Syntax II - Class III
10c - now, even crossing one barrier gives you violation of subjacency.
last week, we abstracted entirely away from the VP, ujntil the end, at which p[oint realized that VP will cause us major trouble.
Whati [IP Bill [VP read ti]
here, Whati must cross VP and IP, so must not be any barriers on movement path, but looks as if there is in fact *two*. VP is both a blocking category and a Barrier. IP becomes both a blocking category and it inherits barrierhood fropm VP. So crosses *two* barriers.
Yet the sentence is perfectly grammatical. Must first LAND on edge of VP so that no longer dominated by VP and then move from there. That is:
Whati [Bill I [VP ti' [VP read ti
ti to ti'. ti' to Whati.
segments (these partial VPs) are never categories, and only categories may be barriers.
The sum total of VP (together) are a blocking category and barrier for ti. But antecedent of ti, that is ti', is not *excluded* by the VP, since it is partially *inside* the VP.
Similarly, ti' is not dominated by the upper VP, since does not entirely dominate it. Thus not a barrier to the antecedent Whati. The IP is a blocking category but presence of blocking category not sufficient to pose a problem for subjacency. The IP is not a barrier.
If had tried to get ti out in one fell swoop, would have had problems.
(Could have otherwise simply claimed that the VP is L-marked. "VC is not a BC or barrier for anything it dominates because VP is L-marked." This is not what Chomsky said. We would need to investigate 2 important ingredients:
1) I theta-marks VP.
2) I has to be or become lexical.
The idea that I theta-marks it VP complement is not a bad idea. It is in fact important point in the theory. Chomsky
John said that he would fix the car. and
"Fix the car, he did"
[VP fix the car]i he did ti
the trace must be properly governed.
in this particular governed, could make it antecedent governed. but there are other exmaples of "VP topicalization" that give you the feeling of subjacency effect and not an ECP effect which suggest theta-governed within its base position.
?? [VP fix the car]i, I wonder whether he will ti
not an ECP violation. therefore ti must be properly governed. can do antecedent or theta-governed. Can it be properly antecedent governed? No, because *whther* will definiately be in between. No point in adjoining to the VP because we are *moving* the entire VP. So cannot perform any intermediate landing, must go in one fell swoop across the CP. So *whether* should certainly give us a minimality effect. Subjacency violation like in 10c. But not so bad, because actually theta-governed. Not by "he." Will have to be "will," which is an I. Apparently I assigned theta-role and allows trace of fronted VP to satisfy the ECP such that it need not find a local antecedent.
So (1) does in fact seem to be true.
But (2), I does NOT seem to be lexical. Because not able to be generated by feature +-V and +-N. So I cannot BE lexical. But can *become* lexical by something lexical moving into it.
If want to add adverbial "probably" there, get "John read probably it."
Aux be (=copular be) and have. but modal verbs not. so would not be able to leave the VP for I. Perhaps as some point "read" moves from VP to I, and then retroactively marks the VP.
Whati [IP Bill I [VP ti' probably [VP read ti]
in languages where this is allowed, probably most likely raises to I. (ignoring comma intonation)
Perhaps in Lf syntactic component, read moves to I? without getting any phonological reflex of it.
Then at LF, would have lexical element, and at LF, would not be a barrier. But extracted the VP earlier than LF, at S-structure. If subjacency condition thought to be condition of LF representations, then OK, so long as L-marked by the end of LF. Generally, we say this is a condition on S-structure, overt representation. And if so, cannot undo the subjacency violation. Could imagine ECP violations like this, but not subjacency.
But if subjacency to be rethinkable as condition on LF, then we could in fact raise the "read" and lift the barrierhood.
In revisiting of theory in Lasnik and Chomsky in black book, "it is sufficient for a node to be selected for it to be a barrier." But it is essential for VP to be a barrier.
[Digression: Why does Chomsky need VP to be a barrier? Section 10 of the greenish-bluish book towards the end.
based on particular parasitic gap construction
Whoi ti [VP warned the men [CP Opi that they were about to arrest ti]
This is out. parasitic gap construction. two chains. Null Opi -ti chain which must be composed with "real" chain Whoi-ti, and the connection between the two chains must be established under zero subjacency = no barrier between the two chains.
