Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
COMPETENCE AND PERFORMANCE
MODELLING OF FREE WORD ORDER
GEERT-JAN M. KRUIJFF & SHRAVAN VASISHTH
Copyright (c) 2001, Geert-Jan M. Kruijf f andShravanVasishth
Addresses:
Geert-JanM. Kruijf f
Computational Linguistics
University of theSaarland
Postfach 15 11 50
D-66041 Saarbruecken(Germany)�[email protected] �
Shravan Vasishth
Departmentof Linguistics
Ohio StateUniversity
Columbus,Ohio (USA)�[email protected] �
This document hasbeentypesetby the authorsusing LATEX 2� with the bookufal class
(whichextendsthestandardLATEX 2� book class).Otherpackages thathavebeenusedare
ChrisManning’s avm packagefor typesettingattribute-valuematrices,Hans-PeterKolb’s
gb4e andcgloss4e packagesfor typesettingtheexamples,PaulTaylor’s prooftree pack-
ageandMakotoTatsuta’s proof package for typesetting proofs, the ����� -LATEX distribu-
tionof theAmericanMathematical Society, theAssociationfor ComputationalLinguistics’
acl packagefor citationcommands,andMichaelCovington’s drs macros for typesetting
discourserepresentation structures. The pictures werecreatedusingXFig andthe epic
family of graphical packages.Finally, theindexeshavebeencreatedusingMakeIndex and
F.W. Long’s multind package.
COMPETENCE AND PERFORMANCE MODELL ING OF FREE
WORD ORDER
Geert-Jan M. Kruijf f & ShravanVasishth
Abstract
Thecourse surveys competenceandperformanceissuesrelating to modelling
grammarsof freeword orderlanguages.
Regarding competenceissues,we focuson two fundamental issues:adjacency
(unbounded dependencies, crossed andnested dependencies),and the power re-
quiredof theformal apparatususedto explain freeword order. In this context, we
survey variousapproaches to modelling freeword orderin natural languagegram-
mars(notable,categorialgrammar).Following thePragueschool of linguistics,we
arguethatvariation in wordorder is astructuralmeansto realize informationstruc-
ture- just like intonationin English, or theuseof specific morphological particles
in Japanese.
Regarding performancemodelling, we first present a summaryof existing re-
search (both theoretical andempirical) on sentenceprocessingin Dutch,German,
Japanese,andKoreanthat focuseson dependency issues(like cross-serial center-
embedding, and scrambling). We also presentnew research on Hindi sentence
processing that sheds new light on the cross-linguistic issuescurrently underin-
vestigationin theliterature.
We concludeby presenting a coherent computational modelof grammarsfor
theselanguagesthattakesinto account thecompetenceandperformanceissuesand
facts discussedearlier in thecourse.
Course prerequisities: None.
CONTENTS
Intr oduction ix
0.1 Trying to understandvariability in word order . . . . . . . . . . . ix
0.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1 Form and function in DependencyGrammar Logic 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Syntactic categoriesandcomposition . . . . . . . . . . . . . . . . 2
1.2.1 Thehead/dependentasymmetry . . . . . . . . . . . . . . 4
1.2.2 Categoriesandcomposition . . . . . . . . . . . . . . . . 7
1.3 Relating form andfunction . . . . . . . . . . . . . . . . . . . . . 12
1.3.1 Modeling morphological form . . . . . . . . . . . . . . . 16
1.3.2 DGL’s linking theory . . . . . . . . . . . . . . . . . . . . 24
1.3.3 Thecomposition of linguistic meaning . . . . . . . . . . 30
1.4 Typology, form, andstructural rules . . . . . . . . . . . . . . . . 35
1.4.1 Structural rules andgradual refinement . . . . . . . . . . 37
1.4.2 Multili ngualnetworks of structural rules. . . . . . . . . . 38
2 Theoriesof Inf ormation Structur e 41
2.1 Informationstructurein linguistic meaning . . . . . . . . . . . . 41
2.2 Informationstructurein thePragueSchool . . . . . . . . . . . . . 46
2.3 Steedman’s Theme/Rheme. . . . . . . . . . . . . . . . . . . . . 52
2.4 InformationPackaging . . . . . . . . . . . . . . . . . . . . . . . 55
2.5 Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.6 Informationstructurein DGL . . . . . . . . . . . . . . . . . . . . 68
3 The category of informativity 77
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2 Basicword order . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.2.1 Hawkins’ typology of basic word order . . . . . . . . . . 80
v
3.2.2 A typological modelof basic word order in DGL . . . . . 81
3.3 Variability in (basic) word ordering . . . . . . . . . . . . . . . . . 84
3.4 Thecategory of informativity . . . . . . . . . . . . . . . . . . . . 92
3.4.1 Thenull hypothesis. . . . . . . . . . . . . . . . . . . . . 93
3.4.2 Predicting a language’s canonical focusposition . . . . . 95
3.4.3 Focusprojection . . . . . . . . . . . . . . . . . . . . . . 98
3.4.4 Changing focus . . . . . . . . . . . . . . . . . . . . . . . 100
4 A formal modelof word order asstructura l indication of informativity 107
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2 Modelsof flexible word order in categorial grammar . . . . . . . 108
4.2.1 Steedman’s CCG . . . . . . . . . . . . . . . . . . . . . . 109
4.2.2 Hoffman’s Multiset-CCG. . . . . . . . . . . . . . . . . . 111
4.2.3 Baldridge’s Set-CCG,modalizedCCG . . . . . . . . . . 112
4.2.4 Modelsof word order in CTL . . . . . . . . . . . . . . . 115
4.3 Variability of word order in DGL . . . . . . . . . . . . . . . . . . 119
4.3.1 Preliminariesto theformulation of thepackages . . . . . 121
4.3.2 OV packages . . . . . . . . . . . . . . . . . . . . . . . . 122
4.3.3 V-First packages . . . . . . . . . . . . . . . . . . . . . . 138
4.3.4 SVO packages . . . . . . . . . . . . . . . . . . . . . . . 143
4.4 Modeling thestrategies . . . . . . . . . . . . . . . . . . . . . . . 151
5 A formal model of tune asstructural indication of informativity 161
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.2 Steedman’s syntax-phonology interface . . . . . . . . . . . . . . 162
5.3 Tunein DGL . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.4 Interacting tuneandword order . . . . . . . . . . . . . . . . . . . 169
6 DGL, topic/focus,and discourse 171
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.2 Dynamicinterpretation of information structure . . . . . . . . . . 172
6.3 Binding across(clausal)boundaries . . . . . . . . . . . . . . . . 179
7 An empirical evaluation of sentenceprocessing models 189
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.2 Whatarecenter embeddingsandwhy arethey interesting? . . . . 190
7.3 Threemodels of sentenceprocessing . . . . . . . . . . . . . . . . 193
Contents /vii
7.3.1 Joshi’s Embedded PushdownAutomaton (1990) . . . . . 193
7.3.2 Predictionsof theEPDA modelfor Hindi CECs. . . . . . 197
7.3.3 Concluding remarksregarding theEPDA model. . . . . . 198
7.3.4 Gibson’s Syntactic Prediction Locality Theory(1998/1999) 199
7.3.5 Lewis’ Interference andConfusability Theory (1998/1999) 200
7.4 Centerembeddingsin Hindi: Threeexperiments. . . . . . . . . . 203
7.4.1 Experiment 1 . . . . . . . . . . . . . . . . . . . . . . . . 203
7.4.2 Experiment 2 . . . . . . . . . . . . . . . . . . . . . . . . 205
7.4.3 Experiment 3 . . . . . . . . . . . . . . . . . . . . . . . . 208
7.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
8 Processing asAbduction+Deduction:
A SentenceProcessingModel 213
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
8.2 Processing asabduction+deduction: Themainproposal . . . . . . 215
8.2.1 Abduction . . . . . . . . . . . . . . . . . . . . . . . . . . 216
8.2.2 Somedefinitions . . . . . . . . . . . . . . . . . . . . . . 218
8.2.3 Thealgorithm . . . . . . . . . . . . . . . . . . . . . . . . 220
8.3 Theempirical coverage . . . . . . . . . . . . . . . . . . . . . . . 223
8.3.1 Japanese . . . . . . . . . . . . . . . . . . . . . . . . . . 223
8.3.2 DutchandGerman . . . . . . . . . . . . . . . . . . . . . 230
8.3.3 Hindi . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
8.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . 233
References 237
List of Figures 249
LanguageIndex 250
NameIndex 251
Subject Index 253
viii � Contents
CHAPTER 0
INTRODUCTION
0.1 TRYING TO UNDERSTAND VARIABILITY IN WORD ORDER
Thiscourseis about trying to understand variability in wordorder. Not just how we
canprovidea modelfor it, but in particular whenandwhy variation occurs, how
that variation interacts with other levels of linguistic structure, and how people
processvariation in word order.
Theprincipalunderstandingweadvancehereis thatvariability in wordorderis
notarbitrarybut sensitive to context – variability is ameansfor speakersto express
information structure. Naturally, this raisesthequestionswhenalanguageactually
allows a speaker to do so,andhow variation might interactwith for example tune.
Thesequestionsarenot easilyanswered. Determining whena languagehasa
certain degreeof variability in wordorderis anissuethathasbarelybeenaddressed
in linguistics.Steele(1978) proposedto definethreediscretedegreesof wordorder
variability: rigid, mixed, and free, elaborating on the work on basicword order
typologydonebyfor exampleGreenberg (1966). However, Steeledoesnotprovide
a satisfactory answerfor how to predict a languagehasa certain degreeof word
order variability . Moreover, we should consider a cline from rigid to free word
orderrather thandiscrete degrees:Both DutchandGermanaremixedword order
languages,yet Germanis freer in its word order than Dutch. Oneof the points
we addressin this courseis how onecould go about constructing a typology that
predicts variability.
Turningto theissueof modeling word ordervariability , we find that mostfor-
malmodels of word order do not take informationstructureinto account asa prin-
ciple to explain why variation occurs(with theexception of for example(Hajicova
et al., 1995; Hoffman,1995a)). Word ordervariation is usually not the only fac-
tor in realizing informationstructure though. Whether we look at languageswith
mixed word order like Dutch or German,or thosewith a relatively free word or-
der like Hindi, Hungarian, or Turkish, we find that variation often interactswith
ix
x � Introduction
tune. Theinteraction guides,or restricts, theexact interpretation of thesentence’s
surfaceform. But not all formal frameworks canmodel this type of interaction.
We argue herethat transformational grammarcannot, that the current versionsof
Combinatory Categorial Grammar(CCG,MCCG – (Steedman,2000c; Hoffman,
1995a))faceproblemsaswell, andthatmostincarnationsof HPSG appear to lack
themeansto give a principledway to explain thephenomenaat hand.
The framework that we proposehere conceivesof informationstructure asa
fundamentalaspect of sentential linguistic meaning, andit explains e.g. word or-
der and tune in termsof how they realize information structure. We present a
variety of competencemodelsthat model this – not just how word orderor tune
may do so on their own, but alsoin interaction. Thereby, an important aspect of
these modelsis theway they areformulated.Ratherthanpresenting fragmentsfor
individual languages, wepresentarchitecturesof fragments. In thesearchitectures,
fragmentscanbeshared acrosslanguages, with which we canmodelcommonali-
ties anddifferencesamong(related)languages.The typological story getsfolded
into thesearchitecturesasfollows. Weconsiderthetypological implications to de-
termine whether or not a languagehasaccessto a particular fragment(modeling
specific phenomena).This leadsto the idea of a “typological universalgrammar”
(Hawkins,1983) that promotesvariety rather thanthetypeof monoliticy advanced
in Chomskian universal grammar. The structure we give it strongly reminds of
morefunctionalist approaches like (Halliday, 1985).
Comment Something on the relation between competence/performance?
Comment Performance modeling
In the remainder of the introduction we give a moredetailed overview of the
individual chapters.
0.2 OVERVIEW
In Chapter 1, Form and function in DependencyGrammar Logic I begin by
showing thehybrid logic of linguistic meaningdevelopedin (Kruijf f, 2001) canbe
tied to the resource-sensitive proof theory of categorial type logic. Hybrid logic
(Blackburn,2000) enables usto give anonthologically fine-grainedrepresentation
of the meaning that a sentenceexpresseslinguistically. By tying hybrid logic to-
gether with categorial grammar, we defineDGL asa framework of grammar. To
Introduction /xi
that end I first of all definethe formal apparatus,in the shapeof a headedcate-
gorial calculusin which theanalysisof form leads to thecompositional formation
of linguistic meaning. Like any other categorial approach,theinferencein DGL’s
calculusis drivenby categoriesthatarereflectionsof theunderlying meaningthey
realize. Traditionally the logical tradition derivescategories from Ty -formulas,
but DGL no longer doesthis - it usesa dependency-based, hybrid-logical spec-
ification of meaning. The main issuethat thus arises, is how to let a category
reflecta predicate-valency structureand its dependency relations. But this is no
new issue. Key observationshave beenmadeby Mathesius, Jakobson,and-later-
Kuryłowicz regarding the relation betweenmorphological form anddependency
relations. I argue how theseobservations canbe fruitfully used to provide a pro-
cedure of category-formationthat is basedon linguistic motivations- a possibility
arising from thePraguianview on dependency grammar.I endthechapterwith a
proposalfor anapproachto cross-linguisticmodeling in categorial grammar,intro-
ducing theconcept of a grammararchitecture.
Whatthisfirst chapterprovidesis adependency-basedgrammarformalism that
finds its linguistic motivation in the PragueSchoolof Linguistics, andwhich can
construct logical descriptions of linguistic meaning in a compositional andmono-
tonic way usinga categorial analysisof a sentence’s form. By virtue of a detailed
description of aspectual categoriesanddependency relations,the formulation of a
sentence’s linguistic meaning in DGL not only elucidates which dependents mod-
ify what heads, but alsowhat causal andtemporal entailmentsaretriggered. On
thecategorial side,DGL developsout aheaded calculus,morphological strategies,
andthenotion of agrammararchitecture. Takenaltogether, this providesuswith a
basisfor thesubsequentchapters.
The overall aim of the next chapters is to provide a preliminary account of
how we canexplain the realizationof information structure across typologically
different languages,andhow suchanexplanation canbeintegratedinto agrammar
framework.
I startwith Chapter 2, Theories of information structur e. In this chapter,
I discuss various theories of informationstructure that have found their way into
formal grammar. Basedon reflections on thesetheories,I motivatewhy I opt for
the Praguian approachandhow it is (conceptually) closely related to Steedman’s
theory of information structure. The chapter ends with definitions spelling out
how information structure is represented in DGL (contextual boundness, topic-
focusarticulation), andafew preliminary remarksonthedynamicinterpretationof
xii � Introduction
linguistic meaning andits informationstructure,to bedefinedlater in Chapter 6.
In Chapter 3, The category of informativity , I discussa basic typological
characterizationof whenlanguagesusevariability in word orderor tuneto realize
informationstructure, thustrying to characterizecontextual boundness asa typo-
logical category of informativity. The characterizationis based on empirical data
from a variety of typologically different languages, anda new typology of vari-
ability in word order. First I formulate a setof typological hypothesesthatpredict
whether a languagehasrigid, mixed,or freeword order, integrating(Steele,1978)
with Skalicka’s languagetypology (SkalickaandSgall,1994;Sgall,1995). Subse-
quently, I argue for a setof hypothesesthatpredict whereto expectthe canonical
focus position, andwhenlanguagesuseword order, tuneor a combinationthereof
to realize information structure. Thesesetsof hypotheses form the typological
basis for thegrammararchitecturesto bepresentedin thenext two chapters.
In Chapter 4, A formal model of word order as structura l indication of
informativity , I elaboratevariousgrammararchitecturesthatmodelvariability in
word order andhow that variability canbe usedasa structural indication of in-
formativity. Thearchitecturesareillustratedon a largenumber of examplesfrom
a variety of languages. The approachI take to modeling variability is motivated
in thebeginning of thechapter, whereI present a discussionof variouscategorial
accountsthathave beenprovidedto modelvariation in word order.Here,I opt for
viewing adjacency asa parameter. This enablesus to consider informationstruc-
tureasaprimaryfactor (parameter)determining wordorder. Information structure
andword orderasa structural indicationof informativity arerelated through the
notion of systemic ordering, which indicatesherewhether dependents are real-
ized in canonical orderor not. This providesa useful abstractionover a concrete
prosodic/syntactic structure, and I show in the next chapter how it enables us to
smoothly integrate tune and word order as structural indication of informativity
into a single model.
Besides word order languagesusually also use tune to realize information
structure– sometimesevenpredominantly so,like in thecaseof English. In Chap-
ter 5, A formal model of tune asstructural indication of informativity , I begin
with a discussionof Steedman’s modelof English tunedevelopedin Combinatory
Categorial Grammar.I thenpresent a moreabstractmodelof tunethatcanbe in-
stantiatedto cover different languages, andthatovercomesa few problemsI note
for Steedman’s proposal. The chapter ends with a discussion of how we canin-
clude the modelof tune in the word orderarchitecturesof Chapter4, to provide
Introduction /xiii
a modelof the interaction betweentuneandword order in realizing information
structure.
Finally, in Chapter 6, DGL, topic/focus,and discourse, I addressthe inter-
pretionof sentence’slinguisticmeaning. After (Peregrin, 1995; Kruijf f-Korbayova,
1998)I definethe interpretation of linguistic meaning dynamically, guidedby the
linguistic meaning’s information structure. I usehybrid logic to provideareformu-
lation of Kruijf f-Korbayova’s TF-DRT (Kruij ff-Korbayova, 1998). On the result-
ing proposalI illustratehow binding acrossclausescanbe modeled using hybrid
logic’s jump-operator.
To recapitulate, I provideherea preliminary, typological account of how word
ordervariability andtunecanrealizeinformationstructure.Thisaccount is coupled
to comprehensivemodelsof wordorder tuneasstructural indicationof informativ-
ity. Thegrammararchitecturescanbeusedin DGL to createa representation of a
sentence’s linguistic meaning, including informationstructure,asthe result of an
analysisof thatsentence’s form. At theend,I show how such a representation can
befurther interpreted onadiscoursemodelthatis sensitiveto informationstructure
(Kruijf f-Korbayova, 1998).
SOURCES
Chapters2 through7 arefrom Geert-Jan Kruijf f ’sdissertationentitledA Categorial-
Modal Logical Architectureof Informativity: Dependency GrammarLogic & In-
formation Structure (Kruijf f, 2001). Chapter 8 is to appear as..., whereas Chapter
9 is publishedas...
xiv � Introduction
CHAPTER 1
FORM AND FUNCTION IN DEPENDENCY
GRAMMAR LOGIC
Dependency GrammarLogic (DGL) is a dependency-basedgrammarframework in which a
categorial calculus is usedto analyze form and deliver the kind of representation of a sen-
tence’s linguistic meaning as discussedin the previous chapters. This chapter develops the
foundationsfor DGL. I introducethecategorial calculus,anddiscusshow it canbeemployed to
deliver a dependency-basedanalysisof form. To that end,I first discusshow head/dependent-
asymmetriescanbecaptured,andhow morphologicalstrategiescanbemodeled. Subsequently,
I show how linguistic meaningcanbeformedcompositionally, in parallel to theanalysisof form.
Thetwo principal issuesin thatdiscussionareDGL’s linking theory, relating predicate-valency
structuresandsyntactic categories,andthe interpretation of wordgroupsasparticular types of
dependents.Finally, I presenta proposal for how to construct multil ingualgrammarfragments
in DGL, introducingtheconcept of architecture.
... ��� ��������������� ���� �"! �#$�%�&� �#'� � ( � �)... [the] wayup [and] down [is] oneand[the] same.
Heraclitus,Diels-Kranz22B 60
1.1 INTRODUCTION
In this chapter I lay thefoundationsfor DependencyGrammarLogic, or DGL for
short. My aim hereis to explain how the discussionof the previous two chapters
can be related to an extended form of categorial grammars (namely, categorial
type logics). The result is the basis for a dependency-basedgrammarframework
that foll ows the Praguian form/function distinction. For one, DGL should enable
us to constructgrammarfragments that modelparticular phenomenafor a given
language,foll owing out a dependency grammarperspective. But there is more:
Dueto thefundamental role thatthe(abstract)relationsbetweenform andfunction
play, DGL alsoenablesusto develop a cross-linguistic perspective on phenomena
in natural languages. This latter point is very salient in Praguianlinguistics but
has,unfortunately, received little or no attention in contemporary formal theories
of grammar.
1
2� Formandfunction in Dependency GrammarLogic
To startoff, I begin in section 1.2with aconsiderationof thenatureof syntactic
categoriesandcomposition in DGL. I presenthow categoriesandcomposition can
bebuilt aroundtheidea of a head-dependent asymmetry, andhow dependency re-
lationsandmorphological informationarerepresented in categories. Subsequently,
I work out thelinking theory andthecategorial-hybrid logical calculusin * 1.3. At
the endof this chapter I develop the ideaof integrating cross-linguistic (or typo-
logical,or multilingual)modeling into DGL.
1.2 SYNTACTIC CATEGORIES AND COMPOSITION
In the lexicon, we assigneachword a syntactic category. That category is either
a basic category or a function category. A basic category is atomic- for example,+- andindicatesthat theword does not rely on thepresenceof further arguments
to be provided for that word to enter into a grammatical composition. On the
contrary, a functional category specifies oneor moreargumentsthat areneeded,
andaresulting category thatis affected onceall theargumentshavebeen provided.
The familiar slashes ,&-/. areused to indicatethe position relative to the function
wheretheargumentis expected.
Historically, therearetwo waysin which functional categoriescanbewritten.
Oneway is dueto Lambek,theother is dueto Steedman.TheLambek-notation is
characterizedby thefactthatall theargumentsexpectedto theleft areplacedto the
left of theresulting category, andsimilarly with all theargumentsexpectedto the
right. On the otherhand,Steedman’s notation putsthe resulting category always
up front, after which all the argumentsfoll ow, againwith slashesindicatingtheir
directionality. Thefollowing examples illustratethedifferences- the(a) examples
usethe Lambek-notation, the (b)-examplesusethe Steedman-notation. To illus-
tratemoreclearly which argumentis what,we specify thetype of dependent each
argumentoughtto be: 021 meanstheargumentis a dependentof type 3 .(1) English
“Actor walks.”
a. walks 4507698;: + ,=<b. walks 4><?,=0 628@: +
(2) English
“Actor givesPatient Addr essee.”
a. gives 4BACA;0D628@:FEHG + ,=<?IC.=0 62J�JCGCKMLNLNKHK + IC.=0PORQ�:�S�KNTU: +
Formandfunction in Dependency Grammar Logic /3
b. gives 4VACA;<?,=0D698;:FENG + IC.=0 69JWJCGCKHLHLXKHK + IC.=0PORQW:�S�KNT�: +(3) Czech“ cteActor Addr esseePatient.” (English: reads)
a. cte 4BACA;<Y.=0DORQW:�S�KNT�: + IC.=0 69J�J�GMKMLHLXKHK + IC.=0P698;:FENG +b. cte 4ZACA;<[.=0 ORQW:�S�KNT�: + IC.=0 69J�J�GMKMLHLXKHK + IC.=0 698;:FENG +
(4) Japanese“Actor AddresseePatient ageta” (English: gave)
a. ageta 4>0DORQW:�S�KNT�: + ,\A;0 62J�JCGCKMLNLNKHK + ,\A;0]698@:FEHG + ,=<^ICIb. ageta 4BACA;<^,=0 628@:FEHG + IC,=0 69J�J�GMKMLHLXKHK + IC,=0 ORQ�:�S�KNTU: +
Throughout the dissertation I useSteedman-style notation. Technically, the
Lambek-notation and Steedman-notation are just notational variants (as can be
easily verified). From the viewpoint of readability, though, it seemsthat Steed-
man-style categories remainmore perspicuous even in the presenceof detailed
information about form - becausethe resulting category is alwaysclearly located
at thebeginning.1
Definition 1 (Categoriesvalid in DGL). Givena setof basic categories _ , a set
of dependencyrelations ` , a setof features a (e.g. to specify aspectsof form),and
a setof modesb . Thenthesetof valid (or well-formed)categories in DGL, c JCdWecanbedefinedasfollows.
1. Everybasiccategory fhgi_ is a well-formedcategory: fjg�c JCdWe .2. Giventwo categories fkSH-�f�ljgmc JCdWe and a mode nogpb , thenthe follow-
ing categoriesare alsoin c J�dWe : ANf S ,Uq�f l I/-rANf S .Uq�f l I/-rANf S9s q�f l I , with f S the
resulting category.
Furthermore,let themodalprefix t of a category bethatsequenceof unary modal-
ities 0 , uwv that prefixes a category C, with C beingeither basic, or of the form
ANf2S;,'f�l�I/-rANf2SN.'fxlUI/-rANfyS s fxlUI . Then,thefollowing categoriesarealsovalid in DGL:
3. Givena dependency relation 3zg{` , anda category t�f , thenthefollowing
categories are also in c JCdWe iff 071 doesnot appear in the modal prefix t :
0P1/t�fjg�c JCdWe .4. Givena feature |}g~a , and a category t�f , thenthe following categories
are also in c JCdWe iff there is no (modalized) feature |r� in t that would -
linguistically- contravenewith | : u v�� tif7-/0 � t�fhg�c JCdWe .1Therearealsosomecomputational “arguments”- Steedman-stylecategoriesaremoreconve-
nientwhenit comesto parsingcategorial typelogics,cf. (Kruijf f, 1999b).
4� Formandfunction in Dependency GrammarLogic
Valid categories are all thosecategories that can be specified on the basis of
steps 1-4; nothing elseis a valid category in DGL. �Remark 1. A few remarksarein place.Steps1 and2 build up c J�dWe in a straight-
forward way. Following the traditional formulations of categorial type logics (cf.
(Hepple,1994; Moortgat,1997a;Morril l, 1994)) we includeproducts s q (pairing),
besidesslashes , q -/. q . Step3 defines categories involving thespecification of de-
pendency relations, in such a way that we exclude the possibility to specify one
argumentto beinterpretable astwo dependency relations:
(5) a. thecategory A;<?,r�9LX8W0]698@:FEHG�0PORQ�:�S�KNTU: + I is invalid, becauseanargument
cannot bea verb’s Actor aswell asits Patient.
b. thecategory ACA;<?,U�9LN8W0P698;:FENG�IC. J�8M� 0PORQW:�S�KXT�: + is valid
Step4 avoids the situation in which oneandthe samecategory getsspecified
ashaving, for example,botha nominative andanaccusative inflection.2 �
1.2.1 THE HEAD/DEPENDENT ASYMMETRY
How do we incorporate the idea of a head-dependent asymmetry into our cate-
gories? To begin with, it hasbeenoften observed in the pastthat the functional
categoriesfoundin categorial grammarincorporatealreadyanideaof adistinction
between headsanddependents. Bar-Hillel, after all, consideredcategorial gram-
marto beadependencygrammarandnotaconstituency grammar. Amongthefirst
to exploretheideaof representing ahead-dependentasymmetry in categorial gram-
marin moredepth wasVenneman(1977). Contemporaryproposalsfor includinga
notion of head in categorial grammarinclude Barry andPickering (1992), Moort-
gatandMorrill (1991), and-basedupon the latterproposal- MoortgatandOehrle
(1994), andHepple(1994; 1996; 1997).
MoortgatandMorrill develop in (1991) a calculus thataimsat combining the
notion of constituency or phrasalstructure(i.e. linearization),andhead-dependent
asymmetry. Although their effort looks similar to what Vennemansetout to do
in his (1977), this time just using the more powerful categorial type logics, this
would neverthelessnot be correct. Whereas Venneman sought to combine con-
stituency/linearizationandahead-dependentasymmetry using thefunction/argument
2Notethatif thecategorywouldbeassignedtoawordthatismorphologically ambiguousbetweenbeingnominativeor accusative,thensuchshouldbecapturedusinganunderspecifiedmorphologicalmarking- cf. (Heylen,1999) andthediscussionaboutunderspecificationbelow.
Formandfunction in Dependency Grammar Logic /5
structureof categories,this is exactly whatMoortgatandMorrill arenot trying to
do:
“The important point hereis that we consider the dependency asymmetry
asanautonomousdimensionof linguistic organization - a dimensionwhich
maycross-cut thedistinctionsthatcanbemadein termsof thefunction/argument
opposition.” (1991)(p.15).
MoortgatandMorrill contrasttheir ideaswith approacheswhereahead-dependent
asymmetryis defined in termsof function/argumentstructure,like BarryandPick-
ering’s (1992), arguing that in suchtheories“[headedness]is a derivative concept
justemployedin elucidation.” (ibid.) Whether weagreewith this position or not is
not thepoint at themoment- let usfirst have a closer look at their proposal,which
formedtheinspirationfor many othersto follow later.
Moortgat andMorrill startoff discussing how constituency canbehandled by
categorial type logics of a fairly limited power - namely, thenonassociative Lam-
bek-calculus NL andthe associative calculus L. Oneof the downsides of the as-
sociative Lambek-calculus L is its insensitivity to domains of constituency: The
immediateconstituency hypothesisgivesrise to a fairly rigid bracketing scheme,
which the L is of course imperviousto dueto its associative character. Moortgat
andMorrill discusshow acombinationof NL andL leadto a (hybrid) calculusthat
not only overcomesthis apparentproblem,but -moreimportantly- givesriseto the
well-knownnotion of “flexible constituency”.
Subsequently, MoortgatandMorrill present anon-associativecalculusin which
it is explicitly represented which of thetwo componentsin abinary structureis the
head. As Morril l clarifies later in his (1994)(p.88ff), the calculus developed in
(Moortgat andMorrill , 1991) is essentially NL with subscripts � and � addedto its
operators. The � marksthat in a binary structure the left constituent is the head,
whereas� indicatesthattheright constituent is thehead.
Thoughsimplein nature, theproposalmakesessential useof thepossibiliti es
of categorial type logics to control the construction of trees. For example, struc-
tural rulesfor associativity aregiventhatshowhow headed-nessis preservedover
rebracketing (even though constituency structure is, obviously, changed). Moort-
gatandMorrill illustratetheir approachonmetrical trees, whicharebinarytreesin
whicheachmother nodemarksonedaughternodeasstrongly stressedandanother
asweaklystressed.
With metrical trees,MoortgatandMorril l try to makeacasefor theirargument
that headedness should be consideredasa primitive concept, not asonederived
6� Formandfunction in Dependency GrammarLogic
from function/argumentstructure.For example,consider theirexample(44)(p.17),
heregivenas(6):
(6) English
“What happened?”
a. Johnarrived.
b. Johnleft.
Moortgat andMorrill point out that the neutral utteranceof (6(a)) hasstress
on “John” - stress on the verb would put the verb ‘in focus’. On the otherhand,
neutral utteranceof (6(b)) hasthe stresson theverb. Using �;.U� to indicatewhich
constituent receivesstress, wecanrepresenttheexamplesasthefollowing metrical
configurations(7).
(7) (MoortgatandMorrill , 1991)(Example45,p.17)
a. � e Johnarrived]
b. � G Johnleft]
Then,to quote MoortgatandMorril l,
“Observethatany attemptto characterize prosodicstructurepurelyin terms
of the function/argumentasymmetry would have to treatthetwo verbson a
par: herethenwe seeanexample of theautonomouscharacterof thedepen-
dency dimension.” (1991)(p.17)
Moortgat andMorrill closetheir discussionwith remarking that the calculus
they develop canmodeldifferent typesof dependency - not just theprosodic per-
spective they take,but for examplealsosyntacticor semantic typesof dependency.
Let usreturnthento MoortgatandMorrill’ spoint that phrasingahead-dependent
asymmetry in termsof function/argument-structuremissesthepoint: function/argument-
structure elucidates linearization, and the head-dependent asymmetry might cut
acrossthat. In otherwords,linearization anddependency aretwo differentdimen-
sions,andshould thereforebekeptseparate. Forexample,considertheprototypical
category for asentential adjunct, like a temporal adverbial: <�,^< . Thiscategory is
a function, taking averbasits argument.However, thehead-dependentasymmetry
is exactly theopposite,astheverbgovernsthetemporal adjunct. A templatefor an
appropriate categorial assignmentwould thusbe <�,/�/�7< .
Sgalletal. makethesamepoint, thoughin adifferentguise- “The relationships
between a head(governor) and its modifications, rather than relative closeness
Formandfunction in Dependency Grammar Logic /7
(constituency) arewhat dependency grammarsarebasedon [...]” (1986)(p.136).
From the viewpoint of dependency grammar,the point that Moortgat and Mor-
rill stress is perhaps not asstriking as it may appear - dependency grammarians
havealwaysconsidereddependency to constituteadifferentdimension. Venneman
phrased this very nicely: constituency deals with horizontal organization, depen-
dency deals with vertical organization. And not only in dependency grammarthis
ideahassurfaced. For example, GPSGandHPSG alsodistinguish two separate
dimensions,asexpressedby their treeadmissibilit y conditions: ID-rules statehi-
erarchical relationships,andLP-rulesspecify linearization relations(Gazdaretal.,
1985;Pollard andSag,1993).
It is this point, that linearization anddependency structurearenot isomorphic
but represent orthogonal dimensions, that we should bearin mind. What Moort-
gat and Morril l can be understoodto argue for is not that we cannot usefunc-
tion/argument structuresto representa head-dependentasymmetry - we can,but
the directionality of the slashesneed not mirror the head-dependentasymmetry,
nor is it the casethat structural rules controlling linearization necessarily leadto
changesin dependency structure.
With that in mind, let us now turn to Hepple’s proposal. Heppleoriginally
developedanapproachof locality (head-domains)in hisdissertation (1990), based
onMorrill ’suseof unary modals to modellocality in thecontext of binding (1990;
1994). Simply put,unary modalswould marktheboundariesof a domain. Hepple
critically reflectson this approachthough in (1994; 1997), makingthe point that
his approachallowed the specification of boundaries to be decoupledfrom other
aspects of structure- thusrendering thespecification rather stipulative. Insteadof
usingunary modals,Hepplethereforeswitches for MoortgatandMorrill ’soriginal
proposalto encode informationregarding head/dependentasymmetriesdirectly on� ,&-/.&- s'� . WhatmakesHepple’sdiscussionin (1994; 1997) particularly interesting
is that Heppleaddssomeexplanatory notions like R-heads, R-dependents, anda
discussionof headdomains - all of which will prove to beuseful.
1.2.2 CATEGORIES AND COMPOSITION
How canwe definehow categoriesdo compose?To begin with, we needto define
a nonassociative calculus that serves as our basis. Again, we take a (labelled)
natural deductionformulation of whatMoortgat andOehrlecall “the logic of pure
residuation”. This logic definesthebasic behavior commonto all� , q -/. q - s q � and
07SX-/u�v�S , andis thereforesometimesalsocalled the “base logic”. Note that we do
8� Formandfunction in Dependency GrammarLogic
not yet definetheoperations on thesemantics isomorphic to theoperationson the
categories.
Definition 2 (Proof calculus of pure residuation). The proof calculus of pure
residuation, equivalent to the formal part of the nonassociative Lambekcalculus
NL, is definedasfollows.Givenanymodality n in b , thesetof modalities,
���}� ������� �Y� ��� � �7�� �%� � �2�"������ � ��� � ��� �o�}� � � �� � � � �2���m�
¡ �o�}�]¢U£/£W££££ � �¤� � �9���m� ¥� ���� � � � � �¦�
� �¤� � �y�§�}�£/£W£ ¡ �o�}�]¢£££ ¥ � ���� � �¨� � �¦�
¡ © �}�ª¢ ¡ « �p¬¦¢£££� ¡��©¯® � « �°¢²±m� ��± � � ® � ¬ª� � ® �� ¡ ��¢¯±}�
��±}� ��±}� ¥ ® �� �%� � �2�"±�� ® � �
For the unary modals we havethe following rules defining residuation. In E ³weallow for theelimination of a diamond ³=´ by a lessspecific ³2µ while retaining
mode¶ .
·"± ¸ ¥ ¹º ·y»;¼2± ¹
¼ ¸ ·"± ¹¼ ¸
¡ © ± ¸]¢£££½ ¡º�© »$¾�¢\± � ¿ ¹½ ¡ ·�¢\± �
º ·2» ¼ ± ¸ ¥=À¦Á¼·�± À Á
¼ ¸·"± À Á
¼ ¸ ¿ À Á¼º ·y»;¼9± ¸
Notethat wehavedefinedthese rulesfor Steedman-style notation of
categories.ÂRemark 2 (Pure residuation definesstrict concatenation.). Thecalculus given
in Definition 2 only enables us to modela very restrictedform of concatenation.
Formandfunction in Dependency Grammar Logic /9
Lateron, we will relax the rigidity that thecalculus imposes, by adding structural
rulesthatenable usto modify structures(representedin thelabels before theturn-
style à ) in a controlled fashion. ÂIn DGL we usethe arrows Ä , Å asa notation for headedmodes.This nota-
tion is reminiscentfrom dependency grammar(for example,seeHudson’s Word
Grammaror Mel’ cuk’s Meaning-Text Model), with the arrow pointing from the
headto thedependent- seeFigure1.1 for two examplesof dependency structures
(arc- versus tree-representation). In a structure Æ$ÇÈÅÊÉ�Ë , Ç is the headand É the
dependent,whereasin Æ$Ç{Ä~É�Ë[Ç is thedependentand É thehead.
Christopher greeted Kathy cheerfully
greeted
Christoper Kathy cheerfully
Figure1.1: Simpledependency structure
Ratherthan a single couple of modes ůÌ�Ä , we distinguish various modes,
depending on thenature of thehead(andsometimes,thatof thecomplementm,as
for ÍrÎ ). Eachmode ÏrÅ or ÄÐÏ comesfully equippedwith aproductandits residuals,
i.e. wehave Ñ�ÒUÓ/Ô?Ì/Õ Ó/Ô?Ì/Ö Ó/Ô?× and Ñ�ÒÙØ9Ó�Ì/Õ'Ø9Ó�Ì/Ö'Ø9ÓU× . Thisnaturally follows from the
basiclaw of residuation Ú>ÛÝÜÞÕ�µ@ß if f Ú�Ö�µ$ÜVÛàß if f ÜZÛÝÚ¨ÒrµXß , and-albeit
intuitively- from the fact that we regarddependency ( ÏPůÌ�ÄiÏ ) and linearization
( Ñ�Ò&Ì/Õ&Ì/Ö'× ) asseparatedimensions.
Remark 3 (Endo-/exocentricity canmodelobligatoriness/optionalit y). Follow-
ing Bloomfield,andVenneman’s discussionin (1977), we canmake a distinction
betweenanendocentric category andanexocentric category. An endocentriccat-
egory is a function category in which theheadof theresulting construction is pro-
vided by the function category itself. On the contrary, an exocentric category is
a function category in which the headof the resulting category is provided by
an argumentof the category.3 Examplesof endocentriccategories arecategories
for verbslike Æ;<áÒÐØ9âN8�³Pã98;:FäNå�æhË or Æ;<�Ò�Ø9âN8W³Pã98;:FäNåçæáËCÕ�è 8MÔ ³PéRê�: µ�ëNì :@æ . Exocentric
categories areusually assigned to adverbials (prototypically of the form <[Õ=< ) or
adjectives(prototypically of theform æ§Õçæ ).
The important point of making this distinction is that it enables us to -in a
sense- complete theaccount that DGL givesof FGD’s valency combinatorics. As
I already discussedearlier, the exact modal character of a dependency relation3Seefor similarperspectives(Malmkjæ,1996)(p.218,276)or (PollardandSag,1993).
10í Formandfunction in Dependency GrammarLogic
in a predicate-valency structuredeterminesits behaviorasan inner participant or
free modifier. Furthermore, obligatorinessof argumentsis modeled in a rather
obvious way, by including themin the endocentric category of the head they are
obligatoryargumentsto. Optionality I modelin DGL using exocentriccategories,
anda lexical meaning thataddsthemeaning of thedependent to thatof theheadit
modifies.An exocentric category for anoptionalargumentof ahead with resulting
category ß is prototypically specified as ߨÕCØ9Ó�ß or ß�ÒªÓ/ÔYß . Themeaningof the
argumentessentially is arecipethat takesthemeaning of theheadandconjoins the
meaning of theargumentto it, using theappropriatedependency relation. ÂTo illustratehow ÏrÅ or ÄÐÏ modeswork,webegin by consideringthestructures
presentedearlier in Figure1.1. Omitting informationabout dependency relations
for themoment,theproof in (8) illustrateshow thesestructureswould bederived.
(8)Christopherîwï
greetedî ð�ñxò�ó=ô$õNï¯ö@÷WõXø�ï Kathy î�ï ù ÷/úð greetedû2õXø kathöxî ð�ñxò�ó=ôüõHï¯ö ù ò úð Christopherû ó=ô$õ ð greetedû õNø kathö@ö�î�ñ cheerfully î�ñ�òMý ø ñ ù ÷ úð@ð ChristopherûDó=ôüõYð greetedû9õNø kathö@ö&û ý ø�þHÿ������ ������� öxî¨ñ
Christoper greeted Kathy cheerfully
< sc
a >
c >
Figure1.2: Simpledependency structure
Thelinear structure(thelabel) ÆCÆ ChristopherÖCØ9âX87Æ greetedÖ�8MÔ kathËCËCÖrê�ÔzÎ ����� ���������$É�Ërepresentsthestructure in Figure1.2. Oneimmediately observabledifferencebe-
tweenthe linearizedform andthe structure in Figure1.2 is that the latter is flat,
whereas the linearizedform hasmoreinternal structure in theform of bracketing.
The bracketing arises from our useof binary composition, whereas flatter struc-
tures like in Figure1.2arise from � -ary composition. Theflatterstructures give a
clearer pictureof thedomainof thehead (or headdomain).
To bridge this apparentgap,we could of courseadd � -ary implications and
products,following (Moortgat,1995), anddefinecomposition betweenheads and
their modifiersin termsof � -ary connectivesrather thanbinary ones. Therelation
between � -ary connectivesandbinary onesis simple,though: � -ary connectives
area generalization of the latter, by usingfunctional composition (“Shoenfinkel’s
Formandfunction in Dependency Grammar Logic /11
trick” ). Yet preciselybecausebinary connectivesaretotal functionsenabling one
to apply functional composition, thegeneralization that � -ary connectivespresent
addsnothing new, except for a different way of writing composition.
Remark 4 (N-ary composition in CTL). A formal calculus for Moortgat’s alge-
braicdiscussion(1995) couldbethefollowing:
��Ã���� Ñ���Ì/ß! �Ì "#"#"�Ì/߯ì�× Î� �Ãiß! %$ $ $WÎçì%Ã�߯ì & � �Ñ��%'YÎ '("#"#"'YÎ ì × � Ã)�A versionin which the type would specify the resulting type to obtain from
argumentscomingfrom theleft aswell astheright is just a notational variant and
we will thereforenot consider it here. Notethata moreincremental version of the
elimination rule above could be madeto allow for argumentsto combine in arbi-
trary order. Theseruleswould thusmimick the rules of Baldridge (1999)’s curly
bracketedtypesin Set-CCG,derived from Hoffman (1995a). However, such an
approachwould not be in keeping with CTL: word order is a phenomenonto be
modelled by structural rules, not by the baselogic. Instead,we obtain incremen-
tality throughthefollowing introduction rule:
Ñ�*,+ Î.-@× � Ã/� + ÎwÃ�ß0- 1 � �Ñ�*32hÎUרÃ��4�=Ñ���Ì/ߤ×Â
I did explore the useof � -ary connectivesin (Kruij ff, 1998a), but afterwards
had to conclude that the useof � -ary connectives led to fairly unreadablestruc-
tural rules detailing feature distribution (asusedfor examplein morphology - see51.3.1below). Therefore, I opt for a different approach,namely the oneadopted
by Hepplein (1997). To enableoneto talk of structureslike we obtained in (8)
above,Heppleintroduces the(sensible) notionsof R-headandR-dependent. Hep-
ple definesthese two notionsrecursively - in a structure like ÆCÆ$É Ä ÇRË~Å76&Ë the
(atomic) Ç is consideredto betheR-head,whereas É and 6 are Ç ’s R-dependents.4
Thedomainof aheadcanthereforebeloosely defined asthestructureof anR-
headandits R-dependents. If we would only have thecalculus givenin Definition
2, this would give rise to a very strict notion of locality, not having any structural
rulesto enable differentordering or rebracketing. Naturally, this provesto be too
restrictive to be linguistically interesting. It is then to structural rules operating
4In otherwords,
and 8 are“the ‘immediatedependents’ of the ‘projections’ of 9 .” (Hepple,1997)(p.6).
12í Formandfunction in Dependency GrammarLogic
on headed structuresthat we must turn, andseehow we cangive moreshapeto
(language-specific modelsof) head-domains and the (flexible) locality they give
riseto.
Remark 5 (Product tr eesversusprocesstr ees.). To round off thediscussionon
headedcomposition, let usconsiderthe differencebetween the linear structure in
(8), andthedependency structure in Figure1.2. Thebracketing in (8) reflectsthe
steps taken in deriving the composed structure, andcanbe used to guide further
derivationsteps.This makesthetreestructureof (8) a treerepresenting a process
perspective. On thecontrary, this is not thechief purposeof a structurelike Figure
1.2. Rather, adependency structurecanbeunderstood to representtheproduct of a
derivation. Thesedifferentperspectives(productversus process)arenot mutually
exclusive, though. It is easy to seethat if we employ Hepple’s notions of R-head
and R-dependent to interpret the linear structure in (8), then we obtain exactly
the dependency structure in Figure1.2. Hepple’s recursive notions abstract away
from theindividualstepsin derivingthecomposition of aheadandits dependents,
leaving us with binary, immediaterelations betweena headand a dependent it
governs. Â
1.3 RELATING FORM AND FUNCTION
In the previous chapter, I discussedthe issueof a sentenceexpressing linguistic
meaning. The basic componentsof a sentence’s linguistic meaning, asdiscussed
in that chapter, wereevent nuclei expressingthe aspectual categories of eventu-
alities,anddependency relations that contribute to the further specification of an
eventuality.5 Following Moensand Steedman(1988) and (Steedman,2000b), I
already briefly explainedhow aspectualcategoriesrelateto theirexpression(form)
in verbaltenseandaspect (seealso(Steedman,2000b)). In the current section, I
focus on the relation betweenform anddependency relations - by what formsare
dependency relationsrealized,or conversely, how do we recognizeby theform of
a word group whattypeof dependentit is?
Within thePragueSchoolof Linguistics, the issue of the relation between the
form of a word and the function of the word’s meaning in the underlying mean-
ing of thesentencehastaken in a central place ever sincethepioneering work by
-amongothers-JakobsonandMathesius. Particularly illustrativeof theimportance
5In thesecondpartof my dissertation,a third componentis added,namelyinformationstructure- following (Sgallet al., 1986).
Formandfunction in Dependency Grammar Logic /13
of distinguishingtheform-function relation is Mathesius’ programmaticcontribu-
tion to theTravauxin 1936. There,Mathesiusoutlinestheadvantageof describing
natural languagegrammarsfrom theviewpoint of functionsrather thanforms(criti-
cizingJespersen’sEssentials ontheway)sinceit is functionsthat aresharedacross
languages:
“If we areto apply analytical comparison with profit, the only way of ap-
proachto different languagesasstrictly comparablesystemsis thefunctional
point of view, sincegeneralneedsof expressionandcommunication, com-
monto all mankind, aretheonly common denominatorsto which meansof
expressionandcommunication, varying from languageto language,canbe
reasonably brought.”
– (Mathesius,1936)(p.95/306)
Here, I build forth on ideas worked out in the PragueSchoolof Linguistics
during theinterbellumby Jakobson,Mathesius,Skalicka,andTrnka,work thatset
thebasis for laterwork by Danes, Dokulil, Kuryłowicz, andSgallandhis collab-
orators - cf. (Sgall et al., 1986)(52.10) and(Panevova, 1994). The principal idea
is to distinguisha morphological category of Case, or abstract case,from actual
morphological strategies. This distinction is similar to the ideaof abstract casein
Government& Binding theory, cf. (Haegeman, 1991).6 An abstract casemediates
betweendependency relations and morphological strategies that express depen-
dency relations. The key idea is that theseCasesabstract away from language-
specificform. Eachlanguagehasits own morphological exponents (Trnka, 1932)
or morphological strategies(Croft, 1990) to express thedifferentabstract Casesat
thelevel of surface form, andthusthedependency relationsassociatedwith them.
For example, a Patient is related to the abstract caseAccusative, andacross lan-
guages we find different ways in which the Accusative canbe expressed:Czech
andGermanuseinflection,Japaneseusesanaffix 24: , andEnglishhasa particular
(canonical) position.
6Two sideremarksshould be made. First of all, Case naturally concerns those dever-bitive/denominative dependentsthat arethemselvesnominalor adjectival groups. Secondly, Caseshouldbe kept apartfrom morphological categorieslike number, genderor delimitation. Sgall etal mentionin (1986)only numberanddelimitationasmorphological categoriesfor nouns(pp.172-173).Mathesiusdiscussesin (1936) four morphologicalcategoriesfor nouns,namelynumber, total-ity, definiteness, andqualitativegender (asopposedto purelymorphological gender). Sgallet al.’sdescriptionof delimitationbasicallycoversMathesius’totality anddefiniteness(wheretotality is thedistinction illustratedby French.un pain, du pain, les pains, despains). The inclusionof genderhereasa morphological category is not Mathesius’qualitative gender, but morphological gender-his qualitative gendercorresponds to our distinctionof genderat thelevel of lexical meaning.
14í Formandfunction in Dependency GrammarLogic
It is worthwhile to reflect a bit on the picture that thus arises. At one end,
we have dependency relations that areessential to structuring linguistic meaning.
At the other far end, we have the outer form of sentences. Now, this apparent
chasm between form anddifferent argumentroles hasbeencriticizedby various
authors,e.g. Dowty, Davis, andWechsler. How could onepossibly recognizethe
role of a particular argument?Theanswerpresentedhere,having its rootsin work
done in the PragueSchoolof Linguistics since the 1930s, is that we candistin-
guish language-specific morphological strategies that realize language-universal
dependency relations, and their relation is mediatedthrough language-universal
morphological categories.7
Let meconsidera few additional examplesto illustratewhatwe have in mind
here (cf. also (Kuryłowicz, 1964; Sgall et al., 1986; Sgall et al., 1996)). The
dependency relation Patient is mostly expressedby an accusative case,which in
Czechis reflectedby inflection:
(9) Czech
HonzaHonza-NOM
koblihudonut-ACC
snedlate
.
“Honzaatethedonut.”
(10) Japanese
Susi-oSushi-ACC
Taro-gaTaro-NOM
tabeta.ate
“Taroatesushi.”
Neither in Japanese(10) nor in Czech(9) the expression of accusative caseis
dependent onwordorder. Thisstandsin contrast to analytic languageslikeEnglish,
which do not have anaccusative inflection but realize theaccusative casethrough
placing thewordform in thedirect complement position (directly after theverb):
(11) English
Christopherreadthethe
book.book-ACC
7Although it might be temptingto saythatmorphologicalcategories“realize” dependency rela-tions, this wouldn’ t be correct. It is themorphological strategiesthat realizedependency relations,andwe view the relationbetweenmorphological strategiesanddependency relationsasmediatedthroughabstractmorphological categories.
Formandfunction in Dependency Grammar Logic /15
Similarly, theActor dependency relation is prototypically realizedby a nom-
inative case(i.e. in sentencesin active voice). Again, synthetic languageslike
Czech(12) or Japanese(13) make useof inflection, whereasanalytic languages
like English(14) employ word order, indicatingnominative caseby placementin
subject position:
(12) Czech
HonzaHonza-NOM
koblihudonut-ACC
snedl.ate
“Honza atethedonut.”
(13) Japanese
Hanako-gahon-okatta.
Hanako-NOM book-ACCbought
“Hanako bought a book.”
(14) English
KathyKathy-NOM
despisesJohnWaynemovies
Otherexamplesaretheuseof thedative caseto realize theAddr esseedepen-
dency relation. In Japanesethedative is formedusing the -ni postposition, Czech
hasa dative inflection, whereas in Englishandin Dutchdative caseis reflectedby
placement in the indirect object position or useof a function word like English
“to” or Dutch “aan”.
All theexamplesabove illustratetheprototypical useof cases to realizea spe-
cific dependency relation. Kuryłowicz (1964) calls thesedependency relationsthe
primary functions of the respective cases - i.e. a case’s primary function is that
dependency relation which it usually realizes(p.16). Opposite to a case’s primary
function is its secondary function - that dependency relation which it can real-
ize aswell, but only in what Sgall et al. call “contextually conditioned items” in
(1996)(p.71). For example, Sgall et al. consider the following oppositions in the
useof theaccusative - once realizing its primary function, Patient (15), andonce
its secondaryfunction, theTime:How Long (16):
(15) Christopherreadtheentire book.
(16) Christopherreadtheentire night.
16í Formandfunction in Dependency GrammarLogic
Sgall et al. alsomention that the Accusative hasin Sanskrit as its secondary
function Dir ection:Where To:
(17) Sanskrit
vanamgo-3-SING
gacchatiforest-ACC
(Sgallet al., 1996)(p.72)
“S/hegoesinto theforest”
Theinterestingpoint about (17) is thatin bothGermaniclanguagesandSlavonic
languagestheAccusative displayssimilar behavior, whencombinedwith particu-
lar propositions.For example, in Germanthepreposition auf whencombinedwith
anominal group in accusativecaserealizesDir ection:WhereTo (here: “onto X”),
andsodoestheCzechproposition na whencombinedwith anaccusative.
To recapitulate, we make a distinction betweenmorphological strategiesand
morphological categories.Weconnectdependency relationsto morphological strate-
giesthroughmorphological categories,wherebywe candiscernprimary andsec-
ondary functions for the latter. Morphological strategies are languagespecific,
andmorphological categoriesareassumedto belanguageuniversal8. Puttogether,
we not only advancethe hypothesisthat with this setupwe can explain the re-
lation betweena sentence’s form and its linguistic meaning. Equally important,
the intention is to present an account that might find a validity that applies cross-
linguistically - andwith that,it goes well beyond theaccountsof Wechsler (1995)
or Davis (1996).
In section51.3.1belowI discussmorphological strategiesin somemoredetail,
explaining how for examplecasemarking, adposition, positioning or linking can
be modeled in DGL. Section51.3.2 continuesthe story: Here, I detail out how
one derives the category of a word, given its lexical meaning. Finally, section51.3.3roundsit all of, completing thecalculuspresentedearlier suchthatlinguistic
meaning is built compositionally.
1.3.1 MODEL ING MORPHOLOGICAL FORM
Following Croft’s discussionin (1990)(Ch.2),onecandistinguishfor examplethe
following morphological strategies(18) that a languagemay employ to realizea
particular morphological category.
8Thoughthey neednot alwaysbe“available” in a particularlanguage.
Formandfunction in Dependency Grammar Logic /17
(18) a. case: The useof bound morphemesor casemarkers to indicate the
morphological category.
b. adposition: A morphological category is signalledby a function word
affixedto thewordform.
c. positioning: Thewordform’s position in theclause,relative to for ex-
amplethemainverb,is anindicationof theunderlying morphological
category.
d. linker: A linker is an invariant marker, or morpheme,that relatesthe
modifier and the modified. Unlike the above three strategies, linkers
arenotusedfor verb-nounmodification;only noun-noun modifications
arelinked.9
Thesestrategiesareillustrated in (19) through (22). Particularly (22) is inter-
estingsinceit exemplifieshow strategiescanbecombined.10
(19) positioning
a. English
Elijah wrotea letter, accusative(directcomplementposition)
b. Dutch
datElijah Kathy eenboekgaf,dative(indirectcomplement position)
(20) case
a. Czechknih-a, nominative; Czechknih-u, accusative
b. German desKind-es, genitive
(21) adposition
a. Dutch aan Kathy, dative
b. JapaneseKathy-ni, dative
(22) linking
a. English
Elijah’s cowboy-boots,genitive
9Note that if a linker morphemeis usedonly for the possessive, and not for either predicate-argumentrelationsor any othermodifier-nounrelation,thenit maybedifficult or evenimpossibletodistinguisha linker from e.g.a casemarker or anagreement marker; cf. (Croft, 1990),p.32.
10Note: exceptfor theJapaneseexamples,the‘-’ in eachexampleonly servesto illustratethecasemarker separatelyfrom theroot. Normally, no hyphens areused.
18í Formandfunction in Dependency GrammarLogic
b. English
thecowboy bootsof Elijah’s, genitive
(linking+adposition), (Croft, 1990)(p.33)
In thenext sectionswe discusshow we canmodelin DGL themorphological
strategiespositioning (51.3.1), case(
51.3.1), andadposition (
51.3.1)
Mor phological strategies: Positioning
I canbe fairly brief about how to modelpositioning in DGL. The reason is that
a system like the Lambekcalculus by its very nature provides all the necessary
ingredients - namely, composition and type-raising. Type-raising is the creation
of a category C’ from a category C suchthat C’ is a function that takes as its
argumenta function that takes C as its argument. CCG includesrules for type-
raising in its basic calculus (theT-combinator, cf. (Steedman,2000c)(p.43ff)). In
categorial typelogic, type-raising is a theoremfor all thosemodesthathaveaccess
to (full ) associativity (cf. (Oehrle, 1994; Moortgat, 1997a)). Example(23) gives
anillustrationof type-raising.
(23) æ ;<+ type-raising- = <[Õ\Æ;<?Ò�æhËSteedmanproposesto usetype-raising to model positioning. The intuitions
aresimple, andcanbe illustratedon example(23). What the type-raising in (23)
effectively doesis turning the noun into a category stating that the nounshould
appear in subject-position. That is, the category specifies the noun asnominative
“case” , asillustratedin (24). Note that we make useof the specification of > as
Ú�Î@?A:�� in diamond elimination.
(24)
BDCFEHG À ÁFIKJMLON ¹QPMRº BDCFEz»
IKJ�LSN G ¹ P R T ¿ À¦ÁVUº BDCFE%»SW�XZY[G ¹QPMR]\<^`_Fa
T b G RcUedº b »Sf Ihg XSijG ¹ f Ihg XSi R T k ¹lU Tr m Gon Á J�Ihg B í�p LOIZq Uhrº b »Of Ihg XZilsut LOIZv G \ T ¿ í
Uº BDCFE%»SW�XSYwsut LOIZv G \ T ¿ ¹lU
º BDCFE%»SW�XSYxG ¹ f Ihg XSi Rzy t LOI@{ ¹ f Ihg XZi R í t LOI \}| T kyFU~r
BDCFEHG À¦Á W�XSY { ¹ f Ihg XZi Rzy t LOI { ¹ f Ihg XZi R í t LOI \}|Z| T kÀ¦ÁVU
Similar categoriescanbe specified for other “cases”, like the English dative
(indirect object position) andaccusative (directobjectposition).
Mor phological strategies: Case
Heylen (1999) proposesanapproachto handling featural information in thesetting
of categorial type logics. The leading partsin Heylen’s approachareplayed by
Formandfunction in Dependency Grammar Logic /19
namedinstancesof the unarymodaloperators ³ and ��� . The ideais to give the
boxes names,like we give namesto modes. Particularly, as nameswe can use
thenamesof morphological features,like fem(feminine), �&�Π(accusative) andso
on. Then,by prefixing a type with such boxes,we canspecify its morphological
features.For example, (25) states that to thetoken Czech“kniha” (English book)
we canassignthetypespecifying “kniha” asa feminine nounin nominative case,
singular.
(25) ���%�Z�D�oÃ�� ��� ëZ� � ��ì ä � � � â µì�� �The � � ’s comeinto play in a proof by eliminating themfrom the type, thus
introducing themasexplicit information in the structure. For example,applying& � � to ���%�Z�D��Ã�� ��� ëZ� � ��ì ä � � � â µ�ì�� �����.:�:�� leadsto thefoll owing:
(26)
���%�Z����Ã�� � � ëZ� � � ì ä � � � â µì�� � & �j�� ���%�Z�D�<� � ëZ� Ã�� � ì ä � � � â µ�ì�� � & �u���� ���%�Z����� � ëS� � ì ä � Ã�� � â µ�ì�� � & � ������ ���%�A�D�<� � ëS� � ì ä � � â µì�� Ã��Now thatwe have themorphological informationexplicit in thestructure,we
canoperateon it. For example, asHeylen showed,it is fairly straightforward to
allow for underspecification. The basicideathereis that we introducestructural
rules that enable us to rewrite the nameof a unary modal operator into another
name,in the appropriate context asspecifiedby the structural rule. For example,
consider � � ì��V�,�@ë å asmeaning thatthetypeis underspecified for �%�����F�V� - i.e. the
tokencanbeinterpreted asbeing eithersingular or plural. Thetoken“sheep” is an
examplein case- �l� ì��V���@ë å � meansthat“sheep” is anoun, either pluralor singular.
Then,structural ruleslike the following canbeusedto specify theunderspecified
feature to either singular or plural (the ³ ’s correspondto the angular brackets� $��
in thestructure):
(27) a. +����Q��Î�Õ��%�����F� ��ÌWÍ��O�Q��- ³Pâ µ�ì�� Ú Û ³ ì��V�,�@ë å Úb. +�������Î�Õ��%�����@�V�UÌ����~����- ³Q�V� � å ÚpÛ ³ ì������@ë å Ú
Usingthefirst rule,wecanfor exampleinfer:
(28)
Í��������5Ã�� � ì��V���@ë å � & � �� Í��������Q� ì������@ë å Ã�� +����Q��Î�Õ��%�����F� ��ÌWÍ��O�Q��-� Í������ �Q� â µ�ì�� Ã��
20í Formandfunction in Dependency GrammarLogic
Note that if we would continue the proof with a1 �D� step,we could derive
the type � � â µì�� � for “sheep”. Hence,oneof the nice advantagesof this kind of
underspecification is that it enablesusto introducelexical generalizations. Rather
thanhaving separatelexical entriesfor “sheep”asa singular noun,and“sheep” as
a plural noun,wecanhave just onelexical entrydefining “sheep” asa nounthat is
underspecified for number.
Naturally, we should be able to modelmorecomplex casesaswell. For ex-
ample,consider the Czech“knihy”. This form’s caseandnumber areambiguous
between genitive singular, or plural with either nominative or accusative. A cate-
gory for “knihy” would thushave both the �%��� and ÎF�\Í�� featuresunderspecified
(e.g. ¡�� � ëZ� ¡�� ì���� ¡��V¢;êçâ ë æ ). Subsequently, we would needstructural rules that
specify e.g.tuples of featuresratherthansinglefeatures,like in (29).
(29)
���A � â µ�ì�� � �/ëXì Û ���
A � ì��V� � ¢Xê/â ë���A � �V� � å � ì ä � Û ���
A � ì���� � ¢;ê/â ë���A � �V� � å � ê.¢O¢ Û ���
A � ì��V� � ¢;ê/â ëA proper consideration of suchmorecomplex casesis presented in (Heylen,
1999)(Ch.8),whereHeylen discussessortal hierarchies that control the specifica-
tion of feature structures.
The way agreement is modeledmimics, in a way, agreement by unification.11
Namely, the idea is that a composite structure can be “assigned” a feature � if
andonly if both of its componentshave that feature � aswell. Thus,an abstract
structuralrule for agreement(concord)would look somethinglike this, in Heylen’s
theory:
(30) ³ � Æ°Ú Ö�ÜÞË]Û ³ � Ú~Ö�³ � ÜIn DGL, wherewe considerheadedness asan inherentaspect of composition,
I employ slightly differentabstractstructural rulesfor handling agreementin gen-
eral:12
(31) a. ³ � Æ°Ú~Ö�Ø[ÜÞË]Û ³ � Ú ÖwØD³ � Üb. ³ � Æ°Ú~Ö�Ô[ÜÞË]Û ³ � Ú ÖwÔD³ � Ü
11Although we shouldhastento say that the model is inherentlymore powerful than the waymorphological informationis donein HPSGor featurelogics in general,andthatwe arenot doingunification.Seealsobelow.
12Therulesin (31) do not intendto cover specificsecondarycasessuchastheremaindersof theDual in Russian,or Czech“kote a stene si hraly” insteadof “hrala”.
Formandfunction in Dependency Grammar Logic /21
Let usconsideranexample.Take thelexical entry �V�`?{Ã������V� � å Æh�§Ò�Ø[Í Ë , and
try to prove that “Sheep eat” is a sentencewith a verbal headthat is plural. For
agreement (plural) we usethefollowing structural rule:
(32) ³��V� � å Æ°Ú Ö�Ø[ÜÞË]Û ³Q��� � å Ú Ö�ØD³Q��� � å Ü
(33)
ôK£�¤�¤¦¥4§z¨@©�ª.«�¬Q ® ¨ ©¯ ô�£�¤h¤¦¥�° ªM«�¬ §� ± ² ¥ ¤$õK³O@´MµQ¶¦¤�·O¸�¥�¹º´M·K»¯ ôK£�¤�¤¦¥�°�¼ ½ «�¾ §4¤ ý ¿ §z¨@© ¼A½ «�¾MÀ �¥<Á�Â'ôKà ® ¨ ©¯ ¤ ý�¿ °�¼A½ «�¾ § À QÁ�ÂÙôKà ® Á¯ ôK£�¤�¤¦¥�°�¼A½ «�¾<Ä Â ¯ ¤ ý�¿ °�¼A½ «�¾ §Rô Å�Æ ·S³h¥�¹º´M·K¸°ó¯ ô�£�¤h¤¦¥ Ä Â�¤ ý�¿ °�¼A½ «�¾ §[ô Ç ¨ ©ôK£�¤�¤e¥ Ä Â�¤ ý�¿ §È¨@© ¼A½ «�¾ ô
Thereis animportant observation thatwehaveto makeabout theproof in (33).
As said,wetried to prove �}���V� � å Í - our“goal type”. Theobservationconcernshow
the � � �V� � å in thegoal typeactually enforcestheagreement.For that weshould read
theproof in abottom-upway. Thegoal type is obtainedby introducing the �V�M�V� � å ,which is only possible if the entirestructure indeed carriesthe corresponding di-
amond(i.e.� $�� �V� � å ). For that to be possible, the agreementrule positsthat both
componentsof the structure have to be decoratedwith that diamond - thus, both
“sheep” and“eat” have to be labeled asplural (���h��� ). Which they are- “eat” by
lexical assignment,“sheep” by specification of an underspecified feature assign-
ment( �%�����F� � ). In other words, subjectandverbagree, andall is well indeed. But
notewhatwould havehappenedif wewould have hada subject in singular. Going
bottom-up,theagreementrule would have posited therebeing a subject in plural,
whereasfrom the lexical entry for the subject we would have obtained singular
case(top-down). This clashwould have resulted in theproof falling through - we
would not beableto derive thegoaltype.13
Themodelso far is asHeylen discussedit his dissertation (Heylen, 1999)and
various papers, for example(Heylen, 1997). In (Kruijf f, 1998a) I introducedan
“extension” to Heylen’s model, in the form of “a-symmetric distribution rules”.
Heylen’s agreementrules, aswe saw themabove, areessentially rules that sym-
metrically distribute a label� $�� � from a construction over its two components.
However, whenoneappliesHeylen’s modelto morphologically rich languages, it
quickly becomesapparent thatwe needdifferentkinds of structural rulesto man-
agefeatureinformation- notonly symmetricdistribution rulesdefiningagreement.13Unless,of course,we would have themorecomplex casesof coordinated,singularsubjects- as
thesewould leadto aplural construction.
22í Formandfunction in Dependency GrammarLogic
Because, it may very well happen that a component carriesmorelabels thanare
neededfor aparticular typeof agreement.For example,averbmaycarryinforma-
tion about ? �V�Y�� , which is informationnot relevant to agreementwith its subject.
This leadsto structuralrulesthat allow labelsto percolateupwards: if a headhasa
feature � that is not relevant for agreement,thenwe canpercolatethat feature up-
wards,distributing it over theentirecomposition. Theserulesarethusa-symmetric
in thatonly onecomponent will berequiredto havesomeappropriate labeling,not
bothcomponents.14
As amatterof fact,eventhoughpercolationrulesarestraightforwardto specify
in categorial type logics (andDGL), they presentan important linguistic general-
ization (together with the way agreement is handled) that feature logics as em-
ployed in HPSG are not capable of capturing (Oliva, p.c.). Namely, for unifi-
cation to work, eachfeature used in a particular type of agreementneeds to be
listed explicitly in the attribute-value matrix. This caneasily leadto doubling of
informationabout morphological information, andto ensure “consistency” theat-
tribute/value pairs areco-indexed to indicate that the values should be identical.
But this is not particularly satisfying - it seemscounterintuitive to have to specify
various attributes� in oneandthe samelexical entry. Our approachto handling
morphological (featural) informationby meansof symmetricagreementrules and
a-symmetricpercolation rulesleadsto a muchmoreintuitive picture: Specifythe
informationonce, andonly once. We cando so becausewe arerewriting, rather
thanusing unification.
Mor phological strategies: Adposition
Finally, let usdiscussadposition. As we already mentionedabove, we understand
by adposition theuseof function wordsto indicatewhatmorphological category a
wordform realizes. Examplesof suchfunction wordsarementionedin (34).
(34) a. Prepositions:
i. English
“to” (+N, dative), “of ” (+N, genitive)
ii. Dutch “voor”/“ aan”(+N, dative), “van” (genitive)
b. Postpositions:
i. Japanese“-ga” (+N, nominative), “-o” (+N, accusative)14Notethatthey aresimilar to the É!Ê�÷.ÉzË rulesdefinedin (Moortgat,1997a).
Formandfunction in Dependency Grammar Logic /23
ii. Korean“-ka” (+N, nominative), “-lul” (+N, accusative)
Again,theapproachto modeling thesephenomenain DGL is relatively straight-
forward.Wedistinguish modes��: Í�?¦Å Ì�Ä������ � for composition between thenoun
andits postposition or apreposition,respectively. Example(35) illustratesthepro-
totypical categoriesfor prepositionsandpostpositions,respectively.
(35) a. Preposition corresponding to caseß : ¡%��Ì7æ§Õ � å ë � æb. Postposition corresponding to case ß : ¡ � ÌPæhÒ � äCâOÍ@æ
Analyzing a combination of strategies
Although we canmodeleachof thedifferentmorphological strategies,how would
they cooperatein ananalysisof a sentencein which variousmorphological strate-
giesareusedat thesametime?For example,consider thesentencein (36).
(36) English
Elijah reads Christopher’s book to Kathy.
The sentence in (36) illustratesthreestrategies: positioning (Elijah, book),
linking (Christopher’s), andadposition (to Kathy). For a proof for (36), the most
illustrative stepsin this context aregivenin (37) below.
(37) a.
¿4ÎeÏeÐ�Ñ�Ò G À¦ÁFIKJ�LON�¹ P Rº ¿ÓÎeÏeÐ�Ñ�Ò » IKJ�LON G ¹ P R T ¿ À¦ÁVUº BDCFEz» W�XZY G ¹ P R \�^�_Fa
T b G RÔU#dº b »Sf Ihg XSiuG ¹ f Ihg XZi R T k ¹}U Tr m GÕn Á J�Ihg B í�p LOIAq Uhrº b » f Ihg XSi s t LSI v G \ T ¿ í
Uº ¿ÓÎeÏeÐ�Ñ�Ò »SW�XSY3s t LOI v G \ T ¿ ¹lU
º ¿ÓÎeÏeÐ�Ñ�Ò »SW�XSYÖG ¹ f Ihg XZi Rzy t LOI { ¹ f Ihg XZi R í t LOI \}| T ky@U r
¿ÓΦÏ#Ð�Ñ�Ò G À¦Á W�XZY { ¹ f Ihg XZi Rzy t LOIM{ ¹ f Ihg XSi R í t LOI \}|S| T kÀ¦Á�U
b.
chris GÕn Á IKJ�LONW¹ P B ‘s GÕn Á IKJ�LON�¹ P B íF× ¼ W�Ø@Ù { n ÁFIKJ�LSN�¹ P B y p J�Ú ¾ n ÁFIKJMLON�¹ P B |chris s × ¼ W�Ø@Ù ‘s GÕn Á IKJ�LONW¹ P B y p J�Ú ¾ n ÁFIKJ�LSN�¹ P B
T í ¿ Ubook Gon Á IKJ�LON�¹ P B{
chris s × ¼ W�Ø@Ù ‘s| s p J�Ú ¾ book GÕn ÁFIKJMLON�¹ P B T y ¿ U
c.
to Gon Á Ú�J g ¹ f Ú�Ú i NSLOLONON B y p�Û i N Û n Á IKJ�LON ¹QP B kathy Gon Á IKJ�LSN ¹QP Bto s p�Û i N Û kathy Gon ÁFÚ�J g;¹ f Ú�Ú i NOLOLONSN B
T y ¿ U
24í Formandfunction in Dependency GrammarLogic
1.3.2 DGL’ S LINKING THEORY
Let mebegin by addressing theissueof a linkingtheory in moredetail. Oneimpor-
tantaspect of theaccount I give hereis that it overcomes thekind of criticism that
hasbeenlevied both againstapproaches within dependency grammar(like Fill-
more’s - cf. (Panevova, 1974;Sgall et al., 1986)), andagainst similar approaches
based onspecifying semanticsin termsof Ü -frames(cf. Davis’sdissertation (1996),
andreferencestherein to discussionsby Wechsler andDowty). Thesecriticismsall
comedown to therebeing no obvious relation between lexical meaning andform
(or syntacticbehavior). Putdifferently, themeaning is renderedstipulative at least
from theviewpoint of therebeing no relation between different syntacticbehavior
(form) anda differentiation in meaning.
This criticism is overcomein DGL by realizing therelation betweenform and
function asmediatedby morphological categories,a relation thathasbeenpointed
out and elaboratedwithin the PragueSchoolof Linguistics ever since the early
1930’s. The reason why we indeedovercomethe criticism, rather than replace
it by another stipulative account, is simple. Morphological categories present a
cross-linguistic generalization of the intuitive relation between function and ab-
stract form, andthey canbestraightforwardly related to themorphological strate-
giesof a particular language.This meansthatmorphological categoriesnot only
capture intuitionsabout languagesinvestigatedby variousmembersof thePrague
school (notably, Slavonic languages, Germaniclanguages,English) - becausethey
are abstract and presentcross-linguistic generalizations, they also make predic-
tions aboutlanguagesthat have not beeninvestigated from this point of view. The
prediction is that,if wecouple themorphological strategiesof a“new” languageto
themorphological categories,thenweexpect thesameintuitionsabout therelation
between form and function to be verified (i.e. the morphological strategies help
realize the sameprimary andsecondary functions of the morphological category
asobserved in other languages). Ratherthanbeing stipulative, theaccount is ver-
ifiable - cross-linguistically, even though the repertoiresof categoriesmay differ
from onelanguageto another.15
To formulatea linking theoryfor DGL, I start from the basic approachadvo-
cated by categorial type logics. Categorial type logic providesa fairly straightfor-
ward modelof lexical semantics. The semantic Ý -term of a word is related to a
15With that, thepresent approachpresentsan account of the relationbetweenform andfunctionthatis morefundamental thanfor exampleWechsler’sapproach, which doesnotappearto lenditselfwell to cross-linguisticgeneralizations.
Formandfunction in Dependency Grammar Logic /25
syntactic type using a Curry-Howard correspondence(cf. for exampleOehrle’s
articles (1994; 1995)). A suitable namefor this kind of modelof lexical semantics
would be logical lexical semantics, which maybeunderstood astrying to explain
therelation betweenform andfunction/meaning in a logical way.
According to someauthors, the Curry-Howard correspondence in categorial
typelogic should take theform of anisomorphism.Thatway categorial typelogic
would answera mathematical-logical ideal. Thus,according to Oehrle’s (1994),
syntactic categoriesandtyped Ý -termsarerelated asperDefinition 3.
Definition 3 (Categories and semantictypesin CTL). Givena mapping that re-
latesbasic categories Þ[ßáà with a corresponding type ?NÉ���Þ in a typedÝ -calculus.
Theassociationbetweenthefull setof categories,built usingthecategory-formation
operators Ñ�Ò@â9Ì/Õ�â9Ì/Ö�â�× , andsemantic typescanthenbedefinedasfollows:
ã associateeach implicational typewith argument Ú andresulting category Ü(like ÜÞÕVâ&Ú²Ì�ÜÞÒ�â&Ú ) with the Ý -type ?NÉ��7ưگ˨Ûä?HÉ��7Æ°ÜÞË , i.e. functions from
?HÉ��7Æ°Ú¯Ë to ?HÉ��7Æ°ÜÞË ;ã associate the product typewith first projection Ú and second projection Ü
with thepairing of ?NÉ��7Æ°Ú¯Ë and ?NÉ��7Æ°ÜÞËSeealsoHepple’s (1994; 1995), andMoortgat’s (1997a). Â
However, I would like to arguethatanisomorphic mapping betweensyntactic
typesand Ý -typesdoesnotappearto bedesirablefrom alinguistic viewpoint.16 Be-
cause, if therewereto exist anisomorphismbetween argumentsthatareobligatory
from theviewpoint of proper “grammaticaluse”,andargumentsthatareobligatory
from theviewpoint of meaning (determining “inferrableinformation”or something
similar), thenevery syntactic argumentshould be reflected in the semantics, and
vice versa.This neednot betrue, in eitherdirection.16Even thoughwhat we argue for resultsin a loss of the isomorphismbetweensyntactictypes
and å -types(or formulasandtypesin the generallogical setting),we neednot losethe possibilityto obtainanisomorphismbetweenproofsand å -terms.TheessentialideabehindtheCurry-Howardisomorphismis that, due to the isomorphism betweenformulasand types,we can alsoobtainanisomorphismbetweenaproofof a formulaandatypedterm.Thelatterisomorphismgivesriseto thepossibility of reconstructingthe å -termoncegiven the proof, andvice versa,given the å -term,wecanreconstructtheproof. It appearsto methat,despitethelossof anisomorphismbetweenformulasandtypes,we canstill obtaintheisomorphismbetweenproofsand å -terms,by:æ addinginformationto the å -term,recordingexplicitly eachstepthatis taken;and,æ requiring that the axiomswe begin with (or reasonbackwards to) are identifiable lexical
items.
26í Formandfunction in Dependency GrammarLogic
Expletive pronouns In various languagesthereareverbsthat require an ex-
pletive pronoun to function assurface subject. A niceexample is theGermanverb
“geben”, which requiresanexpletivepronoun“es” to form asentencelike “Es gibt
einen Student im Kino” (& �ç" “There is a student in the cinema”). Now, if we
would indeedhave an isomorphismbetweenthesyntactic type of “geben” andits
Ý -term,we would have an argumentposition in the Ý -term for the expletive pro-
noun aswell, whichshould befilled by whateversemanticswewouldassign to the
expletive pronoun. But, linguistically speaking, this seemscounter-intuitive. The
expletivepronounis neededto form agrammatical sentence- but semantically, the
expletive canbe argued to be vacuous (cf. the discussion in Sgall et al. (1986)
about function words,andalsoother approacheswhereexpletivesareconsidered
to make no realcontribution to themeaning of sentences- cf. HPSG(Pollard and
Sag,1993), (combinatory)categorial grammar(Jacobson, 1990)).
Oehrlepointed out (p.c.) that onecould perhapsargue for an argumentslot
for theexpletive pronoun if thepronoun wereunderstood asreferring to thelarger
situation in which the event is placed. Erteshik-Shir does in fact present suchan
approach,basedon a very literal interpretation of Heim’s file-change metaphor.
However, given theapproachwe take hereto specifying meaning, theneachverb
canbe given an interpretation relative to world-time pairs. In otherwords, asan
eventuality set in a specific time and place. Trying to understand the expletive
pronounasestablishing that givenagainseemsto beredundant,then.
Relational nouns Oneneednot only consider the isomorphismin the direc-
tion from syntactic typeto Ý -type. An example showing why thereneednot bean
isomorphismfrom Ý -type to syntactictype is providedby relational nouns. These
nouns have semantic argumentsandyet neednot to subcategorize in order to be
usedgrammatically (cf. (Sgallet al., 1986)). Consider thenoun“brother”, which
hasas its semantics �.��:�?����V��Æ$ÇRË with (at least) the necessary argument brother-
of( ÇYÌCÉ ), i.e. Ý�Ç%Ý�É�"�ÆK�.��:�?����V�xÆ$ÇyË(èw�.��:�?����V�é2ê:��]Æ$ÇYÌCÉxË . Paraphrased,whenever a
person Ç is a brother, he is necessarily a brother of someother person É , ÇHë;5É ;
onecould extendthis by saying that Ç is alsoa sonof 6 , with Ç�ë;BÉìë;�6 . But
thesyntactic typeof “brother” doesnot needto subcategorize for a syntactic type
corresponding to theargument É , or 6 , in order for “brother” to beusedgrammat-
ically. As Sgall et al. (1986) point out, the extra argument É is therebecausein
the discoursecontext it should be answerable who Ç is a brother of. Onecannot
sensibly utter that “John is a brother” (understanding brother in the family-sense)
without being ableto answerthequestion “Who is Johna brother of?”, asper the
Formandfunction in Dependency Grammar Logic /27
dialoguetest.
Raising verbs Finally, variousframeworks give anaccount of raising verbs in
which their syntactic type(s) do not correspond isomorphically to their semantic
argumentstructure. This couldcount asanother argumentagainst a strict isomor-
phism,though this very muchdependson one’s linguistic intuitions. For example,
Jacobson(1990) doesgive analternative account of raising using function compo-
sition, basedon a combinatory form of categorial grammar,andpresentsevidence
for thataccount.
Thus,themodelof lexical semantics thatcategorial typelogic providesuswith
appears to be too strict to enable us to present particular linguistic intuitions one
might have. We need to relax the isomorphism criterion betweensyntactic types
and Ý -typesin orderto beableto capture(at least)thecaseswementionedabove.17
DGL’s linking theory is definedasfoll ows.
Definition 4 (DGL’s linking theory). Given a mapping B betweenbasic cate-
gories Þ andthesortsusedfor specifying lexical meaning(e.g. objectsandvarious
kindsof eventualities),a mapping M betweenmorphological categoriesandkinds
of dependency relations(i.e. themorphological categories’ primaryandsecondary
functions),anda mapping S betweenmorphological categoriesandmorphological
strategies.
1. Givena predicate-valency structure for an eventuality, of theform
ÆhíîèðïQñ�è � > ��Æh� Ëlè3$ $ $�è � > µ ��Æh� µ ËCË ,specifying theobligatory arguments for ï . Thesyntactic category correspond-
ing to this predicate-valencystructure is built asfollows.
First, theresulting category is a basic category ò , mappedby B fromtheeven-
tuality in í . Setthepredicate-valencystructure’s category ó to ò . Then,going
fromleft to right throughtheconjunction, for each argument� >#ô���Æh�}ô=Ë , weuse
B to map the sort of the nominal ��ô to a category õ . Then,we also extend
ó with ó^Õ�âk³(ö ÷�õ or ø��HÒ�âx³(ö ÷�õ , depending on (a) canonical surfaceordering,
and(b) S.
2. Givena predicate-valency structure basedon a nominal ù that is not an event
nucleus,of theform Æhùúèûï�üýè � > ��Æ þrË%è�$ $ $�è � > ´ ��Æh� ´ ËCË . First, theresulting
17As a matterof fact, on the dependency grammarpoint of view all function words provide acounterexample,sincethey would bemodeled(in DGL) asfunction typesbut their contribution tolinguistic meaningwould bephrasedentirelydifferently(if it all).
28í Formandfunction in Dependency GrammarLogic
category is a basic category ò , mapped by B from ù . Setthepredicate-valency
structure’s category ó to ò . If anyof thearguments specifiedby thepredicate-
valency structure are required to be mappedto syntactic categories for the
word to beusedgrammatically, thenfollow thesteps outlinedabove.
Finally, to expressa particular wordform’s morphological features like number,
gender, or person, ó is prefixedwith theappropriate ¡�� ’s.
Example. To illustratetheabove linking theory, let usconsidera few simpleex-
amplesfrom EnglishandCzech.First,wesetupthefollowing mappings. For each
mapping we indicatefor whatlanguage(s)it is applicable.
(38) a. B Ñ Cz,En×ÿ��� � if object
Í if eventuality
b. M Ñ Cz,En�ÿ���� �����}:�� if Actor����� if Patient �� if Addr essee
c.
S � En � ���� ����first pre-verbalposition ����� � if nom
direct-complementposition � ��� if acc
indirect-complementposition ��� ��� if dat
S � Cz � ���� ����first post-verbalposition ��� ��� if nom
Actor � Patient � ��� if acc
Actor � Addr essee� Addr essee� Patient � ��� if dat
d. morph.features� Cz,En� � �����! #"#$%���'&)(+*-, if 3rd person,singular
���/.0&)12���43!576!" if 1stperson,plural
Subsequently, considerthefoll owing predicate-valency structuresin (39).
(39) a. 8�9;:<�'=?>'>#@A:CB ACTOR D�8FEHGIGb. 8�9J:LKM> � :CB ACTOR D�8FEHGN:OB PATIENT D�8FPQGIG
(39(a)) translatesinto thecategoriesgiven in (40) for English andCzech.The
resulting lexical entries aregivenin (41).
(40) a. English: �-RTS &0UWVYXZU[1]\ "_^b. Czech: �/` &2UWabVYXZU[1]\ "c^
Formandfunction in Dependency Grammar Logic /29
(41) a. English: sleepd �e� .0&)1 ���f3!576!" 8[�/R�S &0U4VgXZU)1]\ "c^ Gihj8M9k:l�%=[>%>W@m:nB ACTOR D�8FEHGoGb. Czech: spıme d � �T.0&)1 � � 3!576!" 8[�/` &2UWabVYXZU[1]\ "_^ GNh�8T9p:l�%=[>%>W@m:nB ACTOR D�8FEHGoG
Similarly, (39(b)) translatesinto thecategoriesfor English andCzechgivenin
(42). Theresulting lexical entries aregivenin (43).
(42) a. English: 8[�/R/S &0UfVgXoU[1]\ "c^ GW` $ UWa VYqsrf1t(tu2*-1 ^b. Czech: 8[�/` U#avVgqsr41t(tu2*-1 ^ GW` &2U#avVgXoU[1]\ "4^
(43) a. English
readsd ���! #"#$%��� &2(+*', 8W8[�/R�S &0U4VgXZU)1]\ "_^ GW` $ U#a Vgqsr41t(tu2*-1 ^ Gwh8�9J:JKT> � :OB ACTOR D�8FEHGN:CB PATIENT D�8FPQGgGb. Czech
cte d � � #"W$'� �%&)(+*-, 8W8[�-` UWabVYqsrf1t(tu2*-1 ^ GW` &2UWabVYXZU[1]\ "c^ Gwh8�9J:JKT> � :OB ACTOR D�8FEHGN:CB PATIENT D�8FPQGgGWith therelation betweenpredicate-valency structuresandsyntacticcategories
thusestablished,how doweinterpretawordform asaparticularkind of dependent?
Theanswer to this is significantly lessinvolved thanthe previousdiscussion, and
is based on a discussion I providedin (1999a).
Essentially, what we do is introducestructural rules that enable us to rewrite
a modalindicatingthat a structure realizesa particular a morphological category,
to a modalindicatinga type of dependency relation which is theprimary (or pos-
sibly secondary) function of that morphological category. Additionally, function
wordscanbegivenfunction categoriesthathave,astheresult category, a category
indicating a typeof dependency relation.
For example,considerthestructural rules in (44(a)),andthefunction wordsin
(44(b)).
(44) a.
BWB?x�D */\0y D XoU[1]\ "{z BWB?xjD */\|y D|}BWB?x�D r4U2U D qsrf1t(tu0*'1 z BWB?xjD rcU)U D }BWB?x�D $ r41 D X $f$W" u|&|&0u0u z BWB?xjD $ rf1 D|}b.
German in d ����~ \#U)rf1t(+�4u ^ ` 3�" u 3 ����$ rf1 ^ h��j��B LOCATIVE D �German in d � �!� � u " u2�Z\ ^ ` 3!" u 3 � �!r4U)U ^ hm�j�QB WHERETO D �
Thefunction wordsin (44(b)) leadto complement-categories. Alternatively, if
wewantto create adjuncts,thenwe canusethecategoriesasin (45).
(45)German in d V ~ \#U)rf1t(+�4u 8[�-R r!a �-GW` 3!" u 3 � � $ rf1 ^ ho8)� U4� GGerman in d Vg� � u " u2�Z\ 8[�/R r�a �/GW` 3�" u 3 ���!rcU)U ^ h�8)� U � G
30� Formandfunction in Dependency GrammarLogic
Example1.3.3(page35ff.) illustratestheuseof theseGermanfunction words
in a derivation, after I have presentedtheentirebaselogic for DGL.
1.3.3 THE COMPOSITION OF LINGUISTIC MEANING
How do weconstruct linguistic meaningin DGL?Thekey ideafollowedhereis to
build “syntactic structureandsemantic structure in parallel”. In theprevioussec-
tion I already discussedhow thefunctionalstructureof aword’s lexical meaning is
closely reflectedin its syntactical category. Hence,thecomposition of asentence’s
linguistic meaning closely correspondsto how wordscanberelatedat thesurface.
More precisely, DGL adheresto a principle of compositionality, characterized
by Parteeet al. in (1990) asfoll ows:
”The meaning of acompoundexpressionis a function of themeaningsof its
partsandof thesyntacticruleby which they arecombined.” (p.318)
Similar principles of compositionality canbe found throughout formal gram-
marandformal semantics - seeJanssen’s (1997) for ageneraloverview of compo-
sitionality principles,Gamut’s (1991), VanBenthem’s(1991),andMorrill’ s (1994)
for compositionally relating categorial grammarandMontague’s intensional logi-
cal semantics.
A differencebetweenDGL andMontagueGrammaris though that we make
a differencebetween theabsenceof meaning, andtheabsurdity of meaning. This
point datesback to Sgall et al.’s discussionin (1986). In Montague’s approach,
only thosetreesareconsideredto be well-formedwhich canreceive an interpre-
tation from a model for that intensional logic being used. In otherwords,well-
formednessequatesto meaningfullness,with the latter meaning “interpretableon
a model”. Sgallet al., whendiscussingChomsky’s infamous example(46), point
out thatthesentencedoeshavea meaning,andthatthesentencedefinitely is well-
formed- despite thefactthatthemeaningis absurd.
(46) Colorlessgreen ideassleep furiously. (Chomsky, 1957)
Parasitic situationsaside, we would normally not consider (46) to make sense
- yet the very fact that we can make that consideration meansthat (46) at least
hasa linguistic meaning.18 On theotherhand,(47) doesnot evenhave anabsurd18We couldtry to phrase“absurdityof meaning”in transparentintensional logic (Materna,1998)
asthe impossibility to conceive of anobjectthat theconstruction madefor theexpressionthat (46)correspondsto. Absencesimplymeansthatthereis noconstructionfor theexpression. For arelation
Formandfunction in Dependency Grammar Logic /31
linguistic meaning, since it is not grammatically well-formedandhencedoes not
evenenable usto constructa representationof its linguistic meaning.
(47) English
Furiously sleep ideasgreencolorless.
In the remainder of this section, I discusshow DGL builds representationsof
linguistic meaningin acompositional way, andhow weexplain thementioneddif-
ference betweenabsenceandabsurdity of linguistic meaning.
Traditionally, categorial type logics usea (typed) � -calculus for specifying the
meaningof a sentence. A convenientmathematical fact thereby is that thereis
a close correspondencebetweennatural deduction andthe � -calculus- theCurry-
Howard correspondence. An important result establishedby the Curry-Howard
correspondenceis thataneliminationrule,eliminating animplication andthereby
combining two elements, correspondsto functional application in the � -calculus.
Conversely, an introduction rule corresponds to functional abstraction. Thus,for
example,whenwe apply an elimination rule to combinea function andan argu-
ment,we canin parallel apply themeaning of theargumentto themeaningof the
function (which is traditionally specified asa � -term).
Theissue now is, how canwe establish a correspondencebetween natural de-
duction andoperationsin a hybrid logic, so as to composea representation of a
sentence’s linguistic meaningin parallel to ananalysis of thesentence’s form? The
answeris relatively simple, in fact.19 Firstof all, weshould recall thatwhatweare
building arerelational structures.For a head � that meansthat it maybe looking
for an argument. That is, ��:��e�QB[�/D , we have a nominal � that refersto some
statewherethehead’s proposition holds,andfrom whereweshould beableto link
to someother (yet unspecified)state
along a � (dependency) relation. Similarly,
oncewe interpret a word group asa particular type of dependent, we specify that
assaying that it is a dependent that is looking for a head.We have something like�j�QB[�-D , but now � and
arefurther specified andit is the � thatwe needto estab-
lish. In other words,to combinea head � with a dependent, all we needto say
is that
is what � is looking for, andvice versa. For example, consideragainthe
lexical assignmentfor “sleeps”, repeatedbelow in (48).
betweencategorial grammar,dependency grammar,andtransparent intensionallogic seethe briefdiscussionin (Kruijf f, 2000).
19I am very much indebtedto CarlosArecesand to AlexanderKoller for the discussions thateventuallyled thecalculusI presenthere.
32� Formandfunction in Dependency GrammarLogic
(48) sleeps d �e�� #"W$'��� &2(+*', 8[�/R &0UWVYXZU[1]\ "_^ Gwh�8�9;:��'=[>%>#@�B ACTOR D�8FEHGgGHow does it getcombinedwith its Actor? Thestepsarein given(49).
(49) i. ���Q8F9<:J�%�F�����<:CB ACTOR D�8FEHGIGii. � �!� B ACTOR D�8o>�:<���?� �!����Giii. �j�Q8F9�:l�%�F�����i:nB ACTOR D�8FEsGHGo:i�m���2B ACTOR D�8�>�:����?� �!���kGH:i�m�T�Q�iv. Axiom: � ( B ACTOR D?��: � ( B ACTOR D#¡ z �£¢!¡v. �j�Q8�9J:J�'�¤�����¥B ACTOR D�8Z>�:<���?� �!����GWG
Because stating that �¦�T� � meansthat � and � � refer to thesamestate,(49v) is
model-theoretically equivalentto (49iii) togetherwith theaxiom in (49iv). Clearly,
this operationis similar to § -normalization.
Conversely, how do we modelthe analogonof functional abstraction? Func-
tional abstraction correspondsto the application of an introduction rule, which
dischargesan assumption. For that discharge to work, the assumption musthave
beenusedearlier in the derivation. Given the above discussion, this must have
leadto the introduction of a link (@) between theassumption’s ‘meaning’ andan
argument.Discharging theassumption thencanbeunderstood assimply severing
that link: Formally, we replace the link �Y�-E betweentheassumption’s nominal �andthe argument E by � . Becausex¨: � © x , we thus effectively drop the
assumption.20
Definition 5 (Baselogic for DGL). We define the baselogic for DGL in terms
of the proof calculusof Definition 2 (p.8) to which we add operationsacting on
representationsformulatedin a hybrid logic.
ª¬«®°¯-±�²�³ ´¬«Aµ¤¶ � � �·g¯-±�²4¸¤¹ º �7�µ¤ª¼» � ´Z·p«®¶½¯/±�²f¸¾¹À¿�±�²4¸t³´A«¬µ¤¶�Á � ¦·g¯-±�²4¸]¹ ªÂ«®°¯-±�²�³ º Á �µ]´Ã» � ªH·p«®¶½¯-± ² ¸ ¹�¿À± ² ¸ ³
20And, for thatreason,wealsodroptheconjunct.
Formandfunction in Dependency Grammar Logic /33
Ä ªL«Å°¯/±eÆ'Ç�È-É4ÉcÉÉÉÉ µ¾ªÀ» � ´Z·p«®¶½¯-±�²4¸]¹�¿�±�²4¸¤Ç k � �´¬«Aµ¤¶ �7� �·g¯-± ² ¸ ¹
µ]´Ã» � ªH·l«Å¶Ê¯-± ² ¸]¹�¿�± ² ¸¤ÇÉ4É4É Ä ªÂ«®°¯-±eÆ/Ç�ÈÉÉÉ k Á �´A«¬µ¾¶jÁ � �·g¯/± ²f¸ ¹Ä Ë «®¶½¯-± Æ Ç�È Ä Ì «®Íů-± Æ_¸ Ç�ÈÉÉɪ Ä µ Ë�Î � Ì ·[Èm«Å°¯T± ² ¸¾¹ÐÏ�¿�± Æ Çn¿¼± Æ ¸¤Ç ´A«Aµ¾¶ Î � ÍÑ·I¯/± ²WÒ ¹oÓ º Î �ª Ä ´QÈm«Å°¯T±�²f¸¾¹ Ï ¿�±�²f¸¾¹ Ó
ªÂ«®°¯-Ô Ï ´A«®¶½¯%¹ Ó k Î �µ¤ª¼» � ´Z·l«® Î � ¶½¯-±�²4¸¾¹ Ï ¿�±�²fÒc¹ Ó
Notethat wehavedefinedtheaboverules for Steedman-style notation of
categories. Therulesbelow definethebehavior of unary modalsÕ � and Ö that are
semantically relevant (Morrill , 1994). Unary modalsthat are semantically neutral
leavethesemantics untouched.
