Recent work by Bresnan and colleagues Keywords: ditransitives, double object constructions, quantifier scope, weak crossover, locative inversion Ditransitive Constructions In a series of papers, Bresnan and various colleagues 1 A note on terminology: I will refer to the goal or recipient NP as the goal NP, whether it is the first object of the double object construction or the object of the preposition in the prepositional dative construction. I will refer to the other internal argument as the theme NP, and to the external argument as such, where relevant. These terms are used as descriptive labels only; no theory of thematic roles is intended. R E M A R K S A N D R E P L I E S (1) a. We give children candy. b. We give candy to children. Bresnan et al. argue specifically against a family of analyses that claim that the double object construction and the prepositional dative are two distinct constructions, with different syntax and different denotational semantics (some recent representatives of this family include (2) a. The lighting here gives me a headache. b. *The lighting here gives a headache to me. (3) a. The count gives me the creeps. b. *The count gives the creeps to me. In addition, there are some nonalternating verbs like deny: (4) a. The boss denied George his pay. b. *The boss denied his pay to George. Bresnan et al. argue that such contrasts are spurious. They have found examples like the following in studies of corpora of spoken and written language, and they report that speakers accept them under the right conditions (such as when the goal NP is relatively long): (5) a. . . . a stench or smell is diffused over the ship that would give a headache to the most athletic constitution. (Bresnan and Nikitina 2007:4, (15)) b. This story is designed to give the creeps to people who hate spiders. 2 I will refer to this reversal as R-dative shift (suggesting ''rightward dative shift''); on the surface, the result of R-dative shift is identical to the prepositional dative, but the two have very different derivations. 2 I first heard this theory suggested by Alec Marantz in a seminar at MIT sometime around the year 2000. As far as either of us is aware, the idea is original to him. The only discussions I have seen of it in print are brief mentions in footnote 4 of Richards 2001 and on pages 43-45 of Harley 2002 (referring to Richards). To be concrete, I will assume as illustrated in (6) that the double object construction has the goal NP introduced by an Appl(icative) head R E M A R K S A N D R E P L I E S The important difference between these structures is that in (6), the goal and theme NPs are arguments of different heads and appear in different phrases, but in (7), they are arguments of the same head and occupy the same phrase. In addition, in (6), the goal NP asymmetrically c-commands the theme NP, but in (7), the theme NP and a constituent consisting of the preposition and the goal NP (P′) c-command each other. 4 What I am calling R-dative shift involves the structure in (6), not that in (7). In linear order, the two are identical, because of the shift of the goal NP and the appearance of to, but their syntactic structures are very different. For reasons that I will outline below, I will analyze R-dative shift as having exactly the same structure as the typical double object construction, except that the specifier of ApplP is projected on the right as in (8). Both the small clause analysis and the ternary-branching structure can account for the scope asymmetries discussed here. I stick to the small clause analysis in the text simply because most current theories of syntax disallow ternary branching. However, the ternary-branching analysis may fare better in accounting for idiom formation. Since it treats the theme NP and the PP on a par, it can account for the fact that either one can combine with the verb to form an idiom to the exclusion of the other (assuming an approach to idiom formation something like that in O'Grady 1998). There are idioms like When it appears on the right, the goal NP must be marked by the preposition to. 5 This rightward projection does not appear to be freely available; otherwise, examples like (2b), (3b), and (4b) would always be acceptable. I will therefore assume that the rightward projection of ApplP's specifier is licensed only when the goal NP is extracted, via wh-movement, relativization, or some other Ā -movement process. (9) The Extraction Constraint on Rightward Specifiers The specifier of ApplP may be ordered to the right of its sister only if the NP that occupies it undergoes Ā -extraction. One of the most important processes of Ā -extraction here will be heavy shift. Under conditions of prosodic heaviness, the goal NP can move further to the right as shown in (10), right-adjoining to VoiceP, I will suppose (taking with it the preposition to that appears as a consequence of R-dative shift). O'Grady (1998:301) argues that the preposition in idioms like give the lie to X is a crucial part of the idiom. If this is true (it is hard to tell with alternating verbs like give and lend, because they only ever allow the preposition to), then the small clause analysis may actually be better equipped to account for the existing idioms: since the theme NP is not directly an argument of the verb (instead, it is the subject of the small clause), V ‫ם‬ PP idioms are expected to exist (throw X to the wolves), but V ‫ם‬ NP idioms are not expected to exist without the intermediacy of the preposition. 