Decomposing Path
The Nanosyntax of Directional Expressions
Marina Blagoeva Pantcheva
A thesis submitted for the degree Philosophiae Doctor
University of Tromsø
Faculty of Humanities, Social Sciences and Education
CASTL
May, 2011
ii
Acknowledgments
When I told my father a couple of years ago that I got a PhD position at
the University of Tromsø, first thing he did was to try to locate Tromsø
on the world map. As it happens to many people who face the same
task, he glided his finger along the Norwegian coastline, going north, and
more north, and more north, passing the Polar Circle, and then even
more north, while the expression on his face was changing from slightly
amused to deeply worried. When his finger finally stopped its journey
at the 69th parallel, he said: “This town is on the edge of the world! Be
careful not to fall off.”
This little Norwegian island on the edge of the world turned out to
be a marvelously exciting, enriching and vibrant place to live. And,
more importantly, it had CASTL — the Centre for Advanced Study in
Theoretical Linguistics where I got friends for life, a husband, two kids,
and, last, but surely not least, a Dissertation. The last feat required a
great deal of knowledge and skills which I acquired thanks to so many
people. It is these people that I wish to thank now.
First and foremost, I want to express my deep gratitude to my supervisor
Peter Svenonius. There are so many things I want to thank him
for that I really do not know where to begin. Let me start by thanking
Peter for accepting into his Moving Right Along project a PhD student
who knew nothing about generative grammar. I thank him for teaching
me the things I know about it now. Peter is a great teacher, incredibly
patient and always capable of explaining the most intricate linguistic
issues, even to a beginner. I am also grateful to Peter for being such
a devoted and supportive supervisor and for always finding time in his
busy schedule to read my numerous drafts and meet me. Peter’s astonishing
knowledge of linguistics and his constructive and often critical
comments (yet always presented with an amazing delicacy) improved this
work greatly. Finally, I wish to thank Peter for being a wonderful friend
in addition to being a terrific supervisor.
Second, I want to thank Pavel Caha, the person with whom I could
iii
iv ACKNOWLEDGMENTS
discuss linguistic any time of the day. Pavel’s huge linguistic knowledge,
intelligence and insightful comments helped me clarify many questions I
asked myself concerning syntax in general and Nanosyntax in particular.
I am also grateful to Michal Starke, Tarald Taraldsen and Gillian
Ramchand for the inspiring seminars they held and the fruitful linguistic
discussions (among all others).
I greatly acknowledge the feedback I got from linguists I met at conferences
in Tromsø and abroad. Among others, I would like to thank
Marcel den Dikken, Joost Zwarts, Sander Lestrade, Patrik Bye and all
the participants of the Workshop on Locative Case in Nijmegen in 2008,
where I presented the ideas developed in this thesis for the first time.
I would like to thank Klaus Abels for the exciting class in Mathematical
Linguistics. This was the best term I had in Tromsø.
I want to thank all the people at CASTL for making it such a nice
place to be, where both linguistics and friendships thrive. I am particularly
grateful to CASTL’s ex-director Curt Rice and its present director
Marit Westergaard for creating such an outstanding working environment.
I also thank the people who work(ed) at the the Faculty of Humanities,
Social Sciences and Education and at the administration of CASTL
for quickly and efficiently helping me solve administrative and practical
problems. I want to especially mention Ståle Berglund, Christin Kristoffersen,
Tore Benzt, Kaori Takamine, Jorun Nordmo, Elisabeth Eriksen
and Jan Helge Bergheim.
The last five years of my life would have been much less exciting
and enriching had I not shared them with my wonderful fellow PhD students:
Monika Bader, Pavel Caha, Björn Lundquist, Kaori Takamine,
Peter Muriungi, Zhenya Romanova, Yulia Rodina, Éva Dékány, Kristine
Bentzen, Mai Tungseth, Nayouki Yamato, Dragana Šurkalović, Christine
Østbø, Peter Jurgec, Andrea Márkus, Inna Tolskaya, Islam Youssef,
Marleen van de Vate, Sandya Sundaresan, Rosmin Mathew, Pavel Iosad,
Sylvia Blaho, Anna Wolleb, Alexander Pfaff, Helene Andreassen, and
Violeta Martínez-Paricio.
The years at CASTL were great fun also because of the following
colleagues: Christian Uffmann, Martin Krämer, Antonio Fábregas, Minjeong
Son, Thomas McFadden and Irene Franco. Thank you for the nice
time we spent together.
I greatly appreciate having the possibility of sharing an office with
Kaori, Björn, and Éva. Having them as office buddies was a truly enjoyable
experience.
I wish to thank my friends “outside linguistics,” more than half of
v
whom accidentally happen to be linguists too: Gillian, Björn, Monika,
Lucius, Lucie, Yulia, Adnan, Kaori, Tom, Tanya Boteva, Jonas, Isabelle,
Pierre, Anna-Lena, Gunnar, Klaus, Luisa. Thanks for the great picnics,
pleasant trips and lovely dinners.
I owe a very special and very big thank to Karla. Thank you, Karla,
for always finding time to visit us, for all the help with the kids and the
household chores, and for driving Misha in your Golden Car.
I want to especially thank my soul sister Monika. It was so nice
talking to somebody who was in the same situation as me. One “Oh, last
week the kids were sick again” was enough for Monika to understand all
my feelings in the tiniest detail.
As for people outside CASTL, I thank my friends back home for
the nice time we had when I were back in Bulgaria: Maria, Joro, Ina,
Momchil, Emma, Teddy, Elica, and Dani Molle.
I would have never coped without Pavel. I am eternally grateful to
him for the sacrifices he made in order for me to finish this thesis. Enumerating
here all the things he did for me would make this dissertation
look tiny compared to its acknowledgments, so I will just say it in big
letters: THANK YOU! I also thank my children Misha and Jana for being
there and preventing me from becoming enslaved by my thesis. The
numerous joys (and occasional worries) they brought into my life left no
space (nor time) for dissertation panic or despair.
I owe thanks to my good old Mac for not completely crashing, despite
being so overburdened that it would block about once per hour in the
last months of thesis writing.
I am also very grateful (but don’t know whom to thank) for the fact
that there are no heat waves in Tromsø. Given that I feel sick when it is
hotter than 25 ◦
C, the non-winter season in Tromsø is cool enough for my
brain to function. I fact, it was a marvelous experience to work until late
in the “summer” and then bike home in the magic light of the Midnight
Sun, full of admiration for the fantastic nature around.
Talking about the weather, the winter in Tromsø is actually not cold
at all. Still, I truly missed the warmth of my wonderful family. I wish
to thank my mother Parvaneh Danesh and my father Blagoy Pantchev
for encouraging me to drop business and go into academia, for all the
support and dedication, and for always being ready to travel so far north
and take care of grandchildren and household. I thank my sister Mina
and my brother-in-law Ventzi for the lovely holidays we spent together.
Finally, I want to thank my parents for letting me go to Tromsø despite
their anxiety. I dedicate this thesis to them.
vi ACKNOWLEDGMENTS
Contents
Acknowledgments iii
Abstract xiii
Abbreviations xv
1 Introduction 1
1.1 Background and object of study . . . . . . . . . . . . . . . . . . 1
1.2 Core proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Theoretical apparatus . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 How to read the thesis . . . . . . . . . . . . . . . . . . . . . . . 8
I Path expressions: typology, syntax,
and semantics 9
2 Types of paths 11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Paths with transition . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Paths without transition . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Orientation of paths . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Delimited paths . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6 The two types of Route paths . . . . . . . . . . . . . . . . . . . 28
2.7 Classification of paths . . . . . . . . . . . . . . . . . . . . . . . 30
2.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 Background and theoretical assumptions 35
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 The relation between syntax and morphology . . . . . . . . . . 35
3.3 Lexicalization of the structure . . . . . . . . . . . . . . . . . . . 41
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
vii
viii CONTENTS
4 Decomposing Path 45
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Goal vs Source paths . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Routes vs Goal and Source paths . . . . . . . . . . . . . . . . . 51
4.4 Transitional vs non-transitional paths . . . . . . . . . . . . . . 55
4.5 Delimited and non-delimited paths . . . . . . . . . . . . . . . . 58
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5 Decomposed Path semantics 65
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 The semantics of Place . . . . . . . . . . . . . . . . . . . . . . . 65
5.3 The semantics of Goal . . . . . . . . . . . . . . . . . . . . . . . 67
5.4 The semantics of Source . . . . . . . . . . . . . . . . . . . . . . 69
5.5 The semantics of Route . . . . . . . . . . . . . . . . . . . . . . 71
5.6 The semantics of non-transitional paths . . . . . . . . . . . . . 75
5.7 The semantics of delimited paths . . . . . . . . . . . . . . . . . 87
5.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
II The Nanosyntax theory of grammar 93
6 Lexicalization of the structure 95
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2 Capturing the mismatch . . . . . . . . . . . . . . . . . . . . . . 99
6.3 Nanosyntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.3.1 The relationship between morphology and syntax . . . . 106
6.3.2 Spell-out of syntactic structure . . . . . . . . . . . . . . 109
6.4 The Superset Principle . . . . . . . . . . . . . . . . . . . . . . . 118
6.5 The Elsewhere Condition . . . . . . . . . . . . . . . . . . . . . 124
6.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
7 Spell-out driven movement 131
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.2 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.3 Spell-out driven movement . . . . . . . . . . . . . . . . . . . . . 133
7.4 Elaborating Spell-out driven movement . . . . . . . . . . . . . . 143
7.4.1 Timing of Spell-out driven movement . . . . . . . . . . . 143
7.4.2 The landing site of evacuated material . . . . . . . . . . 150
7.4.3 Backtracking . . . . . . . . . . . . . . . . . . . . . . . . 158
7.5 Segment vs category matching . . . . . . . . . . . . . . . . . . . 167
7.5.1 Category matching . . . . . . . . . . . . . . . . . . . . . 168
7.5.2 Segment matching . . . . . . . . . . . . . . . . . . . . . 171
7.5.3 Segment vs category matching . . . . . . . . . . . . . . 172
7.6 Finnish - The Grand Finale . . . . . . . . . . . . . . . . . . . . 176
CONTENTS ix
7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
III Syncretisms and partitioning
of the syntactic structure 193
8 Partitioning of the structure 195
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
8.2 Real and spurious syncretisms . . . . . . . . . . . . . . . . . . . 196
8.3 Lexicalization patterns . . . . . . . . . . . . . . . . . . . . . . . 201
8.3.1 Lexicalizing Route paths . . . . . . . . . . . . . . . . . . 201
8.3.2 Lexicalizing Source paths . . . . . . . . . . . . . . . . . 209
8.3.3 Lexicalizing Goal paths . . . . . . . . . . . . . . . . . . 214
8.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
9 Syncretisms 219
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
9.2 Possible and impossible syncretisms . . . . . . . . . . . . . . . . 221
9.2.1 *ABA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
9.2.2 *A&¬A . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
9.2.3 Syncretism typology . . . . . . . . . . . . . . . . . . . . 225
9.3 Typological studies . . . . . . . . . . . . . . . . . . . . . . . . . 228
9.3.1 The possible patterns of syncretisms: some examples . . 234
9.3.2 The impossible patterns of syncretisms: some counterexamples
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
9.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
10 Conclusion 247
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
10.2 Summary of the thesis . . . . . . . . . . . . . . . . . . . . . . . 248
10.3 Directions for future research . . . . . . . . . . . . . . . . . . . 254
Appendix: Language sample and references 261
Bibliography 265
x CONTENTS
List of Tables
2.1 Spatial case system in Tabasaran . . . . . . . . . . . . . . . . . 21
2.2 Classification of paths . . . . . . . . . . . . . . . . . . . . . . . 30
2.3 Path terminology used in Zwarts (2008a) and in this thesis . . . 33
3.1 Morphological containment of locative expressions inside directional
expressions . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Relation between paths and locations . . . . . . . . . . . . . . . 40
4.1 Spatial case system in Chamalal . . . . . . . . . . . . . . . . . . 47
4.2 Languages where the Source marker morphologically contains
the Goal marker . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 Spatial case system in Akhvakh . . . . . . . . . . . . . . . . . . 53
4.4 Spatial case system in Avar . . . . . . . . . . . . . . . . . . . . 53
4.5 Transitional and non-transitional paths in Avar . . . . . . . . . 56
4.6 Transitional and non-transitional paths in Tabasaran . . . . . . 56
4.7 Delimited and non-delimited paths . . . . . . . . . . . . . . . . 59
5.1 Transitional and non-transitional paths . . . . . . . . . . . . . . 75
6.1 Spatial case system in Lak . . . . . . . . . . . . . . . . . . . . . 97
6.2 The in and on series in Bagvalal . . . . . . . . . . . . . . . . . 102
6.3 Locative case in Chamalal . . . . . . . . . . . . . . . . . . . . . 108
6.4 The on series in Avar . . . . . . . . . . . . . . . . . . . . . . . 124
7.1 Spatial case system in Finnish . . . . . . . . . . . . . . . . . . . 177
7.2 Decomposition of Finnish spatial cases . . . . . . . . . . . . . . 177
8.1 Cases used by languages to express a Route path “across” . . . 202
8.2 Possible partitionings of a Route structure . . . . . . . . . . . . 208
8.3 Possible partitionings of a Source structure . . . . . . . . . . . 209
8.4 Possible partitionings of a Goal structure . . . . . . . . . . . . . 215
9.1 Locative=Goal syncretism in Georgian . . . . . . . . . . . . . . 219
9.2 Possible syncretisms . . . . . . . . . . . . . . . . . . . . . . . . 226
xi
xii LIST OF TABLES
9.3 Impossible syncretisms . . . . . . . . . . . . . . . . . . . . . . . 226
9.4 Possible instantiations of an L=G=S syncretism . . . . . . . . . 227
9.5 Pattern of syncretism for the lexicalization of Location, Goal,
and Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
9.6 Possible syncretisms (repeated from Table 9.2) . . . . . . . . . 234
Abstract
This thesis offers a morphosyntactic analysis of the expressions of directed
motion cross-linguistically. I argue that the syntactic structure
underlying directional expressions is richer than previously assumed and
propose a decomposition of the commonly assumed Path head. The core
idea is that the Path projection splits into several heads and different
types of paths correspond to syntactic structures involving a different
number of these heads.
The proposed Path decomposition is based on a cross-linguistic investigation
of the morphological structure of directional expressions. The
data show that there are languages where Source paths are built on top of
Goal paths. In these languages, the Source expressions are formed by the
addition of a special element to the Goal expression, for which there is
evidence that it is an independent morpheme, thus excluding an accidental
containment relationship. Similarly, there are languages where Route
paths morphologically contain Source paths, and non-transitional and
delimited paths are formed on the basis of the corresponding transitional
path.
Following the cartographic approach, according to which the identification
of a morpheme indicates the presence of an independent syntactic
head corresponding to it, I propose a syntactic structure where Source
paths are built on top of Goal paths by the addition of a dedicated
syntactic head. Likewise, Route paths embed Source paths, and nontransitional
and delimited paths embed the corresponding transitional
path. This leads to a hierarchical sequence of heads of the shape: Place
< Goal < Source < Route < {Scale, Bound[ed]}, where each head dominates
the head to the left. The heads Scale and Bound can come on
top of any other Path head, thus deriving non-transitional and delimited
Goal, Source and Route paths, respectively. I also propose a semantic
function for each of these heads.
Following Chomsky’s (2001) Uniformity Principle, I assume that this
syntactic structure is universal across languages, i.e., the syntactic struc-
xiii
xiv ABSTRACT
ture of Source paths, for instance, contains the structure for Goal paths
also in languages where there is no morphological containment relation
between Source and Goal markers. Evidence for this comes from syncretism
patterns and the way the structure is partitioned in various languages,
as I highlight below.
I investigate the Spell-out of the decomposed Path structure and, following
the Nanosyntax theory of grammar, I propose that morphemes
lexicalize syntactic constituents of various sizes and configurations and
that lexical insertion is governed by the Superset and Elsewhere Principles.
Further, I adopt the idea of Spell-out driven movement, originally
proposed by Starke (2011), according to which a lexical entry can trigger
movement of a given constituent in order to create the right syntactic configuration
for the insertion of this entry. I investigate in detail this type
of movement triggered by lexicalization and explore what constraints apply
to it. In doing this, I develop a precise lexicalization algorithm and
test it against the spatial case system in Finnish.
The Spell-out model I adopt allows for a single morpheme to spell
out multiple syntactic terminals. Combined with the fine-grained internal
structure of Path, the model predicts the existence of various possibilities
of partitioning the syntactic structure underlying a given path, depending
on how many morphemes are employed and what portion of the sequence
they spell out. I explore the possible partitionings and illustrate them
with language data. In doing this, I identify the phenomenon of spurious
syncretisms, where a given lexical item is used in the lexicalization of two
or more types of path, crucially necessitating the use of a “supporting”
lexical item to express the syntactically more complex one(s). Thus,
apart from providing additional evidence for the decomposition of the
Path head into several heads, the various partitionings of the Path fseq
make it possible to offer a syntactic explanation for the observed spurious
syncretisms.
Furthermore, the decomposed Path structure and the Spell-out model
developed in this thesis lead to a prediction about possible and impossible
syncretisms and lexicalization patterns. I test these predictions against
the domain of syncretisms involving Location, Goal, Source and Route
paths, using as a basis several typological studies. I reach the conclusion
that the excluded syncretism patterns are indeed unattested, the
apparent counterexamples being an instance of spurious syncretism.
Abbreviations
1 - First person
2 - Second person
3 - Third person
abl - Ablative
acc - Accusative
adess - Adessive
all - Allative
cl - Clitic
d - Declarative
dat - Dative
def - Definite
det - Determiner
dist - Distal
egr - Egressive
el(at) - Elative
erg - Ergative
exc - Exclusive
fin - Finite
gen - Genitive
ill - Illative
imp - Imperative
inc - Inclusive
iness - Inessive
instr - Instrumental
loc - Locative
mot - Motative
neg - Negation
nom - Nominative
non-trans - Non-transitional
nv - Negative verb
sg - Singular
past - Past
perf - Perfect
perl - Perlative
pl - Plural
poss - Possessive
pres - Present
pret - Preterite
prol - Prolative
prog - Progressive
prt - Particle
prv - Perfective
Pu - Unrestricted particle
Pv - Verbal particle
sub - Subitive
subj - Subjunctive
sup - Superessive
term - Terminative
trans - Transitional
xv
xvi ABBREVIATIONS
Chapter 1
Introduction and outline of the
thesis
1.1 Background and object of study
This thesis is concerned with the syntax, semantics and cross-linguistic
typology of spatial expressions encoding motion, like the one illustrated
in (1).
(1) The boy ran into the house.
It is a common view that such expressions encompass two dimensions,
or components: a stative one, for which I adopt the term Location (also
called Place, Configuration, Orientation, Localiser), and a dynamic one,
for which I adopt the term Path (also called Direction, Mode, Modaliser)
(see Bennet 1975, Jackendoff 1983, van Riemsdijk and Huybregts 2002,
Kracht 2002, Creissels 2006, Zwarts 2008a, Svenonius 2010, den Dikken
2010, among others). The Location component involves the spatial relation
between the object being localized and another, usually bigger
object. The dynamic Path component is concerned with how an object
in motion moves with respect to a stationary object. In the example in
(1), the boy traverses a Path such that, at the beginning of the Path, he
is not at the house and at the end of the Path he is inside the house.
Thus, the type of Location is interior and the Path culminates at that
location.
Motivated by the presence of these two components, van Riemsdijk
(1990) proposed that that the syntactic structure underlying directional
spatial expressions contains a stative head PLoc (which I will call Place,
adopting the terminology in Jackendoff 1983 and Svenonius 2010, inter
1
2 INTRODUCTION 1.2
alia), encoding the particular locative relationship, and a dynamic PDir
head (Path in the terminology of Jackendoff 1983, Svenonius 2010) which
expresses the particular type of movement. This assumption was further
corroborated by the fact that many languages have directional expressions
made out of two independent elements, where each element was
taken to correspond to one of the two syntactic heads in the structure
for Paths (cf., in+to).
As the syntactic structure underlying directional expressions became
more and more detailed, in accordance with the general trend towards
finer-grained syntactic representations, the original [Path [Place]] structure
of van Riemsdijk (1990) was enriched by the addition of a myriad of
new projections. It is now a widely held view that the syntax of Paths
involves many other heads than just Path and Place: Deg[ree], Asp[ect],
Ax[ial]Part, Deix[is], etc. (argued for in the works of Koopman 2000,
Svenonius 2010, den Dikken 2010). The Path head became just one of
many other heads in the structure. Still, it remained the only head which
hosts directional elements, no matter what type of directed motion they
express — a Goal paths like to the house, or a Source path like from the
shop, to mention a few types.
In this thesis, I put the Path head under the knife and cut it into
smaller bits. The methodology I use is to first determine what kind of
paths exist in Chapter 2. I argue for the recognition of eight distinct
types of paths. To the best of my knowledge, this is the richest path
typology proposed until now (cf., Jackendoff 1983, Piñón 1993, Mel’čuk
1994, Kracht 2002, Zwarts 2008a). I then investigate how languages express
the various types of paths. In my investigation, I assume, following
the general guideline of the cartographic framework (Rizzi 1997, Cinque
1999; 2005, Cinque and Rizzi 2008), that morphological complexity is
indicative of syntactic complexity (Chapter 3). Thus, whenever we are
able to isolate a morpheme which adds a given meaning to the expression
it is a part of, this is an indication that the underlying syntactic
structure contains an independent head corresponding to that meaning.
In Chapter 4, I then turn to a study of the morphological composition of
the various types of paths cross-linguistically. The conclusion I reach is
that different paths are of different morphological complexity and, crucially,
subject to a subset-superset relationship. The discovery of this
fact provides the basis for my proposal.
1.2 CORE PROPOSAL 3
1.2 Core proposal
In few words, the main proposal of this thesis can be summarized as
follows: The Path head is not atomic, it has a reacher structure than
previously assumed. Specifically, I show that it can be maximally decomposed
into the sequence of heads presented in (2), which I take to
be universal across languages. Each head in the structure in (2) has a
particular semantic function, which I discuss in Chapter 5.
(2) {Scale, Bound}P
{Scale, Bound} RouteP
Route SourceP
Source GoalP
Goal ...
The various types of paths correspond to different syntactic structures.
For instance, a Source path has the structure in (3).
(3) SourceP
Source GoalP
Goal PlaceP
Place ...
While a Goal path has the structure in (4).
(4) GoalP
Goal PlaceP
Place ...
Thus, syntactically Source paths “contain” Goal paths. This containment
relationship is reflected also in the morphological make-up of Source
expressions in some languages, where the Source expression contains a
morpheme marking a Goal path. A language which illustrates this phe-
4 INTRODUCTION 1.3
nomenon is Imbabura Quechua. There, the Goal marker -man is contained
in the complex marker expressing Source -man-da, see (5).
(5) Goal and Source in Imbabura Quechua (data from Cole 1985:119)
a. Utavalu-man
Otavalo-all
ri-ni.
go-1
‘I go to Otavalo.’ (Goal)
b. Utavalu-manda
Otavalo-abl
shamu-ni.
come-1
‘I come from Otavalo.’ (Source)
Similar morphological containment relationships hold for other paths,
for example, non-transitional paths (towards), obtained by the application
of the Scale head, contain the corresponding transitional paths (to),
suggesting that the syntactic structure of the former is a superset of the
syntactic structure of the latter. Likewise, Route paths (via) morphologically
contain Source paths (from), which is also reflected by the structure
in (2).
1.3 Theoretical apparatus
The assumption that the decomposed Path structure in (2) is universal
raises the question of what happens in languages which have monomorphemic
markers to express paths involving multiple heads, e.g., a language
like Kham, where the apparently non-decomposable Source suffix
-ni corresponds to the syntactic structure in (3) involving three heads.
(6) Kham (Watters 2002)
kuwa-ni
well-abl
hai-ke-o.
pull.out-perf-3sg
‘He pulled him out of the well.’
To capture such mismatches between the number of morphemes in a Path
expression and the number of syntactic terminals in the corresponding
structure, I adopt the Nanosyntax framework developed at the University
of Tromsø (Starke 2005-2009, Ramchand 2008b, Bašić 2007, Fábregas
2007, Abels and Muriungi 2008, Muriungi 2008, Lundquist 2008,
Caha 2009b, Taraldsen 2010, Pantcheva 2010, for a representative collection
of papers see Svenonius et al. 2009), which I present in Chapter 6.
Nanosyntax assumes that the terminals in the syntactic representations
are submorphemic, in fact, each terminal node represents a single feature.
1.3 THEORETICAL APPARATUS 5
These features are ordered in a universal hierarchy called the functional
sequence (fseq). Thus, Nanosyntax fits naturally with the proposed finegrained
decomposition of the Path head, where the terminals are smaller
than morphemes in many languages. Further, Nanosyntax assumes that
a single morpheme can lexicalize syntactic structures comprising multiple
terminals, thus allowing for a mismatch between the number of terminals
and the number of morphemes in a Path expression.
In more technical terms, the lexicalization of multiple terminals is
achieved by assuming Phrasal Spell-out, an idea originally proposed by
McCawley (1968), and adopted by syntacticians working within Nanosyntax
(Starke 2009; 2011, Caha 2009a;b, Fábregas 2009) as well as in other
frameworks (Weerman and Evers-Vermeul 2002, Neeleman and Szendrői
2007). In the Phrasal Spell-out system, lexicalization targets nonterminal
nodes. A lexical entry can then be inserted straight into a
phrasal node, thus expressing all the features contained in it, as I show
in (7) taking as an example the Kham data in (6).
(7) SourceP⇒-ni
Source GoalP
Goal PlaceP
Place DP
In addition to Phrasal Spell-out, I adopt an idea recently put forward
in Starke (2011) and explored in more detail in Caha (2010b), according
to which a lexical entry can trigger an evacuation movement of the
syntactic node(s) which it cannot lexicalize. This movement creates the
right syntactic configuration for the entry to be inserted. In addition, it
creates new syntactic nodes — the nodes resulting from the adjunction
of the evacuated material. As an example, take again the Kham Ablative
suffix -ni, which spells out all the nodes in the structure in (7) without
the DP. Consequently, the DP has to extract, so that -ni can be inserted
at the phrasal SourceP node.
6 INTRODUCTION 1.4
(8) SourceP2
DP SourceP1⇒-ni
Source GoalP
Goal PlaceP
Place tDP
As a byproduct, the correct ordering of the elements is achieved —
the DP is linearized before the suffix -ni.
Chapter 7 is devoted entirely to the investigation of Spell-out triggered
movement. There, I present a detailed analysis of the lexicalization
of different Paths in three languages (Karata, Uzbek and Finnish).
I develop a precise Spell-out algorithm and show how it captures the
intricacies of the spatial systems in these languages.
1.4 Predictions
The decomposition of the Path head and the Nanosyntax model of grammar
that I adopt make a prediction regarding the syncretisms found
among the various types of path expressions, a subject investigated in a
number of recent works (Creissels 2006; 2008, Radkevich 2009, Nikitina
2009, Lestrade 2010). This thesis is a contribution to this topic, as it
devotes Chapter 8 and Chapter 9 entirely to the study of possible and
impossible syncretisms between Route, Source, Goal and Locative ex-
pressions.
The fine-grained Path structure allows us to capture the distinction
between what I call a spurious syncretism and a real syncretism. Spurious
syncretisms arise as the result of there being various possibilities to
partition the structure. For instance, a Source phrase with the structure
in (3) can be partitioned in four different ways, depending on how many
morphemes the Source expression consists of and which heads these morphemes
spell out. This is illustrated in (9), where m1, m2, and m3, are
variables over morphemes.
1.5 PREDICTIONS 7
(9) Source Goal Place
language A
m1
language B
m1 m2
language C
m1 m2 m3
language D
m1 m2





















In language B, the Source phrase is formed by a combination of two
morphemes: m1 which spells out the Source head and m2 which lexicalizes
the Place and Goal heads. Morpheme m2 thus lexicalizes a Goal structure
and can therefore express a Goal path. Importantly, it does not syncretize
Goal and Source. It always spells out a Goal structure, even when it is
part of a Source phrase. In other words, although m2 participates in the
Source expressions of language B, it is an unambiguously Goal morpheme
and therefore it would be incorrect to classify it as a morpheme that
is syncretic between Goal and Source. In Chapter 8, I investigate the
various possible partitionings of the decomposed Path structure and come
across many instances of morphemes that participate in more than one
type of path. I call this phenomenon spurious syncretism, to distinguish
it from real syncretism where a genuine ambiguity is involved.
Real syncretisms are the topic of Chapter 9. The Nanosyntax framework
allows such syncretisms to be defined in terms of structural ambiguity
— a lexical item can spell out a given constituent and any subconstituent
of it (known as the Superset Principle). Combined with a
principle which requires that the most highly specified entry is to be
preferred (the Elsewhere Condition), the system prohibits syncretisms of
the type ABA, that is, syncretisms of two categories across a distinct intervening
category. This rules out syncretisms between the spatial roles
Location and Source to the exclusion of Goal, Goal and Route to the exclusion
of Source, and Location and Route to the exclusion of Goal and
Source. In addition, I argue that a syncretism between Goal and Source
is pragmatically excluded due to the specific semantics of the Goal and
Source heads, proposed in Chapter 5. I then test the prediction against
cross-lingustic data and conclude that the expected asymmetry in the
distribution of the syncretism patterns across languages really exists, the
apparent counterexamples being reanalyzed as instances of spurious syn-
cretism.
8 INTRODUCTION 1.5
1.5 How to read the thesis
This thesis is written as one coherent whole and is best read in its entirety.
Nevertheless, I have attempted to enable selective reading and
have organized the exposition in three main parts. Readers who want to
learn about the syntax and semantics of the decomposed Path structure
are invited to read Part I, where I develop the main idea of this thesis.
Readers interested in Nanosyntax and Spell-out driven movement can
read Part II. In this part, I present the Nanosyntax framework starting
with a lay-out of the basic tenets of the theory and ending with a detailed
exploration of the specifics of Spell-out. Hence, Part II can be
read both by readers who are not well acquainted with Nanosyntax, as
well as those who work within this framework. The third part is mostly
oriented towards a cross-linguistic examination of lexicalization patterns
and might be of interest to those readers who are involved in the study
of syncretisms in the domain of spatial expressions.
Part I
Path expressions: typology,
syntax, and semantics
9
Chapter 2
Types of paths
2.1 Introduction
Directional spatial expressions across languages express various kinds of
motion. A couple of them are exemplified by the sentences below.
(1) a. The children walked along the river.
b. The mosquito flew away from the lamp.
c. The frog jumped into the lake.
The first example (1a) describes a motion proceeding along a trajectory
which is roughly parallel to a given landmark, in this case the river. There
is no specification of the starting point or the end-point of the journey.
Nor do we have information about the direction of the movement — it
could be away from the origin of the river towards its mouth, or the other
way around.
The second example (1b), on the contrary, provides us with information
about the direction of movement. It starts close to the landmark
(the lamp), and proceeds in a manner such that the distance between the
mosquito and the lamp increases. But we cannot identify the precise location
of the starting point and the end-point of the movement although
we do have some information about the starting point — we know that it
is closer to the lamp than any subsequent point in the trajectory followed
by the insect. Still, it could be, for instance, above the lamp or below it.
The third example (1c) conveys the direction of motion, too. Here
we know that the movement of the frog is directed towards the lake.
In addition, we can pin down the end-point of the movement: it is in
the lake. However, the precise starting point remains vague, negatively
defined as being not in the lake.
11
12 TYPES OF PATHS 2.1
The established term for such movement trajectories is path. Thus,
each of the prepositions in (1) encodes a given type of path. The entity
which moves is commonly referred to as the Figure (the children, the
mosquito, the frog), while the entity which is stationary and with respect
to which the Figure moves or is located is called the Ground (the river,
the lamp, the lake) (Talmy 2000). Other terms for Figure and Ground
used in the literature on paths are trajector and landmark, respectively
(Lakoff 1987, Langacker 1987).
Note that each of the paths in (1) relates to some location. For
instance, in (1a) each of the points of the path are located at the river.
In (1b), the location of the starting point of the path is not precisely
defined, but we know that if the path is extended towards its beginning,
then the starting point will end up being at the lamp. In (1c) the endpoint
of the path is in the lake.
Due to this relation between path and location, Jackendoff (1983)
proposes that the conceptual structure of path-denoting phrases can be
broken down into two ingredients — path and place, the latter associated
with the location. Formally, the conceptual structure underlying
path expressions is respresented as shown below:
(2) [Path path-function [Place place-function [Thing y ]]]
The Path-functions can be to, from and via, the Place-functions can be
in, on, under, etc. The thing is the reference object, or the Ground.
A Place-function takes as an argument a thing and gives as an output
a place. The Path-function takes as an argument place and returns
a path. Below, I show how this multi-layered structure applies to the
path expressions in (1).
(3) a. [Path via [Place at [Thing the river ]]]
b. [Path from [Place at [Thing the lamp ]]]
c. [Path to [Place in [Thing the lake ]]]
Jackendoff further proposes a classification of paths. He suggests that
there are three main types of paths. The first one is bounded paths.
They include source paths, typically encoded by the English preposition
from, and goal paths, for which the usual preposition is to. The characteristic
property of bounded paths is that the place is an extreme point
of the path – either its beginning, as in Source paths, or its end, as in
Goal paths. The sentence in (1c) provides an example of a bounded Goal
path. The second type of paths is called directions. Directions, too, can
be subdivided into two subtypes: source directions encoded, for instance,
2.2 PATHS WITH TRANSITION 13
by away from, and goal directions, expressed by the preposition towards.
The difference between bounded paths and directions is that, in the case
of the latter, the place is not a point of the path, but would be if the
path were extended some unspecified distance. We find such a Source
direction in our example in (1b). The last type of paths is routes, represented
by the prepositions along, through and others. Here the place
falls on some intermediate points of the path and the extreme point are
left unspecified. The path expression in (1a) belongs to this type. Jackendoff
distinguishes thus five types of paths, which can be schematically
represented as follows:
Paths
bounded routes directions
past, along
goal paths source paths goal directions source direction
to from towards away from
Figure 2.1: Jackendoff’s (1983) typology of paths
In this chapter, I investigate the typology of paths. I propose that
paths can be classified according to three properties: transition, orientation,
and delimitation. There are paths with and without transition(s),
paths with and without orientation, and paths with and without delimitation.
The interaction of these properties leads to eight types of paths,
as opposed to the five types proposed by Jackendoff (1983). The difference
is due to (i) the additional property of delimitation, and (ii) a
distinction I draw between two types of Route paths, while Jackendoff
has only one.
2.2 Paths with transition
Let us begin with the last path expression in (1), repeated below as (4).
(4) The frog jumped into the lake.
The path expressed by the prepositional phrase is characterized by several
properties. First, it has a direction, namely, it is oriented in the direction
of the lake. The lake is thus the Goal of the movement of the frog.
Second, the end-point of the path is defined as being in the lake. Third,
the starting point of the path, although not precisely located, is not in
14 TYPES OF PATHS 2.2
the lake. According to Jackendoff’s (1983) classification, the path in (4)
is therefore a bounded goal path. Zwarts (2005; 2008a) calls such paths
non-cumulative Goal paths and represents them graphically as in (5),
where the plusses indicate location in the lake, and the minuses indicate
location not in the lake. The points 0 and 1 mark the starting point and
the end-point of the path, respectively.
(5) Goal path
− − − − − + + + + +
0 1 Zwarts (2008a)
As can be seen from the graphic representation in (5), the into the lake
path has a two stage structure: the first stage is not located in the lake,
while the second stage is located in the lake. The path thus contains a
transition from one spatial domain to a complementary spatial domain.
For this reason, I will call that types of paths transitional. The transitional
path in (5) is in addition characterized by the fact that there is a
locative condition on the end-point of the path, namely, it has to be in
the lake.
We find the same type of transition in the path expressed in the
following sentence.
(6) The frog jumped out of the lake.
This path is in a sense the opposite of the path expressed in (4). Here,
the location in the lake is not the end-point of the movement of the
frog, but its starting point. The lake is thus the Source of the motion.
Interestingly, by pinning down the starting point of the path, we lose the
precise definition of the end-point of the path. Now we know that the
end-point is not in the lake, but a negative definition is again the only
information we are left with. The Source path in (6) can be thus seen as
the reverse of the Goal path in (5), as it includes a transition too, but,
contrary to Goal paths, imposes a condition on the initial portion of the
path. This kind of paths are visualized by Zwarts (2005; 2008a) as in
(7).
(7) Source path
+ + + + + − − − − −
0 1 Zwarts (2008a)
There is no restriction that there be only one transition per path. Consider
the following example.
2.2 PATHS WITH TRANSITION 15
(8) The boy ran past the tree.
The path represented by the directional expression in (8) has some intermediate
points at the tree. The starting point and the end-point of the
path though remain unknown. Such paths are called routes by Jackendoff
(1983) and this is the term I will adopt.1
According to Zwarts (2005;
2008a), such paths involve a condition on their middle part and can be
graphically represented as follows.
(9) Route path
− − − − + + + + − − − −
0 1 Zwarts (2008a)
A comparison between (7) and (5), on the one hand, and (9), on the other
hand, reveals one difference and one similarity. The difference is that in
the denotation of Route paths, there are two transitions, while Goal and
Source paths have exactly one. The similarity is that in all three kinds of
paths, there is one unique positive phase – the portion of the path where
the locative relationship between the Figure and the Ground obtains.
As a matter of fact, Route paths look as if they are composed of a
Goal path concatenated with a Source path in this order.
(10) Route path
[− − − + + +] [+ + + − − −] = [− − − + + + + − − −]
0 1 0 1 0 1
Goal Source Route
Still, there is only one positive phase. Interestingly, it seems that no
natural language preposition is composed from the concatenation of a
Source path with a Goal path, which would result in two positive phases.
(11) *[+ + + − − −] [− − − + + +] = [+ + + − − − − + + +]
0 1 0 1 0 1
Source Goal non-existing P
What would a preposition corresponding to the representation in (11)
mean? Recall that the Route preposition past in (8) encodes a path
traversed by the running boy where the boy is not at the tree initially,
1
Sometimes the term path preposition is used to mean route preposition, particularly
in grammar descriptions and computational literature. In the terminology
adopted in this thesis, path prepositions is a cover term for all directional spatial
prepositions (onto, through, from) and Route prepositions are just a subtype of path
prepositions.
16 TYPES OF PATHS 2.3
then is at the tree for some time and then isn’t at the tree again. Let us
imagine a preposition of the type in (11) and call it *tsap, following the
intuition that it is in a way the opposite of the Route preposition in (10).
The boy ran *tsap the tree should mean: the boy was first at the tree,
then not at the tree and after that he was at the tree again (imagine a
situation in which the boy ran away from the tree and then returned to
the tree again). To the best of my knowledge, no language has such a
“return” preposition. In Chapter 5, Section 5.5, I offer a suggestion why
this should be so.
To sum up what has been said so far, I have discussed three types
of transitional paths: Source paths, Goal paths and Route paths, which
have in common the fact that in their denotation they have just one
positive phase. When it comes to the number of transitions, Source and
Goal paths have one transition. Route paths have two transitions which
gave rise to the idea that they are “more complex” and composed out of
a Goal path concatenated with a Source path in this order. Thus, in a
way, the “mono-transitional” Source and Goal paths form a natural class
to the exclusion of the “bi-transitional” Route paths, a difference which
comes up again in Section 2.4.
2.3 Paths without transition
It is not always the case that the place to which the path refers falls on
the path. For instance, in John ran towards the house, the location at the
house does not necessarily coincide with any of the points of the path, but
would if the path were extended some unspecified distance. Jackendoff
calls this class of paths directions. The example below illustrates the
distinction between transitional paths and directions.
(12) a. John ran to the house. (transitional path)
b. John ran towards the house. (direction)
(Jackendoff 1983:165)
In (12a), John has reached the house, that is, the endpoint of John’s
path is at the house. In (12b), John probably hasn’t reached the house,
therefore, the location at the house is not a point on the path. However,
it would be if the path were extended. Notice that the transitional path
in (12a) and the “directional” path in (12b) have something in common;
namely, they are both Goal-oriented, i.e., oriented towards reaching a
final location.
2.3 PATHS WITHOUT TRANSITION 17
Transitional Source paths also have corresponding Source-oriented
directions, as shown below.
(13) a. John ran from the house. (transitional path)
b. John ran away from the house. (direction)
(Jackendoff 1983:165)
The directional expression in (13b) encodes the same type of path as
the sentence in (1b) The mosquito flew away from the lamp. Here, the
location at the house (or at the lamp) is not a point of the path, but
would be if the path were extended towards its beginning.
Zwarts (2005; 2008a) discusses such types of paths and calls them
comparative following the intuition that the distance to the reference object
decreases/increases monotonically or, put informally, each consecutive
location of the Figure is nearer to/further away from the Ground.
Graphically, Zwarts (2008a) represents towards-paths as shown below,
where the deeper shade of gray corresponds to a nearer location to the
house.
(14) towards-path
+ + + + + + + + +
0 1 Zwarts (2008a)
The Source-oriented counterpart of the path in (14) is represented as
follows, if we adopt Zwarts’ visualization.
(15) away from-path
+ + + + + + + + +
0 1 Zwarts (2008a)
The Zwartsian graphic representation of the towards and away from
paths in (15) and (14) involves plusses, which, in (5), (7) and (9), were
employed to encode the fact that the Figure is located at the Ground. As
suggested by Jackendoff (1983), however, the location to which a towards
path and an away from path refers does not fall on the path. That is,
in the case of the path expressed in John ran towards the house, John is
not at the house at the end-point of the path p(1). He is, though, surely
closer to the house at p(1) than he was at the beginning of the path,
p(0). In order to reflect this fact, I suggest to graphically represent such
kinds of paths as a sequence of minuses in order to indicate that at no
point in the path, is the Figure located at the Ground. The representation
for a towards and an away from path will then be as in (16) and
18 TYPES OF PATHS 2.3
(17), respectively, where the deeper shade of gray on a minus indicates a
greater distance from the Ground object.
2.4 ORIENTATION OF PATHS 19
(16) towards-path
− − − − − − − − −
0 1
(17) away from-path
− − − − − − − − −
0 1
I will call this type of paths non-transitional. The path is (16) is
a non-transitional Goal path, and the path in (17) is a non-transitional
Source path.
Apart from Goal and Source-oriented non-transitional paths, there
are non-transitional paths that lack orientation. Such is the path in (18),
repeated from (1a), which represents a non-transitional Route path.
(18) The children walked along the river.
The graphic representation of such a path will be the one in (19), as
suggested by Zwarts (2008a), where every plus indicates a location at
the river.
(19) along-Path
+ + + + + + + + +
0 1 Zwarts (2008a)
The reason for such a representation is that an along-path imposes the
same locative condition on all the points of the path. Thus, any of the
points in the path along the river are seen as being at the river.
2.4 Orientation of paths
In the previous sections, I discussed the division of paths according to
the property of having at least one transition. As has already become
apparent, paths differ also on the basis of their orientation. There are,
on the one hand, non-oriented paths, namely Route paths, where there
is no indication as to the direction of the movement. On the other hand,
there are oriented paths, where we know in what direction the movement
proceeds. The Source and Goal paths discussed above are examples
of oriented paths. Oriented paths involve some asymmetry concerning
the two extreme points of the path. For instance, in Goal paths, the
end-point of the path is related to a particular location, while the precise
location of the starting point is unknown. The reverse holds for Source
20 TYPES OF PATHS 2.4
paths. Route paths instantiate the non-oriented path type. The characteristic
property of non-oriented (i.e., Route) paths is that both extreme
points are equally defined. In transitional Route paths both the starting
and the end-point are not located in the region the path relates to, recall
(9). In non-transitional Route paths, both the starting point and the
end-point are located in the region the path relates to, see (19).
Source and Goal are not the only orientations a path can have. Some
other kinds of orientation we find across languages are up-down, hitherthither,
and north-south.
(20) a. The alpinist climbed up.
b. They went north.
This thesis deals only with the Source-Goal orientation of paths. The
reason for this is that the Source-Goal orientation appears to be more
basic than the up-down, hither-thither, north-south orientation and is
usually linguistically encoded in a different way. In English, for instance,
one uses particles or adverbs to express orientation along the up-down,
north-south, etc. axes, while Goal and Source orientation are encoded by
prepositions. An even clearer illustration is provided by the Daghestanian
language Tabasaran where the other orientation types are more complex
than the Source-Goal orientation.
Like many other Daghestanian languages, Tabasaran has a very elaborate
system of spatial cases (Spivak 1990). Those are presented in Table
2.1, which summarizes data from Magometov (1965), Hanmagomedov
(1967), and Comrie and Polinsky (1998). Tabasaran has seven “locative
series” markers, which encode locative relations like in, behind, under,
etc. These locative markers attach to an Ergative-marked noun.
The Allative (Goal) marker -na and the Ablative (Source) marker -an
attach to a noun suffixed by one of the locative series markers to produce
complex directional expressions meaning to under the mill or from
the mountain, see (21) and (22).
(21) räGy
-ni-kki-na
mill-erg-under-all
‘to under the mill’
(22) daG-ˇZi-l-an
mountain-erg-on-abl
‘from the mountain’
The data below shows how Tabasaran expresses up-down orientation,
2.4 ORIENTATION OF PATHS 21
Series Goal Source
at -xy
-xy
-na -xy
-an
in -’ -’-na -’-an
behind -q -q-na -q-an
under -kki -kki-na -kk-an
on -’il -’in-naa
-l-an
between -γy
-γy
-na -γy
-an
vertical attachment -k -k-na -k-an
a
The Allative of the on-series is -’in-na rather than
’il-na due to regressive assimilation (in-na < il-na).
Table 2.1: Spatial case system in Tabasaran
(23), and hither-thither orientation, (24) (data from Magometov 1965:119).2
(23) Up-down
a. daG-ˇZi-l-an-Gina
mountain-erg-on-abl-up
‘from the mountain up(wards)’
b. daG-ˇZi-l-an-kkina
mountain-erg-on-abl-down
‘from the mountain down(wards)’
(24) Hither-thither
a. daG-ˇZi-l-an-mina
mountain-erg-on-abl-hither
‘from the mountain hither’
b. daG-ˇZi-l-an-tina
mountain-erg-on-abl-thither
‘from the mountain thither’
A morphological analysis of the up-down and hither-thither expressions
shows that the Tabasaran orientation markers, attached to the Ablative
form of the noun mountain, are not atomic. For instance, the marker
kkina expressing downward orientation (see (23b)) is composed out of
the under-series marker kki, suffixed by the Allative (Goal) morpheme
-na, thus deriving the meaning to under. It is basically the same complex
suffix that is attached to the ergative marked noun räGy
‘mill’ to derive
2
Due to the ambiguity of the Russian translation, I do not know whether the
up-down or hither-thither markers in Tabasaran are transitional or non-transitional.
However, this does not bear on the point made here.
22 TYPES OF PATHS 2.5
the expression to under the mill in (21). Hence, the downward-marker in
Tabasaran is structurally different from the Goal and Source case endings,
in that the latter are apparently non-decomposable, while the former is
morphologically complex and contains the Goal morpheme. The same
can be argued to hold of the other orientation markers in Tabasaran, all
of which seem to contain the Allative ending -na.3
I suggest that this is an important fact, indicating that Goal and
Source orientation are more “linguistically primitive” in the sense that
they are encoded by monomorphemic elements, while the other orientation
markers are semantically and morphologically composed of more
basic elements and are hence structurally complex. Later, in Chapter 4,
I will modify my statement, in that I will argue that Source markers are
not as irreducible as suggested here, but can be shown to contain a Goal
structure. Still, I suggest that Goal and Source are the basic orientations
and, in the path typology developed in this chapter, I will abstract away
from the up-down, hither-thither, and north-south orientations, which
are derived by means of combining more primitive elements.
2.5 Delimited paths
There is one more type of paths which finds grammatical expression
in many languages. This is the type of path encoded by the complex
preposition up to in the English sentence in (25).
(25) The boy ran up to the house.
In (25), the boy traverses a path that is oriented towards the house
and stops right before it. Following the terminology used in grammar
descriptions, I will call this kind of paths terminative. The Terminative
path in (25) is quite similar to the one we find in (26).
(26) The boy ran to the house.
Given the two differential properties of paths, that I have established so
far – transition and orientation – the sentences in (25) and (26) come out
as synonymous. The reason is that, in both cases, we have a transitional
Goal path, that is, a path which leads from a location not at the house
3
It is possible to also isolate the morpheme -i- in the composition of the orientation
markers, assuming that it gets deleted for phonological reasons when combined with
kki- ‘under.’ This does not change the argument made here, as, in this case again,
the orientation marker contains the Allative morpheme.
2.5 DELIMITED PATHS 23
to a location at the house. The meaning difference between the two
sentences is indeed very subtle, yet clearly perceptible. In the case of up
to, it is made explicit that the house is the limit of the boy’s running.4
Terminative paths have received scarce attention from linguists. Kracht
(2001; 2002) briefly discusses them and suggests that the contrast between
terminative paths and “simple” transitional Goal paths is purely
aspectual: they differ “in the strength of association with telicity.” Thus,
according to Kracht, both transitional and Terminative paths are telic, as
also evidences by the time-adverbial test in (27), but Terminative paths
have a stronger requirement on telicity.
(27) a. The boy ran up to the house in ten minutes/*for ten min-
utes.
b. The boy ran to the house in ten minutes/*for ten minutes.
Winter (2006) assumes a more formal approach to the problem. He tries
to formalize the contrast between the two types of paths by suggesting
that the key difference between them is the presence or lack of closure.
Winter supports this idea by drawing a parallel between closure of paths
and closure of adjectival scales. He builds on previous work by Rotstein
and Winter (2004), where the authors test for closure of adjectival scales
by application of the almost-modification test: closed-scale adjectives
allow modification by almost, while open-scale adjectives do not, see (28)
(examples from Rotstein and Winter 2004:265)
(28) a. The explanation is almost clear. (closed scale)
b. *The explanation is almost unclear. (open scale)
The application of the almost-test to paths is not trivial. The reason
is that almost-modification of paths can give rise to two interpretations:
one is the counterfactual interpretation, where the Figure almost started
traversing the path, but never really initiated the motion. The other interpretation
is called scalar and this is when the Figure started traversing
the path and got very close to finishing it, but never reached the final
point. According to Winter (2006), it is this second, scalar, interpreta-
4
As pointed out by Peter Svenonius (p.c.), English up to is, as a matter of fact,
ambiguous between a Terminative path and a simple (non-delimited) Goal path to
the house, where the house is at a higher altitude than the boy at the beginning of
the running event (imagine that the boy is at the foot of a hill when he starts running
and the house is on top of the hill). I will be concerned only with the Terminative
meaning of up to here, referring the reader to Svenonius (2010) for an analysis of the
syntactic structure underlying the second reading.
24 TYPES OF PATHS 2.5
tion, that indicates the presence of closure. He presents data from Hebrew
and Dutch and the results show that Terminative paths are acceptable
with the scalar interpretation of almost (as well as the counterfactual
interpretation), while non-terminative transitional Goal paths allow only
the counterfactual interpretation.
(29) Hebrew
a. dan
Dan
kimPat
almost
rac
ran
la’agam.
to.the.lake
‘Dan almost ran to the lake.’ (counterfactual/?scalar)
b. dan
Dan
kimPat
almost
rac
ran
Pad
until
ha’agam.
the.lake
‘Dan almost ran to the lake.’ (counterfactual)
‘Dan ran and almost reached the lake.’ (scalar)
(30) Dutch
a. Dan
Dan
rende
ran
bijna
almost
naar
to
het
the
meer.
lake
‘Dan almost ran to the lake.’ (counterfactual)
b. Dan
Dan
rende
ran
bijna
almost
tot
until
het
the
meer.
lake
‘Dan ran and almost reached the lake.’ (scalar)
On the basis of this test, Winter proposes that the spatial path associated
with prepositions like English to, Hebrew le, or Dutch naar is open. By
contrast, the path associated with Hebrew Pad, Dutch tot (and presumably
English up to, although not specifically mentioned) is closed. The
two types of paths are shown in Figure 2.2.
to, le, naar
Pad, tot
-
Figure
2.2: Open and closed paths (Winter 2006)
This is a somewhat surprising result and it is difficult to find independent
motivation for it from the semantics of the prepositional phrases.
What it means is that, in Goal paths like (The boy ran) to the house, the
endpoint (the house) is not included in the path. This idea that to-paths
are open goes against Zwarts’ (2005) definition of Goal paths. According
to Zwarts, the prepositional phrase to the house denotes the set of paths,
such that there is an interval I⊂[0,1], that includes the point 1 and that
2.5 DELIMITED PATHS 25
consists of all the indices i∈[0,1] for which p(i) is at the house. If we
include in the semantics of the PP to the house the fact that it has to
refer to an open path, we arrive at a contradiction: the PP to the house
will then denote the set of paths, such that there is an interval I⊂[0,1],
that includes the point 1 and excludes the point 1 and that consists of all
the indices i∈[0,1] for which p(i) is at the house.
Although I reject the idea that non-terminative Goal paths do not
include the location in their final point, I agree with Winter’s intuition
that the end-point of a Terminative path is more salient than the endpoint
of a non-terminative Goal path. In my view, the difference between
the two types of paths is that Terminative paths explicitly state that the
Ground is the boundary of the movement. Thus, in The boy ran up to
the house, the boy runs as far as the house, that is, once he reaches it, he
stops. In The boy ran to the house, the boy might stop at the outer wall
of the house, as well as continue running along the wall, or even enter
it. The same meaning difference can be observed for other languages.
Consider, for instance, the Persian minimal pair in (31).
(31) a. be-r-im
subj-go-1pl
be
to
park!
park
‘Let’s go to the park (and enter it)!’
b. be-r-im
subj-go-1pl
ta
until
park!
park
‘Let’s walk as far as the park (and stop there)!’
I suggest that the difference between Terminative and non-terminative
Goal paths is that Terminative paths set the end of the path at the
first point where the location to which the path relates is reached. Nonterminative
paths, by contrast, allow the path to “continue,” while still
being wholly contained in the location specified by the Ground. Graphically,
we can reflect this distinction by including just one plus in the
positive phase of a Terminative path, thus capturing the intuition that
once the location is obtained, the path stops. The positive phase of
non-terminative Goal paths, by contrast, contains a sequence of plusses,
reflecting the fact that the path can “continue” within the location.
(32) up to-path
− − − − − − − − − +
0 1
26 TYPES OF PATHS 2.5
(33) to-path (repeated from (5))
− − − − − + + + + +
0 1
The single plus in (32) renders the end-point of the Terminative path
unique: this is the only point where the location specified by the Ground
holds. This can be connected to Kracht’s (2001, 2002) idea of a “strong association
with telicity,” since, in Terminative paths, the end-point stands
out.
To sum up, I suggest that Terminative paths are delimited at the
end-point, that is, they make explicit reference to the upper bound of
the path. This is the result of the uniqueness of the end-point in a
Terminative path which becomes accentuated by virtue of being the only
positively located point.
So far, I have discussed only Goal-oriented delimited paths. However,
knowing that Source paths are the opposite of Goal paths, we expect that
there are “reversed Terminative paths” – Source paths that have a delimitation.
This delimitation, however, will have to hold of the beginning of
the Source path, because it is the positively located point. This will give
rise to a path meaning starting from (as opposed to delimited Goal paths
meaning ending at). A nice illustration of this type of path comes from
the Permic languages. Alongside a “regular” Source case, which expresses
motion away from the Ground, the languages from this branch have a
special Egressive case, which, too, encodes a path leading away from
the Ground, but there is the additional connotation that the Ground is
the starting point for the motion (Csúcs 1998, Hausenberg 1998, Winkler
2001). I present here an example from the language Komi-Permyak,
which has such an Egressive case, in addition to an Elative case (data
from Lytkin 1962, my glossing)
(34) a. Célj
adj
children
lóktënï
come
škóla-išj
.
school-elat
‘The children are coming from school.’
b. Volgograd-šj
anj
Volgograd-egr
Elj
ba
Elba
vá-ëdz
river-term
‘from Volgograd to Elba river’
On the basis of the function attributed to the Egressive case in the source
cited above, I suggest that the Egressive case morpheme in (34b) marks
a path beginning in Volgograd and sets this location as the lower bound
of the movement. Graphically, then, this path can be represented as in
(35), contrasting with the non-delimited Source path shown in (36) in
2.5 DELIMITED PATHS 27
that the positive phase contains just one point.
(35) starting from-path
+ − − − − − − − − −
0 1
(36) from-path
+ + + + + − − − − −
0 1
Summing up, Terminative and Egressive paths differ from their nondelimited
counterparts in the number of plusses in their positive phase.
The positive phase in these paths is either at the end or at the beginning
of the path. Under this assumption, transitional Route paths are not
expected to have a delimited counterpart. Recall that, in Route paths,
the positive phase is in the middle of the path. Therefore, the extreme
points 0 and 1 are not in a locative relation with the Ground, hence, the
Ground cannot be set as the initial or final boundary for any of them.
Even if the positive phase of a Route path contains just one plus as in
(37), it will not be a positively defined extreme point and therefore it will
not be interpreted as a limit for movement.
(37) − − − − + − − − − −
0 1
A parallel reasoning leads to the conclusion that a Source path can be
only lower bound and a Goal path can be only upper bound. That is,
the limit for movement of a delimited Source path is its starting point,
because the location holds of this point, and the limit for movement of
a delimited Goal path is its end-point, as it is the “positively located”
point. As a result, only transitional paths that have orientation can also
have a boundary.
I labeled paths of the Terminative and Egressive type delimited
path, contrasting them with the non-delimited Goal and Source paths
expressed by the prepositions to and from.5
Delimited paths are thus
transitional paths that explicitly indicate a boundary of the path – the
left boundary for Source-oriented paths, and the right boundary for Goal
5
Given that delimited paths make reference to boundaries, a more appropriate
term is probably bounded and unbounded paths (see Depraetere 1995 for a comprehensive
discussion of (un)boundedness). These labels have been, however, used by
linguists to refer to types of paths that are not Terminative or Egressive. For instance,
Jackendoff (1983) uses the term bounded for the type of paths I call transitional. To
avoid confusion, I adopt the term delimited.
28 TYPES OF PATHS 2.6
oriented paths. Both delimited and non-delimited paths involve a transition
from a positive to a negative phase or vice versa, but the positive
phase in delimited paths involves just one point where the location holds,
and this point is construed as the boundary. By contrast, the positive
phase of non-delimited paths can contain a sequence of points where the
locative relation between the Figure and the Ground obtains.
2.6 The two types of Route paths
As outlined above, Jackendoff (1983) argues for three major types of
paths: bounded paths, directions and routes. The first two types are
subdivided into source and goal paths and source and goal directions.
From a Zwartsian perspective, which is also the view adopted in this
thesis, the difference between a goal path and a goal direction is that
the former encodes a transition, while the latter does not. The same
distinction holds of source paths and source directions. The third type of
paths encompasses route paths, like the ones encoded by the prepositions
past and along.
As it stands, Jackendoff’s typology blurs the difference between the
paths expressed by these two preposition – they both fall under the type
of route paths (see Figure 2.1 on p.13). They are, however, different, as
also suggested by Zwarts. Specifically, a past-path contains transitions,
while an along-path lacks transitions. This fact is further corroborated by
the aspectual test involving temporal adverbials (Verkuyl 1972, Dowty
1979). When an atelic verb is combined with a transitional path PP,
the resulting predicate is telic, hence compatible with the time-frame
adverbial in an hour, (38). With a non-transitional path PP the predicate
remains atelic, as non-transitional paths have no impact on the telicity of
the verb. Therefore, the predicate will be compatible with the time-span
adverbial for an hour, (39).
(38) a. The boy ran to the store in an hour.
b. *The boy ran to the store for an hour.
2.6 THE TWO TYPES OF ROUTE PATHS 29
(39) a. *The boy ran towards the store in an hour.
b. The boy ran towards the store for an hour.
Applied to PPs headed by past and along, the temporal modification test
shows that the former gives rise to telic predicates, (40), while the latter
gives rise to atelic predicates, (41).
(40) a. The boy ran past the tree in one minute.
b. *The boy ran past the tree for one minute.6
(41) a. *The children walked along the river in an hour.
b. The children walked along the river for an hour.
As shown in the examples (38) to (41), the temporal modification test
indicates that past-paths pattern with bounded paths (in Jackendoff’s terminology)
and along-paths pattern with directions. Translated into the
terminology adopted in this thesis, past-paths pattern with transitional
paths and along-paths pattern with non-transitional paths. This strongly
suggest a further division of Jackendoff’s route paths into “transitional
routes” and “non-transitional routes.” In this terminology, the path past
the tree in (40) is a transitional route path. The path along the river in
(41) is an non-transitional route path.
The proposed classification is very much in line with Piñón’s (1993)
path typology. Piñón argues extensively for the recognition of two types
of Route paths – bounded and unbounded (in Piñón’s terminology). He
bases his proposal on the observation that some English Route prepositions
are compatible with both in an hour and for an hour modification,
as also noted by Declerck (1979).
(42) a. The insect crawled through the tube {for two hours / in two
hours}.
b. Mary limped across the bridge {for ten minutes / in ten
minutes}.
As evidenced by the results of the adverbial modification test in (38)
and (39), time-frame adverbials are good with bounded (i.e., transitional)
paths and time-span adverbials are good with unbounded (nontransitional)
paths. According to this test, the prepositions in (42) can
express both a transitional and a non-transitional Route path.
5
This sentence is grammatical under an iterative reading where the boy runs a
couple of times past the Ground and the whole macro-event lasts one minute (P.
Svenonius, p.c.).
30 TYPES OF PATHS 2.7
To recapitulate, I fully agree with Piñon in his claim that Route
paths come in two varieties: transitional (bounded) Routes and nontransitional
(unbounded) Routes. Transitional Route paths are those
Route paths that have both their extreme points outside the reference
location. Non-transitional Route paths are homogeneous with respect to
the reference location, that is, each point of the path is in the same local
configuration with the Ground as any other.
2.7 Classification of paths
So far, I have explored the types of paths and the properties that differentiate
them. Now it is time to systematize them and I propose the
classification of paths presented in Table 2.2.
Oriented Non-oriented
Goal Source Route
Withtransition
Transitional
Cofinal Coinitial Transitive
to X from X past X
−−−−++++ ++++−−−− −−−+++−−−
0 1 0 1 0 1
Delimited
Terminative Egressive
up to X starting from X
−−−−−−−+ +−−−−−−−
0 1 0 1
Notransition
Approximative Recessive Prolative
Non- towards X away from X along X
transitional −−−−−−−−− −−−−−−−−− +++++++++
0 1 0 1 0 1
Table 2.2: Classification of paths
In Table 2.2, the paths are classified according to two main properties.
The first main property is the presence or lack of orientation, that is,
whether the path has an inherent direction or not. According to this
property, Paths can be oriented or non-oriented. The oriented paths are
subdivided into Source-oriented or Goal-oriented. Source paths can be
seen as “reversed” Goal paths (and vice versa). This is illustrated in
Figure 2.3.
2.7 CLASSIFICATION OF PATHS 31
house
6
into the house
house
?
out of the house
Figure 2.3: Source paths as reversed Goal paths
Unlike the oriented Goal and Source paths, the non-oriented (Route)
paths have no inherent directionality and therefore they can be reversed
without any meaning difference, as shown in Figure 2.4.
tree
6
past the tree
?
past the tree
tree
Figure 2.4: Reversibility of Route paths
The second main property reflected in Table 2.2 is the presence or
lack of transition(s) in the denotation of the path. Paths that do not
have transitions are non-transitional paths. Paths with transitions have
two sub-types: transitional paths and delimited paths. The latter type
applies only to oriented (Goal and Source) paths and involves paths with
an explicit boundary. As already discussed in Section 2.5, Route paths
do not have a delimited subtype, because they do not specify the location
either of the starting point or of the endpoint (these points are only
negatively defined as not being in a locative relation with the Ground). I
suggested that as a result of this, the extreme points of Route paths cannot
be set as the limit of movement, hence the non-existence of delimited
Route paths.
To summarize, there are three major properties in the path classification
– orientation, transition, and delimitation – and they lead to the
following divisions:
(43) Divisions in the Path typology
a. Orientation – there is a division between oriented (Goal and
Source) and non-oriented (Route) paths.
b. Transition – there is a division between paths with transition(s)
and paths without transitions.
c. Delimitation – Delimited Goal/Source paths are a sub-type
of the oriented paths with transitions.
32 TYPES OF PATHS 2.7
The interaction of the three characteristic properties of paths I propose
leads to eight distinct types of paths. In order to be able to refer to each
of them separately, I introduce the following terminology:
(44) Types of paths
a. Coinitial – transitional Source-oriented path, e.g., from
the house
b. Egressive – delimited Source-oriented path, e.g, starting
from the house
c. Recessive – non-transitional Source-oriented path, e.g.,
away from the house
d. Cofinal – transitional Goal-oriented path, e.g., to the house
e. Terminative – delimited Goal-oriented path, e.g., up to
the house
f. Approximative – non-transitional Goal oriented path, e.g.,
toward the house
g. Transitive – transitional Route path, e.g., past the house
h. Prolative – non-transitional Route path. e.g., along the
house
The terms “coinitial” and “cofinal” are borrowed from Kracht (2002;
2007) and they correspond to two kind of modes that Kracht argues for.
The term mode refers to the way the Figure moves with respect to a
particular locative configuration. Kracht lists the following modes:
(45) Spatial modes Kracht (2002; 2007)
a. cofinal: The Figure moves into the configuration during
the event time, e.g., into the house
b. coinitial: The Figure moves from the configuration during
the event time, e.g., out of the house
c. approximative: The Figure approaches the configuration
during the event time, e.g., towards the tunnel
d. recessive: The Figure moves away from the configuration
during the event time, e.g., away from the tunnel
e. transitory: The Figure moves in and again out of the
configuration during the event time, e.g., through the tunnel
f. static: The Figure remains in the same configuration during
the event time, e.g., in the house
It is easy to set the correspondence between these modes and some of the
path types I suggest. However, there are a couple of path types that do
2.8 CONCLUSION 33
not find a matching mode. For instance, Kracht does not acknowledge as
a separate mode the Terminative, the Egressive and the Prolative. In this
respect, my typology is better matched by the one proposed in Mel’čuk
(1994). Mel’čuk includes in his enumeration of path types Terminative
paths, alongside the path types that Kracht has. Still, neither Kracht
(2002; 2007) nor Mel’čuk (1994) make a distinction between the two types
of Route paths. Concerning Route paths, my path clasification shares
features with Piñón’s (1993) path typology, as he argues that two types of
Route paths should be recognized. My classification is also reconcilable
with Zwarts’ (2008a) path types. The mapping to the corresponding
terms is shown in Table 2.3.
Terminology in this thesis
Zwarts’ (2008a)
example
terminology
Transitional Goal/Source paths → Transitions from, to
Non-transitional Goal/Source paths → Progressions towards
Transitional Route paths → Cycles across
Non-transitional Route paths → Continuations along
Table 2.3: Path terminology used in Zwarts (2008a) and in this thesis
The difference between the path classification I propose and the one
argued for by Zwarts is that Zwarts does not draw a distinction between
Source and Goal orientation in the categories Progressions and Transitions.
In the same vein, Piñón (1993) groups Source and Goal-oriented
paths together.
As none of the path typologies proposed in the literature so far encompasses
all the types of paths I believe exist, I adopt my own system
of path classification encompassing the eight types of paths I present in
Table 2.2. In Chapter 4, I explore the morphological expression of these
eight types and show how each of the divisions I presented in (43) finds
a morpho-syntactic reflex.
2.8 Conclusion
In this chapter, I explored the typology of paths. The purpose was to
determine what kinds of paths are encoded as linguistic primitives and to
decide the relevant properties that set them apart. I suggested that paths
are to be classified according to three properties: orientation, transition
and delimitation. This gives rise to eight types of paths in total. I showed
34 TYPES OF PATHS 2.8
how this path typology fits into existing classifications of paths, proposed
by other researchers (Jackendoff 1983, Piñón 1993, Mel’čuk 1994, Zwarts
2005; 2008a, and Kracht 2002; 2007) and also discussed the points where
it deviates from them.
Chapter 3
Background and theoretical
assumptions
3.1 Introduction
In this chapter, I lay out the core background assumptions on the basis
of which I develop the syntactic and semantic analyses of paths in
Chapters 4 and 5, respectively. I first present the commonly assumed
syntactic structure for directional phrases and, with this structure in
hand, I present the view adopted in this thesis on the relationship between
the morphological composition of an expression and the syntactic
structure underlying it. In that, I am led by the general guideline of
the cartographic framework (Cinque and Rizzi 2008), stating that each
morpho-syntactic feature corresponds to a unique syntactic head (see also
Kayne’s 2005 suggestion to the same effect). I follow the same framework
also with respect to other basic assumptions, namely, the universality
of syntactic structure and the rigid Specifier-Head-Complement order
(Cinque 2005).
3.2 Path, Place and the relation between syntax
and morphology
The syntactic structure underlying directional spatial expressions has
been commonly assumed to involve at least two projections: a Place
projection encoding static location, and a Path projection, where we find
directional markers (van Riemsdijk 1990, van Riemsdijk and Huybregts
2002, Koopman 2000, Svenonius 2010, den Dikken 2010). Thus, there is
35
36 BACKGROUND AND THEORETICAL ASSUMPTIONS 3.2
a general agreement that the minimal syntactic structure for directionals
looks like the one in (1).1
(1) PathP
Path PlaceP
Place . . .
This syntactic structure reflects Jackendoff’s (1983) conceptual structure
for spatial expressions, discussed in Chapter 2. The decomposition
of directional phrases into two components mirrors their conceptual decomposition
into a path component and a place component.
The two syntactic positions in the structure in (1) have direct morphological
counterparts in many languages, according to van Riemsdijk
and Huybregts (2002). Svenonius (2010), too, discusses languages, where
directional expressions involve two elements, each of which corresponds
to one of the positions in the structure in (1). An example for such a
language is Macedonian, which combines a directional preposition with
a locative preposition in Goal expressions (Tomič 2006:85), see (2) (data
provided by Eva Piperevska, p.c.).
(2) a. Kaj
at
parkot
park.def
sum.
be.1sg
‘I am at the park.’
b. Odam
go.1sg
na-kaj
to-at
parkot.
park
‘I am going to the park.’
In the example in (2), the Goal expression is formed by adding the Dative
preposition na to the general locative preposition kaj, resulting in
a bimorphemic marker na-kaj. The preposition na in (2) can be then
taken to lexicalize the Path head in (1), while the locative preposition
kaj spells out the Place head.
1
Authors vary in their terminology. For instance, van Riemsdijk and Huybregts
(2002) use the labels Dir and Loc, while den Dikken (2010) uses PDir and PLoc. In
this thesis, I follow the terminology in Svenonius (2010) and use the labels Path and
Place for the directional and locative heads, respectively.
3.2 THE RELATION BETWEEN SYNTAX AND MORPHOLOGY 37
(3) PathP
Path
na
PlaceP
Place
kaj
DP
parkot
The same mapping of morphemes onto syntactic structure can be found in
languages where spatial expressions are formed by means of case affixes.
Consider, for instance, the Daghestanian language Tsez, which combines
a locative and a directional case affix, as shown in (4) (data from Comrie
and Polinsky 1998:104).
(4) a. besuro-xo
fish-at
‘at the fish’
b. besuro-xo-r
fish-at-to
‘to the fish’
Commonly, it is assumed that the case suffixes in languages with spatial
case systems lexicalize the same heads Path and Place which are spelled
out by spatial prepositions in prepositional languages (van Riemsdijk and
Huybregts 2002, Svenonius 2010, see also Asbury et al. 2007 who explicitly
argue for this idea). Under this view, the bound morphemes -xo and
-r in Tsez lexicalize Place and Path, respectively, just like their prepositional
counterparts in Macedonian.2
In (5), I present how the base
syntactic structure of the Tsez Goal expression is lexicalized, assuming
a Specifier-Head-Complement structure in line with Kayne (1994)
and postponing the discussion of the linearization of the structure until
Chapter 7, Section 7.2.
2
The morpheme status of -r is supported by the fact that the suffix -r systematically
attaches to all the affixes of the remaining six locative series. Moreover, -r
is also used as a Dative case suffix, that is, it functions independently as a separate
morpheme which attaches straight to the noun, see (i) (Bokarev 1967d).
(i) besuro-r
fish-dat
These two facts suggest that is is highly unlikely that that the -xo-r complex in (4b)
is a non-decomposable Goal morpheme -xor, which accidentally has its initial part
identical to the Locative affix.
38 BACKGROUND AND THEORETICAL ASSUMPTIONS 3.2
(5) PathP
Path
-r
PlaceP
Place
-xo
DP
besuro
Note that the Path head is standardly assumed to be the host for directional
elements expressing all kinds of paths. Thus, we find there, for example,
both Goal-encoding morphemes and Source-encoding morphemes.
The latter case is exemplified in (6), using data from the language Tsez
again.
(6) besuro-x-¯ay3
fish-at-from
‘from the fish’
The tree structure corresponding to the Tsez example in (6) is shown
in (7). It is basically identical to the structure in (5), modulo the fact
that there is a different morpheme hosted by the Path head.
(7) PathP
Path
-¯ay
PlaceP
Place
-xo
DP
besuro
To sum up, there is a general consensus that directional expressions
are built on top of locative expressions by adding to the locative structure
the directional head Path. In a language like Macedonian and Tsez this is
morphologically transparent, as there are dedicated morphemes (na, -r,
-¯ay) that lexicalize the Path head. This pattern is not limited to Macedonian
and Tsez in particular. It is fairly common cross-linguistically
and exhibited by genealogically diverse languages, as can be seen from
Table 3.1.
Each of the directional morphemes presented in the data from Macedonian
and Tsez, as well as in Table 3.1, is associated with a given property
and contributes a certain meaning to the spatial expression. For
instance, the stative morphemes (kaj in Macedonian, -xo in Tsez, -o in
3
The vowel o drops regularly before another vowel (Comrie and Polinsky 1998).
3.2 THE RELATION BETWEEN SYNTAX AND MORPHOLOGY 39
Language Genus Location Goal Source Reference
Garo Baric -o -o-na -o-ni Burling (2003)
Estonian Finnic -l -l-lea
-l-t Viitso (1998)
Lezgian Lezgic -qh
-qh
-di -qh
-aj Haspelmath (1993)
Mwotlap Oceanic l(V) a l(V) mw
E l(V) Crowley (2002a)
Yanesha Arawakan -o -o-net -o-ty
Duff-Tripp (1997)
a
According to the orthographical conventions in Estonian, the double ll of the
Allative ending is written as a single l. Nevertheless, the Allative marker is morphologically
decomposed as shown above (Anna Tamm, p.c.).
Table 3.1: Morphological containment of locative expressions inside directional
expressions
Garo, -l in Estonian, etc.) express location and the “dynamic” goal and
source morphemes (na in Macedonian, -r and -¯ay in Tsez, -na and -ni in
Garo, etc.) express transition to or from this location. The stative and
dynamic morphemes also correspond to a syntactic head in the structure
(Place and Path). Hence, we can conjecture that the number of
morphemes that constitute a given expression serves as an indicator of
the number of heads present in the underlying syntactic structure. This
idea is in line with Kayne’s (2005) suggestion that each morphosyntactic
feature should correspond to an independent syntactic head in the
functional sequence. This is also a principle adopted in the cartographic
framework (Cinque 1999, Rizzi 1997, inter alia), where authors are guided
by the maxim “one (morphosyntactic) property – one feature – one head”
(Cinque and Rizzi 2008). Thus, I assume that whenever we are able to
isolate a morpheme which adds a given property (or meaning) to the expression
it is a part of, this is an indication that the underlying syntactic
structure contains an independent head corresponding to that property
(and morpheme).
The assumed parallel between morphological composition and syntactic
structure strongly suggests a theoretical framework which unifies
morphology and syntax into one component of grammar (cf., Baker’s
1985 Mirror Principle and his proposal to the same effect). I discuss this
question in more details in Chapter 6, where I present one such theory.
Assuming two heads in the syntactic structure of directional expressions
– Place and Path – presupposes also a semantic “decomposition” of
directionals, as each of the heads in the syntactic structure is expected
to have some semantic contribution. We find this concept developed
in the work by Kracht (2002; 2007) and Zwarts (2005; 2008a). Kracht
40 BACKGROUND AND THEORETICAL ASSUMPTIONS 3.2
suggests that locative expressions universally contain two layers: one is
called configuration and describes the way the Figure and the Ground
are positioned with respect to each other, e.g., in, under, above, etc. The
other layer specifies the mode, which describes the way in which the Figure
moves with respect to the configuration (Kracht’s modes are listed in
(45), Chapter 2). These two layers can be easily mapped onto the syntactic
structure, taking the Place head to contribute the configuration,
and the Path head to contribute the mode (abstracting away from the
static mode).
Zwarts (2005; 2008a), too, proposes a semantics for directional preposition
which relates paths to locations. More specifically, Source prepositions
are the ones that include the starting point of a path, p(0), which
can be either in the reference object (for out of ), on the reference object
(off ), or at the reference object (from). Goal prepositions, on the
other hand, include the end point of the path, p(1), which is in, on,
or at the reference object (resulting in into, onto and to, respectively).
Finally, Route prepositions (through, across and past) exclude both p(0)
and p(1) and include the intermediate positions of the path p(i) being
in, on, or at the reference object. Notice that the prepositions out
of, into and through share a property too, namely, they all relate to an
“in-location” with respect to the reference object. Table 3.2, which is
adapted from Zwarts (2005:759), gives a more perspicuous overview of
the decomposability of the prepositions discussed in this paragraph.
in on at
Source Ps p(0) out of off from
Goal Ps p(1) into onto to
Route Ps p(i) through across past
Table 3.2: Relation between paths and locations (Zwarts 2005)
The semantics proposed by Zwarts is clearly compositional and in
line with a syntactic structure where a Path head dominates a Place projection.
Thus, if we are to make the syntactic and semantic hypotheses
converge, it will be fairly obvious to state that the Path head encodes
Goal or Source, while the Place head below it expresses an in, on or at
relationship between the Figure and the Ground. Put in other words,
the in/on/at bit is the semantic contribution of the Place head, while
the semantic content of the Path head specifies whether the locative condition
encoded by the Place head holds of the starting point p(0), the
3.3 LEXICALIZATION OF THE STRUCTURE 41
end point p(1) of the path, or some intermediate point p(i) (Svenonius
2007).
To sum up the discussion so far, I presented the commonly assumed
decomposition of directional spatial expression into a Path head dominating
a Place head.4
I showed evidence for this decomposition coming from
the morphological make-up of directional expressions cross-linguistically
and their associated semantics. Finally, I suggested that the presence
of a morpheme in a given expression is an indication for the presence
of a syntactic head corresponding to that morpheme in the structure
underlying the expression.
3.3 Lexicalization of the structure
In the preceding section, I proposed that whenever we see two independent
morphemes in a given expression, this indicates the existence of
two heads in the syntactic structure, corresponding to each of the two
morphemes. This mapping of morphological structure onto syntactic
structure, however, raises one important question: what happens in languages
where path expressions involve only one morpheme? Consider, for
instance, the Dutch Goal expression in (8) featuring a so called strictly
directional preposition (data from den Dikken 2010).
(8) naar
to
het
the
bos
woods
One way to deal with the mismatch between the number of morphemes in
the expression in (8) and the number of heads in the structure in (1) is to
assume that the syntactic structure for directional expressions in Dutch
does not contain two heads – Path and Place – but just one head. This
one head will then correspond to the single morpheme naar. Since naar is
a strictly directional preposition, i.e., one that has only directional uses,
this syntactic head will have to be the Path head, as shown in (9).
4
These are not the only syntactic projections directional phrases consist of. Various
other heads have been proposed. To name a few, Koopman (2000) argues for a Deg
head, accommodating measure phrases; den Dikken (2010) proposes a Deix head for
deictic elements; Svenonius (2006) proposes an AxPart for spatial elements that very
often have nominal properties, like front in in front of.
42 BACKGROUND AND THEORETICAL ASSUMPTIONS 3.3
(9) PathP
Path
naar
DP
het bos
This assumption has one disadvantage. If we assume that the structure
of Dutch directionals differs from the structure of directionals in
Macedonian and Tsez, we give up the idea of a universal syntactic structure
underlying all natural languages. This goes against Chomsky’s
(2001) Uniformity Principle.
(10) Uniformity Principle (Chomsky 2001:2):
In the absence of compelling evidence to the contrary, assume
languages to be uniform, with variety restricted to easily detectable
properties of utterances.
It is therefore preferable to assume that, in all languages, the syntactic
structure of directional expressions involves a Path projection taking as a
complement a Place projection, even though in many languages this is not
morphologically transparent. Thus, the Uniformity Principle prompts us
to postulate that the syntactic structure of the Goal path expressed by
the Dutch directional preposition naar involves the same heads Path and
Place as the syntactic structure for the Macedonian and Tsez Goal paths
expressed by nakaj and xor, respectively, despite the fact that in Dutch
there are more syntactic heads (two) than morphemes (one). In fact,
it has to be so, as Dutch was one of the languages which gave rise to
the establishment of the structure in (1) on the basis of other syntactic
phenomena like movement and incorporation (see the seminal works of
Koopman 2000 and den Dikken 2010).
Such an approach is in line with the cartographic framework whose
working hypothesis is that all languages share the same universal syntactic
structure. Hence, whenever some language provides evidence for the
existence of a particular head, then this head must be present in all other
languages, even if they do not show direct evidence for its existence. As
discussed in Cinque and Rizzi (2008), this hypothesis might turn out to
be wrong, but it is the strongest hypothesis possible and therefore the
one that should be pursued until evidence is found to the contrary.
The assumption that the structure underlying the Dutch naar-expressions
involves the two heads Path and Place raises the question of
how to deal with the aforementioned mismatch between the number of
morphemes and the number of heads. There are three possibilities.
3.4 LEXICALIZATION OF THE STRUCTURE 43
• Possibility 1: the preposition naar is merged under the Place head
and the Path head is occupied by a silent directional preposition,
as suggested in Koopman (2000).
• Possibility 2: the preposition naar is merged under the Path head
and the Place head is occupied by a silent locative preposition, as
suggested in den Dikken (2010).
• Possibility 3: the preposition naar spells out simultaneously the
Path and the Place head.
The three possibilities are schematized in the tree structures below.
(11) a. PathP
Path
∅
PlaceP
Place
naar
DP
b. PathP
Path
naar
PlaceP
Place
∅
DP
c. PathP
Path PlaceP
Place DP



