The particle suo in Mandarin Chinese: a case of long X⁰-dependency and a reexamination of the Principle of Minimal Compliance

2.1 Long dependency

What distinguishes this X⁰-level dependency involving the particle suo from all other nominal elements in the literature is its correlation with relativization (see also Note 7) and its remarkable ability to extend across unlimited long distances. In essence, this dependency can cross every type of clausal boundary in Chinese, regardless of whether the construction is finite or not, control or raising, assuming that a finite/non-finite distinction holds in Chinese (Law and Ndayiragije 2017; Zhang 2016):^[11]

Due to the lack of morphological clues, identifying raising-to-subject verbs may be challenging (but see Teng [1978]).^[13] Apart from that, Examples (2)–(5) cover all types of complement clauses. In all these contexts, suo has the option of occurring in the matrix clause, while the gap of the RC it is associated with resides inside the complement (e.g., in the (b) examples in (2)–(5)). Particularly noteworthy is (5c). As Chinese lacks tense morphology, one might need to rely on the choice of the matrix verb and the native speakers’ perception regarding whether the embedded verb carries an independent tense value from that of the matrix verb. In (5c), however, the two adverbials yiban ‘generally’ and ceng ‘once’ unmistakably place the two events apart in time.

More examples involving suo occurring upstairs with the matrix verb taking a finite complement are given in (6a)–(6c).

Under Ting’s (2010) analysis, suo is permitted to occur upstairs with a verb taking an infinitive but not a finite complement clause. To account for the observations in (5) and (6), it is claimed that such phenomena do not involve a long dependency on the part of suo. Instead, suo occurring in the upstairs clause is licensed by the relativization of the complement clause, and the modification marker de associated with this relativization can be dropped. The derivation underlying examples like (6a) is illustrated in (7).

According to Ting, the possible presence of de following the verb renwei “think” indicates that (7) actually has two RCs, the first of which modifies the second, making the relativization “two-layered” in a sense. The first relativization applies to the complement clause of renwei ‘think’ in (8), yielding (9a) with its structure sketched in (9b).

In (9b), what is modified is a clausal constituent, simply marked as S (though it could be a nominalized S or something of that nature). In this sense, the structure is no different from an adjunct modifying some constituent X. If one wishes to relativize anything inside X, the relativization would proceed, as is done normally, with another Op moved to the edge of the entire constituent (and thus above both S and RC). This is what happens when we relativize the object of huode ‘obtain’ in a structure like (9b). The overall derivation of (6a) is depicted as in (7), and the presence of suo is licensed by the relativization of the complement clause of the verb renwei ‘think.’ Under this approach, the modifier marker de associated with the first relativization can be dropped when the finite complement clause taken by renwei ‘think’ is relativized. If it is true that the first RC has the option of not being trailed by de as seen in (10), suo in (6a) would not be part of any long dependency but rather would locally bind the gap right after the verb renwei ‘think.’^[18]

We concur with Ting (2010) that the presence of de marks a separate RC modifying another RC and that this de can be omitted, given the optional occurrence of de in contexts with stacking RCs as in (10). However, we will argue that the structure, as in (7), cannot be extended to all instances (e.g., in (5) and (6)) where suo occurs upstairs with verbs taking a finite complement. Crucial evidence comes from instances involving the licensing of negative polarity items (NPI).

As is well-known, there’s a so-called S-structure c-command requirement on the licensing of NPI like any in English (see Hoeksema [2000, 2017]; Uribe-Etxebarria [1995] for discussion). From this requirement, the subject-object asymmetry (11) and the ban on topicalized NPIs (12) in English follow, given that the NPI any would fail to be c-commanded by the negation licensor in the surface position.

Chinese is also subject to this requirement, as shown by the unacceptability of the NPI renhe ‘any’ in the subject position (13) (Kuo 2003; Wang and Hsieh 1996) and in the topic position (14).

Anticipating our forthcoming argumentation, it is necessary to note that, similar to English (15a), NPIs in Chinese (15b) (cf. 15c) can be licensed in a long-distance fashion (Kuo 2003; Wang and Hsieh 1996).

