All of us are familiar with the standard argument that deep structure need not satisfy binding conditions. Who has not done or assigned a homework problem based on an example like (1)?

(1) They seem to each other to be clever

The reciprocal each other requires an antecedent, and the antecedent must c-command it, a special case of Condition A of the binding theory. Under standard assumptions, (2), the deep structure of (1), would violate the condition.

(2) _ seem to each other [they to be clever]

Given that deep structure configuration determines theta-relations, the above argument is simultaneously an argument that binding conditions are, at least in part, independent of argument structure and are, instead, geometric. I propose to examine a range of such cases in light of recent developments in syntactic theory. I will be particularly concerned with the precise configurations created by movement, of both the overt and covert variety, and the binding theoretic properties of those configurations.

It is well known that in raising constructions like (1), an anaphor can be bound in the embedded clause as well as the matrix clause:

(3) They seem to have criticized each other

At this point, no interesting conclusions can be drawn from this additional binding possibility. The following theories (among others) would all be consistent with the facts: (a) Condition A can be satisfied at D-structure or S-structure, with they binding each other; (b) Condition A can be satisfied anywhere in the course of the derivation (as in Belletti and Rizzi (1988)); (c) Condition A is satisfied strictly at S-structure, with the trace of they, rather than they itself, serving as the binder of each other; (d) like (c), except that they is the binder, the governing category of each other potentially extending to the matrix clause.

Possibility (a) is excluded by still more complicated examples of the same general type, but in which the anaphor is in neither the D-structure clause nor the S-structure clause of the antecedent:

(4) They are certain to seem to each other to have criticized Harry

The other three possibilities (among others) remain viable.[1]

So far, I have not considered the possibility that LF is the level relevant to binding conditions, that is, that movement that creates a licit binding configuration might be covert. Such a possibility is of great significance within recent 'minimalist' approaches that dispense with S-structure (and, for that matter, D-structure). The fact that anaphora has obvious semantic aspects has always suggested that its syntax, the binding conditions, should be determined at LF, the syntax- semantics interface level, as minimalism evidently demands. However, Chomsky (1981) presented a very powerful argument that at least Condition C must be satisfied at S-structure. Chomsky observed that examples like the following cast grave doubt on a purely LF approach:

(5)  Which book that John_i read did he _i like

(6) *He_i liked every book that John_i read

(7) *I don't remember who thinks that he_i read
      which book that John_i likes

Chomsky's point is that following QR, the LF of (6) would be structurally parallel to the S-structure (and LF) of (5), where John is outside the c-command domain of he. Thus, as in (5), there should be no Condition C effect if LF is the level relevant to that condition. Similarly for (7) following LF WH-movement. Contrary to the prediction of the LF theory, in both instances the hypothesized LF movement, unlike the overt movement creating (5), has no effect on binding possibilities. This strongly suggests Chomsky's conclusion: that Condition C is a requirement on S-structure.

Based on examples like the following, Barss (1986) extends Chomsky's argument to Condition A, the concern of the initial discussion of the present paper.

(8)  John_i wonders which picture of himself_i Mary showed to Susan

(9) *John_i wonders who showed which picture of himself_ito Susan

(8) shows that an anaphor within the embedded CP SPEC can be licensed by an antecedent in the matrix subject position. Given this fact, the ungrammaticality of (9) is surprising if anaphors can be licensed at LF. On the then standard theory (essentially the one also assumed in Chomsky (1981)), in LF, the WH-phrase in situ, which picture of himself, moves to the embedded Comp (i.e., CP Spec) position, where it takes scope. Thus, at LF, the configurational relation between himself and its antecedent is virtually identical in (8) and (9). Hence, the ungrammaticality of (9) shows that anaphors must be licensed at a level prior to LF, e.g., S-structure (and possibly at LF as well; the argument is silent on this point). (9) is out because the reflexive fails to be licensed at that level. As with Chomsky's argument, the conclusion carries over to QR as well:

(10) *The men think that Mary likes every picture of each other

(11)  The men think that every picture of each other, Mary likes

The results of QR applied to the S-structure of (10) would be very similar to the topicalized version in (11). If Condition A can be satisfied at LF, the acceptability contrast between (10) and (11) is unexpected.

