2.1 Previous approaches

There are mainly three types of explanations regarding the relationship between anaphor complexity and the availability of LD binding: the proposal put forth by Reinhart & Reuland (1991; 1993), the account assuming a movement approach to LD binding, and the account assuming the relativized Condition A approach to LD binding.

Beginning with Reinhart & Reuland (1991; 1993), they propose a substantially revised version of binding conditions. According to them, some anaphors are what they call reflexivizers, which reflexive-mark the predicates that select them. They posit that predicates, with the exception of a specific subset of intrinsically reflexive predicates, must be reflexive-marked by having a reflexivizer as an argument in order to be reflexive, which is their version of Condition B. Within their approach, Condition A can be characterized as a requirement dictating that a reflexive-marked predicate has to be reflexive. As long as they do not violate these conditions, anaphors can freely obtain an antecedent from discourse, resulting in what appears to be LD binding. To illustrate how this operates in cases pertinent to the current discussion, let us consider an example provided in (7).

Suppose the anaphor in (7) is a reflexivizer. In such cases, the anaphor reflexive-marks the predicate and, hence, Condition A forces it to be interpreted as bound by Bill, not allowing LD binding. If, on the other hand, the anaphor is not a reflexivizer, the anaphor does not reflexive-mark the predicate. Thus, Condition B prevents the predicate from being reflexive, disallowing the interpretation where the anaphor is bound by Bill. This, at the same time, implies that Condition A is not applicable and hence, the anaphor can freely get an antecedent from the discourse as long as it is not Bill. Now, importantly for the current discussion, Reinhart & Reuland claim that complex anaphors (which they refer to as SELF anaphors) are reflexivizers while simplex anaphors (which they call SE anaphors) are not. Crucially, according to their proposal, the difference between simplex and complex anaphors in terms of their status as reflexivizers stems from the semantics of the additional morpheme SELF that appears in the complex anaphors.

This line of analysis, however, is not extendable to the Japanese paradigm, at least in a straightforward way. First, as outlined in the core data in (2) above, zibun allows local binding as well as LD binding (see also Kishida 2011). According to Reinhart & Reuland’s analysis, such local binding of a simplex anaphor should be ruled out as Condition B violation, as zibun does not have the ability to reflexive-mark the predicate.⁵ Second, as mentioned above, Reinhart & Reuland attribute the distinction between simplex and complex anaphors to the presence or absence of the morpheme SELF, rather than the complexity of the structure. This point is pivotal in capturing the Japanese data: as noted in Section 1, in Japanese, zibun and zisin are LD anaphors respectively when used as simplex anaphors. However, when they are combined and form a complex anaphor, the resulting anaphor behaves as a local anaphor, suggesting that the difference between local and LD anaphors should be attributed to the complexity itself, rather than a specific morpheme. Thus the approach to complexity advocated by Reinhart & Reuland (1991; 1993), which attributes the reflexivizer function to a particular morpheme, is not straightforwardly applicable to the anaphor complexity in Japanese.

The other two accounts appear to be better aligned with the Japanese data in the sense that the distinction between simplex anaphors and complex anaphors is attributed to the complexity of the structure. What differentiates the two accounts lies in the approach to LD binding they assume.

The type of account based on what I would call the movement approach to LD binding has been proposed by studies such as Pica (1987), Cole et al. (1990), Cole & Sung (1994) and Katada (1991). The movement approach to LD binding claims that LD binding of an anaphor is achieved by covert movement of the anaphor into the same locality domain as its antecedent. Hence, according to this approach, the LD binding in (8a) is achieved by covert movement of the anaphor into the matrix clause as shown in (8b).

