Professional Documents
Culture Documents
ii
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.
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
Glossary 329
Index 331
[ vi ] Contents
P R E FA C E
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investigated in nanosyntax at that time (Starke’s unpublished work; Caha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).
Preface [ ix ]
x
[ 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.
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
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
Structures, Vol. 2. New York: Oxford University Press.
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
[ xvi ] Contributors
PART I
Background
2
CHAPTER 1
Nanosyntax
The Basics
LENA BAUNAZ AND ERIC L ANDER
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.
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
Lexicon
Syntax
PF LF
Articulatory- Conceptual-
perceptual systems intentional systems
[ 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):
N a n o s y n ta x : T h e B a s i c s [ 7 ]
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
(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.
N a n o s y n ta x : T h e B a s i c s [ 9 ]
10
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.
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
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
N a n o s y n ta x : T h e B a s i c s [ 11 ]
12
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
1.2.4 Allomorphy
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].
N a n o s y n ta x : T h e B a s i c s [ 13 ]
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.
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)].
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.
N a n o s y n ta x : T h e B a s i c s [ 15 ]
16
[ 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).
N a n o s y n ta x : T h e B a s i c s [ 17 ]
18
KP
⇒ -ās
NumP
K
⇒ puell-
Num NP
⇒ -lle
KP
⇒ -i
NumP
K
⇒ karhu-
Num
NP
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.
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
N a n o s y n ta x : T h e B a s i c s [ 19 ]
20
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
[ 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).
N a n o s y n ta x : T h e B a s i c s [ 21 ]
22
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):
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
NOM A
ACC B
GEN A A
DAT B
INS A
N a n o s y n ta x : T h e B a s i c s [ 23 ]
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.
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).
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]]
K5 K4 K3 K2 K1 DP
mit –dat
instrumental PP
nom
acc
gen
dat
ins
com
N a n o s y n ta x : T h e B a s i c s [ 25 ]
26
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.
N a n o s y n ta x : T h e B a s i c s [ 27 ]
28
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
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
N a n o s y n ta x : T h e B a s i c s [ 29 ]
30
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.’
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
ZP ⇒ a
XP ⇒ a YP ⇒ b YP
Z
XP XP
Y Y
X
X X
Figure 1.13 *ABA pattern
N a n o s y n ta x : T h e B a s i c s [ 31 ]
32
(L4) (L5)
ZP
YP
YP
b a Z
XP
Y XP
Y
X
X
[ 32 ] Background
(L6) (L7) (L8)
CP
IP ΣP ΣP
/l / / il / / lwi / C IP
IP
Σ Σ
I I I
CP ⇒ lui
ΣP ⇒ il Σ P ⇒ il
IP ⇒ le C
IP ⇒ le IP
Σ Σ
I I I
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.
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
N a n o s y n ta x : T h e B a s i c s [ 33 ]
34
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
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
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.
N a n o s y n ta x : T h e B a s i c s [ 35 ]
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
[ 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).
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).
FP FP
HP HP
F F
GP GP
N a n o s y n ta x : T h e B a s i c s [ 37 ]
38
FP FP
GP
HP HP
F
F
GP
FP
FP HP
F
HP
F
GP
GP
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
N a n o s y n ta x : T h e B a s i c s [ 39 ]
40
DP ⇒ bajdan
Figure 1.22 Spelling out DP
AxPartP ⇒ *
AxPart
DP
SNOWBALL: move DP to the left of AxPart, match AxPartP with < AxPartP on >
bajdan ⇐ DP
AxPartP ⇒
AxPart
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
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
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
N a n o s y n ta x : T h e B a s i c s [ 41 ]
42
Goal
DP AxPartP PlaceP
AxPart Place
Place
SNOWBALL: undo CYCLIC and raise [[DP AxPartP] PlaceP]; match GoalP with < -r GoalP >
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
STAY: no match for entire structure
SourceP ⇒ *
Source
Goal
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.
N a n o s y n ta x : T h e B a s i c s [ 43 ]
44
Route
Goal
CYCLIC: move [[DP AxPartP] PlaceP]; match [RouteP [SourceP [GoalP]]] with
< -gal [RouteP [SourceP [GoalP]]] >
SourceP
AxPart Place Route
Source GoalP
Goal
K6P
K5P
K6
K5
mit K4P ⇒ DAT
DP
K6P
K5P
mit ⇐ K6
K5 K6 tK5
K4P ⇒ DAT
DP
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.
