" THE PATH CONTAINMENT CONDITION AND ARGUMENT STRUCTURE

The internal structure of verb phrases (VPs) are investigated. Using the Path Containment Condition, as developed by May (1985), to establish relations between quantified arguments, this study draws two conclusions about the structure of argument-relations within VPs. First, arguments have binary relations with projections of the verb, and second, verbal modifiers have more proximate D-structure relations with the verb than do the subcategorized arguments of the verb. Contains 18 references. (Author/LB) **********************************************************************t Reproductions supplied by EDRS are the best that can be made from the original document. ********************************************************************rn "PERMISSION TO REPRODUCE THIS MATERIAL HAS BEEN GRANTED BY TO THE EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC)." THE PATH CONTAINMENT CONDITION AND ARGUMENT STRUCTURE Thomas Stroik U 8 DEPARTMENT OF EDUCATION Othce of Educauonai Research and Improvement iTE UCATIONAL RESOURCES INFORMATION CENTER (ERIC) his docurneol has been reproduced es ecouved horn the person or organtzahon orvnalIng II CI Mtnor changes have been made lo improve reproduchon oualay Points& view or oPmene meted .11 this dOcu men! do not neCessanlv represent official OE RI poswon w misty Abstract: This study Investigates the internal structure of verb phrases (VPs). Using the Path Containment Condition as developed by May (1985) to establish relations between (quantified) arguments, this study draws two conclusions about the structure of argument-relations within VPs. First, arguments have binary relations with projections of the verb. And second, verbal modifiers have more proximate D-Structure relations with the verb than do the subcategorized arguments of the verb. This study Investigates the internal structure of verb phrases (VPs). Using the Path Containment Condition as developed by May (1985) to establish relations between (quantified) arguments, this study draws two conclusions about the structure of argument-relations within VPs. First, arguments have binary relations with projections of the verb. And second, verbal modifiers have more proximate D-Structure relations with the verb than do the subcategorized arguments of the verb.


Introduction
develops a theory of logical form that expresses the logical representation of a sentence syntactically.His theory, which Is grounded In the Government and Binding framework, derives the logical form of a sentence from its S-structure through the free adjunstion of logical operators to the categorial nodes stated at S-structure.According to May, it is only at this syntactically-derived level of logical form that the logical properties of a sentence--its *copal and binding relations--can be explained.However, since the free adjunction of operators creates structures that overgenerate logical properties, May posits a mell-formedness condition on IF-representations, the Path Containment Condition (PCC).that constrains permissabie logical forms.
In tWs paper, I will use the two major assumptions of May's theory--free operator adjunction and the PCC--to investigate the structure of Verb Phrases (VPs).I will show that, within May's theory, VPs have binary branching structures and adverbial adjunctions are the most proximate arguments of a verb.
May's Theory of Logical Form May (1977) argues that the ambiguity of (1) follows from the fact that the rule of Quantifier Raising can derive two different logical forms for (1), namely (2a,b).As (2) shows, Quantifier Raising (OR), a rule that adjoins a logical operator to an S-node, can generate LF-strictures that assign broad scope to either quantifier (where the outside quantifier is said to have broad scope).Hence, the ambiguity of (1) obtains from the two syntactic representations that are derivable for (1).May (1985) revises his account of the ambiguity of (1).Noting that the Empty Category Principle, a locality principle on admissable relations between antecedents or heads and their arguments,1 requires empty categories to be properly governed at LF, May shows that LF (2a) Is well-formed but LF (2b) Is not.That is, (2a) satisfies the ECP because all of its empty categories (ECs) are locally - governed; on the other hand, (2b) violates the ECP because one of Its ECs, viz.62, Is prohibited from being locally governed by its A.-antecedent by the presence of an intervening A.-operator (the quantifier some).2Consequently, (1) has only one well-formed logical representation--(2a).To account for the ambiguity of (1), May employs the Scope Principle (3). (3) The Scope Principle (SP).
In a class of occurrences of Operators X, if 01,0j are elements of X and 01 governs 0j, then 01, Oj have free scope; where A governs El 1ff A c-commands 8 and 8 c-commands A, and there are no maximal projection boundaries between A and 8; and where A c-commands 8 Iff every maximal projection dominationg A dominates 8 and A does not dominate B.
Applied to (2a), an LF-representation In which the operator every governs the operator some, the Scope Principle gives free scope to the quantifiers, allowing multiple readings to be assigned to (1).The SP also successfully predicts the lack of ambiguity for (4), which has LF (5).