ª�«Ã°¯-±eÆ%³ k-×Ø ªHÙ)Ú[ÛÑ« × Ú °¯/± Æ Ø I Ù2Û!³ ªN« × Ú °¯-±¦Ü Ø I Ù)³Ä Ë «mݯ'Þ�ÈÉÉÉ´ ÄßØ]Ë Ù?à Û È�«Ã¶½¯%¹ ÄáØ
JÙ Û Þ�È º ×´ Ä ª�ÈQ«â¶½¯TÔ ÄßØ
JÙ2Û�³gÈ
Ø ªHÙ Ú ÛÑ«Ãã¯%ä ÄßÄ I È]Û�³gÈ kTå�æ Úªl« å æ Ú °¯'ä ÄáÄ I È Û ³gȪl« å æ Ú °¯-±¦Ü Ä I È Û ³ º å�æ ÚØ ªoÙ2Ú Û «Ã°¯/±¦Ü Ä
IÈ Û ³
Wekeeptherelationsbetween� � / V and çáè é / B|èßD strictly local. Weobtain this by
labelling a structural modalwith an index ê corresponding to the index given to
theunderlying modalrelation. Observe that weallow for a more specific modeEto replacea lessspecific modeP in therepresentationof linguistic meaning in ë V .
In line with this possibility wedrop theceteris paribuscondition usually assumed
34� Formandfunction in Dependency GrammarLogic
for structural rules: If a structural rule changesthe modeof a structural modal,
thenthemodeof theunderlying modalrelation changes accordingly. Finally, note
that we do not have term constructors or deconstructors in DGL. They can be
considered identity functions,by which we trivially obey the general residuation
lawsfor unary modaloperators (Moortgat,1997a). ìRemark 6 (Putting things together). Theelimination rules in Definition 5 bind
theheadandthedependentasfollows. Thedependentcomeswith a referenceto a
head@ , whereas theheadis statedat @ � . Theresulting conjunction statesthatboth
thedependent andthehead hold at @!� . This is anabbreviationthat is equivalentto
repeating theoriginal formulasandthenequating @ and @ � using � 3 @ � , asdonein
(49iii).
With that, the point could be raisedthat on the onehand,I arguedthat (in a� -calculus) delayed § -normalization would be favorable over including normal-
ization directly in thecalculus,whereas on theotherhandit seemsthatDefinition
5 does includea form of normalization. Is there a contradiction arising from this?
Theansweris, no. Theimportantpoint is not somuchwhether or not to delay§ -normalization - the point is whetherwe areable to explain the differencebe-
tweenthe absence of linguistic meaning andthe absurdity of linguistic meaning.
Unlike standard § -normalization, composition asdefinedin Definition 5 does not
fail whenincompatible meaningsarecombined.Thereasonfor this is thatweonly
state that something to hold. But, recall that as long as we have not bound the
nominalsto specificstatesin amodel,thatstatement is all wehave. Thedifference
between absurdity andabsenceof linguistic meaning (Sgall et al., 1986) is thus
maintained. ìExample. On the basis of Definition 5, how canwe built the linguistic meaning
givenin (50) in parallel to ananalysisof its surfaceform?
(50) a. German
ChristopherChristopher
gehtgoes
insinto-the
Kinocinema
inin
derthe
Stadt.city
“Christophergoesto thecinemain thecity.”
b. 8M9I:gíZîâ:ïB ACTOR D�8 � : Christopher GZ:nB DIR:WHERE TO D�8[¡�:pð��?ñv��òó��G:óB LOCATIVE D�8[�¥:<ð��¤ôcõYGvGTheproof in (51)showshow to employ thecalculusof Definition 5 andrelevant
Formandfunction in Dependency Grammar Logic /35
structural rules to analyze the sentence’s surfaceform, andbuild a representation
of its linguistic meaning.21
(51)
1. Lex ö?÷ « ×Ðø Æ�ù i ù?ú¾û ÷ Á/üM² i ùF² å�æ�ýfþ[þ#×Hÿ ÷ ¯-±¦Ü�µ?±¦Ü��¥¿ ±eÆ Ø WHERETOÙ ÷ ·
2. Lex ����� « åeæ ýfþ[þ × ÿ ÷ Á ü��Wù?ú å�æ þ[ý ù × ÿ ÷ ¯-±��/Ç3. Lex kö?÷�� « å æ þ[ý ù × ÿ ÷ ¯/±�� Ø�� Ùfµ���¿������������M·4. E/(2,3) µ ����� » ü��Wù?ú kö?÷�� ·s« å�æ ýfþ[þ × ÿ ÷ ¯-±�� Ø � Ù4µ���¿��!� �"�����M·5. E/(4,5) µ ö?÷ »ÑüM² i ù¤²Tµ �!�#� »�ü��Wù?ú pöF÷�� ·0·Ð«À±eÆ Ø WHERETO
Ùfµ��w¿��!� �"�����M·6. Lex $�% �'&H«lµ�( � ü )þ ×*) þ ú¾û i ÷ ·0Á þ,+ ×.- Æ�ù i ù0/�û ÷ ¯-±21-µ0µ43Å¿�576�·m¿ Ø
ACTORÙ � ¿ Ø
WHERETOÙ98�·
7. Lex Í:�<; ö9� & �>= � % ;Ñ« åeæ@? û,A ×Hÿ ÷ ¯T±eÆMµ�Bp¿C�ED�FG� H�·8. E
å¦æ(7)
ØChristopherÙ ? û,A « ×Hÿ ÷ ¯-±eÆ�µ Bk¿I�EDJFG� H�·
9. Hyp K «ML10. I
×(9)
ØX Ù ) þ ú¾û i�« ×*) þ ú¾û i L
11. E/(6,5) $�% �'&�» þ,+ µ öF÷ » üM² i ù¤² µ ���#� » ü��Wù[ú pö?÷�� ·2·Ð«lµ�( � ü )þ ×*) þ ú¾û i ÷ ·I¯±21-µ0µ43Å¿N576�·m¿ ØACTOR
Ù � ¿ ØWHERETO
Ùfµ��w¿������������M·0·12. E� (11,10) K »�ü )þ µ $#% �'&�» þ,+ µ ö?÷ »�üM² i ùF²Mµ �!�#� »�ü��Wù?ú kö?÷�� ·0·2·b«M(n¯± 1 µ0µ43Å¿N576�·m¿ Ø
ACTORÙ � ¿ Ø
WHERETOÙfµ��w¿������������M·0·
13. E×
(12,8)ØChristopherÙ ? û,A »�ü )þ µ $#% �'&�» þ,+ µ ö?÷ »�ü�² i ù¤²�µ �!�#� »�ü��Wù?ú kö?÷�� ·0·2·Ð«I(�¯±21-µ0µ43Å¿N576�·m¿ Ø
ACTORÙfµ Bk¿I�ED�FE��H�·m¿ Ø
WHERETOÙ4µ��w¿������������M·0·
14. Lex ö?÷ «lµ�( � ý@+ (з0Á üM² i ù¤² å�æ � ý ú × ÿ ÷ ¯±eÆ ¸ µF±eÆ ¸PO ¿Ê±¦Ü Ø LOCATIVEÙ ÷ ·
15. Lex ��% ;Ñ« åeæ � ý ú ×Hÿ ÷ Á-ü��Wù?ú å�æ�þ[ýQ ù ×Hÿ ÷ ¯-± ý Ç16. Lex (*& ��� &H« åeæ�þ[ý ù ×Hÿ ÷ ¯/±eÆ Ø�� Ùfµ � ¿��!�4R>So·17. E/(15,16) µ �!% ;�»�ü��Wù[ú0(T& ��� &2·Ð« å�æ � ý ú ×Hÿ ÷ ¯-±eÆ Ø � Ù4µ � ¿���� R>So·18. E/(14,17) µ ö?÷ » ü�² i ù¤² µ ��% ;�» ü��Wù?ú (*& ��� &2·2·Ð«lµ�( � ý@+ (зI¯/± Ü Ø LOCATIVE
Ùfµ � ¿N�!�4R>So·19. E� (18,13) UVUVW ChristopherX�Y[Z�\^]7_!`�a>U bdc0egfG]<aih�Ukjml"]7_Gn,o0pPnGUrq,s>t�] _!u9p�v�w jxlQy,zVzVzVzE]7_!{@Ukjml*]7_Gn,o0p4nGU qdci|<] _!u9p�v�} f~s>q,fPzVzVzG� }��± 1 µ0µ43Å¿N576�·m¿ Ø
ACTORÙfµ Bk¿I�ED�FE��H�·m¿ Ø
WHERETOÙ4µ��w¿������������M·¿ Ø
LOCATIVEÙ4µ � ¿���� RdSH·2·
1.4 TYPOLOGY, FORM , AND STRUCTURAL RULES
The ideaI want to explore here is how to build multilingual grammarfragments,
i.e. fragmentsdescribing phenomenaof morethanonelanguage,by distinguishing
in whatlanguage(s)a particular structural rule is applicable.
The approach that I take here with DGL is of course not entirely unique.
Therehavebeenpreviousattemptsatcombiningformalapproachesto grammar(or
rather, to syntax pure) anda typological perspective. Oneapproachwasinstigated
by Chomsky in (1965), andfocusedon constructing a “Universal Grammar”that21BecauseI havenotdealtwith informationstructureyet, I donotspecifyany semanticimport for
determinersin (51).
36� Formandfunction in Dependency GrammarLogic
would arguably underly every existing humanlanguage. A fundamentalproblem
with thatapproachwasthat theUniversal Grammarwasthought to beconstruable
by studying just a single language– English. This naturally led to the criticism
that typological universals cannot be defined on the basis of the results obtained
from a singlelanguage. Later, approachesshiftedtheir emphasisto variation. The
PrinciplesandParameters generativegrammarframework, proposedby Chomsky
in the early 1980’s, definesparameters on variation. The collective possibilities
of how to setthese parametersdefine“the” possible grammars- and,stronger, the
possible humanlanguages.Anotherapproachis Jackendoff’ sX-bar theory (1977).
What the above approaches thus have more or lessin commonis that they start
from a single language, anddescribe possible variation in the build-up of gram-
marsin termsof theobservations doneon that language. This is not theapproach
I take here.Rather, theattempts I make hereshould beplaced in a paradigm that
could (loosely) be called “typological universal grammar”,as advancedby peo-
ple like Greenberg, Keenan, Comrie,andHawkins, andwhich -at times-hashad
close tieswith categorial grammar.Thestarting point of this paradigm is that lan-
guagesdiffer, andthat the taskis to characterize the regularities in that variation.
As this type of variation canbemoreconveniently capturedby a head-dependent
asymmetry distinction thanthe strictly linear character of phrase-structuregram-
mars(cf. (Hawkins, 1983)), it should comeperhapsasno surprise that the above
perspective on cross-linguistic modeling hasbeen tried before in categorial for-
mal grammars. Vennemanproposedaround two decadesagoan approachbased
onacategorial grammarformalismthatincludedahead-dependentasymmetry(cf.
(Venneman,1977; Hawkins,1983)), andSteedman’s CCGprovidesanaccountof
cross-linguistic variation (in Germaniclanguages) in termsof availability of spe-
cific combinatory rules(1996; 2000c). It is this ‘tradition’ that I try to continue
with DGL, combining it with insights from Praguian views on typology and the
systemof languageassuch.
Technically, if we restrict eachrule to being applicablein onelanguageonly,
we obtain a “hybrid” fragment that is simply a setof structural rules,for several
languages, thatdonot interactatall. This is onesense in which wecanunderstand
multili nguality of resources: We have a collection of separate,language-specific
resources.However, if we allow a structural rule to be applicablein various lan-
guages,wegetamuchmoreinterestingperspective. If astructural rule, describing
form, is applicable in several languagesthen that rule canbe understood to indi-
Formandfunction in Dependency Grammar Logic /37
catewhatthese languageshave in common, whereas a structural rule applicablein
only onelanguageindicateshow that languagediffers from all therest.This is the
senseof multili nguality I am interestedin here:merged resources,wheremodels
for different languagesmay shareparticular fragments. (Seealso (Kruijf f et al.,
2000).)
The important point hereis that multilingual grammarfragments enable one
to construct a typological perspective of cross-linguistic comparison (cf. (Croft,
1990), Ch.1). Arguably, it is necessary for a framework to beableto provide this
perspective: Only throughapplicationof agivengrammarframework to modelling
a variety of languagesthe framework can be validated. Modelling phenomena
cross-linguistically elucidateswhethera framework’s mechanismsareindeedgen-
eralenough to beableto bespecific enough.
DGL, andcategorial type logic, provideasetting in whichwecanachievethis.
We assumethat the base logic, defined earlier, is languageuniversal. Thus,we
conjecture that we canbuild modelsfor all natural languagegrammarsthat start
from this commonbasis, defining the relation betweenform and meaning in a
compositional way, to which we canaddstructural rulesdefining moreelaborate
meansof structuralcontrol andstructural relaxation.22 Below, weproposeto model
of cross-linguistic variation andsimilarity asnetworks of structuralrules.
1.4.1 STRUCTURAL RULES AND GRADUAL REFINEMENT
According to Halliday, a grammar is “a theory aboutlanguageasa resourcefor
makingmeaning” (1985)(p.xxvi). Halliday (1985) proposesto build upagrammar
from systems, organizedin networks thatarestratifiedby ranks. A systemmodels
a particular choice, driven by the meaning we want to convey. Descending down
the ranks,the choices madeby the systemsinhabiting theselower ranks become
successively morespecific,dealing with increasingly finer detail. To relate this to
grammatical structure,systemsat a higher rank dealwith general organization of
a sentence,e.g. typeof speech act,clause-complexity, choice of moodandvoice.
Systemsat a lower rank decide aboutmorespecific detail. Next to the notion of
ranksHalliday considersthe notion of delicacy. By the delicacy of a system, or
groupof systems, we understandthe relative generality of the decision a system
makes.A lowerdelicacy meansthatit is moregeneral thanahigh delicacy choice,
22It shouldbenotedthattheclaim concernsherethepossibilityto modelgrammarsthatway. Weareby no meansclaimingthatnatural language grammars all work thatway from a cognitive pointof view. Sucha perspective is hardlywarrantedby, andgoesbeyond, mathematicalmodeling.
38� Formandfunction in Dependency GrammarLogic
andassuch there is somehopethat it holds for morelanguages- though this is,
of course,anempirical matter. Undera non-standard interpretation (Bateman,p.c.)
onecanalsoconsider the useof a syntagmaticnotion of delicacy: that is having
grammatical constituents described in more or lessdetail. With this, we get an
analogy betweenrankanddelicacy: Thehigher therank, thelower thedelicacy.
How canwetranslatethispictureto categorial typelogics? As farasI amaware
of, cross-linguistic modelling (in the above sense) hasnever beendiscussed in
categorial typelogics,only -to someextent- in CCG(cf. (Steedman, 2000c; Kruijf f
andBaldridge, 2000)). Here,I proposeto represent a setof structural rulesasa
network. Wecanannotateeachstructural rule (or packageof structural rules)with
the language(s) for which it holds. Following the ideasof ranking anddelicacy,
structural rules of higher ranking expresscommonalities amonglanguages(if the
setis multil ingual), andlower ranking expressesdifferentiation.
1.4.2 MULTIL INGUAL NETWORKS OF STRUCTURAL RULES
In thecurrent section I discusshow wecanbuild multilingualnetworksof structural
rules. (Part II containsnumerousexamples of suchnetworks.)
As is customaryin categorial type logic, eachstructural rule in a fragment is
given a name- be that something like the rather nondescriptive ç ��E�é or a more
elaboratenamelike ç � > � ���� K � @M@�� ^*� é . Givena name çk��é for a structural rule� z � � , we generally write çk��é � z � � . We extend that representation herewith
a set � indicating the languages to which the structural rule is applicable - i.e.
the structural rule is understood to modelpart of a phenomenon in a way that is
appropriate for the language(s) listed in � . For example,if çk��é is applicable to
languages��� ( ��%¢�� , thenwewrite this asin (52).
(52) çk��é �d� j0� �9�@��� z � �Model-theoretically, this change to the representation haslittle or no impact:
We aremerelyclaiming that çk��é is modelled by the appropriate frame-conditions
both in themodel � � j for language� ( , andin themodel � �9� for �!¢ .23
Givena setof structural rules,eachannotatedfor the language(s)they areap-
plicableto, how do we organize theminto a network? The organizingprinciples
arelaid down in Definition 6.23We do not explore herethe(purely technical)questionwhetherwe couldgenerate a combined
model for a setof languages � , ��� , in which rankingcould for examplerelateto a hierarchyoffiltersasdiscussedin (Kurtonina,1995).
Formandfunction in Dependency Grammar Logic /39
Definition 6 (Ar chitectures: Multilingual networks of structural rules). Given
a setof structural rules, ���C��� . �G�P�P�t�g� * � , wherebyeach structural rule is of the
form çk��é�� � z ��� . Wesaythat two rules � ( and � ¢ are connectedin thenetwork,¡, with � ( dominating �!¢ (written as � (£¢ �-¢ ), iff:
i. ¤r�¦¥ j ¤C§¨¤r©ª¥ � ¤ , i.e. � ( is applicableto at least asmanylanguagesas �[¢ .ii. Theoutput structureof � ( servesasaninput structureto � ¢ , i.e if � ( © � z ���
then �%¢ © ��� � z � .
A connected path « between a structural rule � ( and �'¬ is definedasthenon-
reflexivetransitive closure over ¢ . Theexistenceof a path « between � ( and �'¬ is
written as � ( ¢ �7¬ . Thesetof language(s) that a pathcovers is defined astheset
of languages of �!¬ .
Finally, to ensurethata networkis alwaysfully connected, weadda Startnode.
Thisnode doesnot correspond to a structural rule. It only indicatesthetop of the
network. By definition it dominates every node, and there alwaysexists a path
betweenStartandeveryrule in thegivensetof structural rules. ìTo illustrateDefinition 6, I endthis section with two (abstract) examples. In
thenext Part I discussamoreelaboratelinguistic examples, mostlyinvolving word
order.
Example. Consider thesetof structural rulesin (53),defined for languages � . ��<® .(53) � . �
¯ çk�±°_é �d�@² � �g³[� � � z �,çk�^´'é �d�@² � �g³[� � � � z � � ,çk�^µ'é �d� ² � ��¶ · z � � � ,çk�¦¸/é �d�g³E� � ¶~¹ ·�z ��� �
º
Clearly, wehave thefollowing relations: �±° ¢ �^´ , �^´ ¢ �^µ , �^´ ¢ �ª¸ . Thus,
wecandepict thenetwork asa tree, asin Figure1.4.2below.
[treefit=tight,levelsep=6ex] Startçk�±°_é �d�@² � �g³Q�çk�^´'é �d�@² � �g³Q�çk�^µ'é �d�@²� çk�¦¸/é �d�g³[�Figure1.3: A simplemultil ingual network of structural rules
Furthermore, thelanguage� . is coveredby thepath çk�±°_é ¢ çk�^µ'é , whereas � ®is covered by çk�±°_é ¢ çk�¦¸/é . ìExample. Let usconsiderwhathappenswhenwe extendtheset � . given in (53)
with thefollowing setof rules, givenin (54)
40� Formandfunction in Dependency GrammarLogic
(54) � . � ¯ çk�^»'é �d�@²� � ¶~¼ ·�z � ¶~½ · ,çk�^¾'é��d� ² � � ¶V¿ ·�z � ¶~¼ ·º
We have that �^» ¢ �^¾ but thereis no structural rule thatdominates �^» . This
situationmayfor examplehappenwhentheentry-condition for �^» is arisesfrom a
lexical assignment, rather thanfrom structural reasoning. Thesameis actually the
casewith �±° in theexampleabove!
Thus, the network we now obtain would not be fully connected, without the
Start node. This is thepoint why therealwayshasto bea Startnode. With that in
mind, thenetwork we obtain is givenin Figure1.4.2.
Start�d�@² � �g³[�çk�^»'é �d�@²�çk�^¾'é �d�@²�çk�±°_é �d�@² � �g³@�çk�^´'é �d�@² � �g³@�çk�^µ'é �d�@²� çk�¦¸/é �d�g³@�Figure1.4: A simplemultili ngualnetwork of structural rulesincluding Start
SUMMARY
In this chapter I focusedon the relationbetweenlinguistic meaning andsurfaceform, in
particularthe realization of a linguistic meaning’s dependency structure. To that end, I
discussedhow dependency relationscanbe relatedto morphological strategiesthat real-
ize them. Becausethe relationis mediatedby abstractmorphological categories (like in
Government& Binding’s theory of case),the relationis not language-specificbut cross-
linguistic. In this way, DGL canprovide a linking theory that overcomesthe criticism
thathasit thattheinterpretationof a wordform asa particular “role” is stipulated. Subse-
quently, I focusedonhow wecanprovidealogicalcalculus in whichasentence’slinguistic
meaningis built in a compositional, monotonic way asa reflectionof the analysisof its
surfaceform. I useda resource-sensitive categorial proof theoryfor theanalysisof form,
alike theLambek-stylecalculi usedin categorial typelogic. However, ratherthanoperat-
ing ontype-logical termsto reflectsemanticsusingaCurry-Howardstylecorrespondence,
theproof theory in DGL operateson hybrid logical terms.Usingcategories that indicate
head/dependent asymmetries,anda formalizationof morphological strategies, I showed
how wecanobtainthekindof linguisticmeaningrepresentationsdiscussedin earlierchap-
tersthroughacompositionalanalysisof sententialform.
CHAPTER 2
THEORIES OF INFORMATION STRUCTURE
In thischapterI discussvarioustheoriesof informationstructurethatstresstheimportanceof ex-
plaining boththeexpressionof information structureandhow information structurebearsupon
linguistic meaning. Based onreflectionson thesetheories,I motivatewhy I opt for thePraguian
approach,andI discusscoreconceptslike contextualboundnessandtopic-focusarticulation. At
theendof thechapter, I explainhow contextualboundnesscanbeindicatedin thehybrid logical
formulation of asentence’s linguistic meaning,andhow wecanderivea topic-focusarticulation
from theindividual nodes’indicationsof contextual boundness.I alsopoint out how we arego-
ing to interpret a sentence’s topic-focusarticulation model-theoretically, andwhy dependency
relationsarenecessaryfor explaining (therealization of) information structure.
[the phenomenaat issuehere]have to do primarily
with how themessageis sentandonly secondarily
with themessageitself, just asthepackaging of
toothpastecanaffect salesin partial independence
of thequality of thetoothpasteinside.
– WallaceL. Chafe
2.1 INFORMATION STRUCTURE IN LINGUISTIC MEANING
In general, the purpose of a (declarative) sentence is to communicate meaning.
As mostsentencesareuttered in the context of a larger discourse,there is a side-
condition on this communication: the sentence’s meaning needs to be coherent
with the precedingcontext. Arguably, the claim behind information structureas
a theoretical construct is that it helps us to explain how the meaning a sentence
conveys canbecoherentwith respectto a larger discourse.
Froman abstract viewpoint, informationstructuretries to divide the meaning
of a sentence into several parts. Onesuch part, which I call for the momentthe
Relatum, states how themeaning of thesentencepurports to relate to the already
establisheddiscourse. It helpsto set,as it were,the conditions under which the
meaningof thesentencecanbetrue,providedthese conditions aremet.1
1An importantpointhereis thattheRelatumconditions themeaning- for informationstructuretomakeany explanatoryimpact,wemustdistinguishinformationstructureandthelinguisticmeaningitis partof from thesubsequentinterpretationof thesentencein thesettingof theestablisheddiscourse
41
42À Theoriesof InformationStructure
Next to the Relatumwe candistinguish a part that I call herethe Attributum.
TheAttributumsayssomethingabout theRelatum,by qualifyin g or modifying the
meaning it is relatedto in thecontext. Thus,whereas theRelatum of a sentence’s
linguistic meaning could beunderstoodasspecifying certain ‘given’ information,
it would be only partially correct to perceive of the Attributum as the ‘new’ in-
formation. The Attributum neednot provide information that is entirely new, in
an additive fashion. It may well indicate the needto change, modify, a piece of
informationthathadpreviously beenestablishedin thediscourse.
An importantissue now is how a sentence’s surfaceform realizestheinforma-
tion structureof theunderlying linguistic meaning. After all, whereas themeaning
that is beingcommunicatedis by nature multi-dimensional 2, wordformscanonly
beuttered in a linear order. Thus,weneedto project thecomplex underlying struc-
turesontoasingledimension,andtherebyweareconstrained by language-specific
rules defininggrammaticality.
Thebasic ideais thatformsareiconic of their informativity - they carrywhatI
call herestructural indicationsof informativity. It naturally dependson thetypeof
languagewhatmeansareavailableto indicateinformativity. For example,Slavonic
languageslike Czechor Russianpredominantlyuseword order, structuring a sen-
tence suchthat thewordsrealizing theRelatumappear at thebeginning, followed
by the Attributum - see(55) for somepossibilit ies in Czech,and their English
counterparts in (56).
(55) Czech
a. VceraElijah cetlÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë KatceknihuÁ ÂEà ÄÌ ÉPÉPÍ>ÎPÏiÊ�ÉPÊ�Ë .
b. KatceElijah vcera cetlÁ ÂEà ÄÅ�Æ,ÇxÈÉPÊ�Ë knihuÁ ÂEà ÄÌ ÉPÉPÍ>ÎPÏiÊ�ÉPÊ�Ë .
c. Knihu Elijah vceracetlÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë KatceÁ ÂEà ÄÌ ÉPÉPÍ>ÎPÏiÊ�ÉPÊ�Ë .
“Elijah reada bookto Kathy yesterday.”
Thus,even though Slavonic languageshave a relatively free word order, that
word order is by no meansarbitrary: It indicatesinformativity, andthereforethe
sentence’s felicity mayvary dependingon thecontext.
context. Thepurportedrelation,or contextual referencein abroadsense,is not yetresolved. It is the(im)possibility of resolvingthe referencethat rendersa sentence’s linguistic meaning(in)coherent.Also, notethatinformationstructureis notequalto truth-conditions - (Sgalletal., 1986;Hajicova etal., 1998; Peregrin, 1995).
2In thesensethatconceptual structuresarenot linear.
Theoriesof InformationStructure /43
On theotherhand,a languagelike Englishalready usesword orderasa mor-
phological strategy to realize Case,a rich inflectional system being absent. To
realize informativity, Englishpredominantly uses other means, in particular tune.
Theexamples below (56) illustratetheuseof tuneto realizethesameinformation
structuresasin (55). Pitchaccent is indicatedby SMALL CAPS.
(56) English
a. Yesterday Elijah readÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë A BOOK TO KATHYÁ ÂEà ÄÌ ÉPÉPÍ>ÎVÏiÊ�ÉPÊ�Ë .
b. Elijah readÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë A BOOKÁ ÂGà ÄÌ ÉPÉPÍ>ÎVÏiÊ�ÉPÊ�Ë to Kathy yesterdayÁ ÂEà ÄÅ�Æ,ÇxÈÉPÊ�Ë .
c. Yesterday Elijah readthebookÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë TO KATHYÁ ÂEà ÄÌ ÉPÉPÍ>ÎVÏiÊ�ÉPÊ�Ë .
Besidestune,Englishcanalsousefunction wordsto realize informativity. For
example, a definite determiner prototypically indicatesthat the meaningof the
modifiednoun is contextually given (56c), whereasverbal auxiliariescanbeused
to make themainverbmoremarked(57).
(57) English
a. Yesterday Elijah readÁ ÂEà ÄÅJÆdÇmÈQÉPÊ�Ë A BOOK TO KATHYÁ ÂEà ÄÌ ÉPÉPÍ>ÎVÏiÊ�ÉPÊ�Ë .
b. Yesterday ElijahÁ ÂEà ÄÅ�Æ,ÇxÈÉPÊ�Ë DID READ a bookÁ ÂEà ÄÌ ÉPÉPÍ>ÎPÏiÊ�ÉPÊ�Ë to KathyÁ ÂEà ÄÅ�Æ,ÇxÈÉPÊ�Ë .
Finally, wealsoencounterlanguagesthathavearich nominal morphology and
-hence-arelatively freer word order,which realize informationstructureprimarily
throughaffixation. An often-citedexampleis Japanese,wherethe-wasuffix marks
a contextually given item and -ga is often associatedwith newness(though see
(Heycock, 1993)). Haimanmentionsother languagesthat have similar construc-
tions(cf. (Croft, 1990),p.10).For example, thePapuanlanguageHuausesa suffix
-mo to indicate a sentence’s Relatum. Furthermore, although Turkish normally
usesword order to indicate information structure (Hoffman, 1995b; Hoffman,
1995a) Haimannotes that the -sA suffix canmark contrast (“contrastive topic”).
Interestingly enough, Tagalogalso usesmorphological meansto indicate infor-
mativity but, asKroeger observes, it depends on the dependency relation that is
involvedwhetherthesuffix indicatesthattheitem belongsto theRelatum or to the
Attributum (1993)(pp.64-69,pp.130-131). Applying what Kroeger calls the -ay-
Inversion construction to anActor makestheActor part of theRelatum,but using
44À Theoriesof InformationStructure
-ay-Inversion with any other dependency relation indicatesthedependentspecifies
something new. Finally, Engdahl & Vallduvı mention in (1994) Navajo andVute,
languagesin which Attributa areassociatedwith a particular suffix.
To recapitulate,weseethatthereis aninterestingvarietyin how languagescanre-
alize informationstructure,andthat it is necessaryto distinguishdifferenttypesof
dependency relations to give anadequateaccount. Depending on the type of lan-
guage we aredealing with different typesof structural indications of informativity
arepredominantly used,like word orderin Czech,tunein Englishor a dedicated
morphological suffix in Japanese. ‘Predominantly’ should be stressedhere,be-
cause no languageappearsto bemakingabsolute useof oneandonly onemeans.
For example, if we take the two typologically rather different languagesthat
Sgallet al contrast in (1986), EnglishandCzech,thenwecanseethatEnglish can
useword order-relatedconstructions like topicalization or focal fronting, andthat
Czechcanusetuneto mark contrast. If we take languagesthat we canconceive
of asbeing ‘somewhereinbetween’like Dutch or German,thenwe canobserve
anevenmoreobviouscontinuumbetween theuseof word order andtuneasstruc-
tural indications of informativity. Sgall et al. often presentexamples like (58),
illustrating theuseof word order in English.3 Naturally, any theory of information
structureshould beableto handle these.
(58) a. Christopherwaswriting his dissertation on theweekends.
b. On theweekends,Christopherwaswriting his dissertation.
It is thenthis relativelypredominantuserather thananabsoluteuseof different
structural indications of informativity thathasimportant severalconsequencesfor
a theory of information structure,modeledin a particular grammarframework.4
Firstof all, becausegrammardescribestherelation betweenfunction(linguistic
meaning) andform, the framework underlying thegrammarneedsto bepowerful
enough to modelthevarious strategiesa languagemayadopt asstructural indica-
tionsof informativity, andthepotential interactionbetween thesestrategieswithin
a single language.
Secondly, a theory of informationstructure-asaninherentcomponentof a the-
ory of language- mustbeableto makepredictionsabout how informationstructure
canberealizedcross-linguistically. Naturally, a language’s inventoryof strategies3Observe alsotheword ordervariationin (56)4Certainlyfrom the Praguianpoint of view, aswell asthe othertheoriesdiscussedhere,where
themodelingof informationstructureis a matterof grammar.
Theoriesof InformationStructure /45
to structurally realize informativity depends on its typological characterization.
But, becausestrategiesare relative ratherthan absolute, each languageshares at
leastpartof its inventory with other languages.Consequently, lestthetheory gives
rise to a ratherad hocexplaination of information structure, it should bepossible
to lift the modelof how a particular strategy contributesto realizing information
structure in one languageto a different languageif the latter employs that same
strategy. Thus,onewould for exampleexpect that a modelof informationstruc-
ture for Germanwould show significant overlap with similar modelsfor Dutch
andEnglish. And that,wheredifferences do arise, they would be explainableby
language-specific constraints on grammaticality (or “prosodic well-formedness”,
(Morrill , 1994)).
Below I describevarioustheoriesof informationstructure,andreflectonthemfrom
thesetwo perspectivesof cross-linguistic explanationand(formal) coverage. The
theoriesI describearecontemporaryframeworks that oneoftenencountersin for-
mal grammaror formal semantics: thePraguian theoryof topic-focusarticulation
( Ð 2.2), Steedman’s Theme/Rheme( Ð 2.3), and Vallduvı’s information packaging
( Ð 2.4). Thus,my coverage is by nomeans‘total’ but there is agoodreason for dis-
cussing just thesetheories. Namely, these aretheonly theories thatconsiderboth
the “semantics” of information structure and its modeling in a grammarframe-
work - unlike the ‘degrammatized’theoriesof Karttunenor Rooth,or mostof the
Government& Binding tradition which considersonly the syntax of information
structure and not its reflection in linguistic meaning.5 For other overviews, see
for exampleKruijf f-Korbayova (1998), Hajicova andKruijf f-Korbayova (1999),
or Vallduvı (1990).6
After these discussions, I provide in Ð 2.5 a brief reflection from theviewpoint
of theabove remarks about theoriesof informationstructure,andI presentin Ð 2.6
anoverview of how information structureis modeledin DGL.
5TheorieslikeGrosz& Sidner’s(1986),Groszetal’s(1995)andHahn& Strube’sextensionof thelatter(usingDanes’s theoryof thematicstructures)areall concerned with discoursestructureratherthangrammar, andthereforefall outsidethescopeof thisdissertation.SeeKruijf f-Korbayova (1998)for adiscussionof how thesetheoriesrelateto thePraguiantheorythatI dodiscusshere.For reasonsof time I am not ableto discussZubizaretta(1998)or Lambrecht(1996). Both accountsappear todeserve interest,particularlyZubizaretta’s asshetakesa perspective on informationstructurethat ismodeledon thebasisof word orderphenomenaandtunein GermanicandRomancelanguages.
6Vallduvı’s descriptionof Sgallet al’s theoryof topic-focusarticulation(1986) is, however, de-batable.
46À Theoriesof InformationStructure
2.2 INFORMATION STRUCTURE IN THE PRAGUE SCHOOL
Information structure hassincelong beenan essential ingredient of the view on
languagedevelopedin thePrague Schoolof Linguistics. Nowadays, a distinction
is madebetween the topic of a sentence,andthe focus. Thesetwo termscanbe
traced back to Weil’s work in the nineteenth century (1844). Weil’s work was
resumed by several Germanlinguists in the decades around the turn of the last
century. Subsequently, thePragueSchoolof Linguistics startedpaying systematic
attention to issues of information structure, starting with Mathesius’s work (1936;
1975). Mathesius recognized that the distinction betweentopic and focus was
importantto problemsranging from tuneto wordorder,andformulatedanaccount
on thebasisof a structural comparison of CzechandEnglish(cf. also(Sgallet al.,
1986),p.175)).
Within the Functional Generative Description, the theoryof topic-focus artic-
ulation (or TFA for short)hasbeenelaboratedby Sgall, Hajicova, andtheir col-
laboratorsfor more than four decadesnow. Hajicova presents in (1993) a brief
overview of thedevelopmentsthat includeSgallet al (1973; 1980; 1986) andvar-
ious articles primarily by Sgall andHajicova. A recent dialogueexamining TFA
and its relation to formal semantics canbe found in Hajicova, Partee,andSgall
(1998).
Therearethreeprincipal ingredients to thePraguian theory of TFA:
i. thetopic andfocusdichotomy thatdividesasentence’s linguisticmeaning into
acontextually giventopic (theRelatum)andafocusthatis about thetopic (the
Attributum);
ii. contextualboundness, acharacterizationof anindividualhead’sor dependent’s
informativity, beingeithercontextually boundor contextually nonbound;and,
iii. communicativedynamism, which is a relative orderingover theheads andde-
pendentsmakingup a sentence’s linguistic meaning indicating how informa-
tive they arerelative to oneanother.
Furthermore,we have the closely related conceptsof salience (discourseac-
tivation) andthe Stockof Shared Knowledge. Both play an important role in the
discourseinterpretation of TFA (Sgallet al., 1986;Hajicova,1993).
An importantcharacteristic of FGD’sTFA is thatthetermstopic andfocusare
not primarynotions,like their counterparts in othertheories. Rather, topic andfo-
Theoriesof InformationStructure /47
cusarebased on the structural notion of contextual boundness.7 Eachdependent
andeachhead in a sentence’s linguistic meaning is characterized asbeing either
contextually bound or contextually nonbound. Intuitively, items that have been
activated in the preceding discoursemay function as contextually bound (CB),
whereasnon-activateditemsarealwayscontextually nonbound(NB) (Sgallet al.,
1986)(p.54ff,p187ff). Mostly, an item is activated by introducing it explicitly into
thediscourse. Importantabout contextual boundnessis, though,that it is a linguis-
tic opposition, reflectedin thestructuring of linguistic meaningandits realization
- it is not precise to equate contextual boundness to the discourse(or cognitive)
opposition of given/new. For example,apreviouslyitem(CB) mayoccur in acon-
trastive focus, andwecanpresentitemsasCB if they areactivatedby thesituation
of thediscourseor canbeactivated indirectlyby for exampleassociation(Hajicova
et al., 1998)(p.59). In otherwords,contextual boundnessis an issue of linguistic
presentation.
Giventhis characterization, andtheinternal structureof thesentence’s linguis-
tic meaning, wecanderivetheactual topic andfocus. To thatend,Sgalletal define
in (1986) thefollowing procedure(p.216).
Definition 7 (FGD’s Topic-FocusArticulat ion). Givena tectogrammatical rep-
resentation of a sentence’s linguistic meaning,
Ñ themainverbbelongsto thefocus if it is contextually nonbound, andto the
topic if it is contextually bound;
Ñ thecontextually nonboundnodesdepending on themainverbbelongto the
focus, andsodo all nodes(transitively) subordinatedto them;
Ñ if someof theelements of the tectogrammatical representation belong to its
focusaccording to either of theabovepoints, theneverycontextually bound
daughterof themainverbtogetherwith all nodes(transitively) subordinated
to it belong to thetopic;
Ñ if no nodeof thetectogrammatical representation fulfills thefirst two points
above, thenthe focusmaybe more deeplyembedded; special rules for the
determination of focus are applied in these cases.
Ò7To quoteSgallet al.: “If the notionsof topic andfocus(aspartsof a tectogrammaticalrepre-
sentation)arecharacterizedon thebasisof contextual boundness,thenwedon’t have to worry aboutquestionswhethertopic andfocusarea single(deepor surface)constituent[...].” (1986)(p.188).
48À Theoriesof InformationStructure
In FGD, the scaleof communicative dynamismdefinesa (partial) orderover
thenodes in a sentence’s linguistic meaning, after Firbas’s original notion of com-
municative dynamism (seeFirbas(1992) for a recentformulation). If we project
thelinguistic meaning’s treeto a line, then weobtain a reflection of thatorder. The
topic proper andthe focus proper aretheleastrespectively mostcommunicatively
dynamic elements in a sentence’s linguistic meaning. In the projected(deep) or-
der, thetopic propercorresponds to theleftmost item,whereas thefocusproper is
identified by thetherightmost element.
Hajicova andSgallnote in (Hajicova et al., 1998)(p.56ff) thatthere is a strong
correspondencebetween communicative dynamism and word order (and, indi-
rectly, tune). This certainly holds for languageslike Czech. Dependents that are
contextually nonboundareconsideredto becommunicatively moredynamic, and
occur prototypically after the head, whereascontextually bounddependents are
lessdynamic and should occur before the modified head. The mutual ordering
of contextually nonbounddependents thereby follows what is calledthe systemic
ordering, thecanonical ordering in which complement typesoccur in a given lan-
guage(Sgalletal., 1986;Sgalletal., 1995). Ontheotherhand, FGDconsidersthe
order of contextually bound complements to be only determined by their mutual
communicative dynamism. The examples in (59) give a brief illustration of the
above ideas. Also, recall theearlier examples(58) and(56).
(59) a. CzechCoElijah udelal? (English Whatdid Elijah do?)
ElijahElijah-CB
koupilbought-NB
knihu.book-NB.
“Elijah bought a BOOK.”
Topic=Ó Actor :Elijah Ô , Focus=Ó buy, Patient:bookÔb. CzechCoElijah koupil? (English Whatdid Elijah buy?)Õ
he-CBkoupilbought-CB
knihu.book-NB
“He bought a BOOK.”
Topic=Ó Actor :he,buy Ô , Focus=Ó Patient:book ÔDefinition 7 also covers cases where the focus is deeper embedded. Thus,
the dependent(s) constituting the focus do not modify the main verbal headbut
(transitively) one of its dependents. In the examplein (60), only the dependent
realizedasskapsami(English “with pockets”) belongsto thefocus,therestof the
sentence’s linguistic meaningconstitutesthetopic. Consider also(61).
Theoriesof InformationStructure /49
(60) CzechJake nosı krtek kalhotky?
(English “What trousers doesthemolewear?”)
Krtekmole-CB
nosıwears-CB
kalhotkytrousers-CB
swith
kapsami.pockets-NB
“The molewearstrousersWITH POCKETS.”
Topic=Ó Actor : mole,wear,Patient: trousersÔ , Focus=Ó GenRel:pocketsÔ(Kruij ff-Korbayova,1998)(p.27)
(61) English
(Whatteacherdid you meetyesterday?)Ö(Yesterday×9Ø Ï (I ×dØ Ï (met×dØ Ï (theteacher ×0Ø Ï,ÙPÚ Ö
(of CHEMISTRY ×,Û ÏEÜ Ù Ý– cf. (Sgallet al., 1986), (Hajicova et al., 1998)(p.135)
Thus,the primary notions contextually bound andcontextually nonbound are
recursive in the sensethat contextually nonbound itemscanbe embedded under
contextually bound itemsandviceversa.
In thegeneral case,neithertopic norfocusis asingle item,as(60)or (62)show.
(62) English
(Whathappenedto Jim?)
A burgler INJURED him.
Topic=Ó Patient:heÔ , Focus=Ó Actor: burglar, injureÔ (Hajicova,1993)
Petkevic notesin (1987; in prep) thatDefinition 7 does not cover somespecial
casesof topic-focus articulation that hecalls “split semantemes”.Thetopic-focus
articulation of a sentenceis representedat the level of linguistic meaning, andat
that level we do not have separatenodesfor function wordsor evenmorelocal as-
pectsof form. A sentence’s linguistic meaningonly hasnodesthat representwhat
in FGD arecalled auto-semanticunits or semantemes.However, from the view-
point of a sentence’s topic-focusarticulation it is not only the whole semanteme
assuchthat canbe determinedaseithercontextually boundor contextually non-
bound. Petkevic illustratestheneedfor a morerefinedassignedby examples like
(63).
(63) English
a. I SHALL do it, not thatI HAVE already done it.
b. I saw not only a single mousethere but severalMICE.
50À Theoriesof InformationStructure
For example,Petkevic arguesthatin (63a)thespecificationsof theverbal Tense
of both occurrences of do belong to the focus, whereasboth occurrencesof the
headdo belongto thetopic. A similar picturearisesfrom (63b), only thenfor the
specification of number.
Over time, several proposalshave beenmadehow to formalize FGD’s theory
of TFA. In general, theseproposalseither focus on the (truth-theoretic) interpre-
tation of a sentence’s topic-focusarticulation, or have as their main concern the
grammar’s representation of a topic-focus articulation.
Both representrather long traditions. FGD received its first formalization in
Sgall et al’s (1969), wherethe authors wereconcernedwith providing a “math-
ematically -thus linguistically- interesting description of (linguistic) meaning.”8
Sgall(1980) presentsthefirst formalizationof FGD’sTFA. Sgallfirst constructsan
automaton(roughly a complex pushdown store automaton) that is ableto generate
representationsof asentence’s linguistic meaning,includingmarking of contextual
boundness. Subsequently, a transduceris given that completesthe representation
-asit were-by deriving thesentence’stopic-focusarticulation,based onthecontex-
tual boundnessmarking. Petkevic extendsthis type of description in (1987; 1995;
in prep). Petkevic’s formalization is couchedin a larger reformulation of FGD’s
generative description of linguistic meaning, and includessolutions to several of
theproblemsnoted on thepreviouspage.
After Sgall et al argued the importanceof distinguishing a sentence’s topic-
focus articulation for the felicity of its linguistic meaning in a givencontext, vari-
ousattemptshavebeenmadetowardstheclarification of thisview in logical terms.
Onegroup of such contributions wascarried out within the framework of an in-
tensional logic, namelyTichy’s transparent intensional logic. The basic issues
involved in formulating TFA in transparent intensional logic were discussedby
Materna andSgall (1980) andby Materna,SgallandHajicova (1987). Vlk (1988)
provideda procedure for translating the tectogrammatical representations gener-
atedby FGDinto Maternaet al’s logical representationsof transparentintensional
logic.
Other, more recentdevelopmentsare based on Partee’s tripartite structures
or on a logical dynamic perspective as arising from dynamic semantics. (See
(Muskenset al., 1996) for a general description of logical dynamics and its use
in describing natural languageinterpretation.)
8The kind of grammarthat Sgall et al presentin (1969) still employs phrasestructure-basednotions,contraryto thelaterwork (Sgalletal., 1986).
Theoriesof InformationStructure /51
Peregrin (1995) is the first attemptto construct a more dynamicaccount of
TFA. Following anapproachthatessentially goesbackto Jackendoff (1972), Pere-
grin formalizesthe intuition that the focus says something about the topic as aÞ-term. The topic is modeled asan abstraction, to which the focus-termthencan
beapplied.
To provideanaccount of thesemantic effectsof information structure, Peregrin
providesanextensionaltheory of thetruth of a sentence’s topic-focusarticulation.
In this theory, ßPßáà ßPß stands for the extension of an expression ‘ à ’, wherebyßPß"à ßPß is a truth value if ‘ à ’ is a sentence,an individual if ‘ à ’ is a term,anda
classof individualsif ‘ à ’ is a unary predicate. Then,a proposition whoseexten-
sion is denoted by ßTà ß is associatedwith every expression à (understood asa
presupposition associatedwith à ) asgivenin (64).
(64)
ß�àâß = ßPß�àãßPß if X is a sentence
= ßPß<ä�å Ü åçæNàèßPß if X is a term
= ßPß<ä�å Ü à�é�å�×êßPß if X is a unary predicate
The semantics of a formula ë�ÓGìíÔ , asthe predication of ë corresponding to
the focus-part over a sentence’s topic-part ì , is definedin (65), cf. (Peregrin,
1995)(p.240). Notethat ë£é�ì^× is (the î -normalization of) thestandardapplication
of ë to ì .
(65)
ßPß�ë±ÓGìCÔ±ßPß = true iff ß�ìïß = true & ßPß�ëðé�ìª×êßPßræ true
= false iff ß�ìñß = true & ßPß�ë£é�ìª×êßPß = false
= false iff ß�ìñß = false
Therather simpleexamplesin (66) illustratethebasicidea.
(66) a. JohnWALKS: Walk Ó John Ôb. JOHN walks:
ÞJò Ü ò (John) Ó Walk ÔPeregrin works out an extensional account of negation, basicquantification,
and focus asexhaustive listing. On the basis of the definitions in (64) and (65)
Peregrin definesamoredynamicaccount of Ó<óVÔ . Dynamically, a predication ô£é9õö×is true canbe modeled asa statementsaying that there exists an assignmentof a
valueto a variable ÷ such that ô£é�÷T×öøù÷úæûõ is true. A similar construction can
bedefinedfor Peregrin’s new modeof predication, Ó<óVÔ . Givena concatenator Ô!ø ,ìüÔ!øüë hasatruthvalueif andonly if ì hasatruth value,andit is trueif andonlyìýøNë is true(in thesenseof ôCøMõ asabove).
52À Theoriesof InformationStructure
As Peregrin observeshimself, the definition he gives for the truth of a state-
ment ìþÔ!øþë cannotbeappliedrecursively. Kruijf f-KorbayovaextendsPeregrin’s
proposal to an intensional approach in (1998), andprovidesdefinitions that can
be applied recursively (seeher p.78ff). Kruijf f-Korbayova weavesan intensional
(typed)theory of TFA into a discourserepresentation theory to createTF-DRT. In
Chapter 6 I discussKruijf f-Korbayova’s TF-DRT in moredetail, andI showhow
particular technical problemswith TF-DRT canbesolvedby usinghybrid logic.
Besides a dynamic account, Peregrin also briefly discussesthe possibility to
modelTFA in termsof tripartite structures. The idea of using tripartite structures
wasfirst put forwardby Parteein (1991), andis substantiatedto a larger degreein
Hajicova et al’s (1998). In the latter work, the authors discussin Chapter 2 how
a tripartite structureconstisting of anOperator, a Restrictor, anda NuclearScope
could modela sentence’s informationstructurewhenit involvesa focus-sensitive
operator.9
Many subtletieshave beenglossedover in the above discussion of TFA. For
more thorough exposesseeSgall et al’s discussionin Chapter3 of (1986), and
Hajicova et al’s discussion in (1998). Throughout thenext chapters I devote more
attention to therelation betweeninformation structurein theabovePraguian sense,
in particular to contextualboundnessandits realizationusingwordorderandtune.
Whereappropriate the relevant Praguian referencesare given there. Finally, in
Chapter 6 I discussa model-theoretic account of the discourseinterpretation of
informationstructure,basedon a reworked version of TF-DRT.
2.3 STEEDMAN’ S THEME/RHEME
Steedman(1996; 2000c; 2000a) develops a theory of grammarin which syntax,
information structure, and intonational prosody are integrated into one system.
Steedman’s main aim is to provide an information structure-sensitive composi-
tional analysisof English phrasedasa Combinatory Categorial Grammar.There-
fore, this system is monostratal: the only proper representation of a sentence is
the representation of its linguistic meaning: “... a theory of grammarin which
phrasal intonation and informationstructure arereunited with formal syntaxand
semantics is not only possible, but muchsimpler thanonein which they aresepa-
rated.”(Steedman,2000a)(p.653)
9BecauseI donotdiscussfocalizersin thisdissertation,I omit furtherdiscussion.Seefor example(Hajicova et al., 1998)(p.39)for a fully workedout exampleof this approach.
Theoriesof InformationStructure /53
Steedmanrecognizes two independent dimensions of information structure,
both of which arerelevant to its realization (Steedman, 2000a)(p.655). The first
dimensiondefinesapartitioning into aThemeandRheme. Thisdistinction is simi-
lar to theoneproposedby Mathesius,thePraguian topic-focusarticulation,andthe
Relatum/Attributumcharacterization I gaveearlier- thus, Steedman’sTheme/Rheme
indicate how, informally put, theutterancerelatesto theprecedingdiscoursecon-
text.10
Steedmanalsodefinesa second dimension of informationstructure. This di-
mensionfirst of all partitions theRhemeinto a focusanda background. Thefocus
of a Rhemeis that ‘inf ormation’ that is marked in the surface form, whereasthe
backgroundof theRhemeis its unmarkedpart. In English, this correspondsto the
focusbeingmarkedby a pitch accent, whereasthebackground is unmarked by ei-
ther a pitch or a boundary. In a similar move Steedmandividesthe Themeinto a
focusanda background, with thatdifferenceto theRhemethat theTheme’s focus
is optional. Therecan,but neednot, bea marked element in theTheme’s surface
realization.
This partitioning is related to Halliday’s Given-New dichotomy(1985), andto
thePraguiandivisionof contextual boundnessinto contextually bound/contextually
nonbound. It concernsthedistinction betweenelements in thesentence’s meaning
which contribute to distinguishing the Themeand the Rhemefrom other alter-
natives that the context makes available, in the senseof Rooth’s alternative sets
(Steedman,2000a)(p.656).
Theexamplesbelow illustrateSteedman’scharacterization of information struc-
ture in more detail. Steedmanformalizesthe Themeof a sentence as aÞ-term
involving a functional abstraction, like Jackendoff or Peregrin. The Rhemeis a
term that canbe applied to that abstraction, after which we obtain a proposition.
As CCG is a categorial grammarcombining aÞ-calculus to represent linguistic
meaning, this proposition hasthe samepredicate-argumentstructure asthe com-
position of the canonical sentencewould have resulted in. For example, consider
theexample in (67a) andtherepresentation of its Themein (67b).
(67) English
a. (Whatdid Kathy prove?)
(Kathy proved× É4ÿ�Æ Ë�Æ (P=NP × Ígÿ�Æ Ë�Æ .10It shouldbe observed thoughthat his notion of Themeis not similar to Halliday’s useof that
term- inspiteof Steedman’s criticismof Halliday. For Halliday Themerelatesto thematicstructure,not to theinformationstructureof anindividual sentence.
54À Theoriesof InformationStructure
b.Þ ÷ Ü ��������� ÷ ��������å
Because thefunctional abstractionis closely relatedto theexistentialoperatorä , thecontext of (67a) could instantiatetheexistential asin (68).
(68)
� prove’ undecidability’ Kathy’,
prove’ canonicity’ Kathy’,
prove’ infatomability’ Kathy’
prove’ P=NP’ Kathy’
�The set in (68) is an alternative set, i.e. a setof potential alternative instan-
tiations. Steedmancalls it the rhemealternative set, andit holdsthat the Theme
presupposestherhemealternativesetwhereastheRhemerestricts it.11 Thedistinc-
tion of a focus anda background in theRheme,andpossibly in theTheme,helps
to setit apartfrom otheralternativesavailablein thecontext. In particular, wehave
thatthefocuswithin theRhemerestrictstheRhemealternative setpresupposedby
theTheme.
Furthermore,wecanconsiderthesituationin whichaThemeindeeddoeshave
a focus,realized by a markedform. Steedmangivesin (2000a)(p.659) the follow-
ing example, (69).
(69) English
(I knowthatMarcel likesthemanwho wrotesthemuscial.
But who does heADMIRE?)é MarcelÁ ÂEà ÄÏ0È Ø���� Í��dÊ Û�� ADMIRESÁ ÂEà Ä� � Ø Ê��Á ÂEà ÄÚ ÿ�Æ Ë�Æ×Eé the womanwhoÁ ÂGà ÄÏ0È Ø���� Í��dÊ Û�� DIRECTEDÁ ÂEà Ä� � Ø Ê�� themusicalÁ ÂEà ÄÏ0È Ø���� Í��,Ê Û��Á ÂEà ÄÅ�ÿ�Æ Ë�Æ
×
Steedmanargues that the significanceof having a pitch accent on “admire”
seemsto be in the context offering alternativesthat only differ in the relation be-
tweenMarceland ÷ . A marked Themeis representedasin (70).
(70) ä7÷ Ü�� �! #"%$ �&��'� ÷(")� �&*��,+-The utteranceof (69) would be infelicitous if the context would not contain
an alternative, like the ä7÷ Ü + $/. ��'0 ÷�")� �&*��,+- we have here. The setof alternative
Themesprovidedby thecontext of (69) is givenin (71).11As Steedmannoteshimself,for examplesin (2000a)(p.10),alternative setsareusedfor reasons
of expositionratherthanpresentingtheonly possiblemeansof formalization.For example,it is notdifficult to seehow alternative setsin asense‘extensionalize’accessibilityin a modallogic’s frame.
Theoriesof InformationStructure /55
(71)
� ä7÷ Ü �! #"%$ �#��', ÷�")� �&*��,+- ,ä7÷ Ü + $/. ��', ÷1")� ��*��2+3 �The kind of alternative set given in (71) is what Steedmancalls the Theme
alternative set. TheThemepresupposes alsothis set,andit is the Theme’s focus
thatrestrictsit.
Although Steedman doesnotdiscussrecursivity of focusandbackground, they
appear to berecursive in thesamesenseasFGD’scontextually bound/contextually
nonbounddistinction. For example,consider (72)
(72) English
(“Do you seethatold boatnext to theAmsterdam?”)
I canseea VERY old SHIP next to theAmsterdam.
OnepossibleSteedman-style analysisof (72) is given in (73). Like in (60) or
(61), theRhemeis modifiesa headthat is itself partof theTheme.
(73)I can see a4 576 89;:7<>=0?�@BA/CED2F4 5�6 8GH�I�JKI
VERY
H*4 5�6 8L1A/<MC�N old4 5�6 89;:7<>=0?�@0AOCED2F4 5�6 8PQHEI>JRITALLSHIP
L+H*4 576 8L1AB<MC�N next to the Amsterdam4 576 89 :S<>=0?�@BA/CED2F4 576 8GH�I�JKIIn (1996; 2000c) and in (2000a) Steedmanelaboratesa grammarthat shows
how the above kinds of information structure-enriched representations of a sen-
tence’s linguistic meaningare related to English tune. Hoffman worked out in
(1995b; 1995a) a version of CCGthat modelsTurkish freeword order. Hoffman
coupledthat to aslightly different theoryof information structurethatstandsinbe-
tweenSteedman’s account andVallduvı’s informationpackaging. In Chapter 3 I
devotemoreattention to Hoffman’sproposalfor modeling freewordorder in acat-
egorial grammar,whereasin Chapter5 I return to Steedman’s account of English,
focusing in particularon his modelof tune.
2.4 INFORMATION PACKAGING
Startingwith Vallduvı (1990), variouspeople have contributedto a perspective on
information structurecalled informationpackaging, bothin its aspectsof discourse
interpretation andgrammatical realization. Thebasicidea of informationpackag-
ing canbetracedbackto Chafe’s (1976), wherehe introduced the termexplicitly
asfollows (p.28):
56À Theoriesof InformationStructure
“I have beenusingthe termpackaging to referto thekind of phenomenaat
issuehere,with theideathatthey haveto doprimarily with how themessage
is sentandonly secondarily with the messageitself, just as the packaging
of toothpastecanaffect salesin partial independenceof the quality of the
toothpasteinside.”
Vallduvı definesinformationpackagingin (1990) as“a smallsetof instructions
with which the heareris instructedby thespeaker to retrieve the information car-
ried by thesentenceandenterit into her/his knowledge store.” (p.66) To work out
theperspective emanating from this definition,Vallduvı employs thefile metaphor
from Heim’s File-Change Semantics(1982), constructing a theory of how infor-
mationstructureis interpreted in the larger context of a discourse. Vallduvı’s use
of the file metaphor in (1990; 1994) hasbeencriticized by Dekker andHendriks
in their (1994) andin (Hendriks andDekker, 1995); seealsoKubon’s (1998). It
is for this reason that I devote relatively little attention to how Vallduvı’s infor-
mationpackagingguidesdiscourseinterpretation. Instead, I focus on Vallduvı’s
basic proposal for characterizing informationstructureasa tripartitedivisionof a
sentence’s surfaceform. Cross-linguistic justification for this characterizationhas
beenarguedfor by Vallduvı in (1990) andtogetherwith Engdahl in (1994; 1996).
(Engdahl andVallduvı, 1994; Manandhar, 1994) discussintegration of information
packaginginto HPSG, andHendriks presentsa proposalfor including information
packaginginto a Lambek-style categorial grammar,(1994; 1996; 1997).
Vallduvı reflects in Chapter 3 of his (1990) on various approaches to whathecalls
‘info rmational articulation’. Vallduvı divides theseapproachesinto topic/comment
approaches and focus/ground approaches. Both (typesof) approachessplit a sen-
tence, or rather its meaning, in two parts. The topic/commentapproachsplits the
meaning into a part that thesentenceis about, which is usually realizedsentence-
initi ally, anda comment.To follow Halliday, this ‘topic’ is thepoint of departure
for whatthesentenceconveys.12
According to what Vallduvı termsthe focus/ground approaches, the sentence
is divided into ‘focus’ anda ‘ground’, with the ‘focus’ being the informative part
of the sentence’s meaning. The ground anchors the sentence’s meaning to what
the speaker believes the heareralready knows. The ‘focus’ expresseswhat the
speaker believes to contributes to the hearer’s knowledge. The ‘ground’ is also
knownas‘presupposition’ or ‘open proposition’ - the latter beingexplainable,at
12NotethatHalliday (1985) callsthis ‘topic’ theTheme,(Halliday, 1985)(p.59).
Theoriesof InformationStructure /57
leastformally, by Jackendoff’sÞ-termrepresentation mentionedearlier.
Vallduvı arguesthatboth traditionssuffer from various problems. Asidefrom
terminological confusion, bothtraditionssuffer from thefundamentalproblem(ac-
cording to Vallduvı) that they are incompletein their empirical coverage, neces-
sarily sobecause“a binomial informational division of thesentenceis simply not
enough.” (1990)(p.54) For example,consider the examplein (74) adapted from
Dahl (1974).
(74) a. WhataboutJohn? Whatdoes hedo?ÉT�MUGÎ Øà ÄEÁ ÂJohnÁ ÂEà Ä� Í0�dÊ Û�� Ø �,Ë2Ë2Æ Û Éà ÄGÁ Â
drinks beerÁ ÂGà Ä� � Ø Ê��b. WhataboutJohn? Whatdoes hedrink?ÉT�-U�Î Øà ÄEÁ Â
John
Ø �dË Ë�Æ Û Éà ÄEÁ Âdrinksbeer
JohndrinksÁ ÂEà Ä� Í��dÊ Û�� beerÁ!ÂEÃ!Ä� � Ø Ê��The fact that the two perspectivespartition (74b) differently is taken to show
that“neitherof themis by itself capable of capturing all the informational distinc-
tionspresent in thesentence”(Hendriks,1994)(p.93). Vallduvı noticesthat thereis
acertain overlapin how thetwo perspectivesdivide(74b),andproposesto conflate
thetwo perspectivesinto a single, hierarchically structuredtrichotomy.
Vallduvı’s trichotomyof a sentence’s surface form is centeredaround a binary
division accordinginformativity, in thesenseof the‘focus/ground’ tradition. There
is a ground, thatanchorsthesentence’s meaninginto thepreceding discourse,and
a focus that specifies the ‘new’ information. In addition, the ground is further
divided into a link anda tail. According to Vallduvı, the link specifieswhere to
anchor the information specified by the focus, and the tail indicateshow it fits
there(1994)(p.5). Unlike is thecasewith FGD’s contextually bound/contextually
nonbound-distinctionor Steedman’sfocus/background,Vallduvı’sprimary notions
arenot (entirely) recursive. For example, we do have an information packaging
analysis(75) for (61), with V indicatingthelink and ë thefocus.
(75)ÖYesterdayI mettheteacher Ù W Ö
of CHEMISTRY.Ù�ÝHowever, becauseinformationpackaging partitions a sentence’s surfaceform
(rather thanits linguistic meaning), we have to consider British in (76) ashaving
thesameinformativestatusasits head– see(Vallduvı andZacharski, 1994) for the
argument.
58À Theoriesof InformationStructure
(76) English
(Your system doesnot includeanAMPL IFIER.)
TheBRITISH amplifiercomesHIGHLY RECOMMENDED.
(Prevost, 1995)(ex.5)
(77)ÖTheBRITISH amplifier Ù W Ö
comesHIGHLY RECOMMENDED. Ù ÝVallduvı cited examplesfrom various languagesin (1990), andpresentedto-
getherwith Engdahl in (1996) anindepth study of how alargenumberof languages
may employ different strategiesto realize informationpackaging.13 Consider the
following examples for CatalanandEnglish (78)-(80), cf. (Vallduvı andEngdahl,
1996)(p.42), which illustratethefour abstractrealizations of information structure
thatVallduvı distinguishes.
(78) Link-focus sentences: typical topic-commentstructures, predicate-focus
structures,categorical judgments.
a. ThepresidentÖ Ý hates CHOCOLATE].
El presidentX Ö Ý odiala XOCOLATA �SX ].
b. ThepresidentÖ Ý CALLED].
El presidentX Ö Ý haTRUCAT �YX ].
c. ThepresidentX Ö Ý (I) wouldn’t BOTHER ��X ].
El presidentX Ö Ý no l’ EMPRENYARIA �YX pro ].
(79) All-focus sentences: (a) neutral descriptions, news sentences, sentence-
focus structures,thetic judgments;(b) there-sentences; (c) predicate-focus
sentenceswherethelocusof update is inherited.
a.Ö Ý ThePRESIDENT called].Ö Ý Ha trucat el PRESIDENT ].
b.Ö Ý Thereareprotests in theSTREETS.]Ö Ý Hi haprotestesalsCARRERS.]
c.Ö Ý (He) HATES (it).]Ö Ý L Z ’ ODIA eZ pro.]
(80) Link-focus-tail sentencesandfocus-tail sentences:narrow focus,constituent
focus, typical open-proposition structures.
13Vallduvı andEngdahl illustrate information packagingon English,German,Dutch, Swedish,Catalan,Hungarian,Turkish,and-for completeness’sake- Japanese.
Theoriesof InformationStructure /59
SS IS
LF
PF
DS
Figure2.1: Vallduvi’s grammararchitectureincorporating informationstructure
a. ThepresidentÖ Ý HATES ] chocolate.
El presidentX Ö Ý l Z ’ ODIA t Z t X ,] la xocolataZb. Thepresidenthates
Ö Ý CHOCOLATE.]
El presidentÖ Ý t [ la XOCOLATA t X ,] odia[ .
Originally, Vallduvı (1990) proposedto integrateinformationpackaging into
a GB-stylearchitecture, with informationstructureasa autonomous stratum, next
to deepstructure (DS), logical form (LF), phonological form (PF), and surface
structure(SS).
In (1994), Engdahl and Vallduvı elaboratea different approach,making use
of HPSG. The basic idea is to expand the CONTEXT field with a feature INFO-
STRUCT, asshownin (81).
(81) \]]]]^ CONTEXT
\]]]]^ INFO-STRUCT
\]]]^ FOCUS ...
GROUND\^ LINK ...
TAIL ..._` _baaa` _baaaa` _baaaa`In addition, the PHON field is expanded to specify accent aswell. Following
Jackendoff, EngdahlandVallduvı representthepossiblechoicesof accent asA,B,
or unmarked(u).
(82) a. Word c d \^ PHON—ACCENT A
INFO-STRUCT—FOCUSd _`
b. Word c d \^ PHON—ACCENT B
INFO-STRUCT—L INKd _`
c. Word c \^ PHON—ACCENT u
INFO-STRUCT—FOCUS [] _`
60À Theoriesof InformationStructure
Naturaly, the accent-assignments in (82) are for English. The assignmentof
an accent to a particular structure (not necessarily a word, or a constituent in
the phrase-structure grammarsense) imposes a constraint on how that structure
is (to be) interpreted informatively. Thesespecifications get instantiated, and in-
herited,by meansof a rule schemathat operatesnext to PollardandSag’s Head-
ComplementSchemaandHead-Subject Schema, cf. (Pollard andSag,1993)(p.402).14.
In addition to satisfying these schemas,phrasal signshave to satisfy the INFO-
STRUCT instantiationsgivenby therulein (83), cf. (EngdahlandVallduvı, 1994)(p.58).
(83) INFO-STRUCT instantiation principles for English:
Either (i) if a DAUGHTER’s INFO-STRUCT is instantiated,thenthemother
inherits this instantiation (for narrow foci, links andtails),
or (ii) if the most oblique DAUGHTER’s FOCUS is instantiated, then the
FOCUS of themotheris thesignitself (widefocus).
Usingthis rule,EngdahlandVallduvı illustratehow various informationstruc-
ture patternscan be analyzed, like object NP focus, VP focus, and verb focus.
An advantageousaspectof their proposalis that the informationstructurecancut
acrossstandardphrases:subject/verbfocuscanalsobeanalyzed,i.e. thenotion of
constituency with respectto informationpackaging is flexible.
Vallduvı’s theory of information packaging hasfound its way primarily into
HPSG,dueto (Engdahl andVallduvı, 1994) – for example, see(Kolliakou, 1998;
Alexopoulou, 1999). Hendriks proposesin (1994; 1996; 1997) a categorial gram-
marsystem, based on MoortgatandMorrill’ s D calculus(1991), in which hetries
to capture various insightsof informationpackaging.
2.5 REFLECTIONS
At various occassions people have compared FGD’s theory of TFA, Vallduvı’s
information packaging, and Steedman’s Theme/Rheme-basedinformation struc-
ture. Vallduvı presentsin (1990) adiscussionof variousapproaches,amongwhich
Praguian proposals, anda more recentdiscussion canbe found in (Vallduvı and
Engdahl, 1996). Hajicova andKruijf f-Korbayova comparein (1999) FGD’s topic-
focus articulation to all of the approaches discussedin this chapter, andconclude
that the Praguian viewpoint presents various advantagesover the approachesre-
flectedon there. Finally, Kruijf f-Korbayova and Webberfocus in a number of14From the viewpoint of dependency grammar,it is interestingto observe that thesetwo HPSG
schemataareID schemata- they concernimmediatedominance,not linearization.
Theoriesof InformationStructure /61
paperson CzechandEnglish andrely on a symbiosisof FGD’s topic-focusarticu-
lation andon Steedman’s theory, cf. (2000).
What all the approachesI discussedhere have in commonis that they con-
ceive of grammaras the appropriate place to describe information structure and
its realization. Thereare several interesting consequences that follow from that
perspective.
First of all, the interpretation of informationstructure belongsto the level of
discourse. Hence,the ‘proposition’ expressedby a sentence’s linguistic meaning,
including informationstructure,cannot beassigneda truth-value in thesystemof
grammarfor it doesnotproperly needto expressone.15 Thisobviously goesagainst
the views advancedby Montague Grammarand categorial grammarapproaches
thatarebasedon it, like Morrill’ s (1994). But it is for this reason thatHajicova et
al speak of a “Post-MG” semanticsin (1998), following insightsthatwerealready
present in earlier work like (Sgallet al., 1986).
We cancarrytheconsequencesof this viewpoint further. For one,thecrispdi-
vision of thelanguagesystem into -roughly- syntax,semantics andpragmaticscan
no longer bemaintained,sinceinformationstructuredissolvestheclearborderline
betweenwhatMorris andCarnapconsideredto be“semantics” and“pragmatics”.
This is whatPeregrin calls in (1999) thepragmatization of semantics. Themean-
ing expressedby a sentenceassuchis no longer context-independent,asCarnap
assumed. Thesentence’s linguistic meaning with is informationstructuresignifies
a dependenceon thelarger context in which thesentenceis uttered.
At thesametime,informationstructureis aproperty thatbelongsto thelevel of
individualsentences– it hasno referenceto theorthogonaldimensionof thematic
structure that deals with textual organization. It is to this larger organization that
Halliday’s Themecontributes,independently of the local sentential organization.
Thereis a close relation between informationstructure andthematic structure,as
arguedfor by for exampleHallidayandby Danes,but theirstrategieshavedifferent
aims.16
15I deliberatelyputproposition inbetweenquoteshere,asI donotmeanany moretechnicalnotionby it than“a statementthat hasa truth-value”. Differenttheoriesmay formalizethe meaningof asentencedifferentlyat thelevel of grammar;I amnot concernedwith thosedifferenceshere.
16Considerthe following descriptionof the relationbetweenTheme/Rheme,andHalliday’s in-formationstructure,from (Halliday, 1985)(pp.299-300):“There is a closesemanticrelationshipbe-tweeninformationstructureand thematicstructure(...). Other thingsbeing equal,a speaker willchoosethe Themefrom within what is Given andlocatethe focus, the climax of the New, some-wherewithin theRheme.But althoughthey arerelated,Given+ New andTheme+ Rhemearenotthe samething. The Themeis what I, the speaker, chooseto take asmy point of departure.TheGiven is what you, the listener, alreadyknow aboutor have accessibleto you. Theme+ Rhemeis
62À Theoriesof InformationStructure
Steedman’suseof thetermThememustthereforebedistinguishedfrom Halli-
day’s,asSteedmandoesacknowledgein (2000a).Steedman’sdefinition of Theme
diverges from Halliday’s Themefor examplein that Steedman’s Theme(i) does
nothave to besentence-initial, but canbeorderedeither beforeor aftertheRheme;
(ii) the Themecancontain multiple experiental elements(for example, multiple
circumstantial modifiers,but alsomultiple participants, aswell asthemainverb),
and(iii) disjoint partsof a sentencemight belong to theTheme,cf. (2000a)(p.7).
However, whereas Steedmanthusplaceshis Themesquarely within therealm
of information structure, Vallduvı arguesthat the description of a sentence’s role
in thematic structure andits informationstructure should be conflated into a sin-
gle construct, namelyhis focus/ground. According to Vallduvı, the link-part of
the ground corresponds to Halliday’s Theme(cf. (1990; 1994), seealso (Hen-
driks, 1994)) andarguably “providestheexplanation for sentence-initial topiclike
phrases.” (1990)(p.54)However, to begin with it is notclear how Vallduvı provides
amodelof what,in Halliday’s terms,wouldbethefirst experiental element, which
neednot bethefirst phrase(84a), if it is a phraseat all (84b-c).
(84) English
a. Now, ... (Bateman,p.c.)
b.Ö Ú ÿ�Æ Ë�Æ Fromhouse to house] I wendmy way.
(Halliday, 1985)(p.40)
c.Ö Ú ÿ�Æ Ë�Æ On theground or in theair ] smallcreatureslive andbreathe.
(Halliday, 1985)(p.40)
But, worse, Vallduvı’s argumentfor theneed for sucha conflation is far from
convincing. Therearealternative waysof explaining topic-initial phrases,without
having any direct recourseto Halliday’s Theme,andby obliterating the distinc-
tion betweenthetwo orthogonal dimensionsof thematic structureandinformation
structureVallduvı’s focus/ground in fact no longer enablesus to explain phenom-
enathatarepossible exactly becauseof theabovementionedorthogonality.
For example,Halliday givesin (1985) various examplesof sentencesin which
the Themeis actually locatedin the New information. In (85a), “seen” is used
contrastively, so it cannot be Vallduvı’s link, whereas(85b) purportedly can be
analyzedasan all-focussentence. Finally, example(85c) comesfrom Steedman,
andhasonereading in whichHalliday’sThemecorrespondsto Steedman’sRheme
focus
speaker-oriented,while Given+ New is listener-oriented.”
Theoriesof InformationStructure /63
(85) English
a.Ö e Æ/f Ö Ú ÿ�Æ Ë�Æ I ] haven’t SEEN] you for ages.
(Halliday, 1985)(p.301)
b.Ö e Æ/f Ö Ú ÿ�Æ Ë�Æ Theboy] stoodon theburning DECK.]
(Halliday, 1985)(p.297)
c.Ö Ú ÿ�Æ Ë�Æ NIXON] died.
cf. (Steedman,2000c)(p.119)
Vallduvı doesaddress‘focus-preposing’ (or focalfronting)constructions,which
couldbetakento indicatetheproblemsnotedabove. Basedon Catalandata,Vall-
duvı arguesthat an analysis canbe given whereby the focusdoesremainclause-
final (Vallduvı, 1990)(p.132) becauseanemptycategoryis retained.Thus,Vallduvı
wouldhaveit, noproblemsarisewith focal fronting sincethegeneralinformational
articulation of thesentence(86) canbemaintained.
(86)Ö g7h
linkÖ gSh Ö g7h
focus] tail ]]
(Vallduvı, 1990)(p.132)
However, it canbe seriously doubted whether Vallduvı’s analysis still stands
any groundgiventherejectionnowadaysof emptycategories,arejection prevalent
in HPSG aswell. Moreover, Vallduvı’s argumentstill leaves the issue outstand-
ing thatdistinguishing a sentence’s informationstructurefrom its role in thelarger
thematicstructure doesnot indicatea redundancy, but is an essential difference
betweentwo opposite perspectives. Vallduvı’s strategy of imploding thesetwo
perspectivesinto a single characterization givesrise to morequestions thanit an-
swers.
And, asevenVallduvı pointsouthimself, therearealternativewaysof explain-
ing the relation between thematic structure and information structure. Vallduvı
indicatesthatthePraguian scaleof communicativedynamismcould possibly over-
comethenoted‘f ailure’ to explainsentence-initial topics(1990)(p.55).17. Hajicova
andKruijf f-Korbayova voicethesameopinion in (1999)(p.229).
Anotherconsequenceof placing informationstructure in grammar is that its de-
scription thus getsplaced in the larger context of explaining the (grammatical)17Interestinglyenough,Vallduvı saysthat“it mustbepointedoutthatit [discerningcommunicative
dynamismwithin informationstructure,GJMK] violatesany autonomy-of-levels hypothesis,sinceit bringsalonga direct interactionat th esamelevel betweenthematicandinformationalconsider-ations.” (1990)(p.55) Although I do believe thatVallduvı hasa point here,the commentis slightlysurprisingin thelight of thediscussionabove.
64À Theoriesof InformationStructure
systemof natural language.Oneissuethat thuscomesinto view is that of cross-
linguistic realization of informationstructure. In theapproaches I discussedabove
we can find for exampleMathesius’s early work on English and Czech(1936),
Sgallet al’s contrastive studiesin (1986) and(Hajicova et al., 1998), andVallduvı
andEngdahl’s (1996) andHendriks’ (1994).
However, at least from a functionalist perspective, onewould like to seethis
cross-linguistic study takenfurther: Namely, to thepoint wherea grammarwould
describe informationstructure cross-linguistically in the senseof how languages
differ in realizing information structure and what they have in commonin their
strategies.Thecross-linguistic studiesobserve, but hardlypredict - andprediction
is whatany propertheory should do,including onedescribinginformationstructure
andits realization.
In a certainsenseHendriks thus missesthe point when he states that gram-
mar frameworks need to be powerful enough to describe the different strategies
languagesemploy in realizing information structure (1994; 1997). A grammar
framework, incorporating a cross-linguistically adequatedescription of informa-
tion structure,mustbe powerful enough to explain why languagesdiffer in their
strategies,andpredict what they may have in common. Obviously, this in a step
further from simply realizing that therearedifferent strategiesandthatwe should
beableto modelthem.
All of theapproachesI discussedabove distinguishthemselvesfrom other ap-
proachesby actually explicitly discussing the corefunction grammarperformsin
explaining informationstructure.They do soin differentways,though.
In the above we already arguedthat informationpackaging shows shortcom-
ings on various points. First, its argumentsfor collapsing of thematic structure
and informationstructure aredisputable. Second,its characterization of the pri-
marynotionsof groundandfocus aspartitionsof a sentence’s surfaceform leads
to problemswith recursivity, andappears at oddswith thegenerally acceptedidea
of informationstructurebeing anaspect of linguisticmeaning. Third, Engdahl and
Vallduvı do present in (1994) a proposal for how to integrateinformation pack-
aging into HPSGandrelate it to a modelof tune,but the model remains -as the
authors admit- simple. Neitherdo Engdahl andVallduvı, or Manandhar for that
matter, showhow onecould explain word order asa strategy for realizing word
order. (Kolliakou, 1998) and(Alexopoulou, 1999)do elaboratethe HPSG-based
approachin that direction. However, they do so using purely syntactic devices.
Finally, going back to the Government& Binding modelproposedin (Vallduvı,
Theoriesof InformationStructure /65
1990), or usingSelkirk’s prosody model(cf. (Steedman, 2000a)),would not over-
comeany of theseproblems.In any of these models,predicate-argument structure
and informationstructure areseparated. Steedmanconvincingly arguesthat this
separation is wrong. Also in FGD’s stratificational model, cf. Sgallet al’s (1986)
and Petkevic’s (1995), whereinformation structure is an ingerent component of
linguistic meaning, wedonotfind aseparationof levelsthatareresponsiblefor re-
alizing information structure,apart from othersthatwouldcarefor realizing “bare”
linguistic meaning. Fromtheperspective of FGD, including the larger (Praguian)
viewpoint that surfacesyntactic phenomenalike word order interact with tuneto
realize informationstructure,sucha separationwould go against the fundamental
relation between linguistic meaning andits topic-focusarticulation.
This thenleadsusto considerSteedman’sapproach,andFGD.Regarding their
views on information structure, both place it at the level of linguistic meaning.
They bothemploy primitivenotionsthatarerecursive(CB/NB, focus/background).
Also, they allow for amoderateform of recursivity whereit concernsTheme/Rheme
or topic/focus: informationstructurescanbeembeddedwhenit concernsembed-
dedclauses(Hajicovaetal.,1998)(p.160), (Steedman,2000c)( Ð 5.7.2).18 Finally, it
seemsplausible to consider the contrastive topic marker * (Hajicova et al., 1998)
asthe counterpartof Steedman’s Theme-focus,andthe focus proper asthecoun-
terpart of Steedman’s Rheme-focus.
Hence,at the level of informationstructure thereappear to be variouscorre-
spondencesbetweenSteedman’sapproachandFGD.However, they differ substan-
tially whereit concernsthe underlying views on grammar. Steedmandevelops a
monostratalformalism(CCG)in whichsurfaceform andunderlying meaning(with
information structure)arecompositionally related, (Steedman,2000a; Steedman,
2000c).
FGD,ontheotherhand, proposesastratificationalapproach(Sgalletal.,1986;
Petkevic, 1995; Petkevic, in prep). (Sgall,1980)specifies transducers that gener-
ateasurfaceform givenatopic-focusarticulation,and(Petkevic, in prep)develops
themathematical devices to generaterepresentationsof sentential linguistic mean-
ing with topic-focusarticulation. FGD lacksfurther specifications of transducers
to turn Petkevic’s representations into surface forms, which is an acknowledged
shortcoming(Sgall,p.c.).Furthermore,thereappearsto beaproblematicdifference
betweenthe linguistic view on the grammatical phenomenainvolved in realizing
information structure, andtheir possible technical implementation in a stratifica-
18For anearlierdiscussionof recursivity of TFA in FGD,see(Sgallet al., 1986)(i 3.11).
66À Theoriesof InformationStructure
tional framework. For example,wealready notedearlier thatinformation structure
is oftenrealizedusinga combination of variousmeanslike tune,morphology and
word order. On a stratificational approach,one is -normally- technically forced
to assumethat thereis a relation between thesedifferentmeanswhereword order
restricts morphology, andmorphology restricts tune. However, this seemsimplau-
sible. Thereratherappearsto bean interaction in which differentlinguistic means
mutually restrict oneanother to constructa well-formedsurface realizationof the
underlying informationstructure. We understand(Sgall et al., 1986) to argue for
this interaction linguistically, but without furtherimplementation of theframework
it is difficult to judgewhether this view could bemaintained technically.
Steedman’s CCG providesa grammarformalism in which informationstruc-
tureis compositionally relatedto ananalysisof asurfaceform. (Steedman,2000a)
focusesontuneasastructural indication of informativity, thoughit wouldbeincor-
rectto claim thatSteedman’s theory of information structureis reducedto prosody
asdonein (Hendriks, 1997). Thevariation in focus/background thatSteedman ex-
plainsusingmarkednesscanberelatedits realizationaspitchaccents,but neednot
be. Thefocus/background distinction canbeapplied to explain variationsin word
order realizationaswell, asKruijf f-Korbayova andWebberdo. Kruijf f-Korbayova
andWebber’s approachis purely semantic though, without any referenceto gram-
mar.
CCGhasbeenextended to cover free word order. Hoffman (1995a)presents
multiset combinatory categorial grammar(MCCG), which relates an account of
Turkish word order to an information packaging-inspired theory of information
structure. MCCG is a grammarframework that hasa greater generative strength
thanCCG.Baldridge(1998; 1999)presents Set-CCG,a moreconservative exten-
sionof CCGthat is capable of explaining freeword order(including Turkish)but
which has the sameformal and computational properties as CCG. What CCG,
multiset combinatory categorial grammar,and Set-CCGall have in commonis
Steedman’s Principle of Adjacency, (Steedman,2000c). According to this princi-
plecombinatoryrulesmayonly apply to finitely many phonologically realizedand
string-adjacententities. Theemphasizedphraseis important. As is obvious from
MCCG andSet-CCG,thePrinciple of Adjacency meansthatonemodelsvariabil-
ity of wordorder directly in thelexicon. TheMCCGor Set-CCGlexical categories
imposelessrestrictionson thedirectionality in which argumentsneed to becom-
bined with, andmodelvariability in thatway.
But what doesthat meanfor modeling the effect of word order variability -
Theoriesof InformationStructure /67
namely, its useasastructuralindicationof informativity? AsHoffman(1995a)(p.151ff)
observes,all thepossibleordersthataparticular categoryallows for could becom-
piled out, andthe informationstructureof eachof theseorderscould thusbecap-
tured lexically. However, sucha formalism would not be able to capture the in-
terpretation of adjunctions in different word ordersor with long-distance scram-
bling. To overcomethis problem,Hoffmanproposesto split thegrammarinto two
components: Lexical (linguistic) categories and combinatory rules to derive the
predicate-argument structureof a sentence,andso-calledOrdering categories to-
gether with application rules(andidentity) to derive theinformationstructureof a
sentence.
Essentially, Ordering categories aretemplatesof surfacerealizationsof infor-
mation structure, specifying wherethe focus, topic, and ground componentsof
Hoffman’s informationstructure needto befoundin thesurfaceform. Every word
in thesentenceis associated with a lexical category, which is then associatedwith
anorderingcategory. Thegrammatical analysisof thesentenceis aninferenceover
lexical categories using MCCG’s combinatory rules, andcompositionally builds
the underlying argumentstructure. In parallel to this inference,we have infer-
enceover theorderingcategories(associated to words)thatcompositionally builds
thesentence’s underlying information structure. Hoffmandefines hersystemsuch
thatthegrammaticalinferenceandtheinformationstructureinferencecontrol each
other:Composition in oneinferencecanonly bedone iff it is possible in theother
inference. It is in this way, Hoffman argues,that “syntactic andpragmatic con-
straints work together to determine the surfacestructure and word order of the
sentence.” (1995a)(p.160)
However, compiling outpossible informationstructureinto Orderingcategories
is a rather“extensional” approachto explaining word orderasa structural indica-
tion of informativity –necessitated by CCG’s Principle of Adjacency– andraises
doubts about the possibility of having multiple levels of linguistic information
interact in realizing information structure. For example, Hoffman doesdiscuss
examplesthat illustrate how tune andword order interact to realize information
structure,andshowshow shecanrepresent these informationstructures,yet there
is no formulationof theactual inferencemechanismsthatwould leadto these rep-
resentations. Hoffman refers to Steedman’s earlier work on tune, andarguesthat
herapproachis similar to his. Comparing (Hoffman,1995a) to Steedman’s recent
(2000a) revealsthat there is a substantial difference,though, nowadays. Steed-
manconsidersseparateprosodic categories and lexical categories, but the effect
68À Theoriesof InformationStructure
of the prosodic categoriesis a specification of an INFORMATION feature on lexi-
cal categories. Hence, Steedmancanusea singular setof inferencerules that lets
informationstructurereflectdirectly in thepredicate-argumentstructures.19
As with any suchextensional approach,it canbedoubtedthat this would pro-
vide uswith a flexible enough setting to capturecomplex phenomena.And, aside
from the issuewhetherit is technically elegant to discern separatecategoriesfor
different levels of linguistic information,it doesnot seemto lend itself very well
to cross-linguistic generalizations. In CCG, it is the combinatory rules that help
us specify cross-linguistic patterns, not the categories (Steedman, 2000c; Kruijf f
andBaldridge,2000) - but in theonly existing combinatoryaccount of word order
and information structure (Hoffman, 1995a) the description of the realizationof
informationstructureis decoupled from thesyntactic inference.
To recapitulate, the notionsof informationstructure that FGD andSteedman
proposearesimilar, andarenot subject to theproblemswe cannotefor informa-
tion packaging. FGD andSteedmandiffer in the way they describe information
structureand its realization. Theoretically, FGD canbe understoodto argue for
therealization of informationstructureasan interactionbetweendifferentmeans,
like tuneandword order. They areall parametersin the realization– though we
canquestion in how far it is possible to achieve this interaction technically in a
transformationalapproach.Thecombinatory tradition hasyieldedvariousformal-
izations that showhow either word order (Hoffman, 1995a) or tune (Steedman,
2000a; Steedman, 2000c) canbe related to information structure. However, we
canquestion whetherthe theoretical backgroundleads to descriptionsof informa-
tion structure andits realization that are linguistically intuitive andgeneralizable
acrossdifferenttypesof languages.CCG’sPrincipleof Adjacency seemsto neces-
sitate a formal dissociation of the description of word order from the description
of information structure,which breaks with thegeneral linguistic intuition behind
word orderasa structural indicationof informativity.
2.6 INFORMATION STRUCTURE IN DGL
DGL combinesthebestof two worlds- FGD’s view on informationstructureand
its realization as an interaction betweendifferent means,and the categorial ap-
19Considerthough that the effect of a prosodic Theme category on a lexical category likejlkQm-nYoqprm-s>t�mToupis just thespecificationof theinformationfeaturev to “Theme”
jTwEs, i.e. thelexical
category becomesjlk!m-nSoupKxSs>tSxSoup
. “Unification” of a lexical category with anorderingcategory,if onewereto go thatway, is anentirelydifferentissuethanjust a specificationof features.
Theoriesof InformationStructure /69
proach to formalizing therelation between surfaceform andunderlying linguistic
meaning. However, unlike CCG we take adjacency to be a parameter(Moortgat
andOehrle,1994) – aparameterdeterminedfor animportant partby its purposeas
astructuralindicationof informativity. Chapter4 showshow thisleadsnotonly to a
direct (ratherthandissociated) explanationof word order asa structural indication
of informativity, but alsoto an approachthat is generalizable cross-linguistically.
Chapter5 proposesareformulationof (part of) Steedman’sdescription of tune, and
shows how it canbesmoothly integratedinto theaccount of word order– without
any needfor distinguishing additional levelsof categories. In theremainder of this
chapter we present thebasicdefinitionsfor representing andinterpreting informa-
tion structure.
Definition 8 (Representingcontextualboundness).DGL distinguishesfour types
of contextualboundness: CB andNB, CB* andNB*. CB* correspondsto (Hajicova
etal., 1998)’s y marker or Steedman’s Theme-focus.NB* correspondsto thefocus
proper (Sgall etal., 1986)or Steedman’sRheme-focus. Thesetypesarerepresented
at thelevel of linguistic meaningas zb{ | unary operators,andare reflectedas }&~�� incategories. Formally, they are specificationsof an underspecifiedfeature inf. �Remark 7. A lexical category usually only specifiestheinformativity of theword
as����� . Its informativity featuregetsspecifiedin theprocessof derivation,thebasic
mechanismsof which we already discussedin Chapter 1. Due to the correspon-
dencebetween operationsoncategorial featuresandtheir reflection in theunderly-
ing linguistic meaning,determination of a feature dueto its occurrencein a struc-
tural configuration (e.g.word order, tune) meansthatthis is noted in thelinguistic
meaningaswell. For example,Chapter 4 showshow wecanderiveafull y specified
structure ��� Honza�M�����q��� � snedl����� �0� � koblihu �>�����u��} ~ ���7� , given lexical entries
that just specify ����� like snedl ��}�~�� ��� ��� ��� ��� �Y�R��� ÉT�B �¡ ��¢�� ��� �K£ ÈÉ � Æ � É>¡ � . The
corresponding linguistic meaningis ¤ ¥�¦/§ NB ¨M¦l©uª¬«�#®�¯!ª°§ CB ¨>± ACTOR ²S¦´³Rª¶µ ·!¸�¹�º¯!ª°§ NB ¨M± PATIENT ²7¦3»Kª¬¼�·!¸¾½1®�¯O¯ .�Definition 9 (Topic/focusarti culation in DGL). Like in FGD,a sentence’s topic-
focusarticulation is derivedrecursively fromtheindicationof informativityof the
individual nodes in that sentence’s linguistic meaning. To establish the topic and
thefocusof a sentence, weuserules that rewrite a logical formula just indicating
contextualboundnessto a logical formulaincludinga topic/focuspartition ¿ÁÀÃÂ(Kruijf f-Korbayova, 1998). Theideaof using rewriting stemsfromOehrle(1999),
70Ä Theoriesof InformationStructure
but therecursiontherules implementis essentially Sgalletal’sprocedure(cf. page
47) with theamendmentsof (Hajicova et al., 1998)(p.164).
Formally, we represent a topic/focuspartition as Å�Æ�3ÇÈÀÊÉË� . Ç and É are
conjunctionsof terms,whereby Ç maybe empty(in which casewe write Ì ). By
definition, É mustnot beempty;if at onepoint weobtain anemptyfocus, wewrite
that Í , andtry to find a deeper embeddedNB element to serve asfocus. A rewrite
rule is statedas Î���Ï�ÐÑÓÒÔ� , rewriting Ï into Ò . Notethat containmentÕ�zb{ | is not
recursivehere, but only to thecurrentlevel of conjunction.
(87) If a verbal head of theclauseis CB, thenit belongsto thetopic.ÎÖ��ÅuÆQ��z CB |/�-×ÁØÙÏ;�ÚØÁÛÔ��ÐÑ Å¶ÆQ��z CB |/�-×ÜØÙÏ;��ÀÝÛÞ���(88) If a verbal head of theclauseis NB, thenit belongsto thefocus.ÎÖ��ÅuÆQ��z NB |/�-×ßØÙÏ;�àØÜÛÞ�ÖÐÑ Å¶ÆQ��ÌáÀâz NB |/�-×ßØãÏ;�äØÊÛÞ���(89) If a dependent å of a verbal head is CB, then å belongs to the topic (in-
cluding anynodesit governs).ÎÖ��ÅuÆQ�-ÛæÀçz CB |3åèØÁéq�ÖÐÑ Å¶ÆQ��z CB |3åèØÜÛêÀÝéq���(90) If a dependent å of a verbal headis NB, then å belongsto the focus (in-
cluding anynodesit governs).ÎÖ��ÅuÆQ�-ÛæÀçz NB |3åèØÁéq�ÖÐÑ Å¬ÆQ�-ÛæÀëéìØíz NB |3å����(91) If a CB dependentof type å is anembeddedclause, thenit should beplaced
first (topic proper).ÎÖ��ÅuÆQ�-Ûqzîz CB |/�>å��E�Mï°zî�-×ÚØñð��O|-�O|(Àãéq�àÐÑ Å¶Æ¾��z CB |/�>å��E�Mï(zî�-×äØòð��O|-� Ø°ÛíÀÙéq�(92) If a NB dependentof type å is anembeddedclause, thenit should beplaced
last (focus proper).ÎÖ��ÅuÆQ�-ÛíÀóéËzîz NB |/�>å��E�Mï°zî�-×àØñð��O|-�O|-�äÐÑ Å¬Æ¾�-ÛíÀÙéôØõz NB |/�>å��E�Mï(zî�-×äØòð��O|-���(93) Embeddedfocus: If in ö÷Àìø , ø contains no inner participants ( åúùû
ACTOR, PATIENT, ADDRESSEE, EFFECT, ORIGIN ü ) whereas ö does,
thena NB modification of a CB dependent is part of thefocus:ÎÖ��ÅuÆQ�-Ûqzîz CB |/�>å��E�Mï°zîz NB |/�>å�ýl�E�MþÿØ��ñ�O|-�O|�ÀÃéq�ÖÐÑÐÑ ÅuÆQ�-Û zîz CB |/�>å��E�Mï(z |-�O|�À z CB |/�>å��Ez NB |/�>å ý �E�MþÿØ��ñ�ÚØÁéq���A valid topic-focusarticulation is a structure ö À ø to which we can no
longer applyanyof therewrite rulesgiven in (87) through(93),andwhere ø��� Í .�Remark 8 (The structur eof information structur e). Whatkindsof structuresdo
we obtain using Definition 9? Abstractly, whatwe obtain is a relational structure
Theoriesof InformationStructure /71
wheretherelate maybedistributed acrossthe À operator, while maintaining their
mutualrelationsthrough nominal reference.
Applying therules givenin Definition 9, we obtain thetopic/focusbipartition-
ing aspresentedin (94).
