1

Microsoft Content Management Server Field Guide Apress Francois-Paul Briand, Michael Wirsching Year: 2005 Language: english File: PDF, 1.10 MB Your tags: 0 / 0

2

Parentheticals (Linguistik Aktuell Linguistics Today, 106) Nicole Dehe, Yordanka Kavalova Year: 2007 Language: english File: PDF, 2.45 MB Your tags: 0 / 0

The Bantu–Romance Connection

Linguistik Aktuell/Linguistics Today (LA)
Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph
studies into synchronic and diachronic linguistics. Studies in LA confront empirical
and theoretical problems as these are currently discussed in syntax, semantics,
morphology, phonology, and systematic pragmatics with the aim to establish robust
empirical generalizations within a universalistic perspective.

General Editors
Werner Abraham

University of Vienna / Rijksuniversiteit
Groningen

Elly van Gelderen

Arizona State University

Advisory Editorial Board
Cedric Boeckx

Christer Platzack

Guglielmo Cinque

Ian Roberts

Günther Grewendorf

Lisa deMena Travis

Liliane Haegeman

Sten Vikner

Hubert Haider

C. Jan-Wouter Zwart

Harvard University
University of Venice

J.W. Goethe-University, Frankfurt
University of Lille, France
University of Salzburg

University of Lund

Cambridge University
McGill University

University of Aarhus
University of Groningen

Volume 131
The Bantu–Romance Connection. A comparative investigation of verbal agreement, DPs, and information structure
Edited by Cécile De Cat and Katherine Demuth

The Bantu–Romance Connection
A comparative investigation of verbal agreement,
DPs, and information structure

Edited by

Cécile De Cat
University of Leeds

Katherine Demuth
Brown University

John Benjamins Publishing Company
Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of
American National Standard for Information Sciences – Permanence of
Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data
The Bantu-Romance connection : A comparative investigation of verbal agreement, DPs,
and information structure / edited by Cécile De Cat and Katherine Demuth.
       p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166-0829 ; v. 131)
Includes bibliographical references and index.
1. Bantu languages--Grammar, Comparative--Romance. 2. Roma; nce language-Grammar, Comparative--Bantu. I. De Cat, Cécile. II. Demuth, Katherine.
PL8025.1.B36    2008
496'.39045--dc22
isbn 978 90 272 5514 3 (Hb; alk. paper)

2008023397

© 2008 – John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

Acknowledgements

vii

List of contributors

ix

Introduction

xi

Part 1. Clitic and agreement
Concepts of structural underspecification in Bantu and Romance
Lutz Marten, Ruth Kempson & Miriam Bouzouita

3

On different types of clitic clusters
Anna Cardinaletti

41

Pronominal object markers in Bantu and Romance
Marie Labelle

83

The Bantu-Romance connection in verb movement
and verbal inflectional morphology
Carolyn Harford

111

Part 2. The structure of DPs
DP in Bantu and Romance
Vicki Carstens

131

On the interpretability of φ-features
Roberto Zamparelli

167

Agreement and concord in nominal expressions
Giuliana Giusti

201

A unified syntactic analysis of Italian and Luganda nouns
Franca Ferrari-Bridgers

239



The Bantu-Romance connection: A comparative investigaton

Part 3. Information structure
The fine structure of the topic field
Mara Frascarelli

261

Focus at the interface: Evidence from Romance and Bantu
João Costa & Nancy C. Kula

293

Agreement in thetic VS sentences in Bantu and Romance
Jenneke van der Wal

323

Index of languages

351

General index

353

Acknowledgments
We wish to acknowledge the financial support of the European Science Foundation (grant #EW05-242), the British Academy (grant #BCG-41902) and the Linguistic Association of Great Britain which helped bring to life the Bantu-Romance
Connection ESF Exploratory Workshop (Leeds, May 2006).
Every stage of the workshop and this volume was extensively reviewed. We
would especially like to express our gratitude to the team of Bantu and Romance
scholars who provided extensive comments on the chapters of the book. Each
chapter was reviewed by both a Bantuist and a Romanist. Thanks also to the contributors and participants for having embarked with us on this project, and for
having made it such a stimulating experience.

List of contributors
Miriam Bouzouita
King’s College London
Philosophy Department
The Strand
London, WC2R 2LS
United Kingdom
miriam.bouzouita@kcl.ac.uk
Anna Cardinaletti
Ca’ Foscari University of Venice
Dipartimento di Scienze del Linguaggio
Ca’ Bembo – Dorsoduro 1075
30123 Venezia
Italy
cardin@unive.it
Vicki Carstens
University of Missouri-Columbia
Department of English
107 Tate Hall
Columbia, MO 65211
USA
carstensv@missouri.edu
João Costa
Departamento de Linguistica
Faculdade de Ciências Sociais
e Humanas
Av. de Berna, 26 C
1069-061 Lisbon
Portugal
jcosta@fcsh.unl.pt
Cécile De Cat
Department of Linguistics & Phonetics
School of Modern Languages and Cultures
University of Leeds
Leeds
LS2 9JT
UK
c.decat@leeds.ac.uk

Katherine Demuth
Department of Cognitive & Linguistic
Sciences
Brown University
Box 1978
RI 02912
Providence
USA
Katherine_Demuth@brown.edu
Franca Ferrari-Bridgers
Department of Speech and Communication
Studies
IONA College
18 President Street, New Rochelle, NY
ff244@nyu.edu
Mara Frascarelli
Università degli Studi Roma Tre
Dipartimento di Linguistica
Via Ostiense, 236-00146 Roma
Italy
frascare@uniroma3.it
Giuliana Giusti
Università Ca’ Foscari Venezia
Dipartimento di Scienze del Linguaggio
Ca’ Bembo – Dorsoduro 1075
30123 Venezia
Italy
giusti@unive.it
Carolyn Harford
Midlands State University
P Bag 9055
Gweru
Zimbabwe
harfordc@msu.ac.zw
Ruth Kempson
King’s College London



List of contributors
Philosophy Department
The Strand
London, WC2R 2LS
United Kingdom
ruth.kempson@kcl.ac.uk
Nancy Kula
Department of Language & Linguistics
University of Essex
Wivenhoe Park
Colchester CO4 3SQ
UK
nckula@essex.ac.uk
Marie Labelle
Université du Québec à Montréal
Département de linguistique et de
didactique des langues
Université du Québec à Montréal
C.P. 8888, succ. Centre-Ville
Montréal, QC H3C 3P8
Canada
labelle.marie@uqam.ca

Lutz Marten
School of Oriental and African Studies
Africa Department
Thornhaugh Street
Russell Square
London, WC1H 0XG
United Kingdom
lm5@soas.ac.uk
Jenneke van der Wal
Leiden University Centre for Linguistics
van Wijkplaats 4
2311 BX Leiden
The Netherlands
J.van.der.Wal@let.leidenuniv.nl
Roberto Zamparelli
Dipartimento di Scienze della Cognizione e
della Formazione
Università degli Studi di Trento
Via Matteo del Ben, 5
38068 Rovereto
Italy
roberto.zamparelli@unitn.it

Introduction
The study of Romance linguistic structures has a long and fruitful tradition, with
issues in comparative Romance being hotly debated today. However, the study of
Bantu languages, especially from a more theoretical perspective, has received much
less attention. There are approximately 500 Bantu languages in Africa (Nurse &
Philippson 2003). These form a minor sub-branch of Niger-Khordofanian languages, though geographically they cover much of central, eastern and southern
Africa. These languages are morphologically and syntactically similar in many
respects (much as Romance or Germanic languages are). However, they also
exhibit significant linguistic diversity, providing an especially rich (but as yet largely
untapped) resource for understanding linguistic structure and language change.
Recent funding by the British Academy for the project Bantu Grammar: Theory and
Description, has brought together Bantu researchers from the School for Oriental
and African Studies (SOAS) in London, the University of Leiden, and the Zentrum
für Allgemeine Sprachwissenschaft (ZAS) in Berlin, further stimulating research
and discussion regarding Bantu linguistic structure. The study of Bantu linguistic
structures, and the import this has for the field of linguistics more generally, is
therefore gradually becoming more widely appreciated.
Of particular note is that several aspects of Bantu morphological and syntactic
structure appear to be similar to structures in many Romance languages. Like
most Romance languages (except French), Bantu languages permit null subjects,
with subject agreement being marked on the verb. Likewise, when the object is
pronominalized, both groups of languages show object clitics on the verb. As a
result of these rich “agreement” systems, lexical subjects and objects can appear in
various surface word orders, or not at all. Compare, for example, the Sesotho and
Spanish sentences in (1) and (2) below.
(1) (Thabo) o-rata
basadi.
		 (Thabo) agr-like women
		 ‘Thabo likes the women.’

Oa-ba-rata,
Thabo.
agr-obj-like, Thabo
‘He likes them, Thabo.’

(2) (Juán) conoce
a las mujeres.
		 (Juan) know/agr prt det women
		 ‘Juan knows the women.’

Las conoce,
Juán.
obj-knows/agr Juan
‘He knows them, Juan.’

(Sesotho)

(Spanish)

Recent research on the syntax of issues of subject and object clitics and variability
in word order has concentrated primarily on comparative Bantu or comparative
Romance, often in separate African or Romance conferences (e.g., WOCAL – World

 The Bantu-Romance connection: A comparative investigation

Conference on African Languages, Going Romance). There has been little discussion of these issues across these research communities. At more general conferences there are typically only a few papers presented on these topics, with no
critical mass present to enhance in-depth discussion or comparative study. The
only project to date that has explored similarities between Bantu and Romance
languages (led by Bantuists at SOAS and funded by the AHRB) focused exclusively
on morphological issues.
The time is right to take Bantu research to the broader linguistic community
and to extend the field of comparative investigation to syntax and information
structure, providing a breeding ground for new collaborative research.
The purpose of The Bantu-Romance Connection Exploratory Workshop
(funded by the European Science Foundation), in which the contributions to this
volume are rooted, was therefore to foster further communication between Bantu
and Romance research communities in an effort to better understand the nature of
linguistic structure, and its diversity and constraints. In particular, participants were
invited to address several questions regarding the surface similarities found in these
two language groups, and to identify the linguistic tests needed to determine if the
syntactic structures underlying these surface similarities are actually the same or
not. Presenters were invited to discuss the following (partially overlapping) topics:
1.

Clitics, agreement, object drop
• What is the decisive evidence for the agreement/pronominal status of
subject and object markers/clitics?
• What is the grammatical status of overt expletives, and how are these
realized?
• What are the discourse/syntactic restrictions on null objects?
• What is the connection between object agreement and specificity?

2.

DP structure
• What is the structure of the DP?
• At what level of structure are determiners and/or noun class (pre)prefixes
specified?
• How does this interact with definiteness and/or specificity?
• What is the structure of concord?
• How does this explain, within language groups and across groups, similarities and differences in surface morpho-syntactic structure?

3.

Focus, Topic and Information Structure
• What evidence bears on the structural analysis of lexical subjects? Are
these actually “topics” rather than “subjects”?
• What is the structure for topicalization, clefting, right and left dislocation?
What are the syntactic/discourse restrictions on such structures?

