ABSTRACT
Title of Dissertation: PHASEHOOD AND PHI-INTERVENTION
Sigwan Thivierge
Doctor of Philosophy, 2021
Dissertation Directed by: Professor Maria Polinsky
Associate Professor Omer Preminger
Department of Linguistics
This dissertation investigates the notion of phases in syntactic theory, and offers a
reanalysis of certain phases as instances of phi(?)-intervention. Under the standard view,
phases are syntactic structures that, according to the Phase Impenetrability Condition, are
opaque to operations originating outside of the phase (Chomsky, 2000; 2001). Phasehood
was linked to certain heads such as C0 and (transitive) v0, but several issues arise once the
empirical domain is broadened beyond English.
As more work has turned to unrelated languages, a less stipulative alternative has
presented itself: phases are intervention effects, and are reducible to a more general lo-
cality issue. In Rackowski and Richards? (2005) account of Tagalog wh-movement, for
example, CPs act as phases because they constitute the closest goal. In Halpert?s (2019)
account of Zulu hyper-raising, CPs do not act as phases due to the cyclic nature of Agree.
In Keine?s (2017) account of Hindi long-distance agreement, vPs do not act as phases,
which I argue is because v0 is not a ?-goal. In Georgian, vPs act as phases because v0, in
contrast, is a ?-goal (as I will argue in this thesis). These languages show that XPs act as
phases only when they are potential goals for a syntactic operation. These languages also
illustrate two ways of diagnosing phasehood as ?-intervention: via movement out of the
domain, and via agreement into the domain. These results suggest that phasehood is an
epiphenomenon, and that the interior of the ?phase? is accessible even after the phase is
complete.
In this dissertation, I argue that certain instances of phasehood derive from the
?phase? head bearing a ?-probe: the ?-features on the probe intervene for ?-agreement,
which results in phase-like effects. The empirical data in favour this claim comes from
the Georgian agreement system. I show that subjects in Georgian are base-generated in
different positions, depending on whether they fall under the basic agreement paradigm
or the inverse agreement paradigm. In the basic, subjects are introduced above v0 and are
the closest goal for Agree operations that originate outside the vP domain. In the inverse,
subjects are introduced below v0; in this case, the ?-features that are associated with the
?-probe on v0 constitute the closest goal for Agree.
PHASEHOOD AND PHI-INTERVENTION
by
Sigwan Thivierge
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2021
Advisory Committee:
Professor Maria Polinsky, Co-chair/Co-Advisor
Associate Professor Omer Preminger, Co-chair/Co-Advisor
Professor Juan Uriagereka
Associate Professor Alexander Williams
Professor Kira Gor
?c Copyright by
Sigwan Thivierge
2021
Preface
This work was supported by the Social Sciences and Humanities Research Coun-
cil (Doctoral Scholarship #752-2016-2476). This research was also supported in part
by grants (#BCS-1619857, #BCS-1941733) awarded to Maria Polinsky by the National
Science Foundation.
ii
Dedication
To all my relations, and F-A-V.
iii
Acknowledgments
Not long after I completed my BA program in Linguistics, and just as I was set to
begin my MA, I confided to one of my advisors, Charles Reiss, that I had been feeling
a bit anxious about this next step in my education. In true form, Charles laughed and
replied that I should always be feeling a little nervous; that?s how I would know that I
was moving out of my comfort zone and into something bigger and better. That advice
followed me through to the completion of my MA program, and it?s what prompted me
to continue on to my PhD studies. Even now, at the end of it all and the beginning of
something new, that advice is still my constant companion.
First and foremost, I am hugely indebted to my advisors, Maria Polinsky and Omer
Preminger, for guiding me through all of this five year adventure. They were nothing but
encouraging when I chose to work on the Georgian agreement system, and they didn?t
even bat an eye when that project was extended to phase theory itself. They challenged
me when I needed to be challenged, and they were exceedingly patient and understanding
when life threw some curveballs my way. I could not have asked for better advisors. I
am also thankful to my dissertation committee, Alexander Williams, Juan Uriagereka,
and Kira Gor. Their feedback and support were incredibly helpful, and they made this
dissertation stronger and clearer than I could?ve made it on my own.
iv
Alexander deserves to be singled out for extra thanks since he served as my super-
visor when I decided to investigate indexical shift in Georgian. Even though many of our
meetings consisted of us sitting in confused contemplation and head-scratching, I learned
a lot from his encouragement and enthusiasm. I also really appreciated his lunch-time sto-
ries, our races to get the last of the coffee, and his diverse recommendations for movies
and music.
More broadly, I am very grateful that I had the chance to be part of the linguistics
department at Maryland. I benefited from so many conversations and interactions with
Tonia Bleam, Jeff Lidz, Norbert Hornstein, and Valentine Hacquard. I am also grateful
to Howard Lasnik, Naomi Feldman, Bill Idsardi, Ellen Lau, Peggy Antonisse, Andrea
Zukowski, and Colin Phillips. Kim Kwok, in particular, made sure that everything got
completed and that nothing fell through the administrative cracks. I learned so much
from everyone in this department and community.
I was incredibly lucky to put the S in the SMAART cohort: Mina Hirzel, Anouk
Dieuleveut, Aaron Doliana, Rodrigo Ranero, and Tyler Knowlton. We had the best of
times throughout the years, which were filled with lots of laughter and tears, serious
sciencing and goofing around, and endless support and hugs. I am so happy that we got
to go through this experience together.
I would also like to thank many other fellow students and colleagues at UMD:
Paulina Lyskawa, Max Papillon, Annemarie van Dooren, Suyoung Bae, Chia-Hsuan Liao,
Laurel Perkins, Nick Huang, Gesoel Mendes, Yu?an Yang, Adam Liter, Craig Thorburn,
Masato Nakamura, Dan Goodhue, and (Cardboard) Ted Levin. I would also like to apol-
ogize to the ?first years??Jackie Nelligan, Adam Liter, Hanna Muller, Hisao Kurokami,
v
and Yu?an Yang?for hiding the university folding chairs that they had borrowed for their
first department holiday party. On that note, I would like to thank Tyler Knowlton for
helping me hide the chairs during their second department holiday party.
Even before I came to UMD, I had great mentors and friends who encouraged me
every step of the way. Charles Reiss deserves the utmost of gratitude, as well as Dana
Isac and Alan Bale. Ivanna Richardson and Matthew Schuurman are equally deserving;
their friendship kept me grounded, and their advice kept me sane.
I could not have gotten this far without the support of my family and friends. My
parents, Sharon Hunter and Rodney Thivierge, made sure that I never lost sight of my
studies; my grandparents, Rose Mathias and Jimmy Hunter, were always there to talk to
me about our language. It was hearing them speak Anishinaabemowin that started me
down this path in the first place?kitci migwetc. I am grateful for all the laughter and
strength that Tiffany Boudrias, Derek Decontie-Wawatie, and Brian Brousseau provided
me while I was all the way over here in Maryland; and, last but not least, I am indebted to
Sebastian Duncan and Fav, for making Maryland a home for me.
vi
Table of Contents
Preface ii
Dedication iii
Acknowledgements iv
Table of Contents vii
List of Abbreviations ix
Chapter 1: Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Roadmap of the dissertation . . . . . . . . . . . . . . . . . . . . . . . . . 12
Chapter 2: On phase theory 15
2.1 Motivating phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 The Phase Impenetrability Condition (PIC) and Agree . . . . . . . . . . . 21
2.3 Diagnosing phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Issues with defining phases . . . . . . . . . . . . . . . . . . . . . . . . . 35
Chapter 3: Neutralizing phasal boundaries 38
3.1 The Phase Impenetrability Condition as an intervention effect . . . . . . . 38
3.2 Alleviating the Phase Impenetrability Condition . . . . . . . . . . . . . . 50
3.3 Summary and desiderata . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3.2 Desiderata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Chapter 4: Basic properties of Georgian 62
4.1 Description of the Georgian agreement system . . . . . . . . . . . . . . . 63
4.1.1 General properties . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.1.2 The basic agreement paradigm . . . . . . . . . . . . . . . . . . . 66
4.1.3 The inverse agreement paradigm . . . . . . . . . . . . . . . . . . 73
4.2 The Georgian clause structure . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2.1 Argument positions: The basic agreement paradigm . . . . . . . . 81
4.2.2 Argument positions: The inverse agreement paradigm . . . . . . . 83
4.2.3 Georgian is head-initial and head-final . . . . . . . . . . . . . . . 89
4.3 Georgian case-marking . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
vii
4.4 Pronominal licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Chapter 5: Phasehood as ?-intervention in Georgian 106
5.1 An analysis of Georgian agreement . . . . . . . . . . . . . . . . . . . . . 109
5.1.1 Feature geometric representations . . . . . . . . . . . . . . . . . 111
5.1.2 Deriving the prefixes . . . . . . . . . . . . . . . . . . . . . . . . 112
5.1.3 Deriving the versionizers . . . . . . . . . . . . . . . . . . . . . . 122
5.2 Deriving the suffixes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.2.1 Inverse agreement paradigm . . . . . . . . . . . . . . . . . . . . 124
5.2.2 Basic agreement paradigm . . . . . . . . . . . . . . . . . . . . . 128
5.3 Previous analyses of Georgian agreement . . . . . . . . . . . . . . . . . 139
5.3.1 Basic agreement-only accounts . . . . . . . . . . . . . . . . . . . 140
5.3.2 Accounts including the inverse agreement paradigm . . . . . . . . 145
Chapter 6: Further investigating phases as ?-intervention 150
6.1 A case study addressing the lack of phase effects . . . . . . . . . . . . . . 151
6.2 Remaining questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Chapter 7: Conclusion 160
7.1 A comprehensive analysis of the Georgian agreement system . . . . . . . 160
7.2 Phases as ?-intervention . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Bibliography 165
viii
List of Abbreviations
1 1st person
2 2nd person
3 3rd person
ABS absolutive
ACC accusative
AOR aorist
APPL applicative
AUG augment vowel
AUTH author
CAUS causative
C complementizer
DAT dative
ERG ergative
F feminine
FUT future
GEN genitive
INCH inchoative
INF infinitive
IPFV imperfective
M masculine
NOM nominative
NMLZ nominalizer
O object
PART participant
PERF perfect
PL plural
PRES present
PRV preverb
S subject
SF stem formant
SG singular
SM stem marker
RECI reciprocal
TAM tense/aspect/mood
TS thematic suffix
TAM tense/aspect/mood
VER versionizer
ix
Chapter 1: Introduction
1.1 Overview
This dissertation proposes a reanalysis of certain phases as instances of interven-
tion under locality, particularly with respect to ?-intervention. While the theoretical no-
tion of bounding domains extends far back into pre-Minimalist Program linguistic theory
(Chomsky, 1965, 1973, 1981, 1986), modern phasehood is generally thought to be static
throughout its syntactic environment (pace Dikken, 2007; Gallego, 2006, 2010), with its
nature arising from complete propositional content and theta-role assignment (Chomsky,
2000). Under these assumptions, the standard phasal categories were CP and transitive vP,
as well as DP. The motivations for these phasal categories were both conceptual and em-
pirical. The conceptual advantage was immediate under the Minimalist Program, where
the syntactic derivation is built in cyclic, incremental chunks. Some of those chunks, i.e.
phases, reduce the computational complexity required for the derivation since the system
could manipulate only a subset of the syntactic material needed. As each phase was as-
sembled, the lexical items required for that step of the derivation were exhausted and the
sub-structure was shipped to the interfaces via the operation Transfer. For example, given
the lexical items in the Numeration (N) in (1a), the nominal argument [many dogs] will
1
satisfy the EPP feature of the embedded T, and the expletive there will satisfy the EPP
feature of the matrix T. This derivation results in the acceptable sentence in (1b).
(1) a. N = { {there, is, a, strong, likelihood}, {that, will, be, many, dogs, at, the,
park, right, now} }
b. [CP There is a strong likelihood [CP that many dogs will be at the park right
now ] ]
c. *Therei is a strong likelihood that ti will be [many dogs] at the park right now.
(based on Chomsky, 2000)
Without phases, i.e. subset structures in the Numeration, the derivation would result in
the unacceptable string in (1c). In that particular derivation, the expletive there and the
nominal [many dogs] would both be available to potentially satisfy the EPP feature of
the embedded T. Given the principle of Merge over Move (Chomsky, 1995), which states
that merging syntactic material is more computationally efficient than moving syntactic
material, we would expect the expletive subject to satisfy the embedded EPP feature.
That same expletive subject would then satisfy the EPP feature of the matrix T, since it
is the closest nominal and there is no other expletive subject in the Numeration. Thus,
structuring the Numeration into subset relations derived the acceptable sentence in (1b)
and the unacceptable string in (1c).
The empirical motivation for phases derived from long-distance successive cyclic
movement, where embedded wh-phrases move through the edge of intermediate CPs en
route to the matrix Spec,CP position (Chomsky, 1986, 2000, 2001, 2008; McCloskey,
2
2002). This is shown below for West Ulster English, where a quantifier associated with
an embedded wh-pronoun can be stranded in intermediate CP positions. The phase lent
itself to a straightforward explanation of this phenomena: if CPs are phases, and phases
are syntactic substructures that are cyclically sent to the LF and PF interfaces, then wh-
phrases must vacate the phase domain before it is transferred and rendered inaccessible.
(2) a. [CP What did he say [CP all (that) he wanted] ]?
b. [CP What did he say [CP (that) he wanted all] ]?
c. [CP What all did he say [CP (that) he wanted ] ]?
(McCloskey 2000:61)
However, subsequent developments in theoretical syntax exposed a number of issues with
the definition of phases. As originally formulated, phases are ?convergent objects? that are
both ?independent at the interfaces? and minimally complete in their ?propositional con-
tent? (Chomsky, 2000). These properties may have some intuitive underpinnings, but
they proved to be difficult to formalize in linguistic theory. Defining phases in terms of
convergence results in a look-ahead problem, for example, since the derivation?which is
built in incremental steps?has no access to its endgame form. That is, at any given point
of the syntactic derivation, the derivation cannot know whether the final product will (or
will not) result in an acceptable sentence. Interface independency was also found to be
too vague and arbitrary, particularly since phasehood diagnostics could not be formulated
in terms strictly concerning the interface in question. For example, diagnostics that were
meant to demonstrate a phase?s independence at the LF interface also demonstrated syn-
3
tactic behaviour. Finally, categorizing phases in terms of propositional content turned out
to be arbitrary. Transitive vPs, for example, were assumed to be phasal since the transitive
verb assigns a theta role to both the subject and object (minimally), while unaccusative
and passive vPs were assumed to not be phasal since these verbs lack subjects. How-
ever, it is not clear how theta role assignment for transitive vPs is more complete than
unaccusative and passive vPs (Citko, 2014; see also Legate, 2003). In all cases, all the
available theta roles are assigned within the confines of the verb phrase. If that is the
criterion for phasehood, the unaccusative/passive verb phrases should be phases just like
their transitive counterparts.
Alongside these concerns about phasehood formalization, discussion surrounding
the Phase Impenetrability Condition (PIC) suggested that the interior of phases was not as
impenetrable as previously thought. Under the original formulation of the PIC, the phase
interior is opaque as soon as the construction of the phasal projection is complete.
(3) Phase Impenetrability Condition
In phase ? with head H, the domain of H is not accessible to operations outside
? , only H and its edge are accessible to such operations.
(Chomsky, 2000:108)
This formulation was eventually reworked given that the model overgenerated unaccept-
able strings. For example, the Icelandic agreement system allows nominative direct ob-
jects to control agreement morphology on the matrix verb, as shown below. This relation
4
is not possible under the original definition of the PIC since T0 cannot target a vP-internal
argument.
(4) Henni ho?fu leiDst ?eir. ICELANDIC
3SG.FEM.DAT had.3PL bored.at 3PL.NOM
?She had found them boring.? (SigurDsson, 2002)
Given this development, the two versions of the PIC became known as Strong and Weak.
The Strong PIC (PIC1) directly corresponds to the original formulation of the PIC in
Chomsky (2000), while the implementation of the Weak PIC (PIC2; Chomsky, 2001)
allowed for the derivation of the Icelandic example above to be convergent. While both
definitions concern the phase head H and the phase domain HP, the Weak PIC further
considers the next-higher phase domain, i.e. ZP.
(5) a. The domain of H is not accessible to operations outside HP; only H and its
edge are accessible to such operations.
(Strong PIC/PIC1)
b. The domain of H is not accessible to operations at ZP; only H and its edge
are accessible to such operations.
(Weak PIC/PIC2)
(Chomsky, 2001:13?14)
In recent years, the notions of phases and the PIC have come under more scrutiny. As
a result, the literature features several alternative models that derive the effects of phases
while avoiding some of the bigger issues with standard phase theory (e.g. Epstein et al.
5
1998; Uriagereka 1999, 2003; Grohmann 2003, 2007; Epstein and, Seely 2006; Rack-
owski and Richards, 2005, Halpert, 2019). Some immediate questions concern the the-
oretical need for both versions of the PIC, and whether Transfer is needed as a separate
syntactic mechanism. Halpert (2019), for example, argues that phases are purely inter-
vention effects where the ?phase? is merely the closest goal for the operation at hand.
Under this view, the PIC is entirely epiphenomenal since syntactic material lower than
the ?phase? head is inaccessible due to standard locality considerations, namely, minimal-
ity. Furthermore, Halpert shows that the interior of phases can be accessed provided that
the higher goal is made transparent, e.g. under Cyclic Agree operations (Be?jar, 2003;
Be?jar and Rezac, 2009; Clem, 2018; Keine and Dash, 2018). Under this view, Transfer
is no longer a tenable idea, since the syntactic material in question can be accessed by
operations originating outside the ?phase? domain.
This dissertation contributes to this body of reanalyses of phasehood and the PIC. I
argue that some apparent instances of phasehood are merely intervention effects, and that
the PIC and Transfer are unnecessary components of syntactic theory. I show that, at least
in some languages, ?phasehood? correlates with ?-agreement. That is, if a language has a
?-probe on a head H, then H is a potential goal for ?-agreement. The crux of the evidence
comes from exceptional number agreement and person licensing patterns in Georgian, as
will be shown below.
First, plural number agreement on verbs in Georgian can appear as the suffix ?-t?.
The examples below illustrate a part of its distribution in one of two agreement paradigms
in Georgian: the basic agreement paradigm. In this paradigm, ?-t? can mark 1st and 2nd
6
person person plural subjects, as in (6), but it cannot mark 3rd person plural subjects, as
in (7).
(6) a. c?ven is da-v-p?at?iz?-e-t
1PL.ERG 3SG.NOM PRV-1-invite-1/2.AOR-PL
?We invited him/her.?
b. tkven is da-p?at?iz?-e-t
2PL.ERG 3SG.NOM PRV-invite-1/2.AOR-PL
?You (pl) invited him/her.?
(7) isini mas p?at?iz?-eb-en(-*t)
3PL.NOM 3SG.DAT invite-TS-3PL.PRES(-PL)
?They invite him/her.?
The number agreement restriction with respect to 3rd person plural subjects in the ba-
sic agreement paradigm is not found in inverse agreement, Georgian?s second agreement
paradigm. As shown below in (8), ?-t? can mark 3rd person plural subjects just in case
they appear in the inverse agreement paradigm, and only if the object is also 3rd person.
If the object is 1st or 2nd person, as in (9), ?-t? cannot mark the 3rd person plural subject.
Exceptional person agreement is also required for 1st and 2nd person objects in the in-
verse agreement paradigm, where they are marked by additional agreement affixes on an
auxiliary verb form ?ar? ?be?. These agreement patterns for such objects do not appear in
the corresponding configurations in the basic agreement paradigm.
(8) mat is u-q?var-t
3PL.DAT 3SG.NOM VER-love-PL
?They love him/her.?
7
(9) a. mat me v-u-q?var-v-ar-(*t)
3PL.DAT 1SG.NOM 1-VER-love-1-be.PRES-PL
?*They love me.?
b. mat s?en u-q?var-x-ar-(*t)
3PL.DAT 2SG.NOM VER-love-2-be.PRES-PL
?*They love you (sg)?.
I propose a position-based account of these agreement patterns and restrictions. As will be
shown below, the subject positions bear on the Agree relations that originate from two ?-
probes in the structure: a person-probe on v that searches for [PARTICIPANT] (i.e. 1st and
2nd person arguments) and a person- and number-probe on T that search for, respectively,
[PARTICIPANT] and [PLURAL] (see chapter 5 for more discussion on these points). In
the basic paradigm structure in (10a), subjects are introduced as external arguments in
the specifier of v, and do not intervene for (object) agreement between v0 and the theme
argument. In contrast, in the inverse paradigm structure in (10b), subjects are introduced
as experiencer arguments in Spec,ApplP. They are the closest goal to the probe on v0 and
trigger object agreement morphology, and also intervene for agreement between v0 and
the theme argument (see also Be?jar 2003).
(10) a. [ 0 0 0TP T [ ]PART[ ]PL [vP DP v [ ]PART [VP V DP ] ] ]
b. [TP T0 0 0 0[ ]PART[ ]PL [vP v [ ]PART [ApplP DP Appl [VP V DP] ] ] ]
Furthermore, the position-based account manifests particular licensing requirements in
Georgian for 1st and 2nd person arguments, i.e., arguments that bear a [PARTICIPANT]
feature. As will be further discussed in chapter 4, 1st and 2nd person arguments must
enter into an Agree relation with a ?-probe in their domain (Be?jar and Rezac, 2003). Li-
8
censing requirements are observed crosslinguistically, and are often linked to Person Case
Constraint (PCC) effects (Be?jar and Rezac, 2003; Coon and Keine, 2020; Preminger,
2014, 2019; Rezac, 2008). The PCC itself is a restriction on possible combinations of
internal arguments, and manifests as Weak, Me-First, Total, Strong, Super-Strong, and
Ultra-Strong varieties (Albizu, 1997; Anagnostopoulou, 2005; Bonet, 1991, 1994; Do-
liana, 2014; Graf, 2012; Haspelmath, 2004; Nevins, 2007). In Georgian, the PCC comes
in the Strong variety, which restricts the direct object of a ditransitive verb to 3rd per-
son arguments. Given that we independently observe splits in behaviour with respect
to 1st/2nd persons vs. 3rd person arguments in Georgian, we thus have reason to think
that [PARTICIPANT]-bearing arguments are subject to additional constraints concerning
?-probes and Agree.
Returning to the examples sentences in (10), the positions and licensing require-
ments of the subject and object yield intervention effects in (10b), but not in (10a). In
(10a), the 1st and 2nd person ([PARTICIPANT] bearers) themes can be licensed in situ via
long-distance Agree by v. In the configuration in (10b), however, [PARTICIPANT]-bearing
themes cannot be licensed in their base-generated positions, due to the presence of the in-
tervening subject DP in Spec,ApplP. Participant themes must instead move through the
vP edge for higher licensing, which is morphologically marked by x-/v- prefixed to ?ar?,
as in (11) below.
(11) [ 0 0 0 0TP T [PART] [vP DP[PART] v [ApplP DP Appl [VP V DP[PART]]]]]
9
In contrast, no person licensing requirement applies in the case of 3rd person themes, so
they remain in their base-generated position. Since Spec,vP is empty, the person probe
on T0 first encounters v0, shown in (12). Since v0 bears a ?-probe, it bears ?-features
and so it must count as a target by the higher agreement probe on T0. As an intervener,
v0 cannot immediately be ignored in favour of a lower ?-bearing goal. Although the
vP does not carry the valued feature that the person probe on T0 is seeking, this initial
agreement relation renders v0 transparent (Rackowski and Richards, 2005; Preminger,
2011; Van Urk and Richards, 2015). Under Cyclic Agree, where the Agree operation can
apply more than once under certain conditions, (Be?jar and Rezac, 2009; Keine and Dash,
2018; Clem, 2018) syntactic material that is lower than the now-transparent v0 is visible to
Agree (Halpert, 2019). With the intervener neutralized in this fashion, 3rd person plural
inverse subjects in Spec,ApplP are targeted by the number probe on T0 and marked by -t.
(12) [ 0 0TP T [PL] [vP v [ApplP DP[? , PL] Appl0 [VP V0 DP? ]]]]
One prediction that arises from this analysis revolves around trapping a participant theme
in its base-generated position. Under the proposed analysis, a derivation in which the
theme argument is trapped in such a low position should result in the loss of var/xar agree-
ment morphology. This prediction is borne out in causativized inverse verbs. Causativiz-
ing an inverse verb introduces a causer external argument in Spec,vP, which blocks move-
ment of participant themes through the vP edge. Their only option to escape the resulting
licensing problem is via an object camouflage strategy (Harris, 1981; Rezac, 2011), shown
in (13), which encases the unlicensed participant argument in a PP-like structure. Since
10
the participant theme arguments remain in their base-generated position in this case, the
auxiliary var/xar verb form is not required.
(13) a. Nino s?en c?emstavs g-a-q?var-eb-s
Nino.NOM 2SG.DAT 1SG.REFL 2-CAUS-love-TH-3SG.PRES
?Nino makes you love me.?
b. Nino me s?enstavs m-a-q?var-eb-s
Nino.NOM 1SG.DAT 2SG.REFL 1-CAUS-love-TH-3SG.PRES
?Nino makes me love you.?
Further, derivations with empty Spec,vP positions should allow the number probe on T0 to
target arguments inside the vP domain, provided the person probe on T0 first encounters?
and agrees with?v itself. As will be discussed in chapter 4, adversity causatives?which
lack external causers?provide such a test case (Lomashvili 2011). As shown in (14), 3rd
person plural causees indeed trigger -t.
(14) mat kaghald-i e-glej-in-eb-a-t
3PL.DAT paper-NOM APPL-tear-CAUS-TH-3.AOR-PL
?They are caused to tear paper.?
This dissertation results in two innovations. First, the proposed analysis of the Geor-
gian agreement system is a comprehensive model that captures both the basic and in-
verse agreement paradigms, while only using independently-motivated components such
as Cyclic Agree, intervention, and pronominal licensing. The Georgian agreement system
is highly complex, and most analyses have focused on only the basic agreement paradigm.
I argue that the position of the subjects with respect to the ?-probe on v0 derives the agree-
ment patterns found in the two paradigms. That is, basic agreement morphology arises
11
when the subject is base-generated above v0, and inverse agreement morphology arises
when the subject is base-generated below v0.
Second, the proposed analysis bears on the nature of phases and the PIC, as well
as Transfer. I argue that, in Georgian, vP appears to act as a phase because v0 bears
a ?-probe. This property translates to v0 bearing ?-features, which gives rise to ?-
intervention. In certain configurations in the Georgian inverse agreement paradigm, we
find that v0 does indeed intervene for person agreement, but independent Cyclic Agree
mechanisms in the language allow number agreement to target a vP-internal argument.
This suggests that phasehood, in some if not all cases, can be reanalyzed as ?-intervention.
Furthermore, the proposal necessitates a view where ?phases? do not undergo Transfer to
the interfaces given that syntactic material inside the ?phase? domain is a viable target for
?-agreement. Taken together, these points indicate that the PIC should be dispensed with
as a primitive of syntactic theory.
1.2 Roadmap of the dissertation
In Chapter 2, I present the empirical and theoretical motivations for phases as com-
putation domains. I discuss the ways in which phases were proposed to solve a prob-
lem arising from the Merge over Move Principle, which led to the development of the
Phase Impenetrability Condition where the interior of the phase was deemed inaccessi-
ble because that part of the syntactic structure had already been Transferred to the in-
terfaces. Although some properties of this intermittent Transfer could be detected by
long-distance successive cyclic movement, those diagnostics?and the formalization of
12
phases in general?were not without their own set of issues and concerns. I conclude
that these concerns are worrying enough to investigate alternative ways that phases might
derive from other independent aspects of the syntax.
In Chapter 3, I discuss several recent theoretical proposals that derive phasal bound-
aries as intervention effects. Under this view, the so-called phase is the closer goal for
Agree; ?phasehood? and the PIC thus fall under the rubric of general locality constraints
and, specifically, the effects of minimality. I compare and contrast two broad families
of approaches, one where the PIC is a syntactic primitive and another where the PIC is
derived as an epiphenomenon from A-over-A configurations. Specifically, A-over-A con-
figurations that are a subtype of minimality. I also discuss some predictions that tease
the two families of approaches apart. Namely, the latter approach rules out Transfer as
an additional component of the grammar, and so we should be looking for cases where
syntactic operations access an element that is still below the ?phasal boundary?. I argue
that Georgian number agreement provides such a test case.
In Chapter 4, I provide an overview of the morpho-syntactic properties of Georgian
(South Caucasian). I also discuss the clause structure of Georgian with respect to the po-
sition of the subject in two agreement paradigms: (i) in the basic agreement paradigm, the
subject is introduced as an external argument in Spec,vP, and (ii) in the inverse agreement
paradigm, the subject is introduced as a vP-internal applicative argument in Spec,ApplP. I
also describe the agreement patterns found in both paradigms, focusing on the distribution
of the verbal plural marker between the two.
In Chapter 5, I present a comprehensive analysis of the Georgian agreement sys-
tem. I argue that the position of the subject with respect to v derives the similarities and
13
differences between the basic and inverse agreement paradigms. I show that, in the in-
verse agreement paradigm, the vP layer is the first accessible goal for person agreement.
Once this relation is established, number agreement is then possible with a vP-internal
argument. Given the evidence that Agree crosses a so-called phasal boundary into its in-
terior, I conclude that phases and the PIC are purely A-over-A effects, and that Transfer
is unnecessary as an additional component of the grammar.
In Chapter 6, I extend the proposed analysis to Hindi-Urdu. Given that at least some
instances of phasehood are ?-intervention, we expect phasal boundaries to correlate with
?-features, which we would hopefully be able to independently detect, e.g. through the
relevant head acting as its own ?-probe. I argue that the lack of vP phase effects in
Hindi-Urdu (Keine, 2017) follows from the absence of a ?-probe on v0. Since v0 does not
bear ?-features, it cannot intervene for Agree, and thus does not act as a phasal boundary.
Note that argument is based on the lack of evidence for v bearing a ?-probe in Hindi-Urdu.
While the lack-of-evidence argument may raise some concerns, it at least derives Keine?s
result in a more principled manner since crosslinguistic variation in phasehood does not
have to be stipulated on a per-language basis. (For an argument that, in the domain of
agreement, absence of evidence is indeed evidence of absence, see Preminger, 2019).
Chapter 7 concludes.
14
Chapter 2: On phase theory
The notion of the ?bounding node? has long featured in a large part of generative
linguistic theory, with its preconception as the ?cycle? going back to pre-Minimalist Pro-
gram literature. For example, the notions of bounding nodes and barriers in, respectively,
the Extended Standard Theory and Government and Binding Theory are direct correlates
of modern phases. Many of the phasehood properties discussed today can be traced to
Chomsky (1973), which served to formally model islandhood and successive cyclicity ef-
fects. With the advent of the Minimalist Program, idiosyncratic bounding nodes and bar-
riers were redefined as the more conceptually broad phase, i.e. a pre-determined chunk
of structure built in the narrow syntax, and which exhausted a distinct lexical subarray
(Chomsky, 2000). In that work, phases were assumed to be constant in their properties
regardless of language or syntactic environment (with the one possible caveat being the
transitive vs. unaccusative/passive distinction). Phasal categories were (originally) tied
to propositional content, i.e. vP, CP, and DP as stated in Chomsky (2000). Since then,
the potential phasal categories have been extended to other parts of structure, e.g. PP
(Drummond et al., 2010; Kayne, 1999, 2004) and ApplP (McGinnis, 2001). An imme-
diate theoretical consequence of phase theory concerns the interior of the phase. That
is, if phases constitute chunks of the syntactic derivation that are incrementally built and
15
shipped to the LF and PF interfaces, then elements within the phase X may not be visible
to syntactic operations originating in phase Y. Chomsky (2001) refers to this property as
the Phase Impenetrability Condition (PIC), which stipulates that the complement of the
phase head is inaccessible.