(94) English
Thecatatea SAUSAGE.
i. ¤Ô¥º¦O§ NB ¨>¦T©¶ª¶«�&®E¯ºª°§ CB ¨>± ACTOR ²7¦���ª��2&®�¯ºªò§ NB ¨>± PATIENT ²S¦� ª��#½�Y� º«&¯O¯ii. (88), ¤ ¥ ¦����ߧ NB ¨M¦l©qª¶«�&®�¯!ª°§ CB ¨>± ACTOR ²S¦��;ª��2&®�¯!ª°§ NB ¨M± PATIENT ²7¦��ª�Y�½��� !«�¯/¯iii. (89),¤ ¥�¦/§ CB ¨M± ACTOR ²S¦�� ª��2&®�¯��ߧ NB ¨>¦T©uª¶«�&®E¯!ª°§ NB ¨>± PATIENT ²7¦� ª��#½��� !«&¯O¯�
Wealsoobtainthedesired topic-focusarticulation for examples like (95):
(95) English
I mettheteacherof CHEMISTRY.
i. ¤Ô¥º¦O§ CB ¨>¦T©ÿª��)«�«�®S¯ ª § CB ¨M± ACTOR ²7¦��¶ª��S¯ªò§ CB ¨>± PATIENT ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦���ª���µ «�����Y® �"!1¯/¯O¯ii. (87), ¤ ¥º¦O§ CB ¨>¦T©ÿª��)«�«�®S¯#� § CB ¨>± ACTOR ²S¦��¶ª$�S¯ªò§ CB ¨>± PATIENT ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦���ª���µ «�����Y® �"!1¯/¯O¯iii. (89), ¤ ¥ ¦O§ CB ¨M¦l©ÿª��)«�«�®S¯ ª § CB ¨>± ACTOR ²S¦��¬ª��S¯�ߧ CB ¨M± PATIENT ²S¦��ºª¶®Y«���Eµ�«���ªò§ NB ¨M± APPURTENANCE ²S¦���ª%��µ «�����Y® �"!�¯/¯O¯iv. (89), ¤ ¥ ¦O§ CB ¨>¦T© ª$�)«�«�®S¯ ª § CB ¨M± ACTOR ²7¦��uª$�S¯§ CB ¨>± PATIENT ²7¦��ºªu®�«����µ «���ª°§ NB ¨>± APPURTENANCE ²7¦���ª��Eµ «��&��Y® �"!�¯O¯'�)(Ô¯v. (93), ¤ ¥ ¦O§ CB ¨>¦T© ª$�)«�«�®S¯ ª § CB ¨M± ACTOR ²7¦��uª$�S¯§ CB ¨>± PATIENT ²7¦��ºªu®�«����µ «��2¯'�ܧ CB ¨M± PATIENT ²S§ NB ¨>± APPURTENANCE ²7¦�� ª��Eµ�«�����S®*�"!�¯O¯�
The rewriting in (95) reliescrucially on the rule that handles embeddedfoci,
(93). The formulation of this rule is ’different’ from (Sgall et al., 1986), in the
sensethatit is ageneralization similar to proposalsin Koktova (1995). Therewrite
rule (93) enablesusto dealproperly with exampleslike (96),which areanswers to
so-called double-focus questions.
(96) English
(Whomdid you give whatbook?)
I gave thebookon SYNTAX to KATHY.
72Ä Theoriesof InformationStructure
(97) i. ¤ ¥ ¦O§ CB ¨M¦l©ÿª$ +��,º«�¯ ª § CB ¨M± ACTOR ²7¦��¶ª��S¯ª § CB ¨>± PATIENT ²S¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²S¦�äª� !Q¸®�32 ¯/¯ª § NB ¨M± ADDRESSEE ²7¦�4ôª�5°&®Yµ6! ¯O¯ii. (87), ¤ ¥ ¦/§ CB ¨M¦l©ÿª$ 1��,º«&¯7� § CB ¨>± ACTOR ²S¦��¶ª$�S¯ª § CB ¨>± PATIENT ²S¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²S¦�äª� !Q¸®�32 ¯/¯ª § NB ¨M± ADDRESSEE ²7¦�4ôª�5°&®Yµ6! ¯O¯iii. (89), ¤ ¥�¦/§ CB ¨>¦T© ª$ +��,!«&¯¬ª § CB ¨>± ACTOR ²S¦��¶ª��S¯� § CB ¨>± PATIENT ²7¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²7¦�àª$*!Q¸®Y�2�¯O¯ª § NB ¨M± ADDRESSEE ²7¦�4ôª�5°&®Yµ6! ¯O¯iv. (89), ¤ ¥#¦/§ CB ¨M¦l©ÿª$ 1��,º«&¯¬ª § CB ¨M± ACTOR ²7¦��¬ª��S¯ª § CB ¨>± PATIENT ²S¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²S¦�äª� !Q¸®�32 ¯/¯� § NB ¨M± ADDRESSEE ²7¦�4ê/5°#®Yµ6! ¯O¯v. (93), ¤ ¥ ¦/§ CB ¨M¦l©ÿª$ 1��,º«&¯¬ª § CB ¨M± ACTOR ²7¦��¬ª��S¯ª°§ CB ¨>± PATIENT ²7¦�-1ª8.�·1·�01¯��ߧ CB ¨M± PATIENT ²7§ NB ¨>± APPURTENANCE ²7¦� ª�*!Q¸®Y32�¯ª § NB ¨M± ADDRESSEE ²7¦�4ôª�5°&®Yµ6! ¯O¯�
Moreover, we can combine examples like (95) and (97) to form (98). Also
(99) canbe analyzed,in a straightforward way. Note that informationpackaging
doesnot seem to be ableto analyze (98). It is not entirely clearwhatSteedman’s
treatmentof (98) would belike.
(98) English
(Which teacherdid you give whatbook?)
I gave thebookon SYNTAX to theteacher of ENGLISH.
(99) i. ¤ ¥ ¦O§ CB ¨M¦l©ÿª$ +��,º«�¯ ª § CB ¨M± ACTOR ²7¦��¶ª��S¯ª § CB ¨>± PATIENT ²S¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²S¦�äª� !Q¸®�32 ¯/¯ª°§ CB ¨>± ADDRESSEE ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦���ª�9Þ¸� 1:����µ�¯O¯/¯ii. (87), ¤ ¥ ¦O§ CB ¨M¦l©ÿª$ +��,º«�¯7�í§ CB ¨M± ACTOR ²7¦��¶ª$�S¯ª § CB ¨>± PATIENT ²S¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²S¦�äª� !Q¸®�32 ¯/¯ª°§ CB ¨>± ADDRESSEE ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦���ª�9Þ¸� 1:����µ�¯O¯/¯iii. (89), ¤ ¥ ¦/§ CB ¨>¦T© ª$ 1��,º«&¯¬ª § CB ¨>± ACTOR ²S¦��¬ª��S¯� § CB ¨>± PATIENT ²7¦�-òª/.�·Q·10 ª § NB ¨M± APPURTENANCE ²7¦�àª$*!Q¸®Y�2�¯O¯ª°§ CB ¨>± ADDRESSEE ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦���ª�9Þ¸� 1:����µ�¯O¯/¯
Theoriesof InformationStructure /73
iv. (89), ¤ ¥ ¦O§ CB ¨>¦T© ª$ +��,!«�¯ ª § CB ¨>± ACTOR ²7¦��¬ª/�Y¯ª § CB ¨M± PATIENT ²7¦�-°ª�.�·Q·�0 ª § NB ¨>± APPURTENANCE ²S¦�äª$*!Q¸®Y32�¯/¯�ߧ CB ¨M± ADDRESSEE ²7¦��!ªu®Y«���Eµ «;��ª°§ NB ¨>± APPURTENANCE ²S¦�� ª%9Þ¸� +:���Yµ1¯/¯O¯v. (89), ¤ ¥ ¦O§ CB ¨>¦T© ª$ +��,!«�¯ ª § CB ¨>± ACTOR ²7¦��¬ª/�Y¯ª § CB ¨M± PATIENT ²7¦�-°ª�.�·Q·�0 ª § NB ¨>± APPURTENANCE ²S¦�äª$*!Q¸®Y32�¯/¯§ CB ¨>± ADDRESSEE ²S¦��ºªu®�«���Eµ�«���ª°§ NB ¨M± APPURTENANCE ²7¦�� ª%9�¸� 1:����µ1¯O¯'�)(Ô¯vi. (93), ¤ ¥º¦O§ CB ¨>¦T©qª% 1��,º«&¯ºªò§ CB ¨>± ACTOR ²S¦���ª��S¯!ª°§ CB ¨M± PATIENT ²7¦�-1ª8.�·1·�0Q¯§ CB ¨>± ADDRESSEE ²S¦���ªä®�«���Eµ�«�� ª § NB ¨>± APPURTENANCE ²7¦���ª<9Þ¸� +:���Yµ�¯O¯�í§ CB ¨M± PATIENT ²S§ NB ¨M± APPURTENANCE ²S¦�Úª$ !Q¸®�32�¯/¯vii. (93), ¤ ¥º¦O§ CB ¨>¦T©¶ª� 1��,º«�¯ºªñ§ CB ¨>± ACTOR ²S¦���ª��S¯ª°§ CB ¨>± PATIENT ²7¦�-1ª8.�·1·�01¯§ CB ¨>± ADDRESSEE ²S¦��ºªu®�«���Eµ�«���¯��ߧ CB ¨M± PATIENT ²7§ NB ¨>± APPURTENANCE ²7¦� ª� !Q¸®�32�¯ª § CB ¨M± ADDRESSEE ²7§ NB ¨>± APPURTENANCE ²7¦��ñª�9�¸� 1:����µ1¯O¯�
To recapitulate, the topic/focus structureswe obtain in DGL are -still- rela-
tional structures. Nominals ensure that dependents and heads remainproperly
linked - which is exactly theheartof theproblemin atypedapproachlike (Kruijf f-
Korbayova, 1998) whenwe get to very complex structureslike (99). Like in TF-
DRT, though,weconnect thesentence’s topic andfocususingthe À -operator. Fol-
lowing dynamicapproachesto interpretationof information structure(likeKruijf f-
Korbayova’s,Peregrin’s or Steedman’s), the À -operatorcontrols how thesentence
is interpretedin context givenits informationstructure.Givenahybrid logical for-
mulaof theform Ç ÀèÉ , we interprettheformulaby first evaluating Ç against the
current (discourse)model.Only if Ç canbeinterpreted, we interpret É . Chapter6
discussesthis in moredetail, providingmodel-theoretic interpretationsof CB,CB*,
NB, andNB* andthedescribeddynamiceffect of À . �Remark 9 (The relation to FGD). In thelight of theabove discussion, how does
DGL’s account of informationstructure relateto FGD’s topic-focus articulation?
Like I already pointed out at various points in the discussion, the account given
herestaysclose to FGD, elaborating it whereneeded. The main differencewith
FGD is that in DGL the nodes in a linguistic meaningarenot ordered according
to communicative dynamism. Sgall et al (1986)(p.220ff) discusstheir interpreta-
tion of Firbas’s notion of communicative dynamism, andargue for the semantic
relevanceof communicative dynamism. Communicative dynamism canberelated
to degreesof salience, and is thusarguably relevant for contextual interpretation
in general the interpretation of quantifier scopein particular (cf. also(Hajicova et
74Ä Theoriesof InformationStructure
al., 1998)(pp.158-159)). BecauseI dealneitherwith quantifiers nor focalizers,and
becausethe notion of communicative dynamism is still in needof further devel-
opment (Hajicova et al., 1998), I have not includedit in DGL. At the sametime,
that is not to say that DGL could not provide the basis for a formal account of
communicative dynamism.
Theideaof communicativedynamismasanordering over thenodesin arepre-
sentationof sentential linguistic meaning could beincorporated in DGL alongthe
following lines. In theeliminationof aslashor aproduct,a Ø is introducedinto the
linguisticmeaning to combinethefunction andtheargument.Wecanfirst of all re-
fine this by saying thattheelimination ofû �>=�? ¢ =�?*@;= ü introducesaconnective Ø = .
Furthermore, following Sgall et al’s proposals,we canrelatesurface word order
to the ordering of nodes in theunderlying linguistic meaning. This we caneasily
obtain, by first letting theunderlying order mirror finally obtainedorder, andthen
ensuring thatcliti csandtopical/focal embeddedclausesareproperly ordered.The
latter orderingwecanachievebecauseconstructionsinvolving cliti csor embedded
clausesareindicatedby theuseof particular modes,andthesemodesarereflected
on the Ø ’s.�SUMMARY
In thischapter I discussedFGD’s theory of topic-focusarticulation, Vallduvı’s information
packaging, andSteedman’s Theme/Rheme-basedtheory. All thesetheorieshave in com-
monthat they describe informationstructurein termsof its realization(“syntax”) aswell
asits interpretation(“semantics”)– contrary to many otherapproachesthatconsiderjust
oneor the other. In reflectionon thesetheories,I notedseveral problems. I argued that
informationpackagingis mistakenin its conflationof thematicstructure andinformation
structure,showing examplesthat it cannotsatisfactorily explain. Furthermore,its charac-
terizationof theprimary notions of ground andfocusaspartitionsof a sentence’s surface
form leadsto problemswith recursivity, andappears at odds with thegenerally accepted
ideaof informationstructurebeinganaspectof linguistic meaning. Finally, its relationto
a concretegrammar framework is underdeveloped.It is not clearhow theGB architecture
of (Vallduvı, 1990) or HPSG(EngdahlandVallduvı, 1994) couldbeextendedto explain
wordorder, tuneandtheir interaction asmeansto realizeto informationstructure.
For FGD andCCGweobserved thattheirnotionsof informationstructureareclosely
related.However, they differ substantiallyin their viewsof grammar. FGD adopts a trans-
formationalapproachto explainhow informationstructureactsasaparameterdetermining
wordorder andintonation.CCGis amonostratalformalismin whichsufraceform andun-
Theoriesof InformationStructure /75
derlying meaning (with information structure)are compositionally related. For FGD I
notedthat a transformationalaccount cannot give a principled account of how different
strategies(like tune,word order, morphology) can interact to realizeinformationstruc-
ture.CCGhasbeenextendedto cover tuneandvariability in wordorder, but I arguedthat
CCG’s Principleof Adjacency seemsto necessitatea formal dissociationof the descrip-
tionsof word orderandof informationstructure. This breaks with the general linguistic
intuition of wordorderasa structural indicationof informativity.
Alik e CCG,DGL is a monostratal,compositionalapproach. In DGL we operateon
multidimensionalsignsthatrepresentdifferentlevelsof linguistic information,andthereis
noproblemin lettingdifferentlevelsinteractsimultaneously(like in atransformationalap-
proach).LikeFGD,I considerinformationstructure asanimportant factorin determining
surfacerealization, andI arguedhow we canformalize thatview in DGL’s parametrized
setting(usingmodesandstructural rules). I endedthechapterwith discussinghow infor-
mationstructureis representedat the level of linguistic meaningin DGL. Basedon the
proposalsof (Sgallet al., 1986; Heycock, 1993; Hajicova et al., 1998; Steedman, 2000c)
I considera moderate form of recursivity of information structure.I explained how that
enablesusto cover complex examplesinvolving double foci or embeddedfoci which e.g.
informationpackaging is unable to explain.
76Ä Theoriesof InformationStructure
CHAPTER 3
THE CATEGORY OF INFORMATIVITY
Information structurecanbe realizedusingvariousmeans- but whenandwhy can(or does)a
languageavail itself of thesemeans?In thischapterI discussabasictypological characterization
of whenlanguagesusevariability in word orderor tuneasstrategiesto realize informativity, i.e.
indications of contextual boundness. The characterization is basedon empirical datafrom a
variety of typologically differentlanguages,anda new typology of variabili ty in word order.
Theresultsof this chapterareasetof typological hypothesespredicting whethera languagehas
rigid, mixed,or freeword order,aninformativity markednessprinciple,anda setof hypotheses
thatpredict whenlanguagesusewordorder, tuneor acombinationthereof to realize information
structure. Thesetwo setsof hypothesesform thetypologicalbasisfor thegrammararchitectures
to bepresented in thenext two chapters.
Thedoctrineseemsto bethatwe derive aesthetic
pleasurein comprehendingsomethingasa unified
structure,in finding thata complex of disparate
phenomenacanbeexperiencedasa unifiedwhole.
– ChristopherHookway
3.1 INTRODUCTION
Thegoal in this chapter is to formulateasmallsetof hypothesesthatpredict when
a languagemightuseparticular strategiesto realizeinformationstructure,thustry-
ing to characterizecontextualboundnessasatypological categoryof informativity.
Thesehypothesesarebasedon empirical datafrom a variety of typologically dif-
ferentlanguages,andtake tuneandvariability in word order into account.1 These
hypothesesform thetypological basisfor thegrammararchitecturesto bepresented
in thenext two chapters.
In thischapterwebegin by discussingatypological perspectiveonwordorder.
We start with basicor dominant word order in A 3.2. Thereare two reasons for
doing that. Firstly, variability in word order is variation on dominant word order.1We areawareof thefactthattherearemoremeansthanvariability in word orderandtune.The
hypotheseswe presenthereprovide a basis. We do not claim they arecomplete- they needto betestedonalargeramountof data,andelaboratedwhereneededto covermoreof themeansmentionedin theintroductionto Chapter2.
77
78Ä Thecategory of informativity
Secondly, in a Lambek-stylecategorial grammar,any account of word orderstarts
in thelexicon with assigningcategoriesthatmodeldominant wordorder.2 Because
work in typology hasmainly focusedon dominant word order, like Greenberg
(1966) or Hawkins (1983), we canreadily make useof their findings to formulate
a cross-linguistic procedure to construct lexical categories for basicword classes.
In other words, the typological perspective on word orderstarts in the lexicon, as
onewould expect in a categorial approach.
Subsequently, we turn to variability in word order in A 3.3. Naturally, to be
able to predict when a languagecan usevariability in word order to realize in-
formation structure, we need to know to what extent that languagecan vary its
word order. Unfortunately, thereappears to be no typological account explaining
whena languagecandisplay a particular degreeof variability in word order, cf.
(Croft, 1990;Lehmann, 1993; RamatandRamat,1998). Steele(1978) discusses
different degreesof word order (rigid, mixed and free) but providesno typolog-
ical hierarchy on which one degreeor another would be implied by somebasic
facts about the language. Skalicka’s work does discussa typological characteri-
zation but only distinguishesbetween whatSteelewould termrigid andfreeword
order, cf. (Skalicka andSgall,1994;Sgall,1995). Herewe combine Steele’s char-
acterization, Skalicka’s typology of languages, and observations on a variety of
typologically different languagesto construct an initial proposalfor a typological
characterizationof rigid, mixed,andfreeword ordervariability . We needat least
a three-way distinction of variability to be able to explain the different levels of
interaction betweentune andword order to realize information structure, as ob-
servablein languages like English (rigid), Dutch andGerman(mixed), or Czech
andTurkish(free).
Finally, we proposeseveral hypothesesthat predict, for a languageof a given
type, whatstrategiesit will useto realize information structure. Thestrategieswe
discussherearebased on word order, tuneandtheir interaction. The hypotheses
elaboratevariouspredictionsadvancedby Sgalletal. (1986), andareillustrated on
a numberof typologically different languages.Following practice in languagety-
pology, we conceive of these hypothesesas(initial) explanations of how strategies
like word order or tunerealize a category of informativity. In the next chapters,
we formulate grammararchitecturesthat modelthese strategies– the hypotheses
formulatedherecontrol theaccessibility of the rule packagesmodeling particular
2Unlike thecombinatory tradition,wherelexical categoriesnot only modeldominant word orderbut alsopossiblevariability (Hoffman,1995a;Baldridge,1998).
Thecategory of informativity /79
strategies.
3.2 BASIC WORD ORDER
Thegoalof thissectionis to constructarepresentationatypological modelof basic
word order in DGL. By “basic word order” we understand theplacementof mod-
ifiers relative to their heads - subjectandobject asmodifiersof a verbalhead, and
mostnominal headmodifiers. For example, if we look at theplacementof subject,
object, andverbin variouslanguages,we canobserve distinct orders. Threecom-
monly found orders areSOV, SVO, andVSO, illustrated here in examples (100)
through (102) respectively.3
(100) SOV (e.g.Japanese)
Taroo-gaTaroo-NOM
ringo-oapple-ACC
tabetaate
“Tarooat anapple.”
(101) SVO (e.g.English)
Elijah reada book
(102) VSO(e.g.Welsh)
Lladdoddkill ed
ythe
ddraigdragon
ythe
dynman
“The dragon killed theman.” (Comrie,(Hawkins,1983)(p.1))
Besidesadifferentiation in how a languageordersaverbandits complements,
we canalsoobserve variation in otherorderings. For example,whereasJapanese
usespostpositionslike kooen-made(English “ to thepark”), Englishuses preposi-
tions.4 Similarly, languagesvary in wherethey placenominal modifierslikeadjec-3Thenotionsof ‘Subject’and‘object’ asgrammaticalrolesneedto bespecified,astheiruseis oc-
casionallyconfusing. Here,weadhereto theunderstandingproposedin Manning (1996)andKroeger(1993),goingbackto Dixons’ notionof “pivot”. Both ManningandKroegerusea characterizationthat appliesto ergative languagesaswell, andargue that -even there-we canbroadly understandthenominative verbalargumentto be thesubject.Kroeger discussesseveral teststhatconfirmthis,thesetestsbeingapplicableonly to nominative arguments:Raising,ConjunctionReduction,Pos-sessorAscension,secondary predication,obviation, andnumber agreement (1993)(p.55). For morediscussion,see(Manning, 1996)and(Kroeger, 1993).
4Languagesneednot strictly useeitherprepositionsor postpositions.For example,Dutch andGermanuseboth pre- andpostpositions:German dasHausgegenuber (English “oppositeto thehouse”),versusGerman auf demHaus (English “on the house”). If a language allows for twootherwise“opposite” orderings,it canbe saidto be doubling (Hawkins, 1983). Doubling is oftenunderstoodin termsof languagechange,with onestrategy becomingoutdatedandmakingplacefora new, dominantstrategy.
80Ä Thecategory of informativity
tivals, genitives,or relativeclauses.As Hawkinsremarksin (1983)(p.2),despite all
this variation clearpatternscanbe discerned.Nineteen- andearly-twentieth cen-
tury Germanscholarswerepossibly the first to draw attention to them(Hawkins;
cf. also(Sgall, 1995)(p.52)), andthe work by for exampleGreenberg, Lehmann,
Venneman, andHawkins canbeseenasa continuation of their work.
In thenext sectionwepresent thetwo hierarchiesthatHawkinsproposesto ex-
plain thebasicword order of nounsandtheir modifiers. Thesehierarchiesemploy
standardconnectivesfrom propositional logic to combine implicational universals
into aconcisestatement of theinterrelationsbetweendifferentfactorsdetermining
basic word order. Thenoun/modifierhierarchiescanbecombined with universals
describing thebasic wordorderof theverbandits complements(primarily, subject
and object), to cover the patterns leading to the 24 languagetypes proposedby
Greenberg andfurther exploredby Hawkins andcolleagues.
3.2.1 HAWKINS’ TYPOLOGY OF BASIC WORD ORDER
Hawkins (1983) advancesa typological modelof basic word order that is based
on the rich setof Greenberg’s universalsandwhich sharesVenneman’s concern
with thehead-dependentasymmetry.5 ThereareseveralpointsonwhichHawkins’
modeldistinguishesitself, though.
For one, Hawkins argues that statistical implicational universals should be
avoided,asthey are“theoretically undesirable” (p.60). Instead,nonstatistical im-
plicational universalsshould be used,which areexceptionless. To showthat one
actually can construct an account of basic word order in termsof implicational
universals,Hawkinsbuildshisonavery largecollectionof data. Heusesasastart-
ing point Greenberg’s 30-language sample, andthe sampleof 142 languagesthat
Greenberg usedfor certain (limited) co-occurrencesof basicword order. Hawkins
(andcolleagues) extended the second sample to cover about 350 languages,and
his typology of basic word orderis basedon thedatapresentedby thatsample.
Anotherpoint thatdistinguishesHawkins from GreenbergandVennemancon-
cerns the notion of “word order type”. In Hawkins’ typology, the notion of a
word order type no longer meansa uniform linearization for all different kinds
of head/dependent constructions onemight distinguish in a language. Instead, a
“word order type” is definedas a specificpattern of co-occurrence possibilities
5Greenberg (1966) alsonotesthat the distinctionbetweenheadsandmodifiersis importanttofind an answerto why languagesselectoneparticularbasicword orderover another, but doesnotaddresstheissuein detail.
Thecategory of informativity /81
permitted by theimplicational universalsthatHawkins defines.Eachof thesepat-
ternscontains a commonshared property, like ‘prepositions’, functioning as the
typologicalindicator(1983)(pp.114-115). Wordordertypesarethusno longer tied
to either XV or VX, since correlations to only these word order patterns do not
alwaysgiveriseto explanationswhyotherregularitiesdooccur. Ratherthantaking
XV or VX (or Greenberg’s VOS,SVO, SOV), theentire patterndefinesthe word
ordertype. And, becausethepattern is considered asa whole,it is possibleto say
whatparts, whatparticular co-occurrences, make it unique. Obviously, this leads
to moreprecise generalizations.
Against this background6, Hawkins presents a set of basic word order uni-
versalsthat extends Greenberg’s original classification. From theseuniversals,
Hawkinsderivestwo hierarchiesfor nounmodifiers- thePrepositionalNounMod-
ifier Hierarchy (103),andthe Postpositional NounModifier Hierarchy (104). ( Bindicatesthenominal head, C theadjective, D thegenitive.)
(103) Universal XIV, Prepositional NounModifier Hierarchy
Prep E ((NDem F NNum E NA) & (NA E NG) & (NG E NRel))
(104) Universal XVIII, Postpositional NounModifier Hierarchy
Postp G ((AN H RelN G DemN& NumN)& (DemN H NumN G GN))
Thesehierarchies can be combined with universalson the relation between
noun-modifier word order and the basic word order of the verb and its comple-
ments(primarily, subjectandobject) to explain thepatternsleading to the24 lan-
guagetypesproposedby Greenberg (and further explored by Hawkins and col-
leagues). We will not discussthe full setof basicword order types discussedin
(Greenberg, 1966; Hawkins, 1983). Instead,we briefly point out how we canuse
thesebasicword order types to construct lexical categories. Thatway, westartour
cross-linguistic account of word orderalready in the lexicon, asappropriate for a
lexicalizedapproachlike categorial grammar.
3.2.2 A TYPOLOGICAL MODEL OF BASIC WORD ORDER IN DGL
Theaim in the current section is to relate Hawkins’ typology of basicword order
to a grammar framework like DGL. BecauseDGL is a categorial grammar,this
6For completeness,we shouldalsomentionthat Hawkins introducestwo competingprinciplesto explain the basicword orderof nounsand their modifiers. Theseprinciplesare the HeavinessHierarchyandtheMobility Principle.As thediscussionof theseprinciplesis not directly relevanttoour argumenthere,we refertheinterestedreaderto (Hawkins,1983).
82I Thecategory of informativity
is relatively simple: The ordering given by the typology translatesmore or less
directly into the directionality of the slashesthat we usein categories assigned to
differentword classesin thelexicon.
Theapproachis simple,but notsimplistic: It is simplebecauseof thecategorial
nature of the approach. The approachis far from simplistic since in a categorial
grammarthe grammar is for a fundamentalpart constitutedby the lexicon. The
predictionsthatthetypologymakescanthusbecouchedin termsof cross-linguistic
(architecturesfor) grammarfragments. In relation to thediscussion in Chapter1,
we extendthemapping S in DGL’s linking theory, Definition 4 on page27.7
To reflectthedistinctions that Hawkins’ typology of basic word ordermakes,
wedistinguishdifferent modes. Therelevant modesaregivenin (105-107).Modes
may be headed, as indicated. As we explained in Chapter 1, following (Hep-
ple, 1997), a headed modeexplicitly indicateswherethe head of the construc-
tion is. Thus, in CKJMLONQP mode R is a headed mode,with the arrow S point-
ing to the dependent P away from the head C . Logically, this meansthat a
headed mode R hasassociated to it two products: TUR VXW�JZY[L]\*^�Y[L�\*_�Y[L�`and RaSbVcW�J LON \*^ LMN \*_ LON ` .
(105) VERBAL MODIFIERS:Name Form App. Description
Subject dfe Verb Marksthesubjectposition.
Directcomplement ghe Verb Marksthedirectcomplementposition.
Indirectcomplement i�ghe Verb Markstheindirectcomplementposition.
Complement e Verb,Noun Compositionwith acomplement.
Temporaladjunct j�kml Verb,Noun Compositionwith a temp.adj.
Spatialadjunct don�l Verb,Noun Compositionwith aspatialadj.,headed
(106) NOMINAL MODIFIERS:
7Again, theproposalheredoesnotpretendto beempiricallycomplete.Thepoint hereis to showthat we canusethe findingsof language typology in a (categorial) grammarframework – in spiritsimilarto (Venneman,1977), but significantlyimproving onhisproposalby makinguseof (Hawkins,1983). Becauseweareinterestedin therealizationof informationstructure,wepaymoreattentiontovariability in wordorderratherthanbasicwordorder.Thediscussionof basicwordorderis providedto starta cross-linguisticdiscussionof word orderin (for categorial grammar)theproperplace– thelexicon.
Thecategory of informativity /83
Name Form App. Description
Adjectival lpg�q Adjective Compositionof nounwith adjective.
Genitival rtsvu Noun Compositionof nounwith genitival.
Complement e Verb,Noun Compositionwith a complement.
Temporaladjunct j�kml Verb,Noun Compositionwith a temp.adj.
Spatialadjunct don�l Verb,Noun Compositionwith a spatialadj.,headed
Demonstrative gpsvk Noun Compositionof nounwith demonstrative,headed
Article ltwvj Noun Compositionof nounwith article.
(107) ADJECTIVAL /ADVERBIAL MODIFIERS:Name Form App. Description
Adjectival lpg�q Adjective Compositionof nounwith adjective.
Adverbial lpghx Adverbial Marksconstructionwith adverbial.
In (108)wedefinethecategoriesfor transitiveverbs,for activevoice.Weleaveout
intransitive andditransitive verbs, asthecategoriesfor theseverbscanbeimmedi-
atelyderivedfrom thespecifications in (108).
(108) VERB, SUBJECT, OBJECT (ACTIVE VOICE):
Verbcategory =
yzz{ zz|}�~�� �����3�>��� I��6� �Z}�~h�v�����7���t������}�~v�*�f���7�����O�}�~�� �����3�>�����p������}�~ � ���o�#��� � ����}�~"� ���o�#�X�b���}�~�� �����3�>��� I�� ���f}�~��*�f���#� I���� �f}�~"�v�����7�������
Next, (111) and(112) specify thecategoriesfor adjectival andgeneral geniti-
val modifiersof nominal heads, respectively. Thegenitival categoriesaregeneral
in thatthey only hold for whatcould becalledbare genitival structures,beingcon-
structions that arenot construedusinga function word. An exampleof what we
understandby suchbareconstructions are given in (109), with “non-bare” con-
structionsillustratedin (110).
(109) English
(Kathy’s�>���v���v [¡*�>¢ book
(110) English
thebook (of thelecturer� ¡ �>¢(111) ADJECTIVE, NOMINAL HEAD MODIFICATION:
Adjectivecategory =
y{ | ���%� � � �¤£ �7�¦¥§���� I � �¤£v���#�¨�©¥(112) GENITIVE, NOMINAL HEAD MODIFICATION:
Genitivecategory =
y{ | ����� �1ªv«�¬ �7�¦����� I ªv«�¬ � �7�¨�©
84I Thecategory of informativity
(113) DEMONSTRATIVE, NOMINAL HEAD MODIFICATION:
Demonstrativecategory =
y{ | ���%� � � «�® �7�¦¯%°�±]���� I � «�® ���7�¨�©¯%°�±(114) Hypothesis:Prep & DemN E ArtN
(115) ARTICLE, NOMINAL HEAD MODIFICATION:
Article category =
yzzzzz{ zzzzz|�o}�~"� «>² �%� � ��³���}�~h� « ���7� ¯�°"´+µo¶�µo·�°�¸�¹"·�µ�ºv»�°�¼Z¯%°�±]��o}�~"� «>² � I ��³��o��}�~"� « ���7� ¯�°"´+µo¶�µo·�°�¸�¹"·�µ�ºv»�°�¼Z�©¯%°�±�o}�~�½ ¬ � «�² �%� � ��³���}�~"� « ���7�¿¾;¶[À�°"´+µo¶�µo·�°�¸�¹"·�µ�ºv»�°�¼Z¯%°�±]��o}�~�½ ¬ � «�² � I ��³��o��}�~�� « ���7�¨¾;¶[À�°"´+µo¶�µo·�°�¸�¹"·�µ�ºv»�°�¼Z�©¯%°�±
In the next section I look at variability in basicword order- thus,what possi-
bilit ies areavailable in a languageto alter the dominant word order specified by
lexical categories.
3.3 VARIABIL ITY IN (BASIC) WORD ORDERING
In this section a preliminary account of variability in word orderis presented.Us-
ing a datasample of 22 languages,I try to establish hypothesesthat take theform
“If a languageL hascharacteristics C,C’,... then it hasa rigid/mixed/freeword
order.” Becausethedatasampleis rather small,I donotclaimthehypothesesto be
anything morethan just that- hypothetical explanationsthatarehopefully verified
(with minimal adaptation) in thelong run.
Thetypological literatureis rather sparseonaccountsof whylanguagesvary in
word order flexibil ity. In theliterature(e.g. (Croft, 1990)), Steele’s (1978) is cited
asthe referenceon variation in word order,focusing on word orderof the matrix
clause.8 Steeleproposesa distinctionof threedegrees of word order freedom,be-
ing rigid, mixed and free, but does not present a typological characterization of
whenoneof these degreesis available in a language.Within thePrague School of
Linguistics,Skalicka’s account of languagetypes (cf. Skalicka andSgall’s (1994),
Sgall’s (1995) for a recent formulation) discussesthe relation between morphol-
ogy andvariability in word order in moredetail than found elsewhere, but only
considerstheopposition betweenrigid andfreeword order.
The account I present is based on Steele’s characterization of variation (i.e.
whena language’s word order is rigid, mixed,or free)and,for an important part,8As we alreadysaw above, Greenberg (1966)andHawkins (1983) focusratheron basicword
orderratherthanon variation. Most later typologicaldiscussionsdo not discussvariationof wordorderin any deptheither- cf. (RamatandRamat,1998), (Lehmann,1993).
Thecategory of informativity /85
on Skalicka’s insights. Particularly, the initi al datagathering wasdonewith the
following null hypothesisin mind:
(116) WORD ORDER NULL HYPOTHESIS: Themorea languageallowsfor the
foll owing phenomenato occur grammatically, the higher the likelihood
that the languagehasa relatively free word order: frequent useof null
anaphora, lack of expletive pronouns,a rich casesystem,complex verbal
morphology.
This null hypothesisfoll ows from Skalicka’s work on languagetypes, Sgall
et al.’s commentsin (1986), andfrom work by Hale andby Speasasreferred to
in (Kroeger, 1993)(p.113). The hypothesesI formulateon the basisof the data
work out the null hypothesisin moredetail. The intention with thesehypotheses
is to cometo a characterizationof variability in word orderon thebasisof formal
aspects of a language. At leastto the extent allowed by the relatively small and
eclectic samplewe present here, the hypothesespurport to explain why eachof
theselanguagesdisplaysa particular (in)flexibility in word order.9
For 20 languages I gathered dataabout the following aspects.10 The table in
Figure3.1(page 87) presents thedata.
Word order type: TheGreenberg/Hawkinscharacterizationof thelanguagein
termsof therelativeordering of verb,subjectandobject; genitives(G) andnominal
heads(N); adjectives (A) and nominal heads(N); and whetherthe language is
prepositional (Pr) or postpositional (Po).
Variation: The characterization of a language’s word order (matrix clauses,
dependentclauses)asrigid, mixedor free. To determinevariation,we useSteele’s
proposal as in (1978): Of the constraintsgiven below, if a languagebreaks con-9A null hypothesislike (116) is by no meansuniversallyacceptedasa goodgroundfor trying to
explainwhenvariability in wordorderis possibleatall (andcouldthusbeusedto realizeinformationstructure).For example,Steeleseeksto explicitly rebuke suchnull hypotheses,andtriesto correlateword order freedomwith person-agreementbetweenthe subjectand the verb. However, it is notclearin how far this would besupportedby ergative languages. Moreover, if we want to extent theaccountof word orderfreedomto embedded clauses,Steele’s suggestionis falsifiedby Turkish. As(Hoffman,1995a)pointsout, thereis no agreementbetweentheverbandits subjectof anembeddedclause.
10In alphabetical order, the datacomprisestheselanguages: Biblical Hebrew (Ofir Zussman),BrazilianPortuguese(FernandaAranha,JasonBaldridge),Czech(IvanaKruijf f-Korbayova), Dutch(author),English(author, Mark Steedman),French(author, Mark Steedman),German(author, Ju-lia Hockenmaier),Modern Greek(Nikiforos Karamanis),Modern Hebrew (Ofir Zussman,NissimFrancez,Shuly Winter), Hindi (Shravan Vasishth),Italian (Malvina Nissim),Japanese(TomotsugoKondo, Shravan Vasishth),Korean(Kihwang Lee), Mandarin(Julia Hockenmaier), Russian(So-fya Malamud),Swedish(ElisabetEngdahl,NataliaModjeska-Nygren),Tagalog(JasonBaldridge),Turkish(Hoffman,1995a).
86I Thecategory of informativity
straint A (andhenceA’), andB, thenits word order is free,whereasits word order
is rigid if nonearebroken. If somebut notall constraintsarebroken,wordorder is
mixed.
A. A variation on thebasicword order in which theverboccurs in other thanits
position in thebasicword order is to beavoided.11
A’. A variation on the basicword order in which the verb occurs either initial
or final to the clause is to be avoided, if the verb wasneither initi al nor final
respectively in thebasic order.
B. A variation on thebasic word orderin which theobjectprecedestheverband
thesubject follows theverbis to beavoided.
Casestrategies: Following Croft’s discussion of morphological strategies in
(1990) (cf. also Chapter 1), a specification of the strategies usedfor grammati-
cal roles (primarily, subject anddirect object) andfor nominal modifiers. In the
table in Figure 3.1, the columns5-7 labelled Case, Pos, Tune relate to morpho-
logical strategiesfor verbal arguments,andcolumns8-17relate to morphological
strategiesfor nominalmodifiers.
Article s,expletives,andpro-drop: Inspiredby Haleandby Speas(cf. (Kroeger,
1993)(p.113)), Skalicka (Skalicka and Sgall, 1994), and Sgall et al.’s remarks
in (1986), we checkwhether a languagehasboth definite and indefinite articles,
whether it hasexpletives, and whether it allows the subject (and possibly other
modifiers) to be dropped. Thebasic hypothesisis that the absence of articlesand
expletives, and the possibilit y to drop subjects, are all indicative of a rich case
system,which usually enablesa relatively freeword order, cf. (Sgall,1995).
Verbal morphology: Skalicka’s languagetypesrelatea rich verbal morphol-
ogy to freerwordorder. Here,weindicatewhataspectsof verbal morphology (like
tense,aspect, modality, etc.) aremarked on theverbitself. Theremaining aspects
areusually realized usingauxiliaries(if atall present; e.g.Tagalogappearsto miss
tense,cf. (Kroeger, 1993)).
Agreement: In keeping with Steele’s suggestionthat thereis a relation be-
tweenfreeword orderandPerson-agreementbetween subject andverb,we check
whata verbcan(or does) agreewith in eachlanguage.
To explain the variation in word orderwe observed in the datain Figure3.1, we
formulatea setof variation hypotheses. The variation hypothesesareformulated11We understand“to be avoided” to meanthat if the orderwould be grammaticallypossible, it
would behighly marked.
Thecategory of informativity /87L
an
gu
ag
eW
Oty
pe
G/H
Var.
Ca
sePo
sTu
ne
Ca
seC
om
pJu
xA
gr
Su
pA
dp
Ln
kL
+A
dS
+A
dA
fxA
rt.
Exp
l.D
rop
Verb
.A
gr.
Bib
l.H
ebre
wV
SO
/Pr/N
G/N
A1
R/R
Acc
+-
--
-+
-+
--
-+
def
-+
PG
N/T
Abs
:STa
galo
gV
1/P
r/NG
/NA
1F
/?+
--
+-
--
--
+-
--
+-
+P
l/VA
Erg
:AV
1/P
r/NG
/AN
2A
bs:S
Fre
nch
SV
O/P
r/NG
/NA
9R
/R-
+-
--
-+
++
--
--
++
-P
GN
/TA
Abs
:SM
.H
ebre
wS
VO
/Pr/N
G/N
A9
R/R
Acc
+-
--
-+
-+
--
-+
def
-+
PG
N/T
Abs
:SIta
lian
SV
O/P
r/NG
/NA
9R
/R-
+-
--
-+
++
--
--
+-
+N
/TA
bs:S
Eng
lish
SV
O/P
r/NG
/AN
10R
/R-
+-
--
--
++
++
+-
++
-N
/Pst
Abs
:SS
VO
/Pr/G
N/A
N11
Abs
:SM
anda
rinS
VO
/Pr/G
N/A
N11
R/R
Acc
+-
--
--
-+
--
--
+-
--
-S
wed
ish
SV
O/P
r/GN
/AN
11R
(V2)
/R-
+-
--
-+
++
++
-+
++
-N
G/P
stA
bs:S
Dut
chS
VO
/Pr/N
G/A
N10
M(V
2)/R
-+
--
--
-+
++
-+
-+
+-
N/P
stA
bs:S
Ger
man
SV
O/P
r/NG
/AN
10M
(V2)
/M+
+-
+-
-+
++
+-
+-
++
-N
/Pst
Abs
:S
Inui
tE
rgA
bsV
/Po/
?F
/M+
--
+-
--
--
--
--
--
+P
N/T
ME
rg:A
GN
/AN
Abs
:S
Cze
chS
VO
/Pr/N
G/A
N10
F/F
+-
-+
--
+-
+-
--
--
-+
PN
G/T
AA
bs:S
Rus
sian
SV
O/P
r/NG
/AN
10F
/F+
--
+-
-+
-+
--
--
--
+P
NG
/TA
Abs
:SM
.G
reek
SV
O/P
r/NG
/NA
9F
/F+
--
+-
-+
--
--
--
+-
+N
P/P
stA
Abs
:SB
r.Por
tugu
ese
SV
O/P
r/NG
/NA
9F
/FA
gr+
+-
--
++
+-
-+
-+
-+
PG
N/T
AV
Abs
:SH
indi
SO
V/P
o/G
N/A
N23
F/F
+-
--
-+
++
--
--
--
-+
NG
/Fut
Erg
:AH
unga
rian
SO
V/P
o/G
N/A
N23
F/F
+-
--
++
+-
--
--
-+
-+
NP
/TM
Abs
:SJa
pane
seS
OV
/Po/
GN
/AN
23M
/R+
--
--
--
-+
--
-+
--
+T
(Hon
)K
orea
nS
OV
/Po/
GN
/AN
23F
/F+
--
++
--
--
--
--
--
+-/
TAM
Abs
:ST
urki
shS
OV
/Po/
GN
/AN
23F
/F+
--
+-
--
--
--
--
ind
-+
PN
/TV
MA
bs:S
Exp
lana
tion.
“WO
type
”sta
ndsf
orth
ela
ngua
ge’s
wor
dor
dert
ype,
aspe
rGre
enbe
rg/H
awki
ns,w
ith“G
/H”
the
rele
vant
inde
xin
Gre
enbe
rg’s
App
endi
xII.
The
“Var
”co
lum
nin
dica
tesv
aria
tion,
for
mat
rixcl
ause
/depe
nden
tcla
uses
;“F”
stan
dsfo
rfr
ee,“
M”
for
mix
ed,a
nd“R
”fo
rrig
id.
The
next
thre
ecol
umns
prov
ide
info
rmat
ion
onho
wsu
bjec
tsan
dob
ject
sare
dist
ingu
ishe
d:by
“Cas
e”(p
ossi
bly
just
byA
ccus
ative
mar
king
onth
eob
ject
),po
sitio
ning
(“P
os”)
,orb
yT
une.
The
colu
mns
ther
eafte
rrega
rdm
odifi
ersi
nge
neral
,and
the
mor
phol
ogic
alst
rateg
iesa
vaila
ble
ina
lang
uage
for
real
izin
gth
em(C
roft,
1990
):“C
ase”
case
mar
king,
“Com
p”co
mpo
undin
g,“J
ux”
juxt
apos
ition
,“A
gr”
agre
emen
t,“S
up”
supp
letio
n,“A
dp”
adpo
sitio
n,“L
nk”
linke
r,“L
+A
d”lin
ker+
adpo
sitio
n,“S
+A
d”su
pple
tion+
adpo
sitio
n,an
d“A
fx”
affix
atio
n.T
hene
xtth
reec
olum
nssp
ecify
whe
ther
the
lang
uage
has“
Art
”ar
ticle
s(or
just
one,
orno
ne),
“Exp
l”ex
plet
ives
,and
whe
ther
ital
low
sfo
r“D
rop”
drop
ping
(usu
ally
ofth
esu
bjec
t).T
hefin
altw
oco
lum
nssp
ecify
verb
alm
orph
olog
y.“V
erb.”
give
sm
ore
info
rmat
ion
onve
rbal
form
:whe
ther
ther
eis
any
indi
catio
nof
“P”
pers
on,“
N”
num
ber,
or“G
”ge
nder,
and
“T”
tens
e(po
ssib
lyon
ly“P
st”
past
,“Pre
s”pr
esen
t,or
“Fut
”fu
ture
),“A
”as
pect
,“V”
voic
e,or
“M”
mod
ality
.If
a“T
”,“M
”,“V
”or
“A”
isab
sent
itis
real
ized
usin
gan
auxi
liary
.The
last
colu
mn
indi
cate
swhe
ther
the
verb
agre
esw
ithan
yof
itsco
mple
men
ts,a
ndif
so,w
ithw
hat:
“Abs
:S”a
bsol
utive
subj
ect,“
Erg
:A”
erga
tive
Act
or.
Figure3.1: Word orderdata
88I Thecategory of informativity
like implicational universals,with theexception that they usetheconnective Á to
indicatethat the implication is hypothetical, andnot universal. Furthermore,we
definethe logical relations between free, mixed,andrigid word order asfollows: free à rigid F mixed,  mixed à rigid F free, and  rigid à mixed F free.12
To make the representation of variation hypothesesmore compact, we usea
few useful abbreviations. Strat(X,Y) indicatesthatstrategy Ä is usedto realize Å .
For example,Strat(SubjObj,Pos) meansthat positioning is usedasa strategy to
indicatethesubjectandtheobject of a verb. VerbForm(X) meansthat theverbal
form inflects for Å , Art(K) indicatesanarticle of type Æ , andAgr(X,Y) indicates
agreementbetween Å and Ä . Otherwise,we directly usethecharacteristics men-
tioned in the tablein Figure3.1. Negatinga characteristic meansit is not present
(-), e.g. Â Drop meansa languageis not pro-drop.
Then,on thebasisof thedatafor OV languagesin Figure3.1(i.e. Hindi, Hun-
garian, Japanese,Korean,andTurkish)we canformulate thevariation hypotheses
asin (117). Thesevariation hypothesesspecify whena languagewith OV ordering
(at someclauselevel) hasrigid, mixed,or freeword order.
(117) VARIATION HYPOTHESES FOR OV WORD ORDER
a. VARIATION HYPOTHESIS OV-1:
OV & Strat(SubjObj,Case) & ((Agr(ErgActor,Verb) & Drop) F(Agr(AbsSubj,Verb) & (VerbForm(Tense & (Aspect F Mood FVoice))))) Á free
b. VARIATION HYPOTHESIS OV-2:
OV & Strat(SubjObj,Case) & ( Â Agr(Subj,Verb) F Â Drop) Á mixed
c. VARIATION HYPOTHESIS OV-3:
OV & Â Strat(SubjObj,Case) & Â Drop & Â Agr(Subj,Verb) Á rigid
Looking at thedata,weseethat Hindi, Hungarian,KoreanandTurkishall have
a fairly rich verbal and nominal morphology. From Skalicka’s typology it then
follows that these languageshave a relatively freeword order,which they indeed
do. On theotherhand, Japanesedoeshave a nominal casesystem,but verbs only
inflect for tenseandhave no agreementwith thesubject.13 We understandthat its
mixedword order arisesfrom this combination of the presenceof a nominal case
system and the lack of a rich verbal morphology. This is confirmedby German12In the long run, we expect there to be more of a scaleof variability, rather than a discrete
tripartition into rigid, mixed,andfree.13Unlessoneconsidershonorificationasa kind of agreement.
Thecategory of informativity /89
dependentclausewordorder, which is mixed(aslong asthere is nomorphological
ambiguity; otherwiseit rigidifies).
Finally, the hypothesesalso explain the rigid ordering in Dutch dependent
clauses. The presenceof casemarking is significant to the explanation of why
Dutchdiffers from Germanwith respectto dependent clausewordorder. If it were
not, we couldpossibly obtain a mixedword orderby usingprepositions to realize
Case.However, astheexamplesin (118)illustrate, we cannot.
(118) a. Dutch
...
...dethe
manman
wienswhose
fotophoto
KathyKathy
aanto
ElijahElijah
gaf.gave.
“The manwhosephotoKathy gave to Elijah.”
b. * demanwiensfoto aanElijah Marie gaf.
(119) FURTHER HYPOTHESES ON THE BASIS OF OV DATA
a. ARTICLES:
OV & free Á Â (Art(def) Ç Art(indef)), or
OV & (Art(def) & Art(indef)) Á Â free
b. USE OF CASE:
Strat(NounModif,Case) Á Strat(SubjObj,Case)
c. EXPLETIVES:
Agr(AbsSubj,Verb) & (VerbForm(Tense & (Aspect F Mood F Voice)))Á Â Expl
d. PRO-DROP:
Agr(AbsSubj,Verb) & (VerbForm(Tense & (Aspect F Mood F Voice)))Á Drop
It is hardly surprising that for SVO word order we cangive hypothesesthat
aresimilar to (117). Using the datafor Czech,Dutch, English,French,German,
Hebrew, Italian, Mandarin, Brazilian Portuguese,andSwedish,we formulatethe
variation hypothesesin (120).
(120) VARIATION HYPOTHESES FOR SVO WORD ORDER
a. VARIATION HYPOTHESIS SVO-1:
SVO & Strat(SubjObj,Case) & (VerbForm(Tense & (Aspect FMood))) Á free
90I Thecategory of informativity
b. VARIATION HYPOTHESIS SVO-1’ :
VerbForm(Tense & Aspect & Voice & Mood) Á free
c. VARIATION HYPOTHESIS SVO-2:
SVO & Strat(SubjObj,Case) &
(VerbForm(Tense & Â (Aspect F Mood))) Á mixed
d. VARIATION HYPOTHESIS SVO-3:
SVO & Â Strat(SubjObj,Case) & (VerbForm(Tense & Â Mood)) Árigid
Again, thereis aninterestinginteractionbetween nominal andverbal morphol-
ogy. If a languagehasboth a rich verbalanda rich nominal morphology, thenit
hasfreewordorder– confirmingSkalicka’spredictions.Straight examplesof such
languagesin the dataareCzech,Greek,andRussian, with Brazilian Portuguese
presentingan interestingexception. Although native informants judge word order
in BrazilianPortugueseasfree(in thesenseusedhere), it only shareswith theother
freeSVO languages that it hasa rich verbal morphology. Thereis no rich nominal
morphology, andthe subject andobject areindicated using eitheragreement,po-
sitioning, or tune, rather thancase.ThevariationhypothesisSVO-1’ capturesthe
ideathat if we have a very rich verbal morphology in anSVO language,thenthis
sufficesto concludethelanguagehasfreeword order.
Mixedword order in SVO languagesseemsto beprimarily determinedby the
presenceof a rich nominal morphology, as in the caseof German. Verbalmor-
phology appears to belessimportant here. Frenchhasa richer verbal morphology
thanGerman,but lacksanominal casesystemand–accordingly– hasarigid rather
thanamixedwordorder. Observe thatthis providesaninterestingsimilarity to the
OV-case. The datashows that mixed word order is possible with a rich nominal
morphology, eventhough theverbmayonly show tense– cf. Japanese.
In general, rigid SVO word order appears to ensueassoonasthere is no rich
nominal morphology nor a rich verbalmorphology. Thereis at least onenotable
exception - Dutch.Dutchdoesnothaveacasesystem14 norarich verbalmorphol-
ogy, andyet it doeshaveamixedwordorderlikeGerman(andcontraryto English).
Thedatapresentedin thetablein Figure3.1shows usonly that,if we restrict our-
selvesto TypeX languages,thenwhatdistinguishesDutch (mixed) from English
(rigid) is theabsenceof theLinker+Adpositionstrategy. Thisstrategy for realizing
14Dutchdoesnot have a nominalcasesystemanymore – therearestill variousremnantsof casesthough.
Thecategory of informativity /91
nonpronominal possessorgenitival constructions is secondary in English (Croft,
1990)(p.34ff). However, it may indicatea positional fixation of what languages
with anominal casesystemwould simply usemorphology for. Thisview seemsto
besupportedby the typological characterization of English. Englishis both Type
X andType XI, due to its positional fixation of the different basic strategies for
realizing genitives(Linker: GN, Adposition: NG). Hence,I proposeto reformu-
latevariation hypothesisSVO-2 (120c) into (121) below, to cover (at least)Dutch
mixedword order. Thehypothesisexplains themixedword orderof Dutchon the
basisof Dutchbeing lesspositional thanEnglish(andthusmorelikeGerman,even
though Dutchhasno “overt” nominal casesystem).
(121) REVISED VARIATION HYPOTHESIS SVO-2:
SVO & (Strat(SubjObj,Case) F (NG & Â Strat(NomMod,L+Ad)) &
(VerbForm(Tense & Â (Aspect F Mood)))) Á mixed
(122) FURTHER HYPOTHESES ON THE BASIS OF SVO DATA
a. ARTICLES:
SVO & free Á Â (Art(def) Ç Art(indef)), or
SVO & (Art(def) & Art(indef)) Á Â free,
SVO & (rigid F mixed) Á (Art(def) & Art(indef))
b. USE OF CASE:
Strat(NounModif,Case) Á Strat(SubjObj,Case)
c. PRO-DROP:
free Á Drop, rigid F mixed Á Â Drop
Finally, for OV I lack sufficient datato cometo a proper characterization. To
cover thedatain Figure3.1,I proposethehypothesesin (123).
(123) VARIATION HYPOTHESES FOR VO WORD ORDER
a. VARIATION HYPOTHESIS VO-1:
VO & Strat(SubjObj,Case) & VerbForm(Aspect & Mood) Á free
b. VARIATION HYPOTHESIS VO-2:
VO & Â Strat(SubjObj,Case) & VerbForm(Tense & Â (Aspect &Mood))Á rigid
92I Thecategory of informativity
To recapitulate,I presentedSteele’s characterization of (in)variability in word or-
der as either rigid, mixed, or free. Basedon empirical datafrom typologically
different languages,I formulatedseveralhypothesesthat predict whena language
hasrigid, mixed,or freeword order(thusimproving on (Steele,1978)). Theem-
pirical findings underlying the hypothesesconfirmedto an important degreethe
predictions madeby Skalicka’s (morphological) typology of languages(Skalicka
andSgall, 1994; Sgall, 1995), in showing the interrelation betweenmorphology
andvariability . The proposal advancedheredistinguishesitself from Skalicka in
discerningmixed word orderbesidesrigid andfree word order, andcovering its
predictability. This finer distinction is not vacuous. Mixed word order languages
may realize information structure differently than either rigid or free languages.
Unlike rigid languages,mixedlanguagescanusewordorderto realize information
structure,but not to the degree that free languagescan. Oneconsequence is that
mixed languages have a morecomplex interaction betweentuneandword order
thaneither rigid or free languages. On the otherhand, whereasfree word order
languagesusually lack articles,mixed do have themandarethusableto indicate
contextual boundness thatway.
3.4 THE CATEGORY OF INFORMATIVITY
As I alreadynotedin the introduction to this chapter, we know from variouscon-
trastive studiesthat languagesmayrealizeinformationstructure in differentways.
For many differentframeworks it hasalsobeenarguedhow they areableto repre-
sentinformation structure,taking into account suchcross-linguistic variation. The
aimin thissection is to advanceseveralhypothesesthatpredict whenalanguageof
a given typeavails itself of particularstructural indicationsof informativity, elab-
orating on (Sgall et al., 1986). I restrict myself to the useof just word order and
tune, the relation betweenwhich I assumeto be oneof relative opposition: If a
languagedoesnot useword order, it uses tune.
Below I start in È 3.4.1with a null hypothesisderivedfrom (Sgallet al., 1986),
working towardsa setof hypothesesthat predict what a language’s ‘prefered’ or
canonical focus position is. By a language’s canonical focus position (or CFPfor
short) I understandtheposition in a sentencewherewe would expecttheinforma-
tion structure’s focus (focusproper) to berealized,given anunmarked,canonical
word order or an unmarked intonation pattern. Variousauthors have associated
theCFPwith sentence-finality, for SVO languageslike Czechor English (Sgallet
Thecategory of informativity /93
al., 1986; Vallduvı andEngdahl, 1996) but also for SOV languageslike Sinhala
or Tamil (Herring andPaolillo, 1995). Here, I examinedatafrom VX, XV and
SVO languages, andpresentin È 3.4.2asetof hypothesesthat effectuatessentence-
finality but doesnot by definition imply it.
Naturally, a focusneedsnot alwaysberealizedin thecanonical focusposition.
Theremaybevarious reasonsfor doing so.Thethematicstructureof a text andits
overallorganizationmayfor exampleplayarole,andanobviousfactor is theinfor-
mationstructureto berealized. For example, consider(124). Without any further
indications, people understand“Christopherreada book yesterday” to meanthat
yesterday is the focus proper, andthe focus may extendto any point leftwards.15
Theother sentencesrealizedifferentinformationstructures.In (124b-d) thewords
afterthepitch accent realize(part of) thetopic.
(124) English
Christopherreada bookyesterday.
a. Christopherreada book YESTERDAY.
b. ChristopherreadA BOOK yesterday.
c. ChristopherREAD a bookyesterday.
d. CHRISTOPHER reada book yesterday.
Thus, two questions arisehere: How does a focusproject, andwhendoes a
languageusewhat meansto realize an information structure focus in a position
different from theCFP?I addressfocusprojection in È 3.4.3, primarily on thebasis
of thedatapresentedin (Vallduvı andEngdahl, 1996). Subsequently, I present inÈ 3.4.4a setof hypothesesthat predict whenlanguages useword order, tuneor a
combination thereof to realizefoci in other thanthecanonical focusposition.
3.4.1 THE NULL HYPOTHESIS
To start,we canproposean initial version of a very general hypothesisabouthow
the build-up of a sentencemay reflect its information structure asrepresentedin
theunderlying linguistic meaning, (125).
(125) INFORMATIVITY HYPOTHESIS I (INITIAL VERSION)
In the unmarked case(unmarked mixed, free word order or unmarked15With an unmarked tune,or just an H* pitch accenton “yesterday”, (124) canbe an answerto
“What happened?”.
94I Thecategory of informativity
tune), languagestendto realize(verbal) contextually bound dependents/heads
before contextually nonbound ones, and contextually nonbound depen-
dents in canonical/systemic ordering.
The formulation of INFHYP1 is nothing new. In different guisesit appears
throughout work in thePragueSchoolof Linguistics,notably in FGD’s topic-focus
articulation (Sgall et al., 1986; Hajicova et al., 1995; Hajicova, 1993) andits use
of Firbas’s communicative dynamism(Firbas,1992).
Despite its simplicity, INFHYP1 holdsacrossa surprisingly largerangeof lan-
guage types. For example, it holds for mostof the OV languageswe have con-
sidered so far. In rigidly verb-final languages like Japaneseor Tamil, the focus
proper occupies the immediately preverbal position. In non-rigid verb-final lan-
guageslike Sinhalathefocusproper canalsobethepostverbaldependent(Herring
andPaolillo, 1995). And, INFHYP1 holds also for German,which hasa mixed
OV word order at thesubordinate clauselevel, wherethepreverbal position (right
before the verbal cluster) is usually considered to be the preferedposition for the
informational focus.
(126) Japanese
Taroo-waTaro-TOPIC É susi-oʤË
susi-ACCtabeta.eat-PAST
“TarooatetheSUSHI.”
(127) Tamil
antathat
natt-ilcountry-LOC É oru aracan
¯Ê Ë
onekingiru-nt-an
¯be-PAST
“In thatcountry, there wasA KING.” (Herring andPaolillo, 1995)(p.182)
(128) Sinhala
oyathat
kaelaeae-weforest-LOC
hit.iyaalive-PAST É nariy-ek Ê¤Ë .
jackal- INDEF
“In thatforest livedA JACKAL .” (HerringandPaolillo, 1995)(p.170)
Similarly, wefind thattheSVO languageswehaveconsideredall tendto prefer
to place the focus at the end of the clause (in the unmarked case). This holds
particularly for theSVO languageswith mixedword order, like Dutchor German,
or freeword order, like Czech,Greekor Russian.
Thecategory of informativity /95
However, the initial version of INFHYP1 is not obeyed by freeOV languages
(notably, type XXIII). For example,Hindi, HungarianandTurkishappear to form
acounter-exampleto theiniti al formulation of INFHYP1 (Hoffman,1995a),(Vall-
duvı andEngdahl, 1996), (Vasishth,p.c.). Theselanguagesprefer to placethe in-
formational focus directly before the verb - wherever the verb is placed. In other
words,sentence-finality is notacriterion for theplacementof focus(proper). Con-
sider for example (129), cited in (Vallduvı andEngdahl, 1996), and(130), from
Hoffman(1995a)(p.106).
(129) Hungarian
MariMary
JANOST
John-Acclatta.see-Past
“Mary saw JOHN”
(130) Turkish
a. EsraEsra
kitab-ıbook-ACC
okuyorread-PRESPROG
“Esrais reading theBOOK.”
b. Kitab-ıbook-ACC
EsraEsra
okuyorread-PRESPROG
“As for thebook, it is Esrawho is reading it.”
Only if we leave theverb in final position, like in (129)or (130), INFHYP1 is
obeyed. Changingthewordorderof (129)to (131) meansthatINFHYP1 no longer
applies,eventhoughtheword order assuchis unmarked.
(131) Hungarian
MariMary
lattasee-Past
Janost.John-Acc
“MARY saw John”
3.4.2 PREDICTING A LANGUAGE’ S CANONICAL FOCUS POSITION
Thus,leavingmarkedwordorderconstructionslike “subjectiveordering” (Sgallet
al., 1986) or focal fronting aside, INFHYP1 appears to cover (most)non-canonical
ordersin mixed OV andmixed or free SVO languages,but seemsto fail on free
OV languages.Canwefind a certain systemin theseobservations?TheproposalI
advance hereinvolvesfour aspects:
96I Thecategory of informativity
1. thedominant wordorder (XV/VX/SVO), aspertheGreenberg/Hawkins typol-
ogy;
2. thedegree of variability (rigid/mixed/free),asimplied by the hypothesespre-
sented earlier in this chapter;
3. thepresenceor absenceof a productive prosodic system(tune) andits interac-
tion with thedegree of variability ; and,
4. Venneman’s idea of category consistency (cf. (Venneman, 1977; Hawkins,
1983)).
To start with the last, Venneman’s ideaof category consistency is embodied
in his “Natural Serialization Principle”. This principle states that languageslin-
earize all their operator-operandpairs consistently, thus either (strictly) operator
before operand,or operandafter operator. Hawkins (1983) convincingly argues
that Venneman’s principle as such cannot be usedasan adequateexplanation of
word ordertypology. But the ideathat thereis a certain consistency in lineariza-
tion is perhapsnot entirely without merit. Namely, it doesseemthat languages
tendto havea canonical focusposition relative to theverbalheadthatis consistent
with its dominantword order. This leads to (132).
(132) a. OV Á immediate preverbal position
b. VO Á postverbal position
Subsequently, let usbring tune into thepicture. Languagesusetuneandword
order (amongother means)to relative degrees. This is a perspective already ad-
vanced by Sgall et al. in (1986) andlater work. As saidearlier, we consider the
following -initial- relation betweenvariability in word orderandtune.
(133) (OV F VO F SVO) & rigid Á tune
In otherwords,if a languagehasrigid word order, it is predicted to rely pre-
dominantly on tune,(at least for theunmarked case(s) realizing informationstruc-
ture).
Next, SVO behaveslike OV assoonasverbsecondnessis involved,sinceverb
secondness often leadsto theformation of a clause-finalverbal cluster. Similarly,
the following holds: if we have SVO but no verb secondness, thenSVO behaves
like VO. Fromthese observationsand(132a) we predict that thedefault focus po-
sition canbefound towards theendof thesentencein rigid, mixedandfreeSVO.
Thecategory of informativity /97
(134) a. ((rigid F mixed F free) & SVO & (V2 E VFinal) Á OV)Á immediate preverbal position
b. ((rigid F mixed F free) & SVO & Â (V2 E VFinal) Á VO)Á postverbal position
Basedontheaboveobservations(andtheimplicationsrelating them),wearrive
at the following formulationof our first (proper)hypothesisregarding therealiza-
tion of information structure. The hypothesisreformulates(125), anddetermines
whereweshould expect thecanonical focusposition in aparticular language,given
its type.
(135) INFORMATION STRUCTURE HYPOTHESIS 1 (CANONICAL FOCUS POSI-
TION):
Thecanonical focus position (CFP)is determinedby either of the follow-
ing hierarchies,depending on thelanguage’s type:
a. OV/SVO-UFP-hierarchy:
immediate preverbal position Ì(rigid F mixed F free) & SVO & (V2 E VFinal)F(rigid F mixed F free) & Â tune & OV
b. VO/SVO-UFP-hierarchy:
postverbal position Ì(rigid F mixed F free) & SVO & Â (V2 E VFinal)F(rigid F mixed F free) & Â tune & VO
Remark 10 (Pre-/post-verbal positioning ÍÎ sentence-finality). Therearea few
remarksthatwe should make aboutINFHYP1, (135). First of all, although actual
constructions may give the impression that a preverbalor postverbalposition co-
incides with sentence-finality, we do not imply this. Thehypothesisis deliberately
statedin termsof (immediate)pre-andpostverbalpositioning, on thebasisof the
consistency noted in (132). Sentence-finality is effectuated,in other words, but
it is not defining. And exactly becausewe perceive of it that way, we canrelate
in a coherentway thesuperficially different CFPs of for example SVO languages
like DutchandGerman(behavingclosely like OV in morecomplex matrix clause
constructions)andOV languageslike Hungarian or Turkish.
98I Thecategory of informativity
Secondly, the dataabout VO languages andtheir informationstructure is too
scarce to make any genuine predictions about their structure. We return to this
point againbelow. Ï3.4.3 FOCUS PROJECTION
By focus projection we understand the phenomenonwheremorewordgroupsare
interpreted asrealizing partof thefocusthanjust thewordgrouprealizing thefocus
proper. For example, take theEnglish sentencesin (136) below.
(136) English
a. Elijah left his cowboy-boots É on theTABLE Ê Ë .
b. Elijah left É his cowboy-boots É on theTABLE Ê Ë Ê Ë .
c. Elijah É left É his cowboy-boots É on theTABLE Ê Ë Ê Ë Ê Ë .
Here,“on thetable” realizesthefocusproper. In (136a) theboundarybetween
thefocusandthetopic is directly before thefocusproper. In (136b,c) we“project”
thatboundaryfurtherleftwards,understanding morewordgroupsasrealizing parts
of thefocus. Thesameholdsfor examplefor DutchandGerman,bothatthematrix
clauselevel (SVO) andthesubordinate clause level (SOV) astheexamplesbelow
(for Dutch)illustrate.
(137) Dutch
a. Elijah heeftzijn cowboy-laarzen É op deTAFEL Ê Ë latenstaan.
b. Elijah heeft É zijn cowboy-laarzen É op deTAFEL Ê Ë Ê Ë latenstaan.
(138) a. ...omdatElijah É zijn COWBOY-LAARZEN Ê Ë op de tafel heeft laten
staan.
b. ...omdatÉ Elijah É zijn COWBOY-LAARZEN Ê�Ë#Ê�Ë opdetafel heeftlaten
staan.
All theexamplesabove realizedifferentinformationstructures.They illustrate
the point indicatedearlier, namelythat focus projection or the possibilit y thereof
maybeimportant factor in how to realizeinformationstructureunambiguously.
Interestingly, languagesmayproject foci into different directions. For example,
Hungarianallowsfor rightwards focusprojection if the verbaldependents areall
orderedcanonically (i.e. according systemicordering). Thesentencesin (139)ex-
emplify this (Komlosy, cited by Vallduvı & Engdahl). Vallduvı & Engdahl indicate
thatfocus projection in Hungarian canalsobe leftwards.
Thecategory of informativity /99
(139) Hungarian
a. MariMary É+É+É almat Ê Ë eszik Ê Ë a kertben Ê Ë .
apple-ACC eatthegarden-IN
“Mary eatsapplesin thegarden.”
b. MariMary É+É+É beteg Ê Ë volt Ê Ë a tegnapÊ Ë .
sick be-PAST yesterday
“Mary wassick yesterday.”
The tendency to project a focus rightwardsstands in an interesting contrast
to focusprojection in mostother OV andSVO languages(which tendto behave
OV-lik e in the Mittel- andNachfeld - cf. Dutch, German). Most OV andSVO
languagesappear to project strictly towardsthe left from the focus and the nu-
clearstressit carries. On the otherhand, the requirementsfor focus projection in
Hungarianarepurely word order-related, relying as they do on the presence(or
absence)of a canonical order.
Thattunedoesnotplayasignificantrole in therealizationof informationstruc-
ture in Hungarian at all alsoseemsto be indicated by the following contrastbe-
tweenTurkishandHindi on theonehand, andHungarian ontheother. As Vallduvı
& Engdahl report, thefocusmustbepreverbal in Hungarian (unlesstheverbal head
is thefocus proper, in which caseit is placedclause-initial). If thefocusis formed
by a dependentwhosecanonical position is not immediately preverbal, thenit has
to be“moved” there. Becauseof theminimal role thattune seemsto play, we can-
not leave the dependentin situ andstress it - asthe minimal pair in (140) briefly
illustrate,(Vallduvı andEngdahl, 1996).
(140) Hungarian
a. * AttilaAttila
feltfear-PAST É a FOLDRENGESTOL Ê Ë .
theearthquake
“Attila feared theEARTHQUAKE.”
b. Attila É a FOLDRENGESTOL Ê Ë felt.
Turkish andHindi differ from Hungarian in this respect. There,we canleave
a dependent in situ andusea marked tuneto realize it as(part of) the focus. The
examplesin (141)form a minimal pair illustratingtheTurkish situation.
(141) Turkish
100I Thecategory of informativity
a. Bira
hizmełciservant
masa-nıntable-GEN
uzer-i-netop-POSS-DAT
not-unote-ACC
É YEMEK-tenonceÊÐËmeal-ABL before
bırak-tı.leave-PAST
“A servantput thenoteon thetablebefore lunch.”
b. Bir hizmełci É YEMEK-tenonceÊÑË masa-nınuzer-i-nenot-ubırak-tı.
We would like to proposethe following observations. First of all, thereis the
main point that the possibility of focusprojection influences how a sentencemay
be interpreted asrealizing a particular information structure. We thusregardit as
animportant factor in determining thechoiceof structural indicationsof informa-
tivity.
Secondly, a focuscanbeprojectedover wordgroups(dependents) thatareor-
deredaccordingto systemicordering. Thisalsofollows from (Sgalletal.,1986)(p.194ff),
whereNB elements by definition have to appear in systemic ordering. Here,we
observed this phenomenon for Hungarian (Vallduvı andEngdahl, 1996), andwe
canalsoillustrateon theEnglishexamples in (142).
(142) English
a. Christophergave a book Ò to KATHY ÓÕÔ .
b. Christophergave Kathy Ò a BOOK ÓÖÔ .
Arguably, the focus in (142a) can be projected further leftwards, but not in
(142b) becauseof thedative-shiftedBeneficiary. Otherexamplescanbefound in
(Sgallet al., 1986)(p.194ff).
Thirdly, focus projection can in principle be either leftwards or rightwards.
Given the Hungarian data, and the contrastingdatafrom Hindi andTurkish, the
direction in which a focus mayprojectover verbsandsystemically orderedword-
groupsseemsat leastto dependon theproductivity of tune in thegivenlanguage.
3.4.4 CHANGING FOCUS
In this section we have a look at constructions that realize information structure
wherethe focus proper appears in a position otherthanthe canonical focusposi-
tion. Like we saidearlier, there maybevarious reasonsfor doing so,arising from
the information structure and possible focus projections, thematic structure, etc.
Bearing (132) in mind, having the focus proper in a non-canonical position can
meantwo things. Eitherthefocuselement appears in aposition otherthantheCFP
Thecategory of informativity /101
but thatposition is still consistentwith (132), or it is in aposition thatis neitherthe
CFPnor consistentwith (132).
For example, in anOV languagethefocus could bepreverbal but not immedi-
ately preverbal,in which caseconsistency would be maintained. However, if the
focuswouldnotbeimmediately preverbalnor preverbalatall, thenbothINFHYP1
(describing the unmarked or canonical case) andconsistency would be violated.
Naturally, other factors in a languagesystem determine whether we can obtain
thesedifferent marked casesat all - for example, in a rigidly verb-final OV lan-
guageit is hardly likely thata postverbalfocal elementwould befound.
From a viewpoint of economy, like Sgall et al. discuss in (1986) for de-
pendency grammarin general, we could set up the following INFORMATIVITY
MARKEDNESS PRINCIPLE. We useCC for category consistency, andFP for fo-
cusposition.
(143) INFORMATIVITY MARKEDNESS PRINCIPLE:×CC & UFP ØÙÛÚ Ü ×
CC & NON-CANONICAL FP ØÙÛÚ Ü ×NON-CC & NON-CANONICAL FPØ
In words,theleast markedconstruction is onein which thefocusproper is re-
alizedin the canonical focusposition. A moremarked construction is onewhere
the focus proper is realizedin a non-canonical focus position, but still consistent
with thegeneral operator-operand direction. Moremarkedthaneitherof theprevi-
ousconstructions is onewherethe focusproper is not realized in canonical focus
position,nor category consistency is obeyed.
Intuitively, if we would follow out economy, thenwe would alsoget the pre-
diction that a languagewould first tend to theuseits predominant strategy for re-
alizing informationstructure,to obtain a moremarkedfocus position CC & NON-
CANONICAL FP, (unless the construction would be ambiguous betweena focus
proper realizedin CFPandamarkedFP).To obtaina really marked focusposition
NON-CC & NON-CANONICAL FP the languagewould resortto a different strat-
egy, possibly in combinationwith its predominant strategy. Notethattheremaybe
different strategies for mixedandfreeword orderlanguages.For example, mixed
languageshave articles at their disposal to realize contextual boundness, whereas
free(in general)do not.
For example, take againOV languages. An OV languagelike Sinhalahas
mixedword order, andis non-rigid in its verb-finality. INFHYP1 predicts that the
102Ý Thecategory of informativity
unmarked focusposition is immediately preverbal. Subsequently, we predict that
moremarkedfocuspositionswould eitherbeobtainableusingword order,placing
it towardsthebeginningof thesentence(consistently preverbal), or using tune(and
word order)to place thefocusafter theverb. As it turnsout, this is indeedthecase
- cf. Herring& Paolillo (1995).
Similar observations, consistent with the above proposal, can be madefor
Japanese.Although Japaneseis rigidly verb-final andhasanimmediately preverbal
unmarkedfocusposition (144a),wecanobtainamoremarked focusby ordering it
at thebeginningof thesentence(144b) without having to useany marked tune.16
(144) “What did Taroeat?”
a. Japanese
Taroo-gaTaro-NOM
susi-osushi-ACC
tabeta.eat-PAST
“TaroateSUSHI.”
b. Susi-oTaroo-gatabeta.
Theproposalalsoholds for freeOV languageslike Turkishor Hindi, cf. (Vall-
duvı andEngdahl, 1996) for Turkish.17 Furthermore,wecanobserve this behavior
in mixed SVO languages (V2-case) like Dutch andGerman, andfor a free SVO
languagelike Czechwe already illustratedthis.
On the basis of theseobservations, we formulate the following hypothesis,
INFHYP2. INFHYP2 concerns realizations of informationstructure that aremore
marked dueto their realization of the focus proper in other thanthe canonical fo-
cusposition. We use“ambiguous” to indicatewhethera construction (sentence)
would be ‘ambiguous’ betweena canonical and a non-canonical focus position
16Having saidthat,native speakersmaypreferto put somestresson themarkedfocus,evenwhena -wa particleis usedto indicateexplicitly the topic. Note that the ÞQßpà particledoesnot needtoindicatefocus,cf. (Heycock, 1993).
17At thesametime, the datais slightly inconclusive aboutHungarian.Obviously, becauseHun-gariandoesnot have a particularlyproductive tune,we would not predict to observe a post-verbalmarked focus. Moreover, the constraintthat the focus position has to be immediatelypreverbalwould seemto contradictthepossibilityto obtaina (preverbal) markedfocususingword orderonly,in Hungarian.In the light of thescarcetyof thedataavailableto us,we leave Hungarianout of theequationfor themoment.It maymeanthatwe will have to make a morefine-grainedpredictionona futureoccasion, but thatdoesnot invalidatetheapproachassuch.
Thecategory of informativity /103
without any further structural indicationlike tune.
(145) INFORMATION STRUCTURE HYPOTHESIS 2 (MARKED REAL IZATION)
a. (CC & á UFP & á ambigous)â ((mixed ã free) â word order) & (rigid â tune)
b. ( á CC & á UFP & (ambiguous ãäá ambigous))â ((rigid ã mixed ã free) â word order & tune)
Remark 11 (Mark ed realization has to be grammatical). Quite naturally, the
implications of INFHYP2 all are,ultimately, constrained by what is well-formed
in a particular language. If a languagedoesnot have a very productive word order
system,INFHYP2 should not be interpretedto state that there is onenevertheless.
But, within these limits, INFHYP2 seemsto cover eventherareconstructions like
Y-movementin English. åTo recapitulate,wehave INFHYP1 whichpredictsthattheunmarkedfocusposition
is consistent with thedirectionality of thedominantwordorder(OV/VO,with SVO
split into differentcases). Furthermore, INFHYP2 predicts thatmoremarkedfocus
positions can be obtained using either word order or both word order and tune,
dependingonhow marked(with respectto theInformativity MarkednessPrinciple)
the construction would be and whether it by formal structure alone it would be
ambiguousbetween a canonical anda non-canonical focus position.
Specifically, for VO languages, thehypothesespredict thefollowing structural
indicationsof informativity in OV languages.
(146) Structural indicationsof informativity in OV:
a. Rigidly andnon-rigidly verb-first OV languageshave an immediately
post-verbalunmarkedfocusposition.
b. Rigid VOrealizeinformationstructureusingpredominantly tune;mixed
andfreeVO languages usepredominantlyword order.
c. Non-rigidly verb-initial OV languagescanhaveamarkedimmediately
preverbal focus.
d. Rigidly andnon-rigidly verb-initial OV languageswith mixed or free
wordorder canhavemarked focusposition towardstheendof thesen-
tence,using just word orderunlessthestructureassuchwould beam-
biguousbetweena CFPconstruction.
104Ý Thecategory of informativity
Because dataabout structural indications of informativity in languages with
VO word order is very scarce, it is hard to verify the above predictions. Kroeger
(1993) briefly discussesthe realization of information structure in Tagalog,and
an informal inspection of Biblical Hebrew only revealsa partial picture of what
appearsto bearathercomplex situation. Inspectingwordorder variationin Biblical
Hebrew in context (i.e. in the bible) appearsto indicatethat it hasa rathermixed
word order, andthat it placespronominalcliti csdirectly after theverb (Zussman,
p.c.). It uses word order as its primary structural indicationsof informativity, in
other words. With regardto theunmarkedfocusposition, Biblical Hebrew indeed
seemsto prefer theimmediately postverbalposition. Furthermore,it allows for an
SVO variation, in which thepreverbal dependent in fact realizesa (marked) focus.
Whether thereis any particular tune associated to this fronting is not clear, and
requiresfurther research.
The situation in Tagalogis more complex, due to its rich voice system, and
theinfluenceof adependent’sdefinitenessonwhetherit canactually berealized in
nominative case.In theexamplesbelow, all from (Kroeger, 1993)p.62ff, thegloss
AV meansactive voice,OV objective voice, andIV “indi rectobjective” voice.
(147) Tagalog
AnoWhat
angNOM
kinainPERF.OV-eat
mo?2.SG.GEN
“What did you eat?”
(148) a. KinainPERF.OV-eat
ko1.SG.GEN
Ò ang-isda ÓÐÔNOM-fish
“I atethefish”
b. KumainPERF.AV-eat
ako1.SG.NOM
Ò ng-isda Ó ÔGEN-fish
“I ate(some)fish.”
Thequestion in (147) is formulatedin objective voice (OV). In (148a) wehave
thesamevoice,with aPatient thatis in nominativecaseand(necessarily) definite.
(148b) usesactive voice, makingit impossible for the object to be in nominative
caseandto bedefinite.
(149) Tagalog
AnoWhat
baQUES
angNOM
biniliPERF.OV-buy
mo2.SG.GEN
sa-pamilihan?DAT-market
Thecategory of informativity /105
“What did youbuy at themarket?”
(150) a. BiniliPERF.OV-buy
ko1.SG.GEN
Ò ito-ng damit ÓÐÔ .NOM-this-L INK dress
“I bought this dress.”
b. Ò Ito-ng Ó¤Ô damitangbinili ko.
Now, let usconsideraquestion with theverbin activevoice(rather thanobjec-
tive voice,asabove).
(151) Tagalog
Sinowho
angNOM
gumawaAV.PERF-make
ng-sapatosGEN-shoe
naL INK
iyon?that
“Who madethoseshoes?”
(152) a. ?GinawaOV.PERF-make
Ò ni-Bing ÓæÔGEN-Bing
“Bing made(them).”
b. Ò Si-Bing Ó ÔNOM-Bing
angNOM
gumawaAV.PERF-make
ng-sapatosGEN-shoe
naL INK
iyon.that
“Bing is theonewho madethose shoes.”
Theanswerin (152b) is thepreferredanswerhere,cf. (Kroeger, 1993)(p.63).
Thus,we seethatTagalogcan-in principle- placeits focuseither preverballyand
postverbally. Therebywe might understand the postverbalposition to be the un-
marked one, based on the observation that topicalization using the ay-inversion
construction placesanelementin thepreverbal position,cf. (Kroeger, 1993)(p.67)
SUMMARY
Basedonempirical data,wepresenteda typologicalcharacterisation of variability in word
order,andasetof hypothesesthatpredictwhen(andwhy) languagesmakeuseof strategies
like word orderor tuneto realizeinformation structure. The first hypothesis,INFHYP1,
predicts thatthecanonical focuspositionis theimmediatelypreverbalpositionin OV lan-
guages,andin SVO constructionsthathaveaclause-finalverbal cluster. For VO languages
andSVO constructionswithoutverb-secondness,INFHYP1 predicts thatthecanonicalfo-
cusposition is post-verbal. We observed that sentence-finality may be effectuated,but
that it is not defining. This enablesus to relatethecanonical focuspositionof (complex)
DutchandGermanclausesto therealizationof informationstructure in OV languageslike
Hungarianor Turkish.
106Ý Thecategory of informativity
The secondhypothesis,INFHYP2, predicts how more marked realizations are real-
ized. We notedthat thematicstructure and the possibility of focus projection may de-
terminehow informationstructure is to be realized,andthat only through more marked
constructionssuchrealizationcansometimesbeachieved (e.g. to avoid ambiguity). For
example, INFHYP2 makesthefollowing predictionsaboutrealizingthefocusproper in a
non-canonical focus position. As long astheconstruction cannotbeunderstoodto realize
a focusin thecanonical focuspositionandthenon-canonical focuspositionis placedrel-
ative to the canonical focuspositioncategory consistently, thenword ordercanbe used.
Otherwise,aninteractionbetweentuneandwordorderis predicted.
With respectto thehypotheses,we notedthat thereis a differencein theuseof these
strategiesamong languageswith rigid, mixedandfreeword order, andthatstrategiesare
usedto a relativeratherthananabsolute degree.
Thediscussionin thischapterconfirmedvarious of theprincipal hypothesesadvanced
in thePrague Schoolof Linguistics,andmostrecentlyin FGD, about language typology
(SkalickaandSgall,1994; Sgall,1995) andtherealizationof informationstructure(Sgall
et al., 1986; Hajicova, 1993). Even though we looked at a relatively small number of
languages,the hypotheseshave beenformulatedagainstdatathat is typologically more
diversethanis usuallyconsideredin theliterature.
In subsequent chapterswe look in moredetailat how we canformalize the ideathat
informationstructure is a fundamentalparameter in determining the realizationof a sen-
tence.Chapter4presentsdetailedarchitecturesmodellingrigid, mixedandfreewordorder
andtheuseof word orderto realizeinformationstructure.Chapter5 extendsthe formal
modelsto cover tuneandits useasa structural indicationof informativity (both aloneand
in interactionwith wordorder).
CHAPTER 4
A FORMAL MODEL OF WORD ORDER AS
STRUCTURAL INDICATION OF INFORMATIVITY
In this chapter we develop grammararchitecturesthatmodeltheuseof word orderasstructural
indicationof informativity on thebasisof thedistinctionsof rigid, mixedandfreewordorderas
discussedin Chapter 3. Principal to our formal account is the view of adjacency asparameter
(Moortgat and Oehrle, 1994). The results of this chapter are grammararchitecturesof basic
strategies to realize information structure in VX, XV, andSVO languages,controlled by the
informativity hypotheses. The coreof thesearchitectures is formedby architecturesof rigid,
mixedandfreeword order,controlled by thevariation hypotheses.
4.1 INTRODUCTION
Thecross-linguistic perspective on word order thatwe developedin thepreceding
chapter hasbeenformulated independent of any particular grammarframework.
In principal onecould thustake theframework of one’s liking to implement these
ideas. Here,we naturally focus on we could useDGL. The goal of this chapter
is to modelword orderasa structural indicationof informativity. To that end I
provide grammararchitecturesthatdescribebasic formsof variability in word or-
der,andshow how information structurecancontrol word ordervariation. Theac-
countis basedontherelation betweencontextualboundnessandsystemicordering
(Chapter2), andexploits the view of adjacency asa parameter (with information
structure/contextual boundness asanimportant factor besideswell-formedness).
Below I first provide a brief survey of proposalsfor using categorial grammar
to modelword order phenomenaup to freeword order. As we already pointed out
earlier in Chapter 2, thecombinatorytradition adheresto aPrincipleof Adjacency –
only string-adjacentwordscanbecombined.A direct consequenceof thisprinciple
is that variability must to be modelled in the lexicon. To model variability we
have to usecategoriesthat lexically defineflexibili ty to the directions in which
argumentsaretaken.
107
108Ý A formal model of wordorder asstructural indication of informativity
Thecombinatoryapproachdiffers in this from the logical tradition, wherewe
conceive of adjacency asa parameter. Lexical categoriesdefinecanonical word
order. Variability in word order is achieved through the application of structural
rules that have the possibility to alter the tree-structure. Theapplication of struc-
tural rules is controledby theconfiguration of the tree-structure,andpossibly any
features that individual nodes carry. For one, that enables us to create a very
fine-grainedaccount wherepossible variation in word order can be conditioned
on contexts thatarelarger thanindividual words.Anotherconsequenceis thatwe
canusecontextual boundness asa parameterto control structural rulesmodelling
word order, and have the relation between surfaceform andunderlying linguis-
tic meaning/informationstructure definedcompositionally by the Curry-Howard
correspondencethatholdsfor thecalculus in general.
In ç 4.3 I work out a proposal for capturing word order in DGL. We present
modelsof word order-based strategiesfor realizing information structure,aspre-
dicted by the informativity hypothesespresented in Chapter3. Thesestrategies
structurally control morefundamentalgrammararchitectures thatmodelthebasics
of rigid, mixed, andfree word order in VX, XV, andSVO languages.1 In other
words,thestructural control formally implements theview thatvariability in word
order is paramatrizedby theinformationstructureto berealized.
4.2 MODELS OF FLEXIBLE WORD ORDER IN CATEGORIAL GRAMMAR
The purposeof the current section is to provide a brief survey of proposalsfor
using categorial grammarto modelword orderphenomenaup to freeword order.
What the categorial grammarproposalsdiscussedhere,andthe DGL proposalinç 4.3, have in commonis that they employ a flexible notion of surfacestructure in
a setting thathasa generative power strongerthancontext-freeness.It needslittl e
argumentation thatclassical (context-free) phrase-structuregrammaris wholly in-
adequateto explain variation in wordorder. To ‘all ow’ for variation, alternaterules
would have to begiven that describe the other possible orders. However, assoon
asvariation involvesdiscontinuity (of any type)thereis noway to describe it since
we cannot relate thedisplacedelements to thesitewherethey would normally be
located. Finally, any generalization we canmake over possible orderscannot be
expressedin a phrase-structuregrammar.
1For reasonsof conciseness,we presenttheformulationof thesemodelswithout the è�é�êÖë"ì fea-ture. In theunderlyingarchitectureusedfor theinformationstructurearchitecturesthis featureis ofcoursepresent,andusedfor controllingword order.
A formalmodelof wordorderasstructuralindicationof informativity /109
Below westartwith proposalsthathavebeendevelopedin thecombinatorytra-
dition in categorial grammar:Steedman’s CCG(1996; 2000c), Hoffman’s MCCG
(1995b; 1995a), and Baldridge’s Set-CCG(1998; 1999) and modalized CCG,
(2000). Thereafter, webriefly addressvariousdiscussionsof wordorder by Moort-
gat& Oehrle, setin categorial typelogic.2
4.2.1 STEEDMAN’ S CCG
Combinatory Categorial Grammar(CCG)wasfirst introducedby AdesandSteed-
manin (1982) asa generalization of theearlier categorial grammarframeworksof
AdjukiewiczandBar-Hillel , andwaslatergreatly expandedby Steedmanin for ex-
ample(1996; 2000c). At theheart of CCGwefind asetof combinators thatdefine
composition. In CCG,the combinators areperceived of as rule schemata whose
instantiation canbefine-tunedto thesetting of aparticular language,but whichare
otherwise thesolemeansby which we can-or needto- modelcomposition across
languages.Theschematathushave a certain cross-linguistic flavor, andSteedman
hasgone to greatlengthsshowing thatonecanindeedemploy thecombinatorsto
modela varietyof languages.Thevariation in theinstantiationsof theschematais
thenexplainable with referenceto languagetypology. An interesting exampleof
suchexplanationis Steedman’s (2000c) discussionof the treatmentof dependent
clauseword order in Dutch,German,andEnglish.3
CCG’s combinatorsarebasedon the work of Curry andFeys, andwereorig-
inally intendedfor to modelthe í -calculus. A crucial differencebetween CCG’s
combinatorsandtheir original counterparts is, though, thatthe(recursive) applica-
tion of theformerarerestricted.This limits thepower of CCGpur sangandis the
main reason why Vijayashanker andWeir areableto show in (1994) that CCGis
mildly context-sensitive andparseable in polynomial time.4
Moreprecisely, wecandefineCCG’scombinatorsasfollows (Steedman, 2000c).
2We do not discussHepple’s (1990) proposalas it got superceded by his own later work onhead/dependentasymmetriesin categorialtypelogic (1994; 1996; 1997), whichwediscussedalreadyin Chapter1.
3Furthermore,see(Kruijf f andBaldridge,2000)for abrief cross-linguisticcomparisonregardingtheavailability of particularcombinators in Dutch,English,German,andPortuguese.
4As such,CCGcould be setapartfrom categorial type logics. The latterareTuring Complete,asCarpentershowed in (1995), if they are not restricted. Like CCG’s restrictionleadsto its moreconstrainedbehavior, sothereexiststhepossibilityto restrictcategorialtypelogic though.Weprovedin (Kruijf f andBaldridge,2000) thatwe canconstruct a formal bisimulationof CCGin a categorialtypelogic fragment,with thefragmenthaving a (weak)equivalenceto CCG.Conversely, thearticleshows thata logical interpretationof CCG is possible,counteringfor exampleMorrill’ s criticism in(1994).
110Ý A formal model of wordorder asstructural indication of informativity
CCG extends the Adjukiewicz-Bar-Hill el calculus (AB) by adding rules of syn-
tactic combinationwhich correspond to directionally specific forms of Curry and
Feys’ composition (B), type-raising (T), andsubstitution (S) combinators. These
combinatorsaredefinedby thefollowing equivalencesonpredicate-argumentstruc-
tures:
(153) a. B î�ïñðòí�óõôÖî÷ö�ï�ó�øb. T óùðòíúî�ôÖî�óc. Sî�ïñðXí[ó�ôÖî�óûö�ï�ó�ø
Definition 10 (Combinatory rules for CCG). CCG extends its rule setbeyond
thefunction applicationrulesof AB asfollows:
(154) Rulescorresponding to ü .
a. ý�ü�þ ÿ ��� þ�� ����� þ � � ÿ ��� þ1í���ô �÷ö���Õó�øb. ý�ü�� þ ÿ ��� þ�� ����� þ ��� ÿ ��� þ1í���ô �÷ö� � ó�øc. ��ü�þ ����� þ ÿ ��� þ�� � � ÿ ��� þ1í���ô �÷ö� � ó�ød. ��ü�� þ ����� þ ÿ ��� þ�� ��� ÿ ��� þ1í���ô �÷ö���Õó�ø
(155) Rulescorresponding to � .
a. ý��ùþ ÿ þ�� ��� ��� ö ��� ÿ ø þ1í��õô ���b. ���ùþ ÿ þ�� � � ��� ö ��� ÿ ø þ1í��õô ���
(156) Rulescorresponding to � .
a. ý�� þ ÿ ������� þ�� ����� þ ��! ÿ ��� þ+í���ô �"�õö� � ó�øb. ý��#� þ ÿ ������� þ�� ����� þ �$! ÿ ��� þ6í���ô �%� ö� � ó�øc. ��� þ ����� þ ÿ ������� þ�� � ! ÿ ��� þ+í���ô �"�õö� � ó�ød. ���#� þ ����� þ& ÿ ������� þ�� �$! ÿ ��� þ6í���ô �%� ö���Õó�ø
CCG as suchcannot be usedto model free word order,and exactly for that
reason offspringsasMCCGandSet-CCGhave beenintroduced. Yet,CCGcanbe
successfully applied to modelphenomenafound in mixed word order languages,
likecross-serial dependenciesin Dutch,cf. (Steedman,2000c), Chapter 6 for more
detail.5
5TherearefundamentaldifferencesbetweenSteedman’s account andhow we modelcross-serialdependenciesin DGL. Steedmanassumesthatthebasicword orderof Dutchmatrixclausesis VSO,andthatof dependentclausesSOV. Accordingly, verbsareassigneddifferentlexical categoriesforusein matrix anddependentclauses.On the contrary, in DGL we first of all assumethat Dutchmatrixclauseword orderis SVO. Furthermore,thereis noneedto haveseverallexical categoriesfora verbto mirror its useat differentclauselevels.
A formalmodelof wordorderasstructuralindicationof informativity /111
4.2.2 HOFFMAN’ S MULTISET-CCG
In (1995a;1995b) Hoffman introducesmultisetcombinatory categorial grammar
(MCCG),anextension of CCGto dealwith freewordorder. Thebasic ideabehind
MCCG is to relax the subcategorization requirementsof a verb such that it no
longer needsto specify the linear order in which the argumentshave to occur.
Rather, a verb is assigneda function category that takesa multiset of arguments6
which arenot necessarily assignedany directionality. For example, in MCCG we
specify the category of a transitive verb as '�( ×�)+*-,/.102)43�565 Ø . The verb takes a
subject) *-,/.
anda direct object) 37565
, in any order, resulting in a construction
of category ' (sentence).
To beable to combinea function andits argumentin any order, MCCGcannot
employ thestandardrulesfor functional application. Instead,we have (157).
(157) a. Forward application ( ý ):8 (¤ö:9�;�ï&<>= ×�? Ø�ø ? � 8 ( 9@;;ïA<b. Backward application ( � ):? 8 (¤ö:9�;�ï&<>= ×�? Ø�øB� 8 ( 9@;;ïA<
With the rules as in (157), we cancombinea verb and its arguments in any
order. To obtain a semantics, MCCG co-indexes the category’s argumentswith
the argumentsin the predicatestructure. Thereason being that we canno longer
useordinary í -calculus due to the insensitivity to the order in which arguments
aretaken. For example,thecategory for a transitive verb like readwould become
' þ�;�C�D&E�ö F 0 G ø�( ×�) *#,/. þ F 02) 3�5H5 þ G Ø .For composition ( ü ) wecandefinerulesasin (158), using set-theoretic opera-
tions.
(158) a. Forward composition ( ýMü ):8 (¤ö:9�;�ï&<�IJ= ×�? Ø�ø ? ( 9�;�ï&<LKM� 8 (¤ö:9@;�ï&<�IN=O9@;�ï&<LK§øb. Backward composition ( �Mü ):? ( 9@;;ïA<�K 8 (¤ö:9@;;ïA<�IJ= ×�? Ø�øB� 8 (¤ö:9@;�ï&<�IN=O9@;�ï&<LK§ø
With these composition rules,MCCGcanhandle for example freeword order
of sentential adjuncts( '�( × ' Ø ). Also, by allowingmultiple verbsto composeusingü , MCCGcananalyzecomplex sentenceswith embeddedclauses.
