Owing to different ideas about what counts as an anaphor subject to Condition A, two influential but superficially incompatible versions of Condition A of binding theory have coexisted: Chomsky’s (1986) version, and versions of predicate-based binding theories defended by Pollard and Sag (1992) and Reinhart and Reuland (1993) and modified in various ways since ( Pollard 2005, Reuland 2011). Using inanimate anaphors to independently control for sensitivity to Condition A without the confound of logophoricity, we show that Condition A must be checked at the syntax-interpretation interface and that Chomsky’s (1986) version (an anaphor must be bound within the smallest complete functional complex containing it and a possible binder) is nearly correct, with one amendment: a tensed TP boundary is opaque to the search for an antecedent. Given these results, we argue that Condition A should be reduced to phase theory and we outline how this can be done.
The focus of this article is Condition A of binding theory: how to formulate it and where the binding and locality requirements it imposes come from.
In recent years, there have been several attempts to implement Minimalist guidelines by eliminating Condition A of binding theory (see Drummond, Kush, and Hornstein 2011 for a review) as an independent condition—surely a desirable objective. Before such attempts can be made, however, the correct descriptive generalization for Condition A must be known. In fact, there is no consensus about this. As we will argue, a central confound on the way to determining the descriptively adequate generalization (which ultimately needs to be explained) concerns where, for a given anaphor, the boundary lies between conditions under which it is subject to Condition A and thus behaves like what we will call a plain anaphor, and conditions under which it is exempt from it. To see this, consider the following sentences:
John likes pictures of himself.
John showed Bill pictures of himself.
John showed Bill pictures of themselves.
While Chomsky (1986) assumed that the reflexives in (1a–b) are subject to Condition A, as early as Bouchard 1984 an argument was made that himself in all three sentences is not a ‘‘true anaphor’’—thus not subject to Condition A—but a pronoun. This was due to (in this case) split antecedents being allowed in (1c), the (reasonable) assumption being that allowed split antecedence in a syntactic position shows that this position is exempt from Condition A. To conclude on that basis, as is often the case, that the position occupied by himself must be an exempt position is unwarranted. What such cases show is that it merely can be. Indeed, it may be that himself in such a position behaves ambiguously, either as an exempt anaphor or as a plain anaphor subject to Condition A, the latter case being difficult to detect.1 This confound is pervasive (although not universal; see Fox and Nissenbaum 2004: sec. 3) and biases descriptive statements of Condition A (wrongly, as we will show); witness, for example, Drummond, Kush, and Hornstein’s (2011: 401) statement that ‘‘a reflexive within a picture noun phrase that is bound from outside its containing noun phrase is not a ‘true’ reflexive subject to principle A. . . . Rather, it is a pronominal with special logophoric requirements’’ and their footnote 16 citing major works sharing this assumption such as Pollard and Sag 1992 and Reinhart and Reuland 1993, to which we can add Safir 2004 and Reuland 2011, among others.
In this article, we present a detailed case study of the behavior of (some) anaphors in French, the only language we seriously discuss, in which we aim to control for the plain /exempt distinction. The picture emerging from this case study differs to various degrees from what has typically been assumed in major studies of anaphoric systems (e.g., Chomsky 1986, Pollard and Sag 1992, Reinhart and Reuland 1993, Safir 2004, Reuland 2011).
As far as French is concerned, we conclude that for the core case of Condition A, Chomsky’s (1986) descriptive generalization (an anaphor must be bound within the smallest complete functional complex containing it and a possible binder) is almost correct, with one amendment: a tensed TP boundary is opaque to the search for an antecedent. In particular, anaphors in picture noun phrases like those in (1) are subject (often, but crucially not always, vacuously) to Condition A, as Chomsky (1986) assumed.
Exploring further how these plain anaphors behave, we argue that the locality condition imposed on anaphor-antecedent relations by the core case of Condition A cannot be reduced to the requirement that an Agree relation hold at some point in the course of a derivation, contrary to many recent proposals (see Drummond, Kush, and Hornstein 2011, Reuland 2011, and references therein). Instead, at least for French, it can be and should be formulated in terms of phase theory, as others have argued for some other languages; we will outline how this could be done. More precisely, we will argue that Condition A (a) must be viewed as a syntax-semantics interface condition, and (b) is best stated as a requirement that an anaphor contained in some Spell-Out domain find its antecedent in that domain.
The article is organized as follows.
In section 1, we discuss what binding theory ought to account for in general terms. In particular, we recap why it is necessary to separate plain anaphoric behavior, subject to Condition A, from exempt anaphoric behavior, subject to different restrictions. We next propose that this could in principle be done by studying the difference (roughly) between inanimate anaphors, which we argue must (in French) be plain, and animate anaphors, which need not be.
In section 2, we establish that the descriptive content of Chomsky’s (1986) Condition A is (nearly) correct by examining the behavior of inanimate French anaphors. Given these conclusions, and those of section 4, new thinking is needed on how to derive the locality imposed by Condition A within the current framework (see section 5).
In section 3, we consider how plain and exempt anaphors are distinguished, an important question for the discussion in section 4. Given that inanimates are never exempt (in French), in this section we deal with inanimate anaphors and compare two ideas, concluding in favor of the first one: (a) plain and exempt anaphors are distinguished solely on the basis of the properties of their antecedents; (b) exempt anaphors are those lacking an (eligible) coargument, as proposed by Pollard and Sag (1992), Reinhart and Reuland (1993), Safir (2004), and Reuland (2011).
In section 4, we establish that plain anaphors must be exhaustively bound, while exempt anaphors need not be. Given that our characterization of plain anaphors is distinct from all its predecessors, this is a new result (such a conclusion was false in all previous versions of binding theory; see, e.g., Hicks 2009). That binding for Condition A differs from other types of binding (of exempt anaphors, bound pronouns, etc.) requires an explanation; we discuss this in the next section.
In section 5, we look at boundary conditions on how the results of sections 2 and 4 should be integrated into the grammar. We show why these results entail that the locality condition imposed by Condition A cannot be reduced to derivational Agree, why it should be seen as a syntax-interpretation interface condition formulated in terms of phase theory, and what properties a theory of plain-anaphor binding and phases must meet to be able to deal with plain-anaphorbinding locality.
In section 6, we conclude by briefly discussing unsettled general questions.
1 A Central Problem: Distinguishing Plain from Exempt Anaphors
Consider the following (Standard American) English examples:
The moon spins on itself.
*The moon influences people sensitive to itself.
The contrast here shows that an expression such as itself tolerates a local antecedent, (2a), but not a more distant antecedent, (2b). Call such distance-sensitive expressions plain anaphors.2 In a given language, binding theory seeks to answer the following questions, with hopes of finding crosslinguistically valid answers:
Which expressions are plain anaphors?
What makes an expression ( plainly) anaphoric?
What are the descriptive generalizations concerning the distribution of plain anaphors?
Where do these generalizations come from; that is, how should they be derived from theoretical primitives?
Here, we will not attempt to answer all these questions. We will try to identify a subset of plain anaphors in French, and we will address question 3 (in sections 2 and 4) and question 4 in part (in section 5).3
Clearly, answering question 1, at least partially, is a prerequisite to answering the others; if we do not know what the plain anaphors are, it is difficult to answer questions about them. The difficulty of answering question 1 is illustrated by the English paradigm in (3).4
John likes himself.
*John says that Mary likes himself.
John says that Mary likes everyone but himself.
Himself seems to be the kind of expression needing a local antecedent, as the contrast between (3a) and (3b) shows; yet (3c) is typically judged fine even though the very same element himself is involved, and by reasonable measures (depth of embedding) is further away from its antecedent than in (3b).5 This is why a distinction must be postulated between plain anaphors subject to Condition A and exempt anaphors not subject to it.6 But then, some way must be provided to separate plain instances of anaphors, such as himself, from other, exempt instances.7
Two influential but superficially incompatible types of approaches to characterizing this algorithm coexist: an antecedent-based proposal, compatible with Chomsky’s (1986) version of binding theory, which we call the classical (theory of) Condition A; and a position-based proposal defended by Pollard and Sag (1992) and Reinhart and Reuland (1993) and modified in various ways since (Safir 2004, Pollard 2005, Reuland 2011), which attributes a crucial role to the notion of coargumenthood. These two approaches make substantially different predictions regarding the distribution of plain anaphors. The latter approach contends that the set of positions allowing plain anaphors (roughly, cases in which the anaphor has a coargument) is disjoint from the set of positions allowing exempt anaphors (roughly, cases in which the anaphor does not have a coargument). The former makes no such claim.
To decide the issue, we propose to rely on the substantial amount of (descriptive) work done since these theories were first proposed. In particular, some crosslinguistic generalizations seem to hold widely of exempt anaphors:8 while exactly how exempt anaphora functions is not known (there are many perhaps not incompatible proposals regarding what is involved, such as logophoricity, perspective, point of view, empathy9), there is a wide and robust (albeit usually implicit10) crosslinguistic generalization, namely, that the referent of the antecedent of an exempt anaphor must in principle be capable of speech, of thought, of holding a perspective, of having a point of view, or of being an empathic target. While there are circumscribed exceptions (which, interestingly, appear to be culture-sensitive regarding sentience), this means that such referents must be live persons.11
This simple descriptive generalization regarding antecedents of exempt anaphors provides a possible angle, to our knowledge not exploited to this end before, for directly investigating what is not covered by the term exempt anaphora: if exempt anaphors must be animate, looking at the behavior of inanimate anaphors should reveal the conditions plain anaphors are subject to (e.g., Condition A).
In the next section, we look at the behavior of two French anaphoric expressions: elle-même ‘herself/itself ’ (lit. ‘her-same’, ‘her-even’) and possessor son ‘his/her/its’ as part of the expression son propre ‘his/her/its own’. We will show that each can behave either as a plain or as an exempt anaphor. We will also show that when inanimate, these French elements (a) are never exempt and (b) behave like anaphors subject to the classical Condition A.
2 Locality Conditions on Son Propre and Elle-Même
We now look at the behavior of two elements:
elle-même ‘herself/itself ’ (lit. ‘her-same’, ‘her-even’) and its paradigms (lui-même ‘himsame’, eux-mêmes ‘them-same’, etc.), and
son ‘his/her/its’ as part of the expression son propre ‘his/her/its own’ when it is understood as inducing focus alternatives on the possessor son (e.g., her own and not someone else’s)—henceforth possessor son propre.
We will (a) demonstrate the relevance of (in)animacy for locality, (b) show what the binding domain for such inanimate elements ought to be, and (c) demonstrate that (contrary to an assumption sometimes made; see, e.g., footnote 24) possessive and nonpossessive anaphors behave alike with respect to binding.
2.1 The Relevance of Animacy
The relevance of (in)animacy for binding locality in French can first be established by examining the behavior of possessor son propre ‘her own’.12 To guarantee this reading—that is, son propre roughly meaning ‘her own and not someone else’s’—the examples must be read in contexts that make alternatives to the possessor salient. In most cases, explicit alternatives to the possessor will occur in the sentence itself to make this possessor reading even more salient.
Consider the following contrast:13
[Ce pont]i dispose de soni ( propre) architecte.
‘[This bridge]i has itsi (own) architect.’
[Ce pont]i a l’air très fragile. Soni (*propre) architecte a reçu moins de moyens que les autres architectes de la région.
‘[This bridge]i looks very fragile. Itsi (*own) architect got less means than the other architects of the area.’
[Cet enfant]i a l’air très perturbé. Sai ( propre) mère passe moins de temps à la maison que les autres mères de la classe.
‘[This child]i looks very disturbed. Hisi (own) mother spends less time at home than the other mothers of the children in the class.’
This paradigm illustrates that inanimacy and locality of the antecedent correlate; that is, if the antecedent is inanimate, it must locally bind son propre. In (4a), inanimate ce pont ‘this bridge’ locally binds son propre and the sentence is grammatical. The telling contrast is found between the ill-formed (4b) and the well-formed (4c). In both cases, the antecedent of son propre is in a different clause but only the latter involves an animate antecedent.
The behavior of elle-même makes the same point. This element is not standardly described as a local anaphor. For instance, Zribi-Hertz (1995) assumes that elle-même is specific in that it is a bindable expression unspecified for locality and disjoint reference (which makes very weak predictions). However, the behavior of elle-même becomes unexceptional if we take inanimacy into account, as we did for son propre. Indeed, elle-même is also subject to locality if it is inanimate but not if it is animate, as illustrated by the following sentences using clausemateness:
[La Terre]i tourne autour d’ellei-*(même).
‘[The earth]i revolves around iti*(self ).’
[La Terre]i subit l’effet gravitationnel des nombreux satellites qui tournent autour d’ellei-(*même).
‘[The earth]i is subject to the gravitational effect of the numerous satellites that revolve around iti(*self ).’
Mariei subit l’influence des nombreux politiciens qui tournent autour d’ellei-(même).
‘Maryi is subject to the influence of the many politicians that revolve around heri (self ).’
Thus, the well-formed (5a) contrasts with the ill-formed (5b) because the inanimate antecedent la Terre ‘the earth’ is in the same proposition as elle-même in (5a) but not in (5b). When a nonclausemate antecedent is animate, as in (5c) (Marie), the sentence becomes (more) acceptable.
We conclude that inanimacy and locality correlate both for possessor son propre and for elle-même. These expressions thus seem to exhibit either plain or exempt behavior. Using inanimates, we can now investigate how locality should be defined.
2.2 Assessing C-Command
The first defining criterion for binding is c-command. As we now show, both inanimate son propre and elle-même must be c-commanded by their antecedents. First, consider the following contrast:
[Ce problème]i inclut sai ( propre) solution et celle du problème précédent.
‘[This problem]i includes itsi (own) solution and that of the previous problem.’
Les annexes de [ce problème]i incluent sai (*propre) solution et celle du problème précédent.
‘The appendices of [this problem]i include itsi (*own) solution and that of the previous problem.’
In (6a), the inanimate antecedent ce problème ‘this problem’ c-commands sa propre and the sentence is fully acceptable, as opposed to (6b), where the same antecedent does not c-command sa propre. As (7) makes clear, the relevant notion is indeed c-command and not subject orientation. Like (6), (7) exhibits a contrast with respect to c-command; but in this case, the antecedent appears in an object position instead of a subject position.
J’ai lavé [la fontaine]i avec sai ( propre) eau par souci d’économie.
‘I washed [the fountain]i with itsi (own) water out of concern for saving water.’
J’ai lavé les rebords de [la fontaine]i avec sai (*propre) eau par souci d’économie.
‘I washed the edges of [the fountain]i with itsi (*own) water out of concern for saving water.’
Here, sa propre is c-commanded by its antecedent la fontaine ‘the fountain’ occurring in the object position in (7a), but it is not in (7b), and the absence of c-command correlates with the ungrammaticality of sa propre.
Binding of inanimate son propre is not subject to intervention effects, whether with animates or inanimates.
