|Chain and Binding - A Modification of Reinhart and Reuland's "Reflexivity."*1
Reinhart and Reuland have recently proposed an alternative to the standard Binding Theory (Reinhart and Reuland, 1993, henceforth R&R). The main tenet of their proposal is that in the explanation of anaphora a major role should be played by a non-structural condition defined over predicates. In this paper I argue that although a condition on predicates must exist, its role must be limited. The core anaphora data should still be explained by structural notions such as c-command and locality.
R&R claim that the distribution of anaphors and pronouns is governed by two different modules of the grammar. One module consists of a structural condition defined over a modified notion of an A-chain (R&R's chain condition). The other module consists of two non-structural conditions defined over predicates (R&R's binding conditions (A and B)). In this paper, I argue that the role of R&R's non-structural binding conditions should be limited. More specifically, I argue that R&R's condition A should be reduced to their chain condition.
The paper is organized as follows. I begin, in section 1, with a short and necessarily incomplete presentation of R&R's theory. Section 2 is the core of the paper, in which I present the reduction of condition A to the chain condition. I show that a condition on predicates (such as R&R's condition A) is not powerful enough to account for the distribution of reflexives. The only way to account for this distribution is by a structural condition on anaphoric expressions. In R&R's framework, the only way this can be done is by extending the chain condition. This extension is conceptually motivated, since it is based on a more unified notion of an A-chain as a potential argument. In addition, once such an extension is made, the distribution of reflexives does not require a condition on predicates such as condition A. The proposed reduction is thus achieved. In section 3, I demonstrate that the reduction of condition A to the chain condition is motivated not only by the distribution of reflexives but also by that of pronouns. Due to the reduction, R&R's notion of reflexive-marking plays a more limited role. This allows a simplification of this notion, which is presented in section 4. Section 5 deals with strict sloppy ambiguities in VP deletion contexts. I claim that the limited notion of reflexive-marking which was made possible due to the reduction, plays a crucial role in determining when a reflexive can be provided with a strict or a sloppy identity. This, on the one hand, lends support to R&R's claim that a condition on predicates is necessary, while, on the other hand, demonstrating that the condition must make reference to the more limited notion of reflexive-marking, and thus must be limited in scope.
1. R&R's Theory
As mentioned, R&R claim that two different modules of grammar govern the distribution of different anaphoric expressions. I discuss each in turn.
1.1. Binding: R&R's binding conditions are not structural conditions on the different types of anaphoric expressions. Rather, they are non-structural conditions defined over predicates. The conditions, stated in (1) require that a reflexive marked predicate be reflexive (condition A) and a reflexive predicate be reflexive marked (condition B).
(1) Condition A: A reflexive marked predicate must be reflexive.
Condition B: A reflexive predicate must be reflexive-marked.
A predicate can be reflexive either inherently, when a reflexive meaning is specified in the lexicon, or when two of its arguments in the syntax are co-indexed. A predicate can be reflexive-marked in two ways too: Either inherently in the lexicon or by having a reflexive like himself as its argument in the syntax.
Sentence (2a) is out by condition A. The predicate is reflexive-marked by myself but does not have two of its arguments co-indexed; hence it is not reflexive. Sentence (2b) is out by condition B. The predicate is reflexive (i.e., has two arguments co-indexed) but it is not reflexive marked.
(2) a. *John1 hates myself2.
b. *John1 hates him1.
1.2. The chain condition: The binding conditions make no reference to structural relationships among NPs. There is, for instance, nothing in the binding conditions that can block (3) or (4). In (3) The predicate is reflexive marked by himself and is reflexive, as required by condition A. The sentences in (4) both obey condition B. In (4a), the predicate is reflexive and is appropriately reflexive marked (intrinsically - in the lexicon). In (4b), no (semantic) predicate is reflexive, so condition B is vacuously satisfied2.
(3) *Himself1 likes Bill1.
(4) a. * Mary1 behaved her1.
b. * Mary1 heard [her1 criticize John].
R&R argue that such cases are ruled out independently by a condition on A-chains. In (5a) we have a definition of an A-chain and in (5b) a condition on its well-formedness. So an A-chain is any sequence of co-indexation headed by an A-position, in which each link satisfies antecedent government. For such a chain to be well-formed its head must be the unique element bearing the feature [+R], where [+R] is a feature shared by BT's r-expressions and by pronouns, intuitively, by all elements that can pick their reference independently in discourse3..
(5) a. Definition: A maximal A-chain is any sequence of (two or more) coindexed elements which is headed by an A-position and satisfies antecedent government.
b. General condition on A-chains: A maximal A-chain must contain exactly one link a1 which is both [+R(eferentialy independent)] and case-marked.
The ill-formedness of the sentences in (3) and (4) follows immediately from the chain condition. (3) is out because it contains a chain headed by a [-R] element. The sentences in (4) are out because they contain the chain that has two [+R] elements.
