I argue that there is only one true anaphor in natural language, which takes many shapes. Building on the idea that some pronouns are constructed and others are ‘‘natural-born’’ with features, as suggested by Kratzer (2009), I suggest that all nonlocally anteceded bound variable pronouns that are traditionally bound (c-commanded) are the spell-out of a special, but universally available, dependent form, D-bound. The language-specific spell-out of D-bound in phase-internal contexts is responsible for Principle A effects, but not every language imposes phase-internal morphology for binding (thus, Principle A effects are not universal). This approach resolves morphological paradoxes that arise in ellipsis contexts when distance-bound antecedent forms behave as locally bound and vice versa. Finally, the distribution of Principle B and C effects is shown to be a consequence of the distribution of D-bound—the preferred form for anaphoric relations wherever it is available.
In this article, I will argue that there is only one true anaphor in natural language, which takes many shapes. Building on the idea that some pronouns are constructed and others are ‘‘natural-born’’ with features, as suggested by Kratzer (2009), I suggest further that, just like locally bound variables, all nonlocally anteceded bound variable pronouns that are traditionally bound (c-commanded) are the spell-out of a special, but universally available, dependent form, which I call D-bound.
At the heart of my proposal is the assumption that all bound anaphora, whether bound locally or at a distance, is achieved by a single, universally available, grammatical form, D-bound, the one true syntactically sensitive anaphor. The following assumptions are made about D-bound.
(1) Properties of D-bound
Always a variable: D-bound is the same object in SEM (the syntactic input to semantic interpretation) in all cases; it is interpreted as a bound variable regardless of its φ-features.
Always A-bound: The binder of D-bound (its antecedent) must c-command it from an A-position; that is, the D-bound form is A-bound.
Always feature-compatible: D-bound must be feature-compatible with its antecedent (informally, this property may be termed antecedent agreement).
Morphological shape conditions: Spell-out of the morphological shape of D-bound is potentially sensitive to whether A-binding is phase-internal (essentially following Kratzer 2009).
Agreement compatible with morphological shape may be determined by phase-internal factors locally distinct from antecedent agreement.
D-bound enters the derivation with φ-features arbitrarily assigned to it.
Anywhere phase-internal shape is not required, D-bound receives default pronominal shape.
The first assumption, that D-bound is always a bound variable, means that it is semantically interpreted as λx(...[x[+F]]...), where [+F] are further semantically significant features, if any, that restrict the variable and are not attributable to agreement (such as logophoricity or distributivity) (a point expanded on in section 5). The formula does not assume that c-command conditions semantic binding, only that x is bound in some domain that λx has scope over (a point expanded on in section 3). The local A-binder of the bound variable D-bound is its antecedent, although I will argue that some pronouns can also be interpreted as bound variables even when they are not A-bound.1
The key point is that D-bound spells out differently depending on when its antecedent enters the derivation, how far away from its antecedent it is, and how it agrees with its antecedent. The idea is that the phonological shape of surface pronouns has two origins: either the universal bound form D-bound is given pronominal shape as a default and random φ-features if it does not have an A-binder at the end of a phase, or it is given the shape of a local bindee if it has an A-binder within a phase. D-bound, however, is never well-formed if A-free (even when it looks like a pronoun). Binding of D-bound requires feature compatibility wherever it holds, however compatibility is achieved (unification, matching, nonconflict), as conditioned by local or default shape and the nature of the antecedent. The existence of local anaphoric shape is thus not expected to be a universal, though there are functional pressures for local anaphora resolution that can exploit phase-internal A-binding. Arguments that local A-binding should reduce entirely to the locality of Agree, or that distance A-binding should reduce to context shift, are rejected, but the role of heads in effecting binding proposed in Kratzer 2009 is adopted, primarily as A-binding by Spec,v as effected by v. The details are reserved for later sections.
This perspective makes several theoretical results possible. First, the parallel between the contexts where bound anaphora is possible and the existence of Principle C effects, first posited by Reinhart (1983) and Grodzinsky and Reinhart (1993), can be rehabilitated. A problem faced by that approach has been the existence of bound variable interpretations in what Hornstein (1995) has called ‘‘almost c-command’’ contexts, that is, cases where a quantifier originates (or appears on the surface) in a position that does not c-command the pronoun it binds as a variable. In all such cases, Principle C effects are not observed, contrary to Reinhart’s Generalization. Assimilation of the almost c-command effects to E-type (or D-type) anaphora does not contribute to solving this problem, especially if D-type anaphora is also argued to be available in contexts of c-commanded bound variables (as argued by Elbourne (2005)). The D-type approach renders Principle C effects completely independent of bound variable phenomena and unexplained.
I will argue that Reinhart’s complementarity approach is the right one in general and that it can be implemented straightforwardly (insofar as the syntax is concerned) if we assume the existence of D-bound and distinguish it from other sorts of bound variable or covariation phenomena. By distinguishing between bound variable pronouns and the smaller class of cases where D-bound is involved, it becomes possible to restore Principle C to a simple reflex of lost competition to the D-bound form. Natural-born pronouns and names thus must always be free wherever they would compete with D-bound; and in this respect, natural-born pronouns and names have exactly the same status with respect to obviation in this theory (they both lose to D-bound). By contrast, almost c-command bound variable pronouns are treated here as true pronouns, which presumably have to be shifted or refeatured under variable binding, but this operation is only available where D-bound would not be well-formed (and so this sort of binding of a variable does not induce obviation).
The second potential advantage builds on Kratzer 2009 to show that all anaphoric diversity is morphological: the syntax of D-bound is everywhere the same. All morphological variability in the phonological realization of D-bound is due to morphological shape conditions or language-specific feature compatibility (e.g., syncretisms in paradigms, gender presupposition avoidance). However, the diversity of the morphology is not just arbitrary. I argue here that functional pressures ensure that creases in what Universal Grammar (UG) allows are exploited in a variety of ways that favor certain outcomes. Thus, even though A-binding (always conditioned by c-command) is the same relation, whether local or not, the universal distinction between phase-internal and phase-external relations permits the syntax to express a distinction, one that most languages exploit, perhaps because it permits early anaphoric resolution. The incidence of languages that do not distinguish local anaphora in this way is predicted to be rare, but possible, a result consistent with what is known. Relations beyond phase-internal relations (far vs. very far) do not exist, modulo ‘‘hoisting’’ analyses (term from Safir 2013), where D-bound raises into positions where it can receive a phase-internal antecedent (and its morphological shape can be influenced). Moreover, on this account, binding is distinct from both local agreement (Safir 2010) and context-shifted indexicality (Safir 2004a, Anand 2006), and I will demonstrate that A-binding can and must hold where Agree and indexical shift cannot account for the morphological effects.
The One True Anaphor approach also permits certain morphological shape paradoxes, paradoxes that involve proximity of binding for strict and sloppy readings, to be resolved without appeal to vehicle change (in contrast to proposals in Safir 2004c, for example). These cases involve ellipsis contexts where nonlocally A-bound pronouns (instances of D-bound) interpreted as variables in antecedent VPs are treated as locally A-bound variables in ellipsis sites, even though the pattern of parallelism appears to involve a ‘‘strict’’ reading. This shows that strict readings are not essentially coreferential ones (e.g., as compared with Grodzinsky and Reinhart 1993) and explains how the pattern is derived.
Another important advantage of positing D-bound is that certain agreement paradoxes that involve A-binding of apparently interpretable features can be resolved if features on D-bound serve only to achieve compatibility with their binders—they are not interpreted, since D-bound is always a variable. I demonstrate that logophoric pronouns that do not have the morphological form of indexicals in simple sentences must enter the derivation as D-bound assigned a logophoric feature in addition to whatever other features it is randomly assigned. D-bound, when it takes pronominal form, is also a potential phase-internal A-binder for another D-bound that must then have local shape and agreement consistent with the D-bound that antecedes it. In a language with morphologically marked logophoric pronouns, D-bound in pronominal logophoric form is predicted to be a possible antecedent for a locally bound D-bound that has local shape as well as logophoric morphology (as is the case in Yoruba; see Safir 2013:539). Henceforth, I treat the term pronoun as a descriptive term for a form that is phonologically realized solely on the basis of its feature content; that is, it contains no independently identifiable morphological stem.
The presentation will proceed as follows. In section 2, with fleshed-out assumptions and sample derivations, I show how the system works. In section 3, I defend the view that Principle C effects arise from complementarity with the distribution of D-bound. I demonstrate in section 4 that the distinction between the local and long-distance morphological shape of D-bound is determined independently of its interpretation. In section 5, I argue that long-distance A-binding is morphologically necessary to account for the distribution and interpretation of (at least) logophoric pronouns, distinguishing these pronouns from those that might be treated by other paths to coconstrual (indexicality, indexicality shift, and D-type anaphora). In section 6, I look at the issues arising from phase-internal shape crosslinguistically, including the argument that functional pressures exploit seams in the system generated by UG. A short conclusion follows.
2 How the System Works
The easiest way to understand what this theory predicts about the distribution of D-bound is to see how standard examples would be derived. Derivations in various degrees of detail will be offered for the examples in (2), where italics indicate coconstrual.
John praised himself/*him/him.
The men say that *themselves/they are winning.
The men expect themselves to win.
The men seemed to each other to be guilty.
The men expect each other to question themselves.
(3) Assumptions about phases and Spell-Out
Phase-determining heads are C and v.
Anything merged above the phase head and its complement (sister), but below the next X0 merged to the undominated node dominating the phase head, is in the phase edge of the phase for the phase head in question.
A phase closes when the phase edge is capped (by the ‘‘next X0’’ of (3b)).
The complement of the phase head is sent to Spell-Out (for morphological insertion) and cannot be further altered by Spell-Out.
i. Spell-out of the complement of the phase head is conditioned by copy relations between occurrences in the phase edge and occurrences in the phase complement.
ii. The lower of two copies is unpronounced (not spelled out), either not ever, or sometimes not, depending on further assumptions (e.g., Safir 2010).
A probe in the Agree relation cannot access anything in the complement of the first phase head below it.
There is a v in structure whether v assigns an external argument or not.
Except for (3g), which has been argued for by Legate (2003), these are common assumptions in the literature on Minimalist syntactic architecture, and I will not defend them here (see, e.g., Chomsky 2004 for (3a–b) and Fox and Pesetsky 2005 for (3b,d–e); (3c) is implicit)). Notice that (3f) is a less general assumption than the Phase Impenetrability Condition, since I assume that feature compatibility for pronouns by binding will have to hold over a distance and across phases, as discussed below, though pronouns will have to be spelled out when the phases they are in are capped.
