In some Arabic dialects, preverbal coordinated subjects cause plural agreement on the verb while postverbal ones cause either plural agreement or singular agreement. This paradigm has been addressed by Aoun, Benmamoun, and Sportiche (1994, 1999) and Munn (1999), with varying degrees of success. This article offers an alternative to the previous analyses that utilizes the concept of decomposed Merge (Hornstein 2009), whereby Merge is reanalyzed as two suboperations. Previously unexplained cases that flout the paradigm are explained here by a decomposition of the Extension Condition (Chomsky 1995) and a derivational account of pronoun binding across coordination.
There is a long-standing dispute over Arabic conjunct-sensitive agreement. The two approaches discussed here cover large swaths of the empirical landscape, yet some data still evade explanation. In this article, I offer an account that handles the basic facts and captures recalcitrant cases. Further, I argue for the decomposition of Merge into two suboperations (Hornstein 2009) and pare down the Extension Condition (Chomsky 1995).
The approach offered here is that coordinated subjects can optionally undergo the operation Label when merged together. When Label takes place, plural agreement is effected. When Label does not take place, only one conjunct can be agreed with, effecting singular agreement. Only objects that have undergone Label function as constituents (following Hornstein 2009); and as such, only they can move to preverbal positions. This correctly predicts that preverbal coordinated subjects must effect plural agreement whereas postverbal ones can effect either plural or singular agreement.
In section 1, I discuss the biclausal approach proposed by Aoun, Benmamoun, and Sportiche (1994, 1999) and the monoclausal approach proposed by Munn (1999). In section 2, I revisit the motivation behind the decomposition of Merge and extend this argument to coordination (following Larson 2010). In section 3, I offer a decomposed Merge analysis for the basic paradigm and show that it covers more empirical ground than the previous approaches. In section 4, I address the recalcitrant data that neither previous approach can account for: coordinated subjects in which one of the conjuncts is a quantified noun phrase and the other has a bound pronoun. The new account captures these via the decomposition of Merge coupled with theories of Arabic quantified NPs from Benmamoun 1999 and Mohammad 1988. Further, I argue that quantifier-variable binding across coordination is subject to derivational constraints.
Arabic conjunct-sensitive agreement is exemplified in (1), from Aoun, Benmamoun, and Sportiche 1999:669.1 This sentence has two coordinated subjects, yet only singular agreement appears on the verb. This contrasts with (2), from Aoun, Benmamoun, and Sportiche 1999:678, in which a normal plural noun effects plural agreement on the verb.2
1.1 Advantages of the Biclausal Analysis
According to Aoun, Benmamoun, and Sportiche,3 we have herein a suitable test of their approach. Sentences like (4) in which the subject is syntactically plural at every stage of the derivation should be acceptable with elements that require plural subjects. Sentences like (1) in which the subject is only superficially plural should not be acceptable under the same conditions. Reciprocals require plural subjects, and the prediction is borne out, as (5) and (6) illustrate.
1.2 Disadvantages of the Biclausal Analysis
The biclausal analysis is relatively unconstrained and risks overgeneration. It indeed seems to make false predictions of acceptability, as (7) illustrates.
This sentence shows that, unlike postverbal coordinated subjects, preverbal ones generally obligatorily show plural agreement. Aoun, Benmamoun, and Sportiche (1999:678) admit that they have ‘‘no explanation for why first conjunct agreement is not systematically possible in the SV order.’’
Also unexplained in their account is the fact that subject-initial Arabic conjunct-sensitive agreement is occasionally acceptable, albeit in a severely constrained set of circumstances. As (8) shows, when the first conjunct is a quantified noun phrase and the second conjunct contains a pronoun bound by that first conjunct, singular agreement is possible.4
1.3 Advantages of the Monoclausal Analysis
Munn’s (1999) monoclausal analysis accounts for the paradigm by distinguishing semantic from syntactic plurality. Doing so is fairly straightforward. For example, the English noun group effects singular agreement but, being necessarily composed of multiple entities, is semantically plural. Compare this with the English noun scissors, which effects plural agreement yet is semantically singular. This distinction can be seen in (9a–d), modified from Munn’s examples (1999:645, (12)).