1.3. The distribution of Logophors: A result of R&R's theory is that it offers a definition of the environment where the so-called logophors can appear. Logophoric reflexives -- reflexives that appear to violate the standard condition A -- can appear only in positions that are not argument positions of a syntactic predicate, a notion which is given a definition by R&R. Because condition A of R&R's Binding Theory is defined over syntactic predicates and because only reflexives that are arguments of a syntactic predicate can reflexive mark the predicate, reflexives that are not arguments of a syntactic predicate are exempt from R&R's condition A. These reflexives can appear freely and are only restricted by discourse conditions of accessibility. The reflexives in the (a) sentences of (6-7) are all not arguments of a syntactic predicate. They can, thus, be discourse bound and are not in complementary distribution with pronouns, as demonstrated, respectively, by the (b) and (c) sentences4.
(6) a. Lucie1 counted 5 tourists in the room apart from herself1.
b. Max1 said that Lucie counted 5 tourists in the room apart from
c. Lucie1 counted 5 tourists in the room apart from her1.
(7) a. Lucie1 saw a picture of herself15.
b. Max1 said that Lucie saw a picture of himself1/ myself.
c. Lucie1 saw a picture of her1.
2. The reduction
When going into the finer details of the distribution of pronouns and reflexives, it turns out that R&R's conditions A and B operate at different levels of the grammar. Condition A, which applies to syntactic predicates, must apply at LF, whereas condition B, which applies to semantic predicates, must apply at the interface with some abstract level of semantic representation. Consider the sentences in (8).
(8) a.*The queen1 invited both Max and her1 for dinner.
b. The queen invited both Max and myself for dinner.
Because in (8a) there is a semantically reflexive predicate which is not reflexive marked, the sentence is out by condition B. Condition A, which applies to syntactic predicates, does not rule out (8b). Myself can be used logophorically because it is not an argument of a syntactic predicate. It doesn't reflexive mark anything and condition A is vacuously satisfied.
So, condition A applies at LF and condition B at the translation to semantic interpretation. Yet they are both assumed to be part of the same module of grammar, namely Binding. On the other hand, the chain condition and condition A, which apply at the same syntactic level, are assumed to be totally distinct.
I will try to show that condition A can be reduced to the chain condition. This reduction will result in a more natural division of labor between linguistic modules. On the one hand, there will be one syntactic condition (the chain condition) which operates at LF and makes reference only to well established structural relations; on the other hand, there will be one semantic condition (condition B), narrow in scope, which makes reference to semantic notions (i.e., semantic predicates) and, perhaps, operates at an abstract level of semantic representation. I will show that this proposed reduction is necessary on independent empirical grounds and motivated by further conceptual considerations.
2.1. Problems in R&R's condition A: R&R offer a striking piece of evidence for the claim that condition B is a non-hierarchical condition defined over predicates. The contrast in (9) shows that condition B can be satisfied even when a [-Reflexive] expression is locally bound, as long as there is some reflexive-marking on the predicate.
(9) a. *Jij hoorde [Jan1 zich1 critiseerde]
You heard Jan1 criticize SE1.
b. Jan1 hoorde [zichzelf1 zich1 critiseren]
It is interesting that there is no similar evidence with respect to condition A. In fact, there seems to be counter-evidence, namely, evidence that condition A is not a condition on predicates but rather a condition on arguments. Consider the ill-formed sentences in (10). There is nothing in R&R's system that can block them. In these sentences two arguments are co-indexed. Hence the predicate is reflexive. The predicate is also reflexive marked by a third argument, namely myself. It appears that both condition A and B are met. In R&R's framework it doesn't matter at all that the reflexive marker does not partake in the reflexive relation, because the condition is defined over the predicate with no reference to its arguments.
(10) a. *You1 showed myself to yourself1.
b. *Jan1 showed myself to SE1. (Dutch)
c. *You1 heard myself criticizing you1.
Tanya Reinhart has pointed out (personal communication) that the sentences in (10) could be handled within the framework of reflexivity with a slight modification of the binding conditions. Under such a modification, conditions A and B would be relativised to an index. Condition A would require an i-reflexive marked syntactic predicate to be i-reflexive and condition B would require an i-reflexive semantic predicate to be i-reflexive marked (where a predicate is i-reflexive if two of its arguments bear the index i and is i-reflexive marked by a self-anaphor if that anaphor is an argument of the predicate and bears the index i.). The sentences in (10) are out by condition A. They all contain an i-reflexive marked predicate that is j- but not i-reflexive6.
This is a possible move, but I think that there is something about it that runs against the spirit of R&R's binding conditions. The intuition behind these conditions is that they should only make reference to predicates, not to their arguments. Direct restrictions on arguments should follow solely from the chain condition.