naar
The first possibility depicted in (11a) is untenable, as it predicts that
naar can also function as a locative preposition, contrary to the facts (see
the discussion in den Dikken 2010). The second possibility in (11b) has
also one weakness, namely, it requires the availability of a silent locative
preposition. This locative preposition, however, never surfaces in the
absence of a strictly directional preposition. It is therefore necessary to
somehow restrict its syntactic distribution. As den Dikken concludes in
an earlier version of his paper, it remains unclear how to achieve this.5
Given the problems encountered by the first two possibilities, the
third approach, according to which naar lexicalizes both Path and Place,
seems to be worth pursuing. It is also the approach adopted by Caha
(2010a) for his analysis of the German strictly directional preposition
zu ‘to.’ Thus, I assume that it is possible for one lexical item to lexicalize
more than one head in the syntactic structure. This assumption
has been made in a fair number of previous linguistic works, for example
Williams (2003), Borer (2005a;b) and Ramchand (2008b), who implement
it by using different theoretical apparatuses. How exactly this is
achieved theoretically in this thesis is the topic of Chapters 6 and 7.
5
Den Dikken’s discussion concerns the strictly directional German preposition aus
‘from’, but the same problem arises in the analysis of Dutch naar.
44 BACKGROUND AND THEORETICAL ASSUMPTIONS 3.4
3.4 Conclusion
Before concluding this chapter, let me summarize the main theoretical
assumptions, which I will make use of in my investigation of the syntactic
structure underlying the various types of path expressions.
• Assumption 1: The mapping between morphemes and syntactic
terminals is one-to-many.
• Assumption 2: Syntactic structure is universal.
• Assumption 3: There is a rigid Specifier-Head-Complement order.
According to assumption 1, the identification of a morpheme in a
given language suggests the presence of one or more heads in the syntactic
structure which this morpheme lexicalizes. This entails that one syntactic
head can be lexicalized by just one morpheme, but one morpheme can
lexicalize many syntactic heads.
As stated in assumption 2, a syntactic head that is established on the
basis of morphological data in a given language will be universally present
in all languages, even those which do not offer morphological evidence for
its existence. Thus I adopt the working hypothesis of the cartographic
approach saying that all languages share the same underlying syntactic
structure, a part of Universal Grammar.
Assumption 3 says that the only ordering found in human language
is one where specifiers precede heads and heads precede complements, as
argued for by Kayne (1994). All other orders are derived by movement.
In the next chapter, I investigate the morphological composition of
path expression in a variety of languages, guided by the assumptions
above. I develop the idea that the structure underlying directional expressions
involves more heads than just Path (and Place) and suggest
that different types of paths correspond to different syntactic structures.
Chapter 4
Decomposing Path
4.1 Introduction
In Chapter 2, I explored the question of how many kinds of spatial paths
there are. This provides the theoretical base for my investigation of how
these various types of paths are expressed cross-linguistically. For my
purposes, I studied the grammar descriptions of 81 genealogically diverse
languages.1
The results of the typological study reveals that there is
a certain hierarchy when it comes to the expression of paths. Namely,
different paths are of different morphological complexity and, crucially,
subject to a subset-superset relationship. For instance, in some languages,
Source paths morphologically contain Goal paths, Route paths
contain Source paths, and non-transitional paths contain the corresponding
transitional paths. Interestingly, the reverse pattern appears to be
non-existent. My findings were confirmed by data from languages which
were not included in the initial sample.
Taking morphological complexity to be indicative of syntactic complexity,
as suggested in Chapter 3, I propose a syntactic structure for
directional expressions that can accommodate the linguistic facts. In a
nutshell, I suggest that each of the distinctions in the path typology listed
in (43), Chapter 2, is brought about by a special head in the syntactic
structure underlying directional expressions. Consequently, I decompose
the Path head which has been proposed to be present in the syntactic
structure for directionals into five heads: Goal, Source, Route, Scale, and
Bound.
1
A list of the languages in the sample is presented in the Appendix.
45
46 DECOMPOSING PATH 4.2
4.2 Goal versus Source paths
One of the properties with respect to which paths differ is the orientation
value they have. Paths can be Goal-oriented and Source-oriented. In the
majority of languages these two kinds of paths are encoded by elements
of equal complexity. For example, in Scottish Gaelic both the Goal and
the Source marker are monomorphemic, see (1).
(1) Scottish Gaelic (MacAulay 1992)
a. gu
to
bocsa
box
‘to (a) box’
b. bho
from
bocsa
box
‘from (a) box’
In the languages in Table 3.1, both the Goal and Source paths are encoded
by bi-morphemic markers consisting of a locative element and a
directional element. This type has been already exemplified by Tsez,
repeated below from (4b) and (6) in Chapter 3.
(2) a. besuro-xo-r
fish-at-to
‘to the fish’
b. besuro-x-¯ay
fish-at-from
‘from the fish’
There are, however, languages where the morphological equality between
Goal and Source markers does not hold. One such language is Imbabura
Quechua (Jake 1985, Cole 1985). Imbabura Quechua has a spatial case
system with a Locative case, expressing static location, an Allative case,
expressing Goal-oriented paths, and an Ablative case, expressing Sourceoriented
paths (data from Cole 1985:119).
(3) Imbabura Quechua
a. Utavalu-pi
Otavalo-loc
kawsa-ni.
live-1
‘I live in Otavalo.’ (Location)
b. Utavalu-man
Otavalo-all
ri-ni.
go-1
‘I go to Otavalo.’ (Goal)
4.2 GOAL VS SOURCE PATHS 47
c. Utavalu-manda
Otavalo-abl
shamu-ni.
come-1
‘I come from Otavalo.’ (Source)
The intriguing fact is that the Allative marker in Imbabura Quechua is
-man (3b), while the Ablative marker is -manda (3c). On the face of it,
it seems that the Ablative marker contains the Allative marker.
Another language which demonstrates a containment relationship between
the Source marker and the Goal marker is the Daghestanian language
Chamalal (Magomedbekova 1967b), as shown by the data below
(my glossing).
(4) Chamalal (Gigalt dialect)
a. miky
i-l-a
road-on-loc
‘on the road’ (Location)
b. miky
i-l-u
road-on-all
‘onto the road’ (Goal)
c. miky
i-l-u-r
road-on-all-abl
‘off the road’ (Source)
As shown in Table 4.1, the Source marker is built by adding the morpheme
-r to the Goal marker -u in each of the seven locative series in
Chamalal.
Series Location Goal Source
in -∅ -∅-i -∅-u -∅-u-r
at -x -x-i -x-u -x-u-r
inside -ń -ń-i -ń-u -ń-u-r
behind -n -n-i -n-u -n-u-r
under -kk -kk-i -kk-u -kk-u-r
on -l -l-i -l-u -l-u-r
vertical attachment -tS -tS-i -tS-u -tS-u-r
Table 4.1: Spatial case system in Chamalal (Magomedbekova 1967b)
It should be noted that the Goal morphemes -man in Quechua and
-u in Chamalal are unambiguously directional, that is, they cannot be
used in locative contexts. This excludes the possibility that the Source
48 DECOMPOSING PATH 4.2
elements -da and -r are added to a stative Locative morpheme, as happens
in Tsez and in the languages listed in Table 3.1.
The Source and Goal paths discussed so far are transitional paths
like to and from. It is then a legitimate question whether such a containment
relationship obtains between non-transitional Source and Goal
paths (away from and towards). The answers is yes, and the evidence for
this is provided by languages like Dime (Mulugeta 2008) and Bulgarian.
The Omotic language Dime, spoken in Ethiopia, has a directional
marker -bow that, when attached to a noun marked by Comitative case,
expresses direction towards the Ground (as described by Mulugeta 2008).
In other words, -bow is a marker of a non-transitional Goal-oriented path.
Interestingly, according to Mulugeta (2008), away from-paths (i.e., nontransitional
Source paths) are expressed by adding the Ablative marker
-de to a noun that is already marked by -bow. I reproduce here the
examples he gives, where the proper noun Taddese is marked by the
directional -bow, see (5a), and by the combination of -bow and the Ablative
marker -de, see (5b) (data from Mulugeta 2008:58, the original
translation provided by Mulugeta is preserved).
(5) a. šiftaye
Shiftaye
taddeseka-bow
Taddese.com-dir
tiN-i-n.
go-perf-3
‘Shiftaye went towards Taddese.’
b. šiftaye
Shiftaye
taddeseka-bow-de
Taddese.com-dir-abl
Pád-i-n.
come-perf-3
‘Shiftaye came from the place where Taddese is found.’
Thus, we can say that the Recessive (away) path is built on top of an
Approximative (towards) path, by the addition of the Ablative -de.
Exactly the same pattern is replicated in the Slavic language Bulgarian.
The Bulgarian preposition k@m is unambiguously directional and
means ‘towards, in the direction of’. Thus, the data in (6a) is an example
for a non-transitional Goal path. A non-transitional Source path is
expressed by adding the Source preposition ot to the preposition k@m,
thus deriving the complex preposition otk@m, (6b).
(6) a. k@m
towards
k@shta-ta
house-def
‘towards the house, in the direction of the house’
b. ot-k@m
from-towards
k@shta-ta
house-def
‘from the direction of the house’
4.2 GOAL VS SOURCE PATHS 49
Again we find the Recessive (away) path preposition to morphologically
contain the Approximative (toward) path preposition.
Table 4.2 summarizes all the languages where I have discovered a
Source-Goal containment relationship. So far I haven’t come across a
language where the reversed relationship obtains, that is, where a Goal
expression morphologically contains a Source expression.
Language Location Goal Source Reference
Bulgarian pri k@m ot-k@m Pashov (1999)
Dime -se -bow -bow-de Mulugeta (2008)
Chamalal -i -u -u-r Magomedbekova (1967b)
Ingush -ğ -ga -ga-ra Nichols (1994)
Jingulu -mpili -Nka -Nka-mi Blake (1977)
Mansi -t -n -n-@l Keresztes (1998)
Quechua -pi -man -man-da Jake (1985), Cole (1985)
Uchumataqu -tá -ki -ki-stani Vellard (1967)
Table 4.2: Languages where the Source marker morphologically contains
the Goal marker
Importantly, some of the languages in Table 4.2 provide evidence that
the element attaching to the Goal marker in the Source expression serves
as an independent morpheme. For instance, -stani in Uchumataqu is used
as Comitative suffix (Adelaar 2004). Likewise, in Mansi, the element -l is
in fact the Instrumental case suffix and the reduced vowel @ can be shown
to be inserted for phonological reasons (Keresztes 1998). Even more
convincing are the cases of Bulgarian and Dime, where the morphemes
k@m and -de by themselves express a Coinitial path (i.e., a transitional
Source path from), but attach to the Approximative (toward) marker in
Recessive (away from) expressions.
These facts suggest that the containment relationship obtaining between
Source and Goal expressions, no matter whether transitional or
not, is not accidental, since in many languages it is possible to identify
the bit that is added to the Goal marker as an independent element.
Therefore, I claim that the Source expressions in the languages in Table
4.2 are built on the basis of Goal expressions by adding a morpheme.
Taking morphological complexity to be indicative of syntactic complexity,
as argued in Chapter 3, I suggest that the morphemes attaching to
the Goal expressions in Table 4.2 indicate the presence of an independent
syntactic head corresponding to them. I therefore propose that the
50 DECOMPOSING PATH 4.2
syntactic structure of Source expressions embeds the syntactic structure
for Goal expressions, as illustrated in the tree diagram in (7).
(7) a. Source phrase:
SourceP
Source GoalP
Goal PlaceP
b. Goal phrase:
GoalP
Goal PlaceP
The structure in (7a) can be mapped onto the morphological composition
of the Chamalal Source expression miky
i-l-u-r ‘off the road’ (see
(4c)) in the way depicted in (8). I assume that the series marker -l
‘on’ lexicalizes the AxPart head proposed by Svenonius (2006), where
one finds spatial elements encoding notions like exterior, interior,
top, bottom, etc. (see Muriungi 2006, Takamine 2006, Pantcheva 2006;
2008). In addition, I assume that a single morpheme can spell out more
than one head in the syntactic strucure, as suggested in Chapter 3.2
(8) SourceP
Source
-r
GoalP
Goal PlaceP
Place AxPart
AxPart
-l
DP
miky
i



-u
According to the hypothesis that languages share a universal syntactic
structure, the structure in (7a) underlies Source expressions in all
languages, even those where they are morphologically non-decomposable.
That is, I propose that Source expressions are built on top of Goal expressions
by the addition of a dedicated syntactic head also in languages like
English, Scottish Gaelic and Tsez, where the containment relationship is
not evident. The reason that the syntactic structure in these languages
2
For the lexicalization of the Dime and Bulgarian Recessive expressions, see (32)
in Chapter 5.
4.3 ROUTES VS GOAL AND SOURCE PATHS 51
is not reflected by the morphological composition of the Source marker
is that the Source marker lexicalizes both the Goal and the Source head.
For an illustration, see the tree diagram in (9) showing the lexicalization
of the structure in the Tsez Source expression besuro-x-¯ay (fish-at-from)
‘from the fish.’
(9) SourceP
Source GoalP
Goal PlaceP
Place AxPart
AxPart DP
besuro






-¯ay
-xo
The existence of a special Source head in the syntactic structure for
Source expressions raises the question of what precisely the semantic
function of this head is. I will not address this question here, but defer the
discussion to the next chapter, which is entirely dedicated to establishing
the semantics of each of the heads I am going to postulate in the current
chapter.
4.3 Route paths versus Goal and Source paths
In the previous section, I showed that the split within the oriented paths
into Source and Goal-oriented ones has a morpho-syntactic reflex. It is
interesting to see whether the other divisions in the path typology also
correlate with a difference in the morpho-syntactic structure of the corresponding
directional phrases. In this section, I investigate the split
between the non-oriented and oriented paths, or put differently, I compare
Route paths on the one hand, with Goal and Source paths, on the
other hand.
The first clue about the relationship between non-oriented and oriented
paths comes from the Slovak language. In Slovak, Route phrases
are formed by adding the preposition po to a PP expressing a Goal path
(P. Caha, p.c.).
52 DECOMPOSING PATH 4.3
(10) Na
On
Forum
Forum
Romanum
Romanum.acc
vstupujeme
enter.1pl
po-pod
po-under
oblúk-∅
arch-acc
Tita.
of.Tito
‘We entered the Forum Romanum by going under Tito’s arch.’
(lit.: via under)
The following data shows that without the preposition po, the PP has a
Goal reading.
(11) Slamu
hay
dal
put.3sg
pod
under
stôl-∅.
table-acc
‘He put the hay under the table.’
Thus, the data from Slovak suggests that Route paths are “bigger” than
Goal paths in that the former morphologically contain the latter. Given
that Source paths are also “bigger” than Goal paths, a legitimate question
arises: Is there a relationship between Route and Source paths? As
things stand now, we know that both Route and Source paths are formed
on the basis of Goal paths. It is logically possible, though, that there
is an ordering between Source and Route paths, too. At this point, the
typological investigation of path expressions cross-linguistically meets its
first impediment – many grammar descriptions do not mention Route
paths. Since the current study relies heavily on written sources, it becomes
quite difficult to determine in what ways Route paths relate to
other paths in their morphological make-up.
Despite the scarcity of information, there are two languages that shed
light on the topic. The first language is the Daghestanian language
Akhvakh, described by Magomedbekova (1967a). Like other languages
from the East Daghestanian group, Akhvakh has a rich spatial case system,
like the ones already presented for the related languages Tabasaran
(Table 2.1) and Chamalal (Table 4.1). The noteworthy fact is that Route
paths are formed by the addition of the morpheme -ne to the Ablative
(Source) case ending -u, thus deriving the complex Translative (Route)
case ending -u-ne. Table 4.3 presents the case system in Akvakh, based
on data in Magomedbekova (1967a).
The same phenomenon has been observed for the language Avar by a
number of researchers: Uslar (1889), Charachidzé (1981), Blake (1994).
In Avar, too, the Route case (called Perlative) is formed by suffixing the
Ablative case marker with the element -n. The data is shown in Table
4.4.
The existence of an Ablative base for the Perlative form in Avar is
4.3 ROUTES VS GOAL AND SOURCE PATHS 53
Series Location Goal Source Route
on -g -g-e -g-a -g-u -g-u-ne
at, near -x -xar-i -lir-a -xar-u -xar-u-ne
at -q -q-e -q-a -q-u -q-u-ne
in -l’ -l’-i -l’-a -l’-u -l’-u-ne
under -t@
-t@
-i -t@
-a -t@
-u -t@
-u-ne
Table 4.3: Spatial case system in Akhvakh (Magomedbekova 1967a)
Series Location Goal Source Route
on (top of) -da -da -d-e -da-ssa -da-ssa-n
at -q -q -q-e -q-a -q-a-n
under -ň’ -ň’ -ň’-e -ň’-a -ň’-a-n
in, among -ň -ň -ň-e -ň-a -ň-a-n
in a hollow -∅ -∅ -∅-e -∅-ssa -∅-ssa-n
object
Table 4.4: Spatial case system in Avar (Blake 1994)
supported by the fact that the allomorph of the Ablative case is preserved
in the Perlative case. That is, whenever the form of the Ablative suffix is
-ssa, the Perlative is -ssa-n. And whenever the Ablative suffix is -a, the
Perlative is -a-n. This excludes the possibility that the morphological
containment relationship between the Route and the Source cases is a
phonological accident and justifies the independent morpheme status of
-n
So far, I presented data showing that (i) Route paths contain Goal
paths (Slovak), and (ii) Route paths contain Source paths (Akhvakh and
Avar). This provides us with an answer to the question of how the split
between non-oriented and oriented paths is reflected in morpho-syntax
– non-oriented paths are formed on the basis of oriented path by the
addition of a special head. Hence, the syntactic structures corresponding
to the two types of paths are as shown in (12), where non-oriented paths
are built on top of oriented paths.
(12) a. Non-oriented paths
Pathnon-oriented
Pathoriented PlaceP
b. Oriented paths
Pathoriented PlaceP
54 DECOMPOSING PATH 4.3
Recall from Section 4.2 that the oriented paths themselves split into
Source and Goal paths, which are in a hierarchical relation – Source paths
are built on top of Goal paths (see (13) repeated from (7)).
(13) Oriented paths
a. Source path
SourceP
Source GoalP
Goal PlaceP
b. Goal path
GoalP
Goal PlaceP
The evidence for these structure comes from the morphological containment
facts. This relation of morphological containment has several
properties. It is:
• transitive – if A morphologically contains B and B morphologically
contains C, then A contains C (for instance, in Avar, the Perlative
case contains the Ablative and the Ablative contains the Locative,
hence Perlative contains Locative)
• reflexive – A morphologically contains itself (for instance, the Allative
contains the Allative)
• antisymmetric – if A morphologically contains B and A is distinct
from B, then B does not morphologically contain A (for instance,
in Chamalal the Ablative contains the Allative, but the Allative
does not contain the Ablative)
Let “>” represent the transitive, reflexive, and antisymmetric relationship
of morphological containment. We have already established the
following pairs.
(14) a. Source path > Goal path
b. Route path > Goal path
c. Route path > Source path
The pairs in (14) give rise to the following linear order:
(15) Route path > Source path > Goal path
Translated into syntactic structure, this order gives us the following tree:
4.4 TRANSITIONAL VS NON-TRANSITIONAL PATHS 55
(16) RouteP
Route SourceP
Source GoalP
Goal PlaceP
I suggest that this is the structure universally underlying Route expressions.
Applied to the Avar data, the morphemes map onto syntactic terminals
in the following way (taking as an example the at-series marked
by -da).
(17) RouteP
Route
-n
SourceP
Source GoalP
Goal PlaceP
Place AxPartP
AxPart DP






-ssa
-da
Summing up, I have reached the conclusion that there is a linear
ordering between Goal, Source and Route paths on the basis of the relation
of morphological containment. This ordering corresponds to an
increasing complexity in the syntactic structure underlying these path
expressions. Goal paths are the minimal element, or put informally, the
most “simple” paths, Source paths are formed by the addition of a Source
head, and finally, Route paths are the most complex ones, as they are
built on top of Source paths by the addition of a Route head, whose
semantics I discuss in Chapter 5, Section 5.5.
4.4 Transitional versus non-transitional paths
It is a very common phenomenon cross-linguistically that non-transitional
paths, like the one expressed by the English preposition towards, are more
56 DECOMPOSING PATH 4.4
complex than the corresponding paths expressing transition, like the English
Goal path expressed by to. This can be illustrated by a morphological
analysis of the English prepositions just mentioned: to and towards.
The first one (to) is clearly contained in the second one (to-wards). Furthermore,
we encounter the morpheme -wards in other English spatial
elements, for example outwards, inwards, etc. The contribution of -wards
in all cases appears to be the same: towards means “in the direction of,”
outwards means “in direction (away) from inside.” On semantic grounds
then, the role of -wards seems to be to turn a transitional path into a
non-transitional path, preserving the orientation of the path.
The same phenomenon is observed for a number of other languages.
For instance, Tabasaran and Avar have special morphemes that attach
to a noun marked by the Allative and the Ablative case, which results in
a non-transitional Goal-oriented (towards) and a non-transitional Source
oriented (away from) path, respectively. The morphological structure of
non-transitional Goal (Approximative) and Source (Recessive) paths in
these two Daghestanian languages is demonstrated in the tables below.
Table 4.5 shows how Approximative and Recessive paths are constructed
in Avar for the on-series, Table 4.6 shows how they are expressed in
Tabasaran for the in-series.
Goal Source
Trans -de -da-ssa
Non-trans -de-Gun -da-ssa-Gun
Table 4.5: Transitional and nontransitional
paths in Avar (Uslar
1889)
Goal Source
Trans ’-na ’-an
Non-trans ’-na-di ’-an-di
Table 4.6: Transitional and
non-transitional paths in
Tabasaran (Hanmagomedov
1967)
Tabasaran and Avar clearly illustrate that non-transitional Goal and
Source paths are built on top of the corresponding transitional paths.
Drawing a parallel between oriented paths (Goal and Source) and nonoriented
paths (Route), one would expect that non-transitional Route
(Prolative) paths, too, are formed on the basis of transitional Route
(Transitive) paths. Regrettably, the language sample used in this study
provides no such data. This might be an accidental gap. Alternatively,
there are such languages but since Route paths are not routinely included
in grammars, the data is inaccessible. Nevertheless, in Chapter 9,
Section 9.3.2, I discuss data from English which indirectly confirms the
hypothesis that Prolative paths contain Transitive paths.
4.4 TRANSITIONAL VS NON-TRANSITIONAL PATHS 57
When it comes to the syntactic structure underlying non-transitional
path expressions, I assume the same strategy as before and suggest
that non-transitional paths embed the corresponding transitional paths.
Thus, I argue for the existence of a special Scale head that comes on top
of any transitional path structure and turns it into a non-transitional
path.
(18) a. Non-transitional Goal (Approximative) paths
ScaleP
Scale GoalP
Goal PlaceP
b. Non-transitional Source (Recessive) paths
ScaleP
Scale SourceP
Source GoalP
Goal PlaceP
c. Non-transitional Route (Prolative) paths
ScaleP
Scale RouteP
Route SourceP
Source GoalP
Goal PlaceP
The English morpheme -wards, as well as the morphemes -Gun and -di
in Avar and Tabasaran, respectively, are then lexicalizations of the Scale
58 DECOMPOSING PATH 4.5
head I propose, as shown below for the Approximative English expression
towards the house and the Recessive Tabasaran expression a@rabaji’-an-di
(cart.erg.in-abl-scale3
) ‘in direction out of the cart.’
(19) a. ScaleP
Scale
-wards
GoalP
Goal PlaceP
Place DP
the house



to
b. ScaleP
Scale
-di
SourceP
Source GoalP
Goal PlaceP
Place
’
AxPartP
a@rabaji



-an
Thus, again I reached the conclusion that one of the distinctions in
the path typology is reflected in the syntactic structure by the presence of
a dedicated head. Specifically, the division between transitional and nontransitional
paths is due to the Scale head which is part of the syntax of
the latter. The Scale head thus takes as a complement a transitional path
structure (GoalP, SourceP or RouteP) and derives the corresponding nontransitional
path. The discussion of the exact semantic function of the
proposed Scale head is taken up in Chapter 5, Section 5.6.
4.5 Delimited and non-delimited paths
In this section, I turn to the last distinction in the path typology — the
split between delimited paths, i.e., paths that, apart from expressing a
3
I gloss the morpheme -di scale for reasons I lay out in Chapter 5
4.5 DELIMITED AND NON-DELIMITED PATHS 59
transition, also set one of the extreme points as the limit for movement
(e.g., an up to path), and non-delimited paths, i.e., paths that do not
impose such a condition on their extreme points (e.g., a to path). I will
follow the same approach I have followed until now in that I examine
the morphological make-up of these types of paths and try to establish a
one-way morphological containment relationship.
Recall that delimited paths split into two types: (i) the so-called Terminative
paths, which are delimited at the end-point p(1), e.g., the path
expressed in The boy ran up to the house, where the boy runs as far as
the house and stops at it, and (ii) the so-called Egressive paths, which are
delimited at the starting point p(0), e.g., the path expressed in The boy
ran starting from the house. Terminative paths differ from non-delimited
Goal-oriented (Cofinal) paths in that they explicitly set the location to
which the path relates as the upper boundary for movement. Similarly,
Egressive paths paths differ from non-delimited Source-oriented (Coinitial)
paths in that they set the location as the lower boundary for movement.
In Table 4.7, I present the four types of paths mentioned in this
paragraph in order to remind the reader of the terminology used in the
thesis and give an overview of the relationship between them.
Goal-oriented Source-oriented
Non-delimited Cofinal (to) Coinitial (from)
Delimited Terminative (up to) Egressive (starting from)
Table 4.7: Delimited and non-delimited paths
As we can see from Table 4.7, Terminative and Cofinal paths, while
sharing the property of being Goal-oriented, differ from each other in
that the former are delimited and the latter are not. Likewise, Egressive
and Coinitial paths are both Source-oriented but are distinguished by the
lack of delimitation for the latter type of paths. Terminative and Egressive
paths differ from each other in their orientation. In what follows, I
investigate the question whether these distinctions have a morphological
reflex.
Let us begin the investigation with Terminative paths. In many languages,
Terminative paths are formed on the basis of non-delimited transitional
Goal paths (Cofinal paths), as can be already seen from the composition
of the English Terminative expression up to, which contains the
Cofinal preposition to. Other examples come from Basque, (20), and the
Finnic language Veps, (21).
60 DECOMPOSING PATH 4.5
(20) Basque (Ibarretxe-Antuñano 2004)
a. etxe-ra
house-all
‘to the house’
b. etxe-raino
house-term
‘up to the house’
(21) Veps (Zajtseva 1981)
a. mec-ha
forest-all
‘to the forest’
b. mec-hasei
forest-term
‘up to the forest’
The data in (20)-(21) reveals that both in Basque and in Veps the Terminative
case suffix is complex and consists of the Goal suffix (Allative
-ra in Basque, Illative -ha in Veps) and a “delimiting” morpheme (-ino
in Basque, -sei in Veps). Since I assume that the identification of a
morpheme indicates the presence of at least one syntactic head in the
structure which corresponds to the morpheme, it follows that Terminative
paths are built on top of Cofinal paths by the addition of a special
syntactic head, which I label Bound.
(22) a. Delimited Goal (Terminative) paths
BoundP
Bound GoalP
Goal PlaceP
b. Non-delimited Goal (Cofinal) paths
GoalP
Goal PlaceP
The lexicalization of the structure for a Terminative path is presented in
(23), taking as an example the Veps expression in (21b).
4.5 DELIMITED AND NON-DELIMITED PATHS 61
(23) BoundP
Bound
-sei
GoalP
Goal PlaceP
Place DP
mec


-ha
If it is indeed the case that the Goal-oriented delimited paths (i.e.,
Terminative paths) are derived on the basis of Cofinal paths by the addition
of a special head Bound, then we expect this Bound head to be
operative also in Source-oriented delimited paths. In other words, the
syntactic structure of Egressive paths should be the one presented in
(24), where the delimited Source path embeds the non-delimited Source
path.
(24) a. Delimited Source (Egressive) paths
BoundP
Bound SourceP
Source GoalP
Goal PlaceP
b. Non-delimited Source (Coinitial) paths
SourceP
Source GoalP
Goal PlaceP
The structure in (24a) in turn suggests that there could be languages
where the syntactic containment relationship is also morphologically transparent.
This is indeed what happens in Udmurt, where the Egressive
case ending -i<
´ˇsen morphologically contains the Elative case i<
´ˇs, marking
Coinitial paths.
62 DECOMPOSING PATH 4.6
(25) Udmurt (Winkler 2001)
a. anaj
mother
gurt-i<
´ˇs
village-el
pot-i-z.
leave-pret-3sg
‘Mother came out of the village.’
b. Ižkar-i<
´ˇsen
Iževsk-egr
Moskva-o´ˇz
Moscow-term
pojesdn-en
train-instr
mi<n-i.
go-pret.1sg
‘I went by train from Iževsk to Moscow.’
This containment relationship in Udmurt is not accidental, as the various
allomorphs of the Egressive case (-i<
´ˇsen, -´ˇsen, -la-i<
´ˇsen) correspond to the
different allomorphs of the Elative case (-i<
´ˇs, -´ˇs, -la-i<
´ˇs). I present the way
the Egressive structure is spelled out in Udmurt in (26).
(26) BoundP
Bound
-en
SourceP
Source GoalP
Goal PlaceP
Place DP
Iževsk



