EPP, Labeling and Word Order in Arabic

This paper looks at EPP and word order in Arabic in light of Chomsky’s Labeling Theory, proposed in POP and POP Extensions. In the current framework, EPP—the principle that SpecTP must be filled in—is eliminated. EPP-driven movement is reduced to labeling failure: if T fails to label the structure that arises after E-merge of the subject in Spec of vP [DP vP], then filling SpecTP becomes necessary in order to ‘strengthen’ (the labelability of) T. Chomsky postulates two types of T: Strong and weak. English-type languages, which show poor agreement inflection, have a weak T, and therefore impose the Fill-SpecTP requirement. On the other hand, NSLs, Chomsky claims, have a strong T which can label the TP structure, by virtue of having rich agreement inflection. This paper shows that Chomsky’s approach to EPP makes wrong predictions. Instead, a freezing effect account which also maintains a labeling system can explain the word order facts in Arabic. Crucially, the account proposed does not make resort to Chomsky’s parameter of strength or otherwise of T.

In (3a), two phrases are merged; and in (3b) two heads are merged. Now these structures cannot be labeled because there is no 'nearest head': in (3a) both X and Y, heads of XP and YP, respectively, are equally near, and therefore they are both potential labels. Similarly, in (3b), both X and Y are heads and, in turn, can potentially function as labels. In POP and POP Extensions, Chomsky proposes that this labeling failure can be resolved by (i) either movement of one of the two constituents, or (ii) sharing of some feature(s) between the two heads. Under the first scenario, suppose that in (2a) XP moves. Now only one head remains accessible for LA, namely Y, and the structure is successfully labeled. The same procedure applies if YP moves. On the second scenario, suppose that X and Y agree in some feature, say φ, then this shared feature will project, and the new syntactic object will be successfully labeled as < φ, φ>.
A point is in order here before we proceed. LA crucially applies at the phase level, not at the moment of (E-or I-)merge.
Interestingly, the labeling system has been entertained to explain different phenomena (Note 2). In this paper, we consider one interesting, yet challenging consequence of LA, namely, analyzing EPP in terms of LA. Section 2 is a brief sketch of EPP. Section 3 reviews labeling approaches to EPP. Specifically, it reviews Chomsky's (POP & POP Extensions) analysis of EPP-related phenomena in terms of labeling. Some other labeling-based approaches to EPP, Goto (2017) and Gallego (2017), are also discussed. Section 4 presents a labeling-based alternative to EPP-driven movement in Standard Arabic, without appeal to EPP. It begins with a review of the different subject positions that subjects might occupy along the clausal spine. The section also shows how the present analysis can account for the derivation of the different word orders in Standard Arabic and modern dialects of Arabic. Section 5 is the conclusion.

EPP
Let us consider the sentences in (4).
(4) a. *left (intended 'he left') b. John/He left c. *seems to be a man in the room.
d. There seems to be a man in the room.
(4a) is ungrammatical in English, because it requires an explicit subject, as the grammaticality of (4b) proves.
(4c) is more interesting in this context. Note that the predicate seem does not require a thematic subject; nonetheless, there must exist an overt subject for the sentence to be grammatical. That the subject position can be filled by an expletive there, which has no semantic content, indicates that the (EPP) condition is entirely structural. In fact, it is the grammaticality judgment of sentences like those in (4) that led to formulating EPP.
Null subject languages have always raised a problem for the universality of EPP, as they allow clauses without an overt subject. For instance, (5) is perfect in Arabic (Note 3).  (2000), Soltan (2007), Aoun et al. (2010), Saeed (2011), among others, on Arabic. Now in constructure like (5a) above, htat is, where there is no overt movement of the subject to a preverbal position, analyses seem to go along two lines: on one approach EPP is satisfied by a null expletive pro (Fassi Fehri, 1993;Cardinaletti, 1997Cardinaletti, , 2004Mohammed, 2000;Holmberg, 2005;Sheehan, 2006, among others Vol. 11, No. 2; another approach, the EPP in such constructions is satisfied by the movement of a rich-agreement-inflected verb to T (traditional Arabic grammar; Alexiadou & Anagnostopoulou, 1998), among others. For instance, to derive the SVO word order in Arabic, it can be argued that the subject moves to SpecTP to satisfy EPP (assuming that Agree might take place at a distance). Further, the existence of expletives like hunak in Standard Arabic (and its various forms in modern dialects) well motivates postulating an EPP on T (Note 5).
The picture is not this clean, however. In linguistic theory, and ever since it was introduced in Chomsky (1981), EPP has always been a mysterious universal (Note 6) 'principle'. As initially formulated, EPP states that a clause must have a subject; formally, SpecTP must be filled. Apart from this. it has always been felt that it is a stipulative dubious construct (e.g., Fassi Fehri, 2012; Epstein & Seely, 1999;Boskovic, 2002;Lasnik, 2003), among many others).
There are two aspects of the EPP that have been questioned.