[Ce problème]i amène l’étudiantk à sai ( propre) solution et à celle du problème précédent/à sak ( propre) solution et non pas à celle de son voisin.
‘[This problem]i leads the studentk to itsi (own) solution and to that of the previous problem/to hisk (own) solution and not to his neighbor’s.’
[Les fleuves]i éloignent les déchetsk de leursi/k ( propres) sources.
‘[The rivers]i carry wastek away from theiri /itsk (own) sources.’
In (8a), sa propre can be anteceded either by the inanimate ce problème ‘this problem’ or by the intervening animate l’étudiant ‘the student’. Similarly, in (8b), leurs propres can be bound either by les fleuves ‘the rivers’ or by the intervening c-commander les déchets ‘waste ( pl.)’.
All the same points can be made with inanimate elle-même. First, (9) illustrates the c-command requirement.
[La Terre]i tourne autour d’ellei-*(même).14
‘[The earth]i revolves around iti*(self ).’
Les satellites de [la Terre]i tournent autour d’ellei-(*même).
‘The satellites of [the earth]i revolve around iti(*self ).’
As with son propre, no subject orientation is involved, since the same holds when the antecedent occupies the object position instead of the subject position.
J’ai roulé [le tapis]i sur luii-*(même).
‘I rolled [the carpet]i on iti*(self ).’
J’ai roulé les bords [du tapis]i sur luii-(*même).
‘I rolled the edges of [the carpet]i on iti(*self ).’15
Inanimate elle-même is not subject to intervention either: as long as the antecedent is in the local domain of elle-même, other elements—whether animate or inanimate, singular or plural—can intervene between the two.
[La Lune]i attire [l’eau de la Terre] j vers ellei/j-même.
‘[The moon]i attracts [the earth’s water] j to itselfi/j .’
[La Lune]i attire les êtres humains/l’homme vers ellei-même.
‘[The moon]i attracts human beings/mankind to itselfi.’
In conclusion, inanimate son propre or elle-même must be syntactically bound, but need not be bound by the closest binder, nor is priority given to animate binders over inanimate ones. This will matter for examples showing intervention effects with subjects in section 2.3.
2.3 Calibrating Binding Domains
The antecedent of inanimate son propre must not only c-command it but must also occur in its local domain, which, as we will show for now (we will slightly revise this in section 5.1), can be characterized as the smallest XP with an intervening subject containing the anaphor. This generalization is based on sentences involving TPs, small clauses, and DPs.
First, the status of son propre differs in the following sentences depending on whether its antecedent occurs in the smallest TP containing it or not:
[Cette auberge]i fait de l’ombre à soni ( propre) jardin et au jardin de la maison voisine.
‘[This inn]i gives shade to itsi (own) garden and to the garden of the neighboring house.’
[Cette auberge]i bénéficie du fait que [TP soni (*propre) jardin est plus spacieux que celui des auberges voisines].
‘[This inn]i benefits from the fact that [TP itsi (*own) garden is more spacious than that of the neighboring inns].’
[Cette auberge]i bénéficie du fait que [TP les touristes préfèrent soni (*propre) jardin à ceux des auberges voisines].
‘[This inn]i benefits from the fact that [TP the tourists prefer itsi (*own) garden to those of the neighboring inns].’
In (12a), son propre and its antecedent cette auberge ‘this inn’ belong to the same TP and the sentence is natural. In (12b) and (12c), however, the antecedent cette auberge is the subject of the main clause while son propre occurs in the embedded clause (subject in (12b), object in (12c)), which means that the antecedent is outside the smallest TP containing son propre, and the sentence is degraded. This shows that the local binding domain must at most be the smallest tensed TP containing the anaphor.
This is so whatever type of proposition is involved. (12) involves a complement TP, while (13) and (14) involve adjunct TPs and the same contrast obtains; that is, the sentence is degraded when the antecedent does not occur in the smallest TP including son propre.
[Cette montagne]i est moins réputée pour soni ( propre) sommet que pour le sommet voisin auquel elle donne accès.
‘[This mountain]i is less renowned for itsi (own) summit than for the neighboring summit it gives access to.’
[Cette montagne]i attire beaucoup de gens parce que soni (*propre) sommet est l’un des sommets les plus escarpés du pays.
‘[This mountain]i attracts many people because itsi (*own) summit is one of the steepest summits in the country.’
[Ce problème]i inclut sai ( propre) solution et celle du problème précédent.
‘[This problem]i includes itsi (own) solution and that of the previous problem.’
[Ce problème]i présente peu de difficultés pour que les élèves puissent trouver sai (*propre) solution plus rapidement que celle des problèmes précédents.
‘[This problem]i presents few difficulties so that the students can find itsi (*own) solution more quickly than that of the previous problems.’
The same contrast obtains with nonfinite TPs.
[Cette défaite supplémentaire]i a entraîné sesi propres conséquences.
‘[This additional defeat]i entailed itsi own consequences.’
[Cette défaite supplémentaire]i a poussé les habitants à supporter sesi (*propres) conséquences en plus de celles de l’occupation.
‘[This additional defeat]i led the inhabitants to endure itsi (*own) consequences on top of those of the occupation.’
In (15b), ses propres occurs in an infinitival clause containing a subject (i.e., PRO controlled by les habitants ‘the inhabitants’) while the antecedent cette défaite supplémentaire ‘this additional defeat’ is the subject of the matrix clause (i.e., is outside of the infinitival clause). In this case, the sentence is degraded, as opposed to (15a) where ses propres and the antecedent occur in the same TP. Also, note that this is not due to an intervention effect caused by the animate les habitants ‘the inhabitants’, since (8) has already demonstrated that inanimate son propre is not subject to such intervention effects.
Similarly, the contrast in (16) shows that a small clause also constitutes a binding domain.
[Cette peinture]i possède sesi ( propres) composants.
‘[This paint]i includes itsi (own) components.’
[Cette peinture]i a rendu les ouvriers allergiques à sesi (*propres) composants et à ceux d’un autre type de peinture similaire.
‘[This paint]i made the workers allergic to itsi (*own) components and to those of another type of similar paint.’
In (16b), the subject of the small clause, les ouvriers ‘the workers’, intervenes between ses propres and the antecedent cette peinture ‘this paint’, the subject of the matrix clause. As a result, (16b) is degraded, unlike (16a), which does not involve a small clause.
Finally, the same holds if son propre is located in a DP with a subject (distinct from the anaphor).16
[Cette entreprise]i suscite l’admiration de soni ( propre) patron et la colère des patrons concurrents.
‘[This company]i arouses the admiration of itsi (own) manager and the anger of the competing managers.’
[Cette entreprise]i suscite l’admiration des employés pour soni (*propre) patron et leur colère contre les patrons concurrents.
‘[This company]i arouses the admiration of the employees for itsi (*own) manager and their anger against the competing managers.’
[Cette entreprise]i suscite votre admiration pour soni (*propre) patron et votre colère contre les patrons concurrents.
‘[This company]i arouses your admiration for itsi (*own) manager and your anger against the competing managers.’
In (17b) and (17c), son propre is part of a DP with a subject, and its antecedent is outside this DP; in other words, the subject of the DP, les employés ‘the employees’ in (17b) and votre ‘your’ in (17c), blocks the dependence between son propre and its antecedent, making the sentence unacceptable. This contrasts with (17a), where son propre appears in the same DP with no intervening subject.
The nonpossessive anaphor elle-même displays the same properties, suggesting that possessive and nonpossessive anaphors should be treated alike. (18) shows that inanimate elle-même does not license an antecedent outside the smallest tensed clause it occurs in, whatever the type of clause (complement clause in (18b) and (18c), adjunct clause in (18d) and (18e)) and whatever the position of elle-même (subject in (18b) and (18e), object in (18c) and (18d)).17
[La Terre]i tourne autour d’ellei-*(même).
‘[The earth]i revolves around iti*(self ).’
[La Terre]i pâtit du fait qu’ellei-(*même) n’a pas la priorité sur les hommes.
‘[The earth]i suffers from the fact that iti(*self ) does not get priority over humans.’
[La Terre]i subit le fait que de nombreux satellites tournent autour d’ellei-(*même).
‘[The earth]i suffers from the fact that many satellites revolve around iti(*self ).’
[La Terre]i connaît le phénomène des marées en partie parce que la Lune tourne autour d’ellei-(*même).
‘[The earth]i has tides partly because the moon revolves around iti(*self ).’
[La Terre]i est la seule planète bleue du système solaire parce que contrairement aux autres, ellei-(*même) est dotée d’une atmosphère comportant du dioxygène et est recouverte d’eau liquide.
‘[The earth]i is the only blue planet of the solar system because contrary to the others, iti(*self ) has an atmosphere containing dioxygen and is covered by liquid water.’
Similar judgments obtain if elle-même appears in an infinitival clause with its antecedent in the matrix clause. Thus, in (19b), the PRO subject of the nonfinite clause (controlled by les invités ‘the guests’) intervenes between le tapis ‘the carpet’ and elle-même; and in (19c), the subject les hommes ‘humans’ occurs between elle-même and the antecedent la Terre ‘the earth’. Since ellemême is not in principle subject to intervention, as illustrated in (19), this is a question of domain.
[Le tapis]i est enroulé sur luii-même.
‘[The carpet]i is rolled around itselfi.’
Du fait de sa beauté, [le tapis]i n’incite pas les invités à marcher sur luii-(*même), mais à coˆté.
‘Because of its beauty, [the carpet]i does not lead the guests to step on iti(*self ), but on the side.’
[La Terre]i ne peut pas rendre les hommes responsables d’ellei-(*même).
‘[The earth]i cannot make humans responsible for it(*self ).’
[Cette loi]i a provoqué [la colère des habitants contre ellei-(*même) et contre ses promoteurs].
‘[This law]i aroused the anger of the inhabitants against iti(*self ) and against its proponents.’
[Cette loi]i a provoqué [leur/notre colère contre ellei-(*même) et contre ses promoteurs].
‘[This law]i aroused their/our anger against iti(*self ) and against its proponents.’
[L’enceinte du château]i cache les habitants derrière ellei-(*même).18
‘[The wall of the castle]i hides the inhabitants behind iti(*self ).’
In both cases, the subject of the DP or the PP, les habitants ‘the inhabitants’, intervenes between elle-même and the antecedent, respectively cette loi ‘this law’ or l’enceinte du château ‘the wall of the castle’.
All these examples lead to the same conclusion as the sentences involving son propre: as we have illustrated using tensed TPs, infinitival TPs, APs, and DPs, the domain relevant for anaphoricity appears to be the smallest XP with an intervening subject containing the anaphor.
Descriptively, the behavior of French inanimate possessor son propre and elle-même is simple. They behave like anaphors subject to the classical Condition A locality restriction in the sense that they must be bound within a local domain corresponding to the smallest XP with an intervening subject containing them.
3 Meeting Locality, Defining Exemption
We have concluded that the classical Condition A is, at least for the French anaphors elle-même and son propre, (nearly—see section 5.1) descriptively correct. In this section, we discuss the impact of this finding on theories of the distribution of anaphors, particularly theories of the distribution of the plain/exempt distinction. Two related questions need to be answered: (a) Is Condition A sufficient to explain the distribution of anaphors? and (b) How are we to determine which instances of anaphoric expressions are plain and which are exempt?
3.1 Anaphor Type
Many binding theories distinguish between two types of anaphors subject to different conditions: in Reinhart and Reuland’s (1993) terminology, SELF anaphors (such as English itself ) are deemed to be subject to some version of ConditionA,while SE anaphors (such asDutch zich and Scandinavian sig) are often concluded to fall, not under Condition A, but (descriptively) under the Tensed-S Condition.19 To make sure that we are not comparing apples and oranges, we need to make reasonably sure that the French anaphors we are dealing with are of the SELF type.
There are good grounds to conclude that son propre and elle-même are run-of-the-mill SELF anaphors: a priori, because of their internal makeup; and a posteriori, because of their distribution. Indeed, it is extensively documented (see König and Siemund 2005) that, in language after language, affixing an intensifier or a focus particle to a pronoun turns it into a complex anaphor (e.g., him → himself ) of the SELF type. This is what is found in French with both expressions under consideration. The internal structure of son propre makes it similar to complex SELF anaphors in Reinhart and Reuland’s (1993) or Safir’s (2004) terminology: it is complex, as it comprises a pronoun (son, just like him in himself ) combined with another element (propre, like self in himself ) whose effect is to intensify or create focal alternatives to the denotation of this pronoun (see Charnavel 2012: chaps. 1–2, for detailed discussion). Similarly, elle-même comprises a pronoun elle and the focus particle/intensifier même (and elle-même, much like English himself, can be used as an intensifier). Thus, from the point of view of internal makeup, both conform to a well-attested structural schema for SELF anaphors; it is thus reasonable to conclude that the Condition A they must satisfy is the Condition A that constrains SELF anaphors quite generally.
3.2 Locality and Sorting Out the Plain/Exempt Distinction
3.2.1 The Coargument-Based Condition A
Apart from approaches based on Chomsky 1986, the most influential theories of Condition A are coargument-based theories (e.g., Pollard and Sag 1992, Reinhart and Reuland 1993, Safir 2004, Reuland 2011),20 and they come in a variety of flavors. They all share a core idea, which we may call the Coargument-Based Condition A (CBCA).
A SELF anaphor must be bound by an eligible syntactic coargument (eligibility varies from theory to theory).21 It is exempt if and only if it does not have such a coargument.
First, an anaphor is exempt when it is the single syntactic argument of a predicate, in particular in a DP or a PP. This is illustrated with English reflexives in examples like the following:
In (22), the anaphor herself is the only argument of picture; similarly, in (23), himself is said to be the single argument of the preposition over. In both cases, the anaphor is exempt from the CBCA and lawfully not anteceded by a coargument.
Second, an anaphor is exempt when it is part of an argument as, for example, in coordination.
Here, the anaphor is embedded in an argument: the complement of invited is Lucie and himself or a man like himself. As himself is analyzed as lacking a coargument, it is exempt from the CBCA.
3.2.2 The CBCA Is Too Strong, or Too Weak
The distribution of the French inanimate anaphors discussed here makes it clear that the CBCA is too strong. For example, when elle-même is the single inanimate argument of a predicate, the CBCA predicts it to be exempt and therefore, given our conclusions, to require an animate antecedent. But this is not the case; an inanimate antecedent is well-formed.
[Cette loi]i a entraîné la publication d’un livre sur ellei-même et sur son auteur.
‘[This law]i led to the publication of a book about itselfi and about its author.’
[La Grande Roue]i a éjecté les enfants au-dessus d’ellei-même.
‘[The Ferris wheel]i ejected the children above itselfi.’
Within a CBCA framework, such facts could be handled by assuming, as we do, that inanimate anaphors are never exempt. But this would now make the CBCA too weak, as such anaphors are subject to a local binding restriction illustrated by the following minimally modified sentences:
*[Cette loi]i est si importante que les journalistes prédisent la publication d’un livre sur ellei-même et sur son auteur.
