STRUCTURE BUILDING THAT CANT BE

1
STRUCTURE BUILDING THAT CANT BE Samuel Epstein,
Hisatsugu Kitahara, Daniel Seely sepstein_at_umich.
edu kitahara_at_icl.keio.ac.jp
tseely_at_emich.edu
2

STRUCTURE BUILDING THAT CANT BE Samuel Epstein,
Hisatsugu Kitahara, Daniel Seely sepstein_at_umich.
edu kitahara_at_icl.keio.ac.jp
tseely_at_emich.edu
1. Introduction
Central goal
Explore the form and consequences of a deduction
of cyclic transfer, appealing to Accessibility
Condition Simple, single structure building
operation Merge Strong Minimalist Thesis 3rd
Factor
3
Consider typical counter-cyclic IM at CP phase
C2 C1 T3
Bill
C2
T2
C1
Transmit ? (C1, T1)
?
NOM
IM (T2, Bill)
v Bill
T1
?
Value (T1, Bill)
3rd Sg masc
VP V
NP ate rice
v v
NOM
Claim IM at root and counter cyclic IM are
simply Internal Merge and Internal Merge is
just Merge. There is only one structure
building operation.
Merge (?, ?) ?, ?, ?
4
Counter-cyclic IM Disconnect then Reconnect
Replacement, unlike any other instance of Merge
( EM and IM at the root)
C
C
C
Transmit ? (C, T)
T Bill
IM (T, Bill)
T
C
?
Value (T, Bill)
NOM
v Bill
T
?
3rd Sg masc
v v
NOM
Replacement is required to get the desired
result but replacement is not an instance of
Merge--it involves structure building beyond
Merge.
5
Formal replacement see Kitahara (1997)
T / \ T v
/ \ Bill v / v
C / \ C
Cnew
Form CP C, C, T
/ C
\
Tnew / \ NP T / \ T v
/ \ Bill v / v
Form Tnew T, NP, T
Tnew / \ NP T / \ T v
/ \ Bill v / v
Merge C T
COLD / \ C T / \ T v /
Bill
IM NP to T
Take T of C, put Tnew in place of T thereby
creating Cnew C, C, Tnew category
creating that is NOT output of Merge
Replacement instead
6
We suggest SMT theres Merge, not a separate
structure building operation Replacement. SMT
only (simply) Merge
T, NP, T C, C, T
Note this is not a single object, but two
objects that share a term, T2
C
Tnew
Tnew and C are different set-theoretic objects
T2
C
Bill
There is no root!
Transmit ? (C, T1)
?
NOM
Further EM impossible given Accessibility
IM (T2, Bill)
v Bill
T1
?
Value (T1, Bill)
Thus, Transfer
3rd Sg masc
v v
Thus, cyclic Transfer
NOM
This two peaked effect is distinct from
Multi-Dominance, as defined in important work
of Citko (2005), but rather is of the type
investigated by Groat (1994), Kawashima
Kitahara (1994), Epstein et al (1998).
7
Central components of our analysis
By Chomskys Label Accessibility Condition (AC),
only the label of an entire Syntactic Object, the
root, is accessible to the Narrow Syntax. AC
Zero Search Efficiency 3rd Factor Any
system must access something 3rd Factor
access it with least effort (thus label
designated).
If the label accessed by NS (through
Accessibility) is a Phase Head (C and v), then
operations other than EM are initiated.
Localization of operations to as few points as
possible is more efficient than a free for
all, so consistent with SMT. Again, may be a
3rd Factor consideration.
8
An accessed phase head itself has access only to
elements within its derivational c-command
domain ( terms of the category Merged with the
PH minimal search). Merge comes free.
(only element of UG) SMT maximize the
explanatory effects of Merge. The only
relations available to NS are those that directly
follow from the simple, single, necessary
structure building operation Merge.
Merge comes for free. Thus, the SMT determines
that replacement (an additional structure
building operation) be eliminated again, in
favor of the simplest operation, Merge (X,Y)
X, Y.
9
Thus, counter-cyclic applications of Merge yield
the two-peak "effect." Since "replacement" is
barred by SMT, the two-peaked effect necessarily
results. The two-peaked effect entails
absence of a root if there is no root, then
there is no label of the root, and hence by the
AC, further Merge is impossible. In short The
derivation halts.
To proceed with the derivation, one peak is
eliminated by Transfer, thereby creating a root.
Thus, we derive cyclic Transfer.
10
Consequences
Deduction of cyclic Transfer
Derive the otherwise stipulated invisibility of
Spec T to attraction by C.
Empirical differences between the traditional PIC
and our two-peak-effect-induced transfer the
latter gives apparently correct result for cases
problematic for OP/AUGB system.
For PIC phases are absolute, CP vP For us,
phases relative
11
To review this introduction (!)
C
T3
T2
C
Bill
Transmit ? (C, T1)
?
NOM
IM (T2, Bill)
v Bill
T1
?
Value (T1, Bill)
3rd Sg masc
v v
NOM
12