While the details vary among the studies, what these studies have in common is the idea that anaphor complexity matters for LD binding because it influences the possibility of movement. Pica (1987) and Cole & Sung (1994) claim that the movement involved in (8b) is head-movement. According to them, while simplex anaphors have X^o status and can undergo head-movement, complex anaphors are phrasal and cannot undergo head-movement. Hence, they stay locally and, as a result, complex anaphors cannot be bound by the matrix subject. A slightly different account is proposed by Katada (1991). Katada, who analyzes zibun and zibun-zisin in Japanese, assumes that zibun has a [+op] status and this motivates it to undergo phrasal movement into some VP-adjoined position. In her view, simplex zibun can undergo movement to a VP-adjoined position in a higher clause given that its trace is lexically governed by the predicate that selects it. However, when zibun within the complex anaphor zibun-zisin moves out, its trace is not lexically governed because the presence of zisin between the trace of zibun and the predicate serves as a barrier. Thus, zibun can only move to the local VP, where it can antecedent-govern its trace.⁶ Setting aside the differences in the details, the pivotal aspect shared by these studies is the view that movement of the anaphor enables the apparent LD binding and complexity affects the possibility of such movement.

Another type of account for the effect of anaphor complexity on LD binding assumes what I would refer to as the relativized Condition A approach to LD binding (Manzini & Wexler 1987; Progovac 1993). The relativized Coniditon A approach captures LD binding by relativizing some aspect of Condition A (Manzini & Wexler 1987; Progovac 1993). In particular, Progovac (1993), tries to account for the effect of anaphor complexity from this perspective. Specifically, she relativizes the potential antecedent and claims that LD anaphors take a head Agr as their antecedent, whereas local anaphors take a nominal phrase as their antecedent. (See Aikawa (1993) for the same line of analysis of LD binding of zibun.) Here, Agr is assumed to be coindexed with the subject it agrees with. She further assumes that some Agr can be anaphoric to the higher Agr and form a chain (see also Borer (1989)). Given that the higher Agr agrees with the subject in its specifier, this results in a structure, as schematized in (9). This means that an anaphor bound by Agr1 can be transitively bound by Agr2 and, as a consequence, can be coindexed with a higher subject.

This idea can capture the effect of anaphor complexity by further assuming that relativization of the antecedent is done based on the head/phrasal status of the anaphor: a simplex anaphor, being X^o, requires a head Agr as its local antecedent. On the other hand, a complex anaphor, being phrasal, requires a phrase NP as its local antecedent. As the LD binding is made possible via binding by Agr, this successfully derives the observation that only simplex anaphors undergo LD binding.

While the approach of Reinhart & Reuland (1991; 1993) does not seem to fit the Japanese paradigm, each of the other two accounts of the effect of anaphor complexity appear to work for the Japanese data, given the approach to LD binding that they assume. However, yet another approach to LD binding has recently been proposed and supported, which is what I review in the next section.

2.2 Null mediator approach

An approach to LD binding that recent literature supports is what I would call the null mediator approach (Charnavel 2020a; b; Charnavel & Bryant 2023; Baker & Ikawa 2024). Baker & Ikawa (2024), in particular, argue for this approach based on the data of Japanese. According to this approach, there exists a null pro within the same local domain as the anaphor. This null pro can be coindexed with an NP in a higher clause and bind the anaphor following Condition A. Therefore, in an example of the form in (10a), there is a null pro in a high position within the embedded clause as depicted in (10b). This pro can be coindexed with an argument from higher clauses, John, and bind the anaphor. Given that the pro and the anaphor are in the same clause, this binding relationship does not violate Condition A, similarly to the binding relationship between Bill and the anaphor. I assume that complement clauses can optionally contain this mediating pro (see also Nishigauchi 2014).

When juxtaposed with the approaches to LD binding adopted by the studies reviewed in the previous section, the null mediator approach presents certain advantages. Foremost among these is its ability to effectively capture the resemblance between LD anaphors and logophoric pronouns in African languages. It has long been observed that LD anaphors share similarities with logophoric prnouns in their selection of antecedents (Kuno 1987; Sells 1987; Oshima 2004; Charnavel 2020a; b; Baker & Ikawa 2024). For example, Baker & Ikawa (2024) compare the distribution of potential antecedents for LD anaphors in Japanese and logophoric pronouns in Ibibio and demonstrate that the choice of antecedent shows the same pattern for LD anaphors in complement clauses and logophoric pronouns: in addition to the matrix agentive subject, the matrix source phrases are potential antecedents (11a and 12a). On the other hand, the matrix goal arguments are not ((11b) and (12b)).⁷

Notably, for zibun, this characteristic distribution of the antecedent is only visible in LD binding, and not in local binding. The locally bound zibun (and zibun-zisin) are generally subject-oriented, as noted in Section 1 and, hence, cannot be bound by the source phrase as shown in (13) (see also Oshima 2004). See, for example, the contrast in (13). While zibun in an embedded clause can be LD bound by the matrix source Taroo in (13a), it is not possible for zibun to be bound by the local source argument in a matrix clause in (13b).