‘*[This law]i is so important that the journalists predict the publication of a book about itselfi and about its author.’
*[La Grande Roue]i a été fermée après que des enfants ont été éjectés au-dessus d’ellei-même.
‘*[The Ferris wheel]i was closed after some children got ejected above itselfi.’
The sentences ( paralleling (23)) in which elle-même and the antecedent are clausemates are grammatical, but the sentences in which they are not in the same clause are degraded.
Within a CBCA framework, such facts may merely suggest that some additional condition is at play. Theories such as Pollard and Sag’s (1992), which do not include anything other than the CBCA, are thus insufficient. But proposals such as Reinhart and Reuland’s (1993), Safir’s (2004), and Reuland’s (2011) do include an additional condition: besides their CBCA, Reinhart and Reuland (1993) and Reuland (2011) posit a Chain Condition that roughly enforces the existence of a ‘‘chain’’ minimally containing the anaphor and its (closest) antecedent. This has the effect of requiring that the distance between them be short enough to allow (in principle) for an A-movement relationship (see Reuland 2011:sec. 3.6.2). And in addition to his version of the CBCA,23Safir (2004) posits a local antecedent-licensing condition.
Local Antecedent Licensing (LAL) An anaphor must be c-anteceded in domain D, where domain D for X is the minimal maximal extended projection containing X (where the verb may extend the projection of a P with a dependent complement). (Safir 2004:148, 150)
The fact that these theories need to add a condition to their Condition A raises two questions. First, the analogue to our Condition A is not the CBCA but the sum of these requirements (the CBCA plus the Chain Condition or the LAL); how then do they compare? Second, what role does the CBCA play in conjunction with these additional conditions: is it needed for locality? is it needed to separate plain anaphors from exempt anaphors?
Regarding the first question, it is clear that either sum is too strong: while the CBCA plus the Chain Condition and the CBCA plus the LAL correctly predict that inanimate possessor son propre must be bound locally, they wrongly disallow such cases as (27).
[Cette peinture]i révèle les propriétés de (la combinaison de . . . ) sesi ( propres) composants.
‘[This paint]i reveals the properties of (the combination of . . . ) itsi own components.’
Such examples (where we can recursively embed the pronoun arbitrarily deep) are well-formed (no subject intervenes) even though (a) the domain D for the anaphor under the LAL excludes its licit antecedent (situated outside of the minimal maximal extended projection containing ses, at most the direct object) and (b) the distance between ses—the possessor of the complement of (the complement of . . . ) the direct object—and its antecedent disallows an A-movement step.24
We conclude that our formulation of Condition A is empirically superior.
3.2.3 The CBCA Is Too Strong
Let us now turn to the second question regarding the role of the CBCA. Clearly, lacking an eligible coargument does not guarantee exempt status—witness French inanimate anaphors. But perhaps having an eligible coargument does guarantee plain status. We will show it does not.
In section 3.2.1, we merely reported how the notion of syntactic coargumenthood is applied in coargument-based theories. But first, it should be mentioned that syntactic coargumenthood is a complex notion not otherwise needed and not expressible in any simple way using the primitive notions allowed by Minimalist guidelines. Moreover, it is far from obvious how the notion of syntactic coargumenthood applies, given the development of syntactic theories. Thus, even for a subject–direct object pair of a verb, current analyses take them not to be syntactic (or semantic) coarguments, the direct object being an argument of V, the subject an argument of little v (or Voice). Note next that none of the cases discussed in sections 3.2.1 and 3.2.2 straightforwardly illustrates a lack of syntactic coargumenthood. In (24a), it is unclear why the conjunction and does not have two syntactic arguments, one of them a syntactic subject. In (24b), it is assumed that the element like is not an argument-taking predicate (with a subject). But the natural assumption is that like has a subject and that man like himself is a relative ( possibly head-internal, so that like has two arguments, himself and the subject man). Similarly, it is unclear why a preposition like over, a two-place predicate, does not have a (silent) subject, especially given examples like (20c) (see also footnote 18) showing that prepositions do project syntactic subjects.
Inversely, cases that must be claimed to involve coargumenthood with a subject to prevent exemption do so on dubious grounds. Thus, possessive constructions such as his picture of himself / him, where his is a nonagent possessor, are claimed by Pollard and Sag (1992) and Reinhart and Reuland (1993) to involve coargumenthood (whether syntactic or semantic) between himself and his. This is at best controversial; more plausibly, his is a coargument of picture of himself /him under a possessive/locative relation (see, e.g., Kayne 1993, Den Dikken 1998) and himself /him lacks a coargument. Similarly, a sentence such as Johnk seemed to himself to tk hurt himself, claimed to have a nonexempt reflexive in its main clause, requires taking John to be an argument and a subject of seem even though John is neither (it is the subject of T, not of seem).25 Appeal to syntactic coargumenthood to predict exemption thus appears to be based on shaky grounds.
Second, although it would be cumbersome to show this in detail, it should be reasonably clear that, in terms of a locality restriction, the Chain Condition or the LAL (or our Condition A) is more permissive than the CBCA. This is because movement or the LAL in principle allows dependencies between, say, a subject and a non-coargument position, even if this subject has a coargument (as, for example, in He will [t like soup], where he can move even though he and soup are coarguments, or in possessor-raising cases).
When a (nonsubject) anaphor A with a c-commanding argument B is involved, the Chain Condition and the LAL are equivalent to the CBCA. For the Chain Condition, this is because—in Minimalist terms—B will act as an intervener for an external probe trying to reach A. For the LAL, the minimal maximal extended projection containing A does not contain non-coarguments of A or B in the configuration under discussion (it may with prepositions).
Finally, a subject anaphor A c-commanding a coargument B is predicted to be ill-formed by CBCA approaches, but this prediction is incorrect (regardless of how eligibility is defined). This is shown, for example, by the well-known Chinese cases in which a plain SELF anaphor (ta-ziji, lit. ‘him-self ’) that is subject of a verb can take an antecedent outside of its clause (see Huang and Liu 2001). For French, it is shown by the examples in (17) (see also footnote 16). This means that either the CBCA is not needed in such cases, as it is redundant with the required second condition, or it is too strong to predict the behavior of inanimate anaphors.
In addition, the behavior of animate son propre and elle-même provides straightforward arguments that having an eligible coargument does not guarantee plain status. First, assuming (wrongly, it seems to us, but as needed by some versions of the CBCA26) that animate son and Jean in (28) are coarguments predicts that son cannot be coindexed with anything other than Jean. This is contradicted by the well-formedness of (28).27
Mariei a vendu soni propre portrait de Jeank.
‘Maryi sold heri own picture of Johnk.’
Mariei s’inquiète souvent du fait que ses enfants dépendent d’ellei-même.
‘Maryi is often worried that her children depend on herselfi.’
[L’avenir de Mariei]k ne dépend pas d’elle-mêmei, mais de ses parents.
‘[Maryi’s future]k does not depend on herselfi, but on her parents.’
Mariei s’inquiète souvent du fait que ses enfants se fient à ellei-même.
‘Maryi is often worried that her children trust herselfi.’
[Les voisins de Mariei]k ne se fient pas à elle-mêmei, mais à ses parents.
‘[Maryi’s neighbors]k do not trust herselfi, but her parents.’
Mariei pensait que ses enfants avaient honte d’ellei-même.
‘Maryi thought that her children were ashamed of herselfi.’
[Les parents de Mariei]k n’ont pas honte d’ellei-même, mais de ses amis.
‘[Maryi’s parents]k are not ashamed of herselfi, but of her friends.’
Mariei se demande si ses collaborateurs sont fiers d’ellei-même.
‘Maryi wonders if her coworkers are proud of herselfi.’
[Les collaborateurs de Mariei]k ne sont pas fiers d’ellei-même, mais de ses parents.
‘[Maryi’s coworkers]k are not proud of herselfi, but of her parents.’
In these examples, under the CBCA, elle-même must be considered a syntactic coargument of the subject of, for example, the verb dépendre de ‘depend on’, the verb se fier à, ‘trust’, or the adjective fier ‘proud’, as must all dependents of verbs or adjectives mediated by weak prepositions. Otherwise, the CBCA would wrongly predict (e.g., in English) that objects of weak prepositions (e.g., indirect objects, genitives, objects of on in depend on) should always be exempt. That such prepositions are weak is evidenced by the fact that they can disappear without meaning loss under derivational processes (e.g., se fier à ‘trust’ → fiable ‘trustworthy’), incorporation (e.g., dépendant de ‘dependent on’ → auto-dépendant ‘self-dependent’), and so on.
It would seem, then, that lacking an eligible coargument does not guarantee exempt status—witness French inanimate anaphors. And having an eligible coargument does not guarantee plain status—witness the animate anaphors above.28
3.2.4 Residual Coargumenthood Dependence?
Given the above discussion, the simplest conclusion we could draw is that coargumenthood in general and the CBCA in particular cannot be appealed to in order to constrain anaphora (at least in French). Concomitantly, the simplest algorithm separating plain from exempt anaphors is no algorithm at all beyond requiring what content exempt anaphors must have: they must have an antecedent with the right kind of intrinsic content and discourse role (the reason for this has not been established, but see Charnavel 2014 for a proposal). If such an antecedent is available, it can serve as antecedent without regard to the particular position occupied by the anaphor (there may of course be other constraints that a theory of exempt anaphora would have to deal with). If no such antecedent is available, the anaphor is plain and subject to Condition A.29
Adopting this null hypothesis looks incorrect, as shown by the following minimal pair:
Mariei s’inquiète souvent du fait que ses enfants dépendent d’ellei-même.
‘Maryi is often worried that her children depend on herselfi.’
*Mariei s’inquiète souvent du fait que ses enfants évitent ellei-même.
‘*Maryi is often worried that her children avoid herselfi.’
The ill-formedness of (33b) is unexpected, as the antecedent is identical in all relevant respects to that found in (33a), which obviously meets the relevant requirements for anteceding an exempt anaphor, whatever these requirements may be.
The following pair shows that the deviance of (33b) has nothing to do with anaphora:
*Jeani pense que Marie examinera luii-même.
‘*Johni thinks that Mary will examine himselfi.’
*Jeani examinera luii-même.
‘*Johni will examine himselfi.’
*Le ressorti contracte luii-même.
‘*The springi contracts itselfi.’
Indeed, even if lui-même is locally bound, the sentence remains unacceptable (regardless of animacy); in particular, requiring a coargument binder is not the relevant factor (as the contrast between the French examples and their English counterparts also shows; recall from footnote 15 that actual English judgments are not indicated). What, then, excludes French examples like (33b)?
We believe the answer to this question falls under a generalization extensively discussed and justified by Cardinaletti and Starke (1999). This generalization has very broad crosslinguistic relevance (Romance, Germanic, and Slavic are discussed; many other languages are mentioned) and is not category-specific (pronouns and adverbs are discussed). It is unrelated to binding theory, but can descriptively be stated as follows: all else (that is relevant) being equal, if a weaker form of the target element is available, it must be used (and thus blocks the use of a stronger form).30 In the present instance, all else being equal, an available pronominal clitic (reflexive, accusative, or dative) should block the use of bare elle(-même), and indeed the French data conform exactly to the prediction made by Cardinaletti and Starke’s proposal.
First of all, the unacceptability of lui-même (read without narrow focus or contrast on luimême) in simple clauses correlates with the acceptability of the reflexive clitic se in the following pairs:
*Jeani examinera luii-même.
‘*Johni will examine himselfi.’
‘Johni will examine himselfi.’
??Jeani décrit le paysage à luii-même.
‘??Johni describes the landscape to himselfi.’
Jeani sei décrit le paysage.
‘Johni describes the landscape to himselfi.’
Mariei dépend d’ellei-même.
‘Maryi depends on herselfi.’
*Mariei sei dépend.
‘*Maryi depends on herselfi.’
Marie a présenté Jeani à luii-même.
‘Mary introduced Johni to himselfi.’
Mariek s*i/k’est présenté Jeani.
‘Maryk introduced Johni to himself*i /herselfk.’
Jeani a été assigné à luii-même.
‘Johni has been assigned to himselfi.’
*Jeani si’a été assigné.
‘*Johni has been assigned to himselfi.’
Taking se to be the missing argument, it is well-known that it can only stand for a dative or an accusative object, it must be deep-subject-oriented, and it is incompatible with passive voice (see, e.g., Sportiche 2014 for recent discussion and references). As (35)–(39) illustrate, in all such simple-clause cases se and elle-même are in complementary distribution. Se can stand for a direct object with a subject antecedent (35a–b) or an indirect object with a subject antecedent (36a–b); in such cases, elle-même is excluded. Se is unavailable with other complements (37a–b), with an indirect object with a nonsubject antecedent (38a–b), or in the presence of passive voice (39a–b); in all such cases, elle-même is perfectly acceptable.
In such cases, elle-même is in competition with the pronominal clitic se; by Cardinaletti and Starke’s (1999) generalization, only if se is not allowed to occur is elle-même allowed by itself.32 This type of competition is not limited to anaphors such as elle-même: as Cardinaletti and Starke (1999) discuss in detail (see also Zribi-Hertz 2000 for related considerations in French), it is also found with strong pronouns such as lui ‘him’ (once again without contrastive or deictic accent on it) and their clitic counterparts such as le ‘him’.
*Jeani examinera luiq.
‘*Johni will examine himq.’
‘Johni will examine himq.’
??Jeani décrit le paysage à luiq.
‘??Johni describes the landscape to himq.’
Jeani luiq décrit le paysage.
‘Johni describes the landscape to himq.’
Mariei dépend d’elleq.33
‘Maryi depends on herq.’
*Mariei luiq dépend.
‘*Maryi depends on herq.’
*Mariek a présenté Jeani à luip.
‘*Maryk introduced Johni to himp.’
Mariek luip a présenté Jeani.
‘Maryk introduced Johni to himp.’
*Jeani a été assigné à luip.
‘*Johni has been assigned to himp.’
Jeani luip a été assigné.
‘Johni has been assigned to himp.’
This pattern of competition is duplicated exactly with exempt-anaphor cases where elle-même has an antecedent in a different clause.34
*Jeani pense que Marie examinera luii-même.
‘*Johni thinks that Mary will examine himselfi.’
Jeani pense que Marie li’examinera.
‘Johni thinks that Mary will examine himi.’
??Jeani pense que Marie décrit le paysage à luii-même.
‘??Johni thinks that Mary describes the landscape to himselfi.’
Jeani pense que Marie luii décrit le paysage.
‘Johni thinks that Mary describes the landscape to himi.’
Mariei s’inquiète du fait que ses enfants dépendent d’ellei-même.
‘Maryi is worried that her children depend on herselfi.’
*Mariei s’inquiète du fait que ses enfants lai dépendent.
‘*Maryi is worried that her children depend on heri.’