• A quick and dirty summary of Valuation-induced
• cyclic transfer and an empirical problem with
  it
• from Epstein, Kitahara, Seely (to appear)

Syntactically determined features, like phi of
C/T and Case of NP are lexically Unvalued (by
hypothesis, BEA).
These features start as Unvalued and become
Valued in the course of a derivation
Such features must be removed from an object
bound for SEM (by Transfer), since they are
uninterpretable at SEM and thus (would) induce
LF Crash.
When does Transfer (spell-out) remove them?
13
When does Transfer take place?
Before Valuation is too early Unvalued features
wont be implementable at PHON in effect, they
(would) induce Crash at PF.
After Valuation is too late Once Valued, the
distinction between Valued/Unvalued is lost.
Transfer wont know to spell out, say, the
valued phi features of C/T, since these are
identical to the (inherently) valued phi features
of N.
So, Transfer (in effect) operates during
Valuation.
Transfer must take place internal to a Valuation
operation (EpsteinSeely (2002)) or else
internal to the Phase (BEA, OP, AUGB). Valuation
based Transfer requires at least
some derivational history minimized as much as
possible it must see what goes from unvalued
to valued.
14
This logic (attractively) induces cyclic
Transfer. The logic is also explicitly appealed
to in AUGB to derive CP, and not TP, as a phase
and to derive feature inheritance (from C to T
and v to V).
Adopting observations of Marc Richards, AUGB
states it follows that the PIC entails that
TP cannot be a phase, with operations of
valuation and A-movement driven by properties
of T. Suppose TP were a phase. Then its
interior will be transferred by PIC, but the
head T will retain its valued uninterpretable
features. The derivation will therefore crash
at the next phase. AUGB, p. 13
The logic of Valuation induced Transfer is
clear Syntactically valued features at or
outside the phase head (i.e. at the Edge) induce
Crash.
15

• Epstein, Kitahara, Seely (to appear) argue that
• this same logic leads to a pervasive empirical
  problem with the Valuation induced cyclic
  Transfer system

All instances of movement across a phase boundary
are disallowed and this leads to massive
undergeneration.
If T cannot be a phase head because syntactically
Valued features at or beyond T will induce Crash
at the next phase, then C and v cannot be either
(since syntactically Valued features do occur at
and beyond, i.e. at the Edge of, C and v)
16
Consider a specific example of this problem, with
key derivational points indicated a. whom
did they like? b. the vP phase for "whom do
they like?" c. vP whom v' they v' v'
vlike(valued phi) ACC
VP twhom V' tV twhom
?
?
once valued, they are indistinguishable at
the next phase level from interpretable
features, hence will not be deleted before
reaching the CI interface
17

• An even quicker review of Epstein, Kitahara,
  Seely
• (to appear) solution to this problem.