If VP were not a barrier, then would be no barrier between the two chains (because CP is not a barrier here since it is L-marked). In order to be correctly ruled out, we have to assume that VP is in fact a barrier.
In 1993, forgets about this reason, for he resorts to simpler reason in terms of selection, such that VP not a barrier.
]
But is there any reason to assume this intermediate adjunction? Look at (21). 21a is unambiguous. 21b is ambiguous since John may be bound by himself.
At D-structure, himself can be antecednce governed by BIll. But for (b), need to
[how many pics of himselfi/j]k Johni [CP t'k Billj tk]
is extracted successive cyclicaly
"reconstructing" [how many pic of himself] back into initial trace ti to get Bill reading, or can reconstruct into John, at which point it we can get the John reading.
do this "reconstruction" at LF, but *only* if there was ti' where ti landed during movement. So in order to ensure the ambiguity, we must have that intermediate trace. It can take the matrix subject as its antecedent.
Interestingly, sentences of this type are grammatical if the Spec CP position occupied by WH phrsae
?[how many pics of himselfi/j]k Johni [CP WH Billj tk]
which gives us a subjacency violation, just mildly degraded, plus they are *still degraded.*
In 22b, which is minimal variant of 21b, somewhat degraded because of subjacency, but importantly it is still ambiguous.
If Bill was the only possibility and not ambiguous, Syntax I would have taken care of this.
How do we get the John reading in 22b. Not from a trace in post of spec Cp. We do get it on a more densely structured tree. There is a t'l in the embedded VP and a t''k adjoined to the matrix VP.
?[how many pics of himselfi/j]k Johni [VP t''k [VP [CP WH Billj [t'k tk]
The two traces below Bill will give is the Bill reading. But to get John reading, can reconstruct to t''k. But to do this it is *essential* we do intermediate adjunction to VP.
page 8 on the handout.
this sugegcts strongly that intermediate adjunction is in factg the case.
Of course 22b does violate subjacency condition in going from tk' to t''k. IP a BC. CP is l-marked so not BC, but inherits barrierhood from IP. VP we need not worry about because t''k in VP. So get weak subjacency violation.
But, Question. Why could not intermediately adjoin to IP before extracting out of IP and then move to VP. In notes as well.
[VP t'''k [VP V [CP WH [IP t''k [IP Su I [VP t'k [VP ... tk
another possibility is to adjoin to *CP* on the way up:
[VP t'''k [VP V [CP t''k [CP WH [IP Su I [VP t'k [VP ... tk
neither of which would give us a subjacency violation. how so:
[VP t'''k [VP V [CP WH [IP t''k [IP Su I [VP t'k [VP ... tk
tk we need not worry about because only separated by segment of VP.
the IP together is BC, but t''k no problem because within IP.
t''k only dominated by segment of IP. CP is l-marked, and not BC. and CP cannot inherit barrierhood from IP in this case. and VP in total dominates, but t'''k is within VP.
If so, we are in dire straits, because we do want to show that spec, CP being occupied gives us a WH-island violation.
Therefore, we have to *declare* this derivation illigitimate, because of something else. Otherwise don;t understand why violates subjacency.
Similarly for the second derivation.
[VP t'''k [VP V [CP t''k [CP WH [IP Su I [VP t'k [VP ... tk
VP is only a segment.
IP is a BC. CP inherits barrierhood but t''k is inside of the CP.
And childs play that t''k has no division from its antecedent. VP is a segment and VP is only a segment. Must find a way from outside to block this.
Thus, make rules (27) and (28): adjunction of wh-constituents to IP is illegitimate. And adjunction to arguments is illegitimate.
I think that Maryi, Bill really loves ti.
example of topicalization.
Where could Mary be sitting. no space whatsoever except via adjunction to IP. So adjunction to IP exists.
And transformation of adjunction to IP exists.
