OVERVIEW OF THE TALK
CAN CLOSEST CONJUNCT
AGREEMENT BE DERIVED IN
THE SYNTAX PROPER?
Lanko Marušič, Jana Willer-­‐Gold, Boban Arsenijević, Andrew Nevins
Based on the EMSS project, funded by the Leverhulme Trust
#emssproject
• Closest-Conjunct Agreement
Experimental Evidence
(CCA) in South Slavic:
• Modeling
the basic patterns (closest-conjunct agreement and
highest conjunct agreement)
• Distributed
vs purely syntactic approaches to CCA
• Further
constraining the theory: sandwiched agreement (two
probes, one on each side)
1
2
IS THE MAN WHO IS TALL HAPPY?
METHODOLOGY:
ELICITED PRODUCTION
• Chomsky
1975: Syntax refers to hierarchical, not linear order,
when it comes to auxiliary movement; learners prefer
hierarchically-based generalisations
1983: South Slavic languages show cases of agreement
based on linear order: when two noun-phrases are conjoined, the
verb can sometimes agree with the linearly closest one
Participant sees model sentence
Prevod
je
ovjeren
pečatom.
translation.M.SG. AUX.SG authenticated.M.SG. by.seal
• Corbett
• Bock
& Miller 1991: English speakers show cases of agreement
based on linear order, called ‘attraction’, with the plural
complement of noun phrases (e.g. the key to the cabinets are)
Participant sees replacement noun phrase (&P) displayed onscreen
Molbe
i
rješenja
requests.F.PL. and decisions.N.PL.
Participant produces new sentence aloud and then clicks mouse
Molbe
i
rješenja
su
ovjeren-i/-a/-e pečatom
requests.F.PL. and decisions.N.PL. AUX.PL
• How
experimentally robust is linear conjunct agreement in South
Slavic morphosyntax? Is it distinct from attraction?
3
Responses recorded, classified, tabulated
4
IP
NP2
VP
NP1
NP1
CP
C
METHODS AND MATERIALS
•
Nine &P combinations of two genders; 6 items per condition
•
All NPs inanimate and plural; dependent variable = gender agreement on participle
•
54 distractor items (18 RCs, 18 QNPs, 18 epicene nouns)
•
Exp 1a: SV configurations (preverbal subjects), n=30 * 6 sites
•
Exp 1b: VS configurations (postverbal subjects), n=30 * 6 sites
•
Design and methodology identical across sites with local adaptation of vocabulary and
morphosyntax
•
Participants 18-22, not linguistics students; native speakers who grew up in region tested
VP
IP
GIVEN A HIERARCHICAL &P
IP
&P
NP2
NP1
VP
&
SV
VS
Linearly closest conjunct
NP1
VP
…
…
V
V
&’P
&
NP2
Linearly closest conjunct
VP
&P
V
&’P
&P
NP1
&’P
&
NP2
Distal conjunct still hierarchically higher
NP2
Distal conjunct hierarchically lowest
VP
V
1
&P
NP1
&’P
&
NP2
5
6
CHOOSING THE
AGREEMENT STRATEGY
LINEAR AGREEMENT IS
ROBUST GIVEN &PS (FN/NF)
• The ‘default’ or ‘resolution’ value
masculine
1
Results for SV condition (where first and closer diverge)
for conjuncts with mixed genders is
• When
a &P has M+F, N+M, etc, and we see masculine agreement,
we can’t tell if it’s default agreement or closest-conjunct agreement
the combinations N+F and F+N, there are three
distinct options: first-conjunct agreement (hierarchical), closestconjunct agreement (linear), and default agreement (resolution/
prescriptive)
61%
Highest
Closest
44%
• However, in
• (In VS
contexts, the first-conjunct is the closest conjunct)
7
43%
40%
39%
42%
22%
16%
17%
11%
11%
Ljubljana
Niš
8%
Novi Sad
Sarajevo
8
Zadar
Zagreb
Willer-Gold et al
(in prep)
&P
NP1
V
IP
&PS AREN’T JUST FLAT:
DISTAL CONJUNCT AGREEMENT
IF &PS WERE FLAT…
&IP NPNP
22
VP
VS
NP1
&P
NP2
V…
NP1
VP
&
NP2
VP
V
NP1
NP1
NP2
BUT IF HIERARCHICAL…
VP
V
&P
&’P
&
&
Distal conjunct “three nodes away”
…V
&P
&
&P
&
NP1NP2
Distal conjunct “three nodes away”
VP
V
&’P
5
…
IP V
&P
If &Ps were just flat, then the distal conjunct in both SV and VS
conditions should be equally produced/judged
VPVP
IP
% of distal productions
C
Linearly closest conjunct preferred
…
&’P
&
NP2
Distal conjunct still hierarchically higher
*
&P
NP1
NP2
*
0%
VP
4
Linearly closest
conjunct
preferred
NP1
CP
rating of distal agreement
SV
NP1
3
NP1
2
&P
C
1
NP2
VP
CP
60%
&
NP1
40%
&’P
20%
NP1
SV
NP2
VS
Willer-Gold et al
(in prep)
SV
VS
&Ps have internal hierarchical structure
Distal conjunct hierarchically lowest
9
10
VP
V
1
&P
THEORIES OF
CONJUNCT AGREEMENT:
MARUSIC ET AL (2007/2015)
SO WHERE CAN LINEARITY
TRUMP HIERARCHY?
NP1
&’P
&
NP2
• Key
• In
one principled corner of the grammar, in one small corner
1
of the world
intense research focus is how to model the variability in
agreement strategies within a restricted model of possibilities
(Boskovic 2009, Puskar & Murphy 2014, Marusic et al
2007/2015)
• Two-step Agree: Agree-Link
• An
can be grouped into purely syntactic approaches to
CCA versus distributed approaches to CCA
• These
11
idea: ConjP doesn’t have its own gender
vs Agree-Copy
•A
Probe may establish a relation with a Goal (e.g. Pesetsky &
Torrego 2007), but the actual copying takes place postsyntactically (see also Bhatt & Walkow 2013)
2
• Participles
enact Agree-Link with ConjP, but depending on
how and when Agree-Copy takes place, different parts of
ConjP may be targeted
12
2
MARUSIC ET AL (2007/2015):
AGREE-COPY CHOICE
• ConjP
has its own Number, but not its own Gender
• Speakers
who wish to avoid default Gender must therefore
open up the lid on the ConjP and choose one of the
individual conjuncts during Agree-Copy
• Key
BHATT & WALKOW 2013:
DISTRIBUTED AGREEMENT
• In
Hindi/Urdu, while conjoined subjects show resolved
agreement, conjoined objects show CCA
– Closest
Conjunct
an the
Illusion
• AgreementMurphy
must &bePuškar
partly
syntactic
to Agreement
account isfor
Murphy & Puškar(repeated
– Closest
Conjunct
Agreement
is
an Illusion
Murphy
&
Puškar
– Closest
Agreement
Illusion
below)
thatConjunct
LCA
is onlyis anacceptable
• If Agree-Copy
happens before linearization, HCA results
• If Agree-Copy
happens after linearization, CCA results
• (ThereMurphy
is no
way
to Conjunct
choose
a medial
& Puškar
– Closest
Agreement
is an Illusion conjunct
(17)
(repeated below) that LCA is only acceptable when the conjunct phrase is in preverbal position.
It is entirely ungrammatical if the &P is postverbal.
It is entirely
ungrammatical
theSve
&P ishaljine
postverbal.
syntax,
PF
mechanisms
step ini to sva
furnish
agreement,
and
these,.
(54) (54) [if&P
odela]
su. juče
prodata
[ Sve haljine i sva odela] su juče
prodata
all may
dress.fpl
and
suit.npl
are
sell.prt.npl
(54) [post-syntactic,
Sve haljine i sva odela]
su be
jučeall sensitive
prodata
.to
all
dress.fpl
and
allyesterday
suit.npl
are yesterday
sell.prt.npl
being
linear
order
‘All dresses and all suits were sold yesterday.’
&P
if there 8are 3)
all dress.fpl and all suit.npl are yesterday sell.prt.npl
‘Allwere
dresses
and all suits were sold yesterday.’