6Multiset: a category of thesametypemayoccurmorethanoncein thesameset.
112Ý A formal model of wordorder asstructural indication of informativity
Quite naturally, the question ariseshow all this freedom canbecontrolled, or
restricted. For example, Japaneseallows for a free ordering of argumentsbut is
otherwiserigidly verb-final. Wecanrepresentthis in MCCGby attachinga“direc-
tional feature” to arguments: '�( ×"PRQ) 56SUTHV%WX*-,Y. 0ZPRQ) 56SZTYV%W[3�565 Ø . To make composition
sensitive to these directional features,we require that in× � 0 �Mü#Ø the
?argument
of8
is markedasPRQ?
, andthat in× ý 0 ýMü#Ø it is markedas
Q\ ?. It remains anopen
issue, though, whether morefine-grainedrestrictions canbe captured in this way
aswell - for example,theoccurrenceof Czechclitics in theWackernagelposition,
a criticism mentionedin (Kruijf f, 1999a). Furthermore, there aresubstantialdiffi-
cultieswith the relation betweeninformationstructureandword orderin MCCG.
As I already mentionedin Chapter 2 (p.66ff.), the Principle of Adjacency forces
oneto modelnot only variability of word orderbut alsoits effect asstructural in-
dication of informativity elsewhere thanin therule component. Variability canbe
modelled in the lexicon, but the effect of word orderasa structural indicationof
informativity cannot. In MCCG, thiseventually leadsmoreor lessto adissociation
of information structurefrom word order,contrary to general linguistic intuitions.
4.2.3 BALDRIDGE’ S SET-CCG, MODALIZED CCG
Baldridge presents in (1998; 1999) a framework that incorporates ideasfound in
Hoffman’s MCCG but thatat thesametime retains the formal andcomputational
strengths of CCG. Next to Set-CCG,Baldridge proposesin (2000) a version of
CCG that distinguishesdifferent modesof composition like categorial type logic
does, basedon (Kruijf f andBaldridge,2000). Here,we present bothSet-CCGand
modalizedCCG.Thereason for discussing modalizedCCGis that,dueits affinity
with categorial typelogic, it seemsenvisionableto extent modalizedCCGto cover
Set-CCGandthenusethis “modal-Set-CCG”to overcometheproblemsnotedfor
multiset combinatorycategorial grammar.
JustlikeMCCG,Set-CCGenablesoneto expressthatparticularargumentscan
becombinedwith in any order. However, unlikeMCCG, Set-CCGcategoriesretain
thespecification of directionality (unlike MCCG’s ( ) sothatSet-CCG’s categories
in general look a lot more like the original CCG categories. For example, the
category for a transitive verb (with basicSOV order) is ' ��×�)�]^,Y._02)`SZ565 Ø . Then,
occurring left of the head,both argumentscanbecombined in any orderusing �(Definition 11).
Definition 11 (Set-CCG). Therule schematafor the combinators are defined in
A formalmodelof wordorderasstructuralindicationof informativity /113
Set-CCGasfollows.
(159) a. Backward application( � ):? 8 � ö�a�b ×�? Ø�øB� 8 � ab. Forward application:( ý ):8 � ö�a�b ×�? Ø�ø ? � 8 � a
(160) a. Backward composition:?c� öedObgfQø 8 � ö�a�b ×�?h� d÷Ø�øB� 8 � a � fb. Forward composition:8 � ö�a�b ×�?i� d÷Ø�ø ?j� öedkblfQøB� 8 � a � f
(161) a. Backward type-raising:8 � m ��× m ��× 8 Ø3Øb. Forward type-raising:8 � m ��× m ��× 8 Ø3Ø
Remark 12 (Rigidificati on and Set-CCG’s power). An important point to note
aboutSet-CCGis its rigidification of slash-directionality. For thepurposeof econ-
omy, we areallowed to usethe “up-down” slash ( to specify categories, but once
oneargumentfrom a ( ’d bagis taken in a particular direction, thenall arguments
have to be combined with in that direction. Thus, '�( ×�)1T�02)`, Ø can take either
take its argumentsall to the right, in any order, so that we get a rigid head-initial
structure.Or, we cancombine with all argumentsto theleft.
In otherwords,Set-CCGallows for scrambling of the arguments, but not of
theheaditself. Theheadremainsin afixedposition, andit is this rigidification that
makesit possiblefor Set-CCGto havethesamegenerativestrenghtandparseability
asCCG.åBesidesSet-CCG,BaldridgehasalsoproposedmodalizedCCG, based(Krui-
jf f and Baldridge, 2000). Below, we first briefly describe the intentions behind
(Kruijf f andBaldridge,2000), andthenexplain how Baldridgeemploys it to create
modalizedCCG.
Our goal in (Kruij ff and Baldridge, 2000) is twofold. Firstly, we try to es-
tablish a fragmentin categorial type logic thataccepts exactly thesamestructures
asCCGcanallow for. The fragmentcanthusserve asa logical interpretation of
CCG,countering Morrill ’s criticism levelled against CCGin (1994)(for example,
114Ý A formal model of wordorder asstructural indication of informativity
cf. p.257).Secondly, by proving that the fragmentis anexactsimulation of CCG,
we alsoobtain a (weak)generative equivalenceto CCG’s mild context-sensitivity
andthe possibility to parse with the fragment in polynomial time (Vijayashanker
andWeir, 1994). Hence,the fragment illustrateshow onecanemploy a restricted
form of commutativity that does not lead to a collapse,and which gives rise to
a linguistically interestinggenerative strength. Concisely, the simulation canbe
definedasfollows.
Definition 12 (A simulation of CCG). TheCCG-equivalent fragment definedby
(Kruijf f and Baldridge, 2000) usesmodalities n�o 02pAq , thebasestandard logic NL,
and the following structural rules for simulating application ( � 0 ý ), composition
(B), andtyperaising (T).
(162) Right Associativity:rtsvuwrtxzyj{�|Y|7}@~���� �j�rHr�sMu�x�|�yj{�|7}@~����
(163) Left Associativity:rHr�sMu�x�|�yj{�|7}@~���� ���rtsvuwrtxzyj{�|Y|7}@~����
(164) Right Commutativity:rHr�s�y�x�|���{�|[}@~���� � �����rHr�s��i{�|�y-x�|[}@~����
(165) Left Commutativity:rts�y�r�x>�i{�|H|[}@~���� � �����rtxz��r�s�yi{�|H|[}@~����
The baselogic naturally models ����� . The rules for associativity enable us to
simulate ���g����� , whereasweneedtheadditional, limited form of commutativity
to handle �i�l�������k� . Wedonotneedanyotherrules,astyperaisingis a theorem
of NL already (Oehrle, 1994). �Todefinethemodalizedversion of Combinatory Categorial Grammar,Baldridge
redefines the rules of CCG (Definition 10) to respect the modal behaviorof the
rules which we useto simulatethem. Furthermore,Baldridgedefinesthe setof
modalitiesto be nZ�[�2o�� pAq andusestwo variable modesn��&�%� q , where �h�wnZ�X�2o�� p&qand �l�Bn�o�� p&q .
A formalmodelof wordorderasstructuralindicationof informativity /115
Definition 13 (Modalized CCG). Therules for modalizedCCG are as follows.
Notethat the semantics of the rules are the sameas givenin (153) for pure CCG
andare thusomitted.
(166) TheCCG’base logic’.
a. �%���- �¡�¢ �£ £�¤ ¡b. �%���- £ ¡�¥ �£�¤ ¡
(167) Composition
a. �%���h�� �¡�¢ ��£ £ ¢ p7¦ ¤ § ¡j¢ p�¦b. �%���h�� £ ¥ o ¦ ¡�¥ ��£ ¤ § ¡j¥ o ¦
(168) Crossing composition
a. �%���k�X�- ¡�¢ p £ £ ¥ o ¦ ¤ § ¡�¥ o ¦b. �%��� � �- £ ¢ p�¦ ¡�¥ o&£ ¤ § ¡�¢ p�¦
(169) Type-raising
a. �%��¨©�j �¡ ¤«ªg£ ¢ � � £ ¥ � ¡j�b. �%��¨©�j �¡ ¤ ª £ ¥ � � £ ¢ � ¡j�
Assaid,type-raising is a theoremof NL, but weneedto explicitly state it for Com-
binatory Categorial Grammar. We permit any modality to decorate the slashes
which are created in order to mimicthebehavior of NL. �4.2.4 MODELS OF WORD ORDER IN CTL
Over time, there have beenvariousproposals for dealing with word order-related
phenomenain theLambektradition. With therealizationthatacontext-freeframe-
work doesnotsufficeto explain for examplediscontinuousconstructionslike long-
distancedependencies,people turned their attention to theLambek-Van Benthem
calculus LP. LP is a calculus that is fully associative andcommutative - thusal-
lowing for a muchfreerordering.
Worse,LP actually allows for any ordering. Thus,obviously, this calculus is
too strongand,dueto its context freeness,the original Lambekcalculus L (Lam-
bek, 1958) is too weak. An intermediate position betweenL and LP would be
ideal. Justadding commutativity to L precipitatesa collapseto LP, asMoortgat
(1988) shows. Control thusturns out to bethekeyword.
116¬ A formal model of wordorder asstructural indication of informativity
Early proposalsfor structural control involved theuseof unarymodalopera-
tors like , ® , or ¯ - for example,seeMoortgat’s (1988; 1996), Hepple’s (1990),
or Morrill’ s (1994). Appropriately marked wordsor structureslicensethe appli-
cation of structural rulesthat involve -for example- theuseof commutativity. Be-
cause structural rules thus no longer necessarily apply to all structuresbut only
to specificconfigurations, we cangain a generative strength that goesbeyond the
ordinaryLambekcalculuswithout collapsingto LP.
From the technical viewpoint, Kurtonina and Moortgat show in (1996) that
the thus evolving landscapeof substructural proof logics behavesnicely, having
characteristics like completenessandsoundness. More importantly, Moortgat&
Oehrlediscussin their importantpaper “Adjacency, dependency, andorder” (1994)
whatthecategorial typelogical perspective on word order in factamounts to from
a linguistic point of view.
Thepoint thatMoortgat & Oehrleadvanceis thatadjacency is a parameterof
resourcestructuring. Structural rulescanbeusedto reconfigurestructuressuchthat
elements thatusedto benon-adjacent, now do becomeadjacent. Thus,adjacency
is notconsidered to beanecessarycondition for composition, but is something that
canbebrought about. This view stands in sharp contrastto Steedman’s Principle
of Adjacency for CCG. According to that principle, only string-adjacententities
maybecombined(cf. (Steedman, 2000c),p.54). Whendealing with adjacency as
a parameter, Moortgat& Oehrlepoint out that there areessentially two situations
thatonemight face(170).
(170) a. �:°-±³²µ´%¶6�¸·c¹��`º »c±¼��·g´��¸·c¹��H¶b. °�±¼��²g´��¸·c¹��H¶hº �:»c±³·g´"¶6�¸·c¹��
Eachstructural rulespecifiestheconfigurationweencounter(theLHS), andthe
configuration thatis requiredfor therule to apply (theRHS).Clearly, thetwo cases
in (170) aresymmetric. In (170a) we find that ² ¹ is combined with ° whereas it
should find its proper place with a substructure of ° , namely ²-´ ; (170b) presents
the opposite ‘movement’.7 Moortgat & Oehrlecharacterize (170a) as caseof a
7A brief remarkabout ‘movement’: Like Steedmanpoints out in the introduction to (1996),we use‘movement’heremetaphorically andnot in the senseof Chomskyan linguistics. Thereisno equivalent of “move-½ ” or alike in categorial type logic, becauseof a different relationto log-ical form. We do admit though that, ‘despite’ adheringto a metaphoricalsenseof movement,thestructuralrulesdealingwith word orderdo have a flavor of moving elementsaround. It is perhapsnoteworthy thatpreciselybecauseof thatflavor, categorial typelogic hasbeenusedasatool to modelMinimalism.
A formalmodelof wordorderasstructuralindicationof informativity /117
dependent ²�¹ beingattractedby thehead ²+´ , whereas(170b) illustratesthecase
of ²¾¹ adjoining itself to thehead²�´ .Formally, Moortgat & Oehrlemake useof structural rules that regulate the
interaction betweenheadedness (distinguishing a head-dependentasymmetry as
proposedby Moortgat& Morril l in (1991)) andadjacency. Axiomatically, this type
of interaction is captured-abstractly- by axiomsthatdefinehow differentmodesare
commutativeandassociativewith respect to oneanother:
(171) Interaction - Mixed commutativity:
�"�:¿J�jÀ�Áh�l�ÃÂ�Ä_��ÅhÆ �"�:¿J�ÃÂ�Ä_�µ��À[Ä1�(172) Interaction - Mixed associativity:
�"�:¿J�jÀ�Áh�l�ÃÂ�Ä_��ÅhÆ �:¿«��ÀÇ�:ÁÈ�ÃÂ�Ä1�"�The É modegenerically represents a modeof composition that enablesnon-
adjacency, whereas Ê modelsadjacent composition in termsof a heads anddepen-
dents. Thus,therulesin (171) and(172)areschemata: Moortgat & Oehrleobtain
modelsfor specific constructionsby appropriately instantiating Ê2�HÉ . Moortgat &
Oehrlediscusstwo examplesthatillustratethehead-wrapping typeof construction
in (170a) andthehead-attraction of (170b).
We canbriefly characterizethe understanding of head wrapping explored in
(Moortgat andOehrle, 1994) as foll ows. Headwrappingenables to elements to
becomeadjacent, starting from a composition in which thesetwo elements were
not adjacent. MoortgatandOehrleexplain this usingthetermsinfix andcircumfix
(or host): the infix syntactically adjoins itself to the headof the circumfix. More
specifically, we have that the infix cango beforeor after thehead of thehost, and
caneitherdeterminethe headof the construction (endocentricity) or combine as
a dependent (exocentricity). Consequently, Moortgat & Oehrledistinguish four
“wrapping” modes, labelled Ë�ÌX�2Ë:ÍÎ�"Ï�ÌÐ�"Ï^Í with Ë"�eÏ� indicating that the infix is to
the left (right), andwith Ì��:Í&� indicating that the infix (circumfix) determinesthe
headof theconstruction.
Theinterestingaspect about themodelthatMoortgat & Oehrlethenpresent is
their distinction between basescasesand recursive cases. The base casestell us
underwhatconditions head-wrapping is equivalent to simpledependency adjunc-
tion. On theother hand, therecursive casesestablish thecommunicationbetween
thewrappingmodesanddependency. They useinstantiationsof (171) and(172) to
allow theinfix to “travel” througha treestructureuntil it is at a landing sitewhich
is characterizedby oneof thebasecases. Their modelis givenin (173).Themode
118¬ A formal model of wordorder asstructural indication of informativity
ËeÑ©�eÏ^Ñ+� is a shorthandfor Ë�Ì or Ë:Í ( Ï�Ì or Ï�Í ).
(173) MOORTGAT AND OEHRLE’ S MODEL OF HEAD WRAPPING
a. Basecases:A1: ¿J��ÒAÁ Ó7Æ ¿«�jÒÕÔ�Á A1’: ¿«��Ö#Á Ó�Æ ¿J��ÒØ×jÁA2: ¿J��Ò&Á Ó�Æ ¿«� Ö ×�Á A2’: ¿J� Ö Á Ó�Æ$¿«� Ö Ô@Á
b. Recursive cases:A3: �"�:¿J��Ö%ÙkÁh�l��ÒÎÄ1� Ó�Æ �"�:¿«�jÒÎÄ1�g�-Ö%ÙOÁh�A4: �:¿J� Ö �:ÁÈ� Ö%Ù Ä1�"� Ó�Æ �"�:¿«� Ö Áh�g� Ö/Ù Ä_�A5: �:¿J� Ö �:ÁÚ�jÒ Ù Ä1�"� Ó�Æ �:ÁÛ�jÒ Ù �:¿ � Ö Ä1�"�A6: �"�:¿ �jÒ Ù Áh�l��ÒÎÄ1� Ó�Æ �:¿«�jÒ Ù �:ÁÚ��Ò�Ä_�"�
Moortgat & Oehrleapply their modelto Dutch verb raising, andin their dis-
cussion they briefly reflecton how their modeldiffers from Steedman’s CCG.As
we already pointed out earlier, CCG modelscross-serial dependencies in Dutch
using crossedcomposition, ��� � and ��� � . Although thesecombinatorsareob-
viously not valid in L, thestructuresthey allow for aretheoremsof LP. However,
if we would indeedmodelcrossedcomposition purely asLP, thenany directional
variant would do.
Steedmanrestrictspossible ruleschemataby meansof two principles: theprin-
ciple of Directional Consistencyandtheprinciple of Directional Inheritance - cf.
(1996), p.42ff. It follows from thesetwo principlesthatany rule in CCGneedsto
obey, or project, thedirectionality specified in the lexicon. On theother hand, the
model in (173) restricts the possible orderings directly in termsof the logic. The
orderspossibleon thebasisof ��� � ����� � arederivable astheoremsof (173),8 as
do theprinciples- thereis no need for their meta-theoretical stipulation.
Moortgat& Oehrlealsobriefly discussthesecondcase,headattraction(170b).
We do not repeat their entire discussion here, asMoortgat& Oehrleonly present
fragmentsto dealwith particular casesof head attraction in Dutch andEnglish.
Rather, we just note a few interestingobservations about their fragments.For ex-
ample,associativity (“re-bracketing”) providestheformal meansto block recursion
upor down trees,yieldingempirical consequencesliketheRightRoofconstraint or
the(im)possibility of dangling prepositions.Furthermore,islandconstraintscanbe
modelledmodelledby combining a head-dependentasymmetrywith associativity.
8In fact, they arederivableon even simplerstructuralrulesthatarecloseto (171) and(172),aswe show in (Kruijf f andBaldridge,2000).
A formalmodelof wordorderasstructuralindicationof informativity /119
To conclude, Moortgat & Oehrle’s modelsof head wrapping andheadadjunction
provide an interesting exampleof how word order canbe modeled in categorial
type logic. Theirsis, naturally, not theonly proposal thathasbeenadvanced- but
it bearsa close resemblanceto for example Hepple’s later work (1996; 1997), and
it providesmoredetail thanFoster’s (1992).
Unsurprisingly, the typesof structural rules that Moortgat & Oehrleemploy
aresimilar to theones we find in thesimulationsof CCG(Kruijf f andBaldridge,
2000)aspresentedabove. Thisunderlinestheobservation wemake in (Kruijf f and
Baldridge, 2000), namelythat the differencesbetweenthe combinatory tradition
and the Lambektradition are for a large part a matterof perspective. Here,we
naturally staywithin theLambektradition - but it is not inconceivable that DGL’s
modelof word orderto bepresentedbelow canbemoreor lessdirectly translated
into for examplea modalizedversion of Set-CCG.
4.3 VARIABILITY OF WORD ORDER IN DGL
Ouraimin thepresent section is to developamodelof basicphenomenawefind in
variability in word order. Particular about themodelis that it hasanarchitecture.
Thearchitecturegivesthemodelaninternalstructurethatdeterminesaprecedence,
or interdependence, amongstructural rules. Figure4.1 givesan overview of the
architecture. Many of the decision points in the architecture arecoveredby the
variation hypothesesof Chapter 3 or thedata assuch.9 Thearchitecture in Figure
4.1 provides the foundations on which we build our modelsof word order as a
structural indicationof informativity, seeÜ 4.4.
For example, consider the branching under Rigid OV. We have several options
here,concerning verbfinal clustering, non-rigid OV behavior, andscrambling. To
illustratehow we work with thearchitecture, consider thesubordinate clausecon-
structionsin Dutch, Flemish,andGermanin (174) through(176).
(174) Dutch, Flemish
omdatbecause
ChristopherChristopher
Kathykath
boekenbooks
wilwants
lerento beteach
lezen.to read.
English “BecauseChristopherwantsto beteach Kathy to readbooks.”
(175) a. Dutch omdatChristopherKathy boeken wil lerenlezen.9The only decisionsnot coveredconcernnon-rigidity in verb-finality, andwhena languagehas
cross-serialor nestedlong-distancedependencies.
120¬ A formal model of wordorder asstructural indication of informativity
Rigid
Rigid SVO
VFinal cluster Non-rigid Mixed OV
Rigid OV
Crossed ordering
Nested Ordering
Discont.CrossedOrdering
Scrambling
Free OV
Rigid VO
Mixed VO
V2-position Mixed SVO
Free SVO
Free SVO (nc)
Non-rigid VFirst VFinal
Figure4.1: Thearchitecture of DGL’sword ordermodel
b. Flemish,Dutch* omdatChristopherwil Kathy boekenlerenlezen.
c. Flemish,Dutch* omdatChristopherwil Kathy leren boeken lezen.
(176) a. German weil ChristopherKathy Bucher zu lesenbeibringenmochte.
As wealready discussedearlier, subordinateclauseshave adifferentdominant
word orderfrom matrix clauses- the formerareSOV, whereas the latterasSVO.
In theexamplesabove weseethatall three languagesdisplay verbraising, leading
to verbfinal clusters.What thegrammarsfor theselanguagesthuscanbethought
to havein commonis apackageof structuralrulesVFinalCluster thatenforcesthe
clause-finalordering of verbsin subordinateclauses. But, continuing this line of
thought, once the verbsareclusteredat the end, the languagesdiffer in how the
verbsareto beordered. Both DutchandFlemishsharea further packageCrosse-
dOrdering, on topof VFinalCluster, thatleads to anordering giving riseto cross-
serial dependencies. Germandoesnot have sucha package,but instead has-in
addition to VFinalCluster- a packageNestedOrdering (sharedwith for example
Japanese)thatorderstheverbsin sucha way that wegetnesteddependencies.
Thus,to sumup,we can“instantiate” therelevant partof thearchitecture asin
Figure4.2. Naturally, oncewe have definedthementionedpackagesof structural
rules, we canrecast the picture asa cross-linguistic network like we discussedin
A formalmodelof wordorderasstructuralindicationof informativity /121
Chapter1.
Rigid OV VFinal Cluster
Nested
CrossedOrdering
Ordering German
DutchDiscontinuity
Flemish
Figure4.2: Dutch, FlemishandGermanverbraising
4.3.1 PREL IMINARIES TO THE FORMULATION OF THE PACKAGES
Before we get to the formulation of the packagesof structural rules, I need to
clarify afew general strategiesthatI follow. Firstof all, following Moortgat(1999)
andSteedman(1996; 2000c) we distinguish different clause levels by meansof
features.Therelevantunary modaloperatorsaregivenin (177).
(177) a. Ý�Þ�ßHÒÕà , clause level, áâË�ã1äæå�çkè/é��¸ã�ê7ë�ì .b. Ý@Þâí-î¼ï , matrix clauselevel, çkè/é`ðñáâË�ãc. Ý@Þ�àHò�ó , subordinateclauselevel, ã�ê7ë�ðñáâË�ã
The different unary modal operatorsin (177) help us to obtain a modularity
in defining structural rules. Constraints on the required ordering at a particular
clauselevel canthenobtainedby a linkagerule like ô:¿�õ à6òLó Ó�Æ ô:¿�õYöZ÷ ÀRø^ù Ò , which
specifiesthat theorder in subordinateclauseis verb-final.
Secondly, I adhere to a particular encoding of the formal namesof the struc-
tural rulesdiscussedhere. Thegeneral format is PackageName [.SubPackage]
Number . Description, whereby Description cantakethefoll owing form: çú�eé[� =“move” structure é , ûÇ��ü7�"ç�Í&� = “percolate” feature ü over mode ç�Í , Í���ü7�"ç�ÍA� =“distribute” feature ü overmodeç�Í . Forexample,VFinal.XDep1.p(vhead,mod)
is a rule in theCrossedDependencies(XDep) subpackage of theVerb-Final Clus-
ter (VFinal) package,specifying thepercolation of thevhead feature over a struc-
ture built using mode ç�ý^Í . Theseformal namesare usedin proofs, but we -
naturally- give a more elaborate description of the ideasbehind eachrule when
introducingit.
Finally, to keepthe definitions of the packagesreasonably short, we usually
omit statementsof structural rulesfor modesotherthan ã�á , Íá , and ÊHá - like á , èYç�þ
122¬ A formal model of wordorder asstructural indication of informativity
or ã"û7þ . Rulesfor the latter modesaresimply different instantiations of the same
structureasused for ã�á , Íá , and ÊHá .
4.3.2 OV PACKAGES
TheOV packagesdefinebehavior thatis, possibly, availablein languagesthathave
OV asdominant word order at someor all clauselevels. For example, Japanese
displays OV behaviorat both the matrix clauseandthe subordinate clause levels,
whereas Dutch andGermanonly useOV at the subordinate clause level, having
SVO asthedominantorderof thematrix clause.
Below we present the various packagesrelevant to modeling aspects of OV
word order.Naturally, a languageonly employsaproper subset of thesepackages.
Moreover, evenwhena languagedoes employ a particular packageit need not be
thecasethatall rulesareused - for example, if a languagehasnomodalauxiliaries,
any rule definingthebehavior of the ç�ý�Í modeis superfluous.10
The architecture in Figure 4.1 in fact shows what packagescan be taken in
conjunction (&) andwhich only in (exclusive) disjunction (V). Thus,a language
usually eitherhasCrossed Dependenciesor NestedDependencies.
Themodeling of OV wordorderstartsfrom theassumptionthat wehavelexical
function categoriesthattake their argumentsto theleft. For example,with Subjthe
subject,DObj thedirect object, andVerb theverb,wehavethefollowing template-
like structuresfor OV:
(178) Prototypical OV canonical structure: �6ÿÇê[ë%ÉN��� àYß � ��� ë%Éñ� � ׸ß�����Ï�ë��"�
Definition 14 (Verb Final Clusters, (VFinal)). TheVFinal package definesthe
ordering of verbs towards the end of a clause. The exact ordering is of verbs
within thecluster is definedin thepackagesXDep, NDep, andMxDep depending
on language-specific behavior. All thesepackagesextend thebehavior specifiedby
VFinal. TheVFinal package comprisestherules givenin (179).
10At least,it is superfluous in thesensethatit is not usedin derivationsfor thatlanguage.
A formalmodelof wordorderasstructuralindicationof informativity /123
(179)
ô:¿+õ à6òLó ÆÈô:¿�õYöU÷ À ø�ù Ò ± ÿÇê[ë ÊYã&ü7Ê��Ðþ&Ë�¶ô:¿+õ í#î¼ï ÆÈô:¿�õYöU÷ À ø�ù Ò ±� þè/Ï^Ê6écÊYã�Aü7Ê��Ðþ&Ët¶
¿z��� àHß ô:Áhõ öZ÷ À ø^ù Ò Æ«¿ñ��� àHß ô:Áhõ ö Ô�� ù × ±����+Ê��Ðþ&Ë���� Ë"��&Ì���þ Í���&ü7Ê��Ðþ&Ë6�H¶¿ñ� � ×2ß�ô:ÁhõYöZ÷ À ø^ù Ò Æ«¿ñ� � ×2ß�ô:ÁhõYö Ô�� ù × ±����+Ê��Ðþ&Ë���� Ë"��&Ì���þ Í���&ü7Ê��Ðþ&Ë6�H¶¿z���ÎÀ ß ô:Áhõ öZ÷ À ø^ù Ò Æ«¿ñ���ÎÀ ß ô:Áhõ ö Ô�� ù × ±����+Ê��Ðþ&Ë���� Ë"��&Ì���þ Í���&ü7Ê��Ðþ&Ë6�H¶ô:¿ñ� � àHß ÁhõYöZ÷ À ø^ù Ò Æ«¿ñ� � àHß ô:ÁhõYöZ÷ À ø^ù Ò ±����+Ê��Ðþ&Ë! "� ûÇ��&ü7Ê#�Ðþ Ë/���NãLá��H¶ô:¿ñ� � ×2ß"ÁhõYöZ÷ À ø^ù Ò Æ«¿ñ� � ×2ß�ô:ÁhõYöZ÷ À ø�ù Ò ±����+Ê��Ðþ&Ë! "� ûÇ��&ü7Ê#�Ðþ Ë/���ñÍ á��H¶ô:¿z���ÎÀ ß Áhõ öZ÷ À ø^ù Ò Æ«¿ñ���ÎÀ ß ô:Áhõ öZ÷ ÀRø^ù Ò ±����+Ê��Ðþ&Ë! "� ûÇ��&ü7Ê#�Ðþ Ë/��� ÊHá��H¶
�Remark 13 (Description of the VFinal package). The general strategy is as
follows. For a clause to be well-formed, we in general require it to be either
Ý Þ í#î¼ï ÿ (matrix clause)or Ý Þ à6ò�ó"ÿ (subordinate clause). In a languagethat has
verb final ordering at a particular clause level, we include ± ÿÃê7ëcÊ/ã$&ü7Ê��Ðþ&Ë�¶ or
±� þ èYÏ^Ê6é¾ÊYã&ü7Ê��Ðþ&Ë�¶ (or both) to mirror thatrequirement.
Eachof theserules specifies that if a structure ¿ is a particular type of clause
( ã�ê7ë�¢LçkèYé ), then it hasto be &ü7Ê#�Ðþ Ë . Then,starting with the VFinal1 rules,we
seethat for a structure composed out of ¿ and Á to be &ü7Ê#�Ðþ Ë , we needto have
thatthesubstructure Á needsto be &ü7Ê��Ðþ&Ë . Becauseof thedirection of theheaded
modes(pointing % to thedependenton theleft), we know thattheverbal headhas
to beto theright - corresponding to OV.
In thesimplestcase,defined by theVFinal0 rules, we have that theverbfinal
cluster is formedby the verbal head itself. The examples in (180) illustratesuch
cases,whereas the derivation (181) shows how the goal category ÝÎÞ�àHò�ó2ÿ canbe
derivedfor subordinate clausesasin (180).11
(180) a. Dutch
...(dat)ChristopherChristopher
boekenbooks
leestreads
“... (that)Christopherreads books.”
b. German
...(daß)ChristopherChristopher
Bucherbooks
leßt.reads
“... (that)Christopherreads books.”
(181)11Notethat,for brevity, weglossover any morphological strategiesandfunctionalinterpretation.
124¬ A formal model of wordorder asstructural indication of informativity
subj &�®ÃÞ�ù ß î'�dobj &B® Þ)( ù2î'�
tverb &B®ÃÞ ö Ô�� ù ×�%® Þ ( ù2î'��¥ � ×2ß��%®#Þ�ù ß î'��¥*� àYß ã��"�ô tverbõ�ö Ô�� ù × &4® Þ+( ù2î���¥ � ×2ß��%® Þ ù ß î'��¥*� àHß ã^� ± ® Þ�,�¶dobj � � ׸ß�ô tverbõ ö Ô�� ù × &�® Þ�ù ß î'��¥*� àHß ã ± ¥-,+¶
subj ��� àHß � dobj � � ×2ß�ô tverbõ6ö Ô�� ù × �.&kã ± ¥-,+¶subj � � àHß � dobj � � ×2ß�ô tverbõ6öZ÷ À ø^ù Ò �/&Oã ±��0�+Ê��Ðþ&Ë���� Ë"��AÌ���þ&ÍÎ��Aü7Ê��Ðþ&Ë��H¶subj ��� àHß ô dobj � � ×2ß tverbõ öZ÷ À ø^ù Ò &kã ±��0�+Ê��Ðþ&Ë1 "� ûÃ��&ü7Ê��Ðþ&Ë/���NÍá��H¶ô subj ��� àHß � dobj � � ×2ß tverb�"õ6öZ÷ À ø^ù Ò &Oã ±��0�+Ê��Ðþ&Ë1 "� ûÃ��&ü7Ê��Ðþ&Ë/����ãLá��H¶ô subj � � àHß � dobj � � ×2ß tverb�"õ àHò�ó &kã ± ÿÇê[ë ÊYã2Aü7Ê��Ðþ Ë�¶
subj �� àYß � dobj � � ×2ß tverb�.&B®ÃÞLàHò�ó%ã ± ® Þ�3 ¶
Definition 15(VFinal, Crossed Ordering (VFinal.XDep)). Thefirstextensionto
theVFinal package weconsider here is theXDep package which modelsthetype
of ordering in a verbal cluster that gives rise to cross-serial dependencies. The
XDep package comprisestherulesasgivenin (182).
(182)
Á � � ×2ß��:¿ñ� �Îí54 ×LÄ1�ÇÆ«¿z� �Îí54 ×��:Á � � ×2߸Ä1� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�eç�ý^ÍA�H¶Áæ� � ×2ß��:¿ñ����ù ò ïÄ1�ÇÆ«¿z����ù ò ïÎ�:Á � � ×2߸Ä1� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�:þ ê�éX�H¶Á � �ÎÀ ß �:¿ñ� �Îí54 ×LÄ1�ÇÆ«¿z� �Îí54 ×��:Á � �ÎÀ ß Ä1� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�eç�ý^ÍA�H¶Á ���ÎÀ ß �:¿ñ����ù ò ïÄ1�ÇÆ«¿z����ù ò ïÎ�:Á ���ÎÀ ß Ä1� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�:þ ê�éX�H¶Á ��� àHß �:¿ñ��� (�� Ö Ä1�ÇÆ«¿z��� (�� Ö �:Á ��� àHß Ä_� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�Õû8��Ï�H¶Á � � ×2ß��:¿ñ��� (�� Ö Ä1�ÇÆ«¿z��� (�� Ö �:Á � � ×2߸Ä1� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�Õû8��Ï�H¶Á ���ÎÀ ß �:¿ñ��� (�� Ö Ä1�ÇÆ«¿z��� (�� Ö �:Á ���ÎÀ ß Ä_� ±��0�+Ê��Ðþ&Ë1��6 � �%û7���³çú�Õû8��Ï�H¶ô:¿z� ��ù ò ï ÁhõYö Ô�� ù × Æ«¿z� ��ù ò ï ô:ÁhõYö Ô�� ù × ±��0�+Ê��Ðþ&Ë1��6 � �%û9 "� ûÃ��&Ì���þ Í��2þê�é[�H¶ô:¿ñ� �Îí54 ×�ÁhõYö Ô�� ù × Æ«¿z� �Îí54 ×�ô:ÁhõYö Ô�� ù × ±��0�+Ê��Ðþ&Ë1��6 � �%û9 "� ûÃ��&Ì���þ Í��"ç�ý^Í&�H¶ô:¿ ��� (�� Ö Áhõ ö Ô�� ù × Æ«¿z��� (�� Ö ô:Áhõ ö Ô�� ù × ±��0�+Ê��Ðþ&Ë1��6 � �%û9 "� ûÃ��&Ì���þ Í��:û:��Ï�H¶
�Remark 14 (Description of the XDep package). The XDep packageextends
the VFinal package by determining the exact order of auxiliaries, modal verbs,
modal infinitives and the verbal head(possibly an infinitive itself) in the verbal
cluster. TheXDep0 structural rulesenable thecluster to beformed- andXDep1
imposes the requirementfor that the ordering, extending the VFinal1 rules. The
examples in (183) belowillustratein moredetail the kind of phenomenacovered
by VFinal+XDep.
(183) a. Dutch
A formalmodelof wordorderasstructuralindicationof informativity /125
...(dat)ChristopherChristopher
KathyKathy
wildewanted
kunnento beable to
kussen.kiss
“...(that)Christopherwantedto beableto kissKathy.”
b. Dutch
...(dat)ChristopherChristopher
KathyKathy
ElijahElijah
zagsaw
willento want
kussen.to kiss
“...(that)Christophersaw Kathy wanted to kissElijah.”
Now let us have a closer look at the derivations. In the derivations,we make
useof aslightly moreabstractlexiconwhosefunction categoriesaregivenin (184).
(184)
inf�<;>=@?)ACB�D+ELr#;F=HGHD!I�J ¬-K E�L¸r#;F=MD)L�I�J ¬-K�N L1OHP�J�Q�|H|
mod�r�;>=�D)L�I�J ¬-KRN L!OU|1S K�T�U ELr�;F=MD)L�I�J ¬-KRN L!OHP�J�Q�|
aux�r#;>=MD+L�I�J ¬"K�N L1Oâ|VS K D�WCXr#;F=MD)L�I�J ¬-KRN L!O+P�J�Q�|
modi�r#;>= D)L�I J ¬ K�N L OHP�J�Q�|VS KYT.U E r�;9= D)L�I J ¬ K�N L OHP�J�Q�|
tverb�<;>= ?)ACB�D)E r�;9= GHD!I J ¬ K E�L r�;F= D)L�I J ¬ KRN L Oâ|H|
perc Z*[ ;>=�D)L�I�J ¬-KRN L1O@\1S K G+B�]HOHP�J�Q
Then,for (183a) thederivationis asin (185). Weleaveout thefirst elimination
steps,asthese aretrivial.
(185)
...subj ^5_�`�a@[ aux ^5_�b)c@de[ modi ^5_�f�g1hi[ dobj ^5_Rh�aHj inf kVl)m@n b)h \#\V\Fo0prq s�t�usubj ^ _�`�a [ aux ^ _�b)c@d [ dobj ^ _�h�a [ modi ^ _Yf.g1h j inf kVl)m@n b)h \#\V\Fo0p q v�w�x�y{z"|#} ~��0�V�Y�<} � [ ���M�
\ usubj ^5_�`#aH[ aux ^2_�b�cCd*[ dobj ^2_�h�aHj modi ^5_�f.gVh inf kVl)m@n b)h \#\�o�p q v�w�x�y{z"|#} ~��0�V�7��} � [��"� �Cz-�e�#���M� \ usubj ^ _�`#a [ dobj ^ _�h�a [ aux ^ _Rb�cCd j modi ^ _�f.gVh inf kVl)m@n b)h \#\�o�p q v�w�x�y{z"|�} ~����1�e�*} � [ z-�Y�
\ usubj ^5_�`#aH[ dobj ^2_�h�a@j aux ^5_Rb�cCd*[ modi ^5_�f.gVh inf
\ k#l)mCn b)h \�o�p q v�w�x�y{z"|�} ~����1���i} � [���� �Cz"�e�Vz-�e�\ u
subj ^5_�`#aH[ dobj ^�_�h�a@j aux ^5_Rb�cCd*[ modi ^5_�f.gVh inf\ k l)�@��� b!� \�o�p�q vw�x�y{z"|��<} | [��"� �Cz"�*� �"� x�y{z"|
\ usubj ^5_�`#aHj dobj ^�_�h�a@[ aux ^5_Rb�cCd*[ modi ^5_�f.gVh inf
\V\ k#l)�@��� b!� o�p q vw�x�y{z"|���} � [��"� x�y{z"|��H���i�\ u
j subj ^5_�`#aH[ dobj ^�_�h�a@[ aux ^5_Rb�cCd*[ modi ^5_�f.gVh inf\V\#\ kVl)�@��� b�� o0p q v�w�x�y{z"|#��} � [���� x�y{z"|��+�
p � \ uj subj ^ _�`�a [ dobj ^ _�h�a [ aux ^ _�b�cCd [ modi ^ _Yf.g1h inf
\#\V\ k `�cC o�p q ¡8�R¢:x p �"� x�y{z"|£usubj ^5_�`�a@[ dobj ^2_�h�aH[ aux ^2_�b�cCde[ modi ^2_Yf.g1h inf
\#\V\Fo�¤9¥ `'cM p q¤9¥@¦ u
Theexamplein (183b) illustratesDutchcross-serialdependenciesin awayit is
usually found in theliterature.Thederivationin VFinal+XDep is givenin (186).
126s A formal model of wordorder asstructural indication of informativity
(186)
...subj ^5_�`�a@[ perc ^2_<§ n
] [ subj ^�_R`�a@[ modi ^5_�f�g1hi[ dobj ^5_Rh�aHj inf kVl)m@n b)h \#\V\#\Fo�p q�sMt�usubj ^ _�`�a [ perc ^ _<§ n
] [ subj ^ _R`�a [ dobj ^ _�h�a [ modi ^ _Yf.g1h j inf k l)m@n b)h \#\V\#\Fo�p q ~����1�e�*} � [ ���M�\ u
subj ^ _�`#a [ perc ^ _*§ n] [ subj ^ _�`#a [ dobj ^ _�h�a j modi ^ _�f.gVh inf kVl)m@n b)h \#\V\Fo0p q ~����1���i} � [���� �Cz"�*�V���M� \ u
subj ^ _�`#a [ subj ^ _R`�a [ perc ^ _*§ n] [ dobj ^ _�h�a j modi ^ _�f.gVh inf kVl)m@n b)h \#\V\Fo0p q ~��0�V�Y�<} � [ �e�@¨
\ usubj ^5_�`#aH[ subj ^�_R`�a@[ dobj ^5_Rh�a+[ perc ^5_*§ n
] j modi ^5_�f.gVh inf kVl)m@n b)h \#\V\Fo0p q ~��0�V�Y�<} � [ �e�@¨\ u
subj ^5_�`#aH[ subj ^�_R`�a@[ dobj ^5_Rh�a+j perc ^5_*§ n] [ modi ^5_�f.gVh inf
\ k#l)mCn b)h \V\Fo0prq ~��0�V�7��} � [��"� �Cz-�e���Y�@¨\ u
subj ^5_R`�a@[ subj ^�_R`�a@[ dobj ^5_Rh�a+j perc ^5_*§ n] [ modi ^5_�f.gVh inf
\ k#l)�@��� b!� \V\Fo0p q | [���� �Cz"�*� �"� x�y{z"|\ u
subj ^5_R`�a@[ subj ^�_R`�a@j dobj ^5_Rh�a+[ perc ^5_*§ n] [ modi ^5_�f.gVh inf
\V\ k#l)�@��� b!� \Fo0p q � [���� x�y{z"|#�+�©�-�\ u
subj ^ _R`�a j subj ^ _R`�a [ dobj ^ _Rh�a [ perc ^ _*§ n] [ modi ^ _�f.gVh inf
\V\#\ kVl)�@��� b�� o0p q � [���� x�y{z"|#�+�p � \ u
j subj ^ _R`�a [ subj ^ _R`�a [ dobj ^ _Rh�a [ perc ^ _*§ n] [ modi ^ _�f.gVh inf
\V\#\#\ k#l)�@��� b!� o�p q � [���� x�y{z"|��+�p � \ u
j subj ^5_�`�a@[ subj ^5_�`#aH[ dobj ^2_�h�aH[ perc ^2_<§ n] [ modi ^2_Yf.g1h inf
\#\V\#\ k `'cM o�p q ¡8��¢ªx p ��� x�y{z-|�usubj ^5_�`�a@[ subj ^5_�`#aH[ dobj ^2_�h�aH[ perc ^2_<§ n
] [ modi ^2_Yf.g1h inf\#\V\#\Fo�¤9¥ `'cM p q
¤F¥C¦ u
As it turns out, the packagesVFinal and XDep take a similar approach to
capturing cross-serial dependenciesasthemodelproposedby Moortgatin (1999).«In theexamplesin (175)weillustrated apeculiar contrastbetweenFlemishand
Dutch,whereit concernstheverbfinal cluster.12 In (standard)Dutch,theverbfinal
clustermustbecontinuous- wecannot haveverbalcomplementsinterspersedwith
theverbsmaking up theverbal cluster.
Flemish,on the otherhand, doesallow for that, giving rise to the possibility
of what we call herediscontinuous crossed ordering. The ordering amongthe
componentsmaking up the verb final cluster is still the sameasin Dutch,but we
mayhave that -for example-thePatient (directcomplement) or anAddresseeor
Beneficiary (indirectcomplement) appears inbetweentheauxiliary andtheverbal
head. Thestructuralrulesin theDXDep packageextendtheVFinal.XDep package
to allow for thesealternative orderings.
Definition 16(VFinal, DiscontinuousCrossedOrdering (VFinal.XDep.DXDep)).
TheDXDep package is an extension of VFinal’s XDep package, and covers the
construction of discontinuous verb final clusters with a crossedordering. The
DXDep monotonically extendsXDep with therulesgiven in (187) below.
(187)ô�¬®�¯�°)±�²hõ ö+³�´1µ °�¶ ¬®�¯�°)±�ô�²hõ ö+³�´Vµ ° ·�¸�¹»º�¼�½Y¾V¿�ÀÂÁ$Ã�ÄF¿�ÁÅÀÆÁÇÃ!Ä9È"¿ Ä>É�Ê�Ë�Ã�½eÌ�Í)Ì*Î�Ï#Ðô�¬® ¯{Ñ£± ²hõYö+³�´1µ ° ¶ ¬® ¯{Ñ£± ô�²hõYö+³�´1µ ° ·�¸�¹»º�¼�½Y¾V¿�ÀÂÁ$Ã�ÄF¿�ÁÅÀÆÁÇÃ!ÄF¿�È"¿ Ä>É�Ê�Ë�Ã�½eÌ�Í�º#Î�Ï#Ð
12I would like to thankMichael Moortgatfor pointingthis out to me.
A formalmodelof wordorderasstructuralindicationof informativity /127
«
Remark 15 (Explanation of the DXDep package). TheDXDep packagesimply
follows out the strategy we developed in the XDep. Namely, the distributional
characteristics of the vhead andvfinal featuresdefinethosestructuresto be well-
formedthatallow for proper distribution of thefeaturesover themodalconfigura-
tionsmakingup these structures. If thevhead/vfinal featurescannot bedistributed
over a particular construction, it is ill-for med- at leastfrom theviewpoint of these
packages.
TheDXDep relaxesthedistributional characteristics of the features.Thefea-
turesnot only distribute over continuous verb final clusters,asdeterminedby the
XDep package- therulesgivenin (187)allow now for distribution over the Ì*Î andº#Î modesaswell. In conjunction with the structural rules in the XDep package
this givesrise to thepossible formation of discontinuousclusters,astheexamples
below illustrate.
(188) a. Flemish
...(dat)ChristopherChristopher
wilwants
boekenbooks
lezen.to read.
“...(that)Christopherwantsto readbooks.”
b.
...subj ^5_�`�a@Ò aux ^5_Rb�cCd*Ò dobj ^2_�h�a@j inf k#l)mCn b)hMÓVÓ o�p q s�t�usubj ^�_R`�a@Ò aux ^2_�b�cCdej dobj ^5_Rh�a inf k#l)mCn b)hMÓ o�pÔq v�w�x�y{z-|#} ~Å���1��} �»~����1���i} � Ò���� �Cz"�e�V�-� Ó usubj ^ _R`�a j aux ^ _�b�cCd Ò dobj ^ _Rh�a inf Ó k l)m@n b)h o�p q v�w�x�y{z"|�} ~����1���i} � Ò���� �Cz"�e�Vz-�e�
Ó usubj ^�_R`�a@j aux ^2_�b�cCdeÒ dobj ^5_Rh�a inf Ó kVl)�H��� b!� o�p q v�w�x�y{z"|��*} | Ò���� �Cz"�*� �"� x�y{z"|
Ó uj subj ^ _R`�a Ò aux ^ _�b�cCd Ò dobj ^ _Rh�a inf Ó#Ó k#l)�@��� b!� o�p q v�w�x�y{z"|#��} � Ò���� x�y{z"|��+�
p � Ó uj subj ^5_�`#aHÒ aux ^2_�b�cCd<Ò dobj ^2_�h�a inf ÓVÓ k `'cM o�p q ¡ª��¢:x p �"� x�y{z"|£u
subj ^ _�`#a Ò aux ^ _Rb�cCd Ò dobj ^ _�h�a inf ÓVÓ o�¤ ¥ `'cM p q¤9¥@¦ u
(189) a. Flemish
....(dat)KathyKathy
wilwants
kunnento beableto
SanskritSanskrit
schrijven.to write
“...(that)Kathy wantsto beableto write Sanskrit.”
128s A formal model of wordorder asstructural indication of informativity
b.
...subj ^5_R`�a@Ò aux ^2_�b�cCdeÒ modi ^2_Yf.g1h�Ò dobj ^2_�h�aHj inf k#l)mCn b)hMÓVÓ#Ó o�p q�sMt�usubj ^5_�`�a@Ò aux ^5_Rb�cCd*Ò modi ^5_�f�g1h-j dobj ^5_Rh�a inf k l)mCn b)h ÓVÓ o�pÕq �»~����1�7��} � Ò���� �@z"�e�V�-� Ó usubj ^5_�`�a@Ò aux ^5_Rb�cCd*j modi ^5_�f�g1h-Ò dobj ^5_Rh�a inf Ó kVl)m@n b)hMÓ o�p q ~Å���1����} � Ò��"� �Cz"�*�V���M�
Ó usubj ^5_�`�a@j aux ^5_Rb�cCd*Ò modi ^5_�f�g1h-Ò dobj ^5_Rh�a inf Ó#Ó k#l)mCn b)h o�p q ~Å���1����} � Ò��"� �Cz"�*�1zi�Y�
Ó usubj ^ _�`�a j aux ^ _Rb�cCd Ò modi ^ _�f�g1h Ò dobj ^ _Rh�a inf Ó#Ó k#l)�@��� b!� o�p q v�w�x�y{z-|Ö�<} | Ò��"� �Cz-�e� ��� x�y{z"|
Ó uj subj ^5_�`�aCÒ aux ^5_Rb�cCd*Ò modi ^5_�f�g1h-Ò dobj ^5_Rh�a inf Ó#ÓVÓ kVl)�H��� b!� o�p×q v�w�x�y{z"|��i} � Ò���� x�y{z"|#�+�
p � Ó uj subj ^ _R`�a Ò aux ^ _�b�cCd Ò modi ^ _Yf.g1h Ò dobj ^ _�h�a inf Ó#Ó#Ó k `'cM o�p q ¡8�R¢8x p ��� x�y{z"|£u
subj ^/_R`�aCÒ aux ^2_�b�cCdeÒ modi ^2_Yf.g1hiÒ dobj ^2_�h�a inf Ó#Ó#Ó o�¤>¥ `'cM p q¤F¥C¦ u
«Another way in which languagesmay order the verb final cluster is a nested or-
dering, leading to nesteddependencies. AmongGermaniclanguages for example
Germanhasa nestedordering. This type of ordering is brought about by placing
the verbsmakingup the cluster after theverbal head, rather thanbefore asin the
caseof a crossedordering. The examples in (190) below illustratethe contrast
between a Dutchcrossedordering anda Germannestedordering.
(190) a. Dutch
...(dat)KathyKathy
SanskritSanskrit
wilwants
kunnento beableto
schrijvento write
“...(that)Kathy wantsto beableto write Sanskrit.”
b. German
...(daß)KathyKathy
SanskritSanskrit
schreibento write
konnento beableto
will.wants
“...(that)Kathy wantsto beableto write Sanskrit.”
TheNDep packageextendstheVFinal packageto coverconstructionslike (190b).
Definition 17(Verb Final Clusters,NestedOrdering (VFinal.NDep)). TheNDep
package monotonically extendsthe VFinal package, and models nested ordering.
Thepackage consistsof thestructural rules givenin (191).
A formalmodelof wordorderasstructuralindicationof informativity /129
(191)
²Ø�¯�°)± É�Ù 5Ú5Û!°CÜF¬ Ï9¶ ¬Ý�¯{Ú5Û!° É ²Ø�¯�°)± Ù�Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�à�â"ÌYÏ#в㠯{Ñ£± É�Ù 5Ú5Û!°CÜF¬ Ï9¶ ¬Ý�¯{Ú5Û!° É ²Ø ¯{Ñ£± Ù0Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�à�â"ÌYÏ#вØ�¯�°)± É�Ù µ)ä�å Ü ¬ Ï9¶ ¬Ý ¯ µ)ä�å É ²Ø�¯�°)± Ù�Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�½eæ�ç8Ï#Ð²Ø ¯{Ñ£± É�Ù µ)ä�å Ü ¬ Ï9¶ ¬Ý ¯ µ)ä�å É ²Ø ¯{Ñ£± Ù0Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�½eæ�ç8Ï#вØ�¯�°)± É�Ù �è ´Vé Ü ¬ Ï9¶ ¬Ý ¯ è ´Vé É ²Ø�¯�°)± Ù�Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�Ä:Ã�ê<Ï#Ð²ë ¯{Ñì± É�Ù è ´Vé Ü ¬ Ï9¶ ¬Ý ¯ è ´Vé É ²Ø ¯{Ñ£± Ù0Ï ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä7ß�¿�àáÉ�Ä:Ã�ê<Ï#Ðô�¬® µ)ä�å Ü ²hõYö+³�´Vµ ° ¶ ô�¬+õYö+³�´1µ ° µ)ä�å Ü ² ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä9È"¿ Ä>É�Ê�Ë�Ã�½YÌ{Í)½eæ�çªÏ#Ðô�¬®2Ú5Û!°CÜF²hõ ö+³�´Vµ °�¶ ô�¬+õ ö+³�´1µ ° �Ú5Û!°CÜF² ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä9È"¿ Ä>É�Ê�Ë�Ã�½YÌ{Í�à�â"Ì�Ï#Ðô�¬®�è ´Vé Ü ²hõYö+³�´Vµ ° ¶ ô�¬+õYö+³�´1µ ° .è ´#é Ü ² ·�¸0¹�º�¼�½e¾1¿�ÞÂÁ$Ã�Ä9È"¿ Ä>É�Ê�Ë�Ã�½YÌ{Í�Ä8ÃMê*Ï#Ð
«
Remark 16 (Explanation of the NDep package). Thereis little to explain about
the NDep package,asit follows exactly the samepattern of thinking asdoes the
XDep package.Theonly concretedifferencesherearetheplacementof thecom-
ponentsmakingupthecluster: ratherthanbeing placed beforetheverbalhead, they
areplacedafterit. Therulesin NDep arethus, in otherwords,themirror imageof
thestructural rulesin XDep. Theexamplebelow, repeating(190b), illustratesthe
useof theNDep package.
(192) a. German
...(daß)Kathy Sanskrit schreibenkonnenwill.
b.
...subj ^5_�`�a@Ò aux ^5_Rb�cCd*Ò modi ^5_�f.gVhiÒ dobj ^5_Rh�a+j inf kVl)mCn b)hMÓ#ÓVÓ o�p q s�t�usubj ^ _�`�a Ò aux ^ _Rb�cCd Ò dobj ^ _�h�a Ò#j inf kVl)m@n b)h ^ f.g1hHí modiÓVÓ#Ó o�p q î����1�e�<} � Ò ���M� Ó usubj ^5_R`�a@Ò aux ^2_�b�cCd<Ò dobj ^5_Rh�a+j inf ^�f.gVhHí modikVl)mCn b)hMÓ#Ó o�pÔq î����1����} � Ò��"� �Cz"�*�V���M� Ó usubj ^ _R`�a Ò dobj ^ _Rh�a ÒVj inf ^ f.g1hHí modik l)mCn b+h ^ b�cCd�í auxÓ#Ó o�p q îï���1�Y�<} � Ò zi�Y�
Ó usubj ^5_R`�a@Ò dobj ^5_Rh�aHjVÒ inf ^�f.g1hHí modiÓ ^/b�cCd�í auxkVl)mCn b)hMÓ o�p q îï���1�7��} � Ò���� �@z"�e�Vz-�Y�
Ó usubj ^ _R`�a Ò dobj ^ _Rh�a jVÒ inf ^ f.g1hHí modiÓ ^ b�cCd�í auxkVl)�@��� b!��Ó o�p q v�w�x�y{z"|��*} | Ò���� �Cz"�*� �"� x�y{z"|
Ó usubj ^5_R`�a@j dobj ^5_Rh�aHÒVÒ inf ^�f.g1hHí modiÓ ^.b�cCd�í auxÓ kVl)�H��� b!� o�p q v�w�x�y{z"|��i} � Ò���� x�y{z"|#�+���-�
Ó uj subj ^5_R`�a@Ò dobj ^5_Rh�aHÒVÒ inf ^�f.g1hHí modiÓ ^.b�cCd�í auxÓVÓ kVl)�@��� b!� o�p�q v�w�x�y{z-|#��} � Ò��"� x�y{z"|��H�
p � Ó uj subj ^5_�`�a@Ò dobj ^5_�h�a@Ò#Ò inf ^�f.gVhHí modiÓ ^/b�cCd�í auxÓ#Ó k `'cM o�p q ¡ª��¢:x p ��� x�y{z"|�u
subj ^5_�`�a@Ò dobj ^2_�h�a@Ò#Ò inf ^�f.gVhHí modiÓ ^/b�cCd�í auxÓ#Ó oï¤>¥ `'cM p q¤9¥C¦ u
«Example(Dutch, Flemish,and German subordinateclauses). In thedefinitions
we gave above we stressedthe fact that packagesprovide monotonic extensions.
They thusact asbuilding blocks - fragments that we canuseto build grammars
130s A formal model of wordorder asstructural indication of informativity
to cover a language.Moreover, becauseof a modularity that modelsaspects that
languagesmay have in common,we caneasilybuild multilingual fragments(cf.
Chapter 1). Using Dutch, FlemishandGermanexampleswe already illustrated
how the VFinal, XDep, DXDep andNDep packageswork. Taken together, we
cancreate a multilingual fragment for Dutch ( Á ), Flemish( ¹ ) andGerman( ð )
subordinate clauseword order,asshownin Figure4.3.
[VFinal0.*] ñVò�ó ôeó õ:ö[VFinal1.*] ñVò�ó ôeó õ:ö[VFinal.XDep0.*] ñVò�ó ô7ö[VFinal.XDep1] ñ#ò.ó ô7ö[VFinal.XDep.DXDep0.*] ñ#ô:ö[VFinal.XDep.DXDep1.*] ñVô7ö[VFinal.NDep0.*] ñ#õ8ö[VFinal.NDep1.*] ñVõ:ö
Figure4.3: A multilingualnetwork for subordinateclauseWO in Dutch,FlemishandGerman
Thereare-naturally- morepackagesfor modeling OV behavior, aswe saw in
Figure 4.1 on page120. Below we definethe NrOV packagethat defines non-
rigid verbfinality, which is basedontheMxOV packagethatdefinesscramblingor
mixedword order in OV languages.Finally, we definetheFreeOV package(also
extending MxOV) that allows for free word order in OV languages, like Turkish
(Hoffman,1995a).
Definition 18 (Mixed OV ordering (VFinal.MxOV)). TheMxOV package de-
finesmixedword order for OV languages. For thebasic verb final control mecha-
nisms,MxOV makesuseof VFinal (which it thus extends). MxOV comprisesthe
structural rules givenin (193).
(193)
j�÷�k h�b!ø ^�_ � aHjVj�ù2k b)a'a ^�_�h�a!úk�l)�@��� b!�Rû jÖù5k b)a'a ^�_Rh�a+jVj�÷�k h�b!ø ^�_ � a!úk�l)�@��� b!� q üý�Yþv��<} �$ÿÖ�i�M�Vx'� Ój�÷�k h�b!ø ^ _ � a j#j�ù2k�� gVf ^ _�`�a úk�l)�@��� b!�Rû jÖù5k#� g#f ^ _�`�a jVj�÷�k h�b!ø ^ _ � a úk�l)�@��� b!� q üý�Yþv��<} �$ÿ p ���#x'� Ój#jÖ÷�k h�b�ø ^ _ � a ÿ jÖù5k b)a'a ^ _Rh�a ú Ó k l)�@��� b!� û jVj�ù2k b)a'a ^ _�h�a ÿ jÖ÷�k h�b�ø ^ _ � a ú Ó k l)�@��� b!� q üý�Yþv��<} �$ÿÖ�i�M�Vx'� Ój#jÖ÷�k b+a'a ^ _Rh�a ÿ j�ù5k#� gVf ^ _�`#a ú Ó k�l)�@��� b!�{û jVj�ù2k�� gVf ^ _R`�a ÿ j�÷�k b)a'a ^ _�h�a ú Ó k�l)�@��� b!� q üý�Yþv��<} �$ÿ p ���V�-�j#jÖ÷�k h�b!ø ^�_ � a ÿ j�ù5k#� gVf ^�_�`#a1ú Ó k�l)�@��� b!�{û jVj�ù2k�� gVf ^�_R`�a ÿ j�÷�k h�b!ø ^�_ � a!ú Ó k#l)�@��� b�� q üý�Yþv��<} �$ÿ p ���#x'� Ój�÷�k b)a'a ^�_�h�aHj#j�ù2k�� gVf ^�_�`�a!úk�l)�@��� b!�Rû jÖù5k#� g#f ^�_�`�a@jVj�÷�k b)a'a ^�_�h�a1úk�l)�@��� b!� q üý�Yþv��<} �$ÿ p ���V�-�
«
Remark 17 (Explanation of the MxOV package). Thestructural rulesin (211)
obey theverbfinal characterof theclausein a similar way like Set-CCG.Thedi-
rectionality remains fixed, in that the rightmost element (wherethe verbal head
is located) is never moved. Moreover, conform VFinal, we define the eligible
orders in termsof a distribution of the vfinal feature. Finally, we make the pos-
sibility to scramble dependent on the presenceof case-marking. The reason for
A formalmodelof wordorderasstructuralindicationof informativity /131
doing so is that languagesgenerally tend to rigidify their word order assoonas
case-marking (inflection) is absent, or caseis realizedthroughfunction words.We
canobserve this for examplein Germansubordinateclauses,wherescrambling of
complements is only possible if they areproperly marked for case, and in Turk-
ish, whereword order becomesSOV assoonascasemarking is suppressed (cf.
(Hoffman,1995a),p.50ff).
To illustratethe MxOV package,consider the following Japaneseexamples,
adaptedfrom Tsujimura’s (1996)(p.186). Thesentencein (194) givesthecanoni-
cal ordering of the elements, whereasthe sentencesin give the other possible or-
derings.
(194) Japanese
Kinooyesterday
Taroo-gaTaro-NOM
susi-osushi-ACC
tabeta.eat-PAST
“Tarooatesusiyesterday.”
(195) a. Taroo-gakinoo susi-o tabeta.
b. Taroo-gasusi-o kinoo tabeta.
c. Susi-oTaroo-gakinoo tabeta.
d. Susi-okinoo Taroo-ga tabeta.
e. Kinoo susi-o Taroo-ga tabeta.
All theseorders arederivableusing MxOV andthe following structural rules
thataddbehavior for themode � � (temporal adjunct).
(196)jÖ÷�k�� g#f ^ _�`�a j�ùÂ^ _eø�f.b úk�l)�@��� b!��û ùÆ^ _Yø�f.b j#j�÷�k�� gVf ^ _R`�a úk�l)�@��� b!� q üý�Yþv��<} �Çÿ��'��ze� p � Ó uj#jÖ÷�k b)a'a ^ _Rh�a ÿ ùÆ^ _eø�f.b ú Ó k�l)�@��� b!�Rû jÖùÆ^ _eø�f.b ÿ jÖ÷�k b)a'a ^ _Rh�a ú Ó k�l)�@��� b!� q üý�Yþv��<} �Çÿ��'��ze�1�i� Ó u
j�ùÆ^�_Yø�f�bC÷�k�l)�@��� b!��û ùÆ^�_Yø�f.b"j�÷�k#l)�@��� b�� q v�w�x�y{z"|��i} �:ÿ ��� x�y{z-|#�+���'��z Ó u
(197) a.
...temp ^ _eø�f.b ÿ j subjck#� gVf ^ _R`�a ÿ j dobjck b)a'a ^ _�h�a j tverbkVl)�@��� b!�ÖÓVÓ o�p q v�w�x�y{z-|Ö�<} |#ÿ �"� �Cz-�e� ��� x�y{z"|
Ó utemp ^5_Yø�f�b ÿ j subjckV� g#f ^�_�`�a@jVj dobjck b+a'a ^�_�h�a tverbkVl)�@��� b!�ÖÓ o�p q v�w�x�y{z-|#��} �7ÿ �"� x�y{z"|��H���i� Ó utemp ^ _Yø�f�b jVj subjckV� g#f ^ _�`�a ÿ j dobjck b+a'a ^ _�h�a tverbÓ kVl)�@��� b�� o�p q v�w�x�y{z"|��i} �:ÿ ��� x�y{z-|#�+�
p � Ó uj subjck#� gVf ^�_R`�a@j temp ^5_eø�f.b ÿ j dobjck b+a'a ^�_�h�a tverbÓ kVl)�@��� b�� o�p q üý�Yþv��<} �Çÿ��'��ze� p � Ó uj#j subjck#� gVf ^ _R`�a ÿ temp ^ _eø�f.b ÿ j dobjck b+a'a ^ _�h�a tverbÓVÓ kVl)�@��� b!� o�p q v�w�x�y{z-|#��} �7ÿ �"� x�y{z"|��H�
p � Ó ujVj subjckV� g#f ^�_�`�a ÿ temp ^5_Yø�f.b ÿ j dobjck b)a'a ^._Rh�a tverbÓ#Ó k f9ø�d o�p q ü z��'¨Cx���x p ��� x�y{z-|�uj subjckV� g#f ^�_�`�a ÿ temp ^5_Yø�f.b ÿ j dobjck b)a'a ^�_Rh�a tverbÓ#Ó oï¤>¥ f>ø�d p q
¤F¥C¦ u
132s A formal model of wordorder asstructural indication of informativity
b.
...temp ^5_Yø�f.b ÿ j subjck#� gVf ^�_�`#a ÿ j dobjck b)a'a ^/_Rh�aHj tverbkVl)�@��� b!��ÓVÓ o0p q v�w�x�y{z"|Ö�<} |#ÿ ��� �@z"�e� ��� x�y{z-|
Ó utemp ^5_eø�f.b ÿ j subjck � g#f ^._R`�aCj#j dobjck b)a'a ^�_�h�a tverbk l)�@��� b!� Ó o�pÔq v�w�x�y{z"|#��} �7ÿ ��� x�y{z"|��+���-� Ó utemp ^5_eø�f.b"j#j subjck#� g#f ^._R`�a ÿ j dobjck b)a'a ^�_�h�a tverbÓ kVl)�@��� b!� o�p q v�w�x�y{z"|���} �7ÿ �"� x�y{z"|��H�
p � Ó uj subjck#� gVf ^�_�`#aHj temp ^�_eø�f.b ÿ j dobjck b)a'a ^�_�h�a tverbÓ kVl)�@��� b!� o�p q üý�Yþv��<} �$ÿ��'��zY� p � Ó uj subjck#� gVf ^ _�`#a jVj dobjck b)a'a ^ _Rh�a ÿ temp ^ _eø�f.b tverbÓ kVl)�@��� b!� o�p q üý�Yþv��<} �$ÿ��'��zY�V�-� Ó ujVj subjck#� gVf ^�_�`#a ÿ j dobjck b)a'a ^�_Rh�a ÿ temp ^2_eø�f.b tverbÓ#Ó kVl)�@��� b�� o0prq v�w�x�y{z"|#��} �7ÿ ��� x�y{z"|��+�
p � Ó uj#j subjck#� gVf ^ _R`�a ÿ j dobjck b)a'a ^ _�h�a ÿ temp ^ _eø�f.b tverbÓ#Ó k f>ø�d o�p q ü z��'¨Cx���x p �"� x�y{z"|£uj subjck#� gVf ^�_R`�a ÿ j dobjck b)a'a ^�_�h�a ÿ temp ^5_eø�f.b tverbÓ#Ó o�¤>¥ f>ø�d p q
¤9¥C¦ u
c.
...temp ^ _eø�f.b ÿ j subjckV� g#f ^ _�`�a ÿ j dobjck b)a'a ^ _�h�a j tverbkVl)�@��� b��ÖÓ#Ó o�p q v�w�x�y{z"|��*} |Vÿ ��� �Cz"�*� �"� x�y{z"|
Ó utemp ^5_Yø�f.b ÿ j subjck#� gVf ^�_R`�a@j#j dobjck b)a'a ^�_Rh�a tverbkVl)�@��� b��ÖÓ o�p q v�w�x�y{z"|��i} �:ÿ ��� x�y{z"|#�+���i� Ó utemp ^ _Yø�f.b j#j subjck#� gVf ^ _R`�a ÿ j dobjck b)a'a ^ _Rh�a tverbÓ kVl)�@��� b!� o�p q v�w�x�y{z-|#��} �7ÿ �"� x�y{z"|��H�
p � Ó uj subjck#� g#f ^�_�`�a@j temp ^5_Yø�f.b ÿ j dobjck b)a'a ^�_Rh�a tverbÓ kVl)�@��� b!� o�p q üý��þv��*} �Çÿ��'��zY� p � Ó uj subjck#� g#f ^ _�`�a j#j dobjck b)a'a ^ _�h�a ÿ temp ^ _Yø�f�b tverbÓ kVl)�@��� b!� o�p q üý��þv��*} �Çÿ��'��zY�V�-� Ó uj#j subjck#� g#f ^�_�`�a ÿ j dobjck b)a'a ^�_�h�a ÿ temp ^5_Yø�f�b tverbÓ#Ó kVl)�H��� b!� o�p q v�w�x�y{z"|��i} �:ÿ ��� x�y{z"|#�+�
p � Ó uj#j dobjck b)a'a ^�_�h�a ÿ j subjck#� g#f ^�_R`�a ÿ temp ^5_Yø�f�b tverbÓ#Ó kVl)�H��� b!� o�p q üý�Yþv��<} �$ÿ p ���V�-� Ó uj#j dobjck b)a'a ^ _Rh�a ÿ j subjck#� gVf ^ _�`#a ÿ temp ^ _eø�f.b tverbÓVÓ k f>ø�d o�p q ü z��'¨@x���x p ��� x�y{z"|£uj dobjck b)a'a ^�_Rh�a ÿ j subjck#� gVf ^�_�`�a ÿ temp ^2_eø�f.b tverbÓVÓ oï¤>¥ f>ø�d p q
¤F¥M¦ u«WhereassomeOV languageshavemixedwordorderbut areotherwiserigidly verb
final, like Japaneseor Korean, other OV languagesdo allow for single dependents
(of thematrixverb)to occurafter theverbalhead. For example,Herring& Paolillo
discussSinhala, anIndo-Aryanlanguagethat is SOV (Greenberg/Hawkinstype23)
andwhich is non-rigid in the this way. Consider theexamplesin (198)and(199),
from (HerringandPaolillo, 1995)(pp.169-170).
(198) Sinhala
oyathat
gan-enrivier-INSTR
e-god.athat-bank
ekaone
paetta-k-aside-INDEF-LOC
lokularge
kaelaeaewa-kforest-INDEF
tibunaa.be-PAST.
“On thefar bankof thatrivier wasa large forest.”
(199) a. Sinhala
gan-enriver-INSTR
me-god.a-tthis-bank-also
tibunaabe-PAST
kaelaeaewa-k.forest-INDEF
A formalmodelof wordorderasstructuralindicationof informativity /133
“On this bank of therivier alsowasa forest.”
b. Sinhala
oyathat
kaelaeaewa-weforest-LOC
hit.iyaalive-PAST
nariy-ek.jackal- INDEF
“In thatforest liveda jackal.”
Herring & Paolillo associate the post-verbalpositioning of a dependentwith
(presentational)focus,bringingto theforetheintroductionof new information.Later
we comeback to this useof post-verbalpositioning - for themoment,we arejust
concernedwith thepureform of theconstruction.
Definition 19 (Nonrigid OV ordering (VFinal.MxOV.NrOV)). TheNrOV pack-
age extendstheVFinal andMxOV packages,modeling nonrigid verbfinality. Fol-
lowing out the strategy introducedin VFinal, we have a feature that controls the
postverbal positioning, called nrvfinal. Thefeature interactswith vfinal and en-
suresthat only a single dependentcan be placed postverbally. NrOV consistsof
thestructural rulesin (200).
(200)
ô�¬�õ Ú�� å ¶ ô�¬+õ� é%ö Ñ � µ � · � ½ � ê"º�ç º��2¼�âi¼ªê"º��eº#Ì��rÊ��:º�¼�½Y¾ÖÐô�²Ø�� ±�Ü ¬+õ � é%ö Ñ � µ � ¶ ¬Ý ¯ � ± ô�²hõYö Ñ � µ � · Þ©ê��0¸0ß�¿�àáÉ��iÎMÏ#Ðô�²ë Ñ£±!Ü ¬+õ� é%ö Ñ � µ � ¶ ¬Ý ¯{Ñ£± ô�²cõ ö Ñ � µ � · Þ©ê��0¸0ß�¿�àáÉ�º#Î�Ï#Ðô�²Ø�°)±!Ü�¬+õ � é%ö Ñ � µ � ¶ ¬Ý�¯�°)±�ô�²hõYö Ñ � µ � · Þ©ê��0¸0ß�¿�àáÉ�Ì*Î�Ï#Ð
«
Remark 18 (Explanation of the NrOV package). As we point out in the def-
inition, the packageintroduces a new feature, nrvfinal, that controls postverbal
positioning. By meansof a linking rule, we canhave that a matrix ( à � ç ) clause
hasa postverbaldependent. That is the caseif andonly if a single dependent is
placedafter construction that is otherwiseverbfinal.
Thederivationsin (201) examplify thestructural rulesof NrOV, on (abstract)
clauses similar in form to the examplesin (199). Observe that, becausewe still
have thelink betweenà � ç and Ê��:º�¼�½e¾ aswell (from VFinal) exampleslike (199)
already follow from VFinal+MxOV.
134s A formal model of wordorder asstructural indication of informativity
(201) a.
subjc �����! �"$#��%�'&)(�*,+-subjc. /",# ��� �0&(�* +21 ���'354
dobjc ��� �0&)($( � ��6�&)* +-dobjc. &(�( ���%��6�&7*$+ 1 � � 3�4
.
.
.-iobjc .�8 &7*�9�:<; ( - dtverb.>=)?!@ & 8A�����6�&)*$+%B : 8 (�C ���'&)(�*$+%B :ED (�F�G 1 B)3�4-
dobjc. &)($( 9 : 8 (�C - iobjc . 8 &)* 9�:<; ( - dtverb. =)?!@ & 8 GE��� �0&(�* +�B :ED ( F 1 B)3�4-subjc. /",#H9 :ED ( C - dobjc. &(�(I9 : 8 (!C - iobjc .>8 &7*E9�:<; ( - dtverb.>=)?!@ & 80GJG<�KF 1 B354-subjc. �"$#L9 :ED ( C - dobjc. &(�(59 : 8 ( C - iobjc .�8 &7*�9 :<; ( - dtverb.>=M ; /&N GJGO�KF 1 P�Q5R +/S'T U�V T C W!X'Y S!Z0[ W!\ R +/S'T]G^4-subjc. /",# 9 :ED ( C - dobjc. &)($( 9 : 8 ( ->- iobjc . 8 &7* 9�:<; ( dtverb. =M ; /&N G<�AF 1 P�Q5R +/S'T`_�V a C]W�\ R +/S!T`[cb Red G^4-subjc. /",#L9 :ED ( ->- dobjc. &(�(I9 : 8 ( C - iobjc .�8 &7*�9 :<; ( dtverbGJ.>=M ; /&N �AF 1 P�Q5R +fS'T`_�V a C]W!\ R +/S'T`[,b�Z d Ge4-dobjc. &(�(I9 : 8 ( ->- subjc. /",#L9 :ED ( C - iobjc .�8 &7*�9�:<; ( dtverbGJ.>=M ; /&N �AF 1 g�h0i�P U�V j C F d [�Z d G^4-dobjc. &(�(I9 : 8 ( ->- iobjc .�8 &7*�9�:<; ( C - subjc. �",#k9 :ED ( dtverbGJ.>=M ; /&N �AF 1 g�h0i�P U�V j C F d [ R]d G^4-iobjc . 8 &7* 9�:<; ( ->- dobjc. &)($( 9 : 8 (C - subjc. �",# 9 :ED ( dtverbGJ. =M ; /&N �AF 1 g�h0i�P U�V j C Z d [ Red G^4- C - dobjc. &)(�(�9 : 8 (C - subjc. /",#L9 :ED ( dtverbG`G 95; (�l - iobjc .�8 &7* . fm =M ; �&)N �nF 1 oqp)i�P U�V j C R]d G^4- C - dobjc. &)($(59 : 8 ( C - subjc. �",#k9 :ED ( dtverbGJG 9 ; (cl - iobjc .>8 &)* . #r*Js �KF 1 g S't p7Reh5R F<+/u)+ p7R]v�R Zxw W�\ R +fS'T 4C - dobjc. &(�(59 : 8 (�C - subjc. /",#L9 :ED ( dtverbGJG 95; (�l - iobjc .>8 &7* ���%� #r*`s F 1 ���!y�4
b.
...zsubjc{ +/u)j}| bIF d
~�zdobjc{ S d,d | b�Z d
~�ziobjc{ Z)S't5| b R]d
zdtverb{ W!\ R +fS'T>���r����� �������5�<�J�<� �
~>�O��� �O��� �<� ���5�<� �$�zsubjc{ +/u)j | bIF d
~�zdobjc{ S d,d | b�Z d
z�ziobjc{ Z)S't | b R]d dtverb{ W!\ R +fS'T �r��� � �������5�<�$��� � ~>�<� ���5�<�$�0�}��� �$�z
subjc{ +/u)j | bIF dz�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b R]d dtverb� { W!\ R +fS'T ��� � �������5�<�$��� �~>�<� ���5�<�$�'�}��� �$�z
dobjc{ S d,d | b�Z dz�z
subjc{ +/u)j�| bIF d~�z
iobjc{ Z)S'tI| b R]d dtverb� { W!\ R +fS'T5��� � ���� ¡���<� ¢~ � �f����� �$�z�~�z
subjc{ +fu)j | b�F d~�z
iobjc{ ZS!t | b R]d dtverb����| Z d)£zdobjc{ S d$d { + p W!\ R +/S'T �¤�¥� ¦�§/ ¡���<� ¢ ~ ��� �$�z�~�z
subjc{ +/u)j | bIF d~�z
iobjc{ Z)S't | b R]d dtverb����| Z d£zdobjc{ S d,d { jAt h ��� � ����¨,§/����� � �5©f�5§/�>ªO����« �<� ���5�E� �~�z
subjc{ +/u)j | bIF d~�z
iobjc{ Z)S't | b R]d dtverb���¬| Z d£zdobjc{ S d,d �Hq® jAt h � �
®'¯ �
c.
...zsubjc{ +/u)j | bIF d
~�zdobjc{ S d,d | b�Z d
~�ziobjc{ Z)S't | b R]d
zdtverb{ W!\ R +fS'T ���r����� �������5�<�J�<� �
~>�O��� �O��� �<� ���5�<� �$�zsubjc{ +/u)j | bIF d
~�zdobjc{ S d,d | b�Z d
z�ziobjc{ Z)S't | b R]d dtverb{ W!\ R +fS'T �r��� � �������5�<�$��� � ~>�<� ���5�<�$�0�}��� �$�z
subjc{ +/u)j | bIF dz�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b R]d dtverb� { W!\ R +fS'T ����� �������5�<�$��� �~>�<� ���5�<�$�'�}��� �$�z�~�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b Red dtverb���¬| F d£zsubjc{ +/u)j { + p W!\ R +/S'T �¤�¥� ¦�§/ ¡���<� ¢ ~ � � �$�z�~�z
dobjc{ S d$d | b�Z d~�z
iobjc{ ZS't | b Red dtverb����| F d£zsubjc{ +fuj { jAt h ��� � ����¨,§/����� � �5©f�5§/�>ªO����« �<� ���5�E� �~�z
dobjc{ S d$d | b�Z d~�z
iobjc{ ZS!t5| b R]d dtverb����| F d£zsubjc{ +fu)j��H ® jAt h � �
® ¯ �
«
Definition 20(Freeword order fr om OV (VFinal.MxOV.FreeOV)). TheFreeOV
package extends the behavior to freeword order of verbal complements,starting
froma basic OV order. TheFreeOV package consistsof thestructural rules given
in (202). Themonotonically extend VFinal and MxOV, and can for example be
usedin conjunction with NrOV.
A formalmodelof wordorderasstructuralindicationof informativity /135
(202)
j�ù2k f>ø�dû jÖù5kV�/°!n�n q ü z��'¨Cx���x p � ¨M�C�+ujÖù5kV�/°!n�n û jÖù5k#l)�H��� b!� q w�¨M�C�Mþv��*} |Vÿ � ¨M�C�i� �"� x�y{z"| Ó ujÖù5k �/°!n�n û jÖù5k �O°Vl)�@��� b!� q w�¨M�C�Mþv��*} |Vÿ � ¨M�C�i�VyR¨ �"� x�y{z"| Ó u
j�÷ ^�`�aVí>j�ù5k#� gVf kV�/°!n�n û jVj�ù2k�� gVf ^�_R`�a1÷�k#��°!n�n q w�¨M�C�Mþv»�i} �Çÿ p � Ó uj ÿ ÷ ^�`#aVí:ú Ó ^�h�aVíFj�ù2k b)a'a kV�/°!n�n û j ÿ jÖù5k b+a'a ^�_Rh�aV÷ Ó ^�`�aVí:úkV�/°!n�n q w�¨M�C�Mþv»�i} �Çÿ��-�M� p � Ó uj ÿ ÷ ^ h�aVí ú Ó ^ � aVí j�ù2k h�b!ø kV�/°!n�n û j ÿ jÖù5k h�b�ø ^ _ � a ÷ Ó ^ h�aVí úk#��°!n�n q w�¨M�C�Mþv»�i} �Çÿ�x'�M�1�i� Ó u
ùÆ^ _ � a ÿ ÷ý^ `�aVí ú Ó8û ÿ ùÆ^ _ � a ÷ Ó ^ `#aVí ú q w�¨M�C�Mþv²±<} z pCp ��ÿ p �M�1�i� Ó uj ÿ ùÆ^ � aVí>jÖúk h�b!ø�Ó ^�h�aVíFj�÷�k b)a'a kV�/°!n�n û j ÿ ùÆ^�h�aVíFjÖ÷�k b)a'a�Ó ^ � aVí�jÖúk h�b!ø k#��°1n#n q w�¨M�C�Mþv´³*} �Çÿ�x'�M�1�i� Ó uj ÿ ùÆ^ � aVí9jÖúk h�b!ø#Ó ^�`�aVí>j�÷�k#� gVf kV�/°!n�n û j ÿ ùÆ^�`#aVí9j�÷�k�� gVfÓ ^ � aVí>jÖúk h�b�ø kV��°1n�n q w�¨M�C�Mþv´³*} �Çÿ�x'�M� p � Ó uj ÿ ùÂ^�h�aVíFj�úk b)a'a!Ó ^�`�aVí>j�÷�k#� gVf kV�/°!n�n û j ÿ ùÆ^�`#aVí9j�÷�k�� gVfÓ ^�h�aVíFjÖúk b)a'a k#��°1n#n q w�¨M�C�Mþv´³*} �Çÿ��-�M� p � Ó uj ÿ ùÆ^ � aVí9j�÷�k h�b!ø#Ó ^�`�aVí9jÖúk#� gVf kV�/°!n�n û j ÿ ùÆ^�`#aVí9jÖúk#� g#fÓ ^ � aVí9jÖ÷�k h�b�ø kV��°1n�n q w�¨M�C�Mþv´³*} �Çÿ�x'�M� p � Ó uj ÿ ùÂ^ h�aVí jÖ÷�k b+a'a1Ó ^ `�aVí jÖúk#� gVf kV�/°!n�n û j ÿ ùÆ^ `#aVí jÖúk#� g#fÓ ^ h�aVí jÖ÷�k b)a'a k#��°1n#n q w�¨M�C�Mþv´³*} �Çÿ��-�M� p � Ó u
jÖù5kV�/°!n�n�^ � aVí ÷ û jÖùÆ^ � aVí ÷�kV�/°!n�n q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i�Vx'� Ó ujÖùÆ^ � aVí7÷�kV�/°!n�n û jÖù5kV�/°!n�n�^ � aVí7÷ q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i�Vx'� Ó ujÖù5k#��°!n�n�^�h�aVí�÷ û jÖùÆ^�h�aVí�÷�kV��°1n�n q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i�1�i� Ó ujÖùÆ^�h�aVí�÷�k �/°!n�n û jÖù5k �/°!n�n ^�h�aVí�÷ q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i�1�i� Ó uj�ù2k#��°1n#n�^�`�aVí7÷ û jÖùÆ^�`�aVí7÷�k#��°!n�n q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i� p � Ó uj�ùÆ^ `#aVí ÷�kV�/°!n�n û jÖù5kV�/°!n�n�^ `�aVí ÷ q w�¨M�C�Mþv�µe} �:ÿ � ¨M�C�i� p � Ó u
Remark 19(Explanation of the FreeOV package).TheFreeOV packagebuilds
forth on the MxOV packageby letting the latter handle all the scrambling that
maintains verb finality. FreeOV addsto that behaviorby first of all enabling the
formation of anVO order, using theFreeOV1.* andFreeOV2.* rules.13 Just like
MxOV, wedefinestructural rulesthatallow for scramblingof theargumentsatany
level of the tree,aslong astheentire structureis markedasfree. TheFreeOV3.*
are responsible for that. The FreeOV4.* make sure that we can freely reorder
elementsat the top-most level of the treeaswell asmoreembeddedlevels. The
examplederivationsin (203) illustrateFreeOV on the abstract lexicon with case
marking.
13Thus,in a senseFreeOV takesfurtherthebehavior we definein NrOV.
136s A formal model of wordorder asstructural indication of informativity
(203) a.
subjc �H¶® +fu)j q® S d t �zsubjc{ +/u)j �Hq® S d t � � q®f·��
...zdobjc{ S d$d | b�Z d
~�ziobjc{ ZS't | b Red
zdtverb{ W!X'Y S�Z �r�L¸® S d t �¬¹ b�F d ��� ¹
·��zsubjc{ +fuj | b�F d
~�zdobjc{ S d$d | b�Z d
~�ziobjc { ZS!t | b R]d
zdtverb{ W�X'Y S!Z ���r�k� � ¹ ·��z
subjc{ +/u)j�| b�F d~�z
dobjc{ S d$d | b�Z d~�z
iobjc{ ZS!t5| b R]dzdtverb{ W�\ R +/S!T>���r����� ���������<�J�O� �
~>�O��� �O��� �<� �����<� �$�zsubjc{ +fuj | b�F d
~�zdobjc{ S d$d | b�Z d
z�ziobjc{ ZS!t | b R]d dtverb{ W�\ R +/S!T �r���º� ���������<�$�/� � ~>�<� �����<�$�'�}��� �$�z
subjc{ +fuj | b�F dz�z
dobjc{ S d$d | b�Z d~�z
iobjc{ ZS!t | b R]d dtverb� { W�\ R +/S!T ����� ���������<�$�/� �~>�<� �����<�$�'�}��� �$�z�~�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b R]d dtverb����| F d)£zsubjc{ +/u)j { + p W�\ R +fS'T ��� � ¦¤§� ¡���<� ¢
~ � � �$�z�~�zdobjc{ S d,d | b�Z d
~�ziobjc{ Z)S'tI| b Red dtverb���¬| F d)£
zsubjc{ +/u)j { \ p YcY �¤� � ��§ �0� ¡�¡�O� � ~�� § �0� �7�5§ �<� ���5�<� �$�z�~�~�z
iobjc{ Z)S't | b R]d dtverb��| F d)£zsubjc{ +/u)j ��| Z d)£
zdobjc{ S d$d { \ p YcY �¤� � ��§
�0� ¡�»��� ¢ ~ ���/� � � �$�z�~�~�ziobjc{ Z)S't | b R]d dtverb��| Z d£
zdobjc{ S d,d �¬| F d)£
zsubjc{ +/u)j { \ p YcY �¤� � ��§
�0� ¡�¡¼<� ¢ ~ ���/� � � �$�z�~�~�ziobjc{ ZS!tI| b R]d dtverb�5| Z d)£
zdobjc{ S d$d �¬| F d)£
zsubjc{ +/u)j { jAt h ��� � ����¨c§f����� � � § �0� �~�~�z
iobjc{ Z)S't | b R]d dtverb�¬| Z d£zdobjc{ S d,d �¬| F d£
zsubjc{ +/u)j �L¸® jAt h � � q® ¯ �
b.
...zsubjc{ +/u)j | bIF d
~�zdobjc{ S d,d | b�Z d
~�ziobjc{ Z)S't | b R]d
zdtverb{ W!\ R +fS'T ���r��� � �������5�<�J�<� �
~>�O��� �O��� �<� ���5�<� �$�zsubjc{ +/u)j | bIF d
~�zdobjc{ S d,d | b�Z d
z�ziobjc{ Z)S't | b R]d dtverb{ W!\ R +fS'T �r���º� �������5�<�$��� � ~>�<� ���5�<�$�0�}��� �$�z
subjc{ +/u)j | bIF dz�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b R]d dtverb� { W!\ R +fS'T ����� �������5�<�$��� �~>�<� ���5�<�$�'�}��� �$�z�~�z
dobjc{ S d,d | b�Z d~�z
iobjc{ Z)S't | b Red dtverb���¬| F d£zsubjc{ +/u)j { + p W!\ R +/S'T �¤�¥� ¦�§/ ¡���<� ¢ ~ � � �$�z�~�z
dobjc{ S d$d | b�Z d~�z
iobjc{ ZS!t5| b R]d dtverb����| F d£zsubjc{ +fu)j { \ p Y,Y ��� � ��§ �0� ¡�½�<� � ~�� § �0� �7�5§ �<� �����<� �$�z�~�~�z
iobjc{ Z)S't | b Red dtverb�5| F d£zsubjc{ +/u)j �¬| Z d£
zdobjc{ S d,d { \ p Y,Y ����� ��§
�0� ¡���/� ¢ ~ ���/� � � �$�z�~�~�ziobjc{ Z)S't | b Red dtverb�5| Z d£
zdobjc{ S d,d ��| F d£
zsubjc{ +fu)j { \ p Y,Y ����� ��§
�0� ¡��¼O� ¢ ~ ���/� � � �$�z�~�ziobjc{ Z)S'tI| b Red dtverb�5| Z d£
zdobjc{ S d,d { \ p YcY | F d£
zsubjc{ +/u)j���� � ��§ �0� ¡�¡¾<� � ~�� § �0� � � � �$�z�~
dtverb | Z d)£zdobjc{ S d$d �¬| R]d£
ziobjc { ZS!t { \ p YcY | F d£
zsubjc{ +/u)j������ ��§
�0� ¡���/� ¢ ~ ���/�7��� �$�z�~dtverb | Red)£
ziobjc{ ZS't �5| Z d£
zdobjc{ S d,d { \ p YcY | F d£
zsubjc{ +/u)j ��� � ��§
�0� ¡��¼O� ¢ ~ ���/�7��� �$�z�~�~dtverb | Red)£
ziobjc{ ZS't �5| Z d£
zdobjc{ S d,d ��| F d£
zsubjc{ +fu)j { \ p Y,Y ��� � ��§
�0� ¡�¡¾<� � ~�� § �0� � � � �$�z�~�~dtverb | Red)£
ziobjc{ Z)S't$�¬| Z d£
zdobjc{ S d,d ��| F d£
zsubjc{ +fuj { jAt h ��� � ����¨c§f���¿� � � § �0� �~�~
dtverb | R]d£ziobjc{ ZS't ��| Z d£
zdobjc{ S d,d ��| F d£
zsubjc{ +fu)j �H ® jAt h � �
q® ¯ �
c.
...j subjck#� g#f ^ _R`�a ÿ j dobjck b)a'a ^ _�h�a ÿ j iobjck h�b�ø ^ _ � a j dtverbk1l)�@��� b���Ó#Ó o0p q v�w�x�y{z"|��*} |Vÿ ��� �Cz"�e� ��� x�yj subjck#� gVf ^�_�`#a ÿ j dobjck b)a'a ^�_Rh�a+jVj iobjck h�b!ø ^�_ � a dtverbk1l)�@��� b���Ó o�p q v�w�x�y{z"|��i} �:ÿ ��� x�y{z-|#�+�Çxj subjck#� gVf ^ _�`#a jVj dobjck b)a'a ^ _Rh�a ÿ j iobjck h�b!ø ^ _ � a dtverbÓ kVl)�@��� b!� o�p q v�w�x�y{z-|#��} �7ÿ �"� x�y{z"|��H���i�j dobjc k b)a'a ^/_Rh�aHjVj subjck#� gVf ^�_�`�a ÿ j iobjck h�b!ø ^�_ � a dtverbÓ kVl)�@��� b!� o�p q üý��þv��*} �Çÿ p �M�1�i� Ó uj dobjc k b)a'a ^/_Rh�aHjVj iobjck h�b!ø ^�_ � a ÿ j subjck#� gVf ^/_�`�a dtverbÓ kVl)�@��� b!� o�p q üý��þv��*} �Çÿ p �M�Vx'� Ó u
j ÿ j iobjck h�b!ø ^ _ � a ÿ j subjck#� gVf ^ _�`#a dtverbÓVÓ ^ h�aVí j dobjc k b)a'a k��<°Vl)�@��� b!� o�p q îï¨Mþv��<} �Çÿ��-� Ó uj ÿ j iobjck h�b!ø ^�_ � a ÿ j subjck � g#f ^�_�`�a dtverbÓ#Ó ^/h�aVíFj dobjck b)a'a k ��°1n#n o�p q w�¨M�C�Mþv��*} |Vÿ � ¨M�C�i�VyR¨ �j ÿ#ÿ j subjck#� gVf ^ _�`#a dtverbÓ ^ h�aVí j dobjck b)a'aVÓ ^ � aVí j iobjck h�b!ø k#��°1n#n o�p q w�¨M�@�Mþv0��} �$ÿ�x'�M�V�-� Ó uj ÿ#ÿ j subjck#� gVf ^�_�`#a dtverbÓ ^ � aVí>j iobjck h�b!ø�Ó ^�h�aVíFj dobjck b)a'a k#��°1n#n o�p q w�¨M�@�Mþv¤³<} �$ÿ�x'�M�V�-� Ó uj ÿVÿ j subjck#� gVf ^ _R`�a dtverbÓ ^ � aVí j iobjck h�b!ø�Ó ^ h�aVí j dobjc k b)a'a k f>ø�d o�p q ü z��'¨@x���x p � ¨M�C�HuÿVÿ j subjck � gVf ^�_R`�a dtverbÓ ^ � aVí9j iobjck h�b!ø Ó ^�h�aVí9j dobjc k b)a'a o�¤ ¥ f>ø�d p q
¤F¥M¦ u«
A formalmodelof wordorderasstructuralindicationof informativity /137
Observe that in FreeOV we maintain therequirementthatdependentsneedto
haveexplicit casemarking for themto beableto scramble. Naturally, wecaneasily
relax this constraint - structure of the rulesremains the same,only we no longer
needto have any feature markingon ¬ , ² or Ù . The definition below givesthe
relaxeddefinitionsof MxOV (MxOVnc) andFreeOV (FreeOVnc). Thepackages
MxOVnc andFreeOVnc monotonically extend oneanother, aswell astheMxOV
andFreeOV packages,andcanfor example beusedto modelthe freeword order
of Hindi.
Definition 21 (Mixed/f ree OV word order without casemarking). The Mx-
OVnc package is a relaxedversion of MxOV, in which dependentsno longer need
to be explicitly marked for case. MxOVnc comprises the rules given in (204).
FreeOVnc is a similarly relaxed version of FreeOV, consisting of therulesgiven
in (205).