In recent years, however, phases and the PIC have been reanalyzed as deriving from
other independent properties of the narrow syntax. In the strongest recharacterization of
phases and the PIC, they constitute an ephiphenomenon based on the A-over-A condition
(Chomsky, 2004), which is itself a general locality issue, arguably a particular subcase of
minimality (Halpert, 2019; Rackowski and Richards, 2005). Under these recharacteriza-
tions, the so-called phase is merely the closer goal, and the PIC arises as an intervention
effect. In other models, phases still exist as primitives of the narrow syntax, and the
PIC can be alleviated when syntactic mechanisms?such as Agree?allow the interior
be accessed under certain circumstances. In this sense, the phasal boundary can be un-
locked given that the initial Agree relation satisfies the Principle of Minimal Compliance
(Preminger, 2011b; Richards, 1998; van Urk and Richards, 2015). While I discuss these
recharacterizations in more detail in chapter 2, I will first provide an overview for the em-
pirical and conceptual motivations for phases leading up to and including the Minimalist
Program (Chomsky, 2000). This overview will also include a discussion of the issues and
problems with defining phases and the PIC, which will set the foundation for their recent
reanalyses.
16
2.1 Motivating phases
In the Minimalist Program (MP), language, or the language faculty, is modelled as a
system of syntactic structures and operations that are subject to two sets of constraints and
interfaces: (i) conceptual-intensional (i.e. LF), and (ii) perceptual-sensorimotor (i.e. PF).
The syntactic structure is built by the application of two major operations: Merge, which
creates one syntactic object by combining two syntactic objects, and Move, which moves a
syntactic object from one position in the structure to another. Operating alongside Merge
and Move is Agree, which establishes a relation between a set of uninterpretable/unvalued
features on a probe P with the corresponding interpretable/valued features on a goal G.
The language faculty (i.e. the narrow syntax) interacts with the interfaces at specific points
in the derivation, where a pre-determined chunk of the syntactic structure is shipped?or
Transferred?to the interfaces. Once the content of the phase has been Transferred, that
content cannot be accessed by the narrow syntax. This derives the Phase Impenetrability
Condition (PIC), which has become a crucial component of phase theory and its rechar-
acterization in recent years.
The main conceptual advantage for phases is a computational one. Given that the
language faculty is designed around efficiency and minimal computations under MP,
Chomsky argued that phases reduce the computational cost of derivations (Chomsky,
2000:110?112). When phases made their first appearance in the Minimalist Program,
they were introduced as a solution to a problem concerning the acceptability contrasts be-
low, provided the lexical items available for the derivation below in (2) (Chomsky, 2000;
see also discussion in Citko, 2014). This collection of lexical items, i.e. the Numeration,
17
provided the building blocks for the derivation; no additional items could be used for that
stage of the derivation, and all items in the Numeration must be exhausted. Note that, in
each case in (1a,1b), the derivation is technically sound. In (1a), the EPP feature of the
embedded T is checked by the Merge of the expletive there in Spec,TP. In comparison, the
EPP feature is checked in (1b) via Move: the embedded [many dogs] moves to Spec,TP.
In either derivation, all uninterpretable features are checked/value and all elements in the
Numeration are used. Given that each derivation is convergent, the model needs some
other principle to explain the acceptability of (1a), but not (1b).
(1) a. Therei are likely ti to be [many dogs] at the park right now.
b. *There are likely [many dogs]i to be ti at the park right now.
(2) a. N = {there, T, are, likely, to, be, many, dogs, at, the, park, right, now}
b. [TP toEPP be many dogs at park]
The Merge over Move (MOM) principle was proposed in order to derive the acceptability
contrasts above (Chomsky, 1995). This principle states that merging elements into the
structure is more computationally efficient than moving an element from one syntactic
position to another. Returning to the cases in (1a) and (1b) above, we see that, at the stage
in the derivation where the EPP feature of the embedded T needs to be checked, MOM
dictates that it is more computationally advantageous to Merge an expletive subject (=1a)
than it is to Move a nominal argument into another position (=1b). But, although MOM
derived the contrasts in (1), it could not derive the acceptability contrasts in the more
complex strings in (3) below.
18
(3) a. There is a strong likelihood that there will be [many dogs] at the park right
now.
b. *Therei is a strong likelihood that ti will be [many dogs] at the park right now.
The assumed Numeration for the strings above is provided in (4).
(4) N = {there, is, a, strong, likelihood, that, many, dogs, will, be, at, the, park, right,
now}
Given the Merge over Move principle, the expletive there enters in the derivation at the
stage of the embedded Spec,TP. Since the EPP and uninterpretable features are checked/valued
by the expletive subject, [many dogs] remains in its base-generated position. At the point
in the derivation concerning the matrix T, the embedded expletive subject is the appropri-
ate element to occupy the matrix Spec,TP given Merge over Move, and minimality. This
derivation, although technically sound, results in the unacceptable string in (3b). The
phase, i.e. a derivational cycle, was proposed as the solution to this problem. That is,
the Numeration was recharacterized as a set of sets, rather than simply a set of primitive
elements, as shown below.
(5) a. N = {a, b, c, d, e, f}
b. N = { {a, b}, {c, d}, {e, f} }
Given the additional structure to the Numeration, the problem described above is allevi-
ated by simply manipulating the content of each subset. If the expletive subject there is
19
not included in the subarray for the embedded clause, as shown below, then there is no
derivation that will generate the string in (3b) above.
(6) N = { {there, is, a, strong, likelihood}, {that, will, be, many, dogs, at, the, park,
right, now} }
At the point of the derivation when the EPP and uninterpretable features on the embed-
ded T needs to be checked, only [many dogs] is available since that is the only DP in
the subarray. So, [many dogs] Moves to the embedded Spec,TP position. The expletive
subject there then checks the relevant features of the matrix T, which results in a conver-
gent derivation. Note that the importance of structuring the Numeration into subarrays is
centred around having multiple computational domains in the derivation. That is, even
if the subarray corresponding to the embedded clause contained an expletive subject, the
derivation would still yield an acceptable sentence.
(7) a. N = { {there, is, a, strong, likelihood}, {that, there, will, be, many, dogs, at,
the, park, right, now} }
b. There is a strong likelihood that there will be many dogs at the park right
now.
At this point in the development of the theory, phases were thus essentially lexical sub-
arrays (Chomsky, 2000), i.e. a subset of the overall lexical array specified for the entire
derivation. However, defining phases in this manner is ad hoc: phases are essentially sub-
arrays, but the subarrays are divvied up in whatever fashion delivers the desired empirical
20
result. While various solutions were proposed to address this issue?which will be dis-
cussed in the following section?the most prominent formal characterization of phases
involved tying phasal status to syntactic heads. That is, certain syntactic heads are also
phase heads. Since syntactic heads were also thought to be the loci of uninterpretable fea-
tures, phasehood came to signal trigger points of syntactic operations such as Agree and
Case assignment (Chomsky, 2000; Gallego, 2010; Legate, 2012; Miyagawa, 2011). Im-
portantly, phase heads determine when the structure undergoes Transfer to the interfaces,
which leads to the following questions: If phase heads determine when the structure is
shipped to the interfaces, what happens to the syntactic material inside the phase? An-
swering this question led to the formulation of the Phase Impenetrability Condition (PIC),
which I turn to in the next section.
2.2 The Phase Impenetrability Condition (PIC) and Agree
This section provides an overview of the empirical motivation and theoretical im-
plementation of the Phase Impenetrability Condition (PIC), and discusses its major revi-
sions throughout the development of linguistic theory. The two versions of the PIC (i.e.
Strong and Weak) address different syntactic operations, such as Move and Agree, and
they also cover different empirical phenomena. I discuss the details of these properties
below.
Broadly construed, the PIC renders the syntactic material inside of a phase inac-
cessible to outside operations. This effect was tied to the phase head (i.e., the syntactic
head of the phasal projection), which determined when the syntactic structure was to be
21
shipped to the interfaces. The PIC seemed intuitive, since, if the syntactic material below
a phase head had already undergone Transfer by the time a syntactic operation outside
the phase head occurs, then there is simply no syntactic material to access. Any element
outside of the phase head, though, is accessible, which meant that the specifier position
of the phase can be used as an ?escape hatch? for those elements needing to move out of
the phase domain. This is schematized below.
(8) a. [ZP Z ... [XP X ... [HP ? [ H YP ] ] ] ]
b. [CP C ... [TP T ... [vP DP [ v VP ] ] ] ]
(Chomsky, 2001:13)
Note that, in the abstract structure in (8a), there is an intervening non-phase head between
two phase head. This construction allows us to test the effects of the PIC. In practice, the
relevant phase heads are C and v, and the non-phase head is T, as in (8b). In the original
formulation of the PIC in Chomsky (2000), the interior of a phase is inaccessible as soon
as the construction of the phase is complete.
(9) Phase Impenetrability Condition
In phase ? with head H, the domain of H is not accessible to operations outside
? , only H and its edge are accessible to such operations.
(Chomsky, 2000:108)
Under this definition, the complement of H is spelled out (i.e. shipped to the interfaces
via Transfer) as soon as H no longer projects and HP is built. There are two important
22
consequences for this definition. First, syntactic material being inaccessible means that
any unvalued/unchecked features will remain unvalued/unchecked for the remainder of
the derivation. Second, the specifier of H remains accessible to outside operations since
only the complement of H undergoes Transfer.
(10) Phase edge and spell-out domain
XP
X HP
? H?
H YP
Under this formulation and configuration, then, the only way for elements inside of
the phase (in the complement of the phase head) to be accessible to outside operations
is to Move out of the Spell-Out domain, which must be done by first Moving to the
edge of the phase. But, given the assumptions of the Minimalist Program, movement
has to be triggered by some obligatory feature or design of the system (cf. Move ?;
Lasnik and Saito, 1994). As a locality constraint, the PIC was assumed to enforce this
movement. An EPP feature was additionally presumed to be present on those phase heads
that necessitated movement of an element within the phase to the edge. While this feature
assignment may seem arbitrary at this point of the discussion, we will see some evidence
23
for its presence in the structure through the prism of evidence for successive-cyclicity in
movement.
(11) The head H of phase Ph may be assigned an EPP-feature.
(Chomsky 2000:109).
However, the original formulation of the PIC was found to be too strong. For example, the
PIC did not allow for agreement with nominative objects, since that agreement relation
would span the vP phasal boundary. However, agreement with nominative objects was
found to be possible in Icelandic, which will be further discussed below. This empirical
evidence led to a major revision of the PIC, resulting in Strong and Weak versions.
(12) a. The domain of H is not accessible to operations outside HP; only H and its
edge are accessible to such operations.
(Strong PIC/PIC1)
b. The domain of H is not accessible to operations at ZP; only H and its edge
are accessible to such operations.
(Weak PIC/PIC2)
(Chomsky, 2001:13?14)
While the overall structural configuration is the same between the two versions, where
they differ is on the timing of when the interior of the phase becomes inaccessible. Under
the Strong PIC (PIC1), the complement of H is spelled out as soon as HP is built. In
contrast, under the Weak PIC (PIC2), the complement of H is spelled out when the next
24
phase head Z is merged. Testing the predictions of the two different versions of the PIC
thus centres on structures where there is a non-phase head X in between Z and H. Under
the Strong PIC, X should not be able to probe below H, but under the Weak PIC, X should
be able to probe below H. In natural language, Weak PIC effects should be observable,
for example, in cases where a non-phase head T accesses a vP-internal DP (since the
complement of v will only undergo Transfer when C is merged).
(13) Probe?goal relations under the Strong PIC
ZP
Z
X
?
H YP
7
WP
25
(14) Probe?goal relations under the Weak PIC
ZP
Z
X
?
H YP
WP
To reiterate: in order to see which version of the PIC makes the right predictions,
we need a language where Agree is possible between T and a nominative direct object.
Looking at Icelandic, we see that it is indeed possible to establish Agree between T and a
nominative direct object (a pattern that can also be found in other languages, e.g. Polish,
German, and Georgian). These languages have dative subjects that pass all subjecthood
diagnostics, and the verb shows agreement with the direct object that is marked with
nominative (SigurDsson, 2002, 1996; SigurDsson and Holmberg, 2008; Taraldsen, 1996,
1995).
(15) Henni ho?fu leiDst ?eir.
3SG.FEM.DAT had.3PL bored.at 3PL.NOM
?She had found them boring.? (SigurDsson, 2002)
In the example sentences above, the VP cannot have been spelled out since T agrees with
the nominative direct object. This agreement relation is not possible under the Strong
26
PIC, since the VP spells out (and its contents are therefore inaccessible) as soon as the vP
is complete.
However, this is not a knockdown argument in favour of the Weak PIC over the
Strong. We can get the same effects by varying which subarray contains T: Richards
(2011) argues that that there is no Weak or Strong PIC, per se, but rather we see the effects
of there being two different versions. Under this analysis, the different PIC versions are
epiphenomenal since the effects arise from which subarray T is part of, as demonstrated
below. When T is part of the same subarray as C, we get Strong PIC effects; when C is
part of the same subarray as v, however, we get Weak PIC effects instead.
(16) a. N = { {C, T}, {v, V} } Strong PIC/PIC1
b. N = { {C}, {T, v}, {V} } Weak PIC/PIC2
There are two advantages to model proposed in Richards (2011). First, the model cir-
cumvents the Strong vs. Weak PIC issue by deriving the differences as epiphenomenal
effects. Second, the model avoids the mismatch between (i) what a phase is, and (ii) what
actually gets transferred to the interfaces. Under the conception of phases that we have
been discussing so far, a phase is identified by the (syntactic) head, but it is the comple-
ment of that head that is spelled out and rendered inaccessible. By characterizing phases
as subarrays, we avoid this mismatch issue since the subarray itself is the unit spelled
out. Despite these advantages, however, characterizing phases as subarrays leads us back
to the circular argument of phases being subarrays, which are are themselves phases. I
27
return to this issue in the next section, focusing now on the question of which syntactic
mechanisms are constrained by the PIC.
While there remains a question of whether all syntactic mechanisms are subject to
the PIC, a more specific question asks whether the Move/Internal Merge and Agree op-
erations are constrained by the PIC. Conceptually, constraining the Move/Internal Merge
and Agree operations by the PIC seems reasonable given that these operations are purely
syntactic, and phases are syntactic entities. As with many aspects of phase theory, this
question is still up for debate (Bhatt, 2005; Bos?kovic?, 2007; Bos?kovic? and Lasnik, 2003;
Kratzer, 2009; Wurmbrand, 2017). In this thesis, I will assume that Agree is constrained
by (something like) the PIC. The basis for this assumption draws from Polinsky and Pots-
dam?s (2001) analysis of long distance agreement (LDA) in Tsez. Consider (17), where
it seems that Agree into finite complement clauses is possible in Tsez.
(17) eni-r [uz?-a? magalu b-a?c?-ru-?i] b-iy-xo
mother-DAT boy-ERG bread.III.ABS III-eat-PST.PRT-NMZ III-know-PRES
?The mother knows that as for the bread, the boy ate it.?
(Polinsky and Potsdam, 2001)
However, Polinsky and Potsdam show that that the edge of the embedded domain
must be available if LDA is to target an element originating in the embedded domain. As
shown below, LDA is impossible when the embedded clause has something else, e.g. a
wh-phrase, at its edge.
(18) eni-r [?u micxir b-ok?a?k?-ru-?i] r/*b-iy-xo
mother-DAT who.ERG money.III.ABS III-steal-PST.PRT-NMZ IV/III-know-PRES
28
?The mother knows who stole the money.?
(Polinsky and Potsdam, 2001)
In order to derive these patterns, Polinsky and Potsdam argue that LDA in Tsez is fed by
covert movement of the downstairs absolutive target to the edge of its own CP. This, in
turn, suggests that LDA is very much subject to (something like) the PIC, which allows
us to use ?-agreement as a diagnostic of phasal status. I will revisit this discussion of
?-agreement with respect to phase theory in chapter 3, and turn now to an overview of
phasehood diagnostics.
2.3 Diagnosing phases
This section discusses the most widely used phasehood diagnostics, and shows how
they have been applied in specific languages. Assuming phases are syntactic chunks that
are shipped to the interfaces, many diagnostics address the independence of that syntactic
chunk at LF and PF. This is not without issue, however, since we do not really know
what it means to be independent at the PF or LF interfaces. One potential way forward
is to look at the properties that the standard phases?vP, CP, and DP?have in common.
Those common properties could then be phasehood properties, which could be used to
diagnose other phases. However, it is not clear why vP, CP, and DP should be considered
standard phases in the first place, since it is logically possible that other phrases could be
phases as well (see section 2.4 for more discussion on this point). Characterizing those
properties common to ?standard? phases as phasal properties thus relies on us having an a
priori categorization of phases vs. non-phases, which leads to circular reasoning. Given
29
these difficulties concerning interface independency and a priori phase comparison, most
diagnostics focus on syntactic properties. Broadly, this family of diagnostics focus on
whether movement out of XP has to proceed through the edge of XP, i.e. the edge property
of phasal boundaries.
(19) a. [ZP Z ... [HP ? [ H YP ] ] ]
b. The domain of H is not accessible to operations at ZP; only H and its edge
are accessible to such operations.
(Strong PIC/PIC1)
(Chomsky 2001:14)
Syntactic diagnostics typically test whether the moved element can be (i) interpreted at
the edge of XP, (ii) pronounced at the edge of XP, or (iii) strand anything at the edge of
XP.1 Note that islandhood does not factor into phasehood diagnostics, since the category
of a phrase does not always determine whether it is an island. For example, interrogative
CP complements of verbs are islands (wh-island; Chomsky, 1973) whereas declarative
CP complements of verbs are not. Similarly, (some) definite DPs are islands (Matushan-
sky, 2005), whereas indefinite ones, at least in languages like English, are not (for further
discussion on islandhood, see Mu?ller, 2011). Furthermore, attempting to block move-
ment out of a domain by filling the specifier position of the island phrase is subject to
language-specific properties, since languages vary in whether multiple specifiers are al-
lowed (Boeckx and Grohmann, 2007; Richards, 1999, 2001, 1997). Given the many in-
1In addition to the movement tests, other diagnostics test focus on whether XP is a domain for feature
valuation, or whether X is a source of uninterpretable features. I will not discuss the ramifications of these
diagnostics here, as they will form the basis for much discussion in section 3.
30
dependent properties that are specific to islands, phasehood diagnostics focus on whether
movement out of phases has to proceed through the edge. In the remainder of this sec-
tion, I will discuss the application of the relevant movement diagnostics with respect to the
edge property as they manifest in the CP and vP domains, in three families of phenomena:
(i) edge-filling effects, (ii) reconstruction effects, and (iii) wh-quantifier float.
Turning first to the edge-filling property of CP, we see that this property is active
in successive cyclic long-distance wh-movement. Configurationally, this looks like (20a)
below, with a natural language example in (20b).
(20) a. [CP WHi [C? C [TP ... [CP WHi [C? C [TP ... [CP WHi [C? C [TP ... WHi ... ] ]
] ] ] ] ] ] ]
b. [CP Who did [TP Nino say [CP who that [TP Dato thought [CP who that [TP
Gio saw who ] ] ] ] ] ]?
Edge-filling effects are a useful diagnostic in this respect, since, under the PIC, elements
have to move out of phases in order to be accessible to operations that stem from out-
side of the phase. Since the wh-element is pronounced at the front of the sentence, the
only way for the embedded wh-phrase to reach the matrix Spec,CP position is to pro-
ceed through each intermediate phase edge. Furthermore, we know that wh-movement is
indeed targeting the edge of the phase since filling that position results in an unaccept-
able string (that is, assuming that the language in question does not allow for multiple
specifiers, as in English).
(21) a. *What did you think [CP when Nino devoured what when]?
31
b. *This is the book which Nino wondered [CP who wrote which].
c. *How did you wonder [CP who fixed the sink how]?
Before I turn to the discussions concerning reconstruction effects and wh-quantifier float
as diagnostics of phasal status, I will provide a caveat that these diagnostics should be
viewed with some skepticism. Although reconstruction effects and wh-quantifier float
both occupy prominent roles in the literature on phases (and they are discussed here as
such), these diagnostics are flawed as diagnostics for phasehood in particular. Recon-
struction effects and wh-quantifier float only show where syntactic elements can stop,
not where they must stop; but, phase theory must predict where syntactic elements must
stop. These diagnostics thus provide, at best, circumstantial support for the phasehood of
the relevant category. That is, there is a question of why intermediate Spec,CP positions
consistently behave as possible stopping points for elements undergoing movement; and
one possible answer is that these positions are possible stopping points because they are
obligatory stopping points.
With this caveat in mind, i will now turn to a discussion of reconstruction effects.
Under syntactic approaches of reconstruction, wh-phrases can only be interpreted in in-
termediate Spec,CP position if they have passed through that position. In this sense,
reconstruction is just the interpretation of a lower copy: as an anaphor, that lower copy is
bound by the most local binder. This is demonstrated below.
(22) Ninoi asked [CP [which picture of herselfi] Mariam j thought [CP [which picture
of herself j] that Tamark liked [which picture of herselfk] ] ].
32
Similarly, Fox (1999) shows that reconstruction feeds Principle C effects. In (23a), the
wh-phrase can be interpreted in the intermediate Spec,CP position. This is not possible
in (23b), since the R-expression Ms. Brown2 is c-commanded by she2. Note that these
readings can be derived whether or not the wh-phrase can or must stop in the intermediate
Spec,CP positions.
(23) a. [Which (of the) paper(s) that he1 gave to Ms. Brown2] did every student1
hope t ? that she2 will read t?
b. [Which (of the) paper(s) that he1 gave to Ms. Brown2] did she2 hope t ? that
every student1 will revise t?
(Fox 1999:173, from Lebeaux, 1990)
Finally, we also see instances of wh-quantifier float where a quantifier associated with
a wh-pronoun can be stranded at the edge of intermediate CPs. Again, we cannot use
this phenomenon as a diagnostic of phasehood, though it has been proposed as such in
the relevant literature. As already mentioned in chapter 1, McCloskey (2000) reports this
property in West Ulster English, shown below.
(24) a. What all did you get for Christmas?
b. What did you get all for Christmas?
(McCloskey 2000:58)
Crucially, quantifier float is not limited to just the base-generated or matrix positions. As
demonstrated below in (25), the quantifier can appear in intermediate Spec,CP positions
33
as well. This suggests that the wh-quantifier is indeed moving through the edge of each
CP.
(25) a. What did he say all (that) he wanted?
b. What did he say (that) he wanted all?
c. What all did he say (that) he wanted?
(McCloskey 2000:61)
In summary, we see that phasehood diagnostics typically focus on the availability of the
phase edge position (Spec,CP and Spec,vP). This focus derives from the notion that phase-
internal elements must vacate the phase domain in order to be accessible to syntactic op-
erations and mechanisms that originate outside of the phase. As discussed in this section,
edge-filling effects suggest that Spec,CP must be a landing site for phrases that move
out of each intermediate phase domain, and reconstruction effects and wh-quantifier float
provide circumstantial support for this view. We will also see that there is some evidence
that Spec,vP is an escape hatch as well, as argued to be the case in Dinka by van Urk
and Richards (2015). Note that the phasehood diagnostics discussed thus far all concern
the application of a syntactic operation out of a phase domain (i.e. Move); in chapter 5, I
will show a complementary set of diagnostics that concerns the application of a syntactic
operation into a phase domain (i.e. Agree).
34
2.4 Issues with defining phases
In this section, I discuss remaining issues with the formalization of phasehood. I
will begin first with an overview of the difficulties concerning the definition of phases,
and then turn to issues surrounding their formal properties. This section ends by laying
out some questions concerning the very nature of phases, and asking whether the Phase
Impenetrability Condition is theoretically necessary.
In their original formulation, phases were assumed to be ?convergent objects? that
were ?independent at the interfaces?, and that were complete in their ?propositional con-
tent? (Chomsky, 2000). However, as alluded to throughout this section, each of these
definitions comes with their own set of issues. First, defining phases in terms of conver-
gence requires the syntax to look-ahead and evaluate potential derivations in terms of their
eventual (un)acceptability. Since phases were meant to ensure computational efficiency,
this look-ahead problem raised serious concerns about defining phases in terms of conver-
gence. Further, Preminger (2011a, 2014) shows that the idea of feature-checking being
crucial for convergence?which is the basis of Chomsky?s convergence-based attempt to
define phases?is flawed in the first place.
Second, the assumption that phases are independent at the interfaces is too vague,
given that many (if not all) of the diagnostics concerning the behaviour of phases at PF
and LF are intertwined with syntactic properties as well. Additionally, we saw that two
of the three standard diagnostics only provide circumstantial support for phasehood. That
is, reconstruction effects and wh-quantifier float only show that moving a phrase through
the edge position is a possibility, not that it is a necessity.
35
Finally, determining when a syntactic object is complete in its propositional content
is arbitrary: (Chomsky, 2000) proposes that vP is a phase since its theta roles are assigned,
and that CP is a phase since it includes both tense and force. According to Chomsky
(2000), only transitive and unergative vPs constitute phases; neither unaccusative nor pas-
sive vPs are phases since they lack external arguments. Yet, all of the available theta roles
are indeed assigned across all of these vPs regardless of whether an external argument is
selected or not. As for CPs, there is no independent motivation for grouping both tense
and force together in this definition; one could imagine a theory where tense comprises its
own phasal boundary to the exclusion of force. Gallego (2006) argues for this very view
based on phase-like effects in the TP domain in Spanish and other null subject languages.
Given these issues and concerns surrounding the nature of phases, we must then
question the necessity for the subsequent proposals that developed out of phase theory.
For example, does linguistic theory necessitate the incorporation of both the Strong and
Weak formulations of the Phase Impenetrability Condition? If we push this question
even further, do we need the PIC at all? At this stage of phase theory, there are several
conflicting accounts about the nature of phases; in some models, phasal boundaries can
be manipulated via head-movement (den Dikken, 2007; Gallego, 2006, 2010); in other
models, only specific types of XPs are phases depending on their syntactic environment
(Bos?kovic?, 2014). As will be discussed in the next chapter, others have proposed that there
are no phases at all?rather, what look like the effects of phases are in fact simply A-over-
A effects (Halpert, 2019; Rackowski and Richards, 2005). That is, the ?phase head? is just
the closest goal for a syntactic operation, and the interior of the ?phase? is inaccessible due
to the independent principle of minimality (Rizzi, 1990, 2001). Under this view, phases
36
are not a syntactic object and so the issues that they raise are unproblematic in these A-
over-A approaches. Analyses such as those proposed by Rackowski and Richards (2005)
and Halpert (2019) may thus constitute the better model for linguistic theory, depending
on the strength and accuracy of the empirical predictions and evidence.
37
Chapter 3: Neutralizing phasal boundaries
There are two major approaches to phase unlocking (i.e., the neutralization of phase
boundaries) that I will discuss in this dissertation. One family of approaches derives
phasehood as an intervention effect (e.g. Abels, 2003; Rackowski and Richards, 2005;
Halpert, 2019), where the interior of the so-called ?phase? is inaccessible due to general
principles of syntactic locality: the ?phase? head is a closer goal than anything properly
contained inside the phase. A second family of approaches (e.g. Preminger, 2011; Van
Urk and Richards, 2015) assumes that the Phase Impenetrability Condition is a syntactic
primitive, which can be alleviated by establishing Agree with the phasal head. This chap-
ter discusses these two approaches in detail, and outlines the theoretical issues at stake
under each set of assumptions.
3.1 The Phase Impenetrability Condition as an intervention effect
This section discusses a major approach of neutralizing phasal boundaries that de-
rives the PIC as an epiphenomenal intervention under locality effect, one derived from
minimality and intervention. Under these models, phase boundaries are reanalyzed as in-
terveners for syntactic operations, which block potential lower goals from being targeted
(see e.g. Rackowski and Richards 2005, Halpert 2019, among others). Some aspects of
38
these ideas are found in Abels (2003) as well, particularly in his section 2.2. In this
section, I discuss the phase unlocking mechanisms proposed in two analyses: (i) Taga-
log wh-movement as proposed in Rackowski and Richards (2005), and (ii) Zulu (Bantu)
hyper-raising as proposed in Halpert (2019).
Beginning first with Rackowski and Richards (2005), we see that, in Tagalog, the
verb may agree with the subject or with the embedded clause. Rackowski and Richards
assume that this relation is marked via the case-sensitive agreement morphology. As
shown in (1a), nominative (NOM) agreement morphology surfaces if the verb agrees with
the subject. In contrast, accusative (ACC) agreement morphology appears if the verb
agrees with the embedded clause, as in (1b).
(1) a. M-agsa-sabi ang kalaba [na masarap ang bulaklak]
NOM-ASP-say ANG water.buffalo that delicious ANG flower
?The water buffalo will say that the flower is delicious.?
b. Sa-sabih-in ng kalaba [na masarap ang bulaklak]
ASP-say-ACC CS water.buffalo that delicious ANG flower
?A/The water buffalo will say that the flower is delicious.?
(Rackowski & Richards 2005:586)
The variation in verbal agreement morphology has important consequences for the avail-
ability of wh-movement. As shown below in (2a), wh-extraction from an embedded CP
is only possible when the verb has agreed with the clause. If the verb agrees with the
subject, as in (2b), wh-extraction is prohibited.
(2) a. Kailan sa-sabih-in ng sundalo [na u-uwi ang pangulo]?
when ASP-say-ACC CS soldier that NOM.ASP-go.home ANG president
?When will the soldier say [that the president will go home ]??
39
b. *Kailan m-agsa-sabi ang sundalo [na u-uwi ang pangulo]?
when NOM-ASP-say ANG soldier that NOM.ASP-go.home ANG president
Intended: ?When will the soldier say [that the president will go home ]??
(Rackowski & Richards 2005:586)
Rackowski and Richards (2005) take this agreement restriction on wh-extraction to indi-
cate that the embedded clause is an intervener. That is, the CP is the first goal that is tar-
geted by the syntactic operation responsible for moving the embedded wh-element out of
the embedded clause. The intervener must be rendered inert if the embedded wh-element
is to be extracted. Rackowski and Richards (2005) propose that Agree preconditions
wh-extraction?that is, agreement with the intervening CP layer renders it invisible for
subsequent operations. The set of assumptions that derive this effect is provided below.
(3) a. A probe must Agree with the closest goal ? that can move.
b. A goal ? can move if it is a phase.
c. A goal ? is the closest one to a probe if there is no distinct goal ? such that
for some X (X a head or maximal projection), X c-commands ? but not ? .
d. Once a probe P is related by Agree with a goal G, P can ignore G for the rest
of the derivation (Hiraiwa, 2001; Richards, 1998).
e. v has a Case feature that is checked via Agree. It can also bear EPP-features
that move active phrases to its edge.
f. C[+wh] has a [+wh] feature that is checked via Agree (and sometimes Move).
(Rackowski and Richards, 2005:582)
40
Under this set of assumptions, the Phase Impenetrability Condition is derived as an A-
over-A effect. Specifically, the type of A-over-A configuration adopted here crucially
requires the intervener to dominate (i.e. contain) the goal. This particular structural re-
lation is important for the PIC-as-intervention model since the probe must encounter the
phrase that behaves like a ?phasal? boundary. In the structure below, for example, the
interior of the vP phase domain cannot be accessed because (i) the vP can move, and (ii)
the vP dominates a wh-feature.