(5) Ersomeonee5e2 believes (5.everyone3(e3 WM]] In (5), since a maximal projection boundary (S') Intervenes between the two operators, someone does not govern everyone; therefore, the Scope Principle does not apply to (5).The only scopal relation that can be assigned to (5) Is the one defined configurationally, the one that gives broad scope to someone.What May's theory falls to predict correctly, however, Is an example like (6).(8) Who bought everything for Max? OR will generate LF (7) for sentence (6).Unfortunately, (7) Is an ill-formed logical representation: It violates the ECP because the quantifier prevents the wh-operator from locally governing 62.This leaves (6) without a well-formed LF, making the sentence uninterpretabie. (7)

(2whoe5everythinge5 62 bought 62 for Max333
To derive a grammatical logical representation for (8), May replaces OR with a more general rule of free operator adjunction.Such a rule permits not just S-adjunction, but adjunction to any categorial node.Since free adjunction allows VP-adjunction, LF (8) can be derived for (8). (8) (2,whoe5e2(ypeverything2Npbought 62 for Max3333 Sentence (8) now has a well-formed LF because (8) satisfies the ECP.A further consequence of LF (8) Is that the scopal relations of (8) are correctly predicted by It.That Is, In (8) the wh-operator does not govern the quantifier because the maximal projection boundary VP intervenes between these operators, so when the SP Is applied to (8).It correctly predicts that there will only be configurationally defined scope.Support for the above analysis comes from the contrast between ( 6) and ( 9). ( 9) What did everyone buy for Max?Unlike (8), ( 9) is ambiguous.
It permits the reading In which everyone bought one particular item and the reading In which every individual each bought something (this thing could be different for each person) for Max.An account for the scopal properties just described follows from the rule of free adjunction and the Scopal Principle.
In (10), the LF of ( 9), the wh-operator governs the quantifier; so these operators, given the SP, can engage In free scopal relations.
Besides arguing for the SP and for free operator adjunction, May (1988) also argues that the ECP should be replaced by the Path Containment Condition (PCC).May notes that the ECP makes several incorrect, grammatical predictions.For one, the ECP incorrectly predicts a scopal difference between (11a) and (11b); It predicts that (11a) should permit ambiguity but (11b) should not, since the subject-trace Is properly governed In (11a) but not In Who do you think that everyone saw at the rally For another, because the wh-traces created by LF-movement of the wh-operators In (12a) and ( 12b) are all properly governed, (12a) and (12b) should both be well-formed.However, there Is an obvious difference In the grammaticality of the sentences.

Path Containment Condition (PCC).
Intersecting A'-categorial paths must embed, not overlap--Where a path Is a set of occurrences of successively immediately dominating categorial nodes connecting a binder to a bindee.
Opposed to the ECP, the PCC allows th rul of free operator adjunction to derive 1.F-representations (14a) and (14b), the LFs of (11a) and (11b) respectively.2).Therefore, according to the PCC, these LF-representations are well-formed.Notice that (14a) and (14b) make the same prediction about *copal relations.That Is, since the wh-operator governs the quantifier In both of LF-representations specified In ( 14), the Scope Principle predicts that both LFs allow free scopal relations between the operators.The PCC, then, can account for data that escapes the ECP.
Of equal importance to the fact that the PCC explains data that resists the ECP Is the fact that U.se PCC can account for all the data that the ECP serves to explain.
The PCC also makes correct predictions about the scopal differences between (8) and ( 9), repeated here In (15).(19), so the ambiguity of (18b) follow,' from the SP.) Further, operator adjunction forms two possible LF-structures for (18a).These LFs are given In (20).-is an acceptable LF. (20a) Is an ungrammatical LF because its paths overlap, In violation of the PCC.On the other hand, the paths In (20b) vacuously satisfy the PCC; they do not intersect so the PCC does not rule them out.Since (18a) has (20b) as Its logical representation, the SP applies to (20b) predicting correctly that, given the fact that a VP-boundary separates the wh-operator and the quantifier, (18a) has scopal ambiguity.

Some Cones uences of Ma s Theor of Lo ical Form
In this section, I will apply May's theory as outlined above to sentences with VPs that take multiple arguments.I will show that, under May's analysis, such VPs must be binary branching structures.3 Let us consider a multiple-argument predicate like read.
In sentences such as (21) the operators/arguments of the verb have ambiguous scope.These scopal relations result from the application of the SP to ( 22), an LF-repesentation of (21).4To decide which, If any, of the three logical representations that we have considered should be the logical form of (21), we need to examine other evidence.Relevant evidence comes from (24). (24) Who did John read everything to?Interestingly, (24) differs from (21) In that it Is not ambiguous.