(204)
÷á^�_ � a@jÖùÆ^�_�h�a!úk�l)�@��� b!�Rû ùÂ^�_�h�aHj�÷ ^�_ � a!úk�l)�@��� b!� q üý��þvy{�)�*} �Çÿ��-�M�Vx�� Ó u÷ ^/_ � aCj�ùÂ^�_�`�a!úk l)�@��� b!� û ùÂ^�_�`�a@j�÷ý^�_ � a1úk l)�@��� b�� q üý��þvy{�)�*} �Çÿ p �M�Vx'� Ó u
jÖ÷ ^�_ � a ÿ ùÆ^�_Rh�aVú Ó k�l)�@��� b!�{û jÖùÆ^�_�h�a ÿ ÷ý^�_ � a1ú Ó k�l)�@��� b!� q üý��þvy{�)�*} �Çÿ��-�M�Vx�� Ó uj�÷ ^�_Rh�a ÿ ùÆ^�_�`#a1ú Ó k�l)�@��� b!�{û jÖùÆ^�_�`#a ÿ ÷ ^�_�h�a1ú Ó k#l)�@��� b�� q üý��þvy{�)�*} �Çÿ p �M�1�i� Ó uj�÷ý^ _ � a ÿ ùÆ^ _�`#a ú Ó k�l)�@��� b!�{û jÖùÆ^ _�`#a ÿ ÷ ^ _ � a ú Ó k#l)�H��� b!� q üý��þvy{�)�*} �Çÿ p �M�Vx'� Ó u÷á^ _�h�a j�ùÂ^ _�`�a úk�l)�@��� b!�Rû ùÂ^ _�`�a j�÷ý^ _�h�a úk�l)�@��� b!� q üý��þvy{�)�*} �Çÿ p �M�1�i� Ó u÷ ^�_�`�a@jÖùÆ^�_eø�f.bMúk�l)�@��� b!�Rû ùÂ^�_eø�f.b"j�÷ ^�_R`�a!úk�l)�@��� b!� q üý��þvy{�)�*} �Çÿ��'��zY� p � Ó u
j�÷ý^�_�h�a ÿ ùÆ^�_Yø�f�bMú Ó k l)�@��� b!� û jÖùÆ^�_eø�f.b ÿ ÷ ^�_�h�a1ú Ó k l)�H��� b!� q üý��þvy{�)�*} �Çÿ��'��zY�V�-� Ó u
(205)
j�÷ ^�`#aVí7ù5kV�/°!n�n û jÖùÆ^�_�`�a1÷�k#��°1n#n q w�¨M�C�Mþv»�+y{��} �Çÿ p � Ó uj ÿ ÷ ^�`�aVí:ú Ó ^�h�aVí7ù5kV�/°!n�n û j ÿ ùÂ^�_Rh�a#÷ Ó ^�`�aVí:úkV��°1n�n q w�¨M�C�Mþv»�+y{��} �Çÿ��-�M� p � Ó uj ÿ ÷ý^ h�aVí ú Ó ^ � aVí ù5kV�/°!n�n û j ÿ ùÂ^ _ � a ÷ Ó ^ h�aVí úk#��°!n�n q w�¨M�C�Mþv»�+y{��} �Çÿ�x'�M�1�i� Ó u
ù©^ _ � a ÿ ÷ ^ `#aVí ú Óªû ÿ ùÆ^ _ � a ÷ Ó ^ `�aVí ú q w�¨M�C�Mþv²±My{��} z pCp ��ÿ p �M�1�i� Ó uj ÿ ùÂ^ � aVíªú Ó ^�h�aVí�÷�kV�/°!n�n û j ÿ ùÂ^�h�aVí�÷ Ó ^ � aVí:úk#��°!n�n q w�¨M�C�Mþv´³�y{��} �Çÿ�x'�M�1�i� Ó uj ÿ ùÂ^ � aVí:ú Ó ^�`#aVí7÷�kV�/°!n�n û j ÿ ùÂ^�`#aVí7÷ Ó ^ � aVí:úkV�/°!n�n q w�¨M�C�Mþv´³�y{��} �Çÿ�x'�M� p � Ó uj ÿ ùÂ^�h�aVí7ú Ó ^�`#aVí7÷�k �/°!n�n û j ÿ ùÂ^�`#aVí7÷ Ó ^�h�aVí7úk ��°1n�n q w�¨M�C�Mþv´³�y{��} �Çÿ��-�M� p � Ó uj ÿ ùÂ^ � aVí:ú Ó ^�`#aVí7÷�kV�/°!n�n û j ÿ ùÂ^�`#aVí7÷ Ó ^ � aVí:úkV�/°!n�n q w�¨M�C�Mþv´³�y{��} �Çÿ�x'�M� p � Ó uj ÿ ùÂ^ h�aVí ú Ó ^ `#aVí ÷�kV�/°!n�n û j ÿ ùÂ^ `#aVí ÷ Ó ^ h�aVí úkV��°1n�n q w�¨M�C�Mþv´³�y{��} �Çÿ��-�M� p � Ó u
jÖù5kV�/°!n�n�^ � aVí ÷ û jÖùÆ^ � aVí ÷�kV�/°!n�n q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i�Vx'� Ó ujÖùÆ^ � aVí7÷�kV�/°!n�n û jÖù5k#��°!n�n�^ � aVí:÷ q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i�Vx'� Ó ujÖù5k#��°!n�n�^>h�aVí7÷ û jÖùÆ^�h�aVí7÷�kV��°1n�n q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i�1�i� Ó ujÖùÆ^�h�aVí�÷�k �/°!n�n û jÖù5k ��°!n�n ^�h�aVí7÷ q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i�1�i� Ó uj�ù2k#��°1n#n�^�`�aVí7÷ û jÖùÆ^�`�aVí7÷�k#��°1n#n q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i� p � Ó uj�ùÆ^ `#aVí ÷�kV�/°!n�n û jÖù5k#��°!n�n�^ `�aVí ÷ q w�¨M�C�Mþv�µ�y{��} �:ÿ � ¨M�C�i� p � Ó u
«
138s A formal model of wordorder asstructural indication of informativity
4.3.3 V-FIRST PACKAGES
After having definedpackagesdealing with OV wordorderin theprevioussection,
we now turn our attention to the “mirror” caseof VO word order. Thereasonfor
dealing first with OV andVO beforeaddressingissuesin SVO is thatthelatter may
beconceivedof asa mixture of theformertwo.
Thecanonical structurefor VSOis asgivenin (206a),with asamplederivation
(with goal À ) in (206b).
(206) a. É�É�É�¸�Ã�ê5Á �� ±! À æAÁ!ÃeÏ 2°)±! Á��ÄÁ�ÃYÏ Ñì±'ÂÆÅ �ÄÁ�ÃYÏ
b.
tverb ��� � =?�@ & 8 C`C F�Ç 8 (cl � �6�&)* +%GJÇ D (cl � �'&)(�* +%G-tverb.>=)?!@ & 8A� C FcÇ 8 (cl �%�)6!&7*,+�GJÇ D (cl��%�!&(�*,+ 1 � � 3�4 subje ��� � @ m�È � �0&)(�* +-
subje.J@ m�È �É���0&)(�*$+ 1 � � 354-tverb.>=)?!@ & 8 9 D (cl - subje.>@ mcÈ �KF�Ç 8 (cl ��� 6�&)* + 1 Ç)3�4 dobjc ���%�'&)(�('���6�&)*$+-
dobjc. &(�( ���%� 6!&7* +Ê1 ���'354C - tverb.>=)?!@ & 8 9 D (�l - subje.>@ m�È G 9 8 (�l - dobjc. &)(�( �AF 1 Ç)3�4C - tverb.>= ; ; *�9 D (�l - subje.>@ m�È G 9 8 (�l - dobjc. &)(�( �KF 1 P y7+ R tJU�V T C]W�X'Y S!Z0[ W R + R t>G^4-
tverb9 D (�l - subje. @ m�È . = ; ; * 9 8 (�l - dobjc. &)($( �nF 1 P y7+ R t$_�V a C]W R + R t$[$F dx£ G^4- C tverb9 D (�l - subje.>@ m�È G 9 8 (�l - dobjc. &(�( .J= ; ; * �KF 1 P y7+ R t$_�V a C]W R + R t$[�Z dx£ G^4- C tverb9 D (�l - subje. @ m�È G 9 8 (cl - dobjc. &(�( . #¶*`s �nF 1 g S't p)Reh5R F P w R + R t R S!T 4C tverb9 D (cl - subje.>@ m�È G 9 8 (�l - dobjc. &)(�( �����!#r*JsfFË1 � � y�4
Thepackagescover VO in general, andthusalsocover the(morescarce) VOS
languages.
Definition 22 (Rigid verb-initial word order, VFirst). TheVFirst package de-
finesthe basiccontrols for establishing that a structure is rigidly verb initial (or
V1). It consistsof thestructural rulesgivenin (207), which are essentially mirror-
ing therulesfound in theVFinal package (179).
(207)
j�ù2k f>ø�d û jÖù5k l)�H�`° `�ø q ü zE�'¨Cx���x p vÍÌ � x�¨ p ��ujÖù5k `'cM �û jÖù5k#l)�H�`° `�ø q ¡8��¢ªx p vÎÌ � x�¨ p ��u
jÖù5k#l)�H�`° `�ø ^ `�aVí ÷ û jÖù5k#l)m@n b)h ^ `#aVí ÷ q v�w�x�¨ p ���*} |Vÿ ��� �Cz"�e� ��� x�¨ p � Ó uj�ù2k�l)�@�^° `�ø ^ h�aVí ÷ û jÖù5k#l)m@n b)h ^ h�aVí ÷ q v�w�x�¨ p ���*} |Vÿ ��� �Cz"�e� ��� x�¨ p � Ó uj�ù2k�l)�@�`° `�ø ^ � aVí7÷ û jÖù5k#l)m@n b)h ^ � aVí:÷ q v�w�x�¨ p ���*} |Vÿ ��� �Cz"�e� ��� x�¨ p � Ó ujÖùÆ^�`�aVí7÷�k�l)�@�`° `�ø8û jÖù5k#l)�H�`° `�ø ^�`�aVí7÷ q v�w�x�¨ p ���i} �:ÿ ��� x�¨ p ��� p �¿Ï Ó uj�ùÆ^�h�aVí�÷�k�l)�@�`° `�ø8û jÖù5k#l)�H�`° `�ø ^�h�aVí�÷ q v�w�x�¨ p ���i} �:ÿ ��� x�¨ p ���1�i�¿Ï Ó uj�ùÂ^ � aVí7÷�k�l)�@�`° `�ø8û jÖù5k#l)�H�`° `�ø ^ � aVí7÷ q v�w�x�¨ p ���i} �:ÿ ��� x�¨ p ���Vx���Ï Ó u
«
Remark 20(Explanation of the VFirst package).TheVFirst packagedefinesthe
basic behavior of thecontrol feature ¸0¹�º�ê5� � , andits interaction with thestandard
modes��Î , Ì*Î and ºVÎ aswell asthefeature Ê�Ë�Ã�½eÌ . Thepackagedoesnot defineany
A formalmodelof wordorderasstructuralindicationof informativity /139
variability in word order - it definesrigid SVO. To establish that a clauseindeed
is rigid in this sense, we needtheVFirst packageanda goalcategory Ð�Ñ<ÒÔÓ Ñ é � � À ,
(208).
(208)
...ÿ j tverbk!Õ ô �`° `'ø ^�`�aVí>j subjek1n�°)Ö Ó ^�h�aVí9j dobjck b)a'a o�p q vw�x�¨
p ���<} |#ÿ �"� �Cz-�e�1v�w�x�¨ p � Ó uj tverb ^ `�aVí j subjekVn�°ÖMk�Õ ô �^° `�ø ^ h�aVí j dobjck b)a'a o�p q vw�x�¨ p ����} �7ÿ�v�w�x�¨ p ��� p ��Ï Ó uj ÿ tverb ^/`�aVí9j subjek n�°Ö Ó ^�h�aVíFj dobjck b+a'a k Õ ô �^° `�ø o�p q vw�x�¨
p ����} �7ÿ�v�w�x�¨ p ���1�i�¿Ï Ó uj ÿ tverb ^/`�aVí9j subjekVn7°)Ö Ó ^�h�aVíFj dobjck b)a'a k f>ø�d o�p q ü z��'¨Cx���x p vÎÌÅx�yRxc�'x'z"|£uÿ tverb ^ `�aVí j subjek n7°)Ö Ó ^ h�aVí j dobjck b)a'a o�¤ ¥ f>ø�d p q
¤9¥@¦ u
The derivation in (208) coversVSO order- but the VFirst package is not re-
stricted that thatparticularword order type. It alsocoversVOSword order, aswe
find it in for example TobaBatak,(209).14
(209) Toba Batak
Mang-idasee-ACTIVEV
si ElijahElijah
si Kathy.Kathy
“Kathy seesElijah.”
The derivation for a structure that we could assign to (209) is given in (210)
below. Naturally, to reflectthefact thatVOSis thecanonicalword order, we start
with a verbal category thatfirst takes theobject, andthenthe(ergative) subject.
(210)
ostverb ��� � =)?!@ & 8 CJC FcÇ D (�l � �'&)(�* +%GJÇ 8 (�l � �6�&7* +�G-ostverb.J=)?!@ & 8A� C F�Ç D (cl����'&(�*,+�GJÇ 8 (cl ���6�&)*$+ 1 � � 3�4 dobjc ��� �'&(�( � �6�&7* +-
dobjc. &)($( �����6�&)*$+ 1 � � 354-ostverb. =)?!@ & 8 9 8 (cl - dobjc. &)($( �KF�Ç D (cl � �0&)(�* + 1 Ç)3�4 subje ����� @ m�È/���0&(�*�+-
subje. @ mcÈ ��� �0&)(�* +×1 ���'3�4C - ostverb.J=?�@ & 8 9 8 (cl - dobjc. &(�( G 9 D (cl - subje.>@ mcÈ �KF 1 Ç354C - ostverb.J=M ; m D *�9 8 (cl - dobjc. &(�( G 9 D (cl - subje.>@ mcÈ �KF 1 P�Q5R^p Fct`U�V T C]W!X'Y S�Zf[ W�\ R^p Fct>Ge4-
ostverb9 8 (�l - dobjc. &)($( .J=M ; m D *�9 D (cl - subje.>@ m�È �KF 1 P�Q5R^p Fct�_�V a C W!\ R^p Fct�[JZ dr£ G^4- C ostverb9 8 (�l - dobjc. &)($( G 9 D (�l - subje. @ m�È . =M ; m D * �KF 1 P�Q5R^p Fct�_�V a C W!\ R^p Fct�[>F dx£ Ge4- C ostverb9 8 (�l - dobjc. &)($( G 9 D (�l - subje.>@ m�È . #r*Js �KF 1 g S!t p7R^h5R F P w \ Rep F,t`4C ostverb9 8 (cl - dobjc. &(�( G 9 D (cl - subje.>@ mcÈ ���%�'#r*`sfF 1 ���!y�4
«
Definition 23 (Mixed, rigi d verb-intia l word order, MxVO). TheMxVO pack-
agedefinesthemixedword orderpattern for rigidly verbinitial languages,extend-
ing the VFirst package. MxVO allows for properly case marked dependentsto
occurin anyorder, andcomprisestherules givenin (211).14ACTIVEV=Activevoice,and“si” arepropernamemarkers;cf. (Manning, 1996),p.27ff.
140s A formal model of wordorder asstructural indication of informativity
(211)
z�Ø | R]d£z�Ù { ZS't { W!\ Rep Fct | Z d)£
z�Ú { S d$drÛ z�Ø | Z d)£z�Ú { S d,d { W�\ Rep Fct | R]d£
z�Ù { ZS't � ���I�� ½�O� ¢ ~ ���/�7��� �$�z�Ø | R]d£z�Ù { Z)S't { W�\ Rep Fct | F d)£
z�Ú { +/u)j Û z�Ø | F d£z�Ú { +fu)j { W!\ Rep Fct | Red)£
z�Ù { Z)S't � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�Ø | Z d£z�Ù { S d$d { W�\ Rep Fct | F d)£
z�Ú { +/u)j Û z�Ø | F d£z�Ú { +fu)j { W!\ Rep Fct | Z d)£
z�Ù { S d,d � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�~�Ø | R]d£z�Ù { ZS't ��| Z d)£
z�Ú { S d,d { W�\ Rep Fct Û z�~�Ø | Z d)£z�Ú { S d,d ��| Red)£
z�Ù { Z)S't { W!\ R^p Fct � ���I�� ½�O� ¢ ~ ���/�)��� �$�z�~�Ø | Z d£z�Ù { S d,d ��| F d£
z�Ú { +/u)j { W�\ Rep Fct Û z�~�Ø | F d£z�Ú { +fu)j���| Z d)£
z�Ù { S d$d { W!\ R^p Fct � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�~�Ø | R]d£z�Ù { ZS!t,��| F d£
z�Ú { +/u)j { W�\ Rep Fct Û z�~�Ø | F d£z�Ú { +fu)j���| Red)£
z�Ù { Z)S't { W�\ R^p Fct � ���I�� ½�O� ¢ ~ ���/� � � �$�z�Ø | R]d£z�Ù { ZS't { W!\ Rep Fct | F d£
z�Ú { Y p�v Û z�Ø | F d£z�Ú { Y p7v { W�\ R^p Fct | R]d£
z�Ù { ZS!t � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�Ø | Z d£z�Ù { S d,d { W!\ Rep Fct | F d£
z�Ú { Y p�v Û z�Ø | F d£z�Ú { Y p7v { W�\ R^p Fct | Z d£
z�Ù { S d,d � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�~�Ø | Z d£z�Ù { S d,d ��| F d£
z�Ú { Y p�v { W�\ Rep Fct Û z�~�Ø | F d£z�Ú { Y p7v ��| Z d)£
z�Ù { S d,d { W!\ Rep F,t � ���I�� ½�O� ¢ ~ � �/�7��� �$�z�~�Ø | R]d£z�Ù { ZS't ��| F d£
z�Ú { Y p�v { W�\ Rep Fct Û z�~�Ø | F d£z�Ú { Y p7v ��| Red)£
z�Ù { Z)S't { W!\ Rep F,t � ���I�� ½�O� ¢ ~ ���/� � � �$�«
Remark 21 (Explanation of the MxVO package). The MxVO package allows
for dependentsto bescrambled, aslongasthey bearpropercasemarking. Because
someverbiniti al languageshave ergativecasemarking (rather thanabsolutive) we
have also included behavior for the Ã�êE� feature. The MxVO extends the VFirst
packagemonotonically, anddoes not alter any of its rigidnessin placing the verb
initi ally.
To illustratetheMxVO package,consider theTagalogexamplesin (212). The
sentencein (212a) givesthecanonical order, which we already presenteda deriva-
tion for in (208). The sentence in (212b) is a variation on (212a), differing in
the order of the dependents. The derivation for (212b) necessarily makesuseof
MxVO, andis givenin (213).
(212) Tagalog
a. Nagbabasaread-PAST
angtitserteacher
ng dyaryo.newspaper
“The teacher readthenewspaper”
b. Nagbabasaread-PAST
ng dyaryonewspaper
angtitser.teacher
“The teacher readthenewspaper”
(213)
...ÿ j tverbkVl)�@�^° `�ø ^�`�aVí9j subjekVn�°Ö Ó ^�h�aVíFj dobjck b)a'a o�p q v�w�x�¨
p ���<} |#ÿ ��� �@z"�e�1vw�x�¨ p � Ó uj tverb ^.`�aVí>j subjek1n�°)Ö�k#l)�H�`° `�ø ^�h�aVí>j dobjc k b)a'a o�p q v�w�x�¨ p ����} �7ÿ ��� x�¨ p ��� p ��Ï Ó uj ÿ tverb ^.`�aVí>j subjek1n�°)Ö Ó ^�h�aVí9j dobjck b)a'a k�l)�@�`° `'ø o�p q v�w�x�¨
p ����} �7ÿ ��� x�¨ p ���1�i�¿Ï Ó uj ÿ tverb ^.h�aVíFj dobjck b)a'a!Ó ^�`�aVí>j subjekVn7°)Ö�k�l)�@�`° `'ø o�p q üý�Yv�þ��<} �$ÿ p ���V�-� Ó uj ÿ tverb ^ h�aVí j dobjck b)a'aVÓ ^ `�aVí j subjekVn�°Ö�k f>ø�d o�p q ü z��'¨@x���x p vÎÌ$x�yRxc�'x�z"|�uÿ tverb ^/h�aVíFj dobjck b)a'aVÓ ^�`�aVí9j subjekVn�°Ö o�¤F¥ f>ø�d p q
¤F¥C¦ u
A formalmodelof wordorderasstructuralindicationof informativity /141
«Justlikein thecaseof OV languages, wemayfind thatalanguage-occasionally-
allows for a dependent to occurbefore theverb,without theword order beingfree
assuch. For example, the (Western-Malayo) Polynesian languageChamorroal-
lows for theergative subjectto appear before theverb.
Definition 24 (Non-rigid verb-first languages, NrVO). TheNrVO package de-
finesthepossibilit y for onedependentto occur before theverbal head– non-rigid
verb-firstness.TheNrVO package consistsof therulesin (214).
(214)
Ü ¬LÝ Ú�� å ¶ Ü ¬LÝ� é�Þ Ñ é � � · � ½ � ê"º�ç º�2¼�â-¼���ê"º��*º#Ì�Ê��:º�ê5� � ÐÜ ²Ø ¯ � ± ¬LÝ � é�Þ Ñ é � � ¶ Ü ¬LÝÞ Ñ é � � �� ±!Ü ² · ÞÆê<¸Ë�»ß�¿�àáÉ��iÎ�Ï#Ð
«
Remark 22 (Explanation of the NrOV package). Admittedly, the NrOV is not
very spectacular. It only allowsfor thedependentrealizedassubject to bemoved
beforethe verb. The reason for modeling NrOV this way is that our -admittedly
limited- observationsconcerning VO languagesall regardergative languages,and
thatin suchlanguagesnon-rigidity is usually reservedfor thesubject (with achange
in voiceto alterthedependentthatis realizedassubject). Thisalsoholdsfor a lan-
guagelikeTagalog. Therewecanorderanelementbefore theverbusing aspecific
construction called ay-inversion. Kroegerpointsout thattheinversionis generally
restricted to thesubject (1993)(p.67ff), but thequestion is whether to usea struc-
tural rule to model this phenomenonor achieve it through a lexical assignment
(to ay) asBaldridge (2000) proposes. Given Kroeger’s observations that thereis
aninterplay betweenwhatdependency relation is involvedin ay-inversion andthe
information structure,we proposeto usea structural rule.
Leaving the Tagalogcaseat the momentfor what it is, we have a look at
Chamorro. In Chamorro, we can employ NrVO directly to obtain the desired
analysis. Consider the examplesentencesin (215), from (Chung, 1990), andthe
derivation in (216) for (215b).
(215) Chamorro
a. lumi’e’see-PAST
i lahiman
i palao’an.woman
“The mansaw thewoman”
142s A formal model of wordorder asstructural indication of informativity
b. i lahiman
lumi’e’see-PAST
i palao’an.woman
“The mansaw thewoman”
(216)
...É Ü tverbÝ7Þ Ñ é � � � ±!Ü Ü subjeÝ�´Vé!ß Ï 2°)±!Ü Ü dobjc Ý#µ ±�±�à �·�¸0¹»º�ê5� � ß�¿�¾�É�Ê�Ë�Ã�½YÌ{Í�Ê��:º�ê5� � Ï#Ð
Ütverb ¿� ±�Ü Ü subjeÝ ´#é!ß Ý Þ Ñ é � � �°)±!Ü Ü dobjc Ý µ ±�± à �
·�¸0¹»º�ê5� � È"¿ Ä�É�Ê��:º�ê5� � Íf�iÎká�Ï#ÐÜtverb 5°)±!Ü Ü dobjc Ý�µ ±�± ÝÞ Ñ é � � �� ±!Ü Ü subjeÝ#´#é!ß à �
· � ç8¸â�»ß�¿�àáÉ��iÎ-Í)Ì*Î�Ï#ÐÜ'Ü
subjeÝ#´Vé!ß2 ¯ � ± É tverb �°)±!Ü Ü dobjc Ý�µ ±�± Ï Ý � é�Þ Ñ é � �Éà �· ÞÆê<¸Ë�»ß�¿�àáÉ��iÎ�Ï#Ð
Ü'ÜsubjeÝ ´#é!ß ¯ � ± É tverb 5°)±!Ü Ü dobjcÝ µ ±�±+Ï Ý Ú�� å à �
· � ½ � ê"º�ç º��¼�â-¼���ê"º��*º#Ì�Ê��:º�ê5� � ÐÜsubjeÝ#´#é!ß5 ¯ � ± É tverb 5°)±�Ü Ü dobjc Ý#µ ±�± Ï àäã ÑMÚ�� å �
· ã Ñ Å Ð
«
Definition 25 (Freeword order in VO languages,FreeVO). TheFreeVO pack-
age definesfreeword order, starting froma VO word order. Thepackage consists
of thestructural rules givenin (217). Just like MxVO, werequire that proper case
marking is present for a structureto appear in position differentfromthecanonical
one.
(217)
j�ù5k f>ø�dû j�ù2k#��°!n�n q ü z��'¨Cx���x p � ¨M�@�Huj�ù2k#��°!n�n û j�ù2k�l)�@�`° `�ø q w�¨M�C�Cv�þ��<} |#ÿ � ¨C�C�-� ��� xj�ù2k#��°!n�n û j�ù2k��O°1l)�@�^° `�ø q w�¨M�C�Cv�þ��<} |#ÿ � ¨C�C�-�#yR¨ �
jVj�÷�k � gVf ^ _R`�a ù5k ��°!n�n û j�ùÂ^ `�aVí j�÷�k � gVf k ��°!n�n q w�¨M�C�Cv�þ���} �$ÿ p � Ó uj#j�úk b)a'a ^ _�h�a ÿ ùÆ^ _�`�a ÷ Ó k#��°!n�n û j�ùÂ^ _�`�a ÿ ÷ý^ h�aVí jÖúk b)a'a!Ó k#��°!n�n q w�¨M�C�Cv�þ���} �$ÿÖ�i��� p � Ó uj#jÖúk h�b!ø ^ _ � a ÿ ùÆ^ _�h�a ÷ Ó k#��°!n�n û j�ùÂ^ _�h�a ÿ ÷ ^ � aVí j�úk h�b!ø�Ó kV��°1n�n q w�¨M�C�Cv�þ���} �$ÿÖ�i���#x'� Ó u
ÿ ùÆ^�`#aVí7÷ Ó ^ � aVí8ú û ùÆ^�_R`�a ÿ ÷ ^ � aVí:ú Ó q w�¨M�C�Cv�þ�±�} z p@p ��ÿ p �M�#x'�j#jÖ÷�k b+a'a ^�_Rh�a ÿ j�ù5k h�b!ø ^�_ � a1ú Ó k#��°!n�n û j#jÖù5k h�b!ø ^�_ � a ÿ j�÷�k b)a'a ^�_�h�a1ú Ó kV�/°!n�n q w�¨M�C�Cv�þ»³<} �$ÿ�x'�M�V�-� Ó uj#jÖ÷�k#� g#f ^�_�`�a ÿ j�ù5k h�b!ø ^�_ � a1ú Ó k#��°!n�n û j#jÖù5k h�b!ø ^�_ � a ÿ j�÷�k�� gVf ^�_R`�a!ú Ó k#��°!n�n q w�¨M�C�Cv�þ»³<} �$ÿ�x'�M� p � Ó uj#jÖ÷�k n7°)Ö ^ _�`�a ÿ j�ù5k h�b!ø ^ _ � a ú Ó k ��°!n�n û j#jÖù5k h�b!ø ^ _ � a ÿ j�÷�k n�°Ö ^ _R`�a ú Ó k �/°!n�n q w�¨M�C�Cv�þ»³<} �$ÿ�x'�M� p � Ó uj#jÖ÷�k�� g#f ^ _�`�a ÿ jÖù5k b)a'a ^ _Rh�a ú Ó k#��°!n�n û j#jÖù5k b)a'a ^ _Rh�a ÿ j�÷�k�� gVf ^ _R`�a ú Ó kV�/°!n�n q w�¨M�C�Cv�þ»³<} �$ÿÖ�i��� p � Ó uj#j�÷�k#n�°Ö/^ _�`�a ÿ jÖù5k b)a'a ^ _Rh�a ú Ó k#��°!n�n û j#jÖù5k b)a'a ^ _Rh�a ÿ j�÷�k#n�°Ö/^ _�`�a ú Ó kV��°1n�n q w�¨M�C�Cv�þ»³<} �$ÿÖ�i��� p � Ó u
j�ùÆ^�_R`�a1÷�k#��°!n�n û ùÆ^�_R`�a@j�÷�kV��°1n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-� p � ÓùÆ^�_�`�a@jÖ÷�k#��°!n�n û j�ùÂ^�_�`�a1÷�kV��°1n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-� p � Ój�ùÂ^�_�h�aV÷�k#��°!n�n û ùÆ^�_Rh�a+jÖ÷�k#��°!n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-�V�-� Óù©^ _�h�a jÖ÷�k#��°!n�n û j�ùÂ^ _�h�a ÷�kV�/°!n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-�V�-� ÓjÖùÆ^ _ � a ÷�k ��°!n�n û ùÆ^ _ � a j�÷�k ��°!n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-�#x'� Óù©^ _ � a jÖ÷�k#��°!n�n û j�ùÂ^ _ � a ÷�k#��°!n�n q w�¨M�C�Cv�þ¿µ*} �7ÿ � ¨C�C�-�#x'� Ó
«
A formalmodelof wordorderasstructuralindicationof informativity /143
Remark 23(Explanation of the FreeVO package). Justlikein theFreeOV pack-
age,we control theaccessibility of FreeVO’s structural rulesusinga freefeature.
TheFreeVO packagemonotonically extends theVFirst, MxVO andNrVO pack-
ages,so -again-we may have that in the process of constructing a derivation we
obtainstructuresthatcanbeanalyzedin termsof these morerestrictive packages.
Thederivation in (218) examplifiesthis.
(218)
...ÿ j tverbkVl)�H�`° `�ø ^�`�aVí9j subjekVn�°Ö Ó ^�h�aVíFj dobjck b+a'a o�p q v�w�x�¨
p ���*} |Vÿ ��� �Cz"�*� �"� x�¨ p � Ó uj tverb ^/`�aVí9j subjekVn�°Ö�k�l)�@�`° `�ø ^�h�aVíFj dobjck b)a'a o�p q v�w�x�¨ p ���i} �:ÿ ��� x�¨ p ��� p �¿Ï Ó uj tverb ^ h�aVí j dobjck b)a'a k�l)�@�^° `�ø ^ `�aVí j subjekVn�°Ö o�p q üý�Yv�þ��<} �$ÿ p �M�V�-� Ó u
j#j subjekVn7°)Ö/^�_�`�a ÿ tverb ^.h�aVíFj dobjck b)a'a1Ó k#�O°Vl)�@�`° `'ø o�p�q î ¨�v�þ��*} �Çÿp � Ó u
j#j subjekVn�°Ö.^ _R`�a ÿ tverb ^ h�aVí j dobjck b)a'aVÓ k#��°!n�n o0p q w�¨M�@�Mv�þ��<} |#ÿ � ¨M�C�-�#yR¨ ��� x�¨ p � Ó uj#j dobjck b)a'a ^�_Rh�a ÿ j subjekVn�°Ö�^�_R`�a tverbÓ k1��°!n�n o0p q w�¨M�C�Mv�þ���} �Çÿ��-�M�
p � Ó uj#j subjekVn�°Ö.^ _R`�a ÿ j dobjck b)a'a ^ _�h�a tverbÓ k1��°!n�n o0p q w�¨M�C�Mv�þ»³<} �Çÿ��-�M�
p � Ó uj#j subjekVn�°Ö/^�_�`�a ÿ j dobjck b)a'a ^�_�h�a tverbÓ k f9ø�d o�p q ü z��'¨Cx���x p � ¨M�@�Huj subjekVn�°Ö/^�_�`�a ÿ j dobjck b)a'a ^�_�h�a tverbÓ o�¤>¥ f9ø�d p q
¤9¥M¦ u
We do not definehereany packagesthat allow for mixed or free word order
of elements that arenot properly marked for case.If onewereto observe sucha
language,then theMxVO andFreeVO packagescanbeextendedto MxVOnc and
FreeVOnc by simply leaving off the casefeatures,analagously to the derivation
of MxOVnc andFreeOVnc from MxOV andFreeOV.
4.3.4 SVO PACKAGES
In the current section we definea number of packagesthat describe word order
behavior which might beavailableto a languagethat is SVO at oneclause level or
another. Figure4.4 givesa concise overview of the packagesandtheir interrela-
tions.
Thepackagesarebasedontheassumptionthatthelexiconassignsabasicword
orderasshown in (219).
(219) É ¬ ´åq� ± É�É ² 2°)±! Ù0Ï Ñ£±! ÁÇÏ�ÏBeforewegoanddefinethepackages,weshould first considerwhat“the struc-
ture” of SVO is - or, at least, whatview weadhereto here. For example,Figure4.5
illustratesthepossible waysin which wecanconceive of the(initial, or canonical)
structuring of SVO clauses.
144s A formal model of wordorder asstructural indication of informativity
Free SVO
&
Free SVO (nc)
V2-position
Wackernagel Position Mixed SVO
Rigid SVO
Figure4.4: Architectureof SVO packages
TheRigid SVO picture shows just thebasicword orderasin (219). A slightly
morecomplex situationarises whenwe addtheWackernagelposition to this con-
figuration. Following FGD,we characterizetheWackernagelposition asto be-in
general- theposition right afterthefirst dependent.15
More interesting situationsarise in the caseof verb secondness, either with
rigid or mixed SVO word order. A languagethat hasa verb second SVO word
order alsohasa verbfinal cluster. This splits thestructureup into severaldomains
or “fields”: the domainbefore the second (Wackernagel)position, the verb-final
cluster, andthedomainbetweenthecluster andthesecond position. In Germanic
linguistics,thesedifferent domainsareusually calledtheVorfeld, theNachfeld, and
theMittelfeld respectively.
Languagesthat are verb second SVO are most often mixed word order lan-
guages,like Dutchor German(with Swedishanexception wenotedin Chapter3).
Thus,becausewordordercanbemixed,wegetacommunication betweentheVor-
15Datafrom for exampleCzechshows quite obviously that it is not thefirst constituent, or eventhe first phrase. A phrase-structuregrammardoesnot lend itself to a satisfactorydefinition of theWackernagel position,in otherwords.Sgall(p.c.) notesthat,in a moredetailedway, theview FGDis asfollows: The Wackernagel position is (prototypically, if not always), (i) the surfacepositiondirectly following after the positionof the first item in the uppersubtreeif the surfaceword ordercorresponds to the underlying positionsof the subtree;(ii) if one of the deeper embeddeditemsis shifted to the left (as in Czech “Jirku jsmeplanovali poslatdo Francii”) then the Wackernagelpositionis aftertheheadof theshiftedsubtree(“jsme”); (iii) someothershiftsmaybenecessarytospecifyotherpossibilities.
A formalmodelof wordorderasstructuralindicationof informativity /145
Rigid SVO
Rigid SVO, Wackernagel
V2-SVO, Wackernagel, VFinal
Mixed V2-SVO, Wackernagel, VFinal
Free SVO, Wackernagel
Subject Verb Object
Subject
Subject
Subject
Subject
Object
Object
Object
Wack
Wack V2
V2
Wack Verb Object
Vfinal
Vfinal
Figure4.5: Structuring of SVO word order
feld andtheMittelfeld. Verbaldependents within theMittelfeld canbescrambled,
and can be ‘exchanged’ with the dependent in the Vorfeld (as verb secondness
needsto be maintained). If a languageis free but still does distinguish a Wack-
ernagel position, we get the moregeneral situation that dependents canbe either
placedin thedomainbeforetheWackernagelposition or after it - just aslong the
Wackernagelposition is indeed in its right place.
Definition 26 (Wackernagel Position, SVO.Wack). TheWack package defines
structural rulesthatcharacterizetheWackernagel position. Weconsider theWack-
ernagel position to be-in general- thepositionafter thefirst dependent.Thestruc-
tural rulesgivenin (220) implement that viewpoint, immediately for the rigi d ( æ )andmixed( ç ) cases. Thefree( è ) word order casesare defined in (230)on page
150,in thedefinition of theFreeSVO package.
146é A formal model of wordorder asstructural indication of informativity
(220)
ê�ëíì�îIï,ðfñ�ò�ì¸ó ðô�õ½ö�÷cø�ù ð�úÔû ë�ì�îIï,ðfñ�ê�ò ÷cø�ù ð�ú ìAó ðô�õ½ö ü ýrþ�ÿ�� ������� ñ�� ö � � ñ�� ����������!ö��ê�ëíìÉîIï,ð/ñ�ò�ì��]ðô õ½ö�÷cø�ù ð�úÔû ë�ì�îIï,ðfñ�ê�ò ÷cø�ù ð�ú ì��]ðô õ½ö ü ýrþ�ÿ�� ������� ñ�� ö � � ñ�� ����������!ö��ê�ë�ìÉîIï,ð/ñ�ò�ì�î ù���� õ½ö�÷ ø�ù ð�ú û ë�ì�îIï,ðfñ�ê�ò ÷ ø�ù ð�ú ì�î ù���� õ½ö ü ýrþ�ÿ�� ������� ñ�� ö � � ñ�� ����������!ö��
ê�ëíì�îIï$ð/ñ�ò ì î�� � ó õ½ö�÷cø�ù ð�úÔû ë�ì�îIï,ðfñ�ê�ò ÷cø�ù ð�ú ì î!�"� ó õ½ö ü ýrþ�ÿ�� ������� ñ�� ö � � ñ�� ����������!ö��ê�ëíì�î ó ð/ñ�ò ì ï$ðô õ½ö�÷ ø�ù ð�ú û ë�ì�î ó ðfñ�ê�ò ÷ ø�ù ðcú ì ï$ðô õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%$&�!ö��ê�ëíì î ó ð ñ�ò�ì �]ðô õ½ö�÷ ø�ù ð�ú û ë�ì î ó ð ñ�ê�ò ÷ ø�ù ðcú ì �]ðô õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%$&�!ö��ê�ëíì î!�]ð ñ�ò ì ï$ðô õ½ö�÷ ø�ù ð�ú û ë�ì î!�]ð ñ�ê�ò ÷ ø�ù ð�ú ì ï$ðô õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%'(�!ö��ê�ëíì�î��eðfñ�ò�ì¸ó ðô�õ½ö�÷ ø�ù ð�ú û ë�ì�î!�]ð/ñ�ê�ò ÷ ø�ù ð�ú ì¸ó ðô�õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%'(�!ö��
ê�ëíì î ó ð ñ�ò�ì�î ù���� õ½ö�÷cø�ù ð�úÔû ë�ì î ó ð ñ�ê�ò ÷cø�ù ðcú ì�î ù��)� õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%$&�!ö��ê�ëíì î ó ð ñ�ò ì î�� � ó õ½ö�÷cø�ù ð�úÔû ë�ì î ó ð ñ�ê�ò ÷cø�ù ðcú ì î�� � ó õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%$&�!ö��ê�ë�ì î��eð ñ�ò�ì î ù���� õ½ö�÷ ø�ù ð�ú û ë�ì î!�]ð ñ�ê�ò ÷ ø�ù ð�ú ì î ù��)� õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%'(�!ö��ê�ë�ì î!�]ð ñ�ò ì�î�� � ó õ½ö�÷cø�ù ð�úÔû ë�ì î!�]ð ñ�ê�ò ÷cø�ù ð�ú ìÉî�� � ó õ½ö ü ýrþ�ÿ�� ������� ñ�# ö � � ñ�� �������%'(�!ö��
ê�ê�ë ÷+*�,�- ù ó ì ï,ðô ñ�ò�ì¸ó ðô�õ½ö�÷ ø�ù ð�ú û ê�ë ÷+*�,.- ù ó ì ï,ð)ô ñ�ê�ò ÷ ø�ù ð�ú ìKó ðô�õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!� ���'ö��ê�ê�ë ÷ *�,�- ù ó ìKï,ðôqñ�ò�ì¸ó ðô�õ½ö�÷cø�ù ð�úÔû ê�ë ÷ *�,.- ù ó ìAï,ð)ôAñ�ê�ò ÷cø�ù ð�ú ìKó ðô�õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!� ���'ö��ê�ê�ë ÷ *�,�- ù ó ìAó ð)ô ñ�ò ìnï$ðô õ½ö�÷cø�ù ð�úÔû ê�ë ÷ *�,.- ù ó ì¸ó ðô ñ�ê�ò ÷cø�ù ð�ú ìKï,ðô õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!��$&�'ö��ê�ê�ë ÷ *�,.- ù ó ìAó ðô ñ�ò�ì��]ðô õ½ö�÷ ø�ù ð�ú û ê�ë ÷ *�,.- ù ó ì¸ó ðô ñ�ê�ò ÷ ø�ù ð�ú ì��eð)ô õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!��$&�'ö��ê�ê�ë ÷ *�,�- ù ó ì1�eð)ôAñ�ò ìnï$ðô õ½ö�÷cø�ù ð�úÔû ê�ë ÷ *�,.- ù ó ì1�]ðô¸ñ�ê�ò ÷cø�ù ðcú ìKï,ðô õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!��'(�'ö��ê�ê�ë ÷+*�,.- ù ó ì �]ðô ñ�ò�ì ó ðô õ½ö�÷ ø�ù ð�ú û ê�ë ÷+*�,.- ù ó ì �]ðô ñ�ê�ò ÷ ø�ù ðcú ì ó ð)ô õ½ö ü ýrþ�ÿ�� ������� ñ0/ ö � � ñ�� �����!��'(�'ö��
2
Remark 24 (Explanation of the SVO.Wack package). The Wack packagede-
finesa “generalpurpose”setof structural rulesthatcanbeused in any word order
setting. Thepackageassuchdoesnot defineanyordering. All its structural rules
do is definewhattheWackernagelposition is in varioussettings,with “definition”
meaning thata a 35476&8 featurecanonly bepercolated to a higher level if andonly
if the elementoriginally labeled with that feature is in the right position (i.e. the
Wackernagelposition).
The Wack packageserves as the foundation for various, more specific ac-
counts. Below we detail out theV2nd packages,which extend theWack package
to modelverbsecondness.2
Definition 27(Verb-second position, SVO.Wack.V2nd). TheV2nd package is a
simpleextension to theWack package, to defineverbsecondnessin (rigid/mixed)
SVO languages. TheV2nd package comprisesthestructural rulesgiven in (221).
(221)
9�:<;>=@?(ACBD90:�;�E.F G HJILK%MONQP5N�RTSVU�WYX�Z90:�;�E.F[BD90:�;�\Y].^%_ G SVU�WYX�N%Ra`JI�b)cLZ
:edgfih ^Oj 90k5;�\Y].^%_Tdgf =Tl�m�npo Bq:�dgfih ^�j 90k5;�E.Fadgf =Tl mrnpo G stSvu5w `JI�b)ciwxSVU�WYXzy{w | j(} U!~ �gI�b)c o Z:�d fih ^ j 9�kv;>\Y].^%_Td f ]��OA npo Bq:�d fih ^ j 90k5;�E.Fad f ]��OA npo G stSvu5w `JI�b)ciwxSVU�WYXzy{w | j(} U!~ �gI�b)c o Z:�d f�h ^ j 9�kv;�\Y])^%_ad m�^ � npo Bq:�d fih ^ j 90k5;�E.�O��].mTd m�^ � npo G stSvu5w `JI�b)ciwxSVU�WYXzy{w | j(}!�{� I�X7~>�gI�b)c o Z
A formalmodelof wordorderasstructuralindicationof informativity /147
Becausea requirementfor verbsecondnessis almostinvariably accompanied
bya requirementfor certain verbsto beverbfinal, thefollowing rules fromVFinal
should becombinedwith (221) to account for that, andoneof the *Dep packages
to determinetheorder in theverbfinal cluster.
(222)
:�dgf m�^ 9�kv;�E.�r���]��YB�:edgf m�^ 90k5;�E.�O��].m G SC��N+WYI�|0y�w | j0}��7� I�X{~ }�� N+WYI�| o Z:�dgf ��^ 9�kv;�E.�r���]��YB�:edgf ��^ 90kv;>E.�O��].m G SC��N+WYI�|0y�w | j0}��7� I�X{~ }�� N+WYI�| o Z90:�d f m�^ kv;>E.�����]���B�:ed f m�^ 90k5;�E.�r���z]�� G SC��N+WYI�|��w � j0}�� N+WYI�|�~���Xzb o Z9�:�d f ��^ kv;>E.�����]���B�:ed f ��^ 90kv;>E.�r���]�� G SC��N+WYI�|��w � j0}�� N+WYI�|�~���N+b o Z2Remark 25 (Explanation of the V2nd package). TheV2nd packageis, assaid,
a simple extension to theWack package. Becausethelatter alreadythe important
aspect of defining the“verbsecond” or Wackernagelposition, all theV2nd package
doesis specifying what should appear in that position for that element(and the
clause) to be verb second. Particularly, a clause is verb second if it either has
the verbalheadin the second position (rule ���L�t�{�i� ��¡Q¢i£¥¤�4¦�Y§�35476&8Y¨ ), or if it has
the(modal) auxiliary in thatposition ( �©�L�t�{�i� ��¡Q¢¦�¦§�35476&8Y¨ ). In the latter case, the
verbalheaditself is then -mostly- requiredto bein verbfinal position.
To illustratetheV2nd package,considerfirst theDutchsentencein (223), with
theformal lexical entriesasgivenin (224).
(223) Dutch
ChristopherChristopher
wilwants
boekenbooks
lezen.to read
“Christopher wantsto readbooks.”
(224) Lexicon:
christopher ªY«@¬&�®%¯%�boeken ª"« ¬.°�)¯ �wil ªY« ¬�±)² ¡�¡�« ¬ )®%¯%�@³{´"µ ®�¶L¨�· ´Y¸�¹�º « ¬ ±�»&¼�½ )¾ ¡�« ¬ �®%¯%�@³{´"µ ®�¶¿��Ôèt¨�¨lezen ª¥«a¬&±)À�Á )º ¡�«[¬)°�)¯%�@³ ´¥º.® ¡�«g¬&�®%¯%�@³{´"µ ®�¶¿��Ôèt¨�¨
Observe that themodalauxiliary “wil” imposestwo requirements. First of all,
it needs the infinite (andany of its arguments) to form a verb final structure,due
to the  ¬ ±�»O¼�½ �¾ . Secondly, it statesthat it hasto appear itself in the verb second
position,  ¬&±)² . Becausewe give asgoal-category  ¬ ¸g¯�Ã7Ä which thenrewrites to
Âv¬ ±)² Ä (verb-secondmatrix clause),all theserequirementshave to be met. The
derivation in (225) shows how this is done.
148é A formal model of wordorder asstructural indication of informativity
(225)
...christopher
ìÉîIï$ð/ñ�êwil÷�ø�ù ðcú ì î!� � óOñ
boekenì î ó ð
lezenö�ö¥Åv� ü ýrþ�ÿ�� �������!� þ<Æ.Ç�$�� È ñ�É Æ�� � �����Eö��
êchristopher
ìÔî�ï,ð�ñwil
ì î!� � ó<ñboeken
ì î ó ðlezen
ö�ö�÷ ø�ù ð�ú Åv� ü ýrþ�ÿ�� ������� ñ�� ö � � ñ�� ����������!ö��êchristopher
ì î�ï,ð ñwil
ì�î!�"� ó ñboeken
ìnî ó ðlezen
ö�ö�÷Q*�Ê@Å5� ü þ<Æ.Ç�$a'����������&�êchristopher
ìÉî�ï,ð/ñwil
ì î!� � óOñboeken
ì î ó ðlezen
ö�ö�÷ �1Ë �[Å5� ü ÌÍ�&Î � '(Ï�'��AþgÆ)Ç�$O�christopher
ìnîIï,ð/ñwil
ì î�� � óEñboeken
ì î ó ðlezen
ö�öYÅÑÐ¥Ò �1Ë � � ü Ð Ò�Ó �
2Definition 28 (SVO mixed word order, SVO.MxSVO). The MxSVO package
modelsmixedword order in SVO languages. The structural rules are given in
(226).
(226)
ê�ò ÷cø�ù ð�ú ìÉî ù��)� ñ�ë�ìÉîIï$ð õ½ö û ëíì�î�ï,ð/ñ�ê�ò ÷cø�ù ðcú ìÉî ù��)� õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ ��������Õ�Ï5ö��ê�ò ÷cø�ù ðcú ì î!� � óOñ�ë�ìÉîIï$ð õ½ö û ëíì�î�ï,ð/ñ�ê�ò ÷cø�ù ðcú ì î!� � ó õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ ����� #5Ö $<ö��ê�ò ÷ ø�ù ð�ú ì î ù���� ñ�ë�ìÉî ó ð õ½ö û ëíì�î ó ðfñ�ê�ò ÷ ø�ù ð�ú ì î ù���� õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ $&������Õ�Ï5ö��ê�ò ÷ ø�ù ð�ú ì î�� � óOñ�ë�ì î ó ð!õ½ö û ëíì î ó ð ñ�ê�ò ÷ ø�ù ð�ú ì î!�"� ó õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ $&��� #vÖ $<ö��ê�ò ÷ ø�ù ð�ú ì î ù���� ñ�ë�ì î!�]ð õ½ö û ëíì î��eð ñ�ê�ò ÷ ø�ù ð�ú ì î ù���� õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ '(������Õ�Ï5ö��ê�ò ÷cø�ù ð�ú ì î�� � óOñ�ë�ìÉî!�]ð õ½ö û ëíì�î��eðfñ�ê�ò ÷cø�ù ð�ú ì î!�"� ó õ½ö ü Ì�Ï5ýrþ�ÿ�Ô�� #�ñ '(��� #5Ö $<ö��ò�ì î ó ð ñ�ê�ë ÷cø�ù ðcú ìÉî ù��)� õ½ö û ê�ë ÷cø�ù ð�ú ì�î ù���� ñ�ò�ì î ó ð�õ½ö ü Ì�Ï5ýrþ�ÿ[Æ&� #�ñ ��Õ�Ϧ�%$&�!ö��ò�ì�î!�]ðfñ�ê�ë ÷cø�ù ðcú ìÉî ù��)� õ½ö û ê�ë ÷cø�ù ð�ú ì�î ù���� ñ�ò�ìÉî!�]ð õ½ö ü Ì�Ï5ýrþ�ÿ[Æ&� #�ñ ��Õ�Ϧ�%'(�'ö��ò�ì î ó ð ñ�ê�ë ÷ ø�ù ð�ú ì î!�"� ó õ½ö û ê�ë ÷ ø�ù ð�ú ì î�� � ó<ñ�ò�ì î ó ð!õ½ö ü Ì�Ï5ýrþ�ÿ[Æ&� #�ñ�#5Ö $��%$&�'ö��ò ì î��eð ñ�ê�ë ÷ ø�ù ð�ú ìÉî!�"� ó õ½ö û ê�ë ÷ ø�ù ð�ú ì�î�� � ó ñ�ò�ì î!�]ð õ½ö ü Ì�Ï5ýrþ�ÿ[Æ&� #�ñ�#5Ö $��%'(�'ö��
ò�ì î ó ð ñ�ë�ì îIï$ð õ½ö û ëíì î�ï,ð ñ�ò ì î ó ð!õ½ö ü Ì�Ï5ýrþ�ÿT×�� #�ñ ������$&�'ö��ò�ìÉî�ï,ðfñ�ë�ìÉî!�]ð õ½ö û ëíì�î��eðfñ�ò�ìÉîIï$ð õ½ö ü Ì�Ï5ýrþ�ÿT×�� #�ñ ������'(�'ö��ò�ìÉî��eðfñ�ë�ìÉîIï$ð õ½ö û ëíì�î�ï,ð/ñ�ò ì�î!�]ð õ½ö ü Ì�Ï5ýrþ�ÿT×�� #�ñ ������'(�'ö��ò�ìÉî!�]ðfñ�ë�ì î ó ð!õ½ö û ëíì î ó ð ñ�ò�ì�î!�]ð õ½ö ü Ì�Ï5ýrþ�ÿT×�� #�ñ $&����'(�'ö��ò�ì î ó ð ñ�ë�ìÉî!�]ð õ½ö û ëíì�î��eðfñ�ò�ì î ó ð�õ½ö ü Ì�Ï5ýrþ�ÿT×�� #�ñ '(����$&�'ö��õ ì�î ó ð/ñ�ò ì ï$ðô ë ö û ëíì î�ï,ð ñ�ò ì ó ð)ô õ½ö ü Ì�Ï5ýrþ�ÿaØ� #�ñ ������$&�'ö��õ ì î!�]ð ñ�ò ì ï$ðô ë ö û ëíì î�ï,ð ñ�ò ì �eð)ô õ½ö ü Ì�Ï5ýrþ�ÿaØ� #�ñ ������'(�'ö��ñ�ëíì �eð)ô õ½ö ìAó ð)ô ò û ñ�ëíì¸ó ðô ò ö ì �]ðô õ ü Ì�Ï5ýrþ�ÿaØ� #�ñ '(����$&�'ö��ñ�ëíìKï,ð)ô õ½ö ìAó ð)ô ò û ñ�ëíì¸ó ðô ò ö ìKï,ð)ô õ ü Ì�Ï5ýrþ�ÿaØ� #�ñ ������$&�'ö��ñ�ëíìKï,ð)ô õ½ö ì1�]ðôxò û ñ�ëíì1�]ðô�ò ö ìKï,ð)ô õ ü Ì�Ï5ýrþ�ÿaØ� #�ñ ������'(�'ö��
Remark 26 (Explanation of the MxSVO package). Thereareseveral observa-
tions we should make about thestructural rulesin (226). First of all, thepackage
hasbeen designedsuch that it relieson theWack packageto bepresent.Although
thecontraint that Ù should bemarked with 3v4¦6O8 could berelaxed,thereis a lin-
guistic reason why wemodeled MxSVO with thisconstraint. Thisreason is simple:
It follows from the definition of what it meansto be mixed SVO, (Steele, 1978).
In general,it tendsto hold thatmixedSVO languages have a requirementfor verb
secondness - witnessDutchandGerman.Towardstheendof rigid SVO languages
A formalmodelof wordorderasstructuralindicationof informativity /149
wehave for exampleEnglish, whichdoesnothavesucharequirement,andneither
doesa freeSVO languagelike Czech.
Secondly, we obtain a communication betweentheVorfeld andtheMittelfeld
in the following way. The MxSVO1.* “temporarily move aside”, as it were, to
allow the dependentfrom the Vorfeld to combine with the dependent(s) from the
Mittelfeld. TheMxSVO3.* enable any orderingof thedependents to obtain, after
which the MxSVO2.* rulesmove a dependent from the Mittelfeld back into the
Vorfeld. The latter stephas to bemade,becauseotherwise we would have an el-
ementthat requiresto be in the Wackernagelposition, but is not. And the stepis
made,if wehaveagoalcategory  ¬ ¸g¯�Ã7Ä , becausethewack featureonly distributes
if thesuchmarkedelementis indeed in theright position (rulesSVO.Wack0(m).*).
To illustratetheMxSVO package,considertheDutchexamplesin (227). These
areall valid variations,amongother possible ones. An analysisfor (227b) is given
below, in (228).
(227) Dutch
a. ChristopherChristopher
wilwants
boekenbooks
aanto
KathyKathy
voorlezen.to read
“Christopherwantsto readbooks to Kathy.”
b. Boekenwil ChristopheraanKathy voorlezen.
c. Christopherwil aanKathy boekenvoorlezen.
(228)
.
.
.
christopherÚ¦ÛzÜ�Ý�Þ�ß wil à�á�â Ý�ã Ú Ûä¥å+æ Þ boekenÚ Ûzæ�Ý Þ�Þ aan Ú¦Û&ç�è+é(ç kathyê�Ú Ûë�Ý voorlezenê�ê�êì ïîí ïð7ñ¥ò ó ùðcú ò ð7ô+õ ó ö ò ÷ Þ * ô ø ø�ù ðcú ê�ù
ß wil à�áLâ Ý0ã Ú Ûä¥å+æ Þ christopherÚ ÛzÜQÝ Þ boekenÚ Ûzæ�Ý Þ�Þ aan Ú Û&ç�è+é(ç kathyê�Ú Ûë�Ý voorlezenê�ê�êì ïîí ú � ï�ð¦ñ¥û%ò� Þ ï,ð ø � � ó ê�ù
ß wil à�áLâ Ý0ã Ú Ûä¥å+æ Þ boekenÚ Ûzæ�Ý Þ christopherÚiÛzÜQÝ�Þ�Þ aan Ú¦Û&ç�è+é(ç kathyê�Ú Ûë�Ý voorlezenê�ê�êì ïîí ú � ï�ð¦ñ¦ü�ò� Þ ï,ð ø ó ð ê�ù
boekenÚ Ûzæ�Ý Þ�ß wil à�áLâ Ý0ã Ú Ûä"åQæ Þ christopherÚ ÛzÜQÝ Þ�Þ aan Ú Û&ç�è+é(ç kathyê�Ú Ûë�Ý voorlezenê�ê�êì ï í ú � ï�ð¦ñiô ò� Þ � � ó ø ó ð ê�ù
ß boekenÚ Ûzæ�Ý Þ wil Ú Ûä¥å+æ Þ christopherÚiÛzÜQÝ�Þ�Þ aan Ú¦Û&ç è%é�ç kathyê�Ú Ûë�Ý voorlezenê�ê�ê�à�á�â Ý�ã ì ïýí ïð7ñ¥ò ó ùðcú ö Þ þ>ê ò ÿ Þ ø�ù ðcú ø ó ð ê�ù
ß boekenÚ Ûzæ�Ý Þ wil Ú Ûä¥å+æ Þ christopherÚ ÛzÜ�Ý Þ�Þ aan Ú Û&ç è%é�ç kathyêOÚ Ûë�Ý voorlezenê�ê�ê�à�� � ì ï í ðiô+õ ó �eï ó ù ð�ú ùß boekenÚ Ûzæ�Ý Þ wil Ú Ûä"åQæ Þ christopherÚ ÛzÜQÝ Þ�Þ aan Ú Û&ç è%é(ç kathyê�Ú Ûë�Ý voorlezenê�ê�ê�à ä���� ì ï í ú ù
Ë þ � � �eï ð�ô+õ ó ùboekenÚ Ûzæ�Ý Þ wil Ú Ûä¥å+æ Þ christopherÚiÛzÜQÝ�Þ�Þ aan Ú¦Û&ç è%é�ç kathyêrÚ Ûë�Ý voorlezenê�ê�êì ��� ä���� ï í � � ù
2
Definition 29 (SVO fr ee word order, SVO.FreeSVO). TheFreeSVO package
defines freeword order for SVO languages.It canbeusedin conjunction with the
MxSVO package, andconsistsof thestructural rulesgivenin (229). Theextension
to theWack package, to dealwith full freeword order, is givenbelow in (230).
150é A formal model of wordorder asstructural indication of informativity
(229)
ê�ë ÷ � Ë � û ê�ë ÷cø�ù ð�ú ü Ì���Î � '(Ï Ö��� � � �������&�ê�ë ÷ � Ë � û ê�ë ÷�� þ -+- ü Ì���Î � '(Ïg'�� /L�� � �ê�ò ÷ *�,�- ù ó ì � � ó ô ñ�ê�ë ÷ ø�ù ð�ú ìAï,ð)ô õ½ö û ê�ë ÷ ø�ù ð�ú ì î�� � ó<ñ�ê�ò ÷ *�,�- ù ó ìAï,ð)ô õ½ö ü � �� � ýrþ½ÿ�Ôr� #Ëñ�É��� ��$�� #vÖ $<ö��
ê�ò ÷ *�,�- ù ó ì ù���� ô¶ñ�ê�ë ÷cø�ù ð�ú ìAï,ð)ô õ½ö û ê�ë ÷cø�ù ð�ú ì�î ù���� ñ�ê�ò ÷ *�,�- ù ó ìAï$ðô õ½ö ü � �� � ýrþ½ÿ�Ôr� #Ëñ�É��� ��$��%��Õ�ÏIö��ñ�ê�ò ÷+*�,�- ù ó ì ï$ðô ë ö ì ó ð)ô õ û ëíì î�ï,ð ñ�ê�ò ÷+*�,�- ù ó ì ó ðô õ½ö ü � �� � ýrþ½ÿ�Ôr� #Ëñ�É��� ��$��>���!ö��
ñ�ëíìAó ðô�õ½ö ì ï$ðô ò û ñ�ëíì ï,ð)ô ò ö ìAó ð)ôxõ ü � �� � ýrþ½ÿ�Ôr� #Ëñ $&�)� ���'ö��ë�ì îIï,ð ñ õ ì î ó ð ê�ò ÷ *�,�- ù ó ö û ëíì î�ï,ð ñ�ê�ò ÷ *�,�- ù ó ìAó ðô�õ½ö ü � �� � ýrþ½ÿ�Ôr� #Ëñ�É��� ��$��%$&�!ö��
(230)
9 90:�;�E.�O��].mTd =Tl m�� j k dah ^ � npo ;>�������@B 90:�;�E.�O��].mTd =Tl�m�� j 90k5;�\Y].^%_[dah ^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9>9�:<;�E.�O��])mad =Tl�m�� j k d m�^ �"npo ;>�������@B 90:�;�E.�O��].mTd =Tl�m�� j 90k5;�\Y].^%_[d m�^ �"npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9>9�:<;>E.�r�>].mTd =al�m�� j k d ��^ � npo ;>�������@B 90:�;�E.�O��].mTd =Tl�m�� j 90k5;�\Y].^%_[d ��^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9>9�:<;>E.�r�>].mTd ]��OA�� j k d h ^ � npo ;>�������@B 90:�;�E.�O��].mTd ]��OA� j 9�kv;>\Y].^%_Td h ^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9 90:�;�E.�O��].mTd ]��OA� j k d m�^ � npo ;>�������@B 90:�;�E.�O��].mTd ]��OA� j 9�kv;>\Y].^%_Td m�^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9>90:�; E.�O��].m d ]��OA�@j k d ��^ � npo ; ������� B 90:�; E.�O��].m d ]��OA�@j 9�kv; \Y].^%_ d ��^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c90:�dgf�h ^rj k d m�^ �"npo ;>�������@B�:�dgf�h ^rj 90k5;�\Y].^%_Td m�^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c90:�dgfih ^rj k d ��^ � npo ;>�������@B�:�dgf�h ^rj 90k5;�\Y].^%_Td ��^ �1npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c90:�d f m�^ j k d h ^ � npo ;>�������@B�:�d f m�^ j 90kv;>\Y].^%_[d h ^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9�:�d f m�^ j k d ��^ � npo ;>�������@B�:�d f m�^ j 90kv;>\Y].^%_[d ��^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c90:�d f ��^ j k d h ^ � npo ;>�������@B�:�d f ��^ j 9�kv;>\Y].^%_[d h ^ � npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c9�:�dgf ��^rj k d m�^ �"npo ;>�������@B�:�dgf ��^rj 9�kv;>\Y].^%_[d m�^ �"npo G s SVu5wx`JILb�c!y jQ��o w � j �<ILb�c90:�; ������� B 90:�; E.�O��].m G ��M �O� stSVu5w | j0}��7� I�X{~ � M �O�90:�d h ^ � kv;>������� B 90:�;>�������tdah ^ � k G ��M �O� stSVu5w � jQ� M �r� ~�R�b o Z90:�d m�^ � kv;>������� B 90:�;>�������td m�^ � k G ��M �O� stSVu5w � jQ� M �r� ~�Xzb o Z90:�d ��^ � kv;>������� B 90:�;>�������td ��^ � k G ��M �O� stSVu5w � jQ� M �r� ~ N+b o Z90:�d fih ^ kv;>������� B�:�d f�h ^ 90kv; ������� G ��M �O� stSVu5w � jQ� M �r� ~�R�b o Z90:�dgf m�^ kv;>������� B�:�dgf m�^ 9�kv;>������� G ��M �O� stSVu5w � jQ� M �r� ~�Xzb o Z90:�dgf ��^ kv; ������� B�:�dgf ��^ 90k5; ������� G ��M �O� stSVu5w � jQ� M �r� ~ N+b o Z2Remark 27(Explanation of the FreeSVO package).TheFreeSVO packageex-
tends theMxSVO package,essentially by allowing theverbal headto occurin any
position aswell. Webriefly definetheinteraction with any elementsrequiring to be
placedin theWackernagel position,with themajorityof thecontrol structural rules
being placed in theWack package,(230). Although thenumber of rules in (230)
mayseemrather large, thereader should notethat these areexplicit instantiations
of only a handful of rule schemata- three, to beprecise. With theFreeSVO pack-
age,we obtain completely freeword orderwithin thedomainof a verbalhead, as
illustratedin thederivationsin (231). Languagesthatdisplay such behavior arefor
exampleCzechandRussian,at least in simplecases. I discussedaniniti al proposal
for dealing with a morecomplex caselike Czechclitics in (Kruijf f, 1999a).
A formalmodelof wordorderasstructuralindicationof informativity /151
(231) a.
subjÅÑÐ¥Ò ù ð+Ë Ç
verbÅÑÐ Ò *�,�- ù óOñ�ñ Ð Ò ù ð+Ë Ç�� îIï,ð �0ö�� ó ðô¥Ð Ò ÿ ù Ë Ç�öê
verb÷ *�,�- ù ó Å ñ Ð"Ò ù ð+Ë Ç�� îIï$ð �0ö�� ó ðô¥Ð"Ò ÿ ù Ë Ç
üxÐ"Ò��a�dobj
Å Ð"Ò ÿ ù Ë Çêverb
÷ *�,�- ù ó ìKó ðôdobj
Å Ð"Ò ù ðQË Ç�� î�ï,ð � ü ���a�subj
ì îIï,ð ñ�êverb
÷Q*�,�- ù ó ì ó ð)ôdobj
ö¥Åv� ü ���a�ñ�ê
verb÷ *�,�- ù ó ìKï,ðô
subjö ì¸ó ð)ô
dobjÅv� ü � �� � ýrþ¡ÿ�Ôr� #Ëñ�É��� ��$��>���!ö��
ñ�êverb
÷ � þ -+- ìKï,ðô subjö ìqó ð)ô
dobjÅv� ü � �� � ýrþ¡ÿ�� È ñ�É��� ��$�� /L�� � ö��
êverb
ì ï,ðôsubj
÷ � þ -Q- ì ó ð)ô dobjÅ5� ü � �� � ýrþ½ÿ�� � ñ0/L�� � �>���'ö��
ê�ñverb
ì ï,ðôsubj
ö ì¸ó ðôdobj
÷ � þ -+- Åv� ü � �� � ýrþ½ÿ�� � ñ0/L�� � �%$&�!ö��ê�ñ
verbì¸ï$ðô
subjö ì¸ó ðô
dobj÷ �1Ë �[Å5� ü Ì���Î � '(Ïg'�� /L�� � �
ñverb
ì ï,ðôsubj
ö ìAó ðôdobj
Å Ð Ò �1Ë � � üxÐ Ò Ó �
b.
...��Õ �� �ìÉî�ï,ðfñ�ê�É� .��� ÷ *�,�- ù ó ìKó ðô7$ Ö��� öYÅ �$ Ö��! �ìÉî ó ðfñ�ê�É� .��� ÷+*�,�- ù ó ì ï$ðô ��Õ �� öYÅ � ÌÍÏ5ýrþ¡ÿaØz� #Ëñ ���)��$&�'ö$ Ö��! Éì î ó ð ñ�ê�É� .��� ÷ � þ -+- ì ï$ðô ��Õ �� ö¥Å5� þ �� ��$ '�� /L�� � $ Ö��! Éì î ó ð ê�ñ�É� .���rìAï,ð)ô ��Õ �! ö�÷�� þ -Q- Å5� � �� � ýrþ¡ÿ�� � ñ0/L�� � �>���'öê $ Ö��! Éì î ó ð ñ�É� ����¶ìAï,ð)ô ��Õ �! ö�÷ � þ -Q- Å5� � �� � ýrþ¡ÿ�� � ñ0/L�� � �%$&�'öê $ Ö��! �ì î ó ð ñ�É� .���¶ìAï,ð)ô ��Õ �! ö�÷ �1Ë �gÅ5� Ì�ÎQÏ�'�� /L�� � $ Ö��� Éì î ó ð ñ�É� ����¶ìAï,ðô ��Õ �� ö¥Å#" Ò �1Ë � � Ó " Ò
c.
...��Õ �� �ì î�ï,ð ñ�ê�É� .��� ÷+*�,�- ù ó ìKó ðô7$ Ö��� öYÅ ���Õ �� �ìÉî�ï,ðfñ $ Ö��! �ì î ó ð ê�É� .��� ÷ *�,�- ù ó öYÅ � � �� � ýrþ¡ÿ�� #�ñ�É��� �$��+$&�'ö��Õ �� Ôì�î�ï,ðfñ $ Ö��! Éì î ó ð ê�É� .��� ÷�� þ -+- öYÅ5� þ �� ��$ '�� /L�� � $ Ö��! Éì î ó ð ñ ��Õ �� Ôì î�ï,ð ê�É� .��� ÷ � þ -+-öYÅ5� ÌÍÏ5ýrþ¡ÿT×� #Ëñ ���)��$&�'ö$ Ö��! Éì î ó ð ê�ñ ��Õ �� ÔìÉîIï,ð>É� ���� ö�÷�� þ -+- Å5� � �� � ýrþ¡ÿ�� � ñ0/L�� � �>���'öê $ Ö��! Éì î ó ð ñ ��Õ �� ÉìÉîIï,ð>É� ���� ö�÷�� þ -+- Å5� � �� � ýrþ¡ÿ�� � ñ0/L�� � �%$&�'öê $ Ö��! �ì î ó ð ñ ��Õ �� �ì î�ï,ð É� .��� ö�÷ � Ë � Å5� Ì�ÎQÏ�'�� /L�� � $ Ö��� Éì î ó ð ñ ��Õ �! �ìÉîIï$ð�É� ���� ö�Å#"aÒ �1Ë � � Ó "aÒ
4.4 MODELING THE STRATEGIES
The hypotheseswe formulated in Chapter3 are bestviewed as indicating how
strategies like word order or tunerealize acategory of informativity in aparticular
setting. In this section, we work out the structural rules defining thesestrategies
in more detail. Thesestructural rules extend the basic grammararchitectures I
developedin theprevioussection.
Theapproachto modeling theword order-basedstrategiesto realize informa-
tion structureis asfollows. We already discussedearlier therelation between con-
textual boundnessandthePraguianconception of systemicordering: wordgroups
realizing thecontextually nonbound dependentsof thefocusappear in canonical or
152é A formal model of wordorder asstructural indication of informativity
systemicordering. Thereis no suchconstraint on wordgroupsrealizing contextu-
ally bounddependents,whoseorderingis ratherdependent on theunderlying scale
of communicative dynamism. We alsoobserved the important role that systemic
ordering playsin focus projection.
Therefore, the first stepwe take is to rewrite the grammatical modes ¶6 , �{6 ,etc. into modesthat indicate whether or not the dependents are realized in sys-
temicordering. We introduceheadedmodes¶%$ (systemically ordered)and � ¶ (not
systemicallyordered) to indicatethis. Thenew ¶&$!·z� ¶ -structureabstractsnot only
from theparticular grammatical structure,but also–moreimportantly– from a lan-
guage’s specific systemic ordering.16 This enables to formulate a more general
account of word order asa structural indication of informativity. Oncewe have
a structure indicating the (non-)systemic ordering of dependents, we follow the
informativity hypotheses of Chapter3 to determine contextual boundnessof the
individual wordgroups.
Thestructuralrulesmodeling wordorder strategiesto realizeinformationstruc-
turecontrol themorebasic word orderrulesasfollows. Thegoalcategory we try
to prove states that the (verbal) headof the construction hasto have a specific
informativity, e.g. Â�¬O½�' . The structural rules definehow the contextual bound-
nessof the verbal headinfluences further possible distributions of contextually
bound/contextually nonbound valuesover thedependents. Whetherthese distribu-
tionsarethenderivablefrom thecanonical wordorder dependson thepossibilities
to vary word order.
Besidesbeing anatural wayto go in adependency grammar,definingthemod-
els in termsof systemic ordering ratherthanspecificgrammatical structuresnot
only yieldsa moregeneral perspective – it alsoprovidesfor a smoothintegration
with modelsof tuneasastructural indicationof informativity, asweshow in Chap-
ter 5.
Below we definethe fundamentalrulesfor the models, andillustratethemon
basic examplesrelating to the more linguistically oriented discussionin Chapter
2. Thedefinitionsonly elaboratea simplemapping to ¶&$�·z� ¶ -structures,andonly
consider relatively shallow structures. This is no inherent theoretical problem, as
thedefinitionscanbe(monotonically) extendedto cover morecomplex structures.
16Languagesmay show differencesin their systemicorderings;only within a single language,its systemicorderingis considered universal,(Sgall et al., 1986; Sgall et al., 1995). The mappingfrom modeslike
���,$&�
to� Ö �+Ǧ�
canbe madesensitive to a language’s specificsystemicordering.However, theresulting
� Ö �.Ǧ�-structureis independentof thatspecificsystemicordering,sinceit is
formulatedpurelyin termsof whetheror not somesystemicorderingis obeyed.
A formalmodelof wordorderasstructuralindicationof informativity /153
Definition 30 (Structural indications of informativity in OV). The InfOV de-
finesthebasic structural rulesfor describing structural indication of informativity
in OV languages,given INFHYP1 and INFHYP2. The InfOV monotonically ex-
tendsall theOV packages,andonly definesrulesthat specify feature information.
Thestructural rulesof InfOV are below, without anyreferenceto tune(asweonly
definetune in thenext chapter).
ê�ë�ìÉîIïQ��ñ�ò�ìÉîIïQ� õ½ö�÷ �1Ë � û ê�ë�ìÉîIï$ð/ñ�ò�ì î ó ð�õ½ö�÷ �1Ë � ü Ó Ç / ÿ¡þ¥� ý � �'ÿ � $O�ê�ñ�ë�ìÉîIïQ�0ò ö ìAïQ�'ô õ½÷ �1Ë � û ê�ñ�ë�ìÉîIï$ð�ò ö ìAó ðô�õ½÷ �1Ë � ü Ó Ç / ÿ¡þ¥� ý � �'ÿ � $O�ê�ñ�ë�ì ïQ�'ô ò ö ì ïQ�'ô õ½÷ �1Ë � û ê�ñ�ë�ì ï,ð)ô ò ö ì ó ðô õ½÷ �1Ë � ü Ó Ç / ÿ¡þ¥� ý � �'ÿ � $O�ê�ñ�ëíì õ ïô ò ö ì õ ïô õ½÷ �1Ë � û ê�ñ�ë�ì¸ó ðô ò ö ì ï,ðô õ½÷ �1Ë � ü Ó Ç / ÿ¡þ¥� ( Ö Ç¬ý � �'ÿ � $O�ê�ëíì î õ ï ñ�ò�ì î õ ï õ½ö�÷ �1Ë � û ê�ë�ì î ó ð ñ�ò�ì îIï,ð õ½ö�÷ �1Ë � ü Ó Ç / ÿ¡þ¥� ( Ö Ç¬ý � �'ÿ � $O�
ê�ë�ìÉîIï���ñ�ê�ò ÷ õ�)!* ìÉîIïQ� õ½ö�÷ �1Ë � û ê�ë�ìÉîIï���ñ�ê�ò ÷ � õ � ìÉîIïQ� õ½ö�÷ � Ë � ü Ó Ç / ÿ¡þ¥� Ó#,+a�� þ¥� õ-� + �ê�ëíìÉî õ ï�ñ�ê�ò ÷ õ�).* ì�î õ ï õ½ö�÷ �1Ë � û ê�ë�ìÉî õ ï�ñ�ê�ò ÷ � õ � ìÉî õ ï õ½ö�÷ �1Ë � ü Ó Ç / ÿ¡þ¥� Ó#,+a�� þ¥� õ-� + �
ê�ñ�ê�ë ÷ õ�)!* ìAïQ�'ô¸ê�ò ÷ ð ) ö ìAïQ�'ô õ½÷ �1Ë � û ê�ñ�ê�ë ÷ � õ � ìAïQ�'ôqê�ò ÷ � õ ��ö ìAïQ�'ô õ½÷ � Ë � ü Ó Ç / ÿ¡þ¥� þ ���� � Ö � Õ����ê�ñ�ê�ë ÷ õ�)!* ì õ ïôqê�ò ÷ ð ) ö ì õ ïô õ½÷ �1Ë � û ê�ñ�ê�ë ÷ � õ � ì õ ï)ôAê�ò ÷ � õ � ö ì õ ïô õ½÷ �1Ë � ü Ó Ç / ÿ¡þ¥� þ ���� � Ö � Õ����ñ�ëíì õ ï)ô ê�ò ÷ ð ) ö ì õ ïô ê õ½÷ ð ) û ñ�ë�ì õ ï)ô ê�ò ÷ ð ) ö ì õ ïô ê õ½÷ � õ � ü Ó Ç / ÿ¡þ¥� þ ���� � Ö � Õ����
ê�ñ�ë�ì îIï�� ò ö ì ïQ�'ô ê õ½÷ õ�)!* ÷ �1Ë � û ê�ñ�ë�ì îIï�� ò ö ì ïQ�'ô ê õ½÷ � õ � ÷ � Ë � ü Ó Ç / ÿ¡þ¥� +aÖ ��Î7þY� � + �ê�ë ÷ õ�).* ìÉî�ïQ��ê�ò ÷ õ�) û ê�ë ÷ õ�)!* ì�îIïQ��ê�ò ÷ � õ � ü Ó Ç / ÿ¡þ¥� / �@� $�� +a�)Ö0 �ê�ë ÷ õ�) ìÉîIïQ��ñ�ê�ò ÷ õ�).* ìÉî�ïQ� õ½ö û ê�ë ÷ � õ � ì�îIïQ��ñ�ê�ò ÷ õ�)!* ì�î�ïQ� õ½ö ü Ó Ç / ÿ¡þ¥� 1 � � $�� +a�)Ö� �ê�ë ÷ ð ) ì�î�ïQ�%ê�ò ÷ ð ) û ê�ë ÷ � õ � ì�îIïQ��ê�ò ÷ ð ) ü Ó Ç / ÿ¡þ¥� 2 +a�)Ö� �ê�ë ÷ ð ) ì�î õ ï/ê�ò ÷ ð ) û ê�ë ÷ � õ � ì�î õ ï/ê�ò ÷ ð ) ü Ó Ç / ÿ¡þ¥� 2 +a�)Ö� �
ñ�ë�ì ïQ�'ô ê�ò ÷ ð ) ö ì ïQ�'ô ê õ½÷ ð ) û ñ�ë�ì ïQ�'ô ê�ò ÷ ð ) ö ì ïQ�'ô ê õ½÷ � õ � ü Ó Ç / ÿ¡þ¥� 2 +a�)Ö� �ê�ë ÷ õ�).* ì î�ïQ� ê�ò ÷ ð ) û ê�ë ÷ õ�)!* ì îIïQ� ê�ò ÷ � õ � ü Ó Ç / ÿ¡þ¥� / �@� $�2 +a��Ö� �ê�ë ÷ õ�).* ì î õ ï ê�ò ÷ ð ) û ê�ë ÷ õ�)!* ì î õ ï ê�ò ÷ � õ � ü Ó Ç / ÿ¡þ¥� / �@� $�2 +a��Ö� �
ê�ë ÷ ð ) ìÉîIïQ��ñ�ê�ò ÷ õ�).* ìÉî�ïQ� õ½ö û ê�ë ÷ � õ � ì�îIïQ��ñ�ê�ò ÷ õ�)!* ì�î�ïQ� õ½ö ü Ó Ç / ÿ¡þ¥� ýrÿ ò[Ö Õ�Ç�$&� ��� �ñ�ë�ìÉî�ïQ��ê�ò ÷ ð ) ö ìAï��0ôqê õ½÷ õ�).* û ñ�ë�ìÉîIï���ê�ò ÷ � õ ��ö ìAïQ�'ô¸ê õ½÷ õ�)!* ü Ó Ç / ÿ¡þ¥� ýrÿ ò[Ö Õ�Ç�$&� ��� �
ê�ê�ë ÷ ð ) ìÉî�ïQ�0ò ÷ ð ) û ê�ë ÷ ð ) ìÉî�ïQ��ê�ò ÷ ð ) ü Ó Ç / ÿ¡þ¥� � ñ � � ��3 � Ö ö��ê�ë�ìAï��0ô�ò ÷ ð ) û ê�ë ÷ ð ) ìAïQ�'ô�ò ü Ó Ç / ÿ¡þ¥� � ñ � � ��� Ö54 ö��ê�ë�ì ï��0ô ò ÷ õ�) û ê�ë ÷ õ�) ì ïQ�'ô ò ü Ó Ç / ÿ¡þ¥� � ñ Ç � ��� Ö64 ö��ê�ë�ì õ ï)ô ò ÷ õ�) û ê�ë ÷ õ�) ì õ ï)ô ò ü Ó Ç / ÿ¡þ¥� � ñ Ç � �>Ç7� 4 ö��ê�ëíì õ ïô�ò ÷ ð ) û ê�ë ÷ ð ) ì õ ïôxò ü Ó Ç / ÿ¡þ¥� � ñ � � � Ç7� 4 ö��
ê�ê�ë ÷ ð ) ì�î õ ï�ò ÷ ð ) û ê�ë ÷ ð ) ìÉî õ ï�ê�ò ÷ ð ) ü Ó Ç / ÿ¡þ¥� � ñ � � ��3 Ç7�0ö��ê�ñ�ê�ë ÷ ð ) ìÉî õ ï/ê�ò ÷ ð ) ö ìÉî õ ï õ½÷ õ�).* û ñ�ê�ë ÷ ð ) ìÉî õ ï�ê�ò ÷ ð ) ö ìÉî õ ï/ê õ½÷ õ�).* ü Ó Ç / ÿ¡þ¥� � ñ Ç ��7 ��3 Ç7�0ö��ê�ê�ë ÷ õ�).* ìÉî�ïQ�0ò ÷ õ�) û ê�ë ÷ õ�)!* ì�îIïQ��ê�ò ÷ õ�) ü Ó Ç / ÿ¡þ¥� � ñ Ç � ��3 � Ö ö��
ê�ê�ë ÷ ù ð,ð ÷ � õ � û ê�ê�ë ÷ � õ � ÷ ù ð,ð ü Ó Ç / ÿ¡þ¥� � ñ '(Ç / �>�����!ö��ê�ê�ë ÷ õ � � ÷ � õ � û ê�ê�ë ÷ � õ � ÷ õ �>� ü Ó Ç / ÿ¡þ¥� � ñ '(Ç / �>Ç Ö�# ö��ê�ê�ë ÷ ð ) ì î�ïQ� ò ÷ õ�) û ê�ë ÷ ð ) ì î�ïQ� ê�ò ÷ õ�) ü Ó Ç / ÿ¡þ¥� � ñ Ç � ��3 � Ö ö��
ê�ê�ë ÷ õ�) ìÉî�ïQ�0ò ÷ õ�) û ê�ë ÷ õ�) ìÉîIï���ê�ò ÷ õ�) ü Ó Ç / ÿ¡þ¥� � ñ Ç � ��3 � Ö ö��ê�ë�ì ïQ�'ô ê�ò ÷ ð ) ÷ õ�)!* û ê�ë ÷ õ�)!* ì ïQ�'ô ê�ò ÷ ð ) ü Ó Ç / ÿ¡þY� � ñ Ç �7 ��� Ö64 ö��ê�ëíì õ ïô¸ê�ò ÷ ð ) ÷ õ�)!* û ê�ë ÷ õ�)!* ì õ ïô¸ê�ò ÷ ð ) ü Ó Ç / ÿ¡þY� � ñ Ç �7 �>Ç7� 4 ö��ê�ê�ë ÷ õ�).* ì�î õ ï�ò ÷ ð ) û ê�ë ÷ õ�)!* ìÉî õ ï�ê�ò ÷ ð ) ü Ó Ç / ÿ¡þY� � ñ � � �3 Ç7�0ö��ê�ê�ë ÷ õ�)!* ì�î�ïQ�'ò ÷ ð ) û ê�ë ÷ õ�)!* ìÉîIïQ��ê�ò ÷ ð ) ü Ó Ç / ÿ¡þY� � ñ � � �3 � Ö ö��
154é A formal model of wordorder asstructural indication of informativity
2
Remark 28(Explanation of the InfOV package).To begin with, theInfOV pack-
ageaswe defineit above doesnot have explicit referenceto tune yet. Thereason
being thatwe only deal with themodeling of tunein thenext chapter. Thus,sofar
we only capturetheeffect of word order.
Thestrategy employedin theInfOV packageis ratherstraightforward.Thegoal
category should specify theinformativity of theverbal head.Technically, whatwe
dothenis thatwefirst assignthefocusproper (nb*), andsubsequently distributecb
andnb featuresasappropriate to theconstruction (word order) we have. We need
to endupwith adistribution that assignstheverbal headacontextual boundnessas
required by thegoal category.
To that end,we specify percolation rules that distribute features (the p(X,Y)
rules), and linkagerules that specify features. The ImmPreV.CFP rules realize
the focus proper in the canonical focus position, being immediately before the
verb. The PostV.FP rule covers the case of a postverbally realized focus proper,
as possible in non-rigid verb-final languageslike Sinhala(Herring and Paolillo,
1995). The VerbFocus rules cover cases observable in free OV languageslike
Turkish, Hungarian andHindi wheretheverbformsthefocusproper if thereis no
preverbally placeddependent. Oncewe have establishedthefocus proper,we can
projectthetopic (TProj) andthefocus (FProj).
Froma morelinguistic perspective, we should reada proof bottom-up. At the
bottom, we find a sentencewith given indicationsof informativity. Then,reading
upwards, we seehow the sentence’s information structure determinesits actual
word order, aswe reason how the observed word order canbe establishedon the
basis of thecanonicalword order- a view akin to thegenerative perspective taken
in FGD, cf. Sgallet al’s discussionin (1986) andPetkevic (in prep).
Example. First, let us consider various unmarked cases,wherethe focus proper
is realized in the immediately preverbalposition. The proof in (232) shows the
realizationof an“out-of-the-blue” sentence.
A formalmodelof wordorderasstructuralindicationof informativity /155
(232)
.
.
.ß�ß�ß subjcià�8 åQä à ë 8�9�Ú ÛzÜQÝ Þ�ß�ß dobjcià(â Ý�Ý à ë 8�9YÚ Ûzæ�Ý ß tverbià ë 8�9)ê�à ä���� ì ïß�ß�ß subjci à 8 åQä à ë 8�9 Ú ÛzÜQå Þ�ß�ß dobjcià â Ý0Ý à ë 8�9 Ú ÛzÜ�å ß tverbià ë 8�9 ê�à ä���� ì ï í õ � ñiðò ï�:ï ñ þ ó ù
ß�ß�ß subjcià�8 åQä à ë 8�9YÚ ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå ß tverbià ë 8�9)ê�à ä>�?� ì ï�í õ � ñiðò �1�A@ þ - ðò B�C @ ù
ß�ß�ß subjcià�8 åQä à ë 8�9�Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜ�å ß tverbiàD8�;�ê�à ä��?� ì ï í õ � ñiðò E ø þ ó C @ þ �.F ùß�ß�ß subjcià�8 åQä à�8�;!Ú¦ÛzÜQå�Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú¦ÛzÜQå�ß tverbiàD8�;%ê�à ä��?� ì ïJí õ � ñiðò G ø þ ó C @ þ �HF ùß�ß�ß subjcià 8 å�ä à 8�; Ú ÛzÜQå ß�ß�ß dobjcià â Ý0Ý à 8�;=< Ú ÛzÜQå tverbià 8�; à ä���� ì ï í õ � ñiðò ÿ Þ õ�)%ø î�ïQ� ê�ùß�ß�ß�ß subjcià�8 åQä à�8�;�Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbiê�àD8�;%à ä��?� ì ïýí õ � ñiðò ÿ Þ õ�)%ø
î�ïQ� ê�ùß�ß�ß subjcià�8 åQä à�8�;!Ú ÛzÜ�å Þ�ß�ß dobjcià(â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbiê�àD8�;7ì � � ä���� ï í � �0 ùß�ß subjcià�8 å�ä à�8�;!Ú¦ÛzÜQå�Þ�ß�ß dobjcià(â Ý�Ý à�8�;H<iÚ¦ÛzÜ�å tverbiêì � � 8�; � � ä��?� ï í
� �0 ù
Thenext examplesshowshow wecanrealizeanarrow focus,beingconstituted
by just thefocus proper (233), or a wider focusincluding moredependents, (234).
(233)
.
.
.ß�ß�ß subjcià 8 åQä à ë 8�9 Ú¦ÛzÜQÝ�Þ�ß�ß dobjcià â Ý�Ý à ë 8�9 Ú Ûzæ�Ý ß tverbià ë 8�9 ê�à ä���� ì ïß�ß�ß subjci à�8 åQä à ë 8�9�Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à ë 8�9¥Ú ÛzÜ�å ß tverbià ë 8�9.ê�à ä���� ì ï�í õ � ñiðò ï�:ï ñ þ ó ù
ß�ß�ß subjcià�8 åQä à ë 8�9YÚ¦ÛzÜQå�Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú¦ÛzÜQå�ß tverbià ë 8�9)ê�à ä>�?� ì ï�í õ � ñiðò �1�A@ þ - ðò B�C @ ù
ß�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå ß tverbià ë 8�9�ê�à ä��?� ì ï í õ � ñiðò ï�ñJI � � õ ó ù þ : ùß�ß�ß subjci à�8 åQä à Ý ;!Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå ß tverbià Ý ;%ê�à ä��?� ì ï í õ � ñiðò E ø þ ó�K @ þ �HF ùß�ß�ß subjci à�8 å�ä à Ý ;!Ú ÛzÜQå ß�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbià Ý ;%à ä���� ì ï í õ � ñiðò ÿ Þ ð )%ø î�ïQ� ê�ùß�ß�ß�ß subjci à�8 åQä à Ý ;!Ú¦ÛzÜQå�Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú¦ÛzÜQå tverbiê�à Ý ;%à ä��?� ì ï�í õ � ñiðò ÿ Þ
ð )%ø î�ïQ� ê�ùß�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜ�å Þ�ß�ß dobjcià(â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbiê�à Ý ;iì � � ä���� ï í � �0 ùß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜQå Þ�ß�ß dobjcià(â Ý�Ý à�8�;H<iÚ ÛzÜ�å tverbiêì � � Ý ; � � ä��?� ï í
��� ù
(234)
.
.
.ß�ß�ß subjcià�8 åQä à ë 8�9�Ú ÛzÜQÝ Þ�ß�ß dobjcià(â Ý�Ý à ë 8�9YÚ Ûzæ�Ý ß tverbià ë 8�9)ê�à ä���� ì ïß�ß�ß subjci à 8 åQä à ë 8�9 Ú¦ÛzÜQå)Þ�ß�ß dobjcià â Ý0Ý à ë 8�9 Ú¦ÛzÜ�å.ß tverbià ë 8�9 ê�à ä���� ì ï í õ � ñiðò ï�:ï ñ þ ó ù
ß�ß�ß subjcià�8 åQä à ë 8�9YÚ ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå ß tverbià ë 8�9)ê�à ä>�?� ì ï í õ � ñiðò �1�A@ þ - ðò B�C @ ù
ß�ß�ß subjcià�8 åQä à ë 8�9¥Ú ÛzÜQå Þ�ß�ß dobjcià(â Ý�Ý à�8�;=<¦Ú ÛzÜ�å ß tverbià Ý ;�ê�à ä��?� ì ï í õ � ñiðò E ø þ óLK @ þ �HF ùß�ß�ß subjcià�8 åQä à�8�;!Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<iÚ ÛzÜQå ß tverbià Ý ;�ê�à ä��?� ì ï í õ � ñiðò G ø þ ó C @ þ �HF ùß�ß�ß subjcià 8 å�ä à 8�; Ú7ÛzÜ�å)ß�ß�ß dobjcià â Ý0Ý à 8�;=< Ú¦ÛzÜQå tverbià Ý ; à ä���� ì ï í õ � ñiðò ÿ Þ ð )%ø î�ïQ� ê�ùß�ß�ß�ß subjcià�8 åQä à�8�;!Ú ÛzÜQå Þ�ß�ß dobjcià�â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbiê�à Ý ;>à ä��?� ì ï í õ � ñiðò ÿ Þ
ð )%ø î�ïQ� ê�ùß�ß�ß subjcià�8 åQä à�8�;!Ú ÛzÜ�å Þ�ß�ß dobjcià(â Ý0Ý à�8�;=<¦Ú ÛzÜQå tverbiê�à Ý ;iì � � ä���� ï í ��� ùß�ß subjcià�8 å�ä à�8�;�Ú ÛzÜ�å Þ�ß�ß dobjcià(â Ý�Ý à�8�;H<7Ú ÛzÜ�å tverbiêì � � Ý ; � � ä��?� ï í
� �0 ù
2Example. The structural rules in InfVO also allow for more marked cases, ob-
tainable through word orderalone. For example,we canrealize just the verb as
focusby placing it sentence-initially. We canobserve this behavior in for example
Hungarian(Vallduvı andEngdahl, 1996) or Turkish(Hoffman,1995a). Examples
(235)and(236) differ in theorderin which thedependents arerealized.
156é A formal model of wordorder asstructural indication of informativity
(235)
.
.
.ß�Þ�ß tverbià ë 8�9¥Ú æ�Ý!M ß�ß dobjcià�â Ý0Ý à ë 8�9�ê�Ú ÜQÝ!M ß�ß subjcià�8 å�ä à ë 8�9)à ä��?� ì ïß�Þ�ß tverbià ë 8�9 Ú 8 Ü�M ß�ß dobjcià â Ý�Ý à ë 8�9 êOÚ 8 Ü�M ß�ß subjcià 8 åQä à ë 8�9 à ä>�?� ì ï í õ � ñiðò N� õOïO: ï ñ þ ó ù
ß�Þ�ß tverbiàD8�;=<¦Ú 8 Ü�M ß�ß dobjcià(â Ý�Ý à Ý ;>ê�Ú 8 Ü�M ß�ß subjcià�8 åQä à ë 8�9�à ä��?� ì ï í õ � ñiðò ð - þ ).C �7ð � ï ùß�Þ�ß tverbiàD8�;=<¦Ú 8 ÜPM ß�ß dobjcià(â Ý�Ý à Ý ;�êrÚ 8 Ü�M ß�ß subjcià�8 åQä à Ý ;�à ä���� ì ï í õ � ñiðò ð - þ ).C �7ð � ï ùß�ß tverbi Ú 8 Ü�Miß�ß dobjcià�â Ý0Ý à Ý ;�à�8�;=<¦Ú 8 Ü�Miß�ß subjcià�8 å�ä à Ý ;�à ä���� ì ï í õ � ñ¦ðò ÿ Þ õ�).*�ø�õ ïxô ê�ùß�ß�Þ tverbi Ú 8 Ü�M ß�ß dobjcià â Ý0Ý à Ý ; ê�Ú 8 ÜPM ß�ß subjcià 8 åQä à Ý ; à 8�;=< à ä���� ì ï í õ � ñ¦ðò ÿ Þ õ�).*�ø�õ
ïxô ê�ùß�Þ tverbi Ú 8 ÜPM ß�ß dobjcià(â Ý0Ý à Ý ;�êrÚ 8 ÜPM ß�ß subjcià�8 åQä à Ý ;%à�8�;=<"ì � � ä���� ï í � � ùÞ tverbi Ú 8 Ü�M ß�ß dobjcià(â Ý�Ý à Ý ;>ê�Ú 8 Ü�M ß�ß subjcià�8 åQä à Ý ;¦ì � � 8�;=< � � ä��?� ï í
� �0 ù
(236)
.
.
.ß�Þ�ß tverbià ë 8�9 Ú ÜQÝ!M ß�ß subjcià 8 åQä à ë 8�9 ê�Ú æ>Ý!M ß�ß dobjcià â Ý�Ý à ë 8�9 à ä���� ì ïß�Þ�ß tverbià ë 8�9YÚ�ÜQåQM¦ß�ß subjcià�8 å�ä à ë 8�9)ê�Ú{Ü�åQMiß�ß dobjcià(â Ý�Ý à ë 8�9)à ä>�?� ì ïîí õ � ñiðò ïO:ï ñ þ ó ù
ß�Þ�ß tverbiàD8�;=<¦Ú ÜQåQM ß�ß subjcià�8 åQä à Ý ;�êrÚ ÜQåPM ß�ß dobjcià�â Ý0Ý à ë 8�9)à ä��?� ì ï í õ � ñiðò ð - þ ).C �7ð � ï ùß�Þ�ß tverbiàD8�;=<¦Ú ÜQåPM ß�ß subjcià�8 å�ä à Ý ;�êrÚ Ü�åQM ß�ß dobjcià(â Ý�Ý à Ý ;�à ä���� ì ï í õ � ñiðò K @ þ �HF ùß�ß tverbi Ú ÜQåQM ß�ß subjcià�8 åQä à Ý ;�à�8�;=<¦Ú ÜQåPM ß�ß dobjcià�â Ý0Ý à Ý ;�à ä���� ì ï í õ � ñ¦ðò ÿ Þ õ�).*�ø ïQ�qô ê�ùß�ß�Þ tverbi Ú{ÜQåQMiß�ß subjcià�8 åQä à Ý ;�êrÚ�ÜQåQM¦ß�ß dobjcià�â Ý0Ý à Ý ;�à�8�;=<�à ä���� ì ïýí õ � ñ¦ðò ÿ Þ õ�).*�ø
ïQ�qô ê�ùß�Þ tverbi Ú Ü�åQM ß�ß subjcià�8 åQä à Ý ;�ê�Ú Ü�åQM ß�ß dobjcià(â Ý�Ý à Ý ;>à�8�;=<¥ì � � ä���� ï í � �� ùÞ tverbi Ú ÜQåQM ß�ß subjcià�8 å�ä à Ý ;%êOÚ ÜQåPM ß�ß dobjcià�â Ý0Ý à Ý ;7ì � � 8�;=< � � ä���� ï í
��� ù
Finally, consider the post-verbally realized focus proper in (237). We illus-
trated suchconstructionsin Chapter 3 on Sinhala,a non-rigid verb-final language.
