(Oxford Studies in Comparative Syntax) Lena Baunaz, Liliane Haegeman, Karen de Clercq, Eric Lander-Exploring Nanosyntax-Oxford University Press (2018)

Exploring Nanosyntax
ii
Oxford Studies in Comparative Syntax

Richard Kayne, General Editor
Movement and Silence Functional Structure from Top to Toe: The

Richard S. Kayne Cartography of Syntactic Structures, Volume 9
Comparisons and Contrasts Edited by Peter Svenonius
Richard S. Kayne Chinese Syntax in a Cross-linguistic
Discourse-Related Features and Functional Perspective
Projections Edited by Y.-H. Audrey Li, Andrew
Silvio Cruschina Simpson, and W.-T. Dylan Tsai
Functional Heads: The Cartography of The Architecture of Determiners

Syntactic Structures, Volume 7 Thomas Leu
Edited by Laura Brugé, Anna Cardinaletti, Beyond Functional Sequence: The Cartography
Giuliana Giusti, Nicola Munaro, of Syntactic Structures, Volume 10
Cecilia Poletto Edited by Ur Shlonsky
Adverbial Clauses, Main Clause Phenomena The Cartography of Chinese Syntax: The
and Composition of the Left Periphery: Cartography of Syntactic Structures,
The Cartography of Syntactic Structures, Volume 11
Volume 8 Edited by Wei-Tien Dylan Tsai
Liliane Haegeman Argument Licensing and Agreement
Variation in Datives Claire Halpert
Edited by Beatriz Fernández Rethinking Parameters
and Ricardo Etxepare Edited by Luis Eguren, Olga Fernández-
Locality Soriano, and Amaya Mendikoetxea
Edited by Ian Roberts and Enoch Aboh Deconstructing Ergativity: Two Types of
Aspects of Split Ergativity Ergative Languages and Their Features
Jessica Coon Maria Polinsky
A Comparative Grammar of Borgomanerese The Morphosyntax of Portuguese and Spanish
Christina Tortora in Latin America
Cross-Linguistic Studies of Imposters and Edited by Mary A. Kato and Francisco
Pronominal Agreement Ordóñez
Edited by Chris Collins Dravidian Syntax and Universal Grammar
Japanese Syntax in Comparative Perspective K.A. Jayaseelan and R. Amritavalli
Edited by Mamoru Saito The Syntactic Variation of Spanish Dialects
Micro-Syntactic Variation in North American Edited by Ángel Gallego
English Exploring Nanosyntax
Edited by Raffaella Zanuttini and Laurence Edited by Lena Baunaz, Karen De Clercq,
R. Horn Liliane Haegeman, and Eric Lander
Exploring
Nanosyntax
Edited by Lena Baunaz, Karen De Clercq,

Liliane Haegeman, and Eric Lander
1
iv
1
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2018
All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.
You must not circulate this work in any other form

and you must impose this same condition on any acquirer.
Library of Congress Cataloging-in-Publication Data

Names: Baunaz, Lena, editor. | De Clercq, Karen, editor. |
Haegeman, Liliane M. V., editor. | Lander, Eric, editor.
Title: Exploring nanosyntax / edited by Lena Baunaz, Karen De Clercq,
Liliane Haegeman, and Eric Lander.
Description: New York : Oxford University Press, [2018] |
Series: Oxford studies in comparative syntax |
Includes bibliographical references and index.
Identifiers: LCCN 2017040790 (print) | LCCN 2017050828 (ebook) |
ISBN 9780190876760 (updf) | ISBN 9780190876777 (epub) |
ISBN 9780190876784 (online course) | ISBN 9780190876753 (pbk. : alk. paper) |
ISBN 9780190876746 (cloth : alk. paper)
Subjects: LCSH: Grammar, Comparative and general—Syntax.
Classification: LCC P291 (ebook) | LCC P291 .E985 2018 (print) | DDC 415—dc23
LC record available at https://lccn.loc.gov/2017040790
9 8 7 6 5 4 3 2 1
Paperback printed by WebCom, Inc., Canada
Hardback printed by Bridgeport National Bindery, Inc., United States of America
CONTENTS
Preface vii
List of Contributors xv
PART I. Background
1. Nanosyntax: The Basics 3
Lena Baunaz and Eric Lander
2. Notes on Insertion in Distributed Morphology and Nanosyntax 57
Pavel Caha
3. Spanning versus Constituent Lexicalization: The Case of
Portmanteau Prefixes 88
Knut Tarald Taraldsen
PART II. Empirical Investigations

4. A Note on Kim’s Korean Question Particles Seen as Pronouns 111
Michal Starke
5. Syncretism and Containment in Spatial Deixis 116
Eric Lander and Liliane Haegeman
6. Decomposing Complementizers: The Functional Sequence of French,
Modern Greek, Serbo-Croatian, and Bulgarian Complementizers 149
Lena Baunaz
7. Syncretisms and the Morphosyntax of Negation 180
Karen De Clercq
8. Nanosyntax of Russian Verbal Prefixes 205
Inna Tolskaya
PART III. Theoretical Explorations

9. Complex Left Branches, Spellout, and Prefixes 239
Michal Starke
vi
10. Word Order and Nanosyntax: Preverbal Subjects

and Interrogatives Across Spanish Varieties 250
Antonio Fábregas
11. The Feature Structure of Pronouns: A Probe Into
Multidimensional Paradigms 277
Guido Vanden Wyngaerd
12. Functional Sequence Zones and Slavic L>T>N Participles 305
Lucie Taraldsen Medová and Bartosz Wiland
Glossary 329
Index 331
[ vi ] Contents
P R E FA C E
Nanosyntax (Caha 2009; Starke 2009, 2011) is a formal theory of language

set in the Principles and Parameters framework (Chomsky 1986; Chomsky
and Lasnik 1993; among others). The theory has developed out of the car-
tographic approach (Rizzi 1997, 2004a, 2004b, 2013; Cinque 1999, 2002;
Belletti 2004; Kayne 2005, 2007; Cinque and Rizzi 2008/2010; Haegeman
2012) to grammar and can in many ways be considered a radical implementa-
tion of this approach that bridges the domains commonly identified as syntax
and morphology. Nanosyntax aims above all to identify the fine-grained
structure of language, a goal that, importantly, is framed in terms of the “one
feature–one head” maxim (see Cinque and Rizzi 2008, 50). The outcome of the
nanosyntactic approach is an extremely fine-grained decomposition of mor-
phosyntactic structure, and it has also yielded a set of novel tools for doing
precise empirical research.
In spite of the promising results that have emerged from nanosyntactic
research, publications remain scarce and rather inaccessible, and as a result
the framework is relatively unknown. The present volume is meant to fill this
gap in the literature and make the framework more accessible to a wider au-
dience. It brings together a selection of papers written by senior and junior
scholars working within the nanosyntactic framework. The diversity of the
contributors, the variety of topics discussed, and the wide range of languages
studied provide a well-rounded introduction to the theory.
The goals of the volume are threefold: to introduce the main theoretical
assumptions and the core technical machinery that nanosyntax makes use
of, to highlight some of the results that have been achieved and show the
potential of this line of research for empirical investigation, and to discuss
a number of aspects of the theory that are in need of further elaboration.
Correspondingly, the volume contains three kinds of contributions. The first
three chapters (Part I: Background) address some of the foundational concepts
of nanosyntax; all three chapters also provide some perspective on the rela-
tion between nanosyntax and a competing theoretical approach, namely
Distributed Morphology. A second group of contributions (Part II: Empirical
viii
Investigations) focuses on how the nanosyntactic line of argumentation can

be implemented in empirical research; in some of these chapters the focus is
less on the technical implementation and more on the use of nanosyntax as
a tool for uncovering descriptive generalizations. A third group of chapters
(Part III: Theoretical Explorations) explores more technical aspects of the
theory, considering theoretical issues that are unsettled and currently being
debated among researchers in the framework.
PART I: BACKGROUND
The three chapters making up Part I provide the foundation that will enable
readers to navigate the rest of this volume. These contributions also aim at
locating the theory of nanosyntax within the larger setting of theoretical
linguistics.
The volume starts with a comprehensive introduction to the framework,
“Nanosyntax: The Basics” by Lena Baunaz and Eric Lander. This chapter can
be viewed as an essential reader’s companion, as it sets out most of the tech-
nical details needed for understanding many of the contributions, such as the
concept of submorphemic heads, phrasal spellout, syncretism and the *ABA
theorem, lexical entries, the principles governing the relation between syntax
and the lexicon, and spellout-driven movement. This chapter also crucially
aims to set the framework in the broader context of generative grammar, es-
pecially with regard to cartography and Distributed Morphology.
Pavel Caha’s contribution, “Notes on Insertion in Distributed Morphology
and Nanosyntax,” provides a comparison of nanosyntax and Distributed
Morphology (DM), an approach to morphosyntax with which nanosyntax is
often compared. Specifically, Caha compares the conception of spellout as it is
understood and implemented in DM with the theory of phrasal spellout de-
veloped in nanosyntax. The author takes a partly historical perspective on the
issue, starting back in 2006, when all work done in DM explicitly or implicitly
denied the existence of phrasal spellout. At about this time, the first work in
nanosyntax became public. Caha revisits several early case studies that were
investigated in nanosyntax at that time (Starke’s unpublished work; Caha
2008, 2009), which argued in favor of a theory of phrasal spellout and devel-
oped its basic mechanics, such as The Superset Principle and The Biggest Wins
Theorem. This early work further suggested that adopting phrasal spellout
simplifies the architecture of grammar by immediately eliminating a number
of postsyntactic operations adopted in DM (minimally Fusion and Fission).
Caha further highlights how these developments filtered through in DM.
Although some of those working in DM have ultimately ended up adopting
some version of phrasal spellout (Radkevich 2010), much of the mainstream
[ viii ] Preface
work still opposes it (Embick 2014). The author focuses on some of the
counterarguments and also speculates as to where the debate is heading.
Tarald Taraldsen’s contribution “Spanning versus Constituent Lexicaliza
tion: The Case of Portmanteau Prefixes” compares conceptual and em-
pirical arguments for the “spanning” versus “constituency” approaches
to nanosyntactic spellout. Spanning allows for a sequence of heads to be
lexicalized even if they do not form a constituent, whereas the constituency
approach requires a structure to be a proper constituent if it is to be lexicalized.
He shows that there are both theoretical and empirical reasons to favor the
constituency approach over spanning. The core of the chapter is a case study
in Bantu nominal class prefixes, in which it is shown that the constituency
approach makes correct predictions about prefix structure. Specifically, he
shows that class prefixes in Bantu should be understood as portmanteaus cor-
responding to a constituent made up of Num, Cl, and a classifier-like N (which
is distinct from the head noun).
PART II: EMPIRICAL INVESTIGATIONS
Overall the contributions in this part focus on empirical applications of

nanosyntax, showing how the theory and methodology laid out in Part I can
contribute to a better understanding of certain data patterns. In many cases
the empirical material considered leads to certain theoretical issues being
brought to the fore as well. Some of the tools and concepts discussed in this
part include morphological containment, the relationship between syntactic
and lexical structure, syntactic movement in nanosyntax, and—above all—
syncretism and the ban on ABA patterns.
Michal Starke’s squib, “A Note on Kim’s Korean Question Particles Seen as
Pronouns” is an illustration of how the Superset Principle can help to solve em-
pirical puzzles. Starke discusses how Kim’s (2011) analysis of Korean question
particles as pronouns referring to the Addressee of the question suffers from
one major flaw: If these particles are pronouns they are also expected to occur
in declarative sentences, contrary to fact. Hence it seems impossible to cap-
ture both the interrogative and pronominal nature of these pronouns. Starke
shows how the Superset Principle can overcome this dilemma by assuming
that the lexicon contains lexical items that have the structure of pronouns
contained within the structure of question particles.
In their contribution “Syncretism and Containment in Spatial Deixis,” Eric
Lander and Liliane Haegeman explore a fine-grained morphosyntactic anal-
ysis of spatial deixis. They propose that the universal core of spatial deixis
is based on a three-way contrast: Proximal ‘close to speaker,’ Medial ‘close
to hearer,’ and Distal ‘far from speaker and hearer.’ They then discuss how
Preface [ ix ]
x
crosslinguistic variation in the domain of spatial deixis can be understood in

terms of syncretism: no syncretism (Dist vs. Med vs. Prox), Med/Prox syncre-
tism, Dist/Med syncretism, or total syncretism (Dist/Med/Prox). They also
present several patterns of morphological containment from a wide range of
languages, showing consistently that Distal structurally contains Medial and
that Medial structurally contains Proximal. The data point to a functional se-
quence made up of three additive Dx heads merged in a unique order, in line
with the nanosyntactic approach.
In her chapter titled “Decomposing Complementizers: The Functional
Sequence of French, Modern Greek, Serbo-Croatian, and Bulgarian Com
plementizers,” Lena Baunaz looks at the form of declarative complementizers
(equivalent to English that) in French, Modern Greek, Serbo-Croatian, and
Bulgarian. Starting from the observation that Greek has two declarative
complementizers pu and oti, each with its own distribution and selectional
restrictions, she proposes that French, Serbo-Croatian, and Bulgarian also
distinguish these two complementizer types but that French que, Serbo-
Croatian da, and Bulgarian če happen to be syncretic elements. She proposes
a nanosyntactic analysis of the French, Serbo- Croatian, and Bulgarian
complementizers in terms of complex morphemes, which lexicalize complex
structures of different sizes. Baunaz also shows that, depending on the size of
declarative complementizers, weak, strong, or no islands are created, suggesting
that declarative complementizers are interveners. Thus a nanosyntactic anal-
ysis of complementizers gives us insights into the principles underlying
Relativized Minimality effects (Starke 2001; Rizzi 2004a, 2013, et al.).
Karen De Clercq uses syncretisms as a tool to investigate the morphosyntax
of negative markers. In her contribution “Syncretisms and the Morphosyntax
of Negation,” she explores the negative markers (e.g. English not) of ten
different languages. A crosslinguistic look at these negative markers allows
her to identify four different types of negative markers: (i) negative scalar
quantity markers, (ii) negative classifier markers, (iii) negative focus markers,
and (iv) negative tense markers. Across these four different types of markers,
syncretisms are prevalent. When these markers are ordered with regard
to their scopal behavior (from narrow to wide scope), it turns out that the
syncretisms follow the natural scope order of negative markers. The syncre-
tism diagnostic relies on the assumption that nonadjacent syncretisms are
excluded in principle, that is, by the *ABA theorem. The study of syncretisms
is crucial to determining which features are merged adjacently in the func-
tional sequence: Looking at attested syncretisms across languages permits
the deduction of the linear order of the underlying functional features. From
the syncretism patterns and pursuing an intuition first expressed for sentence
negation by Poletto (2008), De Clercq develops a nanosyntactic analysis of
negation that involves splitting up what is often thought of as an indivisible
[ x ] Preface
unit NegP into at least five different submorphemic heads in a containment
relation.
Inna Tolskaya’s contribution, “Nanosyntax of Russian Verbal Prefixes,”
addresses the issue of widespread polysemy in Russian verbal prefixes and
prepositions (P) and argues that multiple instantiations of a single prefix
share a core conceptual meaning and receive specific denotations as a function
of their syntactic position. She shows that each polysemous P in Russian can
be assigned a single decomposed lexical structure, from which it is possible
to understand its set of related meanings as well as its selectional properties.
She argues that Pantcheva’s (2011) decomposition of spatial paths can also be
applied to scales of change and time. It is shown that the internal structure
of a verbal prefix parallels the structure of the PP complement of the verb,
and that the Superset Principle is crucial in deriving the observed selectional
restrictions of the prefix on the complement.
PART III: THEORETICAL EXPLORATIONS
The four chapters in Part III are of a more conceptual and technical nature and
explore a number of issues in nanosyntactic theory that remain unresolved at
this stage of the theory. These open-ended issues include, broadly, the role of
complex specifiers and constituenthood in spellout, clausal phenomena from
a nanosyntactic perspective, pointers, and the interaction between distinct
functional sequences.
Michal Starke’s second contribution, “Complex Left Branches, Spellout,
and Prefixes,” starts by discussing how to distinguish between prefixes and
suffixes in a principled and nonstipulative way. He then delves into the details
of constructing prefixes. Whereas suffixes are argued to be part of the main
spine in the primary derivation, prefixes are structures created in a sec-
ondary derivation as a last resort and inserted into the main spine as complex
specifiers.
It is fair to say that most nanosyntactic work has thus far concentrated
on the internal structure of words or lexical items. Indeed, a recurring issue
raised by non-nanosyntacticians concerns the relevance and validity of the
framework when it comes to sentence-level syntax. In his chapter “Word
Order in Nanosyntax: Preverbal Subjects and Interrogatives Across Spanish
Varieties,” Antonio Fábregas explores how nanosyntactic proposals can be
applied to the study of a set of phenomena that are standardly considered
to fall within the realm of sentential syntax. In particular, he considers the
distribution of overt subjects in wh-interrogatives across three varieties of
Spanish (European Spanish, Venezuelan Spanish, and Dominican Republic
Spanish). Although Verb–Subject order is compulsory in this kind of sentences
Preface [ xi ]
xii
in European Spanish, some subjects can be preverbal in Venezuelan Spanish

(Mérida), and all kinds of subjects can be preverbal in Dominican Spanish.
Making use of the nanosyntactic tool of phrasal spellout, Fábregas claims that
the variation attested in Spanish can be reduced to the size of the exponents
contained in the lexical repertoire of each variety. Specifically, he shows that
the minimal difference among these three varieties lies in the size of the expo-
nent responsible for subject agreement.
Guido Vanden Wyngaerd’s contribution, “The Feature Structure of
Pronouns: A Probe into Multidimensional Paradigms,” examines the validity
of the *ABA diagnostic, which assumes that noncontiguous syncretisms are
excluded in principle. He discusses both empirical and theoretical issues that
bear on the validity of the *ABA diagnostic. His empirical material comes
from Cysouw’s (2003) extensive study of person marking, including personal
pronoun paradigms and the patterns of syncretism they reveal. At the theo-
retical level, it is shown that morphemes that involve the fusion of multiple
grammatical dimensions (such as person and number) require an analysis in
terms of pointers (Caha and Pantcheva 2012). Because pointers introduce the
possibility of ABA patterns, he also discusses a second analysis, based on a re-
vision of the Superset Principle (the Revised Superset Principle, or RSP), orig-
inally proposed by Pavel Caha. He then discusses the empirical and theoretical
merits of the pointer approach as compared with the RSP approach, showing
that the two make different predictions.
In “Functional Sequence Zones and Slavic L>T>N Participles,” Lucie
Taraldsen Medová and Bartosz Wiland argue in favor of the existence of dis-
tinct “zones” of functional features (fseq zones). Under such an approach,
elements that generally compete for insertion with each other form the same
fseq zone, whereas elements that co-occur together form different fseq zones.
On the basis of participles and thematic suffixes in Polish and Czech, they
identify three such zones: root, theme, and participle. Each zone is argued to
have a complex internal structure, drawing on and paralleling other work in
nanosyntax (Starke 2006; Lundquist 2008) on the decomposition of lexical
categories. Their approach is able to explain why only unaccusative verb roots
in Czech and Polish can build adjectival L-passives, whereas unergative roots
cannot, making crucial use of the “peeling” approach to case and argument
selection in their analysis.
REFERENCES
Belletti, Adriana (ed.). 2004. Structures and Beyond: The Cartography of Syntactic
Structures, Vol. 3. New York: Oxford University Press.
Caha, Pavel. 2008. “The Case Hierarchy as Functional Sequence.” In Scales, edited by
Marc Richards and Andrej Malchukov. Lipsko, Poland: Universität Leipzig.
[ xii ] Preface
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
Caha, Pavel and Marina Pantcheva. 2012. “Contiguity Beyond Linearity: Modeling
Cross- Dimensional Syncretisms.” Talk presented at Workshop on the
Representation and Selection of Exponents, University of Tromsø. June 7.
[online] Available at http://cms.unige.ch/lettres/linguistique/seminaire/media/
220/Caha%20Pantcheva%20231012.pdf
Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press.
Chomsky, Noam and Howard Lasnik. 1993. “The Theory of Principles and Parameters.”
In Syntax: An International Handbook of Contemporary Research, Vol. 1, ed-
ited by Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo
Vennemann, pp. 506–569. Berlin: Walter de Gruyter
Cinque, Guglielmo. 1999. Adverbs and Inflectional Heads. Oxford: Oxford University
Press.
Cinque, Guglielmo (ed.). 2002. Functional Structure in DP and IP: The Cartography of
Syntactic Structures, Vol. 1. New York: Oxford University Press.
Cinque, Guglielmo and Luigi Rizzi. 2008. “The Cartography of Syntactic Structures.” In
CISCL Working Papers on Language and Cognition, 2, edited by Vincenzo Moscati,
pp. 43–59. Siena: CISCL.
Cinque, Guglielmo and Luigi Rizzi. 2010. “The Cartography of Syntactic Structures.”
In The Oxford Handbook of Linguistic Analysis, edited by Bernd Heine and Heiko
Narrog, pp. 51–65. New York: Oxford University Press.
Cysouw, Michael. 2003. The Paradigmatic Structure of Person Marking. Oxford: Oxford
University Press.
Embick, David. 2014. On the Targets of Phonological Realization. Ms., University of
Pennsylvania.
Haegeman, Liliane. 2012. Adverbial Clauses, Main Clause Phenomena and the Composition
of the Left Periphery: The Cartography of Syntactic Structures, Vol. 8. Oxford: Oxford
University Press.
Kayne, Richard S. 2005. Movement and Silence. Oxford: Oxford University Press.
Kayne, Richard S. 2007. “On the Syntax of Quantity in English.” In Linguistic Theory
and South Asian Languages: Essays in Honour of K.A. Jayaseelan, edited by Joseph
Bayer, Tanmoy Bhattacharya, and Musaliyar V. T. Hany Babu, pp. 73–105.
Amsterdam: John Benjamins.
Kim, Chonghyuck. 2011. “Korean Question Particles are Pronominals: A Transparent
Case of Representing Discourse Participants in the Syntax”. [online] Available at
http://ling.auf.net/lingBuzz/001157.
Lundquist, Björn. 2008. Nominalizations and Participles in Swedish. Doctoral disserta-
tion, University of Tromsø.
Pantcheva, Marina. 2011. Decomposing Path: The Nanosyntax of Directional Expressions.
Doctoral dissertation, University of Tromsø.
Poletto, Cecilia. 2008. “On Negative Doubling.” In Quaderni di Lavora ASItMs: pp. 57–
84. Venice: University of Venice.
Radkevich, Nina. 2010. On Location: The Structure of Case and Adpositions. Doctoral dis-
sertation, University of Connecticut.
Rizzi, Luigi. 1997. “The Fine Structure of the Left Periphery.” In Elements of Grammar,
edited by Liliane Haegeman, pp. 281– 337. Dordrecht, The Netherlands:
Kluwer.
Rizzi, Luigi. 2004a. “Locality and Left Periphery.” In Structures and Beyond: The
Cartography of Syntactic Structures, Vol. 3, edited by Adriana Belletti, pp. 223–251.
New York: Oxford University Press.
Preface [ xiii ]
xiv
Rizzi, Luigi (ed.). 2004b. The Structure of CP and IP: The Cartography of Syntactic
Rizzi, Luigi. 2013. “Syntactic Cartography and the Syntacticisation of Scope-Discourse
Semantics.” In Mind, Values and Metaphysics—Philosophical Papers Dedicated to
Kevin Mulligan, edited by Anne Reboul, pp. 517–533. Dordrecht, The Netherlands:
Springer.
Starke, Michal. 2006. “The Nanosyntax of Participles.” Lectures at the 13th EGG
summer school, Olomouc.
Starke, Michal. 2009. “Nanosyntax: A Short Primer to a New Approach to Language.”
Nordlyd: Special Issue on Nanosyntax 36: pp. 1–6.
Starke, Michal. 2011. “Towards an Elegant Solution to Language Variation: Variation
Reduces to the Size of Lexically Stored Trees.” [online] Available at LingBuzz/
001183.
[ xiv ] Preface
CONTRIBUTORS
Lena Baunaz is a postdoctoral assistant at the University of Zurich. She holds

a PhD from the University of Geneva, which she published as The Grammar of
French Quantification (Springer, 2011). Her recent research interests include
the nanosyntax of the subjunctive mood, complementizers, and ontological
categories. She has published in Probus, Studia Linguistica, and others.
Pavel Caha is an assistant professor in the Department of Czech Language at
Masaryk University in Brno. He has worked on case, numerals, adpositions, de-
gree morphology, verbal particles, and other topics. His papers have appeared
in Natural Language and Linguistic Theory, Journal of Linguistics, Glossa, Journal
of Comparative Germanic Linguistics, and others.
Karen De Clercq is a postdoctoral researcher funded by the Flemish Fund for
Scientific Research (FWO) and working at Ghent University. She wrote her
PhD on the nanosyntax of negative markers under the supervision of Liliane
Haegeman. She is currently working on the fine-grained morphosyntax of
Quantity-words (many/much; few/little), adjectives, degree comparison, and
negation.
Antonio Fábregas is professor of Hispanic linguistics at UiT-Norway’s Arctic
University. He is the editor-in-chief of Borealis. An International Journal of
Hispanic Linguistics and associate editor of the Oxford Research Encyclopedia of
Morphology. His work concentrates on the syntactic analysis of so-called mor-
phological phenomena and sometimes looks into semantics and phonology.
Liliane Haegeman was professor of English linguistics at the University of
Geneva (Switzerland) from 1984 to 1999. Between 2000 and 2009 she was
full professor of English linguistics at the University of Lille III (France). Since
2009 she has held a research position at Ghent University. She has worked
extensively on the syntax of English and Flemish.
Eric Lander is a postdoctoral researcher at the University of Gothenburg,
currently working on negation in the history of Scandinavian. His research
xvi
interests include Germanic philology, the NP/DP parameter, demonstratives,

complementizers, and ontological categories. He has earned degrees from
Harvard, Leuven, and Ghent.
Lucie Taraldsen Medová obtained her PhD in Slavic linguistics at Princeton and
she is a driver for Tromsø taxi. Her work includes illuminating conversations
with various customers in her taxi and articles on argument structure of Slavic
and Romance languages. She claims that SE is an antipassive and was con-
vinced she showed so in her dissertation, but the world still keeps turning as
it did before, prior to this marvelous discovery. More than in linguistics, she is
interested in mysteries of people’s souls and in food-and-wine pairing.
Michal Starke is the creator of nanosyntax, of LingBuzz—the standard ar-
chive of linguistic papers—and of the popular Eastern Generative Grammar
summer school. He has taught at NYU, the University of Geneva, and the
University of Tromsø.
Knut Tarald Taraldsen is professor of general linguistics at the University of
Tromsø, where he has been working since 1981. His research focuses on theo-
retical syntax, Scandinavian syntax, Romance syntax, and the morphosyntax
of Southern Bantu languages. He has been a visiting associate professor at
MIT and a visiting scholar at Princeton and the University of Cape Town. His
work has been published in Linguistic Inquiry and Lingua and in several edited
volumes.
Inna Tolskaya recently completed her PhD at the University of Tromsø. Her
research focuses on prefixation in Russian and English, verbal decomposition,
and scalarity. Her other interests include phonology and Manchu-Tungusic
languages. She has published in the Journal of Linguistics, Lingua, and confer-
ence proceedings.
Bartosz Wiland is an assistant professor at the faculty of English at Adam
Mickiewicz University in Poznań and a member of the Young Academy of
Europe. He received his PhD in 2009 and specializes in theoretical and com-
parative morphosyntax of English and Polish, especially the internal syntax of
verbs and sources of word order variation.
Guido Vanden Wyngaerd is a professor of Dutch linguistics at KU Leuven
(Brussels campus). He is the co-author (with Johan Rooryck) of Dissolving
Binding Theory (Oxford University Press, 2011). His recent research has shown
a shift toward nanosyntax, including the study of such topics as the represen-
tation of person, gradability in adjectives, degree comparison, and negation.
[ xvi ] Contributors
PART I
Background
2
CHAPTER 1
Nanosyntax
The Basics
LENA BAUNAZ AND ERIC L ANDER
N anosyntax (Caha 2009; Starke 2009, 2011a, 2011b ) is a generative approach to

the study of language that is in line with the major tenets of the Principles and
Parameters framework of Chomsky (1981, 1986). More precisely, the nanosyntactic
approach is a direct descendant of cartography, as it is anchored in basic carto-
graphic assumptions about the fine-grained nature of the functional projection
and the fundamental simplicity of syntactic structure. Although nanosyntax is
currently in the process of growing and developing as a theoretical framework in
its own right, it has already proven to offer a promising set of methods for doing
detailed empirical research, coupled with an innovative yet restrictive theory of
syntax and its place in the architecture of Universal Grammar (UG).
The first chapter of this volume aims to set the theory of nanosyntax in the
broader context of generative grammar, especially with regard to two leading
frameworks in current generative theory and research: cartography and
Distributed Morphology (henceforth DM). The chapter is written for readers
familiar with generative linguistics. Section 1.1 briefly sketches the history
and basic theoretical underpinnings of cartography, with particular attention
We would like to thank Liliane Haegeman and Tom Leu for their extensive notes on
this chapter. We are also very grateful for invaluable questions and comments from two
anonymous OUP reviewers. All errors are our own. Lena Baunaz’s research has been
supported by the Swiss National Foundation (grant: PA00P1_145313) and FWO pro-
ject 2009-Odysseus-Haegeman-G091409. Eric Lander’s research has been supported
by BOF grant 01D30311, FWO project 2009-Odysseus-Haegeman-G091409, and a
postdoctoral grant from the University of Gothenburg.
4
paid to those facets that have led to the emergence of nanosyntax. Section
1.2 consists of a short overview of the theory and terminology of DM, with
the aim of explicitly pinpointing and exposing some of the core differences
with nanosyntax that could otherwise lead to confusion or misunderstanding.
Section 1.3 provides the reader with an overall picture of nanosyntactic theory
and also introduces the major technical tools needed to navigate this volume
(any additional technical information will be provided where relevant in later
chapters). Section 1.4 is an overview of the nanosyntactic interpretation of
the Principles and Parameters framework. Section 1.5 concludes.
1.1. CARTOGRAPHY: A MAP OF SYNTACTIC CONFIGURATIONS
In earlier Principles and Parameters work, very basic structures were advocated
for clauses and noun phrases (CP-IP-VP and NP, as in Chomsky 1981, 1986),
but the meticulous study of syntax from a crosslinguistic perspective has, over
time, led researchers to postulate more finely articulated structures for clauses
and noun phrases. In many ways this began with Pollock’s (1989) splitting of the
category I on the basis of a comparison between French and English, and Abney’s
(1987) arguments for positing the functional projection DP above the lexical NP
in English, which built on earlier work by Szabolcsi (1981, 1984, 1987) on the
Hungarian noun phrase. It was from this general line of reasoning that the car-
tographic approach to syntax (see Benincà 1988; Cinque 1990, 1999, 2002; Rizzi
1997, 2004b; Belletti 2004) can be said to have emerged. Foundational work
in cartography was done in the 1990s, notably Rizzi (1997) arguing for a fine-
grained left periphery (i.e. splitting CP into further projections) mostly on the
basis of Italian data, and Cinque’s (1999) crosslinguistic study leading to a finely
articulated map of the adverb positions populating the functional domain of IP.
Their main results are summarized in (1):
(1) a. [ForceP [TopP* [FocP [TopP* [FinP [IP . . . ]]]]]] [Rizzi 1997, 15, his (41)]
b. [MoodP speech-act frankly [MoodP evaluative fortunately [MoodP evidential allegedly
[ModP epistemic probably [TP past once [TP future then [ModP irrealis perhaps
[ModP necessity necessarily [ModP possibility possibly [AspP habitual usually
[AspP repetitive again [AspP frequentative(I) often [ModP volitional intentionally
[AspP celerative(I) quickly [TP anterior already [AspP terminative no longer
[AspP continuative still [AspP perfect(?) always [AspP retrospective just
[AspP proximative soon [AspP durative briefly [AspP generic/progressive characteristically
[AspP prospective almost [AspP sg.completive(I) completely
[AspP pl.completive tutto [VoiceP well [AspP celerative(II) fast/early
[AspP repetitive(II) again [AspP frequentative(II) often
[AspP sg.completive(II) completely]]]]]]]]]]]]]]]]]]]]]]]]]]]]]].
(Cinque 1999, 106)
[ 4 ] Background
The goal of the cartographic approach is clearly illustrated in (1), namely to draw
“maps as precise and complete as possible of syntactic configurations” (Rizzi
2013, 1). An important result of cartographic research, then, is the view that the
units of syntax are much smaller, and syntactic representations much more ar-
ticulated, than previously thought. This general notion of decomposition as the
(empirical and theoretical) way forward in mapping out UG is a prominent fea-
ture of nanosyntax as well.
It is commonly assumed in cartography that the map of UG should be
very simple, structurally speaking. First, each syntactico-semantic feature is
assumed to be an independent head that projects. This is known as the “one
feature–one head” maxim (henceforth OFOH) (Cinque and Rizzi 2008, 50; see
also Kayne 2005, ch.12). Second, most researchers have strict assumptions
about how heads project. These assumptions are deeply influenced by the work
of Kayne (1984, 1994): (i) structures are strictly binary-branching and right-
branching, (ii) only one specifier per head is allowed, and (iii) only leftward
movement is allowed. In short, the combination of the OFOH maxim with a
strict Kaynean (antisymmetric) view on structure-building leads to the kinds
of detailed syntactic representations emerging out of the cartographic research
program.
Closely related to this goal of mapping out UG is the strong trend in cartog-
raphy to “syntacticize” domains of grammar (see Section 1.1.1 for references).
The degree to which meaning can and should be syntacticized continues to be
a major point of contention within and between frameworks (see Geeraerts
2010 for an overview). In generative frameworks it is (at least implicitly)
assumed that certain aspects of meaning, often termed grammatical semantics,
belong to the grammar proper (i.e. syntax), whereas other aspects of meaning,
termed extralinguistic or conceptual semantics, fall outside of grammar.1 Typical
examples of the first category are features encoding number, case, tense, as-
pect, and so on; aspects of meaning considered to arise from the social, cul-
tural, or historical context, on the other hand, are seen to fit into the latter
category. Drawing the boundary between the two is an empirical question,
in that only concepts observed to have morphosyntactic encoding across lan-
guages can be considered grammatical(ized) (see Cinque 2010). A major goal
of cartography (and nanosyntax), then, is to determine exactly which parts of
meaning are grammatical and should thus be syntacticized. The great extent
to which semantics is syntacticized in cartography can be described in terms
of a strict mapping between syntax and semantics. This means that syntax is
assumed to be the vehicle for expressing grammatical semantics, and it does
so by means of abstract syntactico-semantic features that are arranged by
syntax into a hierarchy.
1. Although definitions will vary, other terms for this kind of meaning may include
extragrammatical, pragmatic, encyclopedic, etc.
N a n o s y n ta x : T h e B a s i c s [ 5 ]
6
1.1.1 The model of grammar and full syntacticization
The broad-strokes model of grammar currently adopted by most generativists,

including cartographers, is shown in Figure 1.1 (Chomsky 1965, 1981, 1986,
1995; for a cartographic perspective see Rizzi 2013, among others).
The “box of linguistic computations” (as syntax is called by Rizzi 2013,
10) contains a presyntactic repository (or lexicon) storing both functional
and lexical morphemes, made up of (one or more) abstract features like sg,
pl, past, def, etc. The presyntactic lexicon then feeds these “bundles” of ab-
stract features into the recursive syntax. Syntax then computes the gram-
matical representations to be interpreted at the interfaces of phonological
form (PF) and logical form (LF). Typically, phonological interpretation is
achieved at PF. This includes, among other things, the interpretation of spe-
cial prosodic contours relating to topic and focus (see Bocci 2009 for Italian).
Semantic interpretation is achieved at LF, which includes the interpretation of
scope-discourse properties. Beyond these interfaces we find “other (language
independent) systems on both sound and meaning sides, which use grammar-
determined representations for communication, socialization, the expres-
sion of thought, play, art, and whatever use humans make of their linguistic
abilities” (Rizzi 2013, 10). In the former systems the ways in which we artic-
ulate and perceive phonological representations are determined. In the latter
systems the ways in which we understand language are determined. As is clear
from Figure 1.1, these systems are external to syntax: That is, the articulatory–
perceptual systems and conceptual–intentional systems receive input via the
interfaces from syntax. From the point of view of cartography, with so much
of the grammar having been syntacticized, we can state that there is “very
little computation” required postsyntactically for the purposes of interpreta-
tion, because the information received from syntax comes packaged in such
rich syntactic structures (Rizzi 2013, 11).
Lexicon
Syntax
PF LF
Articulatory- Conceptual-
perceptual systems intentional systems
Figure 1.1 Architecture of grammar [based on Rizzi 2013, 10, his (22)]
[ 6 ] Background
1.1.2 The proliferation of functional heads and the fseq
Generative linguists generally assume the Uniformity Principle: “In the ab-
sence of compelling evidence to the contrary, assume languages to be uniform,
with variety restricted to easily detectable properties of utterances” (Chomsky
2001, 2). This principle is at the core of cartography; as a research program, car-
tography aims to identify the complete set of atoms making up grammatical
structures and the hierarchical organization of these structural atoms, both
of which are taken to be universal (Rizzi 1997; Cinque 1999; Cinque and Rizzi
2008). The existence of crosslinguistic variation is due to the way languages
(overtly or covertly) realize these structures, as well as the type of movements
they allow: “the distinct hierarchies of functional projections dominating VP,
NP, AP, PP, IP, etc., may be universal in the type of heads and specifiers that
they involve, in their number, and in their relative order, even if languages
differ in the type of movements that they admit or in the extent to which they
overtly realize each head and specifier” (Cinque 1999, 2002; Cinque and Rizzi
2008, 46, citing Rizzi 1997). Under this hypothesis, “parameters are formal
properties of features” (Shlonsky 2010, 12). This is known as the Borer–
Chomsky Conjecture, which has been formulated as in (2):
(2) The Borer–Chomsky Conjecture

All parameters of variation are attributable to differences in the features
of particular items (e.g. the functional heads) in the lexicon.
(Baker 2008, 353, and also Borer 1984)
So information-structural movement to the left periphery, for instance, is

triggered by the presence of the relevant features and heads, and when the
attracting head has the appropriate triggering properties (say, an EPP fea-
ture). As cartographers admit, this is a strong claim, because it “implies
that if some language provides evidence for the existence of a particular
functional head (and projection), then that head (and projection) must
be present in every other language, whether the language offers overt ev-
idence for it or not” (Cinque and Rizzi 2008, 45, citing Kayne 2005 and
Cinque 2006).
Because not all languages provide overt evidence for all the functional
projections that are postulated, a question that naturally arises under the car-
tographic approach is whether the full fseq is always syntactically represented
and if so, how one handles the fact that not all languages provide overt evi-
dence for its full instantiation. One way to approach the issue of crosslinguistic
variation might be in terms of activation: Although functional categories in
the fseq as such are universal, they may be deactivated or inactive in some
languages but not others, perhaps because of whether certain heads carry in-
terpretable or uninterpretable features (Shlonsky 2010, 426). The concept of
8
truncation has also played a role in trying to answer this question. According
to this view, a structure can be reduced by being “cut off” at a certain layer,
preventing the higher functional categories from projecting (see Rizzi 1994;
Haegeman 2003, 2006b,2006c). The stronger approach, that all functional
categories are always active in every language, is argued for by Cinque (1999,
132–133, 2013). It has also been proposed that variation in the overt instanti-
ation of functional categories can be explained by assuming that the fseq can
to some extent display conflation of two or more syntactic heads (e.g. Rizzi
1997; Zubizaretta 1998), possibly the product of the movement of one head
to a higher head.
Evidence that the fseq is universal comes, on the whole, from detailed
empirical work, often from a comparative perspective. In particular, efforts
have been made to achieve a more fine-grained, syntactic(ized) decompo-
sition of scope-discourse properties in the CP domain (Rizzi 1997; Aboh
2004a; Belletti 2004; Haegeman 2006a, 2012). Additional efforts include
elaborating the precise structural positions for adverbs (Laenzlinger 1998;
Cinque 1999), adjectives (Cinque 2010), subjects (Cardinaletti 1997, 2004),
negation (Haegeman and Zanuttini 1991; Zanuttini 1991; Haegeman 1995),
quantifiers (Beghelli and Stowell 1997; Szabolcsi 1997; Puskás 2000), tense/
aspect/ mood/ modality (Cinque 1999), inflection (Pollock 1989; Belletti
1990), the nominal domain (Abney 1987; Giusti 1997), and more. Over
the course of cartographic investigations there has been a proliferation of
fine-grained functional structures: CP has been split into Force, Top, Int,
Foc, Mod, and Fin (Rizzi 1997, 2001, 2004a; Aboh 2004a), the vP-to-TP re-
gion into a range of modal, temporal, and aspectual projections (Cinque
1999, 2006), the event structure into various sorts of VPs (Larson 1988;
Hale and Keyser 1993; Ramchand 2008), DP into D, Q, Num, A, and so
forth (Szabolcsi 1981, 1984, 1987, 1994; Abney 1987; Ritter 1991; Giusti
1997; Alexiadou, Haegeman and Stavrou 2007). Work has also been done
on refining the internal structure of PPs (Koopman 2000; den Dikken 2010;
Noonan 2010) and APs (Scott 2002; Laenzlinger 2005; Svenonius 2008;
Leu 2015).
The identification of fine-grained syntactic structures is perhaps the most
salient characteristic of cartographic work, but it is important to recognize
why exactly syntactic representations have developed in this direction. As
emphasized by Cinque and Rizzi (2008), fine-grained structures are posited
only insofar as there is morphosyntactic evidence for the functional heads
involved, with the overall result after years of research of a very large in-
ventory of functional categories. For example, Rizzi (1997) demonstrates
that Italian distinguishes separate syntactic positions for topicalized and fo-
cused elements; Aboh (2004a), moreover, shows that Gungbe has particles
that overtly realize the topic and focus heads. This is evidence for discrete
features or projections encoding topic and focus in the syntax. In other words,
[ 8 ] Background
a comparative approach is deployed to assess the universality of the fseq.
Work on crosslinguistic variation often has macrocomparative (typological)
scope, but the systematic study of grammatical phenomena in closely related
languages or dialects has also given rise to a fruitful field of microcomparative
work, notably for the dialects of North Italy (Benincà and Vanelli 1982; Poletto
2000; Manzini and Savoia 2003, 2007, 2011; Benincà and Poletto 2004),
Dutch and its dialects (Haegeman 1992, 2014; Barbiers 2006; Barbiers and
Bennis 2007), Scandinavian languages (Johannessen et al. 2009; Lindstad
et al. 2009), and also for diachronic studies (see Benincà, Ledgeway, and
Vincent 2014 for a recent reference).
1.1.3 Cinque 2005
An influential theoretical development in cartography has been Cinque’s

(2005) reinterpretation of Greenberg’s Universal 20 (Greenberg 1963, 87) (see
Abels and Neeleman 2009, 2012; for an alternative account based on seman-
tics, see Dryer 2009). In his seminal work, Cinque observes that of the 24
mathematically possible orders of demonstrative (Dem), numeral (Nml), ad-
jective (A), and noun (N), only 14 are attested, leaving 10 possible orders un-
attested. He proposes to derive this striking pattern from the following basic
restrictions:
(i) The universal merge order is Dem > Nml > A > N (the extended projec-
tion of the noun; Grimshaw 1991).
(ii) Only leftward movement is allowed (Kayne 1994).
(iii) Only phrasal movement is allowed (i.e. only XPs move; head movement
is disallowed) (see Koopman and Szabolcsi 2000, among others).
(iv) Only phrases containing N may be moved (i.e. remnant movement is
disallowed).
Note that the fourth restriction means that pied-piping is allowed (as long as
N is included in the moved constituent). On the basis of these restrictions,
Cinque demonstrates that the 14 attested orders can be derived whereas the 10
unattested orders are, by the same token, underivable. Importantly, Cinque’s
theory can be applied at the level of morphology as well (see Muriungi 2008;
Caha 2009; Lander 2015a, 2015b); as we see in Section 1.3, virtually every
aspect of the theory has an important impact on the implementation of
nanosyntax. Cinque’s (i) and (ii)—namely the view that the fseq is universal
and right-branching—are commonly assumed in the nanosyntactic approach.
Restrictions (ii), (iii), and (iv) are reflected in the current nanosyntactic
system of phrasal spellout and spellout-driven movement, as elaborated in
Section 1.3.3.4.
10
1.1.4 A summary of cartographic assumptions
Driven by a set of assumptions centering around the OFOH maxim, the fun-
damental simplicity (and antisymmetry) of syntactic projection, and a strict
mapping between syntax and semantics, the cartographic program has by
means of detailed comparative work argued for a particular view of grammar,
essentially summed up as follows: Syntax is made up of a limited set of atoms
that are organized into a single, universal sequence (the fseq). In terms of em-
pirical work, researchers in the framework embrace a comparative approach,
with the goal of mapping out the universal fseq and describing crosslinguistic
variation in a careful and detailed way.
1.2 THEORY AND TERMINOLOGY IN DISTRIBUTED

MORPHOLOGY AND NANOSYNTAX
Terminological differences are common sources of confusion when moving

between theoretical frameworks. Different terms may be used for the same
(or very similar) concepts, and conversely the same term is sometimes used
and understood in quite different ways. For these reasons we think it is worth
having an explicit discussion of terminology in DM versus nanosyntax before
moving on to the particulars of nanosyntactic theory.2 See Caha (Chapter 2)
for a more in-depth comparison of theoretical and analytical issues between
the two frameworks.
1.2.1 Basic architectures compared
DM (Halle and Marantz 1993; Marantz 1997; Bobaljik 2007, 2012, 2015; Embick
and Noyer 2007; Harley 2014; Embick 2015) has played an important and in-
fluential role in the development of nanosyntax. Both frameworks are late-
insertion models (see Section 1.2.2) with a commitment to the idea that syntax
is responsible not only for sentence structure but also for word structure. The
main difference is that nanosyntax seeks to eliminate the various postsyntactic
rules and operations available in the DM model. Nanosyntax also argues for a
different perspective on the lexicon (conceived of as separate “lists” in DM, as
seen in Figure 1.2). Most notably, nanosyntax does away with the presyntactic
list of morphemes that feeds syntax, ultimately because in nanosyntax there is
no distinction between the “features” of morphemes and the “heads” of syntax
(consider OFOH, and the discussion in Section 1.3). The main architectural
2. We are grateful to the anonymous reviewers for their instructive comments and
insightful questions, convincing us to write this section.
[ 10 ] Background
Figure 1.2 Model of grammar according to Distributed Morphology [based on Embick 2015,
20, his (12)]
Syntax
(SMS)
Atomic features
merged as fseq
Spellout
Lexicon
</f n/ [F1[F2]] CONCEPT >
PF LF
Articulatory- Conceptual-
perceptual systems intentional systems
Figure 1.3 Model of grammar according to nanosyntax (Caha 2009, 52; Starke 2011)
differences can be seen by comparing Figure 1.2 for DM versus Figure 1.3 for
nanosyntax.
In Figure 1.3, the abbreviation SMS stands for syntax, morphology, and
semantics, which in nanosyntax are seen as one and the same module, to be
identified with (the cartographic notion of) syntax.3 This idea has a number
of theoretical consequences that are considered in more detail in Section 1.3.
3. Note that the interface with the conceptual– intentional systems may in
nanosyntax be called CF (conceptual form) (e.g. Caha 2009, 52), a way of distinguishing
the nanosyntactic vision of a radically syntacticized formal semantics from the more
12
The rest of this section is organized as a discussion of four (clusters of re-

lated) terms: Morpheme and Vocabulary Insertion (Section 1.2.2); Vocabulary
Insertion/Item/List versus lexical item/entry and lexicon (Section 1.2.3);
allomorphy (Section 1.2.4); and morphophonology, suppletion, and portman-
teau (Section 1.2.5). This is not an exhaustive overview, of course; rather, the
goal is to preempt some common areas of misunderstanding and also hope-
fully to ease the transition into our discussion of nanosyntactic theory in
Section 1.3.
1.2.2 Morpheme and Vocabulary Insertion
In American structuralist approaches (e.g. Bloomfield 1933; Harris 1951), a

morpheme is considered to be the smallest unit consisting of a “sound” or
“form” paired with a “meaning” or “function.” In realizational, late-insertion
theories like DM and nanosyntax, however, sound and meaning are not in-
herently linked but are separate entities, and it is only when the syntactic
derivation reaches a certain point that the meaning is paired with (for some,
replaced by) sound.
The structuralist notion of meaning is modeled in DM as a bundle of
formal syntactico-semantic features, each (language-specific) bundle called
a morpheme.4 These abstract bundles of meaning are fed into the syn-
tactic component, where functional morphemes are merged as syntactic
terminals (say, the morpheme for third person singular present tense [3sg,
pres], merged as the head T0). The (morpho)syntactic representation, now
a syntactic tree structure with complex terminal nodes, then branches
off to PF and the articulatory-perceptual systems. It is in this mapping
between syntax and phonology that phonological forms are inserted, a
process known as Vocabulary Insertion [note that various postsyntactic op-
erations like Morphological Merger, Fission, Fusion, Impoverishment, fea-
ture deletion, and so forth, may need to take place before, and sometimes
after (readjustment rules), Vocabulary Insertion]. The closest analogue of
Vocabulary Insertion in nanosyntax is what is usually called spellout or
lexicalization.
standard sense of “covert syntax” at LF (logical form). In this vein consider also Kayne
(1998) on eliminating LF movement.
4. Note that roots are hypothesized in DM to have different properties (see Embick
2015, 6–7). We mainly focus our discussion on functional morphemes here. For a
nanosyntactic perspective, see Taraldsen Medová and Wiland (Chapter 12) for a rad-
ical decomposition of the root domain, building on ideas from Lundquist (2008) and
Starke (2009) on the internal structure of lexical categories.
[ 12 ] Background
1.2.3 Vocabulary Item versus lexical item/e ntry and lexicon
The correspondence between sound and meaning is in DM referred to as a

Vocabulary Item, and the (memorized) inventory of Vocabulary Items is
called the Vocabulary List. Although Vocabulary Item is sometimes used in
nanosyntax as a term for stored correspondences of this sort, one is more likely
to find the term lexical item or lexical entry. The lexical entry of nanosyntax
is not exactly the same as the Vocabulary Item of DM. One of the main
differences involves the placement of “encyclopedic” (i.e. noncompositional,
extralinguistic) information.5 A Vocabulary Item in DM involves syntactico-
semantic structure and phonology only; noncompositional information
comes from another, separate list called the Encyclopedia. In nanosyntax, on
the other hand, a lexical entry is considered to have three available slots for
storing linguistic information: the first for the phonological form, the second
for the syntactico-semantic structure, and the third for conceptual (encyclo-
pedic) information.
This allows nanosyntax to maintain that there is only a single lexicon (ex-
plicitly denied in DM, with its separate lists). The usage of lexicon and lex-
ical item/entry (to the extent that this terminology is standardized within
the framework) instead of Vocabulary List and Vocabulary Item, then, is ac-
tually motivated by an important difference in theoretical assumptions. As
seen in Section 1.2.1, nanosyntax does not posit a presyntactic list of abstract
morphemes as DM does. Thus the term morpheme is understood differently
in nanosyntax, often being used in the more traditional sense as a sound–
meaning pairing, or as a synonym for lexical entry.
1.2.4 Allomorphy
The term allomorphy in nanosyntax is understood in a restricted sense, as a

phonologically conditioned alternation. A typical example of allomorphy in this
sense is the English plural marker -s, which is phonetically realized as [-s] after
voiceless obstruents (tip-s, boat-s, riff-s, math-s), as [-əz] after (post)alveolar
fricatives (mass-es, praise-s, bush-es, match-es, grudge-s), and as [-z] everywhere
5. For example, even though dog and cat are, syntactically speaking, basically indistin-
guishable (i.e. they are animate singular count nouns), there is a great deal of idiosyn-
cratic, “real-world” information that is not important for the syntax (or the phonology
for that matter) but nevertheless connected to these lexical items: physical shape and
appearance, that dogs are more social than cats, that cats do not like to be walked, etc.
In addition to the idiosyncratic, real-world definition of words, there is also the pos-
sibility of special idiomatic usages that need to be stored as encyclopedic information
[for example, that nouns like ape and dog can be used as verbs (i.e. ‘imitate’ and ‘pursue
intently’) but cat cannot; Bobaljik 2015, 25–26].
14
else (voiced consonants: rag-s, tab-s, tram-s, rail-s, wave-s; vowels: bee-s, tray-

s, etc.). The elsewhere environment is considered to point to the underlying
representation /-z/, which is the phonological form stored in the lexical entry
for the English plural morpheme (the allomorphs of /-z/—i.e. [-s], [-əz], and
[-z]—do not need to be stored, because they are predictable).
As Bobaljik (2015, fn.8) points out, some researchers in DM choose to
use allomorphy to refer to alternations that are lexically or grammatically
conditioned, requiring an analysis in terms of morphology. An example
of allomorphy in this sense might be irregular pasts as they are commonly
analyzed in DM, as seen in (3):
(3) a. [past] ⇔ -t / ]V __ where V = {√dwell, √spell, √dream, . . .}

b. [past] ⇔ -Ø / ]V __ where V = {√speak, √run, √fly, . . .}
c. [past] ⇔ -d / ]V __
[Bobaljik 2015, 6, adapted from his (14)]
This analysis assumes that there are three lexically conditioned allomorphs,
each occurring in its own set of contexts: -t can be used in a subset of irregular
verbs like dwell—dwelt, spell—spelt, dream—dreamt (3a). Ablaut in irregular
verbs like speak—spoke, run—ran, fly—flew, and so forth, is modeled in terms
of a null morpheme (run—ran-Ø, where the vowel change is the result of a
later (morpho)phonological readjustment rule, occurring after Vocabulary
Insertion) (3b). Finally -d is the regular (default, elsewhere) past ending (3c).
Note here that the final element -d may then later on participate in phonologi-
cally predictable allomorphy, for example, devoicing in wash-ed /wɔʃt/, trick-ed
/tɹɪkt/or epenthesis in batt-ed /bæɾəd/, trott-ed /tɹɑɾəd/, and so forth.
The absence of an independently recognized notion of morphology (or
more precisely the series of postsyntactic mechanisms affecting the output
of syntax in the branch to PF) in nanosyntax means that it is impossible
in this framework for allomorphy to denote anything other than a phono-
logically conditioned alternation. In nanosyntax, any kind of contextual
allomorphy that is not phonological–phonetic in nature, such as grammat-
ical or lexical allomorphy, must be encoded in some other way, for example in
terms of a more fine-grained structural difference or a lexical entry storing an
irregular form.
1.2.5 Morphophonology, suppletion, portmanteau
In DM, one may account for the vowel alternation in run—ran in terms of a
somewhat superficial readjustment rule turning /ʌ/into /æ/. A slightly more
complex root alternation like can—coul-d (where -d could be analyzed as the
regular past ending) would be accounted for in terms of suppletion, where a
[ 14 ] Background
particular Vocabulary Item contains information that V0 should be spelled out
as coul- /kʊ-/in a specific context, namely when the verb can is to the imme-
diate left of [Tʹ [past]]. This rule prevents the incorrect (but regularly formed)
*can-d. Extreme cases of morphological irregularity or unpredictability that
are not segmentable at all can be called portmanteau elements.6 For instance,
forms like were and was are portmanteaus consisting of the verb be plus past-
tense (and inflectional) features. Another example would be French contrac-
tions of certain prepositions with the masculine definite article, namely au for
*à le or du for *de le (see Taraldsen in Chapter 3). Portmanteau elements are
analyzed in DM in terms of fusion of syntactic heads/terminals, turning two
(or more) heads into a single head (see Caha in Chapter 2 for references and
discussion).
Although different in nature and applying at different stages postsyntax,
all of these rules and operations are essentially morphophonological. In
nanosyntax, however, there is a very strict division of labor between syntax
and phonology, with no independent morphology of any kind between the
two. This also means that morphophonological rules (applying between mor-
phology and phonology in some sense) have no natural place in the architec-
ture of nanosyntax. So whereas in DM an alternation like tell—tol-d involves
both a lexically conditioned allomorph -d and a morphophonological readjust-
ment rule (/ɛ/ → /oʊ/) (Bobaljik 2015, 7), in nanosyntax it is necessary instead
to posit a more fine-grained underlying structure (see Caha in Chapter 2, fn.
8 and references there for tol- as a portmanteau, plus the regular ending -d)
or the storage of specific structural configurations in the lexicon [for example,
the lexical entry < /geɪv/ ⇔ [V give] + [past] > linking the regularly formed but
incorrect *give-d to the phonological form /geɪv/ (i.e. gave)].
1.3 NANOSYNTAX: THEORY AND METHODOLOGY
At this point we turn to why nanosyntax looks the way it does, with its “strictly
modular” architecture (lacking any independent notions of morphology or
morphophonology and with a single, postsyntactic lexicon). Nanosyntax is
based on the reasoning that the general increase in the inventory of syntactic
projections and the idea that features (rather than feature bundles) are the
atoms or building blocks of syntax have important consequences for the de-
marcation (or lack thereof) between syntax and morphology and thus for the
model of grammar in general. The purpose of this section is to explain the
basic underpinnings and inner workings of Figure 1.3.
6. In practice the distinction between suppletion and portmanteau is, admittedly, not
always clear-cut.
16
1.3.1 Submorphemic heads and phrasal spellout
As a descendant of cartography, nanosyntax assumes a strict syntax–

semantics mapping, the OFOH maxim, and the view that syntactic structures
are fundamentally quite simple. For a morpheme made up of the syntactico-
semantic features X, Y, and Z, for example, it is not possible in nanosyntax
to arrange X, Y, and Z in a “feature bundle” (4a); rather, one is forced to view
these features as heads merged in a binary-and right-branching tree, putting
them in a fundamentally asymmetrical relation with one another (4b) (see, for
instance, Dékány 2009, 51):
(4) a. Unordered bundle (i.e. symmetrical relation)

* [X, Y, Z]
b. Ordered sequence (i.e. asymmetrical relation)
✓ [XP X [YP Y [ZP Z]]]
Many important aspects of nanosyntactic theory can be seen to emerge from

this way of thinking about morphemes.
Let us begin with the well-accepted fact that there is not a strict one-to-
one relationship between abstract features and their phonological realizations
(i.e. morphs). In any one given language, there will always be more featural
distinctions than there are morphs available, that is, there is generally a one-
to-many relationship between morphs and features. Consequently, features
can be described as being submorphemic, because single morphs usually cor-
respond to several formal features. As seen in (4b), moreover, features are
heads merged in a tree structure. If these heads are submorphemic and mul-
tiple heads make up a single morph, then it must be possible for spellout to
target phrases (XPs) and not just heads, which is what is standardly assumed
in frameworks like DM.
As an illustration of this concept, consider the split between agglutinating
languages like Finnish and fusional–inflectional languages like most Indo-
European languages (see also Halle and Marantz 1993: 116). Finnish tends to
have distinct morphs for individual functional categories. For example, the
allative case in Finnish is expressed by the morph -lle, and plural number is
expressed by -i, as seen in (5). In Latin, on the other hand, the categories case
(K) and number (Num) are typically expressed by a single morph. As seen in
(6), the ending -ās expresses both accusative case and plural number (as well
as feminine gender).
(5) a. karhu-lle (Finnish)

bear-all
‘onto the bear’
[ 16 ] Background
b. karhu-i-lle
bear-pl-all
‘onto the bears’ (from Caha 2009, 73)
(6) puell-ās (Latin)
girl-acc.FEM.pl
‘girls.acc’ [from Rocquet 2013, 8, her (1)]
The Latin morph -ās is a portmanteau: The features for K and Num are
submorphemic in Latin, as there is not a direct one-to-one correspondence be-
tween functional category and phonological realization, as there is in Finnish
(where -i is Num and -lle is K).
As mentioned at the beginning of this section, the OFOH maxim requires
positing two projections, KP and NumP. In addition, there are good reasons
to think that K and Num are merged in a strict order. Consider, for instance,
that in languages like Finnish in which K and Num are realized separately, the
Num morph is systematically found closer to the nominal stem than the K
morph is, meaning that the underlying hierarchy of functional categories is
K > Num > N.
This leads to more general considerations of the framework. In the tradi-
tional model of grammar in Figure 1.1 and the DM version of this model in
Figure 1.2, abstract morphemes from the lexicon are inserted at individual
terminal nodes in the syntactic structure. As we just saw, K and Num are re-
quired to be separate heads under nanosyntactic assumptions.7 For Finnish,
then, there is no conflict between terminal insertion and separate K and Num
heads, with one morph per head. For Latin, however, we are forced to say that
the portmanteau -ās corresponds not to a single head but rather to (at least)
two, namely K0 and Num0.
Different ways of handling such mismatches have been proposed, some of
which were briefly encountered in Section 1.2.5, like Fusion; another approach
might be to posit a null morph in either K0 or Num0, with the other head
hosting the overt morph -ās, and a rule specifying the proper contextual
environments for them.8 Caha (Chapter 2) provides a detailed discussion of
these issues in DM versus nanosyntax, but suffice it to say for now that the
nanosyntactic strategy for dealing with portmanteau morphology is to make
use of phrasal spellout. Rather than trying to preserve at all costs the idea that
7. We are of course simplifying for the purposes of exposition. K and Num can both
be decomposed into multiple features, and thus multiple heads.
8. See also Kayne (2005) for application of this general approach to various syntactic
phenomena. Null morphemes are also allowed in nanosyntax, of course, but only if
there is evidence for it and the allomorphic alternation is phonologically plausible (see
Section 1.2.4).
18
morphemes must correspond to syntactic heads (X0s) (and thereby having to

accept morphology-specific operations like Fusion, for example, to account
for more problematic cases), nanosyntax instead adopts a system of spellout
that can target phrases (XPs).9
In a phrasal spellout system, it is possible to model portmanteau mor-
phology as larger chunks of structure, something a system restricting spellout
exclusively to terminals cannot do. Thus the entire phrase [KP K [NumP Num]]
can be targeted for spellout in the case of Latin -ās (Figure 1.4). In Finnish, KP
and NumP are separately targeted for spellout (Figure 1.5).
Note that we choose to represent the Finnish morphemes -i and -lle as
phrases (KP and NumP) rather than as heads (K0 and Num0). The stems puell-
and karhu- are also represented as phrasal constituents (NPs). The reason for
this ultimately has to do with considerations of spellout-driven movement,
KP
⇒ -ās
NumP
K
⇒ puell-
Num NP
Figure 1.4 Spelling out -ās in Latin
⇒ -lle
KP
⇒ -i
NumP
K
⇒ karhu-
Num
NP
Figure 1.5 Spelling out -i and -lle in Finnish
9. Note that phrasal and terminal spellout are not necessarily mutually exclusive. It is
possible to have a system in which both spellout mechanisms coexist (see in particular
Pantcheva 2011, section 6.3.2).
[ 18 ] Background
the details of which we postpone until Section 1.3.3.4. As already sketched,
spellout-
driven movement of these XPs will result in the correct linear
ordering of elements, with movement of NP to the left of K in Figure 1.4,
giving puell-ās, and with roll-up movement in Figure 1.5, giving karhu-i-lle.
1.3.2 Overall consequences for the architecture of grammar
The introduction of phrasal spellout brings with it a deeper shift in the very
architecture of grammar (here following the reasoning of Starke 2011a,
2011b). Phrasal spellout is a way to lexicalize multiple heads as a single unit,
but without destroying the hierarchical ordering of these heads (i.e. the fseq)
“inside of” the morpheme. Thus phrasal spellout allows for a direct and trans-
parent (in fact, one-to-one) correspondence between syntax (the fseq) and
morphology. Morphology is just like syntax in that it is built up by merging
abstract features as heads in an fseq. Thus it is not the case that morphemes
are constructed beforehand and fed into syntax as its primitive building
blocks. Instead it is basically the other way around: Morphemes are built by
syntax, and the primitive building blocks of syntax (from the cartographic
perspective and OFOH) are features.
A consequence of this morphology-as-syntax idea is that there is no
presyntactic lexicon of available feature bundles, because features cannot be
combined before syntax but only in the syntax. Instead this lexicon must be
postsyntactic, because a morpheme [that is, a syntactic (SMS) structure] can
be stored away only if it has already been built in the first place. This should be
thought of primarily in terms of language acquisition, during which the child
must determine which SMS structures to store in her mental lexicon over
time. In other words, the syntactic motor is running, continuously producing
syntactic trees, some of which are considered crucial enough in the linguistic
environment to merit storage in the lexicon. When a new lexical entry is
created to store a certain SMS structure, furthermore, it becomes possible to
link this structure to phonological and conceptual information as well.
As mentioned, the only thing that acts as input to the syntactic computa-
tion is the individual atomic features provided by UG, which syntax merges
together as heads according to the universal fseq, resulting in a syntactic struc-
ture. At each step or cycle of the syntactic derivation, moreover, whatever has
been built by syntax must be lexicalized by appropriate material from the lex-
icon, after which the syntax continues to build, followed by another round of
lexical access, and so on. This spellout loop between syntax and the lexicon can
be seen in Figure 1.3. Henceforth we refer to structures generated by the syntax
(SMS) as syntactic trees or S-trees for short. Syntactic trees which are stored in
lexical entries will be called lexical trees or L-trees. Although both S-trees and
L-trees ultimately have the same source (the SMS component) and are thus
20
made up of the same material, it is nevertheless important to distinguish the

two. This becomes clear in Section 1.3.3.4 for the spellout process, the purpose
of which is to match an S-tree with the appropriate L-tree (which, as one-third
of a lexical entry, is linked to specific phonological and conceptual content too).
1.3.3 The basic tools and technology
In this section we introduce some of the common methodological tools in use

in nanosyntactic research, as well as the spellout mechanism, which is a cru-
cial component of the theory.
1.3.3.1 Mapping the fseq: From linear to hierarchical order
The basic nanosyntactic tools used in mapping out the universal fine-grained
structure of language are the following: (i) semantics, (ii) syncretism, and (iii)
morphological containment. We discuss each in turn.
(i) Semantics. One way of mapping out the universal structure of language
is to study semantic compositionality. For example, in her work on the hier-
archy of Path features, Pantcheva (2011) gives a number semantic arguments
in support of her proposed hierarchy of Path features. Route, for instance,
which can be paraphrased as ‘from X to Y,’ can be seen as being composed of the
features for Source and Goal. That is, in terms of structure, Route can be thought
of as being built on top of Source ‘from’ and Goal ‘to.’ Semantic considerations
like these can thus play a role in establishing fseqs and determining differences
in structural size (see Ramchand 2008 on the semantic classes of verbs; detailed
work on participles by Lundquist 2008 for Swedish and Taraldsen Medová and
Wiland in Chapter 12 for Slavic; Fábregas 2009 on the semantics and mor-
phology of indefinites and interrogatives, among others).
However, semantics on its own may not be sufficient; semantic facts need
to be closely integrated and aligned with the syntactic and morphological facts
as well (just as these need to agree with the semantics).10 In the case of Path,
for instance, Pantcheva (2011) provides empirical support from a broad range
10. Nanosyntax is not a revival of Generative Semantics, as sometimes claimed, as

syntax, morphology, and semantics are all the same module, whereas in Generative
Semantics (Lakoff 1971) there is a clear prioritization of semantics over syntax.
As Cinque and Rizzi (2008, 53) put it: “there is a fairly restrictive universal set of
properties that can be expressed by the functional elements entering into the different
hierarchies associated to clauses and phrases.” This limit on which parts of meaning
are “grammaticalized” or “syntacticized” means that the universal hierarchy of syntax
should not be reduced to semantics. Rather it is syntax that dictates “the pattern and
the seams which delimit meaning and use” (Shlonsky 2010, 14).
[ 20 ] Background
of languages proving that the syntax and morphology of Path do indeed line
up with the semantic facts (see also Tolskaya in Chapter 8 on Path prefixes and
prepositions in Russian). The methodology for deciding “how much meaning”
a feature can encode should be decided on morphosyntactic grounds, meaning
that in principle every head in the fseq should be backed up by morphological
evidence in some language (see Lander and Haegeman in Chapter 5).
(ii) Syncretism. Nanosyntax has been particularly successful as a theory of
syncretism, and the intensive study of syncretism has played a central role in
the development of nanosyntactic theory (see Caha 2009, 2010, 2013; Taraldsen
2009; Pantcheva 2011; De Clercq 2013, Chapter 7; Rocquet 2013; Vangsnes 2014;
Baunaz 2015, 2016, Chapter 6; Baunaz and Lander 2017, to appear; among
others). Caha (2009, 6) defines the phenomenon of syncretism as “a surface
conflation of two distinct morphosyntactic structures.” In other words, syncre-
tism arises when two or more distinct grammatical functions are spelled out by
a single form. As an example, consider the expression of Location, Goal, and
Source readings in English (7) and French (8) (based on Pantcheva 2011, 238).
(7) a. I ran at the sea. Location

b. I ran to the sea. Goal
c. I ran from the sea. Source
(8) a. J’ai couru à la mer. Location/Goal

I.have run at/to the sea
‘I ran at the sea.’ or ‘I ran to the sea.’
b. J’ai couru de la mer. Source
I.have run from the sea
‘I ran from the sea.’
As seen in (7), English prepositions make overt distinctions between Location/

Goal and Source readings. In French, however, a single preposition à expresses
both Location and Goal readings, with Source expressed by the distinct form de,
as seen in (8). In other words, there is a Location–Goal syncretism in French
but not in English. Building on work by Svenonius (e.g. 2010), Pantcheva (2011,
sections 8, 9) investigates syncretism patterns of Location ‘at,’ Goal ‘to,’ Source
‘from,’ and Route ‘via’ readings across languages.
Building on Blake (1994), among others, Caha (2009) performs a detailed
crosslinguistic study of (nominative–accusative) Case systems. He demonstrates
that Case syncretisms are constrained, in that the phenomenon targets only ad-
jacent cases. If we take just five cases (nom, acc, gen, dat, ins), in Russian we
see the syncretisms nom–acc, acc–gen, and gen–dat–ins. We can arrange the
five cases in a table such that syncretism affects only adjacent cells, as seen in
Table 1.1 (shaded cells highlighting syncretism).
22
Table 1.1 SYNCRETISMS IN RUSSIAN (FROM CAHA 2009, 12)
‘window’ (sg) ‘teacher’ (pl) ‘one hundred’
nom okn-o učitel-ja st-o

acc okn-o učitel-ej st-o
gen okn-a učitel-ej st-a
dat okn-u učitel-am st-a
ins okn-om učitel-ami st-a
When Caha expands his empirical coverage to more languages and to more cases,
he finds that it is possible to arrange the Case sequence in a single universal
order such that attested syncretisms are always in adjacent cells. From this he
formulates the generalization in (9):
(9) Universal Case Contiguity (Caha 2009, 49)

a. Nonaccidental11 Case syncretism targets contiguous regions in a
sequence invariant across languages.
b. The Case sequence: nom—acc—gen—dat—ins—com
Note that the observation in (9a) that syncretism affects only adjacent case
layers and the resulting Case sequence in (9b) are not simply a convenient way
to organize the data on syncretism. Rather, they constitute a hypothesis that
makes predictions about possible syncretisms. More precisely, given (9) we
predict that we should not find a language where two noncontiguous cases are
syncretic, for example a gen–ins syncretism with a distinct dat. Indeed, Caha
finds that noncontiguous cases are never (or very rarely) syncretic. Two unat-
tested syncretisms are illustrated in Table 1.2.
The hypothesis that syncretism targets only adjacent cells in a paradigm
is known as the *ABA theorem, first formulated by Bobaljik (2007, 2012) in
his work on comparative and superlative inflection in adjectives across
languages.12
11. This refers to the possibility of homophony. Two morphemes with distinct un-
derlying phonological representations can be affected by phonological rules that cause
them to surface identically at PF. If this happens within a single paradigm, we end
up with two very syncretic-looking morphemes, but this is purely an accident. For the
purposes of the computation, these morphemes are structurally distinct and do not
instantiate a genuine syncretism. For discussion see Caha [2009, 11, his (11)], among
others.
12. Various questions about the *ABA theorem and the possibility of “gaps” in the
functional sequence (e.g. Caha 2009, section 9.3; Starke 2013) have been raised re-
cently, many of which are discussed in this volume (see Vanden Wyngaerd Chapter 11).
[ 22 ] Background
The contiguity/*ABA generalization about syncretism gives us a powerful
tool for probing and teasing out the atomic ingredients of syntax. By looking
at attested syncretisms across languages, it is possible not only to identify
which fine-grained features are present, but also to deduce the linear order of
these features.
(iii) Morphological containment or nesting. Syncretisms do not reveal eve-
rything about the structure of the functional features at stake. For instance,
with respect to Case, although we can identify the feature sequence, we cannot
identify the hierarchy, that is, just taking syncretism into consideration will
not reveal if the underlying sequence is nom > acc > gen > dat > ins > com or
com > ins > dat > gen > acc > nom. That is, syncretism reveals a linear order
of features that can reflect one of two possible hierarchies; what syncretism
cannot explicitly tell us is which of the two possible hierarchies is correct. In
other words, in Figure 1.6 we cannot yet tell if A is nom and F is com, or if A is
com and F is nom.
The hierarchy in Figure 1.6 should be understood as consisting of priva-
tive features that build on each other cumulatively as heads in the fseq. This
Table 1.2 UNAT TESTED SYNCRETISMS:

*ABA (BASED ON ROCQUET 2013, 32)
NOM A
ACC B
GEN A A
DAT B
INS A
K6 > K5 > K4 > K3 > K2 > K1
Figure 1.6 Case as a hierarchy of additive heads
24
means that cases are in superset–subset relations with one another, such
that the lowest case is made up of one feature, the next case in the sequence
is composed of this feature plus a second feature, the third case is composed
of these two features plus a third, and so on. Because of this, Caha (2009)
prefers to use K1, K2, and so on, instead of nom, acc, and so on, as seen in
Figure 1.6.
With respect to the case hierarchy, to determine if A in Figure 1.6 should be
identified with nom or com, other phenomena must be taken into account.
First, the fact that nom is crosslinguistically “unmarked” suggests that nom
is the simplest, structurally and featurally speaking (Caha 2009, 23). Second,
the facts concerning morphological containment (see Bobaljik 2007, 2012),
which we detail presently, suggest that nom is the smallest case in Figure 1.6
as well. For instance, in West Tocharian the acc plural ending -m̥ is found
overtly contained within the gen/dat plural ending -m̥ts, as seen in (10a). In
(10b), we see that the Russian dat.pl ending -am is overtly contained within
the ins.pl ending -ami.
(10) a. West Tocharian (Caha 2009, 69)

[GEN/DAT.PL [ACC.PL -m̥ ] -ts] [gen/dat [acc]]
b. Russian (Caha 2009, 12)
[INS.PL [DT.PL -am ] -i] [ins [dat]]
These and similar morphological facts show that certain cases are contained
within others: acc is contained within gen and dat, and dat is contained
within ins.
Similar facts are found for prepositional phrases (PPs). If we assume that
prepositions are like case morphemes in being composed of K features, then
we see that the way prepositions select their DP complements also reflects a
containment relation. In English the gen preposition of selects an acc com-
plement (11a), as represented in (12a). In Arabic the dat preposition li selects
a gen complement (11b), as represented in (12b). In German the ins preposi-
tion mit selects dat (11c), as represented in (12c).
(11) a. of him (English)

b. li-l-binti (Arabic)
to the girl.gen
c. mit einem Hammer (German)
with a.dat hammer
These facts tell us that gen contains (i.e. is larger than) acc, that dat contains
gen, that ins contains dat, and so on.
[ 24 ] Background
(12) Case selection by prepositions as containment (Caha 2009, 37)
a. English [of + DP-acc]GEN [gen [acc]]
b. Arabic [li + DP-gen]DAT [dat [gen]]
c. German [mit + DP-dat]INS [ins [dat]]
Figure 1.7 is a more detailed illustration of the configuration in (12c) from

German.
The highest layer (K5) in the sequence corresponds to the preposition mit
whereas the lower part (from K4 down to K1) is realized as the dative case
ending. (DP will undergo spellout-driven movement to the left of K4 to pick
up dative inflection.)
As we have seen, morphological containment—when it can be observed—
is an especially clear way of observing the nesting of underlying functional
structure. For Case in particular we have seen that dat is larger than gen, gen
is larger than acc, and so on. Thus we can safely conclude that the correct hi-
erarchy is the one in Figure 1.8.
We turn now to the inner workings of the spellout process in nanosyntax.
K5 K4 K3 K2 K1 DP
mit –dat
instrumental PP
Figure 1.7 Containment of dative within instrumental PP
K6 > K5 > K4 > K3 > K2 > K1
nom
acc
gen
dat
ins
com
Figure 1.8 The Case hierarchy
26
1.3.3.2 The principles of spellout
Syncretism involves a single form that is applicable in more than one struc-
tural environment. For example, French à in (8a) functions either as a Place
marker with the meaning ‘at’ or a Goal marker with the meaning ‘to.’ To put
it differently, there is a single lexical entry (a single morpheme) stored in the
lexicon, with an L-tree that is able to match multiple S-trees. To flesh out this
idea, we need to be more explicit about the interaction between S-trees in SMS
and L-trees in the lexicon. As we will see, the availability of phrasal spellout
becomes crucial here.
As alluded to in Section 1.2.1, a lexical entry is made up of three elements
which are linked together: (i) phonological structure, (ii) syntactic (SMS)
structure (i.e. an L-tree), and (iii) conceptual structure. This is illustrated in
Figure 1.9 for the nonce-item blicket.
ZP
YP
Z
/ blık t / XP BLICKET
Y
Figure 1.9 Lexical entry
Lexical entries are arbitrary in the sense that each language (in fact each id-
iolect; Kayne 2016) will have its own idiosyncratic inventory of lexical entries
pairing phonology, syntax (SMS), and conceptual (extralinguistic, pragmatic)
information. It is the successful storage of these entries over time, essentially,
that constitutes acquisition of language.
Every single possible syntactic structure or S-tree does not necessarily cor-
respond to a specific lexical entry in a given language. Indeed, the fact that
syncretism is prevalent in a language shows that a single L-tree will often
have to map onto multiple S-trees of various sizes. As we discuss toward the
end of this chapter, this says something quite profound about the nature of
crosslinguistic variation.
For our purposes in this chapter we are focusing on the L-tree (the second
slot in a lexical entry). S-trees generated by the SMS module are abstract,
requiring proper lexicalization or spellout. This involves matching an S-tree
with an appropriate L-tree. Because this L-tree is part of a lexical entry, more-
over, the matching process establishes a connection not only between the S-
tree and an L-tree, but also between the S-tree and a certain phonological
form (the first slot) and concept (the third slot), each of which is interpreted
later on when the syntactic derivation “branches off” in the model of grammar
[ 26 ] Background
assumed by generative linguists. In other words, the lexicon, with its three-
slot lexical entries, is in many ways what binds this model together.
There are three principles of spellout governing the proper matching of
L-trees to S-trees: (i) the Superset Principle, (ii) the Elsewhere Principle, and
(iii) the Principle of Cyclic Override (see Starke 2009). We discuss each of
these in turn.13
(i) The Superset Principle. The first principle of spellout is known as the
Superset Principle, stated in (13).
(13) Superset Principle (Caha 2009, 67, but see Vanden Wyngaerd Chapter 11)
A lexical tree L can match a syntactic tree S if L is a superset (proper or
not) of S. L matches S if L contains a node that is identical to a node in
S and all the nodes below are also identical.
Informally put, the Superset Principle allows for an S-tree to be spelled out by
an L-tree as long as that L-tree is the same size or bigger (and assuming that
they are made up of the same features).
We first give an example of how the Superset Principle works by using
Cardinaletti and Starke’s (1999) classification of pronouns (who build on
Kayne 1975). This constitutes a good example of the kind of superset–subset
relations we are interested in here (see also Rocquet 2013). Table 1.3 is an
overview of the French pronominal system, with a distinction made among
strong pronouns, weak (subject) pronouns, and clitic (object) pronouns. Note
that weak pronouns and clitic pronouns are grouped together as “deficient,”
as will become clearer.
According to Cardinaletti and Starke (1999), these three pronoun classes
must be distinguished in terms of semantics, syntax, morphology, and prosody.
We sum up the differences here but refer to their paper for more details.
Semantically, strong pronouns must be referential, whereas weak and clitic
pronouns do not need to be (i.e. deficient pronouns can be expletive and im-
personal). When they refer, weak pronouns and clitics need to be associated
to a prominent discourse antecedent. That is not the case for strong pronouns
(see Cardinaletti and Starke 1999, section 2.5, for a more thorough definition
of referentiality).
The strong pronouns appear in thematic positions, but not the weak and
clitic ones. Syntactically, strong pronouns can be coordinated, be moved to
left-peripheral positions, and be modified by adverbs. Even though weak
13. The reader should note that two of these “principles” (Superset and Cyclic
Override) are sometimes referred to as theorems in the literature (see Starke 2009,
2011b and Starke Chapter 4), as they follow logically from the basic theoretical
assumptions or “axioms” of the nanosyntactic approach. We continue to refer to them
as principles.
28
Table 1.3 FRENCH PRONOUNS
STRONG WEAK (SUBJECT) CLITIC (OBJECT)

SG 1 moi je me
2 toi tu te
3 M lui il le
F elle elle la
PL 1 nous nous nous
2 vous vous vous
3 eux M ils les
F elles
pronouns and clitics are grouped together as deficient, they must also be
distinguished from each other, in that weak pronouns occupy XP positions,
whereas clitics crucially do not. Morphologically, clitics are more deficient
than weak pronouns, and weak pronouns tend to be more deficient than
strong pronouns. In terms of prosody, only deficient pronouns may restruc-
ture, that is, only weak pronouns and clitics can “form a single unit with an
adjacent lexical element” (Cardinaletti and Starke 1999, 159).
Cardinaletti and Starke (1999) analyze these differences in terms of struc-
tural deficiency and structural containment. The more deficient a pronoun is,
the less structure it displays. Whereas strong pronouns contain heads with
referential and human features, weak pronouns and clitics do not. Strong
and weak pronouns share the head that realizes prosodic features (Σ in
Figure 1.10), whereas clitics do not. Finally, clitics are the most deficient as
their structure is composed of phi-features only. Summarizing: The structure
of strong pronouns contains that of weak pronouns, and weak pronouns, in
turn, contain the structure of clitics. This can be expressed as the superset–
subset relations in Figure 1.10.
C > Σ > I
clitic
weak
strong
Figure 1.10 Superset–subset relations in pronouns
Now consider (14), where it can be seen that strong pronoun elle ‘she’
and weak pronoun elle ‘she’ are syncretic. In (14a), elle ‘her’ is in a thematic
[ 28 ] Background
position (after the preposition quant à ‘as for’), and as such it is a strong pro-
noun. Moreover it does not prosodically restructure with the preposition. In
(14b), elle ‘she’ is in the subject position and prosodically restructures with the
verb joue ‘play,’ that is, it is a weak pronoun.
(14) a. Pierre travaille. Marie, quant à elle, joue sur la plage.
Pierre works Marie as to sheSTRONG plays on the beach
‘Pierre is working. As for Marie, she’s playing on the beach.’
b. Elle joue sur la plage.
sheWEAK plays on the beach.
‘She’s playing on the beach.’ [Rocquet 2013, 23, her (41)]
For Cardinaletti and Starke (1999), the structure of elleSTRONG ‘she’ contains the
structure of elleWEAK. In Figure 1.11 we have a very simple lexicon filled with a
single lexical entry, as well as two S-trees, S1 for the strong 3f.sg pronoun and
S2 for the weak 3f.sg pronoun.
(L1)
CP
ΣP
C
/ εl / IP
Σ
I
(S1) S-tree for strong 3f.sg pronoun (S2) S-tree for weak 3f.sg pronoun
CP ⇒ elle
ΣP ΣP ⇒ elle
C
IP
IP
Σ Σ
I I
Figure 1.11 Strong–weak syncretism in elle
By the Superset Principle, L1 can spell out either S1 or S2. For S1

there is a perfect match with L1, so S1 spells out as /ɛ l/ (elle), as this is
the phonological form specified in L1. As for S2, though it is not per-
fectly matched by L1, L1 is nevertheless a superset of S2. That is, the
L-t ree [CP [ΣP [IP]]] contains the S-t ree [ΣP [IP]]. Thus S2 also spells
out as elle.
The Superset Principle is at the heart of how syncretism is accounted for
in nanosyntax, because it is precisely this principle that allows for a single
L-tree to match multiple S-trees. In Figure 1.11, there is a single lexical entry
30
that applies in multiple syntactic environments. The element elle instantiates

a strong/weak syncretism.14
(ii) The Elsewhere Principle. The second principle of spellout is known as the
Elsewhere Principle (see also Kiparsky 1973). This principle guarantees that a
more specific lexical entry will take precedence over a more general lexical entry.
(15) Elsewhere Principle

If more than one L-tree can lexicalize the same S-tree (by the
Superset Principle), then the L-tree with the least amount of
superfluous material is chosen.
This can also be called, more informally, the Best Fit Principle or Minimize Junk.
Let us return to our French example in (14) (see also Rocquet 2013, 24–25).
In French, the strong 3m.sg pronoun is lui ‘him.’ From our preceding discus-
sion of the Superset Principle we might expect that lui ‘him’ will be spelled
out in both strong and weak environments, parallel to elle ‘her’ in (14a) and
(14b). After all, the lexical structure of the strong pronoun lui ‘him’ is [CP [ΣP
[IP]]], which is a superset of the structure of the weak pronoun, [ΣP [IP]].
This is not the case, however, because there is another, separate lexical item
that competes with lui ‘him,’ namely the weak 3m.sg pronoun il ‘he.’ As seen
in (16), after the preposition quant à ‘as for’ the weak pronoun il ‘he’ cannot
be used (16a), whereas it can perfectly occur in the subject position in (16b).
(16) a. Marie travaille. Pierre, quant à lui / *il, joue sur la plage.
M. works P. as to himSTRONG / heWEAK plays on the beach
‘Marie is working. As for Pierre, he is playing on the beach.’
b. Il / #Lui joue sur la plage.

heWEAK / heSTRONG plays on the beach
‘He is playing on the beach.’ [Rocquet 2013, 24, her (24)]
The Elsewhere Principle accounts for this. For this example our lexicon
includes two lexical entries, as seen in Figure 1.12.
As in the previous example, the maximal structure in L2 is once again a
suitable match for S3. The other lexical entry, L3, on the other hand, is not a
suitable match for S3 because L3 lacks the top head, C. Thus S3 spells out as
lui ‘him.’ Now consider S4 in Figure 1.12. Note that by the Superset Principle,
both L2 and L3 are suitable matches for S4. That is, L2 is a superset of S4, and
14. More can be said about the Superset Principle. For example, the Anchor Condition
(Abels and Muriungi 2008; Caha 2009, 89) can be seen as a condition on the Superset
Principle. It states that the lowest feature in an L-tree must be matched by the S-tree,
which has implications for cases in which more than one entry competes to lexicalize
the same feature. Such details, however, go beyond the scope of this introduction (but
see Taraldsen Chapter 3, 90–91).
[ 30 ] Background
(L2) (L3)
CP
ΣP ΣP
C IP
/ lwi / / il / IP
Σ Σ
I I
(S3) S-tree for strong 3m.sg pronoun (S4) S-tree for weak 3m.sg pronoun
CP ⇒ lui
ΣP ΣP ⇒ il
C
IP IP
Σ Σ
I I
Figure 1.12 Strong lui vs. weak il
L3 matches S4 exactly. Here the Elsewhere Principle steps in as referee: L3 is

a better fit for S4 (because L2 has an extra feature C, which is absent in L3),
and for this reason L3 gets to lexicalize the S-tree, and S4 spells out as il ‘he.’
Now that we have discussed and exemplified these two principles, we
are in a position to understand the *ABA theorem, which plays an impor-
tant role in syncretism and thus in determining which features are merged
adjacently in the functional sequence. Caha (2009, section 2.3) shows that
the *ABA theorem, on nanosyntactic assumptions, actually derives from a
combination of the Superset Principle and the Elsewhere Principle.
The *ABA theorem is about syncretism patterns. It states that spellout
patterns such as the one in Figure 1.13 should not be possible.
The principles of spellout can account for the ban on ABA patterns. In
an attempt to generate the pattern in Figure 1.13, we might posit the lexical
entries in Figure 1.14.
By the Superset Principle, L5 can map onto either S5 [XP] or S7 [ZP [YP
[XP]]], which is what would be needed in the ABA pattern in Figure 1.13. L4,
(S5) (S6) (S7)
ZP ⇒ a
XP ⇒ a YP ⇒ b YP
Z
XP XP
Y Y
X
X X
Figure 1.13 *ABA pattern
32
(L4) (L5)
ZP
YP
YP
b a Z
XP
Y XP
Y
X
X
Figure 1.14 Two hypothetical lexical entries for b and a
furthermore, perfectly matches S6 [YP [XP]], as is also required in Figure 1.13.

