ࡱ> c p(jbjbSS 11) ]PPPP 4hZn R,1%zJ JPt    $PPjz z&Towards a Theory of Head-Binding Norvin Richards ms., MIT, February 1996 In this paper I develop a theory of binding based on relations between heads bearing the N-features of arguments. The theory consists of certain fairly straightforward assumptions about the distribution of these N-features, along with certain general syntactic constraints, such as Shortest Move and Economy of Representation. I try to show that a theory of this kind can account for the core binding facts, as well as a variety of facts from other languages which have been the topic of some debate. Phenomena treated here include the blocking effect for long-distance anaphora in Chinese and its absence for long-distance anaphora in various other languages, anti-subject-orientation of pronouns in languages like Norwegian, disjoint reference effects for anaphors in languages such as Dutch, Danish, and Japanese, and certain cases of non-complementarity between anaphors and pronouns. Traditionally, binding theory has concerned itself with relations between various kinds of NPs. In this paper I will develop a theory based on rather different assumptions. The theory will concern itself with the nature of the N-features which are taken in the Minimalist Program to enter into checking relations with nominal elements. The phenomena typically examined by binding theory will fall out as a side effect of well-formedness conditions on these checking relations. Chomsky (1995) suggests that among the things that can move are features of various kinds. In a case like 1, for example, a [+wh] feature is claimed to move into CP in order to enter into a checking relation with a feature on Co: 1. What is he cooking t? In English, however, a [+wh] feature cannot move by itself, presumably for morphological reasons. Thus, all the features associated with the wh-word what are "pied-piped" along with the [+wh] feature (just as in Italian, a wh-word embedded in a prepositional phrase cannot move by itself, and must pied-pipe the entire prepositional phrase along with it). The movement in 1, then, creates a number of related but distinct chains: a feature-chain headed by the feature [+wh], a category-chain headed by the No what, and perhaps others. The Minimalist approach to syntax centers largely on checking relations between specifiers of heads and features on the heads. The postulation of feature-chains raises interesting questions, in the Minimalist program, about what kinds of conditions exist on well-formed checking relations; what kind of relation does a feature-chain have to bear to a specifier for feature-checking to take place? In this section I will suggest a partial answer to this question, examining particularly the behavior of certain kinds of N-features on To and on q-assigners. The result will be a theory of well-formedness conditions on feature checking which yields the behavior of anaphors and pronouns essentially as a side effect. I will assume the basic phrase structure for the clause given in Chomsky 1995, as in 2: 2. CP 4 C' 4 Co TP 4 T' 4 To vP 4 DP v' ! 4 SUBJ vo VP 4 DP V' ! 4 OBJ Vo In 2, the subject and object are assigned their M2hc,rS쇔OGqdWTjЙ~Ta0.m+H j=dML.~|cq-roles by different heads; the subject is base-generated in the specifier of a "light verb" v. The subject then moves to Spec TP to check an N-feature on To; the subject is thus in the unique position of being related both to a q-assigner and to To, which will figure in the account developed here of the distinctive behavior of subjects with respect to binding theory. I will crucially assume, also, that the N-features of the arguments of the sentence appear on the heads which assign q-roles to those arguments, as well as on the head To. For my purposes it will be useful to assume that these N-features come in two types. I will be assuming f-features of the familiar kind having to do with person, number, and gender. In addition, we will need a second kind of features I will call "U-features"; these are unique to a given NP, unlike f-features. Note that U-features are not referential indices; they have no connection to the reference of the noun. All that must be true of them is that no two nouns in a sentence have the same U-features; we may take them as representing the noun's position in the Numeration, for example. Crucially, we will assume that a head may be base-generated "underspecified" for either or both of these kinds of features. Now we can return to the question with which this section began; what must be true of the N-features of a head for it to be able to check off its features with an NP in its specifier? I want to claim that a feature-chain must satisfy two requirements in order to allow feature-checking to take place: 3. a. Accuracy--the feature-chain must be associated with features that accurately reflect the features on the element involved in checking. AND b. Completeness--the feature-chain must be associated with a fully specified set of features. A feature-chain is taken to be "associated", in the relevant sense, with the features of the feature-chains on any head occupied by any member of the chain. Thus, if a H^G$e_T}7xG3J3Оk׉f-feature moves from head X to head Y, it is associated with the f-features (if any) on both heads. Furthermore, we will need to take "association" to be transitive; if a f-feature a is associated with another f-feature chain b, and b is associated with yet another f-feature chain c, then a is associated with c. There will be many cases, of course, in which Accuracy and Completeness are both satisfied by a trivial feature-chain, as in 4: 4. [XP NPf(3sg), U(3) Xf(3sg), U(3)] Here the NP in Spec XP has third person singular f-features and the U-feature 3, and the head Xo has N-features which are the same as those on the NP, satisfying Accuracy, and are not underspecified, satisfying Completeness. This trivial N-feature-chain can therefore enter into a well-formed feature-checking relation and vanish, following standard Minimalist assumptions. Accuracy and Completeness could also be satisfied in a non-trivial feature-chain, as in 5: 5.[XP NPf(3sg), U(3) Xf(3sg), U(3) [YP NPf(3sg), U(5) Yf( ),U(5)]] : 1 z------------------m Here the f-feature-chain on Yo is non-trivial, with members on both Xo and Yo. If the chain consisted purely of the features on Yo, it would satisfy neither Accuracy nor Completeness, since these features are underspecified and do not accurately reflect the features of the NP in Spec YP. The head of the chain, on the other hand, is associated with a set of third person singular f-features that are on the head Xo which the head of this f-feature chain occupies. These features are fully specified and correspond to the f-features of the NP in Spec YP. Thus, the f-feature chain on Yo obeys Accuracy and Completeness, and can enter into a well-formed spec-head relation with the NP in Spec Y and be checked off. Accuracy and Completeness could also be satisfied separately by different members of a feature-chain, as in 6: 6.[XP NPf(3sg), U(3) Xf(3sg), U(3) [YP NPf( ), U(5) Yf( ), U(5)]] : 1 z--------------m In this case, the f-feature-chain satisfies Accuracy, since its underspecified features correspond to the underspecified features on the NP in Spec YP. It also satisfies Completeness, because the head of the chain is associated with completely specified f-features on Xo. Thus, the feature-chain can enter into a well-formed checking relation and be checked off. It may be worth noting that none of the above claims have any semantic consequences to speak of. Of course, Accuracy can force two NPs to have the same f-features; compare 5 with 7: 7.[XP NPf(1sg), U(3) Xf(1sg), U(3) [YP NPf(3sg), U(5) Yf( ),U(5)]] : 1 z------------------m The f-feature chain in 7 fails to satisfy Accuracy, since no member of the chain is associated with f-features that correspond to those of the NP in Spec YP; it does satisfy Completeness, since the head of the chain is associated with the fully specified f-features on Xo, but these f-features are not the same as those on the NP in Spec YP, so Accuracy is not satisfied. Thus, there is a sense in which the two NPs in 7 must have the same f-features if a f-feature chain of the type shown is to be constructed. This is taken here to be due to well-formedness conditions on feature-checking. The theory will make two claims about anaphors. First, all anaphors are taken to have underspecified N-features, so that Accuracy will require feature-chains checked by them to be associated with underspecified N-features, as in 6. Second, the configuration in 8 is taken to be interpretable just in case the N-features a are those of an anaphor: 8. XP 5 NP X q X 1 [N-features a] The interpretive component will contain instructions assigning an interpretation to configurations like 8 involving some semantic relation between the NP in Spec XP and the anaphor that supplied the N-features; for reflexives, for example, this semantic relation will involve some type of referential dependence. Assuming, following Chomsky 1995, that feature-checking takes place only in order to get rid of uninterpretable features, features in this configuration need not be checked off, since they can be assigned an interpretation; thus, they will remain unchecked. These relatively straightforward assumptions will get us the core binding facts. The requirements of Accuracy and Completeness, along with the assumption that anaphors always have underspecified N-features, will force anaphors to be bound. Because feature-movement is subject to Shortest Move, binding will observe certain locality conditions. The Economy condition requiring features to be underspecified whenever possible will give the effect of Condition B, preferring that anaphors be used whenever they can be. The basic facts of binding theory thus fall out from well-formedness conditions on feature-checking, as desired in a Minimalist framework. In fact, a variety of new facts can now be captured, as we will see. 1. The Basic Theory Let us run through some basic examples showing how the approach outlined above gets the core binding facts. Consider 9: 9. a. Stacyi told Susanj about herselfi/j b. Stacy Tstacy vstacy toldsusan Susan aboutherself herself 9.b represents the structure of 9.a which is relevant for the binding theory. The DU+;22]%C g~B۟/bYy;m1p>TO?6/ga Lݬ rq-assigner about bears the N-features of the NP herself. Similarly, the q-assigning light verb v and To both bear the N-features of the subject NP Stacy, and the verb told bears the N-features ofSusan. Because the features on the q-assigner about are in a checking relation with an anaphor, the N-feature-chain cannot be trivial; the chain must satisfy both Accuracy and Completeness by being associated both with an underspecified set of N-features and with a fully specified set of N-features. Thus, we base-generate about with underspecified N-features, satisfying Accuracy, and then move these features to another head with fully specified features, creating a feature-chain that satisfies Completeness. For example, the derivation of a sentence like 9.b might proceed as follows (head-movement and XP-movement is not shown; categories are shown in their positions at Spell-Out): 10. a. TP 4 DP T' ! 