5 There are many possible analyses of the appearance of to. To mention just three: (a) the two NPs must be distinguished when their canonical order is reversed; (b) the null dative case marker corresponding to to is only licensed adjacent to the verb; (c) to obligatorily deletes when adjacent to the verb. I will not attempt to construct an analysis of case here, since my only purpose is to argue for the existence of R-dative shift. I should note, however, that the to that appears as a result of R-dative shift acts in numerous ways like a regular preposition: it can be stranded by wh-movement, but not by heavy shift, for instance. Other types of Ā -extraction, like wh-movement, will also license R-dative shift. The important point of this analysis is that, although R-dative shift looks identical on the surface to the prepositional dative construction, it is really a double object construction. Most crucially, the base position of the R-dative-shifted goal NP asymmetrically c-commands the theme NP, in contrast with the prepositional dative construction in (7) where this is not true. As just stated, one of the environments where R-dative shift happens is under the conditions for heavy shift. So, the following sentence has two possible derivations: (11) I give candy to anyone who offers me a seat on the bus. In one derivation, the goal NP starts out as the object of the preposition to in the prepositional dative construction in (7). In the other, it starts out as the first object of the double object construction in (6), but in a rightward specifier, as in (8) (R-dative shift). It then undergoes heavy shift to rightadjoin to VoiceP, as in (10). Because it started out in a rightward specifier, to appears. This hypothesis then accounts for Bresnan et al.'s data by claiming that phrases like give X the creeps and give X a headache are always double object constructions. The examples in (5) are underlyingly double object constructions, with the goal NP shifted to the right and to appearing as a consequence. This is allowed because all of the goal NPs in these examples are prosodically heavy and undergo further extraction to right-adjoin to VoiceP. The evidence that this hypothesis is correct comes from two sources: scope interactions between quantifiers, and locative inversion. Both phenomena show that examples like (5) pattern with double object constructions and not with prepositional dative constructions. They are therefore not prepositional dative constructions, but a special variant of the double object construction. Scope Interactions In order to test the R-dative shift hypothesis, we have to look at examples like give X the creeps and give X a headache, or deny X Y, where in many situations (like those in (2b), (3b), and (4b)) most English speakers have the intuition that the prepositional dative is not allowed. Cases like (11) are simply ambiguous; since they allow the prepositional dative, that derivation is always possible, and we will not be able to find evidence for the alternative double object derivation. The first type of data I will look at involves the relative scope of quantifiers. As is well known bar to every child, I'm going to need ten different candy bars.'' Because this is a perfectly coherent discourse, the universal quantifier must be able to take scope over the indefinite. Such a context does not work with the double object construction: ''#If I have ten candy bars, and I want to give a different child every candy bar, then I'm going to need ten children.'') Note that readings where the indefinite takes scope over the universal quantifier are not helpful; they are also compatible with the opposite scope order, and so their availability tells us nothing. Surface scope readings are also possible in both constructions, as one would expect. (13) a. I gave every candy bar to a different child. (every Ͼ a) b. I gave every child a different candy bar. (every Ͼ a) This is the general case in English: scope readings corresponding to the linear order of the quantifiers are usually available and the most salient. This will be significant below, where scope readings corresponding to the surface order of the quantifiers are unexpectedly unavailable. In the theory developed in Bruening 2001, the difference between the double object construction and the prepositional dative shown in (12) follows from the trees in (6) and (7). In the double object construction in (6), the goal NP asymmetrically c-commands the theme NP. The goal NP will therefore always move first in any scope-taking movement, given standard locality constraints on movement, and it will be impossible for the theme NP to cross over the goal. In the prepositional dative construction in If this is correct, and R-dative shift is just a double object construction but with the specifier of ApplP on the right, as in (8), then we expect R-dative shift to pattern just like the double