1. Introduction
Wh-movement, focus fronting, and topicalization are posited to involve some feature-checking relation between a head, whether it be an interrogative complementizer (CQ), a focus head, or a topic head, and an element that typically moves to the specifier (Spec) positions of the projections of such respective heads (Chomsky Reference Chomsky, Culicover, Wasow and Akmajian1977, Reference Chomsky1986, Reference Chomsky, Hale and Keyser1993, Reference Chomsky1995, Reference Chomsky2015; Cinque Reference Cinque1981; Cheng Reference Cheng1997; Ouhalla Reference Ouhalla, Eid and Ratcliffe1997; Rizzi Reference Rizzi and Haegeman1997, Reference Rizzi, Cinque and Salvi2001, Reference Rizzi2004). Although relevant research has explored the syntax of constructions involving A’-movement, the incremental derivation of these constructions has received less attention (Bošković Reference Bošković2007; Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001; Nunes Reference Nunes2021, Reference Nunes2022). For instance, one challenging aspect of constructions involving A’-movement relates to the restrictions that are placed on the extraction of embedded wh-subjects (Nunes Reference Nunes2021, Reference Nunes2022). For example, when the subject wh-word is extracted in English, the embedded complementizer must not be used. This particular issue can be illustrated with the following English example:
The extraction of the embedded wh-subject which boys in (1) is permissible when the complementizer (C0) of the embedded clause that is not overt. Interestingly, in Jordanian Arabic (JA), wh-subject extraction exhibits a markedly distinct behavior in similar constructions – the complementizer (C0) of the embedded clause (ʔinn) must be overt with full agreement morphology. Before presenting supporting examples, it is important to emphasize that C0s in embedded declarative clauses may be null or overt in JA, as illustrated in (2a, b). If the C0 is overt, however, it must show full agreement with the embedded nominal, typically the subject, as shown in (2a, c, d).Footnote 1 , Footnote 2
Importantly, the lack of full agreement on the C0 ʔinn in JA results in ill-formedness, as seen in (2c, d).Footnote 3 In Modern Standard Arabic, in comparison, the C0 ʔinna may be null or overt but does not show any form of agreement morphology, as in (3).Footnote 4
If ʔinna is overt, as in (3a), it assigns accusative Case to the subject, but if it is null, as in (3b), the subject receives nominative Case (for more discussion on the distinction between JA and Modern Standard Arabic in this regard, see Jarrah Reference Jarrah2019, Reference Jarrah2020).
As far as wh-subject extraction in JA is concerned, the C0 of the embedded clause (ʔinn) must be overt and show full agreement morphology. The following JA question illustrates this point.Footnote 5
Notice also that the embedded C0 ʔinn must bear an inflectional suffix that copies the phi-features (φ-content) of the extracted subject; otherwise, the resulting sentence would be ungrammatical, whether the embedded C0 ʔinn bears a suffix with a default φ-content (third-person masculine singular) or zero agreement morphology, as evidenced by the following examples:Footnote 6
Therefore, subject extraction in JA is allowed when the C0 ʔinn fully and overtly agrees with the extracted wh-phrase.
A similar behavior of complementizer agreement is manifested in constructions with topicalization. The C0 ʔinn must be overt and fully agree with the subject of the embedded clause, as shown in the examples in (6).
Interestingly, the C0 ʔinn is found to agree with the closest goal that is typically the subject but can be the object if there is no intervening subject, as seen below.
This also applies to constructions with focused or topicalized dislocated objects, as we will show later.Footnote 7 In Section 3, we offer a unified account of the JA data provided above, appealing specifically to Chomsky’s (Reference Chomsky, Hale and Keyser1993) copy theory of movement as well as Jarrah, Al-Deaibes & Hammouri’s (Reference Jarrah, Al-Deaibes and Hammouri2024: 254) Agree Identification Condition (AIC) ‘which requires a phonologically null goal to be φ-identified through a co-varying φ-inflection on its probe’. We assume that the obligatorily overt C0 ʔinn establishes an Agree dependency with an unpronounced copy of the extracted element in the CP area in sentences that involve wh-movement and focus fronting and with a silent pronoun (pro) in sentences that contain topicalized dislocated elements. Moreover, we will shed light on how our findings compare to related phenomena in typologically diverse languages.