* Who do you think that kissed Mary
standard reason bad because that-trace effect C in between t' and t
but imagine were allowed to intermediately adjoin on the way out of IP.
t' in other IP governed as *locally* as possible, and from there to t''. Thus antecedent government should work. The other traces will be eliminated at LF.
so another reason intermediate adjunction not allowed for IP extraction.
(27) is simply a blunt stipulation.
But why is it forbidden. We do not really know. And we really don't know. Luckily not just for one case, but for two. But still really bad. (Could also move out * Who do you think that kissed Mary via vacuous movement hypothesis, discussed on page 9. In which case one stone kills one bird)
If 28 *were* a brute stipulation, would be bad. But can be understood in a relatively profound way and is useful elsewhere.
Can be made to follow from theta-criterion/projection principle.
If I went to Starbucks and had ingredients of cappaccino but also whipped cream, cinnamon, etc., then said I did not order this, I order a cappaccino.
If VP ordered a CP complement, and gets back CP plus a bunch of other stuff, will not be happy. Wanted lower CP but got upper CP as well. Now Verb cannot select the lower CP segment anymore. Buried under whipped cream which it did not want.
Thus, adjunction to arguments is illegitimate.
Projection principle requires that theta-marking must engage not only at D-structure, but even at S-structure.
Not only is 28 not a stipulation that cannot be reduced to something understandable, also useful elsewhere in the theory.
* (30) Who did [NP pictures of t] cause a scandal
[CP Whoi C(did) [IP [NP pix [PP of ti]] I'
in Syntax I, subject island violation
NP is not l-marked and thus BC for ti, and thus also barrier for ti.
The IP is BC. Since NP is BC, IP is also a barrier for ti. And also on the movement path. And gives us a *heavy* subjacency violation.
Now, imagine that ti was not the only trace. Suppose allowed to adjoin to NP on the way out of the IP.
[CP Whoi C(did) [IP [NP t'i [NP pix [PP of ti]]] I'
Now, ti has antecedent in t'i. And t' separated from antecedent by IP which is BC but not a barrier. So this should be completely good.
How do we block this derivation.
Can block this via (28) which says that you cannot adjoin to arguments.
So (28) also needed for (30).
And needed also for (29), extracting out of a Noun phrase.
29a:
?? Who did John make [NP the claim [CP that Bill harassed t
=
?? Whoi ... [VP t''' [VP make [NP the claim [CP t''i [that Bill [VP t'i [VP harassed ti
ti-t'i good
t'i-t''i good
but t''i-Whoi would be problematic.
However, t'''i-t''i would make it OK!
since CP l-marked by claim, algorithm would say it is not a barrier.
Chomsky: "the CP in the complement of N is a barrier."
So CP is a barrier but not a BC. Therefore the NP cannot inherit barrierhood from CP, since CP is not a barrier.
29b is worse. here extracting from relative clause. relative clause is not l-marked.
Whoi ... [VP t''' [VP make [NP the claim [CP Opj [that Bill [VP t'i [VP V tj to ti
that Bill had denied the claim to Renquist.
the claim now is being relativized.
ti to t'i and from there to t'''i.
Here, the IP is BC, which gives barrierhood to CP. (The CP is a relative clause, which are NOT complements - they are *adjoined* to N'. That means these are not L-marked and are thus inherent barriers. CP a BC. Thus, NP inherits barrierhood from CP. Very severe violation.
But imagine could adjoin to the CP or to the NP. Would give you just one barrier crossed. Yet this is too bad for that. But NP is complement of V make, so ruled out by 28.
One more thing need rule out. Why cannot adjoin to CP on the way out of the CP? For CP is not an argument, for it is a relative clause.
Must do something to (28), broaden it, so that it covers this as well.
(28) is about banning adjunction to things that receive theta-roles. Relative clauses do not *receive* theta roles, but are standardly assumed to be *predicates* of the noun phrases connected to, and thus they *assign* theta-roles.
"Adjunction to thematically active material is illegitimate."
rationale, the projection principle, will, continue to cover this extension.