-i<
´ˇs
In sum, I propose that delimited Goal and Source paths involve a
special head, which I label Bound, that takes as a complement the corresponding
non-delimited path. I discuss the semantics of the Bound head
in Chapter 5, Section 5.7.
4.6 Conclusion
In this chapter, I established that in some languages:
• Source paths morphologically contain Goal paths.
• Route paths morphologically contain Source paths.
• Non-transitional paths morphologically contain the corresponding
transitional path.
4.6 CONCLUSION 63
• Delimited paths morphologically contain the corresponding nondelimited
path.
The morphological composition of path expressions cross-linguistically
and the assumption that morphological complexity indicates syntactic
complexity led me to propose that the Path head is not a unique projection
hosting directional elements, but should be decomposed into several
heads. Thus, I proposed that the syntactic structure of paths varies
depending on whether we have a Goal-oriented path, a Source-oriented
path, or a non-oriented (Route) path. The corresponding structures are
shown below.
(27) a. Goal paths
GoalP
Goal PlaceP
b. Source paths
SourceP
Source GoalP
Goal PlaceP
c. Route paths
RouteP
Route SourceP
Source GoalP
Goal PlaceP
Similarly, the syntactic structure for non-transitional paths embeds
transitional paths, and the structure for delimited paths embeds the nondelimited
paths.
64 DECOMPOSING PATH 4.6
(28) a. Non-transitional paths
ScaleP
Scale PathtransitionalP
Pathtransitional PlaceP
b. Delimited paths
BoundP
Bound Pathnon-delimitedP
Pathnon-delimited PlaceP
I take these structures to be universal across languages, although in
many languages the real number of heads in the structure is obscured by
lexicalization. Thus, Source paths are built on top of Goal paths in all
languages – even in those where this in not morphologically transparent.
Likewise, Route paths are universally built on top of Source paths, nontransitional
paths on top of transitional paths, and delimited paths on
top of non-delimited paths.
A potential alternative to the multi-layered syntactic structures underlying
the various types of paths is to assume that Path is a single head
with a complex feature content and that morphemes lexicalize (sub-head)
features instead of syntactic terminals. To my mind, this view just shifts
the complexity of the structure to a different level. The subset-superset
relationships between the various types of paths will still hold, but at
the sub-terminal level. Given that we independently need to assume the
operations which generate syntactic trees, it seems more parsimonious to
avoid the need of operations that organize features into complex terminals,
which then are merged together to build syntactic trees.
The consequence of my proposal is that, with the exception of Goal
paths, directional expressions consist of more than a unique Path head
and a locative projection. The Path projection comprises up to five
distinct heads and in the following chapter, I address the question of
what the semantic contribution of each of these heads is.
Chapter 5
Decomposed Path semantics
5.1 Introduction
In the preceding chapter, I argued for a decomposition of the Path projection
in the syntactic structure of directional expressions into multiple
heads, whose number varies depending on what type of path the structure
expresses. I proposed the existence of a Goal, a Source, a Route, a
Scale, and a Bound head. The evidence for the existence of these heads
came from the morphological composition of the respective directional
expressions. In this chapter, I substantiate the proposal by suggesting
a particular semantic function for each of these heads, thus motivating
their existence from the point of view of compositional semantics.
5.2 The semantics of Place
While decomposing the Path head, I have greatly simplified the way
I represent its complement, namely, the Place projection. The label
“Place,” in fact, stands for a myriad of heads, each of which has its share
in the compositional meaning of the entire locative phrase. I will not go
into details here, as they are not of great significance for the proposal
developed in this thesis, but just briefly discuss the topic.
The semantics of locative phrases is a well studied topic. They are
commonly analyzed in terms of regions (Wunderlich 1991, Herweg and
Wunderlich 1991, Nam 1995, Kracht 2002; 2008, Svenonius 2008). Regions
are modeled as contiguous sets of points in space. For instance,
Wunderlich (1991) defines a function p, which gives for each Ground object
the set of points it occupies (its eigenspace). In addition, Wunderlich
defines a family of functions that give for each Ground object a set of
65
66 DECOMPOSED PATH SEMANTICS 5.2
surrounding or neighboring regions. For instance, ext[v] is the external
region of the Ground object (marked by v), while int[v] is the internal
region of the Ground object. Interior location is then defined as in (1),
where u stands for the Figure and v stands for the Ground:
(1) <u,v> ∈ ||in|| iff p[u] ⊆ int[v]
In prose, u is in(side) v, iff the eigenspace of u is a subset of the region
internal to v. This is a simplified formalization of interior location, as
the use of the preposition in in English (and its counterparts in other
languages, see Levinson and Meira 2003) is not limited to configurations
where the Ground object properly contains the Figure. For instance, in
some cases, an apple is judged to be in the bowl even if it is wholly above
the top rim of the bowl (e.g., on top of a pile of apples). I will not go into
details about how this meaning extension can be explained and resolved,
as this will lead me too far astray, and I refer the interested reader to
Herskovits (1986), Feist (2000) and Zwarts (2008b).
Zwarts (1997) and Zwarts and Winter (2000) note that locative phrases
can be modified by measure phrases, as for example in ten meters above
the roof. For that reason they propose to analyze them in terms of vector
spaces. Thus, a locative phrase like above the roof is associated with the
set of vectors which start from the roof and point upwards. The modified
phrase ten meters above the roof then simply denotes the intersection between
two sets — the set of all the vectors that have their starting point
at the roof and point upwards and the set of all vectors that are ten meters
long. Thus, for a simple (unmodified) locative phrase, Zwarts and
Winter (2000) define the following functions:
(2) a. loc – assigns to the Ground the set of points it occupies (i.e.,
its eigenspace)
b. P – assigns to the set of points occupied by the Ground a set
of vectors
c. loc−
– assigns to the set of vectors a set of points, that is,
turns the set of vectors back into a region.
Svenonius (2008) shows how this semantic analysis of locative phrases
can be implemented in syntax and discusses what bits of information
each of the different projections it consist of adds to the structure. I
refer the interested reader to the works cited and will restrain myself
from an exhaustive discussion, as the main contribution of this thesis lies
elsewhere. As to the semantics of locative phrases, I assume that they
denote regions, which is what both approaches agree on at the end of the
5.3 THE SEMANTICS OF GOAL 67
day.
5.3 The semantics of Goal
In this subsection, I turn to the semantic contribution of the Goal head.
The Goal head takes as a complement a stative Place projection, just like
the Path head in the generally accepted structure [Path [Place]]. It is
therefore worth investigating what semantics has been proposed for the
Path head in the literature.
The are several approaches to Path semantics. According to the mereological
approach, paths are seen as part structures (Krifka 1998, Piñón
1993). Kracht (2002; 2008) analyzes them as spatiotemporal entities.
Finally, there is a family of approaches that treat paths as atemporal
sequences of locations (Wunderlich 1991, Herweg and Wunderlich 1991),
regions (Nam 1995), or vectors (Zwarts 2005).
Yet another approach is taken in Fong (1997). Fong proposes that
directional predications are not necessarily related to movement in space
only. They can apply to various types of ordered structures, as for example,
static orders with no temporal reference (see (3)), or aspectual
(temporal) orders (see the Finnish example in (4)).
(3) a bridge out of San Francisco (Fong 1997:2)
(4) Finnish (Fong 1997:26)
a. Tuovi
Tuovi
unohti
forgot.3sg
kirja-n
book-acc
auto-on.
car-ill
‘Tuovi forgot the book in the car.’
(lit. Tuovi forgot the book into the car)
b. Tuovi
Tuovi
löysi
found.3sg
kirja-n
book-acc
laatiko-sta.
box-ela
‘Tuovi found the book in the box.’
(lit. Tuovi found the book out of the box)
To account for this, Fong develops a very abstract semantics for directionals.
She posits a diphasic structure for the objects and events
they operate on and suggests that directional phrases are sensitive to the
linear ordering of these phases. Diphasic means roughly that there is a
transition from a phase of not-p (¬p) to a phase where p holds (or vice
versa) and it is monotone, that is, once p is obtained, there cannot be a
development back to ¬p. The semantic interpretation of a directional expression
is dependent on the phase in which its truth is evaluated. Thus,
68 DECOMPOSED PATH SEMANTICS 5.3
Illative (i.e., into) predicates take as interval the monotone development
from ¬p to p, where the truth of the predicate loc-in(a,b) is evaluated in
the second phase. Elative (i.e., out of ) predicates take the same diphasic
interval, however the truth of loc-in(a,b) is evaluated in the first phase.
In this way, Fong can account for the use of directional cases in Finnish
in non-motion contexts. The difference between Finnish and English directionals
is that Finnish directional cases encode phasal transitions, but
not necessarily spatial transitions, while English directional PPs always
define spatial transitions.
I agree with Fong’s proposal that directional prepositions, like into
and out of, involve a transition. I therefore suggest that the semantic
contribution of the Goal head is to encode transition from one region
to another. This proposal is further motivated by the fact that Goal
paths have a two-stage structure: a negative phase and a positive phase
where the locative relation between Figure and Ground obtains (see (5),
repeated from (5), Chapter 1).
(5) Goal paths
− − − − − + + + + +
0 1 Zwarts (2008a)
This suggestion is also compatible with analyses developed by Wunderlich
(1991), Zwarts (2005; 2008a) and Kracht (2008), who have transitions
as part of the (Goal and Source) path semantics in one way or another.
Wunderlich (1991), for instance, introduces a change predicate
constant, which expresses transition from one region to another. This
change predicate is part of the general scheme for directional prepositions
proposed by Wunderlich.
(6) λv, λu change(d, loc(u, r[v])) (Wunderlich 1991:603)
In (6), d specifies some dimension in which the transition takes place.
In the first phase of d, loc(u, r[v]) is false, and in the second phase
of d, loc(u, r[v]) is true. d is very often a temporal dimension that is
mapped onto a spatial dimension, e.g., in Mary ran into the room. d can
be, nevertheless, spatial only, as in It is raining into the room. When d
is temporal, then change[d, φ] becomes equivalent to Dowty’s (1979)
become operator.
(7) [become φ] is true at an interval I iff (i) there is an interval J
containing the lower bound of I such that ¬φ is true at J, (ii)
there is an interval K containing the upper bound of I such that φ
5.4 THE SEMANTICS OF SOURCE 69
is true at K ..., there is no non-empty interval I such that I ⊂ I
and conditions (i) and (ii) hold for I as well as I.
Thus, with the help of a change or a become operator, we can capture
the semantics of the Goal head. It will then encode a single punctual
transition from a phase where a particular locative relation does not
hold to a phase where it holds.
To sum up, the semantic content of the Goal head is that of transition
to the spatial region encoded by the Place projection in its complement.
The location denoted by PlaceP consequently holds of the end-point p(1)
of the Goal path.
5.4 The semantics of Source
Let us now turn to the semantics of the Source head. Recall that Goal
and Source paths group together to the exclusion of Route paths by the
property of being mono-transitional (i.e., they both contain one transition).
As I proposed in Section 4.2, in syntax, the Source head takes as
a complement a Goal phrase. The Goal head already expresses a transition,
therefore, the semantics of the Source head cannot be that of a
transition, as that would lead to a path with two transitions, and this is
not a Source path.
As discussed in Chapter 2, Section 2.7, Goal paths and Source paths
are constructed in the same way, but are the mirror images of each other.
In other words, Source paths can be seen as the opposite of Goal paths
(as also suggested in Zwarts 2005; 2008a). Therefore, I propose that
the Source head is the locus of a semantic reversal operation. Thus,
the Source head just reverses the orientation of the path provided by
the [Goal [Place]] configuration keeping all other things equal. More
precisely, the Source head assigns to each point i in the interval [0,1]
the position that is assigned to 1−i in the denotation of the Goal path,
where 0 and 1 represent the starting point and the end-point of the path,
respectively. Thus, the starting point p(0) in the Source path will be
assigned the position of the end-point p(1) in the Goal path. Likewise,
p(1) in the Source path will be assigned the position of p(0) in the Goal
path. The result is that the positively located end-point in the Goal path
becomes the starting point of the Source path. In this way, the spatial
domain encoded by the Place head will hold of the first phase of the
Source path, leading to a path of the type +++−−−.
The reversal encoded by the Source head resembles a negation func-
70 DECOMPOSED PATH SEMANTICS 5.4
tion. In this sense, my proposal is similar to the treatment of Source
phrases by Arsenjević (2006) who analyses the so-called Source modifiers
(for instance Slavic spatial prefixes like iz- ‘from’) as being more
complex than Goal modifiers and crucially involving negation of Goals.
In the same line, Svenonius (2009b) suggests that English Source particles
like out and off are endowed by the feature neg, accounting for the
observation that they license NPIs, which otherwise occur in the scope
of negation. The same idea is also expressed by Dirr (1905), cited in
Kurbanov (1990), and in the Tabasaran grammar by Magometov (1965).
These authors suggest that the special element ud-, which attaches to
locative-directional verbal prefixes to derive source-oriented prefixes in
Tabasaran, is a “negative particle” (Dirr 1905) or an element related to
negative particles (Magometov 1965).
In sum, I suggest that Source paths are semantically more complex
than Goal paths in that the former involve an additional reversal operation.
One could, however, imagine the opposite scenario where Goal
paths are built on top of Source paths. This raises the question of whether
the greater complexity of Source paths is accidental or it is conditioned
by some factors, possibly lying outside the domain of syntax proper. In
the following paragraphs, I develop a tentative suggestion that the more
complex syntax and semantics of Source paths reflects a general property
of human cognition.
My proposal consists in relating the greater complexity of Source
paths to the so called Goal bias, established by Lakusta (2005), Lakusta
and Landau (2005), Assadollahi et al. (2006), inter alia. The Goal bias
is the term for the natural preference to encode end-points (Goals) over
starting points (Sources) in the representations of events. Thus, both
children and adults prefer to linguistically express Goals over Sources
when describing motion events, change of state events, change of possession
events, etc. This perceptual and attentional asymmetry is reflected,
in addition, in language acquisition (see Lakusta and Landau 2005 and
references therein). On the basis of experimental data, Lakusta and Landau
conclude that the Goal bias is a conceptual one and suggest that it
also influences the grammar.
A possible way in which the Goal bias is reflected by the grammar can
be found in Ikegami’s (1982) “Goal over Source principle.” The formulation
of this principle is motivated by Ikegami’s observation that languages
are not symmetrically organized with respect to the notions Source and
Goal. An example of this asymmetry is the fact that Source markers are
consistently a marked term in relation to the Goal marker, e.g., the Goal
5.5 THE SEMANTICS OF ROUTE 71
reading of the English adverbs here and there does not require special
marking, while the Source reading requires the preposition from (from
here and from there).1
The Goal-Source asymmetry in grammar has also
been observed by Levin (1993) who notices that some verbs take a Source
argument only in the presence of a Goal argument.
(8) a. The witch turned him from a prince into a frog.
b. The witch turned him into a frog.
c. *The witch turned him from a prince.
My suggestion is that Source paths are construed as more complex than
Goal paths (and not the other way around) because of this Goal-Source
asymmetry in the cognitive system. In other words, Goal paths, which are
the preferred choice in describing a motion event, correspond to a simpler
syntactic and semantic structure. Source paths, which are apparently
conceptually more complex, correspond to richer syntactic and semantic
structure. From this proposal it follows that the Goal bias is in fact
hardwired in the grammar in that it is reflected by distinct structures
for the two kinds of paths (see, however, Gehrke 2008 for the opposite
view).
5.5 The semantics of Route
In this subsection, I turn to non-oriented paths, derived in syntax by
adding a Route head on top of the Source path structure. A Source path
is, as already discussed, mono-transitional and has its first phase as the
positive phase.
(9) Source path
+ + + + + − − − − −
0 1 Zwarts (2008a)
Route paths are bi-transitional and they have the positive phase “in the
middle,” that is, the locative relation between the Figure and the Ground
holds of some intermediate points of the path. On the basis of these
observations, I suggest that the Route head is another transitional head
which encodes a transition to a positive phase. Applied to a Source path,
1
Interestingly, Ikegami includes in his discussion of the markedness of Source the
adversative prefix un-, suggesting that a verb like bind is related to the notion of Goal,
while un-bind is related to the notion of Source. This is in line with the proposal that
the notion of Source includes a reversal semantics.
72 DECOMPOSED PATH SEMANTICS 5.5
the Route head leads to a path that has the following shape:
(10) Route path
− − − − + + + + − − − −
0 1 Zwarts (2008a)
The whole computation is represented in (11), step by step.
5.5 THE SEMANTICS OF ROUTE 73
(11) The syntactic and semantic derivation of a Route path
a. [Place ...]
b. merger of Goal →
c. [Goal [Place ...]] representing a path of the shape −−−+++
d. merger of Source →
e. [Source [Goal [Place ...]]] → reversal of Goal path → +++−−−
f. merger of Route →
g. [Route [Source [Goal [Place ...]]]] → adding a second transition
→ −−−+++−−−
This particular proposal for Routes reflects the bi-transitional character
of bounded Route paths: the first transition — from the positive (middle)
phase to the negative (final) phase — is contributed by the [Source [Goal
Place]] structure; the second transition — from the negative (first) phase
to the positive (middle) phase — is the contribution of the Route head.
If [Route [Source [Goal [Place ...]]]] is the right universal syntaxsemantic
structure for paths, we have a possible explanation for the
nonexistence of “return prepositions” such as *tsap in (11), Chapter 2,
which are supposed to be of the shape +++−−−+++. In a past Route
path, the location to which the path relates holds of the middle (positive)
part of the path. The first transition, contributed by the Route head, is
from a negative to a positive phase, just like in a Goal path. The second
transition is from a positive to a negative phase, just like in a Source
path. Thus, as already discussed in Chapter 2, a past Route path can be
seen as the concatenation of a Goal path with a Source path. In syntax
this is reflected by the Route head encoding a transition to a positive
phase, like a Goal head, and taking as a complement a Source phrase.
In a *tsap-path, the location to which the path relates holds of the
beginning and the end of the path and the middle portion of the path is
a negative phase. The first transition in a *tsap-path is therefore from
a positive to a negative phase, just like in a Source path. The second
transition is from a negative to a positive phase, just like in a Goal path.
What we then need, in order to derive such a path, is that the Route head
encodes a transition to a negative phase and takes as a complement a Goal
path. The Route head, however, encodes a transition to a positive phase,
hence it cannot apply directly to GoalP, as the semantic computation of
the structure [ Route [ GoalP ]] will fail.
The question then boils down to why there is no transitional head in
addition to the transitional heads Goal and Route that, unlike Goal and
Route, encodes a transition to a negative phase and hence can take as a
complement a Goal structure. Under the view developed in this thesis,
74 DECOMPOSED PATH SEMANTICS 5.6
a transition of the type +++−−− is a derived transition, the result of
reversing a Goal-type transition to a positive phase −−−+++. This
implies that transitions to a positive phase are in a sense “more basic,”
which is, in fact, expected given the Goal bias discussed in the preceding
section. Thus, I propose that the lack of a transitional head encoding
transition to a negative phase by itself is due to the derived status of
such a transition, obtainable only by reversing a Goal-type transition.
Note that even if we add another reversing head on top of the Route
head in the structure [Route [Source [Goal [Place ...]]]], we will not derive
a *tsap-path of the shape +++−−−+++. The result will be again a
past-path −−−+++−−−. The semantics I proposed for the reversing
Source head is that it assigns to each point i in the interval [0,1] the
position that is assigned to 1−i in the denotation of the path in the
complement position, i.e., the Goal path. When we apply this reversal
operation to a Route path −−−+++−−−, what we get is the following:
the starting point 0 of the reversed Route path will be assigned the position
assigned to the endpoint 1 of the non-reversed Route path (taking
i = 0, 1 − 0 = 1), that is, a minus value. The endpoint 1 of the reversed
Route path will be assigned the position assigned to the starting point
0 in the non-reversed Route path (taking i = 1, 1 − 1 = 0), that is, a
minus value again. Hence, a reversed Route path will begin and end with
a negative phase, just like a non-reversed Route path and, crucially, will
not lead to a path of the shape +++−−−+++.
Before concluding this section, let me say a couple of words about
a prediction made by the Route structure defended here regarding the
conceptual bias to encode certain types of paths over others. Recall that
at the end of the preceding section, I suggested that there is a correlation
between the preference for Goal paths over Source paths and the greater
complexity of the syntax-semantic structure underlying the latter. If
this is correct, then we expect that the even greater complexity of Route
paths will correlate with a disinclination to use Route paths compared
to both Source and Goal paths. Indeed, the psycholinguistic study of
Stefanowitsch and Rohde (2004) shows that Source PPs, although less
frequent than Goal PPs in motion event representations, still occur more
often than Route PPs (called trajectory in their terminology). Thus, if
the suggested correlation holds, then this can be seen as indirect evidence
for the proposal that Route paths are built on top of Source paths.
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 75
5.6 The semantics of non-transitional paths
In this section, I turn to non-transitional paths and investigate the semantic
contribution of the Scale head I proposed in Chapter 4, Section 4.4.
In order to remind the reader of the terminology, I present an overview
of the transitional and non-transitional paths in Table 5.1, repeating the
relevant rows from Table 2.2.
Goal Source Route
Transitional
Cofinal Coinitial Transitive
to X from X past X
−−−−++++ ++++−−−− −−−+++−−−
0 1 0 1 0 1
Approximative Recessive Prolative
Non- towards X away from X along X
transitional −−−−−−−−− −−−−−−−−− +++++++++
0 1 0 1 0 1
Table 5.1: Transitional and non-transitional paths
In the preceding chapter, I proposed that, syntactically, non-transitional
Goal, Source and Route paths are derived on the basis of the
corresponding transitional path by the addition of a Scale head, as shown
in (12).
(12) a. Non-transitional Goal (Approximative) paths
ScaleP
Scale GoalP
Goal PlaceP
76 DECOMPOSED PATH SEMANTICS 5.6
b. Non-transitional Source (Recessive) paths
ScaleP
Scale SourceP
Source GoalP
Goal PlaceP
c. Non-transitional Route (Prolative) paths
ScaleP
Scale RouteP
Route SourceP
Source GoalP
Goal PlaceP
I labeled this head Scale, because, in the case of oriented (Goal and
Source) non-transitional paths, it imposes a specific ordering on the
points of the path in its complement. More specifically, Approximative
paths are such that at each subsequent point in the interval [0,1]
the Figure is closer to the Ground. Recessive paths are the mirror image
of Approximative paths, such that at each subsequent point in the
interval [0,1], the Figure is farther away from the Ground. I suggested
that graphically Approximative and Recessive paths are represented as
a sequence of minuses, where the darker shade of the minus reflects a
greater distance to the location denoted by PlaceP (see the last row in
Table 5.1).
In the case of Prolative paths, the Scale head applies to a bi-transitional
path structure to return a sequence of points where the location encoded
by PlaceP holds of each of the points in the newly derived path. Thus,
Prolative paths are different from Approximative and Recessive paths
in that the former contain just plusses (i.e., at each point the locative
relation between the Figure and the Ground holds), while the latter two
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 77
contain just minuses (i.e., at no point in the path does the locative relation
encoded by PlaceP hold). Semantically then, when applying to a
Transitive path, the Scale head has to pick out (an interval from) the
positive phase of the path in its complement, so that the result is a path
made just out of plusses. Conversely, when applying to Coinitial and Cofinal
paths, the Scale head has to pick out (an interval from) the negative
phase, so that we get a path containing just minuses.
Before phrasing this idea in more formal terms, let me introduce an
“auxiliary” function Val which assigns the values 1 and 0 to the points in
the path, depending on whether they are positively or negatively located,
viz., depending on whether the location denoted by PlaceP holds, or does
not hold of the point p(i), respectively.
(13) a. Let Val(p(i))=1 iff p(i) is a positively located point.
b. Let Val(p(i))=0 iff p(i) is a negatively located point.
With the function Val in hand, I propose the following semantics for the
Scale head.
(14) The Scale head is a function from a path p to a path p which
picks out the unique interval I, for which
a. if p(0) is negative, then p (1)=p(in), such that
Val(p(in))=Val(p(in+1))
b. if p(1) is negative, then p (0)=p(in), such that
Val(p(in))=Val(p(in-1))
What (14) says in prose is that the Scale head selects a portion from the
path in its complement such that it is to the left and to the right of the
transition and this portion corresponds to the new path. The transition
is defined in (14) as the point where the value at a point in is different
from the value of the next point in+1, or the preceding point in-1. Note
that, in this way, the Scale head also cuts out the transition itself, thus
capturing the fact that Approximative, Recessive and Prolative paths do
not involve any transition.
Let us now discuss in more detail precisely what intervals the Scale
head picks out when operating on the various kinds of transitional paths.
Starting with Cofinal (Goal) paths, we have to apply the rule in (14a),
because p(0) is negative in these paths. The new path p will be such
that its end-point, p (1), will be the last point before the transition, that
is the point p(in) which has a different value from the point right after it
p(in+1). Thus, the interval picked out by the Scale head will be entirely
in the negative phase of the Cofinal path, either a sub-interval of it, or
78 DECOMPOSED PATH SEMANTICS 5.6
the whole negative phase, as shown in (15a).
When the path on which the Scale head operates is a Coinitial (Source)
path, the rule in (14b) is the relevant one, as Coinitial paths have the endpoint
p(1) as the negative point. In the new path p, the starting point
p (0) will be the first point after the transition, that is the point p(in)
which has a different value from the point before it p(in-1). Thus, again
the interval selected by the Scale head will lie entirely in the negative
phase, as shown in (15b).
Finally, when applying to Route paths, both rules in (14a) and (14b)
are relevant, as Route paths have both extreme points defined as negative
location. This means that the interval picked out by the Scale head has to
be both to the left and to the right of the transitions. The unique interval
which matches these conditions is the positive phase of the Route path,
as it both precedes and follows a transition. Thus, the end-point p (1)
of the derived Prolative path will be the last point in the positive phase
and the starting point p (0) will be the first point of the positive phase
of the Transitive path, as shown in (15c).
(15)
a. Scale applied to a GoalP
−−−−−+++++
0 i5 i6 1
b. Scale applied to a SourceP
+++++−−−−−
0 i5 i6 1
c. Scale applied to a RouteP
−−−++++−−−
0 i3 i4 i7 i8 1
Note, however that the shape of the non-transitional paths suggested
in (15) fits only with the graphic representation of non-transitional Route
paths, like the one encoded by the English preposition along, which are
construed as a sequence of points at each of which the location encoded
by PlaceP holds (see (16), repeated from (19), Chapter 2).
(16) along-Path
+ + + + + + + + +
0 1 Zwarts (2008a)
When it comes to non-transitional Goal and Source paths, they are not
only a sequence of points at which the Figure is not at the Ground,
but there is in addition an ordering between these points such that the
distance between the Figure and the Ground decreases or increases. This
necessitates yet another add-on to the semantics of the Scale head. What
I suggest is that the Scale head maps the picked out negative interval
on a gradient closeness scale, such that the “closer” a point was to the
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 79
positively located extreme point in the original path p, the closer the
Figure is located to the Ground at that point. In more technical terms,
the ordering of the points is defined as follows.
(17) a. if p(1) is positive, then for each pair p (i) and p (j), if
i<j, then at p (j) the Figure is closer to the Ground than
at p (i).
b. if p(0) is positive, then for each pair p (i) and p (j), if
i<j, then at p (i) the Figure is closer to the Ground than
at p (j).
The condition in (17a) gives us an Approximative path of the shape in
(18).
(18) − − − − − − − − −
0 1
While the condition in (18b) derives a Recessive path, shown in (19).
(19) − − − − − − − − −
0 1
Note that, in the case of a Prolative path, derived on the basis of a Transitive
path by the application of the Scale head, the conditions in (17)
do not play a role. The reason is that the extreme points of a Transitive
path are negatively defined. Since the antecedents in the material implications
in (17) are therefore false, no inference can be made concerning
the consequents. This means that we do know know anything as to the
comparative distance between the Figure and the Ground at the points
p(i) or p(j). That is, the Figure can be equidistant from the Ground at
p(i) and p(j), or closer to the Ground at one of the two points or the
other. This is a welcome result, given that a path like along the river
does not necessarily involve points where the distance between the Figure
and the river is constant or increases/decreases monotonically. As
discussed by (Zwarts 2005:752), at some of the points, the Figure can be
farther away from the river than at others, but the path still counts as
an along-path.
Let us now turn to the position of the Scale head in the sequence of
syntactic heads. As shown in (12), it can appear on top of a Goal, a
Source or a Route head and the question is how to capture its varying
positioning. One way is to assume that there is a functional sequence in
the sense of Cinque (1999), where heads are arranged in a rigid hierarchical
order. The Scale head will then have a fixed position above the Route
80 DECOMPOSED PATH SEMANTICS 5.6
projection. This will give the hierarchy in (12c) as the full-blown structure.
However, this is the structure for a non-transitional Route path.
In order to explain how non-transitional Source and Goal paths come to
be, one will have to assume that certain projections from the “middle” of
the structure can be missing. Thus, a non-transitional Source (Recessive)
path will be basically part of the same functional sequence, but, crucially,
the Route projection will be absent. Similarly, a non-transitional Goal
(Approximative) path will be represented by a syntactic structure where
the Route and the Source projections are not there.
Another way to deal with the placement of the Scale head in the
structure is to assume that Scale does not have a fixed position in the
functional sequence but can appear anywhere, as long as its complement
is a kind of a path: a RouteP, a SourceP, or a GoalP. In this way Scale will
resemble Cinque’s (1999) Neg head, which, too, can appear in different
positions in the functional sequence. This assumption entails that Scale
can appear between Source and Goal and between Route and Source. The
possible positions for Scale are given in the structure in (20).
(20)
<Scale>
Route
<Scale>
Source
<Scale>
Goal PlaceP
The consequence of a multiple positioning of the Scale head is that now
we have two ways to construct a Recessive path. Consider the following
two scenarios.
(21) Non-transitional Coinitial path: We first construct a Cofinal
path, then apply the reversing Source head to get a Coinitial path
and then apply the Scale head to get a Recessive path.
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 81
ScaleP
Scale SourceP
Source GoalP
Goal PlaceP
(22) Reversed Approximative path: We first construct a
Cofinal path, then apply the Scale head to get an Approximative
path and then apply the reversing Source head to
get a Recessive path.
SourceP
Source ScaleP
Scale GoalP
Goal PlaceP
The two structures underlying a Recessive path will have the same semantics.
In the first case, (21), the Scale head applies to a Coinitial path
and picks out an interval from the negative phase, and, for each pair p(i)
and p(j), if i<j, the location of the Figure at p(i) will be closer to the
Ground than at p(j) (see the rules in (14b) and (17b)). This leads to a
path with the following graphic representation.
(23) − − − − − − − − −
In the second case, (22), the Scale head applies to a Cofinal path to derive
an Approximative path in a way analogical to the one just described
for Coinitial path (but by applying the rules in (14a) and (17a)). The
reversing Source head then operates not on a transitional Cofinal path
(as discussed in Section 5.4), but on an Approximative path of the shape
in (24).
(24) − − − − − − − − −
0 l k 1
5m 3m
Applying the Source head to the path in (24) will derive the path repre-
82 DECOMPOSED PATH SEMANTICS 5.6
sented in (23). Recall that the Source head assigns to each point i in the
output path p the position at point 1−i in the input path p. Let p (i) be
the third point in the output path. It will be then assigned the position
at point 1 − i in the Approximative path in (24), that is, the position at
the third last point, called p(k) for short. Similarly, let p (j) be the third
last point in the output path p. It will be assigned the position at the
third point in the input path, called p(l) for short.
(25) a. p(k) → p (i)
b. p(l) → p (j)
To make things less abstract, suppose that, at p(l) the Figure is 5 meters
away from the Ground and, at p(k), it is 3 meters away from the Ground.
Then, in the reversed path in (26), at p (i) the Figure will be 3 meters
away from the Ground and, at p (j), it will be 5 meters away from the
Ground. In other words, the distance between the Figure and the Ground
will be getting bigger the closer we are to p(1) and the thus derived path
will have the following shape.
(26) − − − − − − − − −
0 i j 1
3m 5m
To sum up, both scenarios in (21) and (22) lead to a Recessive path of
the same type and the question is which of the two possible solutions to
the positioning of the Scale head is the correct one, the missing heads
solution or the varying positioning of Scale solution.
The empirical data suggests that the second solution is on the right
track, as it correctly captures the variation in the morphological expression
of Recessive paths. Both strategies in (21) and (22) are represented
in the languages of the world.
There are languages where Recessive paths are built by having the
Scale head apply to an already constructed Coinitial path. Such languages
are Tabasaran, where this head is lexicalized by the morpheme
-di, and Avar, where Scale is spelled out by the morpheme -Gun, as shown
in (27) and (28), respectively.
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 83
(27) Tabasaran (Hanmagomedov 1967)
a. a@rabaji-’-na
cart.erg-in-all
‘into the cart’
(Cofinal)
b. a@rabaji-’-na-di
cart.erg-in-all-scale
‘towards the cart’
(Approximative)
c. a@rabaji-’-an
cart.erg-in-abl
‘out of the cart’
(Coinitial)
d. a@rabaji-’-an-di
cart.erg-in-abl-scale
‘away from the cart’
(Recessive)
(28) Avar (Uslar 1889)
a. wáccass-de
brother.erg-on.all
‘onto the brother’
(Cofinal)
b. wáccass-de-Gun
brother.erg-on.all-scale
‘towards the brother’
(Approximative)
c. wáccass-da-ssa
brother.erg-on-abl
‘from the brother’
(Coinitial)
d. wáccass-da-ssa-Gun
brother.erg-on-abl-scale
‘away from the brother’
(Recessive)
The a-examples in (27) and (28) show a transitional Goal path. The
b-examples illustrate how the addition of the morpheme -di and -Gun
to the transitional Goal path derives an Approximative path. The cexamples
represent a transitional Source (Coinitial) path. And, finally,
the d-examples show that Recessive paths are built by adding the same
“Scale” morpheme -di and Gun to the Coinitial path — the case we are
interested in here.
Note that, while the Scale, Allative and Locative morphemes in the
Tabasaran examples in (27) spell out one head each, the Ablative case
suffix -an spells out two heads — Goal and Source. This is a possibility I
discussed in Chapter 3, Section 3.3, where I suggested the need of a theory
which allows a single morpheme to spell out multiple heads. Similarly,
in Avar, the Ablative -ssa lexicalizes both Goal and Source. As to the
Avar Allative -de, one can in principle assume that it, too, lexicalizes two
heads — Place and Goal, However, the composition of the Allative in the
other locative series in Avar (see Table 4.4) rather suggests that -de is a
bimorphemic marker -da-e, where the vowel /a/ does not surface because
of the following /e/. Below, I present the syntactic structure underlying
the Tabasaran Recessive expression and its lexicalization (postponing
84 DECOMPOSED PATH SEMANTICS 5.6
the discussion of the precise linearization of the elements until Chapter
7, Section 7.2).
(29) ScaleP
Scale
-di
SourceP
Source GoalP
Goal PlaceP
a@rabaji’



-an
The second possible way to construct a Recessive path, shown in (22),
is found in languages like Dime and Bulgarian, where the Source head
applies to a pre-constructed Approximative path. For an illustration,
see the data in (30) and (31), repeated from Chapter 4, (5) and (6),
respectively.
(30) a. šiftaye
Shiftaye
taddeseka-bow
taddese.com-dir
tiN-i-n.
go-perf-3
‘Shiftaye went towards Taddese.’
b. šiftaye
Shiftaye
taddeseka-bow-de
taddese.com-dir-abl
Pád-i-n.
come-perf-3
‘Shiftaye came from the place where Taddese is found.’
(31) a. k@m
towards
k@shta-ta
house-def
‘towards the house, in the direction of the house’
b. ot-k@m
from-towards
k@shta-ta
house-def
‘from the direction of the house’
Here, again, we have an instance of one morpheme spelling out multiple
heads, namely, the approximative -bow and k@m, which lexicalize Place,
Goal and Scale. In (32), I illustrate how the base structure is lexicalized
in Dime and Bulgarian, again deferring the discussion of the linearization
to Section 7.2.
5.6 THE SEMANTICS OF NON-TRANSITIONAL PATHS 85
(32) a. Dime
SourceP
Source
-de
ScaleP
Scale GoalP
Goal PlaceP
Place DP
taddesseka
b. Bulgarian
SourceP
Source
ot
ScaleP
Scale GoalP
Goal PlaceP
Place DP
k@shtata



-bow



-k@m
To recapitulate, the hypothesis that the Scale head can appear in
one of several positions in the functional sequence makes the correct
prediction that Recessive paths can be constructed in two ways: as nontransitional
Coinitial paths (Tabasaran, Avar) and as reversed Approximative
paths (Dime, Bulgarian).
When it comes to the structure of non-transitional Route paths, there
are three possible positions for the Scale head in the sequence of heads –
between Goal and Source, between Source and Route and above Route.
The respective structures are shown in (33a-c).
(33) a. RouteP
Route SourceP
Source ScaleP
Scale GoalP
Goal PlaceP
86 DECOMPOSED PATH SEMANTICS 5.7
b. RouteP
Route ScaleP
Scale SourceP
Source GoalP
Goal PlaceP
c. ScaleP
Scale RouteP
Route SourceP
Source GoalP
Goal PlaceP
From the three structures above only the one in (33c) is semantically
computable. Recall from Section 5.5 that the Route head encodes a
transition to a positive phase. In (33a) and (33b), however, the Route
head takes as a complement a Recessive path, which has no such positive
phase (see the graphic representation of Recessive paths in Table 5.1).
For that reason, the semantic computation of (33a-b) will fail. Hence,
the only way to construct a Prolative path is the one in (33c), where first
the Route head applies to a Coinitial path, whose first phase is positive,
and the Scale head subsequently applies to the Transitive path.
To sum up this section, I suggested that the semantic role of the
Scale head is to pick out a unique interval from the path structure below
it, such that it is to the left and to the right of the transition(s). In
addition, it imposes an order in the selected interval such that, at the
points “closer” to the positive extreme points (1 for Goal paths and 0
for Source paths), the Figure is closer to the Ground than at the points
“farther away” from the positive extreme point. Since the extreme points
in Route paths are not positive, there is no mapping on a closeness scale
in non-transitional Route paths.
5.7 THE SEMANTICS OF DELIMITED PATHS 87
5.7 The semantics of delimited paths
I now turn to the semantic function of the last head I proposed in the
decomposed Path structure — the Bound head which is part of the syntax
of Goal and Source-oriented delimited paths (see the structures in (34),
repeated from Chapter 4).
(34) a. Delimited Goal (Terminative) paths
BoundP
Bound GoalP
Goal PlaceP
b. Delimited Source (Egressive) paths
BoundP
Bound SourceP
Source GoalP
Goal PlaceP
As discussed in Chapter 2, delimited paths split into Goal-oriented delimited
paths, which I called Terminative, and Source-oriented delimited
path, which I labeled Egressive. The two kinds of delimited paths contrasted
with Cofinal and Coinitial paths in that delimited paths set the
location denoted by PlaceP as the limit of movement (see Table 4.7 for
an overview over the four kinds of paths). The minimal pair in (35), repeated
from (25) and (26), Section 2.5, illustrates the difference between
a Terminative and a Cofinal path.
(35) a. The boy ran up to the house. (Terminative path)
b. The boy ran to the house. (Cofinal path)
As I suggested in Chapter 2, the difference in the meaning between the
two sentences is that, in (35a), the boy runs as far as the house, that
is, once he reaches it, he stops, whereas in (35b), the boy can continue
along a path that is contained within the location. I proposed that
88 DECOMPOSED PATH SEMANTICS 5.7
the distinction between the two types of paths is that the Terminative
one involves just one point (more precisely, the end-point) where the
locative relation between the Figure and the Ground holds, hence the
interpretation that the location is the limit for the movement along the
path. The positive phase of Cofinal paths, by contrast, can contain a
whole sequence of positive locations, therefore there is no such effect.
Below, I present the graphic representation of the paths in (35a) and
(35b), repeated from Chapter 2, Section 4.7.
(36) Terminative path
− − − − − − − − − +
0 1
(37) Cofinal path
− − − − − + + + + +
0 1
The Source-oriented Egressive paths are the mirror image of Terminative
paths, i.e., paths where the location specified by the Ground is true only
of the starting point of the path (see (38)), thus contrasting with Coinitial
paths (see (39)).
(38) Egressive path
+ − − − − − − − − −
0 1
(39) Coinitial path
+ + + + + − − − − −
0 1
If this suggestion is on the right track, then the semantic function of the
Bound head is to “cut away” all but one point from the positive phase
of the path in its complement. In other words, I suggest that the Bound
head picks out an interval from the path structure it applies to which
contains the negative phase and one point from the positive phase. This
can be formalized as follows.
(40) The Bound head is a function from a path p to a path p which
picks out an interval I, for which
a. if p(1) is positive, then p (1)=p(in), such that
Val(p(in))=Val(p(in-1))
b. if p(0) is positive, then p (0)=p(in), such that
Val(p(in))=Val(p(in+1))
5.7 THE SEMANTICS OF DELIMITED PATHS 89
What (40) says in simple words is that, in the case of Goal paths, where
the end-point p(1) is positive, the Bound head selects an interval whose
end-point is the first point after the transition. In this way, we ensure
that the end-point in the new path is a positively defined point and that
it is unique. Similarly, in Source paths, where the starting point p(0)
is positive, the picked out interval begins with the last point before the
transition, thus leading to a path which begins with a unique positive
point, followed by a transition to a negative phase. Graphically, the
selected intervals can be represented as follows:2
(41)
a. Bound applied to a GoalP
−−−−−+++++
0 i5 i6 1
b. Bound applied to a SourceP
+++++−−−−−
0 i5 i6 1
According to (40), the Bound head does not operate on Route paths,
which have both p(0) and p(1) negatively defined. This correctly captures
the fact that there are no delimited counterparts to Route paths, as
discussed in Chapter 2, Section 4.7.
Before concluding this section, one more remark is in order. Assuming
that the Bound head does not have a fixed position in the Path fseq just
like the Scale head, there are two possibles ways to construct an Egressive
path. Those are presented in (42) and (43) below:
(42) Delimited Coinitial path: We first construct a Cofinal path,
then apply the reversing Source head to get a Coinitial path and
then apply the Bound head to get an Egressive path.
BoundP
Bound SourceP
Source GoalP
Goal PlaceP
2
According to the formulation in (40), it is left unspecified whether the whole
negative phase is included in p, as shown in (41), or just a part of it. In both cases p
will be a delimited path.
90 DECOMPOSED PATH SEMANTICS 5.8
(43) Reversed Terminative path: We first construct a Cofinal
path, then apply the Bound head to get a Terminative path and
then apply the reversing Source head to get an Egressive path.
SourceP
Source BoundP
Bound GoalP
Goal PlaceP
We already saw an example of the Egressive path built according to the
strategy in (42). The example was provided by the Permic language
Udmurt in Chapter 4, Section 4.5, where the Egressive marker morphologically
contained the Coinitial Elative case marker. I repeat the data
below.
(44) Udmurt (Winkler 2001)
a. anaj
mother
gurt-i<
´ˇs
village-el
pot-i-z.
leave-pret-3sg
‘Mother came out of the village.’
b. Ižkar-i<
´ˇsen
Iževsk-egr
Moskva-o´ˇz
Moscow-term
pojesdn-en
train-instr
mi<n-i.
go-pret.1sg
‘I went by train from Iževsk to Moscow.’
So far I haven’t encountered a language which forms Egressive paths according
to the scenario in (43). This is not surprising given that Egressive
paths seem to be rather rare in the languages of the world. In any case,
the lack of data does not disconfirm the hypothesis developed here.
5.8 Conclusion
In this chapter, I discussed how the meaning of each of the paths presented
in Chapter 2 can be compositionally derived in the decomposed
path structure I argued for in Chapter 4. I discussed each of the heads
claimed to be part of the structure underlying directional expressions and
proposed the following semantics:
5.8 CONCLUSION 91
• The Place projection encodes a spatial region.
• The Goal head encodes transition to the spatial region denoted by
PlaceP
• The Source head is the locus of a semantic reversal operation which
reverses the orientation of the Goal path in its complement position.
• The Route head introduces a second transition to the first (positive)
phase of the Source path below it.
• The Scale head picks out a unique interval from the path structure
below it, such that it is to the left and to the right of the transition(s).
In addition, it imposes an order in the selected interval
such that the “closer” a point is to the positive extreme points, the
shorter the distance between the Figure and the Ground is at that
point.
• The Bound head picks out an interval from the path structure below
it which ends or begins with a unique positive point. This positively
located extreme point of the path is interpreted as the limit for
movement.
I also discussed the positioning of the five “Path-heads” in a binary
branching syntactic structure. I suggested that the Route, Source, Goal
and Place heads are rigidly ordered in the syntactic hierarchy, while the
Scale and Bound heads do not have a fixed position. Schematically,
the relations between the various kinds of paths can be represented as
shown in Figure 5.1, where −→ stands for the relation morphological
containment or syntactic dominance. Figure 5.1 also reflects the semantic
compositionality of the various kinds of paths, for instance, Prolative
paths are compositionally derived from Transitive paths, which are derived
from Coinitial paths, etc. The arrow connecting Egressive and
Terminative paths is put in brackets, as there is no linguistic evidence
confirming this relation, however, it is predicted to be possible by the
system.
92 DECOMPOSED PATH SEMANTICS 5.8
Egressive (−→) Terminative
Transitive −→ Coinitial −→ Cofinal −→ Location
Prolative Recessive −→ Approximative
↑ ↑ ↑
↓ ↓
Figure 5.1: The relation of syntactic dominance between the various
types of paths
In the next chapter, I address the question of how the Path structure
I propose is lexicalized. I discuss the apparent mismatch in various languages
between the number of heads in the structure and the number of
morphemes that spell them out, which has come up already at a couple of
occasions. Consequently, I adopt and present the Nanosyntatic view on
syntax developed in Tromsø (Starke 2005-2009, Caha 2009b, Svenonius
et al. 2009) and show how this new approach to language structure sheds
light on various phenomena within the realm of directional expressions.
Part II
The Nanosyntax theory of
grammar
93
Chapter 6
Lexicalization of the structure
6.1 Introduction
So far, I have extensively argued for a universal hierarchy of paths, as
described in Chapters 4 and 5. I claimed that each path is associated
with a specific syntactic structure and decomposed the Path projection
proposed in the literature into several smaller heads. In my investigation,
I was led by the assumption, laid out in Chapter 3, according to which
each morphosyntactic feature corresponds to an independent syntactic
head in the functional sequence (Kayne 2005). As a consequence, I assumed
that the number of morphemes which constitute a given spatial
expression can be taken as an indication of how many terminals are contained
in the syntactic structure underlying this spatial expression. In
other words, I suggested that whenever we see that a given spatial marker
consists of more than one morpheme, this means that the corresponding
syntactic structure contains more than one terminal. I illustrate this
assumption on Lak (Daghestanian), see the example in (1).
(1) Lak (Murkelinskij 1967)
a. k@at-lu-vu
house-erg-iness
‘in the house’
b. k@at-lu-vu-n
house-erg-iness-all
‘into the house’
c. k@at-lu-vu-n-maj
house-erg-iness-all-scale
‘towards (the inside of) the house’
95
96 LEXICALIZATION OF THE STRUCTURE 6.1
In Lak, the Approximative marker -n-maj in (1c) is morphologically complex
— it contains the morpheme -maj plus the Allative morpheme -n —
and attaches to the Inessive morpheme -vu. Thus, according to the assumption
about the correspondence between morphemes and terminals,
the syntactic structure lexicalized by the Approximative marker has to
involve (at least) two terminals: one for the morpheme -n and one for
the morpheme -maj. For this reason (albeit based on data from other
languages), I proposed in Chapter 4 that the “dynamic portion” of the
syntactic structure for Approximative expressions contains two heads:
Scale and Goal. Each of these heads is spelled out by one of the morphemes
which make up the Approximative marker: -n and -maj.
(2) ScaleP
Scale
-maj
GoalP
Goal
-n
PlaceP
k@atluvu
In the tree diagram in (2), the terminals Scale and Goal correspond to a
unique morpheme and we have thus a one-to-one mapping between the
morphological composition of the Approximative marker and the syntactic
structure. The picture is, nevertheless, rarely so neat and this is
so because a monomorphemic expression can also stand for a syntactic
structure that contains more than one terminal (recall the case of Dutch
naar discussed in Chapter 3, Section 3.3).
An illustration of such a one-morpheme/multiple-terminals correspondence
can be found in Lak again, once we zoom in on the locative portion
of the spatial expression in (1). As proposed by Svenonius (2006), locative
expressions contain a Place head dominating an AxPart head where
one finds spatial elements encoding notions like exterior, interior,
top, bottom, etc. (see also Muriungi 2006, Takamine 2006, Pantcheva
2006; 2008). The AxPart head in its turn takes a K[ase]P as a complement.
The minimal syntactic structure of locatives is therefore as shown
in (3).
6.1 INTRODUCTION 97
(3) PlaceP
Place AxPartP
AxPart KP
K DP
The locative Lak expression (see (1a)) contains two morphemes: the
Ergative marker -lu and the Inessive marker -vu. Assuming that the
Ergative marker is the lexicalization of the K[ase] head, the Inessive
morpheme is left to stand for both the Place head and the AxPart head.
(4) PlaceP
Place AxPartP
AxPart KP
K
-lu
DP
k@at
-vu



This suggestion is supported also by the semantics of the Inessive
marker, which has a stative meaning and encodes the notion of interior
as opposed to other Lak stative morphemes that encode spatial relations
like on, under, etc. (see Table 6.1).
Location Goal Source Route Approximative
in -vu -vu-n -v-a(tu) -vu-x -vu-n-maj
on -j -j-n -j-a(tu) -j-x -j-n-maj
behind -x -xu-n -x-a(tu) -xu-x -xu-n-maj
under -lu -lu-n -l-a(tu) -lu-x -lu-n-maj
at -č’a -č’a-n -č’-a(tu) -č’a-x -č’a-n-maj
by -c’ -c’u-n -c’-a(tu) -c’u-x -c’u-n-maj
Table 6.1: Spatial case system in Lak (Zhirkov 1955)
To sum up, the Lak Approximative expression in (1c) is taken to
illustrate the fact that morphemes can correspond to unique terminals
in the syntactic structure (the morphemes -lu, -n, and -maj), as well as
to multiple terminals (as in the case of -vu).
98 LEXICALIZATION OF THE STRUCTURE 6.1
(5) ScaleP
Scale
-maj
GoalP
Goal
-n
PlaceP
Place AxPartP
AxPart KP
K
-lu
DP
k@at
-vu



Consequently, one can expect to find a language where Approximative
expressions are derived from the combination of five independent
morphemes, each of which corresponding to a unique terminal in the
structure in (5). Similarly, there should be languages where all the five
terminals in the structure are spelled out by a single morpheme. While
I have not encountered a language exemplifying the former scenario, the
latter case is found in the Mongolic language Ordos (data from Georg
2003).
(6) ger-lüü
house-towards
‘towards the house’
The Ordos marker -lüü is apparently mono-morphemic, which means that
it has to stand for all the five terminal nodes in the syntactic structure for
an Approximative path. Thus, morphologically, Ordos merges together
all five syntactic heads Scale, Goal, Place, AxPart and K, three of which
are expressed separately in Lak.
6.2 CAPTURING THE MISMATCH 99
(7) ScaleP
Scale GoalP
Goal PlaceP
Place AxPartP
AxPart KP
K DP
ger
-lüü



