ON THE CONDITION OF ADJUNCTION IN BARRIERS

This paper reconsiders Chomsky's Adjunction Condition and suggests some modification of the theory of barriers to solve problems arising mainly from the improper definition of the blocking category and the barrier. It is argued that in S-structure, there is no adjunction in movement except extraposition and topicalization. First, it is suggested that the adjunction conjunction is too powerful in that it allows many ungrammatical sentences. To solve the problems, the assumption was made that there is no adjunction at S-structure except extraposition and topicalization. Next, only inherited barriers were assumed, which is to consider IP, traditionally a defective category, as a normal lexical category. In fact, under the modified barrier theory, subjacency condition can be said to be stronger than it.was in Chomsky (1986). In this study, the modified theory was applied to every sentence that had raised problems for barriers and showed that the new definition works well. (JL) *********************************************************************** Reproductions supplied by EDRS are the best that can be made from the original document. *********************************************************************** U.S. DEPARTMENT OF EDUCATION Off ica of Educational Research and Improvement EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC) NIDThis document has been reproduced as received from the person or organization originating it. O Minor changes have been made to improve reproduction quality. Points of view or opinions stated in this document do not necessarily represent official OERI position or policy.


Introduction
In this paper, I reconsider Chomsky's 'Adjunction Condition (Chomsky 1986:p.6)'and suggest some modification of the theory of barriers to solve some problems arising mainly from the improper definition in Chomsky's barriers.In short, I argue that in S-structure, there is no adjunction in movement except extraposition and topicalization.To support this hypothesis, I assume only inherited barriers.
Problems in Adjunction Condition Chomsky(1986:p.6)suggests the following condition on adjunction which allows adjunction to VP but not to CP are NP; (1) Adjunction Condition Adjunction is possible only to a maximal projection that is a nonargument.
Under the condition, adjunction to an argument NP or CP is not allowed, while adjunction to a nonargtunent VP is allowed.Then let's examine whether adjunction to other nonargwnent maximal projections is possible or not.
Kansas Working Papers in linguistics, Volume 20, 1995, pp.69-83 ry ti (2) has been traditionally considered to violate the Subjacency condition.When we first move 'who', the result will be (3); (3) you wonder [cp who i [c' [+WH] When 'what' is moved in (3), it is first adjoined to the VP in the embedded clause and then to the VP of the main clause, and finally moves to the Specifier position of the main clause.The process is shown in ( 4 Here the problem arises when movement occurs from t1 j to t2 j.In this step the blocking category which dominates t1 j is not VP but IP, and the CP dominating the IP is a barrier.The movement is 1-subjacent because it crosses over one barrier.Actually this prediction contrasts with that of LGB in which it is judged as the violation of the Subjacency.However, when 'what' is moved, if it is first adjoined to VP and then to the IP of the embedded sentence instead of the VP of the higher sentence, the result will be different. (5) [cp what j [c' do k [ip you t k [vp t3 j [vp wonder [cp who [c' e [ip t2 j [ip t [vp t1 j [vp

see t j ]]]]]]]]]]
The movement from t1 j to t2 j is not problematic because no barrier intervened.Look at t2 j carefully.The IP of the embedded sentence is not a blocking category for t2 j because the IP doesn't exclude t2 j .The CP is L-marked by wonder', so CP is not a blocking category.Therefore there is no barrier between t2 j and t3 j and the movement from t2 j to t3 j obeys Subjacency and the ECP.However this derivation must be barred because (2) is not grammatical.That is, in syntactic movement we should allow adjunction to VP but we should not allow adjunction to IP.However, it is ad hoc to maintain that adjunction to nonargument VP is allowed and adjunction to a non-argument IP is barred.In addition, Lasnik and Saito analyze topicalization be a case of IP adjunction and if they are correct, barring IP adjunction is not plausible.