[VP praise D-bound + 3sg]
[vPJohn [v [VP praise D-bound + 3sg]]]
[vPJohn [v [VP praise pronoun + 3sg-self ]]]
[vPJohn [v [VP praise himself ]]]
[TPJohn [T [vPJohn [v [VP praise himself ]]]]]
In (4a), D-bound is merged as the complement to praise and D-bound already has features assigned to it. Before the phase [vPJohn [v [VP praise D-bound]] is capped, John is the antecedent (A-binder) of D-bound; and since John is within the phase edge, D-bound takes on phase-internal shape (pronoun-self ). It is not clear that there is any additional relation of agreement rather than compatibility with an antecedent in this case, but it is possible that additional feature compatibility may be imposed by v (see below). Consistent with a Distributed Morphology approach, I then assume that Spell-Out will pick the form from the pronoun-self paradigm that expresses the features of D-bound + 3sg. Subsequent raising of John to Spec,TP (where by Spec I simply mean the first nonterminal node merged above a head and its complement) will not change anything of relevance here, on the assumption that T can probe the phase edge, particularly Spec,vP, for its agreement features (on agreement with T, which is not crucial in this account, more below). This ensures that Johnpraisedhim is a morphological impossibility, unless him is not bound by John, in which instance it is a DP assigned only φ-features (and presumably Case), hence realized phonologically as a pronoun.
(5) Binding of D-bound is achieved when
D-bound is in the complement of a minimal c-commanding head H, and
Spec,HP is the A-binder of D-bound, and then
Spec,HP is designated as the semantic binder of D-bound in the property formed by H and its complement.2
If Spec,HP is the A-binder of D-bound and H is the minimal phase head c-commanding D-bound, then D-bound has phase-internal morphological form.
The φ-features of v are inherited from its Spec.
D-bound must be feature-compatible with v, where v=H in (6a). (Agree)
(7) D-bound must be feature-compatible with its A-binder.
These assumptions will ensure the outcome in (4c), but notice that antecedent agreement (in (7)) with Spec,vP may be all that D-bound requires, so it is not obvious that Agree (in (6c)) must value D-bound, at least for the English cases. Evidence for (6c) (and its enabling assumption (6b)) is only found where phase-internal agreement is different from agreement outside the phase, and it must come from instances where morphological shape is not the only reason for the difference (e.g., the pronoun-self paradigm is feature-compatible in English in the same way it would be at a greater distance, but sometimes phase-internal shape requires particular agreement for local D-bound). Contrary to Kratzer (2009), I reject (6c) as a general requirement on D-bound, treating instances where (6c) holds as special cases (see sections 5 and 6).
Now let us return to derivations, beginning with (2b). As illustrated in (8), D-bound does not have an A-binder in the v phase, but it is on the edge of the v phase in (8a), so its shape will not be determined until the next phase, which is the C phase, which closes when say is merged as in (8b). Since D-bound is not bound phase-internally in (8), it must spell out in the C phase as pronominal, taking the pronominal form consistent with its randomly assigned features. These features must be compatible with any antecedent that this instance of D-bound has. Only the highest copy of this D-bound in pronominal form will spell out, as in most accounts of how copies spell out. Notice that there is no way to spell out themselves for D-bound since there is no point in the derivation where themselves has a phase-internal binder (and certainly no binder mediated by v); there is no way to add phase-internal-bound morphology.
[vPD-bound + 3pl [v [VP winning]]]
[say [CP [that [TPD-bound + 3pl [T be
[vPD-bound + 3pl [v [VP winning]]]]]]]]
[vP[the men] [v [say [CP [that [TPD-bound + 3pl [T be
[vPD-bound+3pl [v [VP winning]]]]]]]]]]
[vP[the men] [v [say [CP [that [TPthey [T be
[vPD-bound + 3pl[v [VP winning]]]]]]]]]]
Since D-bound need only have an A-binder at some point, it does not matter how long it must wait for the c-commanding binder; in other words, A-binding is crucially unbounded. As it happens, the A-binder in (2b) is the external argument of say (i.e., the men). If the men is the A-binder of D-bound + 3pl, then feature compatibility is required—and it being satisfied here, the result is a grammatical structure where morphology will spell out the highest copy of D-bound+3pl as they (as in (8d), where the underlined copy is unpronounced). SEM, which sees (8c), will interpret D-bound as a bound variable of the men. If the men is not assigned as the A-binder of D-bound, then D-bound is free and the sentence is ill-formed. For an interpretation where they is not semantically bound by the men (i.e., it refers to some other 3pl entity), the external argument of win would have had to originate as a DP consisting only of φ-features (which will spell out as a pronoun)—not as D-bound.
The derivation for (2c) begins the same way as the derivation of (2b); the difference lies in the absence of a phase head between the external argument of expect and D-bound, where the external argument raises to Spec,TP of the infinitive.
[vPD-bound + 3pl [v [VP win]]]
[vPthe men [v [expect [TPD-bound+SELF + 3pl [to
[vPD-bound + 3pl [v [VP win]]]]]]]]
[vPthe men [v [expect [TPthemselves [to
[vPD-bound + 3pl[v [VP win]]]]]]]]
Since A-binding is phase-internal in (9b), phase-internal-bound shape is added; D-bound + SELF + 3pl is feature-compatible with its binder, the men; and then D-bound + SELF + 3pl spells out as themselves, as in (9c).
[v [VP [to D-bound + RCP] [seem
[TP [the men] [to [vPthe men [v [VP BE guilty]]]]]]]]
[vP[the men] [v[VP [to D-bound + RCP] [seem
[TP [the men] [to [vPthe men [v [VP BE guilty]]]]]]]]]
[T[vP[the men] [v[VP [to each other] [seem
[TP [the men] [to [vPthe men [v [VP BE guilty]]]]]]]]]]
For reasons I will not explore here, I assume that each other is not decomposed in English, but a single lexicalized form.4 I also assume that seem is the head of a VP shell complement to v when an experiencer argument is overt and that to D-bound + RCP is the specifier of the V shell. I will not try to explain how raising to EPP (Extended Projection Principle) position ignores the Spec,vP position (of seem) and I will assume that D-bound + RCP does not c-command the lowest copies of the men, as these mysteries are not peculiar to my approach. Rather, this derivation is intended to illustrate how D-bound + RCP becomes A-bound within its phase. This does not occur until the men raises to Spec,v position (10b), where the men must stop (since v heads a phase) on its way to the matrix Spec,TP to satisfy the EPP. By hypothesis, there are no phase heads between the v immediately above seem and the v of the lower clause (i.e., no C node). Since binding holds between the men and D-bound at point (10b) in the derivation, D-bound + RCP will have a phase-internal-bound shape, which is spelled out as each other when the vP phase containing seem is capped, as in (10c). Further movement of the men to Spec,IP changes nothing and I do not illustrate it here, but Agree can value the features of the matrix T because it c-commands the men within the same phase, as in (10c).
Finally, consider (2e), where one anaphor is the A-binder of the other, although the two instances of D-bound differ in their content. The first D-bound is merged without the RCP feature, so it acquires phase-internal-bound shape when it has an A-binder (see (11b)) before it spells out as in (11c). The D-bound that has the RCP feature will not have an A-binder until (11d).
[VP question D-bound + 3pl]
[vPD-bound + RCP [v [VP question D-bound + SELF + 3pl]]]
[TPD-bound + RCP [to [vPD-bound + RCP [v [VP question themselves]]]]]
[vP[the men][v[VP expect [TPD-bound + RCP [to [vPD-bound + RCP [v VP question themselves]]]]]]]]
[TP[the men] [PRES [vP [the men][v[VP expect [TPeach other [to [vPD-bound + RCP [v [VP question themselves]]]]]]]]]]
In the last step, (11e), the higher vP phase is capped and D-bound + RCP spells out as each other, presumably because each other is feature-compatible with any plural (it is not clear whether or not each other needs to carry any plural feature, given its interpretation, but such a feature is easily added if there is evidence for it).
Examples like (2e), where one anaphor binds another and they have different meaning and form, are easily derived in this theory; however, they pose a problem for movement theories of anaphora, which crucially derive locality of anaphora from locality of movement (see Safir 2008, 2013, for discussion). No appeal is made here to φ-feature transmission and so there is no way to identify such transmission with A-binding between the reciprocal and the reflexive. Only if that A-binding is local does phase-internal morphology appear on D-bound; but the morphology is not the A-binding itself, since D-bound is well-formed whether the antecedent is within the phase or not. In certain cases, phase-internal-bound morphology requires more specific feature compatibility than is seen outside the phase (and Kratzer (2009) discusses cases of local binding of fake indexicals with this character), but there is no evidence that these specific requirements amount to the actual binding itself, rather than reflexes of a particular syntactic context where binding holds. That no appeal is made to feature transmission by v suggests that Spell-Out will result in the right features on D-bound even when v, or T above it, has no φ-features to transmit, as is widely the case empirically—for example, with binding relations in infinitives. This only shows that visible local agreement is not necessary to implement A-binding or semantic binding. Later, I will argue that it is also not sufficient to implement local binding, at least not without ad hoc assumptions.
The key point is that A-binding is not required to be a local relation in the syntax for any bound form. A-binding can take place at the first phase in the derivation where the configuration allows it—locality effects are all about spell-out of either phase-bound D-bound (anaphoric shape) or D-bound bound outside the lowest phase that contains it (most typically, with resulting pronominal shape). The role of Agree in this system is minor, affecting spell-out of local morphological shape in some cases, but not in English. I postpone discussion of why there should even be a local/nonlocal morphological shape distinction until section 6.
This account departs notably from accounts like the one in Safir 2004b, where morphological forms of bound elements compete to represent an interpretation with respect to a particular antecedent. The complementarity that plays such an important role in that reasoning is achieved here by the determinants on spell-out options, so in this account, no competitive calculation that considers more than a single node is necessary with respect to morphology. As noted in the next section, however, competition still plays a role in interpretive outcomes.