In addition, it appears that such a move can not really solve the problem that is exemplified in the sentences in (10). In fact, (10) is just an instance of a more severe problem. To see this consider (11). In (11a) we have a reflexively marked predicate that is not reflexive, thus the sentence is appropriately ruled out by condition A. In (11b), by contrast, the predicate wash is intrinsically reflexive. Hence condition A is met. The sentence is totally out and nothing in R&R's system seems to be able to explain this. This time we can't explain the ill-formedness by relativizing condition A to an index, since the reflexive marker does partake in the reflexive relation.
(11) a. *Himself sneezed.
b. *Himself washed.
c. *Himself1 washed himself1.
2.2. A Modification of the chain condition: To see what really goes on in (11b), consider R&R's explanation for the ill-formedness of (11c). In this sentence, there is one chain, , which is headed by a [-R] element and is, thus, ill-formed. It looks like we would want the same explanation for the ill-formedness of (11b). Namely, we would want to say that in (11b) there is a chain, , that is headed by a [-R] element and is, thus, ill-formed.
The problem is that R&R depart from standard assumptions about chains and disallow a chain to consist of one element (for reasons I return to below). Therefore, could not be a chain in (11b) and there is no way to rule out the sentence. But, if, as generally accepted, chains are supposed to be something like the arguments at LF, it is not at all clear why we shouldn't have singleton chains.
So, let us assume that there are singleton chains. This will allow us to rule out (11b) in the same way that we presently rule out (11c). In addition, it will rule out the sentences in (10), without any need to complicate the binding conditions. In these sentences there is a [-R] expression that has to form a singleton chain and again this chain is ill-formed.
Consider now the full paradigm in (12). As mentioned, the ill-formedness of (12a) and (12b) can't be explained by R&R. (12c) and (12d) can be explained, but by two different mechanisms. (12c) violates the chain condition, whereas (12d) violates condition A. (It is reflexive marked but not reflexive).
(12) a. *Himself behaved.
b. *John1 explained myself to himself1.
c. *Himself1 behaved himself1.
d. *Himself ran
If we choose to acknowledge the existence of singleton chains, the four sentences in (12) fall under the same generalization. They all contain an ill-formed chain headed by a [-R] element.
2.3. The reduction: (12d) still stands out as a violation of both condition A and the chain condition. However, this redundancy is readily overcome. It turns out that once we assume singleton chains, we no longer need condition A. Whenever there is a predicate, one of whose arguments is a self-anaphor, this argument will have to have a local antecedent, by the chain condition, and the predicate will necessarily be reflexive7.
2.4. The distribution of Logophors: Now, let us look at the reason R&R have for disallowing singleton chains. Consider the logophoric use of reflexives. If singleton chains were allowed, then himself in (13) would form an ill-formed singleton chain and the sentence would be ruled out.
(13) Bill1 said that I will introduce Mary and himself1 to many people.
To solve this problem, we can make use of R&R's notion of a syntactic predicate and say that A-chains are restricted to argument positions of syntactic predicates. Because himself in (13) is not an argument of a syntactic predicate, it does not form an A-chain, and there is no violation of the chain condition.
How will this be captured formally? Consider R&R's condition on A-chains in (5), modified in (14) so as to allow singleton chains.
(14) Definition: A maximal A-chain is any sequence of coindexation which is headed by an A-position and satisfies antecedent government.
General condition on A-chains: A maximal A-chain must contain exactly one link a1 which is both +R and case marked.
In order to license logophors in the appropriate positions, all we need is to assume that A-positions are argument positions of a syntactic predicate as it is defined in R&R. More specifically, an A-position is either a syntactic argument or the external argument of a syntactic predicate, as these notions are defined in (15) (R&R's 40). A reflexive not in an A-position will never form a singleton chain and thus will never violate the chain condition.
(15) The syntactic predicate of (a head) P is P, all its syntactic arguments and an external-argument of P (subject).
The syntactic arguments of P are the projections assigned theta or case by P8.
2.5. R&R's argument for the two different modules: R&R have offered two strong arguments for the claim that their chain and binding conditions must be distinct. Such arguments appear to go against a reduction of condition A to the chain condition. Here we will see that this is not the case.
The first argument that R&R present is based on the distribution of the so-called long distance anaphors such as the Dutch zich or the Norwegian seg (henceforth, SE-anaphors). These anaphors pattern with pronouns with respect to the binding conditions and with reflexives with respect to the chain condition, and thus demonstrate that the conditions must be distinct.
SE-anaphors are assumed to have the two features [-Reflexive] and [-R]. Because Se-anaphors are [-Reflexive], they cannot reflexive mark a predicate. Like pronouns and unlike reflexives, condition B does not allow them to be arguments of a reflexive predicate, which is not independently reflexive marked. This is demonstrated in (16).