To conclude, the Path head universally decomposes into several terminals.
In some languages each terminal corresponds to one morpheme
in the path expression. In other languages, a single morpheme appears
in expressions whose syntactic structure contains multiple terminals. As
I mentioned in Chapter 3, this one-to-many relationship between the
number of morphemes and the number of syntactic terminals strongly
suggests a theory of Spell-out that allows for one morpheme to lexicalize
multiple terminals. I explore this topic in the remainder of this chapter,
starting with an investigation of two alternative views, which I then
reject.
6.2 Capturing the mismatch
The first alternative solution which aims to capture the mismatch between
the number of morphemes that constitute a given expression and
the number of terminals in its underlying syntactic structure is found in
the theory of Distributed Morphology (DM). The DM framework differs
from Nanosyntax in the some basic assumptions, for instance, in DM
lexical insertion is limited to terminal nodes. Further, Distributed Morphology
does not assume a one-to-one feature/terminal correspondence.
Nevertheless, in certain cases the number of terminals which need to be
lexicalized exceeds the number of lexical entries (called Vocabulary Items
in DM) which actually spell them out. DM deals with these cases by invoking
the operation of Fusion (Halle and Marantz 1993; 1994). Fusion
is an operation that applies after syntax and precedes Spell-out. What
it does is to take two sister nodes that have only grammatical features
100 LEXICALIZATION OF THE STRUCTURE 6.2
(and no phonological content), and fuse them into a single terminal node,
which inherits the features of the original nodes. A single morpheme can
then spell out this newly derived terminal node. The operation can be
repeated multiple times, hence a terminal resulting from the fusion of
two nodes can be itself fused with another node. There is no limit on
how many nodes can be fused into one terminal and we can imagine a
node like the one in (8), which has been created by the Fusion of K with
AxPart, followed by Fusion with Place, followed by Fusion with Goal,
and finally with Scale. The Ordos Approximative marker -lüü can then
be inserted at this terminal.
(8) [Scale, Goal, Place, AxPart, K] ↔ -lüü
The operation of Fusion, however, in some cases leads to the so-called
Fusion paradox, first identified in Chung (2007:fn. 22), and also discussed
in Caha (2009b) and Radkevich (2009). The paradox consists in the fact
that Fusion, on the one hand, precedes lexical insertion, but on the other
hand, it is triggered by the availability of an appropriate portmanteau
lexical item in the lexicon, which expresses the features of the fused nodes.
Finnish spatial morphology can be used to illustrate this paradox.
Finnish has two Goal cases: an Allative case, marked by the morphologically
complex ending -l-le, and an Illative case, marked by the portmanteau
suffix -h(V)n, see the examples below.1
(9) Finnish Allative and Illative cases (Karlsson 1999)
a. kato-l-le
roof-all
‘onto the roof’
b. kato-on
roof-ill
‘into the roof’
Assuming that the morpheme -l spells out the AxPart head, the lexicalization
of an Allative phrase has to involve the Fusion of the Place and
Goal heads into a new node, shown in (10a), which is then lexicalized by
-le, see (10b).
(10) a. Goal
Goal Place
⇒ [Goal, Place]
1
For a justification of this decomposition and a detailed analysis of the Finnish
spatial case system, see Section 7.6 in Chapter 7.
6.2 CAPTURING THE MISMATCH 101
b. [Goal, Place] ↔ -le
In the case of the portmanteau Illative morpheme, which stands for all
the three heads AxPart, Place and Goal, Fusion has to fuse the nodes
AxPart, Goal and Place together, as shown in (11a), so that the insertion
of -(h)Vn can take place under the thus derived terminal, see (11b).
(11) a. Goal
Goal Place
Place AxPart
⇒ [Goal, Place, AxPart]
b. [Goal, Place, AxPart] ↔ -h(V)n
Note that, while Fusion applies to all the three heads in (11), it does
not apply to the AxPart head in the case of -l-le. Hence, Fusion has to
somehow “know” that the lexicon contains an appropriate portmanteau
morpheme for the Illative phrase before it applies to the AxPart head
in the syntactic structure. The same kind of “knowledge” will prevent
Fusion from applying to AxPart in the Allative phrase. Given that Fusion
precedes lexical insertion, however, it is not expected to know in advance
what matching lexical items there are in the lexicon and this is what
constitutes the paradox.
An alternative to the Fusion solution of Distributed Morphology, still
keeping the idea that lexical insertion is limited to terminal nodes and
that each feature corresponds to a terminal, is to assume the availability
of silent morphemes à la Kayne (2004; 2008). Thus, we can assume
that the Ordos morpheme -lüü is inserted under one of the five syntactic
heads constituting an Approximative expression, say Scale, and all the
remaining four heads are spelled out by phonologically null morphemes.
102 LEXICALIZATION OF THE STRUCTURE 6.2
(12) ScaleP
Scale
-lüü
GoalP
Goal
∅
PlaceP
Place
∅
AxPartP
AxPart
∅
KP
K
∅
DP
ger
The disadvantage of this solution lies in the fact that the distribution
of the postulated null morphemes in (12) has to be somehow restricted
to the cases when Scale is lexicalized by -lüü, because they do not seem
to occur otherwise. For instance, a Locative phrase in Ordos is always
marked by -tU or -dU (Georg 2003), while one would expect null marking
to be available too.
Another example of the need to synchronize the co-occurrence of null
morphemes with a particular overt morpheme can be found in the Andic
language Bagvalal (Gudava 1967a). Like the vast majority of Daghestanian
languages, Bagvalal has the so-called locative series expressing
various spatial configurations. In Table 6.2, I present two of them, which
are relevant for the present discussion.
Series Location Goal Source
in -i -i -i -i-ss
on -a -a -a-r -a-ss
Table 6.2: The in and on series in Bagvalal (Gudava 1967a)
While a Goal expression in the Bagvalal on-series is marked by the
morpheme -r, no overt Goal morpheme is present in the in-series. Expressions
involving -i are therefore syncretic between Location and Goal
(see (13a) and compare to the unambiguous examples of the on-series in
(14), data from Gudava 1967a).
6.2 CAPTURING THE MISMATCH 103
(13) a. beq’-i
barn-loc/goal
‘in the barn’ or ‘into the barn’ (Location/Goal)
b. beq’-i-ss
barn-loc/goal-source
‘out of the barn’ (Source)
(14) a. am-a
roof-loc
‘on the roof’ (Location)
b. am-a-r
roof-loc-goal
‘onto the roof’ (Goal)
c. am-a-ss
roof-loc-source
‘off the roof’ (Source)
Given that the series markers -i and -a carry spatial concepts which
are commonly taken to be expressed by AxPart morphemes (Svenonius
2006), it is reasonable to assume that -i and -a spell out the AxPart head.
Consequently, the Place head must be lexicalized by a null morpheme.
Further, another null morpheme has to lexicalize the Goal head, when
AxPart is spelled out by -i, see (15).
(15) Goal
Goal
∅
PlaceP
Place
∅
AxPart
AxPart
-i
DP
When AxPart is spelled out by -a, however, the overt Goal morpheme -r
is inserted at the Goal head, see (16).
104 LEXICALIZATION OF THE STRUCTURE 6.2
(16) Goal
Goal
-r
PlaceP
Place
∅
AxPart
AxPart
-a
DP
This raises the issue of how to make certain that the overt Goal morpheme
-r in (16) does not attach to the -i marker of the in-series. Similarly, we
need to rule out the converse scenario where the null Goal morpheme
of (15) appears with the -a marker of the on-series. In other words, as
Starke (2011) puts it, “the ‘null morpheme’ approach needs to be supplemented
by a mechanism which ensures that the right (null) morphemes
occur with the right overt morpheme.”
The third conceivable solution to the mismatch between the number
of morphemes and the number of syntactic terminals present in the
structure is to adopt the view that one lexical item can lexicalize multiple
terminals in the syntactic tree. In more technical terms, we can assume
that lexical entries can be inserted not only at terminal nodes, but also
at phrasal nodes.
(17) Phrasal Spell-out:
Lexical insertion can target phrasal nodes.
Phrasal Spell-out is, in fact, a rather old idea originating from the generative
semantics theory, where it was proposed by McCawley (1968). More
recently, it has been adopted by Weerman and Evers-Vermeul (2002) to
account for many properties of Dutch and English pronouns. Similarly,
Neeleman and Szendrői (2007) show that Phrasal Spell-out explains the
cross-linguistic distribution of pro drop.
Allowing for morphemes to be inserted straight into phrasal nodes
obviates the need for a null Goal morpheme in the in-series, as well as
of the null Place morpheme. We can now simply assume that -i has the
features Goal, Place, and AxPart and can be inserted at the GoalP node,
as indicated in (18).
6.2 CAPTURING THE MISMATCH 105
(18) GoalP⇒-i
Goal PlaceP
Place AxPartP
AxPart DP
Suppose now that the morpheme -a does not have a Goal feature, but
just Place and AxPart. The Goal head will then be spelled out by a
separate morpheme which has the Goal feature. We can assume that -r
is such a morpheme.
(19) GoalP
Goal
-r
PlaceP⇒-a
Place AxPartP
AxPart DP
Apart from dispensing with the null Goal and Locative morphemes, the
two lexicalization diagrams presented above also provide a possible explanation
for why there is no Goal morpheme in the in-series, (18), but
there is a Goal morpheme -r in the on-series, (19). This is the result
of the assumed feature specification of the entries for -i and -a: the former
has the feature Goal, while the latter lacks it and therefore cannot
express a Goal phrase by itself.
Before concluding this section, I would like to draw the reader’s attention
to the fact that, in addition to Goal of motion, -i also expresses
Location, as shown in example (13a). Thus, the lexical item -i should
be able to lexicalize two distinct phrasal nodes: GoalP (in a Goal expression)
and PlaceP (when used in a Locative expression). Hence, we
need to build into the lexicalization procedure a mechanism to capture
syncretisms. The lexicalization model which thus emerges from the discussion
must have the following properties:
• First, it has to allow for a single morpheme to lexicalize multiple
terminals (Phrasal Spell-out)
• Second, it has to allow for a given morpheme to lexicalize more
than one syntactic structure.
106 LEXICALIZATION OF THE STRUCTURE 6.3
A syntactic framework which meets these requirements is Nanosyntax—
a model of grammar developed at the University of Tromsø (Starke
2005-2009, Ramchand 2008b, Bašić 2007, Fábregas 2007, Abels and Muriungi
2008, Muriungi 2008, Lundquist 2008, Caha 2009b, Taraldsen 2010,
Pantcheva 2010, for a representative collection of papers see Svenonius
et al. 2009). Apart from incorporating the properties listed under
the bullet points above, Nanosyntax also conforms with other assumptions
I made in Chapter 3, for instance the fine-grained decomposition
of syntactic structure, as well as the tight connection between syntax
and morphology. In the next section, I present the basic tenets of the
Nanosyntax theory of grammar, which I will eventually adopt.
6.3 Nanosyntax
The distinguishing feature of Nanosyntax is that syntactic terminals are
very “small,” hence the nano bit in the name of the framework. In fact,
each terminal corresponds to a unique feature. These features are ordered
in a universal hierarchy called the functional sequence. Nanosyntax thus
falls within the cartographic approach and follows its maxim “one morphosyntactic
property – one feature – one head,” discussed in Chapter
3.
6.3.1 The relationship between morphology and syn-
tax
Syntax builds structure by taking the atomic features and arranging them
by Merge into syntactic structures which comply with the hierarchical
order determined by the functional sequence. Under the Nanosyntax
view of grammar adopted here, morphemes are just the reflection of
how chunks of these syntactic structures are stored in the lexicon. For
instance, suppose that syntax combines the atomic features Place and
AxPart into the structure in (20):
(20) PlaceP
Place AxPartP
AxPart DP
This structure can be stored in the lexicon as a unit, which is paired
with a phonological exponent (and optionally also conceptual content, as
6.3 NANOSYNTAX 107
discussed under (25)). This unit, let us label it a, will then represent a
morpheme and its lexical entry will be of the following shape:
(21) a ⇔ </phonological content/, PlaceP
Place AxPartP
AxPart
, (conceptual content)>.
The morpheme a is a portmanteau morpheme which expresses location
as well as a spatial configuration. An example of such a morpheme is
the Bagvalal series marker -a presented in (14a) in the preceding section,
repeated in (22) (data from Gudava 1967c).
(22) a. am-a
roof-loc
‘on the roof’
The lexical entry of -a has the following shape:
(23) -a ⇔ </a/, PlaceP
Place AxPartP
AxPart
, on>
Other languages might not store the structure [PlaceP Place [AxPartP AxPart
]] as one unit in the lexicon, but store just [Place] and just [AxPart].
Each of the two structures will be paired with phonological content and
such languages will have two morphemes: one for Place and one for
AxPart. This scenario is found in Chamalal, presented in Table 4.1 in
Chapter 4, where there is a Locative suffix -i which attaches to the series
markers. I repeat in Table 6.3 the relevant columns from Table 4.1.
Taking as an example the series marker -n (behind), we can assume
that its lexical entry contains the information given in (24).
(24) -n ⇔ </n/, AxPart, behind>.
We can also assume that the lexical entry for the Locative suffix is the
one in (25).
(25) -i ⇔ </i/, Place>
The reason why the entry for -i in (25) has no conceptual content is
because it is not associated with the type of “encyclopedic” information
108 LEXICALIZATION OF THE STRUCTURE 6.3
Series Location
in -∅ -∅-i
at -x -x-i
inside -ń -ń-i
behind -n -n-i
under -kk -kk-i
on -l -l-i
vertical attachment -tS -tS-i
Table 6.3: Locative case in Chamalal (Magomedbekova 1967b)
that distinguishes a cat from a dog or top from bottom. The morpheme
-i carries only “formal semantic” information, like, for example, stative
spatial relation, which comes from the semantic contribution of the
head(s) in the syntactic structure stored in the entry.
The two morphemes -n and -i in Chamalal then combine to lexicalize
the Locative Plural structure, as shown in the fourth row of Table 6.3.
Thus, under the Nanosyntax view on grammar, morphemes are pieces
of syntactic structure stored in the lexicon and combined with phonological
(and conceptual) content. Thus, syntax directly determines the
“shape” of a morpheme, i.e., what features it has and how they are ordered
geometrically. As a result, only morphemes whose structures are
derivable by syntax can exist.
Syntax is also responsible for the way morphemes combine with each
other to build “morpheme complexes” like the Locative marker in the
Chamalal behind-series. More precisely, the reason we get the combination
DP-n-i and not *DP-i-n is because the AxPart projection is
closer to the noun than the Place projection in the syntactic structure in
(20) (see Svenonius 2006). The same effect can be observed in the Lak
Approximative morpheme complex -lu-vu-n-maj (recall the example in
(1c)), where the Kase morpheme -lu is closest to the noun, followed by
the series marker -vu, the Goal morpheme -n, and the Scale morpheme
-maj, thus mirroring the syntactic structure [Scale [Goal [Place [AxPart
[K DP ]]]]].
The fact that the morphological composition of an expression reflects
the syntactic structure underlying it has been stated as a principle by
Baker (1985) in the formulation below:
6.3 NANOSYNTAX 109
(26) The Mirror Principle (Baker 1985)
Morphological derivations must directly reflect syntactic derivations
(and vice versa).
According to Baker, the parallel between syntactic and morphological
derivations strongly suggests a theoretical framework where both morphological
and syntactic processes take place in a single module of the
grammar. In the same vein, Lieber (1992) argues for reducing morphology
to syntax thus eliminating the morphological component assumed
by some scholars (e.g., Ackema and Neeleman 2007), where processes like
word formation take place. Instead, Lieber attempts to derive the properties
of complex words from principles independently needed in syntax.
The nanosyntactic framework adopted here follows this line of research
and advances towards a unification of syntax and morphology, since, under
this view, the shape of morphemes and the way they combine into
words and then phrases are directly determined by syntax.
6.3.2 Spell-out of syntactic structure
Recall from the preceding subsection that the lexicon is simply a list of
entries where fragments of syntactic trees are combined with a phonological
representation and a conceptual content. The Spell-out procedure
can then be defined as a replacement of a piece of the syntactic tree by
a lexical entry from the lexicon, thus supplying the syntactic structure
with the phonological and conceptual content of the entry. In choosing
the appropriate lexical entry, Spell-out is thus concerned with whether it
has a matching syntactic specification, i.e., whether the syntactic structure
stored in the lexical entry matches the syntactic structure the entry
replaces. Matching can be defined as identity for the purposes of this section,
but I will be refining the definition in the course of the discussion,
the final version being presented in (13), Section 7.3, Chapter 7.
(27) A lexical entry matches a node if the syntactic tree in its specification
is identical to the node.
Apart from matching the syntactic structure of the entry to the syntactic
structure to be replaced, the Spell-out procedure also chooses from various
allomorphs, as proposed by Bye and Svenonius (2011). For instance,
when spelling out the D head in English indefinites, Spell-out chooses
between a and an depending on whether the noun it dominates begins
with a consonant (a pear) or a vowel (an apple). For this reason, Bye and
Svenonius (2011) suggest that Spell-out consists of two components: a
110 LEXICALIZATION OF THE STRUCTURE 6.3
syntactic component, which they call L-Match, and a phonological component
called Insert. L-Match precedes Insert and is concerned with
choosing the lexical item(s) that can replace a given syntactic structure,
without being sensitive to phonological information. Insert operates on
the output of L-Match and picks out the allomorphs depending on the
phonological environment. Since I deal only with the syntax of path expressions
in this thesis, I will leave aside the phonological aspect, thus
narrowing the discussion of lexicalization only to the matching component
of Spell-out. In other words, I will be concerned only with that
aspect of lexicalization which involves searching and finding of matching
lexical entries.
Given that the syntactic tree stored in a lexical entry can involve
multiple terminals arranged in a tree structure, a lexical entry can be
used to replace, or spell out, a syntactic structure that contains multiple
terminals. A natural way to implement this idea technically is to
assume that lexical entries can be inserted at phrasal nodes. A lexical
items whose lexically stored tree contains a phrasal node XP can be thus
inserted into an XP node in the syntactic structure (Starke 2009, Caha
2009a;b, Fábregas 2009, among others).2
The adoption of Phrasal Spell-out eliminates the asymmetry between
terminal and phrasal nodes. Both types of nodes are subject to lexical insertion.
When it comes to the succession in which the nodes are targeted
by Spell-out, I take lexicalization to proceed bottom-up, as also assumed
in other frameworks — see, for instance, Bobalijk’s (2000) argument
for root-outwards vocabulary insertion within the theory of Distributed
Morphology. With these two assumptions concerning lexicalization —
Phrasal Spell-out and bottom-up Spell-out — lexical insertion can target
the syntactic nodes in two possible successions, presented in (28) and
(29). (The abstract labels of the nodes reflect the order in which they
are targeted by lexical insertion and I will not specify at this point the
2
Alternative ways to lexicalize multiple terminals have been proposed. One is
based on the idea of morphemes “spanning” a stretch of the syntactic tree, originally
proposed by Williams (2003) and also adopted in some nanosyntactic works, e.g.,
Abels and Muriungi (2008), Taraldsen (2010), Bye and Svenonius (2011). Another
way is by using some form of movement, or Re-merge as in Borer (2005a;b) and also
Ramchand (2008b) who implements it in a nanosyntactic analysis of the verbal phrase.
Under the “spanning” and “multi-attachment” views, lexical items are specified for the
dominance relations between the features only and carry no information about the
particular geometric configuration of these features with respect to each other. In
this thesis, I adopt the mechanism of phrasal Spell-out as it most compatible with the
idea that lexical entries can contain whole syntactic phrases where complex structural
relations obtain.
6.3 NANOSYNTAX 111
real labels reflecting headedness.)
(28) 5
4 3
2 1
(29) 5
3 4
1 2
Technically, nodes which are at the same level could be lexicalized in two
possible orders: left-to-right as in (28), or right-to-left, as in (29). Both
orders are compatible with a bottom-up Spell-out, since in both (28) and
(29) the (higher) mother nodes are lexicalized after their (lower) daughter
nodes. I will nevertheless assume the right-to-left order in (28), because
it is the one compatible with my following two assumptions about Spellout,
to which I turn below. I will then come back to the lexicalization
succession in (28) and discuss how it follows from these assumptions.
The next assumption concerning the lexicalization of syntactic structure
is that it proceeds in a cyclic fashion such that each step of external
Merge defines a cycle (Caha 2010b). In other words, the lexicon is accessed
every time a new feature is added to the derivation.
(30) There is a round of lexical access after each External Merge op-
eration.
Let us now examine the lexicalization of an abstract syntactic structure,
where I use the letters A, B, and C to stand both for features and
for the nodes in the tree.
The first operation in the derivation is the External Merge of feature A
with feature B to create node BP. This is the first cycle in the derivation.3
(31) BP
B A
According to (30), this is followed by a round of lexicalization, where the
lexicon is inspected for lexical entries which match the nodes created in
this cycle, more precisely nodes A, B and BP. The next operation is the
Merge of feature C to the derivation.
3
It is in principle possible that A constitutes a cycle by itself. But if we take a
cycle to be defined by External Merge, which joins two syntactic objects to create
a new one (Chomsky 2001:3, Adger 2003:54), the first cycle in the derivation will
include A, B and the new syntactic object BP.
112 LEXICALIZATION OF THE STRUCTURE 6.3
(32) CP
C BP
B A
Since this is an External Merge, a new round of Spell-out is triggered.
The structure in (32) is thus shipped to the lexicon for the search for
items which can lexicalize it. At this point the question arises whether
the nodes A, B and BP, which were spelled out in the previous cycle,
are inspected for lexicalization again. A “fresh Spell-out” strategy would
entail that the lexicalization procedure starts from the very bottom and
nodes A, B and BP are inspected for lexicalization again (and again and
again; in fact, A, B and BP will be targeted by lexicalization as many
times as External Merge applies). The alternative is that the lexicalization
keeps track of the previous cycles of Spell-out and targets only
the nodes in the new cycle. The answer has been suggested by Starke
(2011) who investigates this question and argues on the basis of idioms
that Spell-out must “remember” the outcome of previous Spell-out oper-
ations.
Adopting Starke’s proposal, I suggest that the lexicalization operation
inspects only the nodes in the second cycle — C and CP. Importantly,
the Spell-out of node BP, which is to the right of node C, happened
in the previous cycle. In other words, BP is targeted by lexicalization
before C because BP is in an earlier cycle. This suggests a right-to-left
lexicalization order of the nodes BP and C. Similarly, node CP will be
lexicalized before the node to its left which will be created when a new
feature, say D, is merged in the next cycle.
Note also that the bottom-up Spell-out, too, follows to a certain extent
from the assumption about cycles—since each Merge of a new feature
triggers lexicalization, lexicalization tracks the growth of the syntactic
tree from its bottom to its top.
The only nodes that remain not ordered are the lowest A and B
nodes, where a left-to-right lexicalization does not clash with the assumption
about cycles, because there is no preceding cycle (see fn. 3).
Given that lexicalization goes right-to-left at the higher levels, I will assume
that the same order holds of the two lowest nodes for the sake
of uniformity. Similarly, within a cycle itself, Spell-out could proceed
top-down or bottom-up. For the sake of uniformity again, I will assume
that lexicalization is bottom-up also within a cycle. In sum, I opt for a
consistently bottom-up right-to-left Spell-out order in (28) both between
6.3 NANOSYNTAX 113
cycles (as a consequence of my assumptions about cyclic Spell-out) and
within a single cycle (for the purpose of uniformity).
My last assumption about the basics of the Spell-out procedure concerns
non-lexicalized features. I assume that each feature in a given cycle
has to be lexicalized before the derivation can proceed to the next cycle.
If there is a feature left unexpressed, the derivation crashes. This requirement
is based on a principle called Exhaustive lexicalization which has
been proposed by Ramchand (2008a) and Fábregas (2007). The content
of this principle is that syntactic structures that contain non-lexicalized
features are ill-formed. Since I assume that this principle should hold at
the end of each cycle, I adopt it in the following formulation:
(33) Cyclic Exhaustive Lexicalization:
Every syntactic feature must be lexicalized at the end of a cycle.
Let us now go back to the structure in (32) and examine how it can
be spelled out. Assuming that the nodes BP and CP are the result of
external Merge, once the derivation reaches these nodes, the syntactic
structure built up to that point will be shipped to the lexicon to be
spelled out.
As already mentioned in the beginning of the section, Spell-out involves
finding a matching lexical entry for the syntactic structure to be
lexicalized and inserting it at the relevant node. To illustrate, suppose
that we have the following lexical entries in the lexicon (from now on I
will use italic small letters a, b, c, etc. to represent lexical entries, and
capital letters A, B, C, etc. to stand for features and/or nodes in the
syntactic tree):
(34) a. a ⇔ </a/, A>
b. b ⇔ </b/, B>
c. c ⇔ </c/, BP
B A
>
d. d ⇔ </d/, C>
e. e ⇔ </e/, CP
C BP
B A
>
114 LEXICALIZATION OF THE STRUCTURE 6.3
The very first stage of the derivation is to Merge A with B thus deriving
node BP. The structure immediately becomes target of lexicalization,
since every external Merge operation triggers a round of lexical access.
The lexicalization procedure, starting from the rightmost bottom element,
according to the assumption, attempts lexical insertion at node
A. There is a matching lexical entry a and it gets inserted, see (35a).
The next target is node B where the matching item b becomes inserted,
(35b). Finally, lexicalization turns to node BP, which is spelled out by
the matching item c, (35c).
(35) First cycle
a. Step 1
BP
B A⇒a
b. Step 2
BP
b⇐B A⇒a
c. Step 3
BP⇒c
B A
Note that, at the second stage of derivation, (35b), the syntactic structure
is pronounced as /b/+/a/. At the third stage (35c), however, the same
syntactic structure is pronounced as /c/. This is so because once we
insert c in node BP, it overrides all previously inserted lexical material at
the lower nodes. As a consequence, the a and b morphemes “disappear.”4
Once the lexicalization of node BP has been accomplished, the derivation
can continue by an external Merge of C to derive the phrasal node
CP. This results in the structure in (28), repeated below.
(36) CP
C BP
B A
At this stage, another round of lexical insertion takes place. Consequently,
the lexical item in (34d) is inserted under node C, which is
followed by an insertion of the entry in (34e) straight into the phrasal
node CP. This is depicted in (37a) and (37b), respectively.
4
To be more precise, I assume that insertion at the phrasal node overrides the
previously inserted phonological material, but has to preserve the conceptual content
of the items inserted in the daughter nodes. In other words, overriding can happen
only when the conceptual contents of the overriding item and the overridden items
match.
6.3 NANOSYNTAX 115
(37) Second cycle
a. Step 4
CP
d⇐C BP⇒c
B A
b. Step 5
CP⇒e
C BP
B A
As already mentioned, I assume that the lexicalization “remembers”
the outcomes of the previous cycles and hence does not need to spell out
nodes A, B and BP in (37) again.
The insertion of the entry e at node CP at the stage depicted in (37b)
overrides the items d and c and the structure is pronounced as /e/ instead
of as /d/+/c/.
An interesting question is what would happen if the lexicon did not
have all the five lexical entries in (34), but just some of them. Consider
first the combination of morphemes in (38), where the lexical entries that
spell out just terminal nodes are left out.
(38) a. c ⇔ </c/, BP
B A
>
b. e ⇔ </e/, CP
C BP
B A
>
As things stand now, no lexical insertion will take place at steps 1, 2 and
4 because I assumed that matching requires identity and no lexical entry
in (38) contains a tree structure that is identical to the targeted node.
The structure can be nevertheless lexicalized at both the first and the
second cycle by inserting c at node BP and e at node CP, as shown in
(39).
(39) a. First cycle
BP⇒c
B A
b. Second cycle
CP⇒e
C BP
B A
116 LEXICALIZATION OF THE STRUCTURE 6.3
The lexicalization is successful because the entry c contains all the
features of the first cycle and the entry e contains all the features of the
second cycle. As a result, Cyclic Exhaustive Lexicalization will be obeyed
since, at each cycle, all the features expressed in the syntactic tree are
lexicalized. The reason for this is that the syntactic features A, B, and
C are all included in the lexical entries.
Let us now consider a case in which the lexicon does not have the
“phrasal” entries c and e, but just the “terminal” entries a, b, and d.
(40) a. a ⇔ </a/, A>
b. b ⇔ </b/, B>
c. d ⇔ </d/, C>
The way the lexicalization will proceed is to insert a at node A and b
at node B in the first cycle, see (41a-b). In the second cycle, d will be
inserted at node C, as shown in (41c).
(41) First cycle
a. Step 1
BP
B A⇒a
b. Step 2
BP
b⇐B A⇒a
c. Second cycle
CP
d⇐C BP
b⇐B A⇒a
In (41), Cyclic Exhaustive Lexicalization is satisfied again, although
the nodes BP and CP are not specifically listed in the lexical entries. This
is so because the nodes BP and CP do not represent syntactic features
distinct from A, B, and C.
The lexicalization scenarios in (39) and (41) depict two different ways
to spell out a given node in compliance with Cyclic Exhaustive Lexicalization.
For example, at node BP, Cyclic Exhaustive Lexicalization can
be satisfied by inserting lexical entries at each terminal (see (41b)) or by
inserting a lexical entry straight into a phrasal node (see (39a)).5
We can then state the condition for when a node counts as successfully
spelled out (that is, in accordance with Cyclic Exhaustive Lexicalization)
5
Naturally, a combination of the two strategies is also viable. Thus, the structure
in (32) can also be spelled out as /d/+/c/ in a language which lacks the “phrasal”
entry e, but has the “phrasal” entry c and the “terminal” entry d.
6.3 NANOSYNTAX 117
as in (42).
(42) A node X is successfully lexicalized if
a. a lexical entry is inserted at X, or
b. the daughters of X are lexicalized.
A node X can thus be successfully lexicalized directly in the case of (42a),
or by inheritance, in the case of (42b). The condition in (42) has one
important implication for the case in (35), repeated below as (43).
(43) First cycle
a. Step 1
BP
B A⇒a
b. Step 2
BP
b⇐B A⇒a
c. Step 3
BP⇒c
B A
At the stage depicted in (43b), the lexicalization procedure has inserted
a under node A and b under node B. Then, lexicalization turns
to the next node BP, which does count as successfully lexicalized, since
all the features contained in that node are lexicalized. Nevertheless, the
lexicalization procedure searches the lexicon for a matching lexical entry
c and, in case there is one, inserts it straight into the phrasal node BP,
as shown in (43c). The question is: Why is it necessary to perform a
search for an item able to spell out node BP in one go, given that all the
features in the cycle are expressed (i.e., Cyclic Exhaustive Lexicalization
is satisfied). The derivation can continue without inserting c into node
BP, as it does in languages which lexicalize the structure as a+b for lack
of an item c.
This question uncovers an important aspect of the Spell-out model I
assume, which I consider necessary to emphasize: lexicalization is attempted
at every node. In other words, the lexicalization procedure
“blindly” proceeds up the syntactic tree trying to match lexical material
to each node, and does not care whether the targeted node is already
lexicalized by inheritance. Consequently, if lexical insertion can happen
at a phrasal node, then it will. The effect of this can be seen in languages
which, apart from a and b, have the item c. Then c will be inserted at
(i) CP
d⇐C BP⇒c
B A
118 LEXICALIZATION OF THE STRUCTURE 6.4
node BP, even though it is already lexicalized by inheritance. Famously,
went will be used instead of *go-ed and worse will be used instead of *bader.
This phenomenon is often referred to in the literature as blocking (for
instance, Andrews’ 1990 Morphological Blocking Principle, Posers’ 1992
Phrasal blocking, and Kiparsky’s 2005 blocking mechanism). Blocking in
general alludes to cases where the existence of a portmanteau morpheme
blocks the occurrence of a sequence of morphemes. The Phrasal Spell-out
system derives many instances of blocking effects, as lexical items which
lexicalize a phrasal node are inserted later than the “smaller” lexical items
which lexicalize terminals and thus override them. This has been formulated
as the Biggest wins theorem in Starke (2009), Caha (2009b) and
Taraldsen (2010). Muriungi (2009) labels it the Union Mechanism with
different formulation but to the same effect.
6.4 The Superset Principle
In the preceding section, I defined Spell-out as the insertion into the
syntactic structure of a matching lexical item, that is, a lexical item
whose entry contains an identical syntactic structure. The phenomenon
of syncretism, however, suggests that the requirement of identity between
syntactic structure and the lexically stored tree is too restrictive.
Consider, for instance, the following Hindi data (fieldwork notes collected
by Peter Svenonius and Minjeong Son).
(44) baccaa
child
kaar-ke
car-gen
saamne-see
front-abl
calaa.
walk.perf
(i) ‘The child walked via in front of the car.’
(ii) ‘The child walked from in front of the car.’
The example above is ambiguous between a Route path reading and a
Source path reading. In Chapter 4, I argued that Route paths and Source
paths have different underlying syntactic representations, shown in (45).
6.4 THE SUPERSET PRINCIPLE 119
(45) a. Route path
RouteP
Route SourceP
Source GoalP
Goal PlaceP
Place ...
b. Source path
SourceP
Source GoalP
Goal PlaceP
Place ...
What this means for the Hindi Ablative marker -see in (44) is that
it is used to lexicalize two different structures: (45a) when it gives rise
to a Route reading, and (45b) when it has a Source path interpretation.
Importantly, the two structures in question are in a superset-subset relationship
– the syntactic features which make up a Source expression are
a proper subset of the syntactic feature in a Route expression. Similar
cases of syncretisms have been successfully modeled in the framework of
Distributive Morphology by assuming underspecification of lexical items,
formulated as the Subset Principle.
(46) The Subset Principle (Halle 1997)
The phonological exponent of a vocabulary item is inserted into
a morpheme in the terminal string if the item matches all or
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.6
Put in simple words, the Subset Principle says that a lexical item lexicalizes
a terminal in the syntactic tree if the feature specification of the item
is a subset of the features expressed in the terminal. This presupposes
that one terminal contains more than one feature, contrary to the assumption.
Further, lexical insertion is restricted only to terminals, while
I have adopted the idea that insertion can also target phrasal nodes.
One would therefore need to reformulate the Subset Principle so that the
domain for lexical insertion is augmented to include phrasal nodes. Nevertheless,
one big conflict remains: the Subset Principle allows features
in the syntactic tree to remain unexpressed. This goes against the Cyclic
Exhaustive Lexicalization principle presented in (33), which requires that
6
The term morpheme in DM is used to mean “terminal node.”
120 LEXICALIZATION OF THE STRUCTURE 6.4
each syntactic feature in the tree should be expressed (at the end of a cycle).
Thus, if we want to capture the Hindi data following the same idea
that there is a subset-superset relations between syncretic morphemes
(and structures), while at the same time keeping the assumptions about
Phrasal Spell-out, it is necessary to change the idea behind the Subset
Principle so that the lexical items are overspecified in order for all syntactic
features in the tree to be realized. The result is called the Superset
Principle and it is formulated as a condition on matching.
(47) The Superset Principle (Starke 2009, Caha 2009b):
A vocabulary item matches a node if its lexical entry is specified
for a constituent containing that node.
Thus, matching now requires that the syntactic tree stored in a lexical
item is identical to or bigger than the node it is inserted into. Given that
the structure of Route paths is bigger than the structure of Source paths,
as argued in Chapter 4, the tree stored in the lexical entry of the Hindi
Ablative -see should be specified for the bigger Route structure. This
will make it possible that it is also used to lexicalize the smaller Source
structure.
(48) Lexical entry for Hindi Ablative -see (under the Superset Prin-
ciple):
-see ⇔ </see/, RouteP
Route SourceP
Source GoalP
Goal PlaceP
Place
>
With this specification, -see can spell out both structures in (45), as
shown below.
6.4 THE SUPERSET PRINCIPLE 121
(49) a. Route path
RouteP ⇒ -see
Route SourceP
Source GoalP
Goal PlaceP
Place ...
b. Source path
SourceP ⇒ -see
Source GoalP
Goal PlaceP
Place ...
In the case of a Route phrase, (49a), -see is inserted into a node
which is identical to the syntactic structure of the lexical entry (48) (ignoring
the complement of Place for now). In the case of a Source path,
-see spells out a sub-constituent of the syntactic tree it is specified for,
thus leaving the Route feature “unused.” In other words, the syncretism
of Route=Source syncretism in Hindi comes out because the Superset
Principle allows a given lexical entry to lexicalize syntactic structures
that are identical to or smaller than the structure it is specified for, crucially
restricting unexpressed features to the lexical entry, and expressing
all the features in the syntactic tree.
Before concluding this section, let me go back to a case of syncretism
I presented in the beginning of this chapter and show how the current
system deals with it. Recall from Section 6.2, that the series markers -i
‘in’ and -a ‘on’ in the Daghestanian language Bagvalal express both a
spatial configuration and Location. In addition, -i expresses also Goal of
motion (see the data below, repeated from (13a) and (14a)).
(50) beq’-i
barn-loc/goal
‘in the barn’ or ‘into the barn’
(51) am-a
roof-loc
‘on the roof’
The ability of -i and -a to express a spatial configuration and Location
by themselves suggests that they are specified for both the AxPart and
Place features. In addition, the entry for -i must have a Goal feature, as
it is used also in Goal expression. The lexical entries I therefore suggest
are presented in (52) and (53).
122 LEXICALIZATION OF THE STRUCTURE 6.4
(52) a. -a ⇔ </a/, PlaceP
Place AxPartP
AxPart
, on >
b. -i ⇔ </i/, GoalP
Goal PlaceP
Place AxPartP
AxPart
, in >
Thus, when expressing a Locative structure, -a makes use of its full lexical
specification and lexicalizes a node that is identical to the tree fragment
stored in the entry.
(53) PlaceP⇒-a
Place AxPartP
AxPart ...
The entry for -i is “bigger” than the entry for -a—it contains a Goal head
in addition, and therefore -i can lexicalize a Goal structure.
(54) GoalP⇒-i
Goal PlaceP
Place AxPartP
AxPart ...
By the Superset Principle, -i can also lexicalize a PlaceP node, leaving
its Goal feature “unused.”
(55) PlaceP⇒-i
Place AxPartP
AxPart ...
6.5 THE SUPERSET PRINCIPLE 123
Note that, due to the specification in (52a), the series marker -a cannot be
used to Spell-out a Goal structure—it does not contain a Goal feature and
by the Superset Principle it cannot be inserted at GoalP. The Bagvalal
case syncretism thus provides a strong argument for the adoption of the
Superset Principle (and Exhaustive Lexicalization, which goes together
with it). Consider what will happen if we import the Subset Principle
into the Phrasal Spell-out model (and consequently restate it to hold of
Phrasal Spell-out plus drop the requirement that each feature is to be
lexicalized). The entries for -a and -i that we then need to assume have
to be underspecified, see (56) and (57).
(56) a. -a ⇔ </a/, PlaceP
Place AxPartP
AxPart
, on >
b. -i ⇔ </i/, PlaceP
Place AxPartP
AxPart
, in >
Both entries can be inserted at PlaceP, spelling out a Locative phrase.
The series marker -i can also be inserted at GoalP, because it is specified
for a subset of the features contained in the tree. The crucial question
is: how do we prevent -a from being inserted at GoalP? Since the entry
for -a contains the same subset of features like -i, it should be equally
eligible for insertion at GoalP.7
As far as I can see, the only way to capture the data without assuming
Superset, is to posit a silent Goal morpheme and restrict it contextually
to -i ‘in’ – an option I explored and rejected in Section 6.2, since it
necessitates the development of a mechanisms which ensures that the
right null morphemes occur with the right overt morphemes.
7
It should be highlighted that the impossibility of applying the Subset Principle
to the Bagvalal case discussed here is because the Subset Principle is not compatible
with a Phrasal Spell-out model. In the framework of DM, the Bagvalal Location/Goal
syncretism in the in-series and the lack of it in the on-series can be captured by
assuming that to and in undergo Fusion, but to and on do not. This, in turn, leads
to the same Fusion paradox I presented in Section 6.2.
124 LEXICALIZATION OF THE STRUCTURE 6.5
6.5 The Elsewhere condition
In the preceding section, I showed how the Superset Principle can capture
syncretisms by allowing a given morpheme to lexicalize a structure which
is identical to or contained in the lexically stored tree of the morpheme’s
lexical entry. This relaxation of the matching condition leads to the
occurrence of situations where more than one lexical item is eligible to
spell out a given syntactic node. Consider, for instance, the case of the
Avar Source marker -ssa, which I discussed in Chapter 4. To refresh
the memory of the reader, I repeat the spatial markers of Avar for the
on-series below (for a full overview of all Avar locative series, see Table
4.4).
Series Location Goal Source Route
on (top of) -da -d-e -da-ssa -da-ssa-n
Table 6.4: The on-series in Avar (Blake 1994)
As suggested in Chapter 4, Section 4.3 (see (17) in the relevant chapter),
the Source marker -ssa lexicalizes both the Goal and the Source
heads. The lexical entry of -ssa is therefore as in (57).
(57) Lexical entry for Avar Ablative -ssa:
-ssa ⇔ </ssa/, SourceP
Source GoalP
Goal
>
According to the Superset Principle in (47), -ssa can spell out a Source,
as well as a Goal structure, when combined with a Locative marker lexicalizing
Place. The reason is that both structures are contained in the
tree stored in the lexical entry for -ssa. This does not happen, though.
Goal expressions are lexicalized by the morphemes -da+-e (pronounced
as /de/), as shown below.
(58) wácc-ass-de
wácc-ass-da-e
brother-erg-on-to
‘onto the brother’ (data from Uslar 1889)
The question is why Avar prefers to use the Allative suffix -e to lexicalize
6.5 THE ELSEWHERE CONDITION 125
the Goal structure and not the Ablative suffix -ssa.
Let us first see what the lexical entry for the Allative -e could be. It
attaches to a Locative marked DP (just like the Ablative -ssa), hence, it
does not spell out the Place head. It does not have a Source semantics,
hence it is not specified for the Source feature either. The result is a
lexical entry of the shape in (59), which spells out just the Goal head.
(59) Lexical entry for Avar Alative -e: -e ⇔ </e/, Goal>
The two possible ways to lexicalize a Goal structure are then shown in
the tree diagrams in (60).
(60) a. GoalP
-e ⇐ Goal Place
-da ⇐ Place ...
b. GoalP
-ssa ⇐ Goal Place
-da ⇐ Place ...
In the first case, the Allative marker is inserted under the Goal head,
(60a). In the second case, the Goal head is lexicalized by the Ablative
marker, (60b), an option made available by the Superset Principle. The
preferred lexicalization is the one in (60a), as shown by the data in (58).
Note that the Allative entry in (59) is a perfect match for the Goal
head in (60). The Ablative entry -ssa in (57) is, on the contrary, “bigger”
than the structure it lexicalizes in (60b) as a result of which the Source
feature on -ssa remains unused. The observation is, therefore, that the
Spell-out procedure chooses to insert the lexical entry that leaves the
least number of “unused” features.
This conclusion can be formalized in the form of the following rule:
(61) When two lexical entries meet the conditions for insertion in a
given node, the item with the fewest features not contained in
the node gets inserted.
The same idea has been more concisely formulated in Taraldsen (2010) as
the Minimize Underattachment constraint, the term “underattachment”
referring to the cases when a lexical entry fails to lexicalize the full-size
structure it is specified for.
As a matter of fact, the generalization in (61) follows from a more
general principle, namely, the Elsewhere Condition of Kiparsky (1973),
shown below (Starke 2009, Caha 2009b, Pantcheva 2010).
126 LEXICALIZATION OF THE STRUCTURE 6.6
(62) Elsewhere Condition (Kiparsky 1973:94)
Two adjacent rules of the form
A −→ B / P___Q
C −→ D / R___S
are disjunctively ordered if and only if:
a. the set of strings that fit PAQ is a subset of the set of strings
that fit RCS, and
b. the structural changes of the two rules are either identical
or incompatible.
Put informally, the Elsewhere Condition says that whenever we have two
rules—one which applies in a more general case, and the other which
applies in a more specific case—the specific rule blocks the general rule
from application.
An alternative way to formulate the same condition is provided by
Neeleman and Szendrői (2007).
(63) Elsewhere Principle (Neeleman and Szendrői 2007)
Let R1 and R2 be competing rules that have D1 and D2 as their
respective domains of application. If D1 is a proper subset of D2,
then R1 blocks the application of R2 in D1.
The generalization in (61) is simply an implication of the Elsewhere Condition.
Consider again the Avar Source and Goal markers. The Ablative
-ssa can spell out a Source structure and a Goal structure. The Allative
-e can spell out just a Goal structure. The domain of application of the
Allative marker is hence a proper subset of the domain of application of
the Ablative marker. Therefore, by virtue of (63), the Allative marker
wins the competition for insertion in a Goal structure.
To summarize, in this subsection, I presented a case where two lexical
items compete to lexicalize a given structure and demonstrated how the
competition is regulated by the Elsewhere Condition of Kiparsky (1973),
which enforces the choice of the more highly specified lexical item.
6.6 Conclusion
In this chapter, I presented a model of Spell out which allows one morpheme
to spell out one or more terminals in the syntactic structure. I
presented the main principles and assumptions governing the lexicalization,
which I repeat below.
6.6 CONCLUSION 127
(64) The shape of the lexical entries is <phonology, syntactic tree,
conceptual content>
(65) Cyclic Spell-out:
Each step of External Merge is followed by lexical access. There
is no lexicalization cycle after Internal Merge.
(66) Cyclic Exhaustive Lexicalization:
Every syntactic feature must be lexicalized at the end of a cycle.
(67) Bottom-up Phrasal Spell-out:
Lexicalization proceeds right-to-left bottom-up starting from the
lowest node in the current cycle. Lexicalization targets both
terminal and non-terminal nodes.
(68) The Superset Principle (Starke 2009, Caha 2009b):
A vocabulary item matches a node if its lexical entry is specified
for a constituent containing that node.
(69) The Elsewhere Condition (reformulated from Kiparsky 1973)
When two lexical entries meet the conditions for insertion in a
given node, the item with the fewest features not contained in
the node gets inserted.
In the model, adopted here, lexical entries are pieces of syntactic structures
which are combined with a phonological representation and a conceptual
content, as stated in (64). All but the syntactic component of
an entry can be missing. Lexical items which are not associated with
encyclopedic information (e.g., functional morphemes) lack conceptual
content. Phonologically null morphemes lack phonological representa-
tion.
According to (65), the lexicon is accessed after each External Merge
operation but not after Internal Merge. Thus every External Merge
defines a lexicalization cycle. As a consequence, the nodes derived by
movement become target to lexical insertion only after External Merge
has applied. For a successful lexicalization it is necessary that every feature
(i.e., terminal) in the syntactic tree be lexicalized at the end of the
cycle, as stated in (66). If this is not the case, the derivation crashes.
The nodes in the syntactic structure are targeted by lexicalization
in a right-to-left bottom-up succession and lexicalization is attempted
at each node, alternating two terminals with one non-terminal node.
The lexicalization operation consists in matching and inserting lexical
entries at the nodes of the syntactic tree thus replacing the node with
the phonological and conceptual content of the entry, under the condition
128 LEXICALIZATION OF THE STRUCTURE 6.6
that the syntactic structure stored in the entry contains the syntactic
structure of the targeted node. When there are more than one matching
item, the one which has the smallest number of superfluous features is
chosen.
Many of the assumptions made here are shared with other frameworks,
perhaps most prominently Distributed Morphology (Halle and
Marantz 1993; 1994). For instance, in both frameworks, syntax operates
on abstract features to combine them into syntactic trees. Unlike DM,
however, in Nanosyntax these features are not pre-packaged into bundles.
Consequently, each syntactic terminal represents a unique feature.
A further similarity between the two frameworks lies in the information
stored in the lexical items. Both frameworks assume that lexical
items are stored in the lexicon as a pairing of a phonemic string (phonological
exponent) and a set of formal features (see, for example, Halle
1997 and Harley and Noyer 1999). The Nanosyntactic lexicon, nevertheless,
differs from the lexicon assumed in DM in two ways. First, the
nanosyntactic lexical entries can also contain a conceptual component
alongside the formal features. Second, in Nanosyntax, the grammatical
features are organized in a binary branching tree structure, while in DM
features are grouped into bundles. When it comes to the first point, the
two frameworks start to converge, since some scholars working in the DM
tradition now assume that root vocabulary items are specified for both
content and grammatical features (e.g., Siddiqi 2006). The second point,
however, still holds.
Other concepts shared by both Nanosyntax and DM are Late insertion
and Cyclicity of insertion, i.e., the idea that lexical items are inserted
into the syntactic structure once its derivation has been completed and
that this happens in cycles. Thus, both frameworks assume that lexicalization
of the structure is a post-syntactic operation. Still, the view on
lexicalization adopted here diverges from DM’s view. Following one of
the main tenets of Nanosyntax, I assume that lexical insertion can target
phrasal nodes. The domain for lexical insertion assumed in DM is, on
the contrary, limited to terminal nodes only, thus necessitating a special
morphological operation called Fusion, whose role is to “fuse” two or more
features into one terminal, whenever we have a given lexical entry spell
out what seems like two (or more) terminals. This in turn leads to the
Fusion paradox pointed out by Chung (2007) and illustrated in Section
6.2 of this chapter. Put briefly, the Fusion paradox is caused by the fact
that the morphological operation Fusion applies only in the presence of
a lexical entry which can spell out the fused nodes, yet the lexicon is not
6.6 CONCLUSION 129
accessed before all syntactic and morphological operations (like Fusion)
are performed.
Finally, perhaps the most conspicuous difference between the two
frameworks concerns the tolerance towards unexpressed features. Working
with underspecification of lexical entries, DM assumes that it is acceptable
to leave features unexpressed in the syntactic structure, while
all features of the lexical entries have to be expressed. Nanosyntax assumes
the exact opposite: every feature in the syntactic tree has to be
expressed (formulated as the Cyclic Exhaustive lexicalization) and the
features which are allowed to remain unexpressed are the ones of the lexical
entry. As a result, the lexical entries are overspecified in Nanosyntax.
This chapter focusses mostly on the system of assumptions ruling the
lexicalization of structure and contains little real language data. In the
following chapter, by contrast, I present a detailed investigation of the
spatial morphology in several languages (Karata, Uzbek, and Finnish)
and show the importance of each of the assumptions made. In addition,
I develop a novel idea originally proposed by Starke (2011) that
the particular shape of the lexical entries can give rise to a syntactic
movement in order to create the right configuration for Spell-out. I show
how this Spell-out triggered movement delivers the suffixal ordering or
spatial markers in the languages examined and successfully captures an
interesting case of allomorphy in Finnish.
130 LEXICALIZATION OF THE STRUCTURE 6.6
Chapter 7
Spell-out driven movement
7.1 Introduction
In Chapter 6, I presented the Nanosyntax framework and outlined in
abstract terms how the processing of Spell-out works and what principles
regulate the competition for lexical insertion. The aim of this chapter is
twofold: first, it deals with the question of how the syntactic structure is
linearized and develops the idea that the movements required to obtain
the suffixal ordering of spatial markers (and suffixes in general) under
a universal Specifier-Head-Complement template (Kayne 1994, Cinque
2009) are the result of the particular shape of the lexical entries stored
in the lexicon. In this, I follow a proposal made by Starke (2011) and
developed in more detail in Caha (2010b) that movement can be triggered
by the need to create the right configuration for Spell-out. The second
aim of this chapter is to test empirically the Spell-out model presented in
Chapter 6 and the Spell-out driven movement suggested in the present
chapter. For this purpose, I explore the spatial case system of several
languages and investigate the ways the system captures phenomena like
portmanteau morphology and syncretisms.
7.2 Linearization
As already mentioned in Chapter 3, I assume, following Kayne (1994)
and work in cartography (Rizzi 1997, Cinque 1999; 2005; 2009), that
syntactic structure is universally ordered such that heads are to the left
of their complements and specifiers are to the left of their heads. When
it comes to the way the syntactic nodes are linearized, I assume Kayne’s
Linear Correspondence Axiom, given below:
131
132 SPELL-OUT DRIVEN MOVEMENT 7.2
(1) Linear Correspondence Axiom (LCA) (Kayne 1994):
If a non-terminal X asymmetrically c-commands a non-terminal
Y, then all terminal nodes dominated by X will precede all terminal
nodes dominated by Y.
Asymmetric c-command means that X c-commands Y, but Y does not
c-command X. The c-command relation is defined as in (2).
(2) C-command (Kayne 1994:16):
X c-commands Y iff X and Y are categories and X excludes Y and
every category that dominates X dominates Y.
Exclusion, in its turn, is defined as in (3).
(3) Exclusion (Chomsky 1986:9):
X excludes Y if no segment of X dominates Y.
As an illustration of how the LCA works, take the structure in (4),
where the projection CP has adjoined to the projection BP, which results
in BP being a category with two segments (May 1985, Chomsky 1986).
(4) AP
A
a
BP
CP
C
c
BP
B
b
DP
D
d
In (4), CP c-commands BP because:
1. CP and BP are categories
2. No segment of CP dominates BP, hence CP excludes BP
3. Every category that dominates CP also dominates BP (for instance
AP)
BP, however, does not c-command CP because it does not exclude it
(there is one segment of BP dominating CP). Hence, CP asymmetrically
c-commands BP and by the LCA whatever is dominated by CP will
7.3 SPELL-OUT DRIVEN MOVEMENT 133
precede whatever is dominated by CP, that is, c will precede both b and
d at Spell-out.
Other asymmetric c-command relations in (4) are: A asymmetrically
c-commands CP, B, DP and D. Thus a will precede b, c, and d. B
asymmetrically c-commands D, therefore b will precede d. Knowing that
c precedes both b and d, the result is that the terminal nodes will be
linearized as a-c-b-d.
The formulation of LCA in (1) allows only for a linearization of the
terminal nodes in the syntactic structure. If lexical items are taken to also
spell out non-terminal nodes, as I have assumed, it becomes necessary to
reformulate the LCA, so that it is stated over both terminal and phrasal
nodes. I will therefore adopt the following version of LCA:
(5) Linear Correspondence Axiom (Phrasal Spell-out formulation):
If a non-terminal X asymmetrically c-commands a non-terminal
Y, then whatever spells out X precedes whatever spells out Y.
With the modification in (5), the asymmetric c-command relation is
mapped onto a linear order of non-terminal nodes and thus becomes applicable
in the Phrasal Spell-out system adopted here. In what follows,
I explore how exactly this happens.
7.3 Spell-out driven movement
As a first example, I take the spatial case system of the Daghestanian
language Karata (Magomedbekova 1971). Karata has six series: -t
˙
‘ on,
-t
˙
‘’ under, -č at, -l‘ inside, -a vertical attachment, and -∅ in,
among.1
It has a Locative suffix with four allomorphs -i, -a, -o and
-∅. The choice of a particular Locative allomorph depends on the series
marker it attaches to, for instance, -a goes with -t
˙
‘ and -t
˙
‘’, -o goes with
č. There is an Allative suffix -r attaching to the Locative marker, and an
Elative suffix -gal syncretic between Source and Route, which attaches
to the Locative marker, too. Although the number of allomorphs of the
Locative case is only less by two than the total number of spatial series,
Magomedbekova draws a morpheme boundary between the series marker
and the Locative ending. Taking her analysis for granted, I therefore
suggest that the lexical specifications associated with the entries of the
1
The distinction between -t’ and -t‘’ lies in the “strength” of the consonant — -t
˙
‘
is the lax variant and -t
˙
‘’ is the tense variant of the lateral affricate (represented as
¹ I and ¹ , respectively, in the Russian orthography).
134 SPELL-OUT DRIVEN MOVEMENT 7.3
Locative, Allative and Elative morphemes are as presented in (6):2
• The locative morpheme lexicalizes the Place head – (6a-d).
• The Allative morpheme -r lexicalizes Goal – (6b).
• The Elative morpheme -gal lexicalizes Source and Goal when it is
used as a Source marker – (6c) – and Route, Source and Goal when
it is used as a Route marker – (6d).
• The series markers -t
˙
‘, -t
˙
‘’, -č, etc. lexicalize the AxPart head –
(6a-d).
(6) a. Place AxPart DP
-a -t
˙
‘
b. Goal Place AxPart DP
-r -a -t
˙
‘
c. Source Goal Place AxPart DP
-gal -a -t
˙
‘
d. Route Source Goal Place AxPart DP
-gal -a -t
˙
‘






