specified (accusative, i.e.
[(þhr)v]), while the Max constraint is underspecified (genitive/accusative, i.e.
(þhr)). The behaviour of the output form varies: in o-stems it is treated as
underspecified, in i-stems it is treated as fully specified, as shown in (40).
(40) Analysis of the Russian accusative per Wunderlich (2004)
o-stems i-stems
input [(þhr)v] fully specified [(þhr)v] fully specified

output [(þhr)] underspecified [(þhr)v] fully specified
Max (þhr) underspecified (þhr) underspecified
However, this variation in the lexical specification of the output forms makes
no difference in the selection of the winning candidate. Both the under-
specified output and the fully specified output satisfy the underspecified
Max constraint. Thus, the tableau in (41) corresponds to those in (37)–(38),
and represents symmetrical syncretism: there is a syncretic genitive/accusative
form, which can be used for the accusative (as well as the genitive). The
tableau in (42) corresponds to (39) and represents directional syncretism: the
genitive form can be used for the accusative (as well as the genitive).
(41) Tableau with underspecified output
input: ACC [(þhr)v] Max (þhr)
NOM [] *!
F GEN/ACC [(þhr)N]
(42) Tableau with fully specified output
input: ACC [(þhr)v] Max (þhr)
NOM [] *!
F GEN [(þhr)N]
Thus, the Max constraint defines a syncretic class of values, while the
absence of a distinct accusative in the output produces a paradigmatic gap,
which is filled by the best available candidate. The effects of this directional
rule do not surface in this particular instance in Wunderlich’s analysis,

because they are blocked by a higher-ranking constraint (Compatibility),
which bars the use of the genitive for the accusative. But directional rules
are nonetheless inherent in the model and, indeed, are implied wherever a
fully specified output interacts with an underspecified Max constraint.
Thus, Wunderlich’s model cannot be seen as offering an alternative to
directional rules, because it already contains them.
4.4 Possible constraints on syncretism

In the preceding section we have looked at the properties inherent in differ-
ent descriptive mechanisms. In this section we look at how these have been
employed to describe syncretism, reviewing some representative proposals.
More particularly, we look at their restrictiveness, as this is a central concern
of most approaches: it is generally assumed that an adequate model will
distinguish between possible and impossible patterns of syncretism. Below
we evaluate these claims in the light of patterns surveyed in Chapter 3: x3.
4.4.1 Variants on hierarchical structures

A number of proposals have been, in effect, claims that feature values are
hierarchically structured, and that syncretism is necessarily a reflection of
this structure. The most restrictive version is offered by Williams (1981,
1994), which does not make use of the elsewhere condition. As we have
shown in Chapter 2: x2.1, this implies that if there are multiple syncretic
patterns in a language, they will be nested within each other, which
Williams (1994) expresses with the claim that ‘when there are multiple
related paradigms, there will be one instantiated paradigm, and all others
will have its syncretic structure, and perhaps some more. But no other
related paradigm will have a contrary syncretic structure, making distinc-
tions where that one does not’ (p. 27). Chvany (1986) and Johnston (1997)
argue for loosening this model by, in effect, allowing underspecification, as
described above for Noon (2) and Gaagudju (6)–(7) The effect is to allow
‘contrary syncretic structures’, but only in as much as they can be described
given a single linear order of features.9 McCreight and Chvany (1991) take the
logical next step and extend this proposal along multiple dimensions.
9
Plank (1991) observes that the quest for a single invariant linear order that would capture all
the major syncretic patterns of a language is implicit throughout the history of grammatical
description.
For example, given a paradigm delimited by two features and drawn as a grid,
only those cells which can be brought next to each other can be described as
syncretic, as in (43a), while (43b) is an impossible pattern.
(43)
a. x y b. x y
X a a X a b
Y a c Y b d
Z b d Z c e
These models remain agnostic about the actual content of feature struc-
ture; what they constrain is the co-occurrence of multiple patterns of
syncretism involving the same feature values. All of the proposals leave a
good amount of material unaccounted for. Thus, what Williams (1994)
describes as ‘contrary syncretic structure’ is in fact common, especially
with case, person and gender. The more relaxed model of Chvany (1986)
and Johnston (1997) largely holds for case (which they were designed to
describe) and person, but not for gender, as seen above in x4.2.1.
McCreight and Chvany’s (1991) model, which is specifically designed to
handle the interaction of multiple features, cannot describe polarity effects
(see Chapter 3: x3.7). Finally, these proposals are all predicated on sym-
metrical rules, so the observations in x4.3 above apply.
A less restrictive model of linear ordering is offerred by Plank (1991). In
a survey of case syncretism in several Indo-European languages, he notes
that although strict linear ordering will not capture all the syncretic pat-
terns for any of the languages, the deviations from linear ordering are
fairly minimal, so that cases which are syncretic with each other can at least
be represented as adjacent on a two-dimensional model.
4.4.2 Carstairs (1987), Carstairs-McCarthy (1998a, b)

Carstairs (1987) divides inflectional homophony (‘syncretism’ in the sense
we have been using it) into two types:
* Syncretism. This occurs where the feature whose value is neutralized
and the feature whose value is the determining context are realized
simultaneously (the inflectional marker exhibits cumulative
exponence). For example, dative/ablative plural in Latin constitutes
syncretism for Carstairs because the neutralized feature (case) and
the conditioning feature (number) are realised simultaneously by
the endings -is, -ibus, which cannot be broken up into distinct

case and number morphemes. This is the most common and
diachronically stable type of inflectional homophony.
* Take-overs. These occur where an inflectional marker that realizes
the morphosyntactic value a in one context realises both values a
and b in some other context.
He proposes the Systematic Homonymy Claim (Carstairs 1987: 123), which
states that all inflectional homophony is either (i) a syncretism or (ii) a take-
over, where the conditioning feature is lower on the relevance hierarchy than
the feature whose values are neutralized. The relevance hierarchy, proposed
by Bybee (1985), ranks inflectional features on their semantic relevance to
the meaning of the root, as manifested in cross-linguistic frequency and
position. Carstairs cites the following as contexts for take-overs, where the
features to the left of the arrows are higher on the relevance hierarchy:
(44)
* tense, mood > person-number
* definiteness > person-number
* number > case
The following schematic noun paradigms illustrate what is possible under
this system. Paradigms (45) and (46) both conform to the Systematic
Homonymy Claim. Syncretism (in Carstairs’ sense) is represented in (45):
genitive and dative are syncretic in the plural, expressed by the form g, which
simultaneously realizes case and number. The paradigm in (46) illustrates a
take-over, with homophony of the genitive singular and genitive plural.
First, there is separate realization of number (p) and case (a–d). Second,
the neutralized feature (number) is higher on the relevance hierarchy than
the conditioning feature (case).
(45) (46)
SG PL SG PL
NOM a e NOM a pa
ACC b f ACC b pb
GEN c g GEN pc pc
DAT d g DAT d pd
(Carstairs 1987: table 4.15, 113)

The paradigms in (47) and (48) represent impossible, or at any rate

unsystematic, patterns. The paradigm in (47), with syncretism of the
genitive and dative in the plural, is not possible, because, although the
distribution of the genitive marker c makes it look like a take-over,
the neutralized feature (case) is not higher on the relevance hierarchy
than the conditioning feature (number), in violation of the Systematic
Homonymy Claim. Finally, the paradigm in (48) is not possible, because
it is not a take-over (e does not serve a non-syncretic function elsewhere), nor
is it a syncretism, because number (p) and case (a–d) are realized separately.
(47) (48)
SG PL SG PL
NOM a pa NOM a pa
ACC b pb ACC b pb
GEN c pc GEN c pe
DAT d pc DAT d pe
(Carstairs 1987: table 4.10, 111) (Carstairs 1987, table 4.11, 110)
Before considering the important revisions to this model made by

Carstairs-McCarthy (1998a, b), let us consider the predictions made so far.
The statistical predominance of what Carstairs-McCarthy terms syncretism is
undeniable. What we have termed directional effects – Carstairs’ take-overs –
are observable only in a minority of cases. Carstairs (1987: 111–12) states that
this follows from the reasonable assumption that the normal state of affairs is
for an inflectional marker to have a consistent function. Thus, in (45), g is
always genitive/dative plural, while in (46), p is usually plural but then some-
times it is singular/plural. However, the banning of (47) and (48) seems less
well motivated. The sort of pattern in (47) is not uncommon; we shall return
to it below. On the other hand the pattern in (48) is indeed rare, though we can
cite an example involving case in the Dravidian language Pengo (49). (The
initial d of the singular endings is the post-consonantal variant of t and n;
compare the vocalic stem singular doNri, doNri-tin, doNri-to, doNri-ni, doNri-taN
‘mountain’.) Though this pattern is rare, that fact may follow from more
general principles of morphological organization than from any specific
constraint on syncretism. Where features are realized separately, feature
conditioned allomorphy is the exception rather than the rule. The rarity of
(48) is then a facet of the rarity of the sort of pattern in (50), where case and
(49) Pengo noun declension, ‘hand’ (Burrow and

Bhattacharya 1970: 32–3)
singular plural
NOM key key-ku

ACC-DAT key-diN key-ku-kaN
LOC key-do key-ku-kaN
GEN key-di key-ku-ni
INS-ABL key-daN key-ku-taN
(50)
singular plural
NOM a pa
ACC b pb
GEN c pe
DAT d pd
number are marked separately, but the genitive, and only the genitive, has a
distinct plural allomorph.
Carstairs-McCarthy (1998a, b) offers a somewhat different analysis of
these phenomena, which result in a somewhat different range of predictions.
The underlying principle is that inflectional meaning should be governed by
the same constraints that obtain for lexical semantics, which leads to four
axioms, summarized below from Carstairs-McCarthy (1998b):
A. Lexical items do not contain meanings consisting of incompatible
disjuncts, e.g. *‘apple OR banana’. By the same token, the mean-
ing of an inflectional marker should not contain incompatible
disjuncts consisting of competing values for the same feature,
e.g. *‘ablative OR locative’. (Carstairs-McCarthy assumes that
feature structure is flat.)
B. Lexical items do not have meanings containing negations, e.g.
*‘not apple’. By the same token, inflectional markers should not
contain negations in their meanings, e.g. *‘not past’.
C. Lexical items may contain compatible disjuncts. For example, the
different senses of climb in
(a) the boy climbed up the tree
(b) the boy climbed down the tree

(c) the snake climbed up the tree
can be reconciled by characterizing its semantics as ‘go, upward OR
clambering’. Sentence (b) contains only the element ‘clambering’ and
sentence (c) only the element ‘upward’, but the two meanings are compat-
ible with each other, as witnessed by sentence (a). (Carstairs-McCarthy
takes this example from Jackendoff 1985.)
D. Special vocabularies may exist where the meanings of otherwise
distinct lexical items may be collapsed; Carstairs-McCarthy cites
the example of the Dyirbal ‘mother-in-law language’, where, for
example, the normal terms nudin ‘cut deeply’ and gunban ‘cut less
deeply’ are combined in the single term dyalnggan ‘cut’. By the
same token, where distinct allomorphs of inflectional markers are
found in special contexts, for example where case marking is
cumulated with plural marking in plural contexts, related mean-
ings which elsewhere have distinct realizations may share the same
realization.
E. No rule can make overt reference to the unmarked value of a
feature.
Axiom D now accounts for ‘syncretism’ as characterized in Carstairs

(1987) and is defined as inflectional homophony that occurs under condi-
tions of allomorphy. Cumulative exponence, by definition, provides a
context for allomorphy, but is not itself the crucial factor – for example,
distinct inflection classes are also relevant contexts.10 Another property of
syncretism under axiom D is that the neutralized meanings be related
(p. 292); however, since feature structure is construed as flat (p. 294),
relatedness of meaning cannot be formalized, so it is not clear what
constraints emerge from this.
The other axioms account for directional effects, largely replacing take-
overs with underspecification. Because of axiom E, which states that the
unmarked value is, quite literally, unmarked, underspecification for any
value always entails homophony with the unmarked value. This yields two
types of directional effect, depending on whether axiom C is invoked.
Where axiom C is not involved, it is predicted that the form asso-
ciated with the unmarked value will prevail in unmarked contexts, as in
10
If, as Carstairs-McCarthy suggests (1998b: 290), inflection class is to be construed as part
of the meaning of an inflectional marker, then this too constitutes cumulative exponence.
example (51), taken from the Chibchan language Ika. First and third
person singular are syncretic in all tense paradigms except the distal past.
The syncretic 1SG/3SG form has no overt person-marking affix, which
makes it look like the 3SG form of the distal past. If we assume that third
person and ‘elsewhere’ are the unmarked values for person and tense,
respectively, this is what the axioms should yield, since the form associated
with the unmarked third person value is extended in the unmarked context.
(51) Ika (Frank 1990)
distal past elsewhere
1SG (stem) -rua -na (stem)

2SG nL-(stem) -na nL- (stem)
3SG (stem) -na (stem)
This can be represented by the rules in (52), which are in accordance with
the axioms above. The ending -rua is specified as the first person distal past
ending, nL- as the second person prefix, unspecified for tense, and Ø is the
general default, unspecified for tense and person.
(52)
-rua 1 distal past

nL- 2
Ø
Carstairs-McCarthy (1998b) illustrates the effects of axiom C with an example

from Hungarian conjugation (53), which had been described in Carstairs
(1987) as a take-over of the 1SG indefinite by the definite in the past.
(53)
present past
1SG INDF vár-o-k vár-t-a-m

1SG DEF vár-o-m ! vár-t-a-m
At issue is the distribution of the 1SG affixes -k and -m, which are isolated
in (54). In the present, -k marks the indefinite and -m the definite, but in the
past -m marks both.
(54)
present past
1SG INDF -k -m
1SG DEF -m -m
Carstairs-McCarthy attributes the distribution of the affixes to the rules:
(55)
-m 1 definite OR past
-k 1
Since the values ‘past’ and ‘definite’ can co-occur, this is a compatible
disjunction; -k is simply an elsewhere form. Although these rules result in
neutralisation both of definiteness and of tense, at the level of the word, it
is only the past tense forms which are identical. This is a secondary effect,
resulting from the fact that there is a separate layer of affixes which
distinguish the two tenses (the distinct theme vowels, plus the past tense
marker -t-), thus breaking up the potential homophony between present
and past.
To summarize, Carstairs-McCarthy’s (1998a, b) model accommodates
two types of non-accidental syncretism: (i) syncretism in the strict sense,
involving a combination of values represented by a distinct allomorph, and
(ii) underspecification. (The continued need for take-overs is acknow-
ledged (1998b: 299) but not directly incorporated into the model.) Does
this model allow for testable predictions?
In the case of ‘syncretism’ (in the sense defined above), the prediction
would appear to be that where we find an affix (or inflectional operation of
some sort) which always combines some set of values wherever it is found,
these values should be related to each other, that is they should have related
meanings. However, without an explicit theory as to what constitutes
‘related’ meanings, we cannot readily assess this prediction. However, in the

case of underspecification, the model implies two clear predictions about the
behaviour of directional effects:
(56) I. Directional effects will always involve the unmarked value of the
feature in question (because of axiom E).
II. Directionality will be of two sorts; either:
a. The form associated with the unmarked value of some feature will
prevail in the context of the unmarked value of some other feature.
–or–
b. The form associated with the marked value of some feature will prevail
in the context of the marked value of some other feature.
Assessing the validity of these predictions is complicated by the fact that

they all revolve around the notion of markedness, which is itself hard to
pin down with precision. Therefore, our observations can only be tenta-
tive. Nevertheless, working with commonly held assumptions about what
should be construed as the unmarked value, we find no general support for
these constraints. We summarize our observations below. Note that we are
merely looking at whether the data conform, superficially, to the predic-
tions. Whether the individual examples even admit of a comprehensive
analysis in terms of Carstairs-McCarthy’s (1998a, b) model is a question
we have not addressed.
Case. Prediction I is met in all instances of what we have called ‘core case
syncretism’ (Chapter 3: x3.1.2), which is indeed the most common type.
The prediction is violated wherever directional effects can be observed
which do not affect nominative or absolutive, and a number of these have
been seen so far, or will be adduced below, e.g. genitive/accusative in
Russian and Classical Arabic (see (32) above); ergative/locative in
Koryak ((13b) in Chapter 3: x3.1.2) and the ergative/genitive in Lak ((14)
in x3.1.2); the gentive/accusative dative-locative/accusative alternation
in Bonan ((15) in Chapter 3: x3.1.2); and dative/illative in Erzja Mordvin
((18) in Chapter 3: x3.1.2). Within those examples which conform to
prediction I, prediction IIa seems to be met in cases of differential object
or agent marking, where accusative or ergative marking is absent for some
set of nominals. Examples which relate to prediction IIb are hard to
intepret. For example, the accusative form seems to prevail in Greek and
Latvian ((3) in Chapter 3: x3.1.2); whether these conform to prediction IIb
depends on whether one interprets neuter (in the case of Latin) and
feminine (in the case of Latvian) as marked values.
Person. Directional effects involving person are outlined in Chapter 3:

x3.2.3; the results are summarized in (57), where ‘U’ indicates unmarked
and ‘M’ indicates marked. The symbol to the left of the slash (‘/’) stands for
the syncretic value, the symbol to the right of the slash represents the
context. Thus U/U means ‘the form associated with the unmarked person
value is extended to a marked value in the unmarked context’, M/M means
‘the form associated with a marked person value is extended to the
unmarked value in a marked context’, and so on.
(57) Directional effects involving person (from Chapter 3: x3.2.3)
1/2 2/3 1/3
a. Burarra M/M
b. Callahuaya M/?
c. Carib M/U
d. Dani M/M
e. Dutch U/M
f. Gujarati ?/? U/?
g. Koiari U/?
h. Kongo M/U
i. Literary Kannada ?/?
j. Murle M/M
k. Nobiin ?/M U/M
l. Old Icelandic M/U
m. Shinassha ?/M
n. Suena M/M
o. Udihe ?/?
If third person is construed as the default value, then prediction I is violated

where directional effects do not involve third person (in Gujarati, Literary
Kannada, Nobiin, Udihe and Shinassha). In evaluating prediction II, we
have assumed that values such as singular, present tense, indicative and
declarative constitute the unmarked values of the contextual features,
though in a number of cases it is not obvious what markedness values to
assume. Possible examples conforming to prediction IIa are found in
Gujarati and Koiari: in both cases the third person form prevails, but it is
not clear whether the context should be construed as unmarked. Examples
conforming to prediction IIb come from Burrara, Dani, Shinassha and
Suena, while Callahuaya and Udihe are possible examples as well. Neither
component of prediction II is met by Dutch or by Nobiin (since the
unmarked third person form prevails in the marked context), nor by the
Carib, Old Icelandic and Kongo examples (because the marked second
person form prevails in the unmarked context).
Number. Assuming a flat feature structure, we have found no good
examples conforming to either prediction. Were one to admit a hierarchi-
cal features structure, with dual as the marked counterpart to plural, then
the Koryak example cited in Chapter 3: x3.5 seems on the surface to
conform to both predictions: the dual form prevails in certain transitive
combinations, which may be construed as the marked context. However,
one should recall that in Koryak, morphologically, it is the plural which is
marked with respect to the dual (by the addition of the suffix -la-).
Tense-aspect-mood. The convincing examples of directionality conform
to prediction I but not to II. Thus Gapapaiwa, Tawala and Loniu (cited in
Chapter 3: x3.6.2) all involve syncretism with what may reasonably be
construed as an unmarked tense, in the context of marked person (non-
third person), but the form which prevails is the zero marking associated
with the unmarked tense.
The intuition behind Carstairs-McCarthy’s predictions is that directional
effects can be attributed to underspecification, and that the possibilities for
underspecification are themselves limited by markedness relationships.
However, as we have seen, directionality does not seem to be generally
predictable on the basis of the values of the features involved. Carstairs-
McCarthy does allow the possibility that apparent counter-examples may
be analysable as rules of referral (1998a: 18–19). In this case, the model
makes no generally applicable predictions about directional effects, since
there is no theory-independent way of identifying which phenomena should
be construed as rules of referral.
4.4.3 Impoverishment
Impoverishment (Noyer 1997, 1998, 2001) is a device allowing for the
constrained alteration of feature values and plays a key role in some
descriptions of syncretism. Bobaljik (2002) writes that ‘[i]mpoverish-
ment . . . admit[s] of predictions about impossible syncretisms cross-
linguistically, predictions that appear to be largely borne out.’
Specifically, impoverishment is designed to account for directional effects.
As an illustration, Bobaljik adduces an example from Stump (1993),
syncretism of the 2SG and 3SG in the two past tenses of Macedonian:
(58) Macedonian (Stump 1993)
present aorist imperfect
1SG padn -am padna -v padne -v

2SG padne -š padna padne -še
3SG padne padna padne -še
1PL padne -me padna -v -me padne -v -me
2PL padne -te padna -v -te padne -v -te
3PL padn -at padna- -a padne- -a
In Stump (1993) this example was used to illustrate a stipulated directional

rule, namely a rule of referral. The syncretic form is interpreted as looking
like the third singular, both for its zero person-number ending and for the
absence of the past tense marker -v-.11 The rule of referral stipulates that
2SG takes the form of 3SG (i.e. 2SG and 3SG form a set, whose form is
determined by the 3SG). Bobaljik offers instead an interpretation in terms
of impoverishment, which illustrates its two key aspects: (i) deletion of
features, and (ii) the assumption of unmarked or default feature values. In
this case, impoverishment deletes the feature second person in the singular
of the past tenses. This deleted value is replaced by the default value,
construed as third person. What is not possible is the replacement of the
third person form by the second, which would be perfectly feasible under
Stump’s model. Thus, impoverishment makes two predictions:
I. Directional effects will involve the unmarked value of a feature.
II. The form associated with the unmarked value will prevail.
Note that prediction I coincides with that of Carstairs-McCarthy (1998a, b),
while II is the opposite. However, an assessment of the validity of these
predictions is contingent on the particular model of feature structure employed,
and this is an issue distinct from the device of impoverishment as such.
As discussed above, different models of feature structure make different
predictions, and so, the effects of impoverishment will differ according to
11
The absence of -v- in the 3PL forms has a different diachronic explanation from that of the
2SG/3SG form. Historically, the -v- descends from -x-. There is no evidence that -x- was ever
present in the singular forms at any stage in the history of the Slavonic languages. Its loss in
the 3PL is relatively recent (post-sixteenth century; Koneski 1996 [1986]: 92), due to the
regular loss of intervocalic -x- in Macedonian.
the feature structure assumed. If we assume flat feature structure, then the
observations above about Carstairs-McCarthy’s prediction I apply, and the
observations about prediction II apply in the reverse: Carstairs-McCarthy’s
positive examples are counter-examples to impoverishment’s predictions,
and vice versa. More elaborate models of feature structure create more
markedness relationships, and so more domains for impoverishment to
operate in. But, as Carstairs-McCarthy points out (1998b: 288), this also
opens up the range of surface effects that can be produced. He illustrates this
with Noyer’s account of number syncretism in the verbal prefixes of the
Papuan language Nimboran. The prefix k-, which is characteristic of the
dual alone for first and third persons, is found for the plural as well in second
person and first inclusive (59).
(59)
person
1, 3 2, 1INCL
SG [þSG -PL] Ø Ø
DU [-SG -PL] k- k-
PL [-SG þPL] i- k-
Noyer’s analysis is illustrated in (60). Number values are treated as

compounds – significant for the present example is that dual and plural
are each treated as further modifications of the value non-singular.
Impoverishment works by deleting the plural value, yielding the under-
lying non-singular. Separately, the prefix i- is construed as overtly plural,
while k- is the more general non-singular, yielding what looks like the
‘take-over’ of the plural by the dual in the second person and first
inclusive.
(60)
impoverishment inflectional rules
[þPL] ! Ø/ [þ2 -SG] i [þPL]

k [-SG]
But, as Carstairs-McCarthy points out, the reverse situation can just as

easily be described by transforming the dual into non-singular, and treat-
ing i- as the generic non-singular marker, as in (61)–(62):
(61)
person
1, 3 2, 1INCL
SG [þSG PL] Ø Ø
DU [SG PL] k- i-
PL [SG þPL] i- i-
(62)
impoverishment inflectional rules
[PL] ! Ø/ [þ2 SG] i [SG]

k [SG PL]
If, as Noyer assumes, dual is marked with respect to plural, then it would
seem that impoverishment makes no consistent predictions in this instance
about directional effects.
4.4.4 Stump (2001) and Zwicky (2000)

Stump (2001) presents a model in which overt stipulation of syncretic sets
of values plays a major role; Zwicky (2000) is a reaction to and adaptation
of this model. Stump assumes flat feature structure, with no markedness
asymmetries. He divides syncretism into three types, each with its own
formal representation:
i. Unstipulated syncretism. This is described by simple underspecifi-
cation for the feature in question. Because of the assumptions
about feature structure, this is effectively limited to instances
where all the values of a feature are syncretic.
ii. Unidirectional and bidirectional syncretism. These are described
by rules of referral. The Bidirectional Referral Principle (p. 219)
stipulates that any rule of referral also entails its inverse. That is, if
a rule of referral states that value x of feature a takes the form of

value y, this implies that value y takes the form of value x, all else
being equal.
iii. Symmetrical syncretism. As with rules of referral, this involves a
stipulated set of values, but here the form is defined across the
whole set rather than in terms of the individual members. In a
sense, symmetrical syncretism can be defined negatively: it is syn-
cretism which does not lend itself to being described by under-
specification (because some but not all values of a feature are
involved) nor by rules of referral (because no directional effects
are in evidence).
These devices can describe any syncretic pattern. However, Zwicky 2000
argues that it is too complex, imputing underlying distinctions to pheno-
mena which are, in his view, epiphenomenal. First, he rejects the distinc-
tion between unstipulated and stipulated syncretism: what Stump terms
unstipulated syncretism is a convenient representational short cut, but no
less stipulative than simply declaring the identity of all the values. Second,
he rejects the distinction between directional and symmetrical syncretism:
directional effects can be seen as a by-product of underspecification
(see x4.4.3 above). Thus, Zwicky proposes retaining Stump’s symmetrical
syncretism as the sole type. However, as pointed out above in x4.3 and its
subsections, a symmetrical approach can only readily describe unidirec-
tional syncretism: what we have described as convergent bidirectional
syncretism cannot be described without assuming further devices, and
divergent bidirectional syncretism cannot be described at all. Thus, there
are inherent limits to what Zwicky’s model can describe.
On the other hand, though Stump’s basic system can describe any
pattern, he imposes a distinct constraint onto this, namely the Feature
Ranking Principle, which reads:
For any language ‘, there is a ranking > of morphosyntactic features in ‘

which satisfies the following condition: for every stipulated syncretism S
in ‘, if the dominant properties of S include a specification of the feature
Fd and the subordinate properties of S include a specification of the
feature Fs, then Fd > Fs. (Stump 2001: 239)
In effect, this states that, within a given language, if you have stipulated
syncretism of values of feature x in the context of feature y, then you
cannot have stipulated syncretism of values of y in the context of feature x.
Empirically, this represents a limit on the number of different syncretic
patterns that can be accommodated in a language, and it appears, on the

whole, to be borne out. However, we know of one possible counter-
example, namely case and number syncretism in Slovene, which seem to
require stipulated syncretism of number in the context of case, and of case
in the context of number. Slovene distinguishes three numbers and six
cases. Number syncretism in the context of case is systematically found in
nouns and adjectives, where dual and plural are syncretic for the genitive
and locative. As we argue below in Chapter 5: x5.1.1, this pattern is best
modelled as a rule of referral, as evidenced by the suppletive pattern of the
word cˇlovek. Case syncretism in the context of number is widespread; in
particular, the dual always has two syncretic forms, though the actual
patterns may vary both across and within word classes. All nominals
have dative and instrumental syncretism in the dual. Nouns and adjectives
additionally show nominative/accusative syncretism.
(63) Declension of Slovene ‘person’
NOM človek človeka ljudje

ACC človeka človeka ljudi
GEN človeka ljudi ljudi
LOC človeku ljudeh ljudeh
DAT človeku človekoma ljudem
INS človekom človekoma ljudmi
Pronouns have accusative/genitive syncretism, while the behaviour of the

locative dual varies. It never has a distinct form of its own in Slovene; the
literary standard allows syncretism either with the dative/locative, or with
the genitive (Toporišič 1976: 241–2).12
12
Genitive/locative syncretism in the dual is an archaism artificially introduced into the
literary language in the nineteenth century (Tesnière 1925: 305). Nonetheless, it is given as
the preferred option in such authoritative grammars as Toporišič (1976). It is interesting to
note that the third person pronoun, whose declension has both pronominal and adjectival
characteristics, has three options for its locative dual: it can be identical to the dative/
instrumental dual (njima), to the genitive dual (naju) or to the locative plural (njih)
(Toporišič 1976: 241–2).
(64) Declension of Slovene ‘I’
NOM jaz midva mi

ACC mene naju nas
GEN mene naju nas
LOC meni naju nama nas
DAT meni nama nam
INS menoj nama nami
If we assume, along with Stump, that feature structure is flat, then one of
the patterns of case syncretism in the dual must be stipulated, since under-
specification for case in the context of number can only be invoked once.
4.5 Summary
4.5.1 Predictions and counter-examples
The contrast is often made between a restrictive theory, which is able to
make predictions about possible and impossible structures, and a descriptive
framework, which is open-ended. The formal models we have reviewed
above aim to provide a restrictive theory of syncretism. In no case, however,
do the predictions correspond precisely to the empirical evidence – there is
always a residue of counter-examples. On a literal-minded approach this
would seem to invalidate all such attempts straightaway. However, we can
assume that most investigators who have proposed constraints are aware of
the possibility of counter-examples, so a more nuanced assessment of such
proposals must focus on how these are treated.
One way of dealing with exceptions would be to treat them as the result of
accidental homophony, and thus beyond the reach of morphological analy-
sis. A clear example of accidental homophony was cited above in Chapter 1
(x1.5), where vowel reduction in Russian leads to the homophony of the
case-number endings -o and -a when unstressed. However, more often than
not the examples we see are ambiguous, so that it is no more natural to
interpret them as phonologically accidental than as morphologically sys-
tematic.13 More importantly, there is diachronic evidence that the boundary
13
As an illustration of the danger of such an approach, we can take Noyer’s (2001) treatment
of syncretism (primarily of gender and number) in the non-Pama-Nyungan Australian
language Nunggubuyu, where the formal model forces systematic syncretism to be treated
between accidental and systematic is fluid: syncretic patterns which arose as

by-products of regular sound change may be reinterpreted as morphologi-
cally systematic, and then extended to environments that were not subject to
the original sound change. Sologub (1983) discusses such an example from
Russian. In Standard Russian, the genitive singular of a-stem nouns is
distinct from the dative/locative: genitive singular zˇeny ‘wife’ versus dative/
locative singular zˇene. In some dialects, the two forms have collapsed,
resulting in genitive/dative/locative singular zˇeny or zˇene (depending on the
dialect). This does not follow from any sound change but is rather the result
of analogy with the i-stems (as in NOM/ACC kost 0 , GEN/DAT/LOC kosti ‘bone’;
see (90) in Chapter 3: x3.8.2), where this collapse was phonologically regular
(p. 86). Other examples that have been regarded as the morphological exten-
sion of originally accidental patterns include some forms cited in Chapter 3,
namely the Dhaasanac B forms (examples (83) and (84) in x3.7.2), the second
singular / third singular present in Old Icelandic (x3.2.4), the first singular /
third singular present in Livonian (x3.2.4), as well as the genitive singular/
as the concatenation of separate instances of accidental homophony. Nunggubuyu shows

systematic syncretism of first exclusive and second person non-singular in person marking
on verbs, alongside a non-syncretic first inclusive form. This occurs in one of the two sets
of subject prefixes (the choice is determined by tense, aspect and polarity (negation)), and
in the object prefixes. According to Noyer, these two person values do not constitute a
natural class (indeed, it is hard to imagine how they would; see the discussion at the end of
x3.2.2). Instead, he treats subject-and object-person syncretism as two separate instances of
accidental homophony. Syncretism of the subject markers is construed as the accidental
by-product of a morphological rule. The rule itself is not given, but the derivations (Noyer
2001: 811) are given below, where nV:- is construed as the first person non-singular subject
marker, nV- is the second person non-singular marker, and wu- as the non-singular
intransitive subject marker. (Note that in Noyer’s analysis, nV- is in fact construed as
unmarked for number; its function as a non-singular marker is due to the existence of a
series of overtly singular second person markers.)
underlying form surface form

1DU F nV:- Ni- ! ni:-Ni-
2DU F nV- wu- Ni- ! ni:-Ni-
1DU M nV:- ni- ! ni:-Ni-
2DU M nV- wu- ni- ! ni:-Ni-
1PL nV:- ru- ! nu-ru-
2PL nV- wu- ru- ! nu-ru-
The syncretic 1PL/2PL object markers have the form na-. This is assumed to result from the
existence of two homophonous prefixes: second person non-singular nV- (as seen above)
and first person non-singular object nV- (Noyer 2001: 760).
accusative plural in Common Slavonic (Meillet and Vaillant 1934: 398) and
the collapse of singular and plural in neuter nouns in Old High German
(Wurzel 1987: 69). If such diachronic interpretations are held to be valid,
then we must admit the possibility that accidental homophony may be
reinterpreted as morphological systematicity. This is not to claim that all
apparent instances of syncretism must necessarily be treated as morpholo-
gically systematic, but rather that such an interpretation should be available
for any apparent instance, to be examined on its individual merits rather
than excluded a priori. A formal model which instead dismisses certain
phenomena from the outset can only be a poor representation of linguistic
reality.
A more inclusive approach would characterize the counter-examples not
as accidental, but rather as ‘marked’. That is, the morphological model can
produce them, but only in a complex or indirect fashion. This approach is
subject to the same observations made above: it is a covert distinction, and
so has no necessary empirical ramifications. While neither approach can
successfully determine the possibility of a given pattern in a given lan-
guage, they do imply statistical and diachronic claims. Systematic patterns
are common, accidental or marked patterns are rare and diachronically
unstable. Both these points invite caution. The causal connection between
formal markedness and rarity is often assumed but remains undemon-
strated (especially since formal markedness is not subject to direct obser-
vation). Further, as the examples in Chapter 3 show, the standard
assumptions about what are common and what are rare patterns are in
need of revision.
The related issue of diachrony is also problematic. For example,
Carstairs (1987: 128–31) gives the example of the syncretism of 1SG/2SG/
3SG in the imperfect indicative in medieval dialects of Italian, which was
resolved in Modern Italian by the creation of new, unsyncretic forms. This
is taken as evidence that the original syncretic pattern was unsystematic.
He observes, though, that unsystematic patterns may be diachronically
persistent. Further, we may note that systematic syncretism may be
resolved over time as well – consider the Old Nubian paradigm cited in
(48) in Chapter 3: x3.2.4, where 2SG/3SG and 1PL/2PL are syncretic. This
constitutes systematic syncretism in Carstairs’ (1987) terms, as it involves
cumulative exponence, and its diachronic stability is attested by its reten-
tion over ten centuries (Browne 2002: 1) in contemporary Nubian lan-
guages such as Dongola Nubian (Armbruster 1960). But in the Nubian
language Nobiin this syncretism was resolved through the creation of new,
distinct second person forms. What this seems to indicate is simply that
homophony of any sort may be resolved over time; underlying systemati-
city plays no necessary role.
4.5.2 Towards a model of syncretism