(16) a. John1 hates himself1.
b. *John1 hates him1.
c. ??Jon1 haat zich1. (Dutch)
Because SE-anaphors are [-R], they can tail an A-chain (as long as condition B is met) and thus sometimes distribute with reflexives. In all the sentences in (17) condition B is met. In (17I), the predicate is inherently reflexive and thus (independently of its arguments) is reflexive marked. In (17II), the predicate is not semantically reflexive and therefore does not need to be reflexive marked. The (a) sentences in (17) are out because they contain an A-chain tailed by a [+R] element, namely a pronoun. The rest of the sentences in (17) are in because the A-chains they contain are tailed by [-R] elements, namely reflexives (in the b sentences) or SE-anaphors (in the c sentences).
(17) I.a1. *Mary1 behaved her1.
b1. Mary1 behaved herself1.
c1. Mary1 gedroeg zich1.
II. a2. *Mary1 heard [her1 sing].
b2. Mary1 heard [herself1 sing].
c2. Mary1 hoorde [zich1 zingen].
The second argument is based on the observation that violations of condition B are uniformly weaker than violations of the chain condition. This observation, which is exemplified in (18) and (19) could be explained if we assume the existence of two conditions with different degrees of immunity to violations (Similar to ECP and subjacecny).
(18) a. ?John and Mary1 like her1.
b. *Mary1 likes her1.
(19) a. ??Jon1 haat zich1. (Dutch)
Jon hates SE.
b. *Jon1 haat hem1.
Jon hates him.
It is easy to see that these arguments only show condition B to be distinct from the chain condition, nothing's been said about condition A. We've seen that SE-anaphors pattern differently with respect to the chain condition and condition B, not condition A. We've seen that violations of chain condition are stronger than violations of condition B, not condition A. The arguments are thus fully consistent with our proposed reduction.
2.6. ECM: Consider the sentences in (20). R&R have to struggle in their explanation of the well-formedness of (20a). In this sentence we should expect a violation of condition A. Himself reflexive marks the predicate criticize and the predicate is not reflexive.
(20) a. John1 heard [himself1 criticize Mary].
b. *John1 heard [Mary criticize himself1].
In order to deal with (20a), R&R suggest that criticize can incorporate at LF to the matrix verb. This incorporation, which is overt in Dutch, forms a new complex predicate (heard-criticize) which is reflexive marked and reflexive and, thus, saves the sentence from a violation of condition A.
The problem is that once such an incorporation is allowed, it seems that the embedded object should count as an argument of the complex predicate, and it is not clear how (20b) is to be ruled out9. Under the proposed reduction of condition A to the chain condition, the problem is overcome. (20a) is good because it contains a well formed chain . (20b) is out, because Mary by relativised minimality (Rizzi 1991) does not allow antecedent government to hold between John and himself, thus creating the ill-formed chain . (20b) is a case where co-arguments of the same predicate are too far apart and cannot form an A-chain (see note 6). This demonstrates that the notion of locality which is relevant for the distribution of reflexives is not the domain of a predicate but the domain of antecedent government. The reduction of condition A to the chain condition is, thus, not only important for the sake of simplicity but also necessary for empirical reasons.
The obvious question is what distinguishes (20b) from (21). That is to say, why does Mary count for relativized minimality in (20b) and not in (21).
(21) John1 explained Mary to himself1.
That there should be a distinction between the two cases with respect to antecedent government seems to me to be evident even independently of anaphora. To see this, note that there are languages such as Albanian that allow passivization over an NP in double-object constructions, but, as far as I know, passivization never crosses the subject of a clause10.
I would like to propose that the distinction between (20) and (21) is parallel to a distinction discussed by Chomsky with respect to movement (Chomsky 1992). Assuming the VP internal hypothesis and the checking theory of case, we need an explanation for the contrast between (22a) and (22b). (22a) is a paradigmatic case of a relativised minimality violation. In Rizzi's terms, the expletive it is a potential antecedent governor of the NP trace and therefore blocks antecedent government by the more distant NP John. In order for this explanation to go through we must distinguish (22a) from (22b). That is, we must explain the fact that in (22b) the subject trace does not block the antecedent government of the object trace.
(22) a. *John1 seems that it is likely [t1 to leave].
b. John1 Mary2 [ t1 likes t2].
Chomsky introduces a notion of equi-distance in order to explain this contrast11. I will not go over the precise definitions, but the idea is that the Verb, when adjoining to AGR-O in (22b), makes Mary and t1 equi-distant from t2. It, thus, extends the domain of possible antecedents for t2 to include Mary. In (22a), by contrast, no such extension is possible and we get a relativised minimality effect.
The notion of equi-distance, as it turns, helps us to distinguish (20b) from (21). In (21) Mary and John are equi-distant from himself, whereas in (20b) they are not. Assuming a Larsonian analysis for (21) (Larson 1988), the verb movement extends the domain of possible antecedents for himself so as to include John. In (20b) even if there is LF adjunction of criticize to wash it is a double adjunction through some functional category. Under Chomsky's proposal the equi-distance domain of an NP could be extended only once by adjunction. Therefore Mary and John are not equi-distant from himself in (20b) and we get a relativized-minimality effect.
