2. Anaphoric binding: TP and vP phases

A number of linguists argue for the reducibility of local binding domains to phases (Lee-Schoenfeld Reference Lee-Schoenfeld2004, Canac-Marquis Reference Canac-Marquis and Müller2005, Quicoli Reference Quicoli2008, Hicks Reference Hicks2009, Safir Reference Safir2011, Despić Reference Despić2011, Antonenko Reference Antonenko2012, Charnavel and Sportiche Reference Charnavel and Sportiche2016). Some assume that the binding domains for anaphors are CPs/TPs and vPs, drawing on Chomsky's (Reference Chomsky and Kenstowicz2001, Reference Chomsky, Freidin, Otero and Zubizarreta2008) proposal that CPs and vPs constitute phases, while some others further argue that PPs and/or DPs can also be phases (Lee-Schoenfeld Reference Lee-Schoenfeld2008; Hiraiwa Reference Hiraiwa2005; Bošković Reference Bošković and Gallego2012, Reference Bošković2014; Despić Reference Despić2015). The basic idea behind phase theory is that a sentence is decomposed into syntactic domains called phases, each of which is spelled out and independently sent off to the interface levels, Logical Form (LF) and Phonetic Form (PF), which assign semantic and phonetic/phonological interpretations to the utterance. However, the phase head and its edge on the one hand, and the complement of the head on the other hand undergo Spell-Out separately. This means that the edge of a phase is accessible to operations in the next higher phase, following Chomsky's (Reference Chomsky and Kenstowicz2001: 14) Phase Impenetrability Condition, given in (9).

1. (9)

   1. a. [_ZP Z … [_HP α [H YP]]]

   2. b. Phase Impenetrability Condition (PIC): The domain of H is not accessible to operations at ZP; only H and its edge (α) are accessible to such operations.

Under PIC, YP in (9) – which is the domain/complement of H – is spelled out within HP, but H and its edge α are spelled out at the ZP level (Chomsky Reference Chomsky and Kenstowicz2001: 13). In other words, the head of a phase and its edge are spelled out along with the complement of the higher phase head. Chomsky (Reference Chomsky and Kenstowicz2001: 13–14) provides two versions of PIC: the strong PIC and the weak PIC. In (9) I quote the weak PIC, which is the version adopted here. This is so because it captures facts that cannot be accounted for by the strong version. Consider the sentence in (10).

Focusing on the distributor baʕḍuhum ‘each’ in (10), it moves from a lower position within a PP to Spec-PP, where it is pronounced (I will discuss this kind of movement within both possessive DPs and PPs in detail in the next two sections). T in (10) agrees with baʕḍuhum, which is itself at the edge of PP whose Spell-Out domain is the VP (the complement of the phase head v). Agreement with T derives the nominative Case marking. That is, baʕḍuhum is still accessible to T under the weak PIC because it is not spelled out until the next phase head is merged (here, C, based on the definition of PIC in (9); see Chomsky Reference Chomsky and Kenstowicz2001, Citko Reference Citko2014). For a somewhat similar phenomenon in Icelandic and Polish, see Citko (Reference Citko2014: 35). Assuming, following Citko, that only phases can have an element pronounced at their edges, the fact that baʕḍuhum is pronounced in Spec-PP provides evidence that such possessive PPs in Arabic count as phases (see sections 3 and 4). In a phase-based binding theory, the binding principles have been restated as follows:

Binding in Arabic adheres to the principles in (11a–b), as has been claimed for some other languages like English (Quicoli Reference Quicoli2008; Despić Reference Despić2011, Reference Despić2015) and German (Lee-Schoenfeld Reference Lee-Schoenfeld2004, Reference Lee-Schoenfeld2008). Consider the Arabic examples in (12) and (13).

The anaphors in (12)–(13), whether reflexive or reciprocal, are locally bound within their vP phase, as illustrated by the LF representations in (12b) and (13b). Essentially, the matrix subject binds the anaphor within the vP before it moves to the specifier of TP in both examples (see Quicoli Reference Quicoli2008); Principle A is thereby satisfied, as predicted. A binding relation between a pronominal and a DP within its vP phase, by comparison, violates Principle B and thus results in ill-formedness, as also indicated in (12) and (13). The sentence is grammatical as long as the principles in (11a–b) are met, whether the antecedent is the subject, as in (12)–(14), or the object, as in (15).

Again, the anaphors in (14)–(15) are bound in their vP either by the subject (14) or by the object (15), in accordance with Principle A. In contrast, the pronominals are ungrammatical because they are locally bound inside the vP, violating Principle B.Footnote ⁵

One question arising here is how binding is achieved in constructions like (14)–(15). To begin, I argue that binding in the above sentences is not a result of an Agree relation between the anaphor and its antecedent (Antonenko Reference Antonenko2012, Charnavel and Sportiche Reference Charnavel and Sportiche2016). This is substantiated by the fact that binding under agreement (probe-goal relation) cannot capture cases involving partial binding (Charnavel and Sportiche Reference Charnavel and Sportiche2016: 65), as illustrated in (16).

Supposing that there is a group of boys (A, B, C, D, etc.) then (16) could mean that A, B, and C saw A, B, C, and D, but D and E saw D, E, and F. However, Agree relations require exhaustive binding; accordingly, if binding is the result of an Agree operation, then (16) is wrongly predicted to be ungrammatical with the interpretation described. It is also not clear how binding by Agree is achieved in sentences like (17).

Assuming that taħaddaθ-a ‘talked’ takes two PP complements maʕa Nuhaa ‘with Nuha’ and ʕan nafsihaa ‘about herself’, with the former being the higher complement c-commanding the latter, it remains puzzling how Nuhaa binds the anaphor under the Agree-based approach. First, DPs are not heads and thus cannot probe features for valuation. Second, v in (17) does not assign accusative Case to either of the two DPs, which suggests that there is no agreement between v and the PP-embedded DP arguments. For more arguments against binding under Agree, see Drummond, Kush, and Hornstein (Reference Drummond, Kush, Hornstein and Boeckx2011) and Charnavel and Sportiche (Reference Charnavel and Sportiche2016). It would therefore seem that binding in (14)–(15) does not result from an Agree relation. Rather, both cases in (14)–(15), whether the anaphor is bound by the subject or object, can be accounted for by the phase-based approach, with binding established at the end the phase. Specifically, binding is taken to be a filter, evaluated when the Spell-Out domain of a certain phase is transferred to the interfaces (Antonenko Reference Antonenko2012).