The null mediator approach is inherently crafted to capture this similarity between LD binding and logophoricity, and as such, provides a straightforward explanation for it. Logophoric pronouns have frequently been argued to have a left peripheral null pro (or an operator) as its antecedent (Koopman & Sportiche 1989; Adesola 2006). Such a view is motivated by the observation that logophoric pronouns in numerous languages are licensed solely under (a certain set of) complementizers (Koopman & Sportiche 1989; Clements 1975). For example, the Ibibio sentence in (14) from Baker & Ikawa (2024) shows that its logophoric pronoun imo is licensed in the complement clause but not in the matrix clause, as shown in (14). The idea that the logophoric pronouns needs to be bound by a null pro introduced by in clause periphery captures this restriction.

In such null pro approaches to logophors, the choice of the apparent antecedent for a logophoric pronoun in (11) is hence contingent upon the choice of the referent of this pro. Crucially, works such as Sundaresan (2012) and Nishigauchi (2014) employ such a null pro approach to LD anaphors to capture the parallel pattern in antecedent choice between the two constructions. The null mediator approach is an extension of this idea to capture the apparent violation of Condition A (Charnavel 2020a; b), with a slight change in the position of pro to an A-position (Nishigauchi 2014; Charnavel 2020a; b; Baker & Ikawa 2024). Thus, within the null mediator approach, one can posit that the referent of pro is determined by the same mechanism as the pro involved in logophors, thereby accounting for the analogous pattern in the choice of the apparent antecedent demonstrated in (11)–(12).⁸

On the other hand, the movement approach and relativized Condition A approach do not easily capture the similarity between LD anaphors and logophoric pronouns. Note first that logophoric pronouns are pronouns, as opposed to anaphors. For example, Baker & Ikawa (2024) give the example in (15), where a logophor imo cannot be locally A-bound. In such a context, an anaphoric form idem-imo needs to be used.

Now recall that the core component of the movement approach is the movement of anaphors to the same locality domain as their antecedents, where it can be bound by the antecedent following Condition A. A similar mechanism cannot be involved in the derivation of sentences with logophoric pronouns, if logophoric pronouns are pronouns and hence, do not follow Condition A. Thus, there is no place in the movement analysis for accommodating the similarity between LD anaphors and logophoric pronouns. For the same reason, the relativized Condition A approach (whatever form it takes) is not compatible with the logophoric paradigm either.⁹

Another problem for Progovac’s version of the relativized Condition A approach also comes from the example in (12). As (12a) shows, it has been pointed out that an LD antecedent for zibun does not have to be a subject (Oshima 2004; Nishigauchi 2014; Baker & Ikawa 2024). However, in Progovac (1993), the LD binding is achieved by binding by Agr which can be anaphoric to a higher Agr. As the index on Agr comes from its agreement with the subject, this anlaysis of LD binding predicts that only the subjects can be the apparent antecedents for LD binding. Given that the source phrase is not in a position to agree with Agr (or T in the current term), Progovac’s approach does not expect a non-subject to be an LD antecedent.

Overall, the null mediator approach seems to hold more promise than the other approaches in explaining the behavior of LD bound anaphors. However, simultaneously, the null mediator approach does not straightforwardly anticipate the differences among anaphors concerning LD binding. First of all, if there exists a local mediator of LD binding that can bind anaphors in accordance with Condition A, why do some anaphors disallow LD binding, failing to be bound by it? If there can be a local pro that is coindexed with a matrix argument and can bind the anaphor as depicted in (10b) above, then the straightforward prediction would be that every kind of anaphor should satisfy Condition A by being bound by it. One possibility that naturally arises here is that some anaphors are incompatible with binding by the null pro due to, for instance, their featural makeup. But such an approach is not available at least in Japanese, given the aforementioned observation that zibun and zisin respectively allow LD binding, while a complex anaphor that consists of them, zibun-zisin, does not. So, why is it that the complexity of anaphors that distinguishes anaphors that can be bound by the mediator from those that cannot in Japanese?