(55)
su yesterday.
prodata ’ [&P sve haljine i sva odela].
‘All dresses and all
suits*Juče
sold
(55)
8
An important starting assumption is that the participle is a single ϕ-probe, which probes for number and gender features of the noun together (following Bejar 2003). Another assumption is that
(17)
Juče
su items
prodata
[&P sva odelaas valued/unvalued
i sve haljine]. and interpretable/uninterpretable
features
on lexical
are characterised
Yesterday are sell.prt.npl
suits.npl and all dresses.fpl
(2007).
The process of Last Conjunct Agreement (17) proposed
in the spirit
of Pesetsky & Torrego all
‘All suits and all dresses were sold yesterday.’
in this account proceeds following the steps in (18) – (21).
Step
1: The probe
establishes
a Match
relation
withis &P
for number
NP1
for gender
An
important
starting
assumption
is that the
participle
a single
ϕ-probe, and
which
probes
for num-(it enters
ber
and
gender features
of the noun
together
(following
Bejar2007).
2003). Another assumption is that
into
Multiple
Agree; Hiraiwa
2001,
Pesetsky
& Torrego
THEORIES OF CONJUNCT
AGREEMENT: BOSKOVIC (2009)
features on lexical items are characterised as valued/unvalued and interpretable/uninterpretable
[PrtP
Participle
... The process of
[&P
NP2gender ]]
uϕ: (2007).
number
gender
& Torrego
Last
ConjunctNP1
Agreement
(17)&proposed
in(18)
the spirit
of Pesetsky
in this account proceeds following the steps in (18) – (21).
Step 1: The probe establishes a Match relation with &P for number and NP1 for gender (it enters
StepMultiple
2: If theAgree;
probeHiraiwa
has an EPP
of the subject is required. Valuators undergo
into
2001,feature,
Pesetskypied-piping
& Torrego 2007).
Multiple Agree
pied-piping, and here pied-piping of the subject fails due to ambiguity of the target for movement
(18)
Participle
... as Serbo-Croatian
[&P number
NP2gender ]]
uϕ:moved
(either[PrtP
&P or NP1
can be
allowsNP1
for gender
violations&of Coordinate
Structure
an entire conjunct phrase which has only inherited the features of one of its conjuncts, in
The central claim of thiswith
paper
is that the nature of ‘closest’ is in Closest Conjunct Agreement
this case, the final NP. In order to have an &P that has only inherited the features of the second
entirely illusory. What we
in fact
have
the case
of LCA,
example,( ↑Agr↑)
is actually
agreement
conjunct,
theclaim
Agreein
operation
targetting
the for
first
will have
toof
fail ‘closest’
to apply. We is in Closest Conjunct Agreemen
The central
of this
paper
isconjunct
that↑Agr↑
theapplies
with an entire conjunct
phrase
which
has only
inherited
the features
one
ofnature
its conjuncts,
can achieve this with
an order
of operations
in which of
too early, i.e.inbefore the
been
The
relevant
in (56).
this case, the finalentirely
NP. Inconjuncts
order
tohave
have
anmerged.
&P that
has
only
inherited
the
features
the second
illusory.
What
we
inorder
factis given
have
in theof case
of LCA, for example, is actually agreemen
conjunct, the Agree operation
targetting
the first
conjunct
( ↑Agr↑) will have to fail to apply. We
(Move) >conjunct
↑Agr↑
> Merge
> ↓Agr↓
with (56)
an entire
phrase
which has only inherited the features of one of its conjuncts, i
can achieve this with an order of operations in which ↑Agr↑ applies too early, i.e. before the
Here,
Merge
applies
after
↑Agr↑,
an
instance
counterfeeding.
Thethat
derivation
thismerged.
case,The
therelevant
finalorder
NP.isIn
order
toof have
an &P
hasproceeds
onlyasinherited the features of the secon
conjuncts have been
given
in (56).