Now, returning to the licensing of suo, consider Example (16a), where the embedded NPI subject renhe ren ‘anyone’ is licensed by the matrix negation, and suo occurs in the matrix clause containing the verb renwei ‘think.’ Under the “two-layered” relativization approach proposed by Ting (2010), examples like (16a) would have a structure like (16b). (cf. the derivation depicted in Example (7)). The finite complement clause of the verb renwei ‘think’ with the NPI subject would be relativized, and this relativization would be argued to license suo. Then, relativization, in turn, applies to the object in the clause modified by the RC [wo suo bu renwei de] ‘that I don’t think,’ resulting in the nominal modified by the RC which is itself modified by an RC containing suo. According to Ting (2010), the higher de following renwei ‘think’ can be dropped. Now, note that this derivation would incorrectly predict the unacceptability of examples like (16a): as shown in (16b), the NPI renhe ren ‘anyone’ would end up in a position where it fails to be c-commanded by its negation licensor since the licensor is inside an adjunct modifying the phrase containing the NPI. This prediction is not borne out.

By contrast, in the single relativization approach depicted in the structure (17), there is no such problem, as the NPI renhe ren ‘any person’ is c-commanded by the negation licensor throughout the derivation. We claim that this is a genuine instance of a long dependency between suo and the gap left by relativization.

2.2 A cross-linguistic perspective

The significance of the long dependency exhibited by suo is best assessed when suo is compared with other nominal heads situated around the T-node and associated with the object, namely clitics of the Romance type.^[20]

In addition to the traits mentioned above, Romance clitics also have the option of undergoing climbing, with the clitic in the matrix clause associated with the embedded object. When this occurs, a few more related facts have been reported, the most obvious of which is that only modal, causative, aspectual, and perceptual verbs allow CC (see de Andrade and Bok-Bennema 2017; Cardinaletti and Shlonsky 2004; Kayne 1989, 1991, 1994, 2002; Li 1990; Ouhalla 1999; Roberts 1991; Rooryck 1994; Rosen 1990; Sportiche 1996; Uriagereka 1995; Villa-García 2019). Verbs such as decide reject CC (18); as expected from its insensitivity to clause types in long dependency, no such restriction is found with suo in Chinese (19):

Another well-studied characteristic of Romance CC is restructuring. It is observed that CC is often correlated with “tense-merge” and exhibits the Specified Subject Condition (or SSC) effects. The former refers to the fact that when the clitic is placed on the matrix verb, the matrix verb and embedded verb cannot be modified by different temporal adverbs, as shown by the contrast in (20) (Napoli 1981). The latter refers to the phenomenon that the placement of the dative clitic and object clitic is not allowed to cross the main clause over the subject of the embedded clause, giving rise to the SSC effects, as in (21) (Sportiche 1996).^[23]

These facts have led to the proposal that the structural context for CC is the merge of a biclausal structure into a (complex) single predicate. This not only explains why different temporal adverbs cannot coexist in (20b) and why an overt embedded subject is incompatible with CC in (21), but also reduces CC to local head movement in a monoclausal structure. For the current discussion, the exact analysis of restructuring is not critically significant (cf. the discussion in Cardinaletti and Shlonsky [2004]; Cinque [2001, 2003, 2006]; Grano [2015]; Huang [2018]; Villa-García [2019]; Wurmbrand and Lohninger [2023]). What matters for our purposes is the observation that suo in Chinese displays none of the properties in (20)–(21) (see also Ting [2010] for discussion). As (5c) demonstrates, suo is well-formed even when the matrix verb and the embedded verb are each modified by an adverbial indicating a time frame. Furthermore, all the examples in (3)–(6) contain an embedded clause with its own subject. A Chinese example comparable to the French causative (21) is given in (22).

