?tekauer Pavol, Lieber Rochelle. Handbook of Word-Formation
Подождите немного. Документ загружается.
C
HOMSKY
’
S
REMARKS
AND THE TRANSFORMATIONALIST HYPOTHESIS
S
12
7
(
3) Sentence: the enem
y
destro
y
ed the cit
y
Noun phrase: Possessive Noun PP [of Ob
j
ect]
|||
sub
j
ect verb ob
j
ect
Each part of the sentence must assume a different syntactic label to make this
work. Each of these solutions involves a wrenching revision of higher structure.
T
here are two broad strategies to deal with
t
hese facts, to cope with what fails to
fit neatl
y
into sentential
g
rammar:
A
bstract Theor
y
: create a more abstract theor
y
in
w
hi
c
h n
o
minalizati
o
n
s
an
d
se
nt
e
n
ces
l
oo
k th
es
am
eo
r in
w
hi
c
h
se
nt
e
ntial
s
tr
uc
t
u
r
e
fall
s
in
s
i
de o
f NP
s
tr
uc
t
u
r
e.
Lexicalist Theor
y
:
attri
bu
t
e
all
o
f th
e odd
iti
es
t
o
hi
s
t
o
ri
c
al r
es
i
due o
r l
e
xi
c
al
exceptions found in a vocabulary list who
s
e properties may reach outside of a
particular language or grammar.
L
ees chose the route of subordination
,
where a Sentence was inside the NP.
C
homsky, on the other hand, chose an abstract syntax (X-bar) which left lexical
r
estrictions to be stated elsewhere. Chomsky’s approach is extended in modern
m
inimalist theories. Lee’s a
pp
roach continues to
be
r
e
fl
ec
t
ed
in th
e
v
i
ew
that th
e
r
e
i
s
a hidden VP in the NP; we will return to this a
pp
roach below.
U
nder Chomsky’s approach a choice of feature (N, V) determined the behavio
r
o
f the label on an XP. If a phrasal node became a VP, then it assigned case (
d
estro
y
the cit
y
); if it became an NP, then it acquired an affix
o
r an
e
xtra
b
ra
c
k
e
t that
blocked the assignment of case to an object, requiring insertion of a preposition to
do the work
(
the destruction of the cit
y
).
(
4
)
XP
Spec X
enem
y
X Comp
N PP
V
NP
d
estroy city
T
his is a beautiful resolution of the core problem. But it created other problems
which continue to be a challenge, as we shall see.
1.2 Transformations
B
eyond phrase-structure, the anchor of transformational grammar is the concept
o
f transformations. If common transformations apply to both sentences and
n
ominalizations, then the ar
g
ument for their common source is much stron
g
er.
12
8
T
O
M R
O
EPER
C
homsky argued that the passive transformation can occu
r
inside
th
e
no
minalizati
o
n
:
(
5) the city’s destruction by the enemy
The object moves into the subject
position and the agent into a
t
by
-p
hrase. This is
the kind of evidence for abstraction that good theories exhibit. Yet questions remain.
W
hy is the subject apparently optional in nominalizations but obligatory in
se
nt
e
n
ces?
2. THE S
U
BJECT ENIGMA
One explanation for the apparent optionality of the sub
j
ect in nominalizations is
that it reall
y
is obli
g
ator
y
but is carried b
y
an invisible PRO,
j
ust as in
V
P structures
[John wants PRO to sin
g
]. If there is a hidden PRO, then it should be able to be
c
ontrolled. Evidence for control of a sub
j
ect position comes from contrasts like fro
m
(
D. Charney – personal communication):
(
6) a. John was in PRO control of the army => John controls
b. John was in the PRO control of the army => the army controls
T
he definite article has the power to block outside control, an interesting
phenomenon for which a deeper theor
y
remai
n
s
elusive. One possibilit
y
is that the
Determiner ‘protects’ the inner PRO from requirin
g
case-assi
g
nment (Chomsk
y–
p
ersonal communication
)
and block
s
obligatory control as well.
