Professional Documents
Culture Documents
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
University of Surrey
GREVILLE G. CORBETT
University of Surrey
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
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
xi
xii Contents
3.1.3 Conclusion 56
3.2 Person 57
3.2.1 Introduction 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
3.3.1 Introduction 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
3.4.1 Introduction 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
3.5.1 Introduction 92
3.5.2 Values 93
3.5.3 Directionality 94
3.5.4 Summary 95
3.6 Tense-aspect-mood 95
3.6.1 Introduction 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
3.7.1 Introduction 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
3.8.1 Introduction 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
6 Conclusion 219
6.1 Taking stock 219
6.2 Results 220
6.3 Consequences 221
References 254
Author index 271
Language index 275
Subject index 279
Preface
xv
xvi Preface
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
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
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
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
o-stem DAT SG -o
i -o
i
LOC SG -oi, -ei
consonant stem DAT SG -ei -i
LOC SG -i
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).
6 The Syntax–Morphology Interface
Proto-Indo-European/
Italic5 > Latin
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
(8) Singular indicative forms of the verb ‘eat’ in Mursi (Turton 1981:
341–2)
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.
INF sein
1SG bin
2SG bist
3SG ist
1PL sind
2PL seid
3PL sind
(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
For the noun mesto ‘place’, the genitive singular form is identical to the
nominative/accusative singular, while for vino ‘wine’, the genitive singular is
10 The Syntax–Morphology Interface
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.
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
12 The Syntax–Morphology Interface
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
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
14 The Syntax–Morphology Interface
(2) Nested syncretism in Upper Sorbian (Šołćina and Wornar 2002: xx 4, 10, 12)
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
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.
16 The Syntax–Morphology Interface
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).
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.
Characteristics of syncretism 17
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.
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
1SG u- Ø- Ø-
2SG a- aw- (stem-initial V:)
3SG i- aw- (stem-initial V:)
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)).
Characteristics of syncretism 19
a. ‘give’ b. ‘live’
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.
possessor
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.
object
6
Some of the forms with 1SG possessor are optionally based on the stem nuu-; these have been
omitted.
Characteristics of syncretism 21
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.
masculine feminine
singular plural singular plural
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).
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
22 The Syntax–Morphology Interface
indicative subjunctive
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).
Characteristics of syncretism 23
positive negative
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?
(18) Latin
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
Characteristics of syncretism 25
(20) Lak (Xajdakov 2001: 354) and Udihe (Nikolaeva and Tolskaya 2001:
212–13)
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)
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)
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
Characteristics of syncretism 27
singular plural
1 wada-de-m wada-de-y
2 wada-de-w wada-de-y
3 wada-de wada-d-iN
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.
Characteristics of syncretism 29
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).
30 The Syntax–Morphology Interface
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.
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’
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.
singular plural
M igral igrali
F igrala igrali
N igralo igrali
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.
32 The Syntax–Morphology Interface
singular plural
1 igraju igraem
2 igraeš 0 igraete
3 igraet igrajut
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.
Characteristics of syncretism 33
M SG N SG F SG
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
36 The Syntax–Morphology Interface
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
37
38 The Syntax–Morphology Interface
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
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
40 The Syntax–Morphology Interface
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.
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
(5) a. Tsakhur (Kibrik 1999: 130, 201) b. Domaaki (Lorimer 1939: 44, 76)
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)).
‘I’ ‘girl’
‘I’ ‘girl’
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.
I II III
‘you.PL’ ‘he’ ‘that.M’
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).
46 The Syntax–Morphology Interface
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.
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
4
Exceptions include Sanskrit mitra ‘friend’ and vr:tra ‘foe’ (Gregory Stump (personal com-
munication), citing Whitney 1889: x1185c).
48 The Syntax–Morphology Interface
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
Cross-linguistic typology of features 49
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.
singular plural
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.
Cross-linguistic typology of features 51
noun pronoun
‘foliage’ ‘he’
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.
52 The Syntax–Morphology Interface
7
One also finds languages where both these case functions are combined in a single case, e.g.
the dative-accusative of Pengo.
Cross-linguistic typology of features 53
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.
‘big village’
(17) German
‘boy’ ‘people’
(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.
animate: man + bad = ‘bad man’ inanimate: grass + bad = ‘bad grass’
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
*
Person distinguished in singular only.
*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).
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
62 The Syntax–Morphology Interface
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).
Cross-linguistic typology of features 63
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.
(26) Macedonian
obligatory mood
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
Cross-linguistic typology of features 65
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
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
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’).