PUSKAR & MURPHY 2015:
have to fail to apply. W
can
achieve
this
with
an
order
of
operations
in
which
↑Agr↑
applies
too
early, i.e. before th
DERIVING
CCA SYNTACTICALLY
conjuncts have been merged. The relevant order is given in (56).
follows: Move does not apply. ↑Agr↑ applies before Merge. Since there is no goal that this
(56)
Constraint according to the author) and they are assumed to count as equidistant.
if Probe has EPP: Pied-Piping ambiguity; can’t move either
(57)
Since NP2 now as a valuator cannot be extracted, the only option is to move the whole &P to
subject position, which results in the LCA pattern.
• (In
to fact
′
(59)
N
N
(60)
Move > ↑Agr↑ > (Merge) > ↓Agr↓:operation ↑Agr↑ and Agree will find a goal in T’s specifier (60):
&′
(57)
↑Agr↑:
&
NP
&P
[gender: !] &P N
suits N
➊
↓Agr↓:
(58)
Merge:suits.
to the target of EPP-driven movement, the gender features on the verb can be provided by NP1
(60)
Move > ↑Agr↑ > (Merge) > ↓Agr↓:
TP
T′
➋
N
&P N
T
are
[ϕ:!]
vP
NP
VP
v
v
&
dresses.F
sold
tV
T
Move
feeds
T
vP
upward
Agree;
NP
are
v with
VP &P
T agrees
[ϕ:!]
suits.
v sold t
t
(and hence
NP2)
′
&′
t&P
➋
N
&P
addition, for 3-conjuncts, predicts medial conjunct agreement…)
15
TP
&&′
[gender:!]
&
NP
dresses.
F
NP
The derivation
ofTFirst[PrtP
Conjunct
Agreement
is essentially
same&process,
the only
(21)
[TP
Participle
.(17)
. . [&Pnumber
NP1the
NP2gender
]]] differuϕ:
gender
ence being that the pied-piping dilemma in (19) does not arise, as the probe does not have an
EPP feature in this derivation. Since there is no conflict between the two valuators with respect
The
derivation
of EPP-driven
First Conjunct
Agreementthe
(17)gender
is essentially
theon
same
thebeonly
differ- by NP1
to the
target of
movement,
features
theprocess,
verb can
provided
ence being that the pied-piping dilemma in (19) does not arise, as the probe does not have an
and the conjunct phrase can stay in situ, thereby yielding postverbal FCA. Thus, the availability
EPP feature in this derivation. Since there is no conflict between the two valuators with respect
of LCA or FCA is tied to the presence of an EPP feature on the probe. For LCA, the EPP feature
F
F
dresses F
gender
F
N
NP
Step 4: Since the probe bears an EPP feature, movement of the &P or one of the NPs is required.
(20)
[
Participle
.move;
.[
NP1 option is &to
NP2
]]] &P
This
&Pcannot
canbe. extracted,
preverbal
CCA
Sincetime,
NP
now[ only
as a valuator
theyielding
only
move
the whole
to
Step 4: Since
the probe bears
an EPP feature, movement
of the &P
one of the NPs is required.
• Disadvantage:
beor impossible,
(21)
[predicts
T preverbal
[PrtP
ParticipleHCA
. . . [to
NP1 gender
& counter
NP2gender ]]]
TP
uϕ:
&Pnumber
Here, Merge
applies
after ↑Agr↑, an instance of counterfeeding. The derivation proceeds a
NP
&
&P
dresses Move
follows:
not apply. ↑Agr↑
applies
before Merge. Since there is no goal that th
& does NP
NP
&
Downward Agree:
[gender: n]
NP
&
operation
can
target,
it
does
not
find
a
goal
(57)
and thus applies vacuously. The next operatio
suits
dresses
&
&
NP
&P
valued by NP2
dresses
&
NP
[gender: n]
[gender:!]
Merge introduces
the
conjuncts
via External
Merge
(58). Finally, ↓Agr↓ applies and the &
suits
[gender:
!]operations must be maintained. There are
At the TP level, the previous
order of
again two possible
Merge: scenarios depending on whether Movesuits
or not.
Move the
takes place,
it willconjunct
feed the next (59). As a result, the &P node bea
head
receives
the genderapplies
value
ofIfonly
lowest
&P (59)
thespecifier
TP level, the
previous order of operations must be maintained. There are again two possible
↓Agr↓:
operation
↑Agr↑
and Agree will find a goal inAtT’s
(60):
scenarios
depending
whether Move applies or not. If Move takes place, it will feed the next
the
featuresorder
of only
the second
conjunct.