N a n o s y n ta x : T h e B a s i c s [ 45 ]
46
Workspace-2 Workspace-1
mit ⇐
K6 K5
mit ⇐
K6 K5 K4P ⇒ DAT
DP
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).
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
Place
Figure 1.31 Macedonian na-kaj
PathP
⇒ naar
Path PlaceP
Place
Figure 1.32 Dutch naar
N a n o s y n ta x : T h e B a s i c s [ 47 ]
48
‘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
[ 48 ] Background
K3P
⇒ GEN-s
K3 K2P
⇒ NOM–acc -Ø
K2 K1P
K1
1.5 CONCLUSION
N a n o s y n ta x : T h e B a s i c s [ 49 ]
50
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.
New York: Oxford University Press.
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
Structures, Vol. 3. New York: Oxford University Press.
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. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
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.
N a n o s y n ta x : T h e B a s i c s [ 51 ]
52
[ 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.
N a n o s y n ta x : T h e B a s i c s [ 53 ]
54
[ 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.
Pantcheva, Marina. 2011. Decomposing Path: The Nanosyntax of Directional Expressions.
Doctoral dissertation, University of Tromsø.
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.
Amsterdam: John Benjamins.
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.
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. 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
Structures, Vol. 2. New York: Oxford University Press.
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.
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.
Starke, Michal. 2009. “Nanosyntax: 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.
N a n o s y n ta x : T h e B a s i c s [ 55 ]
56
[ 56 ] Background
CHAPTER 2
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
[ 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
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 [ 61 ]
62
[ 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.
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 [ 63 ]
64
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.
[ 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.
Neuter Masculine
[+singular] [-masc] [+masc]
4. I am ignoring the potential decomposition into -u-m/-u-s, which does not change
the point.
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 [ 65 ]
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.
Neg
Neg V
ani/mos SLEEP
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
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.
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 [ 67 ]
68
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.
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 KNOW
Figure 2.3 Fusion of Neg and V
Neg
Neg V molu
+NEG KNOW
[ 68 ] Background
2.3.4 The communication breakdown
Let us check. Consider for example the Vocabulary Item for SLEEP, given
in (6).
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?
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
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 [ 69 ]
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.
Cause
V Cause
KNOW li
al
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
Neg
Neg Cause
ani V Cause
KNOW li
al
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 [ 71 ]
72
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
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
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 [ 73 ]
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
V TPAST
broke/gave Ø
Neg
Neg V
Ø 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).
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 [ 75 ]
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.
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
Case sg pl sg pl sg pl Features
[ 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
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 [ 77 ]
78
Case sg pl Features
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’
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.
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 [ 79 ]
80
[ 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).
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 [ 81 ]
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.
[ 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.
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).
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 [ 83 ]
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.
Recall finally the paradigm in Table 2.1, repeated as Table 2.7 for convenience.
How does the Superset Principle deal with this example?
Neuter Masculine
[+singular] [-masc] [+masc]
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
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
REFERENCES
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 [ 85 ]
86
[ 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.
Doctoral dissertation, Ghent University.
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.
Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar, and the Structure of
IP.” Linguistic Inquiry 20 (3): pp. 365–424.
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.
Starke, Michal. 2009. “Nanosyntax: A Short Primer to a New Approach to Language.”
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.
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 [ 87 ]
88
CHAPTER 3
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):
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):
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
(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.
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:
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):
[ 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:
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.
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).
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 [ 91 ]
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:
The containment problem then arises from the assumption that the direct ob-
ject is included in the TP in sentences like (11):
(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:
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.
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)].
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.
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 [ 93 ]
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.
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
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).
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 [ 95 ]
96
prefix
prefix
ba-
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
ba-
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:
Some class 9 nouns also have plurals (only) in class 6, as seen in (19):
7. These are all nouns denoting members of a tribe or ethnic group. The significance
of this remains to be determined.
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 [ 97 ]
98
sg pl
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.
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):
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:
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):
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 [ 99 ]
100
prefix
[NumlP [NumP Pl [ClP Clx [NP Nx ]]] Pl [ClP <NumP> Cl [NP N2 ]]]
prefix
[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
vu-
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):
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 [ 101 ]
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-
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.
prefix
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
[ 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):
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.
