Objects and information structure

Mary Dalrymple (mary.dalrymple@ling-phil.ox.ac.uk)

Centre for Linguistics and Philology
Oxford University
Walton Street
Oxford OX1 2HG UK
telephone: +44 1865 280 403
fax: +44 1865 280 412
Irina Nikolaeva (irina a nikolaeva@yahoo.com)
Department of Linguistics
School of Oriental and African Studies
University of London
Thornhaugh Street, Russell Square
London WC1H 0XG
telephone: +44 (0)20 7898 4250
DRAFT: Do not quote or cite
March 2, 2010

List of Abbreviations
We have regularised some of the glosses and the transcription of the original
sources. We use the following abbreviations:
Abs
Acc
All
AN
Art
Asp
Attr
Aug
Aux
Caus
Cl
Com
Conv
Cop
C
Dat
Decl
Det
Dir
Dur
Du
Erg
Expl
Ez

absolutive
accusative
allative
action nominal
article
aspect
attributive
augmentative
auxiliary
causative
classifier
comitative
converb
copula
completive
dative
declarative
determiner
directive
durative
dual
ergative
expletive
ezafe

Fem
Foc
Fut
Gen
Hab
Hon
Impf
Imprs
Imp
Indef
Indic
Inf
Instr
Interj
Intr
Irr
Loc
Masc
Mod
Neg
NMLZ
Nom
Obj

feminine
focus
future
genitive
habitual
honorific
imperfective
impersonal
imperative
indefinite
indicative
infinitive
instrumental
interjection
intransitive
irrealis
locative
masculine
modal
negation
nominalisation
nominative
object marker or
object agreement

Obl
P
Part
Pas
Perf
Pl
Poss
Postess
Prep
Pres
Prog
Proh
Pron
Purp
Rel
Rep
Sg
Spec
Subj
Sub
Supess
Top
Tr

oblique agreement
potential
participle
passive
perfective
plural
possessive
postessive
preposition
present
progressive
prohibitive
pronominal
purposive
relative form
reportative
singular
specific
subject agreement
subordinate
superessive
topic
transitive

Preface and
Acknowledgements
Our work on this topic began in connection with our participation in the AHRC
project Noun Phrase Agreement and Coordination (AHRC grant MRG-AN10939/APN17606,
2004-2006), a joint project which involved, besides us, Louisa Sadler, Ryo
Otoguro, and Aline Villavincencio at the University of Essex. We are grateful
to Louisa, Ryo, and Aline for helpful comments and discussion as our ideas
developed over the course of the project. Earlier versions of the work were
presented at the 2005 International Lexical-Functional Grammar Conference
(LFG05) in Bergen, the 2005 Linguistic Association of Great Britain conference in Cambridge, the 2005 Colloque de Syntaxe et Semantique a` Paris
(CSSP 2005), the University of Leipzig in December 2007, the University
of Manchester in April 2008, and the workshop Arctic Languages: Syntax,
Morphology, Lexicon, University of Troms in September 2009. An early
version of our work appeared as Dalrymple and Nikolaeva (2005), though
it will be clear to readers of that paper that our ideas have evolved considerably since its publication. For very helpful comments and discussion of
these earlier versions, we are grateful to Farrell Ackerman, Ash Asudeh, David
Beck, Balthasar Bickel, Miriam Butt, Greville Corbett, David Cram, Annahita
Farudi, Ron Kaplan, Tracy Holloway King, Aditi Lahiri, Olivia Lam, Louise
Mycock, Tara Mohanan, Louisa Sadler, Devyani Sharma, Nigel Vincent, Yogendra Yadava, and Annie Zaenen. Farrell Ackerman, Louise Mycock, and
Louisa Sadler provided very helpful comments on drafts of this book as it took
shape. We are grateful to our language consultants for their generous assistance: Ghazaleh Kad and Shamsi Saber for Persian, Dolgor Guntsetseg and
Tuya Bavuu for Khalkha Mongolian, Galina Koreneva and Anna Lamdo for
Tundra Nenets, and Devyani Sharma and Tara Mohanan for Hindi. Finally, we
thank our husbands, Ken Kahn and Simon Carne, for their love and support
during the writing of this book.

iii

Contents
List of Abbreviations

Preface

iii

Introduction
1.1 The phenomenon
1.2 Previous work
1.2.1 Marking as distinguishing arguments
1.2.2 Marking as coding features
1.2.3 DOM in transformational syntax
1.3 Criteria for marking
1.4 Our proposal
1.4.1 Information structure
1.4.2 Syntax
1.4.3 Diachrony and grammaticalisation
1.4.4 Limits of our analysis
1.5 Structure of the book

1
1
2
3
6
8
10
14
14
17
18
19
20

Syntactic assumptions
2.1 Grammatical functions
2.1.1 Grammatical function diagnostics
2.1.2 Grammatical functions and semantic roles
2.2 Objects
2.3 Levels of syntactic representation
2.4 Describing linguistic structures
2.5 Grammatical agreement
2.6 Agreement and pronominal incorporation
2.7 Casemarking
2.8 Nonsyntactic critera for casemarking patterns
2.9 Conclusion

23
24
24
26
27
29
33
36
37
40
43
45

Information structure in grammar

3.1 The content of information structure
3.2 Information structure roles
3.2.1 Focus
3.2.2 Topic
3.2.3 Topicality and topic-worthiness
3.2.4 Secondary topic
3.3 Conclusion

47
47
49
49
50
52
55
59

Syntax and information structure

4.1 Alternative views of information structure
4.1.1 Tree-based representations
4.1.2 Early work in LFG
4.2 Information structure and its role in grammar
4.2.1 Content of information structure: The LFG view
4.2.2 Linguistic encoding of information structure relations
4.2.3 Information structure in relation to semantics
4.3 Our architecture
4.3.1 Glue and the syntax-semantics interface
4.3.2 Information structure features and relations
4.3.3 Levels and equations
4.3.4 A short text
4.3.5 Information structure and its place in grammar
4.4 Conclusion

61
61
61
64
68
68
69
71
74
74
77
81
88
91
93

Topicality and grammatical marking

5.1 Topical marking for different grammatical functions
5.2 Grammatical encoding of topical subjects
5.3 Subjects and topichood
5.4 Conclusion

95
95
100
103
107

Topical marking of nonsubjects

6.1 Casemarking of topical nonsubjects
6.1.1 Persian
6.1.2 Tariana
6.1.3 Topical nonsubject casemarking
6.2 Agreement with topical nonsubjects
6.2.1 Itelmen
6.2.2 Tabassaran
6.2.3 Topical nonsubject agreement
6.3 Conclusion

109
109
109
115
118
119
119
122
125
127

Topicality and DOM

7.1 Objects as grammaticalised secondary topics
7.2 Agreement with topical objects: Tundra Nenets
7.3 Casemarking of topical objects
7.3.1 Tigre
7.3.2 Dolakha Newar
7.4 Conclusion

129
129
131
137
137
141
143

Primary and secondary objecthood and DOM

8.1 Grammatical marking and grammatical function
8.2 Object agreement and grammatical function: Ostyak
8.3 Object casemarking and grammatical function
8.3.1 Mongolian
8.3.2 Chatino
8.3.3 Hindi
8.4 Objects and markedness
8.5 Conclusion

145
145
146
156
156
160
164
170
173

Multiple objects and grammatical alignment

9.1 On the typology of multitransitive constructions
9.2 No ditransitive construction
9.2.1 Nenets
9.2.2 Ostyak
9.2.3 Mongolian
9.3 Ditransitive constructions
9.3.1 Chatino
9.3.2 Hindi
9.4 Other multitransitive constructions
9.4.1 Dolakha Newar
9.4.2 Tigre
9.4.3 Applicatives: Upper Necaxa Totonac
9.5 Topicality and goals
9.6 Conclusion

175
175
178
178
179
182
183
184
185
186
187
191
194
198
200

10 Grammaticalisation
10.1 Case studies
10.1.1 Uralic
10.1.2 Persian and the Iranian languages
10.1.3 Hindi and the Indo-Aryan languages
10.2 Paths of grammaticalisation
10.2.1 Spreading of DOM
10.2.2 Narrowing of DOM

201
201
202
209
210
214
214
218

10.3 Towards a typology

10.4 Conclusion
11 Conclusion

221
224
225

Introduction
1.1 The phenomenon
Many languages exhibit non-uniform grammatical marking targeting objects.
Variations can occur within one and the same language with objects of one and
the same verb. For example, in Turkish (Altaic1 ) the object of the same verb
either takes the accusative suffix or remains unmarked:2
(1) a. Ali bir kitab-1
ald1
Ali one book-Acc buy.Past.3Sg
Ali bought a certain book.
b. Ali bir kitap ald1
Ali one book buy.Past.3Sg
Ali bought a book.

(Enc 1991:5)

In Palauan (Austronesian), the object of the same verb either does or does not
trigger agreement on that verb:
(2) a. Te-illebed-ii
a bilis a rengalek
Subj.3Pl-Perf.hit-Obj.3Sg the dog the children
The kids hit the dog.
b. Te-illebed
a bilis a rengalek
Subj.3Pl-Perf.hit the dog the children
The kids hit a dog/the dogs/some dog(s).
1 Language

(Woolford 2000:5)

families are provided according to the classifications in Ethnologue (Grimes 1999).

are provided for examples that do not come from our own fieldwork.

2 References

Introduction

Such patterns are widely known under the rubric of differential object marking or DOM (a term introduced by Bossong 1985).
We understand DOM as covering both agreement and casemarking (case or
adpositional marking on the object). Though we recognise that agreement and
casemarking differ both historically and synchronically, as noted by Comrie
(1979) and Croft (1988:167168), among many others, we believe that they
share commonalities in DOM, and we will use the cover term (grammatical)
marking to refer to them. This approach is in line with Nichols (1986), who
analyses agreement and casemarking as alternative strategies for encoding the
relation between the head and a dependent, as well as some generative literature, where case and agreement are inherently linked.
The aim of this book is to provide a new view of DOM which encompasses
syntactic, semantic, and information-structural differences between marked
and unmarked objects. We will make the following claims:
Marked objects are associated with the information-structural role of
topic. The association is synchronic in some languages, and may be
historical in others. Where the direct connection between marked objects and topicality has been lost through grammaticalisation, marked
objects in some languages become associated with semantic features
typical of topics (animacy, definiteness, specificity).
In some languages, marked and unmarked objects display an identical
behavioural profile and can be assigned to the same grammatical function. Other languages distinguish syntactically between marked and unmarked objects: marked objects are primary objects, while unmarked
objects are secondary objects. This reflects the tendency for topical
arguments to appear high on the grammatical function hierarchy, and
nontopical arguments to appear lower.
We begin our discussion with a review of previous work; we then present the
essential aspects of our claims, and conclude this chapter with an overview of
the book.
1.2 Previous work
DOM has been studied from a formal, generative perspective as well as a
functional-typological perspective, and has been discussed and analysed in detail by Lazard (1984), Bossong (1985, 1991), de Hoop (1992), Aissen (2003a,b),
Nss (2004), and de Swart (2007), among many others. Many of these analyses concentrate either on differential object agreement or on differential object
casemarking, including both case and adpositional marking.

Previous work

1.2.1 Marking as distinguishing arguments

Analyses of grammatical marking (and in particular casemarking) have often
appealed to two types of functional motivation, referred to as coding/indexing
and discriminatory/disambiguating/distinguishing (Comrie 1979, 1989, de Hoop
and Narasimhan 2005, de Hoop and Malchukov 2007, Nss 2007, Malchukov
2008, and others).
Discriminatory or disambiguating casemarking serves to distinguish between
different categories: for example, between the two arguments (the subject and
the object) of a transitive clause. It encodes the relation between two arguments rather than the properties of an individual argument. The discriminatory
function of casemarking has been argued to provide a functional motivation
for the fact that in many languages casemarking is missing on the single argument of intransitive verbs and on one of the two arguments of transitive
verbs. Silverstein (1976) and Comrie (1977) argue that since the basic purpose of formal marking on core arguments is to distinguish the subject from
the object, the need to overtly mark the object is greater in some cases than in
others because an object with subject-like semantic properties for example,
an animate, specific, or definite object is more likely to be confused with
the subject. Therefore, objects whose semantic features are typical of subjects
are more likely to be overtly marked. This approach relies on the concept of
the transitive prototype, in which the object is prototypically inanimate, indefinite, and/or nonspecific (Comrie 1989), and maintains that the function of
DOM is to signal deviation from the prototype. It also stands in conformance
with the widespread functional view that infrequent (and therefore functionally
marked) categories receive more formal marking, whereas frequent (and therefore functionally unmarked) categories tend to remain formally unmarked; the
explanation for this is based on economy considerations (Haiman and Thompson 1985, Du Bois 1987) and the iconic relation between functional and formal markedness. On this view, DOM is essentially iconic: formal marking
on objects reflects their status as atypical or infrequent objects, and thus their
functional markedness.
The idea that marking serves to distinguish or differentiate between arguments of a predicate has been pursued in depth in the influential work of Aissen (2003a,b). In her approach, languages may appeal to different factors in
DOM, but in all cases the resulting patterns reflect the tension between two
functional principles: (i) iconicity between functional and formal markedness
of objects, and (ii) economy, the pressure to avoid excessive marking. Following much work in functional typology (Silverstein 1976, Givon 1976, Comrie
1977, 1979, 1989, Du Bois 1987, Dixon 1994), Aissen claims that there is a
prototypical association involving grammatical functions and features such as
animacy, humanness, definiteness and specificity/referentiality. Subjects are

Introduction

prototypically high in these features, while objects are low. In other words,
properties that are unmarked for subjects are marked for objects, a relation
known as markedness reversal. On this view, unmarked subjects are animate,
human, definite and specific, while marked subjects are inanimate and/or nonspecific. For objects, the opposite markedness patterns are at work: inanimate
and indefinite/nonspecific objects are unmarked, while nonhuman definite animate objects are more marked, and human objects are most highly marked.
Aissen formalises these correlations as Optimality Theoretic constraints.
Referential properties of animacy, humanness, definiteness and specificity are
organised into two Prominence Scales, the Animacy Scale and the Definiteness
Scale.
(3) Animacy Scale: Human > Animate > Inanimate
Definiteness Scale: Personal pronoun > Proper name > Definite NP >
Indefinite specific NP > Nonspecific NP
Humans are located higher on the Animacy Scale than nonhuman animates,
which in turn are higher then inanimates, and so on. In addition, Aissen introduces a binary Relational Scale, where the subject outranks the object, as
well as several harmonic (or markedness) hierarchies representing the relation
between the Prominence Scales and the Relational Scale. The harmonic alignment constraints for animacy and definiteness features are displayed in (4) and
(5), respectively.
(4) *Su/Inan  *Su/Anim  *Su/Hum
*Oj/Hum  *Oj/Anim  *Oj/Inan
(5) *Su/NSpec  *Su/Spec  *Su/Def  *Su/PN  *Su/Pro
*Oj/Pro  *Oj/PN  *Oj/Def  *Oj/Spec  *Oj/NSpec
The most highly ranked constraints in (4) penalise inanimate subjects and human objects; the constraints ranked one step lower penalise animate nonhuman
subjects and animate nonhuman objects, and so on. The definiteness alignment
constraints in (5) work similarly.
These hierarchies predict the most and least marked patterns of subject and
object marking across languages. Constraints higher on the hierarchy incur
more costly violations than constraints lower on the hierarchy. This means that
if an object at a certain point in the hierarchy is overtly marked, then any object
that is higher on the relevant scale will also be overtly marked. DOM arises
when some but not all objects are marked; this is implemented by correlating
harmonic constraint hierarchies with the constraint *STRUCC , motivated by

Previous work

the needs of economy, which penalises a value for the morphological category
CASE below a certain point on the hierarchy.
The position of *STRUCC in the hierarchy leads to different patterns of object marking across languages. If *STRUCC dominates all the constraints on
both scales, then marking is banned for all objects. If *STRUCC is ranked
at the lowest point on the hierarchy, all objects receive grammatical marking.
Such languages do not display DOM. In languages with DOM, object marking can be based either on the Animacy Scale or on the Definiteness Scale.
For example, Aissen (2003b) shows that in Catalan (Romance) only personal
pronoun objects are casemarked. This is captured in an Optimality Theoretic
account by positioning *STRUCC lower in the Definiteness Scale than the constraint penalising pronominal objects. Similarly, if *STRUCC is ranked lower
than the top-ranked constraint *Oj/Hum in the Animacy Scale, casemarking
is penalised for all objects except the most highly ranked human objects. According to Aissen (2003b:456), such languages are difficult to find, although
there are many languages where marking is penalised for all objects except
animates (including humans): for example, Sinhalese (Indo-Aryan). Further
demotion of the economy constraint *STRUCC yields other patterns of object
marking. In Pitjantjatjara (Pama-Nyungan), pronominal and proper name objects are marked, while other objects, including definite objects, are unmarked;
this is treated by positioning *STRUCC below the constraints penalising marking on pronominal and proper name objects. In Hebrew (Semitic), only definite objects require the object marker et, while indefinite objects are always
unmarked; this means that the economy constraint is ranked lower than the
constraint penalising definite objects. Simultaneous reference to multiple features involves more complicated multidimensional crossing of the scales, but
the basic principle remains the same.
Aissens pioneering proposals have inspired much subsequent work and discussion, including an exploration of patterns that do not fit neatly into Aissens
cross-linguistic picture. For example, Yang and van Bergen (2007) argue that
in Mandarin Chinese (Sino-Tibetan), objects that are obligatorily marked in
the ba-construction are either highly prominent in terms of animacy or, surprisingly, low in prominence in terms of definiteness; for a small category of
objects in-between, casemarking is optional. Importantly, the effect of animacy and definiteness is only noticeable on scrambled objects; Yang and van
Bergen propose that the syntactic position of the object introduces an additional dimension into the prominence hierarchy which can influence marking.
Other works following on from Aissens work, including Morimoto (2002),
propose various readjustments of the original constraint hierarchy, but do not
question the general spirit of the prototype deviation approach to DOM.

Introduction

1.2.2 Marking as coding features

The coding/indexing perspective on marking differs from the discriminatory
perspective in that marking is taken to signal specific semantic and/or pragmatic properties of the relevant argument, rather than a particular relation between one argument and another. In fact, de Hoop and Narasimhan (2005)
claim that a purely discriminatory function for casemarking is cross-linguistically
rare, and that in most instances casemarking serves to mark some aspect of the
interpretive content of a phrase. This is particularly clear for obliques, where
case can bear an important semantic load in signalling the meaning contribution of the casemarked phrase. In the analysis of DOM, the coding/indexing
perspective assumes that the presence of marking is connected to the presence
of certain properties of the object. This view goes back to Hopper and Thompsons (1980) classic study, in which DOM is taken to be one of the signals
indicating high transitivity, rather than deviation from a transitive prototype.
Nss (2004, 2006, 2007) argues that Aissens approach contradicts these
established notions of transitivity, and proposes that the prototypical transitive
clause is one in which the two participants are maximally semantically distinct.
Her definition of semantic distinctness includes several parameters, but the parameter that is especially relevant for cross-linguistic patterns of DOM is affectedness, understood roughly as involvement in the verbal event and change
of state of the participant as a direct result of this event. On this view, prototypical objects are those that are highly affected by the transitive event. According
to Nss, then, formally marked objects are not functionally marked; instead,
they are prototypical highly affected and individuated objects, which tend
to receive more grammatical marking than non-prototypical, less affected
objects.
Other researchers have also appealed to affectedness as a factor in analyses
of DOM: for example, C
etinoglu and Butt (2008) discuss the role of affectedness in object casemarking in Turkish. However, the relevance of affectedness
for DOM has been questioned by de Hoop (2008), who shows that in many
cases object marking is present in sentences in which the object participant is
not affected by the verbal event. For instance, in Mandarin Chinese the object marker ba, often treated as an instance of DOM, is required on objects of
the verbs forget and lose, although the forgotten or lost thing is not usually
affected by the forgetting or losing event. Nss (2004) claims that definite objects are more affected than indefinite ones because the event affects the whole
rather than a part (cf. I drank the milk and I drank some milk). However, in
many cases it is difficult to argue that definite or animate objects are more affected than indefinite or inanimate ones, if affectedness is understood in terms
of a change of state. De Hoop cites the following Hindi (Indo-Aryan) examples
from Mohanan (1994), involving the accusative marker ko:

Previous work

(6) a. Ilaa-ne haar

ut.haayaa
Ila-Erg necklace lifted
Ila lifted a necklace.
b. Ilaa-ne haar-ko
ut.haayaa
Ila-Erg necklace-Acc lifted
Ila lifted the necklace.
(de Hoop 2008, citing Mohanan 1994:80)
According to de Hoop, there is no reason to think that the necklace that is
picked up in (6b) is more affected than the necklace that is picked up in (6a),
even though the former is definite and the latter is indefinite.
In fact, de Hoop and Narasimhan (2005), Nss (2007), and Malchukov
(2008) note that in the case of DOM, disambiguating and indexing approaches
lead to roughly the same result (though they make different predictions with respect to differential subject marking). Affectedness normally presupposes high
individuation of the referent, while individuation in its turn is associated with
definiteness. Animacy may also be relevant for affectedness because the effect
of an action on an animate entity is more salient for human perception than
the effect on an inanimate entity and is more likely to arouse empathy. Thus,
affectedness is ultimately operationalised in terms of the same features of
prominence as are relevant on the disambiguating/discriminatory marking perspective: when a language decides what kinds of objects are affected enough
to be marked, it can choose on the basis of more easily measurable properties
such as animacy and definiteness.
De Swart (2006, 2007) proposes an analysis which, in a sense, combines
the discriminatory and coding approaches. His model is based on the idea that
the speaker takes the perspective of the addressee in order to ensure recoverability of the intended interpretation. In some instances this implies that
the speaker chooses to mark an object, rather than leaving it unmarked, when
he/she wishes to emphasise a certain feature of the object: for instance, definiteness in Hindi. If the speaker intends to highlight definiteness in order to
ensure that the addressee will interpret the object as definite, accusative ko
appears. If the speaker does not want to force a definite interpretation of the
object, casemarking is omitted. Thus, marking on the object is the result of a
principled decision on the part of the speaker and has the coding function.
Recoverability presupposes semantic distinctness between two arguments
(Nss 2004). This explains why casemarking on objects can be influenced by
the semantic properties of the subject and vice versa. However, de Swart does
not account for these patterns in terms of transitivity and, unlike Aissen, does
not appeal to prototypical features of subjects and objects. He illustrates his

Introduction

analysis with data from languages which seem to best support the disambiguating/discriminatory view of marking, in which DOM seems to be primarily determined by the need to differentiate the subject from the object. De Swart
argues that in such languages, sentences with no semantic contrast between
agent and patient must show a morphological contrast between them, in order
to avoid ambiguity. The relevant semantic dimensions involve familiar prominence features, but are largely language-specific. In Malayalam (Dravidian),
for example, object marking is mostly found on animate objects. However,
in sentences which can in principle be interpreted in two different ways, it is
also found on inanimate objects; the reason seems to be that without overt
casemarking, the sentence would be ambiguous.
(7) a. kappal tiramaalakal-e bheediccu
ship.Nom waves-Acc
split.Past
The ship broke through the waves.
b. tiramaalakal kappal-ine bheediccu
waves.Nom ship-Acc split.Past
The waves split the ship.
(de Swart 2007, citing Asher and Kumari 1997)
Such systems are called global because marking depends on properties of the
subject, properties of the object, and the relation between them, along the lines
of the discriminating/disambiguating view of marking. In contrast, local
systems are those in which the presence of object marking is only dependent
on the features of the object itself, along the lines of the coding/indexing approaches. As de Swart notes, global systems present a problem for Aissens
model, since they depend on the simultaneous consideration of properties of
the subject and object rather than properties of the object alone, but can be
accounted for in de Swarts analysis within the framework of Bidirectional
Optimality Theory.
1.2.3 DOM in transformational syntax
Many analyses of phrasal syntax within the transformational paradigm assume
two distinct positions for objects, VP-internal and VP-external, and postulate a correlation between object position and object marking (Diesing 1992,
Dobrovie-Sorin 1994, van Geenhoven 1998, Torrego 1998, Ritter and Rosen
2001, Woolford 1999, 2000, 2001, de Hoop 1992, among others). The distinction is usually said to be semantically driven, dependent on a specific/referential
vs. a nonspecific/nonreferential interpretation of the object. It is generally

Previous work

assumed that VP-internal indefinite/nonspecific objects are syntactically less

visible than VP-external definite/specific ones.
De Hoop (1992) observes that a morphological difference in case in languages with DOM is linked to a semantic difference in the interpretation of indefinites. Her analysis builds on work by Diesing (1992) on the interpretation
of indefinite objects: Diesing suggests that arguments universally excluded
from VP-internal positions bear the feature [+ SPECIFIC ], while VP-internal arguments are interpreted as nonspecific. This is formally modelled by requiring that NPs that are not overtly quantified must move out of VP before LF
to escape existential closure. De Hoop (1992) assumes two categories of NPs,
strong (presuppositional) and weak. Strong NPs are anchors in conversation;
they are semantically characterised as generalised quantifiers and include referential, partitive, and generic expressions. Weak NPs are analysed as existentially quantified. Additionally, there are two kinds of object case: Weak Case,
assigned VP-internally at D-structure, and Strong Case, assigned at S-structure
to [Spec,AgrO]. A strong NP moves out of the VP to get Strong Case, while
Weak Case requires syntactic adjacency to the verb. Overt accusative marking
on VP-external objects is analysed as the morphological realisation of abstract
Strong Case. This is exemplified, for instance, in Turkish, as shown in (1),
where the marked and unmarked object receive specific and nonspecific interpretations, respectively.
These works deal only with casemarking, but since case and verbal agreement are treated as two aspects of the same phenomenon in this framework,
roughly similar analyses have been proposed for differential object agreement.
Both case assignment/checking (depending on ones syntactic assumptions)
and agreement are treated in terms of movement of the object to the specifier
position of the relevant agreement head. In object-agreement languages, agreement serves as a specificity licenser, as argued by Sportiche (1995), among
others.
Subsequent work has made it clear that specificity is not the only feature
responsible for DOM. Adopting the premise that VP-internal and VP-external
object positions may be associated with different semantic properties, Woolford (1999, 2000, 2001) aims to explain which objects occupy which of these
positions and why, taking into account more complex patterns where there is
no single semantic feature that triggers movement out of the VP and agreement. Instead, a VP-external object may have any of several distinct clusters
of features.
The basic premise of Woolfords proposal is that economy keeps objects
in their base VP-internal positions unless that would violate certain Exclusion
Principles. Exclusion Principles are modified versions of Diesings mapping
principles, which exclude NPs bearing certain features from the VP-internal
object position. Woolford assumes a family of Exclusion Principles based

10

Introduction

on different semantic features including specificity, humanness, animacy, and

number; on her view, these are separate principles and cannot be reduced to
one more general principle. Each Exclusion Principle can be independently
active in a language. The economy principle Avoid Movement, which prohibits moving objects out of their base position (Chomsky 1995), ensures that
objects remain within the VP unless this would violate one or more Exclusion Principles. Cross-linguistic differences in restrictions on agreement are
dealt with in terms of different rankings of various Exclusion Principles and
the economy principle Avoid Movement.
In some languages DOM seems to depend on aspectual features of the verb.
The idea that object marking correlates with aspect has been explored by Ramchand (1997), Meinunger (1998), Woolford (2000) and others, who suggest
that the Asp head is involved in Case assignment to the object. Woolford
(2000) claims that the Aspect node in Palauan supplies a Case feature when it is
[+PERFECTIVE]. Ritter and Rosen (2001) provide a more sophisticated analysis,
arguing that in languages with DOM the split in object marking is determined
by the presence or absence of the feature [Q UANT ( ISATION )]. Quantised objects
(their Class I objects) must check their Q UANT feature, forcing such objects
to move out of the VP and triggering agreement or casemarking, while nonagreeing or noncasemarked objects (their Class II objects) are not quantised
and remain within the VP. The specific semantic contribution of the feature
differs from language to language. When [Q UANT ] is an inherent feature of the
verb, it has aspectual meaning: it encodes delimitedness or boundedness of the
event. According to Ritter and Rosen (2001), this situation is exemplified in
Finnish (Uralic), where objects of bounded events stand in the accusative case
and objects of unbounded events take the partitive case; in Palauan, where object agreement correlates with boundedness of the event (as expressed through
perfectivity); and in Mandarin Chinese, where the ba-construction is only possible with delimited events. On the other hand, when [Q UANT ] is a feature of
the functional head Agr, it is uninterpretable and lacks inherent semantic content. In this case DOM is not sensitive to event type; instead, [Q UANT ] enters
into a checking relation with definite/specific/animate objects, as in Turkish,
Hebrew or Bantu. Ritter and Rosen (2001) do not explain why different semantic types are involved in object split in these languages, but emphasise the
importance of treating [Q UANT ] as a feature which bears on the interpretation
of verbs as well as objects, and which can be realised either as object case or
object agreement.
1.3 Criteria for marking
As we saw in the previous section, most existing work on DOM appeals to
inherent referential semantic features of the object noun phrase, such as ani-

Criteria for marking

11

macy, definiteness, or specificity, to distinguish marked and unmarked objects.

Aissens (2003a, 2003b) work is based on hierarchies of animacy and definiteness. For Nss (2004, 2006, 2007), the crucial notion of affectedness also correlates with referential properties of the object, although the correlation is indirect. Most transformational work on DOM is based on the premise that object
marking patterns are defined in terms of semantic features (Diesing 1992, van
Geenhoven 1998, and others). Work by Woolford (1999, 2000, 2001) relies on
a family of Exclusion Principles based on specificity, humanness, animacy, and
number. Ritter and Rosen (2001) use a general notion of boundedness which
encompasses specificity and definiteness as well as event-boundedness.
These criteria are indeed useful in analysing patterns of DOM where objects
that are characterised as semantically strong or definite show more agreement with the verb, or more casemarking, than objects without these properties. For example, semantic factors have been argued to provide the clearest
explanation of patterns of DOM in Turkish or Hebrew, where the distribution of marked and unmarked objects is fairly straightforward and definable
in terms of simple semantic features: Turkish marks specific objects, while
Hebrew marks definite objects.3 Yet these factors do not adequately account
for languages in which objects with the same semantic features can be either
marked or unmarked. Such variation was noted by Bossong (1991:152), who
concluded that in these languages the rules of DOM cannot be formulated precisely, but must allow for a certain degree of variability across speakers and
situations.
Take, for example, Hindi. As discussed by Mohanan (1994:79 ff.), Singh
(1994), and Aissen (2003b), the accusative marker ko in modern Hindi appears on specific human objects (in some dialects, all specific animate objects),
whether definite or indefinite, but is impossible on indefinite inanimate nouns.
3 However,

von Heusinger and Kornfilt (2005) and Kornfilt (2009) argue that the correlation
between specificity and accusative marking in Turkish is imperfect, and Danon (2006) claims that
the definiteness condition on casemarking in Hebrew is a purely syntactic condition, related to the
presence of an article, rather than a semantic property.

12

Introduction

(8) a. Ilaa-ne bacce-ko/*bacca ut.haayaa

Ila-Erg child-Acc/child lifted
Ila lifted the/a child.
b. Ravii-ne kaccaa kelaa kaat.aa
Ravi-Erg unripe banana cut
Ravi cut the/a unripe banana.
c. Ravii-ne kaccaa kele-ko
kaat.a
Ravi-Erg unripe banana-Acc cut
Ravi cut the/*a unripe banana.
(Aissen 2003b, from Mohanan 1994)
In other words, for specific human/animate nouns ko-marking is obligatory,
while for inanimate indefinites, marking is disallowed. However, with inanimate/nonhuman definite objects, apparent optionality arises: such objects are
either marked or unmarked.
On Aissens analysis of Hindi, the constraints that disallow ko on inanimate
nonspecific objects and indefinite inanimates are dominated by *STRUCC ,
which penalises a value for the morphological category CASE, so casemarking is absent. On the other hand, constraints that penalise the absence of
case on human/animate objects dominate *STRUCC , so accusative marking
is required. Optionality of marking on inanimate/nonhuman definite objects
is handled in terms of constraint reranking: the relative ranking of *STRUCC
against the constraints that penalise the absence of case on definite nonhuman/inanimate objects is undetermined. On this view, the patterns of grammatical marking are basically random and would be expected to reflect idiolectal
and/or free variation.
On De Swarts (2007) analysis, definiteness in Hindi does not trigger casemarking but is rather a side effect of the use of casemarking. De Swart claims
that this eliminates optionality; however, his theory does not lead to straightforward and testable predictions for the distribution of ko, and it remains unclear
how it can be formalised in an explicit grammar. Most importantly, no clear
evidence is provided for the hypothesis that the presence of the accusative correlates with the speakers intention to highlight definiteness.
Many languages exhibit similar patterns of optionality. In Sinhala (IndoAryan), inanimate objects are never overtly marked, whereas animate objects
are optionally casemarked.

Criteria for marking

13

(9) mam@ miniha(-w@) daekka

I
man-(Acc) saw
I saw the man.

(Gair and Paolillo 1997:32)

In Kannada (Dravidian), accusative marking is obligatory on animate and/or

definite objects but optional on indefinite inanimates (Lidz 2006). Casemarked
indefinite objects are specific, while noncasemarked objects are ambiguous
between a specific and nonspecific interpretation.
(10) a. Naanu pustaka huDuk-utt-idd-eene
I.Nom book look.for-Nonpast-be-1Sg
I am looking for a book/a certain book.
b. Naanu pustaka-vannu huDuk-utt-idd-eene
I.Nom book-Acc
look.for-Nonpast-be-1Sg
I am looking for a certain book.

(Lidz 2006:11)

Lidz (2006) shows that the specific vs. nonspecific interpretation of noncasemarked objects correlates with their syntactic position, but not exactly in the
way predicted by de Hoops (1992) analysis. Casemarked objects have a specific reading even when they occur inside the VP, while noncasemarked objects
may occur VP-externally, in which case they must receive a specific reading.
Apparent optionality is also observed in languages with differential object
agreement. In Ostyak (Uralic), object agreement may but need not appear with
definite as well as indefinite objects, as shown in (11) (see Nikolaeva 1999,
2001 and the discussion of Ostyak in Chapter 8, Section 8.2):
(11) tam kalaN we:l-@s
/ we:l-s-@lli
this reindeer kill-Past.3SgSubj kill-Past-Obj.3SgSubj
He killed this reindeer.
a:muj kalaN we:l-@s
/ we:l-s-@lli
some reindeer kill-Past.3SgSubj kill-Past-Obj.3SgSubj
He killed a/some reindeer.
These examples demonstrate that in some languages, semantic features such
as animacy, definiteness, or specificity alone are not enough to account for
the distribution of agreement or casemarking on objects. Further, there is no
evidence that the properties of the subject participant (as in global systems)
or the degree of affectedness play a role. Optionality is observed with exactly
the same subjects and exactly the same verbs.

14

Introduction

1.4 Our proposal

We approach DOM from a different perspective. We believe that in many
cases, the seeming unpredictability of variation in DOM is due to the limited
set of features examined in previous work, and that a better account is possible if additional factors are taken into account. In particular, variations in
marking can often be explained by reference to information structure, a level
of sentence grammar where propositions, as conceptual states of affairs, are
structured in accordance with the informational value of sentence elements and
contextual factors. We propose a theory of DOM which emphasises the role
of information structure in the marking patterns of objects. The informationstructural distinction we make explains many of the unexplained areas in semantic analyses of DOM.
1.4.1 Information structure
Specifically, we propose that topicality plays an important role in DOM. Marked
objects are often topical, while unmarked objects are nontopical. For example,
in Ostyak (Uralic), object agreement is required for topical objects:
(12) a. What did he do to this reindeer?
b. tam kalaN we:l-s-@lli
/ *we:l-@s
this reindeer kill-Past-Obj.3SgSubj kill-Past.3SgSubj
He killed this reindeer.
Nontopical, focused objects disallow agreement:
(13) a. Which reindeer did he kill?
b. tam kalaN we:l-@s
/ *we:l-s-@lli
this reindeer kill-Past.3SgSubj kill-Past-Obj.3SgSubj
He killed this reindeer.
This is true independent of semantic features of animacy and definiteness; in
both of these examples, the object is animate and definite.
Following Lambrecht (1994) and others, we understand topicality as a pragmatic relation that holds between a referent and the proposition expressed by
an utterance: topicality has to do with the construal of the referent as pragmatically salient, so that the assertion is made about this referent. Topicality
is not an inherent property of a referent, and although it correlates with the
role played by the referent in the preceding discourse, the correlation is imperfect. It cannot be unambiguously established on the basis of the referential
features of the object either; rather, it depends on the speakers assessment of

Our proposal

15

its saliency within a given communicative context, as we discuss in Chapter 3.

Since we view information structure as part of grammar, we treat grammaticality as including pragmatic well-formedness. For instance, object agreement is
ungrammatical in (13) only if it construed as an answer to the question Which
reindeer did he kill?, though it is possible in the context in (12). In order to
demonstrate the pragmatic role of a referent, we provide examples in context
whenever possible.
In some respects, our view most closely resembles the proposal of Nss
(2004): like ours, her analysis relies not on the semantic features of the object,
but rather on its perceptual role. In fact, Nss (2004) suggests that affectedness, which plays a crucial role in her treatment of DOM, has two aspects:
expressing a part/whole relation, and salience. Salience is a necessary property of topics, so the two notions, though distinct, seem to be closely related
(see Croft 1991:155 for further discussion of the correlation between affectedness and topicality). However, for Nss (2004) salience has to do with which
entities are of more interest for human perception in general. Her analysis
does not, then, address contextual factors. On our approach, salience relates
to the role played by a referent in the pragmatic structure of the proposition,
established within a given communicative context.
Our proposal provides a different functional motivation for DOM from the
discriminatory/disambiguating perspective: we argue that DOM was originally
motivated by the need to highlight similarities between subjects and topical
objects, which tend to be grammatically marked, as opposed to nontopical objects. That is, it arises from the need to give overt expression to a property that
is common to subjects and (some) objects but less typical of other non-core
grammatical functions: languages tend to mark topics, whether subjects or objects, either by agreement or by casemarking. This means that topical objects
are not functionally marked, atypical, or uncommon; in fact, we suggest that
objects are just as likely to be topical (grammatically marked) as nontopical
(focused, unmarked).
The idea that topicality may play a role in DOM has been mentioned in previous work (Aissen 2003b, Leonetti 2003, and others). However, it has not
received extensive elaboration in the family of functionally-typological and
Optimality Theoretic analyses of DOM, nor has it been standardly incorporated into most existing generative or transformational treatments. Some research employs the notion of prominence interpreted more broadly than Aissens Prominence Scales (de Hoop and Narasimhan 2005, de Hoop and Lamers
2006, de Hoop 2008). De Swart (2007:138) defines prominence as a feature
that is concerned with the centrality of an entity in the discourse or with the
readiness with which an entity presents itself to the speaker as a topic of conversation. His analysis distinguishes two levels of prominence: (i) discourse
prominence, which reflects the status of the argument in the discourse, and (ii)

16

Introduction

semantic prominence, which depends on the intrinsic semantic features of the

argument. The notion of discourse prominence seems to be comparable with
our notion of topicality; however, de Swart does not provide a detailed characterisation of discourse prominence, noting only that semantic prominence
influences discourse prominence because an inherently prominent element is
more likely to be topical.
Another exception is Meinunger (1998): working within the Minimalist Program, he suggests that the distinction between topical and nontopical objects is interpreted syntactically, and that the semantic feature that all
VP-external phrases have in common is [+T OPIC]. This feature activates VPexternal agreement projections, so that topical NPs must move to a SpecAgr
position for Case checking. He also assumes an inherent relationship between
verbal agreement and Case, which are seen as two aspects of the same phenomenon, in that both are effects of the movement of an argument to the specifier position of an agreement head. Languages differ in how they mark scrambled topical objects: some languages mark topical and nontopical objects with
morphologically different cases, while in other languages the difference in the
interpretation of the objects is linked to object agreement, and in a third type of
language the difference is only positional. Meinunger views topical arguments
as discourse-linked/familiar and contextually salient. In addition, the speaker
must intend to construe them as topics, as opposed to presentational focus
or comment, which normally contains discourse-new elements. However, his
definition of topic and the relationship between topicality and the referential
features of the object are not clearly spelt out. For instance, Meinunger claims
that topicality is realised as specificity in Turkish and animacy in Hindi, but
does not explain this difference, though object marking clearly depends on
very different semantico-pragmatic conditions in these languages.
Escandell-Vidal (2009) presents a detailed analysis of DOM by means of
the preposition a in Balearic Catalan. Although differential object casemarking in many varieties of Catalan is explained by appeal to semantic features of
animacy or definiteness, in Balearic Catalan referential properties of the object play a secondary role. Pronominal objects are always casemarked, but
for lexical objects casemarking is (partly) determined by topicality. As in our
analysis, Escandell-Vidal understands topicality as aboutness, and therefore
Balearic Catalan seems to behave very similarly to a number of the languages
we discuss in this book. Her analysis differs from ours in that she does not discuss the alignment of grammatical functions of marked and unmarked objects
with information structure roles, though she shows that there are positional
differences between topical and nontopical objects, since Balearic Catalan is
a language with designated topic and focus positions. Topical objects cannot
occur in the canonical focus position, and instead must be either left- or rightdislocated. However, only discourse-old topics, and not all topical objects,

Our proposal

17

are casemarked. This implies that both marked and unmarked objects can occur in topic position, and raises the question of whether they display identical
syntactic behaviour.
1.4.2 Syntax
This leads us to the next point. Our view differs from most previous proposals in another respect: we believe that existing typological analyses of DOM,
which mainly concentrate on its functional motivation and the cross-linguistic
distribution of morphological marking, do not pay sufficient attention to the
syntactic side of the phenomenon. The behavioural properties and syntactic
status of marked vs. unmarked objects are rarely discussed, and to our knowledge, there are few systematic cross-linguistic studies of the differences between them.
We have found that in some languages marked and unmarked objects do not
differ syntactically and, arguably, realise the same grammatical function, the
object. In other languages they not only have different information structure
roles, but also exhibit different behavioural syntactic profiles. For example,
marked and unmarked objects in Ostyak differ in their ability to control coreference with the subject of an action nominal dependent clause, allow possessor
topicalisation, control possessor reflexivisation, and launch floated quantifiers.
In other languages as well, a number of syntactic tests distinguish marked objects and unmarked objects, with the marked object displaying a larger number
of properties associated with core arguments. We suggest that in these languages, marked and unmarked objects bear different grammatical functions.
Our proposal is cast within the framework of Lexical Functional Grammar
(LFG), which distinguishes between the primary object OBJ and the secondary
or semantically restricted object OBJ (Bresnan and Kanerva 1989). These
two types of objects are usually discussed in connection with double object
constructions, but in many languages, both types are available in single object constructions as well, with the choice between them determined by their
information structure role.
As mentioned above, many analyses of DOM assume two distinct positions
for objects, VP-internal and VP-external, and posit a correlation between the
position of the object and grammatical marking (Diesing 1992, Dobrovie-Sorin
1994, van Geenhoven 1998, Torrego 1998, Ritter and Rosen 2001, Woolford
1999, 2000, 2001, de Hoop and Malchukov 2007, among others). Issues of
word order and the positions of objects have been studied thoroughly within
the LFG framework by Butt and King (1996), Choi (1999) and many other
researchers: their work has clearly shown that information structure role can
be relevant for word order constraints, and that information structure has a
strong effect on where objects can appear. However, we do not posit a di-

18

Introduction

rect relation between DOM and the phrase structural position of the object in
all languages, because at least in some of the languages we analyse there is
no obvious connection between the position of the object and its grammatical marking. More generally, we do not assume that grammatical functions or
information structure roles must be identified configurationally. LFG analyses grammatical functions as syntactic primitives, and does not define them in
terms of phrase structure position. Positional generalisations concerning the
behaviour of different grammatical functions or arguments bearing different
information-structure roles can be easily modelled within LFGs projection
architecture (Kaplan 1987, Asudeh 2006), which allows constraints to refer
to and relate different aspects of the structure of an utterance.
1.4.3 Diachrony and grammaticalisation
Our analysis also provides a diachronic explanation for why referential semantic features such as animacy, definiteness, and specificity often play a role in
DOM. We propose that these features are involved in grammaticalisation of
topichood.
Abundant diachronic evidence shows that DOM often originates as a marking device for topics. We take this situation to be historically primary. One possible direction of change involves widening of topical marking, where marking
spreads to certain nontopical objects. In a language with topical object marking, the marking can generalise or extend, applying to nontopical objects which
have features typical of topics. When this happens, marking patterns become
automatic consequences of distinctions at other levels of structure. As a result,
the role of information structure in object marking is diminished: marking is
obligatory not only for topical objects, but also for objects with certain semantic features, independent of their information structure role.
The opposite direction of change involves narrowing of topical marking:
marking becomes specialised for topics which bear certain semantic features.
In the relevant languages, only a subset of topical objects are formally marked,
while nontopical objects remain unmarked. Narrowing usually involves the
most typical members of the set. Objects ranked high on the prominence hierarchies are frequent topics, so topical marking can become restricted to them.
At the next stage of grammaticalisation, the connection to information structure may be completely lost, so that object marking becomes dependent on
semantic features alone.
Thus, we propose that different patterns of DOM arise as a result of different
directions of grammaticalisation of topic marking on objects. Features that are
typical of topics come to be required, or sufficient, for object marking.

Our proposal

19

1.4.4 Limits of our analysis

Our theory of DOM does not extend to all instances of variable marking of
objects. Since we are interested in differences in interpretation that depend
on the presence or absence of grammatical marking, we only consider languages where a subset of objects is grammatically marked (by either case or
agreement), while another subset receives no marking at all, as in the examples
cited above. We will not address those instances of DOM where objects can
take two alternative casemarkers, as attested, for example, in Icelandic (North
Germanic):
(14) a. Hann kloraDi mig
he.Nom scratched me.Acc
He scratched me.
b. Hann kloraDi mer
he.Nom scratched me.Dat
He scratched me.

(Nss 2004, citing BarDdal 2001)

The Icelandic accusative-dative alternation has been reported to reflect a semantic contrast: in (14a) the scratching is perceived as a forceful act of violence, possibly painful for the patient participant, while in (14b) the interpretation involves volitionality on the part of the patient. In other languages,
such as Russian (Slavic) or Finnish, the case of the object differs depending
on aspectual characteristics of the verb. As interesting as such instances may
be, we believe that they deserve separate treatment and do not fall under the
same generalisations as DOM in languages which contrast formally marked
and unmarked objects.
The focus of this book is the grammatical function of object. We limit the
scope of our study to object marking casemarking or agreement in order to more fully explore the relation between the presence and absence of
marking and the status of object arguments at different levels of linguistic representation. We exclude languages where some patient/themes are not syntactic objects at all, but participate in various detransitivising constructions, and
may remain syntactically unexpressed or incorporated into the verb. In other
words, we only deal with proper syntactic arguments. In this way, our work differs from some previous research on DOM such as, for example, Nss (2004),
which treats alternative encodings of patient/theme arguments independently
of their syntactic status.
Furthermore, our analysis does not address the distribution of verbal clitics or incorporated pronouns, although in many languages they are associated
with some but not all objects, as discussed by Bresnan and Mchombo (1987)

20

Introduction

for Chichewa
(Central Bantu), Culy (2000) for Takelma (Penutian), Bowern
(2004) for Bardi (Nyulnyulan), Donohue (2004) for Tukang Besi, and Jaeger
and Gerassimova (2002) and Jaeger (2004) for Bulgarian (Slavic), among others. The status of object markers in a number of Bantu languages is debatable
(see, for example, Seidl and Dimitriadis 1997 and Morimoto 2002 on Swahili,
and Woolford 2001 on Ruwund), so we will not discuss such languages unless
we can demonstrate that the marking on the verb is actually agreement marking and not pronominal incorporation (see Bresnan and Mchombo 1987 and
Chapter 2, Section 2.6 for detailed discussion of this issue).
We also limit our scope to the examination of nominative-accusative languages, and have little to say about languages with an ergative-absolutive casemarking system, where the object is absolutive. Malchukov (2006) notes that
DOM is more typical of nominative-accusative systems, and proposes a principle of argument marking that holds of both types of languages: languages
tend to avoid manipulation of the casemarking of the unmarked argument, i.e.
the nominative argument in nominative-accusative systems, and the absolutive
argument in ergative-absolutive systems. Deemphasis of the absolutive object
tends to give rise to voice alternations such as antipassive, rather than case alternations. We do not take a stand here on whether this view is correct, but the
fact remains that DOM is more frequent in nominative-accusative languages,
and we will concentrate on such languages in this book.
1.5 Structure of the book
The book is structured as follows.
Chapter 2 provides a brief introduction to Lexical Functional Grammar (LFG)
(Bresnan 2001, Dalrymple 2001, Falk 2001), the syntactic framework employed in this book. Chapter 3 presents an overview of the model of information structure which we assume, and in Chapter 4 we present our proposal
for the treatment of information structure in LFG.
In Chapter 5 we discuss the expression of primary topics in syntax, suggesting that although there is no unique alignment between information structure
roles and grammatical functions, there are important cross-linguistic tendencies in the grammatical expression of primary topics.
Chapter 6 addresses languages in which grammatical marking for nonsubject topics occurs with a variety of syntactic roles: it is possible for objects,
possessors, instruments, and other syntactic functions. Comrie (2003) discusses agreement patterns in such languages under the rubric of trigger-happy
agreement, and we show that casemarking can work in a similar manner.
These data constitute the main evidence for the relevance of topichood in the
grammatical marking of nonsubjects.

Structure of the book

21

Differential object agreement and differential object casemarking are dealt

with in Chapter 7, where we examine languages in which grammatical marking
of nonsubject topics is syntactically restricted to objects. In such languages, (a
subset of) topical objects receive special agreement or casemarking, while (a
subset of) nontopical/focused objects remain unmarked.
We then take a closer look at syntactic differences between marked and unmarked objects in languages in which grammatical marking is confined to objects, and discuss cross-linguistically typical patterns of alignment between
syntax and information structure. We show in Chapter 8 that in some such languages, a number of syntactic tests distinguish marked objects and unmarked
objects, with the marked (topical) object displaying more properties associated with core grammatical functions. Chapter 9 suggests some generalisations
concerning ditransitive constructions in languages with DOM.
Chapter 10 proposes a historical explanation for semantic patterns of DOM,
which can arise as a result of different directions of grammaticalisation of nonsubject topic marking. Since features such as definiteness, specificity or animacy are known to be associated with topics, grammatical marking may spread
to nontopical objects with these semantic features, or may narrow to include
only topical arguments which also have these features.
Chapter 11 concludes the book.

Syntactic assumptions
Nontransformational, constraint-based theories of grammar such as Construction Grammar (Croft 2001, Goldberg 2006), HPSG (Sag et al. 2003), Role and
Reference Grammar (Van Valin 2003), Simpler Syntax (Culicover and Jackendoff 2005), and Lexical Functional Grammar (Bresnan 2001, Dalrymple
2001, Falk 2001) represent different aspects of the structure of an utterance
as separate but related grammatical modules. Such theories assume that syntactic structure is related to semantics, information structure, and other linguistic levels not by means of transformational operations, but by constraints
involving one or more levels of structure. Information structure roles may be
associated with particular phrase structural positions, but these positions do not
define the roles (as they often do in tree-based, transformational theories such
as Principles and Parameters Theory or Minimalism: Chomsky and Lasnik
1993, Chomsky 1995). Instead, the relations between grammatical, semantic,
and information structural roles are specified in terms of constraints involving
different levels of linguistic representation.
In LFG, these different aspects of linguistic structure phrase structure,
grammatical functions, information structure are represented by structures
that may be of different formal types. The phrasal structure of the sentence is
represented by a phrase structure tree, the constituent structure or c-structure.
Grammatical functions like subject and object are represented by the functional structure or f-structure. Information structure (Choi 1999, Butt and
King 1996, 2000) is related to other grammatical levels within the projection
architecture of LFG (Kaplan 1987, Asudeh 2006). LFG provides the tools
needed to analyse the relation between grammatical functions and information structure roles and to provide a formal treatment of the grammar of object
marking and its relation to semantics and information structure.

23

24

Syntactic assumptions

2.1 Grammatical functions

A central tenet of LFG theory is the dual nature of syntactic structure. The abstract grammar of all languages is organised around syntactic relations like
subject, object, and adjunct: all languages distinguish arguments from adjuncts, and subjects from nonsubjects, for example. These abstract grammatical relations are overtly expressed in phrasal structure in very different ways
in different languages. Some languages have rigidly fixed word order, while
others allow a great deal of freedom in word order, and languages vary in
terms of where the verb must appear in relation to its arguments. The level at
which word order and phrasal constituency is represented has its own internal
organisational principles, different from the level at which abstract grammatical relations are represented. The two syntactic structures of LFG, constituent
structure and functional structure, reflect this duality.
2.1.1 Grammatical function diagnostics
LFG assumes that abstract grammatical functions such as subject, object, complement, and adjunct are theoretical primitives, not defined in terms of other
levels of structure. That is, the subject of a sentence is not defined as a phrase
appearing in a particular phrase structural position or as an entity bearing a
particular semantic role; instead, subject, object, and other grammatical functions are basic concepts of the grammar, around which various grammatical
properties tend to cluster (Falk 2006).
In LFG, diagnostics identifying particular grammatical functions appeal to
abstract behavioural grammatical properties defined at functional structure. In
many languages, for example, the subject is the argument which agrees with
the verb, or is the antecedent of a reflexive, or is absent in a subordinate clause
under identity with a matrix clause argument. Mohanan (1994) shows that
binding relations for pronouns in Hindi are constrained by the grammatical
function of the antecedent. The antecedent of a pronoun cannot be the subject
of its clause:
(1) Vijay ne Ravii ko uskii saikil par bit.haayaa
Vijay Erg Ravi Obj his bicycle Loc sit.Caus.Perf
Vijayi seated Ravij on hisi,j bike.

(Mohanan 1994:126)

Matsumoto (1996) discusses honorification in Japanese, showing that some

verbs take special forms to honor the subject referent, and not other grammatical functions. The verb form o-V ni naru is used to honor the subject sensei
teacher, and cannot be used to honor a nonsubject:

Grammatical functions

25

(2) a. sensei wa hon o o-yomi

ni narimashita
teacher Top book Acc Honorific-read Cop become-Polite-Past
The teacher read a book.

(Matsumoto 1996:27)

b. *Jon wa sensei no o-tasukerare ni natta

John Top teacher by Honorific-help Cop become-Past
John was saved by the teacher.

(Matsumoto 1996:28)

None of these properties is definitional of subjecthood, and indeed all of these

properties have been found to be associated with nonsubjects as well as subjects, depending on the language. In fact, these properties tend to target the top
of the grammatical function hierarchy in (3), originally proposed by Keenan
and Comrie (1977), and so distinguish subjects from other arguments in some
languages, and core arguments from noncore arguments, or arguments from
adjuncts, in other languages.
(3) subject > object > oblique > adjunct
Because the subject is the argument that stands at the head of the grammatical
function hierarchy, it is often the only grammatical function that can participate
in certain processes, and can be reliably identified in this way.
Diagnostics targeting nonsubject grammatical functions, specifically objects,
also vary from language to language. For example, we have seen that some
languages have object agreement, which can function as a test for objecthood,
and some languages have a special relativisation strategy for objects that is
different from the strategy used for other grammatical functions.
Diagnostics for grammatical functions that are stated in terms of nonsyntactic criteria are more difficult to use, and less reliable. Unlike transformational
theories, LFG does not assume that grammatical functions are defined in terms
of positions in a phrase structure tree. Thus, in many languages the appearance of a phrase in a particular position is not criterial for identification of the
phrase as bearing a particular grammatical function, although in some fixed
word order languages, a particular phrase structure position may be reserved
for a particular grammatical function. Grammatical function alternations are
also unreliable tests; it is often claimed that objects can be identified as those
arguments of an active verb which appear as the subject of the corresponding
passive verb, but although this is often true, there are passive subjects which
are not objects of the active verb (Bill is said to be available/*Someone said
Bill to be available). For more discussion of grammatical function diagnostics,
see Andrews (1985) and, for objects, Hudson (1992); for an LFG perspective,
see Dalrymple (2001: Chapter 2) and, for objects, Borjars and Vincent (2008)
and Lam (2008).

26

Syntactic assumptions

2.1.2 Grammatical functions and semantic roles

Predicates encode an association between grammatical functions and particular semantic roles: for example, the active version of the verb kick associates
its subject with the semantic role of kicker, and its object with the semantic
role of the entity that is kicked (The man kicked the ball). Mapping theory
(Levin 1986, Bresnan and Kanerva 1989, Butt 1995, Alsina 1996) is a theory
of the relation between argument structure (the level at which semantic roles
are represented) and functional structure (the grammatical functions associated
with those roles).
Researchers in mapping theory agree that argument structure encodes a certain amount of semantic information about the arguments of a predicate, but
different researchers have adopted different views of the content and representation of argument structure. Much research is based on the assumption that
argument structure makes available a set of abstract semantic roles like agent,
source, theme, patient, and goal, which are ranked relative to one another on a
thematic hierarchy. Bresnan and Kanerva (1989) propose the thematic hierarchy shown in (4):
(4) Thematic hierarchy (Bresnan and Kanerva 1989):
AGENT

> BENEFACTIVE > RECIPIENT / EXPERIENCER

> INSTRUMENT > THEME/PATIENT > LOCATIVE

These semantic roles are systematically related to grammatical functions (see,

for example, Bresnan and Kanerva 1989 and Bresnan and Zaenen 1990): on
this view, mapping theory relates the agent role (kicker) of kick to SUBJ, and
the patient role (object kicked) to OBJ. Other versions of mapping theory adopt
a different view of argument structure, often in an attempt to define or provide a clearer characterisation of semantic roles. Butt (1995) assumes Jackendoffs (1983, 1990) Conceptual Semantics representation of argument structure, which involves the use of a set of conceptual primitives such as CAUSE,
GO, and BE to encode basic semantic relations. Alsina (1996) works within
a version of the proto-role theory of Dowty (1991), which provides a set of
entailments classifying the arguments of a predicate according to the degree to
which they exhibit proto-agent or proto-patient properties. Our theory of argument marking does not depend on the details of these theories of argument
structure, and so we will appeal when necessary to semantic roles like agent,
patient, theme, source, and goal, without presupposing any particular theory of
their representation or definition at argument structure.
Whatever theory of argument structure is adopted, all LFG researchers agree
that mapping theory must include several very general constraints on the relation between argument structure and functional structure. The first is functionargument biuniqueness (Bresnan 1980), which requires each semantic role to

Objects

27

correspond to exactly one grammatical function, and no grammatical function

to correspond to more than one semantic role. For example, there are no predicates whose argument structure contains two different semantic roles (an agent
and a theme, for example) with both roles mapped to a single grammatical
function, and there are no predicates whose argument structure contains a single semantic role mapped to two different grammatical functions:
(5) Function-argument biuniqueness:
Argument structure:

ALLOWED:
2
1

Functional structure:

GF1

GF2

DISALLOWED:
1
2
GF1

DISALLOWED:
1
GF1

GF2

Another general constraint is the subject condition, which stipulates that a

verbal predicate must have a subject. This is a more controversial condition
which has been challenged by some researchers (for discussion, see Berman
1999), and it may be that the subject condition holds in only some languages
rather than being a universal principle of mapping theory. Though (as far as we
know) the subject condition holds in all of the languages that we will discuss,
we will not take a position on the universality of the subject condition in this
work.
2.2 Objects
Since our focus is the differential marking of objects, we will be most concerned with object grammatical functions. Here, we provide a brief overview
of the standard LFG view of objects. Recent LFG work has explored the object
functions from a variety of perspectives: see Lam (2008) for a thorough discussion of object functions with a focus on constructions with multiple objects,
and Borjars and Vincent (2008) for an LFG-based perspective that differs from
the traditional view in interesting ways, particularly with regard to the analysis
of the English double object construction. We return to a discussion of the
object functions and object mapping in Chapter 9.
LFG assumes that more than one object function can be manifested in a
language. In early LFG, these object functions were called OBJ (object) and
OBJ2 (second object) (Bresnan 1980, Kaplan and Bresnan 1982). In a double
object construction, the object which shares more properties with the object
of a monotransitive verb was called the (primary) object, OBJ, and the other
argument was called the second object, OBJ2.1
1 Dryer (1986) uses the terms primary object and secondary object to draw nearly the same
distinction; see Chapter 9 for further discussion of LFGs object functions in double object constructions.