4 Stacy ToN(Stacy) vP i v' 4 voN(Stacy) VP 1 4 told DP V' ! wu Susan VoN(Susan) PP i P' 4 PoN( ) DP 1 ! about herself b. TP 4 DP T' ! 4 Stacy ToN(Stacy) vP i v' 4 voN(Stacy) VP 1 4 told DP V' ! ep Susan VoN(Susan) PP N( ) i : P' 1 4 z---PoN( ) DP 1 ! about herself c. TP 4 DP T' ! 4 Stacy To vP i v' 4 vo VP 1 4 told DP V' ! wu Susan VoN( ) PP i P' 4 PoN( ) DP 1 ! about herself In a., all the  Q}>߇: @ 5G8&St6QiBLzit Ӆnjx[^A#PS?T: .k\JaJ`B iưR,f|ݲډ [`67Uz0YOm7OڪU{jߜ`(l\Ӊg `:[`xnŘq-assigners and To heads in the sentence are base-generated with N-features accurately reflecting the N-features of the nominals that enter into checking relations with them. In particular, the preposition about is base-generated with underspecified N-features, corresponding to the underspecified N-features of the anaphor herself. All the N-feature-chains in a., then, satisfy Accuracy. In b., the N-features on about move to the verb told in order to become associated with a fully specified set of N-features, thus satisfying Completeness. Now N-feature-checking can take place; all uninterpretable N-features vanish as a result of feature-checking. The results of this are shown in c. Only the anaphoric N-features that moved to the verb remain, since these features can receive an interpretation; the interpretive component interprets this structure as involving referential dependence between herself and Susan. Given that feature-movement is subject to Shortest Move, we might wonder why sentences like 9 are ambiguous; that is, why the anaphoric features on about are not forced to move to told, thus obligatorily creating the semantic effect of referential dependence on Susan. In what follows I will crucially be assuming that To and arch" SHAPE="poly" COORDSq-assigners are distinct types of heads in a sense which is relevant for Shortest Move, and therefore do not interfere with one another for purposes of calculating shortest moves. We will see as the paper progresses that To and q-assigners have quite different binding-theoretic properties, so this claim is not entirely ad hoc. Thus, anaphoric features will always have the option, I assume, of moving either to the closest q-assigner or to the closest To. In 9.b, this creates the ambiguity that we see; the anaphoric features can move either to toldsusan, creating the semantic effect of referential dependence between Susan and herself, or to Tstacy, creating the semantic effect of referential dependence between Stacy and herself. A sentence like 11.a, on the other hand, is only two ways ambiguous, for reasons which are now fairly clear: 11. a. Heleni hopes Stacyj told Susank about herself*i/j/k b. Helen Thelen vhelen hopesCP Stacy Tstacy Prstacy toldsusan Susan aboutherself herself In 11.b, the features on aboutherself need to move to a head with fully specified features, as before, and, as before, have the option of moving either to the closest q-assigner (toldsusan) or to the closest To (Tstacy). They are thus blocked by Shortest Move from moving either to vhelen or to Thelen; this has the semantic result that referential dependence between Helen and herself is impossible. 12.a-b show the effects of Structural Economy: 12. a. Heleni hopes Stacyj told Susank about heri/*j/*k b. Helen Thelen Prhelen hopesCP Stacy Tstacy vstacy toldsusan Susan abouther her In 12 we have chosen to assign a fully specified set of N-features to the sented in Orbis or the SML card catalog.
  • The Law Library maintains a separate online catalog calleq-assigner about. Structural Economy thus militates against interpretations of 12 in which her corefers with Stacy or Susan, since these readings could have been gotten by base-generating about with a more impoverished set of features, as we have just seen. No such restrictions exist on coreference of her with Helen, since base-generation of about with anaphoric features would not yield this reading. 1.1 Why Head-Binding? We have just seen the bare bones of a fairly simple system that seems to get the core facts regarding the distribution of anaphors and pronouns. This system operates under the assumption that the entities whose behavior is regulated by the binding theory are N-features on are available from any networked computer. Please note that mostq-assigners and on To. Before we go any further in the process of elaborating such a system, it might be worth noting some phenomena that seem to suggest that binding theory ought to be formulated along these lines, rather than in more traditional terms of relations between NPs of various kinds. 1.1.1 Indicatives and Subjunctives To begin with, reference to heads of various kinds seems to be necessary in any case when we attempt to define binding domains. 13-14 give one famous example of this. In the sentences in 13, the lower clause is in the subjunctive mood, while in the sentences in 14 it is in the indicative mood. For some reason this has an effect on the possible binding relations; long-distance binding is possible in 13 but not in 14. 13. a. (Icelandic, from Thrinsson 1991 (55)) Jni sagDi aD g hefDi svikiD sigi Johni said that I had-SUBJ betrayed self b. (Italian, adapted from Napoli 1979 (18)) La signorai dice che io giaccia presso di si the woman says that I lie-SUBJ near of self 14. a. (Icelandic, from Thrinsson 1991 (55)) *Jni veit aD ptions
    * Stack directories for Sterling Memorial Library, ܂$TL8EMAv=OɛH@sBq-assigner is being moved, while in 20.a the movement has no effect on binding theory, since no heads of the relevant kinds are moved. At LF, the structures of 20.a and 20.b are as in 21.a and 21.b (irrelevant specifiers and traces are repressed): 21. a. CP 3 Co TP ri DP T' ! ri John Tojohn vP 3 vojohn VP to Vo T(Prep)P g 3 looked DP T(Prep)' ! gi him T(Prep)ohim PP : ty 1 Pohim t 1 1 1 behind z--------------m b. CP 3 Co TP 3 DP T' 1 3 John Tojohn vP 3 vojohn VP 3 Vo T(Prep)P g rp looked DP T(Prep)' % go a picturehim of him T(Prep)opic. PP : 3 1 Popic. t 1 1 1 behind 1 z--------------m 21.a-b clearly show the relevant distinction between 20.a-b; in 21.b, the N-features of him are not c-commanded by the feature-bearing head T (Prep)o, so the fact that they are more inclined to behave like a bound pronoun (that is, to force a reading of non-coreference with the subject) is no surprise. In a theory in which binding relations are between nominals, on the other hand, I can see no way to capture the asymmetry in 20. 1.2 Conclusion This concludes the preliminary sketch of the theory. In what follows I will develop this binding theory in more depth. Section 2 will investigate the behavior of anaphors in different languages, deriving the various cross-linguistic differences from differences in the inventory of anaphors and in the availability of different types of To. In section 3 we will turn our attention to creating a theory of the distribution of pronouns based on a straightforward notion of Structural Economy; this approach will then be expanded to deal with problems involving noncomplementary distribution between pronouns and anaphors and cases of anti-subject-orientation of both pronouns and anaphors. 2. Anaphors In section 1 we saw a preliminary sketch of how this theory deals with reflexives. This sketch cannot be the whole story, of course; if it were, anaphors would exhibit the same behavior in all languages, which they clearly do not. Let us take a look at some of the cross-linguistic differences postulated in this theory. 2.1 To Consider the distinction between Japanese and English illustrated in 22-23 (23.a from Gunji 1983 (131), 23.b from Masanori Nakamura, p.c.): 22. a. *Naomii thinks that sheselfi/herselfi is beautiful b. Naomii thinks that shei is beautiful 23. a. Naomi-wa [zibun-ga utukusii] -to omotteiru Naomi TOP self NOM beautiful that think 'Naomii thinks that shei is beautiful' b. *Naomi-wa [kanozyo-ga utukusii] -to omotteiru Naomi TOP she NOM beautiful that think 'Naomii thinks that shei is beautiful' Here I will adapt a claim made by Borer 1989, Progovac 1992, 1993, and Aikawa 1993, who claim that languages may differ in whether they have "anaphoric Agr" in their agreement paradigms or not. On this account, then, the ill-formedness of NIC violations like 22.a is at least partly attributable to a simple failure of spec-head agreement; to have an anaphor in subject position, a language must have an appropriate agreement morpheme in its inventory that can agree with it, and English, unlike Japanese, lacks such a morpheme. I am following Chomsky (1995) here in assuming that agreement is actually a spell-out of N-features on heads of various kinds; in particular, agreement with the subject is taken here to reflect N-features on To. The account based on the availability of an anaphoric morpheme that enters into a spec-head relation with the subject can be transferred to this way of thinking fairly straightforwardly; some languages can have To morphemes with anaphoric N-features, while others cannot. In a language in which To cannot be anaphoric, To cannot enter into a feature-checking relation with an anaphor in its specifier, since the features of To and the anaphor cannot be identical; thus, well-formedness conditions on feature-checking produce the NIC effect. More specifically, we should say that some To morphemes but not others can be anaphoric, since the NIC effect may appear in some tenses but not others in a given language. Compare 24 with 22-23: 24. a. Naomii wants herselfi to win b. *Naomii wants heri to win English infinitival clauses act like Japanese tensed clauses with respect to the NIC; in this theory, this indicates that English infinitival T may be anaphoric. 2.2 Types of Anaphors The second major type of cross-linguistic difference which is relevant for anaphora has to do with the types of anaphors which are available in the lexicon of a given language. A cursory glance at the literature on anaphora shows that many languages have several different types of reflexives. Here we will make use of the more detailed theory of N-feature types first described in the Introduction. I will be assuming a distinction between the f-features (person, number, gender) and another class of features I call U-features, which differ from f-features in that they are unique to the particular nominal bearing them; several nominals may have the same f-features, but no two nominals have the same U-features. This distinction yields three logically possible types of anaphors: anaphors with defective f-features, those with defective U-features, and those with defective features of both kinds. I propose that all three types are attested. 25 gives some examples of the different kinds of expressions which are relevant for the binding theory (where + indicates that the element in question possesses fully specified features of that kind, while - indicates defective features): 25. f-features U-features Chinese ziji - + Chinese ta-ziji + - Japanese zibun - + Japanese kare-zisin + - Japanese zibun-zisin - - English himself + - English him + + John + + 2.3 Applying the theory Suppose we claim that To is the only possible supplier for defective !,[eH*\ȰÇ#JHŋ3jȱGF!