Even assuming a condition which prevents IP-adjunction, there is another derivation which can allow legal derivation.That is, if 'what' adjoins to the VP of the matrix sentence after 'what' is adjoined to the embedded VP, then it crosses over one barrier, CP.In this case to avoid the barrier, we can assume a derivation in which 'what' is adjoined to the Spec of CP .Then, CP does not exclude 'what' and this derivation has the same effect as it has whim 'what' moves through the Spec of CP.Therefore, the sentence should be grammatical.As May(1985) shows, the element adjoined to the Spec of CP c-commands its trace under the definition of the domination by exclusion.To prevent adjunction to CP we cannot say that because who' in Spec of CP is an argument, adjunction to CP is not permitted.This is because the violation of WH-island constraint is more serious when an adjunct is placed in the Spec of CP.Therefore, adjunction to the Spec of CP must be prevented Adjunct condition: The following sentence violates the adjunct condition.Although Chomsky(1986) says that it can be explained in terms of barriers, he doesn't show the process clearly.In (7), PP is not L-marked, so it is a blocking category and becomes a barrier.IP is also a blocking category and a barrier because it immediately dominates PP.Therefore the movement of 'who' crosses over two barriers which means a violation of subjacency and the sentence is ungrammatical.In this case the internal structure of PP is [pp before [cp t3 [ip PRO [vp t2 [vp interviewing t1 M]].When t3 moves to the Spec of CP, it crosses over two barriers PP and IP.However, there is evidence against this structure.
(8) John said he would leave the party before speaking to anyone, and leave the party before speaking to anyone he certainly did.
In 8), the PP (before -phrase) is a constituent of VP because it is moved forward together with VP by VP-preposing.In (8b) the PP is not the sister of V because the 'do so ' substitution doesn't take place to include the PP.Therefore it is not correct to analyze PP to be immediately dominated by IP.The data below show that PP should be immediately dominated by VP (Tiedeman 1989:19-20).
(9) VP-preposing a. Bill said that John would go home before he saw Mary, and go home before he saw Mary he did.
b.* Bill said that John would go home before he saw Mary, and go home he did before he saw Mary.
(10) WH-cleft a. John went home before he saw Mary.b.What John did was go home before he saw Mary.
(11) VP-deletion John went home before he saw Mary and Bill did too.
Therefore, if PP is within VP in (6), then when a constituent moves from the Spec of the CP, complement of P, it can be adjoined to the higher VP instead of being adjoined to the Spec of higher CP.If this adjunction to VP is possible, the movement crosses over only one barrier, PP.Then the movement has a 1subjacency which is a weak violation of subjacency and the sentence must be judged marginal.In addition, there is another possible derivation which is a more serious problem to the theory.If we accept Chomsky(1986)'s adjunction condition which always allows adjunction to a non-argument position, then there is no way to prevent adjunction to PP.Then, the movement from the embedded [SPEC, CP] to adjoine to PP crosses over no barrier, and when the constituent moves from the adjoined PP to adjoin to the VP of the higher sentence, it doesn't cross over any barrier.Of course, this derivation does not explain the ungrammaticality of the sentence Complex NP constraint phenomenon : Another problem found in Chomsky(1986) is that it can not explain the Compleic NP Constraint phenomenon(CNPC) properly.In general, extraction from a complex NP is ungrammatical.
(12) a. * Which book did John meet [np a child [cp who read t ]] ?
b. ?? Which girl did Tom hear [np the rumor [cp that he met ti]] ?
The structure of the above complex NP's is ( 13 When we extract from the relative clause (13a), it crosses over two barriers CP and NP, which is the violation of subjacency and the sentence is correctly judged as ungrammatical.In (13b) CP is not a blocking category because it is subcategorized and L-marked as a complement of N. NP is not a blocking category either because it is L-marked by V. Therefore, extraction of the WHphrase from the noun complement clause should be permitted.But ( 12b) is marginal at best.To solve this problem, Chomsky(1986) maintains that the CP which receives oblique case from N as a complement of N is an intrinsic bather.
In addition to this, he says that even though the CP which receives oblique case from N is a barrier, it is not a blocking category, so the NP which immediately dominates the CP cannot be an inherited barrier according to the definition of barrier (Chomsky 1986:14).Therefore extraction from a noun complement clause weakly violates subjacency.Even though this explanation, predicts the difference of grammaticality of the two types of sentence, there arise many problems.First, "why does the CP which receives oblique case become a barrier ?" and "is it possible for a sentence to receive a case(cf.Stowell 1981) ?"Second, a relative clause is not subcategorized and a noun complement clause is subcategorized.If wt.. follow the adjunction condition (1), adjunction to the relative clause should be allowed, while adjunction to the noun complement clause should be prevented.
However, if we permit adjunction to a relative clause, then extraction from a relative clause crosses over no barrier.A noun complement clause subcategorized by N and so no adjunction is allowed.

