2 Theoretical framework and the notion of ‘a record’

Given that the current paper uses the Minimalist Program as a theoretical construct to analyze the relevant data, we provide here the relevant assumptions of the Minimalist Program we follow. We also explain the notion of a Record and its motivation.

In the Minimalist Program, heads and lexical items may enter the derivation endowed with (i.e. bearing from lexicon) uninterpretable, unvalued features, such as Case and $\unicode[STIX]{x1D719}$ -features. Such features should be valued (i.e. lexically specified) and deleted before the sentence derivation reaches the two interface levels of PF and LF (where the derivation is interpreted). This condition on valuation and deletion of uninterpretable, unvalued features is forced by the so-called Full Interpretation, a principle that demands nothing but interpretable elements at the two interfaces (see Chomsky Reference Chomsky1986, Reference Chomsky1995: Chapter 4). Valuation of such features is carried out by the Agree operation which ‘establishes a relation (agreement, Case checking) between an LI [lexical item] $\unicode[STIX]{x1D6FC}$ and a feature F in some restricted search space (its domain)’ (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000: 101). A feature set that starts the Agree operation is referred to as a probe, which seeks to establish a relation with another set of ‘matching’ features, called the goal (Fuß Reference Fuß2005: 25). Carstens (Reference Carstens2000: 350ff.) provides the following widely-accepted reformulation of Chomsky’s (Reference Chomsky, Martin, Michaels and Uriagereka2000) definition of Agree:

Through the Agree operation, an agreement relation between two elements can be established at a distance. Consider (2) as an example of the Agree operation.

The probe were (whose underspecified form bears an unvalued Number feature) agrees in Number with the post-verbal subject several books, the goal here, which in turn carries a valued Number feature.

Additionally, under recent assumptions of the Minimalist Program, elements leave their canonical position due to the Move operation, which is a combination of the operations Agree and Merge (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000), i.e. Move to occur there should be an Agree relation between a head (i.e. a probe) whose Spec is the target of the Move operation and an XP element which is attracted by the EPP feature on the head itself to move to its Spec (see Roberts Reference Roberts2010). For instance, when the thematic subject leaves Spec,vP to Spec,TP in English, it should have entered into an Agree relation with $\text{T}^{0}$ prior to its movement to Spec,TP. The movement of the subject to Spec,TP is said to be triggered by the EPP feature on $\text{T}^{0}$ . This implies that Move is parasitic on Agree, i.e. Move is Probe–Goal $+$ EPP.

Furthermore, Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001) argues that a sentence derivation is implemented through phases, i.e. propositional local domains with the ultimate aim of reducing the computational complexity. For Chomsky, there are minimally two phases: a CP and a v*P (i.e. a vP with an external argument such as ergative and transitive verbs, but see Legate Reference Legate2003 for the assumption that passives and unaccusatives are also phases in English). See (3) for a schematic representation of phases:

We shall claim below that ʔinn is a phase head that shares its unvalued $\unicode[STIX]{x1D719}$ -features with its complement $\text{T}^{0}$ .

One of the key notions of the current paper is a PF Record. We essentially argue that overt Case and $\unicode[STIX]{x1D719}$ -agreement in Arabic are prompted because of securing a phonetic record of Agree relations between probes and goals. A record is defined as a phonetically overt realization of the Agree relation between a probe and its goal. For instance, the overt nominative Case morpheme - $u$ on the subject ʔalfaata ‘the girl’ in the following MSA sentence counts as a phonetically overt realization (hence a record in our terms) of the Agree relation between $\text{T}^{0}$ and the post-verbal subject:

Likewise, the suffix - $e$ attached the complementizer datt in example (5) below from Katwijk Dutch is under our approach a record of the Agree relation between the complementizer and the pre-verbal subjects we/jullie/hullie (the example is taken from Haegeman & van Koppen Reference Haegeman and van Koppen2012: 441, cited originally from Barbiers et al. Reference Barbiers, van der Auwera, Bennis, Boef, De Vogelaer and van der Ham2006):

The main motivation for the notion ‘a record’ is that dependency relations between elements should be realized beyond narrow syntax, i.e. at PF. We will argue below that this occurs in Arabic syntax. A record can be secured by overt Case on the goal. When overt Case is not available, a record is secured through a $\unicode[STIX]{x1D719}$ -agreement suffix (i.e. $\unicode[STIX]{x1D719}$ -affix) that expresses the morphological realization of the goal’s $\unicode[STIX]{x1D719}$ -properties on the probe. We shall show that JA and LA make use of the latter strategy (using an $\unicode[STIX]{x1D719}$ -affix) to secure a record of Agree relations due to the lack of overt Case markings in these two Arabic dialects. On the other hand, MSA uses overt Case to perform this task as long as the goal can bear an overt Case marking morpheme. In this way, we account for the apparent complementary distribution between overt Case and $\unicode[STIX]{x1D719}$ -Agree in Arabic. We propose that such a distribution is subject to the postulated interface condition Agree Chain Record, which forces an Agree chain to have a morphological realization through morphological case or, if there is none, $\unicode[STIX]{x1D719}$ -Agree. This essentially speaks for the assumption that overt Case (in Arabic) blocks $\unicode[STIX]{x1D719}$ -Agree instead of being supplementary to Agree as proposed by a number of researchers in related literature.

In the following section, we explore the existing view of the bound forms attached to the complementizer ʔinn, being pronominal elements. This view is shown to be problematic as it suffers from several problems.

3 Setting the scene

The view that enjoys near unanimity among modern and traditional Arabic grammar linguists with respect to MSA ʔinn is that it is a head (i.e. a complementizer) that assigns accusative Case. This view is supported by, for example, the fact that a pre-verbal subject is invariantly assigned accusative Case when it is preceded by ʔinn, as shown in the following pair:

In (6a), the clause-initial subject ʔalwalad ‘the boy’ is assigned nominative Case (i.e. - $u$ ), while it is assigned accusative Case in (6b) (i.e. - $a$ ) being directly preceded by the complementizer ʔinn.

In a clause with VSO word order,Footnote ^[4] an inflectional suffix is forced to appear on ʔinn, irrespective of the subject being used or dropped, as the following examples demonstrate:

Within traditional Arabic grammar, such an inflectional suffix counts as a pronominal element (Ibn Al-Anbari Reference Ibn Al-Anbari1961; see also Owens Reference Owens1988). Likewise, working within the generative practice, Mohammad (Reference Mohammad and Eid1990, Reference Mohammad2000) and Benmamoun (Reference Benmamoun and Mead1993), among others, argue that this suffix in such situations is a phonetic form of an expletive that is assigned accusative Case by ʔinn. The expletive is assumed to be in Spec,TP/IP and is forced to surface because it is assigned accusative Case. Mohammad (Reference Mohammad and Eid1990: 104) states: ‘Since ʔanna has the property of assigning accusative Case, it can be used in order to create a non-nominative context. Such a context will bar pro from occurring in this position and force a lexical pronoun to appear’. It is clear that Mohammad’s (Reference Mohammad and Eid1990, Reference Mohammad2000) and Benmamoun’s (Reference Benmamoun and Mead1993) arguments are actually similar in substance to what traditional Arabic grammarians said long ago on the status of this suffix being a bound pronoun. However, in this article evidence is presented defending an alternative view, i.e. this suffix is not a bound pronoun but an agreement inflection. Despite the intuitive appeal of the traditional Arabic grammar’s ‘pronominal’ view, it suffers from empirical evidence.Footnote ^[5] Neither Arabic prescriptive grammarians nor recent Arabic scholars who maintain the same position support this view with strong (empirical) evidence nor do they show to what extent MSA is similar to other natural languages or even to Arabic dialects in this respect.