Thus, if we had only the Superset Principle to govern spellout, then the ABA
pattern in Figure 1.13 might (in principle) very well be possible.
However, once we take the Elsewhere Principle into account, the derivation
of the ABA pattern is blocked. Even though L5 in Figure 1.14 can map onto
the S-tree [XP] by the Superset Principle, it will be prevented from doing so
because L4 is a better match. This is because L4 has only one extra feature (Y),
whereas L5 has two extra features (Y and Z) compared with the S-tree [XP].
The spellout results are summarized in (17).
(17) S-tree Spellout

[XP] => b (both L4 and L5 match, but L4 is a better fit)
[YP [XP]] => b (both L4 and L5 match, but L4 is a perfect fit)
[ZP [YP [XP]]] => a (only L5 is a match)
Thus the Elsewhere Principle, by constraining the application of the Superset

Principle, ends up blocking the ABA pattern.
(iii) The Principle of Cyclic Override. The third principle of spellout is known
as the Principle of Cyclic Override (or the Biggest Wins Theorem), stated
in (18).
(18) The Principle of Cyclic Override

Previous lexicalizations are overridden or canceled out by later
lexicalizations.
In a derivational system that builds structure from the bottom up,
the application of this principle is quite intuitive. To illustrate let us
complete our paradigm of French pronouns by expanding our lexicon
once more to include three lexical entries (Figure 1.15), one for the
clitic le ‘him’ (L6), one for the weak pronoun il ‘he’ (L7), and one for
the strong pronoun lui ‘him’ (L8).
Consider now the derivational history of lui ‘him’ in Figure 1.16.
[ 32 ] Background
(L6) (L7) (L8)
CP
IP ΣP ΣP
/l / / il / / lwi / C IP
IP
Σ Σ
I I I
Figure 1.15 Lexical entries for lui, il, and le
(S8) > (S9) > (S10)
CP ⇒ lui
ΣP ⇒ il Σ P ⇒ il
IP ⇒ le C
IP ⇒ le IP
Σ Σ
I I I
Figure 1.16 Cyclic Override in the derivation of lui
Structures are built one feature at a time. The first step in building lui ‘him’
is to build IP (S8), which spells out as le ‘him’ because L6 is the best match
by the Elsewhere Principle. Next the feature Σ is added, resulting in S9. This
structure, [ΣP [IP]], spells out as il ‘he’ because L7 is the best match. The
spellout of il ‘him,’ moreover, overrides the now-redundant, previous spellout
le ‘him.’ Finally, the feature C is added (S10). The resulting structure [CP [ΣP
[IP]]] spells out as lui ‘him’ because L8 is the best match. The spellout lui ‘him’
overrides the lower-level spellout il ‘he.’
Note that Cyclic Override does not cancel the SMS structure in any way.
Rather, it ensures that the system stays “up to date” with the latest and most
efficient matches between SMS and the lexicon. This has the most crucial
effect on phonology: Although both grammatical (SMS) and conceptual infor-
mation can be built up compositionally, phonology (being constrained by line-
arization) must be constantly choosing the latest best form for pronunciation.
1.3.3.3 Phrasal spellout and idioms
Idioms with their various idiosyncrasies are often considered to pose problems
for standard theories of syntax. However, in a system that allows for phrasal
spellout, such as nanosyntax, idioms are easier to understand.
Idioms are prima facie an important source of support for [the nanosyntactic
notion of] phrasal spellout. Within the traditional approach, there is no easy way
34
to handle multi-word idiomatic expressions, as witnessed by the clunkiness of

existing attempts at handling idioms while at the same time confining spellout
to terminals. Under phrasal spellout, idioms are natural: they are cases in which
a relatively high-level constituent has been stored.
(Starke 2011a, 6)
Note that this “high-level constituent” is not confined to simple VPs or NPs
but can include functional layers for aspect, tense, definiteness, and so forth.
There are two basic kinds of idioms: phonological and conceptual. Phonological
idioms are cases when a phonologically irregular form replaces a “regular” form,
such as when children replaces *child-s or mice replaces *mouse-s. Conceptual
idioms, on the other hand, involve idiosyncratic conceptual information replacing
the regular concepts in a certain phrase, such as when the basic conceptual infor-
mation associated with the individual items in kick the bucket is replaced by the
concept ‘die,’ or when hold your horses is interpreted as ‘be patient.’
First consider phonological idioms. As seen in Figure 1.17, the irregular
plural of mouse is mice rather than the regularly formed *mouse-s. We can say,
then, that mouse-s is built regularly at some point in the derivation but that
this particular combination of lexical items is phonologically overridden at a
higher node by mice.
In Figure 1.17 we see that there is a lexical entry containing the SMS struc-
ture [NP mouse] + [pl] linked to the phonology /maɪs/ (i.e. mice).15 Learning
an irregular form, then, amounts to storing a particular lexical entry in the
lexicon to ensure that *mouse-s surfaces as mice instead (Starke 2009; see also
Caha Chapter 2 for discussion).
⇒ mice
NumP ⇒ –s
NP ⇒ mouse / ma s / [[NP mouse] [NumP PL]]

PL
Figure 1.17 The form mice as a phonological idiom
15. When a lexical entry itself refers to other, independent lexical entries (i.e.
[NP mouse] and the plural marker), then we are making use of “pointers.” A pointer is a
way to refer to a lexical item within another lexical item. The idiomatic entry for mice
contains (or points to) two other lexical entries: one for the NP mouse and another for
the plural marker. The idiomatic entry for hold your horses ‘be patient’ points to at least
four other lexical entries: hold, your, horse, and the plural marker. It has been suggested
that pointers can also be used to encode more complex paradigmatic patterns involving
multiple dimensions of grammatical features (e.g. case, gender, number) (Caha and
Pantcheva 2012; Starke 2013). See Vanden Wyngaerd (Chapter 11) for more details.
[ 34 ] Background
Consider next a conceptual idiom like hold your horses, with the special in-
terpretation ‘be patient.’ What we need to say about this idiom is that the in-
dividual concepts associated with hold, your, and horses are replaced, or at least
receive an additional interpretation, at the highest node.
In Figure 1.18 we again see that there is a lexical entry containing a cer-
tain SMS structure, basically [VP hold your horse-s], which is linked to special
conceptual information that cannot be accessed or deduced simply from the
conceptual information associated with the individual items hold, your, and
horse(-s). The special interpretation of ‘be patient,’ then, is due to the fact that
there is a lexical entry linking the specific phrase hold your horses to additional
conceptual information concerning patience.
BE PATIENT ⇐ VP
hold ⇐ [VP hold [your [horse-s]]] BE PATIENT

⇒ your
⇒ horse-s
Figure 1.18 Hold your horses as a conceptual idiom
There are some important points about the nanosyntactic view of idioms
that should be highlighted here. First, idioms illustrate that not all lexical
entries have their own phonology and conceptual content. Conceptual
idioms do not have their own phonology because they simply hijack the
phonology of already-existing lexical items like hold, your, and horses.
Similarly, phonological idioms do not have their own conceptual content,
because they refer to the conceptual content of already-existing lexical
items like mouse.16 Either way, however, some kind of SMS structure must
be in the central slot.
Second, the Principle of Cyclic Override is strictly relevant only for phono-
logical idioms, not for conceptual idioms. As a reviewer points out, phonolog-
ical idioms involve obligatory override (in the sense that *mouses instead of
mice or *goed instead of went is simply ungrammatical), whereas conceptual
idioms can have both the literal and idiomatic interpretations available. We
suggest that this is due to the relative complexity of the systems involved.
As previously mentioned, although it is impossible to say two things at once
16. It is interesting to note that this view of the lexicon—that structures of all sorts
and sizes can be stored in the lexicon, and that some lexical entries are “deficient” in
some sense—is similar to Jackendoff’s (2002, ch. 6) conception of the lexicon and lex-
ical storage.
36
(forcing the choice of one form over another), it is perfectly possible to think
about multiple things at once.17
1.3.3.4 Spellout-driven movement
There are three principles of matching and phrasal spellout: the Superset
Principle, the Elsewhere Principle, and Cyclic Override. These principles
and the way they determine how phrasal spellout proceeds result in a dy-
namic view of language that emphasizes “size differences” to account for
a range of syntactic phenomena both within a single language and across
languages.
In this section we show how these principles are put to use in the course of
a full derivation. Importantly, we see that syntactic structure needs to be al-
tered to provide an appropriate structural configuration for spellout to succeed
according to the principles just discussed. This alteration of structure for the
purposes of aiding spellout is known as spellout-driven movement.18 Here we
present one influential view, developed by M. Starke (e.g. Starke 2011b; see
also Starke Chapter 9), which can be characterized as the “strict constituent
hood condition” on spellout. According to this view, only constituents can be
targeted for spellout. Before continuing, a proviso: Not all researchers adopting
nanosyntax share this particular implementation of nanosyntax, spanning
approaches being one prominent alternative. See Taraldsen (Chapter 3) for
discussion of the two approaches.
When a syntactic structure has to be lexicalized, the lexicon is consulted
to see if any lexical entries are available to match the syntactic struc-
ture. This happens in a stepwise fashion: Nanosyntacticians often assume
that structures are built one head at a time, and at each layer the structure
17. Directly related to this is an important issue raised by a reviewer, namely why

functional elements appearing within an idiom often have a completely regular, com-
positional effect on the interpretation of the idiom, especially in terms of lexical as-
pect (see Marantz 1997; McGinnis 2002; Harley 2014; among others), whereas the
nanosyntactic approach might seem to suggest that the idiomatic interpretation
linked to a given constituent would destroy any such internal structural regularities
because of Cyclic Override. It is crucial here to recognize the division between syntactic
(SMS) structure and conceptual information. Any aspectual regularities observed in
the interpretation of idioms clearly belong to the domain of SMS. Cyclic Override does
not cancel SMS structure in any way; rather, it makes sure that the latest phonological
realization is up to date with the derivation. In other words, the SMS structure stays
the same no matter what; any special “encyclopedic” information that may (or may
not) end up becoming associated with this structure is additional, not affecting the
functional or grammatical core of the phrase.
18. Not all movement is spellout-driven. Determining the exact nature of non-
spellout-driven movement (e.g. wh-movement) and how it fits into nanosyntactic
theory is still a topic for future work. See Starke (2011a) for discussion.
[ 36 ] Background
must be successfully lexicalized (an approach known as Cyclic Exhaustive
Lexicalization; Fábregas 2007; Ramchand 2008; Pantcheva 2011). At each
cycle, the structure has multiple attempts at successful lexicalization, corre-
sponding to the steps in (19). Crucially, ultimately there can be no parts of the
structure that remain unlexicalized.
To start with, spellout-driven movement is governed by the algorithm
in (19).
(19) stay > cyclic > snowball
See Aboh (2004b) for Snowball movement and Cinque (2005) for Cyclic and
Snowball movement.
First let us illustrate the application of algorithm (19) in abstract terms, be-
fore providing a concrete example. Suppose that, in the course of a derivation,
[HP [GP] . . . ] has been formed and spelled out and that at the next step the
feature F has been added to the structure [HP [GP] . . . ], as in Figure 1.19. In this
structure [HP [GP] . . . ] has been spelled out, but F has not. To spell out the fea-
ture F, the structure first stays as is, and the lexicon is checked for a lexical entry
containing the structure [FP F [HP [GP] . . . ]] (Figure 1.19). If there is no suitable
entry, then the leftmost daughter of the sister of F, GP, is evacuated to the left of F,
and the lexicon is checked for a lexical entry containing the constituent that is
left over, namely [FP [HP . . . ]] (Figure 1.20). If there is no suitable entry, then
the cyclic movement is undone and the sister of F is evacuated to the left of F, and the
lexicon is checked for the constituent that is left over, namely FP (Figure 1.21).
STAY check lexicon for:
FP FP
HP HP
F F
GP GP
Figure 1.19 stay in the spellout algorithm
Spellout-driven movement is a mechanical procedure for generating new

constituents that are candidates for lexicalization. We note here also that
spellout-driven movement owes much to Cinque’s (2005) U20 theory (see
Section 1.3): A constituent containing the head of the extended projection
(which is embedded within GP above) undergoes phrasal movement to the
38
CYCLIC check lexicon for:
FP FP
GP
HP HP
F
F
GP
Figure 1.20 cyclic in the spellout algorithm
SNOWBALL check lexicon for:
FP
FP HP
F
HP
F
GP
GP
Figure 1.21 snowball in the spellout algorithm
left and may involve different degrees of pied-piping.19 It is often assumed,

following much other work in the framework, that the landing site for spellout-
driven movement is an unlabeled specifier and that this kind of movement
leaves no traces (Starke 2011b, as well as Chapter 9 of this volume).
Let us now turn to a concrete example of spellout-driven movement, using
Pantcheva’s (2011, section 7.3) nanosyntactic account of the ON-series, a
set of morphemes related to the word for ‘on,’ in Karata. In (20) we see the
morphological expression of Place, Goal, and a syncretic Source/Route ‘from/
through’ in Karata.
19. Although in this case the kind of pied-piping involved is of the whose book
type, pied-piping of the book of who type also arises (see Cinque 2005, 321), as
subsequently seen.
[ 38 ] Background
(20) a. bajdan-ʈ’-a
square-on-loc
‘on the square’
b. bajdan-ʈ’-a-r
square-on-loc-goal
‘to the square’
c. bajdan-ʈ’-a-gal
square-on-loc-source/route
‘from/through the square’ (Pantcheva 2011, 137)
Pantcheva proposes the functional sequence in (21a) with the lexical entries
for the morphemes in (21b).
(21) a. Route > Source > Goal > Place > AxPart > . . .DP
b. < -ʈ’ ⇔ AxPartP ⇔ on >
< -a ⇔ PlaceP >
< -r ⇔ GoalP >
< -gal ⇔ [RouteP [SourceP [GoalP]]] >
AxPart stands for ‘axial part,’ referring to an object’s position with respect
to some axis (i.e. ‘front,’ ‘back,’ and so on; see Svenonius 2006). The Karata
morpheme -ʈ’ encodes the AxPart ON. The morpheme -a is a locative marker,
correponding to PlaceP. The morpheme -r is used to express the Goal reading, and
it too builds on top of both the Place and AxPart markers. Thus -r corresponds
to GoalP. Finally, we see that -gal, which is syncretic between Source and Route
readings, corresponds to the full structure [RouteP [SourceP [GoalP]]].
To derive the Route structure bajdan-ʈ’-a-gal ‘through the square,’ the der-
ivation proceeds as follows. We start at Figure 1.22, where DP has been built
and matched with bajdan ‘square.’
Next, in Figure 1.23, the AxPart layer is added to the structure. With the
first two steps in the algorithm not producing a suitable match,20 finally the
third step results in a match for AxPartP.21
20. Although we do not take a stand on whether individual-terminal spellout can or

should be allowed in the spellout system, here we assume that terminals cannot be
lexicalized (Starke 2011b, though see Pantcheva 2011, section 6.3.2, for a system that
consistently allows for terminal spellout). See also Lander (2015b, section 5.1.2) for
discussion.
21. Note that antilocality (Abels 2003) does not apply here, as phases are not relevant
for us. Furthermore, it is important to note that the kind of movement discussed here
(spellout-driven) seems to be different from more traditionally studied long-distance
syntactic movement (not least because the latter leaves traces, and the former does
not; Starke p.c.).
40
DP ⇒ bajdan
Figure 1.22 Spelling out DP
STAY: no match in the lexicon for [AxPartP [DP]]
AxPartP ⇒ *
AxPart
DP
CYCLIC: not applicable (NA)
SNOWBALL: move DP to the left of AxPart, match AxPartP with < AxPartP on >
bajdan ⇐ DP
AxPartP ⇒
AxPart
Figure 1.23 Spelling out the AxPart layer
Next, in Figure 1.24, Place is added to the structure. Once again there is
no suitable match until the third step in the algorithm, when PlaceP can be
lexicalized.
It is worth noting that at this point we have a complete structure expressing
Location and meaning ‘on the square,’ bajdan-ʈ’-a. In this example, however,
the syntax is aiming at the Route expression ‘through the square,’ meaning it
has to build all the way up to RouteP. Thus the syntax continues building the
fseq, with the Goal layer. Note in Figure 1.25 that at the snowball step there
are two matches for GoalP (both < -r ⇔ GoalP > and < -gal ⇔ [RouteP [SourceP
[GoalP]]] >) by the Superset Principle, but that -r wins by the Elsewhere
Principle, yielding the Goal structure.
Next the syntax adds Source. As seen in Figure 1.26, a successful match
occurs at the second step in the algorithm (because by the Superset Principle
the S-tree [SourceP [GoalP]] is a subtree of the L-tree [RouteP [SourceP
[GoalP]]]), yielding the Source structure.
Finally, in Figure 1.27, the derivation reaches the Route layer, and again the
second step in the algorithm delivers a successful match.
The final product in Figure 1.27 is the structure with the Route reading of
-gal. Now, as seen in (20c), -gal syncretizes Route (‘through the square’) and
[ 40 ] Background
STAY: no match for entire structure
PlaceP ⇒ *
Place
DP AxPartP
AxPart
CYCLIC: move DP, no match for [PlaceP [AxPartP]]
bajdan ⇐ DP PlaceP ⇒ *
Place AxPartP
AxPart
SNOWBALL: undo CYCLIC and raise [DP AxPartP]; match PlaceP with < -a PlaceP >
bajdan ⇐ DP
AxPartP ⇒ PlaceP ⇒ -a
AxPart Place
Figure 1.24 Spelling out the Place layer
Source (‘from the square’) readings. This is structurally captured by the L-tree
[RouteP [SourceP [GoalP]]], where Source is a subset of Route.
1.3.3.5 Pre-elements
As an addendum on the topic of spellout-driven movement, it is important

to recognize that the system sketched here is perfectly suited to accounting
for suffixes, that is, “right-adjoined” morphemes. All the morphemes discussed
previously have been suffixes, and these are easily derived because so far we
have always moved parts of the complement to the left to spell out constituents
on the right. On the other hand, the question of how to derive “pre-elements”
like prefixes and prepositions is less clear, because these need to stay in situ and
42

GoalP ⇒ *
Goal
DP AxPartP PlaceP
AxPart Place
CYCLIC: move [DP AxPartP], no match for [GoalP [PlaceP]]
bajdan ⇐ DP AxPartP ⇒ GoalP ⇒ *
AxPart Goal PlaceP
Place
SNOWBALL: undo CYCLIC and raise [[DP AxPartP] PlaceP]; match GoalP with < -r GoalP >
bajdan ⇐ DP AxPartP ⇒ PlaceP ⇒ -a GoalP ⇒ -r
AxPart Place Goal
Figure 1.25 Spelling out the Goal layer
precede the item they combine with rather than being swapped by spellout-
driven movement (which would turn them into suffixes and postpositions).
There are various proposals in the literature. Using the German preposition mit
as an example, we look at three ways to account for pre-elements: spanning,
head movement, and the use of an additional workspace.
On the first view, namely spanning, spellout does not require strict
constituenthood in order for matching to take place. A span can be defined
as “a nontrivial sequence of heads” (Taraldsen Chapter 3, 88). For instance,
[ 42 ] Background
SourceP ⇒ *
Source
DP AxPartP PlaceP GoalP
AxPart Place Goal
CYCLIC: move [[DP AxPartP] PlaceP]; match [SourceP [GoalP]] with

< -gal [RouteP [SourceP [GoalP]]] >
bajdan ⇐ DP AxPartP ⇒ PlaceP ⇒ -a SourceP ⇒ -gal
AxPart Place Source GoalP
Goal
Figure 1.26 Spelling out the Source layer
consider the case of the comitative preposition mit in German, which selects
a dative complement (see Caha 2009, 65–67; also Caha 2010). As seen in
Figure 1.28, comitative mit corresponds to K6 and K5, but there is no node
containing only K6 and K5 to the exclusion of the dative DP. Thus K6 and K5
do not form a constituent. On the spanning view, the preposition can be
lexicalized in situ, even though it does not constitute a strict constituent.22
For this kind of approach see Abels and Muriungi (2008), Ramchand (2008),
and Dékány (2009).
22. Had the preposition been an instrumental, corresponding only to K5, then we
might need to resort to terminal spellout. We do not commit to one or the other
approach.
44

RouteP ⇒ *
Route
DP AxPartP PlaceP SourceP
AxPart Place Source GoalP
Goal
CYCLIC: move [[DP AxPartP] PlaceP]; match [RouteP [SourceP [GoalP]]] with
< -gal [RouteP [SourceP [GoalP]]] >
bajdan ⇐ DP AxPartP ⇒ PlaceP ⇒ -a RouteP ⇒ -gal
SourceP
AxPart Place Route
Source GoalP
Goal
Figure 1.27 Spelling out the Route layer
K6P
K5P
K6
K5
mit K4P ⇒ DAT
DP
Figure 1.28 Spelling out nonconstituent mit by spanning

A second suggestion proposed by Caha (2010) is that prefixal structures
are formed by something like traditional Travis (1984)–Baker (1988) head
movement.23 This would mean that head movement is allowed in certain
cases, contra the strict U20-style approach referred to in Section 1.1.3.24
In the approach illustrated in Figure 1.28, the order of heads in the func-
tional sequence is kept intact (i.e. K6 > K5). In Figure 1.29, however, the heads
are reversed by movement (i.e. K5 > K6). Caha (2010, 28–29, fn.12) points out
that although this approach defies the U20 ban on head movement, “such
a movement is only allowed in case [K5] and [K6] are spelled out by a single
morpheme, which renders their ordering opaque . . . thus, admitting head
movement in such a constrained fashion does not, in fact, lead to orderings
incompatible with the generalizations observed in [Cinque (2005)].” An im-
portant point is that mit is a constituent in Figure 1.29, which makes this
approach stricter than one that requires mit to correspond simply to a span of
features as in Figure 1.28.
K6P
K5P
mit ⇐ K6
K5 K6 tK5
K4P ⇒ DAT
DP
Figure 1.29 Travis-Baker head movement for mit
A third approach that requires constituenthood in order for spellout to ob-

tain has been advocated by Pantcheva (2011), Starke (2013 and Chapter 9 of
this volume), and Taraldsen in (Chapter 3). To illustrate this line of thinking,
we must first imagine that syntactic structures can be built in more than a
single cognitive “workspace.” One way of deriving prefixes, then, would be to
posit that a complex head structure can be constructed in a secondary work-
space (Workspace-2 in Figure 1.30) and then subsequently merged into the
primary workspace (Workspace-1 in Figure 1.30). Because the complex head
23. Baker (1985) observes that the order in which affixes appear correlates with the
order of the syntactic operations they trigger. This is known as the Mirror Principle.
Mirror Principle effects can be derived if complex words are formed by head movement,
which is subject to the Head Movement Constraint (Travis 1984).
24. Note that even if a system disallows individual-terminal spellout, this does not
mean that only XPs can be targeted for spellout. Complex heads (such as the set [K6 K5
K6] in Figure 1.29) can also be targeted because they are made up of more than a single
individual terminal.
46
Workspace-2 Workspace-1
mit ⇐
K6 K5
mit ⇐
K6 K5 K4P ⇒ DAT
DP
Figure 1.30 Building mit in a separate workspace
is in this case constructed separately in its own workspace and does not result
from head movement, its internal ordering of features follows the fseq (i.e.
K6 > K5).
Again, the complex head merged into the primary workspace is a proper
constituent.25,26
Important questions remain, however, as to how exactly we should in-
tegrate the building of complex heads (pre-elements like prepositions and
prefixes) into the spellout algorithm discussed previously. At this stage we
may think of the need to accommodate prefixal or complex head structures
as adding another step to this algorithm, either stay > cyclic > snowball
> head-move (see Figure 1.29) or stay > cyclic > snowball > construct
(see Figure 1.30).
1.4. PRINCIPLES AND PARAMETERS
In nanosyntax, the atoms of linguistic structure are understood to be

syntactico-semantic features merged as syntactic heads according to an in-
variant and universal fseq. To determine what this set of universal features is,
25. Yet another option is brought up by Caha (2009, 66), who discusses a remnant
movement analysis for spelling out prepositions. However, he rejects this option on
the grounds that it does not conform to Cinque’s (2005) U20 rules for movement (i.e.
the requirement that all moved phrases contain the head noun, meaning no remnant
movement is allowed).
26. According to Starke (2013 and Chapter 9, Section 9.1.1), the lexically stored
difference between a suffixal and a prefixal structure, then, can be thought of in terms
of what kind of a set is found at the bottom of the tree. The bottom of a suffixal tree will
be a singleton set, because its complement has been evacuated to the left. The bottom
of a prefixal tree will be a binary set, because this kind of structure is built from scratch,
and merge always joins two elements together (see Kayne 1984).
[ 46 ] Background
nanosyntacticians rely on the detailed study of morphosyntactic phenomena
in a wide range of languages. In terms of the Principles and Parameters model
(Chomsky 1981, 1986), the universal fseq—along with basic architectural
properties of the language faculty, principles of merge and movement, the
spellout algorithm, among others—would constitute the invariant Principles
of language or UG.
As we have seen in Section 1.3, the shape of L-trees determines how
the spellout algorithm proceeds. In other words, because L- trees differ
crosslinguistically, the way S-trees are matched by L-trees during spellout
will also differ across languages. Thus languages spell out structures differ-
ently according to the content of their lexicon. In this way, variation can be
explained purely in terms of differences in the lexicon (see Chomsky 2001,
2; see Starke 2011a for more discussion). Even though the fseq is the same
across languages, lexically stored structures (i.e. the way the fseq is packaged
up) will vary from language to language. This packaging can be thought of as
the Parameters of language.
To take a concrete example, let us consider another example from
Pantcheva’s (2011) study of Path expressions. The Macedonian item nakaj,
‘to(ward),’ can be decomposed into the locative morpheme kaj ‘at’ (call this
Place) and na- ‘to’ (call this simply Path), as sketched in Figure 1.31. Dutch
naar ‘to(ward),’ on the other hand, is not overtly decomposable: It is an indi-
visible portmanteau of Place and Path, as sketched in Figure 1.32.
PathP
na- ⇐ Path PlaceP ⇒ kaj
Place
Figure 1.31 Macedonian na-kaj
PathP
⇒ naar
Path PlaceP
Place
Figure 1.32 Dutch naar
48
Although simplified, this example illustrates how Macedonian and Dutch

package the fseq differently: Macedonian splits it up into two L-trees, one for
na-and one for kaj, whereas Dutch stores it as a single unit naar.
To take another example of crosslinguistic variation, consider Finnish versus
English case endings. Consider the partial paradigms for ‘bear’ in Table 1.4 for
Finnish and Table 1.5 for English.
In Finnish the ending -n shows a gen–acc syncretism, with a distinct nom
ending (-Ø), as illustrated in Figure 1.33. In English, on the other hand, the
ending -Ø shows a nom–acc syncretism, with a distinct gen ending (-s), as
in Figure 1.34.
The way the fseq is lexically partitioned in Finnish versus English leads to a
crosslinguistic difference once spellout occurs.
Table 1.4 FINNISH (CAHA 2009, 115)
‘bear’ (sg)
NOM karhu-Ø
ACC karhu-n
GEN karhu-n
Table 1.5 ENGLISH
‘bear’ (sg)
NOM bear-Ø
ACC bear-Ø
GEN bear-s
K3P
K3 K2P ⇒ GEN–ACC -n
⇒ NOM -Ø
K2 K1P
K1
Figure 1.33 Lexical packaging of nom, acc, gen in Finnish
[ 48 ] Background
K3P
⇒ GEN-s
K3 K2P
⇒ NOM–acc -Ø
K2 K1P
K1
Figure 1.34 Lexical packaging of nom, acc, gen in English
The nanosyntactic approach to variation in terms of lexical storage can

also be usefully applied to phenomena at the clausal level. For an account of
microvariation in wh-movement in different varieties of Spanish, see Fábregas
(Chapter 10).
1.5 CONCLUSION
Nanosyntax is in essence a cartographic approach to linguistic structure,

and the internal structure of morphemes in particular. Like other carto-
graphic approaches, nanosyntax assumes a strict syntax–semantics mapping,
simplicity of syntactic projection (i.e. trees are binary-branching and right-
branching), and the OFOH principle, according to which every syntactico-
semantic feature corresponds to a head in the syntactic spine.
Nanosyntax allows for phrasal spellout, meaning that spellout does not
need to target heads or terminals but can target entire phrases. Phrasal
spellout is the nanosyntactic response to the observation that morphemes
have an internal structure, that is, that syntactic features and heads are
submorphemic. Accordingly, syntax becomes responsible for constructing
morphemes, and thus syntax feeds the lexicon.
A central concern of the theory is to determine precisely how spellout takes
place, that is, how syntactic structures are lexicalized by being matched by
structures in the lexicon. This process of spellout is governed by three prin-
ciples: the Superset Principle, the Elsewhere Principle, and the Principle of
Cyclic Override. In the course of the derivation, syntactic structure can be
altered in particular ways to create the structural constituents that will be
appropriate candidates for being matched by lexically stored structures
(according to the three principles mentioned); these alterations are achieved
by movement. This is known as spellout-driven movement, and it is governed
by the algorithm stay > cyclic > snowball.
50
More broadly, nanosyntax views syntax, morphology, and formal seman-

tics as unified in a single module, the computational system SMS. The SMS
module takes atomic features and merges them as heads according to the
ordering imposed by the fseq. This fseq is taken to be universal, belonging
to the Principles of language, whereas the language-specific way this fseq is
divided up into lexical entries across languages constitutes the Parameters of
language variation. This is, in short, the nanosyntactic view of the Principles
and Parameters framework (Starke 2011a). In this way nanosyntax contributes
to the continuing search for what is universal in language, what is language-
specific, and how the two interact.
REFERENCES
Abels, Klaus. 2003. Successive Cyclicity, Anti-locality, and Adposition Stranding. Doctoral
dissertation, University of Connecticut.
Abels, Klaus and Peter K. Muriungi. 2008. “The Focus Particle in Kîîtharaka: Syntax
and Semantics.” Lingua 118: pp. 687–731.
Abels, Klaus and Ad Neeleman. 2009. “Universal 20 without the LCA.” In Merging
Features: Computation, Interpretation and Acquisition, edited by Josep M. Brucart,
Anna Gavarró, and Ricardo J. Solà, pp. 60–79. Oxford: Oxford University Press.
Abels, Klaus and Ad Neeleman. 2012. “Linear asymmetries and the LCA.” Syntax 15
(1): pp. 25–74.
Abney, Steven Paul. 1987. The English Noun Phrase in Its Sentential Aspect. Doctoral dis-
sertation, MIT.
Aboh, Enoch Oladé. 2004a. The Morphosyntax of Complement- Head Sequences.
Aboh, Enoch Oladé. 2004b. “Snowballing Movement and Generalized Pied-Piping.” In
Triggers, edited by Anne Breitbarth and Henk van Riemsdijk, pp. 15–47. Berlin:
Mouton.
Alexiadou, Artemis, Liliane Haegeman, and Melita Stavrou. 2007. Noun Phrase in the
Generative Perspective. Berlin: Mouton de Gruyter.
Baker, Mark. 1985. “The Mirror Principle and Morphosyntactic Explanation.” Linguistic
Inquiry 16 (3): pp. 373–415.
Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing.
Chicago, IL: University of Chicago Press.
Baker, Mark. 2008. “The Macroparameter in a Microparametric World.” In The Limits of
Syntactic Variation, edited by Theresa Biberauer, pp. 351–374. Amsterdam: John
Benjamins.
Barbiers, Sjef. 2006. Er Zijn Grenzen aan Wat Je Kunt Zeggen. Utrecht, The Netherlands:
Oratie Universiteit Utrecht, Faculteit Geesteswetenschappen.
Barbiers, Sjef and Hans J. Bennis. 2007. “The Syntactic Atlas of the Dutch Dialects: A
Discussion of Choices in the SAND-project.” Nordlyd 34: pp. 53–72. [online]
Available at http://septentrio.uit.no/index.php/nordlyd/index
Baunaz, Lena. 2015. “On the Various Sizes of Complementizers.” Probus 27 (2):
pp. 193–236.
Baunaz, Lena. 2016. “Deconstructing Complementizers in Serbo-Croatian, Modern
Greek and Bulgarian.” In Proceedings of NELS 46 (1), edited by Christopher
[ 50 ] Background
Hammerly and Brandon Prickett, pp. 69–77. Amherst, MA: Graduate Linguistics
Student Association.
Baunaz, Lena and Eric Lander. 2017. “Syncretisms with Nominal Complementizers.”
Studia Linguistica: pp. 1–34. DOI: 10.1111/stul.12076.
Baunaz, Lena and Eric Lander. To appear. “Cross-Categorial Syncretism and the Slavic
Containment Puzzle.” In Balkan Syntax and (Universal) Principles of Grammar
(provisional title), edited by Iliyana Krapova and Brian Joseph. Berlin: Mouton
de Gruyter.
Beghelli, Filippo and Tim Stowell. 1997. “Distributivity and Negation: The Syntax of
each and every.” In Ways of Scope Taking, edited by Anna Szabolcsi, pp. 71–107.
Dordrecht, The Netherlands: Kluwer.
Belletti, Adriana. 1990. Generalized Verb Movement: Aspects of Verb Syntax. Turin,
Italy: Rosenberg and Sellier.
Belletti, Adriana (ed.). 2004. Structures and Beyond: The Cartography of Syntactic
Benincà, Paola. 1988. “L’ordine degli elementi della frase e le costruzioni marcate.”
In Grande Grammatica Italiana di Consultazione, edited by Lorenzo Renzi,
pp. 129–194. Bologna: Il Mulino.
Benincà, Paola and Laura Vanelli. 1982. “Appunti di sintassi veneta.” In Guida ai Dialetti
Veneti IV, edited by Michele Cortelazzo, pp. 7–38. Padua, Italy: CLEUP.
Benincà, Paola and Cecilia Poletto. 2004. “Topic, Focus and V2: Defining the CP
Sublayers.” In The Structure of CP and IP: The Cartography of Syntactic Structures,
Vol. 2, edited by Lorenzo Rizzi, pp. 52–75. New York: Oxford University Press.
Benincà, Paola, Adam Ledgeway, and Nigel Vincent. 2014. Diachrony and Dialects:
Grammatical Change in the Dialects of Italy. Oxford: Oxford University Press.
Blake, Barry. J. 1994 [2004]. Case. Cambridge: Cambridge University Press.
Bloomfield, Leonard. 1933. Language. New York: Holt.
Bobaljik, Jonathan. 2007. “On Comparative Suppletion.” Ms. University of Connecticut.
Bobaljik, Jonathan. 2012. Universals in Comparative Morphology: Suppletion, Superlatives,
and the Structure of Words. Cambridge, MA: MIT Press.
Bobaljik, Jonathan D. 2015. “Distributed Morphology.” Ms., University of Connecticut.
http://bobaljik.uconn.edu/papers/DM_ORE.pdf
Bocci, Giuliano. 2009. On Syntax and Prosody in Italian. Doctoral dissertation,
University of Siena.
Borer, Hagit. 1984. Parametric Syntax. Dordrecht, The Netherlands: Foris.
Caha, Pavel. 2010. “The Parameters of Case Marking and Spell Out Driven Movement.”
Linguistic Variation Yearbook 2010, edited by Jeroen van Craenenbroeck,
pp. 33–77. Amsterdam/Philadelphia: John Benjamins.
Caha, Pavel. 2013. “Explaining the Structure of Case Paradigms by the Mechanisms
of Nanosyntax: The Classical Armenian Nominal Declension.” Natural Language
and Linguistic Theory 31 (4): pp. 1015–1066.
Caha, Pavel and Marina Pantcheva. 2012. “Contiguity Beyond Linearity: Modeling
Cross- Dimensional Syncretisms.” Talk presented at Workshop on the
Representation and Selection of Exponents, University of Tromsø. June 7.
[online] Available at http://cms.unige.ch/lettres/linguistique/seminaire/media/
220/Caha%20Pantcheva%20231012.pdf.
Cardinaletti, Anna. 1997. “Subjects and Clause Structure.” In The New Comparative
Syntax, edited by Liliane Haegeman, pp. 33–63. London: Addison, Wesley,
Longman.
52
Cardinaletti, Anna. 2004. “Towards a Cartography of Subject Positions.” In The

Structure of CP and IP: The Cartography of Syntactic Structures, Vol. 2, edited by
Luigi Rizzi, pp. 115–165. New York: Oxford University Press.
Cardinaletti, Anna and Michal Starke. 1999. “The Typology of Structural Deficiency: A
Case Study of the Three Classes of Pronouns.” In Clitics in the Languages of
Europe, edited by Henk van Riemsdijk, pp. 145–233. Berlin: Mouton de Gruyter.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht, The
Netherlands: Foris.
Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2001. “Derivation by Phase.” In Ken Hale: A Life in Language, edited by
Michael Kenstowicz, 1–50. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 1990. Types of A’ dependencies. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and Inflectional Heads. Oxford: Oxford University
Press.
Cinque, Guglielmo (ed.). 2002. Functional Structure in DP and IP: The Cartography of
Cinque, Guglielmo. 2005. “Deriving Greenberg’s Universal 20 and Its Exceptions.”
Linguistic Inquiry 36 (3): pp. 315–332.
Cinque, Guglielmo. 2006. Restructuring and Functional Heads: The Cartography of
Cinque, Guglielmo. 2010. The Syntax of Adjectives. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 2013. Cognition, Universal Grammar, and Typological
Generalizations. Lingua 130: pp. 50–65.
In CISCL Working Papers on Language and Cognition, 2, edited by Vincenzo
Moscati, pp. 43–59. Universtity of Sienna, distributed by MIT Working Papers
in Linguistics.
De Clercq, Karen. 2013. A Unified Syntax of Negation. Doctoral dissertation, Ghent
University.
Dékány, Éva. 2009. “The Nanosyntax of Hungarian Postpositions.” Nordlyd 36: pp. 41–76.
[online] Available at http://septentrio.uit.no/index.php/nordlyd/index.
Den Dikken, Marcel. 2010. “On the Functional Structure of Locative and Directional
PPs.” In Mapping Spatial PPs: The Cartography of Syntactic Structures 6, edited by
Guglielmo Cinque and Luigi Rizzi, pp. 74–126. New York: Oxford University Press.
Dryer, Matthew. 2009. “On the Order of Demonstrative, Numeral, Adjective, and
Noun: An Alternative to Cinque.” [online] Available at http://attach.matita.net/
caterinamauri/sitovecchio/1898313034_cinqueH09.pdf.
Embick, David. 2015. The Morpheme: A Theoretical Introduction. Boston and Berlin:
Mouton de Gruyter.
Embick, David and Rolf Noyer. 2007. “Distributed Morphology and the Syntax-
Morphology Interface.” In The Oxford Handbook of Linguistic Interfaces, edited by
Gillian Ramchand and Charles Reiss, pp. 289–324. Oxford: Oxford University
Press.
Fábregas, Antonio. 2007. “An Exhaustive Lexicalisation Account of Directional
Complements.” In Tromsø Working Papers on Language & Linguistics: Nordlyd
34.2, special issue on Space, Motion, and Result, edited by Monika Bašić, Marina
Pantcheva, Minjeong Son, and Peter Svenonius, pp. 165–199. Tromsø: CASTL,
[ 52 ] Background
Fábregas, Antonio. 2009. “An Argument for Phrasal Spellout: Indefinites and
Interrogatives in Spanish.” Nordlyd 36: pp. 129–168. [online] Available at http://
septentrio.uit.no/index.php/nordlyd/index
Geeraerts, Dirk. 2010. Theories of Lexical Semantics. New York: Oxford University Press.
Giusti, Giuliana. 1997. “The Categorical Status of Determiners.” In The New Comparative
Syntax, edited by Liliane Haegeman, pp. 95–123. New York: Longman.
Greenberg, Joseph. 1963. “Some Universals of Grammar With Particular Reference to
the Order of Meaningful Elements.” In Universals of Language, edited by Joseph
Greenberg, pp. 73–113. Cambridge, MA: MIT Press.
Grimshaw, Jane. 1991. “Extended Projection.” Ms., Brandeis University.
Haegeman, Liliane. 1992. Theory and Description in Generative Grammar: A Case Study in
West Flemish. Cambridge: Cambridge University Press.
Haegeman, Liliane. 1995. The Syntax of Negation. Cambridge: Cambridge University
Press.
Haegeman, Liliane. 2003. “Conditional Clauses: External and Internal Syntax.” Mind
and Language 18: pp. 317–339.
Haegeman, Liliane. 2006a. “Argument Fronting in English, Romance CLLD and
Left Periphery.” In Crosslinguistic Research in Syntax and Semantics: Negation,
Tense and Clausal Architecture, edited by Raffaella Zanuttini, Hector Campos,
Elena Herburger, and Paul Portner, pp. 27–52. Washington, DC: Georgetown
University Press.
Haegeman, Liliane 2006b. “Clitic Climbing and the Dual Status of sembrare.” Linguistic
Inquiry 37: pp. 484–501.
Haegeman, Liliane 2006c. “Conditionals, Factives and the Left Periphery.” Lingua 116:
pp. 1651–1669.
Haegeman, Liliane 2012. Adverbial Clauses, Main Clause Phenomena and the Composition
of the Left Periphery: The Cartography of Syntactic Structures, Vol. 8. Oxford: Oxford
University Press.
Haegeman, Liliane. 2014. “West Flemish Verb-Based Discourse Markers and the
Articulation of the Speech Act Layer.” Studia Linguistica 68 (1): pp. 116–139.
Haegeman, Liliane and Raffaella Zanuttini. 1991. “Negative Heads and the Neg
Criterion.” The Linguistic Review 8: pp. 233–251.
Hale, Kenneth and Samuel J. Keyser. 1993. “On Argument Structure and the Lexical
Expression of Syntactic Relations.” In The View From Building 20: A Festschrift for
Sylvain Bromberger, edited by Kenneth Hale and Samuel J. Keyser, pp. 53–108.
Cambridge, MA: MIT Press.
Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the Pieces of
Inflection.” In The View From Building 20: A Festschrift for Sylvain Bromberger,
edited by Kenneth Hale and Samuel J. Keyser, pp. 111–176. Cambridge, MA:
MIT Press.
Harley, Heidi. 2014. “On the Identity of Roots.” Theoretical Linguistics 40: pp. 225–276.
Harris, Zellig. 1951. Methods in Structural Linguistics. Chicago: University of Chicago
Press.
Jackendoff, Ray. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution.
Oxford: Oxford University Press.
Johannessen, Janne-Bondi, Joel Priestley, Kristin Hagen, Tor Anders Åfarli, and
Øystein A. Vangsnes. 2009. “The Nordic Dialect Corpus—an Advanced Research
Tool.” In Proceedings of the 17th Nordic Conference of Computational Linguistics
NODALIDA 2009, edited by Kristiina Jokinen and Eckhard Bick, pp. 73–80.
NEALT Proceedings Series Volume 4. Red Hook, NY: Curran Associates.
54
Kayne, Richard S. 1975. French Syntax: the Transformational Cycle, p. 473. Cambridge,

MA: MIT Press.
Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht, The
Netherlands: Foris.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
Kayne, Richard S. 1998. “Overt vs. Covert Movment.” Syntax 1 (2): pp. 128–191.
Kayne, Richard S. 2005. Movement and Silence. Oxford: Oxford University Press.
Kayne, Richard S. 2016. “More Languages Than We Might Have Thought. Fewer
Languages Than There Might Have Been”. Ms., New York University.
Kiparsky, Paul. 1973. ‘ “Elsewhere’ in Phonology.” In A Festschrift for Morris Halle, ed-
ited by Paul Kiparsky and Stephen R. Anderson, pp. 93–106. New York: Holt,
Rinehart and Winston.
Koopman, Hilda. 2000. “Prepositions, Postpositions, Circumpositions, and Particles:
The Structure of Dutch PPs.” In The Syntax of Specifiers and Heads: Collected Essays
of Hilda Koopman, edited by Hilda Koopman, pp. 204–260. London: Routledge.
Koopman, Hilda and Anna Szabolcsi. 2000. Verbal Complexes. Cambridge, MA:
MIT Press.
Laenzlinger Chistopher. 1998. Comparative Studies in Word Order Variations: Pronouns,
Adverbs and German Clause Structure. Amsterdam: John Benjamins.
Laenzlinger, Christopher. 2005. “French Adjective Ordering: Perspectives on DP-
internal Movement Types.” Lingua 115: pp. 645–689.
Lakoff, George. 1971. “On Generative Semantics.” In Semantics: An Interdisciplinary
Reader in Philosophy, Linguistics and Psychology, edited by Danny D. Steinberg
and Leon A. Jakobovits, pp. 232–296. Cambridge: Cambridge University Press.
Lander, Eric. 2015a. “Intraparadigmatic Cyclic and Roll-Up Derivations in the Old
Norse Reinforced Demonstrative.” The Linguistic Review 32 (4): pp. 777–817.
Lander, Eric. 2015b. The Nanosyntax of the Northwest Germanic Reinforced Demonstrative.
Doctoral dissertation, Ghent University.
Larson, Richard. 1988. “On the Double Object Construction.” Linguistic Inquiry 19: pp.
239–266.
Leu, Tom. 2015. The Architecture of Determiners. Oxford and New York: Oxford
University Press.
Lindstad, Arne-Martinus, Anders Nøklestad, Janne Bondi Johannessen, and Øystein
A. Vangsnes. 2009. “The Nordic Dialect Database: Mapping Microsyntactic
Variation in the Scandinavian Languages.” In Proceedings of the 17th Nordic
Conference of Computational Linguistics NODALIDA 2009, edited by K. Jokinen
and E. Bick. NEALT Proceedings Series Volume 4.
Manzini, Maria Rita and Leonardo Maria Savoia. 2003. “The Nature of Comple
mentizers.” Rivista di Grammatica Generativa 28: pp. 87–110
Manzini, Maria Rita and Leonardo Maria Savoia. 2007. A Unification of Morphology
and Syntax: Investigations into Romance and Albanian Dialects. London and
New York: Routledge.
Manzini, Maria Rita and Leonardo Maria. 2011. Grammatical Categories. Cambridge:
Cambridge University Press.
Marantz, Alec. 1997. “No Escape from Syntax: Don’t Try Morphological Analysis
in the Privacy of Your Own Lexicon.” UPenn Working Papers in Linguistics 4:
pp. 201–225.
[ 54 ] Background
McGinnis, Martha. 2002. “On the Systematic Aspect of Idioms.” Linguistic Inquiry 33
(4): pp. 665–672.
Muriungi, Peter. 2008. Phrasal Movement Inside Bantu Verbs: Deriving Affix Scope and
Order in Kîîtharaka. Doctoral dissertation, University of Tromsø.
Noonan, Máire. 2010. “Á to zu.” In Mapping Special PPs, edited by Guglielmo Cinque
and Luigi Rizzi, pp. 161–195. Oxford: Oxford University Press.
Poletto, Cecillia. 2000. The Higher Functional Field: Evidence from Northern Italian
Dialects. New York: Oxford University Press.
Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar, and the Structure of
IP.” Linguistic Inquiry 20 (3): pp. 365–424.
Puskás, Genoveva. 2000. Word Order in Hungarian: The Syntax of A′-positions.
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon: A First- Phase Syntax.
Cambridge: Cambridge University Press.
Ritter, Elizabeth. 1991. “Two Functional Categories in Noun Phrases: Evidence from
Modern Hebrew.” In Perspectives on Phrase Structure: Heads and Licensing, edited
by Susan D. Rothstein, pp. 37–62. San Diego, CA: Academic.
Rizzi, Luigi. 1994. “Early Null Subjects and Root Null Subjects.” In Language Acquisition
Studies in Generative Grammar, edited by Teun Hoekstra and Bonnie Schwartz,
pp. 151–176. Amsterdam and New York: John Benjamins.
edited by Liliane Haegeman, pp. 281– 337. Dordrecht, The Netherlands:
Kluwer.
Rizzi, Luigi. 2001. “On the Position Int(errogative) in the Left Periphery of the Clause.”
In Current Studies in Italian Syntax: Essays Offered to Lorenzo Renzi, edited by
Guglielmo Cinque and Giampaolo Salvi, pp. 287–296. Amsterdam: Elsevier,
North-Holland.
Rizzi, Luigi. 2004a. “Locality and Left Periphery.” In Structures and Beyond:
The Cartography of Syntactic Structures, Vol. 3, edited by Adriana Belletti,
pp. 223–251. New York: Oxford University Press.
Rizzi, Luigi. (ed.). 2004b. The Structure of CP and IP: The Cartography of Syntactic
Semantics.” In Mind, Values and Metaphysics—Philosophical Papers Dedicated
to Kevin Mulligan, edited by Anne Reboul, pp. 517– 533. Dordrecht, The
Netherlands: Springer.
Rocquet, Amélie. 2013. Splitting Objects: A Nanosyntactic Account of Direct Object
Marking. Doctoral dissertation, Ghent University.
Scott, Gary-John. 2002. “Stacked Adjectival Modification and the Structure of Nominal
Phrases.” In Functional Structure in DP and IP: The Cartography of Syntactic
Structures, Vol. 1, edited by Guglielmo Cinque, pp. 91–120. New York: Oxford
University Press.
Shlonsky, Ur. 2010. “The Cartographic Enterprise in Syntax.” Language and Linguistics
Compass 4 (6): pp. 417–429.
Nordlyd 36 (1): pp. 1–6. [online] Available at http://septentrio.uit.no/index.php/
nordlyd/index.
56
Starke, Michal. 2011a. “Towards an Elegant Solution to Language Variation: Variation

Reduces to the Size of Lexically Stored Trees.” [online] Available at LingBuzz/
001183.
Starke, Michal 2011b. Class notes, course on nanosyntax given at Ghent University.
November 7–10.
Starke, Michal. 2013. Auxiliaries and Structural Gaps: Current Issues in Nanosyntax.
Lecture series presented at CRISSP, Hogeschool-Universiteit Brussel. March18,
20, 22.
Svenonius, Peter. 2006. “The Emergence of Axial Parts.” Nordlyd 33 (1): pp. 49–77.
[online] Available at http://septentrio.uit.no/index.php/nordlyd/index
Svenonius, Peter. 2008. “The Position of Adjectives and Other Phrasal Modifiers in
the Decomposition of DP.” In Adjectives and Adverbs: Syntax, Semantics, and
Discourse, edited by Louise McNally and Christopher Kennedy. New York: Oxford
University Press.
Svenonius, Peter. 2010. “Spatial Prepositions in English.” In Mapping Spatial PPs: The
Cartography of Syntactic Structures, Vol. 6, edited by Guglielmo Cinque and Luigi
Rizzi, pp. 127–160. New York: Oxford University Press.
Szabolcsi, Anna. 1981. “The Possessive Construction in Hungarian: A Configurational
Category in a Non- Configurational Language.” Acta Linguistica Academiae
Scientiarum Hungaricae 31 (1-4): pp. 261–289.
Szabolcsi, Anna. 1984. “The Possessor That Ran Away from Home.” The Linguistic
Review 3 (1): pp. 89–102.
Szabolcsi, Anna. 1987. “Functional Categories in the Noun Phrase.” In Approaches
to Hungarian 2: Theories and Analyses, edited by István Kenesei, pp. 167–189.
Szeged, Hungary: JATE.
Szabolcsi, Anna. 1994. “The Noun Phrase.” In The Syntactic Structure of Hungarian, ed-
ited by Ferenc Kiefer and Katalin E. Kiss, pp. 179–275. New York: Academic.
Szabolcsi, Anna. (ed.) 1997. Ways of Scope Taking. Dordrecht, The Netherlands: Kluwer.
Taraldsen, Knut Tarald. 2009. “The Nanosyntax of Nguni Noun Class Prefixes and
Concords.” Ms. CASTL. [online] Available at LingBuzz/000876.
Travis, Lisa. 1984. Parameters and Effects of Word Order Variation. Doctoral dissertation,
MIT.
Vangsnes, Øystein A. 2014. “Indexicals by Nanosyntax: Wh and D Items Apart.” SLE
Workshop on Nanosyntax, Poznań, Poland. September 12.
Zanuttini, Raffaella. 1991. Syntactic Properties of Sentence Negation: A Comparative
Study of Romance Languages. Doctoral dissertation, University of Pennsylvania.
Zubizaretta, Maria Luisa. 1998. Prosody, Focus, and Word Order. Cambridge, MA:
MIT Press.
[ 56 ] Background
CHAPTER 2
Notes on Insertion in Distributed

Morphology and Nanosyntax
PAVEL C AHA
2.1 THREE DIFFERENCES BETWEEN NANOSYNTAX

AND DISTRIBUTED MORPHOLOGY
The theories of nanosyntax (Caha 2009; Starke 2009; Baunaz and Lander in
Chapter 1) and Distributed Morphology (Halle and Marantz 1993; Harley and
Noyer 1999; Embick and Noyer 2007; Bobaljik 2015) each provide a partic-
ular theory of how to get from the basic building blocks of language to com-
plex sentences.1 In practice, the research mostly focuses on linguistic units of
smaller size (the traditional words or small phrases), but both theories share
the idea that a general theory of syntax reaches deep into the structure of
such units.
It is clear that any such theory is going to have multiple components,
and the two theories diverge on a number of important questions. For in-
stance, an important difference (dif 1) concerns the nature of the basic building
blocks. Distributed Morphology (DM) projects its syntactic structures from
complex objects that correspond to prepackaged feature structures (bundles)
taken directly out of the presyntactic lexicon. The main architectural claim of
nanosyntax is opposed to this; the idea is that the only component of grammar
capable of constructing complex feature structures is syntax, which thus has
1. The work on this paper has been supported by the Czech Grant Agency, GAČR,
project number GA 17-10144S (Exploring Linear Contiguity), awarded to Pavel Caha.
I am also grateful to two anonymous reviewers for their comments on previous drafts
of this paper.
58
to start from individual (atomic) features. This leads to a syntax with large
trees, increased number of movement steps, and a quite different outlook
than the tree-structure representations used within DM. Nanosyntax also
tries to dispense with head movement, which leads to an increase in phrasal
movement and remnant movement derivations. As highlighted in the intro-
duction to this volume, nanosyntax shares these features with cartography
and other approaches.
Another difference (dif 2) is the use of phrasal spellout, which in nanosyntax
represents one of the core analytical tools, while at the same time being rarely
used and often argued against in DM, as in Embick and Marantz (2008) or
most recently in Embick (2014).2 The core idea behind phrasal spellout is that
pronunciation is like other operations in natural language (movement, el-
lipsis, etc.) in that it targets potentially nontrivial constituents.
This aspect of nanosyntactic theory is closely related to the fact that
morphemes usually express multiple features, which for nanosyntax auto-
matically means that they express multiple heads. Such morphemes are here
called portmanteau morphemes. Portmanteau morphemes must also be dealt
with in DM (and other theories), and indeed DM has at least two tools to
accommodate portmanteaus, including (i) the postsyntactic operation of
Fusion and (ii) zero morphemes accompanied by conditioned allomorphy or
readjustment rules.
This chapter probes deeper into the particular issue of portmanteaus, be-
cause it seems that in this particular domain, a meaningful comparison can
be made. I argue that neither Fusion (Section 2.3) nor silent morphemes
(Section 2.4) represent adequate tools to deal with portmanteaus. I further
demonstrate that the insertion principle used in DM (the so-called Subset
Principle) is incapable of governing nonterminal lexical insertion, whereas
the nanosyntactic conception (the so-called Superset Principle) handles the
relevant data with ease. The move to nonterminal spellout is also theoretically
attractive; once adopted, it replaces the morphological operations proposed to
deal with portmanteaus (Caha 2009, ch.2).
Once such morphological operations are eliminated, what emerges is
the possibility of a direct mapping between syntax and pronunciation,
2. For clarity, let me mention that phrasal spellout is a type of nonterminal spellout
(phrase is a type of nonterminal node). I mention this here because there seems to be
some confusion about this in the literature. For instance, Haugen and Siddiqi (2013)
write that phrasal spellout “inserts entire phrases into non-terminal nodes rather than
just single Vocabulary Items.” This is incorrect: Phrasal spellout inserts Vocabulary
Items into phrasal nodes.
Please note that what nanosyntax calls phrasal spellout is often best recast in DM
as the spellout of a nonterminal inside a complex head. This is because NS often uses
phrasal movement derivations where DM uses head movement, and so for reasons of
cross-theoretical comparison, it is best to understand phrasal spellout to actually mean
nonterminal spellout.
[ 58 ] Background
mediated only by lexical access. This theoretical goal ultimately represents
a third difference (dif 3) between nanosyntax and DM: Is there (DM), or
is there not (nanosyntax), a separate morphological module of grammar, i.e,
a special M-structure with rules and operations specific to that module?
This issue—independent of dif 2—revolves around questions of whether
the remaining postsyntactic operations used in DM are needed to derive
the surface forms of language, or whether these operations can also be
dispensed with. Once again, nanosyntax is not alone here in its attempts
to avoid using such operations; see, for example, Koopman (2005) or
Kayne (2010).
The specific operations that are covered under dif 3 (on this way of cutting
the pie) are postsyntactic operations that change constituency and/or linear order.
Specifically, Marantz (1989) as well as Embick and Noyer (2001) propose that
in a number of cases, structures of the type [X [Y Z]] can be transformed
onto [[Y X] Z] by Merger, which (according to them) takes place after syntax.
According to my perception of the data, the difference with nanosyntax here
is not so much about what the constituency is: I think that in most cases it
is [[Y X] Z] (i.e., the one that DM derives by Merger). The question is rather
whether the postsyntactic Merger is the only way to derive such structures.
The strategy that has been adopted in some of the nanosyntax literature
is to say that Y and X are in fact merged independently together, and only
later merged with Z, yielding [[Y X] Z] (see Baunaz and Lander in Chapter 1
for the operation Construct, and also Starke in Chapter 9 and Taraldsen in
Chapter 3).3
Obviously, the logically independent issues covered under dif 2 (is there
or is there not phrasal spellout) and dif 3 (what is the constituency of
strings and how do we derive it) interact. If lexicalization targets (poten-
tially phrasal) constituents, then it matters what type of constituent struc-
ture feeds into lexicalization. To give a concrete example: It matters for the
spellout of X and Y whether the constituency is [X [Y Z]] (X and Y cannot
spell out together) or whether it is [[X Y] Z] (X and Y can be spelled out
together).
Most of the discussion of nanosyntax from DM positions has focused
on the issues surrounding constituency (dif 3). So despite the fact that my
main goal here is to look at the topic of insertion (dif 2), it feels wrong to
ignore the concerns that researchers working in DM have about the constit-
uency required by phrasal spellout. So I devote some remarks to this topic in
Section 2.2, before I come back to insertion in Sections 2.3 to 2.5. Section 2.6
concludes.
3. In the context of DM, Bobaljik (2012, 57) also suggests the option of simply base-
generating such structures.
In s e r t i o n i n Di s t r i b u t e d M or p h ol o g y a nd N a n o s y n ta x [ 59 ]
60
2.2 DEVELOPING A THEORY OF SPELLOUT IN A MODEL

OF SYNTAX THAT KEEPS CHANGING
As highlighted under dif 1, nanosyntax and DM have diverging opinions con-

cerning the general outlook on syntactic structure. On the one hand, these
differences are rather limited in scope, pertaining mostly to the number of
heads and the amount and type of movement needed to generate surface
strings. On the other hand, these differences have quite a big impact on the
debate that takes place in the literature. To a certain extent, it may seem sur-
prising that diverging opinions about movement would lead to disagreement
between frameworks that are in practice concerned with traditional mor-
phology. But it is in fact old news that one’s conception of syntax tends to be
rather important for one’s choice of morphological framework.
To illustrate this, let me go back to the 1990s when DM itself was taking
off as a framework. In one of the papers of that time, Pullum and Zwicky
(1992) reflected on DM’s conception of morphology from the positions of
an “A-morphous” approach (based on the idea that words correspond to un-
structured feature matrices and represent the atoms of structure-building;
see e.g. Anderson 1992). The authors noted that the structured morpholog-
ical representations used in DM fare quite well in Pollock’s (1989) system,
where complex words arise by head movement, rather than being simply
pulled out of the lexicon. To the authors, such a system involved “baroque-
style derivations,” where “it is not clear why affixes or stems could not be
moved apart by syntactic rules to yield (e.g.) Affix Topicalization or Heavy
Stem Shift constructions.” They went on to claim that Pollock’s system, “even
if it worked . . . , it would . . . generate strings of the wrong sort to be input to
a morphological module of a theoretically and empirically optimal sort.” In
effect, Pullum and Zwicky took their paper to “cast serious doubt on whether
its basic operation of head-to-head movement was doing any real work.” Some
25 years later, we are where we are. Head movement has become a common
stance in syntactic analyses and DM a point of reference with which to com-
pare new alternatives.
The reason for mentioning Pullum and Zwicky’s criticism here is not to
agree with it; rather, the intention is to show that DM’s reliance on a new syn-
tactic tool (head movement) led to a new style of morphological analyses, and
that researchers who had their doubts about the correctness of the tool were
reluctant to adopt a morphological theory that seemed to require its large-
scale application.
But syntax did of course not stop in the 1990s, and new tools kept arriving.
Consider, for instance, the development from Cinque (1996) to Cinque
(2005). Both papers try to derive the order of nouns, adjectives, numerals,
and demonstratives. In 1996, Cinque proposed that in some such orders, the
noun moved from its base position at the bottom of the tree to the left of
[ 60 ] Background
the modifiers by head movement. In 2005, however, Cinque included many
more languages in his sample and focused primarily on the issue of how
to exclude those orders that are not attested. In this setting, he noted that
head movement is “unable to exclude the unattested orders” (Cinque, 2005,
317) and proposed that the noun actually crossed its modifiers by phrasal
movement. This allowed the construction of a system based exclusively on
phrasal movement that can generate all and only the attested orders. This—
among other things—meant that the complement of the noun (which often
remains last in the string even when the noun moves) has to undergo obliga-
tory extraposition (Cinque 2005, 327). Extraposition here refers to movement
of the noun’s complement to a high position in the extended NP such that it
c-commands and precedes all the noun’s modifiers. After that, in languages in
which complements follow the noun, the whole extended NP undergoes rem-
nant movement, so that the complement ends up to the right of the whole
noun phrase (again).
With such changes in syntax taking place (see the literature emanating
from Kayne 1994, 1998), it became possible to contemplate new models of
morphology that make use of such derivations. To see where they fit in within
nanosyntax, consider the form puell-ās (discussed also in Baunaz and Lander
in Chapter 1), which is an accusative plural (-ās) of the Latin noun ‘girl’ (puell-).
As far as syntax is concerned, there are reasons to think that number and case
occupy distinct slots in the crosslinguistically fixed sequence of projections
(we see some evidence for this in Section 2.4). But in the morphological string,
it is impossible to separate the accusative meaning from the plural meaning;
-ās is a portmanteau for both. To remove tensions such as these (an indivisible
form corresponding to two syntactic positions), nanosyntax proposes that
these two projections are spelled out together as a phrase after the noun has
moved to the left by phrasal movement, as seen in Figure 2.1.
Suppose now that the noun had a complement. What would happen to it?
If the complement did not move, it would be carried along inside the NP and
end up in between the root and the affix (which does not happen). So the
assumption needed to make this work is that the complement undergoes oblig-
atory extraposition to an even higher position (minimally above the landing
site of the noun), followed by remnant movement of the noun with its affix.
There is nothing new here compared with Cinque’s (2005) work: Complements
undergo obligatory extraposition.
NP [ K [ Num NP ] ]
puell as
Figure 2.1 Spelling out -ās as a phrase after noun movement
62
For many researchers, such derivations are simply “too baroque” to be of

any interest. For instance, Embick (2014) correctly points out that “the main
predictions of [phrasal spellout] derive from constituency.” But keeping the
1990s ideas about syntax as a reference point, he points out that, for in-
stance, “synthetic forms should never be found with adjectives that take
complements,” because that would lead to the problem of the intervening
complement in Figure 2.1.
In general terms, Embick concludes that the representations required by
phrasal spellout are “in the crucial cases . . . incorrect.” He notes, though, that
“in all of these, it is possible to incorporate additional assumptions [obligatory
extraposition of complements] to neutralize incorrect predictions,” but for
these “there is little evidence” (Embick 2014, 23). However, keeping Cinque’s
independent conclusions about obligatory extraposition in mind, one cannot
help feeling that we are taking part in a similar debate that took place some
25 years ago: We simply believe in different syntaxes (one standard by now,
the other still too baroque), and so we are talking cross-purposes.
Embick’s paper is one of the few critiques of phrasal spellout that at least
acknowledges these issues. Merchant (2015, 299) fails to mention even
the possibility of extraposition when he writes that “whenever [phrasal
spellout involving the root] must occur, the root should allow for no internal
arguments, a prediction that is clearly false.”
Merchant’s unwillingness to contemplate extraposition in these cases
contrasts with his own work on ellipsis. Consider, for instance, gapping
examples like John likes sandals and Mary likes stiletto heels, concerning which
van Craenenbroek and Merchant (2013, 743) say that “it has become fairly
standard to analyze gapping as involving movement of the remnants [sti-
letto heels] to the left [across likes] followed by deletion . . . of the rest of the
clause.” Clearly, the authors are relying on extraposition to create constituents
(the verb and its inflection) that are later targeted by ellipsis. The reason why
Merchant denies this option for phrasal spellout (in effect subjecting phrasal
spellout to a different theoretical standard than ellipsis) is a mystery to me.
Another construction for which Embick (2014) as well as Haugen and
Siddiqi (2013) think that phrasal spellout does not work are cases in which a
preposition spells out as one piece with the determiner (they mention French
and Spanish, respectively). They start from the idea that because of the scope
relations, the preposition and the determiner cannot form a constituent to
the exclusion of the noun. But because they spell out together, they conclude
that phrasal spellout cannot deal with this fact.
What is slightly puzzling is that Embick (2014) ultimately does adopt the
structure [[P D] N] for his own analysis, that is, a structure that makes his point
about wrong predictions of phrasal spellout irrelevant if evaluated against
that structure. The reason why he denies the phrasal spellout model access to
the correct structure is because he says that, as far as syntax is concerned, the
[ 62 ] Background
structure must be [P [D N]], and the structure that correctly describes the P–D
interaction is derived by Merger. The idea is that phrasal spellout cannot make
use of the postsyntactic Merger that, according to Embick, is the only way to
derive the structure.
There are two observations to be made. The first one is that phrasal
spellout is in principle independent of whether one does or does not have
Merger, and so these things should not be automatically lumped together.
Even more so, they introduce a double standard for comparison in which a
terminal spellout model is evaluated against the correct structure, and the
phrasal spellout model is claimed to make wrong predictions because it is
evaluated against a different (and incorrect) structure. In other words, even
if one adopts a theory with Merger, it is still better (as I argue in Sections
2.3–2.5) to spell out the nonterminal derived by Merger than let insertion
target terminals.
The second observation is related to the issue of whether syntax can or
cannot build structures of the sort [[P D] N]. Here it is worth noting that apart
from the French–Spanish facts, there is evidence for such a type of structure
independent of phrasal spellout. For instance, Baker and Kramer (2014) dis-
cuss the fact that in Amharic, a prepositional marker that scopes over the
whole extended NP is located inside (i.e. in the middle of) a constituent that is
located on the left branch of that NP. This can only be captured if the preposi-
tion is included inside the left-branch modifier ([[P Modifier] N]).
Now given the converging evidence for the possibility of structures like
[[P D] N], the question to ask is this: Can we design our theory of syntax
in a way that can accommodate the evidence without the involvement of a
postsyntactic component?
Proposals addressing such issues are already out there even independently
of nanosyntax, but see also Baunaz and Lander’s (Chapter 1, Section 1.3.3.5)
discussion of the operation Construct and also Starke in Chapter 9 and
Taraldsen in Chapter 3. For instance, Leu (2008) investigates structures anal-
ogous to the unexpected [[P D] N], but for a slightly different case. He looks
at examples for which the evidence suggests that definiteness morphemes
are present in the projection of the adjective, yielding surface structures of
the sort [[D A] N], against the semantically expected [D [A N]]. Leu then
proposes a theory that base-generates the definiteness marker inside the pro-
jection of the adjective. The fact that definiteness scopes over the whole NP
is attributed to a silent D sitting in the main projection line, so we actually
have the structure [[D A] [D [N]‌]], where the boldfaced D is silent because of a
generalized Doubly Filled Comp (Koopman 1996). So once again, it seems that
we independently need the structures that Embick claims to be problematic,
and we also have minimally the beginning of an account. The fact that this
account deviates from the structures used in the 1990s should not distract us
too much.
64
My ultimate point is that reflections on nanosyntax in the DM literature

resemble the welcome that DM got some 25 years ago. The reason for the re-
semblance is that people back then as much as now have different ideas about
what syntax looks like. The syntactic structures of nanosyntax are “too ba-
roque,” and “even if they worked, there is little evidence for them.” I have tried
to do some justice to that debate here; my main point is that the constituent
structures needed for phrasal spellout (and claimed to be problematic) are
needed also for reasons that have nothing to do with phrasal spellout, a point
that I think should not be controversial.
The only thing that is potentially controversial is whether we should
generate such structures in the syntax. I think that the null hypothesis is
that we should minimally try to do so. Ultimately, any empirical debate
about the structures needed for phrasal spellout cannot take the 1990s
syntax for granted. Rather, we have to focus on questions such as these: Is
there any evidence for or against the claim that the constituent struc-
ture needed for phrasal spellout is accessed also by other processes (like
gapping, etc.)?
Whatever the ultimate answers to such questions will be, in what
follows, I sidestep as many of these issues as possible. My goal is to discuss
the differences that exist in the insertion procedure independently of the
diverging opinions about what syntactic structures look like and how they are
constructed. The way I do this is that I assume here only structures independ-
ently adopted in DM and show how the theory can be improved if nonterminal
spellout is adopted.
2.3 INSERTION IN DISTRIBUTED MORPHOLOGY
This section illustrates some of the paradoxes that arise in classical DM, in
which spellout targets terminal nodes only. I also show why it is impossible
to extend the insertion procedure to nonterminals, presenting what I think
are insurmountable puzzles. The solution to the puzzles comes in Section 2.5,
when we turn our attention to the Superset Principle.
2.3.1 The Subset Principle
Vocabulary insertion in DM is governed by the so-called Subset Principle.

Its canonical formulation comes from Halle (1997, 428) and is given in (1).
The principle has two parts, which can be called the Subset Clause and the
Elsewhere Clause. The Subset Clause determines the conditions for the appli-
cability of a Vocabulary Item. The Elsewhere Clause says what happens when
several Vocabulary Items are applicable in a given environment.
[ 64 ] Background
(1) The Subset Principle
[The Subset Clause]: The phonological exponent of a Vocabulary Item is
inserted into a morpheme of the terminal string if the item matches all
or only a subset of the grammatical features specified in the terminal
morpheme. Insertion does not take place if the Vocabulary Item contains
features not present in the morpheme. [The Elsewhere Clause]: Where sev-
eral Vocabulary Items meet the conditions for insertion, the item matching
the greatest number of features in the terminal morpheme must apply.
The main empirical bite of the principle is that it can nicely model syncretism.
As an example, consider the following fragment of the singular declension in
Latin. In Table 2.1, I also indicate one possibility of attributing morphosyntactic
features to the relevant cells, borrowing some of the features from Halle (1997).
What we see here is that one of the exponents (-us) is tailormade for a partic-
ular case and gender.4 There is no intricacy involved in saying that this marker is
the pronunciation of the features [+masc, +sup]. It is much harder to say what
the other marker is the pronunciation of. It can appear both with masculines
and neuters, and it can appear both in the nominative and in the accusative.
The Subset Principle allows us to generate such paradigms easily. It is
enough to say that the marker -um simply corresponds to an underspecified
singular marker, as in (2a). The Subset Principle then predicts that it can
occur in any case and any gender, as long as the [+sg] feature (for which -um is
specified) is present in a given feature bundle or paradigm cell. So in principle,
this marker can occur in all the cells of the paradigm in Table 2.1.
(2) a. -um = [+sg]

b. -us = [+sg, +sup, +masc]
The number of cells where such an “underspecified” marker actually surfaces

depends on what other markers there are. So when we add the nominative
Table 2.1 A FRAGMENT OF THE L ATIN DECLENSION
Neuter Masculine
[+singular] [-masc] [+masc]
NOM -um -us

[+superior]
ACC -um -um

[-superior]
4. I am ignoring the potential decomposition into -u-m/-u-s, which does not change
the point.
66
(+sup) masculine (+masc) -us into the set of Vocabulary Items (2b), this will
lead to a clash between -um and -us in the nominative of the masculine par-
adigm (in all other cells -us does not qualify for insertion on the basis of the
Subset Clause). In this particular environment, -us wins, because it is more spe-
cific, and -um fills the rest of the cells. So the possibility to have underspecified
markers coming in after syntax is very attractive; syntacticians can do their
work on fully specified feature structures, over which syntactic generalizations
are stated. The Subset Principle takes care of the morphological detail.
2.3.2 The paradox
Notice now that the insertion principle in (1) explicitly states that phonolog-
ical exponents are “inserted into a morpheme of the terminal string.” A mor-
pheme in DM is a term for a bundle of features (a terminal), so in effect this
means that insertion applies only at terminal nodes.
Are there any reasons to doubt this conclusion? Suggestions to this effect
were rare in the early days of DM (I think that the first nonterminal spellout
analysis in DM is Radkevich (2010), with a manuscript version going back
to 2008). However, some issues did start appearing. An interesting insight
into the dilemmas that the theory faced at that point (and still faces today) is
provided by a careful study of negation in Korean by Chung (2007).
The starting point of Chung’s discussion is the fact that sentences in Korean
can be negated by attaching one of the two negative prefixes ani or mos to the
verb, as shown in (3). Chung (2007) shows that each of the negations is a
separate head in the clausal spine, and the verb combines with it by syntactic
movement. The negation sits lower than T, so that the structure of the string
mos/an(i)-ca is as shown in Figure 2.2.
(3) a. ca-n-ta

sleep-pres-decl
‘is sleeping’
b. mos/an(i) ca-n-ta
neg sleep-pres -decl
‘can’t sleep /isn’t sleeping’
[Chung 2007, (1), (2), (4)]
Neg
Neg V
ani/mos SLEEP
ca
Figure 2.2 Structure for mos/an(i)-ca
[ 66 ] Background
The second relevant fact is that the verb al-‘know’ (4a) does not combine
with any of these markers (4b), but shows a suppletive form molu-instead, as
seen in (4c).5
(4) a. al-n-ta

know-pres-decl
‘knows’
b. *mos/an(i) al-n-ta
neg know-pres-decl
‘cannot /does not know’
c. molu-n-ta
neg.know-pres-decl
‘cannot /does not know’
[Chung 2007, (45)]
What we see here in abstract terms is that where one meaning (‘not sleep’)
has two markers (Neg and V), another meaning (‘not know’) has a single
nondivisible marker (Neg + V). Chung takes care to show that molu is an actual
negative form; it does not correspond to a lexical verb like ‘ignore, be unaware
of.’ So it seems that molu actually expresses both Neg and V.
But given that insertion targets only terminal nodes, it cannot be the
case that molu is inserted into two terminals at the same time. So the
question is whether molu is inserted under Neg or under V. The choice
feels forced and leads to some obvious issues. If it was inserted under V,
we would expect it to combine with negation (which it does not). If it was
inserted under Neg, we would expect it to combine with a verb (which it
does not either).6
The intuition that the form molu in fact conveys the meaning of both
‘know’ and the negation seems hard to implement. Chung (2007) concludes
that within the confines of the DM system, there is only one possible solution.
What one has to say is that the structure in Figure 2.2 is targeted by a special
operation, Fusion, which turns the layered representation into a flat node.
The procedure is given in Figure 2.3, taken from Chung [2007, (82)]. The lex-
ical entry in (5) is then allowed to apply, because Fusion has collapsed both
features under a single terminal node.
5. The form is obviously similar to mo(s-a)l, and could have easily developed from the
regular form. But synchronically, it is unpredictable and must be stored (to judge from
Chung’s paper).
6. An anonymous reviewer suggests that molu may be a special allomorph for KNOW
under negation and that also negation has a special zero allomorph in the context of
KNOW. I get to this type of approach in Section 2.4.
68
(5) [+neg, KNOW] ⇔ molu
The solution in terms of Fusion does justice to the observation that molu
conveys both Neg and V, but it leads to a paradox (identified already in Chung
2007, fn.22). On the one hand, Fusion must precede lexical insertion, be-
cause lexicalization targets the structures that Fusion creates. On the other
hand, Fusion happens only when the lexicon contains a portmanteau for the
fused heads (for instance, Fusion cannot happen when the verb is ‘sleep,’ as
in Figure 2.2.). Thus, an operation that precedes lexicalization is conditioned
by lexicalization.
2.3.3 Nonterminal spellout as a solution to the paradox
Chung then notes that a natural solution for the paradox would be to say that
the lexical item molu spells out the whole structure in Figure 2.3. The lexical
entry would then look like Figure 2.4.
Once Figure 2.4 is adopted as a possible format for a lexical item, the paradox
disappears. There is no need for Fusion to apply before insertion, because the
nonterminal in Figure 2.3 can be directly mapped onto its pronunciation.
So the natural thing to do at this point would be to simply drop the restric-
tion that Vocabulary Items may be inserted only at terminals. If one could do
that, one would get a more general theory of insertion that applies to all syn-
tactic nodes (without a restriction to terminals, which in any event amounts
to pure stipulation). Moreover, the ugly paradox would disappear; we would
in fact derive the fact that Fusion is driven by the existence of a particular
Vocabulary Item, because insertion at a nonterminal would be available only
for such items. Or wouldn’t it?
Neg
Neg V [+NEG, KNOW]
+NEG KNOW
Figure 2.3 Fusion of Neg and V
Neg
Neg V molu
+NEG KNOW
Figure 2.4 Lexical entry for molu
[ 68 ] Background
2.3.4 The communication breakdown
Let us check. Consider for example the Vocabulary Item for SLEEP, given
in (6).
(6) ca-⇔ [V SLEEP]
Looking at the item, let us ask the question whether this lexical item can spell out
the structure in Figure 2.2. After we have dropped the restriction to terminals, the
Subset Principle now says that the entry can spell out a node “if the item matches
all or only a subset of the grammatical features” specified in that node. And be-
cause (6) corresponds to a subset of the features contained in Figure 2.2, it seems
that ca- should be able to spell out that node, in effect meaning both ‘to sleep’
and ‘not to sleep,’ a consequence that Chung notes in his fn. 22. Generalizing this
observation, Bye and Svenonius (2010, fn.12) point out that this solution would
in fact lead to the expectation that every sentence is pronounced by a single mor-
pheme. I call this consequence “the communication breakdown.”
Once Chung realized the grim prospects of simply extending the Subset
Principle to nonterminal spellout, he made no attempts at developing the
idea further. Compared with the problem of communication breakdown, the
Fusion rule in Figure 2.3 (which he ultimately adopts) looks quite innocent.
But is there really no way to avoid this problem?
2.3.5 Trying to f ix things by adding principles: The VIP
One possibility to avoid communication breakdown has been investigated in

the work by Radkevich (2010) and Bobaljik (2012). They suggest that adopting
the “Vocabulary Insertion Principle” (henceforth, VIP) in addition to the
Subset Principle will eliminate the problem. The VIP is given in (7):
(7) The Vocabulary Insertion Principle (Radkevich 2010, 8)

The phonological exponent of a vocabulary item is inserted at the min-
imal node dominating all the features for which the exponent is specified.
With the VIP in place, some of the most pressing problems of communica-
tion breakdown indeed disappear. For instance, (6) can no longer spell out the
whole structure in Figure 2.2, because the minimal node containing all the
features of (6) is the V node. Consequently, Neg must be spelled out separately
(by one of the negative markers) and we get the right result.
However, (7) ultimately fails in a slightly more complex set of cases. To see
that, consider an additional fact noted by Chung (2007), namely that supple-
tion for negation interacts with causativization. The first piece of the relevant
70
data is in (8), which is a causative form of the verb ‘know,’ meaning ‘let know,
inform.’ The causative component corresponds to the affix -li. The structure
of the causative as proposed in Chung is shown in Figure 2.5. The VIP-based
insertion procedure correctly inserts al-‘know’ at the V node, and -li spells out
the causative.
(8) al-li-
know-caus
‘let know, inform’
The question to ask is what happens when we negate the causative form,
yielding the meaning of ‘not to inform,’ in which the negation scopes over the
causative, as shown in Figure 2.6, reproduced from Chung [2007, (86)].
The prediction of the Subset Principle together with the VIP is that the
lexical item molu in Figure 2.4 can be inserted at the top node of the tree in
Figure 2.6. That is because the features of the lexical entry (Neg and V) corre-
spond to a Subset of the features in Figure 2.6, and the top node in Figure 2.6
is the minimal node containing those features.
But that is the wrong result. As Chung (2007) points out, the structure in
Figure 2.6 is actually pronounced as (9). This means that Figure 2.6 cannot be
spelled out by molu even though the Subset Principle would seem to allow this
even when combined with the VIP.
(9) ani / mos al-li-ess-ta (Korean)

neg know-caus-past-decl
‘did not /could not inform’ [Chung 2007, (58)]
Cause
V Cause
KNOW li
al
Figure 2.5 Structure of the causative
Neg
Neg Cause
V Cause
KNOW
Figure 2.6 Negated causative
[ 70 ] Background
What is the general lesson we learn? The Subset Principle—which has
been widely adopted in DM as a principle governing insertion—fails badly
when used as a principle that regulates insertion at nonterminal nodes.
The literature that (for good reasons) tries to extend DM by nonterminal
spellout, adopted a general strategy of preventing problems by adding
principles. Their goal is to somehow block the unbounded extension of a
particular Vocabulary Item up to a point at which this single item spells
out the whole sentence. The VIP is one possible way to do this, but it fails
in a particular set of cases—namely when a Vocabulary Item contains a
feature at the bottom (V) and at the top of a tree (Neg), with one or more
features intervening in between (Cause). I refer to this as the “problem
with interveners.”
Other principles of a similar sort have been proposed, but they all ulti-
mately fail. I discuss one of these in Section 2.3.6.
2.3.6 Global comparisons
One way to avoid the problem with interveners is to say that there is actu-
ally nothing wrong with the fact that molu matches the structure in Figure
2.6. There is just a better way of expressing the content of Figure 2.6, which
expresses more features than molu alone.
The idea is informally depicted in Figures 2.7 and 2.8. What we see in Figure
2.7 is that when the item molu spells out the whole constituent, it leaves the
causative feature unexpressed (expressed features are in bold). Compare with
this the (correct) spellout in Figure 2.8, which—even though more complex
Neg ⇒ molu
Neg Cause
V Cause
KNOW
Figure 2.7 Spelling out Neg and V
Neg
Neg Cause
ani V Cause
KNOW li
al
Figure 2.8 Spelling out Neg, V, and Cause
72
in terms of the number of markers—expresses more features. (10) makes ex-

plicit the logic that would lead to the result that Figure 2.8 is chosen over
Figure 2.7. The inspiration for the particular shape of the statement is taken
from Haugen and Siddiqi (2016, 369–370).
(10) Spellout preferably targets multiple terminals, but only if doing so

realizes at least as many features as the spellout of individual terminals.
A theory along these lines has been also proposed in Julien (2015), and
it must have been on Embick and Marantz’s (2008, 4) mind when they
wrote that “when the approaches to blocking involving comparison be-
tween otherwise grammatical expressions are made explicit, they all nec-
essarily involve global competitions,” a path that they were not willing to
open for the computational complexity it brings along.
We will see in Section 2.5.1 that this conclusion is not completely
correct and that the Superset Principle provides a solution that avoids
global competitions. Mentioning the quote here serves as an illustration
that by weighing the two scenarios in Figures 2.7 and 2.8 against one an-
other, we are not pursuing a useless intellectual exercise. It seems that
when Embick and Marantz tried to imagine a model of spellout in which
molu blocks mos-al [recall (4)], they considered the “global comparison”
scenario in (10) an unavoidable consequence. This finding is important
when we want to evaluate the contribution of nanosyntax to the devel-
opment of a usable phrasal spellout model. Clearly, Embick and Marantz
did not see it coming as late as 2008 (which is about the same time when
the first nanosyntax papers started appearing in press; see Caha 2007 and
Fábregas 2007, who both give credit to Starke’s unpublished work going
back to the early 2000s).
Now, even admitting that (10) is computationally complex, is it going to
work? There is again a set of cases in which it won’t. In classical instances of
phrasal spellout, the expected and regular combination of A and B is blocked
by an opaque form C. But there are also cases in which the combination of
A and B is blocked by A, that is, a form that is identical to one of the two ex-
pected pieces. I call this the AbAB pattern (A blocks AB).
One example of such a pattern can be found in English plural formation. As
Baunaz and Lander (Chapter 1, Section 1.3.3.3) illustrate, English has some irreg-
ular plurals like mice that (apparently) block regular plurals such as *mouse-s (see
also Siddiqi 2006, ch.3). If we extend this treatment to sheep blocking *sheep-s,
we get AbAB.
Another case of AbAB is attested for irregular past tense formation. In the
classical case, *give-d is blocked by gave. One way to encode this is to say that
gave has the lexical entry in Figure 2.9 and that it blocks the analytical spellout
give-d, as shown in Figure 2.10.
[ 72 ] Background
T
gave
V TPAST
Figure 2.9 Lexical entry for gave
T ⇒ gave
V TPAST
give –d
Figure 2.10 Spelling out gave
Against this analysis, consider now the verb put. Here, the regular form
*put-ed is not blocked by a third form, say *pat. Instead, the past tense is actu-
ally the same as the root (put), so we have a case of AbAB.
How can we implement this? According to the Subset Principle, put cannot
have an entry like Figure 2.9, because it can also be inserted under V only.
Therefore it must be lexically specified as V, just like all regular verbs. But with
such an entry, it is a mystery why put can block *put-ed, whereas regular verbs
cannot. In technical terms, when we run the global comparison model on the
two lexicalization patterns in Figures 2.11 and 2.12, we incorrectly predict
that Figure 2.12 (where no features are left unexpressed) will be chosen over
Figure 2.11 (where past is unexpressed).
The conclusion is that even admitting global competition does not help,
because the signals we are getting from the data are contradictory. For put and
*put-ed, it seems that it is better to use a portmanteau rather than to express
all the features. But for molu and ani-al-li, it seems that it is better to express
all the features, rather than use a portmanteau.
There seems to be no easy way out of the AbAB issue. This conclusion is
(to my mind) supported in Haugen and Siddiqi (2016, 374–375), where an
T ⇒ put
V TPAST
Figure 2.11 Spelling out only V
V TPAST
put -ed
Figure 2.12 Spelling out V and TPAST
74
approach to English past tense along the lines of (10) is developed. The solu-
tion adopted in the paper is to invoke “two homophonous VIs,” one for the past
tense, and the other for the present tense. Although Haugen and Siddiqi con-
sider the need for two entries “not fatal,” I think that such a solution indicates
the inability of the approach to deal with the AbAB pattern in a systematic way.
2.4 ZERO MORPHOLOGY
Given these issues, the response within DM (and other theories) is to maintain
the initial hypothesis and deny the existence of nonterminal spellout to begin
with. Under this standard approach, the AbAB pattern is simply analyzed as
an instance of zero morphology, as seen in Figure 2.13.
This approach can be generalized for all portmanteaus (i.e., beyond the
AbAB pattern) by proposing that the overt terminal has a special allomorph
(either suppletive or derived by Readjustment Rules), as in Figure 2.14. This
is a classical DM solution going back to Halle and Marantz (1993, 124), and it
may be used as an all-purpose mechanical alternative to phrasal spellout. The
analysis of molu ‘not know’ would then be as in Figure 2.15.
This solution is rather popular in a part of the DM literature. For instance,
Trommer (1999 and 2003) suggests that if we make use of a sufficient number
of zeros (along the lines of Figure 2.14), we may get rid of Fusion altogether.
V TPAST
put Ø
Figure 2.13 Zero morpheme with put
V TPAST
broke/gave Ø
Figure 2.14 Zero morpheme with broke and gave
Neg
Neg V
Ø molu
Figure 2.15 Zero morpheme with molu
[ 74 ] Background
Trommer further suggests that also Impoverishment (an operation that
deletes features prior to insertion) may be eliminated in a similar fashion.
Although some researchers have a general feeling that the excessive use
of zero morphology is to be avoided (see e.g. Siddiqi 2006, ch.3), individual
inclinations on this question are likely to vary (as Trommer’s papers indicate).
The natural question to ask in this context is this: Is there any way to empirically
tease apart the phrasal spellout approach and the zero-morphology approach?
To address this question, I turn to a particular type of data in which the ex-
pressive power of silent morphology can be easily observed. These are the so-
called *ABA patterns that were recently investigated both within DM (Bobaljik
2012, among others) and nanosyntax (Caha 2009, among others). My exposi-
tion here is necessarily brief, but the point will hopefully get across, relying on
the background to the *ABA patterns provided in Chapter 1 (Section 1.3.3.2)
of this volume.
The particular case I discuss is a fragment of the Classical Armenian declen-
sion, seen in Table 2.2. Sources on the language include Schmitt (1981), Halle
and Vaux (1998), and Caha (2013).
The first important thing to note about the paradigms is their syncretism
structure (syncretism highlighted by shading). In particular, all the paradigms
in the language obey a constraint on syncretism such that in the order of cases
given in Table 2.2 (i.e. nom—acc—loc), only adjacent cases can be syn-
cretic. Specifically, we have the nom—acc syncretism (azg), the acc—loc
syncretism (azg-s), and also the nom—acc—loc syncretism (hogi). What
is not found is the nonadjacent nom—loc syncretism (see the penultimate
column). The absence of this pattern is usually referred to as the *ABA (* for
absence, ABA for the pattern that is absent).
To derive the *ABA, the decomposition of the cases into features must
proceed as shown in the last column. What we see there is that as we go down
in the paradigm (arranged so as to have all the syncretisms adjacent), the
number of features monotonically grows from [A]‌to [A, B] to [A, B, C] and
so on. If this “cumulative” decomposition is adopted, the *ABA property of
the paradigm is captured regardless of whether we are adopting a DM-style
insertion (see Bobaljik 2012) or a nanosyntax type of theory (Caha 2009).
Table 2.2 A FRAGMENT OF A CL ASSICAL ARMENIAN PARADIGM