3. Arguments for the Reduction of Condition A to the Chain Condition (the Distribution of Pronouns)
We've assumed that A-chains are restricted to argument positions of a syntactic predicate. This assumption was necessary for the reduction of condition A to the chain condition. We will now show that it is independently motivated by the distribution of pronouns. It is thus not just a mechanism that enables us to amend the ruins of eliminating condition A.
Consider the distribution of possessive pronouns. This has been a problem for standard BT, and is not resolved in R&R. In (23) we see that possessive pronouns can be locally bound, and are thus not in complementary distribution with anaphors.
(23) a. [John and Mary]1 like their1 pictures.
b. [John and Mary]1 like each other's1 pictures.
R&R's theory has a way of dealing with cases of non-complementarity between pronouns and anaphors. What the theory does in such cases is show us that the pronouns and anaphors are not in an argument position of a syntactic predicate. Then, based on the fact that R&R's condition A is relevant only for the licensing of anaphors as arguments of a syntactic predicate, the anaphor is exempt from any condition and is licensed as a logophor.
Such an approach appears to be relevant for (23), but it doesn't quite work. The approach appears to be relevant because the non-complimentary in (23) seems related to the fact that the pronoun and the anaphor are not arguments of a syntactic predicate12. However, it cannot be dealt with by exempting the reflexive from some locality condition, because this non-complementarity is due to an offending pronoun. It is the pronoun which seems to be violating a non-locality condition, thus bringing about non-complementarity.
Under R&R's theory the pronoun in (23a) is violating the chain condition. (23a) contains an A-chain which has two [+R] elements. Such an A-chain is ill-formed and the sentence is incorrectly ruled out by the theory, irrespective of the fact that the pronoun is not an argument of a syntactic predicate. However, if A-chains can be formed only by arguments of a syntactic predicate we have no offending A-chain in (23a). Pronouns are exempt from the chain condition when they are not arguments of a syntactic predicate, just like reflexives are. The non-complementarity in (23) is thus explained like all other cases of non-comlemantarity.
Support for this explanation can be found in Mohawk. In Mohawk, just like in English, a possesive inside an object NP can be bound by the subject. This is demonstrated in (24a). The surprising fact about Mohawk is that once the noun heading the object incorporates to the verb, the binding is no longer possible (24b). This fact is easily explained under the assumption that pronouns which are not arguments of a syntactic predicate are exempt from the chain condition. In (24a) there is no violation of the chain condition because the pronoun is not an argument of a syntactic predicate. In (24b), by contrast, the noun incorporates to the verb forming a predicate of which the possessive is an argument13.
(24) a. I?i1 k-ohres ne i?i1 wak-nuhs-a.
I 1sS/3NO-wash Det I 1s-house-SUF
'I washed my house.'
b. * I?i1 k-nuhs2-ohres ne [i?i1 t2]
I 1sS/3NO-house-wash Det I 1s-house-SUF
'I washed my house.'
Aikawa (1993) has pointed out that Japanese has a reflexive morpheme (ziko), which detransitivizes a predicate and is relevant for the licensing of anaphoric relationships. Although the reflexive morpheme detransitivizes the predicate, the suppressed argument of the predicate can still be realized by something like clitic-doubling. This is demonstrated in (25).
(25) a. John-ga ziko-hihansita.
'John defended himself.'
b. John1-ga zibun1-o ziko-hihansita.
John-NOM zibun-acc self-defended.
'John defended himself.'
Now, the Japanese pronoun kare which normally cannot take a local antecedent, can do so when ziko is present. This is demonstrated in (26).
(26) a. *John1-ga Kare1-o nagutta
John-NOM he-ACC hit.
'John hit him'
b. John1-ga kare1-o ziko-hihansita.
John-NOM he-ACC self-defended.
John self-defended him.
'John defended himself.'
This surprising fact cannot be explained by R&R. Although in (26b) there is no violation of condition B (the reflexive predicate being reflexive-marked by ziko), R&R predict a violation of the chain condition. The chain includes two [+R] elements and is thus ill-formed. However, once A-chains are restricted to arguments of a syntactic predicate, the contrast in (26) can be explained. As mentioned, in (26b) ziko detransitivizes the predicate. kare is thus not an argument of the predicate but rather a clitic-doubling of ziko generated in some kind of adjunct position. Therefore, kare in (26b), contrary to (26a), is not an argument of a syntactic predicate and cannot be part of an A-chain. In (26b), contrary to (26a), there is no violation of the chain condition.
Now consider the contrast in Hebrew demonstrated in (27).
(27) a. *Dani1 natan lo1 et hakadur.
Dani gave to him the ball.
b. Dani1 axal lo1 et haoxel.
Dani ate to him the food.
'Dani ate the food leisurely.'
c. Dani1 tiyel lo1 bagina.
Dani walked to him in the garden.
'Dani walked in the garden leisurely.'