??Jeani pense que Marie a présenté Suzanne à luii-même.
‘??Johni thinks that Mary introduced Susan to himselfi.’
Jeani pense que Marie luii a présenté Suzanne.
‘Johni thinks that Mary introduced Susan to himi.’
??Jeani pense que Marie a été assignée à luii-même.
‘??Johni thinks that Mary has been assigned to himselfi.’
Jeani pense que Marie luii a été assignée.
‘Johni thinks that Mary has been assigned to himi.’
Note that this pattern cuts across the direct object/prepositional object distinction: PP objects with a weak preposition (e.g., indirect objects) that cliticize as pronominal clitics are excluded.
The pattern tracks the availability of pronominal clitics in the nonverbal domain too. Thus, all of the following sentences are fine since the pronouns cannot be replaced by a weak pronominal clitic:
Jeani est fier de luii /luii-même.
‘Johni is proud of himselfi.’
Jeani a vendu des photos de luii /luii-même.
‘Johni sold pictures of himselfi.’
Finally, note that—as expected under Cardinaletti and Starke’s view—all the excluded examples with either elle or elle-même improve if the pronouns are deictic or focused (and thus not in competition with weak forms, which are neither).
(?)Jeani pense que Marie examinera LUIi-(MÊ ME).
‘Johni thinks that Mary will examine HIMi(SELF).’
We conclude that nothing about anaphora is involved in excluding examples such as (33b) and that coargumenthood is also not involved.
3.2.5 Extension to English?
Obviously, it would be desirable to extend our conclusions to English (and all other languages), but we are not in a position to do so simply because not enough is known. We briefly outline the reasons here (as a full discussion is beyond the scope of this article).
First, poorly understood variation is reported among English dialects. Thus, example (3b) (the structural counterpart of the French example (33b)) is standardly reported to be robustly illformed in American English (the picture we gathered from our American students is actually much more nuanced) but well-formed in British English (see Zribi-Hertz 1989, Pollard 2005). This may be related to the second point.
Second, according to Reinhart and Reuland (1993), narrow-focus accent on the anaphor in (3b) makes the sentence well-formed; this is surprisingly similar to the French fact in (51). This may be related to the third point.
Third, as Ahn (2012, 2014) discusses extensively, in English, locally bound subject-oriented reflexives (that are in a possible movement relationship with the subject position)35 and pronouns in the same positions cannot bear normal phrasal accent and must instead be deaccented. Thus, John cut himself/him with normal stress on the reflexive or pronoun cannot answer a question such as What happened? This suggests that English may after all have the weak/strong distinctions relevant for Cardinaletti and Starke’s (1999) generalization, thus interfering with anaphoric dependencies in the same way as in French. Ahn (2012, 2014) shows that deaccented reflexives in English (which must be subject-oriented) should be analyzed exactly like French reflexive se. This would immediately explain why (subject-oriented) object reflexives cannot be exempt (the same competition as in French), except when the reflexive is narrowly focused (as in the French cases in (51)).36 This is also related to the fourth point.
Fourth, non-subject-oriented reflexives can be stressed under normal phrasal accent (What happened? can be answered by John showed Mary herself, with main stress on the reflexive). In addition, they seem to be able to be exempt in ways subject-oriented reflexives cannot. Thus, Ahn (2012, 2014) reports the following contrastive judgments:
i. Johni showed Mary himselfi.
ii. *Billi thinks Johnj showed you themselvesi+j .
i. Johni showed Billj themselvesi+j .
ii. Johni thinks you showed Billj themselvesi+j .
As there are good grounds (see section 4.2) to conclude that the possibility of split antecedence is a diagnostic for exempt status, these sentences run counter to what has been standardly assumed regarding exemptability, and they show both that coargumenthood simply cannot be at play in defining nonexempt status and that a better overall assessment of the English facts is needed.
4 When Is Inclusive Reference Possible? (Corroborating the Plain/Exempt Dichotomy)
Lebeaux (1984) discusses what have become classical diagnostics for relations between plain anaphors and their antecedents: c-command, split antecedence, strict/sloppy readings in ellipsis, and pronoun/anaphor complementarity. As Hicks (2009:135–166) shows, none of these diagnostics clearly separates (what he and many others assume to be) plain and exempt anaphors (within a CBCA approach). But it is important to know what properties plain (or exempt) anaphors must have, in order to properly constrain theories of their behavior. In this section, we try to settle some of these questions for French by controlling for the plain/exempt distinction.37 In particular, we discuss two questions:
Has independent evidence been convincingly adduced corroborating the dichotomy between plain and exempt anaphors defended by CBCA theories?
Is there in fact independent evidence corroborating the dichotomy between plain and exempt anaphors as we define it?
4.1 Correlation Failures in CBCA Theories of the Plain/Exempt Dichotomy
Let us ask what the plain/exempt dichotomy correlates with in CBCA theories. Different versions make different claims. For Pollard and Sag (1992), positions allowing plain anaphors disallow pronouns with the same antecedents. For Reinhart and Reuland (1993), plain anaphors must be semantically bound (i.e., must be interpreted as bound variables) while exempt anaphors need not be. Neither of these correlations holds, as we now show.
First, there is no overall complementarity between plain anaphors and pronouns. This can be shown in two ways, both of them well-known. One would expect that in a context in which a pronoun has coarguments and cannot take one of them as antecedent, a plain anaphor with that antecedent should be allowed. But this is incorrect, as (53) shows.
(53) *Johni and Mary like himi /himselfi.
Here, neither him nor himself (which is a plain anaphor in the CBCA view, having the subject as coargument) is allowed with John as antecedent. This suggests that syntactic coargumenthood is not sufficient to explain the joint distribution of pronouns and plain anaphors.38
Conversely, there are simple cases in which plain anaphors occur in the same position as pronouns (and with the same antecedent) as per the CBCA.39
Jeani parle de luii /luii-même.
‘Johni is speaking of himi /himselfi.’
Jeani est fier de luii /luii-même.
‘Johni is proud of himi /himselfi.’
Here, both the pronoun and the anaphor are allowed, with no necessary focus difference.40
Similarly, unlike what Reinhart and Reuland (1993:673ff.) claim, it is unclear whether plain anaphors must be semantically bound, that is, interpreted as bound variables. Thus, both of the following English sentences allow strict and sloppy readings, (55) readily (see, e.g., Hestvik 1995, Kehler 2005), (56) for many speakers (see, e.g., Büring 2005:141):
4.2 Inclusive Reference
We now show that the plain/exempt dichotomy we propose does correlate with the (im)possibility of inclusive reference, unlike the dichotomy CBCA theories propose.
Inclusive reference refers to cases of nonexhaustive binding where the reference of an anaphor is strictly included in ( partial binding) or strictly includes (split antecedence) the reference of an antecedent (see, e.g., Lasnik 1989, Den Dikken, Lipták, and Zvolenszky 2001). Exempt anaphors are typically assumed (correctly so in our view) to allow nonexhaustive binding. Except by Hicks (2009), it has also long been assumed that syntactic binding imposed by (any version of) Condition A is interpreted as referential identity,42 either through coreference or through semantic binding (the choice being regulated by Grodzinsky and Reinhart’s (1993) Rule I, or some descendant of it; see, e.g., Roelofsen 2010).
Inclusive reference is a particularly interesting criterion because there is no reason, other than formal ones, why it should be allowed or disallowed. In particular, constraints on what can act as the antecedent of a plain anaphor should be irrelevant since they impose no interpretive constraints on the content of this antecedent.
The most interesting cases are ones on which theories differ regarding the plain/exempt dichotomy. The parameters of the problem are (a) whether an anaphor is predicted to be exempt either by CBCA theories or by our approach based on the classical Condition A (four possibilities); (b) two types of inclusive reference; (c) the choice of animacy ((in)animate) and type of position ((non)-coargumental) for son propre and elle-même. Given that both expressions behave the same and that both types of inclusive reference also behave the same, the findings are as shown in table 1.
|.||Exempt (by CBCA theories .||Exempt (by classical Condition A) .||Inclusive reference observed .||Examples .|
|2||Yes||No||Not Possible||Non-coargumental animates|
|4||No||No||Not Possible||Coargumental animates|
|.||Exempt (by CBCA theories .||Exempt (by classical Condition A) .||Inclusive reference observed .||Examples .|
|2||Yes||No||Not Possible||Non-coargumental animates|
|4||No||No||Not Possible||Coargumental animates|
The following pattern shows that both hypotheses are right in uncontroversial cases of the plain/exempt distinction; that is, it illustrates lines 1 and 4 of table 1:
Jeani a dit à Pierrem que personne d’autre qu’euxi+m-mêmes ne devrait faire ça.
‘Johni told Peterm that no one but themselvesi+m should do this.’
Ce décreti détourne la loim de luii-même/d’ellem-même/*d’euxi+m-mêmes.
‘This decreei diverts the lawm from itselfi /*from themselvesi+m.’
[Jeani et Marie]m pensaient que personne d’autre que luii-même ne devrait faire ça.
‘[Johni and Mary]m thought that no one but himselfi should do this.’
[La Terre et le Soleili]m dépendent d’euxm-mêmes/*de luii-même pour leur énergie.
‘[The earth and the suni]m rely on themselvesm/*on itselfi for their energy.’
(57a) and (57b) are instances of split antecedence: the reference of eux-mêmes ‘themselves’ is the sum of the reference of the two antecedents Jean and Pierre in (57a) and ce décret ‘this decree’ and la loi ‘the law’ in (57b). Thus, eux-mêmes is not coindexed with any syntactic coargument in (57a). Since it does not have any eligible coargument (it is a subpart of an argument of faire ‘to do’), eux-mêmes is exempt under CBCA theories; and it is exempt under our theory, too, since it is animate. In (57b), eux-mêmes is inanimate. Therefore, it cannot be exempt (under any theory: for CBCA theories because it is in (verbal) coargument position; for our theory because it is inanimate). However, it is not locally bound as required by Condition A or by CBCA theories. Exactly the same obtains, mutatis mutandis, for (57c) and (57d) with partial binding.
For controversial cases (lines 2 and 3 of table 1), the possibility of strict inclusive reference correlates with our version of (non)exemption: inanimates require exhaustive binding; animates, which can freely be exempt, do not. We illustrate this first for inanimate son propre when it is the only argument of the nominal predicate it combines with.
In all of (58a–c), involving inclusive reference, son propre is predicted to be exempt by the coargument view since it has no coargument. But in none of them is partial reference (as in (58a)) or split antecedence (as in (58b) and (58c)) acceptable. Once again, the animacy dimension plays a crucial role. If we modify the sentences in (58) to make the anaphor animate, inclusive reference becomes possible.
The same pattern is found with elle-même. In each of the following cases, illustrating what we take to be nonexempt positions, an inanimate cannot be partially bound, (60a), or allow split antecedents, (60b):
But selecting animate antecedents makes such cases fine.
In sum, exempt anaphors need not be exhaustively bound by a unique antecedent, but plain anaphors indeed must be; this should follow from the right theory of Condition A (see section 5.4.3).
5 Fine-Tuning Condition A and Deriving It
The previous argumentation concludes that the classical Condition A is basically the correct generalization (in French) regulating the distribution of plain anaphors. In section 5.1, we turn to the question of how precisely to formulate Condition A, concluding that Chomsky’s (1986) formulation should be amended to disallow a plain anaphor from taking an antecedent outside a tensed TP containing it. In section 5.2, after examining the properties of the binding relations between plain anaphors and their antecedents, we will conclude that the best strategy for attempting to reduce Condition A to more primitive conditions is to reduce it to phase theory, but not via binding as a derivationally set up Agree relation as many have proposed (see, e.g., Hicks 2009); we will then examine what this entails for phase theory.
5.1 Fine-Tuning Condition A
Taking into account the plain/exempt distinction, we only deal with plain anaphors in this section, which we systematically illustrate with inanimate anaphors.
We can paraphrase Chomsky’s (1986) formulation of Condition A as saying that ‘‘a plain anaphor must be bound within the smallest complete functional complex containing a structural binder for the anaphor.’’ A complete functional complex is understood to be some phrasal projection containing only saturated predicates (i.e., predicates with all of their arguments). Given the Predicate-Internal Subject Hypothesis and trace theory, a complete functional complex for some predicate p is simply the maximal projection of p. Given furthermore that the binding domain of an anaphor must contain a binder for the anaphor, this formulation is equivalent to requiring that ‘‘a plain anaphor must be bound within the smallest XP containing a structural binder for the anaphor.’’
It is easy to see that if the anaphor is not the highest, or included in the highest, projected argument of some head X, XP, the maximal projection of X will be its binding domain. By definition, the subject of an XP is the highest argument of that XP (if this XP has a subject). It follows that if an XP with a plain anaphor has a subject that is not or does not contain it, this plain anaphor in XP will have to be bound within this XP. In other words, as we have been assuming so far and as we have demonstrated to be the case in all instances we have discussed, a subject cannot intervene between a plain anaphor and its antecedent. If however the plain anaphor is the highest, or part of the highest, projected argument of some head X, XP, the maximal projection of X will not be its binding domain, since XP will not, by assumption, contain a binder for the anaphor. In that case, the binding domain will be the smallest YP containing XP and a binder for the anaphor. All of this comes down to requiring that an anaphor and its antecedent be in the smallest XP containing both without a subject intervening between them.
In (62), we schematize the major pattern predicted for plain anaphors. We illustrate with English sentences, and with reflexive anaphors where possible (reciprocals otherwise). (We ignore exemption here for simplicity of exposition, therefore using animates.) (ECM = exceptional Case marking)
We have discussed most of the corresponding cases in French with inanimate—that is, plain— anaphors. They conform exactly to this pattern with two exceptions: cases like (62e) and (62h), discussed for elle-même in section 3.2.4 and involving (as accusatives) competition with weak forms, and cases like (62j) and (62k), to which we now turn.
In Chomsky 1986, (62j) and (62k) with m indices are both predicted to be fine. Since (62j) is actually ill-formed, a special mechanism is needed to exclude it.43 (62k), however, is fine with the indexing indicated. But we do not know whether this is a case of exempt anaphora or plain anaphora (since these examples involve animates). Constructing comparable examples in French with plain anaphors (inanimates) yields deviant sentences.
[Cette auberge]i bénéficie du fait que [TP soni (*propre) jardin est plus spacieux que celui des auberges voisines].
‘[This inn]i benefits from the fact that [TP itsi (*own) garden is more spacious than that of the neighboring inns].’
[Ce musée]i indique que [TP l’équipe de soni (*propre) conservateur collabore avec d’autres conservateurs de musée].
‘[This museum]i indicates that [TP the team of itsi (*own) curator collaborates with other museum curators].’
[La Terre]i pâtit du fait qu’[TP ellei-(*même) n’a pas la priorité sur les hommes].
‘[The earth]i suffers from the fact that [TP iti(*self ) does not get priority over humans].’