EPP and Labeling Theory
Within the labeling framework there have been attempts to eliminate EPP. EPP-driven movement is explained in terms of (failure of) labeling. Of the labeling-based accounts we will consider here Chomsky's POP & POP Extensions, Gallego (2017), and Goto (2017).

EPP in POP and POP Extensions
In POP & POP Extensions, Chomsky does away with EPP (as well as ECP (Note 7)) as a universal principle. EPP-driven movement is reduced to labeling failure. To get a full picture, EPP-driven movement occurs when the derivation reaches the structure in (6) (after the application of External Merge of the external argument in Spec,vP), which cannot be labeled.
Now, to label the structure in (6), either of the XPs moves or the two XPs should agree in some feature. As this is the thematic position of the subject (and not a checking configuration), the latter option is not felicitous. So, one phrase has to move. It happens that the subject moves (Note 8). When the subject moves, the remaining XP is successfully labeled as <v> -the closest head (Note 9).
In POP Extensions, Chomsky goes one step further and does away with (the universality of) EPP altogether. He claims that postulating a universal version of EPP might be "a historical accident" (p. 9). Following Rizzi's generalization of the correlation between pro drop and rich agreement, Chomsky postulates two types of T: strong T and weak T. A strong T, which correlates with rich verbal agreement, is one that can label; a weak T, which occurs with impoverished subject-verb agreement morphology, is one that cannot. Accordingly, NSLs, have a strong T, manifested in rich subject-verb agreement, which can label the structure in (6) as < φ, φ>. Thus, SpecTP need not be filled, and, in turn, movement of the external subject to SpecTP is deemed unnecessary, because T is strong enough to do the job on its own. Non-null subject languages like English, on the other hand, have a weak T (manifested in poor subject-verb agreement) which cannot label the structure in (6). By this logic, then, movement of the external subject to SpecTP is driven by T's inability to label the structure. Having an XP in its Spec by movement of the external subject (or, alternatively, merging an expletive) makes T strong enough to label the structure and avoid a crash at SEM.
Theoretically, LA, as stated in (1) above has the advantage of adhering to the minimalist guidelines, especially SMT: LA is a principle of minimal computation. Further, LA, as shown in POP and POP Extensions has many interesting conceptual consequences, the most important of which is that it reduces Internal Merge to labeling failure-an interpretive principle of the SEM interface.
The main argument is that T in NSLs is by assumption strong, as Arabic is an NSL. The empirical consequences/predictions are: 1) Agreement holds when the subject does not move to SpecTP.
2) Constructions with the subject in Spec,TP may not realize agreement.
Let's first consider agreement and its interaction with subjects. Consider (7) (6). Assuming that the subject DP Ahmed does not move to SpecTP, T, or more precisely the complex [v-T] can label the structure, as T is strong enough (i.e., fully specified for φ).
Consider, however, the well-known phenomenon of agreement asymmetry in Arabic: SVO corresponds with full agreement, and VSO with poor agreement. This asymmetry raises a serious problem for Chomsky's framework. According to POP Extensions, T in Arabic is, by definition, strong (as in other NSLs), and, consequently, agreement will be observed when the subject stays in situ. Interestingly, the asymmetry in (8) yields just the opposite picture: (full) agreement holds only when the subject moves to a preverbal position. In Chomsky's labeling sense, T in Arabic is strong when accompanied by movement of the subject to a preverbal position (Note 10); similarly, T is weak when the subject stays in situ.
In fact, Soltan (2007) proposes an analysis for the agreement asymmetry in Standard Arabic in terms of strength (or otherwise) of T (Note 11). According to Soltan, T is strong/rich, i.e., fully inflected for phi-features, in the SVO word order (where full agreement manifests); whereas in VSO, T is poor/weak (with poor agreement on the surface). This description actually captures the facts in (8). Now, in terms of strength/weakness dichotomy, what is strong for Chomsky is weak for Soltan and vice versa.
Thus, we cannot explain the agreement-subject position facts in Arabic following POP/ POP Extensions model. (2017) Gallego (2017) proposes an interesting labeling-based approach to analyze EPP, without making recourse to Chomsky's feature strength. He argues that SpecTP must be filled in languages like English not because T is poor, but because T is actually a copy of C. Gallego follows POP idea that copies cannot label. According to Gallego, the distinction between NSLs and non-NSLs with respect to EPP boils down to a bundling parameter: whether C and T are bundled (Note 12) together or independently. In Non-NSLs, C and T are bundled together in the lexicon, i.e., there is only one head: T is just a copy of C. in NSLs, on the other hand, C and T are two independent heads in the lexicon (Note 13). Though intriguing, Gallego's account unfortunately predicts that subjects consistently move to SpecTP in NSLs for the sake of labeling the structure in (6), which is not borne out in Arabic. (2017) Goto (2017) points out that Chomsky's labeling approach to the EPP parameter (especially in POP Extensions) is stipulative a lot of stipulations. First, appealing to strong/weak parameter of T violates well-established minimalist guidelines. Goto refers to Chomsky's (1998) statement that "strange properties" such as strength should be eliminated from linguistic theory of FL. Another question raised by Goto relates to the asymmetry that arises in Chomsky's model between T and R, both non-phase heads. Goto wonders why only T has the strong/weak parameter when phases should have the same properties, suggesting that dispensing with it is a minimalist gain (Note 14).