66 The Syntax–Morphology Interface
verb + pronoun +
indicative interrogative pronoun verb
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.
Cross-linguistic typology of features 67
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)
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
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
future,
permis- converbs,
past, sive, present
past subjunct- perfect, and future
participle present ive conditional participles
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.
Cross-linguistic typology of features 69
default subjunctive
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.
(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
70 The Syntax–Morphology Interface
option 1 option 2
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
Cross-linguistic typology of features 71
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
(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.
Bellary Kumta
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).
perfect
Central, Finnish,
present reconstructed Balkan, Vlax Sinti Welsh Northeast
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.
74 The Syntax–Morphology Interface
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.
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
Cross-linguistic typology of features 75
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.
object
‘personal
reference’ 1SG 1PL 2SG 2PL Ø or 3
object
object
object
object
1 2 3 or ø
1 — -N -N
subject
2 n«- -N — n«-
3 n«- -N n«- Ø
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
80 The Syntax–Morphology Interface
Otherwise, the examples are not numerous enough to make any mean-
ingful generalizations.
object
1 2 Ø or 3
1 — ny- ?-
subject
2 ny-m- — m-
3 ny- m- Ø
object
1SG 1PL 2 3
1SG — — ka- a-
subject
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.
82 The Syntax–Morphology Interface
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.
Cross-linguistic typology of features 83
SG PL SG PL
(restrictive non-restrictive)25
singular plural
M e- lo- Ni- lu-
F a- na- Na- na-
N i- ni- Ni- lu-
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).
Cross-linguistic typology of features 85
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.
singular plural
M el ei
F ea ele
N el ele
86 The Syntax–Morphology Interface
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).
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
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
88 The Syntax–Morphology Interface
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).
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%
* 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)
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.
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).
Cross-linguistic typology of features 91
realis verbs
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.
92 The Syntax–Morphology Interface
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.)
27
This also obtains for part of the bound pronominal system, namely in transitive verbs
which mark both subject and object.
Cross-linguistic typology of features 93
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.
94 The Syntax–Morphology Interface
(70) Auxiliary forms for continuous events in Yele (Henderson 1995: 36)
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
Cross-linguistic typology of features 95
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.
96 The Syntax–Morphology Interface
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.
present/past potential
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).
*not attested
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).
Cross-linguistic typology of features 99
‘descend’ ‘show’
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.
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.
100 The Syntax–Morphology Interface
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.
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).
Cross-linguistic typology of features 101
I anterior past
II future
III recent past
IV present
V immediate past
VI potential
VII contemporaneous
VIII negative
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
Note:
* ‘S’ here stands for any verb stem
* trial number (not shown) is distinguished from the plural by the additional affix -bi-
Cross-linguistic typology of features 103
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).
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.)
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.
Cross-linguistic typology of features 105
singular plural
1INCL — A
1 A B
2 B B
3F B A
3M A A
form A form B
masculine feminine
singular plural
1 ?- -Ø n- -n
2 t- -t t- -t
3F t- -t y- -Ø
3M y- -Ø y- -Ø
singular plural
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
110 The Syntax–Morphology Interface
(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-
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.
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.)
Cross-linguistic typology of features 113
(91) The adjective dobry ‘good’ in Lower Sorbian (Stone 1993: 630)
‘grove’
NOM SG dobrava
ACC SG dobravo
GEN SG dobrave
LOC SG dobravi
DAT SG dobravi
INS SG dobravo
NUMBER CASE
Context Syncretic
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.
118 The Syntax–Morphology Interface
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.
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.
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.
‘teach’
non-past past
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
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
Cross-linguistic typology of features 125
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.
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 w- c-
2 f- c-
3 m- c-
4 k- t-
5 p- t-
diminutive j- t-
animate y- ”-
H
1 2 3 4 5 6 diminutive animate
1 2 3 4 5 dimin. animate
realis verbs
present and
adjectives pronoun unmarked
tense object
present with first
first second (intransitive person
declension declension subject) subject
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.
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’.
gender
non-male
inanimate
plant
I II IV III
(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).
Formal representation 131
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)
I þ
II þ þ
III þ þ
IV þ þ þ þ
(11)
a. SG PL b. unmarked marked
M k- t- unmarked U/ U U/M
F t- k- marked M/U M/M
(12)
a. hierarchical b. cross-classifying
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
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
(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.
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
136 The Syntax–Morphology Interface
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.
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.
Formal representation 137
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)
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.
(24)
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.
Formal representation 139
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.