(237)
.
.
.ß�Þ�ß�ß subjcià ë 8�9 à 8 åQä Ú¦ÛzÜQÝ�ß tverbià ë 8�9 êrÚ æ�Ý�M ß�ß dobjcià ë 8�9 à â Ý�Ý à ä���� ì ïß�Þ�ß�ß subjcià�8 åQä à ë 8�9¥Ú ÛzÜQÝ ß tverbià ë 8�9)êrÚ æ�Ý�M ß�ß dobjcià ë 8�9�à�â Ý�Ý à ä���� ì ï í õ � ñ¦ðò ÿ Þ� õ � ø0õ �>� ê�ù
ß�Þ�ß�ß subjcià�8 åQä à ë 8�9¥Ú¦ÛzÜQÝ�ß tverbià ë 8�9)êrÚ æ�Ý�M ß�ß dobjcià�â Ý0Ý à ë 8�9)à ä���� ì ï�í õ � ñ¦ðò ÿ Þ� õ � ø ù ð,ð ê ù
ß�Þ�ß�ß subjcià�8 å�ä à ë 8�9¥Ú ÛzÜ�å ß tverbià ë 8�9.êOÚ ÜQåQM ß�ß dobjcià(â Ý0Ý à ë 8�9�à ä���� ì ïýí õ � ñið�ò ïO:ï ñ þ ó ù
ß�Þ�ß�ß subjcià�8 åQä à ë 8�9�Ú ÛzÜ�å ß tverbià ë 8�9)êrÚ ÜQåQM ß�ß dobjcià�â Ý0Ý à�8�;=<�à ä���� ì ïýí õ � ñiðò@��)ï+Ë ð�ò C @ ù
ß�Þ�ß�ß subjcià 8 åQä à ë 8�9 ÚiÛzÜ�å.ß tverbià Ý ; êrÚ�ÜQåQM�ß�ß dobjcià â Ý0Ý à 8�;=< à ä���� ì ï í õ � ñiðò ïñ�I � � õ ó ù þ : ùß�Þ�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜ�å ß tverbià Ý ;>ê�Ú ÜQåQM ß�ß dobjcià(â Ý0Ý à�8�;=<�à ä>�?� ì ï í õ � ñiðò K @ þ �HF ùß�ß�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜ�å tverbià Ý ;!Ú Ü�åQM ß�ß dobjcià(â Ý�Ý à�8�;H<�à ä��?� ì ï í õ � ñiðò ÿ Þ ð )�ø î�ï�� ê ùß�ß�Þ�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜ�å tverbiêrÚ Ü�åQM ß�ß dobjcià�â Ý0Ý à�8�;=<�à Ý ;�à ä>�?� ì ïîí õ � ñiðò ÿ Þ
ð )�ø ï��qô ê ùß�Þ�ß�ß subjcià 8 å�ä à Ý ; Ú¦ÛzÜQå tverbiêrÚ{ÜQåQM¦ß�ß dobjcià â Ý0Ý à 8�;=< à Ý ; ì ��� ä���� ï í � �0 ùÞ�ß�ß subjcià�8 åQä à Ý ;!Ú ÛzÜQå tverbiê�Ú Ü�åQM ß�ß dobjcià(â Ý�Ý à�8�;H<Yì � � Ý ; � � ä>�?� ï í
� � ù2
Definition 31(Structural indicationsof informativity in VO). TheInfVO defines
the basic structural rules for describing structural indication of informativity in
VO languages, given INFHYP1 and INFHYP2. TheInfVO monotonically extends
all the VO packages, and only definesrules that specify feature information. The
basic setof structural rulesof InfVO is given below, without anyreferenceto tune.
Becauseof lack of sufficient data, asweindicatedalready in Chapter3, the InfVO
is smallerthanthe InfOV and InfSVO packages.
A formalmodelof wordorderasstructuralindicationof informativity /157
ê�ñ�ëíìAïQ�'ô�ò ö ìAïQ�'ô õ½÷ �1Ë � û ê�ñ�ë�ìAï,ð)ô�ò ö ìAó ðô�õ½÷ �1Ë � ü Ó Ç / þ�ÿ�� ý � �0ÿ � $O�ê�ñ�ëíìÉîIïQ�0ò ö ìAïQ�'ô õ½÷ �1Ë � û ê�ë�ìÉîIï$ð/ñ�ò�ìAó ðô�õ½ö�÷ �1Ë � ü Ó Ç / þ�ÿ�� ý � �0ÿ � $O�ê�ñ�ëíì�î õ ï�ò ö ì õ ï)ô õ½÷ �1Ë � û ê�ë�ì î ó ð ñ�ò�ìKï,ðô õ½ö�÷ �1Ë � ü Ó Ç / þ�ÿ�� ( Ö Ç¬ý � �'ÿ � $O�ê�ñ�ëíì õ ïô�ò ö ì õ ï)ô õ½÷ �1Ë � û ê�ñ�ë�ì¸ó ðô ò ö ìAï,ðô õ½÷ �1Ë � ü Ó Ç / þ�ÿ�� ( Ö Ç¬ý � �'ÿ � $O�
ê�ë�ì ïQ�'ô ê�ò ÷ õ�)!* ÷ �1Ë � û ê�ë�ì ïQ�'ô ê�ò ÷ � õ � ÷ �1Ë � ü Ó Ç / þ�ÿ�� +aÖ ��Î�þ¥� õ-� + �ê�ëíì õ ïô ê�ò ÷ õ�)!* ÷ �1Ë � û ê�ë�ì õ ï)ô ê�ò ÷ � õ � ÷ �1Ë � ü Ó Ç / þ�ÿ�� +aÖ ��Î�þ¥� õ-� + �ñ�ëâì ï��0ô ê�ò ÷ õ�) ö ì ï��0ô ê õ½÷ õ�).* û ñ�ë�ì ïQ�'ô ê�ò ÷ � õ � ö ì ïQ�'ô ê õ½÷ õ�)!* ü Ó Ç / þ�ÿ�� � +a�)Ö0 �ê�ë ÷ õ�) ìAï��0ôqê�ò ÷ õ�) û ê�ë ÷ � õ � ìKïQ�'ô¶ê�ò ÷ õ�) ü Ó Ç / þ�ÿ�� � +a�)Ö0 �ê�ë ÷ ð ) ì õ ïôqê�ò ÷ ð ) û ê�ë ÷ � õ � ì õ ïôqê�ò ÷ ð ) ü Ó Ç / þ�ÿ�� 2 +a��Ö� �ê�ë ÷ ð ) ì�î õ ï/ê�ò ÷ ð ) û ê�ë ÷ � õ � ì�î õ ï/ê�ò ÷ ð ) ü Ó Ç / þ�ÿ�� 2 +a��Ö� �ê�ë ÷ ð ) ì ï��0ô ê�ò ÷ õ�) û ê�ë ÷ � õ �Ôì ïQ�'ô ê�ò ÷ õ�) ü Ó Ç / þ�ÿ�� ýxÿ ògÖ Õ�Ç�$&� ��� �ñ�ë�ì î õ ï ê�ò ÷ ð ) ö ì õ ï)ô ê õ½÷ õ�).* û ñ�ë�ì î õ ï ê�ò ÷ � õ ��ö ì õ ïô ê õ½÷ õ�)!* ü Ó Ç / þ�ÿ�� (Lýxÿ ò[Ö Õ�Ç�$&� ��� �
ñ�ë�ì õ ïô ê�ò ÷ ð ) ö ì õ ï)ô ê õ½÷ õ�).* û ñ�ë�ì õ ï)ô ê�ò ÷ � õ � ö ì õ ïô ê õ½÷ õ�)!* ü Ó Ç / þ�ÿ�� (Lýxÿ ò[Ö Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ìÉî�ïQ��ê�ò ÷ ð ) ö ìAï��0ôqê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉîIï���ê�ò ÷ � õ ��ö ìAïQ�'ô¸ê õ½÷ õ�)!* ü Ó Ç / þ�ÿ�� +a�� þ ���� � ògÖ Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ì�î�ïQ�%ê�ò ÷ õ�) ö ìAï��0ôqê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉîIï���ê�ò ÷ � õ ��ö ìAïQ�'ô¸ê õ½÷ õ�)!* ü Ó Ç / þ�ÿ�� +a�� þ ���� � ògÖ Õ�Ç�$&� ��� �ê�ë�ìAï��0ô�ò ÷ õ�) û ê�ë ÷ õ�) ìAïQ�'ô�ò ü Ó Ç / þ�ÿ�� � ñ Ç � ��� Ö64 ö��
ê�ê�ë ÷,ù ð,ð ÷ � õ � û ê�ê�ë ÷ � õ ��÷,ù ð,ð ü Ó Ç / þ�ÿ�� � ñ '(Ç / ���� �'ö��ê�ê�ë ÷ õ � � ÷ � õ � û ê�ê�ë ÷ � õ � ÷ õ �>� ü Ó Ç / þ�ÿ�� � ñ '(Ç / ��Ç Ö)# ö��
ê�ê�ë ÷ ð ) ì î õ ï ò ÷ ð ) û ê�ë ÷ ð ) ì î õ ï ê�ò ÷ ð ) ü Ó Ç / þ�ÿ�� � ñ � � �03 Ç7�0ö��ê�ëíì õ ïô�ò ÷ ð ) û ê�ë ÷ ð ) ì õ ï)ô�ò ü Ó Ç / þ�ÿ�� � ñ � � �%Ç7� 4 ö��ê�ëíìKïQ�'ô¬ò ÷ ð ) û ê�ë ÷ ð ) ìKïQ�'ô¬ò ü Ó Ç / þ�ÿ�� � ñ � � �>� ÖR4 ö��
ê�ê�ë ÷ ð ) ì�î�ïQ�'ò ÷ ð ) û ê�ë ÷ ð ) ì�îIï��%ê�ò ÷ ð ) ü Ó Ç / þ�ÿ�� � ñ � � �03 � Ö ö��ê�ê�ë ÷ ð ) ìÉî�ïQ�0ò ÷ õ�) û ê�ë ÷ ð ) ì�îIï��%ê�ò ÷ õ�) ü Ó Ç / þ�ÿ�� � ñ Ç � ��3�� Ö ö��
S
Example (VO realization of information structur e). We give threebrief exam-
ples- canonical word order (238), noncanonical word order(239), andnon-rigid
verb-firstness(240).
(238)
...9 j 9 tverbi; ����<dah ^ � 9>9 subji
;>�zl>=�;����� o d m�^ � 9>9 dobji;�].^%^ ;>���z�L;>= ?(AUT©R
9 j 9 tverbi; ���z��d h l.� 9 9 subji
;>�l =�;>���� o d h l�� 9 9 dobji;�].^%^�;������;�= ?(AUT�R G VzW � SCu5wxsAW�ROu�MOXZ
9 j 9 tverbi; ���z� dah l.� 9>9 subji
; �l>= ; ���� o dah l.� 9 9 dobji; ].^%^ ; �%X�Y ; = ?(A T�R G VzW � SVu5w Z\[�R�K>Stw n �\Z�Z
9 j 9 tverbi; ����<d h l�� 9 9 subji
;>�l =�;>�%X o d h l.� 9>9 dobji;�])^%^�;>�%X�Y);>= ?(A]T R G VzW � SVu5w �\Z�MO[�^Z
9 j 9 tverbi; �&X"dah l.� 9>9 subji
;>�l>=p;>�%X o dah l�� 9>9 dobji;�].^%^�;��&X!Yr;>=@?(AUT�R G VzW � SCu5w �\Z�M�[�^Z
9>9tverbi
d h l.� 9>9 subji;>�zl>=�;��%X�;>�%X"d h l.� 9 9 dobji
;�].^%^�;��%X�Yr;�= ?(AUT�R G VW � Svu5w � j W>_O~�R�[U` o Z9 9 j tverbi
d h l.� 9>9 subji; �zl>= ; �%X o d h l.� 9 9 dobji
; ].^%^ ; �%X�Y ; �%X ; =@?(A T�R G VW � Svu5w � j W>_O~�R�[U` o Z9 j tverbi
d h l.� 9>9 subji;>�zl>=�;��%X o d h l.� 9 9 dobji
;�].^%^�;��%X�Yr;��%XaTcbed =@?(A R Gfbgd�VZj tverbi
d h l�� 9>9 subji;>�zl>=�;��%X o d h l.� 9 9 dobji
;�].^%^�;��%X�Y5Tcbgd �%X bed = ?(A R Ghbed�VZ
158é A formal model of wordorder asstructural indication of informativity
(239)
.
.
.ß�Þ�ß tverbià ë 8�9¥Ú æ�Ý!M ß�ß dobji à ë 8�9)à�â Ý0Ý êrÚ ÜQÝ!M ß�ß subjià ë 8�9�à�8 åQä à ä���� ì ïß�Þ�ß tverbià ë 8�9 Ú 8 Ü�M ß�ß dobji à ë 8�9 à â Ý�Ý êrÚ 8 ÜPM ß�ß subjià ë 8�9 à 8 åQä à ä��?� ì ï í õ � ðiñ¥ò N� õrïO: ï ñ þ ó ù
ß�Þ�ß tverbià ë 8�9�Ú 8 Ü�M ß�ß dobji à ë 8�9)à�â Ý�Ý êrÚ 8 ÜPM ß�ß subjià�8 åQä à ë 8�9)à ä��?� ì ïýí õ � ðiñ¥ò ÿ Þ� õ � ø(õ �>� ê�ù
ß�Þ�ß tverbià ë 8�9YÚ 8 Ü�M ß�ß dobji à ë 8�9�à�â Ý�Ý ê�Ú 8 ÜPM ß�ß subjià�8 åQä à�8�;=<�à ä���� ì ïýí õ � ð¦ñ¥ò@!�ïQË ðò BiC @ ù
ß�Þ�ß tverbià ë 8�9YÚ 8 Ü�M¦ß�ß dobji à�â Ý0Ý à ë 8�9)ê�Ú 8 ÜPM¦ß�ß subjià�8 åQä à�8�;=<�à ä���� ì ïýí õ � ð¦ñ¥ò ÿ Þ� õ � ø ù ð,ð ê�ù
ß�Þ�ß tverbià ë 8�9 Ú 8 ÜPM ß�ß dobji à â Ý0Ý à Ý ; êrÚ 8 ÜPM ß�ß subjià 8 åQä à 8�;=< à ä���� ì ï í õ � ð¦ñ¥ò N ï�ñJI � � õ ó ù þ : ùß�Þ�ß tverbià Ý ;!Ú 8 Ü�M ß�ß dobji à�â Ý0Ý à Ý ;�êrÚ 8 ÜPM ß�ß subjià�8 å�ä à�8�;=<�à ä���� ì ï í õ � ð¦ñ¥ò K @ þ �.F ùß�ß tverbi Ú 8 ÜPM ß�ß dobji à(â Ý�Ý à Ý ;>à Ý ;!Ú 8 Ü�M ß�ß subjià�8 åQä à�8�;=<�à ä���� ì ï í õ � ðiñYò ÿ Þ ð )%ø(õ ïxô ê�ùß�ß�Þ tverbi Ú 8 ÜPM7ß�ß dobji à(â Ý�Ý à Ý ;>ê�Ú 8 Ü�M¦ß�ß subji àD8 åQä à�8�;=<�à Ý ;�à ä���� ì ïýí õ � ðiñYò ÿ Þ
ð )%ø(õ ïxô ê�ùß�Þ tverbi Ú 8 ÜPM ß�ß dobji à â Ý0Ý à Ý ; êrÚ 8 ÜPM ß�ß subjià 8 åQä à 8�;=< à Ý ; ì � � ä>�?� ï í � �0 ùÞ tverbi Ú 8 Ü�M ß�ß dobji à(â Ý0Ý à Ý ;%êrÚ 8 ÜPM ß�ß subjià�8 åQä à�8�;=<¥ì � � Ý ; � � ä���� ï í
� �0 ù
(240)
.
.
.ß�ß�ß subji à ë 8�9�à�8 åQä Ú ÛzÜQÝ Þ�ß tverbià ë 8�9YÚ æ�Ý�M ß�ß dobji à ë 8�9�à�â Ý0Ý ê�à ä��?� ì ïýí ú ùË þ � � �^ï õ � õkj þ �ml�� ó * � � þ ï+Ë ù
ß�ß�ß subji àD8 åQä à ë 8�9YÚ ÛzÜQÝ Þ�ß tverbià ë 8�9YÚ æ�Ý�M ß�ß dobji à ë 8�9�à�â Ý0Ý ê�à ä��?� ì ï í õ � ð¦ñ¥ò ÿ Þ� õ � ø0õ �>� ê�ù
ß�ß�ß subji à 8 åQä à ë 8�9 Ú¦ÛzÜQÝ�Þ�ß tverbià ë 8�9 Ú æ�Ý�M ß�ß dobji à â Ý0Ý à ë 8�9 ê�à ä��?� ì ï í õ � ð¦ñ¥ò ÿ Þ� õ � ø ù ð,ð ê�ù
ß�Þ�ß�ß subji à�8 å�ä à ë 8�9YÚ ÛzÜ�å ß tverbià ë 8�9�ê�Ú ÜQåQM ß�ß dobji à(â Ý0Ý à ë 8�9)à ä��?� ì ï í õ � ðiñ¥ò ïO:ï ñ þ ó ù
ß�Þ�ß�ß subjià�8 å�ä à ë 8�9¥Ú ÛzÜ�å ß tverbià ë 8�9.êrÚ ÜQåQM ß�ß dobji à(â Ý�Ý à�8�;=<�à ä���� ì ïýí õ � ð¦ñ¥ò@!�ïQË ðò BiC @ ù
ß�Þ�ß�ß subjià�8 åQä à Ý ;!Ú¦ÛzÜ�å.ß tverbià Ý ;%ê�Ú�ÜQåQM�ß�ß dobji à�â Ý0Ý à�8�;=<�à ä���� ì ï í õ � ð¦ñ¥ò @ þ - ð - þ )%ò I � � õ ó ù þ : ùß�ß�ß�ß subjià 8 åQä à Ý ; Ú¦ÛzÜ�å tverbià Ý ; Ú{Ü�åQM�ß�ß dobji à â Ý0Ý à 8�;=< à ä���� ì ï í õ � ðiñYò ÿ Þ ð )%ø î�ïQ� ê�ùß�ß�Þ�ß�ß subjià�8 åQä à Ý ;!Ú ÛzÜ�å tverbiêrÚ ÜQåPM ß�ß dobji à(â Ý�Ý à�8�;H< à Ý ;�à ä���� ì ï í õ � ðiñYò ÿ Þ
ð )%ø ïQ�qô ê�ùß�Þ�ß�ß subjià�8 åQä à Ý ;!Ú ÛzÜQå tverbiêrÚ ÜQåQM ß�ß dobji à(â Ý0Ý à�8�;=<�à Ý ;iì � � ä>�?� ï í ��� ùÞ�ß�ß subjiàD8 åQä à Ý ;�Ú¦ÛzÜQå tverbiê�Ú7Ü�åQMiß�ß dobji à�â Ý0Ý à�8�;=<"ì � � Ý ; � � ä���� ï í
� �0 ù2
Definition 32(Structural indicationsof informativity in SVO). TheInfSVO de-
finesthebasicstructural rulesfor describing structural indication of informativity
in SVO languages,given INFHYP1 and INFHYP2. TheInfSVO monotonically ex-
tendsall theSVO packages,andonlydefinesrulesthatspecify feature information.
Thebasic setof structural rules of InfSVO is given below, without any reference
to tune.Thestrategy followedis thesameasin thepreviouspackages,exceptthat
wepaymore attention here to therelation between non-systemic ordering andthe
boundarybetween (therealizations of) topic andfocus.
ê�ëíì�îIïQ�0ò ÷ �1Ë � û ê�ëíì�îIï,ðò ÷ � Ë � ü Ó Ç / ýrþ�ÿ�� ý � �'ÿ � $O�ê�ñ�ëâì�î�ïQ�'ò ö ìKïQ�'ô õ½÷ �1Ë � û ê�ëíì�îIïQ�%ñ�ò�ìKó ðô�õ½ö�÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ý � �'ÿ � $O�ñ�ëíìKïQ�'ô¬ò ö ìKïQ�'ô õ û ñ�ëíì��eð)ô�ò ö ì¸ó ðô�õ ü Ó Ç / ýrþ�ÿ�� ý � �'ÿ � $O�
ê�ë�ìÉî õ ï!ò ÷ �1Ë � û ê�ëíì î ó ð ò ÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�ê�ë�ì î õ ï ò ÷ �1Ë � û ê�ëíì î!�]ð ò ÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�
ê�ñ�ëíì î õ ï ò ö ì õ ïô õ½÷ �1Ë � û ê�ëíì î õ ï ñ�ò�ì ï,ð)ô õ½ö�÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�ê�ñ�ëíì�î õ ï�ò ö ì�î õ ï õ½÷ �1Ë � û ê�ëíì�îIï,ðfñ�ò�ì î ó ð�õ½ö�÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�ê�ñ�ëíì�î õ ï�ò ö ì�î õ ï õ½÷ �1Ë � û ê�ëíì î ó ð ñ�ò ì�î�ï,ð õ½ö�÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�ê�ñ�ëíì õ ïô�ò ö ì õ ïô õ½÷ �1Ë � û ê�ñ�ëíì¸ó ð)ô ò ö ìnï$ðô õ½÷ �1Ë � ü Ó Ç / ýrþ�ÿ�� ( Ö Ç¬ý � �0ÿ � $O�
A formalmodelof wordorderasstructuralindicationof informativity /159
ê�ë�ìAïQ�'ôqê�ò ÷ õ�)!* ÷ �1Ë � û ê�ë�ìAïQ�'ô¸ê�ò ÷ � õ ��÷ �1Ë � ü Ó Ç / ýrþ¡ÿ�� +aÖ ��Î7þY� õ-� + �ê�ëâì õ ïôqê�ò ÷ õ�)!* ÷ �1Ë � û ê�ë�ì õ ïô¸ê�ò ÷ � õ ��÷ � Ë � ü Ó Ç / ýrþ¡ÿ�� +aÖ ��Î7þY� õ-� + �ê�ëâì�î õ ïfê�ò ÷ õ�)!* ÷ �1Ë � û ê�ë�ìÉî õ ï�ê�ò ÷ � õ � ÷ � Ë � ü Ó Ç / ýrþ¡ÿ�� +aÖ ��Î7þY� õ-� + �
ñ�ê�ë ÷ � õ � ì�î�ïQ�%ê�ò ÷ õ�) ö ìAïQ�'ôqê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉîIïQ��ê�ò ÷ � õ ��ö ìAïQ�'ô¸ê õ½÷ õ�)!* ü Ó Ç / ýrþ¡ÿ�� � +a��Ö� �ñ�ê�ë ÷ õ�) ì î�ïQ� ê�ò ÷ õ�) ö ì ïQ�'ô ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ �Ôì îIïQ� ê�ò ÷ � õ � ö ì ïQ�'ô ê õ½÷ õ�)!* ü Ó Ç / ýrþ¡ÿ�� � +a��Ö� �ñ�ê�ë ÷ � õ �Éì î�ïQ� ê�ò ÷ ð ) ö ì ïQ�'ô ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ �Ôì îIïQ� ê�ò ÷ � õ � ö ì ïQ�'ô ê õ½÷ õ�)!* ü Ó Ç / ýrþ¡ÿ�� ýrÿ ò[Ö Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ì õ ïô ê�ò ÷ ð ) ö ì õ ïô ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ì õ ïô ê�ò ÷ � õ � ö ì õ ï)ô ê õ½÷ õ�).* ü Ó Ç / ýrþ¡ÿ�� (Lýxÿ ògÖ Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ìÉîIïQ��ê�ò ÷ ð ) ö ì õ ïô¸ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉîIïQ��ê�ò ÷ � õ � ö ì õ ïô¸ê õ½÷ õ�).* ü Ó Ç / ýrþ¡ÿ�� (Lýxÿ ògÖ Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ì�îIïQ��ê�ò ÷ õ�) ö ì õ ïô¸ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉîIïQ��ê�ò ÷ � õ ��ö ì õ ïô¸ê õ½÷ õ�).* ü Ó Ç / ýrþ¡ÿ�� (Lýxÿ ògÖ Õ�Ç�$&� ��� �ñ�ê�ë ÷ ð ) ìÉî õ ï�ê�ò ÷ ð ) ö ì õ ïô¸ê õ½÷ õ�).* û ñ�ê�ë ÷ � õ � ìÉî õ ï�ê�ò ÷ � õ ��ö ì õ ï)ôAê õ½÷ õ�).* ü Ó Ç / ýrþ¡ÿ�� (Lýxÿ ògÖ Õ�Ç�$&� ��� �
ñ�ë�ì î õ ï ê�ò ÷ ð ) ö ì î õ ï ê õ½÷ õ�).* û ñ�ë�ì î õ ï ê�ò ÷ � õ � ö ì î õ ï ê õ½÷ õ�).* ü Ó Ç / ýrþ¡ÿ�� (Lýxÿ ògÖ Õ�Ç�$&� ��� �ê�ë ÷ ð ) ì îIï�� ê�ò ÷ ð ) û ê�ë ÷ � õ � ì îIïQ� ê�ò ÷ ð ) ü Ó Ç / ýrþ¡ÿ�� 2 +a�)Ö� �ê�ë ÷ ð ) ì î õ ï ê�ò ÷ ð ) û ê�ë ÷ � õ �Ôì î õ ï ê�ò ÷ ð ) ü Ó Ç / ýrþ¡ÿ�� 2 +a�)Ö� �ê�ê�ë ÷ ð ) ì�îIï��'ò ÷ ð ) û ê�ë ÷ ð ) ì�îIïQ��ê�ò ÷ ð ) ü Ó Ç / ýrþ¡ÿ�� � ñ � � �3�� Ö ö��
ê�ëíìKïQ�'ô¬ò ÷ ð ) û ê�ë ÷ ð ) ìKïQ�'ô¬ò ü Ó Ç / ýrþ¡ÿ�� � ñ � � � � Ö54 ö��ê�ê�ë ÷ ð ) ìÉî�ïQ�0ò ÷ õ�) û ê�ë ÷ ð ) ì�îIïQ��ê�ò ÷ õ�) ü Ó Ç / ýrþ¡ÿ�� � ñ Ç � �3 � Ö ö��ê�ê�ë ÷ õ�) ìÉî�ïQ�0ò ÷ õ�) û ê�ë ÷ õ�) ìÉîIïQ��ê�ò ÷ õ�) ü Ó Ç / ýrþ¡ÿ�� � ñ Ç � �3 � Ö ö��
ê�ë�ì ïQ�'ô ò ÷ õ�) û ê�ë ÷ õ�) ì ïQ�'ô ò ü Ó Ç / ýrþ¡ÿ�� � ñ Ç � � � Ö64 ö��ê�ë�ì õ ïô ò ÷ õ�) û ê�ë ÷ õ�) ì õ ïô ò ü Ó Ç / ýrþ¡ÿ�� � ñ Ç � ��Ç7� 4 ö��ê�ëíì õ ïô�ò ÷ ð ) û ê�ë ÷ ð ) ì õ ïô�ò ü Ó Ç / ýrþ¡ÿ�� � ñ � � ��Ǧ� 4 ö��ê�ê�ë ÷ ð ) ì�î õ ïò ÷ ð ) û ê�ë ÷ ð ) ì�î õ ï/ê�ò ÷ ð ) ü Ó Ç / ýrþ¡ÿ�� � ñ � � �3 Ç7�0ö��
ê�ñ�ê�ë ÷ ð ) ìÉî õ ï�ê�ò ÷ ð ) ö ìÉî õ ï õ½÷ õ�).* û ñ�ê�ë ÷ ð ) ìÉî õ ï�ê�ò ÷ ð ) ö ìÉî õ ï�ê õ½÷ õ�)!* ü Ó Ç / ýrþ¡ÿ�� � ñ Ç ��7 ��3 Ç7�0ö��S
Example. To round off this chapter, we present a few examples of the realiza-
tion of information structure in SVO word order languages. In the next chapter
we seemoreexamplesof informationstructure in SVO languages,using a mix-
ture of tuneandword order. (241) provesthe realization of an “out-of-the-blue”
sentence,whereas(242) and(243) illustratehow systemicorderingplays a role in
establishing a boundarybetweentopic andfocus.
(241)
.
.
.ß�ß subjià ë 8�9 Ú ÛzÜQÝ Þ�ß verbià ë 8�9 Ú æ�Ý!M ß dobji à ë 8�9 ê�à ä��?� ì ïß�ß subjià ë 8�9¥Ú¦ÛzÜQå�Þ�ß verbià ë 8�9¥Ú æ�Ý!M ß dobji à ë 8�9�ê�à ä���� ì ïýí õ � ï�ð7ñ¥ò ïO:ï ñ þ ó ù
ß�Þ�ß subjià ë 8�9YÚ ÛzÜ�å ß verbià ë 8�9�ê�Ú Ü�åQM ß dobji à ë 8�9)à ä��?� ì ïýí õ � ï�ð7ñ¥ò ïO:ï ñ þ ó ù
ß�Þ�ß subjià ë 8�9 Ú¦ÛzÜ�å.ß verbià ë 8�9 ê�Ú�ÜQåQMiß dobji à 8�;=< à ä���� ì ï í õ � ï�ð7ñ¥ò@!�ïQË ðò BiC @ ù
ß�Þ�ß subji à 8�; Ú¦ÛzÜ�å.ß verbià 8�; ê�Ú{Ü�åQMiß dobji à 8�;=< à ä���� ì ï í õ � ï�ð7ñ¥ò C @ þ �HF ùß�ß�ß subji à�8�;!Ú ÛzÜ�å verbià�8�;�Ú Ü�åQM ß dobji àD8�;=<�à ä���� ì ï í õ � ï�ð¦ñYò ÿ Þ õ�)%ø î�ïQ� ê�ùß�ß�Þ�ß subji à�8�;!Ú ÛzÜ�å verbiê�Ú ÜQåQM ß dobji àD8�;H<�à�8�;%à ä���� ì ïýí õ � ï�ð¦ñYò ÿ Þ õ�)%ø
ïQ�¸ô ê�ùß�Þ�ß subjià 8�; Ú¦ÛzÜQå verbiê�Ú{ÜQåQM�ß dobjià 8�;=< à 8�; ì � � ä>�?� ï í � �0 ùÞ�ß subjià 8�; Ú¦ÛzÜQå verbiêrÚ{Ü�åQMiß dobji à 8�;=< ì ��� 8�; ��� ä���� ï í
� �0 ù
160é A formal model of wordorder asstructural indication of informativity
(242)
.
.
.ß�ß dobji à ë 8�9�Ú Ûzæ>Ý Þ�ß verbià ë 8�9YÚ ÜQÝ!M ß subjià ë 8�9�ê�à ä���� ì ïß�ß dobji à ë 8�9 Ú Û 8 Ü Þ�ß verbià ë 8�9 Ú ÜQÝ!M ß subjià ë 8�9 ê�à ä>�?� ì ï í õ � ï�ð¦ñ¥ò N� õOïO: ï ñ þ ó ù
ß�Þ�ß dobji à ë 8�9�Ú Û 8 Ü ß verbià ë 8�9.êrÚ 8 ÜPM ß subjià ë 8�9)à ä>�?� ì ïîí õ � ï�ð¦ñ¥ò N� õOïO: ï ñ þ ó ù
ß�Þ�ß dobji à ë 8�9YÚ Û 8 Ü ß verbià ë 8�9�êrÚ 8 ÜPM ß subjià�8�;=<�à ä>�?� ì ï�í õ � ï�ðiñ¥ò@!�)ï+Ë ðò BiC @ ù
ß�Þ�ß dobji à Ý ;!ÚiÛ 8 Ü�ß verbià Ý ;�êrÚ 8 ÜPM¦ß subjià�8�;=<�à ä���� ì ï í õ � ï�ðiñ¥ò N"ïñ�I � � õ ó ù þ : ùß�ß�ß dobji à Ý ; Ú Û 8 Ü verbià Ý ; Ú 8 Ü�M ß subjià 8�;=< à ä���� ì ï í õ � ï�ð7ñ¥ò ÿ Þ ð )%ø î õ ï ê�ùß�ß�Þ�ß dobji à Ý ;!Ú Û 8 Ü verbiê�Ú 8 Ü�M ß subjiàD8�;=<�à Ý ;�à ä���� ì ïýí õ � ï�ð7ñ¥ò ÿ Þ
ð )%ø(õ ïxô ê ùß�Þ�ß dobji à Ý ;!Ú Û 8 Ü verbiêrÚ 8 ÜPM ß subjià�8�;H<�à Ý ;¦ì � � ä���� ï í � �� ùÞ�ß dobji à Ý ;�Ú¦Û 8 Ü verbiê�Ú 8 Ü�M¦ß subjià�8�;=<¥ì � � Ý ; � � ä��?� ï í
� �0 ù
(243)
.
.
.ß�Þ�ß verbià ë 8�9YÚ æ�Ý�M ß dobji à ë 8�9)êrÚ ÜQÝ!M ß subjià ë 8�9)à ä��?� ì ïß�Þ�ß verbià ë 8�9 Ú 8 Ü�Miß dobji à ë 8�9 êrÚ 8 ÜPMiß subjià ë 8�9 à ä>�?� ì ïîí õ � ï�ð¦ñ¥ò N� õOïO: ï ñ þ ó ù
ß�Þ�ß verbià ë 8�9YÚ 8 Ü�M ß dobji à ë 8�9�êrÚ 8 ÜPM ß subjià�8�;=<�à ä>�?� ì ï í õ � ï�ðiñ¥ò@!�)ï+Ë ðò BiC @ ù
ß�Þ�ß verbià Ý ;!Ú 8 Ü�M ß dobji à Ý ;�êrÚ 8 ÜPM ß subjià�8�;=<�à ä���� ì ï í õ � ï�ðiñ¥ò N"ïñ�I � � õ ó ù þ : ùß�ß verbi Ú 8 ÜPM7ß dobji à Ý ;>à Ý ;!Ú 8 Ü�M¦ß subjià�8�;=<�à ä���� ì ï í õ � ï�ð7ñ¥ò ÿ Þ ð )%ø(õ ïxô ê ùß�ß�Þ verbi Ú 8 ÜPM7ß dobji à Ý ; ê�Ú 8 Ü�Miß subjià 8�;=< à Ý ; à ä���� ì ïýí õ � ï�ð7ñ¥ò ÿ Þ
ð )%ø(õ ïxô ê ùß�Þ verbi Ú 8 Ü�M ß dobji à Ý ;�êrÚ 8 ÜPM ß subjià�8�;H<�à Ý ;¦ì � � ä���� ï í � �� ùÞ verbi Ú 8 Ü�M ß dobji à Ý ;�êrÚ 8 Ü�M ß subjià�8�;=<¥ì � � Ý ; � � ä��?� ï í
��� ù2
SUMMARY
In thischapterI discussedvariousapproachesto modelingwordorderin acategorial gram-
mar- CCG,MCCG,Set-CCG,andcategorial typelogic. I elaboratedonwhywechoosefor
viewing adjacency asa parameterratherthana principle. The“principle”-view yields(in
contemporaryformulationsof CCG)varioustechnical consequencesthatareat oddswith
our linguistic intuitions, like thedissociationbetweentheexplanationsof word orderand
informationstructure in MCCG.The“parameter”-view enablesusto considerinformation
structureasaprimary factor(parameter)determining wordorder. I presentedgrammarar-
chitecturesthatformalizedthebasicaspectsof rigid, mixedandfreeword order predicted
by thevariationhypothesesof Chapter3, including complex phenomenalike (discontin-
uous)cross-serialdependenciesandnesteddependencies.Subsequently, I extendedthese
architectureswith models of how word order principally actsasa structuralindication of
informativity in OV, VO, andSVO, asfar ascoveredby the informativity hypothesesof
Chapter3. Theadoptedstrategy wasto relatethesyntacticmodesto modesthatjust show
whetheror not systemicordering is adhered to, after (Sgall et al., 1986). This provides
usnot only with a very general description of therelationbetweenword order andinfor-
mationstructure.It will alsoenableusto describetheinteraction betweentuneandword
orderin a straightforwardway, aswe shall seein thenext chapter. There, we extendthe
SVO model to cover tuneasastructuralindication of informativity.
CHAPTER 5
A FORMAL MODEL OF TUNE AS STRUCTURAL
INDICATION OF INFORMATIVITY
Besidesword orderlanguagesusuallyalsousetuneto realize information structure- sometimes
even predominantly so, like in the caseof English. In this chapter, we begin by discussing
Steedman’s modelof English tunedeveloped in Combinatory Categorial Grammar. We then
continueby presenting a moreabstractmodelof tunethatcanbe instantiated to cover different
languages,and that overcomesa few problems we may note for Steedman’s proposal. The
chapterendswith adiscussionof how to includetunein themodelof informativity wedeveloped
in thepreviouschapter.
5.1 INTRODUCTION
The goal of this chapter is to develop an abstract model of tune in its role as a
structural indication of informativity, after (Steedman, 2000a), andshow how the
model can be integratedwith the word order account presentedin the previous
chapter. Theintegratedarchitecture enablesus to describe formally how tune and
word order interact to realize information structure.
Most languagesthat do not have a mixed or free word order predominantly
usetune to realize structural indication of informativity, cf. (133) on page96. An
often-cited example of sucha languageis English. Only in (very) marked cases
doesEnglishuseboth word order andtune- otherwise it just places the nuclear
stressin a position other thanthe unmarked oneto realize a different focus. The
examplesin (244)illustratethis useof tune,on narrow focus.
(244) English
a. Elijah gave a book n to KATHY ofp .
b. Elijah gave n a BOOK oDp to kath.
c. n ELI JAH o p gave a book to Kathy.
161
162q A formal model of tuneasstructural indication of informativity
Theuseof tuneto realize informationstructurewasstudiedalready by Math-
esius in theearly nineteenthirties- seefor example(1975), whereMathesiuscon-
trastsEnglish andGerman.Sgallet al. (1986) discussEnglish tuneandhow it can
beunderstood to indicatetheunderlying linguistic meaning’s topic-focusarticula-
tion, a discussion continuedin Hajicova et al’s (1998).
An important,recent contribution to thestudy of tuneandits relation to infor-
mationstructureis Pierrehumbert& Hirschberg (1990). Pierrehumbert& Hirschberg
argue that the interpretation of tune is built compositionally from pitch accents,
phraseaccents, andboundary tones.
Various authors have advanced proposals that formalize this compositional
interpretation in more detail, including Steedman (2000c; 2000a) and Hendriks
(1996; 1997; 1999). Bothwork outmodelsthat arephrasedin categorial grammar,
with Steedmanworking in CCGandHendriks in theLambektradition.1 Herewe
focus on Steedman’s proposal,discussingit in moredetail in r 5.2. The reasons
for opting for Steedman’s CCGaccount ratherthanHendriks’ proposalis that the
former is worked out in more linguistic detail, and that it -surprisingly perhaps-
appearsmoresuitable for recasting in termsof categorial type logic. In r 5.3 we
present the abstractmodelof tunein DGL, andin r 5.4 we show how to integrate
themodelwith theSVO architecture developedin thepreviouschapter.
5.2 STEEDMAN’ S SYNTAX-PHONOLOGY INTERFACE
Steedmanproposesin (2000a) an integration, into CCG,of informationstructure
andits realization throughtune. With that,Steedmanpresentsnot only a compre-
hensive modelof Englishtuneandits relation to linguistic meaning. He alsopro-
vides the ground for the important argumentthat “a theory of grammarin which
phrasal intonation and informationstructure arereunited with formal syntaxand
semantics is not only possible, but muchsimpler thanonein which they aresepa-
rated.”
In other words, Steedmanshows that a categorial model is both possible and
preferable over the generally adhered to GB architectureasproposedby Selkirk.
This architecture is given in Figure 5.1 (adapted from Steedman(2000a)), and
shows how Selkirk proposesan autonomousstructural level called “Intonation
Structure” thatmediatesbetweenPhonologicalForm(PF)andLogical Form(LF).
1Recall from earlierdiscussionsthat, in fact,Moortgat & Morrill alreadyproposeda modelofprosody in (1991). Also Oehrlediscusses prosody, for example in (1991)and(forthcoming).How-ever, noneof thesediscussionsconcern Pierrehumbert& Hirschberg’s proposal.
A formalmodelof tuneasstructuralindication of informativity /163
Important to observe is thatwe alsohave theSurfaceStructure mediating between
PFandLF, asidefrom the Intonation Structure. Theresponsibility of the Surface
Structure still concernsaspectsrelating to theLF’s Predicate-Argumentstructure.
The additional task for Intonation Structure is to definethose aspects of LF that
relateto information structure.2
PF
Intonational
Lexicon
D-Structure
S-Structure
structure
a) Predicate-argument str.b) Information StructureLF:
Figure5.1: Architectureof a GB theory of Prosody
Thearchitecture thatSteedmanadvancesis depicted in Figure5.2. Insteadof
having separate levels for S-Structureand Intonational Structure we now have a
singlemoduledescribing surfacesyntax - CCG.As usual we have that operations
on categories (describing surface structure) are associatedwith a compositional
formation of a sentence’s semantics,cf. (Steedman, 1996). In (Steedman,2000a),
thesesemanticscaptureboththesentence’s informationstructureandits predicate-
argumentstructure.3
Wealready discussedSteedman’stheory of informationstructurein r 2.3(p.52ff.).
Here, we briefly review how Steedman’s Themeand Rhemerelate to different
tunes, afterwhich we turn to Steedman’s formalization in CCG.
In keepingwith Pierrehumbert& Hirschberg, Steedmanconsiderstheinterpre-
tation of tuneto be built from pitch accents,phrase accents, andboundary tones.2Recallthatthisseparationof predicate-argumentstructureandinformationstructuregoesagainst
thePraguianviewpoint,aswe alreadynotedin Chapter2.3Thearchitecturein Figure5.2 shows thata predicate-argumentstructureis in factconceivedof
asthe end-product in (Steedman,2000a). The reasonwhy Steedmanpresentsthe architecturethisway primarily hasto do with his efforts to rebuke theGB architecture.As Steedmannoteshimselfat theendof s 3.2, in practiceonewould not want to have a normalizedterm,but a morestructuredmeaning- for example, like the representationsas proposedby Maternaet al. (1987), or as perPeregrin (1995) andelaboratedin Kruijf f-Korbayova (1998).
164q A formal model of tuneasstructural indication of informativity
Phonetic form
CCG Lexicon
Information Structure
Predicate-ArgumentStructure
phonology
normalization
Figure5.2: Architectureof a CCGtheoryof Prosody
We startwith pitch accents. Steedmanfoll ows BeckmanandPierrehumbertin as-
sociatingthefollowing pitch accents with Themeor Rheme(in English).
(245) a. Realizing Theme:L+H*, L*+H
b. Realizing Rheme:H*, L*, H*+L, H+L*
For example,in (246), “admires” hasapitchaccent L+H* (andaboundarytone
LH%), whereasthefocus of theRhemehasa pitch accent H*.
(246) English
I know thatMarcel likes themanwho wrotesthemuscial.
But who does heADMIRE?
Marcel
t uOv wxzy�{Q|�}�~0�������
ADMIRES
L+H*LH%t u�v wp �0{��&�t u�v w�����L�6�
thewomanwho
t u�v wxzy�{P|�}�~�L�%���
DIRECTED
H*t u�v wp �0{����
themusical
LL%t uOv wxzy�{Q|�}�~0�������t uOv w�����L�6�
Pitchaccentsconcernwordforms, indicatingtheinformativity of asingle (sim-
ple) dependent or head.To enable the projection to larger structures,i.e. “phrase
accents”, Steedmanassumesthatpitch accentsarenotonly related to thelinguistic
meaning, but that they arealsoexpressedin theword’s category. Sucha category
canthenbecomposedwith othercategorieson thecondition thatthis composition
obeys whatSteedmancalls “compatibility with theme-or rheme-hood”.
A formalmodelof tuneasstructuralindication of informativity /165
Below we point out how this works in CCG.Beforethat we should still have
a look at boundarytones, andwherewe get our tunesfrom in the first place. To
begin with thelatter, Steedmanassumeswhathecallsa“pre-syntactic assignment”.
In the context of his (2000a), this naturally resolvesto an assignmentof tunesto
wordsin thelexicon - but this is not a theoretical necessity.4
In building a compositional semantics for the boundary tones LH% andLL%
andtheabove pitch accents,Steedmanfirst of all considerstheconceivable differ-
encebetweenLH% andLL%. Steedmanconjecturesthat an H% boundary tone
indicatesthat the Themeor Rhemeit is associatedto is the hearer’s responsibil-
ity, whereasan L% boundary toneindicatesthat the information is the speaker’s
responsibility.5 For example,by marking information with a boundarytoneH%,
we could realizevarious speech actslike questioning, polite requesting, ceding or
holding theturn,etc.
Steedmansubsequently makes a rather inelegant move, and models bound-
ary tonesasemptystrings,reminiscentof transformational grammar’s emptycat-
egories. A boundarytonehasa functional category, with no realization, thatcom-
bineswith pitch accents. Important hereis thatthecomposition of aboundarytone
category with a pitch accent category allows for the Theme/Rhemedistinction to
beprojectedfrom pitch accentsonto prosodic phrases.
Thus, to recapitulate, we have an inventory of pitch accents that realize either
Themeor Rheme(using Steedman’sterms),andwhichcanproject over larger con-
structionsby composition with eitheraboundarytoneor wordswith unmarkedin-
tonation. Projection is handledby unification: Categoriescarryfeatures� (Theme),� (Rheme)or � (unmarked), which get unified in the usual way asexplained in
(Steedman,1996). The ideaof unification that Steedmanusesis of course very
muchakin to the distribution andpercolation structural ruleswe employ in DGL
(andcategorial type logic in general) to handle feature information. In the next
section, we sketch how we cannot only remodelSteedman’s proposalin termsof
DGL, but actually -by using boxesanddiamonds- getrid of theemptystringsthat
Steedmanusesto modelboundary tones.
4For example,tunescould be obtainedfrom a speech-recognition module, and thencombinedwith the ‘bare’ category informationfrom the lexicon to form the category assignment usedin thederivation. This doespresenta differenceto for exampleHendriks’ (1999), in which a possibleintonationis derivedthrougha rewriting system.
5Thereby, responsibilityis understoodin thesenseof ‘ownership’.
166q A formal model of tuneasstructural indication of informativity
5.3 TUNE IN DGL
Herewe provide a modelof tune in DGL that we envision asan abstract model
of tune. That is, we assume the presence of tunes that either realize contextu-
ally bound information (Steedman’s � -tunes) or contextually nonbound informa-
tion (the � -tunes),andboundarytones,but we leave openhow a specific language
instantiates thesewith its own inventory of tunes.6 Particularly, we assumethe
following unary modal operatorsfor the basicintonation: ��� for tunes realizing
contextually nonboundelements,and ��� for unmarkedintonation. We caninclude
these lexically, or let structural rulesdetermineproper instantiationsof a generally
assigned �P�z� feature. Here,we keepit simpleandassumea lexical assignment. At
thesametime,boundarytones ��� arenot modeled lexically. Instead, we modelthe
presenceof a boundarytoneusing a structural rule that rewritesa �>� or ��� feature
into for exampleeither ��������� or ��������� . This immediately leadsto amoreelegant
proposal,becausewe modelboundary tones directly on elementsin the prosodic
structure.We no longer have to assumeemptystrings thatmodelboundary tones.
A basictunesystemthat implements theseideasfor SVO is thendefinedasfol-
lows. Alik e the modelswe constructed for word order asa structural indication
of informativity, we assumemarking with systemic ordering. Because we only
dealwith the interaction between(non-systemic) word orderandtunein the next
section, thedefinition belowconcernssystemicword order only.
Definition 33 (BasicSVO tune). We definehere an abstract, basicmodelof tune
that is basedon Steedman’s CCG-basedproposalin (2000a).
�f�5���.�Q�J�D�h�� ¢¡Q£� �¤�¥=¦0 �§�£�¨ª©c�f�«���.�P�J�D�h�� �£h¬¤� �f¤�¡� �§>£?¨ ® ¯�°�±�²L³ ´iµ�¶i·�°�¸!¹mºO»�¼P²!½.½�D¾?�D�f�� £h¬¤ f¤�¡ ��¿&�.��D�h�� ¡P£ ¤�¥=¦0À ���H�Q�J�D�h´i ¥¢£?Á&¬£ D ¥ §�£�¨ ©c�D¾��D�f�i £h¬¤ f¤�¡ ��¿&�.��D�h�� £?¬¤ h¤�¡�À ���.�Q�J�D�h´i £?¬¤ h¤�¡ §>£?¨ ® ¯�°�±�²L³ ·�´�µi¶�·>°�¸.¹mºO»�¼P²�D¾?�D�f�� ¡P£ ¤�¥H¦ � ¿&�H� �D�h�� ¥D£?Á&¬£ D ¥PÀ � �.�Q� �D�h´� ¬£ D ¥ §�£�¨ ©c�D¾��D�f�i £h¬¤ f¤�¡ � ¿&�.� �D�h�� £?¬¤ h¤�¡�À � �.�Q� �D�h´i £?¬¤ h¤�¡ §>£?¨ ® ¯�°�±�²L³ ·�´�µi¶�·>°�¸.¹mºO»�¼P²¾m�«� ¿&�.� �D�h�� ¬£�Á&¥¢£ D ¥�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ©c¾m�«� ¿&�H� �D�h�� £h¬¤ f¤�¡�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ® ¯�°�±�²L³ º�ÃJ��ÄL°�±�ÅQƼPÇ�Ⱦm�«� ¿&�.� �D�h�� ¬£�Á&¥¢£ D ¥�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ©c¾m�«� ¿&�H� �D�h�� £h¬¤ f¤�¡�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ® ¯�°�±�²L³ º�ÃJ��ÄL°�±�ÅQƼPÇ�È�D�f�� �¬£?Á�¥D£�  ¥���¿&�.���D�h�� �¬£P �¤�¥%©c�D�f�i �£?¬¤& �f¤�¡���¿&�.���D�h�� �¬£� �¤�¥ ® ¯�°�±�²L³ º�ÃJ��ÄL°�±�ÅQƼPÇ�È�D�f�i ¬£?Á�¥D£  ¥ �J¿&�H���D�h�� ¡P£ ¤�¥H¦ ©c�D�f�i £?¬¤ f¤�¡ ��¿&�.���D�h�� ¡Q£ ¤�¥=¦ ® ¯�°�±�²L³ º�ÃJ��ÄL°�±�ÅQƼPÇ�Ⱦm�«���.�P�J�D�h�� ¬£ ¤�¥�À ���.�P�J�D�h´� ¡P£ ¤�¥H¦ ©c¾m�«���H�Q�J�D�h�� £h¬¤ f¤�¡�À ���.�P�J�D�h´� ¡P£ ¤�¥H¦ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQȾ��D�f�i ¬£ ¤�¥ � ¿&�H� �D�h�� ¬£ ¤�¥�À � �H�Q� �D�h´i ¬£ ¤�¥ ©c¾��D�f�i £h¬¤ f¤�¡ � ¿&�.� �D�h�� £?¬¤ h¤�¡�À � �.�Q� �D�h´� ¬£ ¤�¥ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQȾm�«� ¿&�.� �D�h�� ¬£ ¤�¥�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ©c¾m�«� ¿&�H� �D�h�� £h¬¤ f¤�¡�À � �.�P� �D�h´� ¡P£ ¤�¥H¦ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQÈ�D�f�i ¬£ ¤�¥ � ¿&�.� �D�h�� ¬£ ¤�¥ ©c�D�f�i £?¬¤ f¤�¡ � ¿&�.� �D�h�� ¬£ ¤�¥ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQÈ�D�f�i �¬£� �¤�¥��J¿&�H���D�h�� ¢¡P£� �¤�¥H¦�©c�D�f�i �£?¬¤& �f¤�¡���¿&�.���D�h�� ¢¡Q£� �¤�¥=¦ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQÈ
6More specifically, we assumethat we canmonotonically extendthe package we describeherewith a language-specificpackage that (a) instantiatesthe tunes,and(b) regulatestheir possibleco-occurrencesandmutualorderings- in away similar to for exampleVFinal’sXDep andNDep pack-ages.
A formalmodelof tuneasstructuralindication of informativity /167
�D�f�i ¬£ D ¥ � ¿&�H� �D�h�� ¬£?Á�¥D£ D ¥ ©Ê�D�f�i £h¬¤ f¤�¡ � ¿&�.� �D�h�� ¬£�Á&¥D£ D ¥ ® ¯�°�±�²L³ ¯�¶i¼�ÄHÉQÈ�D�f�i �¬£!  ¥���¿ ¤ �0�D�h�� �¬£?Á�¥D£�  ¥�©Ê�D�f�i �£h¬¤� �f¤�¡���¿ ¤ �0�D�h�� �¬£?Á�¥D£�  ¥ ® ¯�°�±�²L³ ¯�¶i¼�ÄHÉQÈ�D�D�f�i ¤�¥ ��¿&�.���� ¬£ ¤�¥ ©Ê�f�i ¤�¥ ��¿&�.���D�h�� ¬£ ¤�¥ ® ¯�°�±�²L³ Ë�¾m±�Ì.ÍP°�»HÎ!Ï6½.Ä À È�D�D�f�� D ¥ � ¿&�.� �� ¬£ ¤�¥ ©Ê�f�i D ¥ � ¿&�.� �D�h�� ¬£ ¤�¥ ® ¯�°�±�²L³ Ë�¾m±�Ì.ÍP°�»HÎ!Ï6½.Ä À È�D�D�f�� D ¥ � ¿&�.� �� ¬£ D ¥ ©Ê�f�i D ¥ � ¿&�.� �D�h�� ¬£ D ¥ ® ¯�°�±�²L³ Ë�¾Ð¸.Ì.ÍP°�»HÎ!Ï6½.Ä À È�D�D�D�f�i ¬£ D ¥ � ¿&�H� �� ¬£?Á�¥D£ D ¥ ©Ê�D�f�i ¬£ D ¥ � ¿&�.� �D�h�� ¬£�Á&¥D£ D ¥ ® ¯�°�±�²L³ Ë�¾Ð¸.Ì.ÍP°�»�ÑÒÌ=»HÎ�Ï5½.Ä À È�D�f�«� �H�Q� �� ¬£?Á�¥D£ D ¥ ©Ê�D�f�i ¬£?Á�¥D£ D ¥ � �H�Q� � ® ¯�°�±�²L³ Ë�¾Ð¸.Ì.ÍP°�»�ÑÒÌ=»HÎ=½.Ä�Ó À È�D�D�f�i �¤�¥��J¿&�H�0�� ¢¡P£! �¤�¥=¦�©Ê�f�i �¤�¥���¿&�.���D�h�� ¢¡Q£� �¤�¥=¦ ® ¯�°�±�²L³ Ë�¾m±�Ì�Ô�ÍQÕ»HÎ.Ï6½.Ä À È�D�f�«�i�H�Q���� ¡P£ ¤�¥=¦ ©Ê�D�f�i ¡P£ ¤�¥=¦ ���H�Q�i� ® ¯�°�±�²L³ Ë�¾m±�Ì�Ô�ÍQÕ»HÎD½.Ä�Ó À È�D�D�f�i D ¥ � ¿&�H� �� ¡P£ ¤�¥=¦ ©Ê�f�i D ¥ � ¿&�.� �D�h�� ¡Q£ ¤�¥=¦ ® ¯�°�±�²L³ Ë�¾m±�Ì�Ô�ÍQÕ»HÎ.Ï6½.Ä À È�D�f�«� �H�Q� �� ¥D£�Á%¬£ D ¥ ©Ê�D�f�i ¥D£�Á%¬£ D ¥ � �H�Q� � ® ¯�°�±�²L³ Ë�¾Ð¸.Ì.ÍPÌ=»�ÑÒ°�»=΢½.Ä�Ó À È�D�D�f�i ¤�¥H¦ � ¿&�H� �� ¥D£�Á%¬£ D ¥ ©Ê�f�i ¤�¥=¦ � ¿&�.� �D�h�� ¥¢£?Á&¬£ D ¥ ® ¯�°�±�²L³ Ë�¾Ð¸.Ì.ÍPÌ=»�ÑÒ°�»=Î!Ï5½.Ä À ÈÖRemark 29 (Explanation of the SVO.Tunepackage). With the
SVO.Tune packageasabove we provide an illustration, ratherthana fully devel-
opedpackage. It illustrates how (an importantpart of) Steedman’s modelof tune
canbeformulatedin DGL, at points in a moreelegantway (i.e. without having to
resortto emptyelements).
The rules are divided into linkage (or “specification”) rules and percolation
rules.Thelinkagerules assign nuclearstress (either in canonical or non-canonical
position), project the focusor the topic, andassignboundaries. The percolation
rules just percolate the verbal head’s featuresover the entirestructure. Note that
all theserules are definedover the headed ×&Ø modes- non-systemic ordering is
addressedin thenext section. Thedefinition thusprovidesa basicformalization of
theuseof tuneto realizetheinformation structure’s focusproper in a canonical or
noncanonical focus position, without resorting to any word order-relatedmeans.
Theexamplein (247) illustratestherealizationof aso-called“out-of-the-blue”
sentence, using tune as the structural indication of informativity. Reading top-
down, we first realize the focus proper in the canonical focus position, and then
project thefocusall theway leftwards.
(247)
.
.
.�D�D�actor
£?¬¤ h¤�¡ �J¿&�  ¾��D� tverbti £?¬¤ h¤�¡ ��Ù Â � �D� patient
£?¬¤ f¤�¡�À §�£�¨zÚ ½�D�D�actor
£?¬¤ h¤�¡ � ¿&�.� ¾?�D�tverbti
£?¬¤ h¤�¡ � Ù Â � �D� patient £?¬¤ h¤�¡ À §>£?¨ Ú ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£?¬¤ f¤�¡ ��¿&�.��D�tverbti
£?¬¤ h¤�¡�À �iÙ Â � �D� patient £?¬¤ f¤�¡ §>£?¨zÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£?¬¤ f¤�¡ � ¿&�.� �D�tverbti
£h¬¤ h¤�¡�À � �H�Q� �D�patient
£?¬¤ f¤�¡ §�£�¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £h¬¤ f¤�¡ ��¿&�.��D�
tverbti £?¬¤ h¤�¡ À ���.�Q�ª�D�
patient ¡P£ ¤�¥H¦ §�£?¨ Ú ½ ® ¯�°�±�²L³ ´�µi¶�·>°�¸.¹mºO»�¼P²!½!½.È
�D¾��D�actor
�£?¬¤% �h¤�¡��J¿&�H���D�tverbti
D¬£. �¤�¥ À ���H�Q�J�D�patient
¢¡Q£. �¤�¥=¦0 �§�£�¨ Ú ½ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQÈ�D¾��D�
actor ¬£ ¤�¥ � ¿&�.� �D�
tverbti ¬£ ¤�¥PÀ � �.�Q� �D�
patient ¡P£ ¤�¥=¦ §>£?¨kÚ ½ ® ¯�°�±�²Q³ µ�¶�¼�ÄHÉQÈ
�D�D�D�D�actor
¬£ ¤�¥ � ¿&�.�tverbti
¬£ ¤�¥ � �.�Q� �D�patient
¡P£ ¤�¥H¦ §�£?¨eÚ ½ ® ¯�°�±�²L³ Ë�¾m±�Ì.ÍP°�»=Î�ÏR½HÄ À È�D�D�D¾��D�
actor ¬£ ¤�¥ ��¿&�.�
tverbtiÀ �i�.�P�J�D�
patient ¡P£ ¤�¥=¦ ¬£ ¤�¥ §>£?¨ Ú ½ ® ¯�°�±�²L³ Ë�¾m±�Ì.ÍP°�»=Î=½.Ä�Ó À È
�D�D¾��D�actor
�¬£! �¤�¥���¿&�.�tverbti
À �i�.�Q�J�D�patient
=¡P£� �¤�¥=¦0 �¬£. �¤�¥ Ú-Ü�Ý §�£�¨ ½ ® Ü Ý Û�È�D¾��D�
actor ¬£ ¤�¥ � ¿&�H�
tverbtiÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ ¬£iÚaÜ Ý ¤�¥ Ü Ý §>£?¨ ½ ® Ü Ý Û�È
¾��D�actor
¬£ ¤�¥ � ¿&�.�tverbti
À � �.�P� �D�patient
¡P£ ¤�¥=¦�ÚaÜ Ý ¬£ Ü Ý ¤�¥ Ü Ý §�£�¨ ½ ® Ü�Ý Û�È
Next, wepresenttwo proofs with thenuclearstress in canonical focusposition,
168q A formal model of tuneasstructural indication of informativity
but realizing a different information structureby placing boundaries. Theseproofs
illustratethebasic ideabehind formalizing boundarytonesin a moreelegantway.
In (248) theverb is realizedwith a boundary tone,andin (249) theverbal head is
still partof thefocussotheActor is realizedwith a boundarytone.
(248)
.
.
.�D�D�actor
£?¬¤ h¤�¡ � ¿&�  ¾?�D� tverbti £?¬¤ h¤�¡ �iÙ Â � �D� patient
£?¬¤ h¤�¡�À §�£�¨kÚ ½�D�D�actor
£?¬¤ h¤�¡ � ¿&�H� ¾��D�tverbti
£h¬¤ h¤�¡ � Ù Â � �D� patient £?¬¤ h¤�¡�À §>£?¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£?¬¤ f¤�¡ ��¿&�.���D�tverbti
£h¬¤ h¤�¡�À �iÙ Â � �D� patient £?¬¤ h¤�¡ §>£?¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£?¬¤ f¤�¡ � ¿&�.� �D�tverbti
£h¬¤ f¤�¡�À � �H�Q� �D�patient
£?¬¤ f¤�¡ §>£?¨zÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £h¬¤ f¤�¡ � ¿&�.� �D�
tverbti £?¬¤ h¤�¡�À � �.�Q� �D�
patient ¡P£ ¤�¥H¦ §�£?¨eÚ ½ ® ¯�°�±�²L³ ´iµi¶�·>°�¸.¹mº�»?¼Q².½!½.È
�D¾��D�actor
�£?¬¤% �f¤�¡���¿&�.���D�tverbti
¢¬£?Á�¥D£!  ¥ À ���.�P�J�D�patient
=¡P£! �¤�¥=¦0 �§>£?¨ Ú ½ ® ¯�°�±�²L³ º�ÃJ��ÄQ°�±�ÅLƼPÇ�È�D¾��D�
actor ¬£ D ¥ � ¿&�H� �D�
tverbti ¬£�Á&¥D£ D ¥PÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ §�£�¨kÚ ½ ® ¯�°�±�²L³ ¯�¶i¼�ÄHÉQÈ
�D�D�D�D�actor
¬£ D ¥ � ¿&�.�tverbti
¬£�Á&¥D£ D ¥ � �H�Q� �D�patient
¡Q£ ¤�¥=¦ §�£�¨kÚ ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQ°�»�ÑÒÌ=»HÎ!Ï6½.Ä À È�D�D�D¾��D�
actor ¬£ D ¥ � ¿&�.�
tverbtiÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ ¬£�Á&¥¢£ D ¥ §�£�¨kÚ ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQ°�»�ÑÒÌ=»H΢½.Ä�Ó À È
�D�D¾��D�actor
�¬£!  ¥���¿&�.�tverbti
À �i�.�Q���D�patient
=¡P£! �¤�¥H¦0 �¬£?Á�¥D£!  ¥ ÚaÜ�Ý §�£�¨ ½ ® Ü Ý ÛPÈ�D¾?�D�
actor ¬£ D ¥ � ¿&�H�
tverbtiÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ ¬£�Á&¥D£�ÚaÜ Ý Â ¥ Ü Ý §�£�¨ ½ ® Ü Ý Û�È
¾��D�actor
¬£ D ¥ � ¿&�.�tverbti
À � �.�P� �D�patient
¡Q£ ¤�¥=¦�ÚaÜ Ý ¬£�Á&¥¢£ Ü Ý Â ¥ Ü Ý §>£?¨ ½ ® Ü�Ý Û�È
(249)
.
.
.�D�D�actor
£?¬¤ h¤�¡ � ¿&�  ¾?�D� tverbti £?¬¤ h¤�¡ � Ù Â � �D� patient
£?¬¤ h¤�¡�À §�£�¨kÚ ½�D�D�
actor £?¬¤ h¤�¡ �J¿&�H��¾��D�
tverbti £h¬¤ h¤�¡ ��Ù Â � �D� patient
£?¬¤ h¤�¡�À §>£?¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £?¬¤ f¤�¡ � ¿&�.� �D�
tverbti £h¬¤ h¤�¡�À � Ù Â � �D� patient
£?¬¤ h¤�¡ §>£?¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £?¬¤ f¤�¡ ��¿&�.���D�
tverbti £h¬¤ f¤�¡ À �i�H�Q�J�D�
patient £?¬¤ f¤�¡ §>£?¨ Ú ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£h¬¤ f¤�¡ � ¿&�.� �D�tverbti
£?¬¤ h¤�¡�À � �.�Q� �D�patient
¡P£ ¤�¥H¦ §�£?¨eÚ ½ ® ¯�°�±�²L³ ´iµi¶�·>°�¸.¹mº�»?¼Q².½!½.È�D¾��D�
actor £?¬¤ h¤�¡ � ¿&�H� �D�
tverbti ¬£ ¤�¥PÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ §�£�¨zÚ ½ ® ¯�°�±�²L³ µi¶i¼�ÄHÉQÈ
�D¾��D�actor
¬£?Á�¥D£ D ¥ � ¿&�.� �D�tverbti
¬£ ¤�¥�À � �.�Q� �D�patient
¡Q£ ¤�¥=¦ §�£�¨zÚ ½ ® ¯�°�±�²L³ º�ÃJ�iÄQ°�±�ÅQƼPÇ�È�D�D�D�D�
actor ¬£?Á�¥D£  ¥ �J¿&�H�
tverbti ¬£ ¤�¥ �i�H�Q�ª�D�
patient ¡Q£ ¤�¥=¦ §�£�¨ Ú ½ ® ¯�°�±�²L³ Ë�¾m±�ÌHÍQ°�»HÎ�Ï5½.Ä À È
�D�D�D¾��D�actor
¬£?Á�¥D£ D ¥ � ¿&�H�tverbti
À � �.�Q� �D�patient
¡P£ ¤�¥=¦ ¬£ ¤�¥ §�£�¨zÚ ½ ® ¯�°�±�²L³ Ë�¾m±�ÌHÍQ°�»HÎ=½.Ä>Ó À È�D�D¾��D�
actor ¬£�Á&¥¢£ D ¥ � ¿&�H�
tverbtiÀ � �H�Q� �D�
patient ¡Q£ ¤�¥=¦ ¬£ ¤�¥�ÚaÜ Ý §�£�¨ ½ ® Ü�Ý ÛPÈ
�D¾?�D�actor
¬£?Á�¥D£ D ¥ � ¿&�.�tverbti
À � �.�P� �D�patient
¡P£ ¤�¥=¦ ¬£�ÚaÜ Ý ¤�¥ Ü Ý §�£�¨ ½ ® Ü Ý Û�Ⱦ��D�
actor ¬£?Á�¥D£  ¥ �J¿&�H�
tverbtiÀ �i�.�Q�J�D�
patient ¡Q£ ¤�¥=¦ ÚaÜ Ý ¬£ Ü Ý ¤�¥ Ü Ý §>£?¨ ½ ® Ü Ý Û�È
Finally, we canrealize a focusproper that is in canonical word order position
but which is not sentence-final. In this case,the nuclearstress occursin a non-
canonical focusposition. Theproofsbelow illustratetwo simplecaseswhereeither
the Actor (250) or the verbal head (251) is the focus proper in the underlying
informationstructure. Not resorting to word order, we canrealize themherejust
using nuclear stress, like is done for example in English.
A formalmodelof tuneasstructuralindication of informativity /169
(250)
.
.
.�D�D�actor
£?¬¤ h¤�¡ � ¿&�  ¾��D� tverbti £?¬¤ h¤�¡ ��Ù Â � �D� patient
£?¬¤ f¤�¡�À §�£�¨zÚ ½�D�D�
actor £?¬¤ h¤�¡ � ¿&�.� ¾?�D�
tverbti £?¬¤ h¤�¡ � Ù Â � �D� patient
£?¬¤ h¤�¡ À §>£?¨ Ú ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £?¬¤ f¤�¡ � ¿&�.� �D�
tverbti £?¬¤ h¤�¡�À �iÙ Â � �D� patient
£?¬¤ f¤�¡ §>£?¨zÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £?¬¤ f¤�¡ � ¿&�.� �D�
tverbti £h¬¤ h¤�¡�À � �H�Q� �D�
patient £?¬¤ f¤�¡ §�£�¨kÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾?�D�actor
¡P£ ¤�¥=¦ � ¿&�.� �D�tverbti
¥D£?Á&¬£  ¥ À � �.�Q� �D�patient
¬£  ¥ §�£�¨ Ú ½ ® ¯�°�±�²L³ ·�´iµ�¶i·�°�¸!¹mºO»�¼P²!½!½HÈ�D�D�D�D�
actor ¡P£ ¤�¥H¦ �J¿&�H�
tverbti ¥D£?Á&¬£ D ¥ ���.�Q�J�D�
patient ¬£ D ¥ §�£�¨zÚ ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQÌ=»�ÑÒ°�»HÎ!Ï6½.Ä À È
�D�D�D¾?�D�actor
=¡P£! �¤�¥H¦��J¿&�H�tverbti
À �i�.�Q�ª�D�patient
�¬£!  ¥P D¥D£�Á&¬£.  ¥P �§�£�¨ Ú ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQÌ=»�ÑÒ°�»H΢½.Ä>Ó À È�D�D¾��D�
actor ¡Q£ ¤�¥=¦ � ¿&�.�
tverbtiÀ � �H�Q� �D�
patient ¬£ D ¥ ¥¢£?Á&¬£ D ¥�ÚaÜ Ý §�£�¨ ½ ® Ü Ý ÛPÈ
�D¾��D�actor
¡Q£ ¤�¥=¦ � ¿&�.�tverbti
À � �.�Q� �D�patient
¬£  ¥ ¥D£�Á%¬£ Ú-Ü Ý Â ¥ Ü Ý §�£�¨ ½ ® Ü�Ý Û�Ⱦ��D�
actor ¡P£ ¤�¥H¦ �J¿&�H�
tverbtiÀ ���H�Q�J�D�
patient ¬£ D ¥JÚaÜ Ý ¥D£�Á&¬£ Ü Ý Â ¥ Ü Ý §>£?¨ ½ ® Ü Ý Û�È
(251)
.
.
.�D�D�actor
£?¬¤ h¤�¡ � ¿&�  ¾��D� tverbti £?¬¤ h¤�¡ � Ù Â � �D� patient
£?¬¤ f¤�¡�À §�£�¨zÚ ½�D�D�
actor £?¬¤ h¤�¡ ��¿&�.��¾?�D�
tverbti £?¬¤ h¤�¡ ��Ù Â � �D� patient
£?¬¤ h¤�¡�À §>£?¨zÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor £?¬¤ f¤�¡ � ¿&�.� �D�
tverbti £?¬¤ h¤�¡�À � Ù Â � �D� patient
£?¬¤ f¤�¡ §>£?¨zÚ ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ�D¾��D�
actor D£?¬¤& �f¤�¡���¿&�.���D�
tverbti ¢£h¬¤% �h¤�¡ À �i�H�Q�J�D�
patient �£?¬¤& �f¤�¡� �§�£�¨ Ú ½ ® ÛP±�Õ0³ ºOÇ0½�Ã�¼�ÅQÈ
�D¾��D�actor
£?¬¤ f¤�¡ � ¿&�.� �D�tverbti
¡P£ ¤�¥=¦0À � �.�Q� �D�patient
¥D£?Á&¬£ D ¥ §�£�¨kÚ ½ ® ¯�°�±�²L³ ·�´iµ�¶i·�°�¸!¹mºO»�¼P²!½!½HÈ�D¾?�D�
actor ¬£ ¤�¥ � ¿&�H� �D�
tverbti ¡Q£ ¤�¥=¦0À � �H�Q� �D�
patient ¥D£�Á%¬£ D ¥ §�£�¨zÚ ½ ® ¯�°�±�²L³ µi¶�¼PÄHÉPÈ
�D�D�D�D�actor
¬£ ¤�¥ ��¿&�H�tverbti
¡Q£ ¤�¥=¦ ���H�Q�J�D�patient
¥D£�Á&¬£ D ¥ §�£�¨zÚ ½ ® ¯�°�±�²L³ Ë�¾m±�Ì�Ô�ÍLÕ»HÎHÏR½HÄ À È�D�D�D¾?�D�
actor D¬£. �¤�¥���¿&�H�
tverbtiÀ ���H�Q�J�D�
patient D¥D£�Á&¬£!  ¥� ¢¡P£� �¤�¥=¦0 �§�£�¨ Ú ½ ® ¯�°�±�²L³ Ë�¾m±�Ì�Ô�ÍLÕ»HÎ�½.Ä�Ó À È
�D�D¾��D�actor
¬£ ¤�¥ � ¿&�.�tverbti
À � �.�Q� �D�patient
¥D£?Á&¬£ D ¥ ¡Q£ ¤�¥=¦�ÚaÜ Ý §�£�¨ ½ ® Ü Ý ÛPÈ�D¾?�D�
actor ¬£ ¤�¥ � ¿&�H�
tverbtiÀ � �.�Q� �D�
patient ¥D£�Á%¬£ D ¥ ¡P£iÚ-Ü Ý ¤�¥=¦ Ü Ý §�£?¨ ½ ® Ü�Ý Û�È
¾��D�actor
¬£ ¤�¥ ��¿&�.�tverbti
À ���H�Q�J�D�patient
¥¢£?Á&¬£ D ¥�Ú-Ü Ý ¡Q£ Ü Ý ¤�¥=¦ Ü Ý §�£�¨ ½ ® Ü Ý ÛPÈ
5.4 INTERACTING TUNE AND WORD ORDER
The informativity hypothesesin Chapter 3 predict interaction betweentune and
wordorderasstructural indicationsof informativity. In Chapter4 weprovidedthe
architectures for word order asa structural indications of informativity, andin the
current chapterwediscussedtune.Whatboth accountshavein commonis thatthey
havebeendefined, notrelatively to themodesindicatinggrammatical relationslike
subject or object, but relatively to systemicordering. In otherwords,weextend the
Praguianview relating word order asa structural indication of informativity and
systemicordering to cover tuneaswell. Thereis a definiteadvantagein doingso.
We cannow describe the interactionbetween tuneandword order in theseterms
aswell. All we need to do is elaboratethemodelpresentedin Definition 33 such
thattuneis sensitive to bothsystemicandnon-systemic ordering.To roundoff this
chapter, we just illustratetheprincipal ideason a smallsetof rules,(252).
170q A formal model of tuneasstructural indication of informativity
(252)
�D¾m�«� ¤ ��� � À � ¤ �P� ´� §�£�¨ ©Þ�D¾m�5� Ù Â � � À �  � ´i §>£?¨ ® ÛQ±�Õ³ ·>ÄQ±�ºOÇ0½�Ã�¼PÅQÈ�f�5� ¿ ¤ � �� §�£�¨ ©Þ�f�6� ¿ Ù Â �� §�£?¨ ® ÛQ±�Õ³ ·>ÄQ±�ºOÇ0½�Ã�¼PÅQÈ�D¾m�«��¿ ¤ �Q� À � ¤ �P��´� �§�£�¨�©Þ�f�6��¿ ¤ ��¾f�R���  �i´ À �§�£�¨ ® ÛQ±�Õ³ ·>ÄQ±�ºOÇ0½�Ã�¼PÅQÈ�D¾m�5��¿&�.�L� À � ¤ �P��´� �§�£�¨�©Þ�f�6��¿&�H��¾f�5� ¤ ���i´ À �§>£?¨ ® ÛQ±�Õ³ ·>ÄQ±�ºOÇ0½�Ã�¼PÅQȾm�5��¿&�.��� À � ¤ �P�i´Ò©ß�«�J¿&�H��¾f�6� ¤ ����´ À ® ÛQ±�Õ³ ·>ÄQ±�ºOÇ0½�Ã�¼PÅQÈ�D�f�5���.�Q���� ¬£ ¤�¥ ©Þ�D�f�� ¬£ ¤�¥ ���H�Q�i� ® ¯�°�±�²L³ Ë�¾m±�ÌHÍP°�»HÎ=½HÄ�Ó À È�D�f�«� ¤ �P� �� ¬£ D ¥ ©Þ�D�f�� ¬£ D ¥ � ¤ �P� � ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍP°�»HÎD±�½>Ó À È�D�f�5� ¤ ��� �� ¬£?Á�¥D£ D ¥ ©Þ�D�f�� ¬£?Á�¥D£ D ¥ � ¤ ��� � ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍP°�»�Ñ«Ì=»H΢±�½�Ó À È�D�D�D�f�i ¬£ D ¥ � ¿ ¤ � �� ¬£?Á�¥D£ D ¥ ©Þ�D�f�� ¬£ D ¥ � ¿ ¤ � �D�h�� ¬£�Á&¥¢£ D ¥ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍP°�»�Ñ«Ì=»HÎ!Ï6±�½ À È�D¾Ð�6� ¤ �P���D�h�� �¬£�Á&¥D£P  ¥ À � ¤ �P���D�h´� ¢¡P£P �¤�¥=¦0 �§�£�¨ª©Þ�D¾m�5� ¤ �P���D�h�� �£h¬¤� �f¤�¡ À � ¤ �����D�h´i ¢¡Q£� �¤�¥=¦0 �§�£�¨ ® ¯�°�±�²L³ ·�º%Ã��iÄQ°�±�ÅLÆ0¼QÇLÈ�D¾Ð�6��¿ ¤ ���D�h�� ¬£�Á&¥D£ D ¥PÀ � ¤ �P���D�h´� ¡P£ ¤�¥=¦ §�£�¨ ©Þ�D¾m�5�J¿ ¤ ���D�h�� £h¬¤ f¤�¡�À � ¤ �����D�h´i ¡Q£ ¤�¥=¦ §�£�¨ ® ¯�°�±�²L³ ·�º%Ã��iÄQ°�±�ÅLÆ0¼QÇLȾ��D�f�i ¬£ D ¥ � ¿&�.� �D�h�� ¬£ D ¥�À � ¤ ��� �D�h´i ¬£�Á�¥D£ D ¥ ©Þ¾��D�f�i £?¬¤ h¤�¡ � ¿&�H� �D�h�� £h¬¤ f¤�¡�À � ¤ �P� �D�h´i ¬£�Á&¥D£ D ¥ ® ¯�°�±�²L³ ¯�¶�¼�ÄHÉQÈ�f�5� ¤ �P� �D�h�� ¡P£ ¤�¥=¦ §�£�¨ ©Þ�f�6� ¤ �P� �D�h�� £?¬¤ h¤�¡ §>£?¨ ® ¯�°�±�²L³ ´iµ�¶i·�°�¸!¹mºO»�¼P²!½.½.È
For completeness, (252) repeatssomeof the rulesdefinedin InfSVO for de-
scribing theeffect of mixed word order on systemic ordering. We usetheserules
in the examplebelow, (253). Besidesthese rules, we describe the placementof
nuclearstress andof boundaries,topic projection over non-systemic ordering,and
thenecessarypercolation rules.
(253)
.
.
.�D�D�patient
£?¬¤ f¤�¡ � ¿ Ù Â ¾��D� tverbti £h¬¤ h¤�¡ � �  � �D� actor
£h¬¤ f¤�¡�À §�£?¨eÚ ½�D�D�
patient £?¬¤ h¤�¡ �J¿ ¤ �0¾��D� tverbti
£?¬¤ h¤�¡ ���  �J�D� actor £?¬¤ h¤�¡�À §>£?¨zÚ ½ ® ÛP±�Õ0³ ·>ÄQ±�ºOÇ0½!Ã�¼�ÅLÈ
�D¾��D�patient
�£?¬¤& �h¤�¡��J¿ ¤ ���D� tverbti D£?¬¤& �h¤�¡ À � ¤ ���J�D� actor
D£?¬¤& �f¤�¡� �§�£�¨ Ú ½ ® ÛP±�Õ0³ ·>ÄQ±�º�Ç�½�Ã�¼�ÅLÈ�D¾��D�
patient £?¬¤ f¤�¡ � ¿ ¤ � �D� tverbti
£?¬¤ f¤�¡�À � ¤ �P� �D� actor ¡P£ ¤�¥H¦ §�£?¨zÚ ½ ® ¯�°�±�²L³ ´�µi¶�·>°�¸.¹mºO»�¼P²!½!½.È
�D¾��D�patient
£?¬¤ h¤�¡ � ¿ ¤ � �D� tverbti ¬£�Á&¥¢£ D ¥�À � ¤ �P� �D� actor
¡Q£ ¤�¥=¦ §�£�¨kÚ ½ ® ¯�°�±�²L³ ·�º�ÃJ�iÄQ°�±�ÅQƼPÇ�È�D¾?�D�
patient ¬£ D ¥ ��¿ ¤ �0�D� tverbti
¬£?Á�¥D£ D ¥PÀ � ¤ �����D� actor ¡Q£ ¤�¥=¦ §�£�¨zÚ ½ ® ¯�°�±�²L³ ¯�¶i¼�ÄHÉQÈ
�D�D�D�D�patient
�¬£!  ¥���¿ ¤ � tverbti ¢¬£?Á�¥D£�  ¥�� ¤ �����D� actor
¢¡Q£. �¤�¥=¦0 �§�£�¨ Ú ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQ°�»�Ñ-Ì=»HÎ�Ï6±�½ À È�D�D�D¾?�D�
patient ¬£ D ¥ � ¿ ¤ � tverbti
À � ¤ �P� �D� actor ¡Q£ ¤�¥=¦ ¬£�Á&¥¢£ D ¥ §�£�¨zÚ ½ ® ¯�°�±�²L³ Ë�¾Ð¸!ÌHÍQ°�»�Ñ-Ì=»HÎ=±�½�Ó À È
�D�D¾��D�patient
¬£ D ¥ � ¿ ¤ � tverbtiÀ � ¤ �P� �D� actor
¡P£ ¤�¥=¦ ¬£�Á�¥D£ D ¥�ÚaÜ Ý §�£�¨ ½ ® Ü�Ý ÛPÈ�D¾?�D�
patient ¬£ D ¥ ��¿ ¤ � tverbti
À � ¤ ���J�D� actor ¡Q£ ¤�¥=¦ ¬£?Á�¥D£iÚ-Ü Ý Â ¥ Ü Ý §�£?¨ ½ ® Ü Ý Û�È
SUMMARY
In this chapter, we first discussedSteedman’s theory integrating tuneinto a CCG of En-
glish. We criticized Steedman’s useof emptystringsto model boundarytones,andpre-
sentedan abstractmodel of Steedman’s account of tune that shows how we canmodel
boundarytonesascomplex feature labels.A furtherdistinctionis thatthemodelinterprets
tuneson wordgroups relative to whetherthesewordsappearin systemicordering. This
perspective takesthePraguian view on therelationbetweenword order andsystemicor-
deringto tune. Theadvantageof doing so is that the interactionbetweentuneandword
ordercanbe(formally) describedin termsof how tunesshouldbe interpretedrelative to
systemicallyandnon-systemicallyorderedwordgroups. Weendedthechapterwith abrief
discussionof suchadescription,andshowedanexampleinvolving theinteractionbetween
tuneandmixednon-systemicallyorderedrealizations of dependentsin anSVO language
type.
CHAPTER 6
DGL, TOPIC/FOCUS, AND DISCOURSE
A hybrid logic modelingTF-DRT
This chapter explains how we formalize the interpretation of a sentence’s linguistic meaning,
with its information structure, in thecontext of a largerdiscourse.Wearguewhy notonly infor-
mationstructure but alsodependency relationsarefundamentally importantto discourseinter-
pretation. To illu stratetheargument,we proposea rudimentary information structure-sensitive
discoursetheory like (Kruij ff-Korbayova,1998)thathooksup with DGL, andin which we for-
malizethebinding of varioustypes of anaphors. Theoverall effort enablesusto –in principle–
cover theentire trackfrom sentential form to linguistic meaningto discourseinterpretation.
It is theabstractive power of ordinaryspeech
which rendersit morelogically powerful than
any algebraof logic hithertodeveloped.
– CharlesS.Peirce,Logic Notebook, 1898
6.1 INTRODUCTION
Information structure is an essential aspect of a sentence’s linguistic meaning. It
indicateshow thesentence’s linguistic meaning is beingpresentedasbothdepen-
denton theprecedingdiscoursecontext, andhow thatmeaning affects thecontext.
In Chapter 2 we already explainedhow the important ingredients of information
structure (contextual boundness) are representedin DGL, and how we derive a
topic-focusarticulation from thecontextual boundnessof theindividual nodesin a
linguistic meaning. Chapters 3 through 5 elaboratedon how we cananalysestruc-
tural indicationsof informativity asreflecting theunderlying informationstructure.
As an example, consider the proof in (254) and the representation of the corre-
sponding linguistic meaning in (255).
171
172q DGL, topic/focus,anddiscourse
(254)
à ³elijah á Ü Ý £?¬¤ Ü Ý h¤�¡�â&ã  £ ± ä%²!åæ ³ç�elijah
£?¬¤ á Ü Ý f¤�¡ â ã  £ ± Ü Ýè ¾?é Àê ³ç�D�elijah
£?¬¤ h¤�¡ á â ã  £ ± Ü Ý è ¾Ðë Àì ³kathy á Ü Ý £?¬¤ Ü Ý h¤�¡�â ã ÙPÙPí�± ä%²!åî ³ç�kathy
£?¬¤ á Ü Ý f¤�¡�â ã ÙPÙ�í± Ü Ý è ¾Ðï Àð ³ç�D�kathy
£?¬¤ h¤�¡ á â ã ÙPÙ�í ± Ü Ýè ¾mñ Àò ³letters á Ü Ý £?¬¤ Ü Ý h¤�¡�â�ó�ã £ ± ä%²!åô ³ç�letters
£h¬¤ á Ü Ý h¤�¡�â ó�ã £ ± Ü Ýè ¾mõ Àö ³ç�D�letters
£h¬¤ h¤�¡ á â ó�ã £ ± Ü Ý è ¾Ð÷ Àà=ø ³r ùzá ± ú�ÇQËà�à ³ �r ù ã  £ á â ã  £ ± â Û�¾ à=ø ÀàHæ ³
sent á Ü Ý £?¬¤ Ü Ý h¤�¡ Ü Ý0ûQü0ý ã Ù�¾�¾?¾ â&ã  £ ±�þ0¿&�  ½ À ÍQÙ Â � âOó�ã £ ± À Í Â � â ã ÙQÙ�í0± À ä%²!åà=ê ³ �sent £?¬¤ á Ü Ý f¤�¡ Ü Ý ûLü�ý ã Ù ¾?¾�¾ â ã  £ ±�þ ¿&�  ½ À Í Ù Â � â ó�ã £ ± À Í Â � â ã ÙPÙPí ± À Ü Ýè ¾?é!ë Àà ì ³ �D�
sent £?¬¤ h¤�¡ á Ü Ý ûLü�ý ã Ù�¾?¾�¾ â%ã  £ ±�þ¿&�  ½ À ÍLÙ Â � â�ó0ã £ ± À Í Â � â ã ÙPÙ�í0± À Ü Ýè ¾?é!ÿ ÀàHî ³ �D�D�
sent £?¬¤ h¤�¡ ûQü0ý ã Ù á ¾�¾ â ã  £ ±�þ ¿&�  ½ À Í Ù Â � â ó0ã £ ± À Í Â � â ã ÙPÙ�í ± Ü Ýè ¾?é!ï Àà=ð ³
q �>á ± ú�ÇQËà ò ³ �q � ã ÙQÙ�í á â ã ÙPÙ�í ± â Û�¾ à=ð Àà=ô ³ �D�D�
sent £?¬¤ h¤�¡ ûLü�ý ã Ù �  � � q � ã ÙPÙ�í á ¾ â ã  £ ±�þ ¿&�  ½ À ÍQÙ Â � â ó�ã £ ± Í è ¾ àHî Î à ò Àà=ö ³
s � á ± ú�ÇQËæ!ø ³ �s � ó�ã £ á â ó�ã £ ± â Û�¾ à=ö ÀæLà ³ ¾��D�D�
sent �£?¬¤% �f¤�¡� ûQü0ý ã Ù �  � � q � ã ÙPÙ�í À ��Ù Â � � s � ó�ã £ á â&ã  £ ±�þ0¿&�  ½ Í è ¾ à=ô Î æ!ø Àæ�æ ³ �
r ù ã  £ � ¿&�  ¾�¾��D�D� sent £?¬¤ h¤�¡ ûQü0ý ã Ù �  � � q � ã ÙPÙ�í À � Ù Â � � s� ó�ã £.À á ½ þ è ¾ à�à Î æLà Àæ!ê ³ �
r ù ã  £ �J¿&�  ¾��D�D� sent £h¬¤ h¤�¡ �  � � q � ã ÙPÙ�í ûQü0ý ã Ù ��Ù Â � � s� ó0ã £!À á ½ Ë�¾����².Æ0Å�Î��f¸ªÓ À ¾ æ�æ Àæ ì ³ �
r ù ã  £ � ¿&�  �D¾��D� sent £h¬¤ h¤�¡ �  � � q � ã ÙPÙ�í À � Ù Â � � s � ó�ã £ ûQü0ý ã Ù á ½ Ë�¾����².Æ0Å�ÎDÅL¸ªÓ À ¾ æ!ê Àæ�î ³ �D�
r ù ã  £ ��¿&�  ¾?¾��D� sent £?¬¤ h¤�¡ �  � � q � ã ÙPÙ�í À ��Ù Â � � s � ó�ã £ À ûQü0ý ã Ù á ½ Ë�¾����².Æ0Å�Î!Ï6½.¸ À ¾ æ ì Àæ!ð ³ �D�
r ù ã  £ � ¿&�  ¾?¾��D� sent £?¬¤ h¤�¡ �  � � q � ã ÙPÙ�í À � Ù Â � � s � ó�ã £.À §�£?¨ á ½ �5Æ»�¼��få�h½%Ì=Æ0½�f¸&º���ÃR¾ æ�î Àæ ò ³ �D�
r ù ã  £ � ¿&�  ¾?¾��D� sent £?¬¤ h¤�¡ �  � � q � ã ÙPÙ�í À �iÙ Â � �D� letters
£?¬¤ h¤�¡�À §>£?¨ á ½ â è ¾ ö Î à=ö Î æ!ð Àæ!ô ³ �D�r ù ã  £���¿&�  ¾?¾��D� sent
?£?¬¤% �h¤�¡��  � � q � ã ÙPÙ�í À �iÙ Â � �D� letters �¬£!  ¥ À �§>£?¨ á ½ ¹?¾��h±�Õ�ÍQ»�°�±%Î�¸!ÌHÍQ°�» À ¾ æ ò Àæ!ö ³ �D�
r ù ã  £ � ¿&�  ¾?¾��D� sent £?¬¤ h¤�¡ �  � �D� kathy
£h¬¤ h¤�¡ À �iÙ Â � �D� letters ¬£  ¥ À §>£?¨ á ½ â è ¾ ð Î à=ð Î æ!ô Àê�ø ³ �D�
r ù ã  £���¿&�  ¾?¾��D� sent ?£?¬¤& �h¤�¡ª�  � �D� kathy
=¡P£! �¤�¥=¦ À ��Ù Â � �D� letters �¬£!  ¥ À �§�£�¨ á ½ ¹?¾��h±�Õ�ÍQ»�°�±%ÎD±�Ì�Ô�ÍQÕ» À ¾ æ!ö Àê�à ³ �D�
r ù ã  £ � ¿&�  ¾?¾��D� sent ¬£�Á�¥D£ D ¥ �  � �D� kathy
¡Q£ ¤�¥=¦0À � Ù Â � �D� letters ¬£ D ¥QÀ §>£?¨ á ½ ¹?¾��h±�Õ�ÍQ»�°�±%Î�¸!ÌHÍQ°�»�Ñ-Ì=» À ¾ ê�ø ÀêPæ ³ �D�
r ù ã  £ ��¿&�  ¾?�D� sent�  � �D� kathy
¡P£ ¤�¥=¦ ¬£�Á�¥D£ D ¥ ��Ù Â � �D� letters ¬£ D ¥PÀ §�£�¨ á ½ Ë�¾Ð¸.Ì.ÍP°�»�ÑÒÌ=»HÎ �f¸JÓ À ¾ ê�à Àê�ê ³ �D�
r ù ã Â £ � ¿&� Â �D�D¾ sent� Â � �D� kathy
¡P£ ¤�¥=¦0À � Ù Â � �D� letters ¬£ D ¥ ¬£�Á&¥¢£ D ¥ §�£�¨ á ½ Ë�¾Ð¸.Ì.ÍP°�»�ÑÒÌ=»H΢ÅL¸JÓ À ¾ êPæ Àê ì ³ç�D�D�
elijah £?¬¤ h¤�¡ �J¿&�  �D�D¾ sent
�  � �D� kathy ¡Q£ ¤�¥=¦0À ��Ù Â � �D� letters
¬£ D ¥ ¬£�Á&¥D£ D ¥ §�£?¨ á ½ â è ¾ ê Î à=ø Î ê�ê ÀêPî ³ç�D�D�elijah
¬£ D ¥ � ¿&�  �D�D¾ sent�  � �D� kathy
¡Q£ ¤�¥=¦0À � Ù Â � �D� letters ¬£ D ¥ ¬£�Á&¥D£ D ¥ §�£?¨ á ½ ¹?¾��h±�Õ�ÍQ»�°�±%Î�¸!ÌHÍQ°�» À ¾ ê ì Àê�ð ³ç�D�D�D�D�
elijah ¬£ D ¥ � ¿&�  ¾?¾ sent
�  � �D� kathy ¡Q£ ¤�¥=¦0À � Ù Â � �D� letters
¬£ D ¥PÀ ¬£?Á�¥D£ D ¥ §�£�¨ á ½ Ë�¾Ð¸.Ì.ÍP°�»�ÑÒÌ=»HÎ�Ï5½.¸ À ¾ êPî Àê ò ³ç�D�D�D�elijah
¬£ D ¥ ��¿&�  ¾�¾ sent�  � �D� kathy
¡Q£ ¤�¥=¦0À �iÙ Â � �D� letters ¬£ D ¥�À ¬£�Á�¥D£ D ¥ á Ü Ý §�£�¨ ½ Ü Ý Ûi¾mÿ�� Àê�ô ³ç�D�D�
elijah ¬£ D ¥ � ¿&�  ¾�¾ sent
�  � �D� kathy ¡Q£ ¤�¥=¦0À �iÙ Â � �D� letters
¬£ D ¥QÀ ¬£�Á�¥D£ á Ü Ý Â ¥ Ü Ý §�£�¨ ½ Ü Ý Ûi¾mÿ�õ Àê�ö ³ç�D�elijah
¬£ D ¥ � ¿&�  ¾�¾ sent�  � �D� kathy
¡P£ ¤�¥=¦0À � Ù Â � �D� letters ¬£ D ¥QÀ á Ü Ý ¬£�Á&¥¢£ Ü Ý Â ¥ Ü Ý §>£?¨ ½ Ü Ý Ûi¾mÿL÷ ÀÖ
(255) Linguistic meaning of (254):����� n CB o ����������� �"!#� n CB o�$ ACTOR % �'&(����)* +�,.-/!� n NB* o�$ PATIENT % �'01��23,546-87"!#� n NB o�$ ADDRESSEE % �9:��);�<4�4���=6!>!with topic-focusarticulation:?�@�A�B
CB C A�D�E������ �"!#EFBCB C�$ ACTOR % A'&(E���)* +�,.-/!
G BNB C�$ ADDRESSEE % A9HEI)��J4�4��<=K!LEFB
NB* C�$ PATIENT % A'0ME�23,54N-87/!>!
Weneed to makeonefurtherstep. Wealready indicatedin Chapter2 that arep-
resentation of theform?O@�AP G(Q !
is interpreted dynamically. In thenext section
we definethis processof information structure-sensitive dynamic interpretation,
andin R 6.3we presenta basic approachto binding.
6.2 DYNAMIC INTERPRETATION OF INFORMATION STRUCTURE
The proper placefor describing the interpretation of informationstructure is dis-
course,andwith that in mind we proceedin the current section asfoll ows. First,
DGL, topic/focus,anddiscourse /173
weconsideraninformation structure-sensitivediscourserepresentation theory. The
principal ideasbehind ourproposalcomefrom Kruijf f-Korbayova (1998). Wenote
afew problemsfor thetyped approachtaken in (Kruijf f-Korbayova,1998), andwe
discusshow they areovercome(alreadyin DGL). Thereafter, wedefinethemodel-
theoretic dynamic interpretation of informationstructure,giveninformationstruc-
ture-sensitivediscourserepresentations. Thisthemeis continuedin thenext section
(6.3). With that,we have essentially arrivedat a proposalthat in principle covers
theentire track between a sentence’s surface form andits eventual interpretation-
in-context, all from aPraguianview, (without theclaimof course thattheproposal
is any way complete).