It is important to note that the explanations of the effect of complexity provided by the movement approach and relativized Condition A approach cannot be adjusted or expanded to align with the null mediator approach. Firstly, while analyses rooted in the movement approach argue that the complexity of an anaphor hinders its movement, thereby leading to the absence of LD binding, the null mediator approach cannot adopt this kind of idea, given that it does not posit the movement of anaphors. Similarly, the analysis presented by Progovac (1993) based on the relativized Condition A approach to LD binding is also not feasible. The critical factor in Progovac (1993) in terms of the analysis of the effect of complexity is the head status of the binder Agr in LD binding. Since the null pro in the null mediator approach is not a head, the concept of relativizing the antecedent based on the head/phrase status of the anaphor does not align with the null mediator approach. Consequently, a new explanation must be provided within the framework of the null mediator approach to elucidate why certain anaphors, particularly complex ones, fail to undergo LD binding and this is the objective pursued by this paper.

3.1 Internal structure of the anaphors

Based on the data like (16), I have so far discussed that the anaphor complexity impacts the availability of LD binding. But what kind of complexity is it that exactly matters here? In the Japanese paradigm, it can be straightforwardly observed that not just any kind of complexity matters. While I have referred to zibun as a “simplex” anaphor, this “simplex” anaphor can be further decomposed into zi and bun: previous studies such as Tsujimura & Aikawa (1999) and Kishida & Sato (2012) have discussed cases where the morpheme zi appears in verbs such as zi-satsu-suru ‘kill-oneself’ and zi-baku-suru ‘explode onself’. The morpheme bun is used by itself with the meaning “part”, as shown in (17) (Komatsu & Suzuki 2011).

Then what exactly is the notion of complexity that differentiates zibun from zibun-zisin? I claim that it is the syntactic complexity that matters here. This view is aligned to their behaviors with respect to the plural marker -tati. This plural marker can occur in the middle of the complex anaphor zibun-zisin, resulting in zibun-tati-zisin “oneselves” (Noguchi 2020). Under the well-accepted assumption that -tati syntactically composes with a nominal (Kurafuji 1999; Nakanishi & Tomioka 2004; Ueda & Haraguchi 2008; Tatsumi 2017), this observation suggests that zibun and zisin in zibun-zisin are syntactically composed with each other.¹⁰ Notice further that tati cannot occur in between zi and bun (^*zi-tati-bun). I consider zibun to be formed by the root merger between and , resulting in . The complexity derived from root merger does not affect the availability of LD binding while syntactic complexity does.

If it is the syntactic complexity within anaphors that plays a crucial role in terms of LD binding, the next question is as follows: what exactly are the syntactic structures internal to the Japanese anaphors? Again, the behavior of -tati is telling in this regard. Previous studies on this suffix have proposed a structure where this suffix merges with a nominal projection, not a head (Kurafuji 2004; Ueda & Haraguchi 2008, a.o.). For instance, consider the expression gakusei-tati “student-pl”, where the plural marker tati is attached to the noun student. The structure of gakusei-tati is generally considered to be (18), although the exact identity of the heads X and Y varies among analyses as well as the exact kind of interpretation it bears, which do not concern the current argumentation.

Now recall that the suffix -tati can attach to the first element zibun in the complex anaphor zibun-zisin, resulting in zibun-tati-zisin. This suggests that the first element zibun in the complex anaphor zibun-zisin has a phrasal status before it composes with the second element zisin. Therefore, I propose that the complex anaphor zibun-zisin and kare-zisin have structures in (19) and (20) respectively, whereas the simplex anaphor zibun or zisin has a structure in (21).