On the other hand, Uriagereka (1988) presents evidence that covert movement can indeed create new binding possibilities. Chomsky (1986) had proposed that in existential constructions such as (12), the 'associate' of there covertly moves to the position of there, producing (13).

(12) There is a man in the room

(13) A man is t in the room

One of the major arguments for such covert movement is the well-known superficially bizarre agreement paradigms displayed by existential constructions, with the verb agreeing with something that is not its formal subject:

(14) a. There is/*are a man here
       b. There are/*is men here

Based on the acceptability of (15), Uriagereka proposes that anaphors need not be licensed at S-structure, reasoning that the required c-command relation between two knights and each other holds at LF (16) but not at S-structure (15).[2]

(15) There arrived two knights on each other's horses

(16) Two knights arrived t on each other's horses

The general phenomenon underlying Uriagereka's argument is actually independent of expletive replacement per se, since, as discussed in some detail by Lasnik and Saito (1991), even direct objects of transitive verbs generally behave as if they c-command into certain adjuncts:[3]

(17) I saw two men on each other's birthdays

Under conventional (though not universal[4]) assumptions about deep phrase structure configuration, two men would not be expected to c-command each other, parallel to the fact considered by Uriagereka. In certain versions of Case theory, for example one proposed by Chomsky (1991), and examined further by Lasnik and Saito (1991) and Lasnik (1993), in the LF component a direct object raises to Spec of Agr_O, where its Case is licensed. (17) might then be a further example of successful licensing of an anaphor at LF (via A-movement) remedying failure in overt syntax. The near contradiction between Barss's examples on the one hand, and Uriagereka's and Lasnik and Saito's on the other remains to be resolved. I will return to that task.

Before proceeding, I would like to consider a plausible objection to the classic analysis of (1), repeated as (18).

(18) They seem to each other to be clever

Recall that a major argument for the geometric approach to binding rests on the standard assumption that (18) crucially involves movement, raising of the lower subject into matrix subject position. However, Lasnik and Saito (1992) argued, on grounds quite independent of the issues under consideration here, that at least some apparent raising examples are actually structurally ambiguous. According to Lasnik and Saito, (19), for example, is ambiguous between (20) and (21).

(19) John is likely to win

(20) John₁ is likely [t₁ to win]

(21) John₁ is likely [PRO₁ to win]

Lasnik and Saito were concerned with the intriguing contrast between (22) on the one hand and (23), (24) on the other.

(22)  How likely to win is John

(23) *How likely to be a riot is there

(24) *How likely to be taken of John is advantage

Saito (1989), following a suggestion of Tony Kroch, proposed that (23) and (24) are excluded by essentially the Proper Binding Condition of Fiengo (1977), given the representations in (25) and (26).

(25) [how likely t₁ to be a riot]₂ is there₁ t₂

(26) [how likely t₁ to be taken of John]₂ is advantage₁t₂

In neither of these representations is t₁ bound, in violation of the condition. But we should then expect (22) to be bad as well, if it must involve raising. But if it can, as an alternative, involve control, then there is no a priori reason to expect it to be bad. This constitutes the heart of Lasnik and Saito's argument that (at least some) classic raising structures have a base-generated alternative. Note that (23) and (24) are still correctly ruled out, since existential there and idiom chunks cannot appear as controllers of PRO, not being arguments. But if these examples must involve raising, then the Proper Binding Condition cannot be evaded.

While the particular WH-movement paradigm appealed to above is only relevant with raising adjectives, Martin (1992, 1996) argues that the conclusion is actually more general. Martin provides extensive evidence that there are two distinct infinitival Infl's, one associated with raising (including ECM, an instance of raising to Spec of Agr_O, on recent accounts), the other with control. One of the several diagnostics he provides concerns the licensing of VP ellipsis. The Infl associated with PRO licenses ellipsis of its complement; the Infl associated with raising does not. (27) illustrates the former possibility.

(27) John tried to be courageous, and Mary tried to also

And (28) and (29) illustrate the latter, with ECM and passive-raising respectively.

