Fodor & decomposition
Heidi Harley, Ling/Phil 596D
February 5, 2004
Fodor 1971: Three reasons for not analyzing kill as cause to die are not
reasons not to analyze ÔmeltÕ as 'make meltedÕ
1 Fodor's
paper:
1. Some
reasons to believe in syntactic lexical decomposition: pronominal reference to
the embedded event:
a. Floyd
melted the glass, and it surprised me.
b.
Floyd
melted the glass, and I was surprised that he/it would do so.
These sentences
purport to show that there are two subevents in the verb melt, with which the pronoun or do so VP can be coreferential Ñ ei = causing (to melt) and e2
= melting
2. However,
with "kill" as "cause to die", such pronominal reference
isn't possible.
a.
(overt "cause" morpheme):
John
caused Mary to die, and it surprised me that he/she did so.
b.
("kill")
John
killed Mary, and it surprised me that he/*she did so.
3. Why
should "kill" and "melt" behave differently in this
respect?
ˆ at the end of the paper, Fodor suggests
that itÕs because meltintr
is homophonous
with melttr,
and the conditions on pronominalization and deletion care about such homophony.
ˆ Problem: it seems quite well supported
that structural identity is what matters for deletion, although there are also interesting restrictions having to do
with auxiliary-homophony, see LasnikÕs recent work on the matter.
ˆ if we adopt a sensible version of the
Ôcause [X melt]Õ story for melt,
we can keep the structural identity restriction, have a good story for the
pronominal reference, and answer
all of FÕs Ôthree reasonsÕ without really any trouble at all.
4. So
why does F argue that decomposition is inappropriate even for "melt"?
Melttr behaves differently from
its clausal paraphrase in two respects.
Consider
first:
a. Floyd
caused the glass to melt on Sunday
by heating it on Saturday.
b. Floyd
melted the glass on Sunday by heating it on Saturday.
When
the verb is lexical, there is only one event, says Fodor, which may not be temporally distinct.
But when you use "cause to X', there are two events, an event of causing
and a result event, which may be temporally distinct. Hence, says Fodor,
"cause to X" is not part of the representation of "melt"
(or "kill")
5. "subordinated
sentoids" !!
6. Second,
subjects control instrumental adverbials:
a. John
contacted Mary by using the telephone. (John used it, not Mary)
b. John
caused Bill to die by swallowing his tongue. (Bill swallowed Bill's t.)
c. *John
killed Bill by swallowing his tongue. (*Bill swallow Bill's t.)
Hence, causative
verbs are not derived by "collapsing" the corresponding causative
sentences.
7. A version of a modern proposal, taken
from Kratzer 1993:
(After a long excursus on the semantics of idioms and their
interpretation, more next week) Kratzer presents evidence from Malagasy that
external arguments are introduced by a separate morpheme from the main verb, an-.
a.
She concludes that the separate morpheme introducing external arguments is not
a verb itself, but rather a functional projection, VoiceP. For us, it'll be
equivalent to CAUSE.
b.
The structure for "Mittie fed the dog" then, will be the following:
VoiceP<t>
![]()
DP<e> Voice'
![]()
![]()
Mittie Voice VP<e,t>
CAUSE V<?> DP<e>
![]()
![]()
feed the
dog
This way,
"Mittie" is not an argument of "feed", but an argument of
"cause". The semantic representation of (b) is (c):
c. [
Cause(Mittie, event) & Fed(the dog, event)]
which can be
read, roughly, as "There was an event that Mittie caused, and it was an
event of feeding the dog."
How can such a
proposal be saved from Fodor's objections?
8. If event arguments must be bound by Tense, one significant difference
between something like 2a and 2b is the number of Tense nodes that are
available in the syntax:
John
caused (Tpast)) Bill to die (TInf) on Saturday by
poisoning him on Friday.
John
killed (TPast) Bill on Saturday by poisoning him on Friday.
This will ensure
that if there are two events in kill (or melt)
Ñ a causing event and a dying event Ñ they have to be cotemporaneous, because
bound by the same Tense operator. Ditto for Kratzer's treatment of feed.
Another way of
saying this is that a temporal adverbial must compose with (adjoin to) a tense
phrase (or possibly an event-phrase whose temporal reference is fixed by
binding by Tense). Only one tense operator, only one tense adverbial allowed!
(see Demirdache & Uribe-ExtebarriaÕs recent work.)
Te
[CAUSE(Mittie, ei) & Fed (the dog, ei)]
This takes care
of the temporal adverbial problem
(Alternatively,
if event arguments are bound by an existential operator at vP, they will have
to be the same event
in John killed Bill
(only one operator) but could be different events in John caused Bill to die
(two operators)).