If something want to assign theta-role to a subject, ...
last week, we abstracted entirely away from the VP, ujntil the end, at which p[oint realized that VP will cause us major trouble.
Whati [IP Bill [VP read ti]
here, Whati must cross VP and IP, so must not be any barriers on movement path, but looks as if there is in fact *two*. VP is both a blocking category and a Barrier. IP becomes both a blocking category and it inherits barrierhood fropm VP. So crosses *two* barriers.
Yet the sentence is perfectly grammatical. Must first LAND on edge of VP so that no longer dominated by VP and then move from there. That is:
Whati [Bill I [VP ti' [VP read ti
ti to ti'. ti' to Whati.
segments (these partial VPs) are never categories, and only categories may be barriers.
The sum total of VP (together) are a blocking category and barrier for ti. But antecedent of ti, that is ti', is not *excluded* by the VP, since it is partially *inside* the VP.
Similarly, ti' is not dominated by the upper VP, since does not entirely dominate it. Thus not a barrier to the antecedent Whati. The IP is a blocking category but presence of blocking category not sufficient to pose a problem for subjacency. The IP is not a barrier.
If had tried to get ti out in one fell swoop, would have had problems.
(Could have otherwise simply claimed that the VP is L-marked. "VC is not a BC or barrier for anything it dominates because VP is L-marked." This is not what Chomsky said. We would need to investigate 2 important ingredients:
1) I theta-marks VP.
2) I has to be or become lexical.
The idea that I theta-marks it VP complement is not a bad idea. It is in fact important point in the theory. Chomsky
John said that he would fix the car. and
"Fix the car, he did"
[VP fix the car]i he did ti
the trace must be properly governed.
in this particular governed, could make it antecedent governed. but there are other exmaples of "VP topicalization" that give you the feeling of subjacency effect and not an ECP effect which suggest theta-governed within its base position.
?? [VP fix the car]i, I wonder whether he will ti
not an ECP violation. therefore ti must be properly governed. can do antecedent or theta-governed. Can it be properly antecedent governed? No, because *whther* will definiately be in between. No point in adjoining to the VP because we are *moving* the entire VP. So cannot perform any intermediate landing, must go in one fell swoop across the CP. So *whether* should certainly give us a minimality effect. Subjacency violation like in 10c. But not so bad, because actually theta-governed. Not by "he." Will have to be "will," which is an I. Apparently I assigned theta-role and allows trace of fronted VP to satisfy the ECP such that it need not find a local antecedent.
So (1) does in fact seem to be true.
But (2), I does NOT seem to be lexical. Because not able to be generated by feature +-V and +-N. So I cannot BE lexical. But can *become* lexical by something lexical moving into it.
If want to add adverbial "probably" there, get "John read probably it."
Aux be (=copular be) and have. but modal verbs not. so would not be able to leave the VP for I. Perhaps as some point "read" moves from VP to I, and then retroactively marks the VP.
Whati [IP Bill I [VP ti' probably [VP read ti]
in languages where this is allowed, probably most likely raises to I. (ignoring comma intonation)
Perhaps in Lf syntactic component, read moves to I? without getting any phonological reflex of it.
Then at LF, would have lexical element, and at LF, would not be a barrier. But extracted the VP earlier than LF, at S-structure. If subjacency condition thought to be condition of LF representations, then OK, so long as L-marked by the end of LF. Generally, we say this is a condition on S-structure, overt representation. And if so, cannot undo the subjacency violation. Could imagine ECP violations like this, but not subjacency.
But if subjacency to be rethinkable as condition on LF, then we could in fact raise the "read" and lift the barrierhood.
In revisiting of theory in Lasnik and Chomsky in black book, "it is sufficient for a node to be selected for it to be a barrier." But it is essential for VP to be a barrier.
[Digression: Why does Chomsky need VP to be a barrier? Section 10 of the greenish-bluish book towards the end.
based on particular parasitic gap construction
Whoi ti [VP warned the men [CP Opi that they were about to arrest ti]
This is out. parasitic gap construction. two chains. Null Opi -ti chain which must be composed with "real" chain Whoi-ti, and the connection between the two chains must be established under zero subjacency = no barrier between the two chains.