The organization of the paper proceeds as follows. Section 2 provides an overview of the literature on (long-distance) A’-movement, focusing on the derivation of this type of movement. Section 3 discusses the licensing of complementizer agreement with nominals in the left periphery in JA. We show that JA grammar exhibits obligatory complementizer agreement with A-bar elements due to the AIC (Jarrah, Al-Deaibes & Hammouri’s Reference Jarrah, Al-Deaibes and Hammouri2024). This condition requires the probe to bear an agreement inflection when the goal is phonologically null or when the probe agrees with a chain involving two silent links, such as when the complementizer agrees with a wh-subject or focused subject. Section 4 sheds light on how our findings align with or diverge from related patterns observed across typologically diverse languages. Finally, Section 5 concludes the paper, adding support to the view that the morphological expression of Agree relations is dictated by interface conditions, which are also shown to necessitate an overt complementizer in cases where the embedded element (typically the subject) is extracted.
2. Setting the stage
Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001) offered an account of long-distance A’-movement within the framework of the phase theory. An interrogative CQ head of an embedded clause, as in (8) below, cannot directly attract an object wh-phrase to its Spec position, as illustrated in (8b), due to the blocking effect triggered by the presence of the head of the lower phase, i.e., v0, giving rise to a fatal violation of the so-called Phase Impenetrability Condition (PIC), which is provided in (9).
Therefore, the wh-phrase what should first move to the specifier of the verb phrase (Spec,vP), where what becomes accessible to CQ, as represented in (10).
The intermediate derivation in (10) is possible, under Chomsky’s (Reference Chomsky and Kenstowicz2001) view, because the phase head (v0) can be optionally assigned an EPP-type feature that allows what to move to its outer Spec position (Spec,vP), leaving a copy behind in its thematic position (Chomsky Reference Chomsky, Hale and Keyser1993). Such an EPP-feature assignment is permitted only when it is necessary for producing the optimal outcome, as schematized in (11).
Once what lands in Spec,CP of the embedded clause through successive-cyclic A’-movement, and the EPP-feature on the phase heads (C0 and v0) is checked, as in (11), the computation converges at the interface levels of the Phonological Form (PF) and Logical Form (LF).
Although Chomsky’s model provides an invaluable insight into the intermediate steps of A’-movement, it invokes lookahead computations (Bošković Reference Bošković2007, Nunes Reference Nunes2021). To clarify this point, consider the following data and representations in (12) and (13), both from Nunes (Reference Nunes2021: 4).
According to Nunes, the EPP-assignment of the lowest v0, as in (13a), leads to convergence in instances like (12a) but not in instances like (12c), for this causes the derivation to crash at both PF and LF, as indicated in (12b). Nonetheless, the necessity of the EPP-assignment in (13a) may be determined only upon the merger of the heads of higher phases, namely, after the structure of (13b) has been constructed. In other words, the EPP-assignment in (13a) produces a convergent structure if the configuration of K in (13b) merges with elements like John as in (12a), but not with ones like who as in (12b). This goes against the tenets of the PIC and, also, violates Chomsky’s (Reference Chomsky1995: 225) Inclusiveness Condition, which necessitates that the output must not contain anything that is not part of its input, i.e., the properties of the lexical items.
Bošković (Reference Bošković2007) attempted to overcome this issue in Chomsky’s (Reference Chomsky and Kenstowicz2001) analysis, proposing that in languages that involve wh-movement it is the wh-element, rather than the phase head, that hosts a probing uninterpretable feature [u f]. As a result, the wh-element must move successive-cyclically to Spec,CP, so its [u f] feature is checked against the [u f] of the interrogative CQ. This way, there exists no violation of the Inclusiveness Condition because the [u f] is technically a part of the lexical properties of the wh-word. Additionally, instances with partial movement of wh-elements, as in (12b), can be safely eliminated on the ground that the [u f] on what is not checked, a matter that causes the derivation to crash.
Although Bošković’s (Reference Bošković2007) approach to wh-movement is more compatible with the effects of the PIC, it, according to Nunes (Reference Nunes2021, Reference Nunes2022), appears to fall short of explaining that-trace effects involved in constructions like (14).
Under Bošković’s (Reference Bošković2007) view, the structure in (14), drawing on Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000), should converge if who is specified for [u f], irrespective of whether the embedded C0 head that is present or absent, contrary to fact.