28

Syntactic assumptions

In subsequent work, Bresnan and Kanerva (1989) proposed to characterise

grammatical functions in terms of a decomposition of features2 +R and +O,
where +R(estricted) indicates whether the grammatical function must correspond to a particular semantic role (+ R) or may correspond to any semantic
role (R), and +O(bjective) indicates whether the grammatical function is objective (object-like: + O) or not ( O). On this view, grammatical functions are
cross-classified as shown in (6):
(6)
O
+O

+R

SUBJ

OBL

OBJ

OBJ

(Bresnan and Kanerva 1989:25)

The subject SUBJ and unrestricted/primary object OBJ can correspond to any
semantic role (R), while the (+ R) grammatical functions OBJ and OBL must
correspond to particular semantic roles. OBJ stands for the family of objectlike (+O) functions OBJ THEME , OBJ LOC , and so on, while OBL stands for the
family of non-object-like (O) oblique functions OBL LOC , OBL GOAL , OBL AGENT ,
and so on. This decomposition recasts the OBJ/OBJ2 opposition in terms of a
distinction between OBJ and the family of grammatical functions abbreviated as
OBJ , restricted/secondary object functions that are associated with a particular
semantic role. Languages can vary as to which OBJ functions they express:
like many languages, English expresses only OBJ THEME (that is, the OBJ2 of
previous LFG work is now treated as a restricted OBJ which is required to bear
the semantic role of theme), while other languages may have a wider range of
semantically restricted objects.
With the distinction between primary/unrestricted OBJ and secondary/restricted
OBJ in hand, Bresnan and Moshi (1990) examined languages like Kichaga,
which allow promotion of oblique arguments to objects via applicativisation,
as in examples like (7):
(7) a. n-a--ly-`a
k-ely`a
Foc-1Subj-Pres-eat-FinalVowel 7-food
He is eating food.
b. n-a--ly`--`a
m-k`
` a k-ely`a
Foc-1Subj-Pres-eat-Applicative-FinalVowel 1-wife 7-food
He is eating food for the benefit of/to the detriment of his wife.
(Bresnan and Moshi 1990:148)
2 This assumption entails that grammatical functions are not atomic, since they can be decomposed into simpler featural components; grammatical functions are still primitives of the theory,
in that they are not defined in terms of concepts from other levels of structure.

Levels of syntactic representation

29

Example (7a) shows the monotransitive verb eat with a single object, the
class 7 noun food. With the addition of the applicative affix, a benefactive/malefactive applied OBJ wife is added, and food becomes a secondary
OBJ THEME .
Bresnan and Moshi (1990) show that in Kichaga, applicative constructions
with more than two objects are possible, establishing conclusively that a simple
two-way distinction between OBJ and OBJ2 is not sufficient for grammatical
description. Consider example (8):
(8) n-a-l!e -ku-sh-k-kor.--`a
Foc-1Subj-Past-17Obj-8Obj-7Obj-cook-Applicative-FinalVowel
She/he cooked it with them there.
(Bresnan and Moshi 1990:151)
This example contains three object markers, representing a locative object, an
instrumental object, and a patient object. The grammatical structure of this
example cannot be analysed using only the grammatical functions OBJ and
OBJ2, since this example contains three object-like functions. Bresnan and
Moshi show that in (8), the instrumental OBJ is the unrestricted OBJ, while
the locative and patient arguments bear semantically restricted OBJ functions
OBJ LOC and OBJ PATIENT.
In much previous LFG literature, the OBJ/OBJ distinction was assumed to
play a role primarily or exclusively in the analysis of ditransitive constructions,
but we will see in Chapter 8 that the distinction is also relevent for monotransitive predicates. For discussion of other languages in which monotransitive
verbs can take either OBJ or OBJ , see Cetinoglu and Butt (2008) for Turkish
and Dahlstrom (2009) for Meskwaki (Algonquian).
2.3 Levels of syntactic representation
The following brief sketch presents the formal syntactic concepts that will be
relevant in our analysis of DOM; for a more complete introduction to LFG, see
Bresnan (2001), Dalrymple (2001), and Falk (2001).
The constituent structure tree or c-structure represents linear order and
phrasal grouping:

30

Syntactic assumptions

(9) David is smoking.

IP
I

NP
N

VP

is

V

David

smoking
This tree represents the word order and phrasal constituency of the words that
actually appear in the sentence. This tree is not the product of a movement
process or the input to any transformational operation; LFG does not assume
the existence of processes that modify or destroy grammatical structure.
The categories in the tree shown in (9) are standard X -theoretic, endocentric
categories. We have represented the name David as a NP, but since the internal structure of nominal phrases is not of particular concern here, we will not
distinguish between NP and DP.3 All phrase structure categories in English are
endocentric (that is, headed). Some languages also make use of the exocentric
category S in their phrase structure (Kroeger 1993, Bresnan 2001). According to Bresnan (2001:110), S is a non-headed category which is not subject
to normal X -theoretic constraints on c-structure configurations or the relation
between c-structure and f-structure. Not all languages use this category, and
many of those that do have relatively free word order. The Australian language
Warlpiri has notoriously free word order (Simpson 1991); Austin and Bresnan
(1996) give the following c-structure for a typical Warlpiri sentence:
3 See B
orjars et al. (1999) for discussion of functional categories and the NP/DP distinction in
LFG.

Levels of syntactic representation

31

(10) kurdu-jarra-rlu kapala maliki wajilipi-nyi wita-jarra-rlu

child-Du-Erg Pres dog.Abs chase-Nonpast small-Du-Erg
The two small children are chasing the dog.
(Austin and Bresnan 1996:225)
IP
I

NP
N

kurdu-jarra-rlu
child-Du-Erg

kapala
Pres

S
NP

NP

wajilipi-nyi
chase-Nonpast

maliki
dog-Abs

wita-jarra-rlu
small-Du-Erg

Unlike English, the specifier position of IP in Warlpiri is not associated with

the subject function, and any constituent can appear there. Since the daughters
of S can be of any category and can be reordered freely, all orders of words in
this sentence are acceptable, so long as the auxiliary appears in second position
(Simpson 1991, Austin and Bresnan 1996).
The functional structure or f-structure represents syntactic predicate-argument
structure as well as features like case, person, number, gender, tense, and aspect, as shown in (11). This is the level at which grammatical functions like
subject and object are represented.
The f-structure for the sentence David is smoking is given in (11). The fstructure for the subject David appears as the value of the feature SUBJ, and
contains three features: PRED, whose value is the main syntactic predicate
DAVID of the subject phrase; the person feature PERS with value 3; and the
number feature NUM with value SG, or singular. The f-structure for the entire sentence has the feature PRED with value SMOKE SUBJ, indicating that
the syntactic predicate SMOKE requires one argument, the SUBJ(ect). It also
has a TENSE feature with value PRES(ent), and a PROGRESSIVE feature with
value +. LFG researchers generally follow a convention of representing only
the f-structure information that is relevant for the point at hand, and so the fstructures we present will often contain only information that is necessary for
purposes of the current discussion, and will omit all other features.

32

Syntactic assumptions

(11) David is smoking.

SMOKE SUBJ

PRED

TENSE PRES

PROGRESSIVE +

PRED DAVID

SUBJ PERS 3

NUM

SG

LFG assumes a piecewise relation between levels of linguistic structure, so

that parts of one level of structure are related to the corresponding parts of
another level. For example, the parts of the c-structure tree that make up the
subject phrase are related to the SUBJ f-structure by a correspondence function
called . The constituent structure tree and functional structure for the sentence
David is smoking are given in (12), with the relation between nodes of the
c-structure tree and f-structures indicated by arrows.
(12) David is smoking.
IP

NP

N

VP

is

V

David

PRED

SMOKE SUBJ

TENSE PRES

PROGRESSIVE +

PRED DAVID

SUBJ PERS 3

NUM

SG

smoking
Formally, the arrows from c-structure to f-structure represent a function relating nodes of the c-structure to parts of the f-structure. In (12), the IP, I , I,
VP, V , and V nodes of the tree are all related to the f-structure for the entire
sentence, and the NP, N , and N nodes are related to the f-structure for the subject. This configuration adheres to the general principles governing the relation
between c-structure and f-structure presented by Bresnan (2001:102):
(13) a. C-structure heads are f-structure heads. [In (12), the IP, I , and I are
mapped to the same f-structure; the same is true for the VP, V , and
V, as well as for the NP, N , and N.]

Describing linguistic structures

33

b. Specifiers of functional categories are the grammaticalised discourse

functions DF (including SUBJ). [In (12), the NP in the specifier of IP
is mapped to the grammatical function SUBJ.]
c. Complements of functional categories are f-structure co-heads. [In
(12), the I and the VP are mapped to the same f-structure.]
Other linguistic levels are related to c-structure and f-structure in a similar
way, by means of piecewise functions from parts of one structure to parts of
another. These functions are called projection functions, and because of this,
the different levels are also called projections (Kaplan 1987, Asudeh 2006). In
Chapter 4, we will see how information structure is represented, and how it is
related to other linguistic levels by means of a projection function. For more
discussion of the c-structure/f-structure mapping and how it is constrained, see
Bresnan (2001), Dalrymple (2001), and Falk (2001).
2.4 Describing linguistic structures
In LFG, the grammar of a language is a complex description of possible grammatical structures and the relations between them. The phrase structure rules
of a language determine the acceptable constituent structure trees of the language (as originally proposed by McCawley 1968); annotations on those trees
define the f-structures that correspond to them and, as we will see in Chapter 4, other levels as well, including information structure. In the following,
we confine attention to c-structure and f-structure, and show how to define and
constrain the relation between them.
Consider the following portion of the c-structure and f-structure given in
(12):
(14)

IP

NP

SUBJ

[]

I

The rule in (15) licenses the c-structure configuration in (14), requiring it to

consist of a mother IP node and daughters NP and I :
(15) IP

NP I

We also require a specification of the relation between this phrasal configuration and the f-structure it corresponds to. In particular, IP and I must correspond to the same f-structure (that is, the I is the f-structure head of the IP),
and the NP corresponds to the SUBJ of that f-structure. We can impose this
requirement with the following annotated rule:

34

Syntactic assumptions

(16) IP

NP
(

SUBJ)=

I
=

In an annotated phrase structure rule, the up arrow refers to the f-structure

corresponding to the mother node, and the down arrow refers to the f-structure
of the daughter node that it appears on. Thus, the annotation ( SUBJ)= on the
NP means that the f-structure corresponding to the mother IP, , has a SUBJ feature whose value is the f-structure corresponding to the daughter NP, . In other
words, the NP that appears in the specifier of IP corresponds to the subject of
the IP. This rule adheres to principle (13b) governing the relation between cstructure and f-structure, and is correct for English, a configurational language
which associates the specifier of IP position with the subject function. In other
languages, the specifier of IP may be associated with other functional roles;
see Bresnan (2001) and Dalrymple (2001) for more discussion.
The annotation = on the I means that the f-structure corresponding to
the IP ( ) is the same as the f-structure corresponding to the I (). In other
words, the I is the f-structure head of the IP and shares all of its f-structure
information. This follows principle (13a) governing the relation between cstructure and f-structure, which requires c-structure phrases and their heads to
correspond to the same f-structure.
The remaining nodes in the phrase structure tree are annotated with =,
requiring them to have the same f-structure as their mother. This follows from
principle (13c), which requires the VP and the I to correspond to the same
f-structure, and from principle (13a), which requires a phrase and its head
daughter to correspond to the same f-structure.
(17) David is smoking.
IP
I
=

NP
(

SUBJ)=

N
=

I
=

VP
=

N
=

is

V
=

David

V
=

smoking

PRED

SMOKE SUBJ

TENSE PRES

PROGRESSIVE +

PRED DAVID

SUBJ PERS 3

NUM

SG

Describing linguistic structures

35

Lexical entries are annotated with information about the preterminal node that
dominates them: David specifies information about the preterminal N, is specifies information about the preterminal I, and smoking specifies information
about the preterminal V. Preliminary lexical entries for David, is, and smoking
are given in (18):4
(18) Preliminary lexical entries for David, is, and smoking:
David

is

(
(
(
(
(

smoking (

PRED)

= DAVID
=3
NUM ) = SG
PERS)

TENSE)

= PRES

PROGRESSIVE)
PRED)

=+

= SMOKE SUBJ

Adding this information to the tree gives us the following configuration, which
shows where the information in the f-structure is introduced and how it flows
through the tree:
(19) David is smoking.
IP
I
=

NP
(

SUBJ)=

N
=
N
=

I
=

VP
=

is
(

V
=

TENSE)

= PRES
( PROGRESSIVE) = +

PRED

SMOKE SUBJ

SG

V
=

TENSE PRES

PROGRESSIVE +

PRED DAVID

SUBJ PERS 3

NUM

David
( PRED) = DAVID
( PERS) = 3
( NUM) = SG

4 For expository

smoking
= SMOKE SUBJ

PRED)

purposes, the treatment of tense and aspect has been greatly simplified; among
the issues that we will not treat is the requirement for the verb following is to appear in present
participial form. See Butt et al. (1996) and Frank and Zaenen (2002) for discussion and analysis.

36

Syntactic assumptions

2.5 Grammatical agreement

The lexical entry for the verb is must specify subject agreement requirements:
is requires a third-person singular subject. We will see in later chapters that
agreement can depend on the interaction of a number of syntactic, semantic,
and information-structural factors. However, in many languages, including
English, grammatical agreement depends on purely syntactic factors, and can
be described in simple terms, as we now show.
English subject/verb agreement is handled by specifications placed by the
verb on features of its subject, as shown in the more complete entry for is in
(20):
(20) Lexical entry for is, incorporating subject/verb agreement specifications:
is

TENSE)

(
(
(
(

= PRES

=+
=3
SUBJ NUM) = SG
PROGRESSIVE)
SUBJ PERS)

According to this entry, the subject must be compatible with the third person
singular agreement features imposed by is. A partial tree and f-structure incorporating this lexical entry is:
(21)

IP

NP
(

SUBJ)=

I
=
I

is

TENSE

PRES

PROGRESSIVE +

PERS 3
SUBJ

NUM

SG

= PRES
( PROGRESSIVE) = +
( SUBJ PERS) = 3
( SUBJ NUM) = SG
(

TENSE)

Since the person and number values specified for the subject David are compatible with those specified by the verb is, David is an acceptable subject of
this verb. This is why the sentence David is smoking is acceptable, and has the
c-structure and f-structure shown in (12).
In contrast, the sentence They is smoking is ungrammatical. This is due to
incompatible specifications for the value of the NUM feature of the subject, as
indicated by the clashing values SG/PL in (22): they has plural (PL) number,
while is requires its subjects number to be SG.

Agreement and pronominal incorporation

37

(22) Ill-formed f-structure for *They is smoking.

IP
I
=

NP
(

SUBJ)=

N
=

I
=

VP
=

is
N
=

TENSE)

= PRES
( PROGRESSIVE) = +
( SUBJ PERS) = 3
( SUBJ NUM) = SG

they
( PRED) = THEY
( PERS) = 3
( NUM) = PL

V
=

PRED

SMOKE SUBJ

TENSE PRES

PROGRESSIVE +

PRED THEY

SUBJ PERS 3

NUM

SG / PL

V
=

PRED)

smoking
= SMOKE SUBJ

English verb agreement is particularly easy to characterise: the verb agrees

with the subject, regardless of its semantic features or information-structural
role. As we will see, agreement patterns in many languages are more complex and require reference to nonsyntactic levels of linguistic structure, to be
presented in Chapter 4.
2.6 Agreement and pronominal incorporation
Grammatical agreement is crucially different from pronominal object incorporation, though there are clear historical relations between incorporated pronouns and agreement morphology (Givon 1976, Greenberg 1977). Bresnan
and Mchombo (1987) analyse object marking in Chichewa
as what they call
anaphoric agreement, involving the incorporation of an object pronoun. Our
concern is differential object marking as manifested in grammatical agreement
morphology, not the distribution of incorporated object pronouns or clitics,
and so we summarise their analysis mainly to show how anaphoric agreement
is analysed in LFG and how it differs from grammatical agreement.
Chichewa
is a subject pro-drop language: a full noun phrase subject may
but need not appear. Example (23) has subject marking but no object marking.
An object phrase is required, but the subject phrase may be absent:

38

Syntactic assumptions

(23) a. njuchi zi-na-lum-a

alenje
bees 10Subj-Past-bite-Indic hunters
The bees bit the hunters.
b. zi-na-lum-a
alenje
10Subj-Past-bite-Indic hunters
They bit the hunters.

(Bresnan and Mchombo 1987:744745)

Bresnan and Mchombo propose that the verb zi-na-lum-a bit optionally specifies a pronominal subject, and also specifies that the subject must be of noun
class 10:
(24) zi-na-lum-a (( SUBJ PRED) = PRO )
( SUBJ NOUNCLASS) = 10
The equation ( SUBJ PRED) = PRO is optional, and need not be used in the
analysis of the sentence. When there is a full noun phrase subject, as in (23a),
the pronominal contribution provided by the verb is not used, since it would
clash with the specifications provided by the noun phrase subject. In this case,
the verbal morphology behaves as agreement marking: the agreement features
specified by the verb (here, NOUNCLASS) must match the features of the subject,
just as in the English example discussed in the previous section.
When no full noun phrase subject is present, as in (23b), the pronominal
contribution provided by the verb is required, since otherwise the sentence
would lack a subject. In this case as well, the subject is specified as noun class
10, but additionally the verb makes a pronominal contribution associated with
its subject. This is possible because of the optionality of the SUBJ PRED PRO
contribution of the verb. The functional structures for the examples in (23) are
given in (25):

Agreement and pronominal incorporation

(25) a. njuchi zi-na-lum-a

alenje
bees 10Subj-Past-bite-Indic hunters

PRED

39

BITE SUBJ, OBJ

PRED BEES
SUBJ

NOUNCLASS
10

PRED HUNTERS
OBJ

The bees bit the hunters.

NOUNCLASS

b. zi-na-lum-a
alenje
10Subj-Past-bite-Indic hunters

PRED

BITE SUBJ, OBJ

PRED PRO
SUBJ

NOUNCLASS 10

PRED HUNTERS
OBJ

They bit the hunters.

NOUNCLASS

Chichewa
transitive verbs may also show object marking. Bresnan and Mchombo
(1987) provide compelling evidence to show that in Chichewa
examples like
(26a) and (26b), the object agreement affix is actually an incorporated pronoun, unlike the subject agreement affix, which encodes either an incorporated
pronoun or grammatical agreement.

PRED BITE  SUBJ, OBJ

(26) a. njuchi zi-na-wa-luma

PRED BEES
bees 10Subj-Past-2Obj-bite

SUBJ

NOUNCLASS
10
The bees bit them.

PRED PRO

OBJ
NOUNCLASS

PRED

BITE SUBJ, OBJ

PRED HUNTERS
b. njuchi zi-na-wa-luma
alenje TOPIC

NOUNCLASS 2
bees 10Subj-Past-2Obj-bite hunters

PRED BEES
The bees bit them, the hunters.

SUBJ

NOUNCLASS 10

PRED PRO

OBJ
NOUNCLASS

When an overt noun phrase appears and is interpreted as the object, as in (26b),
it is in fact a floating topic phrase, anaphorically linked to the incorporated

40

Syntactic assumptions

pronominal object.5 Bresnan and Mchombo provide a number of tests showing that this is the correct analysis: for example, they show that the relation
between the floating topic phrase and the verb can be nonlocal, since it is the
incorporated pronoun and not the full noun phrase that is the syntactic object
of the verb. These tests allow us to distinguish between anaphoric agreement
(an incorporated pronominal argument) and grammatical agreement (requirement for matching of features such as number, gender, and noun class). The
Chichewa
verb in the examples in (26) is associated with the following lexical
specifications:
(27) zi-na-wa-lum-a ( PRED) = BITESUBJ, OBJ
(( SUBJ PRED) = PRO )
( SUBJ NOUNCLASS) = 10
( OBJ PRED) = PRO
( OBJ NOUNCLASS) = 2
In (27), the subject and object specifications differ: the predicate PRED of the
subject is optionally specified as pronominal by the verb, as indicated by the
parentheses around the specification in the second line of the entry, whereas
the PRED of the object is obligatorily specified. This difference gives rise to the
different behaviour of the subject and object.
In this study we analyse only grammatical agreement, not anaphoric agreement, and whenever possible we provide evidence that demonstrates grammatical and not anaphoric agreement for the constructions we analyse. For
further discussion of constraints on the information structure role of incorporated pronominal arguments, see Bresnan and Mchombo (1987), Culy (2000),
Jaeger and Gerassimova (2002), Jaeger (2004), Bowern (2004), and references
cited there.
2.7 Casemarking
Casemarking has been a major focus of study within LFG, beginning with
Nordlingers (1998) groundbreaking theory of constructive case, and continuing with work by, among others, Butt and King (1999, 2003) and Butt (2008b).
The constructive case view is that case signals grammatical function, and in
fact constructs the f-structure environment in which a phrase is permitted to
appear.
Nordlinger (1998) provides the following treatment of casemarking on the
ergatively-marked noun alanga-ni girl-Erg in Wambaya (Australian):
5 We

discuss the TOPIC role in functional structure and information structure in Chapter 4.

Casemarking

41

(28) alanga-ni ( PRED) = GIRL

( PERS) = 3
( GEND) = FEM
( CASE) = ERG
(SUBJ )
The specification (SUBJ ) is different from the equations introduced earlier, in
that it uses inside-out functional uncertainty (Kaplan 1988) to specify the
environment in which this phrase must appear. These specifications require
alanga-ni to appear in this configuration:
(29)

SUBJ

PRED

GIRL

CASE

ERG

PERS 3

GEND FEM

The first four lines in the lexical entry in (28) specify values for the features
PRED, PERS, GEND, and CASE, as in the English examples discussed in Section 2.3. The specification (SUBJ ) in the fifth line is different: the feature
SUBJ appears before the f-structure designator , rather than after it, meaning
that the f-structure must appear as the SUBJ within some larger f-structure.
The following Wambaya sentence satisfies the requirements in (28), since
alanga-ni is the subject of the sentence:
(30) alanga-ni ngiy-a
dawu darranggu
girl-Erg 3SgFemSubj-Past bite tree.Acc
The girl chopped the tree.

(Nordlinger 1998:85)

This sentence has the following c-structure tree:

(31)

IP
I

NP
N

alanga-ni
girl-Erg

ngiy-a
3SgFemSubj-Past

S
V

NP

dawu
bite

darranggu
tree.Acc

42

Syntactic assumptions

The relevant lexical entries are the following, adapted from Nordlinger (1998:85):
(32) alanga-ni ( PRED) = GIRL
( PERS) = 3
( GEND) = FEM
( CASE) = ERG
(SUBJ )
ngiy-a

dawu

(
(
(
(
(
(

TENSE)

= PAST
=3
SUBJ GEND) = FEM
SUBJ NUM) = SG
OBJ PERS) = 3
SUBJ PERS)

PRED)

= CHOPSUBJ, OBJ

darranggu ( PRED) = TREE

( PERS) = 3
( CASE) = ACC
(OBJ )
The lexical entry for alangani contains a specification for PRED, PERS, CASE,
and GEND; it also contains the constructive case specification (SUBJ ), which
requires alanga-ni to bear the grammatical function SUBJ within its clause.
The auxiliary ngiya specifies tense as well as subject and object agreement:
the subject must be third person feminine singular, and the object must be
third person. The transitive verb dawu simply requires a SUBJ and an OBJ. The
accusative noun darranggu specifies PRED, PERS, and CASE for its f-structure,
and also contains the specification (OBJ ), which requires it to appear as the
OBJ of its clause. These requirements are all met in the f-structure for this
sentence:
(33)

PRED

CHOPSUBJ, OBJ

TENSE PAST

PRED GIRL

CASE ERG

SUBJ
PERS 3

NUM SG

GEND FEM

PRED TREE

OBJ PERS 3

CASE

ACC

Nonsyntactic critera for casemarking patterns

43

In Wambaya, the distribution of case is governed purely syntactically, marking arguments as bearing the SUBJ or OBJ role within their clause. It is clear,
however, that other levels of linguistic structure also govern the distribution of
case, and we now turn to a proposal to handle this within the constructive case
paradigm.
2.8 Nonsyntactic critera for casemarking patterns
The foregoing analyses of agreement and case depend only on syntactic features appearing within the f-structure: person, number, gender, and noun class
features for agreement, and grammatical functions such as SUBJ and OBJ for
constructive case. Nonsyntactic criteria may also be involved in grammatical
marking; indeed, the main theme of our book is how information structure roles
can influence grammatical marking. Here we outline one influential proposal
for handling patterns in which nonsyntactic features influence casemarking.
Butt and King (2003) present a constructive analysis of case which crucially incorporates the traditional distinction between semantic case (where
casemarking depends on or contributes some semantic feature) and structural
case (which depends purely on grammatical function, as in the Wambaya example discussed above); see Butt (2008b) for extensive discussion of the distinction between structural and semantic case and its role in the analysis of
casemarking within LFG. Butt and King provide a constructive case analysis
of Georgian (Kartvelian), in which semantic case is dependent on semantic
inferences over parameters such as volitionality (Butt and King 2003) as well
as grammatical function.
Case patterns in Georgian depend on the tense and aspect of the clause; subjects of transitive aorist verbs are marked with ergative, and objects of present
tense verbs are marked with accusative. Example (34) has a transitive aorist
verb, and so the subject is marked with ergative case, and the object is marked
nominative:
(34) nino-m Ceril-i
daCera
Nino-Erg letter-Nom wrote.3SgSubj.3Obj
Nino wrote a letter.

(Butt and King 2003)

Butt and King (2003) provide the following entry for the ergative case ending:
(35) -m(a)

( CASE) = ERG
(SUBJ )
(EXT- ARG arg-str )
((SUBJ ) TENSE - ASPECT ) = AORIST

This entry specifies that the noun it is suffixed to:

44

Syntactic assumptions
has ergative case;
is the subject of its clause;
is the external argument at argument structure; and
is the subject of a clause whose tense/aspect is aorist.

The first two specifications concern the case and grammatical function of the
ergative noun, and are exactly the same as for ergative nouns in Wambaya,
described above.
The third specification concerns not c-structure or f-structure, but another
level of structure: Butt and King use the notation arg-str to refer to the argument structure representation of the ergative noun, and the specification
(EXT- ARG arg-str ) to require the noun to bear the argument structure role of
external argument (EXT- ARG), in exactly the same way that the specification
(SUBJ ) specifies that the phrase must bear the subject role at f-structure. This
means that ergative case in Georgian is semantic case, specifying not only
a particular f-structure role (subject) but also a particular argument structure
role (external argument). The following configuration shows the c-structure, fstructure, and argument structure that are required in Butt and Kings analysis
for the ergative noun nino-m:
(36)

C-structure:
N

F-structure
(preliminary):


SUBJ

Nino

PRED

N INO

CASE

ERG

Argument structure:


EXT- ARG

PRED

N INO

The fourth specification, ((SUBJ ) TENSE - ASPECT ) = AORIST, ensures that the
clause of which the ergative noun is the subject is aorist. The expression
(SUBJ ) refers to the larger f-structure within which the ergative noun is a
subject; this f-structure is required to have the value AORIST for the feature
TENSE - ASPECT . This results in the following requirements, taking into account
all four specifications for the ergative marker:
(37)

C-structure:
N

Nino

F-structure:

SUBJ

PRED
CASE

TENSE - ASPECT


N INO

ERG

Argument structure:


EXT- ARG

PRED

N INO

AORIST

We adopt the Nordlinger/Butt and King view of constructive case and their
distinction between semantic case and structural case; as we will see, we will

Conclusion

45

find it necessary to augment the treatment of semantic case to encompass information structure requirements as well as requirements at the level of semantic
structure or argument structure.
2.9 Conclusion
We now have our set of basic formal tools for the analysis of differential object marking in hand. Our main concern is the interaction between information
structure and grammatical functions, and so in the following we will be primarily concerned with f-structure and its relation to information structure, to
be defined and discussed in detail in Chapter 4. In the next chapter, we will
present our basic assumptions about information structure and its relation to
other linguistic levels.

Information structure in
grammar
Recent research on information structure addresses two main topics: the content of information structure, and the development of models of grammar
which account for the interactions among information structure, syntax, and
semantics. This chapter briefly deals with the first topic. It is not our goal here
to evaluate theories of information structure in detail (for thorough overviews,
see Vallduv 1992 and Erteschik-Shir 2007). Instead, we offer some observations on how information structure roles are defined in what we consider to be
the most articulated and coherent views of information structure, with the aim
of providing working definitions which can be applied in further analysis. The
second question, the place of information structure in our theory of grammar,
will be addressed in Chapter 4.
3.1 The content of information structure
We view exchange of information as the main function of language. Information structure is the level of sentence organisation which represents how
the speaker structures the utterance in context in order to facilitate information
exchange. Specifically, it indicates how the propositional content of an utterance fits the addressees state of knowledge at the time of utterance. In human
communication, new information is normally added to the already existing
sum of knowledge in the addressees mind. The distinction between familiar
knowledge and the informational contribution of an utterance is manifested
linguistically: propositions can receive different formal expression (are packaged) in accordance with what the speaker assesses as old or new information
for the addressee.
47

48

Information structure in grammar

Vallduv (1992), Vallduv and Engdahl (1996) and Erteschik-Shir (1997,

2007) view information structure as codifying information update. Following
Heim (1982) and Reinhart (1982), Vallduv and Engdahl (1996) and ErteschikShir (2007) understand information states as file-like constructs. Files are collections of file cards corresponding to discourse referents. Each file card has
a number of records that contain information about the entity it denotes; these
records encode what is known about the discourse referent in question. Definite noun phrases correspond to already existing cards, while indefinite noun
phrases introduce new cards. Information structure is represented metaphorically in terms of instructions for manipulating the card files. The cards can
be positioned on top of the file, and their content can be altered or updated.
Different types of information structuring can be viewed as different types of
updating instructions. Like the grammatical relations encoded in f-structure,
these instructions and the information structure units associated with them are
considered universal theoretical primitives, independent of their actual linguistic realisation.
Lambrecht (1994) does not employ the file card metaphor; his theory is conceptualised in terms of the mental representation of discourse referents and
states of affairs that the speaker and the addressee have in their minds. As conceptual representations of states of affairs, propositions added to context are
paired with the speakers assumptions about the addressees state of knowledge and attention at the time of an utterance. Information structure mediates between sentence meaning and form by creating a pragmatically structured proposition which reflects these assumptions. Both morphosyntactic and
prosodic cues may be used to distinguish among such pragmatically structured
propositions.
Lambrecht emphasises that information is propositional in nature and that
by communicating with the addressee, the speaker influences the addressees
mental representation of the world. This representation consists of a set of
propositions, old information, which the addressee knows or believes at the
time of speech. New information consists of propositions (not referents) of
which the addressee does not have a mental representation.
A proposition may in principle be structured into the pragmatic presupposition and the pragmatic assertion. The pragmatic presupposition is old
information: that is, a set of propositions which the speaker assumes that the
addressee knows at the time of the utterance. The pragmatic assertion is new
information, the proposition which the addressee is expected to learn as a result
of hearing the sentence. All utterances express a pragmatic assertion in order
to be informative, but the pragmatic presupposition may be absent. Obviously,
pragmatic structuring is dependent on linguistic and extralinguistic environment: propositions undergo structuring according to the discourse situations

Information structure roles

49

in which they appear. For this reason, we consider sentences not in isolation,
but provided with a context, where possible.
3.2 Information structure roles
Following Lambrecht (1994), we adopt a distinction between the pragmatic
presupposition and the pragmatic assertion, which forms the basis of our definitions of the units of information structure. These concepts underpin our
definitions of focus and topic.
3.2.1 Focus
In generative syntax and semantics, focus is often treated as a kind of opera Kiss 1995),
tor expressing exhaustiveness (Szabolcsi 1981), identification (E.
contrastiveness (Rooth 1992, Fery and Krifka 2008) or the like. This quantificational approach to focus contrasts with the informational approach we
adopt here. Vallduv (1992) and Vallduv and Engdahl (1996) describe focus
as the informative, newsy and contrary-to-expectation part of the proposition
the actual information update potential of a sentence. As such, focus is opposed to ground, the noninformative, known or predictable part. Focus is a
relational notion in the sense that it is not the focus referent itself that is necessarily new for the addressee, but the fact that it participates in the proposition
conveyed by the sentence and fills the informational gap between assertion
and presupposition in a given communicative context. Focus can therefore be
defined as the semantic component of a pragmatically structured proposition
whereby the assertion differs from the presupposition (Lambrecht 1994:213).
The new information conveyed by a sentence (the pragmatic assertion) is the
relation between the focus and the presuppositional part of the proposition.
Since the focus corresponds to an informationally unpredictable part of the
proposition, it must be overtly expressed by one or several sentence elements.
Some syntactic forms are explicit markers of focus; for example, wh-questions
normally target so-called narrow focus or argument focus, which extends over
one (or, in the case of multiple wh-questions, more than one) participant in the
event, as in (1).
(1) a. What is Bill eating?
b. He is eating pizza.
The rough information structure representation of the answer is as follows:
(2) a. pragmatic presupposition: Bill is eating X
b. pragmatic assertion: X = pizza

50

Information structure in grammar

c. focus: pizza

In contrast to narrow focus, wide focus or predicate focus serves to augment

information about a particular referent, as in (3).
(3) a. What is Bill doing? or What about Bill?
b. He is eating pizza in the kitchen.
The representation of (3b) is as follows:
(4) a. pragmatic presupposition: Bill is doing X
b. pragmatic assertion: X = eating pizza in the kitchen
c. focus: is eating pizza in the kitchen
The focus here does not extend over a single participant as in (1b), but instead
over the whole verb phrase. Wide focus corresponds to the traditional notion
of comment. Lambrecht (1994) refers to its syntactic expression as the focus domain, suggesting that every constituent within this domain bears focus
status.
Lambrecht (1994) also distinguishes a third type of focus structure, sentence
focus (sometimes called presentational, news or thetic focus), in which the
entire sentence is focused, as, for example, an answer to a question like What
happened?. The sentence focus type corresponds to linkless structures in the
terminology of Vallduv and Engdahl (1996), or to stage topic structures
in the terminology of Erteschik-Shir (2007), discussed below. Since we are
primarily interested in the status of objects as topic or focus, we will not discuss
sentence focus in detail; neither will we address so-called verb/polarity/verum
focus, restricted to the verb alone.
3.2.2 Topic
Following much previous research (Reinhart 1982, Gundel 1988, Lambrecht
1994, and others), we understand topic as the entity that the proposition is
about. We essentially adopt the following definition from Gundel (1988:210):
An entity, E, is the topic of a sentence, S, iff in using S the speaker
intends to increase the addressees knowledge about, request information about, or otherwise get the addressee to act with respect
to E.

Information structure roles

51

This definition reflects the widespread intuition that utterances normally contain some (known) elements about which the speaker wants to convey new
information to the addressee.
Strawson (1964) in fact suggests that the truth of a statement must be assessed as putative information about its topic. According to Erteschik-Shir
(2007:15), this condition implies that all sentences must have topics, since
all sentences must be assigned a truth-value. This leaves topicless sentencefocus (thetic) sentences unaccounted for. To resolve this contradiction, ErteschikShir introduces the notion of stage topic, absent from Lambrechts model.
Stage topics do not correspond to a referent, but instead describe a situation (time and place) about which the proposition is asserted. These spatiotemporal parameters are contextually defined, but do not necessarily have overt
lexicogrammatical expression; see Erteschik-Shir (2007:1617) for more discussion. However, since we will not discuss or analyse sentence-focus/thetic
utterances in the following, the terminological difference between topicless
structures and structures that involve stage topics is not relevant to us. The
sentences of interest to us are those which have topic-focus structures.
Our notion of topic roughly corresponds to the concept of link proposed
by Vallduv (1992) and Vallduv and Engdahl (1996). Informationally, link
functions as a locus of knowledge update at an address in a card file: it picks
out a specific card where information update is to be carried out, and therefore
an entry under which a new proposition is to be classified. However, as noted
by Erteschik-Shir (2007:11), link is actually defined as a command to switch
to a particular existing address, so it is only involved in a change of address. In
other words, the notion link only describes shifted topics and does not apply
to continuous topics. Erteschik-Shir (2007:44) provides a description of topic
in terms of the cognitive category of attention: if the attention of the addressee
is drawn to a certain referent, then the addressee can be thought of as selecting
the corresponding file card and placing it on top of the file. All existing file
cards are potential topics in the discourse, but the actual topic is located on top
of the file.
We will not treat topic in terms of attention, as we believe that this cognitive
category may extend equally well to the category of focus. Since we do not
use the Heimian card file metaphor either, for us topichood is defined directly
on referents. Like focus, topic is a relational notion: it involves an aboutness
relation between a referent and a proposition. This relation holds if the speaker
assumes that the addressee considers a referent salient enough to be a potential locus of predication about which the assertion can be made. This means
that topic inhabits the presuppositional part of the proposition: it is associated with the pragmatic presupposition of existence, and sentences with topics
pragmatically presuppose that a referent is taken by the interlocutors to be the
centre of current interest in the conversation. This is what Lambrecht (1994)

52

Information structure in grammar

calls topicality presupposition or relevance presupposition. For example,

the topic of the second sentence in (3) (repeated as 5) is the referent of the pronoun he, the individual Bill, because the sentence is construed as information
about Bill. At the time the sentence is produced, the speaker assumes that the
addressee takes Bill to be pragmatically important and can expect to be given
more information about him.
(5) a. What is Bill doing? or What about Bill?
b. He is eating pizza in the kitchen.
This example shows that a potential (though perhaps not universal) diagnostic
for topichood is the what-about (or tell-about or as-for) test first discussed by Reinhart (1982) and Gundel (1988).
As should be clear from this discussion, we understand topic as a part of
sentence grammar rather than a continuous discourse notion. The relationship
between sentence topic and discourse topic is yet to be explored.
3.2.3 Topicality and topic-worthiness
A great deal of research has been devoted to the relationship between topicality and the referential properties of a noun phrase (see, for example, Reinhart
1982, Vallduv 1992, Lambrecht 1994 and references cited therein). An important distinction is to be made between topicality and what has been referred to
as topic-worthiness (Comrie 2003) or natural topicality (Croft 1991). Topicality reflects the informational status of a referent, and has to do with the pragmatic relation that holds between the referent and the proposition. It depends
on the speakers construal of the situation within the given communicative context, rather than on the noun phrases referential properties. The definition of
topic relies on the speakers assumptions about the addressees state of interest with respect to a referent, and does not automatically involve the speakers
assumptions about the addressees familiarity with a given referent (identifiability of a referent), expressed by the grammatical category of definiteness.
In contrast, topic-worthiness is measured in terms of the sorts of semantic features discussed in Chapter 1: features such as definiteness, specificity,
person, animacy, and humanness. They are sometimes called prominence features. These features tend to characterise topics, and indeed most topics are
definite and animate, the most preferred topics being first and second person
pronouns denoting speech act participants. The interaction between topicality
and animacy has been addressed by Givon (1983b), Dahl and Fraurud (1996),
Yamamoto (1999), Leonetti (2003) and de Swart (2007), among others, and
we do not have much to add. In a nutshell, these works argue that animates
are more often topical than other referents because we tend to think of the

Information structure roles

53

world as organised around animate beings which perceive and act upon their
inanimate environment (Dahl and Fraurud 1996:160). That topics tend to be
definite is also commonly known.
However, the correlation between topicality and topic-worthiness is imperfect. First, not all topic-worthy noun phrases are topics, but only those that
are sufficiently salient or relevant (bear a topicality presupposition). Second, although topics tend to be discourse-old, definite, and human, indefinites
and nonhuman referents are not excluded as potential topics, as long as their
referents have a certain pragmatic status for the interlocutors.
The nature of the interaction between topicality and degrees of identifiability of referents has been much discussed. Lambrecht (1994:165171) suggests
the topic accessibility scale, according to which the most acceptable topic
expressions are those whose referents are highly activated in the discourse or
the situation of speech. These often correspond to an unaccented pronominal
or a referential null. Next on the accessibility scale are accessible referents,
which are clearly identifiable by the interlocutors and therefore correspond to
a definite noun phrase. So-called unused referents are also identifiable, but
they are inactive in the discourse and incur a higher cognitive cost when interpreted as the centre of predication. Their accessibility as topics (as well as
their encoding as definite noun phrases) varies greatly among languages and
types of discourse. Brand-new referents are unidentifiable for the addressee
when new information is conveyed about them, and this explains why indefinite noun phrases are unlikely topics.
Yet indefinite topics are not completely excluded. If referential indefinites
are pragmatically anchored (linked) to another identifiable entity in the consciousness of interlocutors, they can be interpreted as topics. Lambrecht calls
such entities brand-new anchored and illustrates his point with the following
contrast:
(6) a. *A boy is tall.
b. A boy in my class is tall.
In (6a) the subject is brand-new but unanchored. The sentence is unacceptable
because it is difficult to imagine a context in which it would be informative to
predicate tallness on an unidentified subject referent. Such sentences violate
the most elementary condition of relevance (Lambrecht 1994:167). However,
acceptability depends on pragmatic factors; in example (6b) the subject is still
formally indefinite, but it corresponds to an brand-new anchored referent. The
additional PP in my class restricts the unspecified set of all boys to the set of
the boys in the speakers class, and therefore links/anchors the referent of the
indefinite subject a boy to the speaker herself. As a member of this smaller

54

Information structure in grammar

relevant set, the referent becomes more identifiable and more easily interpreted
as topic.
Pragmatically anchored indefinites in Lambrechts sense are roughly the
same as specific indefinites as defined by Enc (1986), Portner and Yabushita
(2001), and other authors. This understanding of specificity, based on the idea
of discourse linking, seems to be the most widely accepted (though see Farkas
1995 for a discussion of alternative views). Specificity involves a weak link
to a previously established referent, where a weak link is defined in terms of
a recoverable relation or a subset relation. The pragmatic link ensures that the
referent of the specific indefinite expression is identifiable to the speaker (but
not to the addressee). All definite phrases and some referential indefinites are
specific in this sense.
Another context that makes an indefinite NP specific and topicalisable is
relativisation. Erteschik-Shir (2007:89, 5253) shows, based on patterns of
topicalisation in Danish, that modifying relative clauses render indefinites specific. In Danish it is possible to topicalise an indefinite object if it is modified
by a relative clause, but is impossible to topicalise a nonmodified indefinite
object:
(7) a. E pige som jeg mdte i gar
gav jeg en god bog.
a girl that I met yesterday gave I a good book
I gave a good book to a girl that I met yesterday.
b. *En pige mdte jeg i gaar
a girl met I yesterday
I met a girl yesterday.

(Erteschik-Shir 2007:8)

In (7a), topicalisation by left dislocation is licensed by the fact that the object girl has a specific interpretation. Erteschik-Shir argues that the relative
clause indicates that the speaker has a particular referent in mind, unlike in
nonspecific indefinites. The relative clause in (7a) introduces the referent of
girl and therefore makes it old with respect to the main clause and a possible
candidate for topichood. In Erteschik-Shirs model, this amounts to saying that
relative modifiers cause a new card file to be opened for the indefinite noun and
therefore make it a suitable topic.
We will also see in subsequent chapters that ongoing discourse can play
a similar role in the topical interpretation of indefinites. In some languages
an indefinite noun phrase can be interpreted as topic if the following clause
(for instance, in coordinated constructions) adds more information about the
respective referent. In the absence of such clause, indefinite NPs cannot be
topics. The effect of the following context on topichood is still poorly understood (but see e.g. Portner and Yabushita 1998), but we will assume that it
works in the same way as the relative clause in (7a).

Information structure roles

55

These data indicate that specific indefinites may be interpreted as topics,

although not as easily as definite noun phrases. The situation is different for
nonspecific indefinites, since neither the speaker nor the addressee can identify
the referent; this amounts to there being no referent at all (and no corresponding card in the card file). In other words, topical status forces a specific interpretation of an indefinite noun phrase (see Erteschik-Shir 1997, 2007, Geurts
2002, Leonetti 2003, Portner and Yabushita 2001, and references cited there
for extended discussion).
It is often assumed that if a potential topic phrase has no referent, the statement cannot be evaluated as true or false (Strawson 1964, 1974; Reinhart 1982;
Gundel 1988; Lambrecht 1994:154156). Our definition of topic, as a referent
which the utterance is presupposed to be about, also ensures that nonreferential
expressions do not serve as topics. This requirement excludes expletive subjects and nonreferring quantified noun phrases as potential topics. However,
anaphoric pronominal elements coindexed with a nonreferring phrase may be
topical.
(8) a. I am looking for a husband.
b. He should be rich and handsome.
In (8b) the pronoun he does not refer to a real world entity, but is associated
with a discourse referent that can be identified from the previous context. The
addressee can establish that the assertion is made about this entity and assess
the truth-value of the proposition.
3.2.4 Secondary topic
The topic role is not necessarily unique. More than one referent can be under
discussion at the time of the utterance, so that the utterance simultaneously
increases the addressees knowledge about both of them. Of course, there are
cognitive limitations on the number of relevant entities under discussion in a
particular communicative act. This limitation has to do with the nature of human attention: we cannot keep many entities in the centre of attention and talk
about all of them at the same time. However, cross-linguistic grammatical evidence suggests that utterances can have at least two topics. Whether or not
three topics can appear in the sentence is uncertain (cf. Lambrecht 1981:73
74); for our purposes it suffices to distinguish the primary topic and the secondary topic (Givon 1984a,b, Polinsky 1995, 1998, Nikolaeva 2001).
Consider the discourse in (9), from Lambrecht (1994):
(9) a. Whatever became of John?
b. He married Rosa,

56

Information structure in grammar

c. but he didnt really love her.

(Lambrecht 1994:148)

In (9b) the subject is topical, and the utterance is not assessed to be about the
object referent (Rosa). This is an example of predicate-focus structure: the
utterance is construed as a comment about the topical referent John. From the
point of view of information structure, it can be represented as follows:
(10) a. pragmatic presupposition: John did X
b. pragmatic assertion: X = married Rosa
c. focus: married Rosa
Here the object Rosa is part of the focus domain. In (9c) the situation is different: although it is construed primarily as information about John, it also
increases the addressees knowledge about Rosa, namely, the fact that she was
not loved by her husband John. When (9c) is produced, both Rosa and John
are salient, under discussion, and pragmatically linked in the consciousness of
interlocutors. This pragmatic association between the two referents is established by the previous context, (9b). After (9b) is produced, the speaker can
assume that the addressee is familiar with the referent Rosa and with the relation between Rosa and John (John married Rosa), and that the addressee
can expect this relation to be commented on in further discourse. Thus, the
communicative purpose of this utterance is to inform the listener about the relationship between two salient entities, John and Rosa. The new information
for the addressee associated with (9c) is that John did not love Rosa. This can
be represented in the following way.
(11) a. pragmatic presupposition: John stands in the relation X to Rosa.
b. pragmatic assertion: X = didnt really love
c. focus: didnt really love
As Lambrecht argues, this difference in topicality is formally marked. In (9b)
Rosa is an accented lexical noun phrase within the focus domain. In (9c) the
same referent is expressed by an unaccented pronoun, as is typical of topics
(Givon 1983a, Ariel 1988, Gundel et al. 1993). The same sentence can easily
appear in the what-about context as a test for the topichood of Rosa:
(12) a. And what about Rosa?
b. He didnt really love her.

Information structure roles

57

Thus, both John and Rosa in (9c) may be characterised as topics. The two
topics stand in a certain relation to each other, established before the relevant
sentence is produced. This relation constitutes a part of the presupposed information associated with the sentence, while the new assertion is meant to
update the addressees knowledge about this relation. Nikolaeva (2001:26) defines secondary topic as an entity such that the utterance is construed to be
ABOUT the relationship between it and the primary topic. This definition does
not explicitly reflect the fact that topics are ordered with respect to saliency:
the primary topic is more pragmatically salient then the secondary topic. For
example, although (9c) is construed as being about the relation between John
and Rosa, John is a more salient participant.
This difference in saliency between two topical elements is captured in Vallduvs approach, which is by and large compatible with ours. Vallduv proposes a trinomial articulation of information structure. In his model, new information is termed focus and old information is termed ground; ground is
further splittable into informationally more and less prominent material, link
and tail. As mentioned above, our notion of topic roughly corresponds to
Vallduvs link. The second informational primitive used by Vallduv, tail,
basically corresponds to what we refer to as secondary topic.1 The new information conveyed by the sentence is to be recorded on the file card headed by
link. Tail indicates a more specific means of adding information to the given
address. It entails the presence of a particular record on the file card for link,
and signifies that update is to be carried out by completing or modifying this
record. This ensures that link and tail stand in a certain presupposed relation,
just as was argued above for the primary and the secondary topic. The new assertion completes or modifies the tail entry, and therefore updates information
about the relation between it and the link.
Consider the following Catalan example of the link-focus-tail structure
from Vallduv (1992).
(13) a. How does the boss feel about broccoli?
b. Lamo [lodia], el br`oquil
the.boss it.hates the broccoli
The boss HATES broccoli.

(Vallduv 1992:74)

Given the context in (13a), the interlocutors believe that the boss has some
attitude toward broccoli at the time (13b) is produced: that is, the entry broccoli (tail) is already present on the card for the boss (link), and the relation between them is under discussion. The focus, indicated here with square
1 Erteschik-Shir

(2007:13) explicitly refers to both link and tail as topic types.

58

Information structure in grammar

brackets, substitutes for the missing material in the boss/broccoli relation and
therefore updates the information on the file card for link.
The same structure is illustrated by the following example from ErteschikShir (2007) (bracketing is hers).
(14) What did John do with the dishes?
[hetop [washed themtop ]foc ]foc

(Erteschik-Shir 2007:4748)

Erteschik-Shirs model has only two informational primitives, topic and focus,
but topic and focus are not mutually exclusive, and subordinate structures are
allowed. Erteschik-Shir analyses example (14) as containing two topics, the
primary and the subordinate topic, but adds that primary topic takes precedence, in the sense that the truth-value is calculated with respect to the primary
topic. Unlike our analysis, she assumes that each topic must have a focus
associated with it. Therefore (14) is said to contain two superimposed focus
domains. In our terms, the informational representation of (14) is as follows:
(15) a. pragmatic presupposition: John stands in the relation X to the dishes.
b. pragmatic assertion: X = washed
c. focus: washed
To put it differently, both topical elements, John and dishes, are excluded
from the focus domain. The focus the aspect in which the old and new
information differ only extends over the verb washed, and there is no subordinate focus-structure.
More research is needed on possible secondary topic contexts but, as far as
we can tell, there are two frequent informational types that involve secondary
topic. The first is illustrated in the examples above, where the focus extends
over the predicate alone: didnt really love in (9c), hates in (13) and washed
in (14). These predicates are accented in English. The information structure
associated with these examples is topic - focused predicate - secondary topic:
i.e., they are instances of predicate-focus structure.
The second type is argument focus, as in the following:
(16) a. Where did John kiss Rosa?
b. He kissed her in the kitchen.
(16b) updates the addressees knowledge about the relation between John and
Rosa by adding the information that it was in the kitchen that he kissed her.
This can be represented as follows:
(17) a. pragmatic presupposition: John kissed Rosa in X

Conclusion

59

b. pragmatic assertion: X = the kitchen

c. focus: in the kitchen
In this example, the focus corresponds to the non-predicate phrase in the kitchen,
but both the primary topic (John) and the secondary topic (Rosa) are expressed
by non-accented pronouns, just as in (9c).
3.3 Conclusion
This chapter has explored our understanding of information structure and the
crucial notion of topicality. The difference between topicality and topic-worthiness
(or prominence) is of crucial importance for our analysis: topic-worthiness,
accessed in terms of referential features, only determines the likelihood for a
referent to be construed as topical, and does not automatically make it topical.
The main characteristic of topics is pragmatic saliency (presupposition of
saliency/relevance or topicality presupposition, in Lambrechts (1994) terminology) attributed to a referent by the speaker in a given context. This of
course makes the assessment of topicality in isolated written examples a rather
difficult task; wherever possible, we will use question-answer pairs as well
as the what-about tests mentioned above to elucidate information structure
roles, following standard practise in information structure-related research.

Syntax and information

structure
Chapter 2 showed that patterns of agreement and casemarking are often definable in terms of purely syntactic criteria verbs may obligatorily agree with
their subjects, for example but that criteria for marking may also refer to
other, nonsyntactic levels of representation. LFGs projection architecture
(Kaplan 1987) allows reference not only to the syntactic levels of c-structure
and f-structure but also to other linguistic levels and the relations among them.
Our investigation centres on the role of information structure in regulating object marking patterns; here we discuss the representation and formal treatment
of information structure and its relation to other levels of linguistic structure.
We rely on Mycocks (2009) insight that the traditional semantic structure of
LFG, an important component of the glue approach to the syntax-semantics
interface (Dalrymple 1999, 2001, Asudeh 2004), plays an important role in
representing information structure relations.
4.1 Alternative views of information structure
Various views of the representation of information structure and its relation
to other grammatical modules have been proposed. Here we review some of
these views, including early work within LFG.
4.1.1 Tree-based representations
It is well known that word order and phrasal configuration are important in en Kiss (1995) discusses discourse-configurational
coding information structure; E.

61

62

Syntax and information structure

languages such as Hungarian (Uralic), which have designated phrasal positions for particular information structure roles. Within LFG, the relation between phrase structure position and information structure role has been explored in detail by, among many others, Choi (1999), Butt and King (1996,
2000), and Mycock (2006). As we discuss in Section 4.2.2 of this chapter, this
work takes the standard LFG view that languages may use phrase structure
position (as well as a range of other cues) to signal various kinds of grammatical information, including information structure role. Some languages (those
often termed configurational) associate grammatical functions such as subject and object with particular phrase structure positions, and use other means
for example, prosody or morphological marking to signal information
structure. Other languages associate phrase structure positions with information structure roles (these are the discourse configurational languages), and
use other means, often morphological, to signal grammatical function. Languages may also use a combination of the two methods, with positional and/or
morphological cues signalling both syntactic and information structure roles.
Phrase structure representations play a very different role in approaches such
as Principles and Parameters or the Minimalist Program, in which the phrase
structure tree is the primary means of representing grammatical information.
Researchers working within this paradigm often encode the information structure role of a phrase by assigning it to a particular phrase structure position
(as if all languages were discourse-configurational languages, at least at an
abstract level), though some have proposed the use of features or additional
levels of structure to define information structure roles independent of phrasal
position, as we discuss below. The second view is similar to the LFG-based
approach that we adopt, since it assigns a more autonomous role to information
structure, and does not assume an invariant link between phrasal position and
information structure role; see Mycock (2006) for more discussion of these
issues.
The influential theory of Rizzi (1997) (the cartographic approach), in
which topic and focus appear in specifier positions of TopP and FocP phrases,
exemplifies the first view. Rizzi (1997:297) provides the following schematic
tree to illustrate the approach:

Alternative views of information structure

(1)

63

ForceP
Force

TopP*
Top

FocP
Foc

TopP*
Top

FinP
Fin

IP

This structure was originally motivated by the information structure roles associated with displaced constituents in the Italian clause: any number of topics
can appear before or after the single focus constituent (this is represented by
the Kleene star after TopP, which indicates that TopP is a recursive category).
Practitioners of the cartographic approach often assume that all topics and foci
must move to the appropriate specifier positions, though possibly at a more
abstract level such as LF: covert movement to these positions explains why
focused elements appearing in situ can trigger effects such as weak crossover.
A number of technical problems have been raised for various versions of
the cartographic view. We will not provide a comprehensive catalog of these
problems, but cite only a few representative discussions: Erteschik-Shir (2007)
discusses data from Italian, Hebrew, and English, Neeleman and van de Koot
(2008) discuss Dutch data, and Zwart (2009) discusses Germanic data which
are problematic for the approach. Some researchers working within the general
Minimalist paradigm have explored alternative views, encoding information
structure by means of features annotated on the phrase structure tree, or by
rules which refer to general, abstract phrasal configurations.
Neeleman and van de Koot (2008) reject the cartographic view, and propose an alternative view that is still (partially) based in tree configuration, but
applies more loosely. As in the LFG approach, they assume a separate level
of information structure, and propose abstractly defined mapping rules that
apply to syntactic configurations to encode/signal information structure roles,
similar to the description-by-analysis rules that are sometimes used in LFG
(Dalrymple 2001: Chapter 7). They point out that their approach does not
require appeal to covert movement, though it is compatible with covert movement given some additional assumptions. If covert movement is not assumed,
the approach is in broad terms compatible with an LFG-style approach, in that
rules are formulated to relate syntactic structures representing the surface order
of constituents to separately defined information structure representations.

64

Syntax and information structure

Other approaches are still less tied to phrase structure, and do not rely on an
association between phrasal position and information structure role. For example, Erteschik-Shir (2007) treats TOP and FOC as features which are lexically
assigned to the heads of phrases, and percolate to up to the maximal phrasal
projection; the features are not tied to a particular phrasal position, and may
appear in various places in the tree. Similarly, Buring (2007) assumes that
nodes of the phrase structure tree may be annotated with the privative features
F and T, defining information structure roles of Focus and Topic, and argues
against cartographic-style approaches which force covert movement in cases
where there is otherwise no evidence for it.
Within the Combinatory Categorial Grammar framework, Steedman (2001)
assumes a very different theory of phrase structure and its representation than
is generally assumed in either the Minimalist Program or LFG. Combinatory
Categorial Grammar allows the division of a clause into constituents of various types, including standard (X -theoretic) constituents as well as sequences
of words that are not constituents on either a Principles and Parameters-based
or an LFG-based view. For instance, a sentence like Anna married Manny
can have either of two constituent bracketings, [[Anna married] Manny] and
[Anna [married Manny]]; prosody determines which bracketing is chosen, as
well as determining the information structure role of the constituents. Although information structure roles are in some sense aligned to phrase structure
constituents, the availability of nonstandard constituents and multiple possible
constituent structures for an utterance gives this approach a good deal of flexibility, and makes it crucially different from the cartographic approach. The
CCG approach provides a remarkably successful theory of the relation between prosody and information structure, which differs in interesting ways
from the LFG perspective. Since our focus in this work is the morphological
marking of information structure roles by agreement and casemarking rather
than prosody, we will not address the prosody/syntax interface issues that are
central to the CCG approach, though in Section 4.2.2 below, we briefly discuss
some LFG-based proposals for the treatment of prosody, its interaction with
syntax and information structure, and how it helps to determine information
structure role.
4.1.2 Early work in LFG
There is a large body of work on information structure and information packaging in LFG, beginning with Bresnan and Mchombo (1987) and continuing
with the influential work of King (1995), Butt and King (1996, 2000), and

Alternative views of information structure

65

Choi (1999); King and Zaenen (2004) and OConnor (2006) provide a useful
overview.1
Among the first LFG researchers to discuss information structure and its interactions with syntax was King (1995), in her analysis of topic, focus, and
word order in Russian. King presents f-structure representations like the following:
(2) F-structure for Inna, John claimed that he saw (at the beach) (King
1995:199) :



TOPIC

PRED

I NNA

PRED CLAIMSUBJ, COMP




SUBJ

PRED J OHN

PRED SEE  SUBJ, OBJ




COMP

PRED

HE

SUBJ

OBJ

Here, the f-structure for Inna is the TOPIC; the line connecting it to the object
of the sentential complement (COMP) indicates that it plays not only the TOPIC
role but also the role of the OBJ of the verb saw. This captures the long-distance
syntactic dependency which this sentence exhibits, involving the displacement
of a phrase to the beginning of the sentence (Kaplan and Zaenen 1989); the
usual term for this construction is topicalisation, and the displaced phrase
Inna bears the label TOPIC.2
LFG researchers commonly use the features TOPIC and FOCUS in f-structure
in this way, to label displaced constituents in unbounded dependency constructions topicalisation constructions, relative clauses, and questions and to
establish a relation between the two different roles that displaced phrases play
in such constructions (TOPIC and OBJ in example (2), for example) in order
to impose syntactic constraints on possibilities for displacement. When the
features TOPIC and FOCUS appear at f-structure, they are taken to be gram1 OConnor (2006) refers to the level at which topic and focus are represented as d(iscourse)structure; for him, i(nformation)-structure is a term for relations among multiple levels, including prosody and semantics. We will retain the more standard term information structure (or
i-structure) for this level, since in some other LFG work the term discourse structure refers to
the relations between successive utterances in a discourse, rather than to sentence-internal information packaging (King and Zaenen 2004).
2 Despite the common use of the term topicalisation for this construction, the displaced phrase
need not be an information structure topic in the sense defined in Chapter 3; in English, as discussed by Prince (1981) (see also Lambrecht 1994:31), focused constituents can also appear in
this position.