NP-movement :
In the case of WH-movement, a WH-object adjoins first to VP to avoid the barrierhood of VP and then moves to the Spec of CP.However, in the case of NP-movement, an NP cannot adjoin to VP.If adjunction to VP occurs, the resulting chain cannot be licensed as an argument chain.Therefore, for NP-movement, no adjunction is allowed.Consider the following sentence; (14)  If VP-adjunction is barred in NP-movement, the movement from t i to 'Mary' in (14) comes to cross aver one barrier, VP.To solve this problem Chomsky(1986) introduces 'feature sharing.In ( 14) 'Mary i ' and t i are coindexed by movement ,and [seem-I] j is also coindexed with t j .In addition, 'Mary' and [seem-I] are coindexed by Spec-head agreement,so that i = j.Then t j can antecedent-govern t i because there is no barrier between them and both of them have the same index.In this case, to allow the antecedent-government between two elements, we combine two separate chains into one extended chain (Chomsky 1986:74).
(15) Extended Chain a. 8 = ( al , , an , ) is an extended chain if ( al , an ) is a chain with index i and 0 has index 't b.Chain coindexing holds of the links of an extended chain.
According to the condition in ( 15), an extended chain is licensed when two chains are coindexed and when subjacency and antecedent-government are obeyed between an and 13.This assumption explains the following example (Chomsky 1986:75); (16) a. * Mary seems that it appears to be smart.In the above sentence, the index of the chain ( Mary, t) is shared with seem' after V-raising in the matrix sentence; however the trace of 'seem' , t j , does not govern t i, so the latter yields an ECP violation.That is, the antecedent- government condition is not obeyed because of the barrier CP between t i and t j, so that the sentence is judged ungrammatical.Now consider a simple passive sentence.In this structure VP intervenes in the movement while in the raising structure there is an IP intervening.There is no problem in antecedent-government in the raising structure because IP is not a barrier; however in the passive structure, VP, which is a baffler, intervenes and prevents antecedent-government.
In (17b), t cannot antecedent-govern t i because of VP.To solve this problem Chomsky(1986:76) assumes that "VP is one of the two segments of the verb phrase, not a category in itself that excludes t j, whether it is base-generated or formed by an adjunction rule".Then " t i is properly governed in the extended chain (be, ti,ti) as required, independently of theta-government by 'killed' " (Chomsky 1986:76).Consider the following more complicated sentence; (18) a. Mary i seems to have been killed t i .
To explain the sentences which have the aspectual verbs Chomsky(1986:77) assumes the relation of head-head agreement(= index sharing) between I and the aspectual verbs V* as in "...I [vp V* NP ] ".He says that "there is (indirect) agreement between the subject and each aspectual verb of VP, as a reflex of SPEC-head agreement"(ibid.p.77).That is, based on the SPEC-head agreement, the Inflection 'to' has the same index with t j in the Spec of IP , and then ,based on the head-head agreement, the aspectual verb 'be' has the same index of i .
Then 'be' can antecedent-govern t i .However, his assumption has some problems as follows; first, is it intuitively plausible to assume agreement between to' and 'be' in the above sentence ?Second, is it plausible to assume Spec-head agreement between ' and to' which is a tense-lacking infinitive ?
The problems discussed in this section are mainly, I think, from the implausible definition of adjunction which considers VP as a barrier and doesn't allow the adjunction to VP in passive.Another problem here is that we combine a head-chain with an argument -chain to have an extended-chain, and consider this extended-chain as an argument-chain.However we cannot find other evidence which can support this assumption.
So far I have discussed the implausibility of the adjunction condition in barriers with many counterexamples, all of which show that the condition which allows free adjunction to any non-argument maximal projection is so powerful that it allows many ungrammatical sentences.Especially it is awkward to allow adjunction to a VP to avoid the barrierhood.Therefore from now on I will discuss the possibility that we can maintain the theory of barriers without admitting adjunction in syntactic movement.