Introduction 

It was hoped that the outcome of the workshop would help establish a framework for fostering future communication between scholars of Bantu and Romance
languages. One of the ways it could have a lasting effect on the field was by encouraging presenters to enter into Bantu/Romance collaborative teams. We hoped
that the resulting volume would provide a framework for future workshops and
publications of this type, and would be of interest as a teaching text for higher-level
linguistics seminars.
The intriguing similarities in between these Romance and Bantu languages, in
both their agreement systems and various aspects of their syntactic word-order possibilities and resulting information structures, provide a new, exciting field of research
that should lay the groundwork for forging a better understanding of the structure of
language. In particular it will provide a deeper understanding of interactions at the
morphology-syntax interface. This will in turn contribute to our understanding of
language change, and ultimately the evolution and acquisition of language itself.
In the following section we briefly discuss the theoretical highlights of each
of the chapters.

Part I: Clitics and agreement
Marten, Kempson & Bouzouita explore word order and subject/object clitic similarities between Bantu (Otjiherero) and Romance (Latin, Spanish). This is couched
within the framework of Dynamic Syntax (DS) (Cann, Kempson, Marten 2005).
First, the authors provide a detailed analysis of the different word order possibilities in Bantu and Romance, showing how DS provides a uniform framework for
analysis for word order variation across the two language groups. They then argue
that Bantu subject and object clitics can be analyzed as being similar to Romance
(object) clitics, where unfixed nodes have to be construed within a tightly locally
restricted domain. Support for their analysis comes from restrictions on the
presence of object markers in passive and locative inversion constructions in
Otjiherero, which follow from independent constraints on the availability of
unfixed nodes within a given domain. Thus, despite differences in surface morphology between the two language groups, Bantu and Romance exhibit striking
parallels with respect to the way lexical information and general structure-building
principles of interact within a theory of DS.
Cardinaletti brings to light a restriction on clitic clusters hitherto ignored in
the literature, determining which clusters can appear as proclisis and which as
enclisis. The latter is shown to be available to fewer types of clitic clusters, while
the former is available to all. Another distinguishing property is vowel change
(in Italian): clitic clusters that exhibit vowel change can occur in both positions,
but only those with no vowel change can appear as enclisis. These restrictions are

 The Bantu-Romance connection: A comparative investigation

argued to derive from partially independent properties of clitic clusters involving
their internal structure and whether they are inserted as lexical units or independent words. Cardinaletti argues that the two types of clusters (i.e., those that can
appear anywhere vs. those restricted to the proclitic position) result from different
types of adjunction: adjunction of one clitic to another, or adjunction to different
functional heads. Only the former structure can give rise to enclitic clusters. Two
clitic positions are distinguished: one in the IP layer and one in the VP layer. Different features are checked in each position, and only clitic clusters resulting from
adjunction on the same head are possible in the VP-internal position. Case is
shown to only play a limited role in explaining the cluster-internal ordering of
clitics. The data are mainly from Italian, but comparison with Bantu leads to a
refinement of the proposal with respect to the (possibly universal) merging order
of direct object and indirect object clitics. Differences between the two language
groups are attributed to independent differences in clause structure.
Labelle investigates the status of object markers in Romance and Bantu. She
contends that the traditional split in the literature between analyses in terms of
affix vs. clitic pronoun needs to be overcome if a unified analysis of object markers
is ever to be attained. Based on evidence from French (including substantial diachronic evidence) and Chicheŵa, she shows that many aspects of the distribution
of object markers cannot be captured by a strictly morphological analysis and that
these elements must be visible to the syntactic component. She goes on to propose
a tentative unified analysis. Critically, she proposes that there is a continuum from
independent word to affix, with object clitics situated more on the clitic side in
Romance than in Bantu. By recognizing that clitics have a mixed status, Labelle
shows that one can account for their morphological properties without having to
overlook crucial syntactic properties.
Harford examines a number of morphological and syntactic properties of
WH-extraction constructions and verbal inflectional morphology in two Bantu
languages (Isizulu and Chishona), and two Romance languages (standard French
and Italian). She analyses these properties in terms of a common right-branching
Split-INFL structure, in which the lexical projection of V is dominated by functional projections of inflectional morphemes. She suggests that the relatively higher
degree of fusion in verbal inflectional morphology in the Romance languages can
be understood as a constraint that minimizes the number of suffixal inflectional
morphemes whose postverbal directionality clashes with the basic order of functional and lexical morphemes. This reinforces the idea that Head-to-Head movement may be used to account for the directionality and relative fusion of verbal
inflectional morphology. It also suggests that favoring or disfavoring movement, as
expressed in relative rankings for the Optimality Theoretic STAY, cut across syntactic and morphological constructions. Thus, despite their surface morphological

Introduction 

differences, Harford suggests that Romance and Bantu languages are typologically
very similar.

Part II: The structure of DPs
Carstens compares several aspects of Bantu and Romance DP structure, drawing
especially on data from Swahili and Spanish. First, she argues that the Bantu Noun
Class is a gender system very much like that of Romance. Drawing on data from
Swahili diminutive/augmentative processes, she argues that apparent derivational
properties of Noun Classes and their prefixes are in fact due to derivational zero
morphemes in Bantu with their own gender specifications, much like the phraselevel null element pro found in both Bantu and Romance, with concomitant recoverability/identification requirements. She suggests that Bantu has a greater number
and diversity of null elements than Romance because the larger number of genders
enables them to be unambiguously identified, facilitating recoverability. Carstens
then proposes that Romance and Bantu DPs share a common architecture, and
that both groups of languages exhibit noun raising (though with different landing
sites). She then argues against Cinque’s (2005) anti-symmetric approach to postnominal modifier order, showing that a symmetrical base-generation approach
to modifiers is preferable. Finally, she shows the Agree relation (Chomsky 2000,
2001) in both Bantu and Romance can easily account for the agreement within the
DP (concord). Carstens concludes that all aspects of DPs in the two languages are
therefore highly similar, in keeping with Universal Grammar. This paper therefore
provides the basis for a much larger comparative analysis of these issues, both
within and between each language group.
Zamparelli’s paper explores the relation between the ± interpretable and ±
valued status of features, within the Chomskian Minimalist program of generative grammar. He argues that uninterpretable features are not automatically unvalued, and shows that at least some semantic features must be visible to Agree
(an operation traditionally considered to handle syntactic information exclusively). The paper opens with a discussion of the notions of default and interpretability, and the handling of lexical “exceptions”. It then explores the status of the
person feature (pers) in light of what has been said about defaults, and proposes a
distinction between a genuine 3rd person value for this feature and the assignment
of pers = 3rd to categories without person specification. Turning to gender (gend),
Zamparelli contends this feature should not be part of “core” grammar as it is not
autonomously interpretable in spite of being valued. As for number agreement
(num), this is shown to involve semantic as well as syntactic features. The facts are
argued to require a modification of the Agree system. Zamparelli draws extensively

 The Bantu-Romance connection: A comparative investigation

on Romance evidence (especially from Italian) and identifies how Bantu data might
bear on the issues discussed and advance this research program.
Giusti investigates feature sharing in Nominal Expressions, drawing on data
from Romance (mainly Romanian, Italian and Spanish) and Bantu (mainly Swahili
and Xhosa). Concord on possessors and modifiers in Romance is considered to be
parallel to prefixes in Bantu, and the determiner system in Romance as parallel to
pre-prefix spreading and multiple demonstratives in Bantu. Giusti contends that
feature sharing involves two distinct structural relations: Agreement and Concord.
Concord involves modifiers, is triggered by Number, Gender or Class features, and
never results in movement. Agreement involves subjects, is triggered by Person
features, and is argued to consist of the operations probe and move. However in
both Romance and Bantu, Agreement in Nominal Expressions is shown to seldom
result in movement: this only happens with pronominal and adjectival possessives,
which are argued to have pronominal reference. In this case the goal is exclusively
formed by the probed features and pied-piping to the specifier of the probe takes
place. Additional support for the distinction between Agreement and Concord is
found in the co-occurrence of these two relations on the same element (the genitival article in Romanian and connectors in Bantu).
Ferrari-Bridgers argues in favor of a unified syntactic analysis of Luganda
and Italian simple nouns. Specifically, she shows that, in both languages, all nouns
are formed via the merger of the nominalizer head [n] with a nominal stem [LP],
yielding the nominal structure [nP [n [LP]]]. She further claims that syntactic
movement is necessary in the formation process of Italian nouns to derive the
correct morpheme order. In order to demonstrate that the structure [nP [n [LP]]]
is representative for both languages, she shows that the nominalizer head [n] corresponds to both the Italian gender feature and the Luganda class feature and that,
therefore, gender and class are the same feature with identical inflectional and derivational functions: At the inflectional level, gender and class trigger VP and DP agreement, whereas at the derivational level, gender and class function as n-marked heads
whose merger with an XP yields a noun. She further argues that, unlike Luganda
nouns in general and Italian masculine nouns, Italian feminine nouns are derived
by movement. Drawing on data from diminutive formation, she also argues that
XP movement is the preferred type of movement in derived nominal formation.
Thus, by reinterpreting the gender/class as a nominalizer, it is possible to provide a
uniform treatment for the derivation of both Bantu and Romance nouns.

Part III: Information structure
Frascarelli argues that a one-to-one mapping between syntax, phonology and discourse underlies the encoding of topics and subjects. Three types of topics are

Introduction 

distinguished (aboutness-shift, familiar, and contrastive) and shown to correspond to dedicated functional projections within the C-domain, each of which
is associated with distinct intonational patterns. Syntax plays a central role in this
approach: it determines the interpretation (scope effects, contrast, discourse functions), the word order and the prosody. PF operations are ruled out: prosodic patterns are determined by the phonological output of syntactic structures. On this
basis, apparently right-peripheral elements are argued to occupy the position of
familiar topics in the left-periphery, the rest of the clause being moved to GroundP
via IP-inversion. The strict one-to-one mapping of syntax onto prosody and discourse is shown to also account for subjects in Bantu and Romance. Referential/
specific preverbal subjects are argued to be topics (hosted in the C-domain), while
indefinite/non-specific preverbal subjects occupy the canonical A-position. Each is
associated with particular intonation and interpretive effects. Postverbal subjects are
also argued to be of two types, one occupying a low A-position and resumed by an
expletive subject, the other occupying the Familiar Topic position and resumed by
a thematic pro. In both cases the word order is derived by remnant movement, and
distinct prosodic patterns obtain. Evidence is mainly drawn from a corpus study of
spoken Italian and a review of the Bantu literature (especially on Kinande).
Costa and Kula explore the syntactic and prosodic realization of new information focus in Bantu and Romance. They argue that syntax alone is insufficient to
encode information structure and determine its prosodic manifestations. Rather,
they propose that the syntax generates all possible structures, and that these are
then filtered out at the interface with the phonological component of the grammar.
Contrary to Frascarelli, they contend that the prosodic effects arising from information structure should not be tied to dedicated syntactic projections. According
to Costa and Kula, the locus of cross-linguistic variation lies in the way prosodic
prominence is assigned. In stress-based languages (Romance), a single position
of prominence may be identified syntactically; in tone-based languages (Bantu),
phonological phrasing determines where focus is assigned. The first section of the
paper presents arguments for a prosodic account of focus movement in Romance
(drawing essentially on evidence from Portuguese) and points out shortcomings of
syntactic account of these facts. The second section discusses evidence from Bantu
indicating that the position of focus is predictable from prosody, most notably at
the right edge of the phonological phrase.
Van der Wal investigates the status of thetic, or non-topical, presentational
constructions with post-verbal subjects in Bantu and Romance. She shows that
type 1 languages (Sesotho and French) exhibit default (expletive) “agreement”
in thetic constructions, whereas type 2 languages (Makhuwa and Italian) exhibit
verbal agreement with the postverbal subject. Previous treatment of these facts
attributed these differences to different parameter settings of the Agree system
(e.g., Collins 2004; Carstens 2005; Baker 2008), but van der Wal shows that these