One immediate challenge against this view (i.e. the inflectional suffix attached to ʔinn is a bound pronoun) is that it does not show how this analysis would carry over to other Arabic vernaculars where ʔinn does not retain its MSA behaviour. For example, a suffix in JA is attached to ʔinn even in clauses with SVO word order, a syntactic environment where the suffix is prohibited to appear in MSA. Witness the following examples from JA:

In all examples in (8), an inflectional suffix is attached to ʔinn in the context of SVO word order.

A similar case is found in LA where the suffix has an invariant form, -o [3sg.m], irrespective of the word order used:Footnote ^[6] $^{\text{,}}$ Footnote ^[7]

Aoun et al. (Reference Aoun, Benmamoun and Choueiri2010) argue that the suffix - $o$ appearing on ʔənno is a [3sg.m] inflection, and it is not a lexical component of the complementizer (see their discussion on page 137). See also Aoun, Benmamoun & Sportiche (Reference Aoun, Benmamoun and Sportiche1994), which labels this suffix (i.e. - $o$ ) as an agreement morpheme. Data from JA and LA dismisses the pronominal view of the suffix attached to ʔinn, given that Spec,TP (the potential site of the expletive pro under the pronominal view) is filled with a pre-verbal subject.

Additionally, the pronominal view is conceptually incomplete. What we can understand from it is that pro lexicalization is forced by Case requirements, something that is barely attested outside Arabic language. One might think of a possibility that pro lexicalization is an idiosyncratic property of Arabic, whereby accusative Case is not assigned to elements with no phonological content. Or pro’s lexicalization can be treated as an idiosyncratic property of accusative Case assignment with no relevance to Arabic. However, the two possibilities are directly dismissed when faced with how syntax works. Firstly, the way syntax works predicts that any phonological constraint (on elements receiving Case) should be outside narrow syntax (i.e. a post-Spell-Out operation). This is because pro’s lexicalization is best seen as a phonological requirement rather than a syntactic necessity. Whether or not pro is forced to be pronounced is a PF matter, as argued for by a number of researchers, including Holmberg (Reference Holmberg2005), who states that ‘nullness is a phonological matter: the null subject is a pronoun that is not pronounced’ (p. 538). So we do not need to make recourse to Case, a narrow syntax process when it comes to the choice between nullness and overtness. Additionally, given that pro’s lexicalization occurs after the spellout point, it has no semantic value, a fact that even Mohammad’s (Reference Mohammad and Eid1990, Reference Mohammad2000) approach (advocating the pronominal view) accepts. Even more serious is the assumption that pro should be lexicalized under an accusative Case assigner, a claim that is barely attested beyond Arabic. To the best of our knowledge, it has never been argued that a pro would be forced to surface when it is assigned accusative Case. A case in point is Italian. In this language, an object pro is attested and assumed to receive accusative Case with no lexicalization being forced (Rizzi Reference Rizzi1986). The following sentence illustrates:

We find the same phenomenon in other pro-drop languages such as Imbabura Quechua (a South American indigenous language with the SOV word order) and European Portuguese, as shown in (11a) and (11b), respectively.

The examples in (10)–(11) strongly undermine the assumption that pro lexicalization is sensitive to which Case is assigned to pro. Additionally, the pronominal view of the suffix attached to ʔinn has no way of accounting for why the proposed expletive surfaces as a bound form rather than a freestanding pronoun as it appears in tonic situations. The conclusion we arrive at here is that tying pro lexicalization to Case is less motivated both theory-internally and cross-linguistically, and even erroneous within a unified theorem of pro licencing.

Thus, the natural question to ask at this point concerns the status of the bound suffix attached to ʔinn in the VSO word order in MSA and why such a suffix does not appear when the SVO word order is used. We show below that this state of affairs follows from a condition that demands a morphological realization of Agree relations in Arabic to be obtained. We call this condition Agree Chain Record (ACR); it can be satisfied by overt Case assigned to the goal by the probe. In such situations, there is no need to use a $\unicode[STIX]{x1D719}$ -affix of a goal on the probe. We show that the latter strategy is only used when the goal does not receive overt Case in Arabic. Data from JA and LA supports this proposal.Footnote ^[8] In these two dialects, there are no overt Case markings on nominals, hence Agree relations are recorded by a $\unicode[STIX]{x1D719}$ -affix of the goal on the probe. Discussion of ACR is delayed till Section 6, which investigates the relevant data from MSA, where the alternation between overt Case and $\unicode[STIX]{x1D719}$ -agreement becomes clear.Footnote ^[9] $^{\text{,}}$ Footnote ^[10]

4 Jordanian Arabic (JA)

Let us explore first complementizer agreement in JA. As it is the case with other eastern Arabic modern dialects, ʔinn in JA is only permitted to appear in non-root contexts, unlike in MSA, as the following examples from JA demonstrate:

On the other hand, ʔinn in JA and in other Arabic dialects introduces a wide range of embedded clauses such as clausal complements of verbs, adjectives, and nouns. It is also used to introduce some adverbial clauses (with the subordinator):

ʔinn signals that the adjoining clause is subordinate in the sense that it depends on the matrix clause to form a well-formed sentence. From this it follows that ʔinn is a clause typer (i.e. an element that denotes the type of the clause being declarative, interrogative, imperative, subordinate, etc.).Footnote ^[11] Adopting Rizzi’s (Reference Rizzi and Haegeman1997) split CP system, in which CP is replaced by Force Phrase ${>}$ Topic Phrase ${>}$ Focus Phrase ${>}$ *Topic Phrase ${>}$ Finiteness Phrase, ʔinn counts as the head of the Force Phrase and, as such, a phase head in the sense of Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001).

Following the hypothesis that a phase head is the locus of $\unicode[STIX]{x1D719}$ -features, among other things (Chomsky Reference Chomsky, Sauerland and Gärtner2007, Reference Chomsky, Freidin, Otero and Luisa Zubizarreta2008), we propose that ʔinn being a phase head is endowed, among others, with a set of uninterpretable, unvalued $\unicode[STIX]{x1D719}$ -features. Such features must be valued and deleted before derivations converge at LF, a state of affairs forced by the principle of Full Interpretation (Chomsky Reference Chomsky1986, Reference Chomsky1995). Given that such features are uninterpretable, they must be deleted, otherwise they survive until LF which in turn cannot read them, causing the sentence to crash. Chomsky (Reference Chomsky, Sauerland and Gärtner2007) assumes that the valuation of such features does not occur while they are on $\text{C}^{0}$ ; he proposes that such features pass down to $\text{T}^{0}$ under what is known as feature inheritance.