Still another difference between suo and Romance clitics is that the former is limited to RCs (and comparable environments), while the latter are not. It is inaccurate to consider suo’s occurrence in RCs as merely an erratic incident. As mentioned in Note 7, suo also occurs in the (long) bei-passive, which has been argued to derive from A-bar movement for entirely independent reasons. What is in common between this construction and RCs in Chinese is that both cases involve syntactic (=pre-LF) A-bar movement.^[25] In Section 3, we will propose that suo selects a relative Op in its complement position, but Romance clitics do not; nevertheless, they are both displaced T⁰-clitics.

The chart below summarizes the comparison between suo and Romance clitics.

In other words, the two types of heads (i.e. suo and Romance clitics) closely resemble each other until the concept of long dependency arises. On the surface, Romance CC appears to be a kind of long dependency, as it involves dependency beyond the embedded verbal projection. However, CC is associated with a full cluster of facts that, to say the least, cast doubt on whether true long dependency exists. By contrast, suo is entirely immune to any such restriction and thus exhibits true long dependency.

A couple of clarifications are in order. First, the very fact that suo may occur in the matrix clause while the embedded clause is finite (i.e., (5) and (6)) is sufficient evidence that this morpheme is not AgrO or any functional head near T that is intrinsically connected to the object. Otherwise, one would be claiming that the object of a finite clause triggers agreement outside the clausal boundary. If anything, suo finds the closest resemblance in pronominal clitics, which leads to our second point: Does our comparison of suo with Romance clitics weaken when suo appears optional but pronominal clitics are not?

As pointed out in Ting (2009), suo may appear optional syntactically, but semantically its use is associated with multiple communicative functions (e.g., ideational, (non-)contextual, personal and aesthetic),^[27] much like the way Romance pronominal clitics display various discourse functions (Capone 2003; Gutiérrez-Rexach 2000; Uriagereka 1995). In this respect, suo is just as optional or obligatory as the clitics in Romance.

It is instructive to compare suo with other long-distance heads in the literature (Cheng and Vicente 2013; Harizanov 2019; Harizanov and Gribanova 2019; Hein 2018). As argued by Harizanov (2019), the fronted participles in Bulgarian are displaced heads, undergo A-bar movement on a par with phrases, and can have discourse effects. Such heads can exhibit long dependency, but unlike suo, their long dependency does not hinge on syntactic A-bar movement. Section 3.4 will argue that a phrasal movement approach does not account for suo’s long dependency, emphasizing the significant role played by syntactic A-bar movement in the long dependency of a nominal head (i.e. suo in the present study).^[28]

This leaves us with the last part of the suo-clitic disparity listed in (23): that Romance clitics do not rely on syntactic A-bar movement for proper use. One naturally wonders if this is correlated with other properties among these languages. We explore this question in the next subsection.

2.3 Types of resumptive pronominal elements

Considering that suo is associated with the characteristic gap inside the RC, resulting from the movement of the relative Op, suo may be descriptively viewed as a displaced resumptive pronoun (RP). Research on RPs has identified two types (Boeckx 2003; Chao and Sells 1983; Demirdache 1991; Sells 1984), represented respectively by the English and Hebrew examples in (24) and (25) (examples taken from Demirdache [1991: 13–14]):

The English type is typically found as substitutes for gaps inside islands (24a) and must have an e-type pronoun interpretation as it rejects a QP head noun (24b). In comparison, the Hebrew type occurs where good gaps are (25a) and may act as a bound variable when the head noun is quantificational (25b). (For discussion regarding the incompatibility between an e-type pronoun interpretation and a QP head noun and its implications for analyses of RPs, the interested reader is referred to Chao and Sells [1983]; Sells [1984]). In this respect, suo clearly patterns with the RPs in Hebrew. It is evident that suo is compatible with grammatical gaps in RCs (cf. (1) through (6)). In (26) below, suo is bound to a quantificational head noun:

Meanwhile, as demonstrated in Section 2.2, suo also behaves partially like Romance clitics (cf. (23)). What deserves attention is that, when the latter occur inside RCs, they actually act like the English-type RPs. The examples below are from Italian:^[29]

In (27a), a resumptive clitic (= lo ‘him’) inside the island (= the conditional CP) rescues an otherwise unacceptable gap. This mirrors the typical role of the English-type resumptive. Furthermore, turning the head noun into a quantificational one (= ogni uomo ‘every man’ in (27b)) makes the resumptive clitic significantly worse, a characteristic of the English-type (compare (27b) with (24b)).