Kruijf f-Korbayova (1998) proposesto split DRT’s discourserepresentation struc-
ture (DRS) into two parts - a topic-partanda focus-part.1 Technically, the focus-
box andthe topic-box aredefinedas S -DRSs(Kuschert, 1996). Theboxestyped
non-rigidly, in that it dependson thestructureof thetopic andthefocus which el-
ementsact asargumentsandwhich asfunctors.2 Abstractly, Kruijf f-Korbayova’s
TF-DRS take theform asin (256).
(256) TOPICG
FOCUS
An exampleTF-DRSis given in (257), (Kruijf f-Korbayova, 1998)(p.72-73).
Werepeat Kruijf f-Korbayova’s notation of dependency relations.
(257) a. Czech
Muzman-nom
potkalmeet-Past
vin
parkupark
DIVKU.girl-acc
“The manmeta girl in a park.”T U(Actor :manVXW'Y�Z T [ (meetV�\�Y (Locative:parkVX\�Y (Patient:girl V�\�Y�Z
1As Peregrin notes,thereis anearlierattemptto accountfor topic-focusarticulationin a frame-work at leastsimilar to DRT. This account is Peregrin & Sgall (1986). QuotingPeregrin, “[i]n thisframework, eachsentenceis associatedwith a situation-like structure(the“content” of a sentence);the“meaning”of asentenceis thenunderstoodastheclassof all theembeddingsof the“content” intothemodel.A sentencearticulatedinto atopicandafocusis consideredastrueif everyembeddingofthe“content” of its topic is meaningfullyextensibleto anembedding of the “content” of thewholesentence.” (Peregrin, 1995)(p.237).Kruijf f-Korbayova stayswith DRT (thatis, ] -DRT), developinganintensionallogic aroundPeregrin’s (1995)extensionalaccount of topic-focusarticulation,andisthefirst to proposeto split a DRSrepresentationinto a topic-partanda focus-part.
2This presentsan attemptat generalizingJackendoff’s ideaof viewing the focusalways asanabstraction,anideawhich is alsofollowedin (Peregrin, 1995).
174 DGL, topic/focus,anddiscourse
b. S`_ .
x
man(x)
x=?
P(x)
G S8a�bcSede, y, p
f(meet(e, Actor:u,Patient: y))
Locative(e,p)
park(p)
A TF-DRSlike theonegivenin (257b) canberelaxedto simulatea S -DRS.To
obtain a S -DRS,weregardG
as S -DRT’sapplication operator“@”, afterwhichwe
can g -reducetheduplex T/F-condition into a gih -normalform, (Kruijf f-Korbayova,
1998)(p.84).
However, asentence’stopic-focusarticulation is usually notasneatly separated
as in (257), with just one element in the topic - or the complementcase, with
just oneelementin the focus. For example, consider (258), (Kruijf f-Korbayova,
1998)(p.86).
(258) a. Czech
Muzman-nom
dıvkugirl-acc
potkalmeet-Past
vin
PARKU.park
“The manmeta girl in a park.”T U(Actor :manVXW'Y (Patient:girl VXW'Y'Z T [ (meetV�\�Y (Locative:parkV�\�Y'Z
In the approach that Kruijf f-Korbayova takes, examples like (258) present a
problem. Thedependentswehavein thetopicyield separatepartial S -DRSs.Each
of thesepartial S -DRSsneedto be combined with the verb’s predicative S -DRS.
However, asKruijf f-Korbayova notes,that would meanthat therewould have to
be multiple functional applications joining the topic-partandthe focus-part. Un-
fortunately, the verb’s S -DRS belongs to the focus (in the caseof (258)), andso
the functional applicationdoes not take place in the TF-DRS construction. Con-
sequently, we could potentially end up with multiple S -DRS for eachpart of a
TF-DRS. Kruijf f-Korbayova notesthatthis is undesirable.
Thereareseveral options opento solve this problem. From the viewpoint of
type logic we could think of using pairing andprojection. Alternatively, Peregrin
(p.c.) proposesto usemultiple S -abstraction and conversion in which multiple
argumentscanbe absorbed in a single application andconversionstep. Kruijf f-
Korbayova suggestsa “wrapping operation” (1998)(p.87) that inserts a dummy
variable into thetopicwhich is of thesametypeasthetypeof thepredicative focus
DGL, topic/focus,anddiscourse /175
S -DRS, and vice versa. The result of this operation is that we obtain a partial
S -DRS that contains a variable for the material belonging to the focus, and the
predicative focus S -DRScontainsvariablesfor thematerial belonging to thetopic.
Theproblemwith Kruijf f-Korbayova’sproposalis that, asa result of thewrapping
operation, we obtain two S -DRSsthat areno longer g -reducible: Their types can
no longer beappliedto oneanother.
In DGL’s representationsof linguistic meaningwith topic-focusarticulation, there
is no suchproblem. I usenominals to maintain the original dependency-based
relational structure. This meansthat -in a trivial way- we can obtain the same
effect as g -reduction/application overG
(as in TF-DRT) because we never took
anything apartin thefirst place. Thus,“representation” is no longer a big concern
here– therelevant representationsof sentential linguistic meaning we already get
from the grammarin the caseof DGL. I proposeto employ these representations
immediately asTF-DRSs. Chapter ?? already discussedthe useof hybrid logic
for modelingdiscourserepresentations.RecallthatthereI proposedto conceiveof
nominalsas-essentially- discoursereferents,andof thepropositionsholding at the
statesidentified by thenominals asthediscourseconditions.
The more interesting issues concern the interpretion of theserepresentations
in the context of a larger discourse. To that end,we have to establish the effect
ofG
on interpreting thecontext dependentpartand thecontext affecting partof a
sentential linguistic meaning andits informationstructure. Consider (259) below.
I have added the more elaboratespecification of the verb’s causal and temporal
structure,(259c).
(259) a. Thecatatea SAUSAGE.
b. jlk<m T CB Z'n ACTOR opm;q.rtsNuwvxVzy TNB Zm {|rt}Kuwv~V<r T NB Zn PATIENT opm;��r���u�8��u�<}wV�V
c. jlk<m T CB Z'n ACTOR opm;q�r1s�uwvxV�y TNB Z'm�{�r�jl� W������������ }Kuwv�rMj�� W����c���c���x�x�x�
r�n PAST o��Kq��'��N��'�Vtr TNB Zn PATIENT opm;��r���u�.��u��}wV�V
Following earlierproposalslike(Peregrin,1995) or (Kruijf f-Korbayova,1998),
we consider information structure-discourseinterpretation to be defineddynami-
cally. That is, we first try to update the context with the topic, andonly if that
succeeds,we try to theupdate thecontext with the focus. Now, to beableto pro-
vide this definition, we have to make sure it is clear what a representation like
(259c) includes.
Each representation is considered from a particular “vantage point” - � in
(259c). Fromthis point, everything is related, starting with theevent nucleus. As
176 DGL, topic/focus,anddiscourse
we already saw earlier, e.g. on page??, the internal structureof aneventnucleus
is like (260).
(260) j � W������������|n PREPo��Kq> J� � � ��¡H��¢ �|r�j � W k ��£��p£'¤¥£'\x�"n CONSo¦qp§ ¢ � �~¨~©3��¢ �When we unfold (259c)’s
Dinto (260), it should be clear that the ª<«�¬��®J�¬�¯
nominal is the sameasthe nominal in the event nucleus. Furthermore, from the
specification of the verbal tense, we know that the activity occured in “the” past.
To bemoreprecise, from?±°eA�D�EI?t²>³'´ @¶µp@ ´¸·N��¹5º(EI?|²>³'´ @¶µp@ ´¸·6»¼�½
E¿¾PAST À�ª�«�¬��®<�¬�¯5Á it follows that
?°e¾PAST À�ª�«�¬��®<�¬�¯ , i.e. ª<«�¬��®J�¬�¯ refers to a
point in thepastof � . Furthermore,we of course have thedependents. Theseare
related to � through their dependency relation, andtheir informativity (CB/NB).3
Theway we canview this intuitively is asfollows. Basedon theunderstanding of
the modeling of CB/NB asB�à C modals, � is the vantage point from which the CB
andNB setthecontexts in which thedependents areto beinterpreted.
To recapitulate,whenwe interpreta sentence’s linguistic meaning, we have to
dosoformally from theviewpoint of � . This ideaformsthebasisfor thedefinition
of dynamicdiscourseinterpretation we present below.
Definition 34 (Dynamic discourseinterpretation). We definea discoursestruc-
ture D as a structure¾�Ä#Å~Æ�ÇÉÈ�ÊÅ�˱̥Å�Í À . HB is a hybrid logical back-and-forth
structurewith spatial extension, asdefinedin Definition?? (page??). P is a sorted
structure on which we interpret objects and properties, which mayoverlap with
HB’s sorted spatial structure.Ä
is a setof nominals points in a discourse, andÆ�ÇÎÈ�Êis a setof relationsmodelling modelsin D.
Æ�ÇlÈ�Êincludesat least Ï , there-
lation thatdefinesa total order over thenominals in D. A discoursemodelÐ È is a
tupleAXѱÅ�Ò Á with D a discoursestructureand
Òa hybrid valuation. To interpreta
sentence’s linguistic meaning representedas?�°5AP G(Q Á , Ð ÈiÅ�Ó�Ô ÕÖ?|°eAP G(Q Á
withÓ
thedenotation of � iff Ð ÈiÅ�Ó×Ô Õ�?|°eAP Á .ØDefinition 34 givesthebasics for a dynamicmodelof interpretation, modelingG
in a way similar to (Kruij ff-Korbayova, 1998)(p.79ff) or the dynamic conjunc-
tion discussedin (Muskens et al., 1996). The important stepnow is to definea
notion of discourseaccessibility, which first andformostrelieson how we under-
stand contextual boundness.We already provideda very basic definition of acces-
sibility in Definition ?? on page??. Herewe refineDefinition ?? in the light of3Recall that it furthermoreholds that they arerelatedto the verbalheaddirectly throughtheir
dependency relation.
DGL, topic/focus,anddiscourse /177
Definition34andamoredetailedspecificationof themodel-theoreticsemanticsofBCB C�٠�B NB C�٠.
Definition 35 (Discourse accessibility). We already provideda verybasic defini-
tion of accessibility in Definition??onpage??. HerewerefineDefinition?? in the
light of Definition 34. For Ð ÈÚÅ�Ó×Ô Õ�?|°5AP Á to holdweneedto specify themeaning
of contextual boundness. We first of all have thefollowing standard definitions:
Ð Å�Ó×Ô Õ Ù E×Ûiff Ð Å�ÓÜÔ Õ Ù:ª<ÝiÞ|Ð Å�ÓÜÔ Õ�Û
Ð Å�Ó×Ô ÕÜ¾ß À�Ù iff à ÓÂá'A�ÓÉÇ|â�Óãá�ä Ð Å�ÓÉáÚÔ Õ Ù¦ÁÐ Å�Ó×Ô ÕÜB ß C�Ù iff å Ó á A�ÓÉÇ|â�Ó á`æ Ð Å�Ó á Ô Õ Ù¦Á
For themodalrelation CB wedefine theaccessibility relationsÇ ³�ç
asfollows:ÓÉÇ ³�ç Ó á ä Ð Å�Ó á Ô Õ Ù meansthat there is a stateè inÄ
, èÎÏ Ó , such that at è we
either havethat Ð Å è Ô ÕéBCB C�Ù or Ð Å è Ô ÕFB
NB C�Ù . Theaccessibility relationÇ ê ç
connectsÑ
with˱Ì"Å�Í
, over which it modelsuniversal accessibility. For CB* we
require that?±°5B «xë�ìxC ªîí A�?t°eB «~ë�ìxC ª Eï?|°eB «xë�ìxC;ë EñðO? ² ëxÁ , i.e. if there existsan
accessibleantecedent ª thenthere existsanother accessibleantecedent ë different
from ª .Next, for a dependency relation ò wehavethat theaccessibility relation
Ç5óis
defined fromË�Ì�ô(Í
to˱Ì�ôõÍ
, interpreting¾ ò6À�Ù on a state ö that is of the right
sort given Ù .
Hence, if weconsider thegeneric discourse(modal) relation ÷ to bemodeled
with Ï asits underlying accessibility relation, then wecanspecify discourseacces-
sibility in more detail asfollows.
?�ø5¾�ùûú À�ª�í ?�ø5¾�ùûú À�ª EI?|@XB�ü C ¾ ò6À è Eý?±þ¾ ÷LÀ� Eý?±þJB�ü á C ¾ ò á À�ª ,for any ò Å ò á ÿ��
,üXÅ�ü á�ÿ Æ «xë Å ÝÚë Ê .Ø
Remark 30 (The nature of contextual boundness). The idea that Definitions
34 and 35 describe with so many words is rather simple: A contextually bound
item is an item that wasintroducedearlier in the discourse, and possibly refered
to after its introduction. Consider Figure 6.1. The discourse“progresses” from
left to right, with �:Ït� á Ï�� á á . Under � we introducean Actor ª , asan NB item.
Subsequently, we have a linguistic meaning (under � á ) in which theActor is CB -
referring to theearlier, newly introduced ª , along (1). Finally, we referonce more
178 DGL, topic/focus,anddiscourse
to thecontextually bound item,under � á á - but notethatDefinition 35 enablesusto
just ‘return’ along(2) to thelatestreferenceto ª (which is at ª á under � á ). In other
words,contextual boundnessis a relation over itemsreferedto, or introduced into,
thediscourse.By quantifying over this (transitive) relation we canformally define
thePraguian notion of salience.
Discourse progression
h h’ h"
E
NB
+Actor
Actora
CB
+Actor
a’
NB+Patient
p
E’
NBNB
=
CB+Actor
NB
Actor
E"
a"
= (1)
(2)
Figure6.1: Thenature of contextual boundness
Ø
Definition 36 (Salience). Thesalienceof an item � at a current point in the dis-
course, � , is defined as follows. If � is NB under � , then the salienceof � is 0:
Salience(hBNB C x)=0. If � is CB under � , thenthe salienceof � is defined as fol-
lows.Let Ï�� bethenon-reflexive transitive closure of Ï . Let � ê ç A � Å ��� bethelength
of thepathfrom � to � á , with � á Ï � � and � NB under � á . Let � ³�ç A � Å ��� bethelength
of thepath from � to � á á , with � last CB-referedto under � á á . If there is no h”, then� ³�ç A � Å ��� =0. Thesalienceof a CB item � under � is calculated as in (261), after
(Sgall et al., 1986).
(261) Salience(hTCB Z x)=
� � \�Y m; �����V - � W'Y m� �����V if � is realizedasa pronoun�if � is realizedasa definitenoun
Ø
Remark 31 (The useof salience). Basedon Definition 36 we canestablish the
salience of the items in Figure 6.1 as follows. Assumethat ª á is realized as a
definite nominalhead,and ª á á asa pronoun. Then,thesalienceof ª á is 1, andthe
DGL, topic/focus,anddiscourse /179
salienceof ª á á is ����� =1 aswell. Notethatif wewouldhaveasubsequent reference
to ª , ª á á á , realizedasanotherpronominalexpression, thenthesalienceof ª á á á would
be � . This constitutesa slight differencewith Sgall et al’s proposal,who would
assignª á á á a salienceequalto its previousvalue, i.e. � .Therole salienceplays in thediscoursetheorywe proposehereis thatof pro-
viding anordering over possibleantecedents, from a givenpoint in thediscourse.
If for a CB item under � thereareseveral possible antecedentsto which it could
bereferring, thenby definition we take themostsalientpossible antecedent asthe
antecedentfor thereference.ØFinally, to round off our discussion in this section, we addressthe issue of
how sentential linguisticmeaning getsactually merged with thealreadyestablished
discourse.
Definition 37 (Merging sententialmeaningwith discourse). Givena specifica-
tion of sentential linguistic meaning,?/°5AP G Q � anda discourse � , bothformu-
latedashybrid logical formulas. By definition wehave that Ð ÈõÔ Õ � . Theempty
discourseis modeled as � . Themerge-operator � for merging � and?8°eAP GõQ � ,��� ?|°eAP GõQ � is definedasfollows.
i. If � Õ � , then take a nominal Þ ÿ Ñ, designate Þ as � �"!#! ���8º<A Þ$� , and
interpret?&%AP G Q � (equating Þ and � ,
?'% � ) as per Definition 34. Let� Õ�?(%AP GõQ �ii. If �*)Õ � , thentakea nominal Þ ÿ Ñ for which it holdsthat Þ á ÏÞ , � �+!#! �3�8º<A Þ á � .
Evaluate?,%AP GõQ � asperDefinition34. Let � Õ � E:?'%.-�¾ ÷LÀ�Þ EM?(%<AP G
Q � , andset � �+!/! �3�8º�A Þ0� .Ø
Examplesfollow in thenext section, afterwehavediscussed(anaphoric) bind-
ing in moredetail.
6.3 BINDING ACROSS (CLAUSAL) BOUNDARIES
In the previous sections we discussedin detail how informationstructure(topic-
focusarticulation) is represented in linguistic meaning, andhow it guidestheinter-
pretation of linguistic meaning in a discoursecontext. Particularly, we described
180 DGL, topic/focus,anddiscourse
dynamic,informationstructure-sensitive interpretation,how that interpretation re-
lies on contextual boundnessand salience, and how a discoursecontext can get
extended.In thecurrentsection, we put thesedefinitionsto use.
Ouraimhereis to look in moredetail atthebindingof anaphora. Intrasentential
binding of anaphoraalready hasreceivedafair amountof attention in thecategorial
grammarcommunity andbeyond. Thereis, asOehrle(1999) remarks,already a
well-developed account of bindingphenomena, building onwork by BachandPar-
tee,andChierchia,anddevelopedfurther by people like Szabolcsi, Morrill (1990;
1994), Hepple(1990), Dowty, Jacobson, and -most recently- Jager (to appear).
The commonthreadthrough the various accountsseemsto be to treata pronoun
(or anaphor) asa type that enablesthe pronoun to “travel” through the structure,
copy thesemanticsof its anchor, andthen “travel” backto thepointwhereit should
reside in the structure. Oehrlepoints out various theoretical aswell asempirical
issues(pp.222-223, (Oehrle,1999)) thatcouldberaised against suchanaccount.
Oehrlehimself exploresa different avenue, alongthe lines of dynamic inter-
pretation.4 Thebasicideais to discern two types of bindersin thecontext, namely
a setof discoursebindersÄ
anda setof local binders 1 , andcarry information
from ananchor (a localor adiscoursebinder) in thecontext to theanaphor. Oehrle
achieves this as follows. Firstly, eachelementin the lexicon is marked with a
unary modalthat indicatesits sensitivity to (referential) context, i.e. in whatset(s)
of bindersits antecedent could be looked for. Secondly, different modesof com-
position areusedto control the dynamics(or influence) of contexts in a way that
is similar to theboxesin DRT controlling accessibility. Thirdly, associatedto each
distinguishedunary modalis a rewrite rule thatspecifies how it dependsonÄ
and1 , andhow it affects them. Oncea proof completes,andwe have obtaineda syn-
tactic structureandits correspondingmeaning representation,wecantry to rewrite
the representation we have obtained. Rewriting goesby the rewrite rules for the
unary modals,and results in a representation that hasan input context, filled-in
semantics for antecedents(by meansof “discoursereferents”), andan output (or
updated) context.
Thus,Oehrleprovidesan account of binding that involvesan explicit notion
of context, which in principle enables us to lift the account to the level of dis-
courseand dealwith both intrasentential and intersentential reference. Interest-
4Oehrlepresentsanaccountof bindingphenomenathatcombinescategorial type logic with theideaof dynamicinterpretationasarguedfor by e.g.CrouchandVanGenabithor in this dissertation.Oehrle’s accountdatesbackto 1995. It wasdeveloped independentlyof theaccount in (CrouchandvanGenabith,1998).
DGL, topic/focus,anddiscourse /181
ingly enough, Jager(to appear) discussesa type-logical reformulation of Jacob-
son’s variable-free treatmentof anaphoric reference,andbriefly exploressucha
possible extension to type-logical grammar to cover intersentential anaphoraas
well.
TheapproachI proposehereemploysmechanismsthatwecanapply to modelboth
intrasentential and intersentential binding.5 The basicideabehind thesemecha-
nismswasalreadyintroducedin Chapter ?? - usenominals andjump-operatorsto
modelcontextual reference.For example, recall from page?? that we canmodel
pronominal referenceabstractly asA32ÎE ?54�¾�ù ú À A32 á E76 «98NÝiÞJ�¬��86ÝÚö#�.� : Fromthe
pronoun’s state2
we needto beableto relate to anù ú
accessiblestate2 á
where
«:86ÝiÞJ�¬��8NÝÚö hold.
The differentiation we make hereis in the context wherewe can look for a
nominalto jumpto. Essentially, wefollow uphereontheideato distinguish alocal
anda global context, asemployed for examplein Oehrle(1999). To bring about
this distinctionwemake adifferencebetween discourseaccessibilit y relationsthat
canreach over ÷ (relating different sentences,i.e. global context) andthosethat
cannot (local context). For example, we have specified the accessibility (modal)
relationùûú
asfollows (262).
(262) An antecedent � is accessibleto an anaphor ; if that antecedentis a dependent
occurringunder a headto which ; ’sheadis relatedin theestablishedcontext.?�øe¾�ù ú À�ª�í ?�øe¾�ùûú À�ª EI?�@XB�ü C ¾ ò6À è EI?Âþ¾ ÷LÀ� Eý?±þB�ü á C ¾ ò á À�ª ,for any ò Å ò áiÿ��
,üXÅ�ü áiÿ Æ «xë Å ÝÚë Ê .
If we would specify thesemantics for a reflexive personalpronounusingùûú
asabove (??), thenwe would allow for it to be bound by an antecedent outside
the sentence it appearsin. Clearly, this is not what we want – a reflexive pro-
nounshould be bound locally. Instead ofùûú
we proposeto usea local context
accessibilit y relation � ù ú . � ùûú cannot reachover ÷ .
(263) A dependent < realizedasa reflexivepronoun canbebound by a dependent = of
typeActor, Patientor Addresseethatmodifiesthesamehead� as < does.?±þ<¾ � ù ú À 2 í ?Âþ¾ � ùûú À 2�EI?�@�B�ü C ¾ òKÀ 2�EI?|@XB�ü á C ¾ ò á À?> ,
for any ò ÿ Æ Actor , Patient, Addr esseeÊ, ò á ÿ��
,üXÅ�ü á ÿ Æ «xë Å ÝÚë Ê .
5A disclaimerapplieshere:I do not provide a modelof quantifiersin this dissertation,andthusany interplaybetweenquantifier binding andanaphoric binding is left out of the discussion. Forcategorial grammar-basedproposals for how to treat suchphenomena, seefor example (Morrill,1994),(Moortgat,1997a),(Oehrle,1999),(Jager, to appear).
182 DGL, topic/focus,anddiscourse
The pure formula in (263) statesthat a node2
is locally accessible from >if > and
2aresisters under the samehead,regardlessof their individual contex-
tual boundness,and2
is either an Actor , Patient, or Addr essee(cf. Petkevic
(in prep)(p.98)andreferencestherein to work by Hajicova andby Panevova). If
we give a reflexive pronoun a lexical meaning of the formA32îE ?�4�¾
LXSÀ A32 á E6 «986ÝiÞ�¬��86ÝÚö �.� thenweareableto handle examples like (264).
(264) English
Elijah SHAVES himself.
(265) i.? @ A�B
NB C A�D�E×�/@Ú¹BA.� � EïBCB C ¾ ACTOR À A'¼ EDCFE?G H�¹�@JIDK ª0� ¼ �IML
CB N ¾ PATIENT À9Ocè IQP±ø5¾LXSÀ9Oª IRK ª0� ¼ �.�.�
ii. i. + (263) í P&S O L NB N�OUT IWVX@ ¹BAZY � I[LCB N ¾ ACTOR À9O ¼\I�C(E?G H�¹Z@WI]K ª0� ¼ �IML
CB N ¾ PATIENT À9Ocè IQP±ø5¾LXSÀ9O ¼^IJK ª0� ¼ �.�.�
iii.P_S O L NB N�OUT I`VX@ ¹BAZY � IaL
CB N ¾ ACTOR À9O ¼^I`CFE?G H�¹�@DIJK ªB� ¼ �IMLCB N ¾ PATIENT À9Ocè IQP±ø5¾
LXSÀ9O ¼^I`CFE?G H�¹�@DIJK ª0� ¼ �.�.�However, how about examples like (266)?
(266) English
Elijah told Christopherthatheshaveshimself.
We canmake various observationsabout (266). First of all, theantecedent of
the anaphor bZc canbe locally resolved as eitherElijah or Christopher, or glob-
ally to any suitable antecedent. To achieve that, we have to make it possible fordfeto do so– on (262)only a suitable antecedent in theglobal context would be
found. Secondly, by binding the reflexive pronoun to the Actor , andthe anaphor
to its antecedent, thereflexive pronounshould alsobecomeboundto theanaphor’s
antecedent. Theadditional rule in (267) achievesthepossibility to locally bind an
anaphor. Therule coversthesimplesituationwheretheantecedent is a dependent
of thesameheadthattheembeddedclauseis a dependentof.
(267) If theanaphor occurs in an embeddedclause(a type g -dependent)under a headh, thena dependentsister (of type g�i ) of that dependentunder
hcan serveas
antecedent.P(jlkXSmon�p PFqrLts3u u N k3vXu u m Lts N k3v mow IQP(q�Lts?u N k3vXu mon IQP(j k
XSmonfor any
v�Åxv9u'ÅxvXu u�y�z,sXÅ{s3u'Å{s?u u�y}|X~��NÅ.�"���
.
A morecomplex situation is illustratedin (268).
DGL, topic/focus,anddiscourse /183
(268) English
Elijah liked thesausagethathebought for himself.
Therule in (269)coversthismorecomplex situation. wheretheanaphor occurs
in anembeddedclause(a typez
-dependent)that modifiesanotherdependent (of
typev u
), a dependentsister of whom(typev u u
) canserve asantecedent.
(269) If theanaphoroccurs in anembeddedclause(a type g$i i i -dependent)thatmodifies
another dependent(of type gli ), thena dependentsister(of type gli i ) of that depen-
dentcanserveasantecedent.�(jlkXSmon�p ����k3v�u u u m9� sU��k3v mow�� �(q � s?u���k3vXu m ~ � �(q � s3u u���k3vXu u mon�� �Fjlk
XSmonfor any
v�Åxv9u'Åxv�u u'ÅxvXu u u�y�z,s�Å{s?uÅ{s3u u�y}|X~��NÅ.�"�X�
.
Finally, asaresult of theinferencetrying to establish antecedentsfor contextual
referencesin asentence’s linguisticmeaning weobtainfirst of all links. This is dif-
ferentfrom traditional approaches, wherewe fill in themeaning of theantecedent
ratherthankeeping explicit the relation betweenreference andreferent. By defi-
nition we rule that if we have that a dependent w of typev
refersto an accessible
antecedent n , thentheheadof w alsorelates viav
to n .
(270) If a reference� hasasantecedent� , and � is a dependentof type g modifying a
headh, then� canbeconsideredtobethe g dependentof
h.
�'�rkU� mon�� �(q � s?��k3v m���p �(q � s?��k3v monfor any
v�Åxv u y�z,s\y}|X~���Å.�"�X�
,��y}|X� d}e Šdfe �
Thus, for (264) we obtain the representation as in (271), basedon applying
(270)to (265).
(271)���.� � NB
���U� �D�/�"�B�Z �¡¢�£� CB��k
ACTOR m � c¤�D¥F¦?§ ¨����J�D©ªn � c#¡�M� CB��k
PATIENT m � c^�`¥F¦?§ ¨X���D�J©ªn � c#¡.¡To recapitulate,we distinguish global andlocal contexts by differentiating the
(discourse)accessibility relations thatcan(or cannot) rangeover them. Whenthe
lexical meaningof a (reflexive) pronoun states that it needs to have an accessible
antecedent,thenthis meaning canonly be interpretedif thereis indeed an acces-
sible dependentthat canserve asantecedent. Depending on the kind of pronoun
we aredealing with, the antecedenthasto be in thesamesentence(e.g. reflexive
pronoun case)or maybefound in precedingsentences.Finally, given therulesin
(262) through (270), it is easyto verify that we obtain the usualnondeterminism
for sentenceslike (266). Like (Oehrle, 1999) we obtainthreereadings,(272).
184« DGL, topic/focus,anddiscourse
(272) ¬® ¯�° CB ±?¯�²�³W´¶µ:· ¸�¸�¹º´�° CB ±3» ACTOR ¼½¯¿¾¢´�ÀÁ¸� Ã:ÄlŶ´�ƪÄl¸¿·/¹´Ç° CB ±3» ADDRESSEE ¼x¯�È�´ÊÉ,Å�Ë�¿Ìxµ:Í Î�ÅZ·#ËÏ´�ƪÄl¸�· ¹´Ç° CB ±3» PATIENT ¼½¯�° NB ±3¯Ð²ZÑ¢´�Ì{ÅÒÄ/ÓB·l¹´Ç° CB ±3» ACTOR ¼½¯Ô��´Ê¬ÖÕl» XS¼½¯¿�7´�ÆªÄ ¸�·/¹�¹´Ç° CB ±3» PATIENT ¼½¯Ô�0iÁ´×¬ ÕxØ » XS¼x¯�� i'´�ƪÄl¸�·#¹�¹�¹�¹a. Applying (267), (270), (263),(270):¬ ¯�° CB ±?¯�² ³ ´�µ{· ¸�¸¿¹�´Ç° CB ±?» ACTOR ¼x¯�¾Ù´�À®¸¿Â Ã:ÄlŶ´�Æ�Äl¸�·#¹´Ç° CB ±3» ADDRESSEE ¼x¯�È�´×É,ÅZË�Â�Ì{µ{ÍBÎÒÅÒ·#Ë¢´�ƪÄl¸¿·/¹´Ç° CB ±3» PATIENT ¼x¯�° NB ±3¯Ð²ÒÑÚ´�Ì:ÅZÄ/ÓB·l¹´�° CB ±?» ACTOR ¼x¯�¾Ù´�À®¸¿Â Ã:ÄlŶ´�Æ�Äl¸�·#¹´�° CB ±?» PATIENT ¼x¯�¾Ù´�À®¸¿Â Ã:ÄlŶ´�Æ�Äl¸�·#¹�¹�¹b. Applying (267), (270), (263),(270):¬' ¯�° CB ±?¯�²�³W´�µ{· ¸�¸¿¹�´Ç° CB ±?» ACTOR ¼x¯�¾Ù´�À®¸¿Â Ã:ÄlŶ´�Æ�Äl¸�·#¹´Ç° CB ±3» ADDRESSEE ¼x¯�È�´×É,ÅZË�Â�Ì{µ{ÍBÎÒÅÒ·#Ë¢´�ƪÄl¸¿·/¹´Ç° CB ±3» PATIENT ¼x¯�° NB ±3¯Ð² Ñ ´�Ì:ÅZÄ/ÓB·l¹´�° CB ±?» ACTOR ¼x¯�È�´ÊÉ,ÅZË9¿Ìxµ:Í Î�ÅÒ·/ËÏ´�ƪÄl¸¿·#¹´�° CB ±?» PATIENT ¼x¯�È�´ÊÉ,ÅZË9¿Ìxµ:Í Î�ÅÒ·/ËÏ´�ƪÄl¸¿·#¹�¹�¹c. For somesuitableantecedentÛ that is accessible in theglobal context,
applying (262),(270), (263), (270):¬ ¯�° CB ±?¯�² ³ ´�µ{· ¸�¸¿¹�´Ç° CB ±?» ACTOR ¼x¯�¾Ù´�À®¸¿Â Ã:ÄlŶ´�Æ�Äl¸�·#¹´Ç° CB ±3» ADDRESSEE ¼x¯�È�´×É,ÅZË�Â�Ì{µ{ÍBÎÒÅÒ·#Ë¢´�ƪÄl¸¿·/¹´Ç° CB ±3» PATIENT ¼x¯�° NB ±3¯Ð² Ñ ´�Ì:ÅZÄ/ÓB·l¹´�° CB ±?» ACTOR ¼x¯¿Üf´ÞÝÚ´�ƪÄl¸¿·/¹´�° CB ±?» PATIENT ¼x¯¿Üf´ÞÝÚ´�ƪÄl¸¿·/¹�¹�¹Reflecting on the rules presentedin (262) through (270), we canobserve that
we mostly leave the dependency relations and the informativity of both the an-
tecedent and the (reflexive) pronoun unspecified (except (263)). The criticism
could now beraisedthatthere“thus” would seemlittle usein contextual reference
resolution for a dependency-basedspecification of linguistic meaning.
Ratherthanbeinganunnecessary evil, it –naturally– is a nice feature thatcan
be conveniently usedto deal with intricate phenomena. For example, consider
(273) adaptedfrom (Steedman, 1996)(p.16).
(273) English
a. Thepicturesof herself/her in NewsweekembarrassedKathy.
DGL, topic/focus,anddiscourse /185
b. Thepicturesof *himself/him in NewsweekembarrassedKathy’smother.
Example(273a) illustratesanexemptanaphor. This is a type of anaphor that
canusually besubstituted by anordinary pronoun,makingit differentfrom thenor-
mal bound anaphora. Mostly, exemptanaphorstake astheir antecedents what the
literaturecalls“perceivers”or “experiencers”. Hereweunderstandthemto referto
a Patient. Usinga narroweddown version of rule (263)we cancaptureprecisely
this behavior, (274). If weassign anexemptanaphor a lexical meaning of thekind� Û&� �(ßBkUàÒ� dfe m � Û u �]á ~9â#� wlãåäoã â#�"æ ¡.¡ , then(274) explains theuninterpretability of
(273b).
(274) A dependent ç realizedasan exemptanaphor canbeboundby a dependent Ü of
typePatient thatmodifiesthesameheadè as ç does.�êéBkUàZ� d}e mëÛÚp �êélkUàÒ� dfe mëÛÊ� � � � s?��k PATI ENT mëÛ�� � � � s u ��k3v u m?ì ,
for anyv9u�y�z
,sXÅ{s3u�y}|X~��NÅ.�"���
.
The examples I have presented so far deal with the resolution of (contextual)
referencewithin thescopeof asinglesentence.Theapproacheasily anduniformly
extends to the resolution across sentential boundaries. To round of this chapter, I
discussa simpleexample thatcaptures,in a nutshell, themechanismsdiscussedin
this chapter.
Example (From sentenceto discourseinterpretation). Consider the tiny dis-
course givenin (275).
(275) English
a. Elijah wentTO A STORE.
b. Elijah bought COWBOY BOOTS for himself.
c. He LIKES them.
The grammaranalysesof the sentencesin (275) are trivial. Therefore, we
immediately turn to thelinguistic meaning. First,consider(275a).Wecananalyze
thisasanall-focussentence,whichresults in thelinguisticmeaninggivenin (276).
(276) a.�,q�� � NB
���U� �Jí�î�¡¢�M� NB��k
ACTOR m � c^�`¥F¦?§ ¨����D�Qï¤��¦U 0¡�a� NB��k
DIRECTION:WHERETO m �3æ �`��ðXî�ñ# ò�`ï^�Z¦¿ $¡b. linguistic meaning with topic-focusarticulation:�Fq��åó�ô � NB
���U� �DíZî�¡¢�a� NB��k
ACTOR m � c^�`¥(¦?§ ¨��Z�D�`ï¤��¦U 0¡�a� NB��k
DIRECTION:WHERETO m �3æ �`��ðXî�ñ# ò�`ï^�Z¦¿ $¡
186« DGL, topic/focus,anddiscourse
Merging (276b) with a new discourse õ givesusthediscoursein (277).
(277) õDö �,jl�åó�ô � NB���U� �`í�î�¡Ú�a� NB
��kACTOR m � c^�`¥F¦U§ ¨X���D�Qï^��¦U $¡�a� NB
��kDIRECTION:WHERETO m �3æ �J��ðXîZñ/ ò�Qï¤��¦U 0¡ ,
curr ent(d).
Next, consider thelinguistic meaningfor (275b) given(278). This time,Elijah
is CB, asis theBeneficiary realizedasa reflexive pronoun. Theverbalheadand
thePatient cowboyboots areNB.
(278) a.��qr� � NB
���U� �`÷ùø"ú5¡¢�a� CB��k
ACTOR m � c^�J¥(¦?§ ¨��Z�J�Qï^��¦U B¡�M� NB��k
PATIENT m �3� �Jûlî0ü�÷\î0ú}ýþ÷\î�î�ðX�þ�Jÿ5¦?ø"ñ#��¦3¡�M� CB��k
BENEFICIARY m � ��� ����kLXSm � n��`ï¤��¦U 0¡.¡.¡
b. linguistic meaningwith topic-focus articulation:�(qr� � CB��k
ACTOR m � c\��¥F¦?§ ¨X���W�Úï^��¦U $¡��[� CB��k
BENEFICIARY m � �º� �(�rkLXSm � n_��ï¤��¦U 0¡.¡ô � NB
���U� ��÷ùø"úù¡Ò��� NB��k
PATIENT m �3� �Wûlî0ü�÷ îBú7ýþ÷\î�î�ðX�\�Wÿ5¦?ø"ñ#��¦3¡.¡c. Resolvedlocal contextual reference:�(qr� � CB
��kACTOR m � c\��¥F¦?§ ¨X���W�Úï^��¦U $¡��[� CB
��kBENEFICIARY m � c\��¥(¦?§ ¨��Z�W��ï¤��¦U 0¡ô � NB
���U� ��÷ùø"úù¡Ò��� NB��k
PATIENT m �3� �Wûlî0ü�÷ îBú7ýþ÷\î�î�ðX�\�Wÿ5¦?ø"ñ#��¦3¡.¡Merging (278c) with the existing discourse(277) is straightforward. Thereis
no conflict in updating (277)with thetopic�+q$� � CB
��kACTOR m � c^�`¥(¦?§ ¨��Z�� ï¤��¦U 0¡¤� � CB
��kBENEFICIARY m � c×� ¥F¦?§ ¨���� �*ï^��¦U B¡ . The new discourse
thenbecomesasin (279).
(279) õDö�(jl�åó�ô � NB���U� �`í�î�¡Ù�£� NB
��kACTOR m � c^�J¥F¦?§ ¨X���J�`ï^�Z¦¿ $¡�a� NB
��kDIRECTION:WHERETO m �3æ �J��ðXîZñ/ ò�Qï¤��¦U 0¡ Å� �(jlk � mow u� �Fj���� � CB��k
ACTOR m � c\��¥F¦U§ ¨X���W��ï¤��¦U 0¡Ò�ª� CB��k
BENEFICIARY m � c ��¥(¦?§ ¨��Z�W�Wï¤��¦U 0¡ô � NB���U� ��÷5ø�úÖ¡Á��� NB
��kPATIENT m �3� ��ûlî0ü�÷ îBú}ýþ÷\î�î�ðX� ��ÿ5¦Uø+ñ#��¦3¡.¡
curr ent(d’).
Finally, consider (275c). Only the verb is NB, whereas the two anaphora are
CB, (280).
(280) a.��qr� � NB
���U� �Q¦?§��Z $¡¢�a� CB��k
ACTOR m � ��� �_��kXSm � n��`ï¤��¦U 0¡.¡�M� CB
��kPATIENT m � � u � � � � k
XSm � n u �`ÿ5¦Uø+ñ#��¦3¡.¡.¡
DGL, topic/focus,anddiscourse /187
b. linguistic meaning with topic-focusarticulation:�Fq�� � CB��k
ACTOR m � ��� ����kXSm � n��`ï^��¦U $¡.¡� � CB
��kPATIENT m � � u � � � � k
XSm � n u �Dÿù¦?ø+ñ/�Z¦?¡.¡ ô � NB���U� �R¦?§��� �¡.¡
Updating thediscourseof (279) with the topic of (280b) meanswe have to be
ableto resolve thereferentsfor thetwo anaphora.
(281) i. ��¬�/¯��� ^° NB ±?¯�²ò´�BÍ$¹�´Ç° NB ±3» ACTOR ¼½¯¿¾¢´�ÀÁ¸� Ã{Ä Åò´�ƪÄl¸¿·/¹´�° NB ±3» DIRECTION:WHERETO ¼½¯��¢´�Ì{µ{ÍBË�·7´�ÆªÄ ¸�·/¹��´Ê¬ »��º¼��#i´º¬ Ø ¯�° CB ±3» ACTOR ¼½¯¿¾�´FÀ®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹0´]° CB ±3» BENEFICIARY ¼½¯�¾�´�À®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹ ^° NB ±3¯Ð² ´������Ò¹F´�° NB ±?» PATIENT ¼½¯�� ´��XÍ! "�"Í!�$#%�"Í�ÍBµ{Ì]´¶ÎÒ¸���Ë�Äl¸U¹�¹´Ê¬� Ø »&�º¼��#i i¬ Ø Ø ¯�° CB ±3» ACTOR ¼x¯t��´Ê¬ÖÕl» XS¼½¯¿�7´�Æ�Äl¸�·#¹�¹´�° CB ±?» PATIENT ¼½¯t�0i'´Ê¬\ÕxØ�» XS¼½¯¿� iÁ´òÎ�¸'�ZË9Ä ¸¿¹�¹�¹
ii. Usingrules(262) and(270), resolvefirst theantecedentfor � , andreplacefor
theantecedent’s meaning:
��¬ )( �� ^° NB ± ( ²ò´�BÍ$¹�´Ç° NB ±3» ACTOR ¼ ( ¾¢´�ÀÁ¸� Ã{Ä Åò´�ƪÄl¸¿·/¹´�° NB ±3» DIRECTION:WHERETO ¼ ( �¢´�Ì{µ{ÍBË�·7´�ÆªÄ ¸�·/¹��´Ê¬ »��º¼��#i´º¬� Ø ( ° CB ±3» ACTOR ¼ ( ¾�´FÀ®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹0´]° CB ±3» BENEFICIARY ¼ ( ¾�´�À®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹ ^° NB ± ( ² ´������Ò¹F´�° NB ±?» PATIENT ¼ ( � ´��XÍ! "�"Í!�$#%�"Í�ÍBµ{Ì]´¶ÎÒ¸���Ë�Äl¸U¹�¹´Ê¬ Ø »&�º¼��#i i¬ Ø Ø ( ° CB ±3» ACTOR ¼ ( ¾Ï´�À®¸¿Â Ã:ÄlÅò´�ÆªÄ ¸�· ¹´�° CB ±?» PATIENT ¼ ( �0i'´Ê¬ ÕxØ » XS¼ ( � iÁ´òÎ�¸'�ZË9Ä ¸¿¹�¹�¹
iii. Next, usingagainrules(262) and(270), resolve first the antecedent for �®i ,andreplacefor theantecedent’s meaning:
��¬ )( �� ^° NB ± ( ²ò´�BÍ$¹�´Ç° NB ±3» ACTOR ¼ ( ¾¢´�ÀÁ¸� Ã{Ä Åò´�ƪÄl¸¿·/¹´�° NB ±3» DIRECTION:WHERETO ¼ ( �¢´�Ì{µ{ÍBË�·7´�ÆªÄ ¸�·/¹��´Ê¬�/»��º¼��#i´º¬ Ø ( ° CB ±3» ACTOR ¼ ( ¾�´FÀ®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹0´]° CB ±3» BENEFICIARY ¼ ( ¾�´�À®¸¿Â Ã:ÄlÅ(´FƪÄl¸¿·#¹ ^° NB ± ( ² ´������Ò¹F´�° NB ±?» PATIENT ¼ ( � ´��XÍ! "�"Í!�$#%�"Í�ÍBµ{Ì]´¶ÎÒ¸���Ë�Äl¸U¹�¹´Ê¬� Ø »&�º¼��#i i¬� Ø Ø ( ° CB ±3» ACTOR ¼ ( ¾Ï´�À®¸¿Â Ã:ÄlÅò´�ÆªÄ ¸�· ¹´�° CB ±?» PATIENT ¼ ( �]´*�/Í! "�"Í!�+#,�+Í�Í µ:Ì�´�Î�¸'�ZË9Ä ¸�¹�¹
Thus,we cansuccessfully update thediscoursewith thetopic of (280b). Sub-
188« DGL, topic/focus,anddiscourse
sequently, we cancomplete the discourse with the focus, after which we obtain
(282).
(282) õDö�(jl�åó�ô � NB���U� �`í�î�¡Ù�£� NB
��kACTOR m � c^�J¥F¦?§ ¨X���J�`ï^�Z¦¿ $¡�a� NB
��kDIRECTION:WHERETO m �3æ �J��ðXîZñ/ ò�Qï¤��¦U 0¡.-� �(jlk � mow u� � j � � � CB��k
ACTOR m � c\��¥F¦U§ ¨X���W��ï¤��¦U 0¡Ò�ª� CB��k
BENEFICIARY m � c ��¥(¦?§ ¨��Z�W�Wï¤��¦U 0¡ô � NB���U� ��÷5ø�úÖ¡Á��� NB
��kPATIENT m �3� ��ûlî0ü�÷ îBú}ýþ÷\î�î�ðX� ��ÿ5¦Uø+ñ#��¦3¡.¡� � j ��k � mow u u� j � � � � CB
��kACTOR m � c^�`¥F¦?§ ¨����J�`ï¤��¦U 0¡�Ê� CB
��kPATIENT m �3� � ûlî0ü�÷\î0ú}ýþ÷\î�î�ð/� � ÿ5¦?ø"ñ#��¦3¡ ô � NB
���U� � ¦?§��� r¡.¡curr ent(d”).
This concludestheexample. /
SUMMARY
In this chapterwe presenteda basisfor aninformationstructure-sensitivediscourserepre-
sentationtheory a la (Kruijf f-Korbayova, 1998). Within that proposal,we exploredhow
we cancould give an account for binding anaphora. The approachwe take is different
from moretraditionalproposalslike Morrill (2000) or Jager(to appear), whereresolution
of various kinds of anaphorais performeddirectly in the derivation. Here,alike Oehrle
(1999), we consideranintegrationwith a moredynamicperspective.
The proposalwe advance maintainsa closerelationbetweenthe representationsthe
grammar delivers, andthestructureswe handle in the discoursetheory. We have depen-
dency relations andinformationstructure, bothof which arefundamentallyimportant to
explain variousaspectsof discourseinterpretation: for example, informationstructurefor
coherence,anddependency relations for resolutionof exempt anaphora.
A distinct advantageof our approachis that we areableto relategrammaranddis-
coursewithout having to assumean indexing mechanism that bypassesresolution, like
(VanEijck andKamp,1997) do. At the sametime, we canprovide a compositional ap-
proachto explaining the interpretationof a sentence,bothgrammatically (leading to lin-
guisticmeaning) andasrelatedto discourse(leading to anupdateddiscoursemodel).
CHAPTER 7
AN EMPIRICAL EVALUATION OF SENTENCE
PROCESSING MODELS
An investigationof center embeddingsin Hindi
Data from Hindi center-embedding constructions (CECs)are usedto evaluate threesentence
processing models:Joshi’s EmbeddedPushdown Automaton (EPDA), Gibson’s Syntactic Pre-
diction Locality Theory(SPLT), andLewis’ InterferenceandConfusabilit y Theory(ICT). The
SPLT andICT (but not theEPDA) arefoundto correctly predict several processingfacts about
Hindi. However, the experimental results also reveal a problemfor thesetwo current, wide-
coverage theories: neither modelappears to be ableto account for differences in reading time
observedat nounphrasesin Hindi CECs.A sentenceprocessing modelis proposedin thenext
chapterthatcanin principlebeintegrated with theICT to provideaunifiedaccount of processing
difficulty in thelanguagesinvestigated.
7.1 INTRODUCTION
Several cross-linguistically applicable modelsof sentence processinghave been
proposedover thelastdecadethatattemptto account for processingdifficultiesex-
periencedby humans. Center-embedding constructions(describedbelow in detail)
have beena centerpiece,so to speak, of these models. In this chapter, we discuss
the predictionsof three modelsusingcenter embeddings in Hindi, andshowthat
thesemodelsmake several incorrect predictions regarding the Hindi data. In re-
sponse to this gapbetweenthedataandtheexisting theories,we present a model
of processing(seenext chapter); this model can account for the Hindi data,as
well asthe existing setof dataavailable for Dutch,German,andJapanesecenter
embeddings.
We begin by describing the performanceissues relating to center embeddings
in general. Thenwe present threemodels of sentenceprocessing(developedby
Joshi, GibsonandLewis) andtheir respective predictions for Hindi. Finally, we
evaluate thesemodels usingnew experimentaldatafrom Hindi.
189
190« An empirical evaluation of sentenceprocessingmodels
7.2 WHAT ARE CENTER EMBEDDINGS AND WHY ARE THEY
INTERESTING?
Center-embedding constructions(CECs)involvesentencesin which linguistic ma-
terial is embedded inside another clause. An example is the center embedding
(295a), which has one embedded clause. Chomsky and Miller (Chomsky and
Miller, 1963), amongothers, have observed that double center embeddings like
(295b), which have two embeddedclauses, aremore difficult for English native
speakers to processthan single embeddings (295a) or right-embedded construc-
tions like (295c).
(283) a. Therat [that thecatchased] atethemalt.
b. Therat [that thecat[that thedogchased] killed] atethemalt.
c. Thedogchased thecat[that killed therat [that atethemalt]].
A widely-heldview is that limitationson humanworking memory1 imposestrong
constraints on theprocessingof complex structures like CECs.Theassumption is
that thenoun phrasesmustbetemporarily stored in working memoryuntil verbal
informationclarifies the sentencestructure. Two wide-coveragetheories of sen-
tence processing, Gibson’s Syntactic Prediction Locality Theory (SPLT) (Gibson,
1998; Babyonyshev andGibson, 1999), andLewis’ InterferenceandConfusability
Theory (ICT) (Lewis, 1998), specifically appeal to working memoryconstraints
in explaining the processing of syntactic structures like CECs.Joshi’s Embedded
PushdownAutomaton(EPDA) doesnot appeal to working memoryconstraintsdi-
rectly, but it does rely on thenotion of temporary storageof material. Gibson and
Lewis’ modelsareableto account for many processingfactsin languagessuchas
Dutch(KaanandVasic, 2000), German(Bachet al., 1986), Japanese(Nakatani et
al.,2000), andKorean(UeharaandBradley, 1996), andJoshi’scando thesamefor
a smaller rangeof languages.
Clearly, many other languagesneedto be investigated before theories of sen-
tenceprocessing canclaim truly universal coverage(asthesemodelsaspire to do).
This is the motivation for studying the processing of CECsin Hindi.2 This is a
1Weassumethatworkingmemory, or short-termmemory, is “. . .ashort-durationsystemin whichsmall amountsof informationaresimultaneously storedandmanipulatedin the serviceof accom-plishinga task” (CaplanandWaters,1999).
2Hindi, alsoknown asUrdu,or Hindi-Urdu,is anIndo-Aryan languagespokenprimarily in SouthAsia; it hasabout 424million speakersin India(source:1991Censusof India,www.censusindia.net),andabout10 million in Pakistan(source:www.sil.org).
An empirical evaluation of sentenceprocessingmodels /191
useful languageto investigatesinceit hascertain propertiesnot seenin previously
studied languages. Welook at thesepropertiesnext. Consider first thesingle center
embedding in example(284):
(284) Siitaa-neSita-erg
Hari-koHari-dat
[kitaab(-ko)book(-acc)
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy thebook.’
Here,theergativecasemarker -nemarkstheagent,3 andtheothernounphrases
(NPs)aremarked by theobliquecasemarker -ko, regardlessof theNP’sgrammat-
ical role asindirect or direct object. However, casemarkingon the direct object
(kitaab) is optional: whenpresent, it marksthe NP asspecific, andwhenabsent,
theNPcould bespecificor non-specific(Mohanan, 1994).
For example, in a sequenceof utteranceslike (285), the direct objectkitaab
cannot have casemarking whenit is not salient in the discourse (285a), but can
have it onceit hasbeenmentioned(285b).
(285) a. SiitaaSita
auraand
Hari-neHari-erg
dukaanshop
mein
ekone
kitaabbook
dekhiisaw
‘Sita andHari saw a bookin a shop.’
b. Sita-neSita-erg
Hari-koHari-dat
[kitaab(-ko)book-acc
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy thebook.’
The interesting fact for us is that, in example(284), if -ko casemarking is
present on the direct object (kitaab), the second and third NPswill have phono-
logically thesamesuffix. This is interesting becausepreviousresearchon adjacent
similarly case-markedNPsin JapaneseandKoreanCECsUeharaandBradley (Ue-
haraandBradley, 1996; Lewis andNakayama,1999) have shown thatnominative
casemarkingon adjacent NPsresults in increasedprocessingdifficulty, presum-
ably due to working memoryoverload (this is discussedin detail below). How-
ever, it is an openquestion whethercasemarkings other than nominative affect
processingsimilarly.
Hindi alsohasrather freewordorder in general; thereis only oneconstraint on
the5! orders for thesingle centerembedding in (284): thedirectobject of themost
deeply embeddedverb mustnot appear to the right of this verb,asthe following
exampleshows(unlike analogousEnglishsentenceslike Thecat thedog bit died,3Hindi is a split-ergative language, with an ergative-absolutive casemarkingsystemin the per-
fective aspect.
192« An empirical evaluation of sentenceprocessingmodels
S
NP0ruci-ne
NP1siitaa
�-ko
VP 0S
NP1PRO
�NP2
hariié-ko
VP1S
NP2PRO
éNP3
kitaab(-ko)
VP2V 2
khariid-ne-ko
V 1bol-ne-ko
V 0kahaa
Figure7.1: Example(287)
examples like (286a)areverynatural in Hindi andoccurquitefrequently in a large
text corpus (Vasishth et al., in preparation)).
(286) a. Siitaa-neSita-erg
Hari-koHari-dat
[kitaabbook
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy a/the book.’
b. * Siitaa-neSita-erg
Hari-koHari-dat
[khariid-ne-kobuy-inf
kitaab]book
kahaatold
‘Sita told Hari to buy a/the book.’
This near-absenceof constraints on word orderturnsout to be very useful in
evaluatingtheexisting modelsof sentenceprocessing,aswe shall presently see.
A third property of Hindi center embeddingsis thatthesearecontrol construc-
tions. Thatis, thestructureof adoubleembedding like (287) is asshown in Figure
7.1 (single embeddingshave a similar structure).
(287) Ruci-neRuci-erg
Siitaa-koSita-dat
[Hari-koHari-dat
[kitaab(-ko)book(-acc)
khariid-ne-ko]buy-inf
bolne-ko]tell-inf
kahaatold
‘Ruci told Sitato tell Hari to buy thebook.’
That is, the indirect object of a clause at a given level (matrix or embedded)obli-
gatorily controls a PRO in subject position in theclauseembedded within it. The
syntax of theseconstructionsis discussedin detailelsewhere(Vasishth,2002).
An empirical evaluation of sentenceprocessingmodels /193
Thesethreeproperties(phonologically similar casemarking with dative and
accusative case,relatively free word order, andcenter embeddingsbeingcontrol
constructions)becomerelevant aswe look at Hindi CECsto testthepredictionsof
the EPDA, SPLT, andICT. We will show that the SPLT andICT canonly partly
account for theHindi processing factsandthat theEPDA fails almostcompletely.
Specifically, Gibson’sSPLT canonly partlyaccount for certain reading timediffer-
encesfor NPs. On theother hand, Lewis’ ICT appears to be noncommittal about
NPreading time differences:it assumesthat theprimarysourceof processingdif-
ficulty for CECsoccurs in the retrieval stage,4 asNPsstored in working memory
are retrieved and integratedwith information about the verb. However, findings
from self-pacedreading experimentspresentedin this paper (seeSection4) indi-
catean additional, earlier, moreprominent source of processingdifficulty in the
NPencoding/storage5 stage.On this basis,we arguethatworking-memoryrelated
constraintson parsing areaffectedby bothencoding andretrieval.
Let usnow turn to thethreesentenceprocessingmodels in question.
7.3 THREE MODELS OF SENTENCE PROCESSING
7.3.1 JOSHI ’ S EMBEDDED PUSHDOWN AUTOMATON (1990)
Joshi (Joshi, 1990) presentsa computational model of processingbased on the
results of (Bachet al., 1986); the latter paper showed that Dutch crossed depen-
dencieswereeasierto processfor native Dutch speakersthanGermannested de-
pendenciesarefor nativeGermanspeakers.Examplesof crossed Dutchandnested
Germandependenciesareshownbelow:
(288) a. JanJan
PietPiet
MarieMarie
zagsaw
latenmake
zwemmenswim
‘Jansaw Pietmake Marie swim.’
NP1NP2NP3V1 V2 V3
b. . . .dass. . . that
HansHans
PeterPeter
MarieMarie
schwimmenswim
lassenmake
sahsaw
‘. . . thatHanssaw Petermake Marie swim.’
NP1NP2NP3V3 V2 V14By ‘retrieval’ wemeantheprocessof integrationof NP informationwith a verb.5We usethe term ‘encoding’ to refer to thestagepreceding storageof NPsin working memory
wherebythe NPs are convertedinto somerepresentationalform suitablefor storage. GathercoleandBradley (GathercoleandBaddeley, 1993)presenta discussionrelatingto theworking memoryprocessesassumedhere.
194« An empirical evaluation of sentenceprocessingmodels
The Dutch CECsarecalled “crossed”becauseof the fact that the dependencies
between the verbsandthe subjects form crossing chains(NP1 NP2 NP3 V1 V2
V3), andthe GermanCECsarenested sincethe pattern is NP1 NP2NP3V3 V2
V1.
(Bachet al., 1986) “. . . show that the pushdown automaton (PDA) cannot be
theuniversalbasis for thehumanparsing mechanism” (Joshi, 1990). Theproblem
for the PDA is that in the caseof German,NP3 and the immediately following
V3 cancombine together, but there is no way to tell wherethat structurebelongs
until onegetsto the endof the sentence,andso this structure(and,similarly, the
NP2-V2-(NP3-V3)sub-structure)hasto bestoreduntil ahigherstructurebecomes
available. By contrast, in Dutch, the sub-structures can be built and integrated
incrementally.
Joshi proposesa PRINCIPLE OF PARTIAL INTERPRETATION to overcomethis
problem with PDAs. As heputsit (Joshi, 1990, 4-5):
1. The structure should be a properly integratedstructure with respect to the
predicate-argumentstructure(i.e.,only predicatesandargumentsthatgo to-
gethershould be integrated:ad hocpackagingof predicatesandarguments
is disallowed),andthereshould bea placefor it to go, if it is expectedto fit
into anotherstructure(i.e., thestructureinto which it will fit musthave been
poppedalready).
2. If astructurewhichhasaslotfor receiving anotherstructurehasbeenpopped,
then thestructurethatwill fill this slot will bepoppednext.
Joshi thendevelops an embedded PDA (EPDA) andshows that it canhandle
theDutchandGermanprocessingfacts. Thesignificanceof this is thatEPDAs are
equivalent to the syntactic formalisms TAGs, HPSG,andCCG,all of which are
capable of providing syntactic analysesfor crossedandnesteddependencies.
In the foll owing discussionof Joshi’s model, we assume that the readerhas
a working knowledge of PDAs (see,e.g.,(Hopcroft andUllman, 1979, 107-124)
for details). In an EPDA, the pushdown store is a sequenceof stacks, and new
stacksmaybecreatedaboveor below (to theleft or right) of thecurrent stack. The
specific behavior of EPDAs describedbelowis basedon (Joshi, 1990).
1. Stack head: This is always at the top symbolof the top stack. If the stack
head ever reachesthe bottom of a stack, thenthe stackhead automatically
An empirical evaluation of sentenceprocessingmodels /195
movesto the top of the stack below (or to the left) of the current stack, if
thereis one.
2. Transition functionv u
: Givenaninput symbol,thestateof thefinite control
andthestacksymbol, thisspecifies(a) thenew state;(b) whetherthecurrent
stackis pushedor popped;and(c) new stacksto becreatedabove or below
thecurrent stack.vXu(input symbol, currentstate,stack symbol) =
(new state,æ#� 04- æ#� 1!-4545456- æ/�.7 , push/popon current stack,
æ ä809- æ ä:1!-454545 æ ä<; )where
æ#� 04- æ/� 1=-4545456- æ#�.7 are the stacks introduced below the current stack,
andæ ä�04- æ ä:1=-454545 æ ä<; arethestacks introduced above it.
Notethatduring eachmove,push/popis carried outonthecurrent stack, and
pusheson thenewly createdstack(s).
Next, weillustrateprocessingof theDutchcrossed dependency sequence:NP1
NP2NP3V1 V2 V3 with thefigurebelow showing thevariousstates.Thecolumn
“Stacksequence” contains thenewly created stacks, “Stack” is thestackwe begin
with, andthe column “Pop action” shows how the interpretation is incrementally
built up. Finally, “No. of (input) items” lists a numberthat Joshi usesasa com-
plexity measure to account for the differencein processing Dutch andGerman–
this just involvesadding up the total numberof input itemsin the EPDA at each
move,andlooking at thelargestnumber(in theDutchcase,3).
Input headat Stacksequence Stack Popaction No. of itemsNP1 0NP2 NP1 1NP3 NP1NP2 2V1 NP1NP2NP3 3V1 NP3 NP1NP2 3V1 NP3 NP2 NP1 3V1 NP3 NP2 NP1 3V2 NP3 NP2 V1(NP1,S1) 2V3 NP3 V2(NP2,S2)=S1 1
V3(NP3)=S2 0
Figure7.2: EPDA processingof Dutchdependencies
196« An empirical evaluation of sentenceprocessingmodels
The way this proceedsis as follows. First, NP1 is readin andpushed on to
thecurrentstack,thesamegoesfor NP2andNP3. ThenNP3,NP2,andNP1are
successively poppedout of the current stack andpushed into sequencesof stacks
at the left of the current stack. Then,each NP is popped out of the EPDA and
incrementally builds up the predicate-argument structuresstarting with V1 up to
V3. Thecomplexity never goesbeyond 3.
TheproblematicGermancase(problematicfor PDAs), wheretheorder of NP
andV sequencesis NP1NP2NP3V3 V2 V1 is handledasshown below. In each
case, V > � is a possibly underspecified structureencoding V�
andits argument(s)
(NP�
andpossibly alsoS).Thatis, V > 3 = V1(NP3),V > 2= V2(NP2,S2),andV > 1=
V3(N1,S1).Notethatthemaximumnumber of input itemsin this caseis 6, higher
thanthat in Dutchcrosseddependencies.
Input headat Stacksequence Stack Popaction No. of itemsNP1 0NP2 NP1 1NP3 NP1NP2 2V3 NP1NP2NP3 3V3 V > 3 NP1NP2 4V3 V > 3 V > 2 NP1 5V3 V > 3 V > 2 V > 1 6V2 V > 3 V > 2 V1(NP1,S1) 4V1 V > 3 V2(NP2,S2)=S1 1
V3(NP3)=S2 0
Figure7.3: EPDA processingof Germandependencies
Joshi also discussesthe case of mixed dependencies in German,where the
sequencesare like NP1 NP2 NP3 V1 V3 V2. The complexity measure for this
kind of dependency is claimedto be intermediate between that for crossed and
nesteddependencies(presumablydueto the larger numberof total steps involved
in mixed dependencies). In sucha case, the EPDA behavesexactly like that for
nested dependenciesin Germanuntil we reach V1. Thenit mustbehave like the
EPDA for crosseddependencies.A schematic view is shownbelow:
Onepoint to note hereis that whenV3 is popped out, its argument(NP) is
uninstantiated. This only getsinstantiated when NP3 is popped out in the final
move. Another important point: whenthe input headis at V2, the precedingV3
hasbeeninsertedto theleft of NP2by creating anew stackbehing thestackholding
An empirical evaluation of sentenceprocessingmodels /197
Input headat Stacksequence Stack Popaction No. of itemsNP1 0NP2 NP1 1NP3 NP1NP2 2V1 NP1NP2NP3 3V1 NP3 NP1NP2 3V1 NP3 NP2 NP1 3V1 NP3 NP2 NP1 3V3 NP3 NP2 V1(NP1,S1) 2V2 NP3 V3 NP2 3
NP3 V3 V2(NP2)=S1 2NP3 V3(NP) 1
NP3 0
Figure7.4: EPDA processingof Germanmixeddependencies
NP2,andinserting V2 into this new stack. Thesemovesareallowedby theEPDA
andaccordwith thePPI.
7.3.2 PREDICTIONS OF THE EPDA MODEL FOR HINDI CECS
Joshi’s account raises someinterestingquestionsfor Hindi center-embedding con-
structions.Recalltheissueof specificity markingonthedirectobject, i.e.,minimal
pairslike thefollowing:
(289) a. Siitaa-neSita-erg
Hari-koHari-dat
[kitaabbook
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy a/thebook.’
b. Siitaa-neSita-erg
Hari-koHari-dat
[kitaab-kobook-acc
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy thebook.’
c. Siitaa-neSita-erg
Hari-koHari-dat
[Ravi-koRavi-dat
[kitaabbook
khariid-ne-ko]buy-inf
bol-ne-ko]tell-inf
kahaatold
‘Sita told Hari to tell Ravi to buy a/thebook.’
d. Siitaa-neSita-erg
Hari-koHari-dat
[Ravi-koRavi-dat
[kitaab-kobook-acc
khariid-ne-ko]buy-inf
bol-ne-ko]tell-inf
kahaatold‘Sita told Hari to tell Ravi to buy a/thebook.’
198« An empirical evaluation of sentenceprocessingmodels
Consider now theparsesfor (289a,b):
Input headat Stacksequence Stack Popaction No. of itemsNP1-ne 0NP2-ko NP1-ne 1
NP3-? /-ko NP1-neNP2-ko 2V2 NP1-neNP2-ko NP3 3V1 V > 2 NP1-ne 4
V > 2 V1(NP1-ne,S1) 3V2(NP2-ko,NP3)=S1 0
Figure7.5: Hindi examples (289a,b)
Basedon Table7.5, it is easyto seethat the EPDA predicts the following for
Hindi center embeddings:
@ No difference in processing difficulty at NP3 with respect to specificity-
marking.
@ No differencein processingdifficulty at V2 in bothsentences.
@ Greatest difficulty at final (matrix) verb.
We will presently show thatnone of these preductions areborneout.
7.3.3 CONCLUDING REMARKS REGARDING THE EPDA MODEL
Consider againthe Dutch vs. Germanconstrast. Bachet al. showed that Dutch
crosseddependenciesareeasier to processfor Dutchnative speakers,but German
nested dependenciesareharderfor Germannative speakers. The EPDA models
momentby momentprocessing difficulty (Joshi, personal communication), so it
would predict that thehighestprocessingcostis at the innermostverb in both the
Dutch and Germancases, since in the EPDA the most time is spentthereand
the numberof items present in the EPDA at this point is the largest. However,
experimentalwork hasshownthatthis is not true,at leastnot for Dutch(Kaanand
Vasic, 2000): in Dutch,asin Hindi, themostcostly region seemsto beat thefinal
NP.
Moreover, in the EPDA, structurebuilding doesnot begin until the verbsare
reached; until that point, the NPsaresimply stored in the stack. NPs,however,
generatepredictions(see,e.g.,(Scheeperset al., 19),andreferencescited therein),
An empirical evaluation of sentenceprocessingmodels /199
they arenot just merelystoredin atemporarybuffer (presumablyEPDA is intended
to modelworking memory). Gibson’s SPLT, discussednext, addressesthis issue
of incrementalprocessingandpredictionsat theNPs.
In sum,thereareseveral empirical problemsin theEPDA model: theinability
to predict momentby momentreading timescorrectly for Hindi andDutch(there
arenoreading timestudiesfor GermanCECs,asfarasweknow), andtheassump-
tion of simplestorageof theNPsbefore theverbsareencountered.
7.3.4 GIBSON’ S SYNTACTIC PREDICTION LOCALITY THEORY (1998/1999)
Gibson’s syntactic prediction locality theory (SPLT) (Gibson, 1998; Babyonyshev
andGibson, 1999) hasasomewhatdifferent processing costmetricthantheEPDA.
The SPLT hastwo cost components: INTEGRATION COST and MEMORY COST.
Integration cost is the distancebetween the head-to-be-integrated(e.g., an NP)
and the headto which it connectsin the current structure (e.g.,a verb). This is
quantified in termsof thenumber of discoursereferentsseparating thetwo heads.
Memory cost is the number of all required syntactic headsat a given point. The
memorycost for eachpredictedsyntactic head b increasesas linguistic material
not matching b is processed. The prediction of the top-level predicate (matrix
verb) is assumedto becost-free(sincemostutterancesareheadedby a predicate),
andfor all requiredsyntacticheadsotherthanthetop-level predicate, memorycostA �U� ¡Öö �, where
�is thenumberof new discoursereferentsprocessedsincethat
syntactic headwasiniti ally predicted.6
We illustratethe model’s predictionsby giving a derivation for Hindi double
embeddings.7 In (290), casemarking or the absence thereof on NP3 is indicated
by ? (no casemarking) and-ko. In this discussion, we focuson the memorycost
alonefor easeof exposition; sinceintegrationcostis a function of memorycostin
theSPLT, therelative processingcosts that interest usremainthesame.
(290) NP1-neNP2-ko NP3-ko NP4-? /ko V3-inf V2-inf V1
Here,thepredictedslowestpoint during real-time processingis over NP4,and
no differencebetween thetwo variants (NP4-? versusNP3-ko) is predicted.
6Only finite verbsintroducediscoursereferentsin thismodel(Gibson,personalcommunication).7We follow Babyonyshev andGibson’s derivation for Japanesecenter-embeddings andassume
that the obliquepostpositions/casemarkers for the embeddedverbsarealsopredictedduring realtimeprocessing; however, nothinghingeson this assumption.
200« An empirical evaluation of sentenceprocessingmodels
Prediction NP1-ne NP2-ko NP3-ko NP4-? /ko V3 inf2 V2 inf1 V1V1 0 0 0 0 0 0 0 0 0V2 - M(0) M(1) M(2) M(2) M(2) * - -Inf1 - M(0) M(1) M(2) M(2) M(2) M(2) * -V3 - - M(0) M(1) * - - - -Inf2 - - M(0) M(1) M(1) * - - -
—- —- —- —- —- —- —- —- —-0 0 2 6 5 4 2 0 0
Figure7.6: Processingof (290)
However, wewill presently show thattheslowestreading time is over thefinal
NPonly if it has-ko marking. Thus,thepredictionsof theSPLT appear to beonly
partly correct.
7.3.5 LEWIS’ INTERFERENCE AND CONFUSABILITY THEORY (1998/1999)
This model treats parsing asa short-term memorytask. In the context of center-
embeddingconstructions,thecentral idea is thatretrieval ataverbof anNPduring
real-time processingis affected by two factors: (i) POSITIONAL CONFUSABIL-
ITY; and (ii) RETROACTIVE INTERFERENCE (RI) and PROACTIVE INTERFER-
ENCE (PI).
Positional confusability is the probability of correctly retrieving an NP from
amonga list of NPsseenup to a givenpoint. For example, if NP1NP2NP3NP4
is thelist of NPsseensofar, andif NP3is to beretrieved,theprobability of correct
retrieval will decreaseif NP3andNP4aresimilarly casemarked.This decreasein
probability is dueto theassumption thatitem-recall is with respect to theend-points
(thefirst andlastitem) of a list (independent motivation for this assumption comes
from thepsychology literature, e.g.,(Henson, 1999)). If anend-point NPis similar
to the item being recalled (in our case,‘similar’ meanssimilarly casemarked),
thentheprobability of correct retrieval decreases.Conversely, if theend-point NP
is dissimilarly casemarkedcompared to theNP to beretrieved,theprobability of
correctretrieval increases(i.e, positional confusability is reduced).
Pro-andretroactive interference aredefinedasfollows. Proactive interference
(PI) occurs whentheretrieval of anNP thatsuffers from interference by anNP or
NPspreceding theNPto beretrieved.Retroactive interference(RI) is theopposite:
theretrieval of anNPsuffersfrom interferencefrom itemsthatfollow theNP. There
is a great dealof evidencein the psychology literature for PI andRI in intertrial
An empirical evaluation of sentenceprocessingmodels /201
list recall (see, e.g.,(Muller andPilzecker, ) and(KeppelandUnderwood, 1962)
for someof the earliest findings). It is an assumption of the model that PI and
RI occur within a list of NPs(see(Humphreys andTehan,1998), which provides
independent evidencefor proactive interferencewithin trials).
Wenow illustratethemodel’s behavior.
(291) BC0DB�1E54545DBF; dHG 0 G 1E54545 G 7JIIfI
is thecurrentword(averb),andasyntactic relation needsto beestablished
betweena constituent projected fromI
anda constituent headed by a prior wordd(anoun), thetotal amountof interferenceat
Idependsonthenumber of similar
itemsinterveningbetweend
andI
(RI) andthenumberof similar itemsprecedingd(PI). ‘Similarity’ is understoodto be syntactic similarity, which is determined
by thestructural role to beassignedtod
. For example, ifd
is to beassignedthe
structural position of subject, thenRI occurs due to allG 0 G 1E54545 G 7 which could
alsofill subject positions, andPI occurs dueto all B)0DB�1E54545DBF; which could also
fill subject positions. In addition, positional confusability increasesifd
andG 7
or
BK0 (i.e., oneof theendpoints) is similar tod
. Thetotal amount of retrieval diffi-
culty atI
is thesumof thetwo kindsof interferenceandpositional confusability.
For easeof exposition, we assignsimplenumerical valuesto eachcomponentof
processingcost: e.g.,if therearetwo elements causingRI, thenRI=2, if oneend-
point is increasing positional confusability (POS),thenPOS=1,etc. In theactual
computational implementation, thecostsarenot necessarily simpleintegervalues.
The predictions for Hindi CECs illustratethe model’s operation. The pattern
in (292a) is predictedto beeasier than(292b).
(292) a. NP1-neNP2-ko NP3-ko NP4-? V3 V2 V1
b. NP1-neNP2-ko NP3-ko NP4-ko V3 V2 V1
The following tables illustratehow the modelworks. In eachtable, the first
column lists the item to be retrieved (d
in the template above) at a particular
verbI
, with the itemsG 0 G 1L54545 G 7 interveningbetween
dand
I, andthe items
BK0.B�1E54545DBF; precedingd
. For eachI
theFigurelists thecost of RI andPI,andthe
uparrow( M ) indicatestheitem(s)involvedin causingRI or PI at retrieval.
Here, the retrieval of a subject at a verb results in the other underlying subjects
causing RI or PI.
In the next section we show that Lewis’ model correctly predicts increased
retrieval difficulty at the innermostverb. However, there is another dimension of
202« An empirical evaluation of sentenceprocessingmodels
Retrieveditem NP1-ne NP2-ko NP3-ko NP4-? V3 V2 V1NP3-ko BK0 B�1 d G 0 I
POS=0RI=0
M M PI=2NP2-ko BK0 d G 0 G 1 I
POS=0M RI=1
M PI=1NP1-ne
d G 0 G 1 G 2 IPOS=0
M M RI=2PI=0
Figure7.7: Processing of (292a)
Retrieveditem NP1-ne NP2-ko NP3-ko NP4-ko V3 V2 V1NP3-ko BK0 B�1 d G 0 I
POS=1M RI=1
M M PI=2NP2-ko BK0 d G 0 G 1 I
POS=1M M M RI=2
PI=1NP1-ne
d G 0 G 1 G 2 IPOS=1
M M M RI=3PI=0
Figure7.8: Processingof (292b)
An empirical evaluation of sentenceprocessingmodels /203
processingdifficulty in suchsentences: encoding difficulty of theNPsincreasesif
similarly case-marked NPsareadjacent to eachother. Lewis’ model is agnostic
aboutprocessing difficultiesat NPsandis thusunable to account for this fact.
Weturn now to theexperimentalevidencefrom Hindi.
7.4 CENTER EMBEDDINGS IN HINDI : THREE EXPERIMENTS
Thesecond author of thesenotes (Vasishth) conductedthree experimentto evalu-
atevarious predictions of these three models. Theseexperimentswereconducted
at Jawaharlal NehruUniversity, New Delhi, India during September 2000. There-
searchwasfunded partly through theproject, “EstablishingOhio Stateasa Major
Centerfor LanguageProcessing Research, Ohio StateCenterfor Cognitive Sci-
ence,Departmentof Linguistics, and Department of Computer and Information
Science”andpartly by the Departmentof Linguistics,The Ohio StateUniversity
(OSU),andwasconductedin accordance with thehumansubjectsresearch proto-
col number80B0433specified by theHumanSubjects InstitutionalReview Board,
OSU.8
7.4.1 EXPERIMENT 1
Method and materials
Experiment 1 hada NPOQN factorial design, thetwo factorsbeing level of embedding
(singleor double;compare(293a,b)and(293c,d)),andabsenceor presenceof case
markingon thefinal NP(compare (293a,c)and(293b,d)). In thetestsentences, all
but thefinal NPswereproper names;the final NP wasalways an inanimatecom-
monnoun,suchas‘book’ or ‘letter’. Thiswasapaper questionnaire wheresubjects
wereaskedto rateeachsentenceon a scale from 1 (completely unacceptable)to 7
(completely acceptable).
(293) a. Siitaa-neSita-erg
Hari-koHari-dat
[kitaabbook
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy a/thebook.’
b. Siitaa-neSita-erg
Hari-koHari-dat
[kitaab-kobook-acc
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy thebook.’
8All comparisonspresentedhereafterhave p R .05,unlessotherwisestated.
204S An empirical evaluation of sentenceprocessingmodels
c. Siitaa-neSita-erg
Hari-koHari-dat
[Ravi-koRavi-dat
[kit aabbook
khariid-ne-ko]buy-inf
bol-ne-ko]tell-inf
kahaatold
‘Sita told Hari to tell Ravi to buy a/the book.’
d. Siitaa-neSita-erg
Hari-koHari-dat
[Ravi-koRavi-dat
[kit aab-kobook-acc
khariid-ne-ko]buy-inf
bol-ne-ko]tell-inf
kahaatold‘Sita told Hari to tell Ravi to buy a/the book.’
Four lists were prepared in a counterbalanced, Latin Squaredesign, and 32
fillers wereinsertedbetween16 target sentencesin pseudorandomizedorder. The
fillers consisted of eight examplesof four syntactic structures: relative clauses,
medialgapping constructions,simpledeclaratives,andsentenceswith that-clauses
(all the stimuli andfillers areavailable from the author on request). Fifty-three
native speakersof Hindi participatedin the experiment. Nineteenof these were
Hindi-speaking students at theOhio StateUniversity, andwerepaid 5 US Dollars
eachfor completing the questionnaire; the remaining thirty-four wereundergrad-
uateandgraduatestudentsat Jawaharlal NehruUniversity, New Delhi, India, and
werepaid80 Indian Rupees each(approximately 1.7USDollars).
Predictions
This experiment tested thefollowing predictions:
T Acceptability will decreasewith increasinglevel of embedding. All three
models predict this.
T Lewis’ modelpredicts that direct-object markingwill result in reduced ac-
ceptability, but Gibson’s and Joshi’s models predict that the direct object
marking will have no effect on acceptability.
Results
As Figure7.9shows, theresults indicatethat increasingtheamountof embedding
reducesacceptability ((293c,d)werelessacceptablethan(293a,b)),aspredictedby
Joshi’s,Gibson’s,andLewis’ models. However, casemarking on thefinal NPalso
results in reduced acceptability ((293b), (293d) werelessacceptablethan(293a),
(293c) respectively), whichLewis’ modelpredicts,but Joshi’sandGibson’sdonot.
Thedetails of thestatistical analysisareasfollows: A repeatedmeasuresanalysis
An empirical evaluation of sentenceprocessingmodels /205
1
2
3
4
5
6
7
No Yes
Mean a
ccepta
bili
ty r
ating
U
Case marking on final NP
Embedding
Single embeddingDouble embedding
Figure7.9: Resultsof Experiment 1
of variance (ANOVA) wasdone for subject (F1) anditem (F2) means, with level
of embedding and presence or absence of casemarking on the final NP as the
within-subject factors.Themeanrating for sentenceslike (293a) wassignificantly
higher (mean:6.162) thanthat for sentenceslike (293b) (mean:4.179), F1(1,52)
= 130.969, rating for sentenceslike (293c) wassignificantly higher (mean:3.189)
thanfor sentenceslike (293d) (mean:2.553),F1(1,52)= 13.447,
7.4.2 EXPERIMENT 2
Method and Materia ls
This was a noncumulative self-pacedmoving window reading task (Just et al.,
1982); exactly the samematerials wereusedasfor Experiment 1 (seeexamples
(293)for thefour conditions).
A G3 laptop Macintosh running PsyScope(Cohen et al., 1993) was usedto
present the materials to subjects. Forty-six native speakersof Hindi participated
in theexperiment; no subjects from Experiment1 participated in this experiment.
Thetaskwasto pressthespacekey in orderto seeeach successiveword;eachtime
thekey waspressed, thepreviouswordwould disappear. Readingtime(msec)was
takenasa measureof relative momentaryprocessingdifficulty. A yes/no compre-
hension question waspresentedafter each sentence; these weremeantto ensure
thatsubjectswereattending to thesentences.
206S An empirical evaluation of sentenceprocessingmodels
Predictions
This experiment tested thefollowing predictions:
T Readingtimeattheinnermostverbwouldbeslowerin exampleslike(293a,c)
than in exampleslike (293b,d). This wasbasedon Lewis’ interferencethe-
ory, which statesthat the probability of correct retrieval of the final NP de-
creasesasits positional confusability with anadjacentNP increases.
T Reading time would be slowesteitherat (i) the last NP (SPLT), or (ii) the
innermostverb(EPDA). Prediction (i) is basedon theSPLT, asdiscussedin
Section 7.3.4.Prediction (ii) comesfrom thefact that in theEPDA process-
ing of examples like (293b,d)will proceedasin Germancenterembeddings
(seeFigure7.3),with ahighestcostof 7 at theinnermostverb(sincethere is
onemoreNPthanin theGermanexamplein Figure7.3).
T TheEPDA andSPLT bothpredict thatreading time over thelastNPwill be
unaffectedby whether theNPhascasemarkingor not.
Results
Residual reading time were calculated for eachregion by subtracting from raw
reading timeseachparticipant’s predictedreading time for regions with the same
numbers of characters;this in turn is calculatedfrom a linear regressionequation
acrossall of aparticipant’ssentencesin theexperiment(FerreiraandClifton, 1986;
Trueswellet al., 1994). This was donein order to factor out the effect of word
length on reading time. However, the raw reading timesgave identical results to
theonesdiscussedbelow.
As shown in Figures 7.10 and7.11, the results indicatethat (a) reading time
(RT) increasesat the second of two adjacentsimilarly case-marked NPs; (b) RT
remains slow if two -ko marked NPsarefoll owed by a third -ko marked NP; (c)
RT is fasterif a non-case-markedNP(rather thana case-markedNP) follows a -ko
marked NP; (d) RT at the innermostverb is slower if the last NP is casemarked
thanwhenit isn’t; and(e)theslowest RT is in theregion of thefinal NP, particularly
if it is casemarked.
Thus,thefirst prediction (Lewis’ model),that RTs would beslower at the in-
nermost verb in sentenceswith case-markedfinal NPsthanin sentenceswith non-
An empirical evaluation of sentenceprocessingmodels /207
-200
-100
0
100
200
300
400
500
NP1 NP2 NP3 V2 V1
Mean R
esi
dual R
eadin
g T
ime (
mse
c)
Position
Case Marking
Non-case marked final NPCase-marked final NP
Figure7.10:SingleEmbeddings
-200
-100
0
100
200
300
400
500
600
NP1 NP2 NP3 NP4 V3 V2 V1
Mean R
esi
dual R
eadin
g T
ime (
mse
c)
Position
Case MarkingCase MarkingCase Marking
Non-case marked final NPCase-marked final NP
Figure7.11:DoubleEmbeddings
case-markedfinal NPs,wasborneout.9 Thesecond prediction (Gibson’smodel’s),
that the slowest RT would be at the final NP was partly confirmed,and Joshi’s
model’s prediction that the slowestRT would be at the innermostverb, wasdis-
confirmed.Thethird prediction (Gibson’s andJoshi’s models’), thatRT at thelast
NPwould beunaffectedby casemarking, wasdisconfirmed.
Thus,Lewis’ andGibson’smodelsmakeseveral correct predictions. However,
9It is possiblethat the longerreadingtime at the innermostverb is dueto spillover to the verbregion from processingdifficulty at theNPs.We intendto investigatethis questionfurther in futureresearch.
208S An empirical evaluation of sentenceprocessingmodels
both modelsareunable to capturesomeof theHindi facts:Lewis’ ICT makesno
predictions for the NP reading times,10 and Gibson’s model cannot account for
the different RTs on the final case-marked vs. non-case-marked NPs. Thus, it is
clear that encoding/storing NPsis a componentof processingthat neither model
canaccount for.
We proposeto extend Lewis’ modelso that it canaccount for encoding and
retrieval difficulty; this is discussedin the next chapter. We choose to augment
Lewis’ model rather than Gibson’s because the former makes few assumptions
about the encoding componentof processing andit is straightforward to incorpo-
ratethe ideasproposedin the next chapter, which provide a fairly robust account
of difficulty dueto encoding processes.
We now consider another aspect of Lewis’ model. RecallthatLewis identifies
two sourcesof retrieval difficulty: positional confusability andinterference (Sec-
tion 7.3.5). In experiment 2, therewasno way to distinguish betweenthe two. In
experiment 3 below, we attempt to find evidencefor positional confusability . We
usethe fact that positional confusability predicts that processingwill improve if
similarly case-markedNPsaremadenon-adjacent,by, e.g.,scrambling. We there-
fore testedthis prediction in Experiment 3 by manipulatingadjacency.
7.4.3 EXPERIMENT 3
Method and Materi als
Thiswasanoffline acceptability rating tasksimilar in design to Experiment 1. The
testsentencesweresingle embeddings;onefactorwaspresenceor absenceof case-
markedfinal NPs,andtheother factor wasscrambled(NP2-ko NP1-neNP3(-ko))
or unscrambled(NP1-ne NP2-ko NP3-(ko)) first andsecond NPs(seeexamples
(294a,b)).
Participantsweregiven a paperquestionnaire andasked to rateeachsentence
on a scaleof 1 (completely unacceptable) to 7 (completely acceptable). Sixty-
seven native speakers of Hindi participated; nonehad participated in the earlier
experiments.Therewere16 testitemsand32 fillers.
(294) a. siitaa-neSita-erg
hari-koHari-dat
kitaabbook
khariid-ne-kobuy-inf
kahaatold
‘Sita told Hari to buy a/the book.’10Lewis (personalcommunication) informsmethatthisclaimis incorrect;Lewis’ ICT doesindeed
make predictionsfor NP readingtimes. However, at the time of writing this we do not possessadescriptionof theprecisepredictionsmadeby theICT.
An empirical evaluation of sentenceprocessingmodels /209
b. hari-koHari-dat
siitaa-neSita-erg
kitaabbook
khariid-ne-kobuy-inf
kahaatold
‘Sita told Hari to buy a/thebook.’
c. siitaa-neSita-erg
hari-koHari-dat
kitaab-(ko)book-acc
khariid-ne-kobuy-inf
kahaatold
‘Sita told Hari to buy thebook.’
d. hari-koHari-dat
siitaa-neSita-erg
kitaab-(ko)book-acc
khariid-ne-kobuy-inf
kahaatold
‘It wasHari who Sitatold to buy thebook.’
The conditions (294a) and(294b) wereincludedto establish whether scram-
bled sentencesare in general lessacceptable than unscrambledoneswhen pre-
sentedout of context. It is well-known that scrambled sentences(presentedout
of context) are lessacceptable in languageslike English,German,Finnish,and,
Hungarian, (see(Hyona andHujanen, 1997) for a discussion andreferences).We
would thereforeexpect scrambledsentences(in null contexts) to be involve some
processingcost.Onekey question is whetherpositionalconfusability hasagreater
costcompared to the processingcostof scrambling. If increasingpositional con-
fusability hasahigher relativecostthanscrambling,wewill have evidenceconsis-
tentwith theconfusability theory.
Predictions
Scramblingwas expected to result in reduced acceptability; in addition, adding
casemarkingto the final NP in a scrambledsentencsis predicted by Lewis’ con-
fusability theory to result in a smallerdecreasein acceptability than when case
markingis added to thefinal NP in unscrambled sentences. That is, theunscram-
bledorder NP1-neNP2-ko NP3is predictedto bemoreacceptablethanthescram-
bled order NP2-ko NP1-ne NP3, and the reduction in acceptability whenan NP
sequence like NP1-ne NP2-ko NP3-ko is scrambled to NP2-ko NP1-ne NP3-ko
should be smaller than the casewherea sequence like NP1-ne NP2-ko NP3 is
scrambled to NP2-ko NP1-ne NP3. This is becausethe confusability theory pre-
dicts that in a sequencelike NP2-ko NP1-ne NP3-ko therewill be lessretrieval
difficulty at a verbsincethetwo -ko markedNPsareno longer adjacentandareat
the two endsof the list of NPs(asdiscussedin Section7.3.5,the two ends of the
NP-list aretheindexing positionsfor recalling itemsin a list).
210S An empirical evaluation of sentenceprocessingmodels
1
2
3
4
5
6
7
No Yes
Mean a
ccepta
bili
ty r
ating
U
Case marking on final NP
Word orderWord order
CanonicalScrambled
Figure7.12:Experiment 3 results
Results
Results showedthat itemswith two -ko marked NPswerelessacceptable(repli-
cating findings in Experiment 1). Furthermore, as predicted by Lewis’ model,
scrambling sentenceswith two -ko marked NPsresulted in a smaller decreasein
acceptability thanscrambling sentenceswith only one -ko marked NP; i.e., there
wasaninteractionbetweenthefactors (F1(1,66)= 7.5).
7.4.4 DISCUSSION
Consistent with Lewis’ theory of positional confusability, reducing similarity of
adjacentNPsresulted in asmaller decreasein acceptability. Thus,thedatasuggest
that Lewis’ ICT is completely ableto account for the retrieval-relatedprocessing
facts for Hindi, andthat thetwo key componentsin theICT play a role in account-
ing for thedata.
7.5 CONCLUSION
Weempirically evaluatedthreesentenceprocessingmodelsandshowedthatLewis’
model makes the bestpredictions for the Hindi data. We alsoshowthat Lewis’
model appears to be unable to predict all the processing facts in Hindi. In the
next chapter, weproposeamodelof encodingthatcanbeincorporatedinto Lewis’
sentenceprocessingtheory.
An important point is thatalthoughtheEPDA modelclearly failsfor Hindi cen-
An empirical evaluation of sentenceprocessingmodels /211
terembeddings,this is notsoclearfor Gibson’smodel.Recallthat all thesentences
in all theexperimentswerepresentedoutof context, andsinceweweremanipulat-
ing specificity of theNP, it is possiblethatsubjectswereunable to “accommodate”
thespecific referent. If this wasindeed thesourceof processingdifficulty, thenthe
SPLT’s predictions may turn out to be correct if the sentencesarepresentedwith
appropriatepreceding context. Experimentsarecurrently in progressto determine
whetherthis is thecase. Theinterestedreader is invited to consult (Vasishth, 2002)
for thelatest results.
ACKNOWLEDGMENTS
This research was funded by the Departmentof Linguistics and the Department
of ComputerandInformationScience,OSU.The experimentswereconductedat
Jawaharlal NehruUniversity, New Delhi, India; we thank ProfessorsAyeshaKid-
wai, andR. S. Gupta,andAarti Venkataramanfor logistical help. Without their
cooperation, this research would have beenvery difficult to carry out. We arealso
grateful to ChrisBrew, Michael Dickey, David Dowty, EdwardGibson, Martin Jan-
sche,Aravind Joshi, EdithKaan,KeithJohnson,NealJohnson,BrianJoseph, Ruth
Kempson,Marcus Kracht, Bob Levine, RichardLewis, EdsonMiyamoto, Mine-
haruNakayama,GeraldPenn,ShariSpeer, andHansUszkoreit. Versionsof this
chapter werepresentedat the CUNY 2001sentenceprocessing conferenceat the
Universityof Pennsylvania, at a talk in theDepartment of Computational Linguis-
ticsandPhonetics,TheUniversityof Saarland, Germany, andat the13thEuropean
SummerSchoolfor Logic, Language, andInformation held at Helsinki, Finland;
thanks go to the audiences at thesevenues for their commentsand suggestions.
Any errorsarethesoleresponsibility of Shravan Vasishth.
212S An empirical evaluation of sentenceprocessingmodels
CHAPTER 8
PROCESSING AS ABDUCTION+DEDUCTION:
A SENTENCE PROCESSING MODEL
A sentence processing model is presented, basedon abductive and deductive inference. We
show that the modelmakescorrect predictions for an arrayof datainvolving Dutch, German,
Japanese, and Hindi center-embedding constructions. It hascomparable or better empirical
coverage with respectto several other theories of sentence processing, and canbe integrated
into anexisting wide-coveragemodel,Lewis’ InterferenceandConfusability Theory, to obtain
anintegrated theory of working memoryconstraints on humanlanguageprocessing.
8.1 INTRODUCTION
A well-knownfactabout English(Chomsky andMiller, 1963) is thatcenter-embedded
constructions (CECs)like (295a) are more difficult for humans to processthan
right-embedded constructionslike (295b).
(295) a. The salmon[that the man[that the dog chased] smoked] fell off the
grill .
b. The dog chased the man [that smoked the salmon[that fell off the
grill ]].
SuchCECsoccur in several languages,suchasDutch,German,Japanese, and
Hindi, asthefoll owing examplesdemonstrate.
(296) a. (dat)that
AadAad
JantjeJantje
dethe
leraresteacher
dethe
knickersmarbles
lietlet
helpenhelp
opruimencollect
‘(that) Aad let Jantje help theteacher collect themarbles.’ ((Kaanand
Vasic, 2000))
b. (dass)that
diethe
Mannermen
habenhave
HansHans
diethe
Pferdehorses
futternfeed
lehrenteach
‘(that) the men have taught Hansto feed the horses.’ ((Bach et al.,
1986))
213
214S ProcessingasAbduction+Deduction: A SentenceProcessingModel
c. Keiko-gaKeiko-nom
Tadashi-gaTadashi-nom
Kenji-oKenji-acc
kiraida-tohates-comp
omotteiruthinks
‘K eiko thinks thatTadashihatesKenji.’ ((Uehara andBradley, 1996))
d. Siitaa-neSita-erg
Hari-koHari-dat
[kitaabbook
khariid-ne-ko]buy-inf-acc
kahaasaid
‘Sita told Hari to buy a/the book.’ ((Vasishth,2001))
Several experimental studieshave investigatedDutch,German,Japanese,and
Hindi (see, for example, (Bach et al., 1986), (Kaan and Vasic, 2000), (Lewis,
1998), (BabyonyshevandGibson, 1999), (Ueharaand Bradley, 1996), and (Va-
sishth, 2001)), andasa resultwe now have a body of interesting, empirically de-
terminedfacts aboutrelativedifficultiesin processing these kind of sentences,and
reading time differencesduring real-time processing.
Two theoriesthataddressthequestion of a cross-linguistically robust account
of CEC processing are: Babyonyshev and Gibson’s (Babyonyshev and Gibson,
1999) Syntactic Prediction Locality Theory(SPLT), based on integrationcostand
memorycost; and Lewis’ (Lewis, 1998) Interference and Confusability Theory
(hereafter, ICT), which relies on constraints on working memoryduring compre-
hension.1 Thesetheoriesmake correct predictions for several languages, but are
unable to account for all the processing difficulties in Hindi CECs. This is dis-
cussed in detail in (Vasishth, 2001) (this volume),which showed that (i) although
the ICT canaccount for processing difficulty of verbs, it is unable to account for
differencesin processingnouns;and(ii) theSPLT, whosecomplexity metricrelies
on thenumberof discoursereferents introduced in a sentence,cannot account for
somekey Hindi processing facts.
SinceLewis’ ICT lacks a metric for the processing difficulty of nouns (i.e.,
before the verbsareencountered), but makes the correct predictions for the pro-
cessing of verbs for all thelanguagesunderconsideration, onepossibility is to add
sucha metric to the ICT; this hasthe advantageof maintaining the wide cover-
ageof the ICT andof extending it to account for the Hindi data. We proposethe
abductive-inferencebased modelassuch anaddition to theICT.