On the other hand, several works have challenged feature inheritance given that in some languages complementizers are agreeing elements, something that implies that there is no feature inheritance as such, and this condition should be relaxed. Among these works is the illuminating research by Ouali (Reference Ouali, D’Alessandro, Hrafnbjargarson and Fischer2008, Reference Ouali2011), who convincingly argues that feature inheritance as formulated by Chomsky should entail three logical possibilities: donate, keep, and share. Through donate, $\text{C}^{0}$ passes down its features to $\text{T}^{0}$ without keeping a copy of them; through keep, $\text{C}^{0}$ does not transfer its features to $\text{T}^{0}$ ; and through share, $\text{C}^{0}$ transfers its features to $\text{T}^{0}$ but keeps a copy. Ouali (Reference Ouali2011) shows how each possibility can account for some phenomenon in Tamazight Berber, including clitic doubling and anti-agreement effects. Ouali’s analysis brings insights into our understanding of the behaviour of ʔinn in Arabic dialects. Suppose that ʔinn as a phase head is endowed with a set of unvalued $\unicode[STIX]{x1D719}$ -features which must be valued and deleted due to the principle of Full Interpretation.Footnote ^[12] Suppose also that JA opts for the share possibility and thereby ʔinn percolates down its uninterpretable, unvalued $\unicode[STIX]{x1D719}$ -features to $\text{T}^{0}$ but keeps a copy of them.Footnote ^[13] Note here that Chomsky himself (in Chomsky Reference Chomsky2013) adopts proposals by Ouali (Reference Ouali, D’Alessandro, Hrafnbjargarson and Fischer2008, Reference Ouali2011), assuming that $\text{C}^{0}$ may keep a copy of the $\unicode[STIX]{x1D719}$ -set that is transferred to $\text{T}^{0}$ , which then initiates a separate Agree operation (see Fuß Reference Fuß and Grewendor2014).Footnote ^[14]

An important note here is that examples (13) demonstrate that share does not presuppose that the two heads share the same value of $\unicode[STIX]{x1D719}$ -features as sharing occurs before valuation; so each head can probe separately. Cross-linguistic evidence supporting this view (that $\text{C}^{0}$ and $\text{T}^{0}$ may agree with separate goals) is found in West Flemish External Possessor agreement. Haegeman & van Koppen (Reference Haegeman and van Koppen2012) proposed that $\text{T}^{0}$ and $\text{C}^{0}$ are associated with a separate set of uninterpretable features each. They observed that when the subject is a possessive construction, $\text{C}^{0}$ agrees with the external possessor, which is the most local goal for $\text{C}^{0}$ , while $\text{T}^{0}$ agrees with the possessum, which is the subject here. Consider the following example (from Haegeman & van Koppen Reference Haegeman and van Koppen2012: 4):

The complementizer omda agrees with the possessor die venten ‘those guys’, resulting in the situation that the inflectional suffix attached to omda comes out with the plural form. On the other hand, the Tense of was ‘was’ agrees with the possessum underen computer ‘their computer’. $\text{T}^{0}$ and $\text{C}^{0}$ probe for goals separately (see Haegeman & van Koppen Reference Haegeman and van Koppen2012).

Valuation of ʔinn’s uninterpretable $\unicode[STIX]{x1D719}$ -features is executed through the Agree operation (see Section 2 above). Now let us work out a concrete example. Consider the following sentence:

ʔinn, keeping a copy of its unvalued $\unicode[STIX]{x1D719}$ -features, probes the subject ʔi $\unicode[STIX]{x1D712}$ watuh ‘his brothers’ that bears a valued set of $\unicode[STIX]{x1D719}$ -features (the subject is situated in Spec,TP). As a result, ʔinn’s $\unicode[STIX]{x1D719}$ -features are valued by those of the subject (the goal here), yielding the surface form -hum, whose morphological specification copies that of the subject, that is [3pl.m].

It is worth noting that the realization of $\text{C}^{0}$ ’s agreement features can be spelled out as an agreeing form that is attached to the complementizer is attested in several languages, as shown in the following examples from different languages:

Further empirical evidence that the inflectional suffix attached to ʔinn is sensitive to the $\unicode[STIX]{x1D719}$ -content of the subject comes from the following JA example, where the suffix expresses the same $\unicode[STIX]{x1D719}$ -content of the new subject ʔilbinit ‘the girl’:

Had the inflectional suffix expressed a different $\unicode[STIX]{x1D719}$ -value than that of the subject’s (in an SVO clause), the resulting sentence would be ungrammatical, as demonstrated in the following ill-formed example:

We assume, following the proposals by Haegeman (Reference Haegeman1992), Zwart (Reference Zwart1993), Hoekstra & Smits (Reference Hoekstra, Smits, Hoekstra and Smits1997), Watanabe (Reference Watanabe2000), van Craenenbroeck & van Koppen (Reference van Craenenbroeck and van Koppen2002), Carstens (Reference Carstens2003), and van Koppen (Reference van Koppen2005), that the affix attached to ʔinn is a PF reflex of uninterpretable $\unicode[STIX]{x1D719}$ -features on $\text{C}^{0}$ .

The question that arises now is whether ʔinn probes the subject while the latter is in Spec,vP (the canonical position of subjects in JA; Jarrah Reference Jarrah2017) or Spec,TP (the landing site of a pre-verbal subject in JA; Jarrah Reference Jarrah2017). Given that $\text{T}^{0}$ is an active probe, it is most likely that ʔinn probes the subject while the latter lies in Spec,TP because ʔinn cannot probe over $\text{T}^{0}$ , given relativized minimality (taken here to be a locality intervention that is triggered when the intervener is of the same type as the probe or the goal with respect to the typology condition; see Rizzi Reference Rizzi1990). In this respect, Boeckx (Reference Boeckx2003: 17) notes that ‘an element $\unicode[STIX]{x1D6FD}$ (c-commanding $\unicode[STIX]{x1D6FE}$ and c-commanded by $\unicode[STIX]{x1D6FC}$ ) blocks the establishment of an Agree-relation between two other elements $\unicode[STIX]{x1D6FC}$ and $\unicode[STIX]{x1D6FE}$ even if $\unicode[STIX]{x1D6FD}$ itself could not agree with $\unicode[STIX]{x1D6FC}$ ’. The lower probe (i.e. $\text{T}^{0}$ ) invokes a minimality effect against ʔinn’s probing the post-verbal subject. Furthermore, evidence that ʔinn’s probing is ruled by locality comes from ʔinn’s agreement with a preposed focalized object, as in (19a), or a topicalized object, as in (19b), when the latter appears in a pre-subject position:

In (19), ʔinn agrees with the fronted object rather than the post-verbal subject with which $\text{T}^{0}$ agrees. Additionally, sentences (19) point to the assumption that the Case Activation Principle (a goal should have unvalued Case to enter into an Agree relation) is also not operating in JA, given that a goal can enter into another Agree relation with a different probe even if its structural Case is already assigned (see Carstens Reference Carstens2003 for an argument that the goal’s unvalued structural Case is not a prerequisite of the Agree operation). The preposed focalized object in (19a) is argued to be base-generated as a complement of the verb in Arabic (see Ouhalla Reference Ouhalla, Eid and Ratcliffe1997). Note that it leaves a gap in its base position, an issue that is widely taken as evidence of A-bar movement of the preposed object to its surface position (see Aoun et al. Reference Aoun, Benmamoun and Choueiri2010). Additionally, being a focus, the object here expresses a new piece of information that is not already shared with the hearer, hence the use of a non-specific/indefinite object. Although the object has already had its Case valued, it can enter into another Agree relation with ʔinn. This indicates that ʔinn agrees with either the subject or the object under closest c-command, i.e. the most local goal. As for (19b), ʔinn agrees with the dislocated object which functions here as a topic, expressing old, given information between the hearer and the speaker (hence the use of the object as a definite/specific entity). Note that the topicalized object is generated in situ and doubled by the so-called resumptive clitic on the verb (see Ouhalla Reference Ouhalla, Eid and Ratcliffe1997, among others). This indicates that ʔinn can agree with a topic or a focus. (20a) shows that ʔinn agrees with the subject unless the object intervenes between them (see (20c)/(19a,b)) in which case ʔinn agrees with the closer object. (20b), on the other hand, shows that ʔinn cannot agree with the object while the subject appears between them.

Table 1 shows the paradigm of subject/fronted object–ʔinn agreement in JA.Footnote ^[15]

Table 1 The paradigm of subject/object–ʔinn agreement in Jordanian Arabic.

Let us now turn to JA instances with VSO word order. In VSO word order, the suffix attached to ʔinn in JA is also variant, reflecting the $\unicode[STIX]{x1D719}$ -content of the subject, as the following sentences show:

Given that $\text{T}^{0}$ invokes a minimality intervention effect that blocks ʔinn from probing the post-verbal subject, we cannot assume here that ʔinn agrees with the post-verbal subject even if the latter carries an inflection suffix that expresses the $\unicode[STIX]{x1D719}$ -content of the post-verbal subject. In this regard, several works have argued that in an Arabic VSO sentence, Spec,TP is filled with an expletive pro (Mohammad Reference Mohammad2000: 91–93). We carry over this analysis to JA, assuming that ʔinn agrees with an expletive pro situated in Spec,TP. Evidence in favour of the existence of an expletive pro in Arabic can be adduced from the observation made by Fassi Fehri (Reference Fassi Fehri1993: 40) that pro in Arabic can be surfaced and has $\unicode[STIX]{x1D719}$ -content; pro is inflected for Number and Gender. Fassi Fehri cites some examples from so-called nominal sentences, i.e. sentences that lack overt verbs (in the matrix clause), to support his analysis that expletive pronouns in Arabic bear $\unicode[STIX]{x1D719}$ -content.Footnote ^[16] Consider the following examples:

JA maintains a similar behaviour, as demonstrated below:

For Fassi Fehri, there is an expletive pro that occupies Spec,TP (Spec,IP in his system). The pro in such sentences functions as a grammatical subject that is co-indexed with the thematic subject. The pronunciation of this expletive pro is a marked option that is always associated with emphasis. Let us suppose that Spec,TP in sentences (21) above is filled with a pro which is co-indexed with the post-verbal subject. Given that expletives are assigned Case (Bošković Reference Bošković, Lindseth and Franks1997, Reference Bošković2002; and Martin Reference Martin, Epstein and Hornstein1999) and have $\unicode[STIX]{x1D719}$ -content (Fassi Fehri Reference Fassi Fehri1993), they are qualified as a suitable goal with which ʔinn agrees. This situation eventually results in that the inflectional suffix on ʔinn expresses the $\unicode[STIX]{x1D719}$ -content of the post-verbal subject with which the expletive pro is $\unicode[STIX]{x1D719}$ -co-indexed. ʔinn’s probing the expletive pro in Spec,TP is schematically represented in (24).

The question to be asked here is why the post-verbal subject in (21) above does not move to Spec,TP. The answer to this question lies in an answer to a different question, which concerns the position of a post-verbal subject in JA in VSO word order. One might assume that the post-verbal subject in (21) is located in Spec,vP. However, this assumption is dismissed on the grounds that the post-verbal subject in such cases (i.e. in VSO clauses) should be specific (and in most cases, definite), expressing old, given information. Sentences equivalent to those in (21) above with an indefinite post-verbal subject are ungrammatical or, as some of the informants call it, pragmatically odd (indicated by #).Footnote ^[17]

In order to accommodate the observation that sentences (25) reveal, we resort here to Belletti’s (Reference Belletti and Rizzi2004) hypothesis of the so-called low IP area. Belletti argues for a discourse-related field that is located between TP and vP, where a recursive Topic Phrase and a Focus Phrase can be projected, as shown in the following diagram:

Given that the post-verbal subject in (21) should be specific (and definite), it can be suggested that the subject in such instances is a topic located in the low IP area, after leaving its canonical position, Spec,vP. Some evidence in favour of this option can be adduced from the fact that the post-verbal subject in a VSO clause should be accompanied by a downgrading intonation that, as Belletti (Reference Belletti and Rizzi2004) mentions, is not a property of normal subjects. Assuming that there is no movement from an A-bar position to an A position (Hicks Reference Hicks2009), the topicalized subject in Spec,Topic Phrase (of the low IP area) is unable to move to Spec,TP, hence its inability to fill Spec,TP – the reason, we think, why the expletive pro is called for.Footnote ^[18] The derivation of VSO word order in JA is represented as follows (irrelevant details are omitted) (silent copies are set in <>):

There is one last fact concerning JA complementizer agreement. In JA, when the subject is dropped (i.e. an understood subject/a referential pro), ʔinn should agree with this subject even if there is a preposed object, as shown in the following example:

Although the expectation is that ʔinn agrees with the preposed object being more local to it than the referential pro (located in Spec,vP or in Spec,TP), ʔinn agrees with the latter. This observation is significant in revealing the actual mechanism of ʔinn’s probing in Arabic dialects. We discuss the account of this observation in the following section where the same observation arises in the LA context. We essentially argue, following Rizzi (Reference Rizzi and Haegeman1997), van Craenenbroeck (Reference van Craenenbroeck2004), Branigan (Reference Branigan2011), and Omari & Branigan (Reference Omari and Branigan2014), that ʔinn in Arabic dialects originates in FinP (the lowest layer of Rizzi’s CP system) and then it raises to adjoin to Force $^{0}$ , the head of the Force Phrase. We propose that ʔinn in JA starts probing when it is in Fin $^{0}$ , where it can agree only with the referential pro under the pro– $\text{T}^{0}$ union (we explain this in the following section). If the thematic subject is not a referential pro, there is no agreement between ʔinn and the thematic subject given that the latter does not make a union with $\text{T}^{0}$ . In the next step, ʔinn raises to Force $^{\text{0}}$ , where it also commences probing if its $\unicode[STIX]{x1D719}$ -feature set remains unvalued. Here ʔinn agrees with the most local goal. The same situation occurs in LA, with the exclusion that ʔinn does not probe when it raises to Force $^{0}$ , something that gives rise to the observation that ʔinn in LA only agrees with the referential pro.