(27a) is ruled out by the chain condition since it contains the ill-formed chain . The obvious question is why (27b, and c) are well-formed. We have an immediate answer, assuming A-chains to be limited to arguments of a syntactic predicate. In (27b and c) contrary to (27a) lo is a dative clitic, which is not selected by the predicate. It is thus not an argument of a syntactic predicate and is exempt from the chain condition.
4. Reflexive-marking as Self-movement
Now that we've reduced condition A to the chain condition, the role of reflexive-marking is pretty limited. Reflexive-marking is no longer in charge of the licensing of reflexives; it plays a role only with respect to condition B. This allows a simplification of the definition of reflexive-marking.
We can now avoid the disjunctive definition offered in R&R. We can now say that the only way a predicate is reflexive marked is if its head bears the feature [+reflexive]. The head of an inherently reflexive predicate bears the feature inherently. When the predicate is not inherently reflexive we can say that the [+reflexive] argument (or part of it, say self) has to undergo head movement and adjoin to the head of the predicate, forming a new head which bears the feature [+reflexive].
This unified view of reflexive-marking cannot work for R&R. Self-movement was actually suggested in R&R's "Anaphoric Territories" but abandoned in their "Reflexivity". The problem was that nothing could force self to move. If this movement is only optional, then, in case self stays in its SS position, the predicate would not be reflexive marked, condition A could not do its work, and nothing would force binding of reflexives. In (28a), for instance, condition A would be vacuously satisfied with self staying in its SS position. However, under the suggested modification of R&R's theory, condition A would no longer be responsible for the ill-formedness of (28a). The chain condition would rule out (28a) and condition B would force movement of self in (28b).
(28) (a) * John likes myself.
(b) John1 likes himself1.
5. Strict/Sloppy ambiguities with Reflexives
Let us now move to a discussion that will at first seem remote, but which I will argue is directly linked to the notion of reflexive-marking. It is usually assumed that reflexives in VP deletion contexts (VPDC) can receive only a sloppy identity interpretation. This assumption is based on contrasts, like in (29).
(29) a. John likes his dog and Bill does too.
b. John likes himself and Bill does too.
While the pronoun in (29a) can be interpreted both by strict and sloppy identity, the reflexive has only the second option14.
There are two competing explanations for this fact. According to one explanation (Williams 1977, Sag 1976, Reinhart 1983, and Hestvik 1991) reflexives, contrary to pronouns, must be interpreted as bound variables, and bound variables must receive a sloppy identity interpretation in VPDC15. According to the second explanation (Kitagawa 1991), the contrast in (29b) follows from the fact that the anaphoric elements in the elided VP are identical to those in the antecedent and must obay the binding conditions.
However it seems that, contrary to both views, reflexives do not always need to receive a sloppy identity interpretation in VPDC. Rather, it seems that the interpretation of reflexives in VPDC is affected by the verb which heads the VP. Consider, thus, the contrast in (30). In (30a) we can get both a strict and a sloppy interpretation. That is to say we can mean with (30a) either that Bill expects himself to win the race or that Bill expects Jessie Owens to win the race. By contrast, (30b) has only a sloppy interpretation. It can mean only that Bill promised himself to win the race. If we assume that the explanation for the contrast in (29) is based on some condition on reflexives per se, we have no way to account for (30a) and obviously not for the minimal difference between it and (30b).
(30) a. Jessie Owens1 expected [himself1 to win the race] and Bill did too.
b. Jessie Owens1 promised himself1 [PRO1 to win the race] and Bill did too.
So, what can be the explanation for this contrast between a control and an ECM structure. R&R have also noted an anaphora contrast between Control and ECM. In (31) we see that a SE anaphor can be licensed as the subject of ECM but not as the object of control.
(31) a. Henk1 hoorde [zich1 zingen].
Henk heard [SE sing].
b.Henk1 overreedde zich1 [PRO1 het land te verlaten]
Henk persuaded SE [PRO to leave the country].
R&R's explanation for this contrast has to do with condition B. In (31b) zich and Henk are co-arguments of the same semantic predicate. The predicate is thus reflexive. However, it is not reflexive-marked. (31b), thus, violates R&R's condition B. By contrast, in (31a) zich and Henk are not co-arguments of the same semantic predicate. In (31a) there is thus no semantic reflexive predicate and condition B is vacuously satisfied. 16
I would like to suggest that condition B and the notion of reflexive-marking should also be relevant for the explanation of the contrast in (30). That is to say, I would like to suggest that the correct generalization is that a reflexive must be assigned a sloppy interpretation only when a predicate is reflexive marked.
In (29a) and (30b) we have two co-arguments of the same predicate co-indexed. Therefore condition B forces self to adjoin to the head of the predicate in order to supply it with the necessary [+reflexive] feature. In (30a), by contrast, we do not have two co-arguments of the same predicate co-indexed. Therefore, self doesn't have to move, the predicate doesn't have to be reflexive marked, and a sloppy reading isn't forced.