(4) The PIC as A-over-A
CP
...
C[+wh] TP
...
T vP
...
v whP
In order for the embedded wh-phrase to be extracted, the vP phase boundary must
first be rendered transparent. Being successfully targeted for Agree meets this require-
ment, and the agreement morphology on the matrix verb in (5) reflects this Agree relation.
41
(5) a. Kailan sa-sabih-in ng sundalo [na u-uwi ang pangulo]?
when ASP-say-ACC CS soldier that NOM.ASP-go.home ANG president
?When will the soldier say [that the president will go home ]??
b. *Kailan m-agsa-sabi ang sundalo [na u-uwi ang pangulo]?
when NOM-ASP-say ANG soldier that NOM.ASP-go.home ANG president
Intended: ?When will the soldier say [that the president will go home ]??
(Rackowski & Richards 2005:586)
As demonstrated in the Tagalog sentences above, the embedded wh-phrase can only be
extracted if the embedded clause has been Agreed with. Under this analysis, the PIC is
an epiphenomenon: the interior of the phase in accessible due to general principles of
locality. That is, the phase head constitutes the closest goal, which derives the effect of
the PIC. The Tagalog data shows that, if the phase head is rendered transparent under
Agree, the interior of the phase can indeed be accessed. This suggests that the PIC is not
a syntactic primitive.
Following the conclusions of Rackowski and Richards (2005) with respect to the
PIC, I will now turn to the analysis of Zulu raising proposed in Halpert (2019). Halpert
argues that Zulu hyper-raising also instantiates an A-over-A effect. That is, the ?phase?
itself is a closer goal for agreement, which bleeds the targeting of an XP lower in the
structure. Under this view, the head of the so-called phase bears a feature which satisfies
the initial Agree operation from a higher probe: if no further searches are triggered, then
the result is a phasal boundary effect. If a second cycle of Agree is triggered, however,
then an XP lower in the structure will be targeted for agreement. I discuss this analysis in
more detail in the remainder of this section.
42
Note that, conceptually speaking, Cyclic Agree does not appear to be better vis-a?-
vis look-ahead than phases, since subsequent Agree operations are triggered by an unval-
ued probe that does not crash the derivation. However, Preminger (2011a, 2014) shows
that attempting to derive convergent structures under feature-checking accounts is flawed.
Additionally, look-ahead was problematic in phase theory with respect to computational
efficiency. Building syntactic structures in phases meant that the derivation could not
access its endgame form, and so the computational load was lighter than an alternative
theory where the derivation could track whether the relevant steps would eventually result
in convergence. Look-ahead was never part of the motivation for Cyclic Agree, and so
potential look-ahead issues are not as damaging for these models.
Returning now to Halpert?s analysis of Zulu hyper-raising, consider the data pro-
vided below. As shown in (6a-b), raising out of an embedded CP clause is possible in
Zulu, but raising out of an embedded infinitival TP is prohibited, as in (6c).
(6) a. ku-bonakala [ukuthi uZinhle u-zo-xova ujeqe]
17S-seem that AUG.1Zinhle 1S-FUT-make AUG.1steam.bread
?It seems that Zinhle will make steamed bread.?
b. uZinhlei u-bonakala [ukuthi ti u-zo-xova ujeqe]
AUG.1Zinhlei 1S-seem that ti 1S-FUT-make AUG.1steam.bread
?It seems that Zinhle will make steamed bread.?
c. *uZinhlei u-bonakala [ti uku-(zo-)xova ujeqe]
AUG.1Zinhlei 1S-seem ti INF-(FUT-)make AUG.1steamed.bread
Intended: ?It seems that Zinhle will make steamed bread.? (Halpert
2019:124)
43
The raising profile of Zulu stands in contrast to the raising profile of English, where rais-
ing out of embedded CPs is prohibited, as in (7), and raising out of embedded infinitival
TPs is possible, as in (8).
(7) a. It seems that John eats pizza.
b. *Johni seems (that) ti eats pizza.
(8) a. Johni seems ti to eat pizza.
b. *It seems John to eat pizza.
Halpert derives these patterns under two assumptions: (i) both TPs and CPs bear
?-features, and as such are goals for ? -agreement from matrix T; and (ii) only TPs can
satisfy the EPP in Zulu. Expanding first on (i), we see below that infinitival TPs, like DP
arguments, control both subject and object agreement morphology in Zulu. The sentences
in (9) illustrate this phenomenon in the subject agreement domain: (9a) shows that a
preverbal DP argument that is a member of noun class15 triggers ku- subject agreement
morphology, and (9b) shows that the same agreement marking is required when Spec,TP
is filled by an infinitival TP.
(9) a. ukudla ku-mnandi
AUG.15food 15S-nice
?Food is nice.?
b. uku-xova ujeqe ku-mnandi
AUG.15-make AUG.1steamed.bread 15S-nice
?Making steamed bread is nice.? (Halpert 2019:139)
44
As for object agreement, consider the sentences below. Note that the ya- morpheme has
been argued to indicate that the vP is empty (Buell, 2005; Halpert, 2015). In these cases,
the vP-external arguments can control object agreement: the DP argument in (10a) and
the infinitival TP in (10b) both trigger ku-.
(10) a. ngi-ya-ku-funa]vP ukudla
1SG.S-YA-15/17O-want AUG.15/17food
?I want food.?
b. ngi-ya-ku-funa]vP [uku-xova ujeqe ]
1SG.S-YA-15/17O-want AUG.15/17-make AUG.1steamed.bread
?I want to make steamed bread.? (Halpert 2019:139)
Looking now at the behaviour of embedded CPs, there is evidence that they bear ?-
features as well. First, we see that CP complements may optionally control object agreement?
specifically, class 15/17 agreement morphology. In (11a) below, the ya- morpheme indi-
cates an empty vP, as above in (10). Since object agreement is not possible with vP-
internal elements, the appearance of ku- alongside ya- suggests that the embedded CP is
controlling object agreement on the matrix verb. The sentence in (11b) demonstrates that
this is indeed the case since, without ya-, ku- cannot appear.
(11) a. ngi-ya-ku-cabanga]vP [ukuthi uMlu u-ya-bhukuda manje]
1SG.S-YA-17O-think that AUG.1Mlu 1S-YA-swim now
?I think that Mlu is swimming now.?
b. *ngi-ku-cabanga [ukuthi uMlu u-ya-bhukuda manje]]vP
1SG.S-17O-think that AUG.1Mlu 1S-YA-swim now
?I think that Mlu is swimming now.? (Halpert 2019:140-141)
45
However, although CPs can control object agreement morphology, they cannot control
subject agreement in Spec,TP (unlike infinitival TPs). As shown below, preverbal CPs
cannot trigger ku- subject agreement. Note that, even though subject-agreement and
object-agreement sometimes take the same shape (in some noun classes, such as 15/17),
they can nevertheless be distinguished by their position within the verbal complex.
(12) *[ukuthi w-a-thatha umhlala phansi] ku-ya-ngi-mangaza
that 1S-PST-take AUG.1sit down 17S-YA-1SG.O-surprise
Intended: ?That he retired surprises me.? (Halpert 2019:141)
In summary, both CPs and infinitival TPs bear class 15/17 ?-features in Zulu, given that
they can control object agreement morphology. However, only infinitival TPs can satisfy
the EPP requirement since they?and not CPs?can appear in Spec,TP of a superordinate
clause, and trigger subject agreement. These properties, coupled with the implementation
of Cyclic Agree (Be?jar and Rezac, 2009; Clem, 2018; Keine and Dash, 2018) and feature
interaction and satisfaction (Deal, 2015), derive the Zulu hyper-raising patterns below, as
I will now discuss.1 Note that there is an alternative agreement pattern for the sentence
with hyper-raising in (13b), where agreement morphology for either the raised subject
(noun class 1) and the embedded CP (noun class 17) is possible.
(13) a. ku-bonakala [ukuthi uZinhle u-zo-xova ujeqe]
17S-seem that AUG.1Zinhle 1S-FUT-make AUG.1steam.bread
?It seems that Zinhle will make steamed bread.?
b. uZinhlei u-/ku-bonakala [ukuthi ti u-zo-xova ujeqe]
AUG.1Zinhlei 1S/17S-seem that ti 1S-FUT-make AUG.1steam.bread
1I discuss the motivation and implementation of both Cyclic Agree and feature interaction/satisfaction
in chapter 5.
46
?It seems that Zinhle will make steamed bread.?
c. *uZinhlei u-bonakala [ti uku-(zo-)xova ujeqe]
AUG.1Zinhlei 1S-seem ti INF-(FUT-)make AUG.1steamed.bread
Intended: ?It seems that Zinhle will make steamed bread.? (Halpert
2019:124)
Halpert?s analysis of (13a) assumes that the embedded subject has remained inside the
CP. As an expletive construction, a null expletive is inserted in the matrix Spec,TP which
triggers ku- agreement morphology. The null expletive fulfills the EPP requirement, so
no further operations are necessary.
The patterns in (13b), where the embedded subject has raised out of the CP, is
derived as a second cycle Agree that is initiated following intervention by the CP. We
saw above that CPs bear ?-features; they are thus targeted first by the ?-probe on T. This
results in an A-over-A configuration (Chomsky, 1964) such that the CP intervenes and
blocks agreement with potential goals lower in the structure, as in (14).
47
(14) First cycle Agree and the EPP
...
T0
ku-bonakala CP
ukuthi TP
uZinhle
uzoxova ujeqe
Following Deal (2015), Halpert assumes that the ?-probe on T interacts with the
CP, i.e. the ?-features on CP are evaluated for agreement. Halpert also assumes that CPs
cannot satisfy the EPP in Zulu (since they cannot occupy Spec,TP), and thus the first cycle
of Agree is not satisfied. A second cycle of Agree is initiated, which is now free to ignore
the CP (i.e. the highest A), as in (15). The embedded DP subject is targeted and agreed
with, and the embedded subject moves to the matrix Spec,TP position to satisfy the EPP.
Halpert takes the optionality in exponing u- vs. ku- agreement morphology to indicate
that two cycles of Agree have indeed been initiated. Since two sets of ?-features have
been evaluated for agreement (one by the embedded CP and the other by the embedded
DP subject), both of those sets of ?-features are on T. Either set may surface, i.e. be
?exponed? via the morphosyntactic mechanisms that translate the ?-features on syntactic
heads to actual morphemes.
48
(15) Second cycle Agree and the EPP
...
uZinhle
T0
u-/ku-bonakala CP
ukuthi TP
ti
uzoxova ujeqe
Finally, Halpert derives the lack of raising out of infinitival TPs, as in (13c), as an
instance of first cycle Agree, i.e. there is no second Agree search to satisfy the EPP. As
previously discussed, we saw that infinitivals bear ?-features as well?but, unlike CPs,
TPs can also satisfy the EPP in Zulu. Thus, when they are targeted by matrix T, they serve
as satisfactory goals for the ?-probe on T and move to Spec,TP to satisfy the EPP. Under
this analysis, there is no opportunity to raise embedded DP subjects since the ?-probe will
never target them. Further, it is possible for TPs to appear preverbally in the canonical
subject position, as shown in (16).
(16) a. [uku- (zo)- fika k- ubusika] ku-ya-bonakala
AUG.15- (FUT)- arrive 15ASSOC- AUG.14winter 15S-YA-seem
?Winter?s arrival is evident.?/?We can tell that winter is coming.?
49
b. ?[uku- xova ujeqe kukaZinhle] ku-ya-bonakala
AUG.15- make AUG.1bread 15ASSOC.1Zinhle 15S-YA-seem
?It?s evident that Zinhle is making steamed bread.?
In summary, this section discussed the mechanisms deriving Tagalog wh-movement and
Zulu hyper-raising as A-over-A effects, rather than PIC-driven inaccessibility issues. In
Tagalog, the phase is the closer goal and thus is evaluated first by syntactic operations. If
elements inside the phase are to be extracted, the phase must first be rendered transparent.
This requirement is met under Agree in Tagalog, which is overtly marked as agreement
morphology on the embedding verb. Note that these relations follow from general princi-
ples of locality and agreement; the PIC itself is not a component of the analysis.
As for Zulu, the empirical data were derived under a model where both infinitival
TPs and CPs are ?-goals in Zulu, but only infinitival TPs can satisfy the EPP. Thus,
raising out of infinitives is blocked since the infinitive itself moves to Spec,TP. When
embedded CPs are targeted by matrix T, a second cycle of Agree is initiated since CPs
do not satisfy the EPP. Since the CP layer has interacted with the probe on T, it no longer
behaves as an intervener?thus, the second search allows T to ignore the clause boundary
and target the embedded subject for ?-agreement and movement. In either analysis, the
theoretical notion of the PIC was argued to be an epiphenomenon that derived from other,
independent properties of the syntax.
3.2 Alleviating the Phase Impenetrability Condition
In this section, I will discuss a family of approaches that, in contrast to those dis-
cussed in section 3.1, assumes the PIC is a syntactic primitive. As in the models discussed
50
in chapter 2, phases are assumed to be pre-determined chunks of structure that the syntax
builds. Accordingly, these analyses further assume that the Phase Impenetrability Con-
dition independently governs the inaccessibility of the interior of those phasal structures.
Under these views, the PIC is relaxed just in case the phase head is targeted for Agree. If
that operation is successful, then the interior of the phase can be accessed for subsequent
operations. I will first discuss this model as introduced in Preminger (2011b), followed
by an overview of its extension in van Urk and Richards (2015).
Turning first to Preminger (2011b), we see that, cross-linguistically, person and
number agreement stand in an asymmetrical relationship: In languages that allow long-
distance agreement, person agreement is more fragile than number agreement. This gen-
eralization is formulated below.
(17) Relative Aptitude for Failed Agreement (RAFA)
person at-a-distance >> number at-a-distance (>> any agreement at close range)
Preminger (2011b)
In other words, if number agreement is disrupted in a given language X, then person
agreement is disrupted there as well. Notably, this implicational relationship does not
hold the other way around: disrupting person agreement does not entail that number
agreement is disrupted as well. Preminger proposes that this behaviour follow from a
model where (i) ?-agreement is split into two distinct probes that are situated on their
respective syntactic heads (a person probe on ?0, and a number probe on #0); and (ii) #0
immediately c-commands ?0, as illustrated below.
51
(18) [ ... [#P #0 [ 0?P ? [ ... DP ... ] ] ] ]
(Preminger, 2011b)
This analysis has two advantages. First, any intervener for #0 will also be an intervener
for ?0, whether that intervention arises from the presence of a separate, intervening DP,
or from the presence of a phasal boundary.
(19) a. [ ... [#P #0 [ 0?P ? [ ... INTERVENER DP ... ] ] ] ]
b. [ ... [ 0 0#P # [?P ? [PHASE DP ... ] ] ] ]
(Preminger, 2011b)
Second, the analysis allows for cases where only person agreement is disrupted (and num-
ber agreement is successful). As shown below, a clitic-doubled nominal leaves behind an
A-trace (Anagnostopoulou, 2006) which is invisible for the purposes of Agree. Since
neither the clitic nor the trace is an intervener for #0, the number probe can access the DP
target.
(20) [ ... [#P #0 [ 0?P CLi?? [ ... INTERVENERi>DP ... ] ] ] ]
(Preminger, 2011b)
Preminger assumes that the same general mechanism is applicable to the Tagalog wh-
movement restrictions discussed in Rackowski and Richards (2005), where wh-movement
from an embedded clause is possible just in case the embedded clause has been agreed
52
with. In this scenario, Agree between the probe on ?0 and the phase alleviates the PIC.
The probe on #0 can thus access the interior of the structure and target the embedded DP.
(21) [ ... [ #0#P [?P ?0 [PHASE DP ... ] ] ] ]
(Preminger, 2011b)
As the main focus of Preminger (2011b) is the asymmetrical relationship between person
and number agreement, especially in relation to Baker?s Structural Condition on Person
Agreement (SCOPA) (Baker, 2011), the unphasing aspect of the analysis was not further
developed there. However, van Urk and Richards (2015) assume this property to be a
crucial component of their analysis of Dinka (Nilotic) extraction patterns. I will now turn
to an overview of their analysis for the remainder of this section.
van Urk and Richards (2015) show that, in Dinka, extraction from embedded CPs
must be preceded by movement of the CP to Spec,vP, which is itself parasitic on Agree.
As shown below in (22), both Spec,CP and Spec,vP must be occupied.
(22) a. Bo?l a?-ce? DE?N lE?k a?ke?ko?ol
Bol 3SG?-PRF Deng t?ell story??
?Bol told Deng a story.?
b. Bo?l a?-ce? a?ke?ko?ol lE?k DE?N
Bol 3SG?-PRF story?? t?ell Deng
?Bol told Deng a story.?
c. *Bo?l a?-ce? lE?k DE?N a?ke?ko?ol
Bol 3SG?-PRF t?ell Deng story??
Intended: ?Bol told Deng a story.?
d. *a?-ce? Bo?l lE?k DE?N a?ke?ko?ol
3SG?-PRF Bol t?ell Deng story??
53
Intended: ?Bol told Deng a story.?
(van Urk and Richards 2015:134)
However, if there is an embedded CP complement, as in (23), it appears that both Spec,CP
and Spec,vP can be empty.
(23) a. Bo?l a?-ce? DE?N lE?k [Aye?n a?-ce? k??ta?p GO?Oc]
Bol 3SG?-PRF Deng t?ell Ayen 3SG?-PRF book buy
?Bol told Deng that [Ayen bought a book].?
b. Bo?l a?-ce? lE?k DE?N [Aye?n a?-ce? k??ta?p GO?Oc]
Bol 3SG?-PRF t?ell Deng Ayen 3SG?-PRF book buy
?Bol told Deng that [Ayen bought a book].?
c. a?-ce? Bo?l lE?k DE?N [Aye?n a?-ce? k??ta?p GO?Oc]
3SG?-PRF Bol.GEN t?ell Deng Ayen 3SG?-PRF book buy
?Bol told Deng that [Ayen bought a book].?
(van Urk and Richards 2015:135)
To account for this pattern, van Urk and Richards propose that, in Dinka, CP complements
pattern as DPs: they can check Case in Spec,vP, and further move to the matrix Spec,CP.
Furthermore, CPs linearize to the right. Thus, the positions in Spec,CP and Spec,vP only
appear to be empty in (23b-c)?underlyingly, the embedded CP moves to a rightward
specifier(s) and appears sentence-finally. Some of the evidence for this movement is
provided in (24), below, where filling Spec,vP with another DP blocks movement of the
embedded CP to the matrix Spec,CP. Note that, even though the matrix Spec,CP appears
to be empty in this case, due to the right-hand linearization of CPs (see above), movement
of the embedded CP must pass through Spec,vP en route to the higher landing site.
54
(24) *a?-ce? Bo?l DE?N lE?k [Aye?n a?-ce? k??ta?p GO?Oc]
3SG?-PRF Bol.GEN Deng t?ell Ayen 3SG?-PRF book buy
Intended: ?Bol told Deng that [Ayen bought a book].?
(van Urk and Richards 2015:136)
Additionally, long-distance extraction from finite CPs requires movement through the
intermediate Spec,vP, as illustrated in (25).
(25) a. YeNa? c??i Ya?ar lE?k DE?N [ye? c??i Bo?l tuO?Oc
who P?RF.NS Ya?a?r.GEN t?ell Deng C P?RF.NS Bol.GEN send
wu?ut]?
ca?tt?le.camp.LOC
?Who did Yaar tell Deng that Bol sent to the cattle camp??
b. *YeNa? c??i Ya?ar DE?N lE?k [ye? c??i Bo?l tuO?Oc
who P?RF.NS Ya?a?r.GEN Deng t?ell C P?RF.NS Bol.GEN send
wu?ut]?
ca?tt?le.camp.LOC
Intended: ?Who did Yaar tell Deng that Bol sent to the cattle camp??
(van Urk and Richards 2015:137?138)
van Urk and Richards tie this phenomenon to a precondition on movement?namely,
that movement is parasitic on Agree. Thus, if a wh-element is to be extracted from an em-
bedded CP, v must first Agree with that CP. They argue that the embedded CP projection,
as a phasal boundary, acts as an intervener for the wh-probe (building on Rackowski and
Richards, 2005). Since the embedded CP has a wh-element in its specifier, the CP itself
must have a wh-feature as well (since it triggers wh-movement within its domain). Thus,
under the A-over-A Principle (Chomsky, 1964), the CP is the closer goal to the probe on
the matrix v than the wh-phrase, and acts as an intervener.
55
In order to derive embedded wh-extraction from the CP, van Urk and Richards
adopt a version of the Principle of Minimal Compliance (Rackowski and Richards, 2005;
Richards, 1998), which was formulated to account for the ability of the second, third,
etc. wh-phrase in multiple-wh questions to violate island conditions. That is, in some
languages with multiple wh-movement (e.g., Bulgarian and Romanian), the highest wh-
phrase must move first, and must do so in a way that obeys locality restrictions, including
islands. Once that requirement is satisfied, subsequent wh-phrases may move out of is-
lands to the same CP layer. As characterized below, this property has the effect of allow-
ing subsequent operations to be exempted from syntactic conditions, provided the initial
operation fulfills those requirement(s).
(26) Principle of Minimal Compliance (PMC)
Once a probe P Agrees with a goal G, P can ignore G for the rest of the derivation.
(van Urk and Richards, 2015:142, citing Rackowski and Richards, 2005)
In Dinka, the PMC is met through the initial Agree relation between the matrix v and the
embedded CP, an operation that then triggers movement of that CP to Spec,vP. Having
satisfied the PMC, v is free to ignore the CP node during a subsequent search, leaving v
free to access the wh-phrase in the embedded Spec,CP. The wh-phrase then lands in the
matrix CP.
van Urk and Richards (2015) note that similar restrictions are found in a diverse
set of languages, e.g. Chamorro (Chung, 1998; den Dikken, 2009a) and Hungarian (den
Dikken, 2009b, 2012). It has been argued that, in these languages, both A?- and A-
56
movement is mediated by Agree between the embedding v and the embedded CP?
namely, movement out of the phase is preconditioned by Agree with the phase head.
Broadly, this family of proposals argues that Agree between a head and a phase alleviates
the PIC, which renders the interior of the phase accessible to further operations.
However, the existing literature on mediating phasehood with Agree has largely
focused on extraction constraints from embedded CPs, rather on the ability to probe into
the interior of the phase. What I will propose in the next chapter is that Georgian number
agreement is another manifestation of this pattern. Namely, a probe may Agree with a vP
?phase?, and, once this relation is established, a subsequent probe can search further into
the vP and target previously-inaccessible arguments. Thus, taken together, there are at
least two possible outcomes of undoing phasehood via Agree: (i) Agreeing with a phase
head can allow A- and A?-movement out of that phase, and (ii) Agreeing with a phase
head can allow subsequent probes search inside the phase. In chapter 5, I outline the core
aspects of this view with respect to the Georgian agreement system, showing how this
pattern manifests within the agreement domain.
3.3 Summary and desiderata
This section summarizes the advantages that phase unlocking has provided for our
theory of phases, and outlines the theoretical issues and empirical predications that arise
under the two sub-approaches to phase unlocking.
57
3.3.1 Summary
To begin, the broad family of phase unlocking analyses avoids many of the issues
inherent to formalizing phasehood. As discussed in chapter 2, phases were assumed to
be pre-determined chunks of syntactic structure that were Transferred to the interfaces
independently of the overall structure. While this view lent itself to computational effi-
ciency, it also gave rise to several issues in terms of formalization and definition. Namely,
?phasehood? was tenuously defined in terms of convergence, theta role assignment, and
propositional content, none of which held up to scrutiny. The standard phasal categories
were assumed a priori, and cross-linguistic investigation suggested that phasehood was
not limited to just CP, vP, or DP (for an overview, see Citko, 2014). In addition, the Phase
Impenetrability Condition was shown to have its set of problems (even after several re-
visions and reformulations), and its theoretical necessity has been called into question
given that much of its empirical motivation can be attributed to independently necessary
syntactic properties?most notably, minimality.
Under the phase unlocking models discussed in this section, the PIC is essentially an
intervention effect. As the closer goal for a syntactic operation, the so-called phase head
blocks syntactic operations from accessing material lower in the structure; this property
derives the Phase Impenetrability Condition as a general locality issue. Both intervention
effects and minimality effects are motivated independently of phase theory, so subsuming
phasehood under these phenomena results in a more parsimonious theory. We can also
circumvent the issues surrounding the formal definition of phases and possible phasal cat-
egories by assuming the ?phase? head bears a feature relevant to the syntactic operation in
58
question. As noted in the previous section, there are two distinct models of phase unlock-
ing: (i) the PIC is purely an A-over-A effect, and (ii) the PIC is a syntactic primitive, but
one whose effects can be alleviated. While both approaches take phases to be instances of
intervention, and derive the PIC without referring to Transfer, they differ in how they de-
rive crosslinguistic variation in phasehood. The remainder of this section further outlines
these properties, particularly with respect to accessibility and variation.
3.3.2 Desiderata
As just noted, the two approaches to phase unlocking highlight (at least) two the-
oretical issues that are stake: accessibility of the interior of the phase, and variation
in phasal categories across languages. While phase domains are accessible under both
the PIC-as-intervention and alleviating-the-PIC models?and thus accessibility cannot
be used to adjudicate between the two models?they have different explanations for the
crosslinguistic variation of phases. These are summarized in the table below, and I will
argue in favour of the PIC-as-intervention approach in the remainder of this thesis.
(27) Two models of phase unlocking
PIC-AS-INTERVENTION ALLEVIATING-THE-PIC
IS THE INTERIOR OF Yes Yes
THE PHASE ACCESSIBLE?
HOW IS PHASEHOOD Phasehood correlates Stipulation
DERIVED? with ?-intervention
Turning first to the accessibility issue, we saw that accessing the interior of the
?phase? is possible under both models, and they raise similar issues with respect to Trans-
59
fer and the PIC. Under the alleviating-the-PIC model, Transfer is fundamentally incom-
patible with ?unlocking??and with PMC effects more generally?given that phases would
have to be ?unshipped? from the interfaces once the PMC is met. At first glance, this
may seem desirable since Transfer as an independent mechanism is ruled out under the
alleviating-the-PIC model. However, models that alleviate the PIC cede explanatory
ground since they can no longer derive the PIC from properties of the syntax. In con-
trast, under the PIC-is-intervention model, there is no theoretical requirement for an in-
dependent constraint like the PIC since it can be reduced to minimality. As such, distinct
phase-driven Transfer cycles are unnecessary since locality and intervention derive the
inaccessibility effect previously attributed to Transfer.
In addressing the accessibility issue, I examine cases of phase unlocking in chap-
ter 5 where Agree appears to target arguments that are base-generated below a phasal
boundary, and that crucially stay below that boundary. Specifically, my analysis of the
Georgian agreement system shows that number agreement targets a vP-internal argument
just in case person agreement first targets v itself. My proposal thus necessitates a view
where ?phases? do not undergo Transfer to the interfaces, though this result alone does not
adjudicate between the two models of phase unlocking. In order to distinguish between
the two models, we need to examine cases where they differ. I will focus on this point in
the remainder of this section.
Turning now to explaining crosslinguistic variation in phasal categories, we see that
the two approaches differ with respect to the identity of phases, and whether phases are
cross-linguistically uniform. Under both approaches to phase unlocking, the ?phase? is
merely the closest goal for the operation at hand (e.g., Agree). While the alleviating-
60
the-PIC approach faces the same questions concerning the definition of phases that we
noted with respect to standard phase theory, the PIC-is-intervention approach provides
a clear pathway to track why the ?phase? intervenes. Namely, what look like phasehood
effects arise due to the presence of syntactic features?and, ideally, we would be able to
independently diagnose the presence of these features, whether syntactically or morpho-
logically. In my analysis of Georgian agreement in chapter 5, I show that the vP ?phase?
is a case of ?-intervention, where the ?-probe on v is an intervener with respect to person
agreement. My proposal thus suggests that ?phasehood? across languages correlates with
phi-agreement, i.e. for any given language, vP behaves as a ?phase? just in case v bears
a ?-probe. Thus, languages without a vP ?phase? lack a ?-probe on v, e.g. Hindi-Urdu
(see Keine, 2017; further discussion in chapter 6). Additionally, languages with a non-
canonical ?phase??such as TP?have a ?-probe on that syntactic head, e.g. potentially
Spanish and other null subject languages (see Gallego, 2006; 2010). With these views in
mind, I now turn to an overview of Georgian clausal structure which lays the basis and
foundations for the comprehensive analysis of its agreement system.
61
Chapter 4: Basic properties of Georgian
Georgian, also known as Kartvelian, is a South Caucasian language with approx-
imately four million speakers. It is the official language of the country of Georgia, and
the data provided in this thesis come from my own fieldwork with Georgian-speaking
consultants (unless otherwise cited). This chapter provides an overview of the Georgian
agreement system, as well as its clausal architecture, case alignment, and pronominal
licensing requirements (for more description, see Harris, 1981; Hewitt, 1995, Makharob-
lidze, 2012). (For discussion on Georgian phonology and phonotactics, see Butskhrikidze
(2002) and Begus? (2020), and references therein.) As will be discussed in section 4.1,
Georgian has two major agreement paradigms (the basic and inverse). Section 4.2 shows
that subjects in the two paradigms occupy different structural positions: subjects of verbs
in the basic agreement paradigm are external arguments, whereas subjects of verbs in
the inverse agreement paradigm are experiencer arguments. Section 4.2 also shows that
Georgian has both head-initial and head-final projections. The remainder of this chap-
ter describes the splits in Georgian case-marking and pronominal licensing: section 4.3
describes the Georgian case alignment patterns, and section 4.4 describes the Georgian
person-based licensing requirements for 1st and 2nd person arguments. The descriptions
provided here motivate the structures and mechanisms adopted for the remainder of this
62
dissertation, which are fundamental for the analysis of the Georgian vP ?phase? as purely
?-intervention (which will be discussed in Chapter 5).
4.1 Description of the Georgian agreement system
This section describes the Georgian agreement patterns as they appear throughout
two broad categories of tense: the past, which is generally referred to as the aorist in the
traditional Georgian philological literature, and the present. Before delving into the de-
tails of the Georgian agreement paradigms, note that this dissertation will focus primarily
on verb forms that appear in the aorist since it has overt markers for tense/aspect/mood
(TAM).
Further, the discussion of inverse agreement in section 4.1.3 is limited to verbs that
idiosyncratically take dative subjects. The class of such verbs in Georgian include stems
such as ?nd?, ?to want?, ?q?var? ?to love?, ?xsov? ?to remember?, ?g?viz?? ?to be awake?, and
?z?ul? ?to hate??namely, typical psych verbs. Inverse verbs require additional prefixes that
co-vary with the subject. These prefixes are traditionally referred to as versionizers, and,
outside of the inverse agreement paradigm, they indicate the status of an applied argu-
ment: ?a-? can mark causatives and locatives, ?i-? can have either a reflexive or possessive
meaning, ?u-? appears with benefactive applicative arguments, and ?e-? appears with psy-
chological predicates (Lomashvili, 2011). While the versionizers are productive in the
basic agreement paradigm (i.e. they mark a relevant applied argument), they behave akin
to agreement in the inverse agreement paradigm. That is, inverse subjects (which obli-
63
gatorily trigger versionizers) are not subject to the various applicative interpretations that
basic subjects receive with the same versionizers.