In this chapter we have shown that no attempt to impose formal con-
straints on the description of syncretism is wholly successful. On the other
hand, there are some decided tendencies, remarked on throughout Chapter 3,
most especially the prevalence of certain patterns of case, person and
number syncretism, which have a plausible basis in the structure of the
features in question. The challenge for morphological theory is to allow the
formal model to be open-ended, while still giving an account of the fact
that some patterns are common, and others rare. Below we suggest that the
statistical distribution of patterns is a reflection of diachronic processes.
Consider first the change undergone by Anejom, discussed in (45) in
Chapter 3: x3.2.4, where a four-way person distinction (first inclusive and
exclusive, second and third) and three-way number distinction (dual, trial
and plural) was collapsed into a single form, etymologically the third
plural. Though striking in its rapidity, the direction of change does not
seem unexpected: third person substitutes for the other forms, and plural
substitutes for dual and trial. The basis for the innovative system of
twentieth-century Anejom need not be sought in language-specific mor-
phological quirks. Rather, it follows from quite widespread assumptions
about the unmarkedness of the third person with respect to other person
values, and of plural with respect to other non-singular number values.
Contrast this with the developments in the Cushitic language
Dhaasanac, shown in (83) and (84) in Chapter 3: x3.7.2, where syncretism
of the first plural exclusive form with the 2/3SG F form is extended from one
class of stems (where it was the result of regular sound change) to all stems.
This combination of values cannot plausibly be argued to have any basis in
the structure of person or number. Rather, this accidental disjunction of
values was taken over as a pattern by the morphology.
The change in Anejom yielded a ‘natural’ result, while the change in
Dhaasanac yielded an unnatural result. But in essence the two kinds of
change are the same. The original paradigmatic space was reorganized
according to a new morphological pattern. The difference lies in the source
of the morphological pattern. In the case of Anejom, we can suppose its
source lay in (universal?) properties of feature structure, whereby 3PL is
unmarked with respect to other non-singular person-number values. Thus,
this morphological pattern is shared across most (all?) languages. In the

case of Dhaasanac, the morphological pattern was created by a phono-
logical change that affected one stem class. By definition, this morphological
pattern is language-specific.14 The difference between natural (i.e. com-
mon) and unnatural (i.e. uncommon) syncretic patterns thus need not be
reflected in the formal model. We suggest that contrast between these two
diachronic routes is sufficient to account for the statistical predominance
of morphosyntactically natural patterns. Paradigmatic morphological pat-
terns based on common or universal elements of feature structure (i) are
available to all languages, (ii) can arise spontaneously (as in Anejom) and
(iii) are self-regenerating in case of disruptions. Morphological patterns
based on phonological change are language-specific, and always in com-
petition with morphological patterns based on feature structure. It only
follows that the former type should be more widespread.
On this view, even if we manage to construct a comprehensive model of
feature structure that will account for some syncretic patterns, the possibility
still remains that patterns may be codified which are independent of feature
structure. The crucial question here is whether there are any constraints on
the production of unnatural syncretic patterns. At present we do not have
enough information to decide this question; as the discussion above will
have shown, it can only be resolved empirically.
14
This distinction roughly corresponds to that between ‘system-independent’ and ‘system-
dependent’ morphological naturalness, as discussed within Natural Morphology
(Mayerthaler 1987, Wurzel 1987).
5 Formal framework and case
studies
As we saw in the previous chapter, the challenge for a formal analysis of

syncretism is twofold: to represent the cross-linguistically more general pat-
terns of syncretism in terms of feature structure, and to account for patterns
which occur independently of feature structure. In this chapter we lay out a
formal framework and demonstrate it with three cases studies. When intro-
ducing the formal framework, we show in x5.1 that inferential-realizational
frameworks, such as Network Morphology and Paradigm Function
Morphology, are capable of capturing syncretic patterns which may arise as
the result of underspecification, or as the result of referrals (i.e. independently
of feature structure). One advantage of such approaches is that referrals and
underspecification can be used simultaneously. We shall see when we come to
the second case study that this is just what is required for the avoidance
morphology of Dalabon. Indeed, we show that generalized referrals – where
sets of paradigm cells can refer to sets of paradigm cells – which frameworks
such as Network Morphology allow for, are an important requirement.
We consider the relationship between underspecification (a feature
structure-based constraint) and semantic naturalness. As we saw in
Chapter 4: x4.2.1, underspecification can be used with atomic feature
values (i.e. in a ‘flat’ structure) where the syncretism is the default to the
‘elsewhere’. In such cases, in the absence of other representational
means, only one syncretic pattern can be described for any domain.
Underspecification in such instances need not be equated with semantic
naturalness, as the elsewhere is the residue of what is not specified in the
morphology. Where a hierarchical feature structure is assumed, under-
specification may represent the resort to an intermediate ‘natural class’,
but as we saw in our discussion of Gaagudju in Chapter 4: x4.2.1, some-
times it is not possible to order feature values in such a way that we can
always treat syncretism as a default to an intermediate semantic class.
Our case studies in x5.2–x5.4 concentrate on significant fragments of
individual languages. In x5.2 we show that the Dhasaanac data in
171
Chapter 3: x3.7.2 can be analysed straightforwardly in the Network

Morphology framework. In x5.3 we turn to the verbal system of
Dalabon to show that syncretism which may originally have had a seman-
tic or pragmatic motivation has been morphologized, and demonstrate
that an adequate account requires a combination of referral and under-
specification in order to see how that language’s morphological system-
aticity relates to more general patterns. In x5.4 we look at the nominal
system of Russian to show how the different degrees of morphological
regularity can be represented in terms of the domains to which they apply,
and how morphology, syntax and semantics interact in the animacy-based
syncretisms. In each case, we have implemented our analysis of the lan-
guages in question, an important step, because it allows the reader to check
that the analyses are valid, and so to have a firm basis for evaluating them.
5.1 Network Morphology and syncretism

We shall express our analyses within the Network Morphology frame-
work. It is based on the following principles:
(1) Morphology is lexeme-based.
This means that lexical items in a Network Morphology analysis are
treated as generalizations over their inflectional paradigms.1 Informally,
the rules of morphology are therefore functions from lexical items into sets
of inflected forms. A second important principle associated with Network
Morphology is (2).
(2) Morphology is inferential-realizational.
Network Morphology does not treat affixes (or other inflectional opera-
tions) as lexical entries. Instead these are represented as generalizations in
an inheritance hierarchy. This is a natural way to treat inflection, as it is
therefore straightforward to treat exceptionality and irregularity as asso-
ciated with the lexeme in question, rather than having to encode in each
affix the information about the most irregular item that it can combine
with. We illustrate this further in x5.1.1.
1
We are excluding from this definition idioms and clichés (such as ‘throw down the gauntlet’)
which are listed lexically. However, even for items such as these it is still important to
provide their inflectional variants (such as ‘John threw down the gauntlet’, and so on).
Formal framework and case studies 173
Another property of Network Morphology is that it uses default inheri-

tance networks which can distribute information about morphology in
different parts of the network.
(3) Network Morphology uses default inheritance.
We illustrate how default inheritance works in x5.1.2 The main advantage
of default inheritance theories is that one is able to specify the general
inflectional properties of a language and at the same time include excep-
tions. It also allows one to see the domains over which a particular
phenomenon applies. We shall illustrate this in our case study on
Russian, where we consider the domains of syncretism in that language.
As with all formal frameworks we make a distinction between the fully
specified morphosyntactic paradigm and the particular formal analysis
which accounts for it. The fully specified morphosyntactic paradigm has
the following characteristics:
(4) Feature values are atomic.
(5) Features are ordered relative to each other.2
When we talk of underspecification we mean the relationship of the formal
analysis to the fully specified morphosyntactic paradigm.
(6) The formal analysis may be underspecified relative to the fully
specified paradigms.
Any underspecification of the formal analysis in (6) will have to obey the
constraints of the ordering of features in (5). The ordering of the features
reflects cross-linguistic tendencies, such as those illustrated in our analysis
in Chapter 3: x3.8. Network Morphology uses underspecification and
referrals and combines both as generalized referrals.
(7) Generalized referral:
a. One feature specification may refer to another feature speci-
fication for its realization.
2
There are at least two, logically independent, roles which might be attributed to feature
ordering: (i) underspecification-based syncretism; (ii) the ordering of inflectional elements.
The ordering of features, as we have seen in our discussion of uninflectedness and neutral-
ization in Chapter 2 and the analysis in x3.8, is ideal for underspecification-based syncre-
tism. We are not claiming that all rules of inflectional morphology can be accommodated
with ordered features. Finkel, Shen, Stump and Thesayi (2002) develop KATR a multiset-
based extension of DATR in order to treat instances of morphology for which feature
ordering is not relevant.
b. As with other realization rules, referrals may be underspecified.

c. Extensions of the referring specification will be realized by
extensions of the referred-to specification.
The consequence of (7a) is that there can be instances of directional
syncretism. It follows from (7b) and (7c) that this directionality need not
be limited to individual paradigm cells, but, because referrals are under-
specified, may involve whole sub-paradigms.
While underspecification and generalized referrals, as we have formulated
them, still involve reference to features, we allow for morphomic indexes to
capture non-directional patternings which are independent of features. The
relationship between these indexes and morphosyntax is still constrained by
the ordering of features. In other words, indexed forms must still map to
well-formed feature specifications.
(8) Morphomic indexes must be associated with well-formed feature
specifications.
We shall see how indexes are used in x5.2 in the case study for Dhaasanac.
An important part of the Network Morphology philosophy is that the
analyses should be implementable and testable. For this purpose the
lexical knowledge representation language DATR is used (Evans and
Gazdar 1996). An attractive property of DATR is that it can be used to
represent default inheritance networks, a key organizing principle of the
Network Morphology framework.3
Default inheritance provides for elegant treatments of many problems
within morphology. Corbett and Fraser (1993) and Fraser and Corbett
(1995) showed how default inheritance could be used to analyse animacy,
gender and declensional class assignment in Russian, and this work was
applied to a substantial fragment of Russian nominal morphology. Brown
(1998a) adopted this Network Morphology approach for analysing gender
assignment in Polish, and Fraser and Corbett (1997) implemented a
Network Morphology analysis of gender and morphological class assign-
ment in Arapesh, a language of the Torricelli family, spoken on the north
3
DATR has been used for lexical knowledge representation for a variety of languages. The
DATR-based work on German by Bleiching, Drexel and Gibbon (1996) and Cahill and
Gazdar (1997, 1999) is of particular relevance for syncretism. Cahill and Gazdar (1997:
220–3) discuss rules of exponence and rules of referral in their work on adjectives, determi-
ners and pronouns. This work is extended to nouns in Cahill and Gazdar (1999).
coast of Papua New Guinea. Evans, Brown and Corbett (2002) give a detailed
account of the gender and morphological class assignment system of Bininj
Gun-Wok (a non-Pama-Nyungan language of northern Australia, discussed
above in Chapter 3: x3.5.1). This work included the mother-in-law, or avoid-
ance register, and was applied to a sample of nouns in the language. The
notion of layered defaults was important for that analysis, as certain nouns
require access to what generally holds for their class and to what is the overall
default for nouns as a whole. Other work has shown how the stress system of
Russian can be analysed using default inheritance (Brown, Corbett, Fraser,
Hippisley and Timberlake 1996), and Hippisley (1997, 2001) has applied the
Network Morphology framework to word-formation.
5.1.1 Inferential-realizational theories and morphology

Network Morphology is an inferential-realizational theoretical framework.
This term is due to Stump (2001: 1–30), who develops a typology of morpho-
logical theories which divides them according to two dimensions: lexical
versus inferential, and incremental versus realizational. Lexical theories treat
morphological formants, such as affixes, in the same way as stems and their
associated meanings, by listing both in the lexicon. Inferential theories, on the
other hand, treat the relation between lexical stems and their paradigm of
word forms as a matter of inference, that is of rules or formulas. Incremental
theories require of inflectional morphology that it be information-increasing
so that words build their morphosyntactic specification as the result of
acquiring affixes. Realizational theories, on the other hand, associate a set
of morphosyntactic specifications with lexical stems and these specifications
license morphological exponents, such as affixes.
Stump (2001: 17–27) gives a number of fundamental reasons for prefer-
ring inferential-realizational theories over other theories of inflectional
morphology. What concerns us here are the specific properties of inferen-
tial-realizational theories which make them particularly well suited for
analysing syncretism. Consider the Slovene noun paradigm in (9), repro-
duced from Chapter 4: x4.4.4, which has been discussed by Corbett and
Fraser (1997) in relation to rules of referral.
The question here is how to guarantee that the plural stem ljud- occurs in
the correct contexts in the dual, namely genitive and locative, and not in
the other cases of that paradigm. This is a problem for an incremental
framework, because a narrow feature specification for the stem is the most
natural way of specifying the contexts in which they occur, but this would
stop the addition of the affixes -i and -eh, because they would not
(9) Paradigm of the Slovene noun cˇlóvek ‘person’ (based on Priestly

1993: 401)
NOM človek človeka ljudje

ACC človeka človeka ljudi
GEN človeka ljudi ljudi
LOC človeku ljudeh ljudeh
DAT človeku človekoma ljudem
INS človekom človekoma ljudmi
contribute any new information and would therefore not be information-

increasing. Lexical theories rely on the subcategorization properties of the
affixes which they treat as having lexical entries. The combined suppletion
and syncretism in (9) poses a number of problems for such theories. The
stem ljud- requires a different set of plural affixes from those which would
typically go with a regular noun of cˇlovek’s type. As we can see from the
other cells in the dual paradigm, the stem cˇlovek- can be used in the dual.
The question is what blocks the expected genitive plural/dual and locative
plural/dual affixes for nouns of cˇlovek’s type, -ov and -ih respectively, from
combining with the stem cˇlovek-? The answer for the plural paradigm is
clear enough; there is a lexically specified suppletive plural stem, which is
associated with a different set of endings. So the subcategorization infor-
mation of affixes must be sensitive to the number information on the stem,
even though most nouns do not have different singular and plural stems.
At this point, a lexical framework already has to start listing within
its regular affixes elements of potential exceptionality. For inferential-
realizational theories this problem does not arise, because they do not
treat affixal exponents as lexical entries that require exhaustive informa-
tion on what they may combine with. In fact, inferential-realizational
theories can treat exceptionality where it arises, with the stems of the
lexical items in question, for which the appropriate affixes or other expo-
nents can be inferred. When we continue to the more challenging question
of the dual paradigm, the problems for lexical theories increase. If we
assume that the stem ljud- is specified by some disjunction as also being
the genitive dual and locative dual stem, then the affixes have to be
sensitive to this information. If we do this, of course, then we have encoded
in the affixes information about the most irregular items they can combine
with. If our framework is a lexical-incremental one then things cannot be
worse, because the output information is no more informative than the

subcategorization information. Yet, there is a simple answer to why we
obtain the pattern in (9): for all nouns the genitive and locative dual and
plural are always identical in form. Because they can make reference to sets
of paradigm cells, inferential-realizational theories have no problem mak-
ing simple generalizations like this.
While Network Morphology is an inferential-realizational framework,
it also combines this property with the ability to incorporate varying
degrees of regularity, by analysing morphology in terms of default inheri-
tance, to which we now turn.
5.1.2 Default inheritance

Networks consist of nodes and connections between them. In default
inheritance networks, information is inherited from higher nodes unless
it is specifically overridden. In Figure 3 we present a simple default inherit-
ance network. The diagram is based on a DATR example from Evans and
Gazdar (1996: 176).4 In Figure 3, VERB, EN_VERB, Love, Do, Mow, Sew
and Be are all nodes in the network. The nodes Love, Do, and EN_VERB
inherit from VERB. The nodes Mow, Sew and Be inherit from EN_VERB,
and therefore also from VERB. The nodes in Figure 3 are locations for
information about the morphological paradigms of the items in question.
As such, they generalize the information which classes of lexemes have in
common. The relationships between nodes also make it possible to char-
acterize the degree of exceptionality or lexical idiosyncrasy involved.
VERB
EN_VERB
Love Do Mow Sew Be
Figure 3. A default inheritance hierarchy for some English verbs (based on

Evans and Gazdar 1996)
4
This section discusses default inheritance in DATR. For the purposes of illustration using
English, the examples are based on the fragment published in Evans and Gazdar (1996),
including their use of attributes. This fragment was written to demonstrate DATR, and not
Network Morphology.
A property that all English verbs share is that the passive participle has
the same form as the past participle. Examples (10) and (12) involve a past
participle, while examples (11) and (13) involve passive participles.
(10) I have loved.

(11) I am loved.
(12) I have done this.
(13) This is often done.
We should be able to state this fact as a general property of verbs. The form
of the past tense is generally -ed, but this can be overridden by particular
items, such as do, whose past tense is did. Often, but not always, the past
participle will have the same form as the past tense. There are also subregular
classes, such as the one where the past participle is formed using -en. Default
inheritance allows for a concise treatment of these facts. Evans and Gazdar
(1996: 176) state the following at the node VERB in (14), where we have
omitted some information, as indicated by the ellipses. What is given in (14) is
a representation of the information associated with the top node in Figure 3.
(14) VERB:
<syn cat> == verb
<syn type> == main
<mor past> == "<mor root>" ed
<mor passive> == "<mor past>"
<mor present> == "<mor root>"
<mor present participle> == "<mor root>" ing
<mor present tense sing three> == "<mor root>" s
...
The node name VERB is placed before the colon. Each line containing
‘==’ is a DATR equation. Each left-hand side of a DATR equation
contains paths. Paths contain a combination of ordered attributes. The
right-hand side of the equation may contain values, such as ‘verb’.
Alternatively it may contain paths, or node names, or it may contain a
combination of paths, values and node names.
The first equation at VERB states that the syntactic category of items
belonging to this class is ‘verb’. The equation after this states that the
syntactic type of verb is ‘main’ (i.e. a typical verb is a main verb rather
than an auxiliary). The next equation says that the past is a concatentation
of -ed onto what Evans and Gazdar call the morphological root. The
equation after that says that the passive has the same form as the past.
This is the way referrals are represented in DATR. The equation after the
statement about the passive says that the present uses the morphological
root. This brings us to another important property of DATR: in the
absence of any information to the contrary, we can infer that the value
of a path will be the same as the value for the most specific path of which
that path is an extension. In (15) we give examples of path extension.
(15) <mor present>

<mor present participle>
<mor present tense sing one>
<mor present tense sing two>
<mor present tense sing three>
<mor present tense plur>
All of the paths listed after <mor present> in (15) are extensions of it. At the
node VERB in (14) only two of these extensions are found on the left-hand
side of equations, in addition to the path <mor present> itself. These are:
(16) <mor present participle>

<mor present tense sing three>
This means that we can infer that the values for the first person singular
present tense, the second person singular present tense and the plural
present tense are the same as for the present tense as a whole, namely the
morphological root. This is an example of default inference. The values for
the present participle and third person present tense are not inferred in this
way, as they are already specified at VERB.
From the equation at VERB with left-hand side <mor past> it can be
inferred that the past tense and past participle (which are extensions of
<mor past>) are both formed by suffixing -ed to the morphological root.
We have already seen a referral-based way of stating the identity of the past
participle and passive participle. Here the default syncretism of the past
tense and past participle is the result of underspecification.
We also need to override the default syncretism of the past tense and past
participle. Verbs which have a past participle in -en must inherit the suffixa-
tion of -ed for the past tense but override it for the past participle. This can
be stated at the node EN_VERB which, as we can see from Figure 3, inherits
from the node VERB. Evans and Gazdar (1996: 176) represent this in
DATR as follows:
(17) EN_VERB:
<> == VERB
<mor past participle> == "<mor root>" en.
In (17) the path <> indicates that EN_VERB will inherit all information
from VERB unless otherwise specified. EN_VERB therefore inherits the
specification that <mor past> suffixes -ed onto the morphological root.
This means that the past tense forms for EN_VERB will still be formed
using -ed. As the extension <mor past participle> has a specific value at
EN_VERB, this means that the past participles of verbs of this type will
suffix -en. This is one way in which the default syncretism of the participle
and the past tense forms at VERB is overridden by a particular class of
verbs, and it is represented in terms of path extension, which can be
interpreted as being a strict form of underspecification. It is strict, as
the order of attributes in the path is important for the definition of
extension. (17) also illustrates another point about theories which use
default inheritance, namely that they can involve layered or cascaded
defaults. The overall default for verbs is to have the past participle
and past tense form the same, but another layer of verbs maintains the
default -ed for the past tense, while introducing its own default for the
past participle.
Syncretism can be represented in terms of underspecification and
referral in Network Morphology. An important advantage of this is that
it is possible to combine referrals and underspecification to pick out sets of
cells which are syncretic. In the following sections we discuss how under-
specification relates to semantic naturalness and how to treat systematic
syncretisms. The beauty of an inheritance-based approach, such as
Network Morphology, is that it can represent what occurs over a large
domain of morphology, representing degrees of regularity, from the
domain of a particular word class, or higher, right down to individual
lexical items.
5.1.3 Underspecification and semantic naturalness

Underspecification is often associated with a ‘semantic’ approach to syn-
cretism. However, the feature values involved in syncretism may or may
not be semantically assigned. This is true of gender, for example, in systems
which combine formal and semantic assignment. Under such systems the
relationship between genders x and y might be formal (in that they share
some morphology), and between genders y and z semantic (in that they are
related by meaning). Or genders x, y and z may have a semantic core
complemented by formal assignment for each. Given the cross-linguistic
propensity of genders to collapse (Chapter 3: x3.8), particularly when num-
ber is involved (Chapter 3: x3.4.2), we might also wish to treat gender
syncretism as underspecification for this reason. This means that there are
therefore two different interpretations of what underlies underspecification:
(18) Syncretisms based on underspecification reflect semantic naturalness.
(19) Syncretisms based on underspecification reflect the cross-linguistic ten-
dency of a feature to syncretize (i.e. for different values of that feature to
share identical morphology).
The more appropriate interpretation of underspecification is that in

(19). The reason for this is that underspecification is the natural way of
treating both uninflectedness and neutralization, as we saw in Chapter 2:
x2.4. If we consider the examples of neutralization and uninflectedness
given in Chapter 2: x2.4, none of them can be explained entirely in terms of
the structure inherent to the feature involved. As we have seen in Chapter
3: x3.4.2, there appears to be some involvement of number – typically
plural – and the restructuring of gender, which may sometimes have a
semantic basis. Uninflectedness is a morphological property which is
sensitive to morphosyntactic structure. Under such circumstances under-
specification should be put in the context of other morphosyntactic fea-
tures and is itself not directly interpretable as reflecting a semantic
motivation for syncretism.
It is logical to ask whether syncretism can be determined solely by
semantics. Animacy in Slavonic appeared to be a possible instance. But
as we saw in Chapter 3: x3.1, it is difficult to isolate an on-going semantic
pattern underlying the instances of type 2 syncretism (core case with non-
core case), to which this Slavonic example belongs. Furthermore, the
animacy rule is itself subject to being overridden when items which should
be subject to it belong to a particular inflectional class, showing that
morphology plays a role.5 As Claire Bowern points out (personal commu-
nication) there are instances in various other languages of place names
having their nominative form identical to the locative. However, these are
typically instances of lexically determined syncretism: individual items
have the syncretism, but not all the items in a semantic class. In Tsez,
proper nouns denoting places, provided they are native words, will have
the absolutive identical to one of six local cases (Bernard Comrie, personal
communication). For example, the name of the village Asakh, asaq,
includes the marker -q ‘on (a vertical surface)’ and may function both as
5
The noun djadja ‘uncle’ in Russian, for example, is a masculine animate noun but is not
subject to the animacy rule in the singular, because of its inflectional class membership.
an absolutive and as a local case form. Hence the semantic type allows the
prediction that there will be a syncretism but does not determine which
case will be involved, since this varies from place name to place name.
Moreover, though the form may be used as an absolutive, speakers show
some reluctance here, and prefer to use the name in apposition to a
common noun, which has a clear absolutive. Here again, then, semantics
does not uniquely determine a domain of syncretism. To date we have not
found a language in which a noun denoting a place will necessarily have a
specific syncretism. And in general, we have not found instances of seman-
tics uniquely providing a domain for syncretism.
5.1.4 Systematicity in Network Morphology

Below we offer a model in which systematic syncretism is represented by
the devices of underspecification, indexing and referral, supplemented by
the DATR notion of attribute ordering.
In the analysis which we are proposing underspecification is related to the
ordering of attributes in a path, which in turn is connected with the cross-
linguistic patterns of ordering of features which were highlighted in Chapter
3: x3.8. Hierarchies which rank inflectional features have played an impor-
tant role in other accounts of particular areas of the syncretism spectrum:
for example, Carstairs (1987) treatment of ‘take-overs’ and the Systematic
Homonymy Claim as discussed in Chapter 4: x4.4.2, or Stump’s (2001: 239)
Feature Ranking Principle, as discussed in Chapter 4: x4.4.4. Our formal
analysis is based on the following interpretations:
(20) Network Morphology analysis
a. The order of attributes in a path is associated with cross-linguistic
tendencies of particular features to syncretize. The further to the
right an attribute occurs, the more likely it is to syncretize.
b. The order of attributes in a path will line up with neutralization
and uninflectedness.
c. Underspecification involves resort to an elsewhere form only.
d. Referrals represent directional systematic syncretism within the
language.
e. Indexes represent non-directional syncretism within the language.
f. Syncretisms which violate the order of attributes as required by
(20a) are language-specific referrals (but still systematic).
g. For underspecification-based syncretism there is generally one
form which has more than one function as a default property.
h. For referral-based syncretism there is a default form associated

with a default primary function. The other function of that form
will typically be manifested in a subset of lexical items or mor-
phosyntactic contexts.
i. The adequate treatment of some syncretic patterns requires
simultaneous use of underspecification and referrals, thereby
justifying the theoretical requirement for both. The Slovene
nominal paradigm in (9) and the Dalabon verbal system (x5.3)
involve examples of this type.
The order of attributes in (20a) is established according to cross-linguistic

tendencies, such as those observed in Chapter 3: x3.8. Given atomic values,
we can easily find examples of less straightforward syncretism which line
up with uninflectedness and neutralization, as expected by (20b). Breaking
features up further may obscure this tendency, as we will allow for expla-
nations based more on uninflectedness and neutralization. Because of the
constraint in (20a) and the requirement for atomic values in (5), the
number of possible elsewhere forms is constrained by the number of
features realized. For example, if we considered a language which has
two number values and three or more case values, then it potentially has
two elsewhere forms, a singular one and a plural one. This arises from
ordering number before case (as suggested by the data in Chapter 3: x3.8).
In contrast with this, the potential number of elsewhere forms for Russian
adjective morphology is twelve (a grid of two number values and six case
values), if number is ordered before case, and case is ordered before
gender. If we ignore animacy, then Russian comes close to making use of
the full range of elsewhere forms in the adjectival paradigm: syncretism of
masculine and neuter in the four oblique cases in the singular, and neu-
tralization of all gender distinctions in the plural, leaving just six plural
cells. But this is not the end of syncretism in Russian, because there are
additional patterns related to animacy.
Referral-based syncretism is not constrained by an underlying paradig-
matic design which leads to a limit on the number of elsewhere forms.
However, we would also expect to find some indication of directional effects.
5.2 Case study 1: Dhaasanac

As we saw in Chapter 3: x3.7.2 the Dhaasanac verbal paradigm is an
example of morphological regularity where the syncretized values are
heterogeneous. This is a very good example of ‘morphology by itself’

(Aronoff 1994), as the use of indexes below shows. In the case study we
show how we can capture this autonomous morphological regularity and
still associate it with the relevant morphosyntax. The DATR represent-
ation associated with this analysis is given in Appendix 4. We repeat the
scheme of syncretism in Dhaasanac from Chapter 3: x3.7.2 in (21).
(21) Dhaasanac syncretisms
singular plural
1INCL — A
1 A B
2 B B
3F B A
3M A A
This pattern generalizes across the positive imperfect, positive perfect,

positive dependent and positive simple past paradigms in Dhaasanac. This
is strong evidence for morphological systematicity, which is further rein-
forced by the fact that different verbs have different A and B variants but
still obey the pattern in (21).
The other important aspect of the Dhaasanac data is that they are an
instance of mediated polarity, a term which we introduced in Chapter 3:
x3.7.1. Importantly, the B form is always used for the second person, both
singular and plural. This is therefore an example of underspecification
of number in the presence of second person, but which still needs to
be combined with indexing of the forms. The fully specified paradigms
which we assume for Dhaasanac will be as in (22). Here ‘pos’ stands for
positive, ‘prf’ for perfect, and ‘1st sg’ for first person singular, and so on.
(22) <mor pos prf 1st sg>
<mor pos prf 2nd sg>
<mor pos prf 3rd sg f>
<mor pos prf 3rd sg m>
<mor pos prf 1st_excl pl>
<mor pos prf 1st_incl pl>
<mor pos prf 2nd pl>
<mor pos prf 3rd pl>
The ordering of attributes in (22) reflects the fact that it is negation (or its
absence) and TAM which determine whether there are A and B forms,
because the negated paradigms and the imperative do not distinguish A

and B forms. It is the role of our formal analysis to infer the correct forms
for the positive paradigms.
In (23) we give the lexical entry for the verb fúr ‘to open’, which belongs
to the class of verbs whose stem ends in a coronal consonant.
(23) F˙r:
<> == VERB
<form> == CORONALS
<gloss> == ‘open’
<stem type> == one_vowel
<stem cons_1> == f
<stem vowel_1> == _u
<stem final> == _r.
The node Fúr inherits from the node VERB, which includes the informa-
tion in (24). (We have omitted some of the information given at VERB.)
(24) VERB:
<> ==
<syn> == verb
<index> == _A
<index 2nd> == _B
<index 3rd sg f> == _B
<index 1st_excl pl> == _B
<mor pos imprf> == "<form imprf <index> >"
<mor pos prf> == "<form prf <index> >"
...
Recall from (22) that extensions of the path <mor pos prf>, for example,
will involve attributes for person and number (in that order), and for the
third person singular, also gender. Hence, in the absence of any more
specific information, the form of the perfect paradigm will be determined
by looking at the equation associated at VERB with <mor pos prf>,
namely <mor pos prf> == "<form prf <index> >". The right-hand side of
the equation "<form prf <index> >" involves an evaluable path (the
<index> part). This equation basically means that the morphology of
the positive perfect is determined by looking for the form of the perfect
and inserting the appropriate index for that form. Let us consider the
morphology for the path <mor pos prf 2nd sg>. The rules of inference
mean that, among other things, we can infer (25) from the equation
<mor pos prf> == "<form prf <index> >".
(25) We can infer that:

<mor pos prf 2nd sg> == "<form prf <index 2nd sg> >"
From the information in (24) we can infer (25), and because of the
equation <index 2nd> == _B in (24), and the inference made in (25), we
can infer (26).
(26) <mor pos prf 2nd sg > == "<form prf _B >"
The other TAM, person and number combinations work in the same way.
This is a realizational approach to morphology because it separates out the
morphosyntactic specification from the actual form with which it may be
associated. We should note from (23) that the lexical item Fúr inherits
information about its associated forms from the node CORONALS. Even
though Fúr’s A and B forms may differ from those of other verbs, the
equations at VERB can generalize across all verbs, irrespective of the
specific realization of the A and B forms. In the case of Fúr we can obtain
the forms in (27) for the perfect paradigm.
(27) Fu¤r:<mor pos prf 1st sg> = f _u _r -i.
Fu¤r:<mor pos prf 2nd sg> = f _u _ºº -i.
Fu¤r:<mor pos prf 3rd sg f> = f _u _ºº -i.
Fu¤r:<mor pos prf 3rd sg m> = f _u _r -i.
Fu¤r:<mor pos prf 1st_excl pl> = f _u _ºº -i.
Fu¤r:<mor pos prf 1st_incl pl> = f _u _r -i.
Fu¤r:<mor pos prf 2nd pl> = f _u _ºº -i.
Fu¤r:<mor pos prf 3rd pl> = f _u _r -i.
Thus, by treating the inventory of forms available as separate from the

associated morphosyntax, it is possible to capture the morphological
systematicity of the Dhasaanac verb. At the same time the morphology is
still constrained, because these indexed items have to match up with well-
formed feature specifications.
5.3 Case study 2: The Dalabon verbal system

This case study sets out the core details of syncretism within Dalabon. It is
based on Evans, Brown and Corbett (2001) and uses the same ortho-
graphy. The DATR representation associated with this Network
Morphology analysis is given in Appendix 5.
Dalabon is an Australian language of the Gunwinyguan family, spoken
in central Arnhem Land by a declining population which is now reduced to
about twenty fluent speakers. Dalabon’s verbal system marks intransitive
subjects using bound prefixal morphology, and transitive subjects and
objects using a combination of bound prefixes and clitics. As we saw in
Chapter 3: x3.3 there is strong cross-linguistic evidence to show that

transitive verbs pattern very differently from intransitives in terms of the
person syncretism. This would suggest that transitivity is an important
factor in syncretism on verbs.
Dalabon’s verbal system involves a morphologization of what was
originally a semantically or pragmatically motivated avoidance strategy,
a phenomenon investigated by Heath for Australian languages (Heath
1991) and languages of the Americas (Heath 1998). The avoidance strat-
egy in Dalabon has produced a situation in which certain combinations
of first person subject on second person object or second person subject
on first person object are avoided by the use of person syncretism of the
subject, whereby the appropriate form of the third person subject is used.
There is a directional effect, because the forms used in the syncretism
have third person subject as their primary function, as can be demon-
strated by looking at the intransitive paradigm. Furthermore, the refer-
rals are simultaneously combined with underspecification and predict
whole sets of cells in the extended Dalabon paradigm. This demonstrates
that the best treatment of some syncretisms requires a combination
of referral and underspecification, and so an adequate framework
requires both.
5.3.1 The structure of the Dalabon intransitive paradigm