((24) only has the reading where the wh-operator has broad scope over the quantifier.) If ( 24) Is assigned an LF parallel!to (21)--one In which the VP has a nonbinary branching structure--then the following LF-structure can be derived for (24).6This LF-representation Is perfectly grammatical: its paths fulfill the PCC.However, being well-formed, (25) allows the SP to apply to It with the consequence that, since the wh-operator governs the quantifier, ( 24) Is predicted to be 8 ambiguous.Obviously, such a consequence Is undesirable.
If, however, we assume that the VP must binary branch and that only major phrasal nodes are listed In a path-sat, we can derive LF ( 26). ( (rwhogs everythinges John (wqvread 83] to ) (28), unfortunately, leads to the same conclusion that (25) does.rlat Is, (28) Is a grammatical LF-representatlon that predicts that (24) should have ambiguous scopal readings.But, If ws assume that ths VP must binary branch and that all phrasal nodes are specified within a path-set, we can derive the following LFs for ( 24).
In accordance with the SP, the operators can only have configurationally defined scope--a correct prediction.Further Conse uences of the PCC An interesting consequence of May's theory of Logical Form concerns the structural relationship between adjuncts and VPs.Cosider the *copal relations between the adjunct when and the quantifier everyone In ( 28). (28) When did John see everyone In (28), either operator can have broad scope.This Is confirmed by the fact that (29a) and ( 29b) can be acceptable responses to (28).John saw Mary a week ago; he saw Sarah yesterday; and he saw 8111 earlier this morning.
To account for the ambiguity of ( 28), there must be an LF-representation of (28) that satisfies two conditions.First, since the wh-operator Is In the COMP-node at LF, the quantifier must be able to escape the VP-node that dominates it to insure that the VP-node will not prevent the wh-operator from governing the quantifier (thereby preventing free scope!relations).Second, the PCC must be met.
Satisfying the PCC, however, can be accomplished In two ways: either the operator paths do not intersect or they ars properly embedded.The former case arises In LF (30).In (30), the adjunct when Is an adjunct of S4, a node created at LF-structure (only thls type of LF-representation will guarantee that the adjunct-path will not intersect with the quantifier-path).The adjunct, then, would be only an LF-argument--a possibility not compatible with current theories of predication.7The second case necessitates that the path of the adjunct-operator includes the path of the quantifier.That Is, the path of when must include the VP-node that dominates the quantifier trace.The adjunct, therefore, must be a within the VP, not outside of it.Such conditions are captured In IF (31).Notice that (31) Is not only a well-formed logical representation for (28) because its paths properly embed but also a logical representation that predicts that the logical operators In (28) have free scope!relations.
To decide whether (30) or (31) (or both) Is the correct representation for (28), we need to cosider further empirical data.Relevant data Is provided In (32).(32) Who saw what where Assuming that the adjunct Is an S-adjunct and assuming, as do May (1988) and Chomsky (1985), that wh-in-situ elements are moved into the COMP at the LF-level, we can derive IS (33) for (3)8 If we assume that the adjunct Is a constituent of tn.vp, rather than an adjunct to S, we derive LF (34) for (32).When did Mary send everyone's paycheck to him?That the operators In (35) engage In free *copal relationssee (36) for an example of a broad scope reading assigned to the Quantifier In (35b)--suggests that the adjunct Is at least as deeply embedded In the VP as Is the argument most proximate to the verb.(36) Bill told Sally about Mary's problem yesterday; he told Tom about ittoday; and he told Jean about it Just minutes ago.
Thls Is tho case because If the adjunct were not as deeply embedded as la the direct object In (35a), the paths for the quantifier and the wh-operator would Overlap, as (370 demonstrates. (37a) (5.when2(53 everyone3(52 Bill (vs (v.  ) Note that as represented the paths In (37a), which assume that the adjunct Is as embedded as the direct object, satisfy the PCC.However, if the adjunct Is higher In the VP-node than Is the direct object, as In (37b), then Path(3) CIP,52,041.In this case, the paths will intersect and not embed, In violation of the PCC.It follows therefore that adjuncts, which are constituents of VP*, must be as proximate to the verb as is the closest argument of the verb at LF-structure.The above condition on VP-structure produces two possible logical representations for verb phrases: one In which the adjunct and the closest argument are sisters and one In which the adjunct is a sister to the verb alone.These VP-structures are given In (38). ( The 0-representations In (38) make very different predictions about multiple-wh constructions, so they can be tested for empirical adequacy.