The group was wearing different hats.
The men were wearing different hats.
*The man was wearing a different hat.
*The scissors were different colors.
With the relevant, non-discourse-linked interpretation of different, we clearly see a dissociation of semantic and syntactic plurality. The acceptability of the sentences hinges not upon syntactic plurality (see the differing agreement on the auxiliaries) but upon whether the subjects are semantically multiple.
Munn shows that the reverse is also the case. There are elements that require syntactic plurality, independent of semantic plurality. As (10a–d) show, anaphors must agree with their controllers in their syntactic plurality. For example, though the group is semantically plural, its semantic plurality is insufficient to license the plural reflexive.
*The group is keeping themselves in shape.
The group is keeping itself in shape.
The scissors are by themselves on the table.
*The scissors are by itself on the table.
Munn argues that the coordinated subjects in Arabic are like the group: semantically plural, syntactically singular. The distinction explains the unacceptability of (6). In Arabic, ‘each other’ requires syntactic plurality; semantic plurality is insufficient.
Munn’s analysis argues that agreement may be mediated through what he dubs exceptional government: the relation between a head and its complement’s specifier (akin to the notion of agreement under government found in Mohammad 1988, Benmamoun 1992, and Bahloul and Harbert 1993). Assuming that agreement is mediated through this relation and assuming Munn’s (1993) analysis of coordination as adjunction, we are able to derive sentences in which postverbal coordinated subjects cause singular agreement. The sentence in (1) would have a (simplified) structure like (11).
Munn suggests that this configuration might straightforwardly entail plural agreement—in other words, the fact that both the first conjunct DP and the &P are in the specifier position might be what requires plural agreement in these cases.
These differential agreement mechanisms account for the different agreement patterns. Munn’s analysis predicts that preverbal subjects effect plural agreement and postverbal ones, singular agreement.
1.4 Disadvantages of the Monoclausal Analysis
The monoclausal analysis cannot readily explain the fact that plural agreement is an option with postverbal subjects. Munn (1999:660) in fact suggests that this option may be due to some prescriptive overgeneralization. Ignoring the conceptual disquiet that such an idea stirs, to the extent that a more formalized explanation can be posited, it should be preferred for at least being more easily falsified.
A more interesting disadvantage of this account concerns the collective or distributive readings of coordinated subjects. Aoun, Benmamoun, and Sportiche (1999) note that with coordination of proper names, there is an ambiguity. Example (13) can mean that Alya and Marwaan read a single story (a collective reading). It can also mean that they each read a story, for a total of two stories read (a distributive reading).5
Yet, when a quantified noun phrase is coordinated with a noun phrase containing a bound pronoun, the resulting sentence is unambiguous. It only allows a collective reading. Sentence (14) only has a reading in which each woman-child pair read one story, not two.
The two accounts surveyed here cover a great deal of the data. However, they have two central failings. First, neither account handles the full range of agreement patterns. Aoun, Benmamoun, and Sportiche’s account handles the postverbal agreement facts well, but does not predict the lack of variability in agreement with preverbal subjects. Munn’s account captures the postverbal subject’s singular agreement and the invariability of agreement with preverbal subjects, but does not predict variability with postverbal ones. This is shown in (15), where a check mark indicates that the analysis can handle the relevant data.
Second, neither analysis can account for the data involving quantified noun phrases. Aoun, Benmamoun, and Sportiche’s analysis cannot predict the fact that these allow singular agreement in a preverbal position. Munn’s analysis cannot predict the differences in ambiguity between the cases with quantified and nonquantified noun phrases. In what follows, I present a unified account that handles the full range of data.