Now, why should reflexive-marking be relevant for the forcing of a sloppy interpretation? I would like to suggest that the answer would follow from the explanation we decide to give to a well known phenomenon regarding structural ambiguities in VPDC. It is well known that structural ambiguities do not multiply in VP deletion contexts. Take for example scope ambiguities. A sentence like (32) is only two ways ambiguous. The relative scope of the two quantifiers in the antecedent VP must be the same as that in the deleted one. Thus, the two interpretations in (33a) are possible but those in (33b) are not.
(32) John introduced someone to everyone and Bill did too.
(33a) 1. Someone X, Everyone Y (John introduced X to Y) and
Someone X, Everyone Y (Bill introduced X to Y).
2. Everyone Y, Someone X (John introduced X to Y) and
Everyone Y, Someone X (Bill introduced X to Y)
(33b) 1. Someone X, Everyone Y (John introduced X to Y) and
Everyone Y, Someone X (Bill introduced X to Y).
2. Everyone X, Someone Y (John introduced X to Y) and
Someone X, Everyone Y (Bill introduced X to Y)
There are many possible explanations for this phenomenon, which was already noted in Sag (1976) and Williams (1977). What is clear, however, is that any account of VP deletion must require that the LF structure of the antecedent VP will be the same as that of the deleted one. I would like to suggest that the explanation for the distribution of strict/sloppy ambiguities with reflexives should follow, exactly, from this requirement.
At this point we can assume that condition B requires not only that a reflexive predicate be reflexive marked, but also that a reflexive marked predicate be reflexive, as stated in (34).
(34) Condition B (revised): A predicate is reflexive iff it is reflexive- marked17.
The LF operation of self movement, just like QR, affects structure. Therefore, self movement can take place in the antecedent VP, only if the resulting head chain also exists in the deleted VP. Whenever the antecedent VP is reflexive marked, the elided VP must also be reflexive marked. This explains our generalization. Whenever the antecedent VP is headed by a reflexive predicate it must be reflexive-marked by (34). Structure identity requires the elided VP to be reflexive-marked, and (34) forces a reflexive interpretation - thus sloppy identity.
Whenever the antecedent VP is headed by a non reflexive predicate (e.g. 30a), no reflexive marking is required. Therefore no reflexive interpretation is forced and a strict identity is possible. An obvious question is why does the reflexive itself not exist in the elided VP and why does the chain condition not force local binding resulting in a sloppy identity a la Kitagawa. What I'd like to suggest is that a notion similar to vehicle change, suggested for independent reasons by Fiengo and May (1991), allows the reflexive to be replaced by an NP that does not have to be locally bound 18.
Kitagawa has however argued for his explanation of the interpretation of reflexives in VPDC, not just on the basis of the contrast in (29), but also on the basis of the distribution of pronouns. He has claimed that the contrast in (35) should be explained in the same way as that in (29). The fact that him cannot be co-indexed with Bill follows if standard condition B (or for that matter R&R's chain condition) must apply to the elements in the elided VP. In the elided VP, him is illegally locally bound by Bill.
(35) * Mary likes him and Bill does too (like him)
The contrast in (35) seems to raise a problem for a version of vehicle change that would save the strict reading of (31a) from violating a locality condition. However (35), could also be explained by R&R's condition B. The antecedent VP in (35) is not reflexive marked. By structure identity, the elided VP cannot be reflexive marked. Condition B, thus does not allow the elided VP to be reflexive and (35) is ruled out.
Now consider the contrast in (36). This contrast runs parallel to that in (31). (36a) is good because there is no reflexive predicate in the elided VP and no violation of R&R's condition B. (36b) is bad because there is a reflexive predicate in the elided VP which is not reflexive marked.
(36) a. As for Jessie Owens, Mary expected [him to win the race] and Jessie (himself) did too.
b. *As for Jessie owens, Mary told him [PRO to win the race] and Jessie (himself) did too.
This contrast cannot be explained by Kitagawa and as it turns (35) together with (36) serves as an additional argument for the explanation presented here.
This account of strict sloppy ambiguities with reflexives, has a surprising prediction for Dutch. Recall the contrast in (9) repeated here in (37). Zich which is not licensed in (37a) is licensed in (37b) due to the presence of zichself. This contrast demonstrates that the distribution of zich should not be explained in terms of a constraint on zich but rather in terms of a constraint on the predicate of which zich is an argument. The contrast is, in my view, the strongest argument for the existence of a condition on predicates like R&R's condition B.
(37) a. *Jij hoorde [Jan1 zich1 becritiseren]
You heard Jan criticize SE.
b. Jan1 hoorde [zichzelf1 zich1 becritiseren]
Before getting into the issue of strict and sloppy ambiguity, I need to show how (37b) is licensed with the unified notion of reflexive-marking which I have offered. The question is how can self undergo head movement to the predicate critiseren from its subject position. What I propose happens is that critiseren first has to incorporate to the matrix verb hoorde, and then self adjoins to the amalgamate. The derived LF is presented in (38)
(38) Jan hoorde-criticeren1-zelf2 [zicht2 zich t1]
This has a surprising prediction with respect to strict and sloppy interpretations. Consider the two sentences in (39). In (39a) we have a reflexive predicate downstairs. Therefore in order for condition B to be satisfied we need verb incorporation followed by reflexive-marking. In (39b), by contrast, there is no reflexive predicate. Therefore, condition B doesn't require any reflexive-marking.