To summarize, JA selects the possibility of share of the feature inheritance. $\text{C}^{0}$ passes down its $\unicode[STIX]{x1D719}$ -content to $\text{T}^{0}$ and leaves a copy, a matter that makes $\text{C}^{0}$ an agreeing head. ʔinn agrees with the pre-verbal subject or the preposed object, depending on which one is most local to $\text{C}^{0}$ . This reflects that valuation of ʔinn’s $\unicode[STIX]{x1D719}$ -features is ruled by locality, an expected result under an Agree model to license the formal features. Additionally, this section claims that $\text{T}^{0}$ blocks ʔinn’s probing, creating a minimality effect.

In the next section we investigate the morpho-syntactic form of the inflectional suffix attached to ʔinn in LA.

5 Lebanese Arabic (LA)

In LA, the inflectional suffix attached to ʔinn in most cases has an invariant form, - $o$ . Aoun et al. (Reference Aoun, Benmamoun and Sportiche1994) argue that LA complementizer ʔinnu should be decomposed into a C unit, i.e. ʔinn, and an agreement morpheme - $u$ /- $o$ [3sg.m].Footnote ^[19] Put in the terms developed in this work, the inflectional suffix attached to ʔinn in LA is assigned the default form of agreement. Shlonsky (Reference Shlonsky1997: 264) mentions that this suffix should not be confused with the affix that expresses the truly agreement features [3sg.m]. For him, the former should be an impersonal Agr affix that manifests a default specification of features.

In order to account for the use of the default inflectional suffix on ʔinn in (29), Shlonsky proposes that Spec,AgrC in such situations is filled with a phonetically null it-like pronoun, which enters into agreement with $\text{C}^{0}$ . The problem here, however, is why there should be an AgrC at all in such cases. It is widely attested that default agreement is a sign of lack of agreement (see e.g. Bhatt Reference Bhatt2005 and Baker Reference Baker2008). Contra Shlonsky (Reference Shlonsky1997)), we propose that the default inflectional suffix on ʔinn in LA is used because ʔinn fails to value its unvalued $\unicode[STIX]{x1D719}$ -features. This implies, first, that ʔinn has $u\unicode[STIX]{x1D719}$ -features, whose valuation is realized as an inflectional suffix appearing on ʔinn, albeit with the default form when they are not valued.Footnote ^[20] This also suggests that, like JA, LA opts for the share option, rather than donate or keep. It is not donate because the presence of an inflectional suffix (even in the default form) would be a mystery. Secondly, for instances with a dropped subject (e.g. a referential pro), the inflectional suffix should express the $\unicode[STIX]{x1D719}$ -content of the dropped subject and hence the situation that the suffix has variant forms, as we show below. Note that the keep option is also excluded because $\text{T}^{0}$ is overtly inflected for agreement (i.e. the verb tiiʒi ‘come’ in (29) agrees in $\unicode[STIX]{x1D719}$ -features with its subject). Thus, it is clear that $\text{C}^{0}$ shares its $\unicode[STIX]{x1D719}$ -content with $\text{T}^{0}$ (recall that share does not presuppose that the two heads share the same $\unicode[STIX]{x1D719}$ -features as sharing occurs before valuation– each head can probe separately).

Let us first account for the obvious observation that the $\unicode[STIX]{x1D719}$ -features of ʔinn are assigned the default form (i.e. [3sg.m]) in the presence of an overt subject. What is the reason for that? It cannot be the option that ʔinn always agrees with an expletive (in Spec,TP) whose $\unicode[STIX]{x1D719}$ -features are fixed. Note that the thematic subject is what normally occupies this position in LA (after leaving Spec,vP). Several works on Arabic vernaculars argue extensively that the thematic subject moves to Spec,TP in such dialects in SVO clauses (Mohammad Reference Mohammad and Eid1990, Reference Mohammad2000; Bolotin Reference Bolotin and Eid1995; Benmamoun Reference Benmamoun2000; Harbert & Bahloul Reference Harbert, Bahloul, Ouhalla and Shlonsky2002; and Aoun et al. Reference Aoun, Benmamoun and Choueiri2010). One significant observation at this point is that the default form of the suffix remains as such even in instances that include a preposed object. Consider the following sentence, provided by one LA informant:

Given the default form of the inflectional suffix attached to ʔinn, one possibility suggests itself: ʔinn no longer probes in this Arabic variety. Put succinctly, ʔinn is unable to probe, thereby valuing the $\unicode[STIX]{x1D719}$ -features of ʔinn as default. It is a well-known observation that the default form of agreement is appealed to when the content of unvalued features is not valued (Fassi Fehri Reference Fassi Fehri1993). The question that needs an answer is why ʔinn cannot probe. There should be some factor that prevents ʔinn from probing, hence the result that ʔinn is an inactive probe. If we assume that one probe cannot agree with a goal in the presence of another probe between them, then it follows that ʔinn cannot probe the subject in the presence of $\text{T}^{0}$ . Given that the unmarked word order in Arabic varieties is SVO (see footnote 4), ʔinn should wait until the subject moves to Spec,TP, so escaping the intervention effect invoked by $\text{T}^{0}$ , as is the case in JA. So the question is why ʔinn does not agree with the subject when the latter moves to Spec,TP, or, in other words, why ʔinn does not wait in LA?Footnote ^[21] As we have shown above, an Agree relation between a probe and a goal is impervious to whether the goal has its structural Case valued or not. One might suggest in this context that subject movement to the Spec,TP does not occur in the audible syntax so ʔinn’s $\unicode[STIX]{x1D719}$ -features are valued as default given T’s intervention effect. However, this suggestion would not be plausible when we consider instances with dislocated elements. It is hard to assume that object topicalization/focalization occurs in post-syntactic components. So the possibility of any reason relating to post-syntactic factors is dismissed right away.