Following Antonenko (Reference Antonenko2012), I also argue that the mechanisms by which subject and object binding are derived, respectively, are different in the syntax. In the spirit of Pesetsky and Torrego's (Reference Pesetsky, Torrego, Karimi, Samiian and Wilkins2007) feature-sharing model and building on Reinhart and Reuland's (Reference Reinhart and Reuland1993) view that a predicate is specified as reflexive if one of its arguments is a SELF anaphor, Antonenko contends that an anaphor enters the derivation with an uninterpretable, valued (reflexive) ρ-feature ([uρ, +val]). Since the verb is endowed with an interpretable, unvalued ρ-feature ([iρ, –val]), the ρ-feature on v/V probes the one on the anaphor for valuation purposes. In case of object binding, as in (15), ρ is assumed to be on V, and a λ-operator introduced above the V position binds the variable associated with the uρ in the anaphor. This is shown in (18), based on Antonenko (Reference Antonenko2012: 110).Footnote ⁶

1. (18)

   

Once V merges with the PP in (18), the unvalued ρ on V probes the valued ρ on the anaphor, resulting in the valuation of ρ on V. This is followed by the merger of the direct object (DO) in Spec-VP, at which point the λ-conversion produces a reading where the anaphor inside the PP is bound by the DO, as in (19), adapted from Antonenko (Reference Antonenko2012: 111).

1. (19)

   1. a. ∃e : Theme(e, DO) & DO λx V(e) & PP(e, x)

   2. b. ∃e : Theme(e, DO) & V(e) & PP(e, DO)

This process is accompanied by the valuation of the Case feature on the DO, followed by the valuation of the T-feature on v upon the merger of v and its specifier. At this point, VP is closed as it acquires its phasal status.Footnote ⁷ By comparison, in the cases of subject binding shown in (12)–(14), ρ is argued to be placed on v rather than on V. Example (14) has the structure in (20), adapted from Antonenko (Reference Antonenko2012: 108, 112).

1. (20)

   

Here, the construction of the VP proceeds in the same manner, except that the unvalued ρ does not probe until v is merged, and λ-conversion applies above v. When the phasal status of VP is acquired upon the merger of v and VP is closed, binding has still not taken place. Afterwards, v values its ρ by probing the anaphor. λ-conversion will then apply, yielding subject (Sub) binding, as shown by the derivation in (21) (adapted from Antonenko Reference Antonenko2012: 112).

1. (21)

   1. a. ∃e : Agent(e, Sub) & Sub λx Theme(e, DO) & V(e) & PP(e, x)

   2. b. ∃e : Agent(e, Sub) & λx Theme(e, DO) & V(e) & PP(e, Sub)

The main idea of Antonenko's approach is that phasal status is acquired/closed (e.g., by VP/vP) only when all features are valued, including Case, phi- and ρ-features. Importantly, if a phase is sent off to the interpretive components while ρ is still unvalued, then binding can and will apply in the higher phase.

The binding-by-phase analysis is also predicted to hold in the smallest finite TP. While a finite TP is considered to be the Spell-Out domain of a CP phase by Quicoli (Reference Quicoli2008), Despić (Reference Despić2015), and Charnavel and Sportiche (Reference Charnavel and Sportiche2016), for Antonenko (Reference Antonenko2012) the finite TP itself is a phase. Binding in both approaches is predicted to hold within TP if it passes beyond vP. The major difference between the two is that, while CP counts as a boundary with respect to binding for Quicoli, Despić, and Charnavel and Sportiche, it may not be a boundary for Antonenko. This difference arises specifically when it comes to long-distance binding, as we will see below. In Arabic, embedded CPs block binding from a higher clause, as illustrated in (23). There is no long-distance binding in (22), since the anaphors there are referentially dependent on the embedded subject -hum ‘they’.

Note that -hum ‘they’ in (22) is free in its TP, unlike the reflexive and reciprocal which are locally bound by -hum. The subordinate CP in (22), which acts as a boundary, does not have a vP projection since it is a verbless clause. Such clauses, however, do have a TP projection,Footnote ⁸ in which case the binding domain for the anaphors is the finite TP. See Charnavel and Sportiche (Reference Charnavel and Sportiche2016: 72), who argue that “a finite TP complement of the phase head C” is the Spell-Out/binding domain in English. By contrast, if such TPs contain unbound anaphors, as in (23), they are ungrammatical.

The matrix subject ar-riʤaal-u ‘the men’ in (23) cannot antecede ʔanfusahum ‘themselves’ nor baʕḍahum baʕḍ-an ‘each other’ since they are separated by the subordinate CP boundary; hence the ungrammaticality. This raises the question of how this theory can account for long-distance binding in English, as in (24).Footnote ⁹

1. (24)

   1. a. John wants [_CP [_TP himself to be elected]].

   2. b. John and Bill think [_CP that [ _TP each other's children are idiots]].

First, Quicoli (Reference Quicoli2008) accounts for sentences like (24a) by suggesting that the embedded reflexive is bound in the matrix clause (its phasal domain) which contains a governor as well as a subject (wants, John, respectively). Quicoli, following Chomsky (Reference Chomsky and Kenstowicz2001), assumes that the infinitive complement in such cases is not a CP but rather a nonfinite TP. The structure of the sentence in (24a) should therefore be as in (25).

1. (25) John₁ wants [_TP  himself₁ to be elected].