object construction in scope possibilities. The goal NP asymmetrically c-commands the theme NP even though it follows it, and the theme NP should be unable to take scope over the goal NP. Further movement, like heavy shift in (10), will only take the goal NP to an even higher position; it will have no effect on scope. In this light, consider In other words, in scope possibilities give a headache to X patterns exactly like give X a headache. It does not pattern with the prepositional dative construction, where either scope order is possible. This fact strongly indicates that give a headache to X indeed is the double object construction, but with a different surface variant that happens to look identical to the prepositional dative. The structure that I am suggesting for linear order. (This lighting will move to the surface subject position, Spec,TP, irrelevantly to the point of this article.) The same facts emerge with the nonalternating verb spare. I now use some N or other to try to bring out the distributive reading. . Surprisingly, again, surface scope is unavailable in (17d), just as the same scope reading is unavailable in (17b). The latter, by hypothesis, is the underlying form of (17d). Exactly the same facts emerge with the nonalternating verb deny (which I include because some speakers do not accept R-dative shift at all with spare). Once again, the shifted variants in (18c-d) pattern like the double object constructions in (18a-b). They do not pattern like the prepositional dative construction. As stated above, R-dative shift (i.e., projecting the specifier of ApplP on the right) is licensed by further Ā -extraction, including heavy shift. If heavy shift is always right-adjunction to VoiceP, then we expect similar scope facts to emerge with heavy shift. To show this, we need to find a construction where inverse scope is disallowed, as in the double object construction, but heavy shift is allowed. The spray-load class of verbs is exactly what we need. As is well known, this class of verbs is like the double object construction in not allowing the theme argument to take scope over the goal argument in its with variant, in contrast with its to or on variant where the roles are reversed. R E M A R K S A N D R E P L I E S In Larson 1990 and Bruening 2001, this unavailability of inverse scope is related to the same unavailability in the double object construction; as argued in Bruening 2001, it can receive the same structural account (asymmetric c-command, with the goal argument introduced by an Appl head in the with variant but a small clause structure in the other variant). Unlike the double object construction, however, the spray-load class of verbs permits heavy shift of the goal object, the one that obligatorily takes wide scope. Just as the surface scope reading of R-dative shift is unavailable in (14c), (17d), and (18d), surface scope is also unavailable under heavy shift with this class of verbs. In other words, heavy shift does not change the admissible scope relations between the two arguments. This supports the notion that the unavailability of surface scope in R-dative shift with a prosodically heavy goal NP has a heavy-shift-type etiology. Note, however, that heavy shift is only one of the contexts that licenses R-dative shift. Whmovement of the goal NP is another type of Ā -extraction that does so. In (24a), the pair-list reading is readily available, because the goal everyone can take scope over the theme what. The theme every request in (24b), in contrast, cannot take scope over the goal who, even though this is perfectly possible with an alternating verb in (24c). Again, shifted deny does not pattern with the prepositional dative construction; it patterns with the double object construction. The same holds for give a headache to X, as pointed out by a reviewer. More examples can be given, but they all lead to the same conclusion. Bresnan et al.'s examples do not show that there is no syntactic/semantic difference between the double object construction and the prepositional dative construction. If their examples in (5) were just the prepositional dative construction, there would be no reason that scope readings corresponding to the surface order of the quantifiers would be unavailable. Quantifier scope shows that such examples are actually double object constructions. R E M A R K S A N D R E P L I E Locative Inversion At least one other phenomenon, locative inversion, distinguishes double object constructions that have undergone R-dative shift from the prepositional dative. As is well known, passives are a major source of locative inversion (e.g., Bresnan 1994: 78-79), and this includes the passive of the prepositional dative construction. (26) a. To the guests were given goblets of gold and silver. b. To the people were distributed morsels of consecrated bread. However, R-dative shift does not participate in locative inversion. Phrases like give X hell and give X a good scrubbing that are only grammatical with to under certain circumstances do not permit locative inversion. (27) a. To the generals that lost the battle were given helicopters. b. *To the generals that lost the battle were/was given hell. c. The generals