The term ?inverse/inversion? may also refer to a broader set of agreement patterns
found throughout Georgian, e.g. those involving the perfect as well as the marking of
evidentiality (Harris, 1981). While the agreement marking is the same for verbs in the
perfect/evidential and those with quirky dative subjects, the former require different the-
matic suffixes unrelated to the agreement markers discussed here. Given the additional
complexity of any analysis of Georgian agreement that would address the perfect and
the marking of evidentiality, I leave these additional uses of inverse agreement aside for
further research. I focus here only on verbs whose subjects must be marked dative even
when those verbs are not marked for perfect or evidentiality.
4.1.1 General properties
The Georgian verbal complex features many person- and TAM-markers which ap-
pear as both prefixes and suffixes, many of which are also fusional or zero exponents.
The two major agreement paradigms?basic and inverse?are found throughout eleven
screeves, i.e. verbal paradigms that feature specific fusional TAM morphology such as
the imperfect, future, conditional, and others (see below). The term derives from Geor-
gian ?mts?k?rivi? ?row?, which traditionally appears in Georgian grammars (see, e.g.,
Chikobava, 1950; Kavtaradze, 1954; Shanidze, 1973). TAM morphology typically com-
prises (at least) two types of markers, such as various preverbs (PRV) that signal aspect,
and thematic suffixes (TS) that signal tense and mood.
64
The screeves are further grouped into three series, based on the case-marking of
the arguments involved. As shown below, the three series are (i) the present, (ii), the
aorist, and (iii) the perfect (Harris, 1981:46; Aronson, 1990:41; Melikishvili, 1998, 2001).
Note that the case-marking of the subject dictates the series of the verb: Georgian is
typically described as having seven distinct cases (nominative, ergative, dative, genitive,
instrumental, adverbial, and vocative), but only nominative, ergative, and dative are used
to mark subjects.
(1) Georgian conjugation classes
SUBJECT CASE SERIES SCREEVES
Nominative Ia Present Imperfect Conjunctive present
Future Conditional Conjunctive future
Ergative IIa Aorist Conjunctive past
Dative IIIb Present perfect Pluperfect Conjunctive perfect
(evidential)
a Basic paradigm
b Inverse paradigm
Turning now to agreement, we see below that predicate-argument agreement in
Georgian two-place sentences is marked by a set of prefixes and and a set of suffixes, as
well as plural agreement via ?-t?.
(2) man tkven da-g-p?at?iz?-a-t
3SG.ERG 2PL.NOM PRV-2-invite-3SG.AOR-PL
?She/he invited you (pl).?
Broadly speaking, the prefixes mark the features of one argument, whereas the suffixes
mark the features of the other. The prefixes are tense/aspect-invariant, and, as will be
65
shown in the next sections, there are some cases where they appear to exhibit person hier-
archy effects. I remain agnostic as to their potential status as doubled clitics or agreement
suffixes, and the analysis developed in chapter 5 does not rely on this distinction. In con-
trast, the forms of the suffixes (except ?-t?) show sensitivity to tense/aspect and do not
exhibit person hierarchy effects.
The agreement patterns that I will overview in the following sections appear through-
out different series. The basic agreement paradigm is followed by most verbs in Series I
and II, whereas the inverse agreement paradigm is characterized by Series III conjugation
and verbs that idiosyncratically take dative subjects (see Harris, 1981). Notably, the argu-
ments that the prefixes and suffixes track are, for the most part, flipped between the two
paradigms (hence the name ?inverse?). In the basic agreement paradigm, the prefixes co-
vary with the object while the suffixes co-vary with the subject. In the inverse paradigm,
the prefixes co-vary with the subject in the inverse agreement pattern while the suffixes
co-vary with the object. I discuss the details of these paradigms below.
4.1.2 The basic agreement paradigm
Consider the data below. In each case, the form of the prefix co-varies with the per-
son features of the object: ?g-? marks the 2nd person object in (3a) regardless of number,
?m-? marks the 1st person singular object in (3b), and ?gv-? marks the 1st person plural
object in (3c).
(3) a. me s?en da-g-p?at?iz?-e
1SG.ERG 2SG.NOM PRV-2-invite-1/2.AOR
?I invited you (sg).?
66
b. s?en me da-m-p?at?iz?-e
2SG.ERG 1SG.NOM PRV-1-invite-1/2.AOR
?You (sg) invited me.?
c. s?en c?ven da-gv-p?at?iz?-e
2SG.ERG 1PL.NOM PRV-1PL-invite-1/2.AOR
?You (sg) invited us.?
3rd person objects are generally unmarked, as seen below in (4a-b), except when the
subject is 1st person?in these cases, the prefix ?v-? appears, as in (4c).1
(4) a. s?en is da-p?at?iz?-e
2SG.ERG 3SG.NOM PRV-invite-1/2.AOR
?You (sg) invited him/her.?
b. man is da-p?at?iz?-a
3SG.ERG 3SG.NOM PRV-invite-3SG.AOR
?She/he invited him/her.?
c. me is da-v-p?at?iz?-e
1SG.ERG 3SG.NOM PRV-1-invite-1/2.AOR
?I invited him/her.?
A closer look at the vocalic suffixes in (3-5) shows a difference in their distribution with
respect to person. The suffix ?-e? appears with participant subjects (i.e., 1st and 2nd
person) in the aorist, as in (4c) and (5a). In contrast, (5b) shows that ?-a? appears with
3rd person singular subjects (also in the aorist). The number of the participant subject
1This is not the only way to describe the distribution of this prefix. As characterized here, I am presup-
posing that ?v-? is a a morphologically-conditioned allomorph of a 3rd person object prefix whose appear-
ance is triggered by the 1st person subject. Its distribution could also be described as a 1st person subject
prefix that fails to surface when the object is 2nd person, possibly due to morphological competition. At
this point in the thesis, there is no a priori reason to prefer one characterization over the other. I describe
?v-? as a form specific to 1>3 contexts only for the reason that the distribution of ?g-? is more stable, i.e.
it appears in all cases with a 2nd person object, and the other prefixes seem to be object markers as well.
Ultimately, the characterization pursued in the main text turns out to facilitate a formal distinction between
?v-? and the other prefixes as second vs. first cycle agreement. I will further motivate this characterization
in chapter 5, based on previous work by Be?jar (2003); Be?jar and Rezac (2009) and others.
67
does not affect the form of the suffix?number is marked by a separate agreement marker
?-t??but 3rd person plural subjects do trigger another form altogether, as in (5c).
(5) a. s?en me da-m-p?at?iz?-e
2SG.ERG 1SG.NOM PRV-1-invite-1/2.AOR
?You invited me.?
b. man c?ven da-gv-p?at?iz?-a
3SG.ERG 1PL.NOM PRV-1PL-invite-3SG.AOR
?She/he invited us.?
c. mat c?ven da-gv-p?at?iz?-es
3PL.ERG 1PL.NOM PRV-1PL-invite-3PL.AOR
?They invited us.?
A similar pattern holds for the verb forms used to express the present and future. As
illustrated in (6a), there is no overt suffix for participant subjects, but ?-s? marks 3rd
person singular subjects, as in (6b), whereas ?-en? marks 3rd person plural subjects, as in
(6c).
(6) a. s?en c?ven gv-p?at?iz?-eb-0/
2SG.NOM 1PL.DAT 1PL-invite-TS
?You invite us.?
b. is c?ven gv-p?at?iz?-eb-s
3SG.NOM 1PL.DAT 1PL-invite-TS-3SG.PRES
?She/he invites us.?
c. isini c?ven gv-p?at?iz?-eb-en
3PL.NOM 1PL.DAT 1PL-invite-TS-3PL.PRES
?They invite us.?
68
Now consider the distribution of the plural marker ?-t?. This suffix appears after the
aforementioned person-sensitive suffixes and can mark a 2nd person plural argument in
either the subject position, as in (7a), or the object position, as in (7b).
(7) a. tkven is da-p?at?iz?-e-t
2PL.ERG 3SG.NOM PRV-invite-1/2.AOR-PL
?You (pl) invited him/her.?
b. man tkven da-g-p?at?iz?-a-t
3SG.ERG 2PL.NOM PRV-2-invite-3SG.AOR-PL
?She/he invited you (pl).?
There are two puzzles associated with the verbal plural marker ?-t?, one of which concerns
asymmetrical plural marking, and a second concerning the blocking of ?-t?. First, ?-t? can
appear when a 1st person plural argument is a subject, as in (8a), but not when it is an
object, as in (8b). This is surprising given that, as shown in (7a-b), the appearance of
?-t? can be triggered by the plurality of a 2nd person argument regardless of whether that
argument is a subject or object. That is, we would expect 1st person plural arguments
to trigger the same kind of omnivorous number agreement (Nevins, 2011) as 2nd person
plural arguments do.
(8) a. c?ven is da-v-p?at?iz?-e-t
1PL.ERG 3SG.NOM PRV-1-invite-1/2.AOR-PL
?We invited him/her.?
b. man c?ven da-gv-p?at?iz?-a-(*t )
3SG.ERG 1PL.NOM PRV-1PL-invite-3SG.AOR
?She/he invited us.?
69
Second, 3rd person plural subjects block the appearance of ?-t? even if the object is 2nd
person plural. As we saw in (7b), 2nd person plural objects are marked with ?-t? when
the subject is 3rd person singular, which suggests that the number feature of the subject
plays some role in this particular blocking effect.
(9) mat tkven da-g-p?at?iz?-es(-*t)
3PL.ERG 2PL.NOM PRV-2-invite-3PL.AOR-PL
?They invited you (pl).?
More generally, ?-t? cannot appear with 3rd person plural subjects in the basic agreement
paradigm (Series I and II), as illustrated below in (10). This is not the case for the inverse
agreement paradigm ? (11a) shows that ?-t? can mark 3rd person plural subjects of inverse
verbs, a pattern that will be further discussed in the following section. Note that the
configuration in (11a) is the only configuration where 3rd person plural arguments can be
marked with ?-t?, i.e. when the 3rd person plural argument is the subject, and the object
is also 3rd person. If these roles are reversed, as in (11b), ?-t? can no longer mark the 3rd
person plural argument.
(10) isini mas p?at?iz?-eb-en(-*t)
3PL.NOM 3SG.DAT invite-TS-3PL.PRES(-PL)
?They invite him/her.?
(11) a. mat is u-q?var-t
3PL.DAT 3SG.NOM VER-love-PL
?They love him/her.?
b. mas isini u-q?var-da(-*t)
3SG.DAT 3PL.NOM VER-love-TS-PL
?She/he loves them.?
70
The tables in(12?13) summarize the agreement patterns described in this section. Three
of the four overt prefixes track the features of the object: ?m-? marks 1st person singular
objects, ?gv-? marks 1st person plural objects, and ?g-? marks 2nd person objects regard-
less of number. The distribution of ?v-?, the fourth overt prefix, can be described in two
ways: either it is a 1st person subject prefix that is overridden by a 2nd person object
marker, or it is an allomorph of a 3rd person object prefix that is conditioned by a 1st
person subject (see footnote 2).
(12) Series I ? Basic agreement patterns2
@
@ O
@ 1SG 1PL 2SG 2PL 3SG 3PL
@
S @
@
1SG ? ? g- -0/ g- -0/ -t v- -0/ v- -0/
1PL ? ? g- -0/ -t g- -0/ -t v- -0/ -t v- -0/ -t
2SG m- -0/ gv- -0/ ? ? 0/ - -0/ 0/ - -0/
2PL m- -0/ -t gv- -0/ -t ? ? -0/--t -0/ --t
3SG m- -s gv- -s g- -s g- -t -s -s
3PL m- -en gv- -en g- -en g- -en -en -en
These tables show that the forms of the prefixes are identical across Series I and II,
in contrast to the TAM-conditioned variation in the forms in the suffixes. In Series I (used
to express the present and future), ?-en? marks 3rd person plural subjects and ?-s? marks
3rd person singular subjects; there are no overt suffixes for participant subjects in Series
I. In Series II (used to express the past), ?-es? marks 3rd person plural subjects and ?-a?
2Consultants cannot provide translations for the following argument combinations in a way that cor-
responds to the typical transitive construction: 1SG>1PL, 1PL>1SG, 2SG>2PL, or 2PL>2SG. Rather,
sentences with these combinations require the the pronominal direct object to appear as a reflexive such
as, for example, tSemi tavi ?my head?. Although the corresponding English translations are awkward, they
are possible given an appropriate context, e.g. ?I saw us on TV last night?. While this difference between
Georgian and English is an outstanding puzzle in its own right, I leave it aside for the current purposes of
this thesis (see Lasnik 1981 for more discussion on binding obviations).
71
marks 3rd person singular subjects; in addition, ?-e? marks participant subjects, regardless
of number.
(13) Series II ? Basic agreement patterns
@
@ O
@ 1SG 1PL 2SG 2PL 3SG 3PL
@
S @
@
1SG ? ? g- -e g- -e-t v- -e v- -e
1PL ? ? g- -e-t g- -e-t v- -e-t v- -e-t
2SG m- -e gv- -e ? ? 0/ - -e 0/ - -e
2PL m- -e-t gv- -e-t ? ? -e-t -e-t
3SG m- -a gv- -a g- -a g- -a-t -a -a
3PL m- -es gv- -es g- -es g- -es -es -es
Throughout this discussion, I assume that the markers in Series I and II are iso-
morphic up to the level of phonology. Note that there is a discrepancy in the cells corre-
sponding to a 3rd person singular subject acting on a 2nd person plural object: in Series
I, the ?-t? suffix does not overtly mark the 2nd person plural argument, but the suffix does
appear in Series II. I adopt a phonological account of this discrepancy, rather than a mor-
phosyntactic one: I assume [st] coda clusters reduce to [t] (see section 5.5.2 for further
discussion on this point). The view that this discrepancy in the distribution of ?-t? between
the two paradigms may be conditioned by phonology may come as a surprise, given that
Georgian is known for its complex onset clusters. However, Georgian is conservative in
its coda clusters, which are surprisingly rigid (Begus?, 2020; Butskhrikidze, 2002).
For the remaining cells in both series, the plural marker ?-t? consistently appears
with 1st and 2nd person plural subjects; ?-t? does not mark 1st person plural objects. Note
that there is some ambiguity here in 2PL>1PL contexts?it is unclear which argument
72
?-t? is marking, if it is indeed marking only one argument. In order to maintain a more
uniform characterization concerning 1st person plural objects, I will assume that the 2nd
person plural subject is responsible for the appearance of ?-t? in this case. When it comes
to object marking, ?-t? can appear with 2nd person plural objects, except when the subject
is 3rd person plural. In both series, 3rd person plural arguments cannot trigger ?-t? in
either subject or object position.
4.1.3 The inverse agreement paradigm
As briefly mentioned in section 4.1.1, the arguments that the prefixes and suffixes
mark in the inverse agreement paradigms are, for the most part, flipped compared to the
basic agreement paradigm. Recall that, in Series I and II, ?m-?, ?gv-?, and ?g-? are prefixes
that respectively mark 1st person singular, 1st person plural, and 2nd person objects. In
Series III and inverse verbs, these same prefixes mark the subject?in addition, the ?i-?
versionizer vowel is obligatory with participant subjects in the inverse. These patterns are
shown below in (14).
(14) a. me is m-i-q?var-s
1SG.DAT 3SG.NOM 1-VER-love-3SG.PRES
?I love him/her.?
b. c?ven is gv-i-q?var-s
1PL.DAT 3SG.NOM 1PL-VER-love-3SG.PRES
?We love him/her.?
c. s?en is g-i-q?var-s
2SG.DAT 3SG.NOM 2-VER-love-3SG.PRES
?You (sg) love him/her.?
73
d. tkven is g-i-q?var-t
2PL.DAT 3SG.NOM 2-VER-love-PL
?You (pl) love him/her.?
Similarly, the distribution of ?v-? is also flipped in the inverse agreement paradigm?it
appears only when the subject is 3rd person and the object is 1st person, as shown below.
There are two points to note here: the alternative description of ?v-? as a competing expo-
nent losing out to the 2nd person marker ?g-? is consistent in this paradigm as well; and
the versionizer vowel ?u-? is obligatory with 3rd person subjects (cf. ?i-? with participant
subjects, above).
(15) a. mas me v-u-q?var-v-ar
3SG.DAT 1SG.NOM 1-VER-love-1-be.PRES
?She/he loves me.?
b. mat me v-u-q?var-v-ar
3PL.DAT 1SG.NOM 1-VER-love-1-be.PRES
?They love me.?
As seen in (15) above and shown again in (16) below, another type of agreement marking
appears in inverse clauses with participant subjects, one which did not occur in the corre-
sponding configurations in the basic: the 1st person object is marked as a ?v-? prefix on a
dummy verb ?ar? ?be? (Nash, 1994). This prefix marks 1st person objects, regardless of
number, and it appears to be sensitive to person features only. It is perhaps unsurprising
that the ?subject-oriented? suffixes described in the previous section are sensitive to the
object in the inverse agreement paradigm.
74
(16) a. s?en me g-i-q?var-v-ar
2SG.DAT 1SG.NOM 2-VER-love-1-be.PRES
?You (sg) love me.?
b. mas c?ven v-u-q?var-v-ar-t
3SG.DAT 1PL.NOM 1-VER-love-1-be.PRES-PL
?She/he loves us.?
Similarly, the 2nd person agreement marker ?x-? appears in inverse clauses with 2nd per-
son singular and plural objects, as shown below.
(17) a. me s?en m-i-q?var-x-ar
1SG.DAT 2SG.NOM 1SG-VER-love-2-be.PRES
?I love you (sg).?
b. mas tkven u-q?var-x-ar-t
3SG.DAT 2PL.NOM VER-love-2-be.PRES-PL
?She/he loves you (pl).?
To reiterate, 1st and 2nd person object agreement in the inverse manifests as prefixes on a
dummy verb, a pattern that does not appear in the basic agreement paradigm?except for
copular constructions, as shown below. I will argue in Chapter 5 that there is a principled
reason why the 1st and 2nd person inverse agreement markers in (16-17) and the copular
constructions in (18) are form-identical, and for why this kind of dummy-verb marking
arises in cases like(15?16) in the first place; both will be tied to licensing requirements of
1st and 2nd person arguments.
(18) a. pexs?is?veli v-ar
barefoot 1-be.PRES
?I am barefoot.?
75
b. pexs?is?veli x-ar
barefoot 2-be.PRES
?You (sg) are barefoot.? (Harris, 1981:106?107)
There is no fundamental difference concerning 3rd person agreement between the basic
and inverse agreement paradigms. 3rd person arguments are marked with the same agree-
ment morphemes that appear throughout the basic paradigm, modulo the differences in
number sensitivity and subject vs. object agreement. One point of difference relative to
the basic agreement paradigm, though, is that there is no apparent sensitivity to the num-
ber features of 3rd person objects in the inverse: the suffix ?-s? marks both, as in (19).
(19) a. me is m-i-q?var-s
1SG.DAT 3SG.NOM 1-VER-love-3SG.PRES
?I love him/her.?
b. me isini m-i-q?var-s
1SG.DAT 3PL.NOM 1-VER-love-3SG.PRES
?I love them.?
The data in (16-19) illustrates the split in how 1st/2nd person and 3rd person arguments
are marked in the inverse agreement paradigm: 1st and 2nd persons require exceptional,
complex agreement forms, whereas 3rd person does not. I will return to the role of these
elements throughout the following sections of the dissertation, as this difference will play
a significant role in deriving the agreement patterns both within the inverse paradigm, and
between the inverse and basic.
Before delving into the distribution of the plural marker ?-t?, it should be recapitu-
lated what the main differences between the basic and inverse agreement paradigms are.
Both paradigms have agreement prefixes and suffixes, but their behaviour (i.e. the argu-
76
ment they mark) flips between paradigms. In the basic, the prefixes track the object and
the suffixes track the subject; in the inverse, the prefixes track the subject and the suffixes
track the object. So, keeping the grammatical roles of the arguments constant, the agree-
ment patterns between the basic and inverse paradigms appear to be a mirror image.3
However, once we investigate the behaviour of the plural marker ?-t?, this mirror-image
analogy begins to break down.
Starting from the part of the distribution that is a perfect mirror image, the sentences
in (20) show that 1st person plural inverse objects?but not subjects?can be marked by
?-t?. This is contra the behaviour of ?-t? in the basic paradigm, where 1st person plural
arguments could be marked by -t as subjects, but not as objects. This is shown in (21),
repeated from (8) above.
(20) a. mas c?ven v-u-q?var-v-ar-t
3SG.DAT 1PL.NOM 1-VER-love-1-be.PRES-PL
?She/he loves us.?
b. c?ven is gv-i-q?var-s-(*t)
1PL.DAT 3SG.NOM 1PL-VER-love-3.PRES-PL
?We love him/her.?
(21) a. man c?ven da-gv-p?at?iz?-a-(*t )
3SG.ERG 1PL.NOM PRV-1PL-invite-3SG.AOR
?She/he invited us.?
b. c?ven is da-v-p?at?iz?-e-t
1PL.ERG 3SG.NOM PRV-1-invite-1/2.AOR-PL
?We invited him/her.?
3I use the term ?grammatical role? very broadly here, namely as a way to refer to subject and object
without further reference to thematic roles or structural position.
77
The mirror image analogy breaks down in the following two puzzles. Recall from the
basic paradigm that 3rd person plural subjects block number agreement with 2nd person
plural objects, as in (22). However, this blocking effect does not extend to the inverse
paradigm, as shown in (23).
(22) mat tkven da-g-p?at?iz?-es(-*t)
3PL.ERG 2PL.NOM PRV-2-invite-3PL.AOR-PL
?They invited you (pl).?
(23) a. tkven isini g-i-q?var-t
2PL.DAT 3PL.NOM 2-VER-love-PL
?You (pl) love them.?
b. mat tkven u-q?var-x-ar-t
3PL.DAT 2PL.NOM VER-love-2-be.PRES-PL
?They love you (pl).?
Finally, ?-t? can mark 3rd person plural subjects in the inverse (if the object is also 3rd
person), as in (24). But, as (25) shows, this is not the case for the basic paradigm.
(24) mat is u-q?var-t
3PL.DAT 3SG.NOM VER-love-PL
?They love him/her.?
(25) a. man isini da-p?at?iz?-a(-*t)
3SG.ERG 3PL.NOM PRV-invite-3SG.AOR-PL
?She/he invited them.?
b. mat is da-p?at?iz?-es(-*t)
3PL.ERG 3SG.NOM PRV-invite-3PL.AOR-(PL)
?They invited him/her.?
78
Furthermore, the behaviour of the versionizer vowels also differs between paradigms. In
the basic, they are productive and their appearance is tied to the introduction of an applica-
tive argument (which is not obligatory). There, ?a-? can mark causatives and locatives, ?i-?
can have either a reflexive or possessive meaning, ?u-? appears with benefactive applica-
tive arguments, and ?e-? appears with psychological predicates. In the inverse, however,
they are strictly associated with person features and tense/aspect. In both the present and
in the aorist, ?i-? appears with 1st and 2nd person subjects whereas ?u-? appears with 3rd
person subjects only.
(26) a. me is m-i-q?var-s
1SG.DAT 3SG.NOM 1-VER-love-3.PRES
?I love him/her.?
b. s?en is g-i-q?var-s
2SG.DAT 3SG.NOM 2-VER-love-3.PRES
?You (sg) love him/her.?
c. mas is u-q?var-s
3SG.DAT 3SG.NOM VER-love-3.PRES
?She/he loves him/her.?
(27) a. me is m-i-q?var-d-a
1SG.DAT 3SG.NOM 1-VER-love-TS-3.AOR
?I loved him/her.?
b. s?en is g-i-q?var-d-a
2SG.DAT 3SG.NOM 2-VER-love-TS-3.AOR
?You (sg) loved him/her.?
c. mas is u-q?var-d-a
3SG.DAT 3SG.NOM VER-love-TS-3.AOR
?She/he loved him/her.?
79
The table below summarizes the agreement patterns surveyed in this section. In the in-
verse agreement paradigm, the prefixes track the subject with the ?object? markers from
the basic agreement paradigm: ?m-? marks 1st person singular subjects, ?gv-? marks 1st
person plural subjects, and ?g-? marks 2nd person subjects regardless of number. The
prefix ?v-? can be described as a 3rd person subject marker that only appears when the
object is 1st person, but, as discussed in the previous section, this is not the only way to
characterize this prefix.
(28) Series III ? Inverse agreement patterns (present)
@
@ O
@ 1SG 1PL 2SG 2PL 3SG 3PL
@
S @
@
1SG ? ? m-i- -x-ar m-i- -x-ar-t m-i- -s m-i- -s
1PL ? ? gv-i- -x-ar gv-i- -x-ar-t gv-i- -s gv-i- -s
2SG g-i- -v-ar g-i- -v-ar-t ? ? g-i- -s g-i- -s
2PL g-i- -v-ar-t g-i- -v-ar-t ? ? g-i- -t g-i- -t
3SG v-u- -v-ar v-u- -v-ar-t u- -x-ar u- -x-ar-t u- -s u- -s
3PL v-u- -v-ar v-u- -v-ar-t u- -x-ar u- -x-ar-t u- -t u- -t
Importantly, the distribution of the plural marker ?-t? in the inverse is not straight-
forwardly a mirror image of its distribution in the basic; the plural marking patterns will
provide the main empirical evidence for the argument that the probe responsible for per-
son agreement first evaluates v(P). In particular, the observation that ?-t? marks 3rd person
plural inverse subjects just in case the object is also 3rd person suggests that this combi-
nation of person features and structural positions, and only this combination, allows the
probe responsible for number agreement to access arguments inside the vP.
80
Furthermore, the agreement patterns described in this section suggest that inverse
subjects have lower positions that basic subjects, given that the former triggers object
agreement morphology. In the next section, I show that this is indeed the case. I will also
show that both subjects are higher in the structure than true objects are.
4.2 The Georgian clause structure
This section motivates the assumed structure of the Georgian clause in the basic
and inverse agreement paradigms. In section 4.2.1, I show that Georgian has both head-
initial and head-final structures. In section 4.2.2, I discuss the motivations for adopting an
analysis where basic subjects are introduced in Spec,vP (Be?jar, 2003; Be?jar and Rezac,
2009). In section 4.2.3, I argue that inverse subjects are lower than vP?they are intro-
duced by Appl0, based on evidence from the distribution of the prefixes, interaction with
additional arguments in the clause, and binding behaviour (Be?jar, 2003; Lomashvili and
Harley, 2011; McGinnis, 1995, 1997). In either case, the object is generated in the VP.
4.2.1 Argument positions: The basic agreement paradigm
For subjects of verbs that take basic agreement, I follow a family of analyses that
place the subject in a syntactic position accessible to T0 (Legate, 2008; Nash, 2017) given
that nominative case appears to be closely related to this head. For example, nominaliza-
tions (traditionally referred to as the ?masdar? in the Georgian literature) cannot feature
nominative arguments, as shown below, since T0 is absent in these structures.
81
(29) a. Giorg-i lobian-s c??am-s.
Giorg-NOM lobiani-DAT eat-3SG.PRES
?Giorgi is eating lobiani.?
b. (*Giorg-i) lobian-is c??am-a.
Giorg-NOM lobiani-GEN eat-NMLZ
Intended: ?Giorgi?s eating of lobiani.? (Borise, 2019:96)
Following Nash (2017), I assume that the subject does not move to Spec,TP, but the
analysis developed in section 5 does not rely on this property. The proposed analysis also
does not rely on the specifics of the case-licensing system in Georgian as developed in
Nash (2017) and Legate (2008), i.e. whether case is dependent or inherent (more on this in
section 4.3). I assume that subjects of verbs in Series I and II are base-generated in similar
positions given that they trigger the same agreement patterns (specifically, the prefixes).
As such, I follow Be?jar (2003) and Be?jar and Rezac (2009) and adopt the structure below
for clauses with verbs that follow the basic agreement paradigm.
(30) Structure for the basic agreement paradigm4
...
T0
DPsubj
[ ]PL [ ]PART v0
DPobj V0
[ ]PART
4Note that this structure does not include a VoiceP projection; I remain agnostic as to whether Georgian
has such a layer or not.
82
Positioning the subject in Spec,vP, regardless of case or series, is motivated pri-
marily by the person-marking agreement prefixes. In particular, the distribution of ?v-?
throughout the basic agreement paradigm suggests that it is sensitive to the combination
of 1st and 3rd person arguments. Thus, the probe that expones the prefixal agreement
markers must be able to evaluate both the subject and object. To derive this sensitivity,
I adopt Cyclic Agree approaches where both arguments are targeted by a single probe
(Be?jar, 2003; Be?jar and Rezac, 2009).5 In the remainder of this dissertation, I place the
external argument (i.e. the subject of verbs in Series I and II) in Spec,vP and the internal
argument in VP. A ?-probe on v0 will thus encounter the internal argument first, while
also being able to evaluate the external argument under certain circumstances relating to
the mechanics of Cyclic Agree (see below). Only then can the ?-features of both argu-
ments trigger the spell-out of a unique agreement morpheme, e.g. ?v-?. I discuss further
motivation for Cyclic Agree-based analyses in section 5, as well as the derivations of the
agreement prefixes. In the next section, I argue subjects of verbs that fall under the inverse
agreement paradigm are base-generated lower than v0, namely in Spec,ApplP.
4.2.2 Argument positions: The inverse agreement paradigm
This section focuses on the empirical motivations for placing subjects of inverse
verbs in Spec,ApplP, as shown below.
5Cyclic Agree is not the a priori only way to capture this two-argument sensitivity; one could also
imagine that each argument is targeted by a unique probe, and the prefix is a result of fusion. If that is
indeed the case, then we would expect to see further evidence of fusion throughout the forms of the prefixes.
However, there is a clear preference for object agreement in Georgian, which would be unexpected under
a fusion analysis. Additionally, what looks like fusion is limited to just two combinations of arguments:
1st person acting on 3rd, and 2nd person acting on 3rd. There is no corresponding sensitivity to 3rd person
acting on 1st or 2nd, which we might expect under a fusion analysis of the basic agreement paradigm.
83
(31) Inverse subjects in Spec,ApplP
vP
v0
DPEXP
Appl0
DPTH V0
The first piece of evidence comes from the distribution of the prefixes for inverse
verbs. As was shown in section 4.1.2, the ?m-? prefix appears just in case the subject
is 1st person?but only in the inverse agreement paradigm. In the basic paradigm, ?m-?
marks 1st person objects. The appearance of ?m-? in (32) thus suggests that the 1st person
inverse subject is the first argument encountered by the probe responsible for exponing
the prefixes. I assume that this is because inverse subjects are introduced lower in the
structure than the subjects of non-inverse verbs.
(32) a. me is m-i-q?var-s
1SG.DAT 3SG.NOM 1-VER-love-3SG.PRES
?I love him/her.?
Consider the sentence in (33), which features a ditransitive verb in the basic agreement
paradigm. This construction provides a point of comparison to the sentence in (32), which
I argue feature similar argument structures except (32) lacks an external argument in
Spec,vP. In (33), the clause has an argument in Spec,vP (i.e. a basic subject) and the
prefix tracks the indirect object. This suggests that the probe responsible for exponing
84
the prefixes indeed targets the closest accessible argument, rather than strictly the (direct)
object. In ditransitives in the basic agreement paradigm, it is the indirect argument that
controls this agreement prefix. We can explain this if indirect objects are introduced in a
position that is higher than the direct object and still lower thanv. Spec,ApplP is precisely
such a position, and it is a natural place for an indirect object to be introduced into the
structure.