In what follows I will run the lexicalization of the spatial cases in
Karata, going step by step through the derivation of the “biggest” Route
structure, whose underlying structure is the following (as argued for in
Chapter 4):
2
Note that in the schematization in (6), the morphemes are presented in reverse
order.
7.3 SPELL-OUT DRIVEN MOVEMENT 135
(7) RouteP
Route SourceP
Source GoalP
Goal PlaceP
Place AxPartP
AxPart DP
As an example, I will take the on-series, illustrated below on the noun
bajdan ‘square’ (data from Magomedbekova 1971:73).
(8) a. bajdan-t
˙
‘-a
square-on-loc
‘on the square’
b. bajdan-t
˙
‘-a-r
square-on-loc-goal
‘to the square’
c. bajdan-t
˙
‘-a-gal
square-on-loc-route
‘from/through the square’
At the very first stage of the derivation, we have an AxPart externally
merged with a DP. This step is immediately followed by lexical access
according to the assumption of cyclic Spell-out (recall (65) in Chapter
6).
(9) AxPartP
AxPart DP
The lexicalization procedure first spells out the DP node as bajdan ‘square.’
I will remain agnostic about precisely which node in the extended nominal
projection the entry bajdan lexicalizes, as I would like to focus on
the spatial portion of the functional sequence. For the same reason, I
will not be marking the lexical material inserted in the complement of
prepositional phrases in the diagrams in the following chapters.
The next node in line is AxPart and the lexicalization procedure
searches for an entry in the lexicon which can lexicalize it. As suggested
136 SPELL-OUT DRIVEN MOVEMENT 7.3
above, the series marker -t
˙
‘ lexicalizes the AxPart head, thus a plausible
assumption is that its lexical entry has the following shape:
(10) -t
˙
‘ ⇔ </t
˙
‘/, AxPart, on>
In (10), it is assumed that the entry for -t
˙
‘, which spells out the single
feature AxPart, is specified for a structure consisting of just the head
AxPart. Under this assumption, which I will revise slightly later, the
AxPart node is lexicalized as in (11).
(11) AxPartP
-t
˙
‘ ⇐ AxPart DP
As Spell-out proceeds bottom-up alternating terminal and non-terminal
nodes, the next node which becomes the target for insertion is the root
node, AxPartP. The lexicalization procedure now searches the lexicon
to find a lexical item that can be inserted straight into AxPartP. There
is no such matching item in the lexicon which would spell out AxPart
together with DP, so nothing is inserted.
The outcome of Spell-out at this stage is that each feature in the
syntactic structure is realized at the end of the cycle, in conformity with
the Cyclic Exhaustive Lexicalization requirement in (33), Chapter 6.
Let us now turn to the linearization of the elements in the structure.
As is evident from (8), the DP in Karata precedes the series marker,
hence, the DP must asymmetrically c-command the AxPart head, according
to the LCA. This can be achieved if the DP raises above the
AxPartP node, as in the diagram below.
(12) AxPartP2
DP AxPartP1
-t
˙
‘ ⇐ AxPart tDP
The movement of the DP in (12) creates an adjunction structure
yielding a two-segment category. In order to facilitate the discussion, I
will annotate the lowest segment of a category by a subscript 1 (e.g.,
AxPartP1), and the higher segments by a subscript 2, 3, etc. (e.g.,
AxPartP2). In this new structure, DP c-commands AxPart, but AxPart
does not c-command DP. The asymmetric c-command relation between
DP and AxPart is converted into a precedence relation and the structure
7.3 SPELL-OUT DRIVEN MOVEMENT 137
can be linearized.
A potential drawback faced by this derivation is the lack of an apparent
trigger for the movement in (12) apart from the need to obtain the
correct word order. The Phrasal Spell-out system offers a possible answer
to the question what causes the DP to move in (12). The solution builds
on the idea of Starke (2011) that (this type of) movement is driven by
the need to spell out (see also Caha 2010b who explores the same idea in
the domain of case marking). What we need is to revise the requirement
on matching, plus change the shape of the lexical entry for -t
˙
‘.
Assume, first, that the matching procedure ignores traces (Caha 2009b),
which leads to the following reformulation of the definition in (47), Chapter
6, Section 6.4.
(13) A vocabulary item matches a node if its lexical entry is specified
for a constituent containing that node, ignoring tracesß.
Assume, in addition, that the lexical entry for -t
˙
‘, is not the one in (10),
but the one in (14), where -t
˙
‘ is specified both for the AxPart head and
its projection.
(14) -t
˙
‘ ⇔ </t
˙
‘/, AxPartP
AxPart
, on>
Let us now run the derivation using the entry in (14). The first step is
again the Merge of AxPart and DP.
(15) AxPartP
AxPart DP
Starting from the rightmost bottom-most element, the first two steps in
the lexicalization procedure are identical to the ones described for (11):
first it spells out the DP and then turns to the AxPart node. The lexical
item in (14) matches this node by virtue of the Superset Principle —
its entry is specified for a structure that contains this node — and it is
chosen for insertion. Proceeding up the tree, the Spell-out operation then
tries to lexicalize the next node AxPartP. This time there is a potential
matching lexical entry, the same morpheme -t
˙
‘, which could now make
use of its full lexical specification. However, -t
˙
‘ is not a perfect match
for the AxPartP node, as it does not contain the DP node. In order for
matching to obtain, the DP has to move out and leave a trace, which
will be ignored for the purposes of matching.
138 SPELL-OUT DRIVEN MOVEMENT 7.3
(16) AxPartP2
DP AxPartP1 ⇒ -t
˙
‘
AxPart tDP
The diagram in (16) differs from the one in (12) in that the series
marker -t
˙
‘ is inserted at a phrasal node instead of a terminal. The linearization
of the elements comes out crucially the same: the DP in (16)
c-commands AxPartP, but AxPartP does not c-command DP because it
does not exclude it. Consequently, the lexical material lexicalizing DP
precedes the lexical material lexicalizing AxPartP and -t
˙
‘ correctly comes
out as suffixal.
The attractiveness of this proposal is that now we have a trigger for
the movement in (16). As the lexicalization proceeds from node to node
and tries to match a lexical item at each step, once it hits the nonterminal
AxPartP, it will match the entry in (14) and this will cause the
movement of the DP so that the right constituency is achieved. In other
words, the evacuation of the DP is a Spell-out driven movement caused
by the Spell-out procedure lexicalizing the non-terminal AxPartP node
with the entry in (14). Thus, the trigger for the movement lies in the
shape of the lexical entry for -t
˙
‘: it stores the structure of AxPartP, but
importantly without the DP.
Note that the Spell-out driven movement in (16) is not triggered by
the need to lexicalize all the features present in the tree. The requirement
of Cyclic Exhaustive Lexicalization is met both in (11) and in (16). Thus,
if the entry for -t
˙
‘ were specified only for the AxPart head, as in (10),
then Spell-out would occur without any movement and the AxPartP node
would be lexicalized by inheritance instead of directly. The derivation in
(16) presupposes that direct lexicalization with movement is preferred to
lexicalization by inheritance but without movement. This is reminiscent
of the discussion in the final paragraph of Section 6.3.2, Chapter 6, where
I suggested that the preference of direct lexicalization over lexicalization
by inheritance captures the fact that portmanteau morphemes (e.g.,went)
block analytical expressions which express the same features and concept
(e.g., *go-ed). Thus, here we have the same principle, with the additional
twist of Spell-out driven movement: the lexicalization procedure tries to
match lexical material to each syntactic node in the tree and if there is
a potential match that requires the evacuation of non-matching features,
the evacuation will take place.
7.3 SPELL-OUT DRIVEN MOVEMENT 139
Let us now continue with the derivation of a Karata Place phrase,
whose structure is derived by merging the Place head to the tree in (16).
(17) PlaceP
Place AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
The Place marker -a is suffixal and attaches to a noun marked by one of
the series markers. Applying the same strategy as for the series marker
t
˙
‘, we can assume that -a triggers a movement of AxPartP to a position
from which it asymmetrically c-commands it. The shape of the structure
stored in the entry for -a will therefore have to be the following (I list -i,
-o and -∅ as allomorphs of -a):
(18) -a ⇔ </a, i, o, ∅/, PlaceP
Place
>
Proceeding again node by node up the tree, the lexicalization procedure
first targets the Place head in (17). The entry for -a is chosen for insertion,
as it matches Place by the Superset Principle, i.e., -a is specified for
a structure which contains this constituent. At the next step, the node
PlaceP is subject to lexicalization and the entry -a is chosen again, since
it is a match again, this time using its full specification. In order for the
right configuration for Spell-out to be achieved, AxPartP2 has to move
away, adjoining to PlaceP, as shown in (19).
(19) PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPartP2
In the tree in (19), DP asymmetrically c-commands AxPartP and AxPartP
asymmetrically c-commands PlaceP. As a result, whatever spells
140 SPELL-OUT DRIVEN MOVEMENT 7.3
out DP will precede whatever spells out AxPartP and whatever spells
out AxPartP will precede whatever spells out PlaceP, thus leading to the
correct linearization DP-t
˙
‘-a.
The same strategy can be used also in the derivation of the Goal
expression bajdan-t
˙
‘-a-r. What we need is to assume that the Goal suffix
-r is specified in the way shown in (20).
(20) -r ⇔ </r/, GoalP
Goal
>
The Spell-out of GoalP by -r will trigger an evacuation movement of the
complement of Goal, leading to the structure in (21).
(21) GoalP2
PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPart2
GoalP1⇒-r
Goal tPlaceP2
Thus, the reverse ordering of the morphemes in the Karata Locative and
Goal expressions comes about as the result of successive roll-up movements
triggered by the shape of the lexical entries.
The derivation becomes different when we reach the Source and Route
heads, lexicalized by the morpheme -gal. Similarly to the Goal suffix -r,
-gal attaches to a noun marked by the Locative case. But unlike -r, -gal
corresponds to more than one feature, more precisely, the features Goal,
Source, and Route. The entry for -gal must therefore have the following
shape.
7.3 SPELL-OUT DRIVEN MOVEMENT 141
(22) -gal, ⇔ </gal/, RouteP
Route SourceP
Source GoalP
Goal
>
Such an entry can be used for the lexicalization of a Source expression
by virtue of the Superset Principle. Leaving aside the category/segment
distinction for now, the entry in (22) is a match for the SourceP node
provided PlaceP2 moves out. (Note that -gal is a match also for the
GoalP node, but loses the competition to the more highly specified entry
-r.)
(23) SourceP2
PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPartP2
SourceP1⇒gal
Source GoalP2
tPlaceP2
GoalP1
Goal tPlaceP2
Finally, I turn to the “biggest” structure of Route paths, derived by
the Merge of Route on top of SourceP.
142 SPELL-OUT DRIVEN MOVEMENT 7.3
(24) RouteP
Route SourceP2
PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPartP2
SourceP1⇒gal
Source GoalP2
tPlaceP2
GoalP1
Goal tPlaceP2
Starting again from the lowest node in the cycle, the lexicalization first
targets the Route head, matched by -gal by the Superset Principle. Then
it inspects the RouteP node, which -gal matches again (abstracting away
from the two-segmental nature of the categories GoalP and SourceP),
but PlaceP2 has to evacuate. The evacuation movement takes place and
-gal is inserted at RouteP1, overriding the material inserted below. Thus,
the entry -gal triggers a successive-cyclic movement of PlaceP2 within its
own projection line.
(25) RouteP2
PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPartP2
RouteP1⇒-gal
Route SourceP2
tPlaceP2
SourceP1
Source GoalP2
tPlaceP2
GoalP1
Goal tPlaceP2
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 143
To sum up, the idea of a Spell-out driven movement provides us with
a tool to capture the linearization of suffixes. The Karata spatial markers
come out as suffixal because of the particular configuration of the
trees stored in their entries. The entries trigger evacuation movements
of the nodes which obstruct matching. These evacuation movements are
instantiated in two ways: one way is a roll-up movement which takes
place at the “border” between two lexical entries, e.g., the movement
triggered by the Goal entry -r and the Locative entry -a. The other
way is a successive-cyclic movement triggered by a lexical item within its
own projection line, e.g., the movement triggered by the morpheme -gal,
which covers a span in the structure consisting of more than one feature:
Goal, Source and Route.
7.4 Elaborating Spell-out driven movement
The derivation of Karata spatial expressions raises a number of questions
regarding the nature of the proposed Spell-out driven movement.
In this section, I discuss in more detail the exact mechanics of Spell-out
driven movement. I address three questions, which can be summarized
as When? Where? and What? Specifically, the questions are: (i) when
exactly does the Spell-out driven movement happen in the cyclic lexicalization
model assumed here, (ii) what is the landing site for the evacuated
material, and (iii) what restrictions are there on the nature of the evacuated
constituent(s)? I explore these issues in Section 7.4.1, 7.4.2 and
7.4.3, respectively.
7.4.1 Timing of Spell-out driven movement
The Spell-out driven movement proposed in the preceding section has two
effects. First it creates the right syntactic configuration for the insertion
of a particular lexical entry. Second, it creates new nodes in the syntactic
structure — the higher segments of the categories to which the evacuated
material adjoins. The discussion presented in this section is prompted by
the assumption that these new nodes are subject to lexicalization. The
motivation for this assumption is purely empirical: if the nodes created
by Spell-out triggered movement are subject to lexicalization, then we
can capture an interesting and intricate case of allomorphy in Finnish.
I present this case in Section 7.6, where I show how the lexicalization
of the Finnish Illative phrase by the portmanteau suffix -h(V)n can be
accounted for if we assume that it lexicalizes such a node derived by
144 SPELL-OUT DRIVEN MOVEMENT 7.4
Spell-out triggered movement.
Assuming that the nodes derived by Spell-out triggered movement
are to be lexicalized just like any other nodes in the syntactic tree, I will
now investigate how this idea can be made compatible with the adopted
model of cyclic Spell-out.
Recall that the lexicon is accessed multiple times in the course of the
derivation, more precisely, after each addition of a new feature (External
Merge), as schematized in (26).
(26) External Merge Lexical access
The Spell-out driven movement does not add a feature. It is simply an
evacuation movement of already merged features triggered by the need to
create the right syntactic environment for a given entry to match. There
is therefore no reason to assume that the lexicon is immediately accessed
after this operation. If there is no lexical access after Spell-out driven
movement, it should take place before the External Merge operation in
the cyclic model in (26). Thus, I assume that Spell-out driven movement
(Internal Merge) is followed by the addition of a feature to the derivation
(External Merge), which is then followed by lexical access.
(27)
External Merge
Lexical access
(Internal Merge)
The schematization of a cycle proposed in (27) maintains the idea that
lexical access happens only after a new feature has been introduced to the
derivation. At the same time, it encodes the fact that Spell-out triggered
movement is instigated only after the lexicon has been consulted, that is,
when a lexical item is chosen which would match only if movement takes
place.
The operation of Internal Merge is not an obligatory step in a cycle
and I have marked this optionality by putting it in parentheses. The reason
is that no movement takes place when there is a one-to-one matching
item in the lexicon.
With the revised architectural scheme of a cycle presented in (27),
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 145
I turn to the question of how to implement it in the lexicalization algorythm
in a way such that the nodes derived by Internal Merge are
spelled out. To give a short preview, incorporating the step of Internal
Merge into the cyclic Spell-out system will lead to the conclusion that the
Spell-out driven movement triggered by the matching of a lexical item in
a given cycle actually happens in the next cycle, a view espoused also in
Caha (2010b). I detail the reasoning below.
Consider the lexical entries in (28) and the structure in (29).
(28) a. a ⇔ </a/, A>
b. b ⇔ </b/, BP
B
>
(29) BP
B A
The first cycle in the derivation begins with the External Merge of A
and B. (As already stated in Chapter 6, fn. 3, I assume that A does not
constitute a cycle by itself, following the definition of Merge in Chomsky
2001:3 and Adger 2003:54). According to the assumption in Chapter
6, Section 6.3.2, the lexicalization proceeds right-to-left and bottom-up
within a cycle (as well as between cycles). Therefore, the first node to be
lexicalized in this first cycle is A. The entry a is a match for that node.
The next node in line is B, which, according to the Superset Principle, is
matched by b, since B is a subconstituent of the structure b is specified
for. Finally, b is matched to the BP node, see (30).
(30) BP⇒b
b⇐B A⇒a
In order to spell out BP by b, an evacuation of A takes place and the
node BP2 is created.
(31) BP2
a⇐A BP1⇒b
B tA
The movement of A to BP2 is an Internal Merge operation, therefore it
is not followed by a step of lexicalization.
146 SPELL-OUT DRIVEN MOVEMENT 7.4
The next time the lexicon is accessed is when a new feature, say C,
is added to the derivation.
(32) CP
C BP2
a⇐A BP1⇒b
B tA
Since the Merge of C is an External Merge operation, there is a round
of lexical access. As I already mentioned in the introduction to this
subsection, there is empirical evidence in favor of the idea that the nodes
derived by Spell-out driven movement are subject to lexicalization. Node
BP2 should then be a target for lexicalization, just like the nodes C and
CP resulting from External Merge. According to the assumption made
in Chapter 6, Section 6.3.2, the lexicalization procedure targets only the
nodes in the current cycle and does not “look inside” the preceding cycles,
which have already been spelled out. Therefore, we need to assume that
the BP2 node is part of the second cycle. This, in turn, necessitates that
the movement of A to BP2, which creates the BP2 node and which is
triggered by the matching of b to the BP node in the first cycle, actually
takes place in the second cycle. To illustrate this more clearly, I present
a scheme of the operations applying in Cycle 1 and Cycle 2 in (33).
(33) a. Cycle 1
(i) Merge A and B (External Merge)
BP
B A
(ii) Match A, B and BP (Lexical access)
BP⇒b
b⇐B A⇒a
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 147
(iii) Mark A for extraction
BP⇒b
b⇐B a⇐A<BP2>
b. Cycle 2
(i) Extract A (Internal Merge)
BP2
a⇐A BP1⇒b
B tA
(ii) Merge C (External Merge)
CP
C BP2
a⇐A BP1⇒b
B tA
(iii) Match BP2, C and CP (Lexical access)
The alternative, where the evacuation movement of A takes place in the
same cycle where it has been triggered (i.e., in the first cycle), leads to a
failure to lexicalize the BP2 node under the present set of assumptions.
The reason is that the BP2 node is created in the first cycle, and would not
be accessed by the lexicalization round triggered in the second cycle, if it
sees only the nodes in the second cycle. At the same time, the movement
creating the BP2 node in the first cycle would follow the lexicalization
round of the first cycle, hence the first lexicalization round could not
target BP2. This is represented schematically in (34).
148 SPELL-OUT DRIVEN MOVEMENT 7.4
(34) a. Cycle 1
(i) Merge A and B (External Merge)
BP
B A
(ii) Match A, B and BP (Lexical access)
BP⇒b
b⇐B A⇒a
(iii) Mark A for extraction
BP⇒b
b⇐B a⇐A<BP2>
(iv) Extract A (Internal Merge)
BP2
a⇐A BP1⇒b
B tA
b. Cycle 2
(i) Merge C (External Merge)
CP
C BP2
a⇐A BP1⇒b
B tA
(ii) Match C and CP (Lexical access)
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 149
Hence, a cycle begins with a (potentially vacuous) step of Internal
Merge and ends with lexicalization (see (33b)). Apart from allowing for
the Spell-out of the nodes derived by Internal Merge, this model also has
the advantage that the syntactic module, where Internal and External
Merge are performed, is accessed just once in a cycle.
Note that, in the second cycle of the derivation depicted in (33b),
I have assumed that node A is not inspected for lexicalization. This
assumption is motivated by Starke’s (2011) proposal that the system
remembers the outcomes of previous matching procedures. Since A has
already been matched by a in the first cycle, there is no need to perform
a new search. As far as I can see, nothing hinges on this assumption. No
matter whether the lexicalization procedure matches a to A again in the
second cycle, or just skips A, the outcome is the same.
The idea that Spell-out movement triggered in one cycle actually
happens in the next cycle raises an interesting question regarding the
final cycle in a derivation. Put briefly, the question arises because of the
possibility of triggering movement which would happen after the final
External Merge operation has taken place. The node created by this
movement will therefore not be followed by External Merge. Hence, no
more lexical access can take place. As a result, this node cannot be
lexicalized.
For a more detailed description of this hypothetical scenario, assume
the lexical entries c and d with the lexical specifications in (35), and the
phrase marker in (36).
(35) a. c ⇔ </c/, BP
A BP
B
>
b. d ⇔ </d/, CP
C
>
c. e ⇔ </e/, CP
BP
A BP
B
CP
C
>
150 SPELL-OUT DRIVEN MOVEMENT 7.4
(36) CP2
c⇐BP2
A BP1
B tA
CP1⇒d
C tBP
The matching of d to node CP1 in cycle X will trigger an evacuation of
BP at cycle X+1. Suppose now that this is the end of the derivation,
that is, no new morphosyntactic features are added to the tree. Will
node CP2 then be subject to lexical insertion? If yes, then the phrase
marker in (36) will be spelled out as e, because the matching entry in
(35c) will be inserted at CP2. If not, then it will be spelled out as c+d.
The data necessary to decide which of the two predictions is the correct
one lies beyond the empirical domain of this thesis, which focuses on
the lower portion of the PP. The derivation of a PP does not end with any
of the heads Goal, Source, or Route. There is still more structure above
them, for instance, a little p head introducing the Figure (Romanova
2007), or DegP hosting measure phrases (Svenonius 2010, den Dikken
2010). Furthermore, a PP by itself presumably cannot constitute an
utterance, assuming that sentences consisting of just one PP are cases
of ellipsis. Therefore, one would have to look at the syntax of clausal
structures expecting to find (or not find) a difference in their lexicalization
depending on whether the derivation ends with the clause, or not
(i.e., the clause is embedded). The examination of such data would lead
me too far astray from the main research topic and I will therefore not
investigate it here. Still, it is worth mentioning that a plausible hypothesis
is to assume the availability of an abstract termination symbol of
the type suggested by Kayne (1994:Ch.4), which is merged at the end of
the derivation and which allows or disallows (depending on the empirical
evidence) one more lexicalization round.
7.4.2 The landing site of evacuated material
Now I turn to the questions concerning the adjunction site for the nodes
evacuated due to the need to spell out. Under the view that adjunction
can target any phrasal node, it stands to reason to assume the following
two possible constraints on the landing sites of the obstructing projec-
tions:
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 151
(37) a. Extension Condition: The evacuated material has to adjoin
to the root node of the tree in the workspace.
b. Shortest Move: The evacuated material has to adjoin right
above the node where matching takes place.
The first adjunction site obeys the Extension Condition of Chomsky
(1995:190) according to which syntactic operations apply to the
root of the tree. The second adjunction site complies with the Shortest
Move constraint of Chomsky (1995:90), which requires that movement
should be as short as possible. The precise formulation of Shortest Move
which I present above differs from the one in Chomsky (1995:90), Zwart
(1996:307), and Richards (2001:97), where Shortest Move is phrased as a
ban on a movement of α across a potential landing site for α. The reason
I use the term Shortest Move in (37b) is because, assuming that lowering
operations are not available for Spell-out driven movement, the shortest
possible movement of an obstructing node will place it just above the
node where the lexical item triggering the movement is to be inserted.
In the derivations presented until now, the Spell-out driven movement
satisfies both constraints simultaneously, as adjunction to the root
node is at the same time the shortest move possible. However, as the
tree structures become larger and the derivations more complicated, it
becomes possible to tease apart the two potential landing sites. As far
as I can see, the availability of multiple adjunction sites leads to the necessity
of performing parallel derivations and choosing the optimal one.
In order to avoid such parallel derivations, I suggest in this section that
the number of landing sites should be restricted to one and assume that
the Spell-out triggered movement obeys Shortest Move.
In order to illustrate what type of problems arise if more than one
landing site is available, I present here a hypothetical case, asking the
reader to bear with me through a rather complicated derivation.
Consider again the lexical entries a and b and the abstract structure
in (39), repeated from (30).
(38) a. a ⇔ </a/, A >
b. b ⇔ </b/, BP
B A
>
(39) BP⇒b
b⇐B A⇒a
152 SPELL-OUT DRIVEN MOVEMENT 7.4
Next we merge C and assume the entry in (41).
(40) CP
C BP⇒b
B A
(41) c ⇔ < CP
C
>
The item c triggers an evacuation movement of BP in the next cycle.
Now c can be inserted at the CP node.
(42) CP2
b⇐BP
B A
CP1⇒c
C tBP
Note that at this stage, the CP2 node is still not lexicalized, as it is part
of the next cycle. Suppose now that we merge a head D to derive the
structure in (43).
(43) DP
D CP2
b⇐BP
B A
CP1⇒c
C tBP
Since this is an external Merge, it will trigger a round of lexicalization,
with the first node to be lexicalized being CP2 (assuming that BP is
skipped by virtue of already having been lexicalized in the previous cycle).
Suppose now that there are entries d and e in the lexicon with the
following specifications:
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 153
(44) a. d ⇔ < CP2
BP
B
CP1
C
>
b. e ⇔ < DP
D
>
The entry d matches the CP2 node if A evacuates. There are two possible
adjunction sites where A would not be “in the way”: (i) between CP2
and DP, in which case the Spell-out triggered movement will satisfy the
Shortest Move requirement, but violate the Extension Condition, and (ii)
above DP, in which case the Extension Condition will be obeyed, but not
Shortest Move (assuming that adjunction to CP2 is possible).
Suppose that both options are available, that is, A can adjoin either
to DP or to CP2. Let us then assume that A adjoins to CP2 yielding the
structure below.3
(45) DP
D CP3
a⇐A CP2⇒d
BP
B tA
CP
C tBP
In (45), A has adjoined right above the node where matching takes
place (Shortest Move) thus making it possible for d to spell out the node
CP2. The next node in line is D, which can be spelled out by e, since
its entry contains the D head. Then, the DP node becomes a target for
lexicalization and e is inserted again, this time using its full specification
and triggering an evacuation movement of the CP3 complement.
3
In this, I also assume that subextraction from a left branch is a legitimate operation,
see Corver (1990) and Bošković (2005) for relevant discussion.
154 SPELL-OUT DRIVEN MOVEMENT 7.4
(46) DP2
CP3
a⇐A CP2⇒d
BP
B tA
CP1
C tBP
DP1⇒e
D
The final step in this thought experiment is the Merge of a head E,
which starts a new cycle.
(47) EP
E DP2
CP3
a⇐A CP2⇒d
BP
B tA
CP1
C tBP
DP1⇒e
D
Suppose now there is an entry f with the specification in (48).
(48) f ⇔ < EP
E DP3
CP2
BP
B
CP1
C
DP2
A DP1
D
>
The hypothetical entry f is a possible lexical item, because it can be
generated by syntax. More precisely, this would have been the syntactic
structure generated if A had adjoined to the root node in (43) and CP2
had adjoined to DP2. Given the assumption that adjunction to the root
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 155
node is available, the tree stored in (48) is a licit syntactic structure. It
can be therefore paired with phonological content, stored in the lexicon
and used for lexicalization (provided, of course, that it can be learned).
The entry f is eligible for insertion at the EP node but only if A had
been adjoined to DP in the previous cycle. There is, however, no way for
the system to know in advance what head will be merged to DP in the
next cycle, so that it chooses the suitable landing site for A.
Note that the adoption of Spell-out triggered movement does not
circumvent the problem. Imagine that, by moving A, syntax creates
the right configuration of nodes for f to be inserted at EP, ignoring for
now the segment/category distinction. The trigger would be the same as
the one I suggested for the movement of DP to AxPartP in the Karata
spatial expressions (see the diagram in (16)): movement is induced so
that a node can be spelled out by a particular lexical entry. In order to
create a matching configuration, A will then have to adjoin to DP1, as
shown in (49).
(49) EP
E DP3
CP3
a⇐A CP2⇒d
BP
B tA
CP1
C tBP
DP2
A⇒a DP1⇒e
D
This operation is problematic as it instantiates a “sideways movement,”
where A does not move to a c-commanding position, contra the
proposal in Rizzi (1986), Chomsky (1995). I therefore assume that it is
unavailable.
The same complication arises if we assume that, in the cycle when
the structure in (43) is to be spelled out, the landing site of A is above
DP (Extension Condition). This will create the following structure.
156 SPELL-OUT DRIVEN MOVEMENT 7.4
(50) DP2
a⇐A DP1
D CP2⇒d
BP
B tAP
CP1
C tBP
The insertion of e at the DP node will trigger movement of CP2. Then
E is merged, see (51).
(51) EP
E DP3
CP2⇒d
BP
B tA
CP1
C tBP
DP2
A⇒a DP1⇒e
D
The same problem arises again if we assume that the entry for f
contains the tree in (52), which can be generated if A adjoins to the
node CP2 (Shortest Move).
(52) f ⇔ < EP
E DP2
CP3
A CP2
BP
B
CP
C
DP1
D
>
Again, either the system must have known about the Merge of E and the
existence of f in the previous cycle, so that it could make the right move,
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 157
or A has to undergo sideways movement to adjoin to CP2 (see (53)),
which I assumed to be prohibited (following Rizzi 1986 and Chomsky
1995).
(53) EP
E DP3
CP3
a⇐A CP2
BP
B tA
CP1
C tBP
DP2
A⇒a DP
D
One potential solution is to allow for there to be multiple parallel
derivations with the lexicon as a filter. Under this view, the derivation
will branch once the lexicalization procedure reaches the node CP2 (if
not even before that). In possible derivation #1, the movement of A
triggered by the insertion of d at CP2 will adjoin A to DP. In possible
derivation #2, A will be adjoined between CP2 and DP. At the later point
when the item f is to be inserted into EP, the incompatible derivation will
be filtered out. I am reluctant to adopt this solution, as it is based on
a “global comparison” approach where multiple derivations are processed
in a parallel fashion and the optimal one is chosen at the end. Although
assumed in some frameworks, for instance Optimality Theory (McCarthy
and Prince 1995, Prince and Smolensky 2004), researchers following the
Minimalist Program and related frameworks usually reject this approach
because of its high computational costs (Chomsky 1998, Collins 1997,
Yang 1997).
The alternative solution, which I pursue here, is to eliminate one of
the possible adjunction sites for Spell-out triggered movement so that no
choices need to be made. I will therefore assume that Spell-out driven
movement obeys the Shortest Move requirement. In this, I follow the proposal
put forward in Chomsky (1993), according to which the Extension
Condition does not apply in cases of adjunction.
158 SPELL-OUT DRIVEN MOVEMENT 7.4
7.4.3 Backtracking
Until now, all the instances of Spell-out triggered movement involved
cases when a single obstructing constituent had to move out so that the
right configuration for Spell-out is created. There are, however, cases
when more than one constituent is in the way. One such case is presented
by the lexicalization of a Source expression in Uzbek (and many related
Turkic languages).
Uzbek has three spatial cases (Boeschoten 1998): Locative (expressing
static location), Dative (marking Goal of motion as well as indirect
objects), and Ablative (expressing Source). The corresponding morphemes
are given in (54), where I label the Dative case Allative to stress
its Goal function.4
(54) a. Locative: -D˙a
b. Allative: -G˙a
c. Ablative: -D˙a-n
As can be seen from (54c), the Ablative marker is bi-morphemic. It
consists of the Locative morpheme -D˙a and the morpheme -n. We can
then assume that the morphemes in (54) are mapped to the syntactic
structure of spatial expressions in the following way:5
• The Locative morpheme -D˙a lexicalizes the Place head – (55a,c).
• The Allative morpheme -G˙a lexicalizes the Goal and Place heads –
(55b).
• The Ablative morpheme -n lexicalizes the Source and Goal heads
– (55c).
4
As I suggest in Chapter 10, Section 10.3, Uzbek (and numerous other languages
that use the Dative case to express goal of motion) does have an Allative case, but it
happens to be syncretic with the Dative.
5
In order to keep the discussion simple, I leave aside the AxPart projection and
focus on the Place, Goal and Source heads. The complete derivation will include an
initial cycle where AxPart and DP are merged. I assume that the AxPart head is
included in the specification of the Locative and the Allative morphemes and can be
lexicalized by them together with Place, and Place plus Goal, respectively.
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 159
(55) a. Place DP
-D˙a
b. Goal Place DP
-G˙a
c. Source Goal Place DP
-n -D˙a






Given that the Locative, Allative and Ablative markers are suffixal, we
can assume for them similar entries as for the spatial ending in Karata,
namely, entries which contain phrasal nodes and trigger movements.
(56) Locative suffix:
-D˙a ⇔ </D˙a/, PlaceP
Place
>
(57) Allative suffix:
-G˙a ⇔ </G˙a/, GoalP
Goal PlaceP
Place
>
(58) Ablative suffix:
-n ⇔ </n/, SourceP
Source GoalP
Goal
>
The lexicalization of a locative structure will then proceed as follows.
Leaving aside AxPartP for the sake of simplicity, the derivation begins
with the Merge of Place and DP.
(59) PlaceP
Place DP
There are two entries which can spell out the PlaceP node: the Locative
160 SPELL-OUT DRIVEN MOVEMENT 7.4
morpheme -D˙a in (56) and the Allative morpheme -G˙a in (57). The
suffix -D˙a wins the competition by the Elsewhere Principle (it is a better
match) and triggers a movement of DP to PlaceP2. The DP is accordingly
linearized before -D˙a.
(60) PlaceP2
DP PlaceP1⇒-D˙a
Place tDP
Next, the Goal head is merged into the structure, yielding (61).
(61) GoalP
Goal PlaceP2
DP PlaceP1⇒-D˙a
Place tDP
There are again two entries which can be inserted at the GoalP phrase
(still abstracting away from the category/segment distinction): the entry
-G˙a in (57) and the entry -n in (58). The Goal entry would trigger
an evacuation of the DP, the Source entry would trigger evacuation of
the entire PlaceP2. The Goal entry is chosen over the Source entry by
virtue of being a better match (it does not contain the superfluous Source
feature), see (62).
(62) GoalP2
DP GoalP1⇒-G˙a
Goal PlaceP2
tDP PlaceP1
Place tDP
Finally, the Source head is merged, leading to the structure in (63).
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 161
(63) SourceP
Source GoalP2
DP GoalP1⇒-G˙a
Goal PlaceP2
tDP PlaceP
Place tDP
The Ablative entry (58) can be inserted at the SourceP node, crucially
necessitating two obstructing nodes to move away: the PlaceP2 node and
the DP node adjoined to GoalP2. Let us then see what happens if two
movements indeed take place: first PlaceP2 and then DP, yielding the
structure in (64).
(64) SourceP3
DP SourceP2
PlaceP2
tDP PlaceP1
Place tDP
SourceP1
Souce GoalP2
tDP GoalP1⇒-G˙a
Goal tPlaceP2
The result is a structure which cannot be linearized by the LCA: DP and
PlaceP asymmetrically c-command each other. Hence, assuming that
linearization is governed by the LCA, this is cannot be the structure
underlying the Uzbek Source phrase. The same problem arises if we
assume that the DP moves first, followed by the evacuation of PlaceP2.
Empirically, we know that the DP precedes the Locative marker, and
the Locative marker precedes the Ablative. The LCA then enforces a
structure of a Source phrase like the one in (65).
162 SPELL-OUT DRIVEN MOVEMENT 7.4
(65) SourceP2
PlaceP2
DP PlaceP1
Place tDP
SourceP1
Source GoalP2
tDP GoalP1⇒-G˙a
Goal tPlaceP2
So, the question is what went wrong with the derivation of SourceP and
why did we end up with the non-linearizeable syntactic representation in
(64)?
Plausibly the answer is not in the number of movements that had
to take place — two instead of one. If a lexical entry can trigger one
evacuation movement, then it should also be able to trigger two. Restricting
the number of Spell-out triggered movements per item seems to
me ungrounded.
A comparison between the structure in (65) and the one in (64) reveals
the most likely answer. In (64), the DP and the PlaceP are two separate
constituents, each of which undergoes a movement by itself. In (65),
however, the DP and PlaceP2 form one constituent, suggesting that the
evacuation movement has displaced the entire constituent. Guided by
this observation, I would like to suggest, following Cinque (2005), that
the problem lies in the fact that one of the two movements triggered in
(64) is a movement that does not contain the head noun. More precisely,
the problematic movement is of PlaceP2 as it contains only a trace of
the DP. As Cinque (2005) suggests, word order movements within the
extended nominal projection are restricted by the requirement that the
moved constituent contains the NP.
(66) Restriction on movement (Cinque 2005:321)
Neither head movement nor movement of a phrase not containing
the (overt) NP is possible.
This restriction has been productive in accounting for the attested and
unattested orderings of the nominal modifiers Demonstrative, Numeral,
and Adjective with respect to each other and the NP. Assuming that the
heads Place, Goal, Source, and Route are part of the extended projection
of the noun (an assumption dating back to Grimshaw 1991, who suggest
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 163
that P is part of the extended NP, for a more modern cartographic representation
see, for instance, Bye and Svenonius 2011:ex.(6)), it is logical
to expect that the restriction on movement in (66) applies also to the
word order movements within the structures discussed in this thesis.
Thus, I propose that the requirement that each moved constituent
contains the head noun extends also to the spatial structures explored
here. The impossibility of performing two evacuation movements in (64)
is due to a violation of this requirement. Namely, PlaceP2 does not
contain the head noun.
How does the system then derive a Source phrase? By the time when
Source is merged in (63), the DP has already extracted from PlaceP2
due to the insertion of -G˙a at GoalP. By what mechanism then does the
structure in (65) arise?
There are two conceivable solutions. The first one is to let the system
compute all possible derivations from a given numeration and then
compare them so that the most optimal is chosen. In the pool of possible
derivations there is one where PlaceP and DP move as one constituent
after the Goal head is merged. Specifically, this is the way the derivation
would have proceeded if the Source entry -n were chosen for insertion
at the GoalP node. In other words, when the Goal head is merged,
the derivation line branches: there are two possible continuations of the
derivation: D1 where the sole DP evacuates and the Goal entry is chosen
for insertion at GoalP1 (diagrammed in (62)), and D2 where PlaceP2 is
evacuated and the Source entry is inserted at GoalP1, see (67).
(67) GoalP2
PlaceP2
DP PlaceP1
Place tDP
GoalP1⇒-n
Goal
Computing now two derivations instead of one,6
the derivation continues
with the Merge of the Source head. The only entry specified for the
Source feature is -n, so it will be chosen for insertion in both D1 and D2.
In D1 two evacuation movements will be triggered, leading to a failure to
6
In fact, at least four, as the derivation will have already branched once at the
PlaceP node, where two entries match — the Allative and the Locative. In fact, since
branching can potentially occur every time the lexicon is accessed, the number of
possible derivation grows at an exponential rate.
164 SPELL-OUT DRIVEN MOVEMENT 7.4
linearize the structure. In D2, the movement triggered by the Spell-out
of node SourceP by -n will displace DP and PlaceP together, preserving
the asymmetric c-command relation between DP and PlaceP. When at
the end of the day the system determines the optimal derivation, D2 is
selected over D1.
As already mentioned in Section 7.4.2, such a system is computationally
costly, because there is a huge number of parallel derivations to be
executed and compared globally. As argued for by Collins (1997) and
Yang (1997), global comparisons render the derivations intractable and
I am therefore reluctant to adopt this solution.
An alternative solution involving no global comparison is to allow
the system to retreat to the last step in the derivation when PlaceP2
contains the DP and from that point on follow a derivation path where
they are moved as one inseparable unit. The rationale behind it is the
following: at a certain stage of the derivation two nodes are marked
for extraction: DP and PlaceP2. The latter cannot be moved, though,
because it does not contain the DP. Importantly, the DP has to evacuate
just like PlaceP2, so a way out is to move it together with PlaceP2. The
system therefore backtracks to the point where PlaceP2 and DP form
one constituent. For this purpose the system must remember the path
traversed by the derivation so that it is possible to reconstruct it to the
relevant step, that is, to the step of lexical access after the Goal head is
merged. Although this introduces some complexity, there is still a unique
derivational path to be traversed, albeit one with a step-back. Crucially
no parallel derivations have to be considered as in the global approach.
In more technical terms, the backtrack algorithm can be designed
in the following way: assuming that each of the syntactic operations in
(27) (Internal Merge, External Merge, Lexical access) marks a step in
the derivation, at the step when the lexicon produces the item -n as the
matching entry for the SourceP node, the derivation cannot continue due
to the impossibility of evacuating the PlaceP2 constituent without the
DP being contained in it. The impossibility of moving a node marked
for extraction triggers backtracking. The system thus retreats to the
last step of the derivation where the node marked for extraction can be
legitimately extracted (i.e., where PlaceP2 and DP form one constituent).
This is the step at which the lexicon is consulted right after the Merge
of Goal. In the schematic sequence of operations in Figure 7.1, this step
is marked as number 8, starting the count from the Merge of DP and
AxPart. The impossibility of continuing the derivation is represented by
the symbol “!” and the backtracking is marked as an arrow to Step 8 —
7.4 ELABORATING SPELL-OUT DRIVEN MOVEMENT 165
the last stage in the derivation when DP and PlaceP are a constituent.
Figure 7.1: Backtracking scheme in Uzbek Ablative expressions
8 Choose -G˙a → 9 Move DP → 10 Merge Source → 11 Choose -n → 12 !
8 Choose -n → 9 Move PlaceP2 → 10 Merge Source → 11 Choose -n → 12
Move PlaceP2
?
Back at Step 8, the system pursues an alternative way to lexicalize
node GoalP such that it obviates the need to extract DP from PlaceP2.
Specifically, it chooses the rejected competitor -n, since -n triggers a
movement of the entire PlaceP2 node, thus leading to a successful derivation
in the steps to follow. In the theoretical case where there is more
than one alternative to pursue, I suggest that the choice of alternative
entries is determined by a ranking imposed by the Elsewhere Principle:
the fewer unused features an item would leave, the higher its position on
the list of alternatives is.
Such cases of backtracking occur typically whenever we have two
structures in a subset-superset relation and the smaller structure is spelled
out by a portmanteau, while the bigger structure is spelled out analytically
by a sequence of morphemes. Importantly, the last one of the
sequence of morphemes has to be a portmanteau itself and its feature
specification has to partially overlap with the features of the portmanteau
spelling out the smaller structure. The reason for the backtracking
is the different types of movement triggered by the portmanteau and the
non-portmanteau morphemes. As already pointed out at the end of Section
7.3, portmanteau entries trigger a cyclic movement of an XP within
the structure they span (within the spatial fseq, XP=DP, or a projection
containing the DP).
166 SPELL-OUT DRIVEN MOVEMENT 7.4
(68) BP2
XP BP1
B AP2
tXP AP1
A tXP



entry b
Analytic expressions involve roll-up movements of XP plus pied-piped
material. Such roll-up movements occur at the border between two mor-
phemes.
(69) CP
BP2
XP BP1
B AP2
tXP AP1
A tXP
CP⇒entry c
C tBP



entry b
If the system needs to spell out the CP node in (69) by an entry d which
has both the feature C and B (i.e., overlaps with entry b), it will need to
backtrack to the stage when XP and AP form one constituent and then
move them together.
(70) CP
C BP
B AP2→moved constituent
XP AP1
A tXP






entry d
entry b
Note that, if there is an entry a which matches AP without leaving
7.5 SEGMENT VS CATEGORY MATCHING 167
unused features, it will be used to spell out the displaced AP.
(71) CP2
AP2
XP AP1⇒a
A tXP
CP1⇒d
C BP
B
As a result, the structure it (71) will be spelled out by a+d and not
as b+d. This is the reason why the Uzbek Source marker is stacked
to the Locative suffix (-D˙a-n) and not to the Allative suffix (*-G˙a-n).
Descriptively, it looks as if the -n has to realize its lowest feature. A
requirement to that effect has been proposed by Abels and Muriungi
(2008) and adopted by Caha (2009b) and Pantcheva (2010) under the
name of The Anchor Condition. With the current technology of Spellout
driven movement and backtracking, the Uzbek data are captured
without the need to adopt this restriction.
7.5 Segment-matching versus category-
matching
Until now, I have been ignoring the segment/category distinction within
the lexical entries, allowing for an entry of the sort in (72) (repeated
from (22) in Section 7.3) to spell out the SourceP1 node in the structure
in (73), despite the fact that the GoalP category contains two segments
((73) is repeated from (23) in Section 7.3).
(72) Entry for Karata Ablative suffix:
-gal ⇔ </gal/, RouteP
Route SourceP
Source GoalP
Goal
>
168 SPELL-OUT DRIVEN MOVEMENT 7.5
(73) SourceP2
PlaceP2
AxPartP2
DP AxPartP1⇒-t
˙
‘
AxPart tDP
PlaceP1⇒-a
Place tAxPartP2
SourceP1⇒gal
Source GoalP2
tPlaceP2
GoalP1
Goal tPlaceP2
In this section, I take up the question whether lexical items contain information
only about categories, or whether segments matter too for
matching purposes. On the segment-matching view, the lexical entries
contain information about the exact geometry of the structure they can
replace, including the number of segments a category consists of. On the
less rigid category-matching view, only categories need to be listed in the
lexical entries.
7.5.1 Category matching
Let us first suppose that vocabulary items are specified only for what
heads and maximal projections they can lexicalize and do not store information
about how many segments a category consists of.7
Hence,
the shape of the abstract lexical entries that I use to illustrate various
derivations will be as follows:
(74) a. a ⇔ </a/, A>
b. b ⇔ </b/, BP
B
>
c. c ⇔ </c/, CP
C BP
B
>
7
To conform with this assumption, the matching of items should be stated only
over categories, that is, an item will match the syntactic structure if it contains a
superset of the categories listed in the syntactic structure, in the same geometrical
configuration.
7.5 SEGMENT VS CATEGORY MATCHING 169
Exploring again a hypothetical abstract derivation, let us start the first
cycle by the Merge of B and A. The structure is sent to the lexicon for
Spell-out and the lexicon produces the items a as a match for node A,
and item b as a match for node BP. The node A is then marked for
extraction in the next cycle as a result of the matching of b to BP.
(75) BP⇒b
B A⇒a
The next cycle starts with syntax first moving the nodes that are marked
for extraction, that is A to BP2, and then merging C.
(76) CP
C BP2
a⇐A BP1⇒b
B tA
The lowest unmatched node in this new cycle is BP2 and it is therefore
the first node to be inspected for lexicalization. There are two lexical
entries which match: b and c. The one which wins is b, since it contains
no superfluous features. As a result, node A is marked for extraction to
a position just above BP2. Finally, node CP is spelled out by c, and this
time A is marked for extraction to a node just above CP1. Note that the
BP2 node in (76) is spelled out by b at an intermediate stage, before the
insertion of c at CP overrides it. This is marked by putting parentheses
around the entry b at the BP2 node.
(77) CP2
a⇐A CP1⇒c
C BP2⇒(b)
tA BP1
B tA
The derivation in (77) contains an interesting case of “double evacuation”
which node A has to undergo. In the second cycle of the derivation, A
is marked for extraction twice: once by b matching BP2 and once by
170 SPELL-OUT DRIVEN MOVEMENT 7.5
c matching CP. The question is how this double movement happens in
cycle 3 — does A first adjoin to BP2 and then to CP1, or does it move
in one fell swoop to CP1?
The answer depends on the order in which syntax executes the Spellout
triggered movements once the matching process is accomplished.
Since lexicalization proceeds bottom-up, the node A is first marked for
extraction to BP3 and then marked once again for extraction to CP2. If
syntax performs the movements starting from the first one chronologically,
then A will first move to BP3, and then to CP2. If syntax, on the
contrary, starts with the last one, then A will move to CP2 and, assuming
that lowering is not possible, the movement to BP3 will not be performed
at all.
Computationally it is more economical to take a shortcut and move
A straight to CP2 — this is just one movement instead of two and,
in addition, it obviates the creation of yet another segment in the BP
category.
(78) CP2
a⇐A CP1⇒c
C BP2⇒(b)
tA BP1
B tA
Furthermore, such an assumption is confirmed by the empirical case
of the Finnish Ablative morpheme -tA, which I discuss towards the end
of Section 7.6. Therefore, I adopt the idea that, whenever more than one
movement is necessitated by the need to spell out the structure, syntax
performs them starting from the last one.
The technical implementation of this idea depends on whether we
have one node which has been marked for extraction more than once (as
in the abstract scenario above), or there are two or more distinct nodes
marked for extraction (as in the Finnish scenario presented in the next
section).
In the first case, when one node is marked for extraction twice or
more, I will assume that the later marking cancels out the earlier one.
More precisely, if node A is marked as A<BP3> at the stage when b is
matched to BP2, and then A is marked as A<CP2> when c is matched to
CP, the index <CP2> simply replaces the index <BP3>. As a result, syntax
7.5 SEGMENT VS CATEGORY MATCHING 171
“sees” only the index <CP2> and consequently moves A straight to CP2.
In the second case, when multiple nodes are marked for extraction
(the case of Finnish), the reversed chronology of movements can be
achieved if we assume that syntax proceeds top-down when inspecting
which nodes are to be evacuated. In a bottom-up lexicalization system,
the higher nodes in the tree will be marked for extraction later than
the lower ones. If syntax then evacuates the nodes working from the
top of the tree downwards, the node that was marked for extraction last
will be moved first, and so on. The result is that the movements will be
performed in an order opposite to the order in which they were triggered.
7.5.2 Segment matching
Let us now suppose that segments are included in the computation of
matching and assume the following abstract entries:
(79) a. a ⇔ </a/, A>
b. b ⇔ </b/, BP
B
>
c. c ⇔ </c/, CP
C BP2
BP1
B
>
Starting again with the Merge of A and B, the structure is shipped to
the lexicon for the regular check. The entry a matches node A, b is a
match for node BP and A is marked for evacuation in the next cycle.
(80) BP⇒b
B A⇒a
In the next cycle, first the movement of A takes place, followed by the
Merge of C.
172 SPELL-OUT DRIVEN MOVEMENT 7.5
(81) CP
C BP2
a⇐A BP1⇒b
B tA
Node BP2 is now the target for lexicalization and there is just one lexical
entry which matches — the entry c. It triggers the evacuation of A.
Finally, c lexicalizes the CP node triggering a second evacuation of A to
CP2.
(82) CP2
a⇐A CP1⇒c
C BP2⇒(c)
tA BP1
B tA
Note that, under segment-matching, all Spell-out triggered movements
are encoded in the lexical entries. For instance, the entry c contains
the landing sites for all the movements that A has to undergo prior to
adjoining to the root node, hence the two-segmental BP category in its
specification.
7.5.3 Segment versus category matching
The abstract structures examined in the preceding two sections make different
predictions under the assumed lexical entries. More precisely, the
BP2 node is spelled out by the “bigger” entry c under segment-matching
and by the “smaller” entry b under category-matching. I illustrate this
in the diagram in (83), which conflates the trees in (77) and (82).
7.5 SEGMENT VS CATEGORY MATCHING 173
(83) CP2
a⇐A CP1⇒c
C BP2⇒(c or b)
tA BP1
B tA
This difference does not surface because the entry inserted at BP2 is
overriden by the insertion of c at CP1 in both cases. It is, however, possible
to come up with a scenario where the difference comes to light, for
instance in the case of the Finnish Adessive expression, which I discuss
in Section 7.6. The Finnish Allative shows that the empirically correct
prediction is that BP2 is spelled out by the smaller entry b. As a consequence,
in order to make segment-matching deliver the right results, it
becomes necessary to include one more segment of the BP category in
the entry for b in (79b). In what follows I present this scenario in detail.
The point I would like to make is that both views can lead to empirically
correct results, provided we assume suitable lexical entries. In the absence
of empirical evidence, the choice between category-matching and
segment-matching then needs to be made on the basis of other considerations.
I then opt for category-matching, which is the view involving
less complex lexical entries,
Suppose that, instead of the head C, we merge the head D to the
structure in (81) (=(76)) and assume the entry in (85).
(84) DP
D BP2
a⇐A BP1⇒b
B tA
(85) d ⇔ </d/, DP
D
>
Keeping the entries a and b, assumed in (74), under the category-matching
view, lexicalization first targets BP2 and chooses b for insertion, marking
A for extraction to BP3. Then node DP is spelled out by d, resulting in
174 SPELL-OUT DRIVEN MOVEMENT 7.5
the evacuation of BP2. Syntax then first performs the movement of BP2
to DP2.
(86) DP2
BP2⇒b
a⇐A BP1
B tA
DP1⇒d
D tBP2
Then A is adjoined to BP2. Note that this operation involves an adjunction
within a complex specifier, a movement which is not generally
assumed to be available. There is however nothing in the Spell-out model
developed here which would prohibit it, since I assume, following Chomsky
(1995), that the Extension Condition does not apply in the cases of
adjunction.
(87) DP2
BP3
a⇐A BP2⇒b
tA BP1
B tA
DP1⇒d
D tBP2
Note that the node BP2 is spelled out by b, which wins over c by the
Elsewhere Principle, thus leading to a pronunciation of entire structure
as a+b+d.
Under the segment-matching view and with the entries in (79), lexicalization
again first targets the BP2 node, this time choosing c for spelling
it out as the only matching entry.
7.6 SEGMENT VS CATEGORY MATCHING 175
(88) DP2
BP3
a⇐A BP2⇒c
tA BP1
B tA
DP1⇒d
D tBP 2
The structure will therefore be pronounced as a+c+d. This is an empirically
wrong result, as shown by the Finnish data discussed in the next
section. Therefore, as already announced in the introduction to this subsection,
we need to assume an entry b with a two-segmental BP category
in order to make b spell out BP2 under the segment-matching view, see
(89).
(89) b ⇔ </b/, BP2
BP1
B
>
This entry will be the one used for lexicalizing the BP2 node, as it contains
all the segments which the BP category consists of, and it is also the entry
preferred over c by virtue of being a better match.
In sum, segment-matching and category-matching can both deliver
the right results and the issue of which one is to be adopted is reduced
more or less to the question of where we want to have a greater complexity:
in the shape of lexical entries (as with segment-matching, where
lexical entries contain multi-segmental categories) or in the computation
of derivation (as with category-matching, where the burden falls on computation
by necessitating quite a few evacuation movements, as I show
in the next section).
My decision is to assume simpler lexical entries and a computation
involving lots of movements. In this, I am guided by the common conception
in the field that storage space is costly and computation is cheap.
Thus, I adopt the category-matching view and proceed to the next section,
where I present a detailed derivation of the Finnish Locative, Goal
and Source phrases.
176 SPELL-OUT DRIVEN MOVEMENT 7.6
7.6 Finnish - The Grand Finale
Finnish is a very good language to finish with, not only because it is
homophonous to the verb, but also because its spatial system contains
wonderful twists and turns. It has a revealing case of allomorphy in
one of the two spatial series, the portmanteau morpheme -(h)Vn, which
I have already mentioned in the context of timing of Spell-out driven
movement. Furthermore, Finnish provides us with a rare scenario of
double evacuation movement, which, as mentioned in Section 7.5, suggests
that the execution of Spell-out driven movements happens in an
order opposite to the order of their triggering. Finally, Finnish features
a valuable empirical counterpart to the abstract structures in (87) and
(88), showing that the former one reflects the correct lexicalization, that
is, that the “smaller” entry indeed lexicalizes the counterpart of the BP2
node in Finnish.
Before running a detailed derivation and lexicalization of all Finnish
spatial expressions, I will summarize the principles and assumptions discussed
in this chapter.
(90) a. The linearization of syntactic structure is governed by the
LCA (Kayne 1994).
b. The shape of a lexical entry can trigger movement of a syntactic
constituent such that the right configuration for insertion
is created (Spell-out triggered movement) (Starke
2011).
c. The moved constituent must contain the head noun (Cinque
2005) .
d. The moved constituent adjoins right above the node where
insertion takes place (Shortest Move).
e. The Spell-out triggered movement in one cycle takes place
in the next cycle.
f. The order in which syntax performs the Spell-out triggered
movements is opposite to the order in which they are trig-
gered.
g. Lexical entries store information only about heads and phrases,
7.6 FINNISH - THE GRAND FINALE 177
but not about segments (category-matching).
The assumptions listed under (90d-f) were made to a great extent due
to the empirical evidence provided by the Finnish spatial morphology.
In this chapter, I show how they deliver the right results and why the
alternative assumptions do not.
The Finnish spatial system is presented in Table 7.1 and illustrated
on the noun talo ‘house.’
Series Location Goal Source
in -s -ssa (talo-ssa) -(h)Vn (talo-on) -sta (talo-sta)
on -l -lla (talo-lla) -lle (talo-lle) -lta (talo-lta)
Table 7.1: Spatial case system in Finnish (Sulkala and Karjalainen 1992)
The spatial case morphemes in Finnish can in fact be seen as compositional
with the important exception of -(h)Vn. Comrie (1999) suggests
that Finnish has two “series markers:” -s ‘in’ and -l ‘on.’ These series
markers combine with the Locative ending -CA to express Location
(where C copies the preceding consonant and the vowel A is subject to
vowel harmony). To express Goal, the marker -Ce attaches to the series
marker, and to express Source, the series marker combines with the suffix
-tA. Table 7.2 presents the decomposition of the Finnish spatial cases.
Series Location Goal Source
in -s -s-CA -(h)Vn -s-tA
on -l -l-CA -l-Ce -l-tA
Table 7.2: Decomposition of spatial cases in Finnish
The morphemes in Table 7.2 can be mapped to the syntactic structure
of spatial expressions in the following way:
• The series markers -l and -s lexicalize the AxPart head – (91a-b, d).
• The locative morpheme -CA lexicalizes the Place head – (91a).
• The morpheme -Ce lexicalizes Goal and Place – (91b).
178 SPELL-OUT DRIVEN MOVEMENT 7.6
• The Illative morpheme -(h)Vn lexicalizes AxPart, Place, and Goal
– (91c).
• The morpheme -tA lexicalizes Source, Goal, and Place – (91d).