Uniform
of operations
within
aonderivation:
′
➊
Solution: do Agree again, this time with NP2:
gender
27
↑Agr↑:
&P
Step 3: To prevent a crash due to lack of valuation, another cycle of Agree is instantiated, resulting
in(20)
targeting
NP2, because
NP1 uϕ:
was deactivated
as goal afterNP1
thegender
first Agree &
cycle (i.e.
its gender ]]]
[TP&P and
[PrtP
Participle
. . . [&Pnumber
NP2
uninterpretable gender feature was deleted after Match, unlike the number feature of the &P,
which is interpretable, and therefore not deleted after Match).
&Pnumber
Merge introduces the conjuncts via External Merge (58). Finally, ↓Agr↓ applies and the &-
head↑Agr↑,
receives the
value
only the lowest conjunct
(59). As a result,
the &Pas
node bears
Here, Merge applies after
an gender
instance
ofofcounterfeeding.
The derivation
proceeds
the features of only the second conjunct.
follows: Move does
not&apply.
Merge.
Since there is no goal that this 27
Murphy
Puškar –↑Agr↑
Closestapplies
Conjunctbefore
Agreement
is an Illusion
(57)
↑Agr↑:
operation can target, it does not find a goal (57) and thus applies vacuously. The next operation
&P
Merge introduces
the conjuncts
via External
Merge (58).
Finally,
applies and the &(56)
(Move)
> ↑Agr↑
> Murphy
Merge
>↓Agr↓
↓Agr↓
& Puškar
– Closest Conjunct Agreement is an Illusion
&P&N
➊
head receives the gender value
of only
the lowest conjunct (59). As a result, the &P node bears
[gender:!]
the features of only the second
conjunct.
(58) Merge:
′
(58)
in targeting
NP2, because
was. .deactivated
goal
after the& first NP2
Agree
cycle (i.e. its
(19)
[TP &P and
[PrtP
ParticipleNP1
. [&Pnumber asNP1
uϕ:
gender
gender ]]]
uninterpretable gender feature was deleted after Match, unlike the number feature of the &P,
which is interpretable, and therefore not deleted after Match).
subject position, which results in the LCA pattern.
the
Agree
operation
targetting
the
first conjunct
( ↑Agr↑) will
can>target,
it does
not find a goal (57)
and thus applies
vacuously.
The next operation
(Move) >conjunct,
↑Agr↑operation
> Merge
↓Agr↓
&P N
Step 2: If the probe has an EPP feature, pied-piping of the subject is required. Valuators undergo
(19)
[TP
[PrtP
Participleuϕ:
. . . [&Pnumber
NP1 gender
&
NP2gender ]]]
pied-piping, and here pied-piping of the subject fails due to ambiguity of the target for movement
(either &P or NP1 can be moved as Serbo-Croatian allows for violations of Coordinate Structure
Constraint
the due
author)
and of
they
are assumed
to count
Step 3: To according
prevent a to
crash
to lack
valuation,
another
cycleasofequidistant.
Agree is instantiated, resulting
uϕ:
all dress.fpl and all suit.npl
‘All dresses and all14
suits were sold yesterday.’
Murphy & Puškar – Closest Conjunct Agreement is an Illusion
PrtP
yesterday are sell.prt.npl
*Juče
su prodata
haljine
i sold
sva yesterday.
odela].[&P
(55)
*Juče
su
prodata
sve haljine i sva odela].
&P sveand
‘All [dresses
all suits
were
’
yesterday are sell.prt.npl
all dress.fpl and all suit.npl
yesterday
are
sell.prt.npl
all isdress.fpl
andAgreement
all suit.npl
The central
claim of this
paper
is that the nature of ‘closest’
in Closest Conjunct
‘All dresses and all suits were sold yesterday.’
entirely illusory. What we in fact have in the case of LCA, for example, is actually agreement
13
2
26 conjunct
26
when the
phrase is in preverbal positio
is entirely
ungrammatical
if the
&Paccessible
is postverbal.within the
• Idea: when Itobject
phi-features
are
not
(repeated below) that LCA is only acceptable when the conjunct phrase is in preverbal position.