Dalabon pronominal prefixes mark tense, aspect, mood and clause status.
There are six relevant distinctions: REALIS, SUBORD1, SUBORD2, IRREALIS,
APPREHENSIVE and HORTATIVE. With the exception of certain forms in the
APPREHENSIVE, the six series for tense, aspect, mood and clause status can be
derived by regular rules. The other series can be derived from the SUBORD1
forms. The prefix forms shown in this case study are the REALIS prefixes,
which differ from the SUBORD1 forms in that they terminate with a marker -h.6
The forms linked to certain cells within the transitive paradigm are based
on the corresponding intransitive ones, and so we look at the features of the
intransitive paradigm first. We represent the features involved as a 4 4
matrix, with four person values (FIRST EXCLUSIVE, FIRST INCLUSIVE, SECOND
and THIRD), combined with three values for the number feature: SINGULAR,
6
The values used here are based on those in Evans, Brown and Corbett (2001). Evans has
recently revised the inventory of values, merging SUBORD1 and IRREALIS, and adding a
further value, PURPOSIVE (Nicholas Evans, personal communication). As the relevant issue
for our case study is person syncretism which generalizes across the tense, aspect, mood and
clause status series, this change is not significant for the formal analysis presented here.
DUAL, PLURAL, with an additional feature of harmonicity which has two

values, HARMONIC and DISHARMONIC. This feature marks kinship relations
and is normally restricted in its interpretation to the subjects of intransitive
and transitive verbs, but may occasionally be interpreted in relation to the
object (see Evans, Brown and Corbett 2001: 197–8). Typically, harmonicity
can be seen as a distinction made in the presence of dual number.
The Dalabon intransitive paradigm (REALIS) is given in (28).
(28) The Dalabon intransitive paradigm (REALIS series)
1 EXCL 1 INCL 2 3
SG ngah- — djah- kah-

DU DIS ngeh- djeh- deh- keh-
DU HARM yarrah- yah- narrah- barrah-
PL yalah- ngarrah- nalah- balah-
Logically there cannot be a first person inclusive singular, hence the gap
in the paradigm in (28). From this it follows that person in Dalabon
determines number marking to an extent. We can therefore represent the
relationship between person, number and harmonicity in Dalabon as a
hierarchy in Figure 4. The ordering of attributes in the paths directly
reflects the hierarchy in Figure 4. The hierarchy in Figure 4 is therefore a
representation of the structure of the verbal paradigm for which our
formal analysis should predict the forms. The FIRST INCLUSIVE cannot
have singular forms, and so the ordering allows us to state that there are
no SINGULAR extensions of FIRST INCLUSIVE. It is important to make a
distinction here between the fully specified paradigm in Figure 4, on the
one hand, and the statements required in the formal analysis to infer
the correct forms in that fully specified paradigm. When we turn to the
structure of the transitive paradigm in the next section, we will see that we
need to refer one person value to another, together with its possible
extensions for number. This illustrates the point that referrals and under-
specification are required simultaneously.
5.3.2 The Dalabon transitive paradigm and the inadequacy

of underspecification
There are 102 distinct subject/object combinations, which in turn gener-
ate 102 6 = 612 forms when all TAM combinations are included. This
1ST 1ST INC 2ND 3RD
SG PL DU PL DU SG PL DU SG PL DU
DIS HARM DIS HARM DIS HARM DIS HARM
Figure 4. Paths for person, number and harmonicity in Dalabon
number takes account of the fact that coreferential combinations are ruled
out, as these are encoded by the use of a reflexive/reciprocal suffix with the
intransitive paradigm. There is also a contrast between ‘higher’ and ‘lower’
third person objects when both subject and object are third person singu-
lar. We shall analyse the paradigm section by section.
In (29) we give the forms of the paradigm for the third person transitive
subject, when the object is singular. For comparison, the intransitive
subject forms are given in square brackets above the columns with the
transitive forms. The forms for the second person singular object are
(29) The third person REALIS subject paradigm in Dalabon
subject
3SG 3 DIS 3DU 3PL
object [none kah- keh- barrah- balah-]

1SG kah- keh- burrah- bulah-
2SG djah- djirrah- djirrah- djilah-
bvkah7-
special portmanteau forms which combine information about the subject

and the object. In the case of the paradigm for the second person singular
object it is possible to interpret the syncretism of 3 DIS > 2SG with
7
Used for a higher animacy object.
3DU > 2SG as resulting from the underspecification of the values for
harmonicity.
The disyllabic dual and plural transitive subject forms which occur in
(29) with the first person objects and third person objects are derived from
the intransitive subject forms by the following rule:
(30) CaLah- ! CiLah- if C = y
!CuLah- elsewhere
(where L is a liquid, i.e. l or rr)
The forms derived by rule (30) could be interpreted as marking transitivity,

subject person and number, while being underspecified for the actual
person and number of the object. This is because they can be used with a
third or first person singular object or with non-singular objects of any
person. In the latter case a variety of proclitics are used to mark the object.
The analysis of the forms burrah- and bulah- has consequences for our
analysis of the forms involving the second person singular object. If
djirrah- and djilah- are treated as realizing second singular object combined
with the number of the subject, then the competition between the forms
burrah- and djirrah-, and between the forms bulah- and djilah-, cannot be
resolved using Panini’s principle, as explained in Chapter 4: x4.3, because
the feature specifications would overlap but neither would be more specific
than the other (31).
(31) burrah- (A 3DU), bulah- (A 3PL)
djirrah- (OBJ 2SG, A DU), djilah- (OBJ 2SG, A PL)
The reason that the competition cannot be resolved using Panini’s princi-
ple is that the forms burrah- and bulah- in (31) contain a subject feature
value for third person.
(32) CONFLICT A: which forms should be used for 3DU > 2SG and 3PL > 2SG?
However, we can avoid CONFLICT A, if the value for the subject

person in burrah- and bulah- is deleted, treating these prefixes as marking
only subject number. We would then obtain (33).
(33) burrah- (A DU), bulah- (A PL)
djirrah- (OBJ 2SG, A DU), djilah- (OBJ 2SG, A PL)
Now the specification for burrah- and bulah- is contained in that for
djirrah- and djilah- and so the competition between the two can be resolved
according to Panini’s principle. So far it looks as though the specifications
in (33) will obtain the right result. This appears to be borne out when we
look at other parts of the transitive paradigm, such as that for the first
exclusive subject in (34).
(34) The first person exclusive REALIS subject paradigm in Dalabon
subject
1SG 1DIS 1DU 1PL
object [none ngah- ngeh- yarrah- yalah-]

2SG djah- djirrah- djirrah- djilah-
3SG ngah- ngeh- yirrah- yilah-
The forms for 1DU > 3SG and 1PL > 3SG are obtained from the intransitive
subject forms by the rule in (30). This means that we can assume that they
have the specifications in (35).
(35) yirrah- (A 1DU), yilah- (A 1PL)
Recall from (33) the specifications for djirrah- (OBJ 2SG, A DU) and djilah-
(OBJ 2SG, A PL). As things stand, the specifications for djirrah- and djilah-
overlap with those for yirrah- and yilah- in (35), but neither is more specific
than the other.
(36) CONFLICT B: which forms should be used for 1DU > 2SG and 1PL > 2SG?
The natural resolution of CONFLICT A by underspecification in (32)

was to treat the forms burrah- (A DU) and bulah- (A PL), which are used for
3DU > 2SG and 3PL > 2SG, as not marking subject person. This means that
the option of underspecifying the subject person of yirrah- (A 1DU) and
yilah- (A 1PL) is not open to us in order to resolve CONFLICT B in (36), as
this would then be the same specification as for burrah- (A DU) and bulah-
(A PL). Hence the resolution of CONFLICT A is itself in conflict with the
resolution of CONFLICT B. We could claim that this problem can be
avoided by resort to a principle which prioritizes the realization of the 2SG
object. While the second person singular object is clearly being treated as
special, the problem is that first person objects are not realized by bound
prefixal morphology, and so the putative principle cannot be associated
with a position to the left on a person hierarchy, and would therefore
involve ad hoc stipulation.
There is a further problem for an underspecification analysis when we
turn to the second person subject paradigm in (37).
(37) The second person REALIS subject paradigm in Dalabon
subject
2SG 2DIS 2DU 2PL
object [none djah- deh- narrah- nalah-]

3SG dah- deh- nurrah- nulah-
Like yirrah- and yilah- the forms nurrah- and nulah- are derived by rule (30)
from the intransitive forms and therefore have the feature specifications
in (38).
(38) nurrah- (A 2DU), nulah- (A 2PL)
However, the question arises why the forms burrah- and bulah- are
used for the 2DU > 1SG and 2PL > 1SG respectively. To see this, let us
compare the feature specifications with those of burrah- and bulah- which
were given in (33).
(39) nurrah- and nulah- should fill the 2DU > 1SG and 2PL > 1SG cells according
to Panini’s principle
burrah- (A DU), bulah- (A PL)

nurrah- (2A DU) nulah- (A 2PL)
As the feature specifications in (38) for nurrah- (A 2DU) and nulah- (A 2PL)
are more specific than the feature specifications we assumed for burrah-
(A DU) and bulah- (A PL) in (33), we would expect nurrah- and nulah- to fill
the 2DU > 1SG and 2PL > 1SG cells. However, even if we eliminated the
person information of nurrah- and nulah-, this would not make them less
specific than burrah- and bulah-. Hence, underspecification cannot avoid
the incorrect prediction that nurrah- and nulah- will fill the 2DU > 1SG and
2PL > 1SG cells.
(40) CONFLICT C: which forms should be used for 2DU > 1SG and 2PL > 1SG?
One way to resolve CONFLICT C in (40) might be to alter the informa-

tion content in (38) so that nurrah- and nulah- specify the person of the
object, namely third person. However, a further problem, which is perhaps
not insurmountable, is that the rule in (30) would then be associated
with different morphosyntactic content, depending on the prefix. For
nurrah- and nulah- rule (30) would not only involve the addition of infor-
mation that the prefix is transitive, but also that the person of the object is
third person. On the other hand, for yirrah- (A 1DU) and yilah- (A 1PL) rule
(30) would not involve the addition of information about the person of the
object.
So far we have concentrated on the non-singular subject forms which are
derived by rule (30) from their intransitive counterparts. But it turns out
that the problems we have highlighted for underspecification and the non-
singular forms generalize to all forms, that is to singular and disharmonic
as well. In (30) we present the paradigm of singular subjects and singular
objects.
(41) The paradigm of singular subjects and singular objects (REALIS)
subject
1SG 2SG 3SG
object [none ngah- djah- kah-]

1SG — kah- kah-
2SG djah- — djah-
3SG ngah- dah- kah-
bvkah- (higher animacy object)
For the singular subjects similar problems arise for underspecification. If

djah- may mark the second singular intransitive subject and the object in
the 1SG > 2SG cell (an ergative absolutive pattern), then why is kah- used
in the 2SG > 1SG cell, rather than djah-? We can, of course, specify kah- as
marking a singular subject (with person underspecified), but even then
djah-, which marks second person singular (subject or object), would be
expected to fill the 2SG > 1SG cell, being more specific. This problem is in
some senses worse than CONFLICT A, as in this case it is not that the
feature specifications overlap, but that one is more specific than the other,
and the wrong one (the less specific kah-) wins. The conflict between djah-
(2SG) and ngah- (1SG) to realize the 1SG > 2SG cell can be seen as an
extension of CONFLICT B. Again, we cannot underspecify ngah- for
subject person, as this would then mean it had exactly the same specifica-
tion as kah- (which is the same problem we encountered for yirrah- (A 1DU)
and yilah- (A 1PL)), and here it is not clear that our putative principle, which
treats the marking of second person objects as special, can work, as djah-,
because of its alignment, is not just an object marker. So for a variety of
reasons underspecification fails when used on its own.8
5.3.3 The Dalabon transitive paradigm: a generalized referral analysis

As we showed in Chapter 3: x3.3, the person syncretisms which arise in
transitive verbs are a case apart from intransitives. This suggests that,
where we seek explanations for syncretisms in transitive verbs, these
should be justified on the basis of combinations which are brought about
by transitivity. At the same time we should not dispense with the insights of
underspecification, if it can be applied naturally.
So far we have looked only at those parts of the Dalabon verbal paradigm
involving singular objects. If we lay out the whole paradigm (Table 2), then a
clear pattern emerges.
(42) GENERALIZATION 1: The 1SG > 3SG, 3SG > 3SG, 1 DIS > 3SG and
3 DIS > 3SG forms are the same as the forms for the corresponding
intransitive subject (with the exception of bvkah- which marks 3SG >
3SG with higher animacy objects).
(43) GENERALIZATION 2: The 1DU > 3SG, 3DU > 3SG, 1PL > 3SG, 3PL > 3SG
forms are transparently derived from the intransitive subject forms by
rule (30).
Generalizations 1 and 2 both have a phonological basis since the forms

referred to in (42) are all monosyllabic, and those referred to in (43) are
polysyllabic. It should also be noted that third person singular object
marking cannot be treated as neutralization, as we have defined it in
Chapter 2: x2.4.1, for two reasons. First, third person agreement occurs
with a noun phrase, as in (44) where the comitative verb ‘appear with’
agrees with the noun Naworneng.
(44) ‘‘djarra Naworneng nga-h-ye-burlhmung’’ kah-yini-nj
here Naworneng 1.SG>3.SG-R-COM-appear.PST.PFV 3.SG-say-PST.PFV
‘‘‘Here I’ve brought Naworneng’’ he said.’ (Evans, Brown and Corbett
2001: 200)
8
This conclusion finds independent justification in that Wunderlich (2001a) introduces
‘taboos’ as a constraint in his correspondence-theoretic account of Dalabon. The mechan-
isms involved in this analysis basically replace certain feature values with others by deleting
particular values in the input. This is actually very similar to referrals, thereby demonstrat-
ing that underspecification alone cannot account for the Dalabon data.
Table 2: Paradigm of Dalabon subject þ object combinations, including object clitics; realis TAM series (based on Evans, Brown and Corbett. 2001: 199).
subject
[INTR [/TR]
form] 2 1DU 2DU 3DU 1PL 2PL 3PL
object [free 1 [djah- ] 3 1 DIS 2 DIS 3 DIS [yarrah- ] [narrah- ] [barrah- ] [yalah- ] [nalah- ] [balah- ]
pronoun] [ngah- ] [A dah- ] [kah- ] [ngeh- ] [deh- ] [keh- ] [A yirrah- ] [A nurrah- ] [A burrah- ] [A yilah- ] [nulah- ] [A bulah- ]
1 [ngey] kah- kah- keh- keh- burrah- burrah- bulah- bulah-
2 [njing] djah- djah- djirrah- djirrah- djirrah- djirrah- djilah- djilah-
3[—] ngah- dah- kah- ngeh- deh- keh- yirrah- nurrah- burrah- yilah- nulah- bulah-
BVKAH-
1DU [njerr] njerr njerr njerr njerr njerr njerr njerr njerr bulah-
kah- kah- keh- keh- burrah- burrah- bulah-
2DU [norr] norr norr norr norr norr yirrah- norr norr yilah- norr bulah-
ngah- kah- ngeh- keh- burrah-
3DU bunu bunu bunu bunu bunu bunu bunu bunu bunu bunu bunu bunu
[bu(l)nu] ngah- dah- kah- ngeh- deh- keh- yirrah- nurrah- burrah- yilah- nulah- bulah-
1PL[njel] njel kah- njel kah- njel njel keh- njel njel njel njel bulah-
keh- burrah- burrah- bulah-
2PL [nol] nol nol kah- nol ngeh- nol keh- nol yirrah- nol burrah- nol yilah- nol bulah-
ngah-
3PL [bulu] bulu bulu dah- bulu bulu bulu bulu bulu yirrah- bulu nurrah- bulu bulu yilah- bulu bulu bulah-
ngah- kah- ngeh- deh- keh- burrah- nulah-
The combinations with first inclusive have been omitted from this table (for which see Evans, Brown and Corbett 2001: 199). The prefixal marking in bold is
determined by the referrals (56) and (57).
A second reason is the form bvkah- which marks a higher animacy third
person object. As this form is used uniquely for third person objects
(in the presence of a third person subject), it suggests that third person
is a valid syntactic distinction for objects in Dalabon. Both of these
facts indicate that object person marking in the singular is not irre-
levant for syntax, which therefore means that this cannot be seen as
neutralization.
The next set of generalizations concerns the first person singular object.
(45) GENERALIZATION 3: The 3SG > 1SG and 3 DIS > 1SG forms are the
same as the forms for the corresponding intransitive subject.
(46) GENERALIZATION 4: The 3DU > 1SG and 3PL > 1SG forms are trans-
parently derived from the intransitive subject by rule (30).
(47) GENERALIZATION 5: Generalizations 3 and 4 mean that the forms of
3 > 1SG are the same as the forms of the 3 > 3SG paradigm (with the
exception of bvkah-).
(48) GENERALIZATION 6: Generalizations 3 and 4 mean that there is no
marking of first person singular objects in the bound pronominal
morphology.
Generalizations 3–6 might lead one to assume that first person singu-
lar object is not a morphosyntactically relevant entity. However, if it
played no role, or indeed were non-existent, then CONFLICT C in (40)
could not arise. This point itself is independent of how we actually
choose to treat CONFLICT C, by underspecification or otherwise. If,
for example, first person object were a syntactically non-existent entity
then we would have no need to refer to it in the paradigms and also in the
formal analysis associated with them. But this would mean that the
contrast between the forms kah- (2SG > 1SG) and dah- (2SG > 3SG) in
(41) could not exist. Of course, the prefixes kah- and dah- are used for
other purposes and would not be treated in most theories as fully speci-
fied for the feature values 2SG > 1SG and 2SG > 3SG respectively.
Examination of Table 2 shows that there is always a contrast between
2 > 1 and the 2 > 3 paradigm, and this means that reference must be
made by morphology to first person objects. As for the second person, we
know that objecthood is important there, because of the special port-
manteau forms.
Having demonstrated the inadequacy of underspecification alone, we
now turn to the analysis in Evans, Brown and Corbett (2001) and our
representation of it. That analysis resolves CONFLICT A by specifying
the feature values for burrah-, bulah-, djirrah- and djilah- as in (49).
(49) burrah- (A 3DU), bulah- (A 3PL)

djirrah- (OBJ 2SG, A 3DU), djilah- (OBJ 2SG, A 3PL)
What we have done in (49) is to add the second person object values to
djirrah- and djilah-. This means that djirrah- and djilah- are more specific
and will therefore fill the 3DU > 2SG and 3PL > 2SG cells.
We now turn to the resolution of CONFLICTS B and C. These are both
resolved by the use of rules of referral which are related to the combination
of subject and object persons. Recall that we have seen cross-linguistic
evidence to suggest that person syncretism in two-place verbs is different
from that in intransitives, which strongly suggests that transitivity plays an
important role. Because the rules of referral are formulated in terms of the
combination of subject and object person they account for the syncretisms
observed as resulting in part from transitivity. The rules of referral are
as in (50)
(50) a. 2[n=] > 1[n=] uses the form for 3[n=] > 1[n=]
b. 1[n=] > 2SG uses the form for 3[n=] > 2SG
where n is the number value
(Evans, Brown and Corbett 2001: 207)
The effect of (50a) is to say that the 2 > 1 paradigm uses the forms of
the 3 > 1 paradigm. The second line (50b) states that the 1 > 2SG
paradigm uses the forms of the 3 > 2SG paradigm. Hence, in this case
the original motivation for the referral is the avoidance of combinations
of first and second person objects, the top end of a person hierarchy. The
use of variables , here is for expository purposes. The point is that the
referral does not just involve one cell of the paradigm referring to
another. Rather it is a combination of a referral and underspecification.
A generalized referral of this kind can then be used to predict whole
sub-paradigms.
The referrals in (50) are based on the notion of ‘avoidance’ as elaborated
by Heath (1991 and 1998).
The assorted mechanisms . . . have in common the fact that they
obscure the ‘objective’ relationship between speaker and addressee . . .
The 1st « 2nd combinations are doubly dangerous because they not only
contain the most pragmatically sensitive pronominals, they also combine
them into a syntagmatic structure and thereby necessarily focus attention
on the speaker–addressee relationship . . . The Australian languages . . .
play down the speaker–addressee relationship by omission, substitution,
or skewing of the normal, most transparent, hence also bluntest first and
second person morphemes. (Heath 1991: 86)
Dalabon has actually taken this a step further by morphologizing the

avoidance strategy, so that there are now no longer unique forms for the
2 > 1 and 1 > 2SG paradigms. This means that one can no longer treat
these effects as facts of the ethnography of communication in the Dalabon
community, and therefore in the same way as the use of polite Sie in
German, for example. The politeness rules in German and similar lan-
guages can refer directly to the free pronoun (regardless of whether a verb
is present in the utterance): the primary effect is that one free pronoun is
replaced by another (du by Sie, and so on), and the effects on the morphol-
ogy are secondary, simply involving the verb agreeing with the new pro-
noun in its (etymological) person and number – third plural in German (for
interesting complications see Comrie 1975). However, in Dalabon, free
pronouns themselves are unaffected by the referrals in (50) which only
affect the agreement patterns in verbal prefixation. Consequently, what
was once pragmatically systematic in Dalabon has become morphologi-
cally systematic syncretism.
Under this analysis, CONFLICT B, in both its specific and its general-
ized variants, does not arise, because the 1SG > 2SG, 1DU > 2SG and 1PL >
2SG cells are subject to the referral in (50b). The correct forms are selected
by rule (50b). As djah- is the form used for 3SG > 2SG in (29), it will fill the
1SG > 2SG cell in (41). As djirrah- is the form used for the 3DU > 2SG in (29)
it will be used for the 1DU > 2SG cell in (34), and as bulah- is used in the 3PL
> 2SG cell in (29), it will fill the 1PL > 2SG cell in (34). These correct
predictions all follow from the single referral in (50b).
Conflict C, in both its specific and generalized variants, is resolved by
the rule of referral (50a). The forms used for the 2DU > 1SG and 2PL > 1SG
cells are burrah- and bulah-, because these are the forms used for the 2DU >
1SG and 2PL > 1SG cells in (29). The reason why kah- is used for the 2SG >
1SG cell in (40) is that this is the form used for the 3SG > 1SG cell. In the case
of referral (50a) it generalizes to all first person object cells where the
subject is second person. This can be verified by checking Table 2 where
the dual first person object proclitic njerr and the plural first person object
proclitic njel occur in the non-singular first person object portion of the
2 > 1 paradigm, with in each instance the bound pronominal being the
same as for the corresponding 3 > 1 paradigm (for example, njerr kah-
is both 2SG > 1DU and 3SG > 1DU, and njel burrah- is both 2DU > 1PL and
3DU > 1PL).
With the adoption of the referrals in (50), the conflicts resulting from the
use of underspecification on its own are resolved. However, the referrals
themselves involve a degree of underspecification, because the number of the

argument is underspecified (as indicated by the use of the variables in our
informal representation in (50)). Underspecification of number in the pre-
sence of person makes cross-linguistic sense, of course, because we know that
number marking correlates with person values. The referrals themselves also
imply a directionality, in that burrah- (A 3DU), bulah- (A 3PL), djirrah- (OBJ 2SG,
A 3DU) and djilah- (OBJ 2SG, A 3PL), for example, are now treated as the
primary exponents of third person. There is evidence for this. The forms
barrah- (S 3DU), and balah- (S 3PL), from which burrah- (A 3DU) and bulah-
(A 3PL) are derived by rule (30) are uniquely associated with the third person
in the intransitive paradigm. In the singular, kah- is also uniquely associated
with the third person in the intransitive paradigm, while djah-, which marks
second person singular object in the transitive paradigm, is uniquely asso-
ciated with the second person singular subject in the intransitive paradigm.
There is no direct evidence that the forms djirrah- and djilah- are primary
exponents of 2SG > 3DU and 2SG > 3PL respectively, but as they fill these
cells, the referral (50b) will make reference to them in any event. While
underspecification alone does not work, the generalized (underspecified)
referral approach derives the correct results, is favoured by the paradigm-
internal evidence and also has a potential original motivation in terms of
pragmatically based avoidance based on the top two values of a person
hierarchy.
We now turn to a brief account of the implementation of this analysis
using DATR.
5.3.4 The verbal hierarchy

This section takes one sample Dalabon verb nan ‘to see’. Figure 5 illus-
trates a small part of the lexemic hierarchy in which the verbal lexical items
inherit from a node VERB. Like every other verb, verb nan ‘see’ inherits
from the node VERB. Its lexical entry is given in (51).
VERB
Nan OTHER
VERBS
Figure 5. Dalabon verbal hierarchy

(51) Nan:
<> == VERB
<gloss> == see
<root> == nan.
In (51) the ‘empty path’ <> is paired with the node VERB, on its right-hand
side. This simply means that information which is not specified at the node
9
Nan will be inherited from VERB. In (52) we give the node VERB.
(52) VERB:
<> == undefined
<mor> == MOR_VERB
<syn> == SYNTAX
<syn cat> == verb.
This node serves to bring together the morphological and syntactic speci-
fication of verbs in Dalabon. In (52) the path <mor> refers to the path
<mor>, and all of its extensions, at MOR_VERB, the node which provides
information about the morphology of verbs. It is the information repre-
sented at this node which forms the core of our analysis of the Dalabon
verbal system and it is to this that we now turn in the next section.
5.3.5 The shape of the verbal paradigm

Recall from x5.1.2 that the order of the attributes in our DATR represen-
tation is important, because, as a highly constrained kind of underspecifi-
cation, it determines what can be inferred by default. We saw in (28) in
x5.3.1 that there is no person syncretism in the intransitive paradigm.
However, there are still constraints on the possible combinations of person
and number. Specifically, person imposes certain constraints. First person
inclusive is incompatible with singular, because the lowest cardinality
associated with it is two, although morphologically the first inclusive
dual and the first inclusive plural look like the singular and dual forms
respectively of the other persons.
As person determines which numbers are available, we order the person
attributes before the number attributes, for a given argument (intransitive
subject, transitive subject and object).
9
A path cannot be paired with a node without also making reference to another path and its
extensions at that node. By convention, where a path is paired with another node and no
overt reference is made to a particular path at the node referred to, then the referring path
refers to the identical path at the node referenced. In sum, the empty path at the node Nan
refers to the empty path, and its extensions, at VERB. Furthermore, the value for any
extension of a path which is not already specified at Nan will be found by looking for a
matching path at VERB.
(53) For a given argument of the verb, person attributes are ordered before
number attributes.
Given the assumption that underspecification is related to what is cross-

linguistically general, (53) entails that there should be some reflex of this
cross-linguistically general relationship between person and number in
Dalabon. Typically, this would be syncretism of number in the presence
of person, but this is not the case for Dalabon. The ordering in (53) is
related to our informal description of the referrals in (50). There we used
variables to show that the referral does not just involve one cell of the
paradigm referring to another, but in fact picks out whole sub-paradigms.
The fact that it picks out whole sub-paradigms, of course, shows that the
sub-paradigms are defined in terms of person. While avoidance of 1 > 2SG
and 2 > 1 is more specific to Dalabon, the involvement of whole sub-
paradigms defined by person will follow from the cross-linguistically gen-
eral principle in (53).
The next step is to determine whether the object attributes should occur
before the transitive subject attributes in the paradigm. This question
obviously does not arise for intransitive subjects. If we were to base our
decision on the ordering of affixes, then the fact that non-singular object
clitics appear before the bound prefixes might lead us to claim that object
attributes (followed by their person and number attributes) should occur
first in the path. We argue, in fact, that this is the case for third and second
person object, but not for first person exclusive object.
Only when the object is third or second person may there be clear
marking of the person of that object on the bound pronominal prefixes.
So the second person singular object is clearly marked on the portmanteau
bound pronominal prefix. Furthermore, in the presence of second person
singular object marking, the disharmonic transitive subject is syncretic
with the dual transitive subject. There is also an example of a bound
pronominal prefix clearly marking third person: the third person singular
higher animacy object (Table 2). The form bvkah- is used only when the
object is third person singular, not when it is first person exclusive (41), as
indicated by the contrast between the 3 > 1 and 3 > 3 paradigms. Hence
this form definitely identifies a third person object. In contrast with this
there is no bound pronominal prefix form where an object is clearly
marked as first person. From this we conclude that objecthood triggers
marking of second and third person singular. For the second person
singular it also has an effect on the number marking of transitive subjects.
We therefore order attribute combinations of object with second and third

person before any attribute combinations of transitive subject (and any
associated person and number marking).
There is no unambiguous bound prefixal marking of the first person
exclusive object. The bound pronominal system has the same forms for the
3 > 3SG paradigm and the 3 > 1SG paradigm (Generalizations 3–6). In
other words, for the bound pronominal paradigm third person transitive
subjects condition loss of distinction between first person singular and
third person singular objects. This leads us to claim that, in contrast with
the other persons, the object attributes for first person exclusive occur last
in the path. We shall relate these facts to the referrals in (50). The resulting
combination of values in the paradigm is therefore as in (54)
(54) Second and third person object attributes are ordered before transitive
subject attributes10
Examples:
<mor infl o 3rd sg a 2nd sg>
<mor infl o 3rd du a 2nd sg>
<mor infl o 2nd pl a 1st_excl sg>
In contrast, first person objects are never distinguished by the bound pro-
nominal morphology and so these are treated as occurring last in the path.
(55) First person object attributes are ordered after transitive subject
attributes
Examples:
<mor infl a 2nd sg o 1st_excl sg>
<mor infl a 3rd sg o 1st_excl du>
<mor infl a 2nd du o 1st_excl pl>
In (54) and (55) we have examples of the shape of the fully specified
paradigm. The attribute ordering reflects what is most likely to be under-
specified. We can contrast the avoidance-motivated referral involving
the top two positions on the person hierarchy (first and second) with the
orderings in (54) and (55), which oppose the lower two positions on the
person hierarchy (second and third) with the top position, first person exclu-
sive. The analysis which derives the fully specified combinations, such as those
in (54) and (55), can be underspecified in relation to these combinations.
10
Following Dixon (1994) in our representation we shall use the attributes a, s and o for
transitive subject, intransitive subject and object respectively.
5.3.6 Referral of 1 > 2SG to 3 > 2SG

The referral of the 1 > 2SG to the 3 > 2SG combinations illustrates that
syncretisms can arise from what is essentially a referral with a degree of
underspecification. The underspecified element is the number of the tran-
sitive subject. Note that this referral also fits in with the attribute ordering
we have determined in (54). As the second singular object attributes are
toward the left of the path, they are less likely to be involved in under-
specification-based syncretism. This referral is given in (56) below.
(56) <mor infl o 2nd sg a 1st_excl> == <mor infl o 2nd sg a 3rd>
All extensions of paths for 1 > 2SG can be obtained by referring to the
extensions of 1 > 3SG. These extensions will, of course, be those for number
of the transitive subject, as this is what is not specified in the referral.
5.3.7 Referral of 2 > 1 to 3 > 1

We claimed in (55) that the first exclusive object attributes should be
specified as occurring last in any path. We argued that this had to do
with the fact that there is no person-specific bound pronominal marking
for the first exclusive object, in contrast with the second and third person at
least, where the second singular object is clearly marked and there is
marking of the higher animacy third singular object. As this means that
first exclusive object marking is an extension of the third person paradigm,
the referral of 2 > 1 to 3 > 1 takes the form in (57).
(57) <mor infl a 2nd> == <mor infl a 3rd>
This states that the second person transitive subject finds its extensions
from the third person transitive subject. If third person object attributes
were also ordered after transitive subject attributes, this would mean take-
over of the 3 > 3 paradigm by the 2 > 3 paradigm. But this is not the case,
because from (55) it follows that only first exclusive object attributes can
extend a path which starts with transitive subject attributes, such as in (57).
Note the high degree of underspecification involved. The referring para-
digm of second person transitive subject obtains its transitive subject
number, together with the first exclusive object person and number infor-
mation, from the third person paradigm.
5.3.8 Dalabon: summing up

The referral-based approach allows us to capture the intuition that the
syncretism is motivated by the potential inappropriate combination of first
and second person. It is transitivity which conditions the loss of person

distinction. It also allows us to treat number marking in the transitive and
intransitive paradigms in a similar way. Number distinction is affected by
person (first inclusive) in the intransitive paradigm, which makes full
person distinctions. The best account requires the use of referrals and
underspecification so that whole sub-paradigms may be systematically
syncretic. Thus both underspecification and referral are required. It is
important to check that our Network Morphology analysis does indeed
provide the correct forms for the substantial paradigms of Dalabon. Hence
the need for an implementation, a part of which was presented above. The
full implementation in Appendix 5 does give exactly the right forms for
Dalabon, and only these, and to that extent the Network Morphology
account is valid.
Having seen how underspecification can be combined with referral to
provide the best account of syncretism in the Dalabon verbal system, we
now go on to consider the relationship between position on a morpholo-
gical class hierarchy and the domains of syncretism. For this we shall use
Russian as the main case study, referring to other languages when data
from Russian is insufficient.
5.4 Case study 3: The Russian nominal system

The complex inflectional morphology of Russian offers an illuminating
case study (58). Its system of inflectional classes provides helpful data for
(58) Main paradigms of Russian nouns
class I ‘law’ class II ‘room’ class III ‘bone’ class IV ‘wine’
NOM SG zakon komnata kost0 vino

ACC SG zakon komnatu kost0 vino
GEN SG zakona komnaty kosti vina
LOC SG zakone komnate kosti vine
DAT SG zakonu komnate kosti vinu
INS SG zakonom komnatoj kost0 ju vinom
NOM PL zakony komnaty kosti vina
ACC PL zakony komnaty kosti vina
GEN PL zakonov komnat kostej vin
LOC PL zakonax komnatax kostjax vinax
DAT PL zakonam komnatam kostjam vinam
INS PL zakonami komnatami kostjami vinami
discussing the notions of domain and regularity. Russian has four basic
noun paradigms, as given in (58).
In a traditional account, these paradigms might be treated as monolithic
units. In a Network Morphology account, we would analyse the oblique
plural forms as being shared across the paradigms; thus the dative plural of
a noun consists of its stem plus -am, irrespective of inflectional class. We
would also treat the nominative plural as being stem plus -i (or its ortho-
graphic variant -y), with this being overridden just for inflectional class IV.
At a lower level, we would capture the shared forms of inflectional classes I
and IV. That is, we would have a hierarchy of defaults, the highest apply-
ing very generally, in fact applying to more than just nouns, the lowest
having smaller domains, and at the bottom of the hierarchy would be
lexical items which must contain some idiosyncratic information. A pos-
sible structure is given in Figure 6. In addition to the four lexical items
given, there are thousands of other nouns which inherit from the four
inflectional class nodes. For the detail see Corbett and Fraser (1993). The
implementation on which this case study is based is from Brown (1998b),
and a simplified version of this is given in Appendix 6.11 The relevance of
this approach is that it suggests a range of possible domains for syncretism.
MOR_NOMINAL
MOR_ADJECTIVE MOR_NOUN
N_0
N_I N_IV N_II N_III
zakon … vino … karta … kost´…
Figure 6. An inheritance structure for Russian nominals
11
The fragment rusnoms.dtr at the Sussex DATR archive http://www.cogs.sussex.ac.uk/lab/
nlp/datr/datr.html precedes Brown (1998b) but still involves separate hierarchies. Corbett
and Fraser (1993) gave a single hierarchy. Later papers use a network of hierarchies; in
particular a syntactic hierarchy is added, which is concerned with the syntactic category of
items; by default, items which are, for example, syntactic nouns will inherit information
from the noun section of the morphological hierarchy, but this is not always so.
The higher up the hierarchy that a syncretism is stated, the larger the
domain of syncretism. And the larger the domain, the more systematic
the syncretism.
5.4.1 Domains of syncretism

Given that we find numerous instances of syncretism, what are the possible
domains? We first separate off phonology and then move to the central
potential domains.
5.4.1.1 Phonologically determined ‘syncretism’