(28) ?*I believe John to be courageous, and I believe Mary to also

(29) ?*John is believed to be courageous, and Mary is believed to also

Raising adjectives and verbs at first seem not to conform to this pattern:

(30) ?John is likely to be courageous, and Mary is likely to also

(31) ?John seems to be courageous, and Mary seems to also

However, when we replace the matrix subject with an NP that couldn't be a controller, hence must have raised, the results are in accord with the predicted pattern:

(32) *John said there was likely to be a solution, and there is likely to

(33) *John said there seems to be a solution, and there does seem to

Of direct relevance to the discussion here is Martin's further observation that when 'raising' predicates occur with an experiencer, they pattern strictly as raising, the control properties disappearing. For example, in (34), the addition of the experiencer renders VP ellipsis much less acceptable.

(34) Although John didn't actually hit Bill, he seemed(?*to Mary) to

Since the crucial example (18) necessarily contains an experiencer, by Martin's argument it must involve raising. Thus, the potential alternative to movement into a binding configuration is eliminated.

In recent years, Chomsky has argued at length that movement is invariably driven by the need for formal features to be checked. Thus, it is really only formal features that are involved in the operation of movement. One would then expect that movement would never be of an entire syntactic category, but only of its formal features. However, PF requirements will normally force movement of a category containing the formal features, via a sort of pied-piping, under the reasonable assumption that a bare feature (or set of features) is an ill-formed PF object. If only features 'really' move, and given that movement creates binding configurations, we might be led to conclude that binding involves precisely a relation between features. In fact, Chomsky (1995) argues for exactly that position. I will return to his arguments. Before that, I will consider how this feature movement approach resolves a long-standing problem with expletive replacement. In Chomsky's original proposal (1986), the associate literally replaced the expletive, giving LF (36) for S-structure (35).

(35) There is a man here

(36) A man is t here

But true replacement also gives identical LFs for (37)and (38).

(37) A man is likely to be here

(38) There is likely to be a man here

Yet the interpretive possibilities diverge, as was immediately noted. In (37), a man can evidently have wide or narrow scope with respect to likely, while (38) allows only narrow scope. Chomsky (1991) observes a similarly incorrect scope prediction for (39).

(39) There aren't many linguistics students here

If many linguistics students were to substitute for there, it would be expected to have wide scope with respect to negation, but, in fact, it has only narrow scope. The feature movement analysis of existential constructions has the potential to solve these scope problems. For instance, if in LF, only the formal features of many linguistics students, rather than the entire expression, move to a functional head or heads above negation, it is reasonable to conclude that the quantificational properties remain below negation. Then, if it is this structure that determines scope (that is, if QR either cannot alter these hierarchical relations or does not exist) the desired results are obtained.

Chomsky proposes that binding crucially differs from scope with respect to feature movement. While such movement does not alter scope relations, Chomsky claims that it does alter binding relations. Notice that under this assumption, a version of Uriagereka's account (15)-(16) above, repeated here as (40)-(41) is immediately available.

(40) There arrived two knights on each other's horses

(41) Two knights arrived t on each other's horses

In LF, two knights does not replace there. Rather, the formal features of two knights raise to a functional head (presumably Agr_S, given the agreement properties of the construction) in which position they bear the appropriate command relation to each other. A very similar analysis is apparently available for the Lasnik and Saito example (17), repeated as (42).

(42) I saw two men on each other's birthdays

The features of the direct object two men raise to Agr, Agr_O in this instance, and in that position command the temporal adverb if, for instance, it is adjoined to VP.

Thus far, we seem to have a principled and empirically supported dichotomy: LF movement (at least of the 'A-movement' variety) has binding effects but no scope effects: binding involves merely formal features whereas scope involves quantificational properties that are not (or not just) formal features. Notice too that Barss's Condition A problem is also solved. Recall that in his examples, LF movement of a phrase containing an anaphor fails to create a new licensing configuration for that anaphor, even though seemingly analogous overt movement does create a licensing configuration:

(43) *John_i wonders who showed [which picture of himself_i] to Susan

(44)  John_i wonders [which picture of himself_i] Mary showed to Susan

However, under the feature movement analysis of LF movement, in (44), even if there is LF WH-movement, only the formal features of which picture of himself would move, leaving the anaphor in the lower 'governing category'.