66

Syntax and information structure

maticalised discourse functions whose synchronic function is purely syntactic,

related to but different from the information structure roles of topic and focus
defined and discussed in Chapter 3 (Bresnan 2001, Falk 2001).3 This treatment
of long-distance dependencies allows us to capture the important syntactic differences between sentences in which a constituent is displaced and those with
no displaced constituents, and also encodes the historical relation between the
grammaticalised discourse functions TOPIC and FOCUS in f-structure and the
information structure roles to which they are related (Bresnan and Mchombo
1987). Alsina (2008) criticises this practice on the basis that it needlessly introduces (apparent) information structure distinctions into f-structure, and proposes the term OP rather than TOPIC or FOCUS for all displaced elements; we are
sympathetic to this view, though for compatibility with earlier approaches, we
will continue to use TOPIC and FOCUS in f-structure representations of sentences
with displaced constituents rather than OP.
As noted by King (1995), there was a tendency in early LFG literature to
use the f-structure labels TOPIC and FOCUS to represent not only grammaticalised (and therefore syntactic) functions, but also information structure roles
as defined in Chapter 3. Although the information structure roles of topic and
focus are very different from grammatical functions like subject and object, it
was thought by many researchers to be convenient to co-opt the f-structure to
represent these information structure roles as well as grammatical functions.
When this is done, f-structure is no longer a purely syntactic representation,
but instead represents a combination of information structure and grammatical
structure. In their work on information structure and word order in Urdu and
Turkish, Butt and King (1996) represent information structure topic and focus
as f-structure features, though they note that it may be preferable to represent
them at a separate level. Choi (1999) treats the information structural features
+ NEW and + PROM (inent) as f-structure features in her work, as we will see
in Section 4.2.1 below (see also Cook and Payne 2006), though Choi asserts
that information structure should be thought of as an independent grammatical
representation, the nature of which is left for future work.
3 Bresnan

(2001) also classifies SUBJ as a grammaticalised discourse function. We do not adhere to this classification here, since we believe that the syntactic properties of the grammaticalised
functions TOPIC and FOCUS are quite different from other functions appearing at f-structure: for
example, verbs commonly subcategorize for the functions SUBJ, OBJ, and the oblique functions,
but subcategorisation for TOPIC and FOCUS is much less common, and may not occur at all (for
discussion, see Huang 1989, Her 1991, and Culy 1994). Falk (2001) refers to TOPIC and FOCUS
as grammaticalised discourse functions, and classifies SUBJ, TOPIC and FOCUS as overlay functions (Johnson and Postal 1980), noting that each of the overlay functions relates to its clauses
place in larger syntactic or discourse structures (Falk 2001:59). See Chapter 5, Section 5.3 for
more discussion of the relation between syntactic subject and the information structure role of
topic.

Alternative views of information structure

67

King (1997) provides a detailed critique of the practise of representing information structure roles at f-structure, particularly focusing on mismatches
between f-structure and information structure: cases in which f-structure constituents are either too big or too small to adequately represent topic, focus,
and other information structural constituents. Based on these and other considerations, King argues that a level of information structure separate from
f-structure is necessary, and she is among the first to make an explicit proposal for a separate level of information structure (a separate projection, in
LFG terms: Kaplan 1987, Asudeh 2006) whose parts are related by a projection function to the corresponding parts of other structures. King provides
the abbreviated partial f-structure and i(nformation)-structure shown in (4) for
example (3b):
(3) a. Was it the ex-convict with the red SHIRT that he was warned to look
out for?
b. No, it was an ex-convict with a red [TIE] that he was warned to look
out for.
(King 1997:8, citing Jackendoff 1972:232)
(4) a. F-structure (King 1997:8):

PRED EX - CONVICT

PRED WITH  OBJ

ADJUNCT

PRED TIE


 
OBJ

ADJUNCT
PRED RED

b. I-structure (King 1997:9):

FOC TIE



EX - CONVICT

BCK

WITH
RED

We agree that it is good theoretical practise to separate out different aspects of

linguistic structure and represent them separately, and we follow King (1997),
Butt and King (2000), and many other LFG researchers in assuming that information structure is a separate level of representation, independent from fstructure; see OConnor (2006) and Mycock (2006) for detailed discussion
of these issues. This view also has clear similarities to the HPSG-based proposal of Engdahl and Vallduv (1996), who introduce the features FOCUS and
GROUND within a structure which they call INFO - STRUCT.

68

Syntax and information structure

4.2 Information structure and its role in grammar

4.2.1 Content of information structure: The LFG view
Choi (1999) was among the first to propose a fine-grained representation of
information structure features within LFG, appealing to the features +NEW and
+ PROMINENT to encode distinctions relevant for her analysis of word order. The
feature +NEW categorises what Choi calls discourse-newness: focused arguments are discourse-new (+NEW), and topic and tail (in the sense of Vallduv
1992: see Chapter 3, Section 3.2.4) are discourse-old ( NEW). Choi classifies
topic and contrastive focus as prominent or + PROM, and tail and completive
focus as non-prominent or PROM.
Butt and King (2000) adopt Chois classificatory features to define four information structure roles: topic is [NEW] and [+PROMINENT ], focus is [+NEW]
and [+PROMINENT ], completive information is [+NEW] and [ PROMINENT ], and
background information is [NEW] and [ PROMINENT ]. Butt and Kings classification of roles differs from Chois primarily in the treatment of focus: for
Choi, completive focus is [PROMINENT ], while for Butt and King it is [+PROMINENT ].
This meshes more closely with our understanding of the roles.
In Butt and Kings system, completive information is new to the addressee
but, unlike focus, it is not associated with the difference between pragmatic
assertion and pragmatic presupposition. According to this classification, the
phrase in the kitchen in (5b) is a part of completive information:
(5) a. What is Bill eating?
b.

He

is eating

pizza

in the kitchen.

TOPIC

BACKGROUND

FOCUS

COMPLETIVE

Background information differs from topic in the following way: while topic
is a pointer to the relevant information to be accessed by the addressee, background provides more detailed knowledge that may be necessary for a complete understanding of new (focused) information.
We follow Butt and King in adopting a four-way distinction in information
structure roles, but also rely on the definitions of discourse functions which
were presented and motivated in Chapter 3, rather than the feature-based definitions that Butt and King propose. In particular, it is important to emphasise
that our secondary topic is not the same as what Butt and King (1996, 2000)
refer to as background information, although in some cases these notions can
overlap. There are several differences between Butt and Kings notion of background information and our notion of secondary topic. First,they differ in
terms of prominence (saliency): background information is [ PROMINENT ],
while secondary topic is [+PROMINENT ] (pragmatically salient, in the sense of
Lambrecht 1994), just like the primary topic, although the primary topic is

Information structure and its role in grammar

69

more prominent than the secondary topic. We appeal to a scale of prominence

in distinguishing different degrees of topichood, rather than a simple binary
feature [+PROMINENT ]. Additionally, the secondary topic appears in sentences
which are construed to be about its relation with the primary topic. This implies that the proposition expressed by the sentence must involve at least two
referents. There is no such requirement for background information, as is clear
from example (5b), where the verb is a part of background information; indeed,
some examples cited by Butt and King involve backgrounded one-place verbs
and no referential NPs other than the subject.
Further, Butt and King (2000) do not distinguish between narrow and wide
focus, instead associating focus with one (nonverbal) participant even in cases
like (6b):
(6) a. What is Bill doing?
b.

He

is eating pizza in the kitchen.

TOPIC

FOCUS

On their analysis, (6b) has the same information structuring as (5b). This is primarily motivated by Kings (1997) argument that there are technical difficulties with including verbs as focus or topic in information structure. However,
in cases like (6b) it is difficult to determine any informational (nonsyntactic)
grounds to select one particular participant for this purpose. The informational contribution of the verb in (6b) does not seem to differ from that of the
nonverbal participants; the entire nonsubject portion of the sentence fills the
informational gap between the speaker and the addressee. The informational
role of the verb eating is therefore different from (5b), where it is background
information.
4.2.2 Linguistic encoding of information structure relations
We treat information structure as a separate, independent level of structure,
containing the features TOPIC, FOCUS, BACKGROUND, and COMPLETIVE. We
assume that these information structure roles are fixed by discourse context,
and that assignment of these roles is subject to certain inviolable syntactic and
semantic conditions discussed in Chapter 3: for example, topics must be referential, and foci must be overtly expressed. Various linguistic cues can be used
to signal the information structure of an utterance to the addressee, and these
cues must be consistent with the information structure roles imposed by linguistic context (for more discussion, see Mycock 2006, Erteschik-Shir 2007,
Fery and Krifka 2008, and references cited there). Interestingly, such cues are
often the same as those that languages use to signal grammatical functions:
not only casemarking and agreement, but also word order and phrase structure

70

Syntax and information structure

position. Choi (1999) analyses the influence of information structure roles on

word order in Korean (isolate) and German (Germanic), and Butt and King
(1996, 2000) provide a detailed study of the encoding of information structure roles by phrase structure position in Hindi-Urdu. For Hindi-Urdu, Butt
and King show that topics appear sentence-initially, foci appear immediately
before the verb, and background information is postverbal. These syntactic
positions are associated with information structure roles just as grammatical
functions are: by means of functional annotations on phrase structure rules, as
described in Chapter 2, Section 2.4.
Prosody is also well known to play a role in the encoding of information
structural relations; in this way, grammatical function encoding differs from information structural encoding, since as far as we know languages do not mark
grammatical functions purely by prosodic means. Though prosody provides
important indications of information structure role, it often happens that information structure roles are not unambiguously specified prosodically; Lambrecht (1994) discusses the examples in (7b) and (8b), which exhibit the same
syntactic structure and the same prosody, but are associated with different information structures depending on the linguistic context:
(7)

Argument-focus structure:
a. I heard your motorcycle broke down?
b. My CAR
FOCUS

(8)

broke down.
BACKGROUND

Sentence-focus structure:
a. What happened?
b. My CAR broke down.




(Lambrecht 1994:223)

FOCUS

Linguistic cues signalling information structure role, including casemarking,

agreement, word order, and prosody, must be consistent with the information
structure roles imposed by linguistic context. This accounts for the unacceptability of examples such as (9), where a topical argument is associated with a
prosodic contour which signals focus:
(9) a. What happened to your car?
b. *My CAR
TOPIC

broke down.
FOCUS

Information structure and its role in grammar

71

Since our primary concern is the information structure role of topic and how
it is signalled by means of agreement and casemarking, we will not have much
to say about the positional or prosodic encoding of information structure roles;
for detailed discussion of these issues, and formal proposals within LFG for the
treatment of word order and prosody that are compatible with the grammatical
architecture that we assume here, see Choi (1999), Butt and King (1996, 2000),
OConnor (2006), Mycock (2006), and references cited there.
4.2.3 Information structure in relation to semantics
As discussed in Chapter 3, information structure represents how the propositional content of an utterance is structured in line with the speakers model of
the addressees state of knowledge at the time of utterance: it is concerned with
utterance meaning and its packaging to optimise the effect of the utterance
on the speaker. A formal theory of information structure must, then, involve
reference to the meaning of the parts of an utterance and how they are assigned
information structure roles.
Researchers in formal semantics and information structure have not agreed
on the relation between truth-conditional semantics and information structure,
with some researchers arguing that information structure should be represented
as a completely separate module from truth-conditional semantics, and others
arguing that information structure is best viewed as a means of partitioning
truth-conditional meaning. We take the second view: information structure
partitions sentence meaning into information structure categories, as we describe in Section 4.3 below. In this, our approach resembles structured meaning approaches (von Stechow 1982, Krifka 1992) in some respects, though we
will see that there are important differences between our approach and theirs.
We believe that it is also compatible with Lambrechts (1994) view of information structure as the pragmatic structuring of the proposition; indeed, Lambrecht (1994:341) states that he is not convinced that it is always possible or
even useful to distinguish semantic meaning from pragmatic meaning.
One of the earliest proposals for structured meanings was made by von Stechow (1982), who represents utterance meanings as a list in which the first
element is the topic and the remaining elements are foci. Krifka (1992) represents utterance meaning as a pair, with background as the first member and
focus as the second member. Krifka (2006) assumes a three-part structured
meaning for the VP introduced BILL to Sue, with BILL in focus:
(10) bill, A, x.introduce(sue, x)

(Krifka 2006: example 2)

In (10), the structured meaning is a triple, with the focus bill as first member,
a set of alternatives to the focus A as second member, and the background
to Sue as third member. The set of alternatives
meaning for introduced

72

Syntax and information structure

A contains all the relevant individuals that might have been introduced to Sue,
including Bill (so, for example, in the context under consideration A might be
{bill, fred, chris, sue, ...}). Our approach will also assume that meanings of
the parts of an utterance are separated and classified according to information
structure role.
Importantly, however, we do not adopt some common assumptions that often go along with and provide motivation for structured meaning approaches.
Structured meanings are often posited in the analysis of association with focus (Jackendoff 1972, Rooth 1985); in particular, researchers working within
the structured meaning paradigm often adopt the view that meanings are structured in order to make available distinctions that are needed in the compositional semantics of so-called focus-sensitive operators such as only in examples like (11), from Krifka (2006):
(11) a. John only introduced

BILL

to Sue.

FOCUS

(The only person John introduced to Sue is Bill.)

b. John only introduced Bill to

SUE .
FOCUS

(The only person John introduced to Bill is Sue.)

The truth conditions of examples (11a) and (11b) differ because of focus placement, and proponents of the structured meaning approach argue that this is best
accounted for by assuming that structured meanings combine in a particular
way with operators like only. For example, Krifka (2006) assumes that the
interpretation of only depends on the three-part structure of the meaning of the
VP introduced BILL to Sue given in (10). On this view, the VP adverb only
applies to a structured meaning triple like the one in (10) to give the meaning
in (12):
(12) only (F, A, B) = x .Y A.[B(Y)(x) F = Y ]
(Krifka 2006: example 3)
Combining the meaning for only introduced BILL to Sue with a subject John,
the result is:
(13) Y A.[introduce(sue)(Y )(john) bill = Y ]
(Of the members of the alternative set A, only Bill was introduced to Sue
by John.)
(Krifka 2006: example 4)
A notable property of this result is that the focus/background articulation which
was represented in the structured meaning in (10) is no longer present: it was,

Information structure and its role in grammar

73

in a sense, consumed by the focus-sensitive operator only, and the result is

not a structured meaning, but what Krifka calls a standard meaning.
Our proposal differs in several ways. First, researchers adopting structured
meanings in the analysis of operators like only are often concerned primarily
with the distinction between focus and ground, and not with other information
structure roles such as topic or the background/completive distinction. We believe that all of these information structure roles are important in the pragmatic
structuring of meanings and should be represented at information structure.
Second, we do not believe that the presence of operators such as only results
in the obliteration of the distinctions that are present in structured meanings.
Schwarzschild (1997:2) makes the following observation:
Researchers have pretended that foci embedded under operators
like only and always do not carry the same pragmatic import as
unembedded foci. This is a surprising claim. It means that the language has an elaborate syntactic-phonological system concerned
with moulding an utterance to background discourse and that this
system shuts down as soon as it meets one of these operators.
We agree that this is a surprising and undesirable result. We assume that pragmatic structuring of utterance meaning is relevant for all utterances, and that
this structuring does not disappear in the presence of operators like only; see
Schwarzschild (1997) and Kadmon (2001) for more discussion of this point.
Third, we agree with many other researchers that information structure does
not always provide the relevant distinctions for the interpretation of operators
which have been analysed as focus-sensitive: Vallduv (1992: Chapter 7) provides useful discussion of this point, though we disagree with his conclusion
that information structure must be represented as a completely separate level,
unrelated to truth-conditional meaning. We believe that there are interesting
and important relations between what Kadmon (2001) calls the discourseregulating function of focus and its role in determining the interpretation of
such operators, but we leave open the question of how exactly this should be
worked out. Work by Roberts (1996), Schwarzschild (1997), Kadmon (2001),
and Beaver and Clark (2008) explores this issue in depth, and shows convincingly that although focus in the sense defined in Chapter 3 often plays an important role in the interpretation of so-called focus-sensitive expressions, the
relation is complex, and depends on a thorough understanding of lexical properties of operators, compositional semantics, and information structure roles.
In a similar vein, Hajicova et al. (1998) explore the relation between information structure and scope, showing that while information structure often plays
an important role in the determination of scope, the relation is complex and
not reducible to a simple mapping. We believe that it is important to represent

74

Syntax and information structure

information structure explicitly, and to clearly motivate and define information

structure roles, in order that these connections can be thoroughly explored.
4.3 Our architecture
In the following, we present the formal architecture of grammar which we assume, and we show how to specify information structure relations using the
formal tools of LFG. Our analysis crucially involves semantic structure in the
sense familiar from work on the glue approach to the syntax-semantics interface (Dalrymple 1999, 2001, Asudeh 2004), and we propose a new and
enriched view of the standard glue approach and its relation to information
structure.
4.3.1 Glue and the syntax-semantics interface
LFGs theory of the syntax-semantics interface uses a resource logic, linear
logic, to state instructions for combining meanings of the parts of an utterance
to produce the meaning of the utterance as a whole. This section provides a
brief overview of the standard assumptions of the glue approach, in preparation for presentation of the full formal details of our theory of the information
structure module and its relation to other levels of linguistic structure. We will
also provide some abbreviatory conventions for meaning expressions which
are often used in the glue approach, in an effort to avoid formal overload. For
a more complete introduction to glue, see Dalrymple (1999, 2001) and Asudeh
(2004).
Consider the c-structure and f-structure for the sentence John married Rosa:
(14) John married Rosa.
IP
NP

I

N

VP

V

John

NP

married

N
N

Rosa

PRED

MARRYSUBJ, OBJ

PRED
SUBJ


OBJ

PRED

J OHN
ROSA

Our architecture

75

We follow the normal LFG practise of representing only the features and values of the f-structure that are relevant for current discussion, as noted in Chapter 2, leaving out (among other things) the f-structure features of tense, aspect,
agreement, and case when they are not relevant for the discussion.
According to the glue approach to the syntax-semantics interface, the meaning associated with this sentence is derived via a series of instructions that can
be paraphrased in the following way:
(15) a. The word John contributes the meaning john.
b. The word Rosa contributes the meaning rosa.
c. The word married contributes meaning assembly instructions of the
following form: When given a meaning x for my subject and a meaning y for my object, I produce a meaning marry (x, y) for my sentence.
In slightly more formal terms, glue assumes a level of semantic structure,
sometimes called structure, which is related to f-structure by means of a
projection function from f-structures to semantic structures. Meanings are
related to expressions involving combinations of semantic structures. For the
NP John, the following configuration is usually assumed:
(16)

NP
N

PRED

J OHN

john:[ ]

John
The f-structure for John is related to its corresponding semantic structure by
the function from f-structures to semantic structures, represented by the dotted arrow. In (16), the semantic structure is represented without features or
values, as is common in most glue-based literature. We will propose a set of
features and values for such semantic structures below. This semantic structure
is paired with the meaning for John, represented here simply as the term john.
The expression john:[ ] consists of a meaning expression on the left side and an
expression involving semantic structures on the right side, with the two sides
separated by a colon: this kind of expression is called a meaning constructor.
The meaning constructor for John is represented in the lexicon as in (17),
where the meaning john is associated with the semantic structure projected
from the f-structure . We use the subscript to represent the function relating f-structures to their corresponding semantic structures, which was represented as a dotted line in (16):

76

Syntax and information structure

(17) john:
The glue approach does not prescribe a particular method for the representation of meaning; any method that is adequately expressive for natural language meanings can be used. The only requirement on how meanings are
expressed is that there must be an explicitly worked out way of combining
meanings: in most glue-based treatments, this is function application. Dalrymple et al. (1999) discuss the use of Discourse Representation Structures
(Kamp and Reyle 1993) specifically, Lambda DRT (Bos et al. 1994) in
a glue setting, and Dalrymple et al. (1997) use intensional logic in their gluebased analysis of quantification. For simplicity, we will stick to formulas of
predicate logic in the following explication.
A verb such as married makes a more complicated semantic contribution
than a name like John, since it must provide instructions to combine the meanings of its subject and object to produce the meaning for the entire sentence:
(18)

VP
V

PRED

SUBJ

OBJ

MARRYSUBJ, OBJ

[]
[]

x .y .marry (x, y ):s (o m )

married
The meaning of married is represented simply as x .y .marry (x, y ): a relation between two individuals x and y that holds if x marries y. This expression is paired with the linear logic expression s (o m ), where s is the
semantic structure corresponding to the subject, o is the semantic structure
corresponding to the object, and m is the semantic structure corresponding to
the entire sentence, headed by the verb married. This expression involves the
linear logic operator , linear implication, and can be paraphrased as follows:
(19) If I am provided with the semantic structure of my subject and then the
semantic structure of my object, I produce the semantic structure of the
sentence.
In the lexicon, the meaning constructor for married is represented as follows:
(20) x.y .marry (x, y):(

SUBJ) (( OBJ) )

The meaning constructors that are contributed by the semantically significant parts of an utterance are combined in a linear logic deduction to produce
the meaning of an utterance. Linear implication on the linear logic meaning assembly side (the right side) corresponds to function application on the meaning
side (the left side):

Our architecture
(21)

77

X : f
P : f g
P (X) : g

Following this rule, we can combine the meaning of the subject John with the
meaning of the verb married in the following proof of the meaning of John
married, which still requires a meaning for the object:
(22)

john:s

x .y .marry (x, y ):s (o m )

y .marry (john, y ):o m

We can augment the proof by providing the meaning of the object Rosa, producing the meaning marry (john, rosa) for the entire sentence, as required:
(23)

john:s

x .y .marry (x, y ):s (o m )

y .marry (john, y ):o m
marry (john, rosa): m

rosa:o

The order of combination of meaning constructors does not matter: equally,

we can first combine the meaning of the object Rosa with the meaning of the
verb married in a proof of the meaning of married Rosa, and then combine
this with the meaning of the subject John, as shown in (24).4
(24)

rosa:o

x.y .marry (x, y ):s (o m )

x .marry (x, rosa):s m
marry (john, rosa): m

john:s

4.3.2 Information structure features and relations

Most glue language treatments assume that the meaning constructors that appear in the derivation of the meaning of an utterance are an unordered, undifferentiated collection. Our proposal differs from these: we propose to group
meaning constructors according to information structure role, separating those
that contribute to the focus, those that contribute to the topic, and those that
contribute to background or completive information. Importantly, however, we
retain the requirement of semantic completeness and coherence which other
work within the glue framework assumes (Dalrymple 1999: Chapter 1; Dalrymple 2001: Chapter 9): an utterance meaning must be derivable from the
4 This is an abbreviated version of the proof; in the first step we could be more explicit, and
more correct, by first introducing a hypothetical meaning for the subject and combining it with
the verb, then combining the resulting subject+verb meaning with the object, and then discharging
the assumed meaning for the subject, and abstracting over the variable introduced as the subject
meaning, to produce the verb+object meaning constructor in the second line of the proof. In the
interests of simplicity, we skip these steps.

78

Syntax and information structure

complete set of premises contributed by the parts of the utterance; the utterance meaning should not contain unsaturated expressions, in which some requirements have not been satisfied; and there should be no leftover meaning
constructors which have not been used in the derivation of the meaning of the
utterance.
Our proposal also differs in that we take the semantic structure to be the
proper level for the representation of features relevant for the definition of
the status of discourse referents, in the sense of Lambrecht (1994); in this,
we follow Mycocks (2009) insight that information structure and semantic
structure are closely related. Based on Lambrecht (1994), Liao (2010) proposes a number of information structure features to represent the activation
and accessibility of discourse referents: STATUS, whose values are IDENTIFI ABLE and UNIDENTIFIABLE ; ACTV (activation), whose values include ACTIVE ,
5
ACCESSIBLE , and INACTIVE ; and ANCHORED , a binary feature with a positive or
negative value. Liao shows that these features are crucial in her analysis of the
distribution of overt and null anaphora in Mandarin Chinese, and they also help
to determine information structure role; we propose that semantic structure is
the proper level for the representation of these features. Semantic features
determining topic-worthiness (Chapter 3, Section 3.2.3), including animacy,
humanness, definiteness, and specificity, are also best represented at semantic
structure. Besides these, we introduce a semantic structure feature DF, whose
value is specified by the linguistic context as TOPIC, FOCUS, BACKGROUND, or
COMPLETIVE. This feature will be crucial in our formal analysis of the content and representation of information structure and its relation to semantic
structure. In the following, we will adhere to the same policy with respect to
information structure features as we do with f-structure features: though we
assume that these features belong at semantic structure, if they are not directly
relevant for the discussion, we will not include them in our representations.
The meaning constructors we have introduced so far, with the f-structure for
the sentence John married Rosa, are:

(25)
PRED MARRY SUBJ, OBJ
john:j



SUBJ j : PRED J OHN

x .y .marry (x, y ):j (r m )
m :



rosa:r
OBJ r : PRED ROSA
In (25), the parts of the f-structure have been labelled, with the label m assigned to the f-structure for the entire sentence, j to the subject f-structure,
and r to the object f-structure. The meaning constructors refer to m , j , and
5 OConnor (2006) discusses a similar ACTVN feature, but treats it as binary (with a positive
or negative value); Paoli (2009) makes a similar proposal for a +ACTIVE feature, which she
combines with a + CONTR feature in her analysis of contrastive and new focus.

Our architecture

79

r , the semantic structures related by the function to these f-structures. Instead of representing the function from f-structures to semantic structures
via a dotted arrow connecting the two structures, we will usually represent the
function by means of subscripts on f-structure labels, as we did above. For
example, f is the semantic structure corresponding to f ; in other words, the
function relates the f-structure f to its corresponding semantic structure f .
We represent the syntactic, semantic, and information structural aspects of
the sentence John married Rosa in a context in which the topic is John and the
focus is married Rosa in the following way:
(26)
John

married Rosa.

TOPIC

FOCUS

IP

VP

V

John

I

NP

PRED

SUBJ
m :

OBJ

NP

married

N
N

MARRYSUBJ, OBJ



j : PRED J OHN


r : PRED ROSA

TOPIC
FOCUS

{ john:j }

x .y .marry (x, y ):j (r m )
rosa:r

Rosa
We introduce a new function from semantic structures to information structure. An expression like m , which labels the information structure shown in
(26), is defined in terms of the function , the composition of the function
and the function: m refers to the information structure which is related to
the semantic structure m by the function, or equivalently the information
structure which is related to the f-structure m by the composite function .
Section 4.3.5 below provides more information about the overall architecture
we assume.
We can further cut down on notational overload by introducing additional
abbreviatory devices. It is standard in the glue literature to abbreviate meaning
constructors by means of bold-face labels like the following:

80

Syntax and information structure

(27)

John
married
Rosa
married-Rosa

john:j
x .y .marry (x, y ):j (r m )
rosa:r
y .marry (x, rosa):r m

In (27), the meaning constructor john:j is given the label John, the meaning constructor x .y .marry (x, y ):j (r m ) is labelled married, and
similarly for the meaning constructors labelled Rosa and married-Rosa. We
can use these labels in proofs as abbreviations for the full meaning constructors: the labels are interchangeable with the meaning constructors themselves,
while having the advantage of being simpler and more readable. With these
abbreviations in place, we can present the proof in (24) in a more compact and
readable way:
(28)

married

Rosa
married-Rosa
marry (john, rosa): m

John

We can also use these abbreviations to present the full representation given in
(26) in a simpler way:
(29) John married Rosa.
IP

VP

V

John

I

NP

PRED

SUBJ
m :

OBJ

NP

married

N

MARRYSUBJ, OBJ



j : PRED J OHN


r : PRED ROSA

TOPIC
FOCUS

{ John }

married

Rosa

Rosa
Since the meaning constructor married-Rosa can be deduced via linear logic
proof from the two meaning constructors married and Rosa, we could equally
well represent this configuration as:

Our architecture

81

(30) John married Rosa.

IP
NP

I

N

VP

V

John

SUBJ
m :

OBJ

NP

married

PRED

MARRYSUBJ, OBJ



j : PRED J OHN


r : PRED ROSA

TOPIC
FOCUS

{ John }

{ married-Rosa }

Rosa
This view treats TOPIC, FOCUS, BACKGROUND, and COMPLETIVE as categorising meaning contributions according to their information structure role. By
virtue of its appearance in the TOPIC, FOCUS, BACKGROUND, or COMPLETIVE
set, a meaning (more precisely, the meaning represented on the left-hand side
of a meaning constructor) is assigned a role in affecting the context to fill the
informational gap between the speaker and the addressee. In the representation above, the meaning associated with the phrase John is assigned a topic
role, and the meaning associated with married Rosa is assigned a focus role.
Meaning constructors contributed by the various parts of an utterance are categorised according to their information structure contribution, and appear in the
relevant information role category.
4.3.3 Levels and equations
Our analysis depends on lexical entries like the following (again, for expository
purposes we are working with the simplest possible entries, omitting much
detail):
(31) John

( PRED) = J OHN
john ( ( DF))

This entry specifies that John is a word of category N, and is associated with
a functional description consisting of two parts. The first line says simply
that the node dominating the word John corresponds to an f-structure with the
feature PRED and value J OHN , as in the f-structure in (30). The second line is
crucial in achieving the desired information structure configuration:
(32) john ( (

DF))

82

Syntax and information structure

This specification involves the meaning constructor john:j , which has been
abbreviated as john, in line with the abbreviations introduced in 27. The specification is exactly equivalent to this one:
(33) [john: ] ( (

DF))

Here the meaning constructor john: is written inside square brackets; these
are just delimiters enclosing the meaning constructor expression, and have no
other significance. They could be omitted, but this would detract from readability. The entire expression can be paraphrased in the following way:
(34) The meaning constructor john:j , abbreviated as john, is a member of
the set value of the discourse function signified by ( DF) within the
information structure .
This constraint requires the meaning constructor for John to bear some information structure role.
Now, how can the proper discourse function for John be specified? This
information must not be stated in the lexical entry for John, since it is not an
intrinsic lexical property of John that it plays a particular information structure role: rather, this depends on the discourse context in which it appears on
any particular occasion of its use. As discussed in Section 4.2.2, information
structure roles are determined by the context of utterance, and are linguistically
signalled in a number of ways: through agreement or casemarking, phrasal position, or prosody. Default information structure roles can also be associated
with particular grammatical functions: for example, in many languages the
subject is the default topic, as we discuss in detail in Chapter 5, Section 5.3
(see also Bresnan 2001:98), and we assume that this is the case in English as
well. What is needed is a way to allow specification of these constraints, in
order to determine the information structure role borne by John.
This is accomplished by including an feature DF in the semantic structure,
and allowing specification of the value of DF as TOPIC, FOCUS, BACKGROUND,
or COMPLETIVE. If the value of DF is specified as TOPIC, then the meaning
constructor must be a member of the TOPIC set; similarly, if the value of DF is
specified as FOCUS, the meaning constructor is a member of the FOCUS set, and
similarly for BACKGROUND and COMPLETIVE. Thus, specification of a value for
the semantic structure feature DF determines whether the meaning constructor
is a member of TOPIC, FOCUS, BACKGROUND, or COMPLETIVE at information
structure.6
Our analysis assumes annotated phrase structure rules for English like the
following:
6 This is somewhat reminiscent of the use of the PCASE feature in the specification of the
grammatical role of a prepositional phrase: a preposition specifies a value like OBL GOAL as its
value for PCASE , and this value is then used to specify the grammatical function of the preposi-

Our architecture
(35) IP

83

I
=

NP
( SUBJ)=
=
(( DF)=TOPIC)

With John as the subject of the sentence, and using this rule and the lexical
entry in (31), we have the following partial configuration, encompassing only
the c-structure and f-structure. We have instantiated the and metavariables
in (35) to the particular f-structure names m and j that appear in this sentence.
We have also left out the = arrows which appear on the N and N, which
define the f-structure head relation and ensure that the NP and its head John
correspond to the same functional structure. The arrows represent the familiar
function from c-structure nodes to f-structures.
(36)

IP
NP
(m SUBJ)=j
m =j
((j DF)=TOPIC)
N

I

m:

j:

SUBJ

PRED

J OHN

John
(j PRED) = J OHN
john (j (j DF))
We will omit the c-structure in the following exposition to avoid clutter, retaining only the functional description harvested from the annotated c-structure.
Here is the full f-description:
(37)

(m SUBJ)=j
m =j
((j DF)=TOPIC)
(j PRED) = J OHN
john (j (j DF))

m:

SUBJ

j:

PRED

J OHN

In (37):
tional phrase, by introducing the equation ( ( PCASE))= at the level of the PP. If the value of
( PCASE) for a particular PP is OBL GOAL , the expression ( ( PCASE))= is exactly equivalent to the expression ( OBL GOAL )=. For detailed discussion of PCASE , see Kaplan and
Bresnan (1982) and Dalrymple (2001: Chapter 6).

84

Syntax and information structure

The first line requires the f-structure j to be the subject of m, which is
true of the f-structure shown.
The second line requires the information structure corresponding to m
and j to be the same. We assume that all members of a clause share
the same information structure, and that all phrase structure rules of a
language bear specifications like this one to ensure this. (We leave open
the question of whether a complex, multiclausal utterance has a single
information structure, or a different information structural organisation
for each simple clause). The result is that specifying a particular information structure role for a meaning constructor means that it bears that
information structure role within the entire clause.
The third line provides an optional, default discourse function TOPIC for
the subject; we discuss the relation between SUBJ and TOPIC in detail in
Chapter 5, Section 5.3. If compatible specifications are provided by the
linguistic context (see below), and the prosodic and discourse prominence features of John are consistent with its role as topic, the subject
will be associated with the information structure role of TOPIC.
The fourth line provides additional specification for the subject f-structure,
j: it must have an feature PRED with value J OHN . This is true of the
f-structure shown.
Finally, the fifth line specifies that the meaning constructor john must
bear the role specified by (j DF) at information structure.

We can simplify the final line of these constraints by assuming that the default equation ((j DF)=TOPIC) holds, and simplifying according to the equality
m =j :



(38)
m : SUBJ j : PRED J OHN
(m SUBJ)=j
(j DF)=TOPIC
j :[ DF TOPIC ]
(j PRED) = J OHN


john (m TOPIC)
m : TOPIC { john }
The equations produce the configuration shown:
at f-structure, ms subject is j, and js PRED is J OHN
the semantic structure j corresponding to j has the feature
value TOPIC

DF

with

Our architecture

85

the value of j s DF appears as the feature TOPIC in the information structure for the clause, m
In this way, meaning constructors can be specified in a particular linguistic
context as bearing a particular information structure role.
A more complete set of phrase structure rules for English is provided in (39):
(39) IP

I

I
=

NP
( SUBJ)=
=
(( DF)=TOPIC)

I
=

VP

V
=

V

V
=

"

VP
=

NP

( OBJ)=
=

We do not assume a default information structure role for nonsubject constituents in English (other languages may impose a stricter relation between
grammatical functions and information structure roles, as we will see in subsequent chapters) and so the annotations on these rules are the same as in standard LFG syntactic treatments, except for the specification that the objects
information structure is the same as the information structure for the entire utterance (the equation = in the annotations on the NP daughter of V ).
We also assume the lexical entries in (40) for married and Rosa:
(40) married V
Rosa

( PRED) = MARRYSUBJ, OBJ

[x.y .marry (x, y ):( SUBJ) ((
( PRED) = ROSA
[rosa: ] ( (

OBJ) )]

( (

DF))

We also introduce specifications for the information structure roles of John,

married and Rosa. We assume that the discourse context identifies John as
topical (reinforcing the default specification of the subject as topical in the
phrase structure rule for IP) and married Rosa as in focus, and that the discourse prominence features and prosodic contour reinforce this assignment, or
at least do not conflict with the assignment of these roles. We continue to use
the meaning constructor abbreviations defined in (27).

DF))

86

Syntax and information structure

(41)

IP

NP

SUBJ
m :

(m SUBJ)=j
m =j
((j DF)=TOPIC)

I

N

VP

V

John
(j PRED) = J OHN
john (j (j DF))

PRED

OBJ

married
(m PRED) = MARRYSUBJ, OBJ
married (m (m DF))

MARRYSUBJ, OBJ



j : PRED J OHN


r : PRED ROSA

NP
(m OBJ)=r
m =r
N
N

Rosa
(r PRED) = ROSA
rosa (r (r DF))
Contribution from linguistic and pragmatic context:

(j DF) = TOPIC
(m DF) = FOCUS
(r DF) = FOCUS

We will not make an explicit formal proposal for how the equations labelled
Contribution from linguistic and pragmatic context are introduced on the
basis of the linguistic and pragmatic context. In the context under consideration for (41), the verb and object are in focus, but this may not be unambiguously signalled by casemarking, agreement, phrasal position, or prosody. A
complete theory of the syntax-information structure interface would require a
full specification of how the information structure roles of topic, focus, background, and completive information are determined by discourse context, how
this gives rise to the equations in (41), and how all of these roles are signalled
by agreement, casemarking, word order, and prosody. We hope that our proposals will form the basis of future research into these complex issues. Our
aim in this book is more limited: we are interested in the information structure
role of topic, how it is signalled by means of agreement and casemarking, and
how the relation between topic and grammatical function is constrained; these
are the issues which will be explored in the next few chapters.

Our architecture

87

We can again harvest the functional description from the annotated c-structure
in (41). This time, we reorder the constraints, separating them into those that
refer only to the f-structure, those that refer to semantic structure, and those
that are relevant for information structure. The constraints labelled (A) specify
the f-structure that is shown, and the constraints labelled (B) specify information structure roles. The constraints in (C) define the information structure for
this utterance. We do not repeat the equation specifying John as topic, which
is contextually provided as well as being optionally specified on the phrase
structure rule.
(42)

(A)

(B)

SUBJ
m :

(j

j :[ DF

DF)= TOPIC

(m
(r

DF)

DF)

PRED

OBJ

m :[ DF

= FOCUS

r :[ DF

= FOCUS

john (j (j DF))
married (m (m
rosa (r (r DF))
m =j
m =r

(C)

(m PRED) = MARRYSUBJ, OBJ

(m SUBJ)=j
(j PRED) = J OHN
(m OBJ)=r
(r PRED) = ROSA

MARRYSUBJ, OBJ



j : PRED J OHN


r : PRED ROSA

TOPIC ]
FOCUS ]
FOCUS ]

DF))

We can rewrite and simplify the equations in (C) as we did above, using the
equalities in (B) and in the last two lines of (C) to produce a compact description of the required information structure for this utterance:
(43)

john (m TOPIC)
married (m FOCUS)
rosa (m FOCUS)

(C)

m :

TOPIC
FOCUS

{ John }

married

Rosa

Further simplifying, and using the fact that married-Rosa can be derived by
linear logic proof from married and Rosa:

88

Syntax and information structure



(44)
m :

TOPIC
FOCUS

{ John }

{ married-Rosa }

4.3.4 A short text

With the tools provided so far, we can analyse example (9b) from Chapter 3,
taken from Lambrecht (1994:148) and repeated in (45):
(45) a. Whatever became of John?
b.

He

married Rosa,

TOPIC

FOCUS

c. but he
TOPIC

didnt really love

her.

FOCUS

TOPIC2

The derivation of the meaning of (45b) proceeds exactly as the derivation in

the preceding section of John married Rosa, with John the topic and married
Rosa the focus, except that the topic is realised by an unstressed pronoun rather
than the name John.
The analysis of (45c) is more interesting, since we now have two topical
phrases, the primary topic John and the secondary topic Rosa. To avoid introducing complications related to the analysis of pronouns and adverbs that
are not relevant for current discussion,7 we will simplify the example to John
didnt love Rosa, assuming that the discourse context and information structure
roles are the same as in (45). We then have the following simplified c-structure
and f-structure:
7 For glue-based analyses of pronouns, see Dalrymple et al. (1997), Dalrymple (2001: Chapter 11), and Asudeh (2004, 2005); for a glue-based analysis of adverbial modification, see Dalrymple (2001: Chapter 10).

Our architecture

89

(46) John didnt love Rosa.

IP
I

NP
N

VP

didnt

V

John

LOVE SUBJ, OBJ

POLARITY

NP

love

N

PRED

PRED
SUBJ


OBJ

PRED

J OHN
ROSA

Rosa
Continuing to ignore tense, aspect, and agreement features at both the syntactic
and semantic levels, the c-structure annotations are largely the same as in (41)
above:
(47)

IP

(l SUBJ)=j
l =j
((j DF)=TOPIC)

I

PRED

l :
SUBJ

OBJ

N



j : PRED J OHN


r : PRED ROSA

VP

didnt
(l POLARITY) =
[ P.not(P ):l l ] (l (l

John
(j PRED) = J OHN
john (j (j DF))

LOVE SUBJ, OBJ

POLARITY

NP

V
DF))

NP
(l OBJ)=r
l =r

love
(l PRED) = LOVE SUBJ, OBJ
[x.y .love(x, y ):j (r l )] (l (l

N
DF))

Rosa
(r PRED) = ROSA
rosa (r (r DF))

90

Syntax and information structure

We harvest these functional annotations from the c-structure, obtaining this

result:
(48)

(l SUBJ)=j
l =j
((j DF)=TOPIC)
(j PRED) = J OHN
john (j (j DF))
(l POLARITY) =
[ P.not(P ):l l ] (l (l DF))
(l PRED) = LOVE SUBJ, OBJ
[x .y .love(x, y):j (r l )] (l (l
(l OBJ)=r
l =r
(r PRED) = ROSA
rosa (r (r DF))

DF))

We can rearrange these as before, into those labelled (A) specifying the fstructure, those labelled (B) specifying information structure roles, and those
labelled (C) defining the information structure for this utterance. To the annotations derived from the c-structure above, we add the information, given
in (B), that didnt and love (corresponding to the f-structure labeled l) are in
focus, and John (j) and Rosa (r) are topical:
(49)

(A)

(l SUBJ)=j
(j PRED) = J OHN
(l PRED) = LOVESUBJ, OBJ
(l OBJ)=r
(r PRED) = ROSA

(B)

(j DF)=TOPIC
(l DF)=FOCUS
(r DF)=TOPIC

(C)

l =j
john (j (j DF))
[ P.not(P ):l l ] (l (l DF))
[x .y .love(x, y):j (r l )] (l (l
l =r
rosa (r (r DF))

DF))

The equations in (A) characterise the f-structure shown in (47), as required, and
we do not discuss them further. Simplifying the equations in (C) according to
the equalities in (B) and the first and fifth lines in (C), we have:

Our architecture
(50)

john (l TOPIC)
[ P.not(P ):l l ] (l FOCUS)
[ x.y .love(x, y ):j (r l )] (l
rosa (l TOPIC)

91

FOCUS)

This results in the following information structure:

(51)
l

TOPIC

FOCUS

Rosa


x.y .love(x, y ):j (r l )

P.not(P ):l l

John

Here we have two topical elements, the primary topic John and the secondary
topic Rosa. As discussed in Chapter 3, these are distinguished by prominence:
we assume a scale of prominence on which the primary topic is more prominent than the secondary topic. Prominence is determined by discourse context,
and can be represented by a feature or combination of features at semantic
structure, where other features of discourse referents are represented. We do
not specify a particular method of representing prominence here: for our purposes, it is sufficient to allow a sentence to have multiple topics, since our
concern is the marking of topicality by agreement and casemarking. Indeed, in
most of the languages we examine in subsequent chapters, grammatical marking does not explicitly signal secondary as opposed to primary topic, but only
that the marked argument is topical.
4.3.5 Information structure and its place in grammar
The content of the projections levels of linguistic structure which we
assume, and the relations between them, are different from previous proposals
within the LFG framework. We assume the overall architecture in (52), where
i-structure is information structure, and s-structure is semantic structure, and
the lines connecting the levels are labelled with the name of the function that
establishes the relation between them:
(52)

c-structure

f-structure

s-structure

i-structure

92

Syntax and information structure

This follows standard LFG assumptions in assuming that the function relates parts of the c-structure tree to f-structures, and the function relates fstructures to semantic structures. It differs from the approach of King (1997),
who proposes the following arrangement of linguistic levels:
(53)

c-structure
f-structure

i-structure
s-structure
semantics

Similarly, Mycock (2006) proposes that information structure is directly related to both c-structure and a level of prosodic structure. King and Mycock
share the view that there is a direct connection between i-structure and cstructure, while the connection between f-structure and i-structure is indirect.
This view is based in Kings (1997) exploration of architectural possibilities
for relations between levels, discussed in Section 4.1.2 above, and her observation that f-structure constituents are often either too small or too large to
define information structure roles: this is the granularity problem. Mycock
(2006:91) raises a similar issue in her analysis of constituent questions, noting that the analysis requires reference to units that do not match f-structure
constituents.
Kings (1997) solution to the granularity problem was to posit a direct link
between c-structure and i-structure: c-structure is more fine-grained than fstructure (for instance, V, V , VP, I , and IP are separate c-structure nodes, but
all correspond to the same f-structure), and it seemed that appeal to the more
fine-grained c-structure would allow definition of i-structure constituents of
the proper size. Both King and Mycock posit a link between i-structure and
s-structure, to ensure that all parts of the meaning of an utterance receive an
information structure role.
Our approach is not susceptible to the granularity problem, which plagues
approaches that treat f-structures as akin to meanings and use f-structure representations to encode information structure roles and relations. In our approach,
specifying a particular information structure role for an f-structure constituent
means that the meaning constructors contributed by the lexical head of the
f-structure (or heads, if the f-structure corresponds to a functional category as
well as a lexical category recall the discussion in Chapter 2, Section 2.3)
are associated with the information structure role, and does not entail that the
arguments and modifiers of that head are also associated with that role. This
is unlike approaches which use f-structure units to specify information structure roles, since (as King correctly points out) reference to the contents of an

Conclusion

93

f-structure also involves reference to the arguments and modifiers that appear
in that f-structure. For example, in (51), the meaning constructor associated
with the verb appears in the FOCUS set. This does not entail that the arguments
of the verb must also appear in the FOCUS set; indeed, in (51) the subject and
object of the verb are topical, not in focus.
Our approach also ensures that every meaning constructor bears some information structure role, by associating an equation of the following form with all
meaning constructors:
(54) [meaning-constructor] ( (

DF))

According to this equation, the meaning constructor must play some role at
i-structure, determined by the value of ( DF). In most cases, this value is
unspecified, and determined by linguistic and discourse context (although in
some special cases, the value may be lexically specified: for example, question words may be intrinsically specified as FOCUS). Even though the value
is unspecified, the equation requires some value to be chosen; our analysis
requires all meaning constructors to be integrated into i-structure.8
An advantage of our proposal in the present setting is the direct link between
f-structure and i-structure, and the concomitant ability to specify a direct relation between grammatical function and information structure role; as we will
see in subsequent chapters, such specifications are important in the grammar
and structure of many languages.
4.4 Conclusion
We have presented a formal theory of information structure and its place in the
overall architecture of Lexical Functional Grammar. Information structure is
intimately related to semantic structure, which comports well with structured
meaning approaches as well as with Lambrechts (1994) theory. A pleasant
feature of our theory is that it allows for a simple specification of the information structure role of an argument, by providing a value for the DF feature
within its semantic structure. In the following chapters, we will show how
agreement and casemarking can affect the specification of DF as TOPIC for a
range of arguments in the clause.
8 In

this way, our proposal differs from analyses such as Kwon and Zribi-Hertz (2008), who
assume that certain unmarked subjects and objects in Korean are entirely excluded from information structure; we assume that all of the meaning contributions of a sentence play a role at
information structure. If it is found that certain unmarked or incorporated elements do not bear
one of the four information structure roles that we assume (TOPIC, FOCUS, BACKGROUND,
or COMPLETIVE), it may be necessary to propose an additional information structure role for
extremely backgrounded meaning contributions.

Topicality and grammatical

marking
In a number of languages, topicality affects how a sentence element is grammatically marked. Topical marking (casemarking and agreement) is restricted
to topical subjects in some languages, but can be associated with any of several grammatical functions in other languages. In this chapter, we explore
connections involving grammatical function, grammatical marking, and topicality, demonstrating correlations between grammatical marking and topichood as well as the common association between subjects and primary topics. We will discuss topicality as it affects grammatical marking of arguments
within a single clause, although topicality has also been shown to be important
for cross-clausal long-distance agreement, as Polinsky and Potsdam (2001)
demonstrate in their analysis of Tzez.
5.1 Topical marking for different grammatical functions
Languages often single out topical arguments for grammatical marking. For
example, Bossong (1989) discusses Ayacucho Quechua (Quechuan), a language in which the topic marker qa can combine with any case marker which
signals the syntactic role of clause-level NPs. In (1a), the subject shows topical marking. In (1b), the same marking is found on a locative element, and the
subject is detopicalised and syntactically inverted.

95

96

Topicality and grammatical marking

(1) a. cheqa-paq-mi nina-para-qa chaya-ska-sqa

truth-for-Foc fire-rain-Top arrive-still-Past.3Sg
Really, the fire-rain is still falling.

(Bossong 1989:45)

b. kay-pi-qa sumaq-ta-m wina-n kawsay-kuna

this-Loc-Top good-Adv-Foc grow-3Sg crop-Pl
Here, the crops are growing well.

(Bossong 1989:44)

Similarly, the Japanese topic marker wa occurs on NPs with different grammatical functions. The topic marker replaces the subject and the object markers,
but can appear with some other case affixes/postpositions, as shown in (23):
(2) a. uti no kodomo ga koukou
ni hairu
we Gen child
Nom high.school to enter
Our child will enter high school.

(Fry 2001:138)

b. uti no kodomo wa koukou

ni hairu
we Gen child
Top high.school to enter
As for our child, she will enter high school.

(Fry 2001:139)

(3) a. [nihongo ni] wa yama

toka kawa toka itiminzi
de
Japanese in Top mountain etc. river etc. straight.line Cop
In Japanese, mountain, river, etc. are (written with) straight lines.
(Fry 2001:140)
b. [watasi no toko no ie
kara] wa kanari tooi no
I
Gen place Gen house from Top pretty far Focus
From my house, its pretty far.

(Fry 2001:140)

Given the typical association between topic and subject and more generally,
between topics and grammatical functions that are high on the grammatical
function hierarchy we expect topical marking to appear on subjects in the
majority of cases, even when it is also permitted on nonsubjects. This is confirmed by Fry (2001), who shows that 70% of wa-marked nouns in a corpus
of spoken Japanese are subjects, and less than 4% bear functions other than
subject or object.
Agreement can work in a similar way, and indeed, Comrie (2003) and others
have argued that agreeing elements are highly topical. Morimoto (2009) analyses the so-called subject-object reversal construction in the Bantu languages
Kinyarwanda and Kirundi, showing that the standard term for this construction
is misleading: in fact, no grammatical function change is involved. Instead,

Topical marking for different grammatical functions

97

she provides evidence that the construction involves topical agreement with a
pragmatically prominent argument, which may be either a subject or an object.
Kinyarwanda verb agreement in noun class is expressed by a prefix, and the
agreement controller is normally the argument in immediately preverbal position. In example (4a), the verb shows noun class agreement with the topical
subject boy. In contrast, in (4b) book is topical, and the subject is in focus.
Agreement in this case is with the object.
(4) a. umuhuungu a-ra-som-a
igitabo
1.boy
1-Pres-read-Asp 7.book
The boy is reading the book.

(Morimoto 2009:201)

b. igitabo ki-som-a umuhuungu

7.book 7-read-Asp 1.boy
The BOY (focus) is reading the book (topic).
(Morimoto 2009:201)
Morimoto supports her analysis by demonstrating that the preverbal argument
in (4b) (book) does not have subject status; her evidence comes from syntactic tests such as relativisation, gapping in coordinate structures, and subject-tosubject raising. Further, the postverbal argument in the reversal construction
does not display the usual object properties: it does not have to be located in
the canonical immediately postverbal object position, it does not trigger object
agreement, and it can be neither relativised nor pronominalised. Morimoto
concludes that the postverbal argument in the reversal construction is in fact
the subject, and that agreement in Kinyarwanda and Kirundi is controlled by
the preverbal topic, independently of its grammatical function (subject or object).
Aleut (Eskimo-Aleut) verb agreement also depends on information structure role. According to Golovko (2009), the verb in Aleut always agrees
with a topical argument, which may be the subject, the object, the possessor of the subject or object, or an oblique goal. Aleut has two paradigms of
person-number agreement with the topic, which Golovko calls primary agreement and secondary agreement. Primary agreement is with a topical subject
or a topical possessor of the subject. Secondary agreement is used if a topicalised argument is an object, the possessor of the object, or in some cases
the oblique goal argument of a verb like give.1 For example, in (5a), agreement with the topical subject is expressed via the primary paradigm. In (5b),
the verb shows third person plural agreement, although the subject (teacher)
is singular. Here, agreement is with the fronted topical object, expressed via
1 There are other complications in the Aleut agreement system which we do not discuss here;
see Golovko (2009) for a detailed description.

98

Topicality and grammatical marking

the secondary paradigm; the nontopicalised subject argument takes the special
relative form, which also marks possessors.2
(5) a. ang ag ina-s suna-x ukuxta-ku-s
person-Pl boat-Sg see-NonFut-3Pl
The people see the boat.
b. anik du-s achixana-m achixa-ku-nis
child-Pl teacher-Rel.Sg teach-NonFut-3Pl
The children, the teacher teaches them.

(Golovko 2009)

Possessor topics cannot be overtly expressed within the same clause, and must
be recoverable from the preceding context. However, Golovko argues that the
possessor of the subject in (6) is topical, although it is not expressed by an
independent element; the verb shows singular agreement with the possessor of
the subject, rather than plural agreement with the subject:
asxinu-un
kidu-ku-x
(6) la-n is
son-3Sg.Pl daughter-2Sg help-NonFut-3Sg
His/her sons help your daughter.

(Golovko 2009)

In contrast, overt possessors cannot be analysed as topical, and cannot control

agreement:
la-n is
asxinu-un
kidu-ku-s
/
(7) ajaga-m
woman-Rel.Sg son-3Sg.Pl daughter-2Sg help-NonFut-3Pl
*kidu-ku-x
help-NonFut-3Sg
The womans sons help your daughter.
(Golovko 2009)
There is no independent evidence that the agreeing possessor in example (6)
plays a role as argument of the main predicate; rather, agreement is with the
possessor topic.
The verb may also show agreement with a goal topic of a verb like give:
(8) a. lax
asxinu-m n aan kanfiita-x ag i-ku-x
boy-Sg girl-Rel.Sg to.3Sg candy-Sg give-Nonfut-3Sg
The boy gave the girl the candy.
kanfiita-x n aan ag i-ku-u
b. asxinu-x la-m
girl-Sg boy-Rel.Sg candy-Sg to.3Sg give-Nonfut-3Sg
The girl, the boy gave her a candy.
2 We

have transliterated Golovkos examples from Cyrillic.

Topical marking for different grammatical functions

99

In (8a), the subject is topical, and the verb shows primary third person agreement. In (8b), the goal argument girl is topicalised and fronted, and the verb
shows secondary third person agreement with the goal.
These examples illustrate that grammatical marking of primary topicality
need not be restricted to one grammatical function, but may target various functions including the subject. In Ayacucho Quechua, Japanese and Korean, topical markers can cooccur with nearly every verbal dependent. Some languages
display additional grammatical constraints on marking: in Kinyarwanda and
Kirundi, topical agreement is possible only with the subject and the object, the
two grammatical functions that are highest on the grammatical function hierarchy. Aleut primary agreement is controlled by the subject or its possessor,
while topicalisation of other sentence elements may be expressed by secondary
agreement. These data confirm that grammatical marking (agreement or casemarking) may target (primary) topics, and may sometimes be constrained by
additional syntactic and semantic factors.
The formal analysis of topical marking in these languages is straightforward,
given the theory of information structure and its relation to syntax presented
in Chapter 4. The Quechua topic marker qa and the Japanese topic marker wa
are associated with the constraint in (9):
(9) Topic marking (any grammatical function):
(

DF)

= TOPIC

This constraint ensures that the argument bearing topic marking is associated
with the information structure role of TOPIC, in accordance with the theory presented in Chapter 4, Section 4.3. It does not specify a particular grammatical
function for the argument it is attached to, since topic marking in these languages is syntactically unconstrained: it can be associated not only with the
subject, but with other grammatical functions as well.
Topical agreement may be similarly underconstrained, and may be associated not only with the subject, but with other grammatical functions as well.
For Kinyarwanda and Kirundi, topical agreement marking is controlled by either the subject or the object of the verb. The constraints associated with a verb
with Class 7 agreement morphology are:
(10) Class 1 agreement with topical subjects or objects, Kinyarwanda/Kirundi:
( {SUBJ|OBJ}) = %t
(%t NOUNCLASS) = 1
(%t DF) = TOPIC
This specification uses a local name, which is used when constraints are placed
on an f-structure whose grammatical function is uncertain or undetermined; see

100

Topicality and grammatical marking

Dalrymple (2001:146148) for definition and discussion. Local names always

begin with a percent sign (%). The first line of the specifications defines the
local name %t as referring either to the subject or to the object of the agreeing
verb. The second line requires noun class 1 for %t, and the third line requires
%t to bear the information structure role of topic.
In Aleut, topical agreement is controlled by the subject, the object, the
oblique goal argument of a verb such as give, or a possessor of the subject or object. Third person topical primary agreement involves the following
specifications:
(11) Primary agreement with topical subjects or possessors of the subject,
Aleut:
( SUBJ (POSS)) = %t
(%t PERS) = 3
(%t DF) = TOPIC
This expression also uses the local name %t to define the controller of agreement and to specify it as topical. Parentheses around POSS allow the topical
argument to be the possessor of the subject: the path specifying the agreement
controller can be either SUBJ or SUBJ POSS. First and second person primary
topical agreement is defined similarly, except that the controller of agreement
is specified as first or second person rather than third person. Third person topical secondary agreement is with the object, its possessor, or an oblique goal
argument:
(12) Secondary agreement with topical objects, possessors of the object, or
oblique goals, Aleut:
( {OBJ (POSS)| OBL GOAL }) = %t
(%t PERS) = 3
(%t DF) = TOPIC
Constraints for first and second person verbs are similar, except for the specification of the person value as 1 or 2 rather than 3.
5.2 Grammatical encoding of topical subjects
In some languages, subjects which are also topics are grammatically marked;
this is differential subject marking, or subject marking which depends on
topicality. Topical subjects often receive special casemarking, and also tend
to trigger more agreement than nontopical or focused subjects (Comrie 2003,
Siewierska 2004:159ff, Corbett 2006:197-204). Conversely, Lambrecht and
Polinsky (1997) and Lambrecht (2000) argue that in sentence-focus construc-

Grammatical encoding of topical subjects

101

tions where both the subject and the predicate are in focus, detopicalisation of
the subject may be accompanied by suspended subject-verb agreement.
Somali (Semitic) provides good evidence for these generalisations, as shown
by Saeed (1984, 1987). There are two subject forms in Somali, the absolutive
case and the subject (nominative) case. The absolutive is the basic citation form and the case that marks some subjects, objects and obliques. The
nominative is derived from the absolutive by tonal alternations and sometimes
by adding a final vowel. Crucially, the nominative is found only on topical
subjects, while nontopical or focused subjects must stand in the absolutive
form. Furthermore, with focused subjects the Somali verb must be in the socalled relative form, a paradigm in which agreement is reduced: the relative
paradigm includes only three distinct forms, in contrast to the five or six forms
typically found in paradigms used with topical subjects.
For example, in (13a) and (14a) the subjects are topical and nominative. The
particles wuu and w`ay encode declarative marking and positive polarity, and
both the verb and the declarative particle express agreement with the subject.
In contrast, (13b) and (14b) contain the focus marker ay`aa, indicating that the
subject is focused; the verb is in reduced relative form, and there is reduced
verb agreement.
(13) a. nnku
wuu
imanayaa
man.Nom Decl.3Sg.Masc come.Pres.Prog.3SgMasc
The man is coming.