(azg ‘nation,’ hogi ‘spirit’)
Case Nation, sg Nation, pl Spirit, sg Unattested Decomposition
NOM azg azg-k’ hogi A [A]‌

ACC azg azg-s hogi B [A, B]
LOC azg-i azg-s hogi A [A, B, C]
76
With the background in place, the first relevant point is that the decom-
position proposed guarantees the *ABA only as long as zero morphology
(or Impoverishment) is kept out of the picture. Suppose, for instance, that
there was a zero marker or an Impoverishment rule applying in the locative
cell, which would spell out or delete the features B and C. The effect would be
that loc is effectively turned into a nom, as shown in the middle column of
Table 2.3. The net effect is that we can no longer derive the surface absence of
the *ABA pattern; the impoverished decomposition allows what on the sur-
face looks like an ABA pattern.
Table 2.3 THE INTERACTION BETWEEN *ABA AND ZERO MORPHOLOGY
Case Decomposition Zero Applies A Surface ABA
NOM [A]‌ [A]‌ a
ACC [A, B] [A, B] b

LOC [A, B, C] [A, B, C] a (+ Ø)
I do not speculate here over whether there are some cases in which this
state of affairs is attested in some limited cases or not; the point is simply that
zero morphemes do come with a price.
The second difference starts from the observation that in a paradigm
that decomposes cumulatively, not only the exponents themselves exhibit
the *ABA distribution, but also the morphemes that are conditioned by the
features (Bobaljik 2012; McFadden 2014). This gives us an independent metric
to assess the distribution of other (say, number) morphemes in paradigms
as either compatible or incompatible with the *ABA pattern observed. What
I now want to show is that in Classical Armenian, the zero morphemes re-
quired by the terminal-only spellout system yield morpheme distributions
that are incompatible with the independently observed *ABA restriction.
To show all of this, I elaborate on the Classical Armenian paradigm by
adding the dative and the instrumental in the paradigms, as seen in Table 2.4.
The cases are again placed in an order in which syncretism is still restricted to
Table 2.4 A ZG ‘NATION,’ GET ‘RIVER ,’ BAN ‘ WORD’
Case sg pl sg pl sg pl Features
NOM azg azg-k’ get get-k’ ban ban-k’ [A]‌

ACC azg azg-s get get-s ban ban-s [A, B]
LOC azg-i azg-s get-i get-s ban-i ban-s [A, B, C]
DAT azg-i azg-a-c ’ get-o-y get-o-c ’ ban-i ban-i-c ’ [A, B, C, D]
INS azg-a-w azg-a-w-k’ get-o-w get-o-w-k’ ban-i-w ban-i-w-k’ [A, B, C, D, E]
[ 76 ] Background
adjacent cells (see e.g. azg-i), which leads to the decomposition shown in the
last column.
The new thing in these paradigms is that in the lines below loc, the mor-
phological decomposition is getting increasingly more complex, especially
in the plural. Looking for instance at the dative plural (shaded), we invari-
ably find the marker -c’. Interestingly, this marker does not attach to the root
(whose form can be seen in the nominative singular), but instead to a vowel
whose quality varies depending on the particular root. These vowels are in
bold and appear also in the instrumental (both singular and plural). They are
traditionally analyzed as a separate morpheme, called the theme marker.
The question to ask is what happens to the theme markers in nom-loc,
where we do not see them. In Caha (2013), an analysis in terms of nonterminal
spellout is proposed to explain their disappearance. The gist of the idea is that
those case markers that follow the root directly are portmanteau morphemes
that spell out both the case node and the node where the theme markers
reside, thus causing their absence. In Figures 2.16 and 2.17, I illustrate the
idea on the contrast between the dative and locative plural. Note that the
structures here are the output of several applications of movements, which
I ignore here; see Caha (2013) for details.
azg CLASS K
a c’
Figure 2.16 Dative plural
azg CLASS K
Figure 2.17 Locative plural
What is relevant now is to look at what happens when we try to recast this
analysis in terms of zeros. Such an analysis is depicted in the first two columns
of Table 2.5; all the cells where we see no theme marker are analyzed as having
a perfectly agglutinative structure with zero themes (Øth) conditioned by a
particular case–number combination.
Within each number, the alternation between the zero theme and the
overt theme is a case of a case-conditioned allomorphy. Given the decomposi-
tion of case in Classical Armenian, we expect that such an allomorphy should
78
Table 2.5 THE DISTRIBUTION OF CL ASS ZEROS
Case sg pl Features
NOM azg-Øth-ø azg-Øth-k’ [A]‌

ACC azg-Øth-ø azg-Øth-s [A, B]
LOC azg-Øth-i azg-Øth-s [A, B, C]
DAT azg-Øth-i azg- a -c ’ [A, B, C, D]
INS azg- a -w azg- a -w-k’ [A, B, C, D, E]
observe the *ABA restriction. And so it does: Both zero theme markers and
overt theme markers form contiguous regions in the paradigm. This is easy to
encode: For instance, we can say that in the plural, the overt theme marker is
conditioned by the feature D; so -a- will then appear in all cases that have this
feature in the plural, which is the dative and the instrumental, a contiguous
stretch of cases in the *ABA order.
However, problems appear when we turn to the expression of number. To see
that, consider the fact that in the instrumental plural (see the last row of Table
2.5), we obviously have an agglutinative structure, in which the instrumental
singular (ending in -w) is subject to further affixation by (plural) -k’, yielding a
sequence in which case is followed by number (as in Figure 2.1). This suggests
that there are two separate nodes in the morphosyntactic structure, each with
its own marker and that the noun moves to their left by phrasal movement.
The question is how to analyze the cases in which number and case cannot
be parsed separately. The first thing to note is that this cannot be attributed to
some kind of phonological coalescence. To see that, consider the fact that the
syncretic dat—loc.sg form azg-i has the locative plural counterpart azg-s,
whereas the dative plural is azg-a-c’; it simply cannot be that both are derived
by a phonological process starting from the string [azg-i]-k’. Therefore, both
Halle and Vaux (1998) and Caha (2013) analyze -s and -c’ as portmanteaus for
number and case (recall Figure 2.1 again). In Caha’s approach, the contrast be-
tween the agglutinative instrumental and the portmanteau dative is encoded
as in Figures 2.18 and 2.19, respectively.
azg CLASS
a K NUM
w k’
Figure 2.18 Agglutinative instrumental plural
[ 78 ] Background
N
azg CLASS
a K NUM
c’
Figure 2.19 Portmanteau dative plural
azg CLASS
K NUM
-s, -k’
Figure 2.20 Nominative and accusative–locative plural
For completeness, I also show in Figure 2.20 how the phrasal spellout anal-
ysis handles the nom, acc, and loc cases (which have only a single marker).
The idea is that -s and -k’ spell out all the relevant categories (a combination of
the proposals in Figures 2.16–2.19). See also Caha (2013) for details.
The fact that -k’ appears both as a simple number marker (in Figure 2.18)
and as a portmanteau for several categories (in Figure 2.19) is an instance of
AbAB: The expected combination of -k’ with the case and theme markers is
blocked by -k’ itself. This is important, because the AbAB is the core instance
at which DM makes recourse to zeros (if Fusion or Fission is to be avoided).
There are two ways of distributing the zeros in the plural paradigm:
(i) the zero markers spell out case (see the first column in Table 2.6), (ii) the
zero markers spell out number (see the second column in Table 2.6). In both
options -k’ spells out number.
Table 2.6 THE DISTRIBUTION OF ZEROS IN THE NUMBER AND CASE
Case Zero K Zero Num Features Portmanteau
NOM -ØK-k’ -ØK-k’ [A]‌ K + NUM = -k’

ACC -ØK-s -s-ØNum [A, B] K + NUM = -s
LOC -ØK-s -s-ØNum [A, B, C] K + NUM = -s
DAT -ØK-c ’ -c ’-ØNum [A, B, C, D] K + NUM = -c’
INS -w -k’ -w -k’ [A, B, C, D, E] K = -w NUM = -k’
80
Whichever option we look at, we end up with a *ABA violation exhibited by

the plural marker -k’. Its distribution is impossible to state using the logic il-
lustrated for theme markers, because nom and ins do not form a natural class
in the decomposition required to capture the *ABA pattern. (The same holds
also for Halle and Vaux’s 1998 feature system.)
In the specific decomposition used here, the problem takes the following shape
(relying closely on the logic explored in Bobaljik 2012): For the number marker
-k’ to appear in nom, its context of insertion must be defined by the feature A (it
can also be context free). Therefore, in principle, we expect -k’ in all cases, be-
cause all cases have A. The reason why it does not appear in acc must then be
attributed to the fact that there is a competing number marker (either zero or -s)
that is more specific; it must then be context dependent on A and B; therefore,
it wins over -k’ for insertion in the acc. However, by this reasoning, -k’ should
not appear in ins, because this case also contains A and B, so the more specific
marker should appear here as well. In effect, a language in which all paradigms
comply with the *ABA constraint is turned into one in which the distribution of
zeros introduces a violation of that constraint.
The problem is avoided under the nonterminal spellout analysis and more
generally in any analysis with portmanteau markers. On this type of approach,
the paradigm breaks down into two separate subsystems. From nom to dat,
number and case are spelled out by portmanteaus. In the last column of Table
2.6, these cells are shaded. There is no *ABA violation within this part of the par-
adigm. In the instrumental, there is a different (analytic) subsystem, in which
case and number are separate. Trivially, there is no violation of *ABA in this sub-
system either. Importantly, these two subsystems cannot be directly compared
for *ABA: One has case–number portmanteaus, and the other has case and
number separate.
The approach in terms of portmanteau marking is completely in line with
Halle and Vaux’s (1998) analysis; they too propose that the paradigm is not
surface agglutinative and that the number of morphological slots in the par-
adigm varies exactly as shown in the last column of Table 2.6. For them, the
-k’ marker also blocks a separate (zero) realization of the nominative, being
inserted into a single number–case node just like -s and -c’. However, for them,
the only way to ensure the variable number of slots is to rely on postsyntactic
operations of Fusion and Fission that manipulate the number of insertion slots
by brute force.
The main point is this: Zeros are a tool that DM can use to mimic the effects
of phrasal spellout. In fact, this is possible to the extent that apparently any
nonterminal spellout account may be mechanically translated into an account
that uses conditioned allomorphy and zeros. The main purpose of this section
was to show that the type of allomorphy needed to make this work leads to
patterns that violate an independently observed *ABA constraint. The reason
for this is obviously not that the existence of a *ABA pattern was an illusion
[ 80 ] Background
that is dispelled after the zeros are placed where need be. Rather, the reason is
that the zero morphemes are simply made up, and hence cannot be used as a
general replacement for nonterminal spellout.7
This point quite likely carries over to the past tense of English verbs. As
Bobaljik (2012, 160) as well as Lander (2015, 96) note, there seems to be a
*ABA constraint on the forms of irregular verbs; the sequence is present—
participle—past (consider come—come—came; shine—shone—shone). As
Bobaljik and Lander agree, this can be captured by a cumulative decompo-
sition in which the number of features monotonically grows in the relevant
sequence. As a consequence, we should not be much compelled by an analysis
like the one in Figure 2.14 that leads to parses such as break-Ø, broke-n, broke-
Ø, because such artificial parses once again create a *ABA violation where
none is attested.
At this point, we have gone the full circle. Discarding Fusion, the conclu-
sion was that zero marking offers a way out of the dilemmas. However, the
distribution patterns of such zeros are incompatible with some of the known
constraints on morpheme distributions.
In view of these problems, some of the recent work (Radkevich 2010;
Bobaljik 2012; Haugen and Siddiqi 2016) tries to extend DM by making use of
nonterminal spellout. However, we have seen that the Subset Principle turns
this into a rather difficult task: The subset logic that works so beautifully for
paradigms such as the one in Table 2.1 breaks down when one tries to extend
it to nonterminal spellout.
As mentioned, the strategy has been to avoid these problems by adding
principles that somehow restrict the logic inherent in the Subset Principle; we
have looked at Radkevich’s VIP, or a global comparison model (Julien 2015;
Haugen and Siddiqi 2016). None of these fully work (I think), and so we are
still waiting for an insertion procedure based on the Subset Principle that
would deliver in all the intricate cases. This may well be impossible, as Embick
and Marantz seemed to think back in 2008.
Still, in a sense, all these strategies equal to giving up the Subset Principle for
nonterminal spellout, because their ultimate goal is to allow underspecification
7. An anonymous reviewer thinks that there is something special about making an
argument based on the distribution of silent markers. The reasoning is that although
“there are numberless ways of being pronounced if you are an overt morpheme, there
is only one thing silence sounds like,” with the conclusion that “the *ABA constraint is
expected to be vacuous when it comes to the distribution of zeroes.”
The conclusion does not follow because the premise is confused. Zero is simply one
of the “numberless ways” in which features can be pronounced, /ba/being another. So
saying that “there is only one thing silence sounds like” is like saying that “there is only
one thing /ba/sounds like.” I don’t see why that would constitute a reason to think that
/ba/should not obey the *ABA (and by analogy, I don’t see why the fact that there is
only one way to pronounce a zero morpheme constitutes a reason to think that it will
not obey the *ABA).
82
only at terminal nodes and to neutralize its effects at nonterminals. This is ex-
plicitly hinted at in Bye and Svenonius (2010); in their fn.12, they write, “The
Subset Principle [ . . . ] would lead to the result that every sentence would be
at most one morpheme. To avoid this empirically false result, [ . . . w]e might
for example assume that [an additional principle of] Exhaustive Lexicalization
holds of projecting categories [ . . . ].” The Exhaustive Lexicalization Principle
(Fábregas 2007) says that every feature must be lexicalized; in the passage
just quoted, Bye and Svenonius note that if the principle held only for heads
(each head must be lexicalized), then the communication breakdown would
be avoided. Ultimately, this approach equals to admitting two insertion prin-
ciples: the Subset Principle for terminals and the “Spell Out Every Head”
Principle for nonterminals.
However, doing so first misses Fábregas’s point, which was that the Subset
Principle is a nonstarter from the perspective of the Exhaustive Lexicalization
Principle, because it “has the effect that some syntactic features are not
lexicalized by any piece” (Fábregas 2007, 174). More important, this approach
is also not going to provide a systematic account of the AbAB pattern, in which
one head goes unexpressed.
The last thing to keep in mind is that these points are equally valid for all
nonterminal insertion theories, regardless of whether they apply to sequences
of heads (Bye and Svenonius 2010; Haugen and Siddiqi 2016) or to nodes
(Radkevich 2010). All such approaches run into the same problems unless
they give up the Subset Principle and adopt an alternative approach like the
one developed in nanosyntax.
2.5 THE SUPERSET PRINCIPLE
There is a single reason underlying all these puzzles: The insertion procedure is

not governed by the Subset Principle. To make things work, we have to adopt
the Superset Principle instead (Starke 2009). I give one possible formulation
in (11), which is quite different from Starke’s wording, but it is close to the
original Subset Principle (so that we can see what the difference actually is). It
is also not too remote from Starke’s idea.
(11) The Superset Principle

[The Superset Clause]: The phonological exponent of a Vocabulary Item
is inserted into a node if the item contains all (or a superset of) the
grammatical features contained in the node. Insertion does not take
place if the Vocabulary Item is not specified for all features contained
in the node. [The Elsewhere Clause]: Where several items meet the
conditions for insertion, the item containing fewer features unspecified
in the node must be chosen.
[ 82 ] Background
Potentially, this principle may apply to terminal, nonterminal, and phrasal
nodes. Let me now show how adopting it solves all the puzzles we have seen.
2.5.1 AbAB
I start from the AbAB. With the Superset Principle in place, we can specify both
put and gave in the same way, namely as irregular past tense forms, as in (12a).
With this lexical entry, they can spell out the nonterminal in Figure 2.10, be-
cause they contain all its features. This is not the case for regular verbs, which
have a lexical entry like the one in (12b). They do not contain the T node, and
so they cannot be inserted at the nonterminal in Figure 2.10.
(12) a. put, gave, sang, . . . = [V TPAST]

b. kiss, locate, . . . = V
c. give, sing, . . . = V
Notice now that all the verbs in (12a) can also spell out the V node only, be-
cause V is contained in them. So this explains why put can have the same form
for the past tense and the root, whereas regular verbs cannot.
Finally, the reason why gave and sang do not show up as the root form of
the relevant verb is because of competition. The Vocabulary Items sing and
give (12c) also spell out the V node, and in this capacity, they contain fewer un-
used features, so they outcompete the past tense form in this environment.8
The same logic extends to the AbAB pattern exhibited by sheep and the
Classical Armenian -k’, as well as other such cases (see Caha, 2017b, for a
discussion of yet another AbAB pattern exhibited by the so-called locative
nouns).
2.5.2 Korean Negation
Recall now the problem we had with molu. It was specified as [Neg KNOW]
(recall Figure 2.4) and the question was why it cannot spell out the causative
8. An anonymous reviewer asks about the treatment of forms such as told. If they do
not decompose (see Kaye 1995, 310–312, for some arguments), then their treatment is
the same as for gave in Figure 2.14. If they do decompose, something extra needs to be
said. Space limitations prevent me from going into this here, but the alternative anal-
ysis would postulate at least three heads in the English past tense verb (see Kayne’s
2016 decomposition into bak-ə-d), and propose that tol-is a portmanteau for the lower
two, i.e. bak-ə-, with -d a shared affix. See Caha (2017a) and De Clercq and Vanden
Wyngaerd (2017) for the discussion of forms such as bett-er (which represent an anal-
ogous case of a suppletive form accompanied by what looks like regular morphology).
84
form [Neg [KNOW Cause]] (recall Figure 2.6). The answer is now clear: It is
because the lexical entry in Figure 2.4 does not contain all the features of the
top node in Figure 2.6; Cause is missing.
2.5.3 How to create a simple paradigm
Recall finally the paradigm in Table 2.1, repeated as Table 2.7 for convenience.
How does the Superset Principle deal with this example?
Table 2.7 A FRAGMENT OF THE L ATIN DECLENSION
Neuter Masculine
[+singular] [-masc] [+masc]
NOM -um -us

[+superior]
ACC -um -um
[-superior]
Let me first say that because of independent assumptions (recall dif 1),
there are no complex terminals in nanosyntax, so the following discussion is
artificial with respect to the actual working of the Superset Principle in that
framework. Still, for completeness of the argument, let me show that the
Superset Principle could also be used to handle these cases.
If we want to encode the same intuition as in (2), namely that -um is
a marker that can appear in any cell of the paradigm, it has to contain all
the potential feature combinations. Its specification is thus as shown in
(13a).9
(13) a. -um = [+masc, – masc, +sup, – sup, +sg]

b. -us = [+masc, +sup, +sg]
If we now specify -us as shown in (13b), it will qualify for insertion only in the
nominative singular of the masculine gender, because it does not contain the
features of any other cell. The lexical entries in (13) thus derive the paradigm
in Table 2.7, which shows that we have not lost the cases that were working
fine under Subset Principle–driven insertion.
9. This entry is admittedly quite inelegant, a side effect of the terminal spellout sce-
nario. See Caha (2009) for a system of case decomposition that makes such lexical
items look neater.
[ 84 ] Background
2.6. CONCLUSIONS
By looking at the difficulties of classical DM, we come to realize that nanosyntax

is not just a particular constellation of the bits and pieces that were inde-
pendently available before (like cartography and phrasal spellout). Without
changing the principle of insertion, the combination of these proposals simply
does not take off. In retrospect, employing the Superset Principle and thereby
creating a well-working theory seems trivial. At the same time, when Embick
and Marantz were contemplating these issues back in 2008, the prospects of
any breakthrough in this domain seemed impossible, and recent attempts to
extend the DM model to handle nonterminal spellout still struggle to counter
the inadequacy of the Subset Principle.
From this perspective, the Superset Principle and nanosyntax in
general has opened a whole new set of analytical options to consider for
various phenomena. Unlike the half-working alternatives, it has made
it possible to abandon Fusion and the paradoxes that came with it. By
doing so, it creates the prospect for a simplified architecture of grammar
where syntactic structures are mapped onto their pronunciation by lexical
access only.
REFERENCES
Anderson, Stephen. 1992. A-morphous Morphology. Cambridge: Cambridge University

Press.
Baker, Mark and Ruth Kramer. 2014. “Rethinking Amharic Prepositions as Case
Markers Inserted at PF.” Lingua 145: pp. 141–172.
Bobaljik, Jonathan. 2012. Universals in Comparative Morphology. Suppletion, Superlatives,
and the Structure of Words. Cambridge, MA: MIT Press.
Bobaljik, Jonathan. 2015. “Distributed Morphology.” Ms, University of Connecticut.
[online] Available at http://bobaljik.uconn.edu/papers/DM_ORE.pdf
Bye, Patrick and Peter Svenonius. 2010. “Exponence, Phonology, and Non-
Concatenative Morphology.” Ms. CASTL, University of Tromsø.
Caha, Pavel. 2007. “Case Movement in PPs.” Nordlyd 34 (2): pp. 239–299. [online]
Available at http://septentrio.uit.no/index.php/nordlyd/index
Caha, Pavel. 2013. “Explaining the Structure of Case Paradigms by the Mechanisms of
Nanosyntax.” NLLT 31 (4): pp. 1015–1066.
Caha, Pavel. 2017a. “Suppletion and Morpheme Order: a Unified Account.” Journal of
Linguistics 53 (4): pp. 865–896.
Caha, Pavel.2017b. “How (not) to derive a *ABA: the case of Blansitt’s generalization.”
Glossa: A Journal of General Linguistcs 2 (1): pp. 84–115.
Chung, Inkie. 2007. “Suppletive Negation in Korean and Distributed Morphology.”
Lingua 117 (1): pp. 95–148.
Cinque, Guglielmo. 1996. “The antisymmetric programme: Theoretical and typological
implications.” Journal of Linguistics 32: pp. 447–464.
86

Craenenbroeck, Jeroen van and Jason Merchant. 2013. “Ellipsis Phenomena.” In
The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken,
pp. 701–745. Cambridge: Cambridge University Press
De Clercq, Karen and Guido Vanden Wyngaerd. 2017. “*ABA Revisited: Evidence From
Czech and Latin Degree Morphology.” Glossa: A Journal of General Linguistics 2
(1): 69. pp. 1–32.
Embick, David. 2014. On the Targets of Phonological Realization. Ms., University of
Pennsylvania.
Embick, David and Rolf Noyer. 2001. “Movement Operations After Syntax.” Linguistic
Inquiry 32 (4): pp. 555–595.
Embick, David and Rolf Noyer. 2007. “Distributed Morphology and the Syntax/
Morphology Interface.” In The Oxford Handbook of Linguistic Interfaces, edited by
Gillian Ramchand and Charles Reiss, pp. 289–324. Oxford: Oxford University
Press.
Embick, David and Alec Marantz. 2008. “Architecture and Blocking.” Linguistic Inquiry
39 (1): pp. 1–53.
Fábregas, Antonio. 2007. “The Exhaustive Lexicalisation Principle.” Nordlyd 34 (2): pp.
165– 199. [online] Available at http://septentrio.uit.no/index.php/nordlyd/
index
Halle, Morris. 1997. “Distributed Morphology: Impoverishment and Fission.” In
PF: Papers at the Interface, edited by Benjamin Bruening, Yoonjung Kang, and
Martha McGinnis, pp. 425–450. Cambridge, MA: MIT Press.
Inflection.” In The View from Building 20: A Festschrift for Sylvain Bromberger,
edited by Kenneth Hale and Samuel J. Keyser, pp. 111–176. Cambridge, MA:
MIT Press.
Halle, Morris and Bert Vaux. 1998. “Theoretical Aspects of Indo-European Nominal
Morphology: The Nominal Declensions of Latin and Armenian.” In Mir Curad:
Studies in Honor of Calvert Watkins, edited by Jay Jasanoff, H. Craig Melchert, and
Lisi Oliver, pp. 223–240. Innsbruck, Austria: Institut für Sprachwissenschaft
der Universität Innsbruck.
Harley, Heidi and Rolf Noyer. 1999. “Distributed Morphology.” Glot International 4 (4):
pp. 3–9.
Haugen, Jason H. and Daniel Siddiqi. 2013. “Roots and the Derivation.” Linguistic
Inquiry 44: pp. 493–517.
Haugen, Jason H. and Daniel Siddiqi. 2016. “Towards a Restricted Realization Theory.
Multimorphemic Monolistemicity, Portmanteaux, and Post- Linearization
Spanning.” In Morphological Metatheory, edited by Daniel Siddiqi and Heidi
Harley, pp. 343–386. Amsterdam: John Benjamins.
Julien, Marit. 2015. “Theme Vowels in North Sámi: Spanning and Maximal Expression.”
Lingua 164: pp. 1–24.
Kaye, Jonathan. 1995. “Derivations and Interfaces.” In Frontiers of Phonology, ed-
ited by Jacques Durand and Francis Katamba, pp. 289– 332. London and
New York: Longman.
Kayne, Richard S. 1998. “Overt vs. Covert Movements.” Syntax 1 (2): pp. 128–191.
Kayne, Richard S. 2010. “Towards a Syntactic Reinterpretation of Harris and Halle
(2005).” In Romance Languages and Linguistic Theory 2008: Selected Papers From
[ 86 ] Background
the ‘Going Romance’ Groeningen 2008, edited by Reineke Bok-Bennema, Brigitte
Kampers- Manhe, and Bart Hollebrandese, pp. 145– 170. Amsterdam: John
Benjamins.
Kayne, Richard S. 2016. “What is Suppletive Allomorphy? On ‘went’ and on ‘*goed’ in
English.” Ms., New York University. [online] Available at lingbuzz/003241.
Koopman, Hilda. 1996. “The Spec Head Configuration.” In Syntax at Sunset: UCLA
Working Papers in Syntax and Semantics, edited by Edward Garrett and Felicia
Lee, pp. 37–64. Los Angeles, CA: UCLA Dept. of Linguistics.
Koopman, Hilda. 2005. “Korean (and Japanese) Morphology From a Syntactic
Perspective.” Linguistic Inquiry 36 (4), pp. 601–633.
Lander, Eric. 2015. The Nanosyntax of the Northwest Germanic Reinforced Demonstrative.
Leu, Thomas. 2008. The Internal Syntax of Determiners. Doctoral dissertation, New York
University.
Marantz, Alec. 1989. “Clitics and Phrase Structure.” In Alternative Conceptions of
Phrase Structure, edited by Mark Baltin and Antony Kroch, pp. 99–116. Chicago:
University of Chicago Press.
McFadden, Thomas. 2014. Noun Stem Alternations, the Structure of Case and the Locality
of Allomorphy. Handout of a talk at LAGB, University of Oxford, September 5.
Merchant, Jason. 2015. “How Much Context is Enough? Two Cases of Span-
Conditioned Stem Allomorphy.” Linguistic Inquiry 46: pp. 273–303.
Pullum, Geoffrey K. and Arnold M. Zwicky. 1992. “A Misconceived Approach to
Morphology.” In Proceedings of the West Coast Conference on Formal Linguistics Vol.
10, edited by Dawn Bates, pp. 387–398. Chicago: University of Chicago Press.
Radkevich, Nina V. 2010. On Location: The Structure of Case and Adpositions. Doctoral
dissertation, University of Connecticut.
Schmitt, Rüdiger. 1981. Grammatik des Klassich- Armenischen. Innsbruck, Austria:
Institut für Sprachwissenschaft der Universität Innsbruck.
Siddiqi, Daniel. 2006. Minimize Exponence: Economy Effects on a Model of the
Morphosyntactic Component of the Grammar. Doctoral dissertation, University of
Arizona.
Nordlyd 36 (1): Special Issue on Nanosyntax 36, edited by Peter Svenonius, Gillian
Ramchand, Michal Starke, and Knut T. Taraldsen, pp. 1–6. University of Tromsø.
[online] Available at www.ub.uit.no/munin/nordlyd.
Trommer, Jochen. 1999. Morphology Consuming Syntax’ Resources. Ms, Universität
Potsdam.
Trommer, Jochen. 2003. “Feature (Non- )Insertion in a Minimalist Approach to
Spellout.” In Proceedings of CLS 39, edited by Jon Cihlar, Amy Franklin, David
Kaiser, and Irene Kimbara, pp. 469–480. Chicago: University of Chicago Press.
88
CHAPTER 3
Spanning versus Constituent

Lexicalization
The Case of Portmanteau Prefixes
KNUT TAR ALD TAR ALDSEN
3.1. CONSTITUENT LEXICALIZATION AND SPANNING
In this section, I introduce two different incarnations of nanosyntax, one

that holds that a morpheme can lexicalize only a constituent (constituent
lexicalization) and one that allows a morpheme to lexicalize a nontrivial se-
quence of heads even when those heads do not form a constituent (spanning).
I also offer a fairly general assessment of the relative merits of the two
approaches as a prelude to more detailed discussion in Sections 3.2 and 3.3.
3.1.1 Submorphemic syntactic terminals
According to the classical view of the relation between morphemes (or

Vocabulary Items) and pieces of syntactic structure, a single morpheme
can replace only a single terminal node (head). This view has survived
into Distributed Morphology (DM), the dominant late- insertion model.
Correspondingly, DM also posits a postsyntactic process called Fusion, which
turns a sequence of two adjacent terminals into a single terminal to handle
cases in which a single morpheme plainly replaces two terminals, for example,
went replacing T (= Past) and v/V.
In recent years, the classical view has been challenged, for example, by
pointing out the ad hoc nature of Fusion and, in particular, the fact that Fusion
can apply only when there happens to be a morpheme that can faithfully re-
place its output. By allowing a single morpheme to lexicalize a set of more
than one single terminal, the need for Fusion is eliminated and the effects
attributed to Fusion will depend directly on the existence of morphemes
specified as being capable of lexicalizing a nontrivial set of terminals: The two
heads in [X [Y will be lexicalized as a unit only if there is a morpheme M with
the lexical entry M ⇔ [X [Y. (This is discussed in detail by Caha in Chapter 2.)
In the following sections, I take it for granted that a model allowing a single
morpheme to lexicalize more than a single terminal is superior to the classical
model. However, there are two main implementations whose relative merits
need to be discussed.
3.1.2 Constituent lexicalization versus spanning
The most obvious way of allowing a single morpheme to spell out a nontrivial
set of terminals is to allow lexical insertion to target phrasal nodes. That is,
when M replaces the two terminals XY, lexical insertion applies to the node
labeled XP in (1):
(1) [XP X [YP Y]]
Correspondingly, the lexical entry for M will be as in (2):
(2) M ⇔ [XP X [YP Y]]
Lexical insertion is licensed by a matching relation between a structure built

by the syntax and the structure associated with a morpheme in the lexicon,
as stated in (3):
(3) A syntactic structure S matches a morpheme M with the entry M ⇔ Sʹ

if and only if the root node R of S has the same label as some node N in
Sʹ, and (a) every daughter of R matches a daughter of N and (b) every
daughter of N matches a daughter of R.
Notice that this incorporates the Superset Principle: M with the entry in
(2) can also lexicalize just YP not embedded under X. Notice also that two adja-
cent terminals XY can be lexicalized by M only when they form a constituent.
In (4), YP has a daughter [complement in (4a), or specifier in (4b)] that does
not match a daughter of YP in the structure associated with M in (2):
(4) a. [XP X [YP Y [ZP Z]]]

b. [XP X [YP ZP Y]]
S pa n n i n g v e r s u s C o n s t i t u e n t L e x i c a l i z at i o n [ 89 ]
90
Thus, in the model just described, only a constituent can be lexicalized by a

single morpheme. This is the position taken by Caha (2009), Starke (2009; see
also Chaper 9 of this volume), and Pantcheva (2011). I refer to it as constituent
lexicalization (CL).
But there is a competing view originating from Williams (2003) (and, with
a different implementation, from Brody’s Mirror Theory) and adopted by Abels
and Muriungi (2008), Taraldsen (2010), and Svenonius (2012). According to this
view, any “span” of terminals can be lexicalized by a single morpheme. See (5):
(5) An ordered set < X1, . . . , Xn > of terminals is a span in the structure S
if and only if
a. XnP is the complement of Xn–1 (for n > 1) and
b. every Xi belongs to a single projection of a functional sequence.
In contrast to CL, spanning allows nonconstituents to be lexicalized by a single

morpheme. For example, M with the entry in (2) can lexicalize XY in (4).
3.1.3 Conceptual issues
CL may have only one conceptual advantage over spanning: The basic unit for
lexical insertion is the constituent, an entity independently motivated for ex-
ternal and internal merge. The notion of span, however, would appear to play
a role only for lexical insertion. Hence, parsimony would seem to favor CL.
With respect to predictive power, we have already seen that spanning is
less restrictive than CL when it comes to “chunk lexicalization” of heads in
structures like (4). Spanning, but not CL, allows a morpheme to replace just X
and Y in (4) to the exclusion of Z or ZP.
Based on the definition of matching in (3), CL also predicts that a mor-
pheme M with the entry M ⇔ [XP X [YP Y [ZP Z]]] cannot lexicalize XY in (6),
where Z is missing:
(6) [XP X [YP Y]]
This is because the node YP in the structure associated with M has a daughter
ZP that finds no match in (6) so that requirement (b) in (3) is not satisfied.
More generally, CL based on (3) automatically predicts that any M asso-
ciated with [XP . . . [ZP Z]] can lexicalize only sequences of heads containing the
bottom element Z. By contrast, spanning allows M ⇔ < X, Y, Z > to lexicalize
just XY as well, if the Superset Principle is built in by (7):
(7) A morpheme M with the entry M ⇔ S, where S is a span, can lexicalize

any subspan of S.
[ 90 ] Background
This is because < X, Y > is a subspan of < X, Y, Z > by (5).
Abels and Muriungi (2008), who use spanning, reach the conclusion that
their analysis requires adding the Foot Condition in (8).
(8) A morpheme M with the entry M ⇔ S, with S a span, can lexicalize only
those subspans of S that contain the final element in S.
The empirical observations that led Abels and Muriungi to adopt (8) could be
seen as evidence supporting CL over spanning, because the Foot Condition, as
we have seen, follows directly from (3).
However, the Foot Condition follows from (3) only because (3) includes
clause (b), and clause (b) is not required to restrict lexicalization to
constituents. Even if (3) is replaced with (9), which retains only clause (a),
only constituents can be lexicalized:
(9) A syntactic structure S matches a morpheme M with the entry M ⇔ Sʹ

if and only if the root node R of S has the same label as some node N in
Sʹ, and every daughter of R matches a daughter of N.
Thus the Foot Condition is actually independent of the idea that only
constituents can be targeted by lexicalization. We also note that Caha has
proposed replacing (3) with (9) for reasons having to do with lexicalization
across distinct layers of functional heads.1
Finally, it should be acknowledged that CL is more permissive than
spanning [with spans characterized as in (5)] in that it allows a morpheme
to lexicalize everything inside a Spec-XP in addition to the head X (and its
complements, if there are any).
All things considered, the choice between CL and the spanning version of
nanosyntax cannot easily be made on very general grounds. Hence we now
turn to specific empirical arguments for or against restricting lexicalization to
constituents and begin with a recent argument for spanning.
3.1.4 A purported empirical argument for spanning
The conceptual advantage of limiting lexicalization to constituents will

be nullified, if it turns out that the added restrictiveness makes incorrect
predictions not shared by the spanning approach. In the literature, one
1. As formulated in (3), the matching requirement calls for “pointers” to handle cer-
tain kinds of syncretism (cf. Caha 2009). Alternatively, the bidirectional matching re-
quirement is replaced by a unidirectional one, as suggested by Caha (p.c.), i.e. “and
every daughter of N matches a daughter of R” is removed from (3).
92
striking example of this sort has received some attention (see Embick 2012;
Haugen and Siddiqi 2013; Merchant 2015). This is the so-called contain-
ment problem, which arises as an immediate consequence of the fact that CL
restricts lexicalization to constituents.
Consider, for example, the treatment of the irregular past-tense form ate
within CL. A reasonable assumption is that ate is a single morpheme replacing
(at least) the heads T (= Past), v, and V:
(10) ate ⇔ [TP Past [vP v [VP EAT]]]
The containment problem then arises from the assumption that the direct ob-
ject is included in the TP in sentences like (11):
(11) I ate an apple.
In (12), for example, T, v, and V do not form a constituent excluding an apple:
(12) I [TP Past [vP <I> v [VP EAT [DP an apple]]]]
But the force of an argument against CL based on this is obviously no stronger

than the arguments for assigning the structure in (12) to (11). If (11) should in-
stead be parsed as (13) at the point where ate is inserted, CL has no problems
(provided traces are ignored by the lexicalization procedure):
(13) I [TP Past [vP <I> v [VP EAT < DP> ]]] [[DP an apple ] [ <TP> ]]
In this particular case, there are in fact independent arguments for the
movements involved in deriving the structure required by CL. Johnson’s
(2004) analysis of gapping, for example, assumes both that English deploys
raising of object DPs and raising of a remnant containing the verb across the
raised object.
The more general point, however, is that any instantiation of the contain-
ment problem will be based on assumptions about constituency that are not
obviously valid. Rather, it seems that an evaluation of CL based on contain-
ment makes sense only in the context of competing detailed analyses of de-
rived structures in different languages.
On the other hand, in some cases in which a containment scenario has
been used to motivate a spanning analysis, further considerations may re-
veal that the spanning analysis actually makes incorrect predictions that
would not be shared by a CL account. Consider, for example, the analysis in
Svenonius (2012) of French “contracted” preposition+article sequences like
au ‘on/to the.m.sg’ and du ‘of the.m.sg.’ Svenonius makes a plausible case that
[ 92 ] Background
these should be seen as portmanteau morphemes spanning P and le ‘the.m.
sg.’ Thus, we would have the entries in (14)2:
(14) a. au ⇔ < À, LE >

b. du ⇔ < DE, LE >
Assuming the conventional structures in (15) for au château ‘to/at the castle’
and du livre ‘of the book,’ a CL analysis replacing the spans in (15) with the
constituents [PP À [ArtP LE]] and [PP DE [ArtP LE]] clearly would not work.
(15) a. [PP À [ArtP LE [NP château]]]

b. [PP DE [ArtP LE [NP livre]]]
But the spanning analysis would also predict that a P and a definite article
might contract in a language like Romanian in which the noun precedes the
definite article, provided the noun is a specifier [cf. (5b)].
(16) [PP P [ArtP [NP N ] Art <NP> ]]
In (16), P and Art are a span just as in (15). Thus a language like Romanian
could have portmanteau forms like au and du that would be linearized either
preceding the noun or following it.3 Yet, although Romance languages like
French, Italian, and Portuguese have contracted P + Art sequences, Romanian
does not. If all languages in which the noun precedes the definite article are
like Romanian in this respect, it will appear that the spanning analysis makes
an incorrect prediction.
3.1.5 Empirical arguments for constituent lexicalization
Constituency is also relevant to the evaluation of an empirical observation

that seems to favor CL. Kayne (1994, 52–53) claims that “in a general way”
agglutinating morphology and strictly head-final word order go together. He
proposes that in a strictly head-final language, each head in (17a) will attract its
complement to its specifier position, ultimately yielding (17b) (ignoring traces):
(17) a. [XP X [YP Y [ZP Z LP]]]

b. [XP [YP [ZP LP Z] Y] X]
2. I set aside the necessity to restrict insertion of au and du to environments where

they are immediately followed by a consonant.
3. In a spanning framework, the linearization of a morpheme M replacing X and Y
separated by a specifier is not directly determined.
94
Kayne presupposes that ZYX cannot “fuse” in (17b). But why can a single
portmanteau morpheme M not lexicalize all of ZYX? If lexicalization
targets only constituents, the answer is that ZYX is not a constituent.
With spanning, the answer has to come from clause (a) in (5): In (17b),
YP is no longer the complement of X, and ZP is not the complement of
Y. But as we have seen, clause (a) is redundant in a CL framework. Hence,
CL seems to provide the simplest theoretical account for Kayne’s empirical
observation.
It must be acknowledged, though, that the hedge “in a general way” suggests
exceptions to Kayne’s generalization that may not be consistent with CL, as
pointed out by an anonymous reviewer who mentions Quechua languages as
(partial) exceptions. It remains to be seen whether such languages really are
strictly head-final in Kayne’s sense, in the corners of the system in which two
(or more heads) are represented by a single morpheme.
Another striking prediction of CL is that there should be no portman-
teau prefixes for exactly the same reason CL will not allow À/DE and LE to be
replaced by a portmanteau au or du in (15). In Section 3.2, I will argue that this
prediction is actually borne out in a very specific way.
3.2 NOMINAL CLASS PREFIXES IN SOUTHERN BANTU
This section focuses on certain properties of noun class prefixes in Southern

Bantu languages, primarily Xhosa. These are portmanteau prefixes, and the
goal is to present an argument that they spell out exactly the kind of structures
that CL predicts.
3.2.1 The singular–p lural pairing
In Bantu languages, noun classes are individuated on the basis of the prefixes
attached to nouns and the agreement markers they co-occur with. Typically,
some noun classes are exclusively singular whereas others are plural. This is
illustrated in the paradigm for Xhosa nominal prefixes in Table 3.1, where the
plural classes are placed in the right-hand column.4
Table 3.1 also illustrates how plural classes are paired with singular classes.
A noun in one of the singular classes in the left has its plural form in the class
right next to it (if there is one). For example, a noun in class 1 has a plural form
in class 2, for example, m-ntu ‘person’—ba-ntu ‘people.’
4. This table shows only the basic class prefixes, not taking into consideration phono-
logically induced effects and ignoring the “augments,” i.e. the initial vowels that appear
in front of the basic prefixes in various syntactic environments.
[ 94 ] Background
Table 3.1 XHOSA NOMINAL PREFIXES
Class sg Class pl
1 m 2 ba
3 m 4 mi
5 Li 6 ma
7 si 8 zi
9 N 10 ziN
11 lu
14 bu
15 ku
Table 3.2 SINGUL AR–P LURAL

PAIRING BY GENDER
Gender sg pl
I m ba
II m mi
III li ma
IV si zi
V N ziN
The obvious way to capture this pairing of singular and plural classes is to
say that a noun is associated with the same gender feature both in the sin-
gular and the plural forms and to take the spellout of the prefix to be sensitive
both to number and gender (Carstens 1991). This is shown in Table 3.2, where
Roman numerals represent gender features.
From a nanosyntactic perspective, it therefore becomes plausible to see
the nominal prefixes as portmanteau prefixes spelling out both gender and
number. But if we adopt the idea that there is a one-to-one relation between
features and syntactic heads, this seems to entail that a single morpheme
lexicalizes Num (holding the number feature) and Cl (holding the gender fea-
ture inherited from the N) as a unit in Figure 3.1.5
In Figure 3.1, however, Num and Cl form a span, but not a constituent. But
in the following subsections, I argue that Figure 3.1 is actually not the right
structure.
5. Or the spellout of Num is made sensitive to gender features on the adjacent nouns,
as in Carstens (1991).
96
[NumP Num [ClP Cl [NP N ]]]
prefix
Figure 3.1 Single prefix lexicalizing Num and Cl
3.2.2 Portmanteau pref ixes as specif iers
First, I present a CL-compatible alternative to Figure 3.1, building on a suggestion

by Starke (personal communication, hereafter p.c.). The basic idea is that the
heads Num and Cl spelled out by the prefix are not heads on the extended projec-
tion line of the N, but rather occur inside a separate phrasal constituent, a speci-
fier. In first approximation, this would look like Figure 3.2 for a plural class prefix.
For example, for the plural form ba-ntu ‘people’ (class 2), we would have
Figure 3.3.
(The Roman numeral I in Figure 3.3 represents the gender feature shared
by classes 1 and 2 in Table 3.2.) The implicit assumption is that a version of the
“extended doubly filled COMP constraint” holds: A head is not pronounced,
when its specifier(-chain) is.
However, there is a conflict between the structures in Figures 3.2 and 3.3
and the generally accepted view that functional heads like Num and Cl project
only on top of a lexical head. In particular, nominal functional heads should
project only on top of a lexical N. Adopting this view, Figures 3.2 and 3.3 are
to be replaced with Figures 3.4 and 3.5.6
I now start developing arguments to the effect that nominal class prefixes
in Southern Bantu actually do lexicalize phrasal constituents bottoming out
in a noun distinct from the overt noun that follows the prefix in phonological
form.
[NumP [NumP Plural [ClP Cl ]] Num [ClP <NumP> Cl [NP N ]]]
prefix
Figure 3.2 Plural class prefix as a specifier
[NumP [NumP Pl [ClP I ]] Pl [ClP <NumP> I [NP N ]]]
ba-
Figure 3.3 Class 2 prefix ba-as a specifier
6. I assume that the unpronounced Num and Cl are merged with unvalued features that
get valued by the Num and Cl in the specifier. Thus gender is always inherited from N1.
[ 96 ] Background
[NumlP [NumP Pl [ClP Cl [NP N1 ]]] Pl [ClP <NumP> Cl [NP N2 ]]]
prefix
Figure 3.4 Plural class prefix including a lexical noun
[NumlP [NumP Pl [ClP I [NP N1 ]]] Pl [ClP <NumP> I [NP N2 ]]]
ba-
Figure 3.5 Class 2 prefix ba-including a lexical noun
3.2.3 Irregular plurals
The first clue that Figure 3.5 is the right structure comes from “irregular”
singular–plural pairings. In particular, some nouns in class 1 actually form
plurals (only) in class 6 rather than class 2, as seen in (18)7:
(18) a. u-m-Xhosa ‘a Xhosa’ / a-ma-Xhosa ‘Xhosas’

b. u-m-Zulu ‘a Zulu’ / a-ma-Zulu ‘Zulus’
Some class 9 nouns also have plurals (only) in class 6, as seen in (19):
(19) a. i-n-doda ‘a man’ / a-ma-doda ‘men’

b. i-n-kwenkwe ‘a youngster’ / a-ma-khwenkwe ‘youngsters’
If the spellout of prefixes is conditioned by gender features and number

features as suggested in Table 3.2, the irregular pairs in (18) and (19) must
be handled by positing additional genders and invoking syncretism (see
Carstens 1991). For example, if gender features are decomposed in a suit-
able way, we might obtain a paradigm in which a gender G shares a fea-
ture (value) both with gender I and with gender III, assign Xhosa and Zulu
to gender G and say that {sg, G} syncretizes with {sg, I}, whereas {pl, G}
syncretizes with {pl, III}. The same strategy can also be used with doda and
khwenke.
Outside of Bantu, invoking an extra gender and a two-way syncretism may
provide a plausible account of a sizable class of neuter nouns in Romanian
7. These are all nouns denoting members of a tribe or ethnic group. The significance
of this remains to be determined.
98
Table 3.3 TWO-W AY SYNCRETISM IN SOME

NEUTER NOUNS IN ROMANIAN
sg pl
masc bărbat-Ø bărbaţ-i ‘man’
neut deget-Ø deget-e ‘finger’
fem cas-a cas-e ‘house’
that inflect like masculine nouns in the singular, but like feminine nouns in
the plural, as shown in Table 3.3.8
For example, there might be no inflections specifically targeting all
the gender features that “neuter” is broken up into; instead there are two
underspecified morphemes, one for singular plus the gender feature common
to masculine and neuter, and one for plural plus a different gender feature
common to neuter and feminine.
However, in Section 3.2.4 we see that whereas this type of account is
plausible for the Romanian neuter nouns, it will not work for the irregular
singular–plural pairs in Xhosa.
3.2.4 Agreement with conjoined singular nouns
Acquaviva (2008) proposes a useful way of testing the validity of the idea that
the two members of pairs like deget-Ø ‘finger’ /deget-e ‘fingers’ have the same
gender feature(s) (neuter). The experiment is based on what we see in (20):
(20) a. Nişte bărbaţ-i au fost arestaţ-i

some man-m.pl have been arrested-m.pl
‘Some men have been arrested.’
b. Un bărbaţ şi un copil au fost arestaţ-i
a.m.sg man and a.m.sg child have been arrested-m.pl
‘A man and a child have been arrested.’
We see that when the subject is a conjunction of two singular-masculine

nouns (20b), the agreement ending of the participle is the masculine-plural
ending, just as in (20a), where the subject is a single plural-masculine noun.
This is easily understood if the node that dominates the two conjuncts in
8. Actually accounting for these syncretisms is complicated by the fact that sin-
gular feminine nouns in the dative/ genitive case are also syncretic with the
feminine-plural forms.
[ 98 ] Background
(20b) inherits the shared gender feature(s) of the conjoined nouns and adds
the number feature plural.
By the same token, the agreement marking of the participle in (21b) shows
that the two singular nouns in (21b) must have the same gender features as
the plural noun in (21a), that is, neuter rather than masculine:
(21) a. Nişte deget-e au fost amputat-e

some finger-n.pl have been amputated-n.pl
‘Some fingers have been amputated.’
b. Un deget şi un braţ au fost amputat-e
a.n.sg finger and a.n.sg arm have been amputated-n.pl
‘A finger and an arm have been amputated.’
But if we run the same test on Xhosa singular–plural pairs like u-m-Xhosa /
a-ma-Xhosa and i-n-doda ‘man’ /a-ma-doda ‘men,’ we obtain a different result,
as seen in (22) and (23):
(22) a. A-ma-Xhosa a-sebenza ndawonye

6-6-Xhosa 6-work together
‘The Xhosas are working together.’
b. U-m-Xhosa no-m-Zulu ba-sebenza ndawonye
1-1-Xhosa and.1-1-Zulu 2-work together
‘A Xhosa and a Zulu are working together.’
(23) a. A-ma-doda a-sa-cula

6-6-man 6-still-sing
‘The men are still singing.’
b. I-n-doda ne-n-kwenkwe zi-sa cula
9-9-man and.9-9-youngster 10-still-sing
‘The man and the youngster are still singing.’
By parity of reasoning, we therefore conclude that the gender of singular

nouns in (22b) and (23b) is not the same as the gender of the plural nouns in
(22a) and (23a), respectively. Rather, the singular nouns in (22b) must simply
have the same gender I as ordinary class 1 nouns (as the plural prefix comes
out as regular ba-), and the singular nouns in (23b) must have the same gender
V as ordinary class 9 nouns [as the plural prefix comes out as regular zi(N)-]
(see Tables 3.1 and 3.2).
An immediate consequence of this is that the gender feature(s) deter-
mining the shape of the plural prefix in a-ma-Xhosa ‘Xhosas’ and a-ma-doda
‘men’ cannot originate from the noun following the prefix. Otherwise, we
100
[NumlP [NumP Pl [ClP Cl [NP N1 ]]] Pl [ClP <NumP> Cl [NP N2]]]
prefix
Figure 3.6 Plural class prefix including a lexical noun
[NumlP [NumP Pl [ClP Clx [NP Nx ]]] Pl [ClP <NumP> Cl [NP N2 ]]]
prefix
Figure 3.7 Gender coming from N1
[NumlP [NumP Pl [ClP Cl6 [NP N6 ]]] Pl [ClP <NumP> 6 [NP Xhosa ]]]
ma-
Figure 3.8 a-ma-Xhosa
[NumlP [NumP Pl [ClP Cl6 [NP N6 ]]] Pl [ClP <NumP> V [NP doda ]]]
ma-
Figure 3.9 a-ma-doda
should have *a-ba-Xhosa (ba- = {pl, I}) and *ii-n-doda (ii-/i -zi- = {pl, V}). So
where does the gender feature come from in these irregular plural forms?
The structure proposed in Figure 3.4 above, repeated here as Figure 3.6,
provides an answer.
The gender features that condition the spellout of the prefix are those that
are associated with the head Cl inside the specifier. Because this Cl is merged
on top of N1, the gender associated with it should come from N1 rather than
from N2. In Figure 3.7, the subscript is used to represent this relation, that is,
x ranges over I, II, and so forth.
If we adopt the analysis in fn. 6 of this chapter, we can represent a-ma-
Xhosa ‘Xhosas’ and a-ma-doda ‘men’ as in Figures 3.8 and 3.9.9
Thus, the properties of the irregular singular–plural pairs in Xhosa provide
evidence for the structure in Figure 3.4.
9. I assume that the class 1 prefix m lexicalizes the specifier in a structure like
Figure 3.5 with Pl replaced with Sg. In general, the gender feature determining the
spellout of the class prefix comes from the N1 in Figure 3.4 in all classes. The question
why ba, the plural form of m, is unavailable with nouns like Xhosa and Zulu is left
open here.
[ 100 ] Background
3.2.5 Primary pref ixes and secondary pref ixes
At this point, it is natural to wonder what kind of semantic contribution the

N inside the specifier is making in Figure 3.4. A natural suggestion is that it
acts as a classifier with respect to the second N. In this subsection and the
next, I want to discuss some properties of so-called secondary prefixes that
ultimately seem to support this claim.
In descriptive grammars, a primary prefix just reflects a noun’s class mem-
bership and does not affect what is taken to be its basic meaning, whereas a
secondary prefix changes the noun’s basic meaning in addition to bringing it
into a different noun class, for example, with respect to agreement. By this
reasoning, the class 1 prefix m- is taken to be a primary prefix in u-m-ntu
‘person’ (in Xhosa), whereas the class 14 prefix bu- in u-bu-ntu ‘humanity’ is
considered a secondary prefix. In other words, primary prefixes resemble in-
flectional affixes, whereas secondary prefixes resemble derivational affixes.
Seen in this perspective, it is surprising to find that a secondary prefix
may be added outside a primary prefix. For example, the Manyika counter-
part to Xhosa u-bu-ntu ‘humanity’ is vu-mu-nhu, in which vu- is a secondary
prefix corresponding to Xhosa bu- and mu- is a primary prefix corresponding
to Xhosa m-.
To make this pattern consistent with the general fact that inflectional mor-
phology goes outside derivational morphology, it has been proposed that a
secondary prefix is actually the primary prefix of a silent noun that effects a
change of the basic meaning of the host noun. From this point of view, the
structure underlying vu-in Manyika vu-mu-nhu is like the specifier in Figure 3.4.
This is illustrated in Figure 3.10.
Another striking fact is that a secondary prefix generally does double duty
as the primary prefix for some noun class. For example, the bu- of u-bu-ntu
‘humanity’ is also the primary prefix for some class 14 nouns (see Table 3.1)
that never occur with any other prefix. This is captured in Figure 3.10 by saying
that the noun lexicalized by vu-happens to be in class 14.
[NumP Sg [Cl 14 [NP N14 ]]]
vu-
Figure 3.10 Secondary prefix vu-(Manyika)
However, at this juncture an issue arises if we play by the CL rules as laid out
in Section 1.3, where we found that the definition of matching in (3) entails
the Foot Condition so that a morpheme with the lexical entry in (24) can
lexicalize only constituents containing the exact N14 that occurs in (24):
(24) vu ⇔ [NumP Singular [ClP 14 [NP N14]]]
102
This means that we can capture the identity between secondary bu-/vu- and
primary bu-/vu- only by assuming that this prefix lexicalizes a specifier built
on the same N14 in both cases. So even in u-bu-sika ‘winter’ (Xhosa), a basic
class 14 noun, bu- must lexicalize a specifier containing the same N14 as in
u-bu-ntu ‘humanity,’ as shown in Figure 3.11.
[NumP [NumP Sg [ClP 14 [NP N14 ]]] Sg [ClP <NumP>14 [NP sika ]]]
bu-
Figure 3.11 Primary and secondary prefix bu-(Xhosa)
More generally, all nouns with a primary prefix that also occurs as a sec-
ondary prefix must have a structure conforming to Figure 3.4, with Singular
replacing Plural. See Figure 3.12.
[NumlP [NumP Sg [ClP Cl [NP Nx ]]] Sg [ClP <NumP> Cl [NP Ny ]]]
prefix
Figure 3.12 Structure of noun whose primary prefix also occurs as a secondary prefix
3.2.6 Pref ixes as classif iers
The conclusion just reached obviously begs the question why the N inside the
prefix affects the meaning of the noun following the prefix when the prefix is
used as a secondary prefix, but not when it is used as a primary prefix. This
brings us back to the idea that the Ns inside class prefixes act as classifiers.
Suppose in particular that the meaning of an N inside the prefix corresponds
to a set of properties that things may or may not be endowed with. If the
prefix attaches to a noun denoting things that are conventionally classified as
having those properties, the prefix acts as a primary class prefix. If not, it acts
as a secondary prefix adding a shade of meaning to the host noun.
Instead of trying to present anything like a detailed implementation of the
general idea, I present an observation that suggests that it may be essentially
correct. In Shona, as described by Fortune (1955), the prefix RI- is both the
primary prefix of class 5 nouns and a secondary “augmentative–derogatory”
prefix.10 An example of each kind is given in (25).11
10. RI- is the abstract representation of a prefix whose presence is detectable only

by the way it modifies the initial segment of the following noun, e.g. banga ‘knife’ vs.
ma-panga ‘knives.’
11. The class 9 prefix N- is also detected only by its effect on the following segment,
e.g. mhuka (class 9) ‘beast’ vs. buka (class 5) ‘big beast.’
[ 102 ] Background
(25) a. banga = RI-panga ‘knife’ (class 5, primary RI-)
b. mhuka = N-puka ‘beast’ versus buka = RI-puka ‘big, ugly beast’
(secondary RI-)
The question now is if we can substitute secondary RI-for primary RI-. Asking
whether banga can mean ‘big, ugly knife’ provides no answer to this question,
because banga with primary RI- denotes knives, including the big, ugly ones.
But Shona has a second augmentative zi- that, according to Fortune (1955),
stacks on top of secondary RI- to produce a “second-degree” augmentative
reading, as seen in (26):
(26) a. buka = RI-puka ‘big, ugly beast’

b. zi-buka = zi-RI-puka ‘very big beast’
So if we could substitute secondary RI- for primary RI-, we would be able to

derive a form zi-banga = zi-RI-panga that should unambiguously mean ‘very
big knife’ (with, hypothetically, both the first-degree augmentative meaning
‘big’ coming from secondary RI- and the second- degree augmentative
meaning ‘very’ from zi-). But Fortune notes that when zi- attaches to a class
5 noun, the result is only a first-degree augmentative, for example, zi-banga
means just ‘big knife.’ So it seems that secondary RI- cannot be substituted
for primary RI-.
This is an automatic consequence of the analysis pursued here, which
does not allow for the existence of a secondary RI- distinct from the pri-
mary RI-. On this analysis RI- must lexicalize the structure in (27) in both
cases:
(27) [NumP Sg [ClP 5 [NP N5]]]
N5 is associated with a set of properties relating to appearance that are con-

ventionally attributed to class 5 nouns, but not to nouns in other classes.
3.3 BROADENING THE PERSPECTIVE
I have argued that the properties of the noun class prefixes in Southern Bantu
converge on vindicating the prediction made for portmanteau prefixes by
CL. I now look at some observations suggesting that the prediction may hold
more widely and also outline how portmanteau suffixes may differ from port-
manteau prefixes.
104
3.3.1 Nominal portmanteau pref ixes in other languages
The basic tenet of the CL approach demands that the analysis I have given for
nominal portmanteau prefixes in Southern Bantu be extended to nominal port-
manteau prefixes in any language that has them. This is because a structure like
(28), where Num and Cl form a constituent (modulo traces), cannot be derived
without movement of a constituent not containing a noun, in contradiction to
what Cinque’s (2005) account of Greenberg’s Universal 20 seems to require.
(28) [[NumP Num [ClP Cl <NP> ]] [ NP [ < NumP> ]]]
In other words, basic assumptions lead to the conclusion that a nom-

inal portmanteau prefix always corresponds to a specifier including a
classifier-like noun.
I am not in a position to assess whether this prediction is borne out in
general, but a claim made by Kihm (2001) about the nominal class prefixes
of Manjaku (an African language outside the Bantu group) is suggestive.
According to Kihm, a class prefix in Manjaku is inserted under “little n” as “a
root, i.e. a member of the lexicon, and that to the difference of all other roots,
except v, it is endowed with a category, namely N” adding that “n is a label for
the set of prototypical nouns, or protonouns, each having its own meaning
and exponent” (Kihm 2001, 478). This essentially amounts to saying that a
class prefix contains a noun drawn from a small set of nouns in Manjaku as
well. But much work will be needed to see whether this claim can be extended
to other languages as well.
3.3.2 Pref ixes versus suff ixes
By contrast, the basic principles I have adopted are compatible with two
different structures for nominal class suffixes. Given the assumption that the
lexicalization procedure ignores traces for the purpose of determining con-
stituency, a portmanteau suffix can arise from the structure in Figure 3.13
derived by NP movement across Num and Cl.
For instance, the derived structure in Figure 3.13 is likely to be the right
structure for nouns in Standard Italian, which has the portmanteau morpheme
[ NumP NP [ Num' Num [ ClP Cl <NP> ]]]
suffix
Figure 3.13 Num and Cl as a constituent following NP movement
[ 104 ] Background
-e in plural feminine nouns.12 In varieties like the Colonnata dialect (Manzini
and Savoia 2005), which lacks the portmanteau -e, the structure in Figure 3.13
becomes visible in feminine-plural nouns like dun-y-a ‘woman-pl-fem’ corre-
sponding to Standard Italian donn-e.
In languages that work this way, a class suffix will not contain a classifier-
like noun (one of Kihm’s “protonouns”). In this case, the expectation is that
one will not see any ‘gender change’ between singular and plural forms ex-
cept to the extent that the language has something like the Romanian neuter
discussed in Section 3.2.3.
However, a suffix might also correspond to a specifier similar to the Bantu
class prefixes. In this case, NP has moved across a specifier containing Num
and Cl, as shown in (29):
(29) [XP NP [NumP [NumP Num [ClP Cl [NP N]]] [ClP <NumP> Cl <NP> ]]]
Discussing plural suffixes in Somali, Lecarme (2002, 28) notes: “Somali plural

suffixes, in contrast, can be understood as number categories intermediate
between pure (underspecified) functional categories and plural words.” It thus
seems reasonable to assume that the evidence at Lecarme’s disposal supports
analyzing the Somali plural suffixes as in (29).
In general, the approach to lexicalization taken here predicts that port-
manteau suffixes may or may not show the behavior of an affix containing
a classifier-like noun, but portmanteau prefixes should always reveal the
presence of a classifying noun within them.
3.3.3 Syntax and phonology
Phonologists have often claimed that “some or all prefixes are phonolog-
ical words, e.g. in Germanic, whereas suffixes are not.” (Hyman 2005, 15) As
regards the portmanteau prefixes, CL provides a syntactic basis for this dis-
tinction. Portmanteau prefixes must be phrases in a specifier position, but
portmanteau suffixes need not be. However, the considerations brought to
bear in Section 3.3.2 also lead us to expect that suffixes may also be phonolog-
ical words in some languages, if indeed a phrasal specifier is always mapped
onto a phonological word as is expected if each separate completed extended
projection corresponds to a phonological word. If this expectation is not borne
out, the claim referred to by Hyman will not provide any evidence for the
conclusions I have reached. Again, more empirical work is clearly called for.
12. This contrasts with agglutinating Spanish -a-s, [NumP [ClP NP -a] -s], which, like
(17b), is the outcome of a roll-up derivation.
106
One also notes the hedging “some or all prefixes” in the quote from Hyman.
If indeed not all prefixes behave as phonological words, there are basically two
possibilities. If the prefixes that are not phonological words are also not port-
manteau prefixes, we actually have an even stronger argument for the CL ana
lysis which requires portmanteau prefixes to correspond to phrasal specifiers,
but makes no such claim about prefixes in general. If the prefixes that are not
phonological words include portmanteau prefixes in some language, then we
have to conclude that the correlation between being a phrasal specifier and
being a phonological word is less tight than one might want. However, we
should also notice in this connection that the notion of “phonological word”
may need to be sharpened and relativized to different types of phonological
processes.
On the other hand, if in fact it turns out that all prefixes are phonological
words in all languages, we face two different possibilities. Again, it may be that
phonological wordhood does not depend directly on syntactic phrasehood. Or
all prefixes must be phrasal specifiers.
The second possibility will be consistent with a more radical version of CL
based on (30) (originally suggested by Starke, p.c.):
(30) Lexical insertion targets only phrasal nodes (as opposed to heads).
If (30) holds, a morpheme M with the entry M ⇔ X cannot replace X in (31a),

although it can in (31b) (applying to X′), assuming that the lexicalization pro-
cedure disregards traces:
(31) a. [XP X [YP Y]]

b. [XP [YP Y ] [X′ X <YP>]]
As far as I can tell, (30) would be consistent with the various specific analyses
proposed in this article except for the account of forms like dun-y-a ‘women’
in Colonnata Italian (because the plural marker -y would correspond to a head
in Figure 3.13).
3.4 CONCLUSION
I began by discussing the relative merits of CL and spanning in a general way.

In Section 3.2, I developed a specific empirical argument for CL based on
an examination of nominal class prefixes in Southern Bantu. The argument
flows from the observation that CL, but not spanning, predicts the special
properties that these prefixes were shown to have in Section 3.2.
Obviously, the strength of this argument depends on whether the relevant
properties of the Bantu prefixes are shared by portmanteau prefixes in other
[ 106 ] Background
languages. In Section 3.3, I reported on findings that suggest that they are,
but also pointed out that much further empirical work is needed.
The interim conclusion is that CL is not only conceptually superior to
spanning, but is also empirically superior.
REFERENCES
Abels, Klaus and Peter Muriungi. 2008. “The Focus Particle in Kîîtharaka: Syntax and
Semantics.” Lingua 118: pp. 687–731.
Acquaviva, Paolo. 2008. Lexical Plurals. New York: Oxford University Press.
Carstens, Vicki. 1991. The Morphology and Syntax of Determiner Phrases in Kiswahili.
Doctoral dissertation, UCLA.
Embick, David. 2012. “On the Targets of Phonological Realization.” Talk given at the
MSPI Workshop, Stanford University, October 13, 2012.
Fortune, George. 1955. An Analytical Grammar of Shona. London: Longmans, Green & Co.
Haugen, Jason D. and David Siddiqi. 2013. “On Double Marking and Containment in
Realization Theory.” Ms., Oberlin College and Carleton University.
Hyman, Larry. 2005. “Directional Asymmetries in the Morphology and Phonology of
Words, with Special Reference to Bantu.” Ms., published in 2008 in Linguistics
46 (2): pp. 309–350.
Johnson, Kyle. 2004. “In Search of the English Middle Field.” Ms., University of
Massachusetts Amherst.
Kihm, Alain. 2001. “Noun Class, Gender and the Lexicon- Syntax-
Morphology
Interfaces: A Comparative Study of Niger-Congo and Romance Languages.” Ms.,
CNRS, Laboratoire de linguistique formelle, Université de Paris 7.
Lecarme, Jacqueline. 2002. “Gender ‘Polarity’: Theoretical Aspects of Somali Nominal
Morphology.” In Many Morphologies, edited by Paul Boucher and Marc Plénat,
pp. 109–141. Somerville, MA: Cascadilla Press.
Manzini, Maria Rita and Leonardo Savoia. 2005. I dialetti italiani e romanci. Alessandria,
Italy: Edizioni dell’Orso.
Merchant, Jason. 2015. “How Much Context is Enough? Two Cases of Span-
Conditioned Stem Allomorphy.” Linguistic Inquiry 46 (2): pp. 273–303.
Pantcheva, Marina P. 2011. Decomposing Path: The Nanosyntax of Directional Expressions.
Starke, Michal. 2009. “A Short Primer to a New Approach to Language.” Nordlyd 36 (1):,
pp. 1–6. [online] Available at http://septentrio.uit.no/index.php/nordlyd/index
Svenonius, Peter. 2012. “Spanning.” Ms., University of Tromsø.
Taraldsen, Knut T. 2010. “The Nanosyntax of Nguni Noun Class Prefixes and Concords.”
Lingua 120: pp. 1522–1548.
Williams, Edwin. 2003. Representation Theory. Cambridge, MA: MIT Press.
108
PART II
Empirical Investigations
110
CHAPTER 4
A Note on Kim’s Korean Question

Particles Seen as Pronouns
MICHAL STARKE
K orean marks clauses as interrogatives with a special marker at the end of

the clause, for example, (1) versus (2a).1
(1) ni-
ka ppang- ul mek- ta
ess-
you-nom bread-acc eat-past- decl
‘You ate the bread.’
(2) a. ni-
ka ppang- ul mek- ni
ess-
you-nom bread-acc eat-past- Q
‘Did you eat the bread?’

b. chelswu-nun cikum-to tambay-pi- nya
Chelswu-top now-even cigarette-smoke- Q
‘Does Chelswu still smoke?’
c. chelswu-nun cikum-to tambay-pi- na
Chelswu-top now-even cigarette-smoke- Q

d. chelswu-nun cikum-to tambay-pi-nun ka
Chelswu-top now-even cigarette-smoke-nun Q
roughly: ‘Does Chelswu still smoke, I wonder . . . ’
As illustrated in (2), there is a variety of such particles (with subtly different

semantics). Kim (2011) observes that the four question particles in (2) are
1. All data are from Kim (2011).

112
identical to pronouns: Not only do they look like pronouns but their semantic
interpretation also corresponds to the interpretation of their pronominal
counterpart (see Section 4.1). Accordingly, Kim makes the elegant proposal
that Korean question particles are in fact pronouns that refer to the addressee
of the question. Kim’s implementation of this proposal, however, comes at a
serious cost:
• It loses the fact that those particles are specifically question particles, and
not, for example, declarative particles. As I illustrate in Section 4.1, this is
not a mere accident of Kim’s proposal, rather it looks like one side of the
coin is necessarily lost: Either the correlation with pronouns is captured, or
the interrogativity of the particles is captured, but not both.
• It cannot express the fact that in some dialects, some question particles
diverge from pronouns.
The point of this short chapter is to show that we can keep the beauty of Kim’s
proposal and solve both problems, once we reason in terms of phrasal spellout
and its Superset Theorem (Starke 2002, 2009; Caha 2009, etc.).
4.1 THE COARSE SEMANTIC PROPERTIES OF THE PARTICLES
The four questions in (2) have different semantics, along roughly two
dimensions. First, some of the questions are full-blown questions addressed
to the interlocutor, whereas others express questions that are not meant to
be answered by the interlocutor—somewhat akin to ‘I wonder . . . ’ contexts.
Second, those particles that address the interlocutor necessarily come with a
particular register as interlocutor-addressed speech in Korean is coded for the
relative social status of the interlocutors, speech context, and so forth. A first
partial characterization of the four particles is as follows:
(3) a. ni + hearer-oriented low level

b. nya + hearer-oriented low level
c. na – hearer-oriented
d. ka – hearer-oriented
Kim’s major observation is that those morphemes correspond to independ-

ently existing Korean pronouns, with matching properties:
(4) a. ni 2nd person pronoun low level

b. na 1st person pronoun
c. ka 3rd person pronoun
[ 112 ] Empirical Investigations

The missing nya particle corresponds to the composition of ni and the voc-
ative particle ya, that is, ni + ya > nya. As is obvious from (3) and (4), the
question particles correspond to pronouns both phonologically and semanti-
cally: Exactly those questions that are hearer-oriented correspond to hearer-
oriented pronouns, that is, second persons, also matching their speech level.
Kim also documents that nya does indeed inherit the semantics of vocative,
because of the presence of the vocative morpheme ya and that na and ka differ
in the predicted direction given their differing pronominal roots.
So why do question particles look so much like pronouns? This is what is
expected under the performative hypothesis (Ross 1970), according to which
every question is underlyingly of the form “I ask ADDRESSEE whether p,” if
the question particles are in fact pronouns identifying the addressee. Kim’s
solution is thus that (i) the performative hypothesis is syntactically correct;
questions have a syntactic addressee slot, and (ii) the so-called question
particles in (3) fill that slot. This is a very interesting approach to question
particles.
4.2 THE DILEMMA OF IDENTIFYING PARTICLES

WITH PRONOUNS AND ITS SOLUTIONS
One problem with this solution is that it does not actually capture the
connection between pronouns and question particles. What it captures is a
wider—and inaccurate—generalization: Pronouns will occur as sentence-
final particles in any clause type containing an addressee, interrogative or
not. This is because sentences other than questions also have an addressee
slot, and Kim’s approach predicts pronouns to fill those slots too. Both in
Ross’s (1970) performative approach and in its modern adaptations such as
Truckenbrodt (2006), declaratives also have an addressee slot (roughly, “I tell
ADDRESSEE that p”), and hence if pronouns simply fill the addressee slot,
they are predicted to occur in declaratives as well as interrogatives. Which is
not the case. (Mutatis mutandis for imperatives, etc.)
The problem cannot be solved in the obvious way: We could try to encode
the fact that pronouns occur as sentence-final particles only in interrogative
clauses by adding (the counterpart of) an interrogative feature to them, but
in doing so we would lose the identity between a pronoun (no interrogative
feature) and a sentence-final particle homophonous with a pronoun (with an
interrogative feature). So at first sight, we cannot have it both ways.
But we can. What we want to express is that there is a single lexical entry
covering both pronouns and question markers, and that questions markers
are a superset of pronouns, that is, pronouns + interrogative features. The
Superset Theorem in nanosyntax (Starke 2002, 2009; Caha 2009, etc.; see
Chapter 1 of this volume) yields just that. The superset effect derives from the
K i m ’ s K or e a n Q u e s t i o n Pa r t i cl e s [ 113 ]
114
fact that a syntactic structure can be spelled out by a lexical item if the syn-
tactic structure is contained in the lexical entry for that item (and hence if the
lexical item is a superset of the syntactic material). Thus, if the syntactic lexical
entry for the question particles is the syntactic equivalent of [Q [pronoun]],
it will match both syntactic contexts: qua question particle it will spell out [Q
[pronoun]] (and not just [pronoun], contra Kim); qua pronoun, it will spell
out the syntactic constituent [pronoun]. This resolves the dilemma: There is a
single lexical entry, thus capturing the fact that the question particle and the
pronoun have the same phonological shape, but that lexical entry matches
two distinct syntactic structures, capturing the fact that the entry is restricted
to typing questions.2
4.3 MISMATCHES BETWEEN PRONOUNS
AND QUESTION PARTICLES
Apart from resolving the dilemma in Section 4.2, this solution has the ad-
vantage of explaining cases in which the match between the question par-
ticle and the pronoun is not perfect. Such cases are attested: For instance, in
some dialects of Korean the question particle ka does not correspond to an ex-
isting productive pronoun. In some of these dialects the third person pronoun
alternates between ka and kyay, in others ka has become archaic or formal and
kyay is standard. But in none of these dialects does the question particle be-
come kyay; it always remains ka. Hence within those grammars, the question
particle cannot be equated with the pronoun.
Phrasal spellout and its Superset Theorem predict just that: In these dialects,
the lexical entry for ka is the same as proposed in 4.2, namely [Q [pronoun]];
but contrary to what we discussed in 4.2., this entry now competes with an-
other lexical entry [pronoun] for kyay (again skipping the internal structure
of the pronoun, irrelevant for the present purposes). In these grammars, a
question particle addressed to the hearer works as before: ka is the only candi-
date to spell it out because it is the only lexical item to contain Q. When syntax
builds a second person pronoun, things come out differently: Because of the
Superset Theorem, there are now two candidates to spell out this [pronoun],
ka and kyay. At this point, the Elsewhere Principle comes into play and the
2. The same effect might be achieved by means of the Subset Principle of Distributed
Morphology, depending on the lexical entries for the other 50 or so Korean particles.
The lexical entry for the particle would be the pronoun, without any interrogative fea-
ture, and it would have to be the best match for the combination of Q and pronoun, but
not the best match for declarative + pronoun, etc. All the other particles would thus
have to be more specific and block the pronouns. It is unclear whether this is plausible;
see for instance Kim’s discussion of the clausal particle kka. That difficulty does not
arise under the Superset formulation.

more specific kyay = <+ pronoun> wins the insertion race (see Starke 2002,
2009; Caha 2009 for details; see also Chapter 1 of this volume). Hence in these
dialects, a pronoun will come out as kyay and not as ka, despite the lexical
entry of ka being [Q [pronoun]].
These mismatches come out for free, in fact they are predicted by the
approach: When a language has a lexical item for [Q [pronoun]], it can choose
to add another entry for [pronoun]—or not. The general rules of spellout will
do the rest. Our rephrasing of Kim’s analysis in terms of phrasal spellout thus
allows us to preserve the elegance of his proposal—interrogative particles are
none other than the lexical entry for pronouns—and at the same time capture
the fact that these pronouns are restricted to interrogative clause typing and
can drift away into separate lexical entries.
REFERENCES
Kim, Chonghyuck. 2011. “Korean Question Particles Are Pronominals: A Transparent
Case of Representing Discourse Participants in the Syntax.” [online] Available at
http://ling.auf.net/lingBuzz/001157.
Ross, John R. 1970. “On Declarative Sentences.” In Readings in English Transformational
Grammar, edited by Roderick A. Jacobs and Peter S. Rosenbaum, pp. 222–272.
Washington, DC: Georgetown University Press.
Starke, Michal. 2002. “The Day Syntax Ate Morphology.” Class taught at the EGG
summer school, Novi Sad.
nordlyd/index
Truckenbrodt, Hubert. 2006. “On the Semantic Motivation of Syntactic Verb
Movement to C in German.” Theoretical Linguistics 32 (3): pp. 257–306.
K i m ’ s K or e a n Q u e s t i o n Pa r t i cl e s [ 115 ]
116
CHAPTER 5
Syncretism and Containment

in Spatial Deixis
ERIC L ANDER AND LILIANE HAEGEMAN
5.1 INTRODUCTION
According to Diessel (1999, 6), demonstratives have four main uses, which are
summarized in Figure 5.1 and illustrated for English in (1). For discussion see
also, among others, Fillmore (1971, 1982, 1997), Lyons (1977), Levinson (1983),
Himmelmann (1997, 2001), and Diessel (1999).
(1) a. That book in the window over there is wonderfully illustrated.
(exophoric)
b. My friend’s dog is so friendly. I’ll dogsit that little guy any time.
(anaphoric)
c. . . . and they lived happily ever after. That was the end of our fairy
tale. (discourse-deictic)
d. Have you heard about that terrible measles epidemic? (recognitional)
This chapter1 focuses only on the (gestural) exophoric demonstrative use

(Diessel 1999, 94), that is, the use of demonstratives to point at objects in the
physical environment of the speech participants, with an indication of relative
1. This chapter is a modified and abbreviated version of Lander and Haegeman
(2016). We make use of the same data here; the discussion of the data is also closely
based on the earlier publication. This work has its roots in a nanosyntax weblab held
in November and December of 2012. Special thanks to Michal Starke, Sebastian
Bican-Miclescu, and Bartosz Wiland for discussion of the data. We are grateful to
the members of GIST (Generative Initiatives in Syntactic Theory) for comments on
Pragmatic uses
Exophoric Endophoric
Anaphoric Recognitional
Discourse-deictic
Figure 5.1 Uses of demonstratives
distance from speaker, hearer, or both. Of the four demonstrative functions

in Figure 5.1, the exophoric use is normally considered the most basic or con-
crete one, because endophoric uses are more anchored in the discourse itself,
as opposed to the physical speech situation.2 We use the general term spatial
deixis for the (gestural) exophoric use.
The purpose of this chapter is to investigate spatial-deictic systems in
a wide range of languages from the perspective of two morphological phe-
nomena. In Section 5.2 we discuss the semantic distinctions that have
emerged from the literature. We systematize these and propose a hierarchical
organization that receives support from the observed syncretism patterns.
Section 5.3 delves deeper into the data and the proposed hierarchical classifi-
cation by investigating the phenomenon of morphological containment, for
which we discover crosslinguistically regular patterns clearly showing that
certain components of the spatial-deictic system are structurally larger than
others. In Section 5.4 we embed our findings in a nanosyntactic model: We
show that the patterns observed in Sections 5.2 and 5.3 can be understood in
terms of a simple syntactic structure consisting of three syntactico-semantic
heads (Dx1, Dx2, and Dx3) in a unique merge order. Section 5.5 concludes the
chapter.
a previous version of this talk (May 7, 2014). Thanks to Lena Baunaz and Frédérique
Berthelot for help with French. Thanks also to two anonymous OUP reviewers for
helpful comments. Abbreviations include dem.pro = pronominal demonstrative, dem.
adn = adnominal demonstrative, n(eut) = neuter, f(em) = feminine, m = masculine,
abs = absolutive, sg = singular, pl = plural, du = dual, an = animate, inan = inani-
mate, IE = Indo-European, coll. = colloquial. For the languages discussed we provide
the family in brackets, e.g. Ewe [Niger–Congo]. Eric Lander’s research is supported by
BOF (Bijzonder Onderzoeksfonds) grant 01D30311 awarded by UGent and Odysseus
grant G091409 awarded by FWO–Flanders (Fonds Wetenschappelijk Onderzoek–
Vlaanderen). Liliane Haegeman’s research is funded by Odysseus grant G091409
awarded by FWO–Flanders.
2. For the exophoric–endophoric distinction, Meeuwis and Stroken (2012) prefer the
terms situational and nonsituational, respectively. See also Rauh (1983).
S y n c r e t i s m a n d C o n ta i n m e n t i n S pat i a l De i x i s [ 117 ]
118
5.2 ORGANIZING THE DATA
5.2.1 Crosslinguistic attestation
From the literature a consensus emerges that the encoding of spatial deixis in
natural language encodes relative distance (‘close to,’ ‘far from’) of a referent
and that the distance of the relevant referent is evaluated from a deictic center
(or anchor or origo). Fillmore (1982) and Anderson and Keenan (1985) distin-
guish “distance-oriented” and “person-oriented” systems of spatial deixis. In
distance-oriented systems, only the speaker serves as the deictic center. In
person-oriented systems, in addition to the speaker, nonspeakers such as the
hearer can also serve as the deictic center. According to Diessel (1999, 50) the
two systems differ in terms of the encoding of distance: Distance-oriented sys-
tems have at most a three-way system of distance contrasts, whereas person-
oriented systems may have four (or more) contrasts. As we will see in this
section, there are in fact reasons to be skeptical about systems with more than
three contrasts. More generally, we believe that the distance-oriented versus
person-oriented distinction is a false dichotomy. One natural way to account
for person-oriented deictic systems might be to bundle person–participant
features with distance features. Fortis and Fagard (2010, 10–11), for in-
stance, identify the following crosslinguistically possible anchors: Speaker (S),
Hearer (H), Third Person (Th), and both S and H together.3 On the basis of the
combinations of distance oppositions and person features, there are then at
least twelve different possible combinations (and hence readings) available. In
(2) we provide the relevant feature systems; Table 5.1 shows all of the possible
combinations.
(2) a. Distance features

distal
medial
proximal
b. Person features
[1]‌ = S
[2]‌ = H
[3]‌ = Th
[1]‌+ [2] = S + H
3. Imai (2003, section 2.3) lists three additional anchors: Participant, Non-
participant, and Object. Although we are not in principle denying the possibility of
the existence of such anchors, it remains true that they are very rare in his sample and
marked by idiosyncratic properties. For instance, the Huallaga Quechua morpheme qa
indicating an Object anchor is optional and “should be excluded from a paradigm of
deictics” (Imai 2003, 70, n.5). Indeed, similar issues subsequently arise for the Third
Person anchor, casting doubt on the validity of that particular anchor.