As for why LA ʔinn does not wait until movement to the left periphery is accomplished, as is the case in JA, we suggest that ʔinn in Arabic originates in Fin and then moves to Force (to check the Force feature; see Omari & Branigan Reference Omari and Branigan2014), along the lines of Rizzi (Reference Rizzi and Haegeman1997), van Craenenbroeck (Reference van Craenenbroeck2004), and Branigan (Reference Branigan2011). Suppose that ʔinn in JA still probes while it is in Force, so it can agree with an element moving or being base-generated in the CP domain. On the other hand, ʔinn in LA originates in Fin. Suppose that in LA ʔinn’s Agree only occurs while it is in Fin, ʔinn in LA cannot agree with any element that is dislocated to the left periphery (see our analysis of the two examples in (33) below as empirical evidence). Additionally, it cannot agree with the thematic subject which originates in Spec,vP given the intervention effect of $\text{T}^{0}$ . This state of affairs leads to the situation that ʔinn in LA is typically assigned the default value.

The question to ask here, as an anonymous JL referee notes, is why the derivation does not crash when $\text{C}^{0}$ does not find an accessible goal within its c-command domain, as its unvalued $\unicode[STIX]{x1D719}$ -features are not valued but are instead assigned the default value? If our analysis is on the track, it can be postulated that an Agree relation is triggered to set some context-dependent value to the unvalued $\unicode[STIX]{x1D719}$ -features. When no context-dependent value is possible, e.g. because of the intervention of some categories, such unvalued $\unicode[STIX]{x1D719}$ -features are assigned the default form as a last resort to salvage the derivation. This entails that the default form of agreement is used when Agree fails (see Preminger Reference Preminger2014 for pertinent discussion).Footnote ^[22]

The question that arises now is how we can account for the instances where ʔinn is inflected for agreement in LA. Note first that in such situations, the subject should be phonologically null (i.e. a pro), as demonstrated in the following example:

Consider also the following examples, provided by two LA informants:

Given our discussion of JA agreeing complementizers above (see Section 4), ʔinn should be able to probe here. This correlation between the subject being a pro and ʔinn being able to probe needs an exploration. In order to account for this correlation, we draw on Holmberg’s (Reference Holmberg2009) theory of null subjects and agreement (advocated independently for Arabic in Al-Horais Reference Al-Horais2012). Under this theory, $\text{T}^{0}$ probes the referential pro located in Spec,vP to value its own $u\unicode[STIX]{x1D719}$ -features. The referential pro counts a $\unicode[STIX]{x1D719}$ P subject which mainly consists of a set of valued (i.e. lexically specified) $\unicode[STIX]{x1D719}$ -features. For Holmberg (Reference Holmberg2009: 94–95), ‘when T probes a $\unicode[STIX]{x1D719}$ P subject, and has its unvalued $\unicode[STIX]{x1D719}$ -features valued by the subject, the resulting union of the $\unicode[STIX]{x1D719}$ -features of $\text{T}^{0}$ and the subject yields a definite pronoun’. This union is made possible through incorporation of a $\unicode[STIX]{x1D719}$ P subject into $\text{T}^{0}$ by making the $\unicode[STIX]{x1D719}$ -feature values of the subject pronoun copied by $\text{T}^{0}$ . Holmberg (Reference Holmberg2009: 97) states:

Following the incorporation of a $\unicode[STIX]{x1D719}$ P subject into $\text{T}^{0}$ , $\text{T}^{0}$ becomes now endowed with a set of interpretable $\unicode[STIX]{x1D719}$ -features whose content is identical to that of the referential pro in Spec,vP. We suggest that ʔinn probes $\text{T}^{0}$ in such situations, resulting in that the inflected $\unicode[STIX]{x1D719}$ -content of ʔinn being identical to that of the dropped subject. $\text{T}^{0}$ does not invoke an intervention effect against ʔinn’s probing the post-verbal pro since ʔinn probes $\text{T}^{0}$ itself, which acts here as a goal, due to its union with the referential pro. This argument can be supported by examples with a referential pro subject and a fronted object. In such examples, ʔinn agrees with the referential pro both in JA and LA, as evidenced in the following examples:

One might wonder here why such a mechanism is not available when the subject is an overt DP. In this regard, Roberts (Reference Roberts2009: 76) argues that a goal to be incorporated into its probe should be a defective goal, defined as follows: defective goals always delete/never have a PF realisation independently of their probe. Overt DP subjects always have a PF realization independent of their probes (e.g. $\text{T}^{0}$ ), something that blocks the incorporation of an overt subject into $\text{T}^{0}$ . Here, the inflectional suffix attached to ʔinn is assigned the default form. Consider the following LA examples (adapted from Aoun et al. Reference Aoun, Benmamoun and Sportiche1994: 202):

Examples (34c, d) indicate that ʔinn’s $\unicode[STIX]{x1D719}$ -content must be assigned the default value (i.e. [3sg.m]); otherwise the sentence becomes ill-formed. This can be straightforwardly explained under the analysis developed here. The full DP post-verbal subject cannot constitute a union with $\text{T}^{0}$ , which as a result does not become a goal.

Another issue that should be addressed before we pursue our investigation of ʔinn in MSA concerns the issue of which principle excludes the situation in which $\text{T}^{0}$ locally probes the lower subject and then $\text{C}^{0}$ locally probes $\text{T}^{0}$ , resulting in the correct agreement suffix on $\text{C}^{0}$ without any movement, as raised by an anonymous JL referee. Our solution to this issue is that $\text{T}^{0}$ ’s $\unicode[STIX]{x1D719}$ -features vanish in syntax once they are valued by the post-verbal subject. They cannot count as an eligible goal with which ʔinn can agree, simply because they are no longer syntactically existing (see Holmberg Reference Holmberg2005 along these lines). The morphological suffix that appears on the verb is inserted (or created) in morphology as a reflex of this valuation. Note here that pro-incorporation with $\text{T}^{0}$ is different because $\text{T}^{0}$ in such a situation copies the interpretable valued features of pro, which are still syntactically active, and hence can enter into a further Agree relation with a different probe.

In this section, we have addressed the syntactic behaviour of ʔinn in LA. We have argued that ʔinn starts in $\text{Fin}^{0}$ , where it also starts probing. ʔinn can only agree with the referential pro which makes a union with $\text{T}^{0}$ , given the defective nature of the pro. We have also argued that ʔinn in LA raises to $\text{Force}^{0}$ , where its unvalued $\unicode[STIX]{x1D719}$ -features are assigned the default form when the thematic subject is not a pro. We have effectively argued that ʔinn is unable to probe when it raises to $\text{Force}^{0}$ in LA, contrary to the situation in JA, where ʔinn is able to probe if its $\unicode[STIX]{x1D719}$ -content is not assigned a value. In result, ʔinn can agree with a pre-verbal subject or a preposed object in JA.Footnote ^[23]

In the next section, we explore the morpho-syntactic behaviour of ʔinn in MSA. Here the interaction between overt Case and $\unicode[STIX]{x1D719}$ -Agree becomes evident as MSA still maintains its overt Case markings. We will argue that overt Case and $\unicode[STIX]{x1D719}$ -Agree are used to secure what we call a record (i.e. a morphological realization) of Agree relations. When the goal receives overt Case, there is no need to spell out the valuation of $\unicode[STIX]{x1D719}$ -features of the probe. On the other hand, when the goal is not able to receive overt Case (e.g. a goal is a phonologically null element), $\unicode[STIX]{x1D719}$ -features of the probe are spelled out.