that lost the battle were given hell. (28) a. I gave the tables a good scrubbing. b. I gave a good scrubbing to the tables we inherited from Aunt Selma. c. *To the tables we inherited from Aunt Selma was given a good scrubbing. d. The tables we inherited from Aunt Selma were given a good scrubbing. In (27b) and (28c), locative inversion is ungrammatical, in contrast with passivization in (27c) and (28d). The locative inversion examples (27b) and (28c) are very strange, at best. Similarly, locative inversion with R-dative-shifted give X a headache only has a funny handing-over reading, where someone literally hands over a headache. (29) a. #To anyone who stays too long in this room is given a horrible headache. b. Anyone who stays too long in this room is given a horrible headache by the flickering lights and foul odor. R E M A R K S A N D R E P L I E S In this it contrasts with the passive, where the usual sense of give X a headache holds. Moreover, passives of double object constructions do not participate in locative inversion at all (30) a. At that battle were given to the generals enough helicopters to block out the sun. b. *At that battle were given the generals enough helicopters to block out the sun. This holds equally of R-dative-shifted double object constructions, again distinguishing them from prepositional datives like (30a). (31) a. *At that battle was/were given the generals who lost hell. b. *At that battle was given to the generals who lost hell. (32) a. *In that room is/are given anyone who stays long enough horrible headaches that don't go away. b. *In that room are given to anyone who stays long enough horrible headaches that don't go away. (33) a. *At that time was/were given the tables we inherited from Aunt Selma a good scrubbing. b. *At that time was given to the tables we inherited from Aunt Selma a good scrubbing. (34) a. *With that Supreme Court decision were denied minorities some of the rights they had fought so long to enjoy. b. *With that Supreme Court decision were denied to minorities some of the rights they had fought so long to enjoy. Compare: c. With that Supreme Court decision were taken from minorities many of the rights they had fought so long to enjoy. Again, R-dative shift does not pattern like the prepositional dative. It patterns like the double object construction. Taken together with the scope facts above, this fact indicates that the double object construction is distinct from the prepositional dative, even when they look identical on the surface because of R-dative shift. 8 8 An anonymous reviewer suggests that some unnamed constraints on locative inversion that have been described in the literature will account for the failure of locative inversion with R-dative shift, without the need to posit an operation of R-dative shift. I have been unable to find any constraint in the literature that will do the job. The reviewer suggests one possible constraint, namely, that transfer of possession must be involved for locative inversion to be acceptable. He or she cites To the receptionist was given my name as ungrammatical and as proof that transfer of possession is necessary. However, it is clear that locative inversion in general does not require any type of transfer (Under the bridge lived a fat, ugly troll). The reviewer's example also becomes perfectly acceptable if the postverbal NP is made longer: To the receptionist should be given the name of anyone who comes into the office. Moreover, no constraint on what can be fronted in locative inversion will explain the ungrammaticality of htm). R E M A R K S A N D R E P L I E S 299 The Nature of R-Dative Shift As stated above, I analyze R-dative shift as being identical to the normal double object construction, except that the specifier of ApplP is projected on the right. This rightward projection is only licensed by Ā -extraction of the NP. One such type of extraction is heavy shift: when the goal NP is prosodically heavy, it can move and right-adjoin to VoiceP. An alternative would be to move the goal NP directly to VoiceP, from a normal leftward specifier. R-dative shift would then consist of this rightward Ā -movement, which could be licensed by wh-extraction or by prosodic heaviness. There are several reasons not to take this route, and to distinguish R-dative shift from heavy shift. First, if R-dative shift were rightward movement, adjoining the goal NP to VoiceP, we would expect it to be possible for the goal NP to appear outside of adjuncts, or at least have some freedom of ordering with respect to adjuncts. This is not true, however. When R-dative shift takes place because of extraction of any type besides heavy shift, to must appear inside of all adjuncts. (35) a. Who did he give the creeps to last night? b. *Who did he give the creeps last night to? (36) a. The man that he gave the creeps to last night is over there. b. *The man that he gave the creeps last night to is over there. If adjuncts adjoin higher than ApplP (to VoiceP, say), the rightward-specifier analysis predicts this ordering restriction. When R-dative shift is licensed by prosodic heaviness, then the goal NP can follow adjuncts. (37) a. He gave the creeps last night to