(91) a. Place AxPart DP
-CA -l/-s
b. Goal Place AxPart DP
-Ce -l
c. Goal Place AxPart DP
-(h)Vn
d. Source Goal Place AxPart DP
-tA -l/-s
I therefore assume the following entries for the Finnish spatial morphemes,
adhering to the category-matching view:
(92) Series marker -l:
-l ⇔ </l/, AxPartP
AxPart
, on>
(93) Series marker -s:
-s ⇔ </s/, AxPartP
AxPart
, in>
(94) Locative marker -CA:
-CA ⇔ </CA/, PlaceP
Place
>
7.6 FINNISH - THE GRAND FINALE 179
(95) Goal marker -Ce:
-Ce ⇔ </Ce/, GoalP
Goal PlaceP
Place
>
(96) Source marker -tA:
-tA ⇔ </tA/, SourceP
Source GoalP
Goal PlaceP
Place
>
(97) Illative marker -(h)Vn:
-(h)Vn ⇔ </(h)Vn/, GoalP
AxPartP
AxPart
GoalP
Goal PlaceP
Place
, in>
Let us first examine the derivation of the “biggest” Source expression in
the on-series, which involves the lexical entries -l, -CA, -Ce, and -tA.8
The first step is the Merge of AxPart and DP.
(98) AxPartP
AxPart DP
The lexicon is consulted and the item -l in (92) is chosen for insertion.
This triggers a movement of DP to AxPartP2 as the first step in the next
cycle.
(99) AxPartP2
DP AxPartP1⇒-l
AxPart tDP
8
I assume that -s and -(h)Vn are not considered by Spell-out for the Adessive, Allative
and Ablative expressions, because they are associated with a different conceptual
content, namely, the notion of interior.
180 SPELL-OUT DRIVEN MOVEMENT 7.6
Afterwards, the Place head is merged and the structure is shipped to the
lexicon for the search for matching items.
(100) PlaceP
Place AxPartP2
DP AxPartP1⇒-l
AxPart tDP
The first node in the new cycle for which the lexicalization procedure
searches for a match in the lexicon is AxPartP2 (assuming that the DP
has been spelled out in a previous cycle and the system remembers the
outcome). The series marker -l is chosen as a match and consequently DP
is marked for extraction to AxPartP3 in the next cycle. The second node
to be lexicalized is Place. The Locative entry in (94) and the Goal entry
in (95) match it by the Superset Principle and the entry -CA is chosen by
virtue of having fewer superfluous features. Next, PlaceP is inspected for
lexicalization. The entries -CA and -Ce match this node, too. Again -CA
wins the competition and triggers an evacuation of AxPartP2. Hence the
next cycle begins with two movements: AxPartP2 adjoins to PlaceP1 and
then DP adjoins to AxPartP2 within the specifier of PlaceP. As already
discussed below the example in (86), I assume the movement of DP to
AxPartP3 to be available, as there is nothing in the system of assumptions
that would exlude it.
(101) PlaceP2
AxPartP3
DP AxPartP2⇒-l
tDP AxPartP1
AxPart tDP
PlaceP1⇒-CA
Place tAxPartP2
Next, the Goal head is merged.
7.6 FINNISH - THE GRAND FINALE 181
(102) GoalP
Goal PlaceP2
AxPartP3
DP AxPartP2⇒-l
tDP AxPartP1
AxPart tDP
PlaceP1⇒-CA
Place tAxPartP2
The first node targeted by lexicalization is AxPartP3. This is so because
this node is created as the result of the Spell-out triggered movement
which has taken place in the current cycle. Thus, AxPartP3 has not
been inspected for lexicalization previously. The series marker -l matches
AxPartP3, triggering adjunction to AxPartP4, i.e., to a position right
above the node where insertion takes place. Then, PlaceP2 is inspected
for matching items. The Locative marker -CA and the Goal marker -Ce
are chosen as matching entries and -CA wins by the Elsewhere Principle.
As a result, the AxPartP3 node adjoined to it is marked for extraction
to a position right above PlaceP2. Then the Goal head is targeted by
lexicalization. There are two matching items: the Goal marker -Ce in
(95), and the Source marker -tA in (96). The winner is -Ce by the Elsewhere
Principle. Finally, the phrasal node GoalP is targeted and -Ce is
chosen over -tA again. Subsequently, AxPartP3 is marked for extraction
to GoalP2. The insertion of -Ce at GoalP1 overrides the Locative marker
-CA inserted at PlaceP2.
182 SPELL-OUT DRIVEN MOVEMENT 7.6
(103) GoalP2
AxPartP4
DP AxPartP3⇒-l
tDP AxPartP2
tDP AxPartP1
AxPart tDP
GoalP1⇒-Ce
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP2
Note that the PlaceP2 node is predicted to be “temporarily” lexicalized
by -CA, before the insertion of -Ce at GoalP1 takes place. This
lexicalization of PlaceP2 can in fact be made visible, if this node becomes
embedded under a different projection, for instance a verbal head.
(104) VP
V PlaceP2
AxPartP3
DP AxPartP2⇒-l
tDP AxPartP1
AxPart tDP
PlaceP1⇒-CA
Place tAxPartP2
Suppose, for the sake of the argument, that there is a verbal entry with
the shape in (105).
(105) verb ⇔ < VP
V
>
The lexicalization of the structure will proceed in the following way (to
keep the discussion simpler, I will leave aside the step of the lexicalization
7.6 FINNISH - THE GRAND FINALE 183
dealing with the AxPartP3 node). As already discussed in the paragraph
preceding (104), the PlaceP2 node is matched by both -CA and -Ce
and the former entry is chosen by the Elsewhere Principle, triggering
evacuation of AxPartP3 to PlaceP3. The verbal entry in (105) matches
the VP node and triggers evacuation of the entire PlaceP2.
(106) VP
PlaceP3
AxPartP3
DP AxPartP2⇒-l
tDP AxPartP1
AxPart tDP
PlaceP2⇒-CA
tAxPartP3
PlaceP1
Place tAxPartP2
VP⇒verb
V PlaceP2
This time, due to the verbal material dominating it, -CA will not be
overridden by another morpheme (assuming that the Finnish verbs do
not lexicalize heads from the spatial domain, but see Chapter 8 for a
discussion of languages where this happens).
The structure in (106) is the previously advertised real language scenario
corresponding to the abstract structures in (87) and (88) discussed
in Section 7.5 and repeated below.
(107) DP2
BP3
a⇐A BP2⇒b or c
tA BP1
B tA
DP1⇒d
D tBP2
184 SPELL-OUT DRIVEN MOVEMENT 7.6
The question raised in connection to this diagram was what lexicalized
the BP2 node (corresponding to the PlaceP2 node in (106)): the smaller
entry b or the bigger entry c? Assuming that the smaller Locative marker
-CA is b in (74b) and the bigger Goal marker -Ce is c in (74c), Finnish
shows that the correct answer is b.
Back to the derivation of the Finnish Ablative structure, the next
cycle begins with two Spell-out driven movements: AxPartP3 to GoalP2
and DP to AxPartP4, as shown in the diagram in (103). The adjunction
of AxPartP3 to a node right above PlaceP2, which is supposed to
take place after moving AxPartP3 to GoalP2, does not happen, as this
instantiates lowering from GoalP2.
After these movements take place, the next head, Source, is merged,
see (108).
(108) SourceP
Source GoalP2
AxPartP4
DP AxPartP3⇒-l
tDP AxPartP2
tDP AxPartP1
AxPart tDP
GoalP1⇒-Ce
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
The nodes to be lexicalized are the following. First, AxPartP4 is
spelled out by -l, marking the DP to extract to AxPartP5. Then GoalP2,
matched by -Ce and by the Source marker -tA, is lexicalized by -Ce triggering
movement of AxPartP4 to GoalP3. The Source head is lexicalized
by -tA and, finally, the SourceP node is spelled out by -tA, triggering
movement of AxPartP4 to SourceP2 and overriding -tA inserted at the
Source head and -Ce inserted at GoalP2, see (109).
7.6 FINNISH - THE GRAND FINALE 185
(109) SourceP2
AxPartP5
DP AxPartP4⇒-l
tDP AxPartP3
tDP AxPartP2
tDP AxPartP1
AxPart tDP
SourceP1⇒-tA
Source GoalP2
tAxPartP4
GoalP1
Goal PlaceP2
tAxPartP3
PlaceP
Place tAxPartP3
With this the derivation of the Source expression in the on-series is
completed.
Let us now turn to the lexicalization of the Source case in the inseries.
The derivation starts again with the Merge of DP and AxPart.
There are two items which contain the category AxPart and which are
associated with the concept of interiority: -s and -(h)Vn. The two entries
therefore compete and -s wins by the Elsewhere Principle and spells out
AxPartP1, triggering movement of DP.
(110) AxPartP2
DP AxPartP1⇒-s
AxPart tDP
The derivation then proceeds in a way parallel to the one described for
the on-series, with the only difference that, -(h)Vn also competes with
-CA and -Ce for insertion at Place, PlaceP1, PlaceP2, Goal and GoalP1.
In each case, it loses because it has more superfluous features.
The situation becomes different once we merge the Source head and
create the GoalP2 node by the evacuation movements triggered in the
preceding cycle, see (111).
186 SPELL-OUT DRIVEN MOVEMENT 7.6
(111) SourceP
Source GoalP2
AxPartP4
DP AxPartP3⇒-s
tDP AxPartP2
tDP AxPartP1
AxPart tDP
GoalP1⇒-Ce
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
The first node to be lexicalized is AxPartP4, spelled out by -s thus
triggering evacuation of DP to AxPartP5. The next node is GoalP2.
The entry -(h)Vn is an almost perfect match — it spells out the entire
structure without the DP. Interestingly, the Goal marker -Ce can
also be inserted at GoalP2 (triggering evacuation of AxPartP4) and thus
competes with -(h)Vn. Since there is a tie (both entries leave zero unused
features), the decision of which one to be used has to be regulated
by some other principle than the Elsewhere Principle. A comparison
between the two possible lexicalizations of the GoalP2 shows that the
Spell-out by -h(V)n leads to an expression involving fewer morphemes
(-h(V)n) than the Spell-out by -Ce (-s and -Ce). This result can be captures
if we assume a principle enforcing the use of as few lexical entries
as possible for the Spell-out of a given syntactic structure. A principle to
this effect is the Economy of Derivation requirement of Emonds (1994)
and the Minimize Exponence principle of Siddiqi (2006). I present here
the latter.
(112) Minimize exponence (Siddiqi 2006:14):
The most economical derivation will be the one that maximally
realizes all the formal features of the derivation with the fewest
morphemes.
The adoption of this principle delivers the right result. Since -(h)Vn
realizes more features than -Ce, -(h)Vn is chosen to spell out GoalP2,
thus triggering an evacuation of the DP to a position right above GoalP2.
7.6 FINNISH - THE GRAND FINALE 187
This is however an intermediate stage of the lexicalization, since there
are two more nodes to be spelled out — Source and SourceP. The Spellout
of SourceP by -tA triggers movement of the entire AxPart4 to the
root node. This is the last movement triggered by Spell-out, hence the
first one to be performed by syntax. Moving AxPartP4 to SourceP2
will raise also the DP contained in it, hence the movement of DP to
a position right above GoalP2 triggered by -h(V)n’s matching of GoalP2
will not take place, because it is a downward movement. Finally, the first
movement triggered by the Spell-out procedure in this cycle, namely the
adjunction of DP to AxPartP4 will take place and the resulting structure
will be the one in (113).
(113) SourceP2
AxPartP5
DP AxPartP4⇒-s
tDP AxPartP3
tDP AxPartP2
tDP AxPartP1
AxPart tDP
SourceP1⇒-tA
Source GoalP2
tAxPartP4
GoalP1
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
Importantly, the Finnish Ablative expression can be derived only if we
assume that the order in which syntax performs the Spell-out triggered
movements is opposite to the order in which they are triggered. Consider
what would happen if we assumed the opposite: as lexicalization
proceeds bottom-up in the tree, first DP will be marked for extraction to
AxPartP5, then the same DP will be marked for extraction to GoalP3,
crucially escaping from AxPartP5, and finally AxPartP5 will be marked
for extraction to SourceP2. If the movements take place is the same order,
the last movement of AxPartP5 will displace a constituent which
does not contain the DP, contrary to the assumption in (66).
188 SPELL-OUT DRIVEN MOVEMENT 7.6
(114) SourceP2
AxPartP5
tDP AxPartP4⇒-s
tDP AxPartP3
tDP AxPartP2
tDP AxPartP1
AxPart tDP
SourceP1
Source GoalP3
DP GoalP2
tAxPartP5
GoalP1
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
That this violation of the restriction on movement leads to incorrect
results is clear from the wrong linearization of the nodes in (114): the
AxPartP category precedes the DP, because the former asymmetrically
c-commands the latter. Even more problematic, the entry -tA in fact
does not match the SourceP1 node anymore, due to the DP placed in
the specifier of GoalP. As a consequence, -tA cannot be inserted. Note
that the DP-to-GoalP3 movement takes place in cycle X+1, after -tA was
matched with SourceP1 in cycle X. SourceP1 is therefore inaccessible for
lexicalization anymore and the occurred mismatch cannot be repaired.
In conclusion, as mentioned in the final paragraph of Section 7.5.1,
the lexicalization of the Ablative expression in the Finnish in-series can
be successfully captured by the system of assumptions in (90) only if we
assume also (90f).
Let us now focus on the GoalP2 node in the structure in (113). Just
like in the case of the Source expression in the on-series, the entry matching
the GoalP2 node does not surface once all the morphemes are inserted
into the nodes of the syntactic structure, because it is overridden by -tA
at the SourceP1 node. Still, the lexicalization of GoalP2 by -(h)Vn can
be made visible if we use the same trick as in (106) and embed it under
a verbal head.
7.6 FINNISH - THE GRAND FINALE 189
(115) VP
V GoalP2
AxPartP4
DP AxPartP3⇒-s
tDP AxPartP2
tDP AxPartP1
AxPart tDP
GoalP1⇒-Ce
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
(116) verb ⇔ < VP
V
>
Assuming again a verbal entry which triggers movement (see (116)), the
lexicalization procedure will first target AxPartP4, triggering a movement
of DP to AxPartP5. Then it will inspect the GoalP2 node, where the
entry -(h)Vn will trigger a movement of DP to GoalP3. And, finally, the
Spell-out of VP by the verb will displace the entire GoalP2 phrase. In
syntax, first GoalP2 will raise to VP, then DP to GoalP3 and the first
movement DP-to-AxPartP5 will not take place, since it is downwards.
190 SPELL-OUT DRIVEN MOVEMENT 7.7
(117) VP
GoalP3
DP GoalP2⇒-(h)Vn
AxPartP4
tDP AxPartP3
tDP AxPartP2
tDP AxPartP1
AxPart tDP
GoalP1
Goal PlaceP2
tAxPartP3
PlaceP1
Place tAxPartP
VP⇒verb
V tGoalP2
In such case, -(h)Vn will lexicalize GoalP2 and will not be overridden
by the lexical material inserted at the higher verbal node. As a result,
the lexicalization of the GoalP2 node derived by the Spell-out driven
movements of AxPartP3 becomes visible and lends support to the idea
that such nodes do indeed get lexicalized. This, in its turn, suggests that
the cycle scheme presented in (33b) during the discussion of the timing
of Spell-out driven movement in Section 7.4.1, where a cycle begins with
a step of Internal Merge gives the correct results empirically.
7.7 Conclusion
The main goal of this chapter was to articulate in detail the workings of
the Spell-out system described in Chapter 6 and test it against the empirical
domain of the spatial expressions in the languages Karata, Uzbek
and Finnish. I adopted the idea originally proposed by Starke (2011)
that evacuation movements can be triggered in order for the lexicalization
of a given node to take place. This allowed me to account for the
linearization of the elements making up the spatial expressions in the
abovementioned languages, all of which feature suffixal markers.
This, of course, raises the question of how the system deals with
7.7 CONCLUSION 191
prepositional languages and, in particular, with spatial prepositions that
lexicalize more than one head in the syntactic structure. A solution to
this question has been offered by Caha (2010b) who proposes that the
entries for prepositions lack maximal projections. Consequently, prepositions
cannot lexicalize phrasal nodes and therefore do not induce an
evacuation of the DP, which would result in a postposition. As an example,
take the Persian Goal preposition be in (118), and assume that it
has the entry in (119).
(118) be
to
mædræsse
school
‘to (the) school’
(119) be ⇔ </be/, Goal
Place Goal
>
Caha suggests that instead of forcing a phrasal movement of the DP,
the entry in (119) leads to a structure which is traditionally taken to be
the result of head movement. The preposition be is then inserted at the
complex Goal head, see (120).
(120) GoalP
be⇐Goal
Place Goal
PlaceP
tPlace DP
Thus, the difference between prepositional and postpositional languages
lies not in the setting of a given parameter, but in the variation
of the lexical inventory.
The Phrasal Spell-out system developed in this part of the thesis and
the decomposed Path structure argued for in the first part make certain
predictions as to the possible make-up of spatial expressions in general
and the types of syncretisms one expects to find cross-linguistically.
This is the topic of the next two chapters, where I investigate a number
of languages and show how attested and non-attested syncretisms and
lexicalization patterns can be accounted for by the system.
192 SPELL-OUT DRIVEN MOVEMENT 7.7
Part III
Syncretisms and partitioning of
the syntactic structure
193
Chapter 8
Partitioning of the structure
and spurious syncretisms
8.1 Introduction
In the first part of this thesis, I argued for a particular decomposition
of the Path projection into five distinct heads: Goal < Source < Route
< {Scale, Bound[ed]}. I suggested that the various path expressions
correspond to different syntactic structures involving different number
of these heads and proposed a certain semantic function for each head.
In the second part of the thesis, adopting the Nanosyntax framework,
I explored how these structures are lexicalized in a Phrasal Spell-out
system. I focussed on the lexicalization of the spatial expressions in the
languages Karata, Uzbek and Finnish, running a detailed step-by-step
lexicalization for each expression.
Because of the rather fine-grained structure of path expressions, especially
of the “bigger” Route and Source paths, combined with the possibility
that lexical items lexicalize more than one syntactic head, the
system predicts that there should be various ways in which a given path
expression can be spelled out across languages. For instance, there exist
many ways in which a Route expression can be “partitioned” depending
on how many lexical items (morphemes) are needed to construct it and
where exactly the cut between those lexical items is.
The various types of partitioning of the structure is the topic of this
last part of the thesis. This time, I adopt a broader perspective, as enforced
by the subject, and instead of focusing on the spatial system of one
language, I take a cross-linguistic view. In investigating the lexicalization
patterns of paths cross-linguistically, I explore two topics: (i) what
195
196 PARTITIONING OF THE STRUCTURE 8.2
is the role of other categories in the Spell-out of the spatial structure
and (ii) what various types of syncretism patterns are predicted to exist
(and not exist) by the theory. I deal with the first question in Chapter
8, where I show that, in many languages, the structure underlying directional
expression is lexicalized by the joint collaboration of adpositional
and verbal elements, as also suggested in Son and Svenonius (2008). This
leads to a phenomenon which I label spurious synctretism to distinguish
it from “real” syncretism. In Chapter 9, I investigate the question of what
real syncretisms are expected to be possible and impossible and discuss
some cases of putative counterexamples.
8.2 Real and spurious syncretisms
In Chapter 6, Section 6.4, I discussed the Hindi Ablative marker -see
which syncretizes Route and Source paths, see (1), repeated from (44) in
Chapter 6.
(1) baccaa
child
kaar-ke
car-gen
saamne-see
front-abl
calaa.
walk.perf
(i) ‘The child walked via in front of the car.’ (Route)
(ii) ‘The child walked from in front of the car. (Source)’
As suggested there, the ambiguity of the Hindi -see is due to the fact
that it spells out two distinct structures — a SourceP and a RouteP —
which is made possible by the Superset Principle.
(2) a. Route path
RouteP ⇒ -see
Route SourceP
Source GoalP
Goal Place
Place ...
b. Source path
SourceP ⇒ -see
Source GoalP
Goal Place
Place ...
A closely related language, Persian, is also said to have that type of
ambiguous marker: the preposition æz translated as ‘from.’ The preposition
æz participates in both Route and Source path expressions. In
8.2 REAL AND SPURIOUS SYNCRETISMS 197
Route expressions, however, it is used only in combination with the
verbs gozæshtæn ‘to go/pass by’ or ræd shodæn ‘to pass by” (Mahootian
1997:166).
(3) a. Bæchche
child
æz
from
baq
garden
gozæsht.
pass.3sg
‘The child went via the garden.’
b. Bæchche
child
æz
from
pol
bridge
ræd
pass
shod.
became.3sg
‘The child passed by the bridge.’
When combined with any other motion verb, an æz-PP gives rise only to
a Source interpretation and never to a Route interpretation.
(4) Bæchche
child
æz
from
baq
garden
doid.
ran
‘The child ran from the garden’
*‘The child ran via the garden.’
It is obvious that the Persian facts are quite different from the Hindi facts
in the sense that, in Persian, the Route meaning of the preposition æz
requires a particular “Route-verb,” while, in Hindi, the Route meaning
of -see is available with a rather unrestricted set of manner of motion
verbs. More precisely, Persian does not allow a Route interpretation of
an æz-PP with any other verb than the two verbs in (3), while Hindi
allows a Route interpretation of -see with, for instance, verbs like walk
and fly (see the examples in (1) and (5), respectively).1
(5) ciRijaa
bird
Jhiil-ke
lake-gen
upar-see
above-abl
uRi.
flew
(i) ‘The bird flew via above the lake.’ (Route)
(ii) ‘The bird flew from above the lake.’ (Source) (fieldnotes)
1
Note that, in this respect, Hindi, which is commonly characterized as a verbframed
language in Talmy’s (2000) typology, does not behave like such, since it encodes
the path on the case ending (i.e., on the satellite) rather than on the verb
(see also Narasimhan 2008 for a discussion of the mixed properties of Hindi). Persian,
which in many respects behaves like a satellite-framed language (Feiz 2007), in
this particular case shows properties of a verb-framed language, as the Route path
meaning component is supplied by the verbs translating as pass. These mixed properties
of Hindi and Persian lend support to a more fine-grained typology of verb and
satellite-framed languages in the spirit of Beavers et al. (2010).
198 PARTITIONING OF THE STRUCTURE 8.2
The fact that an æz-expression in Persian has a Source interpretation
with any other verb but the two pass-verbs in (3) leads to the conclusion
that the Persian æz is not really ambiguous between Route and Source,
but expresses Source only. The question is then what makes the Route
meaning in (3) possible. As the Route interpretation is available only in
the presence of one of the two “Route verbs,” it is logical to hypothesize
that the verbs gozæshtæn ‘to pass’ and ræd shodæn ‘to pass by’ lexicalize
the bit of syntactic structure that is necessary for a Source path to
become a Route path. In other words, I suggest that gozæshtæn and ræd
shodæn spell out not only the verbal portion of the functional sequence,
subsumed here under the label VP, but also the Route head, thus leaving
the SourceP in the complement position to be lexicalized by the Source
preposition æz.2
(6) VP
SourceP
æz ⇐ Source
Source Goal
Goal Place
GoalP
tGoal PlaceP
tPlace DP
VP⇒gozæsht
V RouteP
Route tSourceP
In (6), the verb gozæshtæn lexicalizes a constituent containing both the
VP and the Route head. The preposition æz spells out the complex
Source head formed by the head-adjunction of the Place and Goal heads
to it, as proposed in Chapter 7.
Such an analysis captures the fact that Persian verbs that do not
belong to the set of “Route verbs” cannot express Route of motion with
an æz-PP. The reason lies in the Exhaustive Lexicalization Principle presented
in (33), Chapter 6, which states that each syntactic feature has to
be lexicalized. This is exemplified below by the manner of motion verb
doidæn ‘run,’ which, I propose, does not have the feature <Route>. This
2
For a fine-grained decomposition of the VP, see Ramchand (2008b) who proposes
that events contain three subevents (init, proc and res), each corresponding to a
distinct head in the verbal projection. In this chapter, I will collapse init, proc and
res under the label V, since the focus of this thesis is on Path expressions.
8.2 REAL AND SPURIOUS SYNCRETISMS 199
leads to a failure to lexicalize the Route head in (7) and, for that reason,
the Route reading for (4) is unavailable.
(7) VP
SourceP
æz ⇐ Source
Source Goal
Goal Place
GoalP
tGoal PlaceP
tPlace DP
VP⇒doid
V RouteP
Route tSourceP
To sum up, in Persian Route expressions, representatives of two categories
collaborate to lexicalize a given syntactic structure: because there
is no adposition that is specified in the lexicon as big enough to spell out
a Route phrase, a special Route verb lexicalizes the Route head, along
with the heads it lexicalizes in the verbal domain (cf. the analysis in
Son and Svenonius 2008). The structure below the Route head, i.e., the
Source phrase, is then lexicalized by the Source preposition æz, available
in Persian. On the face of it, it then appears that Persian æz syncretizes
a Route and a Source path, but in reality, it always only spells out a
Source structure and the Route=Source syncretism is spurious.
The distinction between real and spurious syncretisms is that real
syncretisms involve lexical items that are used to spell out two, or more,
distinct structures. For instance, as shown in (2), Hindi -see is used to
lexicalize a Route structure (involving four heads) and a Source structure
(involving three heads), thus leading to a real ambiguity. Spurious
syncretisms involve a lexical item that spells out one and the same structure
whithin two or more distinct structures. For instance, the Persian
preposition æz always spells out a Source structure. In the proposed decomposition
of paths, the Source structure is part of a Source expression
(in fact, identical to it) and also a proper part of a Route expression.
In the case of a Source expression the preposition æz itself spells out
the entire structure. In the case of a Route path, æz again spells out a
Source structure and the remaining Route head is lexicalized by a special
verb. Therefore, æz is not really ambiguous between Route and Source
— it always spells out a Source structure, and the Route meaning in
the case of Route paths comes from the special verbs’ ability to lexical-
200 PARTITIONING OF THE STRUCTURE 8.3
ize this head. Finally, languages that do not exhibit syncretisms have
lexical items which are tailor-made for each of the syntactic structures
corresponding to the different types of paths. I illustrate schematically
the three distinctions in (8)-(10), taking as an example the case of Route
and Source paths in Hindi, Persian and English.
(8) Hindi (real syncretism)
V Route Source Goal Place
Route path
verb see
V Source Goal Place
Source path
verb see






(9) Persian (spurious syncretism)
V Route Source Goal Place
Route path
verb æz
V Source Goal Place
Source path
verb æz









(10) English (no syncretism)
V Route Source Goal Place
Route path
verb via
V Source Goal Place
Source path
verb from






As can be seen from the schematization above, “real” syncretisms
involve a structural ambiguity (see (8)). Spuriously syncretic formatives
are, on the contrary, unambiguous. They always spell out the same
structure, and their multiple functions results from the fact that they
appear in different syntactic contexts (see (9)).
8.3 LEXICALIZATION PATTERNS 201
8.3 Lexicalization patterns
8.3.1 Lexicalizing Route paths
The model developed in this thesis predicts the existence of other ways
to partition Route paths. Consider the following sentences which involve
trajectories traversing a Ground, thus representing a type of Route paths.
(11) Finnish, Finnic, Sulkala and Karjalainen (1992)
Pojat
boys
joksevat
run.3pl
talo-n
house-gen
edi-tse.
front-prol
‘The boys are running across in front of the house.’
(12) Tabasaran, Daghestanian, Magometov (1965) (my glossing)
Izu
I
ult.určo
@unuza
jump.over
niri-ll-an.
river.erg-sup-abl
‘I jumped across the river.’
(13) Kayardild, Tangkic, Evans (1995)
Kamar-ra
stone-nom
ngudi-ja
throw-imp
katharr-ir
river-all
jirrka-an-kir!
north-from-all
‘Throw the stone from the north across the river!’
(14) Czech, Slavic (P. Caha, p.c.)
Kluci
boys
pro-běhli
via-ran
před
in.front.of
dom-em.
house-instr
‘The boys ran across in front of the house.’
Focusing on the different case marking that the languages in (11)-(14)
make use of in these Route expressions, we can observe a fairly big range
of choices. Finnish employs the Prolative case, which is a case designated
to the expression of Routes. Tabasaran makes use of the Ablative case,
which typically expresses Sources. Kayardild marks the Ground DP by
the Allative case — a case that otherwise denotes Goals. Finally, in
Czech we have a combination of a preposition and a Ground DP marked
by Instrumental. The same expression is also used to mark Location, as
shown in (15).
(15) Kluci
boys
stáli
stood
před
in.front.of
dom-em.
house-ins
‘The boys stood in front of the house.’ (P. Caha, p.c.)
The data is summarized in the following Table 8.1.
202 PARTITIONING OF THE STRUCTURE 8.3
Language
The expression The expression is
used for Route prototypically used for
Finnish Prolative DP Route paths
Tabasaran Ablative DP Source paths
Kayardild Allative DP Goal paths
Czech P+Instrumental DP Location3
Table 8.1: Cases used by languages to express a Route path “across”
The legitimate question is whether we are dealing here with real or
spurious syncretisms, that is, whether the Ablative in Tabasaran, for
instance, is really ambiguous between a Source and a Route path or
whether it can lexicalize only a Source structure, like the Persian æz,
leaving the Route head to be spelled out by some other lexical item.
Let us start with Czech, which employs a locative phrase in expressions
of Location and Route paths. We are looking for a clue whether
something other than the expression consisting of a preposition and an
Instrumental-DP brings about the Route reading, just like in Persian.
The data in (16) reveals that this is indeed the case.
(16) Czech, P. Caha, p.c.
a. Had
snake
pro-lezl
via-crawled
před
in.front.of
vchod-em.
entrance-instr
‘The snake crawled via in front of the entrance.’ (Route)
b. Had
snake
lezl
crawled
před
in.front.of
vchod-em.
entrance-instr
‘The snake crawled in front of the entrance.’ (Locative)
In (16), the a-example bears a Route meaning and consists of a proprefixed
verb combined with a locative phrase. The b-example has an
unprefixed verb combined with a locative phrase, and the only available
interpretation is the locative one. Thus, a logical conclusion is that prolexicalizes
the heads that are between the verb and the Place projection
in a Route structure, as depicted in the tree diagram below.
3
The Czech Instrumental case by itself, of course, prototypically marks instruments.
The point I would like to make here is that the combination of a spatial
preposition and an Instrumental-marked DP is used to express Location as well as in
expressions of Route paths.
8.3 LEXICALIZATION PATTERNS 203
(17) VP
V
pro- ⇐ Route
Route Source
Source Goal
V
lezl
RouteP
tRoute SourceP
tSource GoalP
tGoal PlaceP
před ⇐ Place DP
So, in Czech we have an instance of a spurious syncretism: although
we have the same combination P+DP-instr in expressions of Location
and Route path, it cannot express both notions. I suggest that P+DPinstr
is locative only and, in the case of Route paths, it needs the
support of a special prefix that lexicalizes the Goal, Source and Route
heads. The Czech prefix pro- thus lexicalizes a portion of the spatial
domain below the verb and raises to adjoin to the verb. This is in line
with analyses where Slavic lexical prefixes spell out prepositional material
in the complement of the verb, as in, for instance, Svenonius (2004) and
Asbury et al. (2007).
Let us now have a look at the other languages in the data set. I turn
to the Tabasaran Route=Source syncretism and take some more data
under investigation in order to see whether something other than the
Ablative case triggers the Route reading, that is, lexicalizes the Route
head.
Consider the data below.
(18) Tabasaran, Magometov (1965) (my glossing)
a. Izu
I
ult.určo
@unuza
jump.over
q@an-di-ll-an.
stone-erg-sup-abl
‘I jumped over the stone.’
b. Izu
I
q@irčo
@unuza
jump
q@an-di-ll-an.
stone-erg-sup-abl
‘I jumped off the stone.’
In this minimal pair, the a-example expresses a Route path, while the
b-example expresses a Source path. A comparison between the two reveals
that the two sentences differ with respect to the prefix attached to
204 PARTITIONING OF THE STRUCTURE 8.3
the verb. Specifically, the verb in the Route expression is prefixed by
ult.ur-, which I gloss as ‘via,’ while the verb in the Source expression is
prefixed by a different affix q@ir-, the meaning of which is not given in
Magometov’s grammar and is impossible to determine precisely due to
insufficient data. Crucially, the verb in the Route expression in (12) also
bears the prefix ult.ur-. It is then highly probable that the prefix ult.uris
the element that turns a Source expression into a Route expression by
lexicalizing the Route head. The tree diagram below represents the proposed
lexicalization of the example in (12), where the morpheme -ll spells
out the PlaceP node including the AxPart projection, triggering cyclic
movement of the DP within this structure. The Source morpheme -an
spells out the SourceP node including the Goal projection, thus triggering
cyclic movement of the PlaceP2 projection, Finally, the Route prefix
ult.ur- spells out the Route head and adjoins to the verb. (I have simplified
the structure by not depicting the movements of the DP to PlaceP3
and PlaceP4 triggered by the insertion of -ll into the highest PlaceP node
at the cycles when the Goal and Source heads are merged.)
(19) VP
V
ult.ur⇐Route V
RouteP
tRoute SourceP2
PlaceP2
DP PlaceP1⇒-ll
Place AxPartP2
tDP AxPartP1
AxPart tDP
SourceP⇒-an
Source GoalP2
tPlaceP2
GoalP
Goal tPlaceP2
As can be seen from the spell-out of the Route structure in (19),
the Tabasaran Route=Source syncretism is a spurious one, too. The
Tabasaran Ablative case marker cannot express a Route path. It can
lexicalize the structure only up to the Source head, and therefore, in
Route phrases, we need a special prefix to lexicalize the Route head.
8.3 LEXICALIZATION PATTERNS 205
Let us now have a look at Kayardild. Considering the fact that neither
in Czech nor in Tabasaran are we dealing with a real ambiguity, we can
ask whether the Kayardild Allative case in (13) really syncretizes Route
and Goal. It is plausible that the Allative lexicalizes Goal only and the
Route meaning comes from the verb. Consider the data in (20) from
Evans (1995:150).
(20) Kurrka-tha
take-imp
nga-ku-l-da
1-inc-pl-nom
natha-r
camp-all
nga-ku-lu-wan-jir!
1-inc-pl-poss-all
‘Let’s take (it) to our camp!’
If the Allative case were ambiguous between Route and Goal, then the
example in (20) would have a second interpretation ‘Let’s take it via our
camp!’ There is no indication of this interpretation being available in
Evans (1995). If the hypothesis about the spurious Goal=Route syncretism
of the Allative is correct, then the prediction is that there is no
Route interpretation of (20). Since I have no access to such data nor to
native speakers of Kayardild, the hypothesis remains to be checked. In
case it is correct, then the verb translated as throw in (13) lexicalizes the
Route and Source heads, which the Allative case fails to lexicalize.
To sum up, I discussed the lexicalization of Route paths in several
languages and showed how the syntactic structure of Routes can be sliced
up in different portions, each of which is lexicalized by a separate lexical
item. Leaving aside the AxPart head for the reason that none of the
languages discussed has a morpheme border between Place and AxPart,
the situation can be summarized like this:
(21) V Route Source Goal Place
Finnish
verb prol
Persian
verb æz
Tabasaran
verb ult.ur abl sup
Kayardild
verb all
Czech
verb pro p+instr





















206 PARTITIONING OF THE STRUCTURE 8.3
In Finnish, the Prolative marker covers the whole structure up to
the Route head. In Persian, two special “Route verbs” lexicalize the
Route head (together with the verbal phrase) and the remaining portion
of the structure is lexicalized by a Source preposition. In Tabasaran, the
Ablative marker spells out the structure from the Goal head to the Source
head, leaving the Route head to be lexicalized by the prefix ult.ur-. In
Kayardild, the Allative marker goes up to the Goal head, and the verb
has to lexicalize the Route and Source heads. Finally, in Czech, the
prefix pro- lexicalizes Route, Source, Goal and the rest of the structure
is spelled out as a locative expression.
The lexicalizations shown in (21) are just a few examples of the logically
possible lexicalizations of the structure. I will give here three more
examples. The first one comes from Slovak and has been already presented
in Chapter 4, Section 4.3. In Slovak, Route expressions contain a
special preposition that combines with a GoalP, see (22) repeated from
(10) in Chapter 4 (data from P. Caha, p.c.).
(22) Na
On
Forum
Forum
Roman-um
Romanum-acc
vstupujeme
enter.1pl
po-pod
po-under
oblúk-∅
arch-acc
Tita.
of.Tito
‘We entered Forum Romanum via under Tito’s arch.’
The following data, repeated from (11) in Chapter 4 shows that without
the preposition po, the PP has a Goal reading.
(23) Slamu
hay
dal
put.3sg
pod
under
stôl-∅.
table-acc
‘He put the hay under the table.’
The second example comes from Yukatek Maya. Yukatek Maya is a
“radically verb-framed” language, where the two spatial prepositions ich
‘in’ and ti’ ‘at’ express exclusively Locations and never encode paths
(Bohnemeyer and Stolz 2006).
(24) Le=kàaro=o’
det=cart=dist
ti’=yàan
prep=exist.3sg
ich
in
le=kàaha=o’
det=box=dist
‘The cart, it is in the box.’
This means that no spatial preposition lexicalizes heads higher than the
Place head, because, if it could do so, then, contrary to the facts, it would
be able to participate in path expressions in combination with manner of
motion verbs without the support of special “path-verbs.” In the Route
8.3 LEXICALIZATION PATTERNS 207
expression in (25), we have such a path-verb combined with a locative
expression, and I suggest that the verb lexicalizes the heads from Route
to Goal (data from Bohnemeyer in prep).
(25) Le=ríiyo=o’
det=river=dist
h-máan
prv-pass.3sg
ich
in
le=bàaye=o’.
det=valley=dist
‘The river, it passed through the valley.’
The third example comes from the language Tetun Dili, spoken in East
Timor. Like many other Austronesian languages, Tetun Dili makes use
of verb serialization for the expression of directed motion, combining
verbs that express manner of motion (walk, run) with verbs expressing
path (enter, exit). The example of a Route expression in (26) features
a manner of motion verb and a path-verb translating as cross and does
not involve any P element (data from Hajek 2006:244).
(26) ami
1pl.exc
lao
walk
esik
cross
ponti.
bridge
‘We walked across the bridge.’
I therefore suggest that, in addition to lexicalizing a verbal projection,
the path-verb esik ‘to cross’ lexicalizes all the heads in the Path domain
down to the DP (cf., the analysis of Svenonius and Son 2009, who propose
that the verb go lexicalizes a P node in Tetun Dili).
The lexicalizations in Slovak, Yukatek Maya and Tetun Dili can then
be represented as follows:
(27) V Route Source Goal Place
Slovak
verb po p+acc
Yukatek Maya
verb ich
Tetun Dili
verb















The stretch between V and Place can be carved up in eight other
ways depending on how many morpheme borders the Route expression
contains and where the “cuts” fall.4
This is schematized in Table 8.2,
where the vertical lines mark morpheme borders.
4
If we include the AxPart head in the picture, the number of remaining possible
partitionings rises to thirty two.
208 PARTITIONING OF THE STRUCTURE 8.3
V Route Source Goal Place
1 Tetun Dili
2 | Finnish
3 | Persian
4 | Kayardild
5 | Yukatek
6 | | Slovak
7 | | Czech
8 | |
9 | |
10 | |
11 | |
12 | | | Tabasaran
13 | | |
14 | | |
15 | | |
16 | | | |
Table 8.2: Possible partitionings of a Route structure
The first row represents a lexicalization pattern where the verb and
all the spatial heads in the P domain are spelled out by one lexical item.
This partitioning type is common among languages with verb serialization,
where directed motion is typically expressed by a combination of
two or more verbs, the last one of which is usually a path-verb. I suggest
that the path-verb covers the entire stretch from V to Place. The next
four rows represent a partitioning of the Route structure which involves
just one morpheme border. The rows from 6 to 11 represent partitionings
which involve two cuts. Such languages are harder to find and are
represented here by Czech and Slovak. Even more difficult to find are
languages which slice the structure into three pieces (type 12-15). The
only one I have come across so far is Tabasaran (type 12). I still have not
found a language making the maximal possible number of cuts (type 16).
Nevertheless, despite the lack of languages representing certain partitioning
types, there are at least eight possible partitionings that are predicted
to be possible and are attested. The fine-grained syntactic structure of
Route expressions argued for in this thesis provides us with a tool to
8.3 LEXICALIZATION PATTERNS 209
handle these diverse lexicalization patterns of Route expressions across
languages. The same strategy can be applied also to the other two types
of paths, which I analyze in the subsequent sections.
8.3.2 Lexicalizing Source paths
Let us now turn to the lexicalization of Source paths cross-linguistically.
There are in total eight logically possible ways to partition the syntactic
structure, as shown in Table 8.3, where I mark the border between morphemes
by a vertical line and again exclude the AxPart head from the
picture.
V Source Goal Place
1 Mandarin
2 | English
3 |
4 | Yukatek
5 | | Laz
6 | | Lezgian
7 | |
8 | | | Hua?
Table 8.3: Possible partitionings of a Source structure
The first lexicalization possibility is found in Mandarin Chinese (data
from Chen and Guo 2009:1751).
(28) W˘o
I
păo
run
ch¯u
exit
le
perf
chúfáng.
kitchen
‘I ran out of the kitchen.’
The Mandarin Source expression in (28) contains a manner of motion
verb păo ‘run’ combined with the path-verb ch¯u ‘exit’ and no adpositional
element. I therefore suggest that the path-verb ch¯u lexicalizes the whole
stretch from Source to Place, thus taking directly a DP as a complement.
The second lexicalization possibility is represented by English.
(29) He ran from the house.
210 PARTITIONING OF THE STRUCTURE 8.3
In (29), the verb run combines with a Source phrase where the Source,
Goal and Place heads are lexicalized by the preposition from (recall the
schematization in (10)).
I have not encountered an example of the third lexicalization pattern.
This would be a language where certain “Source-verbs” combine with
Goal complements to form Source expressions.
Yukatek Maya is a plausible candidate for a language that spells out
the Source structure according to the scenario in row 4, since its spatial
PPs express only Location, as claimed in Bohnemeyer and Stolz (2006).
Indeed, Source paths are encoded exclusively in the verb (Bohnemeyer
and Báez 2008). In the following example, the Source meaning is contributed
by the verb, which combines with a locative expression.
(30) e=kàaro=o’
det=cart=dist
h-hòok
prv-exit.3sg
ich
in
le=kàaha=o’.
det=box=dist
‘The cart, it exited (lit. in) the box.’
Consequently, the verb translated as ‘exit’ in Yukatek Maya spells out
the Source and Goal heads in the syntactic structure leaving the Place
head to be lexicalized by the locative preposition ich ‘in.’
An example of the fifth lexicalization pattern is found in the Kartvelian
language Laz. Laz has two spatial cases: the null-marked Locative with
the phonologically null exponent -∅, (31), and the Motative case, whose
ending is -şa. The latter is used both in Goal and Source expressions,
see (32) (data from Broschart and Dawuda 1999).
(31) Peteri
Peter
livadi-∅
garden-loc
on.
copula.3sg
‘Peter is in the garden.’
(32) a. Peteri
Peter
oxori-şa
home-mot
ulu-n.
go-3.sg.pres
‘Peter goes home.’
b. Peteri
Peter
oxori-şa
home-mot
mulu-n.
come-3sg.pres
‘Peter comes from home.’
The data in (32) arouses suspicion, as a comparison between the Goal
and the Source examples shows that in the latter there is a morpheme
m- prefixed to the verb. Thus, it is conceivable that the Motative in Laz
expresses Goal only and a Source expression needs an additional prefix.
In other words, the examples in (32) give grounds for an investigation of
the question whether the alleged Goal=Source syncretism in Laz in not
8.3 LEXICALIZATION PATTERNS 211
a spurious one. Indeed, according to the data elicited in Kutscher (2001;
2010), the only possible reading of an unprefixed verb taking a Motative
DP is one of Goal of motion, and never Source of motion.
(33) Oxori-şa
house-mot
ulu-n.
go-3sg.pres
‘S/he goes into the house.’
*‘S/he goes out of the house.’ (Kutscher 2010)
Kutscher (2010) attributes the Goal interpretation for the sentence in
(33) to the preference for Goal paths described in various studies (Lakusta
2005, Lakusta and Landau 2005, among others). The interpretation of
the sentence in (33) as Goal-directed rather than Source-directed, however,
appears not to be a matter of preference. As stated by Kutscher
(2010:15), native speakers “negate the possibility of interpreting the utterance
[in (33)] as expressing a movement away from the house” suggesting
that the Source reading is not simply dispreferred but rather
unavailable. This strongly suggests that the Motative case ending -şa
expresses Goal paths only, that is, it spells out the structure as high as
the GoalP, as shown in the diagrams below corresponding to the example
in (33). Given the null phonology of the Locative suffix, it is impossible
to decide on the basis of the available data whether the Motative suffix
is stacked on the Locative suffix, or is attached directly to the noun thus
“overriding” the null Locative morpheme. In (34), I assume that -şa covers
both the Place and Goal heads. Still, an entry for -şa specified for
just the Goal feature is possible, leading to a roll-up derivation type.
(34) VP
V
ulu
GoalP2
DP GoalP1⇒şa
Goal PlaceP2
tDP PlaceP1
Place tDP
The hypothesis that the Motative suffix lexicalizes the Goal head finds
further support in the fact that when the verb translating as go combines
with a Locative-marked DP, the Goal reading is ungrammatical (data
212 PARTITIONING OF THE STRUCTURE 8.3
from Kutscher 2001:170).
(35) a. Neˇkna-şa
door-mot
b-ulu-r.
1sg-go-1sg.pres
‘I go to the door’
b. *Neˇkna-Ø
door-loc
b-ulu-r.
1sg-go-1sg.pres
The reason is a violation of Exhaustive lexicalization, as the Goal head
remains unlexicalized.5
To recapitulate, the data suggests that the Motative case in Laz unambiguously
marks Goals. In the Source expression in (32b), repeated
in (36) below, the Source meaning is contributed by the prefix m- on the
verb. I propose that the prefix m- spells out the Source head in Laz.
Hence, this is an example of a partitioning of the structure according to
type 5 in Table 8.3.6
(36) Peteri
Peter
oxori-şa
home-mot
mulu-n.
come-pres
‘Peter comes from home.’
The tree diagram for the Laz data in (36) is as follows:
(37) VP
V
m-⇐Source V
SourceP
tSource GoalP2
DP GoalP1⇒-şa
Goal PlaceP2
tDP PlaceP1
Place tDP
5
Note that Laz has a null morpheme to mark Location (Broschart and Dawuda
1999, Kutscher 2001), but does not have a null morpheme to mark Goals. If it did,
then the example in (35) would have been grammatical.
6
If we assume that the entry for -şa is specified only for the Goal feature, then Laz
will have a the partitioning pattern presented in row 7 of Table 8.3.
8.3 LEXICALIZATION PATTERNS 213
The fifth lexicalization pattern can be found in the language Lezgian,
where the Source suffix -aj is stacked on top of the locative structure. I
illustrate this in (38), taking as an example the on-series. The example
in (38a) shows that the suffix -l expresses Location, the data in (38b)
exemplifies the addition of the Source ending -aj to the locative structure
in order to express Source of motion (data from Haspelmath 1993, my
glossing).
(38) a. Čur.a-l
pasture.erg-on
wad
five
jac
oxen
amiq’-na.
stay-past
‘Five oxen were still on the pasture.’
b. Nurali
Nurali
buba
father
balk’an.di-l-aj
horse.erg-on-source
ewič-na.
descend-past
‘Father Nurali got off the horse.’
The tree diagram corresponding to the example in (38b) is shown in (39).
(39) VP
SourceP2
PlaceP2
DP PlaceP1⇒-l
Place AxPartP2
tDP AxPartP1
AxPart tDP
SourceP⇒-aj
Source GoalP2
tPlaceP2
GoalP
Goal tPlaceP2
VP⇒ewič-na
V tSourceP2
Until now, I have not come across the lexicalization pattern presented in
row 7 of Table 8.3 (but see fn. 6 in the current chapter).
Finally, a probable candidate for the lexicalization presented in the
last row in Table 8.3 is the Papuan language Hua. According to Kibrik
(2002:49), Hua has two locative suffixes: vi’ ‘in’, and -ro’ ‘at.’ In Goal
expressions, the morpheme -ga is added to the locative suffixes deriving
vin-ga and ro-ga. In Source expressions, the morpheme -ri is attached to
the Goal marker thus forming vin-ga-ri’ and ro-ga-ri’. My uncertainty as
214 PARTITIONING OF THE STRUCTURE 8.3
to whether Hua indeed exemplifies pattern number 8 stems from the fact
that Haiman (1980) differs in his analysis of the Hua spatial morphemes
and suggests that the presence of the morpheme -ga is only optional in
Source expressions, as shown in (40) (my glossing).
(40) zu-ro(-ga)-ri’
work-at-to-from
oe.
come.1sg
‘I have come from work.’ (Haiman 1980:234)
The lexicalizations presented in this section can be summed up in the
following way (assuming that the analysis of Kibrik 2002 for Hua is cor-
rect).
(41) V Source Goal Place
Mandarin
verb
English
verb from
Yukatek Maya
verb ich
Laz
verb m- mot
Lezgian
verb -aj -l
Hua
verb -ri’ -ga -ro’


















8.3.3 Lexicalizing Goal paths
Let us now proceed to the lexicalization of the last structure left to discuss
— a Goal path. There are four general strategies to spell out such a path,
shown in Table 8.4.
Let us start with the first lexicalization strategy, where the verb spells
out the heads in the adpositional domain. As already mentioned, this
patterns is common in serial verb languages. I provide here an example
from Thai, taken from Zlatev and Yangklang (2004:163).
(42) chán
I
khˇaw
enter
hˇOON.
room
‘I went into the room.’
8.3 LEXICALIZATION PATTERNS 215
V Goal Place
1 Thai
2 | Evenki
3 | Yukatek
4 | | Tobati
Table 8.4: Possible partitionings of a Goal structure
The path-verb khˇaw ‘to enter,’ which in example (42) appears on its
own, lexicalizes the Goal and Place projections, in addition to the verbal
projection.
(43) VP
khˇaw⇐V
V Goal
Goal Place
GoalP
tGoal PlaceP
tPlace DP
The second lexicalization strategy is exemplified by Evenki (data from
Nedjalkov 1997:170).
(44) Bejumimni
hunter
hokto-tki
road-all
tuksa-d’ara-n.
run-pres-3sg
‘The hunter is running to(wards) the road.’
In Evenki, the Allative ending -tki spells out the Place and Goal heads,
as shown in (45).
216 PARTITIONING OF THE STRUCTURE 8.3
(45) VP
V
tuksad’aran
GoalP2
DP GoalP1⇒-tki
Goal PlaceP2
tDP PlaceP1
Place tDP
The third lexicalization strategy — the one where a verb takes a
locative expression — can be found in numerous languages where the
path is encoded on the verb rather than on the adpositional phrase. Not
surprisingly, this is also the lexicalization strategy used in Yukatek Maya.
As we already know, the verb in Yukatek Maya encodes the path, thus
a Goal expression consists of a “Goal-verb” plus a locative phrase (data
from Bohnemeyer and Báez 2008).
(46) Le=kàaro=o’
det=cart=dist
h-òok
prv-enter.3sg
ich
in
le=kàaha=o’.
det=box=dist
‘The cart, it entered (lit. in) the box.’
The corresponding tree diagram in shown in (47) (see also the analysis
of Goal expressions in Son and Svenonius 2008).
(47) VP
òok ⇐V
Goal V
GoalP
tGoal PlaceP
ich ⇐Place DP
le=kàaha=o’
Finally, a language exemplifying the fourth lexicalization pattern,
where there is a morpheme border both to the left and to the right of
Goal, is the Austronesian language Tobati. In Tobati, the Allative suffix
-d is stacked onto the Locative -i (data from Donohue 2002:199-200).
8.4 CONCLUSION 217
(48) a. Ntric
1sg
tad-i
sea-loc
nanac.
swim.pl
‘They are swimming in the sea.’
b. Nyiu
coconut
tad-i-d
sea-loc-all
rar.
fell
‘The coconut fell into the sea.’
The structure for the Tobati example in (48b) is presented below.
(49) VP
V
rar
GoalP2
PlaceP2
DP PlaceP1⇒-i
Place tDP
GoalP1⇒-d
Goal tPlaceP2
To sum up, I discussed various ways in which languages partition the
structure underlying Goal paths when they spell it out. The situation
can be summarized as follows.
(50) V Goal Place
Thai
verb
Evenki
verb all
Yukatek Maya
verb ich
Tobati
verb -d -i