This way, himself is at the edge of TP, and is accessible to John since there is no CP blocking binding. This account, however, does not work for (24b) since the subordinate clause has a finite TP, and therefore also a CP, which is a phase boundary. Pollard and Sag (Reference Pollard and Sag1992) argue that possessive reciprocals like the one in (24b) are exempt from the binding theory. Similarly, such examples have not been explained by Quicoli (Reference Quicoli2008), Despić (Reference Despić2015), or Charnavel and Sportiche (Reference Charnavel and Sportiche2016). However, Antonenko (Reference Antonenko2012) offers an account of the subject-internal reflexives of embedded finite clauses in English, which could also explain the facts in (24b). In such long-distance binding constructions, Antonenko assumes that the ρ-feature is on the matrix V, namely, outside the embedded vP and TP phases. On this view, a feature-sharing operation between the anaphor and the matrix V is possible even after the subordinate TP is completed, since the anaphor is in Spec-TP. Accordingly, λ-conversion happens above the matrix V, ensuring that binding holds between the subject-internal anaphor inside the embedded TP and the matrix subject, as shown in (26), adapted from Antonenko (Reference Antonenko2012: 189).Footnote ¹⁰

1. (26)

   

Again, the edge of TP in (26) is a phase edge, so the anaphor is accessible by the matrix subject.Footnote ¹¹ It must be noted, however, that if the English reflexive or reciprocal is the entire subject of the subordinate TP, as in the translation of (23) above, then such binding is impossible, leading to ungrammaticality.Footnote ¹² This long-distance binding does not work for Arabic constructions like (23) and (27a), nor for reflexive/reciprocal possessives, as in (27b).

As for (23), it seems that the ρ-feature is located on the subordinate T, and that λ-conversion must apply above T, which means that the derivation will crash if no binder is found inside the embedded CP. This indeed appears to be the case. On the other hand, (27a) is unacceptable in Arabic, unlike in English. At this point, I have no satisfying answer to why the Arabic C appears to be less transparent than its English counterpart; all we can say for now is that the difference might be attributed to the position of ρ. Perhaps English allows ρ to be outside the embedded CP containing the subject-internal anaphor, but Arabic does not. Another possibility is that Arabic anaphors can only be licensed in an object position, and thus cannot surface in a subject position even if they are embedded within the subject. The reflexive and reciprocal possessives, illustrated in (27b), will be discussed in the next section.

To sum up, we have seen in this section that Arabic reflexives and reciprocals behave alike in vP and TP phases; nonetheless, they do exhibit asymmetries within possessive DPs, a topic to be explored below.

3. Binding asymmetry: DP phase

Before presenting the data and binding facts in this section, I would like to mention that the data in this paper are based on the author's intuitions. However, a grammaticality judgment test has been conducted with eight highly educated native speakers of Arabic. Six of the participants have completed a PhD, one is a PhD candidate, and one has an MA. Two are from Saudi Arabia, two from Jordan, one from Iraq, one from Palestine, one from Egypt and one from Algeria. The Iraqi participant and one of the Jordanians are professors of traditional Arabic grammar, and the former is the author of two books on classical Arabic grammar. Importantly, the participants were unanimous as to the (un)acceptability of the data in this paper. For instance, all of them found the sentences in (30) below ungrammatical, but those in (31) grammatical.

This section concentrates on reflexives and reciprocals that are not direct arguments of the verb predicate, particularly reflexive and reciprocal possessives.Footnote ¹³ Nominal possessive phrases in MSA are formed in two different ways: by attaching a possessive pronominal suffix to a noun, or by using the construct-state/synthetic construction (the combination of two nouns), as exemplified in (28) and (29), respectively.

The synthetic possessive typically consists of the possessum, followed directly by the possessor, as in (29).Footnote ¹⁴ The possessor can also be an anaphor; however, not every anaphor is permitted in this type of possessive construction. Arabic disallows reflexive possessives, but allows reciprocal possessives. Consider (30) and (31).

The sentences in (30)–(31) are asymmetrical in both Arabic and English. Reciprocals in both languages pattern with pronouns rather than with reflexives in this construction, as the examples in (28) and (31) show. According to the participants consulted, the very same linguistic phenomenon is also observed in various nonstandard Arabic dialects, as illustrated in (32)–(35).

JordanianFootnote ¹⁶ and KuwaitiFootnote ¹⁷ Arabic also show this pattern. Note that such reflexive-reciprocal asymmetry applies across the board to all the nonstandard dialects above. Crucially, if possessive pronouns are used instead of reflexives, then the sentences become grammatical, just like in MSA. Below is an illustrative example from Palestinian Arabic.

The other nonstandard dialects exhibit this behaviour as well. One notable difference between MSA and the nonstandard dialects is that the two components of the reciprocal (the range and contrast, baʕḍihim + baʕḍ-an ‘each + other’, respectively) must be used in MSA, but only one reciprocal component is used in the nonstandard dialects. Palestinian, Jordanian, and Algerian Arabic permit the use of either one or two reciprocal components, but Saudi, Kuwaiti, and Egyptian Arabic allow only one. It appears that one of the reciprocal elements in the nonstandard dialects undergoes some kind of ellipsis at some point in the syntactic derivation. Since investigating the reciprocal in the nonstandard dialects is beyond the scope of this article, I restrict my attention to the reflexive-reciprocal asymmetry in MSA. What is important to us is that all the varieties of Arabic mentioned above behave alike in allowing reciprocal possessives and disallowing reflexive possessives.Footnote ¹⁸

Returning to the data in (30)–(31), the distributional difference between Arabic reflexives and reciprocals needs to be explained. Before doing so, let us see how this asymmetry has been accounted for in English. Despić (Reference Despić2011, Reference Despić2015) argues that D is a phase head when it takes a possessive phrase (PossP) as its complement and thus constitutes a binding domain, just like vPs and CPs. This is shown in (37), from Despić (Reference Despić2015: 211).

1. (37)

   

Since possessive DPs form a binding domain, the reflexive anaphor in Spec-PossP cannot be bound in its phasal domain. This explains why reflexive possessives do not occur in English, and why the possessive must take the non-reflexive pronominal form. However, reciprocal anaphors are permitted in English possessive DP structures, as in (38a–d), from Despić (Reference Despić2015: 213).

1. (38)

   1. a.