3 Principle B and Principle C Effects
The most influential account of Principle C effects is that proposed by Reinhart (1983) and its slight reformulation by Grodzinsky and Reinhart (1993). The present account relies on the following generalization from those earlier accounts:
Wherever a pronoun Y can be interpreted as a variable bound by X, X and Y cannot be coconstrued (are obviative) if Y is not a pronoun.
What I am calling Reinhart’s Generalization was a potential advance because without it, Principle C is just a stipulation unrelated to any other generalization that does not produce the same effect. Reinhart also tried to explain the relationship between bound readings and obviation stated in (12) in pragmatic terms with her Rule I, as formulated in Grodzinsky and Reinhart 1993:79.5
NP A cannot corefer with NP B if replacing A with C, C a variable A-bound by B, yields an indistinguishable interpretation.
This formulation can be understood as a competition to represent the bound reading between pronouns, on the one hand, and names and descriptions, on the other, where the losing form is limited in the class of interpretations it can have by comparison with the winning form. Since, on this account, the bound variable interpretation is assumed to be the favored way to achieve coconstrual between a form and its antecedent, any speaker who does not use the form that permits the bound variable interpretation (a pronoun) must not mean to indicate a coconstrual, or must not mean to signal coconstrual in the usual way.
I will set aside discussion of (13) as an explanation of Principle C effects for the time being, however, because Reinhart’s explanation in (13) depends on (12)’s being true—and there are well-known counterexamples to (12) that are set aside by reference to A-bound in the formulation of Rule I. In particular, consider the paradigm in (14), which shows the missing parallel between the bound reading in (14b) and that in (14a), where the existence of a bound reading correlates with a Principle C effect in (14a), but not in (14c).
*He thinks that John is a genius.
Every boy thinks that he is a genius.
His mother thinks that John is a genius.
Every boy’s mother thinks that he is a genius.
Cases like (14d), which illustrate what Hornstein (1995) has called ‘‘almost c-command,’’ are counterexamples to Reinhart’s Generalization, insofar as Principle C effects are predicted for (14c), but not observed (hence the reference to A-bound in Rule I). As noted in Safir 2013, where some of these cases are discussed, Barker (2012) details a much wider range of violations of the c-command condition on the interpretation of pronouns as bound variables, including examples essentially like (15a–c), where the quantifier takes extrawide scope.
The scope of each book has expanded on that of its predecessor.
When the game ended, the amount of wealth that each person had accrued was added to their overall score.
After unthreading each screw, but before removing it, make sure the boards will remain aligned, so you can replace it later.
Barker argues that these are not E-type readings, and I agree, particularly since E-type readings are typically observed only for weak quantifier antecedents, which these are not (but I return to this question at the end of this section).
I will assume that bound variable readings for pronouns are possible wherever D-bound or pronouns are in the scope of the quantifier for which they permit a semantic bound variable reading, and in perhaps a few other cases, modulo weak crossover effects (see Safir 2004c). One reason why almost c-command cases are less common is that quantifier extraction out of a subject is syntactically difficult. What appears to be the easiest extraction is that of the specifier of the subject DP, hence the almost c-command effect. Kayne (1994:22–27) suggests that this c-command is not ‘‘almost’’ but actual c-command, on the assumption that the highest adjoined position in DP (covertly or overtly) is adjoined to DP, hence not dominated by DP on the segment theory of May (1985). Consider one of May’s classic inverse-linking examples, Someone in every city loves its weather. If adjunction to DP repeats the DP node, such that the two DP nodes of (16) are ‘‘segments’’ of DP, and if X only dominates Y if all segments of X dominate Y, then every city in (16) is not dominated by DP.
(16) [every city [someone in every city]] loves its weather
Thus, in Kayne’s system, every city would c-command its. Appeals to allow specifiers to c-command out of the subjects that contain them are harmless if local anaphora is limited by coargumenthood (as in Reuland 2011) (e.g., to avoid Every boy’smother loveshimself ), but this does extend to long-distance cases, where Reinhart and Reuland’s (1993) CHAIN relations do not hold in any case (links are conditioned by locality, unlike in the long-distance cases). To preserve Reinhart’s Generalization, it has to be the case that only quantifiers in a subject DP could adjoin to that DP, which is feasible if Spec,DP is distinguished from adjunction to DP. However, even this move will fail, as I will show below when I reconsider appeals to D-type readings as almost c-command fixes for (12).
From the perspective of the D-bound approach, D-bound is only well-formed when it is embedded in the complement of a head H and the first-merged XPa above H (which is what I define a specifier to be) is the A-binder of D-bound. Thus, an adjunct cannot be an A-binder of D-bound, where an adjunct is an XPb merged to [XPa [H complement]], as in the almost c-command cases.6
It is possible, however, to defend a new version of Reinhart’s approach based on the distribution of D-bound rather than the distribution of the wider class of possible bound variable readings.
If X can be a binder for D-bound in position Y and Y is not D-bound, then X and Y are not expected to be coconstrued (i.e., they are obviative).7
Where a quantifier has scope over D-bound in the absence of a mediating head, as in the relevant almost c-command cases, D-bound is not available, but then binding without c-command can achieve the bound variable readings if the variable is something other than D-bound, namely, a natural-born pronoun.
(18) Everyone’s mother predicts his future success.
every x [[x’s mother] [predicts his future success]]
every x [[x’s mother] [predicts x’s future success]]
In (18), Quantifier Raising (QR) or some scopal device with the same semantic effect will establish binding of the pronoun as a variable, but there is no c-commanding A-binder for the pronoun, so D-bound is not a possibility there. I then predict that the almost c-command cases will allow bound anaphora, but will not induce obviation, since something other than D-bound is in the position of the bound variable (i.e., in (18b)).
Notice that ‘‘something other than D-bound’’ includes both natural-born pronouns, on the one hand, and names and descriptions, on the other, so obviation effects for names (Principle C) fall together with obviation effects for pronouns (Principle B), as in the competition theory proposed in Safir 2004b and, under very different assumptions, Heinat 2006. Both names and natural-born pronouns lose the competition for the bound reading whenever D-bound could be well-formed in the same position. Even if coconstrual is enforced by the character of 1st and 2nd person pronouns, which requires them to independently pick out the same referent in discourse with each use, Principle B effects are observed for identical indexicals, as predicted by Syntax-Induced Obviation.
*I killed me.
*I defended me.
*I voted for me.
Apart from locality effects, where phase-internal-bound shape is imposed, one might ask why languages have D-bound if natural-born pronouns will serve the purpose of supporting bound anaphora more generally. My answer is that it is not always the case that natural-born pronouns can be successful bound variables, even in contexts where they do not compete with D-bound. Heim (2005) observes that an issue arises for 1st and 2nd person features on pronouns that appear to be bound variables. Since these pronouns appear to directly access the context of utterance, their interpretation as variables would appear to require them to shed this interpretive fixity—yet examples like (20a–b) are easy enough to come by.
I can’t do what I want, but others can.
Only I spend my money as I please.
These examples permit bound readings for the second 1st person pronoun, as evidenced by the available sloppy reading (e.g., the elided portion of (20a) can be understood as ‘others can do what they want’, and (20b) can imply that others do not spend their money as they please). Thus, for earlier accounts the question arose how such indexical features could be suppressed or coopted. In the account proposed here, inspired by ideas in Kratzer 2009, these cases are simply instances where the features of D-bound are compatible with those of the binder (suppose they are maximally compatible with the antecedent within the morphology of the language) and the features that appear in the form of a phonological pronoun (not natural-born) are uninterpreted by hypothesis. After all, local binding by a 1st person pronoun results in a reflexive form in English, and the features on the reflexive must be innocuous, so it seems a small step to assume that features associated at a greater distance can be just as empty on the dedicated bound form (especially when arguments against binding as the effect of agreement are introduced). Some cases like (20a–b) might be addressed by some version of a theory that shifts contextual coordinates (= coopted interpretation, as above), as discussed in section 5.2; but, as also discussed there, a significant class of cases seems to require binding rather than shifting.
Nonetheless, since I allow natural-born pronouns to be bound variables when D-bound is not available (i.e., in cases where there is no c-command), the question arises whether indexical pronouns can be bound in these circumstances. If natural-born indexical pronouns cannot shed their features, then they should not support bound variable interpretation as bindees. There does seem to be a contrast between indexical pronouns and natural-born 3rd person pronouns in this respect. Consider (21a–d) and the possible readings they support.8
Students should respect their teachers, but that’s just the difference between me and you.
My students respect me but your students don’t.
—sloppy reading unavailable
Students should respect their teachers, but that’s just the difference between me and Bill.
My students respect me but Bill’s students don’t.
—sloppy reading unavailable
Students should respect their teachers, but that’s just the difference between Bill and Joe.
Bill’s students respect him but Joe’s students don’t.
—sloppy reading available
Students should respect their teachers, but that’s just the difference between Bill
His students respect him but hers don’t.
—sloppy reading available
As (21c–d) show, a bound reading is achievable between a subject possessor and a 3rd person pronoun object so as to support a sloppy reading in the ellipsis context (probably because the possessor is contrastively focused). In these cases, the pronoun must be natural-born because D-bound cannot be licensed where its antecedent does not c-command it. The same sort of binding fails when the natural-born pronoun that would have to be bound is a 1st person indexical. Moreover, if a c-commanding 1st person pronoun is introduced, allowing both the lower pronouns to be instances of D-bound (with the matrix subject as A-binder), the sloppy reading improves.
In those days, students respected their teachers, and when you and I taught then, we were less skeptical.
I used to believe my students respected me, just as you did.
—? sloppy reading available
If these data are properly interpreted, the difference between the non-c-commanding my in (21b) and the c-commanding I in (22) reveals a difference in the behavior of fake indexical pronouns bound as variables and natural-born indexical pronouns where they would have to be bound as variables. Thus, in (22), where there is a c-commanding antecedent for my and me, the latter two pronouns can permit a sloppy interpretation (because they are D-bound spelled out as indexicals), whereas the natural-born indexical pronouns in (21b), where me cannot be A-bound and must therefore be natural-born, does not permit a sloppy reading. The fact that the indexical features on natural-born pronouns are necessarily visible to interpretation, whereas the indexical features on instances of D-bound are only visible to morphology.
The evidence given here supports the view that D-bound is the same syntactic and interpretive entity whether it is A-bound phase-internally or at a distance. In both cases, features on D-bound are invisible to interpretation, but visible to morphology. I have also shown that Reinhart’s Generalization does not work if Principle C effects are linked to contexts where bound variable interpretation is possible; but a revision of it, based on the distribution of D-bound, does make the right prediction about the distribution of Principle C effects.