&P
Juče
su prodata
[&P sva odela
i sve haljine].
Yesterday are sell.prt.npl
all suits.npl and all dresses.fpl
‘All suits and all dresses were sold yesterday.’
TP
2
subject/object asymmetry
idea: linearization is post-syntactic
➊
V
&P
➊
If Move does not apply, thereby leaving the &P in ′its postverbal base position, then ↑Agr↑ will
If Move
apply, thereby leaving the &P in its postverbal base position, then ↑Agr↑ will
& doesbynotMove).
probe upwards butNP
not find a goal (i.e. it will be counterfed
Following our assumptions
probe upwards but not find a goal (i.e. it will be counterfed by Move). Following our assumptions
about the ϕ-probe on T, the derivation crashes as soon as T cannot find a goal.
(61)
F > ↓Agr↓:
↑Agr↑dresses
> (Merge)
TP
16about the ϕ-probe on T, the derivation crashes as soon as T cannot find a goal.
& (61)
[gender: !]
NP> (Merge) > ↓Agr↓:
↑Agr↑
TP
EVALUATING
PUSKAR & MURPHY (2015)
that regardless of adjectival agreement, CCA on the verb is still possible, as shown in (7-8) where two
to the verbal probe as well and necessarily trigger feminine gender agreement in (7-8), contrary to fact.
distinct probes end up with different agreement.
Theories of post-syntactic CCA do not face this problem as both adjectival and verbal
agreement could look for the linearly closest goal, which is the first conjunct for the adjective and the
(7) [Ene na
drugih ležeče
[ žage
in
kladiva]]
so
bila
umazana.
last for the verb. This argument against syntactic theories of CCA is similar to the one presented by
one(9)on
other Marušič
lyingF.PL
hammer
dirty
F.PL
N.PL AUX wereN.PL
examples like
((17) from
et al. saw
2008),
whereand
two verbal
probes
on opposite sides of
theNEU
otherfeaturally
lying sawsdifferent
and hammers
were which
dirty.' is in both cases agreement with
(Slovenian)
subject end'On
up each
realizing
agreement,
the
(8)
[Jedne
na
drugima
ležeće
[
žage
i
kladiva]]
su
bila
prljava.
closest conjunct. Whatever syntactic mechanism one considers, only one set of features should be
one
on
other
lying
saw
and
hammer
AUX
were
dirty
F.PL
F.PL
N.PL
N.PL
NEU
visible to the verbal probes, which should both realize the same agreement.
'On each other lying saws and hammers were dirty.'
(BCS)
(9) Včeraj
so
bile
[ krave
in
teleta ]
prodana.
(Slovenian)
Theories
of
syntactic
conjunct
agreement
predict
that
given
the
structure
in
(5b),
the
entire
DP should
yesterday aux beenF.PL [ cowF.PL
and calfN.PL ]
soldN.PL
also
carry feminine
because the inner ConjP carries them (as in Puškar &
‘Yesterday
cows andgender
calves features––either
were sold.’
Murphy 2015) or else because the derivation resulted in feminine features being copied to the
(Bošković
2009).
Regardless
of the
derivation, theagreement,
subject's gender
should
be visible
The adjective
same logical
argument
comes
from Dutch
complementizer
where features
the C agrees
with
thesingular
verbal conjunct,
probe as well
feminine
gender
agreement in (7-8), contrary to fact.
only to
one
whileand
thenecessarily
lower verb trigger
agrees with
the whole
ConjP:
Theories of post-syntactic CCA do not face this problem as both adjectival and verbal
(10) agreement
Ich dink
de-slook for
[ the
toow
en closest
Marie]goal, which
kump. is the first conjunct for the
(Dutch)
could
linearly
adjective and the
I for think
that-This
youSG and
Marie
comePL of CCA is similar to the one presented by
last
the verb.
against
syntactic theories
2SG argument
‘I think that
Marie
come.’et al. 2008), where(Haegeman
van Koppen
2012) sides of the
examples
likeyou
(9) and
((17)
fromwill
Marušič
two verbal&probes
on opposite
subject end up realizing featurally different agreement, which is in both cases agreement with the
In sum,
closest-conjunct
agreement
cannot be
the result one
of computation
purelyone
within
thefeatures
ConjP and
closest
conjunct. Whatever
syntactic
mechanism
considers, only
set of
should be
agreement
interfacing
with
the ConjP
level,the
as same
such agreement.