Certain syncretisms can be excluded on the grounds that they result solely
from a phonological rule; for example, the collapse of genitive singular
with nominative/accusative singular in some stem-stressed nouns in
Russian, discussed in Chapter 1: x1.5, which is due to the reduction of
unstressed vowels. Such instances can be straightforwardly accounted for
by the phonology and therefore excluded from treatment within morphol-
ogy proper.
5.4.1.2 Lexically determined syncretism

The least systematic type of syncretism (that involving the smallest
domain) would be a syncretism for a single lexical item. We do not find
such an instance in Russian. To illustrate what this would look like,
consider the paradigm of Italian essere ‘be’ in (59).
(59) Italian essere ‘be’ in comparison with a regular verb
‘be’ ‘speak’
1 sono siamo 1 parlo parliamo

2 sei siete 2 parli parlate
3 è sono 3 parla parlano
According to Davide Ricca (personal communication) the verb essere

‘be’, which has first person singular and third person plural both as sono, is
the only verb in the language to have this identity. This then is a case of
lexically determined syncretism, which is equivalent to Coleman’s (1991)
first degree syncretism. Other languages show further unique syncretisms,
though these are rare.
In principle the situation is the same, whether it affects a single item or

several items in an unmotivated way: that is, the syncretism would need to
be specified in the lexical entry for each of them. Here is a further example
of lexically determined syncretism. Numerals in Russian vary in the num-
ber of case distinctions made, but the normal minimal paradigm distin-
guishes three forms. Consider these numerals against that general
background:
While syncretism of nominative and accusative is widespread in Russian
numerals, syncretism of all the remaining case forms is unusual, and is the
only morphological feature these items have in common. This combina-
tion of syncretisms including the instrumental must be lexically specified
for these three items only.12
(60) Lexically determined syncretism in Russian numerals
‘forty’ ‘hundred’ ‘ninety’
NOM sorok sto devjanosto

ACC sorok sto devjanosto
GEN soroka sta devjanosta
DAT soroka sta devjanosta
LOC soroka sta devjanosta
INS soroka sta devjanosta
5.4.1.3 Morphologically determined syncretism

In this part of our case study of Russian we illustrate how morphological
regularity is itself a matter of degree. Referring back to the Russian
paradigms in (58) we now consider morphological domains, starting
from the smallest. The smallest morphological domain above that which
is lexically determined is syncretism within a single inflectional class. We
find this in the class we have labelled class III in (58), where there is, among
other things, identity of genitive singular and locative singular, repeated
here in (61).
12
Sorok ‘forty’ and sto ‘hundred’ are the clear cases; because of vowel reduction all the forms
of devjanosto ‘ninety’ are pronounced identically, thus there is an additional phonolo-
gically induced syncretism of the nominative and accusative with all the other cases, giving
only one phonological form for this item. Note also that GEN/DAT/LOC/INS syncretism is also
characteristic of the singular feminine adjectives; see (67) below.
We could reflect this in the lexical entry. However, we would be suggest-

ing, counterfactually, that the identity is an idiosyncratic fact about the
particular noun kost0 ‘bone’, which is not shared with other items, in other
words that the syncretism is lexically determined. In fact, the syncretism of
(61) Syncretism within a single inflectional class in Russian
class III ‘bone’
NOM SG kost0
ACC SG kost0
GEN SG kosti
DAT SG kosti
LOC SG kosti
INS SG kost0 ju
locative and genitive singular is one of the characteristics shared by all

members of the inflectional class (Coleman’s 1991 second degree syncret-
ism). We should therefore state this syncretism at a higher point in the
inheritance hierarchy in Figure 6, at the node N_III, from which kost0
inherits (as do over 4,000 other nouns). This syncretism is evidently more
systematic than any lexically stipulated syncretism. It is, however, less
systematic than instances we shall come to shortly.
There is also a good reason to treat the syncretism in (61) as directional.
This is represented with the DATR equation in (62). Again, we have
omitted some information, as indicated by the ellipses.
(62) N_III:
<> == MOR_NOUN
<mor sg loc> == "<mor sg gen>"
...
The equation in (62) is a referral which states that for class III the form of the
locative singular is determined by the form of the genitive singular. The
reason for this is that the form of the genitive singular -y in class II, although
orthographically different in (58), can be treated as phonologically identical
with the exponent -i of class III, suggesting that the stem +i combination is
primarily the exponent of genitive singular, and secondarily in class III the
exponent of locative singular. This analysis is further justified when we
consider syncretism which ranges over more than one inflectional class.
In Russian, as can be seen in (58), dative and locative singular are

identical in two inflectional classes (II and III). We repeat the relevant
paradigms in (63).
(63) Syncretism within a subset of the inflectional classes of a word class
II ‘map’ III ‘bone’
NOM SG karta kost0

ACC SG kartu kost0
GEN SG karty kosti
DAT SG karte kosti
LOC SG karte kosti
INS SG kartoj kost0 ju
While the case forms are identical within the inflectional classes, the
inflections involved differ between inflectional classes. This example is
more systematic than that involving just class III. It is more systematic in
two ways. First, many more nouns are involved, all of those in classes II
and III. We state the syncretism at the MOR_NOUN node (and then override
it for the remaining nouns which inherit from N_O in Figure 6, see Brown
(1998b: 257)). Second, the syncretism holds true for two quite different
morphological realizations: those nouns like karte ‘map.DAT SG/LOC SG’ and
those nouns like kosti ‘bone.DAT SG/LOC SG’, as shown in (63).
If we refer back to the example paradigms in (58), we see that the locative
singular realization combining a stem with the ending -e is the default for
nouns as a whole, because it is found in classes I, II and IV. Furthermore,
in classes I and IV, the combination of a stem with the ending -e is reserved
solely for the locative singular. This is repeated in (64).
(64) The ending -e as primary exponent of locative singular
I ‘law’ IV ‘wine’
NOM SG zakon vino

ACC SG zakon vino
GEN SG zakona vina
DAT SG zakonu vinu
LOC SG zakone vine
INS SG zakonom vinom
Examination of the paradigms in (63) and (64) suggest that the combi-
nation of stem and ending -e should primarily be associated with loca-
tive singular as a default for nouns.
The DAT/LOC syncretism is therefore a referral of dative singular to locative
singular. This can be expressed as in (65).
(65) MOR_NOUN:
<> == MOR_NOMINAL
<mor sg dat> == "<mor sg loc>"
<mor sg loc> == "<stem sg>" e "<stress sg>"
...
Class II, which includes nouns such as karta ‘map’, inherits both equations
and therefore combines the syncretic pattern with the default ending -e.
The intermediate class N_O, from which class I and class IV inherit,
overrides the referral by stipulating that the realization of dative singular
is the stem plus the ending -u. Class III, on the other hand, overrides the
default realization of the locative singular, while still inheriting the default
referral of dative singular to locative singular. When the referral in (65) is
combined with the referral specific to class III in (62) we obtain a chain in
which the dative singular is based on the locative singular, and the locative
singular is based on the genitive singular, thereby yielding the collapse of
three case distinctions in class III. Analytically this is important, because
the identity of genitive and dative in class III is the product of combining
the noun default referral for DAT/LOC with the class III referral for LOC/GEN.
With the original Jakobsonian features the DAT/LOC syncretism can be
captured only by leaving the realization specified as +peripheral.
However, this would also include the instrumental case. In contrast, it is
easier to pick out the LOC/GEN syncretism. The GEN/DAT syncretism in class
III is problematic for the Jakobsonian feature system, as these cases do not
share any values under the unmodified version. It is possible to express
some of these generalizations using a modified variant of the Jakobsonian
approach (Chapter 3: x3.1.1). The problem is that a decision must be made
about which generalizations are to be captured, and this involves preclud-
ing other phenomena. For instance, in Müller’s (forthcoming) approach to
syncretism and inflectional allomorphy in Russian, there are two separate
vocabulary items, /e/4 and /e/5, the former used for the dative/locative of
class II and the latter used for the locative of classes I and IV. The reason
for this is that, even though dative and locative form a natural class
[-subj, +obl], Müller also attempts to capture sharing between inflectional
classes by using variables, and the variables for classes I and IV exclude
those for class II. In fact, scaling up underspecification-based approaches
is often a problem when taken in the context of other features. Breaking
the features up even further to make underspecification work brings with it
the potential for massively increasing the number of elsewheres, which the
system underutilizes in any event. When taken in isolation this problem for
theories which manipulate the feature values is not as great as when
number and case, or lexical information about inflection class, are actually
combined, thereby yielding a massive number of potential combinations
which never appear. In contrast, the analysis of the noun syncretisms based
on atomic values and referrals actually captures the licensing effect that
one syncretism may have on others.
Having seen an example of a syncretism which covers more than one
morphological class within a word class, we now turn to a more general
domain, namely to potential examples of syncretism throughout a word
class. For a clear case we turn to Russian’s South Slavonic relative, Serbo-
Croat. Here the accusative plural and genitive plural are syncretic for the
personal pronouns, but not for nouns and adjectives. In (66) we illustrate
this syncretism using the first and third person pronouns (non-clitic).
(66) Syncretism throughout one word class in Serbo-Croat
1PL 3PL
NOM mi oni (M), ona (N), one (F)

ACC nas njih
GEN nas njih
DAT nama njima
LOC nama njima
INS nama njima
The other pronouns behave in the same way with regard to the PL ACC/PL
GEN syncretism. Thus we have an instance which involves all members of
the word class. While the word class involved in Serbo-Croat is small with
respect to the number of lexical items involved, we have already seen
an instance where the lexical class was much larger, namely Finnish
(example (13) in Chapter 3: x3.1.2), where syncretism affects all nouns,
but not pronouns.
LOC PL / GEN PL
MOR_NOMINAL
LOC SG / DAT SG
MOR_ADJECTIVE MOR_NOUN
N_0
LOC SG / GEN SG
N_I N_IV N_II N_III
zakon … vino … karta … kost ′…
Figure 7. The location of syncretism patterns in the Russian inheritance

hierarchy
We also find syncretisms which generalize across more than one word
class. In Russian all adjectives and pronouns have the genitive plural
syncretic with the locative plural. This does not extend to nouns, where
the inflected forms differ, as can be seen in (58).
Finally, it is possible to find examples of syncretism across all potentially
relevant word classes. In Slovene, the dative dual and instrumental dual
are syncretic for nouns, adjective and pronouns (Priestly 1993: 399). That
is to say, these forms are syncretic for anything which can mark them. No
lexical item of Slovene has a dative dual which is distinct from the instru-
mental dual. The type of inheritance hierarchy illustrated in Figure 6
suggested possible morphological domains for syncretism, and we have
found instances of all of them. In Figure 7 we show where the general-
izations are located on the inheritance hierarchy for Russian. The syncre-
tism of locative plural and genitive plural is a property of nominals,
because it occurs for pronouns as well as adjectives, but it is overridden
by nouns. The locative singular and dative singular syncretism is a default
for nouns, but overridden at the node N_O, and the syncretism of the
locative singular and genitive singular is a default for class N_III only and
is specified there. The locations in the hierarchy represent different degrees
of regularity. The higher up the hierarchy a syncretism is stated, the greater
the number of items that are likely to be affected. The second aspect to
regularity, namely that with inflectional classes a syncretism may general-
ize over different realizations, was illustrated in (63) above, and we shall
discuss a more dramatic instance of it in x5.4.2. Before that we must
consider what would be a highly systematic type of syncretism, but one for
which we have found no convincing evidence.
5.4.1.4 Candidates for syntactically determined syncretism

There are various types of situation which at first appear to be syntactic
domains for syncretism. On closer examination they seem to us to be better
analysed in other ways.
In the earlier discussion we talked of ‘noun’, ‘adjective’ and so on, which
suggests that syntactic categories can provide the domain for syncretism.
However, the hierarchy given in Figures 6 and 7 simplifies the position. We
need a separate hierarchy for syntactic categories, though by default items in a
given category inherit their morphological information from a node domi-
nated by a morphological category corresponding to this syntactic category
(nouns from MOR_NOUN, adjectives from MOR_ADJECTIVE, and so
on). The question then arises as to what happens when syntactic and mor-
phological categories fail to match. Which provides the domain for syncret-
ism? Slavonic provides helpful data here. Consider Russian items like
stolovaja ‘dining room’. Syntactically these behave like nouns. However,
they behave morphologically as adjectives (67).
(67) Russian nouns and adjectives
‘map’ ‘dining room’ ‘new’ (F.SG)

syntactically: noun noun adjective
morphologically: noun adjective adjective
NOM SG karta stolovaja novaja

ACC SG kartu stolovuju novuju
GEN SG karti stolovoj novoj
DAT SG karte stolovoj novoj
LOC SG karte stolovoj novoj
INS SG kartoj stolovoj novoj
The syncretisms found with morphological nouns and morphological

adjectives are different in Russian. Stolovaja ‘dining room’ and similar
items have the syncretism of morphological adjectives. This shows that it is
the morphological category which provides a domain for syncretism. To
demonstrate that a syntactic category is a possible domain, we would need
to find the converse situation to that found in Russian, that is a situation in
which morphological category and syntactic category differed for certain

items, and where these items took the syncretisms appropriate for the
syntactic category but not for the morphological category.13
The other area where we might consider syntactic influence on syncret-
ism is with features which may have a role both in morphology and in
syntax. (They are to be distinguished from purely morphological features,
such as features indicating inflectional class.) Typical examples are number
and gender, which are relevant to syntax, notably in agreement, and also to
morphology. To see the role of such features we shall consider the inter-
esting issue of animacy, discussed above in Chapter 3: x3.1.2 and Chapter
4: x4.3.4 (also see Perlmutter and Orešnik 1973, Huntley 1980, Corbett and
Fraser 1993).
Table 3 includes two examples for each of the four main inflectional
classes (see (58) above); each of the eight examples is representative of a
group of nouns. If we look at any of the examples individually, we find an
instance of syncretism. Take just the first example, student. There is syncret-
ism of accusative and genitive singular (conditioned by animacy). More
generally, the singular accusative depends on the gender and animacy of
the item in question in Table 3, and the plural accusative on the animacy
of the item in question. Evaluation of this information allows us to state that
the singular accusative of masculine animates is the same as the singular
genitive, and the plural accusative of animates is the same as the genitive.
The different inflectional paradigms in Table 3 share the same pattern of
identity, even though the particular inflections differ. It would clearly be
inadequate to state the identity of forms separately for each inflectional
class; that would imply that the cases involved could equally well differ
from inflectional class to inflectional class. This regularity is captured in
the Network Morphology account by a statement high up the inheritance
13
Andrew Carstairs-McCarthy (personal communication) points out a possible example of
syntactically conditioned syncretism in Latin. Some third declension adjectives have
syncretism of the dative and ablative singular (with the ending -i) when used adjectivally,
but when used substantivally, these forms are distinct (dative -i versus ablative -e), e.g. a
sapienti viro ‘by a wise man’, a sapiente ‘by a philosopher’ (Kennedy 1955: 40). However,
this alternation also obtains within nouns of the third declension. The third declension is
divided into two subtypes, the i-stems and the consonant stems: i-stems have the syncretic
dative/ablative singular -i, while consonant stems have a distinct dative singular -i versus
ablative singular -e. Thus, the choice of forms is not necessarily correlated with syntax, and
the alternation between syncretic and non-syncretic patterns in the third declension adjec-
tives might better be described as an alternation between i-stem and consonant-stem
declension patterns.
Table 3 The morphological effect of animacy in Russian
‘student’ ‘law’ ‘teacher(F)’ ‘map’ ‘mouse’ ‘bone’ ‘monster’ ‘wine’

animate inanimate animate inanimate animate inanimate animate inanimate
NOM SG student zakon učitel0 nica karta miš kost0 čudovišče vino
ACC SG studenta zakon učitel0 nicu kartu miš kost0 čudovišče vino
GEN SG studenta zakona učitel0 nicy karty miši kosti čudovišča vina
NOM PL studenty zakony učitel0 nicy karty miši kosti čudovišča vina
ACC PL studentov zakony učitel0 nic karty mišej kosti čudovišč vina
GEN PL studentov zakonov učitel0 nic kart mišej kostej čudovišč vin
tree. Since animacy also affects the agreement of adjectives, and the form
of pronouns, the logical place for it would be at the top of the hierarchy in
Figure 7, namely at the MOR_NOMINAL node.
There are two questions here, the first is the type of feature we are
dealing with, and the second, our main concern, is the domain of syncret-
ism. Animacy appears to be a semantic feature, in that the nouns involved
denote entities which live and move (thus insects are animate but plants are
not). The match with this semantic definition is close in Russian (less so in
some other Slavonic languages). There are some interesting borderline
cases, for instance pokojnik ‘the deceased’ is grammatically animate.
Such instances are animate for agreement purposes and for the morpho-
logical matter of syncretism. But we should also capture the fact that the
personal pronouns regularly have accusative/genitive syncretism, whether
or not they are referring to an animate entity. This may be captured by
stipulating that they are grammatically animate. Our conclusion is that the
animacy feature is a morphosyntactic one (albeit one with strong semantic
motivation in Russian, and with lesser semantic motivation in some other
Slavonic languages). In any case, while animacy is a major determining
feature for accusative/genitive syncretism, the syncretism depends on the
interaction of animacy with number and gender (see Fraser and Corbett
1995 for more details). Moreover, it is overridden by morphological con-
siderations (inflectional class II nouns have accusative singular in -u irre-
spective of animacy). In our previous examples the domain given uniquely
specified the syncretism in question. The animacy feature does not do this.
What then is the domain of the syncretism we have just examined? We shall
see in the next section.
5.4.2 Orthogonal specification of syncretism

The clearest domains we have encountered were well described in terms
of an inheritance hierarchy. However, the interesting data on animacy
prove significantly different. The point is that generalizations about
syncretism determined by animacy can be stated high up at the level of
MOR_NOMINAL in the hierarchy in Figure 7. They apply to different
paradigms, giving identities of pattern rather than any phonological
identity. It is not the case that accusative/genitive syncretism is found
with all items in the nominal domain, rather only those that are animate
(whether as a result of their semantics, or being specified as animate, or
being marked as animate as a consequence of agreement). Thus the
specification is orthogonal to the specification for inflectional class.
This captures the common-sense view that we do not wish to claim there
are eight inflectional classes for nouns in Russian, rather that there are
four main classes, with animacy affecting each of them.
The natural way of thinking of domains is in hierarchical terms, and this
was our approach. However, syncretism may require specification which is
orthogonal to the morphological hierarchy. How then does such orthogo-
nal specification of syncretism differ from lexical specification? Lexical
specification means that the individual items must each be specified, in
other words that they are exceptional in this regard. Orthogonal specifica-
tion can be regular (animacy syncretism in Russian shows a very high
degree of regularity). The distinguishing point is that it depends on a
feature which is otherwise irrelevant to inflection, because it affects only
the accusative case, and which cross-cuts the features which determine
inflection. We can specify a domain within which it operates, but not a
domain where it uniquely determines a syncretism.
Animacy in Russian shows that regularity of syncretism is quite a subtle
notion. On the one hand, syncretism based on animacy is systematic,
because its evaluation potentially determines the accusative plural of
every lexical item which is nominal, and therefore it is stated high up in
the hierarchy. Furthermore, it is realized in several different ways which
have no phonology in common. On the other hand, there are many nouns,
for which in the singular it has no effect (as already mentioned, inflectional
class II nouns have accusative singular in -u irrespective of animacy).
5.5 Conclusion
We have examined three case studies of unrelated languages, illustrating
different aspects of syncretism. The Dhaasanac study required the general-
ization of the A/B pattern across different tenses. Furthermore, this was a
pattern which, while systematic, could not be accounted for using mor-
phosyntactic features alone. By separating the realization of the forms
from the morphosyntactic feature specification, Network Morphology
makes it possible to account for such systematicity. As it places constraints
on feature specification, Network Morphology is still able to relate this
language-specific morphological systematicity to the general patterns: the
choice of A/B patterns is determined by tense, aspect, mood and negation.
In our case study of Dalabon we showed that what was originally a
pragmatically determined avoidance of particular forms – namely those
involving combinations of first person and second person in the transitive
paradigm – had become morphologized. Underspecification cannot deal

with this syncretism on its own. The Network Morphology construct of
generalized referral involves a degree of underspecification and is therefore
able to refer one set of cells in the paradigm to another set of cells. This
means that it is not just stating the facts directly, but predicting whole sets
of forms. Finally, our case study of Russian showed that degrees of
morphological regularity can be defined by locating syncretisms at different
places in an inheritance hierarchy. Each of the case studies in this chapter
has an implementation associated with it. The fragment for Dhaasanac in
Appendix 4 generates the positive perfect and imperfect paradigms, as well
as the positive imperative singular, for three example verbs, two of the
coronal type and one of the non-coronal type. The Dalabon fragment in
Appendix 5 covers 117 forms (102 transitive and 15 intransitive) derived for
the verb ‘to see’, and the Russian fragment in Appendix 6 covers a signi-
ficant proportion of the nominal lexicon. Our approach therefore has
justification on two grounds. We have put forward linguistic arguments
which favour these approaches, and we have implemented our analyses to
demonstrate that linguistic arguments are indeed tenable.
6 Conclusion
6.1 Taking stock

It is worth reflecting on how we have reached this point. We began from a
logical typology of syncretism. We asked what might theoretically be
found, working up from the simplest possibilities (Chapter 2). This
approach is feasible because the phenomenon is finite, and particularly
worthwhile because syncretism is complex, and the reality does not corres-
pond to many linguists’ hunches about it. Given this basis we could move
on to the cross-linguistic typology of Chapter 3, where we saw that extend-
ing the range of languages beyond those most usually treated for syncret-
ism changes the picture dramatically. Among other things it shows that
some of the patterns of syncretism which are familiar from Indo-European
are actually somewhat exotic. We relied largely on two samples. First of
all, our own sample for the Surrey Syncretisms Database, with thirty
languages chosen for genetic and geographic diversity but with the entry
condition that they must show instances of syncretism. These were inves-
tigated in great detail in preparation for this book. The second sample is
that of the World Atlas of Language Structures. This sample of 200 lan-
guages, also chosen to avoid genetic and areal bias, provides a useful
balance in that it was externally selected; of course, some of the languages
provided no helpful data since they lack inflectional morphology. While
the breadth of these samples is invaluable in ensuring coverage, so that
unusual phenomena do not fall through the net, we were not tied to them.
This is because the slight differences found between closely related lan-
guages (which would not figure in the same sample) can be of great
significance too. There are instances in the literature of a good deal of
theoretical weight being put on the set-up of a particular language’s
syncretisms, which is shown to be misplaced by the inability of the analysis
to handle the slightly different system of a closely related language. Having
established a solid empirical base we moved on to considering the formal
219
tools available for describing the patterns observed (Chapter 4). We estab-
lished the essential types of pattern, and the basic theoretical devices
employed to model them. We noted that some claims for simplicity of
analysis were undermined by the need to reintroduce complexity by other
means. Some false trails have been introduced into the literature by ana-
lyses of interesting but rather limited selections of inflectional systems. We
suggest that such analyses need to be justified within an account of the
whole system. And given the complexity of some of these systems, the
convincing way to validate an analysis is through implementation. In
Chapter 5, therefore, we discussed full analyses and implementations of
inflectional systems which are particularly significant for the syncretisms
they exhibit.
6.2 Results
When viewed in terms of morphosyntactic features, certain patterns are
repeated across unrelated languages with sufficient frequency to suggest
that there is a basis in function or meaning. Syncretism of case most
typically affects the core cases used to express subject (intransitive and
transitive) and object. One pattern involves collapse of core case distinc-
tions, often correlated with lower animacy or specificity of the argument.
The other major pattern involves the use of some non-core case form in the
object or transitive subject function; which non-core case is used is subject
to cross-linguistic variation. Syncretism of person, which we examined
through subject-person marking on verbs, usually affects non-singular
values, most often combining first and second or second and third person.
In the case of transitive verbs which agree with both core arguments,
syncretism usually involves the collapse of subject-person distinctions in
the presence of certain object values (most typically first or second person).
Gender syncretism does not lend itself to very clear cross-linguistic general-
izations; however, there is a decided tendency for gender values which are
distinct in the singular to be syncretic in the non-singular, often with a
plausible semantic basis. Syncretism of number usually involves collapse of
non-singular values. In Chapter 3: x3.8, analysis of interactions showed
that cross-linguistically some features are more likely to be syncretic than
others.
However, if the prevalence of certain patterns is evidence for a morpho-
syntactic or semantic basis to syncretism, there is also no lack of isolated,
language-specific patterns. Most strikingly, there is evidence of the
Conclusion 221
propagation of patterns that were originally accidental, the result of

regular sound change (see especially Chapter 3: x3.7). This suggests that
syncretism need not have a basis in the meaning or function of the morpho-
syntactic values that have been collapsed. As we have seen in Chapter 4, the
existence of what appears to be morphologically stipulated syncretism has
posed a problem for formal models which attempt to link syncretism
directly to feature structure. Nor do more abstract constraints on morpho-
logical structure, for example the rejection of directional rules, seem to
provide an accurate picture of the sorts of syncretic patterns which do or
do not occur.
6.3 Consequences
In a sense, syncretism can be seen as a minor morphological phenomenon,
a brief detour on the path between function and form. As such, an account
of syncretism need be nothing more than a makeshift patch on an idealized
morphological model. But when we take into account the full range of
syncretic patterns as seen in the languages of the world, we find that their
implications stretch down to the foundations of our conception of morph-
ology. Without syncretism, the structure of inflectional morphology need be
nothing more than a direct link between morphosyntactic values and forms.
Syncretism, in all its variety, argues for the existence of morphological
structure which is, at least in some degree, independent of meaning.
This independence is reflected in the assumption of three morphological
devices: underspecification, indexing and referrals.
Failure to incorporate these devices in a model of morphology will under-
mine any attempt to develop a sound and coherent theory of features.
Without them, we would be required to stretch morphosyntactic features
beyond what is reasonable and consistent in order to account for examples
of syncretism which involve some degree of morphological systematicity.
A final benefit of taking the full range of syncretisms into account is that it
gives us a clearer picture of the interaction between morphology and
syntax. Syncretism is morphology failing syntax, and it does this in inter-
estingly different ways. The three formal devices (underspecification,
indexing and referral) represent successively more extreme deviations
from a one-to-one correspondence between morphosyntax and morpho-
logy. Underspecification is uninformative, in that it fails to honour all of
the relevant functional distinctions. Indexing compounds uninformative-
ness with an additional, autonomous structure which cuts across
morphosyntactic natural classes. Referrals display both these violations,

and are misleading as well, in as much as a form which has the function x in
one context has the syncretic function x/y in some other context.
These interactions between the core components of morphology and
syntax provide a unique window into the internal workings of language.
Appendix 1: Case syncretism in
the World Atlas of Language
Structures sample
Indo-European languages are marked with ‘*’ in Appendix 1.
1.1 Languages with syncretism between core cases only

1. Basque Absolutive/ergative in a number of plural
paradigms (Hualde and Ortiz de Urbina 2003).
2. Boumaa Fijian Nominative/accusative in first exclusive (dual,
paucal, plural) pronouns (Dixon 1988).
3. English* Subject/object in singular neuter pronouns.
4. Harar Oromo Nominative/absolutive in plural pronouns;
absolutive here is an object case, also used for
predicate nominals (Owens 1985).
5. Mangarayi Nominative/accusative in the collective
demonstrative (Merlan 1982).
6. Murle Nominative/accusative for some nouns (Arensen
1971).
7. Paumari Absolutive/ergative in pronouns (Chapman and
Derbyshire 1990).
8. Pitjantjatjara Absolutive/ergative in pronouns, nominative/
accusative in nouns (Bowe 1990).
9. Suena Nominative/accusative in personal
pronouns. These cases are distinct in
interrogative and demonstrative pronouns
(Wilson 1974).
10. Wambaya Absolutive/ergative in pronouns, nominative/
accusative in other nominals (Nordlinger 1998).
11. Warao Nominative/accusative in nouns, 1SG, 1PL and
2SG pronouns do distinguish them (Romero-
Figueroa 1997). (Nouns distinguish other cases
though: dative, locative, allative, ablative.)
223
224 Appendix 1
12. West Greenlandic Essentially the same system as in Yup’ik described

below (Fortescue 1984).
13. Yaqui Nominative/accusative-possessive in the plural of
all nominals except for the third person pronoun
(Dedrick and Casad 1999).
14. Yidiny Absolutive/ergative in pronouns, nominative/
accusative in other nominals and inanimate
deictics (Dixon 1977).
15. Yukaghir Nominative/predicative in proper names,
possessed nouns and third person pronouns. The
predicative marks intransitive subject or transitive
object under focus; otherwise, distinct nominative
and accusative are marked (Maslova 2003).
16. Yup’ik Absolutive/relative in non-singular nouns, and in
first and second person pronouns. The relative
case combines the functions of ergative and
genitive (Jacobson 1995).
17. Yurok Nominative/accusative in plural and third person
pronouns. Only pronouns mark case (Robins 1958).
1.2 Languages with syncretism involving core cases and peripheral

cases
18. Araona Ergative/genitive-possessive for all nominals
except 1SG and 2SG pronouns (Pitman 1980).
19. Armenian* There is no distinct accusative case: inanimate
nouns have nominative/accusative, animate
nouns and pronouns have dative/accusative.
Some declension classes have dative/genitive in
the singular, while the pronoun ‘who’ has
accusative/dative/genitive (Minassian 1980).
20. Burushaski Ergative/genitive in all nominals except for singular
feminine nouns and modifiers (Berger 1998).
21. Chukchi Locative/ergative-instrumental in second
declension nouns, and in the interrogative
pronoun ‘who’ (Skorik 1961–77).
22. Comanche Nominative/accusative in some nouns (nominative
is -Ø, and one of the allomorphs of the accusative
is -Ø). Genitive is marked by a morphophonological
Case syncretism in the WALS sample 225
alternation using the accusative as a base (singular,

and one of the dual paradigms) or the nominative
as a base (plural, and one of the dual paradigms).
This alternation affects only final -v and -r. Thus in
stems ending in other segments, nominative/
genitive or accusative/genitive (Charney 1993).
23. Finnish Accusative/genitive in singular nouns, nominative/
accusative in plural nouns (Fromm 1982).
24. French* Accusative/dative (direct/indirect object) for non-
third person pronouns. (No case distinctions for
first and second plural pronouns, unless
possessive form is considered.)
25. Georgian Absolutive/instrumental/genitive and dative/
adverbial in attributive adjectives. Genitive/
ergative/dative in the so-called Old Georgian
plural form (Hewitt 1995).
26. German* In the definite article and strong adjectives (which
display the maximal range of distinctions),
nominative/accusative in the feminine and neuter
singular, and in the plural. Feminine singular also
has dative/genitive. Nouns mark only the dative in
the plural, and in the masculine and neuter
singular, the genitive and (marginally) the dative.
N-stems (nouns and weak adjectives) distinguish
nominative (masculine) or nominative/accusative
(feminine and neuter) from a peripheral case-
cum-plural form (masculine and neuter n-stem
nouns may distinguish genitive singular).
27. Greek (Modern)* Nominative/accusative in the singular of feminine
and neuter nouns, and in the plural of i-stem
nouns. Accusative/genitive in the singular of
masculine i-stem nouns and in the plural of first
and second person pronouns (Holton, Mackridge
and Philippaki-Warburton 1997).
28. Hindi* Direct/oblique in the singular of feminine nouns
and the plural of first and second person pronouns.
Oblique singular/direct plural in masculine nouns.
Compounds of the previous patterns are found in
other paradigms (McGregor 1977).
226 Appendix 1
29. Ingush Nominative/genitive in the plural of some nouns,

ergative/genitive in the singular of others. All
peripheral cases are syncretic in attributive
adjectives (Nichols 1994).
30. Irish* Both cases (nominative-accusative and genitive)
syncretic in singular of one noun class. The
irregular noun bean ‘woman’ has two forms:
nominative-accusative and one serving the other
functions (Bammesberger 1982).
31. Kayardild Nominative/locative in first declension nouns
(Evans 1995).
32. Krongo Nominative-accusative/dative in pronouns (Reh
1985).
33. Lak1 Absolutive/ergative in pronouns, ergative/
genitive in noun. (Žirkov 1955).
34. Latvian* Nominative/accusative in the plural of feminine
nouns and adjectives. Nominative/genitive in the
singular of some declension classes. Various types
of ‘polarity effects’ between singular and plural
cases (Mathiassen 1997).
35. Lezgian Ergative/inessive in the singular of nouns with
stems in -a or -e, and the 1SG pronoun
(Haspelmath 1993).
36. Martuthunira Accusative/genitive in all but vocalic stem nominals,
and in 1SG and 2SG pronouns (Dench 1995).
37. Nenets Accusative/genitive in the singular of second
person pronoun and nouns with a first person
possessor. Nominative/accusative in the dual of
nouns with second or third person possessor,
and in the plural of any possessed noun.
Combination of both patterns in the dual of
unpossessed nouns, and nouns with a first person
possessor (Salminen 1997).
38. Ngiyambaa Absolutive/ergative in first and second person
pronouns, absolutive/accusative in other
nominals. Ergative/dative in some noun classes
(Donaldson 1980).
1
Note that this value has changed from that found in Baerman and Brown (2005a).
Case syncretism in the WALS sample 227
39. Russian* Wide variety of patterns; see Chapter 3: x3.1 for a

sample.
40. Spanish* Pronouns maximally distinguish three forms
(subject, object and prepositional). The last two
are distinct only for 1SG and 2SG (Butt and
Benjamin 2000).
1.3 Languages where inflectional case marking is absent or minimal

(A language is classed as having minimal case marking if the nominal
paradigm contains no more than two forms. In such a language the absence
of any case distinction is equivalent to the complete absence of case
inflection.)
Abipon, Abkhaz, Acoma, Ainu, Alamblak, Amele, Apurina, Arabic
(Egyptian), Asmat, Awa Pit, Aymara, Bagirmi, Bambara, Barasano, Beja,
Bribri, Bukiyip (Mountain Arapesh), Burmese, Cambodian, Canela-Kraho,
Carib, Cayuvava, Chalcatongo Mixtec, Chamorro, Coast Tsimshian, Cree,
Daga, Dani, Dehu, Diola-Fogny, Ekari, Ewe, Fur, Gooniyandi, Grebo,
Guarani, Haida, Hanis Coos, Hebrew, Hixkaryana, Hmong Njua, Igbo,
Indonesian, Iraqw, Jakaltek (Popti’), Japanese, Kanuri, Kapau, Karo
Batak, Karok, Kawesqar, Kayah Li, Kera, Kewa, Khasi, Khmu, Kilivila,
Kiowa, Kiribatese, Kobon, Kongo, Korean, Koromfe, Koyraboro Senni
Songay, Kunama, Kutenai, Lakhota, Lango, Lavukaleve, Lealao
Chinantec, Luvale, Maba, Malagasy, Mandarin Chinese, Maori,
Mapuche, Maranungku, Marind, Mataco, Maung, Maybrat, Mundari,
Nama, Navajo, Ndyuka, Ngiti, Nkore-Kiga, Otomi, Paamese,
Pasamaquoddy, Persian, Piraha, Rama, Rapanui, Sango, Sanuma,
Selknam, Sentani, Shipibo-Konibo, Slave, Squamish, Supyire, Swahili,
Taba, Tagalog, Tamazight, Tetelcingo Nahuatl, Thai, Tiwi, Tlingit,
Tukang Besi, Tunica, Ungarinjin, Urubu-Kaapor, Usan, Vietnamese,
Wari, Wichita, !Xu, Yagua, Yimas, Yuchi, Zulu (123 languages).
1.4 Languages where inflectional case marking is not syncretic

Bawm, Brahui, Cahuilla, Copainala Zoque, Dongolese Nubian, Epena Pedee,
Evenki, Garo, Hausa, Hungarian, Hunzib, Ika, Imonda, Kannada, Ket,
Khalkha, Koasati, Ladakhi, Lepcha, Maricopa, Meithei, Nez Perce, Nivkh,
Nunggubuyu, Paiwan, Quechua (Imbabura), Sierra Miwok, Southeastern
Pomo, Trumai, Turkish, Una, Wardaman, Witoto, Yoruba (34 languages).
Appendix 2: Person syncretism
in the World Atlas of Language
Structures sample
2.1 Languages with syncretism of subject person in verbs which mark

a single argument
1. Abipon 1/3 in fourth conjugation verbs (Najlis 1966).
2. Amele 2/3 in dual and plural. 1SG/2SG in remote past and
negative past. In some other paradigms these
patterns are combined with syncretism of number
(Roberts 1987).
3. Arabic (Egyptian) 2SG M/3SG F, and 2DU/3DU F in imperfect
paradigms (Mitchell 1962).
4. Awa Pit 2/3 in statements and 1/3 in questions (Curnow
1997).
5. Aymara 1/2 in singular and plural in the subjunctive, and
in the so-called ‘third form’ paradigms (Deza
Galindo 1992).
6. Bagirmi In consonant-intitial stems, 1SG/2SG/3 in the
indefinite aspect, and 2SG/3PL in the definite
aspect. 1PL either is included in the previous
syncretic values, or is syncretic with 2PL. 1PL/3PL
optional in vowel-initial stems (Stevenson 1969).
7. Beja 1SG/3SG M in past tense (Hudson 1974).
8. Burushaski Many different patterns depending on tense/
mood: 1/2 and 2/3 in both singular and plural, and
some paradigms have 1/2/3 in plural (Berger
1998).
9. Canela-Kraho 1/3 in one class of verbs (Popjes and Popjes 1986).
10. Carib 2PL/3PL in present tense copula (Hoff 1968).
11. Cayuvava 1INCL PL/1EXCL PL in copula (Key 1967).
12. Chukchi 2SG/3SG (Skorik 1961–77).
13. Daga 1SG/3PL in past (Murane 1974).
228
Person syncretism in the WALS sample 229
14. Dani 2PL/3 in hypothetical mood (Bromley 1981).