ɓ(S\ɲ%aܰ18s܉B(J,F <32{ Ij5J,2G,9r,P2V/tZ%@(Z p q5s'˩d0N4a"^WO Ey)-RӶV"Lvm I0TbRdNf-features. Anaphors with defective f-features are therefore subject-oriented, since only To can satisfy the defective features on heads coindexed with such anaphors, and the features on To are always those of the subject. Now consider 26-28 (Chinese, from Huang and Tang 1991 (263-4)): 26. Zhangsani renwei Lisij hai-le zijii/j Zhangsan think Lisi hurt-ASP self 'Zhangsani thought that Lisij hurt himselfi/j' 27. Zhangsani renwei woj hai-le ziji*i/j Zhangsan think I hurt-ASP self 'Zhangsani thought that Ij hurt myself*i/j' 28. Zhangsani renwei Lisij hai-le ta-ziji*i/j Zhangsan think Lisi hurt-ASP him-self 'Zhangsani thought that Lisij hurt himself*i/j' Note that Chinese is like Japanese and unlike English in allowing apparent violations of the NIC (Chinese, from Huang 1982 (331)): 29. Zhangsan shuo [ziji hui lai] Zhangsan say self will come 'Zhangsani says that hei will come' By hypothesis, this is an indication that Chinese tensed To can be inherently anaphoric. Up until now, we have used anaphoric heads in all and only those cases where anaphors were related to heads; a aO0M.1@|ae.@#. R,c}0(lG, zbHA4A&L6 K`B(PH%@@ac <q-assigner was anaphoric just in case it was assigning a q-role to an anaphor, and To was anaphoric just in case the subject was an anaphor, as in 29. There is at least one clear syntactic distinction, however, between To and q-assigners. q-assigners enter into the relevant structural relation with the nominals whose features they bear at the beginning of the derivation, while Spec TP is empty at the beginning of the derivation; a nominal must move to Spec TP at some point in the derivation to check the features of To. In a language like Chinese, then, where To may have anaphoric features, there is no reason not to base-generate To with anaphoric features even in cases where the subject is not an anaphor. The only requirement imposed by the syntax should be that the N-feature-chain on To satisfy Accuracy and Completeness at the point in the derivation at which the features on To are checked. Thus, we should be able to base-generate To with anaphoric features even if the subject is not an anaphor, just in case some higher To can supply the anaphoric To with the features it lacks in time for feature-checking to take place (that is, feature-movement will have to take place before movement of the subject to Spec TP). Here I am taking advantage of a property of the system of feature-movement outlined above, namely the distinction drawn between the syntax and the semantics of binding. The syntax of feature-movement, as we saw, has no inherent semantic effects, and the N-features it manipulates have nothing essential to do with the reference of nominals. In the cases we have seen so far, feature-movement has always been associated with referential dependence, because the only features that needed to move were those contributed by anaphors. But according to the theory developed here, this is not a necessary property of feature-movement, which may have semantic consequences of quite different kinds, depending on whether the features in question are those of an anaphor. The interpretive component, in other words, will distinguish between syntactically identical representations, assigning coreferent readings just in case moved features belong to an anaphor. Now we are in a position to account for the data in 26-28. 26 may be base-generated with the structure in 30 (heads and features which do not participate in binding are omitted for clarity's sake, and the feature specification of ziji is included): 30. Zhangsan Tf(3 sg), U(zhangsan) renwei Lisi Tf( ), U(lisi) hai-le f( ), U(ziji) ziji (ziji = -f, +U) 30 shows the topmost To base-generated with the f-features and U-features of Zhangsan, while the To of the embedded clause is base-generated with blank f-features and the U-features of Lisi. The q-assigner hai-le 'hurt-ASP' picks up a blank set of f-features from the direct object ziji 'self', along with a fully specified set of U-features. In 30, the nearest f-features to hai-le f( ) are those on the topmost To; the f-features ofhai-le f( ) may therefore move there, satisfying Completeness and creating the semantic effect of referential dependence between ziji and Zhangsan. The f-features of the embedded To may then also satisfy Completeness by moving the topmost To; since Zhangsan and Lisi have the same f-features, the feature-chain thus created satisfies Accuracy as well as Completeness. By LF, all N-feature chains have entered into well-formed checking relations with nominals and been checked off, except for the f-features of ziji, which, being interpretable, can remain on the To whose specifier is Zhangsan; Zhangsan and ziji are thus interpreted as coreferent. Now consider 31, a possible structure for 27: 31. Zhangsan Tf(3 sg), U(zhangsan) renwei wo Tf( ), U(wo) hai-le f( ), U(ziji) ziji (ziji = -f, +U) 27 exhibits the well-known blocking effect of Chinese; if two potential antecedents for ziji have different f-features, only the closer binder can actually act as an antecedent. 31 shows why. The To of the embedded clause can be base-generated anaphoric for f-features, just as in 30, and the f-features of ziji can satisfy Completeness by moving to the higher To, just as it did in 30. However, the embedded To also has anaphoric f-features, and must also satisfy Completeness by moving to the higher To; this means that the embedded To will have to be associated with the 3rd person singular f-features which are on the higher To, and these features conflict with those of the embedded subject wo 'I'. wo and the embedded To therefore cannot enter into a feature-checking relation, since their features are different, and the structure will violate Accuracy and crash. Long-distance anaphora in 36 is thus correctly predicted to be impossible. Now consider 32, a structure for 28: 32. Zhangsan Tf(3 sg), U(zhangsan) renwei Lisi Tf(3 sg), U( ) hai-le f(3 sg), U( ) ta-ziji (ta-ziji = +f, -U) In 32, as in 31, long-distance binding of the reflexive is impossible. The embedded reflexive in this case is ta-ziji, an anaphor which, by hypothesis, projects blank U-features. For the anaphor to be long-distance bound, therefore, the embedded To must be base-generated anaphoric for U-features. This means that the U-features on the intermediate To will have to move to the higher To to satisfy Completeness. This leads to an ill-formed structure, since Lisi and Zhangsan have different U-features; the embedded To will thus have different features from Lisi, leading to a violation of Accuracy and a failure of feature-checking. In effect, ziji and ta-ziji exhibit the same "blocking effect"; the difference between them is that ta-ziji is sensitive to U-features rather than f-features. Since U-features are unique to the nominal bearing them, no two potential antecedents will ever have the same U-features; the blocking effect for ta-ziji is therefore more frequently triggered. We have seen how a sufficiently rich theory of features, coupled with the assumption that some languages but not others have a variety of To which is base-generated with anaphoric features, can successfully predict the behavior of reflexives in Chinese. Chinese only has two types of reflexives; a reflexive with blank y̜UsDV `',b3~s\V30Jn>H,L +^? +( Yz0PJW@%;`WІ)i( %htm` 4$Z|q+eC9#(M+f-features (ziji) and one with blank U-features (ta-ziji). We have seen that reflexives with blank f-features are subject-oriented, since only To can supply f-features, and reflexives with blank U-features are apparently local. We would expect, then, that a reflexive with underspecified features of both kinds would be both local and subject-oriented. Japanese, which has all three kinds of possible reflexives, confirms this (Japanese, adapted from Katada 1991 (289)): 33. a. Johni-ga [Billj-ga Mikek -ni zibuni/j/*k-no koto -o hanasita to] itta John NOM Bill NOM Mike DAT self GEN matter ACC told that said 'Johni said that Billj told Mikek about himselfi/j/*k' (zibun = -&`E~X0* /_w!=噣h" 0af%$f, +U) b. Johni-ga [Billj-ga Mikek -ni kare-zisin*i/j/k-no koto -o hanasita to] itta John NOM Bill NOM Mike DAT him self GEN matter ACC told that said 'Johni said that Billj told Mikek about himself*i/j/k' (kare-zisin = +f, -U) c. Johni-ga [Billj-ga Mikek-ni zibun-zisin*i/j/*k-no koto -o hanasita to] itta John NOM Bill NOM Mike DAT self self GEN matter ACC told that said 'Johni said that Billj told Mikek about himself*i/j/*k' (zibun-zisin = -f,-U) 33.a and b show that the behavior of Japanese zibun and kare-zisin is parallel to that of Chinese ziji and ta-ziji respectively. Japanese zibun-zisin behaves just as we predicted an anaphor with blank features of both kinds would behave; we can conclude that Japanese has all three of the possible kinds of anaphors predicted by this theory. There is a class of languages that appears to exhibit long-distance binding of a kind that differs from what we have seen so far (Icelandic, from Thrinsson 1991 (51, 55-6)): 34. a. Jni bad Jens, um aD raka sigi Johni asked Jens to shave selfi b. Jni sagDi [aD g hefDi svikiD sigi] Johni said that I had-SUBJ betrayed selfi c. *Jni veit [aD hefur svikiD sigi] Johni knows that you have-IND betrayed selfi Where do these reflexives fit into the typology I have sketched? They appear to be long-distance, subject-oriented reflexives, which puts them in a class with Chinese ziji; that is, they are anaphors which project null f-features. Yet they differ from ziji in at least two ways. First, they exhibit no "blocking effect": 35. a. (Icelandic, from Thrinsson 1991 (55)) Jni sagDi aD g hefDi svikiD sigi Johni said that I had-SUBJ betrayed selfi b. (Chinese, from Huang and Tang 1991 (264)) *Zhangsani renwei wo hai-le zijii Zhangsan think I hurt-ASP self Second, these reflexives must be bound by third person antecedents, whereas ziji can have an antecedent of any person or number: 36. (Icelandic, adapted from Thrinsson 1975 (578)) a. *gi hata sigi I hate self b. Jni hatar sigi John hates self 37. (Chinese) a. Woi hai-le zijii I hurt-ASP self b. Zhangsani hai-le zijii Zhangsan hurt-ASP self Suppose we try to take advantage of this constellation of facts in the following way. Recall the explanation proposed here for the ill-formedness of 35.b: ziji has blank f-features, so that in order for it to be long-distance bound the closer To must be base-generated with null f-features as well. As a result, the closer To is itself anaphoric, and must be associated with the features of the higher To in order to satisfy Accuracy; since the higher To has 3rd person singular f-features, the lower To will fail to enter into a well-formed spec-head relation with its subject wo 'I'. Note, however, that whereas ziji's antecedent can have any f-features, the antecedent of Icelandic sig must be 3rd person singular. This suggests that while ziji may have completely blank f-features, sig's f-features must be at least partially specified. Suppose this is the case; that is, sig has only partially deficient f-features, possessing 3rd person singular features but lacking some other f-features; call these missing features [X]. If this feature is something all potential antecedents have in common, we now have an explanation for the lack of blocking effects in Icelandic. 