Probes can potentially access
visible Probes
to the verbal
probes,only
which
should
both realize
individual conjuncts within the ConjP, in a way crucially sensitive to postsyntactically-determined
linear(9)
order.Včeraj
so
bile
[ krave
in
teleta ]
prodana.
(Slovenian)
yesterday aux beenF.PL [ cowF.PL
and calfN.PL ]
soldN.PL
Aljović, N.‘Yesterday
& M. Begović.
2015.
Morpho-Syntactic
Aspects of First Conjunct Agreement. Paper
cows and
calves
were sold.’
presented at AAB conference Zadar, Croatia. Aoun, J., E. Benmamoun & D. Sportiche. 1994.
Agreement,
word
order,argument
and conjunction
some
varieties
of Arabic. LIagreement,
25:195–220.where
Bhatt,theR.C&agrees
M. with
The same
logical
comes in
from
Dutch
complementizer
Walkow. 2013. Locating agreement in grammar: An argument from agreement in conjunctions. NLLT
only one singular conjunct, while the lower verb agrees with the whole ConjP:
31:951–1013. Bošković, Ž. 2009. Unifying first and last conjunct agreement. NLLT 27:455-496.
Haegeman, L. & M. Van Koppen. 2012. Complementizer agreement and the relation between Co and
(10) Ich dink de-s
[ toow en
Marie]
kump.
(Dutch)
To. LI 43.3: 441-454. Marušič, F., A. Nevins, & A. Saksida. 2007. Linear Agreement in Slovene
I
think that- al. (eds) you
Marie
come
SG and
Conjuncts. In R. Compton et 2SG
FASL
15. Ann Arbor: MichiganPLSlavic Publ. 210-227. Marušič,
‘I
think
that
you
and
Marie
will
come.’
(Haegeman
&
van
Koppen
2012)
F., A. Nevins, & B. Badecker. 2015. The Grammars of Conjunction Agreement in Slovenian. Syntax
18.1:39-77.
Munn, A. 1999.
First conjunct
agreement:
Against
a clausal analysis.
LI 30:in643
– 683.
together
lying.N
[invites.N
and
adverts.F
]]
aux.pl
landed.F
trash
In sum,
agreement
cannot be
the resultis of
purely within
the ConjP and
Puškar,
Z. &closest-conjunct
A. Murphy. 2015.
Closest conjunct
agreement
an computation
illusion. Ms. University
of Leipzig.
agreement Probes interfacing only with the ConjP level, as such Probes can potentially access
individual conjuncts within the ConjP, in a way crucially sensitive to postsyntactically-determined
linear order.
SANDWICHED AGREEMENT
When agreement can be with different parts of &P
Complementizer shows CCA; main verb shows resolved agreement
• Correctly
DCA
Rules in Preverbal HCA & Rules out Postverbal
Leftward participle shows CCA with NP1;
Rightward participle shows CCA with NP2
• Even
rules out impossibility of Medial Conjunct Agreement in
3-conjunct coordinations
• Disadvantage: as
all the action is internal to &P, faces difficulty
when different agreement targets pick different parts of the
conjunction…
17
EXPERIMENTAL DESIGN:
SANDWICHING & DEPENDENT VARIABLES
• PARTICIP1
[Skupaj ležeča [ vabila in reklame ]] so
[
pristale v smeteh
DP-internal adjective shows CCA with NP1; participle shows CCA with NP2
Such
challenge
models
in which
Aljović, N.examples
& M. Begović. 2015.
Morpho-Syntactic
Aspects of First
Conjunct Agreement. Paper
presented at AAB conference Zadar, Croatia. Aoun, J., E. Benmamoun & D. Sportiche. 1994.
all
CCA
is
computed
internally
to
the
&P Bhatt, R. & M.
Agreement, word order, and conjunction in some varieties of Arabic. LI
25:195–220.