2 Decomposed Merge
Here, I revisit the motivation for decomposing Merge (Chomsky 1995). For the sake of space, this is but a sketch of the argumentation in Hornstein 2009. The main motivation comes from the differential targeting of adjuncts. Under bare phrase structure (BPS) (Chomsky 1995), accounting for adjuncts becomes difficult; the decomposition of Merge is an attempt to correct this. Coordination is taken to be adjunction and should be amenable to a decomposed Merge account. I pursue this line of argument with the aim of applying to it Arabic conjunct-sensitive agreement.
2.1 Bare Phrase Structure
Before BPS, there existed categories such as X0, X′, and XP; these had fixed positions, unchanged throughout the derivation ( Jackendoff 1977). Under BPS, categorial labels are no longer extrinsic entities with rigid positions along a derived skeleton. Rather, they are mere clarificational substitutes for lexical items. Instead of (16), we now have (17).
What were nonlexical entities are now lexical ones. In addition to this, the notions of minimal, intermediate, and maximal projection have been relativized. In (16), the X′ level was, and always will be, an intermediate projection. In (17), the middle saw is an intermediate projection, but this was not always so. Before Ivan was merged, saw was a maximal projection. It was the highest projection of saw and as such, maximal.
Given that these terms are now relational, it becomes clear that there can only be one maximal projection of a given head. This causes a problem with adjunction.
2.2.1 A Labeling Problem
Hornstein (2009) presents the following conundrum. Under BPS, there can be only one maximal projection per head. Before BPS, this was not the case; in particular, adjunction extended the tree but did not change the bar-level information. As (18) shows, an adjunct could adjoin to a VP, and the label dominating that structure would in turn be another VP. This was advantageous. Certain operations—say, VP-ellipsis—only work on maximal projections. For the structure in (18), VP-ellipsis can operate on the inner (19a) or outer (19b) VP.
(18) Iris [VP[VP felt good] on Sunday].
(19) Iris felt good on Sunday . . .
. . . and Ivan did on Saturday.
. . . and Ivan did, too.
But under BPS, we can no longer capture these facts. What is considered a maximal projection is now relative and not inherent to any node. As a result, the structure in (18) has only one maximal projection, the outer VP. We no longer have a means of operating on the VP to the exclusion of the adjunct.6
2.2.2 Decomposed Merge
To solve this dilemma, Hornstein (2009) proposes decomposing the Merge operation.7 As construed in Chomsky 1995, Merge takes two syntactic elements and combines them, projecting one of them as the label of said combination (20).
Hornstein instead proposes that the operation in (20) be split into two: Concatenate (21) and Label (22). Concatenate takes two atomic syntactic units and combines them into a complex of atomic units. Label makes that complex atomic itself by choosing one of the elements of the Concatenate operation to serve as the label of the complex.
According to the theory, normally both of these operations are carried out; but with adjunction this is not the case. Adjuncts, not being necessary to the derivation, do not necessarily have to undergo Label.
This decomposition allows for an elegant account of the differential behavior of adverbial modification. When an adverb concatenates with a verb and does not project, as depicted in (23), the verb+adverb complex is, in Hornstein’s (2009:62) words, ‘‘invisible’’ to the rest of the structure.8 So when an operation like VP-deletion targets a VP with a concatenated adverb, the VP deletes, leaving the adverb behind as in (24).
(24) Ivan ran slowly and Iris did quickly.
(26) Ivan ran slowly and Iris did, too.
As shown above, adjunction can be wedded to BPS in an elegant fashion. But more than just adverbs have been argued to be adjuncts. For instance, Munn (1993) argues that coordination is also an adjunction structure, as schematized in (27).
In Larson 2010, I argue for a similar tack with respect to coordination. In (28), it is possible to target both the topmost VP for deletion, and a lower one. And in (29), it seems that the anaphor can be bound by either the topmost DP or a lower one.
(28) Ivan [VP[VP ate an apple] and wrote a letter] in the park . . .
. . . while Ivy [did] in the library.
. . . while Ivy [did] and read a book.
I showed [DP the man] and the woman to [himself] and herself in the pond.