Nunes (Reference Nunes2021, Reference Nunes2022), drawing on Bošković’s (Reference Bošković2007) and Chomsky’s (Reference Chomsky and Kenstowicz2001) models as well as his earlier work (Nunes Reference Nunes, Kosta, Franks, Radeva-Bork and Schürcks2014, Reference Nunes2016), adopts a middle-ground approach, allowing an edge feature (EF = an alternative of EPP) to be optionally encoded on either the phase head or the moving wh-element, which is subject to parametric variation. On the one hand, the EF is optionally part of the lexical specification of wh-words in languages permitting wh-in-situ questions, such as Brazilian Portuguese, where the wh-element bearing the EF must successive-cyclically move to its targeted Spec,CP position but remains in situ otherwise. On the other hand, the EF is optionally encoded on strong phase heads in English-type languages that do not permit wh-words to remain in situ except for constructions with multiple or echo wh-questions. Centering on English-type languages, Nunes argues that a phase head, entering the computation with a lexically specified EF, assigns that EF to a wh-element, as represented in (15) from Nunes (Reference Nunes2021: 8).
When a wh-word is assigned an EF by a phase head, as in (15), it inherits that feature. The lexical features of the assigner and the assignee undergo some sort of a rearrangement, which in turn should feed the computation without violating the Inclusiveness Condition. After the wh-word gets the EF, it is then required to move to a position, where it can license its EF, the uF in the sense of Bošković (Reference Bošković2007); otherwise, the derivation is doomed to crash. Additionally, Nunes’ proposal that strong phase heads are only optionally specified for an EF can account for the (un-)grammaticality of constructions with either a covert or an overt C0 head such as (14) above. Under this view, the EF is not specified on the overt that in wh-constructions involving wh-subject extraction, as shown in (16).
Since that lacks the EF to begin with, it does not assign it to who, and therefore bans who from moving to its edge, which explains the crashing fate of (16). As Nunes points out, this is based on the reasoning that a strong phase head allows a wh-word to move through its edge only if the head assigns an EF to the wh-word. By contrast, the EF is optionally specified on null Cs (∅), and consequently wh-questions, such as (17) converge at the PF-LF interface because the wh-subject after receiving the EF from the embedded null C0 must move up crossing the edge of that C0.
Nunes (Reference Nunes2021) has also extended his approach to account for A’-movement in Irish. Consider the following examples.
Nunes argues that the lower C0 aL in (18) carries a valued EF [EF:val] but the higher C0 in (18) and both C0 heads in (19) bear an uninterpretable EF [EF:u]. In any case, for the C0 to be able to lexically license A’-movement via its edge, it must overtly appear as aL and be specified for an EF whether inherently valued or not. It is reported that aL bears [EF:val] in unmarked situations but [EF:u] in marked situations. To justify the licensing of A’-movement via the edge of a phase head with [EF:u] like aL [EF:u], Nunes (Reference Nunes2021: 23) proposes that ‘a language will only formally specify that a phase head permits movement through its Spec if other phase heads of the same type do not’. In other words, a language resorts to this strategy only when other options are unavailable.
Although this approach advanced by Nunes (Reference Nunes2021, Reference Nunes2022) circumvents some issues of A’-movement in languages like English and Irish, it falls short of explaining relevant JA data presented in the previous subsection for two major reasons. Firstly, a wh-element, according to Nunes, is assigned an EF only once by the closest strong phase head, in accordance with the PIC, which in turn gives it a passing ticket to successive-cyclically move up crossing the edges of higher phase heads. This approach works well for English, as we can see in (20) from Nunes (Reference Nunes2021: 11).
Given that null Cs, under Nunes’ view, bear an EF but the overt that does not, once the extracted who in (20) receives an EF from the lower C∅, it becomes indifferent to the type of C0 it passes across because it already has the license to move to the matrix Spec,CP. Such a line of reasoning, nonetheless, fails to apply to JA data; see (21).
The examples in (21) show that in wh-questions containing two embedded clauses the C0 head in all of the embedded clauses must be overt and carry φ-features agreeing with the extracted subject, as substantiated by the well-formedness of (21a) versus the ill-formedness of (21b–d). In other words, in JA the extracted wh-phrase seems to be always sensitive to the type of C0 it crosses in higher phases, unlike what we saw in the English example in (20). A reviewer suggested that the description of JA – namely, that all crossed complementizers must bear agreement – closely resembles Nunes’ account of Irish complementizer agreement, where C0 bears an unvalued EF. The overarching idea behind the licensing of A’-movement in JA indeed aligns with that proposed for Irish. The two languages, however, employ different mechanisms to accomplish this task. For instance, JA manifests C0 agreement with extracted subjects and objects, albeit under distinct conditions, as will be shown later.