6 Modern Standard Arabic (MSA)

This section is concerned with the morpho-syntactic behaviour of ʔinn in MSA. In the previous sections, ʔinn has been argued to be an agreeing head in JA and LA. The difference between these two varieties can be seen in terms of whether ʔinn’s probing can occur in $\text{Force}^{0}$ . In JA, ʔinn probes elements in $\text{Fin}^{0}$ and $\text{Force}^{0}$ , whereas LA ʔinn does so just in $\text{Fin}^{0}$ , hence the enforcement of the default valuation of ʔinn’s uninterpretable $\unicode[STIX]{x1D719}$ -features in cases where the subject is not a pro. It has been shown already that when the thematic subject is a pro, ʔinn agrees with this pro given its union with $\text{T}^{0}$ .

As for MSA, we assume that this variety selects share, like JA and LA. What appears challenging at the first blush is the observation that an inflectional suffix is only attached to ʔinn in the context of VSO word order. In other word orders beginning with the subject or the object, there is no inflectional suffix whatsoever used on ʔinn. Consider the following examples:

It is evident that when ʔinn is followed by a nominal, there is no inflectional suffix attached to it (witness (35a, b)), whereas this suffix is obligatory when ʔinn is followed by a verb.

In this regard, Mohammad (Reference Mohammad and Eid1990, Reference Mohammad2000) claims that the inflectional suffix is a PF form of the expletive pro, which is situated in Spec, TP. As shown in Section 3 above, this claim suffers from theory-internal problems and lacks cross-linguistic corroboration. The obvious observation regarding the sentences in (35) is that when the structural Case assigned by ʔinn has an overt form, i.e. a morphological from, there is no inflectional suffix used on ʔinn, whilst the inflectional suffix is used when there is no morphological Case. One might assume here that this inflectional suffix is a clitic having the effect of absorbing the Case assigned by ʔinn. But the question that still remains open is why ʔinn’s Case is not assigned to null elements such as pro. We argue that the use of an inflectional suffix in the presence of a pro must be treated from a different perspective, which considers the reason behind Agree and Case assignment themselves.

Postponing the discussion of OVS/OSV sentences for a moment, the derivation of Arabic VSO and SVO word orders has received much attention from researchers. The view that would be qualified as a consensus is that the VSO word order in MSA is unmarked, whereas the SVO word order is marked in the sense that the latter is derived from the former (see Aoun et al. Reference Aoun, Benmamoun and Sportiche1994 and Aoun et al. Reference Aoun, Benmamoun and Choueiri2010). Let us start with the unmarked word order. In the VSO word order, it has been long assumed that Spec,TP is filled with a pro (Mohammad Reference Mohammad and Eid1990, Reference Mohammad2000). This view concurs with Bobaljik & Jonas’ (Reference Bobaljik and Jonas1996) assumptions that Spec,TP is present in all languages as an A position intermediate in an articulated IP structure (see Aoun et al. Reference Aoun, Benmamoun and Choueiri2010: Chapter 3 for further discussion in this matter). If we pursue the line of research that argues that this pro is co-indexed with the post-verbal subject hence sharing the same $\unicode[STIX]{x1D719}$ -content, it can be assumed that the pro has a set of $\unicode[STIX]{x1D719}$ -features which are interpretable. As such, this pro is qualified as a goal with which ʔinn can agree when there is no intervening goal. ʔinn having $u\unicode[STIX]{x1D719}$ -features is a probe that agrees with the pro, assigning the accusative Case to it, just like ʔinn agrees with other pre-verbal elements in JA. As a result of this valuation, a PF reflex of valuation of ʔinn’s uninterpretable $\unicode[STIX]{x1D719}$ -features appears on ʔinn as an inflectional suffix. Following this assumption, we need to resolve the remaining issue of why there is no PF reflex when ʔinn agrees with a nominal (a subject or a fronted object). Given that the pre-verbal subject is assigned Case by ʔinn, it is most likely that the former enters an Agree relation with the latter, resulting in the subject being assigned the accusative Case by ʔinn.

One might suggest that ʔinn shares its $\unicode[STIX]{x1D719}$ -features with its complement $\text{T}^{0}$ , when there is no intervening argument between them, whilst it donates such features to $\text{T}^{0}$ in the presence of the subject (or the object) in a pre-verbal position. Although this suggestion accounts for the presence/absence of the inflectional suffix on ʔinn in MSA, it faces a serious problem of why ʔinn may choose between these two possibilities. Additionally, this suggestion is undermined when JA (and LA) are taken into account, as ʔinn here always shares its $\unicode[STIX]{x1D719}$ -features with $\text{T}^{0}$ . As we have shown earlier, in the case of SVO word order, ʔinn agrees with the subject as evidenced by Case assignment (see Schütze Reference Schütze1997, Chomsky Reference Chomsky and Kenstowicz2001, and Soltan Reference Soltan2006 for the relation between Case and Agree, but see Pesetsky & Torrego Reference Pesetsky, Torrego, Guéron and Lecarme2004 and Al-Balushi Reference Al-Balushi2011 for different proposals). The valuation of ʔinn’s $\unicode[STIX]{x1D719}$ -features is executed by the subject, so the expectation is that the PF content of the features on ʔinn is the same as that of the subject. Let us suppose that ʔinn forms an Agree chain with the subject, resulting in the assignment of the accusative Case on the latter. Let us suppose further that such a chain should be recorded (i.e. having a phonetically overt realization) in the PF, following the effects of one condition, labelled Agree Chain Record, formulated as follows:

Due to economy conditions on presentation, this record (R) is confined to one realization, i.e. $0>R>2$ (0 refers to no record; 2 refers to two records and more). On these grounds, the PF reflex of valuation of ʔinn’s $\unicode[STIX]{x1D719}$ -feature is deleted when it agrees with the pre-verbal subject as ACR is secured through the overt structural Case assigned to the subject by ʔinn. The same analysis extends to instances where ʔinn agrees with the object in the context of a VSO word order, as in (35b) above, reproduced for convenience in (37a), and to similar examples in the context of an OVS word order, in (37b):

ʔinn agrees with the object in (37), resulting in assigning a new structural morphological Case to the object. Given that ACR is secured through Case assignment, the PF reflex of valuation of ʔinn’s $\unicode[STIX]{x1D719}$ -features is dropped in PF. This reasoning automatically accounts for why this reflex is obligatory when ʔinn agrees with the pro. Since the latter does not have overt $\unicode[STIX]{x1D719}$ -content, ACR must be obtained in a different way, which is the PF reflex of valuation of ʔinn’s $u\unicode[STIX]{x1D719}$ -features, i.e. an inflection suffix of goal appearing on the probe. ACR does not force the goal, if null, to appear when the Agree relation is established, but it forces a $\unicode[STIX]{x1D719}$ -realization of this Agree to occur on the probe.