Ivan showed [DP the man and the woman] to [themselves] in the pond.
If only one XP is allowed per projection, there must be some other means of capturing these facts. The decomposition of Merge seems to be sufficient. We simply have structural ambiguity with coordination, just as with traditional adjunction. That is, for (29a), the structure of the coordination is as shown in (30). For (29b), the structure is as shown in (31).
In the following section, I extend this analysis to Arabic coordinated subjects.
3 Bare Phrase Arabic Conjunct-Sensitive Agreement
There are essentially three versions of coordinated subject agreement in Arabic: preverbal with plural agreement (32), postverbal with singular agreement (33), and postverbal with plural agreement (33). In this section, I show how my approach accounts for these.
3.1 Postverbal, Singular
Given the decomposition of Merge and its relation to adjuncts, we now have two ways to compose coordinated subjects. In this section, we will consider coordinated elements in which only Concatenate has applied, as in (34).
Say that (34) was the coordination of Omar and Karim from (33) with singular agreement. In this structure, Omar is singular and as such should precipitate singular agreement on any verb for which it serves as subject. As a DP, Omar can also be targeted as an external argument. However, the complex Omar w Karim is not a targetable atomic entity and cannot combine with a verb as an external argument.9
Another difference between the DP+&P concatenate Omar w Karim and a regular constituent is that the DP does not c-command the &P. There is no branching node dominating DP that dominates &P. Interface conditions not requiring c-command should have no problem accepting the DP+&P concatenate. It only lacks the ability to be targeted as a unit. Conditions requiring c-command cannot be satisfied by a structure like (34). The Linear Correspondence Axiom (Kayne 1994) requires this, and at PF such structures must have been labeled in order to be interpreted. I argue below for optional late labeling.
Given these notions concerning the DP+&P concatenate, we build the structure in (35) by merging (Concatenate and Label) [DPOmar] as the argument to a V. The adjunct phrase, having already undergone Concatenate with the DP, will of course maintain that relationship. The DP Karim will, by hypothesis, receive the same thematic role as the DP that its host &P concatenated with.
A T head will eventually be introduced into the derivation, and the V head will move to that higher position. Following proposals in Pesetsky and Torrego 2004, the T head will scan its ccommand domain and find only [DPOmar] to agree with. This will allow singular agreement to arise on the verb, as in (33).
3.2 Pre- and Postverbal, Plural
Just as it was possible for the coordinated subject to enter into the derivation without having undergone Label, so too it can merge with the V, having undergone both Concatenate and Label. The latter will work like traditional coordination and thus spur plural agreement. In effect, we have explained the fact that postverbal coordinated subjects in Arabic can cause either plural or singular agreement on the verb. Recall that this optionality is something that Aoun, Benmamoun, and Sportiche’s analysis cannot predict.
The initial position of the labeled coordinated subject does not need to be its final one. As an atomic element and a maximal projection, it is a prospective target for a movement operation. Arabic is such that its subjects sometimes do move to Spec,TP, and we thus expect that coordinated subjects can do so as well, as long as they have undergone Label.10 In other words, if a sentence with a coordinated subject has subject-verb order, then it is necessary that the coordinated subject was labeled and thus spurred plural agreement. Recall that Munn’s analysis has no way of predicting this lack of optionality in subject-verb constructions. Under the analysis proposed here, the restriction follows from independent constraints on syntactic operations.
Compared with the previous analyses, the decomposed Merge analysis straightforwardly predicts that agreement varies when the subject is postverbal but is constrained when the subject is preverbal. Expanding the table in (15) shows this graphically.
We can now also explain the reciprocal facts discussed earlier. In (5) and (6), we saw that only plural subjects license reciprocal objects. That is, only coordinated subjects that have undergone Label can fully c-command (and thus license) reciprocals.11 This extends to other instances in which plural agreement correlates with acceptability. Aoun, Benmamoun, and Sportiche (1999) show that ‘‘plurality-seeking’’ elements (in the sense of Schwarzschild 1996:10) like meet (when intransitive) and modifiers like ‘together’ require plural (in this case, labeled) subjects.