(33) man mo-m-ts-a me c?ign-i
3SG.ERG PRV-1-give-3SG.AOR 1SG.DAT book-NOM
?She/he gave me a book.?
(34) Structure for ditransitive verbs
vP
man
?she/he?
v0
m-
me
?me?
Appl0
c?igni V0
?book? ts
?give?
Furthermore, the sentence in (35) provides reason to think that the agreement probe
is indeed on v0. Recall that, in the basic agreement paradigm, the ?v-? prefix appears when
the subject is 1st person and the object is 3rd person. In a ditransitive construction, the
?v-? prefix appears when the subject is 1st person and the indirect object is 3rd person.
If the subject is introduced as an external argument is in Spec,vP, and the indirect object
85
as an applicative argument in Spec,ApplP, then a probe on v0 can potentially target both
arguments. This position between the two arguments captures the appearance of ?v-?. If
the agreement probe were lower, say, between the applicative argument and the internal
argument, we would not expect any prefix to appear since the 3?3 argument combination
does not trigger any overt agreement morphology.
(35) me mas c?ign-i mi-v-e-ts-i
1SG.ERG 3SG.DAT book-NOM PRV-1>3-give-1/2.AOR
?I gave him/her a book.?
(36) Applicative arguments in Spec,ApplP (=30)
vP
DPEA
v0
DPIO
Appl0
DPIA V0
Second, it is not possible for another applied argument to be introduced in inverse
verb constructions (Lomashvili, 2011). This follows from the assumption that Appl0 has
already introduced the so-called inverse subject, and thus any additional arguments must
be introduced as adjuncts, as shown below.
(37) a. *Dato-s Gio-s Nino u-u-q?var-s
Dato-DAT Gio-DAT Nino.NOM APPL-APPL-love-3SG.PRES
Intended: ?Dato loves Nino because of Gio.?
86
b. Dato-s Nino Mariam-is gamo u-q?var-s
Dato-DAT Nino.NOM Mariam-GEN because.of APPL-love-3SG.PRES
?Dato loves Nino because of Gio.?
Third, (38a) shows that inverse subjects can bind nominative-object anaphors, showing
that the dative subjects are higher than the nominative objects (Amiridze, 2003; Harris,
1981; McGinnis, 1995, 1997). Reversing the binder and bindee in terms of their argument
roles/case marking results in an unacceptable string, as in (38b). While this does not
show that the dative argument is in Spec,ApplP per se, it does show that it is higher
than the non-dative one. Coupled with the observations concerning (33) and (35) above,
however, we can pinpoint the location as indeed being Spec,ApplP. Further, (38c) shows
that scrambling the nominative argument to the left of the dative one does not resolve the
binding issue.
(38) a. Vano-s u-q?var-s tav-isi tav-i
Vano-DAT VER-love-3SG.PRES self-GEN self-NOM
?Vano loves himself.?
b. *tav-is tav-s u-q?var-s Vano
self-GEN self-DAT VER-love-3SG.PRES Vano.NOM
Intended: ?Vano loves himself.?
c. *Vano u-q?var-s tav-is tav-s
Vano-NOM VER-love-3sc sg.pres self-GEN self-DAT
Intended: ?Vano loves himself.? (Harris 1981:208)
Finally, McGinnis (1997) shows that the behaviour of inverse subjects parallels other
dative-marked arguments throughout the basic agreement paradigm. McGinnis notes that
dative indirect objects can bind nominative direct objects, as in (39a). Nominative direct
objects cannot bind dative indirect objects, even if the nominative argument is scrambled
87
to the left of the dative, as in (39b), in parallel to (38c) above. This shows that, across
the board, dative arguments are structurally higher than nominative arguments, and that
scrambling does not affect binding relations.
(39) a. Nino-m a-nax-a c?ven pat?ara Gela-s tav-isi
Nino-ERG CAUS-see-3SG.AOR 1PL.POSS little Gela-DAT self-GEN
tav-i
self-NOM
?Nino showed our little Gelai himselfi.?
b. Nino-m a-nax-a Gela tav-is tav-s
Nino-ERG CAUS-see-3SG.AOR Gela-NOM self-GEN self-DAT
?Nino showed him/herselfi/* j Gela j.? (McGinnis 1997:5)
The combination of these facts suggest that inverse subjects are introduced by Appl0 as
experiencer arguments, as illustrated below.
(40) Inverse subjects in Spec,ApplP
vP
v0
DPEXP
Appl0
DPTH V0
Since the inverse subject is base-generated below v0, these arguments are evaluated
first by the ?-probe on that head. This mechanism gives rise to the ?object? agreement
patterns that were described in sections 4.1.2 and 4.1.3?namely, that the prefixes used
to mark objects of verbs in the basic agreement paradigm are also used to mark subjects
88
of verbs in the inverse agreement paradigm. In the inverse, the subject is the closest goal
to the probe on v0 (in contrast to the basic, where the subject is introduced in the spec-
ifier of v). With this structure, we also derive the subject-oriented behaviour of inverse
subjects with respect to binding, since the inverse subject is base-generated above the
direct object. Finally, placing inverse subjects in Spec,ApplP explains why applied argu-
ments cannot appear in inverse verb constructions: the position is already occupied, so
additional arguments must appear as adjuncts.
Note that I have assumed a head-initial structure for all projections excluding the
VP thus far. In the next section, I provide an overview of Georgian as a mixed-headed
language, where, in the verbal domain, only the VP is head-final.
4.2.3 Georgian is head-initial and head-final
This chapter assumed thus far that Georgian is head-final in the VP domain, and
that projections above this layer are head-initial. This section discusses the empirical
facts that motivate this structure. First, as shown below, postpositional phrases (PPs),
genitive + noun combinations, participial relative clauses (RCs), and small clauses (SCs)
all show head-final properties (Borise, 2019; Harris, 1981). The sentences below demon-
strate head-finality in the nominal domain: (41) shows that the postposition must appear
at the end of the PP, and (42) shows that the head noun must follow the genitive.
(41) a. alkimik?os-eb-is-tvis
alchemist-PL-GEN-for
?for the alchemists?
89
b. *tvis alkimik?os-eb-is
for alchemist-PL-GEN
Intended: ?for the alchemists?
(42) a. Amerik?-is s?e-ert-eb-ul-i s?t?at?-eb-i
America-GEN PRV-one-SF-PTCP-NOM state-PL-NOM
?United States of America?
b. *S?e-ert-eb-ul-i s?t?at?-eb-i Amerik?-is
PRV-one-SF-PTCP-NOM state-PL-NOM America-GEN
Intended: ?United States of America?
(Borise, 2019:82-83)
We also see head-final properties in the verbal domain, as shown below (see also Lo-
mashvili, 2011). In (43-44), the predicative element has to appear at the end of the
phrase.6 (Borise, 2019) argues that, if we assume that these clauses do not include any
other functional projection, then the verb must be in its base-generated position. The word
order thus reflects the underlying structure of the VP, and, as seen in the (b) examples, the
OV word order indicates that the VP is head-final in Georgian.
(43) a. [PtcpP Ek?a-s c?a-k?itx-ul-i ] c?ign-i
[ Eka-GEN PRV-read.PRF-PTCP-NOM ] book-NOM
?the book Eka read? (Foley, 2013:8)
b. *[PtcpP c?a-k?itx-ul-i Ek?a-s ] c?ign-i
[ PRV-read.PRF-PTCP-NOM Eka-GEN ] book-NOM
Intended: ?the book Eka read?
(44) a. Manana [SC Gela-s c??k?vian-ad ] tvl-i-s
Manana.NOM [ Gela-DAT smart-as ] consider-SM-3SG.PRES
6See Borise (2019) for more examples of Georgian head-finality in the VP domain, e.g. object + verb
idioms and nonfinite + finite verb constructions.
90
?Manana considers Gela smart.?
b. *Manana [SC c??k?vian-ad Gela-s ] tvl-i-s
Manana.NOM [ smart-as Gela-DAT ] consider-SM-3SG.PRES
Intended: ?Manana considers Gela smart.?
When we look beyond the nominal and verbal domains, however, we see that Georgian
displays head-initial properties. For example, (45) shows that the complementizer of
embedded CPs must appear at the beginning of the phrase. The distribution of relative
pronouns is similar, as they must also appear in initial position, shown in (46).
(45) Marik?a pikrob-s [CP rom Giorgi-m (*rom) mankana-s
Marik?a.NOM think-3SG.PRES [ COMP Giorgi-ERG COMP car-DAT
(*rom) i-q?id-a (*rom) ]
COMP VER-buy-3SG.AOR COMP ]
?Marika thinks that Giorgi bought a car.?
(Borise, 2019:85)
(46) a. [DP k?ac-i [CP romel-ma [C -c [TP ... [vP s?e-gh-eb-a
[ man-NOM [ which-ERG [ COMP [ ... [ PRV-paint-SF-3SG.AOR
saxl-i ] ] ] ] ]
house-NOM ]
?the man who painted the house?
b. [DP k?ac-i [CP vin [C -c [TP ... [vP s?e-gh-eb-a
[ man-NOM [ who [ COMP [ ... [ PRV-paint-SF-3SG.AOR
saxl-i ] ] ] ] ]
house-NOM ] ] ] ] ]
?the man who painted the house?
(Borise, 2019:87-88)
While there is ongoing discussion in the Georgian literature concerning its head-final vs.
head-initial status (Asatiani and Skopeteas, 2012; Lomashvili, 2011), I will not discuss
91
the finer details here since the proposed analysis of Georgian agreement does not rely on
head-initial or head-final properties. I will instead note that Georgian is similar to German
in its clausal structure?namely, that C is head-initial while the verbal domain is head-
final?and that the observed mixed-headedness pattern is consistent with the Final-Over-
Final-Condition (Sheehan et al., 2017). I will adopt the structure below for the remainder
of this thesis, but only for simplicity; to reiterate, the analysis that will be proposed in
chapter 5 does not rely on this exact structure.
(47) Georgian clausal architecture
CP
C0
T0
Aux0
Asp0
v0
Appl0
... V0
4.3 Georgian case-marking
This section discusses general case alignment properties in Georgian. The proposal
developed in the next section does not rely on any particular model of deriving Georgian?s
case alignment; as we will see, every DP goal in the three main cases (ergative, nomina-
tive, dative) is visible to agreement probes. Adjudicating between competing theories of
92
case assignment in Georgian is thus orthogonal to the goals of this dissertation. Rather, I
will lay out a broad description of the case marking patterns as well as two major analyses
of how they arise.
As mentioned in section 4.1.1., Georgian distinguishes between seven cases: nom-
inative, ergative, dative, genitive, instrumental, adverbial, and vocative. Of these seven,
only the first three mark subjects throughout the various series and screeves. As shown
below in (48), repeated from section 4.1.1, subjects in Series I are marked nominative,
(some) subjects in Series II are marked ergative, and (some) subjects in Series III are
marked dative.
(48) Georgian conjugation classes
SUBJECT CASE SERIES SCREEVE
Nominative Ia Present Imperfect Conjunctive present
Future Conditional Conjunctive future
Ergative IIa Aorist Conjunctive past
Dative IIIb Present perfect Pluperfect Conjunctive perfect
(evidential)
a Basic paradigm
b Inverse paradigm
This is not the whole picture, however. First, not all subjects are treated alike within
the three series: in Series I, all subjects, regardless of agency or transitivity, are marked
the same to the exclusion of objects. This is shown below, where the subjects are marked
nominative and the object is marked dative.
(49) a. Giorg-i dghes mankana-s q?idul-ob-s.
Giorg-NOM today car-DAT buy-SF-3SG.PRES
?Giorgi is buying a car today.?
93
b. Elene a-cemin-eb-s.
Elene.NOM VER-sneeze-SF-3SG.PRES
?Elene is sneezing.?
c. Moc?veneba u-c?inar-d-eb-a.
ghost.NOM VER-disappear-INCH-SF-3SG.PRES
?The ghost is disappearing.?
(Borise, 2019:80)
In Series II and III, however, there is an ?activity? split among subjects (Harris, 1985).
Only ?active? subjects (i.e. subjects of transitive and unergative verbs) are marked erga-
tive in Series II and dative in Series III. ?Inactive? subjects (i.e. subjects of unaccusative
verbs) are marked with the same case as objects in Series II and III?namely, they are all
marked with nominative regardless of series. In (50a-b) below, for example, the transitive
and active subject are respectively marked ergative, while the inactive subject in (50c) is
marked nominative.
(50) a. Giorgi-m gus?in mankana i-q?id-a.
Giorgi-ERG yesterday car.NOM VER-buy-3SG.AOR
?Giorgi bought a car yesterday.?
b. Elene-m da-a-cemin-a.
Elene-ERG PRV-VER-sneeze-3SG.AOR
?Elene sneezed.?
c. Moc?veneba ga-u-c?inar-d-a.
ghost.NOM PRV-VER-disappear-SM-3SG.AOR
?The ghost disappeared.?
(Borise, 2019:80)
If we abstract away from the active/inactive subject case-marking, Georgian displays a
TAM-based split-ergative case alignment: in Series I, Georgian exhibits a nominative-
94
accusative case-marking system, whereas in Series II and III, Georgian seemingly dis-
plays an ergative-absolutive alignment. Indeed, the primary axis along which the Screeves
(and thus, the Series) are distinguished is Tense/Aspect, as was shown above in (49-50).
Based on this perspective, many have indeed characterized Georgian as a split-ergative
language, e.g. Hewitt (1987, 1995), Nash (1995, 2017); Nash and Rouvert (1997), Boeder
(1979), King (1994), Tuite (1999), and Andre?asson (2001). However, the active/inactive
subject distinction in Series II and III?and its consequences for case-marking?suggests
that treating Georgian as a split-active language may be more accurate, as argued by
Aronson (1970), Comrie (1973), Klimov (1973, 1977); Klimov and Dzidziguri (1979),
Harris (1981), Asatiani (1982), Amiridze (1998, 2006), and Melikishvili (1998, 2001).
See also Anderson (1984) and subsequent work for more discussion on Georgian case
alignment.
Moving on from the contentious status of Georgian as a split-ergative vs. split-
active language, I will now discuss the general mechanics of case assignment proposed
by Legate (2008) and Nash (2017). While both analyses assume that case in Georgian is
licensed in situ, and that nominative case is related to T0, the crucial difference between
the two models lies in how ergative case is assigned. In Legate (2008), ergative is in-
herent and assigned in situ in Spec,vP. Once the external argument is assigned ergative,
it no longer enters into the case-assignment calculus. The remaining argument is then
assigned nominative under licensing by T0, which explains why nominative case disap-
pears in nominalizations (i.e. structures lacking T0), as discussed in section 4.2.1. This is
demonstrated below.
95
(51) Ergative as inherent case
TP
T0
DPERG
v0
Appl0
DP 0NOM V
In contrast, Nash (2017) argues for a hybrid model of case assignment. Ergative
case in Georgian is dependent, which is assigned to the higher of two arguments that
are both within the same vP domain and have unvalued case features. Once ergative (i.e.
dependent) case is assigned, nominative case is head-assigned viia T0, similarly to Legate
(2008). This is illustrated below.
(52) Ergative as dependent case
TP
T0
DPERG
v0
Appl0
DP 0NOM V
96
While there are many additional details to the analyses of Georgian case assign-
ment proposed by Legate and Nash, I will leave those aside here and refer the reader to
those works. The proposed analysis in chapter 5 does not crucially rely on either model of
case assignment, since all nominals in the three main cases (ergative, nominative, dative)
are potential goals in Georgian. That is, Georgian is not a case discriminating (Bobaljik,
2008) language. While many languages make a distinction concerning which cases are
accessible for agreement, as shown below, Georgian does not. For example, Bobaljik
(2008) shows that Hindi only allows agreement with unmarked case, i.e. arguments with
nominative case. Nepali, however, allows agreement with the first two categories: un-
marked and dependent case, i.e. arguments that are marked nominative or ergative.
(53) Case accessibility
UNMARKED CASE > DEPENDENT CASE > LEXICAL/OBLIQUE CASE
(Bobaljik, 2008:303)
As shown below (repeated from sections 4.1.2?4.1.3), Georgian allows agreement with
all three categories: unmarked, dependent, and lexical/oblique case. That is, agreement
mechanisms can target nominals that are marked nominative, ergative, or dative case
(modulo locality principles).
(54) a. me mas v-p?at?iz?-eb
1SG.NOM 3SG.DAT 1-invite-TS
?I am inviting him/her.? (Series I?nominative subject)
b. me is da-v-p?at?iz?-e
1SG.ERG 3SG.NOM PRV-invite-1/2.AOR
?I invited him/her.? (Series II?ergative subject)
97
c. me is m-i-q?var-s
1SG.DAT 3SG.NOM 1-VER-love-3SG.PRES
?I love him/her.? (Series III?dative subject)
In summary, this section provided a broad overview of the case-marking patterns in
Georgian. In Series I (i.e. the present), Georgian exhibits a nominative?accusative case-
alignment, where, specifically, all subjects are marked nominative and all objects are
marked dative. There are splits with respect to case-marking and subject agency in Series
II and III, however. In Series II (i.e. the aorist), Georgian displays an ergative?absolutive
case-alignment, but only active subjects are marked ergative; inactive subjects, as well as
objects, receive nominative. This activity distinction is found in Series III, as well, i.e. the
perfect. There, only active subjects are marked with dative; in contrast, inactive subjects
and objects are receive nominative. This dissertation does not adhere to any particular
model of case-assignment in Georgian, since, as a non-case-discriminating language, all
nominals may be targeted for agreement regardless of its specific case-marking. I now
turn to a discussion of pronominal licensing in Georgian, particularly the different re-
quirements for 1st and 2nd person arguments in exclusion to 3rd person arguments.
4.4 Pronominal licensing
This section provides an overview of the licensing requirements for Georgian pronom-
inals, drawing from patterns related to dative intervention and Person Case Constraints
(PCC) effects. The term ?PCC? describes a range of phenomena that restricts the com-
bination of internal arguments of a ditransitive, and it comes in at least four types: (i)
Weak, (ii) Strong, (iii) Me-First, and (iv) Ultrastrong. I will not discuss the properties of
98
each variety here, but see, e.g., Nevins (2007) for further discussion. Before continuing,
note that Georgian has Strong PCC effects, which force the theme to be 3rd person in the
presence of an indirect object (see, e.g.,(54a?c) below).
I will begin by overviewing the general properties of ditransitives in Georgian. First,
we know from the previous discussion concerning the distribution of the prefixes and
binding that the indirect object is higher than the direct object, and it is marked with dative
case. Although 1st and 2nd person pronouns are syncretic for all case-marking, (55)
shows that dative case is detectable with 3rd person pronouns (as well as non-pronominal
noun phrases).
(55) man mi-ts-a mas c?ign-i
3SG.ERG PRV-give-3SG.AOR 3SG.DAT book-NOM
?S/he gave him/her a book.?
In Georgian, 1st/2nd person direct objects are prohibited in the context of an indirect
object (Harris 1981; Rezac 2009, 2011). The verb in (56) below appears with the 2nd
person prefix ?g-?, showing that the probe targets the 2nd person indirect object. If the
direct object is 3rd person, as in (54a), the sentence is acceptable. If, however, the direct
object is 1st person, as in (54b), the resulting string is unacceptable. This contrast shows
that a 1st person direct object cannot be licensed in its base-generated position when a
2nd person argument intervenes for agreement. As shown in (54c), there is a possible
repair strategy that allows 1st and 2nd persons to co-occur in the licensing domain of v0:
a 1st person direct object can appear with a 2nd person indirect object (and vice versa)
just in case the direct object is encased in a nominative-marked reflexive, which Harris
99
(1981) refers to as object camouflage (see also Amiridze, 2006; Rezac, 2009, 2011). I
discuss this repair strategy in more detail at the end of this section.
(56) a. man mo-g-ts-a s?en c?ign-i
3SG.ERG PRV-2-give-3SG.AOR 2SG.DAT book-NOM
?S/he gave you a book.?
b. *man mo-g-q?id-a s?en me
3SG.ERG PRV-2-sell-3SG.AOR 2SG.DAT 1SG.NOM
Intended: ?S/he sold me to you.?
c. man mo-g-q?id-a s?en c?em-i tav-i
3SG.ERG PRV-2-sell-3SG.AOR 2SG.DAT 1SG.POSS-NOM self-NOM
?S/he sold me to you.?
I would like to now draw a connection between the properties involved in this Georgian
pattern and those involved in Basque absolutive-displacement phenomena. Rezac (2008)
shows that there are two types of unaccusative constructions in Basque: one where the
dative is higher than the absolutive (DAT>ABS), and another where the absolutive argu-
ment is higher than the dative (ABS>DAT). However, it is only in the DAT>ABS cases
where 1st and 2nd person arguments cannot appear as direct objects, as illustrated below.
Note that this is a strong PCC effect.
(57) *Nii Itxaso-ri j gustatzen ni-a-tzai-o j
I.ABS Itxaso- ?DAT liking 1-TM- D-3
Intended: ?Itxaso likes me.?
There is a possible repair strategy, however: if the 1st/2nd person object is marked with
ergative case (instead of absolutive), the resulting string is acceptable.
100
(58) Itxaso-rii zu-k/*zu j gustatzen d-i-oi-zu j
Itxaso-DAT you-ERG/ABS liking X-?ED-3-2
?Itxaso likes you.? (NB: no agreement form at all available with zu)
Rezac (2008) proposes that 1st and 2nd person arguments are subject to a Person Licens-
ing Condition (PLC), which dictates that [PARTICIPANT]-bearing arguments must enter
into an Agree relation with a functional head in order to be licensed(Be?jar and Rezac,
2003). Thus, assuming a structure like (59) below, 1st/2nd person absolutive arguments
cannot be licensed by v0 since the higher dative argument blocks agreement with the ob-
ject. However, 1st/2nd person absolutive arguments may move into a higher agreement
domain to avoid PLC violations, as in (60). By moving into the agreement domain of T0,
they can receive ergative case there and fulfill their licensing requirement by entering into
an Agree relation with T0.
(59) PLC violation
TP
T0
v0
DPDAT
Appl0
V0 DPABS
101
(60) PLC repair
TP
DPERG
T0
v0
DPDAT
Appl0
V0 DPABS
In striking parallel to what Rezac (2008) shows for absolutive-displacement in
Basque, Georgian too allows otherwise-unlicensed participant pronouns to move?under
certain circumstances?to a higher agreement position, where they can be licensed-by-
agreement. Specifically, in Georgian, the specifier position of a head H0 above vP (and
below TP) is a higher landing?and licensing?site for [PARTICIPANT]-bearing objects
which would otherwise not be licensed in their base-generated positions. The auxiliary
?-var, -xar? verb forms for inverse constructions with 1st and 2nd person objects, re-
peated below, as well as the Strong PCC effects in Georgian ditransitives both suggest
that [PARTICIPANT]-bearing arguments must be licensed.
(61) a. s?en me g-i-q?var-v-ar
2SG.DAT 1SG.NOM 2-VER-love-1-be.PRES
?You (sg) love me.?
b. me s?en m-i-q?var-x-ar
1SG.DAT 2SG.NOM 1SG-VER-love-2-be.PRES
102
?I love you (sg).?
In ditransitives, a non-licensed 1st/2nd person theme must be encased in a PP-like struc-
ture if there is an intervening argument. In the inverse, however, 1st and 2nd person
inverse objects can (and must) move through the edge of the vP phase on their way to be
licensed by this higher head, a relation indicated by agreement on the dummy ?ar? ?be?. I
assume that this movement is triggered by an EPP feature on v0. Specifically, in the verbal
structure corresponding to the basic agreement paradigm, that EPP feature is discharged
when v introduces an external argument (and thus 1st/2nd person themes cannot move for
licensing in ditransitives). Since v0 does not introduce an argument in the verbal structure
corresponding to the inverse agreement paradigm, the EPP feature remains active and
triggers movement for non-licensed 1st/2nd person themes.
Evidence for the view that the introduction of external arguments can block move-
ment for internal ones can be found in causative constructions. Since causative v0 intro-
duces an external argument in its specifier, movement through the vP should be blocked,
given that the EPP feature on v0 has been discharged. We thus expect to lose the aux-
iliary ?-var, -xar? forms of 1st and 2nd person agreement morphology in these cases?a
prediction that is borne out, as shown below.
(62) a. Dato Giorgi-s c?emstavs a-q?var-eb-s
Dato.NOM Giorgi-DAT 1SG.REFL CAUS-love-TS-3SG.PRES
?Dato makes Giorgi love me.?
b. Dato Giorgi-s s?enstavs a-q?var-eb-s
Dato.NOM Giorgi-DAT 2SG.REFL CAUS-love-TS-3SG.PRES
?Dato makes Giorgi love you.?
103
In the sentences in (62), the 3rd person applicative argument Giorgis intervenes for agree-
ment from v0 and prevents licensing of the [PARTICIPANT]-bearing theme argument. In a
non-causatived construction, that theme argument could move through the vP phase edge
to be licensed higher. Here, however, the introduction of the 3rd person external argu-
ment/causer Dato discharges the EPP feature on v0 and blocks movement through the vP
edge. Thus, the [PARTICIPANT]-bearing theme is trapped in its base-generated position,
and ?object camouflage? (Harris, 1981) applies as a last resort operation (as in ditransitive
constructions). While questions remain about the proper analysis of object camouflage,
the empirical data show that a [PARTICIPANT]-bearing theme argument does not need to
be licensed by agreement proper once camouflage applies. It remains in its VP-internal
position, and ?-var, xar? do not appear since licensing from the higher head H0 is not
triggered in the derivation.
Summarizing the conclusions drawn in this chapter, evidence from the distribution
of prefixes, binding, and interaction with additional arguments suggest that inverse sub-
jects are higher than objects. They are also lower than basic subjects, since they trigger
?object? agreement morphology (and block agreement with the object). Taken together,
these facts point to inverse subjects being merged in Spec,ApplP. In contrast, basic sub-
jects are introduced in Spec,vP. In both paradigms, objects are in VP.
The core difference between these two clause structures is the position of ?subjects?,
which interacts with licensing requirements of 1st and 2nd person arguments. Together,
they derive the differences in person- and number-agreement. In the basic paradigm, the
external argument is accessible to the higher probes on T0 (and the internal argument is
accessible to the probe on v0). In the inverse paradigm, however, the experiencer applica-
104
tive argument is inside the vP ?phase? and thus inaccessible; it also blocks agreement with
the internal argument.
The view that vP-internal experiencer applicative arguments intervene for agree-
ment from v0 is a crucial part of the proposed analysis in chapter 5. This intervention
results in 1st and 2nd person inverse objects moving through the edge of the vP in order
to be agreed with?and licensed by?a higher functional head. Importantly, the licensing
requirements exhibited by 1st and 2nd person pronominals in Georgian can be indepen-
dently observed via PCC effects.
In the next chapter, I provide an analysis of the Georgian agreement system that
spans both the basic and inverse agreement paradigms, while also showing that the ap-
parent ?phasehood? of the vP is epiphenomenal. That is, v is a ?-intervener in Georgian
since, as a locus of ?-agreement, it bears ?-features and is a viable target for operations
that originate outside of the vP domain.
105
Chapter 5: Phasehood as ?-intervention in Georgian
This chapter provides a comprehensive analysis of the Georgian agreement system,
focusing in particular on two intriguing agreement patterns between the basic and inverse
agreement paradigms. First, as shown in (1), the verbal plural marker ?-t? can mark 1st
and 2nd person plural subjects, which is not the case for 3rd person plural subjects, as
shown in (2).
(1) a. c?ven is da-v-p?at?iz?-e-t
1PL.ERG 3SG.NOM PRV-1-invite-1/2.AOR-PL
?We invited him/her.?
b. tkven is da-p?at?iz?-e-t
2PL.ERG 3SG.NOM PRV-invite-1/2.AOR-PL
?You (pl) invited him/her.?
(2) mat is da-p?at?iz?-eb-en(-*t)
3PL.ERG 3SG.NOM PRV-invite-TS-3PL.AOR(-PL)
?They invited him/her.?
The restriction in (2) is only found in the basic agreement paradigm, however. In the
inverse agreement paradigm, ?-t? exceptionally marks 3rd person plural subjects?but
only when the object is also a 3rd person argument, as in (3). This number agreement is
one of two puzzles that this chapter will address.
106
(3) mat is u-Xvar-t
3PL.DAT 3SG.NOM VER-love-PL
?They love him/her.?
Second, 1st and 2nd person objects of verbs in the inverse agreement pattern trigger a
distinct set of agreement morphology that is not found in the corresponding configura-
tions in the basic agreement paradigm. As shown below, 1st and 2nd person objects are
marked as additional agreement affixes on the auxiliary dummy-verb ?ar? ?be?. In these
configurations, 1st and 2nd person inverse objects also block number agreement with the
3rd person plural inverse subject. This person agreement is the second puzzle that this
chapter will address.
(4) a. mat me v-u-q?var-v-ar-(*t)
3PL.DAT 1SG.NOM 1-VER-love-1-be.PRES-PL
?*They love me.?
b. mat s?en u-q?var-x-ar-(*t)
3PL.DAT 2SG.NOM VER-love-2-be.PRES-PL
?*They love you (sg)?.
This is not to say that these are the only noteworthy agreement patterns?the Geor-
gian agreement system is notoriously complex and has been studied extensively over a
number of years in several frameworks (e.g. Harris, 1981; Anderson, 1991; Be?jar, 2003,
Be?jar, and Rezac 2009; McGinnis, 2013; Blix, 2016; among many others). A central puz-
zle that this chapter focuses on is the distribution of the plural marker ?-t? with different
combinations of arguments throughout both the basic and inverse agreement paradigms.
This chapter also focuses on the puzzle revolving around the auxiliary verb forms re-
quired by 1st and 2nd person objects in the inverse agreement paradigm. I will go into
107
further detail on Georgian agreement in the following sections, and provide a full set of
derivations. I will also argue that there is value in investigating a well-defined slice of the
empirical pie (viz. the distribution of ?-t?) in this manner.
I will build an analysis that treats ?-t? as an exponent of the NUMBER agreement
probe (Anagnostopoulou, 2003; Be?jar, 2003; Laka, 1993; Preminger, 2011b; Shlonsky,
1989; SigurDsson, 1996; SigurDsson and Holmberg, 2008) whose agreement domain can
be expanded if and only if the PERSON probe (ibid.) agrees with the vP ?phase? (on domain
expansion, see Be?jar and Rezac, 2009; Keine and Dash, 2018; Clem, 2018; inter alia).
This analysis will provide empirical support for the PIC-is-intervention approach. The
crux of this argument derives from the number agreement patterns discussed above, which
arise when person agreement first targets v0. Since this head bears a ?-probe, it also,
by extension, bears ?-features. Furthermore, I assume any probe that has successfully
targeted a DP copies [?] onto itself, and [?] is a viable target for person and number
probes alike.
Thus, as a goal for ?-agreement, v0 is made transparent by the initial Agree relation
when a higher probe on T0 searches for person features. Due to the cyclic nature of Agree
in Georgian, a subsequent search for number features allows the probe on T0 to bypass v0
entirely and target a vP-internal argument. My proposal thus undermines the necessity of
Transfer as a component of the syntactic derivation.
108
5.1 An analysis of Georgian agreement
This section lays out the details of the proposed analysis of the Georgian agreement
system. I adopt the structures below in (5-6) for clauses corresponding to the basic and
inverse agreement paradigms; see chapter 4 for a detailed discussion of their motivations.