Chomsky's Condition C problem receives a parallel solution. In Chomsky's examples, LF movement of a phrase containing an R-expression fails to remove that R-expression from the command domain of a pronoun, while analogous overt movement does remedy the potential Condition C violation:

(45) *He_i liked [every book that John _i read]

(46) *I don't remember who thinks that he_i read
        [which book that John_i likes]

(47)  [Which book that John_i read] did he_i like

By (new) hypothesis, only the formal features of the bracketed expressions in (45) and (46) move, leaving John in the command domain of he in violation of Condition C.

Successful though the approach has been, it cannot be maintained in its present form. Contrary to Chomsky's claim, there is strong evidence that LF feature movement does not affect binding possibilities. Lasnik and Saito (1991) and den Dikken (1995) show that 'expletive replacement' does not in general create new binding configurations. For example, as mentioned above, it is well known that a raised subject can antecede an anaphor in the higher clause (the classic argument that D-structure need not obey Condition A). But the associate of a raised expletive cannot:

(48) Some linguists seem to each other [t to have been given
       good job offers]

(49) *There seem to each other [t to have been some linguists
       given good job offers]

This is entirely unexpected under the binding by features analysis. That analysis predicts an asymmetry between binding and scope, but we find no such asymmetry this time. Binding from the high position of there (or from the associated Agr_S) is no more possible than high scope, as in (50), even though the agreement form of the higher verb shows that the features of the associate have raised.

(50) There seem to have been some linguists given good job offers

Other phenomena known to involve A-binding pattern similarly. (51) illustrates this for weak crossover.

(51)a. Some defendant_i seems to his_i lawyer to have been at the scene
      b. *There seems to his_i lawyer to have been some defendant_i
           at the scene

Overtly raised some defendant can bind his, obviating the potential weak crossover (WCO) effect in (51)a. But there is no such amelioration in (51)b. Once again, if raised features could act as binders, (51)a and (51)b would have the same status.

At this point, we have one strong generalization alongside an apparent paradox. The generalization, valid for all the data considered so far, is that the scope of the associate of there is always 'low': feature movement does not create new scope possibilities. The paradox is that feature movement both does and does not create new binding possibilities. Uriagereka's existential example seems to indicate the binding potential of raised features, as does the Lasnik and Saito transitive example. These two examples are repeated as (52) and (53).

(52) There arrived two knights on each other's horses

(53) I saw two men on each other's birthdays

By hypothesis, in (53) the features of two men raise to Agr_O and in that position bind each other. Likewise, in (52) the features of two knights raise to Agr_S, and in that position bind each other. But seemingly identical movement does not suffice in the ungrammatical (49) and (51)b. Suppose, then, that feature movement to a high position never suffices for binding. (49) and (51)b are then immediately explained, but the well-formed (52) and (53) now demand a new explanation. I turn to the task of finding one.

Larson (1988) provides one possibility. In his footnote 49, Larson suggests that, contrary to usual assumptions, adverbials are projected in innermost complement position, thus are, in D-structure, lower than direct objects. Feature raising would then be irrelevant for the anaphor binding in (52) and (53); the NP complement would itself be in position to bind the reciprocal. (49) and (51)b would be correctly excluded since the to phrase, being an argument of the higher predicate seem could not originate lower than an underlying argument of the lower predicate, some linguists or some defendants. However, several of the paradigms examined by Lasnik and Saito (1991) cast doubt on this Larsonian approach as a general solution. Consider first (54), where the associate of there in an ECM clause fails to bind a reciprocal in the higher clause.

(54) *The DA proved [there to have been two men at the scene]
         during each other's trials

So far, this is just as predicted, since there is no way that the adverbial modifying the higher predicate could have started lower than the complement of the embedded clause. But (55) is not as predicted.

(55) The DA proved [two men to have been at the scene]
       during each other's trials

As discussed by Lasnik and Saito (and much earlier by Postal (1974)), the downstairs ECM subject behaves like an upstairs direct object in its ability to bind an anaphor in the higher clause. This, in fact, was one of the two principal arguments in Chomsky (1995) that raised features can serve as binders, a possibility we have been forced to reject. But the Larsonian alternative now under consideration seems to exclude (55) as much as (54). Once again, there is no obvious way that the upstairs adjunct can originate lower than the downstairs argument.