4 Extending the Decomposed Merge Analysis
In this section, I show that the decomposed Merge analysis handles the intransigent data facing the other accounts. In doing so, I further articulate the theory of decomposed Merge.
In discussing the disadvantages of the biclausal analysis, I noted that a certain type of preverbal coordinated subject can effect singular agreement: namely, a universally quantified noun phrase coordinated with a noun phrase that contains a bound pronoun. Given the discussion above, how can we account for this fact?
For quantifier-variable binding to work, the quantifier phrase must c-command the pronoun. Under the style of coordination assumed here, this is not difficult. This is shown in (38) (for convenience, English words are used in Arabic structures).
But if Label had not taken place, there would be no c-command relation between quantifier and pronoun, and the sentence would be ungrammatical. If it is necessary for Label to take place in this instance, we fail to predict the variability in agreement. Only plural agreement should be possible, pre- and postverbally. But a deeper look at the way quantified noun phrases work in Arabic will save us from this unwanted conclusion.
Benmamoun (1999) argues for an analysis of Arabic quantified noun phrases in which the nominal begins the derivation as specifier to a quantified noun phrase (see ‘boy’ in (39)). The QP head (‘every’ in (39)) head-moves to the commanding D head position.
Deriving possessive phrases in Arabic involves a very similar operation. Ritter (1987, 1991) and Mohammad (1988) derive possessive phrases like (40) from a structure in which the possessor is the specifier to a noun phrase that the possessee heads, as shown in (41).
‘the student’s book’
The DP necessarily undergoes Label so as to be able to create a c-command relation between the conjuncts and license the binding. As stated earlier, this will force plural agreement across the board.
However, this is not the only possible derivation for the subject in (37). Instead of DPs, NPs could be coordinated. That is, the NP in (41) could coordinate with the specifier NP in (39). This is shown in (43) before head movement of the Q to D and cliticization of the possessive pronoun to its host NP. In this case, the NP must undergo Label after concatenating with the &P, for the same reason as above. This time, however, after the quantifier moves to head the ‘‘matrix’’ DP, it is only a single DP that merges as a subject into the overall sentence structure and effects singular agreement. This derivation is a possibility regardless of whether the subject is pre- or postverbal.
Note that this structure will not only effect singular agreement, it will also account for the restrictions in agreement with normal coordinated subjects and preclude the licensing of reciprocals and ‘‘plurality-seeking’’ adverbials like ‘together’. Structures like (43) are not coordinated DPs; rather, they are single DPs headed by a quantifier that involve coordinated inner NPs. As such, they are expected to pattern like regular DPs headed by ‘every’. That is, independent of their collective readings, subjects like the one in (44) will not license elements like ‘together’. They fail to do so for the same reason that the subject in (45) fails to do so: there is no plural item there to license ‘together’.
In this section, I have shown that the decomposed Merge approach can account not only for the restrictions on agreement with normal coordinated subjects, but also for the lack of the same restrictions when the subject has a quantifier as one of its components.
4.2 Ambiguous Readings
In discussing the disadvantages of Munn’s analysis, I noted that normal sentences with coordinated subjects are ambiguous. No matter the verb/subject order or agreement, these sentences allow both collective readings and distributive ones. That is, (46) can mean either that Alya and Marwaan read two stories (one each) or that they read one story (together).
We can capture this ambiguity with the tool developed here: differential application of Label. Assume that labeled coordinations invoke collective readings and unlabeled ones, distributive readings. Labeling collectivizes the DP, and going unlabeled effects a distributive reading. This agrees in spirit with Kratzer’s (2007) proposal that the interpretation of collectivity/distributivity is determined by the plurality of DPs. Kratzer proposes that DPs have features that must vacate the DP in order to be interpreted. These features move to their sister’s head, which in the case of external arguments would be the verb. I posit that a labeled coordinated DP has a feature that marks the verb, and that this mark effects a collective reading (similar to what Kratzer’s proposal does for coordinated DPs).12 Structures like (47) lack this feature. This effects a distributive interpretation: Alya read a story and Marwaan read a story.