Secondly, Nunes’ approach primarily focuses on cases involving A’-movement. In JA, however, the appearance of agreeing φ-features on the C0 ʔinn is triggered not only by wh-movement (21) or focus fronting (22) but also by topicalization (23). (Note that focused elements are indicated by capital letters.)
The agreement between the C0 and the focused element is also borne out in multiple embedded sentences, as illustrated in (24).
This agreement pattern also holds for the resumptive subject pro associated with topicalized, left-dislocated elements, as in (25).
Overall, the agreement paradigm in JA is very consistent in constructions involving wh-movement, focus fronting, and topicalization.
As far as Irish is concerned, Nunes has briefly touched on cases where a resumptive pronoun is A’-bound. See example (26).
Drawing on McClosky (Reference McClosky, Epstein and Seely2002), Nunes assumes that the C0 aN carries a traditional EPP feature but not the type of EPP feature that Chomsky (Reference Chomsky, Martin, Michaels and Uriagereka2000) associates with successive-cyclic movement. Additionally, the EPP feature of aN is assumed to be checked only by via the external merge of an ghirseach ‘the girl’. This is so because na síogaí ‘the fairies’ and í ‘her’ already have their Case features valued and thus cannot move up to check the EPP feature of aN. In comparison, constructions with topicalization in JA manifest C0 agreement through distinct mechanisms. Consider (27).
In JA, unlike in Irish, the C0 agrees with the closest goal, the subject in (27a) but the object in (27b). The parallelism of the behavior of ʔinn in various types of JA sentences like those observed in (21)–(25) and in (27) calls for a unified analysis. To accomplish this task, we call on Jarrah, Al-Deaibes & Hammouri’s (Reference Jarrah, Al-Deaibes and Hammouri2024) AIC, which requires a probe to bear a φ-inflection of the goal when the goal is not pronounced (in the absence of any overt goal DPs within a phase). Such a condition will be considered in reference to Bošković’s (Reference Bošković2007), Chomsky’s (Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001), and Nunes’ (Reference Nunes2021, Reference Nunes2022) insights on phases and movement. This will be the focus of Section 3 below.
3. Complementizer agreement with left peripheral nominals in JA
In JA wh-questions involving subject extraction from embedded clauses, the realization of ʔinn, along with its affixed agreement morphology, is constrained by syntactic factors, as shown in (28a–c), which we have already seen in (4) and (5).Footnote 9
The C0 ʔinn, as pointed out earlier, must be overt and obligatorily carry agreement morphology whether it is local or non-local to the subject extraction site, as shown in (28) and (29), respectively.
The same paradigm also applies to constructions featuring focus fronting, as shown in the examples in (30) and (31).
The interesting question that arises here is why the embedded C0 must be overt and fully copy the φ-content of the extracted element, as seen in (28)–(31). What is equally puzzling here is the fact that declarative sentences permit a covert complementizer as shown in (32) below, but sentences with extracted subjects do not, as was seen in (28b), (29b–d), (30b), and (31b–d) above.
First of all, the JA facts in (28)–(32) cannot be explained under (any of) Bošković’s (Reference Bošković2007), Chomsky’s (Reference Chomsky and Kenstowicz2001), and Nunes’ (Reference Nunes2021, Reference Nunes2022) A’-movement models, as we already mentioned in our discussion in the previous subsection. Alternatively, one may claim, given the systematic agreement pattern observed in (28)–(31), that A’-movement in JA is seemingly possible only when there exists an overt embedded C0 fully agreeing with the moving element, whether extraction is due to wh-movement or focus fronting. Such a claim is, nonetheless, not warranted either, simply because the agreement paradigm in (28)–(31), as we mentioned earlier, is exactly the same as that found in constructions lacking subject extraction but involving topicalization, as exemplified in (33) and (34).Footnote 10
To address all these questions, we draw upon Chomsky’s (Reference Chomsky, Hale and Keyser1993) copy theory of movement and Jarrah, Al-Deaibes & Hammouri’s (Reference Jarrah, Al-Deaibes and Hammouri2024) AIC, which is an interface condition that requires a φ-inflection on the probe when the goal is phonologically null. This condition ensures that the identity of the goal is φ-identified at the PF interface, aligning with Jarrah’s (Reference Jarrah2019) and Miyagawa’s (Reference Miyagawa2009) arguments that an Agree chain should be morphologically recorded at PF. The AIC is an interface condition that requires the probe to bear φ-inflection reflecting the features of the goal when the goal is phonologically null. By contrast, when the goal is overt, the AIC is vacuously satisfied since the goal itself appears at PF and serves as the morphological realization of the Agree chain. This analysis supports the view that the decision to use a φ-inflection on the probe is ruled by PF interface conditions.