(38b) predicts that sentences formed by moving a wh-object and leaving the wh-adjunct in-situ at S-structure will be as grammatical as sentences formed by moving the wh-adjunct and leaving the wh-object in-situ because both types of sentences will have logical representations that meet the PCC.That Is, the LF-representations derived from multiple-wh constructions based on (38b) are either (39a) or (39b), both of which are well-formed.Since the paths In (39a) and (39b) intersect and embed, either type of multIple-wh construction under consideration Is predicted to be well-formed.
It predicts that multlple-wh constructions with the wh-object in-situ should violate the PCC, but such constructions with the wh-adjunct in-situ should satisfy the PCC.This can be seen bi examining the paths for the two constructions under consideration, as given In (40).*What did John buy when Now If the VP-structure Is as expressed In (38b), there should be no grammatical distinction between (41a) and (41b) nor between (41c) and (41d).The fact that there Is a grammatical difference between these pairs suggests that (38b) does not represent the logical structure of Ws.On the other hand, If the VP-structure at LF Is the structure expressed In (38a), then we should expect the construction with the wh-adjunct in-situ to be well-formed and the construction with the wh-object In-situ to be 111-formed.Interestingly, the data does not support thls prediction ithr: the data Is xactly opposite of what it Is predicted to be.

A Re-analysis of Multiple-Wh Constructions
The above results force a re-examination of our earlier assumptions (after all, at least one of our assumptions must be incorrect or we would have one of our predictions supported by, rather than both of them contradicted by, the data).I will argu her that the questionable assumption Is the assumption that wh-in-situ lements move into COMP at LF (note: I am only challenging this assumption for languages that permit wh-movement as S-structure).I will argue that wh-in-situ elements remain in-situ at LF where they function as dependent, lexical variables.
If wh-in-situ lements do indeed move at LF, then we would predict that the wh-operator moved at S-structure and the wh-operator moved at LF la multiple-wh constructions would engage In free scope!relations, In accordance with the Scope Principle.We can see that this Is predicted by examining the LF of (42)--which Is stated In (43).(44b) ?*John was kissing Mary; Bill was kissing Sue; but no man was kissing Sarah.
We can see that (44b), as a response to (42), Is much worse than (44a) Is.This difference Is unexpected If (43) Is the LF of ( 42).After all, LF (43) predicts that the order and the way In which the wh-arguments are Instantiated should not affect grammatIcallty--a prediction not compatible with the evidence given In ( 44).

(rwhIch manes e2 No kiss which woman3]]
In (45), Wh2 and Wh3 are not both Independent operators that can freely choose their referents.Rather, only Wh2 Is an operator; so only Wh2 can freely pick a referent or a non-referent (for example, no man).Wh3, on the other hand, Is a dependent varlable--a variable licensed for a referent If and only If It Is bound to a wh-operator that has chosen a referent (as opposed to choosing a non-referent).(Note that the assumption that wh-in situ expressions are variables dependent on a wh-operator will explain why the absence of a wh-operator In sentences such as "I love who" are uninterpretable on the non-echoic reading.)LF (45) then predicts that If Wh2 selects a referent then Wh3 can freely choose a referent or a non-referent; but If Wh2 does not select a referent, then Wh3 cannot choose a referent Independently.So (45) predicts the following grammaticality Judgments about responses to (42).The fact that the judgments predicted by (45) accord with accepted intuitions about responses to (42), while (43) has no way of differentiating the various responses cited In (48), suggests that (45)--a logical representation that leaves wh-eiements in-situ at LF--has more empirical validity than does (43).
A second argument In support of my wh-in-situ analysis involves scopal relations between conjoined wh-phrases and other logical operators.Consider (47).For some child x, which man y and which woman z are such that x was dancing with x and y (46b) For which man y Is there some child xl and for which woman z Is there some child x2 such that xl loves y and x2 loves z.
The scopal ambiguity of (47), as captured In (48), follows from May's theory of scope assignment.