The English-Hebrew contrast has been attributed to whether an RP is correlated with a separate A-bar movement (cf. the references earlier). Such a correlation exists in Hebrew, explaining why the context is always conducive to A-bar movement of the relative Op. In English, however, the relative Op is base-generated in Spec,CP in relatives with RPs. The lack of movement avoids island effects, and the resumptive indicates where the offending trace would be. With this analysis, we can classify resumptive pronominal elements cross-linguistically based on two properties:

With (28), some intermingled traits of resumptives and clitics start to make sense.^[30]

Firstly, suo’s occurrence in RCs (and the long bei-passive, see Note 7) is no longer a sporadic incident in Chinese. It is simply the consequence of Chinese belonging to the same language group as Hebrew, both of them tying the use of RPs to well-formed syntactic A-bar movement. The two languages differ only in whether the RP is in the clitic form or not, just as what distinguishes Italian from English.

Secondly, the choice to manifest an RP as a T⁰-clitic or not determines whether an RP (and any pronoun for that matter) demonstrates the subject-object asymmetry in monoclausal structure. Given the standard requirement for c-command, a clitic under T⁰ (or some functional head nearby) must bind its trace, effectively ruling out the subject clitic at least in typical SVO languages. This is the case for suo^[31] and the Romance clitics (again recall (23)). On the other hand, in-situ resumptives found in Hebrew and English do not display this asymmetry. The Hebrew example in (25b) shows the subject of a finite clause (= hu ‘he’) to be a resumptive. (29) is a similar English example:

But more pertinent to the central concern of this work is the third fact—the Italian resumptive clitic remains as local as other clitics in the language, despite occurring inside an RC, as illustrated by the ill-formed (30):

Given our understanding of suo, (30) sharpens the central concern of this article: Why don’t Romance clitics behave like suo with respect to long dependency, even when they appear in a syntactic environment apparently identical to the one for suo? This brings us to Section 3, where we will argue that, under the analysis in which suo forms a big-DP with the relative Op, since the movement of the relative Op has satisfied the locality constraint, suo can have a long dependency with its underlying position in accordance with the PMC proposed by Richards (1998).

3.1 The big-DP analysis of the particle suo

Assuming suo to be a Hebrew-type resumptive, as characterized in (28), we adopt Ting’s (2010) approach (see also An and Kuo [2006]) and assign to suo Boeckx’s (2003: 28) big-DP structure of resumptive pronouns:^[32]

(31)

In the comparable structure hosting suo in (32), suo serves as the D head, taking the relative Op as its NP complement.^[33] Given that Chinese RC formation does not involve wh words, the complement NP is assumed to be a null Op under a matching analysis of RCs (cf. Chiu 1995; Ning 1993).^[34],[35]

(32)

Under this big-DP analysis of suo, examples like (1a) have a structure as in (33).

(33)

Some remarks on the derivation concerning the local dependency of suo depicted in the structure (33) are in order. Following Ting (2003, 2010), we assume, that suo, being a clitic-like element, undergoes successive cyclic movement to T or a functional head in the split-INFL domain. Notably, this movement does not carry the adjoined heads with it, analogous to excorporation, as argued for Romance-type clitics by Li (1990) and Roberts (1991, 2010).^[36] For concreteness, we adopt Ting’s (2010) labeling of the functional projection hosting suo as FP, aligning with the term used for a clitic projection by Uriagereka (1995) (cf. Cardinaletti and Shlonsky 2004). Additionally, we posit, in line with Uriagereka (1995), that the head of FP has a strong [uF] feature requiring checking/valuation by PF. Suo bears the [iF] feature, matched to its association with the F projection. Suo’s raising to the F head is attracted to value its [uF] feature. The [F] feature signifies discourse functions, as investigated by Ting (2009) in the utilization of suo across various communicative situations (cf. the description of these functions in Section 2.2).