(Saeed 1987:216)

b. nnka
ay`aa imanaya
man.Abs Foc come.Rel.1Sg/2Sg/3SgMasc/2Pl/3Pl
The MAN is coming.

(Saeed 1987:216)

(14) a. Nimankii w`ay

keeneen
men.Nom Decl.3Pl bring.Past.3Pl
The men brought (it).

(Saeed 1987:217)

b. Nimanki ay`aa keenay

men.Abs Foc bring.Rel.1Sg/2Sg/3SgMasc/2Pl/3Pl
The MAN brought (it).

(Saeed 1987:217)

Other languages pattern similarly. Maslova (2003a) shows that in Kolyma

Yukaghir (isolate), topical subjects trigger agreement in number (singular or
plural) and person (first, second and third). Thus, the verbal paradigm for topical subjects consists of six forms. When the (intransitive) subject bears focus
marking, the verb exhibits reduced agreement, distinguishing only third person

102

Topicality and grammatical marking

plural from the rest. Example (15) illustrates this: the subject is focused, as
indicated by focus marking, and the verb shows reduced agreement:3
(15) o,
nah
a omo-se soromo-k kel-u-l
Interj very good-Attr person-Foc come-EpentheticVowel-SubjFoc
Wow, a very good person has come.

(Maslova 2003a:91)

In contrast, the subject in example (16b) is topical, and the verb shows full
agreement:
(16) a. The lake-king was delighted and let all the fishes go. He sent them
into the river along watercourses.
b. tamun jel
at tude+sam kewe-s
c obul laNin
that after he+self go-Perf.Intr.3Sg sea Dir
And then he went away into the sea himself. (Maslova 2003a:572)
Maslova (2003a) argues that focused and topical subjects in Kolyma Yukaghir
have the same syntactic properties, despite the fact that they bear different
casemarking and trigger different agreement. It is the information structure
role of the subject that determines its casemarking as well as which agreement
paradigm is used.
Formally, we analyse topical subject marking in a way similar to the analysis presented above, for marking of any topical element; the main difference is
the additional requirement that the marked argument must be a subject. Nominative/topical casemarking in Somali is associated with the constraints in (17):
(17) Casemarking on topical subjects:
(SUBJ )
( DF) = TOPIC
The first of these two specifications, (SUBJ ), requires the argument bearing
nominative/topical case (represented by the f-structure metavariable ) to be
the subject of its clause, in line with the constructive case approach discussed
in Chapter 2, Section 2.7. The second line specifies that the casemarked arguments discourse function is TOPIC. Together, these two specifications ensure
that the casemarked argument is the SUBJ at f-structure, and that its semantic
contribution is associated with the information structure role of TOPIC.
Agreement as a marker of topicality works similarly. Each member of the
Yukaghir full agreement paradigm is associated with the following specification:
3 Maslova glosses the focus ending on the focused subject as Pred, for Predicative Case. We
have glossed it as Foc, for Focus marking.

Subjects and topichood

103

(18) Agreement with topical subjects:

((

SUBJ) DF)

= TOPIC

Here ( SUBJ) is the verbs subject, and ( SUBJ) is the subjects semantic
structure. Specifying the value TOPIC for the feature DF within the subjects
semantic structure ensures that the semantic contribution of the SUBJ is associated with the TOPIC information structure role. The reduced agreement
paradigm does not mark the subject as topical, and is not associated with this
constraint.
5.3 Subjects and topichood
Chapter 3 showed that topicality is defined on referents rather than linguistic
expressions, so a topic is not a priori constrained to be encoded as a particular
grammatical function. However, there are certain cross-linguistic tendencies
governing the syntactic expression of topics. We understand these as specifying the preferred mapping (alignment) between information structure and
syntax: even though sentence form is not fully determined by function, some
of its aspects may be motivated by function, including information structure
role.
We have seen that there is a strong association between topic marking and
subjecthood. This is not arbitrary: the correlation between (primary) topic and
subject has been much discussed, at least since Hockett (1958) and Keenan
(1976) (see Givon 1976, Comrie 1989, Bossong 1989, Yamamoto 1999, and
Erteschik-Shir 2007, among others). Givon (1976) suggests that subjects are
grammaticalised primary topics, and Lambrecht (1994:132137) argues that
subjects are unmarked topics. Given Lambrechts typology of focus structures, this implies that the topic-comment articulation (wide or predicate focus) is unmarked information structuring. Predicate focus utterances serve to
augment the addressees information about a referent under discussion. This
is communicatively more common than identifying an argument in an open
proposition (narrow focus) or reporting a new and unexpected event (sentence
focus). The subject-predicate structure, then, iconically reflects the unmarked
topic-comment information structure.
More generally, this reflects a tendency toward isomorphism between pragmatic (informational), semantic, and syntactic prominence. On one hand, the
topic referent is cognitively salient because it is characterised by an aboutness relation to the proposition (see Chapter 3, Section 3.2.2). On the other
hand, the subject normally corresponds to a participant in the described situation that has the most prominent semantic role and is often obligatorily expressed as a syntactic argument. Speakers tend to produce clauses presenting

104

Topicality and grammatical marking

situations which single out one participant. The subject is the argument that is
singled out as syntactically prominent, and hence is a good choice to express
the referent to which the aboutness condition applies. Bresnans (2001:98)
classification of subject as a grammaticalised discourse function is based on
these factors; she also points out that c-structure properties can reflect this
prominence, in that subjects as well as topics and foci can be required to precede or c-command other constituents in the clause.
The correlation between subjects and topics can be manifested in a number of ways, more strongly in some languages than in others. We have seen
that topical marking in some languages is restricted to subjects. Other languages impose semantic constrains on subjects related to the notion of topicworthiness (see Chapter 3, Section 3.2.3): only subjects with topic-worthy
features are acceptable. For instance, Givon (1976) notes that in Malagasy
and Kinyarwanda, subjects must be definite or generic. But even in languages
without semantic restrictions on subjects, such as English, the overwhelming
majority of subjects are definite. Textual counts demonstrate that in coherent
discourse, the majority of subjects are pronominal, and unaccented pronouns
are normally topical (Lambrecht 1987 and others).
Some languages place even stronger restrictions on the relation between syntax and information structure: they disallow nontopical subjects altogether. For
example, Bresnan and Mchombo (1987:778), citing Bokamba (1981), show
that in Dzamba (Bantu), all nonsubject arguments can be questioned in situ,
but the subject cannot be questioned in its initial position: it must be questioned using a headed or headless relative clause.
(19) a. *Nzanyi o -wimol-aki o -Biko e-kondo lOO me?
[Who told Biko a story/tale today?]
b. o -Moto o -wimol-aki o -Biko e-kondo lOO me nzanyi?
The person who told Biko a story/tale today is who?
(Bresnan and Mchombo 1987:778)
Since wh-questions are inherently associated with narrow focus (see Chapter 3,
Section 3.2.1), this indicates that Dzamba does not allow focused subjects, and
indeed Bresnan and Mchombo (1987) claim that Dzamba subjects are grammaticalised topics and for this reason cannot bear the focus role.
Information structure role may affect the grammatical realisation of arguments. We will discuss the effect of topicality on the realisation of arguments
as objects in Chapter 9; here we briefly discuss its effect on the realisation of
subjects. In some languages, a nontopical or focused argument may not be
realised as subject, but must be demoted to a lower function on the hierarchy.
According to Lambrecht (2000), the functional motivation for this is paradigmatic contrast: topicless sentence-focus constructions take on a form which

Subjects and topichood

105

distinguishes them from topic-comment structures in which the subject is topical.

In particular, the thematically highest argument of the verb may be assigned
to different grammatical functions, depending on its information structure role.
In constructions where it is not topical, the thematically highest argument tends
to be associated with the behavioural features of objects. Creissels (2008)
shows that in French impersonal constructions, the argument of an intransitive
verb can appear postverbally:
(20) a. une femme viendra
a woman come.Fut.3Sg
A woman will come.
b. il
viendra
une femme
Expl come.Fut.3Sg a woman
A woman will come (there will be a woman coming).
(Creissels 2008:155)
Here the postverbal NP is located in the canonical object position and patterns with objects, rather than subjects, with respect to a number of syntactic
properties. Creissels concludes that it is a syntactic object, although it has the
same semantic role as the canonical subject in (20a), and Cummins (2000) argues for the same conclusion. The split is motivated by information structure
considerations: while the subject in (20a) is topical, (20b) is a presentational
all-new construction, and the postverbal NP belongs to the broad focus domain. Creissels (2008:157) refers to such cases as pragmatic conditioning on
fluid transitivity.4
Nikolaeva (2001) argues that in transitive clauses in Northern Ostyak (Uralic)
the subject must bear the primary topic role. When the agent argument is nontopical, passivisation is required. This claim is supported by the following
observations. First, topicalisation of a non-agent argument triggers passivisation. In passive clauses, the agent is encoded as a locative NP, while another
topical argument is the subject. Kulonen (1989) shows that arguments with different semantic roles can be realised as the subject in a passive construction:
patient/theme, recipient/benefactive, location, goal, and time. The examples
in (21) demonstrate that the passive construction is required when the primary
4 Following

work by Platzack (1983) and others, Ldrup (1999) proposes a similar analysis
for the presentational focus construction in Mainland Scandinavian, which involves an expletive
in subject position and a postverbal, presentationally focused NP. However, Borjars and Vincent
(2005) argue for an alternative analysis of this construction in which both the expletive and the
presentationally focused postverbal NP bear the subject role.

106

Topicality and grammatical marking

topic, the element that bears the aboutness relationship to the proposition, is
not associated with the agent argument.
(21) a. What about Peter?
b. (luw) Juwan-na re:sk-@s-a
he John-Loc hit-Past-Pas.3SgSubj
John hit him. (literally: He was hit by John.)
c. (luw) Juwan-na ke:si-na ma-s-a
he John-Loc knife-Loc give-Past-Pas.3SgSubj
John gave him a knife. (literally: He was given a knife by John.)
Unlike English, the use of the non-passive construction in this context is strictly
ungrammatical; the active counterparts of (21b,c) cannot be used in this context:
(22) a. *Juwan (luwe:l) re:sk@-s-li
John he.Acc hit-Past-Obj.3Sg.Subj
John hit him.
b. *Juwan luw e:lti ke:si ma-s
John he to knife give-Past.3Sg.Subj
John gave him a knife.
Additional evidence for the topichood of the subject is provided by sentences
with a focused agent. Questions and answers involving the agent require passivisation.
(23) a. kalaN xoj-na we:l-s-a?
reindeer who-Loc kill-Past-Pas.3SgSubj
Who killed the reindeer?
b. Juwan-na we:l-s-a
John-Loc kill-Past-Pas.3SgSubj
John did.
The ungrammaticality of the non-passive counterparts of (23) in this context
immediately follows from the requirement for the subject to be associated with
topic, together with the assumption that focus and topic cannot correspond to
the same sentence element. Finally, as mentioned above, topic expressions

Conclusion

107

must be referential. In Ostyak, quantified expressions such as anybody or nobody do not occur as subjects of transitive clauses. As example (24) demonstrates, when they correspond to an agent-like argument, the clause must be
passivised.
(24) a. tam xu:j xoj-na an wa:n-s-a
this man who-Loc Neg see-Past-Pas.3SgSubj
Nobody saw this man.
b. *xoj tam xu:j an wa:nt-@s
/ wa:nt-@s-li
who this man Neg see-Past.3SgSubj see-Past-Obj.3SgSubj
Nobody saw this man.
These examples show that Ostyak grammar displays a strong correlation between the grammatical function of subject and the (primary) topic role.
Other languages are less strict and allow for nontopical subjects, but only
when special devices are used. That is, the subject referent is interpreted as
topic unless there are syntactic or prosodic cues to the contrary, such as appearance of a topical nonsubject element in a noncanonical sentence-initial
position, or prosodic prominence of focus subjects in event-reporting (sentence focus) sentences. The latter is illustrated by the following contrast from
Lambrecht (1994).
(25) a. Hows your neck?
My neck/It HURTS.
b. Whats the matter?
My NECK hurts.

(Lambrecht 1994:137)

The answer in (25a) is a predicate focus structure: the comment hurts provides
new information about the topical referent under discussion (neck). In contrast, (25b) is an event-reporting sentence focus (or thetic) utterance: here, the
new information is associated with the whole proposition, so the subject is not
topical. This difference is expressed formally: the topical subject corresponds
to an unaccented NP and can be pronominalised, while the nontopical subject
is a prosodically prominent lexical NP. The subject in (25b), as opposed to
(25a), is structurally and functionally marked.
5.4 Conclusion
We have shown that primary topicality can be expressed by case or agreement,
and that in some languages this marking is compatible with any of several
grammatical functions. However, unmarked primary topics are subjects, and

108

Topicality and grammatical marking

in many languages the marking of primary topicality is restricted to subjects:

topical subjects are associated with more grammatical marking (case and/or
agreement) than nontopical subjects. We also find languages where nontopical
arguments cannot appear as subjects, but must be demoted through voice-like
alternations.

Topical marking of nonsubjects

Though the correlation between topicality and grammatical marking has been
well studied for subjects, it is less well studied for nonsubject topics. The languages we discuss in this chapter show clearly that topicality can be a relevant
factor in determining patterns of grammatical marking of nonsubjects. In these
languages, marking is associated with sentence elements bearing any one of a
variety of grammatical functions other than subject, but is determined partially
or completely by reference to topichood. In this sense, these patterns are similar to those discussed in Chapter 5, Section 5.1, except that in the languages
discussed here, marking is permitted only for nonsubjects, and disallowed for
subjects.
6.1 Casemarking of topical nonsubjects
6.1.1 Persian
The Persian (Iranian) postposition ra (with various colloquial and dialectal
variants) marks arguments bearing various grammatical functions, and its distribution has often been argued to be conditioned by information-structural
factors. First, certain time adverbials can be marked by ra, such as, for example, hafteye a yanda next week, emsb tonight or tabestan in summer.

109

110

Topical marking of nonsubjects

(1) a. kmal emsb-o inja mi-mun-e

Kamal tonight-RA here Impf-stay-3Sg
Tonight Kamal is staying here.

(Mahootian 1997:121)

b. hafte-ye a yanda-ro esterahat mi-kon-am

Impf-do-1Sg
week-Ez coming-RA relax
As for the next week, I will relax.

(Karimi 1990:143)

Furthermore, ra may be present on floating topics located at the left periphery of the sentence and cross-referenced by a clitic. The floating topic can
correspond to a number of different grammatical functions. In (2a), it corresponds to an oblique object of the verb laugh, while in (2b) it corresponds to
a comitative adjunct.
(2) a. man-o beh-me mi-xand-e
I-RA at-1Sg Impf-laugh-3Sg
She laughed at me.
raqsid-am
b. sasan-o bah-a
Sasan-RA with-3Sg dance.Past-1Sg
Sasan, I danced with him.

(Karimi 1990:143)

(Karimi 1990:154)

The floating topic phrase is often associated with the possessor of the object,
which is also marked by ra.
bast-am
(3) masin-o dar-es-o
car-RA door-3Sg-RA close.Past-1Sg
As for the car, I closed its door.

(Karimi 1990:143)

Finally, ra is a frequent object marker:

(4) man ketab-ra xarid-am
I
book-RA buy.Past-1Sg
I bought the book.

(Dabir-Moghaddam 1992:557)

Thus, the postposition ra marks direct objects, adverbials and floating topics.
In all of these functions, ra is optional. We maintain that its distribution on
non-objects is determined at least in part by information structure: ra appears
on nonsubject topics as well as on some nonsubject phrases with features of
topic-worthiness, such as definiteness.1
1 Dabir-Moghaddam (1992) mentions one exceptional case where r
a appears on the main
clause subject and also corresponds to a nonsubject gap within the relative clause. We do not
know how this example fits with the overall distribution of ra.

Casemarking of topical nonsubjects

111

Adverbials marked with ra must be interpreted as topical. The relevant utterances are construed as information about the relation between the subject
referent and a certain time span which delimits the action performed by the
subject referent. For instance, in example (5) summer is mentioned in the previous context, and so the speaker has reason to believe that the addressee considers it important and can expect a certain pragmatic relation between summer
and the subject referent to be expressed. The unmarked adverbial tabestan is
impossible or strongly dispreferred:2
(5) a. What are your plans for the summer?
esterahat mi-kon-am
b. tabestan-ra/?*tabestan
Impf-do-1Sg
in.summer-RA/ in.summer relax
In summer/as for the summer, I will relax.
Example (5) contrasts with example (6):
(6) a. When will you finally relax?
b. tabestan/?*tabestan-ra
esterahat mi-kon-am
in.summer/ in.summer-RA relax
Impf-do-1Sg
I will relax in summer.
In example (6), summer is not pragmatically presupposed and constitutes part
of the new information associated with the utterance, so the adverbial in summer is in focus; here, the distribution of ra is opposite to (5).
Similar considerations apply to floating topics. When a constituent marked
by ra is a floating topic, as in (2) and (3), fronting (topicalisation) is an additional clue to its topical status. Oblique objects or possessors are infrequent
topics, but can be interpreted as such under certain pragmatic conditions. Topical possessors marked with ra tend to be inalienable; this is because a statement about an inalienably possessed entity also contributes information about
its possessor, without which the inalienably possessed object cannot be conceptualised. According to our consultants, (3) is possible only in a context
where the relation between the subject referent and the car (possessor of the
object door) is under discussion. For example, it can be construed as an answer to the question What did you do to the car? (What I did to the car is, I
closed its door), but is an inappropriate answer to What have you just done?,
which does not presuppose any pragmatic association between the speaker and
the car. The referent of the object, door, is less pragmatically important than
the referent of the external possessor car. The utterance is not construed as
2 Examples for which the source is not mentioned come from personal communication with
Ghazaleh Khad and Shamsi Saber.

112

Topical marking of nonsubjects

being about the relation between the subject referent and the object door, but
rather about the relation between the subject referent and the possessor car.
Therefore it cannot answer the question What did you do to the door of the
car?.
The situation with direct objects is more complicated, and various theories
have been proposed for the distribution of ra on objects. Windfuhr (1979)
was among the first to suggest that ra marking has to do with topicality, while
Browne (1970) and Karimi (1990) claim that specificity is the relevant factor. Lazard (1992) argues for the importance of the degree of individuation
of the ra-marked argument: the more the object referent is individuated and
affected by the action designated by the verb, the more likely it is that ra is
used. According to Ghomeshi (1997) and Bossong (1991:64), several properties are relevant for ra marking. These include topicality as well as animacy,
definiteness, and affectedness.
We are in agreement with the view that the distribution of ra on objects
is at least partly explainable in terms of topicality. In this we follow DabirMoghaddam (1992), who proposed that the main function of ra is what he
calls secondary topic marking. Though he does not provide a precise definition
of secondary topic, his understanding of this notion seems to be close to ours,
judging from the contexts he provides for his examples. We note two important
differences between his analysis and ours. First, the examples in (7) indicate
that ra can mark the primary topic as well as the secondary topic. The subjects
in (7) are focused, and since there are only two NPs in each of the examples, the
ra-marked object cannot be the secondary topic; rather, it must be the primary
topic:
(7) a. ki masin-i-ra/?*masin-i did
who car-Indef-RA/car-Indef see.Past.3Sg
Who saw a car?
xarid
b. ki ketab-i-ra/?*ketab-i
who book-Indef-RA/book-Indef buy.Past.3Sg
Who bought a book?
Second, we disagree with Dabir-Moghaddams position that topicality is the
only relevant factor in determining ra marking: we believe that topicality is
a factor for some objects, while on other objects the motivation is essentially
semantic, having to do with features of topic-worthiness.
In particular, ra-marking works differently for definite and indefinite objects. As shown by Lazard (1982), all definite objects must be marked, independent of their information structure function. The omission of ra in (4) with
the given translation yields strict ungrammaticality:

Casemarking of topical nonsubjects

113

(8) a. What are you eating?

b. in sib-o/*sib
mi-xor-am
this apple-RA/ apple Impf-eat-1Sg
I am eating this apple.

(Karimi 1990:148)

For the same reason, all personal pronouns, proper nouns and objects with
demonstratives must be marked even if they are in focus (Ghomeshi 1997:137).
Thus, information structure motivations are irrelevant for definite objects; they
must be marked due to their definite status.
This observation accounts for the fact that ra can appear twice in a single
clause. As evidenced by example (2b), ra can be found on both an object
and its possessor. In (9), ra marks the definite object as well as the temporal
expression just this one hour.
(9) faqat in ye saat-o in ketab-o be-xun
just this one hour-RA this book-RA Imp-read
Read this book this one hour!

(Ghomeshi 1997:151)

Karimi (1990) takes examples like (9) to contradict the topic analysis of ra, on
the basis of her assumption that the topic role is unique in a clause. As we have
seen, however, topical arguments need not be unique in the clause: Chapter 3
introduced a distinction between the primary topic and the secondary topic,
and we assume that the presence of a secondary topic entails the presence of a
primary topic. Furthermore, in constructions like (9) the object is definite, and
definite objects must be marked independently of any informational requirements. In fact, the first and the second ra in (9) have different functions: the
first marks the topicality of the temporal adjunct, while the second is licensed
by definiteness.
For indefinite objects, ra is optional. According to Lazard (1982), indefinites that require ra are either partitive or have a a certain X interpretation. That is, they must be pragmatically anchored in the sense of Lambrecht
(1994) or specific in the sense of Enc (1986) (see Chapter 3, Section 3.2.2):
the referent is chosen from a familiar set and linked to entities in the domain
of previous discourse. Indeed, nonspecific (nonreferential) objects cannot bear
ra. But the correlation between ra-marking on indefinite objects and specificity is imperfect: although indefinite objects must be specific to be marked
by ra, not all specific objects are ra-marked.
This can be shown by testing for specificity in an intensional context, using
anaphoric pronouns like it or one. If the antecedent is specific, the anaphor
must be definite, whereas nonspecific antecedents can be referred to by the
indefinite anaphor one. As Ghomeshi (1997:138139) shows, this test indicates that indefinite objects without ra can be specific.

114

Topical marking of nonsubjects

(10) jian mi-xast

ye qalam-i peyda kon-e va peyda-s
John Impf-want.3Sg one pen-Indef find do-3Sg and find-3Sg
kard
do.Past.3Sg
John wanted to find a pen and he found it.

(Ghomeshi 1997:139)

In (10), the pronominal object clitic -s it in the second conjunct refers to the
pen John wanted to find, and so the NP ye qalami a pen must be construed as
specific. Nevertheless, it does not host ra.
Recall Erteschik-Shirs (2007) examples from Danish involving indefinite
objects with and without relative clauses, cited in Chapter 3, Section 3.2.2 (example 7, page 54); we argued there that the relative clause renders the indefinite object specific and therefore potentially topical. The situation in Persian
seems to be analogous: ra marks a specific indefinite object as topical, and
if an indefinite object is not perceived as pragmatically salient, ra-marking is
impossible. Consider the following contrast:
(11) a. ?/*man sib-i-ra
xord-am
I
apple-Indef-RA eat.Past-1Sg
I ate an apple.
xord-am
ke az deraxt oftadebud
b. man sib-i-ra
I
apple-Indef-RA eat.Past-1Sg that from tree fell.Past.3Sg
I ate an apple that fell from the tree.
In (11a), object marking on an indefinite object is impossible. In (11b) it is
possible, though not required. This is because the object is interpreted as specific and therefore can (although need not) be topical.
In the examples discussed so far, the relation of a specific indefinite to another familiar entity is expressed in the same sentence. In example (11b) it is
expressed by a relative clause, while in example (6) of Chapter 3 it is expressed
by a modificational PP. However, in some cases this relationship seems to be
pragmatically assigned by the speaker based on extrasentential factors. In particular, an indefinite object may be interpreted as topical if it is mentioned in
the context immediately following the relevant utterance. Dabir-Moghaddam
shows that if ra is hosted by an indefinite object, the subsequent discourse must
provide more information about the object referent, for example by means of
a coordinate structure.
xarid-am
(12) a. man ketab-i/*ketab-i-ra
I
book-Indef/ book-Indef-RA buy.Past-1Sg
I bought a book.

Casemarking of topical nonsubjects

115

b. man ketab-i/ketab-i-ra
xarid-am
va ...
I
book-Indef/ book-Indef-RA buy.Past-1Sg and
I bought a book and ...

(Dabir-Moghaddam 1992:557)

An appropriate continuation for the sentence in (12) is something like sent it

to my brother. Without this elaboration, a sentence with a ra-marked indefinite object is intuitively incomplete, and speakers perceive it as unacceptable.
Although in (12b) the object referent is new to the addressee and marked as
indefinite, it receives grammatical marking due to its topical status. The topical status of the object is licensed by the fact that its referent plays a role in
the following discourse. We will show in Chapter 8, Section 8.3.1 that this
condition is not unique to Persian: topicality marking on objects in Khalkha
Mongolian works in very much the same way.
In sum, although marking on direct objects correlates with topicality, this
correlation is not absolute, unlike for adverbials and floating topics. The postposition ra marks all definite objects as well as topical indefinite objects. In
Chapter 10, we show that this situation is typical of other languages in which
DOM is conditioned partly by information structure and partly by semantics,
and we suggest that this situation can result from historical spreading of ramarking to nontopical definite objects with features typical of topics.
6.1.2 Tariana
Tariana, an Arawakan language described by Aikhenvald (2003), indicates the
topicality of nonsubject elements by means of the suffix -nuku/-naku. Aikhenvald is not completely consistent in identifying the status of this marker: it is
sometimes referred to as a case marker indicating a syntactic function (Aikhenvald 2003:158) and sometimes as a topical non-subject marker (Aikhenvald
2003:160). It can appear on objects, which are not otherwise casemarked, or on
non-objects together with casemarking indicating syntactic function; we will
treat it as indicating topicality alone. It can also appear on dependent verbs,
in which case it indicates that the subordinate clause headed by the dependent
verb contributes background information: we will not discuss these cases here.
According to Aikhenvald, objects marked with -nuku/-naku must satisfy at
least one of the following conditions: (i) they must be the topic of a narrative,
(ii) they must be referential, specific and/or definite, and (iii) they must be
pragmatically important. Parameter (i) is a strong indication of topicality in
our sense, on the assumption that discourse topics are normally interpreted as
sentence topics (though not the other way around). Parameter (iii) also strongly
suggests topicality. It remains unclear how (ii) interacts with the other two
conditions in particular, whether each of the relevant properties is sufficient
for topical marking of nonsubjects, or whether the combination of properties

116

Topical marking of nonsubjects

in (ii) reflects a minimal condition that all marked nonsubjects must share.
No discussion is provided of cases where the properties listed in (ii) conflict
with each other: for example, there are no examples with specific indefinite
objects. Still, the general tendency is clear: objects marked with -nuku/-naku
are topical. Example (13) is part of a story about gold miners; here, gold is
topical and marked with nuku:
(13) di-he-ta-pidaha
diha
3Sg.Non.Fem-see.Caus1-Caus2-Remote.Past.Reported Art
paiku-nuku
gold-Top
He showed the gold.

(Aikhenvald 2003:145)

Nontopical objects are unmarked if lexical, or marked with a different marker,

-na, if pronominal. The object women in example (14) is nontopical and
unmarked:
(14) mhada ina
pi-awada
Proh woman.Pl 2Sg-think.about
Dont think about women.

(Aikhenvald 2003:153)

In example (15), payment is nontopical and unmarked, while secrets is topical and marked:
(15) di-weni
na-na-ka-pidana
3Sg.Non.Fem-pay 3Pl-want-Sub-Remote.Past.Reported
na-pia-nipe-nuku na-kalite-pidana
du-na
3Pl-hide-Nmlz-Top 3Pl-tell-Remote.Past.Reported 3Sg.Fem-Obj
Wanting a payment, they told their secrets.

(Aikhenvald 2003:145)

Importantly, -nuku/-naku marking is not restricted to objects, but occurs on

other grammatical functions if they are topical. For instance, it may appear
on topical instrumental and locative obliques. Aikhenvald analyses such cases
as involving double case marking, although these cases are clearly different
from canonical instances of case stacking (Plank 1995, Nordlinger 1998):
one of the two case markers is actually a marker of information structure
function, similar to the Japanese examples with grammatical marking followed
by topic marking discussed in Chapter 5, Section 5.1.

Casemarking of topical nonsubjects

117

(16) a. Then she took the pestle, she carefully took her son, and pulled him
out (of a hammock).
b. diha-da
ye:da-ne-nuku dhuepaneta
Art-Cl:Round pestle-Inst-Top 3SgFem.exchange
du-kwe-ta-pidana
3SgFem-hand-Caus-Caus-Remote.Past.Rep
She exchanged (the child) with the pestle and hung it.
(Aikhenvald 2003:159)
The context in (16a) establishes the salient role of the referent pestle in (16b),
an instrumental oblique which is also marked for topicality. An external topic
which is not an argument of the clause can also be marked with -nuku/-naku:
(17) nhua-naku kida-mhana
I-Top
ready-RemotePast.NonVisual
For me, it was over.

(Aikhenvald 2003:146)

Like Persian ra, -nuku/-naku marking is never found on subjects. In fact,

subject topicality is not formally marked in Tariana, though a subject may receive overt marking for focus. Example (18) demonstrates that -nuku marking
on the object may be present if the subject is in focus (and marked as such):
(18) wha-ne wa-kesi-pe-naku hipay yapise-se-naku
we-Foc 1Pl-relative-Pl-Top ground under-Loc-Top
ma-pe-kade-naka
Neg-throw-Neg-Pres.Visual
We dont want to throw our relatives under the ground.
2003:636)

(Aikhenvald

This example also demonstrates that nuku/-naku can mark primary as well as
secondary topics: in example (18) the subject has focus marking and cannot
be the primary topic, and Aikhenvald notes that in the context in which this
sentence was produced, relatives is topical.
Aikhenvald also presents examples showing that -nuku/-naku marking can
appear twice in the same clause. In (19), it marks the object as well as the
(inalienable) possessor of the object:
(19) diha-pasi-nuku di-whida-nuku
du-pisa-taka
he-Aug-Top 3Sg.Non.Fem-head-Top 3Sg.Fem-cut-off
du-pe
3Sg.Fem-leave
She cut off the head of him, the big one.
(Aikhenvald 2003:157)

118

Topical marking of nonsubjects

In (19), the possessor and the possessed need not be adjacent. The possessor
does not have object properties, and does not seem to have been promoted to
a role within the main clause: for example, it cannot be passivised. It may
be that the first nuku-marked phrase in this construction is a floating, clauseinitial topic that is interpreted as the possessor of the object, similar to the
Persian examples in (2) and (3) above.
6.1.3 Topical nonsubject casemarking
The Tariana nonsubject topic marker -nuku/-naku is associated with almost
the same specifications as for the Quechua and Japanese markers discussed in
Chapter 5, except that the marking may appear on any argument except the
subject:
(20) Topic marking (nonsubjects), Tariana:
(SUBJ )
( DF) = TOPIC
The first of these two specifications is similar to the requirement associated
with topical subject marking, presented in (17) of Chapter 5, except for the
crucial presence of the negation operator. The requirement (SUBJ ) ensures
that the argument bearing topical casemarking (the f-structure ) is not the
subject of its clause; it may bear any other grammatical function. As in (17) of
Chapter 5, the second line specifies that the casemarked arguments discourse
function is TOPIC, as described in Chapter 4, Section 4.3.
The specifications associated with Persian ra are more complex, since ra is
not invariably associated with topics; it marks definite objects whether or not
they are topical, as well as arguments that are demonstrably topical: indefinite
objects, some adjuncts, and external/displaced topics, which we treat as bearing the f-structure role of grammaticalised TOPIC (for discussion of the grammaticalised TOPIC role at f-structure for displaced constituents, see Chapter 4,
Section 4.1.2). These constraints are captured in the following specifications:
(21) Topic marking (nonsubjects), Persian:
{(OBJ )
( DEF) = +
|({TOPIC|OBJ|ADJ } )
( DF) = TOPIC}
This specification is disjunctive, reflecting the dual nature of ra as a definite
object marker and a marker of topical elements. The disjunction is surrounded
by curly brackets, and the two parts are separated by a vertical stroke. The

Agreement with topical nonsubjects

119

first part consists of the first two lines, which specify that ra can mark definite
objects: the first line requires the argument marked with ra to be an object, and
the second line requires it to be definite (with the feature DEF +). There is no
requirement for ra-marked definite objects to be topical; in Chapter 10, we discuss the historical relation between topicality and semantic features of topicworthiness, including definiteness. The second part, the third and fourth lines
in (21), requires the ra-marked argument to be either a displaced f-structure
TOPIC, an object, or an oblique, and to bear the information structure role of
TOPIC, as in the Tariana specifications in (20).
6.2 Agreement with topical nonsubjects
6.2.1 Itelmen
In Itelmen (Chukotko-Kamchatkan) as described by Bobaljik and Wurmbrand
(2002), subject and direct object arguments are not casemarked and can be
omitted under pro-drop. The Itelmen verb has two agreement slots, primary
(prefixal) agreement and secondary (suffixal) agreement.3 The basic structure
of the verbal form is as follows:
(22) Agreement1-Mood-Stem-Aspect-Tense-Agreement2
Prefixal agreement (Agreement1) references the subject and can sometimes be
null. We are concerned here with suffixal agreement (Agreement2), which is
influenced by information structural factors: when there is a choice of agreement controller, the Itelmen verb agrees with a nonsubject topic.
Suffixal agreement in Itelmen is obligatory; following Bobaljik and Wurmbrand, we take this to be a morphological fact. There are two distinct sets of
agreement suffixes for third person nonsubject elements: portmanteau suffixes
express the features of a third person direct object and, under some circumstances, the subject, while a separate set of suffixes simply reference a third
person oblique. Suffixal agreement with first and second person nonsubjects
never makes reference to the subject. We will gloss the agreement morphemes
as either Obj or Obl, depending on the grammatical function of the agreement
controller.
Bobaljik and Wurmbrand show that exponence of the features of the subject
in suffixival agreement arises in two situations. In intransitive clauses, subject
agreement can be expressed twice, once in the prefixal slot and again in the
suffixal slot:
3 This oversimplifies the morphology of the Itelmen verb; for more detailed discussion of Itelmen verb morphology and agreement, see Bobaljik and Wurmbrand (2002).

120

Topical marking of nonsubjects

(23) kma t-ko-kicen

I
1SgSubj-come-1SgSubj
I came/arrived.

(Bobaljik and Wurmbrand 2002:(6a))

When the direct object is third person, object suffixal agreement can express
features of both the direct object and the subject:
(24) kza @lcqu-n
na
you see-2SgSubj>3SgObj him
You saw him.
(Bobaljik and Wurmbrand 2002:(12d))
It is also possible for the verb to show oblique suffixal agreement, which is
in complementary distribution with the object agreement suffixes, and does
not involve expression of subject features. The following examples illustrate
secondary oblique agreement: example (25a) shows agreement with an oblique
source argument, which appears with dative/locative casemarking, while (25b)
shows secondary oblique agreement with the otherwise unexpressed possessor
of the subject.
(25) a. kma iplX-enk
t--nen
B@pq-5Pn
I
friend-Dat/Loc 1SgSubj-take-3SgObl fly.agaric-Pl
I took fly agaric (mushrooms) from my friend.
(Bobaljik and Wurmbrand 2002:(14c))
ktxiN qaPt
b. da qusknaqu lj Gw i ploxo le-Gw in
very bad become-3SgSubj head already
Interj Q.
iPte-s-kinen
split-Pres-3SgObl
And Q. began to feel very bad, already his head is splitting.
(Bobaljik and Wurmbrand 2002:(32b))
If there are several potential nonsubject agreement controllers in a clause, the
choice of agreement controller is determined by pragmatic factors. In (26), the
verb give agrees with the subject and one of its two nonsubject arguments,
with the object competing with the oblique for the agreement slot:
(26) a. isx-enk n-z@l-a-um
kza k@ma-nk
father-Loc Imprs-give-Fut-1SgObj/Obl you me-Dat
Will father give you to me?(Bobaljik and Wurmbrand 2002:(14b))
b. isx-enk n-z@l-a-in
kza k@ma-nk
father-Loc Imprs-give-Fut-2SgObj/Obl you me-Dat
Will father give you to me? (Bobaljik and Wurmbrand 2002:(15))

Agreement with topical nonsubjects

121

In (26a) secondary agreement is with the dative indirect object me, and in
(26b) agreement is with the direct object you. The translation is roughly the
same, and Bobaljik and Wurmbrand (2002) emphasise that there is no evidence
from casemarking, word order, or other syntactic criteria that these different
agreement patterns are evidence of an alternation in grammatical functions.
Instead, they demonstrate convincingly that conditions on the choice of the
secondary agreement controller are governed by discourse-pragmatic considerations such as salience. We believe that this indicates that secondary agreement depends on information structure: when there is a choice of agreement
controllers, the verb agrees with topical nonsubjects.
The influence of topicality is clear when contextual information is provided. Bobaljik and Wurmbrand provide example (27), in which the direct
object/theme, the controller of verb agreement, is established as topical in the
context provided by the preceding sentence:
(27) maP k@man Bac? ke-nk
t-z@l-cen?
where my
knife who-Dat/Loc 1SgSubj-give-1SgSubj>3SubjObj
Where is my knife? Who did I give it to?
(Bobaljik and Wurmbrand 2002:(18b))
In this example the knife is under discussion when the second clause is produced, so the object NP triggers secondary agreement. The same verb agrees
with the indirect object in example (28), where the previous context establishes
the goal rather than the theme as topical:
(28) zlatumx piki-in.
@Nqa @nna-nk
t-zel-nen?
brother go-3SgSubj what him-Dat/Loc 1SgSubj-give-3SgObl
My brother left. What did I give to him?
(Bobaljik and Wurmbrand 2002:(19b))
The same distribution is observed in sentences without wh-question words.
(29) a. maP BacP qenu zlatumx-enk
where knife really brother-Dat/Loc
t-z@l-cen
1SgSubj-give-1SgSubj>3SubjObj
Where is the knife? Didnt I give it to my brother?
(Bobaljik and Wurmbrand 2002:(20a))
b. i kma @nna-nk
Bac t-z@l-nen
and I
him-Dat/Loc knife 1SgSubj-give-3SgObl
[My brother came]. And I gave the knife to him.
(Bobaljik and Wurmbrand 2002:(20b))

122

Topical marking of nonsubjects

The context makes it clear that the topic role is associated with the direct object
in (29a) and with the dative oblique in (29b).
Example (25b) above, where agreement is with the possessor/dative argument, additionally shows that the secondary agreement controller is topical.
The discourse fragment consists of two clauses, and the character called Qusklnaqu is the topic of the first clause. The second clause states that Qusklnaqus
head is splitting. Since the head is inalienably possessed by Qusklnaqu, the
second clause provides new information about him. In this context, both referents (Qusklnaqu and his head) are under discussion at the time of the utterance.
Note that the presence or absence of overt (pronominal) agreement controllers
does not affect agreement. Thus, it is information structure rather than grammatical role that determines nonsubject third person agreement in Itelmen: the
verb agrees with nonsubject elements that are topical.
The influence of the person hierarchy on agreement in Itelmen is not yet
clear. Bobaljik and Wurmbrand point out that in the related languages Alutor and Chukchi, agreement is required with first or second person agreement
controllers. However, they demonstrate that in Itelmen, this is only a strong
tendency and not a firm requirement; they provide example (30), in which suffixal agreement is controlled by a third person element even in the presence of
a potential second person controller:
(30) kma xej Pnc kn-ank n@nc m-z@l-cen
not
you-Dat fish 1SgSubj-give-1>3SgObj
I
I wont give the fish to you.

(Bobaljik and Wurmbrand 2002:(28))

Pending further research, we leave open the question of the influence of person
on the determination of agreement controller in Itelmen.
6.2.2 Tabassaran
Kibrik and Seleznev (1980) discuss two types of person agreement in Tabassaran (North Caucasian): the first type is subject agreement, and the second
type is with a nonsubject element which is pragmatically prominent. In the
first type of agreement, the verb agrees with a first or second person subject
which can bear various semantic roles and can be casemarked in several ways
(normally Nominative, Dative or Ergative). With a third person subject, the
verb may receive the affix -(u)v. Kibrik and Seleznev analyse this affix as a
default non-agreement marker rather than third person agreement, but we find
their arguments unclear, and for simplicity we will treat this affix as expressing
third person agreement.
Any prominent nonsubject argument can trigger the second type of agreement, including patient and recipient as well as (arguably) some non-terms.

Agreement with topical nonsubjects

123

Nonsubject agreement is realised differently depending on the case of the

agreement controller: for example, agreement with a first person singular argument is marked either with the affix -is or with -za/-zu, depending on whether
the argument stands in the dative or nominative/ergative, respectively. There
are additional morphological constraints on the realisation of agreement features in nonsubject agreement, the details of which are not relevant for present
purposes.
In some circumstances, the verb can agree with a subject as well as a prominent nonsubject argument, combining the first and second types of agreement.
In particular, the verb can take two agreement affixes when the subject bears
the agent role and is in the first person; in such circumstances, the verb obligatorily hosts subject agreement, and optionally hosts nonsubject agreement
with a prominent nonsubject element, as in example (31):
(31) izu ivu
uv
cunu-zu-vu
I.Erg you.Nom beat-1Sg.Nom/Erg-2Sg.Nom/Erg
I am beating you.

(Kibrik and Seleznev 1980:20)

In all other cases only one agreement affix is present, either subject agreement
or nonsubject agreement. The examples in (32) have third person subjects, and
therefore only one agreement affix is allowed: either the third person agreement suffix -(u)v, or a suffix encoding agreement with a pragmatically prominent nonsubject element.
(32) a. duRu izu
uv
cun-uv / uv
cunu-za
he.Erg I.Nom beat-3
beat-1Sg.Nom
He has beaten me.

(Kibrik and Seleznev 1980:23)

b. duRu izus atnar

uRn-uv / uRn-is
she.Erg I.Dat socks.Nom knit-3 knit-1Sg.Dat
She knitted socks for me.

(Kibrik and Seleznev 1980:23)

c. duRu iziPin alarxun-uv / alarxunu-ziPin

he.Erg I.Supess attack-3
attack-1Sg.Supess
He attacked me.

(Kibrik and Seleznev 1980:23)

hitikin-uv / hitikinu-zuqh
d. duRu izuqh
hide-1Sg.Postess
he.Erg I.Postess hide-3
He hid behind me.

(Kibrik and Seleznev 1980:23)

In (32a) nonsubject agreement is optionally triggered by the nominative patient

argument, in (32b) by the dative recipient, in (32c) by an argument in the
superessive (Supess) case, and in (32d) by the postessive (Postess) NP.

124

Topical marking of nonsubjects

It could be argued that all of the examples in (32) involve predicate-argument

agreement and therefore do not violate the usual assumptions about agreement
domains. However, nonsubject agreement is also possible when the controller
corresponds to a possessor, as shown in (33).
(33) a. jas agaji
dumu uv
cun-uv / uv
cun-as
I.Gen father.Erg he.Nom beat-3
beat-1Sg.Gen
My father has beaten him.

(Kibrik and Seleznev 1980:23)

b. duRu jas agaji-s

kaz
ikv-uv / ikn-as
he.Erg I-Gen father-Dat letter.Nom write-3 write-1Sg.Gen
He wrote a letter to my father.

(Kibrik and Seleznev 1980:23)

hitikn-uv / hitikn-as
c. baj jas c h huka-qh
hide-1Sg.Gen
boy I.Gen shed-Postess hide-3
The boy hid behind my shed.

(Kibrik and Seleznev 1980:23)

d. duqari jas jaku-xi hiturd-uv / hiturd-as

they.Erg I.Gen axe-Com cut-3
cut-1Sg.Gen
They are cutting (wood) with my axe.
(Kibrik and Seleznev 1980:24)
In (33a) agreement is controlled by the possessor of the ergative subject, in
(33b) by the possessor of the dative recipient, in (33c) by the possessor of
the postessive NP, and in (33d) by the possessor of the comitative NP. In all
these examples, nonsubject agreement is indicated by the first person singular
genitive agreement affix -as.
Unfortunately, the precise conditions on nonsubject agreement are not clearly
defined in the sources available to us. Kibrik and Seleznev (1980) state only
that the agreement controller in nonsubject agreement is more prominent or
emphatic than non-agreeing elements. For instance, if the verb in (3233)
shows agreement with the first person singular element, its pragmatic prominence is said to be assessed more highly than in corresponding sentences with
(default) third person agreement. Based on their discussion, we cannot claim
with certainty that agreement indicates topicality. Pragmatic prominence can
be taken to indicate either that the agreeing element is salient in the sense of
being under discussion (and thus topical), or that it is emphasised in the sense
of being selected from a set of alternative candidates (and thus contrastive);
we cannot resolve this issue without additional data, though we believe that it
is cross-linguistically more common for agreement to encode topicality than

Agreement with topical nonsubjects

125

emphasis. For present purposes, the important point is that nonsubject agreement in Tabassaran is not restricted to a particular grammatical function and is
conditioned by information structure, possibly topicality.4
6.2.3 Topical nonsubject agreement
Agreement marking involving nonsubject topics has the following general form,
where GF is any grammatical function, and [GF SUBJ] is any grammatical function other than subject:5
(34) General form of agreement with topical nonsubjects:
(( [GF SUBJ])

DF)

= TOPIC

This is also similar to the constraint for topical subject agreement, given in
(18) of Chapter 5, except for the specification of the agreement controller:
here, agreement is with any grammatical function except subject. We do not
use this very general constraint for Itelmen and Tabassaran, however, since the
agreement affix in these languages varies according to the case or grammatical
function of the topical controller of agreement.
In Tabassaran, the agreement affix varies with the case of the agreement controller. Assuming that agreement does indeed mark topicality in Tabassaran,
specifications for the first person singular dative agreement affix are as given
in (35):
(35) Agreement with first person singular dative topical nonsubjects (Tabassaran):
( [[GF (POSS)] SUBJ]) = %t
(%t PERS) = 1
(%t NUM) = SG
(%t CASE) = DAT
(%t DF) = TOPIC
As with the specification of agreement with topical subjects or objects in Kinyarwanda/Kirundi given in Chapter 5, example (10) (page 99), this specifica4 A similar situation obtains in Maithili (Indo-Aryan) (Stump and Yadav 1988, Comrie 2003,
Bickel et al. 1999), in which the controller of agreement must be prominent in some sense, but can
be a subject, object, possessor, or (in some dialects) the object of a preposition. Dalrymple and
Nikolaeva (2005) suggest, following Comrie (2003), that the relevant factor is topicality, though
further research has indicated that a more general notion of prominence or contrast may better
characterise agreement patterns in Maithili.
5 The expression [ GF SUBJ ] is a regular expression, and the minus operator () is the complementation operator: the expression [GF SUBJ] refers to all strings consisting of any grammatical function, but with the string SUBJ removed.

126

Topical marking of nonsubjects

tion uses a local name, beginning with a percent sign, which is used when
constraints are placed on an f-structure whose grammatical function is uncertain or undetermined. In the first line of these specifications, the local name
%t is defined as any f-structure which bears some grammatical function within
the clause (GF), or a possessor (optional POSS within the GF), but not a subject
(the option SUBJ is removed); this f-structure will be the controller of agreement. The constraints in the second, third and fourth line require the f-structure
named %t to be first person dative. The final line ensures that %t bears the information structure role of topic. Similar specifications are relevant for affixes
marking topicality of arguments with other combinations of person, number,
and case.
Itelmen has two sets of secondary agreement suffixes: one for objects, and
one for obliques. The oblique agreement suffix is always a marker of topicality: intransitive verbs do not agree with nontopical obliques, and transitive
verbs agree with their (topical or nontopical) objects rather than nontopical
obliques. For the third person singular oblique agreement suffix, the following
specifications are relevant:
(36) Third person singular oblique agreement (Itelmen):
( OBL PERS) = 3
( OBL NUM) = SG
(( OBL ) DF) = TOPIC
Other combinations of person and number are treated similarly.
Itelmen object agreement is different: since suffixal agreement is always
required on the Itelmen verb, object agreement is only optionally an indicator
of topicality of the object, though it does indicate that there is no other topical
nonsubject (oblique) element in the clause. In other words, the verb shows
object agreement if the object is the only possible agreement controller, or if
the other potential controllers are nontopical. Constraints associated with the
first person singular object agreement suffix are:
(37) First person singular object agreement (Itelmen):
( OBJ PERS) = 1
( OBJ NUM) = SG
[(( OBL ) DF) = TOPIC]
((( OBJ) DF) = TOPIC)
The first two lines of this specification ensure that the object is first person
singular. The third line states that there may be no oblique phrase which bears

Conclusion

127

the topic role in the clause.6 The fourth line introduces a default specification
of topicality for the object in the presence of object agreement.
6.3 Conclusion
We have shown that topichood of a nonsubject element can be explicitly indicated by casemarking (as in Persian or Tariana) or agreement (as in Itelmen
and possibly Tabassaran) for objects, obliques, possessors, and other nonsubject grammatical functions. The languages we have examined do not require
a unique alignment between information structure role and grammatical function, but provide primary evidence for the relevance of topicality in grammatical marking of nonsubjects.

6 The information structure role of other arguments in the clause, including obliques, may be
specified by casemarking, agreement, prosody, phrase structure position, or discourse context, as
discussed in Chapter 4, Section 4.3.3.

Topicality and DOM

We have seen that some languages treat nonsubject topics specially in terms
of grammatical marking: in such languages, topic marking can apply to a variety of nonsubject elements, and any one of a number of grammatical roles
can be the target of topic marking. In this chapter we discuss languages in
which only topical objects are marked, giving rise to DOM. In some of these
languages DOM depends on topicality alone, while in others topicality-based
DOM works together with semantic factors.
7.1 Objects as grammaticalised secondary topics
Chapter 6 discussed languages in which a whole range of salient nonsubject
elements in the clause can bear topical marking: direct objects, some oblique
objects and adjuncts, and sometimes even possessors. Crucially, objects are
always candidates for marking of this type: we do not know of any language in
which nonsubject topic marking is unavailable for objects, and in fact in many
languages grammatical marking of topical nonsubject arguments is restricted
to objects. This is DOM.
Languages with DOM overtly mark a close association between objects and
topics, just as some languages require and mark a close association between
subjects and topics. In general, objects or, more generally, arguments high
on the grammatical function hierarchy tend to be more topical than other
nonsubject elements. Croft (1991) observes that high topicality is typical of
subjects, while objects are characterised by medium topicality, in contrast to
other grammatical functions. Rude (1986) and Blake (2001:135) also claim
that objects tend to be more topical then obliques and adjuncts, which are
located low on the hierarchy.

129

130

Topicality and DOM

Indeed, as follows from the definition of secondary topic given in Chapter 3,

an utterance with a primary and secondary topic conveys a relation that holds
between two salient participants. Since the primary topic is closely associated
with the subject function (see Chapter 5, Section 5.3), and languages tend to
lexicalise important relations between two participants as transitive verbs, we
would expect that the secondary topic would often be realised as the second
argument of a transitive verb: the object. This means that the direct object
is the result of grammaticalisation of the secondary topic in the same way as
the subject is usually thought to be the result of the grammaticalisation of the
primary topic. In fact, Givon explicitly addresses the grammatical association
of the secondary topic function with objecthood and suggests that in historical
terms, objects are grammaticalised secondary topics (Givon 1983b, 1984a,b,
1990, 2001).
Sasse (1984) discusses the inherent connection between information structure roles (or, in his terminology, pragmatic functions) and syntax. He argues
that subject and object are pragmatically more prominent than other grammatical functions: the canonical pragmatic function of the direct object as a
secondary grammatical function associated with the patient role is to identify
the pragmatic peak corresponding to lower-order (secondary) topicality. The
correlation between objecthood and topicality is so strong that in a number of
languages, nontopical patients cannot be expressed as syntactic objects, and
must undergo incorporation. Sasse shows that this situation is attested in a
number of Eastern Cushitic languages (see Nss 2007 for similar observations).
Conversely, in some languages topical status triggers the promotion of nonobject arguments to the object role, often via applicativisation (see Peterson 2007
and references therein, and the discussion of Upper Necaxa Totonac in Chapter 9, Section 9.4.3). For instance, Rude (1986) shows that in Nez Perce (Sahaptian) a number of nonobject grammatical functions can undergo promotion
to object. This is characteristic of allative, associative and ablative obliques.
In example (1a), river is an oblique argument with allative case, while example (1b) illustrates an alternative encoding of this argument as a direct object
marked with the object marker -ne; here the verb hosts the applicative affix
(allative voice), signalling transitivisation.

Agreement with topical objects: Tundra Nenets

131

(1) a. kaa hi-t`eemik=se

pikuun-x
and 3Nom-go.down-Asp river-All
And she went down to the river.
b. kaa kuus-ne pee-xyuu-ye
and water-Obj 3.Tr-go-All-Asp
And he went to the water.

(Rude 1986:139)

Rude argues, based on a textual study, that promoted/applied objects are

more topical than non-promoted obliques. In fact, he claims that the Nez
Perce direct object is a kind of secondary topic (Rude 1986:148), although
his definition of secondary topic is not actually provided. Rudes operational
criteria for topicality are different from the presuppositional approach we employ here: for him, topicality is a gradient discourse-related notion and can be
measured in terms of referential distance and persistence in discourse. Nevertheless, the two approaches lead to roughly similar results when it comes to
the analysis of narrative texts. A recurrent referent repeatedly mentioned in the
previous discourse is likely to be salient for the speaker and the addressee and
therefore to be topical in our sense as well. In the remainder of this chapter, we
will discuss languages where topicality marking is restricted to objects, parallel to the languages where it is restricted to subjects, as discussed in Chapter 5.
7.2 Agreement with topical objects: Tundra Nenets
In Tundra Nenets1 (Uralic), subject agreement is obligatory and references
both person and number features. Object agreement is optional, and references the number but not the person of the accusative object. Thus, intransitive verbs agree with the subject, while transitive verbs either agree with the
subject alone, or with both the subject and the object. The object marker for
the singular object is always phonologically null, and the marker for singular
objects is a portmanteau morpheme referring both to the subject and the object.
In further glosses, object agreement verbs are glossed simply as Obj, without
indicating the object marker specifically.
Agreeing and nonagreeing objects have different information structure roles.
If example (2d) is understood as an answer to (2a), the whole clause constitutes
the focus domain. If it is understood as an answer to (2b), the focus domain
includes the verb and the object, but excludes the (topical) subject. If it is
1 The

Tundra Nenets data were collected by the second author during fieldwork supported by
the Endangered Languages Documentation Programme, SOAS, London, as well as a grant from
the Academy of Finland (project number 125225).

132

Topicality and DOM

understood as an answer to (2c), the object corresponds to narrow focus. In all

of these contexts, object agreement is disallowed.
(2) a. What happened?
b. What did a/the man do?
c. What did a/the man kill?
/ *xadaoda
d. xasawa ti-m
xadao
man reindeer-Acc kill.3SgSubj kill.Obj.3SgSubj
A/the man killed a/the reindeer.
Similarly, in (3a) the question word xbya-m whom is focused, and object
agreement is impossible. Example (3b) is understood as an answer to (3a), with
the object Peter in narrow focus: again, object agreement is ungrammatical.
/ *lad@oda
(3) a. Wanya xbya-m lad@o
John who-Acc hit.3SgSubj hit.Obj.3SgSubj
Whom did John hit?
/ *lad@oda
b. Wanya Pyetya-m lad@o
John Peter-Acc hit.3SgSubj hit.Obj.3SgSubj
John hit Peter.
In contrast, agreement must be present when the previous context establishes
a topical role for the object. Interpreted as answers to the question in (4a),
the object of kill in (4b) and (4c) is the secondary topic, and the verb must
show object agreement. Thus, Nenets is similar to the languages discussed
in Chapter 6 in that agreement depends on topichood; it is syntactically more
constrained, however, in that only topical objects control secondary agreement,
and not other nonsubject arguments.
(4) a. What did a/the man do to the/a reindeer?
/ *xadao
b. xasawa ti-m
xadao da
man reindeer-Acc kill.Obj.3SgSubj kill.3SgSubj
A/the man killed a/the reindeer.
c. xadao da
kill.Obj.3SgSubj
He killed it.

Agreement with topical objects: Tundra Nenets

133

As shown in (4c), the topical subject and object need not be overtly expressed.
As noted in Chapter 3, topical arguments are often discourse-old, and tend to
receive reduced expression as a pronoun or null element; in fact, a topical third
person pronoun is normally omitted unless it is contrastively stressed. When
there is no object pronoun, the agreement affix is the only overt expression of
the object.
There is no difference in the behaviour of objects with different semantic
features such as definiteness or animacy with respect to object agreement.
Nonreferential objects do not trigger agreement, but this follows from the general condition that topics must be referential, as noted in Chapter 3. Third
person pronouns behave like lexical nouns, as shown in (5), which is grammatical without object agreement if the object is in focus, for example as an
answer to the question Who did John hit? or What did John do?. If it is construed as an answer to the question What did John do to him?, so that the object
has the secondary topic role, object agreement is obligatory.
(5) Wanya syita lad@o
/ lad@o da
John he.Acc hit.3SgSubj hit.Obj.3SgSubj
John hit him.
Note that in some respects, object agreement in Nenets resembles that of Chichewa,

as analysed by Bresnan and Mchombo (1987) (see Chapter 2, Section 2.6). As

in Nenets, Chichewa
object agreement is optional, and correlates with the topicality of the object. However, a closer look reveals that Nenets object agreement is grammatical agreement, whereas Chichewa
involves what Bresnan
and Mchombo (1987) call anaphoric agreement, or pronominal incorporation.
Bresnan and Mchombo (1987) point out that in languages with incorporated
pronominal objects, the verb cannot govern the case of the full noun phrases
that are anaphorically linked to the incorporated pronominals, since these full
noun phrases are not arguments of the verb. This, then, allows us to distinguish
between anaphoric and grammatical agreement. Crucially, objects in Nenets
must appear in accusative case, even if object agreement is present on the verb,
as shown in (5). This shows that object noun phrases in Nenets are governed
by the verb, and that the object agreement affixes represent grammatical agreement and not pronominal incorporation.
In (6), narrow focus is associated with the locative phrase, the topic referent
is John, and the secondary topic referent is Peter; again, object agreement is
obligatory:

134

Topicality and DOM

(6) a. Where did John hit Peter?