The asymmetry seen in (54)-(55) is not limited to reciprocal licensing. All standard c-command phenomena display the same pattern. (56)-(57) illustrate the corresponding weak crossover asymmetry:

(56) The DA proved [no suspect_i to be at the scene of the crime]
      during his_i trial

(57) *The DA proved [there to be no suspect_i at the scene
       of the crime] during his_i trial

And negative polarity item licensing is shown in (58)-(59).

(58) The DA proved [no one to be at the scene] during
      any of the trials

(59) *The DA proved [there to be no one at the scene] during
       any of the trials

The state of affairs thus far can be summarized as follows: In situations of unquestionable overt raising (e.g., subject raising to subject position), we get high binding behavior. In situations of unquestionable non-overt raising (e.g., there constructions), that is, situations where there is covert raising (features only, by hypothesis) if there is raising at all, we get no high binding behavior. Finally, and problematically, with direct objects and ECM subjects, which are generally assumed not to undergo overt raising, we do get high binding behavior. Of course ECM subjects have not always been thought to remain in the lower clause in overt syntax. Postal (1974) presented 434 pages of arguments that such subjects do raise into the higher clause. A number of these arguments were already addressed by Chomsky (1973). But a substantial percentage were not, including in particular, ones based on the high behavior of ECM subjects with respect to anaphor binding. Lasnik and Saito (1991) summarize several of these 'height' arguments, add a few variants on them, and conclude that Postal was basically right, but that his specific analysis was wrong. Postal argued that the lower subject raises to the higher direct object position. However, as hinted above, direct object position is itself not high enough.[5] Lasnik and Saito observe that Spec of Agr_O, the position associated with accusative Case, is plausibly high enough, under the reasonable (though not obviously necessary) assumption that the adjuncts containing the anaphor to be bound are adjoined to VP. An ECM subject, like a higher direct object, would move to that position. Interestingly, though, the Lasnik and Saito paper ended with a paradox of its own: while they assumed that LF raising was of the entire NP and not just its formal features, they presented substantial evidence, some of it mentioned at the beginning of the present paper, that it is S-structure configuration, and not LF, that determines binding potential. And yet that the relevant raising was an LF phenomenon was virtually unquestioned at the time. However, the empirical arguments against this view still seem sound. And added to them, we now have the conceptual economy-based argument that LF movement is just of features.

There is, I believe, a natural way to resolve the evident contradiction. Koizumi (1993, 1995), revising and extending ideas of Johnson (1991), argues that accusative Case is checked overtly in English, just like nominative Case.[6] The accusative NP overtly raises to Spec of Agr_O (with V raising to a still higher head position, the V head of a 'shell' VP). If this is correct, the seemingly paradoxical asymmetry is immediately reduced to the independent pied-piping asymmetry in Chomsky's theory of movement. In the there construction, the only movement is the covert movement of the formal features of the associate to the Agr head. For an ECM subject or the object of a transitive or unaccusative, the movement is overt, hence, of the entire NP. Under this circumstance, all of the semantic/referential properties necessary for licensing anaphors, NPIs, and bound pronouns are raised. Plausibly, when only features move, these properties are left behind, along with the quantificational/scope properties that Chomsky already argued are left behind.

Earlier, I mentioned that Chomsky (1995), contrary to what is argued here, claims that raised formal features do license anaphors. His claim is based on two phenomena. The first is the Lasnik and Saito paradigms that I have discussed in some detail. Those paradigms provide strong evidence that there is relevant raising. But to resolve the apparent contradiction with the inability of the associate of there in existential constructions to display high behavior, I have suggested, following Koizumi, that raising to accusative Case position is actually overt.[7] Since overt raising is of the entire constituent, these facts (and the corresponding facts with simple transitives) provide no evidence for binding potential of mere features.

Chomsky's second argument involves control. He presents an example suggesting that the associate of there behaves as if it is high in its ability to control PRO in an adjunct:

(60) There arrived three men (last night) without [PRO identifying
       themselves

As (61) shows, a typical object cannot control PRO in this construction:

(61) *I met three men (last night) without identifying themselves

Thus, Chomsky reasons that feature raising, in this instance to Agr_S, does create new control configurations, and, a fortiori, new binding configurations. Momentarily, I will question the generality of the control phenomenon, but even if Chomsky turns out to be correct about control, it might not be necessary to draw a broader conclusion about binding. There are significant semantic and syntactic differences between binding and control. This issue would take us too far afield here, but see Lasnik (1992a) for some discussion.