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 [ 103 ]
104
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.
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
suffix
[ 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> ]]]
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.
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 [ 105 ]
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).
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
[ 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.
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
Carstens, Vicki. 1991. The Morphology and Syntax of Determiner Phrases in Kiswahili.
Doctoral dissertation, UCLA.
Cinque, Guglielmo. 2005. “Deriving Greenberg’s Universal 20 and Its Exceptions.”
Linguistic Inquiry 36 (3): pp. 315–332.
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.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
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.
Doctoral dissertation, University of Tromsø.
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.
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 [ 107 ]
108
PART II
Empirical Investigations
110
CHAPTER 4
(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 . . . ’
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.).
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:
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.
REFERENCES
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
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.
Starke, Michal. 2009. “Nanosyntax: 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
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
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)
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
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.
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.
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
Interpretation Label
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
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 [ 119 ]
120
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.
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.
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 [ 121 ]
122
5.2.2.1 Dist ≠ Med ≠ Prox
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 [ 123 ]
124
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.
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 [ 125 ]
126
(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’
5.2.2.2 Dist ≠ Med = Prox
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 [ 127 ]
128
‘close to S or H’
5.2.2.3 Dist = Med ≠ Prox
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
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.
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 [ 129 ]
130
12. In fact, it is explicitly stated that “wâná covers both the medial and the distal
scopes” (Meeuwis and Stroeken 2012, 148).
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
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 [ 131 ]
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.
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.
neutral dem
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 [ 133 ]
134
5.3 MORPHOLOGICAL CONTAINMENT
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.
15. For ogwo, Taylor (1985, 135) also provides the meaning ‘not far removed from
both speaker and hearer.’
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 [ 135 ]
136
Medial Proximal
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.
1 2 3 4 5 6
bá mī mā bī mā dī ‘near S’
1 2 3 4 5 6
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
Distal 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.
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).
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 [ 137 ]
138
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.
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,
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
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 [ 139 ]
140
The forms from Boumaa Fijian show the containment relation sketched in
Figure 5.6.
Distal Medial
Figure 5.6 Distal contains Medial
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.
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.
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 [ 141 ]
142
DeixP
Deix0
Prox
Med
Dist
Dx3P
Dx3 Dx2P
Dx2 Dx1P
Dx1
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
Dx1
Dx1 Dx1
Figure 5.12 Additive heads
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 [ 143 ]
144
5.5 FINAL THOUGHTS
LANGUAGES
REFERENCES
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 [ 145 ]
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.
Cambridge: Cambridge University Press.
Bühler, Karl. 1934. Sprachtheorie: Die Darstellungsfunktion der Sprache. Jena,
Germany: Fischer.
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
Chandralal, Dileep. 2010. Sinhala. Amsterdam: John Benjamins.
Cinque, Guglielmo and Luigi Rizzi. 2008. “The Cartography of Syntactic Structures.”
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.
Amsterdam: John Benjamins.
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.
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 [ 147 ]
148
Decomposing Complementizers
The Functional Sequence of French, Modern Greek,
Serbo-Croatian, and Bulgarian Complementizers.
LENA BAUNAZ
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
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.
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
Second, MG, SC, and Bg have a distinct mood particle marking the subjunc-
tive mood under desiderative ‘wish’-type verbs (6).1
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).
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.
6.1.1 Proposals
2. For syntactic complexity inside of complementizers, see also Leu (2015), who
argues that German dass is two complementizer heads: d-and -ass.
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 [ 153 ]
154
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.
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:
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:
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.).
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 [ 155 ]
156
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.
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.
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).
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 [ 157 ]
158
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).
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
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):
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.
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 [ 159 ]
160
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.
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
English Translation MG SC Bg Fr
Strongly ‘remember’ thimame sjetiti se pomnja se rappeler
veridical ‘regret’ %
žaliti sâžaljavam
‘understand’ comprendre
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 [ 161 ]
162
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.
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).
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 [ 163 ]
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.
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 [ 165 ]
166
(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).
Nonveridical
(verbs of
communication)
SC da
Bg če/da argument No
MG oti/na adjunct Island
Fr que
Nonveridical
(desideratives,etc.)
(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?’
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 [ 167 ]
168
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
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.).
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.
(49) strongly veridical > relatively veridical > nonveridicalind > nonveridicalsubj
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 [ 169 ]
170
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).