15. Diola-Fogny 2SG/PL in the so-called ‘full form’, 3SG/1PL the
‘stripped form’ (Sapir 1965).
16. Dongolese Nubian 2SG/3SG and 1PL/2PL. In the negative indicative,
all values except 3PL are syncretic (Armbruster
1960).
17. Ekari 1SG/3SG F and 2SG/1PL (Drabbe 1952).
18. English 1SG distinct from 3SG and ‘elsewhere’ form in the
verb ‘be’, otherwise 1SG not distinct from
‘elsewhere’ form.
19. Ewe 2SG/3SG optional for verbs in sentence-medial
position (Westermann 1930).
20. French 2SG/3SG and 1SG/2PL in future, 2SG/3SG/3PL in
conditional, 1SG/2SG/3(SG) in present.
21. German 1PL/3PL throughout, 1SG/3SG in preterite, and
3SG/2PL of verbs lacking vowel alternation in
2SG/3SG.
22. Harar Oromo 1SG/3SG M and 2SG/3SG F (Owens 1985).
23. Hebrew 2SG M/3SG F in future (Berman 1997).
24. Hindi 2SG/3SG and 1PL/3PL (McGregor 1977).
25. Hixkaryana 1PL EXCL/3 (Derbyshire 1979).
26. Ika 1PL/2PL except in future and with experiencer
verbs, 1SG/3SG has the same restriction, plus it
does not occur in the distal past (Frank 1990).
27. Iraqw 2SG/3SG F in main verbs, and 1/2 in the selector and
the locative existential verb (Mous 1993).
28. Irish 3SG/2PL in preterite subjunctive, conditional, and
in habitual past, 2/3 in present indicative, and 1SG/
2/3 elsewhere (Bammesberger 1982).
29. Kannada 1SG/2SG in literary past tense, and optionally in
negative (Sridhar 1989).
30. Kapau 2/3 in dual and plural of a number of the
numerous tense/aspect/mood paradigms, 1SG/
2PL/3PL in others (Oates and Oates 1968).
31. Karok 2SG/3SG and 2PL/3PL in negative verbs (Bright 1957).
32. Ket 1SG/3SG M where person-marking prefix precedes a
vowel-initial base (Verner 1999).
33. Kewa 1SG/2SG in the perfect of active verb bases marking
egocentric benefaction (Franklin 1971).
230 Appendix 2
34. Kiowa 1DU-PL/3, where the number value of ‘3’ depends

on the noun class of the referent: plural if class I,
singular if class II, singular and plural if class III
(Watkins 1984).
35. Kobon 2DU/3DU; 2SG/3SG for medial same-subject verbs
(Davies 1981).
36. Kongo 2SG/3SG for indicative verbs (Carter and
Makoondekwa 1979).
37. Kunama 2PL/3PL in suffix conjugation, 1SG/2SG in class IIa
irregulars (Bender 1996).
38. Lak 1/2 in either singular or both singular and plural in
some tense/aspect paradigms, 2/3 in transitives,
except for present imperfective (Xajdakov 2001).
39. Lango 1DU/3PL in subjunctive, plus 1PL in habitual and
progressive, plus the 3SG form used in the perfect
with the 3SG pronoun and relativized subjects
(Noonan 1992).
40. Latvian 2SG/3 in the future, and in the present of some
stem classes. (Mathiassen 1997).
41. Lavukaleve 1INCL DU-PL/2PL (Terrill 2003).
42. Lealao Chinantec 1SG/3 in transitives with an animate object (Rupp
1989).
43. Luvale 2SG/3SG, where 3SG is of class II, and, under some
conditions, of class I (animates). 1PL/3PL, where
3PL is of class VII (Horton 1949).
44. Marind 3SG/2PL in various paradigms, depending on tense
and choice of prefixes (Drabbe 1955).
45. Murle 1SG/3SG/1INCL PL in the subjunctive; 2SG/3 in the
imperfect in one conjugation class (Arensen 1971).
46. Nama 2SG F/3SG F (Hagman 1977).
47. Nenets 2DU/3DU in transitives where the object is dual or
plural (Salminen 1997).
48. Ngiti 1INCL PL/3 indefinite, when an overt pronoun is
used (Kutsch Lojenga 1994).
49. Nivkh 2SG/3SG versus another form for all other values
(Gruzdeva 1997).
50. Nkore-Kiga 1SG/3PL, where 3PL is of class 12/13 (Taylor 1985).
51. Nunggubuyu 1/2 in dual and plural of a number of tense/aspect/
mood paradigms (Heath 1984).
Person syncretism in the WALS sample 231
52. Otomi1 2/3 in the antecopreterite paradigm (Hekking

1995).
53. Sentani 2SG/3DU in the adhortative (Cowan 1965).
54. Spanish 1SG/3SG in the present subjunctive and imperfect
(Butt and Benjamin 2000).
55. Suena 2/3 in the dual and plural of the remote tense
(Wilson 1974).
56. Swahili 2SG/3SG, where 3SG is of class 3/4 or 11/10 (Ashton
1947, Myachina 1981).
57. Tiwi 1PL/2PL; 1SG/3SG F in past (Osborne 1974).
58. Usan 2/3 in both singular and plural of medial switch
reference verbs (Reesink 1987).
59. Yimas 1/2 in paucal, though note that the corresponding
free pronoun also combines these values (Foley
1991).
60. Yukaghir 1SG/2SG in transitive verbs with object focus. All
values except 3PL syncretic in intransitives with
subject focus (Maslova 2003, Nikolaeva and
Xelimskij 1997).
61. Zulu 2SG/3SG, where 3SG is animate or of class III/IV
(Oxotina 1961).
2.2 Languages where subject person is marked inflectionally, but is not

syncretic
(Note that this does not include instances where subject-person
marking interacts with object-person marking, for which see Appendix 3.)
Abkhaz, Acoma, Ainu, Alamblak, Apurina, Armenian, Asmat, Barasano,
Basque, Bawm, Brahui, Bukiyip (Mountain Arapesh), Cahuilla,
Chalcatongo Mixtec, Chamorro, Coast Tsimshian, Copainala Zoque, Cree,
Evenki, Finnish, Fur, Georgian, Gooniyandi, Grebo, Greek (Modern),
Guarani, Hanis Coos, Hungarian, Hunzib, Ingush, Jakaltek (Popti’),
Kanuri, Karo Batak, Kilivila, Koasati, Krongo, Kutenai, Lakhota, Maba,
Malagasy, Mangarayi, Mapuche, Maranungku, Maricopa, Mataco, Maung,
Maybrat, Mundari, Navajo, Nez Perce, Oneida, Paamese, Paiwan,
Pasamaquoddy, Paumari, Persian, Quechua (Imbabura), Russian, Sango,
Sierra Miwok, Slave, Squamish, Taba, Tamazight, Tetelcingo Nahuatl,
1
Data for Otomi added subsequent to Baerman and Brown (in press b).
232 Appendix 2
Tlingit, Tukang Besi, Tunica, Turkish, Una, Ungarinjin, Urubu-Kaapor,

Wambaya, Wardaman, West Greenlandic, Wichita, Witoto, Yuchi,
Yup’ik, Yurok (80 languages).
2.3 Languages with no inflectional marking of subject person

Araona, Bambara, Boumaa Fijian, Bribri, Burmese, Cambodian,
Comanche, Dehu, Epena Pedee, Garo, Haida, Hausa, Hmong Njua,
Igbo, Imonda, Indonesian, Japanese, Kawesqar, Kayah Li, Kayardild,
Kera, Khalkha, Khasi, Khmu, Kiribatese, Korean, Koromfe,
Koyraboro Senni Songay, Ladakhi, Lepcha, Lezgian, Mandarin
Chinese, Maori, Martuthunira, Meithei, Ndyuka, Ngiyambaa, Piraha,
Pitjantjatjara, Rama, Rapanui, Sanuma, Selknam, Shipibo-Konibo,
Southeastern Pomo, Supyire, Tagalog, Thai, Trumai, Vietnamese,
Warao, Wari, !Xu, Yagua, Yaqui, Yidiny, Yoruba (57 languages).
Appendix 3: Syncretism in
two-place verbs in the World Atlas
of Language Structures corpus
Languages marked with ‘þ’ also have person syncretism in one-place verbs
(which may be recapitulated in two-place verbs, but is not recorded sepa-
rately here).
3.1 Languages where a distinct pattern of subject-person syncretism

is found in two-place verbs
1. Bininj Gun-Wok Subject person not distinguished where object is
first or second person, unless the subject is 1SG, or
the object is 1SG and the subject singular (Evans
2003).
2. Georgian In 1 > 2 constructions, where one argument is
plural, the form is the same as 3SG > 2PL. This
applies only to the present tense of verbs with 3SG
ending -s (Hewitt 1995).
3. Gooniyandi Subject person not distinguished when object is
first person (McGregor 1990).
4. Guarani Subject person not distinguished where object is
first person (Gregores and Suárez 1967).
5. Maba Subject person not distinguished where object is
first person. Syncretism of 1/3PL where object is
second person (Tucker and Bryan 1966).
6. Mangarayi Person of subject not distinguished where object is
first or second person (Merlan 1982).
7. Ungarinjin Subject person not distinguished where object is
first or second person (Rumsey 1982).
8. West Greenlandic In conditional, syncretism of 1SG/2SG subject
where object is 3rd person (Fortescue 1984).
9. Yimasþ Syncretism of 2SG/3SG subject where object is 3SG
(Foley 1991).
233
234 Appendix 3
10. Yurok Syncretism of 2/3SG subject where object is 3PL.

(Here a 1PL passive form is used; where object is
3SG, 1PL active intransitive form is used (Robins
1958).)
3.2 Languages with syncretism of object person in two-place verbs

11. Iraqwþ The selector distinguishes object person, with
syncretism of 1SG/2SG M/3PL, and of 3SG M/2SG F –
except that where both arguments are third
person, both subject and object are distinguished
(Mous 1993).
12. Nkore-Kigaþ Syncretism of 2PL/3PL object, regardless of subject
(Taylor 1985).
13. Nunggubuyuþ Syncretism of 1PL/2PL/1INCL DU object where
subject is 3rd person (Heath 1984).
14. Kanuri In conjugation class I, syncretism of 1SG/3rd object
where subject is 3SG, and syncretism of 1/3rd object
where subject is 3PL (Hutchison 1981).
3.3 Languages with both subject and object person syncretism

15. Karokþ Subject is not distinguished if object is 1PL or 2PL.
Syncretism of 2SG/3 object where subject is 1PL. In
the positive indicative and optative, syncretism of
3PL/1PL object where subject is 3PL. In the negative
indicative, syncretism of 2PL/3PL subject where
object is 3SG, and of 2PL/3 subject where object is
3PL (Bright 1957).
16. Marind For all paradigms, syncretism of 2PL/3PL object
where subject is third person. Other patterns
limited by tense/aspect/mood (i) person of plural
subject not distinguished where object is first
person, (ii) syncretism of 2SG/3SG subject where
object is 3SG, (iii) object person not distinguished
where subject is first person, (iv) singular subject
not distinguished where object is 2SG, and (v)
syncretism of 1SG/3SG subject, regardless of object
(Drabbe 1955).
Syncretism in two-place verbs in the WALS corpus 235
17. Sentaniþ Person of dual and plural subject not

distinguished where object is second person.
Syncretism of 3SG/2DU subject regardless of
object. Syncretism of 1SG/3SG object regardless of
subject (Cowan 1965).
3.4 Languages with other patterns of person syncretism in two-place

verbs
18. Asmat There is a single form for 3SG > 1SG/2SG and 3PL >
2PL (Voorhoeve 1965).
19. Caribþ Where both first and second person are involved,
the 1INCL form is used (Hoff 1968).
20. Maricopa The form for 1 > 2 is the same as that for 3 > 1
(Gordon 1986).
Nunggubuyu With the A series prefixes, the form for 1 > 2SG is
(see no. 13 above) the same as that for 3SG F > 3SG M, and the form
for 2PL > 1 is the same as that for 1INCL PL > 3,
where 3 belongs to the ‘ANAwu’ gender (Heath 1984).
3.5 Languages with two-place verbs, where this is not associated with
a distinct pattern of person syncretism
Abiponþ, Acoma, Ainu, Alamblak, Apurina, Basque, Bawm, Bukiyip,
Cahuilla, Cayuvavaþ, Chukchiþ, Coast Tsimshian, Copainala Zoque,
Cree, Daniþ, Ekariþ, Eweþ, Grebo, Hanis Coos, Hixkaryanaþ,
Ket, Kilivila, Kiowaþ, Kongoþ, Kunamaþ, Kutenai, Lakhota,
Lavukaleveþ, Limbu, Maung, Mundari, Nez Perceþ, Oneida, Paamese,
Paiwan, Sierra Miwok, Squamish, Tetelcingo Nahuatl, Tlingit, Tukang
Besi, Tunica, Wambaya, Wardaman, Wichita, Yuchi (45 languages).
Appendix 4: DATR fragment for
Dhasaanac case study
File: dhaasanac.dtr
Purpose: A and B form syncretism in Dhaasanac
Author: Dunstan Brown (March 2004)
University of Surrey, Guildford GU2 7XH
Documentation: Chapter 5: x5.2; data from Tosco (2001)
Here and in Appendices 5 and 6, examples which appear in Chapter 5 are

noted as such by the use of bold in the comments which precede them.
DATR is distinguished from comments by use of courier font.
This fragment generates the positive perfect and imperfect paradigms for
three example verbs, two coronals ( fúr ‘open’ and leét ‘fall down’), and one
non-coronal (rok ‘grind’), as well as the positive imperative singular, which is
the stem plus high tone (Tosco 2001: 114). The positive imperative singular is
the citation form and is therefore used for labelling the example lexical entries.
The purpose of this fragment is to demonstrate that we can generate
different A and B forms for verbs of different types.
The nodes FORMS, CORONALS, NON_CORONALS:

these form a hierarchy, with the nodes CORONALS and NON_CORONALS
inheriting from FORMS. FORMS therefore makes generalisations about the
forms for both types.
The node FORMS:

the first equation states that the perfect ends in -i (see Tosco 2001: 113).
This ending will be preceded by the appropriate consonant grade, which
will depend on whether the form is the A or B form, and whether the verb
inherits from CORONALS or NON_CORONALS.
FORMS:
<form prf> = = "<stem initial> " "<form c grade> " -i
The next equation at FORMS creates the forms for the imperfect. This
is done by evaluating the path <stem type> for each lexical item. In this
236
DATR fragment for Dhasaanac case study 237
fragment either a lexical item will have one vowel associated with its
stem, or two.
<form imprf> = = "<form all "<stem type>" >"
If the lexical item has one vowel associated with it, then it will combine
that one vowel with the values for <stem initial> The values for <stem
initial> are defined at the VERB node, but their effect is to create an
initial stem which has one vowel, if the lexical item has only one vowel,
and two vowels, if the lexical item has two vowels. Hence, for items which
have only one vowel, this vowel is repeated for <form all> (associated
with the imperfect). This gives us a partial implementation of the
Bimoraic Filter (Tosco 2001: 125–6), namely by Stem Adjustment.
Nasal extension and reduplication are not dealt with here. The evaluable
path at the end of the equation deals with vowel harmony.
<form all one_vowel> = =

"<stem initial> " "<stem vowel_1> "
"<form c grade> " "<v_harm "<stem vowel_1> " > "
The equation below concatenates <stem initial>, the appropriate con-

sonantal grade (<form c grade>) and evaluates the final vowel of the stem
to determine the vowel of the imperfect. In this case, stems with two vowels
do not need to be augmented for the imperfect.
<form all two_vowels> = =
"<stem initial> " "<form c grade> "
"<v_harm "<stem vowel_2> " > ".
The node CORONALS:

this inherits from the node FORMS.
CORONALS:
<> = = FORMS
In the two equations below, the stem final element in the lexical entry is
evaluated to determine the A and B forms.
<form c grade _A> = = <form _A "<stem final> ">
<form c grade _B> = = <form _B "<stem final> " >
The equations below represent the consonant gradation for the B forms
of verbs. Consonant gradation assigns the final-stem consonant for B
forms. This means that by default the grade associated with <form _A>
238 Appendix 4
will be the same as what is specified for a lexical item as its final element of
the stem.
<form _A> = = "<stem final>"
The spirantization-II rule (Tosco 2001: 21) has been approximated here as
a rule which entails that the A form of stems ending in -t will be _ð.
<form _A _t> = = _ð
The equations below correspond to Table 6 in Tosco (2001: 128). As an

example, the equation <form _B _r> = = _ºº can be understood as follows:
the B form of a stem ending in r has ºº in place of r. The other equations
are interpreted in the same way.
<form _B _r> = = _ºº
<form _B _t> = = _t
<form _B _r> = = _ºº
<form _B _n> = = _nn
<form _B _s> = = _t
<form _B _l> = = _ll.
The nodes NON_CORONALS:

this inherits from the node FORMS.
NON_CORONALS:
<> = = FORMS
<form type> = = non_coronal
<form all one_vowel> = =
"<stem initial>" "<form c grade>"
"<v_harm "<stem vowel_1> " > "
<form all two_vowels> = =
"<stem initial> " "<form c grade> "
"<v_harm "<stem vowel_2> " > "
Non-coronals drop their last stem consonant when it is followed by a

suffix. This means that the non-coronal is dropped in all verbal forms
with the exception of the imperative singular, and the a-form of the short
past (Tosco 2001: 132).
The B form of non-coronals affix yy before the paradigmatic vowel (Tosco
2001: 134). We have treated these two phenomena together, by associating the
consonantal grade for A forms as truncated, and the consonantal grade for B
forms as yy. If the short past were included in this fragment, it would require
the first of these statements to be overridden.
<form c grade _A> = =

<form c grade _B> = = _yy.
The node VERB (example (24) in Chapter 5: x5.2)

VERB:
<> = =
<syn> = = verb
Example (24) in Chapter 5: x5.2

The path <index> and its extensions associates the morphosyntax with the
appropriate index. Second person, third person singular feminine, and first
person plural are associated with index B. All other person and number
combinations are associated with index A by default.
<index> = = _A
<index 2nd> = = _B
<index 3rd sg f> = = _B
<index 1st_excl pl> = = _B

The forms of the imperfect positive and perfect positive are determined
by evaluating the index associated with the particular person and
number information. For example, the form of the second person
(of either number) associated with <mor pos imprf> – the morphosyntax
of the imperfect positive – will require the form for <form imprf _B>, once
the <index> path is evaluated. On the other hand, the form of the first
person singular, for example, will require the form for <form imprf _A>.
These indexed forms are then determined by the lexical item’s member-
ship of a particular form class (coronal or non-coronal).
<mor pos imprf> = = "<form imprf <index> > "
<mor pos prf> = = "<form prf <index> > "
By default there is only one stem vowel and so the the value for <stem
vowel_2> is ‘undefined’.
<stem vowel_2> = = undefined
The initial form of the stem <stem initial> depends on whether there is
a second vowel, <stem vowel_2>.
<stem initial> = = <stem "<stem vowel_2> " >
As the default is for <stem vowel_2> to be undefined, the stem will consist
of a consonant and a vowel.
240 Appendix 4
<stem undefined> = = "<stem cons_1> " "<stem vowel_1> "
Otherwise, if <stem vowel_2> is specified lexically, the stem will consist of a

consonant and two vowels, together with a value for tone <stem tone>.
<stem> = =
"<stem cons_1> " "<stem vowel_1> "
"<stem tone> " "<stem vowel_2> "
However, as positive verbal forms are unaccented (Tosco 2001: 41), the
default for stem tone is set to lack of tone. This is done purely because we
are dealing with the positive paradigms. But the value could be different, if
this analysis were extended.
<stem tone> = =
The form of the positive imperative singular is the the initial stem plus high
tone, and the final element of the stem.
<mor pos impv sg> = = "<stem initial high_tone> "
"<stem final> "
The position of the high tone is determined by evaluating whether the stem
has two vowels or not.
<stem initial high_tone> = =
<stem high_tone "<stem vowel_2> " >
If the vowel for <stem vowel_2> is ‘undefined’, then the high tone will be
associated with the vowel of the mononsyllabic stem.
<stem high_tone undefined> = =
"<stem cons_1> " "<stem vowel_1> "
"<stem tone high_tone> "
High tone is represented by .

<stem tone high_tone> = =
The equations below implement the vowel harmony rule for the imper-
fect. If the final vowel of the stem is o, then the affix of the imperfect is o.
If the final vowel of the stem is e, then the affix is e. Otherwise, the
imperfect affix is a.
<v_harm _e> = = -e
<v_harm _o> = = -o
<v_harm> = = -a.
Example verb lexical entries

The verb rók belongs to the class of non-coronals. The fragment generates
forms for the positive perfect and imperfect paradigms and positive
imperative singular. These have been checked against Tosco (2001: 451).
However, the B form of the imperfect generated is as given in Tosco (2001:
134). Reduplicated variants are not generated by this fragment.
R˘k:
<> = = VERB
<form> = = NON_CORONALS
<gloss> = = ‘grind’
<stem type> = = one_vowel
<stem cons_1> = = r
<stem vowel_1> = = _o
<stem final> = = _k.
The verb Leét belongs to the class of coronals. The fragment generates forms
for the positive perfect and imperfect paradigms and positive imperative
singular. These have been checked against Tosco (2001: 432).
L¯et:
<> = = VERB
<form> = = CORONALS
<gloss> = = ‘fall down’
<stem type> = = two_vowels
<stem cons_1> = = l
<stem vowel_1> = = _e
<stem vowel_2> = = _e
<stem final> = = _t.

The verb fúr belongs to the class of coronals. The fragment generates
forms for the positive perfect and imperfect paradigms and positive
imperative singular. These have been checked against Tosco (2001: 383).
F˙r:
<> = = VERB
<form> = = CORONALS
<gloss> = = ‘open’
<stem type> = = one_vowel
<stem cons_1> = = f
<stem vowel_1> = = _u
<stem final> = = _r.
Appendix 5: DATR fragment
for Dalabon case study
File: dalvbs.dtr
Purpose: referrals and avoidance in Dalabon
Author: Dunstan Brown (September. 1999)
Documentation: Chapter 5 (x5.3); data from Evans, Brown and
Corbett (2001)
Dalabon is an Australian language of the Gunwinyguan family, spoken in

central Arnhem Land by a dwindling population now reduced to about
twenty fluent speakers.
This fragment implements the analysis given in case study 2 in x5.3 and
the associated work in Evans, Brown and Corbett (2001a).
We show that syncretism of 2 > 1 (second person subject and first person
object) with the 3 > 1 paradigm can be accounted for by the use of referrals
of the 2 > 1 paradigm to the 3 > 1 paradigm.
Similarly, the 1 > 2 sg paradigm refers to the 3 > 2 sg paradigm. This is a
formalization of the idea that ‘person disguise’ is involved, in that the
combinations of second and first person are too direct and need to be
disguised. In Dalabon this strategy has been morphologized.
Our analysis also assumes a particular structure to the Dalabon verbal
paradigm. This means that the object attributes are ordered first in the
path for all persons except first person exclusive, for which they are
ordered last.
This analysis generates 117 forms (102 transitive and 15 intransitive
forms). No TAM marking is included, and so the forms generated
belong to the the subord1 series. The difference between these forms
and those found in Table 2 is that the latter have the marker for realis
attached, namely -h. These realis forms have also been checked and are
correct.
The associated theorem dumps for this fragment can be viewed at http://
www.surrey.ac.uk/LIS/SMG/dalabon/. DATR is distinguished from com-
ments by use of courier font.
242
DATR fragment for Dalabon case study 243
The node VERB

Example (52) in Chapter 5
In a complete analysis of the morphology of Dalabon the lexeme node
VERB would inherit from a higher node WORD information about words
in general. As this fragment covers only part of the verbal system, we have
stated that any other extensions of the empty path for VERB, if not given
at the VERB node, are undefined.
The second equation states that the morphology of verbs is found at the
MOR_VERB node. The third refers to a node SYNTAX, which defines a
verb phrase, and the fourth states that the syntactic category of this class of
items is ‘verb’.
VERB:
<> ¼ ¼ undefined
<mor> ¼ ¼ MOR_VERB
<syn> ¼ ¼ SYNTAX
<syn cat> ¼ ¼ verb.
Verbal morphology defaults to nothing and verbal inflection consists of

some cluster (cl) before a verbal root.
MOR_VERB:
<mor> ¼ ¼
<mor infl> ¼ ¼ <mor cl> "<root>"
Referrals
The 1 > 2 sg object paradigm refers to the 3 > 2 sg object paradigm.
<mor infl o 2nd sg a 1st_excl> ¼ ¼ <mor infl o 2nd sg a 3rd>

For extensions of the second person transitive subject paradigm refer to
the third person transitive subject paradigm. Because of our ordering of
attributes, the first person exclusive object attributes constitute the only
possible extension. This accounts for the 2 > 1 paradigm.
<mor infl a 2nd> ¼ ¼ <mor infl a 3rd>
The 3 dis > 2 sg cell refers to the 3 du > 2 sg cell.

<mor infl o 2nd sg a 3rd dis> ¼ ¼ <mor infl o 2nd sg a 3rd du>
244 Appendix 5
The Morphological Cluster

The morphological cluster will default to bound prefixal (<mor
prefix>) marking.
Clitic forms are followed by a plus sign.
With the exception of the first person exclusive clitics, the clitic forms
are specifically given as being followed by bound prefixal morphology
(<mor prefix>). The realizations of first exclusive dual and plural just
specify the clitic pronoun forms. Another equation specifies that third
person bound prefixes may be preceded by clitics (see formants and equa-
tions for the ‘front’ element of third person prefixes).
Of course, first person exclusive object clitics are also followed by the
bound prefixal morphology, but we state this later in terms of them
preceding it. The attribute o means ‘object’.
<mor cl> ¼ ¼ <mor prefix>

<mor cl o 1st_excl du> ¼ ¼ njerrþ
<mor cl o 1st_excl pl> ¼ ¼ njelþ
<mor cl o 1st_incl du> ¼ ¼ njehþ <mor prefix>
<mor cl o 1st_incl pl> ¼ ¼ ngorrþ <mor prefix>
<mor cl o 2nd du> ¼ ¼ norrþ <mor prefix>
<mor cl o 2nd pl> ¼ ¼ nolþ <mor prefix>
<mor cl o 3rd du> ¼ ¼ bunuþ <mor prefix>
<mor cl o 3rd pl> ¼ ¼ buluþ <mor prefix>
Intransitive prefixes
Subject prefixes are created by putting together front and final elements
(see formants). The front element may combine person and number
marking, and the final element may combine intransitive subject (s) or
transitive subject (a) marking with number. Note that the 1st_inclusive
dual combines a front formant which marks first inclusive dual, and a final
element which marks singular subject. This captures the similarity with an
augmented / unit augmented system.
<mor prefix s 1st_excl> ¼ ¼
<mor prefix front 1st_excl>
<mor prefix final s>
<mor prefix s 1st_incl du> ¼ ¼
<mor prefix front 1st_incl du>
<mor prefix final s sg>
<mor prefix s 1st_incl pl> ¼ ¼

<mor prefix front 1st_excl sg>
<mor prefix final s du>
<mor prefix s 1st_incl dis> ¼ ¼
<mor prefix front 2nd sg>
<mor prefix final s dis>
<mor prefix s 2nd> ¼ ¼
<mor prefix front 2nd>
<mor prefix s 3rd> ¼ ¼
<mor prefix front 3rd>
Transitive prefixes
The prefixes of the transitive paradigm are similar to the intransitive,
having a front and final element which mark person and number.
<mor prefix a> ¼ ¼ <mor prefix s>

<mor prefix a 1st_excl> ¼ ¼
<mor prefix final a>
<mor prefix a 1st_incl pl> ¼ ¼
<mor prefix final a du>
<mor prefix a 2nd> ¼ ¼
<mor prefix front 2nd>
<mor prefix a 2nd sg> ¼ ¼
<mor prefix front 2nd dis>
<mor prefix final a sg>
<mor prefix a 3rd> ¼ ¼
<mor prefix front 3rd>
Object marking
The special 2 sg object marking involves the 2 sg front element combined
with the expected marking of the transitive subject (a) in the final element.
246 Appendix 5
The third person object adds bv_ if it is higher animacy than the subject,
and otherwise has no extra marking, if it is lower animacy.
<mor prefix o 2nd sg a 3rd> ¼ ¼

<mor prefix front 2nd sg> <mor prefix final a>
<mor prefix o 3rd sg> ¼ ¼ <mor prefix>
<mor prefix lower> ¼ ¼ <mor prefix>
<mor prefix higher> ¼ ¼ bv_ <mor prefix>
Formants
Front of Prefix
It should be noted that we have tried to break down the prefixes beyond a
standard analysis of morphemes. We have analysed prefixes as containing
a front element, which marks person and also number.
Note the left-hand side paths: <mor prefix front 3rd sg>, <mor prefix
front 3rd pl>, <mor prefix front 3rd du>. These have a corresponding
reference on the right-hand side to <mor cl> as well as the front element
which marks third person for the appropriate number. This allows for the
combination of third person subjects with the first person exclusive object
clitics.
<mor prefix front 1st_excl> ¼ ¼ ng_
<mor prefix front 1st_incl du> ¼ ¼ y_
<mor prefix front 2nd sg> ¼ ¼ dj_
<mor prefix front 2nd dis> ¼ ¼ d_
<mor prefix front 2nd> ¼ ¼ n_
<mor prefix front 3rd sg> ¼ ¼ <mor cl> k_
<mor prefix front 3rd pl> ¼ ¼ <mor cl> b_
<mor prefix front 3rd du> ¼ ¼ <mor prefix front 3rd pl>
<mor prefix front 3rd dis> ¼ ¼ <mor prefix front 3rd sg>
<mor prefix front 1st_excl pl> ¼ ¼ <mor prefix front 1st_incl du>
<mor prefix front 1st_excl du> ¼ ¼ <mor prefix front 1st_excl pl>
Final Part of Prefix

The final part of the prefix is the same for singular transitive subjects (a)
and intransitive subjects (s). The disyllabic plural and dual formant has
a_ for the intransitive subject and _U/I_for the transitive. The form _U/I_ is
realized as /i/ after palatals and /u/ otherwise.
<mor prefix final a sg> ¼ ¼ <mor prefix final s sg>

<mor prefix final s> ¼ ¼ a_ <mor prefix final>
<mor prefix final a> ¼ ¼ _U/I_ <mor prefix final>
<mor prefix final s dis> ¼ ¼ e_
<mor prefix final a dis> ¼ ¼ <mor prefix final s dis>
<mor prefix final du> ¼ ¼ rra_
<mor prefix final pl> ¼ ¼ la_.