The Bantu-Romance connection: A comparative investigation

proposals cannot account for the Makhuwa facts. Rather, she shows how issues
involving Case, Binding Theory and Information Structure demonstrate that these
language differences are due to the status of subject-verb agreement: In the type 1
languages she suggests that subject-verb “agreement” is actually pronominal, as
evidenced by the use of the “expletive” class 17 marker in Sesotho and the obligatory subject clitic in French. In contrast, subject agreement in the type 2 languages
shares agreement features with the postverbal subject. The paper presents a nice
treatment of comparative Bantu prosodic structure to demonstrate the different
position of the logical subject with respect to the VP, laying the groundwork for
further comparative research in this area.
In sum, this volume has wide empirical coverage across Bantu and Romance
languages, providing some support for universalist claims regarding: the status
of subject and object clitics (Marten, Kempson & Bouzouita), the order of insertion for DO and IO clitics (Cardinaletti), the syntactic restrictions on the nature
of Inflection (Harford), and the existence of a continuum of graded rather than
discrete notions of clitic vs. affix (Labelle). Several of the papers also use data
from both Bantu and Romance languages to explore current theoretical issues,
especially relating to the understanding of the Agree relation. Some suggest that
Agree is not sufficient to account for feature sharing in Nominal Expressions, and
should be complemented with Concord (Giusti). Others suggest that Agree should
be rethought in light of the lack of automatic correlation between unvalued and
uninterpretable features (Zamparelli). Still others propose that its current formulation nicely accounts for Bantu/Romance similarities, though the details of the
analysis vary (Carstens, Ferrari-Bridgers). Finally, several of the papers contribute
to current theoretical debates in the field, such as the role played by syntax in the
overall architecture of grammar. Some suggest that it is the syntax that determines
discourse structure and its prosodic reflexes (Frascarelli), whereas others argue
that interfaces between syntax and discourse and/or syntax and phonology (Costa
& Kula) as well as the pronominal/agreement status of subject clitics (van der Wal)
play a more prominent role. It is hoped that these contributions, with their new
data and comparative Bantu/Romance analysis, will stimulate further debate about
the nature of grammar.
Cécile De Cat
Leeds, UK
April 2008

Katherine Demuth
Providence, RI

Introduction 

References
Baker, M. 2008. The Syntax of Agreement and Concord. Cambridge: CUP.
Cann, R., Kempson, R. & Marten, L. 2005. The Dynamics of Language. Oxford: Elsevier.
Carstens, V. 2005. Agree and EPP in Bantu. Natural Language & Linguistic Theory 23: 219–279.
Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by Step: Essays on Minimalist
Syntax in Honor of Howard Lasnik, H. Lasnik, R. Martin, D. Michaels & J. Uriagereka (Eds),
89–156. Cambridge MA: The MIT Press.
Chomsky, N. 2001. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz (Ed.),
1–52. Cambridge MA: The MIT Press.
Cinque, G. 2005. On deriving Greenberg’s universal 20. Linguistic Inquiry 36(3): 315–332.
Collins, C. 2004. The agreement parameter. In Triggers, A. Breitbarth & H. van Riemsdijk (Eds),
115–136. Berlin: Mouton de Gruyter.
Nurse, D. & Philippson, G. 2003. The Bantu Languages. London: Routledge.

part 1

Clitics and agreement

Concepts of structural underspecification
in Bantu and Romance
Lutz Marten

Department of Africa, School of Oriental and African Studies,
University of London

Ruth Kempson & Miriam Bouzouita

Philosophy Department, King’s College, London

The paper explores parallelisms between Bantu (specifically Otjiherero) and
Romance (through Latin and Spanish) with respect to left and right peripheries,
and subject and object clitics. The analysis is formulated in Dynamic Syntax (DS,
Cann et al. 2005) and centrally involves notions of structural underspecification.
Through providing detailed analyses of different word order possibilities in the
Bantu and Romance languages discussed, we show how DS concepts of structural
growth over initially underspecified tree relations, such as the building of linked
structures and unfixed nodes, provide a uniform basis for analyses of word
order variation across the two language groups. We then extend our analysis to
include Bantu subject/object markers, which we analyze by employing the same
formal tools as used in the analysis of Romance (object) clitics, namely unfixed
nodes which have to be construed within a tightly locally restricted domain.
Empirical support for our analysis comes from restrictions on the presence of
object markers in passive and locative inversion constructions in Otjiherero,
which we show to follow from independent constraints of the availability of
unfixed nodes within a given domain. The analyses of Bantu and Romance
presented show that despite differences in surface morphology between the
two language groups, both exhibit a striking parallel with respect to the way
lexical information and general structure building principles of DS interact. The
difference between Romance clitic systems and the agglutinative morphology
of Bantu subject and object makers is thus seen to be comparatively superficial,
while the DS analysis brings out the strong structural parallelism between the
two language groups.1

. We are grateful to Jekura Kavari, Nancy Kula, Clara Simango, Nhlanhla Thwala and audiences in Groningen, Leiden, Oxford and Leeds for helpful comments on parts of this paper,
as well as to Helsinki University Library for granting access to their Swahili corpus. Parts of



Lutz Marten, Ruth Kempson & Miriam Bouzouita

1. Introduction
In this chapter, we present an explanation of structural parallels between Bantu
and Romance from the perspective of Dynamic Syntax, a grammar formalism
which argues that natural language syntax can be explained as humans’ ability to
build structured representations of meaning from words in context, on a left-toright basis. The model thus purports to reflect both natural language structures
in relation to the linear order of words, and the way information progressively
emerges during the incremental parsing of those words. Given this perspective,
left and right periphery constructions are of specific interest, and we begin by illustrating similarities of periphery constructions in Romance and Bantu. In Section 2,
we introduce the DS analysis of the left and right periphery in more detail, with
reference to Spanish and Latin, based on previous DS analyses. Section 3 shows
how the DS analysis of Romance can be extended to left and right periphery constructions in Bantu, which we illustrate from Otjiherero, Nsenga, siSwati, Swahili
and Tumbuka, and demonstrates how the DS model provides a straightforward
analysis of these constructions, without stipulation of construction-specific principles, which extends equally to Romance and Bantu. In the following Section 4,
we probe further the parallelism between Romance and Bantu, and the formal
space in which this is expressed in the DS model. We develop an analysis of Bantu
subject agreement markers along the lines of Romance (object) clitics, which in
our analysis reflect the more liberal word-order of Latin, and in particular local
scrambling.
Employing formal DS concepts, we construct an analysis of Bantu subject (and
object) markers, which allows them freedom of construal within a very tightly
locally restricted domain – what we call local underspecification. This analysis
contradicts the view of Bantu verbal structure as morphologically fixed, but has
the advantage of offering a principled analysis of subject and object-marking restrictions in passive and locative inversion constructions, which we illustrate from
Otjiherero (Section 5). Our argument thus starts from the more familiar left and
right periphery analyses of Romance, and proceeds to show how they are matched
in Bantu. In a second step, we focus on a less obvious parallel, namely the type of
process induced by Bantu (subject) agreement markers and Romance clitics (those
that are syncretic in form), and show that this parallel can be brought out through
the DS formalization of local underspecification, with the shared restrictions of
locative inversion and passive as evidence.

the research reported in this paper have received financial support from the AHRC (B/RG/
AN8675/APN16312) which is hereby gratefully acknowledged.

Concepts of structural underspecification in Bantu and Romance

1.1 Romance-Bantu similarities on the left and right periphery
It has often been proposed that Bantu and Romance share many structural characteristics despite differences in morphology. For example, objects can be fronted,
with agreement and clitic doubling:2
(1) Alenje
njûchi zi-ná-wá-lum-a.
		 2.hunters 10.bees sm10-past-om2-bite-fv
		 ‘The hunters, bees bit them.’
(2) El coche, María lo compró.
		 the car Maria cl bought.3sg
		 ‘The car, Maria bought it.’

[Chichewa]

[Spanish]

Subjects can be postposed, with either a focus or backgrounding effect, or
when postposed to the right of the full verb phrase, characteristically associated
with contrast:
(3) Zi-ná-wá-lum-a
alenje
njûchi.
		 sm10-past-om2-bite-fv 2.hunters 10.bees
		 ‘They bit the hunters, the bees.’
(4) Canta muy bien María.
		 sing.3sg very well, Maria
		 ‘She sings very well, Maria.’

[Chichewa]

[Spanish]

There are also object-subject inversion effects in both language families, associated with contrastive interpretation:
(5) Alenje
zi-ná-wá-lum-a
njûchi.
		 2.hunters sm10-past-om2-bite-fv 10.bees
		 ‘The hunters, they bit them, the bees.’
(6) Un coche compró
María.
		 a car bought.3sg Maria
		 ‘A car, Maria bought.’

[Chichewa]

[Spanish]

. The term clitic doubling is here used non-technically to mean use of an NP with co-construed
clitic for both right-and left-peripheral NPs. We use the following ab­breviations in the glosses: a =
“personal a”; acc = accusative; appl = applicative; cl = clitic; fut = future; fv = final vowel;
hab = habitual; nom = nominative; om = object marker; pass = passive; pl = plural; sg =
singular; sm = subject marker; 1, 2, 3, = noun-class number. We are grateful to Jekura Kavari,
Clara Simango and Nhlanhla Thwala for providing the Otjiherero, Nsenga and siSwati examples.
Chichewa examples are from Bresnan and Mchombo (1987).





Lutz Marten, Ruth Kempson & Miriam Bouzouita

In all these cases there is variation. For example, fronted objects require doubling in most Bantu languages, while in Spanish, clitic doubling is obligatory only
if the subject intervenes between the object and the verb. However, despite variation both within and across the language families, the overall parallel between
Bantu and Romance is clear. Object agreement markers on the verb, whether an
identifiably separate clitic or an affix, result in increased word-order freedom; subjects and objects broadly differ in the referentiality implications associated with
co-occurrence of the full NP with the agreement marker; and the subject, if rightperipherally placed, may be placed either externally to the VP or, in some languages, immediately following the verb:
(7) Zi-ná-wá-lum-a
njûchi alenje.
		 sm10-past-om2-bite-fv 10.bees 2.hunters
		 ‘They bit them, the bees, the hunters.’
(8) Compró
María un coche.
		 bought.3sg Maria a car
		 ‘She bought, Maria, a car.’