Table 5.1 READINGS RESULTING FROM THE FEATURE COMBINATIONS IN (2)
Feature Bundles Interpretation Label
a. ?
Dist + S ‘far from S’ S-based distal
b. ?
Med + S ‘medium distance from S’ S-based medial
c. Prox + S ‘close to S’ S-based proximal (= Proximal)
d. *Dist + H ‘far from H’ H-based distal
e. *Med + H ‘medium distance from H’ H-based medial
f. Prox + H ‘close to H’ H-based proximal (= Medial)
g. *Dist + Th ‘far from Th’ Th-based distal
h. *Med + Th ‘medium distance from Th’ Th-based medial
i. ??Prox + Th ‘close to Th’ Th-based proximal
j. Dist + Inclusive ‘far from S and H’ Incl-based distal (= Distal)
k. *Med + Inclusive ‘medium distance from S and H’ Incl-based medial
l. ?
Prox + Inclusive ‘close to S and H’ Incl-based proximal
Table 5.2 THREE AT TESTED SPATIAL-D EICTIC READINGS
Interpretation Label
c. ‘close to S’ S-based proximal (= Proximal)

f. ‘close to H’ H-based proximal (= Medial)
j. ‘far from S and H’ Incl-based distal (= Distal)
Based on the empirical descriptions available in the literature, this system

clearly overgenerates. In the descriptions of spatial deixis we have examined,
none of the readings signaled by an asterisk in Table 5.1 have been found so far.
The readings accompanied by question marks in Table 5.1 are reported in the
literature but their status is dubious for reasons to be discussed presently. Out
of the twelve possible readings generated by the combination of person features
with distance contrasts, only three are undoubtedly attested as described in the
literature. These three core readings are given in Table 5.2.
From the discrepancy between Tables 5.1 and 5.2 we conclude that it is un-
likely that encoding of spatial deixis results from the combination of person
features with distance features, and we do not pursue an approach in which
person features are involved in the expression of spatial deixis.4
4. The basic idea that person features are at the core of spatial-deictic contrasts
can be found in Leu (2015, section 2.7.2) and Harbour (2016), among others. We
do not approach spatial deixis from this angle because, as discussed in this section,
when person features are combined with distance contrasts in this way, the system
overgenerates. There are other reasons to keep the domains of person and deixis clearly
120
Let us now consider the (non)attestation of the readings in Table 5.1 in

more detail. Starting with the readings involving S (a–c), it is rather uncon-
troversial to say that crosslinguistically S is universally the primary anchor of
spatial deixis (Imai 2003). The spatial-deictic systems of many languages are
indeed described as being purely S-based, with proximal, medial, and distal
readings with respect to S. There is some reason to doubt the empirical va-
lidity of these descriptions, however. We illustrate this with two examples.
First, consider the case of Venda [Niger–Congo], which has been claimed to
have a four-way distance-oriented system: ‘immediately next to speaker,’ ‘rel-
atively close to the speaker,’ ‘farther away from the speaker,’ and ‘relatively
remote from both the speaker and the person addressed’ (Poulos 1990, 107,
cited in Imai 2003, 97). However, on the basis of native speaker elicitation,
Imai (2003, section 4.3.3) concludes that Venda in fact shows only a three-
way system and that H is the anchor in the medial form; the fourth item
‘immediately next to speaker’ is more accurately identified as an item for em-
phasis or givenness (Imai 2003, 99). Similarly, Malagasy [Austronesian] has
been claimed to have as many as seven S-based distance contrasts (Anderson
and Keenan 1985). Again on the basis of experiments, Imai (2003, section
4.3.2) demonstrates that Malagasy actually has a three-way system, which
also involves the H anchor in parts of the system. The impression of a very
rich distance-oriented system can be attributed to additional parameters out-
side of distance (i.e. boundedness and visibility, as well as the existence of a
separated. For instance, Starke (2013) and Vanden Wyngaerd (Chapter 11 in this book)
find that the person fseq is [1 [2 [3]‌]]. If this fseq were also responsible for spatial
deixis, then it would mean that Prox is larger than Med and Med is larger than Dist,
i.e. [Prox [Med [Dist]]]. However, this is in fact the exact opposite of our findings in
Section 5.3, where we firmly establish the containment relation [Dist [Med [Prox]]]. In
other words, the person–distance hypothesis for spatial deixis not only overgenerates,
but it also appears to clash with the containment facts.
A second reason we think it is unwise to reduce spatial-deictic distinctions to person
features is that certain predictions made by such a hypothesis appear not to be con-
firmed. For instance, if person and deixis are actually the same, then we would expect
that languages with a rich set of pronouns will also have a rich set of deictic contrasts.
By doing a rough search on the World Atlas of Language Structures (WALS) (Dryer
and Haspelmath 2013), where it is possible to combine the value “distance contrasts
in demonstratives” with the value “inclusive–exclusive distinction in independent
pronouns,” we find that this prediction is not borne out. For instance, Navajo [Na-
Dene] and Koasati [Muskogean] have rich deictic systems but no inclusive–exclusive
distinction in pronouns. On the other side of the coin, WALS gives a total of 21 lan-
guages that do have an inclusive–exclusive distinction in pronouns but display only
a two-way deictic contrast. There is even a language, Kera [Afro-Asiatic], that shows
an inclusive–exclusive pronoun system but no distance contrasts in its demonstrative
system. Virtually all combinations in between these extremes are found as well. Thus
there does not appear to be any significant crosslinguistic correlation between person
and deixis.

neutral demonstrative; Imai 2003, section 4.7, 96). Regarding the state of our
understanding with regard to spatial deixis, Imai (2003, 21) points out the
following:
Popular languages that are studied in the literature in more detail than less popular
languages tend to be analyzed as being addressee-anchored: for example, Basque,
Finnish, Hawaiian, Japanese, Korean, and Maori. Less popular languages tend to be
described as a speaker-anchored three-degree distance contrast. It is, however, pos-
sible that some researchers failed to correctly detect the addressee anchor in these
less-studied languages.
We adopt a strong version of Imai’s intuition that the H anchor is

underreported, and we formulate the hypothesis that all languages should be
described not only in terms of an S anchor but also in terms of an H anchor.
In specific cases, syncretisms may obscure the H anchor, of course, a point to
which we return, but nonetheless readings anchored by the H should exist
in every language. We are confident that this hypothesis will be further con-
firmed as the descriptive adequacy of the typological literature improves.
As indicated in readings a–c in Table 5.1, the encoding of S-based prox-
imal ‘close to S’ (reading c) is clearly reported in the literature, but for now,
given the lack of clarity for languages like Venda and Malagasy, we question
the existence of purely S-based medials ‘medium distance from S’ (reading b)
and distals ‘far from S’ (reading a). This is signaled by superscripted question
marks in Table 5.1. We call the S-based proximal (reading c) the Proximal.
With respect to the hearer (H) anchoring of spatial deixis, H- based
proximals ‘close to H’ (reading f) have been identified, as is more fully
demonstrated in Section 5.2.2 We refer to the H-based proximal as the
Medial. H-based medials ‘medium distance from H’ (reading e) and distals
‘far from H’ (reading d) are, as far as we can tell, unattested, as indicated in
Table 5.1 by asterisks.
With regard to the Th(ird person) anchor, according to Imai (2003, 171) no
language expresses more than a single degree of distance from a Th anchor,
namely ‘close to Th’ (reading i). Thus, the Th-based medial ‘medium distance
from Th’ (reading h) and distal ‘far from Th’ (reading g) have been marked by
asterisks. However, even the Th-based proximal (reading i) has rarely been
identified, and there are a number of reasons to doubt that the Th anchor is
actually a valid category. First, though Imai (2003, 25–26) claims that lan-
guages like Kikuyu [Niger–Congo] and Inuktitut [Eskimo–Aleut] have mor-
phologically encoded Th anchors, he himself adds that such readings resemble
“pragmatically conditioned perspective shifts” whereby “the deictic center has
been transferred from the speaker to another person.” Thus the ‘close to Th’
reading seems to have more to do with logophoricity than with spatial deixis.
Moreover, the prefix that Inuktitut uses as its ‘field shifter,’ -ta, is optional and
122
therefore “should be excluded from a paradigm of deictics” (Imai 2003, 70,

n. 5). This casts doubt on the status of Th as an anchor. Furthermore, Imai
(2003, 25–26) observes that, apart from Kikuyu and Inuktitut, no other lan-
guage seems to mark such shifts in their morphology. On the basis of these
considerations, we tentatively conclude that further research on Kikuyu and
Inuktitut would be needed to confirm Th as a deictic anchor.
Let us finally turn to the Incl(usive) readings: ‘far from S and H,’ ‘medium
distance from S and H,’ and ‘close to S and H.’ From the literature it is clear that
that distal reading ‘far from S and H’ (reading j) is attested. This is the reading
we have labeled the Distal. To the best of our knowledge, the Incl-based me-
dial ‘medium distance from S and H’ (reading k) has not been reported, and
we assume it is unattested. For the Incl-based proximal ‘close to S and H’
(reading l), Imai (2003, 22–23) mentions that, though quite rare, it is attested
in languages such as Paamese [Austronesian], Quileute [Chimakuan], Bemba
[Niger–Congo], and various languages of the Philippines [Austronesian].
However, it is unclear how, in fact, the Incl-based proximal reading ‘close to
S and H’ differs from the more clearly attested S-based proximal with the
meaning ‘close to S’ (reading c). Moreover, Imai (2003, 22–23) points out
that the relative proximity denoted by ‘close to S’ as compared with ‘close to
S and H’ is not crosslinguistically consistent. In Binukid [Austronesian], for
instance, the form ʔi ‘close to S and H’ seems to denote an object closer than
the form ʔini ‘close to S,’ but in languages like Waray-Waray [Austronesian]
and Quileute [Chimakuan] the ‘close to S’ form seems to indicate closer prox-
imity than the ‘close to S and H’ form (Imai 2003, 23). Furthermore, in most
languages there is no morphological distinction between encodings of the
two readings (see Janssen 2004, 989–990 on English this, Tahitian teie, and
Japanese ko-). Because of the somewhat peripheral status of the Incl-based
proximal reading and the semantic fuzziness associated with it, we abstract
away from it in this chapter until further research can be conducted.
To sum up, though review of the typological literature may at first sight
appear to lead to a complex inventory of deictic centers and hence a wide array
of spatial contrasts, careful matching of such claims with the available data
reveals—at least at this point—that the crosslinguistic variation is, on the
whole, actually quite modest and fails to reflect such a complexity. Indeed,
in their extensive studies both Diessel (1999, 36, 40) and Imai (2003, 171–
173) conclude that all languages have at least two distance contrasts5 and that
three contrasts are the upper limit for the vast majority of languages (see also
Fillmore 1982, 48–51).6
5. Although some languages have a neutral demonstrative, as we will see, with no

distance contrasts displayed, these languages will often have locative adverbs available
that make at least a two-way contrast.
6. More “exotic” languages like Dyribal, Inuktitut, or West Greenlandic are said to
display dozens of contrasts involving parameters like ‘visible–invisible,’ ‘movement

5.2.2 Syncretism and a universal three-w ay system
In this section we support our classification on the basis of syncretism

patterns. We show that the three-way system posited in Table 5.2 is in fact
universal. More precisely, we show that observed variation across languages
in the encoding of spatial deixis is best understood in terms of syncretism.
Under such a perspective all languages have access to the Proximal, Medial,
and Distal readings in Table 5.2, even though their morphology may not
overtly reflect the three distinct readings. We believe that such a syncre-
tism hypothesis greatly simplifies the assumptions needed to account for the
crosslinguistic variation observed.
5.2.2.1 Dist ≠ Med ≠ Prox
The most straightforward evidence in support of the universality claim of the

availability of three distinct readings identified in Table 5.2 is the fact that
many languages encode exactly such a three-way system in the morphology
of their spatial-deictic systems. In (3)–(19) we provide a sample of 17 lan-
guages that have been reported to display exactly such a system. For the sake
of transparency we have not altered the basic glosses as they appear in our
secondary sources (for the morphosyntactic categories considered, and their
abbreviations, see the titular footnote).
(3)7 Kwakw’ala [Wakashan] (Bach 2006, 270)

dem.pro suffixes
-k ‘1 vis’ [= close to first person and visible]
-uχ ‘2 vis’ [= close to second person and visible]
-iq ‘3 vis’ [= close to third person and visible]
toward–away–across,’ ‘up–down,’ ‘downhill–uphill,’ ‘upriver–downriver,’ ‘north–south

coastline,’ ‘in–out,’ etc. (Diessel 1999, 42–47 and sources cited there; see also Imai
2003, 176 for an exhaustive list). Following Fillmore (1982, 48–51) we assume that
these are not part of the core of spatial deixis. First, it is reasonable to think that
(in)visibility should be considered an evidentiality feature of some kind. The param-
eter of directionality or movement, moreover, is clearly part of the Path system care-
fully studied by Pantcheva (2011), i.e. ‘movement toward’ corresponds to her Goal,
‘movement away’ to her Source, and ‘movement across’ perhaps to her ‘Route.’ Second,
it would appear that many of the remaining parameters here (concepts involving hills,
rivers, coastlines, etc.) belong in the extralinguistic domain, meaning that these bits
do not even belong to UG proper.
7. We consider the gloss ‘3 vis,’ meaning ‘close to third person and visible,’ to be a relic
of Bach’s (2006) particular descriptive terminology and we do not attach too much
importance to it. For our purposes we consider such glosses to conform to the Distal
reading we have established, namely ‘far from S and H.’
124
(4) Passamaquoddy-Maliseet [Algic] (Ng 2002, 94)

dem.pro–adn [Type 3]
an.sg inan.sg
wot yut ‘near speaker’
not nit ‘near addressee’
yat yet ‘away from speaker and addressee’
(5) Wari’ [Chapacura-Wanham] (Everett and Kern 1997, 149)

dem.adn
m–f n
cwa’ ca’ ‘proximate to the speaker’
ma’ ‘proximate to the hearer’
cwain cain ‘far away (distal) from the interlocutors’
(6) Latin [IE] (Bennett 1918, section 87)

dem.pro–adn
m.sg f.sg n.sg
hīc haec hōc ‘this (where I am)’
iste ista istud ‘that (where you are)’
ille illa illud ‘that (something distinct from the speaker)’8
(7) Iraqw [Afro-Asiatic] (Mous 1993, 90–91)

dem.adn
-í (neut -ká) ‘near the speaker’
-síng ‘near the addressee’
-qá‘ ‘near neither of them but still visible’9
(8) Kiswahili [Niger–Congo] (Okombo and Habwe 2007, 82–83)

dem.pro–adn
huyu ‘proximal to the speaker and listener’
hicho ‘proximal to the addressee and distal to the speaker’
kule ‘away from the speaker and addressee’
(9) Sinhala [IE] (Chandralal 2010, 228)

dem.adn
mee ‘proximal to speaker, or to both speaker and hearer’
oyə ‘proximal to hearer’
arə ‘distal from both speaker and hearer; in sight’10
8. More precisely: “le plus éloigné dans l’espace et dans le temps” or “il se réfère à une
troisième personne ou un troisième objet” (Gaffiot 2016: 683).
9. There is also a distal-invisible item -dá‘ that we do not include, as we think evi-
dential features encoding visibility must lie outside the core spatial-deixis fseq we are
trying to uncover here.
10. Here again there is also a distal-invisible ee that we do not include.

(10) Khasi [Austro-Asiatic] (Diessel 1999, 43)
roots
-ne ‘proximal’
-to ‘medial (near H)’
-tay ‘distal’
(11) Korean [isolate] (Diessel 1999, 20–21)

dem.adn
i ‘near the speaker’
ku ‘near the hearer’
ce ‘away from both speaker and hearer’
(12) Japanese [Japonic] (Diessel 1999, 59)

roots
ko- ‘near S’ (and/or hearer; Janssen 2004, 989)
so- ‘near H’
a- ‘away from S + H’
(13) Yimas [Lower Sepik-R amu] (Foley 1991, 112)

roots
-k ‘near speaker’
m- ‘near hearer’
-n ‘near neither speaker nor hearer’
(14) Arapesh [Torricelli] (Aronoff 1994, 98)

dem.pro [gender viii]
SG PL
eñuda’ ešuda’ ‘pro-near-me’
neñuda’ nešuda’ ‘pro-near-you’
ñeiñuda’ šeišuda’ ‘pro-over-there’
(15) Tukang Besi [Austronesian] (Donohue 1999, 137, 147)

dem.pro–adn
ana ‘near the speaker’
atu ‘nearer the addressee than the speaker’
iso ‘at a distance from either the speaker or the listener(s)’
(16) Tahitian [Austronesian] (Tryon 1970, 24)

dem.adn
teie ‘near the speaker’ (and/or hearer; Janssen 2004, 990)
tēna ‘near the person addressed’
tēra ‘not near the speakers’
126
(17) Ket [Yeniseian] (Werner 1997, 137)

dem.pro
ki ‘neben dem Sprechenden’
tu ‘neben dem Zuhörenden’
qa ‘vom Sprechenden und Zuhörenden entfernt’
(18) Basque [isolate] (Hualde and de Urbina 2003, 123)

dem.pro–adn
(h)au(r) ‘this . . . indicates proximity to the speaker’
(h)ori ‘that (just there) . . . [indicates] proximity to the addressee’
(h)ura ‘that (over yonder) . . . [indicates] remoteness from both’
(19) Old Catalan [IE] (Imai 2003, 24, citing Otaka 1987)
dem.pro–adn
m.sg f.sg
aquest aquesta ‘proximal to the speaker ‘
aqueix aqueixa ‘proximal to the addressee’
aquell aquella ‘distal’
We take the languages illustrated in (3)–(19) to be those wearing the universal

spatial-deictic system “on their sleeve.” Put differently, in the relevant lan-
guages the universal system in Table 5.3 is not obscured by syncretism: Each
conceptual category is matched by a formal category.
5.2.2.2 Dist ≠ Med = Prox
In this section we turn to languages whose morphology seems to corre-

spond to a two-way system: One deictic item has the reading ‘close to S or
H’ and the other one ‘far from S and H.’ Examples are given in (20)–(22).
We do not conclude that the system of spatial deixis of such languages is
impoverished, though. Rather, we assume that they illustrate cases of syn-
cretism in which the Medial and Proximal readings are expressed by the
same morphological item, with a second distinct item encoding the Distal
reading only.
Table 5.3 THREE-W AY SYSTEM (DIST ≠ MED ≠ PROX)
Distal Medial Proximal

‘far from S and H’ ‘close to H’ ‘close to S’

(20) Bulgarian [IE] (Imai 2003, 23, citing Yajima 1984)
dem.adn
m.sg f.sg n.sg
toozi taazi tovaa ‘close to speaker or hearer’
onzi onazi onova ‘distal’
(21) Apurinã [Arawakan] (Facundes 2000, 356)

dem.adn
m f
iye oye ‘close to the speaker or to the hearer’
ukira okira ‘far from the speaker and hearer’
(22) Catalan [IE] (Imai 2003, 23-24, citing Hualde 1992)

dem.pro–adn
m.sg f.sg
aquest aquesta ‘proximal to either the speaker or the addressee’
aquell aquella ‘distal’
Evidence for a Med–Prox syncretism in the prepositional system is also found

in Fijian [Austronesian]. Geraghty (1976) reports that the prepositions e and
mai are usually glossed as ‘locative—close to S’ and ‘locative—far from S’, re-
spectively, as seen in (23).
(23) Fijian [Austronesian] e and mai (Geraghty 1976, 513)

a. Sā tiko e waqa na kato.
prt is on boat the box
‘The box is on the boat (Speaker is on boat)’
b. Sā tiko mai waqa na kato.
prt is on boat the box
‘The box is on the boat (Speaker not on boat)’
However, on the basis of information obtained from native informants,

Geraghty is able to identify more precise readings for these prepositions.
He finds that (23a) with the preposition e remains possible in the context in
which the S is not on the boat but the H is; however, (23b) with mai can never
be used if the S, the H, or both, happen to be on the boat (Geraghty 1976,
514–515). Thus he concludes that e refers to a “location close to speaker or
hearer” whereas mai marks “location remote from both speaker and hearer”
(Geraghty 1976, 515). Geraghty’s description, which is an improvement on
previous accounts that did not take into account the H anchor, thus uncovers
another attestation of a Med–Prox syncretism.
128
Table 5.4 MED–P ROX SYNCRETISM

‘close to S or H’
Evidence for a Med–Prox syncretism may also be found in Basque. As al-

ready seen in (18), Basque belongs to the group of languages with a three-way
system. However, one specific fact from the history of this language hints at
the availability of a Med–Prox syncretism. According to Hualde and de Urbina
(2003, 122), Western varieties of Basque have a “proximate article” in the
plural, as shown for instance in gizon-ok ‘the men here.’ In earlier texts, a prox-
imate article was attested also in the absolutive singular, realized in different
variants: -ori, -or, -au, and -o. It should be noted that the variant -au resembles
the Proximal (h)au(r) in (18), whereas the -ori and -or variants are clearly related
to the Medial (h)ori instead. Thus this overlap in the abs.sg proximate article
could be argued to instantiate at least a partial coalescence of Prox and Med.
As already mentioned, then, though at first sight the languages discussed in
this section seem to show a two-way contrast, the underlying system can also
be described in terms of the encoding of three separate readings, whereby two
of them are expressed by the same morphological item (i.e. a Med–Prox syn-
cretism) and the third one has a distinct morphological realization (i.e. Dist).
This is summarized in Table 5.4 in which the shaded cells indicate syncretism.
The next section discusses a different kind of two-way system.
5.2.2.3 Dist = Med ≠ Prox
We now turn to languages that have been identified as displaying a two-way

(so-called distance-oriented) system, in which one deictic item has the reading
‘far from (or not close to) S’ and the other one ‘close to S.’ English is one lan-
guage which has such a system, as shown in (24).11
11. We note that for English and for many other examples in this chapter, more
morphological decomposition is possible (i.e. th-is, th-at). However, when consid-
ering syncretism patterns it is not necessary to have a highly fine-grained mor-
phological segmentation. For instance, we could refine the analysis of English by
saying that the morphemes -at (Dist–Med) vs. -is (Prox) are responsible for making
the deictic distinction here, but the syncretism pattern remains the same if we do
not perform the decomposition (i.e. that vs. this). Similarly, a reviewer suggests that
decomposing the Latin forms (classified as making up a three-way system) appears
to reveal a Dist–Med syncretism in the initial element i- and the final inflectional
component: Dist i-ll-e, i-ll-a, i-ll-ud and Med i-st-e, i-st-a, i-st-ud. Rather than syncre-
tism, however, this seems to be a case of sharing a demonstrative root (consider is,
ea, id meaning ‘he she, it’ or acting as the neutral demonstrative), with deixis being

(24) English [IE]
dem.pro–adn
this ‘close to S’
that ‘not close to S’
With respect to the interpretation of that in English, however, the ‘close to H’

reading exists too: When pointing to a referent that is close to the H, that (and
not this) will be used. It just so happens that the item that will be used for a
referent far from both S and H as well. As a result of this syncretism, the spe-
cific H anchor associated with the Medial reading is obscured in the English
system. However, given that the anchor is overtly available in other languages
(and given that we assume the oppositions to be universal), we postulate that
such a reading must be available in the underlying system.
Additional examples of languages with a morphologically two-way system
with a Dist–Med syncretism are illustrated in (25)–(34). The criterion for in-
clusion as a language displaying a Dist–Med syncretism was not only that a
‘proximal’ versus ‘distal’ system be reported in the secondary source for the
language in question, but crucially that this system be reported as being S-
anchored. Thus, in the following examples, where glosses are not sufficient, we
have also provided the relevant passage reporting the anchor.
(25) Klallam [Salish] (Montler 2007, 411, 419–420)

dem.pro–adn
non-fem fem
tiǝ tsiǝ ‘near’
tǝsǝ ɬǝsǝ ‘far’
“the far and near demonstratives indicate distance from the speaker,
not necessarily the addressee” (Montler 2007, 419)
(26) Epena Pedee [Choco] (Harms 1994, 45)

dem.adn
na ‘this (here) –physically proximate to the speaker’
hã ‘that (there) –physically distant from the speaker’
(27) Macushi [Cariban] (Abbott 1991, 105)

dem.pro–adn
an inan
mîserî – coll. insemoro seni – coll. sîrîrî ‘near to the speaker’
mîîkîrî – coll. inkamoro siini – coll. mîrîrî ‘remote from the speaker’
marked by distinct elements (Dist -ll- and Med -st-), keeping the three-way system
intact. For the sake of presentation, then, we do not enter into such fine-grained
decompositions in these cases.
130
(28) Pirahã [Mura] (Everett 1986, 285)

dem.pro–adn
gíisai ‘this – proximal’
gáihi ‘that – distal’
“distinguished by the proximity of the referent to the speaker”
(Everett 1986, 285)
(29) Lingala [Niger–Congo] (Meeuwis and Stroeken 2012, 148)

dem.adn
óyo ‘close to the speaker’
wâná ‘close to the hearer or away from both speaker
and hearer’12
(30) Mandarin Chinese [Sino-Tibetan] (Yip and

Rimmington 2004, 48)
dem.pro–adn
zhè ‘close to the speaker’
nà ‘away from the speaker’
(31) Limbu [Sino-Tibetan] (Tumbahang 2007, 149–150)

dem.pro stems
ABS.SG
ba- ‘close to the speaker . . . even if it is far away from
the hearer’
hamba- ‘remote . . . from the speaker . . . [even] if the referents
are closer to the
hearer than to the speaker’
(32) Semelai [Austro-Asiatic] (Kruspe 2004, 192)

dem.pro–adn
(ʔ)nɔʔ ‘this’
ke ‘that’
“based on simple distance orientation from the speaker”
(Kruspe 2004, 192)
(33) Gooniyandi [Australian] (McGregor 1990, 144)

dem.pro(?)–adn
ngirndaji ‘this’
ngooddoo ‘that’
“distance with respect to the speaker” (McGregor 1990, 144)
12. In fact, it is explicitly stated that “wâná covers both the medial and the distal
scopes” (Meeuwis and Stroeken 2012, 148).

Table 5.5 DIST–M ED SYNCRETISM

‘far from/not close to S’
(34) Evenki [Altaic] (Nedjalkov 1997, 210–213)

dem.pro–adn
er ‘this –near the speaker’
tar ‘that –not considered by the speaker to be near’
In Section 5.3 we also illustrate the Dist–Med syncretism in Wargamay

[Australian], Gulf Arabic [Afro-Asiatic], and Welsh [IE].
Once again we can understand what may appear to be two-way distance-
oriented systems in terms of a universal hierarchy with three separate
concepts if we assume that two of these concepts are expressed by the same
morphological exponent (Dist–Med) and the third one has its own distinct
morphological realization (Prox), as illustrated in Table 5.5.
5.2.2.4 Dist = Med = Prox
Diessel (1999, 36–39) reports that although all languages have at least
two locative adverbs expressing a contrast like ‘here’ versus ‘there,’ some
languages in fact have ‘neutral’ or ‘unmarked’ demonstratives without ap-
parently encoding distance contrasts at all. The seven languages with such
neutral demonstratives in Diessel’s (1999) sample are Alamblak [Sepik],
Czech [IE], French [IE], German [IE], Koyra Chiini [Nilo-Saharan], Supyire
[Niger–Congo], and Tok Pisin [English creole]. In these languages the neutral
demonstratives are very close to definite articles or the like, and it is pos-
sible that they are developing in this direction (i.e. from exophoric to endo-
phoric). Diessel is careful to mention, though, that for now they retain their
exophoric usage.
A methodological issue arises here. Because our hypothesis that there are
universally three core readings in the spatial deixis domain is not adopted by
all researchers, it is quite difficult to glean precise glosses from the descrip-
tive literature, and thus it is difficult to verify if the three readings we are
concerned with are available for the neutral demonstrative of the language in
question.
However, a simple example from a familiar language can show how
the neutral demonstrative works. In French, there is one adnominal
132
demonstrative, whose form varies in number and gender: m.sg ce, f.sg
cette, pl ces. The form can be used in all three of the deictic contexts we
have postulated. We illustrate this in detail in (35), providing the relevant
discourse contexts.
(35) French [IE]

S and H are sitting facing each other at a table; S points to a book at
location (a, b, c) and says:
Ce livre est vraiment bon; tu devrais le lire.
CE book is truly good you should it read
a. ✓ ‘close to the S’ (directly in front or beside)
b. ✓ ‘close to the H’ (directly in front or beside)
c. ✓ ‘far from S and H’ (on a table at the other side of the room)
For the sake of completeness, we add that our informants report that the
locative adverbs (i)ci (usually glossed as ‘here’) and là (usually glossed as
‘there’) are actually distance-neutral. Rather than distinguishing proximity
or remoteness, these items are in fact used for contrastive purposes when
two or more referents are involved. As such they are not used in any of
the three contexts in (35). The locative adverb là-bas ‘over there,’ however,
does carry a distinct distal or remote-type reading, and may optionally be
used in (35c).13 Thus, in French we have a clear-cut case of a neutral ges-
tural demonstrative, ce(tte)–ces, which can be used in Proximal, Medial, and
Distal contexts. In other words, French ce(tte)–ces displays a Dist–Med–
Prox syncretism.
Within our three-way system, languages like French can be thought of as
having total syncretism among all three readings, with one morphological ex-
ponent for Prox, Med, and Dist, as shown in Table 5.6.
Table 5.6 DIST–M ED–P ROX SYNCRETISM

neutral dem
Even one-way systems, then, can be thought of in terms of the universal

three-way contrast identified in Section 5.2.1.
13. Frédérique Berthelot (p.c.) points out that it is important to distinguish ce livre-là

from ce livre, lá. In the latter structure là is a topic marker and as such can be used in
any context, regardless of distance or number of referents.

5.2.2.5 Interim summary
Despite at least twelve logically conceivable spatial-deictic readings, we have

argued that only three spatial-deictic readings are attested crosslinguistically:
Proximal ‘close to S,’ Medial ‘close to H,’ and Distal ‘far from S and H.’ These
three readings, moreover, are taken to be universal, even though not all lan-
guages show a transparent three-way system. Languages that might be taken
to display impoverished two-way and one-way systems can be understood
in terms of a mismatch between meaning and form in that the underlying
three-way semantic distinction is encoded in one or two forms that display
syncretism.
Let us consider the syncretisms we advocated for in this section more
closely. Out of the five possible syncretism patterns among Dist, Med, and
Prox, in fact only four are attested, as shown in Table 5.7.
One possible syncretism is not attested, namely the pattern in Table 5.8:
There is no language with a two-way system in which one deictic item encodes
either a Distal or Proximal reading (i.e. Dist–Prox syncretism) and the second
deictic item separately encodes a Medial reading.
This missing pattern is the so-called ABA pattern (see Bobaljik 2007,
2012 and Caha 2009). The ABA gap is expected on nanosyntactic grounds,
because it is ruled out by nanosyntactic principles of spellout (see Baunaz
and Lander in Chapter 1, Section 1.3.3.2, for discussion and references).
Because syncretism is restricted to contiguous layers in this way, it is
an indispensable tool for uncovering the linear sequence of functional
heads. To impose hierarchical structure on this linear order, another kind
of phenomenon needs to be taken into account, namely morphological
containment.
Table 5.7 FOUR AT TESTED SYNCRETISMS

Table 5.8 ONE UNAT TESTED SYNCRETISM (*ABA)

*
134
5.3 MORPHOLOGICAL CONTAINMENT
The phenomenon of morphological containment became prominent thanks

to pioneering work by Bobaljik (2007, 2012) in the framework of Distributed
Morphology and was given a nanosyntactic implementation in works by Starke
(e.g. 2013), Caha (2009), and Pantcheva (2011), among others. Morphological
containment can be used to detect underlying hierarchical relations. To take
a simple example of morphological containment from Bobaljik’s (2007,
2012) work, we can clearly see that the basic (or positive) form of an adjective
(Adj), for example, great, is structurally smaller than the comparative (Cmpr)
form, for example, great-er. Because the comparative is bigger than the posi-
tive, we can posit (assuming that syntax builds from the bottom up) the un-
derlying hierarchy Cmpr > Adj or [Cmpr [Adj]].
In the case of spatial deixis, the literature provides abundant crosslinguistic
evidence for unidirectional containment, that is, the reported patterns of
containment systematically show Distal to be morphologically more complex
than Medial and Medial as morphologically more complex than Proximal.
Following nanosyntactic guidelines, we interpret these containment relations
to be significant and to reflect the internal organization of three additive
heads involved in the encoding of spatial deixis.
5.3.1 Medial contains Proximal
In some languages we see that the morphological expression of Proximal is

overtly contained within the morphological expression of Medial (36)–(39).
In Ma’di, for instance, the Proximal morpheme ɗ ɨ̀ is contained within the
morpheme ɨ̀lέɗ ɨ̀ that encodes Medial. The containment relation is sketched
in (36b) using brackets, showing that Medial is bimorphemic, with one of the
morphemes being the Proximal form.
(36) Ma’di [Nilo-Saharan] (Blackings and Fabb 2003, 123)14

a. dem.adn
ɗ ɨ̀ ‘this N (proximal in the physical context)’
ɨ̀lέɗ ɨ̀ ‘that N near you (in the physical context)’
b. containment
[Med [Prox]]
[ɨ̀lέ-[ɗ ɨ̀]]
14. For the sake of completeness we add here that it is not clear from the description
if ɗ ɨ̀ is S-anchored. Importantly, however, the item ɨ̀lέɗ ɨ̀ is explicitly mentioned as
H-anchored.

Similarly in Nkore-Kiga (37), the Proximal morpheme is contained within
the element encoding Medial, with the Prox ingredient first, that is, [[Prox]
Med].
(37) Nkore-Kiga [Niger-Congo] (Taylor 1985, 135-136)

a. dem.pro–adn
ogu ‘near speaker’
ogwo ‘near hearer’15
b. containment
[[Prox] Med]
[[ogu]-o] > ogwo
The pattern is also observed in Boumaa Fijian (38) and Palauan (39).
(38) Boumaa Fijian [Austronesian] (Finegan 2013, 212; dialect A in

Dixon 1988)
a. dem.pro–adn
oŋgo ‘near the speaker’
oŋgori ‘near the hearer’
b. containment
[[Prox] Med]
[[oŋgo]-ri]
(39) Palauan [Austronesian] (Janssen 2004, 989–990)

a. dem.pro
ngile ‘this –related to the first person exclusive’
ngilecha ‘that –related to the second person’
b. containment
[[Prox] Med]
[[ngile]-c ha]
The containment pattern observed is summarized in Figure 5.2.

Note that the linear ordering between the Medial component and the
Proximal component (e.g. Med > Prox in Ma’di [ɨ̀lέ-[ɗ ɨ̀]] but Prox > Med in
Palauan [[ngile]-cha]) is not relevant for establishing the containment relation
in Figure 5.2.
15. For ogwo, Taylor (1985, 135) also provides the meaning ‘not far removed from
both speaker and hearer.’
136
Medial Proximal
Figure 5.2 Medial contains Proximal
5.3.2 Medial and Distal contain Proximal
In other languages we see that a morpheme encoding the Proximal reading is

overtly contained within both the Medial and Distal items.
In Wailevu Fijian (40), the Proximal morpheme is γā; this morpheme is
contained in the bimorphemic Medial γā-ri and also in the bimorphemic
Distal γa-ðei.
(40) Wailevu Fijian [Austronesian] (Ross 2007, 278)

a. dem.pro–adn
γā ‘near speaker’
γāri ‘near addressee’
γaðei ‘distant from both speaker and addressee’
b. containments
[[Prox] Med] [[Prox] Dist]
[[γā]-
ri] [[γa]-ðei] (provided γā surfaces as γa here)
As seen in the paradigms in Tables 5.9 and 5.10, Ewondo also shows contain-
ment of the Proximal within both the Medial and Distal forms. In noun class 2,
for example, the Proximal singular ɲū is contained within the Medial singular
ɲū-lú and also within the Distal singular ɲū-lí. The same pattern is instantiated
in the plural, where the Proximal mī is contained within both Medial mī-lí and
Distal mī-líí.
Table 5.9 dem.pro–a dn IN EWONDO [NIGER-C ONGO]

(DIESSEL 1999, 18, CITING REDDEN 1980)
1 2 3 4 5 6
ɲɔ́ ɲū dī dzī ɲī ɲū ‘near S’
sg ɲɔ́lō ɲūlú dīlí dzīlí ɲīlí ɲūlú ‘near H’
ɲɔ́lí ɲūlí dílí dzílí ɲílí ɲūlí ‘away from S + H’
bá mī mā bī mā dī ‘near S’
pl bálā mīlí mālá bīlí mālá dīlí ‘near H’

bálí mīlíí mālí bīlíí mālí dīlíí ‘away from S + H’

Table 5.10 [[PROX] MED] AND [[PROX] DIST] IN EWONDO
1 2 3 4 5 6
ɲɔ́ ɲū dī dzī ɲī ɲū Prox

sg [[ɲɔ́]-lō] [[ɲū]-lú] [[dī]-lí] [[dzī]-lí] [[ɲī]-lí] [[ɲū]-lú] Med
[[ɲɔ́]-lí] [[ɲū]-lí] dílí dzílí ɲílí [[ɲū]-lí] Dist
bá mī mā bī mā dī Prox
pl [[bá]-lā] [[mī]-lí] [[mā]-lá] [[bī]-lí] [[mā]-lá] [[dī]-lí] Med
[[bá]-lí] [[mī]-líí] [[mā]-lí] [[bī]-líí] [[mā]-lí] [[dī]-líí] Dist
In the singular there are only three forms (the Distal in classes 3, 4, and 5)
that happen not to show this overt containment of Proximal; in the plural, the
containment of Proximal is completely systematic across noun classes.
The patterns previously observed are summarized in the diagrams in
Figures 5.3 and 5.4.
Medial Proximal
Figure 5.3 Medial contains Proximal
Distal Proximal
Figure 5.4 Distal contains Proximal
The containment relation in Figure 5.3 is the same as that in Figure 5.2,
which was established on the basis of data from Ma’di, Nkore-Kiga, Boumaa
Fijian, and Palauan. Wailevu Fijian and Ewondo provide additional evidence
for this containment relation, as well as providing evidence for the relation
in Figure 5.4. This means that the Proximal can be contained not only within
the Medial but also within the Distal. At this point, the relative containment
relation of Medial vis-à-vis Distal has yet to be established.
5.3.3 Distal–M edial contains Proximal
A morpheme with the Proximal reading can also be overtly contained within
an item which is itself syncretic for the Distal and the Medial reading (see
Section 2.2.3).
138
First take Wargamay, which has a two-way system consisting of a syncretic

Dist–Med versus Prox. In (41) we see that the syncretic Distal–Medial form
ɲuŋga-ɠi contains the Proximal form ɲuŋga.
(41) Wargamay [Australian] (Dixon 1981, 44-45)

a. dem.pro–adn
ɲuŋga ‘this one (near speaker)’
ɲuŋgaɠi ‘that one (distant from speaker)’
b. containment
[[Prox] Dist–Med]
[[ɲuŋga]-ɠi]
Similarly, in Gulf Arabic (42), the Proximal forms m.sg haadha and f.sg
(haa)dhi are contained within the syncretic Distal–Medial forms m.sg
(haa)dha-ak and f.sg (haa)dhi-ich.
(42) Gulf Arabic [Afro-Asiatic] (Holes 1990, 172-173)

a. dem.pro
m.sg f.sg
haadha (haa)dhi ‘this –near to the speaker’
(haa)dhaak (haa)dhiich ‘that –remote . . . from the speaker’
b. containment
[[Prox] Dist–Med]
[[haadha]-ak]
[[(haa)dhi]-ich]
In Welsh (43) we also see a version of this phenomenon.
(43) Welsh [IE] (Borsley et al. 2007, 176)

a. dem.pro–adn
m.sg f.sg n.sg
hwn hon hyn ‘proximal – this’
hwnnw honno hynny ‘distal – that’
b. containment
[[Prox] Dist–Med]
[[hwn]-nw]
[[hon]-no]
[[hyn]-ny]
It should be noted that the Welsh system is in a state of flux, as the hwn and
hwnnw systems are falling together semantically. Concurrently with this coa-
lescence, a new Distal form is emerging with the form m.sg hwnna, f.sg honna,

n.sg hynna ‘physically distant from the speaker’ (Borsley et al. 2007, 176).
Here too, though, we observe containment of the Proximal within the new
Distal: [[hwn]-na], [[hon]-na], and [[hyn]-na].
As a final example, in Limbu (44), the Proximal stem is ba-, a morpheme
which is contained within the Distal–Medial stem ham-ba-.
(44) Limbu [Sino-Tibetan] (Tumbahang 2007, 149)

a. dem.pro stems
ba- ‘close to the speaker’
hamba- ‘remote . . . from the speaker’
b. containment
[Dist–Med [Prox]]
[ham-[ba-]]
These patterns may be summarized in Figure 5.5.
Distal / Medial Proximal
Figure 5.5 Distal/Medial contains Proximal
Note once again that establishment of the containment relation in Figure 5.5
is independent of the linear ordering of Dist–Med and Prox (e.g. Dist–Med > Prox
in Limbu [ham-[ba-]] but Prox > Dist–Med in Welsh [[hwn]-nw]).16
5.3.4 Distal contains Medial
The containment patterns discussed in the previous three subsections provide

morphological evidence from a range of languages that the structure encoding
the concept Proximal ‘close to S’ reading is smaller than structures encoding
Medial and Distal. The size of the Medial structure vis-à-vis the Distal struc-
ture, however, is still to be determined.
16. The Scandinavian languages appear to present a counterexample, because Dist–

Med n.sg det, cm.sg den ‘that’ might be analyzed as being contained within Prox n.sg
[[det]-ta], cm.sg [[den]-na] ‘this’ (Swedish). Although there are various issues that make
the Scandinavian data more complicated than they may at first seem, we would like
to point out here that if the colloquial language is to be considered, then the genuine
members of the spatial-deixis system actually involve locative reinforcers meaning ‘there’
and ‘here’: Dist–Med n.sg det där, cm.sg den där, pl dom där ‘that–those’ vs. Prox n.sg
det här, cm.sg den här, pl dom här ‘this–these’ (Swedish). Note that no containment is
observed in these forms. See Lander (2015, appendix V) for more discussion. See also Leu
(2015: sections 2.2, 2.3) for the view that the ‘here’ and ‘there’ reinforcers can be silent.
140
Examples of containment relations involving Medial and Distal compo

nents are harder to come by, but Boumaa Fijian (45) deserves to be discussed
here. In this language the Medial is yā, which is clearly contained within the
bimorphemic Distal ma-yā.
(45) Boumaa Fijian [Austronesian] (Ross 2007, 278; dialect B in Dixon

1988, 58)
a. dem.pro–adn
yā ‘near addressee’
mayā ‘distant from both speaker and addressee’
b. containment
[Dist [Med]]
[ma-[yā]]
The forms from Boumaa Fijian show the containment relation sketched in
Figure 5.6.
Distal Medial
Figure 5.6 Distal contains Medial
Supporting evidence for this containment relation is discussed next.
5.3.5 Distal contains Medial contains Proximal
Consider Ewondo from Tables 5.9 and 5.10 again. This language displays
instances of what might be called “total nesting.”. Recall from Section 5.3.2 that
some parts of the demonstrative paradigm show containment of the Proximal
within both the Medial and the Distal. Additionally, there is a containment re-
lation between the Distal and the Medial: Medial mīlí is contained within the
Distal mīlí-í. As shown in (46), this also applies to the plural of classes 4 and 6.
(46) Ewondo [Niger-Congo] (Diessel 1999, 18, citing Redden 1980)

a. class 2 pl mīlí ‘near H’
mīlíí ‘away from S + H’
class 4 pl bīlí ‘near H’
bīlíí ‘away from S + H’
class 6 pl dīlí ‘near H’
dīlíí ‘away from S + H’

b. containment
[[[Prox] Med] Dist]
[[[mī]-lí]-í]
[[[bī]-lí]-í]
[[[dī]-lí]-í]
In the plural Distal forms of classes 2, 4, and 6, then, we see all three layers of
the spatial-deictic fseq morphologically realized, a case of total nesting.
The Ewondo paradigm shows the containment relation sketched in
Figure 5.7.
The diagram in Figure 5.7 summarizes all of the containment relations
discussed: Proximal is contained within Medial (e.g. Ma’di), Proximal is
contained within Distal (e.g. Wailevu Fijian), Medial is contained within Distal
(Boumaa Fijian), and Proximal is contained within Medial, which in turn is
contained within Distal (Ewondo). In the discussion in Section 5.4 we see that
the fseq of spatial-deictic features directly replicates Figure 5.7.
Figure 5.7 Distal contains Medial contains Proximal
5.4 CAPTURING THE PATTERNS IN A FUNCTIONAL SEQUENCE
Our main claim is that spatial deixis is semantically encoded in Universal

Grammar as a three-way contrast: Proximal ‘close to S,’ Medial ‘close to H,’
and Distal ‘far from S and H.’ This claim is supported by the prevalence of ex-
actly such a three-way system across languages, but also by the fact that other
types of systems can also be understood in terms of this three-way system, if
these systems are analyzed in terms of syncretism.
One possible proposal for representing the syntax of spatial deixis would be
that there is a single functional head like Deix (as in Svenonius 2010) or DxPLACE
(as in Den Dikken 2010), whose value is determined by the presence of one
specific feature. According to Figure 5.8, if the deictic head Deix is projected,
then one of the values Prox–Med–Dist is chosen to the exclusion of the other
two. That is, the three features in Figure 5.8 are in complementary distribution.
Indeed, it is conceivable that the flat structure in Figure 5.8 might in some
way be able to account for the syncretism data discussed in Section 5.2, as Prox
142
DeixP
Deix0
[Prox] [Med] [Dist]
Figure 5.8 Flat structure for spatial deixis
is adjacent to Med (perhaps allowing for a Med–Prox syncretism) and Med is

adjacent to Dist (perhaps allowing for a Dist–Med syncretism) in Figure 5.8.
Crucially, Med intervenes between Prox and Dist, potentially blocking the
ABA pattern from arising.
However, our containment data from Section 3 speak against the flat
structure in Figure 5.8. The containment data clearly reveal an asymmet-
rical relation among Prox, Med, and Dist structures: The three spatial-deictic
distinctions are not on a par, but rather they show an internal hierarchy,
in which Prox is smaller than Med, which in turn is smaller than Dist. See
Figure 5.9 (= Figure 5.7).
To put Figure 5.9 in more formal terms, we assume that spatial deixis is
an independent functional domain made up of formal syntactico-semantic
features (call them Dx features). The featural relation among Prox, Med, and
Dist is formally represented in Figure 5.10.
Adopting a cartographic approach, we assume that the syntactico-semantic
features making up syntactic structure are constrained by the “one feature–one
Figure 5.9 Distal contains Medial contains Proximal
Dx3 > Dx2 > Dx1
Prox
Med
Dist
Figure 5.10 Superset–subset relations in spatial deixis

head” principle (Cinque and Rizzi 2008, 50), meaning that features are as-
sociated with specialized functional heads that head functional projections.
Although these projections could conceivably be either left-branching or
right-branching, we follow Kayne (1984, 1994) in assuming that only binary-
and right-branching structures are possible. The three Dx features in Figure
5.10, then, comprise the fseq in Figure 5.11.
Dx3P
Dx3 Dx2P
Dx2 Dx1P
Dx1
Figure 5.11 Functional sequence of spatial deixis
The heads Dx1, Dx2, and Dx3 are understood to be unary and additive, such
that the structure underlying the Proximal reading corresponds to [Dx1], the
Medial reading corresponds to [Dx2 [Dx1]], and the Distal reading corresponds
to [Dx3 [Dx2 [Dx1]]]. This is shown in Figure 5.12.
Put differently, the Proximal is a subset of the Medial, and the Medial is, in
turn, a subset of the Distal. Thus we replace the flat, tertiary-branching struc-
ture in Figure 5.8 with a binary-branching structure with an internal structure
that directly reflects the containment data just discussed. While syncretism
is normally taken to reveal a linear ordering of features (Dist | Med | Prox),
morphological containment, on the other hand, reveals hierarchical relations,
automatically establishing a single linear order in the process. In this sense it
is perhaps the more helpful diagnostic of the two.
Dx3P ⇒ Distal
Dx2P ⇒ Medial Dx2P

Dx3
Dx1P ⇒ Proximal
Dx2 Dx2 Dx1P
Dx1P
Dx1
Dx1 Dx1
Figure 5.12 Additive heads
144
5.5 FINAL THOUGHTS
We have proposed a universal three-way distinction in the functional domain

of spatial deixis on the basis of syncretism and morphological containment
data. The syncretism data provide the means of determining which functional
layers must be adjacent in a linear order of syntactico-semantic features; the
morphological containment data provide us with a hierarchy of features, re-
vealing in a very clear way the relative sizes of Distal, Medial, and Proximal
structures.
Observe that Prox ‘close to S’ is at the bottom of the spatial-deixis hier-
archy. Importantly, this makes sense from a semantic viewpoint. Adopting the
fseq proposed here, then, corresponds to saying that the speaker’s perspective
is built first, giving spatial deixis an egocentric foundation (see Bühler 1934).
From here the domain of spatial reference is extended to that of the hearer, by
most accounts the next-important participant in the speech situation. Finally,
the domain is extended beyond the immediate surroundings to an area beyond
the speaker and hearer. This is a common way to characterize the semantics of
spatial deixis, but, importantly, we have shown here that morphological form
parallels the semantics, suggesting that form and meaning are closely linked.
LANGUAGES
Apurinã Facundes (2000)

Arapesh Aronoff (1994)
Basque Hualde and de Urbina (2003)
Binukid Imai (2003)
Bulgarian Imai (2003)
Catalan Imai (2003)
Chinese Yip and Rimmington (2004)
English Eric Lander
Epena Pedee Harms (1994)
Evenki Nedjalkov (1997)
Ewondo Diessel (1999)
Fijian Dixon (1988)
Finegan (2013)
Geraghty (1976)
Ross (2007)
French Frédérique Berthelot (personal communication)
Lena Baunaz (personal communication)
Gooniyandi McGregor (1990)
Gulf Arabic Holes (1990)

Iraqw Mous (1993)
Japanese Diessel (1999)
Ket Werner (1997)
Khasi Diessel (1999)
Kiswahili Okombo and Habwe (2007)
Korean Diessel (1999)
Imai (2003)
Klallam Montler (2007)
Kwakw’ala Bach (2006)
Latin Bennett (1918 [1895])
Limbu Tumbahang (2007)
Lingala Meeuwis and Stroeken (2012)
Macushi Abbott (1991)
Ma’di Blackings and Fabb (2003)
Nkore-Kiga Taylor (1985)
Palauan Janssen (2004)
Passamaquoddy Ng (2002)
Pirahã Everett (1986)
Semelai Kruspe (2004)
Sinhala Chandralal (2010)
Swedish Eric Lander
Tahitian Tryon (1970)
Tukang Besi Donohue (1999)
Wargamay Dixon (1981)
Wari’ Everett and Kern (1997)
Welsh Borsley et al. (2007)
Yimas Foley (1991)
REFERENCES
Abbott, Miriam. 1991. “Macushi.” In Handbook of Amazonian Languages, Vol. 3, edited by

Desmond C. Derbyshire and Geoffrey K. Pullum, pp. 23–160. Berlin: De Gruyter.
Anderson, Stephen R. and Keenan, Edward L. 1985. “Deixis.” In Language Typology
and Syntactic Description, Vol. 3, edited by Timothy Shopen, pp. 259– 308.
Aronoff, Mark. 1994. Morphology by Itself: Stems and Inflectional Classes. Cambridge,
MA: MIT Press.
Bach, Emmon. 2006. “Paradigm Regained: Deixis in Northern Wakashan.” SOAS
Working Papers in Linguistics 14: pp. 267–281. [online] Available at http://works.
bepress.com/emmon_bach/28/
Bennett, Charles E. 1918 [1895]. New Latin Grammar, 3rd ed. Norwood, MA:
Norwood Press.
146
Blackings, Mairi and Nigel Fabb. 2003. A Grammar of Ma’di. Berlin: De Gruyter.
Bobaljik, Jonathan D. 2007. “On Comparative Suppletion.” Ms., University of
Connecticut.
Bobaljik, Jonathan D. 2012. Universals in Comparative Morphology: Suppletion,
Superlatives, and the Structure of Words. Cambridge, MA: MIT Press.
Borsley, Robert D., Maggie Tallerman, and David Willis. 2007. The Syntax of Welsh.
Bühler, Karl. 1934. Sprachtheorie: Die Darstellungsfunktion der Sprache. Jena,
Germany: Fischer.
Chandralal, Dileep. 2010. Sinhala. Amsterdam: John Benjamins.
In Studies in Linguistics, University of Siena CISCL Working Papers 2, edited by
Vincenzo Moscati, pp. 42–58. Cambridge, MA: MIT Press.
Den Dikken, Marcel. 2010. “On the Functional Structure of Locative and Directional
PPs.” In The Cartography of Syntactic Structures, Vol. 6, edited by Guglielmo
Cinque and Luigi Rizzi, pp. 74–126. Oxford: Oxford University Press.
Diessel, Holger. 1999. Demonstratives: Form, Function, and Grammaticalization.
Dixon, Robert M. W. 1981. “Wargamay.” In Handbook of Australian Languages 2, ed-
ited by Robert M. W. Dixon and Barry J. Blake, pp. xxiv–144. Amsterdam: John
Benjamins.
Dixon, Robert M. W. 1988. A Grammar of Boumaa Fijian. Chicago and London: University
of Chicago Press.
Donohue, Mark. 1999. A Grammar of Tukang Besi. Berlin: De Gruyter.
Dryer, Matthew S. and Martin Haspelmath (eds.). 2013. The World Atlas of Language
Structures Online. Leipzig, Germany: Max Planck Institute for Evolutionary
Anthropology. [online] Available at http://wals.info. Accessed 2015-03-06.
Everett, Daniel L. 1986. “Pirahã.” In Handbook of Amazonian Languages, Vol. 1, edited by
Desmond C. Derbyshire and Geoffrey K. Pullum, pp. 200–326. Berlin: De Gruyter.
Everett, Daniel L. and Barbara Kern. 1997. Wari’: The Pacaas Novos Language of Western
Brazil. London and New York: Routledge.
Facundes, Sidney da Silva. 2000. The Language of the Apurinã People of Brazil (Maipure/
Arawak). Doctoral dissertation, SUNY.
Fillmore, Charles J. 1971. Deixis Lectures (Santa Cruz). [online] Available at www-
personal.umich.edu/~jlawler/Deixis.html.
Fillmore, Charles J. 1982. “Towards a Descriptive Framework for Spatial Deixis.” In
Speech, Place, and Action: Studies in Deixis and Related Topics, edited by Robert J.
Jarvella and Wolfgang Klein, pp. 31–59. New York: Wiley.
Fillmore, Charles J. 1997. Lectures on Deixis. Stanford, CA: Center for the Study of
Language and Information (CSLI) Publications.
Finegan, Edward. 2013. Language: Its Structure and Use, 7th ed. Stamford, CT: Cencage
Learning.
Foley, William A. 1991. The Yimas Language of New Guinea. Stanford, CA: Stanford
University Press.
Fortis, Jean-Michel and Benjamin Fagard. 2010. Space and Language, Part V: Deixis.
Leipzig Summer School in Typology. [online] Available at htl.linguist.univ-
paris-diderot.fr/fortis/leipzig/5.DEIXIS.pdf.
Gaffiot, Félix. 2016 [1934]. Dictionnaire Illustré Latin-Français, revised and augmented
by Gérard Gréco, with Mark De Wilde, Bernard Maréchal, Katsuhiko Ôkubo.

[online] Available at http://gerardgreco.free.fr/IMG/pdf/Gaffiot_2016_-_
komarov.pdf.
Geraghty, Paul. 1976. “Fijian Prepositions.” Journal of the Polynesian Society 85 (4): pp.
507–520.
Harbour, Daniel. 2016. Impossible Persons. Cambridge, MA: MIT Press.
Harms, Phillip L. 1994. Epena Pedee Syntax. Dallas: SIL and University of Texas at
Arlington.
Himmelmann, Nikolaus. 1997. Deiktikon, Artikel, Nominalphrase: Zur Emergenz
syntaktischer Struktur. Tübingen, Germany: Max Niemeyer.
Himmelmann, Nikolaus. 2001. “Articles.” In Language Typology and Language
Universals, edited by Martin Haspelmath, Ekkehard König, Wulf Oesterreicher,
and Wolfgang Raible, pp. 831–841. Berlin: De Gruyter.
Holes, Clive. 1990. Gulf Arabic. London and New York: Routledge.
Hualde, José I. and Jon Ortiz de Urbina. 2003. A Grammar of Basque. Berlin: Walter de
Gruyter.
Hualde, JoséError! Reference source not found. I. 1992. Catalan. London and
New York: Routledge.
Imai, Shingo. 2003. Spatial Deixis. Doctoral dissertation, SUNY.
Janssen, Theo. 2004. “Deixis and Reference.” In Morphology: An International Handbook
on Inflection and Word-Formation, Vol. 2, edited by G.E. Booij, Christian Lehmann,
and Joachim Mugdan, pp. 983–997. Berlin: De Gruyter.
Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht, The
Netherlands: Foris.
Kruspe, Nicole. 2004. A Grammar of Semelai. Cambridge: Cambridge University Press.
Lander, Eric. 2015. The Nanosyntax of the Northwest Germanic Reinforced Demonstrative.
Lander, Eric and Liliane Haegeman. 2016. “The Nanosyntax of Spatial Deixis.” Studia
Linguistica: DOI 10.1111/stul.12061.
Leu, Tom. 2015. The Architecture of Determiners. New York: Oxford University Press.
Levinson, Stephen. 1983. Pragmatics. Cambridge: Cambridge University Press.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
McGregor, William. 1990. A Functional Grammar of Gooniyandi. Amsterdam: John
Benjamins.
Meeuwis, Michael and Koen Stroeken. 2012. “Non- Situational Functions of
Demonstrative Noun Phrases in Lingala (Bantu).” Pragmatics 22: pp. 147–166.
Montler, Timothy. 2007. “Klallam Demonstratives.” University of British Columbia
Working Papers in Linguistics 20: pp. 409–425.
Mous, Maarten. 1993. A Grammar of Iraqw. Hamburg: Helmut Buske Verlag.
Nedjalkov, Igor. 1997. Evenki. London and New York: Routledge.
Ng, Eve C. 2002. Demonstrative Words in the Algonquian Language Passamaquoddy: A
Descriptive and Grammaticalization Analysis. Doctoral dissertation, SUNY.
Okombo, Duncan O. and John H. Habwe. 2007. “Pragmatic Strategies in the Use of
Kiswahili Demonstratives.” Swahili Forum 14: pp. 81–88.
Otaka, Sumio. 1987. Kataloniago no Bunpo [Catalan Grammar]. Tokyo.
Poulos, George. 1990. A Linguistic Analysis of Venda. Pretoria: Via Afrika Limited.
Rauh, Gisa. 1983. “Aspects of Deixis.” In Essays on Deixis, edited by Gisa Rauh, pp. 6–
60. Tübingen, Germany: Niemeyer.
148
Redden, James E. 1980. A Descriptive Grammar of Ewondo. Carbondale, IL: Department

of Linguistics.
Ross, Malcolm. 2007. “Talking About Space: Terms of Location and Direction.” The
Lexicon of Proto Oceanic: The Culture and Environment of Ancestral Oceanic Society,
2. The Physical Environment, edited by Malcolm Ross, Andrew Pawley, and
Meredith Osmond, pp. 229–294. Canberra, Australia: ANU E Press.
Lecture series presented at CRISSP, Hogeschool-Universiteit Brussel. March 18,
20, 22.
Svenonius, Peter. 2010. “Spatial Prepositions in English.” In The Cartography of
Syntactic Structures, Vol. 6, edited by Guglielmo Cinque and Luigi Rizzi, pp. 127–
160. Oxford: Oxford University Press.
Taylor, Charles V. 1985. Nkore-Kiga. London: Croom Helm.
Tryon, Darrell T. 1970. Conversational Tahitian. Berkeley and Los Angeles: University
of California Press.
Tumbahang, Govinda Bahadur. 2007. A Descriptive Grammar of Chhatthare Limbu.
Doctoral dissertation, Social Inclusion Research Fund/SNV Nepal.
Werner, Heinrich. 1997. Die ketische Sprache. Wiesbaden, Germany: Otto Harrassowitz.
Yajima, Hideo. 1984. Gendai Burugaria-go Nyumon [Introduction to Modern Bulgarian].
Tokyo: Tairyusha.
Yip, Po-Ching and Don Rimmington. 2004. Chinese: A Comprehensive Grammar. London
and New York: Routledge.

CHAPTER 6
Decomposing Complementizers
The Functional Sequence of French, Modern Greek,
Serbo-Croatian, and Bulgarian Complementizers.
LENA BAUNAZ
6.1 INTRODUCTION AND OUTLINE OF THE CHAPTER
It has been reported in the literature (Roussou 2010; Baunaz 2015, 2016a,
among others) that complementizers vary crosslinguistically as to what in-
formation they lexicalize. This variation is illustrated in Table 6.1: French
(Fr) que, Serbo-Croatian (SC) da, and Bulgarian (Bg) če do not have a single
analogue in Modern Greek (MG), but correspond to a number of different
instantiations: pu introduces epistemic factive ‘remember’-type complements
(1a), and oti introduces nonfactive ‘say’-type complements (2a), whereas Fr [(1b),
(2b)], SC [(1c), (2c)], and Bg [(1d), (2d)] display the same complementizer—
que/da/če—in these two contexts. As a first approximation, we conclude that
Fr que, SC da, and Bg če unite properties of multiple items in MG. This pattern
.This chapter is a modified and extended version of Baunaz (2016b), both in

data and analysis. I would like to thank Boban Arsenijević, Metin Bagriacik, Alexia
Ioannidou, Ciro Greco, Liliane Haegeman, Illiana Krapova, Eric Lander, Teodora
Radeva-Bork, Anna Roussou, Vesela Simoneva, Tomislav Sočanac, Andrew Weir, as
well as the audience of the SLE 44 workshop Nanosyntax and NELS 46, for judgments
and/or comments on previous versions of this chapter. I am also very grateful to the
two reviewers of this volume for various insightful comments helping improve this
paper. All remaining errors are my own. This research has been partly supported
by the Swiss National Foundation (grant: PA00P1_145313), as well as FWO project
2009-Odysseus-Haegeman-G091409.
150
Table 6.1 DECL ARATIVE COMPLEMENTIZERS IN MG, FR ,

SC, AND BG, 1.0
Factive Verbs Nonfactive Verbs
MG pu oti
Fr que
SC da
Bg če
(and similar ones noted in the literature) raises the question of how these
different properties get spelled out as morphemes crosslinguistically.
(1) a. Thimame pu se sinandisa stin aghora. (MG)

remember.1sg that you met.1sg at.the market
[Roussou 1992, 125, (3)]
b. Je me rappelle que je t’ai rencontré
I 1sg.acc remember that I you.have meet.past.part
au marché. (Fr)
at.the market
c. Sjećam se da sam te upoznao
remember1.sg that aux.past.1.sg you meet.past.part
na tržnici. (SC)
on the.market
d. Pomnja, če te sreštnax na pazara. (Bg)
remember1.sg that you meet.past.part on the.market
‘I remember that I met you at the market.’
(2) a. O Pavlos ipe oti i Roxani efije. (MG)

the Paul said.3sg that the Roxanne left.3sg
‘Paul said that Roxanne left.’ [Giannakidou 2009, 1886, (7)]
b. Paul a dit qu’il a vu Marie. (Fr)
Paul has said that.he has seen Marie
c. Pavao je rekao da je
Paul aux.past.3sg say.past.part that aux.past.3sg
vidio Mariju. (SC)
see.past.part Mary
d. Pavel kaza, če e vidjal Mary. (Bg)
Paul said that he saw Mary
‘Paul said that he saw Mary’

The facts summed up in Table 6.1 might lead one to think that Fr, SC, and
Bg have two homophonous complementizers; for instance for Fr we could
postulate two ques, que1 and que2, where que1 appears under factive verbs and
que2 under nonfactive verbs, each corresponding to the more specialized MG
complementizers, that is, que1 would correspond to pu and que2 to oti. Rather
than analyzing these overlaps in terms of homophony I consider phonolog-
ically identical complementizers to be cases of syncretism (see Baunaz and
Lander in Chapter 1, Section 1.3.3.1).
Before we try to assess the feasibility of decomposing complementizers into
articulated structures, we need to complete the provisional picture in Table 6.1.
First, MG, SC, and Bg factives can select two different complementizers. Semi-
factives can optionally select oti in MG (see Roussou 2010 and references cited
there), (3a). Emotive factives can select only pu (3b).
(3) a. Thimame oti/pu dhjavaze poli. (MG)

remember.1sg that read-3sg much
‘I remember that he used to read a lot./I remember him reading a lot.’
[Roussou 2010, 590, (17)]
b. O Pavlos lipate *oti/pu efije i Roxani. (MG)
the Paul is.sad.3sg that left.3sg the Roxanne
‘Paul regrets that Roxanne left.’ [Giannakidou 2015, (7)]
In addition to da, SC exhibits a second complementizer, što, that exclusively

appears under emotive factive verbs, verbs under which da is impossible (4a).
In this context, (some) speakers from Croatia (henceforth Hr) have the option
of using da (Tomislav Sočanac, personal communication, hereafter p.c.). Note
also that some (Serbian, henceforth Sr) speakers can optionally select što
under semi-factives (4b) (Boban Arsenijevic, p.c.). In (4) and subsequent, the
% diacritic indicates speaker variation plus the variety in question.
(4) a. Žalim što/%Hrda si povrijedio Ivana. (SC)

regret.1sg that aux.past.2sg hurt.past.part John
‘I regret that you hurt John.’
b. Znam da/%Sršto si bio u Gentu.
know.1sg that aux.past.2sg been in Ghent
‘I know that you’ve been to Ghent./I’m familiar with the fact that
you’ve been to Ghent.’
Če is not the only declarative complementizer in Bg: Deto is used under emo-

tive factive verbs (alternating with če in this context; see Krapova 2010), (5a).
As with SC, there is also speaker variation in Bg: For some speakers, semi-
factive verbs can optionally select deto (5b).
De c o m p o s i n g C o m p l e m e n t i z e r s [ 151 ]
152
(5) a. Naistina sǎžljavam, deto/če ne otedlix poveče vnimanie na (Bg)

postrojkata.
‘I really regret that I did not devote greater attention to the construction.’
[Krapova 2010, 26, (56a)]
b. Pomnja, %deto/če te sreštnax na pazara. (Bg)
‘I remember that I met you at the market/meeting you at the market’
Second, MG, SC, and Bg have a distinct mood particle marking the subjunc-
tive mood under desiderative ‘wish’-type verbs (6).1
(6) a. Thelo na fiji o Kostas. (MG)

want.1sg subj leave.3sg the Kostas
b. Želim da Ivan ode. (SC)
want.1sg subj John leave.3sg
‘I want John to leave.’
c. Iskam da ostanat decata. (Bg)
want.1sg that stay.3pl children
‘I want the children to stay.’ [Krapova 1998, 86, (24b)]
In all these contexts, Fr predicates select que throughout (7). In addition, sub-
junctive mood is expressed by means of verbal morphology, as seen in (7b).
(7) a. Je me rappelle que Marie est partie. (Fr)

I 1sg.acc remember that M. is.ind left
‘I remember that Mary left.’
b. Je regrette/veux que Marie part-e.
I regret/want that M. leave-subj
‘I regret that Mary leaves/I want Mary to leave.’
In view of these observations, Table 6.1 can be modified as Table 6.2, where
three columns have been added (emotive factive/semi-factive/desideratives),
and more language-internal variations are considered.
1. There is a debate concerning the status of SC da: Either there are two homopho-
nous items: ‘declarative’ da and modal da, or only one da. See Todorovic (2012) and
references cited there for details.

Table 6.2 DECL ARATIVE COMPLEMENTIZERS IN FR , SC, BG, AND MG, 2.0
Emotive Factive Semi-Factive Nonfactive Desiderative

Fr que que que que
SC da
% Hr.
da
da da
što % Sr.
što
Bg če če
če da
deto %
deto
MG pu
pu oti na
oti
6.1.1 Proposals
In this chapter, I make four major claims: (i) the complementizer morpheme is

composed of features that are hierarchically ordered according to a functional
sequence (fseq) (see Baunaz 2015, 2016a; Baunaz and Lander to appear)2;
(ii) the complementizer morpheme lexicalizes structures of different sizes;
(iii) the distribution of complementizers is governed by veridicality (see Baunaz
2015, 2016a); (iv) the complementizer morpheme is syntactically active. The
basic template for the internal structure of complementizers is that in (8).
(8) F4 > F3 > F2 > F1
Evidence in favor of (8) comes from crosslinguistic patterns of syncretism and

Relativized Minimality (RM) (Starke 2001; Rizzi 2004).
6.1.2 Syncretism and Relativized Minimality
The nanosyntactic approach to syncretism (Caha 2009) is based on the

idea that features are additive. As a consequence, a fseq can realize sev-
eral structures (see Baunaz and Lander in Chapter 1, [21–22]; Lander and
Haegeman in Chapter 5, [section 5.2]). The structures in (9) can each be
matched to a phonological form. So /na/matches F1P in (9a), /oti/matches
the combination of F1 with F2 (in this order), namely F2P in (9b), and so on. We
speak of syncretism when two (or more) structures are associated to the same
phonological form, as in (9c) and (9d). A syncretism thus represents a one-to-
many mapping between phonological form and syntactic structure.
2. For syntactic complexity inside of complementizers, see also Leu (2015), who
argues that German dass is two complementizer heads: d-and -ass.
154
(9) a. [F1P F1 ] => /na/

b. [F2P F2 [F1P F1]] => /oti/
c. [F3P F3 [F2P F2 [F1P F1 ]]] => /pu/
d. [F4P F4 [F3P F3 [F2P F2 [F1P F1 ]]]] => /pu/
As has become clear from Chapter 1 (Sections 1.3.3.1 and 1.3.3.2, and references
cited there), what has become known as the *ABA generalization plays a crucial
role in the study and interpretation of syncretism and for the identification
of the features and their relative position in the fseq. It states that syncretism
only targets contiguous structural layers, that is, ABA patterns are ruled out.
Consider Table 6.3. Table 6.3 has been stripped from its Sr versus Hr and Bg
variations (i.e. % in Table 6.2) and the columns have been relabeled as in (9),
for simplicity’s sake. See Section 6.3 for a more thorough account.
Table 6.3 DECL ARATIVE COMPLEMENTIZERS IN FR , SC, BG,

AND MG (SIMPLIFIED VERSION)
F4 F3 F2 F1
Fr que que que que
SC što da da da
Bg deto če če da
MG pu pu oti na
The syncretisms displayed by Fr, SC, Bg, and MG in Table 6.3 suggest that
the adjacency relations are as displayed in (10).
(10) F4 | F3 | F2 | F1
The linear ordering in (10) is the only one that can capture these facts without
any *ABA patterns in MG, SC, and Bg (and Fr): što, deto, and oti do not inter-
vene between the different realizations of da, če, and pu. Bulgarian shows that
F3 and F2 must be adjacent and MG that F4 and F3 are contiguous.
What the observed syncretisms and the *ABA theorem leave open is which
hierarchical order in (11) is appropriate:
(11) a. F4 > F3 > F2 > F1

b. F1 > F2 > F3 > F4
The strategies I adopt here to detect the fine- grained structure of

complementizers are based on (i) syncretism (see Chapter 1, Section 1.3.3.1)
and (ii) a development of featural RM (Starke 2001; Rizzi 2004, 2013;
Haegeman 2010; Haegeman and Ürögdi 2010a, 2010b, among others). In par-
ticular, (ii) is based on the following hypothesis: The less semantically marked

an item is, the less structure it has (Starke 2001). In other words, the item
with the smallest structure is semantically unmarked. Depending on which
end of the F4 | F3 | F2 | F1 spectrum the item with less semantics is syncretic
with, we should be able to identify the “smaller” end of the hierarchy.
6.2 STRONG VERSUS WEAK PRESUPPOSITION
In Section 6.1, we have seen that MG semi-factives can select two different
complementizers, and so do SC and Bg semi-factives and emotive factives.
In MG semi-factives may indeed select oti (see Table 6.2), but they do so only
when they involve weak presupposition; when they express strong presupposition
they select pu (see Roussou 2010, and references cited there). The use of oti in
(12) somehow indicates some factivity weakening of the main predicate. The use
of pu, however, commits the speaker to the truth of the embedded proposition.
In (12), the continuation but he is wrong, because I didn’t forces a reading in which
the speaker’s point of view about the truth of the embedded proposition needs
to be taken into account. In the context in (12), only the oti-version is felicitous:
(12) a. O Janis paraponethike oti ton ksexasa; ala

the John complained.3sg that him forgot.1sg but
kani lathos: dhen ton ksexasa.
make.3sg mistake not him forgot.1sg
b. O Janis paraponethike pu ton ksexasa; # ala
the John complained.3sg that him forgot.1sg but
kani lathos: dhen ton ksexasa.
make.3sg mistake not him forgot.1sg
‘John complained that I forgot him; but he is wrong, because I didn’t.’
[Giannakidou 2015, (92)]
Baunaz (2016a) claims that in situations like (12), oti-selection involves a CP that
expresses truth commitment by the subject exclusively (vs. speaker), that is, the
truth of the embedded proposition is relative to the speaker. (Some) MG semi-
factives are then able to shift their meaning from strong to weak presupposition,
that is, these verbs are no longer semi-factive, but range over emotive factives.
As shown in Section 6.1 [see (4)], semi-factives embed da in SC. Some Sr
speakers can also use što in this context. When da is embedded, the embedded
proposition must be true; when što is embedded, the factivity presupposition
is weakened. Emotive factives embed što in SC. Some Hr speakers can also use
da in this context.3 The choice of complementizer coincides with presupposi-
tion strength, again: Da involves weak presupposition and što involves strong
3. Da here is register-bound: It appears in higher and more formal registers (Boban
Arsenijevic and Tomislav Sočanac, p.c.).
156
presupposition. In case što is selected with semi-factives, or da with emotive

factives, factivity is weakened (Boban Arsenijevic and Tomislav Sočanac, p.c.). It
is important to note that when the verb triggers weak presupposition, factivity
is not canceled out, as the embedded sentence needs to be considered true in the
discourse: “the speaker speaks on the condition that it’s true, but does not commit
to it. The truth is based on the information contributed by someone else (the
hearer most typically), not the speaker.” (Baunaz 2016a, 71; Boban Arsenijevic,
p.c.), that is, the embedded proposition is true relative to the speaker.
Bulgarian is very similar to SC in that respect. Recall that Bg has two declar-
ative complementizers: deto and če. Some (factive) verbs appear to optionally
select both. For instance, Krapova (2010) reports that a (sub)type of emotive
factives can select deto as well as če. Some (other) speakers also optionally
accept deto with semi-factives (especially ‘remember’) (Teodora Radeva-Bork,
p.c.). As is the case for Sr and Hr što versus da alternations, the choice of če
versus deto makes a difference for both types of speakers: če involves a weak
presupposition, deto a strong presupposition.
Our findings have been summed up as in Table 6.2, repeated here as
Table 6.4 for conveninence.
Table 6.4 DECL ARATIVE COMPLEMENTIZERS IN FR , SC, BG, AND MG, 2.0
Emotive Factive Semi-Factive Nonfactive Desiderative

Fr que que que que
SC % Hr.
da da
da da
što % Sr.
što
Bg če če
če da
deto %
deto
MG pu
pu oti na
oti
As it stands, the presentation in Table 6.4 does not allow us to make a distinc-
tion between weak versus strong presuppositions in the factive columns. What
SC, Bg, and MG tell us is that (some) semi-factives can shift their meaning to a
weak presupposition interpretation (i.e. these verbs are no longer semi-factive,
but must be ranged as emotive factive instead), whereas some emotive factives
may shift their meaning to a strong presupposition interpretation (i.e. these
verbs are no longer emotive factive, but semi-factive), that is, verbs can switch
classes. The form of the complementizer indicates this switch.
The weak versus strong presupposition discussed here recalls the distinc-
tion relative versus strong veridicality discussed in Baunaz and Puskás (2014)
for Fr. I propose that the notion of veridicality can handle the apparent dual
nature of these verbs insightfully. In other words, Table 6.4 can be refined in
light of the notion of veridicality. This will require our columns to be relabeled.

6.3 VERIDICALITY
On the basis of Fr, Baunaz and Puskás (2014) have argued that the classifi-
cation of verbs involved in the selection of the complementizers in Table 6.4
should be refined in terms of veridicality, rather than factivity. Giannakidou
(2009) defines the notion of veridicality as in (13), that is, an embedded prop-
osition has to be true for at least one individual (the subject of the main verb,
the speaker, or both), in all the worlds of a relevant model.
(13) Veridicality (Giannakidou 2009, 1889)

A propositional operator F is veridical iff from the truth of Fp we
can infer that p is true according to some individual x (i.e. in some
individual x’s epistemic model).
Baunaz and Puskás (2014) investigate the notion of “some individual” in (13)
to understand how it applies to emotive factive, semi-factive, and nonfactive
complements. They show that with semi-factives, the embedded proposition
must be true, both from the subject’s and the speaker’s point of view, as shown
by the continuation in (14):
(14) Paul découvre que Marie est partie, # mais c’est faux (Marie n’est pas partie).
Paul finds out that Marie is left, but it is false (Marie NE has not left
‘Paul found out that Marie left, but it is false (Marie didn’t leave).’
With emotive factives, the embedded proposition must be true from the point
of view of the subject, but not (necessarily) from that of the speaker, as shown
by the continuation in (15):
(15) Jean est persuadé qu’il pleut, et il regrette qu’il pleuve. (but of course
Jean is convinced that it rains, and he regrets that it rain.SUBJ
it’s not raining)
‘Jean is convinced that it’s raining, and he regrets that it’s raining.’
[from Schlenker 2005, 27, fn.14, (i)]
Baunaz and Puskás (2014, 245) “observe that the shift in the relevant epi-
stemic model (i.e. of the Speaker or of the Subject) allows to make different
inferences with respect to the truth of the embedded proposition, i.e. the
veridical status of regretter ‘regret’ ( . . . ) is relative,” as opposed to that of
semi-factives.
Finally, negating the complement of nonfactives (verbs of saying,
desideratives) does not yield contradictory statements, that is, these verbs
do not infer the truth of their complement, by neither the subject nor the
speaker (16).
158
(16) She is convinced that it’s raining, and . . ..

a. veut/suggère qu’il pleuve. (But of course it’s not raining!)
wants/suggests that it rain.subj
b. dit qu’il pleut. (But of course it’s not raining!)
says that it rains.ind
They also observe that there are veridical verbs in Fr that can switch from one
type of veridicality to another, as illustrated in (17). See also Baunaz (2016a).
(17) Pierre wrongly believes that Marie is getting married, and...

a. # réalise/comprend qu’elle ne veut pas rester célibataire
‘realizes/understands that she doesn’t want.ind to stay single’
b. regrette/comprend qu’elle ne veuille pas rester pas célibataire
‘regrets/understands that she doesn’t want.subj to stay single’
The predicates in (17) are all veridical in Giannakidou’s sense, as both (17a)
and (17b) are true according to some individual x, namely Pierre, the speaker,
or both. In addition to the subject, p must also be true for the speaker, as
the continuation shows in (17a). In contrast, if regrette/comprend in (17b)
presupposes that x believes that p: p is true, but only with regard to the ep-
istemic model of Pierre [and not (necessarily) with regard to the epistemic
model of the speaker] (Schlenker 2005). So “the shift in the relevant ep-
istemic model (that is, of the Speaker or of the Subject) [permits] to make
different inferences with respect to the truth of the embedded proposition”
(Baunaz and Puskás 2014, 246). The two uses of comprendre involve different
features (strongly veridical vs. relatively veridical). In that sense there must
be different verbal realizations of the same phonological form (= syncretism).
To sum up, the distinction of the weak versus strong presupposition
is similar to the distinction relatively versus strongly veridical: (i) Strong
presuppositional verbs are strongly veridical, in that they require that the
embedded proposition be true from the point of view of the speaker and from
that of the subject; (ii) weak presuppositional verbs are relatively veridical in
that they require that the embedded proposition be true from the point of view
of the subject, but not (necessarily) from that of the speaker (see Schlenker
2005; see also Giannakidou 1998 for MG and Tóth 2008 for Hungarian);
(iii) verbs of saying and desideratives do not embed propositions whose truth
must be inferred by the subject or by the speaker: They are nonveridical.
I claim that these distinctions also apply to the seemingly ambiguous verbs
discussed in Section 6.2. Semi-factives can have two readings in MG and some
Bg and SC: strongly and relatively veridical. Emotive factives can also involve
two readings in SC and for all Bg: strongly and relatively veridical. Just as
for Fr comprendre in (17), I claim that because the verbal realizations allowing
shift of meaning have the same phonological form, they must be syncretic

items. Crucially this meaning shift correlates with the form of the comple-
mentizer in Bg (deto vs. če), suggesting that it is veridicality that selects the
type of complementizer, as seen in Table 6.5.4
Table 6.5 REL ATIVELY AND STRONGLY VERIDICAL

COMPLEMENTIZERS IN MG, BG, SC, AND FR
Relatively Veridical Strongly Veridical
pu
MG pu
oti
Bg če deto
% Hr.
da da
SC
% Sr.
što što
Fr que que
The syncretism patterns of MG and Bg tell us that the relatively and strongly
veridical columns should be inverted: The relatively veridical and nonveridical
columns must be contiguous (see MG oti and Bulgarian če) to avoid giving rise
to an ABA violation. The syncretism patterns of SC and MG also tell us that
the strongly veridical and relatively veridical columns should be adjacent too
(see SC da, što, and MG pu). In addition, note that the nonveridical column
should be split into two columns, one hosting complementizers selected by
verbs of saying (taking the indicative mood) and another selected by desider-
ative verbs [taking the subjunctive mood in MG, Bg, SC, and Fr; see Sočanac
(2017) for SC]. Nonveridical verbs taking the indicative mood select če in Bg,
whereas nonveridical verbs selecting the subjunctive mood select da. Če being
syncretic with relatively veridical complementizers, the two columns should
be adjacent. This gives us Table 6.6 and the linear ordering in (18):
(18) strongly veridical | relatively veridical | nonveridicalind | nonveridicalsubj
From now on, I call the complementizers selected by strongly/relatively

and nonveridical verbs strongly/relatively and nonveridical complementizers,
respectively.
The syncretism patterns do not give us any insight into the ultimate hi-
erarchical relation of the features in (18), though: It only helps us deduce
4. Recall that when Serbian speakers embed što under ‘remember’-type verbs, factivity
is weakened, i.e. the matrix verb is interpreted as a relatively veridical verb [see Section
6.2. See also (4) for relevant examples]. When SC speakers embed što under ‘regret’-
type verbs, factivity is strengthened, i.e. the matrix verb is interpreted as a strongly
veridical verb. This is why the % is inverted in Tables 6.5 and 6.6 vs. Tables 6.4 and 6.2.
160
Table 6.6 DECL ARATIVE COMPLEMENTIZERS MG, BG, SC, AND FR , 3.0
Strongly Veridical Relatively Veridical NonveridicalIND NonveridicalSUBJ

pu
MG pu oti na
oti
Bg deto če če da
da % Hr.
da
SC da da
što % Sr.
što
Fr que que que que
a linear order (see Section 6.1.2). In Section 6.4 I argue that nonveridical
complementizers occupy the right-hand edge of our fseq (and is structurally
quite small), and that strongly veridical complementizers occupy its left-hand
edge (and is structurally quite big). My proposal is based on featural RM.
6.4 VERIDICAL ISLANDS (FORMERLY KNOWN AS FACTIVE ISLANDS)
Deconstructing the meaning of wh-phrases escaping weak islands (WIs), Starke

(2001) observes that they are always interpreted with wide scope of existential
presupposition [= what he calls specificity, β in (19) and (20)]. He observes that
when extraction out of WIs is tolerated, the moved wh must have something
more than its intervener (which in that case does not intervene at all), that is,
β in (19b); when extraction is blocked, it is either because the intervener has
something more than the moved wh, that is, it is bigger because β is involved,
as in (20a), or it is because the intervener and the wh-extractee share the same
feature(s), that is, they are of the same size [(19a), (20b)]. When extraction is
not blocked, it is because the intervener has something less than the moved
wh, that is, it is smaller because β is not involved (19b) [the examples in (19) are
from Haegeman and Ürögdi 2010a, 126, (23a) and (23b)].
(19) a. * How do you wonder whether John will =α*i . . .αj . . .αi

solve the problem?
b. ? Which problem do you wonder whether = αβ. . .α. . .αβ
John will solve?
(20) a. ?? Who do you wonder which boy likes? = *α. . .αβ. . .α
b. ?* Which girls do you wonder which boy likes? = *αβ. . .αβ. . .αβ
In Starke’s terms, this means that a quantifier α involving β, that is, a specific
wh-phrase like, for instance, which in (19b) and (20b), is semantically marked
(it is “specific”). The more semantically marked, the bigger (the wh-phrase

contains α and β); the less semantically marked, the smaller (the wh-phrase
contains only α).
Drawing on observations from syncretism patterns in Sections 6.2 and 6.3,
we know that complementizers come in various sizes [see (18)], but we do not
know the hierarchical ordering of their features. Featural RM can help us de-
cide which featural hierarchy is correct when it comes to the internal structure
of complementizers: “bigger” complementizers will block movement of wh-
phrases, “smaller” complementizers will allow wh-phrases to move across them.
Because most of the verbs involved in Section 6.1 are generally claimed to
induce weak (factive) islands in languages like English (Melvold 1991, among
others), this section is thus based on long-distance wh-extraction across fac-
tive and nonfactive domains. Recall that (non)factivity should be considered
as (non)veridicality (Section 6.3). The relevant (non)veridical selecting verbs
are exemplified in Table 6.7 for MG, SC, Bg, and Fr.
I propose that the verbs in Table 6.7 select for complementizers of different
“sizes”, or, put differently these verbs create different syntactic domains.
Thinking in terms of cumulative layers, this means that complementizers
with the smallest feature compositions have the least semantics and that
complementizers with the biggest feature compositions have the most seman-
tics. If that is correct, the size of the structure that complementizers realizes
should play a role in strong, weak, or no island configurations, just like the
size of the moved element should also be relevant. In the remainder of this
section, long-distance wh-extractions in MG, SC, Bg, and Fr are closely studied
in the configuration in (21), with the selecting contexts in Table 6.7.
(21) [argument wh/adjunctwh] . . . Comp . . .
Table 6.7 SOME (NON)VERIDICAL VERBS IN MG, SC, BG, AND FR
English Translation MG SC Bg Fr
Strongly ‘remember’ thimame sjetiti se pomnja se rappeler
veridical ‘regret’ %
žaliti sâžaljavam
‘understand’ comprendre
Relatively ‘remember’ thimame %

sjetiti se pomnja regretter
veridical ‘regret’ lipame žaliti sâžaljavam
‘understand’ comprendre
Nonveridical ‘say’ leo reći kazvam dire

‘want’ thelo željeti iskam vouloir
Sections 6.4.1–6.4.3 introduce extraction configurations in Fr, MG, SC, and

Bg. Section 6.4.4 discusses an apparent inconsistency in SC with long-distance
extraction out of što-clauses.
162
6.4.1 Long-d istance extraction across veridical domains
In MG, long-distance wh-extractions across relatively veridical domains (with

pu and oti) result in WI-type violations (22) and (23). Long-distance extraction
across strongly veridical domains (with pu) yields stronger violations, even
when objects are involved [(22) vs. (24)].5
(22) a. Pjoni thimase oti sinandises ti? (MG)

who remember.2sg that met.2sg
‘Whom do you remember that you met?’
b. ?? Potei thimase oti sinandises Maria ti?
when remember.2sg that met.2sg Mary
‘When do you remember that you met Mary?’
(23) a. Pjoni lipase pu pligoses ti? (MG)

who be.sorry.2sg that hurt.2sg
‘Whom do you regret that you hurt?’
b. * Potei lipose pu efijes ti?
when be.sorry.2sg that left.2sg
‘When do you regret that you left?’ [Giannakidou 1998, 220, (70)]
(24) a. * Pjoni thimase pu sinandises ti? (MG)

who remember.2sg that met.2sg
b. * Potei thimase pu sinandises ti Maria ti?

when remember.2sg that met.2sg the Maria
* ‘When do you remember that you met [Roussou 1992, 126, (7)]
Maria?’
In SC spoken in Croatia, long-distance wh-extractions across the relatively ve-

ridical complementizer da is possible only with arguments, (25a), and is even
preferred if koga ‘who’ is D-linked (Tomislav Sočanac, p.c.). Adjunct extraction
is never possible (25b).6 Long-distance extraction across the strongly veridical
complementizer da in SC results in stronger restrictions (even with D-linked
objects) (26).
5. The French data are based on Baunaz and Puskás (2014) and Baunaz (2015). The
MG and SC data are based on Baunaz (2014, 2015). The reader is referred to these
papers for discussion about focus and topic extractions in Fr, SC, and MG, where it is
shown that focus extraction behaves like wh-extraction, but not topic extraction.
6. The SC što-variants are discussed in Section 6.4.4.