Nan ‘see’ is our example lexical entry.
Nan:
<> ¼ ¼ VERB
<gloss> ¼ ¼ see
<root> ¼ ¼ nan.
Recall that this analysis generates the appropriate 117 forms of the
paradigm. The forms can be seen at http://www.surrey.ac.uk/LIS/SMG/
dalabon/.
Appendix 6: DATR fragment
for Russian case study
File: rusnoms7.dtr (simplified version)

Purpose: the nominal system of Russian
Author: Dunstan Brown, Greville Corbett, Norman
Fraser and Andrew Hippisley
Documentation: Chapter 5 (x5.4); data from Brown (1998b)
This appendix contains a simplified version of the DATR fragment which
implements the analysis given in case study 3 in Chapter 5: x5.4. It is based
on rusnoms7.dtr which is discussed in detail in Brown (1998b), where the
full DATR fragment is given. The original fragment was checked on the
first 1,500 noun lexemes from Zasorina’s (1977) frequency dictionary, a
sample of adjectives, and first, second and third person pronouns.
Among other things, we have made the following simplifications:
*omitted a number of nodes which deal with
morphonological interdependencies
stress assignment
*omitted paths which determine stress patterns
*omitted concatenation marking
DATR is distinguished from comments by use of courier font.
Declensional class hierachy

The declensional class hierarchy represented in Figure 6 in Chapter 5: x5.4
is implemented in DATR below.
The node MOR_NOMINAL

This contains equations which give much of the plural morphology: pl acc,
pl nom, pl gen, pl dat, pl inst, pl loc.
It is also at this point in the hierarchy that the forms of the accusative,
singular and plural, are determined. The accusative singular evaluates both
gender and animacy, whereas the accusative plural needs only to evaluate
animacy. The node ACCUSATIVE has been omitted from this simplified
248
DATR fragment for Russian case study 249
appendix. It is used to state that animates use the genitive form for accusa-
tives, and that inanimates use the nominative form for accusatives.
The oblique plural forms require a theme vowel, <mor theme_vowel>,
which is a for nouns and i for adjectives.
As these statements are at the node MOR_NOMINAL they apply to
both nouns and adjectives, unless overridden.
MOR_NOMINAL:
<> ¼ ¼ MOR_WORD
<mor sg acc> ¼ ¼
ACCUSATIVE:< sg "<syn gender>" "<syn animacy>" >
<mor pl acc> ¼ ¼ ACCUSATIVE:< pl "<syn animacy>" >
<mor pl nom> ¼ ¼ "<stem pl nom>" i "<stress pl nom>"
<mor pl gen> ¼ ¼ "<mor pl loc>"
<mor pl dat> ¼ ¼ "<stem pl>"
"<mor theme_vowel>" "<stress pl>" m
<mor pl inst> ¼ ¼
"<stem pl>"
"<mor theme_vowel>" "<stress pl>"
m’i
<mor pl loc> ¼ ¼ "<stem pl>"
"<mor theme_vowel>" "<stress pl>" x.
The node MOR_NOUN (example (65) in Chapter 5: x5.4)

There is a noun default for singular dative, which is the referral to singular
locative.
There is a noun default for the singular locative. The other equations do
not appear in example (65). One requires the evaluation of the noun’s stem
type to determine its genitive plural, and the other states that the theme
vowel for nouns is -a-.
MOR_NOUN:
<> ¼ ¼ MOR_NOMINAL
<mor sg dat> ¼ ¼ "<mor sg loc>"
<mor sg loc> ¼ ¼ "<stem sg>" e "<stress sg>"
<mor pl gen> ¼ ¼ MGP:<"<mor stem hardness>" pl gen>
<mor theme_vowel> ¼ ¼ a.
The node N_O

This node generalizes over classes I and IV, which are exemplified by
zakon and vino in Chapter 5: x5.4. N_O states the realization of singular
250 Appendix 6
genitive, singular dative and the singular instrumental for both types
of noun.
N_O:
<> ¼ ¼ MOR_NOUN
<mor sg gen> ¼ ¼ "<stem sg gen>" a "<stress sg>"
<mor sg dat> ¼ ¼ "<stem sg>" u "<stress sg>"
<mor sg inst> ¼ ¼ "<stem sg>" om "<stress sg>".
The node N_I

N_I inherits from N_O. The formal gender assigned, if there is no assign-
ment by semantics, is masculine. In the full DATR representation the
nominative plural form is partly determined by the stress pattern. The
node NOM_PL is not given in this simplified appendix. The genitive plural
of nouns with hard stems is the stem plus ov.
N_I:
<> ¼ ¼ N_O
<mor formal gender> ¼ ¼ m
<mor pl nom> ¼ ¼ NOM_PL:<"<index>">
<mor hard pl gen> ¼ ¼ "<stem pl>" ov "<stress pl>".
The node N_IV

N_IV inherits from the node N_O. The formal gender of this class is neuter.
The singular nominative is the stem plus o.
The plural nominative is the stem plus a. The plural genitive involves
evaluation of the final element of the stem.
N_IV:
<> ¼ ¼ N_O
<mor formal gender> ¼ ¼ n
<mor sg nom> ¼ ¼ "<stem sg nom>" o "<stress sg>"
<mor pl nom> ¼ ¼ "<stem pl>" a "<stress pl nom>"
<mor pl gen> ¼ ¼ GEN_PL:< "<stem pl final>" >.
The node N_II

N_II inherits from MOR_NOUN. The formal gender of this class is femi-
nine. The singular nominative is the stem plus a. N_II overrides the assign-
ment of accusative singular which would otherwise be inherited from
MOR_NOMINAL. The accusative singular is stem plus u. The genitive
singular is stem plus i. The instrumental singular is stem plus Vj (u), where
V is a vowel o and (u) is the archaic alternative.
The plural genitive involves evaluation of the final element of the stem
and stress information. The node STEMSTRESS is not given in this
simplified appendix.
N_II:
<> ¼ ¼ MOR_NOUN
<mor formal gender> ¼ ¼ f
<mor sg nom> ¼ ¼ "<stem sg nom>" a "<stress sg>"
<mor sg acc> ¼ ¼ "<stem sg>" u "<stress sg acc>"
<mor sg gen> ¼ ¼ "<stem sg>" i "<stress sg>"
<mor sg inst> ¼ ¼
"<stem sg inst>"
"<mor vowel sg>" "<stress sg>" j ‘(‘ u ’)’
<mor pl gen> ¼ ¼
STEMSTRESS: <"<mor stem hardness>" "<stress
pl>">.
The node N_III (example (62) in Chapter 5: x5.4)

This node inherits from MOR_NOUN. All members of this class are
morphologically soft (not given in (62)). The singular genitive is inherited
orthogonally from N_II (not given in (62)). The singular locative refers to
the singular genitive.
N_III:
<> ¼ ¼ MOR_NOUN
<mor stem hardness> ¼ ¼ soft
<mor sg gen> ¼ ¼ N_II
<mor sg loc> ¼ ¼ "<mor sg gen>"
<mor sg inst> ¼ ¼ "<stem sg inst>" ju
<mor formal gender> ¼ ¼ f.
Adjective inflection
The node MOR_ADJ:
this inherits from MOR_NOMINAL, including the the generalizations
about animacy and the accusative, as well as most plural forms.
MOR_ADJ specifies most of the singular forms of adjectives. It also
specifies the theme vowel which is used in the oblique plural forms given
at MOR_NOMINAL.
MOR_ADJ:
<> ¼ ¼ MOR_NOMINAL
<mor sg gen> ¼ ¼
252 Appendix 6
"<stem>" "<mor vowel sg>" "<stress sg>" vo

<mor sg gen f> ¼ ¼ "<mor sg loc f>"
<mor sg dat> ¼ ¼
"<stem>" "<mor vowel sg>" "<stress sg>" mu
<mor sg dat f> ¼ ¼ MOR_NOUN
<mor sg inst> ¼ ¼ "<stem>" i "<stress sg>" m
<mor sg inst f> ¼ ¼ N_II
<mor sg loc> ¼ ¼ "<stem>"
"<mor vowel sg loc>" "<stress sg>" m
<mor sg loc f> ¼ ¼ "<stem>"
"<mor vowel sg>" "<stress sg>" j
<mor theme_vowel> ¼ ¼ i.
The node A_I:

this inherits from the node MOR_ADJ. It accounts for the majority of
attributive adjectives. The singular nominative masculine combines the
singular stem with ij or oj, depending on the stress pattern of the adjective.
(ADJ_VOWEL is not given in this appendix, but it determines the vowel
on the basis of stress.) The singular nominative feminine refers to N_II and
combines this with the augment -ja. The singular neuter inherits in part
from the node N_IV and combines the value with the augment -je. The
plural nominative combines the plural theme vowel with je. The singular
accusative feminine combines the value for singular accusative at N_II
with the augment ju.
A_I:
<> ¼ ¼ MOR_ADJ
<mor sg nom m> ¼ ¼ "<stem sg>"
ADJ_VOWEL:<"<stress sg>"> "<stress sg>" j
<mor sg nom f> ¼ ¼ N_II ja
<mor sg nom n> ¼ ¼ N_IV je
<mor sg acc f> ¼ ¼ N_II ju
<mor pl nom> ¼ ¼ "<stem pl>"
"<mor theme_vowel>" "<stress pl>"
je.
The nodes A_II and A_III:

these inherit from MOR_ADJ. They define possessive adjectival classes
which combine some noun morphology with the default adjectival mor-
phology. This is why they consist almost entirely of inheritance relations
with the noun classes. The class A_III is rare.
A_II:
<> ¼ ¼ MOR_ADJ
<mor sg nom f> ¼ ¼ N_II
<mor sg nom n> ¼ ¼ N_IV
<mor sg acc f> ¼ ¼ N_II.
A_III:
<> ¼ ¼MOR_ADJ
<mor sg nom f> ¼ ¼ N_II
<mor sg nom n> ¼ ¼ N_IV
<mor sg acc f> ¼ ¼ N_II
<mor sg gen n> ¼ ¼ "<mor sg gen m>"
<mor sg dat n> ¼ ¼ "<mor sg dat m>"
<mor sg gen m> ¼ ¼ N_I
<mor sg dat m> ¼ ¼ N_I.
6.3 A note on Russian phonology

The Russian examples in Chapter 5: x5.4 and elsewhere in the book have
been given in transliteration. In contrast, the DATR fragment which this
simplified appendix is based on generates forms in phonological transcrip-
tion. It is therefore important to outline the phonological correspondences
assumed for the DATR fragment.
This note is based on Corbett and Fraser (1993: 114).
The phoneme transcribed as /i/ is retracted to its back allphone [ç] after non-
back hard (unpalatalized) consonants. Thus the nominative plural /zakoni/ is
realized with [ç], but /kost 0 i/ is realised with [i], since [t 0 ] is soft. This difference
is reflected in the writing system, with the inflection of /zakoni/ being written
z, and the inflection for /kost 0 i/ written h. Underlyingly it is assumed that
both items actually share an identical inflection.
All consonants which can be palatalized are automatically palatalized
before /e/. Thus the locative singular of /zakon/, namely /zakone/, will be
realized with a palatalized [n 0 ]. If the consonant is already palatalized as in
genitive plural /kost 0 -ej/, it simply remains palatalized. Some consonants
are always hard (/š, ž, c/), and remain so before /e/. On the other hand, /č 0 /
and /šč 0 /are always soft (palatalized), and naturally remain so before /e/.
The velars /k, g, x/are palatalized before /i/, so that the genitive form
/knigi/ ‘book’, will be realized with palatalized [g 0 ] (which then demands
the front allphone of /i/, namely [i]).
References
This section lists works cited in the preceding text. An annotated bibliography on
syncretism (including some of the works below, but not restricted to them) may be
found in Baerman (2002a).
Abbot, Miriam. 1991. Macushi. In Desmond Derbyshire and Geoffey K. Pullum

(eds.), Handbook of Amazonian languages (vol. III). Berlin: Mouton de
Gruyter. 23–160.
Ackema, Peter and Ad Neeleman. 2003. Context-sensitive spell-out. Natural
Language and Linguistic Theory 21. 681–735.
Aikhenvald, Alexandra Y. and Robert M. W. Dixon. 1998. Dependencies between
grammatical systems. Language 74. 56–80.
Anttila, Raimo. 1972. An introduction to historical and comparative linguistics. New
York: Macmillan.
Arensen, Jon. 1971. A Murle grammar. Linguistic Monograph Papers No. 7.
Khartoum: Sudan Research Unit, Faculty of Arts, University of Khartoum.
Arkadiev, Pëtr M. 2003. Tipologija padežnogo sinkretizma. Ms., Russian State
University for the Humanities.
Armbruster, Charles Herbert. 1960. Dongolese Nubian: a grammar. Cambridge:
University Press.
Aronoff, Mark. 1994. Morphology by itself. Cambridge: MIT Press.
Aronson, Howard I. 1991. Modern Georgian. In Alice C. Harris (ed.), The
indigenous languages of the Caucasus (vol. I: Kartvelian languages). Delmar:
Caravan Books. 219–312.
Ashton, E. O. 1947. Swahili grammar (second edition). London: Longmans, Green
and Co.
Austin, Peter. 1981. A grammar of Diyari, South Australia. Cambridge: Cambridge
University Press.
1986. Structural change in language obsolescence: some Eastern Australian
examples. Australian Journal of Linguistics 6. 201–30.
Baerman, Matthew. 2002a. Syncretism: an annotated bibliography. Available
on-line at http://www.surrey.ac.uk/LIS/MB/Bibliography.htm.
2002b. The Surrey person syncretism database. Available on-line at: http://
www.smg.surrey.ac.uk/personsyncretism/index.aspx.
2004. Directionality and (un)natural classes in syncretism. Language 80.
807–827.
254
References 255
2005. Typology and the formal modelling of syncretism. In Geert Booij and Jaap
van Marle (eds.), Yearbook of morphology 2004. Dordrecht: Kluwer. 41–72 .
Baerman, Matthew and Dunstan Brown. 2005a. Case syncretism. In Martin
Haspelmath, Matthew Dryer, David Gil and Bernard Comrie (eds.), World
atlas of language structures. Oxford: Oxford University Press.
2005b. Verbal person/number syncretism. In Martin Haspelmath, Matthew
Dryer, David Gil and Bernard Comrie (eds.), World atlas of language
structures. Oxford: Oxford University Press.
Baerman, Matthew, Dunstan Brown and Greville G. Corbett. 2002a. The Surrey
syncretisms database. Available online at: http://www.smg.surrey.ac.uk/
syncretism/index.aspx.
2002b. Case syncretism in and out of Indo-European. In Mary Andronis,
Christopher Ball, Heidi Elston and Sylvain Neuvel (eds.), CLS 37: The main
session. papers from the 37th meeting of the Chicago Linguistic Society (vol. I).
Chicago: Chicago Linguistic Society. 15–28.
Bammesberger, Alfred. 1982. A handbook of Irish. Heidelberg: Carl Winter.
Bátori, István. 1990. Die Markierung des Objekts am Verb im Mordwinischen:
morphologische Unterbestimmtheit und Homomorphie. Nyelvtudományi
Közlemeńyek 91. 15–23.
Bavin, Edith L. and Tim Shopen. 1987. Innovations and neutralizations in the
Warlpiri pronominal system. Journal of Linguistics 23. 149–75.
Bechhaus-Gerst, Marianne. 1996. Sprachwandel durch Sprachkontakt am Beispiel
des Nubischen im Niltal. Köln: Rüdiger Köppe.
Beeler, Madison S. 1976. Barbareño Chumash grammar: a farrago. In Margaret
Langdon and Shirley Silver (eds.), Hokan studies. The Hague: Mouton. 251–70.
Béjar, Susana and Daniel Currie Hall. 1999. Marking markedness: the underlying
order of diagonal syncretisms. Paper presented at the Eastern States
Conference on Linguistics, University of Connecticut.
Bender, Lionel M. 1996. Kunama. Munich: Lincom.
Bentley, Delia and Thórhallur Eythórsson. 2001. Alternation according to person
in Italo-Romance. In Laurel J. Brinton (ed.), Historical linguistics 1999.
Amsterdam/Philadelphia: John Benjamins. 63–74.
Berger, Hermann. 1998. Die Burushaski-Sprache von Hunza und Nager (vol. I:
Grammatik). Wiesbaden: Harrassowitz.
Berman, Ruth. 1997. Hebrew. In Robert Hetzron (ed.), The Semitic languages.
London: Routledge. 312–34.
Bierwisch, Manfred. 1967. Syntactic features in morphology: general problems
of so-called pronominal inflection in German. In To honor Roman
Jakobson: essays on the occasion of his seventieth birthday. The Hague:
Mouton. 239–70.
Blake, Barry J. 1994. Case. Cambridge: Cambridge University Press.
Bleiching, Doris, Guido Drexel and Dafydd Gibbon. 1996. Ein Synkretismusmodell
für die deutsche Morphologie. In Dafydd Gibbon (ed.), Natural language
processing and speech technology. Results of the third KONVENS Conference,
Bielefeld. Berlin: Mouton de Gruyter. 237–48.
Blevins, James. 2003 Stems and paradigms. Language 79. 737–67.
256 References
Bobaljik, Jonathan. 2002. Syncretism without paradigms: remarks on Williams

1981, 1994. In Geert Booij and Jaap van Marle (eds.), Yearbook of
morphology 2001. Dordrecht: Kluwer. 53–86.
Boeder, Winfried. 1976. Morphologische Kategorien. In K. Braunmüller and
W. Kürschner (eds.), Grammatik. Akten des 10. Linguistischen Kolloquiums,
Tübingen 1975 (vol. II). Tübingen: Niemeyer. 117–26.
Booij, Geert. 1996. Inherent versus contextual inflection and the split morphology
hypothesis. In Geert Booij and Jaap van Marle (eds.), Yearbook of morphology
1995. Dordrecht: Foris. 1–16.
Bowe, Heather. 1990. Categories, constituents, and constituent order in
Pitjantjatjara. London: Routledge.
Boyeldieu, Pascal. 1982. Deux e´tudes laal. Berlin: D. Reimer.
Breedveld, J. O. 1995. Form and meaning in Fulfulde: a morphophonological study of
Maasinankoore. Leiden: CNWS.
Breen, J. Gavan. 1976. Wagaya. In Robert M. W. Dixon (ed.), Grammatical
categories in Australian languages. Canberra: Australian Institute of
Aboriginal Studies. 590–4.
Bright, William. 1957. The Karok language. Berkeley: University of California
Press.
Brockelmann, Carl. 1908–13. Grundriss der vergleichenden Grammatik der
semitischen Sprachen. Berlin: Reuther and Reichard.
Bromley, H. Myron. 1981. A grammar of Lower Grand Valley Dani. (Pacific
Linguistics C63). Canberra: Australian National University.
Brown, Dunstan. 1998a. Defining ‘subgender’: virile and devirilised nouns in
Polish. Lingua 104. 187–233.
1998b. From the general to the exceptional: a Network Morphology account of
Russian nominal inflection. PhD thesis, University of Surrey.
2001. Constructing a typological database for inflectional morphology: the
SMG database for syncretism. In Steven Bird, Peter Buneman and Mark
Liberman (eds.), Proceedings of the IRCS Workshop on Linguistic
Databases. Philadelphia: Institute for Research in Cognitive Science,
University of Pennsylvania. 56–64 (paper available at: http://
www.ldc.upenn.edu/annotation/database/proceedings.html).
Brown, Dunstan, Greville Corbett, Norman Fraser, Andrew Hippisley and Alan
Timberlake. 1996. Russian noun stress and network morphology. Linguistics
34. 53–107.
Browne, Gerald M. 2002. A grammar of Old Nubian. Munich: Lincom.
Buck, Carl D. 1933. Comparative grammar of Greek and Latin. Chicago: University
of Chicago Press.
Burrow, Thomas and Sudhibhushan Bhattacharya. 1970. The Pengo language.
Oxford: Clarendon Press.
Butt, John and Carmen Benjamin. 2000. A new reference grammar of modern
Spanish. London: Arnold.
Bybee, Joan. 1985. Morphology. Amsterdam/Philadelphia: John Benjamins.
Cahill, Lynne and Gerald Gazdar. 1997. The inflectional phonology of German
adjectives, determiners, and pronouns. Linguistics 35. 211–45.
References 257
1999. The German noun inflection. Journal of Linguistics 35. 1–42.

Cairns, Charles E. 1986. Word structure, markedness, and applied linguistics.
In Fred R. Eckman, Edith A. Moravcsik and Jessica R. Wirth (eds.),
Markedness. Proceedings of the twelfth annual linguistics symposium of the
University of Wisconsin-Milwaukee, March 11–12, 1983. New York: Plenum
Press. 13–38.
Calabrese, Andrea. 1998. Some remarks on the Latin case system and its
development in Romance. In José Lema and Esthela Treviño (eds.),
Theoretical analyses on Romance languages. Amsterdam/Philadelphia: John
Benjamins. 71–126.
Cardona, George. 1964. A Gujarati reference grammar. Philadelphia: University of
Pennsylvania Press.
Cardona, George and Babu K. Suthar. 2003. Gujarati. In George Cardona and
Dhanesh Jain (eds.), The Indo-Aryan languages. London: Routledge. 658–97.
Carmack, Stanford. 1997. Blocking in Georgian verb morphology. Language 72.
314–38.
Carstairs, Andrew. 1984. Outlines of a constraint on syncretism. Folia Linguistica
18. 73–85.
1987. Allomorphy in inflection. London: Croom Helm.
Carstairs-McCarthy, Andrew. 1994. Inflection classes, gender, and the principle of
contrast. Language 70. 737–88.
1998a. How lexical semantics constrains inflectional allomorphy. In Geert Booij
and Jaap van Marle (eds.), Yearbook of morphology 1997. Dordrecht:
Kluwer. 1–24.
1998b. Comments on the paper by Noyer. In Steven G. Lapointe, Diane
K. Brentari, and Patrick M. Farrell (eds.), Morphology and its relation to
phonology and syntax. Stanford: CSLI Publications. 286–301.
Carter, Hazel and João Makoondekwa. 1979. Kongo course: maloòngi makikóongo
(dialect of Zoombo, Angola). London: School of Oriental and African Studies.
Chafe, Wallace. 1997. Sketch of Seneca, an Iroquoian language. In Ives Goddard
(ed.) Handbook of North American Indians (vol. XVII: Languages).
Washington: Smithsonian Institute. 551–79.
Chapman, Shirley and Desmond C. Derbyshire. 1990. Paumarı̀. In Desmond
C. Derbyshire and Geoffrey K. Pullum (eds.), Handbook of Amazonian
languages (vol. III). Berlin: Mouton de Gruyter. 161–352.
Charney, Jean Ormsbee. 1993. A grammar of Comanche. Lincoln: University of
Nebraska Press.
Childs, George Tucker. 1995. A grammar of Kisi: a southern Atlantic language.
Berlin: Mouton de Gruyter.
Chvany, Catherine V. 1986. Jakobson’s fourth and fifth dimensions: on reconciling
the cube model of case meanings with the two-dimensional matrices for case
forms. In Richard D. Brecht and James S. Levine (eds.), Case in Slavic.
Columbus: Slavica. 107–29.
Clairis, Christos. 1985. El Qawasqar: lingüı´stica fueguina, teorıá y descripción.
(Estudios filológicos, Anejo 12.) Valdivia, Chile: Facultad de Filosofı́a y
Humanidades, Universidad Austral de Chile.
258 References
Coleman, Robert. 1976. Patterns of Syncretism in Latin. In Anna M. Davies and

Wolfgang Meid (eds.), Studies in Greek, Italic and Indo-European linguistics
offered to Leonard R. Palmer. Innsbruck: Institut für Sprachwissenschaft der
Universität Innsbruck. 47–56.
1991. The assessment of paradigm stability: some Indo-European case studies. In
Frans Plank (ed.), Paradigms: the economy of inflection. Berlin: Mouton de
Gruyter. 197–211.
Comrie, Bernard. 1975. Polite plurals and predicate agreement. Language 51.
406–18.
1980. Inverse verb forms in Siberia: evidence from Chukchee, Koryak, and
Kamchadal. Folia Linguistica Historica 1/1. 61–74.
1991. Form and function in identifying cases. In Frans Plank (ed.), Paradigms:
the economy of inflection. Berlin: Mouton de Gruyter. 41–55.
Corbett, Greville G. 1983. Hierarchies, targets and controllers: agreement patterns
in Slavic. London: Croom Helm.
1991. Gender. Cambridge: Cambridge University Press.
2000. Number. Cambridge: Cambridge University Press.
2005. The number of genders. In Martin Haspelmath, Matthew Dryer, David
Gil and Bernard Comrie (eds.), World atlas of language structures. Oxford:
Oxford University Press.
Corbett, Greville G., Matthew Baerman and Dunstan Brown. 2002. Domains of
syncretism: a demonstration of the autonomy of morphology. In Mary
Andronis, Christopher Ball, Heidi Elston and Sylvain Neuvel (eds.), CLS
37: The Panels. Papers from the 37th meeting of the Chicago Linguistic
Society (vol. I). Chicago: Chicago Linguistic Society. 385–98.
Corbett, Greville G. and Norman M. Fraser. 1993. Network morphology: a DATR
account of Russian nominal inflection. Journal of Linguistics 29. 113–42.
1997. Komp0 juternaja lingvistika i tipologija. Vestnik Moskovskogo Universiteta
(ser. 9. Filologija, no. 2). 122–40.
Cowan, H. K. J. 1965. Grammar of the Sentani language. ’s-Gravenhage: Martinus
Nijhoff.
Crowley, Terry. 1998. An Erromangan (Sye) grammar. Honolulu: University of
Hawaii Press.
1999. Ura: a disappearing language of southern Vanuatu. (Pacific Linguistics
C156). Canberra: Australian National University.
Curnow, Timothy J. 1997. A grammar of Awa Pit (Cuaiquer): An indigenous
language of South-western Colombia. PhD thesis, Australian National
University.
Curtius, Georg. 1863. Erläuterungen zu meiner Griechischen Schulgrammatik.
Prague: Tempsky.
Cysouw, Michael. 2003. The paradigmatic structure of person marking. Oxford:
Oxford University Press.
Cysouw, Michael. Forthcoming. Syncretisms involving clusivity. In Elena
Filimonova (ed.), Clusivity. Amsterdam/Philadelphia: John Benjamins.
Dalrymple, Mary and Ronald M. Kaplan. 2000. Feature indeterminancy and
feature resolution. Language 76. 759–98.
References 259
Davies, John. 1981. Kobon (Lingua Descriptive Studies 3). Amsterdam: North-
Holland.
de Angulo, Jaime. 1932. The Chichimeco language (Central Mexico). International
Journal of American Linguistics 7. 152–94.
Dedrick, John M. and Eugene H. Casad. 1999. Sonora Yaqui language structures.
Tucson: University of Arizona Press.
Dench, Alan C. 1995. Martuthunira: a language of the Pilbara region of western
Australia. Canberra: Australian National University (Pacific Linguistics C125).
2001. Descent and diffusion: the complexity of the Pilbara situation. In
Alexandra Y. Aikhenvald and Robert M. W. Dixon (eds.), Areal diffusion
and genetic inheritence: problems in comparative linguistics. Oxford: Oxford
University Press. 105–33.
Derbyshire, Desmond. 1979. Hixkaryana (Lingua Descriptive Studies 1).
Amsterdam: North-Holland.
Deza Galindo, Juan Francisco. 1992. Gramática de la lengua aymara. Lima: Artex
Editores.
Dickens, Patrick. 1992. A Ju/0 hoan grammar. Ms., Nyae Development Foundation
of Namibia.
Dimmendaal, Gerrit J. 1983. The Turkana language. Dordrecht: Foris.
Dixon, Robert M. W. 1977. A grammar of Yidi . Cambridge: Cambridge
University Press.
¸
1988. A grammar of Boumaa Fijian. Chicago: University of Chicago Press.
1994. Ergativity. Cambridge: Cambridge University Press.
Donaldson, Tamsin. 1980. Ngiyambaa: the language of the Wangaaybuwan.
Cambridge: Cambridge University Press.
Drabbe, Peter. 1952. Spraakkunst van het Ekagi. ’s-Gravenhage: Martinus Nijhoff.
1955. Spraakkunst van het Marind, zuidkust Nederlands Nieuw-Guinea. Mödling:
Missiehuis St. Gabriël (Reprinted 1966 by Johnson Reprint Corp., New York).
van Driem, George. 1987. A grammar of Limbu. Berlin: Mouton de Gruyter.
Dutton, Tom. 2003. A dictionary of Koiari, Papua New Guinea, with grammar notes.
Canberra: Australian National University.
Elliot, Eric. 1999. Dictionary of Rincón Luiseño. PhD thesis, University of
California at San Diego.
Endzelıns [Endzelin], J
anis. 1922. Lettische Grammatik. Riga: Gulbis.
Engel, Ralph, Mary Allhiser de Engel and José Mateo Alvarez. 1987. Diccionario
zoque de Francisco Leon. Hidalgo: Instituto Lingüı́stico de Verano.
Evans, Nicholas. 1995. A grammar of Kayardild: with historical-comparative notes
on Tangkic. Berlin: Mouton de Gruyter.
2003. Bininj Gun-Wok: a pan-dialectal grammar of Mayali, Kunwinjku and Kune
(Pacific Linguistics 541). Canberra: Australian National University.
Evans, Nicholas, Dunstan Brown and Greville G. Corbett. 2001. Dalabon
pronominal prefixes and the typology of syncretism: a Network
Morphology analysis. In Geert Booij and Jaap van Marle (eds.), Yearbook
of morphology 2000. Dordrecht: Kluwer. 187–231.
2002. The semantics of gender in Mayali: partially parallel systems and formal
implementation. Language 78. 111–55.
260 References
Evans, Roger and Gerald Gazdar. 1996. DATR: a language for lexical knowledge
representation. Computational Linguistics 22. 167–216.
Ezard, Bryan. 1997. A grammar of Tawala: an Austronesian language of the Milne
Bay Area, Papua New Guinea (Pacific Linguistics C137). Canberra: Australian
National University.
Feldstein, Ronald. 2003. On the structure of syncretism in Romanian conjugation.
Ms., Indiana University.
Fennell, Trevor G. 1975. Is there an instrumental case in Latvian? Journal of Baltic
Studies 6. 41–8.
Feoktistov, Aleksandr p. 1966. Erzjanskij jazyk. Jazyki narodov SSSR (vol. III:
Finno-ugorskie jazyki). Moscow: Nauka. 177–98.
Finkel, Raphael, Lei Shen, Gregory Stump and Suresh Thesayi, 2002. KATR: a set-
based extension of DATR (Technical report 346–02). University of Kentucky
Department of Computer Science, Lexington, KY.
Fischer, Wolfdietrich. 1997. Classical Arabic. In Robert Hetzron (ed.), The Semitic
languages. London: Routledge. 187–219.
Foley, William A. 1986. The Papuan languages of New Guinea. Cambridge:
Cambridge University Press.
1991. The Yimas language of New Guinea. Stanford: Stanford University Press.
Fortescue, Michael. 1984. West Greenlandic (Croom Helm Descriptive
Grammars). Dover, New Hampshire: Croom Helm.
Frachtenberg, Leo J. 1922. Siuslaw. In Franz Boas (ed.), Handbook of American
Indian languages (vol. II) (Smithsonian Institution Bureau of American
Ethnology Bulletin). Washington: Government Printing Office. 431–629.
Fradkin, Robert A. 1991. Marking, markedness, and person-gender-number
patterning in the Arabic tenses and moods. Folia Linguistica 25. 609–64.
Frank, Paul. 1990. Ika syntax (Studies in the Languages of Colombia 1). Dallas:
Summer Institute of Linguistics.
Frank, Wright Jay. 1999. Nuer noun morphology. MA thesis, State University of
New York, Buffalo.
Franklin, Karl James. 1971. A grammar of Kewa, New Guinea (Pacific Linguistics
C16). Canberra: Australian National University.
Fraser, Norman M. and Greville G. Corbett. 1995. Gender, animacy and declensional
class assignment: a unified account for Russian. In Geert Booij and Jaap van
Marle (eds.), Yearbook of morphology 1994. Dordrecht: Kluwer. 123–50.
1997. Defaults in Arapesh. Lingua 103. 25–57.
Fromm, Hans. 1982. Finnische Grammatik. Heidelberg: Carl Winter.
Glasgow, Kathleen. 1984. Burarra word classes. In Kathleen Glasgow et al. (eds.),
Papers in Australian linguistics 16 (Pacific Linguistics A68). Canberra:
Australian National University. 1–54.
Goddard, Cliff. 1982. Case systems and case marking in Australian languages: a
new interpretation. Australian Journal of linguistics 2. 167–96.
Gordon, Lynn. 1986. Maricopa morphology and syntax (University of California
Publications in Linguistics 108). Berkeley: University of California Press.
Graudiņa, A. 1969. Liter ar
as valodas un izlokšņu mijiedarbıba. Latviesˇu valodas
kulturas jautājumi. 16–20.
References 261
Greenberg, Joseph H. 1963. Some universals of grammar with particular reference

to the order of meaningful elements. In Joseph H. Greenberg (ed.), Universals
of language. Cambridge, Massachusetts: MIT Press. 73–113.
Gregores, Emma, and Jorge A. Suárez. 1967. A description of colloquial Guarani.
The Hague: Mouton.
Grimm, Jacob and Willhelm Grimm. 2004. Deutsches Wörterbuch (Digital
edition). Frankfurt am Main: Zweitausendeins.
Grjunberg, A. L. and D. I. Edel0 man. 1987. Afganskij jazyk. In Osnovy iranskogo
jazykoznanija (vol. IV: Novoiranskie jazyki: vostocˇnaja gruppa). Moscow:
Nauka. 6–154.
Grosse, Siegfried. 2000. Morphologie des Mittelhochdeutschen. In Werner Besch,
Anne Betten, Oskar Reichmann and Stefan Sonderegger (eds.),
Sprachgeschichte. Ein Handbuch zur Geschichte der deutschen Sprache und
ihrer Erforschung (vol. 1.2). Berlin: Mouton de Gruyter. 1332–40.
Gruzdeva, Elena Ju. 1997. Nivxskij jazyk. In Jazyki mira: paleoaziatskie jazyki.
Moscow: Indrik. 139–54.
Güldemann, Tom and Rainer Vossen. 2000. Khoisan. In Bernd Heine and Derek
Nurse (eds.), African languages. Cambridge: Cambridge University Press. 99–122.
Gvozdanović, Jadranka. 1991. Syncretism and the paradigmatic patterning of
grammatical meaning. In Frans Plank (ed.), Paradigms: the economy of
inflection. Berlin: Mouton de Gruyter. 133–60.
Hagman, Roy S. 1977. Nama Hottentot grammar. Bloomington: Indiana University.
Halle, Morris. 1997. Distributed morphology: impoverishment and fission. MIT
Working Papers in Linguistics 30. 425–49.
Hamel, Patricia J. 1994. A grammar and lexicon of Loniu, Papua New Guinea
(Pacific Linguistics C103). Canberra: Australian National University.
Hansson, Gunnar O. 1996. Productive syncretism and morphological theory: a
case study of North Sami inflectional morphology. Ms., University of
California, Berkeley.
Harbour, Daniel. 2003. Elements of number theory. PhD thesis, Massachusetts
Institute of Technology.
Harley, Heidi and Elizabeth Ritter. 2002. Person and number in pronouns: a
feature-geometric analysis. Language 78. 482–526.
Harvey, Mark. 2002. A grammar of Gaagudju. Berlin: Mouton de Gruyter.
Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter.
Haspelmath, Martin, Matthew Dryer, David Gil and Bernard Comrie (eds.), 2005.
World atlas of language structures. Oxford: Oxford University Press.
Haugen, Einar. 1982. Scandinavian language structures: a comparative historical
survey. Minneapolis: University of Minnesota Press.
Haviland, John. 1979. Guugu Yimidhirr. In Robert M. W. Dixon and Barry
J. Blake (eds.), Handbook of Australian languages. Canberra: Australian
National University Press. 27–180.
Hayward, Richard. 1984. The Arbore language. Hamburg: H. Buske.
Healy, Alan, Ambrose Isoroembo and Martin Chittleborough. 1969. Preliminary
notes on Orokaiva grammar. In Papers in New Guinea Linguistics 9 (Pacific
Linguistics A18). 33–64.
262 References
Heath, Jeffrey. 1984. Functional grammar of Nunggubuyu. Atlantic Highlands