The structure of the paperis as follows. Section8.2 outlines the main pro-
posal: an algorithm, a complexity metric, and the relationship betweenthe two;
Section8.3 illustratestheoperationof themodelby giving severalderivations for
theDutch,German,Japanese,andHindi facts; andSection8.4concludesthepaper.1This is by no meansanexhaustive list of theoriesrelatingto sentenceprocessing–we chooseto
discussthesetwo theoriesbecausethey have wide empiricalcoveragefor thequestions we addresshere.
ProcessingasAbduction+Deduction: A SentenceProcessingModel /215
8.2 PROCESSING AS ABDUCTION+DEDUCTION: THE MAIN
PROPOSAL
Theprocessingmodelweproposefor explaining thecomplexity of center-embedded
constructionsis basedon a combinationof abduction anddeduction.
The basic idea is as foll ows. We assume that we have a grammar, V , for a
particular natural language, W . V defineswhat types of functional categories (or
predicate-valency frames)we can encounter in W , and how thesefunctions can
combinewith their arguments.Throughout this paperwe will assumethat V is a
categorial grammar, andthatthebasictypesof functionalcategoriescanbederived
directly from V ’s lexicon.
We canextract thelist of types of functionalcategoriesfrom V ’s lexicon. Dis-
regarding the specific words that eachof thesefunctional categories have been
assignedto in thelexicon, wecanconsiderthis list essentially asproviding uswith
schemaselucidating how words (of particular categories)canbe combined. For
example, the intransitive verbs give us the schema XZY8[]\_^ , the transitive verbs
XZY8[`\$a!b�[]\cNd^ , andsoon. We regardthis list asour collection of hypothesis for-
mers, e . We employ e in thefoll owing way.
Whenwe processa sentence,we do soby starting at thebeginning of thesen-
tence,andproceeding word by word towardsthe endof the sentence.2 In center
embeddings,we encounterNPsbefore we seea verb. TheseNPsarearguments
for oneor moreverbs.TheNPsthat we have encountered at a givenpoint during
real-timeprocessingresult in unconsciousabductive inferencesaboutthepossible
completion of the sentence(i.e., about the kind of schemaor schemasthat will
apply). Themodelrelieson theassumption thata greater number of abductive in-
ferenceswill result in increasedprocessing difficulty dueto anprocessingoverload
on humanworking memory.3
Putdifferently, whenever we encounter a word that is believed to be an argu-
mentof an asyet unseen verb (function), we assumea hypothetical function that
would explain the occurrence of that word asa (projected) argument. For exam-
ple, if weencounteranoun in nominative casebeforewehave encountereda verb,
we hypothesizea verbal category thatwould take a noun in nominative caseasits
2Computersdo notnecessarilyhave to do so- for examplewhenusinghead-cornerparsingalgo-rithms.
3Weassumethatworkingmemory, or short-termmemory, is “. . .ashort-durationsystemin whichsmall amountsof informationaresimultaneouslystoredandmanipulatedin the serviceof accom-plishinga task” (CaplanandWaters,1999).
216S ProcessingasAbduction+Deduction: A SentenceProcessingModel
argument.It is our list e thatprovidesuswith thepossiblehypothesis-candidates,
since e includesall the (basic) types of functions that we can conceivably en-
counter, given V .
Subsequently, whenever the parser encounters a verb, it tries to matcha hy-
pothesized function or functionsagainst theactual functional category of theverb.
If thereis a match,or if theverbal category subsumes(i.e., is moreinclusive than
but not inconsistent with; see(Shieber, 1986, 14-16) for a precise definition) the
hypothesized function, thenwe caninstantiate the hypothesisasthe encountered
verbal category, andcompose theverbwith thenounasits argument.
Abduction, then, is understood hereasthe kind of unconscious andinstanta-
neous reasoning we useto advancea hypothetical function asthebestexplanation
for the occurrence of an argument, acting on the assumption that we are trying
to processa grammatical sentence.Deduction is usedin theCategorial Grammar
sense, asthemeansto subsequently try to composeanactually encounteredfunc-
tion andany available,suitableargument(s). Theaccountof processing complexity
arises from the number of hypothesescurrently active, and how difficult it is to
matchthemagainstthefunctional categoriesof observedwords.
In the next subsections we discuss the notion of abduction in somemorede-
tail; thenwe present thealgorithm andthecomplexity metric,andtherelationship
between them.
8.2.1 ABDUCTION
Thecontemporary understanding of abduction, asa third form of logical reasoning
next to deduction andinduction (cf. (Josephson andJosephson, 1996)), is usually
traced back to its discussionby theAmericanlogician, Charles S.Peirce(Kruijf f,
1995; Kruijf f, 1998b). Peircedefined abductionasthefollowing kind of inference:
A surprising phenomenonO is observed;
but if H wereto bethecase, thenO wouldfollowasa matterof course.
Therefore, there is reason to believethat H.
Thus, whereas (intuitively speaking) deduction derives a consequence from
given axioms, and induction establishesa rule or generalization, abduction pro-
poses anexplanation for a surprising observation.
The surprise is the key, here. Being surprised meansthat either (a) we did
not expectto observe (anything like) f at all, or (b) we did expect to make some
ProcessingasAbduction+Deduction: A SentenceProcessingModel /217
observation, but it wasnot f . On (a), our knowledgeis incomplete,whereason
(b), our knowledgeis in someway incorrect. Either way, we do not at that time
havesufficient knowledgeto createahypothesisexplaining f - if weknewall along
that f would happen,how comewe got surprised?4 Peirce’s claim wasthat only
through abduction could we obtaingenuinely new knowledge(on theassumption
wewould let ourselvesbesurprised, andacknowledgethatfact).
Here,like in AI in general, we take in a substantially weaker (but morework-
able)position. We assumethatwe have at our disposalall hypothesisformersthat
couldbepossibly abduced.For our application, thelist is assumedto bethesmall-
estonegivenagrammar/lexicon - theset e discussedabove. Weassumethat e is
finite andclosed.5 Theseareall reasonable assumptionssince e cannot have any
redundanthypothesisformers(having beencreated from the grammarrules), and
the list of schemasextractedfrom the grammarwill be finite (if this werenot so,
thesetof grammarruleswould have to infinite).
e is createdon the basis of a compilation of the lexicon; in a lexicalist ap-
proach like Categorial Grammar, the lexicon determineshow words can be put
together. Structural rules, like thecombinatorsin Combinatorial Categorial Gram-
mar(CCG)(Steedman, 2000d) or themeaning postulatesin Multi-Modal Logical
Grammar(MMLG) (Moortgat,1997b), only vary theorder in which wordsoccur
grammatically.
Thecompilation of thelexiconis basedonaprocedureproposedfor linear logic
in (Hepple, 1998), andextendedin (Kruijf f, 1999b) to cover a larger rangeof mul-
tiplicative resource logics usedin MMLG. Originally, compilation wasproposed
for the purposesof efficient chartparsing with Lambek-style grammars,in order
to overcomeproblemswith earlier approaches(e.g.,(Hepple, 1992) and(Konig,
1990)). The result of the procedureis a set of first-order functions to represent
categories(i.e., there areno higher-orderformulas).
Oncewe have a compiled version of the lexicon, we abstractaway from indi-
vidual words, andretainthedifferent functional categoriesthataredefined.Taken
together, thesefunctional categories make up e . The list of hypothesis formers
e is assumedto be partially ordered by a simplicity criterion: simpler structures
4For that reason,Peirceadvancedthe ideathat a hypothesisis createdby a “guessinginstinct”becausewe cannotrely on reasoningfrom our knowledge as such. In this context it is perhapsinterestingto notethat Peircewasnot alonein postulatinga fundamentalrole for somethinglike a“guessinginstinct” in logic. Godeltook thesameline–cf. (Parsons,1995), andthebrief comparisonbetweenGodel’s ideasandPeirce’s in (Kruijf f, 1997).
5This is not to confused with thefactthatthesetof sentences is infinite.
218S ProcessingasAbduction+Deduction: A SentenceProcessingModel
appear beforethemorecomplex ones. Examplesof thesimplicity criterion: mon-
oclausalstructuresaresimpler thanbiclausalones,the intransitive-verbbased hy-
pothesisis simpler thantheditransitive verb-basedhypothesis.This assumption is
not arbitrary; it is based on experimentalevidencefrom (Yamashita, 1997), which
showed that (Japanese) subjects prefer to completesentenceswith verbsthat are
simpler (i.e.,verbsthatresult in monoclausalstructures)rather thanmorecomplex
ones. We take this result to indicate that simpler structuresare moreaccessible
thanmorecomplex ones, andmodel this assumption by the partial ordering (the
ordering is partial becauseit is possible thatthereis nowayto specify relativesim-
plicity betweena givenpair of hypotheses). We leave aside the issue of precisely
defining theordering criteriafor themoment.
8.2.2 SOME DEFINITIONS
Next, we definesometermsthatwe usein theproposedalgorithm.
Abducible structur e(s): An ABDUCIBLE STRUCTURE is a hypothe-
sisbasedon the informationavailablesofar; no morehypothesesare
selectedthanarejustified by the informationavailable up to a certain
point (this will bemadeprecisepresently).
New information results in the replacementof previoushypotheses.Abduced
functions Xhg arepart of the abducible structures that are taken from e , andthus
posit thepresenceof a word with a particular syntactic category. For example,in
Japanese,if only a nominative NP (we representthis as [`\ji kml=npo ) hasappeared
so far, XFg�Y8[`\ji kml=npo�^ is a syntactic hypothesisthat says: an intransitive verb X�gwith thenominative NPwill give a sentence.6
NotethatalthoughanominativecasemarkedNPis in principleconsistent with
aninfinite setof possible continuations,our modelallows for theselection of only
thosehypothesesfrom thehypothesisformers e that areminimallyconsistent with
thenominative NP. Wedefineminimal consistency asfollows:
Minimal consistency:
Therearetwo cases: (i) only a list of NPshasbeenseenso far, (ii) a
list of NPsanda verb,or only a verb,hasbeenseensofar.
6Thesubscripton q is merelya notationaldevice usedin the derivationsin Section8.3 for im-proving readability.
ProcessingasAbduction+Deduction: A SentenceProcessingModel /219
(i) If only a list of NPshasbeenseensofar: A list of hypothesesH’ ,
H’ r H, is minimally consistentwith a given list of nouns NPs
iff eachhypothesish s H’ is ableto take theNP’s asarguments
without positing any new, unseen arguments.
(ii) A list of hypothesesH’ , H’ r H, is minimally consistentwith a
verb,or agivenlist of nounsNPsandaverb,if f eachhypothesis
h s H’ is ableto take any of theseenNP’s asarguments (given
the valency-frame of the verb that hasbeenseen); if the verb
requiresany new, unseen argument(s) and/or is an argumentof
anotheras-yet-unseenverb X g , theunseenargument(s) and/or the
function Xhg areposited. Any unseen argumentsthat thefunction
X�g would require arealsoposited.
An example illustrating the first clauseabove of minimal consistency is as
follows. Suppose that, during the course of processing a Japanesesentence,we
have seenonly one nominative NP so far. In that case, a hypothesis satisfying
minimal consistency is Xtg:Y8[`\ji kml=npo�^ , andoneviolating minimal consistency is:
X�guY8[]\ji kml=npo�b�vC^ , where v is a hypothesized,new, unseenNP. By contrast,if, af-
ter we seethefirst nominative NP, we seea second nominative NP, theminimally
consistenthypothesesarenow X g Y8[]\$wdi kml=npo�b�[`\_xhi kml)npo�^ , where X g is a stative
verb,and Xyg&z|{4Y8[]\$wdi kml=npo�b}X�g&zC~dY8[`\_xhi kml)npo�^�^ , i.e.,acenter-embeddedstructure.
Thesecond clauseof minimal consistency canbeexemplifiedasfollows. Sup-
posewe are processinga sentence in Japanese,and we first seea verb V1 like
itta-to, ‘said[past]-complementizer’. Here,thehypothesiswill be X}g:Y�v�b.�,wdY���b��t^�^ ;sinceV1 is necessarily anembeddedverb(due to thepresenceof thecomplemen-
tizer), there is a function X�g (with somesubject NP v ) that takes a clause headed
by V1 asanargument,andV1 takes as-yet-unseenarguments� and � .Thereis somepsychological motivation for theminimalconsistency constraint.
Yamashita(Yamashita, 1997) hasconductedJapanesesentencecompletion tasks
whereshepresentedsubjects with incompletesentencescontaining only a series
of NPswhich they wereasked to complete. Shefound thatsubjects tended to use
verbsthat subcategorizedonly for theNPspresent, but notverbsthatwouldrequire
adding new, unseen NPs.Thefirst author of this paper obtainedasimilar resultfor
Hindi in a pilot study.
Turning next to the issueof processingverbsafter the nounshave beenseen,
the model usesa processof matching the verb to the hypothesized function or
220S ProcessingasAbduction+Deduction: A SentenceProcessingModel
functions, in themanner definedbelow.
Matching: A verbV MATCHES with a function X�g if f V hasa valency
that is identical with that of Xdg . An NP can matchwith a posited
NP argumentif f its casemarking, person, number, gender, andother
information, is consistentwith thatof theposited argument.
With thesedefinitionsin place,we turn next to thealgorithm, basedon which the
complexity metric is defined.
8.2.3 THE ALGORITHM
Theprocessing algorithm worksasfollows.
Init: Setthequeue datastructure � to � , setthescanning pointer to position 0.
Scan: Scanthenext word � g , moving thepointer to thenext position.
Lookup: Lookupthescannedword ��g in thelexicon of V .
Process: This is themainpartof thealgorithm.
if ����� then
check thecategory C of �Zg :if C is a function category C then
�������`� C �else
��� abduceY�e]b C b.�+^end if
else �=������t�
if thecategory of ��g is a function category C then
��� deduceY C b.�+^or (failing that) �������`� C �
else � C is not a function category C ���� deduceY C b.�+^or failing that ��� abduceY�e`b C b.�+^If thelatter stepfails, or wearriveat thelatter step
and ��g is thelastword in thesentence,thenFAIL.
end if
ProcessingasAbduction+Deduction: A SentenceProcessingModel /221
end if
��� deduceY C b.�+^ :
Givenacategory C andastructure � , if C is anargumentthen try to combine
it with a hypothesis or function in � , starting with the outermosthypothe-
sis/function first (FIFO). Else,C is a function, andtry to matchit againsta
hypothesis � in � , suchthat C is either equalto � or subsumesit. Failing
all that, throw anexception stating that theword with category C cannot be
combinedwith anything in thestructure.
��� abduceY�e]b C b.�+^ : Givena list of possible hypothesese , a category C, and
a structure � , find theminimally consistenthypothesis(or hypotheses)� in
e that takes � asan argument,andwhich canbe combinedwith � either
as an argumentof a hypothesis/function in � , or as a function taking the
outermosthypothesis/function in � asits argument.If no suchhypothesis�canbefound, then FAIL. Otherwise, integrateC and � into � andreturnthe
updated structure. The hypotheses abduced in this stepareorderedby the
simplicity criterion.
Processing starts with Init . Subsequently, we cycle through Scan-Lookup-
Process, either until we FAIL or until we arrive to theendof the sentence. If the
structure � contains no unmatched hypotheses,thenthe sentence is grammatical
andcanbeassigned � ; otherwise,thesentenceis consideredungrammaticalon V .
To repeat anearlier examplefrom Japanese:two nominative casemarkedNPs
starting a sentencecould be followed either by a stative predicate (8.2.3a), or a
nesteddependency construction with a single level of embedding (8.2.3b).
(297) a. XF~�Y8[]\$wdi kml=npo�b�[`\_xhi kml)npo�^b. X���Y8[]\$wdi kml=npo�b}X���Y8[`\_xhi kml)npo�^�^
Thesearetwo hypothesesselected from e . No other hypothesesareselectedbe-
causethesearetheonly two thatareminimally consistent, giventheinformationso
far. Thesehypothesesarebasedon thegrammatical possibilities in Japanese,and
sincea single clause sentencehasa simpler structurethana sentencewith anem-
bedded clause, the hypothesesareorderedasshown above. Next, the appearance
of an accusative case marked NP will result in thesehypothesesbeingdiscarded
andthenew hypothesisbeingselected:
(298) X���Y8[`\jwdi kml)npo�b}XC�dY8[]\cxhi kml=npo�b�[]\c�hi �t�9�.o�^�^
222S ProcessingasAbduction+Deduction: A SentenceProcessingModel
Sincethe numberof hypotheseshasfallen from two to one, the model predicts
faster processingat the accusative NP. This prediction is borne out, asdiscussed
further on. We turn next to thecomplexity metric.
The complexity metric
The complexity metric hastwo components: ABDUCTION COST, the costassoci-
atedwith theabductive process, andMISMATCH COST, thecostassociatedwith a
mismatchbetweenanencounteredverbandabducedfunctions.
Abduction cost: This reflectstheincreasingprocessingloadassentencefrag-
mentsappearincrementally. Theabduction costis thesumof thenumberof NPs
seenso far, the numberof functions X=g that areposited, andthe total numberof
distinct hypothesesabducedat a givenpoint. Thesethree sub-components arein-
tendedto reflecttheloadin working memoryof: (a) storing anincreasingnumber
of NPs;(b) positing functions;and(c) storing hypotheses.
Mismatch cost: We assumethat the (queued) hypothesesare unanalyzable
units at first, andthat whena word appears, the hypotheseshave to be examined
asa whole–hence, we assumea left to right depth first search. Every time a verb
fails to matchwith a hypothesizedfunction, thereis a mismatchcostof one.This
assumption also haspsychological motivation: Neal Johnson(personal commu-
nication) hasconducted experimentswherehe found that subjects tend to store
information in working memory(information suchaswords, diagrams,etc.) as
unanalyzable units. We take this to indicate thathypothesesarestoredin working
memoryasunanalyzablewholes.
The numerical valueassociatedwith eachsub-component in the metric is as-
sumedto be 1 and the components are assumedto be additive. This is merely
a convenience, andnothing crucial hingeson this assumption. In a fully imple-
mentedversion of this model, the unit costs associatedwith eachcomponentwill
beassociatedwith precisereading time predictions.
The complexity metric applies in conjunction with the application of the al-
gorithm: at eachstage whenthe algorithm incrementallybuilds/revisesthe list of
possible hypotheses, thecomplexity metric is usedto computetheprocessing cost
at thatpoint.
In thenext section, we provide someillustrationsof theempirical coverage of
this processing model.
ProcessingasAbduction+Deduction: A SentenceProcessingModel /223
8.3 THE EMPIRICAL COVERAGE
8.3.1 JAPANESE
Notethat in the following discussion, theverbsin nestedsentencesarenumbered
in reverseorder of occurrence,i.e., thematrix verb, which appearslast,is V1. The
numbersdo not reflectthe verbs’ valencies; this reversenumbering conventionis
merelyin order to highlight thedifferencefrom Dutch(discussed later).
Gibson’s (1998) data
Gibson(Gibson, 1998) hasshown that(299a) is lessacceptable than(299b).
(299) a. obasan-gaaunt-nom
bebiisitaa-gababysitter-nom
ani-gabrother-nom
imooto-osister-acc
izimeta-toteased-comp.
itta-tosaid-comp.
omotteiruthinks
‘The aunt thinks that the babysitter said that the elderbrother teased
theyoungersister.’
b. bebiisitaa-gababysttr.-nom
ani-gabrother-nom
imooto-osister-acc
izimeta-toteased-comp.
itta-tosaid-comp.
obasan-gaaunt-nom
omotteiruthinks
‘The aunt thinks that the babysitter said that the elderbrother teased
theyoungersister.’
First,consider theapplicationof thealgorithm for (299a):
Step1:
Input Abduction/deduction Abduction Cost mismatchcost
NP1-ga X�{ (NP1) 1+1+1=3 0
Here,givenonly thefirst NP (obasan-ga), a sentencewith anintransitive verb
(IV), denoted by X�{ , is abduced. This contributes3 to our costso far (abduction
cost,composedof thenumber of NPsseensofar (1), plusthenumber of functions
abduced(1), plusthenumberof hypothesesabduced(1); mismatchcostis currently
0).
Step2:
224S ProcessingasAbduction+Deduction: A SentenceProcessingModel
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X�{ (NP1) 1+1+1=3 0
NP2-ga X�~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
Given the second NP (bebisitaa-ga), andgiven that both the NPsseenso far
arenominative casemarked, the abducible structuresare: a stative predicatetak-
ing two nominative arguments( X ~ (NP1,N2)),anda center embeddedconstruction
( X � (N1,X � (N2))). Theabduction costhere is 7: 2 NPs,3 functions,and2 hypothe-
ses.
Step3:
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X { (NP1) 1+1+1=3 0
NP2-ga X�~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
NP3-ga X � (NP1,X � (NP2,NP3)) 3+5+2=10 0
XK (NP1,XK¡ (NP2,XC¢ (NP3))
Wenow havethreenominativeNPs,andsoweeitherhaveanembeddedstative
predicate,asin Xh� (NP1,XC� (NP2,NP3)),or acenterembedding,asin Xt (NP1,XC¡ (NP2,XC¢ (NP3))).
Theabductioncostis now 10.
Step4:
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X|{ (NP1) 1+1+1=3 0
NP2-ga XC~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
NP3-ga X � (NP1,X � (NP2,NP3)) 3+5+2=10 0
XK (NP1,XC¡ (NP2,XC¢ (NP3))
NP4-o X|{¤£ (NP1,X|{�{ (NP2,X|{¤~ (NP3,NP4))) 4+3+1=8 0
X {¤£ (NP1,X {�{ (NP2,X {¤~ (NP3,NP4))) is abducedbecausethefourth NPis marked
with accusative case,andso there mustbe at least oneembedding with a transi-
tive embedded verb. The abduction cost is now 8; i.e., the model predicts that
processing will take lesstime at this fourth NP, comparedto thethird NP.
Step5:
ProcessingasAbduction+Deduction: A SentenceProcessingModel /225
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X�{ (NP1) 1+1+1=3 0
NP2-ga X�~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
NP3-ga X�� (NP1,XK� (NP2,NP3)) 3+5+2=10 0
X (NP1,X ¡ (NP2,X ¢ (NP3))
NP4-o X|{¤£ (NP1,XL{�{ (NP2,X|{¤~ (NP3,NP4))) 4+3+1=8 0
V3 X|{¤£ (NP1,XL{�{ (NP2,V3(N3,NP4))) 4+2+1=7 2
Here, the next word is izimeta-to, ‘teased-complementizer’, and a deduction is
performedin thefoll owing manner:
(i). V3 tries to match X�{¤£ in
X|{¤£ (NP1,XL{�{ (NP2,X|{¤~ (NP3,NP4)))¥ failure.
This matching attempt fails becausetheoutermostfunction X6{¤£ hasa valency
framethatdoesn’t matchtheactual verb’s.
(ii). V3 triesto match X {�{ in
X {¤£ (NP1,X {�{ (NP2,X {¤~ (NP3,NP4)))¥ failure.
Here,again, thefailure occurs dueto thevalency frameof theverbnot match-
ing thatof thenext function.
(iii). V3 triesto match X�{¤~ in
X|{¤£ (NP1,XL{�{ (NP2,X|{¤~ (NP3,NP4)))¥¦X|{¤£ (NP1,XL{�{ (NP2,V3(N3,NP4)))
This succeedsbecausethe valency frameof the verb matchesthat of the next
function. The costnow is the sumof the abduction cost (7) plus the number of
failed matches(2): 9. Notice that the numberof abducedfunctions is now 2, not
3; this is becauseoneof the abducedfunctions hasalready beenresolved by its
matching with V3.
Step6:
226S ProcessingasAbduction+Deduction: A SentenceProcessingModel
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X|{ (NP1) 1+1+1=3 0
NP2-ga XC~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
NP3-ga XC� (NP1,XC� (NP2,NP3)) 3+5+2=10 0
X (NP1,X ¡ (NP2,X ¢ (NP3))
NP4-o X|{¤£ (NP1,X|{�{ (NP2,X|{¤~ (NP3,NP4))) 4+3+1=8 0
V3 X|{¤£ (NP1,X|{�{ (NP2,V3(N3,NP4))) 4+2+1=7 2
V2 X|{¤£ (NP1,V2(NP2,V3(NP3,NP4))) 4+1+1=6 1
Thedeductive processgoes asfollows:
(i). V2 triesto match X {¤£ in
X {¤£ (NP1,X {�{ (NP2,V3(N3,NP4)))¥ failure.
(ii). V2 triesto match X�{�{ in
X|{¤£ (NP1,X|{�{ (NP2,V3(N3,NP4)))¥¦X|{¤£ (NP1,V2(NP2,V3(NP3,NP4)))
V2 fails to match XC{¤£ , but successfully matchesXh{�{ . The cost is now 7 (the
abduction cost, 6, plusthemismatchcost,1).
Step7:
Input Abduction/deduction Abduction Cost MismatchCost
NP1-ga X|{ (NP1) 1+1+1=3 0
NP2-ga XC~ (NP1,NP2) 2+3+2=7 0
XC� (NP1,X�� (NP2))
NP3-ga XC� (NP1,XC� (NP2,NP3)) 3+5+2=10 0
XK (NP1,XC¡ (NP2,XC¢ (NP3))
NP4-o X {¤£ (NP1,X {�{ (NP2,X {¤~ (NP3,NP4))) 4+3+1=8 0
V3 X|{¤£ (NP1,X|{�{ (NP2,V3(N3,NP4))) 4+2+1=7 2
V2 X|{¤£ (NP1,V2(NP2,V3(NP3,NP4))) 4+1+1=6 1
V1 V1(NP1,V2(NP2,V3(NP3,NP4))) 4+0+0=4 0
Thedeductionin this caseis immediate:
ProcessingasAbduction+Deduction: A SentenceProcessingModel /227
V1 tries to match XC{¤£ in
X|{¤£ (NP1,V2(NP2,V3(N3,NP4)))¥ V1(NP1,V2(NP2,V3(NP3,NP4)))
Here,V1 matchesthe outermostabduced function Xd{¤£ immediately, and the
parseis completed. Thecostat this stageis 4.
Thetotal cost(thesumof thecostsat eachstep)givesusthecomplexity of the
sentencerelative to othersentences.So,in this case, thetotal cost is 48.
By contrast, (299b)’s processingyields a lower total costof 38:
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-ga X { (NP1) 1+1+1=3 0
2 NP2-ga X�~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2)) 0
3 NP3-o X�� (NP1,XC� (NP2,NP3)) 3+2+1=6 0
4 V3 X�� (NP1,V3(NP2,NP3)) 3+1+1=5 1
5 V2 X ( v ,V2(NP1,V3(NP2,NP3))) 4+1+1=6 0
6 NP4-ga X� (NP4,V2(NP1,V3(NP2,NP3)) 4+1+1=6 0
7 V1 V1(NP4,V2(NP1,V3(NP2,NP3)) 4+0+1=5 0
Table1: (299b)
Note that in Step5 above, the appearanceof an embedded verb results in an
abduced hypothesisinvolving a matrix verb anda nominal argument. This is be-
causeV2 hasthe complementizer -to, which requiresit to be an embeddedverb;
i.e., thesecond clausein thedefinition of minimal consistency applies.
Nakatani et al. (2000)
(Nakatani etal.,2000) conductedseveral off-lineacceptability rating questionnaire
experimentswith Japanese;their results maybesummarizedasfollows:7
Nakatani et al. found that double embeddings are less acceptable than left
branching structures.Theexamplesbelowillustratetherelevantstructures.
7Note: theEnglishglossesaresometimesdifferentfrom (Nakataniet al., 2000).
228S ProcessingasAbduction+Deduction: A SentenceProcessingModel
(300) a. [obasan-waaunt-top
[bebiisitaa-gababysitter-nom
[imooto-gasister-nom
naita-to]cried-comp.
itta-to]said-comp.
omotteiru]thinks‘The auntthinks thatthebabysittersaidthattheyoungersister cried.’
b. [imooto-gasister-nom
naita-to]cried-comp.
bebiisitaa-gababysitter-nom
itta-to]said-comp.
obasan-waaunt-top
omotteiru]thinks
‘The aunt thinks that the babysitter said that the elderbrother teased
theyoungersister.’
Our modelmakesthecorrect prediction about this setof examples,asthefol-
lowing two derivationsshow.
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-wa X { (NP1) 1+1+1=3 0
2 NP2-ga X�~ (NP1,NP2) 2+3+2=7 0
X � (NP1,X � (NP2))
3 NP3-ga X�� (NP1,XK� (NP2,NP3)) 3+5+2=10 0
XK (NP1,XK¡ (NP2,XC¢ (NP3)))
4 V3-to X� (NP1,XK¡ (NP2,V3(NP3))) 3+2+1=6 2
5 V2-to X� (NP1,V2(NP2,V3(NP3))) 3+1+1=5 1
6 V1 V1(NP1,V2(NP2,V3(NP3))) 3+0+1=4 0
Table2: Doublenesting, total costis 40 for (300a)8
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-ga X { (NP1) 1+1+1=3 0
2 V3-to X�~ (V3(NP1),v ) 2+1+1=4 0
3 NP2-ga X�~ (V3(NP1),NP2) 2+1+1=4 0
4 V2-to X � (V2(V3(NP1),NP2),� ) 3+1+1=5 0
5 NP3-ga X � (V2(V3(NP1),NP2),NP3) 3+1+1=5 0
6 V1 V1(V2(V3(NP1),NP2),NP3) 3+0+1=4 0
Table3: Left branching, total cost is 25 for (300b)
8In exampleslike (300a),wepredicta fall in readingtimeat V3 dueto a hypothesisbeingelimi-nated.Wedo not have any datayet to confirmor disconfirm thisprediction.
ProcessingasAbduction+Deduction: A SentenceProcessingModel /229
Moreover, Nakatani et al. found that in double embeddings intransitive V3’s
aremoreacceptablethantransitive V3’s. Examplesof these structures areshown
below.
(301) a. haha-gamother-nom
titi -gafather-nom
fukigen-nafussy
akatyan-gababy-nom
naita-tocried-comp.
itta-tosaid-comp.
omotteiruthinks
‘My mother thinks thatmy fathersaidthatthefussybabycried.’
b. obasan-gaaunt-nom
syoojiki-nahonest
bebisitaa-gababysitter-nom
ani-gabrother-nom
imooto-osister-acc
izimeta-toteased-comp.
itta-tosaid-comp.
omotteiruthinks
‘My auntthinks that thehonestbabysitter saidthatmy brother teased
my sister.’
The modelmakes the correct prediction since(301)a hascost40 and(301)b
hascost 48. Seeearlier derivations(Table 2 and the full derivation for (299a))
respectively).
Yamashita(1997)
Yamashita(Yamashita, 1997) investigated theeffect of word orderandcasemark-
ing on the processingof Japanese.Oneof her experimentsis a moving window
taskinvolving threeconditions:
Condition A. Canonical order, with 4NPsand2 verbs:
[NP1-nomNP2-dat[NP3-nomNP4-accV2] V1]
Condition B. Samestructureasin Condition A, but scrambled NP3andNP4:
[NP1-nomNP2-dat[NP4-accNP3-nomV2] V1]
Condition C. Samestructureasin Condition A, but scrambledNP1,NP2,NP3
andNP4:
[NP2-datNP1-nom[NP4-accNP3-nomV2] V1]
Theresults for Condition A areinterestingin thecontext of thepresent model;9
considertheexamplebelow.
9In this paper, we do not discusstheeffect of word ordervariationsincethis introducesissuesofpragmaticsthatthemodelcurrentlydoesnot takeinto account. Themodelcan,however, beextendedto incorporateconstraintsfrom pragmatics;essentially, theideawouldbeto includeinformationfromthepragmaticsof anutterancein theabductive process.
230S ProcessingasAbduction+Deduction: A SentenceProcessingModel
(302) [denwa-dephone-on
hansamu-nahandsome
gakusei-gastudent-nom
sensei-niteacher-dat
[tumetaicold
koibito-gagirlf riend-nom
nagailong
tegami-oletter-acc
yabutta-to]tore-comp.
itta]said
‘On the phone,a handsomestudent told the teacher that the cold-hearted
girlf riendhadtorn up theletter.’
Yamashita found that reading timesrosesteadily in suchexamplestill theac-
cusative markedNP, andthenfell at theaccusative NP.
Thepresentmodelpredicts this pattern,asshown below.
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-ga X { (NP1) 1+1+1=3 0
2 NP2-ni X�~ (NP1,NP2) 2+1+1=4 0
3 NP3-ga X � (NP1,NP2,X � (NP3)) 3+4+2=9 0
XC� (NP1, XC� (NP2,NP3))
4 NP4-o XC (NP1,NP2,XK¡ (NP3,NP4)) 4+2+1=7 0
5 V2 X (NP1,NP2,V2(NP3,NP4)) 4+1+1=6 1
6 V1 V1(NP1,NP2,V2(NP3,NP4)) 4+0+0=4 0Table5: (8.3.1)
Beforestep4, the reading time is predicted to rise steadily. At step4, a fall
in reading time is predictedsincethenumberof hypotheses falls from two to one,
andthenumberof functionsis now one.
8.3.2 DUTCH AND GERMAN
Dutch: Kaan et al. (2000)
Turning next to Dutch,KaanandVasic (KaanandVasic, 2000) conductedseveral
moving windowstudiesandfound thefollowing.
Fact 1: Double embeddingsharder than singleembeddings
Examplesof eachtypeareshownbelow:
(303) a. Dethe
leiderleader
heefthas
PaulPaul
SonyaSonya
hetthe
kompascompass
helpenhelp
lerenteach
gebruikenuse
tijdensduring
dethe
bergtochthike
‘The leader helped Paul teachSonya to usethe compass during the
hike.’
ProcessingasAbduction+Deduction: A SentenceProcessingModel /231
b. Metwith
aanwijzingendirections
vanof
dethe
leiderleader
heefthas
PaulPaul
SonyaSonya
hetthe
kompascompass
helpenteach
gebruikenuse
tijdensduring
dethe
bergtochthike
‘With the leader’s directions Paul taught Sonya to use the compass
during thehike.’
Doubleembeddingshavea costof 50:
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1 X { (NP1) 1+1+1=3 0
2 NP2 XC~ (NP1,NP2),X � (NP1,X � (NP2)) 2+3+2=7 0
3 NP3 XC� (NP1,NP2,NP3)) 3+6+3=12 0
XC� (NP1,Xm (NP2,NP3))
XC¡ (NP1,XK¢ (NP2,X|{¤£ (NP3)))
4 NP4 X {�{ (NP1,X {¤~ (NP2,NP3,NP4)) 4+5+2=11 0
X|{ � (NP1,XL{ � (NP2,XL{¤� (NP3,NP4)))
5 V1 V1(NP1,X�{ � (NP2,X|{¤� (NP3,NP4))) 4+2+1=7 0
6 V2 V1(NP1,V2(NP2,X|{¤� (NP3,NP4))) 4+1+1=6 0
7 V3 V1(NP1,V2(NP2,V3(NP3,NP4))) 4+0+0=4 0
Table6: total costis 50 for (303a)
Singleembeddingshave a lower costof 30.
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1 X|{ (NP1) 1+1+1=3 0
2 NP2 X ~ (NP1,NP2),XK� (NP1,X�� (NP2)) 2+3+2=7 0
3 NP3 XC� (NP1,NP2,NP3)) 3+6+3=12 0
XC� (NP1,Xm (NP2,NP3))
XC¡ (NP1,XK¢ (NP2,X|{¤£ (NP3)))
4 V1 V1(NP1,XC (NP2,NP3)) 3+1+1=5 0
5 V2 V1(NP1,V2(NP2,NP3)) 3+0+0=3 0
Table7: total costis 30 for (303b)
KaanandVasic found that RTs increasedwith eachincoming NP, andfell at
the innermostverb, which is what our modelpredicts. In the present model, the
232S ProcessingasAbduction+Deduction: A SentenceProcessingModel
NPreading timesarepredictedto risedueto theincreasein thenumber of abduced
functions, andafall in reading time is predictedat thefirst verbdueto theelimina-
tion of somehypotheses (seederivationsabove to seehow exactly this happens).
Dutch and German: Bach et al. (1986)
Bachet al. (Bachet al., 1986) showedthatDutchcrosseddependencieswereeas-
ier to processfor native DutchspeakersthanGermannesteddependenciesarefor
native Germanspeakers. Examplesof crossed Dutch andnested Germandepen-
denciesareshownbelow:
(304) a. NP1NP2NP3V1 V2 V3
JanJan
PietPiet
MarieMarie
zagsaw
latenmake
zwemmenswim
‘Jansaw Pietmake Marie swim.’
b. NP1NP2NP3V3 V2 V1
. . .dass
. . . thatHansHans
PeterPeter
MarieMarie
schwimmenswim
lassenmake
sahsaw
‘. . . thatHanssaw Petermake Marie swim.’
TheDutchCECsarecalled crossedbecauseof thefact that theverbs andthesub-
jects they link with form crossing chains (NP1 NP2 NP3 V1 V2 V3), and the
GermanCECsarenestedsince thepattern is NP1NP2NP3V3 V2 V1.
Our model predicts that Dutch center embeddings will be more acceptable
since, asshownin Tables6 and7, in Dutch, therewill be no mismatchcost; in
theanalogousGermanexampls,however, therewill beamismatchcostassociated
with eachembeddedverb.
8.3.3 HINDI
Vasishth (Vasishth,2001) conductedaself-pacedreading timestudy andfoundthat
in center embeddings, accusative casemarkingon direct objects in Hindi (which
marksspecificity in the caseof inanimateobjects), makesprocessingharder. Ex-
amplesof single center embeddingsareshownbelow.
(305) a. Siitaa-neSita-erg
Hari-koHari-dat
[kitaabbook
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy a/the book.’
ProcessingasAbduction+Deduction: A SentenceProcessingModel /233
b. Siitaa-neSita-erg
Hari-koHari-dat
[kitaab-kobook-acc
khariid-ne-ko]buy-inf
kahaatold
‘Sita told Hari to buy thebook.’
Themodelpredictsthatin thecaseof (305a),therewill beonly onehypothesis
by thetimethethird NPis processed, whereasin (305b),therewill betwo hypothe-
sesat the third NP. Thesetwo hypothesesarisebecauseof the fact that both the
dativeandaccusativecasemarkingsin Hindi aremarkedby thesuffix/postposition
-ko, andbecauseHindi hasextremely freeword order. Thephonologically similar
casemarking combinedwith thepossibility of reorderingNP2andNP3results in
two possible hypotheses.
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-ne X { (NP1) 1+1+1=3 0
2 NP2-ko X�~ (NP1,NP2) 2+1+1=4 0
3 NP3 X � (NP1,X � (NP2,NP3)) 3+2+1=6 0
4 V2 XC~ (NP1,V2(NP2,NP3)) 3+1+1=5 1
5 V1 V1(NP1,V2(NP2,NP3)) 3+0+0=3 0
Table8: total costis 22 for (305a)
Step Input Abduction/deduction Abduction Cost MismatchCost
1 NP1-ne X�{ (NP1) 1+1+1=3 0
2 NP2-ko X�~ (NP1,NP2) 2+1+1=4 0
3 NP3-ko X � (NP1,X � (NP2,NP3)) 3+4+2=9 0
XC� (NP1,XK� (NP3,NP2))
4 V2 XC~ (NP1,V2(NP2,NP3)) 3+1+1=5 1
5 V1 V1(NP1,V2(NP2,NP3)) 3+0+0=3 0
Table9: total costis 24 for (305b)
Similar predictionshold for double embeddings,but a discussionis omitted.
For details, see(Vasishth, 2001).
8.4 CONCLUDING REMARKS
A hybrid abductive/deductive modelof humanlanguageprocessing is proposed,
basedon existing psycholinguistic results. An important observation is thatmany
of the mechanismsproposedhave correlatesin other theories. For example, the
234S ProcessingasAbduction+Deduction: A SentenceProcessingModel
number of NPsseenup to a givenpoint arecountedaspartof theabduction cost;
thiscorrespondsto thenumber of discoursereferents,whichis acriticalcomponent
in Gibson’smodel.Ourcontributionis to proposeaverygeneral general perceptual
mechanism–abduction– asthekey processthatallowsanincremental parse,given
a particular grammarV for therelevant languageW .
Themodelfaresbetter thanexisting accountsfor thedataconsidered here.For
example, noneof the existing theoriescancurrently account for the fall in read-
ing times at the accusative verb in Japanese, andat the first verb in Dutch; and
Gibson’s model(Gibson, 1998) appears to make incorrect predictions for the ris-
ing reading timesfor verbs (see(KaanandVasic, 2000)for details). However, it
remains to be seenwhetherthe predictions it makesareall borneout. For exam-
ple, the modelpredicts that therewill be a fall in reading time whenthe number
of abducedhypothesesis reduced in working memoryto oneasa result of new
incoming information. This happensto be the correct prediction for Yamashita’s
JapanesedataandKaanandVasic’s Dutch data,but we do not have enough data
yet to determinewhetherthis is prediction is borneout (for example) for (299b).
Further, we currently do not have a preciseaccount for the scrambling facts
(e.g.,thosepresentedin (Yamashita,1997)). Onereasonthatwe hesitate to extend
our modelfor scrambling is thatword ordervariation is almost always correlated
with a particular discoursecontext, andyet studieson scrambling andprocessing
like Yamashita’s (Yamashita, 1997) assume that processing of a scrambled sen-
tencepresentedto subjectsout of theblue(i.e.,without any discoursecontext) can
becomparedwith unscrambledcorrelates. Pilot sentencecompletionstudiesusing
Hindi, conductedby thefirst author, indicatethat subjectsfind scrambledsentences
lessacceptable thanunscambledones(these werepresented without any preced-
ing discoursecontext). Wemustthereforeawait furtherempirical work before any
valid conclusionscanbedrawn about theprocessingof scrambledsentences.
Therearesomefactsthatour modelfails to capture. For example,Nakatani et
al. found thatasingly nested,5 NPstackwasmoreacceptablethandoubly nested,
3-4NPstacks. Therelevant examplesaregivenbelow andthederivationfor (306a)
is shown in Table4.
(306) a. tuma-wawife-nom
kakarityoo-nichief-clerk-dat
uranaisi-gafortune-teller-nom
otto-nihusband-dat
seekoo-osuccess-acc
yakusoku-sita-topromised-comp.
ziman-sitaboasted
‘The wife boastedto thechief clerk thatthefortune-tellerpromisedthe
ProcessingasAbduction+Deduction: A SentenceProcessingModel /235
husbandthathe’d succeed.’
b. haha-gamother-nom
titi -gafather-nom
fukigen-nafussy
akatyan-gababy-nom
naita-tocried-comp.
itta-tosaid-comp.
omotteiruthinks
‘My mother thinks thatmy fathersaidthatthefussybabycried.’
c. obasan-gaaunt-nom
syoojiki-nahonest
bebisitaa-gababysitter-nom
ani-gabrother-nom
imooto-osister-acc
izimeta-toteased-comp.
itta-tosaid-comp.
omotteiruthinks
‘My auntthinks that thehonestbabysitter saidthatmy brother teased
my sister.’
Input Abduction/deduction Abduction Cost MismatchCost
NP1-wa X|{ (NP1) 1+1+1=3 0
NP2-ni X�~ (NP1,NP2) 2+1+1=4 0
NP3-ga X � (NP1,NP2,X � (NP3)) 3+2+1=6 0
NP4-ni X�� (NP1,NP2,XK� (NP3,NP4)) 4+2+1 0
NP5-o X (NP1,NP2,X ¡ (NP3,NP4,NP5)) 5+2+1=8 0
V2-to XC (NP1,NP2,V2(NP3,NP4,NP5)) 5+1+1=7 1
V1 V1(NP1,NP2,V2(NP3,NP4,NP5)) 5+0+0 0Table10: total costis 41 for (306a)
Our modelincorrectly predicts that (306a) will be lessacceptable than(306b)
(which hascost40) (seeTable2), but correctly predicts that it will be moreac-
ceptable than(306c), which hascost48 (seethe first derivation presented in this
paper). However, we consider our modelto beprimarily a theory of theencoding
processesthat occurduring NP processing, andwe proposeto integrate this the-
ory of encoding-via-abductive-inferencewith Lewis’ InterferenceandConfusabil-
ity Theory(ICT) which is a theoryof the integration of encodedNPswith verbs.
Integrating thepresent theory with Lewis’ ICT givesusa complete account of en-
coding andretrieval processesduring sentenceprocessing;this integratedaccount,
we argue, makesmorecorrect predictions thanother current sentence processing
models.See(Vasishth, 2001) (this volume)for details.
Finally, in relation to other, similar accounts, we contendthat our account is
a useful generalization over accounts like the ones basedon pushdown automata
236S ProcessingasAbduction+Deduction: A SentenceProcessingModel
(Joshi, 1990), or incrementalprocessing by predictionof minimumvalency aspro-
posed in work by Scheepers et al. (Scheepers et al., 19). Implicit in all thesetreat-
mentsis theideaof abductive inference.Our proposalforegroundsabduction, and
demonstrates the considerable predictive power suchforegrounding makesavail-
able to us. In this sense,our model is lessa challengeto existing accounts than
a reformulation of these in more general (although very precise) terms. Future
work will consist of building a computational implementation of the integrated
ICT/abductive inferencemodel.
Acknowledgments
Thanks to Rick Lewis andJohnJosephsonfor their detailed feedback, andto
Chris Brew, TakaoGunji, Martin Jansche, Keith Johnson,Neal Johnson,Brian
Joseph, Mineharu Nakayama,Mark Steedman,andShariSpeer, for many useful
comments. At the time of writing, the second author was a visitor at the Insti-
tute of Communicating andCollaborative Systems(ICCS),Division of Informat-
ics, University of Edinburgh, andwould therefore like to acknowledgehereMark
Steedman’s andBonnieWebber’s hospitality.
An earlier version of this paper waspresentedin Japan at theJapaneseCogni-
tiveScienceMeeting,2000,at theKobeShoinGraduateSchool,andattheESSLLI
2000 Linguistic Theory and GrammarImplementation Workshop, Birmingham,
UK; we thankthe audiencesthere for their comments. The authors alone arere-
sponsiblefor any errors.
REFERENCES
Anthony AdesandMark Steedman. 1982. On theorder of words.Linguistics& Philoso-
phy, 7:639–642.
Theodora Alexopoulou. 1999. The Syntaxof Discourse Functions in Greek: A Non-
Configurational Approach. Ph.D.thesis,Departmentof Linguistics,University of Ed-
inburgh,Edinburgh,UnitedKingdom, March.
Maria Babyonyshev andEdward Gibson. 1999. The complexity of nestedstructuresin
Japanese.Language, 75(3):423–450.
E. Bach,C. Brown, andW. Marslen-Wilson. 1986. Crossedandnesteddependenciesin
GermanandDutch: A psycholinguistic study. Language and Cognitive Processes,
1(4):249–262.
JasonBaldridge. 1998. Local scramblingandsyntacticasymmetries in tagalog. Master’s
thesis,Universityof Pennsylvania.
Jason Baldridge. 1999. Strong equivalence of CCG and Set-CCG. Unpublished
manuscript,November.
JasonBaldridge.2000. A categorial account of extraction asymmetries in tagalog andtoba
batak.Unpublishedmanuscript, November.
GuyBarryandMartin Pickering. 1992. Dependency andconstituency in categorial gram-
mar. In Alain Lecomte,editor, Word Order in Categorial Grammar/L’OrdredesMots
dans lesGrammairesCategorielles, pages39–57. Adosa,Clermont-Ferrand, France.
Patrick Blackburn. 2000. Representation, reasoning, andrelational structures:a hybrid
logic manifesto.Journal of theInterestGroupin PureLogic, 8(3):339–365.
D. CaplanandG. S.Waters.1999. Verbalworkingmemory andsentencecomprehension.
Behavioral andBrain Sciences, 22:77–126.
Bob Carpenter. 1995. TheTuring-completenessof multimodal categorial grammar. Un-
publishedmanuscript. Carnegie Mellon University.
William Chafe.1976. Givenness,contrastiveness,definiteness,subjects,topicsandpoint
of view. In C. Li, editor, SubjectandTopic, pages25–55. Academic Press,New York,
NY.
NoamChomsky andGeorge Miller. 1963. Introductionto the formal analysisof natural
languages.In R. DuncanLuce,RobertR. Bush,andEugeneGalanter, editors,Hand-
book of Mathematical Psychology, pages269–321.JohnWiley.
NoamChomsky. 1957. SyntacticStructures. Mouton, TheHague, theNetherlands.
237
238S References
NoamChomsky. 1965. Aspectsof theTheoryof Syntax. TheMIT Press,CambridgeMas-
sachusetts.
SanderChung.1990. Vp’s andverb movementin Chamorro. Natural Language andLin-
guisticTheory, 8:559–619.
J. D. Cohen,B. MacWhinney, M. Flatt, andJ. Provost. 1993. PsyScope:An interactive
graphic systemfor designingandcontrolling experimentsin thepsychology laboratory
usingMacintoshcomputers. Behavior Research Methods, Instruments andComput-
ers, 25:257–271.
William Croft. 1990. Typology and Universals. Cambridge Textbooks in Linguistics.
Cambridge University Press,Cambridge,UnitedKingdom.
RichardCrouchandJosefvanGenabith.1998. Context change, underspecification, and
thestructureof gluelanguagederivations.In Mary Dalrymple,editor, Semantics and
Syntaxin Lexical Functional Grammar, pages117–189. TheMIT Press,Cambridge
Massachusetts.
OstenDahl. 1974. Topic-comment structurerevisited. In OstenDahl, editor, Topic and
Comment,ContextualBoundednessandFocus, number6 in Papersin Text Linguistics.
HelmutBuske,Hamburg, Germany.
Anthony R. Davis. 1996. Linking andthehierarchical lexicon. Ph.D.thesis,Department
of Linguistics,Stanford University, Stanford CA.
Paul Dekker and Herman Hendriks. 1994. Files in focus. In ElisabetEngdahl, editor,
Integrating Information Structure into Constraint-basedandCategorial Approaches,
DYANA-2 DeliverableR1.3.B.,pages41–79.Institutefor Logic, LanguageandCom-
putation(ILLC), Amsterdam,TheNetherlands.
ElisabethEngdahl andEnric Vallduvı. 1994. Informationpackaging andgrammararchi-
tecture:A constraint-basedapproach. In ElisabetEngdahl,editor, Integrating Infor-
mationStructureinto Constraint-basedandCategorial Approaches, DYANA-2 Deliv-
erableR1.3.B.,pages41–79. Institutefor Logic, LanguageandComputation (ILLC),
Amsterdam,TheNetherlands.
F. FerreiraandJr. Clifton, C. 1986. Theindependenceof syntacticprocessing.Journal of
MemoryandLanguage, 25:348–368.
JanFirbas.1992.FunctionalSentencePerspectivein WrittenandSpokenCommunication.
Cambridge University Press,Cambridge,UnitedKingdom.
JohnC.Foster. 1992. BetweenL andLP:A theory of wordorderin categorial grammar. In
Alain Lecomte,editor, WordOrderin Categorial Grammar. Adosa,Clermont-Ferrand.
L.T.F. Gamut. 1991. Logic, Language, and Meaning: Volume 2, Intensional Logic and
Logical Grammar. TheUniversity of ChicagoPress,Chicago, London.
SusanE. Gathercole and Alan D. Baddeley. 1993. Working Memory and Language.
Lawrence Erlbaum Associates,Hove,UK.
GeraldGazdar, Ewan Klein, Geoffrey K. Pollum, andIvan A. Sag. 1985. Generalized
Phrase-Structure Grammar. Basil Blackwell,Oxford, UnitedKingdom.
References /239
EdwardGibson. 1998. Linguistic complexity: locality of syntacticdependencies.Cogni-
tion, 68:1–76.
JosephH. Greenberg. 1966. Someuniversalsof grammarwith particularreferenceto the
order of meaningful elements. In JosephH. Greenberg, editor, Universals of Lan-
guage, pages 73–114.TheMIT Press,Cambridge, Massachusetts,secondeditionedi-
tion.
BarbaraGroszandCandacaL. Sidner. 1986. Attention, intensions,andthe structure of
discourse.Computational Linguistics, 12:175–204.
BarbaraGrosz,Aravind Joshi,andScottWeinstein. 1995. Centering: A framework for
modeling the local coherence of discourse.Computational Linguistics, 21(2):203–
226.
L. Haegeman. 1991. Introduction to GovernmentandBindingTheory. Basil Blackwell,
Oxford.
Eva Hajicova, Barbara H. Partee,andPetrSgall. 1998. Topic-Focus Articulation,Tripar-
tite Structures,andSemantic Context. Studiesin Language andPhilosophy. Kluwer
Academic Publishers,Dordrecht,Boston,London.
Eva Hajicova and Ivana Kruijf f-Korbayova. 1999. On the notion of topic. In Prague
LinguisticCirclePapers, volume3,pages225–236. JohnBenjamins,Amsterdam,The
Netherlands.
EvaHajicova,HanaSkoumalova,andPetrSgall. 1995. An automaticprocedurefor topic-
focus identification.Computational Linguistics, 21(1):81–94, March.
Eva Hajicova. 1993. Issuesof sentencestructure and discourse patterns, volume 2 of
Theoretical andComputational Linguistics. CharlesUniversity Press,Prague,Czech
Republic.
M.A.K. Halliday. 1985. Introduction to FunctionalGrammar. EdwardArnold, London.
JohnA. Hawkins. 1983. Word Order Universals. Academic Press,New York, London,
etc.
Irene R. Heim. 1982. TheSemanticsof DefiniteandIndefinite NounPhrases. Phdthesis,
University of Massachusettsat Amherst.
HermanHendriks andPaul Dekker. 1995. Links without location. In Proceedingsof the
AmsterdamColloquium, pages339–358.
HermanHendriks. 1994. Informationpackaging in a categorial perspective. In Elisabet
Engdahl, editor, Integrating Information Structure into Constraint-basedand Cate-
gorial Approaches, DYANA-2 Deliverable R1.3.B.,pages41–79. Institutefor Logic,
LanguageandComputation(ILLC), Amsterdam,TheNetherlands.
HermanHendriks. 1996. Intonation,derivation, information.A proof-theoreticframework
for the representation of information packaging. In V.M. Abrusci and C. Casadio,
editors, ProofsandLinguisticCategories: Applicationsof Logic to theAnalysisand
Implementationof Natural Language(Roma1996Workshop).
HermanHendriks.1997. Thelogic of tune. In Geert-JanM. Kruijf f, Glyn V. Morrill, and
240§ References
RichardT. Oehrle, editors, Formal Grammar1997, Aix enProvence,France,August
9-10.
HermanHendriks. 1999. Contour andstructure. A categorial account of alignment. In
GosseBouma,ErhardHinrichs,Geert-JanM. Kruijf f, andRichardT. Oehrle,editors,
Constraints and Resourcesin Natural Language Syntaxand Semantics, pages155–
174.CSLI Press,Stanford CA.
R. N. A. Henson. 1999. CodingPositionin Short-Term Memory. International Journal of
Psychology, 34(5/6):403–409.
Mark Hepple.1990.TheGrammarandProcessingof Order andDependency:A Catego-
rial Approach. Ph.D.thesis,University of Edinburgh.
Mark Hepple.1992. Chartparsinglambekgrammars:Modalextensionsandincremental-
ity. In Proc.COLING’92.
MarkHepple.1994. A general framework for hybridsubstructural categorial logics. Tech-
nicalReport94-14, IRCS,Universityof Pennsylvania, PhiladelphiaPA, September.
Mark Hepple. 1995. Mixing modesof linguistic descriptionin categorial grammar. In
Proceedings EACL-7,Dublin Ireland.
Mark Hepple.1996. Headpromotionandobliquenessin a multimodal grammar. In V.M.
Abrusci andC. Casadio,editors,Proofsand Linguistic Categories: Applications of
Logic to the Analysis and Implementation of Natural Language (Roma1996Work-
shop).
Mark Hepple. 1997. A dependency-basedapproachto boundedandunbounded move-
ment. In Proceedings of the Fifth Meetingon Mathematics of Language (MOL-5).
DFKI-D-97-02.
Mark Hepple.1998. Memoisation for gluelanguagededuction andcategorial parsing. In
Proceedings of theCOLING-ACL’98 Joint Conference, Montreal Canada,August.
SusanC.HerringandJohnC.Paolillo. 1995. Focuspositionin SOV languages.In Pamela
Downing andMichaelNoonan,editors,Word Order in Discourse, volume30 of Ty-
pological Studiesin Language, pages163–198. JohnBenjamins,Amsterdam,The
Netherlands.
CarolineHeycock. 1993. Focusprojection in Japanese.In Merceı Gonzalez,editor, Pro-
ceedings of NELS, volume24,pages159–187.
Dirk Heylen. 1997. Generalisation and coordination in categorial grammar. In Geert-
JanM. Kruijf f, Glyn V. Morrill, andRichardT. Oehrle,editors,Formal Grammar
1997, Proceedings of the Conferenceheld in Aix en Provence, France, August 9-10,
1997.
Dirk Heylen. 1999. TypesandSorts:ResourceLogic for Feature Checking. Ph.D.thesis,
OTS,Universityof Utrecht,TheNetherlands.
Beryl Hoffman. 1995a. Computational Analysisof the Syntax and Interpretation of
“F ree” Word-Order in Turkish. Ph.d.thesis,ircs report 95-17, University of Penn-
sylvania.
References /241
Beryl Hoffman. 1995b. Integrating”free” wordordersyntaxandinformationstructure. In
ProceedingsEACL’95, Dublin, March.
JohnE. Hopcroft andJeffrey D. Ullman. 1979. Introduction to AutomataTheory, Lan-
guagesandComputation. Addison-Wesley, Reading,MA.
M. S. Humphreys andG. Tehan. 1998. Cuesandcodesin working memory tasks. In
C. Izawa, editor, On memory:Evolution, Progress,and Reflections on the 20th An-
niversaryof theAtkinson-Shiffrin model, pages127–151. Erlbaum, Mahwah,NJ.
Jukka Hyona andHeli Hujanen.1997. Effects of CaseMarking andWord Orderon Sen-
tenceParsingin Finnish:An Eye-Fixation Study. QuarterlyJournal of Experimental
Psychology, 50A(4):841–858.
RayJackendoff. 1972. Semantic Interpretationin Generative Grammar. TheMIT Press,
CambridgeMassachusetts.
Ray Jackendoff. 1977. X’-Syntax: A Studyof PhraseStructure, volume 2 of Linguistic
Inquiry Monographs. TheMIT Press,Cambridge Massachusetts.
Pauline Jacobson. 1990. Raising as function composition. Linguistics& Philosophy,
13(4):423–475.
Gerhard Jager. to appear. Anaphora and quantification in categorial grammar. In
Michael Moortgat, editor, Proceedings of LACL 1998, LectureNotesin Computer
Science.Springer Verlag. Available from http://www.let.uu.nl/ Ger-
hard.Jaeger/personal/.
TheoM.V. Janssen.1997. Compositionality. In JohanvanBenthemandAlice terMeulen,
editors, Handbookof Logic and Language. Elsevier ScienceB.V., AmsterdamNew
York etc.
JohnR. JosephsonandSusanG. Josephson. 1996. Abductive Inference: Computation,
Philosophy, Technology. Cambridge University Press,Cambridge,UK.
Aravind K. Joshi.1990. Processingcrossedandnesteddependencies:An automaton per-
spective on thepsycholinguistic results.Language andCognitiveProcesses, 5(1):1–
27.
M. A Just,P. A. Carpenter, andJ.D. Woolley. 1982. Paradigmsandprocessesin reading
comprehension. Journal of Experimental Psychology: General, 111(2):228–238.
Edith KaanandNadaVasic. 2000. Cross-serialdependenciesin Dutch: Evidence for a
discourse-basedmetric? Posterpresentedat CUNY SentenceProcessingConference,
University of California, SanDiego.
G. Keppeland B. J. Underwood. 1962. Proactive inhibition in short-term retentionof
singleitems.Journalof Verbal LearningandVerbal Behavior, 1:153–161.
Eva Koktova. 1995. Wh-extractionand the topic-focusarticulationof the sentence.In
BarbaraH. ParteeandPetrSgall,editors, DiscourseandMeaning: Papers in Honor
of EvaHajicova, pages255–271. JohnBenjamins,Amsterdam,TheNetherlands.
Dimitra Kolliakou. 1998. Linkhood and multiple definite marking. In GosseBouma,
242 References
Geert-JanM. Kruijf f, andRichardT. Oehrle,editors,ProceedingsJoint Conf. onFor-
mal Grammar, HPSGandCG, pages14–24,Saarbrucken,Germany, August.DFKI.
EstherKonig. 1990. The complexity of parsing with extendedcategorial grammars. In
Proc.COLING-90.
Paul Kroeger. 1993. PhraseStructure andGrammaticalRelationsin Tagalog. Disserta-
tionsin Linguistics.CSLI Press,Stanford, CA.
Geert-JanM. Kruijf f andJasonM. Baldridge. 2000. Relatingcategorial type logicsand
ccg. Journal of Language andComputation. Undersubmission.
Geert-JanM. Kruijf f, IvanaKruijf f-Korbayova,JohnBateman,ElkeTeich,andetal. 2000.
Multilingual generation for threeslavic languages. In Proceedings COLING2000.
IvanaKruijf f-Korbayova andBonnieLynn Webber. 2000. Discourseconnectives,infer-
ence,andinformation structure. In JohanBos andMichael Kohlhase,editors, Pro-
ceedings of ICOS-2, Inferencein Computational Semantics, pages105–120, Inter-
nationalConferenceandResearchCenterfor ComputerScience,SchlossDagstuhl,
Germany, July.
IvanaKruijf f-Korbayova. 1998. TheDynamicPotentialof Topic andFocus: A Praguian
Approach to DiscourseRepresentation Theory. Ph.D.dissertation, Facultyof Mathe-
maticsandPhysics,CharlesUniversity, Prague,CzechRepublic,June.
Geert-JanM. Kruijf f. 1995. The Unbearable Demise of Surprise: Reflections
on abduction in Artificial Intelligence and Peirce’s philosophy. Master’s the-
sis, Universiteit Twente, Enschede, The Netherlands, August. Available from
http://kwetal.ms.mff.cuni.cz/˜gj/abduction.html.
Geert-JanM. Kruijf f. 1997. Peirce’s theory of abduction andgoedel’s axioms of infinity
- reflections on the mathematics of creativity. Presentationat LOGIKA’97, Liblice
(Czech Republic). Available from http://kwetal.ms.mff.cuni.cz/˜gj/abduction.html,
June.
Geert-JanM. Kruijf f. 1998a. Basic,dependency-basedlogic grammar. TechnicalReport
TR-1998-05, UFAL MFF UK, Prague, CzechRepublic,May.
Geert-JanM. Kruijf f. 1998b. Peirce’s latetheoryof abduction: A comprehensiveaccount.
Availablefrom http://kwetal.ms.mff.cuni.cz/˜gj/abduction.html.
Geert-JanM. Kruijf f. 1999a. Diamonds for dependencies.In Proceedings of the Third
Workshopon Computational Semantics, pages351–366, Tilburg, The Netherlands,
January.
Geert-JanM. Kruijf f. 1999b. Resourcelogics,grammars,andprocessing.In ChristianRe-
toreandEdStabler, editors,ESSLLI’99WorkshoponResourceLogicsandMinimalist
Grammars, Utrecht,TheNetherlands,August.
Geert-JanM. Kruijf f. 2000. Categories,constructions,anddependency relations.In Pro-
ceedings of Text, Speech andDialogue 2000, CzechRepublic. Springer Verlag.
Geert-JanM. Kruijf f. 2001. A Categorial-Modal Logical Architecture of Informativity:
DependencyGrammarLogic & InformationStructure. Ph.D.thesis,Facultyof Math-
References /243
ematicsandPhysics,CharlesUniversity, Prague,CzechRepublic, April.
PetrKubon. 1998. Informationrepackaging. In GosseBouma,Geert-JanM. Kruijf f, and
RichardT. Oehrle,editors,Proceedingsof theJoint Conferenceon Formal Grammar,
Head-DrivenPhraseStructureGrammar, andCategorial Grammar, Saarbrucken,Au-
gust.
NatashaKurtoninaandMichaelMoortgat. 1996. Structural control. In PatrickBlackburn
andMaartenDe Rijke, editors, SpecifyingSyntacticStructures. CSLI Publications &
FoLLi, StanfordCA.
NatashaKurtonina. 1995. FramesandLabels:A Modal Analysisof Categorial Inference.
Ph.D.thesis,OTS,UtrechtUniversity.
JerzyKuryłowicz. 1964. TheInflectional Categories of Indo-European. Carl WinterUni-
versitatsverlag,Heidelberg, Germany.
Susanne Kuschert. 1996. Higherorder dynamics: Relatingoperationalanddenotational
semanticsfor © -drt. Claus-report 72, Computational Linguistics, University of the
Saarland, Saarbrucken,Germany.
JoachimLambek. 1958. Themathematicsof sentencestructure. AmericanMathematical
Monthly, 65:154–169.
Knud Lambrecht. 1996. InformationStructure andSentenceForm. Number 71 in Cam-
bridge Studiesin Linguistics.CambridgeUniversityPress,Cambridge,UnitedKing-
dom.
Winfred P. Lehmann. 1993. Theoretical Basesof Indo-EuropeanLinguistics. Routledge,
London,UnitedKingdom.
RichardL. Lewis andMineharuNakayama. 1999.Determinantsof processingcomplexity
in Japaneseembeddings:New theoryanddata.In MineharuNakayamaandCharlesJ.
Quinn, editors,Proceedings of theEastAsianPsycholinguisticsWorkshop, Stanford,
CA. CSLI.
RichardL. Lewis. 1998. Interferencein working memory: Retroactive andproactive in-
terferencein parsing. Unpublishedmanuscript,TheOhioStateUniversity.
Kirsten Malmkjæ, editor. 1996. The Linguistics Encyclopedia. Routledge, London,
UnitedKingdom.
SureshManandhar. 1994. Encoding informationpackaginginformationin HPSG.In Elis-
abetEngdahl,editor, IntegratingInformationStructureintoConstraint-basedandCat-
egorial Approaches, DYANA-2 Deliverable R1.3.B.,pages83–87.Institutefor Logic,
LanguageandComputation(ILLC), Amsterdam,TheNetherlands.
Christopher D. Manning. 1996. Ergativity: ArgumentStructure andGrammatical Rela-
tions. Dissertationsin Linguistics.CSLI Press,Stanford, California.
Pavel MaternaandPetrSgall. 1980. Functional sentenceperspective, the question test,
andintensionalsemantics.SMIL-Journal of LinguisticCalculus, (1-2):141–160.
Pavel Materna, Eva Hajicova, andPetrSgall. 1987. Redundant answersandtopic-focus
articulation. LinguisticsandPhilosophy, 10:101–113.
244 References
Pavel Materna. 1998. Conceptsand objects, volume 63 of Acta Philosophica Fennica.
Philosophical Societyof Finland,Helsinki,Finland.
Vil em Mathesius.1936. On someproblemsof the systematicanalysisof grammar. In
Travauxdu Cercle Linguistiquede Prague, volume VI, pages95–107. Reprintedin
(Vachek, 1964), pp.306-319.
Vil emMathesius.1975. Oninformationbearingstructureof thesentence.In S.Kuno, ed-
itor, Harvard Studiesin Syntax andSemantics. Harvard University Press,Cambridge,
Massachusetts.
MarcMoensandMarkSteedman. 1988. Temporalontology andtemporal reference.Jour-
nal of Computational Linguistics, 14:15–28.
TaraMohanan. 1994. ArgumentStructure in Hindi. CSLI Publications,Stanford.
MichaelMoortgat andGlyn Morrill. 1991. Headsandphrases:Typecalculus for depen-
dency andconstituent structure. Unpublishedmanuscript. Availablefrom http://www-
lsi.upc.es/˜morrill/.
MichaelMoortgatandDick Oehrle.1994. Adjacency, dependency andorder. In Proceed-
ingsof theNinthAmsterdamColloquium.
MichaelMoortgat. 1988.Categorial Investigations:Logical andLinguisticAspectsof the
LambekCalculus. Foris,Dordrecht,TheNetherlands.
Michael Moortgat. 1995. Multimodal linguistic inference. Bulletin of the IGPL,
3(2,3):371–402,June.SpecialIssueonDeductionandLanguage.
Michael Moortgat. 1996? Generalizedquantifiersanddiscontinuous type constructors.
In W. SijtsmaandA. vanHorck,editors,DiscontinuousDependencies. Mouton. Also
appeared astechnical report, OTS,University of Utrecht(1991).
Michael Moortgat. 1997a. Categorial type logics. In Johanvan BenthemandAlice ter
Meulen,editors,HandbookofLogicandLanguage. ElsevierScienceB.V., Amsterdam
New York etc.
Michael Moortgat. 1997b. Categorial type logics. In Johanvan BenthemandAlice ter
Meulen,editors,Handbookof Logic andLanguage, chapter2. Elsevier, Amsterdam.
Michael Moortgat. 1999. Meaningful patterns. In Jelle Gerbrandy, Maarten Marx,
Maartende Rijke, andYde Venema,editors, JFAK. EssaysDedicatedto Johan van
Benthemon the Occasionof his 50th Birthday, Vossiuspers.AmsterdamUniversity
Press,Amsterdam.
Glyn V. Morrill. 1990. Intensionality andbinding. LinguisticsandPhilosophy, 13:699–
726.
Glyn V. Morrill. 1994. TypeLogical Grammar:Categorial Logic of Signs. Kluwer Aca-
demicPublishers,Dordrecht, Boston,London.
Glyn V. Morrill. 2000. Type-logical anaphora. Report de Recerca/ResearchReport
LSI-00-77-R, Departament de Llenguatgesi SistemesInformatics, Universitat de
PolitecnicadeCatalunya.
G. E. Muller andA. Pilzecker. Experimentellebeitragezur lehrevomgedachtniss(Exper-
References /245
imentalcontributionson thetheoryof memory). In J.A. Barth,editor, Zeitschrift f ur
Psychologie. Erganzungsband1, pages???–??????,Leipzig.
Reinhard Muskens,JohanvanBenthem,andAlbert Visser. 1996. Dynamics.In Johanvan
BenthemandAlice ter Meulen,editors,Handbook of Logic andLanguage. Elsevier
ScienceB.V., AmsterdamNew York etc.
Kentaro Nakatani, Maria Babyonyshev, andEdward Gibson. 2000. The complexity of
nestedstructuresin Japanese.Posterpresentedat the CUNY SentenceProcessing
Conference,University of California, SanDiego.
Richard T. Oehrle. 1991. Prosodicconstraintson dynamic grammatical analysis. In
DeclarativePerspectiveson Phonology, Edinburgh Working Papersin Cognitive Sci-
ence.ECCS,Scotland.
RichardT. Oehrle. 1994. Term-labelled categorial typesystems.Linguistics& Philoso-
phy, 17(6):633–678,December.
RichardT. Oehrle.1995. Some3-dimensionalsystemsof labelleddeduction. Bulletin of
theIGPL, 3(2,3):429–448, June.SpecialIssueonDeduction andLanguage.
RichardT. Oehrle. 1999. Binding astermrewriting. In GosseBouma,Erhard Hinrichs,
Geert-JanM. Kruijf f, andRichardT. Oehrle,editors, Constraints and Resources in
Natural LanguageSyntax andSemantics, pages221–241.CSLI Press,Stanford CA.
RichardT. Oehrle. forthcoming. Multi-modal type-logical grammar. In RobertBorsley
and R. Borjars, editors, Non-Transformational Syntax. Blackwell, London, United
Kingdom.
JarmilaPanevova. 1974. On verbal framesin functionalgenerative descriptionI. Prague
Bulletinof Mathematical Linguistics, 22:3–40.
JarmilaPanevova. 1994. Valency frames and the meaningof the sentence.In Philip
Luelsdorff, editor, ThePrague School of Structural andFunctional Linguistics, vol-
ume41 of LinguisticsandLiterary Studies in EasternEurope, pages223–243.John
Benjamins,Amsterdam.
CharlesParsons.1995. Platonismandmathematicalintuition in Kurt Godel’s thought.
TheBulletinof Symbolic Logic, 1(1):44–74, March.
BarbaraH. Partee,Alice G.B. Ter Meulen,andR. Wall. 1990. Mathematical methodsin
linguistics. Kluwer Academic Publishers,Dordrecht NL.
BarbaraH. Partee. 1991. Topic, focus andquantification. In S. Moore andA. Wyner,
editors, Proceedings fromSemanticsandLinguisticTheoryI, pages257–280, Ithaca
NY. CornellUniversity.
Jaroslav Peregrin andPetrSgall. 1986.An attempt at a framework for semanticinterpre-
tationof natural linguistics.TheoreticalLinguistics, 13:37–73.
Jaroslav Peregrin. 1995. Topicandfocusin aformal framework. In BarbaraH. Parteeand
PetrSgall,editors,DiscourseandMeaning: Papers in Honor of EvaHaji cova. John
Benjamins,Amsterdam,TheNetherlands.
Jaroslav Peregrin. 1999. Pragmatization of semantics.In Ken Turner, editor, Seman-
246 References
tics/PragmaticsInterfacefromDifferentPointsof View, pages419–442. Elsevier, Am-
sterdam,TheNetherlands.
Vladimır Petkevic. 1987. A new dependency basedspecificationof underlying represen-
tationsof sentences.TheoreticalLinguistics, 14:143–172.
Vladimır Petkevic. 1995. A new formal specificationof underlyingstructures.Theoretical
Linguistics, 21(1):7–61.
Vladimır Petkevic. in prep. Underlying Structure of Sentence Basedon Dependency.
CharlesUniversity, Prague.
JanetPierrehumbertandJulia Hirschberg. 1990. The meaningof intonational contours
in the interpretation of discourse. In J. Morgan P. Cohenand M. Pollack, editors,
Intentions in Communication. TheMIT Press,Cambridge Massachusetts.
Carl PollardandIvanA. Sag.1993. Head-DrivenPhraseStructureGrammar. University
of ChicagoPress.
ScottPrevost. 1995. Contextual aspectsof prosody in monologuegeneration. In IJCAI
WorkshoponContext in NLP, Montreal,Canada.
AnnaGiacaloneRamatandPaoloRamat,editors.1998. TheIndo-EuropeanLanguages.
LanguageFamily Descriptions.Routledge,London,UnitedKingdom.
ChristophScheepers,BarbaraHemforth, andLarsKonieczny. 19??IncrementalProcess-
ing of GermanVerb-FinalConstructions: PredictingtheVerb’sMinimum (!) Valency.
??, ??:??–??
PetrSgall, Ladislav Nebesky, Alla Goralcıkova, andEva Hajicova. 1969. A Functional
Approach to SyntaxIn Generative Descriptionof Language. Elsevier ScienceB.V.,
AmsterdamNew York etc.
PetrSgall,Eva Hajicova,andEva Benesova. 1973. Topic,Focus,andGenerativeSeman-
tics. Kronberg/Taunus.
Petr Sgall, Eva Hajicova, and Eva Buranova. 1980. Aktualnı clenenı vety v cestine
[Topic/FocusArticulationof theCzechSentence]. Academia,Prague,Czechoslovakia.
PetrSgall,EvaHajicova,andJarmilaPanevova. 1986. TheMeaning of theSentencein Its
SemanticandPragmatic Aspects. D. ReidelPublishingCompany, Dordrecht,Boston,
London.
PetrSgall,OskarE. Pfeiffer, Wolfgang U. Dressler, andM. Pucek. 1995. Experimental
researchonsystemicordering. Theoretical Linguistics, 21:197–239.
PetrSgall,Alla Bemova, andEva Hajicova. 1996. Remarkson thesemanticfeaturesof
casesandprepositionsasrelatedto syntax. In EdnaAndrews andYishaiTobin, edi-
tors,Towardsa Calculusof Meaning, volume43 of Studiesin Functional andStruc-
tural Linguistics, pages71–82. JohnBenjamins,Amsterdam,TheNetherlands.
PetrSgall. 1980. A dependency basedspecificationof topic andfocusII. SMIL-Journal
of Linguistic Calculus, (1-2):110–140.
PetrSgall. 1995. Pragueschooltypology. In MasayoshiShibatani andTheodoraBynon,
References /247
editors, Approachesto Language Typology, pages 49–84. Clarendon Press,Oxford,
UnitedKingdom.
StuartShieber. 1986. Introduction to Unification-basedapproachesto grammar. CSLI,
Stanford, CA.
Vladimır Skalicka andPetrSgall. 1994. Praguiantypology of languages. In Philip A.
Luelsdorff, editor, PragueSchool of Structural andFunctional Linguistics, number41
in LinguisticsandLiteraryStudiesin EasternEurope,pages333–357. JohnBenjamins,
Amsterdam, TheNetherlands.
Mark Steedman.1996. SurfaceStructure andInterpretation. TheMIT Press,Cambridge
Massachusetts.
Mark Steedman.2000a. Informationstructure andthe syntax-phonology interface. Lin-
guisticInquiry, 31(4):649–689.
Mark Steedman.2000b. Theproductionsof time. Draft 4.0,July24.
Mark Steedman. 2000c. The Syntactic Process. The MIT Press,Cambridge Mas-
sachusetts.
Mark Steedman.2000d. TheSyntactic Process. MIT Press,Cambridge,MA.
SusanSteele.1978. Word ordervariation: A typological study. In JosephH. Greenberg,
editor, UniversalsofLanguage. Volume4: Syntax, pages585–624.StanfordUniversity
Press,Stanford, California.
Bohumil Trnka. 1932. Somethoughts onstructural morphology. In CharisteriaGuilelmo
Mathesio, pages57–61. Reprintedin (Vachek,1964), pp.329-334.
J.C. Trueswell,M. Tanenhaus,andS.M. Garnsey. 1994. Semantic influences onparsing:
Useof thematicroleinformationin syntacticambiguity resolution. Journal ofMemory
andLanguage, 33:285–318.
Natsuko Tsujimura.1996. AnIntroduction to JapaneseLinguistics. Basil Blackwell.
Keiko Ueharaand DianneBradley. 1996. The effect of -ga sequences on processing
Japanesemultiply center-embeddedsentences. In 11th Pacific-Asia Conferenceon
Language, Information, andComputation.
JosefVachek. 1964. A PragueSchool Reader in Linguistics. Indiana University Press,
Bloomington,IN.
EnricVallduvı andElisabetEngdahl. 1996. Thelinguistic realizationof informationpack-
aging. Linguistics, 34:459–519.
Enric Vallduvı andRonZacharski. 1994. Accenting phenomena, associationwith focus,
andthe recursivenessof focus-ground. In Paul Dekker andMartin Stokhof, editors,
Proceedings of the Ninth AmsterdamColloquium, pages683–702, Amsterdam, The
Netherlands.UniversiteitvanAmsterdam.
EnricVallduvı. 1990. TheInformationalComponent. Ph.D.thesis,University of Pennsyl-
vania,Philadelphia, USA.
Enric Vallduvı. 1994. The dynamics of information packaging. In ElisabetEngdahl,
editor, Integrating Information Structure into Constraint-basedand Categorial Ap-
248 References
proaches, DYANA-2 DeliverableR1.3.B.Institutefor Logic,LanguageandComputa-
tion (ILLC), Amsterdam,TheNetherlands.
JohanF.A.K. Van Benthem.1991. Language in Action: Categories,Lambdas, andDy-
namicLogic. North-Holland,AmsterdamNL.
Jan Van Eijck and Hans Kamp. 1997. Representing discoursein context. In Johan
VanBenthemandAlice TerMeulen, editors,Handbookof Logic andLanguage. Else-
vier, Amsterdam, etc.
Shravan Vasishth,Martin Jansche,and Chris Brew. in preparation. A corpus study of
Hindi centerembeddings.MS.
ShravanVasishth.2001. Evaluating working-memory basedmodelsof sentenceprocess-
ing: An empiricalinvestigation usingHindi center-embedding constructions. In OS-
UWPL, pages??–??,OhioStateUniversity.
ShravanVasishth.2002. Workingmemoryin sentencecomprehension:Anempiricalinves-
tigation of Hindi centerembeddings. Ph.D.thesis,Ohio StateUniversity, Columbus,
OH.
TheoVenneman. 1977. Konstituenz undDependenz in einigenneuerenGrammatiktheo-
rien. Sprachwissenschaft, 2:259–301.
K. Vijayashanker andDavid Weir. 1994. Theequivalenceof four extensions of context-
freegrammar. Mathematical SystemsTheory, 27:511–546.
Tomas Vlk. 1988. Towardsa transduction of underlying structuresinto intensional logic.
PragueBulletinof MathematicalLinguistics, (50):35–70.
StefanWechsler. 1995. TheSemanticBasisof Argument Structure. Dissertationsin Lin-
guistics.CSLI Publications,StanfordCA.
Henri Weil. 1844. TheOrder of Words in the AncientLanguagescomparsedwith that
of the ModernLanguages. JohnBenjamins,Amsterdam, The Netherlands. Second
edition1884; Frenchoriginal 1844, Englishtranslationby CharlesW. Super.
Hiroko Yamashita. 1997. TheEffectsof Word OrderandCaseMarking Informationon
theProcessingof Japanese.Journal of PsycholinguisticResearch, 26(2):163–188.
MariaLuisaZubizaretta.1998. Prosody, Focus,andWordOrder. Number31in Linguistic
Inquiry. TheMIT Press,Cambridge,Massachusetts.
L IST OF FIGURES
1.1 Simpledependency structure . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Simpledependency structure . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 A simplemultilingual network of structural rules . . . . . . . . . . . . . 39
1.4 A simplemultilingual network of structural rulesincluding Start . . . . . 40
2.1 Vallduvi’s grammararchitecture incorporatinginformationstructure . . . 59
3.1 Wordorderdata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.1 Thearchitecture of DGL’s wordordermodel. . . . . . . . . . . . . . . . 120
4.2 Dutch,FlemishandGerman verbraising . . . . . . . . . . . . . . . . . . 121
4.3 A multilingual network for subordinateclauseWO in Dutch,Flemishand
German . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4 Architecture of SVO packages . . . . . . . . . . . . . . . . . . . . . . . 144
4.5 Structuring of SVO wordorder . . . . . . . . . . . . . . . . . . . . . . . 145
5.1 Architecture of aGB theoryof Prosody . . . . . . . . . . . . . . . . . . 163
5.2 Architecture of aCCGtheory of Prosody . . . . . . . . . . . . . . . . . 164
6.1 Thenatureof contextual boundness . . . . . . . . . . . . . . . . . . . . 178
7.1 Example (287) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
7.2 EPDA processingof Dutchdependencies . . . . . . . . . . . . . . . . . 195
7.3 EPDA processingof Germandependencies . . . . . . . . . . . . . . . . 196
7.4 EPDA processingof Germanmixeddependencies. . . . . . . . . . . . . 197
7.5 Hindi examples(289a,b) . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.6 Processingof (290) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.7 Processingof (292a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.8 Processingof (292b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
7.9 Resultsof Experiment1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.10 SingleEmbeddings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.11 DoubleEmbeddings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.12 Experiment3 results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
249
LANGUAGE INDEX
Biblical Hebrew, 111
BrazilianPortuguese,96
Catalan,63,68
Chamorro, 149,150
Czech,3, 14–16, 18, 20–22, 30, 46, 48,
50, 52, 53, 66, 69, 84, 96, 99,
101,109,120,153,157,159
Dutch,16,18,24,48,49,84,86,95–97,
101, 104–106, 109, 113, 117,
119, 126–130, 132–138, 154,
156–158
English,2, 3, 6, 14–16, 18–20, 24, 30,
32, 37, 46–50, 53, 54, 57–60,
63, 65–69, 77, 78, 84–86, 89,
96,97,99,100, 105, 107, 110,
117,127,129,157
Flemish,128,129,135,136,138
French,14,96,97
German,14, 18, 31, 36, 48, 49, 84, 86,
95–97,101,104–106,109,113,
117, 128–130, 132, 136–139,
154,157
Greek,96,101
Hebrew, 96
Hindi, 93, 95, 101, 106, 107, 109, 145,
163
Hua,47
Hungarian, 93, 95, 101, 102, 104–107,
109,113,163,164
Italian,96
Japanese,3, 14–16, 18, 24, 47, 48, 85,
93,95,97,101, 109, 120, 129,
130, 139–141
Korean, 24,93,95,141
Mandarin, 96
Navajo,48
Portuguese,117
Russian,46,96,101,159
Sanskrit,17
Sinhala,99,101, 109,141,163,165
Swedish,96,154
Tagalog, 48,93,111,112,148–150
Tamil, 99,101
TobaBatak,147
Turkish, 48, 60, 72, 84, 91, 93, 95, 101,
102, 104, 106, 107, 109, 113,
139, 163,164
Vute,48
Welsh,85
NAME INDEX
Ades,Anthony, 117
Adjukiewicz, Kazimierz,117, 118
Areces,Carlos,33
Baldridge, Jason,12, 72, 117,120, 122,
123, 150
Bar-Hillel, Yehoshua,4, 117,118
Barry, Guy, 5
Benthem, JohanF.A.K. van,32,124
Bloomfield, Leonard,10
Carnap, Rudolf,66
Carpenter, Bob,118
Chafe,William, 60
Chomsky, Noam,32,37,38
Comrie,Bernard,38,85
Croft, William, 17,92
Curry, Herbert,117, 118
Dahl, Osten,62
Danes,Frantisek,14,50,67
Davis, Anthony R., 15,17,25
Dekker, Paul,61
Dokulil, M., 14
Dowty, David R.,15,25
Engdahl,Elisabet,48,61,63–65,69,70,
105, 106
Erteshik-Shir, Nomi, 27
Feys,Robert,117,118
Fillmore,CharlesJ.,25
Firbas,Jan,52,79,100
Foster, JohnC.,127
Gamut,L.T.F., 32
Greenberg,JosephH., 38,84,86–88,90,
92,94,102, 141
Grosz,Barbara,50
Hahn,Udo,50
Haiman,John,47,48
Hajicova,Eva,50,52,55–57,66,69
Hale,Kenneth,91,92
Halliday, Michael A.K., 39, 57, 58, 61,
66–68
Hawkins,JohnA., 38,84,86–88,90,92,
94,102, 103,141
Heim,Irene,28,61
Hendriks,Herman,61,65,69
Hepple,Mark, 5, 7, 8, 12, 13, 117,124,
127
Herring,SusanC., 109, 141
Heylen,Dirk, 20,21,23
Hoffman,Beryl,12,60,72,73,102, 117,
119,120
Hudson,Richard,9
Jackendoff, Ray, 38,55,58,61,64
Jacobson,Pauline,28
Jakobson,Roman,13,14
Janssen,TheoM.V., 32
Jespersen,Otto,14
Karttunen, Lauri, 49
Keenan,Edward,38
Koktova,Eva,77
Koller, Alexander, 33
Kroeger, Paul,48,85,111,150
Kruijf f-Korbayova,Ivana,50,56,66,69,
72,79
Kubon, Petr, 61
Kurtonina,Natasha,124
Kuryłowicz, Jerzy, 14,16
Lambek, Joachim, 2, 3, 5,19,43,61,84,
124,127
Lambrecht, Knud,50
252 NameIndex
Lehmann, WinfredP., 86
Manandhar, Suresh,70
Manning, ChristopherD., 85
Materna,Pavel, 55
Mathesius,Vil em,13,14,50,57,69
Mel’ cuk, Igor, 9
Moens,Marc,13
Montague,Richard,32
Moortgat,Michael,5–8, 27,65,117,124–
127,129,135
Morrill, Glyn V., 5–8, 32, 65, 66, 118,
122,124,125
Morris, Charles,66
Oehrle,RichardT., 5, 8, 26,27,75,117,
124–127
Oliva,Karel,23
Paolillo, JohnC., 109,141
Partee,BarbaraHall, 32,50,55,56
Peregrin, Jaroslav, 55,56,58,66,79
Petkevic, Vladimır, 54,55,70,71,163
Pickering,Martin, 5
Pollard,Carl,65
Rooth,Mats,49,58
Sag,IvanA., 65
Selkirk,Lisa,70
Sgall,Petr, 7, 14,16,17,27,28,32,48,
50–52, 54, 55, 57, 69–71, 76,
79, 80, 84, 91, 92, 103, 108,
153,163
Sidner, CandaceL., 50
Skalicka,V., 14,84,91–93, 95,96,98
Speas,Margaret,91,92
Steedman,Mark, 2, 3, 9, 13,19, 38,49,
57–60, 62, 65–68, 70–75, 78–
80,117, 119, 124–127, 129
Steele,Susan,84,90–93,98
Strube,Michael,50
Tichy, Pavel, 55
Trnka,Bohumil, 14
Tsujimura,Natsuko, 139
Vallduvı, Enric, 48–50, 60–65, 67–70,
80,105,106
Venneman,Theo, 5, 7, 10, 38, 86, 87,
103
Vijayashanker, K., 117
Vlk, Tomas,55
Webber, BonnieLynn, 66,72
Wechsler, Stefan,15,17,25,26
Weil, Henry, 50
Weir, David, 117
Zubizaretta,MariaLuisa,50
SUBJECT INDEX
R-dependent,12,13
R-head,12,13
action-type,seeAktionsart
adjacency, 124,125
agreement,21
alternative set,58,59
aspect,13,93
attribute-valuematrix,23
background
Steedman, 58,59,62,70,72
case,seemorphology, case
categorial grammar, 1, 4, 5, 27, 28, 32,
38, 58, 60, 61, 65, 66, 84, 88,
115–117, 169
category, seetype
type,12
categorial typelogic,4–6,12,19,20,23,
26,28,33,39,40,43,117,118,
120–122, 124, 125, 127, 169
CombinatoryCategorial Grammar, 19,38,
40,57,58,60,71–74,81,117–
124, 126, 127, 169
modalized,117,121–123
commutativity, 125
compositionality, 32
constituency, 5
Curry-Howard correspondence, 26, 33,
116
delicacy, 40
dependency
head-dependentassymetry, 2, 4, 5,
7, 38,86,125, 127
semanticimport, 36
dependency grammar, 1, 4, 7, 9, 25, 28,
32,65,108,161
dependency relation,2–4,10,11,13–17,
29,31,43,45,48,150
Actor, 2–4, 16,30,33,48,53,54
Addressee,3, 16,30,135
Beneficiary, 107,135
Direction
WhereTo, 17
Patient,3,4,14–16,30,53,54,112,
135
Time
How Long,16
dialogue test,28
dynamic semantics,55
endocentric category, 10
event
eventnucleus,29
eventuality, 13,28,29
exocentric category, 10
file-change metaphor, 28
File-ChangeSemantics,61
focal fronting, 48
focus
canonical focus position, 99, 100,
103, 104, 107,108, 110, 111,
113,163
Steedman,58,59,62,68,70,72,75
Vallduvı, 62,67,68,70,80
form, 1, 13,15,49,54,58,59
function, 1, 17
functionword,18,24,27,28,31,47,54,
89,139
Functional GenerativeDescription, 10,50–
52, 54, 55, 59, 62, 65, 66, 70,
254 SubjectIndex
71, 74, 75, 79–81, 100, 113,
153,163
gapping, seeextraction,gapping
gender, seemorphology, gender
Government& Binding, 14,43,49,70
GPSG,7
ground
Vallduvı, 62,67,68,70,80
headdomain,11
headwrapping, 125
HPSG,7, 21, 23, 27, 61,64, 65, 68,70,
81
hybrid logic, 33,34,43,45,56,79
informationpackaging,49,60–65,70,72,
74,78,80,81
informationstructure,13,36,45–50,55,
57, 58, 60, 61, 63–75, 79–81,
83, 84, 88, 91, 98–100, 103–
108, 110, 111, 113, 115, 116,
120, 150, 160, 161, 163, 168,
169
Attributum,46,48,51,57
focusprojection,100,105–108,113,
161
Relatum,45,46,48,51,57
informativity
communicativedynamism,51–53,69,
79,80,100,161
contextual boundness,45, 51, 55,
57, 58, 75, 83, 99, 109, 115,
116,160,161,163
contextually bound, 51–54, 58, 59,
62,100, 161
contextually nonbound, 51–54, 58,
59,62,100,160,161
structuralindications,46,48,49,72–
75,81,99,107, 110, 111, 113,
115, 120, 128, 161, 162, 165,
167,169,170
lexical meaning, 10,17,25,29,31
lexical semantics,26
logical,26
linguisticmeaning, 1,13,15,17,29,31–
33, 35, 36, 43, 45, 46, 49–55,
57, 58, 60, 62, 66, 70, 71, 74,
75,79–81, 100, 116
link
Vallduvı, 62,67,68
linking theory, 18,19,24–26, 28,29,88
markedness,72
Meaning-Text Model, 9
modality, 93
MontagueGrammar, 32,66
morphology
casemarker, 18
category, 14,15,17,18,24–26, 29,
31,43
delimitation, 14
exponent,14
morphological strategy, 1, 14, 15,
17,19,29,43,47,92,94,132
adposition,17–19, 24
case,17–19
linker, 17,18
position, 17–19, 24
number, 14
multisetcombinatorycategorial grammar,
72,73,117,119–121,169
nominal, seehybrid logic, nominal
number, seemorphology, number
person, seemorphology, person
phrase-structuregrammar, 38, 65, 116,
117, 153
pied-piping,seeextraction, pied-piping
Postpositional NounModifier Hierarchy,
87
PragueSchool of Linguistics,13–15, 25,
50,91,100,113
SubjectIndex /255
predicate-valency structure,1, 10,29–31
Prepositional Noun Modifier Hierarchy,
87
presupposition,55
PrinciplesandParameters,38
relational nouns,28
Rheme
Halliday, 67
Steedman, 49, 57–60, 65, 67, 68,
71,75,80
salience,51
Set-CCG,12,72,117, 119–122, 127, 139,
169
Stockof SharedKnowledge,51
structural control, 124
structural rule, 7, 9, 12, 20, 21, 23, 31,
35, 36, 39–42, 81, 116, 124,
125, 127–130, 133, 135, 137,
139, 140, 142–144, 147, 150,
151, 154–162, 164, 165, 167
associativity, 125
systemicordering,52,100,105,107,115,
160, 161, 167–169
tail
Vallduvı, 62
tectogrammaticalrepresentation,51, 52,
55
tense,13,93,97
thematicstructure,50,57,66–70,80,99,
108, 113
Theme
Halliday, 57,61,66–68
Steedman, 49, 57–60, 65, 67, 71,
73,75,80
topic-focusarticulation, 45, 49–51, 54–
57, 65, 66, 70, 71, 75–77, 79,
80,100
focus,50–56,71,75–77,79,99,101–
103, 105–107, 109, 111, 161,
163,164,167, 168
focusproper, 52,71,75,76,99,100,
105–108,110, 113, 163–165
topic,50,51,53–55, 71,75–77,79,
100,105,109, 163, 167, 168
topicproper, 52,76
Topic/Focus-sensitive DiscourseRepre-
sentationTheory, 56,57,79
topicalization, 48
transparent intensional logic, 32,55
tripartitestructure, 55–57
tune,47, 48, 50, 52, 57, 60, 70–75, 81,
83–85, 99,100,103,106,107,
109–111, 113, 160–163, 165,
167–170
type-raising,19
unarymodaloperator, 20, 35, 124, 129,
130
box,20
diamond, 20
universal,86
implicational, 86,87,93
UniversalGrammar, 37
valency frame,10
freemodifier, 10
innerparticipant,10
verbfinal, 128,129, 132,135, 136, 139,
141,142,144, 153, 155, 156
verbinitial, 147, 148
verbraising,129
verbsecond,103,153–157
Wackernagelposition, 153–155, 157, 159
WordGrammar, 9
wordorder, 12,41,46–48,50,52,57,60,
70–75, 80, 81, 83–85, 87, 88,
90–103,106,109–111,113,115–
117,119, 120, 124,125,127–
131,138, 139, 141,143,145–
256 SubjectIndex
150, 152–156, 158–161, 163,
164,166,168,169
basic,85–88, 119
free,72, 83, 84, 91–103,109–111,
113, 115, 116, 119, 120, 139,
143, 145, 149, 150, 152, 154,
157–159,163,169
mixed,83,84,91–95,97–103,109–
111, 113, 115, 116, 119, 139,
141,152–154,156,157,169
OV (orXV), 87,93,95,97–99,101–
104, 106, 108–111, 113, 115,
116, 130–132, 139, 141, 143,
146,149,162,163,169
rigid, 83, 84, 91–95, 97–99, 102,
103, 113, 115, 116, 153, 154,
157,169
SOV, 85,87,99,105, 119,121, 129,
139,141
SVO, 85, 87, 96, 97, 99, 101–106,
109–111, 115, 116, 119, 129,
131,146,147, 152, 153,155–
158,167–170
V1, 147
variationhypothesis,93,95–98,115,
128,169
VO (or VX), 87, 98, 99, 102–104,
110, 111, 115, 116, 144, 146,
150,165,169
VOS,87,146–148
VSO,85,119,146,147
wordform, 18,24