B
inding facts fall into place as well as a support for the hidden subject PRO:
(
7) a. ? the dressing of himself thrilled the little boy.
b. the dressing of the little boy thrilled hi
m
c. The little boy’s dressing himself thrilled hi
m
In (7b) we
g
et the implication that someone else must have dressed the little bo
y
,
exactl
y
as if there were a PR
O
ar
b
in a hi
g
her position [the PR
O
ar
b
dressin
g
of the
l
ittle boy] which can only be identical with an object if a reflexive is present, while
i
n (7a) it follows that a coindexed PRO is in the higher position [the PR
O
i
dressing
of himself
i
f
f
]
, allowing the reflexive to appea
r
i
n
(
7c
)
and in
(
7a
)(
with a clea
r
m
eaning even for those who find the sentence marginal).
T
he nature of the sub
j
ect role in nominalizations has, however, been the
persistent target of discussion. Wasow and Roeper (1972) assumed that nominal
g
erunds had no sub
j
ect:
(
8) the sin
g
in
g
of son
g
s
C
HOMSKY
’
S
REMARKS
AND THE TRANSFORMATIONALIST HYPOTHESIS
S
129
G
rimshaw
(
1990
)
, Kratzer
(
1996
)
, Alexiadou
(
1999
)
and others have made a
s
imilar assumption.
1
Williams (1987) proposed that the possessive relation can
e
ncompass any relation and therefore one can independently capture the notion of
s
ub
j
ect without a true verbal sub
j
ect position. Evidence in
be
half
o
f thi
sc
laim
co
m
es
fr
o
m f
o
rm
s
lik
e:
(
9
)
John’s idea
where no verb is present but John takes on the A
g
ent role.What would show that
sentence-like sub
j
ect behavio
r
i
s involved? In the best scientific tradition
,
it should
be abstract, indirect evidence which proves the most persuasive. Modern work has
p
rovided new su
pp
ort for the
h
idden subject position
.
F
irst it has been argued by Roeper (1987) that the hidden subject can act as a
co
ntr
o
ll
e
r in
se
nt
e
n
ces o
f th
e
f
o
rm
:
(
10
)
the PR
O
i
destruction of the cit
y
[PR
O
i
to
p
rove a
p
oint]
where it is
p
ossible to claim that the Agent that destroys the city proves a point. This
argument has been criticized by William
s
(1987) and Lasnik (1984), although their
c
riticism seems to apply to the example, not the deeper claim. They observe that it is
possible that the entire nominalization is the subject:
(11) the destruction of the cit
y
proved a point
N
ow it seems that there is no necessity for the Agent, separate from the
nominalization, to be at work. However other examples point exactly at the hidden
Ag
ent:
(12) a. The use of dru
g
s to
g
o to sleep
b. The opening of the side door to enter the roo
m
H
ere it would be impossible to argue that the whole nominalization is the subject
because it would produce these readings:
(13) a. *The use of drugs went to sleep.
b. *The openin
g
of the door entered the room.
F
u
rth
e
r
ev
i
de
n
ce
that th
e
n
o
minalizati
o
n involves a true sub
j
ect position comes
from the argument that it can be blocked by preposing the object:
(14) ?*the drug’s use to go to sleep
1
See Alexiadou (in preparation) for a good summary.
1
30
TO
M R
O
EPER
(15) ?*the city’s destruction to prove a point
These data have been further challenged from the perspective of the view that no
s
ub
j
ect is involved, by claiming that they are Results (Grimshaw 1990). If those
prenominal possessives are linked to Result nominalizations, then the absence of
c
ontrol could have a different explanation, namel
y
, that onl
y
action nominalizations
allow purpose clauses to be controlled.
2
Thus we would no more expect
*
the cit
y
’s
destruction to prove a point
th
a
n
t
*
the city’s façade to prove a poin
t
.
I
s
th
e
r
e
f
u
rth
e
r
e
vidence relevant to this dis
p
ute?