(25) a. (?) Kogai žališ da si povrijedio ti? (SC)
who regret.2sg that aux.past.2sg hurt.past.part
‘Whom do you regret that you hurt?’
b. * Kadi žališ da si otišao ti?.

when regret.2sg that aux.past.2sg leave.past.part
‘When do you regret that you left?’
(26) a. ?? Kogai se sjećas da si upoznao ti? (SC)

who remember.2sg that aux.past.2sg meet.past.part
b. * Kadi se sjećas da si upoznao Mariju ti?

when remember.2sg that aux.past.2sg meet.past.part Maria
* ‘When do you remember that you met Maria?’
Recall that in Bg, the form of the complementizer distinguishes between rela-
tively and strongly veridical domains: Strongly veridical verbs select deto, rel-
atively veridical verbs embed če. Relatively veridical če gives rise to WIs (27)
and (28); when strongly veridical deto is used, extraction is (almost) never
possible (or degraded, compared with extraction across če in similar contexts),
no matter which type of verb is involved (29) and (30).
(27) a. Koji pomniš, če sreštna na pazara ti? (Bg)

who remember.2sg that met at market.the
b. * Kogai pomniš, če sreštna Maria na pazara ti?
when remember.2sg that met Maria at market.the
(28) a. Kakvoi sǎzaljavaš če Ivan e napravil ti? (Bg)

what regret.2sg that Ivan
aux.3sg done
b. * Kǎdei sǎzaljavaš če Ivan e otišǎl ti?
where regret.2sg that Ivan aux.3sg gone [Krapova 2010, (66)]
(29) a.?(?) Koji pomniš, deto sreštna na pazara ti? (Bg)

who remember.2sg that met at market.the
b. * Kogai pomniš, deto sreštna Maria na pazara ti?
When remember.2sg that met Maria at market.the
(30) a. * Kakvoi sǎzaljavaš deto Ivan e napravil ti? (Bg)

what regret.2sg that Ivan aux.3sg done
b. * Kâdei sǎzaljavaš deto Ivan e otišǎl ti?
where regret.2sg that Ivan aux.3sg gone [Krapova 2010, (65)]
In Fr only argument extractions of D-linked wh-phrases are allowed with rela-

tively veridical verbs (31) and (32). Aggressively non-D-linked qui diable ‘who
164
the hell’ is rejected in (31b). With strongly veridical verbs, long-distance wh-
extractions are never possible (33) and (34).
(31) a. Quel tournoii est-ce que Paul regrette que Roger ait vendu ti?
‘Which tournament does Paul regret that Roger sold?’
b * Qui diablei est-ce que Paul regrette que Roger puisse aimer ti?
‘Who the hell does Paul regret that Roger can love?’
c. * Commenti est-ce que Paul regrette que Roger ait gagné le tournoi ti?
‘How does Paul regret that Roger won the tournament?’
(32) a. Quelle photoi est-ce que Jean comprend/admet que Jules prenne ti?
‘Which picture does Jean understand/admit that Jules takes.subj?’
b. ?? Commenti est-ce que Jean comprend/admet que Jules prenne la photo ti?
‘How does Jean understand/admit that Jules takes.subj the picture?’
[Baunaz and Puskás 2014, 236 (6b), (7b)]
(33) a. ??/* Quelle photoi est-ce que Jean se rappelle que Jules prend ti?
‘Which picture does John remember that Jules takes?’
b. * Qui diablei est-ce que Jean se rappelle que Roger peut aimer ti?
‘Who the hell does John remember that Roger can love?’
c. * Commenti est-ce que Jean se rappelle que Jules prend la photo ti?
‘How does John remember that Jules takes the picture?’
(34) a. * Quelle photoi est-ce que Jean comprend/admet que Jules prend ti?
‘Which picture does Jean understand/admit that Jules takes.ind?’
b. * Commenti est-ce que Jean comprend/admet que Jules prend la photo ti?
‘How does Jean understand/admit that Jules takes.ind the picture?’
[Baunaz and Puskás 2014, 235 (4b), (5b)]
Wh-extraction out of veridical domains yields thus either strong or weak is-
lands. In the next section, wh-extraction out of nonveridical domains is studied.
6.4.2 Long-d istance extraction across nonveridical domains
Consider (35)– (42). In these examples, long- distance extraction across

nonveridical domains (including verbs of communication and desideratives) is
unproblematic, irrespective of the language and form of the complementizer,
that is, with both oti and na in MG [(35) and (36)], da in SC [(37) and (38)], če
and da in Bg [(39) and (40)] and que in Fr [(41) and (42)].

(35) a. Pjoni ipe o Pavlos oti idhe ti? (MG)
who said.3sg the Paul that saw.3sg
‘Whom did Paul say that he saw?’
b. Potei ipes oti idhes ton Pavlo ti?
who said.2sg that saw.2sg the Paul
‘When did you say that you saw Paul?’ [Giannakidou 1998, 220, (69)]
(36) a. Pjoni theli o Pavlos na dhi ti? (MG)

who want.3sg the Paul subj see.3sg
‘Whom does Paul want to see?’
b. Pjoni theli o Pavlos na fiji ti?
who want.3sg the Paul subj leave.3sg
‘When does Paul want to leave?’ [Giannakidou 1998, 220, (70)]
(37) a. Kogai je Pavao rekao da je

who aux.past.3sg Paul say.past.part that aux.past.3sg
vidio ti? (SC)
see.past.part
b. Kadi si rekao da si
when aux.past.2sg say.past.part that aux.past.2sg
vidio Pavla ti?
see.past.part Paul
‘When did you say that you saw Paul?’
(38) a. Kogai Ivan želi da vidi ti? (SC)

who Ivan want.3sg subj see.3sg
‘Whom does John want to see?’
b. Kadai Ivan želi da ode ti?
when Ivan want.3sg subj leave.3sg
‘When does Ivan want to leave?’
(39) a. Kogoi kaza Pavel, če e vidjal ti? (Bg)

who said Paul that aux.3sg see.part
b.? Kogai kaza, če si vidjal Pavel ti?
when said that aux.2sg see.part Paul
‘When did you say that you saw Pavel?’
166
(40) a. Koji iskash da dojde ti. (Bg)

who want.2sg subj come.3sg
‘Who do you want to come?’
b. Kogai iskash da dojda ti?
when want.2sg subj come.1sg
‘When do you want me to come?’
(41) a. Quel tournoii est-ce que Paul dit/observe que Roger a gagné ti? (Fr)
‘Which tournament does Paul say/observe that Roger won?’
b. Qui diablei est-ce que Paul dit/observe que Roger peut aimer ti?
‘Who the hell does Paul say/observe that Roger can love?’
c. Commenti est-ce que Paul dit/observe que Roger a gagné le tournoi ti?
‘How does Paul say/claim that Roger won the tournament?’
(42) a. Quelle raquettei est-ce que Paul préfère/désire/veut que Roger vende ti? (Fr)
‘Which tennis racket does Paul prefer/desire that Roger sells? /
Which tennis racket does Paul want him to sell’
b. Qui diablei est-ce que Paul préfère/désire/veut que Roger aime ti?
‘Who the hell does Paul prefer/desire that Roger love? /
Who the hell does Paul want Roger to love?’
c. Commenti est-ce que Paul préfère/désire/veut que Roger gagne le tournoi ti?
‘How does Paul prefer/desire that Roger win the tournament? /
How does Paul want Roger to win the tournament?’
6.4.3 Summary
Extraction facts tell us that there is something that uniformly blocks long dis-
tance wh-movements completely, partially, or not at all. Recall also that when
WIs are created, argument extraction is better with D-linked phrases. So the
constraint must be somehow related to the type of main clause predicate in-
volved. See Figure 6.1.
6.4.4 Serbo-C roatian što
For completeness’ sake it should be added that SC speakers can also embed što
in strongly veridical domains and that Sr što can also be selected under rela-
tively veridical verbs [see Section 6.3, fn.4, as well as example (4) and Table 6.5].
Arguments may be extracted from strongly veridical domains (43), but not from
relatively veridical domains in SC spoken by some Serbians (44).

Strongly SC da *argument Strong
veridical Bg deto *adjunct Island
MG pu
Fr que
Relatively SC %Hr. da argument Weak

veridical Bg če *adjunct Island
MG pu/oti
Fr que
Nonveridical
(verbs of
communication)
SC da
Bg če/da argument No
MG oti/na adjunct Island
Fr que
Nonveridical
(desideratives,etc.)
Figure 6.1 Veridicality, complementizers, and island effects
(43) a. (?) Kogai žališ što si pro povrijedio ti? (SC)
who regret.2sg that aux.past.2sg pro.acc hurt.past.part
‘Who do you regret that you hurt?’
b. * Kadi žališ što si otišao ti?

when regret.2sg that aux.past.2sg leave.past.part
‘When do you regret that you left?’
(44) a. * Kogai se sjećas što si pro upoznao ti? (Sr)

who remember.2sg that aux.past.2sg pro.acc meet.past.part
‘Who do you remember that you met?’
b. * Kadi se sjećas što si upoznao Mariju ti?

when remember.2sg that aux.past.2sg meet.past.part Maria
* ‘When do you remember that you met Maria?’
In view of the examples discussed in Sections 6.4.1–6.4.3, the judgments here

are unexpected.
There are (at least) three reasons to think that što-clauses are not
embedded clause structures. First, što-clauses require resumption (Boban
Arsenijevic, p.c.). This is shown in (45) with resumptive ga ‘it.’ This re-
quirement is rescued with a pro resumptive with nominative and ani-
mate accusatives (Boban Arsenijevic, p.c.). Da-clauses do not require
resumption.
168
(45) Zaista žalim što sam ga se odrekao.

really regret.1sg that aux.1sg it.Cl.gen refl renounced.m.sg
‘I really regret that I renounced it.’ (Boban Arsenijevic, p.c.)
Because (43a) and (44a) involve an animate accusative, both involve an accu-
sative pro (in italics).
Second, relativization with što requires resumption too (46) (Boban
Arsenijevic, p.c.; Mitrović 2012), that is, što-clauses are arguably relative
clause constructions (see Arsenijevic 2009 for SC; see also Aboh 2005 for Gbe;
Krapova 2010 for Bg; and Haegeman and Ürögdi 2010a, 2010b for arguments
in favor of factive complements as relatives).7
(46) Žena što ju je skavi čovek pozvao na večeru

woman that her aux each man invited on dinner
zahvalila mu se. (SC)
thanked him refl
‘The woman every man invited to dinner thanked him.’ [Mitrović 2012, (15b)]
Finally što-clauses can be embedded under veridical verbs. When they are
embedded under žaliti ‘regret,’ they introduce strongly veridical domains (see
Section 6.3). If that is correct, we expect što to be embedded “by default” under
the fact, as in (47). This is not borne out: Only da can appear under činjenica
‘fact’ (Boban Arsenijevic, p.c.).
(47) Zaista žalim činjenicu da/*što sam

really regret.1sg fact that aux.1sg
ga povrijedio. (SC)
him.acc.Cl hurt.m.sg
‘I really regret the fact that I hurt him’
For these reasons, I leave što-clauses out of the discussion here.8
6.4.5 The fseq of complementizers
Section 6.3 told us that the linear ordering of the features constituting the fseq
of complementizers is one in which the relatively veridical and nonveridical
7. Why (43a) is fine and (44a) ungrammatical still remains to be explained, even
under a relativization analysis.
8. See Haegeman and Ürögdi (2010a, 2010b) on operator movement and factivity; see
also Aboh (2005) and Krapova (2010) about some factives being relative clauses. See
also Arsenijevic (2009) and Kayne (2009) for embedded clauses being relative clauses.

features are adjacent, and in which the strongly veridical and relatively verid-
ical features are also adjacent [see (18), repeated here as (48)].
(48) strongly veridical | relatively veridical | nonveridicalind | nonveridicalsubj
What this section teaches us is that strongly veridical complementizers yield

stronger islands than relatively veridical complementizers and that nonveridical
complementizers create no island at all. Taking into account featural RM and the
idea that layers are cumulative (see Section 6.1.2), nonveridical complementizers,
which are (semantically) unmarked (they are non-presuppositional), should be
structurally very small; strongly veridical complementizers, which are the “more
semantically marked” of the complementizers investigated here (they involve
strong presupposition), must be quite big. Relatively veridical complementizers,
involving some kind of “weak presupposition,” are less semantically marked
than strongly veridical complementizers, but more marked than nonveridical
complementizers, and as such must be of an in-between size, by this logic.
Thinking in terms of cumulative layers, then, strongly veridical complementizers
are the biggest of all complementizers and include all the other layers. As such,
they occur at the very left-hand end of the complementizer fseq, and the
smallest, nonveridical complementizers should occur at the very right-hand end
of the complementizer fseq (49). Relatively veridical complementizers are then
sandwiched in between.
(49) strongly veridical > relatively veridical > nonveridicalind > nonveridicalsubj
The hierarchy in (49) is based on syncretism patterns with complementizers

and on wh-extractions out of (non)veridical domains in four different lan-
guages. If the hierarchy is adequate, the labels attributed to the features
appear to be misleading, though: Veridicality is a (semantic) property of
predicates, not complementizers. In Section 6.5, I propose that (non)veridical
predicates can select three types of complementizers, which vary in terms of
existential presupposition, that is, complementizers can be specific, partitive,
or non-presuppositional. What governs what kind of feature(s) is selected is,
I claim, (non)veridicality.
6.5 THE FEATURES OF WH-P HRASES AND COMPLEMENTIZERS
Under the logic used in Section 6.4, if a complementizer blocks wh-

extractions, it must share (some) features with wh-pronouns, features rele-
vant to RM. If it does not block movement, then features relevant to RM are
not involved. Building on Baunaz (2015, 2016b), I claim here that the fseq
of complementizers is basically similar to that of wh-phrases: Both involve
170
an operator that is built independently and that is prefixed to a BaseP, as

schematized in Figure 6.2.9
OpP
BaseP
Figure 6.2 [OpP [BaseP]]
6.5.1 Pref ixed OpP
As seen in Section 6.4, complementizers in MG, SC, Bg, and Fr may block
wh-movement, that is, they may block movement of quantificational phrases.
Wh-phrases minimally involve an Operator (Op) (Rizzi 1997, among others).
So because they intervene in quantificational chains, (some) complementizers
should also involve an operator (or an operator feature; see fn.9 of this chapter,
and Haegeman and Ürögdi 2010a and 2010b). I follow Roussou (2010) and
Manzini and Savoia (2011) and claim that complementizers may involve a
declarative operator [Opprop(ositional), operating over propositional variables
(in the embedded clause)]. Unlike interrogative pronouns, they range over
propositions, not individual variables.10
There is at least one reason to believe that an Op is prefixed to complementiz-
ers: Complementizers in Fr, MG, SC, and Bg are syncretic with a morpheme that
appears in the formation of quantifiers: Fr. cha-que ‘every,’ MG ká-pu ‘somewhere,’
Bg kă-deto ‘where (relative),’ and so forth. These quantifiers are bimorphemic
and involve an overt Op morpheme that is prefixed to -que/-pu/-deto. Note that
the MG complementizer oti is arguably also bimorphemic, consisting of the def-
inite article o-plus -ti ‘thing’ (see Roussou 2010, among others) and conforms to
the previous quantificational pattern; see MG ká-ti ‘something.’11 Extending this
idea to other quantificational elements amounts to saying that these items are
bimorphemic too, with Op being nonovert. In Figure 6.3, Opα (α = wh, foc, ∀, ∃,
9. OpP also spells out an fseq. As this is beyond the scope of this chapter, the reader
is referred to Baunaz and Lander (under review) for an analysis. This means that OpP
is constituted of (potentially various) features.
10. Even though BaseP in Figure 6.2 is at the core of the internal structures of
complementizers, it does not select anything, syntactically speaking: It is the higher
layers that determine where our nanostructure gets inserted in the clause (see also De
Clercq in Chapter 7).
11. Also relevant to this discussion is recent work by Szabolcsi, Whang, and Zu (2014)
and the references cited there (thanks to a reviewer for pointing this out to me).

Øwh/prop/cha-/ká-/o-/kă- ⇐ OpαP
BaseP ⇒ (-)que/(-)ti(-)/(-)pu(-)/(-)deto
Figure 6.3 Lexicalizations of OpP and BaseP
etc.) is a constituent of its own, built in a domain outside of BaseP and subse-
quently prefixed to BaseP as an independent element (see Baunaz and Lander in
Chapter 1, Section 1.3.3.5 and Starke in Chapter 9; see also Baunaz 2016b; Baunaz
and Lander to appear; and Leu’s 2010 proposal on German jeder and Fr. chaque).
If OpP involves information about the semantic meaning of these
constituents (namely interrogative, universal, etc.), what about BaseP? I pro-
pose in Section 6.5.2 that BaseP involves an fseq whose highest property spells
out specificity (that is, β in Starke’s terms discussed in Section 6.4). By the
Superset Theorem see Baunaz and Lander in Chapter 1, 27, smaller pieces of
the structure can also be spelled out, involving other features. These features,
combined with Opα, are what is relevant to RM and will account for the verid-
ical islands presented in Section 6.4.
6.5.2 BaseP
Our working hypothesis, based on featural RM, is that wh-phrases and

complementizers share similar features. In our structure in Figure 6.3, the
features shared by wh-phrases and complementizers cannot be in PreP, as
these features relate to the semantics of these items: wh-phrases are interrog-
ative, complementizers are noninterrogative. Thus the feature we are looking
for must rather belong to BaseP. In this section I show that several features
may appear in the extended projection of BaseP in Figure 6.3. On the basis
of previous work of mine on wh-words in French I propose that the features
making up BaseP are specificity and partitivity. I argue that these features are
ordered above a categorial feature [n(ominal) for wh-phrases, c(ategory) for
complementizers], which constitute the core of these constituents.
6.5.2.1 Specificity, Partitivity, Neutrality
Baunaz (2011, 2015, 2016b) assumes that wh-phrases can receive three
different interpretations according to three different discursive contexts:
partitive, specific, and neutral. These concepts are defined as follows: A
172
partitive interrogative phrase is an object that belongs to a closed set of

presupposed alternatives. A specific interrogative phrase narrows down
the context to familiar individuals, excluding alternatives. As for out-of-
the-blue interrogative phrases, they lack specificity and partitivity, that
is, they do not involve existential presupposition. In that sense, they are
neutral.
(50) exemplifies partitivity, (51) specificity: The referent of qui in (50) can
potentially be any girl belonging to the predefined set of girls, that is, the
blonde one, the red-headed one, or the brunette. In (51) the journalist wants
to identify the individual that all the witnesses recognized.
(50) After the parade, all the girls are standing in front of the jury. Joe, one of the
judges, asks Bob:
Joe: Et toi, qui tu préfères, là?
‘And you, who do you prefer, here?’
Bob: La blonde. /La brune. /# Aucune.
‘The blonde. /The brunette. /# None (of them).’
(51) Witnesses and defendants have been brought face to face. One of the
defendants has been accused by all the witnesses. Before the verdict,
a journalist asks the judge:
J: Et qui les témoins ont reconnu dans le box des accusés?
‘And who is it that all the witnesses recognized in the defendants’ box?’
Both partitive and specific contexts involve existential presupposition: In both

(50) and (51), a negative answer is infelicitous, because it would go against the
existential presupposition that there is an antecedent for qui. Importantly,
these concepts are syntactically encoded: They play a role in Scope Islands
(Baunaz 2011, 2016b). Starke (2001) shows that specificity plays a crucial role
in wh-extraction out of WIs.
Interrogative phrases can also be used in contexts in which the interlocutor
does not know its referent (and if there is a referent available), that is, in neu-
tral context, as in (52).
(52) Stan: Qui as-tu croisé ce matin à l’entrainement ?

‘Who did you meet this morning at practice?’
Nole: ‘Nobody. /Roger.’
Qui can also appear in nonquantificational contexts, as is examplified in (53):
(53) a. Les professeurs rentraient chez eux, qui à Paris, qui à Bruxelles.
the professors returned home who to Paris who to Brussels
‘The professors returned home, some of them to Paris, others to Brussels’

b. Qui apportait un fromage, qui un sac de noix, qui un quartier de chèvre . . .
who brought.3sg a cheese who a bag of nuts who a piece of goat . . .
‘One brought a piece of cheese, one a bag of nuts, one a piece of goat meat . . . ’
[Lipták 2001, 137, (13a)]
In (53), qui is interpreted as a distributee. In (53a), qui refers to professeurs and is

interpreted as partitive (‘some of them’). In (53b), qui has no antecedent: It does
not refer to a term that has been already mentioned, that is, neither specificity
nor partitivity is involved, and a neutral meaning shows up. It is an indefinite.
If we can show that one of these concepts is contained within the other,
then we can argue that they are in a hierarchical relationship. The idea behind
this is that the more semantically marked, the bigger, the less semantically
marked, the smaller (see Starke 2001 and Section 6.4 of this chapter; see also
Chapter 1, Section 1.3.3.1, about semantic composition as a tool to detect hier-
archical structures). I propose that specificity contains partitivity, but not vice
versa. Swedish shows morphological evidence for this claim.12
6.5.2.2 Morphological containment
The semantics of the partitive and specific readings suggests a certain kind of
containment. This can be seen in Swedish, for instance, in which a partitive
DP like en av killarna ‘one of the boys’ in (54a) may be preceded by a defi-
nite article to specify the reference of one of the individuals belonging to the
presupposed set, that is, turning the DP specific (54b):
(54) a. [en av killarna] kom hem till mig igår. (Swedish)
one of guys.the came home to me yesterday
‘One of the guys came to my home yesterday.’
b. [den [en-a (av killarna)]], det vill säga, Eric . . .
the one-def of guys.the that wants say Eric . . .
What (54b) shows is that specificity is bigger than partitivity and is as such
more semantically marked. In specific contexts, partitive DPs can be used. This
suggests that the two features are hierarchically ordered: specificity > partitivity.
Qui thus displays three interpretations: specific, partitive, and neutral (re-
call that the operator part is built as an independent subtree and has by hy-
pothesis nothing to do with our fseq; it is prefixed independently). These are
syncretic in (50)–(53), that is, there is a particular lexical tree (L-tree) for qui
that maps onto a range of syntactic trees (S-trees) (Figure 6.4). The S-trees for
12. Thanks to Eric Lander for discussing this with me and for providing the relevant
example.
174
(a) Lexical tree for qui

specP
partP
spec
/ ki / nP
part
(b) Syntactic trees for qui (with a prefixed operator)
(i) specific quiwh
specP ⇒ qui
Øwh ⇐ OpwhP partP
spec
nP
part
n
(ii) partitive quiwh
partP ⇒ qui
Øwh ⇐ OpwhP nP
part
n
(iii) neutral quiwh
nP ⇒ qui
Øwh ⇐ OpwhP
n
(iv) neutral quiindef
nP ⇒ qui
Figure 6.4 L-tree for qui and S-trees
qui show different feature make-ups and are of different sizes [(b) in Figure
6.4, with the partitive meaning contained within specificity. Noninterrogative
qui, being nonquantificational, nonspecific, and nonpartitive is not prefixed
with an OpP [(biv) in Figure 6.4. No special feature indicates the neutral
reading, because the neutral reading is defined in term of the absence of both

specificity and partitivity. nP indicates that qui is a (pro)nominal constituent
(i.e. it belongs to a nominal fseq).13
A reviewer notes that Figure 6.4 is reminiscent of Cardinaletti and Starke’s
1999 classification of pronouns (see Chapter 1, Section 1.3.3.2). The reminis-
cence is due to the Superset Theorem, which is at the core of how syncretism is
treated in nanosyntax. By the Superset Theorem, the L-tree in (a) in Figure 6.4
can match the S-trees in (b).
6.5.3 The internal structure of complementizers
From the beginning of this section our working hypothesis is that wh-phrases
and complementizers share similar features, and the locus of this similarity
is BaseP. Just as there exist specific, partitive, and neutral qui, I propose that
complementizers may also be specific, partitive, and neutral. On a par with
the fseq proposed for qui, I propose that complementizer que can lexicalize
structures of different sizes. This analysis is extended to MG pu/oti, SC da, and
Bg če/deto. I claim that veridicality governs the distribution of complementizers
and selects different types of declarative complementizers: specific, partitive,
or neutral, as in Figure 6.5 [c means here ‘category,’ as Fr. que and MG ti are
syncretic with nominal categories, whereas MG pu and Bg deto are syncretic
with adverbials, and the category (or categories) of SC da, Bg če is left unde-
fined here14].
Nonveridical verbs select neutral complementizers, which range over
nonfinite sets of propositional variables (neither true nor false). These
complementizers are neither specific nor partitive. Also they are perme-
able to wh-arguments and wh-adjuncts alike, that is, they do not involve
OppropP. Because wh-phrases have a “bigger” feature composition than these
complementizers—wh- phrases at least involve OpwhP— they can move
freely. Recall that nonveridical verbs can select (at least) two distinct types
of complementizers (Section 6.2.3): complementizers selected under verbs
of saying, and complementizers selected under desideratives. So presumably
there is no nonveridical feature. The S-trees of these complementizers should,
though, be different, even though the difference between them is minimal.
Because that distinction is not related to veridicality, I refer to it as X here.
13. This chapter focuses only on features that participate in RM, but there might be
more features sandwiched between PartP and nP, like features specifying n (‘thing,’
‘form,’ ‘body,’ etc.) or phi features, i.e., features that would account for the differences
between que and qui. This is left for future research.
14. But see Baunaz and Lander 2017 for development.
176
Based on syncretism patterns of Bg in Table 6.6, the complementizer selected

by verbs of saying is syncretic with the one selected by relatively veridical
verbs (= če), whereas for the complementizer selected by desideratives (da),
I propose, based on RM (Section 6.4), that it lexicalizes the smallest bit of
structure [(a) in Figure 6.5].
(a) Bare complementizer
cP ⇒ /k /, /na/, /da/
(b) XP complementizer
XP ⇒ /k /, /oti/, /da/, /t∫e/
cP
X
c
(c) Partitive complementizer
partP ⇒ /k /, /pu/, /oti/, /da/, /t∫e/

Øprop ⇒ OppropP XP
part
cP
X
c
(d) Specific complementizer
specP ⇒ /k /, /pu/,/da/, /deto/
Øprop ⇒ OppropP partP

spec
XP
part
cP
X
c
Figure 6.5 Complementizer structures
Complementizers selected by verbs of saying are slightly bigger and

lexicalize X plus the c feature [(b) in Figure 6.5]. Relatively veridical verbs se-
lect partitive complementizers that range over (a given set of) propositional
variables (true or false). Partitive complementizers are quantificational: They
create WIs. The consequence is that extraction is possible, but only for [spe-
cific] wh-arguments (which are “bigger” than partitive complementizers [see
(a) in Figure 6.4]. Wh-adjuncts, which are nonspecific and nonpartitive, are
not extractable (they are “smaller” than partitive complementizers).

Strongly veridical verbs select specific complementizers that locate the
complement proposition with respect to a given point of reference, binding a
single propositional variable, which corresponds to a single truth value (true)
(see Roussou 2010). Specific complementizers are quantificational: They block
any type of wh-movement. In addition, they introduce specific propositions.
Because specificity contains partitivity, the S-tree of specific complementizers
contains a (specific) feature, which dominates a (partitive) feature, as in (d) in
Figure 6.5. With specific complementizers, extraction is never possible: wh-
/focus-phrases are blocked. They are either of the same size as specific
complementizers or smaller.
The size of complementizers is then relevant for intervention effects.
6.6 CONCLUSIONS
In this chapter I have discussed the nanosyntax of Fr, MG, SC, and Bg com
plementizers. From syncretism patterns and long-distance wh-extractions across
complementizers in these languages, I have argued that (i) complementizers are
complex morphemes, lexicalizing structures of different sizes; (ii) their distribu-
tion is governed by veridicality; and (iii) complementizers are syntactically active.
The intervention effects observed in the four languages have been accounted for
in terms of (i) the type of features displayed by the relevant morphemes (com-
plementizer, wh-phrase): That is, OpP plus either the specific or the partitive
feature is what matters to RM; and (ii) the “size” of the morphemes involved.
REFERENCES
Aboh, Enoch. 2005. “Deriving Relative and Factive Clauses.” In Contributions to the 13th
“IGG,” edited by Laura Brugè, Giuliana Giusti, Nicola Munaro, Walter Schweikert,
and Giuseppina Turano, pp. 265–285. Venice, Italy: Cafoscarina.
Arsenijevic, Boban. 2009. “Clausal Complementation as Relativization.” Lingua 119:
pp. 39–50.
Baunaz, Lena. 2011. The Grammar of French Quantification (Studies in Natural Language
and Linguistic Theory 83). Dordrecht, The Netherlands: Springer.
Baunaz, Lena. 2014. “Declarative complementizers under the microscope”. Talk
given at SLE 44, Workshop : Nanosyntax. How going fine-grained enables a better
understanding of language. Adam Mickiewicz University in Poznań, Poland,
September 11–14.
Baunaz, Lena. 2015. “On the Various Sizes of Complementizers.” Probus 27 (2):
pp. 193–236.
Baunaz, Lena. 2016a. “Deconstructing Complementizers in Serbo-Croatian, Modern
Greek and Bulgarian.” In Proceedings of NELS 46 (1), edited by Christopher
Hammerly and Brandon Prickett, pp. 69–77. Amherst, MA: Graduate Linguistics
Student Association.
178
Baunaz, Lena. 2016b. “Wh-phrases in-situ: Interface Strategies.” Acta Linguistica 63

(2): pp. 125–168.
Baunaz, Lena and Genoveva Puskás. 2014. “The Selection of French Mood.” In Variation
Within and Across Romance Languages, edited by Eric Mathieu and Marie-Hélène
Côté, pp. 233–253. Amsterdam: John Benjamins.
Baunaz, Lena and Eric Lander. To appear. “Cross-Categorial Syncretism and the Slavic
Containment Puzzle.” In Balkan Syntax and (Universal) Principles of Grammar, ed-
ited by Iliana Krapova and Brian Joseph. Berlin: Mouton de Gruyter.
Baunaz, Lena and Eric Lander. 2017. “Syncretisms with nominal complementizers.”
Studia Linguistica. DOI: 10.1111/stul.12077. Online ISSN: 1467–958.
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of
Tromsø.
Cardinaletti, Anna and Michal Starke. 1999. “The Typology of Structural Deficiency.
A Case Study of the Three Classes of Pronouns.” In Clitics in the Languages of
Europe, edited by Henk van Riemsdijk, pp. 145–233. Berlin: Mouton De Gruyter.
Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency.
Amsterdam and Philadelphia: John Benjamins.
Giannakidou, Anastasia. 2009. “The Dependency of the Subjunctive Revisited: Temporal
Semantics and Polarity.” [Special issue on Mood] Lingua 119: pp. 883–1908.
Giannakidou, Anastasia. 2015. “Evaluative Subjunctive and Non-Veridicality.” Ms.,
University of Illinois at Chicago.
Haegeman, Liliane. 2010. “The Internal Syntax of Adverbial Clauses.” In Exploring the
Left Periphery, edited by Kleanthes Grohmann and Ianthi Tsimpli. Lingua the-
matic issue, 120: pp. 628–648.
Haegeman, Liliane and Ürögdi, Barbara. 2010a. “Referential CPs and DPs: An Operator
Movement Account.” Theoretical Linguistics 36: pp. 111–152.
Haegeman, Liliane and Ürögdi, Barbara. 2010b. “Operator Movement, Referentiality
and Intervention.” Theoretical Linguistics 36: pp. 233–246.
Kayne, Richard S. 2009. “Antisymmetry and the Lexicon.” Linguistic Variation Yearbook
8 (1): pp. 1–31.
Krapova, Iliana. 1998. “Subjunctive Complements, Null Subjects and Case Checking
in Bulgarian.” University of Venice Working Papers in Linguistics 8 (2): pp. 73–93.
Krapova, Iliana. 2010. “Bulgarian Relative and Factive Clauses with an Invariant
Complementizer.” Lingua 120: pp. 1240–1272.
Leu, Tom. 2010. “The Internal Syntax of `jeder’ (`every’).” Language Variation Yearbook
2009: pp. 153–204.
Leu, Tom. 2015. The Architecture of Determiners. Oxford and New York: Oxford
University Press.
Lipták, Anikó. 2001. On the Syntax of WH-items in Hungarian. Doctoral dissertation,
Utrecht: Landerlijke Onderzoekschool Taalwetenschap.
Manzini, Maria R. and Leonardo M. Savoia. 2011. Grammatical Categories: Variation
in Romance Languages. (Cambridge Studies in Linguistics 128). Cambridge:
Cambridge University Press.
Melvold, Janis. 1991. “Factivity and Definiteness.” In MIT Working Papers in Linguistics
15, edited by Lisa Cheng and Hamida Demirdache, pp. 97–117. Cambridge,
MA: MIT Press.
Mitrović, Ivana. 2012. “Relative Clauses in Serbian. “Handout of a talk given in the
Syntax Lab, University of Maryland, September 11.

Rizzi, Luigi. 1997. “The Fine Structure of the Left Periphery.” In Elements of Grammar: A
Handbook of Generative Syntax, edited by Liliane Haegeman, pp. 281– 337.
Dordrecht, The Netherlands: Kluwer.
Rizzi, Luigi. 2004. “Locality and Left Periphery.” In Structures and Beyond. The
Cartography of Syntactic Structures Vol. 3, edited by Adriana Belletti, pp. 223–251.
Semantics.” In Mind, Values and Metaphysics—Philosophical Papers Dedicated
to Kevin Mulligan, edited by Anne Reboul, pp. 517– 533. Dordrecht, The
Netherlands: Springer.
Roussou, Anna. 1992. “Factive Complements and Wh-Movement in Modern Greek.”
University College London Working Papers in Linguistics 4: pp. 123–147.
Roussou, Anna. 2010. “Selecting Complementizers.” In Exploring the Left Periphery,
edited by Kleanthes Grohmann and Ianthi Tsimpli. Lingua thematic issue, 120:
pp. 582–603.
Schlenker, Philippe. 2005. “The Lazy Frenchman’s Approach to the Subjunctive.”
In Romance Languages and Linguistic Theory 2003: Selected Papers from Going
Romance 2003, edited by Twan Geerts, Ivo van Ginneken, and Haike Jakobs, pp.
269–309. Amsterdam and Philadelphia: John Benjamins.
Sočanac, Tomislav. 2017. Subjunctive Complements in Slavic Languages: A Syntax-
Semantics Interface Approach. Doctoral dissertation, University of Geneva.
Starke, Michal. 2001. Merge Dissolves Into Move. Doctoral dissertation, University of
Geneva.
Szabolcsi, Anna, James Doh Whang, and Vera Zu. 2014. “Quantifier Words and Their
Multi‐Functional(?) Parts.” Language and Linguistics 15: pp. 115–155.
Todorovic, Natasa. 2012. The Subjunctive and Indicative da-Complements in Serbian.
A Syntactic-Semantic Approach. Doctoral dissertation, University of Illinois at
Chicago.
Tóth, Enikő. 2008. Mood Choice in Complement Clauses. Frankfurt: Peter Lang.
180
CHAPTER 7
Syncretisms and the Morphosyntax

of Negation*
KAREN DE CLERCQ
7.1 INTRODUCTION
As already discussed and illustrated in the introductory chapter to this volume

(Baunaz and Lander in Chapter 1, Section 1.3.3.1), syncretisms can be used to
diagnose underlying and hidden structure inside morphemes that might at
first sight be considered indivisible. In the present chapter, syncretisms are
used as a tool to look inside negative markers and discover more about the
functional sequence underlying negative markers.
In Section 7.2, I first look at negative markers from a crosslinguistic per-
spective to distinguish different functions or uses for negative markers, and
I classify negative markers accordingly into four different groups. Section
7.3 discusses the syncretism patterns that can be detected across the nega-
tive markers of ten different languages. In Section 7.4, I argue that negative
markers are internally complex and may consist of up to five different features
depending on their distribution, semantics, and use. Apart from a negative
feature (Neg), hosting a semantic negative operator, negative markers may
spell out additional syntactico-semantic features, such as scalar quantity (Q),
classifier (Class), focus (F), and tense (T).1 Indeed, negative markers never
spell out just the negative feature, but they are always packaged with at least
* This research was funded by the Research Foundation—Flanders (FWO).

1. Poletto (2008) also splits up NegP into several functional projections based on a
study of negative markers in Italian dialects. However, unlike the present chapter her
approach only deals with negative markers at the sentential level.
one of these additional features. Section 7.5 provides a brief preview of how
those featurally complex negative markers take scope in their positions in the
clause, and Section 7.6 is a conclusion to the chapter.
The main hypothesis underlying this chapter is that a negative marker in
natural language is always featurally complex, packaging negation (with its
corresponding semantic operator) with other syntactico-semantic features.
Crucially, I make a distinction between the semantics of negation the one
hand and the syntax and semantics of negative markers in natural language.
These concepts do not coincide. The study of the morphosyntax of negative
markers and the syncretism patterns across them brings out the need for
drawing this distinction.
7.2 CLASSIFICATION OF NEGATIVE MARKERS

7.2.1 Preliminaries
When it comes to classifying negative markers, the literature is not so clear on

how many different types of negative markers there are, as opposed to, for in-
stance, what we find in relation to the classification of case markers (cf. Caha
2009 for case markers and the references cited therein). A lot of the discussion
has focused on whether negative markers are propositional negators, predicate
negators and/or (predicate) term negators (Horn 1989, 140–141), and whether
these markers express contradictory negation or contrary negation (Horn
1989, ch.1). Even though the semantics of negation is never out of sight in the
discussion to follow, the core of the present discussion is the morphosyntax of
negative markers and what it tells us about the semantics of negative markers.
For expository purposes, I limit my discussion to negative markers
that combine with adjectival predicates in predicative position and with
copular verbs.
7.2.2 Classifying properties
In the present section I classify negative markers according to four

properties: (1) their semantics (contradictory or contrary negation), (2) their
scope position, (3) their ability to be stacked, and (4) their function, that is,
how they are used. By means of these properties I classify negative markers
into four different groups, whose label is based on the syntactic position in
which they take scope: (1) negative tense markers, Tneg-markers, (2) nega-
tive focus markers, Focneg-markers, (3) negative classifier markers, Classneg-
markers, and (4) negative scalar quantity markers, Qneg-markers. I first discuss
the four properties on which the classification is based.
S y n c r e t i s m s a n d t h e M or p h o s y n ta x of N e g at i o n [ 181 ]
182
The semantic property that plays a role in the classification is whether neg-
ative markers express contradictory or contrary negation. In what follows
I explain the concepts briefly to be able to show how the discussion affects
the classification of negative markers in this chapter. More concretely, the
semantic distinction between contradictory and contrary negation will turn
out to be decisive in splitting up the group of-what Aristotle used to refer to
as-predicate term negators into two different types of negative markers.
Two sentences are each other’s contradictories when the Law of
Contradiction (LC) and the Law of the Excluded Middle (LEM) apply. Sentences
(1a) and (1b) and (2a) and (2b) cannot be true or false at the same time. They
respect the LEM and the LC and are thus each other’s contradictories.
(1) a. John is bald.

b. John is nonbald.
(2) a. John is married.
b. John isn’t married.
The LC states that a proposition (p) cannot be true and false in the same
circumstances.
(3) Law of Contradiction (LC)

∼(p & ∼p)
(Russell 1940, 259)
The Law of the Excluded Middle says that any proposition is either true
or false.
(4) Law of the Excluded Middle (LEM)

p ∨ ∼p
(Russell 1940, 259)
However, in a pair of sentences like (5a) and (5b) the LC applies but the LEM
does not hold; both (5a) and (5b) can be false at the same time, as illustrated
by (5c). When two propositions can be false together, they are each other’s
contraries.
(5) a. She is happy.

b. She is unhappy.
c. She is neither happy, nor unhappy.
Following Horn (1989, 273–286), I assume that the contrariness of (5a)

and (5b) is encoded by the low-scope predicate term negator un-. The prefix

un-turns the negative predicate term happy into an opposite, invoking a
middle ground between two predicates in a way negative markers expressing
predicate denial, like not or n’t, or predicate term negation, like non, do
not. When negative markers are able to invoke a middle ground between
two predicates, they give rise to contrary negation. Contradictory negative
markers, like non in (1b), usually denote the complement of the predicate
(term) they combine with, whereas contrary negative markers denote the op-
posite of the predicate (term) they combine with, leaving a middle ground
between the negative and nonnegative predicate. Even though certain nega-
tive markers tend to give rise to contrary negation and others to contradic-
tory negation, it is by no means my intention to claim that contradictory and
contrary negation are exclusively and inherently related to a certain negative
marker.2 I use these semantic notions to further delineate a distinction be-
tween types of negative markers that is also visible from other properties
(like, for instance, stacking).
The most important criterion for the classification is the scope a negative
marker has. Scope does not necessarily coincide with the surface position of
a negative marker, as illustrated by the Neg-raising example in (6), which—as
shown in the paraphrase provided—has the low-scope reading of negation.
(6) I don’t think it will rain today

= I think it will not rain today.
(Collins and Postal 2014, 9)
That said, negative markers very often scope in their surface position, as illus-
trated by the examples in (7). In (7a) the negation clearly does not scope over
the subject many students, but is restricted to the predicate happy. In (7b) the
negation scopes over many students, which is immediately in line with the sur-
face position of the negative markers.
(7) a. Many students are not happy. (many > ¬)

b. He didn’t invite many students. (¬ > many)
The structure in Figure 7.1 shows where negative markers in combination with
adjectival predicates tend to surface and which functional projections I asso-
ciate with these surface positions.
2. “If p is a set of possible situations, those in which p is true, and U is the Universe
of possible situations of which p is a subset, then whether q is a contradiction or con-
trary of p is relative to U” (Borschev et al. 2006, 6). Consequently, it is hard to pin down
when negation gives rise to contradictory or contrary negation, because, as noted by
Borschev et al. (2006), this is highly influenced by the Universe and hence by pragmatic
factors.
184
TP
T FocP
Foc vP
V ClassP
Class QP
Q aP
Figure 7.1 Positions of negative markers in the functional projection
These positions are at the same time also responsible for the label the
different markers receive in this chapter. In Section 7.2.3 I elaborate in more
detail on the different surface and/or scope positions of negative markers.
A good probe into the scopal and distributional properties of negative
markers is whether negative markers can be stacked on top of each other.
Stacking is at the same time also the third property on which the classification
builds. When two negative markers cannot co-occur, I take this as evidence
that they compete for the same position. Stacking can thus be used as a test to
see where negative markers can surface, how they can be combined, and hence
where they take scope with respect to each other.
A final property that is used in the classification is how negative markers are
used. I refer to this as the function negative markers have. Negative markers
can deny, emphasize, modify, classify, and characterize, depending on where
they surface, as I illustrate and discuss in Section 7.2.3.
In Section 7.2.3, four different groups of negative markers are distin-
guished based on the properties discussed in this section, and the typology is
applied to the negative markers of English.
7.2.3 Four groups of negative markers
On the basis of the four properties discussed in Section 7.2.2, four groups
of negative markers can be distinguished: (1) negative tense markers,

Tneg-markers, (2) negative focus markers, Focneg-markers, (3) negative classi-
fier markers, Classneg-markers, and (4) negative scalar quantity markers, Qneg-
markers. In what follows I discuss per group how the properties apply. At the
end of this section, a table summarizes the results.
7.2.3.1 Negative tense markers
Negative markers like English n’t (in spoken English) and not (in formal written
English) are usually referred to as markers giving rise to sentence negation
(Klima 1964, 270–276), nexal negation (Jespersen 1917, 42–55), or predicate
denial (Horn 1989, ch.1). These markers usually give rise to contradictory
negation, which was the first property that led to the present classification.
I label them Negative Tense markers (Tneg-markers henceforth), because they
normally take scope over the tensed predicate. Syntactically, their scope po-
sition can be translated as a position dominating TP, a functional projection
hosting tense at the clausal level.3,4 Even though the English negative marker
n’t surfaces below the tensed auxiliary, it scopes over the tensed predicate, as
illustrated by the question tags in (8), a test that Klima (1964, 263) introduced
as a diagnostic for identifying sentence negation.5
(8) It isn’t possible to solve that problem, is it? (Horn 1989,185)
The freestanding not on the other hand can give rise to both positive and neg-
ative tags, see (9).
(9) Kim is not happy, is she /isn’t she? (based on Horn 1989, 490, 517)
3. Ouhalla (1991) argues that the position of negation with respect to T is subject to
parametric variation. For French (Rowlett 1998), Italian (Zanuttini 1996, 1997) and
Spanish (Laka 1994) it has been argued that NegP (also called ∑P (Laka 1990) or PolP
(Culicover 1991) is above TP. However, for English it has been argued, for instance by
Ouhalla (1991), that NegP is below TP. The line taken here is that the negation which
takes scope over the tensed predicate is always in a NegP dominating TP, a proposal in
line with Haegeman (1995) and Holmberg (2003) with respect to English.
4. I focus on negative tense markers that combine with the present indicative tense.
However, there are languages that develop different negative markers depending on
a particular type of tense or mood (Haspelmath 2011; Dryer 2011). If a language has
different negative markers within the finite domain, these markers tend to be in com-
plementary distribution with the negative marker used with the present tense indic-
ative, and will hence never be stacked on top of each other within the same clause.
I assume for now that tense-and mood-related markers belong to the group of nega-
tive tense markers, and I leave a detailed study of them for future research.
5. An exception to this rule are the modal verbs shouldn’t and mustn’t. With these
modal verbs n’t scopes below the modal verb. I consider this a consequence of in-
herent properties of these deontic modals and I will not pursue this further in the
present paper.
186
As we will see, not is ambiguous between a Tneg-marker and a Focneg-marker,

or, in Klima’s terms, a sentence negator and a constituent negator. When not
gives rise to positive tags I consider it a Tneg-marker; when it gives rise to neg-
ative tags I consider it a Focneg-marker (see Section 7.2.3.2).
When we consider the stacking properties of n’t in English, we find that it
can stack on an emphatic NOT and on the negative prefix un-, as illustrated
in (10):
(10) She isn’t NOT unhappy.
It is impossible to stack the same negative marker n’t twice on top of each
other with the same predicate, see (11).6
(11) *She isn’t-n’t happy.
Moreover, the stacking order is fixed: n’t—NOT—un-is the only option, (12):
(12) *She is-un not n’t-happy.
As mentioned before, stacking shows how negative markers can be combined

and hence where they take scope with respect to each other. (10) shows that
the scope of n’t is wider than the scope of not.
Finally, with respect to their functions, Tneg-markers predominantly have
the function of denying an utterance accessible in the discourse. Horn (1989,
203) states that “the prototypic use ( . . . ) of negation is indeed as a denial of a
proposition previously asserted, or subscribed to, or held as plausible by, or at
least mentioned by, someone relevant in the discourse context.”
Summarizing, we identified English n’t and not as Tneg-markers, giving rise
to contradictory negation and taking wide scope over the tensed predicate in
a NegP dominating TP. Moreover, we saw that these markers can stack on at
least two other negative markers, again confirming their wide scope, and that
they function as denials.
7.2.3.2 Negative focus markers
The English low-scope negator not, which we identified as the one giving rise to
negative tags in Section 7.2.3.1, is often also referred to as a negative marker
that gives rise to constituent negation (Klima 1964, 307), or, in Aristotelian
terms, predicate negation. In terms of their semantics, these markers
6. See De Clercq and Vanden Wyngaerd (2017), who explain the ungrammaticality of
(11) as a consequence of a violation of the fseq.

contribute contradictory negation, though their reading can be strengthened
to a contrary interpretation (Horn 1989).
I label this type of marker a Negative Focus marker (henceforth Focneg-
marker) and want to argue that they are in a position for negation above a
low Focus Phrase (FocP),7 a projection dominating vP (Belletti 2001, 2004;
Jayaseelan 2001, 2008; Butler 2003; Kandybowicz 2013 for discussion on low
FocP), see (1).
The link between focus and negation has often been made in the litera-
ture (Jackendoff 1972; Kratzer 1991; Horn 1989; Herburger 2000; Han and
Romero 2001; Haegeman 2000; Butler 2003; Neeleman and Vermeulen 2012;
and many others), though it is far from obvious what the full picture is. It is
beyond the scope of this chapter to discuss the relationship between negation
and focus in detail. Nevertheless, to make clear why I label this group of nega-
tive markers focus negative markers, a brief excursion to a discussion of focus
is necessary.
The focus of a proposition P is that part that would correspond to the wh-
word in a wh-question (Neeleman and Vermeulen 2012, 232; see also Rooth
1985, 1992). Pia is the focus in (13) and John in (14), and they both attract the
main stress in English (Selkirk 1984).
(13) Who did John invite?

He invited PIA.
(14) Who invited Pia?
JOHN invited Pia.
(Neeleman and Vermeulen 2012, 232)
Focus also triggers a set of alternative propositions. This set of alternative

propositions is called the focus value of a sentence, and it differs from the
ordinary value of a sentence. The ordinary value and the focus value of a sen-
tence like (13) are given in (15).
(15) Ordinary value: [John invited Pia]

Focus value: [[John invited Pia], [John invited Bill], [John invited
Sarah], [John invited Tom], . . .]
Neeleman and Vermeulen (2012) propose that in a case of contrastive focus,

as in (16), when contrast is linguistically encoded, an extra component is nec-
essary to account for the behavior of contrastive focus.
7. Even though I am linking the position for low scope negation to focus, I do not
argue that the positions coincide, as will become clear in the analysis proposed in
Section 4.
188
(16) A: John invited Bill.

B: No, you’re wrong. He invited PIA.
The extra component of meaning that contrast contributes is “that there is

at least one alternative for which the proposition expressed by the sentence
does not hold.” The alternative component thus always consists of a negative
marker that is not present in the focus component. The focus component of
B’s reply in (16) can thus be semantically represented as in (17a) by means of
a triplet consisting of a function corresponding to the background, the focus
Pia, and the set of alternatives to focus, which is the singleton Bill in this
case. The alternatives component of B’s reply in (17b) consists of a negation,
not present in the focus component. The meaning contributed by contrastive
focus can be made explicit by the not X, but Y construction, in (17c).
(17) a. <λx[John invited x], Pia, {Bill}>

b. ∃y, y ∈ {Bill}, ¬[John invited y]
c. John invited not Bill, but Pia.
I want to argue that apart from contributing negation, a negative marker

turns the negated predicate into the one necessary “alternative for which
the proposition expressed by a sentence [in the discourse] does not hold”
(Neeleman and Vermeulen 2012, 235). In other words, by turning a predicate
into a negated alternative, negation necessarily also points to an explicit or
implicit focus. Let’s have a look at the sentence in (18):
(18) I am not angry with you for writing such innocent letters as these.
If this is the (negated) alternative for which “the proposition expressed by a

sentence [in the discourse] does not hold,” then it must directly or indirectly
point us to the focus. However, this can be determined only in context. A po-
tential focus could be (19):
(19) I am not angry with you for writing such innocent letters as these, but
[Focus for ignoring me all the time].
As such, we could say that negation, and in particular the low-scope empathic
negative marker, turns a predicate into a negative alternative of an overt or
covert focus.
Corresponding with the position above a low FocP, this group of markers
does not scope over a tensed predicate, but only over an untensed predicate
and hence behaves like a predicate negation (Englebretsen 1980). The negative

tags introduced in (9) in Section 7.2.3.1 can be used to diagnose the low-scope
reading, whereas the positive tags diagnose the wide-scope reading. I inter-
pret the ambiguity in (9) as a case of syncretism of two underlyingly different
markers because in some languages, like Greek (see Section 7.3), there are
different negative markers to express predicate negation (Focneg) and predi-
cate denial (Tneg).
It is important to note that Focneg-markers scope over the entire untensed
or nonfinite predicate, and not just over the predicate term. This is shown by
(20b), where the predicate term happy is modified by very. Not takes scope
over the entire predicate (very happy) and not just over happy.
(20) a. Kim is NOT happy, isn’t she?

b. Kim is NOT very happy(, isn’t she?)
Finally, functionally, Focneg-markers are typically used to give rise to contras-

tive or corrective focus, as in (17c) and because they scope over untensed
predicates, they are also typically used as constituent modifiers, as illustrated
by (21). Moreover, it can also be characterized as the negator used in small
clause complements of the type in (22).8
(21) not long ago, not everybody, not very often

(22) He considers it not pleasant.
Summarizing, I propose to treat low-scope constituent negators like English

not as negators taking scope in a NegP immediately dominating a low FocP.
This type of negative marker typically gives rise to contradictory readings,
induces contrastive or corrective focus readings and is also used as an adnom-
inal or adverbial modifier.
7.2.3.3 Negative classifier markers
The negative marker non-in English is very productive and combines with all pos-
sible adjectives (and nominals) from all possible origins: non-Turkish, nonintuitive,
nonpsychiatric, noncolored, nonwhite, etc., though often it rather combines with
derived adjectives than simplex adjectives and it is more usually used with neu-
tral terms than with positive or negative adjectives (Zimmer 1964, 32–35).
Typically, non-in combination with adjectives triggers contradictory nega-
tion. Consider the example in (23). Given that it does not make sense to say
that a dress is neither red nor nonred, we can conclude that there is no middle
8. I want to thank Michal Starke for pointing this out to me.
190
ground between the nonnegated predicate P and the negated P. Non- hence
gives rise to contradictory negation, respecting the LC and the LEM and di-
viding the world into into P and non-P.
(23) *Her dress is neither red nor nonred.
This semantic property of dividing the world into P and non-P brings us also
to the classifying function of this type of negative marker (Warren 1984, 101;
Kjellmer 2005), illustrated in (24).
(24) a. Nicola believes herself to be a nonangry person and, indeed,

she never loses her temper. (Corpus, ukbooks/08.)
b. Use nonfat milk instead of whole milk. (Corpus npr/07.)
(Kjellmer 2005,162-163)
Its classifying function and its property to divide a set into two subsets are
the main reasons to label this negative marker a Negative Classifier marker
(Classneg-marker). I want to propose that this negative marker takes scope above
a Classifier Phrase (ClassP) (Borer 2005), a functional projection that was orig-
inally proposed as part of the extended projection line of the nominal phrase,
turning mass into countable material. Considering the fact that the syncretism
patterns we are about to discuss in Section 7.3 focus on negative markers with
copular clauses with adjectival predicates, the concomitant claim here is that
I argue that in the extended projection line of APs ClassP is also present.9
With respect to its stacking properties, non-can stack onto adjectives
prefixed with iN-, dis-, or un-(see Section 7.2.3.4), but not onto the Focneg and
Tneg markers not or n’t (25).
(25) a. non-dis-enfranchized, non-in-finite

b. Non-un-happy people are the best. (Jim McCloskey, personal
communication, hereafter p.c.)
c. *John is non-not happy.
d. *John is-non-n’t happy.
9. De Clercq (2013) labelled non-as a Degneg-marker, a Negative Degree Phrase, taking
scope in DegP, a projection for which Corver (1997) argued it is in the extended projec-
tion line of gradable APs. According to Corver (1997) DegP hosts degree items like so
and that that have deictic and determiner-like properties, pointing to a specific degree
on a scale. This intuition is still present in the present Classifier label. However, since
it turns out that what Corver labels as Degneg-markers can be stacked on non-, see (i), it
seems unlikely that the scope position of non-is in or right above DegP.
(i) Goodbye Weber, you and your customer service are so non-professional in helping
your customers. (\https://www.consumeraffairs.com/homeowners/weber_grill.html)

The scopal properties of non-are in line with its stacking properties and non- is
hence restricted to the predicate term. Non-does not take scope over very in
(26a), only over professional. In Aristotelian terms, Classneg-markers hence ex-
press predicate term negation.
(26) She is very nonprofessional.

a. = She is [very [¬ professional]].
b. ≠ She is [¬ [very professional]].
From the stacking and scopal properties of non-in relation to those of n’t and
not we can deduce that not and n’t take wider scope than non. The scope of
non-is limited to the predicate term, but it is slightly wider than the scope of
un-[see Section 7.2.3.4].
Summarizing, Classneg-markers, like non-, express contradictory negation,
they scope above a position ClassP, which I tentatively argue can be part of
the functional superstructure of AP, fulfilling a role comparable to the role T
fulfills with respect to V and D with respect to N.
7.2.3.4 Negative scalar quantity markers
The productive English marker un-(27a) and the less productive dis-(27b) and
iN-(27c) are what I call Negative Scalar Quantity markers, or Qneg-markers.
(27) a. unhappy, untrue, unfriendly

b. disadvantageous, discourteous, disharmonious
c. inhuman, impeccable, immaculate
Dis- and iN-are used for words from foreign origin (Zimmer 1964, 27–32),
that is, iN-combines mainly with words of Latin origin and dis-with words
of French origin. The productive marker un-is used mainly for native English
words (Zimmer 1964, 35). Given that the unproductive prefixes iN-, dis-, and
the productive un-are in complementary distribution, we deduce that they
belong to the same type of negative markers, (28) and (29).
(28) a. He is irrational.
b. *He is unrational.
c. *He is disrational.
(29) a. His behavior is un-American.
b. His behavior is *in-American.
c. His behavior is *dis-American.
192
Like Classneg-markers, Qneg-markers also scope over the predicate term and
again they do not scope over the entire untensed predicate (30). They thus
express predicate term negation in Aristotelian terms.
(30) She is very unhappy.

a. = She is [very [¬ happy]].
b. ≠ She is [¬ [very happy]].
However, their scope position is even closer to the adjectival stem than the
scope position of Classneg-markers. Two arguments for this come from (i) the
distinction between Level I and Level II morphemes (Siegel 1974; Allen 1978;
Lieber 1981; and others; Horn 1989, 283–286) and (ii) stacking of Classneg-
and Qneg-markers. I discuss these two points in turn.
First, some Qneg-markers in English, like iN-, have allomorphs, that is, they
undergo morphophonological change due to the adjectival stem to which they
attach, as illustrated in (31).
(31) in-human, ir-relevant, im-mature, im-portant, il-logical, . . . . (based on

Zimmer 1964, 28–29)
This is a typical property of Level I morphemes, a morpheme that is struc-

turally so close to the root that it adapts to it. Level II morphemes, like non-,
never show this morphophonological change. Qneg-markers, which have lowest
scope, are also the negative markers that give rise to lexicalized readings (Horn
1989, 282). A word like inhuman, for instance, does not mean ‘not human,’ but
‘cruel.’ This tendency to develop lexicalized readings is again a typical property
of Level I morphemes.
A second argument to show that Classneg-markers have wider scope than Qneg-
markers comes from stacking. The English negative markers un- and dis-cannot be
stacked (33a), whereas in (admittedly rather contrived and rare cases) the Classneg-
marker and the Qneg-marker can be stacked, with Qneg-taking lowest scope (32b).
(32) a. *undisloyal, *disunhappy, . . .

b. Nonunhappy people are the best. (Jim McCloskey, p.c.)
With respect to their scope position, I propose that Qneg-markers take scope above
the functional projection QP, which Bresnan (1973) and Corver (1997) argue is
part of the functional superstructure of AP. According to Corver, the functional
projection QP hosts words like more, less, enough, which restrict the denotation
of the adjective by picking a set of degrees on the scale of “intelligence,” as the
example in (33) shows.
(33) John will never be morei intelligent than his sister. (Corver 1997, 132)

In line with this intuition but also deviating from Corver’s proposal I want
to argue that Q° contributes gradability and that Qneg-markers select a set
of degrees at the negative outer end of the scale. Unlike Corver, however,
I assume that more functional heads must be involved in the inner structure
of words like more or less than just Q°, which I take to contribute only grada-
bility and not the comparative for instance.
(34c) illustrates how un-selects a set of degrees at the outer end of the
scale of happiness, thus creating an opposite with respect to happy in (35a).
The negative gradability contributed by un-can give rise to a predicate that
creates an opposition such that in certain contexts neither (34a) nor (34b) is
true. Un-thus evokes a middle ground, as illustrated in (34c). Qneg-markers are
therefore contrary negators.10
(34) a. John is happy.

b. John is unhappy.
c. John is neither happy nor unhappy.
Un-’s ability to give rise to contrary negation also determines its function.
Qneg-markers are usually characterizing or scalar markers (Funk 1971; Kjellmer
2005), as seen in (35), as opposed to the classifying function of non- (Warren
1984, 101; Kjellmer 2005), cf. (24).
(35) a. President Clinton, a man whose liberalism and personal

lapses arouse distinctly unchristian hatred at First Federated.
(Corpus, times/10.)
b. Lainey had a terrible voice, unmusical and sharp, and she
usually pitched herself an octave below the sopranos to
submerge it. (Corpus, usbooks/09.) (Kjellmer 2005,162)
In a context like (36), the Q-negated characterizing adjective is appropriate,

but not the Class-negated classifying predicate. However, for a context like
(37), only the Classneg-marker is appropriate, and the Qneg-marker is not
appropriate.
(36) Context: An American boy is spitting on the American flag. A teacher says:
a. His behavior is un-American. (= inappropriate for an American)
b. # His behavior is non-American
10. Un-can give rise to contradictory negation in certain cases (Zimmer 1964, 35–45).
I abstract away from this discussion for now and refer the reader to Zimmer (1964) and
Horn (1989, 273–295) for discussion of these cases.
194
(37) Context: Amie is an actress. However, she never obtained a degree.

Her friend says:
a. She is nonprofessional.
b. # She is unprofessional.
These contexts show that Qneg-markers evoke contrary readings and create
characterizing or scalar predicate terms, as opposed to Classneg-markers that
give rise to contradictory negation and create classifying referential predicate
terms.
From four properties— (1) semantics (contradictory or contrary), (2)
scope, (3) stacking properties, and (4) function—we have identified four
different types of negative markers, both at the constituent level and the sen-
tential level and discussed their properties with respect to English. Table 7.1
summarizes the different properties that we discussed per different type of
marker. The English markers that correspond to the four distinguished types
have been added to the table.
Table 7.1 CL ASSIFICATION
Tneg Focneg Classneg Qneg
not/n’t not non un-/iN-/dis-
predicate denial predicate negation term negation
Scopes over tensed predicate predicate predicate term
Stacks on Focneg, Classneg, Qneg Classneg, Qneg Qneg -
Semantics contradiction contradiction contradiction contrariety
Function denial contrast/modifying classifying characterizing
7.3 SYNCRETISM PATTERNS
With the classification in Section 7.2 as a tool, negative markers in nine lan-
guages were investigated: Modern Greek, French, Chinese, Persian, Modern
Standard (MS) Arabic, Moroccan Arabic, Dutch, Hungarian, and Czech.11
The approach taken to typological research is the Middle Way (Baker and
McCloskey 2007, 294) approach. Under this approach fewer languages are
studied than typical of typological research, but more languages than typical
11. For more detailed data discussion of eight of these languages see De Clercq (2013,
40–67). However, Dutch is not discussed there. For a detailed discussion of a broader
typological sample with 22 languages, see De Clercq (in prep.).

Table 7.2 SYNCRETISMS
Tneg Focneg Classifier Qneg

Pattern 1 Modern Greek dhen oxi mi a-
Informal English -n’t not non- un-
Bon Usage French ne . . . pas pas non iN-
Pattern 2 Formal English not not non- un-
Colloquial French pas pas non iN-
Pattern 3 Chinese bù bù fei fei
MS Arabic laa laa ghayr- ghayr-
Persian na na qheyr- qheyr-
Pattern 4 Moroccan Arabic ma (ši) muši muši muši
Pattern 5 Dutch niet niet niet- on-
Hungarian nem nem nem -tElEn
Pattern 6 Czech ne- ne ne- ne-
of generative research. As such, a combination of both empirical coverage (a

strength of typological research) and theoretical abstraction (a strength of
generative syntax) can be obtained.
As can be seen in Table 7.2, the languages and their negative markers
can be ordered in such a way that the attested syncretisms are always in ad-
jacent cells: There is no language in the sample that has its Qneg-marker and
Tneg-marker syncretic without also having a syncretic Classneg-and Focneg-marker. The
negative markers can thus be ordered such that there is no violation of the *ABA
restriction on syncretism (see Baunaz and Lander in Chapter 1, Section 1.3.3.2)
between the different markers. The order thus obtained follows the natural se-
mantic scope of negation: going from wide (for instance English n’t) to narrow
scope (un-) or, inversely, going from narrow to wide scope. Consequently, the
syncretism pattern that emerges from the crosslinguistic comparison of neg-
ative markers shows that morphology is not arbitrary: It follows the natural
semantic scope of negation.
Building on nanosyntactic ideas (see Starke 2009, 2011a, 2011b; Caha 2009;
Baunaz and Lander Chapter 1) I propose that syncretisms point to structural
relatedness. Negative markers that are syncretic must be contiguous, because
they share features. This contiguity relationship can be translated in terms of
hierarchically ordered features. Syncretic negative markers thus share hierar-
chically ordered features. Negative markers can thus not be monolithic blocks.
They need to be decomposed into features to capture the syncretism patterns
and the absence of ABA patterns. The morphology of Greek, which is fully
nonsyncretic for the four types of negative markers that we distinguished,
exemplifies the featural complexity of these different types of negative
196
markers and the need for distinctive features, whereas the morphology of
Czech mirrors how, in spite of the featural complexity, negative markers also
share something, that is, negation. In Section 7.4, the internal featural make-
up of negative markers is discussed.
Summarizing, the properties that we have identified to classify negative
markers into different types can also be found in nine languages other than
English. For reasons of space I cannot show this for each language separately,
but I refer the reader to De Clercq (2013, 40–67) for discussion of the data.
Moreover, negative markers can be ordered in such a way that no violation
of the *ABA restriction arises. In addition, Table 7.2 shows that morphology
parallels natural semantic scope and probes into the internal syntax of nega-
tive markers, as is explained in the next section.
7.4 THE INTERNAL SYNTAX OF NEGATIVE MARKERS
I assume that each of the four different groups of negative markers that we
distinguished has a distinctive feature that sets it apart from the other types
of negative markers. In combination with an additional negative feature,
these four different features constitute the internal structure of a negative
marker. I label these features Q, Class, Foc, and T in correspondence with
the overall labels of the respective groups. The syncretisms that we detected
indicate that the different markers—and more specifically their underlying
respective distinctive features—are structurally related and organized in a
particular way. Given that there is no language in the sample that has its Qneg-
marker and Tneg-marker syncretic without also having a syncretic Classneg- and
Focneg-marker, I want to argue that negative markers can be decomposed into
four structurally contiguous syntactico-semantic features: Q, Class, Foc, and
T in addition to a Neg feature, which ensures semantic negativity. I propose
that these features are organized in a “negative nanospine,” representing the
inner structure of a negative marker (Figure 7.2).
TP
T FocP
Foc ClassP
Class QP
Q NegP
Neg
Figure 7.2 The negative nanospine

Before I discuss in Section 7.5 how this nano-fseq is inserted in clausal
syntax, I first focus on the spellout of negative markers at the nanolevel in
English, and I provide support from French for the complex nanospine in
Figure 7.2.
The lexicon contains the following lexical trees for negative markers in
English: the Qneg-marker un-in Figure 7.3, the Classneg-marker non-in Figure
7.4, and the syncretic not in Figure 7.5, which lexicalizes both the Focneg-
marker and the Tneg-marker.12
QP ⇒ un
Q NegP
Neg
Figure 7.3 Qneg-marker un-
ClassP ⇒ non
Class QP
Q NegP
Neg
Figure 7.4 Classneg-marker non-
TP ⇒ not
T FocP
Foc ClassP
Class QP
Q NegP
Neg
Figure 7.5 Syncretic Focneg-marker / Tneg-marker not
Negative markers are merged in a separate workspace, creating a negative

nanospine before it is inserted in the containing clause. So concretely, before a
12. I consider the difference between the Tneg-marker not and n’t in English a
difference between two different registers of the same language. For the present dis-
cussion I only discuss the formal form of English, with not being syncretic as a Tneg- and
Focneg-marker.
198
marker like low-scope not (Focneg-marker) can be inserted in clausal syntax, the
first feature of its nanospine needs to be merged in a separate workspace. This
first feature is Neg°. At the level of the phrase, that is, NegP, the lexicon will be
checked and Figure 7.3 will be inserted. This item is the closest match, following
the Superset Theorem (as discussed in Baunaz and Lander in Chapter 1, Section
1.3.3.2). At the next spellout node, after Q° is merged, the only item that is a
complete match is again Figure 7.3, overriding the previous (same) spellout.
When Class° is merged, the item to be inserted is in Figure 7.4. It overrides the
previous spellout upon insertion, but now the spellout is different, that is, it is
non-. When Foc° is merged in syntax, there is no perfect match in the lexicon,
because in formal English not is syncretic between being a Focneg-marker and
a Tneg-marker (Figure 7.5). In this situation the Superset Theorem kicks in and
regulates insertion of a lexical item that is not a perfect match. Hence, because
of the Superset Theorem, a structure as in Figure 7.6 can be spelled out by the
lexical item in Figure 7.5.
FocP
Foc ClassP
Class QP
Q NegP
Neg
Figure 7.6 FocP (spelled out as not)
An extra piece of support for the negative nanospine developed here comes
from French bipartite negation. A long-standing problem is that of accounting
for the necessary presence of ne for the expression of sentential negation in
le bon usage French (Pollock 1989; Grevisse and Goosse 1993; Rowlett 1998;
Rooryck 2010), as illustrated in (38):
(38) Je *(n’) ai pas faim.

I neg have neg hunger
‘I’m not hungry.’
Under the account developed here, we can treat ne and pas as the spellout of
different parts of the negative nanospine: Ne spells out the TP layer alone,
and pas spells out the lower part. Only in combination can they give rise to
sentential negation, but pas alone can give rise to constituent and contrastive
negation. French hence wears the internal structure of the higher layers of the
nanospine on its sleeve. I refer the reader to De Clercq (2017) for more details
on how the two pieces of the negative spine end up in two different positions
in the clause with the verb in between them.

Summarizing, on the basis of the syncretisms, negative markers can be or-
dered in terms of their contiguity relationship: Syncretic markers are structur-
ally closer than nonsyncretic markers. In structural terms this was translated
as a fseq of features, each feature representing a group of negative markers.
It turns out that negative markers in natural language never just encode se-
mantic negation, but they always package semantic negation, a Neg feature in
the present proposal, with at least one or several other syntactico-semantic
features like (Scalar) Quantity, Classifier, Focus, or Tense.
7.5 THE EXTERNAL SYNTAX OF NEGATION: A BRIEF PREVIEW
At this point, we still need to consider how these complex negative markers,
which are created in a separate workspace, will be inserted into the clausal
functional structure. For reasons of space, this question is not dealt with in de-
tail, but I want to broadly sketch a preview of how the external syntax of nega-
tion in English can be looked at from the perspective developed in this chapter.
The fseq that we uncovered by means of the syncretism patterns is rele-
vant not only for the spellout of negative markers at the nanolevel. The same
functional projections are part of the clausal spine, albeit interspersed with
other functional projections that are part of the fseq. Moreover, in line with
proposals by Cinque (1999) and Starke (2004), I propose that NegP (which
under the present account consists of a complex phrase) can optionally project
on top of every of the four negation-related features in the clausal spine. The
structure in Figure 7.7 shows the clausal spine with the optional projections
positions for (the featurally complex) negation.
With respect to insertion of the complex nanospine or the complex speci-
fier, the idea is that the complexity of the nanospine is indicative of the locus
(NegP)
(Neg) TP
T (NegP)
(Neg) FocP
Foc (NegP)
(Neg) ClassP
Class (NegP)
(Neg) QP
Figure 7.7 Clausal spine with optional projections for negation
200
AgrSP
NP AgrS’
John AgrS NegP
is TP ⇒ not TP
T FocP T vP
Foc ClassP v QP ⇒ happy
Class QP Q aP
Q NegP a √
Neg
Figure 7.8 Tneg-marker (not) inserted in the clausal spine
of insertion in the clausal spine. A similar hypothesis is also present in Endo

and Haegeman (2014) for the matching between internal and external syntax
of adverbial clauses, as well as in De Clercq and Vanden Wyngaerd (2017) for
the matching between the internal and external syntax of negation. See also
Williams (2003, 2009, 6–7) for a similar intuition. More specifically, the hy-
pothesis is that the more structure the nano–spine consists of, the higher it
will be inserted in clausal syntax. The structure I propose for a sentence like
(39), which contains a Tneg-marker, is Figure 7.8: the complex nanospine that
involves all negative features will be inserted in the highest position for nega-
tion in the structure, that is, a NegP dominating TP. I assume that the copular
verb ‘be’ raises from T° to the head of AgrSP and that the subject is in Spec-
AgrSP (Belletti 2000, 489; see also Pollock 1989; Griffin 2003).13
(39) John is not happy.
The same procedure applies with negative markers that consist of fewer
features. In that case, depending on the number and type of features in the
nanospine, a NegP will project on FocP, ClassP, or QP in the clausal spine and
host the relevant complex nanospine in its specifier. I will leave a more de-
tailed account of the procedure that regulates insertion of complex negative
specifiers into the main spine as a topic for future research (see Starke in
Chapter 9 for a proposal on how to insert complex specifiers).
13. AgrSP is nowadays reinterpreted as SubjP (see Cardinaletti 1997; Rizzi 2003).

7.6 CONCLUSION
This chapter illustrates how syncretisms can function as a probe into the in-
ternal structure of negative markers. Syncretisms show how different types
of negative markers must be underlyingly structurally related and share
properties. At the same time, the fact that languages also differ as to which
markers are syncretic mirrors the internal differences between those struc-
turally related negative markers. Morphology thus turns out to be a guide in
uncovering the internal syntax of what is often considered an indivisible unit,
that is, negation. The fact that the syncretisms parallel the natural semantic
scope of negation shows that form–meaning correspondences in the func-
tional domain are not completely arbitrary.
The syncretism patterns discussed in this chapter have revealed that a
negative marker can be decomposed into Neg and up to four additional
features: Q, Class, Foc, and T. Therefore it is fair to say that negativity, encoded
by the Neg feature, hardly ever occurs in its pure form and that consequently
the semantics of negation cannot be equated to the semantics of negative
markers.
REFERENCES
Allen, M. R. 1978. Morphological Investigations. Doctoral dissertation, University of

Connecticut.
Baker, Mark and Jim McCloskey. 2007. “On the Relationship of Typology to Theoretical
Syntax.” Linguistic Typology 11: pp. 285–296.
Belletti, Adriana. 2000. “Agreement Projections.” In The Handbook of Contemporary
Syntactic Theory, edited by Mark Baltin and Chris Collins, pp. 483– 510.
Oxford: Blackwell.
Belletti, Adriana. 2001. “Inversion as Focalization.” In Subject Inversion in Romance and
the Theory of Universal Grammar, edited by Aafke Hulk and Jean-Yves Pollock,
Belletti, Adriana. 2004. “Aspects of the Low IP Area.” In The Structure of IP and CP: The
Cartography of Syntactic Structures, Vol. 2, edited by Luigi Rizzi, pp. 16–51.
Borer, Hagit. 2005. In Name Only. Oxford: Oxford University Press.
Borschev, Vladimir, Elena Paducheva, Barbara Partee, Yakov Testelets, and Igor
Yanovich. 2006. “Sentential and Constituent Negation in Russian BE-Sentences
Revisited.” In Formal Approaches to Slavic Linguistics: The Princeton Meeting 2005
(FASL 14), edited by Hana Filip, Steven Franks, James Lavine, and Mila Tasseva-
Kurktchieva, pp. 50–65. Ann Arbor: Michigan Slavic Publications.
Bresnan, Joan. 1973. “Syntax of the Comparative Clause Construction in English.”
Linguistic Inquiry 4: pp. 275–343.
Butler, Jonny. 2003. “A Minimalist Treatment of Modality.” Lingua 113: pp. 967–996.
Cardinaletti, A. 1997. “Subjects and Clause Structure.” In The New Comparative Syntax,
edited by Liliane Haegeman, pp. 33–63. London: Longman.
202
Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective.

Collins, Chris and Paul Postal. 2014. Classical NEG Raising. Cambridge MA: MIT Press.
Corver, Norbert. 1997. “Much- Support as a Last Resort.” Linguistic Inquiry 28:
pp. 119–164.
Culicover, Peter. 1991. “Polarity, Inversion and Focus in English.” In Proceedings of the
Eight Eastern States Conference on Linguistics ’91, edited by German F Westphal,
Benjamin Ao, and Hee-R ahk Chae, pp. 46–68. Baltimore: ESCOL.
De Clercq, Karen. 2013. A Unified Syntax of Negation. Doctoral dissertation, Ghent
University.
De Clercq, Karen. 2017. “The Nanosyntax of French Negation.” In Negation: Syntax,
Semantics, and Variation, edited by Silvio Cruschina, Katharina Hartmann, and
Eva-Maria Remberger, pp. 49–80. Göttingen, Germany: V&R Unipress/Vienna
University Press.
De Clercq, Karen. In prep. The Morphosyntax of Negative Markers. Ms., Ghent University.
De Clercq, Karen and Guido Vanden Wyngaerd. 2017. “Why Affixal Negation is
Syntactic.” In Proceedings of the 34th West Coast Conference on Formal Linguistics,
edited by Aaron Kaplan et al., pp. 151– 158. Somerville, MA: Cascadilla
Proceedings Project.
Dryer, Matthew. 2011. “Order of Negative Morpheme and Verb.” In The World Atlas of
Language Structures Online, edited by Matthew Dryer and Martin Haspelmath,
Munich: Max Planck Digital Library. [online] Available at URL http://wals.info/
chapter/143.
Endo, Yoshio and Liliane Haegeman. 2014. “Adverbial Clauses and Adverbial Concord.”
In Proceedings of FAJL 7. Formal Approaches to Japanese Linguistics. MIT Working
Papers in Linguistics, edited by Shigeto Kawahara and Mika Igarashi, pp. 25–44.
Englebretsen, George. 1980. “On Propositional Form.” Notre Dame Journal of Formal
Logic XXI (1): pp. 101–110.
Funk, Wolf- Peter. 1971. “Adjectives With Negative Prefixes in Modern English
and the Problem of Synonymy.” Zeitschrift für Anglistik und Amerikanistik 19:
pp. 364–386.
Grevisse, Maurice and Andre Goosse. [1936] 1993. Le bon usage. Paris: Duculot.
Griffin, William E. 2003. “The Split-INFL Hypothesis and AgrsP in Universal Grammar.”
In The Role of Agreement in Natural Language. TLS 5 Proceedings, edited by William
E. Griffin, pp. 13–24. Austin: Texas Linguistic Forum.
Haegeman, Liliane. 1995. The Syntax of Negation. Cambridge: Cambridge University Press.
Haegeman, Liliane. 2000. “Negative Inversion, the Neg-Criterion and the Structure of
the CP.” In Negation and Polarity, edited by Laurence Horn and Yasuhiko Kato,
pp. 29–69. Oxford: Oxford University Press.
Han, Chung- Hye and Maribel Romero. 2001. “Negation, Focus and Alternative
Questions.” In WCCFL 20: Proceedings of the 20th West Coast Conference on
Formal Linguistics, edited by Karine Megerdoomian and Leora Anne Bar-el,
pp. 101–114. Somerville, MA: Cascadilla Press.
Haspelmath, Martin. 2011. “Negative Indefinite Pronouns and Predicate Negation.” In
The World Atlas of Language Structures Online, edited by Matthew S. Dryer and
Martin Haspelmath. Munich: Max Planck Digital Library. [online] Available at
http://wals.info/chapter/115.
Herburger, Elena. 2000. What Counts. Focus and Quantification. Cambridge, MA:
MIT Press.

Holmberg, Anders. 2003. “Questions, Answers, Polarity and Head Movement in
Germanic and Finnish.” Nordlyd 31: pp. 88–115. [online] Available at: http://
septentrio.uit.no/index.php/nordlyd/index
Horn, Laurence. 1989. A Natural History of Negation. Chicago: University of Chicago
Press.
Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA:
MIT Press.
Jayaseelan, Karattuparambil. 2001. “IP-Internal Topic and Focus Phrases.” Studia
Linguistica 55: pp. 39–75.
Jayaseelan, Karattuparambil. 2008. “Topic, Focus and Adverb Positions in Clause
Structure.” Nanzan Linguistics 4: pp. 43–68.
Jespersen, Otto. 1917. Negation in English and Other Languages. Copenhagen: A.F.
Høst & Søn.
Kandybowicz, Jason. 2013. “Ways of Emphatic Scope-Taking: From Emphatic Assertion
in Nupe to the Grammar of Emphasis.” Lingua 128: pp. 51–71.
Kjellmer, Göran. 2005. “Negated Adjectives in Modern English.” Studia Neophilologica
77: pp. 156–170.
Klima, Edward S. 1964. “Negation in English.” In The Structure of Language, edited
by Jerry A. Fodor and Jerrold J. Katz, pp. 246–325. Englewood Cliffs, NJ:
Prentice-Hall.
Kratzer, Angelika. 1991. “The Representation of Focus.” In Semantics: An International
Handbook of Contemporary Research, edited by Arnim von Stechow and Dieter
Wunderlich, pp. 825–834. Berlin: Mouton de Gruyter.
Laka, Itziar. 1990. Negation in Syntax: On the Nature of Functional Categories and
Projections. Doctoral dissertation, MIT.
Laka, Itziar. 1994. On the Syntax of Negation. New York: Garland.
Lieber, Rochelle. 1981. On the Organization of the Lexicon. Doctoral dissertation, MIT.
Distributed by IULe.
Neeleman, Ed and Reiko Vermeulen. 2012. “Types of Focus and Their Interactions With
Negation.” In The Syntax of Topic, Focus and Contrast, edited by Ed Neeleman and
Reiko Vermeulen, pp. 227–264. Berlin and Boston: Walter de Gruyter.
Ouhalla, Jamal. 1991. Functional Categories and Parametric Variation. New York:
Routledge.
Poletto, Cecilia. 2008. “On Negative Doubling.” In Quaderni di Lavora ASIt, edited by
Jacopo Garzonio, pp. 57–84. Venice: University of Venice.
Rizzi, Luigi. 2003. “On the Form of Chains: Criterial Positions and ECP Effects.” In
On Wh-Movement, edited by Lisa L. Cheng, and Norbert Corver. Cambridge,
MA: MIT Press.
Rooryck, Johan. 2010. A Compositional Analysis of French Negation. Ms., Leiden
University.
Rooth, Mats. 1985. Association with Focus, Doctoral dissertation, University of
Massachusetts.
Rooth, Mats. 1992. “A Theory of Focus Interpretation.” Natural Language Semantics 1:
pp. 75–116.
Rowlett, Paul. 1998. Sentential Negation in French. Oxford: Oxford University Press.
Russell, Bertrand. 1940. An Inquiry Into Meaning and Truth. New York: Routledge.
Selkirk, Elizabeth O. 1984. Phonology and Syntax: The Relation Between Sound and
Structure. Cambridge, MA: MIT Press.
204
Siegel, Dorothy. [1974]1979. Topics in English Morphology. New York: Garland.

Starke, Michal. 2004. “On the Inexistence of Specifiers and the Nature of Heads.” In
Structures and Beyond: The Cartography of Syntactic Structures, edited by Adriana
Belletti, pp. 252–268. Oxford: Oxford University Press.
nordlyd/index
Starke, Michal. 2011a. Class Notes, Course on nanosyntax given at Ghent University.
November 7–10.
Starke, Michal. 2011b. “Towards Elegant Parameters: Language Variation Reduces to
the Size of Lexically Stored Trees.” Ms., Tromsø University. [online] Available at
lingbuzz/001183.
Warren, Beatrice. 1984. Classifying Adjectives. Gothenburg, Sweden: Acta Universitatis
Gothoburgensis.
Williams, Edwin. 2003. Representation Theory. Cambridge, MA: MIT Press.
Williams, Edwin. 2009. Regimes of Derivation in Syntax and Morphology. London:
Routledge.
Zanuttini, Raffaella. 1996. “On the Relevance of Tense for Sentential Negation.” In
Parameters and Functional Heads, edited by Adriana Belletti and Luigi Rizzi,
pp. 181–207. Oxford: Oxford University Press.
Zanuttini, Raffaella. 1997. Negation and Clausal Structure: A Comparative Study of
Romance Languages. Oxford: Oxford University Press.
Zimmer, Karl. 1964. Affixal Negation in English and Other Languages: An Investigation of
Restricted Productivity. London: William Clowes and Sons.