/ *lad@o
b. Wanya Pyetya-m pedara-xona lad@o da
John Peter-Acc forest-Loc hit.Obj.3SgSubj hit.3SgSubj
John hit Peter in the forest.
Nenets allows nontopical (focus) subjects in transitive clauses; they are not
morphosyntactically marked, but must bear nuclear stress. If the object is topical, the verb shows object agreement:
(7) a. xbya ti-m
xadao
/ xadao da
who reindeer-Acc kill.3SgSubj kill.Obj.3SgSubj
Who killed a/the reindeer?
b. Wanya ti-m
xadao
/ xadao da
John reindeer-Acc kill.3SgSubj kill.Obj.3SgSubj
JOHN killed a/the reindeer.
There is no subject topic in these examples. This indicates that object agreement is not associated only with the secondary topic function. In (7b), the
agreeing object is the primary topic and is likely to have a definite interpretation. In other words, although in many cases the agreeing object bears the
secondary topic role, the function of object agreement is broader: it can simply
indicate the topicality of the object.
However, there are certain semantic restrictions on agreeing objects in Tundra Nenets. Objects with indefinite determiners never trigger agreement:
/ *pyuroNada
(8) xurkax@wo pyryibtya-m pyuroNa
some
girl-Acc
look.3SgSubj look.Obj.3SgSubj
He is looking for some girl.
Additionally, first and second person singular and plural object pronouns in
Nenets behave differently from all other objects in that they do not trigger
agreement, no matter what their information structure role. Example (9) is ungrammatical with object agreement, even construed as an answer to the question What did John do to you/me?, a context in which the object is associated
with the secondary topic function.
/ *lad@o da
(9) Wanya syiqmo/syito lad@o
John I.Acc/you.Acc hit.3SgSubj hit.Obj.3SgSubj
John hit me/you.

Agreement with topical objects: Tundra Nenets

135

For first and second person pronominal objects, then, patterns of agreement do
not depend on information structure. Instead, they are defined in terms of the
referential status of these objects: first and second person objects never trigger
agreement. We return to this point in Chapter 10.2
We now turn to the question of the grammatical function of agreeing vs. nonagreeing objects, as determined by their syntactic behaviour: we find no behavioural differences that distinguish them, and we analyse both as the primary
object, LFGs OBJ. Objects in Nenets have a number of syntactic properties that
distinguish them from other grammatical functions. They can be promoted to
subject in the passive. Subjects and objects are the only two grammatical functions that can be relativised using the participial strategy; all other grammatical
functions must be relativised by means of another verbal form, the action nominal. These tests, however, will not help us in establishing syntactic differences
between marked and unmarked objects, since there is no object agreement in
the resulting construction.
Another object property relates to control structures. Nenets has a number of
complement-taking verbs which take a dependent null-subject clause headed
by a so-called converb (either the modal converb or the purposive converb).
The dependent subject must be interpreted as coreferential with the matrix object: this is object control. Both agreeing and nonagreeing objects can control
the dependent subject.
(10) a. nysya-da nyu-m-ta
xanyeo
father-3Sg son-Acc-3Sg hunt.Mod.Conv
toxolao /toxolaoda
teach.3SgSubj/teach.Obj.3SgSubj
The father taught his son to hunt.
nya-m-ta
b. x-w@ncyo
leave-Purp.Conv friend-Acc-3Sg
x@lkadotampyi/x@lkadotampyida
persuade.3SgSubj/persuade.Obj.3SgSubj
He is persuading his friend to leave.
Additionally, both agreeing and non-agreeing objects can serve as the antecedent of a possessive reflexive, provided the antecedent linearly precedes
the reflexive.
(11) Pyetya Masha-m pida mya-kona-nta lad@o
/ lad@o da
Peter Mary-Acc she yurt-Loc-3Sg hit.3SgSubj hit.Obj.3SgSubj
Peter hit Maryi in heri yurt.
2 In some varieties of Nenets, third person pronouns behave like first and second person pronouns, and do not trigger agreement.

136

Topicality and DOM

Moreover, Nenets agreeing and non-agreeing objects do not show any obvious positional difference. Nenets is a fairly strictly subject-initial verb-final
language, but word order is otherwise relatively free. Example (12a) demonstrates that both types of objects can appear immediately before the verb, while
in (12b) both types of objects are separated from the verb by the oblique element in the forest.
(12) a. nysya-da pedara-xona wenyako-m lad@o
/ lad@o da
father-3Sg forest-Loc dog-Acc hit.3SgSubj hit.Obj.3SgSubj
His father hit a/the dog in the forest.
/ lad@o da
b. nysya-da wenyako-m pedara-xona lad@o
father-3Sg dog-Acc forest-Loc hit.3SgSubj hit.Obj.3SgSubj
His father hit a/the dog in the forest.
We therefore suggest that marked and unmarked objects realise the same grammatical function: the object. Nenets is a language with only one object function, the OBJ function of LFG (see Chapter 2, Section 2.2). As we might expect, then, Nenets has no double object constructions; the goal argument of
verbs such as give is invariably expressed by a dative-marked oblique:3
(13) Petya Masha-noh ti-m
myiqNa
/ myiqNada
Peter Masha-Dat reindeer-Acc give.3SgSubj give.Obj.3SgSubj
Peter gave Masha a/the reindeer.
Dative obliques never trigger agreement and do not have other syntactic properties of objects: they do not passivise, do not participate in control constructions and are relativised by means of a different strategy.
In formal terms, third person topical object agreement in Nenets is associated with the following constraint:
(14) Agreement with third person topical objects:
( OBJ PERS) = 3
(( OBJ) DF) = TOPIC
This is similar to the constraint for topical subject agreement, given in (18) of
Chapter 5, except that the constraint requires the object rather than the subject
to be topical. As a result of this specification, the semantic structure contributed by the object is associated with the information structure role of topic.
3 In Chapter 8, we will see that there are languages that have more than one object function,
the primary OBJ and the secondary OBJ of LFG, but nevertheless do not have a double object
construction; Ostyak, an Uralic language related to Nenets, is one such language. We discuss these
patterns further in Chapter 9.

Casemarking of topical objects

137

Verbs which agree only with the subject are not associated with this constraint,
since they do not require their object to be topical (recall that first and second
person objects do not control agreement, and may or may not be topical).
7.3 Casemarking of topical objects
In Tigre (Semitic) and Dolakha Newar (Tibeto-Burman), casemarked and noncasemarked objects have different information structure roles: marked objects
are topical, while unmarked objects are nontopical. We know of no behavioural
differences between marked and unmarked objects, and we analyse both as primary objects, LFGs OBJ. Unlike Nenets, both languages have a double object
construction, which we discuss in Chapter 9; here we restrict attention to the
behaviour of monotransitive objects, marked and unmarked, showing that they
bear different information structure roles, but correspond to the same grammatical function.
7.3.1 Tigre
The primary description of the Tigre (Semitic) data comes from Jake (1980).
Tigre objects can be marked by the preposition P1g1l, and can trigger object
agreement in gender. Agreement is determined by definiteness: definite objects always trigger agreement, while indefinites do not. This distribution is not
affected by information structure. Here we are primarily interested in prepositional marking of objects, which is optional on definite objects of monotransitive verbs, but incompatible with indefinite objects (animacy plays no role).
For monotransitive verbs, this results in the following possibilities for object
marking: (i) casemarked definite objects that trigger agreement; (ii) noncasemarked definite objects that trigger agreement, and (iii) noncasemarked indefinite objects that do not trigger agreement.4 In Chapter 9, we return to
a discussion of casemarking and agreement in double object constructions in
Tigre.
The objects in examples (15) and (16) are definite, and must agree with the
verb. The preposition is allowed but not required for the definite object in
(15), and disallowed with the indefinite object in (16). We are interested in
the optionality demonstrated in (15), and the conditions under which definite
objects must be preceded by the preposition P1g1l.
4 Jake (1980) notes that the causee argument displays the same distribution of prepositional
marking as patient/theme objects: indefinite causees are unmarked, while definite causees are
either marked or unmarked. However she argues that the causee differs from the patient/theme
objects in a number of other syntactic properties, from which she concludes that it corresponds to
a different grammatical function. Since the data on causatives are complex, we leave this question
open.

138

Topicality and DOM

(15) has
am
a
P1tt
a (P1g1l) la h1san nadPayu
/ *nadP
a
Hasama.Masc to.her Prep the boy sent.3Masc.3Masc sent.3Masc
Hasama sent the boy to her.

(Jake 1980:72)

(16) Lilat
(*P1g1l) waraqat katbat(*t
a)
Lilet.Fem Prep letter.Fem wrote.3Fem(*3Fem)
Lilet wrote a letter.

(Jake 1980:73)

Jake (1980) is mainly concerned with an explication of conditions on verb

agreement, and she does not provide a detailed characterisation of when the
preposition P1g1l is used on definite objects. However, an analysis of the narrative texts published in Raz (1983) reveals that such objects must be topical.
A definite object introduced into the discourse for the first time remains unmarked, even if it is highly identifiable; we suggest that this is because there
is no pragmatically presupposed relationship between its referent and another
salient referent, and so the object referent is not a secondary topic. This is
shown in examples (17) and (18).5
(17) a. wa k@@nna @t l@waddu m@d@r kanaan warsaw
and like.this while doing land Canaan inherited.3Pl
And while living like this, they inherited the land of Canaan.
(Raz 1983:111)
b. gis wagabilye
m@n @de faron afg@r
go and.my.people from hand Pharaoh bring.out
Go and free my people from the hands of Pharaoh.(Raz 1983:109)
The objects in these examples are highly definite: in (17a) the object is a proper
name (Canaan) and in (17b) it is a possessed definite phrase (my people).
However, there is no presupposed pragmatic relationship between the subject
and the object referent established prior to the time of the utterance. Example
(17a) is the last sentence of a text telling the story of Moses and the people
of Israel, but there is no previous mention of the land of Canaan in the text.
Example (17b) is direct speech: God is addressing Moses. Again there is no
established pragmatic relation between the subject (Moses) and the referent of
my people, since it is the first time God addresses Moses with this request
and it is rather unexpected for Moses.
5 We have followed Razs transcription conventions for the examples taken from his work,
which differ from the conventions used by Jake. Raz transcribes the object-marking preposition
as @g@l.

Casemarking of topical objects

139

On the other hand, discourse-old definite objects are likely to be topical, and
are often marked by the preposition P1g1l. The beginning of the same story
describes how the mother of Moses made a chest, put her baby in the chest,
and put the chest in the Nile. The chest was then found by Pharaohs daughter.
At that stage of the story the NP denoting the referent chest is definite and
identifiable by the interlocutors, since it has been mentioned in the previous
discourse. Example (18) describes the first time the Pharaohs daughter sees
the chest; there is no presupposed pragmatic relationship between her and the
chest, and the object remains unmarked.
(18) wa @ttu m@n rayim lasanduqat salsala @tta ma-y k@rit
and there from afar the.chest reed in.the water placed
r@etta
saw.3Fem.it
And there she saw, from afar, the reed chest placed in the water.
(Raz 1983:108)
Pharaohs daughter then sends her maidservants to retrieve the chest. When
sentence (19) is produced, the interlocutors have a mental representation of a
certain relation that holds between the maidservants and the chest, since the
narrator has already made it clear that the servants were sent for the chest. The
sentence provides new information about the relationship that holds between
the maidservants (primary topic) and the chest (secondary topic), and can be
paraphrased as follows: what the maidservants then did to the chest is: they
opened it. The object NP the chest is prepositionally marked.
(19) wa lawasayfa
@g@l lasanduqat kf@t-k@msal-abalaya g@na
and her.maid.servants Prep the.chest when-opened.3Pl.it a.child
bakke rakbaya
crying found.3Pl
And when her maidservants opened the chest, they found [in it] a child
crying.
(Raz 1983:108)
Note that there is no prepositional marking on the indefinite object child in
the second clause.
Similarly, example (20), taken from a tale about two friends, a cat and a dog,
updates the nature of the relationship that holds between two highly salient
participants (the cat deceived the dog).
talmat
@ttu
(20) daam d@mmu ... @g@l kal@b w@ul
but
cat
Prep dog deliberately she.deceived him
But the cat ... deliberately deceived the dog.

(Raz 1983:104)

140

Topicality and DOM

Here the subject (cat) and the object (dog) are construed as the primary and
the secondary topic, respectively. The secondary topic object exhibits prepositional topic marking.
Thus, casemarked objects in Tigre are always topical. The examples we have
seen indicate that the casemarked object is the secondary topic, but we have no
examples where the subject is in focus, so we do not know if the casemarker
appears on primary topic objects as well. Only a subset of topics is marked,
namely, definite topics; indefinite topical objects are not marked. Tigre object
marking involves, then, a combination or semantic and information structural
factors.
The next question is whether casemarked and noncasemarked objects correspond to the same grammatical function. Jake (1980) shows that casemarked
and noncasemarked definite objects of monotransitive verbs behave identically
in obligatorily triggering agreement, and she does not discuss any other behavioural differences between the two types of monotransitive objects. In
the absence of any evidence of behavioural differences, we believe that casemarked and noncasemarked objects of monotransitive verbs correspond to the
same grammatical function: the (primary) object, LFGs OBJ.
The formal treatment of topical object casemarking for monotransitives is
straightforward:
(21) Casemarking of topical objects, Tigre:
(OBJ )
( DEF ) = +
(

DF )

= TOPIC

These constraints require the marked argument to be an object, to be definite,

and to play the role of topic at information structure. In fact, the treatment of
casemarking is more complicated than this when we take into account patterns
of casemarking with ditransitive verbs; we return to a discussion of ditransitive
marking in Tigre in Chapter 9.
Kifle (2007) discusses the closely related language Tigrinya, and shows that,
although patterns of casemarking and agreement in Tigrinya are similar in very
broad terms to Tigre, the two languages differ in interesting ways. In particular, it is casemarking in Tigrinya that depends on definiteness, while agreement
patterns are determined by information structure role: in this way, Tigrinya is,
in a sense, opposite to Tigre with respect to the roles of agreement and casemarking in signalling semantic information and information structure role. As
in Nenets, agreement depends on topicality for transitive objects. Additionally,
Tigrinya has an applicative construction in which the applied argument must
be topical and verb agreement with the applied object is required; this follows
the general tendency for applicativisation to be triggered by topical status of

Casemarking of topical objects

141

the applied argument, as discussed for Nez Perce in Section 7.1 of this chapter. Interestingly, the applied argument cannot become the subject of a passive
sentence, although either the recipient or the theme of an underived ditransitive verb can passivise; Kifle (2007) analyses this as a problem for the alignment between information structure roles and grammatical functions which we
propose, but we believe that it actually indicates a need for refinement of the
general theory of applicatives and voice alternations, and in particular Bresnan
and Moshis (1990) formal theory of object asymmetries; the Tigrinya data are
otherwise unproblematic for our view.
7.3.2 Dolakha Newar
Object casemarking in Dolakha Newar (Tibeto-Burman), as described by Genetti
(1994, 1997, 2007), shows essentially the same properties as Tigre, except that
definiteness does not play a role. Objects are either unmarked (hence in absolutive case) or marked with the suffix -ta, termed dative by Genetti. The
suffix will be glossed here as Obj, although it should be noted that the same
marking occurs on experiencer subjects. We return in Chapter 9, Section 9.4.1
to patterns of casemarking for with ditransitive verbs.
Objects of monotransitive verbs are either marked or unmarked. According
to Genetti (2007:113), the object is casemarked if (i) the referent is human
and given in the discourse, or (ii) the referent is nonhuman but animate, and
occurs in a clause crucial to the resolution of a narrative plot. This means
that marked objects denoting human referents tend to be discourse-old. In
example (22), the child was mentioned in the immediately preceding sentence
in the discourse (Then they had one small son at that time.). The utterance
makes an assertion about the relationship that holds between two highly topical
referents under discussion, the parent and the child.
(22) am muc
a-ta b
abu-ri-n
muc
a ju-e-l
agin
mury
a-ku
that child-Obj father-Ind-Erg child be-NMLZ-because lap-Loc
ta-ene
put-Part
Because he was a child, the father put the child on his lap.
(Genetti 2007:115)
Notice, however, that discourse givenness is not actually a necessary condition
for object marking. Genetti provides evidence that in some cases the referent
of a marked object is not mentioned in the previous text, but is accessible
through the invocation of a schema. Example (23) is taken from a text about
a crown prince; in Dolakha Newar culture it is generally assumed that crown
princes have wives, so the referent of the phrase your legal wife is accessible

142

Topicality and DOM

and its existence is presupposed. Recall from Chapter 3, Section 3.2.2 that
pragmatic presupposition of existence is a necessary property of topics, while
discourse givenness is not.
(23) chana bihaiti-ta-uri
chin
che-ku
ta-ina
2Sg.Gen legal.wife-Obj-Top 2Sg.Erg house-Loc put-Fut
You will put your legal wife in the house.

(Genetti 1997:48)

On the other hand, not all human objects are casemarked. They typically remain unmarked when first introduced into the text, so humanness is not a sufficient condition for object marking.
For nonhumans, too, object marking indicates that the referent is highly
salient, since such objects occur at a climax in the narrative. Example (24)
appears when the manipulation of the animals is crucial to plot resolution
(Genetti 1994:51). The speaker has related the sons plan to release the calf,
and the release of the calf described here is the culmination of this plan.
(24) kae-uri-n
tapakka s
ac
a-ta phen-ju
son-Indef-Erg all.at.once calf-Obj release-3Sg.Past
Then the son suddenly released the calf.

(Genetti 2007:114)

In our terms, this means that the object in (24) is topical.

Genetti also provides several examples of casemarked inanimate objects,
although she notes that they are rare:
(25) a. ota
ult
a
yeN-an keN-gu
ju-en con-a
this.Obj translate do-Part show-NMLZ be-Part stay-3Sg.Past
ka
Associative
It turns out she translates this, then shows it to people.
(Genetti 1994:51)
b. au luNm
a tuphi-ta
h
ati yer-eu?
now mortar broom-Obj what do-3Pl.Fut
Now what will he do with the mortar and the broom?
(Genetti 1994:114)
In example (25a), the object marker has fused with the pronoun. In both examples the object is under discussion, as can be seem by examination of the
previous context. For example, (25b) is taken from a narrative text. The immediately preceding discourse is: One of them quickly brought in a mortar
and a broom. And put them there. What was born? They said: A mortar and

Conclusion

143

broom were born. When they said that the king stayed silent. In this context,
(25b) describes a pragmatically highly salient relationship between the king
and the mortar and broom, where the coordinated object NP is the secondary
topic. These examples show that inanimate objects may be casemarked if they
are topical.
Thus, casemarking of patient/theme objects cannot be unambiguously described in terms of definiteness, discourse-givenness or animacy. Instead, the
presence of the casemarker requires a certain degree of pragmatic salience:
the referents of the object and the subject must stand in a certain presupposed
relationship established in the context or based on world knowledge. The assertion associated with the sentence where the marked object appears is meant
to update the addressees knowledge about this relationship. Notice that object
marking can co-occur with the topic marker, as shown in example (23) as well
as in (26):
(26) bhut-na janta w
a guli
khy
aN-an tar-ai
ghost-Erg 1SgObj Top how.much scare-Part put-3Sg.Pres
Ghosts scare me so much.
(Genetti 2007:300)
It is not clear on Genettis account whether the topic marker in either of these
examples carries an additional meaning (for instance, contrastivity).
Genetti (2007:315) shows that objects in Dolakha Newar can be distinguished from subjects by means of a number of tests: objects do not trigger
agreement, do not serve as antecedents of regular reflexives, and do not participate in control constructions or certain types of relativisation. The tests do
not distinguish between marked and unmarked objects, however, and she concludes that there is no behavioural syntactic difference between marked and
unmarked objects: in our terms, they are both primary objects, LFGs OBJ.
We adopt a formal treatment of object marking in Newar that is similar to the
one proposed above for Tigre, except that definiteness does not play a role in
marking:
(27) Casemarking of topical objects, Newar:
(OBJ )
( DF) = TOPIC
These constraints require the marked argument to be an object and to fill the
topic role at information structure.
7.4 Conclusion
In the languages we have examined in this chapter, marked and unmarked objects of monotransitive verbs correspond to the same grammatical function:

144

Topicality and DOM

both are primary objects, the OBJ function of LFG. We base this claim on the
lack of observed behavioural differences between them. Of course, further
research on these languages may reveal hitherto undiscovered behavioural differences between marked and unmarked objects, which would necessitate their
reclassification along the lines to be discussed in Chapter 8. There, we will see
that other languages exhibit a different pattern: the marked object of monotransitive verbs can be shown to have different grammatical properties from
the unmarked object.

Primary and secondary

objecthood and DOM
In some languages, differential object marking correlates with a difference in
grammatical function: marked and unmarked objects have different syntactic
behaviour, with marked objects exhibiting more properties of core grammatical functions than unmarked objects. Here we discuss languages where marked
(topical) objects are grammatically distinguished from unmarked (nontopical)
objects. We concentrate on marking patterns with monotransitive verbs in
this chapter; Chapter 9 discusses marking and alignment patterns for trivalent
verbs, with particular attention to languages with ditransitive constructions.
8.1 Grammatical marking and grammatical function
There are two basic patterns of interaction between the grammatical marking
of nonsubject topics (DOM) and grammatical objecthood. In languages of the
first type, a difference in object marking does not correlate with a difference
in grammatical function. Such patterns are not surprising in the context of traditional theories of argument mapping, discussed in Chapter 2, Section 2.1.2,
which define possible relations between grammatical functions and semantic
roles as they are represented at argument structure. On this view, we would
expect marked and unmarked objects which correspond to the same semantic
role to be mapped to the same grammatical function, and there is no expectation that information structure role could affect argument mapping.
The languages discussed in Chapter 7 are of this type. In these languages,
grammatically marked and unmarked objects do not display behavioural syntactic differences; grammatical marking correlates with information structure
differences topical vs. nontopical arguments and does not reflect a dif145

146

Primary and secondary objecthood and DOM

ference in grammatical function. In these languages, DOM is defined by information structural considerations, not grammatical function.
In languages of the second type, grammatical marking of the object signals a difference in grammatical function and a concomitant difference in
grammatical behaviour. This complicates the theory of argument mapping,
but makes the relation of f-structure to information structure more clear, since
different information structure roles correspond to different grammatical functions. We demonstrate in the following that in languages of the second type,
when a predicate takes a single nonsubject nominal argument, it can be encoded as either a topical, marked OBJ, or a nontopical, unmarked OBJ . This
gives rise to patterns of DOM which reflect a difference in syntactic status
between marked and unmarked objects.
Languages of this type are Ostyak, Mongolian, Chatino, and Hindi, to be
discussed in this chapter. In these languages, marked and unmarked objects
show different behavioural profiles. Our proposal allows, and indeed requires,
the restatement of generalisations about object agreement and casemarking in
these languages in purely syntactic terms: the agreeing or casemarked object
is the primary object OBJ, and the nonagreeing or noncasemarked object is the
secondary object, a member of the family of grammatical functions OBJ . OBJ
and OBJ require different patterns of casemarking or agreement, and hence
we have DOM. Languages of this type are in essence just like many other
languages in defining agreement and casemarking patterns in purely syntactic
terms, by reference to grammatical functions rather than information structure. Their distinguishing property is the tight linkage between information
structure and f-structure: marked/primary objects (OBJ) are (secondary) topics,
while unmarked secondary or restricted objects (OBJ ) are nontopical. Though
it appears that secondary agreement in Ostyak and object casemarking in Mongolian, Chatino, and Hindi are determined by information structure, as in the
languages examined so far, this is actually a side-effect of the strong relation
between information structure role and grammatical function.
8.2 Object agreement and grammatical function: Ostyak
Northern Ostyak, also called Northern Khanty, is a Uralic language. The data
here reflect the dialect of Obdorsk and come from the fieldwork of the second
author.
As in the related language Nenets, subject agreement in Ostyak is obligatory,
while object agreement is optional (Nikolaeva 1999, 2001). Intransitive verbs
agree with the subject in person and number (1a). Transitive verbs agree either
with the subject alone (1b), or with the subject and the object, as in (1c)-(1e).
Subject and object pronouns may but need not appear. As in Nenets, object

Object agreement and grammatical function: Ostyak

147

agreement forms indicate the number (but not the person) of the object. There
is no verb agreement other than with subject and object.
(1) a. (ma) je:l@n o:m@s-l-@m
I
at.home sit-Pres-1SgSubj
I am sitting at home.
b. (ma) tam kalaN we:l-s-@m
I
this reindeer kill-Past-1SgSubj
I killed this reindeer.
c. (ma) tam kalaN we:l-s--e:m
I
this reindeer kill-Past-SgObj-1SgSubj
I killed this reindeer.
d. (ma) tam kalaN-@t we:l-s@-l-am
I
these reindeer-Pl kill-Past-PlObj-1SgSubj
I killed these reindeer.
e. (ma) tam kalaN-N@n we:l-s@-Nil-am
I
these reindeer-Du kill-Past-Du.Obj-1SgSubj
I killed these (two) reindeer.
Subject markers differ in these forms. For example, the first person singular
subject marker in the absence of object agreement is -@m, as in (1a) and (1b);
with the singular object it is -e:m (1c), and with the dual or plural object it is am, as in (1d) and (1e). Similar distinctions obtain for the whole paradigm. As
in Nenets, the subject marker with singular objects, such as -e:m, is a portmanteau morpheme referring both to the subject and the object, and we gloss object
agreement verbs as Obj, without indicating the object marker specifically.
As shown by Nikolaeva (1999, 2001), traditional descriptions of Ostyak
(Redei 1965, Honti 1984, and others), which suggest that agreement is conditioned by definiteness of the object, are incorrect: both definite and indefinite objects may but need not trigger agreement. Instead, agreement correlates
with information structure role. Objects that trigger agreement are topical,
while nonagreeing objects are nontopical and share a cluster of semantic and
pragmatic properties that are associated with narrow or wide focus.
We first examine nonagreeing objects. In object questions and answers (narrow focus structures), agreement must be absent, even if the object is definite:

148

Primary and secondary objecthood and DOM

(2) a. u:r-na
mati kalaN we:l-@s
/
forest-Loc which reindeer kill-Past.3SgSubj
*we:l-s-@lli
kill-Past-Obj.3SgSubj
Which reindeer did he kill in the forest?
b. u:r-na
tam kalaN we:l-@s
/ *we:l-s-@lli
forest-Loc this reindeer kill-Past.3SgSubj kill-Past-Obj.3SgSubj
He killed this reindeer in the forest.
Example (2b) is also an appropriate answer to the question What did he do?
(with different prosody). In this case, the object is not the narrow focus, but is
a part of the wide focus domain. Both types of objects are nontopical, and the
verb agrees with neither.
Objects under the scope of focus items such as only or even and contrastively
focused objects, including first and second person pronouns, also fail to trigger
agreement:
(3) a. tamxatl tup wul a:n wa:n-s-@m
/ *wa:n-s-e:m
today only big cup see-Past-1SgSubj see-Past-Obj.1SgSubj
Today I only saw the/a big cup.
b. tamxatl ma-ne:m jir-@s
/
today I-Acc
tie.down-Past.3SgSubj
*jir-s-@lli
anta naN-e:n
tie.down-Past-Obj.3SgSubj not you-Acc
Today he tied me down, not you.
It is well known that in many SOV languages there is an immediately preverbal position which is reserved for a focused element, independent of its
Kiss 1995, Butt and
semantic role and grammatical function (Kim 1988, E.
King 1996). Ostyak is such a language, with a grammaticalised focus position
immediately before the verb. The questioned NP which reindeer in (4) must
appear in this position.
(4) a. *mati kalaN u:r-na
we:l-@s
which reindeer forest-Loc kill-Past.3SgSubj
Which reindeer did he kill in the forest?
The same applies to nonagreeing objects that do not correspond to a whquestion word: they are normally impossible to separate from the verb.

Object agreement and grammatical function: Ostyak

149

(5) *tam kalaN u:r-na

we:l-@s
this reindeer forest-Loc kill-Past.3SgSubj
He killed this reindeer in the forest.
This additionally argues for the focus status of at least some nonagreeing objects, since they are required to appear in focus position.
However, not all nonagreeing objects are in focus, and not all appear in verbadjacent position. Nonreferential objects never trigger agreement, even if they
are not in the focus position and arguably do not bear the focus role. In (5), the
focus function is associated with an oblique element (xalsa where), which
must be immediately preverbal:
(6) li-ti
pil
xalsa kas-l-@m
/ *kas-l-e:m
eat-Part companion where find-Pres-1SgSubj find-Pres-Obj.1SgSubj
Where shall I find a companion to eat with?
Examples such as (6) illustrate the only circumstance in which non-agreeing
objects are separated from the verb by another constituent. Such examples
are important because they show that agreement does not directly depend on
the position of the object. Both agreeing and nonagreeing objects can be located in a position which is not adjacent to the verb, but only nonreferential
nonagreeing objects can appear there.
Given the assumption that topics must be referential, the proper generalisation about Ostyak nonagreeing objects is, then, that they may not be topical.
When they appear in the preverbal focus position, they are in focus. When
they do not appear in the focus position, they are nonreferential and therefore
nontopical. In this case they can be separated from the verb by a focus element.
On the other hand, the position of objects that trigger agreement is fairly
free. The object can be separated from the verb by other elements or, in some
cases, even appear after it. For referential objects that are not immediately
preverbal, object agreement is required:
(7) Pe:tra mo:jp@r u:r-na
wa:n-s-@lli
/ *wa:nt-@s
Peter bear
forest-Loc see-Past-Obj.3SgSubj see-Past.3SgSubj
Peter saw the/a bear in the forest.
Some referential objects appear immediately before the verb, but this only occurs if there is no nonverbal focus constituent, so that either the focus falls on
the verb itself or the verb is included in the focus domain. For example, the
following sentence can be interpreted in two ways: either as an answer to the
question What did you do to this reindeer? (broad predicate focus) or to the
question You didnt kill this reindeer, did you? (polarity/verum focus).

150

Primary and secondary objecthood and DOM

(8) tam kalaN we:l-s-e:m

this reindeer kill-Past-Obj.1SgSubj
I DID kill this reindeer/I KILLED this reindeer.
Nikolaeva (2001) shows that for some Ostyak verbs object agreement is
either highly unlikely or altogether impossible. These are verbs that typically
introduce new (nontopical) participants into the discourse. For example, the
verb taj- have very rarely takes object agreement, as confirmed by textual
counts. Its meaning is such that in most cases its object cannot be interpreted
as topical, cf. the following infelicitous piece of discourse:
(9) a. What do you do to this knife?/What is the relationship between you
and this knife?
b. ??I have it/this knife.
Under certain discourse conditions this verb can take object agreement, but
such cases are rare. One possible context is the so-called verum focus.
(10) a. But you dont have the/a knife!
b. taj-l-e:m!
have-Pres-Obj.1SgSubj
I do (have it/one)!
In such examples the focus extends over the verb alone, while the object is
topical and therefore triggers agreement.
As discussed in Chapter 5, Section 5.2, the primary topic role is often associated with subject. In Ostyak this association is obligatory, at least in transitive
clauses. It follows from this that although agreeing objects are topical, they are
not primary topics, since the primary topic role is always associated with the
subject. Rather, they must be analysed as secondary topics, as confirmed by
the following examples. Clauses with object agreement, as in examples (11b)
and (11c), are felicitous as answers to the question in (11a).
(11) a. What did you do to this reindeer?
b. tam kalaN we:l-s-e:m
/ *we:l-s-@m
this reindeer kill-Past-Obj.1SgSubj kill-Past-1SgSubj
I killed this reindeer.
c. we:l-s-e:m
/ *we:l-s-@m
kill-Past-Obj.1SgSubj kill-Past-1SgSubj
I killed it.

Object agreement and grammatical function: Ostyak

151

This context establishes the secondary topic role of the object in the answer,
and nonagreeing objects are not permitted here. In these examples, the verb is
in focus. Object agreement is also present when the focus falls on a non-object
constituent, to the exclusion of the verb and the object itself.
(12) kalaN xalsa we:l-s-@lli
/ *we:l-@s
reindeer where kill-Past-Obj.3SgSubj kill-Past.3SgSubj
Where did he kill the/a reindeer?
Sentence (12) exemplifies an argument-focus structure in which the focus is
on the oblique constituent where, the object is topical, and object agreement
is obligatory.
As shown in Chapter 3, topical arguments tend to receive reduced expression as a pronoun or null element. Consistent with this claim is the fact that
objects that trigger agreement may correspond to a referential null, as in (11c).
In fact, a corpus count shows that almost half of the clauses with object agreement have no overt object (Nikolaeva 2001). The use of overt NPs as objects
that trigger agreement is a marked option and is usually motivated by the need
to disambiguate between several referents. Moreover, as shown by the text
analysis reported by Nikolaeva (2001), 83% of the objects that trigger agreement have been mentioned in the previous discourse. Only 17% have not been
mentioned previously, although these may refer to entities that are relevant in
the speech situation. Further arguments for the topical status of the agreeing
object are presented by Nikolaeva (2001).
Nikolaeva (1999) and Dalrymple and Nikolaeva (2005) show that Ostyak exhibits clear behavioural syntactic differences distinguishing the agreeing from
the nonagreeing object. Nikolaeva (1999:346) provides the table in (13), which
indicates that the former exhibits more properties of core grammatical functions than the latter.
(13) Syntactic properties of subjects and objects in Ostyak:
Subject
Verbal agreement
Control of coreference in action nominal
clauses
Possessor topicalisation
Control of possessive reflexivisation
Quantifier float

Agreeing
(topical)
object
+

Nonagreeing
(nontopical)
object

+
+
+
+

+
+
+
+

For example, the second line of the table indicates that both the subject and the
agreeing object can control coreference with the subject of an action nominal

152

Primary and secondary objecthood and DOM

(AN) adverbial clause. Such clauses are introduced by various postpositions,

which specify the meaning of the adverbial. In example (14), the main and the
embedded clause subjects are coreferential:
(14) [a:s-e:mi tutjux se:w@r-m-al sis] luwi /i mosa
father-1Sg wood cut- AN-3Sg when he
something
no:m@lm-@s
remember-Past.3SgSubj
When my father was cutting wood, he remembered something.
In (15), the embedded clause subject is coreferential with the main clause
agreeing object.
(15) [xuli u:n u:l-m-al pata] xuli /i nox an ta:l-s-e:m
fish large be-AN-3Sg because fish
out not carry-Past-Obj.1SgSubj
I didnt take out the fish because the fish/it was large.
On the other hand, nonagreeing objects in the main clause cannot control the
embedded subject. Example (16) has no object agreement in the main clause
and is ungrammatical.
(16) *[xuli u:n u:l-m-al pata] xuli /i nox an ta:l-s-@m
fish large be-AN-3Sg because fish
out not carry-Past-1SgSubj
I didnt take out the fish because the fish/it was large.
Other syntactic tests in Ostyak are discussed in detail by Nikolaeva (1999)
and are not repeated here. All of these tests lead to the conclusion that the
agreeing object in Ostyak shows more properties of core arguments and has
more properties in common with the subject, while the non-agreeing object
does not exhibit these properties.
Most previous work on DOM has not provided an account of the correlation between DOM and behavioural syntactic properties distinguishing primary from nonprimary objects. Some transformational analyses establish a
correlation between casemarking, agreement, and position of the two types of
objects, but do not straightforwardly extend to an account of the other syntactic differences summarised in Table (13) for Ostyak. Instead, we suggest
that these generalisations should be expressed by assigning agreeing and nonagreeing objects to different grammatical functions. A similar proposal was
made by Butt and King (1996), who propose that unmarked, focused objects
that are adjacent to the verb correspond to a different grammatical function
from objects that may be marked, and whose position is unconstrained: they
analyse the unmarked, verb-adjacent object as the primary object, the OBJ

Object agreement and grammatical function: Ostyak

153

in LFG, and the positionally unconstrained object as the restricted or secondary object OBJ . We suggest the opposite distribution: that the grammatically marked, topical object corresponds to OBJ, and the nontopical, unmarked
object corresponds to OBJ , as shown in (17).
(17) Marked and unmarked patient/theme objects in Ostyak:
OBJ

Marking
Information structure role
Properties of core grammatical functions

yes
topic
yes

OBJ

no
nontopic
no

We believe that our proposal provides a cross-linguistically more satisfactory

account, since it accounts for the fact that it is the grammatically marked, topical object rather than the unmarked, nontopical object that displays more properties characteristic of core grammatical functions, as shown for Ostyak in the
table in (13).
Our proposal stands in agreement with Butt and Kings (1996) view that the
distinction between primary and secondary objects is relevant not only for ditransitive constructions, but also for monotransitives: monotransitive verbs in
some languages can take a subject and either a primary or a secondary object,
and a secondary object may appear even when there is no primary object in the
clause. Indeed, although Ostyak distinguishes the primary/topical OBJ from the
secondary/nontopical OBJ , it does not have a double object construction: no
verb takes an OBJ and an OBJ at the same time. With verbs like give, either
the goal or the theme must appear as an oblique, resulting in two possibilities:
if the theme is an object, the goal is a dative oblique, as in (18a); and if the goal
is an object, the theme is a locative oblique, as in (18b). We return to an indepth discussion of triadic verbs and cross-linguistically available ditransitive
alignment possibilities in Chapter 9.
(18) a. ma a:n Pe:tra e:lti ma-s-e:m
/ ma-s-@m
I cup Peter to give-Past-Obj.1SgSubj give-Past-1SgSubj
I gave a/the cup to Peter.
b. ma Pe:tra a:n-na ma-s-e:m
/ *ma-s-@m
I Peter cup-Loc give-Past-Obj.1SgSubj give-Past-1SgSubj
I gave Peter a/the cup.
Our account meshes well with standard LFG assumptions about permissible relations between semantic roles and grammatical functions, and makes
the correct predictions for patterns of argument marking with objects bearing
a range of semantic roles. As discussed in Chapter 2, Section 2.2, the primary OBJ of LFG is a semantically unrestricted object, which can bear any

154

Primary and secondary objecthood and DOM

of a number of semantic roles. In contrast, Bresnan and Kanerva (1989) define OBJ as a family of semantically restricted objects OBJ THEME , OBJ GOAL ,
and so on, each of which can be filled only by arguments with particular semantic roles. Crosslinguistically, it is very common for languages to allow
only patient/theme arguments as OBJ : we will refer to this grammatical function by the general term OBJ PATIENT / THEME. In a language whose inventory of
secondary object functions is restricted in this way, the correlation between information structure role and grammatical function may be imperfect, holding
only for patient/theme objects and not for objects bearing other semantic roles.
This is the situation in Ostyak. A patient or theme argument can correspond
either to a primary OBJ or a secondary OBJ PATIENT / THEME, with the choice depending on information structure role. Object arguments with other semantic
roles cannot be secondary objects, since Ostyak has only one restricted object
function, and does not allow secondary objects associated with other semantic
roles such as goal or causee. Instead, goal and causee objects must surface
as primary objects and must trigger agreement, whether or not they are topical; the constraint limiting OBJ to arguments bearing the patient/theme role
may not be violated, independent of the information-structure role of the argument. This is further evidence that agreement in Ostyak is defined purely in
terms of grammatical functions, and not information structure roles: although
mapping rules in Ostyak must be defined by reference to information structure
roles as well as semantic roles and grammatical functions, agreement can be
straightforwardly characterised in terms of the grammatical function OBJ.
This is shown by agreement patterns with verbs such as give, which have
a patient and a goal argument: they allow the goal to be an agreeing OBJ, but
not a nonagreeing OBJ , as shown in (18b). When the goal is an oblique, as in
(18a), the patient can be either an agreeing OBJ or a nonagreeing OBJ PATIENT.
The same is observed in causatives. Ostyak causatives are productively derived only from intransitive verbs. The causee argument is casemarked in the
same way as other objects: it stands in the unmarked nominative case if it is
a lexical noun, or in the accusative if it is a personal pronoun. However, the
causative construction differs from the regular transitive construction in that it
requires object agreement even if the causee object bears the focus function.
The sentences in (20) are construed as replies to the question in (19); in the
question as well as the answers, the causee is the OBJ and cannot be the OBJ ,
and agreement is required:
(19) xoj xoll@-pt@-s-li?
who cry-Caus-Past-Obj.3Sg.Subj
Whom did he make cry?

Object agreement and grammatical function: Ostyak

155

(20) a. ma:ne:m xo:ll@-pt@-s-li

/ *xo:ll@-pt@-s
I.Acc
cry-Caus-Past-Obj.3SgSubj cry-Caus-Past-3SgSubj
He made me cry.
b. Pe:tra xo:ll@-pt@-s-li
/ *xo:ll@-pt@-s
Peter cry-Caus-Past-Obj.3SgSubj cry-Caus-Past-3SgSubj
He made Peter cry.
Thus, object agreement with goal and causee objects does not correlate with information structure: they may not be expressed as OBJ PATIENT / THEME, and there
is therefore no option to make use of the distinction betwen OBJ and OBJ to express their information structure status. In other words, non-patient/non-theme
objects must be realised as OBJ, no matter what their information structure role.
Note that non-patient/non-theme objects share all syntactic properties of
agreeing patient/theme objects. For example, they can control action nominal clauses:
(21) [pasa:n e:lti o:m@s-t-al sis] Juwan a:n-na ma-s-e:m
table at sit-AN-3Sg when John cup-Loc give-Past-Obj.1SgSubj
I gave John a cup when he was sitting at the table.
Other properties listed in Table (13) are also identical, which suggests that
non-patient/non-theme objects and agreeing patient/theme objects represent
the same grammatical function.
In sum, generalisations regarding verb agreement in Ostyak are purely syntactic; the verb agrees with the OBJ, but not with OBJ PATIENT / THEME. The agreement affix -am specifies that the subject is first person singular, and the object
is plural, but does not specify information structure role:
(22) Agreement specifications for the agreement affix -am:
(
(
(

SUBJ PERS)

=1
= SG
OBJ NUM ) = PL
SUBJ NUM)

In Ostyak, it is the mapping relations that are complex, referring not only to
semantic roles, as is usual in theories of mapping, but also to information structure: topical patients and themes map to OBJ, while nontopical patients and
themes are secondary objects OBJ . Agreement patterns then follow directly
from grammatical role.

156

Primary and secondary objecthood and DOM

8.3 Object casemarking and grammatical function

8.3.1 Mongolian
Khalkha Mongolian, as described by Guntsetseg (2008), also exhibits differential object casemarking.1 The accusative -(i)g is obligatory on definite NPs
(including personal pronouns, proper names and universally quantified objects)
and optional on indefinite objects (Guntsetseg 2008). This makes Mongolian
DOM different from DOM in Ostyak, which depends purely on information
structure role and not on particular semantic features, except for the general
requirement for the topic to be referential. DOM in Mongolian is partially dependent on semantic criteria: the accusative marker is required on all definite
objects, regardless of their topicality. The possible genesis of such semantic
conditions in processes of grammaticalisation is discussed in Chapter 10.
The correlation of marking with specificity in Mongolian is indirect. Whereas
the accusatively-marked indefinite object must receive a specific interpretation,
the nominative object can be either specific or nonspecific.
(23) a. Bold neg ohin uns-sen
Bold a girl kiss-Past
Bold kissed a girl/a certain girl.
b. Bold neg ohin-ig uns-sen
Bold a girl-Acc kiss-Past
Bold kissed a certain girl.
In general, animate specific indefinites are more likely to be casemarked than
indefinites, but this is only a tendency: there is no absolute correlation with
animacy. Indefinite inanimates, too, can be either marked or unmarked:
(24) ter neg zahia(-g) bich-sen
he a letter(-Acc) write-Past
He wrote a letter.
Affectedness of the object does seem to be a factor in DOM. Affected objects
of verbs whose semantics implies a change of state of the object (e.g. repair) are more likely to be marked with accusative than objects of intensional
verbs such as search. Again, however, this is only a general tendency: as
shown by the examples above, the same verb can take either a nominative or
an accusative object.
1 Examples for which the source is not mentioned come from personal communication with
Dolgor Guntsetseg.

Object casemarking and grammatical function

157

The nominative is required when the object is clearly in focus. Thus, the
object in (24) is zero-marked if the sentence as construed as the answer to the
question What did he write? or What did he do?. In such focus contexts,
accusative is generally impossible. Example (25) illustrates the same pattern:
(25) a. What did he do? Or: What did he copy?
b. ter neg uguulber(*-ig) huul-san
he a sentence-Acc copy-Past
He copied a sentence.
The accusative is equally impossible on indefinite objects if they have a nonreferential interpretation, as shown by the contrast in (26):
(26) a. bi neg oyutn-ig haij
baina. Ter ih uhaantai
I a friend-Acc look.for Prog he very clever
I am looking for a friend. He is very clever.
b. bi neg oyutan / *ojutn-ig haij
baina. Ter ih uhaantai baih
I a friend friend-Acc look.for Prog he very clever be
yostoi
should
I am looking for a friend. He should be very clever.
In (26b) the object friend is nonreferential: the speaker does not imply that
the referent of this NP exists. The accusative is strictly ungrammatical in this
context.
The question is, then, what triggers the accusative on a referential indefinite
object in (26a). We suggest that the relevant factor is topicality. We argued
in Chapter 3, Section 3.2.2 that under certain conditions indefinite NPs can
be interpreted as topics if they are pragmatically linked to another accessible
referent, or if the neighbouring context provides additional information about
them. Such context may involve relativisation, as in Persian (Chapter 6, page
114, example 11) and Danish (Chapter 3, page 54, example 7) or a coordinated
clause, as in Persian (Chapter 6, page 114, example 12). The same is observed
in Mongolian, as seen in example (26a). Here are two more examples:
(27) a. What did he do?
b. ter neg uguulber-ig / *uguulber huul-san gevch ter ni buruu
he a sentence-Acc sentence copy-Past but it Top wrong
bai-san
be-Past
He copied a sentence but it was wrong.

158

Primary and secondary objecthood and DOM

(28) a. What did you do yesterday?

b. bi uchigdur neg nom-ig / *nom unsh-aad duusga-san. ter ikh
I yesterday a book-Acc book read-Conv finish-Past it very
sonirkholtoi bai-san
interesting be-Past
Yesterday I finished a book. It was very interesting.
In both of these examples, the object is part of the broad predicate focus domain. However, it is topical within this domain, because it is interpreted as
pragmatically salient by the speaker who assesses its saliency when planning
the subsequent clause, in which the same referent is mentioned. In this context the accusative is obligatory. Without the continuation (but it was wrong
and it was very interesting) the accusative would be ungrammatical in this
context. This means that the accusative marks definite objects and indefinite
topical objects. Indefinite nontopical objects remain unmarked.
This conclusion stands in agreement with the suggestion made by Guntsetseg (2008) that the decisive factor determining DOM in Mongolian is discourse
prominence, in the sense of topicality. She tested this by means of an questionnaire experiment given to 320 native speakers for translation, consisting of 75
test sentences relating to DOM and 100 filler/control sentences. The questionnaire contained examples where the object is anaphorically cross-referenced in
the next clause (e.g. John kissed a girl and she slapped him) and examples
without such a relationship (e.g. John kissed a girl. James didnt come to
school today). The results show that the indefinite object is more likely to be
accusatively marked if the speaker intends to give further information about its
referent in the immediate subsequent discourse, where this object is normally
cross-referenced by anaphora.
The marked object in Mongolian may be either the primary or the secondary
topic. In the presence of a focused subject, topical marking is allowed on the
object, as shown in example (29b), where the object is more topical than in
(29a):
(29) a. Hen neg ohin uns-sen be?
who a girl kiss-Past Q
Who kissed a girl? (any girl)
b. Hen neg ohin-ig uns-sen be?
who a girl-Acc kiss-Past Q
Who kissed a girl? (a certain highly salient girl)

Object casemarking and grammatical function

159

As with the other languages discussed in this chapter, marked and unmarked
objects in Mongolian have different syntactic properties, and we assign them to
different grammatical functions based on these differences. First, they differ in
their ability to combine with the particle ni, whose distribution is syntactically
governed: it appears only on subjects and marked objects, and not on unmarked
objects.
(30) bi neg huuhd-ig / *huuhd ni hav-san
I a child-Acc child NI see-Past
I saw a (certain) child (one of the children).

(Guntsetseg 2008)

On subjects and definite marked objects, ni marks a topic which is usually,

but not always, contrastive. An example with ni-marking on the subject can
be found in the second conjunct of (27b). On indefinite marked objects, the
particle normally implies a partitive reading (one of a known set).
The second difference is that, although unmarked constituent order in Mongolian is SOV, accusative objects can be fronted to the pre-subject position;
this is impossible for nominative objects:
(31) neg zahia-g / *zahia bi bich-sen
a letter-Acc letter I write-Past
A letter, I wrote.

(Guntsetseg 2008)

These differences cannot be explained by assuming that the unmarked/nominative

object is required to appear in a particular phrase structure position. Although
there is a tendency for the unmarked object to be immediately preverbal (similar to Ostyak, Turkish, Hindi and many other SOV languages, as discussed
earlier), Mongolian does not seem to have a dedicated focus position located
immediately before the verb. Both the marked and unmarked object can be
separated from the verb by other constituents:
(32) bi neg ohin(-ig) gudamj-ind zohi-son
I a cat-Acc street-Dat/Loc hit-Past
I hit a cat in the street.

(Guntsetseg 2008)

Unlike in Ostyak, this word order does not necessarily require a nonreferential
interpretation of the unmarked object. Further, the focus wh-question word
need not immediately precede the verb, as shown in example (33):
(33) ter haana neg baavgai al-san be?
he where a bear
kill-Past Q
Where did he kill a bear?

(Guntsetseg 2008)

160

Primary and secondary objecthood and DOM

Thus, a phrase structure based account cannot account straightforwardly for

the different behavioural properties of marked and unmarked objects, as in
Mongolian unmarked objects need not appear in a fixed phrase structure position. Instead, we analyse accusative and nominative objects as corresponding
to different grammatical functions, the former to OBJ and the latter to OBJ .
We treat the accusative affix -ig as simply marking its argument as an OBJ, in
line with Nordlingers (1998) constructive case theory, described in Chapter 2,
Section 2.7. As in Ostyak, rules governing argument mapping are complex,
involving semantic role, information structure role, and grammatical function,
but marking can be specified simply in terms of grammatical role:
(34) Specification for the accusative affix -ig:
(OBJ )
The theory of argument linking which specifies how the arguments of a verb
are syntactically realised must take into account the information structure role
of an argument in determining whether it appears as an OBJ or an OBJ , but rules
for determining casemarking are very simple, referring only to the grammatical
function of the marked argument.
8.3.2 Chatino
In Chatino (Zapotecan), described in Carleton and Waksler (2000, 2002), nonpronominal objects are optionally preceded by the preposition jiPi. Carleton
and Waksler (2002) argue in detail that the use of the preposition is determined
by the information structure role of the object. They claim that jiPi marks an
object as focused, perhaps on the basis of the (in our view, incorrect) assumption that a clause cannot contain more than one topic; however, they observe
that their analysis necessitates a wholesale recasting of the notion of focus in
order to account for the fact that the relevant referent is already present in the
previous discourse. We believe that the jiPi-marked object is better accounted
for by assuming that it is in fact topical and that what Carleton and Waksler
refer to as focus is actually a topic according to our definitions.
Carleton and Waksler show that animacy of the object does not correlate
with the distribution of jiPi: both animate and inanimate objects can be marked
with the preposition, but in both cases it is optional. The correlation with specificity is also incomplete; although nonspecific objects do not seem to allow jiPi,
specific objects are either marked or unmarked, as shown in (35):
(35) a. Juan -yuPu-nto:-yu (jiPi) Maria
Juan C-have-eye-3Sg Prep Maria
Juan recognised Maria.

(Carleton and Waksler 2002:159)

Object casemarking and grammatical function

161

b. nkw-ilo
(jiPi) tone Pe
C-rescue.3Sg Prep Loc.home
He rescued the house.

(Carleton and Waksler 2002:159)

Examples (35a,b) also show that definiteness is not enough to ensure prepositional marking, since the objects in these examples are definite but not necessarily marked. Except for a few lexicalised expressions where objects are
always nonspecific, the lexical semantics of the verb is usually irrelevant for
object marking.2
The distribution described above is predicted if we assume that the postposition marks topical objects. As mentioned above, nonreferential (nonspecific)
NPs are impossible as topics, and the postposition is impossible on such objects. As for specific objects, Carleton and Waksler (2002) further argue that
the function of jiPi is to zoom in one character, that is, to bring the object
referent to the addressees attention. They observe that the preposition signals
to the addressee how to identify a location in the discourse model where the
speaker wants to bring the centre of the addressees attention. This indicates
that the object referent is pragmatically salient, as was suggested above for
topical arguments. Moreover, it is highly likely to be present in the previous
discourse.
In (36), the primary topic referent is the man who is the main character of
the narrative:
(36) nka-lo-yu
jiPi na kuchilu-uP ntu-siPyu-yu yane
C-remove-Top Prep Det knife-Spec P-cut-Spec neck
He took (his) knife and began to cut (his) throat.
(Carleton and Waksler 2002:159)
This referent does not receive overt linguistic expression, but the verb bears the
topic marker, indicating that its subject is the topic of the clause. The object
knife is marked as specific and is preceded by the definite article because it
has been mentioned in the previous discourse. The narrative tells the story of
an eagle and a serpent that terrorise the community. The second part of the
narrative describes the plan to kill the eagle and the knife that was specifically
purchased for this purpose. Thus, the pragmatic relation between the main
character and the knife is well-established in the consciousness of the speaker
2 Carleton and Waksler (2002:162) point out that the preposition tends not to occur on objects of
verbs whose lexical semantics entails most or all of the semantic features of the objects. These
are verbs such as extinguish (fire), earn (money) or husk (corn). The objects of these verbs
need not be individuated, since they are perceived as parts of the event denoted by the verb. As
Carleton and Waksler (2002:163) note, the speaker does not need to call the addressees attention
to that object, because its presence is already understood or expected.

162

Primary and secondary objecthood and DOM

and the addressee at the time (36) is produced. Carleton and Waksler state
that, although the referent of knife is not new, the relationship between this
referent and the proposition is, which makes the knife focused. The first
clause in (36) updates this relation: the clause is construed to be about the
relation that holds between the man and his knife. By our criteria, the object is
the secondary topic, and that is why it is marked by the preposition jiPi.
Note that in the second clause and began to cut his throat the object throat
is also definite and specific, as indicated by the specific marker on the verb, but
it has not been mentioned in the previous discourse and is not expected to be
pragmatically salient at the time of the utterance. This object does not have
topical status, but rather inhabits the domain of wide focus, and is not marked
with the preposition.
Examples (37a) and (37b) are taken from the same text.
(37) a. nte-su
nchiPyu nte-su
nchiPyu na nt-yoti-na
Hab-cut.3Sg fruit
Hab-cut.3Sg fruit
Neg Hab-know-1Pl
tukwi nchiPyu nte
what fruit
Hab.be
He cuts fruit, he cuts fruit; we dont know what kind of fruit it is.
b. nku-tyejna ntu-su-kaPa na nyate -e P
jiPi na nchiPyu
C-begin Hab-cut-again Det person-Spec Prep Det fruit
The man began to cut the fruit again.
(Carleton and Waksler 2002:167)
In (37a) the fruit is introduced for the first time, and the object preposition
is not used. In (37b), which occurs later in the text, the relation between the
subject referent and the object referent is already established and the object
is pragmatically salient. This is the same fruit as was mentioned in (37a).
The new information provided by (37b) is that cutting of the fruit by the man
occurred again. This indicates that this fruit is under discussion when (37b) is
produced and corresponds to the secondary topic.
Carleton and Waksler (2002) also discuss example (38), which appears in a
context in which both the subject and the object are under discussion; in this
example, the object receives topic marking.
(38) yna Pa ku-tze-o P jiPi kosa na tzoPo
hence P-fear-3Pl Prep thing Neg good
Hence they fear (those) bad things/creatures.
(Carleton and Waksler 2002:166)

Object casemarking and grammatical function

163

This example occurs in a narrative about hostile semi-human creatures who

prey on the women of the village, so the pragmatic relation between the villagers (the referent of the subject in (38)) and the creatures (the referent of the
object) is well established. Sentence (38) occurs at the end of the narrative,
after a number of other sentences that describe the relation between the villagers and the creatures. The function of the utterance in question is to update
the addressees knowledge about this relation. In fact, Carleton and Waksler
(2002:167) note that the preposition normally occurs at the right boundary of
a scene, sometimes, though not always, corresponding with the end of a narrative discourse. This means that the referents involved in respective propositions are likely to be familiar to the interlocutors and the relationship between
them is pragmatically established.3
Although marking on lexical objects depends on topicality, object personal
pronouns must cliticise to the objective preposition, independently of their information structure status (Carleton and Waksler 2000). Thus, grammatical
marking is extended to all pronominal objects in Chatino:
(39) nk-yaja-nto-o P jiPi-chuP
C-found-eye-1Sg Prep-3SgFem
I saw her (from afar).

(Carleton and Waksler 2000:390)

This is similar to the constraints in Mongolian requiring marking on all definite objects, independent of topicality; we discuss patterns such as these in
Chapter 10.
We do not have much evidence about the syntactic properties of objects in
Chatino. However, at least one piece of data suggests that casemarked and
noncasemarked objects behave differently. Chatino is a verb-initial language,
but Carleton and Waksler (2000) show that under certain pragmatic conditions,
subjects as well as marked objects preceded by the preposition jiPi can be
dislocated by what they call Focus Dislocation into sentence-initial position.
Recall that Carleton and Waksler call focus what we understand as topic, so
it is likely that we are dealing with topicalisation rather than focus here. In
(40), the subject appears sentence-initially:
nchiPyu
(40) tzaka ynate nte-su
one human Hab-cut.3Sg fruit
A man is cutting fruit.

(Carleton and Waksler 2000:395)

3 Carleton and Waksler (2002) also note that objects may be marked when disambiguation is
needed between a complex subject consisting of two words, on the one hand, and the subject and
the object, on the other hand. We do not address such cases here.

164

Primary and secondary objecthood and DOM

Marked objects (those marked with the preposition jiPi) can also be fronted,
while the preposition either appears together with the object or remains stranded.
(41) a. jiPi kiPyu nka-ra kuna Pa
Prep man C-hit woman
The woman hit the man.
b. kiPyu nka-ra kuna Pa jiPi
man C-hit woman Prep
The woman hit the man.

(Carleton and Waksler 2000:396)

On the other hand, objects unmarked with the preposition cannot appear in
sentence-initial position. The position of the marked and unmarked object is
otherwise the same, so this does not seem to be because unmarked objects
must appear in a particular focus position. These patterns are similar to the
Mongolian data discussed above, and seems to reflect a syntactic difference
between the two kinds of objects: subjects and marked objects can be fronted,
but unmarked objects cannot.
We suggest that the assignment of object grammatical functions depends on
pronominality and topicality, and we assume the same treatment of the Chatino
preposition jiPi as for the Mongolian accusative affix - ig, given in (34) above:
-ig marks its argument as OBJ.
8.3.3 Hindi
Hindi (Indo-Aryan) exhibits differential object casemarking coupled with a
complicated agreement system (Mohanan 1994: Chapter 5). In example (42),
the verb agrees with the uncasemarked subject:
(42) a. Ravii
baalak-ko ut.haaegaa
Ravi.Masc boy-KO lift.Fut.3MascSg
Ravi will lift up the boy.
b. Niinaa
baalak-ko ut.haaegii
Nina.Fem boy-KO lift.Fut.3FemSg
Ravi will lift up the boy.

