You are on page 1of 257

Information Structure


general editor: Robert D. Van Valin, Jr., State University of New York, Buffalo
advisory editors: Guglielmo Cinque, University of Venice; Daniel Everett,
Illinois State University; Adele Goldberg, Princeton University; Kees Hengeveld,
University of Amsterdam; Caroline Heycock, University of Edinburgh; David
Pesetsky, MIT; Ian Roberts, University of Cambridge; Masayoshi Shibatani,
Rice University; Andrew Spencer, University of Essex; Tom Wasow, Stanford
1. Grammatical Relations
Patrick Farrell
2. Morphosyntactic Change
Olga Fischer
3. Information Structure: The SyntaxDiscourse Interface
Nomi Erteschik-Shir
In Preparation
The Acquisition of Syntax and Morphology
Shanley Allen and Heike Behrens
The Processing of Syntax and Morphology
Ina Bornkessel and Matthias Schlesewesky
Phrase Structure
Andrew Carnie
Morphology and the Lexicon
Daniel Everett
The PhonologyMorphology Interface
Sharon Inkelas
The SyntaxSemantics Interface
Jean-Pierre Koenig
Syntactic Categories
Gisa Rauh
Computational Approaches to Syntax and Morphology
Brian Roark and Richard Sproat
Language Universals and Universal Grammar
Anna Siewierska
Argument Structure: The SyntaxLexicon Interface
Stephen Weschler
The SyntaxDiscourse Interface
Great Clarendon Street, Oxford ox: odp
Oxford University Press is a department of the University of Oxford.
It furthers the Universitys objective of excellence in research, scholarship, and education by publish-
ing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne
Mexico City Nairobi New Delhi Shanghai Taipei Toronto
With ofces in
Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan
Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
Nomi Erteschik-Shir :oo,
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published :oo,
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or trans-
mitted, in any form or by any means, without the prior permission in writing of Oxford University
Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights
organization. Enquiries concerning reproduction outside the scope of the above should be sent to the
Rights Department, Oxford University Press, at the address above
You must not circulate this book in any other binding or cover and you must impose the same condi-
tion on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd.,
ISBN ,,8oI,,:o:,88 (Hbk.)
,,8oI,,:o:,,, (Pbk.)
I , , , , Io 8 o :
Acknowledgements viii
I. Introduction I
:. Architectures and Information Structure Inventory ,
:.I. Inventory: topics ,
:.I.I. Danish topicalization ,
:.I.:. Catalan: Links and Tails Io
:.I.,. Topics and truth values I,
:.I.. Stage topics Io
:.I.,. Permanently available topics I,
:.I.o. Topic tests I,
:.I.,. Multiple topics ::
:.I.8. Topic drop :,
:.I.,. Scope and topic properties :,
:.:. Inventory: foci :,
:.:.I. Semantic denitions :,
:.:.:. Marking foci by stress ,o
:.:.,. Deriving foci from stress ,:
:.:.. Deriving stress from syntax ,
:.:.,. Deriving syntax from IS ,,
:.:.o. Deriving focus pragmatically ,8
:.:.,. Summing up focus properties o
:.,. Putting it all together: f(ocus)-structure :
:.,.I. Topicfocus interactions ,
:.,.:. Notation ,
:.,.,. Contrastive and restrictive topics and foci 8
:.,.. Indenite topics ,I
:.,.,. Some answers ,,
:.. IS architecture ,,
:..I. Stylistic components ,,
:..:. S-structure ,8
:..,. IS at PF oo
:... Functional features o,
:..,. Lexical features o,
:..o. Multiply-dimensional architectures oo
:.,. Functionalism ,:
:.,.I. Functionalism vs. formalism ,:
:.,.:. Functionalist methodology ,8
,. Congurations 8o
,.I. Congurational languages 8o
,.:. The left periphery 8o
,.:.I. Cartography 8,
,.:.:. An even ner structure of the left periphery ,,
,.,. The importance of information structure IoI
,.,.I. Distinguishing topicalization from left
dislocation in English Io,
,.,.:. Focus preposing and Yiddish movement in English IIo
,.,.,. When the left-peripheral element does not have
information-structural impact II:
,.,.. Fronting non-topics II,
,.,.,. Focusing the subject: existentials II,
,.,.o. Clefts: syntactic mapping of information structure I:I
,.,.,. Other strategies for marking subjects as non-topics I:I
,.,.8. Left-peripheral prospects I:,
,.. Scrambling I:
,..I. Scrambling in Russian I:,
,..I.I. Syntactic topic and focus features and Russian
word order I,o
,..:. Scrambling in Japanese I,I
,..,. Dutch scrambling I,o
,... Scrambling in Persian I,
,.,. Scandinavian object shift I,
,.o. Concluding remarks on word order I,:
. Information Structure Constraints I,
.I. Identication I,
.I.I. Argument identication, a constraint on topicalization I,
.I.:. Identication of IS functions I,,
vi contents
.:. IS constraints on syntax I,,
.:.I. I(denticational)- dependencies I,,
.:.:. Canonical f-structures Ioo
.:.,. The constraint on I-dependencies Ioo
.:.. Wh-topics Io8
.:.,. Topic-islands Io,
.:.o. That-t effects I,I
.,. Superiority I,
.,.I. Superiority in other languages I,8
.,.I.I. Superiority in Hebrew I,,
.,.I.:. Superiority in German I8I
.,.I.,. Superiority in Danish I8,
.. IS constraints on complex NP I8,
..I. Extraposition from NP I8,
..:. Extraction from NP I8,
.,. Processing I8,
.o. Architectural consequences I,,
,. Aspectual Focus I,
,.I. The theory of atoms I,
,.I.I. Aspectual focus I,o
,.:. Meaning components and extraction I,,
,.:.I. Manner-of-speaking verbs I,,
,.:.:. Picture NPs I,8
,.:.,. Datives :oo
,.,. Missing objects :o,
,.,.I. Missing objects in Activities :oo
,.,.:. Object omission in habituals :o8
,.. Contextual binding of lexical constituents :Io
o. The Division of Labor between Syntax and IS :I,
References :I,
Index of terms :,,
Language Index ::
Index of authors :,
contents vii
I would rst like to thank Robert Van Valin both for giving me the oppor-
tunity to write this book and for his constructive comments encouraging
me to broaden the books perspective. I am also grateful to John Davey at
OUP for his encouragement while the book was in the making.
The book has also beneted greatly from discussions with my col-
leagues and graduate students: Rachel Eitan, Anita Mittwoch, Tova
Rapoport, Soe Raviv, Liron Shokty, and especially Lisa Rochman
and Natalia Strahov who also combed through the manuscript in its
I used the draft of this book as a text for a graduate course in Linguistic
Methodology at Ben-Gurion University in the fall semester of . I
am happy to express my gratitude to the students of this course for their
helpful comments, questions, and requests for clarication.
Finally, I thank my family and friends for allowing me to put so much
on hold for so long.
Support for this research was provided in part by grant No. / from
the Israel Science Foundation.
General Preface
Oxford Surveys in Syntax and Morphology provides overviews of the
major approaches to subjects and questions at the centre of linguistic
research in morphology and syntax. The volumes are accessible, criti-
cal, and up-to-date. Individually and collectively they aim to reveal the
fields intellectual history and theoretical diversity. Each book pub-
lished in the series will characteristically contain: () a brief historical
overview of relevant research in the subject; () a critical presentation
of approaches from relevant (but usually seen as competing) theoretical
perspectives to the phenomena and issues at hand, including an objec-
tive evaluation of the strengths and weaknesses of each approach to the
central problems and issues; () a balanced account of the current issues,
problems, and opportunities relating to the topic, showing the degree
of consensus or otherwise in each case. The volumes will thus provide
researchers and graduate students concerned with syntax, morphology,
and related aspects of semantics with a vital source of information and
Nomi Erteschik-Shirs Information Structure: The Syntax-Discourse
Interface explores the interplay of discourse-pragmatics and morpho-
syntax. Beginning with the Prague School Professor Erteschik-Shir
reviews work on the nature of this interaction within different frame-
works up to the present, and considers its wider implications for the
architectures of linguistic theory.
Robert D. Van Valin, Jr
General Editor
Heinrich Heine University, Dsseldorf
University at Buffalo, The State University of New York
In memory of my mother, Trudi Erteschik,
The possibility of optional divergence from basic word order is a univer-
sal property of natural language. While some approaches have tried to
account for this in purely syntactic terms, others have provided purely
pragmatic explanations, and yet others hold mixed views, which allow
for semantic or pragmatic triggers to syntactic operations.
There is a general consensus that the linear order of constituents is,
at least to some extent, determined by notions having to do with what
is contextually known and what is not. The denition of these notions,
and their syntactic implementation, has been the topic of much research
since they were rst introduced by the Prague School in the late s.
Another point of consensus is that intonation and morpho-lexical
elements interact with word order to determine the information struc-
ture (IS) of the sentence, a term originally introduced by Halliday to
account for the distinctions of focus, presupposition, and propositional
attitude toward entities in the discourse conveyed by phrasal intona-
tion. This consensus extends to the status of questionanswer pairs with
respect to information structure. In (), the constituent that answers a
wh-question is the focus (in upper case) and a constituent already intro-
duced in the question is the topic of the answer (in italics):
() a Q: What did John do?
b Q: What did John wash?
A: He washed THE DISHES.
Not all questionanswer pairs are as easily analyzed as shown in ():
() a Q: Who washed the dishes?
A: JOHN washed them.
b Q: What happened to the dishes?
c Q: What happened?
In (:a) the predicate is presupposed. In some approaches the object (the
dishes) is analyzed as the topic, in others the whole predicate is viewed
as a unit. (:b) illustrates a case in which the focus is not a syntactic con-
stituent and the all-focus sentence (:c) has no apparent topic. It is there-
fore not obvious whether there is a mapping from syntax to information
structure, and if there is, how to account for the fact that topics and foci
cannot always be mapped onto constituents.
Examples such as those in (:) provoke a variety of dierent accounts
with dierent basic information structure concepts. The Prague School
(Functional Sentence Perspective, e.g., Firbas I,o:; I,o) employs the
gradient notion, Communicative Dynamism (CD), to account for
information structure. The Prague School argued that Communicative
Dynamism (CD) determines the linear arrangement of syntactic con-
stituents within sentences. The degree of CD of a sentence element is
the extent to which it pushes the communication forward and the ele-
ments with least CD (those that are contextually known) precede those
that have more CD (those that convey new, unknown information). A
special status is assigned to the element with the least CD, the theme
or topic. The remnant, the rheme, can be viewed as predicated of the
topic. In more recent work in the Prague School (e.g., Sgall, Hajiov,
and Panevov I,8o), an analysis in terms of Topic Focus Articulation
(TFA) is adopted in which the Topic is generally contextually bound and
the Focus is not. (For a full historical review see Newmeyer :ooI.)
Although there is general agreement that the basic information
structure units not only play a role in determining word order but also
interact with a variety of other grammatical phenomena, such as inter-
pretation, intonation, morphology, and other syntactic phenomena, not
everyone agrees on the status of these same units with respect to deter-
mining the coherence of texts, beyond the cohesion of the questionanswer
pairs in (I) and (:). Linguists working within the Prague School frame-
work were not only concerned with dening the contextual notions that
determine word order within the sentence, they also showed that these
same notions play a role in determining the coherence of sequences of
sentences. According to Dane I,, there are three types of thematic pro-
gression. (I) Keeping the theme constant, (:) deriving the theme from
the previous rheme, and (,) deriving the theme from a hypertheme.
The rst type of thematic progression, which I refer to as topic chaining,
is illustrated in (,).
: introduction
(,) John likes to read. He is intelligent and industrious and will go far.
In the second sentence in the sequence, the pronoun he stands for John,
the topic of the rst sentence. The topic is thus maintained from the
rst sentence to the second. A new topic can be derived from a focus of
the previous sentence. This is the second type of thematic progression
(which I refer to as focus chaining) illustrated in ().
() a Theres a girl in the class who the teacher likes. She answered
all the questions the teacher asked.
b Fortunately someone knew enough French to help us. He even
lent us his map.
In (a), the indenite, a girl, is focused in the rst sentence and provides
a topic for the second. Similarly, in (b), in which the object French pro-
vides the topic, someone is focused and enables the coreferent pronoun
in the following sentence to play the role of topic.
Danes third type of thematic progression involves derived topics.
These topics are derived from a hypertheme introduced in the previ-
ous sentence. A hypertheme involves a discoursally restricted set of ele-
ments. These can be listed in the previous discourse as in (,a) or they can
be introduced by a term which denes such a set as in (,b).
(,) a Ill tell you about my friends, John, Paul, and Mary. John is an
old friend from school, Paul, I met at college, and Mary is a
colleague at work.
b Speaker A: Tell me about your family.
Speaker B: My mother is a teacher, my father works in an
oce, and my sister is a student.
In (,a), a list of three friends is introduced. Each individual within this
set provides a possible topic for the next sentences. In (,b) each topic
in speaker Bs sentences belongs to the set of family members. Neither
one needs previous mention on its own, because speakers are aware that
families are made up of parents, siblings, etc. Therefore, these family
parts are potential future topics. (See also Erteschik-Shir I,88.)
Dierent terms have been used for IS since Halliday, each with its
accompanying approach: Chafe I,,o coined the term packaging to
refer to a speakers use of a particular syntactic form to serve a particu-
lar pragmatic function, or, in his words, to accommodate states of the
addressees mind. Prince I,8o: :o8 puts it more succinctly: speakers
seem to form their utterances so as to structure the information they
are attempting to convey, usually or perhaps always in accordance with
introduction ,
their beliefs about the hearer: what s/he is thought to know, what s/he is
expected to be thinking about. Vallduv I,,o uses the term information
packaging. Other terms are topic-focus-structure, or focus structure as
in Erteschik-Shir I,,,.
Within modern linguistic theories, the place of information
structure in grammar is far from settled. Within the Principles and
Parameters theories (GB/MP), information structure is generally rel-
egated to the peripheries, making it hard to express its central role with
respect to syntax, semantics, and intonation in a systematic way. These
issues will be discussed in detail in Chapter :.
The architectures of multi-layered theories are better suited to
incorporate information structure due to their parallel correspon-
dence architecture, where phonological, syntactic, semantic, and
(some) pragmatic representations are computed in parallel. These
theories include LFG (e.g., Vallduv and Engdahl I,,,), which uses
both typed feature structures (including information structure fea-
tures) and trees, with a mapping relation between the two modes of
representation; HPSG (e.g., Pollard and Sag I,8,; Sag and Ginzburg
:ooo), which uses feature structure sharing; and CCG (Steedman
I,,o; :ooob) which also analyzes linguistic objects in terms of partial
information structures which mutually constrain possible colloca-
tions of phonological structure, syntactic structure, semantic content,
and contextual factors in actual linguistic situations. Within these for-
malisms, the status of information structure is therefore equal to that
of the other components of grammar. Steedman :ooob, for example,
argues that the Surface Syntax acts as a transparent interface between
the spoken form of the language, including prosodic structure and
intonational phrasing, and a compositional semantics, including log-
ical form and information structure. Steedmans constituents do not
correspond to traditional syntactic constituents. Yet he argues that it
is the interpretations of these constituents that are related to distinc-
tions of Information Structure and discourse focus among the topics
the speaker has in mind and the comments that the speaker has to
contribute (Io,).
Role and Reference Grammar (Van Valin I,,,b; I,,,b; Van Valin and
LaPolla I,,,) employs an architecture in which grammatical struc-
tures are stored as constructional templates, each with a specic set of
morphosyntactic, semantic, and pragmatic properties (Van Valin and
LaPolla I,,,: ,,) naturally allowing for a linking between the three.
A good illustration of how this works is the selection of a privileged
syntactic argument (pragmatic pivot), a construction-specic notion
which diers across languages. The distinction between subject-promi-
nent and topic-prominent languages is argued to be due to their choice
of pivot: topic-prominent languages select the topic as the privileged
argument; subject-prominent languages select the argument with the
most prominent thematic role. RRG thus predicts these word order dis-
tinctions across language types.
Functional theories focus on information structure and what it
contributes to grammar. Examples are Functional Syntax (Kuno ;
Kuno and Takami ) and Functional Grammar (Dik a). Kuno
has, over the years, argued for the need for a large number of functional
constraints in syntax, showing that proposed syntactic constraints
cannot account for all of the relevant data. He has not, however, pro-
posed a general theory of the interface between his functional theory
and syntax. Diks Functional Grammar does oer a well-formulated
functional theory. His syntax is, however, very limited, consisting only
of a series of placement rules.
Chapter examines the basic information structure notions, topic
and focus and their properties, from a variety of theoretical perspectives.
Following the establishment of the basic information structure termi-
nology, various approaches to the architecture of information structure
are critically discussed.
Once the theories of information structure have been outlined, we
proceed in Chapter to look in detail at a number of syntactic phenom-
ena that have been analyzed in terms of information structure within
the various frameworks described above. These are grouped into types
of phenomena: We rst examine syntactic congurations that have been
associated with particular information structures, starting with congu-
rational languages that allocate particular syntactic positions to topics
and foci. Here we address the relation between syntactic representations
of topic and focus in these languages and the corresponding pragmatic
interpretations of these structures. We continue to examine languages
that do not employ such strategies in general, but have a number of con-
gurations which have been associated with a particular assignment of
either topic, focus, or both. These consist of frontings of various kinds,
cleft constructions, and existentials. After rst concentrating on the left
periphery, the chapter continues to examine scrambling in the sentence
as a whole.
Chapter deals with information structure-motivated syntactic
constraints including constraints on extraction.
Chapter examines the lexiconinformation structure interface.
Here the notion of aspectual focus is introduced within the theory of
atoms, a theory of lexical decomposition. Aspectual focus is shown to
interact with extraction facts, completeness facts, missing objects, and
an account is also oered for the special intonation assigned to eventive
The book concludes with a brief examination of the division of labor
between syntax and information structure and in particular whether
information structure belongs in the grammar or not. These issues are
discussed from the perspective of general cognition and also through
the insights gained from language acquisition research.
In view of the massive amount of literature on information structure
within the various approaches, the book does not attempt full cover-
age of these sources, nor does it aim to cover all the topics that have
received IS-theoretical analyses. Instead, the book is meant to introduce
the reader to IS methodology with an emphasis on both descriptive and
theoretical rigor.
Architectures and Information
Structure Inventory
One of the connections between information structure and syntax that
is best known is the propensity for languages to order given, old, or topi-
cal information before new or focused information, and linguists from
a variety of linguistic schools have attempted accounts of this phenom-
Although the phenomenon is well established, it has not been a
simple task to account for it. One reason is the fact that this is only a ten-
dency: New information can, under certain circumstances, precede old
information, and the cross-linguistic circumstances in which this occurs
are rather evasive. Another reason this phenomenon is hard to explain is
that linguists nd it hard to agree on the concepts involved in the gener-
alization: What exactly is a topic, and what exactly is a focus? Sections and
are dedicated to these two concepts and it is concluded that topic and
focus are the only information structure primitives needed to account
for all information structure phenomena. Section demonstrates that
topic and focus interact and the theory of f(ocus)-structure which is
based on this interaction is introduced. Next, various conceptions of how
information structure ts into the architecture of grammar are exam-
ined (section ). The chapter concludes with a section on information
structure methodology (section ).
2.1 Inventory: topics
2.1.1 Danish topicalization
Topicalization is a case of dislocation to the left periphery as shown in
the Danish examples in ():
() a Hun hilste p Ole. Ham havde hun ikke mdt fr
She greeted Ole. Him had she not met before
In addition to the Prague School linguists mentioned in Chapter , see among others
Kuno , Prince a, and Birner and Ward and references cited in these works.
architectures and information structure
b Hun hilste p Ole. Hun havde ikke mdt ham fr
She greeted Ole. She had not met him before
In Danish, topicalization is motivated by the need to overtly mark a
topic. Topics must also be old, i.e., they must appear in the preceding
context, and in this sense topics are identied by the context. That is why
moving the topic is optional as shown in (b).

The examples in () show that pronouns, denites, specic indenites,
generics, and contrastive elements qualify as topics but that non-specic,
non-contrastive indenites do not.
() a Hende mdte jeg i gr.
her met I yesterday
I met her yesterday.
b Pigen mdte jeg i gr.
the girl met I yesterday
I met the girl yesterday.
c Blomster ser man om forret.
owers sees one in the spring
One sees owers in the spring.
d Pigen mdte jeg i gr, drengen mdte jeg frst i dag.
the girl met I yesterday, the boy met I only today
I met the girl yesterday, I met the boy only today.
e En pige som jeg mdte i gr 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.
f *En pige mdte jeg i gr.
a girl met I yesterday
I met a girl yesterday.
If Danish topicalization provides a good diagnostic of topichood,
then all DPs, except for non-specic indenites, qualify as topics. One of
the properties of such indenites is that they cannot occur in a discourse
in which they have been mentioned before:
() A girl appeared on TV. *a girl/the girl/she was talented.
The sequence of sentences in () with the indenite cannot be interpreted
with the two indenites referring to the same girl. In order for such an
Languages may differ in this respect. In English topicalization is rarely employed. In
Hungarian, topicalization is strongly preferred.
architectures and information structure
interpretation to be possible, a denite or a pronoun must occur on second
mention. It follows that an indenite must be new, that is, not mentioned
in the preceding discourse. Denites and pronouns are necessarily given
or old. This property is therefore a good candidate for the denition of a
topic. What about generics, contrast, and specic indenites, the other DP
types that qualify for topichood? Are they old or new? Generics need not be
old (as do pronouns and denites), but topicalized generics must be. (c)
cannot be uttered out of context and must be preceded by some reference
to owers in the preceding context. Contrast presents a slightly more
complicated picture. Even (non-specic) indenites can be topicalized if
they are contrasted, as shown in ().
() Et museum besgte jeg allerede i gr,
A museum visited I already yesterday,
en kirke ser jeg frst i morgen.
a church see I only tomorrow.
I visited a museum already yesterday, I will see a church only
Yet contrast cannot occur out of the blue; it requires a discoursally avail-
able contrast set. () can, for example, be uttered in a context in which
the speaker (a tourist, say) is asked whether she or he has already been
to a museum and a church. What is old in this case is not a particular
museum or church, but rather the set including both elements.
Specic indenites are harder to classify as old. They do not require
any mention in the preceding discourse. Even so, according to the
Danish topicalization diagnostic, they belong to the same class as the
others. What makes the indenite specic in (e) is the modifying rel-
ative clause. It is this modication which licenses topicalization, since
without it, topicalization is illicit (f). In order to see how this works
compare (e) to ().
() Jeg mdte en pige i gr. Hende gav jeg en god bog.
I met a girl yesterday. Her gave I a good book.
I met a girl yesterday. I gave her a good book.
If the content of the relative clause were uttered as a separate preceding
sentence, then the girl would in fact be old by the time the main clause
was uttered. I claim that this is how the relative clause functions: it intro-
duces the girl as the one I met yesterday, and thereby makes her old with
respect to the main clause. A more formal account of this idea can be
found in Erteschik-Shir and below in section ..
architectures and information structure
I conclude that, in this extended sense, all the elements that topicalize
in Danish are in fact old or given. It is only because Danish topicalization
was used as a diagnostic of topichood that we reached the conclusion
that specic indenites should be classied with the other topicalized
elements as being old. Another diagnostic might have led to a dierent
classication. The next section discusses a dierent view of what topic-
hood might be.
2.1.2 Catalan: Links and Tails
Vallduv : distinguishes two kinds of old information,
Links, which are left detached (or topicalized) in Catalan, and Tails,
which are detached to the right. He lists the following examples of left
() a [written on an aerogram; rst line on the extra space overleaf]
ja no hi
comptava t
with-this-little-piece-of-paper anymore no obl s-impf-count-on
This-little piece-of-paper I wasnt counting on anymore.
b Quant al Joan i la Isidora no tho s dir,
as-for the J. and the I. no iobj.obj s-know to-say
doncs el Joan
veiem t
ben poc.
since the J. obj p-see quite little
As for Joan and Isidora I cant say, since Joan we see very little of.
c [after mentioning something nasty that the hearer had done
to the speaker a long time ago]
tinc clavat t
al fons del cor.
this obj have pstppl-stick at.the depth of.the heart
Lit.: This I have it stuck deep in my heart.
This I wont forget how it hurt.
Catalan topicalization has in common with Danish topicalization
that the topicalized element must be old, yet, according to Vallduv,
there is a dierence. In Catalan only elements that are not the topic
of the previous sentence are topicalized. Such new topics are called
switch topics, or shifted topics.
Vallduv calls these fronted elements
Note that a characteristic of Catalan topicalization is that it leaves a coreferent clitic pro-
noun in the sentence. The Links are underlined, the Tails are in capitals. Obl=oblique, Obj =
object, iobj = indirect object, pstppl = past participle.
There are also two kinds of topics in Mayan languages according to Aissen : CP-
external topics which must be shifted topics and CP-internal topics which can be continued
topics; but the correlation isnt strict.
architectures and information structure
Links and views them as address pointers in a le system in which
new information is listed under the address specied by the Link. Since
a Link is a command to go to an existing address, Links only appear
when there is a change of address. This is why continued topics are not
fronted. In the examples in (), the fronted (underlined) elements are all
new topics in this sense. The Link in (a) is made contextually available,
the ones in (b) and (c) are mentioned in the previous sentences, yet
do not function as topics there. It follows that topic chaining (as intro-
duced in Chapter ) will not allow for a topicalized Link, whereas focus
chaining will, since in the latter case the focus of the preceding sentence
becomes the new topic of the following one. Finally, topics derived from
hyperthemes are also new topics as illustrated in () from Villalba .
() A: On va posar les coses?
where past- put the things
Where did (s)he put the things?
B: Em sembla que els llibres, els
To-me seems that the books them-masc
va posar al despatx.
past- put in-the study
It seems to me that (s)he put the books in the study.
In this example, the antecedent for the Link is the things in As question.
In order for the questionanswer sequence to be acceptable, it must be
contextually understood that the books belong to the set of things in the
question. Introducing this set in the question conjures up all its elements,
each of which is then considered to be given and can potentially provide
a Link in the following sentence.
A Link can in this way be both new and
given at the same time. What is newly introduced is the particular mem-
ber of the set in question; what is given is the set that includes it.
In view of the fact that the conditions on Danish and Catalan topical-
ization dier somewhat a question arises: Does the same notion of topic
play a role in both languages (and in languages in general), or does each
language have a dierent array of IS functions? This is the topic of this sec-
tion. Let us therefore compare the two cases more carefully. In Danish top-
icalization is possible in all the cases in which left-detached Links appear
in Catalan, yet continued topics can also be fronted. This is shown in ().
The relation between such a set and its members has been referred to as a poset (partially
ordered set) relation by Hirschberg ; Ward ; Ward and Hirschberg ; Ward and
Prince . See Chapter , section . for further discussion.
architectures and information structure
() A: Hans kan jeg godt lide.
Hans can I good like
I like Hans a lot.
B: (i) Ham kan jeg ogs lide.
him can I also like
I like him too.
(ii) Jeg kan ogs lide ham.
I can also like him
I like him too.
In the context of speaker As sentence, the fronted pronoun in Bs
response in (i) is a continued topic. True, B could also have responded
to A with (ii), yet it would be a less natural response.
Even so, Germanic
languages, including Danish, tend to topicalize shift topics (see Chapter ,
section .) more than continued ones, on a par with Catalan, yet
Germanic topicalization is not as rigidly constrained.
The point of
this comparison is to show that a linguist studying Danish rather than
Catalan may reach quite dierent conclusions as to what a topic is by
generalizing over the elements that topicalize in each language. Clearly,
the optimal answer to the question posed above is that the denition
of topics should be universal. So far, we can say that topics must be old
or given, in the extended sense of these notions as discussed here. Yet
some way has to be found to distinguish shift topics from continued
topics in order to capture the distinctions between topicalization in
Danish and Catalan.
In Catalan given elements, which do not qualify as Links, are post-
posed. Vallduv refers to such postposed elements as Tails.
to Vallduv, both Links and Tails belong to the presupposition or the
ground of the sentence, but Tails must have an antecedent in the preceding
In both (i) and (ii) both arguments are pronominal continued topics. In (i) both argu-
ments are also in canonical topic position: the topicalized position and the subject position.
These positions and the topics which go in them are discussed in Chapter , section .
One other difference between topicalization in the two languages is that in Danish pro-
nouns topicalize as seen in (a). This is not possible in Catalan. Pronouns are old par excel-
lence, yet they require an antecedent in the discourse, and although this antecedent need
not be a topic in the preceding sentence, they still do not prepose in Catalan unless they are
contrastive, in which case the contrast set functions as a hypertheme.
There is a certain overlap between what can be fronted and what can be postposed in
Catalan, as far as I can tell. Links exclude continued topics, Tails can be both continued top-
ics and elements which have been mentioned previously as foci (focus chaining). The latter
also qualify as Links.
architectures and information structure
sentence as shown in () from Villalba which should be compared
to () above to see the dierence between Links and Tails.
() A: On va posar les coses?
where past- put the things
Where did (s)he put the things?
B: #Em sembla que els va posar al
to-me seems that them-masc past- put in-the
despatx, els llibres.
study the books
It seems to me that (s)he put the books in the study.
Replacing the things by the books provides an antecedent for the right-
dislocated phrase and the sentence becomes good:
() A: On va posar els llibres?
where past- put the books
Where did (s)he put the books?
B: Em sembla que els va posar al
to-me seems that them-masc past- put in-the
despatx, els llibres.
study the books
It seems to me that (s)he put the books in the study.
These data provide further evidence for the need to distinguish the two
kinds of given information, the Link and the Tail, yet it would be advan-
tageous if both topic types could be derived from the same primitives.
The next section provides another potential property of topics that
should be taken into account in this endeavor.
2.1.3 Topics and truth values
According to Strawson : , the topic has three central properties:
(a) The topic is what a statement is about.
(b) The topic is used to invoke knowledge in the possession of an
(c) The statement is assessed as putative information about its topic.
The rst two properties are already familiar. Vallduvs Link is what
the statement is about since it is the address where new information
is entered. The new information is therefore about the referent of its
# signals a grammatical sentence which is infelicitous in the specied context.
architectures and information structure
Link. Concerning the second property, this is what we have referred to as
given or old. Strawson species more precisely that it is the audience or
the hearer for whom the information is old. We have, up to now, taken it
for granted that it is the hearer for whom the information must be given.
After all, communication is about imparting information to the hearer
and the speaker must be sensitive to what information is already avail-
able to the hearer and what is not.
Strawsons innovation is in the third property. He uses this prop-
erty together with the other two to explain why truth value gaps come
about. He argues that if the topic is what a statement is about and
therefore the statement is assessed with respect to the topic, the topic
must have reference. If, however, the topic has no reference, the sen-
tence cannot be evaluated as either true or false, since in such a case
a statement which, by hypothesis, is about something is really about
nothing. The following examples from Strawson illustrate this point:
() The King of France is bald.
() The exhibition was visited by the King of France.
In the context of the question in What is the King of France like?
() has no truth value since the sentence cannot be veried. It cannot be
about a non-existent king. In (), however, the exhibition is the topic.
This sentence will therefore be assessed as false if the King of France is
not among the visitors at the exhibition. It follows from the properties
Strawson assigns to topics that the topic is chosen in accordance with the
context. If () is uttered in the context of the question: What bald nota-
bles are there?, then () will be assessed as false, since in that context the
set of bald notables is the topic and just as the King of France would be
missing from the visitors at the exhibition, he will also be missing from
the set of bald notables. The context-dependent choice of topic thus
determines the assignment of truth values. This view of topichood has
been formalized in Reinhart who draws on the notion of a context
set la Stalnaker (): The context set of a given discourse at a given
point is the set of propositions which we accept to be true at this point
(Reinhart : ). Each new assertion, if not rejected as false, adds a
new proposition to the presuppositions in the context set. Reinhart sug-
gests that the context set has internal organization; in particular, propo-
sitions in the context set are classied by their topics. Sentence topics
thus determine under which entry a particular proposition is assessed.
Take for example:
architectures and information structure I,
(I,) A: Tell me about John.
A: Tell me about Mary.
B: John invited Mary to dance.
This sentence can be understood to be about John or about Mary since
it can naturally follow the requests made both by speaker A and by speaker
A. According to Reinhart, Bs sentence, in the context of A, will be assessed
as putative information about John and will be entered under the referen-
tial entry for John. However, in the context of A the same sentence will be
evaluated as information about Mary. Although the truth value of Bs sen-
tence will be the same in both contexts in this case, the way the sentence is
evaluated is dierent.
Reinhart does not adopt Strawsons second property of topics, namely
that topics must be in the possession of the hearer, that is they must be old
or given. There are two reasons for this: the rst one is that topics are some-
times indenite, and indenites are new. An example was given above in
(:e) and an initial explanation of why specic indenites do not contradict
the denition of topics as old was given in section I.I and will be discussed
further in section ,. below. The second problem with the idea that topics
are old, according to Reinhart I,8I: ,:, is that this requires that topichood
be dened directly on referents, or that the topic role of an expression can
be identied by checking properties of its referents. This, she shows, can-
not be the case in view of the fact that the same referent may function both
as the topic and as the focus in one sentence (Reinharts (,,) ):
(I) A: Who did Felix praise?
B: Felix praised HIMSELF.
Reinhart argues that if Felix is the topic in Bs response, then so is the ana-
phoric reexive since the referent of these two NPs is the same. But the
reexive answers the wh-question and therefore must be interpreted as a
focus. It follows that the reexive is both a topic and a focus, a plain contra-
diction, according to Reinhart. In Chapter , section :.I, this contradiction
is resolved. It is therefore possible to maintain Strawsons characterization
of topics including all three properties.
It follows from the denition of topics as the pivot for truth value assess-
ment, that all sentences must have a topic, since all sentences must be
assigned a truth value. This is the main reason why Vallduv I,,:, whose
Others who view topics as the subject of predication are Chafe I,,o and Li and
Thompson I,,o.
Io architectures and information structure
denition of Link is otherwise very similar to Reinharts Topic, rejects the
idea that Links are the locus of truth value assessment. Vallduvs system is
derived from the overt markings of Catalan syntax. Adopting the denition
of topics as the locus for truth value assignment would destroy the one-to-
one relationship between Catalan word order and information structure
in view of the existence of Linkless sentences. In Reinharts model, these
would be evaluated according to the continued topic, i.e., the element which
appears as the Tail in Catalan. More problematic are all focus sentences
which in Catalan have neither a Link nor a Tail. These will be discussed in
the next section.
2.1.4 Stage topics
The status of topics with respect to syntactic constituency comes up in par-
ticular in the context of thetic, or all-focus, sentences as in (I,) (repeated
from Chapter I).
(I,) Q: What happened?
A: John washed the dishes!
Such sentences seemingly have no topic.
Yet if we adopt the Strawsonian
denition of topics, according to which topics are the pivots for assess-
ment, then it is crucial that every sentence have one. Gundel I,, in fact
argues that such sentences do have a topic, namely the particular situa-
tion (time and place) about which it is asserted. Similarly, in Erteschik-
Shir I,,,, thetic sentences are viewed as having implicit stage topics
indicating the spatio-temporal parameters of the sentence (here-and-now
of the discourse). These are contextually dened. In (I,), for example, the
question refers to an event in the (immediate) past (due to the tense used)
and at the current location. The answer should therefore be evaluated with
respect to this spatio-temporal location. The sentence in the answer could
For discussions of the distinction between thetic or all-focus sentences which do not
have an overt topic and categorical sentences which do see among others: Basilico I,,8; :oo,;
van Finkel I,8,; Kuroda I,,:; Ladusaw I,,; Lambrecht :ooo; Rosengren I,,,; Sasse I,8,.
Lambrecht I,, adds a third type of IS: In addition to thetic sentences which have sentence
focus and categorical sentences which have predicate focus, there are also sentences with
argument focus:
(i) Q: I heard your motorcycle broke down?
A: My car broke down.
In this context the subject is contrastive, hence viewed as a focus by Lambrecht. Yet as
seen here, topics can also be contrastive. Therefore the IS of such sentences can in fact be
viewed as categorical with a contrastive subject topic.
architectures and information structure
in principle also be evaluated with respect to either the subject John or
the object the dishes, but such evaluation would contradict the context
specied by the question the sentence is supposed to answer. The follow-
ing two examples, however, can only be evaluated with respect to a stage
() Its snowing.
() Theres a cat outside the door.
How do you know whether () is true or false? You look outside to see.
Your outside is the here-and-now of the discourse. () is similar.
The current time and location again functions as the implicit topic with
respect to which the sentence is evaluated. Outside the door modies
this location, so that the sentence will be true only if there is a cat outside
the door. Compare () to () in which this PP is topicalized.
() Outside the door, theres a cat.
Here the fronted PP is an overt topic which species the location with
respect to which the sentence is to be evaluated. This overt stage topic
is insucient for purposes of evaluation. It is completed by the current
spatio-temporal parameters, which tell us where the door is located and
at which time the sentence is supposed to be true. The assessment of
() and () therefore proceeds in an identical fashion, even though the
stage topic of the former is implicit and the stage topic of the latter is
2.1.5 Permanently available topics
An important property of implicit stage topics is that they are always
available since a conversation always takes place at a specic time and in a
specic place. These parameters therefore function as a potential topic at
the beginning of a conversation, which is why a sentence such as () can
be uttered out of the blue. Such out-of-the-blue sentences, with implicit
stage topics, are not the only sentence type that can occur as conversation
openers. It is also always possible for the speaker and hearer to provide
Reinhart, in fact, speculates in a footnote that this may indeed be the case. For the inspi-
ration for this idea see Kratzer . Overt locative topics can behave like subjects in terms
of agreement. Bresnan and Kanerva , for example, show that this is the case in Bantu
languages. This is evidence that locative topics should be classied with non-locative topics.
The fact that temporal and locative adverbials topicalize in the same way as do arguments
also supports the existence of stage topics. See also Ladusaw ; Junghanns and Zybatow
; De Cat ; . Kiss , among others.
architectures and information structure
the topics of the rst sentence as illustrated in the common conversation
opener in ().
() Q: How are you?
A: I am ne.
In addition to this inventory of permanently available topics (stage top-
ics, speaker and hearer), a number of permanent and temporary xtures
of our world are also available as topics with no need for previous men-
tion. This type includes the president, the moon, the train, and
whatever other world xtures the speaker and hearer happen to have in
Examples are given in ():
() a Its dark. The moon has disappeared.
b I have to go. The train is leaving any minute.
There are also topics which are temporarily available without previous
mention. These are referents that exist or appear on the current scene and
may be introduced deictically. That chair is ugly, for example, can be
uttered out of the blue with that chair as its topic, if theres a particular
chair on the scene of the conversation which the speaker is pointing at.
What these permanently or temporarily available topics have in
common is that they can function as topics without being mentioned
previously. Therefore, they are not old. How does this gel with our gen-
eralization that topics must be old? Heres where the distinction between
old and given comes in. Old means that the referent has been mentioned
in the conversation; given, however, means that the hearer has the refer-
ent in mind or, in Strawsons words, knowledge in the possession of an
audience. We should therefore amend the generalization and conclude
that topics must be given.
We conclude that the knowledge the hearer must have in order to accept
a referent as a topic can come to him/her in various ways. In the case of
old topics, the referent must have been mentioned in the immediate dis-
course, or else it can be derived from a previously mentioned topic as in the
case of hyperthemes (see () above). In such cases the topic is a member of
a set dened by the previously mentioned hypertheme. A topic can also be
derived from a previously mentioned referent through world knowledge.
An example is given in ():
() John heard a beautiful concert. The composer directed it.
See Schaeffer to appear and the references therein for this notion.
architectures and information structure
Once a concert is introduced, its composer and other concert parts
come along with it due to world knowledge. Such parts of a previously
mentioned referent are therefore treated as given and become possible
topics of the following sentence.
Another example where accommodation is required stems from
Roberts : :
() John was murdered yesterday. The knife lay nearby.
According to Roberts, once we accommodate that John was killed by
stabbing, the common ground will entail the existence of a knife as the
murder weapon, licensing a weakly familiar discourse referent to satisfy
the familiarity of the knife. Since any murder weapon can be accom-
modated, it seems the process of accommodation is more general and
follows from the use of the verb murder which implies that a murder
weapon was employed. This is how the hypertheme murder weapon is
derived and since the knife is a member of the set of murder weapons,
its potential topichood is licensed. It follows that deriving hyperthemes
and their subparts is not always equally straightforward. Yet in all cases
the derivation links to available lexical or situational information.
2.1.6 Topic tests
The discussion of topics started with the underlying assumption that
Danish left-dislocated elements are topics. We concluded that topics
must be given. We then extended our discussion to Catalan, and dis-
covered that there, the elements on the left periphery are a subset of
the Danish ones since they exclude continued topics. Other languages
may have identied yet another set of elements with only partial over-
lap with those identied in either of these two languages. In order to
decide which elements should be dened as topics, we must devise an
independent way to identify them. Tests for topichood are available but
they obviously depend on the denition of topics adhered to. Here we
adopt the StrawsonReinhart approach. Topics are therefore what the
sentence is about; they are given; and the truth value of the sentence is
established with respect to them. (The rst two characterizations of top-
ics are relatively uncontroversial. The latter is more so.)
The rst property of topics renders the following two versions of the
aboutness test (from Reinhart ; see also () above):
() A: Tell me about John.
B: Hes very nice.
architectures and information structure
() As for John, hes very nice.
Speaker As request in () constrains Bs response to being about John.
The reference to John (he) in the response is therefore the topic of the
response. The same goes for ().
The second property of topics is that they must be given. Previous
mention is therefore an indication of topichood. There are two prob-
lems with this test. One is that a sentence may include more than one
previously mentioned element. Do we conclude that the sentence has
more than one topic? Or is one of these elements the topic? In the next
section we argue that sentences may have more than one topic, yet they
may not all have the same status. Another problem with identifying top-
ics by previous mention is the fact that some topics are given (in the
extended sense) without previous mention. Examples are implicit stage
topics, specic indenite topics discussed in sections . and ., respec-
tively, and the permanently and temporarily available topics discussed
in the previous section. We can conclude that previous mention identi-
es topics, but that not all topics require previous mention.
The third property of topics, i.e., that truth values are calculated with
respect to them, can be used in two ways. The rst is somewhat intuitive,
and we have already seen it at work in (): A DP without reference (the
King of France) led to a truth value gap when it was a topic, and triggered
a false truth value when it was not. In view of the fact that () necessarily
leads to a truth value gap and () never does, it follows that the subject of
the predicate be bald must be a topic, and that the by-phrase of the passive
excludes a topic interpretation. In both cases, the subject is identied as the
topic of these sentences. Subjects are indeed the unmarked topics, yet cer-
tainly not the only position for topics as we have already seen in the cases of
topicalization in Danish and Catalan ( () and () respectively), as well as
the examples with implicit stage topics in () and (). In these examples,
due to the expletive subjects, the only way to assess the truth value was to
interpret the sentences as predicated of a stage topic. The test can also be
extended to identify potential topics in examples in which several elements
can function as topics. Such an example is () repeated in ().
() Q: What happened?/Why do you look so pleased?
A: John washed the dishes!
The question in this example forces a reading in which the sentence must
be predicated of the current spatio-temporal parameter. Other contexts
in which the sentence may be uttered are, however, also possible as
shown in () and ():
architectures and information structure
() Q: What happened to the dishes?
A: John washed them.
() Q: Why does John look so pleased with himself?
A: He washed the dishes.
The question in () not only refers to the dishes, but also asks what
happened to them. Therefore this element is not only given in the
answer, but also what the answer is about. In () John is the only
given element in the answer, hence the topic. This simple sentence
therefore has at least three potential topics depending on context.
In () and () (examples of focus chaining and topic chaining,
respectively) the topics are pronouns. Repeating the full DP would be
redundant. Non-anaphoric, weak pronouns, according to Erteschik-Shir
, are always topics since they, like topics, must be given. Moreover, if
pronouns are considered to be topics, then the fact that they must have
a contextual coreferent antecedent can be derived. In section ., below, I
demonstrate how this follows.
The second way we can use this property of topics is to identify those
DP types that can be potential topics. It is well known that individual-
level (I-level) predicates cannot be predicated of a stage topic.
The only
possible topic of an intransitive I-level predicate is therefore the subject.
It follows that any DP that can ll this position can function as a topic.
() tests a variety of indenite DP types for topichood in this manner.
() shows that denite DPs always qualify as topics, as expected.
() a #A Frenchman is intelligent.
b Dogs are intelligent.
c A student I know is intelligent.
d A DOG is intelligent, a CAT is not.
e Two/some of the students are intelligent.
() a The Frenchman is intelligent
b He is intelligent.
c John is intelligent.
The examples in () show that the interpretation of indenites in topic
position is restricted. (a) shows that a singular indenite is odd in
this position. Simple indenites are new and cannot qualify as topics.
The subject in (b) can only be interpreted generically, and not as a
See Kratzer for the distinction between individual-level and stage-level predicates and
Erteschik-Shir and the references cited therein for more details on the points made here.
architectures and information structure
simple plural indenite. Generics can be topics because they name spe-
cies and names of species may be given. Whereas plural indenites can
get a generic interpretation with ease, this is not the case for singular
indenites. That is why (a) is odd, but (b) is not. (ce) illustrate
that specic indenites, contrastive indenites, and partitive indenites
all qualify as topics according to this test. The reason this is so will be
given in section . below.
2.1.7 Multiple topics
Another controversial issue concerning the basic inventory of IS-theoretical
concepts has to do with whether a sentence can have more than one topic
and focus. That topic-focus assignment is recursive has been argued
among others by Krifka , Lambrecht : , and Erteschik-Shir
One simple argument for this is that if indeed pronouns are top-
ics, as suggested above and argued below in section ., and there can be
more than one pronoun in a sentence, then there must also (potentially)
be more than one topic per sentence. Take, for example, the following
() Q: Did you see Peter yesterday?
A: Yes, I saw him at school.
In the answer in () there are two pronouns, hence two topics by this
reasoning. Topics, however, are not only given, but also dene what the
sentence is about and provide an address for truth value assessment
if we accept the StrawsonReinhart view of topics. We must therefore
decide whether the sentence is about both of the referents of the pro-
nouns or just one, and if one, which one. The same goes for assessment.
These issues are not simple to decide. In fact, there is no evidence (in this
context) that the sentence is about the subject rather than the object or
vice versa yet the truth value wont necessarily come out the same. If we
replace Peter with the King of France, then the answer will come out
false if we evaluate with respect to the subject, and we will get a truth
value gap, if we evaluate with respect to the object. In other words, if
both topics count for evaluation, the sentence will be ambiguous. In
order to avoid such ambiguity in the case of multiple topics, one topic
must be designated as the main topic. This topic, the main topic, is
Nikolaeva argues that secondary topics in Ostyak are marked by triggering
agreement. These topics are similar in function to Vallduvs Tails. Ostyak and Catalan can
therefore be viewed as marking more than one topic syntactically.
architectures and information structure
then what the sentence is about and is the locus for assessment. The dis-
tinction between main and subordinate topics (and foci, see below) will
be discussed in somewhat more detail below in section .
2.1.8 Topic drop
An interesting property of topics in some languages is that they can be left
out altogether. This is often referred to as topic drop. Huang dis-
tinguishes subject drop (PRO drop or null subjects) from topic drop as
follows: subject drop is dependent on the availability of rich inectional
agreement morphology; topic drop does not exhibit such a dependency.
Instead, topics are recoverable from the discourse. Topic drop is illus-
trated in the Chinese examples in () from Huang : (e stands for
the omitted topic pronoun. The translation of e as he/him is only one
possibility: the pronominal reference is derived from the context):
() a e lai-le.
[He] came.
b Lisi hen xihuan e.
Lisi very like
Lisi likes [him] very much.
c Zhangsan shuo [e bu renshi Lisi].
Zhangsan say not know Lisi
Zhangsan said that [he] did not know Lisi.
d Zhangsan shuo [Lisi bu renshi e].
Zhangsan say Lisi not know
Zhangsan said that Lisi did not know [him].
Whereas the English pronouns in the translation of (d) are ambiguous
between a reading in which they refer to the matrix subject or to a dis-
coursally available referent, the Chinese examples are unambiguous and
the null pronoun must refer to a discourse referent outside the sentence.
This shows that the empty category must indeed be a topic.
(c) is ambiguous in Chinese as it is in English. This is because, as Huang argues,
Chinese, in addition to allowing topic drop, also allows control of PRO subjects in tensed
sentences (on a par with subject control in innitives in languages such as English). Since
PRO can only occur in subject position, the empty category in (d), in which the dropped
category is an object, can only get its reference from the context.
Huang uses examples of topicalization in Chinese to argue that the empty categories are a
result of topicalization followed by deletion of the topicalized element.
: architectures and information structure
German, another well-known topic-drop language, exhibits the fol-
lowing subjectobject asymmetry: whereas Ist, :nd, and ,rd person sub-
jects can be dropped, only ,rd person objects can be dropped. Schulz :oo,
also suggests that dropped topics must be recoverable. She derives that,
in order not to induce a breakdown in communication, a necessary pre-
requisite for such recoverability seems to be that the dropped constituent be
a continued topic. She explains the subjectobject asymmetry by argu-
ing that ,rd person objects are easier to recover than Ist and :nd person
objects in view of the fact that ,rd person is the unmarked case for con-
tinued topic objects.
What is of interest to us here is the claim that topic
drop involves continued topics. A similar claim can be found in Wexler
I,,8: ,,, who claims that dropped subjects have to be very strong topics.
Very strong should probably be taken to mean continued topics. In the
case of languages such as German in which topicalization is common, it
may be the case that the two strategies to mark topics (fronting and omis-
sion) target two dierent topic types: Shift topics are topicalized, whereas
continued topics are omitted. If this is the case, then German makes the
same distinction as does Catalan, but uses dierent strategies: Catalan
postposes the continued topic, German omits it.
According to Huang I,8, English is a hot language in the sense of
Marshall McLuhans (I,o) hotcool division of the media. Hot
media require little audience participation. The fact that English pro-
nouns cannot be omitted means that English sentences provide the
hearer with all she or he needs to understand the sentence. Chinese,
according to Huang, is a cool language since it allows topic drop and
therefore requires the hearer to invoke the context, world knowledge,
and such in order to understand the sentence. The premiss underlying
Huangs categorization of these language types is that accessing context
(and making inferences from world knowledge) is not automatic and
therefore requires something of the hearer. This is the same premiss
upon which theories that exclude IS from the grammar are based. If,
however, context is automatically accessed (and I believe it is), English
might be the cool language, in that its lack of IS marking requires more
of the hearer than does Chinese in which topics are marked by their
absence. Even hotter than Chinese is a language which signals new topics
by fronting and old ones by omission, as may be the case for German.
Schulzs analysis in terms of optimality theoretical harmonic alignment, although ele-
gant, will not concern us here.
Chinese also employs topicalization as well as topic drop. I do not know whether Chinese
and other topic-drop languages also distinguish the two topics types by these two strategies.
architectures and information structure :,
2.1.9 Scope and topic properties
One of the important results of the denition of a topic as the pivot for
truth value assignment is that the scopal relations in the sentence depend
on topic assignment: The topic will have wider scope than other elements
in the sentence because the predicate of the sentence is evaluated with
respect to the topic.
Since the topic is contextually determined, the scopal
relations will also depend on context. Scopal ambiguity is therefore elimi-
nated once a sentence is contextualized. Consider the sentence in (,:).
(,:) Two girls arrested three boys.
Outside of context (,:) has at least three dierent choices of topic: Either
the subject takes wide scope, or the object takes wide scope, or else the
sentence has a stage topic and receives an unscoped interpretation.
subject takes wide scope if it is interpreted as the topic of the sentence. In
view of the fact that the subject is weakly quantied, it can only be inter-
preted as a topic under a partitive interpretation (i.e., two of the girls
under discussion).
This is possible in a context in which a set which
includes the two girls in question is given in the discourse. An example
would be a context in which a set of girls is given an assignment to arrest
as many boys as they can. In this case the set of girls would function in
the same way as the hypertheme discussed in section I.: above in that
the two girls would have to be selected from this larger set in order
The literature on topics and related notions is vast and will not be reviewed here. The
following is a selection of (additional) important contributions to the understanding of
various aspects of the notion: Basilico :oo,; Beaver :oo:; Bring I,,,; Chafe I,,o; Givn
I,,o; Gundel I,,; :oo,; Halliday I,o,; Kidwai :oo; Lambrecht I,,; Li and Thompson
I,,o; Molnr I,,8; Paul :oo,; Polinsky :ooI; Prince I,8Ia; Sasse I,8,; Schwarzschild I,,,;
Taverniers :oo,; Ward and Prince I,,I.
Van Valin and LaPolla I,,,: ::I propose a principle to the effect that topical quantied
NPs have scope over focal quantied NPs. This principle does not, however, follow from the
denition of a topic as it does here. Van Valin and LaPolla cite Sgall, Hajiov, and Panevov
I,8o: ::, who make a similar observation. One of the principles determining quantier
scope in Kuno I,,I is that a topicalized quantier has scope over a non-topicalized one. This
principle also follows from the denition of topics here as long as the topicalized element is
indeed interpreted as a topic. Kunos other principles determining quantier scope, e.g., that
discourse-linking (section ,., below) contributes to wide scope, may not be required with
the broad denition of topics employed here.
See Erteschik-Shir I,,,: chapter , for a more complete analysis of scope as it follows
from topic assignment.
This interpretation requires stress on the numeral. Stress on topics occurs when a focus
is embedded within them. It is also this fact that explains how the partitive interpretation is
derived as shown below in section ,..
:o architectures and information structure
to qualify as a topic of the sentence. The truth value of (,:) would be
assessed in such a context by examining the two girls as to whether
they in fact arrested three boys or not. There are two dierent ways in
which this assessment could be accomplished. One way would be to
view the two girls as a group (the collective reading); then if a group
of two girls (together) arrested three boys, the sentence would be true.
Another way to assess the sentence with the subject as topic would be
to view the two girls as separate entities (the distributive reading). In
that case the sentence would be evaluated by examining whether the two
girls arrested three boys each. If true, there would be six boys arrested, all
told. If the object is the topic a parallel analysis can be given. In that case,
under the distributive reading of the object, each boy would be arrested
by two girls, so that six girls, all told, would be involved in the arresting.
The unscoped reading results if neither the subject nor the object is
interpreted as a topic. In that case the sentence will be interpreted as
predicated of a stage topic, which is interpreted as the current here-and-
now in (,:), or else it can have an overt stage topic as in (,,).
(,,) Today/at o oclock/on the corner, two girls arrested three boys.
(,,) will be true if an event of two girls arresting three boys took place
on the specied stage.
The reading in which the object has wide scope is less accessible than
the other two readings. It has frequently been observed that subjects
are unmarked topics across languages (e.g., Li and Thompson I,,o;
Reinhart I,8I; Andersen I,,I; Lambrecht I,,; Winkler and Gbbel
:oo:). Sentences with stage topics are also unmarked, as can be seen by
the ease in deriving the unscoped reading. Sentences with object topics
are marked. We return to this point in section ,.:.
This section argues that once IS is taken into account scopal ambigu-
ity does not occur. That scope follows from topic assignment also has
repercussions for the architecture of grammar in general and the role of
LF in particular as shown in section ., below.
What has been shown in this section is that the optimal notion of topic
is not easy to nd if we link it to syntactic phenomena such as topicaliza-
tion. This is because these phenomena are not uniform across languages.
In some languages, topics must be syntactically marked, as in Catalan,
for example. Such languages are called topic prominent.
English is an
See Li and Thompson I,,o for this notion.
architectures and information structure
example of a subject-prominent language in which such syntactic mark-
ing is uncommon. In subject-prominent languages, the subject is in fact
the unmarked topic. Danish exemplies a language which is harder to
classify. Topicalization is frequently used, but it is still optional. Another
diculty is that topicalization does not select the same class of elements
even in the two languages examined here. This might lead one to the
unhappy conclusion that there are several kinds of topics, and that lan-
guages may mark some but not all of them.
A variety of phenomena
connected to topichood could be described in this way, but it would not
provide an explanation. Our aim should instead be to derive the vari-
ous kinds of topics from a minimal set of primitives. In the next section,
we dene the other basic primitive, the focus. We continue to show that
these two basic IS primitives, topic and focus, interact in such a way that
all types of topic and focus can be derived from them without the need
for further primitives.
2.2 Inventory: foci
Foci, like topics, have been dened in many ways and from several perspec-
tives (semantic, phonological, syntactic, and pragmatic). In this section,
illustrative examples of each of these perspectives are examined.
2.2.1 Semantic denitions
Chomsky , Jackendo , and Lambrecht , among others,
dene the focus as the non-presupposed information in the sentence.