*[La Terre]i a bénéficié du fait que [TP des photos d’ellei-même et de son satellite ont montré les effets néfastes de la pollution].
‘*[The earth]i benefited from the fact that [TP pictures of itselfi and of its satellite showed the harmful effects of pollution].’
In other words, a French plain anaphor in a tensed TP disallows a (closest) antecedent located outside of this TP (as, in fact, widely assumed since Bouchard 1984). This prohibition is not a blanket prohibition against anaphors, as exempt anaphors are of course allowed. This is illustrated by the following examples with elle-même in or inside the subject position:
Au début, c’est sa belle-soeur qui va venir prendre ses enfants, puis [TP elle-même viendra].
‘In the beginning, it is her sister-in-law who will pick up her children, then [TP she (lit. herself ) will come].’
[Marie]i pâtit du fait qu’[TP ellei-(même) n’a pas la priorité sur les hommes].
‘[Mary]i suffers from the fact that [TP heri(self ) does not get priority over men].’
[Marie]i a bénéficié du fait que [TP des photos d’ellei-(même) et de son frère ont été publiées].
‘[Mary]i benefited from the fact that [TP pictures of heri(self ) and of her brother were published].’
We conclude that the classical Condition A must be amended as follows:
A plain anaphor and its binder must be in the smallest XP containing both without an intervening subject and no larger than a tensed TP.
Although it differs from previous versions, this version is not entirely surprising when seen from the perspective of the history of binding theory. It is important to remember, however, that given the plain/exempt-anaphor confound we have discussed, we believe that previously there was no reliable way to establish that this version was indeed the correct one.
5.2 Are Anaphor Binding Properties Due to Agree?
We now turn to the question of how this condition should be encoded in the grammar. Such an encoding should derive the c-command requirement, the locality requirement, and the exhaustive binding requirement discussed in section 4.2.
Condition A as formulated in (65) imposes a locality requirement on the binding of plain anaphors. Proceeding parsimoniously, we should attempt to reduce this requirement to existing, independently motivated requirements. Current theories (Chomsky 2001, 2008) contain exactly two hypotheses imposing locality effects:
Closest Attract (or probe-goal relations or Agree)
Phase theory (because of the Phase Impenetrability Condition) imposes absolute locality requirements: an element inside the Spell-Out domain of a phase head cannot be syntactically directly related to an element external to this phase. Closest Attract is more restrictive, as it must operate within a phase, but it also imposes relative locality requirements. It can only relate elements that are closest to each other, and in particular it is sensitive to intervention effects: a probe cannot see a goal across an intervening potential goal (when they are all within a single phase, of course). See Zeijlstra 2012 for a survey of approaches to Agree.
Many attempts to reduce Condition A to more primitive theoretical constructs (e.g., Hicks 2009, Reuland 2011, Rooryck and Vanden Wyngaerd 2011) construe binding relations involved in Condition A as versions of an Agree relation and thus reduce it to the constraints on probegoal relations. Roughly, this is done by postulating that anaphors have underspecified features, which must be valued and thus require them to enter into an Agree relation in the course of the derivation.
Invoking Agree as the binding mechanism for anaphors is appealing. Indeed, invoking Agree does not explain the c-command binding requirement but at least reduces it to something else: Agree imposes locality since it must operate within a phase. Agree also seems to encode the exhaustivity of binding, as there typically is a complete sharing of features between a probe and a goal.
However, we now provide reasons why such a simple Agree approach is at best insufficient, at worst inadequate (for French). In a nutshell, here are the reasons, which we address in turn and which (as we will show) impose boundary conditions on the right account, whatever it is:
There is no principled reason why Agree should be involved.
There is no principled reason why Agree should entail binding.
There is no principled reason why Agree should entail exhaustive binding.
Agree is (normally) subject to intervention effects. Anaphor binding is not.
The locality imposed by Agree is too permissive.
Agree is a derivational constraint but Condition A must hold at LF.
5.2.1 Required Binding, Exhaustive Binding, and Locality
In Agree approaches, the fact that plain anaphors must be bound is attributed to the fact that they are underspecified (an idea with various instantiations) and must thus be paired with an antecedent.
A first challenge is to derive the fact that plain anaphors have underspecified features requiring an Agree relation. Failing to do this amounts to stipulating that plain anaphors are plain. This is not a simple challenge to meet. The French inanimate pairs son/son propre and elle/elle-même illustrate the difficulty. The pronoun son must be sufficiently specified not to have to enter an Agree relation (it needs no structural antecedent); it thus becomes necessary to explain why adding propre with focal stress on it yields possessor son propre, where son must be underspecified (the same would have to hold of the pairs elle/elle-même and him/himself ). We know of no plausible way to achieve this result, especially in light of Charnavel’s (2011, 2013) detailed argument that propre is just what it looks like: an adjective meaning ‘specific to’ with focal stress in the ‘‘possessor’’ son propre we are discussing here. This difficulty is made especially salient given that adding propre with focal stress on the head noun does not yield an anaphoric behavior (see footnote 12 and references therein).
Granting that plain anaphors must be bound, a second challenge, noted by Drummond, Kush, and Hornstein (2011), arises from the fact that an Agree relation quite generally (DP/T, DP/participle, etc.) says nothing about reference: it is a formal procedure of featural agreement. Some additional stipulation granting an interpretation to the Agree relation is needed in case two (referential) DPs are involved.44
A third challenge is to derive why plain anaphors do not tolerate partial binding or split antecedents. Is this explained by an Agree approach? The answer is negative, for two reasons.
The first reason is related to the second challenge. Assume an Agree relation between a single plural plain-anaphor goal G and a single plural probe antecedent P. Nothing prevents a partial binding interpretation of this relation (where P merely overlaps in reference with G: A, B, and C saw themselves could mean A and B saw A, B, and C or A, B, and C saw B, C, and D, etc.). Such an interpretation must be excluded by stipulating how the formal relation is interpreted. Let us assume such a stipulation. Partial or split binding would now be excluded under the standard assumption that Agree pairs up a single goal with a single probe.
The second reason is that there is no independent reason why an Agree relation must be so restricted. First, note that multiple probing by the same probe must be allowed—since two plain anaphors can obviously have the same antecedent (without c-commanding each other45), unlike what is assumed for standard probes—a problem for all approaches (e.g., Hicks 2009). Furthermore, in French and elsewhere, a coordination of singular DPs can behave from the point of view of agreement like a syntactically plural DP. Traditionally, this is modeled by saying that the whole phrasal conjunction counts as having feature values summing its parts. Using Agree, this can be readily modeled as a single probe having two goals (intervention cannot block this in the case of binding; see the fourth challenge below), the ‘‘sum’’ of whose features equals the feature composition of the probe. In other words, the summing could be a property of the probe rather than of the goal. In the case of binding via Agree, this would wrongly allow inclusive binding (as, e.g., in *Ces deux satellitesj+k dépendent de lui-mêmej et de celui de gauchek pour leur vitesse ‘*These two satellitesj+k depend on itselfj and the left onek for their speed’).
Inversely, probing of the same anaphor by, say, two distinct singular probes could license a plural anaphor with split antecedents, where this time the features of the goal behave in the same way as the ( plural) coordination of singular DPs. It is important to note that while such a conception could turn out to be incorrect, there is currently no reason, principled or empirical, why it is. This means that the dominant conception of Agree, while it derives the impossibility of split or partial binding, does so by stipulation and thus provides no support for binding as Agree.
A fourth challenge arises from the fact that an Agree relation needed for binding differs from standard Agree with respect to (non)intervention effects. Closest Attract/probe-goal/Agree imposes relative requirements, but antecedent-anaphor binding quite generally does not care about intervening elements. In other words, under an Agree approach (regardless of how it is implemented; e.g., multiple probing by heads (Reuland 2011) or probing by DPs (Hicks 2009), we expect the following type of configuration to be impossible, where > represents c-command:
(66) Antecedent1 > Antecedent2 > Anaphor1 > Anaphor2
Indeed, if Antecedent2 probes Anaphor2, given that multiple probing must be allowed (as a single antecedent can bind two anaphors), Antecedent1 should not be able to probe Anaphor1.
Such a restriction on plain-anaphor binding is in fact not found.
[Ces sondes]m analysent [chaque effet sur ellesm-mêmes]k comme [DP# une combinaison de sak propre magnitude et de leursm propres inerties].
‘[These probes]m analyze [each effect on themselvesm]k as [DP# a combination of itsk own magnitude and theirm own inertias].’
Note first that in (67), the object must be c-commanded by the subject, as it contains a plain anaphor (elles-mêmes) bound by it. Second, the plain anaphor leurs (propres inerties) ‘their (own inertias)’ is bound by the subject across chaque effet ‘each effect’, a quantifier that binds the plain anaphor sa (propre magnitude) ‘its (own magnitude)’, ensuring the requisite c-command relations. This pattern could be duplicated with all sorts of quantifiers (e.g., downward-entailing). We therefore know that the object chaque effet ‘each effect’ must c-command DP#. It thus intervenes between the antecedent ces sondes ‘these probes’ and the plain anaphor leurs (propres inerties) ‘their (own inertias)’ without blocking it.
In sum, an Agree approach explains neither the status of plain anaphors as plain, nor the prohibition against nonexhaustive binding, nor the fact that binding involves reference, nor the lack of intervention effects. This does not mean that an Agree approach is excluded. For the first three challenges, it merely shows that Agree is insufficient. The fourth challenge is more problematic, perhaps crippling, but is not unique to anaphor binding as it stands. In other words, there are no good independent reasons why Agree should be the right mechanism (although it may in fact turn out to be).
5.2.2 Anaphor Binding as Agree: Too Permissive
It is well-known—historically, this is one of the main motivations for the distinction between A- andĀ positions—that anaphors cannot come to be bound because of Ā movement of an antecedent. Thus, consider the following examples:
*These pictures, each other’s creators disliked.
*Quels tableaux leurs propres reproductions (et non pas celles d’autres tableaux du même artiste) ont-elles dévalués?
‘*Which paintings did their own reproductions (and not those of other paintings by the same artist) devalue?’
*Quels tableaux des reproductions d’eux-mêmes ont-elles dévalués?
‘*Which paintings did reproductions of themselves devalue?’
Such examples are ill-formed. The deviance is not due to weak crossover, as which-phrases (see, e.g., Hornstein 1995) trigger weaker (if any) crossover effects, and referential antecedents do not trigger any (see Lasnik and Stowell 1991). Under a Closest Attract/probe-goal/Agree approach, this is unexpected as antecedent and anaphor are (in a single phase and) not separated by any intervener.46
5.2.3 Condition A Holds at LF
Finally, we provide reasons why an Agree approach to plainanaphor binding is insufficient to derive Condition A locality effects (in French), as this condition must be seen not (only) as a derivational syntactic constraint but as an LF interface constraint.
The logic of the argument is as follows. If plain-anaphor binding is due to Agree, then binding should be able to take place in the requisite configuration (say, a probe locally c-commanding a goal) during a derivation. Binding locality would thus only need to hold at the point in a derivation where the Agree relationship is established. This configuration could be destroyed by further structural changes, such as movement of (a container of) the goal. We thus predict that the scope of (a container of) the goal should be independent of the position of the goal at the point at which it is probed. As we will show, this makes incorrect predictions. However, if Condition A is an interface condition holding at LF,47 the right predictions are made.
The first paradigm we discuss is based on the fact that the availability of a de dicto reading for some element requires that this element be interpreted in the scope of some intensional item. Consider (69a).
Pierrei savait combien d’effet négatif sur luii-même ces médecins pensaient qu’il existait à cause d’une boucle de feedback, tout en sachant qu’il s’agissait en fait d’un effet positif.
‘Peteri knew how much negative effect on himselfi these physicians thought there was due to a feedback loop, knowing full well it was in fact a positive effect.’
Peter knew [how much x [these physicians thought [there was x negative effect on himself] due to a feedback loop]]
In (69a), an anaphor, lui-même, is locally bound and thus satisfies Condition A in its postmovement position but not in its premovement position. In this sentence, it is possible to understand that the physicians are incorrect in thinking that the effect in question is negative: this is what they think, but in fact the effect is positive (and Pierre knows this). It is thus possible to continue with tout en sachant qu’il s’agissait en fait d’un effet positif ‘knowing full well that it was in fact a positive effect’. In other words, it is possible to interpret effet négatif ‘negative effect’ as de dicto (this is what the physicians think), not de re (they are mistaken). This means that it is possible to interpret the restriction of the wh-quantifier (effet négatif sur lui-même ‘negative effect on himself’) in the scope of the intensional verb penser ‘think’: it can be fully (radically) reconstructed at LF, yielding the representation (given in English) in (69b).
By contrast, such a reading is not available in (70a).
[La fréquence du courant]i détermine combien d’interférence négative avec ellei-même ces physiciens pensaient qu’il existait à cause d’une boucle de feedback. #Il s’agissait en fait d’interférence positive!
‘[The frequency of the current]i determines how much negative interference with itselfi these physicists thought there was due to a feedback loop, #It was in fact positive interference!’
The frequency of the current determines [how much x [these physicists thought there was [x negative interference (*with itself )] due to a feedback loop]]
Here, interférence négative ‘negative interference’ must be read de re (and either de dicto or not—depending on the physicists’ beliefs). Thus, the continuation Il s’agissait en fait d’interférence positive ‘It was in fact positive interference’ is bizarre. Remarkably, if the anaphor is omitted, the continuation is fine, signaling that a de dicto, not de re, reading of the restriction interférence négative ‘negative interference’ is possible again. In other words, sentence (70a) can have the LF representation given in (70b).
An explanation for this pattern correlating anaphor binding and scope is straightforward, provided that Condition A does apply at LF, and mysterious otherwise. In (69a), the anaphor is exempt, as it is animate and also has a perspective-holder antecedent (Pierre). Since no distance-to-the-antecedent requirement is imposed, the anaphor is able to (fully) reconstruct back to a position that is not included in the local binding domain as defined above; it can thus license a de dicto, not de re, reading. However, if the anaphor is not exempt, as in (70a), and Condition A must apply at LF, this reconstruction should not be possible. This prediction is borne out, showing both that Condition A holds at LF and that a derivational Agree approach is too permissive.
In this section, we have argued that if plain-anaphor binding is due to Agree, then (a) its properties are left unexplained (namely, that anaphors need to enter into an Agree relation, that reference is involved in plain-anaphor binding, that binding must be exhaustive); (b) plain-anaphor binding differs in several ways from standard Agree (multiple probing, insensitivity to interveners, possibly probing by XPs instead of heads).48 Most importantly, a standard Agree approach offers no explanation for (c) the failure of Ā movement to provide plain-anaphor binders. In addition, the derivational nature of Agree makes it unable to predict the pattern of reconstructability of moved anaphors, which suggests that (d) Condition A must apply at LF. We take this to mean that at a minimum some other mechanism than Agree must be responsible for binding, and the locality requirement must come from phase theory. And the fact that anaphor binding is insensitive to intervention suggests that Agree is in fact not involved.