[Chichewa]

[Spanish]

In both language families there is variation as to the extent to which “agreement” for non-subject expressions patterns alongside subject marking or more like
a regular anaphoric process, hence giving rise to referentiality effects. In Spanish,
for example, the dative construction behaves more like an agreement phenomenon
than an anaphoric linkage between clitic and full NP that it doubles; and in Bantu,
languages vary as to whether or not object marking, unlike subject marking, is
subject to a referentiality restriction, or is an invariant agreement-like device.
Such left and right periphery effects in Romance have been the subject of considerable research, both in orthodox frameworks where the work is well known
(Rizzi 1997 and others following; Monachesi 2005 in HPSG) and in the emergent
Dynamic Syntax framework (DS; Cann et al. 2005; Kempson et al. 2006, 2007);
and some work has been done extending these claims to the Bantu case (see Cocchi
2001 for a minimalist analysis; Cann et al. 2005; Marten 2007; Marten & Kempson
2002 for DS analyses). Certainly, the parallelism between the two language families
even down to the level of individual-language variation suggests the phenomena
should emerge as the consequence of interaction of general principles. The relevance of the DS perspective in this connection is the claim that syntax is no more
than the progressive construction of semantic representation. In particular, the
concept of building “unfixed” nodes as part of the ongoing construction process
which is the heart of the DS claim, defines a family of relatively weak relations; and
these provide the basis for an integrated characterization of left and right periphery
effects on a broad cross-linguistic basis (Kempson et al. 2007). There is in particular: (i) the construction of a highly restricted, structurally underspecified relation

Concepts of structural underspecification in Bantu and Romance

where a node is introduced as unfixed but necessitates update within an individual
predicate-argument array, the basis for local scrambling; (ii) the construction of
an unfixed node requiring update within an individual tree, the basis for long-distance dependency and long-distance scrambling; and (iii) the building of paired
(“linked”) trees whose relation to each other is not that of mother/daughter at all
but merely a relation of anaphoric connectedness. These three forms of structural
underspecification are put together with the fact that pronouns themselves only
provide a partial specification of node decorations, and so can be used as a basis
for unifying any weak structural relation to provide the necessary update that will
fix any such initially “unfixed” node. The result is an account of a range of intermediate effects such as clitic and pronoun doubling. With just the additional and
independently motivated assumption that pronoun decorations may vary as to
whether they allow further structural development of the node they decorate (the
DS account of expletives: Cann et al. 2005), a further range of intermediate effects
is obtained, in particular associated with expletives, and with doubling as displayed by Spanish datives. The overall result is that interactions between general
processes of anaphora construal and the construction and update of structural
relations provide a principled explanation of what are otherwise taken to be topic
and focus effects, while nevertheless providing an analysis which is sufficiently
fine-grained to provide a basis for the full range of cross-linguistic effects. Indeed
Kempson et al. (2006, 2007) argue that the various topic and focus effects should
be seen as grounded in such interactions rather than requiring these notions as
syntactic primitives. The account thus makes a bid to be explanatory in a way that
an account in terms of stipulating as many discrete formal structures (such as features, categories, or functional projections) as there are distinct patternings, as in
many current alternative analyses, fails to match (e.g., Rizzi 1997; Cardinaletti this
volume; Rivero forthcoming).
However, there are idiosyncracies associated with the interaction of object
marking and local clause-internal variation in Bantu not displayed in Romance,
which threaten this claim, suggesting that the commitment to a single set of universally available principles for inducing the relevant data may not be sustainable. First of all, in the majority of Bantu languages, there can only be one object
marker, whether direct or indirect object marking:
(9) Ngi-m-nik-e
kudla.
		 sm1sg-om1-give-past 15.food
		 ‘I gave him/her food.’
(10) Ngi-ku-nik-e
		 sm1sg-om15-give-past
		 ‘I gave it to Jabulani.’

Jabulani.
1.Jabulani

[siSwati]





Lutz Marten, Ruth Kempson & Miriam Bouzouita

(11)

*Ngi-ku-m-nik-e.

(12)

*Ngi-m-ku-nik-e.

Secondly, in the majority of Bantu languages where only one object clitic is
allowed, passives preclude the presence of object markers, even when the verb
is di-transitive, an equally puzzling restriction (although see Woolford 1995 for
exceptions):
(13) òmbápírà y-á-tjàng-èr-w-á
òvá-nátjè
		 9.letter
sm9-past-write-appl-pass-fv 2-children
		 ‘A letter was written to the children (by Katenaa).’

(í
(by

Kàténáà).
Katenaa)

(14) *òmbápírà y-é-và-tjàng-èr-w-á.
		 9.letter
sm9-past-om2-write-appl-pass-fv
		 Intended: ‘The letter was written to them.’

Thirdly, as is known from Bresnan and Kanerva (1989) and others, locatives
undergo inversion, and in so doing, induce “subject” agreement on the verb, suggesting that the concept of subject in Bantu is exceptional in that canonical subject
and the locative share a potential to act as “subject”:
(15) m-Òn-djúwó mw-á-hìtí
é-rúngà.
		 18–9-house
sm18-past-enter 5-thief
		 ‘Into the house entered the thief.’ (or, ‘There was a thief entering the house.’)

Finally there is the puzzle that unlike the canonical subject and like a passive
subject, such pre-posed locatives cannot co-occur with object markers:
(16) *m-on-djuwo
mw-a-ri-hiti.
		 18–9-house
sm18-past-om5-enter
		 Intended: ‘He/she entered the house’

This suggests that there may be cross-linguistically distinct bases for relations we
informally understand as subject and object, so that the concepts of subject and
object themselves continue to elude us even though they are so very familiar.
However, we shall argue to the contrary in this paper that the family of concepts of structural underspecification which are defined in DS as replacing concepts of movement equally apply to explain these apparent Bantu idiosyncracies,
while retaining a universalist methodology on the syntactic processes themselves.
In so doing, we shall provide an integrated analysis of Bantu passive and locative
inversion that is not available in other frameworks. We will argue that the Bantu
subject and object markers should be seen as associated with underspecified treerelations which are restricted to requiring resolution within a local domain, an
underspecification which, until resolved, debars any other such underspecified

Concepts of structural underspecification in Bantu and Romance

tree relation. Moreover, as we shall show, this turns out to constitute a further
parallelism with the Romance data. Kempson and Cann (2007) have argued
that the family of patterns displayed by Romance clitics involves the construction of locally underspecified tree relations as a reflex of earlier Latin scrambling,
an underspecification which is most clearly displayed by syncretic forms (e.g.,
Spanish me, te). The non-co-occurrence of such syncretic forms within an individual cluster is explained by the tree-logic restriction that only one token of a
structurally underspecified relation is possible at any one time.3 What we shall
argue is that exactly the same explanation carries over to the non-occurrence of
more than one Bantu object clitic, and indeed to the restriction on both passive
and locative inversion that object clitics are generally precluded. The consequence
is that these arguments will provide strong and novel endorsement of the parallelism between Romance and Bantu, while at the same time providing an integrated account of what have been taken to be heterogeneous and puzzling Bantu
data. Given that the concept of syntactic subject defined as a locally underspecified
tree-relation will be new both within and beyond the DS framework (though see
Wu 2005 for a related conclusion), we close by reflecting on what this tells us about
the concept of subject in general.
However, before setting out our analysis, we provide a short introduction to
DS in the following section, using a combination of Latin and Contemporary Castilian Spanish as the languages of illustration for the various concepts of paired
“linked” trees, unfixed tree relations and consequent characterizations of such
structures as Hanging Topic Left Dislocation, long-distance scrambling, pronoun
doubling, short-distance scrambling. We will also look at the account provided
within DS of the morphological template behavior of clitic clusters.4

2. The dynamics of language processing
The methodology implicit in the Dynamic Syntax formalism is to take the constructs used in semantics and define them as tree-structure representations with
. Potential complications for this account arise with se, which can co-occur with both me
and te and may give rise to clitic clusters (Cuervo 2002; Heap 2005). But there is reason to think
that in all such cases se is an ethical dative use, for which there is considerable evidence of its
requiring an independent adjunct analysis (Cuervo 2002).
. It is not possible in the space provided to give a full introduction to DS, nor to provide full
justification for the Romance analyses we discuss. See Cann et al. 2005; Kempson et al. 2006;
Kempson & Cann 2007; Bouzouita 2007; Bouzouita in preparation; Bouzouita & Kempson
2006; which present more detail.





Lutz Marten, Ruth Kempson & Miriam Bouzouita

a concomitant tree-growth process of establishing some interpretation in context.
The central claim is that syntactic properties of natural language reside exclusively
in the progressive growth of such tree-structure representations strictly following
the dynamics of left-right processing: no additional level of syntax is needed. The
type of tree-growth process we assume as the process of building up an interpretation for a string such as (17) involves a growth from just a single-node tree with
a requirement to build-up a propositional structure, as in the left below in (18)
through a sequence of transitions to yield a final tree, being the tree on the right
in (18):5
(17) María compró
un coche.
		 Maria bought.3sg a car
		 ‘Maria bought a car.’
(18)

[Spanish]

Parsing María compró un coche in (17)

		

Ty(t), Comprʹ(Cocheʹ)(Maríaʹ),

?Ty(t),

→

Maríaʹ,
Ty(e)

Comprʹ(Cocheʹ),
Ty (e → t)
Cocheʹ,
Comprʹ,
Ty(e) Ty(e → (e → t))

In example (18), diagrammatically displaying this process, a lot of implicit
content is packed into , which symbolizes the concept of tree growth. Central
to the concept of tree-growth is the concept of requirement: ?X for any decoration X.
Decorations on nodes such as ?Ty(t), ?Ty(e), ?Ty(e → t) etc. express requirements
to construct formulae of the appropriate type on the nodes so decorated (propositions, terms and predicates respectively), and these drive the subsequent treeconstruction process.6
The general dynamic is first to unfold a tree structure imposing such requirements following a mixture of top-down general tree-growth strategies and bottom-

. By convention, nodes decorated by functor types are on the right, nodes decorated by argument types are on the left. In all tree displays, we give only such tree decorations as are needed
to demonstrate the point in question.
. In any partial tree, there is one node indicated by a pointer, ◊, as the node under development. In this framework all noun phrase construals are taken to be of type e, matching arbitrary
names manipulated in natural deduction proofs. Accordingly, the terms onto which words map
are lambda terms within the epsilon calculus (the epsilon calculus provides the formal study
of arbitrary names, see Meyer-Viol 1995). In general, we ignore the internal structure to be assigned to such type e names.