(23) Every farmer who owns a donkey beats it.
Without going into the large and historic literature on this construction, the key idea (also explored, e.g., in Büring 2005) is that perhaps all bound readings of 3rd person pronouns are achieved in the same way. For this project to be successful, the differences between donkey anaphora, where the antecedent must be indefinite, and the more general cases of bound readings, which permit universals as the binding quantifiers, must be accounted for; and, in Elbourne’s proposal, which treats pronouns as radically reduced descriptions, a theory of ellipsis that is syntactically plausible is required. But whether these results are achievable or not, the success of such a project would lose the relationship between Principle C effects and the distribution of D-bound that is captured by Syntax-Induced Obviation (17). Since I am aware of no account of Principle C effects that is derivable from the distribution of D-type readings, the expansion of D-type anaphora, from the cases where the antecedent c-commands to those where it does not, seems a step backward.
I am treating obviation of the sort identified for Principle C effects as an expectation of non-coconstrual, one that can be overcome if, given the right pragmatic conditions, bound anaphora is to be avoided in order to achieve coconstrual. A version of Higginbotham’s (1980:234–235) famous example, discussed in Fiengo and May 1994:10 and Safir 2004c:9, suffices to make the point.
(24) We are at a café and I spot John across the room about to leave. You think it isn’t John, which is baffling to me because it is so obvious. I say:
OK, explain to me why he is putting on John’s coat.
OK, explain to me why he is putting on his coat.
Bound reference is not intended, and he does not get special stress here. The obviation effect is being exploited in (24a) to indicate that the only person who would be putting on John’s coat is John, so the addressee is supposed to infer the identity of the individual from the fixed reference of the owner of the coat. A statement like (24b) cannot be interpreted this way. Thus, binding is avoided, but coconstrual is intended. Under the understanding of obviation suggested above, this is just as expected, though my particular interpretation of what obviation is is not crucial to my proposal.
To summarize, D-bound is the same element when it is A-bound locally as when it is A-bound at a distance, as indicated by the fact that indexical features it bears are neutralized in the interpretive component. Insofar as D-bound is only well-formed when it is in a head-mediated binding relation (A-binding), it is never A-bound by something that does not c-command it, nor by something that does c-command it, but is not the designated argument of a head (e.g., an adjunct, or another Ā-binder; see footnotes 1 and 6). Principle C effects arise for any position that could antecede position X in the event that X is D-bound.
4 D-Bound and Ellipsis Contexts: Neutralizing Local Morphology
The theory developed so far has been supported with evidence based on the strict versus sloppy readings that are possible in cases of VP-ellipsis, but I have not offered a theory of how D-bound interacts with VP-ellipsis. In this section, I will flesh out my assumptions about ellipsis contexts a little further and then use this account to formulate another argument that long-and short-distance A-binding are not different in kind, but only in morphology. The essence of the argument is that where morphology is neutralized by ellipsis, locality does not figure in the well-formedness of D-bound.
I begin by assuming that parallelism, however it is achieved (for discussion, see e.g. Fox 2000), is an underlying requirement in ellipsis contexts like those in (25), so that the elided portion, whether it consists of structurally articulated but silent structure, or a structure deleted under identity with the antecedent DP, must permit the anaphoric element in the antecedent VP (VPa) to match a similar element in the elided VP (VPe).9
John [VPa loves his mother] and Bill does [VP e] too.
John [VPa loves D-bound + 3msg’s mother] and Bill does
[VPe love D-bound’s mother] too
John [VPa loves D-bound + 3msg’s mother] and Bill does
[VPe love 3msg’s mother] too
Under the coconstrued reading for VPa (coconstrual indicated by italics), the VPa is identical in (25b–c),10 but the readings obtained differ depending on whether D-bound is copied (or matched) or whether only the syntactic features of D-bound are copied (or matched). When D-bound is copied, it must have an A-binder, and Bill is available as the antecedent; the reading is sloppy. When only the features are copied, it is parallelism that requires an interpretation matching (25a); however, a bound variable interpretation is not possible without D-bound in VPe (because a pronoun is obviative where A-binding is possible), so parallelism can enforce only a strict reading for what will amount to a pronoun (a DP consisting only of features). Since feature compatibility only matters for the phonology, the features of D-bound do not have to be copied (matched) in an ellipsis context.
The same account can be applied to D-bound when it is morphologically reflexive. In (26a–b), the sloppy reading is always available, but the strict reading seems to be disfavored when the relation between the sentence containing VPa and the one containing VPe is most symmetric, as in (26a). By contrast, (26b), which is connected by a comparative rather than a conjunction, clearly permits the strict reading (for discussion and examples, see Hestvik 1995, Safir 2004c:30, 2013).
Zelda judges herself harshly and Alice does too.
Zelda judges herself more harshly than I ever do.
What is of interest here is that the strict reading is possible when conditions are right. The current account of it works just like it does for (25), where (27a–b) illustrate the two possible candidates for VPe.
Zelda [VP judges D-bound + 3fsg] more harshly than I ever do
[VPe judge D-bound]
Zelda [VP judges D-bound + 3fsg] more harshly than I ever do
[VPe judge 3fsg]
The sloppy reading is achieved when D-bound is copied, again without its features, which would not match I, but which are neutralized by ellipsis anyway. The strict reading also works just as it does in (25b), in that only 3fsg is copied/matched and parallelism forces the resulting pronominal to align semantically with a value in VPa.11
The account so far makes an interesting prediction, one that requires some preliminary exposition. First, notice that the theory predicts, as it should, that (28a) has only a sloppy reading.
No one loves his puppy more than Arthur does.
no one [VP loves D-bound’s puppy] more than Arthur does
[VPe love D-bound’s puppy]
If only the features of D-bound are copied, parallelism has no way to enforce a parallel reading, given that no referent is provided by the antecedent except a bound variable interpretation—but that interpretation fails because the resulting pronoun is not in the scope of the quantifier no one, and would have to be D-bound in any case to be a bound variable in that context (there is a local A-binder). That scope is the problem is proved by the success of (29a), which permits both a near sloppy reading (29b) and a distant sloppy reading (29c) but no strict reading.
Every boy believes that he loves his puppy and that St. Francis does too.
every boy believes that D-bound + 3msg [VP loves D-bound + 3msg’s puppy] and that St. Francis does [VPe love D-bound’s puppy] too
every boy believes that D-bound + 3msg [VP loves D-bound + 3msg’s puppy] and that St. Francis does [VPe love D-bound’s puppy] too
In this example, taken from Safir 2008:342–343, the only possible form in VPa is D-bound, hence an instance of a bound variable, and so only a sloppy reading is possible, since parallelism has nothing else to work with. The ambiguity arises because either the sloppy reading can result from local A-binding, as in (29a), or it can be a long-distance sloppy reading where, as in (29c), D-bound of VPe is A-bound by every boy (or its trace after QR), which has D-bound in its scope.
The prediction of interest here arises when the antecedent in VPa is a reflexive locally bound by a quantifier, but VPe allows a long-distance sloppy reading, one that would normally be represented by a pronoun. Such an example would show that the overt phase-internal-bound morphology (pronoun-self ) is irrelevant to determining whether D-bound is well-formed in an ellipsis context—D-bound must only meet the requirement that it have an A-binder. The examples in (30) and (31) bear this prediction out, in that they easily permit long-distance sloppy readings, where the object of elided promote, if morphologically copied from (30a), would be a reflexive form that would fail to be locally D-bound. No strict reading is possible because both of the potential antecedents are quantifiers. Therefore, the distant and local sloppy readings (plain and boldface italics, respectively) are all that is possible, just as in the case of (29a). Thus, (30) and (31) show that the phase-internal-bound morphology of D-bound in VPa is irrelevant to the interpretive well-formedness of D-bound in VPe.
Every athlete thinks he promotes himself
better than any agent could [promote him/himself ].
every athlete thinks D-bound+3msg [VP promotes D-bound+3msg]
better than any agent could [VPe promote D-bound/D-bound]
I have always believed that I can represent myself
better than any lawyer could [represent me].
I have always believed that D-bound+1sg can [VP represent D-bound+1sg]
better than any lawyer could [VPe represent D-bound/D-bound]
We have now seen that the licensing of D-bound is indifferent to the distance between itself and what binds it, even when it is copied from a counterpart that shows the morphology induced by a local A-binder. In other words, D-bound locally A-bound in VPa can be copied and nonlocally bound in VPe. The logic of the argument would lead us to believe that in a context where D-bound is nonlocally A-bound in VPa, it could be locally A-bound in VPe. As we will see, if one takes a surface pronoun to always be susceptible to Principle B (derived by Syntax-Induced Obviation (17)), then the sort of case I describe should always be excluded. I will show, however, that pronouns that are pronunciations of D-bound behave differently from pronouns that are not.
Looking at natural-born pronouns, we see that Principle B holds as usual.
(32) *Mary hates him and John does too.
(32) is consistent with the proposed theory, in that the only option here is to copy 3msg from him—which also results in a pronoun, not D-bound, so John cannot A-bind him. However, there are contexts where VPa contains a bound variable pronoun that is not locally bound and that can be A-bound by an antecedent that c-commands both VPa and VPe. The prediction we should make is that nonlocally A-bound D-bound (outwardly pronominal in form) in VPa should license locally A-bound D-bound (outwardly reflexive in form) in VPe.
?It turns out that only I believed that John could represent me better than I could (myself).
?John has praised me more convincingly than I ever could have.
In (33a), me is obviative locally with respect to John and can be a bound variable of only I. VPe, however, permits the interpretation ‘I could represent myself’, which is a locally bound reading, and even without elision, parallelism permits the nonparallel morphological form.
Before we declare victory, however, we must look at a complication. It is important to distinguish the natural-born 1st person pronoun here from D-bound with 1st person features. Suppose that (33a) is a mixed case, where the second I is natural-born and the bound variable in VPe is just D-bound directly bound by only I; that is, it is actually distance-bound. A similar attempt might be made to distinguish two paths to coconstrual in (33b), though the case is a bit harder to make here because it is not clear that me c-commands the object of VPe. To be sure that we have the crucial example, we must avoid using indexical pronouns, as in (34).
(34) He’s the sort of star athlete who believes that no agent can [VP promote him] any better than he can [VPe promote himself ].