Furthermore, apart from the v-to-V movement and the potential movement of a VP internal subject, two chains are present in the derivation: one involving the head movement of suo to adjoin to F⁰ and the other involving A-bar movement of the relative Op from NP complement position to Spec,CP of the RC. The two chains adhere to the Path Containment Condition (PCC, Pesetsky [1982], see also Richards [1998] for using the PMC to capture the PCC effects) in exhibiting nesting and avoiding the crossing of dependencies.^[37]

We shall now explore how the subject-object asymmetry in the presence of suo in monoclausal structures, as illustrated by the contrast in (1), can be accounted for (cf. Ting 2003: 134). Assuming that subjects are generated VP-internally (Fukui and Speas 1986; Huang 1993; Kitagawa 1986; Koopman and Sportiche 1991) and that all structural Case is assigned/checked/valued by functional heads via Agree (Chomsky 2001), the big-DP in VP-internal subject position has its uninterpretable nominative Case feature valued by T via Agree. The subsequent movement of the big-DP subject to Spec,TP is driven by an [EPP] feature T bears. Attempting to value the uninterpretable feature of the F head through downward movement of suo from the D head (of the big-DP in Spec,TP) to F⁰ would lead to a crashed derivation.^[38] In a system where all operations occur cyclically and derivationally, downward/lowering movement is naturally excluded (Epstein et al. 1998).

An alternative derivation for the big-DP, originating from the Spec,vP position, was raised by an anonymous reviewer and must be rejected. If we assume that suo moves into FP and the remnant of the big-DP (i.e. the trace left by suo along with its Op complement) moves to Spec,TP to check the EPP feature of T (with its Nominative Case having been valued by T), this derivation would inaccurately predict that suo is possible with local subject relativization, indicating a lack of subject-object asymmetry in monoclausal contexts. However, this proposed derivation conflicts with the observed generalization for remnant movement in (34) according to Takano (2000) and his associated explanation.

In this alternative derivation, the element that has undergone movement is the head, suo, of the big-DP. The subsequent remnant movement of the big-DP, containing the trace of suo, is anticipated to be impossible, as outlined in (34). Consequently, we must reject this alternative derivation.

We shall now shift our focus to suo’s long dependency, which can be represented as follows:

Here, t_i is the trace resulting from moving Op away and t _j is the D-gap directly forming an X⁰-dependency with suo. The question arises: why does the dependency [suo_j, t_j] appear boundless when nominal X⁰-dependencies, including Romance clitics, are typically known to be highly local?

We propose that this is due to the minimality-compliant Op movement, which exempts [suo_j, t_j] from the minimality requirement, taking into account the PMC proposed by Richards (1998).

3.2 The PMC and suo

Richards’ (Richards 1998: 601) PMC is given below:

For the PMC to work with minimality, Richards (1998: 614) defines Attract below:

The minimal domain MD of F is defined as the head H containing F and the Spec, complement and all adjuncts of H. In essence, the PMC states to the effect that if X is a member of some minimal domain MD and X satisfies some principle P, then all other members of MD become invisible for P.

The PMC is applied to a wide range of phenomena, one of which is Icelandic stylistic fronting (Richards 1998: 618):

The past participle byrjað ‘begun’ may be fronted across the tensed auxiliary höfðu ‘had’ only if the subject has been extracted from the Spec,TP. In (39b), this does happen because the subject of the RC, in the form of Op, moves to Spec,CP (and eventually to be bound by the head NP ymsa roskna trésmiði ‘various old carpenters’). In (39a), the subject trésmiðirnir ‘the carpenters’ stays in situ, and no stylistic fronting is allowed to apply to byrjað ‘begun.’

The relevant structure of (39b) for Richards is as follows:

The movement of the subject Op to Spec,CP obeys Shortest Move coded in the definition of Attract in (38). The minimal domain (i.e. TP), of which the pre-movement subject is a member, consists of all the direct components of TP, which, according to the definition of Attract, are relevant for Shortest Move. Therefore, a member of this minimal domain, specifically T, can be ignored for another operation involving T that also complies with Shortest Move. Consequently, V (= the past participle in (39b)) can move directly past T, as T is now rendered invisible to V under the PMC. In summary, the crucial point here is that “movement of V⁰ over T⁰ is possible … only when the subject has been extracted from Spec,TP, thus rendering the minimal domain TP (including its head, T⁰) invisible for Shortest Move” (Richards 1998: 619).