Syntactically Valued features are PHON
Transfer spell out all and only PHON
features. After Valuation is not too late.
SMT features are either SEM or non SEM (PHON).
Unvalued features are not SEM.
They are therefore PHON.
Once PHON always PHON feature conservation.
Thus, syntactically Valued features are PHON.
18
With this much in place, the pervasive empirical
problem revealed above, is eliminated.
Consider again the structure troublesome for
OP/AUGB whom did you visit whom you v whom
visit whom AccCase
19
SUMMARY SO FAR
Valuation induced Transfer doesnt work.
Syntactically Valued features as PHON
features solves the after is too late problem.
But, interestingly, we now must return to the
question we started with
When does cyclic Transfer apply and why?
20
5. Deducing Cyclic Transfer
How deduce cyclic Transfer?
If not from Valuation, which doesnt work, from
what can Cyclicity be deduced or must we
STIPULATE it?
21
Recall our basic idea for cyclic transfer
T
C
DPsubj NOM
T
C
u?
v
T
u?
?
V
v
DPsubj
uCase
NOM
DPobj ACC
v
V
u?
DPobj
V ate
uCase
ACC
u?
?
22
Deducing Cyclicity I SMT, 3rd Factor and Label
Accessibility
Strong Minimalist Thesis (SMT) language is an
optimal solution to interface conditions that FL
must satisfy that is, language is an optimal
way to link sound and meaning If SMT held fully,
UG would be restricted to properties imposed by
interface conditions. A primary task of the MP
is to clarify the notions that enter into SMT
and to determine how closely the ideal can be
approached. Chomsky On Phases, p. 2.
3rd Factor Principles not specific to the
language faculty. principles of structural
architecture and developmental constraints
including principles of efficient computation,
which would be of particular significance for
all computational systems NC, 2005, p. 6
23
By Chomskys Label Accessibility Condition (AC),
only the label of an entire Syntactic Object, the
root, is accessible to the Narrow Syntax. AC
Zero Search Efficiency 3rd Factor Any
system must access something 3rd Factor
access it with least effort (thus label
designated).
MIN PROG Page 248 We assumed earlier that Merge
applies at the root only. In the bare system, it
is easy to see why this is expected. Suppose the
derivation has reached stage SIGMA, with objects
ALPHA and BETA. Then Merge may eliminate ALPHA
and BETA from SIGMA in favor of the new object K
gamma alpha , beta, with label GAMMA. That
is the simplest kind of merger. We might ask
whether Chl also permits a more complex
operation given alpha and beta, select K within
beta (or within alpha, its immaterial) and
construct the new object Gamma alpha, K which
replaces K within beta. That would be an
application of Merge that embeds alpha within
some construction beta already formed. Any such
complication (which would be quite serious)
would require strong empirical motivation. I know
of none, and we therefore assume that there is
no such operation. Merge always applies in the
simplest possible form at the root.
24
1. Merge is indispensable part of the human
faculty of language. It comes for free.
2. But, how Merge operates is constrained by 3rd
factor considerations.
Under the assumption (2), we take seriously the
idea of minimal search. The basic intuition is
that the accessible domain for NS is highly
restricted, so that computational complexity is
greatly reduced. Under minimal search, EM can
access only the labels of the full objects (
either lexical items or roots thus far
constructed), and IM can access the label of the
full object and finds the second object in the
search domain of the accessed label by the
probe-goal algorithm.

• So, we assume (3)
• 3. The label accessibility algorithm and the
  probe-goal algorithm
• are part of minimal search, imposed under 3rd
  factor
• consideration.