The presupposition is dened as the information in the sentence that is
assumed by the speaker to be shared by him and the hearer (Jackendo
: ). In view of the fact that topics are given (in the extended sense
of given) they are generally presupposed. It follows from this denition
of focus that the two notions, topic and focus, are mutually exclusive.
According to Chafe , a topic in a topic-prominent language such as Chinese sets a
spatial, temporal, or individual framework within which the main predication holds. This
is a property of topics in some languages; in others topics are what the sentence is about. The
two properties, according to Chafe, do not go hand in hand, as Strawson would have it, and
different languages mark different topic properties.
This is not quite accurate. Chomsky : in fact denes the focus as follows: The
focus is the phrase containing the intonation center; the presupposition, an expression
derived by replacing the focus by a variable. Each sentence, then is associated with a class of
pairs (F,P) where F is a focus and P is a presupposition, each such pair corresponding to one
possible interpretation. In fact, Chomsky derives the focus from intonation. See section .
for discussion of this tack.
architectures and information structure
This turns out not to be accurate as we show in section .. It also follows
that foci must be new in the discourse. The questionanswer test also
follows since the answer to the wh-question is necessarily new, and the
rest of the sentence, which must be present in the question, is presup-
posed (the focus is in capitals, the presupposition in italics):
() a Q: What did John do?
b Q: What did John wash?
A: He washed THE DISHES.
c Q: Who washed the dishes?
A: JOHN washed them.
d Q: What happened to the dishes?
e Q: What happened?
f Q: What did John do with the dishes?
A: He WASHED them.
In (a), only the subject is presupposed and is therefore the only can-
didate for topichood. The VP is the focus. In this example, the syntax
IS mapping is therefore perfect and, as we shall see in Chapter , this
is in fact the unmarked case. In (d) the topic is also identied with
the presupposition, yet the focus does not form a syntactic constituent.
This is a problem for those theories that consider foci to be a property of
syntactic constituents, but is not a problem for the semantic denition
which identies foci as the complement of the presupposition. In (b)
and (c) the presupposition and the topic do not overlap. These sen-
tences are evaluated with respect to John and the dishes, respectively.
It follows that the topic is contained within the presupposition in these
examples, but cannot be identied with it. (e) illustrates an all-focus
sentence with no presupposition. These sentences have a stage topic as
argued above. In (f) the presupposition is John did something with
the dishes, leaving the verb as the focus.
Jackendo recognizes the role of focus in interpretation (association
with focus, VP-anaphora), and oers a preliminary account employing
this denition of focus as the complement to the presupposition. Rooth
; , developing the ideas in Jackendo, gears his denition of
focus to a more formal account of these issues. Here we set aside these
semantic issues, yet Rooths denition is still of relevance. Rooth employs
architectures and information structure
the notion of p-sets: the sets of alternatives under consideration in the
discourse. The focus semantic value of a phrase, under this view, is the
set of propositions obtainable from the ordinary semantic value by mak-
ing a substitution in the position corresponding to the focused phrase
(Rooth : ). Examine (), as an illustration:
() Q: Which laundry did John wash?
A: He washed THE WHITES.
The question sets up a set of alternatives such as: John washed X laun-
dry, where X can be the colored laundry, the white laundry, the ne
laundry, the woolens, etc. In the answer the white laundry is selected
to be substituted for X, and is therefore the focus value by Rooths de-
nition. Notice rst that it also follows from Rooths denition that the
presupposition and the focus complement each other. Another point
is that the set of alternatives are dened contextually. The question
in () makes this clear since a which-question must range over a dis-
coursally available set. What happens when such a set is not readily
derivable from the context? The what-question in (b), for example,
does not in and of itself engender such a well-dened set. Certainly,
the answer must range over a set of washables, but this set is only
vaguely dened in the context of this question. According to Rooth,
these two types of sets need not be distinguished. We shall see in sec-
tion . that constituents which range over contextually restricting
sets, as illustrated in (), have dierent IS properties from those that
dont. We will also see how this distinction is derivable from dierent
topic and focus assignments.
In the previous section, we noted that topics can be contrastive. This is
also the case for foci. () is just like (), except that the set of alternatives
is specied overtly.
() Q: Which laundry did John wash, the white or the colored?
A: He washed the WHITE laundry.
A contrastive focus focuses one element of the contrast set and elimi-
nates the other alternatives. Contrastive foci are therefore included
under Rooths denition. They are analyzed in section . as a subset of
the foci which range over contextually restricted sets. Contrastive foci
are often referred to as narrow, exhaustive, or exclusive foci. Non-
contrastive foci are referred to as informational foci or presentational
foci (when they occur in existentials).
architectures and information structure
An interesting take on the semantic approach stems from the idea
in Heim a that the division between presupposition and non-pre-
supposition is semantically represented as a division into a restrictive
clause and a nuclear scope. Heims semantics is intended to regulate the
interpretation of denites and indenites. Partee recognizes that
this division is in fact to be viewed as a distinction between topic and
focus, with the topic identied as the restrictive clause and the focus as
the nuclear scope. Diesing proposes an algorithm which derives
this division from the syntactic representation: VP-external material is
mapped onto the restrictive clause and VP-internal material is mapped
onto the nuclear scope. One advantage of Diesings approach is that sub-
ject topics are structurally distinct from subjects which are within the
domain of the focus as in an all-focus sentence. The former are located
outside the VP, the latter inside it. This aords an explanation of the
dierent properties of individual-level predicates for which the subject
must be VP-external and stage-level predicates for which the subject can
either be VP-internal or VP-external (see section .).
2.2.2 Marking foci by stress
Just as it is common in languages to mark topics by fronting, so it is
common to mark foci by stress.
Jackendo : denes this rela-
tionship as follows:
() If a phrase P is chosen as the focus of a sentence S, the highest stress in
S will be on the syllable of P that is assigned highest stress by the regu-
lar stress rules.
At the time, the regular stress rules included the Nuclear Stress Rule (NSR)
which assigns main stress to the nal constituent. Simplifying a great deal,
this means that sentence-nal stress leads to focus ambiguity (capitals indi-
cate stress):
() a Maxwell killed the judge with [a HAMMER]
b Maxwell [killed the judge with a HAMMER]
c [Maxwell killed the judge with a HAMMER]
In the three examples in (), the stress is on the sentence-nal con-
stituent, a hammer. This means, according to Jackendos rule, that the
focus can be this DP as in (a), or the whole VP as in (b), or the whole
Duration, amplitude, and pitch combine to give the effect of perceived pitch accenting.
The term stress is used for convenience throughout.
architectures and information structure ,I
sentence as in (,8c). This also means that the full sentence answer to the
three questions in (,,) will sound the same, i.e., it will have stress on
hammer, since the answers to these three questions are (,8a), (b), and
(c), respectively.
(,,) a What did Maxwell kill the judge with?
b What did Maxwell do?
c What happened?
Answering a wh-question with a full sentence, without pronominalizing
the topic, is highly articial. Even so, it is not dicult to elicit intuitions as
to the relative appropriateness of the answer in (,8) to each of the ques-
tions in (,,). The answer with only nal stress is most appropriate as an
answer to (,,a). It is much less successful as an answer to (,,b) and (c).
The answers to these questions require additional stresses as shown in
(oa) and (b), respectively.
(o) a Maxwell KIlled the JUdge with a HAmmer
b MAxwell KIlled the JUdge with a HAmmer
Although the nal stress on hammer is highest in all three cases under
discussion, the additional stresses cannot be left out totally, although the
middle stresses can be somewhat reduced depending on the speed in
which the sentence is uttered.
(I) is probably the most natural answer
to (,,c):
(I) MAxwell killed the judge with a HAmmer.
Without the initial stress on the subject, the sentence, in the desired con-
text, would be awkward indeed. A stress rule that renders these results,
from Erteschik-Shir I,,,; Erteschik-Shir and Lappin I,8,b, is:
(:) Assign stress to the focus constituents.
This rule together with a rhythm rule which allows the reduction of inter-
mediate stresses in rapid speech results derives stress distributions such
as the one illustrated in (I): The beginning of the focused constituent
is minimally marked with a low accent the end of this constituent with
The fact that out-of-the-blue sentences such as (ob) are assigned a sequence of high pitch
accents on each of the major constituents has been noted by Schmerling I,,, and Steedman
I,,I: :8o among others. More recently Hockey I,,8 concludes from an experimental study of
pitch, duration, and amplitude that information structure focus is marked on all the words
within a focus constituent, calling into question the idea that there is ambiguity about how
large a constituent is marked as focus by pitch accent placement on a particular word.
,: architectures and information structure
a high accent, and the intermediate accents are lowered to a minimum
down to nothing.
The motivation behind the stress rule in (:) is that stress is a way to
mark foci, just as fronting is a way to mark topics. Since the rule applies
to the whole focus, and not only its nal constituent, it in fact marks the
focus unambiguously by minimally marking its beginning and its end in
fast speech (due to the rhythm rule) and maximally marking each one of
the major constituents in slow speech.
Both Jackendos and Erteschik-Shirs stress rules take as their input
sentences in which the focus constituent(s) are marked. Jackendo I,,::
:o considers how this marking may come about: He introduces F, a
syntactic marker which can be associated with any node in the surface
structure. Stress assignment and interpretation are then derived in this
way from the free assignment of a focus feature.
2.2.3 Deriving foci from stress
The opposite view, namely that focus assignment is derived from the
distribution of pitch accents, is also common. As an illustration of this
view, we examine Selkirk I,,,.
In order to derive the focus from pitch
accents, Selkirk I,,,: ,,,oI proposes a Basic Focus Rule which states
that An accented word is F(ocus)-marked. (This is the exact reverse
of the rule in (:) which derives stress from focus assignment.) Selkirk
then proposes the following three rules of Focus Projection:
(,) a F-marking of the head of a phrase licenses the F-marking of
the phrase.
b F-marking of an internal argument of a head licenses the
F-marking of the head.
c F-marking of the antecedent of a trace left by NP- or wh-
movement licenses the F-marking of the trace.
It follows from these rules that if either a verb or an object is accented,
then Focus projection licenses the interpretation of the VP as the focus.
According to the Basic Focus Rule, accent on the verb means that the verb
is F-marked. According to (,a), once the verb is F-marked, the F-marking
Other proponents of this type of approach which Selkirk draws on are Gussenhoven
I,8; Rochemont I,8o; Schmerling I,,,; Selkirk I,8. More recent proponents of this view
are Cinque I,,,; Reinhart I,,,; I,,o; Selkirk I,8; Neeleman and Reinhart I,,8; Szendri
Focus which does not project is often referred to as narrow focus.
architectures and information structure
of the VP is also licensed. (b) ensures that F-marking of the object
licenses F-marking of the verb, which, in turn, according to (a), licen-
ses F-marking of the VP. The F-marking of the VP, according to Selkirk,
licenses the F-marking of the whole sentence via licensing of the various
intervening heads. If, however, the subject is accented, focus projection
to other elements of the sentence is blocked because the subject is an
external argument and neither (a) nor (b) licenses projection from
external arguments. Rule (c) is intended to explain why focus does project
from certain accented subjects. An example is given in () which is a
possible answer to Whats been happening?
() The SUN came out.
Selkirk adopts the hypothesis that the subjects of unaccusative predi-
cates are derived by movement from object position leaving a trace.
According to (c), this trace can be F-marked since its antecedent is.
Once the object trace is F-marked, so is the VP.
Rules such as (c) are
highly counterintuitive and should probably be avoided. Yet stress rule
() which derives stress from focus assignment fares no better. If the
sentence is all-focus, both major constituents will be stressed and not
just the subject as required. An account of these cases is proposed in
Chapter , section . There it is shown that this intonation pattern is
determined contextually and is not limited to unaccusative verbs, nor
does it depend on a syntactic movement analysis.
Although (a) and (b) also do not seem to follow from any general
principles, deriving focus assignment from rule-governed pitch accent
has one major advantage: It does not require the introduction of focus
features in the derivation, a problem for other theories as we shall see
below in section . It does, however, depend on an independent mech-
anism of pitch accent assignment, yet pitch accent assignment depends
at least in part on what the intended focus is. Therefore, theories such as
Selkirks turn out to be somewhat circular.
A dierent perspective on the derivation of focusbackground
interpretation from intonation comes from von Heusinger , who
introduces a mechanism into Discourse Representation theory which
allows the derivation of foregroundbackground interpretation
For this case, Selkirk need not invoke (c): Since the subject is in fact an internal argu-
ment, focus projection also follows from (b).
For comments on this type of approach and further developments see, among others,
Schwarzschild and Bring .
, architectures and information structure
from intonation:
Pitch accent marks the focused constituent which
is replaced by a designated variable in the background. The background
representation marks the part of the sentence that is already anchored
or given in the discourse.
The two viewsderiving stress from focus and deriving focus from
stresscan, in principle, be viewed as two sides of the same coin: the
former reects the perspective of the speaker, who uses stress to mark
the focus of the utterance, and the latter the perspective of the hearer,
who uses stress to discern the focus of the utterance heard (Van Valin,
personal communication). However, most of the researchers who use
one or the other perspective to analyze the relationship between focus
and stress do not view their analyses as reecting actual processing, but
rather as a purely theoretical approach to deriving the correct correla-
tion. The dierent views should therefore be evaluated purely on the
basis of their success in this endeavor. This is not to say, of course, that
an analysis in terms of processing from the two perspectives wouldnt be
2.2.4 Deriving stress from syntax
Cinque I,,, derives stress prominence syntactically by viewing it as a
reection of depth of embedding: The most deeply embedded phrase
is the one that is most prominent.
In this way, Cinque is able to derive
the dependence on position of prominence from the headedness of
languages. His prediction is that in phrases that are right branching,
stress prominence is on the rightmost constituent (as in English),
whereas left-branching phrases have stress prominence on the left.
German has mixed headedness: NPs are head initial (right branching)
and therefore stress nal; VPs are head nal (left branching) and the
most deeply embedded constituent will be the one immediately to the
left of the verb as shown in (,a) with the example in (,b), both from
Cinque I,,,: :,I.
Discourse Representation theory, or File-Change Semantics, originated with Heim I,8:b
and Kamp I,8I. The key idea in the way of thinking about the semantics of discourse is that each
new sentence or phrase is interpreted as an update of the context in which it is used.
In addition, von Heusinger argues that boundary tones indicate breaks in the construction
of the DRS. The resulting representations can be used to express discourse relations or can be
used as arguments for semantic operators. See also Kadmon :ooI for a different theory of the
interpretation of accent placement.
For other proponents of the syntax-based approach to prosody and the prosodyfocus
relation see, among others, Culicover and Rochemont I,8,; Ladd I,8o; Liberman I,,,.
architectures and information structure
b da Fritz einem Kind GELD gegeben hat
that Fritz a child money given has
The parallel sentence in English, however, will be stressed on the nal
constituent which is most deeply embedded, as shown in () adapted
from Cinque : .
IP a ()
() a IP
b Fritz gave a child MONEY
In order to derive the correct results for examples of this sort in right-
branching languages, Cinque adopts a VP-shell analysis of these con-
structions in English (but not in German). Cinques syntactic account
of stress prominence is hard to evaluate in cases in which the syntac-
tic account of a particular construction is controversial, but would in
principle have great potential as an explanation of the cross-linguistic
dierences between right- and left-branching languages and struc-
tures with respect to stress prominence were it not the case that it only
accounts for default stress in all-focus sentences, and does not, in and
of itself, oer an account of the connection between stress prominence
and focus assignment. In order to do that Cinque stipulates a separate
,o architectures and information structure
discourse grammar procedure for stress assignment on focus constitu-
ents. Cinque has the discourse procedure interact with the syntactic der-
ivation of prominence as follows: The discourse procedure determines
that stress on focused constituents wins out (relatively) over stress on
presupposed constituents. In other words, syntax is responsible for
stress assignment, but discourse principles can overrule this assignment.
One might wonder whether syntactic-determined stress assignment
and discourse-determined stress assignment are both necessary in the
grammar. Cinque argues that they are since the syntactic rule applies
blindly to both focused constituents and presupposed constituents. The
discourse rule, however, is responsible only for stressing focused con-
stituents. If stress prominence is a way of signaling focus, the appear-
ance of stress on presupposed elements is indeed problematic. It would
be an achievement if such stress were merely a result of a syntactic rule.
Cinques example is given in (,) (his (o); I,,,: :,8); the accent indi-
cates stress prominence on the focus, the hatchet indicates the additional
detectable prominence within the presupposition):
(,) a (Any news of John?)
Our poor chld] [
is in bed with a ].
b (Whos in bed with a u?)
Our poor chld] [
is in bed with a ].
Cinque argues that both the presuppositions in these two examples (the
NP in (,a) and the VP in (,b) ) as well as the focus constituents have a
detectable prominence which, in the null theory, follows from the sen-
tence grammar procedure that applies to all phrases blindly (i.e., inde-
pendently of the discourse grammar rule).
One of the foremost strengths of Cinques proposal is its ability to
predict the dierent stress assignments on right- and left-branching
languages without any stipulation. Details aside, Cinques rules will
always assign main stress to the most deeply embedded constituent
in a phrase. In a right-branching language, this constituent will be to
the right, and in a left-branching language, to the left, predicting the
observed distinctions on purely structural grounds. I propose in sec-
tion ,. below that the additional stresses on the presupposed part of
the sentence (,a) can in fact be derived by a theory which derives stress
from focus assignment, once the theory allows for multiple topic and
focus assignments. In (,b), the presuppositional part of the sentence
is repeated from the question. Such repetitions inherit their stress from
architectures and information structure ,,
the stress pattern in the question, explaining the (weak) stress on the
nal element of the VP.
This would leave unexplained the dierence
in prominence between left- and right-branching languages. Here the
rhythm rule introduced in section :.I comes into play. It is not dicult
to envision how such a rule can be parameterized such that the rhythm
of the language is dependent on its structure. Right- and left-branch-
ing structures and languages would then dier only in the application
of the rhythm rule and we could still maintain that focus constituents
are universally marked, either by stress, morphology, or word order or
some combination of these features.
Cinque holds the common view that foci are constituents. It is not
obvious that this is so as shown in the next section.
2.2.5 Deriving syntax from IS
Steedman :oooa argues that intonational phrase boundaries and surface
syntactic boundaries coincide, rendering the unconventional syntactic
structure in (8):
For more details on the pronunciation of such repetitions, see Erteschik-Shir I,,,:
Steedmans (:oooa: o,,) proposal is couched in Combinatory Categorial Grammar which
directly pairs phonological and logical forms without intermediary representational levels.
He argues that such a theory in which phrasal intonation and information structure are
reunited with formal syntax and semantics is not only possible, but much simpler than one
in which they are separated.
MarCEL proved completeness MARcel proved completeness
(8) a b
According to Steedman the rhythm rule (not the rhythm rule discussed
here) which applies in certain dialects to move the stress onto the rst
syllable of Marcel applies to the intonational phrase Marcel proved
(on a par with Marcel Proust), requiring, within his framework, that
the latter be analyzed as a syntactic constituent as in (8b). It is a well-
known fact that prosodic rules cannot apply across syntactic boundaries
and so Steedmans syntax follows. This is only one of Steedmans argu-
ments that syntactic structure does not map onto prosodic structure
and therefore must be amended.
,8 architectures and information structure
According to Steedman :oooa: o, both (8a) and (b) are possible, each
as an answer to a dierent question:
(,) Q: I know who proved soundness. But who proved completeness?
A: (Marcel) (proved completeness)
(,o) Q: I know which result Marcel predicted. But which result did
Marcel prove?
A: (Marcel proved) (completeness)
In (,) the subject Marcel is focused in view of the fact that it answers
the wh-question. In addition the object completeness is contrasted (with
soundness), hence the stress on these two constituents. In (,o), however the
object completeness answers the wh-question and is therefore focused. In
addition, the verb is contrasted and therefore stressed. It is this fact which
causes the prosody in which the subject and verb form a constituent, allow-
ing the rhythm rule to apply within it.
Any account in which syntax mediates between prosodic structure and
IS will have to take such issues into account. A possible solution which
does not require structures such as (8b) is outlined in section ,., below.
2.2.6 Deriving focus pragmatically
According to Dik I,,,: ,:o, the focal information in a linguistic expres-
sion is that information which is relatively the most important or salient
in the given communicative setting, and considered by [the] S[peaker]
to be most essential for [the] A[ddressee] to integrate into his prag-
matic information. Such information is often new, and if not new it
re-emphasizes information already available to A rendering explicit or
implicit contrast.
Diks denition of focus is along the same lines as that of Erteschik-
Shir I,,,, based on Erteschik-Shir I,,, and Erteschik-Shir and Lappin
(,I) The FOCUS of a sentence S = the (intension of a) constituent c
of S which the speaker intends to direct the attention of his/her
hearer(s) to, by uttering S.
The denition of focus in terms of speakers intentions entails that it is a
discourse property which is assigned to a constituent in a context of con-
versation. For any sentence several focus assignments will generally be
possible, one of which is realized in discourse. A sentence, in discourse,
has only one main focus which is assigned to a syntactic constituent. This
architectures and information structure
constituent may be an NP, a VP, or even the whole S (as in an out-of-the-
blue sentence). The topic of a sentence is excluded as a focus because it is
by denition already in the hearers attention. Hence, the focus constitu-
ent is selected freely among the non-topic constituents of the sentence.
The fact that the focus is dened as the constituent to which the hearers
attention is drawn enables the constituents contained in it to provide the
topics of the following sentences since these constituents have become
part of the domain of what the hearer is now attending to, allowing for
focus chaining to occur.
The focus, under this denition, is also identied by the question
answer test since the constituent that answers the wh-phrase is indeed
the constituent to which the hearers attention is drawn.
Another test which follows from the denition in terms of the hearers
attention is the lie-test:
() LIE-test:
Speaker A: John said that he knows Peter.
Speaker B: a. Thats a lie, he didnt.
b. Thats a lie, he doesnt.
Each response indicates a dierent IS: (a) takes the subject as a TOPIC
and the whole VP as the FOCUS. The fact that (b) is a possible response
indicates that the lower sentence can be the one that the speaker directs
the hearers attention to and hence the focus can fall on or within it.
This test is therefore not a test for focus per se, but rather a test for the
domain(s) within which the focus potentially may occur. The actual
focus in a particular utterance is restricted to these domains.
The pragmatic denition diers from the semantic one in that the
complement of the focus is not necessarily presupposed. This can be
seen in (b) in which the upper clause does not belong to the focus or to
the presupposition. Another example is given in ():
() Q: Where is the book?
A: I gave it to Mary.
The focus of the answer is to Mary, yet it is not presupposed that I
gave it to someone.
Since the pragmatic denition of focus does not depend on interpre-
tation or on stress assignment, these properties must be derived from
(topic and) focus assignment. In () (section .), a stress rule was
proposed which derives stress from focus assignment. Interpretation
in this framework follows mainly from topic assignment as seen above.
o architectures and information structure
Focus plays a role only in its interaction with the topic. The denition
is syntactically anchored in that foci must be syntactic constituents.
Something further therefore must be said about a sentence such as (,d) in
which the focus consists of the subject and the verb and is therefore seem-
ingly not a constituent. This issue will be addressed in section ,., below.
2.2.7 Summing up focus properties
Any denition of focus must measure up to the requirement that it be
universal. This excludes any denition of focus which is based solely on
intonation, since not all languages use stress to mark the focus. Just as top-
ics can be marked by word order (fronting), so foci, in certain languages,
take up a particular syntactic position. An example is Hungarian in which
foci and wh-question words are positioned preverbally (see the discussion
of Hungarian in Chapter ,, section I). In still other languages, topics and/
or foci are marked by morphological means. Dik I,,,: ,,,o oers two
illustrations. The rst is Wambon, a Papuan language of Irian Jaya.
It has
one focus marker -nde which marks foci and question words:
(,) A: Jakhove kenonop-nde takhim-gende?
they what-Foc buy-,pl:pres:nal
What do they buy?
B: Ndu-nde takhim-gende.
Sago-Foc buy-,pl:pres:nal
They buy SAGO.
Dik also shows that -nde marks contrastive foci and also contrastive top-
ics. These are illustrated in (,,) and (,o), respectively.
(,,) A: Mbitemop ndune ande-tbo
Bitemop sago eat-,sg:past:nal
Bitemop ate sago.
B: Woyo, nekheve ndu-nde e-nogma-tbo
No he sago-Foc neg-eat-,sg:past:nal
No, he didnt eat SAGO,
Diks Wambon data are from Vries I,8,.
Observe that wh-question words which identify the focus in the answer are treated as
though they themselves are foci both in Hungarian and in Wambon. In languages which
mark focus by intonation, however, the wh-constituent is not treated like a focus and is not
stressed. Instead, it is marked by fronting. For an analysis of focus and stress in wh-questions
see Erteschik-Shir I,8o and the references cited therein. See also Erteschik-Shir I,,,: chapter ,
on the different interpretations of wh-constituents in questions.
architectures and information structure I
nekheve ande-nde ande-tbo
he banana-Foc eat-,sg:past:nal
he ate BANANAS.
(,o) A: Nombone ndu-ngup ande-ngup?
This sago-and banana-and
What about this sago and bananas?
B: Wembane ndu-nde takhima-tbo
Wemba sago-Foc buy-,sg:past:nal
Wemba bought the SAGO,
Karolule ande-nde takhima-tbo
Karolus bananas-Foc buy-,sg:past:nal
Karolus bought the BANANAS.
Bs response in (,,) refutes the focused object and replaces it, and both the
refuted object and the one replacing it are marked with the focus marker.
In (,o), however, the sago and the bananas have been contextually dened
as the topic. In Bs response, which contrasts the two, these elements are
also marked with the focus marker. A characterization of focus following
from the distribution of the focus marker in Wambon has to include not
only the regular foci which replace wh-constituents in the answer, but
also contrastive foci, and contrastive topics.
Contrastive topics are also
stressed in languages in which foci are marked by stress (see (:8d) above).
Contrastive topics must therefore be classied both as topics and as foci.
In section ,.,, we show how this can be accomplished.
The second illustration of focus marking oered by Dik (I,,,: ,,o8)
is Aghem, a Grasselds Bantu language of Cameroon.
to Dik, Aghem is a tone language which does not use stress to mark
focus. Instead, it marks focus by means of constituent order and a
focus-marking particle (n). (In addition, it uses clefts and a special
Focus-bound Past tense marker.) The focus position is postverbal and
it hosts wh-constituents as well as regular foci. Elements in the post-
verbal focus position are contrastive. Contrastive subjects, however,
remain in situ. In a multiple focus reading such as the answer to Who
met who?, the subject does appear in the focus position and the other
focus (which answers the other wh-phrase) appears either preverbally
or in nal position, X and Y, respectively, in the schema in (,,) modi-
ed from Dik.
In order to understand focus marking in Wambon fully, one would also have to know
how VP foci and sentence foci are marked.
Dik cites Watters I,,, for these facts about Aghem but does not give any examples.
architectures and information structure
() S Aux X V Foc O Y
It follows that it must be possible to distinguish contrastive foci and
multiple foci since they appear in dierent positions in Aghem.
When the focus marker follows a constituent, it leads to contrast. If
the constituent marked by this particle is already contrastive, it is inter-
preted as only, excluding alternative options. In order to get the contras-
tive meaning of the focus morpheme, it could be viewed as an operator
which selects the focused element and eliminates the other member of
the contrast set from discussion. That, after all, is what contrast means.
If the same operator applies to a contrasted element, it again eliminates
the other element of the contrast set, emphasizing the uniqueness of
the focused element. The function of the Aghem focus morpheme is
therefore not just a way to mark foci, it is much narrower than that.
Although an account of stressfocus correlations is needed in some lan-
guages, this cannot be the basis of a denition which applies to all languages.
We are therefore left with a choice between a semantic and a pragmatic
characterization of the focus, since neither of these denitions depends on
the language-particular means of focus marking. The dierence between
these two approaches comes into play mainly in sentences with parts
that are neither focused nor presupposed. More importantly, as we have
already shown, the division of a sentence into focus and presupposition,
without acknowledging the function of a topic as well, will leave us without
sucient tools to account for important IS-properties of sentences.
The next section shows how the two IS primitives, topic and focus,
suce to characterize the dierent topic and focus types found in natu-
ral language.
2.3 Putting it all together: f(ocus)-structure
Several approaches to IS assume a binary division of the sentence. An
example is the semantic view of focus which sees the focus and the pre-
supposition (or the background) as complementary. The other type of
binary approach takes the topic as its point of departure and refers to the
rest of the sentence as the comment. A dierent terminology for the
same division is theme and rheme.
The previous sections have shown
that the IS notions topic and focus are both linguistically signicant in
For themerheme or topiccomment divisions, see Firbas ; Gundel ; Halliday
; and Jacobs among others. Hallidays denition of theme is the rst element in
architectures and information structure ,
that they can be marked intonationally, syntactically (by word order),
and morphologically. The binary approaches, even if they allow for
both divisions, tend to ignore the interactions between topic and focus
and view them as separate, independent entities.
In the next section
I introduce Erteschik-Shirs I,,, model of IS, f(ocus)-structure, which
examines the interaction between topic and focus, and show that this
theory enables an account of a variety of topic and focus types with-
out recourse to more than these two primitives. This model allows for
non-binary divisions of the sentence: The topic and focus do not neces-
sarily complement each other and the remainder of the sentence forms
the background as illustrated in examples (,:) and (,,) in section :.o,
repeated here for convenience.
(,8) John said that he knows Peter.
(,,) Q: Where is the book?
A: I gave it to Mary.
As shown above, the matrix of (,8) need not be attended to and hence
neither topic nor focus will be assigned to it. In (,,), the topic is I and
the focus is Mary. Both examples include a backgrounded part, namely
the part of the sentence to which neither topic nor focus is assigned.
2.3.1 Topicfocus interactions
F-structure is a structural description, annotated for topic and focus,
which interfaces with syntax and both semantics and intonation. Due
to topic marking this structural description is scopally unambiguous
(see section :., above), making LF redundant. Following Reinhart I,8I,
the common ground or context set (see section :.,) is (metaphorically)
the sentence. Therefore, his notion of theme differs signicantly from that of the others.
For focusbackground divisions, see, in addition to Chomsky I,,I and Jackendoff I,,:
cited above, Chafe I,,o; Prince I,8Ia; Rochemont I,8o; Ward I,88; Zubizaretta I,,8, among
others. See also Casielles-Surez :oo; Erteschik-Shir I,,,; Vallduv I,,; and Winkler and
Gbbel :oo: for overviews of the various approaches.
An exception is Steedman :ooob who makes a distinction between background/focus
as well as theme/rheme. For him both theme and rheme can be partitioned into background
and focus.
See Bring I,,, for the same division.
architectures and information structure
represented by a set of le cards.
Each le card represents a discourse refer-
ent. Entries on each card correspond to what is presupposed about the
discourse referent in question. The cards are organized so that the most
recently activated cards are to be found on top of the stack of cards. (Some
people nd it hard to envision a stack of cards with more than one card
on top. Another metaphor that works equally well is that of a drawer with
le cards, where the front section of the drawer is equivalent to the top of
the le.) These are the discourse referents which provide potential topics
in the discourse. How do cards get to be on top of the le? This follows
implicitly from the denition of the focus. If the attention of the hearer is
drawn to (the referent of) X, then the hearer (metaphorically) selects the
card for X and puts it in a place of prominence, namely on top of his stack
of le cards. The card can be selected from among the already existing le
cards if it is denite and therefore represents an existing referent. If this
is not the case, the hearer is required to make out a new card for this new
(indenite) referent and again, this card is positioned on top of the stack.
This is how the interaction between topic and focus comes into play. It is
also the explanation for both topic chaining and focus chaining. A topic
remains on top of the le and can therefore be continued. A focused ele-
ment can become a topic in the next sentence since focusing a discourse
referent requires the positioning of its card on the top of the stack.
The le system thus involves locating cards on top of a stack (topics) or
positioning them there (foci). In addition, each card manipulated through
the processing of an utterance is also updated with the information con-
tained in the utterance. Topic and focus are therefore dened as triggering
instructions to manipulate a stack of ling cards, each of which represents a
referent available in the discourse. The rules apply to referential constituents
within top/foc domains:
() F-structure Rules
a TOPIC instructs the hearer to locate on the top of his le an
existing card with the appropriate reference.
b FOCUS instructs the hearer to either
(i) open a new card and put it on the top of the le. Assign it
a new label (for an indenite) or
(ii) locate an existing card and put it on the top of the le (for
a denite).
See De Cat for a comparison between Reinhart , Lambrecht , Vallduv
, and Erteschik-Shir . All except for Lambrecht employ le card metaphors for dis-
course referents.
architectures and information structure
c UPDATE instructs the hearer to enter the focus on the topic card
and then to copy all entries to all cards activated by the focus rule.
Let us illustrate with the questionanswer pairs in () modied here:
() a Q: What did John do?
A: He WASHED THE DISHES. They are now clean.
b Q: What did John wash?
A: He washed THE DISHES. They are now clean.
c Q: Who washed the dishes?
A: JOHN washed them. He is very pleased with himself.
d Q: What happened to the dishes?
A: JOHN WASHED them. He is very pleased with himself.
e Q: What happened?
A: JOHN WASHED THE DISHES. He is very pleased with
himself. They are now clean.
The question in (a) presupposes that the subject John has already
been introduced and that a card for this referent is available on top of the
le. The dishes are part of the focus and so the card for this referent is
positioned on top of the stack. This enables focus chaining licensing the
continuation. (Continuations with topic chaining are not listed.)
In section ., it was suggested that pronouns must be interpreted as
topics. We are now ready to demonstrate why this is so. In processing the
answer in (a) the card for John which is available on top of the le is
updated with the content of the utterance as shown in ().
he washed the dishes
The subject pronoun is interpreted as coreferent with the referent of
the label of a card upon which it is entered. Since the pronoun he is
entered on the card labeled John, the pronoun is interpreted as corefer-
ent with John. If such a card were not available, the pronoun would not
be interpretable. The availability of a topic card is therefore necessary to
interpret a pronoun. It follows that pronouns must be topics.
Pronouns are sometimes ambiguous as illustrated in (i).
(i) John
gave a book to Bill
. He
loved it.
Ambiguity occurs when two cards are available on top of the le upon which the pronoun
could be interpreted.
architectures and information structure
In (b), the dishes are again focused and the account is the same. In
(c), it is the card for the dishes which is already on top of the stack, and
the subject John is focused. The card for John is therefore positioned on
top of the le, allowing the continuation with John as a topic. Similarly
for (d). The question in (e) triggers an answer with a stage topic.
Both the subject and the object are part of the focus and their cards must
therefore be put on top of the stack, licensing both continuations listed.
Remember that certain cards are always available as topics as dis-
cussed in section .. These are the cards for the speaker, the hearer, and
the current stage. All three follow from the discourse situation which
always involves a speaker, a hearer, and the spatio-temporal parameters
they are in. The speaker and hearers attention is automatically drawn
to these elements and therefore attention need not be drawn to these
cards by focusing. In section ., a further set of discourse referents were
singled out. These are the permanent xtures such as the moon, the
president, etc. These are members of the common ground shared by
speaker and hearer, which are in and of themselves prominent and can
therefore be positioned on the top of the stack with the help of a bit of
context, but without specic mention. In () (=(a) ), for example, the
moon is part of the inventory of the situation: darkness in a night-time
situation outside.
() Its dark. The moon has disappeared.
Yet another set of elements were shown to be potential topics with-
out previous mention. These are elements in the discourse situation
which can be pointed at. Pointing can be considered a form of focus-
ing, and so it is not surprising that pointing at a referent brings it to
the top of the stack.
The top of the le consists of several cards at any point in the dis-
course. These include not only the permanently available cards, but also
additional cards which have been positioned on top of the stack in the
course of the discourse. Which one is chosen at any particular point
and the lifespan of each card in that position is likely to be restricted by
broader discourse considerations which may impose an internal struc-
ture on the set of cards which belong to the top of the stack. For example,
in the processing of a sentence, a new card for a focus will be placed on
top of the card for the topic. In the following sentence, if the referent
for this uppermost card is selected as the topic, a shift topic is derived.
If, however, the card beneath it is selected, a continued topic is derived.
More complex considerations concerning the level of activation of
architectures and information structure
referents have been widely discussed, but will not be taken up here, since
they have little impact on syntactic structure.
2.3.2 Notation
It should now be clear how to assign topic and focus in simple sentences.
(We will return to matters of architecture and to the question as to how
these features arise in section .) In order to facilitate discussion, we use
the following notation. Topic constituents are subscripted top, and
focus constituents are subscripted foc. We again illustrate with our
questionanswer pairs:
() a Q: What did John do?
A: he
[washed the dishes]
b Q: What did John wash?
A: he
washed [the dishes]
c Q: Who washed the dishes?
A: John
washed them
d Q: What happened to the dishes?
A: [John washed them
e Q: What did John do with the dishes?
A: [he
[washed them
f Q: What happened?
[John washed the dishes]
The f-structures in (a), (b), and (f) are unmarked as suggested in
section .. This is due to the fact that in these f-structures the topic pre-
cedes the focus. In (a) and (b) the subject is the topic and in (f),
the implicit stage topic (the subscripts s and t stand for the spatial
and temporal parameters respectively) can easily be taken to precede the
sentential focus. When the object is the topic as in (c) and (d), the
identity of the topic cannot be resolved until later in the sentence, lead-
ing to the markedness of the construction.
(d) requires a comment
since the following f-structure might have been expected:
Among the many references dealing with various aspects of the activation level of ref-
erents are the articles in Walker, Joshi, and Prince and the references cited therein; Ariel
; Chafe ; Dryer ; and Shimojo .
The f-structures (c), (d), and (e) in which the object is the topic were assigned
the following f-structures in Erteschik-Shir :
(i) Q: Who washed the dishes?
A: top
washed them
architectures and information structure
() Q: What happened to the dishes?
A: [John washed]
(d) and () are in fact notational variants. In the former, the topic
is included within the focus, in the latter it is not. Yet a topic embedded
within a focus will still behave like a topic and therefore its inclusion in the
focus makes no interpretative dierence. The reason to prefer (d) to ()
is that in the latter, the focus, the subject + the verb, does not form a syn-
tactic constituent. By choosing the notation in (d) , we can maintain the
assumption that the foci and overt topics must be associated with syntac-
tic constituents. Doing so requires a slight modication of the stress rule
since it does not apply to topics embedded in this way within the focus.
This leaves (e) which has two topics. Each topic must have a focus
associated with it. (For convenience, I refer to each such top-foc pair as
an f-structure.) It follows that this sentence has two f-structures super-
imposed on each other: one of them is (a) (marked in bold) and the
other is (d). I assume that one of these takes precedence, and this is
the one for which the truth value is calculated. I also assume that the
f-structure that takes precedence is the unmarked one. These assump-
tions are intuitively plausible, but nothing hangs on them. The distinc-
tion between main f-structures and subordinate ones does play a role
in other cases, as we shall see below.
2.3.3 Contrastive and restrictive topics and foci
Contrast is contextually constrained to occur only if a contrast set is
available. This is the case both when the focus is contrasted as in ()
(= () ) and when the topic is contrasted as in ().
() Q: Which laundry did John wash, the white, or the colored?
A: He washed the WHITE laundry.
() B: Tell me about your brothers John and Bill.
A: JOHN is the smart one.
(ii) Q: What happened to the dishes?
A: [John washed them
(iii) Q: What did John do with the dishes?
A: top
[washed them
These are in fact a notational variant of the ones in (). The focus structure of sentences with
object topics under this annotation is then isomorphic to structures with topicalization. The ini-
tial implicit topic forms a chain which is pronounced at its tail. The premiss behind this notation
is that any departure from topic-focus order requires that the topic must be identied elsewhere.
This is what the chain notation indicates.
architectures and information structure
In both responses, one member of the contrast set provided in the context
is selected and in both answers, this element is stressed.
In () the
contrastive element answers a wh-question and is therefore a contrastive
focus. In (), the contrastive element is diagnosed as a topic by the
topic test. Even though the subject in () could also answer a contrastive
wh-question (Who is the smart one, John or Bill), it must be analyzed as a
contrastive topic, rather than a contrastive focus, because it is the subject
of an intransitive individual-level predicate and as such must be a topic as
shown in section ..
The theory of f-structure outlined in the previous section provides an
explanation for the fact that contrastive elements can function as both
topics and foci: The contextually available set in (), {John, Bill}, pro-
vides a topic, and the element selected from this set provides a focus.
This is shown in ().
() {John
, Bill}
The contrastive element has a subordinate f-structure which includes
both a topic and a focus. This is why contrastive elements can function
both as topics and as foci. The full f-structure of () is shown in ().
() [{John
, Bill}
[is the smart one]
The stress rule applies to the subordinate focus explaining the stressed topic.
The contrastive interpretation results from selecting one element and elimi-
nating the other from consideration. This subordinate topic, Bill, can be
dropped or pronounced:
() John, not Bill, is the smart one.
A further focus type with distinct properties is what I refer to as a
restrictive focus. Restrictive foci, like contrastive ones, require a context-
specied set:
() Q: Which one of his friends wants to meet John?
B: JANET wants to meet John.
Here the focus, Janet, is selected from the contextually specied or restric-
tive set of Johns friends, which provides a hypertheme in the terminol-
ogy used above in section .. (It is d(iscourse)-linked in Pesetskys ()
Contrastive foci are distinct from plain foci in that they are not constrained to syntactic
constituents. In (i), for example, the contrastive focus is on the prefex un-.
(i) He didnt TIE his shoes, he UNtied them.
architectures and information structure
terms.) Restrictive foci dier from contrastive ones in that the context set
need not be as clearly dened and therefore the complement of the selected
element is not eliminated. Janet in this case is not contrasted with any
other particular individual belonging to the set of Johns friends. Here
again, the set of Johns friends provides the topic of a (subordinate) focus
structure in which Janet is the focus. Restrictive elements are like contras-
tive ones in that they combine topic properties (they range over a context-
ually specied set) and focus properties (one element of this set is focused).
This is why they too can function both as topics and as foci. Restrictive foci
have a dierent distribution from non-restrictive ones. One example is the
fact that multiple wh-questions necessarily range over restrictive sets:
() Q: Who read what?
A: John read The Times, Peter read the New Yorker, and Susan
read The Economist.
The question in () is licensed in a context in which the set of readers
(John, Peter, and Susan) are under discussion. The set of papers (The
Times, the New Yorker, The Economist) may or may not be contextually
available. In Chapter , this observation will be shown to be partially
responsible for superiority eects.
Contrastive topics and foci are similar to restrictive ones, but they dier
in one property. Whereas the elements of restrictive sets need not be speci-
ed overtly and can be derived from world knowledge, contrasted ele-
ments must be clearly dened. This is because contrast divides the topic
set into two subsets; one such subset is selected, the other is eliminated.
To see this compare () and (). The interpretation of () involves not
only John being smart, but also the assertion that Bill is not smart. Bill is
thus eliminated as a possible topic of the predicate. In (), however, the
selection of Janet as the one who wants to meet John doesnt necessarily
mean that none of the other friends wants to meet him. The dierence
does not show up in the f-structure of the sentence; it is derived from the
manner in which the topic set is discoursally dened.
The approach outlined here aords an explanation for why contrastive
and restrictive elements can function as both topics and foci. There is there-
fore no need for more than one notion of topic and one notion of focus. It
is the interaction of topic and focus and the possibility of imposing one
f-structure on the other that derive the various topic and focus readings.
The particular f-structure assigned to contrastive and restrictive elements
is associated with a particular interpretation as outlined above. In addition,
it is not surprising that these elements are associated with particular stress
architectures and information structure ,I
patterns: in view of their focal component, they are stressed, whether they
function as topics or foci in the main f-structure, and in view of their
topic component, further tonal elements can be added, generating the
characteristic tunes associated with contrast and restriction. Bring I,,,
gives the following parallel examples in English and German respectively:
(,,) A: Did your wife kiss other men?
B: [My]
wife [didnt]
kiss other men.
(,) A: Hat deine Frau fremde Mnner gekt?
B: [Meine]
Frau hat [keine]
fremden Mnner gekt.
According to Bring, when (German) topics are pronounced with a ris-
ing pitch contour, then other alternatives to the one indicated by the
topic are considered to be relevant. This is the case even if the alternatives
have not been introduced in the context. (,,) is another illustration from
Erteschik-Shir I,,,: I,o (adapted from the analysis in Pierrehumbert
and Hirschberg I,,o of Jackendos I,,: classical example). The intona-
tion is marked using the notation introduced in Pierrehumbert I,8o: H
and L stand for high and low tones respectively, * stands for pitch accent,
and % stands for boundary tone. The combination L+H is the notation
for the fall-rise tune observed on the restrictive element here.
(,,) Q: What about the beans? Who ate them?
A: Fred
ate [{[the beans]
; [other foods]}
H* L L+H* L H%
According to Pierrehumbert and Hirschberg a set of salient foods includ-
ing the beans is evoked by the question. The beans is therefore assigned
a restrictive f-structure in the answer: The set consisting of the beans and
the other foods under discussion provides the subordinate topic, and the
beans provides the subordinate focus. This constituent, in turn, provides
the main topic of the utterance with Fred as its focus. The pitch accent
on the high tone is assigned by the focus stress rule, the L tone preceding
it is assigned due to the availability of the subordinate topic set.
2.3.4 Indenite topics
As mentioned in section I.o, contrastive and partitive indenites can be
topics. The examples in (,o) are repeated from (:8). (,oa) is contrastive
and (,ob) is restrictive.
For more details of the correspondence between f-structure and intonation see
Erteschik-Shir I,,,: chapter ; Erteschik-Shir and Lappin I,8,b; Erteschik-Shir I,,,.
architectures and information structure
() a A DOG is intelligent, a CAT is not.
b Two/some of the students are intelligent.
In both cases, the indenite foci are selected from discoursally available
sets allowing the constituents to function as topics (as well as foci) as
discussed in the previous section.
Specic indenites are another type of indenite which is licensed as
a topic. We have already given two pieces of evidence for this. Specic
indenites topicalize in Danish (section .) and they occur as subjects of
intransitive individual-level predicates (section .). An indenite can-
not be associated with a card in the le, since it represents a new entity,
and cards represent entities already introduced in the discourse. If there
is no card associated with an indenite, there cannot be a card on top
of the le for it, and hence, from what we have said so far, indenites
cannot be interpreted as topics. The fact that specic indenites can be
interpreted as topics therefore requires an explanation. The explanation
must have to do with specicity, since non-specic indenites are not
possible topics. Whereas an indenite is new to both speaker and hearer,
a specic indenite is new only to the hearer. Specic indenites contain
a modier which minimally indicates that the speaker has a particular
referent in mind. Examples are given in ().
() a A person I know is famous
b A friend of mine is famous.
c A certain student loves linguistics.
The subject of (a) which functions as the topic of the sentence has a
subordinate f-structure of its own:
() A person
In this f-structure the indenite is in fact the focus (and is therefore
stressed), and the topic is found within the relative clause (the st per-
son subject which is permanently available on top of the stack). The f-
structure of this constituent is therefore equivalent to the f-structure
of the simple sentence in ().
() I
know [a person]
These f-structures require the positioning of a new card on top of the
stack with the label person and the entry I know_. The entry indicates
that the speaker knows the person in question or more generally that
the speaker has this person in mind. If this f-structure is processed
architectures and information structure
before the f-structure of the complete sentence in (a), then a card for
the specic indenite will in fact be available on top of the stack, by the
time the sentence is processed. This does not mean that an ordering on
the processing of f-structures has to be imposed in this system. If the
opposite order is applied, the sentence will fail to get evaluated, since in
this case no card for the topic will be available on top of the stack. The
requirement that a new card is made out for an indenite focus enables
a new card to be positioned on top of the stack for a specic indenite.
The referent of this card can then play the role of topic of the sentence
as a whole.
The other two examples receive similar analyses. The adjective cer-
tain is interpreted as meaning that the speaker has someone in mind. It
is this interpretation of the adjective which enables the opening of a new
card for the student.
It is the fact that the f-structure system allows for subordinate f-struc-
tures which in turn trigger the introduction of new cards that allows a
simple explanation for the licensing of specic indenites as topics.
Subordinate f-structures of this type are not restricted to indenites
and also explain stress on complex denite topics such as (a) discussed
in section . above and repeated in (a) and (b).
() a (Any news of John?)
b [
Our poor chld] [
is in bed with a ]
c Our
[poor child]
(c) is the subordinate f-structure of the subject topic. The st per-
son subordinate topic our belongs to the set of permanently avail-
able cards and the subordinate focus on poor child explains the
stress on this constituent.
2.3.5 Some answers
We are now ready to tackle the syntaxIS relationship challenged by
Steedman a. Remember that Steedmans syntactic structures
are based on intonational phrasing and involve constituents such as
subject+verb. These structures are repeated in ().
MarCEL proved completeness MARcel proved completeness
() a b
architectures and information structure
Instead of positing such unorthodox syntactic structures, I propose
a mapping of f-structure onto intonational phrases.
Since foci are
anchored to syntactic constituents, but not to all syntactic constituents,
only partial isomorphism is predicted between syntax and prosody. The
intonational phrasing in (a) is consistent with an f-structure assign-
ment in which the subject is the topic and the VP is focused. Since the VP
is an f-structure constituent, it must be pronounced as an intonational
phrase. (b), with the intonational phrasing indicated (i.e., Marcel
proved forms an intonational phrase), can be an answer to: I know
which result Marcel predicted. But which result did Marcel prove?
This phrasing is also derivable from f-structure of the sentence. The f-
structure annotated tree structure is presented in ().
Here the distinction between intonational and intermediate phrases may become
In order for the interpretation to follow, the verb must be contrasted.
This is accomplished by assigning both top and foc to it. (In fact, the
topic set is {predicted/proved} with the latter focused.) My claim is
that intonational boundaries are derived from f-structure boundaries
and that the contrastive f-structure boundary between the verb and
its focused object forces an intonational boundary in that position.
In order to account for the full mapping of f-structure to intona-
tional phrases an investigation of which f-structure boundaries are
phonologically visible must be pursued. My point here is merely that
such an investigation should be made before positing special syntac-
tic structures for this purpose.
We started out by showing the need to distinguish shift topics from
continued topics, since these provided Links in Catalan. Shift topics
include contrastive and restrictive topics and these are easy to charac-
terize in view of their distinct f-structure. But how do we distinguish
continued topics from cases of focus chaining? Continued topics involve
repeated selection of the same card from the top of the le, and focus
chaining involves the selection of the most recently added card to the top
of the le. In order to distinguish these, a special status must be assigned
to the card most recently added to the top of the le. One would hope
() vp
proved[top]/[foc] completeness[foc]
architectures and information structure
that this kind of bookkeeping should not be necessary to explain natural
language phenomena. A more careful examination of exactly what the
class of shift topics encompasses within and across languages is needed
before any such move is made.
So far IS has been treated as though it stands apart from the rest of the
grammar. The next section attempts to tackle the question as to which
architecture best accommodates IS.
2.4 IS architecture
It is now clear that topic and focus play a pervasive role in grammar and
interact with phonology (intonation), morphology (top/foc markers),
syntax (linear order), and interpretation (quantier scope). These
notions are also critical in information structure in that they constrain
the possible sequence of sentences in discourse. In spite of the many
roles played by topic and focus there is little agreement as to how and
where they are introduced in the grammar. The architectures of multi-
level models of grammar (e.g., HPSG, LFG, CCG, RRG) allow for IS to
be integrated as a level on a par with argument structure, semantics, etc.
and therefore naturally incorporate IS. The architectures of grammar
posited in the Principles and Parameters model (Chomsky ) and
developed into the model of the Minimalist Program (Chomsky ;
; ), however, pose two obvious problems with integrating IS. In
this model, syntactic computation interfaces with both the phonological
component (PF) and the semantic component (LF), yet there is no inter-
face between PF and LF as shown in ().
As we have seen, IS has both phonological and interpretative repercussions.
It follows that IS features must be available to both PF and LF and must
therefore be present already in the syntax. Here is where the second prob-
lem comes in. As has often been pointed out (e.g., Zubizaretta : ), if
topic and focus features are not introduced lexically, they violate the con-
dition of inclusiveness, since they will have to be added in the derivation:
() Lexicon
,o architectures and information structure
(8) Inclusiveness
Any structure formed by the computationis constituted of
elements already present in the lexical items selected for the
numeration; no new objects are added in the course of computa-
tion apart from rearrangements of lexical properties. (Chomsky
I,,,: ::8)
One way of getting around this is to derive focus from stress which, in
turn, is assigned to syntactic congurations. This approach was dis-
cussed in section :.: and was found lacking. In addition to the prob-
lems mentioned there, this approach does not oer an account of topic
assignment and therefore also no account of the interaction between
topic and focus. Another approach which also does not violate inclu-
siveness is Diesings (I,,:) in which topic and focus are derived from
the syntactic structure (see section :.I). Diesings approach, however, is
geared to explain certain word order alternations, but does not aord an
explanation of the relationship between focus and stress.
Another problem which comes up in this connection is the optionality of
movement triggered by IS features. In the Minimalist Program (MP) move-
ment is triggered by obligatory feature checking, and optional movement
is therefore excluded. A number of dierent approaches to these issues are
available, each involving a dierent view of the theory of grammar.
Chomsky :ooI addresses these issues to some extent in his ana lysis of
Scandinavian object shift. According to Holmberg I,,,, Scandinavian
object shift moves topical objects across preceding adverbs. Holmberg
further proposes an adjacency requirement on this rule (see the next
section). It follows that object shift has both interpretative (the shifted
object is a topic) and phonological properties (adjacency). In view of the
split between LF and PF (see (8,) ), such a combination of properties is
problematic. Chomsky proposes the following account. First, he intro-
duces the economy principle, that surface semantic eects are restricted
to narrow syntax. Chomsky further claims that whereas certain displace-
ment rules do not involve surface semantic eects, and can therefore be
assumed to be phonological, object shift is driven by the semantic inter-
pretation of the shifted object (it is a topic) and must, at least partially, fall
within narrow syntax. But in view of the adjacency requirement, it must
also be phonological. Chomsky explains object shift and the distinction
between languages in which object shift occurs and those in which it
does not by positing the interpretative features INT and INT assigned
according to (8,) and (8o), where only object shift languages observe (8o):
architectures and information structure ,,
(8,) The EPP position of v*P is assigned INT. (Universal)
(8o) At the phonological border of v*P, XP is assigned INT.
(Object shift languages)
Although Chomsky does not admit to this, INT and INT equal topic and
focus respectively. (8,) is therefore equivalent to the idea that predicates
must have a topic. This according to Chomsky is a universal principle .
(8o), however, is a property only of languages that have object shift. It
states that the focus occurs at the edge of the VP. An unshifted object will
therefore be interpreted as a focus and not as a topic as required.
Chomskys INT and INT features do not play a role in narrow syntax,
but are interpreted at the interfaces. Their assignment according to the
rules in (8,) and (8o) does therefore not violate the inclusiveness prin-
ciple. Rule optionality is regulated according to (8,).
(8,) Optional operations can apply only if they have an eect on
Since object shift does have an eect on outcome, its optionality is
One problem with this approach is that it is not embedded in a gen-
eral theory of topic and focus, and it is, for example, unclear how foci
which are not at the VP edge are assigned. These issues could poten-
tially be worked out. Yet, as far as I can see, there is no way to connect
the assignment of the INT features at the C-I (conceptualintentional)
interface to their intonational properties at the A-P (articulatoryper-
ceptual) interface.
2.4.1 Stylistic components
Holmberg I,,, argues for a stylistic component, containing rules
dependent on focus structure. According to Holmberg, Scandinavian
object shift involves the movement of [Foc] objects. Such movement
is licensed when all phonologically visible non-adjunct material has
been removed from between the launching site and the landing site of
the movement, regardless of how it is removed. This type of constraint
on object shift excludes a syntactic analysis according to Holmberg and
points to a phonological one. Holmberg I,,,: , however, rejects the idea
that OS is a phonological operation, since it does not make reference
See Chomsky :ooI for the properties of v*P. These are not relevant for our purposes.
The idea that certain word order phenomena are stylistic stems from Ross I,o,.
architectures and information structure
to phonological primitives such as vowels, feet, nasality, etc. Nor is it a
morphological process, since it does not make reference to primitives
such as stems or axes etc. Holmbergs view of phonology is rather
strict and could be loosened to include movement which is restricted
by clearly phonological parameters such as adjacency. Adding an extra
component to the grammar as Holmberg proposes has to be weighed
against modifying the domain of phonology in this way.
This is exactly the tack taken by Kidwai . She gets around the
Inclusiveness Condition by arguing for an architecture in which the
focus feature is PF Interpretable and generated by UG. This type of
feature can be checked only in the phonological component, which
interfaces with a discourse component. Kidwais view is supported
by the fact that word order triggered by focus structure is constrained
by a variety of adjacency requirements. Such requirements are taken
to be characteristic of phonology rather than syntax. Her architecture
does not include topic features and does not allow for the derivation
of scope from f-structure since the discourse component does not feed
LF. Kidwai revises this architecture and posits that PF derivational
outputs interface with both the C-I and the A-P systems. In principle
this could enable interpretation of IS; however, Kidwais revised system
does not allow for LF access to the discourse component.
2.4.2 S-structure
Zubizaretta notes that subject foci in Germanic and French trig-
ger destressing of the non-focused constituents whereas the same focus
assignment in Spanish and Italian triggers the movement of the non-
focused elements, leaving the focused subject in sentence-nal pos-
ition. Compare the English and Spanish examples in () and () from
Zubizaretta : , both answers to Who ate an apple?
() Jhn ate an apple.
() Comi una manzana Juan.
In order to account for such data, Zubizaretta weakens the inclusive-
ness principle to allow for the features [F] (=focus) and [prosodic
prominence] to be introduced in the derivation at the point where
a single phrase marker has been derived. She refers to this point as
-structure. The computation continues and the the NSR (Nuclear Stress
Rule, see section .) and the Focus Prominence Rule (FPR) apply
here. (The FPR assigns more prominence to a sister node which is
architectures and information structure
marked [+F] than to one marked [F].) If these two rules have
contradictory outputs one of two things may happen: In certain
languages (Germanic and French) defocalized ([F] marked) con-
stituents become invisible with respect to the NSR (as in () ); in
other languages (Spanish and Italian) these constituents undergo
movement (as in () ). Zubizaretta refers to this kind of movement
as prosodically motivated movement or p-movement. P-movement is
therefore a way of moving non-focused material out of a focus position,
along very similar lines to Chomskys (: ).
At this point (-structure) the derivation branches to PF and Assertion
Structure. F-structures are associated with one or more Assertion Structures
and these Assertion Structures are in turn associated with explicit or implicit
context questions. Zubizaretta further identies -structure with LF, since
p-movement has an impact on LF. Zubizarettas model is diagrammed in
() (Zubizaretta : ).
Bailyn invokes prosodically motivated movement in functional form which is
equivalent to Zubizarettas Assertion Structure. Such movement is invoked in contradictory
situations in which Assertion Structure (i.e., context) requires a certain word order and
(sets of phrase markers, feature checking)
(F-marking, stress rules, p-movement)
-structure = LF
PF Assertion Structure
There are a number of aspects of this model that one might want to
question, the most critical is the need for a post-syntactic, pre-phono-
logical set of rules (the stretch between -structure and -structure).
These rules do not belong in the syntax, since they do not adhere to syn-
tactic constraints. They also do not belong in the phonology since they
are formulated in terms of hierarchical structure. This is why Zubizaretta
invents a special place in the grammar for them. This might be justi-
ed if Zubizaretta had singled out a set of rules which had something in
common. This cannot be said for such diverse rules as focus assignment,
stress assignment, and p-movement, each with its own properties and
constraints. Although Zubizaretta accounts for a broad range of cross-
linguistic phenomena, adding such a new component is a high price
to pay to accomplish this.
An additional peculiarity of Zubizarettas
architectures and information structure
architecture is that LF provides the input to the phonology. This is a side
eect of having F-marking aect LF as required. Yet the purpose of the
split between PF and LF is that the output of LF movement is not pro-
nounced. Not enough detail is given about Assertion Structure to evalu-
ate its function and position in the architecture. Winkler and Gbbel
in their extensive review of Zubizaretta make another important
point: Zubizarettas account of topicalization assumes a syntactic topic
feature. In addition F-marking in the post-syntactic component marks
topics as [F]. This duplication of features seems redundant. Winkler
and Gbbel consider how Zubizaretta would handle syntactic focus
movement in languages which move foci to the left periphery. Since
the focus feature is only introduced at -structure, it is not available in
the syntax. An additional syntactic focus feature is therefore required in
these cases, causing even more duplication in the system.
2.4.3 IS at PF
Neeleman and Reinhart set out to explain why topics must scramble
in Dutch.
(a) illustrates the normal OV order in Dutch. (b) illus-
trates scrambling in which the object precedes an adverb.
() a dat Jan gisteren het boek gelezen heeft
that Jan yesterday the book read has
b dat Jan het boek gisteren gelezen heeft
They adopt Cinque s procedure which derives the most prominent
stress on the most deeply embedded constituent. It follows that both
in languages with basic VO order and in languages with basic OV (e.g.,
Dutch), the main stress will fall on the object. In scrambled structures
(restricted to OV languages) the main stress will fall on the verb.
According to Neeleman and Reinhart, potential focus assignments
constitute a focus set which consists of the constituents containing the
main stress of the IP. The object is included in the focus set in the unscram-
bled structure, but not in the scrambled structure. () and () show the
focus sets for the two structures.
neutral intonation is used. He notes that infelicitous order is not ungrammatical and sees
this as evidence that such sentences are not the result of syntactic feature-checking crash.
According to Bailyn, languages differ as to which mechanisms are available for dealing with
such contradictions.
For a general discussion of scrambling see Chapter , section .
architectures and information structure oI
(,:) Non-scrambled Structure
a Syntax: [
AdvP [
DP V] ]
b Focus set: {IP, VP, Object}
c Object: Stressed
(,,) Scrambled Structure
a Syntax: [
DP [
AdvP V] ]
b Focus set: {IP, VP, V}
c Object: Destressed
Yet although it is included in the focus set, the object in the unscrambled
structure can be a topic. It must therefore be destressed. This is accom-
plished by destressing, one of two stress-shifting operations posited by
Neeleman and Reinhart:
(,) a stress strengthening: economy entails that stress strengthen-
ing applies only to derive foci not in the focus set.
b anaphoric destressing: applies to a constituent i it is d-linked
to an accessible discourse entity.
Stress-shifting operations are marked and only apply when necessary
for discourse reasons. Reinhart :ooo argues that optional operations,
like stress shift at the PF branch, or QR at the LF branch, are always econ-
omy violations.
If using an uneconomical operation is the only way to
satisfy a certain interface need, the derivation results in a grammatical
sentence. Yet although it is included in the focus set, the object in the
unscrambled structure can be a topic. It must therefore be destressed by
(,b), which is an uneconomical operation. Since an alternative struc-
ture is available in Dutch in which the object has scrambled and is there-
fore not included in the focus set, uneconomical destressing is ruled out.
Marking the object as a topic is therefore for free. It follows that topics
must shift in Dutch. In English, however, scrambling is not available.
Therefore, when the verb needs to be the sole focus, English must use the
stress-shifting operation in order to strengthen the verb:
(,,) a I have read the book yesterday, and did not tear it up
b # I have read the book yesterday, and did not tear it up.
The reasoning behind Neeleman and Reinharts approach can be viewed
in two dierent ways:
See (8,) for Chomskys version of this constraint.
o: architectures and information structure
(,o) a Under a movement analysis: move in order to avoid
b Under non-movement analysis: destress in order to avoid
In view of the minimalist exclusion of optional movement, Neeleman
and Reinhart opt for the second analysis and allow the base genera-
tion of both the scrambled and the unscrambled word orders, thus
accounting for the optionality of scrambling in Dutch. Since both
orders are base generated, neither order is more costly than the other.
Choosing the order that avoids destressing, a costly PF operation, is
the economical thing to do. Only in languages in which no other order
is available can discourse-motivated destressing apply. Their approach
is thus the opposite of Zubizarettas, discussed in the previous sec-
tion, which proposes the equivalent of (,oa). Zubizaretta avoids the
optionality problem: On the one hand, p-movement is, in fact, obliga-
tory, and on the other, p-movement is not syntactic and hence is not
required to be obligatory. Neeleman and Reinharts explanation of
Dutch scrambling is similar to Chomskys account of object shift, yet
the latter is too sketchy to enable serious comparison. Their approach
also does not violate inclusiveness, since the focus set is derived from
stress assignment which, following Cinques procedure, can be read o
the syntactic tree.
Van Gelderen :oo,: I, criticizes Neeleman and Reinhart for the look-
ahead problem with their approach: Since IS eects are essentially PF
eects, yet the choice between which structure to merge is in the syntax,
syntax has to look ahead to PF. In order to overcome this problem van
Gelderen suggests late adjunction (maybe even at PF) of adverbs fol-
lowing Lebeaux I,88. This raises a new problem since adverb order may
determine scope and therefore aects LF. Van Gelderen concludes that
MP architecture must be relaxed to allow syntax to be sensitive to both
PF and LF, a somewhat radical modication of MP theory. According to
van Gelderen :oo,, information structure is part of the PF Interface and
is free to linearly order the various constituents as it sees t. PF imposes
order on partial structures delivered to it by early spell-out. Yet only lan-
guages with certain particular case systems allow early spell-out. The
free word order resulting from PF ordering is therefore restricted to such
Details of Van Gelderens account of scrambling will be given in Chapter ,, section .
architectures and information structure o,
2.4.4 Functional features
One way is to introduce information structure notions (topic, focus) into
the syntax as functional features which project syntactic structure and trig-
ger movement (e.g., Horvath I,8I; Rizzi I,,,). In MP, triggers for word order
phenomena are morphosyntactic in nature. In the presence of an active fea-
ture, movement must occur, and in its absence it cannot occur. Optional
movement, in MP, cannot be syntactically motivated. Since IS-motivated
movement is optional, Rizzi allows for the optional projection of topic and
focus phrases which, once they are projected, necessarily trigger movement.
The topic and focus features will survive to the C-I interface in view of the
fact that they are interpretable, yet for the same reason they wont be visible
at PF. Intonation, in this framework, therefore remains unaccounted for.
There is also no account of the fact that topics and foci may remain in situ.
The only way Rizzi can account for this option is by not assigning them topic
and focus features. But in that case they will fail to be interpreted as topics or
foci as well. (Rizzis proposal is discussed in detail in Chapter ,.)
2.4.5 Lexical features
Erteschik-Shir I,,, assumes annotation of f-structure on the output of
syntax. F-structures provide the input to PF where stress is assigned and
also to interpretation. Since scope is calculated directly from f-structure,
LF is rendered superuous. A serious problem with this view is that it
violates the principle of Inclusiveness.
In order to overcome this problem, I argue in Erteschik-Shir :oooa that
top and foc features are introduced as a part of lexical selection where
each selection of a lexical item licenses an optional assignment of a top
or foc feature.
Rather than assuming that top and foc are heads that
project their own phrases (as in, e.g., Rizzi I,,,), I view top/foc as fea-
tures which are optionally assigned to lexical items. By analogy with
-features, they may percolate to the maximal projection of the lexical
item they are assigned to.
Breul :oo: : contends that topic and focus features cannot be lexical due to the fact
that constituents are formed in the derivation which do not exist at initial merge. Breuls
point is well taken and the theory of lexical insertion of these features goes hand in hand
with a theory of syntax which does not allow for the composition of new constituents in the
Breul himself assumes that the choice of which phrase is topic or focus is made in the
conceptual-intentional system of the speakers mind. He also posits a [ foc]-feature which
projects the FocP. This feature is assigned to the nite root verb.
I assume the possibility of percolation from N to its extended projection, DP. Similarly,
percolation is possible from I to CP allowing foc assignment to a full sentence.
architectures and information structure
(b) illustrates the set of assignments appropriate to the discourse
in (a): the subject is assigned top and the object is assigned foc. These
features percolate to the DP with no further percolation possible. (c)
is the merged (rough) tree structure, what I refer to as the f-structure of
the sentence. (This structure and the ones below are grossly simplied.
Most functional projections, including IP and CP, are ignored for pre-
sentational purposes.)
() a Q: What did John wash?
A: He washed the dishes..
b select dishes assign [foc]
select the no assignment
select wash no assignment
select he assign [top]
The f-structures derived in this fashion are interpreted according to
assumptions outlined in section . In view of the fact that the question
identies John as the topic and forces focus assignment on dishes,
the answer to the wh-question, the f-structure in (c) is indeed the only
appropriate one in the context of this question. (For sample derivations
of a variety of f-structure in this manner see Erteschik-Shir a.) In
this example neither top nor foc projected. Feature projection is illus-
trated in () in which the object is assigned top.
() a What happened to the dishes?
John washed them.
b select them assign [top]
select washed assign [foc]
select John no assignment
wash dp
the dishes[foc]
This example is symmetric to the previous one: in both cases the topic
is embedded in the focus domain, yet in the former case if the topic is
washed[foc] them[top]
architectures and information structure
removed, the remaining focus is a constituent, in the latter it is not. Since
the focus rule applies vacuously to the topics contained in it, the eect is
that what is interpreted as a focus in the latter case is not a constituent. It
follows that even though a syntactic assignment of focus to a non-con-
stituent is impossible we in fact derive a focus interpretation of a non-
constituent with a purely syntactic assignment procedure.
F-structure assignments can also result in a subordinate f-structure.
() shows how foc embedded in top is interpreted contrastively. In this
case, top is contextually construed as a set of two alternatives and foc on
one of these alternatives eliminates the other.
() a Q: What did John wash, the pots or the pans?
A: He washed the pots (not the pans).
b select pots assign [foc]
select the assign [top]
select washed no assignment
select he assign [top]
See Erteschik-Shir and Rapoport a; ; in preparation, for a theory in which
-assignment and structural/lexical case can be read off structure merged and projected by
verbal meaning components.
The contrastive interpretation depends on the contextual availability of
a contrast set, i.e., a set introduced overtly or implicitly (here, {the pots;
the pans}). The card for this set (positioned on top of the le) consists of
a conjunction of two individual cards. The assignment of top to the DP
must therefore be interpreted as a reference to the topic set as a whole.
The assignment of foc to pots and its percolation to DP make the same
syntactic constituent a focus as well. The focus rule applies to one member
of the conjunctive set and positions it, by itself, on top of the le. Update
enters the rest of the sentence on the card, and the correct interpretation
is derived, selecting the pots and not the pans as the answer to the
The architecture which follows from these assumptions is shown in
vp c
washed dp[top]/[foc]
the[top] pots[foc]
architectures and information structure
The output of narrow syntax is delivered to the C-I interface, which
depends on the conguration of initial merge and f-structure. At the
A-P interface, structures are linearized and may no longer be transpar-
ent to interpretation. Interpret thus applies to the merged structure, a
view reminiscent of the Aspects model of deep structure. This is possible
since LF functions are taken over by f-structure.
In view of the fact that topicalization is optional, I propose that it is
to be analyzed as part of the phonological linearization process. Other
optional dislocation processes should probably also be analyzed in this
way. This is particularly appropriate in cases such as this one in which
dislocation is to the edge of a prosodic domain. Dislocation is therefore
a property of PF computation. (For some evidence for these ideas see
Erteschik-Shir b; a; a.) Once PF dislocation is licensed, the
question arises whether all dislocation is phonological or whether syn-
tactic dislocation still applies. A model which allows only one kind of
dislocation is obviously preferable, yet removing dislocation from the
domain of syntax is a drastic measure for which much more evidence
must be found.
A comparison between this approach and the one involving top/foc
annotations of syntactic structures reveals that the two approaches
derive exactly the same f-structure. It follows that they are notational
variants, the only dierence being that lexical insertion does not vio-
late inclusiveness. In order to decide which approach is preferable, one
would have to reach a better understanding of the theoretical signi-
cance of the principle of inclusiveness at least as it pertains to topic and
focus features. We return to the special status of topic and focus in the
conclusion to this book.
2.4.6 Multiply-dimensional architectures
What the theories in this section have in common is that syntax, pho-
nological, and semantic information is encoded in parallel. Therefore,
PF computation
Initial merge (incl. top/foc) ()