8.4 Conclusion
In this chapter, I discussed two important issues related to the way the
fine-grained syntactic structure I proposed for directional expressions can
be lexicalized: spurious syncretisms and the partitioning of syntactic
218 PARTITIONING OF THE STRUCTURE 8.4
structure.
The term spurious syncretism refers to cases when one particular lexical
item lexicalizes the same piece of structure in the syntactic structures
underlying different directional expressions, leaving the non-lexicalized
head(s) to be spelled out by other “supporting” lexical items. On the face
of it, it then appears that this lexical item is ambiguous between two (or
more) notions. However, the ambiguity is not real, since at least one of
these notions is expressed only with the help of a “supporting” element.
By contrast, a real syncretism arises when one and the same morpheme
lexicalizes two distinct structures in a superset-subset relationship.
An interesting question is how this “syncretism typology” relates to
Talmy’s (2000) typology dividing languages into satellite-framed and
verb-framed depending on which element encodes the notion of path.
The short answer is that there is no one-to-one correspondence between
the two typologies. Both types of syncretisms can occur in a satelliteframed
language, that is, a language where the path is encoded by some
satellite of the verb (preposition, particle, case affix): real syncretisms
arise if there is a satellite that spells out more than one structure, and
spurious syncretisms arise if the “supporting” element is a satellite as well,
e.g., a verbal prefix or a preposition. Regarding verb-framed languages,
the only type of syncretism I have encountered until now is spurious
syncretisms, still the model predicts the possibility of there being verbs
that are structurally ambiguous. In sum, no clear parallel can be drawn
between Talmy’s typology and the distinction between real and spurious
syncretisms.
The collaboration between the spatial markers and the “supporting”
elements motivated a more detailed analysis of how the syntactic structure
can be partitioned to be spelled out by various elements. I presented
data from a number of languages, each of which represents a particular
type of partitioning, and showed how the syntactic structure is lexicalized
in order for the partitioning in question to obtain. The various
types of partitioning present independent evidence for the decomposed
Path structure, in addition to the evidence coming from the morphological
composition of individual spatial markers.
Chapter 9
Syncretisms
9.1 Introduction
In the preceding chapter, I explored the ways in which the structure underlying
Goal, Source and Route expressions can be partitioned among
various elements, sometimes giving rise to a phenomenon I dubbed spurious
syncretism. I discussed the difference between real and spurious
syncretisms, which can be summarized as follows: real syncretisms arise
when a given lexical item a spells out two or more distinct syntactic
structures; spurious syncretisms involve a lexical item a which always
lexicalizes the same syntactic structure, the impression of ambiguity is
caused by the fact that this syntactic structure can be part of a bigger
structure where the heads not lexicalized by a are lexicalized by a
different entry b.
Traditionally, syncretism is understood as the failure of a given formative
to make a morphosyntactic distinction (Spencer 1991, Baerman
et al. 2005). An example for such a lack of distinction can be found in
Georgian, which, according to Creissels (2008), syncretizes Location and
Goal of motion.
in at
Location
-ši -tan
Goal
Source -dan -gan
Table 9.1: Locative=Goal syncretism in Georgian (Creissels 2008)
Table 9.1 suggests that the formatives -tan and -ši (called secondary
219
220 SYNCRETISMS 9.1
case-endings in Vogt 1971) are used to express both Location and Goal
of motion. Thus, they fail to make a syntactically relevant distinction,
which results in a mismatch between syntax and morphology. According
to the definition of syncretism in Baerman et al. (2005:2), this is a runof-the-mill
example of syncretism.
The definition of syncretism I proposed in Chapter 8 as an instance
of structural ambiguity, however, raises the question whether the Inessive=Illative
and Adessive=Allative syncretism in Georgian is real. More
precisely, the question is whether -ši and -tan can express both Location
and Goal of motion by themselves. Put alternatively, we need to know
whether they can spell out both a Goal and a Locative structure. If they
do not, then there is no syncretism. If they do, then the Location/Goal
ambiguity should arise in the same context, i.e., in a combination with
the same verb, in a fashion analogous to the Hindi example in (44) in
Chapter 6.
Unfortunately, the data I was able to find does not give a clear answer
to this question. In Hewitt (1995), -ši and -tan are translated as ‘in’ and
‘at, by’, respectively, and no mention is made as to their use in Goalof-motion
context. Vogt (1971) provides ample data of the sort given in
(1) and (2),1
but gives no example which would be ambiguous between
a locative and a Goal reading.
(1) a. kalak-ši
town-ši
vcxovrob.
live
‘I live in the town.’
b. kalak-ši mivdavar.
town-ši go.past
‘I went to the town.’
(2) a. čems
I.gen
megobar-tan
friend-tan
viq’avi.
was
‘I was at my friend’s place.’
b. čems
I.gen
megobar-tan
friend-tan
mivedi.
go.perf
‘I have gone to my friend’s place.’
From the Georgian data above, it can be concluded that the Locative
interpretation of -ši and -tan is available with stative verbs like live and
be, and the Goal interpretation is available with inherently directed verbs
like go. However, this is not a suitable test-case for a true syncretism,
1
Vogt does not provide glosses for the examples and I have restrained from glossing
-tan and -ši in order to remain neutral as to their function.
9.2 POSSIBLE AND IMPOSSIBLE SYNCRETISMS 221
since it is possible that -tan and -ši are purely Locative case suffixes and
the Goal reading in the b-examples is the result of the motion verb go
lexicalizing the Goal head. In order to probe into the potential Location=Goal
syncretism, it is necessary to create an ambiguous sentence of
the sort found in English (see (3)).
(3) The mouse ran under the table. (Location or Goal)
The case of the Georgian endings -ši and -tan raises a more general worry
about the approach to syncretisms found in grammar descriptions. Very
often it is the case that both spurious and real instances of syncretisms are
subsumed under the label “syncretism.” As a result, works in theoretical
linguistics drawing data from such grammars take cases of what might
well be spurious syncretisms to involve truly syncretic lexical entries.
The concern raised above will come up again and again throughout
this chapter, since I will be dealing with cross-linguistic investigations of
the syncretisms found within the spatial domain. The goal of this chapter
is to show which syncretism patterns are predicted to be impossible
and which ones are expected to exist by the decomposed Path structure
proposed in Part I and the Spell-out system developed in Part II. After
doing that, I discuss a couple of counterexamples and show that they
could well be instances of spurious syncretisms. Finally, I turn to an
interesting instance of a syncretism predicted to be impossible found in
English and shown how it can be explained away.
9.2 Possible and impossible syncretisms
9.2.1 *ABA
This subsection puts forward the view that syncretisms in the spatial
domain can target only adjacent heads in the syntactic structure —
an idea elaborated in detail by Caha (2008), who draws inspiration
from Bobaljik’s (2007) investigation of comparative suppletion across languages
and Michal Starke’s unpublished work on -ed/-en allomorphy in
English. Using the syntactic structure for Location, Goal and Source
phrases which I argued for in Chapter 4, I will reproduce the reasoning
for why this should be the case.
Let us assume the hypothetical lexical entries in (4).
222 SYNCRETISMS 9.2
(4) a. a ⇔ < SourceP
Source GoalP
Goal PlaceP
Place
>
b. b ⇔ < GoalP
Goal PlaceP
Place
>
Consider now how a language with such entries lexicalizes a Source
phrase, which has the structure in (5).
(5) SourceP
Source GoalP
Goal PlaceP
Place ...
There is only one lexical entry which is eligible for insertion at the SourceP
node — the entry a — because it is the only entry which is specified
for a constituent containing that node.
(6) SourceP⇒a
Source GoalP
Goal PlaceP
Place ...
Turning now to a Goal phrase, there are now two items which can be
used to spell it out: a and b.
9.2 POSSIBLE AND IMPOSSIBLE SYNCRETISMS 223
(7) GoalP⇒a or b?
Goal PlaceP
Place ...
Both a and b contain GoalP in their lexical specification. The Elsewhere
Principle enforces the use of the more highly specified entry, in this case,
of the entry b, which is a better match.
(8) GoalP⇒b
Goal PlaceP
Place ...
The same competition between a and b arises for the lexicalization of a
Locative phrase, since they both contain PlaceP.
(9) PlaceP⇒a or b?
Place ...
The winner is again b, because it contains fewer superfluous features
(one) than a (two).
(10) PlaceP⇒b
Place ...
The result is a paradigm of the shape abb, where a spells out the biggest
Source structure and b spells out both the intermediate-sized Goal structure
and the smallest Locative structure. Importantly, a cannot be used
to lexicalize the smallest Locative structure because it loses the competition
with b by the Elsewhere Principle.
This is the gist of the *ABA generalization of Bobaljik (2007). The
*ABA generalization states that it is not attested among languages that
a structure [X [Y [Z]]] is lexicalized by using A, [Y [Z ]] is lexicalized by
using B, and [Z ] is lexicalized by using A again. As explained by Caha
(2008) in nanosyntactic terms, the reason is that whatever lexical item is
used for the lexicalization of the intermediate-sized structure, the same
item will be used for the smallest structure by virtue of the Elsewhere
Principle, thus leading to a pattern ABB.
The restriction that syncretisms target adjacent heads predicts the
224 SYNCRETISMS 9.2
impossibility of a lexical entry to syncretize (i) Location and Source
across a Goal head, (ii) Location and Route across the heads Goal and
Source, and (iii) Goal and Route across a Source head. This is a prediction
to which I will come back in Section 9.2.3
9.2.2 *A&¬A
Let us now turn to one more syncretism which I suggest is not expected
to be found across languages. Consider first the lexical entries in (11).
(11) a. b ⇔ < SourceP
Source GoalP
Goal PlaceP
Place
>
b. a ⇔ < PlaceP
Place
>
The way a language with such entries will lexicalize a Locative phrase is
by using the “perfect match” a.
(12) PlaceP⇒a
Place ...
The Goal expression can be spelled out only by b, since a does not have
the feature <Goal>.
(13) GoalP⇒b
Goal PlaceP
Place ...
The same entry b will also be the only match for a Source phrase.
9.2 POSSIBLE AND IMPOSSIBLE SYNCRETISMS 225
(14) SourceP⇒b
Source GoalP
Goal PlaceP
Place ...
Under the Path decomposition analysis argued for in this thesis, and
the semantics proposed for the Path heads in Chapter 5, the Source head
is the locus of a reversal operation which applies to the Goal phrase.
Thus, in a sense, a Source path is the “opposite” (or the negation) of a
Goal path. This means that a language with a Goal=Source syncretism
has one spatial marker that expresses a certain meaning and its exact opposite.
I suggest that from a pragmatic point of view it is unacceptable
to have such a “contradictory” lexical item.2
Therefore, we would expect
that the Goal=Source syncretism is unattested, although it is grammatical.
Notice that this does not exclude the possibility that a given lexical
item a includes the features <Source> and <Goal> in its feature specification.
As long as there is a disambiguating (i.e., more highly specified)
lexical item b that limits the use of a to one of the spatial roles only, the
item a is not used in a contradictory way.
To conclude, the semantic contribution of the heads in the decomposed
Path structure I propose is such that a syncretism pattern where
Goal and Source paths are expressed by the same marker is expected to
be non-existent due to a pragmatic constraint enforcing disambiguation.
9.2.3 Syncretism typology
The restrictions on syncretisms discussed in the preceding two sections
significantly reduce the number of potentially possible syncretism patterns
among the expressions of Location, Goal, Source and Route. More
precisely, out of the 15 logically possible syncretisms, only 4 are allowed
2
There are potential counterexamples, for example those verbs that have both an
ornative and a privative meaning like seed, trim, etc.
(i) a. seed the grapes = remove seeds from the grapes
b. seed the lawn = put seeds in the lawn
Buck (1997) investigates that type of verbs in English and convincingly argues that
they do not express opposite meanings. Hence, these verbs are not an example of
contradictory lexical items.
226 SYNCRETISMS 9.2
by the system.
Type 1 Location=Goal=Source=Route
Type 2 Location=Goal=Source=Route
Type 3 Location=Goal=Source=Route
Type 4 Location=Goal=Source=Route
Table 9.2: Possible syncretisms
The syncretisms which are predicted to be impossible are presented
in Table 9.3. The rightmost column gives the reason why the relevant
syncretism pattern is excluded. In order to unambiguously refer to the
various types of syncretism patterns in the upcoming discussion, I continue
the numbering from Table 9.2.
Type 5 *Location=Source=Goal=Route *ABA
Type 6 *Location=Source=Goal=Route *ABA
Type 7 *Location=Source=Goal=Route *ABA
Type 8 *Location=Route=Goal=Source *ABA
Type 9 *Location=Route=Source=Goal *ABA
Type 10 *Location=Route=Goal=Source *ABA
Type 11 *Location=Route=Goal=Source *ABA and *A&¬A
Type 12 *Location=Goal=Source=Route *A&¬A
Type 13 *Location=Goal=Source=Route *A&¬A
Type 14 *Location=Goal=Source=Route *A&¬A
Type 15 *Location=Goal=Source=Route *A&¬A
Table 9.3: Impossible syncretisms
The syncretism typology in Tables 9.2 and 9.3 in fact presents a very
simplified picture of the conceivable lexicalization patterns. Taking as an
example the first three spatial roles in Type 1, a Location=Goal=Source
“syncretism” can be instantiated in various ways, depending on what
lexical items and combinations of lexical items a languages uses (see
Table 9.4).
All lexicalization patterns in Table 9.4 instantiate a syncretism pattern
where the notion of Location is expressed differently from the notion
of Goal of motion, which in turn is expressed differently from the notion
of Source of motion. The common feature between all these languages
is that they all have three distinct morphemes a, b and c, which can
spell out the Location, Goal and Source structures either by themselves
9.2 POSSIBLE AND IMPOSSIBLE SYNCRETISMS 227
Location Goal Source Example
a b c English: at, to, from
a a + b a + c Estonian: -l, -l-le, l-t (Viitso 1998)
a b b + c Quechua: -pi, -man, -man-da (Cole 1985)
a b a + c Uzbek: -D˙a, -G˙a, -D˙a-n (Boeschoten 1998)
Table 9.4: Possible instantiations of an L=G=S syncretism
or in combination with each other. Similar versatility also holds of the
other possible syncretism patterns, see Pantcheva (2010) for a discussion
of some of them.
The inclusion of the different combinations of lexical items into the
picture combined with the (to my mind just) assumption that a combination
a+b represents a distinct lexicalization from a lexicalization by
only a, raises the question whether some of the impossible syncretism
patterns predicted to be impossible do not become possible. Imagine a
language which has the following two lexical items:
(15) a. a ⇔ < SourceP
Source GoalP
PlaceP
Place
GoalP
Goal
>
b. b ⇔ < GoalP
Goal
>
On the face of it, we can expect that this language will use a to lexicalize
Location, since no other lexical item has the relevant feature <Place>.
The structure for Goal path will be lexicalized by a combination of a and
b (triggering a roll-up movement) and the structure for Source structure
will be lexicalized by a again, because it is the only entry which has the
Source feature. The obtained pattern will be the one in (16), suspiciously
resembling the prohibited *ABA pattern.
228 SYNCRETISMS 9.3
(16)
Location Goal Source
a a + b a
A detailed derivation shows, though, that the lexicalization pattern in
(16) is impossible under the Spell-out system adopted here. Consider
the stage of derivation when b is matched to GoalP.
(17) GoalP2
PlaceP2
DP PlaceP1⇒a
Place tDP
GoalP1⇒b
Goal tPlaceP2
At the next cycle, which begins with the evacuation movement of PlaceP2,
GoalP2 is targeted for lexicalization. Consequently the lexical item a is
chosen as a match for GoalP2 thus overriding the combination a + b. In
other words, the lexicalization of a Goal phrase as a + b never surfaces
since it is overridden by a at the next stage. In fact, a language with
the lexical items in (15) is predicted to be impossible by the *A&¬A
constraint, because all spatial roles will end up being lexicalized by a.
The aim of the discussion above is to illustrate the fact that once
all possible lexicalization patterns are taken into account, the picture
becomes very complicated. Still, I would like to maintain that the impossible
syncretism patterns in Table 9.3 remain impossible, despite the
numerous ways in which they could be instantiated, since it appears that
the Spell-out system adopted here excludes any lexicalization pattern
that would correspond to a forbidden syncretism pattern.
9.3 Typological studies: testing the predictions
about syncretisms
In the preceding section, I suggested that there are only four possible
syncretism patterns involving the notions of Location, Goal, Source and
Route. Such a claim should be empirically verifiable in that one expects
to find them all across the languages of the world. In reality, it
turns out that the confirmation of this hypothesis requires a great deal
9.3 TYPOLOGICAL STUDIES 229
of fieldwork, because such data can hardly be retrieved from grammar
descriptions. First, reference grammars rarely present the expressions
of directed motion in a detailed enough way. Consequently, it becomes
impossible to determine whether a syncretism is real or spurious, as I
already demonstrated for the case of Georgian in the introduction to this
chapter. Second, the expressions of Route of motion are often straight
out omitted from grammar descriptions, thus leading to a partial information
as to the syncretisms involving Route.
The other side of the coin is that syncretisms that are predicted to
be impossible are claimed to exist, again without data being presented
which would exclude a spurious syncretism.
By and large, linguistic studies dealing with the expression of the spatial
roles Location, Goal and Source (among other thematic roles) confirm
the asymmetry in the distribution of the logically possible syncretism patterns
among languages. This asymmetry has been stated most concisely
by Andrews (1985), who discusses the distribution of the syncretism patterns
involving Location, Goal and Source cross-linguistically. According
to this author, there is a clear skewing in favor of some patterns, while
others seem to be unattested.
A particularly interesting tendency [...] is for certain groups
of notions but not others to be expressed by the same marker
in many different languages. Thus sometimes one finds the
same np-marker coding the Locative, Goal and Source roles
[...], sometimes one finds Locative and Goal expressed by the
same marker, with a different one for Source [...], and sometimes,
as in Warlpiri, different markers are used for all three
locative roles. But one doesn’t seem to find one marker used
for Locative and Source, with a second for Goal; or one for
Source and Goal, with a different for Locative.
(Andrews 1985:97)
Thus, leaving aside the notion of Route, Andrews states the syncretism
patterns Location=Source=Goal (Type 5-6) and the pattern
Location=Goal=Source (Type 11-13) do not seem to be found (in compliance
with the predictions), while there is a general trend towards the
syncretisms Location=Goal=Source (Type 1-2), Location=Goal=Source
(Type 3-4), and Location=Goal=Source (Type 14-15). The last one conflicts
with the prediction of this thesis and I will come back to it shortly.
The same observation has been made in other works where the topic
of syncretisms within the spatial domain is being discussed (Creissels
230 SYNCRETISMS 9.3
2006; 2008, Radkevich 2009, Nikitina 2009, Lestrade 2010). Various typological
studies dealing with these three spatial roles also show that the
general tendency supports the predictions. I briefly review these studies
here.
Blake (1977)
Blake (1977) examines the syncretism tendencies against a sample of
Australian languages. In Appendix I, Blake lists the case forms for 115
Australian languages. Of those, 85 languages are listed as having all three
spatial cases: Locative (encoding Location), Allative (encoding Goal of
motion), and Ablative (encoding Source of motion). A survey of these 85
languages reveals that 91% (77 languages) have a special form for each
spatial case, that is, follow the pattern Location=Goal=Source. Nine
percent (8 languages) have one shared case affix for Locative and Allative
and a separate one for Ablative, exemplifying the syncretism pattern
Location=Goal=Source. No language from this sample exhibits the
syncretism patterns Location=Goal=Source, Location=Goal=Source or
Location=Source=Goal, thus complying with all the predictions made.
Noonan (2008)
The tendency for languages to lexicalize Location, Goal and Source of
motion according to the patterns in Table 9.2 finds further support in the
typological study conducted by Noonan (2008). He examines the overall
patterns of syncretism in 76 Tibeto-Burman languages. The results
show a very robust Locative=Allative syncretism (44 languages): 58%
of the examined Tibeto-Burman languages pattern according to the type
Location=Goal=Source. The pattern Location=Goal=Source is represented
by 25 languages (33%). However, four percent (3 languages) are
claimed to use the same marker for all three functions, thus exemplifying
the type Location=Goal=Source. Further, two languages (2,5%) exhibit
the pattern Location=Goal=Source and two languages (2,5%) follow the
pattern Location=Source=Goal, which goes against the predictions of
the theory. In sum, five languages challenge the *A¬A constraint, which
has a pragmatic motivation, and two languages go against the *ABA
constraint, for which there is a structural explanations. Unfortunately,
Noonan (2008) does not give the names of the languages in question and
this hinders further investigation of the syncretism patterns in these languages,
which could turn out to be spurious.
9.3 TYPOLOGICAL STUDIES 231
Pantcheva (2010)
In Pantcheva (2010), I report the study of 53 languages comprising 22
genera (which in their turn represent 14 language families), as well as
two language isolates. The results are the following: 28 of the languages
examined follow the pattern Location=Goal=Source, which constitutes
53%. The pattern Location=Goal=Source is represented by 34% of the
languages. No language exhibits either of the patterns Location=Source=Goal
or Location=Goal=Source. Finally, thirteen percent of the languages in
the sample have one unique spatial marker that is used in the expression
of all three spatial roles (pattern Location=Goal=Source).
Rice and Kabata (2007)
The typological study in Rice and Kabata (2007) takes a different perspective
on the syncretism patterns, but still allows one to state some
generalizations. Rice and Kabata take as a starting point the Allative
marker (regardless of whether it is a case affix or an adposition) and examine
what other functions it can have (for example, Locative, Ablative,
Purposive, Benefactive, etc.). They examine the models of Allative syncretisms
in 44 genealogically diverse languages. The upshot is that ten
languages (23%) use the same marker for Allative and Locative (pattern
Location=Goal=Source). Five languages (11%) use the same marker for
the Allative, Locative and Ablative functions, thus exhibiting the pattern
Location=Goal=Source.
The remaining 29 languages (66%) are the ones where the Allative
marker is syncretic neither with Locative nor with Ablative. Since this
study of syncretism patterns aims to answer the question which functions
the Allative marker can express other than the Allative, no information
can be retrieved as to whether Locative and Ablative can be syncretic to
the exclusion of the Allative, that is, Location=Source=Goal. Similarly,
nothing can be said about languages in the sample that follow the pattern
Location=Goal=Source. Therefore, no conclusion can be drawn concerning
the distribution of the 29 remaining languages between the patterns
Location=Goal=Source and Location=Source=Goal. Nevertheless, in
the sample of Rice and Kabata (2007), there is not a single language that
uses the Allative marker to express also the Ablative function, but has a
separate Locative marker (i.e., the pattern Location=Goal=Source).
Table 9.5 summarizes the cross-linguistic lexicalization patterns, ar-
232 SYNCRETISMS 9.3
ranged according to their frequency. The numbers in the brackets show
the actual number of languages: the first number is the number of languages
that are claimed to have the relevant pattern, the second number
is the total number of languages included in the respective sample.
Blake (1977) Noonan (2008) Pantcheva (2010)
Rice and
Kabata (2007)
L=G=S 91% (77/85) 33% (25/76) 53% (28/53) —
L=G=S 9% (8/85) 58% (44/76) 34% (18/53) 23% (10/44)
L=G=S 0% (0/85) 4% (3/76) 13% (7/53) 11% (5/44)
L=S=G 0% (0/85) 2,5% (2/76) 0% (0/53) —
L=G=S 0% (0/85) 2,5% (2/76) 0% (0/53) 0% (0/44)
Table 9.5: Pattern of syncretism for the lexicalization of Location, Goal,
and Source
The proposed distinction between real and spurious syncretism raises
the question of how many of the syncretism cases in Table 9.5 involve
truly ambiguous spatial markers. A closer inspection reveals that many
of them turn out to be spurious. For instance, Pantcheva (2010) includes
Classical Tibetan in the group of languages with a Location=Goal=Source
syncretisms, as suggested in DeLancey (2003). However, Beyer (1992:268)
states that, although Classical Tibetan uses “the same locus particles with
verbs of location and verbs of motion[, t]here is no confusion because, of
course, the verbs are different.” Given that the interpretation of a noun
marked by -la (general location) or -na (interior location) in Classical
Tibetan is always disambiguated by the verbs, as suggested by Beyer, it
is more plausible that we are dealing with a case of spurious syncretism
rather than a real one.
Similarly, all of the Location=Goal=Source syncretisms that I have
been able to check turn out to be spurious. To give a few examples,
Rice and Kabata (2007) present Tagalog as a language where the unique
spatial marker sa carries a Locative, an Allative and an Ablative function.
While it is true that sa-phrases participate in expressions of Location,
Goal and Source, they by no means do that alone. According to Schachter
and Otanes (1972:260), Source expressions are built of a noun marked
by sa combined with one of the following verbs or adverbs: buhat, galing,
mula or tubo, all translated as ‘from.’ The fact that a supporting frommorpheme
is needed in order to construct a Source expression suggests
that the Goal=Source syncretisms in Talagog is spurious. Concerning
the Locative=Goal syncretism in Tagalog, there is evidence that it is
9.3 TYPOLOGICAL STUDIES 233
a real one, since “[i]n a sentence with a verbal predicate, a sa phrase is
sometimes ambiguously interpretable as a locative adverb or a directional
complement” (Schachter and Otanes 1972:450). In sum, Tagalog rather
represents a Location=Goal=Source syncretisms, if we take syncretism
to be an instance of structural ambiguity.
Similarly, Pantcheva (2010) takes the Tibeto-Burman language Lahu
to have a Location=Goal=Source syncretisms, since it has several noun
particles of general locative meaning that are neutral with respect to
directionality. However, their interpretation depends on the built in semantic
features of the clause’s verb. Consider, for instance, how the
directional interpretation of a phrase marked by the postposition -lo (attaching
to inanimate Grounds) is determined by the verb it is combined
with (data from Matisoff 2003:216, the original glossing represented).
(18) a. há-q¯o
cave
lo m1
sit
chˇE
prog
ve.
Pu
‘He is sitting in the cave.’ (Location)
b. há-q¯o
cave
lo lòP
enter
e
Pv
ò.
Pv
‘He has already gone into the cave.’ (Goal)
c. há-¯qo
cave
lo tˇOP
emerge
e
Pv
ò.
Pv
‘He has already come out of the cave.’ (Source)
As Matisoff (1973:162) puts it, “[w]hether a [N+lo] is to be translated as
‘to/towards/into N,’ or ‘in/at N,’ or ‘from/out of/away from N’ depends
entirely on the semantics of the clause’s main verb.” Such dependence
is characteristic of spurious syncretisms where the verb lexicalizes the
projections which come on top of the Place head in Goal and Source
expressions.
To summarize, many of the alleged syncretisms in Table 9.5 turn out
to be spurious upon closer inspection, that is, they require the support
of an additional element in order to express Goal and Source paths. As
far as I was able to determine, in none of the languages included in
Rice & Kabata’s and Pantcheva’s samples is there a real Goal=Source
syncretism. Regarding languages with a Location=Goal=Source syncretism,
I suggest that such languages have, in fact, a unique spatial
marker with a default locative interpretation. In order for this marker
to acquire a Source or a Goal meaning, it has to occur with the right
verb. Similar observations have been made also for other languages, not
included in the samples, e.g., the Bantu language Tswana (Creissels
234 SYNCRETISMS 9.3
2006), Yukatek Maya (Bohnemeyer and Stolz 2006), etc. The examples
of Location=Goal=Source syncretisms which I encountered most often
feature pairs where the same (locative) morpheme appears in combinations
with verbs translated as come and go. Such verbs are not well
suited for testing real syncretisms. Many languages have pseudo-deictic
and non-deictic verbs come and go which have the property of being
inherently Goal-oriented and Source-oriented, respectively (Ricca 1993).
Under the current account, an inherently Goal-oriented verb which takes
a locative phrase as a complement lexicalizes the Goal head, and an
inherently Source-oriented verb lexicalized the Source and Goal heads.
9.3.1 The possible patterns of syncretisms: some ex-
amples
As discussed in Section 9.2.3, there are only four syncretism patterns
that are predicted to be possible by the model developed in this thesis,
see Table 9.6, repeated from Table 9.2. In this section, I present a couple
of languages which instantiate the possible syncretism patterns. In doing
this, I focus mainly, but not exclusively, on the general topological notions
at, on, in and leave aside the more complex projective terms like
under, behind, among (see Levinson and Meira 2003, Zwarts 2008a
for a hierarchy of spatial terms).3
Type 1 Location=Goal=Source=Route
Type 2 Location=Goal=Source=Route
Type 3 Location=Goal=Source=Route
Type 4 Location=Goal=Source=Route
Table 9.6: Possible syncretisms (repeated from Table 9.2)
The first type of syncretism is found in English, where the prepositions
at, to, from and via give rise to different spatial interpretations,
even when combined with the same verb.
(19) a. I ran at the stadium. (Location)
b. I ran to the stadium. (Goal)
3
It should be noted that very often a given language instantiates more than one
syncretism pattern. For instance, the English canonical spatial expressions (at, to,
from) instantiate Location=Goal=Source, while the spatially more descriptive prepositions
under, behind, etc., show a Location=Goal=Source syncretism (Jackendoff
1983).
9.3 TYPOLOGICAL STUDIES 235
c. I ran from the stadium. (Source)
d. I ran via the stadium. (Route)
The second type of syncretism has already been illustrated by the Hindi
Ablative marker -see, which syncretizes Source and Route, as I showed
in Chapters 6 and 8 (I repeat the data in (22)). In order to lexicalize
a Goal phrase, Hindi uses the Dative marker -koo and for a Locative
phrase it uses one of the locative markers -mee ‘in’ or -par ‘on’ (data
from Narasimhan 2008).
(20) a. kamree-mee
room-loc
‘in the room’(Location)
b. pharsh-par
floor-loc
‘on the floor’ (Location)
(21) meez-koo
table-dat
‘to the table’(Goal)
(22) a. baccaa
child
kaar-ke
car-gen
saamne-see
front-abl
calaa.
walk.perf
‘The child walked from/via in front of the car.’(Source or
Route)
b. ciRijaa
bird
Jhiil-ke
lake-gen
upar-see
above-abl
uRi.
flew
‘The bird flew from/via above the lake.’ (Source or Route)
Other languages which have been claimed to have a Source=Route syncretism,
but no Location=Goal or Goal=Source syncretism are Qiang,
Tibeto-Burman (LaPolla 2003), Karata, Daghestanian (Magomedbekova
1971), Basque, isolate (Hualde and de Urbina 2003), and Marathi, IndoIranian
(Pandharipande 1997). Unfortunately, the data presented in the
sources cited does not allow to determine whether the Source=Route
syncretism in these languages is real or spurious.
Turning now to the third type of possible syncretism, it turns out to be
surprisingly rare to come across an undoubtedly real Location=Goal syncretism
of the sort found with the English projective prepositions under,
behind and others (see Jackendoff 1983 and example (3) in the current
chapter). Although the results from the typological studies presented in
Section 9.3 show that this is the second most common syncretism pattern
cross-linguistically, it is very difficult to determine which languages have
236 SYNCRETISMS 9.3
a real Location=Goal syncretism (as opposed to a spurious one). As I
already complained, the main problem is the lack of data, since grammar
description generally do not take into consideration the possibility
of there being a spurious syncretism.
A good candidate for such a syncretism is Tagalog, as suggested by
the fact that a phrase marked by the spatial preposition sa can be ambiguously
interpretable as a locative adverb or a directional complement
(Schachter and Otanes 1972:450). Another language where this syncretism
type is found is French, see the data in (23).
(23) French (data from Nikitina 2009, confirmed by M. Starke, p.c.)
a. J’ai
I.have
couru
run
au
at/to.the
stade.
stadium
‘I ran at the stadium’ (Location)
‘I ran to the stadium’ (Goal)
b. J’ai
I.have
couru
run
du
from.the
stade.
stadium
‘I ran from the stadium’ (Source)
Finally, let us turn to the last syncretism type Location=Goal=Source=Route.
A language with this pattern would have one spatial marker which is
ambiguous between Location and Goal of motion, as in French, and a
second marker ambiguous between Source and Route, as in Hindi. Until
now, I have not been able to find a good example of such a syncretism
type, for there is a double impediment. First, the information provided
in grammars about Route expressions is very sparse. Second, the usual
issue with the lack of information about whether a given syncretism is
real or spurious obtains. Good candidates for such a pattern seem to
be the Caucasian languages Godoberi (Gudava 1967c), Bagvalal (Gudava
1967a), Tindi (Gudava 1967d), Hunzib (Bokarev 1967b) and Bezhta
(Bokarev and Madieva 1967). Many Caucasian lanugages are claimed to
have a Source=Route syncretism in Vinogradov (1967), and all the abovementioned
languages are in addition claimed to have a Location=Goal
syncretism. This hypothesis remains to be confirmed.
9.3.2 The impossible patterns of syncretisms: some
counterexamples
In Section 9.2, I suggested that there are two general types of syncretisms
that are predicted by the model to be impossible. One is the *ABA-type
of syncretism, where two heads are syncretic across another. The other
9.3 TYPOLOGICAL STUDIES 237
type is a syncretism of Goal and Source, i.e., a syncretism involving two
“opposite” notions. This section is devoted to the discussion of some
languages which have been claimed to exhibit a prohibited syncretism
pattern. I will show that these syncretisms are more correctly analyzed
as spurious.
Starting with the Goal=Source syncretism, Laz is a language which
has been reported to have a spatial marker ambiguous between the two
notions (Broschart and Dawuda 1999). As I showed in Chapter 8, Section
8.3.2, basing the discussion on data from Kutscher (2001; 2010), the
Goal=Source syncretism in Laz is more correctly analyzed as a spurious
syncretism. Further, in the beginning of the current section, I discussed
the question of how real the Location=Goal=Source syncretisms
reported in Rice and Kabata (2007) and Pantcheva (2010) are. I showed
that a closer inspection of Tagalog and Lahu reveals that at least the
Goal=Source part of the pattern is a spurious syncretism, conforming to
the pragmatic constraint *A&¬A suggested in Section 9.2.2. I reached
the same conclusion when I investigated the sample of “Goal/Source indifferent
languages” presented in Wälchli and Zuñiga (2006), where we find
languages like Lahu and Yukatek Maya with a spurious (Loc=)Goal=Source
syncretism. In sum, I can say that so far I have not come across any convincing
case of a real Goal=Source syncretism.
Let us now turn to the *ABA-type of syncretism. Abstracting away
from the Route expressions, for which very little is said in the literature
on syncretisms, the prohibited syncretisms are the ones where Location
and Source are expressed by the same marker to the exclusion of Goal.
There are several languages which have been claimed to exhibit this
syncretism pattern: Veps (Radkevich 2009), Dinka (Creissels 2008, Nikitina
2009, Lestrade 2010), Nivkh (Nikitina 2009, Radkevich 2009, Lestrade
2010), Kanuri and Old Georgian (Creissels 2008, Lestrade 2010), Iraqw
(Creissels 2006, Lestrade 2010), and Nukuoro (Lestrade 2010). I discuss
them briefly below.
The alleged Location=Source syncretism in Veps is rather a terminological
issue stemming from the particular labeling of two spatial cases:
Addesive-Ablative and Inessive-Elative. Despite their names (which reflect
a diachronic development), these two cases have purely locative
functions. The Source expressions in Veps are formed by adding a dedicated
morpheme -pei, to an Addesive-Ablative or Inessive-Elative marked
noun (Zajtseva 1981). Below, I give an example of an Adessive-Ablative
marked noun and show how the corresponding Source phrase is formed.
238 SYNCRETISMS 9.3
(24) Veps (Zajtseva 1981, my glossing)
a. käde-s
hand-iness.elat
‘in the hand’ (Locative only)
b. käde-s-pei
hand-iness.elat-from
‘out of the hand’ (Source only)
The cases of Dinka and Nivkh have a greater potential of being real counterexamples,
however, the available data does not provide clear evidence
for such a claim. Consider the Dinka data in (25).
(25) Dinka (Andersen 2002)
a. Myè
¨
et”
food
à
˜
-t`O
¨d-be.present
m´E
¨
EEc.
fire.iness
‘The food is on the fire.’ (Location)
b. R`E
¨
Ec
fish
à
˜
-mù
¨
ul
d-crawl
bè
¨
y
out.all
m´E
¨
EEc.
fire.abl
‘The fish is crawling out from the fire.’ (Source)
The form of the noun m´E
¨
EEc ‘fire.iness/abl’ is indeed identical in both
sentences. The sentence in (25b), however, suggests that we need the
support of a special Allative-marked particle bè
¨
y to get the Source reading.
M´E
¨
EEc by itself seems not to be enough. This casts doubts on the
claim that the Locative form in Dinka can be also used to express a
Source path. The doubts are reinforced by the data in (26).
(26) Dinka (Nebel 1948)
a. GEk
cattle
atO
is
luEk.
byre.loc
‘The cattle is in the byre.’ (Location)
b. WeN
cow
aci
past
riN
run
luEk.
byre.loc
‘The cow ran into the byre.’ (Goal)
If Dinka really had a Location=Source syncretism, then the expected
interpretation of the sentence in (26b) would be Source-directional and
not Goal-directional, contrary to the fact.
A similar suspicion can be raised regarding the Location=Source syncretism
in Nivkh. The only data on Nivkh that I found is also that cited
by Nikitina, taken from Gruzdeva (1998) (see (27)). As the data shows,
the Source reading of a locative expression is available when the verb
9.3 TYPOLOGICAL STUDIES 239
encodes a Source path (e.g., come out).
(27) Nivkh (Gruzdeva 1998)
a. T’ivla¸n
cold
čag-ux
water-loc/abl
¸nat’x-Ø
foot-nom
vezla-d.
cramp-fin
‘[I] have a cramp in [my] foot in the cold water.’
b. Umgu-Ø
woman-nom
n’o-x
barn-loc/abl
p’u-d’.
come.out-fin
‘A woman came out from the barn.’
In sum, it is debatable whether Dinka and Nivkh really exhibit the syncretism
pattern Location=Source=Goal. On the basis of the data in
(25)-(27), I suggest that they do not, although they do use the same case
marker in expressions of Location and Source. The Source expressions,
however, take more than just the case ending, for instance a Sourceencoding
verb or a particle. The lack of a Location/Source ambiguous
example and the apparent need of a supporting element suggest that the
Location=Source syncretisms in Nivkh and Dinka are spurious.
The same lack of convincing data marks the other languages mentioned
as having a Location=Source=Goal syncretisms. In the Kanuri
grammar of Hutchison (1981), I found both Locative and Source expressions
where the same postposition lán marks the Ground DP. However,
they feature different verbs: non-motion verbs (work, live) in locative
contexts and directed motion verbs (mostly leave and come) in Source
contexts. Thus, it is plausible that lán is a purely locative postposition
and the Source reading is due to the directed-motion verbs lexicalizing
the Source and Goal heads. This proposal is strongly supported by the
fact that the only interpretation of a noun marked by lán in non-verbal
contexts is the locative one (Hutchison 1981:176).
The Old Georgian, Iraqw and Nukuoro examples provided in Lestrade
(2010) do not contain data where a spatial expression is ambiguously
interpreted as expressing Location or Source of motion either. Rather,
the examples feature different verbs and/or additional particles in the
Source expressions. Consider, for instance, the example below taken
from Lestrade (2010:102)
(28) Nukuoro (Lestrade 2010, citing Ross Clark, p.c.)
a. Kai
and
kilaateu
they
ka
past
teletele
sail.sail
ai
prt
i
on
te
the
moana.
sea
‘And they kept sailing on the open sea.’ (Location)
240 SYNCRETISMS 9.3
b. Ka
past
hulo
go.pl
kee
away
i
from
Kapingamaalangi.
K.
‘[They] left Kapingamarangi.’ (Source)
c. Ka
past
lava
finish
ka
past
hulo
go.pl
ki
to
Luuku
L.
ma
and
Motolako.
M.
‘Then they went to Truk and the Mortlocks.’ (Goal)
While it is true that the same preposition i is used in both Locative,
(28a), and Source expressions, (28b), it is not excluded that the particle
kee translated as ‘away’ present only in (28b) is the lexical item that
lexicalizes the Goal and Source heads. If this is so, then the Nukuoro
Location=Source syncretism is a spurious one.
The last potentially real Location=Source=Goal syncretism that I
have come across in my investigations is found in North Sámi. Northern
Sami has an Illative (Goal) case and a Locative case. The latter is said to
syncretize Inessive and Elative (i.e., Location and Source). In the Locative
singular, marked by -s(t), the Inessive-Elative syncretism is seen as
an accidental homophony resulting from the phonological development
of the Proto-Sámi Inessive and Elative endings *-sn¯e and *-st¯e, respectively.
The merger of the two cases arose due to an apocope followed
by a devoicing and desonorization of the n in the Inessive (*-sn¯e > *-sn
> *-sn
˚
>*-st) (Sammallahti 1998:66). The problem is that the InessiveElative
syncretism has been extended to other parts of the grammar. For
instance, the morphological opposition between Inessive and Elative was
lost also in the plural paradigm and with spatial adverbs and postpositions
by analogy to the singular paradigm and crucially not because
of a phonologically conditioned development (Sammallahti 1998, Hansson
2007). Thus, North Sámi presents a real challenge to the account
developed in this thesis, which bans such a syncretism.
Svenonius (2009a) conducts a thorough investigation of the North
Sámi Locative case, collecting rich data involving Locative and Source expressions.
The detailed data analysis he performs shows that the Source
reading is available whenever the Locative is not and, importantly, requires
the presence of an additional element encoding transition in some
way: a Goal expression, or an Aktionsart operator, like the ‘subitive’ l
presented in (29b).
(29) North Sámi Locative (data from Svenonius 2009a)
a. Joavnna
Jon
viegai
ran
viesus.
house.loc
‘Jon ran in the house.’ (Locative)
9.3 TYPOLOGICAL STUDIES 241
b. Joavnna
Jon
viehka-l-ii
run-sub-past
viesus.
house.loc
‘Jon suddenly ran off from the house.’ (Source)
Svenonius then develops a fine-grained semantic and syntactic analysis
for the English and North Sámi Source expressions, showing that they
are in fact very different. Importantly, he concludes that the North Sámi
Locative phrase always spells out a PlaceP, even in Source expressions,
thus making the Location=Source syncretism in North Sámi a spurious
one.
The last potential counterexample I will discuss is an *ABA-type
of syncretism found in English involving the spatial roles Location and
Route. In what follows, I will argue that the Location/Route ambiguity
actually does not involve a syncretism at all.
In his paper on spatial prepositions in English, (Svenonius 2010) establishes
the class of so called PathPlace prepositions, listed in (30).
(30) PathPlace prepositions: around, through, across, along, over, un-
der
The reason for the label “PathPlace” is that these preposition combine
properties of Path prepositions with properties of Place prepositions.
More precisely, they can be used both in Route-path and in Locative
contexts, see (31a) and (31b), respectively. It is then conceivable that
the PathPlace prepositions can lexicalize a Route projection and a Place
projection.
(31) a. We walked through the forest. (Route path)
b. The pencil is all the way through the cushion. (Location)
It is important that the only path which through and the other PathPlace
prepositions can express are Route paths and no other. Given that in
the Path decomposition I proposed, the heads Place and Route are separated
by the Goal and Source heads (see (32)), it is expected that the
intermediate Goal and Source structures are also lexicalized by through.
This is clearly a bad prediction. First, because it is not true, see (33).
Second, because I argued in Section 9.2.2 that a Goal=Source syncretism
is impossible.
242 SYNCRETISMS 9.3
(32) The syntactic structure of a transitional (bounded) Route path
RouteP
Route SourceP
Source GoalP
Goal PlaceP
(33) a. We walked into the forest. (Goal)
b. We walked out of the forest. (Source)
The problem can be resolved once we look more carefully at the PathPlace
prepositions and the kind of Route path and Location they denote.
In what follows, I argue that PathPlace preposition in fact do not lexicalize
PlaceP even on the static reading and the locative interpretation
they receive comes from elsewhere.
The first important observation is that the PathPlace class contains
prepositions which can express an non-transitional (unbounded) Route
path, either as the only path they express (like along, see (34)), or together
with a transitional (bounded) Route path (see (35), data from Piñón
1993).
(34) We walked along the river {for an hour, *in an hour}.
(35) a. The insect crawled through the tunnel {in two hours, for
two hours}.
b. Mary limped across the bridge {for ten minutes, in ten min-
utes}.
In order to remind the reader, I present the structure of a non-transitional
Route path below, repeating it from (18c) in Chapter 4. It is the same as
the structure of a transitional Route paths (see (32)), plus the additional
Scale head on top.
9.3 TYPOLOGICAL STUDIES 243
(36) Non-transitional (unbounded) Route path
ScaleP
Scale RouteP
Route SourceP
Source GoalP
Goal PlaceP
Thus, all PathPlace preposition can spell out the structure in (36).
The second important observation pertains to the type of Location
expressed by the members of this preposition class. Consider the data in
(37) taken from Svenonius (2010).
(37) a. The pencil is all the way through the cushion.
b. There is a fence around the house.
c. We found a log across the stream.
d. The cloth lay over the table.
The locative interpretations of the English sentences above are “paraphrasable
as ‘occupying the whole of a path’—for example, a pencil which
is through a cushion occupies the whole of the path which goes through
that cushion” (Svenonius 2010). Svenonius calls this type of location
extended location.
In sum, all PathPlace prepositions denote non-transitional Route
paths and the Location they express is an extended Location. The issue
which arises is what syntactic structure underlies the extended Locations.
Recall from Chapter 2 that non-transitional Route paths impose the
same locative condition on all the points of the path. For instance, any
of the points in the path along the river in (34) are seen as being at the
river (Zwarts 2008a).
(38) along-Path
+ + + + + + + + +
0 1 Zwarts (2008a)
Hence, a non-transitional Route path is more like a stative location
than a dynamic path involving change. A non-transitional Route marker
is, however, different from a simple locative marker in that the non-
244 SYNCRETISMS 9.4
transitional Route path picks out a contiguous sequence of points in the
vector space, while the latter picks out one or more points, which are not
necessarily contiguous. Therefore, I suggest that non-transitional Routes
are in fact extended Locations, although they contain path structure.
The equation of non-transitional Route paths with extended Locations
removes the problem posed by the Route=Location syncretism to
the exclusion of Goal and Source. Since non-transitional Route paths
and extended Locations are the same thing and correspond to the same
syntactic structure in (36), then any preposition which can express the
former can also express the latter. Further, there is the prediction that
prepositions that denote only transitional Route paths cannot be interpreted
as extended Locations. The reason is that such prepositions are
lexically specified to spell out the structure in (32) and consequently
cannot lexicalize the Scale head. This prediction is obeyed by the English
preposition past which expresses a transitional Route path only and
consequently has no extended Location reading.
The proposal that non-transitional Route paths and extended Locations
correspond to the same syntactic structure predicts that any language
with a non-transitional Route path will allow a static reading for it
of the type found in the English examples in (37). This prediction seems
to be borne out, with the additional peculiarity that many languages
cannot express extended Location by using copula verbs like to be, but
need a (fictive) motion verb (translating as go), or posture verbs (lie,
stand, etc.) in such constructions. Consider, for instance, the Norwegian
examples below (provided by Tarald Taraldsen, p.c.).
(39) a. Bro-en
bridge-def
går
goes
over
over
elv-en.
river-def
‘The bridge is/goes over the river.’
b. Stokk-en
log-def
ligger
lies
over
over
elv-en.
river-def
‘The log is across the river.’
Thus, in English, the copula to be can mediate the mapping of the
extended Location (or non-transitional Route path) onto the extent of
the Figure (cf., Gawron 2006). In Norwegian, this mapping requires verbs
like go and lie, which presumably serve the same purpose as the English
copula, given the lack of an event reading in (39).
9.4 CONCLUSION 245
9.4 Conclusion
In this chapter, I analyzed the logically possible syncretism patterns involving
the spatial roles Location, Goal, Source and Route. I argued, following
Bobaljik (2007) and Caha (2008), that syncretisms are restricted
to targeting only structurally adjacent heads. In this way I eliminated
six syncretism patterns (Type 5-11) in Table 9.3. I also suggested that
a Goal=Source syncretism should be impossible, although the two heads
are adjacent, because of the specific semantics I proposed for the Source
head. This excluded four more syncretism patterns (Type 12-15).
I tested the predictions against cross-linguistic data. I performed a
typological investigation, which by and large confirmed the predictions:
very few languages are claimed to exhibit the forbidden syncretism patterns.
The results of the typological survey, however, were not as neat as
desired, since a couple of languages, although very few in number, were
claimed to exhibit impossible syncretisms. Upon a closer data analysis,
I suggested, however, that these syncretisms were not real, because they
did not involve genuinely ambiguous spatial markers.
The important result of the investigation in this chapter is that there
undoubtedly are cases of real Location=Goal and Source=Route syncretisms,
while no convincing instances of the Goal=Source and *ABAtype
of syncretisms were found. This asymmetry in the syncretisms distribution
across languages lends support to the decomposed Path structure
I defend, as it reflects the adjacency relation between the various
Path heads and can be connected to their semantics.
246 SYNCRETISMS 9.4
Chapter 10
Conclusion
10.1 Introduction
In the past decade, linguistic research has taken a big interest in the
syntax and semantics of directed spatial expressions, resulting in a multitude
of scientific studies: papers (van Riemsdijk and Huybregts 2002,
Kracht 2002, Zwarts 2005; 2008a), doctoral dissertations (Asbury 2008,
Gehrke 2008, Lestrade 2010, Radkevich 2010), and specialized volumes
(Svenonius and Pantcheva 2006, Asbury et al. 2008, van der Auwera
2010, Cinque and Rizzi 2010), containing influential papers on the topic
(Koopman 2000, republished in Cinque and Rizzi 2010, Svenonius 2006,
den Dikken 2010, to mention some of them). The present work is a
contribution to this recent body of publications. It follows the trend towards
finer-grained syntactic representations and decomposes the Path
head into several projections.
In addition to articulating the internal structure of Path, this dissertation
contributes to the development of the Nanosyntax model of grammar
(Starke 2005-2009, Ramchand 2008b, Bašić 2007, Fábregas 2007,
Abels and Muriungi 2008, Muriungi 2008, Lundquist 2008, Caha 2009b,
Taraldsen 2010, Pantcheva 2010, for a representative collection of papers
see Svenonius et al. 2009). This is the framework I adopt to deal with the
highly fine-grained internal structure of Path and in doing this I develop
a detailed Spell-out algorithm compatible with the main tenets of the
theory.
The main ideas of this thesis are summarized in this chapter. At the
end of it, I discuss some topics for future research, emanating from the
particular proposals made in the dissertation.
247
248 CONCLUSION 10.2
10.2 Summary of the thesis
I start my investigation of the internal structure of Path by exploring in
Chapter 2 what kinds of spatial paths are found in languages worldwide,
with the aim to establish the core properties that distinguish them
from each other. I suggest that there are three such properties: orientation,
transition and delimitation.
The first property leads to a classification of paths into non-oriented
and oriented. Non-oriented paths, usually referred to as Route paths
(via), have no inherent directionality. Oriented paths involve directionality,
which is the result of there being an asymmetry between the two
extreme points of the path — a particular location holds of one of the
extremes and does not hold of the other. Depending on which one of
the extreme points is subject to a particular locative condition, we distinguish
Goal paths (to), where the location holds of the endpoint, and
Source paths (from), where the location holds of the starting point.
The next property, transition, reflects the fact that some paths have
a two or three-stage structure, which contains “positive phases” (where
a certain location holds) and “negative phases” (where the location does
not hold). The term transition refers to the passage from one phase of
the path to the next phase. I called paths with transitions transitional
paths. Transitional Goal paths (to) are paths where there is a transition
between the first, negative, phase of the path to the second, positive,
phase of the path. Transitional Source paths (from) are the reverse of
transitional Goal paths: the first phase is positive and there is a transition
to the second, negative, phase. Finally, transitional Route paths (via)
involve two transitions and the phase “in the middle” is the positive phase.
Paths without transition are non-transitional paths. Non-transitional
Goal paths (towards) thus have no transition between a negative and
a positive stage, but still involve an asymmetry in the location of their
extreme points, such that the endpoint is nearer to the location than the
starting point. The reverse holds of non-transitional Source paths (away
from). Finally, non-transitional Route paths (along) impose the same
locative condition on all points of the path.
The last property, delimitation, applies only to transitional oriented
paths and indicates a boundary of the path: an upper bound for delimited
Goal paths (up to) and a lower bound for Source paths (starting
from).
In Chapter 3, I lay out the basic theoretical assumptions concerning
syntax and morphology and the relationship between the two. Following
work on cartography (Cinque 1999, Rizzi 1997) and Nanosyntax (Starke
10.2 SUMMARY OF THE THESIS 249
2005-2009, Caha 2009b), I assume that each morpheme corresponds to
at least one terminal head in the syntactic structure, but possibly more
than one. Thus, the identification of a morpheme in a given language
is evidence for the presence of a corresponding syntactic head. Further,
following Chomsky (2001), I assume that all languages share the same
underlying syntactic structure. As a consequence, once the existence of
a terminal X has been established for one language on the basis of the
morphological data provided by this language, the result carries over to
all other languages. The X projection becomes universally present, also
in those languages which do not provide morphological evidence for its
existence. Finally, I assume, following Kayne (1994), that the syntactic
structure is universally ordered such that specifiers precede heads and
heads precede complements. All other orders are derived by movement.
Having examined what types of paths there are in Chapter 2 and
guided by the system of assumptions made in Chapter 3, I proceed to
Chapter 4 where I investigate how they are expressed in various languages.
The data analysis leads to a novel generalization: each of the
divisions in the path typology is brought about by a special head in the
syntactic structure underlying the various directional expressions. The
establishment of each syntactic head is motivated by the morphological
composition of the examined directional expressions. For instance, in a
number of languages Source paths are built on top of Goal paths. This
is achieved by the addition of a special element, for which there is evidence
that it is an independent morpheme, thus excluding an accidental
containment relationship. Similarly, there are languages where Route
paths are formed on the basis of Source paths, and non-transitional and
delimited paths are formed on the basis of the corresponding bounded
path.
The result from the data investigation leads me to propose that the
Path head is not a unique projection hosting all kinds of directional
elements, no matter what path they express, but can be decomposed into
several heads: Place, Goal, Source, Route, Scale, and Bound[ed]. On the
basis of the morphological analysis, I suggest that the Place, Goal, Source
and Route heads are rigidly ordered in the syntactic hierarchy such that
the Route head dominates a SourceP, the Source head dominates a GoalP
and the Goal head takes as a complement a Place projection. I argue
that each type of path corresponds to a particular syntactic structure.
250 CONCLUSION 10.2
(1) a. Goal path
GoalP
Goal PlaceP
Place ...
b. Source path
SourceP
Source GoalP
Goal PlaceP
Place ...
c. Route path
RouteP
Route SourceP
Source GoalP
Goal PlaceP
Place ...
The Scale and Bound heads, on the contrary, do not have a fixed
position and can appear on top of any other Path head, thus deriving
non-transitional and delimited Goal, Source and Route paths, respectively.
As an example, I present in (2) the structures underlying a nontransitional
and a delimited Goal path.
(2) a. Goal path
GoalP
Goal PlaceP
Place ...
b. Delimited Goal
BoundP
Bound GoalP
Goal PlaceP
Place ...
c. Non-transitional Goal
ScaleP
Scale GoalP
Goal PlaceP
Place ...
Since I assume that the decomposed Path structure is universal, it
applies also in languages where there is no morphological containment
relationship between the various kinds of path expressions.
Chapter 5 deals with the semantic function of each of the heads I
propose and lends further support to the universality of the Path structure,
as it accounts for the semantic compositionality of paths crosslinguistically.
I suggest that the semantic contribution of the Goal head
is to denote transition to the spatial region encoded by the static locative
projection in its complement. The Source head is the locus of a
reversal operation, which reverses the orientation of the Goal path in its
complement position. The Route head introduces a second transition to
the first (positive) phase of the Source path below it. The Scale head
picks out an interval from the path structure below it, which is to the
left and to the right of the transition. Consequently, when applying to a
10.2 SUMMARY OF THE THESIS 251
Route path, the selected interval consists of the positive phase, where the
locative relationship with the Ground holds. When applying to a Goal
and Source path, the selected interval contains only “negatively located”
points, that is, points where the locative relationship with the Ground
does not hold. In addition, the Scale head imposes an order in the selected
negative interval such that the “closer” a point has been to the
positive extreme point in the original Goal or Source path, the shorter
the distance between the Figure and the Ground is at that point. Finally,
the Bound head picks out an interval from the path structure to which
it applies such that this interval falls in the negative phase and ends or
begins with a unique positive point to derive delimited Goal or Source
paths, respectively. This unique positively located extreme point of the
path is interpreted as the limit for movement.
In Chapter 6, I address the question of how the fine-grained Path
structure I propose is lexicalized. I discuss the apparent mismatch in
various languages between the number of heads in the structure and the
number of morphemes that spell them out. I adopt and present the
Nanosyntax theory of grammar and show how this approach to language
captures the lexicalization of the proposed syntactic structure. I assume
the model of cyclic Phrasal Spell-out proposed by Starke (2005-2009)
where lexicalization targets both terminal and non-terminal nodes and
proceeds bottom-up. Its cyclic nature is a consequence of the assumption
that the lexicon is consulted after each External Merge operation. By
contrast, there is no lexical access after an Internal Merge operation. In
addition, I adopt the requirement that every feature in a cycle has to
be lexicalized before the derivation can proceed to the next cycle, thus
adapting the Exhaustive Lexicalization principle of Ramchand (2008a) to
a cyclic Spell-out system.
In Nanosyntax, lexical entries store fragments of syntactic trees (as
well as phonological and conceptual information). These lexically stored
trees are matched to the nodes in the syntactic tree which are to be lexicalized.
The match need not be perfect in order for a lexical entry to