In the structure for the complex anaphor zibun-zisin in (19), the first element zibun by itself forms an nP. This nP further merges with and the entire complex is categorized again as nP by n^o. Following Marantz (2007), I consider the categorizing head n^o to be phasal. Given that there are two nPs in this structure, I refer to the inner nP as nP₁ and the outer nP as nP₂ just for conveninence. I also propose a parallel structure for kare-zisin, as shown in (20). In contrast, the simplex anaphors have one layer of nP as exhibited in (21). Recall that complex anaphor has a single accent as a whole (see Section 1). I consider such compound accent to be derived from the structures in (19) and (20) by the head movement of or to at PF. I assume that head-movement at PF does not affect binding relationship and, therefore, will not discuss this movement in the rest of this paper.

Building upon the perspective that Japanese generally lacks D projections (Fukui 1986; Hoji 1998, a.o.), I posit that nP instead of DP forms a full anaphor or a full pronoun in Japanese. This assumption is supported by the observation that Japanese anaphors can be modified by adjectives (or nominal adjectives) as illustrated in (22a) (also see Noguchi (1997) and Kim & Yoon (2008)). The same is true for the pronoun kare, as shown in (22b).

At this point, it is important to highlight the distinction between the structure in (21), which contains only one layer of nP, and the structures presented in (19) and (20), which feature two layers of nP. Considering that nP already represents a complete anaphor or pronoun, this implies that complex anaphors possess a structure wherein a simplex anaphor zibun or a pronoun like kare has an extra nP layer over it.¹¹ Note that, under the above-mentioned assumption that the categorizer head n^o is phasal, this extra layer introduces a phase boundary, which becomes crucial in the subsequent section. Given that nP₁ by itself already constitutes the simplex anaphor zibun, I consider that nP₁ and nP₂ in (19) both serve as anaphors and each of them needs to enter into binding relationship (c-command + coindexation) with its antecedent following Condition A.¹²

3.2 Binding Condition and the phase

Recall that the fundamental premise of the null mediator approach is that all the anaphors, including those appearing to undergo LD binding, adhere to Condition A. As my analysis is grounded in the null mediator approach, it is crucial to establish the definition of the binding domain before delving into the discussion of the avaiability of LD binding within this framework. This section is dedicated to accomplishing this objective.

While various notions of locality have been proposed as candidates for binding domains, such as the Specified Subject Condition and/or Tensed-S Condition, many recent works try to reduce the binding domain to the phase-based locality (Lee-Schoenfeld 2008; Quicoli 2008; Hicks 2009; Safir 2014; Despić 2015; Charnavel & Sportiche 2016, a.o.). In this context, Condition A can be defined as a requirement that an anaphor must have an antecedent accessible under the Phase Impenetrability Condition. Drawing from these studies, I will adopt a phase-based definition of the binding domain, especially due to its appeal in not necessitating stipulation of a locality domain defined specifically for binding. Incidentally, various studies have implemented this idea in different ways. For instance, Hicks (2009) claims that the binder and the anaphor form an Agree relationship, which is restricted by the phase-based locality. On the other hand, Safir (2014) claims that anaphor morphology remains underspecified until it gets bound: only if it gets bound locally, the anaphor acquires self morphology. For the purpose of this discussion, I will abstract from the specifics of these different implementations. What becomes pivotal in the rest of this paper is the concept that the binding domain is defined based on the PIC within Narrow Syntax.

The subsequent issue is the exact definition of the PIC. What I claim to be at work here is the version of the PIC in Chomsky (2001), which is defined as shown in (23). According to this definition, XP in (24), which is in the HP phase, becomes inaccessible once the next phase head Z is merged into the structure. Note that XP still remains accessible when the elements in YP is merged into the structure, even though YP is outside HP.

Adopting this definition of the PIC, the strict definition of Condition A now looks like (25).

Thus, if XP in (24) is an anaphor, it has to be bound before the merger of Z.

At this point, one may question the justification for adopting the version of the PIC in (23), especially given that previous studies that have proposed the phase-based approach to Condition A (Lee-Schoenfeld 2008; Hicks 2009; Safir 2014; Charnavel & Sportiche 2016; Saito 2017) assume implicitly or explicitly the stronger version of PIC as defined in (26) by Chomsky (2000), instead of the version in (23). The crucial difference between the definitions in (23) and (26) lies in the fact that, under the latter formulation, XP in (24) becomes inaccessible once the HP is complete, making it impossible for an element in YP to access XP.