architectures and information structure o,

in contradistinction to generative grammars, IS is easily integrated into
these theories in parallel to the other components. Frameworks which
have such parallel architectures to a greater or lesser extent include Head
Driven Phrase Structure Grammar (HPSG) (e.g., Sag and Pollard I,,),
Lexical Functional Grammar (LFG) (e.g., Bresnan :ooI), Dynamic
Syntax (e.g., Kempson, Meyer-Viol, and Gabbay :ooI), Jackendos :oo:
Parallel Grammar, Construction Grammar (e.g., Croft :ooI; Goldberg
I,,,), Representation Theory (Williams :oo,), and Role and Reference
Grammar (e.g., Van Valin and LaPolla I,,,).
The contributions of these
frameworks to IS vary. Here only a few examples of these architectures
and their contribution to IS research are oered.
In HPSG, according to Engdahl :ooI: I-I,, the building blocks of
the theory are multidimensional signs, containing information about
phonology, syntax, semantics (or content) as well as about the context
in which the sign is used. According to Engdahl, the HPSG architecture
allows an account of the focusprosody connection without recourse to
a syntactic feature as required in MP and its precursors, where phonol-
ogy and LF are necessarily mediated by syntax. In HPSG this connection
can be expressed directly in the sign by a constraint which says that any
constituent which contains a focus accent, must be interpreted as pro-
viding focal information and vice versa. The inclusion of the context
dimension also means that there is a natural way of incorporating IS
into the grammar.
In RRG, syntactic knowledge is stored in the form of constructional
templates. The templates give the morphosyntactic, semantic, and prag-
matic properties specic to a given type of construction. The templates
thus provide a way for syntactic constructions to be linked to focus struc-
ture. This framework thus allows formally for an interaction between
syntax and IS. According to Van Valin :oo,: I,I, there are two dierent
cases for the assignment of focus structure to sentence types, special-
ized templates and unspecialized ones. The former account for cases
in which a particular focus structure is associated with a sentence type.
In such cases, the focus structure is stored with a particular template.
Van Valin gives two examples: (IoIa) illustrates a wh-question which,
according to Van Valin, always has narrow focus on the wh-phrase in
the initial (precore) slot. (The dotted lines indicate the potential focus
RRG integrates Lambrechts theory of focus structure in, e.g., Lambrecht I,,.
Lambrechts work touches on many of the topics in this book and is discussed where rel-
evant. See in particular, Chapter .
architectures and information structure
Neither focus structure assignment is uncontroversial: Erteschik-Shir
argues that wh-phrases (except in echo questions and multiple wh-
questions) are not foci, and the preposed locative in locative inversions
may best be analyzed as a stage topic, excluding it from the focus domain
of the sentence.
This is, however, besides the point. What is theoretically
interesting is the option of having specialized templates to account for
structures which are by denition associated with a particular focus structure.
(Here RRG bears some similarity to Cognitive Grammar.) RRG also allows
This may indeed be Van Valins intention. He states that locative inversion structures
are always sentence focus, yet the template indicates that the actual focus excludes the PP.
() a
domain and the triangle the actual focus domain.) (b) is the special-
ized template for locative inversions (e.g., Into the room ran a mouse)
which are assumed to have sentence focus.
architectures and information structure o,
for unspecialized templates for simple transitive sentences which can be
associated with a number of dierent focus structures depending on con-
text. For these some mechanism must be devised to ensure that the correct
selection of focus structure is made with respect to a particular context.
For this purpose Van Valin incorporates von Heusingers I,,, DRT theory
of the interpretation of intonation introduced in Section :., above.
LFG models syntax as linked parallel structures (c(onstituent)-struc-
ture, f(unctional)-structure (subjectobject functions), and a(rgument)-
structure). In addition multiple instances of top, foc, and adjunct are
allowed. top and foc indirectly map to the argument structure by being
identied with, or anaphorically linked to, another syntactic function
(Bresnan :ooI: ,,). LFG thus allows for a mapping of these functions as
well. According to Payne (I,,,: I,, fn ,) in LFG, order does not fall out
from assumptions about congurational constituency. Rather, the very
LFG architecture opens the door to the possibility of mapping relation-
ships between order and (grammaticized) cognitive or discourse statuses,
grammatical relations, argument structure, or constituency.
Even though LFG architecture easily accommodates topic-focus assign-
ment, King I,,, points out that the approach fails when a head is assigned
either topic or focus and the domain of the assignment projects to the
sub-f-structure containing the head, yet the discourse function is in fact
contained within the domain.
One such case is when a verb is contrasted,
yet the focus domain cannot be narrowed down to just the verb. Kings
solution to the lack of overlap between f-structure constituency and the
assignment of topic and focus functions is to posit a separate level of
i(nformation)-structure. She thus reaches the same conclusion for LFG as
others have reached in generative grammar (see sections .I, .: above).
LFG has been enriched with Optimality Theory (OT). This enables
violable alignment constraints, such that the propensity for a focus to be
sentence nal could be expressed by there being a constraint aligning the
focus to the right. Since constraints are ranked, another constraint may
take precedence, so that in fact, foci are not required to be placed at the
right periphery in all cases.
Chapter , of King I,,, is an implementation of LFG to Russian word order. She pro-
poses a basic phrase structure in which all nodes are potentially optional. These nodes are
ordered by linear precedence rules. Annotations on the phrase structure determine the
grammatical and discourse functions of the constituents. These annotations may employ
functional uncertainty so that a range of possible functions is licensed.
See Broadwell I,,, for an example of how this is done and de Hoop :ooo; :oo, and Choi
I,,, for the application of optimality theoretical approach to the IS phenomena involved in
architectures and information structure
Williams : introduces Representation Theory (RT) which is
also based on the idea that sentences consist of several structural levels
or sublanguages in Williamss terminology: Theta Structure, Case
Structure (CS), Surface Structure (SS), Quantication Structure, and
Focus Structure (FS). The syntax of a sentence, in this theory, is a col-
lection of structures from each of these sublanguages, and a set of shape-
conserving mappings among them. When two structures, each from a
dierent sublanguage, are isomorphic, shape-conserving mappings
apply and one such structure is said to represent the other. A simple
example to illustrate this theory is Williamss (: ) account of the
dierence between English and Spanish with respect to the position
of the focus. In English, the focus need not appear sentence nally; in
Spanish it does. This means that English allows FS to be misrepresented
by SS in order to maintain an accurate CS representation; in other words,
shape conserving is maintained between SS and CS, but not between SS
and FS: Case structure wins over focus structure when it comes to word
order. In Spanish, the focus is restricted to sentence-nal position and
so shape conserving is maintained between SS and FS in this language.
Williams : recognizes the similarity between the architectures of
RT and LFG. According to Williams one of the main dierences between
the two is that the matching between levels is not an economy principle
within LFG as it is in RT where the exact isomorphism is the goal of
the relations that hold between successive levels.
Kempson, Cann, and Kiaer argue that the Dynamic Syntax
framework (DS) can capture the construal of topic and focus without
adding a special level of information structure to the grammar. In their
framework the dynamics of the parsing process, which incrementally
builds an interpretation as the sentence is processed, constitutes the
grammar formalism and provides the necessary tools to explain topic
and focus eects cross-linguistically. Importantly, all context-dependent
aspects of interpretation are represented. Although intuitively sim-
ple, the formalism of the framework is too intricate to introduce here.
Instead, I have selected one example from their work in order to infor-
mally demonstrate the power of the DS system. In view of the fact that
sentences are processed from left to right, the DS system can easily pre-
dict dierences in interpretation between elements on the left periphery
and on the right:
() As for John, he was sick
() She talks too fast, Ruth Kempson.
architectures and information structure
According to Kempson et al., topic is a structure that provides an
antecedent for later identication by some anaphoric device. This
explains the role of the initial topic element in () and the anaphoric
pronoun which identies it later in the sentence. In (), the right-
peripheral element is also interpreted as topical, but it has the eect of a
reminder. This follows from the order in which the topic and anaphoric
pronoun are processed in the two constructions. When the pronoun
comes rst as in () it (being a pronoun) already has a contextually
provided value, and the element on the right periphery has to agree
with it. Such a reading is not available at the left periphery since the
left-peripheral element itself provides the context for the value of the
The architecture of this framework, in which contextually sensitive syn-
tactic tree building is accomplished though processing, is therefore poten-
tially suitable to account for the ISsyntax interface, and the insights it
provides follow from its architecture, in particular the fact that the gram-
mar is built up incrementally.
This section demonstrates that architectures can be more or less
congenial to incorporating IS into the grammar. It is particularly prob-
lematic to incorporate it into generative grammar, yet exactly because
of this diculty, interesting research has been engendered. A problem-
atic architecture is therefore in and of itself not a bad thing, although
serious research into IS within MP may trigger serious revision of the
architecture itself. The multidimensional architectures accommodate
IS with ease and as a result tend to focus on IS and its interaction with
other levels to a large extent, leading to much interesting IS research. An
example is RRG, which incorporates the IS theory of Lambrecht ,
which could in principle be incorporated into one of the other theories.
Yet the pairing of this theory with RRG has in fact triggered the develop-
ment of both. The contrast between actual vs. potential focus domain
is not found in Lambrechts work, for example, and the distinction has
led to work on focus structure in dierent types of embedded clauses,
discussed further in Chapter , section .. The incorporation of IS as an
integrated level into RRG has also led to interesting cross-linguistic research.
A basic claim in RRG theory is that IS plays a role in the grammar of all
languages, but the particular role of IS in each language varies.
See Wedgewood for an application of Dynamic Syntax to Hungarian word order.
See Van Valin and LaPolla for discussion.
,: architectures and information structure
2.5 Functionalism
2.5.1 Functionalism vs. formalism
According to Newmeyer I,,,: 8:,
The functionalist literature is replete with discussions of the difculties in dening and
identifying such constructs as topic, focus, theme, and so on. And formal syntacti-
cians, on the relatively rare occasions when they choose to make reference to cognitive
and discourse-based units, tend to look to the rich literature in formal semantics. Clearly,
a major task aheadand one in which formalists and functionalists can work jointlyis
to arrive at even good descriptive generalizations about the nature of these units.
Newmeyer proposes that such units should be derived from indepen-
dently-motivated human cognitive attributes. Yet he complains that
there is no theory of cognition in existence from which the properties,
say of discourse topics (whatever they may be) fall out as a special case.
Functionalism is a cover term for a number of linguistic approaches
which seek to construct a pairing between form and function.

Functionalists are, according to Newmeyer I,,,, empiricists and they
therefore generally have an interest in data derived from actual language
use. Since sentences are used in context, functionalists necessarily take
into account contextual factors in the analysis of sentences. It follows
that functionalists are necessarily concerned with the dynamics of dis-
course and the factors which govern it. It is therefore natural to look for
work on the syntaxIS interface in the work of functional linguists.
Unfortunately, much functionalist work is not easily accessible to formal
syntacticians of the generative persuasion and vice versa. (The purpose of
Darnell et al. I,,, is to mend this rift.) Newmeyer I,8: illustrates the prob-
lem in his discussion of Givn I,,,. According to Newmeyer, the purpose
of Givns book is twofold: rst to show that transformational generative
grammar is useless and second to motivate an approach to syntax which is
based on communicative function and discourse pragmatics. Givn thus
belongs to the post-generative tradition which adheres to the belief that
generative grammar has no place for function. Newmeyer points out that
this is, in fact, a fallacy and quotes Chomsky I,,,: ,o8 as follows:
Cognitive Grammar, Construction Grammar, and Role and Reference Grammar
are considered to be functionalist by some although others nd them too theoretical.
References to these works can be found throughout this book. Here I list only a small sample.
Construction Grammar: Croft :ooI; Goldberg I,,,; Cognitive Grammar: van Hoek I,,,;
Langacker I,8,; I,,I; :ooI; Role and Reference Grammar: Van Valin I,,,; I,,,b; :oo,; Van
Valin and LaPolla I,,,. Most functionalist work is independent of a particular model of
grammar. Examples are Givn I,,,; I,8; I,8o; Gundel I,,; I,88; I,,8; :ooo; Kuno I,8,;
Kuno and Takami I,,,; Lambrecht I,,; :ooo; Prince I,8Ia; I,8Ib; I,,,; I,,8.
architectures and information structure
I have never suggested that there is no interesting connection between the structure
of language and its purpose, including communicative function, nor have I arbitrarily
assumed that use and structure do not inuence one another. Surely there are sig-
nicant connections between structure and function; this is not and never has been
in doubt Where it can be shown that structures serve a particular function, that is a
valuable discovery.
There was, however, good reason for Givns belief at the time, since func-
tion was generally far from the concern of generative grammarians then.
Even though functional notions such as topic and focus have been intro-
duced into generative grammar and play an important role in syntax (e.g.,
Rizzi ), they have been introduced without much notice paid to their
functional role in the discourse. Unfortunately, many linguists who did
take function and discourse seriously took their anti- generativism to an
extreme and also became anti-formal. As a result, they investigated func-
tion and discourse without much concern for structure, and so serious
research into the interface between formal syntax and information struc-
ture has suered from neglect from both research perspectives.
According to Payne : , functionalism is an approach which views
language as a tool for communication, and in which both the communi-
cative jobs to be accomplished vis--vis the hearer, and the general cogni-
tive makeup and language processing constraints of the speaker/hearer,
can aect the online production. Across time, these factors also aect
the grammaticalization of linguistic forms. Payne views the Chomskian
approach to word order as incompatible with the functionalist approach,
yet other formalist approaches such as LFG are potentially compatible
with it. There is therefore no clear dividing line between functionalism and
formalism as such. One of the characteristics of functionalism, according
to Payne, is that
functionalists are interested in understanding speakers motivations for the choices
between propositionally equivalent but formally distinct structures, especially when
there is a rule-governed relationship between a functional factor and the formal struc-
ture (e.g. motivations for choosing a passive over an active construction).
Generative formalists, according to Payne, allow for discourse to play
a role in explaining phenomena such as word order, yet they consider
such explanation as external to grammar in opposition to her own
view. Newmeyer : points out that linguists such as Ellen Prince
(e.g., Prince ; Lambrecht ), who investigate formfunction
Bolinger expresses this succinctly: When we say two things that are different, we
mean two different things by them.
architectures and information structure
interaction, also adhere to a modular view of syntax and argue that the
formfunction interface cannot be derived from extra-linguistic abili-
ties. Architecture is therefore also not a clear diagnostic of functionalism
versus formalism.
An illustration of what Payne has in mind is in order. Payne :
makes the following generalization with respect to a Wyampi text: Once
a new topic is established, nominal references to it follow the verb, except
when such a topic participant occurs in a clause which serves to mark
o the end of a thematic unit, signaling that the current major topic is
ending its tenure, and that a dierent participant will be the next major
topic. Focused (contrasted and questioned) phrases, however, occur
preverbally. It is feasible that a formalist theory can account for the
postverbal position of topics and the preverbal position of foci, yet the
textual marking of the last occurrence of a topic in preverbal position
would probably go unobserved.
Payne presents formalist generalizations made in Hale ; Hale and
Selkirk . These generalizations follow from the need of the preverbal
complement to be lexically governed by the verb explaining asymmetries
concerning intonation, determiner allomorphy, and extraction. Payne
would like to see a comprehensive analysis which covers both types of
generalization. Yet the potential for a formalist explanation of the func-
tional data, and the potential for a functional explanation of the formal-
ist data, is very small. The linguistic literature rarely oers the possibility
of comparing a formalist and a functionalist explanation that cover
exactly the same data, since it is in the nature of the two approaches to
focus on data of dierent types.
Another example in Payne : is a series of rules that account for
word order in Oodham (Papago). It is the combination of the set of rules
that predicts word order, yet each particular rule is followed in a high per-
centage of cases, but not in all cases. Two of these rules are listed in ():
() Information is preverbal when it is:
a pragmatically marked. This includes information which is
contrastive, questioned, answers to information questions,
and other similar statuses.
b Nonidentiable and discourse-topical/important.
A formalist will probably baulk at terms such as pragmatically marked
or important and will probably wonder why important goes with
topical rather than with non-topical or focused.
architectures and information structure ,,
DeLancey :ooI, in accounting for topicalization, attempts an explana-
tion for why it is that topics tend to occur in initial position. His account
is intended to avoid the type of circularity built into Paynes rules.
DeLancey points out that in a wide range of languages, across various
language types, we nd a construction in which a constituent occurs in
sentence-initial position, which ordinarily would occur elsewhere in the
sentence, and that in language after language, this construction is used
when the constituent is a contrastive or resumptive topic. He gives a
couple of examples including the following English one:
(Io,) Costello Id hire in a minute.
According to DeLancey, a legitimate explanation for the typological
facts here must oer an account which provides a principled reason for
the association of topic function with initial positionotherwise it is
not an explanation, merely a description. His explanation is based on
the psychological fact that the initial element in a series has a perceptu-
ally privileged position in the research of Gernsbacher (e.g. Gernsbacher
I,,o; Gernsbacher and Hargreaves I,88; and, in more recent work, Kim,
Lee, and Gernsbacher :oo). One of the early experiments showed that
participants recognition latencies to the probe Tina were faster in (Ioo)
than in (Io,):
(Ioo) Tina beat Lisa in the state tennis match.
(Io,) Lisa beat Tina in the state tennis match.
It follows that the initial component is most accessible. In most English
sentences, the initial element is the subject and/or the agent, hence it is
dicult to separate out the role of the initial position versus the role of sub-
jecthood or agenthood in these experiments. Kim, Lee, and Gernsbacher
:oo is a report of similar experiments in Korean which has much freer
word order than English and in which there is no correlation between ini-
tial position and subjecthood. In Korean, it is also possible to distinguish
initial position from agent in that the passive is marked morphologically
and not by word order. The results of the Korean experiments strongly
conrmed the previous results. The question then is, what are the linguis-
tic repercussions of the fact that rst position is most easily accessible?
Kim, Lee, and Gernsbacher :oo: ,o suggest (based on Gernsbacher
and Hargreaves I,88 and Green I,8, (see also Givn I,,,: 8,) ) that rst
mentioned information provides importance and summons attention.
According to such a functional account, speakers and writers use the
architectures and information structure
passive voice or grammatical inversion to emphasize the patient or
object of a sentence. For example in the case of spoken English, speak-
ers usually put the important constituentthe focus, topic or theme of
the sentencerst, and listeners are assumed to construct a representa-
tion of the sentence based on the shared pragmatic knowledge of lan-
guage use. The rst point, namely that rst position is accessible because
it summons attention, is testable and maybe even has been tested. Let us
assume, for the sake of argument, that this correlation is solid. The sec-
ond point, namely that therefore the important constituent, the focus,
topic, or theme, comes rst, brings us back to the question we raised con-
cerning Paynes generalization in () as to what it means to be linguistic-
ally important. Kim et al. list topic and focus as both being important,
Payne only topical. Going back to DeLancey, remember that topicaliza-
tion targets not only regular topics, but also contrastive topics. A generous
reading of Kim et al. would consider their mention of focus to refer to
contrastive topic, since for some researchers stress is automatically associ-
ated with focus, and the properties of contrast are ignored. We could then
derive that topics (including contrastive ones) are placed in rst position
because they require attention. Yet this is quite counterintuitive. After all
a topic is already attended to by not only the speaker but also the hearer,
and so there is no need to further draw the attention of the hearer to it. If
there were, we would predict that topics would never be dropped as they
commonly are. Something is clearly amiss in this interpretation of the
functional explanation at hand and I dont think it would pass DeLanceys
requirements for what counts as an explanation.
The authors enhance their explanation with the following from
Gernsbachers Structure Building Framework according to which
rst-mentioned information is more accessible because it serves as
a cornerstone to enable integration (mapping) of further information
(Kim, Lee, and Gernsbacher : ).
This is intuitively plausible,
but does not pick out topics as the element which must be perceptu-
ally privileged as DeLancey would have it. The search for a functionalist
explanation of topicalization, based on general cognitive properties, was
therefore not successful in this case.
DeLancey proposes that the gureground organization of percepts
is another cognitive property which may have been adapted to language. The
sentences in () and () are organized dierently with respect to gure
and ground:
Compare, among others, MacWhinney ().
architectures and information structure
() The bank is next to the Post Oce.
() The Post Oce is next to the bank.
() describes the location of the bank, using the Post Oce as a refer-
ence point, and () describes the location of the Post Oce, using the
bank as a reference point. According to DeLancey, the subject of each
sentence denotes the referent to which the speaker wishes to draw the
addressees attention, and the oblique NP denotes a referent used as a
background against which the subject can be identied. DeLancey thus
draws a parallel between visual perception and the organization of the
elements in these two sentences, and it makes sense that the dierent
choice of subjects makes for a dierent point of view. DeLancey is careful
not to associate the gureground distinction with other linguistically
relevant functions, e.g., topic or focus.
This can be tested by applying
the question test: () and () answer the questions in () and ()
respectively and the opposite questionanswer pairing is impossible.
() Where is the bank? It is next to the Post Oce. #The PO is next to
the bank.
() Where is the Post Oce? It is next to the bank. #The bank is next to
the PO.
The test shows that subjects are topics and that the PPs, which answer
the wh-question, are focused. This result is exactly the opposite of the
approach taken in the work of Erteschik-Shir. There gure and focus
are associated since both are what the attention of the perceiver/hearer
is drawn to.
It seems then that the linguistic description of a visual
percept is not divided in a parallel way into gure and ground, in that
the visual focus and the sentential focus are not aligned. I conclude that
although I agree with DeLancey that the gureground distinction plays
an important role in linguistic theory, it is not the role that DeLancey
envisions. This may also be the reason why focus of attention has been
mistakenly related to topic as noted above.
It is possible that this is due to his belief that there is no predened set of functions. A
more formal approach would seek to narrow down the set of functions as much as possible.
It is not obvious that there is an association between ground and some linguistic con-
cept. This is not strange since the ground is the complement of gure and it does not have
any interesting properties of its own as far as I know.
architectures and information structure
To conclude this section, I might add (in spite of my own work in this
direction) that it may be premature to search for a cognitive basis for
IS concepts such as topic and focus at this stage. As shown in this chap-
ter, there is little agreement among linguists as to what IS theory should
be, and therefore it is dicult to employ psycholinguistic ndings to
explain it.
2.5.2 Functionalist methodology
In the functionalist literature, one can often nd criticism of what counts
as data among those formalists who rely on grammaticality judgements
of informants. The criticism is aimed not only at this particular meth-
odology, but also at the fact that judgements are solicited on sentences
outside of context or without considering their intonation patterns. It is
obvious that in order to decide whether discourse functions such as topic
and focus play a role in word order, for example, one would have to have
judgements for the contextual naturalness of sentences. Grammaticality
outside of context is irrelevant in such cases.
Functionalists often base their generalization on real data, such as
that found in recordings or in written texts. Such sources are considered
to be authentic and therefore not open to questioning. This method is
also problematic since one cannot conclude from the absence of a par-
ticular structure in such sources that it is not a possible sentence in the
language. Functionalists also often rely on statistical judgements. This
goes hand in hand with the functional assumption that cognitive pro-
cessing during actual discourse production can aect the order of elem-
ents (and, in a historical way, the eventual grammaticalization of new
structures) (Payne : ). Finally, as shown in the previous sec-
tion, functionalists also rely on psycholinguistic experiments that enable
them to link linguistic properties to other cognitive processes.
Poulsens thesis includes a whole chapter on methodology in
which he questions the reliability of real data used by functionalists as
well as data gathered from corpuses. He discusses experimental methods
that have been used to further standardization in the use of acceptabil-
ity data.
He concludes that each methodology (subjective judgements,
objective judgements, corpus studies, and reading time experiments)
has its advantages and disadvantages.
The topic of Poulsens thesis is what he refers to as sentence inter-
twining in Danish. These are cases of extraction out of subordinate
See references in Poulsen .
architectures and information structure
clauses often involving violations of island constraints.
He compares
the data in various studies of the phenomenon and nds that the data
in Erteschik-Shir ; Erteschik-Shir and Lappin are as reliable as
data gleaned from a corpus. It seems to me that there is at least some
correlation between the reliability of linguistic data and the care with
which they are solicited. It is probably very hard for a linguist to evaluate
data in a language in which she or he does not have native intuitions, yet
as has often been observed, relying on ones own native intuitions often
leads to bad data. One way to get around this is for a linguist who does
have native or near-native intuitions in some language to solicit data
from other native speakers in order to get the data corroborated. Data
related to the syntaxIS interface require even more of an eort. For this
type of data, context and intonation must be controlled for as well. And
as Paulsen states there is no easy way to accomplish this.
Although functionalism seems, at least in principle, to be best geared to
probe the ISsyntax interface, valuable insights are to be gained from
the whole spectrum of theories that deal with the issue. Unfortunately,
much energy on all sides has been spent on linguistic wars between the
proponents of the various theories. Rarely have these wars produced
interesting ideas; instead, they have generally been waged with very little
understanding of the tenets of the other side, leading to only more and
more misunderstanding. I hope this book will help further a more con-
structive dialogue leading to a better understanding of the division of
labor between syntax and IS.
This chapter provides an outline of an array of IS theories (without
attempting anything like a full survey) with the aim of familiarizing the
reader with the methodologies employed to account for syntaxIS inter-
actions. The next chapter deals with word order, a topic where IS factors
play a central role. It follows that any analysis of word order both shapes,
and is shaped by, the particular architecture of grammar assumed. The
direction of future linguistic research, as pointed out by Bailyn a,
may therefore in part be determined by the results of ndings in this
See Erteschik-Shir and Chapter , this volume, for an account of extraction and
an explanation of why Danish (and other Scandinavian languages) allow extraction out of
more subordinate sentence types than English, for example. Traditional Danish grammar-
ians refer to the result of such extraction as knot-sentences because a knot is tied by intro-
ducing an element from a subordinate clause into the main clause.
In the previous chapters, we have seen that topic and focus play a central
role in all languages, but that these notions can be signaled in a variety of
ways. In this chapter, we examine the relationship between word order
and topic-focus articulation. First, we examine languages in which the
basic word order functions to signal topic and focus and contrast them
with languages in which basic word order functions to signal syntactic
argument type. We continue to investigate the role of the left periphery of
the sentence in determining topic and focus roles (section :). Fronting to
the left periphery can also be altruistic. In such cases the fronted element
itself does not play a particular IS role, yet its movement enables another
element to take on an IS role it could otherwise not have. Altruistic front-
ing is examined in section ,. In section , we show that word order also
marks information structure in cases of scrambling. Scrambling is then
compared to Scandinavian object shift (section ,).
3.1 Congurational languages
Some languages allow a variety of word orders. Such languages are commonly
referred to as non-congurational languages with a at, non-hierarchical
structure. In these languages, there is no direct mapping between structure
and grammatical role.
Instead, word order is determined by information-
structural considerations.
Languages in which word order is highly restricted are referred to as
congurational languages. In such languages the overriding function
of word order is to identify the subject and the verbal complements.
English exemplies a congurational language. The word order in such
According to Carnie and Harley :oo,, this conclusion led many functionalists and for-
malists (as in the LFG and RG traditions) to abandon structural accounts of word order phe-
nomena. Ross I,o, proposed that such languages had a stylistic or purely phonological rule
which took a fully specied hierarchical structure and scrambled its elements. Hale I,8:
proposed a congurationality parameter which allowed free ordering of elements.
configurations 8I
a language is relatively xed and the topic and focus constituents are
marked by intonation and prosody. This does not mean that word order
never plays a role in identifying topic and focus in English. The examples
in (I) illustrate cases in which word order marks constituents as topics
(in italics) and the examples in (:) illustrate cases in which word order
marks the focus (in capitals):
(I) a The rst part of the movie John missed out on. (Emonds :oo: 8,)
b Into the pool Mary jumped. (Emonds :oo: 8o)
c Mary, why must she always be late?
(:) a There is a FLY in my soup.
b It was a MOSQUITO that bit me.
c What I saw was A BIG WASP.
In section :, it will be shown that the left-peripheral constituents, such
as the ones in (I), are interpreted as topics. The existential structure in
(:a) marks an informational focus (Chapter I, section :.I) whereas the
clefts in (:b) and (c) mark a contrastive focus. These constructions are
discussed further in sections ,., and ,.o, respectively.
Although constructions exist in English that mark topic and focus,
English sentences in general do not require such marking and word
order is mostly determined strictly by syntax.
In Hungarian, in contrast, the focus necessarily occurs in a desig-
nated position immediately preceding the verb and overt topics occur
sentence initially. Hungarian word order is therefore governed by IS
and not by the need to identify the subject and the object. Hungarian is
therefore non-congurational in this sense. Yet Hungarian is congu-
rational in the sense that its word order is restricted by topic and focus
marking. Hungarian can therefore be categorized as a discourse con-
gurational language (e.g. . Kiss I,,,). In such languages, word order
is not determined by syntactic function and the position of subjects
and objects is not xed. The following Hungarian examples are from
. Kiss I,,8: o8::
(,) a Jnos [
IMRT mutatta be Zsusznak]
John Imre.ACC introduced prev Susan.DAT
John introduced to Susan IMRE.
b Zsuzsnak [
JNOS mutatta be Imrt]
Susan.DAT John introduced prev Imre.ACC
Susan was introduced to Imre by JOHN.
c Imrt [
ZSUZSNAK mutatta be Jnos]
Imre was introduced by John to SUSAN.
According to . Kiss, (a) is a statement about John, (b) is about Susan,
and (c) is about Imre. (In each case, stress on the preverbal focus is
marked in capitals.) It follows that the topic cannot only be the subject
(the most common case), it can also be any one of the fronted objects.
The word orders illustrated in () are S O V IO, IO S V O, and O IO V S,
respectively. Discourse-congurational languages such as Hungarian are
therefore considered to have free word order. Free-word-order languages
in this sense may have hardly any restriction on the order among their
verb, arguments, and adjuncts, yet word order is restricted according to
the position of the topic and the focus.
Van Valin c: in fact proposes a four-way contrast. As men-
tioned in Chapter , section ., in RRG languages (and construc-
tions) are distinguished according to their potential focus domain.
According to Van Valin, one distinction is whether or not the lan-
guage has a rigid syntax, the other is whether or not it has a rigid
focus structure: languages in which the potential focus domain
is the entire main clause in simple sentences will be considered to
have exible focus structure, whereas those in which the potential
focus domain is restricted to a subpart of the main clause will be
considered to have rigid focus structure. The following table is
Van Valins categorization of several languages according to these
Rigid Focus Structure Flexible Focus Structure
Rigid Syntax French, Toba Batak English, Toura
Flexible Syntax Sesotho, Italian Russian, Polish
Even with this four-way distinction, it is still dicult to draw a strict line
between congurational languages and discourse congurational lan-
guages. Take, for example, Germanic languages which require the verb to be
in second position and in which the topic occupies the preverbal position
( () and () are in Danish):
() a Peter s jeg igr.
Peter saw I yesterday
I saw Peter yesterday
configurations 8,
b Jeg s Peter i gr.
I saw Peter yesterday
(o) a Ham s jeg igr.
him saw I yesterday
b Jeg s ham igr.
I saw him yesterday
As noted in the previous chapter, topicalization is optional in Danish.
The sentences in (,) and (o) can all be used in a context in which Peter
has been introduced in the previous sentence, e.g., Tell me about Peter.
On the one hand, fronting marks a constituent as the topic of the sen-
tence; on the other, the topic can equally well be interpreted in situ.
Danish therefore probably should not count as a discourse-congura-
tional language in spite of the fact that topicalization occurs in a very
large proportion of naturally uttered sentences.
Languages may choose to mark only the topic or only the focus con-
gurationally. A diagnostic of the former, also referred to as a topic-
prominent language, is that categorical and thetic judgements are
syntactically distinguished (. Kiss I,,,: 8).
(,) illustrates categorical
judgements (in Hungarian) in which the subject is the topic. In such
cases, the subject precedes the VP. The sentences in (8) are thetic. Here
the subjects follow the verb.
(,) a [
Fido] [
rg egy scontot]
Fido chews a bone
Fido is chewing a bone.
b [
A kutya] [
the dog domestic-animal
The dog is a domestic animal.
(8) a [
Esik az es]
falls the rain
It is raining.
b [
Van egy kutya a szobban]
is a dog the room-in
There is a dog in the room.
See Chapter :, section I. on the distinction between thetic and categorical sentences.
See Li and Thompson I,,o for denitions of subject-prominent and topic-prominent
c [
Bejtt egy kutya a szobba]
came a dog the room-into
A dog has come into the room.
In the sentences in (), the topic is identied with the subject, yet it is
not always obvious whether such examples are to be analyzed as Topic-
VP structures rather than Subject-VP structures. The way to tease these
apart is by examining cases in which the subject is non-specic. . Kiss
oers the following examples:
() a [
Fido] szerintem [
szereti a csontot]
Fido according-to-me likes the bone
Fido, according to me, likes bones.
b *[
Minden kutya] szerintem [
szereti a csontot]
every dog according-to-me likes the bone
According to me, all dogs like bones.
What is wrong with (b) is that a quantied phrase does not occur in
topic position in Hungarian. Only a non-quantied, specic phrase is
licensed there. The position before the adverbial is reserved for topics.
Therefore, the presence of the adverbial disambiguates the f-structure of
the sentence.
Thetic sentences, as argued in Chapter , section ., are predicated
of an implicit stage topic and their word order does not exhibit reorder-
ing for the sake of topic or focus marking. In categorical judgements,
however, an argument plays the role of topic. This topic will always be
fronted in a topic-prominent language. That is why the word order of
thetic and categorical judgements will always dier in topic-prominent
languages. By this criterion, Danish is not topic prominent, since top-
icalization is not obligatory and therefore the word order of categorical
and thetic sentences does not necessarily dier.
. Kiss : illustrates word order variation in Hungarian by show-
ing that the sentences in () will appear in a dierent order if one of the
elements is specic and can function as a topic:
() a [
a szobba ] [
bejtt egy kutya]
the room-into came a dog
Into the room came a dog.
b [
A csontot] szerintem [
minden kutya szereti]
the bone-ACC according-to-me every dog likes
In my opinion, bones are liked by every dog.
(a) is predicated of an overt stage topic, (b) is predicated of the
specic object. The preadverbial position ensures that the initial elem-
ent is indeed interpreted as a topic.
The word order of free-word-order languages such as Hungarian is
therefore free in terms of the order of the arguments and the verb, but
it is xed with respect to focus structure. That is why . Kisss term
discourse-congurational languages is so apt. Discourse-congura-
tional languages can be divided into those which are topic-prominent
languages (they are discourse congurational with respect to topics),
and those which mark foci by word order (they are discourse congu-
rational with respect to foci). Hungarian illustrates a language which
is discourse congurational with respect to both functions. Catalan,
described in Chapter , is discourse congurational only with respect
to topics. Catalan, and to some extent Danish, are languages which
mark topics by word order, but which mark foci by intonation. An
interesting question is whether there are languages which mark only
foci by word order, but mark topics in some other fashion or do not
mark them at all. Celtic languages such as Welsh, Irish, and Scottish
Gaelic, all VSO languages, may t in this category. According to . Kiss
, these languages are neither topic prominent nor subject promi-
nent. Yet Welsh allows argument preposing of either a topic or a focus.
Further investigation is needed to investigate whether fronting a focus
in Welsh is limited to contrast. If so, this type of fronting may fall under
the general category of topicalization which includes both plain topics
and contrastive ones.
Another question we have touched on is the obligatoriness or option-
ality of topic and focus marking. Intonational focus marking, where
it exists, is strictly required. Topic marking, by movement to the left
periphery, is more often optional. This may be because topics are also
identied by context. Languages may dier in this respect. In English
topicalization is rarely employed. In Danish, it is common, but not
obligatory. In Hungarian, topicalization is strongly preferred. The rea-
son for the stronger eect in Hungarian may be the fact that the foci
are dislocated to the preverbal position and topicalization is needed in
order to get the topics out of the way of the foci.
See Chapter , sections . and ., for the view that contrastive elements can be analyzed
as topics as well as foci.
3.2 The left periphery
The idea that information structural categories such as focus project
syntactic structure originates with linguists working on Hungarian syntax,
e.g., Horvath ; ; . Kiss . The rst to introduce FocP was Brdy
. As noted in the previous section, Hungarian has a dedicated focus
position preceding the verb. Movement to this position has been argued
to be syntactic A-bar movement in view of the fact that this movement
passes diagnostics of A-bar movement such as island sensitivity. There are
two properties of the preverbal focus position in Hungarian that will be
discussed here. The rst is the interpretation associated with the preverbal
position. The other is the complementary distribution of verbal particles
and foci in the preverbal position: When the verbal particle is preverbal
(rather than postverbal) the sentence does not have a narrow focus.
The preverbal focus position has traditionally been associated with
exhaustive identication or contrast rather than new information.

According to Horvath : , for example, the preverbal focus exhaus-
tively identies the proper subset of a contextually relevant set of entities
as the one for which the predicate holds. In other words, the Hungarian
preverbal focus is a restrictive focus. Horvath illustrates this as follows:
() Q: Kit hvtak meg?
Who did they invite?
a [JNOST] hvtk meg t
John-acc invited-pl Perf
They invited JOHN (and nobody else).
b Meghvtk *(pldul/tbbek kztt) JNOST.
Perf-invited-pl for-example/among others John-acc
They invited JOHN, for example/among others.
(b) provides evidence that a focus may remain in situ, yet unless it is
preposed it cannot receive the identicational reading. Horvath conclude
that Hungarian has grammaticalized the notion of exhaustive identica-
From the previous section, we gathered that topics are preposed in
Hungarian. Here we see that identicational/restrictive foci occur in pre-
verbal position. Vastly simplifying the organization of the left periphery in
Hungarian, we derive the following linear order of elements:
See, for example, . Kiss ; Kenesei .
Horvath in this paper therefore dispenses with the analysis of Hungarian focus prepos-
ing triggered by a focus feature and replaces it with an exhaustive identication operator
which triggers overt movement.
configurations 8,
(I:) Top Restrictive-Foc V . Informational-Foc
Szabolcsi I,,, discusses the full set of left-peripheral elements in
Hungarian. According to Szabolcsi, the preverbal position may host
either FocP or PredOpP. The latter contains counting operators like few
boys or less than six boys. According to Surnyi :oo: ,,, the elements
which were before relegated to PredOp (counting quantiers) because
they are not identicational, could be considered to be foci and therefore
in FocP, if FocP is taken to house not only contrastive foci, but other foci
as well. This would explain the complementary distribution of PredOp
and FocP, a problem for approaches which generate designated positions
for each element type in the left periphery. Yet postverbal informational
foci would still be unaccounted for.
Sentences in which no topic or restrictive focus occurs can be verb
initial as illustrated in the sentences in (8) above. Verbal particles play
an interesting role in such cases. This can also be seen in the examples
in (II). Note that the verbal particle meg follows the verb in (IIa), but
precedes it in (IIb). This distinction in the order of the verbal particle
correlates with the occurrence of a preverbal focus. In (IIa) the object is
in preverbal focus position, and the particle is postverbal. In (IIb), how-
ever, no element is in focus position and the particle is preverbal. The
complementary distribution of the two elements has led researchers to
the assumption that the focus and the preverbal particle ll the same
slot. Yet neither the verb nor the particle is necessarily interpreted as the
focus when the particle is preverbal. An account of the Hungarian focus
position, which employs a focus feature as the trigger for movement, will
not be able to capture the distribution of particles without further ado.
3.2.1 Cartography
In the previous sections, we saw that, at least in some languages, the left
periphery is the position in which we nd topics. There is a long tra-
dition in Germanic linguistics that sentences are organized linearly
according to function (the topological elds theory starting in the early
nineteenth century).
One reason why this theory may have developed
among Germanic linguists is the fact that the verb in main clauses occurs
Szendri (:ooI) argues that focus movement in Hungarian is stress driven. For a critique
of Szendri see Horvath :oo,.
For a recent account within this framework see Kathol :ooo. The approach started in
Germany and was applied by Diderichsen I,oo; I,o8 (rst published in I,o) to Danish. For an
implementation of topological eld theory within LFG see Ahrenberg I,8, cited in Kathol.
in second position in the linear order of the sentence, the phenom-
enon known as V-. The element which occurs in the rst position (the
Vorfeld or pre-eld) includes a variety of elements, some of which can
be identied as topics, others of which do not play an obvious informa-
tion-structural role. What distinguishes the topological view from the
accepted syntactic view is that the former is linear and the latter is hierar-
chical. We will return to the role played in Germanic by elements in this
position in section below. Many linguists have also noted that given
information precedes new information as already mentioned in the
previous chapter. As pointed out in Prince a, dierent construc-
tions conspire to achieve this result. Movement motivated by discourse
function has been discussed within the generative framework notably
by Jackendo , Chomsky , Culicover and Rochemont , and
Rochemont .
For syntacticians the question arises as to how such
ordering is to be accounted for. If topics, for example, are fronted, how is
this movement motivated and are topics on the left periphery adjoined
or are they to be found in dedicated specier positions? In his ground-
breaking article Rizzi argues for the latter position and shows that
the various positions on the left periphery in Italian can be mapped
for topic-focus functions as shown in () (* indicates that the node is
Chomsky employs a rule of predication to limit the choice of left-dislocated NPs and
guarantee their interpretation as being what the sentence is about.
() Force P


configurations 8,
According to Rizzi, the following four kinds of elements typically
occur in the left periphery: interrogative and relative pronouns, topics,
and focalized elements. Rizzis system is embedded in the minimal-
ist approach in which movement to a specier position is triggered
by the satisfaction of feature requirements of the head, in the cases at
hand, features with an interpretative import. A constituent with topic
or focus features must therefore end up in the Spec position of the
relevant head.
The tree structure in (I,) therefore predicts that topics
can occur both preceding and following the focus in Italian. However,
there can only be one focus. (ForceP and FinP are ignored here.) In
fact, no order is predicted between foci and topics since both sets of
topic projections are optional. Foci can therefore either precede or fol-
low topics, and can also occur in their midst. According to Rizzi I,,,:
:8o, the specier of a TopP is the topic and its complement is the com-
ment as shown in (I).
See also Chomsky :ooI for the view that optional operations can apply only if they have
an effect on outcome, i.e., they must be driven by the interpretative needs of the constituent
in question.
TopP (I)
XP = topic
YP = comment
(I) thus represents the structural mapping of the topiccomment divi-
sion (see Chapter :, section ,). Rizzis structure of the left periphery also
maps the division into focuspresupposition. The specier of the FocP
is the focus, and the complement is the presupposition (Rizzi I,,,: :8,):
FocP ZP = Focus
WP = Presupposition
ZP Foc
Rizzi therefore maps both binary divisions, and mapping makes strong
predictions with respect to both word order and interpretation.
important prediction Rizzi makes is that FocP cannot be recursive,
since if it were, it would give rise to a structure in which a Focus is
embedded within a presupposition (the complement of a higher FocP),
causing an interpretative clash. Such an interpretative clash does not
arise when topic phrases are embedded within each other, since the
comment does not exclude a topic within it.
An examination of the data Rizzi brings to bear on these predictions
raises problems which do not lend themselves to obvious solutions. In
order to see this we will examine Rizzis data with an eye to test the correla-
tion of topic-focus function and cartographical position. Rizzi cites the
following two types of fronting:
() Il tuo libro, lo ho letto
the your book, it I have read
() IL TUO LIBRO ho letto (, non il suo)
the your book I read not the his
According to Rizzi, () is a case of Clitic Left Dislocation (CLLD), with
the clitic lo coreferential to the topic on the left. The constituent to the
left in () is a contrastive focus and cannot be interpreted as conveying a
non-contrastive informational focus. Italian also allows an in situ lower
stressed focus which may or may not be contrastive as shown in ():
() Ho letto IL TUO LIBRO (, non il suo)
I read the your book not the his
In order to explain weak crossover, Rizzi assumes, following Chomsky
, that this focus moves to Spec,FocP in LF.
The structure Rizzi
proposes for Italian is not all that dierent from the Hungarian left
periphery: In both a topic occurs sentence initially, and in both a
A nice outcome of Rizzis : system is that it allows for languages with focus and
topic particles. In these languages the top and foc heads are lled. However, if top and foc con-
stituents are pronounced in situ in these languages, LF movement will not rescue the deriva-
tion in which top and/or foc particles are attached to topics and foci pronounced in situ.
Chomskys classical examples are:
(i) His
mother SHOT John
(ii) ??His
mother shot JOHN
In (ii) focusing generates a weak crossover effect. This would follow if the focus movement
applied to the object in (ii).
contrastive/restrictive focus follows the topic, whereas an informational
focus remains in situ. The two languages dier in that the second topic
position in Italian is not available in Hungarian and that the Italian focus
position is not preverbal.
Rizzi, following Cinque outlines the following four distinctions
between topic and focus constructions:
. A topic can involve a resumptive clitic within the comment. In Italian,
the clitic is obligatory when the direct object is topicalized. (The left-
peripheral element in () can only be interpreted as a topic.)
. A topic does not give rise to weak crossover effects. Rizzi, following
Lasnik and Stowell , employs WCO as a diagnostic for A relations
involving quantication. A binding, under this view, comes in two
versions: either it involves a quantier which binds a variable (as in
(a) ) or it is non-quanticational and binds a null epithet or a null
constant (as in (b) ). Only the former is sensitive to WCO:
() a ?* Who does his mother really like t (=vbl)?
b John, who his mother really likes t (=nc)
Rizzi concludes that Focus involves quanticational A binding while
Topic does not. This is his explanation for the fact that topics do not
produce weak crossover eects. It also aords an explanation for the
rst property of topics and foci: foci do not allow resumptive clitics
since a clitic does not qualify as a variable and if foci are quantica-
tional, they must bind a variable. It also follows that topics require a
resumptive clitic since without the clitic, the object trace would not be
licensed since it cannot be a variable (the non-quanticational topic
cannot identify it), nor can it be any other type of empty category. It
follows that topicalization must give rise to a resumptive clitic, yet
Rizzi says this is only a requirement for object topics and posits a null
constant in the cases in which the resumptive clitic can be left out. He
does not assume any dierences between cases of topicalization with
and without overt resumptive pronouns and, it seems, predicts that
WCO eects should surface in both cases.
Both relative clauses and topicalization in Hebrew exhibit WCO
eects only when the optional resumptive pronoun is absent as pre-
dicted by Rizzi:
Rizzi lists ve properties. Only the rst four are of relevance here.
() a ?? Hayeled e-imo mefaneket kol hazman
The-boy that-mother-his spoils all the-time
meod mefutax.
very developed
The boy that his mother spoils all the time is very developed.
b Hayeled e-imo mefaneket oto kol hazman
The-boy that-mother-his spoils him all the-time
meod mefutax.
very developed
The boy that his mother spoils all the time is very developed.
c Hayeled (e-)oto imo mefaneket kol hazman
The-boy that-him mother-his spoils all the-time
meod mefutax.
very developed
The boy that his mother spoils all the time is very developed.
(a), without the optional resumptive pronoun allowed in Hebrew relative
clauses, exhibits a weak crossover eect. (b) and (c) both have a resump-
tive object pronoun; in (b) it is in situ and in (c) is fronted and incorpo-
rated in the complementizer (used in written style). Both (b) and (c) are
much better than (a), showing the improvement rendered by the resump-
tive pronoun. Hebrew thus exhibits a correlation between the lack of WCO
eects and the presence of a resumptive pronoun. () illustrates the same
eect in Hebrew topicalization:
() a ?? Hayeled haze, imo mefaneket kol hazman
boy this mother-his spoils all the-time
This boy, his mother spoils all the time.
b Hayeled haze, imo mefaneket oto kol hazman.
boy this mother-his spoils him all the-time
This boy, his mother spoils him all the time.
If Rizzis null constant is identical to an overt resumptive pronoun, such
correlations cannot be explained. The Hebrew data also demonstrate that
the WCO eect is sensitive to the absence of resumptive pronouns both
in relative clauses and in cases of topicalization. It is hard to imagine how
Rizzis account of topicalization could be extended to relative clauses, and
the correlation between resumption and WCO may be orthogonal to the
analysis of topicalization as such.
But see Grosu ; Kuno for a possible relation between topicalization and
Rizzis prediction also poses the following problem: the correlation
between topicalization and resumptive clitics or pronouns does not hold
across and even within languages. Rizzi distinguishes languages in which
topicalization gives rise to a resumptive clitic (Italian) and those that do not
(English) by claiming that the former lack a null anaphoric operator. It is this
operator which replaces the clitic in English. He also explains optionality of
resum ptive pronouns in Italian by arguing that PPs in fact do not exhibit
optional resum ptives in that each case is a result of a different structure. In
Hebrew resumptives are optional with direct objects and obligatory with PPs.
Rizzi might surmise that Hebrew only has one type of PP, the one that requires
resumptives, whereas Italian also has the type that does not allow resumptives.
Cross-linguistic evidence for these two types of PPs is however lacking.
. Bare quanticational elements (no one, all, etc.) cannot be topics in
CLLD constructions. They do allow focalization as shown by the examples
in ():
() a *Tutto, lo ho fatto.
everything it I did
b Tutto ho fatto.
Everything I did
According to Rizzi, these data are more complex than they appear.
(a) is much improved if the quantied expression includes a lexical
() Tutti i tuoi libri, li ho rimessi a posto.
all your books them I-have put-back in place
Rizzis account involves QR movement in both cases. In the restricted
case () the quantier binds the variable within Spec, Top; in the non-
restricted case (), the variable is not in an A position and does not
qualify as a variable violating Full Interpretation.
. The number of peripheral topics is unlimited, but there can only be
one fronted focus. This focus can be both preceded and followed by topics
as shown in ():
() A Gianni, QUESTO, domani, gli dovrete dire
To Gianni, THIS, tomorrow, to-him you-should tell
Rizzi lists the following possible permutations for this example. Both
topics, in either order, can precede or follow the focus and the two top-
ics can also occur in both orders when the focus intervenes. Rizzi rejects
the idea that topics should be adjoined because they are recursive and
, configurations
argues for an X projection of topics on a par with foci. Instead, he claims
that the dierent properties of topics and foci follow from their dierent
interpretative properties, as noted above. This is an interesting explana-
tion, but note that it precludes the existence of multiple (fronted) foci
universally, most likely an unwelcome result. Rizzi I,,,: ,:, claims in
footnote I, that this result holds in Italian:
(:,) a A GIANNI ho ditto t che dovremmo leggere il tuo libro.
TO GIANNI I said that we should read your book.
b Ho ditto a Gianni che IL TUO LIBRO dovremmo leggere.
I said to Gianni that YOUR BOOK we should read.
c *A GIANNI ho ditto che IL TUO LIBRO dovremmo leggere
TO GIANNI I said that YOUR BOOK we should read.
The data in (:,) show that either main clause or subordinate focus can
be fronted within its clause, but that it is impossible to front both. These
data cannot be accounted for by a syntactic constraint which blocks
fronting in both main and subordinate clauses since this is possible, as
shown in (:o) in which the same two constituents have been fronted
within their clauses, but here they are both interpreted as topics:
(:o) A Gianni, gli ho ditto che il tuo libro, lo dovremmo leggere.
To Gianni, I said to him that your book, we should read it.
In order to be able to draw clear conclusions with respect to this type
of data, it would have been useful if Rizzi had presented similar cases
in which one of the fronted constituents was focused and the other was
not. His prediction would be that both these cases should be possible
since the only excluded case is a focus embedded within the comple-
ment of another focus.
Another problem with Rizzis approach arises in a system in which
contrast is marked as both topic and focus, as suggested in Chapter :,
section ,.,: contrastive elements are fronted in many languages. Rizzis
system cannot handle elements which are marked both topic and focus,
because topics and foci necessarily land in separate projections. This
is argued very convin cingly by Ritter and Rosen :oo, by showing that
English not- topicalization cannot receive an account within Rizzis carto-
graphic system since not-topics have both topical and focal content:
(:,) a I think Ill smoke a cigarette.
b [Not [in Bills car] ]
you wont (smoke a cigarette t
According to Ritter and Rosen, the fronted XP (excluding the negative
element) is topical old information, and the negative element itself is
focused, triggering stress on the whole XP. As topics, the fronted XPs
dont induce WCO eects and cannot consist of bare quantiers; as foci
they dont license resumptive pronouns. Ritter and Rosen conclude
that a system which denes separate functional projections for topics
and foci cannot accommodate not-topics which have the properties of
both constituent types. This argument against a cartographic account of
topic and focus placement is a serious one. No syntactic account of topic
and focus placement which assumes functional projections for topic
and focus can accommodate elements which are marked both topic and
focus. We either have to give up the account of restrictive and contras-
tive elements as having a complex f-structure, or else we have to give up
the syntactic account.
3.2.2 An even ner structure of the left periphery
Beninc and Poletto show that taking Rizzis cartographic system
to its logical conclusion, in which the functional projections are more
accurately associated with pragmatic functions, requires a larger num-
ber of FPs than envisioned by Rizzi. They replace Rizzis recursive TopP
projections with a series of such carefully dened FPs, both within the
topic and the focus elds.
First they show that in Italian the lower topic position is in fact an
extension of the focus eld ():
() a Un libro di poesie, A GIANNI, lo regalerete
A book of poems TO GIANNI, it you will give
You will give a book of poems to Gianni.
b *A GIANNI, un libro di poesie, lo regalerete
TO GIANNI a book of poems it you will give
In (a) the order is topic followed by focus. The reverse order (b) is
strongly ungrammatical. They argue that Rizzis evidence for the occur-
rence of topics below foci is restricted to temporal adverbs (see ()
above). Temporal adverbs, according to Beninc and Poletto, are distinct
from DPs and PPs in that they can occur at the IP edge. A test is pro-
posed for the position of temporal adverbs: Subject clitics are apparently
optional. If it is assumed that the presence of a subject clitic indicates
left dislocation of the subject and its absence indicates that the subject is
in situ then adverbs can intervene between subjects and inected verbs
both in the presence of a subject clitic and in its absence. In the latter
case, the adverb must be in a position lower than the usual subject posi-
tion. In the former case, it could be in a topic position. Temporal adverbs
,o configurations
therefore do not provide evidence for the presence of a lower topic position
since they are structurally ambiguous between a topic and post-subject
Beninc and Poletto made a point similar to the one made above con-
cerning contrast. They argue that stress does not necessarily indicate a
focus, since preposed contrast can allow a resumptive object clitic, indi-
cating a left-dislocated topic (their (8):,o):
(:,) A: Mi ha ditto che il tappeto, lo compra lanno prossimo.
He has told me that the carpet he will buy it next year.
B: No, ti sbagli, IL DIVANO lo compra lanno prossimo.
No, you are wrong, THE SOFA he will buy it next year.
Beninc and Poletto also show that a contrastive topic such as the one in
(:,) is not subject to WCO eects, indicating that stress is not necessar-
ily an indication of focus status (their (Io):,,):
(,o) A: Mario
, suo padre
non lo vede mai.
Mario, his father not him sees never
His father never sees Mario.
, suo
padre non lo vede mai.
No, Gianni his father not him sees never
No, his father never sees Gianni.
Foci do exhibit WCO and can be distinguished from topics, not by their
stress, but by their not permitting resumptive clitics. They conclude
that in all examples in which the sequence is seemingly [Topic Focus
Topic], it is in fact [Topic FocusI Focus:] with the two foci each having
a particular function. Their ndings thus contradict Rizzis in two ways.
First, more than one focus is allowed, and second, all topics precede the
foci. The following example is an illustration of this distribution (their
(,I) A GIORGIO, questo libro, devi dare.
TO GIORGIO this book you must give
You must give this book to Giorgio.
Beninc and Poletto analyze questo libro as an informational focus which
is licensed by the presence of a contrastive focus (a Giorgio) preceding it.
Southern Italian does not require a contrastive focus as a trigger for preposing an
informational focus. It is difcult to imagine a cartographic account that can predict such
dialectical variation.
According to Beninc and Poletto the topic eld is also complex and
consists of two dierent topic types: the hanging topic (HT) and left dis-
location (LD) illustrated in (a) and (b) respectively (their ():):
() a Mario, non ne parla pi nessuno.
Mario not of-him talks anymore nobody
Mario, nobody talks of him anymore.
b Di Mario, non (ne) parla pi nessuno.
Of Mario not (of-him) talks anymore nobody
Of Mario, nobody talks of him anymore.
These two types of topicalization dier in a number of ways:
. HTs can only be DPs, but LDs can also be PPs (as in (b)).
. There can only be one HT, but there can be several LDs.
. HTs always require resumptive pronouns (or a copy) which
agree with them in number and gender, but not in case. LDs only
require a resumptive pronoun (which agrees in number, gender,
and case) when they correspond to direct or partitive objects.
. The HT copy can be a topic pronoun or an epithet. The copy of
the LD must be a clitic.
. If both types of topicalization occur, HT must precede LD.
. HTs are more restricted in embedded clauses and do not occur in
relative clauses.
Beninc and Poletto propose to account for these dierences by introducing
a separate FP for each with the HT above the LD projection.
Scene-setting adverbs (stage topics) occur higher than LDs but below
HT, but it is not clear whether they require an independent FP since they
cannot be distinguished from LDs. (The authors refer to the HT and the
scene-setting adverbs together as the frame.)
Beninc and Poletto also propose to distinguish another type of
LD, namely contrast, within a given set (list interpretation, LI). ()
(their ():) should be understood in the following context: a farm
producing a set of goods that are known to the people involved in the
() La frutta la regaliamo, la verdure la vendiamo.
The fruit it we-give-for-free, the vegetables it we-sell
We give the fruit for free, while we sell the vegetables.
This type of LD, they propose, is located below the LD projection. (They
refer to the LD and the LI projections together as the theme.)
Instead of recursion, a series of distinct FPs, each specialized for a par-
ticular interpretation, are proposed. These FPs belong to two main elds, the
topic eld and the focus eld, with the former preceding the latter. They con-
sider this order, proceeding from information shared by speaker and hearer
to new information, to be likely to be universal. This guiding principle also
governs the order within each eld. Thus, the contrastive focus which relates
to a contextual context set is more topical than the informational focus
which has no such contextual relation. Here again old precedes new.
Although this article presents a more articulated, and therefore more
precise, analysis of the Italian left periphery, in which care is taken to
contextually analyze each peripheral position, it also illuminates the
problems with the cartographic program quite clearly.
Rizzis view of the left periphery clearly needs to be examined cross-
linguistically. An example of an article which attempts to analyze the
Korean left periphery within this framework is Gill and Tsoulas .
These authors demonstrate that it is not tenable to accept that the
positions hosting topics and foci in Korean are dierent and dedicated
to the distinct discourse functions that these elements fulll in specic
sentences. They thus reject the cartographic approach and propose
instead that independent operations of narrow syntax and phonology
conspire in order to produce the observed results (). (a) and
(b) illustrate short (clause-internal) and long (across one or more CP
boundaries) topicalization in Korean (their ():):
() a I chayk-un [Swunja-ka e sassta]
this book-TOP Swunja-NOM bought
As for this book, Swunja bought (it)
b I chayk-un [Younghee-ka [Swunja-ka e sassta]-ko
This book-TOP Younghee-NOM Swunja-NOM bought-COMP
As for this book, Younghee said that Swunja bought (it)
The authors demonstrate that the TOP morpheme-(n)un is pronounced
with a rising tone followed by an intonational break when the topic is
fronted. The topicalized element is interpreted either as a plain or as a
contrastive topic.
When topicalization does not apply, the subject itself
n(un)-marked elements can also be interpreted as contrastive foci if left in situ.
is interpreted as a topic. This positions Korean as a language which is
halfway between topic and subject prominent.
Gill and Tsoulas refer to fronting without n(un) marking as scram-
bling. (a) and (b) illustrate short and long scrambling respectively.
() a I chayk-ul
Younghee-ka e
this book-ACC Younghee-NOM read
Younghee read this book.
b I chayk-ul
[Chelswu-ka [Younghee-ka e
this book-ACC Chelswu-NOM Younghee-NOM read-COMP
Chelswu said that Younghee read this book.
Scrambling and topicalization interact in an interesting way: it is pos-
sible to scramble an object over a n(un)-marked subject as illustrated in
() I chayk-ul
[Chelswu-nun e
This book-ACC Chelswu-CF bought
It is CHELSWU who bought this book.
Scrambling across such a subject has the eect of forcing a contras-
tive focus reading of the subject (marked CF). It is this property of the
subject in () which challenges the cartographic model. If we assume
a dedicated position for the scrambled element, then the head of this
projection is not only responsible for rendering an appropriate interpre-
tation for the scrambled element itself, it must also determine the inter-
pretation of the subject which is situated in a position below it. As Gill
and Tsoulas point out, it would be surprising if such a functional head
were to exist.
Another point which is problematic for a cartographic
approach, and Rizzis proposal in particular, is the restriction in
Korean that only one topic may be fronted. It is not clear what type of
parameter could distinguish between Italian, which, according to Rizzi,
allows the fronting of more than one topic, and Korean, which does not.
Gill and Tsoulas propose that I is endowed with a [+stress] feature which
forces the rising tone on the element moved into Spec,IP. They justify
Issues such as these lend themselves well to an optimality theoretical approach. Choi
shows that there is only a minor difference in the ranking of the constraints between
Korean and German. We return to his analysis in the context of our discussion of German in
section below.
their proposal by assuming, along the lines of Holmberg , that the
EPP is phonologically motivated. This approach divorces syntactic posi-
tion from topic/focus interpretation. Instead, they argue that interpreta-
tion is at least partially derived as a consequence of general constraints
on accentual phrasing. In (), for example, the contrastive interpreta-
tion results from a contrast between the expected pitch accent in non-
sentence-initial position and the pitch accent assigned there relative to
the rst accentual phrase().
This analysis of Korean may or may not
be on the right path. What this paper makes clear, however, is that the
cartographic approach does not provide insight into the analysis of a
language such as Korean.
A potentially good test case for Rizzis approach is Hungarian, in
which topics and foci are strictly ordered on the left periphery. At the
beginning of this section several properties of the preverbal focus posi-
tion in Hungarian were outlined. Problematic for Rizzis approach is
the particular interpretation which preverbal foci receive in Hungarian
as well as the distribution of counting quantiers and verbal particles.
There is a wealth of work on these issues in Hungarian; however, there
is, as yet, no accepted view of their analysis. Even those that adopt Rizzi-
type functional projections do not agree on which functional categor-
ies to employ. For some recent discussion of these issues with respect
to Hungarian, see . Kiss ; to appear; Horvath ; Surnyi ;
Szendri ; , among others.
As pointed out by Andrew Carnie in a Linguistlist review of Rizzi
b ( January ) the program allows for an examination
of microvariation among closely related dialects, but it has yet to
be examined within a broader cross-linguistic perspective. This is
particularly critical, I believe, in view of the large number of functional
categories that must be introduced in the universal inventory (e.g., HT,
LD, LI, contrastive focus, informational focus, scene- setting adverbial,
etc.), and the availability of evidence to the language learner of the
occurrence of each such category in the language being acquired. Maybe
even more serious is the optionality problem discussed in the previous
chapter in section : Since movement to the left periphery of these
categories is often optional, the Minimalist Program must often resort
to rather baroque analyses in order to account for which ones of these
are optional and which are obligatory and what, if any, pragmatic dierence
Their approach thus belongs to those theories of information structure which derive
function from stress rather than vice versa (see Chapter , section .).
there is between fronting one of these categories and leaving it in situ. In
sum, lacking explanatory power, the cartographical model is reduced to
descriptive adequacy.
3.3 The importance of information structure
In the previous section, it was shown that in order to account for the
variation in the order of fronted elements, a large number of functional
categories were introduced, each with its own information-structural
characterization. If we constrain information structure theory and limit
the basic notions involved to topic and focus, then the number of dierent
functional projections is automatically constrained as well. Beninc and
Poletto, as just mentioned, were forced by their data to conclude that cer-
tain stressed elements are contrastive topics rather than foci as assumed
by Rizzi. If we adopt the analysis of contrast in Erteschik-Shir (see
Chapter , section .), then all cases of contrast can be viewed as either
topics or foci. Topicalizing languages may therefore dier as to whether
contrastive elements are viewed as topics, and consequently fronted, or
whether they are viewed as foci, and therefore not allowed to be fronted.