      

   2. b. Mary's friends

   3. c. each other's friends

   4. d. *their's friends

Despić assumes that pronouns and reciprocals are licensed in different structural positions, and that possessive lexical DPs (non-pronominal DPs) and possessive reciprocals pattern together in English, in the sense that they are both licensed in a structural position higher than the position occupied by pronominal and reflexive possessors, as illustrated in (38). Since non-pronominal possessive DPs and possessive reciprocals are positioned at the leftmost edge of the DP phase – unlike reflexive possessives, which are positioned below the D head, as in (37) – reciprocals can be bound within the vP domain containing the DP. Although this analysis provides an elegant syntactic account of why reflexive possessives like *himself's son are prohibited in English as opposed to reciprocal possessives like each other's sons, it falls short of explaining the asymmetry between *the sons of himself versus the sons of each other, given that in these phrases, the two anaphors (the possessors) occupy the same structural position. The discussion below will attempt to account for the asymmetry in Arabic, and then extend the generalizations to capture some facts in English that Despić has not touched on.

I take as a point of departure Despić's (Reference Despić2011, Reference Despić2015) proposal for English reflexive and reciprocal possessives, and Chomsky's (Reference Chomsky and Kenstowicz2001, Reference Chomsky, Freidin, Otero and Zubizarreta2008) claim that the phase edge is accessible to operations in a higher phase. With these elements in hand, an account of the reflexive-reciprocal possessive asymmetry in Arabic that draws on the phasehood of possessive DPs seems promising. This is driven by the asymmetrical behaviour of possessive reflexives and reciprocals in both Arabic and English, as shown in (39) and (40).

However, there is an independent issue in Arabic that requires explanation. In Arabic possessive phrases, the possessum always precedes the possessor, as in (39), whereas in English, the reverse order is also possible, as in (40). Despić's (Reference Despić2015) proposal that the English reflexive and reciprocal in each other's friends/*themselves’ friends are generated in different base positions within the DP clearly does not apply to Arabic, nor does it explain the asymmetry in English between *the sons of themselves versus the sons of each other. The question here is how the structure in (39) can be accounted for; in other words, how can we explain the grammaticality of reciprocal possessives versus the ungrammaticality of reflexive possessives in Arabic, given that the reflexive and reciprocal appear to occupy the same syntactic position in this language? The first step in addressing this question is to look into the structure of construct states in Arabic. Previous analyses (Ritter Reference Ritter, McDonough and Plunkett1987, Reference Ritter1988, Reference Ritter and Rothstein1991; Siloni Reference Siloni, Bobaljik and Bures1991, Reference Siloni, Belletti and Rizzi1996, Reference Siloni1997; Benmamoun Reference Benmamoun2000; Longobardi Reference Longobardi, Baltin and Collins2001) have proposed that the Semitic construct states involve an additional functional category Num (number) heading NumP (number phrase), the raising of the possessor DP to Spec-NumP and the raising of N (the possessum) over Num to D, which ultimately incorporates with Agr.Footnote ¹⁹ Once Agr is morphologically supported, it assigns genitive Case to the argument in Spec-NumP under government. The Num head, first posited by Ritter (Reference Ritter and Rothstein1991), is assumed to carry the number specification of the noun (Siloni Reference Siloni1997). The derivation of construct states like (41a) is shown in (41b, c) (see Ritter Reference Ritter and Rothstein1991: 45, Siloni Reference Siloni1997: 38, Benmamoun Reference Benmamoun2000: 143).

1. (41)

   1. a. ʔabnaaʔ-a   r-raʤul-i

      sons-acc   def-man-gen

      ‘the man's sons’

   2. 

Once movement occurs, the adjacency requirement on the Case assignment of r-raʤul-i ‘the man’ is met under government (Chomsky Reference Chomsky1981, Stowell Reference Stowell1981, Borer Reference Borer1984, Ritter Reference Ritter and Rothstein1991, Hazout Reference Hazout1992, Rizzi Reference Rizzi, Belletti and Rizzi1996, Siloni Reference Siloni1997). The adjective phrase, if present, is adjoined to NP, as standardly assumed, and therefore follows the moved possessor. Fassi Fehri (Reference Fassi Fehri1993) offers a slightly different analysis of Arabic construct states, postulating PossP instead of NumP as the additional functional category between DP and NP. He proposes that the maximal projection DP expressing ownership is a possessive phrase, on a par with Despić (Reference Despić2015). However, neither of these analyses of Semitic construct-states can explain the reflexive-reciprocal possessive asymmetry if we assume that the two anaphors are treated in the same manner. One proposal that may capture the binding facts of the Arabic data relies on the movement of one of the reciprocal elements. Heim, Lasnik, and May (Reference Heim, Lasnik and May1991) argue that each in an English reciprocal construction like (42a) moves and adjoins to its antecedent phrase at LF, as the representation in (42b) indicates (see also Belletti Reference Belletti1982, Roberts Reference Roberts, Moore and Wyner1991).

The derived LF representation in (42b) is assumed to correspond to the surface form the men each saw the other, with each serving as a distributor and other designating distinctness (i.e., x saw y and y saw x). Heim, Lasnik, and May also claim that e in [e other] is an anaphor bound by each, while the reciprocal other is an instance of the pronominal other. As a whole, [e other] is an R-expression, which explains why each and e in (42b) bear the same index but [e other] bears a different index. Their hypothesis is thus that the semantics of each other is inherited from the non-reciprocal elements each and other. Crucially, the reciprocal each other is interpreted as consisting of two parts, one anaphoric and one non-anaphoric, an analysis that reflects its complexity.

In order to explain why reflexive possessives like the one in (43) are prohibited in Arabic but reciprocal and pronominal possessives like those in (44)–(45) are permitted, I adopt the copy theory of movement (Chomsky Reference Chomsky, Hale and Keyser1993, and much subsequent work, e.g., Nunes Reference Nunes1995, Bošković and Nunes Reference Bošković, Nunes, Corver and Nunes2007, Corver and Nunes Reference Corver, Nunes, Corver and Nunes2007, Stjepanović Reference Stjepanović, Corver and Nunes2007). I also assume that possessive DPs are phases (Hiraiwa Reference Hiraiwa2005; Despić Reference Despić2011, Reference Despić2015) and that the reciprocal element other is pronominal (Heim, Lasnik, and May Reference Heim, Lasnik and May1991). The pertinent data are given in (43)–(45), repeated from (30a), (31a) and (28a).