(Mohanan 1994:103)

When the subject is casemarked, and there is a noncasemarked object, the verb
agrees with the object:
(43) Ravii-ne rot.ii
khaayii
Ravi-Erg bread.Fem eat.Perf.3FemSg
Ravi ate bread.

(Mohanan 1994:103)

Object casemarking and grammatical function

165

When both the subject and the object are casemarked, the verb shows neutral
(third person masculine singular) agreement:
(44) Ravii-ne baalak-ko ut.haayaa
Ravi-Erg boy-KO lift.Perf.3MascSg
Ravi ate bread.

(Mohanan 1994:103)

Mohanan (1994) demonstrates that agreement patterns in Hindi are dependent

on casemarking patterns: on her analysis, the verb agrees with the uncasemarked argument which is highest on the thematic hierarchy. Casemarking
on subjects of transitives and some intransitive verbs depends on aspect and
agentivity. Our main concern in the following is the distribution of the casemarker ko on objects; ko is also possible on experiencer subjects and on subjects of passive verbs, but we will not be concerned with these uses of ko (see
Mohanan 1994 for discussion of these patterns). We return below to a discussion of agreement patterns and how they can best be characterised.
Recall the basic pattern of object marking in Hindi, repeated here from
Chapter 1.4 As observed by Mohanan (1994), objects that bear ko are necessarily interpreted as specific. Human/animate specific objects are always marked,
inanimate/nonhuman indefinite objects are always unmarked, and inanimate/nonhuman
definite objects may be either marked or unmarked.5
(45) Ilaa-ne bacce-ko/*bacca ut.haayaa
Ila-Erg child-KO/child lifted
Ila lifted the/a child.

(Mohanan 1994:80)

(46) Ravii-ne kaccaa kelaa kaat.aa

Ravi-Erg unripe banana cut
Ravi cut the/a unripe banana.

(Mohanan 1994:87)

(47) Ravii-ne kaccaa kele-ko

kaat.a
Ravi-Erg unripe banana-KO cut
Ravi cut the/*a unripe banana.

(Mohanan 1994:88)

Note that human/animate specific objects always take ko even if they are nontopical/focused, as shown in (48) and (49):
(48) Hassan kis-ko/*kaun maaregaa?
Hassan who-KO/who kill.Fut
Whom will Hassan kill?
4 Examples for which the source is not given come from personal communication with Devyani
Sharma and Tara Mohanan.
5 There seems to be a certain amount of (dialectal) variation in this. For example, Butt (1993)
states that object marking is possible for specific indefinite inanimates.

166

Primary and secondary objecthood and DOM

(49) Hassan kisi-ko/*koi

maaregaa
Hassan someone-KO/someone kill.Fut
Hassan will kill someone.
For inanimate/nonhuman definite objects, casemarking appears to be optional;
however, we argue that casemarking patterns are best explained by taking
information structure roles into account. This fits with observations by McGregor (1972), who claims that object marking requires individualisation and
a certain degree of contextual importance for the object, and with Butt and
Kings (1996) proposal that marking patterns are related to information structure roles. In their analysis, focused objects are unmarked and are licensed in
the immediately preverbal position, which also hosts nonspecific objects. This
is confirmed by our own data. Definite inanimate objects remain unmarked,
or at least ko is strongly dispreferred, if the object is focused. The following
sentences were produced as the answers to the question What happened?, and
thus exemplify wide focus.
(50) a. What happened?
b. Hassan-ne meraa kalam/?kalam-ko tod. diyaa
break gave
Hassan-Erg my.Masc pen/pen-KO
Hassan broke my pen.
(51) a. What happened?
b. [jis kitaab ke-bare-me aap baat kar rahe the] [voh
which book about-Loc you talk do Prog Past that
kitaab/?kitaab-ko mai-ne kharid li]
book/book-KO I-Erg buy take.Past
I bought that book you were talking about.
In the following examples, the object is in narrow focus.
(52) a. What did Hassan sell?
b. Hassan-ne voh kitaab/?kitaab-ko bec diyaa
Hassan-Erg that book/book-KO sell gave
Hassan sold that book.
jo dukaan-me
c. Hassan-ne voh kursi/*kursi-ko bec diyaa
Hassan-Erg that chair/chair-KO sell give.Past Rel shop-Loc
thi
be.Past
Hassan sold that chair that was in the shop.

Object casemarking and grammatical function

167

(53) a. What is Anu doing?

b. Anu kitaab pad.h rahi hai
Anu book read Prog Pres
Anu is reading the book.

(Dayal 2003:84)

Again, there is no ko-marking on focused objects.

Specific non-focused objects can be either marked or unmarked. However,
Butt and King (1996) say little about the conditions on their marking; they
notice that accusative objects are generally more compatible with telic constructions and affected readings, but do not elaborate on this point. Our data
show instead that aspectual differences are at best secondary, whereas contextual factors play a prominent role. In contexts where the object is salient and
the utterance updates the addressees knowledge about the relation that holds
between the subject and the object referents, ko-marking is required, as in example (52) above and the following example:
(54) a. What did Hassan do to the book?
b. us kitaab-ko/us-ko/*voh kitaab bec diyaa
that book-KO/that-KO/that book sell gave
He sold it/that book.
In (52) the object is in narrow focus and fills the informational gap in the
presupposed open proposition Hassan sold X. In contrast, (54) is acceptable
in the context of a certain pragmatically established relationship between the
referent of Hassan and the referent of the book. The utterance updates the
information about this relationship and thus can be interpreted in our analysis
as a secondary topic construction. There is no apparent aspectual difference
between the two. The contrast between (52b) and (54b) indicates that objects
of the same verb in the same tense/aspect can be either marked or unmarked,
depending on their information structure role.
The examples below provide further illustration of the requirement for topical objects to be marked. Some of these examples additionally demonstrate
that affectness is not directly relevant for marking: the object of the verb see
is the least affected by the event described in the sentences, but it behaves just
like the highly affected object of the verb break with respect to grammatical
marking.
(55) a. You should punish Hassan for breaking your pen.
b. mere kalam-ko/?kalam Ravi-ne tod.aa Hassan-ne nah
Ravi-Erg broke Hassan-Erg no
my.Obl pen-KO/pen
It was Ravi who broke my pen, not Hassan.

168

Primary and secondary objecthood and DOM

(56) a. Did you read the book I was talking about?

hi kharidaa
b. abhi nah, mai-ne us kitaab-ko/?kitaab kal
now no I-Erg that book-KO/book yesterday Foc bought
Not yet, I only bought that book yesterday.
(57) a. Did you buy that pen?
b. nah, lekin mai-ne kalam-ko/us-ko/*kalam dukaan-me dekhaa
no but I-Erg pen-KO/that-KO/pen
shop-Loc saw
No, but I saw that pen/it in the shop.
Consider also the following contrast:
(58) a. ham mez puch dege
I.Hon table wipe give.Fut
I am going to wipe the table.
b. mez-ko puch do
table-KO wipe give.Imp
Wipe the table!
Both examples in (58) are acceptable in a situation in which the table has
not been mentioned in the previous discourse but is present in the situation
of speech and therefore identifiable and definite. However, the information
structure role of the object differs. According to our consultants, the communicative purpose of the first utterance is to inform the addressee about the
speakers intention. The object referent is construed as part of the event and
is not individuated as a pragmatically salient element. Informationally, it is
a part of wide focus, so ko-marking is absent. In the second case, the event
is presented as a plan. The utterance is construed as an instruction to the addressee to perform a certain action with respect to the table, so the table plays a
prominent pragmatic role in the situation. The object NP is topical and marked
with ko. Thus, while indefinite nonhuman/nonanimate objects are unmarked,
for definite inanimates ko-marking is not merely optional: it is determined by
topicality of the object.
Marking with ko is also possible when the subject is in focus:
(59) kalam-ko kis-ne tod.aa?
pen-KO who-Erg broke
Who broke the pen?

Object casemarking and grammatical function

169

This example demonstrates that ko-marking is not necessarily confined to the

secondary topic.
Butt and King were among the first to propose that monotransitive verbs can
take either OBJ or OBJ , depending in part on information-structural factors. On
their analysis, the factor determining the grammatical function of the object is
position: an unmarked, nontopical object which is immediately preverbal has
a different grammatical function from an object which may be marked, and
whose position is unconstrained. They propose to analyse unmarked objects
that are immediately preverbal as OBJ, and marked or unmarked objects whose
position is unconstrained as OBJ .
Our analysis differs from Butt and Kings in two respects. First, we believe
that when marking correlates with a difference in grammatical function, the
marked, topical object tends cross-linguistically to be associated with properties typical of core arguments, and the unmarked, nontopical object tends to
exhibit fewer core properties: in the languages we have examined in this chapter, the marked, topical object is the primary object, OBJ, while the unmarked,
focused object is a secondary object, OBJ . This is opposite to the distribution
that Butt and King suggest for Hindi.
Second, we do not agree with Butt and King that the Hindi data motivate the
postulation of a distinction in grammatical function based on phrase structure
position, since we know of no independent evidence that would motivate such
a distinction. Instead, we believe that Hindi is like the other languages examined in this chapter: the grammatical function of an argument is signalled by
its case. Marked objects are OBJ, and unmarked objects are OBJ . This analysis
makes possible a very simple statement of agreement patterns in Hindi. As
shown in examples (4244), the verb agrees with the subject unless it is casemarked; in that case, it agrees with the object unless it is casemarked; if both
the subject and the object are casemarked, the verb shows neutral agreement.
Mohanan (1994) analyses this pattern in terms of the position of the arguments
of the verb on the thematic hierarchy: the verb agrees with the uncasemarked
argument which is highest on the hierarchy. If we assume that marked objects
are OBJ and unmarked objects are OBJ , we can restate this generalisation in
purely syntactic terms: the verb agrees with the subject unless it is casemarked,
in which case it agrees with OBJ , which is always unmarked.
The formal analysis of ko is exactly as in Mongolian and Chatino: ko marks
the primary object, OBJ. Mapping rules for Hindi require animate and topical
definite objects to be realised as OBJ, while inanimate and nontopical objects
are OBJ .

170

Primary and secondary objecthood and DOM

8.4 Objects and markedness

The discussion in this and previous chapters raises the question of functional
markedness: which objects, topical (formally marked) or nontopical/focused
(formally unmarked), count as prototypical, canonical, or functionally unmarked?
The prototypical alignment between syntax and information structure assumed in many functional and typological works is the SUBJ/TOPIC, OBJ/FOCUS
alignment. In other words, it is commonly assumed that prototypical/canonical/functionally
unmarked objects are in focus (Du Bois 1987, Bossong 1989, Lambrecht 1994,
Maslova 2003b, Bossong 1991, Jelinek and Carnie 2003, and many others).
Bossong (1991:158), for instance, argues that objects often form an integrated part of the verbal complex, so that the object and the verb belong to
the same pragmatic domain (in our terminology, focus domain or comment).
This idea is implicitly present in many analyses of DOM, in particular those
that highlight its disambiguating function (see Chapter 1, Section 1.2.1), even
though not many of them explicitly address information structure. For example, in Aissens (2003a, 2003b) analysis, prototypical (functionally unmarked)
objects lack features that are ranked high on the Prominence Scales. But these
features are known to characterise topics, so functionally unmarked objects are
assumed to be nontopical/focused.
However, we have argued that objects tend to be more topical than other
nonsubject grammatical functions. This idea obviously stands in contrast with
the frequent claims that objects are prototypically in focus. Our analysis avoids
this contradiction by relying on a more elaborated view of information structure than it is usually assumed and on the notion of secondary topic, as well
as a richer inventory of objects, including primary objects (OBJ) as well as
secondary objects (the family of grammatical functions OBJ ).
We have shown that while subjects are normally primary topics, (primary)
objects are the best candidates for secondary topichood. In Chapter 3 we discussed informational contexts in which secondary topic objects are likely to
appear. They include argument-focus structures in which the focus is associated with an oblique element (e.g. Where did he read the book?) and structures
in which focus is associated with the verbal element alone, to the exclusion of
all nonverbal elements (the so-called verum focus, e.g. He DID read the
book, and probably other structures as well. However, contexts that involve
nontopical, focused objects are also frequent. They include utterances where
the narrow focus falls on the object alone (answers to questions like What did
he write/read/eat/etc.?) or broad focus structures where the object is included
in the broad focus domain but is not pragmatically presupposed (What did he
do? He read a book).

Objects and markedness

171

This functional duality has a syntactic corollary in some languages: this

chapter has examined languages in which the distinction between marked and
unmarked objects goes beyond grammatical marking and correlates with a difference in grammatical function. Topical, grammatically marked objects often
pattern with subjects, and show more characteristics of core grammatical functions than nontopical objects, and we analyse the distinction in terms of the
difference between the primary object OBJ and the secondary, restricted OBJ .
In languages of this type, grammatically marked objects are primary objects:
they are topical and can express various semantic roles. In contrast, unmarked
objects are semantically restricted, nontopical (focused) secondary objects.
Assuming a grammatical function hierarchy with obliques toward the bottom and SUBJ/OBJ at the top, topics tend to link to arguments high on the hierarchy, while focus tends to go with lower ranked elements. Since objects
are located in the middle of the hierarchy, they demonstrate a split: they are
equally appropriate as topics or foci. This gives the following principles of
default alignment between f-structure and information structure:6
(60)

TOPIC

TOPIC2

SUBJ

OBJ

FOCUS

OBJ / OBLIQUE

In other words, the dual linkage of objects in the languages discussed in this
chapter reflects a cross-linguistic tendency for splitting the object into two distinct grammatical functions depending on information structure role: a topical
OBJ and a focused OBJ . This split is not universal, though: the languages discussed in Chapter 7 do not align information structures roles and grammatical
functions, but rather display one informationally undifferentiated object function. In such languages semantic role takes priority over information structure
role in mapping relations.
On this more articulated view, topical objects are not functionally marked;
rather, we suggest that objects are equally likely to be topical and nontopical (focused). Thus, our analysis does not relate formal markedness on objects with their functional markedness. In fact, topical objects can be clearly
shown to be functionally unmarked, if markedness is reflected in textual infrequency: topical objects are common in human discourse, and formally marked
objects are just as frequent in languages with DOM as formally unmarked objects. This is supported by corpus data and corpus counts. In Givons (1979)
text counts, 50% of objects were found to be definite; definiteness, of course,
6 B
orjars and Vincent (2008) propose to do away with the primary OBJ/secondary OBJ distinction, and treat all objects as instances of OBJ . Following this suggestion would not allow us
to express the alignment generalisation in (60), which we believe to be an important fact about the
information structure/syntax interface.

172

Primary and secondary objecthood and DOM

strongly correlates with topicality. More direct evidence comes from Tundra
Yukaghir (isolate), as presented by Maslova (2003b). Yukaghir has a grammaticalised system of focus marking on subjects and objects in which the case
marker of the subject or object indicates its focus vs. nonfocus role. This
makes it particularly easy to identify the information structure role of an argument, based on its casemarking. Maslova presents the following frequency
counts of information structure patterns in finite transitive clauses:
(61)

topical subject, topical object:

topical subject, focused object:
focused subject, topical object:

ca. 65%
ca. 35%
less than 1%

These data demonstrate that topical objects (65%) are in fact more frequent
than focused objects (35%), at least in the narrative texts on which the count
was based.7
Roughly similar results have been obtained from Ostyak textual counts. The
table below, updated from Nikolaeva (1999), shows object agreement patterns
attested for transitive clauses in the collection of Ostyak texts recorded and
published by Papay (1906-1908).
(62)
total
(S)OV
(S)OXV

no object agreement
403 (40%)
329 (74%)
74 (15%)

object agreement
611 (60%)
197 (26%)
414 (85%)

Agreeing/topical objects are slightly more frequent than nonagreeing/nontopical

objects (60% in the first line of the table), so in this sense agreeing/topical objects cannot be considered functionally marked.
The table also demonstrates that in most (S)OV clauses clauses containing an object and no other nonsubject elements the object does not agree
(74%); such objects can be analysed as focused. In the remaining 26% of
(S)OV clauses the agreeing object is topical, while the focus domain includes
the predicate alone, as in the English example (9c) discussed in Chapter 3
(page 55) and the Ostyak example (8) of this chapter. In (S)OXV clauses, the
object normally agrees (85%); on our analysis, such objects are secondary topics. The element X is an oblique constituent, which is focused and normally
located in the preverbal focus position.8 This indicates that in clauses containing no other material than the subject, object and verb, the object is indeed a
likely focus. Importantly, however, when an oblique constituent is present, it
7 Maslova (2003b) states that although more than 97% of focus elements in transitive clauses
are objects, most objects are encoded as topics.
8 We also have a small number of (S)XOV examples, which require special explanation. Their
number is insignificant, however.

Conclusion

173

is this constituent that tends to be associated with focus.9 In such cases, the
object has the status of secondary topic.
This suggests that prototypical alignment, which relates objects with the
focus status, is mostly valid for clauses with no oblique constituents (SOV).
When a clause contains one or several obliques, the focus function is typically
associated with this oblique, while the object may be topical. This situation is
predicted by our analysis and the notion of secondary topichood. We can see,
then, that objects are just as likely to be topics as to be in focus.
8.5 Conclusion
Objects have two prototypical/canonical/unmarked information structure functions: focus and secondary topic. This, together with the fact that topics tend to
be grammatically marked, provides an explanation for patterns of DOM in the
languages we addressed in Chapter 7. Further, some languages restrict nonsubject topic marking to objects, and the realisation of an argument as a primary
object (OBJ) or a secondary object (OBJ ) can depend on its information structure role: in the languages examined in this chapter, topical arguments are
realised as OBJ and nontopical arguments are realised as OBJ , insofar as this is
consistent with additional constraints on argument realisation and expression
and the inventory of grammatical functions in the language.

9 In 15% of SOXV clauses, the object is in focus, while X is a component of a complex predicate
and does not bear a distinct information structure role; see Nikolaeva (1999) for details.

Multiple objects and

grammatical alignment
Although the syntax of ditransitives and verbs with multiple complements is
not the main focus of this book, we offer here some general observations on
multiple object constructions in languages with DOM, and a review of alignment possibilities involving grammatical function, information structure role,
semantic role, and marking for monotransitive and multitransitive constructions. For a thorough discussion of double object constructions in an LFG
setting, see Borjars and Vincent (2008) and Lam (2008).
9.1 On the typology of multitransitive constructions
In Chapter 2, Section 2.2, we reviewed the classic LFG theory of grammatical functions, which includes a distinction between primary and secondary
objects, originally called OBJ and OBJ2 (Bresnan 1980, Kaplan and Bresnan
1982). The theory was subsequently refined by Bresnan and Kanerva (1989),
who proposed to replace the OBJ/OBJ2 distinction with a distinction between
the primary object OBJ and the family of secondary object functions OBJ .
The primary/secondary object distinction is also explored by Dryer (1986),
who proposes a rough typology of argument linking patterns with trivalent
verbs like give,1 which take a theme and a goal. Dryer distinguishes two
classes of give-type constructions. His first class draws the traditional distinction between a theme direct object and a goal indirect object; in LFG, this
would be analysed as involving an OBJ theme and an oblique goal (OBL GOAL ),
1 Lam (2008) points out that in some languages, the verb give is actually the least prototypical
member of the class of trivalent verbs, rather than the most canonical member. We do not believe
that this is true of the languages we discuss in this chapter.

175

176

Multiple objects and grammatical alignment

as in (1a). This alignment pattern has also been referred to as indirective.

Dryers second class distinguishes primary and secondary objects (OBJ vs.
OBJ THEME ), as in (1b). This pattern is commonly referred to as ditransitive,
and also been termed secundative (Haspelmath 2007, Malchukov et al. 2007;
see also Lam 2008).
(1) a. I gave a book to Bill. (indirective)
b. I gave Bill a book. (secundative)
These can be schematically represented as in (2):
(2)
indirective (Dryers class 1):
secundative (Dryers class 2):

THEME

GOAL

OBJ

OBL GOAL

GOAL

THEME

OBJ

OBJ THEME

Dryers analysis of English examples like (1b) has been challenged by Borjars
and Vincent (2008), who argue (following work by Hudson 1992) that the
theme in examples like (1b) behaves more like a primary object than the goal
does. The behavioural evidence for the grammatical status of object-like arguments which we have examined in Chapters 7 and 8 underpins our discussion
of alignment patterns for trivalent give-type verbs in the following, though
more work is clearly needed on the syntactic status and classification of objects
in multitransitive constructions, even for the English construction exemplified
in (1b).
Kittila (2006a) elaborates on Dryers typology, exploring casemarking patterns with recipients, goals, patients, and themes and how they are affected
by the animacy of these arguments; for him, animacy is a cover term which
also encompasses definiteness and a general notion of prominence. He identifies three casemarking strategies for the complements of monotransitive and
give-type trivalent verbs. In the role-based strategy, the theme of a givetype verb and the patient of a monotransitive verb are marked in the same
way, and the goal is marked differently. In the second type, which Kittila calls
object-based, the patient argument of a monotransitive verb and both the
goal and the theme of a give-type trivalent verb bear object marking, independent of their animacy; this strategy requires all object-like arguments to be
marked in the same way. In the third type, called animacy-based, animate
(definite, prominent) patients, themes and goals are marked, while inanimate
patients, themes and goals are unmarked. For monotransitives, this gives rise
to DOM. There are two subtypes of this strategy for trivalent verbs, under the

On the typology of multitransitive constructions

177

assumption that the goal is always animate and marked: in extended differential object marking, both the goal and the animate theme are marked, while
the inanimate theme remains unmarked; in shifted differential object marking, the goal is marked, but the theme is unmarked, even if it is animate. In a
related paper, Kittila (2006b) explores animacy-based marking patterns in examples where both the goal and theme of a give-type verb are animate. For
such examples, he distinguishes theme-prominent languages, those in which
the theme is an object and the goal is an oblique, from recipient-prominent
languages, where the recipient is an object and the theme bears a different
grammatical function.
Kittilas classifications are primarily based on morphological marking (case
and/or agreement) associated with core arguments, which is taken to be definitional as a diagnostic of grammatical function for give-type verbs. The
same is true of Haspelmath (2007) and Malchukov et al. (2007), who discuss
patterns of ditransitive alignment from the point of view of what they refer
to as coding (case or adpositional marking, agreement and position). From
a marking/coding standpoint, there is also room for an additional neutral
alignment, in which the monotransitive theme, the ditransitive theme and the
goal all pattern in the same way.
However, the picture is more complicated if we recognise that grammatical
marking and grammatical functions are independent parameters: LFG does
not assume that identical casemarking and agreement patterns entail identical
grammatical properties (Mohanan 1982, Andrews 1982, Zaenen et al. 1985).
Work within LFG on English and many other languages shows clearly that
grammatical marking, including casemarking, agreement, and phrasal position, does not necessarily correspond one-to-one with grammatical function.
Indeed, in our analysis, theme-prominent languages may be divided into
two subtypes.2 In the first type, marking differences do not correlate with a
difference in grammatical function, and the marked and the unmarked theme
both correspond to the primary object, as discussed in Chapter 7. In the second
type, the marked theme is OBJ and the unmarked theme is OBJ , as discussed
in Chapter 8. Kittila and many other authors do not explore the possibility that
a difference in marking for the object of a monotransitive verb might correlate with a difference in grammatical function, although we have shown that in
some languages DOM itself can signal a difference in grammatical function.
More generally, existing typologies do not extend to languages like those examined in Chapter 8, in which a monotransitive verb can take either a primary
OBJ or a secondary OBJ (see also C
etinoglu and Butt 2008 and Dahlstrom
2009).
2 We have not observed a similar split in recipient-prominent languages: our data contains no
clear examples in which the ditransitive goal exhibits DOM.

178

Multiple objects and grammatical alignment

Our goal in the following is not to present a typology of ditransitive constructions in languages with DOM, but to show that it is likely to be more
elaborate than has been previously thought. A complete typology of ditransitive constructions must take several factors into account: the inventory of
grammatical functions available in the language, the mapping rules relating
grammatical functions in f-structure to semantic roles at argument structure,
the relationship between grammatical functions and grammatical marking, and
the availability of double object constructions.
9.2 No ditransitive construction
Several of the languages we have examined in Chapters 7 and 8 lack a ditransitive/secundative construction, and make use only of indirect alignment
for give-type verbs, involving a theme object and a goal oblique argument.
In Nenets there is only one object function, and a ditransitive construction is
therefore not available. Surprisingly, there are languages which have two object functions, and in which monotransitive verbs can take either a primary
OBJ or a secondary OBJ , but which nevertheless do not have a double object
construction: in these languages, a verb cannot take a primary object and a secondary object at the same time. Ostyak and Mongolian exemplify this pattern.
These languages show that the availability of a double object construction in
a language may depend on factors beyond the inventory of grammatical functions in the language.
9.2.1 Nenets
As shown in Chapter 7, Section 7.2, marked and unmarked monotransitive objects in Tundra Nenets correspond to the same grammatical function. Nenets
has only one object function, LFGs OBJ. Alignment involving semantic role,
information structure role, grammatical function, and grammatical marking
for Nenets monotransitives with third person objects can be graphically represented as in (3); recall that the verb never agrees with first or second person
objects.
(3) Nenets monotransitives with third person objects:
patient/theme
topic

nontopic
OBJ

agreement

no agreement

No ditransitive construction

179

A third person object argument maps to OBJ, whether or not it is topical.3 The
presence or absence of agreement depends on information structure role: the
verb agrees with topical objects, and not nontopical objects. Agreement patterns are, then, determined by a combination of person, grammatical function,
and information structure.
The goal argument of give-type verbs is expressed by a dative-marked
oblique:
(4) Petya Masha-noh ti-m
myiqNa
/ myiqNada
Peter Masha-Dat reindeer-Acc give.3SgSubj give.Obj.3SgSubj
Peter gave Masha a/the reindeer.
The dative-marked oblique goal, OBL GOAL , displays no object properties: it
cannot be passivised or relativised by means of the primary strategy relativising objects, and it does not participate in object control constructions. The
alignment pattern for give-type verbs is shown in (5): topical/agreeing and
nontopical/nonagreeing objects are OBJ, while the goal argument is an oblique
goal OBL GOAL , independent of its information structure role:
(5) Nenets give-type verbs with third person objects:
theme
topic

goal
nontopic

OBJ

agreement

topic or nontopic
OBL GOAL

no agreement

dative

As this diagram shows, information structure does not affect the syntactic realisation of the goal argument: it is an oblique goal, whether or not it is topical.
Further, there is no effect of the presence of the goal argument on agreement
possibilities for the OBJ: exactly as in the monotransitive construction, the verb
agrees with a topical OBJ and not a nontopical OBJ.
9.2.2 Ostyak
In Chapter 8, Section 8.2, we saw that agreeing and nonagreeing monotransitive objects in Northern Ostyak realise different grammatical functions: agreeing objects are primary objects, OBJ, while nonagreeing theme objects are secondary objects, OBJ THEME . Agreement depends purely on grammatical function. Information structure role determines whether theme arguments of mono3 As far as we know, there are no verbs in Nenets which take an object with a thematic role
other than patient/theme.

180

Multiple objects and grammatical alignment

transitive verbs are realised as OBJ or OBJ THEME : topical themes are OBJ, while
nontopical themes are OBJ THEME .
(6) Ostyak monotransitives with patient/theme objects:
patient/theme
topic

nontopic

OBJ

OBJ THEME

agreement

no agreement

We assume that Ostyak has only one restricted object, OBJ THEME , and no other
objective functions: no OBJ GOAL or OBJ CAUSEE, for example. Since Ostyak has
only OBJ THEME , objects with other semantic roles for example, causees
must be realised as agreeing OBJ, since the grammatical role of OBJ THEME is
reserved for themes. Alignment for monotransitives in Ostyak is as follows:
(7) Ostyak monotransitives with causee or goal objects:
goal, causee, ...
topic or nontopic
OBJ

agreement
Even though Ostyak has two object functions, they cannot appear together,
as arguments of the same verb: Ostyak does not have a double object construction. With verbs like give, either the goal or the theme must appear as
an oblique, resulting in two possibilities. If the theme is an object, the goal
is a dative oblique, as in (8); in that case, the goal can be either topical or
nontopical.
(8) [What did you do to the cup? or What did you do to Peter?]
ma a:n Pe:tra e:lti ma-s-e:m
/ ma-s-@m
I cup Peter to give-Past-Obj.1SgSubj give-Past-1SgSubj
I gave a/the cup to Peter.

No ditransitive construction

181

(9) Ostyak give-type verbs with theme object:

theme

goal

topic

nontopic

topic or nontopic

OBJ

OBJ THEME

OBL GOAL

agreement

no agreement

dative case

If the goal is realised as an object, it must be topical, and must agree; in that
case, the theme is a locative oblique, as in (10). Recall that the subject in
Ostyak is closely associated with the primary topic; hence, when the goal is
topical, the theme is associated with a nontopic role (assuming that utterances
with three topical arguments subject, object, and oblique are very rare or
nonexistent).
(10) [What did you do to Peter?]
ma Pe:tra a:n-na ma-s-e:m
/ *ma-s-@m
I Peter cup-Loc give-Past-Obj.1SgSubj give-Past-1SgSubj
I gave Peter a/the cup.
(11) Ostyak give-type verbs with goal object:
theme

goal

nontopic

topic
OBL THEME

OBJ

locative case

agreement

It is clear that agreement patterns depend entirely on grammatical function:

the verb agrees with OBJ, but not with OBJ THEME or obliques. As in Nenets, the
presence of an oblique goal argument does not affect agreement possibilities
for object themes: the verb agrees with the topical OBJ, but not the nontopical
OBJ THEME . When the object is a goal, as in (10), or a causee, as in examples (19)
and (20) in Chapter 8 (page 154), it is a primary object, OBJ, and must agree
with the verb. In this case, the theme cannot appear as OBJ, since that role is
filled by the goal and cannot be doubly filled. Ostyak differs from Nenets in
allowing the theme to be expressed as an oblique, OBL THEME ; this allows the
pattern diagrammed in (11), where the goal is an agreeing primary OBJ, and the
theme is an oblique. The existence of this possibility means that the relation
between information structure role and grammatical function is obscured even

182

Multiple objects and grammatical alignment

for a verb like give, since the primary OBJ in the give-construction can be
either a topical theme, or a goal with unspecified information structure role.
9.2.3 Mongolian
As shown in Chapter 8, Section 8.3.1, Mongolian monotransitive theme objects are primary objects (OBJ) and casemarked if topical or definite, and secondary objects (OBJ THEME ) and noncasemarked if nontopical and indefinite.
Diagrammatically, alignment between different grammatical levels in Mongolian monotransitives is similar to Ostyak:
(12) Mongolian monotransitives:
patient/theme
topic or
definite

nontopic and
indefinite

OBJ

OBJ THEME

casemarking

no casemarking

Like Ostyak, Mongolian does not have a ditransitive construction; again, we

see that languages can have both primary and secondary objects in their inventory of grammatical functions without also having a ditransitive construction,
where the two kinds of objects appear together. The goal argument of a threeargument verb such as give in Mongolian must be expressed by an oblique,
marked with the dative -d, and not an object:
(13) Bi Toya-d nom(-ig) og-son
I Toya-Dat book(-Acc) give-Past
I gave Toya a/the book.
The possibilities for expression of the arguments of the verb give in Mongolian are similar to Nenets, and more restricted than in Ostyak. The primary object in Ostyak can correspond to the goal, with the theme realised as
a locative oblique. In Mongolian the goal cannot be realised as the object of
a trivalent verb such as give, because Mongolian does not have a ditransitive
construction and also does not allow oblique themes. The goal of Mongolian
give-type verbs must be marked by dative case, as in Nenets, and we analyse
the dative-marked argument as an oblique goal, OBL GOAL .

Ditransitive constructions

183

(14) Mongolian give-type verbs:

theme

goal

topic or
definite

nontopic and
indefinite

OBJ

OBJ THEME

OBL GOAL

casemarking

no casemarking

dative

topic or nontopic

Since the goal argument is realised as an oblique phrase in this construction,

there is no obstacle preventing the theme of a give-type verb from surfacing
as either a primary OBJ or a secondary OBJ THEME : it may be either marked (OBJ)
or unmarked (OBJ THEME ), depending on definiteness and information structure
role.
All three languages examined in this section, Ostyak, Nenets and Mongolian, can be classified as theme-prominent languages from Kittilas (2006b)
perspective. However, only in Nenets does the theme invariably correspond
to the primary object. In Ostyak and Mongolian, grammatical functions are
closely aligned with casemarking, and the theme is either marked OBJ or unmarked OBJ THEME .
9.3 Ditransitive constructions: Goal as primary object, theme as secondary object
Although LFG does not treat casemarking as an unambiguous indicator of
grammatical function in every language, casemarking (together with other factors) is often a good indication of grammatical function, as demonstrated by
casemarking patterns in Ostyak and Mongolian. Similarly, object casemarking
in Chatino and Hindi correlates with grammatical function, as shown in Chapter 8: the marked object of a monotransitive verb is the primary OBJ, and the
unmarked object is the secondary OBJ . In languages which have a ditransitive
construction and also exhibit a tight alignment between marking and grammatical function, we do not expect to find both objects of a ditransitive verb
marked in the same way, since it is not possible for both objects to correspond
to the same grammatical function. Further, we expect the primary object of
a ditransitive verb to be marked in the same way as the primary object of a
monotransitive verb. This is what we find in Chatino and Hindi: in both languages, the goal of a give-type verb is marked in the same way as the OBJ of a
monotransitive verb, while the theme remains unmarked. This alignment corresponds to the recipient-prominent type in Kittilas (2006b) typology. In fact,
it is cross-linguistically common for goals and topical monotransitive objects

184

Multiple objects and grammatical alignment

to be marked similarly, and Section 9.5 of this chapter discusses the pragmatic
and historical basis for this similarity.
Alignment patterns for Chatino and Hindi are, then, similar in some respects
to give-type verbs in Ostyak with goal objects, diagrammed in (11) of this
chapter. The crucial factor distinguishing Ostyak from Chatino and Hindi is
that while Ostyak does not have a double object construction, Chatino and
Hindi do.
9.3.1 Chatino
In Chapter 8, Section 8.3.2, we saw that the object of a monotransitive verb in
Chatino is marked with the preposition jiPi if it is either pronominal or topical,
and we argued on the basis of the Focus Dislocation construction that marked
and unmarked objects correspond to different grammatical functions, OBJ and
OBJ respectively.
(15) Juan -yuPu-nto:-yu (jiPi) Maria
Juan C-have-eye-3Sg Prep Maria
Juan recognised Maria.

(Carleton and Waksler 2002:159)

This can be diagrammatically represented as in (16):

(16) Chatino monotransitives:
patient/theme
topic or
pronoun

nontopic and
nonpronoun

OBJ

OBJ THEME

prepositional
marking

no marking

The same preposition marks not only objects of monotransitive verbs, but also
goal/beneficiary objects of verbs such as give or write. Here, the preposition
is obligatory:
(17) nka-ta
na kiPyu jne
jiPi/* Jua
C-give.3Sg Det man money Prep Juan
The man gave the money to Juan.

(Carleton and Waksler 2000:388)

(18) Nka-sa Pa
kiti iPi/* Jua
C-write.3Sg letter Prep Juan
She wrote a letter to Juan.

(Carleton and Waksler 2000:389)

Ditransitive constructions

185

Carleton and Waksler do not explicitly state that prepositional marking of the
theme in these examples is impossible, but they do not provide any examples
in which both arguments are marked. Under the assumption that only the goal
can be marked with preposition jiPi and the theme must remain unmarked, we
can analyse the goal as the primary OBJ, in keeping with the tight correspondence in monotransitive constructions between jiPi-marking and the OBJ function. We treat the unmarked theme as the OBJ THEME , just as in monotransitive
constructions.
(19) Chatino give-type verbs:
theme

goal

topic or nontopic

topic or nontopic
OBJ THEME

OBJ

no marking

prepositional marking

9.3.2 Hindi
Hindi, like Chatino, marks primary and secondary objects differently in monotransitive constructions: the casemarked object of a monotransitive verb in
Hindi is the primary OBJ, and the uncasemarked object is the secondary OBJ ,
as discussed in Chapter 8, Section 8.3.3.
(20) a. Ravii-ne kaccaa kelaa kaat.aa
Ravi-Erg unripe banana cut
Ravi cut the/a unripe banana.

(Mohanan 1994:87)

kaat.a
b. Ravii-ne kaccaa kele-ko
Ravi-Erg unripe banana-KO cut
Ravi cut the/*a unripe banana.

(Mohanan 1994:88)

In standard (prescriptive) Hindi (Masica 1982:20, Kellogg 1955 [1893]:399),

ditransitive constructions cannot contain two ko-marked objects: the goal must
be marked and the theme must be unmarked. As in Chatino, this is exactly what
we expect if the two objects correspond to different grammatical functions,
signalled by different casemarking patterns.
diyaa
(21) a. ilaa-ne maa -ko yah haar
Ilaa-erg mother-KO this necklace gave
Ila gave this necklace to mother.

(Mohanan 1994:85)

186

Multiple objects and grammatical alignment

b. *ilaa-ne maa -ko is haar-ko
diyaa
Ilaa-erg mother-KO this necklace-KO gave
Ila gave this necklace to mother.

(Mohanan 1994:85)

Assuming that casemarking is a reliable indicator of grammatical function in

Hindi, we analyse Hindi in the same way as Chatino: the marked goal is OBJ,
and the theme argument is OBJ THEME (exactly as shown in 19 for Chatino).
However, at least some nonstandard varieties of Hindi do allow the marking
of both objects by the same element ko, as shown in example (22):
(22) aadmii-ne us
kitaab-ko aurat-ko diyaa
man-Erg that.Obl book-KO woman-KO give.Past.MSg
The man gave that book to the woman.

(Kittila 2006b:302)

Similar data are cited by Bhatt and Anagnostopoulou (1996), who argue that,
despite identical marking, the theme and the goal argument have different
properties: for instance, only the theme object can be moved to the pre-subject
position via syntactic topicalisation.
(23) Tim-ko Theo-ne Sita-ko diyaa
Tim-KO Theo-Erg Sita-KO give.Past.MSg
Tim, Theo gave to Sita.
NOT: *Sita, Theo gave to Tim.
(Bhatt and Anagnostopoulou 1996:(19))
In these varieties of Hindi, ko-marking cannot be analysed as unambiguously
marking arguments as OBJ; instead, marking depends on a combination of syntactic and information structure role, as in the languages discussed in Chapter 7. In Kittilas classification, standard Hindi belongs to the recipient-prominent
type (goal is the primary OBJ). Nonstandard Hindi is likely to be a themeprominent language (theme is the primary OBJ, and goal bears some other
grammatical function), with no one-to-one alignment between grammatical
function and casemarking, similar to Nenets. Further work is needed to explore alignment patterns and the syntactic behaviour of marked and unmarked
objects in these nonstandard varieties, but it is possible that, unlike Nenets,
nonstandard Hindi has a double-object construction with the theme as primary
object and goal as a restricted object (possibly OBJ GOAL ).
9.4 Other multitransitive constructions
Alignment patterns with give-type verbs in the languages we have examined
in the previous sections are, for the most part, easily classified in terms of

Other multitransitive constructions

187

the two types of constructions examined by Dryer (1986), and echo the familiar patterns of the English dative alternation exemplified in (1). In Nenets,
Mongolian, and the theme=object alignment pattern in Ostyak, examined in
Section 9.2, the goal of a give-type verb corresponds to an oblique phrase,
and the theme behaves in the same way as a monotransitive object, with its
casemarking and, in Ostyak and Mongolian, its grammatical function (OBJ or
OBJ THEME ) dependent on topicality. This is the object/oblique or direct object/indirect object alignment, shown for English in example (1a) of this chapter. In contrast, the languages examined in Section 9.3 (Chatino and Hindi)
employ the primary object/secondary object, ditransitive alignment, with the
goal as the primary object and the theme as the secondary object, shown for
English in example (1b) of this chapter. In Chatino and Hindi, grammatical
function is tightly aligned with marking: the goal is marked in the same way
as the monotransitive OBJ, and the theme is marked in the same way as the
monotransitive OBJ THEME . We now turn to a discussion of languages that differ in interesting ways from these familiar patterns, though all of them allow
constructions with multiple objects.
We showed in Chapter 7 that marking in Dolakha Newar and Tigre for
monotransitive verbs does not correlate with grammatical function, but only
with information structure role: marking depends on topicality, but marked
and unmarked objects both correspond to the same grammatical function, OBJ.
Since there is a looser connection between casemarking and grammatical function for objects in these languages, it is not surprising that the theme and the
goal in a give-type construction can both be marked in these languages, as in
the nonstandard varieties of Hindi discussed in the previous section. In both
languages, the theme and the goal of a give-type verb both exhibit a range of
properties associated with objects, and it proves difficult to make a conclusive
identification of the object-like grammatical function borne by goal and theme
objects. We provide some discussion of object tests and the classification of
objects of give-type verbs for both languages.
Many languages have an applicative construction, briefly discussed in Chapter 7, Section 7.1, where a nonobject argument is promoted to object via applicativisation; the resulting construction often has more than one object. In
some languages, applicativisation is obligatory when a nonsubject argument is
topical. In Section 9.4.3, we discuss object marking and topicality in one such
language, Upper Nexaca Totonac (Totonac-Tepehua).
9.4.1 Dolakha Newar
Chapter 7, Section 7.3.2 discussed casemarking patterns in Dolakha Newar,
and concluded that (as in Nenets) there is no evidence for a difference in grammatical function between marked and unmarked theme objects in monotransi-

188

Multiple objects and grammatical alignment

tive verbs: both are OBJ, with casemarking patterns depending on information
structure role and pronominality (pronouns are obligatorily marked).
(24) Dolakha Newar monotransitives with nonpronominal objects:
patient/theme
topic or
pronoun

nontopic and
nonpronoun
OBJ

casemarking

no casemarking

With give-type verbs in Dolakha Newar, as in Chatino and Hindi, grammatical marking of the goal argument is morphologically identical to the marking
on the monotransitive topical patient/theme. Unlike the monotransitive object,
the goal must be marked, no matter what its pragmatic or semantic characteristics; omission of the object marker is reported to be impossible. This is true
not only for objects of give-type verbs, but also for objects of exchange and
beneficiary arguments:
(25) a. y
a-ta
dy
ab
a bi-en
ta-u-
rice-Obj money give-Part put-Past.1Sg
I had given money for the rice.

(Genetti 1994:52)

b. in
agu kh
a-ri
guntaN
da-hat
this.type matter-Ind nobody.Obj Proh-say
Dont tell anyone about this type of matter.

(Genetti 2007:114)

Given the lack of a tight correlation between grammatical function and marking in monotransitives, we might expect identical morphological marking on
the goal and the theme of a ditransitive clause, if appropriate informational
conditions are met. This is exactly what is observed: object marking can occur
twice within the same ditransitive clause, as shown in (26):
(26) ale amta
bh
anche-ta bir-ju
then 3Sg.Obj cook-Obj give-Past.3Sg
Then he gave her (in marriage) to the cook.

(Genetti 2007:316)

According to Genetti, patient and goal objects in ditransitive constructions

share many syntactic properties, and she analyses them as instantiating the
same grammatical function: object. On this view, Dolakha Newari represents
the so-called neutral alignment, where all three object-like arguments pattern
identically. This conclusion is based on the following three considerations.

Other multitransitive constructions

189

This conclusion is based on the following three considerations. First, the two
types of object can bear identical casemarking, as evidenced by example (26),
though casemarking on non-patient/non-theme objects is obligatory and not
optional.
Second, both types of object can antecede the emphatic reflexive element
ame tuN his/her own.
(27) a. r
am-na muc
a ame
tuN m
a-ta
bir-ju
Ram-Erg child 3Sg.Gen Foc mother-Obj give-Past.3Sg
Ram gave the child to his own (childs) mother.
(Genetti 1994:317)
b. r
am-na krisna-ta
ame
tuN kit
ab bir-ju
Ram-Erg Krishna-Obj 3Sg.Gen Foc book give-Past.3Sg
Ram gave Krishna his own (Krishnas) book. (Genetti 1994:317)
In (27a) the reflexive is controlled by the unmarked theme object, and in (27b)
by the marked goal object. Third, both types of object are relativised using the
same relativisation strategy, and are the only arguments to be relativised in that
way.
We draw a different conclusion from the results of these tests: we believe
that the objecthood tests discussed by Genetti pick out the entire class of object functions, primary OBJ as well as secondary OBJ . In other words, Genettis
tests pick out those arguments which are objective (+O) in LFG terms: see
Chapter 2, Section 2.2. In an LFG setting, Genettis claim that the patient and
the goal object of give-type verbs correspond to exactly the same grammatical function is untenable; two different semantic roles cannot correspond to the
same grammatical function. For theory-internal reasons, then, we must analyse
the goal and the theme as corresponding to two different object-like grammatical functions. In fact, there is at least one behavioural difference between the
two: marking on the goal argument is obligatory and independent of information structure role, while marking on the theme argument is optional and
correlates with topicality. We believe that this provides empirical motivation
for the claim that the theme and the goal correspond to different object-like
functions. The next question is what those object-like functions are.
The first possibility is that the object of a monotransitive verb and the theme
object of a give-type verb correspond to OBJ THEME , and the goal argument
corresponds to OBJ (this does not match the analysis which we presented in
Chapter 7, Section 7.3.2). On this analysis, casemarking is obligatory for OBJ,
and correlates with topicality for OBJ THEME .

190

Multiple objects and grammatical alignment

(28) Analysis 1 (to be rejected), Dolakha Newar:

Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ THEME

OBJ THEME

OBJ

We discount this possibility on two grounds: it is implausible that the theme

argument would never correspond to the primary OBJ (see Borjars and Vincent
2008 for discussion of the close relation between OBJ and theme), and it is also
implausible that primary OBJ would appear only in a ditransitive construction,
and not the monotransitive construction.
The second possibility is that the object of a monotransitive verb and the
theme object of a give-type verb are OBJ, and the goal argument is OBJ GOAL .
This would make give-type constructions in Dolakha Newar the reverse of
Hindi or Chatino:
(29) Analysis 2:
Monotransitive:
Dolakha Newar:

For comparison,
Hindi, Chatino:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ

OBJ GOAL

Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ THEME

OBJ

On Analysis 2, marking patterns are easy to account for: marking on OBJ depends on topicality, while marking on OBJ GOAL is obligatory. This is the analysis we hypothesized in Section 9.3.2 for non-standard Hindi.
The third possibility is that alignment in Dolakha Newar is exactly like Hindi
and Chatino: the goal is OBJ, and the theme is OBJ THEME :
(30) Analysis 3, Dolakha Newar:
Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ THEME

OBJ

This analysis has the virtue of making Dolakha Newar conform to alignment
patterns attested in other languages: not only Hindi and Chatino, but also English, at least on the secundative analysis advocated in traditional LFG analyses
(Bresnan 1980, Kaplan and Bresnan 1982) and by Dryer (1986). The cost is

Other multitransitive constructions

191

a considerable complication of the constraints on marking: on this analysis,

-ta-marking is obligatory for primary objects which are goals, and dependent
on topicality for theme objects, whether OBJ or OBJ THEME . Without additional
data, we are unable to determine whether Analysis 2 or Analysis 3 better reflects the behaviour of objects in give-type constructions in Dolakha Newar.
9.4.2 Tigre
As in Dolakha Newar, marking of monotransitive objects in Tigre depends not
on grammatical function, but on topicality and definiteness. We showed in
Chapter 7, Section 7.3.1 that prepositionally marked objects are definite and
topical, and unmarked objects are indefinite or definite nontopical:
(31) Tigre monotransitives:
patient/theme
definite
topic

definite nontopic
or indefinite
OBJ

prepositional marking

no marking

The Tigre verb shows agreement with definite objects, but this does not depend on topicality: definite objects control agreement, while indefinite objects
do not. Interestingly, with give-type verbs the controller of agreement is the
leftmost object in terms of its linear position, independent of its semantic role
as theme or goal. In both (32a) and (32b) the leftmost argument controls agreement: the goal in (32a), and the theme in (32b):
la waraqat habetto
/
(32) a. Lilat
P1g1l c ali
Lilet.Fem Prep Ali.Masc the letter.Fem gave.3Fem.3Masc
*habett
a
gave.3Fem.3Fem
Lilat gave Ali the letter.

(Jake 1980:75)

habett
a
/
b. Lilat
la waraqat P1g1l c ali
Lilet.Fem the letter.Fem Prep Ali.Masc gave.3Fem.3Fem
*habetto
gave.3Fem.3Masc
Lilat gave the letter to Ali.

(Jake 1980:75)

192

Multiple objects and grammatical alignment

Agreement, then, does not distinguish the theme object from the goal object.
We believe that object agreement in Tigre is best analysed as a test picking out
all objective (+o) functions, similar to the object tests discussed for Dolakha
Newar in the previous section.
Object marking in Tigre resembles Dolakha Newar in two respects. First,
marking on goal objects does not depend on definiteness or information structure role: goal objects must take the preposition P1g1l independently of whether
they are topical and/or definite. The goal object in (33) is indefinite and cannot
control agreement, but nevertheless must be marked with P1g1l:
(33) has
am
a
P1g1l/* P1ss1t k1ta
ab
kaba
a(*y
a)
Hasama.Masc Prep
woman book.Masc gave.3Masc(*3Fem)
Hasama gave a woman a book.

(Jake 1980:73)

In example (33), both objects are indefinite, and object agreement is impossible; nevertheless, the goal object must be casemarked. This is exactly as in
Dolakha Newar, where marking on the goal object of a give-type verb cannot
be omitted.
Second, under certain information structure conditions casemarking may appear twice in ditransitive constructions. Although in most cases there is only
one P1g1l in ditransitives (marking the goal), it is not impossible to have both
objects casemarked, as shown in example (34).
(34) a. wa @g@l @mmu
@g@l musa ams@at

and Prep his.mother Prep Moses brought.3Fem

And she brought the mother of Moses to him.

(Raz 1983:108)

b. @g@l lak@t
ab @g@l man ams@akahu

Prep the.book Prep who brought.it.2Sg

For whom did you bring the book?

(Raz 1983:83)

Interestingly, there is another test in Tigre that distinguishes the theme object
in a give-type construction from both the goal argument and the monotransitive object. According to Jake (1980), object clitics are possible with verbs
with two complements, a goal and a theme, though not with monotransitives:
the object clitic cross-references the theme object only when the verb agrees
with the goal object. Goal objects never control cliticisation.
habetto
t
a
(35) Lilat
P1g1l la P1n
as
la sac at
Lilet.Fem Prep the man.Masc the watch.Fem gave.3Fem.3Masc 3Fem
Lilat gave the man the watch.

(Jake 1980:77)

Other multitransitive constructions

193

(36) Lilat
la sac at
P1g1l la P1n
as
habett
a
Lilet.Fem the watch.Fem Prep the man.Masc gave.3Fem.3Fem
(*tu)
(*3Masc)
Lilat gave the watch to the man.

(Jake 1980:78)

Thus, the theme object of a give-type verb is uniquely distinguished by the

object cliticisation test, and the goal object is uniquely distinguished by obligatory marking with P1g1l.
The same possibilities of analysis are open to us for Tigre as for Dolakha
Newar. Discounting the cross-linguistically implausible Analysis 1 (represented for Dolakha Newar in 28), we are left with the two possibilities we
considered for Newar in the previous section, repeated here:
(37) Analysis 2, Tigre:
Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ

OBJ GOAL

Analysis 2 gains most plausibility from casemarking patterns: as in Newar,

obligatory marking is associated with OBJ GOAL , while marking for primary OBJ
depends on topicality. Object cliticisation is harder to characterise in a non-adhoc manner, however: object cliticisation is possible for a theme object only in
a ditransitive construction, in the presence of an OBJ GOAL .
(38) Analysis 3, Tigre:
Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ THEME

OBJ

Analysis 3 has the opposite problem: object cliticisation can be neatly characterised as applying to OBJ THEME , but casemarking must be characterised
as obligatory for goal OBJ, and dependent on topicality for theme OBJ and
OBJ THEME .
For Tigre, it may be sensible to consider a third possibility: that the two
object arguments of a give-type verb both correspond to different objectlike functions from the object of a monotransitive verb. This corresponds to
the rare tripartite alignment discussed by Haspelmath (2007) and Malchukov
et al. (2007):

194

Multiple objects and grammatical alignment

(39) Analysis 4 (tripartite), Tigre:

Monotransitive:

Ditransitive:

THEME

THEME

GOAL

OBJ

OBJ THEME

OBJ GOAL

On this analysis, both casemarking and object cliticisation are easy to characterise: object cliticisation is possible only for OBJ THEME , and object casemarking is obligatory only for OBJ GOAL . This may provide the most satisfactory
analysis of Tigre, though examination of more data may reveal additional patterns which would favor Analysis 2 or 3 over this one.
9.4.3 Applicatives: Upper Necaxa Totonac
Beck (2006, 2007, 2008) discusses the syntactic and pragmatic properties of
the applicative construction in Upper Necaxa Totonac,4 showing that applicitivisation promotes topical nonobject arguments to object, and that the result is
a multitransitive construction in which the basic object and the applied objects
share object properties.
The Totonac verb has two person agreement slots and one number agreement slot. The verb can agree with the subject only, the subject and the object,
or with two objects. Subject and object agreement morphemes constitute distinct sets, with the subject affixes reflecting the person and the number of the
subject, and the object affixes reflecting the person and the number of the object in the two distinct slots. The number of the object is indicated by zero in
the singular and -ka:- in the plural, however, only one of the objects can agree
for number, even if both are in the plural and agree for person.
Person agreement is obligatory with first and second person objects. If the
verbs takes two first and/or second person objects, both of them control agreement. There is no subject agreement in this case.
(40) wix kuch:lu kn-li:-lhtuku-ya:-n
cha:-tin hotni"
you knife 1Obj-Instr-stab-Impf-2Obj Cl-one drunk
You, knife, a drunk stabs me with you.

(Beck 2008:2)

With third person objects, agreement is optional. Since third person singular
object agreement is null, only third person plural objects are overtly marked
for agreement.
4 We thank David Beck for clarifying some aspects of his analysis and providing us with additional data in personal communication. Examples for which the source is not given come from
personal communication with David Beck.

Other multitransitive constructions

195

With monotransitive verbs, agreement with the third person plural object is
very frequent, but can be absent if the object is low in what Beck calls topicworthiness, but which we believe is better analysed as topicality. In (41), the
object is nontopical, and the verb does not agree with it, showing only subject
agreement:
(41) lhu:wa" ik-lak-chuku-ma:lh
kim-pahlhcha"
much 1SgSubj-Intensive-chop-Prog 1Poss-tomato
Im slicing a lot of tomatoes.
Competition for object agreement in Totonac arises if two (or more) objects are
third person: any of them can but does not have to control agreement. In (42),
the verb bears two applicative affixes, and takes three object-like arguments:
theme, instrumental and comitative. Beck (2008) shows that each of them can
control agreement, as shown in (42a), (42b), and (42c), respectively.
(42) a. pu:lak-kauj kin-la:xax
na-ik-ka:-ta:-li:-tanka:
Class-ten 1Poss-orange Fut-1SgSubj-PlObj-Com-Instr-fell
kin-ta:sa:kwa wama: hen-tin kin-mach:ta"
1Poss-peon this
Cl-one 1Poss-machete
My peon and I will cut down ten orange trees with this machete.
b. hen-tu:tun machi:ta na-ik-ka:-ta:-li:-tanka:
Class-three machete Fut-1SgSubj-PlObj-Com-Instr-fell
pu:lak-tin kiwi wama: chixku
Cl-one
tree this
man
With three machetes I and this man will cut down a tree.
c. na-ik-ka:-ta:-li:-tanka:
pu:lak-tin kiwi chixku-win
Fut-1SgSubj-PlObj-Com-Instr-fell Cl-one
tree man-Pl
kin-machi:t-kan
1Poss-machete-Pl.Poss
I and the men will cut down a tree with our machete. (Beck 2008)
As with monotransitives, object agreement patterns depend on information
structure factors. First, objects higher in animacy are more likely to control
agreement than inanimate objects, although this is only a preference. More
important is discourse saliency. When an inanimate object is discourse salient
(i.e., topical), it becomes a legitimate controller. This is shown by the following contrast, where both (43a) and (43b) exemplify the comitative applicative.

196

Multiple objects and grammatical alignment

(43) a. *chin-lh kin-puska:t na-ik-ka:-ta:-puzta

arrive-Perf 1Poss-wife Fut-1SgSubj-PlObj-Com-look.for
hen-tu: kin-machi:ta
Class-two 1Poss-machete
My wife is here, we are going to look for my two machetes.
b. akxni chin-lh
kin-puska:t na-ik-ka:-ta:-puzta
when arrive-Perf 1Poss-wife Fut-1SgSubj-PlObj-Com-look.for
hen-tu: kin-machi:ta
Class-two 1Poss-machete
[My machetes got lost.] When my wife comes she and I will look
for them.
(Beck 2008)
In (43a) number agreement with the inanimate theme object is impossible because it is not sufficiently salient. But agreement becomes acceptable if the
context is modified: in (43b) the context establishes the topical role for the
referent of machetes. The second sentence is about machetes, so the object
pronoun referring to machetes can control agreement. Similarly, agreement is
possible with contrastive topics:
(44) a. cha:-tin chixku asta hen-tu: kuchilu li:-lhtuku-lh hotni
Cl-one man
even Class-two knife Instr-stab-Perf drunk
The drunk stabbed a man with two knives.
b. asta hen-tu: kuchilu cha:-tin chixku ka:-li:-lhtuku-lh
even Class-two knife
Cl-one man
PlObj-Instr-stab-Perf
hotni
drunk
With two knives the drunk stabbed a man.

(Beck 2008)

In (44) agreement is triggered by the instrumental topical object.

Beck (2008) proposes that the rules of object agreement obey the hierarchy
in (45):
(45) 1, 2 > discourse-salient 3 > animate 3 > inanimate 3
We believe that this hierarchy is best reformulated as the following generalisation: agreement is obligatory with first and second person objects and with
topical third person objects. The correlation with animacy is only a tendency
and is independently motivated: as mentioned in Chapter 3, animate referents
are frequent topics. Nontopical third person objects do not trigger agreement.
Note that the semantic role of the object plays no role in the object agreement
pattern.