Under this more restricted view, information structure-motivated move-
ment to the left periphery can be motivated either by topic assignment
(possibly including contrast) or by focus assignment (possibly including
contrast). There may also be languages which restrict movement to the
left periphery to contrastive elements alone.
Frey to appear also couches his account of topicalization in German
in cartographic terms, but argues that not all cases of movement to the
left periphery have information-structural import. (a) and (b) below
illustrate topicalization in German (his () ):
() a (Ich erzhle dir etwas ber Hans.)
I tell you something about Hans
Den Hans wird eine polnische Grn heiraten
theACC Hans will a polish countess marry
A polish countess will marry Hans.
b Dem Hans hat Maria nicht geholfen, wohl aber dem Otto
theDAT Hans has Mary not helped PRT but theDAT Otto
Mary has not helped Hans, but rather Otto.
The latter case is rare and may have to do with the unacceptability of sentence-initial
According to Frey, the initial element in (a) is a topic, in the given
context. The fronted element in (b) is contrastive. Fronted contras-
tive elements can answer wh-questions, and are therefore assumed to
be foci, yet the type of wh-question which they answer is restricted
to one which identies a discourse-specied set ( (b), for example,
answers a question such as Which of her friends did Mary help?).
Since the answers to such questions range over this set, this contras-
tive element can equally well be analyzed as a (contrastive) topic. Data
such as these could therefore equally well lead to the generalization
that what is topicalized is a topic. According to Frey, however, contras-
tive elements must be distinguished from both topics and foci. This,
he argues, is necessary in order to explain the fact that movement of
a constituent from an embedded clause (long movement) to the left
periphery necessarily results in contrast, whereas short movement of
the same element does not (Frey to appear: ; his () ):
() Ich erzhle dir was ber Max.
a Den Max sollte unsere Gruppe untersttzen
Max should our group support
b Den MAX
meint Maria, dass unsere Gruppe t
Max thinks Maria that our group
untersttzen sollte
support should
Frey notes that whereas Max in (a) need not receive a contrastive reading
nor must it be stressed, Max in (b) must be stressed and interpreted
contrastively. Freys account of this distinction is that Contrast (but not topic
and focus) projects a functional projection between CP and FinP, which
hosts A-moved contrastive elements and checks their contrast feature.
Spec, FinP hosts constituents moved from the local middle eld, i.e., the
set of constituents between the nite verb or the complementizer and the
verbal elements at the end. CP hosts certain discourse-type adverbials. These
are merged directly in Spec,CP or else the head C hosts a complementizer.
According to Frey, each of these three phrases (CP, ContrastP, FinP) is
associated with an EPP feature, each with its particular characteristics, yet
only one EPP feature occurs in a particular sentence. The verb then moves
to the head which carries the active EPP feature, positioning the verb in
second position as required. Although Freys set of projections correctly
distinguishes the properties of contrastive elements, the theoretical
consequences are problematic. The constraint that only one EPP feature
is realized when three dierent EPP features are available is ad hoc and so
configurations Io,
is the association of verb movement and EPP. Another problem is policing
the optional movement of contrastive elements. Freys analysis thus suers
from the same type of problems as do other cartographic proposals.
Let us therefore have another go at the idea that fronted elements
in German are necessarily topics and that topics may be contrastive.
This would save at least one dedicated functional projection on the left
periphery and would unify Freys contrast projection with FinP.
Here Svenonius :oo comes to our aid. He argues that the left-periph-
eral element in Germanic as a whole is a switch topic or a shifted topic,
and not a continued topic.
Such elements, according to Svenonius
:oo: :I,, redirect the emphasis of the discourse. Svenonius oers the
following illustrations (his (,o) and (,I); :oo: :Io):
(,,) Varmt og kaldt vann var det jo ikke. Norwegian
Warm and cold water was it after.all not
(o) Swedish
a Ja, for tre veckor sen ck professorn besk av
Yes for three weeks ago got the.professor visit of
en herre som talade med stark brytning
a gentleman that spoke with strong accent
Yes, three weeks ago a man with a strong accent came to see
the professor.
b De talade lnge.
they spoke long
They had a long talk.
c och sedan packade professorn och gav sig i vg
and then packed the.professor and gave rfx in way
med den frmmande herrn
with the unknown gentleman
Then the professor packed his luggage and left with the stranger.
The fronted element in (,,) is indenite and has not been mentioned
in the preceding discourse, and so could be considered to be new, yet it
connects to the previous discourse in that it would have to be uttered in
See Chapter :, section I.: for these notions. Lambrecht I,, makes a similar point.
According to Poulsen :oo,, both focal and topical elements get fronted in Danish and therefore
he nds it difcult to dene the function of fronting. Instead, he suggests that the rst position
marks linkage with the previous clause. As argued in the previous chapter, however, the function
of fronted elements in Danish can be viewed as a topicalization process if one takes into account
that topicalization applies to switch topics, including contrastive and restrictive ones which may
have been mistakenly analyzed as foci by Poulsen.
a context in which accommodations are under discussion. The fronted
element is necessarily viewed as one of the relevant features of accommo-
dation, hence a topic according to the denition of topic used here. (See
also Chapter , section ..) In () the obvious topic is the professor,
yet this element does not appear in initial position in the discourse. This,
according to Svenonius, is because the professor is a continued topic.
The time adverbial in () is a stage topic (bound by the current here
and now) which shifts attention to this point in time. In (b), the plural
pronoun groups the professor and the gentleman, both of whom are now
discoursally available as topics. In (c) again the choice is not to use any
of the available topics but rather to again employ a stage topic. This text is
not unusual, but it is easy to nd fronted continued topics in Germanic as
well, as shown in () repeated from Chapter , section ..
() A: Hans kan jeg godt lide.
Hans can I good like
I like Hans a lot.
B: (i) Ham kan jeg ogs lide.
him can I also like
I like him too.
(ii) Jeg kan ogs lide ham.
I can also like him
I like him too.
In the context of (A), the fronted pronoun in Bs response in (i) is a
continued topic. If stressed, it becomes contrastive and other candidates
under discussion will be excluded. B could also respond with (ii) which
is slightly less natural. The arguments in both responses are pronominal
continued topics and in canonical topic positions, the topicalized position
and the subject position, respectively. Let us therefore examine a case in
which one argument is a continued topic and the other is a switch topic.
This is illustrated in () in which the object is a switch topic and the
subject is a continued topic and in () where it is the other way around.
() A: Hvor har hun lagt tingene?
Where has she put the.things
B: a Bogen har hun lagt p bordet. has she put on the.table
She put the book on the table.
Svenonius admits that continued topics occur, and that they are usually subjects.
b #Hun har lagt bogen p bordet.
She has put on the table
() A: Hvor har brnene lagt bogen?
Where have the.children put
B: a #Bogen har Hans lagt p bordet. has Hans put on the.table
Hans put the book on the table.
b Hans har lagt bogen p bordet.
Hans has put on the.table.
As question in () establishes a set of things of which the book in Bs
answer is a member. The subject pronoun is a continued topic. The answer
in which the switch topic (the book) is topicalized (a) is much better than
the one in which the continued topic is in initial position and the switch
topic is left in situ. From these data, we can conclude that the order switch
topiccontinued topic is preferred. () renders the same result: The order
in which Hans, one of the children, hence a switch topic, is in situ in
subject position, yet the continued topic the book is topicalized, is less
good than the one in which the switch topic precedes the continued topic.
Further research is required to determine whether switch topics always
precede continued topics and if not what the circumstances are in which
the order can be reversed. What is clear, however, is that topics (in the
broad sense) occupy rst position in Germanic.
Choi , who works within LFG augmented with OT (see Chapter
, section .), argues that the canonical or unmarked word order in a
scrambling language is determined by a set of syntactic constraints
whereas another set of constraints control the mapping from informa-
tion structure to phrase structure.
The IS features employed by Choi are shown in ().
+Prom Prom
New Topic Tail
+New Contrastive Focus Completive Focus
Weskott employs the notion elaborating topic to characterize topicalized elements
in German which are members of a restrictive set or hypertheme in the preceding context. Such
topics would here receive a restrictive f-structure and would not require a special category.
Ioo configurations
The information structure constraints (Choi I,,,: I,,) are shown in (,).
(,) a new: [New] should precede [+New]
b prom: [+Prom] should precede [Prom]
The syntactic constraints may be in conict with the IS constraints.
Such conicts are resolved in OT by having a set of violable universal
constraints which are ranked so that a higher-ranked constraint is given
preference over a lower-ranked constraint. The grammar of a language
in OT is in essence the way it ranks these constraints. In German, accord-
ing to Choi I,,,: I,,, PROM is ranked above NEW. The syntactic con-
straints are:
(o) cnI: SUBJ should be structurally more prominent than (e.g.,
c-command) non-SUBJ functions.
cn:: Non-SUBJ functions align reversely with the c-structure
according to the functional hierarchy.
In German, these rules are ranked with respect to the information struc-
turing constraints as follows (Choi I,,,: II):
(,) Constraint Ranking (German)
prom >> cnI >> new >> cn:
This ranking allows for IS-motivated scrambling because IS precedence
constraints are ranked higher than the syntactic constraints. The fact
that cnI is ranked fairly high predicts that the scrambling of subjects
is less common than that of the other constituent types. According
to Choi I,,,: I:, topics, which are [New, +Prom] (see () ), will
be sentence initial; non-contrastive foci (completive foci in Chois
terminology) will be placed immediately before the verb because they
have the features [+New, Prom]. What about contrastive elements?
These are [+New, +Prom] and in view of the positive value of the Prom
feature will scramble over Prom elements. Chois system thus classies
contrast with topics in that they are both +Prom and therefore he
predicts that they both topicalize. Choi I,,,: 8o further distinguishes
shift topics from continued topics, which he asserts are not +Prom, but
are merely anaphoric (Tail in () ). Since +Prom elements topicalize,
he predicts that shift topics and contrastive topics will take precedence
over continued topics in initial position, the conclusion which was
reached immediately above. His view of shift topics as +Prom is similar
to Svenonius cited above, for whom shift topics redirect the emphasis
configurations Io,
of the discourse. Chois OT analysis is more elegant and less stipulative
than the cartographic approach. It may fail, however, when faced with
the German data discussed in section ,. below where an element with
no IS function is fronted.
Chois account of German word order carries over to Korean with
only a slight dierence in ranking (I,,,: I88):
(8) Constraint Ranking (Korean)
prom >> new >> canon
Korean diers from German in that both syntactic constraints are
superseded by the IS constraints. This explains both the similarity in
word order between the two languages and also the slight dierence
between the two: In German the subject is harder to scramble over than
other arguments. Splitting canon in two in German, but not in Korean,
accounts for this dierence.
Chois system oers an elegant account of the problematic example in
(,o), also noted by Gill and Tsoulas :oo, in which topicalization of an
object over a nun-marked subject forces a contrastive reading of the subject.
Choi I,,,: 88 argues that nun is a prominence marker in Korean in that
it marks both topics and contrastive foci. The nun-marked subject should
therefore have both interpretations, but in fact receives only the contrastive
interpretation in this case. Choi argues as follows: assume that the subject is
[New] and the object is [+New]. In this case the subject would be a topic
since it is also [+Prom]. The object would lose to the subject since both
the features of the subject [New] and [+Prom] urge it to precede other
elements and it doesnt matter whether the object is a complete focus or a
contrastive focus. If the object doesnt scramble in this case no constraint is
violated, yet if the object scrambles, canon is violated as well as new (and
also prom) if the object is contrastive. If, however, the subject is contras-
tive, i.e., it is [+New] and [+Prom], and the object is a topic and is also
[+Prom], then the object can precede the subject. This word order violates
canon, but satises new which is a higher-ranked constraint.
Ideally rankings should be motivated. If for example the dierence
in ranking between German and Korean could be linked to some other
property distinguishing the two languages, the theory would be much
3.3.1 Distinguishing topicalization from left dislocation in English
Left dislocation is characterized by there being a resumptive pronoun
where topicalization has a gap. According to Prince I,,, left dislocation
Io8 configurations
in English fullls three dierent functions: (i) simplifying discourse
processing by fronting a non-topic from a position which is usually
reserved for topics, a strategy which we will return to in section ,.,; (ii)
amnestying an island violation by leaving a resumptive pronoun in
cases of topicalization in which leaving a gap would cause an island vio-
(iii) triggering a (po)set inference (see Chapter :, section I.:
(n. ,) ). It is only in this last case that a particular discourse function is
associated with the fronted element itself. It is therefore interesting to
compare the function of this type of LD to that of plain topicalization.
Examples (Princes (,cf) ) are given in (,).
(,) a [Reply to a note whose sender says he is considering buying
Miss All-American Beauty, Touch of Class, Folklore, Pristine,
Brigadoon, Red Lion, and Sheer Elegance.] David, Forget
Miss All American Beauty, not a show rose. Touch of Class
and Folklore are. Pristine is if you can get to the show before
it opens. Brigadoon, maybe with older plants, Red Lion

[e] dug it
years ago, Sheer Eligance [sic], [e] see a few in the
shows but no queens here. To nd out the best show roses for
your area you must check on the shows and keep record of
what wins there this year, not ten years ago.
b This I dont call cooking, when you go in that refrigerator
and get some beans and drop them in a pot. And TV dinners
they go stick them
in a pot and she say she cooked. This is
not cooking.
c She had an idea for a project. Shes going to use three groups
of mice. One
, shell feed them
mouse chow, just the regular
stu they make for mice. Another
, shell feed them
And the third shell feed junk food.
d My father loves crispy rice, says Samboon, so we must have
it on the menu. And Mee Grob, too
, he loves it
just as much.
Mee Grob ($.,,) is a rice noodle
According to Prince (her (Io)), the function of these poset LDs is to
trigger an inference on the part of the hearer that the entity represented
by the initial NP stands in a salient partially-ordered set relation to
some entity or entities already evoked in the discourse model. Poset
See Erteschik-Shir I,,: for an account of resumptive pronouns in islands.
Prince notes that the left-dislocated constituents of this type have a characteristic fallrise
prosody which requires further study. (For the sources of these examples see Prince I,,,.)
configurations Io,
relations include subset and subtype relations. In (,a), for example, a
set of rosebushes has been evoked and enumerated. The reply lists the
members and asserts a property of each one. (,c) evokes a set of three
groups of mice, and the LD refers to one of these groups. In (,b) and
(d) the context does not supply a set explicitly and so this set must be
constructed by accommodation. In these two cases, the introduction of
a refrigerator and a menu in the context facilitates accommodation of
the relevant sets: things contained in refrigerators and things contained
on menus, respectively. From this characterization of posets, we can
conclude that they have much in common with restrictive sets and will
assume that they are identical. We can therefore summarize Princes
generalization as saying that the referent of a left-dislocated constituent
must be interpreted as a member of a restrictive set.
Prince uses (,c), in which two LDs are followed by a case of topicaliza-
tion, to illustrate the dierent functions of the two strategies. Following
Prince I,8Ib she denes the discourse function of topicalization in two
parts, of which the rst part is identical to the denition of (poset) LD.
The second part attributes a focus/presupposition information struc-
ture to the topicalized sentence similar to the one found in clefts (see
section ,.o below). Yet the topicalized element itself is not the focus; the
focus is to be found within the clause. This can be seen in the topicalized
(last) sentence in (,c). By the time this sentence occurs, we not only have
a contextually evoked set of groups of mice, we also know that each elem-
ent of this set is being paired with a type of food. The focus of the sen-
tence is therefore presupposed to be a food type. According to Prince, the
new information in the topicalized example is that it is junk food that the
third group will be fed. Another way of putting this is to say that topical-
ization is employed to assert a pairing between elements of two restric-
tive sets, whereas LD involves the listing of the members of one such set.
According to Princes view that topicalization is more specialized than
LD, all cases of topicalization should also be able to appear as LD con-
structions. Gregory and Michaelis :ooI show that this prediction is false
since topicalization, but not LD, is possible with anaphoric elements.

This is illustrated in (,o) (their (,,); :ooI: I,o:):
(,o) Context: A has just outlined some possible policies for the local
school board.
Gregory and Michaelis :ooI disagree with Princes view of the function of topicalization
for technical reasons which will not be dealt with here.
IIo configurations
B: Uh huh. Thats some pretty good ideas. Why dont you do
something with those? You should run for local school board
A: That Im not so sure about. Ive got a lot of things to keep me
busy. (TOP)
A: *That Im not so sure about it. Ive got a lot of things to keep
me busy. (LD)
Gregory and Michaelis rst point out that the preposed pronoun is con-
trastive and therefore expresses a poset relation. According to Prince,
both topicalization and LD should therefore be possible, yet LD is clearly
bad. Gregory and Michaeliss view of LD diers from that of Prince. They
explain the restriction against LD fronting of anaphoric expressions by
the denition of LD. According to these authors, LD is a topic promot-
ing construction. Since pronouns are by denition topics, they cannot
be promoted in this way. What they mean by topic promoting (following
Lambrecht I,, among others) is that the referent is not in the discourse
context prior to the utterance and remains in the discourse context after
its utterance. Topic promoting does not contradict the poset relation. The
discourse-available set allows the introduction of a new referent while
at the same time forcing a well-dened connection to the previous dis-
course. Gregory and Michaeliss notion of topic promotion may therefore
boil down to the same notion that Svenonius :oo refers to as a switch
topic (see immediately above), but whereas Germanic exhibits a prefer-
ence for switch topics in topic position, English LD is fully restricted to
such topics. Further research is required to test this analogy.
3.3.2 Focus preposing and Yiddish movement in English
According to Birner and Ward I,,8: 8, the following examples illustrate
focus preposing and not topicalization:
(,I) a I made a lot of sweetbreads. A couple of pounds I think I made
for her.
b I: Are there black kids in that school now?
S: Not many. I had two really good friends. Damon and Jimmy
their names were.
c I promised my fatheron Christmas Eve it wasto kill a
Frenchman at the rst opportunity I had.
For the sources of the utterances of these sentences, see Birner and Ward I,,8: 8. For
an early observation of this point see Keenan-Ochs and Schieffelin I,,o.
One of the prominent dierences between topicalization and focus prepos-
ing is the fact that the preposed constituent is accented. However, preposed
foci and topics have in common their inferential linking to an antecedent
in previous discourse. In (a), for example, the preposed amount is ren-
dered salient, according to Birner and Ward, in view of the previous men-
tion of a quantity of sweetbreads. Similarly in (b), the mention of people,
and the expectation that people have names, allows for the preposing of the
two names. Finally, in (c), the mention of an event makes salient the time
of its occurrence. Birner and Ward also dene the fronting of contrastive
elements as focus preposing (their (a); : ):
() Waitress: Did you want tea?
Customer: Coffee I ordered, I think.
This is described as follows: the tea and coee represent alternate values
in the inferred anchoring poset {beverages}, rendered salient in light of
the waitress query regarding the customers beverage order. In our terms,
in the context of the waitresss question concerning tea, the contrast set
tea/coee becomes available, allowing the contrastive interpretation of
coee in (). If focus preposing in fact is restricted to the preposing of
contrastive or restrictive foci, then focus preposing can in fact be collapsed
with topicalization, the only dierence being that a focus is contained
within the topic set. In view of the fact that foci are stressed, the dier-
ence between plain topicalization and contrastive topicalization follows.
In order to see whether this is indeed the case, the examples in () should
be examined in order to verify that they are indeed restrictive. According
to Birner and Ward, they all depend on the availability of a poset rela-
tion, i.e., a contextual restrictive set, from which the focused element is
selected. (This is the case in (a) in which a set of quantities is salient, one
of which is picked. In (b) a set of two friends is introduced, the name of
each of these friends is listed (i.e., all the elements of the set are selected),
and in (c), the promising event allows for a set of times, one of which is
selected.) Focus preposing can therefore be seen as one type of topical-
ization, not a category on its own. The requirement that the assertion be
salient follows from the contextual requirement on restrictive sets.
Yet another type of fronting, Yiddish movement, was rst analyzed
in Prince b.
(a) is from Birner and Ward : , (b) and (c)
are from Prince (also cited in Birner and Ward).
This construction is restricted to those English dialects which were inuenced by
II: configurations
(,,) a Then came bald Uncle Hymie, one st shaking violently in
the airlike Lenin he looked!
b A: Hows your son?
B: Dont ask! A sportscar he wants!
c A: Hows your son?
B: Dont ask! #A sportscar he stole!
The preposed element in Yiddish movement is stressed and it can be
brand new (,,a). The distinction between (,,b) and (c), according
to Prince, is that someones wanting something can be presupposed
through general knowledge of the world, yet someones stealing some-
thing cannot. The only restriction on Yiddish movement seems to be
that the assertion should be generally known or plausible. It need not be
salient as is the case for focus preposing in other dialects of English.
3.3.3 When the left-peripheral element does not have information-
structural impact
As mentioned in section ,.I, Prince I,,, argues for a type of LD, a simpli-
fying LD, in which fronting is not directly motivated by the discourse sta-
tus of the fronted element. An example is given in (,) (Princes (,) ).
(,) a there wont be any dead up there. Therell just be tomb-
stones setting there. Because the coal is under the graves.
An old preacher down there
, they augered under the grave
where his
wife was buried. And hes nearly blind and he
prayed and everything.
b I know what this piece of equipments raised to do. Any
, if theyre worth I,o million dollars you dont need
to think for a minute they
re not gonna know what youre
doing. They didnt get there that way.
c My sister got stabbed. She died. Two of my sisters were living
together on I8th Street. They had gone to bed, and this man, their
girlfriends husband, came in. He started fussing with my sister
and she started to scream. The landlady
, she
went up, and he
laid her out. So sister went to get a wash cloth to put on her
These all involve the fronting of discourse-new entities, yet discourse
newness is not a sucient condition for fronting, as is shown in (,,), a
variant of (,c) (Princes (,) ).
For the sources of these examples see Prince I,,,.
configurations II,
(,,) My sister got stabbed. She died. Two of my sisters were living
together on I8th Street. They had gone to bed, and this man, their
girlfriends husband, came in. He started fussing with my sister
and she started to scream. The landlady went up, and he laid her
out. #So a wash cloth
, sister went to get it
to put on
According to Prince, the dierence between (,,) and (,c) is the position
which the initial NP would occupy if the sentence were in canonical form.
In all cases, the NP would be in a position that is strongly disfavored for
NPs evoking Discourse-new entities. One such position is the subject
position (e.g., (,b) and (c) ). The other is the position of a possessive
(e.g., (,a) ). In (,,), however, the fronted element is fronted from object
position, a position which favors discourse-new elements. LD is therefore
disfavored. The purpose of LD can therefore be seen as maintaining a
canonical f-structure for the sentence, since once the discourse element
has been introduced at the left periphery, its copy within the sentence is
discourse old, the favored role in the syntactic positions in question.
3.3.4 Fronting non-topics
According to Frey to appear, not all initial elements are topics. FinP, in
Freys framework, hosts elements moved from the initial position of the
middle eld. Topics often front to this position, and when they move on
to initial position, their topic status is maintained.
Yet the initial posi-
tion of the middle eld does not necessarily host a topic, hence Frey pre-
dicts that the peripheral element in FinP need not be a topic. Such cases
are illustrated in (,o). Here, according to Frey, the fronted elements are
not referential and require contextualization:
(,o) a Fast Jeden Kollegen ndet der berhmte Linguist
nearly every
colleague thinks the famous linguist
(is) nice
The famous linguist thinks nearly every colleague is nice.
b In einem Garten hat Maria den Hund gefttert
in a garden has Mary the
dog fed
Mary fed the dog in a garden.
According to Frey, (,oa) in the context in (,,) is perfect:
See Jger I,,, and Meinunger :ooo for the view that it is topics that are scrambled to
the initial position of the middle eld. See also section below.
() Hans fhlt sich wohl an seinem neuen Arbeitsplatz.
Hans feels REFL ne at his new working place
Fast jeden Kollegen ndet der berhmte Linguist sympathisch.
Note that the preceding sentence introduces the place of work into the
context, allowing it, and any of its natural parts, to be the topic of the fol-
lowing sentence. Colleagues are clearly a natural part of a place of work
and therefore qualify as the topic of (a). This follows if elements which
are subsets of previously mentioned constituents are dened as topics, a
denition not adopted by Frey. Frey does not supply a context for (b) but
assumes that the preposed locational PP cannot be a topic. Yet in a frame-
work which allows stage topics, a PP may in fact play the role of a topic.
(b) cannot answer the question: Where did Mary feed the dog? It follows
that the PP does not play the role of focus. Moreover, the fact that () is
well formed indicates that the PP in fact must be the topic, since the other
sentence constituents are indenite and therefore do not qualify as poten-
tial topics and a sentence must have at least one topic to be interpreted.
() In einem Garten hat ein Mdchen einen Hund gefttert.
in a garden has a girl a dog fed
It follows that (b) and () are predicated of a stage topic in which a gar-
den restricts the location dened by the discoursally available current stage.
Yet arguments for topichood do not extend to (a) and similar
examples in which the fronted elements are argued by Frey to have no
information-structural impact.
() a Leider hat keiner dem alten Mann geholfen.
unfortunately has nobody the
old man helped
Unfortunately, nobody has helped the old man.
b *Leider
sagte Karl, dass t
dem alten Mann keiner
unfortunately said Karl that the old man nobody
geholfen hat
helped has
Karl said that unfortunately nobody helped the old man.
(a) exemplies a sentence adverbial which cannot be topical. (b)
shows that such elements cannot be fronted from within a subordinate
clause, an indication that A movement is not involved. Since A move-
ment must be motivated by semantic or pragmatic import, lack of move-
I exclude from discussion Freys examples in which the initial element is a subject.
Subjects are unmarked topics, but it is not necessarily the case that subjects must be
configurations II,
ment is thus predicted. Frey concludes that the mechanism which allows
for the local fronting of such elements is purely formal and is motivated
by the EPP requirement which can also be fullled by expletives. That
adverbs such as leider are not likely topics is uncontroversial, yet it is
not obvious that the various word orders listed in (oo) are equivalent
from an information-structural perspective ( (ooa) = (,,a) ):
(oo) a Leider hat keiner dem alten Mann geholfen.
unfortunately has nobody the old man helped
b *Keiner/ KEINER hat leider dem alten Mann
nobody has unfortunately the old man
c Dem alten Mann hat leider keiner geholfen.
the old man has unfortunately nobody helped
d *Dem alten Mann hat keiner leider geholfen.
the old man has nobody unfortunately helped
(oob) shows that the subject keiner can be topicalized only when it is
contrastive. This German quantier thus does not qualify as a non-
contrastive topic.
The denite object dem alten Mann, predictably, can be
topicalized whether contrasted or not as shown in (ooc). (ood) is ruled out
because keiner has scrambled to the left, and, as shown in section ., below,
scrambling targets (subordinate) topics. The only word order in which the
left-peripheral element is not a topic is therefore (ooa). A straightforward
conclusion is that fronting an element which does not function as a topic
indicates that none of the other elements in the sentence is to be inter-
preted as a topic. This would leave the sentence topicless, not an option
in a framework in which every sentence must have a topic for truth value
assignment to take place. The only other option is that sentences such as
these must be predicated of a stage topic. The sentences in (oI) show that
an initial adverb is possible in sentences in which all the arguments are
indenite and therefore cannot be interpreted as topics. This proves that
these sentences are indeed to be interpreted as having a stage topic.
(oI) a Glcklicherweise hat ein Mdchen einem alten Mann
Fortunately has a girl an old man
Fortunately a girl helped an old man.
Erteschik-Shir I,,,: I: for arguments showing that quantiers of this type cannot be
b Leider hat ein Hund einen alten Mann gebien.
Unfortunately has a dog an old man bitten
Unfortunately, a dog bit an old man.
I conclude that fronting a non-topic marks the sentence as having a stage
topic in German. It follows, that, contra Frey, fronting adverbs to the left
periphery does have information-structural impact. This impact how-
ever is not associated with the fronted element itself.
The following Danish data give the same results:
() a Desvrre kom Hans/han ikke til selskabet.
unfortunately came Hans/he not to the party
Unfortunately Hans/he didnt come to the party.
b Hans/han kom desvrre ikke til selskabet.
Hans/he came unfortunately not to the party
Hans/he unfortunately didnt come to the party.
Here again, only (a) can be employed out of context. (b), however,
requires that the subject is interpreted as a topic, and is therefore a
good response to Tell me about Hans. (a) is not a possible continu-
ation in this context. We can therefore conclude that in these Germanic
languages when the initial element does not qualify as a topic, then the
sentence is interpreted with an implicit stage topic. Fronting a non-
topic signals a particular IS, namely one in which none of the overt
elements is a topic.
Svenonius claims that the initial position is not a simple topic
position as argued above; he claims it is a switch topic. He includes
in this category contrastive foci, speaker-oriented adverbials, discourse
connectives, scene-setting adverbials, and actual switch topics. If no
switch topic is available, a continued topic, often the subject, is placed in
initial position, and if neither a shift topic nor a continued topic is avail-
able, an expletive may appear. Svenonius description of the elements
in initial position can be captured by the following generalization: The
initial element in Germanic is either a topic or else the sentence is inter-
preted as having a stage topic. Since the class of topics includes contin-
ued topics, switch topics, contrast, and overt stage topics, topicalization
in Germanic in these cases can be seen as motivated by the movement of
a topic. It is only when the fronted element is itself not a topic that such
motivation fails. This is one of the reasons why Svenonius rejects the
cartographic approach and also the reason why Frey separates topical-
ization from movement to FinP which is not motivated by information
configurations II,
The idea that formal or stylistic fronting may have a discourse func-
tion, albeit not one connected to the fronted element itself, has also
been proposed by Holmberg :ooo, following . Kiss I,,8.
His pro-
posal is that subject llers, such as those created by stylistic fronting
as well as expletives, create a structure which allows the merger of what
we have here referred to as a stage topic.
. Kiss :oo: II, develops this argument further and rephrases the EPP
as a requirement that sentences express predication about a constituent
in SpecTopP. She allows this constituent to be an overt or implicit stage
topic and explains the unacceptability of Hungarian sentences with indi-
vidual-level predicates with no constituent in SpecTopP as a violation of
this version of the EPP: Since individual-level predicates do not allow stage
topics, no predication of an implicit stage topic is possible.
. Kiss fur-
ther argues that her version of the EPP also applies to a non-topic-promi-
nent language such as English. Evidence is drawn from the placement of
speaker-oriented sentential adverbials in English (her (,:): II,):
(o,) a *[
A baby boy luckily was born]
b Luckily [
a baby boy was born]
c John luckily [
was born on time]
(o,a) shows that a non-specic subject cannot be followed by a sen-
tence adverbial and (o,b) shows that the adverb must precede such
a subject. (o,c) shows that a specic subject may precede the adverb.
According to . Kiss the preverbal subject in this example is located in
SpecTopP. (o,b), however, is predicated of an implicit stage topic (also
situated in SpecTopP). These data are similar to the German data in
(ooa) and (oI). There too an initial sentence adverb indicates a stage
topic. The question that arises is how this word order is motivated. For
. Kiss the adverb demarcates the border between topic and predicate,
i.e., the topic projects above the adverb. When the topic is an implicit
stage topic, the adverb shows up in initial position.
For Frey, however,
the adverb moves, hence his assumption that its movement cannot be
motivated by information structure.
Although worthwhile, a comparison between Freys formal fronting and Holmbergs
stylistic fronting will not be undertaken here.
See Chapter :, section I.o for an account of the differences between individual-level and
stage-level predicates in terms of information structure.
According to . Kiss :oo, existential there is also fronted to SpecTopP, forcing an inter-
pretation with an implicit stage topic.
II8 configurations
Rizzi :ooa also discusses fronting of non-topic adverbials and argues
for another (recursive) left-peripheral projection, Mod, to accommo-
date their distribution (his (oo) ):
(o) Force Top* Int Top* Focus Mod* Top* Fin IP
Rizzi assumes that salience is associated with this projection. He doesnt
clarify what he means by that, although it seems that all left-peripheral
elements are salient in this sense. Although examples are given of adver-
bials which function as topics and as foci, and therefore land in the appro-
priate projections, most adverbials do not have informational properties.
Rizzi distinguishes topics from such fronted adverbials which, although
they share intonational properties with topics, cannot be said to be what
the sentence is about. In order to explain the dierence between these
two categories with respect to extraction (topics do not trigger minimal-
ity interactions, adverbials do), Rizzi oers two possible explanations. (I)
It follows from the recursive property of topics. Hence he predicts that
in languages which do not allow multiple topics, the minimality inter-
actions with topics should appear, but he has found no case to test this
prediction. (:) Rizzi associates the left-peripheral projections with the
features + Arg, +Mod, +Q each with its own dening features.

According to Rizzi, only topics can be viewed as dened in a purely neg-
ative way with respect to this system. They are neither argumental, nor
quanticational, nor modicational. Since relativized minimality (RM)
eects depend on sameness of structural type expressed by a positive
feature specication, topics would be correctly excluded from the system
of positions which exhibits RM eects. Both attempts at explaining the
dierence between topics and other left-peripheral elements seem highly
Rizzi :ooa also suers from the same criticisms outlined
with respect to Rizzi I,,, in sections :.I and :.: above.
Another addition to the I,,, scheme is Int, the landing site of higher wh-elements
such as perch (why).
The features associated with argumental elements are person, number, gender, and
case. Those associated with quanticational elements are Wh, Neg, focus, etc., and Modiers
include the following: evaluative, epistemic, Neg, frequentative, declarative, measure, and
For another explanation of why topics are not subject to the same constraints as other
elements when moved see Erteschik-Shir I,,,, recapitulated in Chapter , section :..
For further discussion and criticism of Rizzis view of the left periphery and its applica-
tion to a variety of languages, see among many others Paul :oo:; De Cat :oo:; Massam :oo:;
Lpez :oo,; Aboh :oo.
configurations II,
All the cases listed in this section are illustrations of a left-peripheral
element which does not itself have information-structural properties,
yet it signals a particular information structure, namely one with a stage
topic. These structures also have another property which may explain
their information-structural status: Due to the fact that the left-peripheral
element is not an argument, the subject is never in its canonical position
in which it is interpreted as a topic. The motivation for these constructions
may therefore be to oust the subject from this position in order to enable
a non-topic interpretation of the subject, rendering, in this case, an out-
of-the-blue interpretation.
Another reason to postpose the subject is to
enable its interpretation as a focus. This is the topic of the next section.
3.3.5 Focusing the subject: existentials
One way of marking a sentence as being all focus and having a stage topic
is therefore for a non-topic to occupy the left-peripheral position. According
to . Kiss :oo and Holmberg :ooo, existentials employ exactly this strategy.
The outcome is an all-focus sentence predicated of a stage topic:
(o,) a (s)Top
[There is a/*the dog in my garden]
b sTop
[There is a/*the meeting at two oclock]
c (s)Top
[There are many/*all people who like ice cream]
In such an f-structure, the full sentence is entered on the card for the
current here-and-now which provides the stage topic (using f-struc-
ture). When the stage topic is lacking in contextual denition, i.e., either
the place or the time is not contextually available, then a new stage
is dened by adding these parameters to the stage. This can be seen in
(o,). In (o,a) the location is not given contextually and in (o,b) the time
is missing in the context. In (o,c) no locative parameter is contextually
available, yet this parameter is not provided in the sentence either; the
new stage is accommodated to mean the whole world.
The denition
For the observation that, cross-linguistically, topics have a strong tendency to coincide
with subjects see among others Gundel I,,,; Givn I,,o; Li and Thompson I,,o; Reinhart
I,8I; Lambrecht I,,.
The elements following be must provide a focus. Without focusable elements the sen-
tence will be incomplete, explaining why sentences such as
(i) There are people.
(ii) There are cats.
seem incomplete unless they receive an ontological interpretation. Such an interpretation is
less feasible for (i) than for (ii).
I:o configurations
of a new stage requires new inventory. Denites presuppose a referent
associated with a location. Located referents are therefore incompatible
with the interpretation of a new stage. This is the explanation for the
deniteness eect in existentials. The denition of a new stage in this
way also provides an explanation for when the deniteness eect applies.
(oo) illustrates examples in which it does not hold:
(oo) a Theres city hall, the museum, and the park.
b Theres the meeting at : oclock and the oce event at .
Such existentials generally provide a list of elements contained in a certain
place, or time: (ooa) could be a description of the sights in a given town.
(oob) could be a response to a request for the days schedule at the oce.
In both cases, the context must include reference to the stage in question,
namely the town and the oce events, respectively, but whats special about
these stages is that they are unpopulated, and it is in this sense that they are
new. The inventory which is listed in the existential may be given, yet it is
new to the stage in question. An obvious dierence between the sentences
in (o,) and (oo) is that the former lack at least one of the parameters of the
stage and the latter require full contextual specication of the stage (e.g.,
for (ooa), a particular city, and for (oob), a particular day at work). Since the
stage is not new, the inventory on it need not be new either. The denite-
ness eect is therefore predicted to hold only of new stages.
According to Wagner :oo,, structures which exhibit a deniteness effect are in fact
not incompatible with denites but only with DPs that are given. He gives the following
(i) a # Theres the mother in the garden.
b This soap opera is just like Hamlet. Theres the mother, who conspires to kill her
Wagners illustration ts the account given here.
For a characterization of denites in existentials see also Hannay I,8,.
The same deniteness effect is also found in locatives such as (ia) and (b) but not in
possessives such as (ic):
(i) a My soup
[has a/
the y in it]
b John
[has a/
the hat on]
c John
[has a/the hat (in his hand)]
In (ia) and (b) the subjects are interpreted as locations and therefore function as stage topics.
Their f-structure is therefore parallel to that of the sentences in (o,) in that these stage topics
also require the lling in of the location by a prepositional phrase. (ic) differs in that the sub-
ject is interpreted as a possessor and not as a location. The deniteness effect does not apply
and the addition of a locational prepositional phrase is optional.
3.3.6 Clefts: syntactic mapping of information structure
According to Gundel clefts are also employed in order to mark
subjects as non-topics: information structure is more directly mapped
onto syntactic structure in the sense that the focus ( information-
structural predicate) is in post-copular position and is kept out of
syntactic subject position. Gundels paper poses the question as to
why Norwegian clefts (b) are so much more frequent than English
clefts (a).
() a It was Nielsen who won.
b Det var Nielsen som vant.
It was REL won
This is an important question in view of the fact that the information-
structural properties of clefts in the two languages are the same:
In both languages, the cleft clause encodes information that is presupposed or taken
for granted in some sense, so that it may be conceptualized as a single, individuated
entity, with a referential givenness status of at least uniquely identiable. With respect
to topic-focus structure, both languages allow clefts where the cleft clause encodes
the topic and the clefted constituent is the information focus/comment. ()
Both languages also allow the reverse situation in which the cleft clause
is the focus (or part of the focus). According to Gundel, the source of
the dierence is that there is a stronger tendency in Norwegian to map
information structure directly onto syntactic structure.
In the case of
clefts it allows the material in the cleft clause to be treated on a par with
other uniquely identiable material, such as the content of a denite
description even when it is not the topic of a sentence as well as keeping
the focused material out of subject position.
3.3.7 Other strategies for marking subjects as non-topics
Existential constructions and clefts are thus ways for subjects to be
marked as non-topics. Existentials and clefts are constrained in a
number of ways so that other strategies for marking the subject as a
non-topic are also needed. According to Lambrecht : , such stra-
tegies involve cancellation of those prosodic and/or morphosyntactic
subject properties which are associated with the role of subjects as topic
See also Erteschik-Shir in which a similar point is made in comparing Danish and
I:: configurations
expressions in PF (predicate focus) sentences.
Lambrecht :ooo: o:,
allows for two dierent implementations of this strategy. One way is by
assigning prosodic prominence to the subject (prosodic inversion), the
other is by dislocating the subject (syntactic inversion). The former tack
is taken by English, the latter by Italian (Lambrechts (I8) ):
(o8) a Her husband is SICK.
b Il suo marito e malato. (Italian)
(o,) a Her HUSBAND is sick.
b E malato il suo marito.
Lambrecht argues that the purpose of these inversions is the marking of
paradigmatic contrast by assigning the subject object properties, rather
than the iconic view that the focused subject must occur towards the end
of the sentence in order to be nearer to subsequent mention of its refer-
ent as a continued topic. This explanation does not apply to the English
case in (o,a).
Syntactic accounts of postposing focused subjects have had to
overcome the ban on rightward movement. Zubizaretta I,,8: I,8, for
example, accounts for the inversion of focused subject in Italian as
follows: the focused subject is in the specier position of a functional
projection above TP (in the SVO order). Then, p(rosodically
motivated)-movement left-adjoins the TP (VO) to the FP rendering
the VOS order and the correct interpretation of the subject as a focus.
One of the arguments in favor of an account in terms of p-movement
is that Italian is sensitive to the relative weight of the constituents
which invert. P-movement applies in cases of prosodic conict,
i.e., when dierent prosodic rules assign conicting prominence
to two metrical sister nodes.
Samek-Lodovici :oo, supports the
prosodically driven analysis originating with Zubizaretta which
explains the nal position of focused subjects as the position which
According to Lambrecht, a distinctive property of all-focus sentences is the absence of a
topiccomment relation between the subject and the predicate. He does not invoke stage
For an analysis of this phenomenon within RRG theory see Van Valin and LaPolla I,,,:
I8I, and Van Valin :oo,: I,,.
Zubizaretta argues that in spite of its prosodic motivation, p-movement is syntactic. She
also gives an account of Spanish postposed subjects in terms of p-movement, yet her analysis
of Spanish differs signicantly from that of Italian. For an account of Spanish subject place-
ment, see also Casielles-Surez :oo.
is compatible with stress assignment. Samek-Lodovici shows that
seeming counterexamples to Zubizarettas generalization, cases in
which the focus is not rightmost in the sentence, can be accounted
for if it is assumed that such cases result from the dislocation of
discourse-given constituents to the right of the focus. Such right-
dislocated material is pronounced with an intonational pause and
remains unstressed. An example of a non-nal focused subject in
Italian is given in () (from Samek-Lodovici, his ()). (Note that
right dislocation applies to given, yet heavy material, explaining
Zubizarettas observation that the positioning of the focused subject
in nal position in Italian is sensitive to the relative weight of the
subject and the constituents that follow it.)
() Context: We were told that Bill took the children to the zoo.
No. Li ha portati allo zoo GIANNI, i bambini
No. Them has taken to-the zoo John, the children
No. JOHN took the children to the zoo.
Samek-Lodovici thus argues for a xed position for foci at the right
periphery of the IP. Clause-internal and initial foci arise due to the
dislocation of constituents to the right of the focus. The more right-
dislocated constituents there are the further to the left the focus will be
found. This analysis of the position of Italian foci invalidates the idea
of a left-peripheral focus position la Rizzi.
Samek-Lodovicis conclusions concerning the right-peripheral position
of foci in Italian, if correct, may signal that movement to the left periphery
is restricted to the identication of topics (including constrastive ones). If
this is so, then this may be because the unmarked case in languages is one
in which there is a one-to-one mapping between grammatical subject-
predicate and information-structural subject-predicate structures.
3.3.8 Left-peripheral prospects
One of the questions that arise again and again in the study of the infor-
mation-structural import of elements at the left periphery is their pre-
cise function in information structure. In many languages, the leftmost
element is often dened as the topic. In some languages the topic may
be contrastive as well, yet in others, only a non-contrastive topic can
occupy this position. In yet other languages, there is a preference for
switch topics.
See Erteschik-Shir and Gundel among others for the role of such mapping.
I: configurations
Similarly, foci fronted to the left periphery are not uniform. According
to most researchers, the Hungarian preverbal focus position hosts foci
which are selected from a discourse-specied set. These restrictive foci,
as argued in Chapter :, section ,.,, have properties of topics, allowing
them to function as topics, at least in some languages. Similarly,
the Italian preposed focus is also restrictive. Informational foci are
usually excluded from the left periphery. If this is so, the left periphery
seemingly hosts only topics of various kinds with signicant variation
across languages in terms of which types of topics occur there, and what
the order between them is. If the particular topic and focus types are
carefully teased apart with these questions in mind, a quite dierent
picture of the left periphery may emerge.
Another point made here is that movement of information-structural
inert elements need not mean that the structure in which they occur in
fronted position does not dier in information structure from a struc-
ture in which these elements remain in situ. Fronting such elements may
render a thetic focus structure with a stage topic. In other cases, an elem-
ent is fronted altruistically in order to focus another element such as the
A lot of work has been produced especially in the last decade exam-
ining the impact of information structure on word order, yet beyond
the well-known fact that topics tend to precede foci, not much progress
has been made as yet. Some of the syntactic frameworks discussed in
Chapter : may be better suited to make generalizations concerning
the word order of particular languages than minimalism, the syntactic
framework most discussed here. Yet, an explanatory cross-linguistic
account of information structure-triggered word order is still
3.4 Scrambling
According to Bailyn :oo:a: 8:, scrambling is a general cover term for
the process that derives non-canonical word order patterns in so-called
free word order languages such as Japanese, Russian, German, Hindi,
and many others. In such languages, constituents can appear in a variety
of surface orders, without changing the core meaning of the sentence.

The term originates with Ross I,o, who treated it as a stylistic rule. For
many researchers it has remained extra-grammatical both because of
Bailyn :oo:a; :oo:b is an excellent review of the literature on scrambling. Here I deal
only with those issues that are signicant with respect to IS.
configurations I:,
its optionality and because it has no semantic import. Scrambling is
distinguishable from topicalization in that it is not necessarily restricted
to root clauses, yet it is not always easy to tease apart particular cases in
which an element is scrambled to the left periphery. Since Rosss initial
observations, movement of constituents has been distinguished along a
number of parameters, one of them the distinction between A-bar and
A movement. Topicalization has been accounted for either in terms of
A-bar movement or as Adjunction at the left periphery. Scrambling has
received mixed accounts, with a large body of work dedicated to decid-
ing whether it should be analyzed as A-bar or A movement. There is little
agreement on this issue. Nor is there agreement concerning whether
there is one uniform scrambling phenomenon, or whether there are
several. Van Gelderen :oo, denes scrambling as (syntactic) movement
driven by information structure. She therefore includes topicalization
as a type of scrambling. At least three types of scrambling have been
dened: (I) short scrambling (within the VP), (:) medial scrambling (to
a VP-external position), and (,) long-distance scrambling (across a CP
In spite of the interest this issue has engendered, a theory
which predicts the occurrence of one or more of these cross-linguistically
is not yet available.
For most syntactic accounts, the optionality issue is
the most dicult to resolve. Optional reordering is rarely truly optional
and is often motivated by information structure.
Whether this is the
case is the topic of this section.
Van Gelderens account of scrambling in Russian, Japanese, and Dutch
provides a good starting point for the discussion.
3.4.1 Scrambling in Russian
Russian is known to be non-congurational, with few restrictions, if any,
on word order. The following set of examples from van Gelderen :oo,:
,, illustrates that a sentence consisting of a subject, verb, and object can
occur in all six possible orders:
See, e.g., Takano I,,8 for these distinctions. The discussion here will mostly be limited
to short scrambling. Languages differ as to whether scrambling is clause bound (German) or
not (Russian, Japanese). See Bailyn :oo:b for more details and references.
According to Bailyn :oo:b, Slavic languages do not conform to the idea originating with
Fukui I,,, and supported by Neeleman and Reinhart I,,8 that scrambling languages must
be head nal. But see van Gelderens account of the scrambling parameter in section .,.
Bailyn :oo:b lists the following research on the relation between scrambling and dis-
course function: Rudin I,8,, Miyagawa I,,,, and Haider and Rosengren I,,8, as well as
Neeleman and Reinhart I,,8, which view this relation as a property of the PF interface.
() a Ivan kupil knigu
Ivan.NOM bought book.ACC
b Ivan knigu kupil
Ivan book bought
c Knigu Ivan kupil
book Ivan bought
d Knigu kupil Ivan
book bought Ivan
e Kupil Ivan knigu
bought Ivan book
f Kupil knigu Ivan
bought book Ivan
Ivan bought the/a book
What interests us here is rst and foremost the information-structural
properties of these word orders.
In view of the fact that van Gelderen
: denes scrambling as movement driven by IS, she is a good
source for this endeavor.
By using the questionanswer test for focus (but abstracting away
from the propensity in Russian of replying to a question with an
identical word order), van Gelderen : concludes that SVO and
SOV, as well as OVS and VOS, allow the subject to be an informational
focus, but that OSV and VSO never allow it. The object can also be an
information focus in the SVO and SOV orders. A focused object also
occurs in VSO sentences, but not in OSV, VOS, and OVS. V can be
assigned informational focus in SVO and SOV and the entire sentence
can be assigned informational focus in SVO and SOV. The following
generalization emerges here: Both SVO and SOV function as unmarked
word orders and allow all informational focus assignments. In all other
cases, the informational focus is restricted to sentence-nal position.
Van Gelderen : provides the following evidence that the topic
is in initial position in the unmarked orders:
() A: Kto-to vse -taki etu knigu kupil
someone.NOM all-this-way this book.ACC bought
Still, someone bought this book.
B: Etu knigu kupil Ivan (OVS)
this book.ACC bought Ivan.NOM
This book, Ivan bought it.
Van Gelderen : provides convincing arguments for why a syntactic derivation of
the unusual orders is not feasible.
For some reason van Gelderen does not test VSO for topic position, although she says
that the initial element in all the marked word orders can be a topic.
configurations I:,
(,,) A: Kto-to vse -taki etu knigu kupil
someone.NOM all-this-way this book.ACC bought
Still, someone bought this book.
B: Kupil etu knigu Ivan (VOS)
bought this book.ACC Ivan.NOM
Ivan bought this book.
(,) A: Gde eta kniga voob e-to?
where this book.NOM generally -this
Mne kaetsja ona zdes leala
me.DAT seems she.NOM here lay
Where is this book anyway? It seemed to me that it was lying here.
B: Etu knigu Ivan vzjal (OSV)
this book.ACC Ivan.NOM took
This book, Ivan took it.
She concludes that the marked orders all exhibit initial topic and nal focus.
According to King I,,,, however, the basic word order of unemotive
sentences is VSO and this is also the word order which allows sentential
King I,,,: ,,8 gives the following examples:
(,,) a el dod
go rain
Rain was falling.
b [Nelyno proletela kakaja-to neizvestnaja ptica]
noiselessly y past some unknown bird
Some type of unknown bird ew noiselessly past.
c [Prislal mu dengi.]
sent husband money
My husband sent (me) the money.
Note that these examples all depict changes of location and can
therefore be analyzed as unaccusatives. If the only instantiation of
VSO out-of-the-blue sentences is of this type, then Kings argument is
See also King I,,,; Erteschik-Shir and Strahov :ooo; :oo for the generalization that
the sentence-nal position is designated for foci.
According to Natalia Strahov (personal communication) OSV sentences can still have
stress on the object. However, in this case, the object is stressed just as it would have been
stressed in sentence-nal position. These sentences are known as emphatic sentences which
are emotionally colored.
King derives this order by employing the VP-internal subject analysis and forcing the
verb to raise to I.
I:8 configurations
severely weakened. Bailyn :oo,, following most researchers of Russian,
argues for SVO as the unmarked order of Russian sentences. He
claims that this is the order of all-focus sentences and that this order
allows for ambiguity with respect to topic and focus assignments. It
also allows falling (neutral) intonation and is the word order found
most frequently. Van Gelderen :oo,: also argues that the unmarked
word order in the language is SVO using similar tests. Bailyn and van
Gelderen thus agree that SVO is an unmarked word order in Russian.
According to van Gelderen, SOV is also unmarked in that it allows for
an all-focus interpretation. King, however, disagrees with both these
authors and assumes that VSO is unmarked. She argues (King I,,,:
I,,) that the reason subjects frequently occur in preverbal position is
that the element selected as the subject of predication (i.e., the topic)
is the thematically most prominent one, namely the subject. All three
authors agree, however, that generally topics occur in initial position
and that foci occur sentence nally.
Van Gelderen allows early spell-out in certain languages, so that
the linear order of elements which are not fully merged in the syntax
is determined according to information structure requirements at the
PF interface (see Chapter :, section .,). The parameter that licenses
early spell-out is rich case morphology. It follows that Russian, but not
English, will allow early spell-out. She also allows optional IS-driven
syntactic movement. This is possible in languages which do not license
early spell-out, yet languages which do allow early spell-out, such as
Russian, may also allow such syntactic processes.
SVO is the basic word order in Russian, according to van Gelderen. She
claims, however, that SOV also functions as an unmarked word order with
respect to focus assignment. Van Gelderen :oo,: ,, also points out a num-
ber of properties that SOV has in common with SVO including that they
both allow embedding and neutral (non-contrastive) intonation. The
orders VSO, VOS, and OVS cannot be embedded and can only have marked
intonation. The order SOV is parallel to cases of object shift in other lan-
guages. As for the OSV order, it seems to belong to both classes, depending
on whether there is a comma intonation following the object or not.
When the order OSV occurs with an intonational break between
the object and the rest of the clause, it is akin to topicalization in other
See also Strahov :ooI for an analysis of Russian word order and Erteschik-Shir and
Strahov :ooo; :oo for the generalization that the sentence-nal position is designated for
foci. For an account of Russian word order within RRG, see Rodionova :ooI.
configurations I:,
languages. Van Gelderen :oo,: o, argues that the fronted object in the
O,SV order is interpreted as a contrastive topic (her (,I) ):
(,o) A: Nu i kak, kupili knigu ili kartinu?
well and how bought.PL book.ACC or painting.ACC
Well, did you buy the book or the painting?
B: Knigu, Ivan kupil, a kartinu, net
book.ACC Ivan.NOM bought but painting.ACC not
As for the book, Ivan bought it, but as for the painting, he didnt.
According to van Gelderen, this order is formed in the syntax by topical-
ization to Spec,TopP.
King I,,,: o,,o also argues for two dierent types of topicalization.
Following Aissen I,,:, she refers to these as internal and external top-
ics respectively, the latter being equivalent to van Gelderens contras-
tive topic which is separated from the sentence by comma intonation.
According to King, the function of such topics is that of a shift topic
which she denes as having been mentioned in the previous utterance,
but not as a topic. (Contrast is, of course, one type of shift topic, but not
the only one.) Both types of topic can occur in Russian sentences. In this
case, the shift topic precedes the repeated topic.
The orders SOV, SVO, and O,VS, in van Gelderen, are therefore accounted
for by normal syntactic means. The other three orders, according to van
Gelderen, are a result of linearization at PF according to IS requirements.
This is possible in Russian in view of the fact that it allows early spell-out.
Allowing for a division between word orders that are generated in the syn-
tax and those that are not has certain interesting consequences. For example,
Van Gelderen explains the asymmetry between main and subordinate clauses
as follows: subordinate clauses are formed in syntax, hence they instantiate the
hierarchical structure which the syntax provides. No reordering is therefore
possible at PF. This predicts that only those orders derived in the syntax will
occur in subordinate clauses. This is why only SVO, SOV, and O, SV can be
embedded. Similar explanations hold for the other asymmetries.
Topicalization for contrastive purposes is not unusual in languages, and
the fact that it occurs in embedded clauses is not surprising if these clauses
are root clauses as dened in Emonds I,,o; :oo, since these function as the
main assertions of the sentence. In fact, the examples van Gelderen oers are
The contrastive topic feature thus triggers syntactic movement. Her approach to con-
trastive topics thus differs from her account of contrastive foci which are assigned the feature
[+CF] and are interpreted in situ. See Chapter :, section ., for discussion of the conceptual
basis of van Gelderens account.
all embedded under the verb think, which allows for embedded root clauses.
Whether contrastive topicalization in Russian is syntactic as van Gelderen
claims is, however, not as clear as she would have it. In order to see this, we must
rst examine van Gelderens other case of syntactic reordering in Russian,
object shift. Van Gelderen quotes Svenonius who claims optionality and
IS motivation as dening properties of syntactic object shift. Since the SOV
order is optional, and since it also has the properties of an unmarked order,
van Gelderen : concludes that object shift is a syntactic rule which
must also be motivated by IS. Surprisingly, however, van Gelderen shows
consistently that there are no IS properties that dierentiate SVO and SOV. If
this is so, then it is hard to understand what she means by IS motivation. Note
that even if focus projects in the SOV order as it does in the SVO order so that
the main f-structures of the two orders are identical, the role of the objects
in the two orders may still dier in their subordinate f-structure. A careful
examination of the two orders and their IS properties is therefore in order.
The account of Russian word order, according to van Gelderen, is
therefore twofold: Contrastive topicalization and object shift are syntac-
tic, all other reorderings are the result of linearization at PF where topics
are linearized to the left and foci are linearized to the right. Syntactic topic and focus features and Russian word order The
distinctions van Gelderen makes between syntactic and PF reordering, both
motivated by IS, can be captured by the framework introduced in Erteschik-Shir
a and discussed in Chapter , section .. In this framework, topic and
focus features are selected in the lexicon and optionally project by feature
projection through their syntactic projection. This captures the connection
between syntactic structure and interpretation and can potentially account
for the Russian scrambling data via linearization in the phonology without
recourse to IS-motivated syntactic movement. This is because the projection
of topic and focus features is xed in the syntax. The basic word order, SVO,
will allow projection of a focused V to the dominating VP and IP. Reordering
at PF optionally takes any topic and linearizes it to the left periphery and
takes any focus and linearizes it to the right.
The characteristic f-structure
of contrastive topics on the left would also trigger comma intonation (an
eect van Gelderen does not account for). Since the main f-structure of the
SOV order is identical to that of the base order, i.e., it is the main verb which
carries the focus feature, reordering here would have to depend on the
subordinate f-structure assigned to the object as indicated in the previous
section. There is no reason why the position of the object in this case could
For evidence that the sentence-nal position in Russian is a designated focus position,
see Erteschik-Shir and Strahov .
not be established by PF linearization as well. It is reasonable to assume that
main topics and foci will be marked by word order only in main or root
clauses. This oers a simple explanation for why linearization to the periph-
eries is not available in subordinate clauses in Russian.
3.4.2 Scrambling in Japanese
Remember that the parameter that distinguished Russian, which allows
early spell-out and therefore scrambling, and English which does not,
is the availability of rich case morphology in Russian. According to van
Gelderen, Japanese is also a scrambling language, yet in spite of its case
morphology, it does not allow early spell-out. This is because it is rigidly
a verb-nal language and early spell-out would predict word orders in
which the verb would not remain in nal position.
Another reason for its rigid word order is that scrambling creates scope
ambiguity in Japanese where none exists in the unscrambled versions.
Scrambling must therefore feed LF, i.e., it must be syntactic.
Van Gelderen
is careful to point out that the dierent orders in Japanese have dierent
information-structural impact. (Here she diers from many others who
have investigated scrambling in Japanese.) Van Gelderens model there-
fore includes both information structure-motivated syntactic movement
(Japanese scrambling, Russian topicalization and object shift), as well
as information structure-motivated linearization at PF (Russian scram-
bling). In this way, she can maintain that the parameter for scrambling is
that the language be head nal. Russian is not an exception since its scram-
bling capability is due to early spell-out (). This move puts her proposal
at a disadvantage with respect to other models which do not require such
rich machinery. More signicantly, although she gives an account of why
syntactic scrambling is optional in terms of case checking, she does not
explain the particular information structure eects of Japanese scram-
bled structures. Her account is a sophisticated cross-linguistic account of
scrambling which at least supercially views IS as the motivation behind
changes in word order. IS motivation comes for free if scrambling is a lin-
earization eect at PF as it is in Russian. IS motivation is also necessary,
in van Gelderens model, for syntax-internal scrambling (Japanese), since
otherwise she would have no explanation for its peculiar properties. It is
possible that, had she examined the IS properties of Japanese scrambling
with at least as much care as she did with Russian scrambling, a more direct
connection between IS and scrambling could have been made.
Van Gelderen offers much more evidence for the fact that scrambling in Japanese must
be pre-spell-out. I refer the reader to her work for the details.
The basic scrambling data in Japanese are given in () (from Ishii ;
the foci are marked in bold).
(a) gives the basic word order in Japanese,
SOV. (b) shows that the object can be fronted across the subject.
() a Taroo-ga sono hon-o katta. (SOV)
Taro-NOM that book-ACC bought
Taro bought that book.
b Sono hon-o
Taroo-ga t
katta. (OSV: O scrambled)
that book-ACC Taro-NOM bought
According to Ishii, the preverbal position in Japanese is the unmarked posi-
tion for informational foci.
Therefore, as expected, (a) allows not only
focus on the direct object, but also on the VP and on the whole sentence.
The scrambled (b), by fronting the object, positions the subject in prever-
bal position and enables its interpretation as the informational focus. This
is the only possible focus in this sentence.
Here, as in the Germanic cases
examined in section above, scrambling of the object is altruistic, since the
object itself does not gain a particular IS status, yet the movement allows
the subject to be interpreted as a focus. Another way to look at this type of
scrambling is to adopt Ishiis view that scrambling is motivated as a type of
de-stressing process since the scrambled element is always presupposed.