(49) (rwhich man and which woman2 rooms child3(03 was dancing with 12]] Since the conjoined wh-operators govern the quantifier, free scopal relations arise between the logical operators.(Mote: the reading of (47) given In (48b) follows from a principle of operator distribution developed In Barwise and Cooper (1981).They demonstrate that connected operators that have wide scope over another operator distribute.This can be represented formally: (01 02)03 0103 02 03.Hence In (49), the wide scope reading for the conjoined wh-operators (i.e., (Which man and which woman) (some child)) Is equivalent to the reading given In (48b): ( (which man)(some child) and (which woman)(some child).)Now If wh-ln-situ element move at LF, then we would expect (50) to have the same scopel ambiguities as does ( 47).Notice that LF (51)--the LF for (50) in the move-wh at LF analysis--has the same government relation between the conjoined wh-operators and Wh3 as (49) has between the conjoined wh-operators and the quantifier.Since ( 49) and ( 51) have the same government relations between operators, we would predict that they should have the same range of readings.( 50), however, does not have all the scopal possibilities of (47).It lacks ( 52), the equivalent of (48b).
(52) Which man y for which child xl and which woman z for which child x2 are such that xi loves y and X2 loves Z The move-wh at LF analysis, then, overgenerates scopal possibilities and, therefore, needs to be questioned.
A better analysis of ( 50) Is one that assumes that wh-in-situ elements do not move at LF.This analysis would give LF-representation (53) to ( 50).
(53) (s.which child2 (02 (vploves which man and which woman]]] LF (53) does not permit ambiguous scopal relations because It has but one operator--this necessarily prohibits a mUltIplICItY of scopal configurations.The only reading that (53) allows then Is the reading In which the wh-operator first selects its referent and subsequently the wh-variable makes a referent choice.So possible answers to (50) consists of a set of order pairs (which child, which man and which woman>, where the value of the first member of the ordered pair determines the value of the second member of the pair.But such answers, as predicted by ( 53), are the only answers to (50) that are well-formed.Although the wh-ln-situ at LF analysis does account for scopal data (especially ( 47) and ( 50)) that resist the move-wh at LF analysis, there does appear to be some evidence In support of the latter analysis.In particular, sentences such as (54) seem to have scopal relations determined by a rule that moves wh-eiements at LF. Since there Is a VP-boundary between the quantifier and the wh-operators, the Scope Principle correctly permits only configurationally defined scope.
The success that the move-wh analysis has In explaining the *copal relations of (54), however, does not carry over to other types of multiple operator structures.Consider (58), which under May's analysis has LF (57).Given that the operators In (57) govern one another, (57) In accordance with the SP permits free scopal relations between the operators.May's analysis, then, predicts that all the sentences In ( 58) could be well-formed responses to (56).analysis that assumes that wh-in-situ elements remain in-situ at LF can account for the scopal relations of both i54) and ( 58).
If wh-in-situ elements are lexicallzed LF-variables that are value-dependent upon the value selected by a wh-operator and are not Independent operators, then (54) will have LF (59).Notice that since the wh-in-situ element Is not an operator.It does not directly participate In scopal relations.Rather, as a dependent variable, its scope Is a function of the scope of the wh-operator upon which it Is value-dependent.Consequently, the fact that the wh-operator In (59) has broad scope over the quantifier necessitates that the in-situ variable also has scope over the quantifier (hence, this analysis correctly predicts the scopal relations In ( 54)).This analysis naturally extends to account for the *copal relations In (58).That Is, because the in-situ wh-elements have their value attached to the wh-operator In LF (80), they must indirectly have the same scopei relations with respect to the quantifier as does the wh-oparator.In (80), then, the only scopal relations possible are the relations between the quantifier and which book, and these relations are free because the wh-operator governs the Quantifier.Further, the In-situ wh-elements, which are variables that do not overtly participate In scopal relations, have their values set by the wh-operator.By having their values set by the wh-operator, the in-situ wh-elements indirectly absorb the scopal relations of the wh-operator.Therefore, the wh-olements all either have broad scope or narrow scope with respect to the quantifier, but they cannot have mixed scope, as In (58c,d).The possible scopal relations In (54) and in (58), than, accord with the predictions this analysis makes about scope.
Disallowing the general move-wh rule complicates my analysis of VP-structure.After all, I have appealed to multiple-wh structures to motivate the assumption that VP-adjunctions are VP-internal and to argue that such adjuncts are In fact more proximate to the verb at LF than the subcategorized arguments of the verb are.Without move-wh as a general rule, multiple-wh constructions can no longer be enlisted as evidence to show what VP-structures the PCC mandates.
In what follows, I will introduce nsw evidence to support my claims that (I) VP-adjuncts are VP-internal and (II) these adjuncts are sister-related to V at LF.