Concepts of structural underspecification in Bantu and Romance

up lexical actions which contribute concepts and other aspects of structure, and
then compositionally to combine those concepts in a strictly bottom-up fashion to
yield the overall interpretation, in which all requirements must have been satisfied.
For any language, the process of growth of structure is strictly monotonic, and for
any one interpretation, hence wellformedness, there must be at least one sequence
of progressively enriched partial trees between input tree and resulting logical
form in which all requirements are met.
The formal system underpinning the partial trees which are constructed is a
logic of finite trees (LOFT: Blackburn and Meyer-Viol 1994). There are two basic
modalities, 〈↓〉 and 〈↑〉, such that 〈↓〉α holds at a node if α holds at its daughter;
and the inverse, 〈↑〉α, holds at a node if α holds at its mother. Functor and argument relations are distinguished by defining two types of daughter relation:
〈↓0〉 for argument daughters, 〈↓1〉 for functor daughters (with their inverses 〈↑0〉,
〈↑1〉). There is also an additional link operator, 〈L〉, which relates paired trees,
with a link relation from a node in one tree to the top node of another. This tree
language plays a critical role in defining the individual steps of tree growth; and
procedures are defined for step-wise building up of such structures either by computational actions or by lexical or even pragmatic actions. All are defined in the
same vocabulary, a set of context-relative actions for updating representations of
interpretation. Such formal tree languages by definition provide characterizations
of such structural relations as dominate; and in LOFT, the concept of dominate is
defined in the following terms: a node can be described as dominated by a node
Tn(a) when 〈↑∗〉Tn(a) holds at that node, that is when the node identified as Tn(a)
is along some sequence of mother relations from the present node.
Such structural relations will play an important part in what follows, but we start
with the characterization of how predicates are built from lexical specifications.
2.1 Lexical information provided by verbs
As in other frameworks, verbs are the major projector of structure, for which
actions are defined that induce some or even all of the propositional template they
express.
(19)

Result of running lexical actions of compró

					

?Ty(t)Tns()
U,Ty(e),
? x.Fo(x),

?Ty(e → t)
?Ty(e)

Ty(e → (e → t)),
Comprʹ,[↓]⊥





Lutz Marten, Ruth Kempson & Miriam Bouzouita

In Spanish, a subject pro-drop language, verbs project full propositional
structure, in which: (i) a subject argument node is decorated with a placeholder
i.e., a metavariable of the form U,V, ... of type e, which stands for some value to be
assigned either from context or from the construction process; and (ii) an object
node which requires subsequent development, as expressed by the requirement
?Ty(e) in (19).
Taking first the subject node decoration, these meta-variables all have an
associated requirement, ?∃x.Fo(x) which guarantees they must be updated. The
pointer is positioned at this node, since this allows the subject to be identified
contextually as the very next step. Strings made up of either a SVO sequence or a
VSO sequence are thus taken to be unmarked (we shall see in the next section how
left-peripheral expressions may be treated as decorating distinct nodes, external
to the propositional structure itself and identified solely through replacement of
the meta-variable).7
The effect is that verbs induce a sequence of actions which might equivalently
be expressed by a pairing of verb plus subject pronoun. The decoration of argument nodes with a meta-variable, for example, is the intrinsic property of pronouns, underspecification with respect to content being their hallmark. Whether
provided by parsing a pronoun or a pro-drop verb, all such place-holding devices
must be supplied with an assigned value (notice the requirement for a full formula
?∃x.Fo(x)); and different types of anaphoric expression can be defined according
to the different constraints on that process which they impose. Values for reflexive
pronouns have to be updated within a given single predicate-argument structure,
values for other pronouns outside such locally defined structure. Meta-variables
projected as part of the intrinsic specification of the verb, on the other hand, lack
any such restriction, and can have their value identified either locally or nonlocally, indexically, or from some term provided later on in the construction process,
where the choice of value involves general cognitive constraints such as relevance
(Sperber & Wilson 1995). The object node, on the other hand (the other node
induced by the verb’s actions in (19)), has only a requirement that the node be
developed so further linguistic input is essential (with no provided meta-variable,
there is no license to use context-provided values).
By virtue of the grounding in LOFT, trees are not however taken as syntactic
primitives as in other frameworks, but are built up by explicit tree-growth procedures, defined as actions which induce the structure in (19), and then feed into

. As Zagona (2002: 27) points out, the acceptability judgments for the VSO order in finite
declaratives vary from speaker to speaker, some reporting it as archaic.

Concepts of structural underspecification in Bantu and Romance

whatever structural or pragmatic operations are suitably triggered. The so-called
“bottom-node restriction”, [↓]⊥, which is part of the decoration of the predicate
node in (19), is definitive of regular lexical specifications that the node to which
the formula decoration is provided constitute a terminal node in the resulting tree,
closing off development of the node so decorated, so that no further expansion is
possible. As we shall see, this restriction may get lost in some words.
2.2 Context-dependence and lexical specifications for pronouns
The process of providing such a meta-variable with a value is a substitution process,
its replacement having to be selected from the set of terms made available during
the construction process, characteristically from the context. For example, in (20),
lo is naturally understood as picking out the same individual as picked out by the
antecedent use of Pablo:
(20) ¿Quién ama
a Pablo? María lo ama.
		 who
love.3sg A Pablo Maria cl love.3sg
		 ‘Who loves Pablo? Maria loves him.’

The concept of the context-dependence of anaphoric expressions in language
is familiar enough. What is less orthodox is the assumption that it is to be defined
as a tree-update process; and with contexts also represented as (partial) trees,
anaphora resolution can apply equally to the update of a pronoun from antecedent
terms within the structure under construction:
(21) Pablo cree
que María lo ama.
		 Pablo think.3sg that Maria cl love.3sg
		 ‘Pablo thinks that Maria loves him.’

For the identification of lo in (21), the context relative to which that interpretation process takes place includes the partial structure containing the subject node
with its decorations. So there is no distinction between grammar-internal and
discourse uses of pronouns: both are analyzed alike as a tree update process in
which the meta-variable gets to be provided with a term as value by substitution.
In canonical uses of pronouns, like all other content words, the pronoun has ‘the
bottom restriction’ that whatever value it is assigned must be taken as decorating
a terminal node in the tree.
However, by adopting a representationalist stance, with pronoun construal
seen as a substitution process, we can integrate anaphoric and apparently nonanaphoric uses of pronouns into a single form of explanation. In Spanish, as
already briefly discussed above, there is widespread clitic doubling of dative expressions which occur either pre- or post-verbally (e.g., the pre-verbal dative NP





Lutz Marten, Ruth Kempson & Miriam Bouzouita

is doubled by les in (22)), sometimes referred to as an agreement phenomenon
(Franco 2000 inter alia):
(22) A familias de pocos medios les ofrecieron queso.
		 to families of small means cl offered.3pl cheese
		 ‘To low-income families, they offered cheese.’

From a DS perspective, such clitic uses can be seen as anaphoric devices of just the
same sort as more regular pronouns, with just a minor loosening of the defining
properties of what it is to correspond to a regular lexical item of the language:
they have lost this bottom restriction, with the result that the meta-variable they
provide, in (23) represented as U, can be updated by structure as part of the
construction process.8 As a result, these dative clitics have broader distribution,
allowing identification either from context or from the construction process, retaining their anaphoric properties but in a modified form.
(23)

Parsing A familias de pocos medios les ofrecieron queso in (22)

			

?Ty(t)
?Ty(e → t)

Funcionariosʹ
Quesoʹ

?Ty(e → (e → t)),
FPMʹ,
U

Ofrecʹ

The effect, however, is that doubling in such cases will not impose a referentiality constraint, as the meta-variable provided by the clitic can be given a value by
any term whatever of appropriate type, even allowing indubitably quantified NPs
to be doubled:
(24) A nadie le devolvió
María su
manuscrito
		 to nobody cl returned.3sg Maria his/her manuscript
		 ‘Maria didn’t return anyone their manuscript.’

This renders them equivalent to the decorations provided by the verb for its
subject in subject pro-drop languages, hence their supposed agreement-displaying
properties.

. Details of individual terms are omitted, including the internal structure of the composite term projected from familias de pocos medios, diagrammatically represented as projecting the predicate FPM'.
We assume that the subject position is annotated with Funcionarios', supplied from the context.

Concepts of structural underspecification in Bantu and Romance

2.3 The dynamics of long-distance dependency
The dynamics of initially constructing some incomplete specification and subsequently providing its update apply also to the projection of structure. The core
claim of DS is that all syntactic mechanisms can be seen in terms of tree growth
and update, and in particular, discontinuity phenomena are modelled by defining
structural concepts of underspecification and update as a direct analogue of the
formula underspecification and update, taken to be the underpinning of anaphora
resolution. In (26), for example the expression un coche is construed as providing
a term for the resulting logical form, but the node which it decorates does not yet
have its relation to the root, Tn(0), fixed within the overall structure (indicated by
the dashed line):9
(25) un coche compró
		 a
car
bought.3sg
		 ‘A car, he/she bought.’
(26)
		

Parsing un coche in (25)
Tn(0), ?Ty(t),
Cocheʹ,
·≠*ÒTn(0),
? x.Tn(x)

Once this unfixed node is decorated, the actions of the verb can be used to project
a full template of propositional structure; and, with the subject having been contextually identified, say as someone called Maria, the pointer returns to the object
node for further development. This move provides the necessary input for fixing
the unfixed node, with this delayed update also solving the subcategorization requirement of the two-place predicate projected by the verb:
(27)

Parsing un coche compró in (25)
?Ty(t),Tn(0)

		
Cocheʹ ,
·≠*ÒTn(0),
? x.Tn(x)

?Ty(e Æ t)

Marίaʹ,
Ty(e)
?Ty(e ),

Comprʹ

. Formally, this is defined using the Kleene star operation defined on the daughter-mother
relation: 〈↑∗〉Tn(a) is a node dominated by a node Tn(a), where Tn(a) is along an arbitrary sequence of mother relations from the current node to Tn(a). Adding the requirement ?∃x.Tn(x)





Lutz Marten, Ruth Kempson & Miriam Bouzouita

In addition, given that the only restriction on this unification process is that
the result be an update, we expect that it can also apply in the presence of a pronoun
decorating the object node, as long as that pronoun has no bottom restriction. So
we predict the availability of long-distance dependency effects in Spanish with a
preposed dative expression and dative doubling, as in (22): the so-called Clitic Left
Dislocation (CLLD) phenomenon.
Like the process of anaphora resolution, the process of constructing an unfixed
node is in principle available at any point, allowing the construction of an unfixed
node at some relatively late point in the parse. There is an asymmetry between early
and late application of such construction processes, however, by virtue of the tree
either being radically underspecified (early in a parse) or all but complete (late on in
the parse process). In particular, DS assumptions lead us to expect a more restrictive
result once an emergent tree is complete apart from completing the formula decorations on its nodes. General restrictions on tree development dictate that once the
pointer is at a node with some imposed requirement(s), it cannot move up from
that node except as particular licence is defined to enable it do so. One such case is
in the parsing of pronouns. In parsing pronouns, although the formula value may
remain to be provided at some late stage, a type value is established, and so the construction process can proceed. However, the delay in completing the decorations
on the node which this gives rise to has to be essentially local, as a strict bottomup compositionality requirement on the containing predicate-argument structure
must nevertheless be satisfied. In consequence, though the pointer can move away
from a type-decorated node as long as it has some interim place-holder decoration,
it will have to return to the partially decorated node so that a fixed value for that
node is provided when the semantic information on the tree is compiled. This late
construction of a subtree can be achieved by a process of building an unfixed node
of a type that matches the node from which it is built, decorating it suitably, and
then unifying the two (so-called Late*Adjunction). This sequence of steps will, for
example, apply in deriving a construal for (28):
(28) Les ofrecieron queso a familias de pocos medios.
		 cl offered.3pl cheese to families of small means
		 ‘They offered cheese to low-income families.’