With distributive interpretations, sentences with structures like (47) would be true if there were a total of two stories read. But they would also be true if Alya and Marwaan read only one story (the same story), and read it separately. In short, the distributive reading can be true in the two situations yet only have a single reading (see Pietroski and Hornstein 2002 for a similar point). That is, coordinated subjects causing singular agreement are merely functionally ambiguous; they have only a distributive syntax and consequently a distributive logical form. That the collective reading can be inferred from the distributive one obscures the lack of structural ambiguity.
This same idea can capture the apparent tension in this analysis with respect to singular agreement and collectivity. Elements requiring plurality like ‘together’ are not allowed with singular agreement. However, ‘together’ usually tracks collective interpretations, which are possible with singular agreement. If it is the case that collective agreement interpretations with singular agreement only come about via entailment and that there is in fact no collectivity encoded in the syntax or semantics, then it is predicted that elements like ‘together’ and ‘each other’ will be illicit in ostensibly collective environments. This prediction is borne out, as (48) shows.
In contrast, coordinated subjects that cause plural agreement are actually structurally ambiguous generally. One-half of the ambiguity is easy enough to explain. The subject could have merged with the V, having already undergone Label and thus forcing both plural agreement and a collective reading. This is the only reading available, since collective readings do not entail distributive ones.
The subject could also have merged without having undergone Label. This would allow for the distributive reading—but what about the plural agreement? In the following section, I examine how the plural agreement could be derived.
4.3 Decomposing the Extension Condition
At first glance, once the structure in (49) has been built, it cannot proceed to the structure in (50) without violating Chomsky’s (1995) Extension Condition. Structure building is not applying to the root. The D head ‘Omar’ is projecting as the label of the coordination, having already merged with the verb. But we will see that this sort of operation should be allowed in a system with Concatenate and Label as operations.
In Chomsky’s (1995:189) original formulation of the Extension Condition, he couches it as a generalized transformation. To paraphrase: take phrase marker K and add Ø (which must be external to K); then substitute α for Ø, forming a new phrase marker K*.
This process, viewed through the lens of decomposed Merge, reveals a pivot point. After the substitution operation, Chomsky assumes that a new phrase marker labeled K* emerges. This emergence—that is, this occurrence of Label—is neither necessary nor necessarily immediate. Given the above conceptualization, Label does not necessarily take place directly after the substitution operation to avoid violating the Extension Condition. Once ∅ (which is external to K) has been added, the Extension Condition has been obeyed. Any further ( potentially optional) steps in the process are formally independent of the fundamental extensional aspect of the Extension Condition. Labeling is one of these further steps.
We have now whittled down the Extension Condition to merely apply to Add and Substitute. These in turn are equivalent to Concatenate. We have reduced the Extension Condition to its minimal parts, and Label is not one of them.
Given our new understanding of the Extension Condition, we can explain how the derivation can licitly proceed from (49) to (50). Once the verb and the DP have merged, the distributive reading is possible. The verb (or whatever analogous object determines thematic roles) registers this initial merger. Following Kratzer (2007), the DP ‘‘releases’’ features determining collectivity/ distributivity. Merge the singular ‘Omar’ with the verb, and a feature that effects a distributive reading is released onto the verb; merge the plural ‘Omar and Karim’, and a collective reading is released.