VP-structure Revisited
There two types of data that support the claim that VP-adjuncts are VP-internal: binding data and VP-deletion data.Some evidence In support of the above claim comes from the binding relations involving R(eferential)-expressions. In the Government and Binding framework, Binding Principle C states that an R-expression must be A-free.9This means that an R-expression cannot be coindexed with any element that c-commands It from an A-position.Principle C, then, predicts the binding In ( 61).The fact that the sentences In (79) are grammatical under the 6tIpulated b:nding relations, while the one In (81a) Is not, suggests that Path(pronoun) In the LF for (79) cannot be undefined.To express a well-defined path Path(pronoun) for (79), we must assume that the adjunct Iles within Vsthis will allow the pronoun to form a path with the quantifier that It Is coindexed with.From the above assumption, we can derive two well-formed logical representations for the sentences In ( 79).At this point In our argument, we have facilitated two possible logical structures for adjuncts: one where the adjunct Is the sole sister of the verb (82b) and one where the adjunct shares V-sisterhood with the most proximate argument of the verb (82a).There are two typos of evidence that can help decide between the variant logical representations.The first type of evidence comes from data generated by the VP Rule.As previously discussed, VP Deletion shows that (82a), an LF-representation In which a verb , Its NP-object, and an adjunct are sister within a V'-constItuent, Is ill-formed and that an IF-representation, which has binary sisterhood as expressed In (82b), Is well-formed.The second type of evidence comes from sentences that hive multiple adjuncts.If (82a) Is the correct representation, then all adjuncts 21ust be sisters with the verb, with the most proximate argument of the verb, and with one another.
If (82b), on the other hand, Is the correct reprzoentation, then adjuncts need not be sisters with the NP-argument nor with one another; In fact, If VP-structure is binary In nature, It would be expected that  (Note: I articulate the deleted elements In the gapped constituent In (92) rather than just the constituent itself--(v.e3--becausestructures that are like (88) but do not have a variable present In the second conjunct are ill-formed: the examples In (89) show the necessity of having an NP-variable present In the second conjunct In order to have a grammatical structure.)Now although the variable In the gapped constituent Is bound by the wh-operator, it cannot take its interpretation directly from the operator because the gapped element, of which the variable Is a constituent, has to be bound to and take its interpretation from some antecedent, the le-constituent In the first conjunct.Therefore, the gapped constituent In (92) must take as its antecedent (v.read el.The interpretation given to V In the second conjunct, than, has a variable In it that Is not directly bound (again, the variable Is bound In (92)--this explains the grammaticality of (88a)--but it Is not constrained In the interpretation it takes within the gapped W-constituent).Consequently, the variable can be interpi.tedindependent of its bound counterpart In the first conjunct.
Crucial to the concerns of this paper Is not the claim that the variable In the second conjunct of (92) Is a parameterized variable (an interesting claim In its own right) but the claim that the differences In the interpretations of (86a) and (86b) depend on the fact that the variable in the second conjunct is part of the gapped constituent.This latter claim, therefore, Is one that needs further verification.Support for the claim under consideration COMBS from ( 93) and ( 94).In (86a).
However, the variable In the second conjunct of ( 93) is interpreted like the varibie In (86b), nOt Ilk@ (86a).
That Is, In non-gapped sentences, the variable Is directly bound to the wh-operator.We are forced to conclude then that the interpretation of the variable In (86a) is dependent upon its relationship with the verb.Further, the V', more particularly an adjacent sister of V.
My fourth, and final, argument for LF-representations In which adjuncts are the most proximate arguments of a verb Is taken from the data given In (98).Where dld John meet Mary after 8111 dld e Jean What needs to be explained In (98) Is why the empty element In (98b) can Include, within Its Interpretation, the wh-adjunct, while (98a) cannot do so with Its wh-adjunct.It Is possible to see these different Interpretations more clearly when we recast (98) as ( 99).
(99a) sWhy dld John meet Mary before 8111 met Jean for that reason Where dld John meet Mary before Bill met Joan there Notice that in (99a) the wh-adjunct position cannot be fIlled In the adjunct-clause, but In (99b) the wh-adjunct can be filled In the adjunct-clause.
An explanation of the contrasts In ( 98) and (99) follows from a condition on the deletion of arguments In adjuncts.Thls condition can be extracted from the evidence In (100).In particular, since the wh-adjunct Is not a sister with the verb In (lola).It cannot be deleted with the verb; therefore (98a) cannot have the why-adjunct present In the verb phrase of the adjunct-clause, explaining why (98a) lacks an interpretation that permits the why-adjunct to be part of adjunct-clause VP.Conversely, the deletion of the verb and Its wh-adjunct Is acceptable In (98b) because these two elements form a constituent.As a consequence, the verb phrase In the adjunct-clause of (98b) can be interpreted as including the wh-adjunct.