36.a-b have the structures shown in 38.a-b' (heads and features which do not participate in binding are omitted for clarity's sake): 38. a. Jni sagDi aD g hefDi svikiD sigi a'. Jn Tf(3sg, +X) sagDi aD g Tf(1sg) hefDi svikiDf(3sg) sig (sig = partial f, +U) b. *Zhangsani renwei wo hai-le zijii b'. Zhangsan Tf(3sg, +X) renwei wo Tf( ) hai-lef( ) ziji (ziji = -f, +U) In 38.b', ziji projects completely null f-features, and the nearest To must therefore be base-generated with completely null f-features as well for ziji to be long-distance bound; this leads to a violation of well-formedness conditions on feature-checking, as we have seen. In 38.a', on the other hand, sig already has some f-features, lacking only the feature [X]. The intervening To can therefore be base-generated with person and number features, as long as its [X] feature is blank. The lack of a blocking effect follows. In deriving the difference between Chinese and Icelandic from a lexical difference in the anaphors in question, we leave open the possibility that there might be a language that had both kinds of anaphors. That is, there might be a language with two long-distance subject-oriented anaphors, one of which exhibited a blocking effect and could be bound by antecedents of any person and number, while the other had no blocking effect and could only have antecedents of a certain specified person and number. In fact, Korean appears to be such a language. Korean casin seems to be an anaphor of the first kind, like Chinese ziji, while Korean caki acts like Icelandic sig. Casin, but not caki, exhibits blocking effects (Korean, adapted from Cole, Hermon, and Sung 1990 (18-19)): 39. a. *Chelswui-nun [nay-ka casini-ul sarangha-n -ta] -ko sayngkakha-n-ta Chelswu TOP I NOM self ACC love Prs Dec Comp think Prs Dec 'Chelswui thinks I like selfi' (casin =-f, +U) b. Chelswui-nun [nay-ka cakii-ul sarangha -n -ta ]-ko sayngkakha -n -ta Chelswu TOP I NOM self ACC love Prs Dec Comp think Prs Dec 'Chelswui thinks I like selfi' (caki = partial f, +U) Also, casin may have antecedents of any person, while caki may only have a 3rd person antecedent (Korean, adapted from Cole, Hermon and Sung 1990 (20)): 40. a. Nai (Nei)-nun casini-ul sarangha-n-ta I (you) TOP self ACC love Prs Dec 'Ii (youi) like selfi' (casin=-f, +U) b. *Nai (Nei)-nun [Inho-ka cakii-lul miweha-n -ta]-ko mit -nun-ta I (you) TOP Inho NOM self ACC hate Prs Dec Comp believe Prs Dec 'Ii (youi) believe Inho hates selfi' (caki = partial f,+U) The fact that Korean has one long-distance subject-oriented anaphor which acts like Icelandic sig and another which acts like Chinese ziji would seem to militate in favor of analyzing the differences in behavior between the Icelandic and the Chinese anaphors as lexical differences in the anaphors in question, rather than attempting to derive these differences from more general features of the languages involved. 2.5 Summary of section 2 Let me sum up this section by sketching again the bare bones of the binding theory whose essentials I have just outlined. NPs share N-features with - : -A - ? - & -  ' q-assigning heads and with To. Anaphors are NPs with defective features of some kind (either defective f-features, or defective U-features, or both). In order for a feature-chain to enter into a well-formed checking relation with a nominal, it must satisfy the requirements of Accuracy and Completeness; that is, the feature-chain must be associated both with a set of features that accurately reflect the features on the nominal entering into a checking relation, and with a fully specified set of features. In the case of checking relations with anaphors, these two requirements conflict, and a feature must therefore move to another head, so that the feature-chain can be associated with two different sets of features; this gives the effect of Condition A, requiring anaphors to be bound. There are also cases, as we have seen, in which To heads entering into checking relations with non-anaphors bear defective features, and these features must therefore also be "bound" in the sense developed here; that is, they must move to a head with fully specified features. Feature-movement is constrained by Shortest Move, which gives us the relevant notion of binding domain. Cross-linguistic variation in the behavior of anaphoric elements can be partly attributed to the behavior of To; some languages have morphemes in their To paradigms which are anaphoric. Furthermore, To is the only kind of head which can supply f-features; this accounts for the existence of subject-oriented reflexives. Table 41 is an expanded version of Table 25: 41. f-features U-features Chinese ziji - + Chinese ta-ziji + - Japanese zibun - + Japanese kare-zisin + - Japanese zibun-zisin - - English himself + - English him + + John + + Icelandic sig  partial + Korean caki partial + Korean casin - + Korean caki-casin partial - Korean ku-casin + - 3. Pronouns and Structural Economy We have seen that a rather large part of the phenomena which have traditionally fallen within the scope of binding theory can be made to follow from an interaction of independent linguistic principles; Shortest Move and some natural conditions on feature-checking allow us to dispense with any conditions which are solely concerned with anaphora. Given certain assumptions about linguistic structures, then, we apparently no longer need a Condition A at all. This would seem to be a welcome result. Let us see whether we can similarly dispense with Condition B by reducing it to Structural Economy. 3.1 Structural Economy On this account, then, anaphoric features must be understood as being "cheaper" than fully specified features, and therefore to be preferred whenever possible: 42. a. *Johni hurt himi b. John Tjohn vjohn hurthim him On this account, 42.a is ill-formed because in 42.b, the q-assigning head hurthim bears fully specified N-features, while base-generating hurt with underspecified N-features would yield a sentence whose meaning is identical to that of 42.b. The structure therefore violates Structural Economy, on the reasonable assumption that underspecified, anaphoric features are more economical than fully specified, non-anaphoric features. Suppose we take the following principle, then, as a preliminary attempt at dealing with the distribution of pronouns in this framework: 43. Anaphoric Structural Economy Principle (version 1 of 2): Features must be base-generated anaphoric whenever a given reading allows it. On the face of it, there are a number of cases which appear to violate this principle. One of the more obvious ones is listed below: 44. a. Johni thinks hei is a wombat b. John Tjohn vjohn thinksCP [CP he The vhe iswombat a wombat] In 44.b, The apparently violates Structural Economy, since it is in a position to be bound by Tjohn and still fails to be anaphoric. Note that the problem is only with The , not with vhe ; the nearest T to vhe is The , and the nearest L &  (%&@ ?(@1(@1%!6(@   q-assigner is thinksCP . vhe and The are both associated with a single NP-chain, and on the assumption that binding theory deals with relations between chains (cf. Chomsky 1986), rather than relations within a chain, the first of these is not a problem for Structural Economy. Neither is the second, of course, since the features are distinct. The thus appears to be the only problematic head in this sentence. We might try to get around this, then, by postulating some kind of exemption for T; Structural Economy, for some reason, would only apply to   B@  JKKJ  B JKJJ  B HJ   B H@J  B HJ  B IHJJ  B H J  q-assigners. This would be somewhat ad hoc, and moreover it appears to be an unwise move for empirical reasons: 45. a. Johni believes himselfi/*himi to be an extremely handsome marsupial b. John Tjohn vjohn believesCP him Thim vhim to bemarsupial an extremely handsome marsupial c. (Norwegian, from Hestvik 1992 (557)): Johni liker hans*i/j bil John likes his car d. John Tjohn vjohn likescar his This vhis car e. (Japanese, Masanori Nakamura, p.c.): Naomii-wa [kanozyo-ga*i/j utukusii] -to omotteiru Naomi TOP she NOM beautiful that thinks 'Naomi thinks that she is beautiful' f. Naomi Tnaomi vnaomi thinksCP she Tshe vshe is beautiful These are all cases in which we want to say that the matrix T is in a position to bind the lower T, and that Structural Economy demands that an anaphor be used if referential dependence is intended. The lower vs are all apparently safe; in each case, there is an T and a q-assigner intervening between the lower v and the heads coindexed with the higher subject. For example, in 45.b, the closest T to vhim is Thim, and the closest q-assigner is believesCP; thus, neither Tjohn nor vjohn is in a position to bind Thim. We cannot say that the lower clauses in these cases lack T entirely, since that would predict ill-formedness in the following cases: 46. a. Johni believes the teacher to be fond of himi b. John Tjohn vjohn believesCP the teacher Tteacher vteacher to be fondhim of him c. (Norwegian, from Hestvik 1992, 557): Johni likte Marits bilde av hami John liked Mary's picture of him d. John Tjohn vjohn likedpicture Mary's TMary vMary picturehim of him e. (Japanese): Naomii-wa [Taroo-ga kanozyoi-o hihansita]-to omotteiru Naomi TOP Taro NOM her ACC criticized that thinks 'Naomi thinks that Taro criticized her' f. Naomi Tnaomi vnaomi thinksCP Taro Ttaro vtaro criticizedher Here the lower T is responsible for shielding the head bearing the features of the pronoun from binding by the higher T, so we clearly cannot assume that T is absent in the sentences in 45; we must assume that it is present, and that it is subject to an Structural Economy condition which requires it to be anaphoric when the T of the matrix clause bears features that can bind it. Thus, we cannot exempt T from Structural Economy. How, then, are we to deal with sentences like 44 (repeated here as 47), which apparently violate Structural Economy? 