140 The Syntax–Morphology Interface
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.
Formal representation 141
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).
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.’
Formal representation 143
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
144 The Syntax–Morphology Interface
I II III IV V
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.)
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.
Formal representation 145
(35)
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
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
Formal representation 147
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.
student-Ø [ ] *!
F student-a [(þhr)]
F stol [ ] *
stol-a [(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
148 The Syntax–Morphology Interface
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.
Formal representation 149
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).
o-stems i-stems
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).
NOM [] *!
F GEN/ACC [(þhr)N]
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
150 The Syntax–Morphology Interface
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.
Formal representation 151
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.
(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
(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)
singular plural
(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
Formal representation 155
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.
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)
(53)
present past
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
(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
158 The Syntax–Morphology Interface
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.
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 ?/?
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:
Formal representation 161
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.
162 The Syntax–Morphology Interface
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-
(60)
(61)
person
1, 3 2, 1INCL
SG [þSG PL] Ø Ø
DU [SG PL] k- i-
PL [SG þPL] i- i-
(62)
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.
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
Formal representation 165
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).
166 The Syntax–Morphology Interface
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
Formal representation 167
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).
168 The Syntax–Morphology Interface
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,
Formal representation 169
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.
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
171
172 The Syntax–Morphology Interface
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
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.
174 The Syntax–Morphology Interface
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).
Formal framework and case studies 175
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.
VERB
EN_VERB
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.
178 The Syntax–Morphology Interface
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.
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
Formal framework and case studies 179
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.
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:
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.
180 The Syntax–Morphology Interface
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.
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).
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.
182 The Syntax–Morphology Interface
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.
singular plural
1INCL — A
1 A B
2 B B
3F B A
3M A A
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,
Formal framework and case studies 185
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> >".
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.
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.
188 The Syntax–Morphology Interface
1 EXCL 1 INCL 2 3
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.
SG PL DU PL DU SG PL DU SG PL DU
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
subject
7
Used for a higher animacy object.
190 The Syntax–Morphology Interface
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 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?
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
Formal framework and case studies 191
look at other parts of the transitive paradigm, such as that for the first
exclusive subject in (34).
subject
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?
subject
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
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?
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.
subject
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
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).
196 The Syntax–Morphology Interface
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).
Formal framework and case studies 197
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)
198 The Syntax–Morphology Interface
VERB
Nan OTHER
VERBS
(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.
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.
Formal framework and case studies 201
(53) For a given argument of the verb, person attributes are ordered before
number attributes.
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.
Formal framework and case studies 203
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.
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.
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
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.
206 The Syntax–Morphology Interface
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.
‘be’ ‘speak’
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.
208 The Syntax–Morphology Interface
NOM SG kost0
ACC SG kost0
GEN SG kosti
DAT SG kosti
LOC SG kosti
INS SG kost0 ju
(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.
Formal framework and case studies 209
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).
I ‘law’ IV ‘wine’
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
Formal framework and case studies 211
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).
1PL 3PL
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.
212 The Syntax–Morphology Interface
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
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
Formal framework and case studies 213
consider what would be a highly systematic type of syncretism, but one for
which we have found no convincing evidence.
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
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
216 The Syntax–Morphology Interface
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.
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
218 The Syntax–Morphology Interface
219
220 The Syntax–Morphology Interface
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
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
222 The Syntax–Morphology Interface
223
224 Appendix 1
1
Note that this value has changed from that found in Baerman and Brown (2005a).
Case syncretism in the WALS sample 227
228
Person syncretism in the WALS sample 229
1
Data for Otomi added subsequent to Baerman and Brown (in press b).
232 Appendix 2
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).
233
234 Appendix 3
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)
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.
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
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> " > "
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
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> "
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.
DATR fragment for Dhasaanac case study 241
The verb Le´et 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.
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)
242
DATR fragment for Dalabon case study 243
VERB:
<> ¼ ¼ undefined
<mor> ¼ ¼ MOR_VERB
<syn> ¼ ¼ SYNTAX
<syn cat> ¼ ¼ verb.
MOR_VERB:
<mor> ¼ ¼
<mor infl> ¼ ¼ <mor cl> "<root>"
Referrals
Example (56) in Chapter 5
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>
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>
DATR fragment for Dalabon case study 245
Transitive prefixes
The prefixes of the transitive paradigm are similar to the intransitive,
having a front and final element which mark person and number.
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.
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>
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
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.
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.
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 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>">.
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
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.
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).
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
Közleme´nyek 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
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
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
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
275
276 Language Index
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
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