2
.1 Passive -ability Nominalizations
N
ew evidence points again to the e
x
i
stence of the sub
j
ect position in
nominalizations
(
van Hout and Roeper (to appear)):
(16) a. the learnabilit
y
of
g
rammar b
y
children
b. the heritability of IQ by children
c. *children’s learnability of gramma
r
d. *children’s heritability of IQ
If the sub
j
ect position is free, then it ought to allow an Agent reading. Why
wou
l
d
-ability
n
ominalizations, in particular, block these possessives? What stands
o
ut is that the sub
j
ect position is blocked for the ad
j
ective in sentences as well:
(
17
)
*children are learnable [like *children are learned]
(
18
)
*children are heritable
One possible explanation is that there is a form of passive hidden in the suffix
–
able
which involves a requirement that the sub
j
ect be filled by a
T
HEME or an ob
j
ect,
e
xactl
y
as in the
g
rammar of s
e
n
tences. Therefore we have onl
y
:
(
19) grammar is learnable (like: the grammar is learned)
(
20
)
IQ is heritable.
How would this system translate into nominalizations? Here again, the most
abstract features of grammar come into play. The question is this: since the ob
j
ect is
2
There is some subtlety here too. We do have cases like the man to do the job but it is not very general.
S
entences like:
a. ?*the
g
oal to win
b. the goal of winning
s
hows that bare nouns do not usuall
y
take such clauses, and therefore result nouns mi
g
ht not either.
C
HOMSKY
’
S
REMARKS
AND THE TRANSFORMATIONALIST HYPOTHESIS
S
131
not in the possessive/sub
j
ect position what blocks the pro
j
ection of a sub
j
ect in that
position? In a number of instances a no
t
i
o
n
o
f
cove
rt m
ove
me
nt ha
s
r
ece
i
ved
support. In Germanic languages, the subject
often is filled by an expletive, but
t
features of the sub
j
ect move invisibl
y
to accomplish sub
j
ect/verb a
g
reement:
(
21
)
there are three men
th
e
r
e
i
so
n
e
man
there were
p
ushed three me
n
3
H
ere the object of a passive remains unmoved, but it seems to have moved
c
overtly because number agreement is carried out.
I
f the same operation applies inside passive -abilit
y
nominalizations
,
then it can
e
xplain the blocka
g
e of an a
g
ent sub
j
ect in those positions:
(22) Covert-Ob
j
ect Movement:
the [obj] learnability of gramma
r
<
Í
=============
the [grammar]’s learnability of gramma
r
grammar <========== t
Not onl
y
do we capture the facts, but we support the claim that both a sub
j
ect
p
osition exists and a movement o
p
eration h
a
s occurred inside the nominalization
,
m
uch as Chomsk
y
ori
g
inall
y
proposed, which is the obli
g
ator
y
heart of passive in
both sentences and nominalizations.
W
hereas overt movement is necessary for the
o
bject in sentences, in nominalizations the movement is overt or covert.
This analysis can shed light on many of t
h
e old facts as well. It has always been
m
ysterious that nominalizations allowed a
by
-
phrase in all environments:
(23) a. the city’s destruction by the enemy
b. the destruction of the cit
y
b
y
the enem
y
.
I
f covert movement is allowed
,
then w
e
can see wh
y
the b
y
-phrase can appear;
t
he sentence has a passive as one reading (see Borer 1998). If subjects cannot appea
r
i
n the subject position, then no other element should either. We find this to be true of
t
em
p
oral adverbials as well for -abilit
y
:
4
(
24) a. the deniability of tenure to a beloved professor last year [was easie
r
because of budget demands]
b. *last
y
ear’s deniabilit
y
of tenure to a beloved professor.
c. last’
y
ear’s denial of tenure to a beloved professor.
3
Such sentences are ver
y
awkward in En
g
l
ish but common in other Germanic lan
g
ua
g
es.
4
Blocked
p
ronominal tem
p
oral
p
osse
s
sives are discussed in Roe
p
er
(
1987
)
, van Hout and Roe
p
er
(
1998
)
and Sn
y
der (1998).
13
2 T
O
M R
O
EPER
W
here the object has not moved forward, it a
pp
ears that the tem
p
oral adverb can
o
ccupy the subject position (as it can in other Germanic languages). Since there is a
n
on-
p
assive form fo
r
-
tion
(
the enemy’s destruction of the cit
y
)
, -
tion
allows only
optional passive, which means optional movement of the ob
j
ect to the sub
j
ect,
which in turn means that the sub
j
ect position can be open for a temporal adverb
when passive is not present
.