BaseP ⇒ (-)que/(-)ti(-)/(-)pu(-)/(-)deto
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
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
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 [ 171 ]
172
(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?’
(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’
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.
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 [ 173 ]
174
partP
spec
/ ki / nP
part
specP ⇒ qui
Øwh ⇐ OpwhP partP
spec
nP
part
n
(ii) partitive quiwh
partP ⇒ qui
Øwh ⇐ OpwhP nP
part
n
nP ⇒ qui
Øwh ⇐ OpwhP
n
nP ⇒ qui
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
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.
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 [ 175 ]
176
cP ⇒ /k /, /na/, /da/
(b) XP complementizer
XP ⇒ /k /, /oti/, /da/, /t∫e/
cP
X
c
Figure 6.5 Complementizer structures
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.
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 [ 177 ]
178
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 [ 179 ]
180
CHAPTER 7
7.1 INTRODUCTION
7.2.2 Classifying properties
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.
The LC states that a proposition (p) cannot be true and false in the same
circumstances.
The Law of the Excluded Middle says that any proposition is either true
or false.
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.
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.
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.
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 [ 183 ]
184
TP
T FocP
Foc vP
V ClassP
Class QP
Q aP
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.
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,
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
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.
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 [ 185 ]
186
It is impossible to stack the same negative marker n’t twice on top of each
other with the same predicate, see (11).6
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.
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.
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 [ 187 ]
188
(18) I am not angry with you for writing such innocent letters as these.
(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
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
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 [ 189 ]
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.
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).
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).
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)
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.
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.
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.
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 [ 191 ]
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.
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).
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)
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).
(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.
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 [ 193 ]
194
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.
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.).
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 [ 195 ]
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.
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
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
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.
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 [ 197 ]
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
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):
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.
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
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 [ 199 ]
200
AgrSP
NP AgrS’
is TP ⇒ not TP
T FocP T vP
Class QP Q aP
Q NegP a √
Neg
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).
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
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 [ 201 ]
202
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 [ 203 ]
204
8.1 INTRODUCTION
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.’
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-).
1. Note that -oto-is the allomorph of -ot-that appears before consonant clusters.
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
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
8.2.1 Decomposed Path
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
Figure 8.5 Source structure
Goal
PlaceP ...
Figure 8.6 Goal structure
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.
Meaning
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 [ 209 ]
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.
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.
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.
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.
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 [ 211 ]
212
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.
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).
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 [ 213 ]
214
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.
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.
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).
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 [ 215 ]
216
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)].
8.3.3 Route
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 [ 217 ]
218
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).
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.
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).
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).
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 [ 219 ]
220
With do-it is also possible to specify goal, both source and goal, and the dis-
tance (19):
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):
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.
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):
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 [ 221 ]
222
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.
8.4 SYNTACTIC ANALYSIS
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
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.
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 [ 223 ]
224
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
PrefP PathP
Path PlaceP
Path PlaceP
Place
Place GROUND
PP
DP PP ⇒ pere
river
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).
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.
PP
PrefP PathP
Path Place
Path PlaceP
Path Place tPlace Path Place tPlace DP
GROUND
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 [ 225 ]
226
Place GROUND
PP
Figure 8.10 Lexical entry of vy-and spelling out prefP (but not PathP) in (28)
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.
PrefP PathP
Route Route
Source Source
Goal PlaceP
Goal PlaceP
Place
Place GROUND
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):
PP
Place GROUND
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.
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
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 [ 227 ]
228
PP
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].
PathP PlaceP
PathP ⇒ v dom PlaceP
Goal Place tPlace GROUND Goal Place tPlace Place
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).
PP
tPlace Place
Source tPlace GROUND Source
Goal Place Goal Place
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 [ 229 ]
230
PP
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
lentils
Path PlaceP ScaleP
Path
Source tPlace GROUND lentils ScaleP
Goal Place
Scale
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).
PP
ot- ⇐ PrefP PP
lentils PP ⇒ do-
Path PlaceP
Goal
Source tPlace GROUND tScaleP
Place
Goal Place
readiness
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.’
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 [ 231 ]
232
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
(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.
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
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 [ 233 ]
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
Init ProcP
Proc
Figure 8.22 Source/completive superlexical prefix
PP
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.
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. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
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.