In this case, A-binding of him by the trace of who in the antecedent clause cannot skip the intervening he after than on the grounds that it is a shifted indexical (since it is not an indexical); therefore, the binding in this case is local A-binding, and a bound variable reading is possible only if we assume that the object of promote is D-bound by the trace of who. If we judged the antecedent pronoun in VPa to be a natural-born pronoun, then copying it into the parallel position in VPe should result in a Principle B violation (obviation). Instead, the reflexive reading of (34), a local sloppy one, is almost perfect.
The argument that sloppy readings can be anteceded by VPs in which pronouns are A-bound, but not locally, can also be made with the paradigm illustrated in (35), where (35b) allows a reading for VPe whereby every student would believe in himself.
*Every student believes in him.
?Every student hopes that if Mary would just believe in him, then he would too.
Finally, consider (36a), where a local reflexive and an emphatic reflexive (with heavy stress on the emphatic) cooccur. In simple sentences, the bound variable interpretation requires the form with phase-internal-bound morphology, which is why (36b) is not possible.
Every politician praises himself himsélf though not always so convincingly.
Every politician praises *him himsélf though not always so convincingly.
Every politician hires a publicist whose job it is to praise him more convincingly than he can [VPe praise himself ] himsélf.
In (36c), however, where the first him is a bound variable (and thus D-bound), the bound variable interpretation extends not only to the subject of VPe, but also to its object. These cases establish that the Principle B effect can be absent where D-bound in pronominal shape is copied into an environment where it is locally bound and where it would have a local anaphor shape if overt.
I conclude from these demonstrations that D-bound is unselective about how locally it is A-bound and that only its morphological form, not its interpretation, is sensitive to locality. In ellipsis contexts where D-bound’s morphology is never pronounced, the morphology of its antecedent does not limit how locally it is A-bound. This is further evidence that D-bound is the same bound variable element everywhere, however it is pronounced.
5 Binding and Shifting
Kratzer (2009) argues that local binding is distinct from Agree, but she does not commit herself to the existence of phase-external binding. Since some current approaches (see, e.g., Reuland 2001, 2011, Heinat 2006, Hicks 2009) treat agreement as the device that achieves binding, it is perhaps useful to make clear that maintaining such a position requires a number of ad hoc assumptions about the presence of Agree relations when there is no overt sign of them. I address this issue in section 5.1 and turn to overt evidence of phase-external A-binding in section 5.2, as opposed to the covert evidence for phase-external A-binding offered in section 4.
5.1 Agreement Is Not Necessarily Correlated with Binding
Kratzer (2009:196) argues that local bound variable readings have distinctive agreement because they are locally bound by a lexical head that inherits features from its specifier. The head then imposes the features on the bindee by Agree, which is bounded by phases. One purported advantage of such an account is that Principle A of the binding theory is then derived as a consequence of Agree.
In Safir 2010:97, I make an explicit argument for distinguishing Agree from binding on the basis of certain Icelandic constructions in which an oblique argument is bound by a nonnominative subject, while the verb in the same clause agrees with a nominative nonsubject. This is illustrated in (37a–b), from Zaenen, Maling, and Thráinsson 1990:102, 112.
Henni þykir bróðr sinn / *hennar leiðinlegar.
she.DAT thinks brother SIN / her boring
‘She finds her brother boring.’
Konunginum voru gefnar ambáttir í höll sinni / ?hans.
king.the.DAT were given slaves in palace SIN / his
‘The king was given slaves in his palace.’
As pointed out in Safir 2010:97:
[T]he structural subject is Dative, so the verb agrees with the post-verbal Nominative. The possessive anaphor SIN in Icelandic (inflected for agreement with what it modifies), which must normally be bound by a structural subject, is bound by the Dative subject. . . . Moreover, it is also clear that verb agreement is not establishing the anaphoric relation (since the verb usually agrees with a structural subject that is Nominative). . . . If Agree on a functional node determines verb agreement, then a different functional node must be responsible for anaphor-binding.
Notice also that SIN is not bound by nonsubjects, so an appeal to the origin of the dative in a double complement structure must also explain why SIN has the subject-oriented shape. Thus, on this analysis, binding must occur after the dative element moves to something that counts as a subject position, which, in the theory presented here, is Spec,vP.12 Thus, the same functional node seems to be determining agreement with one constituent and binding with another.13
I will henceforth assume that binding is achieved as described in (5)–(7) and that the phenomenon known as (strict) subject orientation (for discussion and references, see Safir 2004b:170–173) is the morphological reflex recorded in some languages that can emerge when a phase-internal form is bound from Spec,vP position (recall that Spec,vP is just the name for the first element merged above v and its complement). Notice that what used to be Principle A is now reduced to the influence of phases on points in the derivation when the shape of morphology on a bindee is determined. There is no independent Principle A, and the Principle A effect is therefore not predicted to be a universal.
This result does not preclude the possibility that A-binding, when mediated by a local head, can affect the agreement as well as the local shape of D-bound. The cases of fake indexicals in German that Kratzer (2009) presents would presumably be instances of this kind, such that locally bound fake indexicals are sensitive to subject-verb agreement in a way that distant ones are not. I will not explore those cases here, as (37) shows that agreement can be independent from morphological shape induced by A-binding within a phase.
5.2 Phase-External Binding
As mentioned above, Kratzer (2009) is noncommittal about the existence of phase-external binding, but she proposes that feature transmission, which ensures agreement, is always restricted to local domains. Thus, if there is agreement at a distance beyond a phase, the mechanism of binding cannot be its source within her theory. She suggests instead that fake indexicals are born as pronouns, but that their indexical features are shifted, so that instead of picking out the utterer in the speech event, indexical picks out the utterer of reported speech, for example. In such cases, as Schlenker (2003) argues for Amharic, embedded 1st person pronouns behave like logophors, picking out the agent of the reported speech (or thought) instead of the utterer in the speech event. Kratzer further suggests that the apparent cases of binding at a distance that do not fall under context shift may turn out to be accounted for by Elbourne’s (2005) account of D-type readings, and that if so there is no real binding of pronominals outside of local domains. Thus, while her theory does not preclude nonlocal binding, it cannot account for long-distance agreement, and so she suggests that, if shifting and D-type readings account for these cases, there may be no need to appeal to nonlocal (A-)binding.
If indeed there is no real A-binding of outwardly pronominal elements outside of their local domain, then there is no reason to posit a nonlocal D-bound to contrast with local D-bound, and binding would appear to stand a better chance of being reduced to agreement. In this section, I will provide some evidence for nonlocal A-binding by showing that there are long-distance, structurally sensitive, antecedent agreement effects that neither context shift nor D-type readings can account for.
5.2.1 The Essentials of the Context-Shifting Account
Kratzer (2009) suggests that pronouns bound as variables at a distance are born as pronouns, unlike local forms bound as variables, which originate without features. In cases such as these, 1st and 2nd person features are present, and are problematic for interpretation unless their relation to the context (the utterer and the addressee involved in the speech act, respectively) is somehow neutralized or shifted. On the assumption that 1st and 2nd person pronouns are interpreted relative to a head that determines the contextual commitment of 1st and 2nd person, such that they bear their designated roles in that context, the proposal is that the context-bearing head does not have to pick out the context of utterance, under certain conditions, and can instead be shifted to the context of reported saying or belief, presumably in the manner ascribed to logophoric interpretation. To see how this works, consider (38).
(38) I believe that I am smart, just as others do.
On the strict reading, others believe the speaker of the sentence is smart, and in that instance, the features on the subordinate subject are born on it and unchanged by interpretive mechanisms. The 1st person pronouns each pick out the speaker in the context of utterance, which is the same context for both matrix and subordinate clauses. On the bound reading, the reading where others also believe that they are smart, the subordinate subject I is a ‘‘fake’’ indexical in Kratzer’s terminology. In effect, the context of utterance is shifted to the context of reported belief or reported saying, and so whoever the believer is, it is that person who is 1st person in the reported context. Thus, the subordinate I will pick out the believer in that context, and the believer in that context (the logophoric antecedent) is just the same as the utterer of the matrix sentence. The key point, however, is that the context-setting head (HCS) in question is not a binder; rather, it affects the interpretation of every natural-born pronominal in its scope, such that the person feature(s) in question is (are) reset for all of the natural-born pronouns below HCS. The analysis is illustrated schematically in (39).
(39) [I think [HCS [I am smart]]]
John jägna näNN yt-lall.
Johni hero Ii.am says-3MSG
‘Johni says that hei is a hero.’
There are various versions of this theory that weaken it. For instance, it has been assumed that other indexical elements (e.g., temporal ones, such as today) can be unaffected when the contextual coordinates of the logophoric operator are shifted; or, as suggested by Anand (2006), it is possible that some contextual coordinates can shift, but others cannot (e.g., the 1st person can shift without affecting the 2nd person) (see also Kratzer 2009:213 on a proposal made in Cable 2005). However, the key assumption is that all pronouns in the scope of the HCS with shifted features will have their features shifted in the same way.
The alternative to the context-shifting account is an operator-binding theory. On this theory, there is a logophoric operator that intercedes between the logophoric antecedent and the logophoric pronoun, but the logophoric operator is a binder of the pronoun that shows the logophoric morphology, whether it is 1st person morphology that is manifested on the bound variable pronoun or some other logophoric marking.
These two theories predict different empirical patterns, and we can test to see which is correct, at least for a given construction or for such constructions in a given language. All that needs to be shown for present purposes is that binding at a distance by an operator is required for some cases, since these cases justify the assumption made earlier that long-distance A-bound superficial pronouns are actually D-bound, not natural-born pronouns.
5.2.2 The Insufficiencies of the Context-Shifting Account
The context-shifting account of Amharic has sometimes been viewed as a model for the distribution of logophoric pronouns. Logophoric pronouns, so named by Hagège (1974), are those that are coconstrued with the subject of a propositional attitude verb, are embedded in the propositional complement selected by the verb, and are required to have what has been identified as a de se reading, although the cases discussed here could as easily be characterized as reflecting the contrast between speaker’s reference and reported speaker’s reference (see Safir 2004a). If these pronouns are actually shifted indexicals, they would be expected to look just like 1st and 2nd person pronouns, as they do in Amharic and some other languages; but in fact, the morphological shapes of pronouns interpreted logophorically are diverse. In Gokana, as reported by Hyman and Comrie (1981), the difference is reflected on subordinate agreement relations, and in Ewe, as reported by Clements (1975), there is a dedicated form for logophoric pronouns distinct from pronouns of other sorts. The form used for logophoric interpretations in subjunctive contexts in Icelandic is identical to the normally bounded anaphor sig, and in Yoruba, the form used logophorically is otherwise employed as a pronoun bound to focus. The first argument against the context-shifting account is that the range of morphological forms is in no way predicted by the contextual shift that is proposed.