5
The covert movement anal
y
sis provides a strai
g
htforward explanation for this
co
ntra
s
t a
swe
ll
:
(
25
)
a. the a
pp
earance of John/John’s a
pp
earance
b. *there’s a
pp
earance of John
I
n (25a) the covert movement occurs and therefore the
j
ustification fo
r
the
r
e
-
i
n
se
rti
o
n i
se
liminat
ed.
Th
e occu
rr
e
n
ce o
f
the
r
e
in the sentential syntax is reduced to
t
he need for case-assignment.
The followin
g
facts – thou
g
h tenuous themselves – at first seem to contradict this
account, but once a
g
ain fall into line with the anal
y
sis:
(
26) a. ?there’s appearing to be a proble
m
b. “I mean, there's appearing stiff on camera. And then there’s being
de
a
d
”
W
e even find cases like (26b) from the internet in a commentary on the
presidential candidate John Kerr
y
and the war in Iraq. If these are acceptable, then
wh
y
is there a distinctive decline in acceptabilit
y
for (27):
(
27) *there’s appearing of a proble
m
W
e can explain the acceptabilit
y
of (26) as derived from (28):
(
28) ? the appearing of there to be a proble
m
(
28) requires raising the lower expletive to the upper position, which is what
h
appens overtl
y
in (26). Therefore it is
gr
ammatical because it is a moved-expletive
n
ot an expletive inserted to capture case. A number of subtle facts fall into place
under the covert movement anal
y
sis.
2
.2 -ing Nominalizations
A
longstanding puzzle about
-ing
is solved by this analysis, namely that the
g
passive is not available for all nominalizing affixes. The
-ing
affix does not allow
g
any object in subject position:
5
Note that temporal adverbs can occup
y
sub
j
ect position in Germanic lan
g
ua
g
es in man
y
constructions.
C
HOMSKY
’
S
REMARKS
AND THE TRANSFORMATIONALIST HYPOTHESIS
S
133
(
29) the destroying of the city
(
30) *the city’s destroying
W
e can explain this fact too now that we have identified the familiar passive
o
peration inside nominalizations hidden w
i
thin th
e
a
bs
tra
c
t
c
oncept o
f
covert
move
m
e
nt
.
Th
e
-ing
is purely transitive, not passive, and therefore does not project a
g
THEM
E
role into the sub
j
ect position that must be sub
j
ect to Feature-checkin
g
.
T
herefore no movement to that
p
osition is allowed.
A
new taxonomy emerges: if -abilit
y
c
arri
es
an obligator
y
p
assive rule, -
tion
c
arri
es
an
optional
one, it is not a surprise to find an affix which has
l
no
p
assive
property, -in
g
.
We will turn to structural differences that these affixes reflect shortly.
3.
C
A
S
E A
SS
I
G
NMENT
Hovering in the background of this discussion is the question of how case is
assi
g
ned to both the sub
j
ect and the
o
b
j
ect. In each instance a variet
y
of possessive
m
arkin
g
occurs. Here is where an ‘exceptionalit
y
’ approach is common.
A
special form of o
f
-insertion has been proposed for ob
j
ects and default
g
enitive
h
as been proposed for subjects. Neither is a straightforward reflection of any deepe
r
principles of grammar. The direction of
explanation we would advocate should be
f
c
lear: a more abstract theory of case-assignment would indicate that these genitives
are
j
ust as natural as Nominative and Accu
s
ative in the sentence domain.
W
e are fa
r
from such an analysis, but the genitive and of-insertion are currently widely studied
and ma
yy
ield a more natural anal
y
sis (Ka
y
ne 2002).
3.1 Coping with Exceptions
There are, however, other intriguing facts which suggest that nominalizations are
‘
exceptional’ and therefore marginal to grammar. Anderson (1976) observes that
th
e
r
e
i
s
an
Affected Object Constraint
that i
s
a
c
ti
ve
in n
o
minalizati
o
n
s:
t
(31) the knowled
g
e of al
g
ebra
(32) *algebra’s knowledge
(33) algebra was known
If the sentential passive allows movement of this ob
j
ect, wh
y
is it ruled out in
no
minalizati
o
n
s?