25
Label Accessibility Condition (AC) NS has
access to only the label of the entire
Syntactic Object, i.e. the root.
What is The Root?
K is the root iff for any Z, Z a term of K,
every object that Z is a term of is a term of K
Term (Chomsky 1995) for any structure
K, (a) K is a term of K, and (b) if L is a term
of K, then the members of the members of L are
terms of K
26
Consequences of Label Accessibility for EM?
NS has access to only the highest label, and
hence only a root can be EMed
Cant reach inside an object and pull out a
category for EM.
So no Sidewards Movement nor Multi-Dominance
Please forgive us!
27
Consequences of Label Accessibility for IM?
If the label accessed by NS (through
Accessibility) is a Phase Head (C and v), then
operations other than EM can be initiated.
C and v inherently bear phi features (and EF),
Chomsky OP, AUGB.
What are the relevant operations (other than EM)?
Inheritance C to T and v to V
Valuation/Agree Value phi and Case
Both under Probe-Goal
28
An accessed phase head itself has access only to
elements within its derivational c-command
domain ( terms of the category Merged with the
PH minimal search). Merge comes free.
(only element of UG) SMT maximize the
explanatory effects of Merge. The only
relations available to NS are those that directly
follow from the simple, single, necessary
structure building operation Merge.
Recall this eliminates replacement
Merge comes for free. Thus, the SMT determines
that replacement (an additional structure
building operation) be eliminated again, in
favor of the simplest operation, Merge (X,Y)
X, Y.
29
Inheritance probing features i.e. the
uninterpretable features phi and EF are
transmitted to the closest accessible head.
Why are these features transmitted C to T and v
to V?
TV inherently has the transitive property
T bears tense thus, unlike Chomsky/Richards we
assume T is not entirely empty.
Case Valuation is done through the feature
complex phi, (EF), tense NOM phi, (EF),
transitive ACC
So, we resurrect and maintain the standard
Case-theoretic assumption finite T assigns Nom
while transitive V assigns Acc
30
More detailed story of Inheritance for the QA
Presumably, in languages like West Flemish, C
exhibits phi-agreement but not Tense. AUGB
continues If that is basically accurate, then
there are two possibilities. One is that Tense
is a property of C, and is inherited by T. The
other is that Tense is a property of T, but
receives only some residual interpretation unless
selected by C (or in other configurations, e.g.,
in English-like modal constructions). Note
that the first option (C bears tense and T
inherits it from C) is adopted by Richards (2007)
(without discussion). Here notice, that if
option two is adopted (T inherently bears tense),
feature-inheritance is necessary for
Case-valuation. Keeping this in mind One
advantage of the latter option Tense is a
property of T is that T will then have at least
some feature in the lexicon, and it is not clear
what would be the status of an LI with no
features (one of the problems with postulating
AGR or other null elements). Another advantage
would be an explanation for why C never
manifests tense in the manner of phi-features
(if that is correct).
31
In Sum feature inheritance is required for Case
Valuation
But, then, how motivate Internal Merge?
EPP the double EFs (one inherent, and one
from probing-feature transmission) borne by T
and by V after probing-feature
transmission (from C and v respectively) require
satisfaction Merge.
So, the EPP is, in effect, a derived property,
emerging only with feature Inheritance.
32
To sum up the details
T
C
DPsubj NOM
T
C
u?
v
T
u? EP EP
?
V
v
DPsubj
uCase
NOM
DPobj ACC
v
V
u?
DPobj
V ate
uCase
ACC
u? EP EP
?
33
6. Consequences of the analysis
Derive Cyclic Transfer, maximizing SMT and 3rd
Factor
Derive the otherwise stipulated invisibility of
spec T to C
Derive certain empirical advantages.
34
Deriving Cyclic Transfer
SMT theres Merge, and not a separate structure
building operation Replacement. SMT only
(simply) Merge
T, NP, T C, C, T
Note this is not a single object, but two