One could imagine that all pronouns without features that are in the scope of a context-shifting operator with particular features must have a certain shape that does not correspond to the standard morphology for person, but that the relation of scope to shape is not binding because the context-shifting operator does not select any particular pronouns in its scope. It is not obvious how one could manufacture such a device, but let us suppose for the sake of argument that one could. We are comparing the shifting device to binding, where the logophoric operator is somehow anteceded by the logophoric antecedent, and the operator binds particular instances of D-bound in its domain that must, by virtue of being bound, have a logophoric shape to be feature-compatible. This structure is schematically presented in (41), where italics indicate coconstrual.
(41) John thinks [pro Log-Op [that he-Log is smart]]
On this analysis (see Safir 2004a and references cited there), the Log-Op is the mediating head that is inducing logophoric shape, potentially at long distance (as in the cases presented below), and either the Spec,v of the matrix verb (John) or a pro that it controls in the left periphery of the complement clause (possibly Spec,CP) is the designated antecedent. Sticking closely to the definitions in (5) and (6), the pro must be the antecedent that binds the logophoric pronoun, where the pro is bound by Spec,v (John), mediated by v. Thus, Spec,v is the ultimate A-binder.
The unselective scope-induced morphology account and the logophoric-binding account make markedly different predictions in interleaving examples (see Safir 2004a, Anand 2006). In an interleaving example, one logophor-inducing verb is embedded in the complement of another, and the lowest complement contains two logophoric pronouns, one associated with the highest logophoric antecedent and the other with a distinct lower one. If the context is shifted with each embedding, then all the logophoric pronouns in the lowest embedding should connect to their referents in exactly the same way; that is, they should be coconstrued. The interleaving theory predicts that the pronouns in the lowest clause could have different logophoric antecedents, depending on which logophoric operator they happen to be bound by. In other words, they could both be bound by the lowest operator; they could both be bound by the highest; or, crucially, one could be bound by the lowest operator and the other by the highest (interleaving). The evidence presented below shows that the binding theory makes the right prediction for Yoruba and that the context-shifting theory does not.
As mentioned earlier, Yoruba is a language with special logophoric morphology (see Adésolá 2004 for discussion and references). In both (42a) and (42b), the pronouns are interpreted as bound pronouns (the pronoun in (42b) optionally, and only with the particular context about to be explained), but only the ‘‘strong’’ pronoun òun in (42a) is logophoric. (The data in (42) are from Olúsèye Adésolá (pers. comm.); in (42)–(44), diacritics below segments have been omitted.)
Enìkòòkan rò pé ìyá òun níwà.
everyone think that mother his have.character
‘Everyone thinks that his mother is nice.’
Enìkòòkan rò pé ìyá rè kò níwà.
everyone think that mother his NEG have.character
‘Everyone thinks that his mother is mean.’
The nonlogophoric (‘‘weak’’) pronoun rè in (42b) can be understood to mean that the speaker has observed this, even for cases where none of the individuals who do or do not like the woman they know are aware that that woman is their mother (e.g., ‘Every orphan who meets a woman who he does not know is his mother finds that woman mean’; often the context must be provided before speakers accept that the anaphora is possible). Example (42a) with the ‘‘strong’’ pronoun òun, however, requires the interpretation that each thinker is aware that the woman he finds mean is in fact his mother. The rè cases show that shifting will never account for pronouns bound as variables where there is no logophoric reading,14 but we will return to these cases, for which Kratzer has made a different sort of suggestion.
The first observation to make is that the logophoric pronoun can appear in a clause that does not shift 1st person pronouns.
Olú so pé òun rí ìyá mi.
Olu say that he.LOG see mother me
‘Olu said that he saw my mother.’
For examples like (43), it would be necessary to appeal to the device that shifts all pronouns in its scope to a shifted 1st person reading without inducing 1st person morphology; instead, the device induces a special morphology without selective binding of pronouns in its scope. Once this is established, the interleaving example in (44) shows that the context-shifting account clearly doesn’t work for Yoruba logophoric pronouns.
Olú rò pé Ade so pé òun rí ìya òun.
Olu think that Ade say that he see mother his
‘Olu thinks that Ade said that he saw his mother.’
The strong pronouns must refer to Olu or Ade (not to someone else); but both can refer to Olu, both can refer to Ade, or one (either one) can refer to Olu while the other refers to Ade. This last possibility, the interleaving effect, is exactly what we expect if logophoric coconstrual is achieved by variable binding, since one operator can be within the scope of another without one scope overriding the other. Arguments of this kind can presumably be made for a number of other languages with logophoric pronouns.
The point I am making is not that the context-shifting account of fake indexicals is not a possible analysis of logophoricity in at least some languages. Rather, in those languages, we would expect to see different properties from those where shifting is employed, as Anand (2006) has shown.15
Kratzer (2009:218) leaves open whether or not there is interleaving in English with respect to fake indexicals, but I think the following context allows for interleaving of just the sort that would not be expected if all fake indexicals that are nonlocally A-bound were shifted. The bound pronouns in (45) are italicized, and italicized coconstrual relations are distinguished by whether or not they are boldfaced. Consider the following scenario: Janet has just married into a family she thinks her friends will not like. She is addressing her friends, warning them, in effect, not to say anything in front of her new husband about what they might think of his family; but she is doing so by trying to make them understand her situation as if from their own perspective.
(45) If I were any one of you and any one of you were me, then I would be careful to tell only me what I thought about my new in-laws.
If the coordinates on the context-setting head are all that can shift, how can there be two different shifted readings for 1st person within the same domain? It is certainly possible if the two substitution statements (if x were y) introduce operators, each binding different instances of D-bound, filling both with 1st person features. Whether other readings of (44) are possible is not important for the point being made here, which is this: only binding relations can distinguish the 1st person pronouns in a way that gives the intended reading. Thus, it must be possible for fake indexicals to be bound at a distance (hence not always shifted), and if so, D-bound can be bound nonlocally as well as locally, as long as its features are compatible with those of its A-binder.
5.2.3 The Insufficiency of D-Type Readings as a Substitute for Variable Binding
As noted in the last section, bound variable readings do not have to have logophoric interpretations, as illustrated in (46).
(46) Every hero like Oedipus believes that his mother is the perfect bride.
The whole point of the Oedipus story and others like it is that the hero is not aware that the person he marries is, in fact, his own mother; but in most such stories, the hero believes he has made the right choice when he marries. Cases like (46) are not, then, reported-speaker or de se readings of the pronouns, but they are bound readings nonetheless. Kratzer (2009:216) suggests that the apparent bound readings in these cases are achieved not by binding, but by an extension of the D-type reading originally assigned to donkey pronoun sentences, distinguished from normal variable binding by Evans (1980).
(47) Most men who own a donkey beat it.
Two key features of (47) are that it may be interpreted to mean that most of the men who own a donkey beat the donkey that they own, which means that it varies with choice of owner, and a donkey does not c-command it. The proposal made by Elbourne (2005) and others cited by Kratzer (2009) is that all pronominal variable-binding phenomena are actually cases where the pronoun is represented as a definite description with an elided restricting clause (= the donkey that he owns in (47)). I will avoid reproducing the details of this account here, but there is a notable difference between the cases where it has been employed in the absence of c-command and the cases where c-command appears to play a crucial role: namely, donkey sentence structures do not permit a bound variable reading when the antecedent is a universal.
(48) Most men who own every donkey (in their district) beat it/them.
No one accepts it in (48) with the bound reading, even though bound neuter pronouns in English are singular with every (e.g., Every donkey in the district loves its mother)—and to my ear, them is only marginally acceptable in (48). The failure of the universal with the singular, which is general (and well-known), is enough to show that the D-type reading has different requirements for agreement than variable binding in the more general case, which usually respects c-command.16
There is, however, another strong theory-internal reason to assume that A-bound pronouns bound as variables are not definite descriptions, but instances of D-bound.
If X can be a binder for D-bound in position Y and Y is not D-bound, then X and Y are not expected to be coconstrued (i.e., they are obviative).
There is no way to recover the force of Syntax-Induced Obviation if, apart from indexical pronouns, all pronouns are definite descriptions—that is, if some bound pronouns c-commanded by their antecedents are not instances of D-bound. If D-type readings are the same as those that determine coconstrual in (49), then the distribution of D-type readings cannot be correlated with the distribution of Principle C effects; for example, (49) does not require obviation between him and Bill.
(49) Every man who saw him hated Bill.
Thus, the D-type theory generalized to all bound variable readings would need a completely separate account of Principle C effects.17
5.3 Summary: The Case for Binding
We still need binding—specifically A-binding, as opposed to just local Agree—for at least the following four reasons:
Antecedent agreement must hold in clauses where subject-verb agreement also must hold, and yet the two must be distinct.
Special morphology (not 1st person) for distance-bound logophors does not follow from context shift, but must appear only where licensed by A-binding.
Special morphology for distance-bound logophors allows interleaving rather than scope override.
Nonlogophoric bound anaphora is not context shift, and appears incompatible with D-type interpretations.
Mediation by a head has been shown to contribute to additional conditions for feature compatibility between antecedent and D-bound, but antecedent agreement appears to be independently necessary to predict local morphology, as in the Icelandic case of binding by a dative antecedent. If the D-type account is not responsible for them, other cases of A-binding at a distance that require compatible features are common (e.g., Forany girlto tell Bill to marryherwould be counterproductive).
6 On the Sources of Phase-Internal Morphology for D-Bound
The proposal I make about D-bound means that there is only one anaphor in natural languages, which happens to have many morphological shapes. Closure for this term—that is, special morphology assigned within a phase—can (and does in most languages) distinguish interpretations where D-bound is A-bound at the first opportunity from interpretations where it is not.