Reducing binding domains to phases, including the DP, predicts that an anaphor should be bound but a pronoun should be free in its phase. This prediction is borne out in (43) and (45), but apparently violated in (44), given that both the reflexive and the reciprocal are anaphoric. However, upon closer scrutiny of the structure of the reciprocal, we find that baʕḍihim ‘each’ in (44) may move overtly to Spec-DP or even higher, as in (46). When movement applies, a distributive reading results. That is, there are two semantic interpretations available for a sentence like (44): collective and distributive. The collective reading, corresponding to the men hugged each other's sons collectively/simultaneously, arises if the verb takes scope over the reciprocal quantifier baʕḍihim ‘each’. If, on the other hand, the quantifier has scope over the verb, as in (46), then a distributive reading arises.

This collective-distributive distinction is induced by the movement of the reciprocal quantifier as well as by scope variation (see Heim, Lasnik, and May Reference Heim, Lasnik and May1991, Roberts Reference Roberts, Moore and Wyner1991, Lasersohn Reference Lasersohn1995, Moltmann Reference Moltmann1997, Carlson Reference Carlson and Rothstein1998, Büring Reference Büring2005, among others). Now, under the copy theory of movement, multiple copies of the distributor in (44) and (46) are created during the derivation, as sketched below.

Two remarks are in order here, concerning Case and binding. In the lowest position in (47), the distributor baʕḍihim ‘some.3pl.m.gen’ has genitive Case (-i-i-), as in (43), but the higher copies are nominative (-u-u-), as in (46) and (47). This means that the higher copies have a different Case. Following Béjar and Massam's (Reference Béjar and Massam1999: 74) analysis of multiple Case checking in languages like Niuean, I assume that the lowest copy of the distributor is assigned genitive by Agr on the D head. When the distributor moves up, the higher copy enters into a Case-checking relation with a higher head (here, T), resulting in nominative Case marking. The lower Case (genitive) is still there but will not be pronounced. This way, both the lower and higher Cases are interpretable by configurationally being in a checking relation with functional heads (Agr and T, respectively), but only the higher Case is phonologically realized. This is exactly what happens in Niuean, according to Béjar and Massam (Reference Béjar and Massam1999).Footnote ²⁰ Regarding binding, it is possible for baʕḍuhum ‘each’ in (47) to be bound by the antecedent phrase and to bind its lower copy baʕḍihim₂ located below the D head, if we assume that Principle A applies cyclically at the end of each phase (Lee-Schoenfeld Reference Lee-Schoenfeld2008, Quicoli Reference Quicoli2008), and further assume that all copies must have a single antecedent because they are part of the same chain, in accordance with the Chain Uniformity Condition (Chomsky Reference Chomsky1995: 91; see also Freidin and Vergnaud Reference Freidin and Vergnaud2001, Quicoli Reference Quicoli2008). To illustrate how binding proceeds in (47), let us first consider the tree diagram in (48) and then those in (49) below.Footnote ²¹

1. (48)

   

First, the distributor baʕḍihim ‘each’ in (48) moves from Spec-PossP to Spec-DP by internal merge, leaving a copy behind. It is at this point, upon the completion of the DP phase, that Principle A first applies at LF (Lee-Schoenfeld Reference Lee-Schoenfeld2008, Quicoli Reference Quicoli2008, Charnavel and Sportiche Reference Charnavel and Sportiche2016) and the PossP complement of D undergoes Spell-Out. To be more precise, the anaphoric copy of the distributor in Spec-PossP is now properly bound in its phasal (DP) domain by the higher copy in Spec-DP, to which Principle A has not yet applied. Once the distributor occupies the edge of the DP phase, it is handed over to the vP phase and becomes accessible to the matrix subject (its antecedent phrase), in conformity with Chomsky's (Reference Chomsky and Kenstowicz2001) PIC. Whether the distributor remains and gets pronounced in Spec-DP or moves again, as shown in (49) below, its binding domain is now the vP phase, where Principle A applies again.

1. (49) vP phase

   1. 

   2. 

Thus, after moving to Spec-DP, the distributor may undergo further movement, first to Spec-VP and then to Spec-vP where it finally adjoins to its antecedent phrase, as shown in (49a–b). Notably, the binding of both of the copies in (49b) occurs within their local domain (the vP phase) since the antecedent is in Spec-vP; the cyclic application of Principle A in vP is therefore satisfied.Footnote ²² The core idea here is that the derivation crashes if Principle A is violated in any phasal domain. For it to successfully apply in a higher phase, it must first have been satisfied in the lower phase. Overall, whether the lower or the higher copy of the distributor is pronounced, binding within the A-chain proceeds cyclically at the end of each phase (DP and vP), as described above. I now turn to the second reciprocal element, baʕḍ-in ‘other’ in baʕḍihim baʕḍ-in ‘each other’ (see (47) above). Since it, like English other, has pronominal properties (Heim, Lasnik, and May Reference Heim, Lasnik and May1991), it must be free in its local domain. In (47), that domain is the DP phase. Accordingly, there are no violations of Principle A or Principle B, and the derivation converges at the PF and LF interfaces, correctly predicting that (47) is grammatical. Crucially, the D head and Spec-DP on the one hand, and the PossP on the other, are spelled out at different points in the derivation. Overall, the suggestion that each atomic element of the reciprocal inherits its semantics from its non-reciprocal usage (Heim, Lasnik, and May) appears consistent with the facts. This is because both elements can surface independently of the other in a sentence, as demonstrated below. Note that baʕḍuhum ‘each’ is always anaphoric, whether in reciprocal or non-reciprocal constructions, but baʕḍ-in ‘other’ is pronominal and hence grammatical when free, as in (50), supporting the current analysis.

Building on Antonenko (Reference Antonenko2012), it seems reasonable to assume that the ρ-feature is placed on the D head and that λ-conversion applies above D, as the structure in (51) shows.

1. (51)

   

If this is correct, then (51) will yield the semantic derivation in (52).

1. (52)

   1. a. ∃e : Agent(e, Sub) & Sub λx D(e) & PossP(e, x)

   2. b. ∃e : Agent(e, Sub) & D(e) & PossP(e, Sub)

On this view, the DP structure in (51) behaves like a full clause in that it has an anaphoric element baʕḍihim ‘each’ in Spec-PossP as well as the binder of baʕḍihim – the higher copy – in Spec-DP. That copy is itself accessible to the matrix subject in the higher phase. Moreover, ρ on D is valued through feature-sharing. Finally, baʕḍuhum ‘each’ is considered the Agent subject, since possessors actually correspond to Agents (see den Dikken Reference Den Dikken1995, Reintges and Lipták Reference Reintges, Lipták and Frascarelli2006).