NJ: Humanities Press (and Canberra: Australian Institute of Aboriginal
Studies).
1991. Pragmatic disguise in pronominal-affix paradigms. In Frans Plank (ed.),
Paradigms: the economy of inflection. Berlin: Mouton de Gruyter. 75–90.
1998. Pragmatic skewing in 1 « 2 pronominal combinations in Native American
languages. International Journal of American Linguistics 64. 83–104.
Heine, Bernd. 1981. The non-Bantu languages of Kenya. Berlin: Reimer.
1982. African noun class systems. In Hansjakob Seiler and Christian Lehman
(eds.), Apprehension: das sprachliche Erfassen von Gegenständen (vol. I).
Tübingen: Narr. 189–216.
Hekking, Ewald. 1995. El Otomı´ de Santiago Mexquititlán: desplazamiento
lingüı´stico, pre´stamos y cambios gramaticales. Amsterdam: Institute for
Functional Research into Language and Language Use.
Helimski, Eugene. 1998. Selkup. In Daniel Abondolo (ed.), The Uralic languages.
Helmbrecht, Johannes. 1999. The typology of 1st person marking and its cognitive
background. In Masako K. Hiraga, Chris Sinha and Sherman Wilcox (eds.),
Cultural, psychological and typological issues in cognitive linguistics.
2003. Ikonizität in Personalpronomina (Arbeitspapiere des Seminars für
Sprachwissenschaft der Universität Erfurt nr. 5). Erfurt: Seminar für
Sprachwissenschaft der Universität Erfurt.
Henderson, James. 1995. Phonology and grammar of Yele, Papua New Guinea
(Pacific Linguistics B112). Canberra: Australian National University.
Herbermann, Charles G., Edward A. Pace, Condé B. Pallen, Thomas J. Shahan
and John J. Wynne. 1907–18. The Catholic encyclopedia. New York: Robert
Appleton.
Hetzron, Robert. 1967. Agaw numerals and incongruence in Semitic. Journal of
Semitic Studies 12. 169–97.
Hewitt, Brian G. 1995. Georgian: a structural reference grammar. Amsterdam/
Philadelphia: John Benjamins.
Hewson, John. 1989. Motivated syncretism. Journal of the Atlantic Provinces
Linguistic Association/Revue de l‘Association de Linguistique des Provinces
Atlantique 11. 39–56.
Hippisley, Andrew. 1997. Declarative derivation: a Network Morphology account
of Russian word formation with reference to nouns denoting ‘person’. PhD
thesis, University of Surrey.
2001. Word formation rules in a default inheritance framework. In Jaap van
Marle and Geert Booij (eds.), Yearbook of morphology 1999. Dordrecht:
Kluwer. 221–61.
Hjelmslev, Louis. 1935–7. La cate´gorie des cas: e´tude de grammaire geńe´rale (Acta
Jutlandica VII/1). Aarhus: Universitetsforlaget.
Hoff, B. J. 1968. The Carib language. The Hague: M. Nyhoff.
Holton, David, Peter Mackridge and Irene Philippaki-Warburton. 1997. Greek: a
comprehensive grammar of the modern language. London: Routledge.
References 263
Horton, A. E. 1949. A grammar of Luvale. Johannesburg: Witwatersrand

University Press.
Hualde, José Ignacio and Jon Ortiz de Urbina (eds.). 2003. A grammar of Basque.
Berlin: Mouton de Gruyter.
Hudson, Richard A. 1974. A structural sketch of Beja. In D. W. Arnott (ed.), African
language studies. London: School of Oriental and African Studies. 111–42.
Huntley, David. 1980. The evolution of genitive-accusative animate and personal
nouns in Slavic dialects. In Jacek Fisiak (ed.), Historical morphology. The
Hague: Mouton. 189–212.
Hutchison, John P. 1981. The Kanuri language: a reference grammar. Madison:
African Studies Program, University of Wisconsin.
Innes, Gordon. 1966. An introduction to Grebo. London: School of Oriental and
African Studies.
Ivanov, Vjačeslav V. 2001. Xettskij jazyk (second edition). Moscow: URSS.
Jackendoff, Ray S. 1985. Multiple subcategorization and the u-criterion: the case of
climb. Natural Language and Linguistic Theory 3. 271–95.
Jacobson, Steven A. 1995. A practical grammar of the Central Alaskan Yup’ik
Eskimo language. Fairbanks: Alaska: Alaska Native Language Center,
University of Alaska.
Jakobi, Angelika. 1990. A Fur grammar: phonology, morphophonology, and
morphology. Hamburg: Helmut Buske.
Jakobson, Roman O. 1936 [1971]. Beitrag zur allgemeinen Kasuslehre:
Gesamtbedeutung der russischen Kasus. Reprinted in Selected writings (vol. II:
Word and language). The Hague: Mouton. 23–71 (originally in Travaux du
Cercle Linguistique de Prague VI. 240–99).
1958. Morfologičeskie nabljudenija nad slavjanskim skloneniem. In American
contributions to the Fourth International Congress of Slavists, Moscow.
’s-Gravenhage: Mouton. 127–56.
Johnston, Jason. 1997. Systematic homonymy and the structure of morphological
categories: some lessons from paradigm geometry. PhD thesis, University of
Sydney.
Kachru. Braj B. 1969. A reference grammar of Kashmiri. Urbana: Department of
Linguistics, Univerity of Illinois.
Kariņš, A. Krišj
anis. 1994. Functionalism and linguistic change in Latvian verb
morphology. Linguistica Baltica 3. 109–20.
Karlson, Fred. 1999. Finnish: an essential grammar. London/New York:
Routledge.
Kathol, Andreas. 2002. Nominal head-marking constructions: two case studies
from Luiseño. In Frank Van Eynde, Lars Hellan and Dorothee Beermann
(eds.), Proceedings of the eighth international conference on Head-Driven
Phrase Structure Grammar. Stanford: CSLI Publications. 189–201.
Kennedy, Benjamin Hall. 1955. The revised Latin primer (edited and further revised
by Sir James Mountford). London: Longmans, Green and Co.
Kettunen, Lauri. 1938. Livisches Wörterbuch mit grammatischer Einleitung.
Helsinki: Suomalais Ugrilainen Seura.
Key, Harold H. 1967. Morphology of Cayuvava. The Hague: Mouton.
264 References
Khubchandani, Lachman M. 2003. Sindhi. In George Cardona and Dhanesh Jain

(eds.), The Indo-Aryan languages. London: Routledge. 622–58.
Kibrik, Aleksandr E. 1997. Ierarxii, roli, nuli, markirovannost0 i ‘anomal0 naja’
upakovka grammatičeskoj semantiki. Voprosy Jazykoznanija 4. 27–57.
(ed.) 1999. Èlementy caxurskogo jazyka v tipologicˇeskom osvesˇcˇenii. Moscow:
Nasledie.
Kienast, Burkhart. 2001. Historische semitische Sprachwissenschaft. Wiesbaden:
Harrassowitz.
Kiparsky, Paul. 2001. Structural case in Finnish. Lingua 111. 315–76.
Klenin, Emily. 1983. Animacy in Russian: a new interpretation. Columbus: Slavica.
Koneski, Blaže. 1996. Istorija na makedonskiot jazik (reprint of 1986 edition).
Skopje: Detska Radost.
Kotyczka, Jozef. 1980. Kurze polnische Sprachlehre. Berlin: Volk und Wissen.
Krishnamurti, Bhadriraju and John P. L. Gwynn. 1985. A grammar of modern
Telugu. Delhi: Oxford University Press.
Kuryłowicz, Jerzy. 1949 [1960]. De la nature des procès dits ‘analogiques’.
Reprinted in: Esquisses linguistiques. Wroctław: Zaktład narodowy im.
Ossolińskich. 66–86.
Kutsch Lojenga, Constance. 1994. Ngiti: A Central-Sudanic language of Zaire.
Cologne: Rüdiger Köppe.
Lakämper, Renate and Dieter Wunderlich. 1998. Person marking in Quechua – a
constraint-based minimalist analysis. Lingua 105. 113–48.
Lamberti, Marcello. 1993. Die Shinassha-Sprache. Heidelberg: Carl Winter.
Lefebvre, Claire and Anne-Marie Brousseau. 2002. A grammar of Fongbe. Berlin:
Mouton de Gruyter.
Leumann, Manu. 1977. Lateinische Laut- und Formenlehre (fifth edition). Munich:
Beck.
Liddell Henry G. and Robert Scott. 1996. A Greek–English lexicon. Oxford:
Clarendon Press.
Lindstrom, Lamont and John Lynch. 1994. Kwamera. Munich: Lincom.
Lipkind, William. 1945. Winnebago grammar. New York: King’s Crown Press.
Lorimer, David L. R. 1939. The Dum aki Language: outlines of the speech of the
Doma, or B ericho, of Hunza. Nijmegen: Dekker and van de Vegt.
Luraghi, Silvia. 2000. Synkretismus. In Geert Booij, Christian Lehmann and
Joachim Mugdan (eds.), Morphologie: ein Handbuch zur Flexion und
Wortbildung. Berlin: Mouton de Gruyter. 638–47.
Lynch, John. 2000. A grammar of Anejom. Canberra: Australian National University.
Magometbekova, Z. M. 1967. Karatinskij jazyk. In E. A. Bokarev and
K. V. Lomatidze (eds.), Jazyki narodov SSSR (vol. IV: Iberijsko-kavkazskie
jazyki). Moscow: Nauka. 323–35.
Maiden, Martin. 1992. Irregularity as a determinant of morphological change.
Journal of Linguistics 28/2. 285–312.
Malécot, André. 1963. Luiseño, a structural analysis II: morphosyntax.
International Journal of American Linguistics 29. 196–210.
Maslova, Elena. 2003. A grammar of Kolyma Yukaghir. Berlin: Mouton de
Gruyter.
References 265
Mathiassen, Terje. 1997. A short grammar of Latvian. Columbus: Slavica.

Matras, Yaron. 2002. Romani: a linguistic introduction. Cambridge: Cambridge
University Press.
Mayerthaler, Willi. 1987. System-independent morphological naturalness. In
Wolfgang U. Dressler (ed.), Leitmotifs in natural morphology. Amsterdam/
Philadelphia: John Benjamins. 25–58.
McCarthy, John J. 2002. A thematic guide to Optimality Theory. Cambridge:
Cambridge University Press.
McCreight, Katherine and Catherine V. Chvany. 1991. Geometric representation
of paradigms in a modular theory of grammar. In Frans Plank (ed.),
McGregor, Ronald S. 1977. Outline of Hindi grammar. Delhi: Oxford University
Press.
McGregor, William. 1990. A functional grammar of Gooniyandi. Amsterdam: John
Benjamins.
McGuckin, Catherine. 2002. Gapapaiwa. In John Lynch, Malcom Ross and Terry
Crowley (eds.), The Oceanic languages. Richmond: Curzon. 297–321.
Meillet, Paul and André Vaillant. 1934. Le slave commun. Paris: Institut d’études
slaves.
Meiser, G. 1992. Syncretism in Indo-European languages. Transactions of the
Philological Society 90/2. 187–218.
Merlan, Francesca. 1982. Mangarayi (Lingua Descriptive Studies 4). Amsterdam:
North-Holland.
Miller, Raymond. 1990. Form and function of the peripheral cases in some Slovene
dialects. Slovene Studies 12/1. 5–22.
Minassian, Martiros. 1980. Grammaire d’Armenien oriental. Delmar: Caravan
Books.
Mitchell, Terence F. 1962. Colloquial Arabic: the living language of Egypt. London:
English Universities Press.
Moravcsik, Edith and Jessica Wirth. 1986. Markedness – an overview. In Fred
R. Eckman, Edith A. Moravcsik and Jessica R. Wirth (eds.),
Markedness: proceedings of the twelfth annual linguistics symposium of the
University of Wisconsin-Milwaukee, March 11–12, 1983. New York: Plenum
Press. 1–11.
Morgenstierne, Georg. 1941. Notes on Phal ura, an unknown language of Chitral.
Oslo: Hos Jacob Dybwad (Skirfter utgit av Det Norske Videnskaps-Akademi i
Oslo II. Hist.-Filos. Klasse. No. 5).
Mous, Maarten. 1993. A grammar of Iraqw (Kuschitische Sprachstudien 9).
Hamburg: Buske.
Müller, Gereon. Forthcoming. A Distributed Morphology approach to syncretism
in Russian noun inflection. In Olga Arnaudova, Wayles Browne, Maria Luisa
Rivero and Dejan Stojanović (eds.), Proceedings of Formal Approaches to
Slavic Linguistics 12. Ann Arbor: Michigan Slavic Publications.
Munro, Pamela. 1976. Mojave syntax. New York: Garland.
Murane, Elizabeth. 1974. Daga grammar: from morpheme to discourse. Norman:
Summer Institute of Linguistics.
266 References
Muysken, Pieter. 1997. Callahuaya. In: Sally G. Thomasan (ed.), Contact

languages. Amsterdam/Philadelphia: John Benjamins. 427–48.
Myachina, E. N. 1981. The Swahili language: a descriptive grammar (translated by
G. L. Campdell). London: Routledge.
Najlis, Elena. 1966. Lengua abipona (vol. I). Buenos Aires: Centro de Estudios,
Universidad de Buenos Aires.
Neidle, Carol J. 1988. The role of case in Russian syntax. Dordrecht: Kluwer.
Neue, Friedrich and C. Wagener. 1902. Formenlehre der lateinischen Sprache
(vol. I: Das Substantivum). Leipzig: O. R. Reisland.
Newmark, Leonard. 1982. Standard Albanian: a reference grammar for students.
Stanford: Stanford University Press.
Nichols, Johanna. 1994. Ingush. In Rieks Smeets (ed.), The indigenous languages of
the Caucasus (vol. IV). Delmar, New York: Caravan Books. 79–145.
Nikolaeva, Irina and Evgenij A. Xelimskij. 1997. Jukagirskij jazyk. In Jazyki mira:
paleoaziatskie jazyki. Moscow: Indrik. 155–68.
Nikolaeva, Irina and Maria Tolskaya. 2001. A grammar of Udihe. Berlin: Mouton
de Gruyter.
Noonan, Michael. 1992. A grammar of Lango. Berlin: Mouton de Gruyter.
Nordlinger, Rachel. 1998. A grammar of Wambaya, Northern Territory (Australia)
(Pacific Linguistics C140). Canberra: Australian National University.
Noreen, Adolf. 1923. Altislandische und altnorwegische Grammatik. Halle: Max
Niemeyer.
Noyer, Rolf. 1997. Features, positions and affixes in autonomous morphological
structure. New York: Garland.
1998. Impoverishment theory and morphosyntactic markedness. In Steven
G. Lapointe, Diane K. Brentari, and Patrick M. Farrell (eds.),
Morphology and its relation to phonology and syntax. Stanford: CSLI
Publications. 264–85.
2001. Clitic sequences in Nunggubuyu and PF convergence. Natural Language
and Linguistic Theory 19/4. 751–826.
Nussbaum, Loren V., William W. Gage and Daniel Varre. 1970. Dakar Wolof: a
basic course. Washington, DC: Center for Applied Linguistics.
Oates, William J. and Lynette F. Oates. 1968. Kapau pedagogical grammar (Pacific
Linguistics C10). Canberra: Australian National University.
Osborne, Charles R. 1974. The Tiwi language. Canberra: Australian Institute of
Aboriginal Studies.
Owens, Jonathan. 1985. A grammar of Harar Oromo (northeastern Ethiopia).
Hamburg: Buske.
Oxotina, Natalija V. 1961. Jazyk zulu. Moscow: Nauka.
Perlmutter, David M. and Janez Orešnik. 1973. Language-particular rules and
explanation in syntax. In Steven R. Anderson and Paul Kiparsky (eds.),
A festschrift for Morris Halle. New York: Holt Rinehart. 419–59.
Pike, Kenneth. 1965. Non-linear order and anti-redundancy in German
morphological matrices. Zeitschrift für Mundartforschung 32. 193–221.
Pitman, Donald. 1980. Bosquejo de la gramatica araona (Notas Lingüı́sticas, no. 9).
Riberalta, Bolivia: Instituto Linguistico de Verano.
References 267
Plank, Frans. 1991. Rasmus Rask’s dilemma. In Frans Plank (ed.), Paradigms: the
economy of inflection. Berlin: Mouton de Gruyter. 161–96.
Plungian, Vladimir. 1995. Dogon. Munich: Lincom.
Pokorny, Julius. 1923. A historical reader of Old Irish. Halle: Max Niemeyer.
Popjes, Jack, and Jo Popjes. 1986. Canela-Kraho. In Desmond C. Derbyshire and
Geoffrey K. Pullum (eds.), Handbook of Amazonian languages (vol. I). Berlin:
Mouton de Gruyter. 128–99.
Pott, August. 1836. Etymologische Forschungen auf dem Gebiete der Indo-
Germanischen Sprachen (vol. II). Lemgo: Meyer’sche Hof-Buchhandlung.
1859. Etymologische Forschungen auf dem Gebiete der Indo-Germanischen
Sprachen (vol. I: Prepositionen). Lemgo and Detmold: Meyer’sche Hof-
Buchhandlung.
Priestly, T. M. S. 1993. Slovene. In Bernard Comrie and Greville G. Corbett (eds.),
The Slavonic languages. London: Routledge. 388–451.
Reesink, Ger P. 1987. Structures and their functions in Usan: a Papuan language of
Papua New Guinea. Amsterdam/Philadelphia: John Benjamins.
(ed.), 2002. Languages of the Eastern Bird’s Head. Canberra: Australian
National University.
Reh, Mechthild. 1985. Die Krongo-Sprache (Niino Mo-Di). Berlin: Dietrich
Reimer.
Rice, Keren. 2000. Morpheme order and semantic scope. Cambridge: Cambridge
University Press.
Ringe, Donald. 1995. Nominative-accusative syncretism and syntactic case. Penn
Working Papers in Linguistics 2. 45–81.
Roberts, John R. 1987. Amele. London: Croom-Helm.
Robins, R. H. 1958. The Yurok language. Berkeley: University of California Press.
Romero-Figueroa, Andres. 1997. A reference grammar of Warao. Munich:
Lincom.
Rothe, Wolfgang. 1957. Einführung in die historische Laut- und Formenlehre des
Rumänischen. Halle: Max Niemeyer.
Rounds, Carol. 2001. Hungarian: an essential grammar. London/New York:
Routledge.
Rumsey, Alan. 1982. An intra-sentence grammar of Ungarinjin, North-Western
Australia (Pacific linguistics B86). Canberra: Australian National University.
Rupp, James E. 1989. Lealao Chinantec syntax (Studies in Chinantec Languages 2).
Dallas: Summer Institute of Linguistics.
Saeed, John. 1999. Somali. Amsterdam/Philadelphia: John Benjamins.
Salminen, Tapani. 1997. Tundra Nenets inflection (Mémoires de la Société Finno-
Ougrienne 227). Helsinki: Suomolais-Ugrilainen Seura.
Sapir, J. David. 1965. A grammar of Diola-Fogny (West African Language
Monographs 3). Ibadan: Cambridge University Press in association with the
West African Languages Survey and the Institute of African Studies.
Sasse, Hans-Jürgen. 1976. Dasenech. In M. Lionel Bender (ed.), The non-Semitic
languages of Ethiopia. East Lansing: African Studies Center, Michigan State
University. 196–221.
Schenker, Alexander M. 1964. Polish declension. The Hague: Mouton.
268 References
Serzisko, Fritz. 1982. Numerus/Genus-Kongruenz und das Phänomen der

Polarität am Beispiel einiger ostkuschitischer Sprachen. In Hansjakob Seiler
and Franz Josef Stachowiak (eds.), Apprehension: das sprachliche Erfassen von
Gegenständen (vol. II). Tübingen: Narr. 179–200.
Sharma, D. D. 1988. A descriptive grammar of Kinnauri. Delhi: Mittal Publications.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Robert
M. W. Dixon (ed.), Grammatical categories in Australian languages.
Canberra: Australian Institute of Aboriginal Studies. 112–71.
Sims-Williams, Nicholas. 1982. The double system of nominal inflection in
Sogdian. Transactions of the Philogical Society 80. 67–76.
Skorik, Pëtr Ia. 1961–77. Grammatika cˇukotskogo iazyka (vols. I and II).
Leningrad: Izdatel0 stvo Akademii nauk SSSR.
Smith, Lawrence R. 1979. Labrador Inittut inverted number marking, exchange
rules and morphological markedness. Linguistics 17/1–2. 153–67.
Šołćina, Jana and Edward Wornar. 2002. Obersorbisch im Selbststudium: ein
Sprachkurs für Unerschrockene. Budyšin: LND (on-line edition available at
http://www.serbski-institut.de/).
Sologub, A. I. 1983. O sinkretizme form v sklonenii suščestvitel0 nyx ženskogo roda
edinstvennogo čisla po dialektnym dannym. In Ruben I. Avanesov (ed.),
Russkie narodnye govory. Moscow: Nauka. 82–8.
Soukka, Maria. 2000. A descriptive grammar of Noon: a Cangin language of
Senegal. Munich: Lincom.
Spencer, Andrew. 1991. Morphological theory. Oxford: Blackwell.
2000. Agreement morphology in Chukotkan. In Wolfgang U. Dressler, Oskar E.
Pfeiffer, Markus A. Pöchtrager and John R. Rennison (eds.), Morphological
analysis in comparison. Amsterdam/Philadelphia: John Benjamins. 191–222.
Sridhar, Shikaripur N. 1989. Kannada. London: Routledge.
Stevenson, R. C. 1969. Bagirmi grammar (Linguistic Monograph Series No. 3).
Khartoum: Sudan Research Unit, University of Khartoum.
Stone, Gerald. 1993. Sorbian. In Bernard Comrie and Greville G. Corbett (eds.),
The Slavonic languages. London: Routledge. 593–686.
Stump, Gregory T. 1993. On rules of referral. Language 69. 449–79.
2001. Inflectional morphology. Cambridge: Cambridge University Press.
Sunik, O. P. 1997. Udègejskij jazyk. In Jazyki mira: mongol0 skie jazyki,
tunguso-man0 čzˇurskie jazyki, japonskij jazyk, korejskij jazyk. Moscow:
Indrik. 236–48.
Szemerényi, Oswald. 1989. Einführung in die vergleichende Sprachwissenchaft.
Darmstadt: Wissenshaftliche Buchgesellschaft.
Taylor, Charles V. 1985. Nkore-Kiga. London: Croom Helm.
Terrill, Angela. 2003. A grammar of Lavukaleve. Berlin: Mouton de Gruyter.
Tesnière, Lucien. 1925. Les formes du duel en slove`ne. Paris: Champion.
Timberlake, Alan. 1975. Hierarchies in the genitive of negation, Slavic and East
European Journal 19. 123–38.
Todaeva, Buljaš X. 1997. Baoan0 skij jazyk. In Jazyki mira: mongol0 skie jazyki,
tunguso-man0 čzˇurskie jazyki, japonskij jazyk, korejskij jazyk. Moscow: Indrik.
29–36.
References 269
Tokarski, J. 1993. Schematyczny indeks a tergo polskich form wyrazowych (revised

and edited by Zygmunt Saloni). Warsaw: Wydawnictwo Naukowe PWN.
Toporišič, Jože. 1976. Slovenska slovnica. Maribor: Obzorja.
Tosco, Mauro. 2001. The Dhaasanac language. Köln: Rüdiger Köppe.
Trask, R. L. 1997. A student’s dictionary of language and linguistics. London:
Arnold.
Tucker, Archibald N. and Margaret A. Bryan. 1966. Linguistic analysis: the non-
Bantu languages of North-Eastern Africa. London: Oxford University Press.
Turton, David. 1981. Le Mun (Mursi). In G. Manessy (ed.), Les langues de
l’Afrique subsaharienne (Les langues dans le monde ancien et moderne, vol.
1.1). Paris. Editions du CNRS. 335–49.
Upadhyaya, U. Padmanabha. 1976. A comparative study of Kannada dialects:
Bellary, Gulbarga, Kumta, and Nanjangud dialects. Mysore: Prasaranga,
University of Mysore.
Uspensky, Boris A. 1965. Strukturnaja tipologija jazykov. Moskva: Nauka.
Verner G. K. [Werner, Heinrich]. 1999. Ketskij jazyk. In Jazyki mira:
paleoaziatskie jazyki. Moscow: Indrik. 177–87.
Viitso, Tiit-Rein. 1998. Fennic. In Daniel Abondolo (ed.), The Uralic languages.
Voorhoeve, C. L. 1965. The Flamingo Bay dialect of the Asmat language.
’s-Gravenhage: Martinus Nijhoff.
1975. Central and western Trans-New Guinea phylum languages. In Stephan
A. Wurm (ed.), Papuan languages and the New Guinea linguistic scene (Pacific
Linguistics C38). Canberra: Australian National University. 345–459.
Wackernagel, Jacob. 1924. Vorlesungen über Syntax (vol. I). Basel: Birkhäuser.
Watkins, Calvert. 1962. Indo-European origins of the Celtic verb. Dublin: Institute
of Advanced Studies.
Watkins, Laurel J. 1984. A grammar of Kiowa. Lincoln: University of Nebraska Press.
Watters, David E. 2003. A grammar of Kham. Cambridge: Cambridge University
Press.
Wegera, Klaus Peter. 2000. Morphologie des Frühneuhochdeutschen. In Werner
Besch, Anne Betten, Oskar Reichmann and Stefan Sonderegger (eds.),
Sprachgeschichte: ein Handbuch zur Geschichte der deutschen Sprache und
ihrer Erforschung (vol. 1.2). Berlin: Mouton de Gruyter. 1313–22.
Werner, Roland. 1987. Grammatik des Nobiin (Nilnubisch). Hamburg: Helmut
Buske.
West, Dorothy. 1973. Wojokeso sentence, paragraph and discourse analysis (Pacific
Linguistics B28). Canberra: Australian National University.
Westermann, Diedrich. 1930. A study of the Ewe language. London: Oxford
University Press.
Whitney, William Dwight. 1889. Sanskrit grammar. Cambridge: Harvard
University Press.
Wiese, Bernd. 1996. Iconicity and syncretism. In Robin Sackmann (ed.),
Theoretical linguistics and grammatical description: papers in honour of Hans-
Heinrich Lieb on the occasion of his 60th birthday. Amsterdam/Philadelphia:
John Benjamins. 323–44.
270 References
2003. Zur lateinischen Nominalflexion: die Form-Funktion-Beziehung. Ms.,

Institut für deutsche Sprache, Mannheim.
Williams, Edwin. 1981. On the notions ‘lexically related’ and ‘head of a word’.
Linguistic Inquiry 12/2. 245–74.
Williams, Edwin. 1994. Remarks on lexical knowledge. Lingua 92. 7–34.
Wilson, Darryl. 1974. Suena grammar (Workpapers in Papua New Guinea languages
8). Ukarumpa, Papua New Guinea: Summer Institute of Linguistics.
Windisch, Rudolf. 1973. Genusprobleme im Romanischen: das Neutrum im
Rumänischen. Tübingen: Präzis.
Wright, Joseph. 1930. Grammar of the Gothic language. Oxford: Oxford University
Press (reprint of 1910 edition).
Wunderlich, Dieter. 2001a. A correspondence-theoretic analysis of Dalabon
transitive paradigms. In Geert Booij and Jaap van Marle (eds.), Yearbook of
morphology 2000 Dordrecht: Kluwer. 233–52.
2001b. How gaps and substitutions can become optimal: the pronominal affix
paradigms of Yimas. Transactions of the Philological Society 99. 315–66.
2004. Is there any need for the concept of directional syncretism? In L. Gunkel,
G. Müller and G. Zifonun (eds.), Explorations in nominal inflection. Berlin:
Mouton de Gruyter. 373–395.
Wurm, Stephan A. 1975. The Trans-Fly (sub phylum level) stock. In Stephan
A. Wurm (ed.), Papuan languages and the New Guinea linguistic scene
(Pacific Linguistics C38). Canberra: Australian National University. 323–44.
Wurzel, Wolfgang U. 1987. System-dependent morphological naturalness in
inflection. In Wolfgang U. Dressler (ed.), Leitmotifs in natural morphology.
Xajdakov, S. M. 1980. Principy immennoj klassifikacii v dagestanskix jazykax.
Moscow: Nauka.
2001. Lakskij jazyk. In M. E. Alekseev (ed.), Jazyki mira: kavkazskie jazyki.
Moscow: Academia. 347–57.
Zaliznjak, Andrej A. 1973 [2002]. O ponimanii termina ‘padež’ v lingvističeskix
opisanijax. In Russkoe imennoe slovoizmenenie. Moscow: Jazyki slavjanskoj
kul0 tury. 613–47 (originally in Andrej A. Zaliznjak (ed.), Problemy
grammaticˇeskogo modelirovanija. Moscow: Nauka. 53–87).
Zasorina, L. N. 1977. Častotnyi slovar 0 russkogo jazyka. Moscow: Russkij jazyk.
Žirkov, L. I. 1955. Lakskij jazyk. Moscow: Izdatel0 stvo Akademii Nauk SSSR.
Žukova, Alevtina N. 1972. Grammatika korjakskogo iazyka: fonetika, morfologija.
Leningrad: Nauka.
Zwicky, Arnold. 1985. How to describe inflection. In Proceedings of the eleventh
annual meeting of the Berkeley Linguistics Society. 372–86.
1991. Systematic versus accidental phonological identity. In Frans Plank (ed.),
2000. Describing syncretism: rules of referral after fifteen years. Presentation at
the annual meeting of the Berkeley Linguistics Society (26), University of
California, Berkeley.
Author Index
Abbot, Miriam 18 Brockelmann, Carl 143

Ackema, Peter 65 Bromley, H. Myron 67, 229
Aikhenvald, Alexandra Y. 111 Brousseau, Anne-Marie 60
Allhiser de Engel, Mary 25, 26 Brown, Dunstan 10, 11, 37, 78, 95, 111,
Anttila, Raimo 51, 71 113, 114, 122, 131, 144, 174, 175, 186,
Arensen, Jon 65, 223, 230 187, 194, 196, 197, 205, 209, 226, 231,
Arkadiev, Pëtr M. 37 236, 242, 248
Armbruster, Charles Herbert 23, 168, 229 Browne, Gerald M. 74, 168
Aronoff, Mark 103, 184 Bryan, Margaret A. 233
Aronson, Howard I. 55, 56 Buck, Carl D. 6, 7, 23
Ashton, E. O. 231 Burrow, Thomas 154
Austin, Peter 51, 54, 144 Butt, John 231
Bybee, Joan 152
Baerman, Matthew 10, 11, 37, 59, 62, 95,
110, 111, 114, 122, 226, 231 Cahill, Lynn 174
Bammesberger, Alfred 226, 229 Cairns, Charles E. 22–3
Bátori, István 37 Calabrese, Andrea 39
Bavin, Edith L. 37 Cardona, George 70
Bechhaus-Gerst, Marianne 75 Carmack, Stanford 37
Beeler, Madison S. 79 Carstairs, Andrew see Carstairs-McCarthy,
Béjar, Susana 133 Andrew
Bender, Lionel M. 230 Carstairs-McCarthy, Andrew 37, 56, 111,
Benjamin, Carmen 227, 231 126, 151, 161–3, 168, 182, 214
Bentley, Delia 71 Carter, Hazel 66, 230
Berger, Hermann 224, 228 Casad, Eugene H. 224
Berman, Ruth 229 Chafe, Wallace 84
Bhattacharya, Sudhibhushan 154 Chapman, Shirley 223
Bierwisch, Manfred 39, 126 Charney, Jean Ormsbee 225
Bindseil, Heinrich 4, 7 Childs, George Tucker 88
Blake, Barry J. 55 Chittleborough Martin 26
Bleiching, Doris 174 Chvany, Catherine V. 37, 130, 150, 151
Blevins, James P. 15, 37 Clairis, Christos 60
Bobaljik, Jonathan 63, 138, Coleman, Robert 5, 72, 208
160, 161 Comrie, Bernard 7, 11, 42, 78, 181, 198
Boeder, Winfried 37 Corbett, Greville G. 10, 11, 29, 37, 78, 82,
Booij, Geert 120, 123 86, 87, 92, 95, 104, 109, 111, 114, 122,
Bowe, Heather 223 123, 138, 144, 147, 174, 175, 186, 187,
Boyeldieu, Pascal 84 194, 196, 197, 205, 214, 216, 236, 242,
Breedveld, J. O. 87 248
Breen, J.Gavan 44 Cowan, H. K. J. 231, 235
Bright, William 229, 234 Crowley, Terry 98
271
272 Author Index
Curnow, Timothy J. 228 Greenberg, Joseph H. 27, 29, 83, 88, 89, 114
Curtius, Georg 4 Gregores, Emma 76, 233
Cysouw, Michael 37, 60, 61, 68, 92 Grimm, Jacob 3
Grjunberg, A. L. 107
Dalrymple, Mary 131 Grosse, Siegfried 72
Davies, John 21, 230 Gruzdeva, Elena Ju. 230
de Angulo, Jaime 101 Güldemann, Tom 89
Dedrick, John M. 224 Gvozdanović, Jadranka 37
Dench, Alan C. 44, 226 Gwynn, John P. L. 42
Derbyshire, Desmond 223, 229
Deza Galindo, Juan Francisco 22, 228 Hagman, Roy S. 230
Dickens, Patrick 89, 90 Hall, Daniel Currie 133
Dimmendaal, Gerrit J. 81, 84 Halle, Morris 39
Dixon, Robert M. W. 111, 202, 223, 224 Hamel, Patricia J. 97
Donaldson, Tamsin 226 Hansson, Gunnar O. 107
Drabbe, Peter 229, 230, 234 Harbour, Daniel 37
Drexel, Guido 174 Harley, Heidi 60, 61, 64
Dryer, Matthew 7, 11 Harvey, Mark 91, 128
Dutton, Tom 64 Haspelmath, Martin 7, 11
Haugen, Einar 70
Edel 0 man, D. I. 107 Haviland, John 43
Elliot, Eric 49 Hayward, Richard 107
Endzelins, Janis 100 Healey, Alan 26
Engel, Ralph 25, 26 Heath, Jeffrey 37, 76, 187, 197, 230, 234, 235
Evans, Nicholas 11, 37, 78, 93, 144, 175, Heine, Bernd 81, 82, 108, 109
186, 187, 188, 194, 196, 197, 226, 233, Hekking, Ewald 231
236, 242 Helimski, Eugene see Xelimskij, Evgenij A.
Evans, Roger 174, 177, 178, 179 Helmbrecht, Johannes 37, 64
Eythórsson, Thórhallur 71 Henderson, James 93–4
Ezard, Bryan 96, 97 Herbermann, Charles G. 3
Hetzron, Robert 104
Feldstein, Ronald 142 Hewitt, Brian G. 225, 233
Fennell, Trevor G. 16 Hewson, John 37
Feoktistov, Aleksandr P. 54 Hippisley, Andrew 175
Finkel, Raphael 47, 173 Hjelmslev, Louis 4, 37, 48, 114
Fischer, Wolfdietrich 143 Hoff, B. J. 67, 228, 235
Foley, William A. 20, 60, 110, 231, 233 Holton, David 225
Fortescue, Michael 224, 233 Horton, A. E. 230
Frachtenberg, Leo J. 105 Hualde, José Ignacio 223
Fradkin, Robert A. 37 Hudson, Richard A. 228
Frank, Paul 156, 229 Huntley, David 214
Frank, Wright Jay14 Hutchison, John P. 234
Franklin, Karl James 229
Fraser, Norman 11, 138, 147, 174, 175, Innes, Gordon 85
205, 214, 216, 248 Isoroembo, Ambrose 26
Fromm, Hans 47, 50, 225 Ivanov, Vjačeslav V. 39
Gage, William W. 60 Jackendoff, Ray S. 155