47. a. Johni thinks hei is a wombat b. John Tjohn vjohn thinksCP [CP he The vhe iswombat a wombat] In fact, we already have a way of dealing with these sentences. Recall that the difference between English tensed To and Japanese tensed To was that Japanese has an anaphoric To element in its To paradigm, while English lacks such an element. This was supposed to account for the distinction in 48: 48. a. *Naomii thinks that sheselfi/herselfi is beautiful b. Naomi-wa [zibun-ga utukusii] -to omotteiru Naomi TOP self NOM beautiful that thinks 'Naomii thinks that shei is beautiful' 48.a is bad because it violates conditions on the well-formedness of feature-checking relations (the feature-chain associated with the lower To cannot satisfy Accuracy, since English To can never be anaphoric), while 48.b is good because Japanese has a To morpheme which can have anaphoric features. But now the cause of the well-formedness of 47 is clear. English tensed To acts as though it is exempt from Structural Economy, not because Structural Economy does not apply in general to To, but because English tensed To cannot possibly be anaphoric. Even if it is structurally in a position to be bound by another head, it cannot be base-generated with defective N-features, and thus fails to violate Structural Economy. We can thus retain our maximally general statement of Structural Economy; all heads must be anaphoric if they possibly can be. 3.2 Noncomplementarity Cases of noncomplementarity between pronouns and anaphors seem problematic for this kind of outlook. If the basic idea is that anaphors are used whenever possible, it is unclear why there should ever be a free choice between an anaphor and a pronoun as equally grammatical ways of expressing a proposition. Space constraints prevent me from considering all such cases, unfortunately. Suppose we look at one case, which arises in structures of long-distance anaphora: 49. (Icelandic, from Thrinsson 1991, 55): a. Jni sagDi aD g hefDi svikiD sigi John said that I had betrayed self b. Jni sagDi aD g hefDi svikiD hanni John said that I had betrayed him 50. (Chinese): a. Zhangsani renwei Mali ai zijii Zhangsan think Mary love self b. Zhangsani renwei Mali ai tai Zhangsan think Mary love him Such non-complementarity is typical in long-distance anaphora; see Koster and Reuland (1991) for an overview. Let us review the derivation of sentences like 50.a. The sentence is base-generated as in 51.a. Binding of the verb ai 'love' by the To coindexed with Zhangsan then takes place across the To coindexed with Mali, as shown in 51.b, and finally the intermediate To is bound by the higher To, as in 51.c. (I list only relevant X` @t @ \~߇@0qSX)f, -U) b. Vi fortalte Joni om ham selvi we told Jon about him self 75. (Danish, from Vikner 1985 (16) at Susani fortalte Annej om hende selv*i/j that Susan told Anne about her self (hende selv=+f, -U) We apparently cannot claim, then, that the ill-formedness of 69.b and 70.b, and is due to the fact that the anaphors involved cannot be bound within the clause; in fact, they cannot be bound by the clausemate subject. When dealing with a subject-oriented anaphor such as zich, of course, the claim that it cannot be bound within its clause and the claim that it cannot be bound by the clausemate subject make the same predictions. We have a choice, then, of claiming that the anti-subject-orientation of Dutch zich and of Norwegian ham selv are completely different phenomena, or of trying to unify them in a single account. Clearly the latter course is preferable, all other things being equal. 3.3.2.1 Economy extended From the standpoint of the theory developed here, the facts in 68-72 are surprising; all the sentences above ought to be well-formed. In fact, this theory is set up in such a way that any anaphor ought to be able to be bound by the clausemate subject. In other words, the theory so far leads us to believe that in some languages, sentences involving an anaphor bound by a clausemate subject ought to have several possible translations. It seems, however, that some factor is making the choice among anaphors less arbitrary than this. Let us see if we can isolate this factor. Dutch has two anaphors: zich, a partial     " # # % & & () * . / 0 2 56 7";#<#>&@'D)Ff-feature anaphor, and zichzelf, an U-feature anaphor. Only the latter can be bound by a clausemate subject (from Reinhart and Reuland 1993 (690)): 76. a. Willemi bewondert zichzelfi Willem admires himself (zichzelf=+f, -U) b. *Willemi bewondert zichi Willem admires self (zich=partial f, +U) Thus, whatever factor is forcing a choice between anaphors prefers U-feature anaphors over partial f-feature anaphors: 77. [+f, -U] >> [partial f, +U] Norwegian is a language with three anaphors: seg, a partial f-feature anaphor, ham selv, a U-feature anaphor, and seg selv, a language which is partially deficient in f-features and fully deficient in U-features. Only the last of these may be bound by a clausemate subject (Norwegian, from Hellan 1988 (104, 130)): 78. a. *Joni foraktet segi Jon despised self (seg=partial f, +U) b. *Joni respekterer ham selvi Jon respects him self (ham selv=+f, -U) c. Johni foraktet seg selvi John despised himself (seg selv=partial f, -U) Thus, Norwegian appears to offer evidence that the factor choosing between anaphors prefers [partial f, -U] anaphors over both [partial f, +U] anaphors and [+f, -U] anaphors: 79. [partial f, -U] >> [partial f, +U], [+f, -U] Collapsing 77 with 79 gets us 80: 80. [partial f, -U] >> [+f, -U] >> [partial f, +U] The scale in 80 looks eerily familiar. The "best" anaphor, the one to be preferred above all others if it is available, is precisely the most underspecified one, while the "worst" anaphor is the one which is only partially anaphoric for one type of features. So far we have only used Structural Economy to distinguish between anaphors and non-anaphors, forcing us to use anaphors when they are available. These facts suggest that Structural Economy actually forces a choice of the most anaphoric element available. Because the clausemate subject is the one binder which is available in principle as a binder to all kinds of anaphors, it is in the case of binding by the clausemate subject that the effects of Structural Economy are seen. 3.3.2.2 Variable binding There is a contrast between Japanese and Norwegian with regard to these effects which offers some support for the account developed here. 81. (Japanese, a-b. from Aikawa 1993 (41-42), c. from Aikawa (p.c.)) a. *Darekai -ga zibuni-o tunetta someone NOM self ACC pinched 'Someone pinched himself' (zibun=-!$@0#%$$!'$(%,, $$$(!''0&&@ %,, %,,@''f, +U) b.?Johni -ga zibuni -o tunetta John NOM self ACC pinched 'John pinched himself' (zibun=-f, +U) c. Johni-ga kare-zisini-o tunetta John NOM him self ACC pinched 'John pinched himself' (kare-zisin=+f, -U) 82. (Norwegian, from Hellan 1988 (104, 130)): a. *Joni foraktet segi Jon despised self (seg=partial f, +U) b. *Joni respekterer ham selvi Jon respects him self (ham selv=+f, -U) There are two main differences between Japanese and Norwegian that we will examine. On the one hand, the ill-formedness of 81.b is rather weak and apparently varies from speaker to speaker; what is generally agreed, apparently, is that sentences involving zibun locally bound by a quantifier are very bad, but binding by a non-quantifier is apparently not as bad and for some speakers is perfectly well-formed. In Norwegian, on the other hand, no such variability is reported. Also, Japanese differs from Norwegian in that its U-feature anaphor, kare-zisin, can apparently be locally bound by the subject with no ill effects, while the Norwegian U-feature anaphor, ham selv, is anti-subject-oriented. So far, we would expect both languages to behave like Norwegian; both languages have anaphors with defective %&&&  f- and U-features (zibun-zisin in Japanese, seg selv in Norwegian) which should be preferred over the less economical U-feature anaphors when both choices are available. On the story developed here, this can be related to another semantic difference between Japanese and Norwegian anaphors, namely their status with respect to variable binding. Aikawa (1995) notes that the Japanese anaphors kare-zisin and zibun-zisin yield quite different interpretations in a sentence like 83 (Aikawa 1995, 7-8): 83. a. John-dake-ga kare-zisin-o hihansita John only NOM him -self ACC criticized 'Only John criticized himself (that is, no one else criticized John)' b. John-dake-ga zibun-zisin-o hihansita John only NOM self self ACC criticized 'Only John criticized himself (that is, no one else performed self-criticism)' 83.a and 83.b have different truth values; in a situation in which John, Bill, and Mary all criticized John, for example, 83.a is false and 83.b is true (since no one but John performed self-criticism, but several people other than John performed criticism of John: namely, Bill and Mary). Furthermore, the sentences are not ambiguous; neither can have the other's meaning. Aikawa (1995), following much work on the semantics of only, interprets this difference as a difference in the properties of the reflexives kare-zisin and zibun-zisin; the former cannot be interpreted as a bound variable, but must induce a reading of coreference, while the latter must be interpreted as a bound variable, and cannot allow a coreference reading. Aikawa (1993) notes that zibun can be bound by quantificational elements, suggesting that it can have a bound-variable reading (Aikawa 1993, 45): 84. Daremo -ga [John -ga zibun-o semeta to] itta everyone NOM John NOM self ACC blamed that said 'Everyonei said that John blamed selfi' Furthermore, zibun interacts with dake 'only' to yield ambiguities that suggest that zibun can be interpreted as related to its antecedent either by coreference or by variable binding (Aikawa, p.c.): 85. John-dake-ga [Mary-ga zibun-o hihansita to] itta John only NOM Mary NOM self ACC criticized that said 'Only Johni said that Mary criticized selfi... a....no one else said that Mary criticized John' OR b....no one elsei said that Mary criticized selfi' Thus, in Japanese, zibun-zisin [-IHJJ  B HJ  B@@@HJ B@@HJ B@B@HIJ B@B@H@J BBB@Bf, -U] apparently cannot corefer with its antecedent and must be a bound variable, while kare-zisin [+f, -U] cannot be a bound variable and must corefer, and zibun [-f, +U] can either be a bound variable or corefer with its antecedent. Let us move on to consider the situation in Norwegian. In Norwegian, the equivalents of kare-zisin and zibun-zisin are not so distinct. The anaphor ham selv (+f, -U), unlike kare-zisin, may trigger either a bound-variable or a coreference reading (Asbjrn Bonvik, p.c.): 86. Vi fortalte bare Joni om ham selvi we told only Jon about him self 'We only told Jon about himself... a....we didn't tell anyone else about Jon' OR b...we didn't tell anyone else about himself' Similarly, depending on context, seg selv (partial @ H    JHIHIH    J HIJKJHH  JHH HIJH   JJHHHJ  JJHHJ  HHJ    HJ    HHIHJ f, -U) may be interpreted either as a bound variable or as coreferring with its antecedent (Arild Hestvik, p.c.): 87. Bare Jon respekterer seg selv only Jon respects self self 'Only Jon respects himself... a....no one else respects Jon' OR b....no one else respects himself' Here, again, Norwegian differs from Japanese; the closest Japanese equivalent to seg selv, zibun-zisin, can only be a bound variable, as we have seen. Finally, Hellan (1988, 1991) notes that Norwegian seg, like Japanese zibun, can give either a bound-variable or a coreference reading (Hellan 1991, 44): 88. John hadde hrt meg snakke nedsettende om seg, John had heard me talk depreciatorily about self og det hadde de som stod