With the clitic pronoun decorating a third argument node of ofrecieron early
on in the parse, Late*Adjunction can induce a node of type e (Ty(e)) for the parse

as an additional decoration on that node imposes the require­ment that in all successful completions of the tree, this underspecified characterization is replaced by a fixed tree relation. Cf.
Kaplan & Zaenen (1989) for use of the Kleene star in defining the related concept of “functional
uncertainty” in LFG.

Concepts of structural underspecification in Bantu and Romance

of a familias de pocos medios which, once established, unifies with that node. The
result of this introduction and decoration of an unfixed node at this point in the
derivation yields as a consequence the very strict locality restriction at the right
periphery (sometimes called “The Right Roof Constraint”), which, although well
known, has never been satisfactorily explained. As a bonus, this account provides
a basis for analyzing the inversion phenomenon characteristic of the Romance
languages:
(29)

un coche, compró

María.

		 a car
bought.3sg Maria
		 ‘A car, Maria bought.’

Early on in the derivation, an unfixed node is introduced, decorated by the
parse of un coche. This weak tree relation is not updated immediately but only after
the verb has been processed, hence with a delay in its construal. The projection
of content for the subject node is also delayed, this node only being introduced
as part of the information provided by the verb, with its content provided once
a predicate has been compiled from verb and object in combination. Once the
pointer has returned to the subject node, Late*Adjunction can then be used to
provide an intervening node for parsing the end-placed subject expression and so
establishing the construal of that subject.
(30)

Parsing Un coche compró María in (29)

		

?Ty(t)
U,Ty(e),
? x.Fo(x),
ʹ ʹ
Maria

Comprʹ(Cocheʹ),
Ty(e Æ t)
Cocheʹ

Comprʹ

This finally leads to a tree no different from that which could have been
derived by parsing (17) (see the the right-hand side tree in (18)). As mentioned
above, wellformedness in DS is not defined by the final tree of a derivation, but
rather by the monotonic transition from an initial to a final tree.
There is, however, an important constraint, which is an immediate consequence
of the concept of partial trees. Just like any individual fixed node, any unfixed node
is nevertheless identified by its tree-node value – in these very weak cases solely by
the dominate relation that defines it. But this means that there can effectively never
be more than one unfixed node at a time, because any two such identified nodes
cannot be distinguished, and so would lead to incompatible tree decorations. The
derivation of (29) notably meets this constraint: at no point in the sequence of trees
over which its interpretation is built up is there ever more than one unfixed node.





Lutz Marten, Ruth Kempson & Miriam Bouzouita

2.4 Constructing trees in tandem
The one missing part of the DS jigsaw of interacting mechanisms is the details of
how paired so-called ‘linked’ trees are constructed. These are licensed by a mechanism which induces paired trees on the assumption that they must share a term
which occurs in both. The two trees are, in effect, introduced in tandem, using
LOFT modal requirements to ensure an anaphoric substitution process across
the pair of trees. Relative clause construal provides a core example (see Kempson
et al. 2001; Cann et al. 2005). But this very same device applies equally to provide a
basis for Hanging Topic Left Dislocation (HTLD) structures (Anagnostopoulou et al.
1997). The left-peripheral expression is analyzed as providing the trigger for introducing a tree linked to the rootnode to be of type e, which that expression duly decorates. There is an additional restriction that the root node with requirement ?Ty(t)
must now be constrained to contain a copy of the term projected from the left-peripheral noun phrase so as to satisfy the sharing-of-terms requirement dictated by the
LINK relation. This modal form of requirement determines the presence of a suitably
construed pronoun in the twinned structure, which the facts of HTLD corroborate:
(31) Este coche, Pablo lo compró
		 this car
Pablo cl bought.3sg
		 ‘This car, Pablo bought it.’
(32) *Este coche, Pablo compró
			 this car Pablo bought.3sg

The only exception to this required presence of a lexical pronoun is the case
of subject pro-drop structures: i.e., just those cases where the verb projects its argument nodes decorated with only a meta-variable as formula value, exactly as
though a lexical form of pronoun had been present, since these can satisfy that
same requirement by identifying the meta-variable appropriately.
The result is that a range of strategies is available in the opening stages of a
parsing process, any one of which can apply. This gives us a basis for explaining the
blurring effect associated with the subject position in pro-drop languages: that it is
able to be construed either as a backgrounding device, or as a focussed term, or more
neutrally, with concomitant difficulty in some cases of differentiating what have been
distinguished as HTLD and CLLD structures (De Cat 2007). In particular we expect
both the availability of building an unfixed node, decorated, and presumed to be
incorporated into the single emergent structure as it unfolds, and the building of a
pair of linked structures, with the second structure suitably construed as having an
interpretation dictated by the decoration on the first, as shown below.
It is notable that, if these are the only two options available for parsing preverbal subject expressions in Spanish, we derive the result observed in Zubizarreta
(2001) that Spanish subjects are invariably in some sense external to the clause:

Concepts of structural underspecification in Bantu and Romance

(33) Martín lo escribió.
		 Martin cl wrote.3sg
		 ‘Martin wrote it.’
(34)

Parsing Martín (a) as decorating an unfixed node or (b) as a linked structure

		 (a)

Tn(0), ?Ty(t),

Martίnʹ, ·≠*ÒTn(0)

(b)
Martίnʹ,
·LÒTn(0)

·L-1ÒTn(n), ?Ty(t)
? ·Ø*ÒMartίnʹ,

The two strategies of building linked structures or an unfixed node within an
individual structure are in principle also available at the right periphery, though
to rather different effect. If an independent linked structure is constructed, it will
be the value assigned to the meta-variable of the pronoun that is identified from
context: the construal of the end-placed NP will then have to be fixed to be coextensive with value of the meta-variable in order to match the linked-structure
requirement of shared term in the paired structures. Hence the reported backgrounding effect that can be conveyed by so-called Pronoun Doubling. Alternatively, an unfixed node can be introduced by Late*Adjunction whose decoration
provides a value for the meta-variable. This typically gives rise to contrastive or
new information effects by virtue of that late construction (see Kempson & Cann
2007). This predicted flexibility at both left and right peripheries is a bonus of
the parsing perspective since, unlike in more conventional grammar formalisms,
there is no commitment to a single assignment of structure for an unambiguous
string. On the contrary, this framework provides a range of tree-growth strategies,
which may feed each other, giving rise to a mixed array of effects.
Taking a step back from the details, what is striking in the set of explanations which this account makes possible is how few stipulations there are: no
single identifiable structure is defined by a mechanism individual to that structure.
Instead we have general principles of tree development interacting with general
principles for anaphora construal: it is these together that determine the range of
effects associated with HTLD, CLLD, Pronoun Doubling, expletives and so on.
2.5 Scrambling and locality constraints on structural under-specification
Locality constraints on actions for tree-growth are not all merely an epiphenomenon
of the stage in the construction process at which update must have taken place. In
free word-order languages, in particular, there is evidence of local processes of tree
construction in anticipation of the verb. To express this in a principled way, we push
the parallelism with anaphora resolution yet further, and extend the articulation
of different locality restrictions to structural processes of tree growth, articulating
analogous restrictions on update of tree growth, defining a sub-type of structural





Lutz Marten, Ruth Kempson & Miriam Bouzouita

underspecification which requires update within a single propositional domain (socalled Local*Adjunction). Local*Adjunction applies to a type-t-requiring node. It
involves the construction of an unfixed node which, in effect, has to be updated to
a fixed relation within a given local scope domain. What the rule induces is one
fixed argument daughter node immediately dominated by a node whose relation
to the node of introduction is an underspecified relation across functor relations
∗
(〈↑1〉 picks out the functor spine along which argument nodes can be constructed – the
unfixed tree-relation is diagrammatically indicated by a dashed line):
(35)
		

The effect of Local*Adjunction
Tn(a), ... ?Ty(t)

·≠1*ÒTn(a)

·≠0Ò·≠1*ÒTn(a),
?Ty(e), ? x.Tn(x),

Proposition node

Unfixed functor node

Argument node

The node introduced by these actions has a requirement for an argument term
(of type e), a description of its tree relation to the point of departure, and a requirement for a fixed value. As we shall see, this rule is used to induce structure for local
scrambling effects. To see the general application of this, we need to revert to Latin,
the source language from which Romance languages developed, as this displays the
relatively free word order variation symptomatic of free use of Local*Adjunction,
applying in conjunction with constructive use of case. What the case specification
ensures is immediate update of any such unfixed node, fixing the structural relation of the node decorated by the expression well before the occurrence of the verb,
thereby allowing the rule to apply again to introduce a further unfixed node. So as
in (36), we can license the building of structure from first servum and then Xerxes:
(36) Serv-um Xerxes
cecidit
		 slave-acc Xerxes.nom killed.3sg
		 ‘Xerxes killed the slave.’

This sequence of words can be in any order. This is where the restriction
imposed by the system that there be only one unfixed structural relation of a
type at a time has a role to play, since any duplication of the process without
any such update would lead to immediate collapse of the two unfixed tree relations, yielding just one argument node for a predicate simultaneously characterized as a subject and an object argument. The effect of case as a constructive
mechanism for fixing any such underspecified tree relation is therefore essential

Concepts of structural underspecification in Bantu and Romance

to the effective re-use of Local*Adjunction as a strategy.10 Of course, once any
one relation is fixed, (as in (37b) below), another unfixed node can be introduced, following through on the same sequence of actions; and the interim
result (37d) – is a partial tree with a set of argument nodes but as yet lacking any
predicate node with which to combine. Notice in particular the pair of argument
nodes (without having parsed a verb) in (37d), a pattern reflected by the later
clitic clusters that emerged in the Romance languages. The verb then follows,
filling out the remainder of the propositional structure to yield the appropriate
output tree with Xerxes' as subject argument Serv' as object argument to the
predicate Caed'. Unlike two case-distinguished unfixed nodes, either subject or
object nodes induced by actions of the verb harmlessly collapse with those introduced as unfixed and updated through constructive use of case (Nordlinger
1998), because annotations provided by the verb are compatible with those provided by computational actions used in parsing the NPs: the formula decorations
provided on the verb-induced argument nodes are meta-variables, compatible
with all formula updates. So the tree projected by the lexical actions of cecidit
can be constructed by applying those actions to the tree (37d). This allows “free”
word order effects without any necessary interpretational difference, with pragmatic constraints free to determine preferential orderings.
(37)

(a)

Locally unfixed node

(b)

Tn(0), ... ?Ty(t)

(c)

Parsing Serv-um
Tn(0), ... ?Ty(t),

·≠1*ÒTn(0)

·≠1ÒTn(0)
?Ty(e Æ t),

·≠0Ò·≠1*ÒTn(0),
?Ty(e),

·≠0Ò·≠1ÒTn(0),
Ty(e), Servʹ

Locally unfixed node
Tn(0), ?Ty(t)
·≠1*ÒTn(0)

·≠0Ò·≠1*ÒTn(0),
?Ty(e),

·≠1ÒTn(0),
?Ty(e Æ t)

·≠0Ò·≠1ÒTn(0),
Ty(e), Servʹ

(d)

Parsing Xerxes
Tn(0), ?Ty(t),

·≠0ÒTn(0),
Ty(e), Xerxesʹ

·≠1ÒTn(0),
?Ty(e Æ t)

·≠0Ò·≠1ÒTn(0),
Ty(e), Servʹ

. Instead of, or in addition to, case specifications there may be pragmatic and prosodic clues,
which are, however, likely to be less secure.