9. An
aside: in fact, with an adverbial that can appropriately modify either event,
in fact you can get two readings for a sentence like He emptied the bottle
again, as F notes in a
rather cute counterargument against treating the temporal adverbial problem as
an effect of ordering adverb-adjunction with the lexicalization treatment (his
example is slowly but
itÕs the same point).
AGAIN[Cause
(He, e) & empty (the bottle, e)] or
[Cause
(He, e) & AGAIN[empty (the bottle, e)]
That is, either he twice caused the bottle to be empty, or
Mary (for instance) caused it to be empty once, and then it was refilled and he
emptied it again.
10. What about the
instrumental adverbials? If
a. John
killed Bill by swallowing his tongue
is like this:
Johni
CAUSE Billj DIE [PP by PROi/*j swallowing hisi/j
tongue].
This is still a
puzzle. The question is, why can't Bill be the one swallowing his tongue here?
That is, why does John have to control PRO in (a), but not in (b) below?
b. Johni caused Billj to die by PROi/j swallowing hisi/j tongue?
Again, this
seems fairly straightforward. You have to make a couple of assumptions about
control theory. Assume that PRO has to be controlled by the closest available
c-commanding subject. Further, assume that by PPs have to attach to vP (the
equivalent of VP in older notation).
Note
that this makes sense if a) eventiveness comes from v and b) by PPs must semantically modify an event,
not a state.
c. John
died/*was dead by swallowing his tongue.
(This in fact is
a main difference between adjectival & eventive passives.)
If that's true,
the difference between 10a) and 10b) just comes down to standard structural
ambiguity, which we're agonizingly familiar with from such examples as Mary
watched the man with the telescope.
The structure of a) is below, and the two possible structures for b) are below:
(I haven't illustrated it as a vP structure but you can draw it yourselves as
an exercise).
![]()
a. TP
![]()
DP T'
![]()
![]()
Johni T VP
![]()
![]()
![]()
[+pst] VP PP
![]()
![]()
V DP P VP
![]()
![]()
![]()
![]()
killed Bill by DP V'
PROi/*j V DP
![]()
![]()
swallowing hisi/j tongue
![]()
b'. TP
![]()
DP T'
![]()
![]()
Johni T VP
![]()
![]()
![]()
[+pst] VP PP
![]()
![]()
![]()
V IP P VP
![]()
![]()
![]()
![]()
![]()
![]()
![]()
caused DP I' by DP V'
Bill I VP PROi/*j V DP
![]()
![]()
![]()
to V swallowing hisi/j tongue
die
![]()
b''. TP
![]()
DP T'
![]()
![]()
Johni T VP
![]()
[+pst] V IP
![]()
![]()
![]()
caused DP I'
![]()
Bill I VP
![]()
to VP PP
![]()
![]()
V P VP
![]()
![]()
![]()
![]()
die by DP V'
![]()
PROj/*i V DP
swallowing his
tongue
in a), there's only one subject/SUBJECT
in Spec-IP, and if control is sensitive to that relation, which it undoubtedly
is then that's why Bill
is not available as a possible controller for PRO. In b), given that there's two possible
attachment sites for the by
phrase, there's two possible controllers.
Remaining
questions:
Why do melt and kill differ w/r to pronominal reference
possibilities (1st ex) ?
Pronouns can
refer back to a state...
11. Moreley was
dead, and it made Scrooge grumpy.
Melt probably encodes an endstate (being
melted) which can be coreferred to with the same verb; I suspect that kill doesn't (so for kill, Fodor's right). Kill is a verb like run,
hop, sneeze, push, etc.
-- it has a zero-derived event-naming nominal form. Arguably, kill has a structure like H&K's structure
for unergative verbs, with an incorporated N root. Consequently, no embedded
state for it to refer
to.
In fact, the
real problem that Fodor sees with lexical decomposition doesn't have to do with
empirical arguments like the above, but rather has to do with a more fundamental
question: if CAUSE doesn't mean cause (which takes a sentential complement), what does it mean?
In what way is it explanatory to posit a head whose contribution to the
semantics is imprecise? (This is the "plus-X" problem).
In the ideal
world, we'd be able to show exact paraphrase equivalents. Since Kratzer's
representation is essentially a causing event + a state, you'd think the
semantics of melt
should be close the semantics of make X liquid, or any other make [state] structure.
a) *John
made Mary sick on Tuesday by poisoning her on Monday.
b) John
made Mary sick, and it surprised me that he did so.
*John
made Mary sick, and it surprised me that she did so.
If this is right, the v paraphrased as 'cause' with a clausal complement really means something more like 'make' with a stative complement.