objects that share a term, T2
C
Tnew
Tnew and C are different set-theoretic objects
T2
C
Bill
There is no root!
Transmit ? (C, T1)
?
NOM
Further EM impossible given Accessibility
IM (T2, Bill)
v Bill
T1
? EF EF
Value (T1, Bill)
Thus, Transfer
3rd Sg masc
v v
Thus, cyclic Transfer
NOM
35
Deriving the invisibility of spec T to C
Since C was never Merged with Tnew, it does not
derivationally c-command Spec Tnew. When C was
Merged with T2, there was no Spec of T2.
C
Tnew
T2
C
Bill
?
NOM
v Bill
T1
?
3rd Sg masc
v v
NOM
36
Deriving the invisibility of spec T to C
Suppose that CHL has constructed TP T vP NP
v' v VP. As a next step, External Merge
(EM) merges C to TP T vP NP v' v VP,
forming CP C TP T vP NP v' v VP. At
this point, T inherits unvalued phi-features and
EF from C (which C bears lexically).
Subsequently, Agree/Valuation values
phi-features on T and Case on N, and Internal
Merge (IM) merges NP to TP T vP NP v' v
VP, forming TP NP T' T vP NP v' v VP.
37
Given the unique property discussed above, NP
occupying SPEC-T (derivationally) c-commands
every term of TP T vP NP v' v VP, but C
does not c-command NP occupying SPEC-T, because,
recall that C was EM merged to a "SPEC-less" TP,
namely TP T vP NP v' v VP, and the
(derivationally determined) c-command domain of C
was established at that point of the derivation.
In fact, C was not Merged with the TP that
includes NP in spec TP position as far as C is
concerned, NP in Spec T is not there.
38
Relative to such counter-cyclic IM, Chomsky
(2005) notes "the edge-feature EF of C
cannot extract the PP complement from within
SPEC-T if it could, the subject-condition
effects would be obviated. It must be, then,
that the SPEC-T position is impenetrable to EF,
and a far more natural principle would be that
it is simply invisible to EF." No analysis of
this optimal result is offered in OP/AUGB. But,
the present analysis explains the "invisible"
status of the SPEC-T position (formed by
"non-cyclic" merger) as a natural property of
the derivational system. IMed spec TP is in
fact invisible to C since C has no relation to
this spec.
39
Consider who left
This cannot be first wh-move to Spec T and then
wh-move from Spec T to Spec C. EF of C cant
probe (and attract) Spec of T, since Spec of T is
naturally not there for C.
C
who
C
Tnew
T2
C
who
?
NOM
v who
T1
?
3rd Sg masc
v v
NOM
40
Invisibility of Spec T to C, and the OP/AUGB
paradigm
(5)(i) it was the CAR (not the TRUCK) of which
they found the (driver, picture) (ii) of
which car did they find the (driver,
picture)? Can extract from object. (6) (i) it
was the CAR (not the TRUCK) of which the
(driver, picture) caused a scandal (ii) of
which car did the (driver, picture) cause a
scandal Cant extract from subject. (7)(i) it
was the CAR (not the TRUCK) of which the
(driver, picture) was found (ii) of which car
was the (driver, picture) awarded a prize Can
extract from derived subject (18) (i) it is the
CAR (not the TRUCK) of which the (driver,
picture) is likely t to t cause a scandal
(ii) of which car is the (driver, picture)
likely t to t cause a scandal (19) of which
car did they believe the (driver, picture) to
have caused a scandal Can extract
Ideally No extraction from Spec of a phase
head. Is extraction from other Specs.
For the ideal to hold, its crucial that spec T
not be visible to C.
Precisely what we derive.
41
Further empirical advantages some speculations
Traditional PIC phases are absolute (i.e.
categorial CP and vP are phases. Period.
But, for EKS phase-hood is relative it requires
IM and that, in turn, requires Inheritance (for
Case)
For example, Spell out at vP (of VP) occurs for
us only when v Transmits its phi features to V
for Case. If V doesnt have Acc property
(passive, unaccusative), then theres no two
peak effect and VP will not be spelled out yet.
(i)Joni likuðu þessir sokkar Jon.DAT
like.PL these socks.NOM "Jon likes these
socks."
Jonsson 1996143
42
SUMMARY AND CONCLUSIONS
Attempted to maximize the explanatory value of
SMT and 3rd Factor.
With potentially greater empirical coverage.
43
(No Transcript)