Upon revisiting the case of suo, we observe that the Icelandic instances of stylistic fronting, as examined by Richards for the PMC, is comparable to the scenario of suo in (35), which also involves a Shortest-Move-compliant phrasal movement (of Op) and results in the X⁰-dependency of suo being exempt from obeying minimality. Specifically, in the relevant structure (35), the minimal domain containing the good Op trace (marked by t_i) is the collection of the direct components of the object DP, with the D head (marked by t_j) being a member of this domain. Based on the earlier-discussed logic, a plausible preliminary explanation is that t_j is ignored for the purpose of Shortest Move. This would naturally account for why t_j can be separated from suo by an unlimited distance. Nevertheless, it becomes apparent that a reconsideration of Richards’ (1998) original implementation of the constraint is necessary to fully capture suo’s long dependency.

3.3 How exactly does the PMC operate in UG?

With UG being derivational,^[39] as Richards assumes (1998: 615), if an element X is ignored during the derivation due to its relevance to a good dependency and X is also the tail of a movement yet to occur, then having X ignored would not, as we might expect, let X participate in the subsequent movement freely of the constraint at issue. In the case of suo in the structural context (Examples (33) and (35)), if suo starts as D (recall the resumptive big-DP analysis of suo embodied by (32)) and gets ignored for its being a member of the minimal domain containing the good Op-trace t_i, it would become invisible to whatever functional head (T/F) attracting (the feature in) suo. Given the technical simplicity and conceptual straightforwardness of explaining the long dependency of suo via the PMC, and considering the absence of empirical evidence from Richards indicating that tails of a chain cannot be ignored, we will explore a feasible solution to maintain the PMC account.

Our proposal is to elucidate what it really means for an element to be “ignored” by a locality constraint c. Richards takes this to be a feature f becoming invisible to any subsequent application of c. However, what requires explicit clarification in his theory is that, despite the exemption of f from c, movement still occurs. In other words, when an element is said to be ignored by a constraint c, it does not mean the element itself gets ignored. In fact, the element (along with its trace) still participates in movement, but it is rendered immune to the locality constraint at issue. Consider (41), which is Richards’ (1998; 605, Example (16b)):

For Richards, the matrix C attracts the local subject and thereby obeys the Subjacency. This operation exempts the matrix C for further Subjacency considerations, including its relation with which car in the island. While this C may be exempted, it still needs to attract the in-situ object which car (or the relevant feature in it) at LF; otherwise, the latter would not be properly interpreted as a wh-phrase with the matrix CP being its scope. In conclusion, regardless of how one chooses to define the PMC and the relevant constraints, the exempted feature must still participate in motivating movement.

This clarification is sufficient to resolve the problem at hand. In a monoclausal structure (see Example (33)), with suo moving locally and creating t_j as its trace, this is well-formed, as typical clitics are: the PMC is still operational but has no detectable effects on the derivation since nothing relies on being saved via exemption. Now, consider the pertinent structure for suo in (35), representing long dependency, repeated below:

The Op NP moves from the complement position of suo through the embedded Spec,CP. Since this step of Op movement satisfies all movement-related locality constraints, the trace of Op makes all members of the minimal domain (=DP) exempt from the same constraints, including D (where suo is situated; cf. ((32)). Recall from the previous discussion that the PMC only makes a constituent immune from a constraint but does not prevent it from participating in movement. As a result, when the derivation reaches the matrix T/F, suo in the embedded clause is still attracted as usual, the only difference being that suo is entitled to the locality exemptions and thus may move directly to the matrix T/F, skipping the otherwise intervening heads such as the embedded T/F.^[40]

Interestingly, our account explains an intriguing contrast in suo’s licensing in relativization of a subject in local versus long-distance contexts. While suo’s presence is illicit where a local subject is relativized, as shown in (1b), repeated here, it can be licensed by relativization of the subject in a downstairs embedded clause,^[41] as shown in (42). In such instances, the long dependency of suo can still be captured by a parallel configuration to that in (35), which has been utilized to explain the long dependency of suo involving relativization from an embedded complement. More specifically, examples like (42) can now be represented as the configuration in (43).