Other multitransitive constructions

197

For multitransitive constructions, Beck concludes that the grammatical function of object is not unique in Upper Necaxa Totonac: he suggests that in multitransitive clauses all objects are of a similar grammatical status because any
object can potentially control agreement. This implies that object agreement is
independent of grammatical role.
We believe that a more satisfactory analysis is available. Bresnan and Moshi
(1990) discuss object symmetries and asymmetries in Bantu, distinguishing
two types of languages. In the asymmetrical object type, only one of a verbs
complements can display primary object properties, while in the symmetrical
object type, any of several of the verbs complements is eligible for the primary object role. They show that the distinction can be formally modelled in a
very simple way in terms of intrinsic assignments of grammatical functions to
arrays of semantic roles: the asymmetrical object type disallows the intrinsic
association of an array of semantic roles with two unrestricted grammatical
functions, while symmetric object type languages do not have this restriction.
We can analyse Totonac as a symmetrical object language in their terms. Taking object agreement to be a primary object property, this means that any of
several arguments can assume the primary object role and control verb agreement, though only one argument at a time can do this, since in any particular
sentence there is only one primary object.5
This view comports well with our view that information structure can play
a crucial role in the mapping of semantic roles to grammatical functions. We
suggest that secondary agreement in Upper Necaxa Totonac is triggered by
the primary OBJ, which can correspond to a variety of semantic roles. In contrast, secondary objects OBJ do not trigger agreement. In monotransitive constructions, the choice between OBJ and OBJ is determined by the referential
status of the object (first or second person vs. third) and, for the third person object, topicality. Applicative operations produce clauses with up to five
objects, but if they are all third person, only one argument controls verb agreement: the agreeing object is the primary OBJ. Non-agreeing objects are semantically restricted and correspond to OBJ THEME , OBJ GOAL , OBJ BENEFACTIVE,
OBJ INSTRUMENTAL and so on. The semantic type of each secondary object is
indicated by applicative morphology on the verb, except for OBJ THEME , which
does not require morphological applicativisation. This analysis makes Upper
Necaxa Totonac essentially similar to Ostyak: in both languages, agreement is
defined by grammatical functions. Both languages impose a strict alignment
between grammatical functions and topicality: the primary object is topical,
while the secondary object is a nontopic. The difference between the two languages lies in the number of semantically restricted secondary objects in the
5 MacKay

and Trechsel (2008) argue that the closely related language Misantla Totonac is a
symmetrical language in the sense of Bresnan and Moshi (1990).

198

Multiple objects and grammatical alignment

inventory of grammatical functions: Ostyak only has OBJ THEME and expresses
other semantic roles by means of casemarked obliques, whereas Upper Necaxa
Totonac allows a number of semantically restricted objects with various functions and therefore allows for a variety of multi-object constructions. However, since there is tight alignment between marking and grammatical function,
doubling of grammatical marking of the kind observed in Tigre, Newar and
non-standard Hindi is impossible for Upper Necaxa Totonac.
Beck (2007) considers a potential object test that distinguishes the theme/basic
object of the verb from applied objects: the theme is the only object that can
be targeted by the object suppression voice, which blocks the syntactic expression of the object. Object suppression cannot target the applied object.
According to Beck (2007), object suppression is not a good test of primary
objecthood, since verbs in the object suppression voice can still agree with
applied objects. Under the assumption that object suppression makes the basic/theme object syntactically inactive, and given Bresnan and Moshis theory of symmetrical vs. asymmetrical object languages, it is not surprising that
verbs in the object suppression voice can show object agreement: when the
theme object is suppressed, the applied object can fill the primary OBJ role.
However, the existence of this test leads to a consideration of an alternative
analysis of the data: it is possible that object suppression does indeed target the primary, non-applied object, and that verb agreement in Totonac can
be controlled by any topical object argument, whether OBJ or OBJ . We have
not encountered other languages exhibiting this pattern, and it goes against
our claim that topical marking is associated with grammatical functions that
are high on the grammatical hierarchy (for objects, OBJ vs. OBJ ). However,
the possibility cannot be ruled out in principle; further exploration of object
patterns in Totonac may provide evidence indicating whether this alternative
analysis is viable.
9.5 Topicality and goals
We have seen in this chapter that topical patient/themes often share properties
in common with goals of give-type verbs; these properties may be limited to
marking, or may extend to grammatical role. The patient/theme object marker
is often identical to the marker of the goal argument in constructions with
give-type verbs, independent of grammatical function. For instance, in nonstandard Hindi and possibly in Dolakha Newar and Tigre, the theme and goal
correspond to different grammatical functions, but still bear the same grammatical marking. In languages where the secundative/double object construction
is possible, the similarities go further: in standard Hindi, Chatino and some
constructions in Ostyak the goal is not only encoded identically to the topical
theme object, but represents the same grammatical function.

Topicality and goals

199

This is a cross-linguistically widespread phenomenon and has often been

noted in previous work. Bossong (1991:157) observes that the dative marker
is a common (but not the only) source of innovated accusatives in languages
with DOM. Masica (1982) provides an overview of a number of languages
(predominantly in Asia) where object casemarking coincides with the marking of the goal argument in ditransitive constructions. Apart from the languages discussed in this book, similar patterns are observed in Romance and
Semitic. In Klamath (Penutian, Rude 1982) patient/theme objects are either
marked or unmarked for case, while for goal and beneficiary objects the same
marking is obligatory. Similar polyfunctionality is present in Lezgian (NakhDaghestanian) and Imonda (Papuan) (Heine and Kuteva 2002:38) and many
languages cited by Kittila (2006a,b). Languages with object agreement often
show the same pattern, e.g. Palauan (Austronesian), Huichol (Uto-Aztecan)
and a number of other languages discussed by Georgopoulos (1992).
Kittila (2006a) suggests that these marking patterns follow from the fact that,
on the one hand, DOM depends on animacy and, on the other hand, there is a
strong correlation between animacy and recipient (goal) (see also Haspelmath
2007). Recall that for Kittila, animacy is a cover term for topic-worthiness
or topicality, including definiteness and prominence as well as animacy. In
fact, ditransitive goals are overwhelmingly high in topic-worthy features, but
in our view animacy effects are subsidiary and follow from independent requirements on information structuring. In particular, we follow previous research in analysing goals in give-type constructions as unmarked secondary
topics.
Dryer (1986), Polinsky (1998) and Haspelmath (2004) claim that ditransitive goals are inherently more topical than theme arguments. According to
Polinsky (1998), the goal is associated with a pragmatic presupposition of independent existence at least prior to the event of transfer, and does not require
assertion in the sentence where it appears. In contrast, the existence of the
patient is not necessarily presupposed; this matches the prototypical properties of the theme of monotransitive verbs, whose existence is not necessarily
presupposed prior to the event in question.
Like primary topic, secondary topic is under discussion at the time of the utterance and is expected to carry a pragmatic presupposition of existence. This
semantic difference corresponds to an asymmetry in the relative informational
status of the two arguments: the goal is superior to the patient in topicality, and
other things being equal, the patient is more likely to be interpreted as focus
and the goal as secondary topic. In other words, secondary topic is the default information structure role for the goal in give-constructions. The same
is true of the causee argument, which prototypically carries a presupposition
of existence because it is indispensable for the development of the causative
event (Polinsky 1995).

200

Multiple objects and grammatical alignment

Since the ditransitive goal is a frequent secondary topic, it is likely to bear

secondary topic marking. We will see in Chapter 10 that in Iranian and IndoAryan DOM started as secondary topic marking on goals but later spread to
topical patient/theme objects, due to the fact that objects are more often topical
that other nonsubject grammatical functions. Thus, the uniformity of grammatical marking on topical monotransitive objects and ditransitive goals observed
in some languages has a historical explanation, even though, as we have seen
in this chapter, the grammatical functions associated with these arguments are
not necessarily the same.
9.6 Conclusion
We have seen that languages can express the goal and theme arguments of
give-type verbs in various ways, depending on the grammatical functions
which the language can deploy, the mapping rules relating grammatical functions to sematic roles, constraints on marking possibilities, and the availability
of double object constructions. These are, in principle, independent factors.
In languages of the type discussed in Chapter 8, grammatically marked
objects correspond to OBJ and are typically topical, while grammatically unmarked objects correspond to OBJ and are typically nontopical. In monotransitive constructions, the complement is realised as OBJ if topical, and OBJ if
nontopical. In languages without a ditransitive construction where the goal
argument of a give-type verb is expressed as an oblique, possibilities for expression of the theme argument are exactly as for monotransitives, since the
presence of an oblique goal does not constrain the marking or grammatical
function of the theme object.
In multitransitive constructions, when more than one nonsubject argument
is present, mapping possibilities may be restricted by the inventory of restricted/secondary objects in the language, in which case the correlation between marking and information structure is obscured. In Hindi and Chatino,
the goal argument of a ditransitive verb is the primary object (OBJ) and the
theme is the secondary object (OBJ THEME ); the theme must be unmarked and the
goal must be marked, independently of their information structure role. This
means that there is no DOM in ditransitive constructions: the theme and the
goal correspond to two different object functions and are associated with different grammatical marking. On the other hand, in Tigre and Dholaka Newar,
identical marking of the two objects is possible because grammatical marking
and grammatical functions are not in tight correspondence.
Data from all of the languages we have examined in this chapter reinforce
our basic claim: objecthood, and particularly primary objecthood, is inherently
associated with topicality.

10

Semantic features, topicality

and grammaticalisation
As discussed in Chapter 1, most previous work on DOM is based on semantic
features such as person, animacy, definiteness and specificity (Diesing 1992,
van Geenhoven 1998, Ritter and Rosen 2001). However, we have seen that
these features are not enough to explain object marking patterns in many languages, and we have argued that in many cases the appropriate generalisations
involve the notion of topicality, especially secondary topicality. The topicality
of a referent depends on how the speaker construes the situation within the
given communicative context; features of topic-worthiness determine only the
likelihood for the object to be construed as topical.
What our account still must explain is why in some languages DOM does in
fact depend on semantic features of topic-worthiness rather than the informationstructural notion of topicality, sometimes in combination with information
structure and sometimes independent of it. In this chapter we propose a historical explanation for these patterns, based on the notion of grammaticalisation.
We suggest that semantic patterns of DOM can arise as a result of different
stages and directions of grammaticalisation of topical marking on objects.
10.1 Case studies
We sketch the historical evidence for the emergence of DOM in Uralic, Persian and Hindi. Presenting a fuller diachronic picture will of course require a
much more detailed investigation; we view this chapter as the first step in this
direction.

201

202

Grammaticalisation

10.1.1 Uralic
In this section we examine grammaticalisation of topical marking in three genetically related Uralic languages, building on our discussion in Chapters 7 and
8 of grammatical marking via object agreement in Ostyak and Tundra Nenets,
and we outline a proposal for the historical development of object marking in
Uralic.
Ostyak is a member of the Ugric branch of Uralic. Tundra Nenets is a
member of the Samoyedic branch of Uralic, and is therefore distantly related
to Ostyak. Besides Ostyak (manifested in a number of dialects), the Ugric
branch includes Ostyaks closest linguistic relative Vogul (or Mansi); these
form the Ob-Ugric subgroup of Ugric, which also includes the more distantly
related language Hungarian. The Samoyedic group consists of the Northern
Samoyedic subgroup (Tundra Nenets, Forest Nenets, Tundra Enets, Forest
Enets and Nganasan) and the Southern Samoyedic subgroup, whose only living representative is Selkup.
(1)

Uralic

Ob-Ugric
Ostyak

Vogul

...

Samoyedic

Ugric
Hungarian

Northern Samoyedic

Southern Samoyedic

... Tundra Nenets, Nganasan ...

Selkup

Importantly, the Samoyedic and Ugric languages are geographically close and
demonstrate a number of common features (Xelimskij 1982). They are sometimes jointly referred to as Eastern Uralic languages.
Xelimskij (1982) and others have shown that object agreement in Ugric and
Samoyed has a common origin (although it was probably absent in Western
Uralic): at least some agreement markers go back to the same etymological source. It is also likely that conditions on agreement were identical at
some stage. However, conditions on agreement differ in the modern Ugric and
Samoyedic languages.
We suggest that the Ostyak system of DOM based solely on information
structure is the most archaic and can probably be hypothesised for ProtoEastern-Uralic that is, those Proto-Uralic dialects from which the Ugric
and Samoyedic languages developed. We discuss the situation in Samoyedic
below. For the Ob-Ugric branch, this claim is supported by the fact that apart
from Ostyak, the information structure-driven system is found in the related
language Vogul (Skribnik 2001). In this language, object agreement works
similarly to Ostyak. It does not directly depend on definiteness, as shown by
the examples in (2): in (2a) the object is definite, but there is no agreement,
and in (2b) the object is indefinite but agrees.

Case studies

203

(2) a. X
urum l
unt joXt-s-@t
os tawe kins-
e,-@t
three goose come-Past-3PlSubj and he.Acc search-Pres-3PlSubj
Three wild geese appeared and are looking for him.
b. j
uswoj-t t
an os m
an p
asi,-kwe-t
now-i-jan@l
eagle-Pl they and little reindeer-Dim-Pl catch-Pres-PlObj.3PlSubj
Eagles, they hunt little reindeer, too.

(Skribnik 2001:226)

Skribnik (2001) argues that agreement is governed by the topicality of the direct object. As in Ostyak, agreeing objects in Vogul can have a variety of
semantic roles. For instance, in (3) the object corresponds to a goal or beneficiary (the doll), which is clearly topical given the preceding context and
corresponds to a referential null.
(3) [The Mos-woman has something that looks like a small child: a doll.]
sa,i-l
w
ar-i-te,
w
ai-l
coat-Instr make-Pres-SgObj.3SgSubj shoe-Instr
w
ar-i-te
make-Pres-SgObj.3SgSubj
She is making a coat for it, she is making shoes for it.
(Skribnik 2001:229)
In fact, topicalisation by means of object agreement goes even further in this
language, because a wider variety of semantic roles than in Ostyak can be
realised as an agreeing object. Topicalised and nontopicalised location arguments are presented in (4), whereas in (5) we show topicalised and nontopicalised instruments.
(4) a. taw X
ap-@n joXt-@s
he boat-Dat come-Past.Subj.3Sg
He came to the boat.
b. taw Xap joXt-@s-te
he boat come-Past-Obj.Sg.Subj.3Sg
He reached the boat.
(5) a. am tul
owl-um-@l r
atasl-
e,-um
I finger-1Sg-Instr tap-Pres-1SgSubj
I am tapping with my finger.

(Skribnik 2001:229)

204

Grammaticalisation
b. am tul
owl-um r
atasl-i-l-um
I finger-1Sg tap-Pres-SgObj-1SgSubj
I am using my finger for tapping.

(Skribnik 2001:229)

In both cases a topical element may not appear as an oblique element, but must
be promoted to object (on our analysis, primary object) and must agree with
the verb. In Ostyak, the primary object can correspond to a patient/theme, goal,
causee or beneficiary, while in Vogul objects can also bear other semantic roles
such as location and instrument.
Thus, in both Ob-Ugric languages, Ostyak and Vogul, object agreement is
conditioned by information structure, and we hypothesise that this was also
true in Proto-Ob-Ugric. There are no known semantic restrictions on agreeing
objects in Ob-Ugric.1
The situation in the third Ugric language, Hungarian, is different. First and
second person object pronouns in Hungarian never trigger agreement; this is
similar to the Nenets pattern discussed in Chapter 7, Section 7.2:
(6) lat-nak
/ *lat-jak
teged / engem
see-3PlSubj see-Obj.3PlSubj you me
They see you/me.

Kiss 2002:54)
(E.

Unlike Nenets, however, third person agreement is not determined by information structure. Instead, the triggering feature is definiteness. Nouns preceded by the definite article a/az, possessed nouns, proper nouns, complement
clauses, third person personal pronouns, and a number of other types of pronouns, including reflexive, reciprocal, and some interrogative pronouns, all
count as definite objects and trigger object agreement. Indefinite nouns preceded by the indefinite article egy or quantifiers, as well as bare objects, do not
trigger agreement.
(7) a. Janos zoldre fest-ett-e
/ *fest-ett
a
J.
green paint-Past-Obj.3SgSubj paint-Past.3SgSubj Det
kapu-t
gate-Acc
Janos painted the gate green.

Kiss 2002:70)
(E.

b. Bicikli-t
sok lany lat-ott
/ *lat-ta
bicycle-Acc many girl see-Past.3SgSubj see-Past.Obj.3SgSubj
Bicycle, many girls saw.

Kiss 2002:22)
(E.

1 However, in some Eastern varieties of Ostyak the distribution of object agreement is similar
to Nenets: agreement is impossible for first and second person objects.

Case studies

205

Thus, object agreement in modern Hungarian does not depend on information

structure at all: topical and focused third person objects trigger agreement
if they are marked as definite, whereas first and second person objects never
do. For example, in (7b) the object is fronted and appears in the topic position
Kiss 2002), but since it lacks the definite article, object agreement is absent.
(E.
Xelimskij (1982:84-94) claims that the situation in Proto-Ob-Ugric was the
same as in Hungarian; however, no clear argumentation for this claim is provided. We suggest that the modern Hungarian system of definiteness marking
is in fact an innovation, which developed after the language acquired the grammatical category of definiteness and grammatical articles. On this view, earlier
Hungarian was closer to Ostyak and Vogul than modern Hungarian in this respect.
This claim is supported by the data from earlier stages of Hungarian. Marcantonio (1985) cites examples from the 15th-16th century Hungarian codex
literature, in which the use of object agreement deviates from the modern pattern. She shows that the verb could be marked for object agreement even if
a non-possessed object was not preceded by the definite article, which is impossible in modern Hungarian. In fact, the use of the definite pronoun az as a
definite article was not yet completely established at that stage (Barczi 1980).
(8) a. sebes szamszereg-et ragad-t-a
vala
quick lance-Acc
grasp-Past-Obj.3SgSubj be.3Sg
He gripped the quick lance.

(Marcantonio 1985:289)

b. a llatok mi-t
ur-unk Krisztus hamar meggyog-t-a
animals what-Acc lord-1Pl Christ quickly cure-Past-Obj.3SgSubj
the animals which our Lord Christ cured quickly.
(Marcantonio 1985:292)
Example (9) is taken from the oldest known Hungarian text, Hallotti Beszed
(end of the 12th century). Here, too, the object paradise does not take the
definite article but triggers agreement.
(9) Odu-tt-a
vola neki paradisum-ut
give-Past-Obj.3SgSubj be.3Sg him paradise-Acc
He gave him paradise.

(Marcantonio 1985:293)

According to Marcantonio (1985), the objects in (8) and (9) are likely to be
topical. For instance, she provides a discourse context for (8a) in which the
quick lance is the centre of discussion.
Conversely, some definite objects did not trigger agreement. In (10a) and
(10b), the object is a possessed noun and therefore definite, but object agreement is absent. Such examples suggest that topicality rather than definiteness

206

Grammaticalisation

was the triggering feature for agreement in early Hungarian: although the objects in (10a,b) are definite, they are unlikely to be topical because these objects
introduced novel discourse participants.
(10) a. an-nak szabadulas-a-t
e n szuv-em kvan
that-Dat liberation-3Sg-Acc I heart-1Sg desire.3SgSubj
My heart desires his liberation.

(Marcantonio 1985:290)

b. es tarsibeli kiralok es zigetek aiandok-a-i-t

aiandokoz-nak
and friendly kings and islanders present-3-Pl-Acc give-3PlSubj
And friendly kings and islanders give their presents.
(Marcantonio 1985:290)
Object complement clauses did not necessarily trigger agreement either. Marcantonio (1985) argues that such examples can be found even in more recent
Hungarian literature, especially in poetry. They show that the rules of object agreement in earlier Hungarian deviate from modern Hungarian and were
likely to be more dependent on information structure.
An information-structure-based system of object agreement is also present
in Samoyedic, as shown by the Tundra Nenets data presented in Chapter 7,
Section 7.2, although Tundra Nenets stands somewhat in-between Hungarian and Ob-Ugric: object agreement marks topical arguments, as in Ostyak,
but first and second person pronouns never agree, as in Hungarian. The other
Samoyedic languages, including Selkup and Nganasan, are similar to Nenets
in this respect. In these languages object agreement is only possible with 3rd
person objects, as shown by the Selkup examples in (11):
(11) a. T@p sinty
qontyrtEnta
/ *qontyrtEntyNyty
he you.Acc see.Fut.3SgSubj see.Fut.Obj.3SgSubj
He will see you.

(Kuznecova et al. 1982:235)

b. T@p kanap qontyrtEnta

/ qontyrtEntyNyty
he dog.Acc see.Fut.3SgSubj see.Fut.Obj.3SgSubj
He will see a/the dog.

(Kuznecova et al. 1982:235)

According to Kuznecova et al. (1982), object agreement marks definite third

person objects, although the examples they provide do not demonstrate clearly
whether we are dealing with definiteness or topicality. If it is indeed definiteness, Selkup is just like modern Hungarian, except that it does not have grammaticalised expression of definiteness in the form of articles. For Nganasan,
Tereshchenko (1979) explicitly says that object agreement with third person

Case studies

207

objects depends on logical emphasis, which is her terminology for information structuring, whereas definiteness does not play a decisive role. For instance, in both examples in (12) the object is definite (possessed); Kuznecova
et al. (1982) state that the object in (12a) is more salient than in (12b), and
agreement is required only in (12a).
(12) a. Bukurjakumti myt@mi@g@j
byZ@dja najb@g@
net.Acc.Sg.3Du pushed.Obj.3DuSubj water.Dat long
nirkut@nu
willow.branch.Instr
They (dual) pushed their net into water with long willow branches.
(Tereshchenko 1979:190)
b. Bukurjakumti najb@g@ nirkut@nu
byZ@dja
net.Acc.Sg.3Du long
willow.branch.Instr water.Dat
myt@mi@Zi
pushed.3DuSubj
They (dual) pushed their net into water with long willow branches.
(Tereshchenko 1979:191)
In Samoyedic and Old Hungarian, then, object agreement was restricted to
third person topical objects. In other words, the scope of grammatical marking
of topics was reduced. This situation contrasts with Ob-Ugric, where agreement marks topical objects independently of person.
Comrie (1977:10) suggests a functional explanation for the lack of agreement with first and second person controllers: first and second person pronouns
are inherently definite, so there is no need to mark them explicitly. This explanation is based on the premise that the primary function of object agreement
in Eastern Uralic is the marking of definiteness. However, in all Eastern Uralic
languages except modern Hungarian, information structure and not definiteness plays the primary role in patterns of object agreement. The Hungarian
situation thus is likely to be secondary, as is also confirmed by the Old Hungarian data.
An alternative explanation for first/second person agreement in Uralic is
as follows. The referents of the first and second person pronouns are highly
salient in human communication. In fact, they occupy the highest position on
scales of topic-worthiness, and are the most likely topics. However, the primary topic is more salient than the secondary topic by definition, so first and
second person pronouns are more likely to correspond to the primary topic
than the secondary topic. Given the default alignment between the primary
topic and the subject, first and second person pronouns tend to be encoded

208

Grammaticalisation

as subjects: in fact, in Ostyak they rarely occur as topical objects triggering

agreement. If they correspond to the patient/theme of a transitive verb, passivisation is preferred even in contexts which in principle require secondary
topicalisation of the object. For example, in the context Where did John hit
Peter? the answer (given in 13) will invariably be in the active, and the secondary topic object Peter will trigger object agreement. But in the context
Where did John hit you? there is room for variation: some speakers prefer to
use the active construction with object agreement (example 14a), while other
speakers employ passivisation (example 14b).
(13) Juwan Pe:tra xo:t-na
re:sk@-s-li
John Peter house-Loc hit-Past-Obj.3SgSubj
John hit Peter in the house.
(14) a. Juwan mane:m xo:t-na
re:sk@-s-li
John I.Acc house-Loc hit-Past-Obj.3SgSubj
John hit me in the house.
b. (ma) Juwan-na xo:t-na
re:sk@-s-aj-@m
I
John-Loc house-Loc hit-Past-Pas-1SgSubj
I was hit by John in the house.
Presumably the choice depends on the individual assessment of the saliency of
relevant referents, but the point is that there is an asymmetry between (13) and
(14).
On this view, the Samoyedic languages (Nenets, Selkup and Nganasan) and
Old Hungarian have grammaticalised the tendency for first and second person
pronouns to be likely primary topics and unlikely secondary topics.2 Therefore
they cannot correspond to the primary object, which is strongly aligned with
secondary topic in these languages. There are no such restrictions for third
person objects. The next historical stage is represented by Hungarian (and
possibly Selkup): grammatical marking of third person topical objects extends
to nontopical definite objects. This process illustrates the spreading of grammatical marking to nontopical objects bearing topic-worthy features, and the
concomitant loss of a connection to information structure.
2 This tendency has been discussed elsewhere in the literature. For instance, Schulz (2005)
argues, based on facts from German pro-drop, that unmarked topical objects are third person,
not first or second person. There is no such asymmetry for subject topics. Haspelmath (2004)
cites frequency data indicating that first and second person themes are less common and less
prototypical than third person themes, both in monotransitive and ditransitive constructions, and
that in some languages they cannot map as objects at all.

Case studies

209

10.1.2 Persian and the Iranian languages

DOM exists to some degree in almost all Iranian languages except for Pashto,
where the object is never marked, and Kurdish, where it is always marked by
the accusative case. Conditions on DOM differ across languages. Moreover,
according to Bossong (1985), there are several etymological sources of DOM
in Iranian. In this section we summarise the history of Persian ra based on
the work of Bossong (1985:58ff.), Karimi (1990), Dabir-Moghaddam (1992),
Haig (2008), and Paul (2003).
Persian ra has cognates in Balochi, Gazi, Tat, Luri and a number of other
Iranian languages. Phonologically, this item has undergone a reduction from
the Old Persian radiy (6th century BC to 3rd century BC) to Middle Persian
ray (ca. 225 AD to 651 AD) and New Persian ra. Semantically and functionally, it has undergone a change of status from a noun to a postposition with
mixed semantic and syntactic functions, then to a (secondary) topic marker
restricted to certain grammatical functions, and then established as an object
marker several centuries after the loss of the old Proto-Iranian oblique case
used for direct objects.
The ancestor of ra was a noun meaning something like reason, aim. It was
attested with the postpositional meaning for the sake of, on account of, by,
due to, because of in Old Persian. Later, in Middle Persian, this postposition
came to be used as a marker of the beneficiary, purpose, goal, external possessor and a few other oblique functions. We can see the reflexes of this situation
in modern Persian as well. As was shown in Chapter 6, Section 6.1.1, ra can
mark the external possessor in modern Persian. However, not all of its earlier functions are preserved. For instance, Bossong (1985) cites the following
Middle Persian sentence:
(15) ma-ra dar sahr dust-an besyar-and
I-RA in town friend-Pl many-3Pl
I have many friends in town.

(Bossong 1985:61)

In this example ra marks the predicative external possessor, but modern Persian employs the verb have in predicative possessive constructions.
In the Early New Persian period, also called Classical Persian (around 1000
AD), ra was mainly used on indirect objects. Its use on direct objects appears
to be an innovation, though sporadic cases of ra on definite direct objects were
attested in the 10th century. In the Classical Persian example below ra is rendered orthographically as r.
(16) k=s
yn ddtn r bgft
when-3Sg this story RA tell.Past.Irr
when he told this story

(Haig 2008:128)

210

Grammaticalisation

However, most objects remained unmarked even if they were highly definite:
(17) xosro tus be du dad
king Tus to him gave
The king gave Tus to him.

(Karimi 1990:110)

The equivalent of (17) would be ungrammatical in standard Modern Persian,

although it may be acceptable in contemporary dialects. In late New Persian,
ra completely lost its indirect object function and came to be an obligatory
marker for definite direct objects and some non-object topical elements (see
Chapter 6, Section 6.1.1), though its earlier oblique function is still preserved
in some lexical items such as why?.
We hypothesise that ra on objects originates as a topicality marking device. As discussed in Chapter 9, beneficiary and goal arguments are unmarked
secondary topics, and external possessors are also highly topical (Payne and
Barshi 1999). Before it began appearing on objects, ra was systematically
used in these two functions. Because of the close association of these functions with topicality, ra is likely to have been reanalysed as a topicality marker
instead of a marker of a grammatical function. It subsequently began appearing
on topical objects, giving rise to DOM.
Interestingly, ra was first attested on objects high in topic-worthiness. Haig
(2008:152-153) argues that as an object marker ra first appeared on first and
second person pronouns (as is still observed in some related Iranian languages),
and was then extended to lexical nouns. According to Paul (2003:182), in the
Early New Persian period it was mostly found on animate definite nouns, while
inanimate definites remained unmarked. So although the syntactic functions of
ra were reduced from the Old Persian period, its use as an object marker can
be characterised as a continuous expansion: it first appeared on first and second person pronouns, then spread to animate definite objects, then all definite
objects, but was still optional. Later, in modern Persian, it became obligatory
on all definite objects and topical indefinite objects.
10.1.3 Hindi and the Indo-Aryan languages
The Indo-Aryan languages are distantly related to the Iranian languages, together constituting the Indo-Iranian branch of Indo-European. The Indo-Aryan
languages, including Hindi, are descended from Sanskrit; the Old Indo-Aryan
period, encompassing Vedic and Classical Sanskrit, extended from 1500 BC to
600 BC, and was followed by Middle Indo-Aryan, which extended from 600
BC to 1000 AD (Masica 1991). Middle Indo-Aryan lost almost all case inflections which had been present in Sanskrit, including the original accusative

Case studies

211

which merged with the nominative, but the new Indo-Aryan languages developed new postpositional or clitic-like markers of major grammatical functions.
These typically originated from lexical nouns and verbs.
According to Beames (1966 [1872-79]), Hindi ko goes back to the Sanskrit noun kaksha armpit, side. As argued by Butt (2008a) and Butt et al.
(2008), the original function of this element was purely locational: the word
meaning armpit grammaticalised as a spatial postposition. Indeed, in related
Iranian languages its cognate denotes location. For instance, the Iranian language Pashto has a locative in kii/ke, which goes back to Avestan kaase. The
latter is the locative form of kaasa armpit, etymologically related to Sanskrit
kaksha (Hewson and Bubenik 2006:150).
The early Hindi forms of this postposition, kaham, kum, kaum, kau etc., all

further phonological
reflect the Old Hindi accusative form k
akha. Following
changes, it developed into Hindi ko, Bengali ke, Oriya ku, Singhi khe and Siraiki kon (Masica 1982). In spite of their common origin, the exact conditions
on the object-marking use of this element differ in modern Indo-Aryan languages. For instance, the range of functions of the object marker in Bengali is
narrower than in Hindi: Bengali ke is mostly used on animate objects, and is
not compatible with pronouns referring to things.
The oldest documented examples of ko in Hindi come from the early 13th
century (Beames 1966 [1872-79], Kellogg 1955 [1893]). Beames examines
the early occurrences of this postposition, which at this stage signalled the
purposive and the recipient or goal of ditransitive verbs like give. There
are also a few examples where ko marks objects. In most cases, such objects
can be interpreted as unattained goals/endpoints, for instance with the verb
seek. Such examples are not numerous, however. According to Butt (2008a),
this usage continues over centuries, with a handful of locative/possessive uses
documented in the 1500s. The following example represents Divan-e-Hasan
Shauqi (Deccani Urdu of 1564).
kAdh zArA Uni
pet ko
(18) na bandh iya
Obj
not tie.Perf.MascSg ever armour Pron.3.Gen belly
He never tied armour to/on his belly.

(Butt 2008a:16)

In the 1800s, ko appears with verbs of directed motion such as go or reach.

(19) s mAnzl-ko
kAb poaco-ge
that destination-Obj when reach-2Sg-Fut
When will you reach that destination?

(Butt 2008a:17)

It is also well known that ko marks experiencer subjects and the subjects of
modal verbs expressing obligation (Mohanan 1994, Butt 2008a), although this
usage seems to be fairly recent.

212

Grammaticalisation

Butt (2008a) presents the following hypothesis for the development of the
functions of ko in Hindi/Urdu. After this element was drawn into the system of
spatial postpositions, its development went in two directions. On one hand, it
developed a dynamic meaning denoting the endpoint of a situation (to) and,
on the other hand, it acquired a stative meaning (at). The next step was a
metaphorical extension by which the spatial concepts were reanalysed in the
domain of events and participants. The dynamic final point interpretation of ko
was extended to mark participants as being associated with the final part of an
event, in particular, recipient/beneficiary arguments of ditransitive verbs. Such
arguments are often thought of as abstract goals. Animate locations (at) were
reanalysed as subjects (Butt et al. 2006).
The next question is how the spatial/goal ko became an object marker. This
development is not immediately obvious (cf. Ahmed 2006). A number of authors relate the accusative usage of ko to its spatial meaning through a type
of metaphoric extension. For instance, Mohanan (1994) argues that accusative
marking in Hindi is used for logical objects towards which an action or event
is directed. That is, it can be seen to mark the endpoint or the goal of a (bound)
action. According to Butt (2008a), in the modern language (the Urdu variety of
Hindi/Urdu) ko marks specific objects. This meaning derives from its function
to express (not necessarily attained) endpoints that are abstract but specific.
Roughly speaking, then, the objective marking has its roots in spatial marking,
and objects are reanalysed endpoints.
We do not disagree with the essence of this explanation, but believe that
additional factors may play a role.
Importantly, Hindi is not alone. Heine and Kuteva (2002) show in their substantial survey of grammaticalisation paths that allative and dative are the two
most common sources of object marking, and we demonstrated in previous
chapters that in many other languages the goal argument of a verb like give is
grammatically marked in the same way as the transitive patient/theme. However, not all these markers have a spatial origin. As we saw in the previous
subsection, the Persian object marker developed from the marker for the ditransitive goal, but the latter does not go back to a locative noun (its source
is a noun meaning something like reason). Deo (2008) analyses the origin of the accusative-dative -l
a in Modern Marathi, tracing it back to the Old
, which expressed purpose or benefaction. This postMarathi adposition l
agim

position is attested in Middle

Marathi poetry, but always with a purely benefactive/purposive use. It did not indicate spatial proximity or control over the
situation. In the late nineteenth and early twentieth century it emerged as the
default dative case marker in the language, and in modern Marathi it is used as
the differential object marker. Interestingly, this object marker replaced the already existing accusative marker -si: that is, the dative marker -l
a replaced the
existing device for accusative marking rather than filling a gap. Thus, evidence

Case studies

213

from Marathi seems to suggest that dative marking tends to be diachronically

extended to accusative contexts even when it does not have a spatial origin,
and despite the presence of a morphologically distinct accusative. Even more
striking is that accusative-dative homophony exists in languages where it is
marked by agreement, which can hardly be thought of as having a spatial origin. Rather, we claim that it originates from the need to keep track of referents
and often cross-references topics (see Section 10.2.1 of this chapter). In sum,
we believe that there must be more to the accusative-dative connection than the
idea that objects are endpoints reanalysed from the spatial domain to a more
abstract domain. Also recall that, on our account, specificity as such does not
explain the distribution of objective ko in modern Hindi (the situation may be
different in modern Urdu). As we argued in Chapter 8, nonspecific objects are
unmarked, but specific inanimates are either marked or not, depending on topicality. This implies that the explanation for the accusative-dative homophony
may have to do with information structuring.
We believe that when ko started to function as the regular marker of direct
objects which could co-occur with any transitive verb, it first marked topical
objects. According to Masica (1982), ko became established as a general object marker fairly late, appearing only once in the early prose of Braj Bhasha,
the literary form of Medieval Hindi (17th century). At that stage it was mostly
found on pronominal objects, although it is difficult to judge how grammaticalised it was in this function. Non-human objects were very rarely marked,
and even definite human patient/theme objects could remain unmarked, as
shown by the following examples from medieval Hindi:
(20) a. harase Lakhana dekhi dou bhr
at
a
Lakshman see two brothers
rejoice
Lakshman rejoiced to see his two brothers.

(Masica 1982:44)

b. ma-ta Bharatu goda baitha-re

mother Bharatu lap seated

The mother seated Bharatu in her lap.

(Masica 1982:19)

Specific human objects are not obligatorily marked with ko even in later Hindi,
including the twentieth-century examples in (21) and (22):
(21) a. [Context: I am a sinner before God, Professor Saheb!]
b. ma-ne baccaa badal diyaa
I-Erg children switch give.Past
I switched children.

(Masica 1982:20)

214

Grammaticalisation

(22) brit.en ke mantriman.d.al ne apne tiin pratinidhi

bhaarat bheje
Britain Poss cabinet
Erg its three representatives India send
the
Past
Britains Cabinet had sent its three representatives to India.
(Masica 1982:19)
Example (21) is from a novel by Rangeya Radhava, written in 1961, and example (22) is from a text by Yash Pal, written in 1963. The objects in (21) and
(22) are likely to inhabit the focus domain and be nontopical. Similar examples
are cited by Kellogg (1955 [1893]) and McGregor (1972).
In sum, evidence indicates that earlier varieties of Hindi may have patterned
differently from the modern language as far as object marking is concerned.
The distribution of ko as an object marker was much more restricted: it was not
required on all specific human objects, as it is in the modern language (McGregor 1972:185). We suspect that, as in Persian, the Hindi object marker originated as a marker of secondary topicality on ditransitive goals (prototypical
secondary topics). Later, it started to be used to indicate topical patient/theme
arguments, being mostly used for highly likely topics (personal pronouns).
From this function it expanded to mark specific human objects independently
of their information structure function, reflecting the fact that topics tend to be
human/animate. As in Persian, then, the role of the referential semantic features of the object has increased, and the connection to information structure
has weakened.
10.2 Paths of grammaticalisation
Based on the evidence discussed in the previous section, we argue that DOM
first emerges as an information structure marking device, at least in some languages. Only later does it come to depend partially or completely on referential
features of the object. In its essence this idea is similar to Danons (2006:1005)
suggestion that DOM might initially arise out of functional factors, and later,
as grammaticalisation proceeds, become syntactically governed. The difference is that Danon includes referential semantic features together with pragmatic and cognitive considerations in the list of possible functional factors,
while on our account functional factors are limited to information structure
alone.
10.2.1 Spreading of DOM
One possible direction of change involves widening of grammatical marking,
where topical marking of objects extends to certain nontopical objects. This
process can be schematically represented as follows:

Paths of grammaticalisation
(23)

topical

nontopical

topical

215

nontopical

>
marked

unmarked

marked

marked

unmarked

That predicate-argument agreement often originates as a topic-marking device

is well known. The diachronic connection between free-standing pronominal
topics, incorporated pronouns or clitics, and agreement affixes has been much
discussed since at least Givon (1979). For instance, Bresnan and Mchombo
(1987:777) argue that object marking in Bantu emerged as anaphoric topic
marking and that this function is still active in Chichewa,
as we saw in Chapter 2. In Chichewa,
object markers are incorporated pronouns referring to topics. In other Bantu languages, object markers have partially or totally lost their
pronominal reference and have undergone grammaticalisation as agreement
morphemes; in some Bantu languages, agreement applies unexceptionally either to all objects or to certain semantic classes of objects, independently of
their topicality.
Morimoto (2002) presents a historical view of object agreement marking in
Bantu that is in many respects compatible with our view: object agreement
originates as a topic-marking device, and later comes to mark certain semantic
features associated with topicality, such as definiteness. When incorporated
pronouns develop into full-fledged grammatical agreement morphemes, they
become obligatory for all objects or for a subset of objects bearing relevant
semantic features. Topic-anaphoricity may still be visible but, given that linguistic change is gradual, it is observed to different degrees in different Bantu
languages. In addition, there is a great degree of inter-speaker variation in most
languages. Such variation is expected, since synchronic variability is a sign of
an unstable situation in a diachronic process of transition from pronominal incorporation to grammatical agreement. Morimotos account differs from ours
in assuming that agreement is optional in some cases, in line with much OTbased work, since objects with the same semantic features either do or do not
trigger agreement; she treats this optionality in terms of constraint reranking
in an OT setting, along the lines discussed in Chapter 1, Section 1.3. On this
view, the optionality of DOM is explained by the fact that transitional stages
in the diachronic process exhibit different properties both across the family
and within individual languages. In our account, real optionality never arises:
conditions on agreement are not formulated in terms of referential semantic
features of the object. Rather, they are defined in terms of the information
structure role of the object (sometimes in combination with semantics) or its
grammatical function, but in either case there is no variation. For instance,
Ostyak objects with the same semantic features may but need not agree. However, agreement is not optional in any syntactic sense: it is triggered by primary

216

Grammaticalisation

objects aligned with the information structure role of topic, while secondary
objects aligned with nontopics do not show agreement.
Nevertheless, we agree with Morimoto that patterns of DOM in different
languages reflect different stages of grammaticalisation, and that agreement
arises as an indicator of topicality, and only later comes to depend on referential semantics. Her analysis and ours agree on the direction of the grammaticalisation process: topical marking extends to nontopical objects with particular
semantic features. In Section 10.1.1 of this chapter, we saw that Old Hungarian agreement was triggered by third person topical objects, and was later
reanalysed as definiteness marking and extended to definite third person objects, whether they are topical or not. At this stage of grammaticalisation the
connection to information structure was totally lost, so that object marking became dependent on semantic features alone: objects with the relevant semantic
features are obligatorily marked. Grammatical agreement in some Bantu languages appears to have developed in the same way.
We also saw that casemarking can work in a similar manner, as was already
noticed by Bossong (1991). For instance, grammatical marking of Hindi topical objects spread onto animate specific objects. In Persian, DOM originated
as a marker of topical objects denoting speech act participants, but later spread
onto all definite objects, including nontopical ones, and even some indefinite
objects.
This spreading scenario is similar to what Mithun (1991) has proposed for
active/agentive casemarking patterns. Such patterns have a semantic basis, but
the initial semantic motivation can be obscured by processes of grammaticalisation. In Caddo (Caddoan), case marking of the first argument of an intransitive verb is dependent on the notion of control: arguments that are in control
of the event are classified as grammatical agents and receive agent marking,
while arguments that are not in control are classified as grammatical patients
and receive patient marking. However, verbs with the causative suffix automatically appear with the agent case, regardless of the degree of control involved in the situation. The reason is that in most cases the causative situation
does presuppose that the agent (the causer) can control the process. The agent
marking on causative agents starts as a tendency reflecting the frequent association between causation and control, but later generalises to the whole class
of causative verbs.
Mithun (1991) further mentions that expansion may even be restricted to
individual lexical items. For example, the verb lose in Caddo behaves irregularly in the sense that its first argument is always encoded as a grammatical
agent. In a similar manner, in languages with DOM the marking can become
intrinsically connected with certain lexical items. For instance, question words
can behave differently from other objects with respect to DOM if they bear
features such as animacy. In Hebrew, where DOM is generally conditioned

Paths of grammaticalisation

217

by definiteness, the question word who? in the object role obligatorily bears
the object marker (Aissen 2003b:453). Browne (1970) and Karimi (1990:154)
note that the grammatical marker ra in Persian must co-occur with the object
question words ki who and is optional with ci what as well as objects modified by the interrogative kodum which. These elements are clearly in focus,
and cannot be analysed as topical. We analyse the spreading of ra-marking to
animate/human ki who as generalisation of grammatical marking to nontopical elements. Since topical arguments are likely to be animate, the inherently
animate who in the object function has acquired grammatical marking, even
though it is nontopical. For ci what and NPs with kodum which, ra-marking
appears if the NP refers to a specific entity. For example, (24a) can only be
interpreted as a general question, while (24b) presupposes a choice between
several known objects.
(24) a. ce xord-i?
what eat-2Sg
What did you eat?
xord-i?
b. ce-ra
what-RA eat-2Sg
What did you eat?
These examples exhibit spreading of topical marking onto a specific nontopical
object.
The examples discussed above clearly demonstrate that the grammaticalisation of object marking involves the familiar features of topic-worthiness:
casemarking or agreement spreads onto nontopical objects that show topicworthy features of definiteness, animacy and/or specificity. The spreading of
grammatical marking to nontopics with the semantic features typical of topics is what Harris and Campbell (1995) would probably call extension: a
change in syntax that requires generalising a rule. As they note (Harris and
Campbell 1995:101), observed extensions generalise to a natural class based
on categories already relevant to the sphere in which the rule applied before it
was extended. The causal mechanism for this type of historical development
is frequency (see Haspelmath 2004 and references therein on the general relevance of frequency for grammaticalisation). Topic-worthy objects are most
frequent topics and therefore are most often marked at the first stage of grammaticalisation. Subsequently, the frequent association of marking with particular classes of objects changes from being preferred to being obligatory. Similar
reasoning applies to individual lexical items. To account for such processes,
Haspelmath suggests the Frequency Condition on Grammaticalization:

218

Grammaticalisation
The more frequent a candidate for grammaticalisation is relative
to other competing candidates, the more likely it is that grammaticalisation will take place.
(Haspelmath 2004)

According to this condition, patterns which are more likely to be produced by

speakers are therefore more likely to be entrenched and automatised, which
ultimately leads to grammaticalisation.
10.2.2 Narrowing of DOM
In some languages only a subset of topical objects is formally marked, while
nontopics must remain unmarked. We believe this is due to the historical process of narrowing, by which the marking becomes available only to some
topics. This process is opposite to spreading, because it involves the retraction
of grammatical marking. It can be schematically represented as follows:
(25)

topical

nontopical

topical

nontopical

>
marked

unmarked

marked

unmarked

unmarked

As was shown above, in Old Hungarian and Samoyedic the original topicalitybased patterns of object agreement, still found in Ob-Ugric languages, were replaced by a system where the marking of topicality was significantly reduced
and restricted to third person topical objects only. We suggested that this may
be due to the frequent association of secondary topicality with the third person.
This process illustrates the narrowing of topical marking to a subset of topics,
while all nontopical objects remain unmarked.
Narrowing processes are also observed in casemarking languages with DOM.
As is well known, a number of Romance languages have differential object
casemarking involving the preposition a (Bossong 1985 and others). As in
Indo-Aryan and Iranian, it originates as an indirect object marker, but in later
stages marks discourse-prominent direct objects. Further development differs
across Romance languages.3 A relevant example of narrowing is presented by
Catalan. In Old Catalan (the period from the Middle Ages to the 19th century) a was attested in more contexts than those accepted by current standards:
it was found on pronominals, proper nouns and human objects, although in
all of these functions it was optional and correlated with topicality. This is
3 Aissen (2003b) and Escandell-Vidal (2009) show that topicality-based DOM was present in
Medieval Spanish. The patterns of DOM in modern Spanish have been extensively discussed in the
literature, but are rather elusive, and contradictory evidence is cited in available sources; moreover,
conditions on marking differ across the dialects of Spanish. We will not attempt to account for
DOM in Spanish, though see Mayer (2008) for some discussion.

Paths of grammaticalisation

219

still observed in some varieties of the language: in Balearic Catalan casemarking appears on pronominal objects and (a subset of) lexical topical objects
(Escandell-Vidal 2009). But in other dialects of modern Catalan, object marking became more restricted. In Central Catalan, the colloquial variety spoken
in Barcelona and adjacent areas, DOM is based on humanness/animacy (Nss
2004). In modern Standard Catalan (the literary variety created by the normalisation process that took place at the beginning of the 20th century) only
pronominal objects take the preposition a, and in this function it is obligatory
(Escandell-Vidal 2009, Aissen 2003a,b). This is independent of information
structure requirements, as the referential status of the object unambiguously
determines its marking. We can conclude, then, that these varieties of Catalan
demonstrate regression of object marking. At an early stage, a marked topical objects. But in Central Catalan and modern Standard Catalan casemarking
was narrowed to human and pronominal objects, respectively, and finally lost
its connection to information structure.
These examples show that, like spreading, narrowing usually involves the
most typical members of the set. Since objects ranked high on prominence
hierarchies are frequent topics, grammatical marking can become restricted
to them. The narrowing scenario also demonstrates the non-arbitrary relation between grammaticalisation and frequency: in the words of Du Bois
(1987), grammars code best what speakers do most. Both spreading and
narrowing are based on the idea that topicality frequently correlates with topicworthiness. The difference is that in the spreading scenario topical marking
expands onto objects with frequent features of topics, while in the narrowing
scenario it is retracted from topical objects with infrequent features of topics.
In both cases, however, the information structure based system of DOM is lost
(partially or completely) and the role of referential semantics increases.
Like spreading, the retraction of topical marking may apply in syntactic contexts determined by individual lexical items. For instance, Mahootian (1997)
notes that the objects of certain Persian verbs never take ra, even if they are
very high on the Prominence Scales. The object of the complex predicate
dombal-e gstnlook for must remain unmarked even if it is animate and
definite.
(26) dombal-e madr-m / *madr-m-o mi-grd-m
after-Ez mother-1Sg mother-1Sg- RA Dur-turn-1Sg
Im looking for my mother.

(Mahootian 1997:199)

This may be due to the fact that the verb look for often requires an indefinite
nontopical object - at least in the meaning look for something new. Therefore it frequently cooccurs with unmarked objects. Then, by analogy, topical
marking is retracted from all objects of this verb, even when they are definite.

220

Grammaticalisation

Note that the standard markedness account also predicts that variations in the
cross-linguistic patterns of DOM have to do with the extent to which prominence features are relevant. In Aissens (2003b) Optimality-Theoretic analysis,
historical changes are assumed to occur due to re-ranking of constraints. Differences in constraint ranking in various historical periods are characterised as
the demotion or promotion of the economy constraint *STRUCC with respect
to markedness constraints. But independently of the direction of this historical change, if an object at some rank of the prominence hierarchy is formally
marked, then higher ranked objects are also marked. In Aissens own words,
the generalisation that more prominent direct objects are always more likely
to be casemarked than ones of lower prominence should hold at all stages
(Aissen 2003b:471).
However, we have seen some patterns which would be difficult to account
for using Prominence Scales. In Tundra Nenets, Nganasan and Selkup, object
agreement is only observed with third person topics, while first and second person objects never trigger agreement even if they are topical. As demonstrated
in Section 10.1.1 of this chapter, Hungarian shows further historical development, but again only third person objects participate in agreement. Yet first
and second person pronouns outrank third person NPs on the definiteness hierarchy (Silverstein 1976, Aissen 1999), so they would be expected to receive
more marking than the latter. There are other instances where object marking
is restricted to the lower segments on the scales. For instance, Nganasan lexical
objects take the accusative case while personal pronouns lack it (Tereshchenko
1979). The verb in Waris (Trans-New Guinea) agrees with animate objects if
they are nouns or third person pronouns, but does not agree with first and second person pronominal objects (Brown 1988). Siewierska (2004:150) identifies a number of other languages (including Sursunga (Oceanic), Nanggu
(Papuan), Waura and Parecis (Carib), and more) where object agreement targets only third person objects, while first and second person objects do not
agree.
Such cases have been referred to as inverse differential case marking
(Jager 2003:253). The predictions that Aissens (2003b) analysis makes are
not borne out in these languages, and they are generally problematic for any
markedness account, as noted by Bickel (2008:204). Optimality-style analyses based on Prominence Scales cannot provide a systematic explanation for
inverse differential casemarking and, indeed, any idiosyncratic facts.
In contrast, we maintain that the direction of change cannot be predicted
with certainty. Our approach is based on the idea that linguistic constructions result from individual historical processes conditioned by various factors which may be in conflict, and need not conform to markedness principles. Cross-linguistic patterns of DOM arise through the interaction of general
grammaticalisation tendencies and language-particular constraints on individ-

Towards a typology

221

ual constructions, as is consistent with the theory of grammatical archetypes

proposed in Ackerman and Webelhuth (1998). These constraints may have different diachronic sources of a phonological or morphological nature, and these
give rise to a certain amount of non-predictable variation.
For instance, we have argued that Samoyedic and Hungarian languages have
grammaticalised the tendency for the first and second person topics to be preferred primary topics rather than secondary topics and therefore not to be expressed as (agreeing) objects. The tendency has grammatical consequences
only in some languages: other languages grammaticalise the tendency for first
and second person elements, as opposed to non-speech act participants, to be
highly topical. The explanation for the Nganasan casemarking restricted to
lexical objects is different. Filimonova (2005), citing personal communication
with Eugene Helimsky, states that in older varieties of Nganasan object pronouns were accusatively marked, and that the loss of the accusative is due to
the analogical levelling of the case paradigm. It is not clear how a markedness
analysis would account for these differences.
10.3 Towards a typology
The diachronic scenarios outlined in the previous sections give us the following
three types of languages with DOM:
Type 1 Languages where DOM is regulated solely by information structure;
correlations with semantic features are only tendencies (no spreading or
narrowing).
Type 2 Languages where DOM is regulated solely by semantic features; correlations with information structure are only tendencies (loss of connection to information structure role via narrowing or spreading).
Type 3 Languages where DOM is regulated both by information structure and
semantics:4
(a) Languages where DOM applies to topical objects and nontopical
objects with certain semantic features (spreading to arguments with
topic-worthy features, while retaining connection to information
structure role).
(b) Languages where DOM applies to topical objects only if they have
certain semantic features (narrowing to arguments with topic-worthy
features bearing the appropriate information structure role).
4 Presumably this is what is implied in the following quotation from Aissen (2003b:460461):
even in zones where DOM is optional, it is often the case that the probability of casemarking
varies depending on the particular properties of the object.

222

Grammaticalisation

As we have seen, analyses that depend purely on semantic factors cannot fully
account for the patterns of DOM in languages of the first and third type.
The first type is represented by Ostyak and Vogul. There are no semantic
restrictions on DOM in these languages: all semantic types of objects can trigger secondary agreement. The only exception is provided by nonreferential
objects, but this follows from an independent requirement for topics to be referential. Such languages exhibit what de Hoop and Malchukov (2007) calls
the fluid type of differential marking. In the fluid type the same NP can receive
alternative encoding depending on the context, with a concomitant pragmatic
difference.
In the second type, which de Hoop and Malchukov (2007) call split, different classes of NPs induce different marking, so that the marking patterns
depend entirely on inherent properties of the objects. This type is exemplified
by modern Hungarian. Other examples include Hebrew accusative marking on
definite objects (Aissen 2003b), object agreement restricted to first person singular objects in Imbabura Quechua (Quechuan) (Cole 1982) and casemarking
on first and second person objects in Yidiny (Pama-Nyungan) (Comrie 1979).
More complicated patterns involving multiple features are also found. For example, Palauan object agreement appears (in the perfective aspect) if the object
is either human or singular specific (Woolford 2000). In Komi-Zyrjan (Uralic),
accusative case occurs either on animate or on definite objects (definiteness is
marked by the third person possessive affix), while inanimate indefinite objects
remain in the nominative (Toldova and Serdobolskaya 2008). We do not claim
that all of these instances of semantically-based DOM originate as information
structure marking: it is possible that object marking was always triggered by
semantic features. The historical scenarios outline above cannot be excluded;
however, since expansion and narrowing involve similar features, in the absence of clear historical evidence it is often impossible to tell which processes
have taken place in languages where the connection to information structure is
totally lost.
The third type seems to be the most common. In these languages, DOM
is generally motivated by referential semantics, but some semantic classes allow apparent optionality: objects with the same semantic features are either
marked or unmarked, depending on their information structure role. This type
can be labelled mixed, i.e. it is both fluid and split. As we have seen, examples of such languages are Hindi, Chatino, Khalkha Mongolian, Tundra
Nenets, Dolakha Newar and Tigre.
Subtype 3a is represented by some Bantu languages. KiSwahili object agreement is obligatory for animate objects, and optional otherwise. For inanimate
objects, agreement marks discourse salience (topicality) (Seidl and Dimitriadis 1997). In the Imithupi dialect of Makua, the object marker on the verb is
optional for objects of non-human classes and obligatory for human objects,

Towards a typology

223

even if the overt object is nontopical (Morimoto 2002). This is demonstrated

in example (27), in which the object is clearly focused but the object marker
on the verb cannot be omitted.
(27) Araa rima a-*(n-)-lh-re
mpani?
Araarima Subj-Obj-feed-Tense.Asp who
Who did Araarima feed?

(Morimoto 2002:(4a,b))

Given the historical scenario we have outlines, we can suspect that topical
marking spread onto human objects in Imithupi, and all animate objects in
KiSwahili.
In two related Austronesian languages, Selayarese and Makassarese, the
verb shows object agreement with topical definite objects (Finer 1997). Focused definite objects and indefinite objects do not trigger agreement.5 The
restriction of topicality marking to definite topics reflects the high frequency of
definite topics in discourse, although indefinite topics are in principle possible
(see Chapter 3). Finer (1997) does not explain how indefinite topics behave,
but in any case agreement is impossible with indefinite objects. Similarly, Aissen (2003a) shows that Sinhala casemarking is optional on animate objects,
but impossible on inanimates. On our account, the Sinhala pattern may have
resulted from narrowing of topicality marking: only animate topics come to
be casemarked, while inanimate topics remain unmarked. Sinhala, Selayarese
and Makassarese all belong to Subtype 3b, as does Tigre, where casemarking
is restricted to definite topical objects.
Roughly the same categories appear to be present in languages where agreement and casemarking are not restricted to the grammatical function of object.
For instance, topicality marking in Tariana does not seem to depend on any semantic restrictions, so this language would instantiate Type 1. The split Type 2
is represented by the extinct language Tangut (Tibeto-Burman), where agreement on the verb is triggered by first and second person arguments alone; the
agreement controller can correspond to a variety of grammatical functions including the subject, the object, the possessor of the subject and the possessor
of the object (Kepping 1979).6
5 Focus

and nonfocus/topical objects differ in their linear position in Seyalarese.

(1979) does not provide glosses for these examples; we reproduce Comries (2003)
glosses for these examples, including for the morpheme glossed only as ?.
6 Kepping

224

Grammaticalisation

(28) a. th In2 sIwo2 tsia1

ndziwo2 mi1 ndziu1 -nga2

why
virtuous people
Neg love-1Sg
Why do I not love virtuous people?

(Kepping 1979:268)

b. nga2 .in1 sa 1 men1 kwie1 ndI2 -khion-nga1

give-?-1Sg
to.me sramana fruit
Give the fruit of the sramana to me.

(Kepping 1979:269)

c. ndziwo2 ngI2 nga2 .in1 lda2 kI1 -zwon2 -nga2

" grasped-?-1Sg

someone
my
hand
Someone grasped my hand.

(Kepping 1979:270)

Thus, in Tangut only first and second person arguments trigger agreement, but
agreement is trigger-happy in terms of grammatical functions. The mixed
Type 3 includes Persian and a number of other languages.
10.4 Conclusion
Though we have not provided an explicit formalisation of the processes of linguistic change we have proposed in this chapter, we agree with Vincent (1997)
that LFG provides a fruitful theoretical setting for the exploration of these processes. As Vincent notes, LFG is well equipped to handle the lexical basis and
lexically specified exceptions to the processes we describe. Our precise formal
model of the relations between c-structure, f-structure, semantic structure, and
information structure provides a good basis for observing interactions among
these levels and the effect of features at different levels on the formulation and
reinterpretation of constraints involving the levels.
Our theory of the historical genesis of DOM is not complete: we are aware
that in many languages other factors play a role in DOM, for instance, volitionality and control on the part of the agent participant and the factors related to
the structure of event (apect, telicity, and incremental theme). The relationship
of these factors to topicality and DOM is yet to be explored.