Application of the New Theory
Since-we allowed adjunction to VP in barriers, we can think of the possibility that VP is not a blocking category and not a barrier ,either.Actually it is hard to find data which show that the ungrammaticality of a sentence is caused by the barrierhood of VP.VP is considered a blocking category because it is not Lmarked.Therefore if we find evidence which shows that VP is L-marked, that is, theta-governed, then it will enable us to give up the necessity of adjunction to Since we assume I as a lexical category, IP is not a 'defective' maximal category any more.However, WH-movement moves a wh-element to the Spec of CP, crossing over IP which is now considered as a non-defective maximal category.To avoid this unfavorable result, I assume that while IP is a blocking category, it is not a barrier itself.Instead, the maximal projection immediately dominating lP becomes a barrier.And the inheritance is effective only once.Under this assumption, there is no problem arising from WH-movement inside CP.In addition to this, because IP is assumed to transmit its barrierhood to its immediately dominating maximal projection, this should be generalized to the other maximal projections.Then the resulting definition for the new blocking category and barriers would be as follows;1) (20) Blocking category a is a blocking category for f3 iff: That is, under Chomsky(1986), the sentence should be awkward because in the second step of movement, what' crosses over one extra barrier.However this prediction is not borned out and therefore Chomsky(1986)'s assumption for the extra barrier can be said to be invalid.Since (b) is 2-subjacent, we can easily predict that (b) is worse than (a).(here under the modified theory, we should assume that 1-subjacency is nearer 'ungrammaticality' than 'awkwardness' as we can see from the ungrammaticality of (a) ) The new theory works in the problems above.
Consider the sentence below(=( 6 Here PP is not L-marked, so it is a blocking category.When Who' moves to the Spec of CP within PP, it crosses over no barrier.Then when it moves to the Spec of CP, it cannot adjoin to the non-argument PP by the new condition which bars the adjunction in syntactic movement.Therefore 'who' must move directly to the Spec of the matrix CP.In this movement, it crosses over one barrier, IP which immediately dominates the blocking category PP.Therefore , the sentence is 1subjacent and is predicted as ungrammatical as it is.Of course we can think of another possibility for the derivation as I showed in section 2.2.That is, we can think of PP generated within VP.However the result is the same with the case of the PP immediately dominated by IP.That is, if we assume PP within VP, then a VP which immediately dominates a blocking category (here the PP ) will be a barrier.In this case , IP cannot be a barrier because IP does not immediately dominates a blocking category but a bather by definition.Therefore the movement crosses over only one barrier.So the new theory also works for this problem.
1*  (Chomsky 1986:34-35) Consider (a) first.As we have discussed in (13), in (a) IP is not L-marked, so it is a blocking category and CP is not L-marked, so it if, also a blocking category.
When which book' moves, it crosses over two barriers, CP which immediately dominates the blocking category IP and NP which immediately dominates the blocking category CP.Therefore the sentence is correctly predicted to be ungrammatical.However the modified theory is confronted with a difficulty in (b).That is, it cannot correctly predict the grammaticality of (b).(Chomsky(1986) explained the difference in grammaticality between (a) and (b) in terms of the oblique case from N).
Let's apply the modified theory to (b).Since the N 'rumor' L-marks CP, CP is not a blocking category ,nor is NP.Then which book' can move from the Spec of the embedded CP to the Spec of the matrix CP freely.
The only thing we can depend on in this stage to check the grammaticality of sentences is the status of traces.In fact, so far linguists have depended on two methods to explain linguistic data , one of which is the binding relation, especially between an antecedent and its trace, and the other is the government of the trace by a head.
Let's look at the structure of (b) carefully.
(27= 26b) ?? Which book i did John hear [Np a rumor [cp t' i le that lip you read t ?In fact, the antecedent 'which book' can antecedent-govern ri in the Spec of CP because there is no barrier between them under the modified theory.It is clear that the problem is not from the binding relation.Rirti(1989:196) gives a very helpful explanation for this case.Rizzi(1989:197), accepting Cinque's proposal, says that " nouns should be inadequate governors.... a maximal projection which is not selected by a I+Vj head is a barrier.... ".The condition he suggests is as follows; (28) XP is a barrier if it is not directly selected by an Xo non-distinct from [+V] 1C) ti /30 This condition implies that " nouns, contrary to verbs and adjectives, will never be able fo-govern inside a lower maximal projection".Consider the following example DP he gives and one corresponding sentence.
(29) a. ?*ED? John's [np 1 En' appearance [ip t to be sick BB (Rizzi 1989:196) b. .1..)11n i I [vp appear [ip t; to be sick 1] In (a) there is no problem in movement.No barriers are crossed.In terms of antecedent-government, there is no problem.Clearly t' can do it.Therefore the problem should be from the relation between N' appearance' and the trace t i in the Spec of IP.According to Rizzi, we can make an explanation as follows ; even though the N appearance' L-marks its complement IP, it cannot L-mark t in the Spec of IP since it is [-V] as Rizzi defined in ( 28).Then the trace t in the Spec of IP is not licensed, therefore the sentence is judged as ungrammatical.On the other hand , in (b) the verb 'appear' can L-mark IP and also L-mark t i in the Spec of IP by Spec-head agreement as defined in (28).In the line of Rizzi's proposal,consider (27).The N 'rumor' L-marks IP, but since it is [-V], it cannot L-mark the Spec of IP.Therefore t' i cannot be licensed.Then we can explain the difference in grammaticality between (26a, b) .First, (26a) is 2-subjacent, so it is ungrammatical.In (26b), even though it doesn't cross over any barrier, one of the traces is not licensed by a head, so it is not grammatical.However, because (26a) seriously violates subjacency, (26a) is worse than (26b) which violates only a licensing condition.2) The following sentences are showing NP-movement.