In (43), the big-DP is base-generated (or first-merged) in the embedded Spec,vP and raises to the embedded Spec,TP to check the [EPP] feature of T. So far, the derivation is identical to that involving the impossible association of suo with local subject relativization. However, in cases like (43), the FP that attracts suo’s movement to value its strong [uF] feature is upstairs, in a clause different from the one containing the subject big-DP. A-bar movement in Chinese does not encounter any difficulties taking place out of an embedded subject (see Cheung [2008] for preposed wh-phrases being derived by movement), as demonstrated in (44). Consequently, it is reasonable to claim that the relative Op can move out of the big-DP in the embedded Spec,TP and land in the matrix Spec,CP, without issue. This step of Op movement allows for the licensing of suo’s landing in the matrix FP in examples like (42), in accordance with the PMC, which is comparable to the long dependence of suo licensed by embedded objects, as seen in examples such as (5) and (6).

3.4 Ruling out alternative approaches to LD dependency of suo

In this subsection, we will argue against several alternative approaches to the long dependency of suo, as suggested by an anonymous reviewer. Firstly, is it conceivable for suo to be base-generated in the matrix TP domain and relate to the relativization site in the embedded clause? One previous analysis along this line is Chiu (1995), where suo is base-generated as a functional head comparable to AgrO, and its overt realization is triggered by movement of the relative Op passing through the specifier of the functional projection. However, as pointed out toward the end of Section 2.2, suo cannot be AgrO or any functional head near T that is intrinsically connected to the object, since the object of a finite clause couldn’t possibly trigger agreement outside the clausal boundary.

Now, let’s consider another potential derivation for the long dependency of suo in which the big-DP hosting suo and its relative Op complement undergoes A-bar movement to the matrix Spec,FP, followed by Op movement to the matrix Spec,CP, leaving suo stranded in the matrix FP. This derivation would be akin to quantifier-float with A-bar movement in West Ulster English. As McCloskey (2000) argues, in the underlying structure [_QP all wh-NP] with the Q head all, the whole QP containing the wh word moves (45a) or the wh word may move by itself, stranding the quantifier (45b).

We will argue that the big-DP headed by suo is unable to proceed in A-bar movement, contra the expectation under the approach of quantifier-float off wh-elements. Neither the complement of suo nor the suo head itself is endowed with the feature that can license A-bar movement of the big-DP in a pied-piping fashion. First of all, the complement of suo, i.e. the relative Op in Chinese, does not have the wh-feature that can percolate up (see the treatment in Chomsky [1973]; Cowper [1987]; Grimshaw [2000]; Webelhuth [1992], among others) to enable the big-DP to undergo long-distance movement in a pied-piping fashion. In English RCs, while the wh relative Op can undergo pied-piping and move in a larger PP (46a) or DP (46b) to Spec,CP in the RC, the relative Op without the wh feature cannot. There is thus good reason to believe that the big-DP headed by suo could not move as a wh phrase by having the relative Op complement.

Furthermore, could the [F] feature of suo percolate up to the big-DP and enable it to undergo A-bar movement in a manner analogous to [Focus] features that drive A-bar movement, as observed in English sentences such as It is [JOHN]’s book that I haven’t read (I read MARY’s book), as suggested by an anonymous reviewer? We contend that the answer to this question is negative. To support our claim, we argue that the big-DP headed by suo does not exhibit behavior that is comparable to the clause-internal (cf. clause-external) Focus/Topic in Chinese (using the terminology of Paul [2005]). It is well-known that in Chinese, movement of the object in the canonical SVO order lands between the subject and verb, leading to the SOV order (see Huang et al. 2009) and the preposed object acquires a contrastive focus or topic interpretation. This movement is clause-bound, as evidenced by the contrast between (47a) and (47b-c),^[42] which is presumably due to A-movement (see Badan and Del Gobbo 2015; Qu 1994; Shyu 2001, 2014; Ting 1995; cf. Paul 2002, 2005).