rundt ogs and it had those who stood around also 'Johni had heard me talk depreciatorily about selfi, and so had those who were standing around' According to Hellan, sentences like 88 can have either a strict or a sloppy reading (that is, the people who were standing around could have heard me talking either about John or about themselves). Thus, we have a robust semantic distinction between the behavior of the anaphors of the two languages, which can be put to use in our theory. Recall that the notion of structural economy developed above makes crucial reference to the meaning of the sentence; the most economical way of expressing a particular meaning must be chosen. If the relevant notion of meaning is sensitive to the contrast between the bound-variable and coreference readings of anaphors, then the anti-subject-orientation contrasts described above fall out straightforwardly. A speaker of Japanese who wishes to express a sentence involving local referential dependence on a subject (say, referential dependence of the clausemate object on the subject) has the following options, in principle, depending on whether she wishes to communicate a bound-variable or a coreference reading: 89. bound-variable coreference zibun-zisin (-f, -U) kare-zisin (+f, -U) zibun (-f, +U) zibun (-f, +U) A speaker of Norwegian in the same situation has the following options: 90. bound-variable coreference seg selv (partial f, -U) seg selv (partial f, -U) ham selv (+f, -U) ham selv (+f, -U) seg (partial f, +U) seg (partial f, +U) Let us take the Norwegian case first, as it is the simplest. Here the contrast between the bound-variable and the coreference reading is irrelevant; all Norwegian anaphors can have either reading, so the set of possibilities is the same in each case. The most underspecified Norwegian anaphor is seg selv, which is at least partially underspecified for both kinds of features. Thus, this anaphor must be used whenever possible (that is, whenever binding is by a clausemate subject), and both of Norwegian's other anaphors are thus correctly predicted to always exhibit "anti-subject-orientation", in the sense of being unable to be bound by a clausemate subject. Next let us move on to the Japanese case, which is more interesting. We have seen that Japanese differs from Norwegian in that kare-zisin, unlike its Norwegian equivalent ham selv, exhibits no anti-subject-orientation. In 90, we can see why. Ham selv is anti-subject-oriented because it must compete with the more underspecified anaphor seg selv, as we have seen. Kare-zisin, on the other hand, has semantic properties which are quite distinct from those of zibun-zisin; kare-zisin must corefer and cannot be a bound variable, while zibun-zisin must be a bound variable and cannot be linked to its antecedent by coreference. Thus, the two anaphors are never in competition, and the lack of anti-subject-orientation follows. The peculiar behavior of zibun follows in a similar way. We can see in 108 that zibun competes with zibun-zisin when it acts as a bound variable, and with kare-zisin when it corefers with its antecedent. In the former competition, it will lose; in the latter, it will tie. Thus, we expect zibun to be well-formed when it corefers with a clausemate subject, but to be ill-formed when it acts as a variable bound by a clausemate subject. As we noted before, this is the case; zibun exhibits at worst a very weak anti-subject-orientation when its antecedent is not a quantifier, but when its antecedent is a quantifier (that is, when its antecedent can only be related to it by variable binding), it becomes ill-formed (81.a-b, repeated as 91): 91. a. *Darekai -ga zibuni-o tunetta someone NOM self ACC pinched 'Someone pinched himself' (zibun=-f, +U) b. ?Johni -ga zibuni -o tunetta John NOM self ACC pinched 'John pinched himself' (zibun=-f, +U) Recall that zibun can perfectly well be bound long-distance by a quantifier (85, repeated as 92): 92. Daremo -ga [John -ga zibun-o semeta to] itta everyone NOM John NOM self ACC blamed that said 'Everyonei said that John blamed selfi' This, of course, is what we expect; in 92, zibun is not in competition with zibun-zisin, which can only be locally bound. Thus, the contrasts between Japanese and Norwegian with regard to anti-subject-orientation provide striking evidence for an account of this phenomenon in terms of competition between anaphors; anti-subject-orientation is found only when two different anaphors could potentially be used for the same job. The account of anti-subject-orientation in anaphors developed here is incomplete, of course; for example, I have no account to offer of the facts of 73.c-d. Still, the simplicity of the account sketched here, as well as its applicability to a wide variety of languages, would appear to make it rather attractive as a base for further research into these questions. 4. Conclusion If the story I am telling here is right, there is probably no need for a syntactic theory of binding. All we need are some general principles regulating feature-checking and some statements about the lexicons of various languages (in particular, statements about the feature specifications of various T morphemes and anaphors). It appears, then, that even without any principles which are specifically concerned with binding theory, we can expand the coverage of the theory to problems that were previously beyond its grasp. Let me close, then, by summarizing again the general framework I have tried to develop here. I have assumed, following much recent work, that nominals share N-features with q-assigning heads and with To. Anaphors are elements whose N-features are deficient in some way, and well-formedness conditions on feature-checking therefore require features on heads associated with anaphors to move to c-commanding heads with fully specified features by LF. Which heads are possible landing sites for feature-movement is determined by Shortest Move. In addition, there is a requirement of Structural Economy which checks at LF to make sure that features which were not supplied by anaphora could not possibly have been. This requirement is blind to S-structure relations, so various operations can "save" structures which violate Structural Economy in early stages of the derivation. This leads to cases of noncomplementarity between pronouns and anaphors. Generally, if the set of possible binders for a given head changes in the course of a derivation, we expect to find noncomplementarity. Structural economy, then, will give us the effects of Condition B, as well as deriving cases of so-called "anti-subject-oriented" anaphors. Cross-linguistic differences in the behavior of anaphora are partly or completely attributable to differences in the lexicon; more specifically, in the kinds of To morphemes and anaphors available in the lexicon. If a particular To paradigm lacks an anaphoric To morpheme, long-distance anaphora will be impossible across To of that kind, an NIC effect will obtain (because of well-formedness conditions on feature-checking), and a pronoun coindexed with that kind of To will be able to corefer with the nearest c-commanding nominal controlling agreement, in apparent violation of Structural Economy. Examples of paradigms which contain anaphoric To morphemes include English infinitival To, Japanese tensed To, Icelandic subjunctive and infinitival To, Norwegian nominal and prepositional To, and Hungarian and Vata prepositional To. Examples of paradigms lacking anaphoric To morphemes include English tensed To, nominal To, and prepositional To, Icelandic indicative To, and Hungarian and Vata nominal To. The possible kinds of anaphors are defined in terms of the familiar ЧrP<nUcz5"ZT`o٣e_R96ЍǝśuBf-features, along with U-features, a type of N-feature which is unique to the nominal bearing it. Anaphors which are deficient only for f-features are subject-oriented, since only To can supply f-features, and can be long-distance bound if the nearest To can be anaphoric. Anaphors which are deficient only for U-features are not subject-oriented and are local. Anaphors which are deficient for both kinds of features are both subject-oriented and local. Bibliography Aikawa, Takako: 1993, Reflexivity in Japanese and LF Analysis of Zibun Binding, unpublished Ph.D. dissertation, Ohio State University. Distributed by MIT Working Papers in Linguistics. Aikawa, Takako: 1995, Remarks and Replies: "Reflexivity" by Reinhart and Reuland (1993), unpublished manuscript, MIT. Aoun, Youssef, and Yen-hui Audrey Li: 1990, Minimal Disjointness, Linguistics 28, 189-203. Anderson, Stephen: 1986, The Typology of Anaphoric Dependencies: Icelandic (and Other) Reflexives, in L. Hellan and K. K. Christensen (eds.), Topics in Scandinavian Syntax, D. Reidel, Dordrecht, pp. 65-88. Avrutin, Sergey: 1993, The Structural Position of Bound Variables in Russian, unpublished manuscript, MIT. Bobaljik, Jonathan: 1993, Ergativity and Ergative Unergatives, in C. Phillips (ed.), Papers on Case and Agreement II: MIT Working Papers in Linguistics, vol. 19, MIT Working Papers in Linguistics, Cambridge, Massachusetts, pp. 45-88. Bobaljik, Jonathan: 1995, The Syntax of Verbal Inflection, unpublished Ph.D. dissertation, MIT. Distributed by MIT Working Papers in Linguistics. Borer, Hagit: 1989, Anaphoric Agr, in O. Jaeggli and K. Safir (eds.), The Null Subject Parameter, Kluwer, Dordrecht, pp. 69-109. Bowers, John: 1993, The Syntax of Predication, Linguistic Inquiry 24, 591- 656. Burzio, Luigi: 1989, On the Non-Existence of Disjoint Reference Principles, Rivista di Grammatica Generativa 14, 3-27. Burzio, Luigi: 1991, The Morphological Basis of Anaphora, Journal of Linguistics 27, 81-105. Burzio, Luigi: to appear, The Role of the Antecedent in Anaphoric Relations, to appear in R. Freidin (ed.), Current Issues in Comparative Grammar, Kluwer, Dordrecht. Chomsky, Noam: 1986, Knowledge of Language, Praeger, New York. Chomsky, Noam: 1991, Some Notes on Economy of Derivation and Representation, in R. Freidin (ed.), Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Massachusetts, pp. 417-54. Chomsky, Noam: 1993, A Minimalist Program for Linguistic Theory, in K. Hale and S. J. Keyser (eds.), The View from Building 20, MIT Press, Cambridge, Massachusetts, pp. 1-52. Chomsky, Noam: 1995, Categories and Transformations, unpublished manuscript, MIT. Cole, Peter, Gabriella Hermon, and Li-May Sung: 1990, Principles and Parameters of Long-Distance Reflexives, Linguistic Inquiry 21, 1-22. Fiengo, Robert, and Heasun Kim: 1980, Binding and Control in Korean: Structural Restrictions on Anaphora in a Non-Configurational Language, Journal of Linguistic Research 1, 59-73. Grodzinsky, Yosef, and Tanya Reinhart: 1993, The Innateness of Binding and Coreference, Linguistic Inquiry 24.69-101. Gunji, Takao: 1983, Generalized Phrase Structure Grammar and Japanese Binding of Reflexives, Linguistics and Philosophy 6, 115-156. Harbert, Wayne: 1989, Subjects of Prepositions, unpublished manuscript, Cornell University. Harbert, Wayne: 1991, Binding, SUBJECT and Accessibility, in R. Freidin (ed.), Principles and Parameters Comparative Grammar, MIT Press, Cambridge, Massachusetts, pp. 29-55. Harbert, Wayne: 1995, Binding Theory, Control, and pro, in G. Webelhuth (ed.), Government and Binding Theory and the Minimalist Program, Blackwell, Oxford, pp. 177-234. Heim, Irene: 1992, Anaphora and Semantic Interpretation: a Reinterpretation of Reinhart's Approach, SfS-Report 07-93, Seminar fr Sprachwissenschaft, Tbingen. Hellan, Lars: 1988, Anaphora in Norwegian and the Theory of Grammar, Foris, Dordrecht. Hellan, Lars: 1991, Containment and Connectedness Anaphors, in J. Koster and E. Reuland (eds.), Long-Distance Anaphora, Cambridge University Press, Cambridge, pp. 27-48. Hestvik, Arild: 1992, LF Movement of Pronouns and Anti-Subject-Orientation, Linguistic Inquiry 23, 557-594. Howard, Irwin, and Agnes Niyekawa Howard, Semantics of the Japanese Passive, in M. Shibatani (ed.), Syntax and Semantics 5: Japanese Generative Grammar, Academic Press, New York, pp. 201-237. Huang, C.