everyone he engaged in conversation with. b. This lighting gave headaches for years to whoever had to work in its harsh glare. This indicates that heavy shift operates on top of R-dative shift, moving the goal NP even further to the right than the position where it appears in (35a). Second, as pointed out by an anonymous reviewer, Ā -movement typically renders a PP impermeable to further extraction (e.g., Wexler and Culicover 1980). (38) a. I think that, to a person like him, you should never give anything. b. *Who do you think that, to t, you should never give anything? This holds of heavy shift to the right, as well. (39) a. I gave some money on Saturday to every boy who had helped me clean the yard. b. *Who did you give some money on Saturday to? In contrast, as discussed above, one of the licensing contexts for R-dative shift is precisely extraction of the goal NP, stranding the preposition to, as in (35a). If R-dative shift is rightward Ā -movement, like heavy shift, this would not be expected to be possible. Hence, R-dative shift must consist of base-generation in a rightward specifier. Furthermore, when prosodic heaviness shifts a goal NP further, over adjuncts, then the goal does become impermeable to further extraction. So, again, we need to distinguish R-dative shift, base-generation in a rightward specifier, from an additional operation of heavy shift. To summarize, then, R-dative shift simply consists of projecting the specifier of ApplP in a double object structure to the right. This rightward projection is licensed only when the goal NP undergoes Ā -extraction from the rightward specifier. Heavy shift is one type of licensing extraction, moving the goal NP to right-adjoin it to VoiceP. Some Consequences for Weak Crossover and Variable Binding The topic of this article has been operations that reorder elements to the right, like R-dative shift and heavy shift. Because rightward reordering processes have not, to my knowledge, been systematically investigated in the large literature on weak crossover and variable binding, I end this article with some data on these topics. Theories of these phenomena will have to incorporate these data; hopefully they will lead to improved understanding. It is not my purpose here either to construct a complete account of the phenomena or to provide a comprehensive overview of the extensive literature on these topics, so the presentation will be rather sketchy. The first observation is that wh-extraction with R-dative shift gives rise to weak crossover. (41) a. *Who 1 did you spare his 1 ordeal to? b. *Who 1 did the boss deny his 1 pay to? Both of the questions in (41) are reasonably grammatical if his is interpreted as referring to someone else; (41a) is still odd, though, because it is strange to spare someone someone else's ordeal. But (41b) is perfect on the nonbound reading. If R-dative shift is identical to the double object construction except in linear order, hierarchical accounts of weak crossover (e.g., Moreover, existing linearity accounts may need to be modified. If the condition is that the trace of the moved quantifier must precede the pronoun that it needs to bind, as in Chomsky 1976, Higginbotham 1980, and An anonymous reviewer suggests that variable binding becomes degraded with heavy shift in general, and this generally seems to be true. Similarly, Phillips (1997) notes that heavy shift also seems to give rise to weak crossover ( (44) a. I gave money to every boy 1 on his 1 birthday. In all of these cases, the base position of the shifted element precedes (and probably also c-commands) the pronoun. Linear accounts of weak crossover based on leftness No NP mentioned in the structural index of a transformation may be reordered by that rule in such a way as to cross over a coreferential NP. (Ross 1967:73, (4.30)) Ross's Crossover Condition makes no reference to directionality, so crossing over a pronoun in either a leftward or a rightward direction is expected to violate it. 9 However, I disagree with this reviewer (as do my consultants) in finding the effect only when the heavy NP shifts over an adjunct PP, as in (i) a. John put back on its 1 pedestal [every sculpture that the rioters had knocked down] 1 . b. John gave back to its 1 author [every unpublished manuscript in the collection] 1 . Examples (43a) and (ia-b) are from the reviewer; he or she gives (ia-b) two question marks, but I and others I have asked find them acceptable, in contrast with (43a-c). Another possibly relevant difference between (ia-b) and (43a-c) is the choice of the pronoun: its versus him/her. However, the following example appears to indicate that the argument/ adjunct distinction is more important than the choice of the pronoun: (ii) ??Thelma broke in front of its 1 creator [every sculpture that she had commissioned earlier that year] 1 . R E M A R K S A N D R E P L I E S However, there is another potential reason for the failure of variable binding in these heavy shift examples. This arises from the fact that R-dative shift plus heavy shift also seems to destroy the possibility of variable binding, strongly for some speakers and more weakly for others. The other response, which I will advocate here, is