We essentially assume that in JA sentences involving subject extraction from an embedded clause to a matrix clause (due to wh-movement or focus fronting) the obligatorily overt complementizer of the embedded clause establishes an Agree dependency with an unpronounced copy of the extracted subject it c-commands. This Agree dependency between the C0 ʔinn and the subject is triggered because ʔinn is endowed with a set of uφ-content that should be valued before the sentence derivation converges at the interface levels (see Carstens Reference Carstens2003, Haegeman Reference Haegeman1994, Hoekstra & Smits Reference Hoekstra and Smits1999, van Craenenbroeck & van Koppen Reference van Craenenbroeck and van Koppen2002, Zwart Reference Zwart1993).Footnote 11 We assume that such an agreement operation is motivated by the AIC. This Agree dependency between ʔinn and the subject can be schematized as follows (irrelevant details are skipped):
As a result of this Agree dependency between the C0 ʔinn and the subject, the uφ-content of C0 ʔinn is, for example, specified as [3p.m.pl] when the subject is a plural masculine element. The crucial issue here is that when the wh-subject or the focused subject moves from tense phrase (Spec,TP) of the embedded clause to the root clause, it leaves behind a copy of itself. This movement should take place before the relevant structure (the embedded clause) is sent to the interface due to the effects of the PIC. Once the wh-subject or focused subject moves to the root clause via intermediate Spec,CP, and given the assumption that the complement domain of a phase is shipped to the interface once it is complete, the Agree dependency between C0 ʔinn and the wh-subject or focused subject is interpreted in the PF interface as a dependency between a probe and a silent goal, as schematized in (36). Therefore, according to Jarrah, Al-Deaibes & Hammouri’s (Reference Jarrah, Al-Deaibes and Hammouri2024: 254) AIC ‘which requires a phonologically null goal to be φ-identified through a co-varying φ-inflection on its probe’, the C0 ʔinn must carry an overt inflection that displays the φ-content of the silent goal.
The Agree inflection of the goal (the subject in this case) on the C0 ʔinn is a necessary step to identify the properties of the silent copy of the subject within the embedded CP phase at PF. In other words, the syntactic operation in (36) is required in order to record (i.e., phonologically manifest) the identity of the silent subject DP at PF.
Let us now consider in more detail how the AIC can be applied to the JA phenomena at hand, beginning with the step-by-step derivation of wh-questions. Witness the sentence in (37a) and its step-by-step derivation in (37b), with all irrelevant details ignored.
We assume that in the tree structure (37b) the wh-phrase ʔaj bana:t ‘which girls’ undergoes cyclic A’-movement to the matrix Spec,CP in order to check/value its wh-feature in a Spec-head configuration with the root C0 (Bošković Reference Bošković2007; Nunes Reference Nunes2021, Reference Nunes2022), leaving behind unpronounced copies, in accordance with Chomsky’s (Reference Chomsky, Hale and Keyser1993) copy theory of movement. Notice that the subject wh-word ʔaj bana:t passes through the edges of strong phases, in line with Nunes’ (Reference Nunes2021, Reference Nunes2022) step-by-step derivation of A’-movement. Additionally, prior to the completion of the embedded CP phase, namely before the domain of the embedded C0 (the lower TP) is shipped to Spell-Out (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, Reference Chomsky and Kenstowicz2001), the C0 ʔinn, that bears uninterpretable features [u f] that need to be checked/valued by interpretable instances of those features [i f], probes the lowest copy of ʔaj bana:t situated in Spec,TP. Once the DP ʔaj bana:t moves to a higher position, and the derivation is sent to the interface, the PF reads off the Agree relation between the C0 ʔinn and the copy of subject wh-word as an agreement between a probe and a silent goal, triggering the effects of the AIC. Therefore, a φ-inflection of the goal should appear on the probe.Footnote 12 This syntactic operation satisfies the identification property in that the identity of the lowest copy of the wh-phrase is recorded on ʔinn at PF. Importantly, if the AIC is violated, then ill-formedness immediately arises, as can be seen in (28b) and (29b–d) above. An anonymous reviewer asked whether the agreement marking on C0 in JA might be purely a post-syntactic (PF) phenomenon, as proposed in some earlier research, including Ackema & Neeleman (Reference Ackema and Neeleman2004), Haegeman & van Koppen (Reference Haegeman and van Koppen2012), Kathol (Reference Kathol2001), Miyagawa (Reference Miyagawa2009), and Zwart (Reference Zwart, van Koppen, Hendriks, Landsbergen, Poss and van der Wal2006). In Flemish, for example, complementizer agreement may occur only when the agreeing C0 is adjacent to the subject at PF (38a). However, if the C0 is separated from the subject, agreement cannot hold (38b).