To recapitulate, the copy theory of movement, the phasehood of possessive DPs, and the pronominal status of baʕḍ-in ‘other’ taken together, explain the binding facts in MSA. This type of movement is not possible for the reflexive and pronominal in such syntactic environments; therefore, the reflexive is left unbound in its phasal domain, causing ungrammaticality, while the pronominal is free, leading to grammaticality. On the whole, this analysis indeed seems to correctly explain the binding asymmetry in Arabic.

That Principle A applies successively at the end of each phase can also be shown for English, as illustrated in (53) and (54), from Quicoli (Reference Quicoli2008: 307).

Quicoli argues that himself in (54a) moves to Spec-vP in the first vP phase, leaving a copy of himself behind. It is here at the end of the first vP phase where Principle A first applies, with himself binding the lower copy. Principle A does not apply to the higher himself in (54a) at this point because the reflexive is in the subject position of the infinitive, which is not its local domain. The binding of himself by John occurs in the second vP phase in (54b) through new cyclic application of Principle A. It must be mentioned here that the syntactic derivation of the Arabic sentence in (48) differs from the derivation of the English sentence in (54), since the movement of himself in (54) is motivated by Case reasons, as suggested by a reviewer. This Case-based explanation in (54) does not work for Arabic, but nonetheless shows that binding applies cyclically.

By extending the above analysis to English, we can also capture the reflexive-reciprocal asymmetry illustrated in (55).

1. (55) The men₁ hugged the sons { of each other₁/*of themselves₁ }.

Despić's (Reference Despić2015) account cannot explain this asymmetry because both anaphors (the possessors) follow the possessum, as mentioned earlier. Under the current analysis, however, it seems reasonable to assume that the reciprocal is licensed in (55) because each moves covertly to a position outside DP, which is evident from the fact that (56a) can have either the distributive reading in which each takes scope over the verb and the hugging subevents took place in turn, or the collective reading, where each has lower scope and the hugging subevents took place simultaneously. Since it is possible for each to move as shown in (56b), binding can also be captured if the lowest each is locally bound within its DP phase by its copy in Spec-DP, and that copy is itself at the edge of the phase, accessible for higher binding.

The question is why the lowest each in (56b) is the only copy to be pronounced. This may be explained by appealing to the reasoning that pronouncing the higher copy in (56b) triggers some violation. To illustrate, let us first consider Bošković and Nunes’ (Reference Bošković, Nunes, Corver and Nunes2007: 17–18) view that pronouncing the lower copy of a chain results from PF considerations. An illustrative example of this from Romanian is given in (57).

Bošković and Nunes state that all wh-words in Romanian, unlike in English, must be moved to the left periphery and pronounced there, except in constructions involving two homophonous wh-words, in which case only one wh-word is pronounced in Spec-CP, as seen in (57). Bošković and Nunes, following Franks (Reference Franks1998), add that the lower copy ce in (57c) surfaces in order to avoid a PF violation. By a similar reasoning, we could also hypothesize that the lower copy of each in (56b) is pronounced because the phonological realization of the higher copy would lead to a violation in the sense that it allows of and other to appear in a string that is not possible in English if they are the only elements in the PP, as shown in (58).

1. (58) The men each hugged the sons [*of other].

Theoretically, there appears to be nothing wrong with the string of other except that English does not allow it if other is not followed by an NP, as seen above. This is inspired by the fact that this string is possible in Arabic, as demonstrated below.

Notice that the string li-baʕḍ-in ‘of other’ in (59) can surface in Arabic after baʕḍihim ‘each’ is moved and pronounced in a higher position. This contrasts with its English counterpart, as indicated by the English translation in (59b). Thus, it indeed seems that the possibility of pronouncing each copy in an A-chain is governed by language-specific requirements, as the above data from Romanian, English, and Arabic show.

Overall, it is the complex nature of the reciprocal that allows the English distributor each to move to, or be interpreted in, different positions, as shown in (56b). Again, this movement allows each to be properly bound in both the DP and vP phases. By contrast, the reflexive in English cannot move, and thus remains free in its phasal domain, leading to the ill-formedness of (60).

1. (60) *The men₁ hugged the sons of themselves₁.

This is also the case in Arabic, as we have seen above. Possessive DPs do in fact have peculiar properties, distinguishing them from other DPs. While there is evidence corroborating the phasehood status of Arabic DPs, as we will see in the following subsection, evidence that English DPs like the one in (56b) are phases is lacking at this point. None of the phasehood diagnostics provided by Citko (Reference Citko2014) distinguish the DP in (56b) from other English DPs, except that the DP in (56b) forms a binding domain. A full exploration of this issue must therefore await future research. The D head in possessive DPs like [_DP each other's sons], by comparison, triggers ellipsis of its NP complement sons as in John and Mary hugged Bill's sons as well as each other's sons. This confirms that such DPs are phases (Despić Reference Despić2015).

3.1 The phasehood status of Arabic possessive DPs

There is evidence that possessive DPs are phasal in Arabic. First, such complex DPs contain a possessor and a possessum, as in (61), which in a sense respectively correspond to the Agent and Theme arguments in clausal possessives (den Dikken Reference Den Dikken1995, Reintges and Lipták Reference Reintges, Lipták and Frascarelli2006).

This property is not found in other Arabic DPs, even in non-possessive construct-state DPs that are also considered complex. Consider (62).

The deverbal noun taṭwiir-a ‘improvement’ forms a construct-state DP with ʔabnaaʔ-i-him ‘their sons’. Since this DP lacks a possessor, unlike the one in (61), all anaphors are licensed in this environment but pronouns are not, as shown in (63) and (64) respectively.

One crucial difference between non-possessive construct states like (63) and (64), and their possessive counterparts, as in (61), is that the non-possessive construct-state DPs lack a subject. One natural hypothesis, raised by a reviewer, is that the DPs in (63) and (64) have a null subject binding the anaphors, along the lines of Chomsky's (Reference Chomsky1986: 167) “PRO in NP” analysis. However, there is reason to reject this hypothesis. If there were, in fact, a null subject inside the DP, the reflexive and the reciprocal in (65) would have a binder in the DP phase, wrongly predicting that (65) should be grammatical.