5.3 Boundary Conditions on Deriving Binding Locality from Phase Theory
Descriptively, this is essentially the conclusion that we have reached:
(71) Binding domain for (French) plain anaphors
A plain anaphor
must be bound in the smallest TP containing it, and
cannot be separated from its antecedent by a subject.
Phase heads include C and v.
Tensed TP is the Spell-Out domain of the C phase.
Everything in vP (resp. DP) but the subject is the Spell-Out domain of the v phase (resp. D).
This immediately suggests the central idea we want to pursue: TP is a Spell-Out domain and also an opaque domain for Condition A. Quite generally, we could take binding domains to be Spell-Out domains. This brings us to the following two proposals for unifying binding locality and phase theory:
(72) Proposal 1
There is a locality domain because Condition A
applies at the LF interface, and
(73) Proposal 2
The binding domain for Condition A is the Spell-Out domain of a phase.
A plain anaphor must be bound within the Spell-Out domain containing it.
Proposal 1, essentially taking Condition A to apply at LF cyclically, is not new; this cyclic, phase-based idea is explicit in Lee-Schoenfeld 2008 and Quicoli 2008, for example.50 The LF part has been argued for in Fox 2000 and Fox and Nissenbaum 2004, among other works, and we have provided an independent argument in section 5.2.3.
Proposal 2 has two ingredients. First, it crucially refers to Spell-Out domain and not to phase, in effect what Lee-Schoenfeld (2008) proposes. This is necessary to our deriving the fact that tensed TP boundaries are opaque for Condition A (see section 5.2.2) and consistent with our conclusion that Condition A is an interface condition. Second, we will construe the notion ‘‘contain’’ literally, and this will help us explain why movement can increase the binding possibilities of an anaphor (e.g., it can escape tensed TPs by wh-movement) despite the copy theory of movement: in such cases, a single object can be simultaneously present both inside and outside a Spell- Out domain and is thus not contained in that Spell-Out domain.
5.3.1 The Case of TP
Let us illustrate the hypothesis that binding domains are Spell-Out domains with the case of tensed TP (here, we will write examples in English to simplify the presentation, but they are meant to be French examples with plain anaphors). In this case, illustrated by (74a–b), examples involve a finite TP complement of the phase head C, as structure (74c) indicates.
Once the phase is completed, the Spell-Out domain TP is transferred. This domain contains an anaphor. By Condition A as formulated in (73), this anaphor must be bound within that TP. Since the antecedent is external to TP, this type of representation crashes, ruling out both (74a) and (74b).
Note that this proposal immediately derives why movement such as wh-movement (or topicalization, 51 both superficially unbounded) to phase edges cannot supply new antecedents for anaphors: if the moved phrase did not qualify as an antecedent (e.g., via c-command) for the anaphor prior to movement, movement to the edge will not supply one as, by assumption, this moved phrase will not remain within TP, the Spell-Out domain. This is illustrated in (68d) for sentence (68c).
5.3.2 The Case of vP and Similar XPs with Subjects
The case of vP is somewhat more complicated. Consider the case of a verbal XP embedded, say, under a causative verb.
This sentence represents a general pattern where
X can be v, an ECM T, a small clause head, a D, an N, and so forth;
the subject of XP, DPk, can be or can contain an anaphor bound from outside XP (e.g., by DPm);
a more deeply embedded object, DPp, can be or can contain an anaphor bound by the subject of XP, DPk, but not anything further away (e.g., DPm).
Thus, in (75a) sa propre atmosphère ‘its own atmosphere’ can antecede sa (propre composition) ‘its (own composition)’: XP must be no bigger than a Spell-Out domain. But sa (propre composition) ‘its (own composition)’ cannot be bound by la Terre ‘the earth’: given our proposal, there must be a Spell-Out domain containing the former but not the latter. In other words, there must be a phase boundary between the subject of XP and the subject of vP. However, sa (propre atmosphère) ‘its (own atmosphere)’ can be bound by la Terre ‘the earth’: they must be in the same Spell-Out domain. We thus reach a contradiction if the structure is as indicated.
This means the structure of (75a) cannot be as shown in (75b). The subject of XP must be both in the same Spell-Out domain as la Terre ‘the earth’ and in the same Spell-Out domain as sa (propre composition) ‘its (own composition)’, but these two DPs cannot be in the same Spell- Out domain.
In fact, the standard assumption about such structures is nearly all we need to resolve this apparent contradiction: such structures are ‘‘raising-to-object’’ structures in which the embedded subject has raised into the main clause. Thus, rather than (75b), a better representation of (75a) is (75c).
In other words, there are two occurrences of the DPk sa propre atmosphère ‘its own atmosphere’: one inside the projection YP, which is the trace of the one raised outside of YP. To handle the binding possibilities in a way that is consistent with our construal of Condition A, it suffices to take YP to be ( part of ) the Spell-Out domain, and to take DPk to have moved out of it to the edge of some phase head (or beyond)—say, X here for concreteness.52 This proposal makes two predictions:
DPp (or something it contains) has a possible antecedent within its Spell-Out domain YP, namely, the trace
DPk. But DPm is outside of this Spell-Out domain and thus cannot antecede DPp.
DPk is actually not contained in YP (although one occurrence of it is); indeed, we are dealing with a single object with two occurrences,53 one inside YP, the other not. Not being contained within YP, it is not subject to Condition A within YP, and it (or an anaphor it contains) can thus be bound by DPp.
This way of construing containment (as a property of objects and not of occurrences) is in fact independently motivated. For example, wh-moving, say, the container of a plain anaphor increases its binding options. This has long been assumed, for such examples as (76).
(76) They know [which pictures of themselves [I like
which pictures of themselves]].
Such sentences are well-formed even though the lower copy of the anaphor does not satisfy Condition A; what matters is the highest (unreconstructed) copy.54 We can demonstrate the same effect in French, controlling at the same time for the plain status of the anaphor involved (something that should be done for English too).
La Terre, de par ses propriétés, soulève la question de quel effet sur sa propre inclinaison (et sur celle de la Lune) le vent solaire a eu
quel effet sur sa propre inclinaison.
‘Because of its properties, the earth raises the question of what effect on its own tilting (and on the moon’s) the solar wind has had
what effect on its own tilting.’
Here, the wh-moved phrase allows the inanimate possessor sa propre to take as antecedent the main clause subject, a relationship that would not be allowed without movement.55
The fact that what matters is the highest unreconstructed copy is illustrated by (70a), relevant parts of which are repeated here:
[La fréquence du courant]i détermine combien d’interférence négative avec ellemêmei ces physiciens pensaient qu’il existait.
‘[The frequency of the current]i determines how much negative interference with itselfi these physicists thought there was.’
The relevant syntactic structure is the lower CP phase in (78) (using English to represent the French).
(78) . . . [CP how much [interference with itself ]] [TP there existed how much [interference with itself ]]
The de dicto interpretation means that the (relevant portion of the) higher copy is deleted, yielding (79) as input to Condition A.
(79) . . . [CP
how much interference with itself] [TP there existed how much [interference with itself ]]
This deletion makes the plain anaphor fully contained in the TP: Condition A should be satisfied but is not, yielding the ill-formedness of such a reading.56
The consequences of this discussion are several. First, extending the logic of the TP case to this (and other cases) adds a source of evidence regarding the inventory of phase heads. The usual one is based on Move/Agree—that is, on Closest Attract/probe-goal configurations that fail in the absence of any intervention, revealing the presence of a(n absolute) phase boundary. This new source of evidence would be based on the binding possibilities for plain anaphors: if a c-commanding (local) binding relation between an antecedent and a plain anaphor is well-formed, there cannot be a phase boundary between them, and if it is ill-formed (even though all conditions are met except locality), there should be.
Second, the above conclusions require a slight rethinking of the earlier conceptions of the ‘‘vP’’ edge, as the subject of vP needs to start inside the vP Spell-Out domain and move out of it. The reason why the vP subject is usually considered to be merged at the edge is not completely clear to us, but an equivalent outcome would arise (with all the same advantages and drawbacks) if there were a phase head outside the vP Spell-Out domain attracting (probing) DPs: it would (in general) attract the closest DP first—that is, the subject.57
More generally, this would apply to any XP with a subject, which, unlike in the tensed TP case, tolerates being or containing a plain anaphor bound from the outside: ECM infinitives and small clauses (both of which would involve ‘‘raising to object’’58), vPs, and DPs with subjects that would have to involve a phase head (see Svenonius 2004). In all cases, the subject would have to move from a Spell-Out-domain-internal position to the edge of the immediately superordinate phase head associated with this Spell-Out domain.
Summarizing the discussion of XPs with a subject, we reasoned that if (a) XP’s subject is bindable from outside, it is outside the Spell-Out domain (at or beyond the phase edge), and that if (b) XP’s subject can bind an anaphor within XP, it is or starts inside the Spell-Out domain. We concluded that in cases where both (a) and (b) hold simultaneously, the subject of XP is both inside the Spell-Out domain and at the edge: if XP has a subject, XP must be (included in) a phase, and the subject has raised to its edge from some lower position inside the Spell-Out domain of this phase head. This conclusion is graphically represented as follows:
More generally, given our line of reasoning, the presence of a subject should always signal the presence of a phase. The bindability of this subject qua anaphor will depend on whether it is located at the edge of this phase or not; for all the cases above, it is. In the case of TP, it is not, but note that given the Extended Projection Principle (tensed TPs always have subjects),59 we predict that TP is not a phase but that a projection immediately above it (e.g., CP) should be.
5.3.3 The Case of XPs without Subjects
Let us now turn to the case of XPs lacking subjects. Empirically, what we have observed is that in such cases (VP, DP, NP), binding of a plain anaphor is allowed into XP from outside XP. Two English examples illustrating this possibility are (81a–b).
Mary saw [DP a picture of herself ].
Maryk T [VP seem to herselfk [to be tk a good candidate]]
As they stand, such examples are of course not telling, as we need to make sure that herself is not exempt. Here are some structurally close equivalents in French, where we control for nonexemption (with inanimates):
[La Terre]m subit [DP le réchauffement [DP de sam propre surface et de celle de la Lune]].
‘[The earth]m suffers from [DP the heating [DP of itsm own surface and of that of the moon]].
[L’horloge]m a [VP semblé ralentir à sonm propre fabricant et au fabricant de cette montre].60
‘[The clock]m [VP seemed to slow down to itsm own maker and to the maker of this watch].’
From this we must conclude that neither the DPs nor the VP headed by sembler ‘seem’ can be Spell-Out domains. Otherwise, the plain anaphor and the (derived) subjects would not belong to the same Spell-Out domains and binding would be excluded. Once again, this type of assumption is standard: (some) control infinitives are assumed to be phases (as they exclude movement out of them probed by a higher T; they are introduced by a C phase that turns the infinitive TP into a Spell-Out domain). Raising infinitives, on the other hand, are not phases for a symmetrical reason: the embedded subject can be probed by a higher T, so no phase boundary can intervene.
In order to explain what happens with DPs, we need to assume that a DP must be a phase only if it licenses a genitive subject; otherwise, it is not. Thus, the phase-inducing head must be that responsible for prenominal possessive pronouns (in French; and, presumably, for prenominal genitives in English).61 Similarly, in the case of the sembler/seem-headed VP: it cannot be a phase (as is standardly assumed given that the main T can probe into the infinitive), since the experiencer can come to be bound by the raised subject.
5.3.4 Binding Locality from Phases: Predictions for Extraction
The hypothesis that binding domains are Spell-Out domains makes predictions for extraction. Note first that different phases show different extraction possibilities.
Unlike the CP edge, the vP edge is tolerant: while the CP edge in languages like French (or English) only tolerates one (wh-)element (giving rise to some islands), under the standard view the vP edge tolerates multiple elements (the subject, possibly an object, a wh-phrase—on its way to the C edge).
A French DP, on the other hand, is intolerant: just as with CP, only one element can be attracted to the edge, if there is a phase head licensing a prenominal genitive. Given our reasoning above, this attracted element must be the subject of the DP if there is one. Once again, such an assumption is independently justified. Indeed, as is well-known for French (see, e.g., Sportiche 1990, Valois 1991), wh-extraction from inside DP is possible, but only if the extracted phrase can independently become the subject of the DP: this follows if the edge of the DP phase is intolerant, like the edge of the CP phase, only allowing one element, which must be the subject of DP (as will be exemplified below).
We end up with the following picture: extraction from inside XP is possible or not, depending on how tolerant of multiple elements the edge is.
We are claiming that the presence of the subject of an XP typically reveals the presence of a phase boundary, for binding reasons.
This claim establishes a connection between binding theory and movement theory in the following way: if a subject reveals the presence of a phase boundary, we should expect to see island effects due to the presence of this phase boundary, which correlates with the presence of a subject. But we should only expect to see evidence of such effects in intolerant phases—that is, DPs or CPs (as tolerant phases allow multiple extractions anyway)—and for constituents showing phase variability.
Basically, this means that DPs may or may not be phases, as opposed to CPs, which are always phases. In other words, we should expect to find cases in which
a given DP without a subject tolerates extraction (it is not a phase), and
the apparently same DP with a subject precludes extraction (it is a phase and extraction is blocked).
Such effects are indeed observed in French and are illustrated by the following cases of encliticization:
Pierre examine [DP la/une photo de cet immeuble].
‘Peter examines the/a picture of this building.’
Pierre en examine [DP la/une photo
Peter en examines the/a picture
‘Peter is examining the/a picture of it.’
Pierre examine [DP ma photo de cet immeuble].
‘Peter examines my picture of this building.’
*Pierre en examine [DP ma photo
Peter en examines my picture
‘Peter is examining my picture of it.’
Paradigm (84a–d) shows that extraction of en, the genitive clitic DP complement of the head noun, is sensitive to the presence of the possessive subject: a subject (here ma ‘my’) blocks extraction (the same pattern would hold if the extracted DP were a wh-phrase).
What is crucial is that even in cases where a DP c-commands a genitive en, movement is not blocked (so that a violation of Closest Attract is unlikely). This is shown by the following case in which the small clause (SC) is also a phase, but a tolerant one:
Pierre croit [SC Jean [capable de ça]].
‘Peter believes John capable of that.’
Pierre en croit [SC Jean [capable
In such a case, the intervening subject Jean does not block the movement of en, an unexpected outcome if an intervening DP blocked probing.
Furthermore, en-extraction is blocked by a DP subject of DP even if en is not plausibly analyzed as a DP, thus not plausibly probed by the same element as this subject.
Pierre lit [DP trois livres (de cuisine)].
Peter reads three books (about cooking)
‘Peter is reading three (cook)books.’
Pierre en lit [DP trois
en] (, de livres de cuisine).
Peter en reads three
en(, of books about cooking)
‘(As for (cook)books,) Peter is reading three.’
*Pierre en lit [DPmes trois
en] (, de livres (de cuisine)).