Walkow. 2013. Locating agreement in grammar: An argument from agreement in conjunctions. NLLT
31:951–1013. Bošković, Ž. 2009. Unifying first and last conjunct agreement. NLLT 27:455-496.
Haegeman, L. & M. Van Koppen. 2012. Complementizer agreement and the relation between Co and
To. LI 43.3: 441-454. Marušič, F., A. Nevins, & A. Saksida. 2007. Linear Agreement in Slovene
Conjuncts. In R. Compton et al. (eds) FASL 15. Ann Arbor: Michigan Slavic Publ. 210-227. Marušič,
F., A. Nevins, & B. Badecker. 2015. The Grammars of Conjunction Agreement in Slovenian. Syntax
18.1:39-77. Munn, A. 1999. First conjunct agreement: Against a clausal analysis. LI 30: 643 – 683.
Puškar, Z. & A. Murphy. 2015. Closest conjunct agreement is an illusion. Ms. University of Leipzig.
18
RESULTS: SANDWICHING OK,
AND SO IS DOUBLE HCA
[C1 & C2] PARTICIP2
• Split Agreement: Participle
1 agrees with C1 and Participle 2
AgrF FN AgrN
AgrF FN AgrF
agrees with C2
*
• Double
HCA: Both participles agree with C1
• Double
LCA: Both participles agree with C2
Split Agreement
19
Double HCA
20
AgrF NF AgrF
Double LCA
THIS LATTER ASYMMETRY IS
EASY FOR MARUSIC ET. AL
• Agree-Copy After
• Agree-Copy
• Lack
Linearization Yields Split Agreement
Before Linearization Yields Double HCA
of Agree-Copy Yields Double Default
• There
is no way in our theory to get Double DCA
PURELY SYNTACTIC THEORIES
FOUNDER ON SANDWICHING
• Split
agreement is already a problem necessitating revisions
• But
even given those, there is no obvious syntactic way to rule
in Double HCA while ruling out Double DCA
• Feature
deactivation (with multiple cycles of Agree) would
help rule in Double HCA, but would then rule out Double
Default
21
22
THREE FUTURE EMPIRICAL
PATHS WE’LL INVESTIGATE
OVERALL CONCLUSIONS
• The
Marusic et al 2015 model predicts masculine agreement
to be freely available in split agreement cases (e.g. CCA on
one side and default on the other). This is worth testing; it
may be that we overgenerate.
• We’ve
shown two ways of sandwiching; are there different
results for the DP-internal adjective cases? Is concord
inherently different (cf. Norris 2014)
• Slovenian
already has more CCA than BCS without
sandwiching; we should test the relative amount of double
HCA vs split agreement in BCS
23
• Closest-conjunct
agreement is the result of external Probes
that can access individual parts of the ConjP
• We’ve
proposed three (and only three) strategies, based on
Agree-Copy: No Agree-Copy yields default, prelinearization
Agree-Copy yields HCA; postlinearization Agree-Copy yields
CCA
• The
real kicker comes with sandwiched agreement, which
vindicates the possibility of double CCA (different results on
each probe) and double HCA, which are not easily derivable
in purely syntactic approaches
24
THANK YOU!
SELECTED REFERENCES
• Marušič, Franc, Andrew
Nevins & Bill Badecker (2015): ‘The Grammars of Conjunction Agreement in
Slovenian’, Syntax 18(1), 39–77.
• Bhatt, Rajesh
& Martin. Walkow (2013): ‘Locating agreement in grammar: an argument from agreement in
conjunctions’, Natural Language and Linguistic Theory 31(4), 951–1013.
• Bošković, Željko. (2009): ‘Unifying
first and last conjunct agreement’, Natural Language and Linguistic Theory
27(3), 455–496.
• Puskar, Z. &
Murphy, A. (2015). Closest Conjunct Agreement is an Illusion. Ms., Universität Leipzig.
• Norris, Mark
(2014). A theory of nominal concord. Phd, UCSC.
• Haegeman, L.
& M. van Koppen. Complementizer agreement and the relation between Co and To. LI 43.3:
441-454.
25
27
!!
THE!EMSS!Team!
!
26
28
!