-T. James: 1982, Logical Relations in Chinese and the Theory of Grammar, unpublished Ph.D. dissertation, MIT, distributed by MIT Working Papers in Linguistics. Huang, C.-T. James, and C.-C. Jane Tang: 1991, The Local Nature of the Long- Distance Reflexive in Chinese, in J. Koster and E. Reuland (eds.), Long- Distance Anaphora, Cambridge University Press, Cambridge, pp. 263-282. Katada, Fusa: 1991, The LF Representation of Anaphors, Linguistic Inquiry 22, 287-313. Kiss, Katalin: 1987, Configurationality in Hungarian, Akadmiai Kiad, Budapest. Koizumi, Masatoshi: 1993, Object Agreement Phrases and the Split VP Hypothesis, in J. Bobaljik and C. Phillips (eds.), Papers on Case and Agreement I: MIT Working Papers in Linguistics, vol. 18, MIT Working Papers in Linguistics, Cambridge, pp. 99-148. Koizumi, Masatoshi: 1995, Phrase Structure in Minimalist Syntax, unpublished Ph.D. dissertation, MIT, distributed by MIT Working Papers in Linguistics. Koopman, Hilda: 1983, The Syntax of Verbs, Foris, Dordrecht. Koster, Jan: 1985, Reflexives in Dutch, in J. Guron, H.-G. Obenauer, and J.- Y. Pollock (eds.), Grammatical Representation, Foris, Dordrecht, pp. 141- 167. Koster, Jan and Eric Reuland: 1991, Long-Distance Anaphora: an Overview, in J. Koster and E. Reuland (eds.), Long-Distance Anaphora, Cambridge University Press, Cambridge, 1-26. Li, Yafei and Chioko Takahashi: 1993, On the Logophoric and Syntactic Nature of Binding of Reflexives, unpublished manuscript, Cornell University. Maling, Joan: 1984, Non-Clause-Bounded Reflexives in Modern Icelandic, Linguistics and Philosophy 7, 211-241. Marcz, Laszlo: 1984, Postposition Stranding in Hungarian, Groninger Arbeiten zur Germanistischen Linguistik 24, 127-161. Napoli, Donna Jo: 1979, Binding of Reflexives across Clause Boundaries in Italian, Journal of Linguistics 15, 1-28. Progovac, Ljiljana: 1992, Relativized SUBJECT: Long-Distance Reflexives without Movement, Linguistic Inquiry 23, 671-680. Progovac, Ljiljana: 1993, Long-Distance Reflexives: Movement to Infl vs. Relativized SUBJECT, Linguistic Inquiry 24, 755-772. Rappaport, Gilbert: 1986, On Anaphor Binding in Russian, NLLT 4, 97-120. Reinhart, Tanya, and Eric Reuland: 1993, Reflexivity, Linguistic Inquiry 24, 657-720. Richards, Norvin: 1993, Tagalog and the Typology of Scrambling, unpublished Honors thesis, Cornell University. Richards, Norvin: 1994, Head-Binding, in E. Duncan, D. Farkas, and P. Spaelti (eds.), Proceedings of WCCFL 12, Center for the Study of Language and Information, Stanford, pp. 421-437. Richards, Norvin: 1995, Toward a Feature-Movement Theory of Long-Distance Anaphora, paper read at the OTS Optionality Workshop. Richards, Norvin: to appear, The Principle of Minimal Compliance, to appear in Proceedings of SCIL 7. Ritter, Elizabeth: 1993, Where's Gender?, Linguistic Inquiry 24, 795-803. Ritter, Elizabeth: 1995, On the Syntactic Category of Pronouns and Agreement, NLLT 13, 405-443. Rizzi, Luigi: 1990, On the Anaphor-Agreement Effect, Rivista di Linguistica 2, 27-42. Rgnvaldsson, Eirkur: 1986, Some Comments on Reflexivization in Icelandic, in L. Hellan and K. K. Christensen (eds.), Topics in Scandinavian Syntax, D. Reidel, Dordrecht, pp. 89-102. Schneider-Zioga, Patricia: 1992, Agr(eement) and Phi-Feature Binding, unpublished manuscript, University of Southern California. Sigler, Michele: 1995, Subject Positions in Armenian, paper presented at NELS 26. Thrinsson, Hskuldur: 1975, Some Arguments against the Interpretive Theory of Pronouns and Reflexives, unpublished manuscript, Harvard University. Thrinsson, Hskuldur: 1991, Long-Distance Reflexives and the Typology of NPs, in J. Koster and E. Reuland (eds.), Long-Distance Anaphora, Cambridge University Press, Cambridge, pp. 49-75. Vainikka, Anne: 1989, Deriving Syntactic Representations in Finnish, unpublished Ph.D Dissertation, University of Massachusetts at Amherst. Vikner, Sten: 1985, Parameters of Binder and of Binding Category in Danish, unpublished manuscript, University of Geneva. Watanabe, Akira. 1994. Agr-Based Case Theory and its Interaction with the A- bar System. unpublished Ph.D. dissertation, MIT. Distributed by MIT Working Papers in Linguistics. Yang, Dong-Whee: 1983, The Extended Binding Theory of Anaphors, Language Research 19, 16-192. Yoon, Jeong-Me: 1989, Long-Distance Anaphors in Korean and their Cross- Linguistic Implications, in R. Graczyk, et al (eds.), CLS 25: Papers from the 25th annual regional meeting of the Chicago Linguistic Society, Chicago Linguistic Society, Chicago, pp. 479-495. (Acknowledgements to be added.)  This tradition is not without its exceptions, of course: see Reinhart and Reuland 1993, in particular, for a different viewpoint.  That is, the facts regarding the distribution of pronouns and anaphors. I will be unable to discuss Condition C here.  Theories assuming that subjects and objects are base-generated in different maximal projections include Bowers (1993), Koizumi (1993, 1995), Bobaljik (1995), and Chomsky (1995).  These requirements are probably too simple; in fact, we will see later that Accuracy must apparently be satisfied at the point in the derivation at which a nominal enters into a spec-head relation with the feature-chain. If we were to assume a version of Watanabe's (1994) Three-Layered Case Theory suggested by Chomsky (class lectures, 1994), in which N-features must be checked twice, we could claim that these two requirements hold of the two different instances of feature-checking.  We might think of the relevant distinction as one between lexical and functional heads (on the assumption that the light verb v is lexical).  The acceptability of sentences like 23.b varies from speaker to speaker; for Katada (1991), for example, they are apparently well-formed. This is one of several problematic cases of noncomplementarity between pronouns and anaphors, which I will discuss in section 3.2. I will not discuss sentences like i., which are apparently well-formed for all speakers: i. Naomi-wa [e utukusii]-to omotteiru Naomi TOP beautiful that think The approach outlined here would lead us to conclude that the empty category in sentences like i. can be anaphoric, at least for speakers who reject 23.b. See Aoun and Li 1990, Vainikka 1989 for arguments for a similar conclusion in Chinese and Finnish, respectively.  This is based on the distinction introduced in Huang and Tang 1991 between f-features (person, number, gender) and R-features (aspects of reference other than f-features), the only difference being that the N-features assumed here are purely syntactic; they are not intended to represent the reference of the nominal.  We might wonder why anaphors which only lack f-features need to corefer with their antecedents, rather than simply agreeing with them in f-features. Recall that we are assuming that any structure in which the N-features of an anaphor move to a head whose specifier is a possible antecedent for the anaphor is given a reading involving coreference by the interpretive component. Ensuring that coreference is enforced is not the duty of the syntax. We will shortly see another case in which the syntax and semantics of binding diverge; feature-movement does not always force coreference.  This may be related to the fact that To is the one functional head that bears N-features (the others all being lexical(arguably, in the case of vo)); for arguments associating s\\7Dz\d^ cWI"Ѝev=لd/$` 8wߜa^96jSeRq!7+[vO)G6B^"f-features with functional projections of various kinds, see Ritter 1993, 1995 and references cited there.  In Icelandic, as in a number of other languages, long-distance binding is restricted to binding into clauses of certain tenses or moods: in this case, subjunctive and infinitival clauses, but not indicative clauses, allow long-distance binding. In terms of the theory developed here, Icelandic infinitival and subjunctive To, but not indicative To, has an anaphoric morpheme in the paradigm. Given that infinitival and subjunctive clauses are generally subordinate clauses, while main clauses are typically indicative, the relative frequency with which the Icelandic pattern shows up (as compared, for example, to the reverse pattern, which as far as I know has never been attested) may be explainable in terms of functional load; forms of To which are typically in subordinate clauses are more likely to be in a position to be bound, and an anaphoric morpheme is thus more useful.  For example, the feature [+N], or [+D].  The facts of Korean anaphora are apparently subject to considerable dialectical variation. Judgments compatible with the facts reported here can be found, however, in Yoon 1989, Cole, Hermon and Sung 1990, and Li and Takahashi 1993.  See Aikawa 1993 for discussion of a blocking effect in Japanese long-distance binding, which suggests that zibun, like ziji in Chinese, is a full f-feature anaphor. Aikawa argues convincingly, however, that the Japanese blocking effect is distinct from the Chinese one in interesting ways. I will have to leave this problem for future research.  According to Anderson (1986), there are at least some dialects of Icelandic in which sig is ambiguous between a partial f-feature anaphor and an R-feature anaphor. In these dialects, sig can be bound short-distance by non-subjects (Icelandic, from Anderson 1986 (66)): i. g sendi Haraldii ft sigi I sent Harold-DAT clothes for self  See Burzio (1989, 1991, to appear), in particular, for accounts pursuing this line.  I revert here to the less detailed notation for features on heads which was used at the beginning of the paper, simply for ease of processing. The notation is intended as a mnemonic, not as an exact representation of the features on the head.  The question of when two sentences have identical meanings in the relevant sense is not a simple one. See Grodzinsky and Reinhart 1993, Heim 1992 for discussion.  Here I assume that there are nominal equivalents of v and To; the exact identity of these heads is unimportant here.  