However, one might ask why local anaphora should, in so many languages, be marked by special morphology, especially because there is no theoretical reason in this approach why there should have to be forms specialized for local anaphora. One might reasonably speculate that the manifestation of this sort of morphological within-phase distinction, expressible in the syntax, might be reinforced by functional pressure for systems where semantic binding at the most local opportunity is distinguished interpretively from semantic binding that requires a longer store. After all, as long as c-command is respected, A-binding is unbounded in this account. The prediction made by this theory in combination with my speculation about functional pressure is (51).
(51) Languages that lack local bindees distinct from pronouns are possible (and attested), but relatively rare.
This statement seems to square very well with the known crosslinguistic facts.
From this perspective, the prevalence of complex reflexives, particularly pronoun + stem, is a form of morphological conservation reacting to functional pressure for local anaphora resolution where possible. The manifestation of the form as complex is just the division of the sources of morphological information, so the agreement form associated with the local shape is expressed as a pronoun, and not a special one, but one generally found in the language, and perhaps a stem with a separate usage. The functional pressure to distinguish local interpretation is thus addressed without a special form of D-bound. D-bound is thus paired with an otherwise familiar paradigm (e.g., the φ-features of a body part reflexive are the same as those used when that morpheme literally denotes a body part). In some languages, these paradigms become lexicalized and the anaphor is fused (as in English), leading in many cases to grammaticalization in the direction of a simplex form. If we think about the historical trajectory in this way, there is no need to posit a grammaticalization stage where a fully invented nonce form is used to distinguish the local interpretation.
Moreover, the distribution of complex forms or formerly emphatic forms for local anaphora takes on a slightly different significance here than in some other proposals in the literature. König and Siemund (2000), for example, following a line of reasoning that has been pursued by others (see Gast 2006 for discussion and references), argue that complex reflexives derive from the grammaticalization of emphatic pronouns; that is, the form used to convey meanings like that of John did this himself is eventually lexicalized to effect reflexivity. However, it does not follow from positing an emphatic source that there is any pressure for locality of interpretation. The same form ought to be just as able to form anaphoric readings at a distance. From this perspective, we would expect that, in languages where the emphatic complex anaphor (used in contexts like John, himself, did this) can form nonemphatic reflexive readings, the same form can be bound at a distance by picking up a discourse-salient form.
However, in Lubukusu, a Narrow Bantu language of Kenya, the form AGR-eene (where AGR-agrees in noun class with its antecedent and -eene, in nonanaphoric usage, means ‘owner’) is used in emphatic contexts such as (52) but cannot take an antecedent within its phase, as in (53)—that is, where it would correspond to a direct object (though in those positions it can take a salient discourse antecedent—preferably, ‘the owner’).18 The local reflexive reading for a direct object is achieved by a reflexive marker (RFM) prefixed to the verb stem, as in (54), in which case AGR-eene can cooccur with the RFM postverbally. When AGR-eene cooccurs finally with the RFM, it fulfills a function different from that of an adjunct, as the contrast between (55) and (52) shows.
Wekesa o-mu-eene a-a-ch-a engo (*o-mu-eene).
Wekesa C1-C1-owner SM.C1-PST-go-FV home C1-C1-owner
‘Wekesa himself went home (*himself).’
(example ID3750 in the Afranaph Database)
Wekesa a-a-p-a o-mu-eene.
‘Wekesa beat him/*himself.’
Yohana a-a-i-yonak-a o-mu-eene.
‘John destroyed himself.’
Wekesa o-mu-eene a-i-siim-a o-mu-eene.
Wekesa C1-C1-own SM.C1-RFM-love-FVC1-C1-own
‘Wekesa himself loves himself.’
In the absence of the doubling prefix on the verb stem, AGR-eene must be interpreted as locally emphatic or as having an extraclausal antecedent, much like a pronoun, unless it is a prepositional object. Where it is a prepositional object, it contrasts with a simple pronoun in some cases with respect to whether the local subject can bind it, as illustrated by the contrast between (56a) and (56b). While (56a) allows either local anaphora or a reading that refers to some ‘owner’ in the context, the pronoun cannot have Yohana as its antecedent.
Yohana a-a-lom-a khu o-mu-eene.
Yohana SM.C1-PST-speak-FV about C1-C1-own
‘John spoke about himself/the owner.’
Yohana a-a-lom-a khu niye.
Yohana SM.C1-PST-speak-FV about PRON.C1
‘John spoke about him/*himself.’
It is also possible for AGR-eene to take a distant c-commanding antecedent or to pick out someone salient in the discourse (like the unbounded dependent forms discussed in Safir 2004b). The former possibility is illustrated in (57), where o-mu-eene can be anteceded by Jack.
Jack a-kanakan-a a-li Lisa a-many-il-e a-li o-mu-eene a-siim-a Alice.
Jack SM.C1-think-FVC1-that Lisa SM.C1-know-TNS-FVC1-that C1-C1-own SM.C1-like-FV Alice
‘Jack thinks that Lisa knows that he likes Alice.’
Thus, the AGR-eene form is complex, potentially locally anaphoric, potentially long-distance-bound, potentially emphatic, but never locally anaphoric on its own unless a preposition intervenes, probably because it introduces a phase boundary.
Although the Lubukusu data involve a number of issues and loose ends that are of interest, I present them here because Lubukusu AGR-eene would seem to show that there is no direct connection between the development of a complex emphatic form for anaphoric readings (associating its morphology with D-bound) and the use of the emphatic form exclusively for locally bound argument readings. The association of the complex form exclusively with local reflexivity appears to be an independent step, and the crease in the theory that permits it is the opportunity to associate the emphatic form with phase-internal binding morphology—a possible association, but not a necessary one. Given the robust use of the local RFM in Lubukusu, there is no pressure in that language for the complex form to be recruited for phase-internal anaphora (even though AGR-eene is licensed to appear in direct object position with a local interpretation just in case the RFM is also present.)
Various points of analysis need to be clarified. Both pronouns and AGR-eene are possible morphological realizations of D-bound in Lubukusu, where D-bound does not spell out as AGR-eene in the absence of the RFM locally, and is subject to discourse restrictions when it is bound at a distance. AGR-eene appears also to have a natural-born counterpart that does not require a sentence-internal binder, although its distribution is limited (it is most often associated with a subject or object marker on the verb when its antecedent is sentence-external).
The existence of forms like Lubukusu AGR-eene does not conform to the generalization that complex anaphoric forms are local, which is fine as a generalization (e.g., Faltz 1977, Pica 1986), but difficult to defend if it is to be enforced as a theoretically necessary result (see, e.g., Reuland 2011 for discussion and references). Simply claiming that AGR-eene is not anaphoric at all, in spite of the counterevidence presented, would make it difficult to formulate a counterexample to the generalization in question. (For rich details about the distribution of Lubukusu AGR-eene, see Safir and Sikuku 2011 and Sikuku 2011.)
From the perspective of the One True Anaphor account, it is unsurprising that D-bound should frequently have underspecified morphology, since its interpretation is not determined by its morphology, but vice versa (i.e., however it is bound). The account does not make any prediction, however, about whether underspecified morphology on D-bound would favor local or more distant antecedency. Icelandic sig (morphologically unspecified for number) is employed for logophoric interpretation in subjunctives (see Reuland and Sigurjónsdottir 1997, Safir 2004a), yet it is possible, and frequently the case, that underspecified forms are exploited as targets for phase-internal Agree, plausibly as a means of responding to the functional pressure for local anaphora resolution.
Nothing I say here affects theories of predicate reflexivity/reciprocity for any case where D-bound is not involved. Systems where local reflexivity does not use D-bound at all, if they exist, would deserve some discussion, but I am not sure that there are such cases, beyond inherently reflexive or inherently reciprocal predicates (but see Siloni 2012). Whether the reflexive affix of Lubukusu (ubiquitous in Narrow Bantu) is phase-internal morphology and/or a manipulation of the argument structure of the predicates to which it is attached is beyond our immediate concerns.19
The main point, however, is that D-bound, by its nature, is indifferent to locality, although UG can be exploited to distinguish local domains from more distant ones by identifying the morphological spell-out of D-bound with special phase-internal morphology.
This article can be seen as an elaboration and emendation of Kratzer 2009 that extends the exploration of the distribution of anaphora to a variety of patterns that must be handled differently, and, if I am right, not as well, in other approaches. The One True Anaphor hypothesis obliterates any syntactic or semantic distinction between anaphors that is not about the realization of their morphology (e.g., there is no Principle A) or the additional semantic conditions attached to them (e.g., reciprocity). All bound forms are D-bound where D-bound is possible; Principle B and Principle C effects are derived from contexts where D-bound is available for a binding relation but is not used for that relation (Syntax-Induced Obviation). Anaphors, local or distant, are bound the same way and by the same devices, as the evidence from ellipsis in English has shown. The morphology of D-bound may be influenced locally or nonlocally by the heads associated with the specifier positions that bind it; however, the morphology does not have to make this distinction, and in particular, binding does not reduce to phase-internal Agree. The effects of Principle A are reduced to conditions on phase-internal Spell-Out for bindees that the theory of grammar does not require. Functional pressure for local anaphora resolution can exploit the option provided by UG to recruit special morphology for phase-internal relations, but even in languages that make the distinction (which is most of them), A-binding is the same whether phase-internal morphology is manifested or not. Many new questions now arise as to whether the anaphoric phenomena that have been analyzed from other perspectives can be better explained (or at least accounted for) within the One True Anaphor approach, but I leave these explorations for future work.
This research was supported by NSF BCS-0919086 (for the Afranaph Project). I would like to thank the audience at the University of Maryland, where this work was first presented at Mayfest, 2011, and also Matt Barros, Carlo Linares, Naga Selvanathan, Edwin Williams, and Justine Sikuku, for useful discussion.
1 I will not enter into the rich literature on how to identify A-binders versus Ā-binders here. Instead, I will assume that all of the specifier (Spec) positions on the verbal spine from V to TP are A-positions, that certain possessor positions in nominals are A-positions, and that the landing sites of QR and wh-movement are Ā-positions. Any theory that separates A-from Ā-binders in this way is likely to be compatible with what I have to say in this article. Also see footnote 6.
2 I am assuming that designation here means the creation of the bound variable representation for SEM such that Spec,HP is the binder of D-bound and that this takes place at the moment the phase is capped by Merge of a head above the left edge of HP, as in (3c). This means I am assuming that there is phase-cyclic interpretation, as argued in Safir 2010, among other works.