This reasoning helps us resolve the puzzle that although the uninterpretable $\unicode[STIX]{x1D719}$ -features are expected to delete, they may survive at PF (see Chomsky Reference Chomsky1995, among many others, on the deletion of uninterpretable $\unicode[STIX]{x1D719}$ -features). Under the analysis proposed here, such features are just a record of an Agree relation, forced by ACR. As far as Arabic is concerned, the PF reflex of valuation of ʔinn’s $u\unicode[STIX]{x1D719}$ -features is held to manifest a record of the Agree operation when Case falls short of doing so. One might wonder here why ACR should exist at all. The answer to this question is tied to an answer of another question which is why there is Agree at all. Miyagawa (Reference Miyagawa2010) argues that Agree occurs to establish a dependency relation, while Move is used to keep a record of the dependency relation beyond narrow syntax (so that semantic interpretation and information structure can make use of it) (p. 33). Contra Miyagawa, we propose that dependency relations between elements cannot be recorded only through Move. They can be recorded through, if any, overt Case assignment and PF reflexes of $\unicode[STIX]{x1D719}$ -features, which are more economical strategies. The examples in (38) below are ungrammatical with a suffix attached to ʔinn because the Agree relation between ʔinn and the pre-verbal subject/pre-verbal object is recorded twice, using overt Case and $\unicode[STIX]{x1D719}$ -agreement, hence violating economy conditions on presentation.

It appears that, as an anonymous JL referee notes, deletion of the extra realization of the Agree relation is cheaper than spelling out both links of the Agree chain.

One piece of empirical evidence in favour of the assumption of securing one PF record of Agree operation comes from the notorious facts on the subject–verb agreement in MSA. One of the heavily investigated phenomenon in MSA is that the observation that the verb shows partial agreement with the subject in VSO, as in (39a) below, while it shows full agreement in SVO word order, as is the case in (39b) (Mohammad Reference Mohammad and Eid1990, Reference Mohammad2000; Fassi Fehri Reference Fassi Fehri1993; Aoun et al. Reference Aoun, Benmamoun and Sportiche1994; Shlonsky Reference Shlonsky1997; Benmamoun Reference Benmamoun2000; and Aoun et al. Reference Aoun, Benmamoun and Choueiri2010; the examples are adapted from Musabhien Reference Musabhien2009: 23).

Before showing how this interplay between agreement and word order is empirical evidence in favour of ACR, one remark on the base-generation of the pre-verbal subject is in order. Several studies both in traditional Arabic grammar (i.e. the Basran School, see Al-Balushi Reference Al-Balushi2011 for discussion) and in the Arabic generative enterprise (see Al-Ghalaayyini Reference Al-Ghalaayyini1981 and Soltan Reference Soltan2007) argue that the pre-verbal subject in SVO word order is a topic rather than a genuine subject.Footnote ^[24] Soltan (Reference Soltan2007) in particular shows that the subject in this situation is base-generated in its surface position rather than a product of movement. Soltan (Reference Soltan2007) provides several pieces of evidence (from binding, idioms, Case, overt resumption, etc.) that Spec,vP is filled with a pro in an SVO clause. Recall from our discussion above that Spec,TP in the VSO word order in MSA is filled with a pro. Following this reasoning SVO and VSO sentences are represented as follows:

Within these structures in (40), $\text{T}^{0}$ (or the verb) agrees with the post-verbal pro in SVO sentences, while it agrees with the DP subject in a VSO clause. Given ACR, a record of the Agree operation between $\text{T}^{0}$ and the DP subject in a VSO clause and between $\text{T}^{0}$ and the pro in as SVO clause must be obtained. In the former case, i.e. a VSO clause, the Agree relation is recorded through the overt nominative Case assigned to the subject by $\text{T}^{0}$ . As such, there is no need for $\text{T}^{0}$ to have a PF reflex of valuation of its $u\unicode[STIX]{x1D719}$ -features by the subject, something that results in the impoverished agreement between the verb and the subject. In an SVO clause, the Agree relation is held between $\text{T}^{0}$ and the pro, a matter that forces this PF reflex on $\text{T}^{0}$ . Whatever Case is assigned to pro, it has no PF value, making Case insufficient to secure ACR. This line of analysis dispenses with the exception postulated by Shlonsky (Reference Shlonsky1997: 188) to account for subject–verb agreement in MSA. For Shlonsky, subject–verb agreement in MSA is lexical in the sense that a verb is selected from the lexicon bearing subject agreement. In our terms, subject–verb agreement is an instance of Agree whose output is subject to ACR, which should be secured to the minimum limit.Footnote ^[25]

One complication at this point comes from the fact that Gender and Person features still appear on $\text{T}^{0}$ . We suggest that this realization of Gender and Person features is forced because nominative Case, being the default Case in Arabic (Mohammad Reference Mohammad, Brentari, Larson and MacLeod1988 and Ouhalla Reference Ouhalla, Lightfoot and Hornstein1994), is not enough alone to qualify as a record. It can be qualified as a record as long as other features such as Gender and Person are spelled on the probe.

The question that arises here concerns how JA and LA react to ACR. Given that these varieties lack overt morphological Case (see Brustad Reference Brustad2000: 27 and Aoun et al. Reference Aoun, Benmamoun and Choueiri2010: 15), the PF reflex of valuation of ʔinn’s $u\unicode[STIX]{x1D719}$ -features is the method that secures ACR. Once there is an Agree relation between ʔinn and some other entity, the PF reflex of valuation of ʔinn’s $u\unicode[STIX]{x1D719}$ -features is triggered, given that spoken Arabic has no overt Case-marking system.

In view of this and on the grounds of the Arabic varieties under discussion, it can be postulated that there are two stratagems available to secure ACR, namely an overt Case marking and $\unicode[STIX]{x1D719}$ -agreement. The choice between these stratagems is not free, but subject to a strict order, governed by economy principles of language. In Arabic, this order is shown as follows: overt Case ${>}$ PF reflex. To the extent that this reasoning is correct, it accounts for why the option of partial agreement is missing in Arabic vernaculars (in VSO), as in Moroccan Arabic (Fassi Fehri Reference Fassi Fehri1993), LA (Aoun et al. Reference Aoun, Benmamoun and Sportiche1994), Palestinian Arabic (Mohammad Reference Mohammad2000), Egyptian Arabic (Jelinek Reference Jelinek, Ohalla and Shlonsky2002) Tunisian Arabic (Mahfoudhi Reference Mahfoudhi2002), and JA (Jarrah Reference Jarrah2017). As Case is no longer morphological in Arabic varieties, the Case option is excluded, hence the recourse to the $\unicode[STIX]{x1D719}$ -agreement of valuation of T’s $u\unicode[STIX]{x1D719}$ -features, even if the subject appears post-verbally. (See the appendix on the interaction of ʔinn with coordinated subjects and its behaviour in constructions with unbounded dependencies.)