11

Conclusion
We have explored the effect of information structure on grammatical marking, presenting evidence from languages that treat topics specially in terms
of grammatical marking. Topicality is a relational property of a referent, determined by the speakers assessment of its relative saliency, and cannot be
measured in terms of inherent semantic features such as animacy: topical
referents are what propositions are construed to be about. Crucial to our analysis is the possibility for more than one sentence element to be topical. We
distinguish between the primary topic and the secondary topic; both are topics, but the primary topic is more pragmatically prominent. Although there
is no unique alignment between information structure roles and grammatical
functions, there are important cross-linguistic tendencies in the grammatical
expression of primary and secondary topics: in particular, we have argued that
while subjects are prototypical/canonical primary topics, objects tend to be
associated with secondary topics.
In the simplest cases of apparently optional case- and adpositional marking and agreement, the factor determining the presence of marking is whether
a sentence element is topical. In some languages, casemarking and agreement
mark the topical status of any grammatical function, subjects as well as nonsubjects. Other languages grammatically mark topicality for a range of nonsubject NPs. Restrictions on marking in these cases have often been treated in
syntactic terms, but we believe that some of these apparent syntactic restrictions may be better thought of as a consequence of independent constraints on
how topics can be syntactically realised.
Some languages place additional syntactic requirements on the grammatical
marking of nonsubject topics, restricting marking to object-like grammatical
functions. This gives raise to differential object marking (DOM): casemarking and agreement patterns in many languages with DOM distinguish topical
225

226

Conclusion

objects, which are grammatically marked, from nontopical, grammatically unmarked objects. In emphasising the role of information structure in DOM, our
analysis accounts for cases of apparent optionality which have not been fully
addressed by semantically-driven proposals. In particular, we account for patterns of DOM in languages with local systems of object marking with no
obvious semantic motivation, showing that such patterns are determined by
information structure role. Optimality-theoretic approaches which rely on referential semantic features of the object handle such cases in terms of reranking
of constraints, leading to unexplained optionality of marking. On our analysis, true optionality does not arise: objects with the same semantic features
are either grammatically marked or unmarked, depending on their information
structure role.
Our approach provides a a unified account of topical marking that accounts
not only for DOM, but also for languages where marking of a variety of grammatical dependents depends on topicality: it accounts unproblematically for
languages like Persian, Itelmen and Tariana, where topic marking is not restricted to objects, as well as languages exhibiting the classic patterns of DOM.
It is not clear how previous analyses of DOM can easily be extended to account
for these languages.
Altogether, our analysis provides a motivation for DOM that is different
from the claims of much previous research. Most work on DOM assumes
that object marking originates from the need to differentiate the object from
the subject. However, we claim that DOM actually marks similarities rather
than differences between subjects (canonical topics) and topical objects: topics
tend to bear grammatical marking, no matter what their grammatical function.
Thus, our analysis does not relate the formal markedness of objects with their
functional markedness, at least if the latter is assessed in terms of frequency or
typicality. Instead, it highlights the coding or indexing function of marking
as an indicator of topicality. Our approach stands in opposition to the common
view that objects are prototypically aligned with the focus function: we have
argued that the SUBJ/topic, OBJ/(secondary) topic alignment is equally likely,
where both core arguments are topical. In support of this view, we have discussed evidence that shows that topical objects are as least as frequent in discourse as focused objects, and in this sense cannot be considered functionally
marked. In this, our analysis is in line with recent work emphasising the coding
function of marking in the related phenomenon of Differential Subject Marking: many of the contributors to a recent collection of papers on Differential
Subject Marking (de Hoop and de Swart 2009) argue that not all Differential
Subject Marking effects can be attributed to the disambiguating function of
grammatical marking, and this is exactly what we have found for DOM.
While DOM patterns in some languages are defined in purely information
structural terms, in other languages they depend on semantic features such as

227
animacy, definiteness, and the like, or on a combination of information structure and semantic features. We have suggested that marking based purely on
information structure roles is historically primary, at least in some languages,
and that the importance of semantic features emerges as a result of different
directions of grammaticalisation of topic marking. Grammatical marking is either extended to nontopical objects characterised by certain semantic features,
or restricted to a subset of topical arguments. The role of semantic features
in DOM is thus explained by historical patterns of grammaticalisation: typical
semantic features of topics come to be relevant for grammatical marking.
Another crucial difference between our work and many previous analyses
of DOM is that we do not discuss only grammatical (morphological) marking
of objects, but pay special attention to their behavioural syntax. Typologically
based work does not usually address the syntax of objects, while most generative research concentrates on positional differences. Our analysis does not
define DOM in terms of object position because we do not assume that syntactic roles are defined configurationally: following the standard LFG view, we
take grammatical functions to be primitives which are not defined in terms of
their syntactic position.
In our investigation of the grammatical behaviour of grammatically marked
and unmarked objects, we found that languages differ: in some languages they
are both primary objects, while in other languages they bear different objectlike functions. In languages like Ostyak, Khalkha Mongolian and Chatino,
grammatical marking of objects may seem to depend on information structure:
topical objects are marked, while nontopical objects are unmarked. However,
closer examination reveals that, in fact, marking patterns in these languages
are defined in completely syntactic terms, just as in English or Latin. The distinguishing characteristic of these languages is the obligatory linkage between
grammatical functions and information structure: primary objects are always
topical, while secondary or restricted objects are nontopical. This means that
in some cases grammatical structure may arise diachronically under pressure
from information structure constraints. The need to distinguish two types of
information structuring (with and without a topical object) has led to grammatical differences that go beyond patterns of agreement or casemarking.
This means that theories of argument mapping, relating semantic roles to
grammatical functions, must take into account not only the semantic role of the
argument of a predicate but also its information structure role in determining
its grammatical function. This is in line with work by Morimoto (2009:212),
who proposes to distinguish topic-prominent languages in which the topic is
linked to subject from subject-prominent languages in which the thematically
highest argument is linked to subject. We argue that the same is observed
with objects: in some languages, linking to the grammatical function of object

228

Conclusion

is strongly influenced by information structure role, while in other languages

semantic role is the sole determinant of linking patterns.
We have also clarified the position of information structure within the grammatical architecture of LFG, and shown how constraints from various levels of
linguistic structure can combine to specify and determine information structure role. LFG distinguishes grammatical marking, grammatical function, and
information structure role, which has been of crucial importance in formulating our theory. Abstract grammatical functions are not assumed to correlate
one-to-one with case or agreement morphology. They are not defined in terms
of their phrase structural position in the sentence or in terms of morphological
properties.
In sum, our approach is different from much previous work, in which casemarking and agreement patterns are treated in terms of purely syntactic factors
(sometimes in combination with semantics), by reference to syntactic roles
like subject and object. This works unproblematically for languages where
agreement patterns do not depend on information structure role. Our work
shows that these processes in fact can make reference to other levels of structure, but that careful examination is needed to determine whether they relate
directly to information structure, or only indirectly, by virtue of a tight alignment between informational roles and grammatical functions. Thus, we see
that although agreement and casemarking are syntactically constrained, different languages can exploit the syntax-information structure interface in different
ways to determine patterns of grammatical marking. Examining these patterns
illuminates our understanding of the syntax-information structure interface.

References
Ackerman, Farrell and Webelhuth, Gert 1998. A Theory of Predicates. Stanford, CA: CSLI Publications
Ahmed, Tafseer 2006. Spatial, temporal and structural uses of Urdu
ko.
URL http://csli-publications.stanford.edu/LFG/11/
lfg06ahmed.pdf. Workshop on case and aspect in South Asian languages,
LFG06, Konstanz
Aikhenvald, Alexandra Y. 2003. A grammar of Tariana, from Northwest Amazonia. Cambridge, UK: Cambridge University Press
Aissen, Judith 1999. Markedness and subject choice in optimality theory.
Natural Language and Linguistic Theory 17(4):673711. Also in Legendre
et al. (2001)
Aissen, Judith 2003a. Differential coding, partial blocking, and bidirectional
OT. In Nowak, Pawel and Yoquelet, Corey (eds.) Proceedings of the 29th
Annual Meeting of the Berkeley Linguistics Society, University of California, Berkeley: Berkeley Linguistics Society
Aissen, Judith 2003b. Differential object marking: Iconicity vs. economy.
Natural Language and Linguistic Theory 21(3):435483
Alsina, Alex 1996. The Role of Argument Structure in Grammar: Evidence
from Romance. Stanford, CA: CSLI Publications
Alsina, Alex 2008. A theory of structure-sharing: Focusing on long-distance
dependencies and parasitic gaps. In Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the LFG2008 Conference, URL
http://csli-publications.stanford.edu/LFG/13/lfg08.html

Andrews, III, Avery 1982. The representation of case in modern Icelandic. In

Bresnan, Joan (ed.) The Mental Representation of Grammatical Relations,
Cambridge, MA: The MIT Press, pp. 427503
229

230

REFERENCES

Andrews, III, Avery 1985. The major functions of the noun phrase. In
Shopen, Tim (ed.) Language Typology and Syntactic Description, Cambridge, UK: Cambridge University Press
Ariel, Mira 1988. Referring and accessibility. Journal of Linguistics 24:67
87
Asher, R. E. and Kumari, T. C. 1997. Malayalam. London and New York:
Routledge
Asudeh, Ash 2004. Resumption as Resource Management. Ph.D. thesis, Stanford University
Asudeh, Ash 2005. Relational nouns, pronouns, and resumption. Linguistics
and Philosophy 28(4):375446
Asudeh, Ash 2006. Direct compositionality and the architecture of LFG. In
Butt, Miriam, Dalrymple, Mary, and King, Tracy Holloway (eds.) Intelligent Linguistic Architectures: Variations on themes by Ronald M. Kaplan,
Stanford, CA: CSLI Publications
Austin, Peter and Bresnan, Joan 1996. Non-configurationality in Australian
Aboriginal languages. Natural Language and Linguistic Theory 14(2):215
268
Barczi, Geza 1980. A magyar nyelv multja e s jelene [The past and present of
the Hungarian language]. Budapest: Gondolat
BarDdal, Johanna 2001. Case in Icelandic: A synchronic, diachronic, and
comparative approach. In Lundastudier i Nordisk Sprakvetenskap, Department of Scandinavian Languages, Lund University, volume 57
Beames, John 1966 [1872-79]. A Comparative Grammar of the Modern Aryan
Languages of India. Delhi: Munshiram Manoharlal
Beaver, David I. and Clark, Brady Z. 2008. Sense and Sensitivity: How Focus
Determines Meaning. Chichester: Wiley-Blackwell
Beck, David 2006. Control of agreement in multi-object constructions in
Upper Necaxa Totonac. In Fujimori, Atsushi and Silva, Maria Amelia Reis
(eds.) Proceedings of the 11th Workshop on Structure and Constituency in
the Languages of the Americas, Vancouver, pp. 111
Beck, David 2007. Voice and agreement in multi-object constructions in
Upper Necaxa Totonac. In Memorias de IX Encuentro Internacional de
Lingustica en el Noroeste, Hermosillo

REFERENCES

231

Beck, David 2008. Sorting out grammatical relations in multi-object constructions in Upper Necaxa Totonac. On-line festschrift for Bernard Comrie
Berman, Judith 1999. Does German satisfy the Subject Condition? In
Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the
LFG99 Conference, URL http://csli-publications.stanford.
edu/LFG/4/lfg99.html

Bhatt, Rajesh and Anagnostopoulou, Elena 1996. Object shift and specificity:
Evidence from ko-phrases in Hindi. In Papers from the Thirty-Second
Regional Meeting of the Chicago Linguistic Society, University of Chicago:
Chicago Linguistic Society, pp. 1122
Bickel, Balthasar 2008. On the scope of the referential hierarchy in the typology of grammatical relations. In Corbett, Greville and Noonan, Michael
(eds.) Case and grammatical relations, Amsterdam: John Benjamins, pp.
191210
Bickel, Balthasar, Bisang, Walter, and Yadava, Yogendra P. 1999. Face vs.
empathy: the social foundation of Maithili verb agreement. Linguistics
37(3):481518
Blake, Barry J. 2001. Case. Cambridge, UK: Cambridge University Press,
second edition
Bobaljik, Jonathan David and Wurmbrand, Susi 2002.
Notes on
agreement in Itelmen. Linguistic Discovery 1(1). URL http://
linguistic-discovery.dartmouth.edu

Bokamba, Eyamba G. 1981. Aspects of Bantu syntax.

manuscript, cited in Bresnan and Mchombo (1987)

Unpublished

Borjars, Kersti, Chisarik, Erika, and Payne, John 1999. On the justification
for functional categories in LFG. In Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the LFG99 Conference, URL http:
//csli-publications.stanford.edu/LFG/4/lfg99.html

Borjars, Kersti and Vincent, Nigel 2005. Position vs. function in Scandinavian
Presentational Constructions. In Butt, Miriam and King, Tracy Holloway
(eds.) On-line Proceedings of the LFG2005 Conference, URL http://
csli-publications.stanford.edu/LFG/10/lfg05.html

Borjars, Kersti and Vincent, Nigel 2008. Objects and OBJ. In Butt,
Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the
LFG2008 Conference, URL http://csli-publications.stanford.
edu/LFG/13/lfg08.html

232

REFERENCES

Bos, Johan, Mastenbroek, Elsbeth, McGlashan, Scott, Millies, Sebastian, and

Pinkal, Manfred 1994. A compositional DRS-based formalism for NLPapplications. In International Workshop on Computational Semantics,
Tilburg. Also published as Verbmobil Report 59, Universitat des Saarlandes, Saarbrucken, Germany
Bossong, Georg 1985. Empirische Universalienforschung: Differentielle Objektmarkierung in den neuiranischen Sprachen. Tubingen: Gunter Narr
Bossong, Georg 1989. Morphemic marking of topic and focus. In Kefer,
Michel and van der Auwera, Johan (eds.) Belgian Journal of Linguistics 4:
Universals of Language, Amsterdam: John Benjamins
Bossong, Georg 1991. Differential object marking in Romance and beyond.
In Wanner, Dieter and Kibbee, Douglas A. (eds.) New Analyses in Romance Linguistics, Amsterdam: John Benjamins, pp. 143167
Bowern, Claire 2004. Bardi Verb Morphology in Historical Perspective. Ph.D.
thesis, Harvard University
Bresnan, Joan 1980. Polyadicity. In Hoekstra, Teun, van der Hulst, Harry, and
Moortgat, Michael (eds.) Lexical Grammar, Dordrecht: Foris Publications,
pp. 97121. Republished in revised form in Bresnan (1982)
Bresnan, Joan (ed.) 1982. The Mental Representation of Grammatical Relations. Cambridge, MA: The MIT Press
Bresnan, Joan 2001. Lexical-Functional Syntax. Oxford: Blackwell Publishers
Bresnan, Joan and Kanerva, Jonni M. 1989. Locative inversion in Chichewa:

A case study of factorization in grammar. Linguistic Inquiry 20(1):150.

Reprinted in Stowell et al. (1992)
Bresnan, Joan and Mchombo, Sam A. 1987. Topic, pronoun, and agreement
in Chichewa.

Language 63(4):741782
Bresnan, Joan and Moshi, Lioba 1990. Object asymmetries in comparative
Bantu syntax. Linguistic Inquiry 21(2):147186
Bresnan, Joan and Zaenen, Annie 1990. Deep unaccusativity in LFG. In
Dziwirek, Katarzyna, Farrell, Patrick, and Mejas-Bikandi, Errapel (eds.)
Grammatical Relations: A Cross-Theoretical Perspective, Stanford, CA:
CSLI Publications, pp. 4557
Brown, Robert 1988. Waris case system and verb classification. Language
and Linguistics in Melanesia 19:3780

REFERENCES

233

Browne, Wayles 1970. More on definiteness markers: Interrogatives in Persian. Linguistic Inquiry 1(3):359363
Buring, Daniel 2007. Semantics, intonation, and information structure. In
Ramchand, Gillian and Reiss, Charles (eds.) The Oxford Handbook of Linguistic Interfaces, Oxford: Oxford University Press, pp. 445474
Butt, Miriam 1993. Object specificity and agreement in Hindi/Urdu. In
Beals, K. et al. (eds.) Papers from the Twenty-Ninth Regional Meeting of
the Chicago Linguistic Society, University of Chicago: Chicago Linguistic
Society, pp. 89103
Butt, Miriam 1995. The Structure of Complex Predicates in Urdu. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and corrected version of 1993 Stanford University dissertation.
Butt, Miriam 2008a. From spatial expression to core case marker: Ergative
and dative/accusative. URL http://ling.uni-konstanz.de/pages/
home/butt/. MS, University of Konstanz
Butt, Miriam 2008b. Theories of Case. Cambridge, UK: Cambridge University
Press
Butt, Miriam, Ahmed, Tafseer, and Poudel, Tikaram 2008. Development
of case in South Asian languages. URL http://ling.uni-konstanz.
de/pages/home/tafseer/dgfs08-hnd.pdf. Paper presented at DGfS
Workshop Sprachwandelvergleich Bamberg, February 28th, 2008
Butt, Miriam, Grimm, Scott, and Ahmed, Tafseer 2006. Dative subjects. URL http://ling.uni-konstanz.de/pages/home/butt/
nijmegen-hnd.pdf. Paper presented at NWO/DFG Workshop on Optimal Sentence Processing
Butt, Miriam and King, Tracy Holloway 1996. Structural topic and focus without movement. In Butt, Miriam and King, Tracy Holloway
(eds.) On-line Proceedings of the LFG96 Conference, URL http://
csli-publications.stanford.edu/LFG/1/lfg1.html

Butt, Miriam and King, Tracy Holloway 1999. The status of case. In Dayal,
Veneeta and Mahajan, Anoop (eds.) Clause Structure in South Asian Languages, Dordrecht: Kluwer Academic Publishers, pp. 153198
Butt, Miriam and King, Tracy Holloway 2000. Null elements in discourse
structure. In Subbarao, K. V. (ed.) Papers from the NULLS Seminar,
Delhi: Motilal Banarsidass

234

REFERENCES

Butt, Miriam and King, Tracy Holloway 2003. Case systems: Beyond structural distinctions. In Brandner, Ellen and Zinsmeister, Heike (eds.) New
Perspectives on Case Theory, Stanford, CA: CSLI Publications, pp. 5387
Butt, Miriam, Nino, Mara-Eugenia, and Segond, Frederique 1996. Multilingual processing of auxiliaries in LFG. In Gibbon, D. (ed.) Natural Language Processing and Speech Technology: Results of the 3rd KONVENS
Conference, Mouton de Gruyter, pp. 111122. Universitat Bielefeld, 7 - 9
October 1996
Carleton, Troi and Waksler, Rachelle 2000. Pronominal markers in Zenzontepec Chatino. International Journal of American Linguistics 66(3):381395
Carleton, Troi and Waksler, Rachelle 2002. Marking focus in Chatino. Word
53(2):157171

Cetinoglu, Ozlem
and Butt, Miriam 2008. Turkish non-canonical objects. In
Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the
LFG2008 Conference, URL http://csli-publications.stanford.
edu/LFG/13/lfg08.html

Choi, Hye-Won 1999. Optimizing Structure in Context: Scrambling and Information Structure. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and corrected version of 1996 Stanford University dissertation.
Chomsky, Noam 1995. The Minimalist Program. Cambridge, MA: The MIT
Press
Chomsky, Noam and Lasnik, Howard 1993. The theory of principles and
parameters. In Jacobs, Joachim, von Stechow, Arnim, Sternefeld, Wolfgang, and Vennemann, Theo (eds.) Syntax: An International Handbook of
Contemporary Research, Cambridge, MA: Walter de Gruyter
Cole, Peter 1982. Imbabura Quechua. Amsterdam: North Holland
Comrie, Bernard 1977. Subjects and direct objects in Uralic languages: A

functional explanation of case-marking systems. Etudes

Finno-Ougriennes
12:517
Comrie, Bernard 1979. Definite and animate direct objects: A natural class.
Linguistica Silesiana III:1321
Comrie, Bernard 1989. Language universals and linguistic typology. Chicago:
University of Chicago Press, second edition

REFERENCES

235

Comrie, Bernard 2003. When agreement gets trigger-happy. Transactions of

the Philological Society 101(2):313337
Cook, Phillipa and Payne, John 2006. Information structure and scope in German. In Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the LFG2006 Conference, URL http://csli-publications.
stanford.edu/LFG/11/lfg06.html

Corbett, Greville G. 2006. Agreement. Cambridge, UK: Cambridge University

Press
Creissels, Denis 2008. Remarks on split intransitivity and fluid intransitivity.
In Bonami, Olivier and Hofherr, Patricia Cabredo (eds.) Empirical issues in
syntax and semantics, CSSP, volume 7, pp. 139168. URL http://www.
cssp.cnrs.fr/eiss7/index_en.html

Croft, William 1991. Syntactic Categories and Grammatical Relations: The

Cognitive Organization of Information. Chicago: University of Chicago
Press
Croft, William A. 1988. Agreement vs. case marking and direct objects.
In Barlow, Michael and Ferguson, Charles A. (eds.) Agreement in Natural
Language: Approaches, Theories, Descriptions, Stanford, CA: CSLI Publications, pp. 159179
Croft, William A. 2001. Radical Construction Grammar: Syntactic Theory in
Typological Perspective. Oxford: Oxford University Press
Culicover, Peter W. and Jackendoff, Ray 2005. Simpler Syntax. Oxford University Press
Culy, Christopher D. 1994. Extraposition in English and grammatical theory.
MS, University of Iowa
Culy, Christopher D. 2000. An incorporated topic marker in Takelma. In
Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the
LFG2000 Conference, URL http://csli-publications.stanford.
edu/LFG/5/lfg00.html

Cummins, Sarah 2000. The unaccusative hypothesis and the impersonal construction in French. Canadian Journal of Linguistics 45:225251
Dabir-Moghaddam, Mohammad 1992. On the (in)dependence of syntax and
pragmatics: Evidence from the postposition -r
a in Persian. In Stein, Dieter
(ed.) Cooperating with written texts: the pragmatics and comprehension of
written texts, Berlin: Mouton de Gruyter, pp. 549573

236

REFERENCES

Dahl, Osten
and Fraurud, K. 1996. Animacy in grammar and discourse. In
Fretheim, Thorstein and Gundel, Jeanette K. (eds.) Reference and referent
accessibility, Amsterdam: John Benjamins, pp. 4764.
Dahlstrom, Amy 2009. OBJ without OBJ: a typology of Meskwaki objects. In Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the LFG2009 Conference, URL http://csli-publications.
stanford.edu/LFG/14/lfg09.html

Dalrymple, Mary (ed.) 1999. Semantics and Syntax in Lexical Functional

Grammar: The Resource Logic Approach. Cambridge, MA: The MIT Press
Dalrymple, Mary 2001. Lexical Functional Grammar, volume 34 of Syntax
and Semantics. New York, NY: Academic Press
Dalrymple, Mary, Kaplan, Ronald M., Maxwell, III, John T., and Zaenen, Annie (eds.) 1995. Formal Issues in Lexical-Functional Grammar. Stanford,
CA: CSLI Publications
Dalrymple, Mary, Lamping, John, Pereira, Fernando C. N., and Saraswat, Vijay A. 1997. Quantifiers, anaphora, and intensionality. Journal of Logic,
Language, and Information 6(3):219273. Reprinted in Dalrymple (1999)
Dalrymple, Mary, Lamping, John, Pereira, Fernando C. N., and Saraswat, Vijay A. 1999. Overview and introduction. In Dalrymple, Mary (ed.) Semantics and Syntax in Lexical Functional Grammar: The Resource Logic
Approach, Cambridge, MA: The MIT Press
Dalrymple, Mary and Nikolaeva, Irina 2005. Nonsubject agreement and discourse roles. In Oxford Working Papers in Linguistics, Philology, and
Phonetics, University of Oxford: Centre for Linguistics and Philology, pp.
7394
Danon, Gabi 2006. Caseless nominals and the projection of DP. Natural
Language and Linguistic Theory 24(4):9771008
Dayal, Veneeta 2003. Bare nominals: Non-specific and contrastive readings
under scrambling. In Karimi, Simin (ed.) Word order and scrambling,
Oxford: Blackwell Publishers
de Hoop, Helen 1992. Case configuration and noun phrase interpretation.
Ph.D. thesis, University of Groningen. Reprinted by Garland Press, New
York, 1996
de Hoop, Helen 2008. Against affectedness. In Cognitive and Functional
Perspectives on Dynamic Tendencies in Languages, Tartu: University of
Tartu and ECLA, pp. 212213

REFERENCES

237

de Hoop, Helen and de Swart, Peter (eds.) 2009. Differential Subject Marking.
Dordrecht: Springer
de Hoop, Helen and Lamers, Monique 2006. Incremental distinguishability of
subject and object. In Kulikov, Leonid, Malchukov, Andrej, and de Swart,
Peter (eds.) Case, valency and transitivity, Amsterdam: John Benjamins,
pp. 269287
de Hoop, Helen and Malchukov, Andrej 2007. On fluid differential case marking: A bidirectional OT approach. Lingua 117:16361656
de Hoop, Helen and Narasimhan, Bhuvana 2005. Differential case-marking in
Hindi. In Amberber, Mengistu and de Hoop, Helen (eds.) Competition and
variation in natural languages: the case for case, London and Amsterdam:
Elsevier, pp. 321345
de Swart, Peter 2006. Case markedness. In Kulikov, Leonid, Malchukov, Andrej, and de Swart, Peter (eds.) Case, valency and transitivity, Amsterdam:
John Benjamins, pp. 249267
de Swart, Peter 2007. Cross-linguistic Variation in Object Marking. Ph.D.
thesis, Radboud University Nijmegen
Deo, Ashwini 2008. Datives: Locations and possessors. Case syncretism
in Marathi diachrony. URL http://ling.uni-konstanz.de/pages/
home/tafseer/workshop/deo_case_2008.pdf. Paper presented at
SFB 471 Case Workshop University of Konstanz, May 8-10 2008
Diesing, Molly 1992. Indefinites. Cambridge, MA: The MIT Press
Dixon, Robert M. W. 1994. Ergativity. Cambridge, UK: Cambridge University
Press
Dobrovie-Sorin, Carmen 1994. The syntax of Romanian. Berlin: Mouton de
Gruyter
Donohue, Mark 2004. A voice opposition without voice morphology. Paper
presented at AFLA 2004, Berlin
Dowty, David R. 1991. Thematic proto roles and argument selection. Language 67(3):547619
Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative.
Language 62(4):808845
Du Bois, John W. 1987.
63(4):805855

The discourse basis of ergativity.

Language

238

REFERENCES

Kiss, Katalin (ed.) 1995. Discourse Configurational Languages. Oxford:

E.
Oxford University Press
Kiss, Katalin 2002. The syntax of Hungarian. Cambridge, UK: Cambridge
E.
University Press
Enc, Murvet 1986. Towards a referential analysis of temporal expressions.
Linguistics and Philosophy 9(4):405426
Enc, Murvet 1991. The semantics of specificity. Linguistic Inquiry 22:126
Engdahl, Elisabet and Vallduv, Enric 1996. Information packaging in HPSG.
In Grover, Claire and Vallduv, Enric (eds.) Studies in HPSG, Edinburgh:
University of Edinburgh, volume 12 of Edinburgh Working Papers in Cognitive Science
Erteschik-Shir, Nomi 1997. The dynamics of focus structure. Cambridge, UK:
Cambridge University Press
Erteschik-Shir, Nomi 2007. Information Structure: The Syntax-Discourse Interface. Oxford: Oxford University Press
Escandell-Vidal, M. Victoria 2009. Differential object marking and topicality:
The case of Balearic Catalan. Studies in Language 33(4):832884
Falk, Yehuda N. 2001. Lexical-Functional Grammar: An Introduction to Parallel Constraint-Based Syntax. Stanford, CA: CSLI Publications
Falk, Yehuda N. 2006. Subjects and Universal Grammar: An Explanatory
Theory. Cambridge, UK: Cambridge University Press
Farkas, Donca 1995. Specificity and scope. In Nash, L. and Tsoulas, G. (eds.)
Actes du Premier Colloque Langues and Grammaire, Paris, pp. 119137
Fery, Caroline and Krifka, Manfred 2008. Information structure: Notional
distinctions, ways of expression. In van Sterkenburg, Piet (ed.) Unity and
diversity of languages, Amsterdam: John Benjamins, pp. 123136
Filimonova, Elena 2005. The noun phrase hierarchy and relational marking:
Problems and counterevidence. Linguistic Typology 9:77113

Finer, Daniel L. 1997. Contrasting A-dependencies

in Selayarese. Natural
Language and Linguistic Theory 15:677728
Frank, Anette and Zaenen, Annie 2002. Tense in LFG: Syntax and morphology. In Kamp, Hans and Reyle, Uwe (eds.) How we say WHEN it happens: Contributions to the theory of temporal reference in natural language,
Tubingen: Niemeyer

REFERENCES

239

Fry, John 2001. Ellipsis and WA-Marking in Japanese Conversation. Ph.D.

thesis, Stanford University
Gair, James W. and Paolillo, John C. 1997. Sinhala. Munchen: LINCOM
Europa
Genetti, Carol 1994. A Descriptive and Historical Account of the Dolakha
Newari Dialect. Tokyo: Institute for the Study of Languages and Cultures
of Africa and Asia
Genetti, Carol 1997. Object relations and dative case in Dolakha Newari.
Studies in Language 21(1):3768
Genetti, Carol 2007. A grammar of Dolakha Newar. Mouton Grammar Series
40. Berlin: Mouton de Gruyter
Georgopoulos, Carol 1992. Another look at object agreement. In Proceedings of the 22nd Conference of the North Eastern Linguistic Society
(NELS22), University of Massachusetts at Amherst, pp. 165177
Geurts, Bart 2002. Specific indefinites, presupposition and scope. In Bauerle,
Rainer, Reyle, Uwe, and Zimmermann, Thomas Ede (eds.) Presuppositions
in discourse, Oxford: Elsevier. To appear
Ghomeshi, Jila 1997. Topics in Persian VPs. Lingua 102:133167
Givon, Talmy 1976. Topic, pronoun, and grammatical agreement. In Li,
Charles (ed.) Subject and Topic, New York: Academic Press, pp. 5798
Givon, Talmy 1979. On Understanding Grammar. New York: Academic Press
Givon, Talmy 1983a. Introduction. In Givon, Talmy (ed.) Topic Continuity in Discourse: A Quantitative Cross-Language Study, Amsterdam: John
Benjamins
Givon, Talmy (ed.) 1983b. Topic Continuity in Discourse: A Quantitative
Cross-Language Study. Amsterdam: John Benjamins
Givon, Talmy 1984a. Direct object and dative shifting: semantic and pragmatic case. In Plank, Frans (ed.) Objects: Towards a theory of grammatical
relations, London: Academic Press, pp. 151182
Givon, Talmy 1984b. Syntax: A Functional-Typological Introduction, volume I. Amsterdam: John Benjamins
Givon, Talmy 1990. Syntax: A Functional-Typological Introduction, volume II. Amsterdam: John Benjamins

240

REFERENCES

Givon, Talmy 2001. Syntax. Amsterdam: John Benjamins

Goldberg, Adele E. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford University Press
Golovko, Evgenij 2009. Aleutskij jazyk v Rossijskoj Federacii (Struktura,
funkcionirovanie, kontaktnye javlenija). [The Aleut language in the Russian Federation: Structure, functioning, contact phenomena.]. Ph.D. thesis,
Saint Petersburg Academy of Sciences
Greenberg, Joseph H. 1977. Niger-Congo noun class markers: Prefixes, suffixes, both or neither. Studies in African Linguistics, supplement 7 pp. 97
104
Grimes, Barbara F. 1999. ETHNOLOGUE: Languages of the World. SIL
International. URL http://www.sil.org/ethnologue/
Gundel, Jeanette K. 1988. The role of topic and comment in linguistic theory.
New York: Garland
Gundel, Jeanette K., Hedberg, Nancy, and Zacharski, Ron 1993. Cognitive
status and the form of referring expressions in discourse. Language 69:274
308
Guntsetseg, Dolgor 2008. Differential object marking in Mongolian. In
Schafer, F. (ed.) SinSpec: Working Papers of the FSB732 Incremental
Specification in Context, Universitat Stuttgart, volume 1, pp. 5369
Haig, Geoffrey 2008. Alignment change in Iranian languages: a Construction
Grammar approach. New York: Mouton de Gruyter
Haiman, John and Thompson, Sandra A. 1985. Natural Syntax: Iconicity and
Erosion. Cambridge, UK: Cambridge University Press
Hajicova, Eva, Partee, Barbara H., and Sgall, Petr 1998. Topic-focus articulation, tripartite structures, and semantic content. Dordrecht: Kluwer Academic Publishers
Harris, Alice and Campbell, Lyle 1995. Historical syntax in cross-linguistic
perspective. Cambridge, UK: Cambridge University Press
Haspelmath, Martin 2004. Explaining the ditransitive person-role constraint:
A usage-based account. Constructions 2/2004. URL http://www.
constructions-online.de/articles/35

REFERENCES

241

Haspelmath, Martin 2007. Ditransitive alignment splits and inverse alignment. Functions of Language 14(1):79102. Special issue on ditransitives,
guest edited by Anna Siewierska
Heim, Irene 1982. The Semantics of Definite and Indefinite Noun Phrases.
Ph.D. thesis, University of Massachusetts at Amherst
Heine, Bernd and Kuteva, Tania 2002. World Lexicon of Grammaticalization.
Cambridge, UK: Cambridge University Press
Her, One-Soon 1991. Topic as a grammatical function in Chinese. Lingua
84(1):123
Hewson, John and Bubenik, Vit 2006. From case to adposition: the development of configurational syntax in Indo-European languages. Amsterdam:
John Benjamins
Hockett, Charles 1958. A course in modern linguistics. New York: Macmillan
Honti, Laszlo 1984. Chrestomathia ostiacica. Budapest: Tankonyvkiado
Hopper, Paul J. and Thompson, Sandra A. 1980. Transitivity in grammar and
discourse. Language 56:251299
Huang, Chu-Ren 1989.
Subcategorized TOPICs in Mandarin Chihttp://cwn.ling.sinica.edu.tw/churen/
nese.
URL
MS, Academia
SubcategorizedTOPICsinMandarinChinese.pdf.
Sinica
Hudson, Richard 1992. So-called double objects and grammatical relations.
Language 68(2):251276
Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar.
Cambridge, MA: The MIT Press
Jackendoff, Ray S. 1983. Semantics and Cognition. Cambridge, MA: The
MIT Press
Jackendoff, Ray S. 1990. Semantic Structures. Cambridge, MA: The MIT
Press
Jaeger, T. Florian 2004. Topicality and superiority in Bulgarian wh-questions.
In Arnaudova, Olga, Browne, Wayles, Rivero, Maria Luisa, and Stojanovic,
Danijela (eds.) Proceedings of FASL-12: The Ottawa Meeting 2003, pp.
207228

242

REFERENCES

Jaeger, T. Florian and Gerassimova, Veronica A. 2002. Bulgarian word order and the role of the direct object clitic in LFG. In Butt, Miriam and
King, Tracy Holloway (eds.) On-line Proceedings of the LFG2002 Conference, URL http://csli-publications.stanford.edu/LFG/7/
lfg2.html

Jager, Gerhard 2003. Learning constraint subhierarchies: the Bidirectional

Gradual Learning Algorithm. In Blutner, Reinhard and Zeevat, Henk (eds.)
Optimality theory and pragmatics, Basingstoke: Palgrave Macmillan, pp.
251298
Jake, Janice L. 1980. Object verb agreement in Tigre. Studies in the Linguistic Sciences 10(1):7184
Jelinek, Eloise and Carnie, Andrew 2003. Argument hierarchies and the mapping principle. In Carnie, Andrew, Harley, Heidi, and Willie, Mary Ann
(eds.) Formal Approaches to Function in Grammar: In Honor of Eloise
Jelinek, Philadelphia: John Benjamins, pp. 265296
Johnson, David E. and Postal, Paul M. 1980. Arc Pair Grammar. Princeton:
Princeton University Press
Kadmon, Nirit 2001. Formal Pragmatics. Malden: Blackwell Publishers
Kamp, Hans and Reyle, Uwe 1993. From Discourse to Logic: An Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and
Discourse Representation Theory. Dordrecht: Kluwer Academic Publishers
Kaplan, Ronald M. 1987. Three seductions of computational psycholinguistics. In Whitelock, Peter, Wood, Mary McGee, Somers, Harold L.,
Johnson, Rod, and Bennett, Paul (eds.) Linguistic Theory and Computer
Applications, London: Academic Press, pp. 149188. Also published as
CCL/UMIST Report No. 86.2: Alvey/ICL Workshop on Linguistic Theory
and Computer Applications: Transcripts of Presentations and Discussions.
Center for Computational Linguistics, University of Manchester. Reprinted
in Dalrymple et al. (1995:337367)
Kaplan, Ronald M. 1988. Correspondences and their inverses. Presented
at the Titisee Workshop on Unification Formalisms: Syntax, Semantics and
Implementation, Titisee, Germany
Kaplan, Ronald M. and Bresnan, Joan 1982. Lexical-Functional Grammar:
A formal system for grammatical representation. In Bresnan, Joan (ed.)
The Mental Representation of Grammatical Relations, Cambridge, MA:
The MIT Press, pp. 173281. Reprinted in Dalrymple et al. (1995:29130)

REFERENCES

243

Kaplan, Ronald M. and Zaenen, Annie 1989. Long-distance dependencies, constituent structure, and functional uncertainty. In Baltin, Mark
and Kroch, Anthony (eds.) Alternative Conceptions of Phrase Structure,
Chicago University Press, pp. 1742. Reprinted in Dalrymple et al.
(1995:137165)
Karimi, Simin 1990. Obliqueness, specificity, and discourse functions: Ra in
Persian. Linguistic Analysis 20:139191
Keenan, Edward L. 1976. Towards a universal definition of subject. In Li,
C. N. (ed.) Subject and Topic, New York: Academic Press, pp. 303333
Keenan, Edward L. and Comrie, Bernard 1977. Noun phrase accessibility and
universal grammar. Linguistic Inquiry 8(1):6399
Kellogg, Samuel Henry 1955 [1893]. A grammar of the Hindi language. London: Routledge and Kegan Paul
Kepping, Ksenia B. 1979. Elements of ergativity and nominativity in Tangut.
In Planck, Frans (ed.) Ergativity, London: Academic Press, pp. 263277
Kibrik, Aleksandr and Seleznev, A. 1980. Sintaksis i morfologija glagolnogo
soglasovanija v tabasaranskom jazyke [The syntax and morphology of verbal agreement in Tabassaran]. In Zvegincev, Vladimir (ed.) Tabasaranskie
etjudy. Materialy dagestanskoj ekspedicii 1979 g., Moscow: MGU, pp. 17
33
Kifle, Nazareth Amlesom 2007. Differential object marking and topicality
in Tigrinya. In Butt, Miriam and King, Tracy Holloway (eds.) On-line
Proceedings of the LFG2007 Conference,
Kim, Alan 1988. Preverbal focus position in type XIII languages. In Hammond, Michael, Moravcsik, Edith, and Wirth, Jessica (eds.) Studies in syntactic typology, Amsterdam: John Benjamins, pp. 148171
King, Tracy Holloway 1995. Configuring Topic and Focus in Russian. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and
corrected version of 1993 Stanford University dissertation.
King, Tracy Holloway 1997. Focus domains and information structure. In
Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the
LFG97 Conference, URL http://csli-publications.stanford.
edu/LFG/2/lfg97.html

244

REFERENCES

King, Tracy Holloway and Zaenen, Annie 2004. F-structures, information

structure, and discourse structure. In Butt, Miriam and King, Tracy Holloway (eds.) On-line Proceedings of the LFG2004 Conference, URL
http://csli-publications.stanford.edu/LFG/9/lfg04.html

Kittila, Seppo 2006a. Object-, animacy- and role-based strategies: A typology

of object marking. Studies in Language 30:132
Kittila, Seppo 2006b. The woman showed the baby to her sister: On resolving humanness-driven ambiguity in ditransitives. In Kulikov, Leonid,
Malchukov, Andrej, and de Swart, Peter (eds.) Case, Valency and Transitivity, Amsterdam: John Benjamins, pp. 291308
Kornfilt, Jaklin 2009. DOM and two types of DSM in Turkish. In de Hoop,
Helen and de Swart, Peter (eds.) Differential Subject Marking, Dordrecht:
Springer, pp. 79111
Krifka, Manfred 1992. A framework for focus-sensitive quantification. In
Barker, Chris and Dowty, David (eds.) The Proceedings from the Second
Conference on Semantics and Linguistic Theory: SALT II, Department
of Linguistics, Ohio State University, pp. 215236. Ohio State University
Working Papers in Linguistics, Number 40
Krifka, Manfred 2006. Association with focus phrases. In Molnar, Valerie
and Winkler, Susanne (eds.) The Architecture of Focus, Berlin: Mouton
de Gruyter
Kroeger, Paul 1993. Phrase Structure and Grammatical Relations in Tagalog.
Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised and
corrected version of 1991 Stanford University dissertation.
Kulonen, Ulla-Maja 1989. The passive in Ob-Ugrian. Helsinki: Finno-Ugrian
Society
Kuznecova, Ariadna, Xelimskij, Evgenij, and Gruskina, Elena 1982. Ocherki
po selkupskomu jazyku [Studies in Selkup]. Moscow: MGU
Kwon, Song-Nim and Zribi-Hertz, Anne 2008. Differential function marking, case and information structure: Evidence from Korean. Language
84(2):218257
Lam, Olivia S.-C. 2008. Object Functions and the Syntax of Double Object
Constructions in Lexical Functional Grammar. D.Phil. dissertation, Oxford
University

REFERENCES

245

Lambrecht, Knud 1981. Topic, antitopic and verb agreement in non-standard

French. Amsterdam: John Benjamins
Lambrecht, Knud 1987. On the status of SVO sentences in French discourse.
In Tomlin, Russell S. (ed.) Coherence and Grounding in Discourse, Amsterdam and Philadelphia: John Benjamins, pp. 217261
Lambrecht, Knud 1994. Information structure and sentence form: Topic, focus, and the mental representation of discourse referents. Cambridge, UK:
Cambridge University Press
Lambrecht, Knud 2000. When subjects behave like objects: An analysis of
the merging of S and O in sentence-focus constructions across languages.
Studies in Language 24:611682
Lambrecht, Knud and Polinsky, Maria 1997. Typological variation in sentence
focus constructions. In Papers from the Thirty-Third Regional Meeting of
the Chicago Linguistic Society: Panels on Linguistic Ideologies in Contact,
University of Chicago: Chicago Linguistic Society, pp. 189206
Lazard, Gilbert 1982. Le morph`eme ra en persan et les relations actancielles.
Bulletin de la societe de linguistique de Paris 73:177208
Lazard, Gilbert 1984. Actance variations and categories of the object. In
Plank, Frans (ed.) Objects: Towards a theory of grammatical relations,
London: Academic Press, pp. 269292
Lazard, Gilbert 1992. A grammar of contemporary Persian. Costa Mesa, CA:
Mazda Publishers
Legendre, Geraldine, Vikner, Sten, and Grimshaw, Jane (eds.) 2001.
Optimality-Theoretic Syntax. The MIT Press
Leonetti, Manuel 2003. Specificity and object marking: the case of Spanish a. In von Heusinger, Klaus and Kaiser, Georg A. (eds.) Proceedings
of the Workshop Semantic and syntactic aspects of specificity in Romance
languages, Konstanz: University of Konstanz, pp. 67101
Levin, Lori S. 1986. Operations on Lexical Forms: Unaccusative Rules in
Germanic Languages. Ph.D. thesis, MIT
Liao, Gwen W.-T. 2010. An LFG Account of Empty Pronouns in Mandarin
Chinese. D.Phil. dissertation, Oxford University
Lidz, Jeffrey 2006. The grammar of accusative case in Kannada. Language
82(1):1032

246

REFERENCES

Ldrup, Helge 1999. Linking and optimality in the Norwegian presentational

focus construction. Nordic Journal of Linguistics 22(2):205229
MacKay, Carolyn J. and Trechsel, Frank R. 2008. Symmetrical objects in Misantla Totonac. International Journal of American Linguistics 74(2):227
255
Mahootian, Shahrzad 1997. Persian. London and New York: Routledge
Malchukov, Andrej 2006. Transitivity parameters and transitivity alternations: Constraining co-variation. In Kulikov, L. I., Malchukov, Andrej, and
de Swart, Peter (eds.) Case, valency, and transitivity: A cross-linguistic
perspective, Amsterdam: John Benjamins, pp. 330357
Malchukov, Andrej, Haspelmath, Martin, and Comrie, Bernard 2007. Ditransitive constructions: A typological overview. URL http://email.
eva.mpg.de/haspelmt/DitransitiveOverview.pdf. Draft for the
Conference on Ditransitive Constructions, November 2007
Malchukov, Andrej L. 2008. Animacy and asymmetries in differential case
marking. Lingua 118:203221
Maling, Joan and Zaenen, Annie (eds.) 1990. Syntax and Semantics: Modern
Icelandic Syntax, volume 24. San Diego: Academic Press
Marcantonio, Angela 1985. On the definite vs. indefinite conjugation in Hungarian: A typological and diachronic analysis. Acta Linguistica Academiae
Scientiarum Hungaricae 35:267298
Masica, Colin 1982. Identified object marking in Hindi and other languages.
In Koul, O. N. (ed.) Topics in Hindi linguistics, New Delhi: Bahri Publications, pp. 1650
Masica, Colin 1991. The Indo-Aryan languages. Cambridge, UK: Cambridge
University Press
Maslova, Elena 2003a. A Grammar of Kolyma Yukaghir. Berlin: Mouron de
Gruyter
Maslova, Elena 2003b. Information focus in relational clause structure. In
Tsunoda, Tasaku and Kageyama, Taro (eds.) Voice and Grammatical Relations: Festschrift for Masayoshi Shibatani, Amsterdam: John Benjamins,
Typological Studies in Language, pp. 175194

REFERENCES

247

Matsumoto, Yo 1996. Complex Predicates in Japanese: A Syntactic and Semantic Study of the Notion Word. Ph.D. thesis, Stanford University. Studies in Japanese Linguistics series, CSLI Publications/Kuroiso Publishers,
Stanford/Tokyo. Revised and corrected version of 1992 Stanford University
dissertation, On the wordhood of complex predicates in Japanese.
Mayer, Elisabeth 2008. Clitics on the move: From dependent marking to split marking. In Butt, Miriam and King, Tracy Holloway
(eds.) On-line Proceedings of the LFG2008 Conference, URL http:
//csli-publications.stanford.edu/LFG/13/lfg08.html

McCawley, James D. 1968. Concerning the base component of a transformational grammar. Foundations of Language 4:243269. Reprinted in
McCawley (1973)
McCawley, James D. 1973. Grammar and Meaning: Papers on Syntactic and
Semantic Topics. Tokyo: Taishukan
McGregor, R. S. 1972. Outline of Hindi Grammar. Oxford/Delhi: Oxford
University Press
Meinunger, Andre 1998. Topicality and agreement. In Darnell, Michael,
Moravcsik, Edith A., Noonan, Michael, Newmeyer, Frederick J., and
Wheatley, Kathleen M. (eds.) Functionalism and formalism in linguistics,
Amsterdam: John Benjamins, volume 2, pp. 203219
Mithun, Marianne 1991. Active/agentive case marking and its motivation.
Language 67:510546
Mohanan, K. P. 1982. Grammatical relations and clause structure in Malayalam. In Bresnan, Joan (ed.) The Mental Representation of Grammatical
Relations, Cambridge, MA: The MIT Press, pp. 504589
Mohanan, Tara 1994. Arguments in Hindi. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Reprinted version of 1990 Stanford University
dissertation.
Morimoto, Yukiko 2002. Prominence mismatches and differential object marking in Bantu. In Butt, Miriam and King, Tracy Holloway
(eds.) On-line Proceedings of the LFG2002 Conference, URL http:
//csli-publications.stanford.edu/LFG/7/lfg2.html

Morimoto, Yukiko 2009. From topic to subject marking: Implications for a

typology of subject marking. In de Hoop, Helen and de Swart, Peter (eds.)
Differential Subject Marking, Dordrecht: Springer, pp. 199221

248

REFERENCES

Mycock, Louise 2006. The Typology of Constituent Questions: A Lexical

Functional Grammar analysis of wh-questions. Ph.D. thesis, University
of Manchester
Mycock, Louise 2009. What do you do?: Variation in interrogative predicates. Paper presented at the workshop Blurring Component Boundaries:
levels of analysis or growth of information?, LFG09, Cambridge, July 2009

Nss, Ashild
2004. What markedness marks: The markedness problem with
direct objects. Lingua 114:11861212

Nss, Ashild
2006. Case semantics and the agent-patient opposition. In Kulikov, Leonid, Malchukov, Andrej, and de Swart, Peter (eds.) Case, valency
and transitivity, Amsterdam: John Benjamins, pp. 309327

Nss, Ashild
2007. Prototypical Transitivity. Amsterdam: John Benjamins
Neeleman, Ad and van de Koot, Hans 2008. Dutch scrambling and the nature of discourse templates. Journal of Comparative Germanic Linguistics
11(2):137189
Nichols, Johanna 1986. Head-marking and dependent-marking grammar.
Language 62(1):56119
Nikolaeva, Irina 1999. Object agreement, grammatical relations, and information structure. Studies in Language 23:331376
Nikolaeva, Irina 2001. Secondary topic as a relation in information structure.
Linguistics 39:149
Nordlinger, Rachel 1998. Constructive Case: Evidence from Australian Languages. Dissertations in Linguistics. Stanford, CA: CSLI Publications. Revised version of 1997 Stanford University dissertation.
OConnor, Robert 2006. Information Structure in Lexical Functional Grammar: The Discourse-Prosody Correspondence. Ph.D. thesis, University of
Manchester
Paoli, Sandra 2009. Contrastiveness and new information: A new view on
focus. MS, University of Oxford

Papay, Jozsef 1906-1908. Eszaki-osztj

ak nyelvtanulmanyok [Studies on
Northern Ostyak]. Nyelvtudomanyi Kozlemenyek 3638:345398, 5279,
164195, 258275, 111150, 313329

REFERENCES

249

Paul, Ludwig 2003. Early Judaeo-Persian in a historical perspective: The

case of the prepositions be, u, pa(d), and the suffix r
a. In Paul, Ludwig
(ed.) Persian Origins - Early Judaeo-Persian and the emergence of New
Persian, Wiesbaden: Harrassowitz Verlag, pp. 177194
Payne, Doris L. and Barshi, Immanuel 1999. External possession: What,
how, and why. In Payne, Doris L. and Barshi, Immanuel (eds.) External
possession, Amsterdam: John Benjamins, pp. 329
Peterson, David A. 2007. Applicative Constructions. Oxford: Oxford University Press
Plank, Frans (ed.) 1995. Double Case: Agreement by Suffixaufnahme. Oxford:
Oxford University Press
Platzack, Christer 1983. Existential sentences in English, Swedish, German
and Icelandic. In Karlsson, Fred (ed.) Papers from the Seventh Scandinavian Conference of Linguistics, University of Helsinki, Department of
General Linguistics, pp. 80100
Polinsky, Maria 1995. Double objects in causatives: Towards a study of coding conflict. Studies in Language 19:129221
Polinsky, Maria 1998. A non-syntactic account of some asymmetries in the
double object construction. In Koenig, Jean-Pierre (ed.) Conceptual structure and language: Bridging the gap, Stanford, CA: CSLI Publications, pp.
403423
Polinsky, Maria and Potsdam, Eric 2001. Long-distance agreement and topic
in Tzez. Natural Language and Linguistic Theory 19(3):583646
Portner, Paul and Yabushita, Katsuhiko 1998. The semantics and pragmatics
of topic phrases. Linguistics and Philosophy 21(2):117157
Portner, Paul and Yabushita, Katsuhiko 2001. Specific indefinites and the
information structure theory of topics. Journal of Semantics 18:271297
Prince, Ellen 1981. Topicalization, focus-movement, and Yiddish-movement:
a pragmatic differentiation. In Proceedings of the Seventh Annual Meeting
of the Berkeley Linguistics Society, Berkeley Linguistics Society
Ramchand, Gillian 1997. Aspect and Predication: The Semantics of Argument
Structure. Oxford University Press
Raz, Shlomo 1983. Tigre grammar and texts. Malibu: Undena Publications

250

REFERENCES

Redei, Karoly 1965. Northern Ostyak chrestomathy. The Hague, Bloomington: Indiana University Press
Reinhart, Tanya 1982. Pragmatics and linguistics: An analysis of sentence
topics. Philosophica 27:5394
Ritter, Elizabeth and Rosen, Sara Thomas 2001. The interpretive value of
object splits. Language Sciences 23:425451
Rizzi, Luigi 1997. The fine structure of the left periphery. In Haegeman, Liliane (ed.) Elements of Grammar, Dordrecht: Kluwer Academic Publishers,
pp. 281337
Roberts, Craige 1996. Information structure in discourse: Towards an integrated formal theory of pragmatics. In Kathol, Andreas and Yoon, Jae-Hak
(eds.) Ohio State University Working Papers in Linguistics, Ohio State
University, volume 49, pp. 91136
Rooth, Mats 1992. A theory of focus interpretation. Natural Language Semantics 1:75116
Rooth, Mats E. 1985. Association with Focus. Ph.D. thesis, University of
Massachusetts at Amherst
Rude, Noel 1982. Promotion and topicality of Nez Perce objects. In
Macaulay, Monica, Gensler, Orin D., Brugman, Claudia, Civkulis, Inese,
Dahlstrom, Amy, Krile, Katherine, and Sturm, Rob (eds.) Proceedings of
the Eighth Annual Meeting of the Berkeley Linguistics Society, University
of California, Berkeley: Berkeley Linguistics Society, pp. 463483
Rude, Noel 1986. Topicality, transitivity, and the direct object in Nez Perce.
International Journal of American Linguistics 52(2):124153
Saeed, John I. 1984. The syntax of topic and focus in Somali. Hamburg:
Helmut Buske
Saeed, John I. 1987. Somali Reference Grammar. Wheaton, MD: Dunwoody
Sag, Ivan A., Wasow, Thomas, and Bender, Emily 2003. Syntactic Theory: A
Formal Introduction. Stanford, CA: CSLI Publications
Sasse, Hans-Jurgen 1984. The pragmatics of noun incorporation in Eastern
Cushitic. In Plank, Frans (ed.) Objects: Towards a theory of grammatical
relations, London: Academic Press, pp. 243268
Schulz, B. 2005. Tune in drop out: Harmonic alignment as a prerequisite
for German topic drop. Cited by Erteschik-Shir (2007:24)

REFERENCES
Schwarzschild, Roger 1997. Why some foci must associate.
University

MS,

251
Rutgers

Seidl, Amanda and Dimitriadis, Alexis 1997. The discourse function of object
marking in Swahili. In Papers from the Thirty-Third Regional Meeting of
the Chicago Linguistic Society: Panels on Linguistic Ideologies in Contact,
University of Chicago: Chicago Linguistic Society, pp. 373387
Siewierska, Anna 2004. Person. Cambridge, UK: Cambridge University Press
Silverstein, Michael 1976. Hierarchy of features and ergativity. In Dixon,
R.M.W. (ed.) Grammatical Categories in Australian Languages, Canberra:
Australian Institute for Aboriginal Studies, pp. 112171
Simpson, Jane 1991. Warlpiri Morpho-Syntax: A Lexicalist Approach. Dordrecht: Kluwer Academic Publishers
Singh, Mona 1994. Thematic roles, word order, and definiteness. In Butt,
Miriam, King, Tracy Holloway, and Ramchand, Gillian (eds.) Theoretical
perspectives on word order in South Asian languages, Stanford, CA: CSLI
Publications, pp. 217235
Skribnik, Elena 2001. Pragmatic structuring in Northern Mansi. In Seilenthal, Tonu (ed.) Congressus nonus internationalis fenno-ugristarum. 7.13.8.2000 Tartu. Pars IV. Dissertationes sectionum: Linguistica III, Tartu:
Tartu University, pp. 222239
Sportiche, Dominique 1995. Clitic constructions. In Rooryck, Johan and
Zaring, Laurie (eds.) Phrase Structure and the Lexicon, Dordrecht: Kluwer
Academic Publishers, pp. 213276
Steedman, Mark J. 2001. The Syntactic Process. Cambridge, MA: The MIT
Press
Stowell, Tim, Wehrli, Eric, and Anderson, Stephen R. (eds.) 1992. Syntax and
Semantics: Syntax and the Lexicon, volume 26. San Diego: Academic Press
Strawson, Paul F. 1964. Identifying reference and truth-values. In Steinberg,
Danny D. and Jakobovitz, Leon A. (eds.) Semantics: An interdisciplinary
reader, Cambridge, UK: Cambridge University Press, pp. 8699
Strawson, Paul F. 1974. Subject and predicate in logic and grammar. London:
Ashgate

252

REFERENCES

Stump, Gregory T. and Yadav, Ramawatar 1988. Maithili verb agreement and
the Control Agreement Principle. In Papers from the 24th Regional Meeting of the Chicago Linguistic Society, Part 2: Parasession on Agreement in
Grammatical Theory, volume 24, pp. 304321
Szabolcsi, Anna 1981. The semantics of topic-focus articulation. In Groenendijk, Jeroen, Janssen, Theo, and Stokhof, Martin (eds.) Formal Methods
in the Study of Language, Amsterdam: Mathematical Centre, pp. 513540
Tereshchenko, Natalja 1979. Nganasanskij jazyk [The Nganasan language].
Leningrad: Nauka
Toldova, Svetlana and Serdobolskaya, Natalia 2008. Differential object marking in Uralic languages: a multiparameter analysis. Paper presented at Cognitive and Functional Perspectives on Dynamic Tendencies in Languages,
University of Tartu, Estonia, May 29June 1, 2008
Torrego, Esther 1998. The dependencies of objects. Cambridge, MA: The MIT
Press.
Vallduv, Enric 1992. The Informational Component. New York: Garland
Vallduv, Enric and Engdahl, Elisabet 1996. The linguistic realization of information packaging. Linguistics 34:459519
van Geenhoven, Veerle 1998. Semantic incorporation and indefinite descriptions: Semantic and syntactic aspects of noun incorporation in West Greenlandic. Stanford, CA: CSLI Publications
Van Valin, Robert D. 2003. Exploring the Syntax-Semantics Interface. Cambridge, UK: Cambridge University Press
Vincent, Nigel 1997. LFG as a model of syntactic change. In Butt, Miriam
and King, Tracy Holloway (eds.) Time over Matter: Diachronic Perspectives on Morphosyntax, Stanford, CA: CSLI Publications, pp. 142
von Heusinger, Klaus and Kornfilt, Jaklin 2005. The case of the direct object
in Turkish: Semantics, syntax and morphology. Turkic Languages 3:344
von Stechow, Arnim 1982. Structured propositions. Report of the Konstanz SFB. URL http://www2.sfs.uni-tuebingen.de/arnim10/
Aufsaetze/Structured%20Prop%201.pdf

Windfuhr, Gernot 1979. Persian grammar: History and state of its study.
Trends in linguistics. The Hague: Mouton de Gruyter

REFERENCES

253

Woolford, Ellen 1999. Animacy hierarchy effects on object agreement. In

Kotey, Paul (ed.) New dimensions in African linguistics and languages
(Trends in African Linguistics 3), Trenton: Africa World Press, pp. 203
216
Woolford, Ellen 2000. Object agreement in Palauan: Specificity, humanness,
economy and optimality. In Paul, Ileana, Phillips, Vivianne, and Travis,
Lisa (eds.) Formal issues in Austronesian linguistics, Dordrecht: Kluwer
Academic Publishers, pp. 215245
Woolford, Ellen 2001. Conditions on object agreement in Ruwund (Bantu).
In Benedicto, E. (ed.) The UMass volume on indigenous languages (University of Massachusetts Occasional Papers in Linguistics 20), Amherst,
MA: GLSA. URL http://people.umass.edu/ellenw/Woolford%
20Conditions%20on%20Object%20Agreement.pdf

Xelimskij, Evgenij A. 1982. Drevnejshie vengersko-samodijskie jazykovye

paralleli [Ancient linguistic parallels between Hungarian and Samoyed].
Moscow: Nauka
Yamamoto, Mutsumi 1999. Animacy and reference. Amsterdam: John Benjamins
Yang, Ning and van Bergen, Geertje 2007. Scrambled objects and case marking in Mandarin Chinese. Lingua 117:16171635
Zaenen, Annie, Maling, Joan, and Thrainsson, H. 1985. Case and grammatical
functions: The Icelandic passive. Natural Language and Linguistic Theory
3(4):441483. Reprinted in Maling and Zaenen (1990:95164)
Zwart, Jan-Wouter 2009. Uncharted territory? Towards a non-cartographic
account of Germanic syntax. In Alexiadou, Artemis, Hankamer, Jorge,
McFadden, Thomas, Nuger, Justing, and Schafer, Florian (eds.) Advances
in Comparative Germanic Syntax, Amsterdam: John Benjamins