The major difference between the two is the position of the subject: in the basic paradigm,
the subject is introduced by v0, whereas the subject is introduced by Appl0 in the inverse.
In both cases, the subject is above the object. As will be argued in the following section,
the differences and similarities between the basic and inverse agreement paradigms follow
from placing the basic subject above v0 and the inverse subject below v0. Namely, the
person probe on T0 will necessarily target the subject first in the basic since it is in Spec,vP.
In contrast, the person probe will necessarily first target v in its entirety in the inverse since
the experiencer applicative argument is lower, in Spec,ApplP; this position also explains
why inverse subjects trigger the appearance of agreement markers which, in the basic
agreement paradigm, mark objects rather than subjects.
(5) Structure for the basic
...
T0
DP
[ ]PL [ ]PART v0
DP V0
[ ]PART
109
(6) Structure for the inverse
...
T0
v0
[ ]PL [ ]PART DP
[ ]PART Appl0
DP V0
In both structures, the subject is above the object, which captures the ability of the subject
to bind anaphors regardless of whether they fall under the basic or inverse agreement
paradigm (section 4; see also Harris, 1981; Amiridze, 2003). In the structure in (5),
the external argument in Spec,vP will always be accessible to the probes on T0. I will
argue that this accessibility straightforwardly derives the subject agreement patterns of
the basic paradigm suffixes, since those suffixes reflect the agreement relation established
between T0 and the external argument. In contrast, placing the subject experiencer DP
argument inside the vP phase, as in (6), has the effect of phase unlocking in the structure
corresponding to the inverse agreement paradigm. Specifically, an Agree relation can be
established with the vP since there is no DP at the vP edge to halt the search. Once v0 has
been targeted?and is therefore no longer a viable intervener?the search space for the
number probe on T0 includes the complement of v. Consequently, the number probe can
target the argument in Spec,ApplP. If that argument is plural, this results in plural inverse
subjects triggering the appearance of ?-t??including 3rd person plural ones.
110
Importantly, much of the motivation for this analyses derives from licensing re-
quirements that 1st and 2nd person pronominals (and only 1st and 2nd persons) are sub-
ject to. As was discussed in chapter 4, the observation that 1st and 2nd person arguments
must be licensed in Georgian was independently attested in Strong PCC effects, which
indicated that 1st and 2nd person arguments must be targeted for agreement. In this the-
sis, I will assume that licensing can be achieved via (i) Agree from a licensing head, and
(ii) being introduced by v.
The remainder of this chapter is organized as follows. In section 5.1.1, I discuss the
representation of pronominal ? -features adopted in this paper. The remainder of this sec-
tion discusses the derivations of the agreement slots: the prefixes (5.1.2), the versionizers
(5.1.3), and the suffixes (5.1.4).
5.1.1 Feature geometric representations
I adopt the simplified feature geometric representations of 1st, 2nd, and 3rd per-
son arguments for Georgian below. Each argument is minimally represented as ???, a
property that allows them to be targeted for evaluation by ?-probes. Consequently, ?-
arguments cannot be skipped over by a ?-probe, even if they do not bear the relevant spe-
cific person/number/gender features (see below for a refinement of this statement, in terms
of Deal?s 2015 Interaction and Satisfaction framework). 1st and 2nd person arguments
additionally carry a [PARTICIPANT] feature. Finally, 1st persons also carry [AUTHOR]. In
each case, the argument may bear a [PLURAL] feature to represent plurality.
111
(7) 1st (8) 2nd (9) 3rd
? ? ?
PART (PL) PART (PL) (PL)
AUTH
One could imagine a representation where 2nd persons bear an additional [AD-
DRESSEE] feature. Harley and Ritter (2002) argued that such a representations are pos-
sible, but McGinnis (2005) shows that this makes the wrong prediction for languages
without clusivity distinctions. In such languages, the 1st person plural inclusive is al-
ways conflated with the 1st person plural, never the 2nd person plural. If the [AUTHOR]
and [ADDRESSEE] features were both generally available, we would expect to find cases
where the 1st person inclusive is conflated with 2nd person plurals cross-linguistically.
Since this is not the case, McGinnis proposes that the primary dependent of [PARTICI-
PANT] is [AUTHOR], which is activated if the language has a 1st person vs. 2nd person
distinction. If the language also has a clusivity distinction, the [ADDRESSEE] feature is
specified for 2nd persons as a secondary dependent. Since Georgian does not distinguish
1st person exclusive from 1st person inclusive, I assume the representations above.
5.1.2 Deriving the prefixes
This section discusses the placement of the prefixes on v0, which has become fairly
standard in syntactic analyses of Georgian, following Be?jar (2003); Be?jar and Rezac
112
(2009). The structure corresponding to the basic and inverse agreement paradigms are
repeated below in (10-11).
(10) Structure for the basic
...
T0
DP
[ ]PL [ ]PART v0
DP V0
[ ]PART
(11) Structure for the inverse
...
T0
v0
[ ]PL [ ]PART DP
[ ]PART Appl0
DP V0
Since the subject is introduced higher than v0 in the basic, v0 targets the object first
(and then potentially the subject). In the inverse, however, v0 first targets the subject
in Spec,ApplP (and then potentially the object).
113
Before continuing this particular discussion, I will briefly overview what is meant
by ?potentially?. Deal (2015) shows that, in Nez Perce complementizer agreement, the
probe searches until satisfied, which, in the Nez Perce case, means until it encounters a
2nd person goal. For example, if the 2nd person argument is in subject position, as in
(12a), the only agreement on the complementizer is with the 2nd person. In contrast, if
the 2nd person argument is the object, the complementizer shows agreement with the 2nd
person and the non-2nd person subject, as in (12b).
(12) a. ke-m kaa prosub j cewcew-te?etum proob j
C-2 then PRO.2SG telephone-TAM PRO.1SG
?When you call me.?
b. ke-m-ex kaa prosub j cewcew-te?etum proob j
C-2-1 then PRO.1SG telephone-TAM PRO.2SG
?When I call you.?
(Deal 2015:6)
These data points provide the core empirical evidence in favour of a feature interaction
and satisfaction model of agreement (see Deal 2015 for details of the proposal). That is,
probes halt once they are satisfied by the feature that values them, but they interact with
non-satisfactory features that they encounter during search.
I analyze the Georgian prefix ?v-? as indicative of satisfaction with a 1st person
argument following interaction with a 3rd person argument. As previously discussed, we
know that ?v-? and ?0/ -? respectively mark a particular combination of arguments, rather
than a single argument. In the basic paradigm, ?v-? and ?0/ -? indicate that there is a 1st/2nd
person subject and a 3rd person object; in the inverse, they indicate that the subject is
114
3rd person and the object is 1st/2nd person. This pattern can be captured by placing a
person probe on v0: in the basic paradigms, the probe is between the external and internal
arguments; in the inverse paradigm, the probe is above both arguments. In either case,
the probe searches until satisfied by a [PARTICIPANT] feature and interacts with each
argument along the way (i.e. 3rd persons). In the basic, the probe interacts with the
3rd person internal argument and is satisfied by the 1st/2nd person external argument.
In the inverse, the 3rd person experiencer argument is targeted first by the probe; the
probe is then satisfied by the 1st/2nd person internal argument. I lay out aspects of these
derivations in more detail below.
Feature interaction/satisfaction alone will not capture the distribution of the ?v-?
and ?0/ -? prefixes, however. That is because, in the basic, in cases where the object does
not bear [PARTICIPANT] and can therefore not satisfy the requirements of v, targeting the
subject requires expanding the search domain of the probe. To achieve this, we also need
Cyclic Agree (Be?jar and Rezac, 2009; Clem, 2018; Keine and Dash, 2018). These models
hold that there are multiple rounds of probe searches, which is allowed just in case the
previous cycle was unsuccessful. Be?jar and Rezac (2009) argue that the probe searches
upward for the second cycle; Keine and Dash (2018) propose instead that unvalued fea-
tures on a probe may project. This reworking retains standard notions of the directionality
and locality of Agree?probing is still limited to c-command under the Keine and Dash
(2018) account, since intermediate heads can target an argument in specifier position un-
der sisterhood (which is a subcase of c-command) once the unvalued features on the head
project to the ?bar level?. In this sense, ?projection? is essentially ?percolation?. Under
Bare Phrase Structure (Chomsky, 1994), v0, v?, and vP are all the same syntactic object: If
115
Agree is satisfied during the first cycle, there is no unvalued feature left to trigger another
Agree cycle, since v0 and v? are the same object. Conversely, if Agree is not satisfied
during the first cycle, then there are unvalued features ?on? v?, and so another Agree cycle
is triggered.
Adopting the shared insight behind these proposals, I adopt the following model
for deriving the Georgian prefixes, which are the exponents of v0. In the first cycle of
Agree, the probe targets the closest argument it c-commands; if the probe remains unval-
ued, those features may project and search again. In the second cycle of Agree, the probe
will target the argument in the specifier of the head that hosts the probe. In order to derive
the specific agreement markers, I adopt the system of Vocabulary Insertion (VI) as for-
malized in Distributed Morphology (DM) (Halle and Marantz, 1993). In DM, Vocabulary
Insertion is the mechanism by which the syntax is mapped to phonological forms; specif-
ically, VI pairs syntactic terminals with phonological underlying representations (which
may be null). For Georgian prefixes, I assume that they spell-out according to the VI rules
below.1
(13) Vocabulary Items, prefixes on v0
v- ? [AUTHOR] / [?PARTICIPANT]
0/ - ? [+PARTICIPANT] / [?PARTICIPANT]
gv- ? [AUTHOR, PLURAL]
m- ? [AUTHOR]
g- ? [+PARTICIPANT]
1Note that the 1st person subject marker ?v-? appears to also be present in the 1st person plural object
marker ?gv-?. This may have arisen diachronically throughout the development of Modern Georgian?that
is, ?gv-? may be historically comprised of the 2nd person prefix ?g-? plus the 1st person subject prefix
?v-??but I will assume that, synchronically, ?gv-? is not decomposed.
116
In characterizing the distribution of the vocabulary items in(13), I am assuming
that morphological rules?unlike syntactic ones?can make reference to the absence of
[PARTICIPANT], represented here as [?PARTICIPANT]. Some version of this must be pos-
sible in order to account for, say, the distribution of the English simple present 3rd person
singular /-z/. This particular morpheme is restricted to the environment corresponding to
finite, nonpast, indicative, 3rd person, and singular?a distribution which almost certainly
requires reference to at least some categories that, syntactically speaking, are represented
simply as the absence of a more marked option. In other words, it is exceedingly likely
that at least one of these categories is syntactically represented as simply the absence of
[INFINITIVE], [PAST], [SUBJUNCTIVE], [PARTICIPANT], and/or [PLURAL], respectively.
Given this, I assume VI rules may refer to syntactically underspecified features.
The derivation of the 1st person plural ?gv-?, 1st person singular ?m-?, and 2nd per-
son ?g-? are all instances of first-cycle Agree. In the basic paradigm, the [PARTICIPANT]
probe on v0 searches its domain and is satisfied by the 1st/2nd person arguments. It does
not interact with any other argument since, at this point in the derivation, the only argu-
ment in its domain is the object. This is illustrated in (14).2
2For ease of exposition, I will not include the person and number probes on T0 in the following structures
since they are not discussed in this section.
117
(14) First-cycle Agree (basic)
...
T0
DP
v0
...
DP V0
[PART]PART
?
PART
(AUTH)
In the inverse paradigm, a similar derivation obtains. The [PARTICIPANT] probe
on v0 searches its domain for an appropriate goal, but it instead encounters the inverse
subject first in Spec,ApplP. If the inverse subject is 1st or 2nd person, as in (15), the
search halts?having been satisfied by a [PARTICIPANT] feature?and does not interact
with any other argument.
118
(15) First-cycle Agree (inverse)
...
T0
v0
DP
[PART]PART Appl0
?
DP V0
PART
...
(AUTH)
The remaining prefixes, ?v-? and ?0/ -?, are instances of second-cycle Agree. As
before, the [PARTICIPANT] probe on v0 searches its domain and is satisfied by 1st/2nd
arguments. But, in the basic paradigm, the probe v0 first encounters the object, which?in
this case?is 3rd person, as in (16). Although the 3rd person object does not bear [PAR-
TICIPANT], it?like all arguments?is represented as ???. It is this feature that the probe
interacts with, i.e. the probe necessarily encounters a ? feature but remains unvalued. The
probe is thus not satisfied and its unvalued features project to the level we could descrip-
tively call v?. The probe searches its domain again, where the first argument it encounters
is the 1st/2nd person argument in subject position (under sisterhood with the intermedi-
ate projection). Now the probe is satisfied, crucially following its interaction with a 3rd
person object.
119
(16) Second-cycle Agree (basic)
...
T0 vP
DP v
? v0
DP V0
PART [PART]PART
?
(AUTH)
For the inverse paradigm, shown in (17), the [PARTICIPANT] probe on v0 first en-
counters the 3rd person subject in Spec,ApplP, and?as above?the probe is not satisfied
and the search halts. Its unvalued features project and the probe searches again. Before
this search, however, the 1st/2nd person inverse object moves to Spec,vP for licensing re-
quirements (see section 4.4 in the previous chapter for discussion on why 1st/2nd objects
need to move in the first place). The second search thus results in feature satisfaction?
since the moved 1st/2nd person arguments bear [PARTICIPANT] features?after prior fea-
ture interaction with the 3rd person subject.
120
(17) Second-cycle Agree (inverse)
...
T0 vP
DP v
? v0
DP
PART [PART]PART Appl0
?
DP V0
(AUTH)
?
PART
(AUTH)
In summary, I argued that the distribution of the prefixes between the Georgian
basic and inverse agreement paradigms provides evidence for the following properties:
(i) unvalued features on a probe project and trigger a second cycle of Agree, and (ii) ?-
probes interact with intervening arguments while searching for satisfactory features. I
analyzed the ?gv-?, ?m-? and ?g-? prefixes as instances of first-cycle Agree; second-cycle
Agree yields ?v-? and ?0/ -?. Finally, I argued that the exponence of ?v-? and ?0/ -? indicates
interaction with a 3rd person argument. They spell-out as a result of the [PARTICIPANT]-
bearing probe on v0 searching for a satisfactory goal (i.e. 1st or 2nd person arguments),
while interacting with other arguments encountered in the process (i.e. 3rd persons).
121
5.1.3 Deriving the versionizers
This section discusses the derivation of the versionizer vowels, which I take to be
the exponence of Appl0 reflecting the argument introduced in its specifier. As discussed in
chapter 4 (specifically, section 4.1), there is overt morphosyntactic evidence that inverse
verb constructions feature both vP and ApplP projections. The prefixes that appear in
the basic agreement paradigm, for example, appear in the inverse agreement paradigm as
well, and I argued that the prefixes spell-out on v0. I also showed that the versionizers, i.e.
applicative morphemes, are obligatory in the inverse agreement paradigm. Given that both
of these markers co-occur in inverse verb constructions, I conclude that the corresponding
syntactic structures include both vP and ApplP, as shown below.
(18) 1st/2nd persons in Spec,ApplP
vP
...
v0 ApplP
DP Appl?
? Appl0 VP
i-
PART ...
(AUTH)
122
(19) 3rd persons in Spec,ApplP
vP
...
v0 ApplP
DP Appl?
? Appl0 VP
u-
...
Recall that the applicative morphology is productive in the basic paradigm, but
not in the inverse. I therefore assume that ApplP is ?deficient? in the inverse paradigm
since the applicative morphology behaves more like subject agreement.3 That is, the ?i-?
versionizer invariably appears with 1st/2nd persons whereas ?u-? invariably appears with
3rd persons.
(20) Vocabulary Items, ?versionizers? on Appl0
i- ? [+PARTICIPANT]
u- ? elsewhere
The spell-out of ?i-? and ?u-? is just the exponence of Appl0 based on the argument
it introduces in its specifier. Note that this is the case only for the ?deficient? Appl0 that
appears with verbs in the inverse agreement paradigm. In the case of a fully productive
3This deficiency may be a side effect of the well-defined class of lexical verbs in the inverse paradigm,
i.e. psych-verbs.
123
Appl0, such as a causative, I assume that the more-specified causative Appl0 exponent
?a-? overrides the person-marking ?i-, u-?.
5.2 Deriving the suffixes
Deriving the suffixes makes use of the property that the person probe on T0 can
enter into an Agree relation with the unvalued [PARTICIPANT] feature on v0. The overall
intuition and mechanics behind this proposal are as follows. In the basic, Spec,vP is
always filled by the subject, and so the person and number probes on T0 will always find
a DP argument. In the inverse, however, this will only occur when 1st/2nd person themes
move through Spec,vP to fulfill their licensing requirements. In any other case, Spec,vP
will be empty and so the probes will necessarily encounter vP in its entirety first. As the
?vP? and ?v0? are instances of the same syntactic object (Chomsky, 1994), the person
probe may target the v phase head and establish an Agree relation with it. This has the
effect of ?unlocking? the phase for further Agree operations from the number probe, which
renders the vP-internal 3rd person arguments accessible for agreement.
5.2.1 Inverse agreement paradigm
Recall that in the inverse paradigm, 1st and 2nd person objects are respectively
marked by the prefixes ?v-? and ?x-? on the dummy ?ar? (Nash, 1994; Lomashvili and
Harley, 2011), and 3rd person objects are marked by the suffix ?-s?. I adopt the VIs below
for the distribution of the inverse person-marking suffixes.
124
(21) Vocabulary Items, suffixes on T0
v- ? [AUTH] / H
x- ? [PART] / H
-s ? elsewhere
For now, I assume that 1st and 2nd person inverse objects obligatorily move through
Spec,vP for licensing from a higher head H0 (Rezac, 2008). As previously mentioned,
T0 lacks the ability to license [PARTICIPANT]-bearing arguments in Georgian, since we
otherwise might expect an Agree relation to rescue the derivation in these contexts. That
is, we would not expect the auxiliary ?-var, -xar? verb forms to appear just in case a
1st/2nd person argument is the object of an inverse verb, if T0 was indeed generally
available to license such arguments. For consistency, I assume that T0 is not a licensing
head throughout the Georgian agreement system (see also Lomashvili and Harley, 2011
for arguments that the exponents of T0 do not constitute ?-agreement).4. Movement
of 1st and 2nd person internal arguments through the edge of the vP phase to a higher
licensing position is triggered by the EPP feature on v in applicative unaccusative (viz.
inverse) constructions, a feature that remains active since an external argument was not
introduced. Once the 1st/2nd person themes are in their higher positions, they are in range
of the [PARTICIPANT]-bearing person probe on T0; since they bear the relevant feature,
the probe is satisfied and an Agree relation is established. There, the 1st/2nd person
theme is accessible to a [PLURAL]-bearing number probe on T0 as well. If that argument
plural, the plural marker ?-t? is exponed. Since the person and number probes are on T0,
I assume that the exponence of the inverse-specific ?v-, x-? affixes is conditioned by the
4As for the basic agreement paradigm, I assume licensing occurs for 1st/2nd person external arguments
by virtue of being introduced into the structure by v
125
presence of the H(P) where 1st/2nd person arguments are licensed; ?ar? is the exponent
of the licensing head itself.
(22) 3rd person subject and 1st/2nd person plural object (inverse)
...
T0
DP
[PL]PL [PART]PART H0 vP
?
DP v
PART PL
? v0def
(AUTH)
DP
PART PL [PART]PART Appl0
?
DP V0
(AUTH)
?
PART PL
(AUTH)
A different state of affairs arises when the object is 3rd person. In this case, the
object does not require licensing and therefore remains low and inaccessible to the person
probe on T0. Since there is no argument in Spec,vP, the person probe first encounters vP
in its entirety. Since v0 bears a ?-probe and thus bears ?-features, it must be evaluated
by the probe and cannot immediately be ignored in favour of a lower goal. However,
v0 does not carry the feature that the person probe is seeking, so this agreement relation
renders v0 transparent for subsequent agreement operations. With no further interveners,
126
3rd person inverse subjects in Spec,ApplP are now accessible to the number probe, which
will spell-out as ?-t? if it finds a plural feature, as in (23).
(23) 3rd person plural subject and 3rd person object (inverse)
...
T0 vP
[PL]PL [ ] 0PART v
DP
[ ]PART Appl0
?
DP V0
PL ?
This analysis captures the fact that 3rd person plural subjects trigger ?-t? in the
inverse. As we will see in section 5.2.2, this is not the case in the basic, a fact that the
current analysis is also able to derive. This analysis also captures the fact that number
distinctions among 3rd person plural objects are not tracked by the agreement system in
the inverse (cf. the basic agreement suffixes, which show a number distinction). A person
probe expands the agreement domain for the number probe (both on T0), allowing the
number probe to target 3rd person plural experiencers in Spec,ApplP. Since the person
probe never finds a DP argument in contexts with 3rd person themes, ?-s/-a? will always
be exponed as ?person? agreement (i.e. as the outcome of failed agreement; Preminger,
2014).
127
Additionally, this analysis predicts that any structure with (i) no external argument
in Spec,vP, and (ii) a 3rd person plural argument below v0 will generate the verbal plural
marker ?-t?. Adversity causatives provide such a test case: since these constructions do not
contain a syntactic causer argument in Spec,vP, the ?-probe on v will intervene for person
agreement (for further discussion of the structure of adversity causatives, see Lomashvili,
2011). This initial Agree relation renders the interior of the vP domain accessible for
further Agree operations, as evidenced by the marking of 3rd person plural causees with
?-t?. As shown below, this prediction is shown to be correct.
(24) mat kaghald-i e-glej-in-eb-a-t
3PL.DAT paper-NOM APPL-tear-CAUS-TH-3SG.AOR-PL
?They are caused to tear paper.?
5.2.2 Basic agreement paradigm
Deriving the basic suffixes makes use of the same machinery as the inverse, but
the structure of basic-paradigm clauses ensures that the person probe on T0 necessarily
interacts with an argument since Spec,vP is always filled. At first glance, this might seem
at odds with the phase unlocking accounts discussed in chapter 3. Under those accounts,
phases were derived A-over-A effects?that is, intervention effects?that arise via the
structural relation of dominance. That is, what counts as the intervener for the syntactic
operation at hand is the maximal projection, since it dominates the goal whose features
are being sought after. This is shown below, repeated from the discussion concerning
Rackowski and Richards (2005) in section 3.1.
128
(25) The PIC as A-over-A
CP
...
C[+wh] TP
...
T vP
...
v whP
7
In Georgian, the person probe on T0 will necessarily encounter the vP layer first,
even in structures corresponding to the basic agreement paradigm. Under the A-over-A
configuration illustrated above, we might then expect the vP to act as an intervener. How-
ever, this is not what we see in the Georgian basic agreement paradigm, where both the
person and number probes T0 target the external argument in Spec,vP, and only the exter-
nal argument. Given this agreement, it must be the case that the vP maximal projection
does not intervene. Rackowski and Richards?s (2005) model offers a solution to this issue,
which derives from the definition of closest goal below.
129
(26) A goal ? is the closest one to a probe if there is no distinct goal ? such that for
some X (X a head or maximal projection), X c-commands ? but not ? .
(Rackowski and Richards, 2005:582)
This definition allows the highest specifier of a phrase to be accessible to Agree. Thus,
the DP argument in Spec,vP constitutes the closest goal to the probes on T0, as illustrated
in (27).
(27) Structure for the basic
...
T0 vP
[ ]PL [ ]PART DPsubj
v0
V0 DPobj
[ ]PART
I will now address the differences in the forms of the suffixes between the basic and
inverse agreement paradigms, particularly regarding 3rd person plural subjects, as shown
below. In the basic paradigm, 3rd person plural subjects are marked by a single suffix
which expresses person and number features as well as tense (e.g. ?-dnen? in 28a), but, in
the inverse, they are marked by separate suffixes which independently express person and
number features (e.g. ?-da? and ?-t?, in 28b).
130
(28) a. mat is xedav-dnen
3PL.ERG 3SG.NOM see.IMPF-3PL.IMPF
?They saw him/her.?
b. mat is u-q?var-da-t
3PL.DAT 3SG.NOM APPL-love-3.IMPF-PL
?They loved him/her.?
This difference correlates with whether the probes on T0 target the same syntactic element
or not. That is, both of the probes target the 3rd person plural argument in Spec,vP in the
basic; in contrast, the probes target different elements in the inverse?the person probe
agrees with the v phase head, and the number probe agrees with the 3rd person plural
argument in Spec,ApplP. The model proposed here seeks to capture this correlation.
I assume that VI potentially proceeds in two cycles (Deal, 2015). The first cycle
expones as many features as possible?provided they originate from the same source?
deriving the 3rd person plural specific morphemes in the basic agreement paradigm. A
second cycle of VI targets individual features left unexponed after the first cycle and
expones any appropriate feature separately. This second cycle derives the marking of 3rd
person plural subjects in the inverse agreement paradigm, where they appear to be marked
by both ?-da? and ?-t?. However, the ?-da? morpheme is the exponent of the Agree relation
established between the person probe on T0 and the unvalued [PARTICIPANT] feature on
v. The plural marker ?-t? is the morpheme that marks the 3rd person plural argument, i.e.
the exponent of the successfully-valued number probe on T0.
Multiple cycles of VI also addresses an issue concerning 2nd person plural basic
objects. These arguments trigger ?-t? (modulo phonological considerations) regardless of
the ?-features of the subject, as shown below, even though they are targeted for agreement
131
by v0 (as indicated by the ?g-? prefix). Under the proposed analysis, the 2nd person plural
object is licensed in its base-generated position, so there is no reason for this argument to
move to a higher position where it can be targeted by the number probe on T. Later in this
section, I will argue that this particular occurrence of ?-t? derives from a second VI cycle,
as described above.
(29) man tkven da-g-p?at?iz?-a-t
3SG.ERG 2PL.NOM PRV-2-invite-3SG.AOR-PL
?She/he invited you (pl).?
There are, admittedly, some stipulations to this approach, such as multiple cycles of VI
that include an additional mechanism that fissions unexponed features. There are two ad-
vantages, however. First, it captures the 3rd person plural ?-dnen? in the basic paradigm vs.
?-da? + ?-t? distinction in the inverse, both under the same set of vocabulary items. Since
3rd person plural arguments are in Spec,vP, they are accessible to the number probe on T0
and so we might expect both suffixes to co-occur, e.g. ?mat is da-p?a?t?iZ-es-(*t)? ?They
invited him/her.? This is not the attested pattern, so there must be some fusion/fission
operation at play here. While we could posit a null plural marker in these contexts, this
would be ad hoc.
Alternatively, we could appeal to a phonological rule that reduces [st] and [nt] clus-
ters such that [t] deletes in these contexts. Indeed, the analysis proposed here implicitly
adopts a similar rule; recall from section 2 that both agreement markers for 3SG?2PL
appear in the past tense (?-a? and ?-t?, respectively) but not in the present (where we
would expect ?-s? and ?-t?, but only ?-t? appears). Since final [st] clusters appear to be
132
absent across the board in Georgian, I have assumed that both markers are underlyingly
represented in the present tense, and that the cluster phonologically reduces to [t] (see
Butskhrikidze, 2002; Begus?, 2018; and references therein). If we were to adopt this toy
phonological reduction approach, however, then the [st] cluster would have to reduce to
[s] alone in some circumstances, and to [t] alone in others, despite the fact that the dis-
tinction between the environments is not phonological but rather morphosyntactic. That
is, the underlying [st] cluster in 3PL?2PL (3rd person plural ?-es? with 2nd person plural
?-t?) would result in [s]. But, 3SG?2PL would result in [t] (from the 3rd person singular
?-s? and 2nd person plural ?-t?). Furthermore, there is no parallel for final [nt] clusters (3rd
person plural ?-en? with 2nd person plural ?-t?), and such word-final clusters do appear in
Georgian.
Second, this model captures the differences between the two paradigms without re-
quiring vocabulary items that make reference to a feature like [paradigm: basic/inverse].
That is, the current proposal avoids the need for one set of vocabulary items that would
apply for verbs in the basic paradigm, and another set that would apply for verbs in the in-
verse paradigm. The proposed analysis derives the basic/inverse distinction via the struc-
tural position of arguments (Spec,vP vs. Spec,ApplP), licensing requirements for 1st/2nd
person arguments, and vP intervention. What may at first glance appear to be the tense-
sensitivity of the agreement morphemes with respect to the ?basic? or the ?inverse? falls
out from whether the probes together expone as many features as possible, or separately
expone what they can.
133
The following derivations demonstrate the workings of this model. First, consider a
structure with 1st or 2nd person singular external arguments in Spec,vP. In this case, only
the person probe on T0 is successfully valued, and so ?-e? is exponed.
(30) 1st/2nd person singular subjects (basic)
...
T0 vP
[ ]PL [PART]PART DP
v0 ...
?
PART
(AUTH)
In the case of 1st and 2nd person plural external arguments, both the person and number
probes on T0 are successful. Since there is no VI for [PARTICIPANT, PLURAL] as a single
bundle, a fission operation applies and ?-e? and ?-t? are exponed separately in the second
cycle of vocabulary insertion.
134
(31) 1st/2nd person plural subjects (basic)
...
T0 vP
[PL]PL [PART]PART DP
v0 ...
?
PART PL
(AUTH)
Consider now cases with 3rd person plural external arguments. While the person probe on
T0 fails to be valued, the number probe succeeds. Since there is a VI for [? PARTICIPANT,
PLURAL], ?-es? is exponed in the first vocabulary insertion cycle (capturing its marking
of person and number in a single morpheme).
135
(32) 3rd person plural subjects (basic)
...
T0 vP
[PL]PL [ ]PART DP
v0 ...
?
PL
The remaining two derivations concern 1st and 2nd person plural internal arguments that
co-occur with 3rd person singular external arguments. Under the proposed analysis, 1st
and 2nd person plural internal arguments are both targeted for agreement and licensed
by v0, and thus remain in their base-generated positions inside the VP. Consider first
the derivation featuring a 3rd person singular subject and a 1st person plural internal
argument. The derivation for in this case is straightforward: the person and number
probes on T0 target the 3rd person singular argument in Spec,vP. Both probes fail since
the 3rd person singular external argument lacks both [PARTICIPANT] and [PLURAL], and
thus ?-a? is exponed as the elsewhere. Furthermore, as was discussed in section 5.1.2., the
Agree relation established between v0 and the 1st person plural internal argument triggers
the exponent ?gv-?.
136
(33) 3rd person singular subjects, 1st person plural objects (basic)
...
T0
DP
[ ]PL [ ]PART v0
?
V0 DP
[PART]PART
?
PART PL
AUTH
Finally, consider the derivation featuring a 3rd person singular subject and a 2nd
person plural internal argument. As in the previous derivation, the 2nd person plural in-
ternal argument is targeted for agreement by v0, but, in contrast to the previous derivation,
there is no VI that expones the person and number features of the 2nd person plural ar-
gument in a single morpheme. There is only a VI rule exponing the person feature, i.e.
?g-?. Following Halle and Marantz (1993) and Harbour (2016), I assume in this thesis that
the unexponed number feature undergoes fission. Under this view, the two exponents that
mark 2nd person plural internal arguments, i.e. ?g- ... -t?, are underlyingly from the same
source even though they appear to be separate on the verbal stem.5
At this point, it seems as though we have two accidentally homophonous plural
markers in Georgian: ?-t? that originates from the number probe on T0, and ?-t? that
5Pretheoretically, Georgian has many circumflexes. However, ?g- ... -t? is the only circumflex in the
verbal paradigm; other circumflexes are in the adjectival or nominal domain.