Under this view the preverbal focus is derived from stressing the most deeply
embedded constituent as in Cinque (Chapter , section .). According
to Ishii, elements scrambled to the left periphery function as switch topics of
the type discussed in Chapter and in section above. This is probably also
what Miyagawa : had in mind when he describes the function of the
fronted object as a focus, by which he means a phrase which requires some
sort of emphasis. In certain cases, the emphasis may presuppose a previous
utterance of the phrase in the conversation, or something comparable. In
these cases it may be more accurate to use the term topic.
Scrambling of the direct object in a double object construction allows
focus on the indirect object in (b).
I exclude long scrambling (across a sentence boundary) from discussion here.
Kim has already shown that in SOV languages, the unmarked focus position is preverbal.
Van Gelderen discusses the fact that focus projection is possible in cases of scrambling
in Japanese. In section .., I suggested that focus projection in Russian stems from the
selection of the focus feature on the main verb and that scrambling of this focus structure
can occur if it is motivated by subordinate focus structure. In order to see whether this
hypothesis can explain the differences in focus projection between the two languages, more
details of the f-structure properties of the two languages are needed.
Ishii follows the general approach of Neeleman and Reinhart (Chapter , section
.), yet he argues that scrambling is a result of movement and not of base generation.
() a Taroo-ga Hanako-ni sono hon-o ageta.
Taro-NOM Hanako-to that book-ACC gave
Taro gave Hanako that book. (S IO DO V)
b Taroo-ga sono hon-o Hanako-ni ageta.
Taro-NOM that book-ACC Hanako-to gave
(S DO IO V:DO scrambled)
() illustrates that the indirect object can also be focused in situ by
stressing it.
() Taroo-wa Hanako-ni hon-o agemasita
Taro-TOP Hanako-to book-ACC gave
Taro gave a book to Hanako.
According to Ishii, (), in which stress shift has occurred, can only be inter-
preted with narrow focus on the indirect object. He claims that the reason for
this constellation is the need for presupposed elements to be contiguous. If the
giving of a book is presupposed, then the order and stress in () is preferred.
Finally, the object can also scramble across an adverb as shown in (b).
() a Taroo-wa hon-o isoide yonda (no desu).
Taro-TOP book-ACC quickly read NL COP
Taro quickly read a book.
Here again, according to Ishii, the adverb and the verb form the presuppo-
sition and must therefore be contiguous. This requirement forces the focus
to precede the adverb. Note that in both () and () the subject is marked
with the contrastive topic marker wa. Topics are of course presupposed,
but the contiguity requirement does not apply to them. The contiguity
requirement is stipulated and cannot, as far as I can see, be made to follow
from Neeleman and Reinharts theory. Although Ishiis article does not
provide a full account of scrambling in Japanese, its careful examination
of the associated IS properties makes it clear that an account of Japanese
scrambling without taking IS into account is doomed to failure.
One of the puzzles van Gelderen solves without recourse to proper-
ties of IS is the fact that scrambling generates quantier scope possibili-
ties not available in the basic word order. In Chapter , section . it was
pointed out that the denition of topic in terms of predication predicts
that the topic will always take wide scope. If this is the case then scram-
bling which xes the interpretation of a certain element as topic should
Van Gelderen : posits the base order to be S DO IO V. The examples given here
are from Ishii who gives the base as S IO DO V.
I, configurations
also have a xed scope interpretation with the derived topic taking wide
scope. The prediction is that movement for the purpose of topicalization
would necessarily limit the scope possibilities, not extend them, as is the
case in Japanese. Yet if scrambling does not x the position of the topic, it
would not have this eect. In order to see whether these f-structure pre-
dictions hold for Japanese, let us examine the data van Gelderen presents.
Van Gelderen :oo,: Ioo, rst points out that scope follows the surface
order of elements in Japanese unscrambled sentences:
(8I) Daremo-ga dareka-o aisiteiru
everyone-NOM someone-ACC love
Everyone loves someone.
x [y [x loves y] ]

*y [x [x loves y] ]
The scrambled examples in (8:) with OSV order, however, allow both
scopal readings:
(8:) a Dareka-o daremo-ga aisiteiru
someone-ACC everyone-NOM love
Everyone loves someone.
x [y [x loves y] ]
y [x [x loves y] ]
b Daremo-o dareka-ga aisiteiru
everyone-ACC someone-NOM love
Someone loves everyone.
x [y [x loves y] ]
y [x [x loves y] ]
The puzzle becomes even more intriguing when we observe that scram-
bling in double object constructions does not have this eect. Here both
the base order (8,a) and the scrambled order (8,b) allow both scopal
interpretations (the data are from Miyagawa I,,,: I:):
(8,) a Hanako-ga dareka-ni daremo-o syookaisita.
Hanako-NOM someone-DAT everyone-ACC introduced
Hanako introduced everyone to someone.
b Hanako-ga dareka-o daremo-ni syookaisita
Hanako-NOM someone-ACC everyone-DAT introduced
Hanako introduced someone to everyone.
Miyagawa marks the wide scope reading of the second QP as questionable. Hoji I,8,
marks it with a star (see van Gelderen :oo,: I:8). This may have to do with the propensity of
the quantier someone to take wide scope together with the fact that it precedes the uni-
versal quantier. Clearly, such effects should be controlled for.
configurations I,,
Miyagawa comments on this as follows: In fact the wide scope reading
of the rst QP is preferred in both word orders. The wide scope read-
ing of the second QP is most easily obtained if the rst QP receives focus
Remember that what Miyagawa refers to as focus is likely to be a
switch topic which ranges over a contextually available set. Also according
to Ishii :ooI, the preverbal position is the unmarked focus position. The
focus structure associated with scopal ambiguity in these cases is that the
rst QP functions as a switch topic and the second functions as a focus.
Let us now return to the initial scope ambiguity in (8:). Here again,
as mentioned above, the fronted object functions as a switch topic.
Therefore, the only dierence between the scrambled order in (8:) and
the base order in (8I) is the interpretation of the object as a switch topic
in the former but not in the latter. Abstracting away from whether or not
movement has occurred, the focus structure required for the second QP
to scope over the rst one is that the rst one be interpreted as a switch
topic, i.e., a topic selected from a contextually available set. If my inter-
pretation of the points made in Ishii and Miyagawa is correct then the
explanation for the availability of scope ambiguity in Japanese must be
the availability of this particular f-structure.
It is still not clear why a switch topic QP would allow a following focused
QP to take scope over it. This is particularly surprising in view of the fact
that were the rst QP a regular topic it would itself take wide scope, by
denition of topics, and therefore the interpretation in which the second
QP takes wide scope would be blocked. One possibility is that the switch
topic requires a contextually available set over which the QP ranges. The
inverse scope reading would be enabled if the second QP were also to
receive a restrictive reading and a pairing between the elements of each of
these sets resulted.
Yet another question remains. Why is there no scope ambiguity in the
base order (8I)? If the line of reasoning presented here is in the right
direction, the reason would have to be that the subject in left-periph-
eral position cannot be interpreted as a switch topic. Future research
will have to verify this. Interestingly, sentences such as these are in fact
ambiguous according to some speakers as noted in Van Valin :oo,: 8.
There it is argued that -ga marks both topics and foci and therefore the
scope ambiguity is predicted.
Miyagawa allows for the base generation of both orders as well as focus movement, so
that the examples in (8,) each have two derivations, one base generated and one scrambled.
If the wide scope of the second QP is achieved by stressing the rst one, then the rst one
must have been scrambled. See also Miyagawa :oo, for an updated analysis of these facts.
For similar points but different data see Kuno I,8: and Erteschik-Shir I,,,: I8:,.
I,o configurations
The point of this speculative section was to show that it might be worth-
while to investigate f-structure properties more carefully in order to see
whether the explanation for interpretative eects such as scope (and
probably also other properties such as binding) can receive an explanation
independently of the syntactic analysis of scrambling. This is an impor-
tant issue to decide since it is exactly these types of phenomena upon
which arguments for particular syntactic analyses of scrambling rest. It is
also important in order to reach a better understanding of linguistic varia-
tion. Van Gelderen :oo,: Io, uses the fact that scrambling modies the
scope possibilities in Japanese but not in Russian to argue for two dierent
accounts of scrambling in the two languages. But this dierence may have
to do with independent f-structure properties of the two languages and
may therefore not warrant van Gelderens conclusion.
Teasing apart f-structure properties and syntactic properties of scrambling
will therefore have to await further research, research that will take more seri-
ously the language-specic f-structure properties and their account.
3.4.3 Dutch scrambling
In Dutch direct objects can occur both before and after an adverb. This
is illustrated in (8) from van Gelderen :oo,: Ioo (her (I,) ):
(8) a Iedereen weet dat Jan op zondag het boek heeft gelezen
everyone knows that Jan on Sunday the book has read
b Iedereen weet dat Jan het boek op zondag heeft gelezen
everyone knows that Jan the book on Sunday has read
Everyone knows that Jan read the book on Sunday.
Van Gelderen adopts Neeleman and Reinharts I,,8 account of Dutch
scrambling discussed in Chapter :, section .,. This requires the base
generation of both orders, and therefore a dierent approach from that
oered for the other cases of scrambling she discusses.
In order to evaluate this approach, we rst examine the information-
structural properties of the two word orders in more detail. According
to de Hoop I,,: only strong DPs scramble.
This is illustrated in (8,)
from de Hoop :ooo: I,.
See among others Eckardt :oo,; Haider and Rosengren :oo,; Wagner :oo, and the refer-
ences therein for accounts of German scrambling that relate to information structure. German
is very similar to Dutch with respect to adverb-DP order. The two languages however differ
in their scrambling of double objects. This may have to do with availability of overt case in
German, but not in Dutch. Here the discussion is limited to scrambling across adverbs.
De Hoop proposes that the type of DP that scrambles is a generalized quantier. All
strong DPs belong to this type. When weak DPs such as two men are scrambled, they get a
strong (partitive or referential) reading, i.e., two of the men or those two men.
configurations I,,
(8,) a dat ik gisteren alle krakers/ twee krakers heb gesproken
that I yesterday all squatters/ two squatters have spoken
b dat ik alle krakers/ twee krakers gisteren heb gesproken
that I all squatters/ two squat yesterday have spoken
that I talked to all squatters/two squatters yesterday
c omdat iedereen (*een plas) nog *(een plas) moet doen
since everyone still a piss must do
since everyone still has to take a piss
(8,a) and (b) show that strong quantiers or weak quantiers that can
receive strong readings scramble optionally (with only the strong read-
ing of the weak quantier available in the scrambled version (8,b) ).
(8,c) shows that a weak DP which excludes a strong reading cannot
scramble. An additional factor is contrast: weak DPs which receive a
contrastive reading can occur before the adverb as shown in (8o) from
de Hoop I,,,.
(8o) a Wist jij dat Els gisteren een EENHOORN heft gezien?
Knew you that Els yesterday a unicorn has seen
b Wist jij dat Els een EENHOORN gisteren heft gezien?
Knew you that Els a unicorn yesterday has seen
Did you know that Els saw a UNICORN yesterday?
Since strong and contrastive DPs are potential topics, a plausible
account of Dutch scrambling is that it involves topicalization. Yet de
Hoop I,,,; :ooo argues that Dutch scrambling is truly optional and
that there are no interpretational dierences between the scrambled
and unscrambled versions. Therefore, if she is right, this generaliz-
ation cannot hold.
De Hoop rst shows that a context which forces a topic reading of the
object still allows both word orders. This is shown in (8,) from de Hoop I,,,.
(8,) A: Hoe zit het met Petra/je oude klasgenoten/de koningin?
Spreek je die nog weleeens?
How about Petra/your old classmates/the queen? Do you
still see her/them/her?
B: a Ja, ik heb vorige week Petra/al mn klasgenoten/
yes I have last week Petra/all my classmates/
twee van mn klasgenoten/de konigin opgebeld
two of my classmates/the queen called
b Ja, ik heb Petra/al mn klasgenoten/twee van mn
yes I have Petra/all my classmates/two of my
klasgenoten/de koningin vorige week opgebeld
classmates/the queen last week called
Yes, I called Petra/all my classmates/two of my classmates/
the queen last week.
The context in A asks about the objects and clearly forces a topic read-
ing of the object in the answer. The fact that this topic can occur both
postverbally as in (a) and preverbally in (b) indicates, according to
de Hoop, that topics need not scramble, although they may. Although
de Hoop employs a context to make her point, the discourse she cites
is not a natural one, since the full DP and not a pronoun is repeated in
the answer. Were a pronoun used, only the scrambled version would be
acceptable. Yet the fact that the rst set of answers in B are at all possible
does back up de Hoops claim that scrambling is truly optional.
De Hoop is further puzzled by the following data:
() a omdat ik zelden de kat aai
since I seldom the cat pet
b omdat ik de kat zelden aai
since I the cat seldom pet
since I seldom pet the cat
() a omdat ik gisteren de koningin zag
since I yesterday the queen saw
b omdat ik de koningin gisteren zag
since I the queen yesterday saw
since I saw the queen yesterday
() a omdat ik altijd de bus neem
since I always the bus take
b omdat ik de bus altijd neem
since I the bus always take
since I always take the bus
According to de Hoop, the sentences in () are well behaved with respect
to the topicality of the scrambled object. In the scrambled version, (b),
the unmarked reading is that the cat is linked to the previous discourse,
and in (a) the cat is not contextually available. (However, as shown
immediately below, context and intonation may allow for dierent topic
assignments.) () and () dier in this respect. In neither sentence does
the scrambled or the unscrambled version require that the object be con-
textually available. In () the queen is taken to be the current queen,
and in () the bus is taken to be whatever bus is relevant to transport-
ing the speaker in the reported event. De Hoop takes this as evidence that
scrambling in these two sentences does not function to signal a topic in
preadverbial position. However, as discussed in Chapter , section ., ref-
erents such as the queen, the sun, and even the bus are contextually
available between speaker and hearer with no need for previous mention.
It follows that they can always function as topics (namely, The sun is shin-
ing as an out-of-the-blue sentence). Since they dont have to function as
topics, either word order works.
The good behavior of () breaks down, according to de Hoop, when a
disambiguating context is supplied. She gives the following two contexts:
() a Recently, Paul seems to be under stress.
b Paul has a cat that seems to be under stress, recently.
Both orders are equally acceptable in both contexts. The context in (a)
does not oer an antecedent for the cat and therefore it should not be
interpreted as a topic and should not scramble. The context in (b)
introduces a cat into the context and therefore the object must be inter-
preted as a topic in the continuation, yet again both orders are good.
De Hoop concludes that adding a specic context limits the number of
interpretations and, as a result, there is no need for the word order to
mark the topic. Another way to look at these data is to assume that since
the deniteness of the object requires that the cat be known to both
speaker and hearer, the cat (just like the queen, above) can be interpreted
as a topic even when it is not introduced overtly as in (a). Further,
when it is introduced as in (b), it should be able to scramble, and the
question that remains is why it doesnt have to.
Another piece of evidence presented by de Hoop for her claim that
scrambling is not equivalent to topicalization is that no scopal eects
are associated with scrambling, as would be expected if the scram-
bled order necessarily has a dierent IS from the unscrambled one:
Although the unmarked case is that the word order does reect the
scopal order (and therefore scrambling does seem to have the expected
eect), this interpretation can be overruled by stress. The following
examples are also from de Hoop (the translations indicate the
scopal interpretation):
() a dat Jan drie KEER alle meisjes kuste
that Jan three times all girls kissed
All girls, Jan kissed three times.
b dat Jan alle MEISJES drie keer kuste
that Jan alle meisjes drie keer kuste
Three times, Jan kissed all girls.
In (a) the stressed adverbial three times precedes the object, yet it
has narrow scope, and in (b) the stressed object precedes the adver-
bial, yet it too has narrow scope. An obvious response to de Hoop is
that intonation changes information structure and therefore changes
the scopal interpretation. If, for example, the object in (b) is contras-
tive (as it would be if we want to maintain the idea that the scrambled
object is a (contrastive) topic), then the reading indicated is what we
would expect.
More serious are the following examples from de Hoop
which show that an unscrambled object can take wide scope (a)
and that an adverb can take scope over a preceding scrambled object.
() a dat Jan minstens n keer per jaar al zn vrienden bezoekt
that Jan at least once a year all his friends visits
All his friends, Jan visits them at least once a year.
b dat Jan alle brieven een keer versheurd heft
that Jan all letters once torn up has
Once, Jan tore up all letters.
The reason for these scopal interpretations, according to de Hoop, is that
the other scopal interpretation is pragmatically implausible. Although de
Hoop employs this example as further evidence for her claim that scram-
bling is optional in Dutch, one might also conclude that, in these cases, the
accepted reading is accommodated in view of the implausible reading that
is derived from the linear order of the quantiers in the sentence.
De Hoop takes her conclusion further. She argues that in Dutch the rela-
tion between IS and syntactic structure is exible. Dutch thus diers from
languages such as Catalan, Italian, Turkish, and Urdu in which this relation
is strict.
More specically, if there is a conict between context and scram-
bling in Dutch, context denitely wins. This conclusion leads her to an opti-
mality theoretical analysis. A lot therefore rests on de Hoops interpretation
of her data. Another interpretation of the same data is to note that the func-
tion of scrambling in Dutch may be to mark a subordinate topic and not the
See Chapter , section . for an account of how scope can be read off f-structure.
See de Hoop for the relevant references.
main topic of the sentence. In the examples listed above, the subject is the
main topic of the sentence. In principle, some other constituent fronted to
the left periphery could also play this role. It follows that, although scram-
bling may be restricted to topical elements, these never function as the main
topic of the sentence, but only as a subordinate one. It is plausible that the
importance of marking the main topic and focus is more critical for pro-
cessing than marking subordinate ones. This may be why scrambling is
not required in all cases in which the object may be interpreted as a topic.
It is unlikely that this is a sucient explanation, however, in view of the fact
that there are scrambling facts that are rigid in Dutch: Pronouns, which are
necessarily interpreted as topics, must scramble and non-specic, non-con-
trastive indenites which cannot be interpreted as topics cannot scramble.

That is why exibility only shows up in the case of denites.
According to Schaeer ; ; to appear, the optionality of the scram-
bling of denites is only apparent. She makes three observations which lead
her to this conclusion. First she shows that the requirements for scrambling
dier depending on whether scrambling is over negation or over a senten-
tial adverb. () shows that scrambling of denites over negation is obliga-
tory and () shows that scrambling over a sentential adverb is seemingly
optional (examples from Schaeer to appear: chapter ):
() a dat Hella het boek niet gelezen heeft
that Hella the book not read has
that Hella hasnt read the book
b *dat Hella niet het boek gelezen heeft
that Hella not the book read has
() a dat Saskia waarschijnlijk de krant gelezen heeft
that Saskia probably the newspaper read has
that Saskia probably read the newspaper
b dat Saskia de krant waarschijnlijk gelezen heeft
that Saskia the newspaper probably read has
that Saskia probably read the newspaper
Schaeers second observation is that when the sentence includes both
a sentential adverb and negation, the scrambled object can occur in two
dierent positions, one before negation but after the adverb (a) and
one before both.
According to Schaeffer ; Dutch weak pronouns cliticize to the verb on their
() a dat Saskia waarschijnlijk de krant niet gelezen heeft
that Saskia probably the newspaper not read has
that Saskia probably didnt read the newspaper
b dat Saskia de krant waarschijnlijk niet gelezen heeft
that Saskia the newspaper probably not read has
that Saskia probably didnt read the newspaper
Schaeer concludes from this evidence that there must be two di-
erent landing positions for scrambled items. Schaeers third observa-
tion is that when the context provides an antecedent for the referent of
the object, scrambling over sentential adverbs is also obligatory (or at
least that these cases exhibit a strong preference for scrambling):

() A: Wat zei je dat Saskia met de krant
what said you that Saskia with the newspaper
gedaan heeft?
done has
What did you say Saskia did with the newspaper?
B: (a) Ik zei dat ze de krant waarschijnlijk
I said that she the newspaper probably
geLEzen heeft.
read has
I said that she probably read the newspaper.
(b) ??Ik zei dat ze waarschijnlijk de krant
I said that she probably the newspaper
geLEzen heeft.
read has
When such an antecedent is not contextually available, scrambling is
strongly dispreferred:
() A: Wat zei je dat Saskia gedaan heeft?
what said you that Saskia done has
What did you say Saskia did?
B: (a) ??Ik zei dat ze de krant waarschijnlijk
I said that she the newspaper probably
gelezen heeft.
read has
I said that she probably read the newspaper.
In all cases in which scrambling is either preferred or dispreferred, the data are not abso-
lute; this is often the case when grammaticality judgements depend on whether an element is
perceived as a topic or not.
(b) Ik zei dat ze waarschijnlijk de krant
I said that she probably the newspaper
gelezen heeft.
read has
Schaeer singles out the class of denites which do not scramble across
sentential adverbs (but do scramble across negation). This class, accord-
ing to Schaeer, includes the sun, the moon, the newspaper, the president,
the mailman, the queen, and proper names such as Susan. (But see de
Hoops : (b).) These referents are available in the discourse with-
out requiring overt mention in the discourse. They can therefore always
be selected as topics, but in view of the fact that they have not been pre-
viously mentioned, they force a topic shift. Schaeer (personal commu-
nication) also agrees that shift topics, selected from a context-specied
set, belong to this class. Examine the following discourses:
() De familie Jansen woont in de Hoofdstraat.
The family Jansen lives in the Hoofdstreet
a Ik heb gisteren de vader ontmoet.
I have yesterday the father met
b ??Ik heb de vader gisteren ontmoet.
I have the father yesterday met
c Ik heb de vader niet ontmoet.
I have the father not met.
d *Ik heb niet de vader ontmoet
I have not the father met
In all cases, the referent for the father is derivable from the introduction of the
family to which he belongs. (a) and (b) show that such topics do not shift
across adverbs. (c) and (d) show that they must shift across negation.
Schaeer concludes from the fact that there are two scrambling posi-
tions that denites, whether they are regular topics or shift topics, always
scramble. The former scrambles to the higher scrambling position and
the latter to the lower one. Schaeers approach is thus along the lines
of Rizzis in that functional projections provide landing sites for fronted
items and each such projection is endowed with the relevant interpreta-
tion. Schaeers distinctions thus enable a view of scrambling which leaves
no room for optionality and which does not lend itself to an OT account.
Schaeers generalizations, however, do not explain why scrambling
across negation should have dierent properties from scrambling over
adverbs. Here the scopal properties of negation might be at stake. Negation
diers from adverbs in that it is associated with focus, therefore, if the object
is to be interpreted as a topic (and here it doesnt matter what type of topic it
is), it cannot stay in the scopal domain of the negative element. The motiva-
tion for scrambling across an adverb must be dierent and seems to be to
remove the topic from the focused VP domain in line with Diesings analy-
sis (see Chapter , section .).
Shift topics, it seems, are tolerated in this
domain (below the adverb, but above negation). One possible reason for
this is that scrambling must be examined in the context of other reorder-
ing operations such as topicalization. If it is the case in Dutch, as in other
Germanic languages, that topicalization applies to switch topics, then switch
topic objects should be topicalized rather than scrambled.
This type of
interaction should be studied further to see if this claim is valid.
How do these conclusions concerning the properties of Dutch scram-
bling fare with the Neeleman and Reinhart approach? As outlined
in Chapter , section ., Reinhart and Neeleman allow for the base
generation of both the scrambled and the unscrambled orders.
assume, following Cinque , that stress is derived from syntactic
structure such that the most deeply embedded element is stressed. The
focus set is derived by listing all constituents which project from this set.
It follows that in Dutch unscrambled OV structures main stress will fall
on the object, which is therefore included in the focus set. In scrambled
structures, the main stress will fall on the verb and the object is no lon-
ger included in the focus set (a). If the object is a topic and it is not
scrambled, it will have to be de-stressed in order to avoid being included
in the focus set as shown in (b).
() a Ik heb het boek gisteren gelezen, en niet verscheurd:
I have the book yesterday read, and not torn-up
b #Ik heb gisteren het boek gelezen, en niet verscheurd
The motivation for scrambling across an adverb in Dutch cant be prosodic, since
contrastive (stressed) topics scramble as do regular unstressed ones (e.g., (b) and (b)
As noted above, section , Frey to appear argues that the initial element in the middle
eld in German feeds topicalization. Since this is also the position of scrambled elements,
the idea that the two positions compete as hosts for topics is supported. German should
be examined to see if indeed the left periphery hosts shift topics and the middle eld hosts
regular topics as predicted.
The terms scrambled and unscrambled order are used to indicate the linear order of the
constituents and do not imply that these orders are derived by movement.
(a), with stress on the object, would be good if the object, not the verb, were contras-
tive (on a par with (b) ). Although Neeleman and Reinhart do not mention such cases,
they could account for this type of case by a stress-shifting operation which assigns stress to
foci not included in the focus set and associate such stress assignment with contrast.
configurations I,
De-stressing the topic, according to Neeleman and Reinhart, is uneco-
nomical, as are all stress-shifting operations. It follows that topics must
scramble. If de Hoop :ooo; :oo, were right and the scrambling of top-
ics is truly optional, then Neeleman and Reinharts account fails, as de
Hoop herself argues. If, however, Schaeer is right, then scrambling is
obligatory and de Hoops argument against Neeleman and Reinhart
fails. Yet how would Neeleman and Reinhart incorporate the ne dis-
tinctions required to account for Schaeers data? First of all, in order
to derive (,oa), Neeleman and Reinhart would have to base generate a
third structure, namely the one in which the object occupies the lower
scrambling position between the adverb and negation. Second, they
would have to add some machinery that would distinguish switch top-
ics and regular topics. Otherwise, since neither belongs in the focus
set, both types of topic would be predicted to occur before the adverb.
Neither of these additions to their framework is desirable and so their
framework fails as well. This leaves Schaeers Rizzi-type approach
which is open to all the same weaknesses Rizzis approach suers
3.4.4 Scrambling in Persian
Karimi :oo, and Roberts :oo, demonstrate that the free word order of
Persian is also determined by IS. The discussion here is limited to their
accounts of the position of direct objects. According to Karimi :oo,: Io,,
the base position of specic objects, marked by the specicity marker
-r, is Spec,VP. He argues at length that non-specic objects dier in
their syntactic, semantic, and morphological properties and therefore
have a dierent base position immediately preceding the verb. The two
orders are derived by the phrase structure rules in (IoI).
(IoI) a [
PP V] ]
b [
PP [
V] ] ]
Karimi goes on to demonstrate that specic direct objects may also scram-
ble above an adverb indicating that it has scrambled to a position external
to the VP. This is illustrated in (Io:) and (Io,) from Karimi :oo,: Io8.
(Io:) Kimea porsid [
ke [ket-ro]
diruz [
Kimea asked that book-r yesterday Sepide
koj gozsht] ] ]
where put-,sg
Lit.: Kimea asked, as for the book, where did Sepide put (it)
Io configurations
(Io,) Kimea porsid [
ke [ket-ro]
diruz [
Sepide [
Kimea asked that book-r yesterday Sepide
be ki dd] ] ]
to whom gave
Lit.: Kimea asked, as for the book, to whom did Sepide give (it)
As can be seen from the literal translation, the scrambled object is inter-
preted as a topic. Scrambling of both specic and non-specic objects is
possible if these are stressed and interpreted contrastively. This is shown
for a non-specic object in (Io) from Karimi :oo,: Io,.
(Io) Kimea [(ye) ketb]
bar -sh t
Kimea (a) book for -her bought
Kimea bought a BOOK for her.
Karimi posits that although the non-specic is new, it can still scramble
when contrasted, since contrast involves discourse linking. As shown in
the previous chapter, contrast falls naturally in the class of topics under
the view that contrast involves a subordinate focus structure, hence the
fact that non-specic contrastive elements can be topicalized is not sur-
prising. Karimi anchors his analysis of scrambling and the two dierent
base positions for specic and non-specic objects by showing the differ-
ent scope and binding properties of the dierent positions. If, however,
these eects are derived directly from the f-structure properties of the
constituents involved as proposed in Erteschik-Shir I,,,,
then it may be
possible to do away with the dierent base positions and oer an account
in terms of topicalization to the left periphery of the VP.
Roberts :oo, goes into more detail concerning the IS properties of
the various word orders in Persian. He argues that Chomskys theories
of Principles and Parameters or Minimalism are typologically inad-
equate for accounting for the languages of the world (,,). His main
reason for rejecting these theories is that they require constituent struc-
ture and hence are inadequate to handle free word order languages
such as Persian. Instead, he proposes an analysis in terms of RRG which
incorporates Lambrechts I,, approach to IS. In order to account for
examples such as (Io) in which a non-specic object has scrambled and
is interpreted contrastively, Roberts :oo,: ,, posits the specialized tem-
plate in (Io,). As shown in Chapter :, section .o, specialized templates
are employed for constructions which are associated with a particular
See also Chapter :, section I., for the analysis of scope in terms of f-structure and
Chapter , section : for the analysis of anaphora as a dependency governed by f-structure.
focus structure. Such specialized templates can be viewed as the RRG
parallel to syntactic top/foc projections in Minimalism.
The dotted lines at the bottom of the diagram indicate the potential
focus domain, in this case the whole structure. The solid lines below the
diagram indicate the actual focus in the sentence. The template in () is
surprisingly reminiscent of Karimis analysis of scrambling illustrated in
()(). Both authors stipulate a specic landing site for the scram-
bled object. In addition, Roberts distinguishes the templates for scrambled
indenites and scrambled denites because the former must be identi-
ed as contrastive foci whereas the latter are contrastive topics under his
analysis. Both Karimis and Robertss analyses suffer from the same weak-
ness in that the mapping from syntax to IS is stipulated. Roberts shows
convincingly that RRG notation allows for all the required mappings, but
doesnt assume any constraints on such mapping. Therefore, although all
languages can be accounted for in this way, it is not clear how the system
imposes constraints on these mappings, predicting that any mapping
whatsoever is possible in natural language. Karimis analysis offers more:
he shows correlations between his mappings and interpretative effects.
It is hard to imagine how these correlations can be made to follow from
specialized RRG templates.
In the next section, we examine object shift in Scandinavian. Object
shift also involves the ordering of objects and adverbials, yet, as we shall
see, it is usually distinguished from scrambling in the other Germanic
languages. It is possible, however, that an analysis of object shift can shed
light on how to approach these types of scrambling as well. We return to
this possibility at the end of the next section.
3.5 Scandinavian object shift
In Scandinavian languages, objects also scramble across adverbials. In the
mainland Scandinavian languages and in Faeroese only pronouns shift,
) = unmarked focus position
but in Icelandic full DPs also shift as long as they are interpreted as topics.
Scandinavian object shift (OS) has been distinguished from scrambling
mainly because OS is restricted by Holmbergs generalization and scram-
bling is not. According to Holmbergs generalization OS depends on the
occurrence of verb movement which positions verbs in Scandinavian in
second position in main clauses. It follows that OS does not occur in
subordinate clauses or in main clauses with auxiliaries in which the aux-
iliary rather than the verb itself moves. The following examples illustrate
the classic cases of object shift in Danish:
() Peter mdte ham ikke.
Peter met him not
Peter didnt meet him.
() a Peter har ikke mdt ham.
Peter has not met him
b at Peter ikke mdte ham.
that Peter not met him
In () both the verb and the object have moved. In the sentences in
(), neither the verb nor the object has moved. In (a), the presence
of the auxiliary blocks verb movement and (b) illustrates the lack of
verb movement in subordinate clauses.
That IS is involved in OS is obvious only in Icelandic which allows full DPs
to shift. It has often been observed that only denites which qualify as topics
undergo OS in Icelandic. In the other Scandinavian languages only pronoun
topics shift, and therefore the reason for their linear position need not follow
from their being topics. Icelandic OS is illustrated in () and ():
() a g les aldrei nyjar bkur.
I read never new books
I never read new books.
b *g les njar bkur aldrei
I read new books never
() a Jn las bkina ekki.
Jon read the books not
b Jn las ekki bkina.
Jon read not the books
(b) shows that non-generic indenites cannot undergo OS and ()
shows that denites do so optionally. Here OS depends on context: it
presupposes recent mention, i.e., topichood.
Holmberg argues that OS applies in a post-spell-out component of
stylistic rules, containing rules dependent on focus structure. Movement
is triggered by the need of foc constituents to be governed by a +foc fea-
ture assigned to the verb and is only licensed when all phonologically visible
non-adjunct material has been removed from between the launching site
and the landing site of the movement. This is how Holmberg links OS to
verb movement. Holmberg derives the linkage between verb movement and
OS from the assignment of an inherent +foc feature to verbs. The restriction
on OS in terms of phonological visibility calls for a phonological account,
yet Holmberg rejects the idea that OS is a phonological or a morphological
operation since it does not make reference to phonological or morphologi-
cal primitives. Chomsky , although tempted by a phonological analysis
of OS, rejects it at least in part precisely because it is driven by the semantic
interpretation of the shifted object (see Chapter , section ). But in view of
the fact that OS cannot apply across a phonologically visible category it must
also be phonological. Chomsky explains OS and the distinction between OS
and non-OS languages by positing the interpretative features INT and INT.
INT is assigned in OS languages only according to ():
() At the phonological border of v*P, XP is assigned INT.
Chomsky doesnt specify what the features INT and INT are. For his
purposes it suces that these features have interpretative import. If the
assignment of INT is to explain OS, then INT must be Focus (and INT
must be topic). Therefore () is equivalent to the idea that OS lan-
guages have a designated focus position at the VP edge. Optionality is
resolved by the economy principle in ():
() Optional operations can apply only if they have an eect on outcome.
Since the optional assignment of the feature INT has an eect on inter-
pretation, it has an eect on outcome, and the optional assignment of the
feature is licensed. Chomskys approach shows how an optional reorder-
ing process with IS-interpretative eects as well as adjacency constraints
can be handled within a minimalist syntactic approach, as long as infor-
mation-structural features are allowed to be assigned in the derivation.
Note that the interpretative eect cannot be dealt with at LF since the
assignment of the INT features depends on phonological boundaries
and therefore must occur post-spell-out.
Numerous papers have been written on OS and its theoretical repercussions. For a
recent collection of articles on the subject, see the special issue of Theoretical Linguistics on
object shift, (/), .
Erteschik-Shir a; b examines OS from a dierent angle:
instead of attempting an account of the placement of the object, she
seeks to understand the placement of adverbs. According to Erteschik-
Shir, Scandinavian adverbs must incorporate prosodically into an avail-
able host. This is also true of pronouns. A potential host for pronoun
incorporation is the verb. This is shown schematically in ():
() a V+Adv DP
b *V+Adv pronoun
c V+pronoun+Adv
(a) shows the order with the verb linearized in second position. The
adverb incorporates into the verbal host. The full DP object does not
incorporate. In (b) the adverb incorporates into the verb and the
pronoun is no longer able to incorporate as required. That is why this
sequence is ruled out. In (c), however, the pronoun incorporates rst.
The verb+pronoun sequence, probably due to the phonological weakness
of the incorporated pronoun, can still serve as a host for incorporation,
allowing the adverb to incorporate. In this order both the adverb and the
pronoun are incorporated as required.
When the verb remains in situ,
the pronoun is adjacent to the verb and can incorporate without the need
for a dierent linear order. This explains Holmbergs generalization.
This story does not automatically carry over to Icelandic, since it is
not at all obvious how the placement of full DPs can receive a prosodic
account. There is, in fact, no evidence that full DPs in Icelandic incor-
porate into the verb. However, the adverb and the object seem to form a
prosodic unit in both orders in (). I therefore assume that Icelandic
adverbs, like all Scandinavian adverbs, are weak and must prosodically
incorporate (PI). Yet whereas Danish adverbs PI into the verb, Icelandic
adverbs PI into the object (either pro- or enclitically) in these cases:
() a bkina+ekki
books-the not
b ekki+bkina
As expected heavy noun phrases do not shift:
() a g s ekki rherrann sem allir eru a kvarta yr.
I saw not minister-the who all are to complain over
I didnt see the minister who everyone was complaining about.
There is more to this than revealed here. See Erteschik-Shir a for details.
It is in fact hard to detect a difference in stress between the two. This may be due to the
fact that phrasal stress is nal.
b ??g s rherrann sem allir eru a kvarta yr ekki.
I saw minister-the who all are to complain over not
This follows naturally from an account in terms of prosodic
Adverb placement and prosodic incorporation occur in the computa-
tion to PF. Since the interpretative eects follow from the interpretation
of f-structure and do not directly depend on the linear order of the elem-
ents, this approach is not linked to a particular architecture. Under this
view OS is not optional. Full DP topics must shift. Non-topics cannot
But why must topics shift in Icelandic? Erteschik-Shir a
claims that shifting the object identies it as a topic. The ID constraint as
dened in the next chapter is concerned with the identication of the syn-
tactic roles of arguments as well as the identication of topic and focus.
The identication of foci is either through intonation or through linear
position or both; the identication of topics is either through fronting or
through de-stressing. Several factors interact to determine which of these
tactics are employed in each particular case. Word order can be used to
identify topics and foci if it is not also used to identify syntactic category.
A language such as Icelandic which has morphological case is therefore
free to use word order for IS purposes and hence allows scrambling of
full DPs where other caseless Scandinavian languages do not. Another
factor is whether the type of topic that needs identication is prosodi-
cally uniform. If regular topics and contrastive topics form a class, then
this class cannot be prosodically identied.
In Scandinavian languages, an element fronted to the initial posi-
tion in the sentence is identied as the main (shift) topic of the sentence
as discussed above. An element fronted within the VP is interpreted
as a subordinate topic. Another way to mark subordinate object top-
ics is by not stressing them. This is the strategy used in the Mainland
Scandinavian languages and in English. The fact that Icelandic can iden-
tify subordinate topics by movement is due to the fact that it has overt
morphological case marking, not available in Mainland Scandinavian.
There may therefore indeed be a connection between the potential for
ID-ing topics and foci by word order and the availability of overt case as
commonly assumed.
In certain dialects of the Mainland Scandinavian languages OS of pronouns is
optional. For a phonological account of this type of optionality, see Erteschik-Shir a:
For an explanation of why Faeroese, which exhibits overt morphological case, does not
allow OS of full DPs, see Erteschik-Shir a: .
In other Germanic languages such as German and Dutch, (subordi-
nate) topics undergo scrambling. Scrambling has much in common with
Scandinavian OS, yet it occurs independently of verb movement. Neeleman
and Reinhart predict that VO languages do not allow scrambling.
They argue that case can be checked either in prosodic (requiring adja-
cency) or in syntactic domains. Preference is given to prosodic checking
since this allows the derivation to reach the articulatory system sooner.
The VO order, but not OV order, allows prosodic case checking which
therefore must take place. VO languages will therefore not allow scram-
bling to check case. It follows, under this view, that scrambling is restricted
to OV languages such as Dutch and German and that Scandinavian OS
cannot be a subcase of scrambling.
If OS is in essence a matter of PF lin-
earization dependent on language-specic incorporation processes, then
the dierence between OS and scrambling may in fact boil down to di-
erences in prosody which in turn may be connected to the dierences in
basic word order. An examination of the prosodic properties of scrambling
along the lines outlined here for OS should therefore be fruitful.
3.6 Concluding remarks on word order
In spite of the growing body of work on IS and word order, this research
is only at its initial stages. Much of the syntactic research suers from
insucient attention to the signicance of IS features and focuses more
on the repercussions for syntactic theory.
Even so, some supercial tendencies can be observed. That topical
information precedes focal information has been known at least since
the days of the Prague School. The leftmost peripheral position is there-
fore often dedicated to the topic. The particular type of topic found on
the left periphery may dier across languages and also within a language
if it has several types of topicalized constructions, i.e., with and with-
out resumption. In addition, certain types of languages also scramble
objects, which function as secondary topics to the left, to a position at
the VP boundary.
If, as suggested speculatively in section ., it turns
out that topicalization bleeds scrambling, i.e., a switch topic object
topicalizes and therefore is not available for scrambling, then the syntac-
tic theory which accommodates this sort of relationship will have very
Wagner argues that scrambling and OS can in fact be seen as one and the same
In Chapter , section . the order of multiple objects within the VP will also be shown
to be determined to some extent by information structure.
dierent properties from the currently available ones. One likely option
is that both scrambling and topicalization involve PF linearization and
that linearization takes into account the complete string and does not
position each element separately. We have also observed that leftward
movement for IS purposes need not be motivated by the properties of
the moved element itself; the motivation may be more global, resulting
in a linear order in which a focus is positioned to the right. This type of
process is not amenable to the cartographic approach since movement
is not triggered by a feature of the moved element. This is yet another
reason to reject a purely syntactic account of those aspects of word order
which are IS motivated. More promising avenues of investigation are
those that involve PF processes. IS-determined word order involves lin-
ear order and not hierarchical order; it also interacts with prosody and
prosodic boundaries rather than syntactic boundaries. Cross-linguistic
dierences may have to do at least in part with language-specic pro-
sodic properties. If, for example, the analysis of object shift proposed
in Erteschik-Shir a is on the right track, then variation in the
prosody of adverbs within and across languages may provide at least a
partial account of their placement within the sentence. Another exam-
ple might be provided by the designated focus position in Hungarian,
which is preverbal. Non-contrastive verbs are prosodically weak. A
stressed element will therefore be more prominent if it is placed next
to the verb. This explanation could account for those languages in
which the designated focus position is adjacent to the verb, yet it does
not suce to explain why in some of these languages the focus precedes
the verb and in others, it follows it. In Hungarian word stress is word ini-
tial. If we assume that the focus+verb is pronounced as a prosodic unit,
then it follows that if the focus were to appear after the verb, initial stress
would not be possible and the sequence would not be pronounceable.
This is a potentially interesting explanation for why the designated focus
position is preverbal in Hungarian, an explanation which may extend to
other languages with a focus position adjacent to the verb.
All these factors point to the need for a phonological account.
Research that examines the ISPF interface and which does not neglect
IS interpretation may not only advance our understanding of word
order as such, it may also make more precise the role played by syntax in
This idea originates with Szendri who argues for stress-driven focus move-
ment. See also Costas comments on this approach in Szendri and Costa which
raises important questions relevant to the syntaxprosodyIS interfaces.
Information Structure Constraints
The purpose of this chapter is to show that the processing of informa-
tion structure plays an important role with respect to constraints which
have traditionally been viewed as syntactic constraints. The rst section
demonstrates that word order in simple sentences is constrained by
the need to avoid an ambiguous parse of the linear string. This is dem-
onstrated for Danish and Hebrew. The second section deals with con-
straints on extraction including the sentential subject constraint and the
that-t eect and argues that dependencies can be processed only within
canonical f-structures. The third section shows that the fact that canoni-
cal f-structure is parameterized across languages aords an explana-
tion for the dierence in superiority eects in dierent languages. The
languages discussed in this section are English, Hebrew, German, and
Danish. Section shows that the constraint on extraction introduced in
section not only accounts for extraction from complex NPs, it also pre-
dicts extraposition from NP. Section discusses evidence for an account
of extraction constraints in terms of processing.
4.1 Identication
4.1.1 Argument identication, a constraint on topicalization
In Chapter , section , it was demonstrated that Danish topicalization
applies optionally to topics with a preference for switch topics. Danish
topicalization, however, is further constrained.
The examples in ()
show that even denites, which qualify as topics, are sometimes ruled
out (a) or less than optimal (b).
() a *Marie/Pigen mdte Peter igr.
Marie/the girl met Peter yesterday
Peter met Marie/the girl yesterday.
The material in this section is to a large extent drawn from Erteschik-Shir a.
information structure constraints
b ?Marie/Pigen mdte jeg igr.
Marie/the girl met I yesterday
I met Marie/the girl yesterday.
Due to V-, topicalized sentences in Danish consist of a DP V DP string
just as non-topicalized transitive sentences do. As a consequence, the
arguments are not identied as subjects or objects by word order and are
potentially ambiguous. (a) would be ambiguous if the topicalized read-
ing were acceptable. Ruling out the topicalized reading thus prevents
ambiguity. In (b) the subject is a case-marked (nominative) subject
pronoun. In view of the fact that the subject is identied by case mark-
ing, the sentence is not ambiguous. When an accusative object pronoun
is fronted, the result is also unambiguous and therefore acceptable:
() Hende mdte Peter/jeg igr.
Her met Peter/I yesterday
Peter met her yesterday.
Dislocation to the left periphery in Danish is thus not only restricted to con-
stituents that qualify as topics, the result must also lead to an unambiguous
parse. Note that the ambiguity restriction cannot be dened in terms of the
fronted element only, in view of the fact that (b) is only slightly degraded.
An analysis restricting fronting to overtly case-marked elements is therefore
doomed to failure.
Further evidence that parsing is at stake is the fact that
topicalization is unrestricted with verbs which select one animate and one
inanimate argument as in ().
() a Den skue har drengen malet.
that drawer has boy-the painted
b Den bog har drengen ogs lst.
that book has boy-the also read
Here the inanimate fronted topics can only be parsed as objects. Context
can also disambiguate in cases which would otherwise be ambiguous:
() Hvem var det Sarah mdte, Peter eller Thomas?
Who was it Sarah met, Peter or Thomas?
*Peter mdte Sarah.
Peter met Sarah
Sarah met Peter.
Surprisingly, even though neither is ambiguous, (b) is slightly degraded compared with ().
This could be because the correct parse is signaled earlier in the latter than in the former.
information structure constraints
In (), the context has been set up to induce the reading with Sarah
interpreted as a subject: The context forces a contrastive reading of the
topicalized object and still allows for a topic reading of the subject. Yet
Danish informants reject the sentence in this context because they still
want to assign the reading in which Peter is the subject.
I propose a parsing constraint on the identication of arguments (ID)
in (). Parsing constraints are most naturally viewed as constraints on the
interface with the articulatory-perceptual system, i.e., as PF constraints.
() ID:
In a string, X V Y, identify X as subject and Y as object if neither is
marked otherwise.
ID would identify the topicalized object in (a) as a subject and the post-
verbal argument as an object, indeed the only interpretation of the string.
In (b) the postverbal argument is identied as a subject, allowing the pre-
verbal argument to be interpreted as an object by ID. Similarly, in () the
initial element is accusative, identifying it as the object. In (), the elements
are identied by the argument structure of the verb, and in () identica-
tion is accomplished contextually, yet such identication is not possible in
Danish. The same sequence in Norwegian would be ne. Danish is partic-
ularly rigid in this respect. A disambiguating context can serve to identify
subjects and objects more easily in Swedish and Norwegian.
The dierent
sensitivity to the ID constraint among the Scandinavian languages may be
due to the dierences in the distribution of pronominal case morphology
in these languages.
Raviv argues that ambiguity is prevalent among
Norwegian pronouns. Pronouns therefore do not play a major role in
Note that the contrastive stress on the topic does not function to disambiguate the sen-
tence since the same intonation would be used if Peter were a contrastive subject.
That languages may differ as to whether interpretation depends on context or on word
order is also clear from de Hoops observations mentioned in the previous chapter, section ..
See Erteschik-Shir b and Raviv for arguments to this effect.
Ambiguous wh-questions are not ruled out, even in Danish:
(i) Hvem mdte Sarah?
Who met Sarah
Who did Sarah meet?/Who met Sarah?
Wh-questions are processed differently from declaratives with respect to ID: The initial
wh-phrase triggers a search for a trace and is therefore not automatically identied as a sub-
ject. Further, since wh-questions never occur out of the blue, their interpretation is neces-
sarily linked to a disambiguating context. Finally, ambiguous wh-phrases are rarely found in
Danish. They are replaced by circumlocutions such as that in (ii):
(ii) Hvem er det (som) Sarah mdte? /Hvem er det som/der mdte Sarah?
Who is it (that) Sarah met? /Who is it that met Sarah?
information structure constraints
disambiguating DP V DP strings in Norwegian. Without context lending
a helping hand, topicalization would be blocked by ID in most cases in
Norwegian. That is probably the reason why context is licensed as an iden-
tier in Norwegian, but not in Danish, where the pronoun system serves
well to disambiguate these strings.
Topicalization in Hebrew exhibits similar constraints to the ones found
in Danish and these data are also accounted for by the ID constraint. There
are two types of Hebrew topicalization, one in which the verb is fronted to
V- position and one in which it is not. (Hebrew preverbal subjects are
topics; postverbal subjects are foci.) Ambiguity only results with V- in
which the two arguments are separated by a V rendering a potentially
ambiguous string. In this word order the fronted element is interpreted
contrastively. () shows that fronting a denite is perfectly acceptable.
() et haYELED paga hai.
the-boy met the-man
The man met the boy.
Denite objects in Hebrew are marked by et (the exact nature of this
elem ent is irrelevant here), and indenite subjects lack this marking, so
if one of the two arguments is denite the sentence is unambiguous. In
(), the subjects are denite. Since they lack marking by et, they cannot
be interpreted as objects.
() a ?YELED paga hai.
boy met
The man met a boy.
b YELED paga haia.
boy met
The woman met a boy.
Although neither is ambiguous, (b) may be slightly better than (a).
This is due to the feminine subject-agreement marking on the verb
which identies the postverbal element as a subject. The reason (a) is
slightly degraded may be that there is no overt marking that identies
the arguments. It is only the lack of et marking on the second argu-
ment which forces it to be interpreted as a subject, and this missing
marking comes rather late in the sentence (compare (b) ).
According to Rahkonen , who did a corpus analysis in Swedish, there were few NP V NP
sequences in the corpus intended as OVS structures to which the reader was tempted to assign
an SVO word order in their actual context. Rahkonen concluded that it is a necessary condition
for the use of OVS in Swedish that it not be misunderstood. This accords nicely with our obser-
vations concerning the same facts in Danish.
information structure constraints
When both the subject and the object are indenite, the result is never perfect:
() a *YELED paga i, lo yalda.
boy met
man, not girl
A man met a boy, not a girl.
b ??YELED paga ia, lo baxur.
boy met
woman, not youth
A woman met a boy, not a youth.
(a) violates ID. (b) does not, due to the feminine agreement morphol-
ogy on the verb which identies the woman as the subject. The reason
(b) is degraded may be that the topicalized argument is indenite and
hard to contextualize contrastively.
() a I naax kelev.
man bit dog
A dog bit a man.
b ??KELEV naax i
dog bit man
A man bit a dog.
(a) is interpreted easily with the postverbal argument as the subject in
view of the fact that dogs normally bite people and not the other way
around. For the same reason (b) is odd. (b) is, of course, improved in a
context in which a man is contrasted with a dog as being the biter.
We can conclude that overt morphological marking on some element
of the sentence identies displaced DPs. This marking need not appear
on the displaced element itself but could appear on the verb or on the
element which remains in situ. The constraint must therefore be formu-
lated in such a way that disambiguating overt marking on any element
of the sentence qualies to ID the displaced element. The fact that the
relevant feature is not necessarily marked on the dislocated element, as
well as the fact that overt morphological marking is visible only at PF, is
further evidence for viewing the ID constraint on IS-motivated disloca-
tion as a parsing constraint at PF.
Languages dier as to how rigidly the ID constraint applies. Danish is
extremely rigid in this respect: The SVO reading of an ambiguous string
is strongly preferred and contextual clues can only overrule this interpre-
tation with diculty. In other languages context can more easily come
into play. More research into the role of the ID constraint in a variety of
languages is needed to decide what additional factors are responsible for
these dierences among languages and whether focus displacement is
also subject to this constraint.
information structure constraints
4.1.2 Identication of IS functions
Yet another aspect of the sentence which requires identication is the
information structure of the sentence, i.e., topic and focus. These, as
we have seen, are marked positionally, morphologically, and/or by into-
nation. In the previous chapter we saw that most accounts of IS-moti-
vated word order are syntactic. Positional marking is thus singled out,
leaving morphological and intonational marking to PF. We also saw that
syntactic accounts suer from serious theoretical and empirical weak-
nesses. A natural conclusion is therefore to view all IS marking, includ-
ing positional marking, as PF operations. This approach also provides a
natural account of the optionality of IS-motivated word order. The fact
that IS-motivated dislocations are sensitive to the ID constraint further
strengthens the argument for their analysis as occurring at PF.
The ID constraint reects that parsing of subject and object functions
is facilitated by overt morphological identication of these functions.
Parsing would be facilitated even further if no dislocation were to
take place. Dislocation must therefore also serve some function. Both
here and in the previous chapter, it was shown that particular linear
positions serve to identify a variety of IS functions cross-linguistically.
In Chapter , we showed that IS functions are sometimes marked
morphologically and that focus is often marked intonationally.
Discourse-congurational languages favor the identication of IS
functions by word order. If the ID constraint is observed, then these
languages must have sucient morphological marking for the various
dislocations to adhere to the ID constraint. Identifying IS roles is thus
as important as identifying thematic roles. The tension created by the
need to identify syntactic roles as well as identifying IS functions is thus
mediated by the ID constraint.
4.2 IS constraints on syntax
4.2.1 I(denticational)-dependencies
On the basis of data of the type shown in (), I argued in Erteschik-Shir
that extraction is completely determined by IS constraints:
() a Who did John say that he had seen?
b ?Who did John mumble that he had seen?
c *Who did John lisp that he had seen?
There it was argued that only focus domains are transparent for purposes
of extraction. The intuition behind this idea was that potential focus
Ioo information structure constraints
domains are processed dierently from non-focus domains in that gaps
are only visible in the former.
Whether a particular subordinate class
provides a focus and therefore allows extraction depends, among other
factors, on the clause under which it is embedded. It also depends on
the context in which the sentence is uttered. (Iob) is improved in a context in
which mumbling has been mentioned (e.g., following At our meetings
everyone always mumbles). (Ioc) is acceptable in a context in which it
is known that John lisps. This is because such a context enables the main
verb to be defocused and consequently enables the subordinate that-
clause to be focused. If intuitions are elicited out of context, judgements
for sentences of this kind will depend on whatever context the informant
happens to come up with. The examples in (II) illustrate other kinds of
contextual factors that interact with focus assignment and the concomitant
acceptability judgements.
(II) a ??What did the paper editorialize that the minister had done?
b *What did you animadvert that he had done?
(IIa) would sound much better if uttered by a member of an editorial
board, and (IIb) probably cant be contextually improved due to the fact
that highly infrequent items such as animadvert are necessarily focused.
Contrastive contexts also interact with extraction judgements:
(I:) a ?Who did John SAY that he had seen? [=contrastive]
b Who did JOHN say that he had seen? [=contrastive]
Contrastive focus on the main verb, as in (I:a), or on another constituent
of the main clause, as in (I:b), does not preclude focus on the subordi-
nate clause. Therefore these sentences are ne with contrastive interpret-
ations. The reason (I:a) is slightly more degraded than (I:b) is because it
is harder to construe a likely contrast set for the verb say.
A number of different syntactic solutions have been suggested over the
years to account for such squishes in grammaticality. This type of solution,
however, does not explain the contextual eects. Speakers judgements
with respect to data of the kind illustrated in (Io)(I:) are rarely stable:
Dierences are found across speakers and sometimes the responses of the
same speaker change. This type of instability occurs whenever grammati-
cality is context dependent because the judgements in such cases are also
See Erteschik-Shir and Lappin I,8,a: 8, for this proposal which also provides an expla-
nation in terms of processing for parasitic gaps and for why resumptive pronouns salvage
islands. For a different pragmatic/processing account of resumptive pronouns see Cann,
Kaplan, and Kempson :oo,.
information structure constraints IoI
context dependent. No syntactic account of data of the type in (Io)(I:)
will be able to predict the contexts which improve acceptability. Syntactic
constraints will therefore always fail empirically.
The idea that extraction depends on focusability is also to be found
within RRG in Van Valin I,,,b; I,,,a; Van Valin and LaPolla I,,,. The
connection between wh-questioning and focusability is obvious under
their view. They explain the connection between topicalization and rela-
tivization and focusability by means of the following aboutness constraint
derived from Kuno I,,o; I,8,:
(I,) Pragmatic aboutness condition on topicalization and relativization
The sentence fragment following a topical element in the precore
slot or a restrictive relative clause must be pragmatically interpre-
table as being about the precore slot element or the head noun.
(Van Valin and LaPolla I,,,: o:,)
Van Valin and LaPolla derive from the denition of topics in Reinhart I,8I
the idea that in order for a sentence to be about an element (the topic),
there must be an alternative form of the sentence in which the topic func-
tions as a focus, i.e., it must be within the focus domain of the sentence.
Van Valin and LaPolla make a distinction between the potential focus
domain and the actual focus domain. The potential focus domain is
syntactically determined by language-specic principles. In English, for
example, the whole sentence is a potential focus domain and the actual
focus can fall anywhere in the sentence. In Italian, French, Chinese,
and Sesotho, however, the subject is excluded from the potential focus
domain (Van Valin and LaPolla I,,,: :I:). One would indeed want some
way of distinguishing a language such as Italian in which a preverbal
subject cannot be focused, and a language such as English in which it
can. One way of doing this is to dene dierent potential focus domains
for the two languages. Another way would be to argue for dierent lin-
earization patterns in the two languages such that Italian (subject) foci
are necessarily linearized at the right periphery. Unless the focusability
facts can be linked to other properties found in the languages in ques-
tion, both these accounts remain stipulative.
There is a special constraint on the potential focus domain in complex
sentences. This is stated in (I) (Van Valin and LaPolla I,,,: 8,, origi-
nally in Van Valin I,,,b).
For discussion concerning the aboutness constraint versus a constraint in terms of
focusability see Erteschik-Shir and Lappin I,,,; I,8,a; Grosu I,8:.
Io: information structure constraints
(I) A subordinate clause may be within the potential focus domain if
it is a direct daughter of (a direct daughter of ) the clause node
which is modied by the illocutionary force operator.
In some languages, subordinate clauses are never included in the potential
focus domain. In such languages (I) is irrelevant. In view of the fact that
relative clauses are not modied by the illocutionary force operator, they
will never allow extraction. This works very well for Lakhota which does
not have wh-movement and in which the same morpheme (tku) is used
for wh-words and specic indenite pronouns. This, in general, leads to
ambiguity, but not inside relative clauses where only the specic-indenite
reading occurs. Relative clauses, according to Van Valin and LaPolla I,,,:
,I:, are necessarily excluded from the potential focus domain. This is
due to the fact that relative clauses are dominated by an NP and therefore
there is no relationship between the embedded clause and the clause node
modied by the illocutionary force operator (the main clause). Since the
interpretation of the pronoun as a wh-word is ruled out outside of potential
focus domains, only the specic-indenite interpretation of the pronoun
is licensed. Note that it is the interpretation of the question rather than wh-
extraction which is blocked. It follows that wh-interpretation in general, and
not only wh-extraction, is restricted to potential focus domains.
Van Valin and LaPolla are aware of the fact that some relative clauses
allow extraction in Danish as pointed out in Erteschik-Shir I,,,:
(I,) a Det er der mange der kan lide.
that are there many that can like
There are many who like that.
b Det hus kender jeg en mand som har kbt.
that house know I a man that has bought
I know a man who has bought that house.
c Den stil har jeg nogle elever der har skrevet.
that composition have I some pupils that have written
I have some pupils that have written that composition.
In these relative clauses, the matrix is relatively empty, allowing the relative
clause itself to be the focus of the utterance. Such extraction is not altogether
bad in English either as shown in (Io), also from Erteschik-Shir I,,,.
There is no limit on the number of direct daughters.
See also Jensen :ooo for arguments that this type of data is incompatible with an analysis
within RRG, and Van Valin :oo,: :,o admits that pragmatic functions and lexical semantics
are at stake.
information structure constraints Io,
(Io) This is the kind of weather that there are many people who like.
Van Valin and LaPolla I,,,: o,I admit that these facts, as well as data
of the type illustrated in (Io)(I:), indicate that the constraints on wh-
question formation cannot be purely structural, but must also involve
pragmatic functions, and lexical semantics. Lexical semantics is involved
in extraction with verbs of manner-of-speaking in (Io) (see section
:., below), and context was seen to be involved in the acceptability of
(Ioc) and (IIa), above. Since lexical and contextual factors can override
the syntactic denition of potential focus domains in (I), and since
these factors are determined by focusability, it is not obvious whether
a syntactic constraint is needed in addition to the IS constraint on
extraction. Arguments for this syntactic constraint are supplied by cross-
linguistic variation. Van Valin :oo, proposes that the constraint in (I)
is relevant only in those languages in which the potential focus domain
extends beyond the main clause. German is an example of a language
where extraction out of nite subordinate clauses is all but impossible.
This would be predicted if the potential focus domain in German were
limited to main clauses. One might therefore want to parameterize (I)
to distinguish languages such as English and German since English
allows for much freer extraction out of that-clauses than German
does. Yet some dialects of German allow extraction out of a subset of
those that-clauses which allow extraction in English. Danish is even
freer than English in that extraction in English is possible in a subset
of those cases in which extraction is possible in Danish but both these
languages would be restricted by (I) in the same way. Having some
languages governed by (I), but not others, is therefore not sucient to
explain the cross-linguistic data. One possible explanation for the fact
that extraction in English is possible in the same cases as in Danish, but
the results are comparatively degraded, is the fact that topicalization
is prevalent in Danish but not in English, as argued in Erteschik-Shir
I,8:. This explanation, however, does not carry over to German where
topicalization is as common as in Danish, yet extraction is highly
constrained. There is, however, another dierence between German and
Danish. German allows scrambling in subordinate clauses and thus has
a way to mark topics in subordinate clauses without extraction. One
possible venue for future research into the factors which determine the
range of extraction in a particular language may therefore be the various
mechanisms the language has for marking topics. The restriction
that extraction requires focusability seems to hold across languages,
Io information structure constraints
In Chapter :, section :.o, we introduced the lie-test which tests clauses
for whether they are focusable or not. This test distinguishes those that-
clauses embedded under manner-of-speaking verbs which do allow
extraction from those that do not. It also allows for in-between cases.
Consider, for example, the lie-test in (I,):
(I,) A: John said that he had seen Peter.
B: a Thats a lie, he didnt.
b Thats a lie, he hadnt.
(I8) A: John mumbled that he had seen Peter.
B: a Thats a lie, he didnt.
b ?Thats a lie, he hadnt.
(I,) A: John lisped that he had seen Peter.
B: a Thats a lie, he didnt.
b *Thats a lie, he hadnt.
The test also does a fairly good job distinguishing those relative clauses
which allow extraction in Danish from those that do not. Yet the lie-test
is problematic since speakers tend to lose their intuitions after a very few
test sentences. In addition, it is almost impossible to control for contextual
eects which, as shown above, make a big dierence. The correlation among
speakers and within speakers between the results of the lie-test and the
extraction facts, although quite good, is therefore not perfect. Van Valin and
LaPolla provide a similar procedure which tests constituents for focusability
by negating them, rather than testing the clause in which they are embedded.
The idea behind both tests is the same, and they make the same predictions,
namely that focusability is dependent on lexical elements and context. Only
a few structures are immune to focusability and extraction. One of these
(sentential subjects) is discussed in section :., below. Another test can be
derived from the observation in Morgan I,,, that fragment replies to
utterances cannot be used if the corresponding sequence in the utterance
is an island. (:o) is Morgans illustration of the ungrammaticality of a
fragment reply when the original utterance includes a relative clause.
(:o) a Did the man who Tricia red leave town?
b *No, Thelma.
c No, the man who Thelma red (left town).
If indeed fragment replies must be focused, then Morgans observation
leads to a test of focusability. The fact that fragment replies are unac-
ceptable inside islands also provides further evidence for the correlation
between focusability and islandhood. Goldberg :ooo provides a variation
information structure constraints
to Morgans observation and demonstrates that such answers cannot be
expressed within islands including relative clauses and complements of
manner-of-speaking verbs as follows:
() Why was Laura so happy?
A: #The woman who thought she was dating someone new lives
next door.
A: the woman who lives next door thought she was dating
someone new.
() A: #John shouted that she was dating someone new.
A: John said she was dating someone new.
Goldbergs adaptation of the test clearly indicates that the correlation
between the answers and extraction is robust. Since answers must be
focused, her test can be used to diagnose focusability.
It thus follows from both approaches that only focus domains
are transparent for purposes of extraction. In Erteschik-Shir I
argue that this constraint on extraction falls under a more general
constraint on Identicational dependencies or I-dependencies, which
include anaphora, wh-trace dependencies, multiple wh-dependencies,
negation and focus of negation, and copular sentences. What all these
dependencies have in common is that the dependent is identied in
the construction, either by its antecedent or by an operator.
The need
for the dependent to be identied contributes to the processing load of
the sentence.
In order to ease this load, dependencies are restricted to
canonical f-structures. These are dened in the next section.
Goldberg herself argues that backgrounded constructions are islands. Since backgrounded
constituents are dened as constituents that do not correspond either to the primary topic
or to part of the potential focus domain (equivalent to Vallduvs Tail), this constraint covers
both the requirement that extraction occur out of focusable constituents as well as the fact that
subjects, which, according to Goldberg, are the primary topic, can be extracted. Although the
whole subject is not backgrounded, elements within it are, and can therefore not be extracted.
The main difference between Goldbergs approach and one that relies only on focusability
concerns the status of subjects. Although subjects in the unmarked case do function as topics,
they also can be focused. Hence extraction of subjects is predicted by both approaches.
In the discussion of Lakhota above, it was noted that question interpretation is restricted
to focus domains even though the wh-expressions do not move. The constraint on I-depen-
dencies is formulated so that it pertains both to wh-t dependencies and to dependencies
between an operator and its dependent. In Lakhota pronominal forms depend on a question
operator (instantiated by an overt morpheme) in order to be interpreted as question words.
Hence, the fact that question interpretation is subject to constraint on I-dependencies is
predicted. Neither account (Van Valin, Erteschik-Shir) requires covert movement in LF to
account for island effects in languages without wh-movement.
See section for discussion.
Ioo information structure constraints
4.2.2 Canonical f-structures
Not all f-structure assignments are equally good. (:,) illustrates a well-
known asymmetry: objects are harder to interpret as topics than subjects
(in languages with xed word order and no morphological marking of
(:,) Tell me about John:
a He is in love with Mary.
b ??Mary is in love with him.
In view of the fact that this constraint gures prominently in lan-
guages such as English which have xed word order, I propose that the
reason for this asymmetry is that in such languages there is a preference
for aligning f-structure with syntactic structure. There are two canonical
f-structures in English:
(:) Canonical f-structures (English):
[ X ]
b sTOP
[ X ]
In other words, an unmarked f-structure is one in which syntactic
structure is isomorphic with f-structure: Either the subject is the topic
and the VP is the focus (or the focus is contained in the VP) as in (:a)
or else there is a stage topic and the remaining sentence is the focus (or
the focus is contained in it).
It follows that a marked f-structure is
one in which an object is the topic. Marked f-structures in English are
thus characterized as having f-structures which do not match syntactic
Isomorphism between syntax and IS is a requirement on the process-
ing of dependencies. This was argued in Erteschik-Shir I,,,. The intuition
behind this idea is that the processing load of a marked f-structure together
with the processing of the dependency leads to processing overload. This
not only predicts many well-known constraints on dependencies, it also
predicts the fact that these constraints are often context dependent.
4.2.3 The constraint on I-dependencies
I-dependencies are therefore restricted to canonical f-structures as shown
in (:,).
See, among others, Andersen I,,I; Lambrecht I,,; Li and Thompson I,,o; Reinhart I,8I.
The idea of a canonical f-structure which presupposes isomorphism between syntactic
structure and f-structure is similar to the idea of shape conservation within Representation
Theory (Williams :oo,).
information structure constraints Io,
[ X ]
The intuition behind this constraint is that dependents must be identi-
ed and that a canonical f-structure, in which IS functions and syntactic
functions are unied, enables the processing of this identication.
the case of wh-traces, for example, the trace must be identied with the
fronted wh-phrase. The proposed constraint restricts such identication
to canonical f-structures. The constraint is thus couched in processing
terms in which f-structure plays a critical role.
Let us rst examine how the constraint applies to the extraction facts
in (Io)(I:). In Erteschik-Shir I,,,, verbs that allowed focus on the
subordinate clause and therefore allowed extraction were called bridge
verbs because they allowed movement out of an island. This term has
caused some confusion since the bridge property was understood on a
par with a feature and therefore verbs could be associated with bridgeness
or not. This contradicts the spirit of the original insight which was that
whether or not a matrix verb requires focus is highly context dependent.
Still some verbs never require a context in order for extraction to occur
from their complement, and other verbs do, albeit to various extents. A
lexical analysis is therefore called for. Such an analysis, based on Erteschik-
Shir and Rapoport in preparation, is oered in Chapter ,, section I. For
now we will assume that there is a lexical property of manner-of-speaking
verbs which attracts focus, but that the same property can be defocused
contextually, enabling focus on the subordinate clause, which then meets
the requirement on extraction, since, according to the constraint on I-
dependencies, the dependent (the trace) must be contained in the focus
domain. It follows that, out of context, only that-clauses under say allow
extraction. All the other manner-of-speaking verbs require some sort of
contextualization in order for the verb to be defocused, thus allowing the
The canonical f-structure proposed in (:a) is in fact the same as the predicate focus
structure in Lambrecht I,,; :ooo; and adopted in RRG. The predicate focus structure is
universally the unmarked focus type and is dened as follows (Lambrecht :ooo: oI,):
(i) Predicate focus structure: Sentence construction expressing a pragmatically struc-
tured proposition in which the subject is a topic (hence within the presupposition)
and in which the predicate expresses new information about this topic. The focus
domain is the predicate phrase (or part of it).
Van Valin and LaPolla I,,,: ::,,I show that the analysis of anaphora relies on this congu-
ration. This analysis of anaphora and the one proposed in Erteschik-Shir I,,, which relies
on canonical f-structure are very similar.
Io8 information structure constraints
subordinate clause to be focused. Extraction is judged acceptable in these
cases to the extent that the context enables such a focus assignment.
According to Van Valin and LaPolla I,,,: o,o the fact that manner-of-
speaking verbs block extraction out of their complements follows from
Grices I,,, Cooperative Principle: the choice of a verb which highlights
the way in which something is said rather than what is said causes the
focus to shift to the verb in the main clause, because of the maxim of
relevance: the speakers choice of an informationally richer expression
(murmur) over another more neutral possibility (say) only makes sense
in terms of the Cooperative Principle if the manner of expression is in
fact highly relevant to the main point of the utterance. Hence the focus
must fall on the matrix verb, keeping the complement from being the
actual focus domain.
4.2.4 Wh-topics
Extractability does not only depend on the content of the matrix clause,
it also depends on the type of wh-phrase extracted. Cinque demon-
strates that extracted phrases which can be interpreted as being d-linked
(restrictive in our terms, see Chapter :, section ,.,) render superior
results to those in which the extracted phrases cannot be interpreted in
this way. Compare (:o) and (:,).
(:o) a ?Which book did you wonder whether John bought?
b ??Which place did you wonder what John put?
(:,) a ??What did you wonder whether John bought?
b ?*Where did you wonder what John put?
D-linking depends on whether a contextual referent for the wh-phrase is
available. Hence in a context in which a set of relevant books (:oa) or a
set of relevant places (:ob) is available, extraction is improved over that
of (Io) in which the wh-phrases do not require a restrictive context and
hence extraction renders worse results. Cinque builds this notion of
referentiality into his syntax and predicts that when the context provides
the required referent, the extraction should be perfectly acceptable. What
is missing in Cinques approach is an explanation of why referentiality
should interact with syntactic constraints, such as subjacency, in this way.
Such an explanation is available once the IS signicance between restrictive
and non-restrictive wh-phrases is observed: Restrictive wh-phrases range
See Featherston :oo: for experimental evidence that the quality of being a bridge verb is
the same quality which permits a verb to take a verb-second complement clause in German.
information structure constraints Io,
over a context-specied set. Such a set, as noted in Chapter :, section ,.,,
enables a constituent to function as a topic. Let us assume that topics
do not need to be reconstructed to trace position and therefore do not
form a dependency with their trace. We would then predict that topic-wh-
phrases would not be sensitive to island constraints and that the extraction
data in (:o) and (:,) should be perfect, which is clearly not the case. The
reason this is so has to do with superiority, the topic of section 3.
4.2.5 Topic-islands
The constraint on I-dependencies, which constrains dependencies accord-
ing to whether the syntactic structure and the f-structure are aligned in a cer-
tain way, may be context dependent. This is not always the case. Extraction
out of sentential subjects is always ungrammatical and cannot be contextu-
ally ameliorated. (:8) gives the f-structure assigned to such a case:
(:8) *Who is [that John likes t]
In order to comply with the constraint on I-dependencies, the subject,
in this case a sentential one, must be assigned topic. Since dependents
must be in the focus domain, they cannot be identied within topics
and extraction will always be blocked.
The extraction from a senten-
tial subject of a restrictive wh-phrase is also blocked in spite of the fact
that reconstruction is not required in such a case. This is because such
extraction will leave an incomplete topic as shown in (:,).
(:,) *[Which girl]
[is [that John likes e]
The subordinate topic (top) contains a gap which prevents it from pro-
viding a referent and hence disqualies it as a topic. Sentential subjects are
thus absolute islands irrespective of the nature of the extracted item.
The application of the constraint on I-dependencies to sentential sub-
jects depends on an f-structure analysis in which the sentential subject
is necessarily analyzed as a topic. Erteschik-Shir I,,, oers a number of
arguments to this eect. Miller :ooI provides much more evidence: He
rst argues against Bolinger I,,,: ,:, who claims that the content of sen-
tential subjects must not be topical and that sentences such as (,oa) are
impossible, as opposed to (,ob) and (,oc), because the anaphoric link
entails topic status. The sentences in (,o) should be considered in the con-
text of the question: What do you think of running him as a candidate?
For the idea that topics are islands see also Meinunger :ooo.
But see Kluender :oo for examples of sentential subjects which present graded extrac-
tion results.
information structure constraints
() a *To do that would be a good idea.
b To run him as a candidate would be a good idea.
c It would be a good idea to do that.
(a) contains the anaphoric element that and therefore extraposition
is obligatory as shown in (c). (b) is possible because, according to
Bolinger, the speaker is turning the question over in his mind and treating it
as his own idea. Miller employs a corpus investigation and shows that
Bolingers judgements are not replicated in naturally occurring sentences.
One example from Miller which illustrates that anaphoric elements can
occur in non-extraposed innitivals is given in () (his (c) ).
() Neither had a choice other than to accept the invitation.
To have refused would have been political suicide.
According to Miller the correct distinction between extraposed and
non-extraposed clauses is that they must be discourse-old or directly
inferable from the previous discourse context in order to remain
in subject position. If this condition does not hold, extraposition is
obligatory. Consider the following example from Miller (his () ):
() [] But we must never forget, most of the appropriate heroes
and their legends were created overnight, to answer immediate
needs. [] Most of the legends that are created to fan the res of
patriotism are essentially propagandistic and are not folk legends
at all. [] Naturally, such scholarly facts are of little concern to
the man trying to make money or fan patriotism by means of
folklore. That much of what he calls folklore is the result of beliefs
carefully sown among the people with the conscious aim of produc-
ing a desired mass emotional reaction to a particular situation or
set of situations is irrelevant.
Miller predicts the acceptability of this complex sentential subject since
its purpose is to sum up the content of the preceding forty lines and
thus make it available as a discourse referent for the judgment is irrel-
evant. This example also clearly shows, as pointed out by Miller, that
it cannot be mere heaviness that accounts for the dierence between
extraposition and non-extraposition as is sometimes assumed. Whereas
all the examples of sentential subjects in Millers corpus are discourse old
or inferable, extraposed sentential subjects can be either discourse new,
For the sources of Millers examples and many more illustrations of the points made by
Miller see his paper.
information structure constraints I,I
discourse old, or inferable. (,,) (Millers (Io) ) is an example of a dis-
course-new extraposed sentential subject.
(,,) European Central Bank Row Wont be Last
PARISIt is astonishing that the real questions about Europes
new single currency, the euro, and about the new European
Central Bank were never addressed during the I:-hour row
among European governments that ended in Sundays sad
compromise on the new banks president. Those questions are:
Can this bank truly be independent? And, if it tried to be truly
independent, could it survive? The answer to both clearly is no.
(Herald Tribune, , May I,,8, rst lines of the article)
Miller shows that the non-extraposed variant of this sentence would be
infelicitous were it to occur discourse initially (his (Io) ):
(,) #That the real questions about Europes new single currency,
the euro, and about the new European Central Bank were never
addressed during the I:-hour row among European govern-
ments that ended in Sundays sad compromise on the new banks
president is astonishing.
Millers claim that sentential subjects are discourse old is thus corrobo-
rated, and the claim made here that they are topics is too.
Since canonical f-structure is asymmetric, with a special status
assigned to subjects, it predicts the existence of subjectobject asymmet-
ries such as the absolute islandhood of sentential subjects, but not of
other sentential complements. In the next section, the that-t eect which
has been accounted for by the ECP in terms of proper government turns
out also to receive an account in terms of IS.
4.2.6 That-t effects
Bayer :oo, employs the idea that topics are islands to account for the
that-trace eect which blocks extraction of a subject in the presence of
an overt complementizer as in (,,a):
According to Bayer, previous accounts (including ECP) are based on the subject having
a special role. Bayer shows that any account which singles out subjects will fail in German
where the that-trace effect arises not only with nominative subjects as shown in (i) but also
with non-nominative subjects (ii):
(i) *Wer glaubst du, da raucht?
believe you that smokes
I,: information structure constraints
(,,) a *Who do you believe that t
likes Mary?
b Who do you believe t
likes Mary?
c Who do you believe (that) Mary likes t
In (,,b), without the complementizer, extraction is ne. (,,c) illustrates
that object extraction is insensitive to the presence of the complemen-
tizer. The eect is one of the well-known cases of subjectobject assymetries
subsumed under the ECP. It is well known that the that-trace eect is
suspended when an element intervenes between the complementizer
and the trace (the anti-adjacency eect):
(,o) Leslie is the person who
I said that under no circumstances
would t
run for president.
Bayer argues that these cases involve an f-structure with a stage topic (see also
Erteschik-Shir I,,,: :,), allowing for a non-topical interpretation of the
subject. The that-trace eect thus boils down to a constraint on extracting
topics (constrained to subject topics in English.) This generalization does
not, however, solve the problem of why the constraint only applies when
the complementizer is overt. Bayer argues that when the complementizer is
left out, the main clause functions as a parenthetical and is ignored when it
comes to extraction.
Bayer shows this for English as well:
(,,) Wh-object
a Who did John believe Susan will meet?
b *Who did John believe Susan will meet?
(,8) Wh-subject
a Who did John believe will meet Susan?
b Who did John believe will meet Susan?
(,,) shows that the main clause material intervening between the wh-
phrase and the subordinate clause cannot be left out, and so cannot be
considered to be parenthetical when the object is extracted. Object extrac-
tion is therefore not predicted to depend on the presence or absence of an
overt complementizer as is indeed the case. (,8) shows that when the sub-
(ii) *Wem glaubst du da Angst ist?
believe you that fear is
Who do you believe is afraid?
Bayer uses examples of Wasfr split to argue for a that-trace effect for objects as well. The
reader is referred to Bayers paper for these data.
Bayer cites Reis I,,, for this idea.
information structure constraints I,,
ject is extracted, it is possible to leave out the main clause, and therefore it
can be analyzed as a parenthetical.
Bayer also explains why it is that the
eect is weaker in German than it is in English: In English the subject is
more strongly identied with the topic than it is in German and other lan-
guages with relatively free word order. Bayers explanation has to do with
the dierent canonical f-structures of the two languages. The canonical
f-structure of German is discussed in section ,.I.: below.
Bayers account also predicts the following results of the focusability
test in Van Valin and LaPolla I,,,: o,: (their ,.I,,):
(,,) a Speaker I: Scully said Mulder talked to the detective.
Speaker :: No, Skinner. (= Skinner said, = Scully said
Skinner talked, = Mulder talked to Skinner)
b Speaker I: Scully said that Mulder talked to the detective.
Speaker :: No, Skinner. (= Skinner said, = Mulder talked
to Skinner, but ?? Scully said Skinner talked)
According to Van Valin and LaPolla, Skinner can replace any of the
three NPs in Speaker Is utterance. In the presence of the complemen-
tizer in (,,b), however, it is dicult to interpret Skinner as replacing
Mulder. It follows that it is harder to interpret the subject of the that-
clause as a focus in the presence of the complementizer than it is in its
absence. Van Valin and LaPolla corroborate this evidence by the follow-
ing observation: It is easiest to interpret Mulder as the focus in (b) if
that is destressed and Mulder is stressed (o,,). Van Valin and LaPolla do
not provide an explanation for these observations, but they do predict a
correlation between focusability and extraction. These observations are
not surprising in view of Bayers argument to the eect that extraction
of a subject under that violates the topic constraint. If the element in
question must be a topic, then the fact that it cannot be focused follows.
Van Valin :oo,: :,: supplies the following explanation for why sub-
jects can be extracted in the absence of a complementizer: Note that
when the complementizer is absent, the subject of the embedded
clause occurs in eect in a position equivalent to the nal position in
the matrix core, a position which is the most unmarked focus position,
and in this case marked narrow focus on the subject of the embedded
clause is indeed possible According to this explanation it is predicted
As noted in Erteschik-Shir I,,,: :,, the complementizer is optional in English only
with bridge verbs. These verbs make for natural parentheticals, providing further evidence
for Bayers view.
information structure constraints
that extraction from the subject of an embedded complement without
a complementizer is degraded if there is a temporal or other adjunct in
the main clause as illustrated in () (Van Valins .).
() a Who did Kim say talked to Dana?
b *Who did Kim say yesterday talked to Dana?
In such cases, the subject position cannot be argued to be in the unmarked
focus position of the matrix. This type of example is also predicted by Bayers
idea that subject extraction depends on the main clause being interpreted as
a parenthetical:
() a Who did Kim say talked to Dana?
b *Who did Kim say yesterday talked to Dana?
It follows from Van Valins discussion that focus on a subject of a sub-
ordinate clause is marked, yet an explanation for the fact that extraction
of such a subject in the presence of a complementizer is totally blocked is
not oered. Van Valins and Bayers approaches are not all that different.
For Van Valin it is the focusability of the subject which is at stake; for
Bayer it must not be a topic. Either way the that-t eect can be seen to
follow from the constraint on I-dependencies since both the require-
ment of focusability and the topic constraint are built into it. The con-
straint on I-dependencies based on the idea that canonical f-structures
reduce processing load also predicts superiority eects as shown in the
next section.
4.3 Superiority
Superiority provides a good test case to demonstrate how f-structure
theoretical constraints predict both the data and their contextual depen-
dence. The examples in () show superiority eects in English:
() a Who read what?
b *What did who read?
c Which boy read which of the books?
d Which of the books did which boy read?
e ?What did which boy read?
f ?*Which of the books did who read?
The answer to a multiple wh-question forms a paired list, as demon-
strated in ():
The account of superiority in English is from Erteschik-Shir . The extension to
Hebrew, German, and Danish is from Erteschik-Shir b.
information structure constraints
() Q: Who read what?
A: John read the Odyssey and Peter read Daniel Deronda.
Such an answer can be viewed as identifying each object (answer to
what) with one of the subjects (answer to who). In this sense the
multiple wh-question itself forms an I-dependency in which one wh-
phrase is dependent on the other.
Superiority eects are the result of two I-dependencies in the same