(39) a. Jan1 hoorde [zichzelf1 zich1 becritiseren]
b. Jan1 hoorde [zichzelf1 Henk becritiseren]
There is thus a surprising prediction, with respect to the availability of strict and sloppy interpretations in VP deletion contexts. The prediction is that a construction like (39b), in which reflexive-marking isn't forced, should allow for both a strict and a sloppy interpretation. A construction like (39a), in which reflexive-marking is obligatory, should allow only a sloppy interpretation. That this is the case is demonstrated in (40).
(40) a. De lerares1 zag [zichzelf1 Mary strafen] en de andere kinderen ook.
The teacher+fem saw [herself M. punish] and the other children too.
'The teacher imagined [herself punishing Mary] and the other children did too.'
b. De Capucijner1 monnik zag [zichself1 zich1 strafen] en Bill ook.
The Capuchin monk saw [himself SE punish] and Bill too.
'The Capuchin monk imagined [himself punishing himself] and Bill did too.'
In (40a) there is no reflexive predicate. Thus no reflexive-marking is required and we get either a strict or a sloppy interpretation. (40a) could, thus, imply either that the children imagined themselves punishing Mary or that they imagined the teacher punishing Mary. In (40b), by contrast, punish is a reflexive predicate. In order for condition B to be satisfied punish must incorporate to imagine and only then can the complex predicate be reflexive marked Accordingly, at least as I was told by the four speakers I inquired, (40b) can only have a sloppy interpretation. I.e. It can only Imply that Bill imagined himself punishing himself. It cannot imply what would pragmatically make more sense (all of us knowing that Bill is not the kind of fellow that would consider self-punishing) i.e. that Bill imagined the capuchin monk punishing himself.
This contrast in (40) is pretty surprising. It shows that the existence of zich determines whether zichself is to be interpreted with a strict or a sloppy identity. It thus shows that the question of strict and sloppy identity is not a question about the relationship between an anaphor like zichself and its antecedent, but rather a question of the environment, a question regarding the predicate and its reflexive-marking.
I have argued that condition A of R&R can be reduced to their chain condition. This reduction leaves us with a very simple theory in which the chain condition does most of the work with a small residue for a condition on the licensing of reflexive predicates. Such a system deals in a very natural way with certain problems in R&R's theory regarding the distribution of reflexives. In addition, it requires a modification in R&R's chain condition, which turns out to be independently motivated by the distribution of pronouns. Once all of this is done the definition of reflexive-marking can be unified and certain observations regarding the availability of reflexives to partake in strict sloppy ambiguities are readily explained.
Aissen, Judith. 1974. "Verb raising", LI 5, 325-366.
Baker, Mark. 1988. Incorporation. A theory of Grammatical Functional Changing, Chicago, IL: University of Chicago Press.
Chomsky, Noam. 1986. Knowledge of Language, its Nature, Origin and Use, New York: Praeger.
Chomsky, Noam. 1992. A Minimalist Program for Linguistic Theory, Cambridge, MA: MITWPL.
Fiengo, Robert and Robert May. 1991. Indices and Identity, to be published at MIT press.
Heim, Irena.1991. Anaphora and Semantic Interpretation: A reinterpretation of Reinhart's Approach, Ms. MIT.
Haikawa, Takako 1993, Reflexivity in Japanese: An LF Analysis of Zibun Binding, Cambridge, MA: MITWPL.
Hestvik Arild. 1992 "Subordination and strict identity interpretation of reflexives", In Proceadings of the Stuttgart Ellipsis Work-shop.
Kitagawa, Yoshihisha. 1991. "Copying identity", NLLT, 9, 495-536.
Lasnik, Howard. 1993. "On Antecedent Contained Deletion", unpoblished ms. Uconn.
Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation, London: Croom Helm.
Reinhart, Tanya and Eric Reuland. 1991. "Anaphoric Territories", unpublished ms., Tel Aviv University and Utrecht University.
Reinhart, Tanya and Eric Reuland.1993. "Reflexivity" , to appear in LI, No 4.
Rizzi, Luigi. 1991. Relativized Minimality, Cambridge, MA: MIT Press.
Sag, Ivan. 1976. Deletion and Logical Form, Doctoral dissertation, MIT.
Williams, Edwin. 1977. "Discourse and Logical Form", LI 8, 101-139
Department of Linguistics and Philosophy
Massachusetts Institute of Technology
77 Massachusetts Ave.
Cambridge, MA. 02138