Pantcheva, Marina. 2011. Decomposing Path: The Nanosyntax of Directional Expressions.
Doctoral dissertation, University of Tromsø.
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 [ 235 ]
236
Theoretical Explorations
238
CHAPTER 9
K3
K2
K1
Z
Y
X W ...
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
K2 K1 Z
Y
X W ...
9.1 PRE
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
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
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
T Asp
Figure 9.7 Prefix structure
T
Asp
Figure 9.8 Suffix structure
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.
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
TP
C
Wh
Figure 9.10 Spellout-driven movement of TP
TP
C Wh
Figure 9.11 ‘Specifier’-like complementizer
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 [ 243 ]
244
cause
become
state
PRT1 PRT2
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
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
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
(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.
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 [ 245 ]
246
(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).
1. /sugar/
2.
Z
Y ...
X
3. SUGAR
1. /of/
2.
K3
K2
K1 Z
Figure 9.17 Lexical entry for of
K2
K1 Z
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 ...
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 [ 247 ]
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 ...
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
REFERENCES
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
Starke, Michal. 2011. “Towards elegant parameters: language variation reduces to
the size of lexically stored trees.” Ms. Tromsø University. [online] Available at
lingbuzz/001183.
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 [ 249 ]
250
CHAPTER 10
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
* 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?’
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)
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 ...
IntP
Int ...SubjP
Subj ...vP
WhP
Wh IntP
Int ...SubjP
Subj ...vP
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.
W or d Or de r a n d N a n o s y n ta x [ 253 ]
254
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
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
2. Suñer (1994) argues that some adverbs are allowed between the wh-element and
the verb in interrogatives:
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).
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.
W or d Or de r a n d N a n o s y n ta x [ 255 ]
256
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).
(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]
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.
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:
W or d Or de r a n d N a n o s y n ta x [ 257 ]
258
(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).
(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.
Qué dicen?
?
WhP
what say.3pl
Wh IntP ⇒ (dice-)/-n/
qué
‘what’ Int ...SubjP
Subj TP
3pl
T ...vP
ellos ...
‘they’
*ellosi Subj
‘they’
Subj TP
3pl
T ...vP
ti ...
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.
W or d Or de r a n d N a n o s y n ta x [ 259 ]
260
ellos X
‘they’
X ...SubjP
(dice)-n
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
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.
(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.
W or d Or de r a n d N a n o s y n ta x [ 261 ]
262
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
(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:
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
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) 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.
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.
W or d Or de r a n d N a n o s y n ta x [ 263 ]
264
10.4.2 ParticipantP
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.
W or d Or de r a n d N a n o s y n ta x [ 265 ]
266
(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.
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
FinP ⇒ -mos
‘1pl’
Fin SubjP
Subj
1pl
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.
WhP ⇒ qué
Wh IntP
Int ContrP
él Contr
Fin SubjP
él Subj
Subj
3sg
WhP ⇒ qué
Wh IntP
Int PartP
tú Part
Fin SubjP
tú Subj
Subj
2sg
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
W or d Or de r a n d N a n o s y n ta x [ 267 ]
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
WhP ⇒ qué
Wh IntP
Int ...FinP
Fin SubjP
Ø
uno Subj
‘one’
Subj TP
3SG
(dice)-Ø
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.
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):
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:
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
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.
W or d Or de r a n d N a n o s y n ta x [ 269 ]
270
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).
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.
(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):
W or d Or de r a n d N a n o s y n ta x [ 271 ]
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).
sg pl
1 -u -u
2 -u -end
3 -u (m) /-en (f) -end
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):
W or d Or de r a n d N a n o s y n ta x [ 273 ]
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.
Oxford: Oxford University Press.
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.
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
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.
Cambridge, MA: MIT Press.
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
W or d Or de r a n d N a n o s y n ta x [ 275 ]
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.
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. 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.
Amsterdam: John Benjamins.
Sundaresan, Sandhya. 2013. Context and (Co)Reference in the Syntax and Its Interfaces.
Doctoral dissertation, University of Tromsø.
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,
pp. 43–70. New York: Oxford University Press.
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.
Oxford: Oxford University Press.
11.1 INTRODUCTION
sg pl sg pl
1 A B 1 A A
2 A C 2 B C
3 D E 3 D E
* 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.