(30) a. Mary seems to be smart (=( 14 In (30a) ,,instead of assuming feature-sharing , we can explain the grammaticality of (a) in terms of subjacency and the licensing condition.
[seem-I] L-marks the trace t"; in the Spec of VP' because it L-marks VP'.
3) The movement from t' in the Spec of IP to the matrix clause obeys subjacency condition of the modified theory.Therefore (30a) is predicted to be grammatical.
On the other hand, (30b) has a different story.Ev'n though all the traces are licensed by [+V] category, the movement from the, Spec of the most deeply embedded IP to the Spec of matrix VP violates subjacency.That is, the movement crosses over one barrier CP which immediately dominates the blocking category IP', so it is judged as ungrammatical as it is.In (30c), subjacency is obeyed, and traces are also well licensed by [+V] category such as I and verbs.Therefore, the sentence is grammatical.

Conclusion
So far I pointed out some problems which arose from the improper definition of the adjunction condition of barriers.As I discussed in section 2, the adjunction condition is too powerful in that it allows many ungrammatical sentences.To solve the problems, I first assumed that there is no adjunction at S-structure except extraposition and topicalization.Secondly, I assumed only inherited barriers , which is to consider IP, traditionally a defective category, as a normal lexical category.Under the new assumption, unlike I which L-marks VP, C and D which lack the ability of theta-government make IP and NP a blocking category, and the immediately dominating category becomes a barrier.In fact, under the modified barrier theory, subjacency condition can be said to be stronger than it was in Chomsky(1986).I applied the modified theory to every sentence which raised problems for barriers and showed that the new definition works well.NOTES 1. Recall that I am now extending the theory in the line that there is no adjunction in syntactic movement except extraposition and topicalization which can be considered as stylistic phenomena.
2. I am not sure whether I can extend this licensing condition to the other categories such as C, or D. I assumed in (30) that I L-marks its complement VP and also L-marks the Spec of VP (I am now assuming VP-internal hypothesis especially in explaining NP-movement as an experiment to extend and testify Rizzi's proposal.I think this licensing condition can be called roughly a headlicensing condition because in this condition the role of head-category such as verb or noun etc.to license traces is critical.I think it works well with the data so far.If this licensing condition is fully developed , then I think it will be very helpful for us to judge grammaticality of linguistic data together with subjacency condition.In fact, Chomsky(1986:24) also mentioned a similar suggestion with Rizzi(1989).Revising his L-marking condition, he mentioned that a lexical category can L-mark not only its complement but also the Spec of the complement.His condition for L-marking is as follows ; "where a is a lexical category, a L-marks b iff b agrees with the head of r that is theta-governed by a".
3 In fact, this process can raise some argument.,since it is not clear whether I has the feature [+V] or [-V].If we think of the Spec-head agreement , especially in terms of number, then I seems to be nearer to [-V].On the other hand, if we consider that verbs raise to I and theta-govern VP, then it seems to be more like [+V].For this case of (30a,b,c), I must be [+v] to L-mark the trace in the Spec of VP.
(6) * who i did you meet Mary before interviewing t

(
17) a. Mary was killed t 1 b.[ip Mary i [be-I] j [vp' ti [vp killed t ]]] Now let's see how the new theory explains this problem.In (a), when what' moves , it cannot adjoin to VP's because our theory disallows adjunction in movement.Therefore 'what' should move from the, Spec of CP directly to the Spec of the matrix clause.Then the movement crosses over one CP' barrier which inherits its barrierhood from LP'.In (b), when 'what' moves , it must move directly to the Spec of the matrix clause and during movement, it crosses over two barriers CP' and CP which immediately dominate IP' and IP respectively.
a. Mary seems to be smart [ip Mary [seem-I] j [vp t) [ip t i to be smart ]]]