What did who read t
One I-dependency is between the fronted wh-phrase and its trace. The
other one is between the two wh-phrases. As () illustrates, the depen-
dent is identied in two dierent dependencies at once. This results in
an interpretative clash, thus blocking the processing of the sentence.
The constraint on I-dependencies is not violated, however, since the
subject wh-phrase can be assigned topic (the question ranges over a
discourse-specied set; it is d-linked) and the trace can be analyzed as within
the focus domain. Since it is not f-structure assignment which rules out
the sentence, context should not have an eect on cases of superiority. This
prediction is false, however, as shown by the following well-known example:
() I know that we need to install transistor A, transistor B, and
transistor C, and I know that these three holes are for transistors,
but Ill be damned if I can gure out from the instructions where
what goes! (Pesetsky , from Bolinger )
The answer to this puzzle lies in a proper understanding of the distinc-
tion between d-linked and non-d-linked questions, however not the one
proposed in Pesetsky (see Erteschik-Shir ; ). (a) and (b)
illustrate a non-d-linked and a d-linked question, respectively, and the
f-structure of each one:
() a What did you choose?
[What] did you
[choose t]
b Which book did you choose?
[Which book]
[did you choose t]
In the non-d-linked question in (a), the fronted wh-phrase and its trace
form an I-dependency and the trace is interpreted as an anaphor. Such a
question must therefore conform to the constraint on I-dependencies. In
(b), however, the fronted wh-phrase functions as a topic in that it ranges
I,o information structure constraints
over a contextually available set (of books). The trace can therefore be inter-
preted on a par with a coreferent pronoun, since the set over which it ranges
is discoursally available. Since no I-dependency is dened, the constraint is
not violated, hence no superiority eects are predicted with which-phrases,
which must be interpreted as d-linked. Questions with simple wh-phrases
can be interpreted as being d-linked if the context provides a set over which
they must range. That is why superiority violations such as (,) can be con-
textually ameliorated. They are always degraded, however. The reason is
that both wh-phrases have to be interpreted as topics as shown in (,):
(,) [Where]
[ [what
[goes t]

The subject wh-phrase forms an I-dependency with the trace in order to
render the pair-list reading. The constraint on I-dependencies requires
the subject to be a topic. The fronted wh-phrase must be interpreted as a
topic because otherwise it will form an I-dependency with the trace which
will then be doubly identied as in (). Bolingers detailed context allows
for such an interpretation. The question will be viewed as degraded rela-
tive to whether the context forces a topic reading on both wh-phrases or
not. Note that both wh-phrases must be interpreted as d-linked. Which-
phrases are necessarily d-linked. Therefore, multiple wh-questions involv-
ing only which-phrases are perfect, as shown in (:d). When only one of
the wh-phrases is a which-phrase, the other depends on context to receive
a d-linked interpretation. This is why (:e) and (:f) are degraded.
Sag et al. :oo, present a number of real occurring illustrations of
violations of the superiority eect which do not involve d-linking.
A selection of these follows:
(8) A: Did you know that there are no licensing laws or sales taxes in
B: I did not. What did who bring back?
(,) Although nothing on this planet (or any other) can compete
with the utter horror that is cilantro! Where the heck did who
the heck come up with adding that gawdawful weed to otherwise
civilized hote-cue-zeen?
(,o) We watch for a few more minutes as the recriminations begin:
what did whom [sic] say and what did who hear?
[Tomato is Coming, by Crispin Oduobuk]
information structure constraints I,,
(,I) I must have missed something. What did who do to Pierre
According to these authors, what characterizes these examples is the
fact that the initial wh-expression continues the topic thread and gen-
erally asks the more pertinent and relevant question in relation to the
previous discourse. If this generalization is correct, then these data
are in fact predicted by the constraints advocated here and in partic-
ular the idea that topic-wh-phrases do not require reconstruction in
order to be interpreted. The only change is to allow for non-d-linked
topic-wh-phrases which, it seems, do occur.
The examples in (,:) provide further evidence for the analysis of
su periority eects proposed here. They illustrate that superiority eects
also arise in single wh-questions when the subject is a non-specic
indenite, i.e., a subject which cannot be interpreted as a topic:
(,:) a *What did a boy nd?
b (?)Which book did a boy nd?
c What did a certain boy nd? (:e)
d What did a BOY nd?
e What do boys like?
(,:a) violates the constraint on I-dependencies because the subject
cannot be interpreted as a topic. In (,:b), the fronted wh-phrase
is d-linked and therefore does not form an I-dependency with its trace.
It is degraded on a par with a sentence with an indenite subject and a
denite object as in (,,):
(,,) a (?)A boy found the book.
b A BOY found the book.
(,,) is degraded because it is a non-canonical f-structure (cf. (:) ) in
which the object is the topic. Note that contrastive stress on the subject
as in (,,b) enables its interpretation as a topic, rendering a canonical
f-structure. (,:c), (,:d), and (,:e) do not violate the constraint on
I-dependencies because specic, contrastive, and generic indenite sub-
jects are interpretable as topics.
Kaynes (I,8) facts in (,) and (,,) show that, surprisingly, an extra
wh-phrase improves superiority violations:
Sag et al. :oo, offer a processing account of superiority effects. Their account is discussed
in section ,.
information structure constraints
() What did [who]
[hide t where]

() Who knows what who
[saw t]

This is because the extra wh-phrase makes it possible to circum-
vent doubly identifying the trace. In (), for example, the fronted
wh-phrase forms an I-dependency with the trace. This dependency is
licensed by the constraint on I-dependencies since the subject is inter-
preted as a topic and the trace is embedded in the focus. Another I-
dependency is formed between the two remaining wh-phrases. This
I-dependency is also licensed by the constraint since the dependent is
embedded in the focus. The presence of the extra wh-phrase enables
the formation of two separate I-dependencies without forcing a double
identication of the trace as in the classic case in (). This is how the
extra wh-phrase saves the construction.
Although Kayne-type questions are an improvement on the classical
case, they are still quite degraded. There are two reasons for this. First,
the subject wh-phrase has to be contextualized as ranging over a topic
set. Second, the integration of the two separate dependencies poses a
heavy processing load: One I-dependency in () is between who and
where allowing for the pair-list interpretation of these two wh-phrases.
However, in order to process the question the fronted wh-phrase what
must also be accommodated so that the interpretation of the question is
that it asks for a triple list reading.
The account of superiority eects proposed here thus aords an expla-
nation of when context can improve acceptability and when it cannot and
predicts the ne distinctions in acceptability evident in the English data.
4.3.1 Superiority in other languages
The account of the English superiority data extends to other languages,
once the nature of their canonical f-structures is determined. In the next
sections Hebrew, German, and Danish superiority data are discussed and
it is demonstrated that superiority eects are determined by the same
considerations as in English. Dierences are due to variation in the appli-
cation of the constraint on I-dependencies which is in turn determined by
the particular canonical focus structure of the language in question.

Triple dependencies are not derivable in this framework, a desirable result since they
do not render an optimal output.
information structure constraints I,, Superiority in Hebrew The rst observation concerning
Hebrew is that although topicalization may result in OSV, superiority
violations are licensed only in the order OVS, as shown in (,oa) and (b)
from Fanselow :oo:
(,o) a ma kana mi?
what bought who
b *ma mi kana?
what who bought
c mi kana ma?
who bought what
(,oa) is only licensed in a d-linked context in which a set of goods are context-
ually specied and (,oc) requires a d-linked context in which a set of buyers
are contextually specied. Note that there is no parallel to a which-phrase
in Hebrew. The phrase that comes closest, eize X, is best paraphrased as
what X and it does not require a contextual set. The fronted d-linked wh-
phrase (,oc) does not form an I-dependency with its trace. Therefore only
one I-dependency is at work in such Hebrew multiple wh-questions, namely
the one that renders the paired reading. (,,) shows this for (,oa).
(,,) ma kana mi

The constraint on I-dependencies constrains I-dependencies to the canon-
ical f-structure of a particular language. In English, the canonical f-structure
is one in which syntactic structure and f-structure are aligned. The fact that
the OVS and SVO orders of (,oa) and (c) are equally good in Hebrew and that
the OSV order of (,ob) is ruled out may mean that it is the OSV word order
which is the culprit. The dierence between OSV and OVS in Hebrew is asso-
ciated with the function of the subject when the object is fronted. When it is
interpreted as a topic, it is placed preverbally, and when it is focused, it is placed
after the verb. The examples in (,8)(oo) demonstrate that this is the case:
(,8) a et hasefer moshe kana.
the-book Moshe bought
Moshe bought the book.
b et hasefer kana moshe.
et marks denite objects. mi (= who) in object position is most naturally marked
with et whereas ma (= what) is not. There is some indication that when object wh-words
are marked with et they are d-linked. The distinction between mi and ma in this respect
may be connected to the fact that it is more common to contextually specify sets of humans
than sets of objects.
information structure constraints
() a *et hasefer yeled exad kana
the-book boy one bought
Some boy bought the book.
b et hasefer kana yeled exad.
the-book bought boy one
Some boy bought the book.
() a et hasefer hu kana.
the-book he bought
He bought the book.
b *et hasefer kana hu
the-book bought he
He bought the book.
() shows that a denite subject which can function as both a topic
and a focus can occur both preverbally and postverbally. () shows
that an indenite subject which cannot be interpreted as a topic is
restricted to the postverbal position. (), in turn, shows that a subject
pronoun, which must be interpreted as a topic, can only occur prever-
bally. (a) and (a) also require contextualization in view of the fact
that both the topicalized object and the preverbal subject are inter-
preted as topics. Since every sentence requires a focus, this forces the
verb to be focused or else one of the arguments must be interpreted
contrastively. In either case the f-structure is marked. To complete our
investigation of the unmarked f-structure in Hebrew, we must also
examine the untopicalized cases:
() a moshe kana et hasefer/sefer
Moshe bought the-book/(a) book
b ?yeled exad kana et hasefer
boy one bought the book
The most natural f-structure of (a) is one in which the subject is the topic
and the VP or object is focused. (b) with the denite object interpreted as
a topic is marked. The results of both orders are schematized in ():
() a *O
b ?O
c O
d *O
e S
f ?S
information structure constraints I8I
(o:c) and (e) are the only unmarked cases. I conclude that the unmarked
focus structure in Hebrew is one in which the topic precedes the verb
and the focus follows it. Hebrew dependencies therefore do not depend
on the syntactic structure of the sentence, but only on the linear
order of topic and focus with respect to the verb. The constraint on
I-dependencies, parameterized to reect the Hebrew canonical f-structure,
is shown in (o,):
(o,) An I-dependency can occur only in a canonical f-structure:
V [ Y ]
(o,) correctly rules out (,ob) because its f-structure (equivalent to
(o:b) ) is not canonical as shown in (o):
(o) ma
The I-dependency between the two wh-phrases is therefore not licensed.
Both (,oa) and (c) are licensed since they conform to the canonical
f-structure in (o,). It also follows that they are restricted to d-linked
contexts since the initial wh-phrase must be a topic.
I conclude that multiple wh-questions in Hebrew are governed by the
same considerations as they are in English. Dierences between the two
languages follow from their dierent canonical f-structures. Superiority in German According to many authors, German
lacks superiority eects. Wiltschko I,,8 not only argues that this is not
the case, but also explains why German superiority eects have been over-
looked. One of the reasons she oers is that controlling for d-linking is
dicult since discourse-related contrasts are often rather subtle (,).
Along these lines, Featherston :oo,b performed an experiment in which
informants were asked to grade the data according to an open-ended
scale. His results showed that superiority eects are robustly active in
German. It turns out, then, that German does not dier signicantly from
English in this respect. Fanselow :oo, although aware of Featherstons
results, still distinguishes the status of English and German with respect
to superiority eects. Fanselow points out that in German the superiority
eect does appear when the subject wh-phrase is in Spec, IP (his (,,) ):
(o,) a wann hats wer gesehen
when has it who seen
b ?*wann hat wers gesehen
who saw it when?
In this section sentences with stage topics have been ignored.
I8: information structure constraints
In (o,a) the subject follows the object clitic, indicating its VP-internal
position. In (o,b), it precedes the object clitic and so must be outside the
VP. These data are reminiscent of the Hebrew facts just discussed: German
subjects in Spec, IP must be interpreted as topics, whereas VP-internal
subjects are interpreted as foci. D-linking is also required, as noted by
Wiltschko. Fanselow :oo gives the following illustration (his (:) ):
(oo) wir haben bereits herausgefunden
we have already found out
a wer jemanden gestern anrief, und wer nicht
who.nom someone.acc yesterday called and who.nom not
b wen jemand gestern anrief, und wen nicht
who.acc someone.nom yesterday called and who.acc not
Aber wir sind nicht eher zufrieden, bis wir auch wissen
But we are not earlier content until we also know
a wer WEN angerufen hat
who.nom who.acc called has
who called who
b wen WER angerufen hat
who.acc who.nom called has
According to Fanselow, OSV order is licensed only if the object is dis-
course linked, but SOV order is also allowed in an out-of-the-blue
multiple wh-question (his (,) ):
(o,) Erzhl mir was ber die Party.
Tell me something about the party.
a Wer hat wen getroen?
who.nom has who.acc met?
who met who?
b ??Wen hat wer getroen
who.acc has who.nom met?
Fanselows example cannot, however, be considered out of the blue. A
party necessarily involves a set of participants. These are what the wh-
phrases range over in the questions following the initial sentence. Since
both wh-phrases range over the same set of party participants, they
are equivalent. (o,a), in which no reordering has occurred, is therefore
From these data I gather that the German canonical f-structure is
similar to the one proposed for Hebrew, with only one small dier-
ence: German, too, requires that the rst argument be the topic and
information structure constraints I8,
the second be the focus, yet the status of the subject is determined di-
erently: German subjects are interpreted as foci when they are VP-inter-
nal, and as topics when they are not, as shown in (o,). The position of
the subject is transparent only in the presence of adverbials or other ele-
ments that mark the VP boundary.
In many of the examples in which
such elements are absent, the linear position of the subject wh-phrase
gives no clue as to its syntactic position. In those cases, the subject will be
interpreted according to contextual clues. An I-dependency is therefore
licensed between two wh-phrases in German when the rst one is inter-
preted as a topic and the second as a focus. Superiority in Danish According to Fanselow :oo, Swedish does
not exhibit superiority eects (his (I:) ):
(o8) Vad koepte vem
what bought who
In Danish, the same question is degraded:
(o,) a Hvem kbte hvad?
who bought what
b ?Hvad kbte hvem?
what bought who
Overt d-linking signicantly improves the question:
(,o) Hvilken bog kbte hvilken pige?
Which book bought which girl?
Danish may have a preference for overtly marking d-linked wh-phrases
instead of just depending on contextual clues. Danish is like English in
this respect, except that the preference in English is even stronger. Danish
diers from English in that superiority eects in subordinate clauses are
not ameliorated by overtly d-linked wh-phrases:
(,I) a *Jeg ved ikke hvad hvem kbte
I know not what who bought
b *Jeg ved ikke hvilken bog hvilken pige kbte
I know not which book which girl bought
Danish generally marks the topic by fronting it to sentence-initial posi-
tion. This is also the case if the topic is located in the subordinate clause.
Topicalization within a subordinate clause is therefore excluded.
See Diesing I,,: for this effect.
Topicalization is licensed in subordinate clauses under a few bridge verbs such as think.
In such cases the syntactically subordinate clause functions as a main clause.
information structure constraints
follows that whereas word order may signal the f-structure of the main
clause, the order within subordinate clauses does not. This is the explanation
I propose for the dierent behavior of Danish main and subordinate
clauses with respect to superiority eects. Scrambling languages such
as German dier: scrambling positions topics outside the VP in subor-
dinate clauses as well as main clauses. No dierence between main and
subordinate clauses is predicted in scrambling languages. This predic-
tion is borne out for German.
Fanselow rejects the idea that the
availability of scrambling is what explains the lack of superiority eects
because there are non-scrambling languages which also lack superiority
eects. I would not be surprised if non-scrambling languages exhibit the
same dierence between main and subordinate clauses as Danish.
Since the verb in Danish main clauses must appear in second position,
the canonical f-structure is identical to the one proposed for Hebrew.
The only dierence between Hebrew and Danish is the preference for
overtly d-linked wh-phrases.
What is common to the languages examined here is the need for
d-linking at least one of the wh-phrases in multiple wh-questions. That
is why such questions are always sensitive to context. Variation among
languages follows from three parameters: the canonical f-structure, the
availability of topicalization and scrambling processes, and the array of
wh-phrases available in a particular language. As I have shown here, all
three must be taken into account in order to predict the cross-linguistic
distribution of superiority eects.
In view of the fact that the constraint on I-dependencies is parameter-
ized according to the canonical f-structure of the language in question,
dierent predictions are made for each language. These predictions are
extended to all I-dependencies and it remains to be seen if indeed the other
I-dependencies work according to the canonical f-structure assigned to
each language. Here we have examined the wh-trace dependency as well
Hebrew is like Danish in this respect. Since Hebrew is not a scrambling language, this is
what is predicted. Since English is not a scrambling language, English should also exhibit a
difference between main and subordinate clauses. This is not the case:
(i) Which book did which boy buy?
(ii) I dont know which book which boy bought.
The difference between main and subordinate clauses in Danish arises because only in the
former is f-structure marked by word order. English main clauses do not differ from subor-
dinate clauses in this way. This may explain why no difference in superiority effects between
main and subordinate clauses can be detected.
For a different cross-linguistic processing account and a discussion of superiority in
German, Icelandic, Hungarian, and Dutch, see Haider .
information structure constraints I8,
as dependencies between multiple wh-phrases and shown how the con-
straint predicts islands and superiority eects. The constraint also makes
interesting predictions concerning anaphora and logophora as well as
association with focus as shown in Erteschik-Shir I,,,: chapter o.
4.4 IS constraints on complex NP
4.4.1 Extraposition from NP
Guron I,8o shows that PP extraposition depends on interpretation.
Except for (,,), the following examples are from Guron (the extraposed
PPs are in italics):
(,:) a A man with green eyes appeared.
b A man appeared with green eyes.
(,,) a The man with the green eyes appeared.
b *The man appeared with the green eyes.
(,) a A man with green eyes hit Bill.
b *A man hit Bill with green eyes.
(,,) a John read a book by Chomsky over the summer.
b John read a book over the summer by Chomsky.
(,o) a John read that book by Chomsky over the summer.
b *John read that book over the summer by Chomsky.
Guron I,8o: o,I makes the following generalization: the extraposed PP
must be marked Complement of Focus NP.
Whether the subject is
focused or not depends on whether the sentence is predicational or
Relative clause extraposition is also sensitive to the constraints Guron proposes for PP
extraposition. See also Ziv and Cole I,, and Maynell :oo,. Maynell argues that extraposed
phrases must answer the immediate question under discussion (see Roberts I,,8 for this
notion), i.e., it must be a focus. The head of the relative clause must be a topic and therefore,
in order for the head and clause to match informatively, the relative clause turns out in gen-
eral to be contrastive. In order for the relative clause to be focused, the main clause VP must
also be presupposed or informatively empty. (i) is an example from Maynell:
(i) THOSE DOCTORS were sulking who had been at the FOOTBALL game the night
The context includes a contrast between doctors who went to the football game and those who
went to a concert, as well as the fact that some of the doctors were in a disagreeable mood.
In fact, she generalizes this statement to say that the extraposed constituent must be
interpreted as a Complement of Focus, where focus can be any X' node to cover extraposi-
tion from AP and ADV as in (i) from Jackendoff I,,,:
I8o information structure constraints
presentational. These are interpreted according to the following prin-
ciples (Guron I,8o: o,,):
(,,) a Predication. The subject refers to an individual or object (or set
of these) whose existence in the world of the discourse is pre-
supposed: thematic subject. The VP describes a property of the
thematic subject.
b Presentation S. The VP denotes, essentially, the appearance of
the subject in the world of discourse.
It is not dicult to translate these two generalizations into IS terms:
Predication is a relation between a topic subject and a focus VP, i.e., it is a cat-
egorical statement with a canonical f-structure. A Presentation S is a sentence
in which the subject is introduced into the world of discourse. In IS terms,
the subject must be focused and the topic is a stage topic. This corresponds
to the analysis of existential sentences in Chapter ,, section ,.,. Existential
sentences are therefore presentational in this sense. It follows from these
denitions that extraposition from the subject of a presentation sentence is
predicted to be possible whereas extraposition from the subject of a predica-
tional sentence will be ruled out. The data in (,:) and (,) follow: (,:) with
the verb appear is presentational, and its subject is therefore focused. (,),
with the verb hit, is not, which is why extraposition is not licensed. Guron
I,8o: o,, stresses that the nature of the verb does not in and of itself predict
whether the sentence will be presentational. A number of non-presenta-
tional verbs can be interpreted as predicating no more of the subject than
its appearance (or disappearance) in the world of the discourse. The fol-
lowing are Gurons illustrations of this observation:
(,8) a A man walked in from India.
b A train chugged past with many passengers.
c A bird darted by with golden wings.
d Some books were burned by Pablo.
e A man died from India.
Guron argues that the verbs in the sentences in (,8) can be understood
as pragmatic synonyms of appear (or disappear), in some cases with the
help of context. (,8d) and (,8e), for example, are completely acceptable
in the following contexts, respectively:
(i) a I was sorry all day about what had happened
b How certain are you that your horse will win?
c Jim lived near the bridge last year, where there are some owers growing.
information structure constraints I8,
(,,) First the Chilean military burned the books of all political gures
sympathetic to the Allende government. And then some books
were burned by Pablo Neruda.
(8o) Several visitors from foreign countries died in the terrible accident.
A woman died from Peru and a man died from India.
These contexts enable the verb to be pragmatically emptied of all
semantic content beyond that of appearance in the world of the dis-
course (o,). This contextual emptying of the meaning of the verb is
reminiscent of the same process with the verbs of manner-of-speaking
discussed in section :., above and also allows the verb not to be focused,
allowing the subject to be focused instead.
The distinction between (,,) and (,o) also falls out from the principle
that the extraposed PP must be linked to a focus: The indenite object
in (,,) is a potential focus, the denite object in (,o) is an unlikely focus.
Guron I,8o: o,, points out, however, that (,ob) is acceptable in a con-
text in which that book by Chomsky is interpreted as something alluded
to but not present in the discourse context.
(8I) John nally read that book over the summer by Chomsky, you
remember, the one you had recommended to him.
Gurons generalization is therefore robust. Extraposition depends on
the focusability of the constituent from which extraposition occurs
and focusability, in turn, depends on lexical and contextual factors. It
is not obvious, however, what the reason for this constraint might be.
One possibility, which connects extraposition to extraction, is that the
gap left behind by extraposition is only visible within a focused
constituent as diagrammed in (8:):
(8:) [ e
That gaps are not visible within non-focused constituents has been observed
in the context of the constraint on I-dependencies. If the gap is viewed as a
dependent here as well, (8:) falls under this constraint as well.
4.4.2 Extraction from NP
Guron I,8o: ooo, draws a connection between extraposition from NP
and extraction from NP:
(8,) a Who did you see a picture of?
b *Who did you see that picture of?
I88 information structure constraints
(8) a Books were sold on linguistic theory.
b *Those books were sold on linguistics theory.
Guron proposes the name constraint to cover these cases. The name con-
straint rules out gaps within names. A Name is dened as follows: A Name is
a complete referring expression. It designates a unique object or individual (or
set of these) in the world of the discourse, either directly, through the use of
proper names or deictic expressions (John, that man), or indirectly, by means
of complements containing direct referring expressions (the girl who sits next
to you, some of those books). The denition comes close to being a denition
of topics and the continuing discussion indicates that topics is what Guron
has in mind. The name constraint is therefore equivalent to the topic con-
straint introduced in sections :., and :.o above. This constraint on extraction
from NPs does not cover the following data from Erteschik-Shir I,8I: I,:
(8,) a What did John write a book about?
b ?What did John revise a book about?
c *What did John destroy a book about?
In spite of the fact that extraction in these sentences occurs out of non-top-
ics, extraction is still less than optimal in two of the cases. As argued in that
paper, extraction from NP depends on the focusability of the NP from which
extraction occurs and that focusability depends, among other things, on the
nature of the verb and the context in which it occurs. (The reader should by
now be able to nd contexts in which both (8,b) and (8,c) become totally
acceptable.) These data follow easily from the constraint on I-dependencies,
which also covers the data pertaining to the topic constraint as noted above.
It is therefore unnecessary to maintain two dierent constraints on extrac-
tion. Note that the extraposition data mirror the extraction data as expected:
(8o) a John wrote a book last year about Chomsky.
b ?John revised a book last year about Chomsky.
c *John destroyed a book last year about Chomsky.
We therefore conclude here that what constrains both extraction out of
NPs and extraposition from them is the constraint on I-dependencies.
Another set of data that goes hand in hand with the extraposition data
and the extraction data is the completeness facts in (8,).
See also Erteschik-Shir I,,, for similar data and observations.
For similar data and discussion see Grimshaw and Vikner I,,, (who made the initial
observations); Erteschik-Shir and Rapoport :ooo; and Goldberg and Ackerman :ooI.
information structure constraints
() a *The book was written.
b The book was revised.
c The book was destroyed.
(a) becomes perfect if something is added to the sentence as shown in ().
() The book was written last year/in /by Chomsky.
What is wrong with (a) must therefore be that it is incomplete in that it
lacks a focus. Once an element is added to supply such a focus, the sentence
becomes good. The verbs in (b) and (c) do not need additional elements
in order to be complete. They provide a focus in and of themselves.
(), in which the denite subject is replaced by an indenite, is
() A book was written.
Here the topic is a stage topic and the whole sentence is in focus. With the
subject included in the focus, the sentence is complete. The problem with
the verb in (a) is that it does not have any content beyond make or cre-
ate when it appears with the object book. The other two verbs in () have
more content and therefore provide likely foci. In the next chapter, we exam-
ine the connection between focusability and predicate type in more detail.
4.5 Processing
The ID constraint and the constraint on I-dependencies have been
couched in processing terminology, but no arguments have been given
to support that processing is in fact at stake when it comes to the depen-
dency data. One argument assumed here is that grammatical constraints
are context independent and non-violable. In view of the graded and
context-dependent data involved with dependencies, the source of the
constraints imposed on them can therefore not be grammar as such.
Research pertaining to processing of islands is available: Kluender
views (unbounded) dependencies as instances of predication.
These dependencies must therefore comply with principles of predica-
tion. Kluender : posits the following principle:
() Initial argument expression NPs must be as referentially specic
as possible; all heads and speciers occurring in complex predi-
cates must be as non-specic in reference as possible.
For argumentation along these lines see also Featherston a and Haider . For
argumentation that processing of island constraints is subject to grammatical principles see
Sadeh-Leicht . Sadeh-Leicht does not relate to the graded data predicted by IS.
I,o information structure constraints
The predication principle distinguishes d-linked wh-phrases as being
referentially specic and is also sensitive to the ability of predicates to
provide bridges (see Chapter , for more discussion).
This principle is, grosso modo, very similar to the constraint on
I-dependencies posited here if specicity is taken to match topichood,
and lack thereof is taken to be focusability. This is not surprising in view
of the fact that Kluender integrates the focusability constraint on depen-
dencies from Erteschik-Shir I,,, with the topichood condition in Kuno
I,,o mentioned in section :.I above.
The main signicance of this work is that Kluender employs the Noo
component of event-related brain potentials (ERPs) as a measure for
investigating the role of referential specicity in dependencies. A rela-
tively large amplitude of the Noo indicates processing load. The initial
results of these experiments were promising, showing that processing
load increases cumulatively when the ingredients determining predica-
tion are lacking. This type of experimentation therefore is potentially
useful in supporting the view that dependencies are governed by pro-
cessing constraints dened in terms of IS principles.
Kluender I,,8 uses the same methodology to argue that the same
processing mechanism is involved in both weak and strong islands. An
example of a strong island, according to Kluender, is extraction out of a
relative clause, which here has been analyzed as a violation of the constraint
on I-dependencies, in particular, the fact that a dependent cannot be
interpreted within an unfocusable constituent. An example of extraction out
of a weak island is extraction out of an embedded question. Extraction
out of embedded questions is analyzed in Erteschik-Shir I,,, on a par
with the superiority eect discussed in section , above which is due to
the interpretative clash involved in the simultaneous interpretation of
the dependent in two dierent I-dependencies. Both constraints restrict
the interpretation or identication of the dependent, yet whereas the
dependent in a strong island cannot be identied at all, the dependent in
a weak island is doubly identied. Kluender agrees that islands result
from interpretative problems, but he argues that it is the same mechanism
which explains both strong and weak islands. In this paper, Kluender
proposes that the diculty involved in processing islands stems from the
need to maintain the antecedent of the dependency (the ller) in working
memory while encountering an additional referential processing burden
at the embedded clause boundary. Kluenders ERP data substantiated
his claim that the same increased load on working memory was to be
found in both types of islands. It is not clear, however, whether these
results in fact argue against the two somewhat dierent mechanisms
information structure constraints
argued for here since it is not obvious that the dierence between them
would show up as a dierence in processing load. As Kluender himself
comments, more ne-tuned research is required to settle these issues.