If it is assumed that the coordinated subject is introduced into the specifier position of a phase head (in the sense of Chomsky 2001 and subsequent work), the phasal projection can undergo Spell-Out and the verb can maintain its collective/distributive reading. If the degree of opacity (the strength of the Phase Impenetrability Condition of Chomsky 2001) after Spell-Out of the phase is sufficient to preclude any overwriting of the collective/distributive reading on the verb, then the DP+&P complex can undergo Label and become available for plural agreement with the T head once it is merged into the structure. With cyclic Spell-Out, the phase in which collective/distributive readings are determined can see the subject as it appears in (49), while the phase in which agreement is determined can see the subject as it appears in (50). We can thus explain why postverbal coordinations with plural agreement are ambiguous. The subject can also move to a preverbal position, and thus preverbal coordinations are predicted to be ambiguous.
4.4 Quantifiers Redux
As we saw above, regular coordinated subjects are ambiguous with respect to distributivity and collectivity, whereas the quantified subjects are unambiguous. Neither the biclausal nor the monoclausal analysis accounts for this. But the decomposed Merge analysis suggests a simple account. Pronouns bound by quantifiers must be c-commanded by them as a derivational condition. In order to effect c-command, the coordination must undergo Label immediately and in turn cause collective readings.
Contra proposals such as those found in Chomsky 1993, Fox and Nissenbaum 2004, and Lebeaux 2009, there are alternative arguments that binding conditions can be reduced to syntaxinternal principles as opposed to LF interface ones (Lidz and Idsardi 1998, Hornstein 2001, Reuland 2001, Kayne 2002, Zwart 2002, Hicks 2008, Takano 2010, Drummond, Kush, and Hornstein 2011). According to these, the relevant condition on binding must hold throughout the derivation. At no applicable time in the derivation of a grammatical sentence is the condition violated.
The derivational approach to binding that I adopt is that at some point in the derivation, coreferential elements are sisters to each other. As in (51), in the course of the derivation α will move and serve as the antecedent to β. The binding constraints are then predicted to be the same as the movement restrictions on α (roughly as in Kayne 2002).
(51) [α β] → [αi [γ [ . . . ti β . . . ]
The restrictions on quantifier-pronoun binding across coordination can similarly be accounted for in a manner that tracks movement restrictions. The derivation of (52) involves a stage where every boy and his are sisters, as in (53). As the configuration in (53) is not yet coordinated with anything, it should not be subject to the Coordinate Structure Constraint. Movement of every boy to the root is allowed, as in (54).
(52) every boy and his dog
(53) [& [[every boy his] dog]]
(54) [[every boy]i [& [[ti his] dog]]]
However, if instead of the initial movement of every boy a different noun were introduced, the full coordinate structure in (55) would be created. Movement of every boy should no longer be possible, and binding should not occur across nonadjacent conjuncts. Indeed, the pronoun in (56) does not have the bound reading.
(55) [& [the instructor [& [[every boy his] dog]]]]
(56) every boy (and) the instructor and his dog
It is possible to give a derivational account of the restrictions on this particular binding configuration. The movement must result in a c-command relation between the moved element and the pronoun. When a pronoun in the second conjunct is not c-commanded by the relevant quantified NP, the sentence lacks a bound pronoun reading.
*[A lunch lady from [every school]i] and itsi janitor attended.
*[[Every presidenti’s] daughter] and hisi wife attended.
If the antecedent must c-command the pronoun for cross-coordination binding purposes and if this is part of a derivational constraint, then it must be that the Label operation occurs as soon as possible in these cases. This binding requires c-command, and the c-command in turn requires Label. The fact that Label takes place immediately here has repercussions for meaning. We will never have a situation like (49) and thus will never have the possibility of distributive readings with this sort of subject. The Label operation always occurs prior to merger of the coordinated subject and verb. From this we can see that the only grammatical sentences with this sort of coordinated subject must have collective readings. They can never avoid Label and in turn distributivity.
This makes a prediction. If a language does not have a c-command condition on the binding of pronouns by quantifier expressions across coordination, then it should allow ambiguous collective/distributive readings of those constructions (and vice versa). If a language has such a c-command condition, then it will only allow collective readings.