If VP were not a barrier, then would be no barrier between the two chains (because CP is not a barrier here since it is L-marked). In order to be correctly ruled out, we have to assume that VP is in fact a barrier.
In 1993, forgets about this reason, for he resorts to simpler reason in terms of selection, such that VP not a barrier.
]
But is there any reason to assume this intermediate adjunction? Look at (21). 21a is unambiguous. 21b is ambiguous since John may be bound by himself.
At D-structure, himself can be antecednce governed by BIll. But for (b), need to
[how many pics of himselfi/j]k Johni [CP t'k Billj tk]
is extracted successive cyclicaly
"reconstructing" [how many pic of himself] back into initial trace ti to get Bill reading, or can reconstruct into John, at which point it we can get the John reading.
do this "reconstruction" at LF, but *only* if there was ti' where ti landed during movement. So in order to ensure the ambiguity, we must have that intermediate trace. It can take the matrix subject as its antecedent.
Interestingly, sentences of this type are grammatical if the Spec CP position occupied by WH phrsae
?[how many pics of himselfi/j]k Johni [CP WH Billj tk]
which gives us a subjacency violation, just mildly degraded, plus they are *still degraded.*
In 22b, which is minimal variant of 21b, somewhat degraded because of subjacency, but importantly it is still ambiguous.
If Bill was the only possibility and not ambiguous, Syntax I would have taken care of this.
How do we get the John reading in 22b. Not from a trace in post of spec Cp. We do get it on a more densely structured tree. There is a t'l in the embedded VP and a t''k adjoined to the matrix VP.
?[how many pics of himselfi/j]k Johni [VP t''k [VP [CP WH Billj [t'k tk]
The two traces below Bill will give is the Bill reading. But to get John reading, can reconstruct to t''k. But to do this it is *essential* we do intermediate adjunction to VP.
page 8 on the handout.
this sugegcts strongly that intermediate adjunction is in factg the case.
Of course 22b does violate subjacency condition in going from tk' to t''k. IP a BC. CP is l-marked so not BC, but inherits barrierhood from IP. VP we need not worry about because t''k in VP. So get weak subjacency violation.
But, Question. Why could not intermediately adjoin to IP before extracting out of IP and then move to VP. In notes as well.
[VP t'''k [VP V [CP WH [IP t''k [IP Su I [VP t'k [VP ... tk
another possibility is to adjoin to *CP* on the way up:
[VP t'''k [VP V [CP t''k [CP WH [IP Su I [VP t'k [VP ... tk
neither of which would give us a subjacency violation. how so:
[VP t'''k [VP V [CP WH [IP t''k [IP Su I [VP t'k [VP ... tk
tk we need not worry about because only separated by segment of VP.
the IP together is BC, but t''k no problem because within IP.
t''k only dominated by segment of IP. CP is l-marked, and not BC. and CP cannot inherit barrierhood from IP in this case. and VP in total dominates, but t'''k is within VP.
If so, we are in dire straits, because we do want to show that spec, CP being occupied gives us a WH-island violation.
Therefore, we have to *declare* this derivation illigitimate, because of something else. Otherwise don;t understand why violates subjacency.
Similarly for the second derivation.
[VP t'''k [VP V [CP t''k [CP WH [IP Su I [VP t'k [VP ... tk
VP is only a segment.
IP is a BC. CP inherits barrierhood but t''k is inside of the CP.
And childs play that t''k has no division from its antecedent. VP is a segment and VP is only a segment. Must find a way from outside to block this.
Thus, make rules (27) and (28): adjunction of wh-constituents to IP is illegitimate. And adjunction to arguments is illegitimate.
I think that Maryi, Bill really loves ti.
example of topicalization.
Where could Mary be sitting. no space whatsoever except via adjunction to IP. So adjunction to IP exists.
And transformation of adjunction to IP exists.