Goto
Though maintaining a labeling approach to the analysis of EPP, Goto (2017) rejects the strong/weak parameter on T proposed in POP Extensions. Instead, he adopts Fukui and Takano (1998) and Toyoshima's (2001) Head-to-SPEC movement hypothesis (Note 15). According to this hypothesis, every movement-be it of a phrase or a head-targets a spec position (Note 16). Further, following Alexiadou and Anagnastopoulou (1998), Goto assumes that EPP in NSLs can be satisfied by head movement of v to T, although the movement here targets SpecT. According to, Goto, the parameter that sets non-null languages like English, where SpecTP must be filled by (E-/I-)merging a DP in Spec apart from null-subject languages as Arabic, where SpecT might not be overtly filled, is overtness of agreement morphology on the verb. This is how the scenario goes on. In non-null subject languages like English, V moves to SpecTP (Note 17), but because the verb is not overtly morphologically inflected for phi-features, it cannot label the SpecTP construction (phi, phi). In null subject languages, on the other hand, the verb is fully inflected, and therefore movement of the verb to T serves to label the SpecTP structure.
The main problem for Goto's analysis is that it implicitly assumes a lexicalist approach. This is evident from his speaking of the verb with full phi-features. The standard view is that Morphophonological realization of features does not take place in syntax. In fact, frameworks such as Distributed morphology and nanosyntax argue that even words are formed derivationally.

The Alternative
To account for the derivation of the different word orders in Arabic, I would like to propose a labeling approach, coupled with feature inheritance and criterial freezing. Crucially, the analysis does away with the strong/week dichotomy of T proposed in POP Extensions. We will, however, maintain Chomsky's suggestion that EPP is not universal, as also suggested in Fassi Fehri (2012). Here I would like to propose a labeling-based approach to the derivation of subjects in Arabic. The analysis makes recourse to criterial freezing and feature inheritance.

Subject Positions, EPP and Labeling
Subjects can be found in different positions in the clausal skeleton. Under mainstream analysis, subjects are E-merged in Spec,vP. This is presumably the thematic position reserved for subjects, necessitated by s-selectional properties of predicate. In addition to the thematic position, Belleti (2004) argues for a position at the left periphery of (the phase) vP, (below the TP inflectional layer, on a par with the left periphery of CP) that can host subjects (Note 18). In fact, Belleti shows that this position hosts the postverbal subjects in NSLs like Italian. A third position for subjects is the well-known Spec,TP, reserved normally for subjects in English-type languages. Subjects may also target a position above Spec,TP (Note 19) (and below CP), known as Spec,SubjP (Cardinelleti, 2004;Rizzi, 2004Rizzi, , 2006. Similarly, discussing the syntax of copular constructions, Shlonsky and Rizzi (2018) argue for two subject positions in the higher inflectional domain (apart from another (criterial) subject position outside of PredP). Aoun et al. (2010) review the different analyses that have been proposed to account for the distribution of subjects in Standard Arabic and modern Arabic dialects. As these linguists show, in the SVO word order, there are two positions that may host the subject: SpecTP or SpecTopP, depending on whether the preverbal DP is analyzed as a subject or a topic (Note 20). In the VSO word order, the subject may be in SpecVP, the thematic position (Note 21). However, based on facts from Moroccan Arabic related to the syntax of copular constructions, Aoun et al. propose that the subject may be in SpecTP, even in the VSO word order. Hence, the positions that Aoun et al. refer to as hosts for the subject are: Spec,VP, Spec,TP and SpecTopP.