137
originates from v0 under fission. But note that ?-t? expones a lone [PLURAL] feature in
each case, and so the apparent accidental homophony can perhaps be subsumed under
a last-resort strategy. I leave this puzzle for future work, however, and rather discuss
the derivation featuring 2nd person plural internal arguments. As above, the person and
number probes on T0 remain unvalued since the 3rd person external argument in Spec,vP
does not bear the relevant features. Thus, ?-a? is exponed as the elsewhere case. The Agree
relation established between v0 and the 2nd person plural internal argument triggers the
exponent ?g-?, and ?-t? spells-out as well following fission of the [PLURAL] feature.6
(34) 3rd person singular subjects, 2nd person plural object (basic)
...
T0
DP
[ ]PL [ ]PART v0
?
V0 DP
[PART]PART
?
PART PL
In summary, I have shown that the differences between the basic and inverse agree-
ment paradigms follow from variations in three properties: (i) the structural position of
6For an alternative analysis, see Bondarenko and Zomp??, (2020). They propose that ?-t? derives from
leftover agreement, where a probe on T0 agrees with the unexponed number feature on v0, which orginates
from the 2nd person plural object. Their account thus involves interleaving syntactic mechanisms (such as
Agree) and morphological operations (such as VI).
138
the ?subjects?, (ii) licensing requirements for 1st and 2nd person arguments, and (iii) ren-
dering v(P) transparent for cyclic Agree. In verbs corresponding to the so-called basic
paradigm, a 1st/2nd person theme will never need to move for licensing requirements in a
monotransitive construction since it is accessible to and licensed by Agree from v0. Relat-
edly, 1st/2nd person external arguments are licensed in situ by virtue of being introduced
by v0. Importantly, the presence of an external argument in Spec,vP ensures that the per-
son and number probes will always find a ?-bearing goal. The two-cycle approach to
Vocabulary Insertion (Deal, 2015) derives the distribution of the strictly subject-oriented
suffixes. 3rd person plural external arguments are marked by ?-es?, an instance of first
cycle VI, i.e. there exists a VI exponing [? PARTICIPANT, PLURAL] as a single bundle.
There is no such exponent for [+ PARTICIPANT, PLURAL], however, so 1st and 2nd person
plural external arguments are marked by two cycles of VI, i.e. ?-e? and ?-t? separately.
5.3 Previous analyses of Georgian agreement
In this section, I will provide overviews of various influential analyses of Georgian
agreement, and discuss details of these analyses as they pertain to the inverse agreement
paradigm. The broad issue at stake for the proposed analysis outlined in this chapter is
the relation between the basic and inverse agreement paradigms, which is close to, but
not exactly, a mirror-image relation. As will be shown in the following sections, current
analyses of Georgian agreement are either (i) too geared towards the basic paradigm (and
so, the only options they leave for the inverse is to be an exact copy of the basic, or the
mirror image of the basic, neither of which is accurate), or (ii) so powerful as to allow
139
any relation (and so, the inverse paradigm could have in principle been any 6x6 table of
forms), and its close-if-imperfect relation to the mirror image of the basic comes out as
an accident. Additionally, there is also the narrow issue of what allows 3rd person inverse
subjects to be marked by ?-t?, and only when the theme is also 3rd person. This pattern is
not predicted to be possible in any of the accounts surveyed here.
5.3.1 Basic agreement-only accounts
The Georgian agreement system has received a wealth of attention, with analyses
ranging from purely morphological (Halle and Marantz, 1993), purely syntactic (Be?jar,
2003; Be?jar and Rezac, 2009), or a mix of the two (Lomashvili and Harley, 2011). Fur-
ther, analyses have been attempted using several different frameworks, e.g. Optimality
Theory (Foley, 2017) and Nanosyntax (Blix, 2016, 2021), as well as the standard Agree
framework (McGinnis, 2013). I provide a brief overview of these accounts below.
Working within the Optimality Theory (OT) framework (McCarthy and Prince,
1995; Smolensky and Prince, 1993), Foley (2016) draws from the notion of phonolog-
ical conspiracies (Kisseberth, 1970) to derive Georgian agreement. Foley argues that the
blocking patterns observed in Georgian agreement with respect to *-s-t, *en-t, *es-t, and
*gv-t constitute a morphological conspiracy against multiple exponence of the person and
number features of a single argument. Given that multiple exponence of person and num-
ber is possible in corners of the Georgian agreement system, as shown below, it is unclear
how robust this conspiracy is.
140
(35) mas c?ven v-u-q?var-v-ar-t
3SG.DAT 1PL.NOM 1-VER-love-1-be-PL
?She/he loves us.?
Furthermore, each of the agreement markers in Foley?s analysis are specified as subject
or object agreement. As such, extending the analysis to the inverse paradigm requires
a separate (but identical) set of the prefixes, i.e. the ?object? markers in the basic agree-
ment paradigm must be specified as ?subject? markers in the inverse agreement paradigm.
Beyond the question of how appealing such an approach may or may not be, it faces a
problem that is arguably more serious: while the distribution of the prefixes in the in-
verse paradigm may be a perfect mirror image of the basic, this is not the case for the
distribution of the suffixes. For example, the 3rd person plural suffixes ?-es, -en?, which
consistently mark the role of these arguments as subjects throughout the basic agreement
paradigm, do not appear anywhere in the inverse agreement paradigm. These facts are
captured by the analysis put forth here, but it is not clear how they would be derived in
Foley?s system.
Finally, Foley analyzes the verbal plural marker ?-t? as omnivorous number agree-
ment exponing #0, which is relativized to [PLURAL]-bearing arguments. The analysis thus
predicts that, in the inverse paradigm, ?-t? will mark 3rd person plural objects when the
subject is singular. As shown below, this prediction is incorrect?only 3rd person plural
subjects trigger ?-t?, with the added restriction that the object is 3rd person singular.
(36) a. mas isini u-q?var-s/*t
3SG.DAT 3PL.NOM VER-love-3SG.PRES
?She/he loves them.?
141
b. mat is u-q?var-t
3PL.DAT 3SG.NOM VER-love-PL
?They love him/her.?
Blix?s (2016) analysis of Georgian agreement is based in the Nanosyntax framework
(Caha, 2009; Pantcheva, 2011; Starke, 2009, 2011), which derives Georgian agreement
via spans.7 That is, heads and features are organized into binary branching structures
which are operated on by syntax to spell out spans of heads. This approach keeps the
major insights of previous syntactic analyses in that the prefixes track the ? -features of
the object by placing the licensing probe low in the structure, below the subject but above
the object. The distribution of the affixes are captured by a cyclic and bottom-up spell
out. Although Blix?s analysis can account for the basic agreement paradigm, possible
extensions to the inverse paradigm remain unexplored. Since the analysis requires sub-
jects to be higher than objects in order to derive the distribution of the prefixes, modelling
the inverse agreement paradigm would require the object to be higher than the subject.
However, the data discussed in chapter 4 shows that this is not the right structural relation
between the arguments in the inverse.
Halle and Marantz (1993) model Georgian agreement purely in the morphological
module, via a series of fission and fusion operations that, respectively, (i) split a single
node into two, and (ii) create a single node out of two. They assume that the prefixes
begin as a cluster of clitics, which are fused into a single node; they further argue that
plural features fission off (except in the case of the 1st person plural prefix ?gv-?). For
example, the 2nd person plural triggers the insertion of both the person prefix ?g-? and the
7Blix?s analysis also derives the agreement patterns in Laz, a closely-related language to Georgian. I
will not discuss the Laz agreement system here, and rather refer the reader to Blix?s (2016) thesis.
142
number suffix ?-t?; since there is no ?gt-? exponent realizing both 2nd person and plural,
the plural feature is fissioned off and surfaces as ?-t? alone.
However, Halle and Marantz?s account does not derive the preference for object
agreement in Georgian (specifically in the basic agreement paradigm) where the prefixes
generally behave like object agreement markers in basic transitive clauses. As was dis-
cussed in the previous section, the prefixes ?m-, gv-,? and ?g-? respectively mark 1st person
singular, 1st person plural, and 2nd person objects. As was also discussed, there is an-
other marker for 1st person arguments, ?v-?, which marks their role as subjects. Given
that there are two distinct agreement morphemes for 1st person arguments that mark their
roles as subjects vs. objects, we can use their distribution to conclude that object agree-
ment morphology is indeed preferred over subject agreement. The sentence in (37) below
shows that the 1st person subject marker ?v-? cannot appear when the object is 2nd per-
son; rather, the 2nd person object marker ?g-? must appear in the prefixal slot. Under
Halle and Marantz?s system, this preference for object agreement morphology is com-
pletely coincidental?it does not follow from any aspects of the model.
(37) me s?en da-g/*v-p?at?iz?-e
1SG.ERG 2SG.NOM PRV-2-invite-1/2.AOR
?I invited you (sg).?
In a similar framework, McGinnis (2013) analyzes Georgian number marking as a syn-
tactic competition between uninterpretable [#, Group] features on T0 that are subject to
fission if they are not fully valued during first search; this competition derives the plu-
ral suffixes specific to 3rd person plural subjects as well as the ?*-s-t? blocking effect.
143
McGinnis proposes that, since the Georgian TAM suffix marks tense, aspect, and mood,
the syntactic heads C0, T0, and Asp0?i.e., the heads which bear those features?fuse
to form a single morphosyntactic node. Crucially, this fusion occurs before Vocabulary
Insertion (VI), where the Vocabulary entry for ?-t? is specified as [#, Group] and the entry
for ?-s? is the single feature [#]. The Vocabulary entry for 3rd person plural ?-es? is even
more specified, bearing at least [Aorist, Group]. The blocking effect that the plural suffix
?-t? has over the default number suffix ?-s? is thus captured by placing them inside the
same set of VI rules for spell-out on T0: since the Vocabulary entry for the ?-t? is more
specified than ?-s?, ?-t? will block ?-s? in the observed contexts (3SG?2PL). Finally, 3rd
person plural subjects never trigger ?-es? and ?-t? together since the Vocabulary entry for
the former is more specified than the latter.
However, it is likely not the case that this blocking effect arises in the morphosyn-
tax: there is reason to think that ?-s? is blocked in the presence of ?-t? because of phono-
logical/phonotactic restrictions. For example, there are no [st] word-final clusters in Geor-
gian. In addition, the tense-sensitive suffix and the number suffix can co-occur in the past
tense where the tense-sensitive suffix is the vocalic ?-a?; this further suggests a phonotac-
tic restriction rather than a morphosyntactic one. Furthermore, McGinnis? (2013) account
is subject to the same set of problems faced by the accounts proposed by Foley (2016)
and Blix (2016). Namely, by focusing solely on the basic agreement paradigm, possible
extensions to the inverse agreement paradigm are unclear, and the relationship between
the two paradigms is unexplained.
To summarize: although the analyses surveyed in this section laid the groundwork
for a morphosyntactic approach to the Georgian agreement system, restricting the empir-
144
ical desiderata to the basic agreement paradigm leaves the inverse agreement paradigm
unexplained. Additionally, the proposed blocking effects concerning multiple exponence
and plural marking were argued to be inconclusive concerning the actual workings of the
Georgian agreement system, and phonotactic restrictions.
5.3.2 Accounts including the inverse agreement paradigm
The first model to be discussed in this section is Be?jar and Rezac (2009), who
develop a Cyclic Agree account of the prefixes in Georgian agreement where agreement
morphology can reflect first- or second-cycle Agree. They propose a low ?-probe on v0
that necessarily targets the internal argument first. If that argument fully values the probe,
first-cycle agreement obtains and object agreement surfaces. If the internal argument does
not satisfy the probe, however, then the remaining unvalued features on the probe project,
and the probe targets the (external) argument in its specifier (see also Keine and Dash,
2018; Clem, 2018). If the external argument values any of the remaining features on the
probe, then second-cycle agreement obtains and subject agreement surfaces.
The analysis proposed in Be?jar and Rezac (2009) focuses exclusively on the distri-
bution of the prefixes in the basic agreement paradigm; the TAM-sensitive suffixes are not
discussed nor the number suffix ?-t?, and so their account does not serve as a comprehen-
sive analysis. While there is also some discussion of number agreement and the inverse
agreement paradigm in Be?jar (2003), it is also limited to the prefixes. Following Marantz
(1989) and McGinnis (1995, 1997), Be?jar situates inverse subjects lower than v0 (see also
145
chapter 4); in this position, the dative experiencer argument is necessarily targeted first
by the probes on v0 and T0.
However, the analysis does not capture the ability of 3rd person plural inverse sub-
jects to trigger ?-t? just in case the object is also 3rd person. Although the phenomenon
is noted to be possible, it is dismissed as an innovation possible only among a subset of
Georgian speakers (see the author?s footnote 17, p. 134; see also Carmack, 1997). How-
ever, innovative or not, it is entirely robust among the Georgian consultants I have worked
with, and so the pattern needs to be accounted for.
The second model discussed in this section is Lomashvili and Harley (2011), who
provide a unified analysis of the basic and inverse paradigms. The differences between
the agreement patterns are derived via morphotactic templates and drawing parallels to
discontinuous bleeding (Noyer 1992), where the insertion of a more-specified exponent
in one position bleeds the insertion of a less-specified exponent that would otherwise
be inserted in a different position. The two agreement slots in Lomashvili and Harley?s
analysis are (i) prefixal, which expones the features of the subject or the object (i.e., they
may compete in a particular context), and (ii) suffixal, which expones the number feature
of the object (i.e., this is where ?-t? is spelled-out; Lomashvili and Harley do not treat
the tense/aspect suffixes as agreement markers). Lomashvili and Harley assume that the
exponence of these affixes takes place in the (post-syntactic) morphological module.
(38) Verb agreement template (Lomashvili and Harley 2011:244)
AGREEMENT PREFIX ? Verb stem ? AGREEMENT SUFFIX
146
Following Be?jar (2003), they place a low probe on v0 that necessarily targets the in-
ternal argument; they also place a higher probe on T0 that necessarily targets the external
argument first. In either case, the probes may only agree with Case-active DPs, and they
probe for both person and number features. Finally, they argue that a phase-based ac-
count of Spell-Out derives the blocking effects of the agreement morphemes. Finally, Lo-
mashvili and Harley assume that, in Georgian, only arguments that bear [+PARTICIPANT]?
i.e. 1st and 2nd persons?can value a ?-probe. So, for the authors, 3rd person arguments
cannot be agreed with, which is a point that I will discuss in short order.
The logic of their analysis of the basic agreement paradigm is as follows: (i) the
probe on v0 agrees with the internal argument; (ii) the vP phase spells-out and Vocabulary
Insertion (VI) for the agreement slots is triggered; (iii) the probe on T0 agrees with the
external argument in Spec,vP; (iv) construction of the matrix CP results in spell-out of
the remaining structure, triggering a second round of Vocabulary Insertion for both the
agreement slots. Crucial to their analysis is the cyclic nature of phasal spell-out?since
VI occurs within the vP phase, object agreement is privileged. Thus, spell-out of the vP
phase can potentially result in bleeding the exponence of subject agreement, which is in
the TP domain.
Lomashvili and Harley also offer an account of the inverse agreement paradigm,
which uses the same machinery proposed for the basic. For these verbs, they assume that
both the experiencer and theme arguments are VP-internal, i.e. dyadic unaccusatives (or
unapplicative accusatives), a structure that I adopted here, as well.
Given these argument positions, the probe on v0 will necessarily target the higher
experiencer DP first, resulting in what looks like object agreement morphology with the
147
logical subject. Further, Lomashvili and Harley assume that the vP does not instantiate
a phase for these structures, and so the vP does not spell-out until the matrix CP is con-
structed. If there is a 1st person experiencer argument and a 2nd person theme (or vice
versa), there is a morphotactic competition concerning the 1st person singular prefix ?m-?
and the 2nd person ?x-?. Lomashvili and Harley propose that, since ?m-? and ?x-? compete
for the same position in one and the same phase (cf. the basic paradigm, where the vP
phase has Spelled-Out by this point), a Last Resort mechanism of ?ar?-insertion applies
which creates a second prefixal agreement slot. In this case, then, there is no templatic
restriction blocking one prefix over the other: ?m-? fills the prefixal agreement slot for
the main verb, and ?x-? fills the prefixal agreement slot for the dummy auxiliary ?ar?,
which Lomashvili and Harley take to be the default realization of T0 (as with English
do-support).
While Lomashvili and Harley go a long way in providing a unified analysis of the
two agreement paradigms in Georgian, they assume that 3rd person arguments cannot
enter into Agree relations, and thus cannot be targeted by ? -probes. While this view
is not unheard of in the current literature on agreement, their account predicts that the
plural marker ?-t? will never mark 3rd person plural arguments. As shown in section
2.3.1, however, ?-t? can indeed mark such arguments, albeit in a very specific context:
when the 3rd person plural argument is the experiencer argument of a verb in the inverse
agreement paradigm, and the theme is also 3rd person. Lomashvili and Harley do not
discuss this phenomenon at all, and in fact explicitly say that 3rd person plural arguments
never trigger ?-t?. n summary, this section showed the limits of previous analyses that
included the derivation of the inverse agreement paradigm. These analyses included the
148
assumptions, shared with the current account, that inverse subjects are indeed higher than
inverse objects, and that phasal boundaries play a role in the differences between the
two agreement paradigms, but the models fall short of deriving 3rd person plural number
agreement in the inverse paradigm. Each account assumed that 3rd person arguments are
invisible for agreement in Georgian, whether that be for Case or ?-feature reasons. Even
if 3rd person plural inverse subjects were indeed visible to agreement in these accounts,
then, as [PLURAL]-bearing arguments, they should trigger ?-t? across the board. However,
we know this is not the case and so these analyses are faced with both under- and over-
generation issues.
Given that the proposed analysis of the Georgian agreement system?and its main
competitors?have been discussed in detail, I now turn to a discussion of extending the
PIC-as-intervention model to Hindi-Urdu, which has been argued to lack vP phasehood
(Keine, 2017). I will show that the Hindi-Urdu data suggests that phasehood correlates
with ?-probes, and thus ?-features on the ?phase? head in question. In this case, v0 does
not bear a ?-probe, and so the lack of phase effects results from the absence of ?-features
on v0.
149
Chapter 6: Further investigating phases as ?-intervention
The discussion in this chapter addresses two topics. First, I provide an overview
of Keine (2017), which argues that vP is not a phase. Drawing from Hindi-Urdu, Keine
shows that long-distance agreement can span multiple vP boundaries, but notably cannot
span any CP boundaries.1. This suggests that only CPs are phases. I will argue that
these effects arise for reasons that are independently diagnosable from the Hindi-Urdu
agreement system. The loci of ?-agreement in the language is only T0. Since v0 does
not bear a ?-probe in Hindi-Urdu, it does not bear ?-features. Thus, it does not act as an
intervener. This property straightforwardly explains the absence of phase effects in the vP
domain in Hindi-Urdu.
In the remainder of this section, I discuss some remaining questions with respect
to the PIC-as-intervention approach. Namely, I outline potential investigations into the
status of derived vs. base-generated ?-features, as well as selectional conditions related
to operational triggers and pronominal licensing.
1A similar effect is found in wh-licensing phenomena in Hindi-Urdu.
150
6.1 A case study addressing the lack of phase effects
This section discusses the status of vP phases in Hindi, as analyzed in Keine (2017).
The analysis tests the status of the vP as a phasal category, using the Agree operation
as a diagnostic. Given the formulation of phases and the PIC, we should not find ?-
agreement relations across multiple vP clauses if vP is a phase. Keine shows that we do
find such agreement relations in Hindi, and subsequently concludes that vP is not a phase.
I argue that v in Hindi does not bear its own ?-feature, and that, consequently, vP does
not behave as a phasal category. Under the framework adopted in this dissertation, v is
not a ?-intervener in Hindi, and so the absence of vP ?phase? effects is entirely expected.
Before delving into the details of Keine?s analysis, I will first set up the issue at
hand. The main theoretical question addressed in Keine (2017) is the status of vP phases,
particularly with respect to the Phase Impenetrability Condition (Chomsky, 2001). Given
the structure and phasal boundaries outlined below, a prediction of Chomsky?s proposal
is that agreement relations cannot obtain across multiple vP projections.
(1) [TP DP T[u?] [vP v [VP V DP[?] ]]]
That is, embedding multiple vPs between TP and VP should block agreement from T with
an internal argument, since the lowest VP will spell out as soon as the higher of two vP
structures are merged.
(2) [TP DP T[u?] [vP v [vP v [VP V DP[?] ]]]]
151
As will be shown in the remainder of this section, data from Hindi ?-agreement shows that
these relations span across multiple vP boundaries, which constitutes evidence against vP
phases. To begin, note that the verb agrees with the structurally highest non-case-marked
argument in Hindi. As shown in the sentences below, ensuring that the subject is case-
marked forces agreement with the object. As part of an idiomatic expression, the object in
(3b) resists movement. We can thus conclude that the verb is agreeing with an argument
that remains in the vP domain.
(3) a. raam-ne bhains ke aage biin bajaa-yii
Ram-ERG buffalo in front.of flute.F.SG play-PERF.F.SG
?Ram did something futile.? (lit. ?Ram played the flute in front of buffalo.?)
b. #biini raam-ne bhains ke aage ti bajaa-yii
flute.F.SG Ram-ERG buffalo in front.of t play-PERF.F.SG
?The flute, Ram played in front of buffalo.? (idiomatic reading deviant)
(Keine, 2017:178?179)
Hindi also has long-distance agreement, where Agree targets an argument that is further
embedded in a complement clause. Certain complement clauses in Hindi obligatorily
contain a vP projection (Bhatt, 2005; Davison, 2010) and crucially lack a CP layer (Bhatt,
2005; Chandra, 2007; Dayal, 1996). Under the view that vPs constitute phases, long-
distance agreement with an object embedded within a vP complement clause should not
be possible. But, as shown below, such agreement relations are possible, which suggests
that vPs do not constitute phasal boundaries.
(4) raam-ne [bhains ke aage biin bajaa-nii] caah-ii
Ram-ERG buffalo in front.of flute.F.SG play-INF.F.SG want-PERF.F.SG
152
?Ram wanted to do something futile.? (idiomatic reading possible)
(Keine, 2017:179)
(5) [TP T[u?] [vP v [VP V [nonfinite clause [vP v [VP V DP[?] ]]]]]]
Keine further shows that long-distance agreement may also span three vP boundaries,
albeit the resulting sentence is degraded due to processing complexity.
(6) ?Raam-ne [[bhains ke aage biin bajaa-nii] shuruu kar-nii]
Ram-ERG [[buffalo in front.of flute.F.SG play-INF.F.SG] start do-INF.F.SG]
caah-ii
want-PERF.F.SG
?Ram wanted to start doing something futile.? (idiomatic reading possible)
(7) [TP T[u?] [vP v [VP V [nonfinite clause [vP v [VP V [nonfinite clause [vP v [VP V DP[?]
]]]]]]]]]
Contrast this with the behaviour of CPs with respect to long-distance agreement in sim-
ilar configurations. As shown in (8), long-distance agreement cannot target an argument
inside an embedded CP. This suggests that Hindi does have phasal categories, but they are
limited to CPs.
(8) larko?-ne soc-aa/*-ii [CP ki monaa-ne ghazal
bo?ys-ERG think-PERF.M.SG/-PERF.F.SG [ that Mona-ERG ghazal.F
gaa-yii thii ]
sing-PERF.F.SG be.PAST.F.SG
?The boys thought that Mona had sung ghazal.?
(Keine, 2017:181, citing Bhatt, 2005:776)
153
Keine concludes that the behaviour of embedded vP and CP complement clauses with
respect to long-distance agreement in Hindi are indicative of their phasal status: CPs are
phases, and vPs are not. The behaviour of vPs with respect to long-distance agreement in
Hindi provides a testing ground for identifying the universal status of phases. Under the
model of phasehood that I have argued for in this dissertation, the Hindi behaviour follows
from the location of ?-agreement in the structure. That is, all ?-agreement originates
from T in Hindi, not from v. Since v does not bear a ?-probe, it does not constitute a
potential goal for ?-agreement?thus, it is not an intervener. The lack of the phase effects
Hindi is thus expected under this notion of phasality.
Before moving on, I will note that Hindi-Urdu has agreement both on main verbs/auxiliaries
and on participial verbs. At first glance, this looks like a language with ?-probes on both
T0 andv0, which would contradict my claim that all ?-agreement in Hindi-Urdu originates
from T0, and only from T0. But, there is reason to believe that agreement on participles
in Hindi-Urdu does not, in fact, originate onv0 but is instead ?parasitic? onT0, which is
where Agree originates. In order to illustrate this property, consider the arguments put
forth by Bhatt (2005). We see in (9) below that long-distance agreement in Hindi-Urdu
co-occurs with agreement on the embedded infinitival verb.
(9) Ram-ne [rotii khaa-nii] chaah-ii thii
Ram-ERG bre?ad.F eat-INF.F want-PFV.F be.PST.F.SG
?Ram had wanted to eat (the) bread.?
(Bhatt, 2005:769)
154
The sentences below show that the agreement relations in (9) are obligatory. That is, long-
distance agreement without infinitival agreement is not possible, as in (10a); and neither
is infinitival agreement without long-distance agreement, as in (10b).
(10) a. *Ram-ne [rotii khaa-naa] chaah-ii thii
Ram-ERG bre?ad.F eat-INF.M.SG want-PFV.F be.PST.F.SG
Intended: ?Ram had wanted to eat (the) bread.?
b. *Ram-ne [rotii khaa-nii] chaah-aa thaa
Ram-ERG bre?ad.F eat-INF.F want-PFV.M.SG be.PST.M.SG
Intended: ?Ram had wanted to eat (the) bread.?
(Bhatt, 2005: 770)
While I will not discuss the details of the Bhatt?s (2005) analysis here, note that these
agreement restrictions suggest that infinitival agreement (and participial agreement as
well; see Bhatt, 2005 for discussion) is indeed parasitic on long-distance agreement. That
is, the ?-agreement exhibited by embedded infinitival verbs in Hindi-Urdu obligatorily
draws from the ?-agreement exhibited by the matrix verb. Bhatt concludes that only
(finite) T0 bears a ?-probe in Hindi-Urdu. This claim, which I have adopted in this thesis
as well, thus explains both the LDA restrictions and the lack of phase effects in the vP
domain in Hindi-Urdu.
Moving beyond Hindi-Urdu, we might also expect to find phase effects around
syntactic heads that (i) are not canonically considered to be phase heads, and (ii) bear
a ?-probe. Two families of analyses may bear on this prediction, specifically those of
phase extension (den Dikken, 2007) and phase sliding (Gallego, 2006, 2010). Both of
these approaches expand the limits of the phase domain under head-movement?namely,
155
the vP phase domain includes the TP domain just in case v0 undergoes head-movement
to T0. I will not discuss these approaches in greater detail here, but I will note that, in
the phase sliding literature, much of the arguments are based on the relationship between
T0 and ?-agreement in Spanish and other null subject languages. This property hints at
the possibility that, in these languages, T0 acts as a phase head because T0 is the locus
of ?-agreement. If this is indeed the case, then the empirical data underpinning the
phase sliding (and possibly phase extension) literature can be straightforwardly captured
under the approach pursued in this dissertation. I leave this possibility aside for now,
however, and turn to some other remaining questions concerning phases as instances of
?-intervention.
6.2 Remaining questions
In this section, I outline some issues and questions that are raised by interven-
tion accounts of phasehood. While there are many potential research avenues to further
explore, perhaps the most pressing ones at the outset concern the relationship between
wh-licensing and ?-intervention. Or, more generally, A?-dependencies and ?-agreement.
I will discuss two broad questions below.
An obvious source of tension between the standard diagnostics of phasehood and
the view of phasehood put forth in this dissertation concerns the A? nature of evidence
for phasal boundaries. Most (if not all) standard phasehood diagnostics are based on
wh-movement and relativization, and other A?-related processes (Boeckx, 2008; Felser,
2004; Fox, 1999; Lahne, 2008; Legate, 2003; McCloskey, 2006). Notably for our pur-
156
poses, these diagnostics are unrelated to ?-features and intervention. If ?-intervention
is indeed the underlying mechanism to phase effects, then it is not clear how?or why?
the presence of ?-probes on a given functional head would pre-empt the possibilities of
wh-extraction and relativization strategies within that language. As discussed in chap-
ter 3, however, there is robust evidence that ?-intervention in languages such as Tagalog
does indeed bear on A?-operations such as wh-movement. One logical possibility that
addresses the A- vs. A? tension involves categorizing [wh] as a kind of ?-feature. In fact,
the phenomenon of ?wh-agreement? suggests that this may be more than just a logical
possibility. In Abaza, for example, wh-agreement manifests in cases where ?-agreement
is overridden with a special, dedicated exponent just in case the targeted DP is a wh-
phrase (see section 2.2 of Baier, 2018). This suggests that wh- and ?-agreement can be
unified under a single model, as argued by Baier (2018). If we categorize [wh] as a kind
of ?-feature, then capturing A? phase effects in terms of ?-intervention becomes quite a
bit easier.
The canonical restrictions on A?-operations across the CP/vP boundary suggests
that there is indeed an accessibility issue?namely, that CPs and vPs constitute phasal
boundaries, at least in some languages. Under the view of phases argued in this project,
the ?phasal boundary? must constitute a ?-goal; ideally, we would be able to detect those
?-feature correlates alongside the restrictions in A?-operations. While I argued that this
is the case for Hindi CPs and vPs, future work should include a wider crosslinguistic
investigation of these correlates.
The phasehood-as-intervention model should also be extended to the DP phase.
Although the thesis did not discuss this particular phasal domain, potential extensions
157
are quite obvious since DPs are the canonical ?-goals. Namely, it should be possible
to use the same logic proposed in this thesis to make DP phasehood epiphenomenal as
well, and the results of that exploration could perhaps bear on the organization of ?-
features throughout the DP domain. For example, we could imagine that all ?-features
are specified at the maximal DP projection. If this were the case, then we would not
expect extraction out of the DP to be possible since the full bundle of ?-features would
always intervene for Agree. We could also imagine that ?-features are hosted on different
syntactic heads throughout the DP domain, as argued by Kramer (2015) and subsequent
work. In this case, we would expect extraction out of the DP to be possible; in a language
with Cyclic Agree and separate person- and number-probes (and perhaps a gender-probe
as well), the initial Agree search could render the DP layer transparent for further Agree
operations.
Another potential future exploration concerns the nature of ?-features and their be-
haviour with respect to being base-generated or derived. For instance, the CP/TP/vP itself
is taken to be a goal for ?-agreement under intervention accounts of phasehood and the
PIC. One research question that I leave for further exploration is whether the derived ?-
features of functional phrases, which originate from a ?-probe, are subject to the same
syntactic conditions as base-generated ?-features, which originate inside a nominal. Any
results of this investigation will thus bear on questions of locality, agreement, and projec-
tion, given the identity issues between functional and nominal phrases.
Further investigations should also focus on selectional conditions related to oper-
ational triggers and pronominal licensing. In Georgian, for example, pronominal argu-
ments that are unlicensed in their base-generated positions must vacate the vP domain,
158
but this movement strategy is only possible when the edge of the domain is empty. This
restriction only applies to A-movement, however, as A?-movement in Georgian occurs
whether the edge is empty or filled. One possibility for this difference between the two
types of movement concerns the needs of the functional head: under A-movement, the
selectional conditions enforced by v are met since it has selected a complement and in-
troduced its external argument, and so no further operations are triggered. In contrast,
the A?-movement cases are solely mitigated by the conditions of a functional head which,
crucially, is not related to v. Investigating these phenomena will thus address issues con-
cerning selection, movement, and the A/A?-distinction.