This prediction is borne out. LA and MA only allow collective readings in the relevant constructions. They also require quantified expressions to c-command their coindexed pronouns across coordination. In (58a–b), the relevant pronoun is feminine and the only c-commanding antecedent is masculine. Native speakers find that the feminine pronoun necessarily refers to a third party female and cannot be bound by the quantified expression, which is not in a c-commanding position.
However, a reviewer notes that the analogous sentence in Egyptian Arabic does not require c-command.
As is predicted, for the reviewer the equivalent of ‘Every man and his son got-in.SG’ is ambiguous between collective and distributive readings in EA. That is, c-command is not an obligatory condition on the binding of pronouns by quantified expressions across coordination in that dialect. Without the need for c-command, Label should be optional here; and if Label is optional, we expect ambiguity.
I have argued that previous accounts of Arabic conjunct-sensitive agreement are inadequate. Instead, I have offered an analysis that explains the data more completely by means of decomposed Merge. Further, a new conception of the Extension Condition emerges. Labeling is not relevant to whether an operation obeys the condition. Instead, it is only important that the root is targeted for the introduction of a new element; other operations can apply to nonroots, only appearing to do so countercyclically.
I would like to thank Norbert Hornstein for his helpful comments on an earlier draft of this article. The article was also made vastly better by the insightful comments and questions of two anonymous LI reviewers. Novel judgments and confirmation of previous ones are presented here thanks to Abdelaziz Boudlal, Mossaab Bagdouri, and Mohamed Yeou for Moroccan Arabic and Lina Choueiri, Ghada Khattab, and Rania Habib for Lebanese Arabic.
2 For simplicity’s sake, the generic use of Arabic in this article and in the title refers to Moroccan Arabic and Lebanese Arabic. The examples are mostly from Moroccan Arabic (MA), but the generalizations extracted from the data carry over to Lebanese Arabic (LA). I note the few examples for which the speakers differ.
4 Aoun, Benmamoun, and Sportiche, an anonymous LI reviewer, and my informants note that singular agreement in these instances correlates with an intonation break between the conjuncts. Another reviewer finds example (8) ungrammatical. I do not have an account for these facts, though I do address the difference between plural and singular agreement in these instances in a later section.
5 The interpretations here hold no matter how word order and plurality marking are permuted. It should be noted that one Lebanese Arabic speaker found singular marking to correlate unambiguously with a distributive reading. I have no account for this fact. My Moroccan Arabic–speaking informants uniformly found the sentences ambiguous.
6 See Hornstein 2009 for arguments against Chomsky’s reformulation of adjunction as pair Merge, which could potentially avoid this problem.
8 Take the dashed line to indicate Concatenate plus Label. Note that there is no c-command relation between the two atomic elements.
9 The question understandably arises as to exactly what kind of syntactic object the DP + &P concatenate is. The only real distinction between it and a traditional constituent is that, being unlabeled, it cannot be targeted as a single entity or unit. Its constituent parts, the two entities concatenated, can indeed be targeted, but since the grammar can only manipulate constituents, no operation can address these entities together.
10 The analysis here crucially relies on syntactic movement of the subject from a VP-internal position to a VP-external one. This is an assumption that is not without controversy. Demirdache (1989) argues (as do Fassi Fehri (1993) and Soltan (2006, 2007)) that subject-verb word order in Arabic is derived via base generation of the subject in the left periphery. I follow Aoun, Benmamoun, and Sportiche (as well as Tucker 2007) in assuming that the subject moves to Spec,TP from a lower, VP-internal position.
11Harbert and Bahloul (2002) note that this does not extend to Standard Arabic. There, singular agreement with reciprocals is licit. The analysis offered here cannot account for this. See Soltan 2006, 2007 for a good account for this and other Standard Arabic agreement facts. Soltan’s approach is similar to the present one. Soltan argues for optional late merger of coordinated subjects qua adjuncts. However, this approach cannot (among other things) account for the differences in agreement with quantifiers discussed here.
12Kratzer’s (2007) system is considerably more complex than this, but the idea that the nature of the arguments determines semantic features of the verb is maintained here.