This adjacency condition, nevertheless, is not required for agreement in JA, as in (39).
The examples in (39) show that in JA, unlike in Flemish-type languages, the agreement between a probing C0 and its goal takes place within narrow syntax, rather than being solely a post-syntactic (PF) effect. This conclusion does not contradict the AIC because we assume that the AIC stipulates that C⁰ must exhibit agreement inflection at PF that reflects the goal’s φ-features; this inflection is, thus, a consequence of operations in narrow syntax (see Carstens Reference Carstens2003).
Up to this point, our discussion has centered on wh-movement. Nevertheless, since both wh-questions and focus fronting rely on A’-movement, the analysis naturally extends to cases involving subject focus fronting as in (40), repeated from (30a).
Similar to constructions with the subject wh-word, as in (37b), the focused subject BANA:T ‘girls’ in (40) moves to check/value its focus feature (Chomsky Reference Chomsky2015). It undergoes successive-cyclic movement to the Spec position of focus phrase of the root clause (Spec,FocP). The C0 ʔinn agrees with the focused element bana:t while the latter is in Spec,TP of the embedded clause. Given the assumption that the focused element bana:t moves to the matrix clause, the Agree dependency between the C0 ʔinn and the focused bana:t includes a silent goal, which according to the AIC must have a φ-inflection on its probe in the PF interface. Therefore, we propose that it is the effects of the AIC that rule in sentences like (30a)/(40) and (31a) but rules out those like (30b) and (31b–d).
In both types of constructions (i.e., wh-extraction and focalization), we assume that the probing C0 ʔinn, as per the AIC, must fully inflectionally display the φ-features of the goal (the subject), only when the subject lacks phonetic content. This justifies why the C0 ʔinn must both be overt and fully agree with the extracted subject.
In constructions with no subject extraction or subject focalization, the embedded C0 ʔinn may either be phonologically null or fully agree with the embedded subject, as shown in the following examples.
In these constructions, the Agree relation is established between the C0 ʔinn and the subject, which is located in Spec,TP. The goal of the Agree dependency between ʔinn and the subject is not silent once the relevant derivation is shipped to the interface levels, hence no AIC effects arise. Therefore, ʔinn may not even surface as an overt element in the Agree chain between ʔinn and the overt subject (i.e., the subject itself).
Moreover, our current approach has a further benefit in that it also captures instances involving long-distance movement, as evidenced by the examples in (42) and (43) repeated from (29) and (31), respectively.
The well-formedness of (42a) and (43a) versus the ill-formedness of (42b–d) and (43b–d) can, once again, be accounted for by appealing to the AIC effects. The Agree dependency between the C0 ʔinn and the subject wh-word or the focused subject should be recorded (phonologically manifested) as a φ-inflection of the goal on the probe. We assume that agreement between ʔinn and the goal takes place when the latter lands in Spec,TP prior to moving to the CP layer.
Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024) address instances where the effects of the AIC would arise. Such instances occur when the verb agrees with the copy of wh-word or focused element. In such cases, there is φ-inflection of, for example, the object wh-word on the verb. Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024) differentiate between an Agree chain and a movement chain (whose links are copies of the entity moved). An Agree relation holds between a probe and a chain, not between a probe and one link of the chain, even if one link of the chain is what values the uF on the probe. Following this assumption, Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024) can account for why there exists no object agreement on the verb in such situations. Although the object, such as the wh-element ʃu: ‘what’ in (44), moves from its thematic position and leaves a covert copy of it in its base position, the chain is overt because its first link, i.e., the higher copy, is overt.
This situation crucially differs from instances where the goal is a single-link chain in the object position, which has no overt link, as illustrated in (45) adapted from Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024: 12).