At this point, I would like to examine a bit further the interesting control phenomenon brought up by Chomsky. Chomsky implies that the associate of there in (60) is behaving just as an overtly raised subject, as in (62), would.

(62) Three men arrived (last night) without PRO identifying themselves

But already there is some difference. While (62) is perfect, (60) is somewhat degraded for many speakers. This contrast is heightened if the adverbial is fronted:

(63) Without PRO identifying themselves, three men arrived

(64) ?*Without identifying themselves, there arrived three men

Further, under raising, the contrast between structures like (60) and (62) greatly intensifies. In the following examples, the adverbial is intended as being in the higher clause, along with the raised subject or there:

(65)  Someone seems to be available without PRO seeming to be eager
       to get the job

(66) *There seems to be someone available without PRO seeming
        to be eager to get the job

It is not obvious why this should be so if raised features of the associate of there can control. Finally, other examples structurally identical to the original (60) are considerably less acceptable, as seen in (67).

(67) ?*There arrived three men (last night) without PRO saying hello

Given all of these facts, though I agree with Chomsky that (60) is reasonably good, I suggest that we should at least consider the possibility that that example is not representative, and that even for purposes of control, feature raising doesn't suffice.

One problem would still remain with respect to the control constructions though. Consider once again Uriagereka's example, repeated here:

(68) There arrived two knights on each other's horses

I have rejected expletive replacement as the explanation for the binding here. Further, I have rejected feature raising as the correct account, for reasons discussed at length above. That seems to leave no account at all. But recall now Lasnik and Saito's observation that even objects of standard transitives have similar binding potential:

(69) I saw two men on each other's birthdays

The account I have presented for the latter fact involves overt raising of the complement of the verb to Spec of Agr_O. Elsewhere (Lasnik 1992b, 1995c) I have argued, following Belletti (1988), that unaccusative verbs do license a Case on their complements (though not accusative Case). Belletti gave a strong argument that the Case in question is inherent rather than structural. Now it is generally assumed that inherent Case is licensed in the head complement configuration, hence does not involve any raising. However, Lasnik(1995c) questions that assumption, arguing that all Case, whether structural or inherent, is licensed in the same Spec configuration. There, I assumed that the movement to that position is covert, but given the perspective of the present paper, it might well be overt. Below, I will have more to say about the claim that inherent Case is structurally parallel to structural Case, but for the moment, note that if it is correct, (68) can potentially receive exactly the same treatment as (69), with two knights raising overtly to Spec of Agr_O, where it can bind the anaphor, and arrive raising still higher. Chomsky's example (62) could be nicely accomodated in just this way. Further, the ungrammaticality of (66) is still predicted: the associate of there remains in the lower clause. But (61) is mysterious, as is (67). The first of these two examples plausibly receives a non-structural account, as control is known to involve (in addition to structure) thematic properties.The second, though, I will have to leave for future research.

Evidence that inherent Case is 'structural' is presented by Stjepanovi (1996). Following Lasnik(1995c), Stjepanovic pursues 'symmetry' in the Case system, the idea that all Case is licensed in the same structural configuration. Recall that early minimalist arguments against 'government' had the same basic motivation.[8] Stjepanovi explores inherent Case in Serbo-Croatian (SC) in detail, and presents compelling evidence that inherently Case marked NPs are just as high as structurally Case marked NPs. Stjepanovi first observes that SC has many verbs with the lexical property of licensing Case other than accusative (hence presumably inherent) on their complements. For example, 'vladati' (rule) and 'ovladati' (master) license instrumental, 'pomo i' (help) licenses dative, and 'sjetiti' (remember) licenses genitive. With this much as background, consider first the binding behavior of accusative object:

(70) Slikao je Samprasa i Ivanisevic a za vrijeme me
       meca jednog protiv drugog

       'He photographed Sampras and Ivanisevic (Acc.) during
       each other's matches'

As in English, the accusative object can bind into the adverbial. Now observe that a dative object has the same binding potential:[9]

(71) Pomogao je Samprasu i Ivanisevicu u za vrijeme
       meca jednog protiv drugog