Lutz Marten, Ruth Kempson & Miriam Bouzouita

In the modern Romance languages, with case no longer being expressible in
the morphology of the full NPs, the effects of the actions of Local*Adjunction
are now restricted to its use with clitic pronouns.11 Firstly, in finite clauses, they
invariably occur in an early position, prior to the verb, exactly as the effects of
multiple applications of Local*Adjunction plus case update (see Bouzouita 2007,
in preparation; Bouzouita & Kempson 2006; and Kempson & Cann 2007 for a
diachronic account of clitic placement). Furthermore, being stored as a lexical
device, each is associated with a distinct scrambling mechanism. For example,
the third person accusative clitics fix their structural relation to the dominating
type-t-requiring node immediately. However, many of the clitics display syncretic properties, e.g., French me, te and Castilian Spanish le. These fail to identify whether the argument introduced is a direct or an indirect object, and are
assigned lexical actions which induce the construction of a locally unfixed node
without immediate update. The syncretic form thus matches the weakeness
of the update they provide. It is these actions of which the agreement forms of
Bantu are redolent; so we will return to this type of update action in due course.
The characteristic rigid pre-verbal positioning, allowing only other clitics to
intervene between the clitic and the verb, echoes the multiple applications of
Local*Adjunction plus update displayed in scrambling, a grouping which over
time became calcified in the lexicon as a single look-up, hence eventually a single
lexical entry. In some languages, e.g., Italian, these are written as one word, glielo;
in other languages, e.g., Spanish, they may become associated with idiosyncratic
and noncompositional forms of interpretation.12 (This effect is missing from the
Bantu languages where the almost invariant case syncretism has not led to any
such multiple clustering.) The composite effect achieved by the Dynamic Syntax
analysis is a characterization of clitic template phenomena as a lexicalization
of the earlier free word order system. The update actions which had licensed
flexibility of NP ordering in the earlier Latin system, where they were freely
available, are now lexically associated with individual clitics or clitic clusters.
In this view, it is the retention of case specifications only in the lexicon with the
clitic pronouns that ensured that the effects of Local*Adjunction are reflected
in clitic placement. Furthermore, because these are now all individual lexical
stipulations, each can only reflect one of the possible mechanisms which give
rise to scrambling effects.

. We leave on one side whether enclitic forms are directly lexically specified or in­duced via
Local*Adjunction.
.

“I gave it to them.” or “I gave them to him/her.” or “I gave them to them.”

Concepts of structural underspecification in Bantu and Romance

3. Left and right periphery effects in Bantu
The structural possibilities at the right and the left periphery available in Romance
are mirrored by similar structures in Bantu, as already shown in Section 1. We can
thus use the Romance analyses of the previous sections for developing analyses of
similar structures in Bantu, and so in this section, we give a cross-Bantu survey
to show how the range of variation encountered is similar to variation across
Romance. For instance, the use of *Adjunction and LINK is exploited in Bantu as
well, given the optional doubling phenomenon at least with object markers; and,
as in Romance, these general mechanisms interact with the tree-update actions
which the subject and object markers provide (see Marten 2007). Bantu subject
and object markers are sometimes referred to as agreement markers, but are better
analyzed from a semantic perspective as quasi-independent pronominal elements
(e.g., Bresnan & Mchombo 1987; Marten & Kempson 2002). They behave like
pronouns in terms of their referential properties much as Romance clitics, and
although they appear to be more morphologized in Bantu than in Romance in
terms of positional restrictions, we will argue in the following section that, like
the Romance clitics, they should be analyzed by employing the concept of locally
unfixed node, but with greater systematicity, hence demanding a more general
form of explanation. At the left periphery, in some languages (such as Tumbuka (38)),
object-argument nodes can be introduced through *Adjunction without co-referring object marker; although, in many languages, the more common, or sometimes only, strategy is to have an object marker (39):
(38) Ngóoma ti-zamu-limilír-a
namchéero.
		 9.maize sm1pl-fut-weed-fv tomorrow
		 ‘Maize we will weed tomorrow.’ (Downing 2006: 62)
(39) Q: Ba-ntfwana, ba-ba-nik-e-ni?
			
2-children sm2-om2-give-past-what
			
‘What did they give to the children?’

[Tumbuka]

[siSwati]

		 A: Tin-cwadzi, ba-ti-nik-e
ba-ntfwana.
			
10-books
sm2-om10-give-past
2-children
			
‘Books, they gave (them) to the children.’

In the siSwati example, tincwadzi ‘books’ is projected onto a linked structure, hence the object marking. In addition, as we will see shortly, siSwati object
markers have not lost their bottom restriction, indicative of the node having to be
a terminal node in the resulting tree. At the right periphery, subject and object expressions are found, and the phenomenon of clitics losing their bottom restriction
may seem to have extended further than in Romance since, in some languages, the





Lutz Marten, Ruth Kempson & Miriam Bouzouita

doubling phenomenon is generalised to all object clitics, and even in some cases
internally to the structure under construction (rather than between pairs of trees).
First, there is the distribution which is widespread in Romance: in siSwati, object
expressions tend to occur with object clitics only when they are right dislocated,
as the adverb placement in (40) indicates. But in Swahili, object clitics can occur
with co-referential objects internally to the verb-phrase sequence, indicating that
in Swahili object clitics have lost the bottom restriction:
(40) Ng-a-yi-bon-a
kahle inja.
		 sm1sg-past-om10-see-fv well 10.dog
		 ‘I saw the dog well.’
(41) *Ng-a-yi-bon-a
(42)
		
		
		
		

inja

[siSwati]

kahle.

Gidyoni a-li-kuwa
h-a-ja-mw-on-a
huyo ki-jana
Gidyoni sm1-past-be neg-sm1-perf-om1-see-fv dem 17-youth
vizuri.
well
‘Gidyoni had not seen the youth well.’ (Mvungi n.d.: 126)

[Swahili]

However, in apparent marked contrast to the Romance pattern, many Bantu languages have a restriction on the number of object markers available in the verbal
template. Thus languages like siSwati, Swahili and Otjiherero allow only one object
marker per verb, although multiple object markers are found, for example, in Tswana,
Rundi and Kinyarwanda. While the restriction on the number of object markers is
often taken as morphological, we will argue below that it is in fact syntactic, in that
object markers project locally unfixed nodes, from which the restriction to only
one object marker at a time follows immediately.13 The analysis involving locally
unfixed nodes predicts furthermore that there should be no significant restrictions
on the case or thematic role of the object clitics, in contrast to Romance, where
clitics are a reflection of an older case system, and this prediction is borne out:
(43) ú-térék-èr-à
òvá-éndà ònyàmà p-òngàndà.
		 sm1-cook-appl-fv 2-guests 9.meat 16-9.house
		 ‘S/he cooks meat for the guests at home.’

[Otjiherero]

(44) ú-vé-térék-èr-à
ònyámà p-òngàndà.
		 sm1-om2-cook-appl-fv 9.meat 16-9.house
		 ‘S/he cooks them meat at home.’

. Multiple object markers, under this view, are only possible if they are taken to induce a
complex of nodes built from a single intermediate propositional node, itself unfixed (a characteristic of scrambling languages: Kiaer 2007), or if the node each one induces is fixed immediately, involving a pragmatic notion of constructive case (McCor­mack 2008). However, we will

Concepts of structural underspecification in Bantu and Romance

(45) ú-í-térék-èr-à
òvá-éndà p-òngàndà.
		 sm1-om9-cook-appl-fv 2-guests 16-9.house
		 ‘S/he cooks it for the guests at the house.’
(46) ú-pé-térék-èr-à
òvá-éndà ònyàmà.
		 sm1-om16-cook-appl-fv 2-guests 9.meat
		 ‘S/he cooks meat for the guests there.’

Although Otjiherero allows only one object marker per verb, object markers
can mark the dative (44), direct (45) or locative (46) complement of the verb. Object
markers are formally distinguished for different noun classes, but not for case. This is
directly explicable on the assumption that the object markers themselves induce the
unfixed node and provide it with a meta-variable decoration, without any structural
update of that node relation taking place. In effect, the only structural information
they provide is that they are an argument of the predicate corresponding to the verb.
Subject expressions, too, as in the Romance languages, are found in post-verbal, inverted position and often carry presentational focus by virtue of this late placement,
or alternatively, may be associated with an afterthought interpretation (48):
(47) à-léndò
à-fwík-à.
		 2-visitors sm2.past-arrive-fv
		 ‘(The) guests have arrived.’

[Nsenga]

(48) à-fwík-à
à-léndó.
		 sm2.past-arrive-fv 2-visitors
		 ‘Guests have arrived.’/‘They have arrived, the guests.’

Nsenga subject markers have lost their bottom restriction, and so update both
through *Adjunction and by constructing a linked node is possible, accounting for
the two different readings. Again, we find variation across Bantu, as, for example, in
Otjiherero, subject markers cannot be associated with update by Late*Adjunction
(49), and post-verbal subjects with agreeing subject marker can only be introduced through a LINK structure yielding (50), with its associated co-referring
afterthought interpretation. However, since the different update possibilities of
Otjiherero clitics are specified lexically, we would expect variation within the language as well as cross-linguistic variation, and indeed, update within the propositional structure is possible with (grammaticalized) locative subject markers (51)
indicating that they have lost their bottom restriction:
(49) *v-á-hìtí
òvá-ndú.
		 sm2-past-enter 2-people

[Otjiherero]

focus on one object marker languages, more specifically on Otjiherero in the following sections,
and will leave multiple object markers to one side for the present.





Lutz Marten, Ruth Kempson & Miriam Bouzouita

(50) v-á-hìtí,
òvà-ndú.
		 sm2-past-enter 2-people
		 ‘They entered, the people.’
(51) p-á-hìtí
òvá-ndú.
		 sm16-past-enter 2-people
		 ‘There entered people.’