CHAPTER 8
Nanosyntax of Russian Verbal Prefixes

INNA TOL SKAYA
8.1 INTRODUCTION
The aim of this chapter is to demonstrate the explanatory potential of a

nanosyntactic approach applied to the case of Russian polysemous prefixes. A fine-
grained decomposition of P elements (prefixes and prepositions) makes feasible
an analysis that preserves a single lexical entry per prefix, with shared syntactic
information, while deriving the semantic difference from different structural
positions. The most novel contribution of the analysis presented here is that the
compatibility of a prefixed verb with PP complements is linked to the inner struc-
ture of the prefix. The PP complement is argued to mirror the prefix structure.
It turns out that different uses of a polysemous prefix not only have related
meanings, but also their selectional properties coincide, even when meaning
similarities are far from obvious. This supports the idea that different uses of
a polysemous prefix share a single lexical entry containing syntactic informa-
tion. If the prefixes were simply homophonous, such systematic coincidence
of combinatorial properties would be unexpected.
When a prefix has a spatial meaning, a prefixed verb often combines with a
path complement, headed by a preposition, related to the prefix. For example,
consider the source–goal asymmetry in Markovskaya (2006). The verbs with
goal prefixes (e.g. ‘into’) may combine with a goal PP (1a), but not with a source
PP (e.g. ‘out of’) (1b). This contrasts with source prefixes, in which case both
source (1c) and goal (1d) are available.
(1) a. My za-šli v dom.

we into-went in house.acc
‘We went into the house.’
206
b. * My za-šli iz doma.
we into-went out.of house.gen
‘*We entered out of the house.’
c. My oto1-šli ot doma.
we from-went from house.gen
‘We went away from the house.’
d. My oto-šli v tenj.
we from-went in shade.acc
‘We went away into the shade.’
I suggest that this asymmetry has a structural explanation: The source-

prefixed verb may (context allowing) combine both with a goal and with a
source PP, whereas a goal prefix may combine only with a goal PP because
source is structurally more complex than goal, as seen in Figures 8.1 and 8.2.
Source
Goal
PlaceP ...
Figure 8.1 Source structure
Goal
PlaceP ...
Figure 8.2 Goal structure
I show that this asymmetry is not unique to the path dimension, but is
also observed with respect to scalar change and in the temporal domain (with
superlexical prefixes discussed in Section 8.2.2). In (2a,b) we see that the verb
prefixed with the goal prefix za-is compatible with goal state PP, but not with
source. In (2c,d), on the other hand, we see that the source prefix ot- is com-
patible with PPs specifying both source (ot-) and goal (do-).
(2) a. On za-stiral rubašku do dyr.

he into-washed shirt till holes
‘He washed the shirt ragged.’
b. * On za-stiral rubašku ot kraski.
he into-washed shirt from paint
intended: ‘He washed the paint off the shirt.’
1. Note that -oto-is the allomorph of -ot-that appears before consonant clusters.

c. On ot-stiral rubašku do belizny.
he from-washed shirt till whiteness
‘He washed the shirt white.’
d. On ot-stiral rubašku ot kraski.
he from-washed shirt from paint
‘He washed the paint off the shirt.’
I argue that the decomposed structure of the lexical entry of a polyse-

mous P element (prefix or preposition) remains the same under different
interpretations, which accounts for the parallelism of semantic and syntactic
properties. The differences arise from the different sites of attachment of the
P-element to the separate subevents in the syntactically represented event
structure of the prefixed verb (following Ramchand’s 2008a event decomposi-
tion). The prefix may attach either in the result phrase or in the aspect phrase,
as seen in Figure 8.3.
In Section 8.2 I briefly present the theoretical background of the study,
particularly Pantcheva’s (2011) path decomposition, the distinction between
lexical and superlexical prefixes, and path–scale–time parallel. In Section 8.3
I present the data, comparing goal, source, and route prefixes in three different
contexts in which they combine with a path, with a scale, and with the tem-
poral trace of the event. In Section 8.4, I present the nanosyntactic analysis
of the data.
Asp1P
subj
Asp1 PP
Perf.
subj
P ScaleP
prefix
Scale AspP
subj
Asp InitP (causign projection)
Imperf.
DP3
subj. Init ProcP (process projection)
DP2
subj Proc ResP (result projection)
DP3
Res PP
subj
subj
P Scale ei
Prefix
Figure 8.3 Prefix in result phrase or aspect phrase
N a n o s y n ta x of R u ss i a n V e r b a l P r e f i x e s [ 207 ]
208
8.2 THEORETICAL BACKGROUND
To derive the semantic and syntactic properties of prefixes from structure,

I adopt a neoconstructivist viewpoint in which the meaning of a morpheme
rests partly on what is brought from the lexicon and partly on the syntactic
structure (see Borer 2005 and Ramchand 2008b).
Once we have two sources for the meaning of a single morpheme, the con-
ceptual entry may remain the same across different instantiations of a single
prefix. In nanosyntax “a lexical entry is made up of three elements which are
linked together: (i) phonological structure, (ii) syntactic (SMS) structure (i.e.
an L-tree), and (iii) conceptual structure” (Baunaz and Lander Chapter 1, [26]).
Thus, although a prefix can be inserted at several sites in the verbal structure
interacting with different scales, what remains constant is the conceptual struc-
ture (specifying the relationship of the event to a scale) and the syntactic struc-
ture (accounting for the kinds of scales the prefix is compatible with).
As specified in its lexical entry, a prefix uniformly interacts with a scale (see
Kagan’s 2013 Scale Hypothesis), lexicalized by the verb or the verbal comple-
ment. The possible scalar complements include paths, volume–extent scales,
incremental themes, scales of change lexicalized by verbs, and temporal scales.
8.2.1 Decomposed Path
This chapter extends Pantcheva’s (2011) analysis of path to scales of change

and to the temporal dimension. The fine-grained substructure of events and
prefixes allows for more complex syntactic interaction between them, so that
a range of interpretations of a polysemous prefix may arise from syntax,
whereas the lexical entry remains sufficiently abstract and constant.
From both semantic considerations and crosslinguistic syncretism patterns,
Pantcheva (2011) argues for a universal path hierarchy. The Path Phrase is
decomposed into Goal, Source, and Route, in which Route is syntactically the
most complex projection, containing both Source and Goal (Figure 8.4).
Route
Source
Goal Place
Figure 8.4 Route structure
Source contains Goal and Place (Figure 8.5), and Goal is the simplest of the
three, containing the Place projection (Figure 8.6).
The classification can be visualized as in Table 8.1 (Zwarts 2005). The pluses
indicate being at a place, and the minuses indicate not being there. The points
0 and 1 mark the starting point and the end point of the path, respectively. For

Source
Goal
PlaceP ...
Figure 8.5 Source structure
Goal
PlaceP ...
Figure 8.6 Goal structure
Table 8.1 VISUALIZING GOAL , SOURCE,

AND ROUTE (ZWARTS 2005, 34)
Goal Source Route
– – – +++ +++ – – – – – – +++ – – –
0..............1 0..............1 0 .....................1
example, if the goal path is ‘to the lake,’ at the starting point 0 the figure is not
at the lake, as indicated by minuses, and at the end point 1 the figure is at the
lake, as indicated by pluses. For a source path, for example, ‘from the lake,’ the
reverse is true: The figure is at the lake at the starting point 0 and away from
the lake at the end point 1. In the case of a route path, for example, ‘across the
bridge,’ there are two transitions: At the starting point 0 the figure is not on
the bridge, and at the end point 1 the figure is also off the bridge (hence the
minuses), but for a period between start and end points, the figure is on the
bridge, as indicated by the pluses.
The prefixes that this chapter concentrates on are classified into goal,
source, and route as in Table 8.2.
Table 8.2 PREFIX CL ASSIFICATION
Meaning
Prefix Lexical Superlexical
Goal za- Illative Inceptive

– – – +++
Source ot- Ablative Completive
+++ – – – s- Superelative ‘there and back’
Route pro- Perdurative Duration

– – – +++ – – – pere- Translative excessive duration
do- Adlative Completive
210
In Tolskaya (2014a, 2014b) I argue that each transition type may be applied
to the spatial or temporal domain. In this chapter I extend the analysis to other
domains and argue for a syntactic explanation. The argument is based, cru-
cially, on the distinction between lexical and superlexical prefixes presented
in Section 8.2.2.
8.2.2 Lexical and superlexical pref ixes
It has been widely recognized that verbal prefixes in Slavic languages form
a heterogeneous class as to their semantic and syntactic properties, falling
into at least two types, lexical prefixes and superlexical prefixes, or internal
and external. The heterogeneity of P elements is not unique to Slavic, as,
for example, Swedish verb–particle constructions can also be divided into
two classes (named predicative and non-predicative by Vinka 1999) based on
whether a resulting state is denoted.
For Russian, this distinction was established and extensively motivated
in Isačenko (1960), Schoorlemmer (1995), Babko-Malaya (1999), Romanova
(2004), Svenonius (2004a), Di Sciullo and Slabakova (2005), and Zaucer
(2009).
Lexical prefixes, as potential argument-structure modifiers, are generated
in a position inside VP. Superlexical prefixes modify the event itself and do
not change the argument structure or the core meaning of the base verb and
are therefore syntactically higher, above the aspect head (Pereltsvaig 2006).
According to Romanova (2004), lexical prefixes attach mostly to perfective
or telic stems (if the verb is supplied with the option), allow the verb to form
secondary imperfectives, cannot stack, do not measure over objects, and can
change the argument structure of the verb. This behavior corresponds to a low
prefix position inside VP.
Superlexical prefixes attach to imperfective or atelic stems, do not allow
the verb to form secondary imperfectives, can stack, can measure over events
or objects, and do not change the argument structure of the verb, referring
to the temporal component of the event, without affecting the meaning of
the main verb. Superlexical prefixes are structurally reminiscent of English
particles, which are claimed by Newell (2008) to be late-adjoined to a null
aspectual head.
Lexical prefixes are closer to the root and change the lexical meaning of the
verbal stem. In the First Phase Syntax framework (Ramchand 2008a), lexical
prefixes are in the ResP (Result Phrase). Contra Ramchand (2008a), however,
I suggest that the prefix does not lexicalize the res head, but is a phrase in the
specifier of the PP adjoined to the res head (based on Svenonius 2008 and
Pancheva 2012). I suggest that the prefix phrase combines with a scale meas-
uring the extent of the result state.

For example, in (3a) the result is mapped to the end of the directed path
covering five kilometers, and in (3b) the result of reading is mapped to the end
of five pages. The prefix is not obligatory here; the ScaleP corresponds to five
kilometers or five pages.
(3) a. On pro-šel pjatj kilometrov.

he through-walkeddir five kilometers
‘He walked five kilometers.’
b. On pro-čital pjatj stranic.
he through-read five pages
‘He read five pages.’
A superlexical prefix, on the other hand, attaches above AspP, mapping the
event to the relevant subpart of the temporal trace (e.g. to the end of the tem-
poral interval of five hours). The crucial difference is that there is no ResP in
this case. In (4) we see the nondirectional form of the the root, and the prefix
is superlexical, whereas in (3a) the verb root is directional and the prefix is
lexical.
(4) On pro-xodil pjatj chasov.

he through-walkednon-dir five hours
‘He walked for five hours.’
Crucially, the same vocabulary item may occur in different positions and so
act both as lexical and superlexical, with interpretations so different that they
have been claimed to be homophonous. However, there is an abstract piece of
meaning that they have in common, that is, establishing a specific relation-
ship between the event and the scale, and the difference comes from the kind
of scale that this relationship is applied to and is hence systematically predict-
able from structure.
8.2.3 Scale–p ath parallel
I argue in Tolskaya (2014a) that polysemous prefixes uniformally serve to re-

late the event to a scale, but the scale may be contributed by the verb’s lexical
entry, by an incremental theme, by syntax, or by context, and this variation is
responsible for the differences in prefix interpretation.
The lexical entry of a prefix specifies a relationship between an event and a
scale with respect to which the event is measured out.
Structurally, scales and paths are very similar and may be classified
according to their shape in a parallel fashion. Both scales and paths can be open
or bound on one or both ends. Thus the scale typology of degree-modifiable
212
adjectives in (5), suggested by Kennedy and McNally (2005), closely mirrors

the path typology in Table 8.1 suggested by Zwarts (2005).
(5) Scale Typology (Kennedy and McNally, 2005)

1. <D(0,1),R,Δ> (TOTALLY) OPEN SCALE (#slightly, #perfectly, #almost tall)
2. <D[0,1),R, Δ> LOWER CLOSED SCALE (slightly, #perfectly, #almost dirty)
3. <D(0,1],R, Δ> UPPER CLOSED SCALE (#slightly, perfectly, almost clean)
4. <D[0,1],R, Δ> (TOTALLY) CLOSED SCALE (slightly, perfectly, almost full)
Evidence for this typology comes from the distribution of degree modifiers
such as ‘slightly’ and ‘perfectly,’ which pick up the minimum and maximum
degrees on the scale. Thus, lower closed-scale adjectives, such as ‘dirty,’ pick
out a minimum on the scale, for example, if the object is minimally dirty it is
true that it is dirty. The upper closed-scale adjectives, such as ‘clean,’ pick out
the maximum point on the scale, so the statement that something is clean is
true only if the maximum cleanness for that object is reached.
Path classification is notably reminiscent of the scale classification.
A source path is a subcase of a lower closed scale, as both have a minimum
point, to which the initial subevent can be mapped, but both are open on the
other end. A goal path is a subclass of an upper closed scale, as both have a
maximal point.
8.3 PREFIX–P P COMPATIBILITY
In this section I show how the prefixes combine with path or scale and how it
may (and may not) be depicted with complement PPs.
8.3.1 Goal
The prefix za- is notoriously versatile, but its diverse meanings may be uni-
fied as denoting a transition into a new place, a state, or an activity. The new
state is usually recoverable from context or explicitly stated. The subject can
be viewed as a figure, in which the place, state, or activity is the ground.
With directional verbs, the prefix modifies path, so that the figure enters
into some closed space, as in (6a). With nondirectional verbs the figure
enters a new activity, for example, the clock enters a working state in (6b).
In (6c) the figure is tea: The tea leaves and water gradually enter a new state
in the process of brewing, that is, of becoming tea. The goal place (‘into the
house’) or the goal state (e.g. the taste or color achieved by the tea) may be
specified.

(6) a. Čelovek za-šel v dom.
man into-walkeddir into house.acc
‘The man walked into the house.’
b. Časy za-xodili.
clock into-walkednon-dir
‘The clock started to work.’
c. Xozjain za-varil čaj (do černoty / do goreči
host into-boiled tea (till blackness / till bitterness)
‘The host brewed tea (black/bitter).’
As a lexical prefix, za- denotes a monotransitional path into a goal location

with motion verbs (6a) and into a new state with verbs of change (6c). As a
superlexical prefix, it gives rise to an inceptive meaning (6b). Thus the seem-
ingly incompatible completive (6c) and inceptive (6b) meanings are derived
from the same lexical entry (which includes a plus to minus transition) when
it attaches in different positions: A transition into a new state is interpreted
as a completion of an eventuality, whereas a transition into a new activity is a
beginning of that activity.
8.3.2 Source
The goal prefixes contrast sharply with source prefixes. The Source projec-
tion is syntactically more complex than Goal and takes Goal as an argument
(as shown in Figures 8.5 and 8.6). Source is the locus of a semantic reversal
operation (Pantcheva 2011), that is, it takes a minus-to-plus Goal path and
reverses it into a plus-to-minus monotransitional Source path. Thus, a path
directed into a certain location becomes a path directed out of the location.
See (7).
(7) The syntactic and semantic derivation of a Source path (Pantcheva,

2011, 72)
a. [Place...]
b. merger of Goal →
c. [Goal [Place . . . ]] representing a path of the shape − − − +++
d. merger of Source →
e. [Source [Goal [Place . . . ]]] → reversal of Goal path +++ − − −
Similarly, the completion projection reverses the meaning of the incep-

tion projection, creating a new type of transition event, in which the lower
and upper bounds change places. The subject is involved in the activity (this
214
corresponds to pluses in Zwarts’s visualization) at the lower bound and not

involved (corresponds to minuses) at the upper bound.2 See (8):
(8) The syntactic and semantic derivation of completion

a. [Aspect [Init [Proc]]] provides the activity and initiator
b. merger of Inception →
c. [Inception [Aspect . . . ]] introduces the transition − − − +++
d. merger of Completion →
e. [Completion [Inception [Aspect . . . ]]] →
f. reversal of the previous transition: +++ − − −
In (9a) the boy is near the fire at the start of the event and away from the fire
at the end of it, so the plus-to-minus transition is from +near fire to –near fire.
In (9b) the transition is from +dirty to –dirty. Just as the position near the fire
is interpreted as undesirable in (9a), the tablecloth is in an undesirable dirty
state at the start of the event and is not dirty in the final result state. In (9c)
there is also a plus to minus transition, that is, the grove is involved in the ac-
tivity (+talking) before the event, but transitions into silence.
(9) a. Maljčik ot-

skočil ot kostra.
boy from-jumpeddir from fire
‘The boy jumped away from the fire.’
b. Xozjajka ot-stirala skatertj.
hostess from-washed table-cloth.acc
‘The hostess washed the tablecloth’.
c. Ot-govorila rošča zolotaja.
from- talked grove golden
‘lit.: The golden grove completed talking.’
‘The golden birch-tree grove has fallen silent.’
[Sergey Esenin, The Golden Grove (1895–1925).
Translation by А. Vagapov)
Another prefix involving the same kind of transition is s- . The prefix
s-involves displacement from a “regular” initial location in case of direc-
tional verbs (10a), a change from the initial state with verbs of change
(10b), and a brief trip there and back, with subsequent return, in the case
2. Intuitively, one might expect the parallel among goal and completion and source
and inception, but the goal prefix za-clearly denotes inception and the goal prefix ot-
clearly denotes completion, and the visualization with pluses and minuses helps to
explain this counterintuitive alignment.

of nondirectional verbs (10c). Тhe meaning component of ‘return’ seems
unique to nondirectional motion verbs and may be derived from the com-
bination of the meaning components contributed by the verb and the
prefix: Source prefix contributes completion of the activity (transition from
+running to –running); nondirectionality of the motion verb ensures the
return (getting to a goal and staying there is not a part of the verb meaning,
whereas a series of there-and-back trips is a possible meaning of the nondi-
rectional verb begat’ ‘run’), uniqueness and briefness of the trip are ensured
by the briefness or shortness of the transition that is a part of the prefix’s
lexical entry.
(10) a. Poezd so-shel s reljs.

train off-walked from rails
‘The train derailed.’
b. Osenju ix so-greet ogonj v kamine.
in.fall them off-warm fire in fireplace
‘In the fall the fire in the fireplace will make them warm.’
c. S-begaj za pivom!
off-runnon-dir for beer
‘Run get some beer (quickly, and then return)!’
The path is of the Source type (+++ – – –, i.e. the figure is at a certain loca-
tion at the beginning of the path, and away from it at the final point). The
initial relationship between figure and ground (‘near’ in the case of ot- and
‘on’ in the case of s-) must be a part of the lexical entry of the prefixes that
makes them different, in spite of the shared-path-type features. In (10b) there
is also a plus-to-minus transition, that is, movement from an original state of
being cold to a noncold, more comfortable state. In the superlexical usage, il-
lustrated in (10c), there is a punctual transition from running to not running,
that is, a plus-to-minus type of transition.
An important piece of evidence for the hierarchy and for the parallelism
of properties demonstrated by prefixes across space and time comes from
considerations of the possibility of modifying each transition point with prep-
ositional phrases.
Thus if a lexical prefix contains the Goal projection, the verb may be mod-
ified by a Goal PP, as in (11a) (see Markovskaya 2006). Crucially, the source is
not available for specification with a PP. However, with a Source prefix, both
Source and Goal are available for modification (11c)–(11f).
(11) a. My za-šli v dom.

we into-went in house.acc
‘We went into the house.’
216
b. * My za- šli iz doma.

we into-went out.of house.gen
‘*We entered out of the house.’
c. My oto- šli ot doma.
we from-went from house.gen
‘We went away from the house.’
d. My oto- šli v tenj.
we from-went in shade.acc
‘We went away into the shade.’
e. Deti s-katilisj s gorki.
children off-slid off slide.gen
‘The children slid down from the slide.’
f. Deti s-katilisj na pol.
children off-slid on floor.acc
‘The children slid down onto the floor.’
The same asymmetry holds with PPs modifying scales, lexicalized by the verbs.
A verb prefixed with a goal prefix may combine only with a PP containing a goal
state [(12a) and (12b)], whereas a source-prefixed verb may (context allowing)
combine both with a goal and with a source PP [(12c)–(12f)].
(12) a. On za-stiral rubašku do dyr.

he into-washed shirt till holes
‘He washed the shirt ragged.’
b. * On za-stiral rubašku ot kraski.
he into-washed shirt from paint
intended: ‘He washed the paint off the shirt.’
c. On ot-stiral rubašku do belizny.
he from-washed shirt till whiteness
‘He washed the shirt white.’
d. On ot-stiral rubašku ot kraski.
he from-washed shirt from paint
‘He washed the paint off the shirt.’
e. so-gretj moloko do komnatnoj temperatury
off-heat milk till room temperature
‘to heat milk till room temperature’
f. so-gretj ruki ot xoloda
off-heat hands from cold
‘to warm hands from the frost’

In (12a) and (12b) we see that a multitude of scales may be associated with the
verb. Normally, the verb ‘to wash’ lexicalizes a scale of cleanness, starting from a
dirty state and ending at a completely clean state. However, the prefix za-makes
it possible to invoke a new, contextually defined scale, for example, of the damage
done to the fabric as the result of overly frequent washing. Similarly, although the
verb varitj ‘cook, boil’ lexicalizes a scale of readiness, with the prefix za-and the
direct object ‘tea,’ a new scale becomes relevant, which is not a part of the lexical
entry of the verb ‘to boil,’ because the scale can measure darkness or bitterness.
The invoking of a nonlexicalized scale might be a special property of za-, as we
see a somewhat similar picture when it modifies path. The location entered does
not have to be the logical end of the trajectory along which the figure moves.
For example, za-jti v magazine ‘ZA-walk into shop’ can refer to a brief digression
from the path, for example, stopping by a shop on the way home. If the shop is
the main goal of the journey, rather than a stop along the way, a different goal
prefix may be preferred to denote the arrival, for example, vo-or pri-.
There is a minus-to-plus transition in each case, from not being in the shop
to being in the shop, or into a new state of being tea (6c), or into a new state of
having holes (12a). What is special about za-and what makes it so polysemous
is that the goal of the transition is context dependent.
Furthermore, what makes za-, as a goal prefix, different from source and
route prefixes is that it is incompatible with the overtly specifying source,
whether it is the source of the path as in (11a) or the original state as in (12a).
To sum up, the source prefixes, as opposed to goal prefixes, are compat-
ible with PPs specifying both source and goal, as we see in (12c)–(12f). In
these cases the prefix denotes a transition along the scale lexicalized by the
verb: cleanness scale with the verb ‘to wash’ and temperature scale with ‘to
heat.’ The source prefix denotes a plus-to-minus transition: from dirtiness to
cleanness, from being cold to normal temperature. Either the source (paint
stains, cold) may be overtly stated or the goal (whiteness, room temperature).
8.3.3 Route
The Route projection, according to Pantcheva (2011), contains both Source

and Goal projections. With route prefixes, it is possible to specify both source
and goal and to measure the route in addition.
When the route projection combines with the source projection, the
second transition is added, so the monotransitional source path turns into a
bitransitional path bounded on both ends.
The syntactic and semantic derivation of a route path (Pantcheva, 2011, 72)
looks as in (13):
(13) a. [Place...]

b. merger of Goal →
218
c. [Goal [Place . . . ]] representing a path of the shape – – –+++

d. merger of Source #
e. [Source [Goal [Place . . . ]]] # reversal of Goal path +++ – – –
f. [Route [Source [Goal [Place . . . ]]]] →
adding a second transition # – – – +++– – –
Duration in the superlexical domain is derived in a parallel fashion (14).
(14) a. [Aspect [Init [Proc]]] provides the activity and initiator

b. merger of Inception →
c. [Inception [Aspect . . . ]] introduces the transition – – –+++
d. merger of Completion →
e. [Completion [Inception [Aspect . . . ]]] →
reversal of the previous transition: +++ – – –
f. [Duration [Completion [Inception [Aspect . . ..]]]] →
adding a second transition: – – – +++ – – –
The route prefixes discussed here are pere-, pro-, and do-. The Route prefixes pere-
and pro- do not have corresponding prepositions and make it possible for the
complement of the verb they combine with to appear without a preposition. The
examples in (15a) and (15b) include the route prefix pere-. As illustrated in (15a)
both source and goal may be specified simultaneously. Furthermore, pere- and
pro-may introduce a direct object referring to the route traversed, expressed by
a DP complement (15b). Contrast this with za-/ot-/s- in (15c) and (15d).
(15) a. Mnogie pere-bežali iz odnogo lagerja v drugoj.

many over-ran from one.gen camp to other.acc
‘Many fled from one side to the other.’
b. Belka pere-bežala dorog-u.
squirrel over-ran road-acc
‘A squirrel ran across the road.’
c. * Mnogie za-/ot-3/s-bežali iz odnogo lagerja v drugoj.
many into-/from-/off-ran from one.gen camp to other.acc
‘Many fled from one side to the other.’
3. I have found a class of potential counterexamples, such as (i).

(i) Nemcy oto-šli čerez most za reku.
Germans from-walked across bridge beyond river.
‘The Germans crossed the river over the bridge.’
However, the PP čerez most is a modifier, not the complement of the verb in these
cases: There is usually a goal PP present or possible, and the PP čerez most specifies
means, not goal.

d. * Belka za-/ot-/s-bežala dorog-u.
squirrel into-/from-/off-ran road-acc
‘A squirrel ran across the road.’
Similarly, with change-denoting verbs like ‘heat,’ the goal may be specified
(16a) or the difference may be measured along the lexicalized scale, for ex-
ample, the distance from the boiling point (16b) on the temperature scale.
(16) a. pere-gretj par do 500 gradusov

over-heat steam till 500 degrees
‘to heat the steam up (lit. ‘overheat’) to 500 degrees’
b. pere-gretj vodu na desjatj gradusov vyše točki kipenija
over-heat water on ten degrees higher point boiling
‘to heat the water up (lit. ‘overheat’) to ten degrees above its boiling point’
The prefix pro- may also introduce the beginning and the end of a path, as in
(17a), an argument specifying the goal, as in (17b), and an argument meas-
uring the trajectory, as in (17c).
(17) a. Nikto ne smog pro-bežatj ot odn-ogo konc-a

no.one not could through-run from one-gen end-gen
do drugogo.
up.to other.gen
‘No one managed to run from one end to the other.’
b. On pro-bežal v svoj kabinet.
he through-ran in his office.acc
‘He ran by into his office.’
c. Ja pro-exal odnu ostanovku po Arbatu.
I through-rode one bus.stop.acc along Arbat.dat
‘I went one stop along Arbat Street by bus.’
Similarly, when the prefix pro- refers to a thorough performance along a scale
of change, both source (18a) and goal (18b) may be modified. With incre-
mental theme, the amount covered may be measured (18c).
(18) a. pro-mytj glaza ot pyli

through-wash eyes from dust
‘wash the dust out of one’s eyes thoroughly’
b. pro-
mytj steklo do prozračnosti
through-wash glass till transparency
‘wash the glass clear’
220
c. pro-čitatj tri stranicy / ot načala do konca

through-read three pages / from beginning till end
‘to read three pages /from beginning till end’
With do-it is also possible to specify goal, both source and goal, and the dis-
tance (19):
(19) a. Ona do-šla ot mašiny do podjezda.

she up.to-walked from car to entrance
‘She walked from the car to the entrance.’
b. Tem ne menee Sonya rešila do-beža-tj distanciju.
it.instr not less, Sonya decided up.to-run-inf distance
‘Nevertheless, Sonya decided to complete running the distance.’
Similarly, both source and goal (20a) and distance (20b) may be specified in
reference to a scale.
(20) a. Ne mogu ničego do-delatj ot načala do konca.

not can nothing up.to-do from beginning till end
‘I can’t complete anything from beginning till end.’
b. Nado do-delatj desjatok del.
need up.to-do ten deeds
‘It is necessary to complete a dozen things.’
Thus we can observe a tight relationship between the complexity of the
prefix with the complexity of the path-denoting PP selected by the prefix. If
the prefix contains only the Goal projection, only a Goal PP may be added. If
the prefix contains a Source, then both Source and Goal PPs may be added.
If the prefix contains a Route, then Goal, Source, and Route are allowed, as
well as a measure of the path length.
Table 8.3 summarizes the classification of the lexical prefixes and what may
be modified.
Table 8.3 COMPATIBILITY OF LEXICAL PREFIXES WITH PPS
Prefix Goal Source Route
– – – +++ za- yes * *

+++ – – – ot- yes yes *
+++ – – – s- yes yes *
– – – +++ – – – pro- yes yes yes
– – – +++ – – – pere- yes yes yes
– – – +++ – – – do- yes yes yes

8.3.4 Superlexical pref ixes
At the superlexical level, the starting point, the end point, and the duration of
an event may be modified by PPs or adverbs, parallel to the Source, Goal, and
Route at the lexical level. On the spatial level, the lowest projection denotes a
punctual transition from not being in a place or state to being there, that is,
minus-to-plus transition. On the temporal level, the shape of the lowest pro-
jection is parallel but the transition is from not being involved in an activity
to starting the activity, that is, inception. The Goal prefix za- thus can denote
inception when it appears in this position. The inception point can be modi-
fied by punctual adverbs only, as is shown in (21):
(21) a. Vdrug on za-begal po komnate.

suddenly he into-ran non-dir
along room
‘suddenly, he started running around the room.’
b. V 6 utra vse vstali, za-begali, za-sobiralisj . . ..
in 6 morning all woke.up into-rannon-dir into-get.ready
‘At six in the morning, everyone woke up, began to run about
and started to pack . . ..’
c. * S dvux do trex on za-begal po komnate.
from two to three he into-rannon-dir along room
‘Between two and three o’clock, he started running around
the room.’
Thus both Goal and Inception positions allow only one transition denoting an
adverbial expression or PP, so the shape of the temporal scale in this case is
similar to the shape of the goal path.
With completion prefixes such as ot- and s-, the point of completion may
be modified with ‘in an hour’ phrases, as illustrated in (22), and with phrases
specifying a single point in time, rather than a period of time. Duration may
not be measured, as the Route projection (corresponding to duration) is not
available.
(22) Tuda za desjatj minut s-bega-tj možno!

there in ten minutes off-run-inf possible
‘It is possible to run there and back in ten minutes!’
When the prefix specifies a bitransitional path with both the beginning and the
end of the activity, it is possible to measure its duration, as illustrated in (23):
(23) a. On desjatj lišnix minut pere-plaval.

he ten extra minutes over-swamnon-dir
‘He swam for ten extra minutes.’
222
b. On pro-plaval dva časa.

he through-swam two hours
‘He swam for two hours.’
c. On do-plaval dva časa.
he up.to- swam two hours
‘He completed swimming the two hours.’
Table 8.4 sums up the properties of the superlexical prefixes, resulting in the
tripartite classification coinciding with the classification of the lexical prefixes
in Section 8.3.3. Goal prefixes turn out to be compatible with ‘suddenly’ and
specific time modifiers in their superlexical use; source prefixes are compat-
ible with ‘in an hour’–type modification. Furthermore, the inception and com-
pletion verbs may be modified by phrases denoting exact time, and then the
point will coincide either with inception or with completion. Route prefixes
allow duration of an activity to be measured when they are superlexical.
Table 8.4 COMPATIBILITY OF SUPERLEXICAL PREFIXES WITH MODIFIERS
Prefix Specific Time ‘in an hour’ Duration

- - - +++ za- yes * *
+++ --- ot- yes yes *
+++ --- s- yes yes *
---+++ --- pro- * * yes
---+++ --- pere- * * yes
---+++ --- do- * * yes
8.4 SYNTACTIC ANALYSIS
In this section I show how a nanosyntactic framework allows an explanation

of the data in Section 8.3.
In cartographic approaches it has been claimed that Slavic prefixes, like
Germanic particles, are prepositional elements originating in a position
within the PP from which they move into the verbal domain (Svenonius
2004b; Romanova 2007; Pantcheva 2012; but see also Starke in Chapter 9,
Section 9.1.3 for discussion). As we have seen in section 8.3, prefixes are very
often identical to prepositions, but there are prefixes without a prepositional
counterpart (e.g. vy- and pri-), and some prefixes (pro- and pere-) allow the
complement of the prefixed verb to appear without a preposition.
Pantcheva (2012) presented an account of doubling, nondoubling, and
alternating prefixes based on nanosyntax, in which a prefix may lexicalize a
large chunk of structure corresponding to both prefix and preposition, and

may spell out prefix, preposition, or both, as long as the spelled out structure
is its subconstituent, according to the Superset Principle.
In this section I show how this account in combination with path decom-
position predicts the acceptable prefix–PP combinations in spatial, scalar, and
temporal domains.
8.4.1 Deriving doubling, nondoubling, and argument-c hanging pref ixes
Pantcheva’s (2012) proposal is that variation in prefix properties is the result

of the specific shape of the lexical entries. The syntactic structure is the same,
but it is spelled out differently by the three types of prefixes or particles.
The Superset Principle (24) allows a P’s lexical entry to be inserted into
different nodes as long as the nodes are subconstituents of the lexical structure.
(24) The Superset Principle

A lexical tree L can match a syntactic tree S if L is a superset
(proper or not) of S
(see Chapter 1, Section 1.3.3.2).
More informally put, the Superset Principle allows for an S-tree to be spelled
out by an L-tree as long as that L-tree is the same size or bigger.
Pantcheva (2012) makes the following assumptions: Slavic prefixes are
phrasal (Svenonius 2004b, 2008), they originate in a position inside the PP,
from where they move to the verbal domain (Ramchand and Svenonius 2002;
Svenonius 2004b, see also Starke in Chapter 9 for a discussion of prefixes),
and they incorporate an abstract GROUND element (Svenonius 2004b).
The following structure emerges, in which PrefP is spelled out by the prefix
and PathP by the preposition. This entire PP,4 projected by the preposition (the
Place head on the right), is in the complement of ResP, as in Figure 8.7.
PP
PerfP PathP
Path PlaceP Path PlaceP

Place DP
Place GROUND
Figure 8.7 Structure of prefix and preposition
4. I keep Pantcheva’s (2012) phrase labels, in which the prefP and PP are projected
by the Path head spelled out by the prefix (the box on the left), to avoid the confusion
arising from the matching labels of the prefix and the path phrases.
224
The argument-changing prefixes or particles allow the complement to

appear with or without a preposition (25), because they can lexicalize both
PrefP and PathP.
(25) a. pere-plytj rek-u

over-swimdir river-acc
‘to cross the river (swimming)’
b. pere-plytj čerez rek-u
over-swimdir across river-acc
‘to swim across the river’
In the tree in Figure 8.8 we see that the lexical entry of the prefix pere-
includes both PrefP and PathP, and so it may lexicalize either both the prefix
and the preposition as an argument-changing prefix (25a) or just the prefix
(25b). The same argument applies to pro-, which also allows for the comple-
ment to appear without a preposition. Note that the DP complement [river
in (25)] is not a part of the lexical entry of the prefix, so the PathP with the
DP complement is not a subconstituent of the prefix. So pere- could not be
inserted at the PP node if the DP were the complement of Place, because
such a DP is not a subconstituent of the tree stored in its lexical entry. This
problem is solved by spellout-driven movement (26), because inserting the
The lexical entry of pere:

pere- ⇔ </pere/, PP >
PrefP PathP
Path PlaceP
Path PlaceP
Place
Place GROUND
PP
DP PP ⇒ pere
river
pere ⇐ PrefP PathP ⇒ čerez

Place tDP
Place GROUND
Figure 8.8 Lexical entry of pere-and spelling out (25a) and (25b)

preposition pere- at PathP node necessitates evacuation of DP. Because the
Superset Principle (following Starke 2005–2011; Caha 2009) ignores traces,
the trace of the DP does not prevent insertion of the preposition.
(26) Spellout-driven movement

The shape of a lexical item can trigger movement of a syntactic constituent
such that the maximally matching configuration for insertion is obtained.
(Caha 2011; Pantcheva 2011; Starke 2014)
So, because the argument-changing prefixes contain both PrefP and PathP,
they have two options available: They may either lexicalize just the prefix part
or both prefix and preposition, in which case the direct object may appear
without a preposition.
This makes them different from doubling prefixes, which are identical to the
corresponding preposition (27). The lexical entry of a doubling prefix may spell
out either the prefix or the preposition, but, unlike the argument-changing
prefix, not both simultaneously. The Place node moves up to the Path node
by head-to-head movement. This spellout-driven movement ensures that the
prefP matches the lexical entry of the prefix (Figure 8.9).
(27) oto-jti ot doma

from-walk from house.gen
‘To walk away from the house’
The prefix can therefore be inserted at the PrefP node, and it can also spell
out the complex Path head under PathP (by the Superset Principle). When it
does, we have doubling.
The Lexical enrty of ot-: PrefP >

ot– ⇔ < /ot/,
Path PlaceP
Path Place GROUND
PP
PrefP PathP
Path Place
Path PlaceP
Path Place tPlace Path Place tPlace DP
GROUND
Figure 8.9 Lexical entry of ot-and spelling out doubling in (27)
226
The nondoubling prefixes (28) have lexical entries corresponding to PrefP,

so that PathP is not a subset of it; hence they can spell out a prefix, but not a
preposition (Figure 8.10).
(28) Jussi vy-pal iz okna.

Jussi out-fell from widnow
‘Jussi fell out from the window.’
The lexical enrty of vy-: PrefP >

vy- ⇔</vy/,
Path PlaceP
Place GROUND
PP
vy- ⇐ PrefP PathP⇒ vy-

Place DP
Place GROUND
Figure 8.10 Lexical entry of vy-and spelling out prefP (but not PathP) in (28)
A nondoubling prefix may be inserted at the PrefP node, but it cannot

spell out PathP. The PrefP in Figure 8.10 with GROUND does not count
as a superset of the (otherwise) identical tree without GROUND. The lex-
ical entry includes a GROUND node as a complement of Place, so even if
the DP is moved up, PathP does not match PrefP (because a legitimate
subconstituent of prefP must include the projecting head Place). The prep-
ositional structure (PathP) can instead be lexicalized by a preposition, for
example, iz ‘from.’
8.4.2 Decomposed pref ix and decomposed path
Now recall that the PathP may be decomposed into Route, Source, and Goal.
So the fully decomposed lexical entry of an argument-changing route prefix
looks as is shown in Figure 8.11.
Once the DP is evacuated (from the bottom of PathP) and adjoined to the
PP, the PP is a subset of the lexical entry of the prefix, and when no preposi-
tion is present, the entire structure is lexicalized by the prefix. Alternatively,
the PrefP is also a subset of the prefix’s lexical entry, in which case PathP may
be spelled out by a preposition.

The lexical entry of pere: PP >
pere- < /pere/,
PrefP PathP
Route Route
Source Source
Goal PlaceP
Goal PlaceP
Place
Place GROUND
Figure 8.11 More detailed lexical entry of pere-
Crucially, when the prefix spells out only the PrefP, the PathP does not have
to contain the same projections as the PrefP, but may contain a subset of the
path structure contained in the prefix, that is, RouteP, SourceP, or GoalP. An
example of a configuration (Figure 8.12) in which a route prefix combines with
a goal PP is given in (29):
(29) pere-bežatj v drugoj lagerj

over-run into other camp
PP
PrefP ⇒ pere- PathP
Route Path ⇒ v- PlaceP

Source Goal Place tP lace DP
Goal PlaceP
Place GROUND
Figure 8.12 pere-(route) and v (goal)
However, if the prefix does not contain the Route projection, the PathPs do
not match, so the Route projection is ruled out in the PathP as well [see (30)
and Figure 8.13.
(30) * za-jti čerez dorogu

into-go over road
It looks like there is a requirement for the complement path to match the
prefix. Because there is only one scale or path that the entire PP refers to, the
228
PP
za- ⇐ PrefP *PathP ⇒ čerez
PathP PlaceP Route

Source
Goal Place tPlace GROUND Goal PlaceP
Place
Figure 8.13 PathPs do not match
subparts of the PP may not be in conflict about bounding it. The prefix bounds
the path, so it establishes the type of path (goal, source, or route), whereas the
PathP specifies the subparts of the path introduced by the prefix. Thus, if the
prefix introduced the goal subpart of the scale, the PathP may overtly specify
this goal as, for example, ‘into the house,’ but because the prefix did not intro-
duce the source, the PathP refers to the path that does not contain a source, and
cannot specify it. Hence the framed part of the tree is not licensed by the goal
PrefP, and the route preposition is ungrammatical. If the prefix contains only a
goal projection, the PathP will also contain only a goal [see (31) and Figure 8.14].
(31) za-jti v dom

into-walk in house
PP
za- ⇐ PrefP PathP
PathP PlaceP
PathP ⇒ v dom PlaceP
Goal Place tPlace GROUND Goal Place tPlace Place
Figure 8.14 Matching PathPs (goals)
We see that in each case the structure lexicalized by the preposition is a

subconstituent of the structure lexicalized by the prefix, and the PathP mirrors
the PrefP. Thus the PathP cannot contain any projections not contained in the
PrefP, so a goal prefix does not combine with a source PP [see (32) and Figure 8.15.
(32) * za-jti ot doma

into-walk from house

*PP
za- ⇐ PrefP PathP
PathP PlaceP PlaceP

*PathP ⇒ ot doma
tPlace Place
Goal Place tPlace GROUND
Source
Goal Place
Figure 8.15 PathP of source PP contains structure not in the PathP of goal prefix
The source prefixes (ot-, s-) may combine with either source or goal
prepositions. (33) shows the source prefix ot-combined with a source PP (see
Figure 8.16 with matching PathPs). (34) shows the source prefix ot-combined
with a goal PP, which is possible because the PathP lexicalized by the preposi-
tion is a subset of the PathP lexicalized by the prefix (see Figure 8.17).
(33) oto-jti ot doma

from-walk from house
PP
ot- ⇐ PrefP PathP
PathP PlaceP PathP ⇒ ot doma PlaceP
tPlace Place
Source tPlace GROUND Source
Goal Place Goal Place
Figure 8.16 PathP of source prefix matches the PathP of the goal PP
(34) oto-jti v tenj

from-walk into shade
To sum up, a prefix is phrasal and originates in a PP in the verbal comple-

ment, from where it raises to adjoin to the verb. Both the prefix and the PP
contain a PathP, which may contain Route, Source, and Goal, hierarchically
ordered. Crucially, the restriction emerges against the PathP containing extra
projections not licensed by the PrefP, that is, the PathP lexicalized by the prep-
osition must be a subset of the path lexicalized by the prefix.
230
PP
ot- ⇐ PrefP PathP
PathP PlaceP PathP ⇒ υ tenj PlaceP
Goal Place tPlace Place

Source tP lace GROUND
Goal Place
Figure 8.17 PathP of source prefix contains all of the structure in the PathP of the goal PP
8.4.3 Decomposing scale
In this section I show that scale may also be decomposed into hierarchically or-
dered projections (Route, Source, Goal), similarly to path. ScaleP is also a possible
complement of the prefix, and in this case the prefix measures the extent of the
event on a scale, mapping the result state to the relevant subpart of the scale.
Goal projection, broadly interpreted, does not have to denote a spatial
transition, but an abstract minus-to-plus transition into the goal state on the
scale. Source projection, similarly, reverses the transition, resulting in a tran-
sition from the start state.
Thus, combining the decomposed path (Pantcheva 2011), decomposed prefix
(Pantcheva 2012), and scalar (Tolskaya 2014a) approaches, we get the structure
in Figure 8.18 for a prefixed scalar verb such as ot-varitj ‘to boil’ (35), which
ResP
lentils
Res PP
ot- ⇐ PrefP PathP
lentils
Path PlaceP ScaleP
Path
Source tPlace GROUND lentils ScaleP
Goal Place
Scale
Figure 8.18 Structure of prefixed scalar verb ‘to boil’

denotes a plus-to-minus transition along a scale of readiness from the original
raw state. The PP containing the prefix is the complement of the Res head and
takes the ScaleP as a complement, so the PP maps the result state to the scale.
The direct object (e.g. ‘lentils’) moves from the position in the specifier of the
ScaleP to the specifier of the PP to specifier of the ResP, and to the specifier of
ProcP, so that it has multiple semantic roles simultaneously. The object must
be a part of the ScaleP, as the scale depends on the object, as the readiness
of lentils is quite different from the readiness of an egg, so the direct object
provides the reference standard for the scale. The object is also the specifier
of the PP, as it is the lentils undergoing the transition along the scale, denoted
by the PP. The direct object is also the resultee and the undergoer of the event.
(35) ot-varitj čečevicu

from-cook lentils
In Figure 8.18 the scale and path are unpronounced, the scale is lexicalized
by the verb, as the verb ‘to cook’ is a scalar change verb, and hence provides
a property scale (Levin and Rappaport Hovav 2006, 2010; Rappaport Hovav
2008). However, path and scale can also be overtly specified, as in (36), where
the source prefix combines with a goal preposition, just like with a spatial
path. Figure 8.19 shows the PP containing the trace of the readiness scale
(which has moved up because of spellout-driven movement).
(36) ot-varitj čečevicu do gotovnosti

from-cook lentils till readiness
PP
ot- ⇐ PrefP PP
lentils PP ⇒ do-
Path PlaceP
Goal
Source tPlace GROUND tScaleP
Place
Goal Place
readiness
Figure 8.19 Path and scale overtly pronounced
When the prefix pro- combines with an incremental theme verb, the in-
cremental theme acts as a scale. For instance, in ‘to read a book,’ the book
provides the scale of how much was read. In this case the direct object is in a
tight relationship with the prefix, as in ‘to read through the book.’
232
(37) a. pro-čitatj knig-u

through-read book-acc
b. pere-gretj rastvor
over-heat solution.acc
The book in (37a) is not only the resultee and the reference set, as with
the overheated solution in (37b), but is in the ScaleP in the complement
of the prefix. The prefix pro- then lexicalizes both the PrefP and the
PathP, as we saw it did with path when the path (road, busstop) appeared
without a preposition. The incremental theme, like the path direct ob-
ject, evacuates to adjoin to the PP, allowing pro- to lexicalize the entire
PP (Figure 8.20).
PP
ScaleP PP ⇒pro-
book
PrefP PathP
Route Route
Source Source
Goal PlaceP Goal PlaceP
Place tScaleP
Place GROUND
Figure 8.20 Evacuation of ScaleP and spelling out pro-
As with path, pro-can introduce unselected objects (38).
(38) a. pro-guljatj lekciju
vs. *guljatj lekciju
through-walk class
‘skip a class’ vs. *to walk a class
b. pro-spatj ekzamen vs. *spatj ekzamen
through-sleep exam
‘to oversleep an exam’ vs. *to sleep an exam
Such a configuration is possible only with pro-, because pere-, the other
argument-changing prefix, is semantically incompatible with an incremental
theme. It means ‘to exceed,’ but an event of, for example, reading a book,
cannot exceed the length of the book.

8.4.4 Superlexical pref ixes
Superlexical prefixes spell out the PrefP subconsitutent when its complement
is not the PP, but a Scale phrase, which introduces a temporal scale based on
the temporal trace of the event.
Ramchand (2008b) suggests that Asp is a function that yields a t variable in
the temporal trace of e, and t may be definite (if perfective) or indefinite (im-
perfective). If t is definite, that is, the event is perfective, a superlexical prefix
may map the event to a definite point on the temporal scale: A goal prefix
maps t to the transition into the event’s temporal trace (inception), a source
prefix maps t to the transition out of the event’s temporal trace (completion),
and a route prefix maps it to the length of the temporal trace (duration).
Because of the Superset Principle, this structure may be lexicalized by the
same prefix as path, as the superlexical prefix is a subconstituent of the P
lexical entry. For example, inception, that is, the transition into the activity,
may be lexicalized by a goal prefix, for example, za-prygatj ‘begin to jump.’ See
Figure 8.21.5
PP
PrefP ScaleP
Goal Place
Scale AspP
Asp InitP
Init ProcP
Proc
Figure 8.21 Goal/inceptive superlexical prefix
Because the abstract GROUND node is not a part of the superlexical prefix,
we get a temporal meaning. Nondoubling prefixes cannot be inserted as
superlexical prefixes for the same reason they do not spell out prepositions: They
do not contain a subconstituent that would exclude GROUND.
Source prefixes such as ot- and s-contain source and goal projections and
combine with a temporal scale, similar in shape to a source path. Recall the
visualization of a source path (39):
(39) +++–––
0.........1
5. This whole PP is obligatorily the complement of a second aspect head, ensuring it

is perfective. The aspect head below the prefix is imperfective, as superlexical prefixes
do not combine with perfective verbs.
234
In the temporal domain, this means that at point 0 (the beginning point) the
figure is involved in the activity, and at point 1 (the final point) the activity
is over.
Figure 8.22 shows the structure of ot-plavatj ‘complete swimming.’
PP
PrefP ScaleP
Source Scale AspP

Goal Place
Asp InitP
Init ProcP
Proc
Figure 8.22 Source/completive superlexical prefix
A route or duration superlexical prefix is a valid subconstituent of the

prefixes pro- and pere-, as in the structure of pere-plavatj ‘to swim too much’
(Figure 8.23). The same structure will apply to pro-plavatj ‘to swim for a certain
duration.’
PP
PrefP ⇒ pere- AspP ⇒ plavatj
Asp InitP
Route
Source Init ProcP
Goal PlaceP
Proc
Place
Figure 8.23 Route/duration superlexical prefix
Thus we see that the syntactic structure of the lexical entry underlying the
three types of instantiations of prefixes is the same. The same lexical entry can
spell out a path prefix, a scale prefix, and a superlexical temporal prefix.
8.5 CONCLUSION: THE BENEFITS OF A NANO-S YNTACTIC

APPROACH
We have demonstrated the possibility of explaining the chaotic behavior of

Russian verbal prefixes with the help of nanosyntax. The link was demonstrated
between the inner structure of a prefix and the PP complement of the prefixed

verb. Such a connection is expected if both the prefix and the PP complement
spell out parts of the PP projected by the prefix.
The path–scale–time parallel is established and supported not only by se-
mantic similarities, but also by the parallel syntactic behavior of the prefixes
at each level.
The Superset Principle allows the same lexical entry to spell out different
syntactic configurations, thus predicting both the parallels between the prefix
behavior at each level and the differences in selectional restrictions of prefixes.
REFERENCES
Babko-Malaya, Olga. 1999. Zero Morphology: A Study of Aspect, Argument Structure and
Case. Doctoral dissertation, Rutgers University.
Borer, Hagit. 2005. Structuring Sense, Vol. I: In Name Only. Oxford: Oxford University
Press.
Caha, Pavel. 2011. The Parameters of Case Marking and Spell-Out Driven Movement. In
Linguistic Variation Yearbook 10 2010, edited by Jeroen van Craenenbroeck, pp.
33–77. Amsterdam and Philadelphia: John Benjamins.
Di Sciullo, Anna Maria and Roumyana Slabakova. 2005. “Quantification and Aspect.”
In Perspectives on Aspect, edited by Angeliek van Hout, Henriette de Swart, and
Henk Verkuyl, pp. 61–80. Dordrecht, The Netherlands: Kluwer.
Esenin, Sergey. 1924. Russian Poetry Classic (1895–1925) Collection of Poems. Bilingual
Version (Russian–English). Translated from the Russian by Alec Vagapov. [online]
Available at http://samlib.ru/w/wagapow_a/yesen.shtml
Isačenko, Alexander. 1960. Grammatičeskij Stroj Russkogo Jazyka. Morfologija. Častj
Vtoraja. Bratislava: Vydavatelstvo Slovenskej Akadémie vied.
Kagan, Olga. 2013. “Scalarity in the Domain of Verbal Prefixes.” Natural Language and
Linguistic Theory 31(2): pp. 1–34.
Kennedy, Chris and Lisa McNally. 2005. “Scale Structure and the Semantic Typology of
Gradable Predicates.” Language 81(2): pp. 345–381.
Levin, Beth and Malka Rappaport-Hovav. 2006. “Constraints on the Complexity of
Verb Meaning and VP Structure.” In Between 40 and 60 Puzzles for Krifka, edited
by Hans-Martin Gärtner, Regine Eckardt, Renate Musan, and Barbara Stiebels.
Berlin: ZAS. [online] Available at: http://www.zas.gwz-berlin.de/fileadmin/ma-
terial/40-60-puzzles-for-krifka/.
Levin, Beth and Malka Rappaport Hovav. 2010. “Lexicalized Scales and Verbs of Scalar
Change.” Paper presented at the 46th Annual Meeting of the Chicago Linguistic
Society, University of Chicago, Chicago, IL, April 8–10, 2010.
Markovskaya, Evguenia. 2006. “Goal-Source Asymmetry and Ryussian Spatial Prefixes.”
In Nordlyd: Special Issue on Adpositions, Tromsø Working Papers in Linguistics 33
(2), edited by Peter Svenonius, pp. 200–219. Tromsø, Norway: University of
Tromsø.
Newell, Heather. 2008. Aspects of the Morphology and Phonology of Phases. Doctoral dis-
sertation, McGill University.
236
Pantcheva, Marina. 2012. “Deriving Doubling, Non-Doubling and Argument-Changing

Prefixes.” Talk at Prefix Verbs: The Impact of Preposition-like Elements on the Syntax
and Semantics of Verbs Stuttgart, July 12–13, 2012.
Pereltsvaig, Asya. 2006. “Small Nominals.” Natural Language & Linguistic Theory 24: pp.
433–500.
Ramchand, Gillian. 2008a. Verb Meaning and the Lexicon. A First Phase Syntax.
Ramchand, G. 2008b. “Perfectivity as Aspectual Definiteness: Time and the Event in
Russian.” Lingua 118 (11): pp. 1690–1715.
Ramchand, Gillian and Peter Svenonius. 2002. “The Lexical Syntax and Lexical
Semantics of the Verb-Particle Construction.” In Proceedings of WCCFL 21,
edited by Line Mikkelsen and Christopher Potts, pp. 387–400. Somerville,
MA: Cascadilla Press.
Rappaport-Hovav, Malka. 2008. “Lexicalized Meaning and the Internal Structure of
Events.” In Theoretical and Crosslinguistic Approaches to the Semantics of Aspect,
edited by Susan Rothstein, pp. 13– 42. Amsterdam and Philadelphia: John
Benjamins.
Romanova, Eugenia. 2004. “Superlexical vs. Lexical Prefixes.” In Nordlyd 32 (2),
Special Issue on Slavic Preﬁxes, editied by Peter Svenonius, pp. 255–278. Tromsø,
Norway: University of Tromsø.
Romanova, Eugenia. 2007. Constructing Perfectivity in Russian. Doctoral dissertation,
Schoorlemmer, Maaike. 1995. Participial Passive and Aspect in Russian. Doctoral disser-
tation, University of Utrecht.
Starke, Michal. 2005–2011. Nanosyntax. Class lectures, Center for Advanced Study of
Teaching and Learning, University of Tromsø.
Starke, Michal. 2011. “Towards Elegant Parameters: Language Variation Reduces to
the Size of Lexically Stored Trees.” Transcipt from a talk at Barcelona Workshop
on Linguistic Variation in the Minimalist Framework. [online] Available at
LingBuzz/001183
Svenonius, Peter. 2004a. “Slavic Prefixes and Morphology: An Introduction to the
Nordlyd Volume” In Nordlyd 32.2, edited by Peter Svenonius, pp. 177–204.
Tromsø, Norway: University of Tromsø.
Svenonius, Peter. 2004b. “Slavic Prefixes Inside and Outside VP.” Nordlyd 32(2): pp.
205– 253. [online] Available at http://septentrio.uit.no/index.php/nordlyd/
index
Svenonius, Peter. 2008. “Russian Prefixes are Phrasal.” In Formal Description of Slavic
Languages, edited by Gerhild Zybatow, Luka Szucsich, Uwe Junghanns, and
Roland Meyer, pp. 526–537. Frankfurt am Main: Peter Lang.
Tolskaya, Inna. 2014a. Verbal Prefixes: Selection and Interpretation. Doctoral disserta-
Tolskaya, Inna. 2014b. “Verbal Prefixes in Russian With the Verbs of Motion: Conceptual
Structure Versus Syntax.” Journal of Linguistics 51(1), pp. 1–31.
Vinka, Mikael. 1999. “Predicative and Non-Predicative Verb Particle Constructions.” In
WCCFL 18 Proceedings, edited by Sonya Bird, Andrew Carnie, Jason D. Haugen,
and Peter Norquest, pp. 570–585. Somerville, MA: Cascadilla Press.
Zaucer, Rok. 2009. A VP-Internal/Resultative Analysis of 4 “VP-External” Uses of Slavic
Verbal Prefixes. Doctoral dissertation, University of Ottawa.
Zwarts, Joost. 2005. “Prepositional Aspect and the Algebra of Paths.” Linguistics and
Philosophy 28: pp. 739–779.

PART III
Theoretical Explorations
238
CHAPTER 9
Complex Left Branches, Spellout,

and Prefixes
MICHAL STARKE
M uch of what is known as nanosyntax follows from taking seriously a core

result suggested by the last few decades of research: Syntax operates
on individual features (and not on prefabricated bundles). As a result, the
representations typically investigated look like Figure 9.1.
K3
K2
K1
Z
Y
X W ...
Figure 9.1 Typical nanosyntactic representation
This derivation always involves a specific type of merge operation: Merge a

feature with the current derivational object. Let us call that merge-f.
To make things concrete, let us take this derivation to be the Caha (2009)
analysis of the genitive case on a noun. The genitive suffix corresponds to
K1 + K2 + K3, that is, a lexical entry with [K3 [K2 [K1]]]. When the (simplified)
derivation hits the stage in Figure 9.1, the tree cannot be spelled out:
K1 + K2 + K3 does not form a constituent and hence cannot be spelled out by
the lexical entry [K3 [K2 [K1]]], and nothing else can spell out K1, K2, and K3.
To rescue the situation, spellout-driven movement displaces ZP (see Starke
2011) producing (in several steps compressed into one here) Figure 9.2.1
1. To focus on the issue relevant here, I use such compressed or simplified derivations
until the last section, where issues around the detailed step-by-step derivations are
taken up.
240
Z K3
Y K2
X K1
W ...
Figure 9.2 Spellout-driven movement of ZP
The expressiveness of human language, however, builds on another type

of grouping: merging a previously built phrase with the current derivational
object, merge-XP. Merge-f on its own produces only linear sequences, the
equivalent of strings. Merge-XP is where relevant recursion is introduced—
the result is not a simple linear sequence, and is hence not trivially equivalent
to a string. With merge-XP, a (base-generated) representation may look like
Figure 9.3.
K2 K1 Z
Y
X W ...
Figure 9.3 A representation with merge-XP
What do these structures correspond to in a world in which terminals are

single features? And what is the spellout algorithm for such structures?
9.1 PRE
One surprising consequence of derivations such as Figure 9.2 is that all

terminals (features) except the lowest ones will always be spelled out as a
suffix: Moving their complement across them is the only way for them to
be a constituent and hence to be spelled out. What about functional heads
that are standardly taken to spell out as pre–V/N/Adj function words (some
complementizers, articles, negation, etc.)? According to the logic in the
opening section, they should end up as suffixes, contrary to fact. The obvious
conclusion is that their syntactic structure is different, and a natural candi-
date is Figure 9.3.
Prefixes are another class of elements that look surprising from the point of
view of derivations such as Figure 9.2. Everything that was said about suffixes
previously also applies to them, so they should also end up as suffixes. Again,
their underlying syntactic structure must be different from that of suffixes,
and Figure 9.3 is a natural candidate.
[ 240 ] Theoretical Explorations

9.1.1 The pref ix–s uff ix asymmetry
Before diving into Figure 9.3, let us pause on a point about prefixes and
suffixes that should be much more obvious than it currently is (see also dis-
cussion in Taraldsen Medová Chapter 3): The notion of prefix (or suffix) is a
distributional notion. Being a prefix (or a suffix) is very much like being pre-
verbal (or postverbal), prenominal (or postnominal). We want to derive ‘being
preverbal/postverbal,’ and similarly we want to derive ‘being prefixal/suffixal.’
Hardcoding that an affix is a prefix or a suffix, and then using that hardcoded
information to position it, is about as insightful as having an idiosyncratic nota-
tional device to mark whether subjects (and objects) are preverbal or postverbal,
and then using these notational devices to drive the syntactic derivation. Every
current approach, however, stipulates the prefix–suffix distinction: Whether a
given affix is prefixal or suffixal is hardcoded through a dedicated notational
device, and that notational device is then used to place the affix in the correct
position with respect to its stem. Which misses the point.
How do we handle prefixes then? To make things a bit more concrete, let
us take a language in which some tense + aspect markers are prefixal, whereas
others are suffixal. Given that the tense + aspect markers carry (at least) a
tense and an aspect feature, their lexical entry is at least [T [Asp]]. A com-
pressed derivation of the suffixal version would be along the lines of Figure 9.4.
T
Asp VP
Figure 9.4 Deriving tense + aspect suffix
This cannot be spelled out, triggering spellout-driven movement:
VP
T
Asp
Figure 9.5 Spellout-driven movement of VP
T + AspP is now a constituent (Figure 9.5) and can be spelled out by the
suffix. Notice the exact shape of the lexical entry: The verbal root has evacuated
from the [Asp VP] constituent, leaving only [Asp], a unary grouping. The en-
tire lexical entry would thus be [T [Asp]].
How does the prefix escape the same fate? A natural hypothesis is that it
instantiates a different structure.
VP
T Asp
Figure 9.6 Deriving tense + aspect prefix
C o m p l e x L e f t B r a n c h e s , S p e l l o u t, a n d P r e f i x e s [ 241 ]
242
Given the structure in Figure 9.6, there is no need for spellout-driven

movement: T + Asp is already a constituent, without dislocating the verbal
stem. The [T Asp] constituent thus remains in front of the verb, correctly
yielding a prefix (assuming no other operation ever moves anything between
V and [T Asp] in this language). Note that the structures for the prefix and
the suffix are now different. The prefix is as shown in Figure 9.7, whereas the
suffix is as shown in Figure 9.8.
T Asp
Figure 9.7 Prefix structure
T
Asp
Figure 9.8 Suffix structure
There is a singleton grouping in the lowermost position of suffixes (see

Figure 9.8) but not of prefixes (see Figure 9.7).
Note that both merge-f and merge-XP are binary operations: they attach a
new feature or a new phrase to a previously built object (at the first derivation
step, this means attaching a feature or a phrase to a single feature). The only
way to create a unary grouping is to evacuate something from what used to be
a binary grouping.2
As a result, the suffixal entry (Figure 9.8) can only spell out structures in
which Asp used to have a complement, but that complement has been evacuated
by spellout-driven movement. The prefixal entry (Figure 9.7), on the other hand,
has no missing complements in its foot (no unary grouping), and hence cannot
have originated on the main spine (where Asp would have had V as its comple-
ment); it can spell out only “specifier-like” structures, complex left branches.
We have now derived prefixal versus suffixal positioning of lexical items
without introducing any dedicated notational device: A lexical entry with a
singleton set (unary grouping) in its lowermost position will spell out a struc-
ture in which the lowermost complement has moved out, making that lexical
entry suffixal, whereas a lexical entry with a binary grouping in its lowermost
position will spell out a corresponding base-generated structure with no com-
plement moved out, that is, a prefix (or a stem—in both cases the lowermost
part of a functional sequence). The notion of unary–binary grouping is inde-
pendent of the prefix–suffix notion: It is both more generic and developed
independently of the prefix–suffix asymmetry. The prefix–suffix asymmetry is
thus derived without any technology dedicated to it.
2. Spellout-driven movement is assumed not to leave a trace for very much the same
reason that head movement is sometimes assumed to not leave a trace: It has no
apparent semantic import, no reconstruction effect, the would-be traces do not trigger
discernible intervention effects, etc.

Before going into more details of the relevant derivations, let us look into
function words occurring in pre–V/N/Adj positions.
9.1.2 Functional “heads”
Function words typically present very much the same problem. Take an inter-
rogative complementizer in a language like English or French. It must have at
least a feature making it interrogative, and one making it a complementizer
(as opposed to e.g. an interrogative pronoun). Given one feature per syntactic
head, this means that an interrogative complementizer spells out at least two
functional layers in syntax.
A (compressed) derivation whereby the complementizer spells out two
heads in the main derivational spine looks like Figure 9.9.
C
Wh TP
Figure 9.9 Deriving a complementizer made up of at least two heads in the main spine
It is impossible to spell this out, as C + Wh do not form a constituent, so

spellout-driven movement will dislocate TP over C/W h.
TP
C
Wh
Figure 9.10 Spellout-driven movement of TP
Figure 9.10 nicely derives complementizer- final languages. But what

about English-style complementizers? Again, it must be that their struc-
ture is different, and a natural candidate is a “specifier”-like structure like
Figure 9.11.
TP
C Wh
Figure 9.11 ‘Specifier’-like complementizer
If so, complementizers of C-final languages have a unary branch as their

foot, [C [Wh]], and spellout on the main spine, whereas complementizers
of C-initial languages have a binary grouping as their foot, [C Wh], and spell
out an embedded spine (a specifier-like structure). Similarly for prenominal
determiners, pre–N/V/Adj negation, and so forth, to the extent that those are
generated higher up in the functional sequence (fseq).
244
9.1.3 Verbs and particles
Let us briefly note a curious possibility: It is sometimes (plausibly) argued that

the particles of particle verbs originate lower than the verb on the main der-
ivational spine; verbs are therefore not the lowermost element of their fseq.
If verbs spell out features on the main fseq—as is usually taken for granted—
they would end up following the particle.
To make this concrete, let us use the traditional cause- become- state
features to describe verbal roots. With these, a (compressed) derivation looks
like Figure 9.12.
cause
become
state
PRT1 PRT2
Figure 9.12 Verbal cause-become-state features generated above particle (PRT) features
This cannot be spelled out with a verb expressing cause + become + state as
there is no constituent covering exactly those features, and therefore spellout-
driven movement ultimately creates the structure in Figure 9.13.
PRT1 PRT2
cause
become
state
Figure 9.13 Structure resulting from spellout-driven movement
The verb can now spell out the constituent cause + become + state given
a lexical entry [cause [become [state]]], and the resulting order is PRT-V. But
what about English-style particle verbs, in which V precedes PRT? Again, it
must be that verbs in such languages realize a specifier-like structure, that is,
complex heads, as in Figure 9.14.
PRT1 PRT2
cause
become state
Figure 9.14 ‘Specifier’-like verb

We are thus left with the interesting consequence that verbs in English-
style languages do not spell out the main spine, the main verbal fseq, but a
complex left branch, a specifier-like structure hanging below the “verbal fseq”
(if the analysis of particles as originating below verbs is correct).
9.2 THE SPELLOUT ALGORITHM
The driving force of a derivation is to extend the currently built fseq with a new
label or type higher in fseq. Fseq is a given (either as a single rigid hierarchy, or
perhaps as a set of possible hierarchies derived by principles), and each deriva-
tion builds instances of such a sequence. If a derivation has reached functional
projection N, the next derivational step is to build layer N + 1 of the functional
sequence (on approaches in which no layer can be skipped), or more plausibly
N + x (on approaches in which layers can be skipped). At each step we are thus
looking for a provider of a given label.
Previous work on the exact derivational steps driven by spellout focused on
the case in which the label is provided by merge-f, directly inserting a feature.
The resulting algorithm was
(1) a. Insert feature and spell out.

b. If fail, try a cyclic (spec-to-spec) movement of the node inserted at
the previous cycle and spell out.
c. If fail, try a snowball movement of the complement of the newly
inserted feature and spell out.
If none of the steps in (1) yields a successful spellout, it must be that a prior

choice turned out to be wrong, and hence needs to be undone:
(2) If the spellout algorithm in (1) fails, go back to the previous cycle, and
try the next option for that cycle.
All of this, however, deals with merge-f. How does merge-XP enter the picture?
There is a deep asymmetry between merge-f and merge-XP, given a
bottom-up derivational mechanism: Merge-XP requires spawning a new
derivation, merge-f does not. This is because merge-XP trivially requires
building an XP to be merged into the current derivation. And building such
an XP is itself a derivation, a separate derivation from the current, main der-
ivation. Merge-XP is thus a heavy operation, whereas merge-f is an atomic,
light operation.
This suggests that merge-XP is the ultimate last-resort operation, attempted
only when the backtracking step in (2) fails.
246
A straightforward reductio ad absurdum argument leads to the same conclu-

sion: If merge-XP was preferred to merge-f, at each step of the derivation the
first operation would be merge-XP, that is, “please try to build an XP providing
(projecting) feature X.” But the attempt to build an XP providing feature X
would itself be a derivation, and hence would itself start with “please try to
build an XP providing (projecting) feature X,” ad infinitum.
If a provision was made to heal this simple version of infinite recursion
(maybe by making a special clause for the first merger), we would still hit a re-
ductio ad absurdum. Assume again that merge-XP is preferred to merge-f, and
consider a situation in which merge-f would have been successful. The attempt
to continue the derivation with feature X would again be “please try to build
an XP providing (projecting) feature X.” But because merge-f would have been
successful, we know that it is indeed possible to build such an XP, namely by
rebuilding whatever has been built up to now, and inserting the relevant fea-
ture on top of that—by assumption. It thus follows that whenever merge-f
would have been successful, it is never going to be tried (either at the top level,
or ever, depending on the how the simpler version of the reductio ad absurdum
is resolved). Hence ordering merge-XP before merge-f yields the wrong result.
Let us therefore extend the spellout algorithm with
(3) If merge-f has failed to spell out (even after backtracking), try to spawn
a new derivation providing feature X and merge that with the current
derivation, projecting feature X to the top node.
I leave aside here the important question of the ordering of remerge (internal
merge, move) in the spellout algorithm, hoping to come back to it separately.
Notice that there is a subtle difference between the last-resort steps in (1b)
and (1c) and the last resort of merge-XP in (3): The former apply recursively
as we backtrack, the latter applies only to the current step of the derivation,
after all backtracking has finished (and failed). This is because we do not want
to override a previously successful spellout of merge-f (a previous morpheme)
with a new derivation (a phrase).
9.3 THE “COORDINATED GROWTH” PROBLEM
Finally, let us take a quick look at step-by-step derivations for specifier-like

derivations, in order to highlight a challenging issue. To make things concrete
again, let us look at derivations for the genitive, as in Figure 9.15, but now
in languages in which it is spelled out by a PRE element, such as the preposi-
tion of in English, in the Caha (2009) analysis. To highlight the relevant issue,
let us work with the target structure in Figure 9.15 and the lexical entries in
Figures 9.16–9.18.

K3 Z
K2 Y
K1 Z X ...
Figure 9.15 Target structure in a ‘specifier’-like derivation
1. /sugar/
2.
Z
Y ...
X
3. SUGAR
Figure 9.16 Lexical entry for sugar
1. /of/
2.
K3
K2
K1 Z
Figure 9.17 Lexical entry for of
K2
K1 Z
Figure 9.18 Lexical entry for null nominative/accusative
The first lexical entry (Figure 9.16) is for the word sugar, with a stand-in for
its phonological representation as /sugar/in (a), its syntactic representation
in (b), and its conceptual representation summarized by ‘SUGAR’ in (c). The
entry in Figure 9.17 is the English prepositional genitive of, and the entry in
Figure 9.18 is the null accusative or nominative.
What happens after ZP is built? K1 needs to be provided, but inserting the
feature K1 is not going to be successful because there is no lexical element [K1]
or containing [K1]. (Inserting the feature K1 would be successful in Czech,
Polish, Russian, etc., which do have lexical elements with [K1], surfacing as
nominative suffixes.) The desperation strategy kicks in, spawning a separate
workspace in the hope that a phrase with K1 can be built. Indeed it can be,
the minimal one being [K1 Z], and the derivation is successfully saved, as in
Figure 9.19.
K1 Z Z
Y X ...
Figure 9.19 Building [K1 Z] saves the derivation
248
The next step is to introduce the feature K2. And there lies the problem.
As before, merge-f will not help. The desired outcome is that the lexical
entry in Figure 9.18 is used, [K2 [K1 Z]], providing both K1 and K2, as in
Figure 19.20.
K2 Z
K1 Z Y X ...
Figure 9.20 Desired outcome, with [K2 [K1 Z]]
But (i) the logic previously given yields a different result, namely the
spawning of another new derivation and the corresponding insertion
of Figure 9.18 above the already existing [K1 Z], as in Figure 9.21, and
(ii) we have no technology for the actually desired result, namely “growing”
the existing [K1 Z] into [K2 [K1 Z]] (or perhaps directly inserting
[K2 [K1 Z]] at the previous derivational step). In other words, the con-
stituent built by the spawned derivation must “grow” in tandem with the
mainline derivation.
K2
K1 Z KI Z Z
Y ...
X
Figure 9.21 Unwanted result, with both [K2 [K1 Z]] and [K1 Z]
Furthermore, the usual elsewhere or overriding spellout mechanism must

apply to this “grown” subderivation: At the next step, K3 is needed. The desired
outcome is our target structure in Figure 9.15. The complex left branch will
now be spelled out by Figure 9.17, that is, [K3 [K2 [K1 Z]]], overriding Figure
9.18 as per the usual spellout mechanism.
To address this coordinated growth issue, I offer the following tentative
suggestion: Spawning a separate subderivation is so costly that it is kept around
as long as possible rather than immediately closed and its result merged in the
mainline. On the next derivational step, that same subderivation is asked to
provide the subsequent feature. At the K2 stage, we thus have a subderivation
[K1 Z] and ask whether that subderivation can provide K2. The normal merge-
f mechanism applied to the root of that subderivation will indeed produce
[K2 [K1 Z]]. Finally, when K3 is requested, merge-f on the subderivation will
produce the desired [K3 [K2 [K1 Z]]], which will correctly be lexicalized by
Figure 9.17.

It is only at the subsequent step that the subderivation will not be able to
provide whatever next feature is requested (because Figure 9.17 has no higher
feature and no other lexical entry with K1/K2/K3 does). At that point, all pos-
sible juice has been squeezed from the costly subderivation; the subderivation
is thus terminated and its result is incorporated into the main line.
If so, we have a representationally simple account for the prefix–suffix
asymmetry, as well as an account for function words that stay pre-V/N/
Adj: They are complex left branches. In contrast, suffixes and post-V/N/Adj
function words directly spell out the main derivation line (and have a unary
grouping as their foot).
REFERENCES
Starke, Michal. 2011. “Towards elegant parameters: language variation reduces to
the size of lexically stored trees.” Ms. Tromsø University. [online] Available at
lingbuzz/001183.
250
CHAPTER 10
Word Order and Nanosyntax

Preverbal Subjects and Interrogatives
Across Spanish Varieties*
ANTONIO FÁBREGA S
10.1 OUTLINE OF THE CHAPTER AND MAIN CLAIMS
The goal of this chapter is to show how the tenets and principles of nanosyntax
(see Baunaz and Lander in Chapter 1 for references) can be used to account
for patterns that have traditionally fallen into the realm of “phrasal syntax”.
The phenomenon that we discuss here is the position of overt subjects in
wh-interrogatives across Spanish varieties. European Spanish displays a
Verb–Subject ordering in a sentence like (1a); however, in the Spanish of the
Venezuelan Sierra (Mérida) some subjects can be preverbal (1b), whereas in the
Spanish of the Dominican Republic all kinds of subjects can be preverbal (1c).1
(1) a. ¿Qué * {usted / uno} dice? European Spanish

what you.pol / one say.3sg
* The research that underlies this chapter is financed by projects FFI2013-41509-P and
FFI2014-56968-C4-2-P, from the Spanish Ministry of Economy and Competitiveness.
I am grateful to Peter Svenonius, Tarald Taraldsen, Julio Villa-García, José Luis
Mendívil, Kristine Bentzen, Martin Krämer, and two anonymous reviewers for
comments to earlier drafts of this chapter. All disclaimers apply.
1. Even though in this chapter we concentrate on pronominal forms, the facts are the
same with DPs containing a lexical N: Being all third person, none is allowed in a pre-
verbal position in European Spanish; all are allowed in Dominican Republic Spanish,
and only contrastive ones are allowed in Mérida Spanish.
b ¿Qué {usted /* uno} dice? Mérida (Venezuela) Spanish
what you / one say.3sg
c. ¿Qué {usted / uno} dice? Dominican Republic Spanish
what you / one say.3sg
‘What do you /does one say?’
My proposal is that in European Spanish, Int(errogative)P is spelled out

through phrasal spellout by the exponent that also materializes subject agree-
ment (Figure 10.1) at SubjP. This prevents any overt additional constituent not
included in the lexical entry of the exponent to be merged in the area between
IntP and SubjectP, blocking a potential position for the subject only in inter-
rogative sentences. Dominican Republic Spanish, in contrast, spells out IntP
with its wh-elements, and its (impoverished) subject agreement lexicalizes
only SubjP (Figure 10.2); this allows overt specifiers between IntP and SubjP.
Mérida Spanish, in contrast, spells out with agreement a constituent that goes
up to Fin(iteness)P, but does not include IntP, which is also spelled out by wh-
elements in this variety (Figure 10.3). The entries in Figures 10.1 and 10.2 and
10.3 are simplified for the time being.
IntP ⇒ /-o/
Int PartP
Part FinP
Fin SubjP
Subj (TP)
1sg
Figure 10.1 European Spanish
SubjP
Subj (TP)
1sg
/-o/
Figure 10.2 Dominican Spanish
FinP ⇒ /o/
Fin SubjP
Subj (TP)
Figure 10.3 Mérida (Venezuela) Spanish
W or d Or de r a n d N a n o s y n ta x [ 251 ]
252
Thus the variation attested in Spanish, I claim, can be reduced to the size of
the exponents contained in the lexical repertoire of each variety.
It has been noted (e.g. Henríquez Ureña 1939; Lipski 2002) that the ab-
sence of inversion in interrogatives partially correlates with how frequent, and
in which contexts, overt pronominal subjects are in the variety in question.
Although here I do not examine pro-drop as a global phenomenon, I believe
that my proposal throws light on this particular aspect of the correlation: The
bigger the material spelled out by subject agreement in one variety, the fewer
positions would be available to host overt pronominal subjects in the same
varieties even in declarative sentences, pushing (some) overt pronominal
subjects up the tree to purely constrastive positions.
Before I start, I need to make explicit some assumptions about the basic
structure of the clause that are relevant for my proposal. I follow Pollock (1989),
Chomsky (1991) and the other early decompositions of InflP in the claim that the
temporal part of InflP has to be differentiated from the head responsible for sub-
ject agreement. The proposal that there is an AgrS position became unpopular
during the mid-1990s with the formulation of the Minimalist Program (Chomsky
1995). The problem, essentially, was that a specialized agreement position should
not be possible in a system in which uninterpretable features are erased before
transfer to LF. Here, we are going to assume that at least two positions have to be
distinguished, T and SubjP, the second hosting agreement (Figure 10.4).
SubjP
Subj TP
T ...
Figure 10.4 Two positions: TP and SubjP
This decision is in practice internally forced in nanosyntax (and any mono-

tonic approach to morphology), because of the existence of distinct exponents
for these two layers; consider (2), from Spanish.
(2) a. cantá-ba-s

sing-past.impf-2sg
b. canta-re-mos
sing-fut-1pl
As nanosyntax cannot use postsyntactic morphological operations like

Fission (Halle and Marantz 1993), splitting one single syntactic head in two,
data like (2) essentially force the conclusion that (at least) two heads have
to be distinguished in the structure to accommodate the two exponents.
But note, additionally, that SubjP does not need to be interpreted as a head

where agreement is the only thing that takes place; see Svenonius (2008) for
a proposal.
Second, here we also divide a wh- interrogative into two layers, one
Int(errogative)P, which is common to all kinds of interrogatives, and a second
layer where wh-elements are placed. IntP can be viewed as one particular in-
stantiation of Force whose content encodes an instruction to the addressee
to provide an answer. In this sense, IntP is common to total and partial (or
wh-) questions (Figures 10.5 and 10.6). The WhP layer, in contrast, is exclusive
to wh-questions (see Figure 10.6) and is not present in total questions (see
Figure 10.5).
(3) ¿Dices tú eso? (total interrogative)

say.2sg you that
‘Are you saying that?’
IntP
Int ...SubjP
Subj ...vP
Figure 10.5 Total interrogative with only IntP
(4) ¿Qué dices tú? (partial / wh-interrogative)

what say.2sg you
‘What are you saying?’
WhP
Wh IntP
Int ...SubjP
Subj ...vP
Figure 10.6 Partial interrogative with both WhP and IntP
With this background in mind, let us now move to the empirical data that
are the core of this chapter. In Section 10.2 I present the facts. In Section 10.3
I present arguments that the Verb–Subject order in Spanish is produced when
the subject stays low, not when the verb moves high. Section 10.4 argues for
a specific ordering of the heads involved in the area, introduces the relevant
exponents and their differences across the three varieties, and derives from
that why the subject has to stay low under certain conditions. Section 10.5
presents some consequences and conclusions.
254
10.2 THE POSITION OF SUBJECTS IN SPANISH
The central fact discussed in this chapter is the different availability of pre-
verbal subjects in wh-interrogatives across Spanish varieties. Here we see
that at least three varieties with different inversion patterns have to be
distinguished.
10.2.1 European Spanish
In European Spanish, wh-questions are associated with Verb–Subject ordering,

irrespective of the referential properties of the subject (5).
(5) a. ¿Qué dices (tú)?

what say.2sg you
‘What do you say?’
b. *¿Qué tú dices?
what you say.2sg
Subjects are not the only elements that cannot appear between the verb
and the wh-element. Similarly, focused elements or topics are excluded (6).
Adverbs of any type are also excluded (7).2
(6) a. *¿Quién A MARÍA ha visto?

who acc María has seen
‘Who has seen MARIA?’
b. *¿Quién las manzanas las ha traído?
who the apples them.acc has brought
‘The apples, who has brought them?’
2. Suñer (1994) argues that some adverbs are allowed between the wh-element and
the verb in interrogatives:
(i) ¿A quién siempre ves?

acc whom always see.2sg
‘Who do you always see?’
However, I do not share this judgment. To the extent that (i) is grammatical, it is con-
firmatory or exclamative: (i) cannot be used in an out-of-the-blue context, but crucially
implies that the addressee has already made the claim that there are people that he
always sees. Instead of the characteristic L% boundary tone of wh-questions, it must
carry the more complex nuclear configuration H + L*HH% (see Prieto and Roseano
2009). One could assume that this complex nuclear configuration is part of the pho-
nological information of an exponent that spells out IntP when other heads related to
echoic questions are present.

(7) a. *¿Qué ya has leído?
what already have.2sg read
b. ¿Qué has leído ya?
what have.2sg read already
‘What have you already read?’
10.2.2 Spanish in Mérida (Venezuela)
Under some conditions, our informants3 from Mérida (Mountain Region in

Venezuela) allow some preverbal subjects in wh-questions. However, the avail-
ability of these subjects is conditioned by (i) the referential properties of the
subject and (ii) contrastiveness.
Pronouns referring to speech participants can occupy the preverbal po-
sition without a contrastive interpretation. Note that usted, the polite ‘you,’
behaves as a participant pronoun even though it triggers third person agree-
ment (see (8)).
(8) a. ¿Qué tú diceh?4

what you say.2sg
b. ¿Qué uhted dice?
what you say.3sg?
‘What do you (polite) say?’
With third person pronouns, this variety allows the overt pronoun provided
that it carries a contrastive interpretation. Two tests show this: First, generic
pronouns (which are semantically maximal) cannot occupy the preverbal posi-
tion in interrogatives, as maximality is incompatible with contrastiveness (9);
second, (10) is ungrammatical in a context in which there is only one contex-
tually available referent for the subject (e.g. if we are talking about the Pope’s
latest interview and we are just discussing what he declared).
(9) *¿Qué uno sabe?

what one knows.3sg
‘What does one know?’
(10) #¿Qué él dijo después?

what he said.3sg then
‘What did he say next?’
3. For this chapter, I had access to five speakers born and raised in this region.
4. Phonologically, Mérida Spanish debuccalizes /s/at the coda; we represent it as an
aspiration.
256
10.2.3 Dominican Republic Spanish
It has been noted that Dominican Republic Spanish lacks Subject–Verb in-
version in interrogatives (Henríquez Ureña 1939, 1940; Kany 1951; Jiménez
Sabater 1975; Silva Corvalán 1982; Morales 1986). Toribio (2000) reports that
the speakers she interviewed allowed all kinds of subjects in preverbal po-
sition in interrogatives, crucially involving noncontrastive expressions (in-
cluding generics) (11); such constructions have also been documented with
generic pronouns in electronic newspapers (12).
(11) a. ¿Cuánto un médico gana?

how.much a doctor earns
‘How much does a doctor earn?’
b. ¿Qué número tú anotaste?
which number you wrote.down.2sg?
‘Which number did you write down?’ (apud Toribio 2000)
(12) ...le importa saber cuánto uno gana y cómo uno vive.
him cares know how.much one earns and how one lives
‘he cares to know how much one earns and how one lives’
[Foro Univisión, Dominican Republic]
10.3 HOW IS ORDER FORCED? AGAINST T-T O-C

MOVEMENT IN SPANISH
My proposal is that the varieties are differentiated by the size of the exponent
that materializes subject agreement, particularly whether it is also the expo-
nent that materializes IntP. Crucially, in European Spanish the Subj-Agr ex-
ponent is the one that spells out Int(errogative) force, thus making positions
between Int and Subj not present in its lexical entry unavailable. One crucial
aspect of our analysis, then, is that what seems to be wh-inversion in Spanish
is not due to T-to-C movement, as proposed for instance in Zagona (2002),
but the result of a configuration in which the “subject” remains low in the
structure (perhaps vP, although see Ordóñez 2007 for an alternative proposal
in which the postverbal subject moves to an XP position lower than T; this is
orthogonal to our analysis) [see (13)]. It follows from this account that there
are multiple subject positions in the clause, in line with previous claims by
Cardinaletti (2004) and Vangsnes (2002), among others.
(13) [WhP qué . . . [TP . . . dices . . . [XP . . . tú . . . ]]]

what say.2sg you.2sg

Crosslinguistically, it is not unusual that wh-sentences exhibit phenomena in
which one can see that there is an impoverishment of the material otherwise
available between T and C. We have seen (Section 10.2.1) that, in European
Spanish, topics and focused elements between T and C are also excluded in
wh-interrogatives. To give a genetically unrelated example, in the Omotic lan-
guage Dima (Mulugeta 2008), person agreement, which is necessary in declar-
ative sentences (14), is ungrammatical in interrogatives (15).
(14) wótú sól-is-im k’óm-déé-*(t)

1pl injera-def-acc cook-impf-1pl
‘We are cooking the injera.’
(15) wótú sól-is-im k’óm-déé-(*t)?

1pl injera-def-acc cook-impf-1pl
‘Shall we cook the injera?’
Descriptively, it seems that in Dima a temporo-aspectual exponent can either

combine with person agreement or with a head imposing interrogative force
on the clause: This suggests that the temporo-aspectual exponent lexicalizes
material that makes the subject agreement position unavailable. I propose
that what happens in European Spanish is in essence the same, involving only
a different exponent: The subject agreement exponent lexicalizes a constit-
uent that includes the position where the preverbal subject could be merged,
making its appearance ungrammatical.
Thus my claim is that what seems to be inversion is, in fact, the subject not
moving higher than TP. Let us now review several pieces of evidence against
T-to-C movement in Spanish and in favor of an analysis in which the subject
stays below TP (see for similar proposals and other arguments Suñer 1994;
Ordóñez and Treviño 1999; Buesa García 2008; Zubizarreta 2012; and Villa
García 2015).5
5. One argument given in the literature is the following (Suñer 1994; Ordóñez and
Treviño 1999): If the verb were in C in an interrogative, as the V-to-C analysis would
claim, one should expect that the verb would be part of the remaining material under
sluicing, on the standard assumption that sluicing involves erasing material below
C (Sag and Hankamer 1984; Lobeck 1995; Lasnik 2001; Merchant 2001; Aelbrecht
2010; alternatively, if sluicing does not involve building material below C, which is or-
thogonal to the argument, see Hardt 1993; Ginzburg and Sag 2000; or Culicover and
Jackendoff 2005). However, this is not the case:
(ii) Juan pudo comprar un libro, pero no sé cuál (*pudo).

Juan could buy a book but not know which could
However, other languages, like Norwegian, behave like this (Kristine Bentzen, per-
sonal communication), and they seem to involve V-to-C movement in some cases,
so the argument is weak. In a cartographic approach in which C is split into many
258
(i) Torrego (1984) noted that in subordinate interrogatives in European

Spanish, the Verb–Subject order is also enforced. This contrasts with English
and other Germanic languages, in which V2 phenomena, expectedly, are not
attested in subordinate clauses. Given that we know that T-to-C movement
tends to be blocked inside subordinate clauses, the European Spanish pattern
is typologically quite surprising.
(16) a. I don’t know what {he eats /*does he eat}. English
b. No sé qué (*él) come (él). European Spanish
not know.1sg what he eats he.
‘I don’t know what he eats’
Note, in contrast, that Spanish does exhibit the other Penthouse Effect phe-
nomena in certain classes of subordinate clauses (Ross 1973; Haegeman 2012;
with one exception noted in Haegeman 2006, and to which we return in
Section 10.4.1).
(17) No sé qué (#, francamente,) dice.

not know.1sg what frankly eats.3sg
Intended: ‘I don’t know what he frankly says.’
(ii) Torrego (1984) and Suñer (1994) note that certain wh-elements do not
trigger inversion (see Rizzi 2001 for Italian cases involving perchè ‘why’).
Although all these elements are interrogatives, only some of them would
trigger T-to-C movement in Spanish.
(18) a. ¿En qué medida la constitución permite un referéndum?

in what measure the constitution allows a referendum?
‘To what extent does the constitution allow a referendum?’
b. ¿Por qué el presidente se ha negado a responder?
for what the president SE has refused to answer?
‘Why has the president refused to answer?’
(iii) If the subject stays low in European Spanish interrogatives, we expect

some adverbs that cannot precede the subject in a declarative to be able to do
so in an interrogative. This is precisely what we see in (19), with siempre ‘al-
ways’; the contrast between (19a) and (19b) makes it impossible to analyze the
pattern as V moving up to C, because that would not account for the different
ordering of the subject with respect to the adverb.
(19) a. ??/* Siempre Juan dice esto.

always Juan says this

b. ¿Qué dice siempre Juan?
what says always Juan?
It is now the moment to introduce my analysis. I propose that in European

Spanish IntP is spelled out by the same exponent that spells out subject agree-
ment, as represented in Figures 10.7 and 10.8, marking that material in bold. By
the Superset Principle, the exponent (-n in Figures 10.7 and 10.8, for 3pl agree-
ment) cannot be used if the constituent IntP contains features not lexicalized
by this exponent, such as a strong pronoun or DP in Spec-SubjP (Figure 10.8).
Note that here the verb stem is represented between parentheses; we discuss
its position in Section 10.4.4.6
Qué dicen?
?
WhP
what say.3pl
Wh IntP ⇒ (dice-)/-n/
qué
‘what’ Int ...SubjP
Subj TP
3pl
T ...vP
ellos ...
‘they’
Figure 10.7 3pl agreement -n corresponds to IntP and SubjP
WhP * Qué ellos dicen?