The ungrammaticality of (65) therefore supports the hypothesis that the actual binder in (63) and (64) is the matrix subject, not a null subject in Spec-DP.

Second, one of the diagnostics of phasehood in Citko (Reference Citko2014: 110) is that an element moving out of a phase can be pronounced at the phase edge. By applying this diagnostic to Arabic DPs, we find that the distributor moving out of a DP can be pronounced at the edge only in possessive DPs. Compare (66) to (67).

Baʕḍihim ‘some.3pl.m.gen’ in (66a) is in Spec-PossP, but in (66b) it has moved to Spec-DP, where it is realized as the nominative form baʕḍuhum ‘some.3pl.m.nom’. In contrast, baʕḍihim ‘some.3pl.m.gen’ in (67a) cannot undergo movement to Spec-DP or be pronounced there, as shown by the ungrammaticality of (67b). Arabic verbless equational sentences, which have a TP projection but no v/VP (Benmamoun Reference Benmamoun2000, Aoun, Benmamoun, and Choueiri Reference Aoun, Benmamoun and Choueiri2010), contain a subject and a non-verbal predicate. The predicate cannot move or adjoin to the subject in Spec-TP, since it is the complement that provides information about the subject. If we were to assume that baʕḍuhum ‘some.3pl.m.nom’ lands in Spec-TP by adjoining to its antecedent phrase ʔulaaʔika/hadaf-u-hum ‘those/their goal’, then we would wrongly predict both sentences to be grammatical. This provides evidence that baʕḍuhum ‘some.3pl.m.nom’ is pronounced at the edge of the predicate DP rather than in Spec-TP, and in turn supports the idea that movement out of possessive DPs proceeds through their edge in (48) above. This is in line with Citko's (Reference Citko2014: 10) proposal about DP phases.

A third diagnostic of the phasehood of a DP in Citko (Reference Citko2014: 112, 123) is whether that DP is a binding domain. Accordingly, the fact that reflexive possessives induce ungrammaticality while pronominal possessives do not, as discussed above, suggests that possessive DPs form phases. Reciprocals, unlike reflexives, are locally bound in their DP phase due to the movement of the distributor, as noted above. This pattern also holds in English possessive DPs, discussed earlier.

It is worth mentioning that Citko (Reference Citko2014) provides more diagnostics distinguishing DP phases from regular DPs, such as whether an element can be stranded at the edge of DP due to the movement of the other DP-internal elements, whether ellipsis can target the complement of D, or whether the D head is a locus of uninterpretable features. The first two diagnostics do not apply to all Arabic DPs. The third one, on the other hand, holds for all Arabic construct-state DPs, whether or not the DP involves PossP, since construct states are complex DPs with multiple functional projections, involving the valuation of features such as phi-features and Case (see, for example, Ritter Reference Ritter and Rothstein1991, Fassi Fehri Reference Fassi Fehri1999, Danon Reference Danon2011).

There is also strong cross-linguistic evidence that possessive DPs are phases. For example, in Somali and Halkomelem Salish, the sentential tense morphology is also employed in possessive phrases, which, as argued by Wiltschko (Reference Wiltschko2003), indicates that T is interpreted on D in such nominal possessive domains, as in (68b).

By the same token, Szabolcsi (Reference Szabolcsi, Kiefer and Kiss1994) shows that possessive DPs in Hungarian contain Infl, and therefore have sentence-like structures. In some other languages, as pointed out by Blake (Reference Blake1994) and Hiraiwa (Reference Hiraiwa2005), the same Case-marking is employed for both agents and possessors. One language showing this parallelism is Yup'ik Eskimo, as seen in (69), from Blake (Reference Blake1994: 151).

Similarly, Reintges and Lipták (Reference Reintges, Lipták and Frascarelli2006) point out that in Egyptian-Coptic, clausal possessives and nominal possessives have parallel structures, with the former derived from the latter via movement of both the possessor and the P head. These properties of nominal possessive phrases make it unsurprising that possessive DPs have the characteristics of phases. Accordingly, it has been hypothesized (Hiraiwa Reference Hiraiwa2005, Despić Reference Despić2011) that it is the combination of two syntactic heads, the D-Poss complex, that makes the DP a phase. This pairing corresponds to the C-T pair in CPs or the v-V pair in vPs.

In summary, I have argued above that a possessive DP in MSA constitutes a phase, and that Principle A is satisfied only when an anaphor is locally bound within its DP phase. Reflexives cannot appear in such DPs because they remain unbound, but reciprocals can appear there, due to the overt movement of the distributor, which is itself a binder that antecedes its own copy within the DP.Footnote ²⁴ Pronominals, on the other hand, are free inside their phasal domain and thus Principle B is respected. Another syntactic environment in which reciprocals can, but reflexives cannot, appear involves constructions with possessive PPs, a topic I turn to next.

4. Binding asymmetry in possessive PPs: PP phase

Reflexives and reciprocals in Arabic can both be the object of a preposition, as shown in (70) and (71).

Both types of anaphors, unlike pronouns, are licensed in (70) and (71). However, this is not the case if they are preceded by prepositions denoting possession, as in (72) and (73).

The reflexives and reciprocals in (72) and (73) exhibit asymmetrical behaviours in that reflexives cause ungrammaticality but reciprocals do not. Interestingly, Arabic possessive pronouns, like reciprocals, are licensed in such environments. Consider (74).

Two questions arise here. First, why is there a discrepancy between reflexives and reciprocals in constructions involving possessive prepositions, as in (72) and (73)? Second, what distinguishes such prepositions (Ps) from others like those in (70) and (71)?