      'He helped Sampras and Ivanisevic (Dat.) during
       each other's matches'

Similarly, Stjepanovic shows that a dative quantifier, just like an accusative quantifier, can successfully bind a pronoun within an adverbial, thus obviating WCO:

(72) Ona kritikuje svakog_i bez njegovog_i znanja
       'She criticizes everyone (Acc.) without his knowing'

(73) Ona pomogne svakom_i bez njegovog_i znanja
       'She helps everyone (Dat.) without his knowing'

Stjepanovic concludes from a range of facts like these that both structurally and inherently Case marked NPs raise to Spec of Agr_O.[10] She does not commit herself on whether the movement is overt or covert, but given the discussion above, I conclude that it is overt.

Given the conceptual argument for symmetry in the Case licensing system, and the facts from SC in support of it, there is reason to doubt a conjecture of mine in Lasnik (1993). There, I suggested that reflexives are licensed via movement to Agr, to be in a Spec-Head relation with their antecedents in Spec of Agr. I further conjectured that in languages where reflexives are strictly subject oriented, non-nominative Cases are all inherent, and therefore licensed in situ. S.-W.Kim (1994) has a somewhat similar proposal. Kim shows that the height effects seen in English ECM constructions do not arise with Korean ECM and concludes that the ECM subject does not raise to Spec of Agr_O (in fact, that there is no Agr_O in Korean). He gives several paradigms, one of them centering on caki, one of the Korean reflexives. The crucial ECM example is reproduced in (74).

(74) *Ku kemsa-nun [Chelswu_i-lul ywucoy-lako] caki_i-uycaypan-eyse
       cungmyengha-ass-ta
       'The prosecutor demonstrated [Chelswu (Acc.) to be guilty]
       in his (reflexive) trial'

However, J.-S. Kim (1996) offers another interpretation of the facts. J.-S. Kim suggests that Korean differs from English not in how Case is licensed, but rather in when it is. In English, raising to the Case position is overt (driven by a strong feature). If the relevant feature is weak in Korean, then the movement will be covert, and hence just of formal features. As seen above, new binding configurations would then not be created.

This would be a good place to end this discussion, except that conscience forces me to acknowledge a puzzling array of facts in Japanese, presented by Yatsushiro (1996). So far, with the possible exception of one rather unclear control example, (67), an overwhelmingly consistent pattern has emerged: scope and binding go together, and both are 'low' except when there is overt raising. The analysis, slightly modifying Chomsky (1995), is that covert raising affects only formal features, but that scope (Chomsky's proposal) and binding (my extension) involve more than formal features. The mysterious fact about Japanese is that scope and binding apparently diverge, and in just the way that Chomsky claims (evidently incorrectly) that they do in English. Yatsushiro provides strong arguments that in Japanese unaccusative constructions, the complement of the unaccusative verb invariably remains in its underlying position at S-structure. It is then not surprising that the complement has 'low' scope (even if it raises in LF, since that raising would be just of formal features). According to Yatsushiro, in (75) daremo-ga ('everyone') necessarily takes narrow scope with respect to dokoka-ni ('somewhere').

(75) Dokoka-ni daremo-ga ita
       somewhere-Loc everyone-Nom be-past
       'Everyone was somewhere'

       somewhere>everyone
       *everyone>somewhere

What is not at all expected is that the complement can bind into the locative, but that is just what happens:

(76) [Otagai-no heya|-ni [Uli to Susi]-ga ita
       each other-gen rooms-Loc Uli and Susi-Nom be-past
       'Uli and Susi were in each other's rooms'

Yatsushiro concludes, plausibly enough, that the features of the NP complement covertly raise, and they license the anaphor from that raised position, but do not create high scope. To distinguish Japanese from English, she observes that in the relevant English constructions, the 'Extended Projection Principle', which demands that a clause have a subject, has been satisfied in overt syntax (by there). She further suggests that subject position is completely unfilled overtly in the corresponding Japanese constructions, the 'EPP feature' being weak. The covert raising in English would then be just of agreement features, while in Japanese it would be of these plus the EPP feature. Either the combination, or the EPP feature alone, would then provide the crucial ability to license the anaphor. This is a potentially promising compromise between Chomsky's account and the one advocated above. The question of whether it can handle the full range of phenomena awaits further research.