Kluender focuses on subject islands and again argues for a processing
explanation in terms of working memory constraints on storage and
discourse reference. He adds another parameter which is that subjects
are inherently dicult to process, and that complex ones are even more
so due to the fact that they contain multiple subconstituents that escalate
discourse referential processing costs sentence-initially ().
Research is also available on the processing of canonical word order
as opposed to non-canonical word order. According to MacDonald and
Christiansen , the well-known fact (since Bever ) that object rela-
tives are more dicult to process than subject relatives follows from the
fact that subject relatives exhibit canonical word order whereas object rela-
tives do not.
This supports the general idea behind the ID constraint, yet
cross-linguistic research examining the particular canonical f-structures
proposed here would be necessary to establish the validity of the proposal.
Kluender provides some evidence that it is linear distance between
ller and gap which contributes to processing load, yet he also maintains
that the non-canonical word order of object dependencies plays a role.
This can be seen clearly in SOV languages in which rich morphological
case marking is available, yet displacing the object to the left still has pro-
cessing consequences as reected in brain responses. Kluender nds this
surprising in view of the fact that ambiguity is resolved early on in view
of the availability of overt case marking on the object. These facts call for
more research which takes into account the IS functions of the sentence
elements as well. If, for example, it turns out that fronting of topics and
d-linked wh-phrases reduces the processing load, then the analysis pro-
posed here may by supported. An issue which may bear on this matter
is the trade-o between subject and verb phrase length in adult Italian.
Kluender : cites Hyams and Wexler : n. , who suggest
Goodall offers a processing account of certain properties of wh-questions in
Spanish and makes some interesting predictions concerning the processing of d-linked wh-
phrases: when extracted they ease processing; when they intervene between the extracted
element and the gap, they raise the processing load. I refer the reader to Goodalls paper for
the details.
For more research along the same lines see Arnon et al. .
Hsiao and Gibson argue that the reason Chinese object relatives are easier to
process than Chinese subject relatives is because Chinese relative clauses precede their
head nouns. They conclude that these data are consistent with the theory that processing
of embedded clauses is easier when their word order is canonical. In view of the fact that
canonical word order is not necessarily equivalent to canonical f-structure, this evidence
may or may not support the view proposed here.
information structure constraints
that the trade-o is between topical and non-topical information. A topi-
cal subject will have a non-topical predicate and vice versa, and non-topi-
cal constituents are longer than topical ones, which are often pronominal.
This line of reasoning predicts that null topics will end up with the lon-
gest predicates. Kluender argues against this explanation and argues that
the IS status of the elements cannot aord an explanation of the ndings
in Clark and Wasow which showed that disuency rates in the pro-
nunciation NPs are highest sentence initially and decrease during the
course of the sentence. In addition, according to Kluender, ERP studies
indicate that the beginning of a sentence entails elevated processing costs
in both comprehension and production (). Kluender claims that these
facts cannot be given an explanation in terms of IS, since, following the
reasoning in Hyams and Wexler , one would expect topics and sub-
jects to present less of an informational burden than objects. One factor
which Kluender does not take into account which may be of signicance
is that complex subjects generally are also information-structurally com-
plex and contain a subordinate f-structure.
The prediction would be that
a simple (pronominal or null) subject topic would provide no processing
load, whereas a complex NP subject with a subordinate focus structure
creating its topic status on the y would in fact elevate processing costs to
a large extent. In order to decide this issue it is therefore important to take
into account not only the main topic and focus status of the subject and
object, but also the IS function of the elements they contain.
Sag et al. argue for a processing account of superiority eects. They
assume the following wh-processing hypothesis: Given the choice between
several orders of wh-expressions, speakers disprefer those that (given the
context) are associated with greater processing cost. Such a processing
hypothesis is basic to any processing account. They argue for the following
processing factors which combine to derive the total processing cost:
() I: Gaps that are further from the ller are harder to process.
II: Less accessible llers are harder to process.
III: Less accessible intervenors are harder to process.
The rst factor is unrelated to IS. The other two factors are and can be
rephrased in the terminology used here. If we take accessible to mean
topical and non-accessible to mean focal, we derive that topical llers are
easier to process than non-topical ones and that focal constituents must
See Chapter , section . for examples of indenite subject topics with subordinate
information structure constraints
not intervene between the ller and the gap. The former goes along with the
idea that topical llers need not be reconstructed and the latter is predicted
by the constraint on I-dependencies: If focal constituents precede the gap,
then the gap position becomes non-focusable, violating the constraint
on I-dependencies. As with any IS processing constraint, graded data are
also predicted. The ndings presented in this article therefore provide a
modicum of support for the processing constraints argued for here and in
previous work. What Sag et al. cannot predict, without further ado, is the
kind of cross-linguistic variation described in section .
4.6 Architectural consequences
One of the conclusions drawn by Sag et al. is that since the superi-
ority eect is a consequence of wh-processing, there is no grammatical
constraint blocking superiority violations. The grammar thus remains
simple and the eect follows from general cognitive principles. If pro-
cessing is sensitive to IS notions as assumed here, the division of labor
between grammar and processing is not quite as obvious.
IS processing explains linear order as well as island constraints. This
brings the division between competence and performance into ques-
tion: Whereas competence is governed by universal principles of gram-
mar, performance is governed by processing principles which may not
be language specic. The processing principles proposed here involve
focusing, which is not a language-specic perceptual mechanism, yet
does take on particular linguistic features. F-structure as a whole is, how-
ever, unique to language and is therefore likely to belong to grammar.
Yet if f-structure constraints are in fact processing constraints, and, as
we have seen, their output could easily be viewed as a matter of perfor-
mance, then we either have to give up the competenceperformance dis-
tinction, or else the domain of grammar will be much less encompassing
than previously thought. Goodall views this new division of labor
as a good thing: If many problematic issues receive an account in terms
of processing, what is left for the grammar can potentially receive a more
minimalistic account.
Aspectual Focus
In the last chapter, we saw several instances in which lexical properties of verbs
have an eect on f-structure: Manner-of-speaking verbs block focusing on the
subordinate clause, and hence on extraction (Chapter , section .). Creation
verbs allow focusing on the complement of the picture noun phrase and hence
license extraction (Chapter , section .). The same eect was also seen with
respect to extraposition (Chapter , section .). Finally, sentence completeness
was seen to be aected by properties of the verb as well (Chapter , section .).
In addition, the lexical properties themselves can be aected by context: We
saw that a meaning component can be backgrounded if its content is context-
ually prominent and that contrast and modication can force focus on a verb
even though the verb itself does not require it. This chapter discusses these and
similar phenomena and oers a framework within which they can be analyzed.
Section briey outlines the lexical theory of atoms and introduces the notion
of aspectual focus. Section elucidates the lexical properties of manner-of-
speaking verbs predicting both the extraction facts and completeness facts. It
goes on to distinguish the lexical properties of the dierent verb types which
take picture NPs as their complements, again deriving their extraction and com-
pleteness properties. Then f-structure constraints on extraction in double object
constructions are shown to follow from their lexical structure. The topic of sec-
tion is missing objects which form structures with characteristic information
structure properties. Finally, section accounts for the unusual stress patterns
assigned to out-of-the-blue change-events pointed out in Chapter , section ..
5.1 The theory of atoms
In Erteschik-Shir and Rapoport ; ; ; a; and in preparation,
we oer a lexical analysis of verbs in terms of meaning components which we
refer to as the theory of atoms (AT).
There are three types of meaning com-
Note that in this theory, meaning components project syntactic structure deriving struc-
tures that are similar in many ways to the one in the work of Hale and Keyser (e.g., Hale and
aspectual focus
ponents: Manner (M), State (S), and Location (L). Verbs are composed of either
one or at the most two such components. Meaning components are inter-
preted in one of two ways:
. the component projects an activity
. the component modies adverbially
In AT, all components have these interpretative options in principle.
S-components, for example, project change-of-state events and modify
adjectivally and L-meaning components project change-of-location
events, and only modify locations. M-components project intransitive
activities, and when a verb decomposes into an M-component as well as
an S-or an L-component, theM-component modies the resulting cause
Each component type has a light version. Light verbs are the basic mem-
bers of their class and their meaning component has no additional meaning
other than class membership.
One dierence between the distribution of a
light verb and the other members of its class is that the light meaning com-
ponent must project a complement. This is because a meaning component
which has no content cannot modify and so must project in order to get
interpreted. Put and become, for example, are light verbs of the classes in
which shelve and (intransitive) break, respectively, are members:
() a John put the book *(on the (top) shelf).
b John shelved the book (on the
(top) shelf).
() a The vase became *(broken/shiny).
b The vase broke (into little pieces).
(a) shows that put, which contains a light L-component, requires a projected
(overt) PP, whereas for shelve, which has a heavy L-component, the comple-
ment is optional. Similarly become requires an adjectival complement in view
of its light S-component, whereas for break, which has a heavy S-compo-
nent, the complement is optional. An additional observation is the fact that
the optional complement of the heavier verbs cannot replicate the meaning
component itself, since it wouldnt add anything not expressed already by the
Keyser ; ). AT theory is only roughly outlined here. According to AT, the decomposi-
tion of verbs into atomic meaning components and their syntactic projection predicts syntac-
tic structure, aspectual interpretation, and argument selection without recourse to functional
projections or movement. For details see Erteschik-Shir and Rapoport ; ; ; a;
in preparation.
The term light verb is here used in the sense that the verb has little content. This use
should not be confused with the same term used for little v.
aspectual focus
verb. That is why (b), for example, requires an adjectival modier in order to
be acceptable. (Note further that the optional complement of heavy words is,
in fact, what is also known as a cognate object or complement.)
5.1.1 Aspectual focus
One of the advantages of AT theory is that the decomposition into
meaning components and the structures projected allow for the fore-
grounding of one meaning component of the verb (and its associated
structure) and the backgrounding of another. This type of focus is
here referred to as aspectual focus and should not be confused with
informational focus, although the latter is to some extent constrained
by the former.
Consider the well-known fact that accomplishments
(cause structures) can receive either a telic or an atelic reading as
shown in ().
() a John chopped the carrot for minutes.
b John chopped the carrot in a second.
Aspectual focus has been coined by Smith / to signify focusing on parts of event struc-
ture in order to explain aspectual properties. It is possible that the way the notion is used here is
in fact the same although it is applied to explain different types of data.
I use the term cause or accomplishment structure loosely to designate structures such as
(c), even though, depending on aspectual focus, the interpretation can be atelic.
The two copies of the verb form a chain of which only the rst copy is pronounced.
V c
S foc
the carrot
V (M)
In (c), the S-component projects, rendering the change (to a cut state)
interpretation, and cuts instrumental M-component modies the
V-V structure, rendering the interpretation that the change is caused by
a sharp instrument.
This complex structure aords the possibility of
aspectually focusing either of the components as indicated. When the M-
component is focused the activity reading is foregrounded allowing for
(a) and when the S-component is focused the change is foregrounded,
aspectual focus
rendering the telic reading in (b). The examples in () show that aspec-
tual focus on a component can be forced by modifying it.
() a ??John chopped the carrot into little pieces for three minutes.
b ??John carefully chopped the carrot in a second.
In (a) the S-component is modied making the sentence incompatible
with a durational time adverbial. In (b) the M-component is modied
adverbially forcing the atelic reading.
Punctual verbs do not exhibit these distinctions as illustrated in (a)
for the verb break. Such verbs require aspectual focus on the lower
V-projection as evidenced by (b).
() a John broke the vase *for three minutes/*in three minutes.
b John broke the vase into little pieces.
Aspectual focus on the upper V-projection depends on the nature of the
M-component of the verb. The M-component of break is force which is
what makes it punctual and hence does not allow aspectual focus.
Aspectual focus is not the same as sentential focus. Both sentences in
() can be assigned the same array of f-structures and stress patterns. Yet,
as shown in the next section, the two are sometimes related.
5.2 Meaning components and extraction
5.2.1 Manner-of-speaking verbs
The following data, repeated from Chapter , illustrate extraction out of
the complement of dierent manner-of-speaking verbs:
() a Who did John say that he had seen?
b ?Who did John mumble that he had seen?
c *Who did John lisp that he had seen?
These verbs have a manner-of-speaking (M
) meaning component. The
verb say is the light verb of this class and therefore requires a complement,
whereas the verbs containing heavier manner components in this class do
not. This is illustrated in ().
() a John said *(that he had seen Mary).
b John mumbled (that he had seen Mary).
c John lisped (that he had seen Mary).
The data in () can therefore be seen as the opposite, in a sense, of
the data in (): When the M-component is light, it does not have sucient
aspectual focus
content to be aspectually focused. The complement is needed in
order to provide such a focus. It follows that the complement of a
light verb is necessarily focused, and therefore allows extraction.

When the M-component is not light, yet the verb has a complement,
the meaning component is aspectually focused and it modies the
whole predicate, which provides the focus of the sentence, prevent-
ing the elements of the complement from being focused indepen-
dently, thus blocking their potential for extraction.
M-components can be defocused contextually, enabling focus on the sub-
ordinate clause, which then meets the requirement on extraction. This
is the case in (c) if the subject is a lisper and therefore identies the M-
component of the verb, as mentioned in the previous chapter. Another
example is () which improves in the context of an editorial board.
() a ??What did the paper editorialize that the minister had done?
b *What did you animadvert that he had done?
Context therefore interacts with the lexical representation of the
verbs. In AT the lexical representation consists of meaning components
which can be contextually manipulated. As shown in the previous chap-
ter, contrastive stress on the verb, but not on the subject, also inuences
() a ?Who did John SAY that he had seen?
b Who did JOHN say that he had seen?
The contrasted subject in (b) is still a topic and hence does not inuence
the f-structure of the predicate. When the verb is contrasted, however, it
makes it more dicult to focus on the elements within the complement
which explain the extraction facts.
5.2.2 Picture NPs
The examples in () show that extraction out of picture NPs is licensed
with perception verbs and creation verbs, but not with other verbs.
() a What did John paint a picture of ?
b What did John see a picture of ?
c *What did John burn a picture of ?
The light component still needs to be interpreted. Recall that light components merely
identify their verb as belonging to a certain verb class. In this case say is categorized as a manner-
of-speaking verb (M
), yet it does not have an interpretable manner. This is why its meaning
component can be identied with the required complement, i.e., the utterance itself.
aspectual focus
Following Gruber : see is analyzed as an abstract verb of
motion. Thus, one sees to a certain location. This means that see has an
L-component and it projects the structure in ():
We here depart from the traditional view of verbs such as paint (in both its uses) as change-
of-state verbs.
In the projection of the non-creation use of paint, the fence is associated with the
V ()
L a picture
Here only one meaning component is involved, the location or endpoint
of seeing, namely the picture. Aspectual focus therefore can only fall on
the object of seeing. Here the relationship between aspectual and infor-
mational focus comes into play. If the verb is (informationally) focused,
then the focus will be on whatever is aspectually focused within it. In
this case, there is only one option, namely the object, and so extraction
from the object is possible.
The verb paint has two meaning components, M (paint-matter) and
an L-component.
The same two meaning components render both the
interpretation in which paint is applied to a surface and the creation
interpretation in ().
V (M)
What is special about () is the fact that the theme picture instanti-
ates the M-component in that this artefact consists of paint-matter. The
presence of this theme therefore allows the identication of the picture
and the M-component very much in the same way as a lisper identies
:oo aspectual focus
the M-component of the verb lisp as argued with respect to (oc) above. In
this way the M-component of paint is rendered light in the context of a
theme artefact. In addition, the L-component is identied as the current
stage rendering the interpretation that the picture comes into being. In
other words, paint is applied to create a picture. The association of the
picture with the M-component also blocks aspectual focus on the lower
V-projection as shown in (I,).
(I,) a John painted a picture in an hour.
b *John painted a picture for an hour.
Here the creation use of the verb diers from the non-creation use in (I).
(I) a John painted the fence in an hour.
b John painted the fence for an hour.
The M-component of the verb is thus rendered light which explains
both extraction and incompleteness.
It should now be pretty obvious why it is that the verb burn blocks
extraction from its picture NP complement. Burn is an M, S verb like
chop and therefore projects the same structure (,c). This structure does
not aord a way to lighten the M-component which therefore remains
active and extraction is blocked. The M-component can, however, be
manipulated by context. Imagine, for example, that each member of a
group is ordered to burn a book. In such a context (I,) is licensed.
(I,) Who did you burn a book about?
Here the M-component of the verb burn is contextually bound making it
light as required for extraction.
Extraction is thus blocked in the presence of heavy M-components.
Extraction is licensed when the verb does not contain an M-component as
was the case with the verb see and also when the M-component is rendered
light, either by the projected structure or contextually. The opposite facts
hold for completeness. There an active M-component is required to make
the predicate focusable.
5.2.3 Datives
An observation due to Erteschik-Shir I,,, is that the dative object in dou-
ble object constructions cannot be extracted whereas it can be extracted in
its PP form:
These sentences should be interpreted in their creation use and not as referring to a spe-
cic picture which is being copied or lled in.
aspectual focus
() a *Who did John give a book?
b Who did John give a book to?
This was argued there to be due to the IS properties of the two objects
in the double object construction in (b): The indirect object must be
interpreted as a topic and the direct object must be focused.
() John gave [Mary
[the book]
This f-structure assignment can now be derived from the projection
of the verbs meaning components which are M and L. The particular L-
component of the verb give is a goal location instantiated by the preposi-
tion to. These meaning components project the following two causative
Similar accounts of double object constructions since Larson abound. For discussion
of a selection of these, see Erteschik-Shir and Rapoport in preparation. The preposition with,
following Hale and Keyser , represents the preposition of central coincidence. See their
work for details.
V (M)
D (L)
P (L)
V (M)
aspectual focus
Let us now examine the possibility of assigning a subordinate f-structure
to the lower V-projections in the two structures.
Extraction possibilities
depend on the focusability of the elements themselves, hence it is necessary
to uncover the f-structure properties of the two objects in each structure
in order to predict their potential for extraction. In (a) no constraints on
the assignment of (subordinate) f-structure are imposed. Both arguments
can be topics or foci. (The unmarked assignment will be one in which the
book is a topic and the PP to Mary is the focus.) Alternatively, to Mary
can function as a topic, and the book will then be a focus. This is due to
the fact that to Mary is identied with the L-meaning component, a goal
location and therefore a potential stage topic. The following examples, in
which a pronoun xes the topic and an indenite the focus, illustrate that
such assignments are indeed possible:
() a John gave it
[to a girl]
b John gave [a book]
[to her]
(b), however, is more restricted. Here the L-component is identied
with the theme rendering the possessive interpretation in which the
possessor Mary is the locus of the book. This possessive interpreta-
tion is employed to account for the f-structure of double object con-
structions. Possession constructions are individual level (see Chapter ,
section .) and cannot involve a stage (or alternatively the possessor is
itself the stage location).
Therefore, the PP (with) book is not a pos-
sible stage, and also not a possible topic. This is also evidenced by the
xing of topic and focus as before:
() a John gave her
[a book]
b *John gave [a girl]
Such an assignment is not required. It is possible, for example, to assign focus to the whole
VP in both structures (and indeed both arguments can be indenite in both).
Both objects can be also be topics:
(i) John gave it to her.
For a similar view of the difference between the two alternative forms see Basilico .
See also Clifton and Frazier for processing experiments that show that processing dou-
ble object constructions is facilitated by having a denite argument preceding an indenite.
Processing of NP-PP sequences is not facilitated by ordering denites and indenites in this
way. These results are predicted by the view that only the double object construction xes
Certain dialects of English allow (i) with both pronouns pronounced incorporated into
the verb:
(i) John gave+her+it.
aspectual focus
Contrast on the indirect object is of course possible as is the case for
any topic:
() John gave a GIRL (not a BOY) the book.
In English, the association of a subordinate f-structure with the lower
V-projection of the double object construction is rigid. This is not the
case across languages since the projection in (b) is not available cross-
linguistically. In languages in which it is not, the option of two dierent
orders for the two objects is determined simply by the need for topics
to precede foci. This is the case in many languages including Hebrew,
Finnish, and Russian.
Prosodic properties of languages may also play a
role as argued in Bring for German.
5.3 Missing objects
Goldberg : cites the following examples to illustrate object
() a The chef-in-training chopped and diced all afternoon.
b Tigers only kill at night.
c The singer always aimed to
d Pat gave and gave, but Chris just took and took.
e These revolutionary new brooms sweep cleaner than ever.
Together with others Goldberg points out that the missing object must
be interpreted as non-specic and must be predictable from context.
Another characteristic of constructions with missing objects, according
to Goldberg, is that they describe either iterative ( (a), (d) ) or generic
actions ( (b), (c), (e) ).
She argues that these properties follow from
the status of the missing object which must be low in discourse promi-
This status of the missing object enables attention to be shifted to
The f-structure that gives rise to (i) is one in which the VP is focused and in which both object
arguments are topics. No subordinate f-structure is assigned to the lower V-projection by itself.
The reason this string is not possible in most dialects has to do with the differences among dia-
lects as to phonological incorporation properties, an area in which dialects tend to vary. (See
Erteschik-Shir a on this general topic.)
For a discussion of Finnish ditransitive verbs see Kaiser and for Russian ditransitives
see Erteschik-Shir and Strahov .
(d) is not only iterative, it is also habitual. See section . for a discussion of missing
objects in habituals.
Goldberg makes the same point and refers to constructions in which the patient
argument is missing as the deproled object construction.
:o aspectual focus
the action. Goldberg argues that the missing object is neither a focus nor a
topic. That a missing element cannot be a focus is fairly self-evident. That
it is not a topic is less so, since topics are also non-prominent, and can in
some languages be dropped. Still, it would be strange if English allowed
object topics to drop when it does not allow subject topics to drop, a much
more common phenomenon across languages. Goldberg makes the point
that topics provide discourse antecedents, whereas omitted objects do not.
This is illustrated in (:,) (from Goldberg :ooI: ,II).
(:,) The chef-in-training chopped and diced all day. *They were put
into a large salad.
(:) Tigers only kill at night. *They are easily caught at that time.
If Goldbergs analysis is correct, then the construction is motivated by IS
factors yet a special IS category (high or low in discourse prominence)
has to be dened to account for it. Goldberg rejects the idea that what
licenses object omission is that the focus is on the activity itself as
proposed by Rice I,88: :oo. One reason this idea does not hold up,
according to Goldberg, is that object omission should be equally good
if another element in the sentence were focused. This is not the case
as shown in (:,) (from Goldberg :ooI: ,I,) in which the verb must be
repeated, and thus focused, even in the presence of a focused adjunct.
(:,) a They claimed that Alice killed ??(someone) YESTERDAY.
b She heard that the singer impressed *(an audience) last NIGHT.
Goldberg continues her argument with proof that object omission can
occur without the verb being focused:
(:o) A: When do these animals hunt?
B: Beavers kill during the day, but tigers only kill at night.
The focus in B, identied by the wh-question in A, is the time adverbi-
als, not the verb. Goldbergs argument, that the verb is not focused, is
well taken, yet all researchers working on object omission agree that the
It seems that the following sequence is acceptable:
(i) I already ate. It was delicious.
Such cases are rare, and the verb eat may be exceptional in allowing discourse reference to the
missing object. Eat also allows a telic particle:
(ii) Eat up! We have to go.
In both cases the missing object is accommodated.
aspectual focus :o,
activity expressed by the verb is in fact focused. This generalization must
therefore be expressed in some way. AT provides the necessary tool: the
activity is aspectually focused, yet does not function as an informational
focus as well. How this comes about is shown in section ,.I.
According to AT, however, every meaning component of a verb must be
interpreted. This is how the number of arguments selected by a particu-
lar verb is xed. Certain transitivity alternations are predicted, however.
Consider for example the transitivity alternations in (:,) and (:8).
(:,) a John broke the vase.
b The vase broke.
(:8) a John cut the bread.
b *The bread cut
Both break and cut are M,S verbs. There are in principle two dierent
ways in which these two meaning components can project. The transi-
tive projection was illustrated in (,c). There the M-component modied
the upper V-projection. The intransitive version is shown in (:,). There
the M-component modies the lower V-projection.
For a discussion of why the middle is possible with cut (e.g. This bread cuts easily) see
Erteschik-Shir and Rapoport :ooo; :oo.
the bread
V (M)
The reason break can project this structure but cut cannot lies in the
dierence between their M-components. Breaks manner is force and
cuts manner is sharp instrument. Both of these are possible modiers
of the transitive cause-event since the causer can employ force and can
also use a sharp instrument. In (:,), however, the M-component modi-
es a change-event. Such an event can be forceful, but cannot be with a
sharp instrument in the absence of the wielder of the instrument. This
is how the AT system rules out the projection in (:o) for cut, but not for
break. The AT theory, as it stands, therefore imposes strong restrictions
on whether or not an argument can go missing. These restrictions are
seemingly violated in cases of object omission.
:oo aspectual focus
5.3.1 Missing objects in Activities
Mittwoch :oo, divides cases of object omission into two types: object
omission in episodic sentences and object omission in habitual sen-
The rst type is illustrated in (,o), the second is discussed in
section ,.:.
(,o) a John is eating/reading/drinking.
b He is polishing/chopping.
(,oa) illustrates what Mittwoch refers to as the Activity use.
points out that these cases form activities in the aspectual sense and also in
the literal sense: they can be used to describe what a person is engaged or
occupied in doing at a particular moment or interval, just like verbs such
as work and rest that do not take objects at all. The omission of the object
in the Activity use does not require much contextualization and is there-
fore a standard use of the verbs involved. The examples in (,ob) dier in
this respect. They are not acceptable answers to What is he doing? as are
the examples in (,oa), but they can be uttered in a more specic context in
which cleaning and cooking, respectively, have been mentioned.
These cases of object omission provide a challenge to AT. Consider
rst the Activity use. The verbs listed by Mittwoch (:oo,: :,8; (:) ) all
seem to be M,L verbs. They include verbs of creation (paint, compose,
cook, bake), verbs of consumption (eat, drink), verbs of cleaning (iron,
mend, dust, sweep).
We have already seen how the L-component of
creation verbs functions. The L-component of consumption verbs has
the opposite function, in that the matter consumed is removed from the
current stage.
Similarly verbs of cleaning can also be viewed as verbs of
removal: Ironing involves the removal of creases, mending involves
This article is preceded by Mittwoch I,,I; I,8: which deal with different aspects of this
See Volpe :oo for interesting arguments that verbs such as eat in fact project an unerga-
tive structure and that their objects, when they appear, are cognate objects.
Goldberg :oo, lists verbs of emission, ingestion, and contribution as verb types which
allow object omission. She therefore excludes the verb sweep which does not belong to any of
these classes and predicts the ungrammaticality of (i):
(i) *Phil swept onto the oor.
According to Mittwoch, He is sweeping is a possible answer to What is he doing? which is why
for her this verb is included among the verbs which allow object omission.
Anita Mittwoch (personal communication) points out that the L-component of verbs of
consumption such as eat may in fact refer to the location where the food ends up. This is dem-
onstrated in (i):
aspectual focus
the removal of holes or tears, and dusting involves the removal of dust.
Now, according to Mittwoch these verbs are interpreted as activities
when their objects are omitted. In AT, intransitive activities are projected
by verbs with an M-component. This is illustrated in ():
(i) A: What happened to that sandwich?
B: I ate it.
It seems to me that the interpretation of Bs response in this context could be something like I
am responsible for the removal of the sandwich. This interpretation would be consistent with
the analysis of the verb as a verb of removal.
On a par with (), in which an M-component modies a change-event,
I propose that L-components which are bound by the current stage (and
only these) can also be interpreted as modifying activity structures. This
is shown in ().
V (L)
D (L)
In view of the fact that the L-component modies the activity structure,
it is necessarily aspectually defocused leaving only the activity interpret-
ation. Since the current stage is available in any context, there are no
contextual restrictions on object omission in these cases.
The examples in (b), according to Mittwoch : , require a
context which includes household chores like cleaning and cooking,
but even in such contexts they still feel elliptical or coerced. It seems
therefore that these M,S verbs can be accommodated when provided with
a cleaning or cooking stage to be interpreted as other removal and
creation verbs are. Yet they still feel elliptical since their S-component is
left uninterpreted.
Another verb discussed by Mittwoch is the verb build. This verb also
does not allow the omission of its object without further ado, even
though it is a creation verb. This is shown in the examples in (), also
from Mittwoch.
aspectual focus
() a ??He is building.
b John is building on the empty lot at the bottom of the road.
Mittwoch comments that build diers from the creation verbs such as
cook, which allow object omission without a special context, in that it
takes much longer to build something than it takes to prepare a meal,
and that maybe this explains why it is harder to conceive of building as
an activity. As support for this intuition she notes that (a) can only
be understood as referring to (a child) building with building blocks.
It might also be that the M-component of build is the light verb in the
class of construction verbs, and therefore it cannot be aspectually
focused as required for projection as an activity. In (b), however, a
location is added which on the one hand binds the L-component of the
verb, and on the other supplies a focus.
5.3.2 Object omission in habituals
The examples in () are all from Mittwoch.
() a She cleans and polishes all day.
b In Mediterranean countries they build mainly on the hilltops.
c Psychiatrists certify as a last resort.
d They murdered, raped, and plundered.
e You only take, you never give.
All the verbs that allow object omission in the Activity use also allow
object omission in habituals.
Habitual sentences permit a much larger
range of verbs to drop their objects. This follows, according to Mittwoch,
from the fact that they generalize over an unlimited number of situa-
tions, and therefore also over an unlimited number of instantiations
of the denotee of the missing object, e.g. houses or villages for [b],
people for [c]. It is therefore natural that the missing object should
As Mittwoch mentions, the verb cook is similar to build in that they both involve many
different activities. Yet cook does allow object omission easily. The reason the analysis of build
as a light construction verb does not carry over to cook may have to do with the shorter time
frame of cooking.
Mittwoch notes that the meaning of object omission in the Activity use can differ from
that in the habitual use. Compare (i) and (ii).
(i) He is writing.
(ii) He writes.
(i) as an answer to What is he doing? denotes making marks with a pen, chalk, etc.; (ii) could
involve using a typewriter, a computer, or even dictation. Similar for drink which in the habit-
ual use is restricted to alcohol.
aspectual focus
be understood as a bare plural (). Event plurality is also achieved
by conjunction as in (d) and also by iteration as illustrated in (d).
According to Mittwoch contrast, illustrated in (e), is the best enabler
of object omission. Mittwoch argues that the purpose of object omis-
sion in such cases is to make the verb prominent. Mittwoch concludes
that the missing object is a phonologically null pro-NP with the features
[+plural, human]. Alternatively, she suggests, one could claim that,
depending on context, object arguments could be saturated in the lexi-
con rather than in the syntax.
In order to show how Mittwochs insights can be incorporated
into AT, we must rst briey examine how plurality functions in this
framework. In Erteschik-Shir and Rapoport a, we show that each
meaning component also has a plural variant. Incremental verbs such
as cool and advance have plural S- and L-components, respectively,
because there is a plural number of increments included in the events
they describe. Similarly, iterative verbs such as ash and beep are also
plural. This is why these achievement verbs allow modication by for
adverbials. As is well known, the same eect is achieved by a plural
theme as shown in ().
() a The soup cooled for an hour.
b Glasses /*a glass broke for an hour.
In the same paper, we show that lexical plural also can aect lexical
structure in that plural verbs, although only consisting of a single (plu-
ral) meaning component, can also project a transitive accomplishment
() a John cooled the soup.
b The ocer advanced the regiment.
We argue that it is the plural which modies the V-V structure and ren-
ders the meaning that the agent initiates the plural event.
The availability of a plural lexical component ts very nicely with
Mittwochs conclusion if we argue that what licenses the omission of the
object in habituals is not a syntactic null proform, but rather the lexical plu-
ral which projects the missing theme argument as shown in () for (a).
This is an extension of the approach in Erteschik-Shir and Rapoport a, where the
plural component does not in fact project a theme on its own, but only as part and parcel of a
plural DP.
aspectual focus
The projection of the lexical plural is licensed, following Mittwoch, only
in cases of habituals. Note that there is no need for the feature [human]
since the value of this feature of the theme can be derived from the par-
ticular meaning components of the verbs. Mittwochs two alternative
accounts, the syntactic proform and lexical saturation, are both instanti-
ated in the analysis proposed here.
One more property of these constructions should still be discussed
here and that is the fact noted by Mittwoch that contrast is the best
enabler of object omission in habituals (e.g., (e) ). This follows from
the generic nature of these sentences and that therefore they cannot be
predicated of a stage topic. Since the object is unavailable, only the sub-
ject remains as a candidate for topichood. If the sentence is unmodi-
ed by adjuncts, then the focus must fall on the verb. Yet the activity
indicated by the M-component is also unavailable in habituals, since no
particular instance of an activity can be focused on. Focusing on a single
verb therefore does not provide a sucient focus. Contrast, iteration,
conjunction of several verbs, or an adjunct are therefore necessary to
supply the sentence with a focus.
5.4 Contextual binding of lexical constituents
The idea that components can be bound by the current stage, proposed
in section ., also supplies an explanation of the unexpected stress pat-
tern of intransitive unaccusatives when they are uttered out-of-the blue
discussed in Chapter , section .. Compare the sentences in () to
those in ():
() a The boy LAUGHED.
b The baby CRIED.
c The babys CRYING.
V ()
aspectual focus
() a A BOY left.
b My BIKE broke.
c The BABYs crying.
In view of the fact that all these sentences are all-focus sentences, they
should, according to the stress rule, be pronounced with stress on the
whole sentence, with highest stress on the nal constituent, in this case
the verb. The sentences in () are pronounced as expected with some
stress on the subject and highest stress on the verb.
The sentences in (), however, are unexpectedly pronounced with
stress only on the subject and so seemingly present a mismatch between
prosodic prominence and focal interpretation. (a) and (b) illustrate
change-events and are explained in the same way. (a) projects the
structure in ().
a boy
Out-of-the-blue change-events are used to answer the question What
happened? and therefore presuppose that some change-event has
occurred (for example: I hear a door slam and ask what happened and
am presented with the answer in (a) ). The parameters of the stage
topic are therefore identied with those of the change-event and in this
sense the stage topic binds the change-event as indicated in (). Note
that the particular changed state or location specied is still aspectually
focused (in this case a change away from the current stage). The verb
therefore plays a conicting role: it expresses both the change (topic)
as well as the particular change (aspectual focus). Since the stress rule
applies to sentential foci and not to aspectual ones, the verb is excluded
from stress assignment as is any topic embedded in a focus as argued in
Chapter , section .. The stress rule will therefore apply solely to the
subject rendering the desired stress pattern. As is the case with any topic,
the verb can, of course, be contrastively stressed as in ():
() What happened? A child LEFT.
Whereas change-events can naturally answer the question What happened?,
activities such as () do not. This is because this question, in addition to
aspectual focus
forcing an out-of-the-blue context, also presupposes that an event occurred
as just argued. Such a presupposition is generally incompatible with the
activity reading assigned to unergatives. Yet, depending on context, some
unergatives can in fact occur in the unexpected stress pattern as shown in
(c). In AT the projected structure is restricted only by the requirement
that all components are interpreted. Consider therefore that the unergative
(c) projects a change structure such as ().
For a more detailed account see Erteschik-Shir and Rapoport in preparation. For a differ-
ent account of these data see Zubizaretta .
the baby
This structure not only imposes a change-event interpretation, but also
imposes an interpretation in which the crying manner is interpreted as a
state. Such an interpretation is appropriate in the case of babies since they
enter a crying-state naturally and the sentence then means: The baby is
now (changed its state to) crying. The fact that this projection is licensed
for this verb in the relevant context accounts for the fact that the sentence
can occur in the intonation pattern associated with the change structure.
Eventive all-focus sentences are thus uttered in a context in which some-
thing has happened: a loud noise, an expression of horror or pleasure, etc.,
i.e., a reaction to a contextual event. Such a context may also prompt an
explicit question such as What happened? or What was that? Both questions
presuppose the occurrence of an event. This event (together with the cur-
rent spatio-temporal parameters) constitutes the topic of the so-called
all-focus sentence.
That lexical properties of verbs interact with IS has been known for
some time. The exact nature of this interaction has been hard to dene.
This is partially due to the fact that neither theories of IS nor theories
of the lexicon are geared to probe the ISlexicon interface. AT theory is
designed for this purpose, in that it employs lexical decomposition in a
way that allows dierent assignments of aspectual focus which in turn
interact with IS. Although this theory could only be supercially out-
lined here, its potential to elucidate the lexiconIS interface has hopefully
been made clear through the investigation of extraction, completeness,
missing objects, and the intonation assigned to eventive sentences.
The Division of Labor between
Syntax and IS
Research into the syntaxIS interface has been hampered by the fact that
no agreement has been achieved among researchers as to what a theory of
IS should look like to the extent that denitions of the basic IS notions
are still controversial. This is not to say that the role of IS in determining
syntactic properties such as word order has not been recognized, yet the
precise nature of how this comes about is still quite vague across linguis-
tic frameworks. In view of this state of aairs, I thought it best to try to
promote research in this area by emphasizing the particular methodol-
ogy needed to sharpen both the theory of IS as well as its application
to syntactic questions. In so doing I have also attempted to show how
far-reaching the purview of IS is within grammar. Yet although some
theories view IS as an integral part of grammar, others relegate it to the
interfaces. Within minimalism, for example, the Rizzi-type approach to
the left periphery discussed in Chapter integrates topic and focus fea-
tures into the grammar, while other minimalists (e.g., Reinhart ,
Chomsky in Stemmer ) view IS as operating on the interfaces.
Chomsky expresses his view as follows:
I suppose it is possible to argue that the computational-representational system accesses
features of language use, though what such a system would look like, I have no idea.
Suppose, for example, we consider the (plainly correct) fact that in a linguistic inter-
change, new/old information is a matter of background that participants assume to be
shared (what is some-times misleadingly called discourse; there need be no discourse
in any signicant sense of that term). Suppose further (as appears to be correct) that old/
new information relates to displacement effects in narrow syntax. And suppose further
(merely for concreteness) that we take these displacement effects to be expressed in nar-
row syntax by transformational operations. Should we then say that the operations of
object-shift, topicalization, and so on literally access shared background information?
This seems close to incoherent; any clarication of these intuitive ideas that I can think of
yields computational systems of hopeless scope, compelling us to try to formulate what
amount to theories of everything that cannot possibly be the topic of rational inquiry.
division of labor between syntax and is
A more reasonable approach, I think, is to take the operations to be autonomous, i.e.,
syntax in the broad sense, and to understand pragmatics to be a theory concerned with
the ways properties of expressions (such as displacement) are interpreted by language-
external (but person-internal) systems in terms of old/new information. That leaves us
with manageable and coherent questions. (Stemmer : )
Neither of these approaches can accommodate the full array of ISsyn-
tax interactions for which evidence has been presented here. Whether IS
should in fact be integrated into grammar or not must therefore still be
left as an open question within this framework.
A possible way of investigating the issue is to examine universal
properties of IS, beyond the accepted principle that topics tend to
precede foci. An initial attempt at such an investigation is oered in
Chapter . There it was argued that the constraint on dependencies
must be couched in terms of unmarked f-structure, where unmarked
f-structures are parameterized across languages. Since the denition
of an unmarked f-structure depends on the mapping of IS onto syn-
tactic structure, it is hard to envision how such a mapping can take
place without integrating IS as part of grammar. And, even if this were
possible, detaching IS from grammar would also remove constraints
on word order and on dependencies from the purview of syntax, leav-
ing very little work for syntax itself.
Another issue which pertains to this question is the role of processing,
also discussed in Chapter . Constraints on word order and constraints
on dependencies were dened as processing constraints, and processing
has generally been considered to be an interface property and again not
something that is part and parcel of grammar proper.
One possible way of reconciling the two points of view, one forcing
IS outside of grammar, the other keeping it inside, is to consider IS as
part of the phonological computation. This is a plausible tack to take
since IS processing is what controls the linear order of elements in the
sentence and linearization is conceivably a phonological rule. This is in
fact the position argued for in Erteschik-Shir b where I propose the
following (uncontroversial) initial division of labor between syntactic
and phonological computation:
() Syntax Phonology
lexically determined morphologically determined
hierarchical linear
c-command adjacency, left/right edges
syntactic constituents prosodic constituents
division of labor between syntax and is
The factors listed under Phonology in () are not properties of syntactic
computation and should therefore not play a role in syntax. To the extent
that factors such as these do play a role in the account of word order, the
account cannot be syntactic, but must rather be phonological. The idea
that IS belongs in a post-syntactic component was discussed in Chapter
in the context of theories that derive focus assignment from syntactic
stress assignment, yet in those theories the purpose of such a compo-
nent was to leave syntax intact and to provide a home only for those IS
phenomena that resist a syntactic account. These proposals also had to
exclude from the post-syntactic component any phenomena with inter-
pretative import, since these would have to be computed by LF syntax.
In the framework presented in Erteschik-Shir , it was argued that
phenomena accounted for by LF syntax, such as quantier scope, can
be directly read o f-structure. This follows, as also shown in Chapter
, from the fact that topics take wide scope. If LF-syntax is unnecessary,
then it is possible that the output of a phonology which includes IS-pro-
cessing will be visible to both the conceptual-intentional and the articu-
latory-perceptual interfaces. Such a model would reect the fact that IS
contributes both to interpretation and to pronunciation. (See Chapter ,
section . for such an architecture.)
Incorporating IS into the grammar in this way means that topic and
focus features must also be integrated into the computational system.
In view of the fact that both features are marked across languages by
a variety of linguistic tools including morphology, intonation, and
word order, this is not surprising, yet as discussed in Chapter , it is
not obvious how the features are introduced into the computation. In
Erteschik-Shir and Lappin we argue that focusing is a non-modu-
lar process which identies the foregrounded constituent in represen-
tations of all modular systems including vision and (non-linguistic)
auditory perception. Focusing is therefore viewed as a general cogni-
tive mechanism which interacts with linguistic computation as it does
with other modes of perception. In that sense, it is external to gram-
mar. Yet the particular properties it takes on are specic to language
although these properties are to some extent parasitic on the proper-
ties of auditory perception: High pitch and amplitude, for example,
draw our attention to any sound, not just linguistic ones. One might
wish to examine theories of visual and auditory perception to see if
focusing is considered an integral part of the computation of these
types of percept. If so, linguists might want to adopt this view as well.
division of labor between syntax and is
The notion of topic is dierent. It does not play a role in non-
communicative perception. Since it is language specic, it is likely that
it is indeed an integral part of linguistic computation. Yet there are
also cognitive requirements on an appropriate use of topic. Jelinek and
Carnie : surmise that we may have a general cognitive sense of
whats new and whats old but languages dier in how they lexical-
ize this into formal features in the lexicon. Assume that Jelinek is right
and that the ability to distinguish whats old or to recognize what we
have encountered before is part of general cognition. Does it follow that
topichood is part of general cognition? The answer to this question is
probably negative. True, the denition of topics presupposes the ability
to make this distinction, yet this ability could conceivably be available
without it playing a role in language in the way topics do.
This question has been the focus of some research into language
acquisition starting with Baker and Greeneld , who have shown
that children are already able to distinguish new from old information
in their rst year. In single-word utterances, children report on or about
new information. At the two-word stage totally presupposed informa-
tion was also omitted. Baker and Greeneld view the newold distinction
as the perceptual/cognitive foundation for the topic-comment and the
presupposition-assertion distinctions in later language. Schaeer ,
however, argues that childrens pragmatic system is underdeveloped in
that children cannot always distinguish shared and non-shared infor-
mation. This, she argues, explains non-adult-like object scrambling in
Dutch children. For the same reason subject omission in Russian is over-
extended by children under the age of according to Gordishevsky and
Schaeer . Under this view children overextend topic assignment,
yet they have the capacity to correctly mark (by omission or scrambling)
the topics that they do identify. Dyakonova argues that informa-
tion structure seems to be acquired in a parallel rather than consecu-
tive fashion compared to syntax. She examines OV and VO order in
Russian child language. Object topics occur preverbally whereas object
foci occur postverbally in Russian. According to Dyakonova the child
exhibited sensitivity to these constraints and she explained those cases
in which non-topics were positioned preverbally as due to a misanalysis
by the child of the shared knowledge in the situation at hand. Her con-
clusion is therefore very much along the lines of Schaeer .
De Cat makes the point that children appear to have the required
competence to encode topics from the onset of the multiword stage. She
observes that children consistently dislocate non-pronominal subjects
division of labor between syntax and is :I,
in individual-level predicates in which the subjects must be interpreted
as topics, but such dislocation is not found in all-focus sentences. She
concludes that childrens spontaneous production of these construc-
tions indicates that they have the ability to identify and encode topics in
a target-like fashion.
Although the acquisition research is still inconclusive, the results so
far indicate that IS notions play a central role in language acquisition as
well. This does not come as a surprise to linguists working in frameworks
in which IS is integrated in the grammar. Van Valin and LaPolla I,,,: oI
put the point of view of such theories succinctly: Researchers investi-
gating language acquisition and development from the communication-
and-cognition perspective do not assume that the cognitive endowment
that enables children to learn language is autonomous. Rather, they take
the question of unique cognitive structures for language learning to be
an open question to be decided by empirical research and not by theo-
retical at. It seems that this perspective stands a better chance of arriv-
ing at an understanding of the cognitive functions of topic and focus
and their role in language acquisition, yet so far not much work has been
done on the acquisition of IS within integrated frameworks. One ques-
tion that looms high on the wish-list for future research is the question
of IS bootstrapping for language acquisition: If focusing is a general
cognitive mechanism which interacts with linguistic computation as it
does with other modes of perception, it may be an excellent candidate as
a bootstrapping mechanism for language acquisition.
Bever :oo,: ,,I, in the context of a discussion of functionalist explana-
tions of linguistic structures, poses the question: Why should a linguist
care about these abstruse potential extra-linguistic sources for linguis-
tic structures? and continues: I cannot dictate an answer, but I do
know that the truth shall set you free. And he concludes: we want to
understand the potential relations between linguistic and general cogni-
tive properties, when exploring the biological and ontogenetic bases for
either one.
Further research into the theory of information structure as well as
its acquisition is therefore called for. This book is intended to encourage
just that.
This page intentionally left blank
Aboh, E. O. (:oo). Left or Right? A View from the Kwa Periphery. In D. Adger,
C. De Cat, and G. Tsoulas (eds.) Peripheries: Syntactic Edges and their
Effects. Dordrecht: Kluwer, Io,,o.
Ahrenberg, L. (I,8,). A Formal Field Grammar. NLPLAB Research Report
LiTh-IDA-R-8,-, Department of Computer and Information Science,
Linkping University, Linkping.
Aissen, J. (I,,:). Topic and Focus in Mayan. Language o8(I): ,8o.
Andersen, T. (I,,I). Subject and Topic in Dinka. Studies in Linguistics I,(:):
Ariel, M. (I,,o). Accessing Noun-Phrase Antecedents. London: Croom Helm.
Arnon, I. et al. (:oo,). Rethinking Superiority Eects: A Processing Model.
Paper presented at the CUNY Sentence Processing Conference, University of
Bailyn, J. F. (:oo:a). Scrambling to Reduce ScramblingPart I. Glot
International o(): 8,,o.
(:oo:b). Scrambling to Reduce ScramblingPart :. Glot International
o(,): Io,:.
(:oo,). Does Russian Scrambling Exist? In S. Karimi (ed.) Word Order
and Scrambling. Malden, Mass.: Blackwell Publishing, I,o,o.
Baker, N. and P. Greeneld (I,88). The Development of New and Old
Information in Young Childrens Early Language. Language Sciences Io(I):
Basilico, D. (I,,8). Object Position and Predication Forms. Natural Language
& Linguistic Theory Io(,): ,I,,.
(:oo,). The Topic of Small Clauses. Linguistic Inquiry ,(I): I,,.
Bayer, J. (:oo,). That-Trace without Reference to the Subject. :oth
Comparative Germanic Syntax Workshop, Tilburg: invited lecture.
Beaver, D. (:oo:). Accommodating Topics. In H. Kamp and B. H. Partee
(eds.) Context-Dependence in the Analysis of Linguistic Meaning. Amsterdam:
Elsevier, ,,8o.
Beninc, P. and C. Poletto (:oo). Topic, Focus, and V:: Dening the CP
Sublayers. In L. Rizzi (ed.) The Structure of CP and IP: The Cartography of
Syntactic Structures,vol. ii. Oxford: Oxford University Press, ,:,,.
Bever, T. G. (I,,o). The Cognitive Basis for Linguistic Structures. In J. R. Hayes
(ed.) Cognition and the Development of Language. New York: Wiley, :,,,o:.
(:oo,). Deconstructing Functionalist Explanations of Linguistic
Universals. In A. Carnie, H. Harley, and M. Willie (eds.) Formal Approaches
to Function in Grammar. Philadelphia: John Benjamins, ,,,,:.
::o references
Birner, B. J. and G. Ward (I,,8). Information Structure and Noncanonical Word
Order in English. Amsterdam: John Benjamins.
Bolinger, D. (I,,,). Meaning and Form. London: Longman.
(I,,8). Asking More Than One Thing at a Time. In H. Hiz (ed.) Questions.
Dordrecht: Reidel, 8,Io,.
Bresnan, J. and J. Kanerva (I,8,). Locative Inversion in Chichewa: A Case
Study of Factorization in Grammar. Linguistic Inquiry :o: I,o.
Bresnan, J. W. (:ooI). Lexical-Functional Syntax. Oxford: Blackwell.
Breul, C. (:oo). Focus Structure in Generative Grammar: An Integrated Syntactic,
Semantic and Intonational Approach. Amsterdam: John Benjamins.
Broadwell, G. A. (I,,,). The Interaction of Focus and Constituent Order in San
Dionicio Zapotec. In M. Butt and T. H. King (eds.) Proceedings of the Lfg99
Conference, the University of Manchester. Stanford, Calif.: CSLI Publications.
Brdy, M. (I,,o). Some Remarks on the Focus Field in Hungarian. UCL
Working Papers :: :o::,.
Bring, D. (I,,,). Topic. In P. Bosch and R. van der Sandt (eds.) Focus:
Linguistic, Cognitive and Computational Perspectives. Cambridge: Cambridge
University Press, I:o,.
(:ooo). Lets Phrase It! Focus, Word Order, and Prosodic Phrasing in
German Double Object Constructions. In G. Mller and W. Sternefeld
(eds.) Competition in Syntax. Berlin: Mouton de Gruyter, IoI,,.
(:ooo). Focus Projection and Default Prominence. In V. Molnr and
S. Winkler (eds.) The Architecture of Focus. Berlin: Mouton de Gruyter.
Cann, R., T. Kaplan, and R. Kempson (:oo,). Data at the GrammarPragmatics
Interface: The Case of Resumptive Pronouns in English. Lingua II,(II):
Carnie, A. and H. Harley (:oo,). Introduction: Formalizing Functionalism.
In A. Carnie, H. Harley, and M. Willie (eds.) Formal Approaches to
Function in Grammar. Philadelphia: John Benjamins, I8.
Casielles-Surez, E. (:oo). The SyntaxInformation Structure Interface:
Evidence from Spanish and English. London: Routledge.
Chafe, W. (I,,o). Givenness, Contrastiveness, Deniteness, Topics and Point
of View. In C. Li (ed.) Subject and Topic. New York: Academic Press, :,,,.
(I,,). Discourse, Consciousness, and Time: The Flow and Displacement
of Conscious Experience in Speaking and Writing. Chicago: University of
Chicago Press.
Choi, H.-W. (I,,,). Optimizing Structure in Context: Scrambling and
Information Structure. Stanford, Calif.: CSLI.
Chomsky, N. (I,,I). Deep Structure, Surface Structure and Semantic
Interpretation. In D. Steinberg and L. Jakobovits (eds.) Semantics: An
Interdisciplinary Reader in Philosophy, Linguistics and Psychology. Cambridge:
Cambridge University Press, I8,:Io.
(I,,,). Reections on Language. New York: Pantheon.
references ::I
Chomsky, N. (I,,o). Conditions on Rules of Grammar. Linguistic Analysis ::
(I,,,). On Wh-Movement. In P. W. Culicover, T. Wasow, and A. Akmajian
(eds.) Formal Syntax. New York: Academic Press, ,II,:.
(I,8I). Lectures on Government and Binding: The Pisa Lectures. Dordrecht:
(I,,,). The Minimalist Program. Cambridge, Mass.: MIT Press.
(:ooo). Minimalist Inquiries: The Framework. In D. Michaels and
J. Uriagereka (eds.) Step by Step: Essays on Minimalist Syntax in Honor of
Howard Lasnik. Cambridge, Mass.: MIT Press, 8,I,,.
(:ooI). Derivation by Phase. In M. Kenstowicz (ed.) Ken Hale: A Life in
Language. Cambridge, Mass.: MIT Press, I,:.
(:oo). Beyond Explanatory Adequacy. In A. Belletti (ed.) Structures
and Beyond: The Cartography of Syntactic Structures, vol. iii. Oxford: Oxford
University Press, Io,I.
Cinque, G. (I,,o). Types of a'-Dependencies. Cambridge, Mass.: MIT Press.
(I,,,). A Null Theory of Phrase and Compound Stress. Linguistic
Inquiry :(:): :,,,,.
Clark, H. H. and T. Wasow (I,,8). Repeating Words in Spontaneous Speech.
Cognitive Psychology ,,: :oI:.
Clifton, C., Jr. and L. Frazier (:oo). Should Given Information Come before
New? Yes and No. Memory & Cognition ,:(o): 88o.
Croft, W. (:ooI). Radical Construction Grammar. Oxford: Oxford University Press.
Culicover, P. W. and M. Rochemont (I,8,). Stress and Focus in English.
Language ::(,,): I:,o,.
Dane, F. (I,,). Functional Sentence Perspective and the Organization of the
Text. In F. Dane (ed.) Papers on Functional Sentence Perspective. Prague:
Academia, Ioo:8.
Darnell, M., et al. eds. (I,,,). Functionalism and Formalism in Linguistics,
General Papers. Amsterdam: John Benjamins.
De Cat, C. (:oo:). French Dislocation. Ph.D. dissertation, University of York.
(:oo). Early Pragmatic Competence and the Null Subject
Phenomenon. In B. H. R. Bok-Bennema, B. Kampers-Manhe, and
P. Sleeman (eds.) Romance Languages and Linguistic Theory :oo:: Selected
Papers from Going Romance :oo:. Amsterdam: Benjamins, I,,:.
de Hoop, H. (I,,:). Case Configuration and Noun Phrase Interpretation.
Ph.D. dissertation, University of Groningen.
(I,,,). Scrambled Denites. In P. Dekker, J. van der Does, and H. de
Hoop (eds.) De Dag: Proceedings of the Workshop on Denites. Utrecht:
Utrecht Institute of Linguistics OTS, ,,,,.
(:ooo). Optional Scrambling and Interpretation. In H. Bennis
and M. Everaert (eds.) Interface Strategies. Amsterdam: Holland Academic
Graphics, I,,o8.
::: references
de Hoop, H. (:oo,). Scrambling in Dutch: Optionality and Optimality.
In S. Karimi (ed.) Word Order and Scrambling. Malden, Mass.: Blackwell
Publishing, :oIIo.
DeLancey, S. (:ooI). Lectures on Functional Syntax. University of Oregon. www.
Diderichsen, P. (I,oo). Helhed og Struktur. Copenhagen: C. E. C. Grads Forlag.
(I,o8). Elementr Dansk Grammatik. Copenhagen: Gyldendal.
Diesing, M. (I,,:). Indenites. Cambridge, Mass.: MIT Press.
Dik, S. C. (I,,,). The Theory of Functional Grammar, Part 1: The Structure of the
Clause. Berlin: Mouton de Gruyter.
Dryer, M. S. (I,,o). Focus, Pragmatic Presupposition, and Activated
Propositions. Journal of Pragmatics :o(): ,,,:,.
Dyakonova, M. (:oo). Information Structure Development: Evidence
from the Acquisition of Word Order in Russian and English. In
K. Bentzen (ed.) Nordlyd ,:.I: Troms Working Papers in Language
Acquisition, 88Io,.
Eckardt, R. (:oo,). Manner Adverbs and Information Structure: Evidence from
the Adverbial Modication of Verbs of Creation. In E. Lang, C. Maienborn,
and C. Fabricius-Hansen (eds.) Modifying Adjuncts. New York: Mouton de
Gruyter, :oI,oo.
. Kiss, K. (I,8,). Congurationality in Hungarian. Dordrecht: Reidel.
(I,,,). Discourse Congurational Languages: Introduction. In K. E. Kiss
(ed.) Discourse Congurational Languages. Oxford: Oxford University Press,
(I,,o). The Focus Operator and Information Focus. Working Papers in
the Theory of Grammar ,.
(I,,8). Discourse-Congurationality in the Languages of Europe.
In A. Siewierska (ed.) Constituent Order in the Languages of Europe. Berlin:
Mouton de Gruyter, o8I,:8.
(:oo,). Division of Labor in the Derivation and Interpretation of Focus
Constructions. GLOW, Lund.
(:oo). The EPP in a Topic-Prominent Language. In P. Svenonius (ed.)
Subjects, Expletives, and the EPP. Oxford: Oxford University Press, Io,:.
(:ooo). Focussing as Predication. In V. Molnar and S. Winkler (eds.)
The Architecture of Focus. Berlin: Mouton de Gruyter.
Emonds, J. (I,,o). A Transformational Approach to English Syntax. New York:
Academic Press.
(:oo). Unspecied Categories as the Key to Root Constructions. In
D. Adger, C. De Cat, and G. Tsoulas (eds.) Peripheries: Syntactic Edges and
their Effects. Dordrecht: Kluwer, ,,I:o.
Engdahl, E. (:ooI). The Role of Syntactic Features in the Analysis of
Dialogue. In C. Rohrer, A. Rodeutscher, and H. Kamp (eds.) Linguistic
Form and its Computation. Stanford, Calif.: CSLI Publications, III:.
references ::,
Erteschik-Shir, N. (I,,,). On the Nature of Island Constraints. Ph.D. disserta-
tion, MIT, Cambridge, Mass.
(I,,,). Discourse Constraints on Dative Movement. In T. Givon (ed.)
Syntax and Semantics: Discourse and Syntax. New York: Academic Press,
(I,8I). On Extraction from Noun Phrases (Picture Noun Phrases).
In A. Belletti, L. Brandi, and L. Rizzi (eds.) Theory of Markedness in
Generative Grammar: Proceedings of the 1979 Glow Conference. Pisa:
Scuola Normale Superiore di Pisa, I,o,.
(I,8:). Extractability in Danish and Pragmatic Principle of Dominance.
In E. Engdahl and E. Ejerhed (eds.) Readings on Unbounded Dependencies
in Scandinavian Languages. Sweden: Ume, I,,,:.
(I,8o). Wh-Questions and Focus. Linguistics and Philosophy ,: II,,.
(I,88). Topic-Chaining and Dominance-Chaining. In Y. Tobin (ed.)
The Prague School and its Legacy. Amsterdam: John Benjamins, I,,,.
(I,,:). Resumptive Pronouns in Islands. In H. Goodluck and
M. Rochemont (eds.) Island Constraints, Theory, Acquisition and Processing.
Dordrecht: Kluwer, 8,Io8.
(I,,,). The Dynamics of Focus Structure. Cambridge: Cambridge
University Press.
(I,,,). Focus Structure Theory and Intonation. Language and Speech
:(:,): :o,:,.
(:oo,a). The Sound Patterns of Syntax: The Case of Object Shift.
Theoretical Linguistics ,I(I/:): ,,.
(:oo,b). What Is Syntax? Theoretical Linguistics ,I(I/:): :o,,.
(:oooa). On the Architecture of Topic and Focus. In V. Molnr and
S. Winkler (eds.) The Architecture of Focus. Berlin: Mouton de Gruyter, ,,,,.
(:ooob). Whats What? In C. F. Gisbert Fanselow, Matthias Schlesewsky,
and Ralf Vogel (eds.) Gradience in Grammar. Oxford: Oxford University Press.
and S. Lappin (I,,,). Dominance and the Functional Explanation of
Island Phenomena. Theoretical Linguistics o: I8,.
(I,8,a). Dominance and Extraction: A Reply to A. Grosu.
Theoretical Linguistics Io: 8I,o.
(I,8,b). Under Stress: A Functional Explanation of Sentence
Stress. Journal of Linguistics I,: I,,,.
(I,8,). Dominance and Modularity. Linguistics :,: o,I8,.
and T. Rapoport. (I,,,). A Theory of Verbal Projection. In G. Matos
and M. Miguel (eds.) Interfaces in Linguistic Theory. Lisbon: APL/Edines
Colibri, I:,8.
(:ooo). Aspectual Focus. GLOW, Bilbao.
(:oo). Bare Aspect: A Theory of Syntactic Projection. In J. Guron
and J. Lecarme (eds.) The Syntax of Tense. Cambridge, Mass.: MIT Press,
:: references
Erteschik-Shir, N. and T. Rapoport. (:oo,a). Path Predicates. In N. Erteschik-Shir
and T. Rapoport (eds.) The Syntax of Aspect. Oxford: Oxford University Press.
eds. (:oo,b). The Syntax of Aspect: Deriving Thematic and Aspectual
Interpretation. Oxford: Oxford University Press.
(in preparation). The Atoms of Meaning: Interpreting Verb Projections.
Ben Gurion University.
and N. Strahov (:ooo). Focus Structure Architecture and P-Syntax.
UCL, London.
(:oo). Focus Structure Architecture and P-Syntax. Lingua II(,):
Fanselow, G. (:oo). The Mlc and Derivational Economy. In A. Stepanov,
G. Fanselow, and R. Vogel (eds.) Minimality Effects in Syntax. Berlin: Mouton
de Gruyter, ,,I:.
Featherston, S. (:oo:). Bridge Verbs and V: Verbs: The Same Thing in Spades?
Eberhard-Karls-Universitt, Tbingen.
(:oo,a). Magnitude Estimation and What It Can Do for your Syntax:
Some Wh-Constraints in German. Lingua II,(II): I,:,,o.
(:oo,b). Universals and Grammaticality: Wh-Constraints in German
and English. Linguistics ,(): oo,,I:.
Finkel, K. von (I,8,). Theticity in Generative Grammar. Papers on
Firbas, J. (I,o:). Notes on the Function of the Sentence in the Act of
Communication. Journal of Studies of the Philosophical Faculty of the
University of Brno Series A, Io: I,,8.
(I,o). On Dening the Theme in Functional Sentence Perspective.
Travaux linguistiques de Prague I: :o,8o.
Frey, W. (to appear). The GrammarPragmatics Interface and the German
Preeld. In V. Molnar and S. Winkler (eds.) The Architecture of Focus. Berlin:
Mouton de Gruyter.
Fukui, N. (I,,,). Parameters and Optionality. Linguistic Inquiry :: ,,,:o.
Gelderen, V. van (:oo,). Scrambling Unscrambled. Ph.D. dissertation, University
of Leiden.
Gernsbacher, M. A. (I,,o). Language Comprehension as Structure Building.
Hillsdale, NJ: Erlbaum.
and D. Hargreaves (I,88). Accessing Sentence Participants: The Advan-
tage of First Mention. Journal of Memory and Language :,: o,,,I,.
Gill, K.-H. and G. Tsoulas (:oo). Peripheral Eects without Peripheral Syntax:
The Left Periphery in Korean. In D. Adger, C. De Cat, and G. Tsoulas (eds.)
Peripheries: Syntactic Edges and their Effects. Dordrecht: Kluwer, I:I:.
Givn, T. (I,,o). Topic, Pronoun and Grammatical Agreement. In C. N. Li
(ed.) Subject and Topic. New York: Academic Press, I,:II.
(I,,,). From Discourse to Syntax: Grammar as a Processing Strategy.
Department of Linguistics. Los Angeles: UCLA.
references ::,
(I,,,). On Understanding Grammar. New York: Academic Press.
Givn, T. (I,8). Syntax: A Functional-Typological Introduction, : vols.
Amsterdam: John Benjamins.
(I,8o). The Pragmatics of Word Order: Predictability, Importance, and
Attention. Amsterdam: Benjamins.
Goldberg, A. E. (I,,,). Constructions: A Construction Grammar Approach to
Argument Structure. Chicago: University of Chicago Press.
(:ooI). Patient Arguments of Causative Verbs Can Be Omitted: The Role
of Information Structure in Argument Distribution. Language Sciences
:,(,): ,o,:.
(:oo,). Constructions, Lexical Semantics and the Correspondence
Principle: Accounting for Generalizations and Subregularities in the
Realization of Arguments. In N. Erteschik-Shir and T. Rapoport (eds.) The
Syntax of Aspect: Deriving Thematic and Aspectual Interpretation. Oxford:
Oxford University Press.
(:ooo). Constructions in Context. Oxford: Oxford University Press.
and F. Ackerman (:ooI). The Pragmatics of Obligatory Adjuncts.
Language ,,(): ,,88I.
Goodall, G. (:oo). On the Syntax and Processing of Wh-Questions in
Spanish. In B. Schmeiser, V. Chand, A. Kelleher, and A. Rodriguez (eds.)
WCCFL :, Proceedings. Somerville, Mass.: Cascadilla Press, IoII.
Gordishevsky, G. and J. Schaeer (:oo:). On Null Subjects in Child Russian. In
Y. Otsu (ed.) The Proceedings of the Third Tokyo Conference on Psycholinguistics.
Tokyo: Hituzi Syobo, II,,,.
Green, G. (I,8,). Pragmatics and Natural Language Understanding. Hillsdale,
NJ: Erlbaum.
Gregory, M. L. and L. A. Michaelis (:ooI). Topicalization and Left-Dislocation:
A Functional Opposition Revisited. Journal of Pragmatics ,,: Ioo,,oo.
Grice, P. H. (I,,,). Logic and Conversation. In P. Cole and J. Morgan
(eds.) Syntax and Semantics III: Speech Acts. New York: Academic Press,
Grimshaw, J. and S. Vikner (I,,,). Obligatory Adjuncts and the Structure of
Events. In E. Reuland and W. Abraham (eds.) Knowledge and Language:
Concepts in Linguistic Structure and Beyond. Dordrecht: Kluwer, I,,,.
Grosu, A. (I,8:). Extragrammatical Motivation for Certain Island Pheno-
mena. Theoretical Linguistics ,: I,o,.
Gruber, J. S. (I,o,). Look and See. Language ,(): ,,,,.
Guron, J. (I,8o). On the Syntax and Semantics of PP Extraposition. Linguistic
Inquiry II(): o,,,8.
Gundel, J. K. (I,,). The Role of Topic and Comment in Linguistic Theory. Ph.D.
dissertation, University of Texas, Austin.
(I,,,). Left Dislocation and the Role of Topic-Comment Structures in
Linguistic Theory. OSU WPL I8: ,:I,I.
::o references
Gundel, J. K. (I,88). Universals of Topic-Comment Structure. In
M. Hammond, E. A. Moravcsik, and J. R. Wirth (eds.) Studies in Syntactic
Typology. Amsterdam: John Benjamins Publishing Co., :o,,,.
(I,,8). Centering Theory and the Givenness Hierarchy: Towards
a Synthesis. In M. A. Walker, A. K. Joshi, and E. F. Prince (eds.) Centering
Theory in Discourse. Oxford: Clarendon Press, I8,,8.
(:oo,). Information Structure and Referential Givenness/Newness: How
Much Belongs in the Grammar? In S. Mller (ed.) Proceedings of the Hpsgo,
Conference, Michigan State University, East Lansing. Stanford, Calif.: CSLI
Publications, I:::.
(:ooo). Information Structure and Distinctions between Grammar and
Use: Some Evidence from Clefts in English and Norwegian. In V. Molnr
and S. Winkler (eds.) The Architecture of Focus. Berlin: Mouton de Gruyter.
Gussenhoven, C. (I,8). On the Grammar and Semantics of Sentence Accents.
Dordrecht: Foris.
Haider, H. (:oo). The Superiority Conspiracy: Four Constraints and a
Processing Eect. In A. Stepanov, G. Fanselow, and R. Vogel (eds.) The
Minimal Link Condition. Berlin: De Gruyter, I,,o.
and I. Rosengren (I,,8). Scrambling. Sprache und Pragmatik, 49. Lund.
(:oo,). Scrambling: Nontriggered Chain Formation in OV
Languages. Journal of German Linguistics I,(,): :o,o,.
Hale, K. (I,8:). Preliminary Remarks on Congurationality. Proceedings of
NELS: 8o ,o.
and J. Keyser (I,,,). On Argument Structure and the Lexical Expression
of Syntactic Relations. In K. Hale and J. Keyser (eds.) The View from Building
:o: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, Mass.:
MIT Press, ,,Io,.
(:oo,). Aspect and the Syntax of Argument Structure. In
N. Erteschik-Shir and T. Rapoport (eds.) The Syntax of Aspect: Deriving
Thematic and Aspectual Interpretation. Oxford: Oxford University Press.
Hale, K. L. (I,,:). Basic Word Order in Two Free Word Order Languages.
In D. Payne (ed.) Pragmatics of Word Order Flexibility. Amsterdam: John
Benjamins, o,8:.
and E. Selkirk (I,8,). Government and Tonal Phrasing in Papago.
Phonology Yearbook : I,I8,.
Halliday, M. A. K. (I,o,). Notes on Transitivity and Theme in English: Part :.
Journal of Linguistics ,: I,,:.
Hannay, M. (I,8,). English Existentials in Functional Grammar. Dordrecht: Foris.
Heim, I. (I,8:a). File Change Semantics and the Familiarity Theory of
Deniteness. Stanford University.
(I,8:b). The Semantics of Denite and Indenite Noun Phrases. Ph.D. dis-
sertation, University of Massachusetts, Amherst.
references ::,
Hirschberg, J. (I,,I). A Theory of Scalar Implicature. New York: Garland.
Hockey, B. A. (I,,8). The Interpretation and Realization of Focus: An
Experimental Investigation of Focus in English and Hungarian. Ph.D. disserta-
tion, University of Pennsylvania.
Hoek, K. van (I,,,). Conceptual Reference Points: A Cognitive Grammar
Account of Pronominal Anaphora Constraints. Language ,I(:): ,Ioo.
Hoji, H. (I,8,). Logical Form Constraints and Congurational Structures in
Japanese. Ph.D. dissertation, University of Washington.
Holmberg, A. (I,,,). Remarks on Holmbergs Generalization. Studia
Linguistica ,,(I): I,,.
(:ooo). Scandinavian Stylistic Fronting: How Any Category Can Become
an Expletive. Linguistic Inquiry ,I(,): ,8,.
Horvath, J. (I,8I). Aspects of Hungarian Syntax and the Theory of Grammar.
Ph.D. dissertation, UCLA, Los Angeles.
(I,8o). Focus in the Theory of Grammar and the Syntax of Hungarian.
Dordrecht: Foris.
(:ooo). Interfaces vs. the Computational System in the Syntax
of Focus. In H. Bennis, M. Everaert, and E. Reuland (eds.) Interface
Strategies. Amsterdam: Royal Netherlands Academy of Arts and Sciences,
(:oo,). Movement, Economy and the PF-Interface: The Case of So-
Called Focus-Movements. Ben Gurion University Io.,.
Hsiao, F. and E. Gibson (:oo,). Processing Relative Clauses in Chinese.
Cognition ,o(I): ,:,.
Huang, C.-T. J. (I,8). On the Distribution and Reference of Empty Pronouns.
Linguistic Inquiry I,(): ,,I,.
Hyams, N. and K. Wexler (I,,,). On the Grammatical Basis of Null Subjects in
Child Language. Linguistic Inquiry :(,): :I,,.
Ishii, Y. (:ooI). Presuppositional Eects of Scrambling Reconsidered. COE
International Symposium, Kanda University of International Studies.
Jackendo, R. (I,,:). Semantic Interpretation in Generative Grammar.
Cambridge, Mass.: MIT Press.
(I,,,). X'-Syntax: A Study of Phrase Structure. Cambridge, Mass.: MIT
(:oo:). Foundations of Language. Oxford: Oxford University Press.
Jacobs, J. (:ooI). The Dimensions of Topic-Comment. Linguistics ,,():
Jger, G. (I,,,). Topics in Dynamic Semantics. Ph.D. dissertation, University of
Munich, CIS-Bericht-,o-,:.
Jelinek, E. and A. Carnie (:oo,). Argument Hierarchies and the Mapping
Principle. In A. Carnie, H. Harley, and M. Willie (eds.) Formal Approaches to
Function in Grammar. Philadelphia: John Benjamins, :o,,8.
::8 references
Jensen, A. (:ooo). Sentence Intertwining in Danish: A Challenge to the RRG
Account? University of Copenhagen, Denmark.
Junghanns, U. and G. Zybatow (I,,,). Syntax and Information Structure
of Russian Clauses. In E.W.B. et al. (eds.) Annual Workshop on Formal
Approaches to Slavic Linguistics: The Cornell Meeting. Ann Arbor Michigan
Slavic Publications, :8,,I,.
Kadmon, N. (:ooI). Formal Pragmatics: Semantics, Pragmatics, Presupposition,
and Focus. Oxford: Blackwells.
Kaiser, E. (:oo:). The SyntaxPragmatics Interface and Finnish Ditransitive
Verbs. In M. van Koppen, E. Thrift, E. J. van der Torre, and M. Zimmermann
(eds.) Proceedings of Console IX, 2000.
Kamp, H. (I,8I). A Theory of Truth and Semantic Interpretation. In J. J.
Groenendijk, T. M. V. Janssen and M. Stokhof (eds.) Truth, Interpretation and
Information. Dordrecht: Foris, II.
Karimi, S. (:oo,). On Object Positions, Specicity, and Scrambling in Persian.
In S. Karimi (ed.) Word Order and Scrambling. Malden, Mass.: Blackwell
Publishing, ,II:.
Kathol, A. (:ooo). Linear Syntax. Oxford: Oxford University Press.
Kayne, R. S. (I,8). Connectedness and Binary Branching. Dordrecht: Foris.
Keenan-Ochs, E. and B. Schieelin (I,,o). Foregrounding Referents: A
Reconsideration of Left-Dislocation in Discourse. BLS :: :o,,.
Kempson, R., R. Cann, and J. Kiaer (:ooo). Topic, Focus and the Structural
Dynamics of Language. In V. Molnr and S. Winkler (eds.) The Architecture
of Focus. Berlin: Mouton de Gruyter.
W. Meyer-Viol, and D. Gabbay (:ooI). Dynamic Syntax: The Flow of
Language Understanding. Oxford: Blackwell Publishers.
Kenesei, I. (:ooo). Focus as Identication. In V. Molnr and S. Winkler (eds.)
The Architecture of Focus. Berlin: Mouton de Gruyter.
Kidwai, A. (I,,,). Word Order and Focus Positions in Universal Grammar.
In G. Rebuschi and L. Tuller (eds.) The Grammar of Focus. Amsterdam:
John Benjamins, :I,.
(:oo). The Topic Interpretation in Universal Grammar. In V. Dayal
and A. Mahajan (eds.) Clause Structure in South Asian Languages.
Dordrecht: Kluwer, :,,8,.
Kim, A. H. O. (I,88). Preverbal Focusing and Type XXIII Languages. In
M. Hammond, E. Moravcsik, and J. Wirth (eds.) Studies in Syntactic Typology.
Amsterdam: John Benjamins, I,,:.
Kim, S.-I., J.-H. Lee, and M. A. Gernsbacher (:oo). The Advantage of First
Mention in Korean: The Temporal Contributions of Syntactic, Semantic,
and Pragmatic Factors. Journal of Psycholinguistic Research ,,(o): ,,,I.
King, T. H. (I,,,). Conguring Topic and Focus in Russian. Stanford, Calif.:
references ::,
King, T. H. (I,,,). Focus Domains and Information Structure. Proceedings of
the LFG97 Conference. Stanford, Calif.: CSLI Publications.
Kluender, R. (I,,:). Deriving Island Constraints from Principles of
Predication. In H. Goodluck and M. Rochemont (eds.) Island Constraints:
Theory, Acquisition and Processing. Dordrecht: Kluwer, ::,,8.
(I,,8). On the Distinction between Strong and Weak Islands: A
Processing Perspective. In P. W. Culicover and L. McNally (eds.) Syntax and
Semantics: The Limits of Syntax. San Diego: Academic Press, :I,,.
(:oo). Are Subject Islands Subject to a Processing Account? In
B. Schmeiser, V. Chand, A. Kelleher, and A. Rodriguez (eds.) WCCFL 23
Proceedings. Somerville, Mass.: Cascadilla Press, IoI:,.
Kratzer, A. (I,8,). Stage-Level and Individual-Level Predicates. Papers on
Quantication. Amherst: Department of Linguistics, University of Massachusetts.
Krifka, M. (I,,:). A Compositional Semantics for Multiple Focus
Construction. In J. Jacobs (ed.) Informationsstruktur und Grammatik.
Opladen: Westdeutscher Verlag, I,,,.
Kuno, S. (I,,I). The Position of Locatives in Existential Sentences. Linguistic
Inquiry: ,,,,8.
(I,,o). Subject, Theme, and the Speakers Empathy: A Reexamination
of Relativization Phenomena. In C. Li (ed.) Subject and Topic. New York:
Academic Press, I,.
(I,8:). The Focus of the Question and the Focus of the Answer. Papers
from the Parasession on Nondeclarative Sentences from the Regional Meeting of
the Chicago Linguistics Society: :,o,,,.
(I,8,). Functional Syntax, Anaphora, Discourse and Empathy. Chicago:
University of Chicago Press.
(I,,I). Remarks on Quantier Scope. In H. Nakajima (ed.) Current
English Linguistics in Japan. Berlin: Mouton de Gruyter, :oI8,.
and K.-I. Takami (I,,,). Grammar and Discourse Principles: Functional Syntax
and Gb Theory. Left Periphery Deletion. Chicago: University of Chicago Press.
Kuroda, S.-Y. (I,,:). The Categorical and the Thetic Judgment (Evidence from
Japanese). Foundations of Language ,: I,,8,.
Ladd, D. R. (I,8o). The Structure of Intonational Meaning. Bloomington:
Indiana University Press.
Ladusaw, W. (I,,). Thetic and Categorical, Stage and Individual, Weak and
Strong. Proceedings from SALT IV, Cornell University.
Lambrecht, K. (I,,). Information Structure and Sentence Form. Cambridge:
Cambridge University Press.
(:ooo). When Subjects Behave Like Objects: An Analysis of the Merging
of S and O in Sentence-Focus Constructions across Languages. Studies in
Language :(,): oII8:.
Langacker, R. W. (I,8,). Foundations of Cognitive Grammar, vol. i. Stanford,
Calif.: Stanford University Press.
:,o references
Langacker, R. W. (I,,I). Foundations of Cognitive Grammar, vol. ii. Stanford,
Calif.: Stanford University Press.
(:ooI). Discourse in Cognitive Grammar. Cognitive Linguistics I:(:):
Larson, R. J. (I,88). On the Double Object Construction. Linguistic Inquiry
I,(,): ,,,,I.
Lasnik, H. and T. Stowell (I,,I). Weakest Crossover. Linguistic Inquiry ::():
Lebeaux, D. (I,88). Language Acquisition and the Form of the Grammar.
Ph.D. dissertation, University of Massachusetts, Amherst.
Li, C. and S. Thompson (I,,o). Subject and Topic: A New Typology. In C. Li
(ed.) Subject and Topic. New York: Academic Press, ,,,o.
Liberman, M. (I,,,). The Intonation System of English. Ph.D. dissertation, MIT,
Cambridge, Mass.
Lpez, L. (:oo,). Steps for a Well-Adjusted Dislocation. Studia Linguistica
,,(,): I,,:,I.
MacDonald, M. C. and M. Christiansen (:oo:). Reassessing Working Memory:
Comment on Just and Carpenter (I,,:) and Waters and Caplan (I,,,).
Psychological ReviewIo,: ,,,.
McLuhan, M. (I,o). Understanding Media: The Extensions of Man. New York:
MacWhinney, R. (I,,,). Starting Points. Language ,,: I,:o8.
Massam, D. (:oo:). Questions in Niuean. Paper presented at AFLA IX, Cornell
Maynell, L. A. (:oo,). Discourse Constraints on Extraposition from Denite
NP Subjects in English. Ohio State University, Columbus. www.ling.
Meinunger, A. (:ooo). Syntactic Aspects of Topic and Comment. Amsterdam:
John Benjamins.
Miller, P. H. (:ooI). Discourse Constraints on (Non)Extraposition from
Subject in English. Linguistics ,,(): o8,,oI.
Mittwoch, A. (I,,I). Idioms and Unspecied NP Deletion. Linguistic Inquiry :(:):
(I,8:). On the Dierence between Eating and Eating Something: Activities
Versus Accomplishments. Linguistic Inquiry I,(I): II,::.
(:oo,). Unspecied Arguments in Episodic and Habitual Sen tences.
In N. Erteschik-Shir and T. Rapoport (eds.) The Syntax of Aspect: Deriving
Thematic and Aspectual Interpretation. Oxford: Oxford University Press.
Miyagawa, S. (I,,,). Against Optional Scrambling. Linguistic Inquiry :8(I): I:,.
(:oo,). A-Movement Scrambling and Options without Optionality.
In S. Karimi (ed.) Word Order and Scrambling. Malden, Mass.: Blackwell
Publishing, I,,:oo.
references :,I
Molnr, V. (I,,8). Topic in Focus: On the Syntax, Semantics, Phonology and
Pragmatics of the So-Called Contrastive Topic in Hungarian and German.
Acta Linguistica Hungarica ,(I:): 8,Ioo.
Morgan, J. L. (I,,,). Some Interactions of Syntax and Pragmatics. In P. Cole
and J. L. Morgan (eds.) Syntax and Semantics: Speech Acts, vol. iii. New
York: Academic Press, :8,,o,.
Neeleman, A. and T. Reinhart (I,,8). Scrambling and the PF Interface.
In M. Butt and W. Geuder (eds.) The Projection of Arguments: Lexical and
Compositional Factors. Stanford, Calif.: CSLI Publications, ,o,,,.
Newmeyer, F. (I,8:). On Discourse-Based Syntax: A Critical Examination of
on Understanding Grammar by Tamly Givn. Journal of Linguistic Research
:(I): II:.
(I,,,). Some Remarks on the FunctionalistFormalist Controversy in
Linguistics. In M. Darnell, E. Moravcsik, F. Newmeyer, M. Noonan, and
K. Wheatley (eds.) Functionalism and Formalism in Linguistics, i: General
Papers. Amsterdam: John Benjamins, o,8o.
(:ooI). The Prague School and North American Functionalist
Approaches to Syntax. Journal of Linguistics ,,: IoI:o.
Nikolaeva, I. (:ooI). Secondary Topics as a Relation in Information Structure.
Linguistics ,,(I): I,.
Partee, B. H. (I,,I). Topic, Focus and Quantication. SALT I: :,,8o.
Paul, I. (:oo:). The Syntactic Encoding of Topic and Focus. Georgetown
Working Papers in Formal Linguistics :: :,oI.
(:oo,). Multiple Topics: Evidence from Malagasy. Austronesian Formal
Linguistics Association IX, Cornell Working Papers in Linguistics.
Payne, D. (I,,,). What Counts as an Explanation? Functionalism and
Formalism in Linguistics, i: General Papers. Amsterdam: John Benjamins,
Pesetsky, D. (I,8,). Wh-in Situ: Movement and Unselective Binding. In
E. Reuland and A. T. Meulen (eds.) The Representation of (in)Deniteness.
Cambridge, Mass.: MIT Press.
Pierrehumbert, J. (I,8o). The Phonology and Phonetics of English Intonation.
Ph.D. dissertation, MIT, Cambridge, Mass.
and J. Hirschberg (I,,o). The Meaning of Intonational Contours
in the Interpretation of Discourse. In P. R. Cohen, J. Morgan, and
M. E. Pollack (eds.) Intentions in Communication. Cambridge, Mass.:
MIT Press, :,I,II.
Polinsky, M. (:ooI). Information Structure and Syntax: Topic, Discourse-
Linking, and Agreement. Third Workshop on Spoken and Written Texts.
Austin: Texas Linguistic Forum.
Pollard, C. and I. A. Sag (I,8,). Information-Based Syntax and Semantics,
i: Fundamentals. Stanford, Calif.: CSLI.
:,: references
Poulsen, M. (:oo,). Sentence Processing and Grammaticality in Functional
Linguistics: Extraction in Danish as an Example. Ph.D. dissertation, University
of Copenhagen.
Prince, E. (I,8Ia). Toward a Taxonomy of Given-New Information. In P. Cole
(ed.) Radical Pragmatics. New York: Academic Press, ::,,o.
(I,8Ib). Topicalization, Focus-Movement, and Yiddish-Movement: A
Pragmatic Dierentiation. BLS ,: :,o.
(I,8o). On the Syntactic Marking of Presupposed Open Propositions. In
A. Farley, P. Farley, and K.-E. McCullough (eds.) Papers from the Parasession
on Pragmatics and Grammatical Theory, 22nd Regional Meeting. Chicago
Linguistic Society: :o8::.
(I,88). Discourse Analysis: A Part of the Study of Linguistic
Competence. In F. J. Newmeyer (ed.) Linguistics: The Cambridge Survey,
ii: Linguistic Theory: Extensions and Implications. Cambridge: Cambridge
University Press, Io8:.
(I,,,). On the Functions of Left-Dislocation in English Discourse.
In A. Kamio (ed.) Directions in Functional Linguistics. Philadelphia: John
Benjamins, II,.
(I,,8). On the Limits of Syntax, with Reference to Left-Dislocation
and Topicalization. In P. W. Culicover and L. McNally (eds.) Syntax and
Semantics: The Limits of Syntax. San Diego: Academic Press, :8I,o:.
Rahkonen, M. (:ooo). Some Aspects of Topicalization in Active Swedish
Declaratives: A Quantitative Corpus Study. Linguistics (I): :,(,,).
Raviv, A. S. (:oo,). Identifying and Processing Topicalization in Danish. MA thesis,
Ben-Gurion University, Beer Sheva.
Reinhart, T. (I,8I). Pragmatics and Linguistics: An Analysis of Sentence
Topics. Philosophica :,: ,,,.
(I,,,). Interface Strategies. Ots Working Papers. Utrecht: Utrecht
(I,,o). Interface Economy: Focus and Markedness. In C. Wilder,
H. M. Gaertner, and M. Bierwisch (eds.) The Role of Economy Principles in
Linguistic Theory. Berlin: Academic Verlag, Ioo,.
(:ooo). Interface Strategies: Reference-Set Computation. Cambridge, Mass:
MIT Press.
Reis, M. (I,,,). Extractions from Verb-Second Clauses in German? In U. Lutz
and J. Pafel (eds.) On Extraction and Extraposition in German. Amsterdam:
Benjamins, ,88.
Rice, S. (I,88). Unlikely Lexical Entries. In S. Axmaker, A. Jaisser, and
H. Singmaster (eds.) Proceedings of the Ith Annual Berkeley Linguistics
Society :o:I:.
Ritter, E. and S. T. Rosen (:oo). Topicalization, Focalization and the Syntax of
Not-Topics. University of Calgary and University of Kansas, manuscript.
references :,,
Rizzi, L. (I,,,). The Fine Structure of the Left Periphery. In L. Haegeman (ed.)
Elements of Grammar. Dordrecht: Kluwer, :8I,,,.
(:ooa). Locality and the Left Periphery. In A. Belletti (ed.) Structures
and Beyond. The Cartography of Syntactic Structures, vol. iii. Oxford: Oxford
University Press, ::,,I.
ed. (:oob). The Structure of CP and IP: The Cartography of Syntactic
Structures, vol. ii. Oxford: Oxford University Press.
Roberts, C. (I,,8). Focus, the Flow of Information and Universal Grammar.
In P. W. Culicover and L. McNally (eds.) The Limits of Syntax, Syntax and
Semantics :,. San Diego: Academic Press, Io,oo.
(:oo,). Uniqueness in Denite Noun Phrases. Linguistics and Philosophy
:o: :8,,,o.
Roberts, J. R. (:oo,). Scrambling in Persian: An RRG Approach. SIL
Rochemont, M. S. (I,8o). Focus in Generative Grammar. Amsterdam: John
Rodionova, E. V. (:ooI). Word Order and Information Structure in Russian
Syntax. MA thesis, University of North Dakota.
Rooth, M. (I,8,). Association with Focus. Ph.D. dissertation, University of
Massachusetts, Amherst.
(I,,:). A Theory of Focus Interpretation. Natural Language Semantics
I: ,,IIo.
Rosengren, I. (I,,,). The Thetic/Categorical Distinction Revisited Once
More. Linguistics ,,(,): ,,,,.
Ross, J. R. (I,o,). Constraints on Variables in Syntax. Ph.D. dissertation, MIT,
Cambridge, Mass.
Rudin, C. (I,8,). Aspects of Bulgarian Syntax: Complementizers and
Wh-Constructions. Columbus, Oh.: Slavica.
Sadeh-Leicht, O. (:oo,). Parsing Self-Embedding as Islands: A Phrasal
Approach. Utrecht Institute of Linguistics OTS.
and J. Ginzburg (:ooo). Interrogative Investigations: The Form, Meaning
and Use of English Interrogatives. Stanford, Calif.: CSLI Publications.
and C. Pollard (I,,). Head-Driven Phrase Structure Grammar. Stanford,
Calif.: University of Chicago Press and CSLI Publications.
et al. (:oo,). Processing Accounts for Superiority Eects. Stanford
Samek-Lodovici, V. (:oo,). When Right Dislocation Meets the Left-Periphery:
A Unied Analysis of Italian Non-Final Focus. Lingua. In press.
Sasse, H. (I,8,). The Thetic/Categorical Distinction Revisited. Linguistics :,:
:, references
Schaeer, J. (I,,,). Direct Object Scrambling in Dutch and Italian Child
Language. Ph.D. dissertation, UCLA, Los Angeles.
(:ooo). The Acquisition of Direct Object Scrambling and Clitic Placement:
Syntax and Pragmatics. Amsterdam: John Benjamins.
(to appear). Specic Language Impairment as Evidence for the Division of
Labor between Pragmatics and Grammar. Amsterdam: Benjamins.
Schmerling, S. F. (I,,,). Aspects of English Sentence Stress. Ph.D. dissertation,
University of Illinois.
Schulz, B. (:oo,). Tune inDrop Out: Harmonic Alignment as a Prerequisite for
German Topic Drop.