In the discussion of reflexive and reciprocal possessives in section 3, I proposed, following Hiraiwa (Reference Hiraiwa2005) and Despić (Reference Despić2011, Reference Despić2015), among many others, that the DP qualifies as a phase, and consequently constitutes its binding domain, only when the DP contains a PossP. Following the same line of reasoning, the PPs in (70) and (71) do not contain a PossP, and accordingly, cannot qualify as phases. This explains the grammaticality of the reflexive in (70) and the reciprocal in (71), since they can both seek an antecedent outside DP, within vP (their phasal binding domain). The pronouns cannot appear in such PPs if they are not free, since this would violate Principle B. The PPs in (72) and (73), on the other hand, denote possession, which indicates that there is a maximal projection PossP embedded in the PP. This suggests that these possessive PPs also count as phases, on a par with possessive DPs. This idea is motivated by the fact that anaphors and pronouns display the same binding patterns in possessive DPs and PPs; that is, pronouns and reciprocals are licensed, but reflexives are not. A logical explanation for this distinct behaviour of reciprocals stems from the ability of the distributor to appear in a higher position, just as in possessive DPs. This is illustrated for a possessive PP in (75).

Given that multiple copies of baʕḍuhum ‘each’ are created during the course of the derivation, binding in (75) can be accounted for. As in possessive DPs, the lowest copy of the A-chain is locally bound by the intermediate copy. Assuming that Principle A applies cyclically, as previously mentioned, then the derivation in (75) will proceed as in (76):

1. (76) PP phase:

   

Again, the distributor baʕḍuhum ‘each’ at the edge of PP serves a dual purpose. First, it binds its copies within the PP, satisfying the first application of Principle A. Once the PP phase is completed, the PossP complement of P is immediately sent to Spell-Out. Second, baʕḍuhum in Spec-PP is accessible to the matrix subject (to which it optionally adjoins after further movement) for binding purposes, as shown in (77).Footnote ²⁵

1. (77) vP phase:

   1. a.

      

   2. b.

      

Since the P and its edge are spelled out along with the complement of the higher phase head (v), in conformity with the PIC, Principle A applies again within the vP phase. At that point, the antecedent phrase ar-riʤaal-u ‘the men’ in Spec-vP locally binds the three copies of baʕḍuhum ‘each’ in this phase. In principle, Antonenko's (Reference Antonenko2012) approach should also work here if we assume that the ρ-feature is on P, and that λ-conversion applies above P. Importantly, Principle A, which applies successive-cyclically, is respected here, and the derivation converges at the interfaces. The other reciprocal element baʕḍ-in ‘other’, the one fulfilling the contrast requirement, is pronominal and hence free in the PP phase, thereby satisfying Principle B. Movement then serves as an escape hatch for reciprocals, saving the structure of reciprocal possessives from violating Principle A. As before, this overt movement is not available for reflexives, which accounts for the ill-formedness of structures in (72). This analysis is thus congruent with that of possessive DPs provided in (48) above.

Possessive Ps do in fact differ from non-possessive Ps in many ways. First, Ps like ʕinda or ladaa ‘at the disposal of/in the possession of’ not only express possession but also have an adverbial (locative) function (e.g., ʕinda l-bajt-i ‘next to the house’), which makes them overlap with nouns in the sense of Bresnan (Reference Bresnan1994) and Kayne (Reference Kayne2005). Second, ʕinda and ladaa have a triliteral root (ʕ-n-d and l-d-j, respectively), as nouns do (see Ryding Reference Ryding2005). Third, possessive PPs pattern with possessive DPs with respect to binding, as demonstrated in (78) and (79).

Note that the reflexive is impossible in both possessive DPs and possessive PPs, suggesting that possessive PPs, like possessive DPs, are phase-defining (Citko Reference Citko2014).Footnote ²⁶ The licensing of reciprocals is possible because of the overt movement of the distributor, as argued above. Fourth, as shown in (80), an element moving out of a possessive PP can be pronounced at the edge of PP (which is another one of Citko's (Reference Citko2014) diagnostics for phasehood). Again, possessive PPs parallel possessive DPs.

Fifth, as pointed out by a reviewer, the fact that possessive Ps have an adverbial function suggests that they are not strictly subcategorized by the verb, and that they assign their own theta-role to the object. For Lee-Schoenfeld (Reference Lee-Schoenfeld2004, Reference Lee-Schoenfeld2008), this is a phase-defining property of PPs in German.

Non-possessive Ps, on the other hand, share no properties with nouns; they do not, for instance, have a triliteral root. Lacking an adverbial function, they are subcategorized by the verb, and thus assign a theta-role to their object through the verb, not on their own. Additionally, no reflexive-reciprocal asymmetry is observed in PPs headed by non-possessive Ps, in contrast to possessive DPs or PPs, as shown in (81).

Given that possessive Ps, unlike non-possessive Ps, display nominal-like features, it is not surprising that possessive Ps behave like nouns as far as binding is concerned. The inability of an element moving out of a non-possessive PP to be pronounced at the edge of the PP, as in (82), is yet another important characteristic distinguishing non-possessive from possessive PPs.

Since the non-possessive PP in (82) is not a phase, baʕḍuhum ‘each’ cannot be pronounced in Spec-PP, unlike in (80) above. This is shown in (82b). It is worth mentioning that the distributor can precede, and take scope over, the verb in constructions involving either possessive or non-possessive PPs, but only in possessive PPs can the distributor be phonologically realized at the edge of PP.

Finally, and importantly, the possessive PPs discussed above also seem to contain both the possessor and possessum. Unlike possessive DPs, which display these two elements overtly, as demonstrated in the previous section, possessive PPs seem to have an overt possessor, but some kind of implied possessum. Put differently, the PP in (83) implies that the man has something, though unknown, so the possessum is implicit, as indicated by the English translation.

If that is correct, which seems so, then the PP in (83) should be aligned with possessive DPs with respect to its phasehood status. This same pattern is found in English, where reciprocals but not reflexives are licensed in the PP at one's disposal, as in at each other's disposal versus *at themselves’ disposal, just as in possessive DPs (each other's books versus *themselves’ books). In addition, the idea that PPs exhibit phasal properties is attested in other languages. Bošković (Reference Bošković2014), for example, argues that PPs are phases in both Serbo-Croatian and English. Similarly, Lee-Schoenfeld (Reference Lee-Schoenfeld2008) suggests that in German, PPs form phases if headed by Ps that assign their own θ-roles to their DP complements (see also Lee-Schoenfeld Reference Lee-Schoenfeld2004). This is also true of Arabic PPs.