1
34
TO
M R
O
EPER
L
ebeaux (1986) attempts to explain this observation by suggesting that there are
t
wo kinds of movement: thematically-ba
s
ed
m
ove
m
e
nt an
dc
a
se
-
b
a
sed
m
ove
m
e
nt
.
These forms occur together as a rul
e
,
but nominalizations allow a fine
r
discrimination among them.
(34) destro
y
=> destro
y
[
+accusative, +Affected THEME
]
destruction =>
[
+Affected Theme
]
know => [Accusative]/Abstract com
p
lement
knowledge => ø
The nominalization subcategorizes only fo
r
thematic
r
oles
,
not case. Case
r
equires an independent subcategorizer, which is carried by
of
. Therefore we can say
f
f
the knowledge of algebra
.
H
oweve
r
s
in
ce
know
does not affect an Ob
j
ect, it cannot
pro
j
ect one. Rather it pro
j
ect
s
onl
y
accusative case in the verbal form and somethin
g
m
ore complex than a sin
g
le thematic role in the nominal form, with no case. In the
fo
r
m
*
al
g
ebra’s knowled
g
e
t
he preposition is
g
one, therefore no case-assi
g
nment
c
an be generated. Moreover, there is also no
theme
projection which in a case like
the cit
y
i
’
s destruction
(
THEME-trac
e
i
)
i
i
all
ows
a th
e
mati
c
t
r
ace to occur in object
p
osition. Therefore what is needed to
r
econstruct the origin of the phrase is
c
ompletely unavailable. From this perspective, the theory of empty categories
r
eceives a more refined interpretation
:
the thematic and case properties are
potentiall
y
splittable, as revealed in nominalizations.
A th
e
mati
c
-tra
ce
i
sd
iff
e
r
e
nt fr
o
m a
c
a
se
-trace. It is natural in a modular theor
y
t
hat the set of empt
y
cate
g
ories also reflects modular differences. It remains to be
s
een if this logic extends to other modules as well.
W
hat does a verb like
know
project? Under this view of thematic roles, it
projects a semantically open object that ta
k
es case, but allows
p
ro
p
ositions as well,
a
s
in
J
ohn knows the truth
or
J
ohn knows that Bill is here
.
Thi
sv
i
ew
fit
s
th
ec
lai
m
t
hat traditional Agent/Theme type thematic roles are derivative from the notion of
Event. If no Event is present, the semantic structure is different.
3.2 Thematic-binding
I
f the thematic system is separable from the case-system, then we should see it
work independently. Williams (1994) has a
r
g
ued that forms of thematic-binding are
possible, indicating that they are a definable subsystem: one verb’s thematic
pro
j
ection, like Agent, can bind another ver
b
’
s Theme. In fact, they appear in the
i
nvisible pro
j
ections of nominalizations. We find that the difference is clearly
ev
i
de
nt in
:
(
35
)
John needs Bill’s su
pp
ort
which requires us to take
J
ohn as the ob
j
ect-Theme of support. Where there is an
affix that gives case, like -ing
,
then a sub
j
ect blocks ob
j
ect binding. Clark (1985)
ar
g
ued that control operates onl
y
on the sub
j
ect position, therefore the empt
y
ob
j
ect
C
HOMSKY
’
S
REMARKS
AND THE TRANSFORMATIONALIST HYPOTHESIS
S
135
m
ust move to the sub
j
ect position of
supporting
(listed as PRO to indicate control in
g
37b = 37a). The consequence is that a fill
e
d
sub
j
ect position as in (36) will block
o
b
j
ect control. However this does not apply to Williams-style thematic-binding
which allows the A
g
ent role of sp
ec
ifi
ed ve
r
bs
t
osee
k a
c
-
co
mman
d
THEM
E
r
o
l
e
in a
lowe
r th
e
mati
cdo
main
:
(36) *John needs Bill’s supporting
(37) a. John needs supportin
g
=
b. Joh
n
1
needs PR
O
1
supportin
g
t
1
In (38) the subject of
p
revention is filled, blocking control by another NP
disease
, which is possible when the subject is gone (39):
(38) *the disease calls for the city’s prevention
(39) the disease calls for prevention
The ob
j
ect control is also blocked when hi
g
her DP structure, above NP, is
i
nvoked as these examples reveal where plural entails more than a bare noun:
(40) the president needs thought (=thinking about him)
(41) the president needs thoughts (= he thinks)
W
e can see that ob
j
ect control is sensitive to the same thematic system because
where non-affected ob
j
ects occur, the
y
are excluded:
(43) the disease calls for knowledge
This does not specificall
y
mean knowled
g
e o
f
the diseas
e
but rather knowled
g
e
i
n
g
eneral. We find that a para
s
i
tic-
g
ap relation can also exist amon
g
n
ominalizations,
j
ust where Affected Ob
j
ects can occur:
(44) a. No taxation __ without representation __
b. Ideas should not undergo presentation without preparation
c. Ideas should not undergo presentation without preparations
A plural again generates an unspecified interpretation, where here the
preparations
m
eans more than preparin
g
the ideas.