Selectional conditions may also bear on some remaining questions for the Tagalog
wh-extraction patterns discussed in chapter 3. In Tagalog, and elsewhere in Austrone-
sian languages, wh-extraction from adjunct clauses is not subject to the same agreement
constraints that were observed in wh-extraction from complement clauses. Thus, phase
unlocking via Agree appears to be optional for adjuncts in Austronesian. I cannot offer
a solution at this juncture, though it seems likely that the agreement restriction between
the two types of wh-extraction patterns is related to c-selection, since only complement
clauses are selected by the embedding verb in order to fulfill some set of conditions.
159
Chapter 7: Conclusion
This dissertation investigated two broad phenomena: (i) the Georgian agreement
system, specifically the distribution of the verbal plural marker between the basic and
inverse agreement paradigms, and (ii) its ramifications and contributions to the reanalysis
of phases and the Phase Impenetrability Condition as (?-)intervention effects. I overview
the core properties of these investigations in the remainder of this chapter.
7.1 A comprehensive analysis of the Georgian agreement system
This dissertation offered a novel analysis of the Georgian agreement system, which
was primarily based on (i) the inverse agreement paradigm, and (ii) the differences in
the distribution of the verbal plural marker ?-t? between the basic and inverse paradigms.
This addressed long-standing puzzles concerning some asymmetries in person and num-
ber agreement in Georgian. The person agreement concerning 1st and 2nd person inverse
objects, which triggers the appearance of auxiliary verb forms that are not found in the
corresponding configurations in the basic paradigm, was derived from pronominal licens-
ing requirements for 1st and 2nd person arguments (which can be independently seen
in Georgian PCC effects). In particular, I argued that these licensing requirements force
1st/2nd person inverse objects to move to a high position, where they subsequently block
160
number agreement with 3PL inverse subjects. The number agreement concerning 3rd per-
son plural inverse subjects, which only occurs when the object is also 3rd person, was
derived under an intervention and Cyclic Agree approach. In particular, I argued that the
?-features on v0 (which originate from its ?-probe) intervene for person agreement since
inverse subjects are merged below v0. Given that initial Agree relation, v0 is transpar-
ent for further operations, which results in the 3rd person plural argument in Spec,ApplP
being the closest goal for number agreement.
More broadly, I showed that the differences between the Georgian basic vs. inverse
agreement paradigms and the distribution of the verbal plural marker cannot be reduced
to flipping the overall patterns. Rather, characterizing Georgian inverse verbs as dyadic
applicative unaccusatives captures the near-mirror image agreement patterns of the agree-
ment paradigms. It also provides more insight into the basic agreement patterns, which is
the default agreement paradigm. In both cases, a low probe on v0 first evaluates the clos-
est c-commanded argument (i.e. the object in the basic, and the subject in the inverse); if
unsatisfied, it targets the argument in its specifier (i.e. the subject in the basic, or a moved
1st/2nd person object in the inverse). Then, a higher person probe on T0, above the vP,
targets the subject in the basic, and the object in the inverse (modulo some restrictions
based on the PLC and accessibility, as discussed).
The proposed analysis builds on standard tools of syntactic and morphological anal-
ysis, such as Agree (Chomsky, 2000) and Vocabulary Insertion (Distributed Morphology;
Halle and Marantz, 1993). There are also several elements that are somewhat newer, such
as as cyclic domain expansion (Be?jar and Rezac, 2009; Clem, 2018; Keine and Dash,
2018) and feature interaction and satisfaction (Deal, 2015) which have nevertheless been
161
motivated elsewhere in unrelated languages (e.g. Nishnaabemwin, Nez Perce, Basque,
and Hindi). Making use of these tools here provides more crosslinguistic support for
these growing bodies of analyses, and allows us to describe the relationship between the
basic and inverse agreement paradigms in Georgian. Crucially, this allows us to capture
the ways in which inverse agreement deviates from being the mere mirror image of ba-
sic agreement?namely, in ways that are associated with other things we know about the
syntax of dative arguments and applicatives crosslinguistically. There is plenty more to
be done, but these are the initial steps of what our theory of agreement needs to look like
in order to model one of the most complex known agreement systems in natural language.
7.2 Phases as ?-intervention
A crucial part of the analysis of the Georgian agreement system involved analyzing
the vP as an intervening goal for ?-agreement. That is, in Georgian, v0 bears a ?-probe,
which translates to v0 bearing ?-features. As the closest goal for ?-agreement, the inte-
rior of the Georgian vP ?phase? is inaccessible under general principles of minimality and
intervention. Once that intervener is made transparent under Agree, syntactic material
lower in the structure can then be accessed in subsequent cyclic Agree operations. This
proposal joins and contributes to a growing family of approaches that subsumes phase-
hood and the PIC under more general locality constraints (Rackowski and Richards, 2005;
Halpert, 2019, and others).
More specifically, the proposed analysis of Georgian agreement adjudicates be-
tween two broad families of phasehood and the PIC, in particular with respect to Trans-
162
fer and deriving the crosslinguistic variation in phasal categories. Under the PIC-as-
intervention model, there is no theoretical requirement for an independent constraint like
the PIC since its effects can be reduced to minimality. In this dissertation, I argued that
Georgian number agreement targets a vP-internal argument just in case person agreement
first targets v0 itself. My proposal thus necessitates a view where ?phases? do not undergo
Transfer to the interfaces.
Finally, I argued that the PIC-as-intervention approach provides a clear explana-
tion as to why the ?phase? intervenes. This is especially notable in contrast to the the
alleviating-the-PIC approach, which inherits the problems regarding defining phases from
standard phase theory. Namely, under PIC-as-intervention approach, there will hopefully
be observable correlates of the (morpho-)syntactic features that derive the intervention
effects. In my analysis of Georgian agreement, I showed that the vP ?phase? is a case
of ?-intervention, where the ?-probe on v is an intervener with respect to person agree-
ment. My proposal thus suggests that ?phasehood? across languages may correlate with
phi-agreement. Thus, for any given language, vP behaves as a ?phase? just in case v bears
a ?-probe. Languages without a vP ?phase? lack a ?-probe on v, which I argued to be the
case in Hindi-Urdu.
There are several remaining questions given the scope of this project, as well as its
consequences for linguistic theory. This thesis does not aim to completely subsume all of
phase theory under general locality constraints; it may be the case that some phrases are
indeed special in the sense that they signal points in the derivation where certain opera-
tions can and must apply (e.g. Cyclic Linearization; Fox and Pesetsky, 2004). Rather, this
dissertation aims to recategorize certain phase-based phenomena as intervention effects
163
by discussing what the morphosyntactic correlates of such a recategorization might look
like. Thus, we gain a better understanding what counts as a phase crosslinguistically,
which will hopefully result in clearer insights into phase theory itself.
164
Bibliography
2006. Transformations and derivations. Cambridge: Cambridge University Press.
Albizu, Pablo. 1997. Generalized person-case constraint: A case for a syntax driven
inflectional morphology. Anuario del Seminario de Filolog??a Vasca Julio de Urquijo
(ASJU, International Journal of Basque Linguistics and Philology) XL 1?33.
Amiridze, Nino. 1998. K?oncep?t?ualuri arket?ip?ebi da mati asaxva morposint?akss?i
[conceptual archetypes and their reflection in morphosyntax]. Doctoral Dissertation,
Tbilisi State University.
Amiridze, Nino. 2003. The anaphor agreement effect and Georgian anaphors. In Em-
pirical issues in formal syntax and semantics, ed. Claire Beyssade, Oliver Bonami,
Patricia Cabredo Hofherr, and Francis Corblin, volume 4, 99?114. Paris: Presses de
l?Universite? Paris-Sorbonne.
Amiridze, Nino. 2006. Reflexivization strategies in Georgian. Doctoral Dissertation,
Utrecht.
Anagnostopoulou, Elena. 2003. The syntax of ditransitives: Evidence from clitics, vol-
ume 54. Walter de Gruyter.
Anagnostopoulou, Elena. 2005. Strong and weak person restrictions: A feature checking
analysis. Clitic and affix combinations: Theoretical perspectives 199:199?235.
Anagnostopoulou, Elena. 2006. Clitic doubling. In The Blackwell companion to syntax,
vol. 1, ed. Martin Everaert and Henk van Riemsdijk, 519?581. Oxford: Blackwell.
Anderson, Stephen R. 1984. On representations in morphology case, agreement and
inversion in georgian. Natural Language & Linguistic Theory 2:157?218.
Anderson, Stephen R. 1992. A-morphous morphology, volume 62. Cambridge: Cam-
bridge University Press.
Andre?asson, Daniel. 2001. Active languages. BA Thesis.
Aronson, Howard I. 1970. Towards a semantic analysis of case and subject in georgian.
Lingua 25:291?301.
165
Aronson, Howard I. 1990. Georgian: A reading grammar. Slavica.
Asatiani, Rusudan. 1982. Mart?ivi c?inadadebis t?ip?ologiuri analizi [typological analysis
of a simple sentence]. Tbilisi: Mecniereba.
Asatiani, Rusudan, and Stavros Skopeteas. 2012. Information structure in Georgian. In
The expression of information structure, ed. Manfred Krifka Renate Musan, volume 5,
127?158. Berlin ; Boston: De Gruyter.
Baier, Nico. 2018. Anti-agreement. Doctoral Dissertation, University of California,
Berkeley.
Baker, Mark C. 2011. When agreement is for number and gender but not person. Natural
Language & Linguistic Theory 29.
Begus?, Gas?per. 2020. Segmental phonetics and phonology. In The Oxford Handbook of
Languages of the Caucasus, ed. Maria Polinsky.
Be?jar, Susana. 2003. Phi-syntax: A theory of agreement. Doctoral Dissertation, Univer-
sity of Toronto.
Be?jar, Susana, and Milan Rezac. 2003. Person licensing and the derivation of PCC effects.
Amsterdam Studies in the Theory and History of Linguistic Science Series 4 49?62.
Be?jar, Susana, and Milan Rezac. 2009. Cyclic Agree. Linguistic Inquiry 40:35?73.
Bhatt, Rajesh. 2005. Long distance agreementi n Hindi-Urdu. Natural Language and
Linguistic Theory 23:757?807.
Blix, Hagen. 2016. South Caucasian agreement: A spanning account. Master?s thesis,
University of Vienna.
Blix, Hagen. 2021. Spans in south caucasian agreement: Revisiting the pieces of inflec-
tion. Natural Language & Linguistic Theory 39:1?55.
Bobaljik, Jonathan David. 2008. Where?s phi? agreement as a post-syntactic operation.
In Phi-theory: Phi features across interfaces and modules, 295?328. Oxford University
Press: Oxford.
Boeckx, Cedric. 2008. Understanding minimalist syntax: Lessons from locality in long-
distance dependencies. Malden, MA: Blackwell Publishers.
Boeckx, Cedric, and Kleanthes K. Grohmann. 2007. Remark: Putting phases in perspec-
tive. Syntax 10:204?222.
Boeder, Winfried. 1979. Ergative syntax and morphology in language change: The south
caucasian languages. In Ergativity. towards a theory of grammatical relations, ed.
F Plank, 435?480. London: Academic Press.
166
Bonet, Eulalia. 1991. Morphology after syntax: Pronominal clitics in romance. Doctoral
Dissertation, Massachusetts Institute of Technology.
Bonet, Eulalia. 1994. The person-case constraint: A morphological approach. In The
morphology-syntax connection, ed. Heidi Harley and Colin Phillips, 33?52. MIT.
Borise, Lena. 2019. Phrasing is key: The syntax and prosody of focus in georgian. Doc-
toral Dissertation, Harvard University.
Bos?kovic?, Z?eljko. 2007. On the locality and motivation of Move and Agree: An even
more minimal theory. Linguistic Inquiry 38:589?644.
Bos?kovic?, Z?eljko. 2014. Now i?ma phase, now i?m not a phase: On the variability of
phases with extraction and ellipsis. Linguistic inquiry 45:27?89.
Bos?kovic?, Z?eljko, and Howard Lasnik. 2003. On the distribution of null complementizers.
Linguistic Inquiry 34:527?546.
Buell, Leston. 2005. Issues in Zulu verbal morphosyntax. Doctoral Dissertation, UCLA.
Butskhrikidze, Marika. 2002. The consonant phonotactics of Georgian. Netherlands
Graduate School of Linguistics.
Caha, Pavel. 2009. The Nanosyntax of Case. Doctoral Dissertation, University of Tromso.
Carmack, Stanford. 1997. Blocking in georgian verb morphology. Language 314?338.
Chandra, Pritha. 2007. (dis)agree: Agreement and movement reconsidered. Doctoral
Dissertation, University of Maryland, College Park.
Chikobava, Arnold. 1950. Kartuli enis zogadi daxasiateba [a general characterization of
the georgian language]. Kartuli enis ganmart?ebiti leksik?oni [Explanatory Dictionary
of the Georgian Language] .
Chomsky, Noam. 1964. Current issues in linguistic theory. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Massachusetts:
M.I.T. Press.
Chomsky, Noam. 1973. Conditions on transformations. In A festschrift for Morris Halle,
ed. Stephen R. Anderson and Paul Kiparsky, 232?286. New York: Holt, Rinehart, &
Winston.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht, The Nether-
lands: Foris Publications.
Chomsky, Noam. 1986. Barriers. Cambridge, Massachusetts: MIT Press.
Chomsky, Noam. 1994. Bare phrase structure. MIT Occasional Papers in Linguistics 5,
Cambridge, MA: MITWPL.
167
Chomsky, Noam. 1995. The minimalist program. Cambridge, Massachusetts: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step: Essays on
minimalist syntax in honor of Howard Lasnik, ed. Roger Martin, David Michaels, and
Juan Uriagereka, 89?156. MIT Press.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, ed. Michael
Kenstowicz, 1?52. Cambridge, MA: MIT Press.
Chomsky, Noam. 2004. Beyond explanatory adequacy. In Structures and beyond: The
cartography of syntactic structures, ed. Adriana Belletti, volume 3, 104?131. Oxford:
Oxford University Press.
Chomsky, Noam. 2008. On phases. Current Studies in Linguistics Series 45:133.
Chung, Sandra. 1998. The design of agreement: Evidence from chamorro. University of
Chicago Press.
Citko, Barbara. 2014. Phase theory: An introduction. Cambridge University Press.
Clem, Emily. 2018. Cyclic expansion in Agree: Maximal projections as probes. Paper
presented at NELS 49.
Comrie, Bernard. 1973. The ergative: variations on a theme. Lingua 32:239?253.
Coon, Jessica, and Stefan Keine. 2020. Feature gluttony. Linguistic Inquiry 1?56.
Davison, Alice Louise. 2010. Long distance agreement and restructuring. Ms., University
of Iowa.
Dayal, Veneeta. 1996. Locality in wh-quantification. Dordrecht: Kluwer.
Deal, Amy Rose. 2015. Interaction and satisfaction in ?-agreement. In Proceedings of
NELS, volume 45.
den Dikken, Marcel. 2007. Phase extension contours of a theory of the role of head
movement in phrasal extraction. Theoretical Linguistics 33:1?41.
den Dikken, Marcel. 2009a. Arguments for successive-cyclic movement through SpecCP.
Linguistic Variation Yearbook 9:89?126.
den Dikken, Marcel. 2009b. On the nature and distribution of successive cyclicity. Paper
presented at NELS 40, MIT.
den Dikken, Marcel. 2012. On the strategies for forming long A-dependencies: Evidence
from Hungarian. Ms.
Doliana, Aaron. 2014. The super-strong person-case constraint: Scarcity of resources by
scale-driven impoverishment. In Proceedings of the 22nd conference of the student or-
ganization of linguistics in europe (console 22), ed. Eleanor Dutton Martin Kohlberger,
Kate Bellamy, 58?80. Leiden: SOLE.
168
Drummond, Alex, Norbert Hornstein, and Howard Lasnik. 2010. A puzzle about P-
stranding and a possible solution. Linguistic Inquiry 41:689?692.
Epstein, Samuel David, Erich M Groat, Ruriko Kawashima, and Hisatsugu Kitahara.
1998. A derivational approach to syntactic relations. Oxford University Press.
Felser, Claudia. 2004. Wh-copying, phases, and successive cyclicity. Lingua 114:543?
574.
Foley, Steven. 2013. The syntax of Georgian relative clauses. NYU Senior honors thesis.
Foley, Steven. 2017. Morphological conspiracies in Georgian and optimal vocabulary
insertion. In Proceedings of CLS 52, ed. Jessica Kantarovic, Tran Truong, and Orest
Xherija. Chicago, IL: Chicago Linguistics Society.
Fox, Danny. 1999. Reconstruction, binding theory, and the interpretation of chains. Lin-
guistic Inquiry 30:157?196.
Fox, Danny, and David Pesetsky. 2004. Cyclic linearization of syntactic structure. Theo-
retical Linguistics 31:1?46.
Frampton, John, and Sam Gutmann. 2000. Agreement is feature sharing. Ms., Northeast-
ern University .
Gallego, A?ngel J. 2006. Phase sliding. Manuscript.
Gallego, A?ngel J. 2010. Phase theory, volume 152. John Benjamins Publishing.
Graf, Thomas. 2012. An algebraic perspective on the person case constraint. In Theories
of everything. in honor of ed keenan (ucla working papers in linguistics 17), ed. Anna
Szabolcsi Jos Tellings Thomas Graf, Denis Paperno, 85?90.
Grohmann, Kleanthes K. 2003a. Prolific domains: On the anti-locality of movement
dependencies. John Benjamins Publishing.
Grohmann, Kleanthes K. 2003b. Transfer vs. spell-out and the road to pf. Linguistic
Analysis 33:176?194.
Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of in-
flection. In The View from Building 20, ed. Kenneth Hale and Samuel Jay Keyser.
Cambridge, MA: MIT Press.
Halpert, Claire. 2015. Argument licensing and agreement. Oxford University Press.
Halpert, Claire. 2019. Raising, unphased. Natural Language & Linguistic Theory 37:123?
165.
Harbour, Daniel. 2016. Impossible Persons. Cambridge, MA: MIT Press.
Harley, Heidi, and Elizabeth Ritter. 2002. Person and number in pronouns: A feature-
geometric analysis. Language 78:482?526.
169
Harris, Alice C. 1981. Georgian syntax: A study in relational grammar. Cambridge
University Press.
Harris, Alice C. 1982. Georgian and the unaccusative hypothesis. Language 58:290?306.
Harris, Alice C. 1984. Inversion as a rule of universal grammar: Georgian evidence. In
Studies in relational grammar 2, 259?91. University of Chicago Press Chicago.
Harris, Alice C. 1985. Diachronic syntax: The Kartvelian case. New York: Academic
Press.
Haspelmath, Martin. 2004. Explaining the distransitive person-role constraint. Construc-
tions .
Hewitt, Brian George. 1987. The typology of subordination in Georgian and Abkhaz,
volume 5. Berlin; New York: de Gruyter.
Hewitt, Brian George. 1995. Georgian: A structural reference grammar. John Benjamins
Publishing.
Hiraiwa, Ken. 2001. Multiple agree and the defective intervention constraint in Japanese.
MIT working papers in linguistics 40:67?80.
Kalin, Laura. 2017. Dropping the f-bomb: An argument for valued features as derivational
time-bombs. In Proceedings of the 47th annual meeting of the North East Linguistic
Society, 119?132.
Kavtaradze, Ivane. 1954. Zmnis dziritadi k?at?egoriebis ist?oriistvis dzvel kartuls?i (on the
history of the basic verbal categories in old georgian). Tbilisi: Sakartvelos SSR Mecn.
Ak?ad. Gamomcemloba.
Kayne, Richard S. 1999. Prepositional complementizers as attractors. Probus 11:39?73.
Kayne, Richard S. 2004. Prepositions as probes. In Structures and beyond: The cartog-
raphy of syntactic structures, ed. Adriana Belletti, volume 3, 192?212. Oxford: Oxford
University Press.
Keine, Stefan. 2017. Agreement and vP phases. A schrift to fest Kyle Johnson 177?185.
Keine, Stefan, and Bhamati Dash. 2018. The ups and downs of agreement. Ms., Univer-
sity of Southern California.
King, Tracy Holloway. 1994. SpecAgrP and case: Evidence from Georgian. In The
morphology-syntax connection, ed. Heidi Harley and Colin Phillips, 91?110. MIT.
Klimov, Georgij A. 1973. Oc?erk obs?c?ej teorii e?rgativnosti [an essay on the general theory
of ergativity]. Moscow: Nauka.
Klimov, Georgij A. 1977. Tipologija jazykov aktivnogo stroja [typology of active-
alignment languages]. Moscow: Nauka.
170
Klimov, Georgij A., and Shota Dzidziguri. 1979. O donominativnom komponente
kartvel?skoj jazykovoj struktury [on pre-nominative component of kartvelian language
structure]. In Collection of papers in linguistics (dedicated to the memory of g.
machavariani), 130?145. Tbilisi: Mecniereba.
Kramer, Ruth. 2015. The morphosyntax of gender. Oxford: Oxford University Press.
Kratzer, Angelika. 2009. Making a pronoun: Fake indexicals as windows into the prop-
erties of pronouns. Linguistic Inquiry 40:187?237.
Lahne, Antje. 2008. Where there is fire there is smoke: Local modelling of successive-
cyclic movement. Doctoral Dissertation, University of Leipzig.
Laka, Itziar. 1993. The structure of inflection: A case study in x0. Generative studies in
Basque linguistics 105:21.
Lasnik, Howard, and Mamoru Saito. 1994. Move ?: Conditions on its application and
output. MIT press.
Lebeaux, David. 1990. Relative clauses, licensing, and the nature of the derivation. In
Proceedings of the North East Linguistic Society 20, ed. Juli Carter, Dechaine Rose-
Marie, Bill Philip, and Tim Sherer, 318?332. University of Massachusetts at Amherst:
Graduate Linguistic Student Association.
Legate, Julie Anne. 2003. Some interface properties of the phase. Linguistic Inquiry
34:506?516.
Legate, Julie Anne. 2008. Morphological and abstract case. Linguistic inquiry 39:55?101.
Legate, Julie Anne. 2012. The size of Phases. In Phases, ed. A?ngel J Gallego, 233?250.
Berlin: de Gruyter.
Lomashvili, Leila. 2011. Complex predicates: The syntax-morphology interface, volume
174. John Benjamins Publishing.
Lomashvili, Leila, and Heidi Harley. 2011. Phases and templates in Georgian agreement.
Studia Linguistica 65:233?267.
Makharoblidze, Tamar. 2012. The georgian verb. Lincom Europa.
Marantz, Alec. 1989. Relations and configurations in georgian. Ms., University of North
Carolina, Chapel Hill.
Matushansky, Ora. 2005. Going through a phase. In Perspectives on phases, ed. Martha
McGinnis and Norvin Richards, 157?181. Cambridge, MA: MIT Working Papers In
Linguistics.
McCarthy, John J, and Alan Prince. 1995. Faithfulness and reduplicative identity. Lin-
guistics Department Faculty Publication Series 10.
171
McCloskey, James. 2000. Quantifier float and wh-movement in an Irish English. Linguis-
tic Inquiry 31:57?84.
McCloskey, James. 2002. Resumption, successive cyclicity, and the locality of operations.
In Derivation and explanation in the minimalist program, ed. Samuel David Epstein
and T. Daniel Seely, 184?226. Malden, Massachusetts: Blackwell Publishing.
McCloskey, James. 2006. Questions and questioning in a local English. In Crosslinguistic
research in syntax and semantics. negation, tense and clausal architecture, ed. Rafaella
Zanuttini, He?ctor Campos, Elena Herburger, and Paul H. Portner, 87?126. Georgetown
University Press.
McGinnis, Martha. 1995. Projection and position: Evidence from Georgian. In Pro-
ceedings of ConSole IV , ed. & Reuben van der Vijver Joa?o Costa, Rob Goedemans,
203?220. Leiden.
McGinnis, Martha. 1997. Case and locality in L-syntax: Evidence from Georgian. In
The UPenn/MIT Roundtable on Argument Structure, ed. Heidi Harley. MIT Working
Papers in Linguistics.
McGinnis, Martha. 2001. Phases and the syntax of applicatives. In North East Linguistic
Society, ed. Minjoo Kim and Uri Strauss, 333?349. Georgetown University: GLSA.
McGinnis, Martha. 2005. On markedness asymmetries in person and number. Language
81:699?718.
McGinnis, Martha. 2013. Agree and fission in Georgian plurals. Distributed Morphology
Today: Morphemes for Morris Halle 38?58.
Melikishvili, Damana. 1998. zmnur pormata i, ii da iii seriis urtiertmimartebisatvis
kartvelur enebs?i. nominat?iur da ergat?iul (akt?iur) k?onst?rukciata urtiertmimarte-
bisatvis [on the interaction of the verb forms of the series i, ii and iii in kartvelian
languages. on the interaction of the nominative and ergative (active) constructions]. In
Works of the chair of the georgian language of the tbilisi state pedagogical university,
volume 4, 127?144. Tbilisi: Tbilisi State Pedagogical University Press.
Melikishvili, Damana. 2001. kartuli zmnis ughlebis sist.ema [the conjugation system of
georgian verb]. Tbilisi: Logos Presi.
Miyagawa, Shigeru. 2011. Genitive subjects in Altaic and specification of phase. Lingua
121:1265?1282.
Mu?ller, Gereon. 2011. Constraints on displacement. Benjamins, Amsterdam .
Nash, Le?a. 1994. On BE and HAVE in Georgian. In MIT Working Papers in Linguistics
22, ed. Heidi Harley and Colin Phillips, 153?171. Cambridge, MA: The MIT press.
Nash, Le?a. 1995. Porte?e argumentale et marquage casuel dans les langues SOV et dans
les langues ergatives: l?exemple du ge?orgien. Doctoral Dissertation, Paris 8.
172
Nash, Le?a. 2017. The structural source of split ergativity and ergative case in Georgian. In
The Oxford Handbook of Ergativity, ed. Jessica Coon, Diane Massam, and Lisa Travis,
175?204. Oxford: Oxford University Press.
Nash, Le?a, and Alain Rouvert. 1997. Proxy categories in phrase-structure theory. In
Proceedings of North East Linguistic Society, ed. Kiyomi Kusumoto, 287?304. McGill
University: Graduate Linguistic Student Association.
Nevins, Andrew. 2007. The representation of third person and it consequences for person-
case effects. Natural Language and Linguistic Theory 25:273?313.
Nevins, Andrew. 2011. Multiple agree with clitics: Person complementarity vs. omnivo-
rous number. Natural Language & Linguistic Theory 29:939?971.
Pantcheva, Marina. 2011. The Nanosyntax of directional expressions: Decomposing path.
Doctoral Dissertation, University of Tromso.
Polinsky, Maria, and Eric Potsdam. 2001. Long-distance agreement and topic in Tsez.
Natural Language and Linguistic Theory 19:583?646.
Preminger, Omer. 2011a. Agreement as a fallible operation. Doctoral Dissertation, Mas-
sachusetts Institute of Technology.
Preminger, Omer. 2011b. Asymmetries between person and number in syntax: A com-
mentary on Baker?s SCOPA. Natural Language & Linguistic Theory 29:917?937.
Preminger, Omer. 2014. Agreement and its failures, volume 68. MIT Press.
Preminger, Omer. 2019. What the PCC tells us about ?abstract? agreement, head move-
ment, and locality. Glossa .
Rackowski, Andrea, and Norvin Richards. 2005. Phase edge and extraction: A Tagalog
case study. Linguistic Inquiry 36:565?599.
Rezac, Milan. 2008. The syntax of eccentric agreement: The person case constraint and
absolutive displacement in Basque. Natural Language & Linguistic Theory 26:61?106.
Rezac, Milan. 2009. On the unifiability of repairs for the person case constraint: French,
Basque, Georgian and Chinook. Anuario del Seminario de Filolog??a Vasca? Julio de
Urquijo? 43:769?789.
Rezac, Milan. 2011. Phi-features and the modular architecture of language. Dordrecht:
Springer.
Richards, Marc. 2011. Deriving the Edge: What?s in a Phase? Syntax 14:74?95.
Richards, Norvin. 1998. The principle of minimal compliance. Linguistic Inquiry 29:599?
629.
173
Richards, Norvin. 1999. Featural cyclicity and the ordering of multiple specifiers. In
Working minimalism, ed. Samuel Epstein and Norbert Hornstein, 127?158. Cambridge,
Massachusetts: MIT Press.
Richards, Norvin. 2001. Movement in language: interactions and architectures. Oxford:
Oxford University Press.
Richards, Norvin W., III. 1997. What moves where when in which language? Doctoral
Dissertation, Massachusetts Institute of Technology, Cambridge, Massachusetts.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, Massachusetts: MIT Press.
Rizzi, Luigi. 2001. Relativized minimality effects. In The handbook of contemporary
syntactic theory, ed. Mark Baltin and Chris Collins. Oxford: Blackwell.
Shanidze, Akaki. 1973. Kartuli enis grammat?ik?is sapudzvlebi [fundamentals of the
grammar of the georgian language], volume 1. Tbilisi: Tbilisis Universit?et?is
Gamomcemloba, second edition edition.
Sheehan, Theresa Biberauer Ian Roberts, Michelle, and Anders Holmberg. 2017. The
final-over-final condition: A syntactic universal, volume 76. Cambridge, MA: MIT
Press.
Shlonsky, Ur. 1989. The hierarchical representation of Subject Verb Agreement. Ms.
SigurDsson, Haldo?r A?rmann. 2002. To be an oblique subject: Russian vs. Icelandic.
Natural Language and Linguistic Theory 20:691?724.
SigurDsson, Halldo?r A?rmann. 1996. Icelandic finite verb agreement. Working Papers in
Scandinavian Syntax 57:1?46.
SigurDsson, Halldo?r A?rmann, and Anders Holmberg. 2008. Icelandic dative intervention:
Person and number are separate probes. Agreement restrictions 251?280.
Smolensky, Paul, and A Prince. 1993. Optimality theory: Constraint interaction in gener-
ative grammar. Optimality Theory in phonology 3.
Starke, Michael. 2009. Nanosyntax: A short primer to a new approach to language.
Nordlyd 36:1?6.
Starke, Michael. 2011. Towards elegant parameters: Language variation reduces to the
size of lexically stored trees. LingBuzz/001183.
Taraldsen, Knut Tarald. 1996. Reflexives, pronouns and subject/verb agreement in ice-
landic and faroese. Microparametric syntax and dialect variation 189?212.
Taraldsen, Tarald. 1995. On agreement and nominative objects in Icelandic. In Studies
in comparative Germanic syntax, ed. Hubert Haider, Susan Olsen, and Sten Vikner,
307?327. Dordrecht: The Netherlands: Kluwer Academic Publishers.
174
Tuite, Kevin. 1999. The myth of the caucasian sprachbund: The case of ergativity. Lingua
108:1?29.
Uriagereka, Juan. 1999. Multiple spell-out. In Working minimalism, ed. Samuel Epstein
and Norbert Hornstein, 251?282. Cambridge, Massachusetts: MIT Press.
van Urk, Coppe, and Norvin Richards. 2015. Two components of long-distance extrac-
tion: Successive cyclicity in Dinka. Linguistic Inquiry 46:113?155.
Wurmbrand, Susi. 2017. Feature sharing or how I value my son. In A pesky set: Papers for
David Pesetsky,, ed. Hadas Kotek Coppe van Urk Claire Halpert, 173?182. Cambridge,
MA: MITWPL.
175