3 A reviewer points out that argument anaphor reciprocals such as each other in English, which are treated here as [D-bound + RCP], would require a great deal of information about interpretation of the RCP feature to be achieved in the semantics. On the morphological side, the RCP feature will suffice to predict the local shape of the argument anaphor in the system as proposed here, but that local shape brings with it the interpretation imposed on D-bound by virtue of its RCP feature. The reviewer suggests that this is more like two true anaphors, since the RCP feature brings about a change in both shape and interpretation.
The source of reciprocal interpretation and its relation to outward morphology is a puzzle that I cannot hope to solve here, but it is worth pointing out that many languages use a single argument anaphor for both reflexive and reciprocal readings (e.g., Yoruba, Gungbe). A Yoruba example is given in (i) (diacritics are missing in this example, drawn from the Afranaph Database (Afranaph Project, ongoing) and retrievable with ‘‘simple search’’ for ID1219). For a similar Gungbe example, search for ID3583 in the Afranaph Database.
Won ri ara won.
they see body.of them
‘They saw themselves.’ or ‘They saw each other.’ or ‘They saw their bodies.’
It is not obvious how these interpretations are retrieved from the anaphoric relation formed by the (A-)binding of these argument anaphors (setting aside the literal interpretation). If the Yoruba form is actually two anaphors, a reflexive one and a reciprocal one, which accidentally share the same morphological shape, then it is necessary to add some mark to the anaphor if it is to be interpreted by the semantics as reciprocal. Whatever this marking is, a feature or a more elaborate property, it is the same marking that I am positing for English. Thus, the origin of reciprocal interpretation is no more mysterious in the theory presented here than it is for any other theory. At least in this theory, polysemy of this kind (reflexive-reciprocal argument anaphor) is expected if local-binder morphology on the one true anaphor is not articulated in a language. In Yoruba and Gungbe, the result is one morphological form for the local anaphor and two interpretations. If German sich, which can be reciprocal or reflexive, should count as an argument anaphor, as most suppose, then it is another example for which the same distinction must be posited.
In work in progress, I am exploring the hypothesis that a null operator in the verbal affix position determines local reciprocal interpretation even when an overt anaphor appears—in which case, whether the anaphor in argument position has reciprocal morphology or not is reduced to the theory of agreement (with the operator and antecedent) and the reciprocal anaphor makes no contribution to interpretation (even in English), other than to be spelled out as D-bound. This theory would simplify and support the One True Anaphor theory, if it turns out to be correct.
4 Some support for the view that the anaphor each other is idiomatic rather than composed is based on examples like (i), explored by Brasoveanu (forthcoming), in comparison with examples like (ii).
(i) The men told each girl a different story.
(ii) The men told each other a different story.
The preferred interpretation of (i) is one where each girl hears a story that the other girls do not hear, though it is also possible that the men are telling the same story to each girl (perhaps one at a time), but the story is one that is different from some discourse-salient story. There is no interpretation of (ii) that corresponds to a situation in which each man tells a story different from what the other men tell; rather, the story that is told does not vary across the men, but is different from some discourse-salient story. If each other were composed, we would expect the same ambiguity allowed by (i) to be found in (ii).
5 As Grodzinsky and Reinhart (1993) regard both pronouns and anaphors to be potential variables, their account makes no prediction concerning obviation for Principle B. For those cases, Reinhart and Reuland (1993) have a different account. The approach to Principle B taken here is distinct from theirs, as described in the text.
6 However, as mentioned in footnote 1, it is necessary to distinguish between A-binding and Ā-binding with respect to Spec,CP (as opposed to, say, Spec,IP), since Spec,CP does not count as a binder for D-bound; therefore, I still assume this part of the A-binding stipulation for the purposes of this article. In work in progress, for which Safir 2011 is an abstract, I seek to eliminate all reference to the A/Ā distinction in linguistic theory. I argue that quantified nominals are usually not DP specifiers at the point in the derivation where their scope is established. This distinction, I argue, interferes with syntactic binding of D-bound (which ultimately requires a DP antecedent), but does not influence binding of variables within the scope of the quantifier. I assume, however, that any theory that derives the A/Ā distinction will not result in a class of possible A-binders that is significantly different from those posited here.
8 Benjamin Bruening (pers. comm.) questions my interpretation of the facts in (21)–(22), arguing that the sloppy interpretation is always available. He provides several examples from the Internet, including (i), which convinces me that there are those who share his judgments.
John: My wife doesn’t understand me. Does yours, Bill?
Bill: I don’t know, John. I’ve never heard her even mention you.
The humor depends on the unexpected strict reading, so the sloppy reading must be presumed to be the likely one. I actually missed the joke on my first reading, but I was able to accommodate it to see what was intended. Other than appealing to accommodation, I have no account of the pattern of judgments Bruening documents. The judgments reported in the text that support my distinction between 3rd person, on the one hand, and 1st and 2nd, on the other, were tested many times, especially in lectures before what was at stake was revealed, since I consider them subtle.
9 For example, the recovery of the interpretation of VPe is dependent on parallelism that need not be local to VPa, as in the case of wh-extraction.
(i) *I don’t know what Mary thinks Bill saw, but I know what he did [VPe see].
(ii) I don’t know what Mary thinks Bill saw, but I know what she says he did [VPe see].
There are ways to improve examples like (i) (e.g., I accept I don’t know what Mary THINKS Bill saw, but I know what he ACTUALLY did); but in any case, these conditions on the right theory of parallelism are beyond the concerns of this article.
10 This approach differs from that of Reinhart (2006), who treats the strict versus sloppy contrast as a difference in the form of binding in VPa. On her account, only the sloppy reading is anteceded by a VPa with a bound pronoun; the strict reading is a mere coreference reading. That is, the status of the antecedent pronoun is different, and the difference permits the two readings. (See Safir 2008:342–343, for discussion.) I am assuming here that the difference between the strict and sloppy readings does not arise because the two readings have different possible binding configurations or indexing in their VP antecedents. Rather, I follow Dalrymple, Shieber, and Pereira (1991) and Williams (1995a), who argue that the content of VPe can be retrieved from VPa in two different ways. Examples of a type pointed out by Dahl (1972), Dalrymple, Shieber, and Pereira (1991), Fiengo and May (1994), and Fox (2000) show that a VPe retrieved as sloppy can in turn be the VPa foraVPe assigned a strict reading.
(i) Smithersi thinks that hisi job sucks. Homer does too. Homer’s wife, however, doesn’t.
As Fox (2000:117n8) remarks about (i), the first ellipsis (the Homer sentence) can be interpreted as sloppy, while the Homer’s wife sentence can be interpreted as strict (Homer’s wife thinks Homer’s job sucks). Dalrymple, Shieber, and Pereira (1991) and Williams (1995a,b) make different proposals to account for these ‘‘mixed’’ readings, but what is crucial for this article, especially in Williams’s approach, is that these authors conclude that strict and sloppy interpretations can both be retrieved from the bound variable relation in VPa. This reinforces the point made here that even if D-bound is licensed locally in VPa, its overt morphology does not block a strict interpretation for VPe.
12 I do not consider the phasal status of nodes inside nominals here—for example, whether the node D is a phase or whether there is some other node inside nominals that counts as one. Presumably, an assumption of this kind is necessary for examples like John’s attack on himself, where himself must take John as its antecedent, and him in place of himself would have to refer to someone other than John. A reviewer points out that, on the account argued for here, it must be the case that John is in the same phase as himself for himself to have the shape it does. For recent exploration of the role of phasal structure on binding in nominals, see Despić 2013.
13 A reviewer notes that the shape of anti-subject-oriented local anaphors, such as those found in Norwegian, is not predicted by whether or not D-bound has a binder within the same phase. This will be true of both subject-oriented and anti-subject-oriented local anaphors. It is possible to say that there is an additional shape condition that is sensitive to whether or not the antecedent is a subject in some languages, but this only describes the difference rather than deriving it, as done in Safir 2004b. Moreover, if an object is a binder for a prepositional object anaphor (e.g., John told Mary about herself ), the theory defended here must appeal to the presence of an additional head that has Mary in its Spec, because Mary must be treated as an A-binder. The issues are interesting, but too complex to explore here.
14Koopman and Sportiche (1989), reporting on logophoricity in Abe, specifically distinguish between bound variable pronouns that are logophoric and those that are not.
15 In Safir 2004a, I express doubt that the context-shifting account is ever warranted, and I show that the binding account appears to be necessary, in part on the basis of interleaving examples. By contrast, Anand (2006) shows that both analyses appear to be required, each producing a systematic pattern of facts. For purposes of the current argument, it is only necessary to show that the binding account is necessary, which establishes that there are instances where D-bound is A-bound at a distance.
16 Even for speakers like me who accept (48) with them, the agreement involved is different from that which holds in the canonical bound variable environments. It appears that D-type readings, in terms of both the antecedents they allow and the antecedent agreement relations they impose, are distinct from the usual bound variable phenomena in a way that does not support the view that both kinds of readings are achieved by the same device.
17 It appears that any theory that exploits a version of Reinhart’s Generalization in (12) will not be compatible with the generalized D-type account of variable binding.
18 In (52)–(57), tone marking is missing, although the Afranaph Database from which these examples are drawn may be updated to include tone information sometime after this article goes to press. The subject marker (SM) is generally thought of as a kind of subject-verb agreement, and the final vowel (FV) is usually -a, unless the sentence is subjunctive. All Bantu languages have a rich noun class system, where singular and plural are represented by different genders (agreement affix classes and paradigms). Singular for humans is class 1 (C1) and plural for humans is C2. In most contexts, Lubukusu typically has two class prefixes before nouns, the pre-prefix, which is usually the shape of agreement for that class, and a noun class prefix, which also identifies the class but normally is not the morpheme used to show agreement on the SM, a complementizer, or an adjective, for example.
19 For discussion of the issues distinguishing predicate-affix-based reflexivity and reciprocity, see Reinhart and Reuland 1993, Siloni 2012, and Safir 2013. For discussion of these issues in Lubukusu, see Safir and Sikuku 2011, Sikuku 2011, and Baker, Safir, and Sikuku, forthcoming. For a contrast to the latter analyses within Bantu, see, for example, Mchombo 2004. As noted in the text, the claims in this article are restricted to reflexive and reciprocal readings induced by argument anaphors (freestanding forms that appear to fill argument position slots in sentence structure).