The clitic -ha on the verb in (45) is reported to be φ-inflection triggered by the Agree relation between the v0 and the object pro. If an Agree chain must be recorded at PF by either the agreement inflection on the probe as in (45) or by the goal being overt as in (44), then either the φ-inflection on the probe or the goal being the whole movement chain containing an overt link (the wh-element) will suffice to satisfy the AIC. However, Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024) do not discuss cases where both links of the chain are silent. In such situation, the effects of AIC arise, as predicted. In our cases, the agree relation between the C0 ʔinn and the subject wh-word or the focused subject include an Agree relation between ʔinn and a chain whose both links are silent, given the movement of the latter to the root CP.
We have thus far discussed how the current approach can explain the complementizer-subject agreement in constructions with wh-movement and focus fronting. The question that arises here is how this agreement phenomenon can be explained in constructions that lack any form of movement, such as those involving topicalized, left-dislocated elements, as in (46), repeated from (33).
Following Jarrah, Al-Jarrah & Al-Shawashreh (Reference Jarrah, Al-Jarrah and Al-Shawashreh2022); Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024); and Ouhalla (Reference Ouhalla, Eid and Ratcliffe1997), we propose that the dislocated subject el-bana:t ‘the girls’ in (46) is a topic, which externally merges in the Spec position of Topic Phrase (Spec,TopP).Footnote 13 Under this assumption, Spec,vP of the embedded clause is not filled with the dislocated subject el-bana:t ‘the girls’ in (46). Instead, Spec,vP (and later Spec,TP of the embedded clause) is filled with a pro that acts as the subject of the embedded clause (see Soltan Reference Soltan2007). Following this line of analysis, the C0 ʔinn establishes an Agree relation with the subject pro rather than a copy of the dislocated subject that merges externally in its surface position, as shown in the following tree structure. (We suppose that the subject pro moves to Spec,TP.)
The presence of the silent pro in (47) that carries the third-person feminine plural features is evidenced by the fact that this pro surfaces overtly when it needs to be emphatically pronounced, as exemplified in (48).
An important point to emphasize here is that, in this article, we adopt Holmberg’s (Reference Holmberg2005: 538) proposal that a referential pro is a true pronoun specified for a set of interpretable features and can therefore value the uF of C0 via an Agree relation. We propose that pro in (46a)/(47) merges in Spec,vP and then moves to Spec,TP for EPP. This movement justifies the situation that in emphatic contexts the pronoun must appear before, but not after, the verb, which is assumed to adjoin to T0 (Benmamoun Reference Benmamoun2000).
We claim that the C0 ʔinn establishes an Agree relation with pro in order to get its unvalued φ-content valued as proper. Because the goal of this relation (i.e., pro) is phonologically silent, the effects of the AIC arise. This also justifies the ill-formedness of (46b) above, reproduced below for convenience.
In (49), there is no agreement inflection at all, leading the sentence to crash for the same reason because of the violation of the AIC.
As far complementizer-object agreement is concerned, such an agreement is possible only in the absence of the subject. This is because the subject, upon agreeing with C0 in Spec,TP, is closer to C0 than the object. This justifies why we have only C0-subject agreement in (50).
If there is no subject, however, the AIC requires C0-object agreement as in (51).
The Agree operation between the C0 and the silent copy of the object in (51) may be represented as in (52), with irrelevant details skipped.Footnote 14
The diagram above illustrates that ʔinn must register agreement morphology with the moved object (the closest element in its c-command domain), as there is no intervening subject in the embedded clause. The AIC is thus satisfied, resulting in the convergence of the derivation at the interface levels.
4. Typological notes on C⁰ agreement with left peripheral nominals
We have already shown that the embedded C0 in JA must be overt and fully agree with the closest goal in constructions involving extraction (wh-movement and focus fronting) and topicalization, as shown in (53)–(55), respectively.
This agreement paradigm in JA, as discussed earlier, is motivated by the AIC. Notably, the idea that complementizer agreement of the embedded clause may correlate with a nominal in the matrix clause has already been explored in other languages. In this section, we want to situate our findings within a broader typological framework to highlight cross-linguistic patterns and generalizations. To begin with, the phenomenon of complementizer agreement with a silent pronoun/operator is found in the Bantu language Lubukusu. Diercks (Reference Diercks2013), for example, has reported that the embedded C0 in Lubukusu agrees with a null subject-oriented pronoun associated with the matrix subject, as clarified below.
Diercks has proposed that an empty operator category (labeled as OP) occupies the embedded Spec,CP position. This operator is bound by the matrix subject and enters into an agreement relation with the C0 head, as schematized in (57) from Diercks (Reference Diercks2013: 359).