Derivation of SVO
Let's recall the initial structure we get when a subject is (E-)merged, reproduced in (9).
Note that this structure cannot be labeled. A way to salvage the (labelibility of the) structure, suppose that the subject DP moves. As a consequence, the order that arises is SVO, and, importantly, the symmetric structure β is successfully labeled. The question that is generally raised here is what triggers movement of the subject. Some linguists (e.g., Fassi Fehri, 1993;Aoun, 2010;Saeed, 2011) claim that movement of the subject is triggered by an EPP on T. Others, e.g., Soltan (2007), claim that the preverbal DP in the SVO order is directly merged in an A-bar position (Note 22).
Here I would follow Chomsky's labeling framework which dispenses with the EPP as a principle of FL (Note 23). In fact, Fassi Fehri (2012) questions the universality of EPP; he argues that EPP, if it exists, does not apply in Arabic. Fassi Fehri demonstrates that there is no Nominative subject in the preverbal position in Arabic, on a par with the one found in non-null subject languages such as English. For instance, an overt expletive cannot appear in raising and modal constructions (ideal structures for postulating expletives) (Note 24).
(10) a. badaa anna r-rajul-a qaliq-un  Alahdal (2018). Following a well-established fact to the effect that preverbal subjects in NSLs show, in addition to their A-properties, A-bar properties (e.g., Ordonez & Trevino, 1999;Rizzi, 2006;Soltan, 2007), Alahdal argues that SpecTP becomes an A-bar position when T inherits EF (a criterial feature in Rizzi's sense, say TOP, which gets satisfied by movement of a discourserelated element to SpecTP) from C (Note 25). Notice that SpecTP turns out to be a criterial position in this case.
After the subject is E-merged, T is merged (along with C). Now T inherits its features from C. Suppose that T also inherits a criterial feature TOP from C and in turn SpecTP becomes an A-bar position. Now, the structure in (6) cannot be labeled. Suppose that the subject DP moves to SpecTP, triggered, not by EPP, but by TOP on T.
Interestingly, an indirect consequence of this movement is that the structure downstairs is successfully labeled as both XPs-the subject DP and TP (more accurately, T) share a prominent feature TOP.

Derivation of VSO
In the previous section we saw how labeling can help derive the SVO word order in Arabic. We saw that one of the options that lead to labelability of the structure in (6) is when the subject DP moves, a scenario Chomsky takes to deduce EPP-driven movement from a labeling requirement. Another possibility is that some element other than the subject moves. Suppose that v moves to T, leaving behind There are two scenarios to derive the VSO word order in Arabic. Under one scenario, the subject DP remains inside the lexical projection, i.e., in SpecvP, while the verb moves to T (e.g., Koopman & Sportiche, 1991;Mohammed, 2000;Benmamoun, 1992;Fassi Fehri, 1993). In the labeling framework, however, movement of the verb head does not seem to salvage the structure, as LA will find D as the closest head and then the structure will be labeled DP, an unwanted result. Another problem that arises here is that V-to-T movement takes place only in the past tense constructions, an idea proposed in Benmamoun (2000).
However, based on copular sentences and existential constructions, Aoun et al. (2010) argue that even in VSO constructions in Moroccan Arabic, the subject seems to be outside the lexical projection, dubbed FP. In fact, the idea that the vP has a left periphery on a par with the CP phase has been proposed by Jayaseelan (2001), Belleti (2004) This shows that postverbal subjects might carry new information focus. So we can safely suggest that the subject DP in VSO constructions is in SpecFocP at the left periphery of vP. Again in labeling terms, postverbal subjects are in a criterial position.
Another scenario under which the structure in (6) above can be labeled is when vP moves rather than the subject DP (Note 26). This seems to be the case in VOS constructions. The adverb in (5), ideally thought to be adjoined to vP, is placed above the subject, which might indicate that the subject remains in situ. If this derivation is on track, then this answers Chomsky's inquiry in POP Extensions as to why it is only the subject that moves out of vP and not vP.

Conclusion
In this paper it has been shown that Labeling Algorithm (LA), as a minimalist construct, motivated by a SEM interface condition, along with criterial freezing, serves derive the different word orders in NSLs, Arabic in particular. What is interesting about the current proposal is that it does not resort to the 'non-minimalist' parameter of strong vs. weak T.