Additional support for the conclusions Just derived can be found In multiple-wh constructions.The following sentences give the relevant evidence.?Who ate when why (103d) Who ate why when The fact that the ordering of the wh-eiements In (103) Is crucial to the well-formedness of the sentences suggests that these wh-eiements cannot have equivalent logical relations with the verb.That is, (103a,b) show that structures are grammatical If where Is more proximate to the verb than when is, but ungrammatical if when Is more proximate than where.Similar results obtain for where and th In (103c,d).Importantly, the above relations are exactly those predicted by (101).
The arguments that I have put forth In this section converge to the same point: VPs binary branch In such a way that their verbs accept arguments one at a time, beginning with all the adjuncts and ending with the NP-arguments.On some Intuitive level, this conclusion seems correct.After all, In (104), the NP-object Marx seems more like the argument of the extended predicate see after 8111 left, as represented In (104), than an argument of see.In this paper, I have argued that VP-structures binary branch and that VP-adjuncts are the most proximate arguments of V at LF.These conclusions raise some interesting questions about the relationship between the levels of representation posited In G8 and some of the principles of grammar hypothesized In G8 (In particular Case Theory, Th-Criterion, and the Projection Principle).For one, what needs to be explained Is why VP-adjuncts are discontinuous with the verb at S-structure when they are continuous with the verb at LF. Now there seems to be an answer to this question.The reason for this S-structure discontinuity follow* In a straightforward way from Case Theory within the G8-framework.According to Case Theory, structural casr, Is assigned at S-structure.
Further, Case Is only assigned under conditions of adjacency.For case assignment of the direct object within a VP, the above conditions require the object to be adjacent to Its case assigner (the verb) at S-structure.
It Is, therefore, the case that the verb and Its "logical" sister (the adjunct) cannot be sisters at S-structure or else the assignment of structural Case of the object will be prohibited.
Although we can suggest an answer to problem that my analysis raises for Case Theory, there are some questions that arise that cannot be resolved so easily.These questions have to do with the D-structure position of VP-adjuncts.Are the adjuncts D-structure sisters of the verb?If so, doesn't that configuration interfere with th-marking?(Relatedly, can X'-elements, as well as X°-elements, th.mark complements--as N' may do with Its arguments when NO Is modified by an adjective?)If not, what Is the mechanism through which an adjunct comes to be the "logical" sister of a verb?Such questions, although very interesting, are however beyond the scope of this paper and Kansas Working Papers in Linguistics.Vol.13, 1988.pp.139-173.

?
Whom did you tell that Harry saw who (12b) Who did you tell whom that Harry saw May accounts for the fact that both sentences In (11) ars ambiguous by replacing the empirically inadequate ECP with Pesetsky's Path Containment Condition (13).
Is well-formed; Its paths satisfy the PCC because they embed.(Note the wh-operator governs the quantifier, In . ,b) are ungrammatical.The paths In these LF-structures violate the PCC because Path(2) intersects and overlaps with Path(3).So neither (27a) nor (27b) Is a possible LF-structure for (24).(This an important result because If either of these LFs would bs well-formed they would Incorrectly predict that (24) should allow free scopal relations.)(27c), on the other hand, does not violate the PCC.
The PCC, then, forces us to analyze the structure of 9 BEST COPY MAE VPs headed by verbs subcategorized for multiple arguments In terms of binary branching structures.
), Path(3) and Path(4) intersect but they do not embed.Consequently, thls logical representation Is an ill-formed IS-representation because It violates the PCC.
Np[what2 where3]who4lls e4 Evp *40 v2 0333] Path(2) (VP,S,S',NP) Path(3) CVP,S,S',NP) Pathpaths embed In (34), LF (34) satisfies the PCC and Is, therefore, a well-formed logical representation if (32).The consequence of the above argument Is that adjuncts, at least at the LF-lavel, are within the verb phrase.Given that our previous arguments demonstrate that VPs have binary branching structures for th4 arguments of V, the question arlses: what Is the branching relationship between adjuncts and subcategorized arguments within the VPs? Sentences that immediately bear upon this question are: (35a) When dld Mary read a book to everyone?(35b) When did Bill tell everyone about Mary's problem?(35c) ,VP,S,S',NP] 44p(wh-adjunct2)wh-NP3](3...(vp(y.V 02] 03 ) satisfies the PCC: its paths properly embed.LF (40b), on the other hand, has paths that intersect and overiap--a PCC violation.So, If the logical representation for multipie-wh constructions Is as stated In (38a), then such constructions are predicted to be grammatical If the wh-adjunct Is left In-situ at S-structure and to be ungrammatical If the wh-object Is left in-situ at S-structure.The above predictions can be tested by the data presented In (41).