* Who do you think that kissed Mary
standard reason bad because that-trace effect C in between t' and t
but imagine were allowed to intermediately adjoin on the way out of IP.
t' in other IP governed as *locally* as possible, and from there to t''. Thus antecedent government should work. The other traces will be eliminated at LF.
so another reason intermediate adjunction not allowed for IP extraction.
(27) is simply a blunt stipulation.
But why is it forbidden. We do not really know. And we really don't know. Luckily not just for one case, but for two. But still really bad. (Could also move out * Who do you think that kissed Mary via vacuous movement hypothesis, discussed on page 9. In which case one stone kills one bird)
If 28 *were* a brute stipulation, would be bad. But can be understood in a relatively profound way and is useful elsewhere.
Can be made to follow from theta-criterion/projection principle.
If I went to Starbucks and had ingredients of cappaccino but also whipped cream, cinnamon, etc., then said I did not order this, I order a cappaccino.
If VP ordered a CP complement, and gets back CP plus a bunch of other stuff, will not be happy. Wanted lower CP but got upper CP as well. Now Verb cannot select the lower CP segment anymore. Buried under whipped cream which it did not want.
Thus, adjunction to arguments is illegitimate.
Projection principle requires that theta-marking must engage not only at D-structure, but even at S-structure.
Not only is 28 not a stipulation that cannot be reduced to something understandable, also useful elsewhere in the theory.
* (30) Who did [NP pictures of t] cause a scandal
[CP Whoi C(did) [IP [NP pix [PP of ti]] I'
in Syntax I, subject island violation
NP is not l-marked and thus BC for ti, and thus also barrier for ti.
The IP is BC. Since NP is BC, IP is also a barrier for ti. And also on the movement path. And gives us a *heavy* subjacency violation.
Now, imagine that ti was not the only trace. Suppose allowed to adjoin to NP on the way out of the IP.
[CP Whoi C(did) [IP [NP t'i [NP pix [PP of ti]]] I'
Now, ti has antecedent in t'i. And t' separated from antecedent by IP which is BC but not a barrier. So this should be completely good.
How do we block this derivation.
Can block this via (28) which says that you cannot adjoin to arguments.
So (28) also needed for (30).
And needed also for (29), extracting out of a Noun phrase.
29a:
?? Who did John make [NP the claim [CP that Bill harassed t
=
?? Whoi ... [VP t''' [VP make [NP the claim [CP t''i [that Bill [VP t'i [VP harassed ti
ti-t'i good
t'i-t''i good
but t''i-Whoi would be problematic.
However, t'''i-t''i would make it OK!
since CP l-marked by claim, algorithm would say it is not a barrier.
Chomsky: "the CP in the complement of N is a barrier."
So CP is a barrier but not a BC. Therefore the NP cannot inherit barrierhood from CP, since CP is not a barrier.
29b is worse. here extracting from relative clause. relative clause is not l-marked.
Whoi ... [VP t''' [VP make [NP the claim [CP Opj [that Bill [VP t'i [VP V tj to ti
that Bill had denied the claim to Renquist.
the claim now is being relativized.
ti to t'i and from there to t'''i.
Here, the IP is BC, which gives barrierhood to CP. (The CP is a relative clause, which are NOT complements - they are *adjoined* to N'. That means these are not L-marked and are thus inherent barriers. CP a BC. Thus, NP inherits barrierhood from CP. Very severe violation.
But imagine could adjoin to the CP or to the NP. Would give you just one barrier crossed. Yet this is too bad for that. But NP is complement of V make, so ruled out by 28.
One more thing need rule out. Why cannot adjoin to CP on the way out of the CP? For CP is not an argument, for it is a relative clause.
Must do something to (28), broaden it, so that it covers this as well.
(28) is about banning adjunction to things that receive theta-roles. Relative clauses do not *receive* theta roles, but are standardly assumed to be *predicates* of the noun phrases connected to, and thus they *assign* theta-roles.
"Adjunction to thematically active material is illegitimate."
rationale, the projection principle, will, continue to cover this extension.
If something want to assign theta-role to a subject, ...
posted by joshwaxman at 7:54 AM
0 Comments:
Post a Comment
<< Home