In many of the previous studies on binding, it is not explicitly argued which version of the PIC is more preferred in capturing binding. Does the choice make a crucial difference in capturing the canonical local binding phenomena? When the anaphor is contained within a Voi(ce)P, which I assume to be the phase, this choice does not affect the range of potential antecedents. The version of the PIC in (26) confines the potential binders to the subject, which is base-generated within VoiP. The version of the PIC in (23) expands the domain to the projection just beneath the phasal C head, but this expansion does not add any new canonical arguments as potential binders (apart from the null pro postulated by the null mediator approach, which I discuss below). The choice becomes more substantial in other cases, especially when the anaphor occupies a subject position and, therefore, is positioned above the VoiP phase. In Section 3.4 below, I will argue that (23) in fact better captures the behaviors of anaphors in such cases.

To add one more comment on this choice, within the framework of the null mediator approach to LD binding, an LD anaphor must fall within the same domain as the mediating pro. Therefore, if we assume that the mediating pro occupies a position higher than TP as illustrated in (27b), Condition A formulated based on the stronger version of the PIC in (26) becomes incompatible with the idea that the anaphor in the object position can be bound by the mediating pro (at least unless one stipulates that the object covertly moves to the edge of VoiP): in (27b), the anaphor self resides within the VoiP phase, while the pro is outside of it.

One can circumvent this problem for the Condition A based on (26) by assuming that the mediating pro is in the edge of every phase, as opposed to just in a clause periphery (Charnavel 2020a; b). Under such assumption, the structure for (27a) would look like (28), where an additional pro appears inside VoiP.

However, recall that one prominent strength of the null mediator approach is its ability to capture the similarity between LD anaphors and logophoric pronouns in African languages. In terms of the licensing of logophoric pronouns, it has been pointed out that the involvement of CP is crucial: as shown in (29), only the embedded CP, but not the embedded DP, licenses the use of a logophoric pronoun within it. This observation aligns with the idea that only CP serves as the locus of the pro capable of binding logophoric pronouns.

Thus, the postulation of the mediating pro in every phase to make it compatible with the Condition A can make the mechanism of LD bound anaphors and logophors less alike and diminish the strength of the null mediator approach.

With these considerations as a background, I will proceed with the definition of Condition A based on the version of the PIC in (23). I will now enter into the analysis of the interaction between anaphor complexity and the (un)availability of LD binding.

3.3 Analysis

Now that I have examined the internal structures of the anaphors and set up theoretical assumptions, I return to the contrast in (30). The pattern is that the simplex anaphor zibun or zisin allows LD binding by John, while the complex anaphor zibun-zisin or kare-zisin cannot. I claim that this pattern straightforwardly follows from the internal structures and the definition of Condition A set up in the previous two sections.

I start from the analysis of zibun-zisin in contrast with simplex anaphors, and later deal with kare-zisin.

3.4 Further predictions

One essential component of the phase-based account so far was the presence of the VoiP phase boundary: the phasal Voi, combined with the nP₂ phase in the complex anaphor, prevents the first element in zibun-zisin, zibun_nP1, from being bound by the mediating pro. It is then expected that the behavior of the anaphors will vary when the Voi phase is absent between the anaphor and the mediating pro. This prediction can readily be examined by looking at the behavior of anaphors in the subject position, assuming that the subject is in Spec TP. More specifically, there are three predictions: one is for simplex anaphors, another is for zibun-zisin, and the third is for kare-zisin. Let me examine these predictions one by one.

First, when a simplex anaphor zibun or zisin is in Spec TP, it is predicted that these anaphors should allow binding by a matrix argument without mediation. The structure would look like (42).

This structure shows that zibun should now be able to find its antecedent until the matrix Voi head is merged following Condition A: given the version of the PIC from Chomsky (2001), zibun_simp can access elements higher than the first phase head above it, namely C, until the second phase head (the matrix Voi) is merged. As an implication, zibun should now be able to be directly bound by the matrix subject John even without the presence of pro. This is in contrast to the structure for zibun_simp in the embedded object position in Section 3.3 above, where binding by a matrix argument needed to be mediated by the mediating pro.