Bantu languages thus make use of the same strategies as the Romance languages, with minor variations. Overt NPs, both subjects and objects, can be introduced into the parse sequence early or late, with the nodes that they serve to
decorate either being introduced within a single tree by *Adjunction or being
taken to be a linked structure. As in Romance, these different modes of introduction interact with the lexical specifications of the corresponding subject and
object markers, introducing locally unfixed nodes so that these can be unified
with an already constructed unfixed node, or, by virtue of some bottom restriction, allowing no such update and forcing any co-referring NP to be processed as
decorating an independent linked structure. As expected, these different strategies
can be used together. As a result, an object expression can be processed by early
application of *Adjunction to create an unfixed node, which can then come to be
fixed as the object node once that is introduced by actions of the verb. The subject
marker, on the other hand, which provides a type value but only a meta-variable
as formula value, will need update from application of Late*Adjunction after the
object node is completed; and the decorations of that subject node will then be
provided by using a post-verbal subject-marked expression to decorate the introduced unfixed node. An illustration of this is (52), with class 10 subject marking
and post-posed agreeing subject, under the assumption that subject markers in
Tumbuka have lost their bottom restriction and that the locative term is construed
as argument of the predicate:
(52) pa-mu-páanda zi-ka-dúk-a
mbúuzi.
[Tumbuka]
		 16-3-wall
sm10-past-jump-fv 10.goats
		 ‘Over the wall jumped goats.’ (‘The goats jumped over the wall.’) (Downing
2006: 62)

The data presented in this section have served to show the parallelism of
Romance and Bantu in terms of word-order freedom (through the application of *Adjunction and the building of linked structures), with restrictions on
this freedom imposed by the lexical constraints encoded by subject and object
markers. However, we have not yet addressed the question of the representation of
these subject and object markers, having concentrated merely on their interaction
with early and late placed NPs. In the following section, we address this question
and argue, taking the conceptual underspecification of Bantu clitics as a starting
point, that they are also structurally underspecified, inducing a locally unfixed

Concepts of structural underspecification in Bantu and Romance

node which they decorate with a place-holder, with both structural relation and
formula value needing to be updated.
4. Inducing locally unfixed nodes: Otjiherero subject markers
We argued above that Romance clitics may decorate locally unfixed nodes, the clitic
system overall being a reflex of the historic case system of Latin and its constructive
use in scrambling. We will employ the same notion of locally unfixed nodes for an
analysis of subject and object markers in Otjiherero, arguing synchronically with
evidence from passive and locative inversion constructions that the Bantu subjectmarking system parallels the scrambling-induced actions of Romance.14 From this
it also follows that in contrast to subject pro-drop in Spanish, where we have analyzed the verbal actions as providing both a subject node and a meta-variable as
decoration, we analyze subject “pro-drop” in Bantu as resulting from the lexical
actions of the subject markers, similar to the actions provided by pronominal object
clitics in Romance. Since subject markers are obligatory cross-Bantu, this means
that overt NP subjects are always taken either to decorate a linked structure, with
a copy of the term they provide having to be constructed in the primary structure,
or an unfixed node, in which case that node will have to unify with the node decorated by the subject marker. The analysis is motivated initially through the parallelism with Romance, and as formal reflex of the observation that Bantu clitics are
pronoun-like. However, crucial to the analysis to be given is that the subject marker
induces the building of a locally unfixed node which it decorates with a meta-variable
as formula value without updating that structural relation. The following steps of
the derivation of (53) illustrate the analysis.
(53) Kàtènáà
w-á-kòtòk-á.
		 1a.Katenaa sm1-past-return-fv
		 ‘Katenaa returned.’
(54)
		

[Otjiherero]

Parsing Kàtènáà w- in (53)
Tn(0), ?Ty(t),
Kàtènáàʹ ,
Ty(e), ·≠*ÒTn(0)

Uw,Ty(e), ·≠0Ò·≠1*ÒTn(0)

. The notion of locally unfixed nodes is independent of the notion of case, and so we are not
proposing that Bantu languages have grammatical case, since it is widely known that they do
not. On the other hand, there are in fact isolated instances of morphological differences between
subject and object markers, e.g., Proto-Bantu class 1 (3rd sing) subject *á vs. object *mù, 1st pl.
subject *tù vs. object *tú (Meeussen 1967), so the point might be worth revisiting.





Lutz Marten, Ruth Kempson & Miriam Bouzouita

We assume that the subject NP decorates an unfixed node introduced by
*Adjunction. We could equally have chosen to introduce an independent linked
structure into the emergent overall structure, since both strategies are available.
However, as mentioned above, under the locally unfixed node analysis for subject
markers, full NP subjects will never be taken to provide the decorations for a fixed
node because no fixed node is available if the subject precedes the verb. Following
the construction of the unfixed node decorated by the preposed expression,
we assume, as with the Romance object clitics, that it is the subject marker that
induces the building of a locally unfixed node, to which it adds a meta-variable
decoration. We assume, furthermore, that the subject marker encodes as formula
value a pronominal meta-variable restricted by the associated class information
(in this case expressed as a subscripted W), and we leave to one side the problem of
how this information is spelled out (see Cann et al. 2005: chapter 7). Note that the
two unfixed nodes can unify at this stage; and indeed in this derivation they do so,
with the actions induced by the subject marker having initiated the local domain
within which the left-peripheral expression will be interpreted. The next step is to
scan the lexical information from the tense morpheme. We assume that the tense
marker not only provides semantic information about the time of the event (an
annotation on the root node), but also provides a skeletal predicate frame with
subject- and predicate-requiring nodes, reflecting the probable historical origin of
many Bantu tense markers as verbs:
(55)

Parsing Kàtènáà w-á in (53)

		

Tn(0), ?Ty(t),Tns()

Kàtènáàʹ ,
Ty(e),
·≠0Ò ·≠*1ÒTn(0)

·≠0ÒTn(0)
?Ty(e)

·≠1ÒTn(0),
?Ty(e Æ t),

With the new presence of a fixed subject node, the unfixed node-relation
could now be fixed by unifying the two nodes, and in this simple case they do so
(with the passive and locative cases that we shall see, this option is not taken up).15
Hence the decoration of the subject is completed at this stage, and the information from the verb then annotates the predicate node as the next step. The final

. By definition of the Kleene star operator, the set of relations denoted can be empty, in this
case, the possible sequence of functor relations being null, allowing enrichment as the subject
relation.

Concepts of structural underspecification in Bantu and Romance

step in the derivation is the parsing of the final vowel, which induces the eventual
construction of the predicate, and then all semantic information in the tree is duly
compiled up progressively so that a final propositional formula of type t is derived
as decoration to the top node, satisfying the overall requirement. While the eventual tree for (53) looks like an ordinary subject-predicate structure, it is important
to keep in mind that its derivation involved construction steps at which first one
unfixed node and then one locally unfixed node were part of the tree.
One of the reasons for analyzing subject markers in Otjiherero as triggering
the building of unfixed nodes was to extend a parallelism between Romance
and Bantu clitics so that it applies also to the Bantu subject clitic (which not all
Romance languages display).16 We have so far discussed similarities between the
two language groups with respect to the different word-order possibilities of full
NPs, which we have analyzed as resulting from the interplay of linked structures,
*Adjunction and different lexical specifications of co-referential pronominal elements. Now we propose that in terms of structural and referential properties
of their pronominal elements also, the two language groups are similar in that,
in both groups, pronominal elements are typically positioned close to the verb
stem because they trigger the building of locally unfixed nodes. However, whereas
in Romance, the projection of locally unfixed nodes for a clitic coincides with
remnants of a case system, so that locally unfixed nodes can be fixed if enough
case information is available, thereby licensing the occurrence of multiple clitics,
in Otjiherero, locally unfixed nodes are only fixed if a fixed tree node address is
provided independently, either by lexical information from the tense marking, or
from the verb. This distinction provides the basis for the constraint in Otjiherero,
as in many Bantu languages, that only one object marker can be present in the
inflected verb form. Moreover, it explains why, in many noun-classes in the Bantu
languages, there is no difference in morphological marking between subject and
object marker: all the marker does is to initiate and decorate a locally unfixed node,
with some other expressions providing its decoration, and construal as subject
or not is only ensured either by the following tense markers (in the case of the
subject), or by the following verb in the case of the object marker. Confirmation of
this analysis now comes from passive and locative inversion in Otjiherero, which
pose the additional puzzle of precluding object prefix-marking, which this analysis
now promises to solve.

. In French, there is clear evidence that the pronoun does not decorate an unfixed node, as
the referentiality restriction on subject clitic doubling remains completely undisturbed. In the
Northern Italian dialects however, an account of subject clitic doubling might arguably follow
lines similar to those developed here (cf. Poletto 2000).





Lutz Marten, Ruth Kempson & Miriam Bouzouita

5. Passive and locative inversion
The passive construction was one of the constructions first argued in generative grammar to involve movement, based on the older descriptive observation
that the logical object of the verb becomes the grammatical subject in the corresponding passive. Accordingly, most analyses of the passive involve the matching
of the function of the relevant NP at two different levels: the logical or semantic
level, and the grammatical level. The challenge in Dynamic Syntax for an analysis
of the passive is that no level of grammatical function is defined: the trees built in
Dynamic Syntax are logical or semantic trees, and thus the eventual tree structure associated with both the active and the passive will be a transitive structure,
although in the passive the formula value for the logical subject might be an existentially quantified term (“someone”). However the semantic representation of the
logical object will be the same for both active and passive, as it will be associated
with an argument node below the predicate node. The question then is how to
derive such a transitive structure from the passive form of the verb. In the light of
the preceding discussion, a natural hypothesis to explore is associating the passive
suffix with a delay in unifying the unfixed node provided by the subject marker
with the logical-subject node provided by the tense marker.17 Delay in unifying the
unfixed node is unproblematic: all rules are constraints so not taking up the option
of unification is always a possibility. So the unfixed node, now identified as locally
unfixed, can remain unfixed until a point at some later stage in the parse when
another putative unification site arises, such as, for example, the logical object
node supplied by the lexical information from the verb. Thus, we assume that in a
parse of (56) the structural option to unify the locally unfixed node after the introduction
of the tense marker which supplied the fixed subject node, is NOT taken up in passives, but
rather the pointer is moved directly to the predicate node as in (57):

(56) òmbàpírà y-á-tjàng-w-á
(í òvá-nátjè).
		 9.letter
sm9-past-write-pass-fv by 2-children
		 ‘The letter was written (by the children).’
(57)

Parsing òmbàpírà y-á in (56)

		

Tn(0), ?Ty(t),Tns()

Òmbàpίràʹ,Ty(e),
·≠0Ò ·≠*1ÒTn(0)

·≠0ÒTn(0)
?Ty(e)

·≠1ÒTn(0),
?Ty(e Æ t),

. An alternative analysis in the system would be to invoke steps of inference between
different, albeit related, conc