Peter en reads my three (, of books (about cooking))
‘(As for (cook)books,) Peter is reading my three.’
Pierre en lit [DPles trois premiers
en] (, de livres).
Peter en reads the three first (, of books)
‘(As for books,) Peter is reading the first three.’
*Pierre en lit [DPmes trois premiers
en] (, de livres).
Peter en reads my three first (, of books)
‘(As for books,) Peter is reading my first three.’
A comparison of (86a) and (86b) illustrates that much like one-replacement in English (this
bigbook about cooking/this big one about cooking/this bigone about cooking/this big one about cooking), the clitic en can replace a bare noun or a bare noun plus some or all of its dependents, but not numerals. It can thus be analyzed as a pro-NP, moving from DP-internal position to clitic position in the T domain. Note in particular that such a pro-NP can never be realized as the possessor of a DP.
What the paradigm in (86c–e) shows is that the presence of an intervening DP subject blocks extraction of this NP. Because pronominal possessors and definite articles are conflated in French, the interpretation of (86c) is unclear. However, (86d) (with a definite article but without a possessor, where the extraction is possible) and (86e) (identical to (86d), but with a possessor, where the extraction is blocked) show that the blocker is not the definite article but the blockage is related to the presence of the possessor. This blocking effect once again is not plausibly an intervention effect on probing, as the DP subject and the NP are not attracted by the same elements. If, however, the presence of a possessor is made possible by a phase-inducing head, all these blocking effects are just reflexes of the intolerant character of this DP phase.
We have proposed to reduce the locality imposed on the relation between a plain anaphor and its local binder by requiring that an anaphor be bound in its Spell-Out domain.
This proposal has led to slight modifications of phase theory (regarding both the inventory of phases and the properties of phase edges), most independently justified. These modifications, in turn, have allowed us to account (e.g., in the case of extraction from DPs) for phenomena seemingly showing a mixed behavior: an intervening subject selectively blocking extraction of a DP or of non-DP XPs.
Our proposal also allowed us to derive two properties of the interaction between movement and binding. First, we have derived why movement of a potential antecedent to the edge never provides new binders for plain anaphors. Second, we have predicted that movement of a plain anaphor to the edge does increase the set of potential antecedents for this anaphor.
5.4 Some Consequences and Questions
In the context of the theory of exempt anaphora that he defends, Reuland (2011:93) claims that there is no need for reconstruction, at least as far as ( plain-) anaphor binding is concerned. Thus, in examples such as (87a–b) there is no need to assume that any particular structural relation has to hold between the anaphors and their antecedents, because these anaphors are all claimed to occupy exempt positions.
Given the theory developed here, however, this conclusion does not hold. Indeed, the positions of anaphors are not exempt, but because they are all animate, the anaphors themselves may be exempt. To decide whether reconstruction is needed, we would have to decide whether these anaphors are exempt in such cases. This would require a theory of when exemption is allowed—that is, precisely what conditions an antecedent must meet to make the anaphor it binds exempt (see Charnavel 2014 for a proposal).
But there are other ways to decide the issue. We have already seen one: the availability of a de dicto, not de re, reading in (69a) shows that radical reconstruction of (the restriction of) the moved wh-phrase is possible.
We can construct another argument by looking at what happens with inanimate anaphors in French, which we have shown cannot be exempt. Consider the following pairs of examples involving possessor son propre and elle-même that must be locally bound by inanimate antecedents:
[Cette loi]i a entraîné la publication d’un livre entier sur sesi propres conséquences et celles des décrets associés.
‘[This law]i led to the publication of a whole book about itsi own consequences and those of the related decrees.’
Je me demande de quel livre sur sesi propres conséquences et celles des décrets associés [cette loi]i a entraîné la publication.
‘I wonder which book about itsi own consequences and those of the related decrees [this law]i led to the publication of.’
[La Terre]i impose des forces latérales sur ellei-même du fait de sa rotation.
‘[The earth]i imposes lateral forces on itselfi because of its rotation.’
Quel genre de forces sur ellei-même [la Terre]i impose-t-elle du fait de sa rotation?
‘What kind of forces on itselfi does [the earth]i impose because of its rotation?’
In both cases, possessor son propre (resp. elle-même) must be a plain anaphor. Since (88b) and (89b) are well-formed, this anaphor should be c-commanded by its antecedent, but it is not. However, this structural requirement is met prior to the movement of the wh-phrase. Given our earlier conclusion that Condition A must hold at LF, reconstruction must be hypothesized to feed Condition A.
5.4.2 The A/Ā Distinction
As we have discussed, certain kinds of movement can feed Condition A by providing a c-commanding antecedent not available otherwise. Classically, the difference between instances of movement that can feed Condition A in this manner and ones that cannot was expressed in terms of the A/Ā position difference. Anaphors were required to be A-bound so that only A-movement could be such a feeder. Even though this amounted to an undesirable stipulation (why is this difference relevant?), it was at least possible to define the A/Ā difference in a non–ad hoc way, namely, by taking A-positions to be potential θ-positions. Indeed, the major motivation for defining A-positions in terms of potential θ-positions was the fact that subjects of TP could be anaphor binders even though they were not always θ-positions—for example, in the following type of raising-to-subject sentence:
(90) Maryi T [VP seem to herselfi [to be ti a good candidate]]
Because of the exempt-anaphora confound (the reflexive is animate, hence possibly exempt), we repeat the French example (82b) with an inanimate anaphor to illustrate this point without this confound.
[L’horloge]m a [VP semblé ralentir à sonm propre fabricant et au fabricant de cette montre].
‘[The clock]m [seemed to slow down to itsm own maker and to the maker of this watch].’
However, this type of definition has been unavailable since the introduction of the Predicate-Internal Subject Hypothesis, which uniformly makes the subject of a TP, or the relevant position from which son propre is bound, never a θ-position.
The proposal we are making in effect redefines the A/Ā distinction not in terms of differences between landing positions, but in terms of movement span (at least as far as binding theory is concerned), in the following way:
Ā movement is movement to the edge of a phase.
A-movement is movement within the Spell-Out domain of a phase head.
Historically, the A/Ā distinction is not invoked solely for its role in feeding binding. It has been claimed to play a role in several other issues. For example:
Improper movement (is improper movement—Ā movement followed by A-movement of the same element—allowed?)
Weak crossover (A-movement does not trigger weak crossover effects;Ā movement can)
Licensing of parasitic gaps (A-movement does not license parasitic gaps; Ā movement can)
All such correlations are controversial but would be worth exploring in the context of our characterization of the A/Ā distinction.
5.4.3 Deriving Condition A: What It Would Mean
Taking the previous conclusions into account, we conclude that the correct descriptive generalization for Condition A is that proposed in (73), repeated here:
A plain anaphor must be bound within the Spell-Out domain containing it.
Assuming this formulation, a theory of Condition A should explain the following properties:
Why certain expressions are plain anaphors (= require an antecedent)
Why Condition A holds of plain anaphors at LF (i.e., at the syntax-interpretation interface)
Why the antecedent of the plain anaphor
c-commands the anaphor,62 and
must be in the same Spell-Out domain as the anaphor
Why the antecedent of the plain anaphor has a referential value
either exactly identical to that of the plain anaphor
or weakly (proxy) identical
Property 7 takes into account Jackendoff ’s (1992) Madame Tussaud ‘‘proxy’’ cases, in which the anaphor can be interpreted as referring to a proxy of the reference of its antecedent (as in John fell on himself, where John the person falls on his wax statue).63 Crucially, such proxy binding can behave like standard plain-anaphor binding (e.g., with inanimate elle-même: La lune fait de l’ombre sur elle-même ‘The (real) moon casts a shadow on (a wax representation of) itself ’).
Given current formalizations, no proposal in the literature explains properties 1, 2, and 5.
Explaining property 1 means explaining why (e.g.) inanimate French son (resp. French elle), combined with the adjective propre inducing focal alternatives on son (resp. with même), turns son (resp. elle) into a plain anaphor—especially given Charnavel’s (2013) conclusion that this ought to be a synchronic, compositional process. It also means explaining why animate versions of the very same items are seemingly allowed to escape Condition A.
Explaining property 2 means deriving why this condition must hold at the LF interface rather than, say, derivationally only.
Explaining property 5 means deriving the size of the locality domain for plain-anaphor binding, and crucially why the antecedent cannot be at the edge of a phase in an Ā position.
Explaining property 4 could follow from taking binding to be Agree or Move, but each proposal faces problems. (Derivational) Agree by itself fares the worst, as it does not explain the uniqueness of the antecedent ( property 3), the fact that reference is involved, and, when it is, the fact that exhaustive binding is required ( property 6) (as already mentioned in section 5.2).
Movement approaches fare better, although some ( possibly fixable) problems arise.
The movement approach proposed by Hornstein (1995, 2001), discussed by Drummond, Kush, and Hornstein (2011), assumes that a plain anaphor simply is a trace of its antecedent. It does derive properties 3 and 6. But it seems to make a principled treatment of property 1 difficult to attain (why should traces be spelled out the way they are—fairly consistently across a wide variety of languages?). Furthermore, the movement approach may be difficult to reconcile with property 7: unless the syntax of such cases is not what it seems (definitely an option), this approach requires that an antecedent and its trace be allowed to differ in interpretation in a way never possible otherwise. In addition, such an approach seems too restrictive: while interpreted copies of a moved element must have exactly the same interpretation, an antecedent and an anaphor need to have the same denotation but not the same interpretation. Thus, in The children looked at themselves, both the children and themselves must refer to the same set, but this need not mean (under a nondistributive reading) that, for example, all the children were looked at; a vaguer interpretation is possible whereby the set of looked-at children can count as the whole set of children.
Kayne’s (2002) suggested movement approach to anaphor binding invokes doubling: the antecedent and the anaphor are generated as a double (e.g., in English [[John him] self ], with the antecedent John moving away). It shifts to properties of doubling the questions about the uniqueness of the antecedent, about the fact that reference is involved, and about exhaustive binding ( properties that actual doubling cases—for example, clitic left-dislocation—do exhibit). It can thus reasonably be said to reduce properties 3, 4, and 6 to independent factors. It does not preclude explanatory answers for property 1 but faces the same kind of problem as Hornstein’s (1995, 2001) approach regarding property 7 or the nonidentity of meaning just discussed.
The goal of this article was to investigate the behavior of anaphors based on French data involving son propre and elle-même with the aim of (a) circumscribing the empirical generalizations applying to plain anaphors and subsumed under Condition A of binding theory, (b) exploring boundary conditions on explanatory accounts of Condition A, and (c) ultimately deriving its effects. Unsurprisingly, French anaphors clearly support the need to distinguish between plain anaphors, which obey Condition A, and anaphors that are exempt from it (hence the need for a theory of exempt anaphora).
Because, as crosslinguistic work suggests, exempt anaphors seem to need to refer to live persons, restricting attention to inanimate anaphors should be a useful tool for circumscribing the scope of Condition A. It turned out this way for French: making use of this independent criterion, we demonstrated that (a) the plain/exempt distinction correlates with inclusive/exhaustive binding possibilities, and (b) a plain anaphor must be bound within a local domain that roughly corresponds to the smallest XP with an intervening subject containing the anaphor and that is no larger than a tensed TP.
We believe that this type of investigation, clearly and reliably separating plain from exempt anaphors, must be carried out in a fair sample of languages before trying to formulate a general theory of Condition A locality of binding.
Further, we argued that this view of binding domains should be integrated into the grammar in terms of phase theory: we formulated Condition A as requiring that a plain anaphor must be bound within the Spell-Out domain of a phase containing it, and we pointed out that appeal to Spell-Out (or Transfer) domains—a notion exclusively relevant at the syntax-interpretation interface—makes sense of the fact that Condition A must be an interface condition. This also allowed us to derive various properties of anaphor binding (e.g., when movement can feed or bleed Condition A), as well as providing new empirical grounding for the notion of phase, the inventory of phase heads, and the paths of syntactic derivations.
Many questions remain that we cannot address here. Some are of a general nature. One question concerns the exact nature of the binding relation between an antecedent and a plain anaphor, which some authors (Hicks 2009, Reuland 2011, Rooryck and Vanden Wyngaerd 2011) propose to reduce to an Agree relation, and others (Hornstein 1995, 2001, Kayne 2002) account for by invoking movement (plus doubling). Another, related question, raised earlier, asks what makes an expression anaphoric, and beyond this how the behavior of plain and exempt anaphors, which can be intrinsically identical, can superficially differ rather radically in terms of locality or inclusive reference, for example (a question discussed in part in Reuland 2011).
Tying these last two concerns together, Charnavel (2014) proposes that apparent longdistance binding is always mediated by an intermediate, Spell-Out-domain-internal logophoric operator LOp, so that even exempt anaphors obey Condition A. The fact that an exempt anaphor allows inclusive binding as shown in section 4.2 is thus consistent with its being exhaustively bound by LOp, and thus with a movement approach, with inclusive binding being a property of the referential dependence between LOp and its own antecedent(s).
Some remaining questions are tied to our particular proposals. How do our findings about the distribution of these French anaphors generalize to other anaphors in French and in other languages? These (French) inanimate anaphors are never exempt from Condition A: Are inanimates always plain crosslinguistically? Are animates always ( potentially) exempt? While our conclusions are consistent with some findings (e.g., with what Huang and Liu (2001) report about the Chinese anaphoric system), it remains to be seen how generally they hold and in particular how they can be integrated with the substantial body of work on anaphora such as Safir 2004, Hicks 2009, Reuland 2011, and Rooryck and Vanden Wyngaerd 2011 into some coherent whole.
For their comments, suggestions, and feedback, thanks to our Harvard and UCLA students, anonymous reviewers, and the audiences at GLOW in Asia IX (2012), the 19th International Congress of Linguists in Geneva (2013), the Generative Initiatives in Syntactic Theory (GIST) 6 conference in Ghent (2012), and the 2012 FoM Thomasberg Workshop. This material is based upon work supported by the National Science Foundation under grants 1424054 and 1424336.
1 This difficulty comes from the fact that exempt anaphors are often, but not always, more permissive than plain anaphors in terms of what antecedent they allow.
3 One common answer to question 2 is that anaphors are referentially deficient, which strikes us as nearly tautological. We would take as explanatory a compositional theory of what makes an expression such as himself an anaphor based on the properties of its parts (a pronoun plus self ).
4 Unless it is explicitly mentioned (see section 4), the bound vs. coreferential difference is irrelevant to our discussion. Throughout, the data would be the same whether the antecedent of an anaphor under discussion is quantificational—thus requiring semantic binding—or not.
6 Thus, the four questions above are really eight questions, four for plain anaphoric behavior and four for exempt anaphoric behavior.
7 For an argument that it is not (e.g., historically) accidental that the very same elements exhibit both behaviors, see Charnavel 2013.
8 As a first approximation, we take exempt anaphors to be those expressions that can, like plain anaphors, be bound locally (unlike pronouns) and long-distance. This definition will be refined in the course of the discussion.