be eligible for insertion: it is enough that the node to be lexicalized is
a sub-constituent of the lexically stored tree, formulated as the Superset
Principle (Starke 2005-2009, Caha 2009b). Thus, the tree stored in a
lexical entry can be bigger than the syntactic structure this entry lexicalizes.
This opens the possibility for the existence of multiple entries that
match a given node, because there can be many entries whose lexically
stored trees contain this node as a sub-constituent. The competition
between them is resolved by the Elsewhere Condition (Kiparsky 1973),
252 CONCLUSION 10.2
which requires the most specific item to be chosen. Thus, when two or
more items compete for insertion at a given node, the one which wins is
the one that leaves the least number of unused features.
In Chapter 7, I elaborate on the lexicalization of syntactic structure
and test the system against empirical data from three languages:
Karata, Uzbek and Finnish. I adopt the idea of Spell-out driven movement,
originally proposed by Starke (2011), according to which a lexical
entry whose lexically stored tree is not a good match for a given node
can trigger an evacuation movement of the obstructing material in order
to create the right configuration for matching and subsequently lexical
insertion. To allow for this possibility, I assume, following Caha (2009b),
that traces are ignored for the purposes of matching.
In addition to creating the right configuration for lexicalization, the
Spell-out driven movement creates new nodes in the syntactic structure
— the higher segmental nodes resulting from the adjunction of the extracted
material to the category which is being lexicalized. Allowing for
these nodes to be targeted by lexical insertion has beneficial effects, as
it makes it possible to account for the allomorphy found in Finnish Goal
expressions on a purely syntactic basis. The implementation of this idea
in the system of assumptions made in Chapter 6 motivates the proposal
that the Spell-out driven movement triggered in a given cycle takes place
in the next cycle. As regards the specific shape of the lexical entries,
I assume that their lexically stored trees need not contain information
about the exact number of segments a category consists of. The lexical
entries store information only about categories.
In the remainder of the chapter, I conduct a detailed investigation of
what constraints apply to the movement triggered by the need to spell
out and argue for a number of restrictions. First, I suggest, following
the insights in Cinque (2005), that the constituent that extracts in order
to make the matching of a given node possible must contain the head
noun. Second, regarding the landing site for the evacuated material, I
assume Shortest move — the moved material adjoins right above the node
where insertion takes place. Finally, I assume that when more than one
movement is necessitated by the need to spell out, syntax performs them
in an order opposite to the order in which they are triggered. At the end
of the chapter, I run a detailed lexicalization of the spatial expression in
Finnish and show how the assumed model accounts for its intricacies.
The lexicalization model developed in Chapters 6 and 7 allows for a
single morpheme to spell out one or more syntactic terminals. Combined
with the fine-grained internal structure of Path argued for in Chapter
10.2 SUMMARY OF THE THESIS 253
4, this predicts various possibilities to “partition” the syntactic structure
underlying a given path, depending on how many morphemes are employed
and which particular heads they spell out. Thus, for example, a
Source structure can be spelled out by a single lexical entry in one language,
but by two entries in another language. Moreover, languages that
have bi-morphemic Source expressions can place the “cut” differently. For
instance, one language can have a morpheme border between the heads
Place and Goal, while another can have a morpheme border between
Goal and Source. In Chapter 8, I lay out all 28 logically possible partitionings
of Route, Source and Goal paths predicted to exist and present
real language data that exemplify eighteen partitioning types. Despite
the lack of data exemplifying the remaining ten possible partitionings,
the eighteen attested partitioning types provide additional evidence for
the decomposition of the Path head into several heads and the proposal
that each kind of path is associated with a unique syntactic structure.
In Chapter 8, I also discuss the phenomenon of spurious syncretism,
which arises in some languages due to the particular partitionings of
paths they employ. Informally put, spurious syncretisms involve two
(or more) structures in a subset-superset relationship, where the smaller
structure is lexicalized by a given entry a, and the bigger structure is
lexicalized by a combination of the same entry a plus a “supporting”
entry b. On the face of it, it then appears that the lexical item a is
ambiguous between two (or more) notions. However, the ambiguity is
not real, since one of these notions is expressed with the help of the
“supporting” element b. I analyze several cases of spurious syncretisms
and show that, in many languages, the structure underlying directional
expressions is lexicalized by the joint collaboration of adpositional and
verbal elements, which leads to such spurious syncretism and reflects the
various ways to partition the structure.
Real syncretisms are the topic of Chapter 9. Real syncretisms involve
two (or more) structures in a subset-superset relationship, where
the smaller structure is lexicalized by an entry a and, importantly, a also
spells out the bigger structure without the support of other entries. This
is possible by virtue of the Superset Principle which allows for a given
lexical item to lexicalize structures that are as big as or smaller than the
syntactic tree stored in its entry. In this chapter, I analyze the logically
possible syncretisms involving the spatial roles Location, Goal, Source,
and Route. I propose two constraints on syncretisms: one motivated by
the particular semantics of the heads I assume, the other resulting from
the lexicalization model applied to the suggested syntactic structure for
254 CONCLUSION 10.3
paths.
The first restriction disfavors a syncretism between the spatial roles
Goal and Source. The justification behind it is that these two paths are
construed as exact opposites of each other — Source paths are derived
by the application of a reversal operation to a Goal path. I then suggest
that it is pragmatically unacceptable that a given lexical item expresses
two opposite notions.
The second constraint, which has a syntactic rationale, states that
syncretisms are restricted to structurally adjacent heads, thus banning
a paradigm of the type ABA, where a lexical item A spells out a given
structure [X [Y [Z]]], B spells out a smaller structure [Y [Z]], and A again
spells out the smallest structure [Z] (Bobaljik 2007, Caha 2009b). The
*ABA constraint prohibits syncretisms of Location and Source to the
exclusion of Goal (Location=Source=Goal), Location and Route to the
exclusion of Goal and Source (Location=Route=Goal & Source), and
Goal and Route to the exclusion of Source (Goal=Route=Source).
The rest of the chapter focuses on testing the predictions. I first
examine several typological studies and conclude that there is indeed
an asymmetry in the cross-linguistic distribution of syncretism patterns,
such that the ones predicted to be possible are common, while the ones
predicted to be impossible are very rare, arguably non-existent. I then
present some languages which instantiate the syncretism patterns predicted
to be possible. Afterwards, I turn to the impossible syncretisms
and discuss languages which have been claimed to have them. Upon a
closer data analysis, I conclude that the offending syncretisms are not
real but spurious, since they all seem to require a supporting element to
express one of the spatial roles.
10.3 Directions for future research
This dissertation aims to develop an empirically well-grounded and theoretically
solid account for the diversity of directional spatial expressions
across languages. Still, some issues were inevitably left open. In this
section, I briefly present three topics for future research.
The first topic relates to the necessity for a strong empirical confirmation
of the predictions discussed in Part III. In Chapter 8, I lay out the
28 possible partitionings of paths allowed by the model, but provide real
language data for only 18 of them. Similarly, in Chapter 9, I discuss the
four syncretisms patterns predicted to be available, but do not illustrate
one of them (Location=Goal=Source=Route). The finding of language
10.3 DIRECTIONS FOR FUTURE RESEARCH 255
data which exemplify all the possible lexicalization and syncretism patterns
will provide important support for the ideas in this thesis. As I see
it, it will require extensive fieldwork, since grammar descriptions hardly
ever include Route paths in the presentation of directional expressions
and rarely control for spurious syncretisms.
The second topic for future research emerging from the ideas in this
thesis concerns the incorporation of the suggested Path hierarchy into the
Case hierarchy proposed by Caha (2009b). In his doctoral dissertation,
Caha (2009b) convincingly argues for a decomposition of case into a
fixed hierarchy of universal features, where each feature corresponds to
a terminal node in a syntactic tree. The individual cases correspond to
phrasal constituents built out of these terminals, as presented in (3).1
(3) The Case hierarchy (Caha 2009b)
Comitative
Com Instrumental
Ins Dative
Dat Genitive
Gen Accusative
Acc Nominative
Nom DP
Paths are often expressed by case marking across languages. Translating
the Path hierarchy into a hierarchy of spatial cases, delivers the structure
in (4), where I use the labels Ablative, Allative and Locative to refer to
cases marking Source, Goal and Location in general.
1
Caha (2009b) labels the individual features A, B, C, etc. For ease of exposition,
I label them here Nom, Acc, Gen, etc. to facilitate the association of a given feature
to the case it yields.
256 CONCLUSION 10.3
(4) The Spatial Case hierarchy
Ablative
Source Allative
Goal Locative
Place DP
Conceivably, the two hierarchies belong to the same functional sequence
(fseq). This motivates an attempt to merge them together. The *ABA
generalization, according to which syncretisms target adjacent heads,
provides a useful tool to determine the positioning of the spatial cases
among the case projections in Caha’s hierarchy (in fact, this is one
of the strategies used by Caha to establish the sequence in (3)). For
instance, the syncretism between the Dative and the Allative is very
widespread cross-linguistically. In fact, often grammars do not list the
Allative as a separate case but attribute the Goal function to the Dative.
A Dative=Allative syncretism is found in many Caucasian languages
(Tsez, Khvarshi, Avar, Lak), Turkic languages (Uzbek, Turkish),
Quechua, Tamil, Cebaara, Armenian, Finish, Itelmen, Chukchi,
and others. The Dative case is also often syncretic with the Locative:
this happens in Japanese, Tshangla, Latin, and other languages.
One can even find languages, where the two syncretisms occur simultaneously
— the Mongolic languages which are claimed to have a Locative=Dative=Allative
syncretism (Janhunen 2003). Thus, it is a likely
hypothesis that the place of the Dative in the case-fseq is between the
Locative and the Allative.
(5)
... Allative
Source Dative
Dat Locative
Loc ...
The hierarchy in (5) suggests that if a language has a Locative=Allative
syncretism, then the Dative should also be syncretic with the Loca-
10.3 DIRECTIONS FOR FUTURE RESEARCH 257
tive/Allative.2
In addition, one expects to find languages where the
Allative is built on the basis of the Dative by the addition of an independent
morpheme. A potential example of such a language is Bidubi
(Blake 1977), where the Dative case ending -ku is contained in the Allative
ending -kutu.
Apart from the Locative=Dative syncretism mentioned in the preceding
paragraph, another syncretism involving the Locative is the Locative=Genitive
syncretisms found in Tschangla, Latin, Sanskrit and Baagangji,
among other languages. This suggests a positioning of the Locative
between the Dative and the Genitive projections, as depicted in (6).
(6)
... Dative
Dat Locative
Loc Genitive
Gen ...
Regarding the position of the Ablative case in the extended case hierarchy,
the rather common Ablative=Instrumental syncretism (Creissels
2008, Noonan and Mihas 2010) suggests that the Ablative and the Instrumental
are adjacent projections in the case fseq. Further, given that
many languages syncretize Instrumental and Comitative to the exclusion
of the Ablative, the *ABA constrains requires that the Ablative does not
intervene between the two projections. Hence, we can hypothesize that
the position of the Ablative is below the Instrumental, as shown in (7).
(7) Comitative
Com Instrumental
Ins Ablative
Source ...
The extended case hierarchy which emerges once we merge Caha’s case
2
A different implication has been proposes by Blansitt (1988), according to whom
if the Dative is syncretic to the Locative, then so must be the Allative, suggesting
that the Allative is to be placed between the Locative and the Dative. To the extent
that I have been able to check the data examined by Blansitt, the generalization does
not hold once spurious syncretisms are excluded.
258 CONCLUSION 10.3
decomposition with the decomposed Path is presented in (8).
(8) The Extended Case hierarchy
Comitative
Com Instrumental
Ins Ablative
Source Allative
Goal Dative
Dat Locative
Place Genitive
Gen Accusative
Acc Nominative
Nom DP
The confirmation of the structure in (8) requires an in-depth study of
the above-mentioned syncretisms aiming their verification, as well as an
investigation of other syncretisms the spatial cases participate in. If
correct, the Extended Case hierarchy raises a lot of questions, two of
which I formulate below.
• How do we capture the meaning distinctions in spatial cases expressing
the same general type of path? For instance, many languages
have an Illative case (to in) and an Allative case (to at),
but so far I have not found evidence for one of the two cases being
“bigger” than the other. It is then conceivable that the difference
in the meaning is encoded lower down in the structure, e.g., in the
AxPart head.
• What implication does the Extended Case hierarchy have for languages
that use spatial prepositions? Caha (2009b) demonstrates
that the Case hierarchy accounts for the cross-linguistic variation
10.3 DIRECTIONS FOR FUTURE RESEARCH 259
in the amount of case marking: if a language has case marking
for a case X, then it will use case marking for all cases which are
lower than X on the hierarchy and prepositions for the ones that
are higher. Hence, if a language has a Source preposition, then we
predict that it cannot use case marking to express Comitative.
The third class of issues which opens a whole new research venue pertains
to the Spell-out driven movement I adopted in the development of
the detailed lexicalization algorithm within the Nanosyntax theory. In
particular, it would be interesting to set the Spell-out driven movements
against the background of the “classical” (feature-driven) movements (like
wh-movement, for example) and explore the question of how to distinguish
them from each other. A potential line of research one can pursue
stems from a suggestion, made by Starke (2011). Starke proposes that
wh-movement is universal, i.e., applies in all languages, but Spell-out
driven movement is “parametric” in that it is contingent on the lexical
resources of a given language. Whether the wh-movement is overt or
covert, however, depends on the particular size of the trees stored in the
wh-entries of the language. Needless to say, I leave this topic for future
research.
260 CONCLUSION 10.3
Appendix: Language sample
and references
Altaic (genus given in parentheses)
Nanai (Tungusic) Avrorin (1968)
Ulcha (Tungusic) Sunik (1968)
Chuvash (Turkic) Clark (1998)
Turkish (Turkic) Göksel and Kerslake (2005)
Uzbek (Turkic) Boeschoten (1998)
Buryat (Mongolic) Skribnik (2003)
Dagur (Mongolic) Tsumagari (2003)
Kalmuck (Mongolic) Bläsing (2003)
Khalkha (Mongolic) Svantesson (2003)
Ordos (Mongolic) Georg (2003)
Arawakan
Yanesha (Arawakan) Duff-Tripp (1997)
261
262 APPENDIX
Austronesian
Iaai (Oceanic) Lynch (2002a)
Jabêm (Oceanic) Ross (2002b)
Kaulong (Oceanic) Ross (2002c)
Longgu (Oceanic) Hill (2002)
Mwotlap (Oceanic) Crowley (2002a)
Niuafo‘ou (Oceanic) Early (2002)
Raga (Oceanic) Crowley (2002b)
Sye (Oceanic) Crowley (2002c)
Tobati (Oceanic) Donohue (2002)
Xârâcúú (Oceanic) Lynch (2002b)
‘Ala‘ala (Oceanic) Ross (2002a)
Cholon
Cholon (Cholon) Adelaar (2004)
Chukotko-Kamchatkan
Itelmen (Southern Chukotko-Kamchatkan) Volodin and Žukova (1968)
Chukchi (Northern Chukotko-Kamchatkan) Skorik (1968)
Dravidian
Telugu (South-Central Dravidian) Krishnamurti (1998)
Gond.i (South-Central Dravidian) Steever (1998)
Jivaroan
Jivaroan (Jivaroan) Juank (1982)
Indo-European
Breton (Celtic) Ternes (1992)
Scottish Gaelic (Celtic) MacAulay (1992)
Mayan
Yukatek Maya (Mayan) Bohnemeyer and Stolz (2006)
APPENDIX: LANGUAGE SAMPLE AND REFERENCES 263
Nakh-Dagestanian
Akhvakh (Avar-Andic-Tsezic) Magomedbekova (1967a)
Avar (Avar-Andic-Tsezic) Charachidzé (1981)
Bagvalal (Avar-Andic-Tsezic) Gudava (1967a)
Bezhta (Avar-Andic-Tsezic) Bokarev and Madieva (1967)
Botlikh (Avar-Andic-Tsezic) Gudava (1967b)
Chamalal (Avar-Andic-Tsezic) Magomedbekova (1967b)
Godoberi (Avar-Andic-Tsezic) Gudava (1967c)
Hinukh (Avar-Andic-Tsezic) Bokarev (1967a)
Hunzib (Avar-Andic-Tsezic) Bokarev (1967b)
Karata (Avar-Andic-Tsezic) Magomedbekova (1971)
Khvarshi (Avar-Andic-Tsezic) Bokarev (1967c)
Tindi (Avar-Andic-Tsezic) Gudava (1967d)
Tsez (Avar-Andic-Tsezic) Comrie and Polinsky (1998)
Icari Dargwa (Lak-Dargwa) Sumbatova and Mulatov (2003)
Lak (Lak-Dargwa) Zhirkov (1955)
Archi (Lezgic) Haidakov (1967)
Khinalug (Lezgic) Desheriev (1967)
Lezgian (Lezgic) Haspelmath (1993)
Rutul (Lezgic) Djeitanishvili (1967)
Tabasaran (Lezgic) Magometov (1965)
Udin (Lezgic) Panchvidze and Djeirashvili (1967)
Niger-Congo
Zulu (Bantu) Taylor (1996)
Wan (Mande) Nikitina (2009)
Nivkh
Nivkh (Nivkh) Gruzdeva (1998)
Quechuan
Quechua (Quechuan) Cole (1985), Jake (1985)
264 APPENDIX
Sino-Tibetan
Garo (Baric) Burling (2003)
Chantyal (Bodic) Noonan (2003a)
Kham (Bodic) Watters (2002)
Nar-Phu (Bodic) Noonan (2003b)
Classical Tibetan (Bodic) Beyer (1992)
Tshangla (Bodic) Andvik (2003)
Cogtse-Gyarong (rGyalrong) Nagano (2003)
Hakha Lai (Kuki-Chin) Peterson (2003)
Akha (Lolo-Burmese) Hanssson (2003)
Burmese (Lolo-Burmese) Wheatley (2003)
Lahu (Lolo-Burmese) Matisoff (1973)
Qiang (Qiangic) LaPolla (2003)
Uralic
Estonian (Finnic) Viitso (1998)
Finnish (Finnic) Sulkala and Karjalainen (1992)
Mari (Finnic) Kangasmaa-Minn (1998)
Mordva (Finnic) Zaicz (1998)
Permyak (Finnic) Lytkin (1962), Riese (1998)
Komi (Finnic) Hausenberg (1998)
Veps (Finnic) Zajtseva (1981), Tikka (1992)
Khanty (Ugric) Abondolo (1998)
Mansi (Ugric) Keresztes (1998)
Nenets (Samoyedic) Salminen (1998)
Uru-Chipaya
Uru (Uru-Chipaya) Vellard (1967)
Isolates
Basque Hualde and de Urbina (2003)
Ibarretxe-Antuñano (2004)
Mapudungun Adelaar (2004)
Wälchli and Zuñiga (2006)
Bibliography
Abels, Klaus and Peter Kinyua Muriungi. 2008. The focus marker in
Kîîtharaka: Syntax and semantics. Lingua 118 5: 687–731.
Abondolo, Daniel. 1998. Khanty. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 358–
386. Routledge, London and New York.
Ackema, Peter and Ad Neeleman. 2007. Morphology = syntax. In The
Oxford Handbook of Linguistic Interfaces, edited by Gillian Ramchand
and Charles Reiss. Oxford University Press, Oxford.
Adelaar, Willem F. H. with the collaboration of Pieter C. Muysken. 2004.
The Languages of the Andes. Cambridge Language Surveys. Cambridge
University Press, Cambridge.
Adger, David. 2003. Core Syntax: A Minimalist Approach. Oxford University
Press, Oxford.
Andersen, Torben. 2002. Case inflection and nominal head marking in
Dinka. Journal of African Languages and Linguistics 23 1: 1–30.
Andrews, Avery. 1985. The major functions of the noun phrase. In Language
Typology and Syntactic Description, edited by Timothy Shopen,
vol. 1, pp. 62–154. Cambridge University Press, Cambridge.
Andrews, Avery. 1990. Unification and morphological blocking. Natural
Language & Linguistic Theory 8 4: 507–557.
Andvik, Erik. 2003. Tshangla. In The Sino-Tibetan Languages, edited
by Graham Thurgood and Randy J. LaPolla, pp. 439–455. Routledge,
London and New York.
Arsenjević, Boban. 2006. Inner Aspect and Telicity: The Decompositional
Nature of Eventualities at the Syntax-Semantics Interface. Ph.D. thesis,
Leiden University.
Asbury, Anna. 2008. The Morphosyntax of Case and Adpositions. Ph.D.
thesis, Utrecht University, Utrecht.
Asbury, Anna, Jakub Dotlačil, Berit Gehrke, and Rick Nouwen (eds.).
2008. Syntax and Semantics of Spatial P. No. 120 in Linguistik Aktuell.
John Benjamins, Amsterdam and Philadelphia.
265
266 BIBLIOGRAPHY
Asbury, Anna, Berit Gehrke, and Veronika Hegedűs. 2007. One size
fits all: Prefixes, particles, adpositions and cases as members of the
category P. In UiL OTS Yearbook 2006, edited by Cem Keskin, pp.
1–17.
Assadollahi, Ramin, Wilhelm Geuder, and Matthias Weisgerber. 2006.
‘Convergent’ vs. ‘divergent’ verbs and the asymmetry between source
and goal. Paper presented at Sinn und Bedeutung 11 in Barcelona,
September 2006.
van der Auwera, Johan (ed.). 2010. Linguistics: Special Issue on Spatial
Case, vol. 48. Mouton de Gruyter, Berlin/New York.
Avrorin, V. A. 1968. Æ Ò °× ° ÞÝ [The Nanai language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 5, pp. 129–148. Æ Ù [Nauka], Moscow.
Baerman, Matthew, Dunstan Brown, and Greville G. Corbett. 2005. The
Syntax-Morphology Interface: A Study of Syncretism, vol. 109 of Cambridge
Studies in Linguistics. Cambridge University Press, Cambridge.
Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation.
Linguistic Inquiry 16 3: 373–415.
Bašić, Monika. 2007. Serbian Ps with and without iz and the Superset
Principle. In Tromsø Working Papers on Language and Linguistics:
Nordlyd 34.2, Special issue on Space, Motion, and Result, edited by
Monika Bašić, Marina Pantcheva, Minjeong Son, and Peter Svenonius,
pp. 300–319. University of Tromsø, Tromsø. CASTL, Tromsø. Available
at http://www.ub.uit.no/baser/nordlyd/.
Beavers, John, Beth Levin, and Shiao Wie Tham. 2010. The typology of
motion expressions revisited. Journal of Linguistics 46: 331–377.
Bennet, David C. 1975. Spatial and temporal uses of English prepositions:
An essay in stratificational semantics, vol. 17 of Longman Linguistic
Library. Longman, London.
Beyer, Stephan V. 1992. The Classical Tibetan Language. SUNY Series
in Buddhist Studies. State University of New York Press.
Blake, Barry J. 1977. Case Marking in Australian Languages. No. 23 in
Linguistic Series. Australian Institute of Aboriginal Studies, Canberra.
Blake, Barry J. 1994. Case. Cambridge University Press, Cambridge,
second edn.
Blansitt, Edward L. 1988. Datives and allatives. In Studies in Syntactic
Typology, edited by Michael Hammond, Edith Moravcsik, and
Jessika R. Wirth, vol. 17 of Studies in Languages, pp. 173–191. John
Benjamins, Amsterdam/Philadelphia.
Bläsing, Uwe. 2003. Kalmuck. In The Mongolic Languages, edited by
BIBLIOGRAPHY 267
Juha Janhunen, Routledge Language Family Series, pp. 229–247. Routledge,
London and New York.
Bobaljik, Jonathan. 2000. The ins and outs of contextual allomorphy.
In University of Maryland Working Papers in Linguistics, edited by
Kleanthes K. Grohmann and Caro Struijke, vol. 10, pp. 35–71.
Bobaljik, Jonathan. 2007. On comparative suppletion. Ms., University
of Connecticut.
Boeschoten, Hendrik. 1998. Uzbek. In The Turkic Languages, edited
by Lars Johanson and Éva Ágnes Csató, Routledge Language Family
Descriptions, pp. 357–378. Routledge, London and New York.
Bohnemeyer, Jürgen. in prep. When going means becoming gone:
Framing motion as state change in Yukatek. Available at
http://www.acsu.buffalo.edu/∼jb77/CoL_model_chapter_yuk.pdf.
Bohnemeyer, Jürgen and Gabriela Pérez Báez. 2008. Object to path in
Mesoamerica: Semantic composition of locative and motion descriptions
in Yucatec Maya and Juchitán Zapotec. In Memoria del IX
Encuentro Internacional De Lingüística En El Noroeste, vol. 2, pp.
269–284. Editorial UniSon, Hermosillo, Sonora, Mexico.
Bohnemeyer, Jürgen and Christel Stolz. 2006. Spatial reference in
Yukatek Maya: A survey. In Grammars of Space, edited by Stephen C.
Levinson and David P. Wilkins, pp. 273–310. Cambridge University
Press, Cambridge.
Bokarev, Evgenij A. 1967a. ÒÙ × ° ÞÝ [The Hinukh language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 436–454. Æ Ù [Nauka],
Moscow.
Bokarev, Evgenij A. 1967b. ÙÒÞ × ° ÞÝ [The Hunzib language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 472–487. Æ Ù [Nauka],
Moscow.
Bokarev, Evgenij A. 1967c. ÀÚ ÖÜ Ò× ° ÞÝ [The Khvarshi language].
In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of
the USSR], edited by V.V. Vinogradov, vol. 4, pp. 421–435. Æ Ù
[Nauka], Moscow.
Bokarev, Evgenij A. 1967d. Þ ° ÞÝ [The Tzes language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 4, pp. 337–349. Æ Ù [Nauka], Moscow.
Bokarev, Evgenij A. and I. G. Madieva. 1967. Ø Ò× ° ÞÝ [The
Bezhta language]. In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the
Peoples of the USSR], edited by V.V. Vinogradov, vol. 4, pp. 455–471.
268 BIBLIOGRAPHY
Æ Ù [Nauka], Moscow.
Borer, Hagit. 2005a. In Name Only (Structuring Sense, vol. I). Oxford
University Press, New York.
Borer, Hagit. 2005b. The Normal Course of Events (Structuring Sense,
vol. II). Oxford University Press, New York.
Bošković, Željko. 2005. On the locality of left branch extraction and the
structure of NP. Studia Linguistica 59 1: 1–45.
Broschart, Jürgen and Carmen Dawuda. 1999. Beyond nouns and verbs:
A typological study in lexical categorization. Arbeiten des Sonderforschungsbereichs
282 “Theorie des Lexikons” 113.
Buck, Rosemary A. 1997. Words that are their opposites: Noun to verb
conversion in English. Word 48 1: 1–14.
Burling, Robbins. 2003. Garo. In The Sino-Tibetan Languages, edited
by Graham Thurgood and Randy J. LaPolla, pp. 387–400. Routledge,
London and New York.
Bye, Patrik and Peter Svenonius. 2011. Non-concatenative morphology
as epiphenomenon. To appear in The Morphology and
Phonology of Exponence, edited by Jochen Trommer, available at
http://ling.auf.net/lingBuzz/001099.
Caha, Pavel. 2008. The case hierarchy as functional sequence. In Scales,
edited by Marc Richards and Andrej L. Malchukov, no. 86 in Linguistische
Arbeits Berichte, pp. 247–276. University of Leipzig, Leipzig.
Caha, Pavel. 2009a. Classical Armenian declension. In Nordlyd 36.1:
Special issue on Nanosyntax, edited by Peter Svenonius, Gillian Ramchand,
Michal Starke, and Knut Tarald Taraldsen, pp. 77–112. University
of Tromsø, Tromsø. Available at www.ub.uit.no/munin/nordlyd/.
Caha, Pavel. 2009b. The Nanosyntax of Case. Ph.D. thesis, University
of Tromsø.
Caha, Pavel. 2010a. The German locative-directional alternation: A
peeling account. Journal of Comparative Germanic Linguistics 13 3:
179–223.
Caha, Pavel. 2010b. The parameters of case marking and spell-out driven
movement. In Linguistic Variation Yearbook 2010, edited by Jeroen
Van Craenenbroeck. John Benjamins.
Charachidzé, Georges. 1981. Grammaire de la langue Avar, vol. 38 of
Documents de linguistique quantitative. Éditions Jean-Favard, Paris.
Chen, Lian and Jiansheng Guo. 2009. Motion events in chinese novels:
Evidence for an equipollently-framed language. Journal of Pragmatics
41 9: 1749–1766.
Chomsky, Noam. 1986. Barriers. MIT Press, Cambridge, Massachusetts.
BIBLIOGRAPHY 269
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In
The View from Building 20: Essays in Linguistics in Honor of Sylvain
Bromberger, edited by Kenneth Hale and Samuel Jay Keyser, pp. 1–52.
MIT Press, Cambridge, Massachusetts.
Chomsky, Noam. 1995. The Minimalist Program. Current Studies in
Linguistics, Cambridge, Massachusetts (CSLing); 28. MIT Press, Cambridge,
Massachusetts.
Chomsky, Noam. 1998. Minimalist inquiries: The framework. MIT Occasional
Papers in Linguistics 15: 1–56.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A Life in
Language, edited by Michael Kenstowicz, pp. 1–52. MIT Press, Cambridge,
Massachusetts.
Chung, Inkie. 2007. Suppletive negation in Korean and Distributed Morphology.
Lingua 117 1: 95–148.
Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A CrossLinguistic
Perspective. Oxford University Press, New York.
Cinque, Guglielmo. 2005. Deriving Greenberg’s Universal 20 and its
exceptions. Linguistic Inquiry 36 3: 315–332.
Cinque, Guglielmo. 2009. The fundamental left-right asymmetry of natural
languages. In Universals of Language Today, edited by Sergio
Scalise, Elisabetta Magni, and Antonietta Bisetto, vol. 76 of Studies
in Natural Language and Linguistic Theory, pp. 165–184. Springer
Netherlands.
Cinque, Guglielmo and Luigi Rizzi. 2008. The cartography of syntactic
structures. In STiL (Studies in Linguistics). CISCL working papers on
language and cognition, edited by Vincenzo Moscati, vol. 2, pp. 43–59.
University of Siena, Siena.
Cinque, Guglielmo and Luigi Rizzi (eds.). 2010. The Cartography of
Syntactic Structure, vol. 6. Oxford University Press, Oxford.
Clark, Larry. 1998. Chuvash. In The Turkic Languages, edited by Lars
Johanson and Éva Ágnes Csató, Routledge Language Family Descriptions,
pp. 434–452. Routledge, London and New York.
Cole, Peter. 1985. Imbabura Quechua. Croom Helm, London.
Collins, Chris. 1997. Local economy. No. 29 in Linguistic Inquiry Monographs.
MIT Press, Cambridge, Massachusetts.
Comrie, Bernard. 1999. Spatial cases in Daghestanian languages.
Sprachtypologie und Universalienforschung 52 2: 108–117.
Comrie, Bernard and Maria Polinsky. 1998. The great Daghestanian
case hoax. In Case, Typology and Grammar: In honor of Barry J.
Blake, edited by Anna Siewierska and Jae Jung Song, pp. 95–114.
270 BIBLIOGRAPHY
John Benjamins, Amsterdam.
Corver, Norbert. 1990. The Syntax of Left Branch Extractions. Ph.D.
thesis, University of Tilburg.
Creissels, Denis. 2006. Encoding the distinction between location, source
and destination: a typological study. In Space in Languages, edited
by Maya Hickmann and Stéphane Robert, Typological Studies in Language,
Vol. 66, pp. 19–28. John Benjamins, Amsterdam/Philadelphia.
Creissels, Denis. 2008. Spatial cases. In The Oxford Handbook of Case,
edited by Andrej Malchukov and Andrew Spencer, pp. 609–625. Oxford
University Press, New York.
Crowley, Terry. 2002a. Mwotlap. In The Oceanic Languages, edited
by John Lynch, Malcolm Ross, and Terry Crowley, Curzon Language
Family Series, pp. 587–598. Curzon, Richmond, Surrey.
Crowley, Terry. 2002b. Raga. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 626–637. Curzon, Richmond, Surrey.
Crowley, Terry. 2002c. Sye. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 694–722. Curzon, Richmond, Surrey.
Csúcs, Sándor. 1998. Udmurt. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 276–
304. Routledge, London and New York.
Declerck, Renaat. 1979. Aspect and the bounded/unbounded
(telic/atelic) distinction. Linguistics 17 9/10: 761–794.
DeLancey, Scott. 2003. Classical Tibetan. In The Sino-Tibetan Languages,
edited by Graham Thurgood and Randy J. LaPolla, pp. 255–
269. Routledge, London and New York.
Depraetere, Ilse. 1995. (Un)boundedness and (a)telicity. Linguistics and
Philosophy 18 2: 1–19.
Desheriev, Yu. D. 1967. Ò ÐÙ × ° ÞÝ [The Khinalug language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 659–675. Æ Ù [Nauka],
Moscow.
Den Dikken, Marcel. 2010. On the functional structure of locative and
directional PPs. In The Cartography of Syntactic Structure, vol. 6,
edited by Guglielmo Cinque and Luigi Rizzi, pp. 74–126. Oxford University
Press, New York.
Dirr, A. 1905. Ö ÑÑ Ø Õ × °ÓÕ Ö Ø × Ö Ò× Ó Ó ÞÝ [A Grammar
Sketch of Tabasaran]. SMOMPK, Tiflis.
Djeitanishvili, E. F. 1967. ÊÙØÙÐ × ° ÞÝ [The Rutul language]. In
BIBLIOGRAPHY 271
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 580–590. Æ Ù [Nauka],
Moscow.
Donohue, Mark. 2002. Tobati. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 186–203. Curzon, Richmond, Surrey.
Dowty, David R. 1979. Word Meaning and Montague Grammar: The
Semantics of Verbs and Times in Generative Semantics and in Montague’s
PTQ. Reidel, Dordrecht.
Duff-Tripp, Martha. 1997. Gramática del idioma Yanesha’, vol. 47 of
Serie Lingüística Peruana. Instituto Lingüístico de Verano, Lima.
Early, Robert. 2002. Niuafo‘ou. In The Oceanic Languages, edited by
John Lynch, Malcolm Ross, and Terry Crowley, Curzon Language
Family Series, pp. 848–864. Curzon, Richmond, Surrey.
Emonds, Joseph E. 1994. Two principles of economy. In Paths towards
Universal Grammar: Studies in Honor of Richard S. Kayne, edited by
Luigi Rizzi Guglielmo Cinque, Jean-Yves Pollock and Raffaella Zanuttini,
pp. 155–172. Georgetown University Press.
Evans, Nicholas. 1995. A Grammar of Kayardild. Mouton de Gruyter,
Berlin.
Fábregas, Antonio. 2007. The Exhaustive Lexicalization Principle. In
Tromsø Working Papers on Language and Linguistics: Nordlyd 34.2,
Special issue on Space, Motion, and Result, edited by Monika Bašić,
Marina Pantcheva, Minjeong Son, and Peter Svenonius, pp. 165–
199. University of Tromsø, Tromsø. CASTL, Tromsø. Available at
http://www.ub.uit.no/baser/nordlyd/.
Fábregas, Antonio. 2009. An argument for phrasal spell-out: Indefinites
and interrogatives in Spanish. In Nordlyd 36.1: Special issue
on Nanosyntax, edited by Peter Svenonius, Gillian Ramchand, Michal
Starke, and Knut Tarald Taraldsen, pp. 129–168. University of Tromsø,
Tromsø. Available at www.ub.uit.no/munin/nordlyd/.
Feist, Michele I. 2000. On In and On: An Investigation into the Linguistic
Encoding of Spatial Scenes. Ph.D. thesis, Northwestern University,
Evanston, Illinois.
Feiz, Parastou. 2007. The Expression and Conceptualization of Motion
Through Space and Manner of Motion in Persian and English: A Comparative
Analysis. Ph.D. thesis, The Pennsylvania State University.
Fong, Vivianne. 1997. The Order of Things. What Syntactic Locatives
Denote. Ph.D. thesis, Stanford University.
Gawron, Jean Mark. 2006. Generalized paths. In Proceedings of the 15th
272 BIBLIOGRAPHY
Conference on Semantics and Linguistic Theory (SALT-XV), Ithaca,
New York. Cornell University, Department of Linguistics. CLC Pub-
lications.
Gehrke, Berit. 2008. Ps in motion: On the Semantics and Syntax of P Elements
and Motion Events. Ph.D. thesis, Utrecht University, Utrecht.
Georg, Stefan. 2003. Ordos. In The Mongolic Languages, edited by Juha
Janhunen, Routledge Language Family Series, pp. 193–209. Routledge,
London and New York.
Göksel, Aslı and Cecilia Kerslake. 2005. Turkish: A Comprehensive
Grammar. Routeledge, London and New York.
Grimshaw, Jane. 1991. Extended projections. Ms. Brandeis University.
Gruzdeva, Ekaterina. 1998. Nivkh, vol. 111 of Languages of the World.
Lincom Europa, Munich.
Gudava, T.E. 1967a. Ú Ð Ò× ° ÞÝ [The Bagvalal language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 351–367. Æ Ù [Nauka],
Moscow.
Gudava, T.E. 1967b. ÓØÐ Ü× ° ÞÝ [The Botlikh language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 293–306. Æ Ù [Nauka],
Moscow.
Gudava, T.E. 1967c. Ó Ó Ö Ò ° ÞÝ [The Godoberi language]. In
ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR],
edited by V.V. Vinogradov, vol. 4, pp. 307–321. Æ Ù [Nauka],
Moscow.
Gudava, T.E. 1967d. Ì Ò Ò× ° ÞÝ [The Tindi language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 4, pp. 368–383ß. Æ Ù [Nauka], Moscow.
Haidakov, S. M. 1967. ÖÕ Ò× ° ÞÝ [The Archi language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 4, pp. 608–626. Æ Ù [Nauka], Moscow.
Haiman, John. 1980. Hua: A Papuan Language of the Eastern Highlands
of New Guinea. No. 5 in Studies in Language Companion Series. John
Benjamins.
Hajek, John. 2006. Serial verbs in Tetun Dili. In Serial Verbs Constructions:
A Cross-Linguistic Typology, edited by Alexandra Y. Aikhenvald
and Robert M. W. Dixon, Explorations in Linguistic Typology,
pp. 239–253. Oxford University Press, Oxford.
Halle, Morris. 1997. Impoverishment and fission. In Papers at the Interface,
edited by Benjamin Bruening, Yoonjung Kang, and Martha
BIBLIOGRAPHY 273
McGinnis, no. 30 in MITWPL, pp. 425–449. MIT, Cambridge, MA.
Halle, Morris and Alec Marantz. 1993. Distributed Morphology and
the pieces of inflection. In The View from Building 20: Essays in
Linguistics in Honor of Sylvain Bromberger, edited by Kenneth Hale
and Samuel Jay Keyser, pp. 111–176. MIT Press, Cambridge, Mas-
sachusetts.
Halle, Morris and Alec Marantz. 1994. Some key features of Distributed
Morphology. In MIT Working Papers in Linguistics, edited by Andrew
Carnie and Heidi Harley (with Tony Bures), vol. 21, pp. 275–288. MIT
Press, Cambridge, Massachusetts.
Hanmagomedov, B. G.-K. 1967. Ì × Ö Ò× ° ÞÝ [The Tabasaran
language]. In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples
of the USSR], edited by V.V. Vinogradov, vol. 4, pp. 545–561. Æ Ù
[Nauka], Moscow.
Hansson, Gunnar Oláfur. 2007. Productive syncretisms in Saami inflectional
morphology. In Saami Linguistics, edited by Ida Toivonen and
Diane Carlita Nelson, pp. 91–135. John Benjamins, Amsterdam and
Philadelphia.
Hanssson, Inga-Lill. 2003. Akha. In The Sino-Tibetan Languages, edited
by Graham Thurgood and Randy J. LaPolla, pp. 236–2251. Routledge,
London and New York.
Harley, Heidi and Rolf Noyer. 1999. State-of-the-article: Distributed
morphology. GLOT International 4 4: 3–9.
Haspelmath, Martin. 1993. A Grammar of Lezgian. No. 9 in Mouton
Grammar Library. Mouton de Gruyter, Berlin.
Hausenberg, Anu-Reet. 1998. Komi. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 305–
326. Routledge, London and New York.
Herskovits, Annette. 1986. Language and Spatial Cognition: An Interdisciplinary
Study of the Prepositions in English. Cambridge University
Press, London and New York.
Herweg, Michael and Dieter Wunderlich. 1991. Lokale und Direktionale.
In Semantik, edited by Arnim von Stechow and Dieter Wunderlich,
pp. 758–785. Walter de Gruyter, Berlin/New York.
Hewitt, George B. 1995. Georgian: A Structural Reference Grammar.
John Benjamins, Amsterdam.
Hill, Deborah. 2002. Longgu. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 538–561. Curzon, Richmond, Surrey.
Hualde, José Ignacio and Jon Ortiz de Urbina (eds.). 2003. A Grammar
274 BIBLIOGRAPHY
of Basque. Mouton de Gruyter, Berlin/New York.
Hutchison, John P. 1981. The Kanuri Language: A Reference Grammar.
African studies program, University of Wisconsin, Madison.
Ibarretxe-Antuñano, Iraide. 2004. Polysemy in Basque locational cases.
Belgian Journal of Linguistics 18 1: 271–298.
Ikegami, Yoshihiko. 1982. Source vs. goal: A case of linguistic dissymmetry.
In Language and Cognitive Styles, edited by Richard Hoops
and Yvan Lebrun, no. 11 in Neurolinguistics, pp. 292–308. Swets and
Zeitlinger B.V., Lisse, The Netherlands.
Jackendoff, Ray. 1983. Semantics and Cognition. MIT Press, Cambridge,
Massachusetts.
Jake, Janice L. 1985. Grammatical relationships in Imbabura Quechua.
Garland, New York.
Janhunen, Juha (ed.). 2003. The Mongolic Languages. Routledge Language
Family Series. Routledge, London and New York.
Juank, Aijíu. 1982. Ajumatsatai yatsuchi: Manual de aprendizaje de la
lengua shuar, vol. 1. Mundo Shuar, Ecuador.
Kangasmaa-Minn, Eeva. 1998. Mari. In The Uralic Languages, edited
by Daniel Abondolo, Routledge Language Family Descriptions, pp.
219–248. Routledge, London and New York.
Karlsson, Fred. 1999. Finnish: An Essential Grammar. Routledge, London
and New york.
Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press,
Cambridge, Massachusetts.
Kayne, Richard S. 2004. Here and there. In Lexique, Syntaxe et LexiqueGrammaire/Syntax,
Lexis & Lexicon-Grammar: Papers in Honor of
Maurice Gross, edited by C. Leclère, E. Laporte, M. Piot, and M. Silberztein,
pp. 253–275. John Benjamins, Amsterdam.
Kayne, Richard S. 2005. Some notes on comparative syntax, with special
reference to English and French. In The Oxford Handbook of Comparative
Syntax, edited by Richard S. Kayne and Guglielmo Cinque, pp.
3–69. Oxford University Press, New York.
Kayne, Richard S. 2008. Comparative syntax and the lexicon. DIUG
lecture, University of Groningen, December 9/10, 2008. Available at
http://www.let.rug.nl/zwart/diug/docs/kaynediug.pdf.
Keresztes, László. 1998. Mansi. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 387–
427. Routledge, London and New York.
Kibrik, Aleksandr E. 2002. Nominal inflection galore: Daghestanian, with
side glances at Europe and the world. In Noun Phrase Structure in the
BIBLIOGRAPHY 275
Languages of Europe, edited by Frans Plank, pp. 37–112. Mouton de
Gryuter, Berlin.
Kiparsky, Paul. 1973. ‘Elsewhere’ in phonology. In A Festschrift for
Morris Halle, edited by Paul Kiparsky and Steven Anderson, pp. 93–
106. Holt, Rinehart and Winston, New York.
Kiparsky, Paul. 2005. Blocking and periphrasis in inflectional paradigms.
In Yearbook of Morphology 2004, edited by Geert E. Booij and Jaap
van Marie, pp. 113–135. Springer, Dordrecht.
Koopman, Hilda. 2000. Prepositions, postpositions, circumpositions, and
particles. In The Syntax of Specifiers and Heads, edited by Hilda Koopman,
pp. 204–260. Routledge, London.
Kracht, Marcus. 2001. The structure of local cases and its relevance for
the study of Uralic languages. In Proceedings of the IX. Congress of
Fenno-Ugrists. Tartu.
Kracht, Marcus. 2002. On the semantics of locatives. Linguistics and
Philosophy 25: 157–232.
Kracht, Marcus. 2007. Language and space. Ms. UCLA.
Kracht, Marcus. 2008. The fine structure of spatial expressions. In Syntax
and Semantics of Spatial P, edited by Anna Asbury, Jakub Dotlačil,
Berit Gehrke, and Rick Nouwen, no. 120 in Linguistik Aktuell, pp.
35–62. John Benjamins, Amsterdam and Philadelphia.
Krifka, Manfred. 1998. The origins of telicity. In Events and Grammar,
edited by Susan Rothstein, pp. 197–235. Kluwer, Dordrecht.
Krishnamurti, Bhadriraju. 1998. Telugu. In The Dravidian Languages,
edited by Sanford B. Steever, Routledge Language Family Descriptions,
pp. 203–240. Routledge, London and New York.
Kurbanov, K. K. 1990. ËÔÓ×Ó Ý ÚÝÖ Ò ÔÖÓ×ØÖ Ò×ØÚ ÒÒÝ Óع
ÒÓÜ Ò ° Ú Ø × Ö Ò× ÓÑ ÞÝ [The ways of expressing spatial relations
in Tabasaran]. In ÎÝÖ Ò ÔÖÓ×ØÖ Ò×ØÚ ÒÒÝ ÓØÒÓÜ Ò °
Ú ÞÝ ×Ø Ò [The Expression of Spatial Relations in the Daghestanian
Languages], edited by E. V. Spivak, pp. 113–122. Æ ËËËÊ
[Academy of Science USSR], Å Õ Ð [Mahachkala].
Kutscher, Silvia. 2001. Nomen und nominalen Syntagma im Lasischen:
Eine deskriptive Analyse des Dialekts von Ardeşen. Lincom Europa,
Munich.
Kutscher, Silvia. 2010. When ‘towards’ means ‘away from’: The case of
directional-ablative syncretism in the Ardesen-variety of Laz (SouthCaucasian).
Sprachtypologie und Universalienforschung 63 2: 252–271.
Lakoff, George. 1987. Women, fire, and dangerous things: what categories
reveal about the mind. University of Chicago Press, Chicago, Ill.
276 BIBLIOGRAPHY
Lakusta, Laura. 2005. Source and Goal Asymmetry in Non-linguistic Motion
Event Representations. Ph.D. thesis, Johns Hopkins University,
Maryland.
Lakusta, Laura and Barbara Landau. 2005. Starting at the end: The
importance of goals in spatial language. Cognition 96: 1–33.
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, vol. 1:
Theoretical Prerequisites. Stanford University Press, Stanford.
LaPolla, Randy J. 2003. Qiang. In The Sino-Tibetan Languages, edited
by Graham Thurgood and Randy J. LaPolla, pp. 573–587. Routledge,
London and New York.
Lestrade, Sander. 2010. The Space of Case. Ph.D. thesis, Radboud
University, Nijmegen.
Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary
Investigation. University of Chicago Press, Chicago, Ill.
Levinson, Stephen and Sérgio Meira. 2003. ‘Natural concepts’ in the
spatial topological domain — adpositional meanings in crosslinguistic
perspective: an exercise in semantic typology. Language 79 3: 484–516.
Lieber, Rochelle. 1992. Deconstructing Morphology: Word Formation in
Syntactic Theory. University of Chicago Press, Chicago.
Lundquist, Björn. 2008. Nominalizations and Participles in Swedish.
Ph.D. thesis, University of Tromsø.
Lynch, David. 2002a. Iaai. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 776–791. Curzon, Richmond, Surrey.
Lynch, David. 2002b. Xârâcùù. In The Oceanic Languages, edited by
John Lynch, Malcolm Ross, and Terry Crowley, Curzon Language
Family Series, pp. 765–775. Curzon, Richmond, Surrey.
Lytkin, V. I. (ed.). 1962. ÃÓÑ ¹È ÖÑ ° ÞÝ ÚÚ Ò ¸ ÓÒ Ø ¸
Ð × ÑÓÖ ÓÐÓ [The Komi-Permyak Language: Introduction,
Phonetics, Lexis and Morphology]. Komi-Permyackoe Knizhnoe Izdatel’stvo,
Kudymkar.
MacAulay, Donald. 1992. The Scottish Gaelic language. In The Celtic
Languages, edited by Donald Macaulay, Cambridge Languages Surveys,
pp. 137–248. Cambridge University Press, Cambridge.
Magomedbekova, Zagidad. 1967a. Ú × ° ÞÝ Ö ÑÑ Ø Õ × °
Ò Ð Þ¸ Ø ×Øݸ ×ÐÓÚ Ö [The Akhvakh Language: Grammatical
Analysis, Texts, Dictionary]. Å Ò Ö [Mecniereba], Tbilisi.
Magomedbekova, Zagidad. 1967b. É Ñ Ð Ò× ° ÞÝ [The Chamalal
language]. In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples
of the USSR], edited by V.V. Vinogradov, vol. 4, pp. 384–402. Æ Ù
BIBLIOGRAPHY 277
[Nauka], Moscow.
Magomedbekova, Zagidad. 1971. Ã Ö Ø Ò× ° ÞÝ Ö Ñ Ø × °
Ò Ð Þ¸Ø ×Øݸ×ÐÓÚ Ö [The Karata Language: Grammar Analysis,
Texts, Glossary]. Å Ò Ö [Mecniereba], Tbilisi.
Magometov, Aleksandr Arnarovich. 1965. Ì × Ö Ò× ° ÞÝ ××Ð ¹
ÓÚ Ò Ø ×ØÝ [The Tabasaran Language: Investigation and
Texts]. Å Ò Ö [Mecniereba], Tbilisi.
Mahootian, Shahrzad. 1997. Persian. Routledge, London.
Matisoff, James A. 1973. The Grammar of Lahu. University of California
Press, Berkeley.
Matisoff, James A. 2003. Lahu. In The Sino-Tibetan Languages, edited
by Graham Thurgood and Randy J. LaPolla, pp. 208–221. Routledge,
London and New York.
May, Robert. 1985. Logical form: its structure and derivation. Linguistic
inquiry monographs; 12. MIT Press, Cambridge, Massachusetts.
McCarthy, John and Alan Prince. 1995. Faithfulness and reduplicative
identity. In Papers in Optimality Theory, edited by Jill Beckman,
Laura Wals-Dickie, and Suzanne Urbanczyk, pp. 249–384. Amherst
University, Massachusetts.
McCawley, James D. 1968. Lexical insertion in a transformational grammar
without Deep Structure. In Papers from the fourth regional meeting
of the Chicago Linguistic Society, edited by B. J Darden, C.-J. N.
Bayley, and A. Davidson, pp. 71–80. University of Chicago, Chicago.
Mel’čuk, Igor’ Aleksandrovič. 1994. Cours de moprhologie générale, vol.
II: Significations morphologiques. Les presses de l’université de Montréal,
Montreal.
Mulugeta, Seyoum. 2008. A Grammar of Dime. Ph.D. thesis, University
of Utrecht.
Muriungi, Peter Kinyua. 2006. Categorizing Adpositions in Kîîtharaka.
In Tromsø Working Papers on Language and Linguistics: Nordlyd
33.1, special issue on Adpositions, edited by Peter Svenonius, pp. 26–
48. University of Tromsø, Tromsø.
Muriungi, Peter Kinyua. 2008. Phrasal movement inside Bantu verbs:
Deriving affix scope and order in Kîîtharaka. Ph.D. thesis, University
of Tromsø.
Muriungi, Peter Kinyua. 2009. The Union Spell-out Mechanism. In
Nordlyd 36.1: Special issue on Nanosyntax, edited by Peter Svenonius,
Gillian Ramchand, Michal Starke, and Knut Tarald Taraldsen,
pp. 191–205. University of Tromsø, Tromsø. Available at
www.ub.uit.no/munin/nordlyd/.
278 BIBLIOGRAPHY
Murkelinskij, G.B. 1967. Ä × ° ÞÝ [The Lak language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 4, pp. 488–507. Æ Ù [Nauka], Moscow.
Nagano, Yasuhiko. 2003. Cogtse Gyarong. In The Sino-Tibetan Languages,
edited by Graham Thurgood and Randy J. LaPolla, pp. 469–
489. Routledge, London and New York.
Nam, Seungho. 1995. The Semantics of Locative Prepositional Phrases
in English. Ph.D. thesis, UCLA.
Narasimhan, Bhuvana. 2008. Motion events and the lexicon: A case
study of Hindi. Lingua 113: 123–160.
Nebel, A. 1948. Dinka Grammar (Rek-Malual dialect) with texts and
vocabulary. Instituto Missioni Africane, Verona.
Nedjalkov, Igor. 1997. Evenki. Routledge, London and New York.
Neeleman, Ad and Kriszta Szendrői. 2007. Radical pro-drop and the
morphology of pronouns. Linguistic Inquiry 38 4: 671–714.
Nichols, Johanna. 1994. Ingush. In North East Caucasian Languages,
Part 2, edited by Rieks Smeets, vol. 4 of The Indigenous Languages of
the Caucasus. Caravan books, Delmar, New York.
Nikitina, Tatiana. 2009. Subcategorization pattern and lexical meaning
of motion verbs: A study of the Source/Goal ambiguity. Linguistics
47 5: 1113–1141.
Noonan, Michael. 2003a. Motion events in Chantyal. In Motion, Direction
and Location in Languages, edited by Erin Shay and Uwe Siebert,
pp. 211–234. John Benjamins, Amsterdam/Philadelphia.
Noonan, Michael. 2003b. Nar-Phu. In The Sino-Tibetan Languages,
edited by Graham Thurgood and Randy J. LaPolla, pp. 336–352. Routledge,
London and New York.
Noonan, Michael. 2008. Patterns of development, patterns of syncretism
of relational morphology in the Bodic languages. In The Role of
Semantics and Pragmatics in the Development of Case, edited by
Jóhanna Barðdal and Shobhana Cheliah, pp. 261–281. John Benjamins,
Amsterdam.
Noonan, Michael and Elena Mihas. 2010. Areal dimensions
in case syncretism: Ablatives and genitives.
Unpublished manuscript, downloadable at
https://pantherfile.uwm.edu/noonan/www/Ablatives%26Genitives.pdf.
Panchvidze, B. H. and E. F. Djeirashvili. 1967. Í Ò× ° ÞÝ [The
Udin language]. In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages of the
Peoples of the USSR], edited by V.V. Vinogradov, vol. 4, pp. 676–687.
Æ Ù [Nauka], Moscow.
BIBLIOGRAPHY 279
Pandharipande, Rajeshwari. 1997. Marathi. Routledge, London.
Pantcheva, Marina. 2006. Persian preposition classes. In Tromsø Working
Papers on Language and Linguistics: Nordlyd 33.1, special issue
on Adpositions, edited by Peter Svenonius, pp. 1–25. University of
Tromsø, Tromsø.
Pantcheva, Marina. 2008. The Place of PLACE in Persian. In Syntax and
Semantics of Spatial P, edited by Anna Asbury, Jakub Dotlačil, Berit
Gehrke, and Rick Nouwen, no. 120 in Linguistik Aktuell, pp. 305–330.
John Benjamins.
Pantcheva, Marina. 2010. The syntactic structure of Locations, Goals
and Sources. Linguistics 48: 1043–1082.
Pashov, Petâr. 1999. Ð Ö× Ö Ñ Ø [A Grammar of Bulgarian].
Hermes, Plovdiv.
Peterson, David A. 2003. Hakha Lai. In The Sino-Tibetan Languages,
edited by Graham Thurgood and Randy J. LaPolla, pp. 409–426. Routledge,
London and New York.
Piñón, Christopher J. 1993. Paths and their names. In Papers from
the 29th Regional Meeting of the Chicago Linguistic Society, edited by
Katharine Beals, Gina Cooke, David Kathman, Sotaro Kita, Karl-Erik
McCullough, and David Testen, vol. 2, pp. 287–303. CLS, Chicago.
Poser, William J. 1992. Blocking of phrasal constructions by lexical
items. In Lexical Matters, edited by Ivan A. Sag and Anna Szabolcsi,
pp. 111–130. Center for Study of Language & Information, Stanford,
CA.
Prince, Alan and Paul Smolensky. 2004. Optimality Theory: Constraint
Interaction in Generative Grammar. Blackwell, Malden, Mas-
sachusetts.
Radkevich, Nina. 2009. Vocabulary insertion and the geometry of local
cases. Available at http://ling.auf.net/lingBuzz/000958.
Radkevich, Nina. 2010. On Location: The Structure of Case and Adpositions.
Ph.D. thesis, University of Connecticut.
Ramchand, Gillian. 2008a. Lexical items in complex predications: Selection
as underassociation. In Tromsø Working Papers on Language
and Linguistics: Nordlyd 35.1, special issue on Complex Predication,
edited by Peter Svenonius and Inna Tolskaya, pp. 115–141. University
of Tromsø, Tromsø.
Ramchand, Gillian. 2008b. Verb Meaning and the Lexicon: A First Phase
Syntax. Cambridge University Press.
Ricca, Davide. 1993. I verbi deittici di moviento in Europa: Una ricerca
interlinguistica. La Nuova Italia, Firenze.
280 BIBLIOGRAPHY
Rice, Sally and Kaori Kabata. 2007. Crosslinguistic grammaticalization
patterns of the allative. Linguistic Typology 11: 451–514.
Richards, Norvin. 2001. Movement in Language: Interactions and Architectures.
Oxford University Press, Oxford.
van Riemsdijk, Henk. 1990. Functional prepositions. In Unity in Diversity,
edited by H. Pinkster and I. Genée, pp. 229–241. Foris, Dordrecht.
van Riemsdijk, Henk and Riny Huybregts. 2002. Location and locality.
In Progress in Grammar: Articles at the 20th Anniversary of the Comparison
of Grammatical Models Group in Tilburg, edited by Marc van
Oostendorp and Elena Anagnostopoulou, pp. 1–23. Meertens Instituut,
Amsterdam.
Riese, Timothy. 1998. Permian. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 249–
275. Routledge, London and New York.
Rizzi, Luigi. 1986. On chain formation in the syntax of pronominal clitics.
In Syntax and Semantics, edited by Hagit Borer, vol. 19, pp. 65–96.
Academic Press, New york.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of
Grammar: Handbook in Generative Syntax, edited by Liliane Haegeman,
pp. 281–337. Kluwer, Dordrecht.
Romanova, Eugenia. 2007. Constructing Perfectivity in Russian. Ph.D.
thesis, University of Tromsø.
Ross, Malcolm. 2002a. ‘Ala‘ala. In The Oceanic Languages, edited by
John Lynch, Malcolm Ross, and Terry Crowley, Curzon Language
Family Series, pp. 347–361. Curzon, Richmond, Surrey.
Ross, Malcolm. 2002b. Jabêm. In The Oceanic Languages, edited by John
Lynch, Malcolm Ross, and Terry Crowley, Curzon Language Family
Series, pp. 270–296. Curzon, Richmond, Surrey.
Ross, Malcolm. 2002c. Kaulong. In The Oceanic Languages, edited
by John Lynch, Malcolm Ross, and Terry Crowley, Curzon Language
Family Series, pp. 387–409. Curzon, Richmond, Surrey.
Rotstein, Carmen and Yoad Winter. 2004. Total adjectives vs. partial adjectives:
Scale structure and higher-order modifiers. Natural Language
Semantics 12: 259–288.
Salminen, Tapani. 1998. Nenets. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 516–
547. Routledge, London and New York.
Sammallahti, Pekka. 1998. The Saami Languages: An Introduction.
Davvi Girji OS, Kárásjohka.
Schachter, Paul and Fe T. Otanes. 1972. Tagalog Reference Grammar.
BIBLIOGRAPHY 281
University of California Press, Berkeley.
Siddiqi, Daniel. 2006. Minimize Exponence: Economy Effects of the Morphosyntactic
Component of the Grammar. Ph.D. thesis, University of
Arizona.
Skorik, P. J. 1968. ÉÙ ÓØ× ° ÞÝ [The Chukchi language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 5, pp. 248–270. Æ Ù [Nauka], Moscow.
Skribnik, Elena. 2003. Buryat. In The Mongolic Languages, edited by
Juha Janhunen, Routledge Language Family Series, pp. 102–128. Routledge,
London and New York.
Son, Minjeong and Peter Svenonius. 2008. Microparameters of crosslinguistic
variation: Directed motion and resultatives. In Proccedings
of the 27th West Coast Conference on Formal Linguistics, edited by
Natasha Abner and Jason Bishop, pp. 388–396. Cascadilla, Somerville,
Ma.
Spencer, Andrew. 1991. Morphological Theory: An Introduction to Word
Structure in Generative Grammar. Blackwell, Oxford.
Spivak, E. V. (ed.). 1990. ÎÝÖ Ò ÔÖÓ×ØÖ Ò×ØÚ ÒÒÝ ÓØÒÓÜ Ò °
Ú ÞÝ ×Ø Ò [The expression of spatial relations in the Daghestanian
languages]. Æ ËËËÊ [Academy of Science USSR], Å ¹
Õ Ð [Mahachkala].
Starke, Michal. 2005-2009. Nanosyntax. Class lectures, CASTL, University
of Tromsø.
Starke, Michal. 2009. A short primer to a new approach to language.
In Nordlyd 36.1: Special issue on Nanosyntax, edited by Peter
Svenonius, Gillian Ramchand, Michal Starke, and Knut Tarald
Taraldsen, pp. 1–6. University of Tromsø, Tromsø. Available at
www.ub.uit.no/munin/nordlyd/.
Starke, Michal. 2011. Towards elegant parameters: Language variation
reduces to the size of lexically stored trees. Transcipt from a talk at
Barcelona Workshop on Linguistic Variation in the Minimalist Framework.
Available at http://ling.auf.net/lingBuzz/001183.
Steever, Sanford B. 1998. Gond
˙
i. In The Dravidian Languages, edited
by Sanford B. Steever, Routledge Language Family Descriptions, pp.
270–297. Routledge, London and New York.
Stefanowitsch, Anatol and Ada Rohde. 2004. The goal bias in the encoding
of motion events. In Motivation in Grammar, edited by Klaus-Uwe
Panther and Günter Radden, pp. 249–268. Mouton de Gruyter, Berlin
and New York.
Sulkala, Helena and Merja Karjalainen. 1992. Finnish. Routledge, Lon-
282 BIBLIOGRAPHY
don.
Sumbatova, Nina R. and Rasul O. Mulatov. 2003. A Grammar of Icari
Dargwa. 92. Lincom Europa, Munich.
Sunik, O. P. 1968. ÍÐ × ° ÞÝ [The Ulcha language]. In ÞÝ
Ò ÖÓ ÓÚ ËËËÊ [The Languages of the Peoples of the USSR], edited
by V.V. Vinogradov, vol. 5, pp. 149–171. Æ Ù [Nauka], Moscow.
Svantesson, Jan-Olof. 2003. Khalkha. In The Mongolic Languages, edited
by Juha Janhunen, Routledge Language Family Series, pp. 154–176.
Routledge, London and New York.
Svenonius, Peter. 2004. Slavic prefixes inside and outside VP. In Nordlyd
32.2: Special issue on Slavic prefixes, edited by Peter Svenonius,
pp. 205–253. University of Tromsø, Tromsø. Available at
www.ub.uit.no/munin/nordlyd/.
Svenonius, Peter. 2006. The emergence of Axial Parts. In Tromsø Working
Papers on Language and Linguistics: Nordlyd 33.1, special issue
on Adpositions, edited by Peter Svenonius, pp. 49–77. University of
Tromsø, Tromsø.
Svenonius, Peter. 2007. Scaling Right Along seminar. Class lecture,
March 20, 2007, CASTL, University of Tromsø.
Svenonius, Peter. 2008. Projections of P. In Syntax and Semantics of
Spatial P, edited by Anna Asbury, Jakub Dotlačil, Berit Gehrke, and
Rick Nouwen, no. 120 in Linguistik Aktuell, pp. 63–84. John Benjamins,
Amsterdam and Philadelphia.
Svenonius, Peter. 2009a. Location and Source in North Sámi. Talk given
at Arctic Languages: Syntax, Morphology, and Lexicon, September
24-26, 2009, University of Tromsø, September.
Svenonius, Peter. 2009b. Space Seminar. Class lectures, CASTL, University
of Tromsø.
Svenonius, Peter. 2010. Spatial prepositions in English. In The Cartography
of Syntactic Structure, vol. 6, edited by Guglielmo Cinque and
Luigi Rizzi, pp. 127–160. Oxford University Press, Oxford. Prepublication
draft available at http://ling.auf.net/lingBuzz/000001.
Svenonius, Peter and Marina Pantcheva (eds.). 2006. Tromsø Working
Papers on Language and Linguistics: Nordlyd 33, special issue on Adpositions.
University of Tromsø, Tromsø.
Svenonius, Peter, Gillian Ramchand, Michal Starke, and Knut Tarald
Taraldsen (eds.). 2009. Nordlyd 36.1: Special issue on
Nanosyntax. University of Tromsø, Tromsø. Available at
www.ub.uit.no/munin/nordlyd/.
Svenonius, Peter and Minjeong Son. 2009. Directed manner of mo-
BIBLIOGRAPHY 283
tion in verb serialization: A comparative study of Indonesian/Malay
and Tetun Dili. Talk presented at the International Symposium on
Malay/Indonesian Linguistics XIII, held in Senggigi, Lombok, Indonesia,
June 6-7.
Takamine, Kaori. 2006. The Axial Part Phrase in Japanese. In Tromsø
Working Papers on Language and Linguistics: Nordlyd 33.1, special
issue on Adpositions, edited by Peter Svenonius, pp. 78–97. University
of Tromsø, Tromsø.
Talmy, Leonard. 2000. Toward a Cognitive Semantics: Concept Structuring
Systems, vol. I. MIT Press, Cambridge, Massachusetts.
Taraldsen, Knut Tarald. 2010. The nanosyntax of Nguni noun class
prefixes and concords. Lingua 120: 1522–1548.
Taylor, John R. 1996. The syntax and semantics of locativized nouns
in Zulu. In The Construal of Space in Language and Thought, edited
by Martin Pütz and René Dirven, pp. 287–305. Mouton de Gruyter,
Berlin/New York.
Ternes, Elmar. 1992. The Breton language. In The Celtic Languages,
edited by Donald Macaulay, Cambridge Language Surveys, pp. 371–
452. Cambridge University Press, Cambridge.
Tikka, Toivo. 1992. Vepsän suffiksoituneet postpositiot. Kieliopillisiin
sijoihin liittyvä suffiksoituminen [Agglutinating postpositions in Veps.
Suffixation to grammatical cases]. No. 22 in Studia Uralica Upsaliensia.
Uppsala Universitet, Uppsala.
Tomič, Olga Mišeska. 2006. Balkan Sprachbund: Morpho-Syntactic Features.
Springer, Dordrecht.
Tsumagari, Toshiro. 2003. Dagur. In The Mongolic Languages, edited
by Juha Janhunen, Routledge Language Family Series, pp. 129–153.
Routledge, London and New York.
Uslar, P.K. 1889. Ú Ö× i° ÞÝ [The Avar Language]. £ØÒÓ Ö i
Ã Ú Þ [The ethnography of Caucasus]. TipografIja Kancel’arIi
Glavnonachal’stvujushtago grazhdanskoju chastIju na Kavkaze, Tbil-
isi.
Vellard, Jahen A. 1967. Contribución al estudo de la lengua Uru. Universidad
de Buenos Aires, Facultad de Filisofía y Letras, Buenos Aires.
Verkuyl, Henk J. 1972. On the Compositional Nature of the Aspects.
Reidel, Dordrecht.
Viitso, Tiit-Rein. 1998. Estonian. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 115–
148. Routledge, London and New York.
Vinogradov, V.V. (ed.). 1967. ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages
284 BIBLIOGRAPHY
of the Peoples of the USSR], vol. 4. Æ Ù [Nauka], Moscow.
Vogt, Hans. 1971. Grammaire de la langue géorgienne. Universitetsforlaget,
Oslo.
Volodin, Alexander. P. and A. N. Žukova. 1968. ÁØ Ð Ñ Ò× ° ÞÝ
[The Itelmen language]. In ÞÝ Ò ÖÓ ÓÚ ËËËÊ [The Languages
of the Peoples of the USSR], edited by V.V. Vinogradov, vol. 5, pp.
334–351. Æ Ù [Nauka], Moscow.
Wälchli, Bernard and Fernando Zuñiga. 2006. Source-goal (in)difference
and the typology of motion events in the clause. Sprachtypologie und
Universalienforschung 59 3: 284–303.
Watters, E., David. 2002. A Grammar of Kham. Cambridge University
Press, Cambridge.
Weerman, Fred and Jacqueline Evers-Vermeul. 2002. Pronouns and case.
Lingua 112: 301–338.
Wheatley, Julian K. 2003. Burmese. In The Sino-Tibetan Languages,
edited by Graham Thurgood and Randy J. LaPolla, pp. 195–207. Routledge,
London and New York.
Williams, Edwin. 2003. Representation Theory. MIT Press, Cambridge,.
Winkler, Eberhard. 2001. Udmurt, vol. 212 of Languages of the World.
Lincom Europa, Munich.
Winter, Yoad. 2006. Closure and telicity across categories. In Proceedings
of Semantic and Linguistic Theory XVI , edited by Christopher Tancredi,
Makoto Kanazawa, Ikumi Imani, and Kiyomi Kusumoto. CLC
Publications, Fort Washington, PA.
Wunderlich, Dieter. 1991. How do prepositional phrases fit into compositional
syntax and semantics? Linguistics 29: 591–621.
Yang, Charles. 1997. Minimal Computation: Derivation of Syntactic
Structures. Master’s thesis, MIT, Cambridge, Massachusetts.
Zaicz, Gábor. 1998. Mordva. In The Uralic Languages, edited by
Daniel Abondolo, Routledge Language Family Descriptions, pp. 185–
218. Routledge, London and New York.
Zajtseva, Maria. 1981. Ö ÑÑ Ø Ú Ô×× Ó Ó ÞÝ ÓÒ Ø
ÑÓÖ ÓÐÓ [A Grammar of the Veps Language: Phonetics and Morphology].
Æ Ù [Nauka], Leningrad.
Zhirkov, L. I. 1955. Ä × ° ÞÝ ÓÒ Ø ÑÓÖ ÓÐÓ [The Lak
Language: Phonetics and Morphology]. ÁÞ Ø Ð³×ØÚÓ Ñ Æ Ù
ËËËÊ [Publishing House of the SSSR Academy of Science], Moskow.
Zlatev, Jordan and Peerapat Yangklang. 2004. A third way to travel: The
place of Thai in motion event typology. In Relating Events in Narrative:
Typological and contextual perspective, edited by Sven Strömqvist
BIBLIOGRAPHY 285
and Ludo Verhoeven, vol. 2, pp. 159–190. Lawrence Erlbaum, Mahwah,
New Jersey.
Zwart, C. Jan-Wouter. 1996. “Shortest Move” versus “Fewer Steps”. In
Minimal Ideas. Syntactic Studies in the Minimalist Framework, edited
by W. Abraham, S.D. Epstein, H. Thráinsson, and C.J.W. Zwart,
no. 12 in Linguistik Aktuell, pp. 305–327. John Benjamins, Amsterdam
and Philadelphia.
Zwarts, Joost. 1997. Vectors as relative positions: A compositional semantics
of modified PPs. Journal of Semantics 14: 57–86.
Zwarts, Joost. 2005. Prepositional aspect and the algebra of paths. Linguistics
and Philosophy 28: 739–779.
Zwarts, Joost. 2008a. Aspect of a typology of direction. In Theoretical
and Crosslinguistic Approaches to the Semantics of Aspects, edited by
Susan Rothstein, pp. 79–106. John Benjamins, Amsterdam.
Zwarts, Joost. 2008b. Priorities in the production of prepositions. In
Syntax and Semantics of Spatial P, edited by Anna Asbury, Jakub
Dotlačil, Berit Gehrke, and Rick Nouwen, no. 120 in Linguistik Aktuell,
pp. 85–102. John Benjamins, Amsterdam and Philadelphia.
Zwarts, Joost and Yoad Winter. 2000. Vector space semantics: A modeltheoretic
analysis of locative prepositions. Journal of Logic, Language,
and Information 9: 169–211.