A similar phenomenon of complementizer agreement is also attested in several other Bantu languages such as Chokwe and Luchazi (for details, see Kawasha Reference Kawasha, Payne and Peña2007).
In a similar vein, Zwart (Reference Zwart1993) has indicated that in Frisian (a west Germanic language) referential pro-drop of a subject in a subordinate clause is allowed if there is complementizer agreement with that subject (cf. van Alem Reference van Alem2024), as shown in (58).
Such pro-drop is, nonetheless, not available if overt complementizer agreement is absent, as in (59).
Bavarian German also patterns with Frisian in this respect, as demonstrated in (60).
Another related phenomenon is complementizer agreement with extracted and topicalized elements in Irish. Consider (61).
Observe that the C⁰ aL, as McCloskey points out, appears in clauses that involve A’-binding of a trace, whereas its counterpart aN is used in clauses where A’-binding targets a resumptive pronoun. The Irish examples in (61) show that the form of C0 varies depending on the bound element.
Overall, the complementizer agreement behaviors observed in JA, Lubukusu, Frisian, Bavarian German, and Irish suggest that there is a connection or dependency between the realization of overt agreement inflection on C0 and extracted or silent elements. In other words, if an element is extracted from an embedded clause or if it is silent, some agreement morphology associated with that element must be recorded on C0 at PF, which in one way or another corroborates the AIC we proposed for JA. How the AIC is achieved across languages is, however, subject to parametric variation. That is, languages vary with respect to the complementizer agreement strategies they adopt in their grammars. For example, embedded C0s in JA, Lubukusu, Frisian, and Bavarian German bear phi-agreement features, whereas Irish employs different morphological forms of C0.
Finally, one theoretical question that arises from this analysis concerns how information about Agree dependencies is transferred from syntax to phonology after Spell-Out. This question is particularly relevant given that our AIC makes explicit claims about the visibility of syntactic relations at the PF interface. In line with post-syntactic models such as Distributed Morphology (Embick & Noyer Reference Embick and Noyer2001, Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993), where morphophonological realization follows narrow syntactic operations, we can view the AIC as an interface condition that ensures syntactic Agree relations leave a phonological footprint. Specifically, when the goal of Agree is phonologically null, the AIC requires the probe – in this case C⁰ – to carry overt φ-feature inflection, thereby preserving the recoverability of the dependency at PF. This reasoning aligns with Jarrah’s (Reference Jarrah2019) and Miyagawa’s (Reference Miyagawa2009) proposals that syntactic chains must be morphologically ‘registered’ at PF to maintain interpretability. Crucially, when the goal is overt, the AIC is vacuously satisfied since the goal itself provides the necessary phonological content. From this perspective, the complementizer agreement patterns observed cross-linguistically reflect a broader grammatical tendency to make silent syntactic relations interpretable at the interface.
5. Conclusion
This paper explored the mechanisms underlying the licensing of complementizer agreement with nominals in the left periphery of JA. It argues that the obligatory complementizer agreement with A-bar elements in JA is a direct consequence of the AIC proposed by Jarrah, Al-Deaibes & Hammouri (Reference Jarrah, Al-Deaibes and Hammouri2024). Given the AIC, we showed that the φ-inflection of the extracted element on the C0 ʔinn is required because its goal is silent being either a copy or pro. This analysis provides strong empirical support for the hypothesis that the morphological realization of Agree dependencies is governed by interface conditions. In particular, it demonstrates that the visibility of agreement relations at PF is not an arbitrary feature of narrow syntax but is rather a consequence of interface-driven processes. The AIC, which mandates that an agreement inflection is realized when the goal is phonologically null, highlights the crucial role of PF in ensuring that agreement chains are morphologically represented. This suggests that the interface between syntax and phonology actively shapes the morphosyntactic output, enforcing specific conditions on when and how φ-features are overtly marked. Such a perspective aligns with broader theoretical proposals that emphasize the role of interface conditions in regulating syntactic operations, thereby providing a unified explanation for the distribution of complementizer agreement and other morphosyntactic phenomena in a range of languages. This analysis underscores the importance of viewing Agree dependencies as part of a broader interface system, where both syntax and phonology work in tandem to shape linguistic expressions.
Acknowledgments
We sincerely thank the editors and anonymous reviewers for their valuable time, insightful feedback, and constructive suggestions, which significantly improved the quality of this manuscript.
Open access