Does the availability of direct binding by a matrix argument make a difference from an empirical point of view? Interestingly, the literature has reported an observation that verifies this prediction.¹⁷ Abe (2014) reports that the subject zibun and the object zibun show slightly different interpretations, when they are coindexed with the matrix subject. He reports that when zibun is in the object position, binding by the matrix subject requires a de se reading, while a de se reading is not required when it is in the subject position. That is, in the indicated context, (43a) is acceptable while (43b) is not.¹⁸

As mentioned in Section 3.3.2, associating pro with some perspectival semantics is one way of accounting for the de se effect in the null mediator approach. This account predicts that the de se effect appears exactly when zibun is bound by the pro, while the de se effect would disappear when the binding does not have to involve pro. Now recall that when the simplex anaphor occurs in the subject position, it should be able to access the matrix argument directly in the absence of pro, while there is no such possibility when the simplex anaphor is in the object position. This predicts that the simplex anaphor in the subject position should be bindable by the matrix argument without any perspectival effect, whereas the object should not, and this prediction is precisely borne out by the findings in (43).¹⁹

Similar support comes from zibun in a subject position in a relative clause. In a relative clause, LD bound zibun is known to be sensitive not to de se perspective, but to empathy (Oshima 2004; 2007). That is, the referent of LD bound zibun in a relative clause has to be the one the speaker empathizes with. One way this constraint surfaces is the inability of zibun to cooccur with a first person pronoun as demonstrated in (44) (Kuno 1996; Oshima 2004; 2007). Following Charnavel (2020a; b), I posit that this empathy effect also arises from the perspectival semantics associated with the null pro.²⁰

Now, the current account expects that, when zibun appears in a subject position inside a relative clause modifying the object, such an empathy effect disappears: again, zibun can directly access the matrix subject without the mediation by pro given the structure in (45). This is borne out as shown in (46).

So far, I have demonstrated how a simplex anaphor behaves when it occurs in a subject position. What happens when the complex anaphor zibun-zisin occurs in the subject position? In fact, as shown in (47), the subject zibun-zisin can be bound by the matrix subject (see also Katada (1991: 289) for a similar judgement).

This is exactly what the current proposal predicts: the structure up until the matrix VoiP would look like (48).

Crucially, there is only one phase boundary nP₂ between the mediating pro and zibun_nP1. Thus, nothing prevents the first element zibun_nP1 from being bound by the mediating pro following Condition A. Hence, the current account correctly predicts that LD binding of zibun-zisin should become possible in the subject position.

One might suspect that zibun-zisin is simply directly bound by the matrix subjects rather than being bound by the mediating pro, which is indeed possible under the binding domain defined by the Specified Subject Condition (Chomsky 1973). Nonetheless, there is evidence suggesting that the binding here is mediated by the pro, as predicted by the current definition of the PIC. As discussed in Section 2, the antecedent for LD binding exhibits different distribution compared to the antecedent for local reflexive binding. More specifically, locally bound zibun or zibun-zisin cannot be bound by a non-subject, including the source phrase. The relevant example is repeated in (49). Zibun or zibun-zisin cannot be directly bound by the local non-subject phrase Taroo.

Now notice that, in (47), zibun-zisin=Taroo reading is also possible in addition to zibun-zisin=Hanako reading. This implies that the binding in (47) displays the characteristics of LD binding rather than local binding in terms of the choice of its antecedent. Note further that not every non-subject matrix argument can bind zibun-zisin in the subject position: the matrix goal argument cannot.

As discussed in Section 2 with (12b), this is also the pattern exhibited by LD binding of zibun in the object position. Since the choice of antecedent is determined by the referent of pro under the null mediator approach (refer to Section 2), the pattern of antecedent choice exhibited by zibun-zisin in a subject position can be explained by assuming that what binds zibun-zisin here is the mediating pro, similar to other instances of LD binding. This implies that zibun-zisin in the subject position can in fact be bound by the pro.