Given the fact that suo, associated with discourse functions such as focus, also occurs between the subject and V, the proposal that the big-DP, undergoing long-distance focus-driven (A-bar) movement plus Op movement, stranding suo, can’t be correct. This is because the big-DP headed by suo would be expected not to be capable of moving long-distance across finite clausal boundaries, presumably because it’s the so-called A-movement, contrary to fact (cf. Example (5b)–(5c)). Independently of the ultimate account for the clause-internal Focus/Topic construction, what is crucial for our purposes here is that, in reality, suo can indeed occur in the main clause, away from its base position inside the embedded finite clause (whether this base position is the subject or object); the observed distribution of suo diverges from what one would expect from a clause-internal Focus/Topic-related element in Chinese but follows from our PMC-based account.^[43] Building on the preceding discussion, we conclude that the long dependency of suo cannot be accounted for by the stranding under A-bar movement approach (cf. McCloskey 2000).

Another potential explanation to consider for suo’s long dependency is the long head movement approach, suggesting that “suo moves on its own via some specifier position.” As mentioned in passing, recent studies have demonstrated that verbal heads, such as Chinese doubled verbs (Cheng and Vicente 2013) and Bulgarian fronted participles (Harizanov 2019; Harizanov and Gribanova 2019) can undergo long movement comparable to phrasal movement. In this scenario of long head movement, suo would reach the matrix clause position via Spec,CP by A-bar movement, and this would have nothing to do with being exempt from the minimality requirement by the PMC. While considering this proposal for suo’s long dependency seems appealing, it ultimately faces challenges, as there’s another element in the structure, namely the relative Op complement of suo, which also requires the specifier of CP as an escape hatch to meet the minimality requirement. There is empirical evidence in Chinese indicating a limitation on (intermediate) multiple specifiers. As pointed out by Huang et al. (2009: 264), sentences like (48) only allow shenme ‘what’ to have a matrix scope reading and weishenme ‘why’ an embedded scope reading, but not vice versa.

A plausible explanation for the interpretation contrast in (48) is that when shenme ‘what’ (or the associated Op) occupies the specifier of the embedded CP, weishenme ‘why’ (or the associated Op) has no way to reach the matrix Spec,CP. If the embedded CP had multiple specifiers, then we would anticipate that weishenme ‘why’ could utilize an available specifier in the embedded CP, allowing such questions to be interpreted as direct questions about weishenme ‘why’ – contrary to observed facts. This finding indicates that Chinese does not behave similarly to multiple-wh-fronting languages such as Bulgarian and Romanian, which permit multiple wh-phrases in Spec,CP and do not exhibit island effects (see Rudin [1988: 456–459] for discussion). Therefore, we conclude that the long head movement approach to explaining the long dependency of suo may not be viable, as there is no feasible way for both suo and its relative Op complement to undergo A-bar movement via an intermediate Spec,CP. In essence, none of the alternative accounts considered above can adequately elucidate the long dependency of suo.

To end Section 4, let’s explore why the resumptive clitic in Italian categorically rejects long dependency (cf. (30)), thus presenting a sharp contrast with suo. As highlighted in (28), Italian, along with English, falls into the category of languages that utilize resumptive pronouns exclusively in the absence of legitimate A-bar movement. When illustrated in the structure (31), the position of NP should be replaced with a gap g and bound by an Op in the matrix CP, with both g and Op being base-generated. Since this dependency, [Op, …, g] is not established through movement, it is not subject to Subjacency or minimality (which, ultimately, underscores the rationale behind utilizing resumptives in these languages). The absence of activation of locality conditions translates to the absence of activation of the PMC. Consequently, the clitic (or its trace) does not receive exemption despite its occurrence inside an RC, a context that may seem superficially identical to that of suo.