?
what they say.3pl
Wh IntP ⇒ (dice-)*/-n/
qué
‘what’ Int SubjP
*ellosi Subj
‘they’
Subj TP
3pl
T ...vP
ti ...
Figure 10.8 3pl agreement -n not possible with DP subject in Spec-SubjP
heads, there is no guarantee that an alleged V-to-C movement targets the same head as
sluicing. In fact, see Leu (2015) for arguments that German dass ‘that’ should be split
into several heads, in which the highest one does not intervene in verb fronting (I am
grateful to an anonymous reviewer for making me note this work).
6. Here we assume, with Caha (2009), that the constituent spelled out by an expo-
nent is defined ignoring the material that has already been spelled out in previous
operations. In other words: On the assumption that spellout works bottom up, the ma-
terial below SubjP has already been spelled out by distinct exponents and is therefore
ignored to define the constituent spelled out together with SubjP.
260
In a noninterrogative context, in which IntP is not spelled out, the Anchor

Condition (Abels and Muriungi 2008) and the Superset Principle guarantee
that preverbal subjects will be possible (thus, no inversion) (Figure 10.9).
XP
ellos X
‘they’
X ...SubjP
(dice)-n
Figure 10.9 Noninterrogative with preverbal subject
In the case of wh-elements like por qué ‘why’ or en qué medida ‘to what extent’
[see (18)], my suggestion is that they should be treated like idioms that are
the only wh-words whose lexical entry materializes in European Spanish both
WhP and IntP (see Rizzi 2001 for a proposal in the same spirit, but relating
base-merging with absence of movement). Once IntP is not lexicalized by the
subject agreement exponent, the exponent can shrink and leave space for a
preverbal subject with a contrastive reading (Figure 10.10).7
WhP ⇒ por-qué
Wh IntP
Int XP
ellos X
‘they’
X ...SubjP ⇒ (dice)/-n/
3pl
Figure 10.10 Preverbal subject with por qué
7. I admit, however, that this suggestion about por-qué has one potential shortcoming,
which has sometimes been noted in nanosyntactic accounts that treat the features
spelled out by a lexical item as a matter of accident and potentially subject to variation
from language to language. In the same way that por qué in Spanish does not trigger
inversion, its Italian counterpart and its English equivalent how come do not trigger it
either (Bakovic 1998). In my approach, this is an accident: Each lexical item in each va-
riety in principle can lexicalize slightly different features. However, the three languages
seem to suggest that there is a deeper connection between being a cause wh-word and
the absence of inversion. Nothing in my account is incompatible with it: It might be
that, by accident, the three languages happen to spell out Int as part of the entry of the
cause wh-word, but if a wider typological study shows that cause wh-words tend to have
this behavior, treating the lack of inversion with them as a lexical accident has the flavor
of a missing generalization. One option in that case would be to associate the cause se-
mantics as an emergent interpretation at the interfaces that is triggered when an Int
head is missing under certain conditions, but I leave the matter open, pending a wider
typological study. Note, however that en qué medida does not have a cause semantics and
does not trigger inversion either, which still favors a lexical account of the Spanish facts.

10.3.1 Initial plausibility of the proposal
One straightforward prediction of this proposal, before introducing the

exponents, is that postverbal subjects in interrogatives should lack some
readings or not be subject to some constraints that preverbal subjects have: In
my analysis, some subject positions become unavailable in interrogatives. The
competing analysis, in which postverbal subjects are still high and the verb (or
a constituent containing the verb) moves even higher, does not make this pre-
diction. Let me show that the facts support the approach in which preverbal
subject positions become unavailable in interrogatives.
(i) Third person overt pronouns in European Spanish must be interpreted
as animate (a property of Strong Pronouns in Cardinaletti and Starke’s 1999
sense) if they are in a preverbal subject position (20a); however, in the inter-
rogative, in which the same pronoun is postverbal, the animacy effect does
not take place (20b). This is expected if the high subject positions that become
unavailable in interrogatives are associated with strong pronouns: Staying
low, the overt pronoun can escape this constraint.
(20) [Talking about a table, una mesa (feminine)]

a. #Ella tiene patas.
she has legs
b. (?) ¿ Qué tiene ella?
(?) what has she
‘What does it (= feminine) have?’
(ii) Adapting a test that Cardinaletti (2004) used for Italian, expletive subjects
will be able to appear only in the highest subject positions, because they are
introduced for purely formal reasons. European Spanish has only one context
in which there is an expletive overt subject (21a), in a semi-archaic expression.
However, (21b), with inversion, is unavailable: In such contexts the expletive
pronoun simply does not appear.
(21) a. Ello es que María tiene una hija.

that is that María has a daughter
‘It just so happens that María has a daughter’
b. ¿Es (*ello) que María tiene una hija?
is that that María has a daughter
Intended: ‘Is it the case that María has a daughter?’
Both contrasts receive a common explanation if the high subject position

licenses expletive subjects (unavailable lower, perhaps because they are not
introduced inside the verb) and imposes a strong pronoun interpretation on
pronouns; as this position becomes unavailable, the effects disappear.
262
10.4 MOTIVATING A HIERARCHY AND DETERMINING

THE LEXICALIZED PARTS
To get deep into the details of the proposal, let us examine more closely the va-
riety of Mérida (Venezuela), for which not all subjects can appear in preverbal
position in interrogatives.
Remember that in Mérida (Venezuela) preverbal subjects are possible
under two conditions:
(i) They must contain participant features, that is, correspond to the
speaker or the addressee.
(ii) Or, if they are third person, they must be contrastive.
Let us examine these two conditions in turn and see how they are reflected in
the structure.
10.4.1 Topic positions
To examine the condition on contrastiveness, the proposal that there

are different preverbal topic positions becomes relevant. Frascarelli and
Hinterhölzl (2007) have argued that there are three distinct high topic
positions in the left periphery (contra the view that topic positions are simply
iterated; Rizzi 1997):
(i) Familiarity topic, which recovers a referent already given in the pre-
vious discourse;
(ii) Contrastive topic, which is also given but also establishes a contrast be-
tween that referent and another referent in the universe of discourse;
(iii) Aboutness-shift topic, which is also given and contrastive, but changes
the topic that the statement is about.
In the following example (22), we see a sentence with all three of them:
(22) [Acerca de la cena]shift, [nosotros]cont [el postre]fam lo hemos traído.

about of the dinner we the dessert it have brought
‘Talking now about dinner, WE have brought the dessert’
This seems to suggest that the ordering among these three topics is Shift >
Contrastive > Familiarity (see Frascarelli 2007; Frascarelli and Hinterhölzl
2007; Bianchi and Frascarelli 2010; Frascarelli 2012); in this chapter, the
topics are differentiated because at each higher level a new feature is added to
the lower levels: [given] (common to all topics) is the only property defined in

familiarity topics, which can be taken to be “vanilla” topics; adding [contrast]
produces contrastive topics; adding to these two [shift] produces aboutness-
shift topics.
Placing these positions with respect to IntP is not uncontroversial. It is
clear that aboutness-shift is above interrogatives, as shown among other
things by the fact that it precedes wh-words (23):
(23) a. Acerca de la cena, ¿ quién trae el postre?

about of the dinner who brings the dessert
‘Talking now about dinner, who brings dessert?’
b. ?? ¿ Quién trae, acerca de la cena, el postre?
who brings about of the dinner the dessert
However, Frascarelli (2012) places contrastive topics also above IntP. Here
I would like to argue that the contrastive topic position is lower than IntP. In
fact, in the varieties in which preverbal subjects can appear in interrogatives,
a contrastive topic subject is included inside the interrogative sentence (24):
(24) ¿Qué él dice?

what he says.3sg
‘What does HE say?’
(24) can be used in a context in which we are discussing the opinion of two of
our friends and oppose the opinion of one of them to one of the other. This
contrasts with (25), outside the interrogative, that can be used only in a con-
text in which we have been discussing the opinion of one friend, and now we
move the discussion to a second friend whose views we were not talking about
before.
(25) (Y) él, ¿qué dice?

and he what says.3sg
‘And him, what does he say?’
Independent evidence for the relative position of contrastive topic and in-
terrogative comes from data reported in Constant (2014: 366 et passim).
Discussing data from Mandarin Chinese, Constant explores the meaning
of the contrastive topic marker -ne inside wh-questions. He convincingly
argues that the availability of this marker in that context is subject to the
possibility of dividing the wh-question into subquestions that in turn ask,
contrastively, about each one of the potential answers. In (26), for instance,
adding the particle -ne equals implictly asking whether the children are at
school, or whether the children are in the park, or whether they are in the
cinema, and so forth.
264
(26) Tāmen dàodǐ zài năr ne?

they eventually at where NE?
‘Where (the hell) are they?’
The particle -ne cannot be used in out-of-the-blue contexts in which there is

no given set of potential answers; (27) cannot be used to ask the time of a
stranger in the street. It could be appropriate in a situation in which we know
that there are trains arriving at 15.30, 15.45, and 16.00, and, seeing a train
arrive, we ask an acquaintance who is also aware of this fact that one of these
three times it is in order to identify the train.
(27) Qǐngwèn xiànzài jǐ-diǎn le (#ne)?

sorry now how-many-o’clock asp NE?
‘Excuse me, what time is it?’
But if the combination of a contrastive topic marker and a wh-interrogative

produces an interpretation in which one is asking to identify one out of a set
of contrastive potential referents, then it follows that the contrastive topic
has to be under the scope of the interrogative, as in Int > Contrastive.
Finally, we have familiarity topics, which are the lowest of all the topic
elements. If contrastive topics are below IntP and familiarity topics are below
contrastive ones, by transitivity, familiarity topics will be below IntP. But how
much lower exactly? Haegeman (2006) notes that Clitic Left Dislocations in
Romance, which are the prototypical manifestations of familiarity topics, are
allowed in a number of subordinate contexts, which she uses to convincingly
argue that the familiarity topic position is licensed independently of ForceP
and the higher CP elements, like SubordinatorP. In contrast to Benincà (2001)
and Benincà and Polletto (2004), she argues with Frascarelli and Hinterhölzl
(2007) that Romance Clitic Left Dislocations are lower than FocusP, and in
fact (Haegeman 2006, 47) proposes that the familiarity topic position should
be related to FinitenessP, as the internal logophoric center of the clause. Thus,
combining all these elements, we get the ordering of heads in (28).8
(28) [ShiftP [IntP [ContrP [FamP (= FinP) [TP...]]]]]
10.4.2 ParticipantP
A second factor in Mérida (Venezuela) is whether the subject includes par-

ticipant features. The nature of participant features has been extensively
8. For explicitness, I follow Haegeman 2006 in the claim that the dislocated element
is in FinP; about the clitic double, see Section 10.4.4.

discussed in the literature (Bianchi 2003, 2006; Speas and Tenny 2003;
Sigurdsson 2004, 2009, 2010; Hill 2007; Giorgi 2009; Collins and Postal 2012,
among others). All these approaches, different as they are, agree in assigning a
special role to participant features in the clause, and in proposing that (to the
extent that their deictic nature involves anchoring to the discourse) they have
to be licensed inside or through a high clausal position.
Bianchi (2003) proposes that Finiteness is the syntactic head that encodes
the speech event, including reference to speaker and addressee. When the
clause is nonfinite, as in an infinitive, Bianchi argues, the speech event is not
specified in Fin, from where two properties follow: First, the infinitival clause
cannot refer deictically to a time interval ordered with respect to the speech
time, so it has to get anchored anaphorically to a time interval denoted by the
main verb; second, person agreement becomes impossible.
A shortcoming of this approach is that the availability of deictic par-
ticipant features and the availability of deictic tense do not always match.
Leaving aside languages in which infinitives do agree in person (such as
Portuguese; Raposo 1987), intuitively there is an asymmetry between tem-
poral deixis and participant deixis: Although tense is not deictic in an infini-
tival clause, a speaker or addressee pronoun contained inside the same kind
of clause can still refer to the speaker or addressee as defined in the main
speech event. In other words, a logophoric interpretation of a first person
pronoun, although possible under certain conditions (see Sundaresan
2013), is not forced in an infinitival clause. This suggests that if Finiteness
is responsible for anchoring tense, a different head should be in charge of
anchoring the participants.
This is, in fact, what Sigurdsson (2004) proposes. In his view, there are a
number of projections identifying the features of the speech event: Two of
them identify the logophoric agent (= speaker) and the logophoric patient
(= addressee). A first or second person feature in a pronoun contained in
the inflectional area of the clause defines its reference by coreferring with
their local logophoric feature in this area (Sigurdsson and Maling 2006,
10). These speech-event-related projections are below Force and above
Fin (29):
(29) [ForceP [Λagent [Λpatient [ . . . TP]]]]
Here we follow Sigurdsson’s proposal rather than Bianchi’s: The participant

layer is distinct from the layer in which the deictic tense is defined, possibly
Fin, which, following Haegeman (2006), is also the position for dislocated fa-
miliarity topics in Romance. Putting all this together, we obtain the sequence
in (30):
(30) [ShiftP [IntP [ContP [ParticipantP [FinP (= FamP) [SubjP [TP]]]]]]]
266
The relative position between contrastive topics and the anchoring of

participants is difficult to determine. Here we are taking the decision of
placing ContrP over PartP because it seems that there are cases in which the
reference of a second person pronoun inside the same utterance can vary if
the second person is taken as contrastive. Consider the example in (31), in
which each instance of you identifies, contrastively, a different member of the
group to which the speaker is talking.
(31) Youi will bring the milk, youj will buy the flour, youk will help me make
the cake, and youl will decorate it.
Such cases suggest that the identification of the addressee can fall inside the
scope of the contrastive topic operator, forcing it to vary. However, nothing
crucial in our analysis depends on the relative ordering of these two heads or,
for that matter, on the possibility that (as Sigurdsson proposes), ParticipantP
could be split into two distinct heads.
10.4.3 The exponents in the varieties
With the previous sequence in mind, let us see how a nanosyntactic account
with phrasal spellout manages to restrict the range of pronouns that can act as
preverbal subjects in wh-interrogatives in Mérida (Venezuela). My proposal is
that in this variety the wh-word spells out WhP and IntP (Figure 10.11), and the
subject agreement exponent spells out FamP (= FinP) and SubjP (Figure 10.12).
WhP ⇒ qué
‘what’
Wh IntP
Int
Figure 10.11 Wh-word corresponds to WhP and IntP in Mérida
FinP ⇒ -mos
‘1pl’
Fin SubjP
Subj
1pl
Figure 10.12 Subject agreement corresponds to FinP and SubjP in Mérida
Consider how this explains the facts. First, on the crucial assumption that
no other exponent spells out FinP in this variety, we predict that in Mérida a
preverbal subject will not be able to occupy the specifier of SubjP, because that
would break the constituent necessary to introduce the subject inflection.

A preverbal subject, then, would have to carry some additional property that
licenses it in a higher position where it does not break the constituent spelled
out by the inflection.
Take first third person pronouns in a contrastive reading. The relevant tree is
shown in Figure 10.13: as ContrP is higher than FinP and lower than IntP, a pronoun
can be spelled out in this position without breaking any of the two constituents.
WhP ⇒ qué
Wh IntP
Int ContrP
él Contr
Contr FinP ⇒ (dice)-Ø
Fin SubjP
él Subj
Subj
3sg
Figure 10.13 Contrastive third person pronoun
Take now participant pronouns; assume that entering in a checking relation

with PartP in the high clause means that they can move to its specifier. This posi-
tion also removes the preverbal subject from Spec-SubjP. Consequently, spellout
is possible (see Figure 10.14).
WhP ⇒ qué
Wh IntP
Int PartP
tú Part
Part FinP ⇒ (dice)-h
Fin SubjP
tú Subj
Subj
2sg
Figure 10.14 Participant pronoun moves to Spec-PartP from Spec-SubjP
The consequence is going to be, then, that third person preverbal subjects will
be licensed only in a contrastive interpretation, whereas participant preverbal
subjects will be licensed even without a contrastive interpretation, just by their
268
nature as pronouns that take their reference from the speech participants. This is
precisely what the consulted speakers report in their judgments.
If we go back to European Spanish, we need to slightly revise the lexical
entry of the subject agreement exponent to include PartP (Figure 10.15).9
WhP
Wh IntP ⇒ (dice)-s
qué
Int PartP
Part FinP
Fin SubjP
Subj TP
2sg
Figure 10.15 Subject agreement includes PartP in European Spanish
Given the constituent that gets lexicalized by subject agreement in an in-

terrogative, the (right) prediction is that European Spanish will not be able to
save a preverbal subject in an interrogative by giving it a contrastive reading,
a familiarity topic reading or by virtue of their participant nature. The only
way of having a subject before the verb in a wh-sentence will be to treat it as
an aboutness-shift topic, in which case it will also precede the wh-word.
Finally, we have seen that in Dominican Republic Spanish any subject can
be preverbal, even without a contrastive reading. Our proposal is that in this
variety, again, the wh-word spells out IntP; as for the subject agreement expo-
nent, it is the spellout of just SubjP (marked in bold in Figure 10.16).
WhP ⇒ qué
Wh IntP
Int ...FinP
Fin SubjP
Ø
uno Subj
‘one’
Subj TP
3SG
(dice)-Ø
Figure 10.16 Agreement spells out only SubjP in Dominican Republic Spanish
9. As the attentive reader will have noticed, in these entries we are taking the non-
trivial decision that specialized topic and focus positions are not projected as part of
the obligatory series of heads that form the spine of a sentence. This contrasts with a
Cinque (1999) style view, in which all heads are present in the tree and when they are
not used are projected with a negative value.

10.4.4 The verb stem and clitics
Finally, a residual issue is how the verb stem (with optional tense markers)
ends up to the left of the subject agreement exponent, as in (32):
(32) ¿Quiénes viene-n?

who.pl come-3pl
‘Who is coming?’
The first question is what kind of constituent can appear to the left of the subject
agreement exponent in an interrogative. The right generalization is not that it is
a syntactic constituent: Note (33), where we can see that the object clitics and the
negation—itself a clitic—can also appear in this position:
(33) ¿Quiénes [no te lo da]-n?

who.pl not you.dat it.acc give-3pl?
‘Who does not give it to you?’
We are not aware of any syntactic analysis in which negation, object clitics, and
the verb form a single syntactic constituent to the exclusion of the other material
contained under TP, NegP, or FP (in which, following Uriagereka 1995, we assume
object clitics move syntactically). This makes it very unlikely that this ordering is
obtained by moving the verb in syntax up to a position between WhP and IntP.
The right generalization is prosodic: The material appearing to the left of
the subject exponent forms a single prosodic constituent. In other words: The
subject exponent must end up being a suffix of the complex formed by the
verb and any potential clitic that forms a prosodic unit with it. Nanosyntax is
a proposal in which one cannot use morphological operations to account for
the position of morphemes, but lexical entries do have phonological informa-
tion. Inspired by Richards (2014), I propose that the subject exponent entries
contain, as part of their phonological information, the right boundary of a
prosodic constituent (for explicitness, a foot) [see (34)].10
(34) ‘]ϕ mos’ (1pl)
The effect of this phonological condition is that after narrow syntax has de-
fined the main constituent order and the principles of spellout have restricted
10. An anonymous reviewer, to whom I am grateful, rightly notes that if the prosodic
condition is defined for each exponent separately we could expect that within the same
paradigm agreement could be suffixal or prefixal depending on the specific entry, a sit-
uation that is attested, but not common. This is a shortcoming of the approach, unless
we resort to a principle of regularity within an exponent set that favors the extension
of the same prosodic condition across entries that are related by their function.
270
the available positions allowed by those entries, there is prosodic reordering

at the phonological component that acts under strict linearity and operates
completely cyclically, so that the exponent in IntP must satisfy its phonolog-
ical conditions with the material that has already been spelled out (that is, the
verb stem to its right or the verb stem plus the clitics, if the entries of those
exponents forced them already to form a prosodic constituent with the verb
stem). (35a) represents the order of exponents before this prosodic operation
(where ClG = Clitic Group); (35b) shows the order required to satisfy the pho-
nological requisite of the exponent: The verb stem (plus clitics) is reordered;
and (35c) the rightmost end anchors its foot boundary to the right boundary
of the prosodic constituent (a foot structure without clitics; possibly a Clitic
Phrase with attached clitics).11
(35) a. ]ϕ mos ^ [(no lo) [canta-]ϕ]ClG

b. [[(no lo) [canta]ϕ]ClG ^ ]ϕ mos
c. [[(no lo) canta]ϕ mos]ClG
10.5. CONCLUSIONS AND FURTHER PROSPECTS
In this chapter I have argued that the variation attested in the available subject
positions for wh-interrogatives across three Spanish varieties can be accounted
for using the nanosyntactic tool of phrasal spellout. Specifically, the minimal
difference is the size of the subject agreement exponent (see Table 10.1).
Table 10.1 SUBJECT AGREEMENT EXPONENTS
IntP PartP FinP SubjP
European
Subject agreement exponent
Spanish
Mérida
(Venezuela) Subject agreement exponent
Spanish
Dominican Subject
Republic agreement
Spanish exponent
11. The crucial assumption here, as the reader has noted, is that any number of clitics
do not form a foot, so that displacing them without the verb stem cannot satisfy the
prosodic conditions of the exponent; note that clitics, no matter how many, never get
rhythmic stress assigned in Spanish, which supports the claim.

As expected from the Superset Principle and the Anchor Condition, in
European Spanish the exponent shrinks in declarative contexts, for which
there is no IntP.
Let us take a moment to see some of the implications of this proposal. The first
immediate consequence is that we not only expect overt adverbs and subjects
to be unavailable in preverbal position in interrogatives, but also we expect
properties that are licensed by Fin, or the other heads involved in the constituent
that is spelled out as subject agreement, to be unavailable in interrogatives. On
this note, consider the following sentence from European Spanish (36):
(36) No sé quién viene.

not know who comes
‘I don’t know who is coming’
The surprising property of this sentence, and all subordinate interrogative

clauses, is that they depend on main predicates expressing doubt or uncer-
tainty; this class of predicates systematically introduce subjunctive clauses,
as can be seen in (37). The only instance in European Spanish of an uncer-
tainty predicate selecting indicative is precisely subordinate interrogatives
like (36).
(37) {No creo /No pienso /Dudo de} que {esté /*está} enfermo.
not believe /not think /doubt of that is.subj /is.ind sick
‘I {don’t believe /don’t think /doubt} that he is sick’
Now, licensing the subjunctive requires that FinP has a marked value as the
internal center of the clause (Giorgi and Pianesi 2005), codifying that the
subordinate clause is not ascribed to the mental model of the subject of the
main predicate (Quer 1998). But in the interrogative, as the subject agreement
exponent has to lexicalize up to IntP, including FinP, there is no structural
place to introduce the marked Fin value that licenses the subjunctive; sub-
junctive might involve an extra structural layer above Fin, in which case that
layer would intervene and prevent lexicalization, or a special “flavor” of FinP,
which would be different from the one lexicalized inside the subject exponent.
Either way, subjunctive is correctly expected to be blocked inside interrogative
clauses in European Spanish.
Moreover, if our analysis is on the right track, we expect Dominican
Republic Spanish to allow subjunctive inside subordinate interrogatives, be-
cause their exponents do not lexicalize Fin; see (38):
(38) No sé quién sea usted.

not know who are.subj you
‘I don’t know who you are’ (El Nuevo Diario, January 1, 2014)
272
In Mérida, as the exponent lexicalizes Fin, but not Int, we expect that—
mutatis mutandis—the subject agreement exponent will also be able to shrink
if Fin codifies subjunctive and the subjunctive morphology spells out Fin; our
informants from Mérida accept (38) as natural also.
As a second prediction, in which the Spanish subject agreement spells
out interrogatives, my approach predicts that there should exist languages
in which there are at least two sets of person or number markers for the
subject, one in declaratives and one in interrogatives. In fact, the Omotic
language Gimira (Breeze 1990, 30–37) is one such language. Table 10.2
shows the set of person or number markers in declaratives; Table 10.3
shows the set used in questions (in which,furthermore, polar and wh-
questions are differentiated, lending additional support to my claim that
WhP has to be distinct from IntP; here I show only the set of forms for
wh-questions).
Table 10.2 GIMIRA AGREEMENT

MARKERS: DECL ARATIVES
sg pl
1 -u -u
2 -u -end
3 -u (m) /-en (f) -end
Table 10.3 GIMIRA AGREEMENT MARKERS: INTERROGATIVES
sg pl
1 -ao -ao (exclusive) /-eo (inclusive)
2 -ao -eo
3 -eo (m) /-ao (f) -eo
Given these facts, Spanish would be like Gimira in this respect, only that,
not having distinct items for interrogatives, syncretism emerges at the surface.
I make a further prediction with respect to subordinate clauses. In non-wh-
subordinate questions, all varieties can have preverbal subjects (39):
(39) Me pregunto si vosotros vendréis.

me wonders if you will.come.2pl
‘I wonder whether you will come’
This is expected if the fact that si ‘if’ is an interrogative complementizer

means that it spells out IntP. Assuming a head Sub(ordinate)P introducing
subordinate clauses, si would spell out Sub plus Int, and the subject

agreement exponent would shrink, allowing the subject to precede the
verbal complex (40):
(40) a. [Sub [Int [Part . . . ]]]

b. [ si [vosotros . . . ]]
This explanation is plausible given that si ‘if’ is incompatible with wh-words,

as expected if WhP is between SubP and IntP [see (41) and (42)].
(41) Me pregunto (*si) quién vendrá.

me wonders if who will.come.3sg
‘I wonder (*whether) who will come’
(42) a. [Sub [wh [Int [Part . . . ]]]]

b. [ *si [quién]]
This fine-grained distinction supports an approach in which the word ordering

facts follow from the size of specific Vocabulary Items.
Our account could be extended to two similar, but not identical, phe-
nomena that also set these varieties apart and have to do with the avail-
able positions for subjects. In Mérida (Venezuela) and Dominican
Republic Spanish, pronominal subjects can be preverbal also in infinitival
constructions (43):
(43) Se cambió la ley para él jurar el cargo.

SE modified the law to he swear.inf the position
‘The law was modified for him to swear the position’
This is ungrammatical in European Spanish. It is tempting to relate this

difference to the word order in interrogatives through the different size of
the infinitival exponent -r and the size of the exponent that materializes
the complementizer. Similarly, the two mentioned varieties (but not
European Spanish) allow for resumptive subject pronouns in relative
clauses.
(44) una chica que ella es muy guapa

a girl who she is very pretty
‘a girl who is very pretty’
It is not inconceivable that this construction is possible in these varieties

because of a parallel impoverishment in the size of the relative pronoun ex-
ponent. However, these aspects of the grammar of the two varieties are not
explored in this chapter, and will be left for future research.
274
REFERENCES
Abels, Klaus and Peter Muriungi. 2008. “The Focus Particle in Kîîtharaka: Syntax and
Semantics.” Lingua 118: pp. 687–731.
Aelbrecht, Lobke. 2010. The Syntactic Licensing of Ellipsis. Amsterdam: John Benjamins.
Bakovic, Eric. 1998. “Spanish Codas and Overapplication.” In Romance Linguistics:
Theoretical Perspectives: Selected papers from the 27th Linguistic Symposium on
Romance Languages (LSRL XXVII), Irvine, February 20–22, 1997. edited by
Armin Schwegler, Bernard Tranel, and Myriam Uribe-Etxebarria, pp. 13–23.
Amsterdam and Philadelphia: John benjamins.
Benincà, Paola. 2001. “The Position of Topic and Focus in the Left Periphery.” In
Current Studies in Italian Linguistics, edited by Guglielmo Cinque and Giampaolo
Salvi, pp. 39–64. Dordrecht, The Netherlands: Foris.
Benincà, Paola and Cecilia Poletto. 2004. “Topic, Focus and V2: Defining the CP
Sublayers.” In The Structure of CP and IP, edited by Luigi Rizzi, pp. 52–75.
Bianchi, Valentina. 2003. “On Finiteness as Logophoric Anchoring.” In Tense and Point
of View, edited by Jacqueline Guéron and Liliane Tasmowsky, pp. 213–246.
Paris: Université Paris X Nanterre.
Bianchi, Valentina. 2006. “On the Syntax of Personal Arguments.” Lingua 116:
pp. 2023–2067.
Bianchi, Valentina and Mara Frascarelli. 2010. “Is Topic a Root Phenomenon?” Iberia
2: pp. 43–88.
Breeze, Mary J. 1990. “A Sketch of the Phonology and Grammar of Gimira (Benchnon).”
In Omotic Language Studies, edited by Richard J. Hayward, pp. 1–67. London:
School of Oriental and African Studies, University of London.
Buesa García, Carlos. 2008. The Subject-Gap Restriction in Spanish Wh-Questions:
Some Empirical and Theoretical Consequences. Second General Examination.
Ms., University of Connecticut.
Cardinaletti, Anna. 2004. “Towards a Cartography of Subject Positions.” In The Structure
of CP and IP, edited by Luigi Rizzi, pp. 115–165. Oxford: Oxford University Press.
Cardinaletti, Anna and Michal Starke. 1999. “The Typology of Structural Deficiency: A
Case Study of the Three Classes of Pronouns.” In Clitics in the Languages of
Europe, edited by Henk van Riemsdijk, pp. 145–233. Berlin: De Gruyter.
Chomsky, Noam. 1991. “Some Notes on Economy of Derivation and Representation.”
In Principles and Parameters in Comparative Syntax, edited by Robert Freidin, pp.
417–454. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and Functional Heads. Oxford: Oxford University Press.
Collins, Chris and Paul Postal. 2012. Imposters: A Study of Pronominal Agreement.
Constant, Noah. 2014. Contrastive Topic: Meaning and Realizations. Doctoral disserta-
tion, University of Massachusetts Amherst.
Culicover, Peter and Ray Jackendoff. 2005. Simpler Syntax. Oxford: Oxford University
Press.
Frascarelli, Mara. 2007. “Subject, Topic and the Interpretation of Referential Pro.”
Natural Language and Linguistic Theory 25: pp. 691–734.
Frascarelli, Mara. 2012. “The Interpretation of Discourse Categories. Cartography for
a Crash-Proof Syntax.” In Enjoy Linguistics! Papers Offered to Luigi Rizzi’s on the

Occasion of His 60th Birthday, edited by Valentina Bianchi and Cristiano Chesi,
pp. 180–191. Siena, Italy: CISCL Press, Università degli Studi di Siena.
Frascarelli, Mara and Roland Hinterhölzl. 2007. “Types of Topics in German and
Italian.” In On Information Structure, Meaning and Form, edited by Susanne
Winkler and Susanne Winkler, pp. 87–116. Amsterdam: John Benjamins.
Ginzburg, Jonathan and Ivan Sag. 2000. Interrogative Investigations: The Form, Meaning
and Use of English Interrogatives. Stanford, CA: CSLI Publications.
Giorgi, Alessandra. 2009. About the Speaker. Oxford: Oxford University Press.
Giorgi, Alessandra and Fabio Pianesi. 2005. “Credo (I Believe): Epistemicity and the
Syntactic Representation of the Speaker.” University of Venice Working Papers in
Linguistics 15: pp. 105–152.
Haegeman, Liliane. 2006. “Argument Fronting in English, Romance CLLD and the
Left Periphery.” In Cross-Linguistic Research in Syntax and Semantics, edited
by Raffaella Zanuttini, Héctor Campos, Elena Herburger, and Paul Portner,
pp. 27–52. Washington, DC: Georgetown University Press.
Haegeman, Liliane. 2012. Adverbial Clauses, Main Clause Phenomena and the Composition
of the Left Periphery. Oxford: Oxford University Press.
Inflection.” In The View From Building 20, edited by Kenneth Hale and Samuel J.
Keyser, 111–176. Cambridge, MA: MIT Press.
Hardt, Daniel. 1993. Verb Phrase Ellipsis: Form, Meaning, and Processing. Ph.D. Thesis,
University of Pennsylvania.
Henríquez Ureña, Pedro. 1939. “Ello.” Revista de Filología Hispánica 1: pp. 209–229.
Henríquez Ureña, Pedro. 1940. El Español en Santo Domingo. Buenos Aires: Biblioteca
de Dialectología Hispanoamericana.
Hill, Virginia. 2007. “Vocatives and the Pragmatics-Syntax Interface.” Lingua 117: pp.
2077–2105.
Jiménez Sabater, Max. 1975. Más Datos Sobre el Español de la República Dominicana.
Santo Domingo: Ediciones Intec.
Kany, Charles. 1951. American Spanish Syntax. Chicago: Chicago University Press.
Lasnik, Howard. 2001. “When Can You Save a Structure by Destroying It?” In
Proceedings of NELS 31, edited by Minjoo Kim and Uri Strauss, pp. 301–320.
Amherst, MA: University of Massachusetts, Graduate Linguistics Students
Association.
Leu, Thomas. 2015. “Generalized x-to-C in German.” Studia Linguistica 69: pp. 272–303.
Lipski, J. 2002. Subject Pronoun Usage Among Spanish Dialects. Ms., Penn State University.
Lobeck, Anne. 1995. Ellipsis: Functional Heads, Licensing and Identification.
Merchant, Jason. 2001. The Syntax of Silence. Oxford: Oxford University Press.
Morales, Amparo. 1986. Gramáticas en Contacto: Análisis Sintácticos Sobre el Español de
Puerto Rico. Madrid: Playor.
Mulugeta, Seyoum. 2008. A Grammar of Dime. Utrecht, The Netherlands: Landelijke
Onderzoekschool Taalwetenschap.
Ordóñez, Francisco. 2007. “Cartography of Postverbal Subjects in Spanish and Catalan.”
In Romance Languages and Linguistic Theory 2005, edited by Sergio Baauw, Frank
A.C. Drijkoningen, and Manuela Pinto, pp. 259–280. Amsterdam: John Benjamins.
Ordóñez, Francisco and Esthela Treviño. 1999. “Left Dislocated Subjects and the Pro-
Drop Parameter.” Lingua 107: pp. 39–68.
Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar and the Structure of
IP.” Linguistic Inquiry 20, pp. 365–424.
276
Prieto, Pilar and Paolo Roseano. 2009. Atlas Interactivo de la Entonación en Español. [on-
line] Available at http://prosodia.upf.edu/atlasentonacion/tobi/index.html
Quer, Josep. 1998. Mood at the Interfaces. The Hague: Holland Academic Graphics.
Raposo, Eduardo. 1987. “Case Theory of INFL to COMP: The Inflected Infinitive in
European Portuguese.” Linguistic Inquiry 18: pp. 85–109.
Richards, Norvin. 2014. Contiguity Theory. Ms., MIT. [online] Available at http://ling.
auf.net/lingbuzz/002247.
edited by Liliane Haegeman, pp. 281–337. Dordrecht, The Netherlands: Kluwer.
Rizzi, Luigi. 2001. “On the Position Int(errogative) in the Left Periphery of the Clause.” In
Current Studies in Italian Syntax Essays offered to Lorenzo Renzi, edited by Guglielmo
Cinque and Giampaolo Salvi, pp. 286–296. Amsterdam: North Holland.
Ross, John R. 1973. “The Penthouse Principle and the Order of Constituents.” In You Take
the High Node and I’ll Take the Low Node, edited by Claudia W. Corum, Thomas C.
Smith Stark, and Ann Weiser, pp. 397–422. Chicago: Chicago Linguistic Society.
Sag, Ivan and Jorge Hankamer. 1984. “Toward a Theory of Anaphoric Processing.”
Linguistics and Philosophy 7: pp. 325–345.
Sigurdsson, Halldór Á. 2004. “The Syntax of Person and Person Features.” Italian
Journal of Linguistics 16: pp. 219–251.
Sigurdsson, Halldór Á. 2009. “Language Quarks.” Iberia 1: pp. 169–183.
Sigurdsson, Halldór Á. 2010. “On EPP Effects.” Studia Linguistica 64: pp. 159–189.
Sigurdsson, Halldór Á. and Joan Maling. 2006. Argument Drop and the Empty Left Edge
Condition. Ms., University of Lund.
Silva Corvalán, Carmen. 1982. “Subject Expression and Placement in Mexican-American
Spanish.” In Spanish in the United States: Sociolinguistic Aspects, edited by Jon
Amaste Elías-Olivares, Lucía, pp. 93–120. Cambridge: Cambridge University Press.
Speas, Margareth and Carol Tenny. 2003. “Configurational Properties of Point of View
Roles.” In Asymmetry in Grammar, edited by Anna Maria di Sciullo, pp. 315–343.
Sundaresan, Sandhya. 2013. Context and (Co)Reference in the Syntax and Its Interfaces.
Suñer, Margarita. 1994. “V-Movement and the Licensing of Argumental wh-Phrases in
Spanish.” Natural Language and Linguistic Theory 12: pp. 335–372.
Svenonius, Peter. 2008. “Interpreting Uninterpretable Features.” Linguistic Analysis
3–4: pp. 375–413.
Toribio, Almeida Jacqueline. 2000. “Setting Parametric Limits on Dialectal Variation in
Spanish.” Lingua 110: pp. 315–341.
Torrego, Esther. 1984. “On Inversion in Spanish and Some of Its Effects.” Linguistic
Inquiry 15: pp. 103–129.
Uriagereka, Juan. 1995. “Aspects of the Syntax of Clitic Placement in Western
Romance.” Linguistic Inquiry 26: pp. 79–123.
Vangsnes, Øystein. 2002. “Icelandic Expletive Constructions and the Distribution of
Subject Types.” In Subjects, Expletives and the EPP, edited by Peter Svenonius,
Villa García, Julio. 2015. The Syntax of Multiple-que Sentences in Spanish. Amsterdam:
John Benjamins.
Zagona, Karen. 2002. The Syntax of Spanish. Cambridge: Cambridge University Press.
Zubizarreta, María Luisa. 2012. “A Note on the Spanish Left Periphery.” In Functional
Heads: The Cartography of Syntactic Structures 7, edited by L. Brugé, Anna
Cardinaletti, Giuliana Giusti, Nicola Munaro, and Cecilia Poletto, pp. 112–125.

CHAPTER 11
The Feature Structure of Pronouns

A Probe Into Multidimensional Paradigms*
GUIDO VANDEN W YNGAERD
11.1 INTRODUCTION
Personal pronoun paradigms are multidimensional, as they involve (at least)

the feature dimensions of person and number. Syncretisms in multidimen-
sional paradigms may be horizontal, vertical, or both (Table 11.1).
Table 11.1 VERTICAL AND HORIZONTAL

SYNCRETISMS IN PRONOUN PARADIGMS
sg pl sg pl
1 A B 1 A A
2 A C 2 B C
3 D E 3 D E
As I demonstrate in Section 11.4 below, deriving the horizontal or cross-

number syncretism is unproblematic given the Superset Principle, but deriving
the cross-person or vertical syncretism requires an extension of the theory.
The structure of this chapter is as follows. In Sections 11.2 and 11.3, I lay out
some assumptions I make concerning the internal structure of the person and
number feature complex, respectively. Section 11.4 examines some attested
horizontal and vertical syncretisms and shows how the horizontal ones can
* I am grateful to the anonymous reviewers of OUP for their feedback and suggestions.
For useful discussion and suggestions, I wish to thank Pavel Caha, Karen De Clercq,
Michal Starke, and the audiences at SLE 47 in Poznan and at SinFonIJA9 in Brno.
278
be derived and why the vertical ones are problematic. Section 11.5 examines
the solution to this problem in terms of the nanosyntactic mechanism of
pointers, showing how they allow the derivation of ABA patterns. Section 11.6
discusses another solution in terms of a revision of the Superset Principle.
The different predictions these two approaches make are examined in detail
in Section 11.7. Section 11.8 addresses the issue of possible but unattested
syncretisms. Finally, Section 11.9 explores some further consequences of the
Revised Superset Principle (RSP).
The empirical evidence on which this chapter is based is mostly taken from
the typological study by Cysouw (2003). I have nevertheless applied a number
of restrictions to this material. Cysouw discusses all kinds of person marking,
that is, both independent pronouns as well as inflectional person markers.
I focus exclusively on person marking in independent pronouns. This restric-
tion is made possible by the fact that Cysouw is generally quite explicit on this
issue in his description of the data.
It is obvious that verbal agreement morphology also bears on the issue of
the internal structure of the person feature complex. Yet at the same time there
are some important empirical differences between both systems. As we will see,
syncretisms in the personal pronouns are extremely rare in the singular, almost
to the point of being nonexistent. The picture in the plural is more complicated,
in ways that will become clear as we proceed. Syncretisms in the verbal inflec-
tion, on the other hand, are extremely common, both in the singular and in the
plural (see, for example, the findings by Baerman et al. 2005; Aalberse 2007;
Aalberse and Don 2009; Ackema and Neeleman 2013; Baerman and Brown 2013).
These studies have found that none of the possible person syncretism patterns
is actually unattested in the verbal inflection, though not all of them are equally
common. A second empirical difference between pronominal paradigms and in-
flectional agreement paradigms is that the latter are typically sensitive to feature
dimensions lacking in the pronouns, such as tense and verb class, so that there
may be several paradigms, each with its own syncretism, within the same lan-
guage. Independent pronouns do not display any tense or class sensitivity and
therefore constitute a clearly distinct case. It is clear that a full investigation of
the person feature complex will eventually have to incorporate verbal inflection as
well, but for now the restriction to independent pronouns seems well motivated.
Another empirical domain that I leave aside is the inclusive–exclusive dis-
tinction. Inclusive pronouns probably involve a composition of the features
of first and second person pronouns, whereas exclusive ones involve only the
features that enter into the make-up of the first person. However, because this
is not the central topic of this chapter, I leave this matter aside for now.
11.2 THE PERSON FEATURE COMPLEX
I adopt the proposal put forth by Béjar (2003). This proposal assumes that
there are three privative features, [speaker], [participant], and [person],

which are moreover cumulative, that is, they stand in a containment relation-
ship. The first person personal pronoun has all three of these features, be-
cause the first person is the speaker, a participant, and a person. The second
person pronoun has only participant and person, and the third only person.
For expository purposes, I refer to these features by numbers (1 = [speaker],
2 = [participant], and 3 = [person]). The feature trees for the first, second, and
third person personal pronouns are given in Figure 11.1 as (a), (b), and (c),
respectively.
(a) 1P (b) (c)
1 2P 2P
2 3P 2 3P 3P
3 3 3
Figure 11.1 Syntactic trees for first, second, and third person personal pronouns
Given these syntactic feature trees, we expect to find syncretisms between

1 and 2 (AAB), 2 and 3 (ABB), and 1, 2, and 3 (AAA), but no syncretism of 1 and
3 across 2 (*ABA). Consider a hypothetical example of an ABB pattern, that
is, a 2–3 syncretism. This pattern can be shown to result from lexical items as
given in Figure 11.2.
In a 1P syntactic tree, (a) of Figure 11.2 will be the only candidate for
insertion by virtue of the Superset Principle, because the lexical tree
(b) does not contain the syntactic tree as a subtree (I am assuming here
the definition of the Superset Principle as given in Caha 2009, 55). In
the second person, both lexical items in Figure 11.2 are candidates, but
(b) will win the competition over (a) by virtue of the Elsewhere Principle,
as it contains less superfluous material than (a). The same is true in the
third person: Both items are candidates for insertion, but (b) will win as
it is a closer match.
(a) </A/, 1P> (b) </B/, 2P>
1 2P 2 3P
2 3P 3
Figure 11.2 Lexical items in 2/3 syncretism
T h e F e at u r e S t r u c t u r e of P r o n o u n s [ 279 ]
280
11.3 NUMBER
Before going on to a discussion of the actual syncretism data, we need to say

something about the syntactic representation of number. Some languages
form the plural of pronouns with the same morpheme that is used with nouns
(or certain noun classes). This is illustrated for Mandarin Chinese in (1) and
Table 11.2 (Corbett 2000, 76).
Table 11.2 PRONOUN PARADIGM OF

MANDARIN CHINESE
sg pl
1 wǒ wǒ-men
2 nı̌ nı̌-men
3 tā tā-men
(1) a. xuésheng

student
b. xuésheng-men
student-pl
Other languages displaying this pattern include Sierra Popoluca (Elson 1960),
Trumai, Korean, Canela-Kraho, and Miskitu (Corbett 2000; Cysouw 2003; see
Daniel 2013 for additional discussion).
Exploiting the analogy illustrated in Table 11.2 and (1), I conclude that the
plural number sits on top of the person feature complex, as shown in Figure 11.3.
In Mandarin Chinese, a lexical item as in (2) can be assumed, which spells
out the plural morpheme.
(a) NumP (b) (c)
pl 1P NumP
1 2P pl 2P NumP
2 3P 2 3P pl 3P
3 3 3
Figure 11.3 Plural on top of person feature complex

(2) < /men/, [NumP pl] >
Spellout-driven movement ensures that the complement of pl in Figure 11.3 moves

into the Spec of NumP, after which -men spells out NumP. We will have reason to
modify the trees in Figure 11.3 later, but for now they represent a good initial
assumption on the position of the number dimension in the feature hierarchy.
11.4 ATTESTED SYNCRETISMS
11.4.1 Types of patterns
Having outlined some of the background theoretical assumptions of this

chapter, we are now in a position to take a closer look at the types of
syncretisms that we can observe in the personal pronouns. Basically, three
general types of patterns can be distinguished, numbered I–III in Table 11.3.
Table 11.3 VERTICAL , HORIZONTAL , AND NON-

LINEAR SYNCRETISMS IN PRONOUN PARADIGMS
I II III
sg pl sg pl sg pl
1 B C A A A A
2 D A B C B A
3 E A D E C D
In the first pattern, we find vertical, or cross-person, syncretisms. The second

type of pattern instantiates horizontal, or cross-number, syncretism. The third
type of pattern is a nonlinear one (i.e. cross-person and cross-number). All of
these patterns are attested, and I now discuss them in turn, starting out with
the horizontal syncretisms, as they are the simplest ones to derive.
11.4.2 Horizontal syncretisms
The horizontal syncretism can be restricted to a single person, two persons, or

arise in all three persons. Table 11.4 lists the attested patterns.1
1. Small capitals indicate language families rather than individual languages. Quite
a number of languages feature no third person personal pronouns, but replace
them by demonstratives instead. I leave it as an open question whether this state of
affairs is to be considered an instance of a lexical gap, filled by the demonstratives, or
whether we are dealing with a syncretism between third person personal pronouns
and demonstratives. The systematic nature of the filling of the gap by demonstratives
suggests that the latter interpretation is correct.
282
Table 11.4 AT TESTED PAT TERNS OF HORIZONTAL SYNCRETISMS

3P Sinhalese, Sentani, Asmat, salish
2P (rare) English, Xokleng
1P (rare) Marind
2P and 3P Berik, Kuman
1P and 3P (rare) Tairora
all persons Salt-Yui (3P: demonstratives)
Table 11.5 PRONOUN PARADIGM OF BERIK
sg pl
1 ai ne
2 aame aame
3 je je
By way of example, consider the case of Berik (New Guinea) in Table 11.5.

The relevant lexical items are given in Figure 11.4.
The item aame (a) can spell out both 2sg and 2pl, by the Superset Principle:
The tree of the singular pronoun is a subtree of the plural pronoun tree. For
the same reason, (b) can spell out 3P, both singular and plural. A problem
arises, however, when we consider the lexical item for the 1sg pronoun ai.
The problem is that in 2sg, there is a tie between aame of Figure 11.4 and
ai of Figure 11.5: aame and ai each contain exactly one feature more than the
(a) </aame/, NumP > (b) </je/, NumP >
pi 2P pi 3P
2 3P 3
Figure 11.4 Lexical items for aame and je
</ai/, 1P >
1 2P
2 3P
Figure 11.5 Lexical item for ai

(a) </ai/, Num1P > (b) </aame/, Num2P >
Num1 1P Num2 Num1P
1 2P Num1 2P
2 3P 2 3P
3 3
Figure 11.6 Two Num features
syntactic node of a 2sg pronoun. We can solve this problem by assuming that
the number projection is internally complex, that is, singular number also
involves the presence of a number feature (Num1), and plural number involves
two features (Num2 and Num1), as in Figure 11.6.2
Now aame can still spell out 2P, both singular and plural: In the plural it is a
perfect match, and in the singular the lexical tree contains the syntactic tree as a
subtree (highlighted in (b) of Figure 11.6). However, ai can no longer spell out the
2sg, because it does not contain the syntactic tree as a subtree. This solution cru-
cially requires that singular pronouns contain a Num1 feature: The presence of Num1
in (a) of Figure 11.6 prevents the tree from being a match for a 2P syntactic tree.
Informally, we can describe this as the impossibility of a lexical tree to “shrink” in
the middle to match a syntactic tree. Although trees can shrink at the top to match
a syntactic tree, they cannot shrink in the middle. The other attested patterns of
horizontal syncretism work in the same way and can be derived in the same way.
Summarizing, the horizontal syncretisms support the claim that the singular
number is not the absence of number, but the presence of a singular number
feature. The existence of horizontal syncretisms further rests on the possibility
of building trees with an incomplete person fseq, that is, with person features
missing at the top of the person sequence, as well as the possibility of shrinking
the number projection at the top of the tree, as allowed by the Superset Principle.3
2. I leave open the question of the precise nature of these number heads. A. Rocquet (per-
sonal communication, hereafter p.c.) suggests that the first number head (Num1), which
is always present in the pronouns, could be a layer of “definiteness” or a “count” layer, be-
cause personal pronouns usually refer to definite or count entities. The number projection
is certain to be more complex than suggested here, as it does not make provisions for the
well-known property of dual number. Because the analysis of the number projection is not
the topic of this chapter, I leave this as an issue for further research.
3. The metaphor of tree shrinking used here is unrelated to the concept of impov-
erishment in DM. Rather, it refers to the fact that a lexical tree may map onto a syn-
tactic tree that is smaller, that is, the lexical tree may be overspecified in comparison
with the syntactic tree; as such, it is the nanosyntactic counterpart of the Subset
Principle of DM.
284
11.4.3 Vertical syncretisms
Cross-person syncretisms in the singular pronouns are extremely rare. Cysouw

(2003) finds only two languages (out of the 90 for which independent pro-
noun paradigms are listed) showing ABB or AAB: Qawesqar and Winnebago.4
Vertical or cross-person syncretisms do occur in the plural, however. Some
languages that display them are listed in (3):
(3) a. AAB: many Athabascan languages (e.g. Slave, Chiricahua Apache,

Navaho, Kato, Hupa), Awa, Southern Haitian Creole
b. ABB: Nez Perce, Warekena, Wolof (object pronouns), Mauritian Creole5
The account of the AAB and ABB syncretisms is not straightforward. Consider
the AAB pattern in Slave in Table 11.6 (Cysouw 2003, 124).
The lexical tree for the 1pl pronoun is given in Figure 11.7.
Table 11.6 PRONOUN PARADIGM

OF SL AVE
sg pl
1 sį naxį
2 nį naxį
3 ʔedį
ʔegedį
</naxį/, Num2P >
Num2 Num1P
Num1 1P
1 2P
2 3P
Figure 11.7 Lexical item for naxį
4. Concerning Qawasqar, M. Cysouw (p.c.) informs me that work by Oscar Aguilera

reveals that the data in Clairis (1985) are probably not entirely reliable and that
Qawasqar has different forms for 1sg and 2sg and either a null pronoun or a demon-
strative in 3sg. A language not mentioned by Cysouw is Sanapaná (Gomes 2013),
which has an ABB pattern both in the singular and the plural.
5. According to Baker (1972) and Stein (1984), but not Adone (1994), who gives an
ABC pattern in the plural.

This can spell out the syntactic tree of a 1pl pronoun, but not of a 2pl one,
because the latter lacks the 1P node and is therefore not a subtree of Figure
11.7. To derive the AAB pattern, the tree would have to be able to shrink
in the middle (from 1P to 2P). For the same reason, the ABB pattern (not
exemplified here) cannot be derived either, as the lexical item for 2P cannot
shrink to 3P.
A solution to the problem of the existence of both horizontal and ver-
tical syncretisms in multidimensional paradigms has been proposed by
Caha and Pantcheva (2012), which relies on the mechanism of pointers (see
Baunaz and Lander, in Chapter 1 fn. 15). A different solution relies on a re-
formulation of the Superset Principle, which can be applied to the problem
of multidimensional paradigms as well. I now discuss these two solutions
in turn.
11.5 POINTERS
11.5.1 Suppletion
A pointer is a node in the tree of a lexical item that points to another, existing,
lexical item (Starke 2011, 2014). Starke proposes pointers as a mechanism to
account for cases of suppletion. For example, the lexical item of the suppletive
past tense brought contains a pointer (represented by the numerical indices in
(4a)) to the lexical items for bring on the one hand, and the past tense mor-
pheme on the other:
(4) a. <24 /brought/, [XP 22 23] >

b. <22 /bring/, V >
c. <23 /-ed/, [PastP Past] >
Informally, the pointer may be represented in a tree using an arrow instead of

a plain branch (see Figure 11.8).
Each of the lexical items pointed to is subject to independent cyclic spellout.
This creates bring + ed, which is overwritten at the top node by brought, but
only in case the nodes dominated by XP have previously been spelled out by
bring and -ed, respectively, because we do not want brought to spell out the
XP ⇒ brought
bring ⇒ V22 Past P23 ⇒ -ed
Figure 11.8 XP for brought, with pointers to bring and to -ed
286
combination of walk + ed (or any other verb, for that matter). Given the Past–
Perfect–Passive syncretism illustrated in (5), we must conclude that -ed has
more internal structure, so that instead of (4c), we have Figure 11.9.
(5) a. They elected George. (Past)

b. They have elected George. (Perfect)
c. George was elected. (Passive)
The -ed suffix can spell out a past tense, a perfect, or a passive syntactic tree:
This is because the lexical tree can shrink at the top, because of the Superset
Principle. Now the suppletive form brought shows the same Past–Perfect–
Passive syncretism. This means that in the lexical item with the pointers (4a),
one of the items pointed to by means of a pointer, namely lexical item 23
(shown in Figure 11.9), must also be allowed to shrink to any subtree.
That is, the shrinking that can take place only at the top in Figure 11.9
can also take place in the middle of the tree in Figure 11.10. The substructure
dominated by the pointer defines its own superset domain for the purposes of
lexicalization candidacy. This is a natural consequence of the mechanism of cy-
clic spellout: The item pointed to is an independent lexical item and can shrink
at the top, even if it is contained in a larger lexical tree by virtue of the pointer
relationship. As a result, the lexical item brought can spell out three different
syntactic trees, just like the suffix -ed can.
<23/-ed/, PastP >
Past PerfP
Perf PassP
Pass
Figure 11.9 More internal structure for -ed
XP ⇒ brought
bring ⇒ V22 PastP ⇒ -ed
Past PerfP
Perf PassP
Figure 11.10 XP for brought, with pointers to bring and to more complex structure of -ed

11.5.2 Multidimensional paradigms
Let us return now to the vertical syncretisms in the pronoun paradigms in

the plural. As we saw in Section 11.1.2–11.1.4., pronouns spell out multiple
feature dimensions: Case, number, person, and gender. Let us assume that
the lexical items for pronouns can also contain pointers at the juncture of
the dimensions (in the tree in Figure 11.11, KP represents the Case sequence,
NumP the number sequence, and ΠP the person sequence).
KP
K NumP
Num ∏P
Figure 11.11 Pointers at Case, number, and person junctures
This assumption allows the derivation of the problematic vertical

syncretisms, because the tree can now shrink in the middle (from 1P to 2P to
3P). Recall the lexical tree for the Slave pronoun naxi̜, syncretic for 1pl and 2pl
(see Figure 11.7). We now add a pointer to this tree (Figure 11.12).
</naxį/, Num2P >
Num2 Num1P
Num1 1P
1 2P
2 3P
Figure 11.12 Lexical tree for naxį with pointer
This representation allows us to derive the AAB pattern. The lexical item
in Figure 11.12 can spell out a 1pl pronoun, but also a 2pl one, because of the
presence of the pointer. The lexical item for the 3pl pronoun ʔegedi̜ (not illus-
trated here) does not contain the 1P and 2P projection. As a result, it will win
the competition over Figure 11.12 in the 3pl because of the Elsewhere Principle.
In a similar fashion, vertical ABB syncretisms can be derived. Assume a
lexical item like Figure 11.12 but without a pointer. This is the A-pronoun,
which can spell out only 1pl, because it does not contain a pointer. In
addition, assume a B-pronoun like Figure 11.12 (with a pointer) but without
the 1P node. This B-pronoun does not compete with the A-pronoun in the first
288
person, because it lacks the 1P node. The B-pronoun contains a pointer and
can spell out both 2pl and 3pl, winning against the A-pronoun because of the
Elsewhere Principle.
11.5.3 Pointers introduce ABA
Allowing pointers also allows a certain type of ABA pattern in the plural, in
agreement with an abstract prediction made by Taraldsen (2012). Consider
Table 11.7.
Let us see how the derivation of this pattern is allowed by the system
outlined in Section 11.5.2. For A, assume a lexical entry with a pointer, as in
(a) of Figure 11.13, which is flexible at the number–person joint: Because of
the shrinkability of items with pointers, the lexical item for A can spell out all
the plural pronouns. The lexical item for B, given in (b) of Figure 11.13, does
not contain a pointer and is therefore rigid (i.e. not shrinkable in the middle).
If the syntactic tree is 3pl, (a) A is the only candidate, because (b) B cannot
shrink in the middle to spell out 3pl. If the syntactic tree is 2pl, B wins the
competition over A because of the Elsewhere Principle. This is because the
lexical item (b), without the pointer, contains less superfluous material than
Table 11.7 ABA PAT TERN

DERIVABLE WITH POINTERS
sg pl
1 C A
2 D B
3 E A
(a) </A/, Num2P > (b) </B/, Num2P >
Num2 Num1P Num2 Num1P
Num1 1P Num1 2P
1 2P 2 3P
2 3P 3
Figure 11.13 Lexical item for A (with pointer) and B (without pointer)

the lexical item (a), with the pointer. If the syntactic tree is 1pl, B is not a com-
petitor because it lacks a 1P node; A can (and does) spell out the tree. These
findings agree with an abstract prediction made by Taraldsen (2012), who
argues that ABA patterns may arise in multidimensional paradigms, given
Caha and Pantcheva’s (2012) analysis in terms of pointers.
In sum, we have seen that vertical syncretisms in the pronouns can be derived
by means of assuming the presence of pointers in lexical representations. But
assuming pointers opens the door to ABA patterns. Given the central role that
syncretisms play in establishing feature hierarchies, this is a potentially undesir-
able conclusion.6 I therefore now turn to another solution, which involves a revi-
sion of the Superset Principle and does not allow the derivation of ABA patterns.
11.6 REFORMULATING THE SUPERSET PRINCIPLE
In this section, I propose a reformulation of the Superset Principle, which will

allow us to capture both horizontal and vertical syncretisms in multidimen-
sional paradigms, but at the same time does not make ABA derivable. It relies
on a reformulation of the Superset Principle (6), which is inspired by (but not
identical to) a proposal made by Caha (2014):
(6) Revised Superset Principle (RSP)

A lexical entry L may spell out a syntactic node SN iff the features of L
are a superset of the features dominated by SN.
6. It is an open question whether ABA is actually attested in the independent

pronouns. I am aware of two claims in the literature to that effect. Baerman et al.
(2005) quote the case of Dakar Wolof (Nussbaum et al. 1970) as an example of a lan-
guage with a 1–3 syncretism in the independent pronouns. However, this case is moot.
Stewart and Gage (1970, 38) give the paradigm in Table 11.i for the Dakar Wolof inde-
pendent pronouns, which does not show the 1–3 syncretism.
Table 11.i PRONOUN PARA-

DIGM OF DAKAR WOLOF
sg pl
1 man ñum
2 yow yeen
3 moon ñoom
There is, to be fair, a 1–3 syncretism in the plural of the so-called dependent pronouns,
but these behave more like clitics than like independent pronouns, to the point at
which one could wonder if they had not better be analyzed as inflectional markers. The
second case involves Bagirmi, a Nilo-Saharan language spoken in Chad (Gaden 1909),
which Cysouw (2003) claims has an ABA pattern in the plural pronouns. This does,
however, appear to be an isolated case.
290
The most striking difference between the RSP and the classical Superset
Principle is the absence in the RSP of a reference to a subtree of the lexical tree
as a condition on the spellout of syntactic trees. The RSP replaces this require-
ment with a weaker superset condition on terminals or features of the tree.
This reformulation will allow the “shrinking in the middle” that was needed to
derive the vertical syncretisms in pronominal paradigms. Consider the abstract
example in Figure 11.14, where (a) is a lexical item and (b) a syntactic tree.
The RSP makes (a) a possible spellout for the syntactic tree (b), even though
the syntactic tree is missing the C and D nodes that occur in the middle of the
lexical tree. All the RSP requires is that the features of the lexical item (which are
the terminals {B, D, F, G}) be a superset of the features of the syntactic tree (which
are {B, F, G}). This condition is obviously met in Figure 11.14. Put more informally,
the lexical tree can shrink at any point to become identical to the syntactic tree.
(a) </α/, A > (b) A
B C B E
D E F G
F G
Figure 11.14 Lexical item α and a syntactic tree
Table 11.8 PRONOUN
PARADIGM OF SL AVE
sg pl
1 sį naxį
2 nį naxį
3 ʔedį ʔegedį
The RSP derives both horizontal and vertical syncretisms in multidimen-

sional paradigms without the need for pointers and at the same time makes
ABA patterns underivable. To see how this works, reconsider the AAB pattern in
Slave (Table 11.8), with the lexical tree for the 1pl pronoun given in Figure 11.15.
The lexical item in Figure 11.15 will be able to spell out a syntactic tree for
1pl, as both trees are fully identical. What we need to show is that Figure 11.15
can also spell out the syntactic tree for 2pl, given the RSP. The relevant syn-
tactic tree is given in Figure 11.16.
The lexical item in Figure 11.15 can spell out the syntactic tree in Figure
11.16, given that the features of the lexical tree (Num2, Num1, 1, 2, 3) are a su-
perset of those of the syntactic tree (Num2, Num1, 2, 3). Note that the lexical

</naxį/, Num2P >
Num2 Num1P
Num1 1P
1 2P
2 3P
Figure 11.15 Lexical item for naxį
Num2P
Num2 Num1P
Num1 2P
2 3P
Figure 11.16 Syntactic tree for 2pl
tree does not contain the syntactic tree as a subtree: In this example, the node
1P of the lexical tree is not found in the syntactic tree. Although this situation
is ruled out by the classical Superset Principle, it is allowed by the RSP.
The RSP does not allow the derivation of ABA patterns. Recall the ABA
pattern previously discussed (see Table 11.9), and the corresponding lexical
items in Figure 11.17 (but now without a pointer).
Given the RSP, both A and B can shrink in the middle, that is, both can
now spell out 2P and 3P. As a result, both lexical items will compete in 2pl
and 3pl. The Elsewhere Principle now ensures that B will win the competition
both in the 2pl and the 3pl, as B contains fewer superfluous features. As a
result, *ABA holds in full generality, that is, ABA patterns are underivable in
principle.
Table 11.9 ABA PAT TERN
sg pl
1 C A
2 D B
3 E A
292
(a) </A/, Num2P> (b) </B/, Num2P>
Num2 Num1P Num2 Num1P
Num1 1P Num1 2P
1 2P 2 3P
2 3P 3
Figure 11.17 Lexical items for A and B, both without pointers
In sum, both the approach in terms of pointers and the RSP allow the deriva-
tion of multidimensional paradigms. However, the pointers approach opens the
door to the derivation of ABA patterns. The RSP is more restrictive and does not
allow the derivation of ABA patterns. In what follows, I investigate a series of non-
linear syncretisms, showing where both approaches make different predictions.
11.7 NONLINEAR SYNCRETISMS
11.7.1 Shapes and sizes
In this section, I discuss syncretisms that are not exclusively horizontal and
not exclusively vertical either. These come in a variety of shapes and sizes,
which are summed up in (7):
(7) a. L-shaped, contiguous

b. double L, without ABA
c. diagonal (noncontiguous)
d. L-shaped, with ABA
e. double L, with ABA
The interest of the discussion is that the two approaches for dealing with mul-
tidimensional paradigms previously outlined make different predictions as to
the possible existence of these patterns. In what follows, attested patterns are
illustrated by an example and unattested patterns by an abstract representa-
tion of the syncretism type in terms of capital letters.
11.7.2 L-s haped, contiguous
There are basically four different, logically possible L-shapes, only one of which is
attested. This is the type of L that is found in Usarufa (Cysouw 2003) (Table 11.10).

Table 11.10 PRONOUN
PARADIGM OF USARUFA
sg pl
1 ke ke
2 e ke
3 we ye
This syncretism is derivable under the pointers approach: The 1P lexical

item ke contains a pointer, which means that it can spell out all persons and
numbers. However, ke loses the competition to more specific lexical items
without pointers (e, we, ye).
The pattern is also derivable with the RSP: Here too, ke can spell out all
persons and numbers. In 2sg, ke loses the competition to the more specific
lexical item e, which lacks the 1P node. In 2pl, ke is the only candidate, because
e lacks the plural number node, and we and ye lack the 2P node. In 3P, we and
ye win against all the other items because they lack the 2P and 1P nodes and
therefore have fewer superfluous features.
The second type of L is one that is the mirror image of the first one
(Table 11.11).
Table 11.11 MIRROR IMAGE PAT TERN

OF THE USARUFA PARADIGM
sg pl
1 A A
2 A B
3 C D
This pattern is also derivable under the pointers approach: The A item
can spell out all persons and numbers because it can spell out 1pl and be-
cause it contains a pointer. It will lose against the more specific B, C, and D
items where these occur because these do not contain a pointer. In the RSP
approach, however, this pattern is underivable. The A item is maximal and
can spell out all persons and numbers. The B item can spell out anything that
is non-first-person. As a result, the Elsewhere Principle will yield a win of the
B item over the A item in both 2sg and 2pl. This makes the pattern in Table
11.11 effectively underivable.
Both of these patterns have a variant in which the L shape is one row lower
in the paradigm. Our conclusions for Tables 11.10 and 11.11 extend directly to
these variants, so that we do not discuss them in detail here.
Two other types of L-shaped patterns involve a 180° rotation of the two
previous patterns (Table 11.12).
294
Table 11.12 ROTATED L-PAT TERNS
sg pl sg pl
1 A B B A
2 A A A A
3 C D C D
These are both derivable under the pointers approach, for essentially the
same reason as previously given: A can spell out all persons and numbers and
contains a pointer, but loses against B, C, and D where these occur because they
do not contain pointers. Neither pattern is derivable under the RSP approach,
however. The left-hand one has an A item that can spell out all persons and
numbers because it can spell out 1P and plural number. But the B item can do
so, too, as it spells out 1pl. As a result, there will be a tie between A and B in
all cells of the paradigm that are non-third-person. The paradigm on the right
of Table 11.12 is underivable because A cannot win against B in 2sg, as B is
more specific in never occurring in the plural. The same conclusions hold for
the variants of these patterns in which the L is one row lower in the paradigm.
In sum, of the four logically possible L shapes (and their variants), only one
is actually attested. This is also the only one derivable under the RSP, whereas all
four types are derivable under the pointers approach. The RSP approach there-
fore provides a much better fit with the data here than the pointers approach.
11.7.3 Double L, without ABA
This abstract pattern is illustrated in Tables 11.13 and 11.14.
Table 11.13 DOUBLE L PAT TERN
sg pl
1 A A
2 B A
3 B B
Table 11.14 DOUBLE
L PAT TERN, ROTATED
sg pl
1 A A
2 A B
3 B B

Neither of these patterns is derivable under the pointers approach. The
reason is that there are two competing items, which both contain pointers. The
A item is maximal and flexible and can express all persons and numbers. The B
item is more specific because it does not contain a 1P projection, but it is also
flexible and can apply to both 2 and 3 and to both singular and plural. This will
ensure a win of A in the first person, but also a win of B in all the other persons.
In particular, in the second person the B item will win both in the singular and
in the plural, because the A item applies to more cases than the B item.
The same reasoning applies to the approach in terms of the RSP. Because B can
express both sg and pl, it will win against A both in the singular and in the plural.
11.7.4 Diagonal
Diagonal syncretisms contradict spatial accounts of syncretism (e.g. McCreight

and Chvany 1991). As we will see, there are quite a number of logically possible
types of diagonal syncretisms, but only one of these types is attested. This is
the pattern illustrated by the example from Suki in Table 11.15.
Table 11.15 PRONOUN PARADIGM

OF SUKI
sg pl
1 ne e
2 e de
3 u i
This pattern is derivable under the pointers approach. The lexical tree of the
e-pronoun is maximal and flexible, i.e. shrinkable at the joint (from 1P to
2P). As a result, e can express all the persons and numbers, but it loses the
competition against the rigid items for the other persons and numbers be-
cause they contain less superfluous material.
Diagonal syncretisms of this type are underivable with the RSP. Here too,
the lexical item e can spell out all persons and all numbers. In 2sg, there are
two more specific items, however: ne (because it cannot spell out plural) and
de (because it cannot spell out first person). We now have three candidates for
spellout, of which e is the one with most superfluous structure; we therefore
do not predict e in this position in the paradigm.
Now look at another type of diagonal in Table 11.16, which is minimally
different from Table 11.15, but is unattested.
This type of diagonal is also derivable with pointers (for the same reason
as the Suki diagonal). However, it is underivable with the RSP. Both the A and
the B items can spell out all numbers and all persons: B because it spells out
296
Table 11.16 DIAGONAL PAT TERN

MINIMALLY DIFFERENT FROM SUKI
sg pl
1 A B
2 C A
3 D E
the maximal tree of 1pl, and can consequently spell out any smaller tree; like-
wise, A can spell out any person or number, as it can spell out 1P (the maximal
person) as well as plural (the maximal number). As a result, the lexical trees
of A and B are identical. This results in a tie between A and B in 1sg and 1pl,
and in 2pl, as both A and B contain the same amount of superfluous material.
I take this to be an inadmissible situation in the domain of pronouns.
Two other types of diagonal patterns are also unattested (for the sake
of brevity, I show the two distinct syncretism patterns in a single table,
Table 11.17).
Table 11.17 TWO MORE

SIMPLE DIAGONAL PAT TERNS
sg pl sg pl
1 B C B C
2 D A A D
3 A E E A
As before, A contains a pointer and is therefore flexible in the middle, but

it will lose to the nonflexible items D and E exactly where it has to. The RSP
approach does not allow the derivation of these patterns. The two diagonals in
Table 11.17 have basically the same shape as the first two in this section, and
the same reasoning applies.
A combination of both diagonals yields the double diagonal patterns in
Table 11.18, both unattested; the right-hand one is derivable with pointers,
whereas the RSP allows the derivation of neither of these.
Table 11.18 COMBINATIONS
OF DIAGONAL PAT TERNS
sg pl sg pl
1 A B B A
2 C A A C
3 A D D A

Summarizing so far, there are six types of diagonal syncretisms allowed
by the pointers approach, only one of which is actually attested and derivable
with pointers. All types of diagonals are ruled out by the RSP. I come back to
this state of affairs in Section 11.7.7.
We now proceed with a discussion of some further types of diagonals,
which combine the diagonal with an L-shaped syncretism. These syncretisms
are underivable under either approach. An overview of the possible patterns
is given in Table 11.19.
Table 11.19 COMBINATIONS OF DIAGONALS WITH L SHAPED

PAT TERNS
sg pl sg pl sg pl sg pl
1 C A A C B B B B
2 A B B A B A A B
3 B B B B A C C A
These are underivable with pointers because both A and B contain a

pointer. The problem arises in the rows where A competes with B in the same
row: Either B wins the competition both in the singular and the plural (as in
the two patterns on the left), or A wins both in the singular and the plural (as
in the two patterns on the right). The patterns are also underivable with the
RSP, as are all patterns involving diagonals.
There are some further logically possible diagonal syncretisms, which
combine a diagonal syncretism with an ABA pattern, as illustrated in
Table 11.20.
Table 11.20 DIAGONAL
WITH ABA
sg pl
1 C A
2 D B
3 B A
This is not derivable under the pointers approach, again because both
A and B must be assumed to contain pointers, both the vertical and the diag-
onal syncretisms requiring pointers, as previously explained. As a result, B will
win against A in the 3pl. Neither are syncretisms with ABA derivable under
the RSP, for the reasons already explained. There are more logically possible
patterns that combine a diagonal (or a double diagonal as in Table 11.18) with
an ABA (or a double ABA), but because they are all underivable I do not discuss
them any further here.
298
11.7.5 L-s haped, with ABA
An L-shaped syncretism with ABA would be one instantiating the abstract

pattern in Table 11.21.
Table 11.21 L SHAPE
WITH ABA
sg pl
1 A A
2 C B
3 D A
The syncretism is derivable in the pointers approach: The A item is maximal

and flexible, and it loses out to the more specific C, B, and D items where these
occur. The pattern is underivable with the RSP, as are all cases involving ABA,
for the reasons outlined earlier.
11.7.6 Double L, with ABA
One instance of this abstract pattern is illustrated in Table 11.22.

Table 11.22 DOUBLE WITH
ABA
sg pl
1 A A
2 B B
3 B A
This is not derivable under the pointers approach for the same reason as the
other double-L patterns discussed in Section 11.7.3. There are two competing
items, both containing pointers, with the tree for B smaller (and therefore
more specific) than the tree for A. This will ensure a win for A in 1P, as re-
quired. The problematic cell is 3pl, where B wins against A because its tree is
smaller than the tree of A. The pattern is also underivable under the approach
in terms of the RSP, because it involves an ABA pattern in the plural.
11.7.7 Summary of f indings
Table 11.23 presents an overview of the findings in this section. The top line
represents the syncretism patterns by their table number (e.g. 10 refers to

Table 11.23 SUMMARY OF DERIVABLE AND AT TESTED
SYNCRETISM PAT TERNS
10 11 122 13 14 15 16 172 182 194 20 21 22

PTR ✓ ✓ ✓ ✗ ✗ ✓ ✓ ✓ ✓ ✗ ✗ ✓ ✗
RSP ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
ATT ✓ ✗ ✗ ✗ ✗ ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗
Table 11.10); a superscripted number indicates the number of different syncre-

tism patterns the table instantiates, if there is more than one. The second line
(PTR) indicates whether a pattern is derivable with pointers, and the third if
the RSP allows it. The bottom line (ATT) indicates whether a pattern is attested.
What the table shows is that both the approach in terms of pointers and
the one in terms of the RSP face some empirical problems. The RSP approach
is more restrictive in general, but perhaps too strongly so, in that it rules out
the existence of a pattern that is actually attested. This is the diagonal pattern
found in Suki (Table 11.15). According to Cysouw (2003, 211), this pattern is
“commonly found in the contemporary Aztecan languages,” although the evi-
dence he quotes involves subject prefixes rather than independent pronouns.
On the other hand, the pointers approach may be said to be too liberal, in
that it allows far more syncretisms than are attested. This is particularly true
in the diagonal syncretisms, in which only one type is attested, but many
more are allowed by the pointers approach. The situation can be represented
in a subset–superset hierarchy, in which neither the approach in terms of
the RSP nor the one in terms of pointers provides an exact match with the
attested data.
(8) logically possible ⊃ derivable with pointers ⊃ attested ⊃ derivable with RSP
On conceptual grounds, the approach in terms of the RSP seems preferable, as

it rules out all cases of ABA, and it therefore leaves the syncretism diagnostic,
a cornerstone of the nanosyntactic method, fully intact. At the same time,
it remains an empirical question whether in certain cases ABA patterns are
attested. The empirical domain investigated here is that of the independent
pronouns, in which they are indeed vanishingly rare, but in the domain of
verb agreement the existence of 1–3 syncretisms is uncontroversial (see e.g.
Ackema and Neeleman 2013). Note further that even under the pointers
approach, ABA patterns could not arise in the dimension that is structurally
highest. In the nominal domain, this is arguably the Case dimension. Even if a
lexical item could contain a pointer in the Case sequence, this would not lead
to any ABA patterns in the Case dimension. It would only create the possibility
of spelling out sequences that the syntax cannot create, assuming there to be
300
no gaps in the functional sequence for Case. The possibility of ABA is created
only in dimensions that sit lower in the hierarchy than Case, such as number
and person.
11.8. POSSIBLE BUT UNATTESTED SYNCRETISMS
In this section, I want to briefly address the matter of possible but unattested
syncretisms. There is a rather striking such case in the data discussed so far,
which is that of the vertical syncretisms in the singular. The fact that such syn-
cretism is attested in the plural suggests that the theory should allow it, yet it is
virtually unattested in the singular. The question is why this should be so. A pos-
sible answer comes from an observation by Baerman et al. (2005, 44), who ob-
serve that case syncretisms are sensitive to an animacy hierarchy, given in (9).
(9) 1sg > 2sg > 1incl du > 1incl pl > 2du > 2pl > 3 > this > that >
indef > animate > meat, vegetable > other inanimate
What they note is that in nominative and accusative case systems, nom–acc
syncretisms are more likely to occur toward the right-hand side of the hierarchy,
that is, in the less animate nominals. For instance, Indo-European languages
regularly have nom-acc syncretism in nouns but typically lack it in pronouns.
The systematic nom-acc syncretism found in neuter nouns in Indo-European
languages (e.g. Greek and Latin) can also be seen as an instance of this corre-
lation, insofar as these nouns typically have inanimate referents. In languages
with split ergativity, on the other hand, the correlation is opposite, that is, erg –
abs syncretisms are more likely to occur with items higher on the hierarchy.
The hierarchy in (9) in fact conflates a number of different subhierarchies,
seen in (10):
(10) a. Person: 1 > 2 > 3

b. Number: sg > du > pl
c. Animacy: animate > inanimate
d. Definiteness: definite > indefinite
e. Word class: pronoun > demonstrative > noun
Given the number subhierarchy (sg > nonsingular), one expects this correla-
tion to also be found with respect to number, that is,a nonsingular number
should favor nom–acc syncretism, and singular number erg–abs syncretism.
However, this prediction is not borne out. Instead, both patterns of syncre-
tism are more likely in the nonsingular than in the singular. Similarly, gender
syncretism occurs more easily in the nonsingular than in the singular, as in,

for example, Germanic languages, in which gender distinctions are typically
lost in the plural. Obviously, this tendency for syncretism to occur more easily
in the nonsingular than in the singular strikingly resembles the one we ob-
serve in the pronoun paradigms under discussion here. Another reflex of the
correlation between animacy and syncretism can be found in the fact that hor-
izontal (cross-number) syncretism is far more common in the third person
than in the first and second persons (see Table 11.4). These observations do
not in and of themselves constitute an explanation for derivable but unat-
tested (or rarer) syncretisms, but they do point to more general patterns that
can be observed in other domains as well.
11.9 FURTHER CONSEQUENCES
This final section discusses both a problem and a consequence of the RSP. The
problem concerns horizontal syncretisms in the non-first person. The abstract
pattern is given in Table 11.24.
Table 11.24 HORIZONTAL SYNCRE-

TISM IN THE NONFIRST PERSON
sg pl
1 A B
2 C C
3 D E
Here, both the lexical items A and C are candidates for spelling out a 2sg
syntactic tree, but the Elsewhere Principle does not designate either one as
a winner. C can spell out 2sg because it can spell out 2pl, that is, it contains
exactly one more feature than the syntactic tree. The same is true for A, how-
ever: assuming either pointers or the RSP, A can spell out 1sg and therefore
also 2sg; like C, the lexical item for A contains exactly one feature more than
the syntactic tree. Because A and C contain an equal amount of superfluous
material, the Elsewhere Principle cannot determine a winner, an unwelcome
result. This is in fact a problem that we encountered earlier (in connection
with the horizontal syncretism of Berik discussed in Section 11.4.2) and that
we solved by assuming that singular number involved a separate number fea-
ture, rather than the absence of number. This solution worked as long as we
considered only horizontal syncretisms, that is, trees that only could shrink
at the top. But now that we have introduced mechanisms that allow trees to
shrink in the middle as well, the problem reappears. In the pointers approach,
there is an easy way to solve it: The only assumption needed is that A contains
no pointer and therefore is not a candidate for spelling out 2sg. C has to
302
shrink only at the top and will therefore spell out 2sg. In the RSP, however,
an additional assumption needs to be made to ensure that C is the winner.
This follows from the fact that the RSP gives up the subtree requirement that
is present in the classical Superset Principle and looks at only the features.
Because C and A both have exactly one feature more than the syntactic tree,
they are both candidates, and both contain exactly the same amount of super-
fluous material. A solution consists in claiming that the Elsewhere Principle
must be supplemented with the principle in (11).
(11) Prefer spellouts in which the mismatch between the lexical tree and
the syntactic tree is restricted to the top of the tree.
An informal version of this principle could be formulated as “Prefer High

Junk,” or “Minimize Gaps.” This will ensure a win of C against A in 2sg: C
differs from the syntactic tree only at the top, whereas A differs from it in
the middle. In sum, to derive horizontal syncretisms, the RSP needs to be
supplemented with the principle in (11).
The second point to be discussed in this section concerns a consequence
of the RSP for the way the lexicon is organized. As stated earlier, the main
difference between the RSP and the classical Superset Principle is the absence
in the RSP of a reference to a subtree or constituent requirement. Since the
RSP gives up the concept of a subtree, an important piece of motivation is
lost for the assumption that lexical items contain trees at all. The alternative
would be to assume that lexical items simply contain sets of features. To see
what this would look like, reconsider the abstract example in Figure 11.14
discussed in Section 11.6, where the syntactic tree contained the terminals B,
F, G, and the lexical entry an additional terminal D between B and F. An alter-
native lexical entry for (a) in Figure 11.14 would contain a mere unordered set
of terminals (or features), as shown in (12):
(12) < /α/, {B, D, F, G} >
This will be insertable in a syntactic tree with the terminals {B, F, G} by the RSP.
Obviously, some information is lost in a representation like (12) as compared
with the richer kinds of representations with lexical trees. But what gets lost
is largely redundant information. For one thing, we lose the nodes that are
the projections of these heads in the tree; but this is redundant information,
as their presence follows from the general principle that each feature is a syn-
tactic head that projects. For another, we lose the hierarchical order of these
heads, but this is also easily recoverable from the universal fseq. Because it is
not clear at this point whether lexical representations as in (12) are more than
mere notational variants of the lexical representations containing trees, I do
not dwell on the matter any further here.

11.10 CONCLUSION
I have shown that the analysis of syncretism in multidimensional paradigms

requires an extension of classical nanosyntactic theory. I have discussed two
such extensions, one in terms of pointers, and another in terms of a RSP.
Both approaches make different empirical predictions, notably with respect
to the possible existence of ABA patterns and of certain types of diagonal
syncretisms. Empirically, neither approach provides a perfect fit with the avail-
able data. At the conceptual level, the RSP, though ruling out any kind of ABA
in principle, requires an additional principle over and above the Elsewhere
Principle to derive horizontal syncretisms in non-first-persons.
REFERENCES
Aalberse, Suzanne. 2007. “The Typology of Syncretisms and the Status of Feature
Structure. Verbal Paradigms Across 355 Dutch Dialects.” Morphology 17:
pp. 109–149.
Aalberse, Suzanne and Jan Don. 2009. “Syncretism in Dutch Dialects.” Morphology 19:
pp. 3–14.
Ackema, Peter and Ad Neeleman. 2013. “Person Features and Syncretism. Natural
Language and Linguistic Theory 31: pp. 901–950.
Adone, Dany. 1994. The Acquisition of Mauritian Creole. Amsterdam and Philadelphia:
John Benjamins.
Baerman, Matthew, Dunstan Brown, and Greville Corbett. 2005. The Syntax-Morphology
Interface: A Study of Syncretism. Cambridge: Cambridge University Press.
Baerman, Matthew and Dunstan Brown. 2013. “Syncretism in Verbal Person/Number
Marking.” In The World Atlas of Language Structures Online, edited by Matthew S.
Dryer and Martin Haspelmath. Leipzig: Max Planck Institute for Evolutionary
Anthropology. [online] Available at http://wals.info/chapter/29.
Baker, Philipp. 1972. Kreol. London: C. Hirst.
Béjar, Susana. 2003. Phi- Syntax: A Theory of Agreement. Doctoral dissertation,
University of Toronto.
Caha, Pavel. 2014. “Funky Genitives.” Class lectures, Generative Linguistics in the Old
World Spring School, Brussels.
Caha, Pavel and Marina Pantcheva. 2012. “Contiguity Beyond Linearity.” Talk at
Decennium: The First 10 Years of the Center for Advanced Study in Theoretical
Linguistics.
Clairis, Christos. 1985. El Qawesqar: Lingüística Fueguina, Teoría y Descripción, Vol. 12
Estudios Filológicos, Anejo. Valdivia: Universidad Austral de Chile.
Corbett, Greville. 2000. Number. Cambridge: Cambridge University Press.
Cysouw, Michael. 2003. The Paradigmatic Structure of Person Marking. Oxford: Oxford
University Press.
Daniel, Michael. 2013. “Plurality in Independent Personal Pronouns.” In The World
Atlas of Language Structures Online, edited by Matthew S. Dryer and Martin
Haspelmath. Leipzig, Germany: Max Planck Institute for Evolutionary
Anthropology. [online] Available at http://wals.info/chapter/35.
304
Elson, Ben. 1960. “Sierra Popoluca Morphology.” International Journal of American

Linguistics 26: pp. 206–223.
Gaden, Henri. 1909. Essai de grammaire de la langue baguirmienne. Paris: Leroux.
Gomes, Antonio Almir Silva. 2013. Sanapaná, Uma Língua Maskoy: Aspectos Gramaticais.
Doctoral dissertation, Universidade Estadual de Campinas.
McCreight, Katherine and Catherine Chvany. 1991. “Geometric Representation of
Paradigms in a Modular Theory of Grammar.” In Paradigms: The Economy of
Inflection, edited by Frans Plank, pp. 91–112. Berlin: Mouton de Gruyter.
Nussbaum, Loren, William Gage, and Daniel Varre. 1970. Dakar Wolof: A Basic Course.
Washington, DC: Center for Applied Linguistics.
Starke, Michal 2011. Class Notes, Course on Nanosyntax given at Ghent University.
November 7–10.
Lecture series presented at Center for Research in Syntax, Semantics and,
Phonology, Hogeschool-Universiteit Brussel. March 18, 20, 22.
Starke, Michal. 2014. “Cleaning up the Lexicon.” Linguistic Analysis 39: pp. 245–256.
Stein, Peter. 1984. Kreolisch und Französisch. Tübingen, Germany: Niemeyer.
Stewart, William and William Gage. 1970. “Notes on Wolof Grammar.” In Dakar
Wolof: A Basic Course, edited by Loren Nussbaum, William Gage, and Daniel
Varre, pp. 355–412. Washington, DC: Center for Applied Linguistics.
Taraldsen, Tarald. 2012. “*ABA and the Representation of Features in Syntax.” Talk
presented at Brussels Conference on Generative Linguistics 7, Brussels.