Gazdar, Gerald 174, 177, Jacobson, Steven A. 14, 48
178, 179 Jakobi, Angelika 83, 98, 99
Gibbon, Dafydd 174 Jakobson, Roman O. 4, 39, 52, 210
Gil, David 7, 11, 136 Johnston, Jason 37, 127, 130, 150, 151
Glasgow, Kathleen 68
Goddard, Cliff 37, 42, 44 Kachru, Braj B. 24
Gordon, Lynn 235 Kaplan, Ronald M. 131
Graudiņa, A. 99 Kariņš, A. Krišjanis 99, 100
Author Index 273
Karlson, Fred 62 Najlis, Elena 18, 228

Kathol, Andreas 49 Neeleman, Ad 65
Kennedy, Benjamin Hall 214 Neidle, Carol J. 39
Kettunen, Lauri 71 Neue, Friedrich 139, 140
Key, Harold H. 228 Newmark, Leonard 82
Khubchandani, Lachman M. 21 Nichols, Johanna 53, 226
Kibrik, Aleksandr E. 37, 43, 76 Nikolaeva, Irina 25, 68, 231
Kienast, Burkhart 143 Noonan, Michael 230
Kiparsky, Paul 37, 39, 45, 132 Nordlinger, Rachel 223
Klenin, Emily 49, 226 Noreen, Adolf 66
Koneski, Blaže 161, 182, 214 Noyer, Rolf 37, 62, 63, 64, 76, 160, 162–3,
Kotyczka, Jozef 30 166–7
Krishnamurti, Bhadriraju 42 Nussbaum, Loren V. 60
Kuryowicz, Jerzy 70, 71
Kutsch Lojenga, Constance 62, 230 Oates, Lynette F. 229
Oates, William J. 229
Lakämper, Renate 37 Orešnik, Janez 148, 214
Lamberti, Marcello 69 Ortiz de Urbina, Jon 223
Lefebvre, Claire 60 Osborne, Charles R. 231
Leumann, Manu 6 Owens, Jonathan 223, 229
Liddell, Henry G. 3 Oxotina, Natalija V. 231
Lindstrom, Lamont 62
Lipkind, William 60 Pace, Edward A. 3
Lorimer, David L. R. 43 Pallen, Condé B. 3
Luraghi, Silvia 5, 38 Perlmutter, David M. 148, 214
Lynch, John 62, 72, 73 Philippaki-Warburton, Irene 225
Pike, Kenneth 8–9, 104
Mackridge, Peter 225 Pitman, Donald 224
Magometbekova, Z. M. 82 Plank, Frans 150, 151
Maiden, Martin 108 Plungian, Vladimir 27, 29, 83, 88, 89,
Makoondekwa, João 66, 230 114
Malécot, André 56 Pokorny, Julius 16
Maslova, Elena 224, 231 Popjes, Jack 228
Mateo Alvarez, José 25, 26 Popjes, Jo 228
Mathiassen, Terje 15, 42, 99, 226, 230 Pott, August 4, 7
Matras, Yaron 73 Priestley, T. M. S. 176, 212
Mayerthaler, Willi 170
McCarthy, John J. 22 Reesink, Ger P. 60, 224, 231
McCreight, Katherine 37, 150, 151 Reh, Mechthild 226
McGregor, Ronald S. 225, 229 Ricca, Davide 206
McGregor, William 233 Rice, Keren 62
McGuckin, Catherine 96 Ringe, Donald 40
Meillet, Paul 168 Ritter, Elizabeth 60, 61, 64
Meiser, Gerhard 4, 5 Roberts, John R. 228
Merlan, Francesca 223, 233 Robins, R. H. 42, 48, 224, 234
Miller, Raymond 39 Romero-Figueroa, Andres 223
Minassian, Martiros 47, 224 Rothe, Wolfgang 85
Mitchell, Terence F. 228 Rumsey, Alan 233
Moravcsik, Edith 22 Rupp, James E. 230
Morgenstierne, Georg 15–16
Mous, Maarten 229, 234 Saeed, John 19, 104
Müller, Gereon 210 Salminen, Tapani 48, 226, 230
Munro, Pamela 80 Sapir, J. David. 229
Murane, Elizabeth 228 Sasse, Hans-Jürgen 107
Muysken, Pieter 66 Schenker, Alexander 39
Myachina, Ekaterina N. 231 Scott, Robert 3
274 Author Index
Serzisko, Fritz 108, 110, 132 Vaillant, André 168

Shahan, Thomas J. 3 van Driem, George 120
Sharma, D. D. 62 Varre, Daniel 60, 100
Shen, Lei 47, 173 Verner G. K. see Werner, Heinrich
Shopen, Tim 37 Viitso, Tiit-Rein 71
Silverstein, Michael 44 Voorhoeve, C. L. 60, 235
Sims-Williams, Nicholas 53 Vossen, Rainer 89
Skorik, Pëtr Ia. 224, 228
Smith, Lawrence R 110 Wackernagel, Jacob 4
Šoćina, Jana 14 Wagener, C. 139
Sologub, A. I. 167 Watkins, Calvert. 63
Soukka, Maria 127 Watkins, Laurel J. 94, 109, 110, 230
Spencer, Andrew 2, 37, 78 Watters, David E. 79
Sridhar, Shikaripur N. 69, 229 Wegera, Klaus Peter 72
Stevenson, R. C. 228 Werner, Heinrich 229
Stone, Gerald 113 Werner, Roland 66
Stump, Gregory T. 8, 47, 63, 114, 131, West, Dorothy 105
134, 136, 138, 142, 148, 160, 161, Westermann, Diedrich 229
163–6, 173, 182 Whitney, William Dwight 47
Suárez, Jorge A. 76, 233 Wiese, Bernd 39
Sunik, O. P. 68 Williams, Edwin 14, 150, 151
Suthar, Babu K. 70 Wilson, Darryl 48, 67, 223, 231
Szemerényi, Oswald 5, 72, 208 Windisch, Rudolf 86
Wirth, Jessica 22
Taylor, Charles V. 230, 234 Wornar, Edward 14
Terrill, Angela 230 Wright, Joseph 22, 92
Tesnière, Lucien 165 Wunderlich, Dieter 37, 45, 142, 145, 147,
Thesayi, Suresh 47, 173 148, 149, 150, 194
Timberlake, Alan 175 Wurm, Stephan A. 102
Todaeva, Buljaš X. 51 Wurzel, Wolfgang U. 170
Tokarski, J. 30 Wynne, John J. 3
Tolskaya, Maria 25, 68
Toporišič, Jože 165 Xajdakov, S. M. 25, 87, 230
Tosco, Mauro 106, 236, 237, 238, Xelimskij, Evgenij A. 20, 231
240, 241
Trask, R. Lawrence 2 Zaliznjak, Andrej A. 15, 37, 42
Tucker, Archibald. N. 233 Zasorina, L. N. 248
Turton, David 8, 65 Žirkov, L. I. 50, 226
Žukova, Alevtina N. 50, 77, 78, 95
Upadhyaya, U. Padmanabha 72 Zwicky, Arnold 96, 131, 134, 135, 136, 144,
Uspensky, Boris A. 114 145, 163–4
Language index
The index includes languages discussed in the text or appendices. It does

not feature languages which appear listed in the appendices but are not
overtly discussed. Sources for these latter languages may be found in
Baerman and Brown (2005a, b).
Abipon (Guaicuruan) 17–18, 228 Callahuaya (Quechua based) 159, 160

Albanian (Indo-European, Albanian) 82–3, Camus (Nilo-Saharan, Nilotic) 81, 108,
86 109, 110, 111
Aleut (Eskimo-Aleut) 59 Canela-Kraho (Ge-Kaingang) 228
Amele (Trans-New Guinea, Madang) 59, 60, Carib (Carib) 67, 159, 160, 228, 235
61, 228 Cayuvava (Cayuvava) 228
Andi (Nakh-Dagestanian, Dagestanian, Chichimeco (Oto-Manguean, Otopamean)
Avaro-Andi-Dido) 82, 86–7 100, 103
Anejom (Austronesian, Oceanic) 72, 73, 169 Chitimacha (Isolate) 59
Arabic (Afroasiatic, Semitic) 122, 142–3, Chukchi (Chukotko-Kamchatkan) 224, 228
158, 228 Comanche (Uto-Aztecan, Numic) 224
Arabic, Classical see Arabic
Arabic, Egyptian see Arabic Daga (Trans-New Guinea, Central and
Araona (Tacanan) 52, 224 Southeast New Guinea) 228
Arapesh (Torricelli) 174 Dalabon (Gunwinyguan) 78, 171, 172,
Arbore (Afroasiatic, Cushitic) 107 183, 186–204, 217, 218, 242–7
Armenian (Indo-European, Armenian) 47, Dani (Trans-New Guinea, Dani-Kwerba)
50, 52, 224 67, 159, 229
Asmat (Trans-New Guinea, Central and Danish (Indo-European, Germanic) 74
South New Guinea) 235 Dhasanaac (Afroasiatic, Cushitic) 106, 107,
Atakapa (Isolate) 59 167, 169–70, 171, 174, 183–6, 217, 218,
Awa (Trans-New Guinea, East New Guinea 236–41
Highlands) 60 Diola-Fogny (Niger-Congo, Northern
Awa Pit (Paezan, Barbacoan) 228 Atlantic) 229
Aymara (Aymara) 21–2, 62, 228 Diyari (Pama-Nyungan) 54, 143–4
Dogon (Niger-Congo, Dogon) 26, 59
Bagirmi (Nilo-Saharan, Bongo-Bagirmi) 228 Domaaki (Indo-European, Dardic) 42, 43,
Barbareño (Chumash) 79 48
Basque (Isolate) 48, 223 Dulong see Trung
Beja (Afroasiatic, Beja) 228 Dutch (Indo-European, Germanic) 65, 70,
Bininj Gunwok (Gunwinyguan) 92, 93, 175, 159, 160
233 Dyirbal (Pama-Nyungan) 155
Bonan (Altaic, Mongolian) 51, 52,
136–8, 158 Eastern Armenian see Armenian
Boumaa Fijian (Austronesian, Oceanic) 223 Egyptian Arabic see Arabic
Burarra (Burarran) 59, 62, 68, 159 Ekari (Trans-New Guinea, Wissel
Burushaski (Isolate) 52, 122, 224, 228 Lakes-Kemandoga) 229
275
276 Language Index
English (Indo-European, Germanic) 3, 26, Kalam (Trans-New Guinea, East New

95–6, 100, 101, 103, 180, 223, 229 Guinea Highlands) 60
Erzja Mordvin (Uralic, Finnic) 54, 158 Kamoro (Trans-New Guinea, Central and
Estonian (Uralic, Finnic) 71 South New Guinea 60
Ewe (Niger-Congo, Kwa) 229 Kannada (Dravidian) 69, 72, 74, 159, 229
Kanuri (Nilo-Saharan, Saharan) 234
Finnish (Uralic, Finnic) 47, 48, 49, 50, 51, 62, Kapau (Trans-New Guinea, Angan) 59, 229
132, 211, 225 Karata (Nakh-Dagestanian, Dagestanian,
Fongbe (Niger-Congo, Kwa) 60 Avaro-Andi-Dido) 82–3
French (Indo-European, Italic) 62, 225, 229 Karok (Hokan, Karok) 229, 234
Fula (Niger-Congo, Northern Atlantic) 87–8 Kashmiri (Indo-European, Dardic) 23–4,
Fur (Nilo-Saharan, Fur) 83 122
Kawasqar (Qawesqar) 60
Gaagudju (Gaagudju) 90–1, 127, 128–9, 130, Kayardild (Tangic) 226
150, 171 Ket (Ket) 229
Gapapaiwa (Austronesian, Oceanic) 96, 103, Kewa (Trans-New Guinea, East New
160 Guinea Highlands) 59, 229
Georgian (Kartvelian) 53, 225, 233 Kinnauri (Sino-Tibetan, Western
German (Indo-European, Germanic) 8–9, Himalayish) 62
19, 53, 59, 71, 72, 74, 83, 168, 198, Kiowa (Aztec-Tanoan, Tanoan) 93–4, 109,
225, 229 110, 111, 230
Gooniyandi (Bunaban) 233 Kisi (Niger-Congo, Southern Atlantic) 83,
Gothic (Indo-European, Germanic) 21–2, 88–9
23, 74, 92 Kiwai (Trans-New Guinea, Trans-Fly) 59,
Grebo (Niger-Congo, Kru) 85 101, 102, 103
Greek (Indo-European, Greek) 3, 5, 38–9, Kobon (Trans-New Guinea, East New
40, 42, 47, 139, 140, 158, 225 Guinea Highlands) 21, 23, 59, 60, 61,
Greek, Ancient see Greek 230
Greek, Modern see Greek Koiari (Trans-New Guinea, Central and
Guambiano (Paezan, Barbacoan) 59 Southeast New Guinea) 59, 64, 159
Guarani (Tupi-Guarani) 76, 233 Kongo (Niger-Congo, Bantu) 66, 159, 160,
Gujarati (Indo-European, Indic) 70, 230
136–60 Korafe (Trans-New Guinea, Binanderean) 60
Guugu Yimidhirr (Pama-Nyungan) 42–3 Koryak (Chukotko-Kamchatkan) 49, 50, 94,
95, 114, 115, 119, 158, 160
Harar Oromo (Afroasiatic, Cushitic) 223, Krongo (Kordofanian) 226
229 Kunama (Nilo-Saharan, Kunama) 230
Hebrew (Afroasiatic, Semitic) 229 Kwamera (Austronesian, Oceanic) 61, 62
Hindi (Indo-European, Indic) 59, 225
Hixkaryana (Carib) 229 Laal (Unclassified) 84, 85
Hungarian (Uralic, Ugric) 156–7 Lak (Nakh-Dagestanian, Dagestanian,
Hunzib (Nakh-Dagestanian, Dagestanian, Lak-Dargwa) 24, 50, 52, 136, 158, 226,
Avaro-Andi-Dido) 59 230
Lango (Nilo-Saharan, Nilotic) 230
Icelandic, Old (Indo-European, Germanic) Latin (Indo-European, Italic) 4–5, 6, 23, 24,
66, 70, 71, 142, 159, 160, 167 46, 55, 86, 90, 111, 134, 135, 139–42,
Ika (Chibchan, Aruak) 156, 229 148, 151, 158, 214
Ingush (Nakh-Dagestanian, Nakh) 53, 59, Latvian (Indo-European, Baltic) 15–16, 40,
226 42, 46, 98–100, 103, 158, 226, 230
Iraqw (Afroasiatic, Cushitic) 229, 234 Lavukaleve (East Papuan, Solomons East
Irish (Indo-European, Celtic) 16, 226, 229 Papuan) 230
Irish, Old see Irish Lealao Chinantec (Oto-Manguean,
Italian (Indo-European, Italic) 206 Chinantecan) 230
Lezgian (Nakh-Dagestanian, Dagestanian,
Ju/ 0 hoan (Khoisan, Northern Khoisan) 83, Lezgic) 226
88–90, 89, 91 Limbu (Tibeto-Burman, Kiranti) 52, 120
Language Index 277
Livonian (Uralic, Finnic) 64, 71, 167 Pengo (Dravidian) 52, 153
Loniu (Austronesian, Oceanic) 97, 160 Phalura (Indo-European, Dardic) 15, 52
Luiseño (Aztec-Tanoan, Uto-Aztecan, Pitjantjatjara (Pama-Nyungan) 223
Takic) 49, 56 Polish (Indo-European, Slavonic) 30–1, 39,
Luvale (Niger-Congo, Bantu) 230 174
Prinmi (Sino-Tibetan, Qiangic) 59
Maba (Nilo-Saharan, Maban) 233 Pumi see Prinmi
Macedonian (Indo-European, Slavonic) 63,
65, 161 Quechua (Quechua) 66
Macushi (Carib) 18, 35
Mam (Mayan) 62 Rawang see Trung
Manchad (Tibeto-Burman, Kiranti) 59 Romani (Indo-European, Indic) 73
Mangarayi (Mangarayi) 48, 223, 233 Romanian (Indo-European, Italic) 83, 85–6,
Mansim (West Papuan) 60 142
Maricopa (Hokan, Yuman) 235 Russian (Indo-European, Slavonic) 1–3,
Marind (Trans New Guinea, Marind) 230, 9–10, 28–9, 30, 31–2, 34, 38–9, 40, 42,
234 47, 49, 50, 52, 53, 55, 57, 83, 112, 136,
Martuthunira (Pama-Nyungan) 51, 226 139, 140, 142, 145–9, 158, 166, 167, 172,
Mayali see Bininj Gun-Wok 173, 174, 175, 181, 183, 204, 214, 215,
Meyah (East Bird’s Head) 60 218, 227, 248–53
Mojave (Hokan, Yuman) 80
Murle (Nilo-Saharan, Surma) 64, 65, 159, Sami (Uralic, Finnic) 107–8, 110
223, 230 Sango (Niger-Congo, Adamawa-Ubangian)
Mursi (Nilo-Saharan, Surma) 8, 65 59, 60
Selkup (Uralic, Samoyedic) 20, 23
Nama (Khoisan, Central Khoisan) 230 Seneca (Macro-Siouan, Iroquoian) 84, 85
Nenets (Uralic, Samoyedic) 47, 48, 226, 230 Sentani (Trans-New Guinea, Sentani) 231,
Nez Perce (Penutian, Sahaptian) 59, 60, 61 235
Nganasan (Uralic, Samoyedic) 60 Serbo-Croat (Indo-European, Slavonic) 211
Ngiti (Nilo-Saharan, Balendru) 62, 230 Shinassha (Afroasiatic, Omotic) 69, 159
Ngiyambaa (Pama-Nyungan) 51, 52, 226 Sindhi (Indo-European, Indic) 21
Nimboran (Trans-New Guinea, Nimboran) Siuslaw (Penutian, Siuslaw) 105
162–3 Slave (Nadene, Athapaskan-Eyak) 60
Nivkh (Isolate) 59, 230 Slovene (Indo-European, Slavonic) 33–4, 39,
Nkore-Kiga (Niger-Congo, Bantu) 229, 230, 59, 92, 114, 116, 165–6, 175–7, 183, 212
234 Sogdian (Indo-European, Iranian) 53
Nobiin see Nubian Somali (Afroasiatic, Cushitic) 18–19, 104,
Noon (Niger-Congo, Northern Atlantic) 105, 110, 117, 122, 132
127–8, 129, 130, 150 Sorbian (Indo-European, Slavonic) 14, 112,
Norwegian (Indo-European, Germanic) 74 118, 119
Nubian (Nilo-Saharan, Nubian) 22, 23, 59, Spanish (Indo-European, Italic) 227, 231
65, 69, 74, 75, 159, 160, 168, 229 Suena (Trans-New Guinea, Binanderean)
Nubian, Dongolese see Nubian 48, 67, 159, 223, 231
Nubian, Old see Nubian Swahili (Niger-Congo, Bantu) 231
Nuer (Nilo-Saharan, Nilotic) 14–15, 17 Swedish (Indo-European, Germanic) 74
Nunggubuyu (Nunggubuyu) 62, 72, 166–7, Sye (Austronesian, Oceanic) 98
230, 234, 235
Tawala (Austronesian, Oceanic) 96–7, 103,
Old English see English 160
Old High German see German Telugu (Dravidian) 40, 42, 47, 85
Orokaiva (Trans New Guinea, Binanderean) Tetun (Austronesian, Timor) 59
26 Tiwi (Tiwi) 62, 231
Otomı́ (Oto-Manguean, Otomian) 231 Tokharian (Indo-European, Tokharian) 86
Trung (Tibeto-Burman, Nungish) 78, 79
Pashto (Indo-European, Iranian) 106, 107 Tsakhur (Nakh-Dagestanian, Dagestanian,
Paumari (Arauan) 223 Lezgic) 42, 43, 48, 117
278 Language Index
Tsez (Nakh-Dagestanian, Dagestanian, West Greenlandic (Eskimo-Aleut) 224, 233

Avaro-Andi-Dido) 181–2 Winnebago (Macro-Siouan, Siouan) 60
Turkana (Nilo-Saharan, Nilotic) 80, 81, 84 Wojokeso (Trans-New Guinea, Angan) 105
Turkish (Altaic, Turkic) 55 Wolof (Niger-Congo, Northern Atlantic) 60
Udihe (Altaic, Tungusic) 24, 62, 68, Yaqui (Aztec-Tanoan, Uto-Aztecan,

159, 160 Taracahitic) 48, 224
Ungarinjin (Wororan) 233 Yele (East Papuan, Solomons East Papuan)
Ura (Austronesian, Oceanic) 97–8, 103 93–4
Usan (Trans-New Guinea, Adelbert Range) Yidiny (Pama-Nyungan) 224
231 Yimas (Sepik-Ramu, Nor-Pondo) 20, 60,
231, 233
Wagaya (Pama-Nyungan), 43–4 Yukaghir (Yukaghir) 224, 231
Wambaya (West Barkly) 223 Yup’ik, Central Alaskan (Eskimo-Aleut)
Wambon (Trans-New Guinea, Central and 13–14, 17, 48, 93, 224
South New Guinea) 59 Yurok (Algic, Yurok) 40, 42, 48, 224, 234
Warao (Isolate) 223
Warekena (Maipurean) 60 Zoque (Mixe-Zoquean) 25, 59
Waskia (Trans-New Guinea, Madang) 59 Zulu (Niger-Congo, Bantu) 231
Subject Index
ablaut 98–9 morphosyntactic 4–7

affix suppression 96–100, 101 phonologically conditioned 4–7, 70, 161,
agglutinative morpology 55, 56 167, 170, 221
agreement 28, 32, 33, 82, 89, 92, 109, 112, conjugation see inflectional class
120, 122, 123, 124, 216 cumulative exponence 151, 155
analogy 7, 63, 70, 71, 72, 105, 107, 161–8,
220 DATR 177–80, 183, 236, 253
animacy 38, 45, 47, 50, 56, 57, 76, 81, 86, 89, declension see inflectional class
90, 109, 127, 145, 146, 147, 148, 181, defaults 26, 45, 61, 71, 74, 88, 89, 100, 109,
214–16, 217 127, 130, 132, 135, 139, 140, 142, 144,
Animacy Hierarchy 44–9, 56, 76, 81, 84, 146, 148, 159
92, 109 default inheritance 173, 174–5, 177–80
number as a component of 48–9 morphological default 26, 45, 46, 64, 70,
word class as a component of 47–8, 50 96, 143, 148, 183
aspect see TAM definiteness 38, 45, 54, 62, 84, 117, 142
attribute ordering 180, 182, 183, 184, 200, differential agent marking 147, 158
201–2, 203 differential object marking 147, 158
avoidance strategy 171, 175, 187, 197–8, directionality 13, 23, 24–5, 70–5
202, 217 directional effects 25, 57, 63–70, 74, 75,
134, 135, 145, 153, 155, 158–61, 164,
basic form see defaults, morphological 166–7, 183
blocking 135, 176 directional rules 133, 139, 143, 144, 145,
147, 148, 149, 150, 221
canonical inflection 27 directional syncretism 136–44, 145, 148,
cardinality of features see features, number 149, 150, 160, 163, 174
of values see also take-overs
case 1–2, 13–16, 20, 28, 35, 37, 38–57, 111, disjunction 81, 126, 154–5, 157, 176
112, 113, 114, 116, 117, 118, 119, 123,
124, 134, 136, 137, 143, 144, 145, 146, elsewhere condition 135, 150
151, 152, 158, 169, 220, 223–7 elsewhere form 127, 129, 130, 157, 182, 183
core cases 40, 42, 47, 48, 49, 52, 53, 56, 124, see also Panini’s principle
144, 158, 223–7 exceptionality 172, 176, 177
peripheral cases 40, 49, 51, 52, 56,
224–7 Feature Ranking Principle 114, 158–60,
semantic cases 52 164–6
syntactic cases 52 feature structure 13, 103, 126, 127, 132, 133,
case systems 151, 161–2, 169, 170, 171, 181, 210–11
ergative-absolutive 40, 42, 47, 49 cross-classifying 126, 127, 130–1, 132, 133
nominative-accusative 40, 42, 49 flat 126, 127–9, 130, 155, 160, 162, 163,
change, diachronic 52, 54, 55, 56, 57, 166, 171
60, 61, 62, 63, 70–5, 82, 89, 91, 92, 95, hierarchical 17, 35, 61, 64, 75, 126,
98, 105, 124, 125, 145, 152, 168–9, 127–8, 129–30, 132, 133,
169–70, 216 150, 160, 171
279
280 Subject Index
features markedness 22–3, 46, 49, 56, 57, 63, 64, 94,
as a context for syncretism 19–23, 20, 29, 110, 132, 133, 158–60, 162, 163, 164–6,
30, 31, 32, 33, 34, 35, 38, 59, 61, 90, 91, 168
111, 112, 114, 116, 117, 118–19, 120, see also unmarkedness
121, 122, 136, 151 mood see TAM
contextual 124 morphological class 17–19, 204
feature dependency 111, 113 morphomes see indexes, morphomic
feature ordering 35, 113, 114, 118, 119, mother-in-law language 155, 175
151, 172, 173, 174, 182
inherent 109, 112, 120, 123 natural classes 103, 108, 111, 126–31, 136,
nominal 113–19 146, 171
non-autonomous values 15, 43, 54, 83 see also unnatural classes
number of values 123 negation 22–3, 74, 120, 122
verbal 119–22 Network Morphology 11, 12, 113, 138, 171,
frequency 37, 75, 110, 117, 152 172–80, 188
neutralization 13, 27, 28, 29, 30, 31, 32, 33,
gender 21, 28–9, 31, 32, 33, 34, 35, 37, 34, 35, 36, 76, 83, 92, 114, 122, 173, 181,
53, 81–91, 106, 110, 111, 112, 182, 183, 194–6
113, 114, 116, 117, 118, 119, number 21, 23, 25, 26, 31, 32, 35, 37, 38, 45,
120, 121, 122, 123, 124, 127, 47, 48, 49, 59, 60, 61, 68, 69, 72, 79,
128, 129, 130, 132, 135, 140, 82–90, 91, 96, 97, 98, 100, 101, 107, 108,
151, 180–1, 216, 220 109, 110, 111, 112, 113, 114, 116, 117,
convergent systems 82–3, 84, 127 118, 119, 120, 121, 122, 123, 124, 132,
crossed systems 82–3 151, 152, 161–2, 163, 165, 169, 181, 188,
semantics of 81, 84–5, 86–7, 88, 199, 200–1, 216, 220
89, 91 Number Hierarchy 95
glossing conventions 11–12
object 40, 42, 56, 57, 75, 76, 77, 78, 79–80, 81,
harmonicity 188 91, 92, 120, 128
homophony 5, 151, 152 Optimality Theory 45, 132, 137, 145–50
accidental 9–10, 24, 61, 81, 141, 142, 148,
160–3, 166, 167, 168, 206 Panini’s principle 135, 138, 190, 192
avoidance of 70, 157, 168 see also elsewhere condition
Paradigm Function Morphology 138, 171
impoverishment 63, 160–3, 166 person 11, 17–18, 19–22, 25, 26, 31, 32, 37,
inclusive–exclusive distinction 57, 61, 62, 45, 48, 57, 61, 62, 63, 75–81, 92, 93, 94,
110, 111 95, 96, 97, 98, 100, 101, 104, 105, 106,
indexes, morphomic 103, 135, 174, 182, 184, 107, 109, 110, 111, 117, 120, 121, 122,
221 124, 134, 136, 151, 156, 159–60, 169,
individuation 44, 50, 109 197, 199, 200–1, 204, 220, 228, 235
inferential-realizational theories 171, 172, polarity 16, 17, 35, 103–11, 122, 132–3, 151,
175–7 184
inflectional classes 38, 47, 48, 49, 59, 91, 93, prepositions, case government of 55
97, 98, 99, 100, 108, 114, 115, 119, 134, prosody 98
135, 139, 140, 141, 142, 145, 148, 155,
204–5, 207–11, 214 regularity 23–4, 172, 183, 207, 212, 218
inflection–derivation distinction 3, 8 relevance hierarchy 152–3
inverse 76, 78, 81, 94, 109 rules
irregularity 176 rule ordering 138, 139, 140, 141, 145, 148
rules of exponence 135, 137
KATR 173 rules of referral 134, 135, 137, 160, 161,
164, 165, 167, 171, 172, 173–4, 182,
Lexical-Functional Grammar 131 183, 187, 188, 197–9, 201, 202, 203,
lexical knowledge representation 174 204, 210, 218, 221, 222
lexical semantics 3, 154–5 symmetrical rules 133–50
Subject Index 281
semantics 38–9, 54, 56, 57, 60, 61, 62, 63, 82, morphological 105–8, 111, 157, 166, 167,
89, 91, 92, 95, 98, 105, 124, 125, 152, 216 168, 184, 186, 198, 217
semantic naturalness 61, 171, 180–2 semantic 108–11
semantic values 55, 56
see also systematicity, semantic take-overs 152–3, 155, 156, 162, 182
sound change; see change, phonologically see also directionality
conditioned TAM (tense-aspect-mood) 26, 37, 57, 59,
split ergativity 42, 43, 44, 45, 48 61, 62, 63, 64, 65, 66, 67, 70, 72, 73, 91,
stems 53, 97, 98–100, 106, 107, 139, 148 93, 95, 110, 111, 120–2, 124, 156, 160,
bare stems see defaults, morphological 164
stress see prosody tense see TAM
subject 40, 43, 57, 65, 75, 76–8, 79, 80, 81, 91, tone see prosody
92, 104, 105, 106, 109, 110, 120, 128 transitivity 187, 194, 197, 204
suppletion 176
Surrey Syncretisms Database 10, 111, underspecification 35, 59, 61, 62, 63, 64, 81,
114–22, 124, 219 129, 146, 147, 155, 157, 158, 163, 164,
syncretism 171, 172, 173, 180–2, 183, 184, 187, 188,
block 8 191–4, 196, 198, 199, 200, 201, 202, 203,
canonical 27, 33–5, 38 204, 211, 218, 221
complete 59, 62 uninflectedness 13, 27, 30–3, 32, 33, 34, 35,
contrary 15–17, 35, 150, 151 36, 173, 181, 182, 183
convergent 145 unmarkedness 23, 25–7, 44, 45, 56, 63, 128,
history of 3–4 136, 145, 146, 155–6, 158, 161, 169
lexically determined 206–7 see also markedness
morphologically determined 207–12 unnatural classes 103, 108, 131–2
nested 14, 17, 35, 129, 150 see also natural classes
simple 13, 17, 35, 83
syntactically determined 213–16 verbs, two-place 75–81, 197
partial 59, 61, 62, 63, 64, 81, 129, 146, 147, voice 120
157, 158, 163, 164, 171, 172, 173, 182,
183, 184, 187, 188, 194, 196, 198, 199, Watkins’ Law 63
200, 201, 202, 203, 211, 218, 221 word class 38, 45, 47, 48, 211
syntactic class see word class World Atlas of Language Structures 7, 11, 12,
Systematic Homonymy Claim 152–4, 182 37, 57, 75, 76, 79, 219
systematicity 9–10, 22, 24, 139, 141, 148,
153, 157, 160–1, 166–7, 168–9, 182–3, zero ending see defaults, morphological
217 zero morphology see defaults, morphological

The Syntax-Morphology Interface

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Syntax-Morphology Interface

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

THE SYNTAX–MORPHOLOGY INTERFACE

Syncretism – where a single form serves two or more morphosyntactic

M A T T H E W B A E R M A N is Research Fellow at the University of Surrey. He

D U N S T A N B R O W N is Senior Lecturer in Linguistics at the University of

G R E V I L L E G . C O R B E T T is Distinguished Professor of Linguistics at the

Earlier issues not listed are also available

The Syntax–Morphology Interface

Cambridge University Press

© Matthew Baerman, Dunstan Brown, and Greville G. Corbett 2005

This publication is in copyright. Subject to statutory exception and to the provision of

First published in print format 2005

- ---- eBook (EBL)

- ---- hardback

3 Cross-linguistic typology of features 37

3.8.4.1 Verbal paradigms 120

4 Formal representation 126

5 Formal framework and case studies 171

5.3.6 Referral of 1 > 2sG to 3 > 2sG 203

Appendix 1: Case syncretism in the World Atlas of Language

This book has an interesting history of collaboration. It began life in

and Technology, Lancaster University, University of Catania, Second

- used to form compound names for morphosyntactic

We might expect that a language’s sentence-structure and word-structure

(4) Na stole ležit pis 0 mo

inflectional distinctions which may be treated similarly. We shall ask when

1.1 History of the notion

fashionable in other realms as well. Pott (1836) is generally credited with

1.2 Delimiting the notion of syncretism

(5) Nominative/accusative singular in Vulgar Latin (Coleman 1976: 50–4)

Classical Latin > Vulgar Latin

first declension NOM SG luna luna

Merger resulting from morphosyntactic change is illustrated by the devel-

(6) Dative/locative singular in Ancient Greek (Buck 1933: 180–5)

Proto-Indo-European > Greek

and consonant-stem nouns, so their merger cannot be attributed to phono-

Thus, the Vulgar Latin example represents a superficial inflectional homo-

second declension (o-stems) ABL SG -o

is evidence from early inscriptions that monophthongization took place

1.3 Scope of the investigation

1.3.2 Selection of forms

individual components of inflected words, e.g. by comparing prefixes to

indefinite aspect definite aspect

1SG ka- baagio wa- ka- baaka

(9) German sein ‘to be’

Within these forms, however, eight distinct patterns of identity can be

(note that orthographic d in (10) is indistinguishable from t in word-final

1.4 Accidental versus systematic homophony

(11) a. stem-stress ‘place’ b. end-stress ‘wine’

orthographic phonetic orthographic phonetic

NOM/ACC SG mesto :mje.st e vino vji.:no

1.5 Using this book

a more limited phenomenon over a much broader range of languages,

1.5.2 Glossing conventions

1.5.3 Structure of the book

We shall map out the range of data to be considered. We include problems

2.1 Syncretic paradigms

(1) Simple syncretism in Central Alaskan Yup’ik (Jacobson 1995: 469–71)

singular dual plural

ABS nuna nunak nunat ‘land’

If simple syncretism is compounded across different environments, the

plural singular dual

NOM žony žona žonje ‘wife’