While I have shown so far that the complex anaphor can be bound by pro, using the possibility of LD binding by the non-subject, I have yet to confirm whether it is impossible at all for zibun-zisin to be bound by the matrix subject without mediation by pro. Unlike the simplex anaphors, zibun-zisin should not be able to directly access the matrix arguments: given the structure in (48) above, there are two phase boundaries, the nP₂ and the CP, between the first element zibun_nP1 and the matrix arguments. Since binding by pro enforces the perspectival effect, as we have seen, I can confirm this prediction using this effect. Starting with the cases with zibun-zisin in complement clauses, the prediction is that the non-de se reading in (51) is not possible. The judgment is subtle, but the prediction holds true. An informant who accepted the non-de se reading in (43a) judged the non-de se reading in (51) to be possible, but crucially reported that it requires contrastive reading unlike (43a). Recall that zibun-zisin can be either the intensified form of the simplex anaphor zibun or the true complex anaphor zibun-zisin, with the former resulting in the contrastive reading (see Section 1). The observation that the non-de se reading becomes possible only in the contrastive reading suggests that complex anaphor zibun-zisin, which is our focus here, does not allow non-de se reading, as predicted by the current analysis.

Similarly, zibun-zisin in a subject position of a relative clause is predicted to fail to cooccur with a first person pronoun, due to the obligatory perspectival (empathy-locus) interpretation. This is borne out in (52): the sentence with the first person pronoun is not allowed without a contrastive reading.

Thus, the current proposal correctly predicts the behavior of zibun-zisin in the subject position.

Finally, let me turn to the complex anaphor with a pronoun, kare-zisin. Recall that this anaphor must inherently adhere to the same locality requirement as the simplex anaphors: its first element is not an anaphor and therefore does not follow Condition A. The difference from the simplex anaphors is that the first element of kare-zisin, kare, avoids being bound by the mediating pro due to its anti-logophoricity. I explained earlier that the absence of LD binding for kare-zisin in the object position stems from this inability to be bound by pro.

Now, we have observed that zibun in a subject position can be directly bound by a matrix clause without the mediation by pro, following Condition A. This suggests that Condition A should similarly permit kare-zisin to be directly bound by the matrix arguments once it occupies the subject position. Importantly, the anti-logophoric property of kare does not prevent such a binding relationship, as mediating pro is not involved. The structure envisioned here is depicted in (53).

Here, the whole anaphor kare-zisin needs to find its antecedent prior to the merger of the second phase head, which is the matrix Voi. This means that a matrix argument should be able to be its antecedent without the mediation by pro. This prediction is confirmed by previous studies. Katada (1991) prodvides the example in (54).²¹

In this example, kare-zisin is bindable by the matrix arguments, John or Bill. Especially notable here is the possibility of binding by Bill. Recall that LD binding mediated by pro does not allow the matrix goal argument to be the surface antecedent (see Section 1, (12b)). This means that kare-zisin, when interpreted as referring to Bill, is directly bound by the matrix dative goal argument Bill. Given that kare-zisin, unlike other anaphors, does not show subject-orientation, the non-subject status does not prevent Bill from binding it.

Overall, the current proposal correctly predicts how the behavior of each of the anaphors changes when they are in the subject position. It is important to note that the detailed differences in antecedent choices among the three anaphors in the subject position cannot be captured merely by asserting that the embedded subject position falls within the same binding domain as the matrix arguments (cf. the Specified Subject Condition). Rather, the internal structure of the anaphors, the PIC-based binding domain, and the null mediator approach together derive the intricate differences among these anaphors.

Notice also that the discussion in this section provides additional evidence supporting the adoption of the version of the PIC repeated in (55).

Temporarily setting aside the possibility of binding by pro and the existence of complex anaphors, Condition A, as defined according to the version of the PIC in (55), and Condition A, as defined based on the stronger version in Chomsky (2000), yield different predictions regarding an anaphor in a subject position, as depicted in (56).

An anaphor can be bound by the matrix arguments in accordance with Condition A based on (55). On the other hand, the stronger version of the PIC predicts that the anaphor cannot be directly bound by the matrix arguments. As we have seen that zibun and kare-zisin can be directly bound by the matrix arguments (i.e., without the mediation by pro), at least the Japanese data clearly prefer Condition A defined based on the weaker version of the PIC.²² Thus, the current discussion motivates the adoption of the weaker version of the PIC in (55).