Schwarzschild, R. (I,,,). Givenness, Avoid F and Other Constraints on the
Placement of Focus. Natural Language Semantics ,(:): II,,.
Selkirk, E. (I,8). Phonology and Syntax. Cambridge, Mass.: MIT Press.
(I,,,). Sentence Prosody: Intonation, Stress, and Phrasing. In
J. A. Goldsmith (ed.) The Handbook of Phonological Theory. London: Basil
Blackwell, ,,oo,.
Sgall, P., E. Hajiov, and J. Panevova (I,8o). The Meaning of the Sentence in its
Semantic and Pragmatic Aspects. Dordrecht: Reidel.
Shimojo, M. (I,,,). Focus Structure and Morphosyntax in Japanese: Wa and Ga
and Word Order Flexibility. Ph.D. dissertation, SUNY, Bualo, NY.
Smith, C. S. (I,,I/,). The Parameter of Aspect. Dordrecht: Kluwer.
Steedman, M. (I,,I). Structure and Intonation. Language o,(:): :oo,o.
(I,,o). Surface Structure and Interpretation. Cambridge, Mass.: MIT Press.
(:oooa). Information Structure and the SyntaxPhonology Interface.
Linguistic Inquiry ,I(): o,8,.
(:ooob). The Syntactic Process. Cambridge, Mass.: MIT Press.
Stemmer, B. (I,,,). An On-Line Interview with Noam Chomsky: On the Nature
of Pragmatics and Related Issues. Brain and Language o8(,): ,,,oI.
Strahov, N. (:ooI). A Scrambling Analysis of Wh-Questions in Russian. In
Formal Approaches to Slavic Linguistics: The Bloomington Meeting, :ooo. Ann
Arbor: Michigan Slavic Productions, :,,,II.
Strawson, P. F. (I,o). Identifying Reference and Truth-Values. Theoria ,o: 8o,,.
Surnyi, B. (:oo). The Left Periphery and Cyclic Spellout: The Case of
Hungarian. In D. Adger, C. De Cat, and G. Tsoulas (eds.) Peripheries:
Syntactic Edges and their Effects. Dordrecht: Kluwer, ,,.
Svenonius, P. (:ooI). On Object Shift, Scrambling and the Pic. In E. Guerzoni
and O. Matushansky (eds.) MIT Working Papers in Linguistics. Cambridge,
Mass.: MIT, :o,8,.
(:oo). Subject Positions and the Placement of Adverbials. In P.
Svenonius (ed.) Subjects, Expletives, and the EPP. Oxford: Oxford University
Press, :oI:.
references :,,
Szabolcsi, A. (I,,,). Strategies for Scope Taking. In A. Szabolcsi (ed.) Ways of
Scope Taking. Dordrecht: Kluwer, Io,,,.
Szendri, K. (:ooI). Focus and the SyntaxPhonology Interface. Ph.D. disserta-
tion, University College London.
(:oo,). A Stress Based Approach to the Syntax of Hungarian Focus.
Linguistic Review:o: ,,,8.
(:oo). Focus and the Interaction between Syntax and Pragmatics.
Lingua II(,): ::,,.
and J. Costa (:oo,). Focus and the SyntaxPhonology Interface. GLOT
,(,/Io): :oI8.
Takano, Y. (I,,8). Object Shift and Scrambling. Natural Language and
Linguistic Theory Io: 8I,8,.
Taverniers, M. (:oo,). Subjecthood and the Notion of Instantiation. Language
Sciences :,(o): o,I,8.
Vallduv, E. (I,,o). The Informational Component. Ph.D. dissertation, University
of Pennsylvania, Philadelphia.
(I,,:). The Dynamics of Information Packaging. Manuscript: University
of Edinburgh Published (I,,) in E. Engdahl (ed.) Integrating Information
Structure into Constraint-based and Categorial Approaches. ESPRIT Basic
Research Project , Dynamic Interpretation of Natural Language. DYANA-
Deliverable R ..B. ILLC. University of Amsterdam, I:,.
and E. Engdahl (I,,,). Information Packaging and Grammar
Architecture. In J. N. Beckman (ed.) Proceedings of the North East Linguistic
Society 25, University of Pennsylvania. Amherst: GLSA, ,I,,,.
Van Valin, R. D. ed. (I,,,a). Advances in Role and Reference Grammar. 8:.
Current Issues in Linguistic Theory. Amsterdam: John Benjamins.
(I,,,b). A Synopsis of Role and Reference Grammar. In R. D. Van Valin
(ed.) Advances in Role and Reference Grammar. Amsterdam: John Benjamins,
(I,,,a). A Brief Overview of Role and Reference Grammar. http://wings.
(I,,,b). Generalized Semantic Roles and the SyntaxSemantics Interface.
In F. Corblin, C. Dobrovie-Sorin, and J.-M. Marandin (eds.) Empirical Issues in
Formal Syntax and Semantics 2. The Hague: Thesus, ,,,8,.
(I,,,c). A Typology of the Interaction of Focus Structure and Syntax. In
E. Raxilina and J. Testelec (eds.) Typology and the Theory of Language: From
Description to Explanation. Moscow: Languages of Russian Culture, ,II:.
(:oo,). Exploring the SyntaxSemantics Interface. Cambridge: Cambridge
University Press.
and R. J. LaPolla (I,,,). Syntax: Structure, Meaning and Function.
Cambridge: Cambridge University Press.
Villalba, X. (I,,8). Right Dislocation Is Not Right Dislocation. In D. O.
Fullana and F. Roca (eds.) Studies on the Syntax of Central Romance
:,o references
Languages: Proceedings of the III Symposium on the Syntax of Central Romance
Languages. Girona: Universitat de Girona, ::,I.
Volpe, M. (:oo). Aected Object Unergatives. Snippets 8.
von Heusinger, K. (I,,,). Intonation and Information Structure.
Habilitationsschrift, University of Konstanz.
Vries, L. de (I,8,). Topic and Focus in Wambon Discourse. In A. M. Bolkestein,
C. De Groot, and J. L. Mackenzie (eds.) Syntax and Pragmatics in Functional
Grammar. Dordrecht: Foris, I,,8o.
Wagner, M. (:oo,). Holmbergs Generalization in an OV Language.
Manuscript, MIT.
Walker, M. A., A. K. Joshi, and E. F. Prince. eds. (I,,,). Centering Theory in
Discourse. Oxford: Oxford University Press.
Ward, G. (I,88). The Semantics and Pragmatics of Preposing. New York:
and J. Hirschberg (I,8,). Implicating Uncertainty: The Pragmatics of
FallRise Intonation. Language oI(): ,,,o.
and E. F. Prince (I,,I). On the Topicalization of Indenite NPs. Journal
of Pragmatics I,(8): Io,,8.
Watters, J. R. (I,,,). Focus in Aghem: A Study of Its Formal Correlates and
Typology. In L. M. Hyman (ed.) Aghem Grammatical Structure. Los Angeles:
University of Southern California, I,,,,.
Wedgewood, D. J. (:oo,). Predication and Information Structure: A Dynamic
Account of Hungarian Pre-Verbal Syntax. Ph.D. dissertation, University of
Weskott, T. (:oo:). Topicalization in Context: Processing the Left Periphery
of German V:-Sentences. Conference on Information Structure in Context,
Stuttgart University.
Wexler, K. (I,,8). Very Early Parameter Setting and the Unique Checking Constraint:
A New Explanation of the Optional Innitive Stage. LinguaIoo: :,,,.
Williams, E. (:oo,). Representation Theory. Cambridge, Mass.: MIT Press.
Wiltschko, M. (I,,8). Superiority in German. In E. Curtis, J. Lyle, and
G. Webster (eds.) WCCFL Io: The Proceedings of the Sixteenth West Coast
Conference on Formal Linguistics. Stanford, Calif.: CSLI, ,I,.
Winkler, S. and E. Gbbel (:oo:). Review Article: Zubizaretta, M.L. (I,,8)
Prosody, Focus, and Word Order, Linguistic Inquiry Monograph ,,.
Massachusetts: MIT Press. Linguistics o(o): II8,:,.
Ziv, Y. and P. Cole. (I,,). Relative Extraposition and the Scope of Denite
Descriptions in Hebrew and English. In M. W. LaGaly, R. Fox, and A. Bruck
(eds.) Papers from the Tenth Regional Meeting of the Chicago Linguistic Society.
Chicago: University of Chicago, ,,:8o.
Zubizaretta, M. L. (I,,8). Prosody, Focus and Word Order. Cambridge, Mass.:
MIT Press.
A-bar movement 8o, I:,
Aboutness I,, IoI
Accent - see also pitch accent ,I,,
, n.,,, o,, III
accentual phrasing Ioo
accommodation ,, I,, ,,, o8, ,o, ,,, Io,
Io,, II8I,, Io, I,:, I,8, :o n. I8,
:o,, :I
activity I,,o, :o8, :Io, :I:
adverb I, n.I:, ,o, oo, o:, 8,, ,,, ,,,
Ioo, Io:, Io, III8, I,,, I,o,,
Io,, I,, I,oI, I,,, I8,, I,,, I,,,
:o, :o,
all-focus :, Io, :8, ,o, ,,, ,,, II,, I:8,
:III:, :I,
A movement I:,
antecedent III,, :I, ,:,, ,o, III, I,,,
I:, Io,, I,o, :o
architecture ,, ,, :o, ,, ,,, ,8, oo, o:,
o,,o, ,,, ,,, I,I, :I,
aspectual focus ,, I,, I,o:oo,
:o,8, :III:
assertion I, ,o, ,,oo, IIII:, I:,, :Io
background ,,, :,, ,o, Io, n.II, I,,
I,o, :I,
bare plural :o,
boundary tone , n.,,, ,I
canonical f-structure II,, I,, Io,,, I,I,
I,,, I,,,, I8I:, I8, I8o, I,I
cartography 8,, ,o, ,,, ,o n.I, ,8IoI,
Io,, Io,, IIo, I,,
categorical Io n.II, 8,, I8o
causative :oI
change of location I:,, I,,, :II
change of state I,,, I,, n.,, :III:
cleft ,, I, 8I, I:I
clitic Io n.,, ,o,, ,,o, II n.,I, I8:
Clitic Left Dislocation (CLLD) ,o, ,,
cognate object I,o, :oo n.:I
cognition o, ,I, :IoI,
Combinatory Categorial
Grammar (CCG) , ,, n.,o, ,,
common ground I,, ,, o
Communicative Dynamism (CD) :
Completeness o, ,,8, ,,, I88, I,,
:oo, :I:
configurational language ,, 8o,, 8,,
I:,, I,,
constituents I:, , :, :8:, , ,8,
, n.8, ,I, ,8o:, o, n.,,, o,, o,,
o8 n.o:, ,,, 8I, 8,, 8,,o, ,,, Io:,
Ioo, Io8 n.:,, Io,, III, II, II,, I:I,, I,:,
II, I, Io, I,, I,,, Ioo, Io, Io, n.II,
Io,, I8, n.,, I8,, I,o,, :IoII, :II,
alignment o,
functional ,
I-dependency Io,8, Io,, I,,, I8I,
I8, I88,o, I,,, :I
Identification (ID) I,I, I,o,
information structure motivated
Ioo,, I,,o,, I8,,, I,,
island ,8, Io,, I8,, I8, n.,,, I,,
on extraction ,, I,, Io,, I,:, I88
phonological ,,
processing ,:, I,, I,o, I,,, :I
syntactic ,, ,,oo, oI n.,,, ,, Io:,
Io,,, IoI, Io,, Io8, I8,, I,,
topic I,,, I88
Construction Grammar o,, ,I n.oo
context I:, , 8,, III:, I, n.,, II,,
:Io, :,, ,I, ,,, , n.,:, ,8, ,, o, 8,I,
,, n.,,, o,, o,8, ,oI, ,,8, 8,, 8,o,
,,8, ,, n.Io, Io:, Io,, Io,II, II,I,
IIo, II,:o, I:,, I,,, I,,o, I:,
I8, I,,8, IooI, Io,, Ioo,o, I,,o,
I,88o, I8,, I,:, I,, I,8, :oo, :o,,
:ooIo, :I:, :I,, :I,
context set I, ,, ,o, ,8
continued topic I: n., 8, Io, :, o, ,,
Io,o, IIo, I::
Index of terms
:,8 index of terms
contrast ,, ,8, :, 8, ,oI, ,I, ,,,
8o:, 8,o, ,, ,o,, Ioo,,
Ioo, III, IIo, I::, I:,, I,,, I, Io,
I8, n.,,, I,, :o,, :o,Io
contrast set ,, I: n.,, :,, :, 8,, o,, III,
contrastive focus :,, :, ,, 8I, 8,, ,o, ,o,
,8Ioo, Io,,, IIo, I:, n.,o, I,, Ioo
contrastive topic oI, ,,o, ,,,
,o, ,8, IoI:, Ioo, II, , I:,, I:,,o,
I,,, Io, I,, I,I
coreferent pronoun, see pronoun
d(iscourse)-linked ,, oI, Io8, I,,,, I,,,
I8I, I8,, I,oI
definiteness effect I:o
distressing ,8, oI:, I,:, I,, I,I
discourse configurational
language 8:,, 8,, I,,
discourse referent I,, :,, , o,
I,o, I,I
Discourse Representation
Theory (DRT) ,,, o8
d-linked, see d(icscourse linked)
double object construction I,:, I,, I,,
Dynamic Syntax o,, o,, ,o n.o
EPP ,,, Ioo, Io:, II,, II,
eventive sentences o, :I:
exhaustive focus :,
exhaustive identification 8o
existential ,, :,, 8I, II,, II,:I, I8o
extraction ,, ,,, ,8, II8, I,, I,,o,,
Io,,, I,I, I8,8, I,oI, I,,
I,,:oo, :o:, :I:
extraposition I,, I,o, I8,8, I,
f(ocus)-structure canonical,
see canonical f-structure
f(ocus)-structure main, see main
f(ocus)-structure restrictive,
see restrictive f-structure
f(ocus)-structure subordinate,
see subordinate f-structure
f(ocus)-structure boundaries ,
fall-rise tune ,I
figure-ground ,o,
file system II, o, ,:, ,, o,
focus aspectual, see aspectual focus
focus contrastive, see contrastive focus
focus exhaustive, see exhaustive focus
focus identificational,
see identificational focus
focus informational, see informational
focus narrow, see narrow focus
focus presentational, see presentational
focus preverbal, see preverbal focus
focus restrictive, see restrictive focus
focus sentential, see sentential focus
focus tests :8, ,,, ,o, I:o, Io
focus chaining ,, III:, :I, ,,, ,, ,
focus feature ,:,, ,,, ,8, oo, o,,
oo, 8o n.,, 8,, 8,, I,o, I,: n.oI,
:I,, :I,
focus marking (F-marking) ,:,, o:,
,,oo, 8I, 8,
focus preposing 8o n.,, IIoI:
focus projection ,:,, I,: n.oI
focus-definition I, :,,, ,8o, , Io,
focusability IoI, Io,,, I,,,
I8,,o, :o:
foreground ,,, I,o, :I,
functional features o,
Functional Grammar ,, o,
Functional Sentence Perspective :
Functional Syntax ,
Functionalism ,I,, ,8
Generic 8,, ::, I8, I,,, :o,, :Io
Given ,, ,I,, I8::, :,, :,, ,, 88, ,,, II,,
I:oI, I:,
ground I:, I,, ,o, ,o,
habitual :o,, :oo, :o8Io
Head Driven Phrase Structure
Grammar (HPSG) , ,,, o,
high tone ,I
Holmbergs Generalization I8, I,o
hypertheme :,, III:, I8I,,
:,, ,, Io,
I(dentificational)- dependencies Io,,,
Io,, I,,, I8I, I8,, I8,,o, I,,
index of terms :,,
Identification ,o, 8o,, I:,, I,I, I,o, I,,,
Io,, I,8, I,o, I,,
argument identification I,o
identification of IS functions I,,
Identification (ID) constraint,
see constraints
identificational focus 8o
I-level, see individual level predicate
inclusiveness ,,8, o:,, oo
incorporation ,I, ,:, I,o:, :o:,
indefinite topic :o
individual level predicate :I, ,o, ,, ,:,
II,, :I,
information packaging
information structure
primitives ,, :,, :
informational focus :,, 8I, 8,, ,oI, ,o,
,8, Ioo, I:, I:o, I,:, I,o, I,,, :o,
interface o, ,, ,,, ,,8, oI,, oo, ,:,,
,8, I:,, I:8, I,,, I,o, :I:I,
intonation (see also prosody) I:, ,
o, :, n.:,, ,,, ,,, o, ,, ,I, ,,, ,,,
,,, ,,, o,, o8, ,,, ,,8, 8I, 8,, II8,
I:8,o, Io, I,I, I,o,
I,,, :I:, :I,
intonational boundaries ,
intonational phrase , ,,, ,,
islands ,8, 8o, Io8, Ioo n.,, Io,, Io,,
Io,, I,I, I8,, I8,,I, I,,
iteration :o,, :o,Io
language acquisition o, :IoI,
left detachment IoII
Left Dislocation (LD) I,, 88, ,,,, Ioo,
Io8Io, II:I,
left periphery ,, ,, I,, oo, ,o, 8o, 8,,o,
,,, ,8, Ioo:, II,, IIo, I:,,, I,o, I,:,
II, I, Io, I,:, I,,, :I,
left-branching ,,
lexical decomposition o, :I:
lexical features o,
Lexical Functional Grammar (LFG)
, ,,, o,,, ,,, 8o, 8,, Io,
lexical structure I,, :o,
lexicon o, ,,, I,o, :o,, :I:, :Io
lie-test ,,, Io
light verbs I,,, I,,8, :o8
linearization oo, I:,,I, I,:,,
IoI, :I
Logical Form (LF) , :o, ,,, ,,
,,, ,8o,, oo,, ,o, I,I,
I,, I,8, :I,
long distance scrambling I:,
low tone ,I
main f-structure 8, ,I, I,o
main topic :,, ,I, I,I, II, I,:
manner-of-speaking verbs Io,, Io,,
I,, I,,, I,8 n.o
mapping between syntax and
information structure :, , :8,
Io,, I:o, I:,, I,, :I
meaning component o,, I,,, :oI:,
:o,, :o,Io
merge o:, o, n.,,, oo, Io:, II,, I:8
middle field Io:, II,
Minimalist Program (MP) , ,,o, o:,,
o,, ,o, Ioo, I:, :I,
multi-level models , ::, ,,, ,o
multiple topics ::, ,o, II8
multiple wh-questions ,o, o8, I,,o, I,,,
I8:, I8
narrow focus :,, ,: n.:,, o,, 8o,
I,,, I,,
negation II,, Io,
new information ,, II, I,, 8o, 88, ,8, Io,,
Io,, :I,I, :Io
nuclear scope ,o
Nuclear Stress Rule (NSR) ,o, ,8,
object shift ,o,, o:, 8o, I:8, I,oI,
I,,,, :I,
old information ,, Io, ,, :I,, :Io
Optimality Theory (OT) :, o,,o, ,,,
Io,o, Io, I,
out of the blue ,, I,I8, ,I, I:,, I,,, I,o,
I8:, I,, :III:
Parallel Grammar o,
parsing constraint I,o, I,8
partitive ::, :,, ,I, ,,, I,o
PF (Phonological Form) ,,o, ,8,,
o:,, oo, I::, I:,, I:8,I, I,I,, I,8
PF computation oo
phonological computation :I
:o index of terms
picture NPs I,, I,8
pitch accent ,o, ,I n. :,, ,:, ,I, Ioo
p-movement (prosodically motivated
movement) ,,oo, o:, I::
poset II, Io8II
potential focus domain o8, ,I, 8:, I,,
IoI,, Io,
PP-extraposition I8,
Prague School I:, , n.I, I,:
Pre-field (see also Vorfeld) 88
presentational focus :,
presupposition I, I:, I, :,,o, ,o, ,,, :,
8,,o, Io,, I,,, Io,, :I:, :Io
preverbal focus 8:, 8o,, Ioo, I:, I,:
Principles and Parameters , ,,, Io
processing ,, o, ,,, ,o, ,,, Io8, II,
I,, Ioo, Io,,, I,,, I,, n. :,, I,8,
I8, I8,,,, :o:, :II, see also
constraints (processing)
prominence ,, :o,, ,,, , o, ,8, oo,
8,,, ,,, Ioo,, III, II,, I::, I,,, Ioo,
I,, :o,, :o,, :II
pronoun ,, 8Io, I:, :I, ,, 8,, ,I, ,,
oo, ,o, 8,,,, ,,, ,,8, Io,, Io,8,
IIo, I,8, II, I8, I,oI, I,,, I,,,, Ioo,
Io:, I8o, :o:
pronoun coreferent ,, Io n.,, I,o
prosody , ,, ,,8, ,, ,8,, oo,,
8I, Io8, I::, I n.,,, I,o,, :o,,
:II, :I
quantifier scope ,,, I,,, :I,
question-answer pairs I:, II, :8, ,,, ,,
,, ,o, I:o
referent, referent discourse, see
discourse referent
Representation Theory ,,, , n.,:, o,,
o,, Ioo n.I,
restrictive clause ,o
restrictive focus ,,o, 8o,, ,I,
restrictive set ,,o, Io,, Io,, III
resumptive ,, ,I,, ,,,, Io,, Io8, I,:,
Ioo n.,
rheme :, :,
right detachment Io
right dislocation I,, I:,
right-branching ,,
Role and Reference Grammar (RRG)
,, ,,, o,8, ,I n.oo, 8:, I:8 n.,,,
Io,, IoI, Io: n.Io, Io, n.Io
root clauses I:,, I:,,I
salience ,8, ,I, Io8, IIII:, II8
scope :,o, ,o, ,8, o,, o,, I,I, I,,o, Io,
Io, :I,
scrambling ,, oo:, ,o n.o,, 8o, ,,,
Io,,, II,, II,, I:o, I,o8, I,I,,
Io,, I8, :Io
scrambling long distance, see long
distance scrambling
sentential focus ,, ,,, I:,, I,,
set of alternatives :,, o,
shifted topics Io, Io,
S-level, see stage level predicate
spatio-temporal IoI,, :o, o, :I:
specific ,, 8Io, I,, I,, :o, ::, o,
,:,, o,, 8,, ,8, II,I8, I,o, I,,I,
I,,, I,:,, IoI:, I,,, I8,,o, I,,,
:o,, :oo, :I,Io
specific indefinites 8Io, I,, :o, ::, ,:,,
Io:, I,,
spell-out ,,, o:, I:8,, I,I, I,
stage level predicate :I n.I, ,o, II, n.,:
stage topic, implicit stage topic IoI,,
:oI, :,o, :8, o,, o8, 8,, ,,,
Io, III,, II,, I:o, I:,
Ioo, I,:, I8I, I8o, I8,, :o:,
stress :,, ,o:, ,,,, ,o, ,8o:,
8:, 8,, ,o, ,,o, ,,Io:, Io, IIII:,
I:,, I:, n.,,, I:8 n.,,, I,:,,
I,,, I,,o, Io, I,oI, I,,,
I,o, I,,, I,,, I8o, I,, I,,8,
:IoI:, :I,
stress shift oI, I,,, I,
stylistic fronting II,
subject prominent ,, :,, 8,, 8,, ,,
subject-object asymmetry :
subordinate f-structure ,,o, ,:,, o,,
I,o, I,:, Io, I,:, :o:
subordinate topic :,, ,, ,I, ,,, II,, Io,
I,I:, Io,
superiority ,o, I,, Io,, I,,, I8I, I8,,,
I,o, I,:,
switch topics, see shifted topics
Tail Io, I:I,, Io, ::, 8, Io,o, Io, n.II
Telic I,o,, :o
that-t(race) I,, I,I:, I,
thematic progression :,
theme :, ,, ,I, ,,, ,,, I,,:oo, :o:,
thetic Io, 8,, I:
topic continued, see continued topic
topic contrastive, see contrastive topic
topic indefinite, see indefinite topic
topic main, see main topic
topic stage, see stage topic
topic subordinate, see subordinate
topic chaining :, II, :I, ,
topic feature ,8, oo, I:,
Topic Focus Articulation (TFA) :, 8o
topic prominent ,, :o, :, n.:, 8, n.:,
8,, II,
topic shifted, see shifted topics
topic switch, see shifted topics
topic tests I,::, ,, ,o
Topicalization ,I:, I,, :o, :,,, 8,
,:, oo, oo, ,o, 8,,, ,I,
,,,, IoIII, II,Io, I:,, I:8,I,
I,, I,,, I,,, I, Io, I,:8,
IoI, Io,, I,,8o, I8, n.,o,
I8, :I,
topic-definition ,, I:, I,Io, I,, Io, I,,,
I,,, IoI, I88, :Io
topic-drop :,
topics multiple, see multiple topics
topological fields 8,
truth value I,Io, I,:o, ::, :,o,
8, II,
truth value gaps I, :o, ::
unaccusative ,,, I:,, :Io
unergative :oo, :I:
universal I, I:, ,,, o, ,,, ,, ,8, Ioo, Ioo,
I,, Io,, I,,, :I
V-: (verb second) 88, Io8
Vorfeld (see also pre-field) 88
weak crossover (WCO) ,o:, ,,o
wh-movement ,:, Io:, Io,
wh-topics Io8
word order canonical (see also word
order unmarked) Io,, I:, I,I
word order unmarked Io,, I:o, I:8
Yiddish movement IIoI:
index of terms :I
Language Index
Aghem I:
Catalan IoI:, Io, I,:o, :: n.:,, :, :o,
,, 8,, Io
Celtic languages 8,
Chinese :,, :, n.:, IoI, I,I n.o
Danish ,I:, I,:o, :,, ,:, ,8, 8:,, 8,,
8, n.,, Io, n.I,, IIo, I:I n.:, I8, I,o,
I,,, Io:, I, n.:, I,8, I8,
Dutch oo:, I:,, I,o,, IoI, I, I,:,
I8 n.,:, :Io
English 8 n.:, II n.:o, :,, :o, ,,, ,I, ,8,
oI, o,, ,,, ,8 n.,:, 8o:, 8,, ,,,
Io,8, IIo, II:, II,, I:I:, I:8, I,I,
I,I, I,, IoI,, Ioo, I,:, I,8,, I8I,
I8,, :o:
Faeroese I,, I,I n.8:
French ,, :I, ,8,, ,, IoI
German :, ,,, ,I, ,, n.Io, IoI, Io,,
Io,,, II,I,, I:,, I,o n.o,, I n.,,
I,:, I,, Io,, Io8 n.I,, I,I n.:I, I,,,
I, n.:, I,8, I8I, :o,
Germanic I:, ,8,, 8:, 8,8, Io,,, IIo,
IIo, I,:, I, I,, I,:
Hebrew ,I,, I,, I,,, I, n.:,
I,8I8:, I8, :o,
Hungarian 8 n.:, o, ,o n.o, 8I8:,
8,, ,oI, Ioo, II,, I:, I,,,
I8 n.,:
Icelandic I8, I,oI, I8 n.,I
Irish 8,
Italian ,8,, 8:, 88,I, ,,,, ,o n.I,
,8,, I::, Io, IoI, I,:
Japanese I:,, I,Io
Korean ,,, ,8Ioo, Io,
Lakhota Io:, Io, n.I:
Norwegian Io,, I:I, I,o,
Oodham ,
Persian I,o
Russian o8 n.o:, 8:, I:o, I:8,I, I,o,
:o, n.I,, :Io
Scandinavian ,o,, ,8 n.,:, 8o, I,8,
I,o:, I,o
Scottish Gaelic 8,
Spanish ,8,, o,, I:: n.,, I,I n.,8
Swedish Io,, I,o, I,, n.o, I8,
Wambon oI
Welsh 8,
Wyampi ,,
Index of authors
Aboh, E. O. II8 n.,,
Ackerman, F. I88 n.,,
Ahrenberg, L. 8, n.,
Aissen, J. Io n., I:,
Andersen, T. :o, Ioo n.I
Ariel, M. I n.8
Arnon, I. I,I n.,,
Bailyn, J. F. ,, n.,,, ,,, I:,
I:, n.8,o, I:8
Baker, N. :Io
Basilico, D. Io n.II, :, n.I,, :o: n.I,
Bayer, J. I,I
Beaver, D. :, n.I,, :o n.:o
Beninc, P. ,,,, IoI
Bever, T. G. I,I, :I,
Birner, B. J. , n.I, IIo n.:,, III
Bolinger, D. ,, n.o,, Io,,o, I,,o
Bresnan, J. I, n.I:, o,8
Breul, C. o, n.,,
Broadwell, G. A. ,o n.o,
Brdy, M. 8o
Bring, D. :, n.I,, ,, n.,I, , n.,, ,I, :o,
Cann, R. o,, Ioo n.,
Carnie, A. 8o n.I, Ioo, :Io
Casielles-Surez, E. , n.I, I:: n.,
Chafe, W. ,, I, n.Io, :, n.I,, :, n.:,
, n.I, , n.o
Choi, H.-W. ,o n.o,, ,, n.Io, Io,,
Chomsky, N. :,, , n.I, ,,,, ,,,
oI:, ,:, 88, 8, n.,, ,o, Io, I,,
I8,, I8,,, :I,
Christiansen, M. I,I
Cinque, G. ,: n.:8, ,,, oo, o:, ,I,
I,:, I, Io8
Clark, H. H. I,:
Clifton Jr, C. :o: n.I,
Cole, P. I8, n.,,
Costa, J. I,, n.8,
Croft, W. o,, ,I n.oo
Culicover, P. W. , n.,, 88
Dane, F. :,
Darnell, M. ,:
De Cat, C. I, n.I:, n.,
II8 n.,,, :Io
de Hoop, H. ,o n.o,, I,oo, I,,
I,, I,o
DeLancey, S. ,,
Diderichsen, P. 8, n.,
Diesing, M. ,o, ,o, I, I8, n.:,
Dik, S. C. ,, ,8, o:
Dryer, M. S. , n.o
Dyakonova, M. :Io
. Kiss, K. I, n.I:, 8Io, Ioo, II,, II,
Eckardt, R. I,o n.o,
Emonds, J. 8I, I:,
Engdahl, E. , o,
Erteschik-Shir, N. ,, ,, Io, :I:,
:, n.:I, ,I:, ,, n.,,, ,8, o n.,8,
,, n., , n.,, ,I, o,, o, n.,,,
oo, o8, ,,8, IoI, Io8 n.::, II, n.,o,
II8 n.,o, I:I n.:, I:, n.o, I:, n.,,,
I,o, I,,, Io, I,oI, I,,, I, n.I, I,o n.,,
I,8 n.,,, I,,, Ioo n.,, IoI n.8, Io:,,
Io,,, Io,, I,:, I,, n.:,, I, n.:, I,,,
I8,, I88, I,o, I,, I,, n.I, :oo, :oI n.Io,
:o, n.II,, :o, n.I,, :o,, :I: n.:,,
Fanselow, G. I,,, I8I
Featherston, S. Io8 n.I,, I8I, I8, n.,,
van Finkel, K. Io n.II
Firbas, J. :, , n.I
Frey, W. IoI,, II,I,, I n.,
Fukui, N. I:, n.,
van Gelderen, V. o:, I:,o, I:8,I,
I,: n.oI, I,,, I,o
Gernsbacher, M. A. ,o
Gibson, E. I,I n.o
Gill, K.-H. ,8,, Io,
Ginzburg, J.
: index of authors
Givn, T. :, n.I,, ,I n.oo, ,:, ,,, II, n.,8
Goldberg, A. E. o,, ,I n.oo, Io,, I88 n.,o,
:o,, :oo n.::
Goodall, G. I,I n.,8, I,,
Gordishevsky, G. :Io
Green, G. ,,
Greenfield, P. :Io
Gregory, M. L. Io,Io
Gbbel, E. :o, , n.I, oo
Grice, P. H. Io8
Grimshaw, J. I88 n.,o
Grosu, A. ,: n.I,, IoI n.8
Gruber, J. S. I,,
Guron, J. I8,8
Gundel, J. K. Io, :, n.I,, , n.I, ,I n.oo,
II, n.,8, I:I, I:,
Gussenhoven, C. ,: n.:8
Haider, H. I:, n.,o, I,o n.o,,
I8 n.,:, I8, n.,,
Hajicov, Eva :, :, n.:o
Hale, K. L. ,,, 8o n.I, I, n.I, :oI n.Io
Halliday, M. A. K. I, ,, :, n.I,, , n.I
Hannay, M. I:o n.I
Hargreaves, D. ,,
Harley, H. 8o n.I
Heim, I. ,o, , n.,:
Hirschberg, J. II n.,, ,I
Hockey, B. A. ,I n.:,
van Hoek, K. ,I n.oo
Hoji, H. I, n.o
Holmberg, A. ,o8, Ioo, II,,
II,, I8,o
Horvath, J. o,, 8o, 8, n.o, Ioo
Hsiao, F. I,I n.o
Huang, C.-T. J. :,
Hyams, N. I,:
Ishii, Y. I,:,, I,,
Jackendoff, R. :,8, ,o, ,:, , n.I, ,I, o,,
88, I8, n.,
Jacobs, J. , n.I
Jger, G. II, n.:8
Jelinek, E. :Io
Jensen, A. Io: n.Io
Joshi, A. K. , n.o
Junghanns, U. I, n.I:
Kadmon, N. , n.,,
Kaiser, E. :o, n.I,
Kamp, H. , n.,:
Kanerva, J. I, n.I:
Kaplan, T. Ioo n.,
Karimi, S. I,,
Kathol, A. 8, n.,
Keenan-Ochs, E. IIo n.:,
Kempson, R. o,, o,,o, Ioo n.,
Kenesei, I. 8o n.
Keyser, J. I,, n.I, :oI n.Io
Kidwai, A. :, n.I,, ,8
Kim, A. H. O. I,: n.oo
Kim, S. ,o
King, T. H. o8, I:,,
Kluender, R. Io, n.I,, I8,,:
Kratzer, A. I, n.I:, :I n.I
Krifka, M. ::
Kuno, S. ,, , n.I, :, n.:o, ,I n.oo, ,: n.I,,
I,,, IoI, I,o
Kuroda, S.-Y. Io n.II
Ladd, D. R. , n.,
Ladusaw, W. IoI, n.III:
Lambrecht, K. Io n.II, ::, :, n.I,,
:,, n., o, n.oo, ,I, ,,,
Io, n.I,, IIo, II, n.,8, I:I:, Ioo n.I,
Io, n.Io
Langacker, R. W. ,, n.oo
LaPolla, R. J. , :, n.:o, o,, ,I n.o,o,
I:: n., IoI, Io, n.Io, Io8, I,,, :I,
Larson, R. J. :oI n.Io
Lasnik, H. ,I
Lebeaux, D. o:
Lee, J. ,o
Li, C. I, n.Io, :, n.I,, :o, 8, n.:, II, n.,8,
Ioo n.I
Liberman, M. , n.,
Lpez, L. II8 n.,,
MacDonald, M. C. I,I
Massam, D. II8 n.,,
Meyer-Viol, W. o,
Maynell, L. A. I8, n.,,
Meinunger, A. II, n.:8, Io, n.I8
Michaelis, L. A. Io,Io
Miller, P. H. Io,,I
index of authors :,
Mittwoch, A. :ooIo
Miyagawa, S. I:, n.,o, I,:, I,,
Molnr, V. :, n.I,
Morgan, J. L. Io,
Neeleman, A. ,: n.:8, oo:, I:, n.,,o,
I,: n.o:, I,,, I,o, I,, I,:
Newmeyer, F. :, ,I,
Nikolaeva, I. :: n.I,
Panevov, J. :, :, n.:o
Partee, B. H. ,o
Paul, I. :, n.I,, II8 n.,,
Payne, D. o8, ,:,, ,,
Pesetsky, D. ,. I,,
Pierrehumbert, J. ,I
Poletto, C. ,
Polinsky, M. :, n.I,
Pollard, C. , o,
Poulsen, M. ,8 n.,I, Io, n.I,
Prince, E. F. ,, , n.I, II n.,, :, n.I,,
, n.I, , n.o, ,I n.oo, ,,,
88, Io,I,
Rahkonen, M. I,, n.o
Rapoport, T. o, n.,,, Io,, I88 n.,o,
I,,, :oI n.Io, :o, n.I,, :o,,
:I: n.:,
Raviv, A. S. I,o
Reinhart, T. IIo, I, n.I:, I,, ::, :o,
,: n.:8, ,, n., oo:, II, n.,8,
I:, n.,,o, I,: n.o:, I,,, I,o,
I,, I,:, IoI, Ioo n.I, :I,
Reis, M. I,: n.::
Rice, S. :o
Ritter, E. ,,
Rizzi, L. o,, ,:, 88,o, ,8IoI, II8,
I:,, I,, I,, :I,
Roberts, C. I,, I8, n.,,
Roberts, J. R. I,,
Rochemont, M. S. ,: n.:8, , n.,,
, n.I, 88
Rodionova, E. V. I:8 n.,,
Rooth, M. :8,
Rosen, S. T. ,,
Rosengren, I. Io n.II, I:, n.,o, I,o n.o,
Ross, J. R. ,, n.,:, 8o n.I, I:
Rudin, C. I:, n.,o
Sadeh-Leicht, O. I8, n.,,
Sag, I. A. , o,, I,o, I,, n.:,, I,:,
Samek-Lodovici, V. I::,
Sasse, H. Io n. II, :, n.I,
Schaeffer, J. I8 n.I,, II,, I,, :Io
Schieffelin, B. IIo n.:,
Schmerling, S. F. ,I: n.:,
Schulz, B. :
Schwarzschild, R. :, n.I,, ,, n.,I
Selkirk, E. ,:,, ,,
Sgall, P. :, :, n.:o
Shimojo, M. , n.o
Smith, C. S. I,o n.,
Steedman, M. , ,I n.:,, ,,8,
, n.:, ,,
Stemmer, B. :I, I
Stowell, T. ,I
Strahov, N. I:, n.,,, I:8 n.,,,
I,o n.,,, :o, n.I,
Strawson, P. F. I,I, I,, ::, :,
Surnyi, B. 8,, Ioo
Svenonius, P. Io,, Ioo, IIo,
IIo, I,o
Szabolcsi, A. 8,
Szendri, K. ,: n.:8, 8, n.o,
Ioo, I,, n.8,
Takami, K.-I. ,, ,I n.oo
Takano, Y. I:, n.8
Taverniers, M. :, n.I,
Thompson, S. I, n.Io, :, n.I,,
:o, 8, n.:, II, n.,8,
Ioo n.I
Tsoulas, G. ,8,, Io,
Vallduv, E. , Io, I:, I,Io,
:: n.I,, , n.I, n.,
Io, n.II
Van Valin, J. , :, n.:o, ,, o,8,
,I n.o,o, 8:, I:: n., I,,,
IoI, Io, n.I:, Io, n.Io, Io8,
I,,, :I,
Vikner, S. I88 n.,o
Villalba, X. II, I,
Volpe, M. :oo n.:I
von Heusinger, K. ,,, , n.,,
Vries, L. d. o n.,,
Wagner, M. I:o n.o, I,o n.o,,
I,: n.8,
Walker, M. A. , n.o
Ward, G. , n.I, II n.,, :, n.I,,
, n.I, IIoII
Wasow, T. I,:
Watters, J. R. R. I n.o
Wedgewood, D. J. ,o n.o
Weskott, T. Io, n.:I
Wexler, K. :, I,:
Williams, E. o,, o,, Ioo n.I,
Wiltschko, M. I8I:
Winkler, S. :o, , n.I, oo
Ziv, Y. I8, n.,,
Zubizaretta, M. L. , n.I, ,,, ,8oo,
o:, I::, :I: n.:,
Zybatow, G. I, n.I
:o index of authors