As Harbert (1995) points out for Gothic and Anderson (1986) and Maling (1984) for Icelandic, infinitive clauses in those languages do not exhibit the complementarity discussed above with respect to long-distance binding; only anaphors are possible in those contexts. This might be a case in which postulating relevantly different types of To would be warranted, although justifying this move will take much more work.  Fiengo and Kim (1980) present a very interesting set of facts, which I cannot do justice to here, partly because my informants disagree with several of their crucial judgments. I will have to leave this problem to further research.  It is probably worth noting that for some speakers, at least, the phenomenon described in 57-59 can be overruled by semantic factors (Korean, Yoo-Kyung Baek, p.c.): i. Johni-un [Billj-i [cakij-ka cakii-pota te ttokttokha-ta- ko] John TOP Bill NOM self NOM self than more smart Dec Comp sayngkakha-n -ta -ko] malha-ess -ta think Prs Dec Comp say Past Dec 'Johni thinks Billj said selfj is smarter than selfi' i. can have the reading shown, where the two instances of caki fail to corefer, presumably because the reading where they corefer would be semantically anomalous (since people cannot be smarter than themselves).  See Richards 1995 for some further discussion of this problem.  I ignore here the anaphor hemzelf, the behavior of which is completely mysterious to me. See Koster 1985 for an account.  There is apparently some debate about this, and various tests for bound variable-hood give different results; for example, many Norwegian speakers can only get the sloppy reading for sentences like i. (Arild Hestvik, p.c., Hellan 1988, 1991): i. Jon respekterer seg selv, og det gjre Bjrn ogs Jon respects self self and it does Bjorn also 'Jon respects himself, and so does Bjrn' For these speakers, seg selv appears to only be able to operate as a bound variable here. On the other hand, there are speakers who can get a strict reading (in which Bjrn respects Jon) for sentences like i., and there are speakers who only get the sloppy reading in i but agree with the judgment in 87 (Arild Hestvik, p.c.). There is obviously a great deal of work to be done in sorting these matters out, but the general outline of the facts above seems clear; Japanese and Norwegian anaphors differ in crucial ways with regard to their status as bound variables which can be used to explain the difference between them with respect to anti-subject-orientation.  To the extent that coreferring zibun exhibits an anti-subject-orientation, I would have to attribute the effect to a preference for a bound-variable reading over a coreference reading whenever possible.  See Richards (to appear) for an attempt at dealing with these facts.  !#,-T U I M  "#%&,-67=>?@JKLWXlmno{|<>@B#$-5XZlnOJQJOJQJ6EH j0JUCJEH>*Y#3KL (.9AQYfd$#3KL (.9AQYfp}F&z |!!$n%%&@&+,H-\--^01Z2p22H7::::0:8:@:Z:z:{=U@i@@AQAENPPP͊Ί ~_j $6jDdfp}F&z |!!$n%%&@&+,H-\--^01Z2p22H7::d  6 8 F H p r !!!!!!!!*","""t%x%~%%%%%%%%%%%%%%%>&R&T&z&|&&&&&D'F'>)@)))))Z*\*****,,,,,,,- ----(-.-0-B-H----////111OJQJEHCJEHOJQJCJEHOJQJ6Y1111111222"2$2:2B2D2T2Z2222h3j344444466264699::0:8:B:Z:t:v:U@h@@@@@ AAA$A%A&A+A0A5AAAHABBB B`BnBBBBBBB,C6CTC^CCCCCCC EEFFFF65EHEH5EHOJQJOJQJCJEHOJQJCJEHV:::0:8:@:Z:z:{=U@i@@AQAE|FFFFFFFFFFFFFd dFFFFFFFFFFFFFFFFFFFGG GGG-G.G8G9G=G?GJGNGwGxGGGGGGGGGGGGGGGGGGGGGGGG HHHHHHHHHH H$H%H&H*H,H2H6HSHTH]HaHjHkHmHnHsHuH{H|HHHHHHHHHCJOJQJ6 5CJEHEH^FGG%G/GBGkGlGvGyGGGGGGGGGGGHHHH/H7HIHRHUH dUH]HbHpHvH~HHHHHHHHHHH!I"INTVUUUUXXXYd dHHHHHHHHHHHHHHIIJJ"J$JKKELLLLLLLNNNN4O9OTOXOOOOOPPNPPPQQQQZS\SSS)T-T2TsTxT}TTTTTTTTcUdUpUqU|U}UUUUUUUUUUUUUUUEH 56EH j0JU6OJQJCJ5EH 5CJEHEHOJQJTUUUUV$V2V6W8WNWVW`WWWWWXXXX"XeXjXoXvXXXXXXXXXYYY YYYY$Y%Y&Y+Y0Y5YAYEYZZZ ZZZZZZZz[[1\4\:\?\Z\_\\\\\^^(^*^*`M`,b-b2b3b6b7bBbCbLbMbTbUbOJQJ5>*5EHEHOJQJ 56EH65EHWYIY\\*`N`a(bVbbbbcEcHddd^eeggi@iziiijCjjd dUb[b\bbbbbJcKcXcYc\cdd2d4dDdFdeeZe\effggMiPiiiiiiijjEkGkOkRkVkWkkkkkllmmmmanbnnnnnnnZo[ooooooo~ppppppppqqqqqqSrXrrrssssss6OJQJ5EHOJQJOJQJEH[jjj'kkklElllmnYnn2oCoPoo=p[pwppqHqhqqq*ssdd st6tQtttFucuu{{-|N|{| !08LSny d dsZt[tbtctttttttttttjukutuuu}u~uuuwwww,y.y0z2z~zzzz{{9|:|L|M|Y|Z|y|z|~~ 37ABFIJQR]^_cvx͂΂т݂ނ 5EHOJQJ56EHOJQJEHEHZւ.9=HM\cw~ڃ"H[r$4d  ,;<?@KL_blmqtu|}  !)*./3>?CDNOXZ`qʄ̈́$3;AB͊Ί)*JK[\EH j0JU5EH565EHEHOJQJY;Cъ 8sЋ Bkq6\| 6ě+,A[vwd \OP'(|}DESThist 6 ʘ̘ԛ֛ $5:JTdoŜߜLN|~12>?ORŠƠ֠ڠ !-1?6 j0JUOJQJ5EHEHEH [ƜƟ$S~ܠ3fǡJpש$z>d ?@LMaešrtƥȥ 9:!"~/0@BhЯ*,^`°İ.0px4<ܲ޲D EHOJQJ5EH5EHOJQJ6OJQJEHEHXDFTVnz|~pxشڴPRbr|RTnvطܷ&08,JLNP`rtpxLNƻȻ (0 "fh26DHnp 5CJEH5EH5EHOJQJ EHOJQJEH6OJQJEHV>*x`7Q5c<bFld  &(<JRT>L)*89`e$2 bp"$<@  8:PTt~lnEHOJQJ65EH5EH5EHOJQJZ "$46Xd`b(-2<\`el%&24ABablmqrvw:;48xpr|~OJQJ j0JU6EHOJQJ\QRjkOPYZz{hp`dZ\ "FJ JNbhnrZ^BD55EH5EHOJQJOJQJ j0JUEHOJQJ6EHVG[r.~_j $6jd &(.0>@BLlt02:FHNPv~ "z{46\^hrxz j0JUEH5OJQJ6EH5EH5EHOJQJOJQJV (*DNLNXZpr~8:BJ`b(0x8:NOjl"&$(bl 6Zh-7K j0JUEH5OJQJ6EH\DPlRzHJx& d @ d  d  d PlRzHJx&D[ I   E  zJi\R 2WHH7Y8Y,^-^^^$%uH>_Cx` d&D[ I   E  zJid T@ d T@ d @ d d @ KS[~:BJdf, - 7 8 Q U V W [ b d f h m o p q s v |       0 1 3 ? @ B V W Y ] ^ `    ( , 4 6 : D F J [\^ *, 56EHOJQJ5EHEH j0JU56YTU`d*,:<DFNPX_` 56EHOJQJ j0JUEH65EHZ\R 2W,V.d T@ d T d T@ 01cghinszbghiouw~   #jk - .   !!9!:!""$$$5EHEH 6EH5EH].[/""t$$$$#%P%`%%%%&(4)6)))) T@ d T@ d T@ $$$$$$$$$$$$%%%%%%%%!%"%n%o%%%%%%%&&&& ''3'4'D'H'z'{'''((((((((())) )")()X)Z)d)))))))))* **0*2*8*>*@*F*\******6+>+r+OJQJ5EH5EHOJQJOJQJEH 6EHOJQJW)R*l****B.//0m00V1@222T3j3d78;99999 >>? T@ d T@ r+t+9,=,,, --3-4---22$2N2P2f2l2n2222233 3 3"3(3*3L3\3f37484@4B4l4m444[5\55555 66.6/6]6_66688 9 9999 9.9/999999999::.:3:<<F=G=B>C>!?"?+?,?5?6?EHOJQJOJQJ5EH5EHOJQJ6EH Y6?=?>?Q?R?e?f?????????@@*@+@=@>@W@X@@@@@@@@@AA)A*A9A:ALAMAAAAAAAAABB%B&B>B?BIBJBBBBBBBBB=D>DDDDDTGVGGGGGH~HHHHJFJPJRJZJdJfJhJ5EHEH j0JUOJQJEH EHOJQJY?h???@b@@@AXAAABYBBBH JJJJJKK~KKK T@ d T@ hJpJzJ|J~JJJJJJJJJJJJJJJJK2K4KZLZMZYZZZmZnZ~ZZZZZZZZZZZZZZZZ[[[[[[ \ \\\*\+\;\<\P\Q\b\c\q\r\\\\\\\\\\\\\\\]]]],^EH j0JUEH`Z%[o[[[B\\\]^sddd e4eSepeeeefkgkkkkkl3laln$oDomoooop'p?pXpvpd T@ d vpppp!qZqqqq-rQrurrrs*tuluuuDvxvv$wwwBxxd T@ uuuuXvZvtvvvvvvvvv w"wnw~wwwxx>x@xxxxxxx4y6yyyyyzzzzj{l{{{$|4|8|:|}}l~t~RSij[\qr68VX DLЃԃփdž$% j0JU5EHOJQJ6]x8yyz`zz{\{{{F|~/Go#RsZNd T@ NȆ *r܋x8pʐ np֓+EҘ?d T@ %'+.>FHnpʋ΋Ћbj|~PR 8:^nPRlnܐސ L\`bܒޒ"$NPƓȓ68NPxz+D%(34 $8<5EHOJQJ6`?dVPb>X'Wʥ./Z"d T@ 46vxԜڜ:<מܞrz&<Xh|~OY^iHM;<WXgl|ǬȬ56LWƮ̮ή<FLN:N*+EH6EHOJQJ`"d̬8H,Tyڲ*l QŷFod T@ +qy{"#OPPlv޽&(̾24jl>F/9[c )}%/gl(,@DEH>*EH j0JUOJQJ6\oH4ľ x̿(t^ Mxd T@ d T@ ",YZuv*+op^_;<OPyz'(=>VWyzln&'.\g%IJ^i>*EH EH5 j0JUEH6OJQJZ~uH>_d d dd T@ d T@ eFcdw-&8Zlf&z"8Eghzi|Xkl!7KN XkloNRtx[x>*6cCx`enO/{Y^d enO/{Y^)k""x`  6767ghln  l!6p  !/![!!i""(#$$V$$$Y''(n(o(p(d^)k""x`  d 7v6767UWghln      N P   Z[\XY$&flCJEHCJEH >*CJCJEH CJOJQJ6CJCJ j0JUCJEH6R66g)l l!6p d T@ d!"67x y       b!c!o!p!z!{!!!!!!!i"j"""""(#)#$$Y'Z'{'''(((n(p(CJEHCJEHCJEH 6CJ j0JUCJ:p  !/![!!i""(#$$V$$$Y''(n(o(p(dd T@ - 000hP/ =!p"p#p$p%|HH(FG(HH(d@- 000hP/ =!p"p#p$p%|HH(FG(HH(d@- 000hP/ =!p"p#p$p%|HH(FG(HH(d@`8  D@    [0@0NormalCJOJQJmH <A@<Default Paragraph Font0+@0 Endnote TextCJ6*@6Endnote ReferenceH*,`,Header  !<&`!<Footnote ReferenceCJEH2`22 Footnote TextCJ&)`A& Page Number0R0jdbtext$ LbLnormal + tab stops at 0.25$ ,T U4bmnq7!ʵn'h"=N S ""S"c Kvc0~$B. N      !"#$%&'()*+,-./0123456789:;<=>?@ABCDE %$   !"#$%&'()*+,-./0123456789:;<=>?@AB CFDEFGHIJKLM`` /VN     !"#$%&'()*+,-./0123456789:;<=>?@A B$C DE   T#*-/7`=BaGJPUYm\bi^o-rw~ØG9CK|@F ` H !&N+06e<@+GNSZ`````````9fkp3v{( \  n K  > Uh;X.\a4  !%"#$%&`'t()*+,G-.;/(01C2*3.4f56F78Y9-:;<\=>?@DABCDEE F}GHIJMKbLM 1FHUUbs \?DK$r+6?hJX,^u%+p(f:FUHYjs>&.)?KZhvpxN?"o^p p(p(Norvin Richards/Mac HD:Desktop Folder:head-binding distribution@yyH<yy rrrrrrrrrrrrr r!r"r#)*G/G0.4.5.6.8.9:;=>ABRSTUXYklmnoqruv~7777777ּֽֿDD{"{#{${%{&{'{(,-. 1 2 4 5f:f;f<s=s>sDsFsG_O_P_V_W"_"`"""" @@@@ @@@@@@4@@@ @"@$@&@(@*@,@.@0@2@4@6@8@:@x@@B@@@J@@@P@R@@@V@@@Z@@@^@@@d@@@v@x@z@@@@@@@@@@8@@@@@@@L@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @@@@$@@@@0@@(@*@X@@2@h@@F@H@J@L@N@@@R@@@t@v@x@z@|@@@@@@@@@@@@,@@@@<@@@H@@@P@@@@d@@@l@@@@@@@@@@@@@@@@ @ @@@@$@@@@4@ GTimes New Roman5Symbol3 Arial9Arboreal9IPATimesaM Times-Roman2Times New RomanWMIPAKielTimes New Roman9New York3Times"))8LwtB50gwqqMm9I4FDU"Ej ßq6 1qv+Pmw"$3盩QJ'oEXjf24Lbh*A·mvBcKTsM^jB/'25jFƮ{ڛ] Wf0NsxW J*oI>։TfhIs=zo K0:]}9JBqi 2dYHz%_2a.YzygHۅ 8qmGij41L\9~9A!i}P/x`0zmYNF5ao Y67LDkXyEolE`@0U+*Ӵ%ד{ X?gxWmުuW"ez '%MR`)O/"CX{iMxI} E0!)i @|W8@gѠKڛH vPq5FH3o]nfJ}ݜaL o_'M6p .xZiXFp~ou`ORaج>1q],Ù(Xn'^fD(+$i0(kr@U^|/s^43ny)6@Ig2YX$F eRQW0kݖIV[T@Ͱ4>0m2˧ߙ JUWNrvE,ӏl pe {p*Fg.2R9v3t3 +*Fs+$ݦ?.:~ .﫻ǔ!@st$F_-&sݭD̞( 2|l3cy.Ȧ0>$jD$:j1[3837 w4C)Ң>C}YdE q1|EOH(K"3UH>Pq\/ vak|d&xMhǷ(*cF1;lɑrfoW[*[TjH,3׮nܢjq- Q]ȸ0sC<`;uA;5C  94bo Va{}%B)tއ+us<ǯ9v *{^;.oGjEN>dGdP锍Bi?v/b`at@ʙ>zLl5  } } >/(tmދU4RyuMǗLaUkxxfco`tpy,]IrmSpA(M[5:2m,ĄÉl$7fGB,:|?=3pvudZMQT'ñkuG7>n׏rx@#F_D .jQOavJ}eZ0P|<0$|=V1;dЌjywZnb{DDH`:"Y*DӔO ؉ދ9=@ sa|%/(m*y|!].a98߅,7k6r{jďMbd!=oWxlBz'bXϡoz0U H77@Igl6Z2掤Wd5 C 5D8"av4(jŤb;U&Z4©rlI{傭^b>z2ڊa_L=Wu?}ʐKu X }iV格DUAb:CakᢦƢ >TECdvp\ҏI,~(՚[5>%A@ZIƜ V4Ѐ0s Ԇ(RL :afasj GQ@~|ud;^C@&߁o7[1>>*%0 ZKbk/,Kߔ3:?D9Χ){Sgm[|uS{Ƃ:'䥣w3gNhs;)TL1kn<4 K!$Z^JiN.sU̕,+\_`'nfBğ /UgcI ֗)| ]=LO}6&[}@aҽ&뙏mQ's 17fdV҆lh3X=$dAP\m񣱺6H= Dwg8&m+`0C` n݈Tm4}u|naN9أF% ^DI͓Ǹ>s&">>Fge9e0jR*-`SPThX}7(6gR~ H ՜.+,D՜.+,< hp|  'hr : BT shrinkwrapped Title 6> _PID_GUID'AN{10329684-4E1B-11D2-A668-0005028CDF3A}SN?H'] rq(ZZNuk++[XQL3;٤\2wڇP3*wSz&B5vMA]f( ~ML]A`d04eHUa1!O]qM K$=Eї[}1MPd>*7Op7LyT@=j͐m`] vm͢鹐[wӀEkz{ɼM':_D(JLy\818l`v4Vun !Jgdi1)u7fF&W%(4vn$DU+;22]%C g~B۟/bYy;m1p>TO?6/ga Lݬ r0 -x@w2(?ޣN%q}Mz$>HymVYJfwU ݠp@`H%a"i+_}^7Q[ 0GE ^˫?tO[6NhC3$+x1NCBnH l)sc;Y7uZd/Fm^CC9f9 EQ3@Byi4a)%,g` 9Ṋܥm["ޣńP)km{bè<jǒ^<7zF%gy?YhhhhTmǫ$u&$|Ŷ..f4},!'MllNBuÓ@|`d.ȵ(E\&|}Y@}G=C Xk'|2Z3aT7`կ#h\L/_fH3bXpE81Y{L!CRnUY9h8]1GL`tmˏpaII>ն:uי}|::jSm9')K (N*ꂲ<w31R0'_yw,mufH\I0.UaF={Յ!}*c>[NT\"W(%rjO#ݘi+X^R%7#ja}3DKo G}4 AWWQ+84JPUx$z4 ˡ#%-1#Yd|"9Aiy1 Oʺx3~{XIǫEwNٞ)tD7w TiX(ȵͳU U0ᢷGuHb@$u6#%W%ݷC= jO碯ID..QQ򒇢tܥ|oA:غIyl:_kOpzUz<4ų@)E?Nl}E:P5H.a5L޽m&fT29fzI+9m4r qs!(BtU;3Ã8)ʋ;' B,N6OD.x'p[T gyu\ܝzGN=vks*e0 WPLVEq*67]@op#A|` Q}>߇: @ 5G8&St6QiBLzit Ӆnjx[^A#PS?T: .k\JaJ`B iưR,f|ݲډ [`67Uz0YOm7OڪU{jߜ`(l\Ӊg `:[`xnŘՍ3oB._[e(L"ȟ#kv&ˍ܉%*זĚ T]7T@ʠ^*򧪫kgOm5XRi( ~)a >S}^QA{wF啑h zCxs`#nu mL;cofQ[^|5K?޶0{wb;JPIђ;+p3f)ښ# 􂝫D.Vs!}"׭-av2.b6O+kٙtr/`chѝ綞؉Ms&߸̀ʕHo#w YCvmJ%}IvN+܃pXv/d/ux_mzx,U 0J@ԄTjpRSFSJzC}p-߱6;XBf &fO0ɾ RŎJ6?1=:GF@ e_sHЄ{uszy h@c/ \#`0yk_*_.(QZj$V77E736,_ݘK_>fۣ/(6 D.*k:ԥ'?Id:1oWY,` ]uTJ8dER%`QY&ڦUٽqwYZim;$ {+{. im"DVoj-2|͔\+3,bJ^ݼ+}%~ZT<8bZC+,*Ƅhэi=MiD(WM O}=\mI;̕u5Y K*}0)UG%6Va[_ YQ9zu!2le2n.^9lWKA^Z% 2f_17 cH~JY2kl,1rJ;'`LY[KvU  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~     Root EntryBnHn / /*/$NR6. n  0L^N Fn( @Bu(+ :1Table*m"(Y Nn*@5.*B2*B3*B4*B5K%^ /m"Y Nn*APCTtcvieOpenVNQ*Hn(<.&@BI-WordDocumentL @!r@B@1v. n&@' m"0@7"p@7#|$'J& 0@'m" /Y n*N8xSummaryInformationCI--f<0yP`. (%p@5&L +l%@.@. @% 0$ /"mDocumentSummaryInformationI--8 @%m 05@p&@5+ L&@%l|"!CompObj6L^NtNxCTCPolesQ*VNH<n(@B<0fyN m0n$XObjectPool" /m"Y Nn*6m"dY Nn*XL @!Fu(u(NY n*6p@% pL @!B`pL @!B0L^NtNxCTRPleaeesQ*VNH<n&.8A-@Jf/./ HnN<:f0P L (JL$0gJR 8g FMicrosoft Word DocumentNB6WWord.Document.8jKP9Hvf1go"/Z BSfwZXBq0]8yє+d^Hj—]vAq3A~EOnxHfqynzZmEۼIvuNÀBnt7?sF_&H _L+zfΆ(xYfhk{t"Zu_!*\<''q/*}\@ [:0&/7A4ى0 ZHMd$%wDؠ hW<$֝HO$"G;*2 @3RPo8&p6|EX0_緃ꩫBczc&R/2(";gle>mF