) shows that under the assumption that wh-in-situ elements move at LF the wh-operators govern one another; therefore they should have free scopal relations.14Nowlet us consider possible responses to (42) In order to check how free the scopal relations In It really are.Note the answers given to (42) In (44).
Mary; Bill was kissing Sue; but Tom was kissing no woman.
which woman was some child dancing with Example (47) Is two-ways ambiguous, having the readings given In (48).
(50)    Which child loves which man and which woman 1 l; (54)    Who took everyone to which restaurantThe fact that the wh-operators both have scope over the quantifier Is explained by (55), an LF-representation formed by the general move-Wh rule.
Np[which boy and which girl] ehich book3)(overyone44 04 read 03 to 823]] Bible to John and Mary; and Sarah read the Koran to Jean and Harry (58c) ?*Peter read the Bible to John and Mary; and Sarah read It to Joan and Harry. (58d) Peter read the Bible to John and Mary; and Sarah read the Koran to them Two of the above responses--(58a), where the the wh-elements have broad scope over the quantifier, and (58b), where the quantifier has broad scope over the wh-elements--are well formed.The other two responses, where the quantifier has narrow scope with respect to one wh-eiement and broad scope with respect to the other wh-element, are less well-formed and perhaps even ill-formed.Since the data In (58) contradict th predictions made by LF (57), a logical representation that employs the general move-wh rule, there Is reason to suspect that wh-movement at LF Is not a permissabie rule.Unlike May's move-wh analysis of logical form, an 18 BEST COPY AVAILABLE only) John's mother Binding between the piano and It In (61a) Is well-formed, In part, because Principle C Is satisfied.That Is, since the c-command domain of the pronoun Iles within the adjunct-clause, the pronoun does not c-command the R-expression John: so John is A-free.The binding relations specified In (61b) are also well-formed because the pronoun, which has its c-command domain restricted to the NP of which It Is a constituent, does not c-command John, thereby preserving Principle C. Opposed to the binding relations illustrated In (61a,b), the binding relations In (61c) are ungrammatical.The pronoun In this sentence has as Its c-command domain the entire S; consequently, John Is coindexed with and c-commanded by an element In an A-position--an obvious violation of Principle C. If we apply Principle C to sentences with VP-adjuncts, we can discover something about the structural relationship between adjuncts and verb phrases.Consider the sentences In (62).
adjucts would have a structure like: (vp ( vm (w V Adjuncti] AdjunctO NP-objectn].With these predictions In mind, let us consider (83) and some Gapping data associated with it, as illustrated In (84).

?
Which book was John reading to Mary and Bill reading to Sue Now If the free interpretation of the variable In (86a) Is independent of the relationship between the gapped verb and the variable, then we would expect that In sentences like (93) (sentences without gapped verbs) the range of interpretation for the variable would be the same as it Is Mary read to John before 8111 dld to Jean (100b) Mho woman that Mary talked to about Tom's problem before Jean dld to about Bill's problem The examples In (100) demonstrate that, In adjuncts with deleted verbs, an argument of the verb can be deleted only If the If It Is an adjacent sister of the verb.Given the above condition on deletion within an adjunct, we could hypothesize that the ungrammaticallty of (99a) and the inability to assign a wh-adjunct reading to the VP within the adjunct-clause arises because the wh-adjuncts are not sisters of the verb.If we accept the above assumptions, we will poslt the following logical representations for the VPs In (99).(101a) (Nip No (le V before-adjunct] why-adjunot]] (101b) (vp (vle (v.V where-adjunct] before-adjunct]] Although the LFs In (101) are the LFs for the matrix VPs In (99), they are also the LFs for the VPs In the adjunct-clause.To see this, note that the before-clause cannot take a before-clause of its own--which Is naturally explained If a before-clause Is already present In adjunct-clause.(102) Where did John meet Mary before Bill did Jean before It rained So the adjunct-clause VPs In (98), under the interpretations given In (99), have the same structure as do the matrix clauses: the structure expressed In (101).Given that the adjunct-clauses In (98) have the VP-structures stated In (101), we can explain the differences In interpretation between (98a) and (98b).