I adopt the proposal put forth by Béjar (2003). This proposal assumes that
there are three privative features, [speaker], [participant], and [person],
1 2P 2P
2 3P 2 3P 3P
3 3 3
Figure 11.1 Syntactic trees for first, second, and third person personal pronouns
1 2P 2 3P
2 3P 3
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
sg pl
1 wǒ wǒ-men
2 nı̌ nı̌-men
3 tā tā-men
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.
pl 1P NumP
1 2P pl 2P NumP
2 3P 2 3P pl 3P
3 3 3
11.4 ATTESTED SYNCRETISMS
11.4.1 Types of patterns
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
11.4.2 Horizontal syncretisms
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.
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 [ 281 ]
282
sg pl
1 ai ne
2 aame aame
3 je je
pi 2P pi 3P
2 3P 3
</ai/, 1P >
1 2P
2 3P
Figure 11.5 Lexical item for ai
1 2P Num1 2P
2 3P 2 3P
3 3
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.
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 [ 283 ]
284
11.4.3 Vertical syncretisms
sg pl
1 sį naxį
2 nį naxį
3 ʔedį
ʔegedį
Num2 Num1P
Num1 1P
1 2P
2 3P
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:
XP ⇒ brought
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 [ 285 ]
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.
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.
Past PerfP
Perf PassP
Pass
XP ⇒ brought
Past PerfP
Perf PassP
Figure 11.10 XP for brought, with pointers to bring and to more complex structure of -ed
KP
K NumP
Num ∏P
Num2 Num1P
Num1 1P
1 2P
2 3P
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
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 [ 287 ]
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
sg pl
1 C A
2 D B
3 E A
Num1 1P Num1 2P
1 2P 2 3P
2 3P 3
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.
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 [ 289 ]
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.
B C B E
D E F G
F G
Table 11.8 PRONOUN
PARADIGM OF SL AVE
sg pl
1 sį naxį
2 nį naxį
3 ʔedį ʔegedį
Num2 Num1P
Num1 1P
1 2P
2 3P
Num2P
Num2 Num1P
Num1 2P
2 3P
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.
sg pl
1 C A
2 D B
3 E A
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 [ 291 ]
292
Num1 1P Num1 2P
1 2P 2 3P
2 3P 3
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):
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.
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).
sg pl
1 ke ke
2 e ke
3 we ye
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).
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 [ 293 ]
294
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
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
11.7.4 Diagonal
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
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 [ 295 ]
296
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).
sg pl sg pl
1 B C B C
2 D A A D
3 A E E A
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
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
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.
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 [ 297 ]
298
Table 11.21 L SHAPE
WITH ABA
sg pl
1 A A
2 C B
3 D A
11.7.6 Double L, 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.
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
(8) logically possible ⊃ derivable with pointers ⊃ attested ⊃ derivable with RSP
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 [ 299 ]
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.
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):
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,
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.
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
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 [ 301 ]
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.
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.
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. 2009. The Nanosyntax of Case. Doctoral dissertation, University of Tromsø.
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.
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 [ 303 ]
304
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.
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.
(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 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
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.
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).
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)]:
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.
PastTP
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.
F2
F1 theme zone ⇔ {∅, E, EJ, NU, AJ, OVA, I}
F2
F1 root zone ⇔ {adj, noun, verb}
F2 F1
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.
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
N3 GET
N2 HAVE
N1
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):
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’.
Verb
Cat3 Noun
Cat2 Adjective
...
rootN-NGIVE
... N2 HAVE
N1 t
rootA-NGET
... N2 HAVE
N1 t
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.
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
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:
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
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.
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
(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 ]]]]]]
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.
KaseP
K2 NomP
root zone theme vowel
K1 NP Ø
z-mar-
król
F2Ppass⇔T/N
F2 F1Ppass
F1 KaseP
K1 NP
król
KaseP F2Ppass⇔T/N
F2 F1Ppass
AccP VP
F1 t
K2 NomP z-mar-
K1 NP
król
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):
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
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
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.
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
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
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.
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).
GIVE
N3 GET
N2 HAVE
N1
N1 t K1 NP F1 t
Karel
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
12.6 CONCLUDING SUMMARY
REFERENCES
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.
[ 330 ] Glossary
INDEX
Note: Tables and figures are indicated by t and f following the page number.
[ 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
[ 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
[ 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
[ 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
[ 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
[ 342 ] Index
344