CHAPTER 12
Functional Sequence Zones

and Slavic L>T>N Participles
LUCIE TAR ALDSEN MEDOVÁ
AND BARTOSZ WIL AND
12.1 INTRODUCTION
The fact that adjectival passive L-participles are formed only from unaccusative
verbs constitutes a long-standing puzzle in the syntax of Czech and Polish. We
provide a structural solution to this puzzle based on the idea of morphemes
understood as individual zones of functional sequence in syntax, or fseq zones
for short. Under this view, morphemes that form the same fseq zone com-
pete for insertion with each other, whereas if two morphemes co-occur, we
take it as a hallmark that they form two different fseq zones in a syntactic
representation.1
We illustrate this approach to morphemes on the basis of the participle
zone that is projected on top of verb stems in Slavic languages and argue that
this zone spells out as L, T, or N, depending on its size and internal constit-
uent structure.
1. A reviewer ask whether co-occurrence of two morphemes can be a case of a single

fseq zone that happens to be lexically split up into two or more morphemes. The cru-
cial point is that the morphemes have to compete for insertion—and that, in turn,
they can do only if they are part of the same fseq. In other words, if a particular se-
quence can be lexicalized by two morphemes, it means that there are two syntactic fseq
zones. For example, if Tense is lexicalized as a fusional morpheme with phi-agreement
features in a language, then temporal and phi-features form a singleton fseq zone in
that language.
306
Consider (1), which is the format of the Slavic verb:
(1) (prefix) -root -theme -participle -agr

a. u - děl - a - l - a active: L-participle
pref - do - aj - L - f.sg
‘(she) did’ (Cz)
b. u - děl - á - n - o passive: T-/N-participle
pref - do - AJ - N - n.sg
‘(it was) done’ (Cz)
There are seven themes in both Polish and Czech: Ø, E, EJ, NU, AJ, OVA, and
I. Together with the root they merge with, they encode the verbal argument
structure. For example, the theme E builds stative stems [as in the Polish ex-
ample in (2)], the theme I builds causative stems [as in (3)], and NU builds
semelfactives [as in Czech (4a)], and degree achievements [as in (4b)].
(2) Stative E
My nie chcemy o tym słysz-e-ć. (Pol)
we not want about it hear-E-inf
‘We don’t want to hear about it.’
(3) Causative I
Jan po-sadz-i-ł-Ø dziecko na stole. (Pol)
Jan pref-sit-I-L -3.m.sg child on table
‘Jan made the child sit on the table.’
(4) a. Semelfactive NU
Petr kop-nu-l-Ø psa. (Cz)
Petr kick-NU-L -m.sg dog.acc
‘Petr kicked the dog (once).’
b. Degree achievement NU
Petr hloup-nu-l-Ø.
Petr stupid-NU-L -m.sg
‘Petr was getting more and more stupid.’
Despite the fact that L-[see (1a)] and T-/N-suffixes [see (1b)] form different types
of participles, we discuss evidence indicating that the actual morpheme lexicalizes
parts of a single fseq zone because all three compete for insertion in the syntactic
structure of participles. In particular, the superset of heads that make up this
zone spell out as the active L-participle [as in (5)]. The subsets that spell out as
the adjectival L-passive as in (6) and T-or N-passives as in (7) and (8) lexicalize
projections that make up eventive (verbal) and adjectival (stative) passives.

(5) Active (nonpresent) L-participle
a. Wczoraj kop-a-ł-em piłkę z kolegami. (Pol)
yesterday kick-AJ-L -1.msc.sg ball with friends
‘Yesterday I played some soccer with friends.’
b. Karel hod-i-l-
Ø boty do kouta. (Cz)
Karel throw-I-L -m.sg shoes into corner
‘Karel threw the shoes into the corner.’
(6) L-passive
a. Król jest zmar-ł-y i nic tego nie zmieni. (Pol)
king.nom is die-L-msc.nom and nothing this not change
‘The king is dead and nothing will change that.’
b. Ta treska je z-mrz-l-á na kost. (Cz)
this codfish.f.sg is pref-freeze-L -f.sg to bone
‘The codfish is frozen solid.’
(7) T-passive
a. Piłka zosta-ł-a kop-nię-t-a. (Pol)
ball.f.sg became-L -f.sg kick-NU-T-f.sg
‘The ball was kicked.’
b. Karel by-l kop-nu-t-Ø do břicha (Petrem). (Cz)
Karel be-L kick-NU-T-msc.sg in stomach Peter.ins
‘Karel was kicked in the stomach (by Peter).’
(8) N-passive
a. Ta dziura zosta-ł-a wczoraj wy-kop-a-n-a. (Pol)
this hole.f.sg become-L -f.sg yesteday pref-dig-AJ-N-f.sg
‘This hole was dug out yesterday.’
b. Boty by-l-y hoze-n-y do kouta (Petrem). (Cz)
shoes be-L -msc.pl throw-N-msc.pl into corner Petr.ins
‘The shoes were thrown into the corner (by Petr).’
We focus particularly on the long-standing puzzle in Polish and Czech mor-

phosyntax (see Cetnarowska 2002 for Polish), namely the fact that only stems
of unaccusative verbs build adjectival L-participles (whereas unergatives and
transitives can only form N-or T-participles). For instance, unaccusative verbs
like vlhnout ‘get wet’ (Cz) or blednąć ‘become pale’ (Pol) will form L-participles
z-vlh-l-ý ‘wet’ or po-blad-ł-y ‘pale,’ whereas unergative verbs like dupnout
‘stamp’ (Cz) or ziewnąć ‘yawn (once)’ (Pol) will not (*dup-l-y, *ziew-ł-y). We
argue that this results from the fact that the lexical entry for L includes case
peels left by the movement of the unaccusative NP argument of the verb stem.
F u n c t i o n a l S e q u e n c e Z o n e s a n d S l av i c L > T > N Pa r t i c i p l e s [ 307 ]

308
We also argue that, interestingly, case peeling also takes place in the deriva-
tion of unergative stems, but in these, the accusative case peels are spelled out
as part of the theme vowel, that is, a morpheme lexicalizes the fseq zone lower
than the participle fseq.
12.2 PASSIVE PARTICIPLES
12.2.1 Verbal and adjectival passives
Passive participles can be verbal (eventive) or adjectival (stative). In German,

these two kinds of passives are morphosyntactically distinguished, with verbal
passives occurring with werden ‘get/become’ and adjectival passives with sein
‘be,’ as in (9) and (10), respectively (see Kratzer 2000; Maienborn 2007).2
(9) Die Reifen werden auf-gepumpt.

the tires get/become up-pumped
‘The tires are being inflated.’
(10) Die Reifen sind auf-gepumpt.

the tires are up-pumped
‘The tires are inflated.’
Kratzer (2000) further distinguishes between two types of adjectival

passives: Target States, which can be modified by a temporal adverbial immer
noch in German or ‘still’ in English, as in (11), and Resultant States, which re-
sist such a modification in German as well as in English, as in (12).
(11) Das Gebäude ist (immer noch) geräumt.

the building is still evacuated
‘The building is (still) evacuated.’
(12) Das Theorem ist (*immer noch) bewiesen.

the theorem is still proven
‘The theorem (*still) is proven.’
Although the modification with ‘still’ remains a reliable diagnostic help to dif-
ferentiate between Resultant State and Target State passives, some languages
distinguish them morphologically as well. In Swedish, for instance, Target
2. The labels ‘target state passive’ and ‘resultant state passive’ used by Kratzer (2000)
are derived from Parsons’s (1990) distinction between target state and resultant state
perfects.

State passives can come out as underived adjectives (adj), as in the following
examples from Lundquist (2008, 148).
(13) a. Dörr-en är fortfarande öppen/??öppnad.

door-def is still open.adj/??open.de
‘The door is still open.’
b. en redan öppnad/??öppen dörr
an already open.de/??open.adj door
‘an already opened/??open door’
(14) a. Dörr-en är fortfarande stängd.

door-def is still close.de
‘The door is still closed.’
b. en redan stängd dörr
an already close.de door
‘an already closed door’
The examples in (13a) and (14a) include Target State participles öppen and
stängd, which is exhibited by ‘still’ modification. The forms in (13b) and (14b)
are Resultant State participles: The derived (de) Resultant State form öppnad
is incompatible with ‘still’ and, conversely, the underived Target State öppen is
incompatible in a resultative context in (13b).3
Several other diagnostics reported in the literature on the participles
have been proposed, including modification with by-phrases.4 Because verbal
passives denote events initiated by agents, they can be modified by agentive
3. In (14), there is no morphological distinction between the Target State and
Resultant State: The example provides a purely contextual (with ‘still’-modification)
example for the validity of the Target State vs. Resultant State distinction in Swedish
on top of the morphological distinction given in (13).
4. Among other well-known modification tests that distinguish between eventive
and stative passives is the compatibility with degree-modifying adverbials, which is
impossible with eventive passives, as in (i), but possible with statives, as in (ii).
(i) a. ?*The cart is completely pushed.

b. ?*The bottle is half emptied.
(ii) a. The equipment is completely damaged.
b. The window is half opened.
There is a considerable body of work on both semantic and morphosyntactic
contrasts between both verbal and adjectival as well as Target and Resultants State
passives, which includes modification by für-PPs, availability of reflexive readings
(Kratzer 2000; Alexiadou and Anagnostopoulou 2008), the presence of transitivizing
morphology in German (Alexiadou et al. 2014), or the contrast in eventive vs. stative
readings in Russian short and long forms of participles as in pokrašen vs. pokrašennyj
‘painted’ (Borik 2013).

310
by-phrases as in (15a). Such a modification is ill-formed in the case of adjec-

tival passives, as in (15b),5 as states do not have agentive implication.
(15) a. The door was (recently) opened (by John).

b. The door is (recently) opened (*by John).
12.2.2 The English -e d suff ix
English verbal and adjectival passives are all formed with the -ed suffix (and
its allomorphs) on the root.6 In other words, the -ed morpheme is not only an
exponent of Past Tense [as in (16a)], but also of eventive verbal passives [in
(16b)], Resultant State passives [in (16c)], and Target State passives [in (16d)]:
(16) a. The gardener mow-ed the grass at 3 p.m. yesterday.

b. The grass {gets, is being} mow-ed by the gardener.
c. The mailbox is (*still) empti-ed.
d. The building is (still) evacuat-ed.
In Starke’s (2006) analysis of participles, this fact indicates that the -ed mor-
pheme lexicalizes an fseq of projections that grammatically encode passive
and Past Tense morphology. Using “the more you do the bigger you are”
logic, he argues for the following hierarchy of projections that make up the
participle zone.
The fact that the Past Tense participles and the Eventive Passive, Resultant,
and Target States are all spelled out as the -ed morpheme follows from the
Superset Principle, the major tenet of nanosyntax, which regulates the inser-
tion of the lexical–phonological material into syntactic nodes.7
5. Our judgments for (15b) are from British English; a reviewer reports that in her or
his variety of English the form open is better in this example. More work, obviously,
needs to be done on the acceptability of such forms and the source of the variation.
6. Suffice it to say, the formation of different types of passives may require different
categories of roots. We see this, for instance, in (16c) and (16d), where the Resultant
State passive is based on an underived adjective empty and the Target State participle
is based on an underived verb evacuate.
7. The Superset Principle and Match have been successfully applied to the domain
of case (Caha 2009), directional adpositions (Pantcheva 2011), Bantu class-markers
(Taraldsen 2010b; Taraldsen in Chapter 3), Slavic prefixes (Wiland 2012), Czech
numerals (Caha 2013), as well as to an extended theory of feature lexicalization in
paradigms (Taraldsen 2012), among others. See also Baunaz and Lander (Chapter 1,
Section 1.3.3.2) for an illustration of how the structural subset–superset relation
works in the lexicalization patterns of strong, weak, and clitic pronouns in French in
Cardinaletti and Starke’s (1999) system of the tripartition of pronouns.

(17) The Superset Principle (Starke 2009)
A phonological exponent is inserted into a syntactic node if its
lexical entry has a (sub)constituent which matches that node,
where matching is defined as follows:
(18) Match (Caha 2009, 67)
A lexical constituent matches a node in the syntactic representation
if it is identical to that node (ignoring traces).
The Superset Principle determines that a phonological exponent of a

Vocabulary Item can in principle lexicalize different syntactic representations.
The representations that match the lexical entry in Figure 12.1 are the
following constituents, where (19a) is the superset and (19b)–(19d) are the
proper subsets of (19a).
PastTP
Fn+3 Eventive Passive
-ed ⇔ Fn+2 Resultant State
Fn+1 Target State

Fn ...
Figure 12.1 Starke’s hierarchy for the participle zone
(19) a. -ed ⇔ [ PastTP [ Eventive Passive [ Resultant State [Target State]]]]

b. -ed ⇔ [ Eventive Passive [ Resultant State [ Target State]]]
c. -ed ⇔ [ Resultant State [ Target State ]]
d. -ed ⇔ [ Target State ]
One more remark about the shape of a lexical entry and the Superset Principle
is in order, in the context of the participle fseq zone. Namely, if more than
one Vocabulary Item observes the conditions on lexicalization, it is the item
that contains fewer features unspecified in the representation that wins the
competition for the lexical insertion. This basic principle in nanosyntax (and
quite an intuitive one, given the fact that smaller representations are built
before the bigger ones) is often informally referred to as the Biggest Wins
Theorem.8
8. See also an illustration of how this theorem works on the example of French
pronouns le, il, and lui in Baunaz and Lander (Chapter 1, Section 1.3.3.2), under the
name of the Principle of Cyclic Override.

312
Although we do not observe how lexical overriding works in the parti-

ciple fseq in English as the entire zone in Figure 12.1 (i.e. from the bottom
Target State layer up to the top Past Tense layer) is lexically specified as -ed,
in Polish and Czech the participle fseq zone is lexicalized by three different
exponents: L, T, and N, as listed in the examples in (5)–(7).
12.3 PARTICIPLE FSEQ IN POLISH AND CZECH

12.3.1 Fseq zones in Slavic
Given the format of a Slavic verb in (1) and the bottom-to-top derivation, L, T,

and N spell out the participle fseq zone that is projected on top of a separate zone,
namely the one that spells out as theme vowels. As briefly illustrated in (2)–(4) and
discussed at length elsewhere (see Jabłońska 2007 for Polish, in particular), the-
matic morphemes spell out the fseq, that is—when combined with a particular cat-
egory of the root—responsible for the argument structure properties of the stem
including the case of its NP argument. (Later in the chapter, we refer to the bare
verb stem, that is, a constituent made of a root and a theme vowel, simply as VP.)
The theme vowel zone is projected on top of a root zone, and the positioning
of the three fseq zones that make up the Slavic verb stem (excluding the agree-
ment morphology) is given in Figure 12.2.
Participle zone ⇔ {L, T, N}
F2
F1 theme zone ⇔ {∅, E, EJ, NU, AJ, OVA, I}
F2
F1 root zone ⇔ {adj, noun, verb}
F2 F1
Figure 12.2 Fseq zones in the Slavic verb
Although it is descriptively clear that roots and themes form two separate
morphemes in the structure of the verb stem in Polish and Czech (as well as in
all Slavic languages), it is less clear whether roots are always morphologically
simplex. There are at least two ways in which Polish and Czech roots look to
be structurally complex.
12.3.1.1 Excursus on the root zone
First, we argue in Taraldsen Medová and Wiland (2015) that what has been
traditionally referred to as a theme vowel NU [as in the forms in (4)] is made of

GIVE
N3 GET
N2 HAVE
N1
Figure 12.3 Hierarchy for light verbs
two separate morphemes N and U. According to that analysis, N has the struc-
ture as in Figure 12.3 and spells out the light verb GIVE or GET, depending
on the amount of syntactic structure it lexicalizes. (In Figure 12.3, the N1–N3
labels refer to the morpheme -N-of the N + U sequence.)
Namely, N spells out the light verb GET in degree achievements, which
explains their GET-readings, as for instance in (4b), where the infinitive form
hloup-n-ou-t literally means ‘to get stupid,’ or in (20):
(20) Petr slep-n-u-l-Ø. (Cz)

Petr blind-N-U-L -msc.sg
‘Petr was getting blind.’
In semelfactives, N spells out the light verb GIVE, which is bigger than the
light GET present in degree achievement stems and defines their GIVE
readings, as in (4a) and repeated as (21), where the infinitive kop-n-ou-t liter-
ally means ‘give a kick’.
(21) Petr kop-n-u-l-Ø psa. (Cz)

Petr kick-N-U-L -msc.sg dog.acc
‘Petr kicked the dog (once).’
The separate morpheme U, which in Czech surfaces either as /-u-/or as /-

ou-/because of a structurally defined phonological lengthening, is an actual
thematic morpheme and contributes to the argument structure properties of
the stem.
Second, ongoing work on lexical categories in nanosyntax put forward by
Starke (2009) and Lundquist (2008) tends to indicate that lexical categories are
not primitive but structurally complex. According to this view, sometimes re-
ferred to as the containment theory of lexical categories, adjectives are smaller
than nouns that are, in turn, smaller than verbs, as shown in Figure 12.4, slightly
simplified.
Because in Czech and Polish semelfactives are based on nominal roots, the
root zone of semelfactive stems is made of a nominal root and the light N
suffix with the GIVE reading as in Figure 12.5. In contrast, Czech and Polish

314
Verb
Cat3 Noun
Cat2 Adjective
...
Figure 12.4 Containment theory of lexical categories
rootN-NGIVE
Noun ⇒ kop- GIVE ⇒ -n-
Cat2 Adjective N3 GET
... N2 HAVE
N1 t
Figure 12.5 Root zone of a semelfactive stem kop-N- ‘give a kick’
rootA-NGET
Adjective ⇒ slep- GET ⇒ -n-
... N2 HAVE
N1 t
Figure 12.6 Root zone of a degree achievement stem slep-N- ‘get blind’
degree achievements are based on adjectival roots; hence their root zones in-
clude an adjectival root and the light N suffix with the GET reading, as in
Figure 12.6.
Although no ingredient of our analysis in this chapter relies on the contain-
ment theory of lexical categories, for the reasons just listed we simply refer
to a morphological root as an entire separate fseq zone in the representation
in Figure 12.2. Essentially, however, the entire root zone as in Figure 12.5 or
Figure 12.6 is lower than the theme vowel zone and the participle zone.
12.3.2 T-/N -p assives
The fact that the participle fseq zone appears to be lexicalized by a singleton
exponent in English and by three exponents—L , T, and N—in Polish and

Czech, is by no means an unusual situation (for instance, the case fseq is
lexicalized by a number of exponents depending on a declension class within
Slavic languages). However, we argue that a participle fseq zone in Polish and
Czech includes a subconstituent made up of case peels; these are then a part
of a lexical entry.
Although in both Polish and Czech T and N morphemes form eventive and
stative passives [see (7) and (8)], Czech distinguishes between them morpho-
syntactically. Eventive passives in Czech have short-form (SF) morphology and
can be modified by an agentive by-phrase, which in Czech involves an agent
marked with an instrumental case, as in (22). In turn, adjectival passives in
Czech have long-form (LF) morphology and resist the agentive modification,
as in (23).9
(22) Ten članek je přelože-n-Ø do italštiny (Karlem). (Cz)

this article.msc.sg is translate-N-msc.sg.sf into Italian Karel.inst
‘This article is translated into Italian by Karel.’
(23) Ten članek je přelože-n-ej do italštiny (*Karlem). (Cz)

this article.msc.sg is translate-N-msc.sg.lf into Italian Karel.inst
‘This article is translated into Italian (*by Karel).’
According to Medová and Taraldsen (2007), Czech additionally distinguishes

between Resultant State and Target State passives in that only the former
takes a locative PP, instead of an expected directional PP. Certain Czech verbs
of induced motion require a directional PP, as in (24), which is a property of
periphrastic verbal passives, as in (25).
(24) Jan hodil boty [do kouta]dir / *[v koutĕ].loc (Cz)

Jan.nom threw.3.sg boots.acc into corner.gen / in corner.loc
‘Jan threw boots into the corner.’
9. Polish also exhibits sensitivity to agentive by-phrase modification with eventive

and stative passives, as in (iii) and (iv), respectively. Polish differs from Czech in that
the agentive phrase is closer to its English equivalent in that it includes a preposition
przez ‘by’ and an accusative agent.
(iii) Piłka zostala kopnię-t-a (przez bramkarza).

ball.nom bacame kick-T-f.sg by goalkeeper.acc
‘The ball was kicked by the goalkeeper.’
(iv) Maria jest przebra-n-a za klauna (*przez swoją mamę).
Maria.nom is dress-N-f.sg for clown-gen by her mom.acc
‘Mary is dressed up as a clown (*by her mom).’
Polish does not distinguish morphologically between eventive and stative participles
as Czech does.

316
(25) Boty byly hoze-n-y [do kouta]dir /*[v koutĕ].loc (Cz)

boots.acc were thrown-N-sf.pl into corner.gen / in corner.loc
‘The boots were thrown into the corner.’
An adjectival passive, which stands out as the one that uses the long form of
the participle in (26), however, takes the locative instead of the directional PP:
(26) Boty byly hoze-n-ý [v koutĕ]loc / ??[do kouta]dir (Cz)

boots-.nom.pl were.3.pl throw-N-lf.pl in corner.loc / into corner.gen
‘The boots were thrown in the corner.’
Using diagnostics involving ‘still’ modification, durative adverbials, and HAVE

passives that distinguish between Resultant and Target States, Medová and
Taraldsen (2007) argue that only adjectival passives with Target State reading
will take locative PPs in contexts like (26).
For the present purposes, we omit any further discussion of the lowest part
of the participle fseq that builds T-and N-passives and simply assume that a
lower part of the participle fseq zone that spells out verbal and (both kinds of)
adjectival passives is lexicalized by T or N, depending on its internal make-up.
For this reason, we continue to refer to this lower part of the participle fseq
simply as the passive T–N area. Instead, we focus on the higher part of the par-
ticiple fseq outlined in Figure 12.1 in the remainder of the chapter.
12.3.3 L-s yncretism and the unaccusativity puzzle
Although all verb stems produce active nonpresent Tense L-participles, verb
stems with different argument structure properties produce different types of
T-and N-passives. For instance, transitive and accusative semelfactive stems
formed with the NU theme produce T-passives in both Polish and Czech, as in
(27), and transitive and accusative causative stems formed with the I theme
produce N-passives, as in (28).10,11
10. There exists a degree of variation in the formation of T-and N-passives between

Czech and Polish in that certain closely related stems that form T-passives in Czech, as,
for instance, hřát ‘warm up’—vy-hřá-t-ý ‘warmed up,’ will form N-passives in Polish, as
grzać ‘warm up’—wy-grza-n-y ‘warmed up.’ Also, we have found at least one reflexive
verb in Polish, namely bać się ‘to be afraid + SE-reflexive,’ that does not produce any
passive participle *wy-ba-n-y, *wy-ba-t-y, whereas it does in Czech, as in bát se –?vy-bá-
t-ý. We do not explore the nature of this variation in this chapter. What is essential,
however, is that there is no variation between Polish and Czech with respect to forma-
tion of active L-participles only by unaccusative verb stems.
11. The theme vowel I disappears in the participial forms both in Polish and Czech;
compare the (a) and (b) and (c) and (d) examples in (28). We leave this variation aside,
but see Medová (2012).

(27) a. Maria ścis-nę-ł-a cytrynę. (Pol)
Maria.nom squeeze-NU-LTense-f.sg lemon.acc
‘Maria squeezed a lemon.’
b. Cytryna zosta-ł-a ściś-nię-t-a.

lemon.f.sg.nom become-LTense-f.sg squeeze-NU-T-f.sg
‘The lemon got squeezed.’
c. Marie si skříp-(nu)-l-a12 prst do dveří. (Cz)

Marie.f.sg SELF.dat squeeze-(NU)-LTense-f.sg finger.msc.sg in door
‘Marie pinched her finger in the door.’
d. Prst byl skříp-nu-t-ý ve dveřích.

finger.msc.sg.nom be.LTense.msc.sg squeeze-NU-T-msc.sg in door
‘The finger got pinched in the door.’
(28) a. Mama kro-i-ł-a warzywa. (Pol)

mom cut-I-LTense-f.sg
vegetables.acc
‘Mom was chopping the vegetables.’
b. Warzywa są już po-kro-jo-n-e.

vegetables are already pref-cut-I-N-n.pl
‘The vegetables are already chopped.’
c. Trenérka od-stran-i-l-a překážky. (Cz)

coach.f.sg pref-side-I-LTense-f.sg hurdle.acc.pl
‘The coach put aside the hurdles.’
d. Překážky už jsou od-stran-ĕ-n-é.

hurdles already are pref-side-I-N-pl
‘The hurdles are already put aside.’
By contrast, unaccusative verb stems produce adjectival L-passives (and do

not produce either T-or N-passives). The fact that only unaccusatives, but
not unergatives, can form adjectival L-participles has been used as a diag-
nostic to distinguish between these two verb classes in Polish and Czech
(e.g. Cetnarowska 2000; Medová 2012). This contrast is illustrated by the
formation of the adjectival L-participle by the stem of an unaccusative verb
umrze-ć ‘to die’ [used either prenominally as a participial resultative adjec-
tive in (29b) or with a copular być ‘be’ in (29c)] and the lack of such forms
12. Notice that we leave completely aside the fact that the N + U sequence is optional
in some participial forms for some verbs; as far as we know, there is no satisfying ex-
planation for this optionality in the literature.

318
based on an unergative stem of semelfactive wark-ną-ć ‘to growl (once)’ in

(30b) and (30c).13
(29) a. Król zmar-ł-Ø. (Pol)

king.nom die-LTense-msc.sg
‘The king died.’
b. zmar-ł-y król
die-L -msc.nom king
‘dead king /the king is dead’
c. Król jest zmar-ł-y.
king.nom is die-L -msc.nom
‘The king is dead.’
(30) a. Pies wark-ną-ł-Ø. (Pol)

dog.nom growl-NU-LTense-msc.sg
‘The dog growled.’
b. *wark-ł-y pies
*growl-L -msc.nom dog.nom
‘*growled dog /*The dog is growled.’
c. *Pies jest wark-ł-y.
dog.nom is growl-L -msc.nom
‘*The dog is growled.’
We thus have a picture in which adjectival L-participles are syncretic with

active L-participles rather than T-or N-passives but, unlike the latter, are
produced only by unaccusative verb stems.
We can explain this puzzle if we assume Caha’s (2009) theory of case,
whereby arguments of unaccusative verbs are selected as NPs with layered case
projections on top and move to a higher subject position by upward movement
that peels off the higher case layers. Given the case fseq as in Figure 12.7,
Caha (2009) argues that case peels stranded by the NP movement will be
spelled out as part of a different lexical item. Figure 12.8 shows the way this
sequence is lexicalized in the paradigm of the Polish singular masculine noun
pan ‘man,’ which has syncretic exponents for accusative–nominative and
dative–locative.
13. Four theme vowels build unaccusative stems: nonspecific with respect to a situ-

ation type (and unproductive in present day Czech or Polish) Ø, the stative E, degree
achievement EJ, and degree achievement NU (as opposed to the semelfactive NU; see
Taraldsen Medová and Wiland 2015 for details). In (29) and in some later examples, we
use the Polish unaccusative verb umrze-ć ‘to die,’ which is based on a stem with the Ø
theme, for ease of exposition.

INSTP
K6 DATP
K5 LOCP
K4 GENP
K3 ACCP
K3 NOMP
K1 NP
...
Figure 12.7 Caha’s case fseq
NP InstP ⇒ em
pan- K6
... DatP ⇒ u
K5
... LocP
K4
... GenP ⇒ a
K3
... AccP
K2
... NomP ⇒ ∅
K1 t
Figure 12.8 NP movement
The NP movement in Figure 12.8 is motivated by the shape of the lex-

ical entries in the sense that the placement of the NP on top of NomP, AccP,
GenP, and so forth, allows the spellout of its sister node as a constituent. The
movement of the NP in the paradigm of the Polish pan ‘man’ spells out the
following lexical entries (31):
(31) Lexical entries for cases in the paradigm on pan ‘man’ (msc.sg)
a. /Ø/ ⇔ [ K1 ]
b. /a/ ⇔ [ K3 [ K2 [ K1 ]]]
c. /u/ ⇔ [ K5 [ K4 [ K3 [ K2 [ K1 ]]]]]
d. /em/ ⇔ [ K6 [ K5 [ K4 [ K3 [ K2 [ K1 ]]]]]]

320
We argue in Section 12.4 that it is precisely the accusative case peels left by the
movement of the NP argument of unaccusatives that are a part of the lexical
entry of the adjectival L-participle.
12.4 CASE PEELS INSIDE A PASSIVE PARTICIPLE
Consider the step-by-step derivation of L-passive participle based on the

unaccusative verb stem, as in zmarły król ‘dead king’ of (29b).
Under the case-peeling approach, the verb stem made up of the root zone,
which includes a verbal root and a prefix z-, and the Ø theme, as outlined in
Figure 12.2, select an accusative NP. In Figure 12.9 we label the node of the
merger of the accusative NP and the verb stem (indicated as VP) simply as
KaseP, without determining which of the sister nodes projects the head, as its
label is largely irrelevant to the present discussion.
With the root and the theme vowel zone lexicalized, the layers of the higher
participle fseq zone are merged. As discussed earlier, the lower layers of the
participle fseq build the T-/N-passives, as seen in Figure 12.10.
At this point, neither the participle fseq nor the case fseq can be lexicalized,
as none of them form a constituent. To facilitate spellout, the KaseP
KaseP
AccP VP(verb stem)
K2 NomP
root zone theme vowel
K1 NP Ø
z-mar-
król
Figure 12.9 Merger of verb stem and accusative NP
F2Ppass⇔T/N
F2 F1Ppass
F1 KaseP
AccP VP(verb stem)

K2 NomP z-mar-
K1 NP
król
Figure 12.10 Layers of participle zone are merged

F3P
KaseP F2Ppass⇔T/N
F2 F1Ppass
AccP VP
F1 t
K2 NomP z-mar-
K1 NP
król
Figure 12.11 Spellout-driven movement of KaseP
constituent raises to the top of the tree as in Figure 12.11; see Caha (2011) and
Baunaz and Lander (Chapter 1, Section 1.3.3.4) on spellout-driven movement
in nanosyntax. The projection of the fseq derived by spellout-driven movement
is labeled simply as F3P, the next layer of structure in the hierarchy.
Although the NP (except the case fseq, which does not form a constituent at
this point), the VP, and the F2P subconstituents of Figure 12.11 can be spelled
out at this point as they are identified by the lexical entries, the entire structure
cannot. Given the hierarchy in Figure 12.1, the Polish and Czech exponent L is
bigger than T and N because it spells out the highest active L-participles. Because
L also spells out smaller L-passives in unaccusatives, its lexical entry must in-
clude a lower layer of structure—but exclude the verb stem and the nominative-
marked NP. Thus L-passives have the following shape of lexical entry (32):
(32) Lpass /ł /⇔ [[ K2 ][ F2 [ F1 ]]]
To match the lexical entry, two evacuating movements must take place: the
movement of the VP (as in Figure 12.12) followed by the movement of NomP
containing the NP, which peels the AccP layer (as in Figure 12.13).
F4P
VP F3P
z-mar-
KaseP F2Ppass⇔T/N
AccP t F2 F1Ppass
F1 t
K2 NomP
K1 NP
król
Figure 12.12 Evacuation of VP

322
NomP F4P
K1 NP
VP F3Ppass⇔L
król
z-mar-
KaseP F2Ppass⇔T/N
AccP tvp F2 F1Ppass
K2 t
F1 t
Figure 12.13 Evacuation of NomP
The extraction of NomP derives a constituent that matches the lexical

entry in (32), which overrides the insertion of T and N.14
Although the lexical entry for adjectival L-passives as in (32) coupled with
the theory of a participle fseq zone whereby L spells out bigger constituents
than T and N already at this point explains the dependency between L-passives
and unaccusative verb stems, there is still one more morpheme in the struc-
ture of these participles, namely the adjectival agreement.
Essentially, the agreement morpheme merged with the participle con-
stituent shows the adjectival declension pattern and includes case as well as
number and gender features. For the case at hand, that is, z-mar-ł-y król, the -y
is an adjectival agreement marker of m.sg.15
As a result of the subsequent spellout-driven movement of F4P to the
top of the tree, the agreement subtree surfaces as a suffix on the participle
zone. Notice that the Agr -y will come out as a suffix no matter if the phi-
features form a sequence of individual heads or a subconstituent triangle
as indicated now in Figure 12.14 and Figure 12.15. In this way, the deri-
vation in Figure 12.15 looks similar to the proposal in Leu (2015), where
the participle moves as a remnant constituent in front of the agreement
suffix.
The movement of the constituent comprising the verb stem (VP) and
the subset of the participle zone F3P > . . . > F1P (without yet a higher layer
of structure building active nonpresent L-participles, as predicted by the
hierarchy in Figure 12.1) on top of the phrase marker also explains why
14. Note that once extracted, the NomP ends up as a nominative suffix on the NP,
assuming Caha’s (2009) approach to case derivation.
15. Active nonpresent L-participles show the verbal nominal inflection pattern. This
agreement morpheme, we take it, sits higher than the adjectival agreement discussed
in the text—in accordance with the general picture in Figure 12.4.

AgrP
y
NomP F4P
K1 NP
król VP F3Ppass⇔L
z-mar-
KaseP F2Ppass⇔T/N
AccP tvp F2 F1Ppass
K2 t
F1 t
Figure 12.14 Merger of AgrP
F4P
AgrP
y NomP t
VP F3Ppass⇔L K1 NP
z-mar-
KaseP król
F2Ppass⇔T/N
AccP tvp F2 F1Ppass

K2 t F1 t
Figure 12.15 Movement of F4P to the left of Agr
even nominative NPs are obligatorily placed after participles [as long
as there is no copular BE present in the sequence, as already indicated
in (29)].16
(33) a. zmar-ł-y król

dead-L -nom.3.sg king.nom
‘dead king’
b. *król zmar-ł-y
king.nom dead-L -nom.3sg.msc
16. Note that Polish allows for both prenominal and postnominal placements of
adjectives; hence the L-participle > NPnom order does not simply follow from a general
constraint on adjective placement.

324
12.5 WHY UNERGATIVES DO NOT BUILD L-P ASSIVES
We have advanced in Taraldsen Medová and Wiland (2015) that both

unaccusative verbs and unergative verbs select accusative objects, that is,
NPs with projected AccP > NomP sequences on top, which subsequently raise
by case-peeling movement.17 The difference between Slavic unaccusatives
and unergatives, as we argue there, is that unaccusatives are syntactically
smaller than unergatives and merge the accusative NP on top of the theme
vowel zone (thus, on top of the entire verb stem, as in the derivation just
outlined). In contrast, the accusative NP is merged as a layer of structure in-
side the theme vowel zone that builds unergatives. Although the arguments
of both unaccusative and unergative verb stems end up as nominative-marked
subjects as a result of case-peeling movement that strands the AccP layer, the
stranded AccP peel is spelled out as part of the unergative semelfactive theme
vowel NU or activity AJ. Because the accusative peel is part of a lexical entry
of unergative stems, such stems do not produce L-passives, which include the
accusative peel as part of their own lexical entry, as in (32).
Consider the derivation of an unergative stem syknout ‘hiss’ as in the Czech
sentence in (34):
(34) Karel syk-nu-l-Ø. (Cz)

Karel.nom hiss-NU-L -msc.sg
‘Karel hissed.’
One remark about the thematic suffix NU is in order before we illustrate the
spellout of the accusative peels as part of unergative stems. As already pointed
out in Section 12.3.1.1, we have argued extensively in Taraldsen Medová and
Wiland (2015) that the sequence NU, which builds both semelfactives [as in (4a)]
and degree achievements [as in (4b)], is made of two separate morphemes: N
and U. The N morpheme spells out the light verb GIVE in semelfactives or
smaller GET in degree achievements. The lexical entry of the light N morpheme
is given in Figure 12.3 (repeated in Figure 12.16 for convenience).
Following the established assumption about the hierarchical unergative >
transitive–accusative > unaccusative argument structure, verb stems that spell
out as unergatives form a bigger fseq zone than verb stems that spell out as
unaccusatives. Thus the size of a bigger unergative stem includes the structure
of an unaccusative stem, which comprises the AccP argument as in Figure 12.9.
17. Such an analysis of Czech and Polish unergatives is in line with Taraldsen (2010a),
who shows that Norwegian unergative participles have agentive GET-passive readings.
This leads Taraldsen to conclude that arguments of unergative verbs are introduced
VP-internally (in particular, this VP-internal projection is identified as ProcessP in
Ramchand’s 2008 framework of an articulated eventive verbal structure).

The root zone of an unergative semelfactive like that in (34) includes a nom-
inal root and the light N with the GIVE reading as we saw in Figure 12.6. The
root zone merges with an unergative theme vowel U, which is bigger than (any)
theme vowel that builds unaccusatives, as outlined in Figure 12.17. Notice that
Fn indicates the heads in the theme vowel zone; we have used a similar notation
for embedding in the participle zone above.
If an accusative argument (= AccP, i.e. Karel in Figure 12.17) is part of an
unergative theme vowel, then to spell out the theme vowel zone to the exclu-
sion of the NP argument, the latter must be evacuated. The nominative NP
raising takes place exactly as in the derivation of the unaccusatives in Figure
12.13, in that the movement of the NP strands the AccP peel and surfaces
as a nominative-marked subject, as in Figure 12.18. The difference is that in
unergative stems, the AccP peel is spelled out as a part of the theme vowel, the
U-theme, in the case at hand.
Note that the U-theme present in the light N + U sequence builds not only
unergatives but also unaccusatives. In the latter case, as we argue in Taraldsen
Medová and Wiland (2015), U is structurally smaller in that it spells out only
a subset of projections of the unergative U to the exclusion of the AccP layer.
This makes the correct prediction about the fact that not only unaccusative
Ø-stems as in (29b) but essentially also degree achievement unaccusative NU-
stems build adjectival L-passives, as we see in (35b).
GIVE
N3 GET
N2 HAVE
N1
Figure 12.16 Hierarchy for light verbs
rootN - NGIVE F3P
Noun ⇒ syk- GIVE ⇒ -n- F3
Cat2 Adjective N3 GET AccP F2P
N2 HAVE K2 NomP F2 F1P
N1 t K1 NP F1 t
Karel
Figure 12.17 First steps in derivation of unergative

326
NomP
K1 NP
rootN - NGIVE F3P ⇒ -u-
F3
Noun ⇒ syk- GIVE ⇒ -n-
AccP F2P
Cat2 Adjective N3 GET K2 t F2 F1P
... N2 HAVE F1 troot–Ngive
N1 tNoun
Figure 12.18 Evacuation of NomP
(35) a. Moje koty z wiekiem będą głuch-ną-ć. (Pol)

my cats.nom with age be/will deaf-NU-inf
‘My cats will be getting deaf with age.’
b. Moje koty będą ogłuch-ł-e.
my cats.nom be/will dead-Lpass-agr
‘My cats will be deaf.’
12.6 CONCLUDING SUMMARY
We have argued for the approach to morphemes as fseq zones in nanosyntax

in that each zone occupies its own position with respect to other zones and the
boundary of one zone marks the beginning of another, higher zone. We have
discussed two scenarios for the lexicalization of participle zones in Polish and
Czech. The first one involved a case in which a lexical entry of the participle
zone includes accusative case peel left by the movement of the nominative NP
argument of the verb stem, which is lexicalized as a separate morphological
constituent. The second involved a situation in which the accusative case peels
become spelled out as part of the verb stem, a constituent lower than the par-
ticiple zone. This ultimately accounts for the contrast between unaccusative
and other types of stems, in that only the first can build L-passives that in-
clude the accusative case peel left as part of their lexical entry.
REFERENCES
Alexiadou, Artemis and Elena Anagnostopoulou. 2008. “Structuring Participles.” In

Proceedings of WCCFL 26, edited by Charles B. Chang and Hannah J. Haynie, pp.
33–41. Somerville, MA: Cascadilla Press.

Alexiadou, Artemis, Berit Gehrke, and Florian Schäfer. 2014. “The Argument Structure
of Adjectival Participles Revisited.” Lingua 149: pp. 118–138.
Borik, Olga. 2013. “Past Participle Formation and the Eventive/Adjectival Passive in
Russian.” In Sinn und Bedeutung 17 Proceedings, edited by Emmanuel Chemla,
Victor Homer, and Gregoire Winterstein, pp. 115–132. Paris: ENS.
Caha, Pavel. 2011. “The Parameters of Case Marking and Spell Out Driven Movement.”
Linguistic Variation Yearbook 10, edited by Jeroen van Craenenbroeck, pp. 32–77.
Amsterdam and Philadelphia: John Benjamins.
Caha, Pavel. 2013. “Czech Numerals and No Bundling.” Ms. CASTL/University of
Tromsø.
Cardinaletti, Anna and Michal Starke. 1999. “The Typology of Structural Deficiency.
A Case Study of the Three Classes of Pronouns.” In Clitics in the Languages of
Europe, edited by Henk van Riemsdijk, pp. 145–233. Berlin: Mouton de Gruyter.
Cetnarowska, Bożena. 2000. “The Unaccusative/Unergative Split and the Derivation
of Resultative Adjectives in Polish.” In Formal Approaches to Slavic Linguistics 8,
edited by Tracy H. King and Irina Sekerina, pp. 78–96. Ann Arbor: Michigan
Slavic Publications.
Cetnarowska, Bożena. 2002. “Unaccusativity Mismatches and Unaccusativity
Diagnostics From Derivational Morphology.” In Many Morphologies, edited by
Paul Boucher, pp. 48–81. Somerville, MA: Cascadilla Press.
Jabłońska, Patrycja. 2007. Radical Decomposition and Argument Structure. Doctoral dis-
sertation, University of Tromsø.
Kratzer, Angelika. 2000. “Building Statives.” Ms., University of Massachusetts
Amherst.
Leu, Thomas. 2015. The Architecture of Determiners. Oxford: Oxford University Press.
Maienborn, Claudia. 2007. “Das Zustandspassiv: Grammatische Einordnung—
Bildungsbeschränkung—Interpretationsspielraum.” Zeitschrift für germanistische
Linguistik 35: pp. 83–144.
Medová, Lucie. 2012. “Anticausatives Are Derived Unergatives.” In Slavic Languages in
Formal Grammar, edited by Markéta Ziková and Mojmír Dočekal, pp. 291–306.
Frankfurt am Main: Peter Lang.
Medová, Lucie and Tarald Taraldsen. 2007. “1, 2, se.” In Czech in Generative Grammar,
edited by Mojmír Dočekal, Markéta Ziková, and Jana Zmrzlíková, pp. 119–138.
Munich: LINCOM Academic Publishers
Parsons, Terence. 1990. Events in the Semantics of English: A Study in Subatomic
Semantics. Cambridge, MA: MIT Press.
Ramchand, Gillian. 2008. Verb Meaning and the Lexicon; A First Phase Syntax.
Starke, Michal. 2006. “The Nanosyntax of Participles.” Lectures at the 13th Eastern
Generative Grammar summer school, Olomouc.
Nordlyd 36: pp. 1–6. [online] Available at http://septentrio.uit.no/index.php/
nordlyd/index.
Taraldsen, Knut Tarald. 2010a. “Unintentionally Out of Control.” In Argument Structure
and Syntactic Relations: A Cross-linguistic Perspective, edited by Maia Duguine,

328
Susana Huidobro, and Nerea Madariaga, pp. 283– 302. Amsterdam: John
Benjamins.
Taraldsen, Tarald. 2010b. “The Nanosyntax of Nguni Noun Class Prefixes and
Concords.” Lingua 120: pp. 1522–1548.
Taraldsen, Tarald. 2012. “Modeling the Neighborhood Hypothesis for Syncretisms.”
Handout for a talk given at the 43rd annual meeting of the North East Linguistic
Society (NELS), CUNY, October 2012.
Taraldsen Medová, Lucie and Bartosz Wiland. 2015. “Semelfactives are Bigger Than
Degree Achievements.” Ms., Adam Mickiewicz University in Poznań.
Wiland, Bartosz. 2012. “Prefix Stacking, Syncretism, and the Syntactic Hierarchy.” In
Slavic Languages in Formal Grammar, edited by Mojmír Dočekal and Markéta
Ziková, pp. 307–324. Berlin and Frankfurt: Peter Lang.

GLOSSARY
*ABA generalization: A descriptive generalization stating that syncretism is cons

trained, in that it targets only adjacent cells in a paradigm.
Allomorphy: A phenomenon whereby a single morpheme displays alternating forms
that are predictable strictly in terms of phonological rules.
Best Fit Principle, see Elsewhere Principle.
Biggest Wins Theorem, see Cyclic Override.
Containment: A phenomenon in which a smaller structure is nested within a larger
structure.
Cyclic Override: Also called the Principle of Cyclic Override, a theorem stating that
earlier cycles of lexicalization are overridden by later cycles.
Elsewhere Principle: A principle governing lexicalization that guarantees that an
S-tree will be spelled out by the L-tree with the fewest unmatched features.
Functional sequence (fseq): A universal hierarchy of atomic features merged as heads.
Lexical entry: An element of the lexicon, which is composed of (i) a phonological rep
resentation, (ii) a syntactic structure (i.e. an L-tree), and (iii) conceptual information.
Lexicalization: Also called matching or spellout, the process of linking a syntactic struc
ture (S-tree) with a certain lexical structure (L-tree) and its associated phonological
exponent.
Lexicon: A postsyntactic module where lexical entries are stored.
L-tree or Lexical tree: The syntactic structure stored in the lexical entry.
Matching, see lexicalization.
Minimize Junk, see Elsewhere Principle.
Morpheme: A sound–meaning pairing, the smallest such unit available in a language.
A lexical entry represents minimally a single morpheme.
OFOH: Acronym for “one feature– one head,” a maxim according to which each
semantic feature corresponds to a head (with its own position) in the functional
sequence.
Paradigm: A list or table of related forms organized according to a common set of
grammatical features.
Phonological exponent: The phonological representation stored in a lexical entry.
Phrasal spellout: An implementation of spellout that targets nonterminal
(i.e. branching) nodes.
Pointer: A node in an L-tree referring or “pointing” to another lexical entry. Pointers
can be used to account for irregular forms (suppletion).
Portmanteau morphology: When two or more distinct grammatical features or
heads are spelled out as a single form; as this kind of form encodes multiple featural
330
distinctions but is not overtly segmentable, portmanteau morphology is often a rele

vant concept in accounts of suppletion.
SMS: An acronym for the Syntax–Morphology–Semantics module, the generative com
ponent of the language faculty responsible for merging grammatical features according
to the functional sequence.
S-tree or Syntactic tree: The syntactic structure generated by SMS, which is lexicalized
by becoming matched with an L-tree.
Span: Any contiguous region of (two or more) adjacently merged heads in a functional
sequence.
Spellout, see lexicalization.
Superset Theorem: Informally called the Superset Principle, a theorem of lexicalization
according to which an L-tree can spell out an S-tree if the S-tree matches a subtree of
the L-tree.
Suppletion: A morphological phenomenon whereby part of a paradigm displays irreg
ularity, i.e. a phonologically unpredictable form (typically a root or stem).
Syncretism: The phenomenon whereby multiple grammatical distinctions in a para-
digm are lexicalized by a single phonological exponent.
Terminal: A nonbranching node, i.e. a syntactic head.
[ 330 ] Glossary
INDEX
Note: Tables and figures are indicated by t and f following the page number.
AAA pattern, 279 Moroccan, 194–96, 195t

AAB pattern, 279, 284, 287, 290 morphological containment
AbAB pattern, 72–74, 79, 82, 83 and, 24, 25
*ABA generalization, viii, ix, x, 22–23, Arapesh, 125, 144
22n12, 31–32, 31f Arawakan languages, 127
complementizers and, 154 argument-changing prefixes, 223–26
defined, 329 Aristotelian views, 182, 186, 191, 192
formulation of, 22 Armenian. See Classical Armenian
negative markers and, 194, 195, 196 Asmat, 282t
pronouns and, xii, 278, 279, 288–89, Athabascan languages, 284
289n6, 290–91, 298, 299–300, 303 attested syncretism, 23, 133t,
spatial deixis and, 133 281–85, 299t
zero morphology and, 75–81, 76t Australian languages, 130, 131, 138
ABB pattern, 279, 284, 285, 287 Austro-Asiatic languages, 125, 130
aboutness-shift topics, 262, 263, 268 Austronesian languages, 120, 122, 125,
adjectival passives, 308–10, 316 127, 135, 140
Afro-Asiatic languages, 120n4, 124, 131, 138 Awa, 284
Alamblak, 131 axial part (AxPart), 39, 40f
Algic languages, 124 Aztecan language, 299
allomorphy, 58, 74, 77, 80
defined, 329 Bagirmi, 289n6
negative markers and, 192 Bantu. See Southern Bantu languages
principles of, 13–14 BaseP, 170, 170f, 171–75, 171f, 175
Altaic languages, 131 Basque, 121, 126, 128, 144
Amharic, 63 Bemba, 122
anaphoric demonstratives, 116, 117f Berik, 281–82, 282t, 301
Anchor Condition, 30n14, 260, 270 Best Fit Principle. See Elsewhere
anchors (in spatial deixis), 118, 118n3. Principle
See also Hearer; Speaker; Biggest Wins Theorem. See Cyclic
Third Person Override Principle
animacy, 300–301 Binukid, 122, 144
A-pronouns, 287–88, 292f Borer-Chomsky Conjecture, 7
Apurinã, 127, 144 Boumaa Fijian, 135, 137, 140, 141
Arabic B-pronouns, 287–88, 292f
Gulf, 131, 138, 144 broad-strokes model
Modern Standard, 194–96, 195t of grammar, 6, 6f
332
Bulgarian prefixed OpP and, 170–71, 171f, 177

complementizers of, x, 149–55, 150t, Relativized Minimality and, x, 153–55,
153t, 154t, 156, 156t, 158–59, 160, 161, 169, 171, 177
159t, 160t, 161, 161t, 163, 164, semi-factive, 151, 152, 153t, 155, 156,
170, 177 156t, 157
spatial deixis in, 127, 144 specific complementizers, 160–69,
by-phrases, 309, 310, 315, 315n9 171–73, 174–75, 177
strong vs. weak presupposition,
Canela-Kraho, 280 155–56, 158
Cariban languages, 129 syncretism and, 151, 153–55, 158–59
cartography, vii, viii, 3 veridicality and (see veridical islands;
emergence of, 4 veridicality)
goal of approach, 5 complex left branches, xi, 242, 245,
a map of syntactic 248, 249
configurations, 4–10 conceptual idioms, 34–35, 35f
case peels, xii, 307, 308, 315, 318, conceptual semantics, 5
320–23, 326 conceptual structure, 14, 26
Case systems, 21–22, 23–25, 23f, 25f constituent lexicalization, ix, 88–107
Catalan, 126, 127, 144 conceptual issues, 90–91
cause-become-state features, 244, 244f defined, 88
Chapacura-Wanham languages, 124 empirical arguments for, 93–94
Chimakuan languages, 122 portmanteau prefixes and (see
Chinese portmanteau prefixes)
Mandarin, 130, 263, 280 spanning vs., 89–90, 107
negative markers in, 194–96, 195t containment, defined, 329. See also
spatial deixis in, 130, 144 morphological containment
Chiricahua Apache, 284 containment problem, 92
Choco languages, 129 containment theory of lexical categories,
chunk lexicalization, 90 313–14, 314f
Classical Armenian, 75–80, 75t, 83 Contradiction, Law of. See Law of
Classneg. See negative classifier markers Contradiction
clausal spine, 199–200, 199f contradictory negative markers, 182,
Clitic Left Dislocations, 264 183, 185, 187, 189–90, 191,
clitic (object) pronouns, 28f, 28t 193n10, 194
Cyclic Override Principle and, 32–33 contrary negative markers, 182, 183,
in Spanish word order, 269–70 187, 193, 194
Superset Principle and, 27–28 contrastive topics, 262, 263, 264, 266,
Colonnata Italian, 105, 106 267, 267f, 268
complementizer-final languages, 243 coordinated growth problem, 246–49
complementizer-initial languages, 243 Creole
complementizers, 149–77 English, 131
BaseP and, 170, 170f, 171–75, 171f, 175 Mauritian, 284
desiderative, 152, 153t, 156t, 158, 159, Southern Haitian, 284
175, 176 crosslinguistic attestation, 118–22
factive (see factive complements) Cyclic Exhaustive Lexicalization, 37
features of, 169–77 Cyclic movement, 37, 38f, 40f, 41f, 42f,
functional sequence of, 153, 43f, 44f, 46, 49, 245
168–69, 175 Cyclic Override Principle (Biggest Wins
internal structure of, 175–77, 176f Theorem), viii, 32–33, 36, 49, 311
nonfactive, 149, 150t, 151, 153t, 156t defined, 329
[ 332 ] Index
explained, 32 double L syncretism, without ABA,
idioms and, 35 292, 294–95
Czech, 247 doubling prefixes, 222, 223–26
functional sequence zones Dutch, 9, 47–48, 47f, 194–96, 195t
and, 312–20 Dyribal, 122n6
negative markers in, 194–96, 195t
participles of, xii, 305, 306, 307, Elsewhere Clause, 64, 65, 82
312–20, 321, 324n17, 324, 326 Elsewhere Principle, 30–32, 33, 36, 40, 49
spatial deixis in, 131 defined, 329
explained, 30
dative plural, 77, 77f, 78, 79f Korean question particles and, 114–15
decomposed path, 207, 208–10, 226–30 pronouns and, 30–31, 279, 287, 288,
decomposed prefixes, 226–30 291, 293, 301, 303
decomposing complementizers. See Encyclopedia, 13
complementizers endophoric demonstratives, 117f
decomposing scale, 230–32 English, 4, 13–14, 48, 48t, 243,
desideratives, 152, 153t, 156t, 158, 159, 245, 246–47
175, 176 morphological containment and, 24, 25
diagonal (noncontiguous) syncretism, negative markers in, 184, 185–86,
xii, 292, 295–97 189–91, 194, 195t, 197–98
Dima, 256 participles of, 308, 310–12, 314
discourse-deictic demonstratives, 116, 117f pronouns of, 282t
Distal, ix–x, 118, 122, 141–43 spatial deixis in, 116–17, 122, 129, 144
Medial contained by, 139–40, 140f syncretism and, 21
Medial contained by, contains Vocabulary Insertion and, 72–74, 81, 83
Proximal, 140–41, 141f word order in, 258
≠ Medial ≠ Proximal, 123–26, 126t English Creole, 131
Proximal contained by, 137–39, 137f Epena Pedee, 129, 144
Proximal contained by Medial Eskimo-Aleut languages, 121
and, 136–37 European Spanish, xi–xii, 250, 251, 251f,
Distal–Medial–Proximal syncretism, 254–55, 256, 257, 258–59, 260,
131–32, 133t 268, 268f, 270t, 273
Distal–Medial syncretism, 128–31 Evenki, 131, 144
distance-oriented spatial deixis, 118, 119 Eventive Passive, 310, 311
Distributed Morphology (DM), vii, 3–4, Ewondo, 136–37, 136t, 137t, 140–41, 144
88, 134 Excluded Middle, Law of. See Law of the
basic architecture, 10–12 Excluded Middle
model of grammar according to, 11f Exhaustive Lexicalization Principle, 82
theory and terminology in, 10–15 exophoric demonstratives, 116–17, 117f
three differences between nanosyntax extralinguistic semantics, 5
and, 57–59
Vocabulary Insertion in, viii, 58, factive complements, 149, 150t, 151, 152,
64–82, 85 (see also Subset 153t, 155, 156, 156t, 157, 158
principle; zero morphology) factive islands. See veridical islands
D-linked phrases, 162, 163–64, 165 familiarity topics, 262, 264, 268
DM. See Distributed Morphology Fijian, 127, 144
Dominican Republic Spanish, xi–xii, 250, Boumaa, 135, 137, 140, 141
251, 251f, 256, 268, 268f, 271, Wailevu, 136, 137, 141
270t, 273 fine-grained syntactic structures, ix, 3, 4,
double L syncretism, with ABA, 292, 298 8–9, 15, 20
Index [ 333 ]
334
Finnish, 48, 48t Germanic languages, 258

phrasal spellout and, 16–17, 18, 18f Gimira, 272, 272t
spatial deixis in, 121 Goal, 20, 21, 26, 39, 40, 41
First Phrase Syntax framework, 210 Russian verbal prefixes, 205–6, 206f,
Fission, viii, 12, 79, 80, 252 208–9, 209f, 209t, 212–13, 215,
Focneg. See negative focus markers 217, 218, 220, 220t, 221, 222,
FocP, 198f, 200 226–30, 233, 233f
Foot Condition, 91, 101 spelling out, 42f
French, 4, 15, 62–63, 243 Gooniyandi, 130, 144
complementizers of, x, 149–55, 150t, grammatical semantics, 5
153t, 154t, 156t, 157–58, 159, Greek, 300. See also Modern Greek
159t, 160t, 161, 161t, 162n5, 163, GROUND element, 223, 226, 233
164, 170, 171, 175, 177 Gulf Arabic, 131, 138, 144
negative markers in, 185n3, 194–96, Gungbe, 8
195t, 197–98
pronouns of, 27–29, 28t, 30–31, 32–33 Hawaiian, 121
spanning and, 92–93 head movement, 42, 45, 45f, 45n23, 46,
spatial deixis in, 131–32, 144 58, 60–61
syncretism and, 21, 26 Head Movement Constraint, 45n23
fseq. See functional sequence heads. See functional heads;
function (negative markers), 181, 184, submorphemic heads
186, 189, 192, 193, 194, 195t Hearer (H), 117, 118, 120–21, 122
functional heads, 7–9, 240, 243 homophony, 22n11
functional sequence (fseq), 10, 18, horizontal syncretism, 277, 281–83, 285,
46–49, 50, 244, 245 290, 301, 303
of complementizers, 153, Huallaga Quechua, 118n3
168–69, 175 Hungarian, 4, 158, 194–96, 195t
defined, 329 Hupa, 284
mapping: from linear to hierarchical
order, 20–25 idioms, 33–36
of negative markers, 199 Imai languages, 127
principles of, 7–9 Indo-European languages, 16, 124, 126,
of pronouns, 283 127, 129, 132, 138, 300
of spatial deixis, 141–43, 143f instrumental plural, 77, 78, 78f
functional sequence (fseq) zones, xii, interrogatives
305, 310, 312–20, 312f, 321, 322, partial, 253, 253f
324, 326 Spanish, xi–xii, 250, 251, 252, 253,
Fusion, viii, 12, 17, 18, 58, 79, 80, 81, 254, 255, 256, 258–59, 260, 263,
85, 88–89 264, 266, 268
defined, 88 total, 253, 253f
Korean negation and, 67–68, wh- (see wh-interrogatives)
68f, 69, 74 interveners, problem with, 71
Inuktitut, 121–22, 122n6
Generative Semantics, 20n10 Iraqw, 124, 145
German Italian, 8, 93, 261
complementizers of, 153n2, 171 Colonnata, 105, 106
morphological containment and, 24, 25 Northern dialects, 9
participles of, 308 Standard, 104–5
spatial deixis in, 131
spellout-driven movement, 42–46 Japanese, 121, 122, 125, 145
[ 334 ] Index
Karata, 38–41 syntactic trees stored in
Kera, 120n4 (see lexical trees)
Ket, 126, 145 verbs and, 244
Khasi, 125, 145 Vocabulary Item vs., 13
Kikuyu, 121–22 lexicalization, 12. See also constituent
Kiswahili, 124, 145 lexicalization; Match; spellout
Klallam, 129, 145 chunk, 90
Koasati, 120n4 Cyclic Exhaustive, 37
Korean defined, 329
negation in, 66–74, 83–84 lexical prefixes, 207, 209t, 210–11,
pronouns of, 280 213, 220t
question particles, ix, 111–15 lexical storage, 47–49
spatial deixis in, 121, 125, 145 lexical trees (L-trees), 19–20, 47, 48
Koyra Chiini, 131 complementizers and, 173, 174f, 175
Kuman, 282t defined, 329
Kwakw’ala, 123, 145 interaction with syntactic trees, 26–33
negative markers and, 197
Latin, 300 pronouns and, 283, 284, 286, 287,
phrasal spellout and, 16, 17, 18, 290–91, 302
18f, 61 Russian verbal prefixes
spatial deixis in, 124, 128n11, 145 and, 208, 223
Vocabulary Insertion and, 65–66, shrinkage of, 283, 283n3, 286, 287,
65t, 84t 288, 290, 301
Law of Contradiction (LC), 182, 190 lexicon, viii, 6, 7, 15, 17, 18, 19, 26, 27,
Law of the Excluded Middle (LEM), 29, 30, 36, 37, 49, 89, 104
182, 190 Cyclic Override principle and, 33
Level I morphemes, 192 defined, 329
Level II morphemes, 192 Distributed Morphology and, 10, 57
Lexical categories, xii, 12 Fusion and, 68
containment theory of lexical idioms and, 34
categories, 313–14, 314f negative markers and, 197, 198
lexical entries/items, viii, 14, 15, 19, 27, Russian verbal prefixes and, 208
30, 36, 37, 70, 83, 84, 89, 101 Vocabulary Item vs., 13
for case, 319 light verbs, 313, 313f, 325f
“deficient,” 35n16 Limbu, 130, 139, 145
defined, 329 Lingala, 130, 145
elements of, 26 Location, 21, 40, 127, 213, 214, 215, 217
idioms and, 34–35 locative plural, 77, 77f, 78, 79f
for je, 282f logical form (LF), 6, 12, 252
Korean question particles and, 113, logophoric agent, 265
114, 115 logophoric patient, 265
phonological information and, 269 long-distance extraction
prefixes and, 223, 226, 239, 246 across nonveridical domains, 164–66
pronouns and, 279, 279f, across veridical domains, 162–64
288–89, 288f Lower Sepik-R amu languages, 125
Russian verbal prefixes and, 205, 207, L-participle (active), 306, 307, 316,
208, 211, 213, 215, 217, 223, 224, 317, 318, 320, 321, 322,
225, 226, 233, 234, 235 322n15, 323n16
Slavic participles and, 311, 320, 322 L-passive, xii, 305, 306, 307, 317, 320,
suffixes and, 241, 242 321, 322, 326
Index [ 335 ]
336
L-shaped, contiguous Modern Standard Arabic, 24, 25,

syncretism, 292–94 194–96, 195t
L-shaped syncretism, with ABA, Moroccan Arabic, 194–96, 195t
292, 298 morphemes
L-syncretism, 316–20 defined, 12, 329
L-trees. See lexical trees Levels I and II, 192
nanosyntax use of term, 13
Macedonian, 47–48, 47f null, 17, 17n8
Macushi, 129, 145 syntax and, 18
Ma’di, 134, 135, 137, 141, 145 Vocabulary Insertion and, 12
main spine, xi, 242, 243, 243f, 244, 245 morphological containment
Malagasy, 120, 121 complementizers and, 173–75
Mandarin Chinese, 130, 263, 280 mapping the structure of language
Manjaku, 104 with, 20, 23–25
Manyika, 101 in spatial deixis, ix–x, 117, 133–41
Maori, 121 morphology, 18
Marind, 282t Distributed (see Distributed
Match, 310n7, 311. Morphology)
See also lexicalization portmanteau (see portmanteau
Mauritian Creole, 284 morphology)
Medial, ix–x, 118, 121, 141–43 silent, 58, 75
Distal containing, 139–40, 140f zero, 58, 74–82
Proximal contained by, 134–35, morphophonology, 14–15
137–39, 137f, 140–41, 141f morphosyntax of negation, x–xi, 180–201.
Proximal contained by Distal See also negative markers
and, 136–37 multidimensional paradigms, xii, 277,
Medial–Proximal syncretism, 285, 287–88, 290. See paradigms
127–28, 128t Mura languages, 130
merge-f, 239, 242, 245–46, 248 Muskogean languages, 120n4
Merger, 12, 59, 63
merge-XP, 240, 240f, 242, 245–46 Na-Dene, 120n4
Mérida (Venezuelan Spanish), xi–xii, nanospine, 196, 196f, 197, 198, 199–200
250, 251, 251f, 255, 262–67, nanosyntax
266f, 270t, 272, 273 basic architecture, 10–12
participant features, 264–66 defined, vii, 3
topic positions, 262–64 model of grammar according to, 11f
Middle Way approach, 194 theory and methodology, 15–46
Minimalist Program, 252 theory and terminology, 10–15
Minimize Gaps, 302 three differences between Distributed
Minimize Junk. See Elsewhere Principle Morphology and, 57–59
Mirror Principle, 45n23 Navajo, 120n4, 284
Mirror Theory, 90 negation. See Korean, negation in;
Miskitu, 280 morphosyntax of negation;
Modern Greek negative markers
complementizers of, x, 149–55, negative classifier markers (Classneg), x,
150t, 153t, 154t, 155, 156, 156t, 181, 185
158–59, 159t, 160t, 161, 161t, internal syntax of, 196, 197, 197f
162, 162n5, 164, 170, 175, 177 properties of, 189–91
negative markers in, 189, syncretism and, 194, 195t
194–96, 195t Negative Degree Phrase, 190n9
[ 336 ] Index
negative focus markers (Focneg), x, North Italian dialects, 9
181, 185 N-passive, 305, 306, 307, 316n10, 317,
internal syntax of, 196, 197, 318, 320, 321, 322
197f, 198 functional sequence
properties of, 186–89 zones and, 314–16
syncretism and, 194, 195t NP movement, 318–20, 319f, 320, 320f
negative markers, 180–201 null morphemes, silent morphology,
classification of, 181–94 zero morphology, 17, 17n8,
contradictory (see contradictory 58, 74–82, 74f
negative markers)
contrary, 182, 183, 187, 193, 194 Object, 118n3
external syntax of, 199–200 OFOH. See one feature-one head maxim
internal syntax of, 196–99 Omotic languages, 256, 272
syncretism and, x–xi, 180, 190, one feature-one head maxim (OFOH),
194–96, 195t, 199, 201 vii, 5, 10, 16, 17, 49, 329
negative scalar quantity markers (Qneg),
x, 181, 185 Paamese, 122
internal syntax of, 196, 197, 197f Palauan, 135, 137, 145
properties of, 191–94 paradigms
syncretism and, 194, 195t *ABA generalization (see *ABA
negative tense markers (Tneg), x, 181, 184 generalization)
in clausal spine, 200, 200f creation of simple, 84
internal syntax of, 196, 197, 197f, 198 defined, 329
properties of, 185–86 multidimensional, xii, 277, 285,
syncretism and, 194, 195t 287–88, 290
NegP, xi, 198, 199, 200 partial interrogatives, 253, 253f
nesting, 23–25, 140, 141 Participant, 118n3
neutral complementizers, participant features, 264–66
171–73, 174–75 participles. See Slavic participles
Nez Perce, 284 particles, 244–45. See also Korean
Niger-Congo languages, 120, 121, 122, question particles
124, 130, 131, 135, 136t, 140 partitive complementizers, 171–73,
Nilo-Saharan languages, 131, 134 174–75, 176
Nkore-Kiga, 135, 137, 145 Passamaquoddy-Maliseet, 124
NomP movement, 322, 322f, 326f passive participles.
non-, 183, 189–91, 192, 193, 195t, 197, See Slavic participles
197f, 198 Path, 20–21, 47, 123n6, 205, 206
noncontiguous syncretisms. See diagonal decomposed path, 208, 226, 230
(noncontiguous) syncretism peeling approach to case. See case peels
nondoubling prefixes, 222, 223–26 Penthouse Effect, 258
nonfactive complements, 149, 150t, 151, Persian, 194–96, 195t
153t, 156t person marking, xii, 278
nonlinear syncretism, 292–300 person-oriented spatial deixis, 118,
Non-participant, 118n3 119–20n4
nonspellout-driven movement, 36n18 Philippines languages, 122
nonterminal spellout, 58, 64, 66, 74, phonological exponents, 65, 66, 69, 82,
81–82, 83, 85 311, 329
in Classical Armenian, 77, 80 phonological form, 6, 13, 26
in Korean negation, 68, 69, 71 phonological idioms, 34–35, 35f
as type of phrasal spellout, 58n2 phonology, 105–6
Index [ 337 ]
338
phrasal spellout, viii–ix, 16–20, 49, 83 coordinated growth problem, 246–49

consequences for architecture of portmanteau (see portmanteau
grammar, 18–20 prefixes)
defined, 329 primary, 101–3, 102f
in Distributed Morphology vs. Russian verbal (see Russian verbal
nanosyntax, 58–59 prefixes)
idioms and, 33–36 secondary, 101–3, 102f
Korean question particles and, spellout and, 240, 245–46
112, 114 suffixes and asymmetry with (see
nonterminal spellout as type of, 58n2 suffix-prefix asymmetry)
Spanish word order and, xii, 251, prepositional phrases (PPs)
266–68, 269–70 morphological containment and,
terminal spellout compared with, 18n9 24–25, 25f
pied-piping, 9, 38, 38n19 Russian verbal prefixes and, 205,
Pirahã, 130, 145 212–22, 234–35
Place, 26, 40, 41f, 208, 218, 224 preverbal subjects, xi–xii, 241, 250,
plural pronouns, 278, 280–81, 280t, 254–56, 260f, 261, 262,
282t, 283, 284t, 291t, 293t, 295t, 266–68, 271
296t, 297t, 301t primary prefixes, 101–3, 102f
pointers, xii, 34n15, 91n1, 278, 285–89, Principles and Parameters framework,
285f, 293, 294, 295, 296, 297, vii, 3, 4, 46–49, 50
299, 301, 303 problem with interveners, 71
ABA introduced by, 288–89 pronouns
defined, 285, 329 clitic (object) (see clitic (object)
Polish, 247 pronouns)
functional sequence zones Cyclic Override Principle and, 32–33
and, 312–20 Elsewhere Principle and, 30–31, 279,
participles of, xii, 305, 306, 307, 287, 288, 291, 293, 301, 303
312–20, 321, 323n16, 17, 326 feature structure of, xii, 277–303
portmanteau morphology Korean question particles as,
defined, 329–30 ix, 111–15
Distributed Morphology and, 58, as multidimensional paradigms (see
74, 78, 80 multidimensional paradigms)
phrasal spellout and, 17–18 number, 280–81 (see also plural
prefixes (see portmanteau prefixes) pronouns; singular pronouns)
principles of, 14–15 person feature complex, 278–79, 280f
suppletion distinguished from, 15n6 strong, 27–29, 28f, 28t, 32–33
portmanteau prefixes, ix, 93, 94–107 Superset Principle and, 27–29, 28f,
as classifiers, 102–3 28t, 30–32, 277, 279, 282, 283,
primary and secondary, 101–3, 102f 285, 286, 302
as specifiers, 96–97 syncretism and, xii, 277, 278, 279,
suffixes distinguished from, xi, 281–85, 290, 292–301, 303
103, 104–5 weak (subject), 27–29, 28f, 28t, 32–33
syntax and phonology, 105–6 Proximal, ix–x, 118, 121, 141–43
Portuguese, 93, 265 Distal containing, 137–39, 137f
postverbal subjects, 241, 261 Medial and Distal containing, 136–37
Prefer High Junk, 302 Medial containing, 134–35,
prefixes, 239–49 137–39, 137f
complex left branches and, xi, 242, Medial containing, Medial contained
245, 248, 249 by Distal, 140–41, 141f
[ 338 ] Index
Qawasqar, 284, 284n4 semantics
Qneg. See negative scalar quantity extralinguistic (conceptual), 5
markers generative, 20n10
question particles. See Korean question grammatical, 5
particles mapping the structure of language
Quileute, 122 with, 20–21
of negative markers, 181, 182–83, 188,
readjustment rules, 12, 14, 15, 58, 74 194, 195t
recognitional demonstratives, 116, 117f Semelai, 130, 145
Relativized Minimality (RM), x, 153–55, semi-factive complements, 151, 152,
160, 161, 169, 171, 177 153t, 155, 156, 156t, 157
Resultant States, 308–9, 308n2, 309n3, Sentani, 282t
310, 310n6, 311, 315, 316 Sepik, 131
Revised Superset Principle (RSP), xii, Serbo-Croatian complementizers, x,
278, 289–93, 294, 295, 296, 297, 149–55, 150t, 153t, 154t, 156,
298, 299 156t, 158, 159, 159t, 160t, 161,
explained, 289 161t, 162, 162n5, 164, 166–68,
problem and consequence of, 301–2 170, 175, 177
RM. See Relativized Minimality Shona, 102–3
Romanian, 93, 97–98, 98t, 105 Sierra Popoluca, 280
root zone, xii, 312–14, 314f, 320, 324 singular pronouns, 280t, 282t, 283,
Route, 20, 39, 40, 41 284t, 291t, 293t, 295t, 296t,
Russian verbal prefixes, 208–9, 208f, 297t, 301t
209t, 217–20, 220t, 221, 222, Sinhalese, 124, 145, 282t
226–30, 233, 234f Sino-Tibetan languages, 130, 139
RSP. See Revised Superset Principle Slave, 284, 287, 290
Russian, 21–22, 22t, 24, 247 Slavic languages, 20, 210, 223
Russian verbal prefixes, xi, 205–35 Slavic participles, xii, 305–26. See also
argument-changing, 223–26 L-participle (active); L-passive;
benefits of nanosyntactic approach N-passive; T-passive
to, 234–35 case peels and, xii, 307, 308, 315, 318,
doubling, 222, 223–26 320–23, 326
lexical and superlexical (see lexical functional sequence zones and, xii,
prefixes; superlexical prefixes) 305, 310, 312–20, 312f, 321, 322,
nondoubling, 222, 223–26 324, 326
prepositional phrases and, 205, verbal and adjectival, 308–10, 316
212–22, 234–35 SMS, 11, 18–20, 26, 50, 208.
syntactic analysis of, 222–34 See also syntactic trees
theoretical background, 208–12 Cyclic Override Principle and, 33
defined, 330
Salish languages, 129, 282t idioms and, 34
Salt-Yui, 282t Snowball movement, 37, 38f,
scale, decomposing, 230–32 40, 40f, 41f, 42f, 43f,
scale-path parallel, 207, 211–12 44f, 46, 49, 245
Scandinavian languages, 9, 139n16 Somali, 105
Scope Islands, 172 Source, 20, 21, 39, 40, 41
scope position (negative markers), x, Russian verbal prefixes, 205–6,
181, 183–84, 185, 186, 188–89, 206f, 208–9, 209f, 209t,
191, 192, 194, 195t 213–17, 218, 220, 220t,
secondary prefixes, 101–3, 102f 221, 222, 226–30, 233
Index [ 339 ]
340
Southern Bantu languages Standard Italian, 104–5

agreement with conjoined singular Stay (in spellout algorithm), 37, 37f, 40f,
nouns, 98–100 41f, 42f, 43f, 44f, 46, 49
irregular plurals, 97–98 S-trees. See syntactic trees
nominal class prefixes in, ix, 94–103, strong islands, x, 161, 167f
104, 106–7 strong pronouns, 27–29, 28f, 28t, 32–33
singular-plural pairing, 94–95, 95t subjects
Southern Haitian Creole, 284 order of in Spanish, 250, 251, 252–53,
Spanish, 49, 62–63, 185n3. See also 252f, 254–56, 258, 262, 266–68,
Dominican Republic Spanish; 270–73, 270t
European Spanish; Mérida; word postverbal, 241, 261
order (Spanish varieties) preverbal (see preverbal subjects)
spanning, ix, 42–43, 44f, 88–94 submorphemic heads/syntactic
conceptual issues, 90–91 terminals, viii, 16–18, 88–89
constituent lexicalization vs., Subset Clause, 64, 65, 66
89–90, 107 Subset Principle, 58, 64–66, 73, 81–82
defined, 42, 90, 330 explained, 65
purported empirical argument Korean question particles and, 114n2
for, 91–93 spatial deixis and, 142f
spatial deixis, 116–45 Vocabulary Insertion Principle added
crosslinguistic attestation to, 69–71, 81
and, 118–22 suffix-prefix asymmetry, xi, 103, 104–5,
distance-oriented, 118, 119 240, 241–43, 249
functional sequence of, 141–43, 143f superlexical prefixes, 207, 209t, 210–11,
morphological containment in, ix–x, 213, 221–22, 222t, 233–34
117, 134–41 Superset Clause, 82
person-oriented, 118, 119–20n4 Superset Principle, viii, 27–29, 36, 40,
syncretism in, 123–33 49, 171, 175
Speaker (S), 117, 118, 120, 121, 122 constituent lexicalization and, 89, 90
spellout, xi, 12, 47. See also lexicalization; defined, 330
nonterminal spellout; phrasal Elsewhere Principle compared
spellout; terminal spellout with, 30–32
prefixes and, 240, 245–46 explained, 27
principles of, 26–33 Korean question particles and, ix,
(see also Cyclic Override 112, 113–14
Principle; Elsewhere Principle; negative markers and, 198
Superset Principle) pronouns and, 27–29, 28f, 28t, 30–32,
Slavic participles and, 320–21 278, 279, 282, 283, 285,
Vocabulary Insertion and, 71–74 286, 302
spellout-driven movement, viii, 18, 36–41 revised (see Revised Superset
algorithm, 37, 49, 245–46 Principle)
pre-elements, 41–46 Russian verbal prefixes and, xi, 223,
prefixes and, 241, 242, 243 225, 233, 235
Russian verbal prefixes and, 224–25 Slavic participles and, 310n7, 311
of Slavic participles, 321f Spanish word order and, 259, 260, 270
of TP, 243, 243f spatial deixis and, 142f
of VP, 241, 241f Vocabulary Insertion and, 72,
stacking properties (negative markers), 82–84, 85
181, 183, 184, 186, 190–91, 192, suppletion, 285–86
194, 195t defined, 330
[ 340 ] Index
Korean negation and, 69 prefixes and, 240
portmanteau distinguished Vocabulary Insertion and, 66, 67, 69
from, 15n6 terminal spellout, 39n20, 43n22, 45n24,
principles of, 14–15 63, 64, 83, 84
Supyire, 131 in Classical Armenian, 76
Swedish, 20, 210 in Korean negation, 71
complementizers of, 173 phrasal spellout compared
participles of, 308–9 with, 18n9
spatial deixis in, 139n16, 145 theme markers, 76–77
syncretism, viii, ix, 26, 75 Third Person (Th), 118, 121–22
complementizers and, 151, Tneg. See negative tense markers
153–55, 158–59 Tok Pisin, 131
defined, 21, 330 topic positions, 262–64
diagonal (noncontiguous), xii, Torricelli languages, 125
292, 295–97 total interrogatives, 253, 253f
horizontal, 277, 281–83, 285, 290, total nesting, 140, 141
301, 303 T-passive, 305, 306, 316n10, 317, 318,
L-, 316–20 320, 321, 322
mapping the structure of language examples of, 307
with, 20, 21–23 functional sequence zones
in multidimensional paradigms, and, 314–16
277, 303 Trumai, 280
negative markers and, x–xi, 180, 190, Tukang Besi, 125, 145
194–96, 195t, 199, 201
nonlinear, 292–99 UG. See Universal Grammar
portmanteau prefixes and, 97–98 unaccusative verbs, xii, 307, 316–20,
pronouns and, xii, 277, 278, 279, 323–25, 326
281–85, 290, 292–301, 303 unattested syncretism, 22, 23t, 133,
in spatial deixis, ix–x, 117, 121, 123–33 133t, 278, 299–301
Subset Principle and, 65 unergative verbs, xii, 323–25, 325f
unattested, 22, 23t, 133, 133t, 278, Uniformity Principle, 7
299–301 Universal Grammar (UG), 3, 5, 19,
vertical, 277, 281, 284–85, 290 47, 141
syntactic trees (S-trees), 19–20, 47, 177 Universal 20 theory, 9, 37, 104
complementizers and, 173–75, 174f Usarufa, 292
defined, 330
interaction with lexical trees, 26–33 Venda, 120, 121
syntax Venezuelan Spanish. See Mérida
morphemes and, 18 verbal passives, 308–10, 316
portmanteau prefixes and, 105–6 verbs, 244–45
syntax, morphology, and semantics. light, 313, 313f, 325f
See SMS order of in Spanish, 250, 253,
254, 258
Tahitian, 122, 125, 145 unaccusative, xii, 307, 316–20,
Tairora, 282t 323–25, 326
Target States, 308–9, 308n2, 309n3, unergative, xii, 323–25, 325f
310, 310n6, 311, 312, 315, 316 verb stems
terminals, 12, 15, 17, 18, 49 Slavic, 312, 316–20, 323–25, 326
constituent lexicalization and, 89, 90 Spanish, 269–70
defined, 330 veridical islands, 160–69
Index [ 341 ]
342
veridicality, 157–60, 177 weak islands, x, 160–69, 167f, 172

non-, 158, 159, 160, 160t, 161, 161t, weak (subject) pronouns, 27–29, 28f,
164–66, 167f, 168, 169, 175 28t, 32–33
relative vs. strong, 156, 158, 159, Welsh, 131, 138–39, 145
159t, 160t, 161t, 163, 166, 167f, West Greenlandic, 122n6
168, 169 West Tocharian, 24
vertical syncretism, 277, 281, wh-interrogatives, xi, 250, 251, 253, 254,
284–85, 290 255, 256, 258, 260, 263, 264,
VIP. See Vocabulary Insertion Principle 266, 266f, 270–73
Vocabulary Insertion, 12, 14 wh-movement, 36n18, 49
in Distributed Morphology, viii, wh-phrases
58, 64–82, 85 (see also Subset extraction of, 160–69
principle; zero morphology) features of, 169–77
Superset Principle and, 72, 82–84, 85 Winnebago, 284
Vocabulary Insertion Principle (VIP), Wolof, 284
69–71, 81 word order (Spanish varieties),
Vocabulary Items, 15, 65, 88 xi–xii, 250–73
lexical item/entry and exponents in, 266–68
lexicon vs., 13 T-to-C movement, 256–61
Russian verbal prefixes and, 211 hierarchy and lexicalized
Slavic participles and, 311 parts, 262–70
Spanish word order and, 273 participant features, 264–66
Vocabulary List, 13 topic positions, 262–64
workspace, 42, 45–46, 46f, 197–98, 199
Wailevu Fijian, 136, 137, 141
Wakashan languages, 123 Xhosa, 94, 95t, 99, 100, 101
Waray-Waray, 122 Xokleng, 282t
Warekena, 284
Wargamay, 131, 138, 145 Yeniseian languages, 126
Wari’, 124, 145 Yimas, 125, 145
[ 342 ] Index
344

(Oxford Studies in Comparative Syntax) Lena Baunaz, Liliane Haegeman, Karen de Clercq, Eric Lander-Exploring Nanosyntax-Oxford University Press (2018)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Oxford Studies in Comparative Syntax) Lena Baunaz, Liliane Haegeman, Karen de Clercq, Eric Lander-Exploring Nanosyntax-Oxford University Press (2018)

Uploaded by

Copyright:

Available Formats

Exploring Nanosyntax

Oxford Studies in Comparative Syntax

Movement and Silence Functional Structure from Top to Toe: The

Functional Heads: The Cartography of The Architecture of Determiners

Edited by Lena Baunaz, Karen De Clercq,

Published in the United States of America by Oxford University Press

© Oxford University Press 2018

All rights reserved. No part of this publication may be reproduced, stored in

You must not circulate this work in any other form

Library of Congress Cataloging-​in-​Publication Data

PART II. Empirical Investigations

PART III. Theoretical Explorations

10. Word Order and Nanosyntax: Preverbal Subjects

Nanosyntax (Caha 2009; Starke 2009, 2011) is a formal theory of language

Investigations) focuses on how the nanosyntactic line of argumentation can

PART II: EMPIRICAL INVESTIGATIONS

Overall the contributions in this part focus on empirical applications of

crosslinguistic variation in the domain of spatial deixis can be understood in

PART III: THEORETICAL EXPLORATIONS

in European Spanish, some subjects can be preverbal in Venezuelan Spanish

Lena Baunaz is a postdoctoral assistant at the University of Zurich. She holds

interests include Germanic philology, the NP/​DP parameter, demonstratives,

N anosyntax (Caha 2009; Starke 2009, 2011a, 2011b ) is a generative approach to

1.1. CARTOGRAPHY: A MAP OF SYNTACTIC CONFIGURATIONS

1.1.1 The model of grammar and full syntacticization

The broad-​strokes model of grammar currently adopted by most generativists,

Figure 1.1 Architecture of grammar [based on Rizzi 2013, 10, his (22)]

(2) The Borer–​Chomsky Conjecture

So information-​structural movement to the left periphery, for instance, is

An influential theoretical development in cartography has been Cinque’s

1.1.4 A summary of cartographic assumptions

1.2 THEORY AND TERMINOLOGY IN DISTRIBUTED

Terminological differences are common sources of confusion when moving

1.2.1 Basic architectures compared

Figure 1.3 Model of grammar according to nanosyntax (Caha 2009, 52; Starke 2011)

The rest of this section is organized as a discussion of four (clusters of re-

1.2.2 Morpheme and Vocabulary Insertion

In American structuralist approaches (e.g. Bloomfield 1933; Harris 1951), a

The correspondence between sound and meaning is in DM referred to as a

The term allomorphy in nanosyntax is understood in a restricted sense, as a

else (voiced consonants: rag-​s, tab-​s, tram-​s, rail-​s, wave-​s; vowels: bee-​s, tray-​

(3) a. [past] ⇔ -​t /​ ]V _​_​ where V = {√dwell, √spell, √dream, . . .}

1.2.5 Morphophonology, suppletion, portmanteau

1.3 NANOSYNTAX: THEORY AND METHODOLOGY

1.3.1 Submorphemic heads and phrasal spellout

As a descendant of cartography, nanosyntax assumes a strict syntax–​

(4) a. Unordered bundle (i.e. symmetrical relation)

Many important aspects of nanosyntactic theory can be seen to emerge from

(5) a. karhu-​lle (Finnish)

morphemes must correspond to syntactic heads (X0s) (and thereby having to

Figure 1.4 Spelling out -​ās in Latin

Figure 1.5 Spelling out -​i and -​lle in Finnish

1.3.2 Overall consequences for the architecture of grammar

made up of the same material, it is nevertheless important to distinguish the

1.3.3 The basic tools and technology

In this section we introduce some of the common methodological tools in use

1.3.3.1 Mapping the fseq: From linear to hierarchical order

10. Nanosyntax is not a revival of Generative Semantics, as sometimes claimed, as

(7) a. I ran at the sea. Location

(8) a. J’ai couru à la mer. Location/​Goal

As seen in (7), English prepositions make overt distinctions between Location/​

Table 1.1 SYNCRETISMS IN RUSSIAN (FROM CAHA 2009, 12)

‘window’ (sg) ‘teacher’ (pl) ‘one hundred’

nom okn-​o učitel-​ja st-​o

Library of Congress Cataloging-in-Publication Data

PART II: EMPIRICAL INVESTIGATIONS

PART III: THEORETICAL EXPLORATIONS

interests include Germanic philology, the NP/DP parameter, demonstratives,

1.1. CARTOGRAPHY: A MAP OF SYNTACTIC CONFIGURATIONS

1.1.1 The model of grammar and full syntacticization

The broad-strokes model of grammar currently adopted by most generativists,

(2) The Borer–Chomsky Conjecture

So information-structural movement to the left periphery, for instance, is

1.1.4 A summary of cartographic assumptions

1.2 THEORY AND TERMINOLOGY IN DISTRIBUTED

1.2.1 Basic architectures compared

1.2.2 Morpheme and Vocabulary Insertion

else (voiced consonants: rag-s, tab-s, tram-s, rail-s, wave-s; vowels: bee-s, tray-

(3) a. [past] ⇔ -t / ]V __ where V = {√dwell, √spell, √dream, . . .}

1.2.5 Morphophonology, suppletion, portmanteau

1.3 NANOSYNTAX: THEORY AND METHODOLOGY

1.3.1 Submorphemic heads and phrasal spellout

As a descendant of cartography, nanosyntax assumes a strict syntax–

(5) a. karhu-lle (Finnish)

Figure 1.4 Spelling out -ās in Latin

Figure 1.5 Spelling out -i and -lle in Finnish

1.3.2 Overall consequences for the architecture of grammar

1.3.3 The basic tools and technology

1.3.3.1 Mapping the fseq: From linear to hierarchical order

(8) a. J’ai couru à la mer. Location/Goal

As seen in (7), English prepositions make overt distinctions between Location/

nom okn-o učitel-ja st-o

1.3.3.2 The principles of spellout

Figure 1.10 Superset–subset relations in pronouns

Figure 1.11 Strong–weak syncretism in elle

b. Il / #Lui joue sur la plage.

(17) S-tree Spellout

1.3.3.3 Phrasal spellout and idioms

to handle multi-word idiomatic expressions, as witnessed by the clunkiness of

Spellout-driven movement is a mechanical procedure for generating new

left and may involve different degrees of pied-piping.19 It is often assumed,

20. Although we do not take a stand on whether individual-terminal spellout can or

As an addendum on the topic of spellout-driven movement, it is important