10 This fact is explicitly mentioned for Icelandic in Sigurðsson 1990.
11 From now on, we will encode the relevant distinction as animate vs. inanimate. However, it should be kept in mind that this is too rough a characterization, as seemingly nonanimate terms can be used as proxies for people (e.g., the parliament, Washington, Japan)—that is, as referents capable of thought, of holding a point of view, and so on.
12 The expression son propre can yield different interpretations: for example, possessor son propre, possessum son propre. Only the former, characterized by its focus properties, exhibits a correlation between animacy and binding locality. (For a detailed discussion, see Charnavel 2012.) Roughly, stress falls on the head noun in possessum son propre, which we do not discuss here, while it falls on propre in possessor son propre, which we do discuss here. The import of this correlation between stress (focus) and anaphoricity is discussed in Charnavel 2011, 2013.
13 As is standard, the star (*) is used contrastively: starred sentences are significantly more degraded than corresponding sentences without a star.
The French data come from two sources: elicitation judgments and a statistically analyzed online magnitude estimation judgment task questionnaire presented in Charnavel 2012:chap. 1, appendix.
14 Note that the sentence with c-command of the antecedent is degraded in the absence of même (an effect of Condition B), while a similar sentence with son propre is not degraded in the absence of propre.
15 The judgment of the corresponding English sentence would be different here (i.e., iti*(self )). But note that throughout the article, the English translations are meant as glosses of the French examples, so that the (absence of ) stars indicated in the English reflect(s) the French judgments. We do not take any stand on the judgment of the corresponding English sentences.
16 Such examples constitute a problem for Reuland’s (2011) view extended to the French cases. In his view, syntactic predicates are redefined as needing to have an event role, and, (some?) Ns are hypothesized not to have an event role. This problem is not manifest in Reuland 2011, where only himself, not itself, is examined. For English, examples should be found to test whether the antecedent must occur within the DP when itself appears in a DP with subject.
17 Note that all the deviant sentences that follow in this section would be well-formed if they were minimally modified to make the antecedent a (live) person.
18 Examples like (20c) suggest that Ps must have ( possibly silent) syntactic subjects (as it looks like the direct object is controlling a silent subject of the PP). This is relevant to the discussion of coargumenthood in section 3.2.3. As a reviewer points out, such cases must be distinguished from those in (8), in which the object is transparent even though it is likely to be the subject of some kind of small clause. We attribute this transparency to the incorporation of the predicate of the small clause.
19 We do not necessarily endorse this conclusion, as we believe that proper controls to separate exempt SE anaphors from nonexempt SE anaphors have not been generally conducted. A significant exception is Huang and Liu’s (2001) demonstration that the Chinese SE anaphor ziji is subject to Condition A when nonexempt.
20 Here is what Reinhart and Reuland (1993:671–672) say: ‘‘Syntactically, there is just one type of SELF anaphor, whose occurrence is governed solely by Condition A, as stated in (12″). This condition rules out argument SELF anaphors that occur in nonreflexive predicates’’ (i.e., are not coindexed with a coargument).
Similarly, Reuland (2011:83) characterizes Condition A as follows: ‘‘A reflexive-marked syntactic predicate is reflexive.’’ This means that if a predicate has a reflexive syntactic argument, the antecedent of this reflexive must be an argument of this predicate.
Pollard and Sag (1992:266) write, ‘‘An anaphor must be coindexed with a less oblique coargument, if there is one.’’
22 A third case where anaphors are exempt according to Reinhart and Reuland (1993) corresponds to focus anaphors. The CBCA is claimed to apply at LF, and the focused expression undergoes movement at LF; in examples like (i), the anaphor is no longer in an argument position, as shown in the representation in (ii), and is therefore deemed to be exempt.
This letter was addressed only to myselfi.
myselfi (this letter was addressed only to ei)
It is in fact not clear that in (i), the object needs to be focused for the sentence to be well-formed. See also footnote 27.
23 Namely, the Locally Reflexive Principle: ‘‘An identity-specific anaphor (SELF-form) is dependent on its coargument antecedent if it has one’’ (Safir 2004:108).
24Reinhart and Reuland (1993:658n4) exclude possessive anaphors from their discussion, but Reuland (2011:167) discusses them: ‘‘Binding of POSS anaphors . . . can be straightforwardly accounted for by general principles of chain formation. The POSS phrase is realized in the left periphery of the DP-projection, and hence in principle is accessible for chain formation with the verbal functional system, unless some other factor intervenes. . . . Chain formation via the extended verbal projection explains that POSS anaphors are subject oriented.’’ Clearly, the claim that chain locality holds between the anaphor and its antecedent is contradicted by (27) given that A-movement (e.g., possessor raising) is possible only for the possessor of a direct object, not for any more deeply embedded possessor. Furthermore, no subject orientation is found in French (we suspect that Reuland’s discussion may more correctly apply to SE possessive anaphors, which are subject-oriented, but crucially not to SELF possessive anaphors).
25 Note that Reuland’s (2011) proposal does not face these particular problems, as Reuland takes eligible coarguments to be arguments of a predicate with an event variable. The example Johnk seemed to himself tk to hurt himself in the text suggests that coargumenthood in fact plays no role. In addition, John needs to count as an argument of hurt, given the embedded reflexive object. Given Hartman’s (2009) analysis of tough-movement constructions, pairs such as He believes himselfk to be tough for himk to trust tk/She believes herselfk to be tough for him to trust tk pose a problem. In the former, the reflexive ought to count as a coargument of trust via its trace, which seems incompatible with the well-formedness of the latter.
27 Supposing that the anaphor is a focused anaphor, which Reinhart and Reuland (1993) propose is exempt, would be too permissive: it would wrongly predict that son propre should always be exemptable from Condition A, even when inanimate.
28 Supposing that elle-même is a focused anaphor in (29)–(32) would correctly predict it to be exempt under the CBCA theory (see footnote 22) and would correctly predict the sentences to be grammatical. This can be controlled for. For example, (29a) is perfectly well-formed as an answer to the question Quand Mariei s’inquiète-t-elle du fait que ses enfants dépendent d’ellei-même? ‘When is Maryi worried that her children depend on herselfi?’, where the focus both of the question and of the answer is on the temporal adjunct (quand ‘when’ in the question, souvent ‘often’ in the answer) and not on elle-même. Furthermore, if focus was at play, inanimate elle-même should be able to behave the same, but it is not.
[La Terre]i est dégradée par les êtres humains même si leur avenir ne dépend que d’ellei-(*même).
‘[The earth]i is degraded by human beings even if their future only depends on iti(*self ).’
Les habitants de [la Terre]i dépendent d’ellei-(*même).
‘The inhabitants of [the earth]i depend on iti(*self ).’
Both sentences (with même) are degraded (locality violation in (i), c-command violation in (ii)).
29 In principle, we should be able to observe Condition A effects with an animate anaphor when no suitable antecedent is available. In the absence of an explicit theory of exempt anaphora, it is difficult to illustrate such a case and justify it. However, here is an example, which contrasts with (33a):
*Marie dit de Jean que ses collaborateurs sont fiers de lui-même.
‘*Mary says of John that his coworkers are proud of himself.’
(i) is degraded since Jean is not a suitable antecedent (i.e., not a logophoric center). See Charnavel 2014 for further discussion.
30 More precisely, a weaker form excludes a stronger form if the latter is more specified than the former. This is why the French clitics le (accusative ‘him’) and lui (dative ‘him’) exclude strong lui (dative/accusative ‘him’), but the clitics en (‘of him/her/it’) and y (‘at/to him/her/it’)—analyzed as prepositional clitics (see, e.g., Kayne 1975)—do not (because en and y incorporate Case information that the strong pronouns do not).
31 We say that se stands for an argument and place an index on it to facilitate presentation, without any claims about its actual role.
32 When se is available, elle-même can also be acceptable if it is added to se. This gives rise to focusing of the relativized object (and irrelevantly can also yield an exclusive reading of the subject in (i), as emphatic himself can do).
Jeani si’examinera luii-même.
‘Johni will examine HIMSELFi.’
Jeani sei décrit le paysage à luii-même.
‘Johni describes the landscape to HIMSELFi.’
This is compatible with—in fact, predicted by—Cardinaletti and Starke’s (1999) proposal.
34 As expected, this is true even if the antecedent is not a subject. For example:
??Marie a informé Jeani qu’on présenterait Suzanne à luii-même.
‘??Mary informed Johni that one would introduce Susan to himselfi.’
Marie a informé Jeani qu’on luii présenterait Suzanne.
‘Mary informed Johni that one would introduce Susan to himi.’
36 Crucially, Ahn (2014) shows that deaccenting of reflexives works differently from deaccenting of pronouns.
37 The only proper attempt to do this that we know of is by Fox and Nissenbaum (2004:sec. 3). Discussing examples involving animate reflexives in English, they conclude that their facts would follow if the reflexives were either subject to Condition A or exempt from it, a conclusion very much along the lines of what we conclude for French. If they are right, binding Condition A theories must be thoroughly revised for English, hopefully along the lines that we suggest for French (as hinted at in section 3.2.5).
40 It may well be that complementary distribution—hence, some kind of competition approach (see, e.g., Safir 2004)—is right once meaning is taken into account, but it cannot be based purely on position. It may also be that plain anaphors as we define them are in complementary distribution with pronouns. We leave this open here (deciding is difficult because of various confounding factors), but it would be relevant for the treatment of Condition B.
41 This assumes the standard view (adopted in Reinhart and Reuland 1993) that if reflexives are semantically bound, only a sloppy reading is available. As a reviewer points out, this standard view may be false. If so, ellipsis provides no argument that reflexives must be semantically bound.
42 Identity must be understood in the right way to account for binding by nonreferential expressions and also modulo proxy interpretation (viz., Jackendoff ’s (1992) Madame Tussaud cases). This extends to reciprocal expressions for which referential identity is also required in the following sense: in for example They like each other, which we take to mean roughly ‘Each of them likes an other of them’, the sets over which each and other range (namely, them) must be identical to ||they||, where they is the antecedent of the reciprocal.
44Drummond, Kush, and Hornstein (2011) also discuss other reasons why Agree does not suffice for anaphor binding. In particular, they discuss some technical reasons for distinguishing standard Agree from ‘‘binding’’ Agree, such as the required presence of (un)valued and (un)interpretable features on both probe and goal not obviously satisfied by ‘‘binding’’ Agree.
45 A configuration Antecedent1 > Anaphor1 and Antecedent1 > Anaphor 2 but not Anaphor1 > Anaphor2 (where > represents c-command), clearly possible, shows that multiple probing cannot be reduced to Anaphor1 probing Anaphor2 and being probed by Antecedent1.
46 For the system outlined in Reuland 2011, which treats some anaphor binding as a case of a single head probing both the antecedent and the anaphor, the impossibility of such cases reduces to, for example, heads in the C system not being allowed to act as anaphoric probes, an unexplained gap (especially given the existence of agreeing Cs in, for example, Bantu languages like Kilega; see Kinyalolo 1991).
47 That Condition A must be an LF condition is not a new idea, but we do not believe that arguments based on the de re/de dicto distinction have been proposed before. Arguments for viewing Condition A as an LF condition are found in Fox and Nissenbaum 2004 and Lebeaux 2009, among other works; also see references therein. Hicks (2009) argues otherwise, unconvincingly in our view: the only arguments showing that Conditions A and B cannot hold at LF (a) only concern Condition B and (b) are based on Condition B’s being sensitive to properties (e.g., phonological ones) deemed not to be visible at LF. However, Hicks gives no arguments that these properties could in fact be reflexes of syntactically encoded structural differences, hence visible at LF.
50 We do not take a stand here on the validity of the grounds under which such proposals were put forth, as ours are different. Lee-Schoenfeld (2008) takes reflexives, what we call plain anaphors, to have to be bound within the phase containing them, even though she at one point takes tensed TP to be a boundary ( p. 289, (15a)).
52 DPk could also have moved past the edge, which would not change the essence of what we present. For the sake of simplicity, we will ignore this option here.
53 It is necessary (under the copy theory of movement) to distinguish the case of Remerge/Move of some item from the superficially similar case of two objects being identical in all respects, including their index, and being merged independently. The former is the case of Move, subject to phase locality; the latter is not. Thus, the notion of ‘‘single object with multiple occurrences’’ is needed.
55 The property of movement to the edge to extend the binding of anaphors could perhaps be exploited to explain why, in languages like Chinese, an anaphor in subject position can take an antecedent outside of its clause even when it is plain (as is well-known for the SELF anaphor ta-ziji—as noted earlier—and as is shown in Huang and Liu 2001 for the SE anaphor ziji): if such subjects were allowed to move into the edge, this would be expected (footnote 49 may be relevant here).
56 Technical questions arise in a derivational bottom-up cyclic Spell-Out view, a view we do not commit to. If such a view were adopted, the information about which one is the highest copy (given the issue of containment) must be available at the point at which Condition A applied, however Condition A is formalized. In an Agree approach, the plain anaphor in (77) still has not ‘‘Agreed’’ with its antecedent by the time it is transferred, yet the derivation does not crash. In our approach, it must count as not being contained in the Spell-Out domain. In (70a), the deletion of the higher trace at the edge must be allowed (but not required) to be triggered by the transfer of the Spell-Out domain, to yield an anaphor fully contained in this domain.
57 In slightly different ways, this is argued by many authors—for example, by Hinterhölzl (2006) (the phase is AspP above vP) and by Coon, Mateo Pedro, and Preminger 2014 and Deal (2013) (the phase is vP but subjects are introduced lower in VoiceP). That subjects do move through a vP-external intermediate position is evidenced by (i), given Sportiche’s (1988, 1997) analysis of floating quantifiers (with tout ‘everything’ at the vP edge).
[[Les enfants]k ont [tous tk [tout [tk lu]]]].
‘??Mary informed Johni that one would introduce Susan to himselfi.’
[[the children]k have [all tk [everything [tk read]]]]
‘The children have all read everything.’
59 The case of infinitival TPs is more complex and controversial. Whether the Extended Projection Principle holds at the TP level will depend on further considerations (whether control is movement or not, whether raising infinitives involve some kind of restructuring or not, etc.).
60 Unlike English, French requires extraposition of the experiencer in raising cases. This sentence must thus be read with the appropriate intonation.
61 This may be tied to when possessor raising is allowed/required: precisely when a possessor lacks DP-internal Case licensing and can thus escape the nonphasal DP.
62 It is sometimes reported that c-command is not necessary (see, e.g., Drummond, Kush, and Hornstein 2011:414ff. and fn. 37, and references therein). Such cases all involve animate reflexives, hence are not properly controlled for the plain/exempt distinction. Furthermore, it is not even clear that the reported cases do not in fact involve LF c-command.
63 But not vice versa (John fell on himself cannot mean that wax-John fell on real John), an asymmetry that must be accounted for independently.