(45) As for disease, prevention__ is better than curing__
1
36
T
O
M R
O
EPER
A
t a very subtle level we can see our argument reflected again:
(
46) *the disease’s prevention is better than curing.
(
46
)
fails because there is a lack of parallelism, since
curing
has no analogous
g
preposed form. However in this case, where covert movement invisibly converts
the
prevention of diseas
e
int
o
the disease’s
p
revention we achieve the
p
arallelis
m
needed:
(
47) the prevention of disease is better than its curin
g
.
W
e find again that nominalization provides a sharper representation of core
l
inguistic phenomena.
One criticism that might be made of this pr
o
posal is that it distorts thematic roles
o
r does not honor their traditional definition. Our approach here is different, though
we leave it as
j
ust a suggestion. Once again, if we take nominalizations to reflect
c
entral properties of
g
rammar, then we need to revise the theor
y
of thematic roles to
m
ake a sharp distinction between those entailed b
y
Events and another set of
s
emantic relations that are en
g
a
g
ed in a different manner for verbs like
know
.
It is
q
uite
p
ossible that the dee
p
est i
n
s
ights into Case will come from languages
with richer case systems. We have reason to believe that Case will also be
appropriately abstracted by seeing nominalizations as a core phenomenon. Chomsky
(lectures) once remarked that “modes of execution should be resolved into leading
i
deas” when we have a full understanding. What looks like a technical solution
s
hould have the
g
erm of a principle within it. If
g
rammar is in some sense a ‘perfect’
sy
stem, then all exceptionalit
y
should be re
s
olved in nominalizations
,
which is too
r
obust a phenomenon to be rele
g
ated to the peripher
y
. The capacit
y
of
n
o
minalizati
o
n
s
t
oco
n
ve
rt th
e
nominative
assigned by an auxiliary into a genitiv
e
s
hould follow from a
p
rinci
p
le
(
like transformation
)
which ca
p
tures other case-
c
onversion cases, like the capacity of affixes to convert dative into accusatives
(
be-
i
n German). Yet the insight is not yet there.
4
. INTRI
GU
IN
G
I
SSU
E
S
: A
S
PE
C
T
U
AL DIFFERENTIATI
O
N
O
F
NO
MINALIZATI
O
N AFFIXE
S
W
hat emerges from this analysis is that the process of derivational affixation
c
arries all the properties of syntax, the natural prediction that follows fro
m
C
homsky’s original claims. The fact that affixes change category should not
o
vershadow an important question for the fu
tu
r
e: w
hat
o
th
e
r inf
o
rmati
o
n i
sc
arri
ed
b
y
affixes? Wh
y
should there be more than one nominalizin
g
affix? We find there
ar
e
tran
s
iti
ve
affix
es:
-
in
g
,p
assive affixes -abilit
y
and ambi
g
uous ones: -
tion
,
-
ment
,
-
ence
.
(
48) John’s enlargement of the house