to suggest that there is a linear constraint on variable binding in addition to a scope (c-command) constraint, along these lines: (48) A quantifier Q can bind a pronoun P as a variable only if a. Q takes scope over P (Q c-commands P) and b. Q precedes P. The condition in (48a) is inviolable; the strength of (48b) varies from speaker to speaker. Then, even though the goal NP takes scope over the pronoun in (46b) and (47b), it will be unable to bind it as a variable because it does not precede it. The reason to prefer this account is that it unifies judgments on (46b) and (47b) with similar judgments where there is no rightward shift of any kind. For example, as we saw above, the goal NP can easily take scope over the theme NP in the prepositional dative construction (49a), but many people find variable binding degraded in (49b), which is cited as ungrammatical by Williams (1995:155, (94b)). (49) a. I gave a copy of the syllabus to every boy. However, the same people who find (46b) and (47b) only slightly degraded, as I do, make the same judgment about (49b). It therefore appears desirable to account for them in the same way, with a (weak) precedence condition like that in (48b). Note that the constraints in (48) are not meant to account for weak crossover. For speakers like me, weak crossover is quite strong with overt wh-movement, but the effects of (48b) are very weak in examples like (46b), (47b), and (49b). It therefore seems undesirable to formulate (48b) in such a way that it would also encompass weak crossover. But now the question arises of whether the failure of variable binding with heavy shift illustrated above is due to weak crossover or to the condition in (48b). If it is not due to weak crossover, then existing linearity accounts of weak crossover are perfectly adequate, and there is no need to use Ross's Crossover Condition. I will not attempt to decide this here, but will simply end with the conclusion that variable binding has a strong linear component. There appears to be a precedence condition (48b) on top of a scope requirement (48a) for variable binding, and a linear approach to weak crossover fares much better than a hierarchical one for the data discussed here. The best account of weak crossover may refer to leftness, or it may refer to crossing, but either way, it is about linear order and not hierarchical order. Conclusion I have argued here that some apparent instances of the prepositional dative construction are actually double object constructions that have undergone R-dative shift. In scope possibilities and their interaction with locative inversion, they pattern exactly like double object constructions and do not pattern like prepositional dative constructions. Hence, there are nonalternating uses of give, like give X a headache and give X the creeps, and nonalternating verbs like deny and spare, which truly do never alternate. When they appear with to, as in Bresnan et al.'s examples, they are still double object constructions, but they have undergone R-dative shift, which I have suggested involves simply projecting a specifier on the right rather than on the left. The facts presented here therefore indicate that the grammar distinguishes the double object construction from the prepositional dative construction, both syntactically and semantically. The two constructions are not syntactically and semantically indistinguishable, as Bresnan et al. argue. They appear to have distinct semantics, such that phrases like give X a headache can only be double object constructions, and can never be prepositional dative constructions. 10 Finally, the rightward reordering operations investigated here indicate that constraints on variable binding, like weak crossover, need to be formulated in linear rather than hierarchical terms. 10 The other main claim made by Bresnan et al. is that the factors governing the choice of syntactic frame, and a probabilistic algorithm for making that choice, are part of speakers' grammar. However, this claim is never backed up with evidence that the choice between syntactic frames could not be part of language use rather than grammar. Speakers have to make such choices in every situation, not just those involving ditransitives. They must choose whether to use an active or a passive, for instance, to describe the same event (The dog chased the cat vs. The cat was chased by the dog). Many of the same factors identified by Bresnan et al. go into making that choice: discourse accessibility, animacy, givenness, definiteness, and so on (e.g., Ferreira 1994 and references there). Speakers must decide whether to apply optional transformations, like topicalization; whether to use a simple declarative or a cleft; whether to use control (He decided to leave) or a finite clause (He decided that he should leave); and so on. Bresnan et al. provide no compelling argument that such choices must be internal to the grammar, rather than part of language use. R E M A R K S A N D R E P L I E S References

Abstract not found

Abstract not found

1 When a complement PP goes missing: a study on the licensing of Swiping

operation “move α ” in the previous framework is reconsidered as a combination of two simpler