While subjects with accusative Case in languages such as English are restricted to contexts where embedded subjects originate in defective or infinitival clauses (TPs), abundant cross-linguistic evidence shows that subjects can indeed receive accusative Case across a CP boundary (Kuno 1972; 1976; Şener 2008; Fong 2019; Wurmbrand 2019). The primary goal of this paper is to examine the restrictions on subjects with accusative Case through an in-depth study of Japanese and explore their theoretical implications.

Japanese has several types of CP complements, including those headed by to for reports (ReportP) (1a) and those headed by ka for questions (ForceP) (1b) (Saito 2010; 2012; Hiraiwa 2010a; Yoshida 2019; Tomioka 2020; Nasu et al. 2024):1

While the complement clause in (1a) is headed by the Report head to, the one in (1b) is headed by the Force head ka, which receives accusative Case.2 The subject Hanako in (1a) and (1b) receives nominative Case. Interestingly, subjects that receive accusative Case do not behave uniformly with the clausal complements introduced above; while both accusative subjects and those with the object marker no koto (henceforth accusative no koto subjects) are allowed with ReportPs, only accusative no koto subjects are allowed with ForcePs.

I argue that the asymmetry in the distribution of subjects with accusative Case arises from an interaction between phasal transfer (Bošković 2016a; Saito 2017a; b; 2020; 2024a; Takahashi 2023) and anti-locality (Bošković 1994; 2005; 2015; 2016a; b; Ishii 1999; Saito & Murasugi 1999; Abels 2003; Grohmann 2003; Takahashi 2023). Accusative subjects cannot be selected by either omow- ‘think’ or tazune ‘ask’, indicating that they must originate within complement CPs. They can move out of ReportP into matrix VPs because this movement does not violate the ban on improper movement owing to phasal transfer. However, the subjects cannot move out of ForceP because moving out of ForceP places them “too close” to the matrix VP, resulting in an anti-locality violation. By contrast, accusative no koto subjects can be selected by both omow- ‘think’ and tazune ‘ask’, showing that they can be base-generated within matrix VPs. They can appear with both ReportP and ForceP because their derivation does not violate any constraint on movement. I also suggest that when an accusative subject is licensed across a “complex” CP involving both the Report head to and the Force head ka, to is externally pair-merged with ka, rendering the latter “invisible” (for external pair-merge of heads, see Epstein et al. 2016; Saito 2020).

The remainder of the paper is organized as follows. In Section 2, I present core observations concerning the distribution of accusative (no koto) subjects. In Section 3, I present an analysis of the observations based on phasal transfer and anti-locality. In Section 4, I discuss cases where ForcePs apparently lack Case and show that such ForcePs pattern with Case-marked ForcePs, providing further support for the proposed analysis. In Section 5, I examine cases where a ForceP accompanied by the Report head to allows accusative subjects and argue that the Report head to in such cases is externally pair-merged with the Force head ka. Section 6 presents the conclusion.

This section focuses on the distribution of accusative (no koto) subjects with ReportP and ForceP. I show that while both accusative subjects and those with the object marker no koto (accusative no koto subjects) are allowed with ReportP, only accusative no koto subjects are allowed with ForceP. I also argue that the ban on accusative subjects with ForceP cannot be fully captured by the well-known constraints on multiple accusatives.

Before discussing accusative (no koto) subjects, some remarks on no koto phrases are in order. No koto ‘-GEN fact’ appears with objects and is classified into optional and obligatory types (Sasaguri 1999; Takubo 2007; Kishimoto 2018; 2021). First, when an object is selected by verbs such as nagur- ‘hit’, the presence of no koto on the object is optional:

Example (2) is acceptable regardless of the presence or absence of no koto. Second, when an object is selected by verbs such as hanas ‘talk’, the presence of no koto on the object is obligatory:

Importantly, no koto cannot appear with subjects:

The unacceptability of (4), where the subject Mary is accompanied by no koto, clearly shows that no koto appears with objects but not with subjects.

We now proceed to the discussion on accusative (no koto) subjects (Kuno 1972; 1976; 2007; Takemura 1975–1976; Saito 1983a; Kitagawa 1985; Kaneko 1988; Oka 1988; Ueda 1988; Hoji 1991; 2005; Sakai 1998; Hiraiwa 2001; 2005; Kobayashi & Maki 2002; Tanaka 2002; Takano 2003; Takeuchi 2010; Takahashi 2016; 2021a; b; Kishimoto 2018; 2021). As is well known, ReportP allows both accusative subjects and accusative no koto subjects (for accusative no koto subjects, see Saito 1983a; Kitagawa 1985; Kishimoto 2018; 2021; Takahashi 2021b).

Hanako in (5) receives accusative Case and can be accompanied by no koto. Interestingly, ForceP prohibits accusative subjects (Kitagawa 1985; Hiraiwa 2010a; Takeuchi 2010; Takahashi 2016) but (marginally) allows accusative no koto subjects:

Example (6a), which involves the accusative subject Hanako, is unacceptable. Example (6b), which involves the accusative no koto subject, is slightly degraded. Significantly, although the judgment can be delicate, (6b) is more acceptable than (6a). Importantly, both (6a) and (6b) contain two accusative phrases. The relative unacceptability of (6a) and (6b) thus seems to relate to the two well-known double-o constraints (henceforth DoCs), which prohibit multiple accusatives in different ways. Following Poser (2002), I call them the surface Double-o Constraint (henceforth the surface DoC) and the deep Double-o Constraint (henceforth the deep DoC). The surface DoC refers to cases where multiple accusative phrases yield relatively mild degradation (Harada 1973; 1975; Kuroda 1988; Poser 2002; Hiraiwa 2010b; Saito 2016; 2017b).

Both Hanako and atama ‘head’ receive accusative Case in (7), and the presence or absence of no koto on the first accusative phrase Hanako does not affect the acceptability of this example. Importantly, the violation of the surface DoC can be circumvented by various strategies, including clefting (Harada 1973; 1975; Kuroda 1988; Hiraiwa 2010b; Saito 2016; 2017b):

Hanako(-no koto) in (8a) and atama ‘head’ in (8b) are in the focus position of cleft construction. Examples (8a) and (8b) are more acceptable than (7). Notably, (8a) and (8b) are acceptable regardless of whether Hanako involves no koto.3 The deep DoC refers to cases where multiple accusative phrases lead to complete unacceptability (Kuroda 1965; 1978; Harada 1973; 1975; Williams 1981; Saito 1982; 2020; Poser 2002; Takano 2004; Ochi 2009; Miyagawa 2017). When the causative predicate -(s)ase ‘cause’ selects an intransitive verb such as hasir- ‘run’, the causee can receive either dative Case or accusative Case. The accusative causee can be accompanied by no koto.

When the causative predicate -(s)ase ‘cause’ selects a transitive verb such as yom- ‘read’, the causee cannot receive accusative Case, which the deep DoC captures (for a discussion on examples other than causatives, see Miyagawa 2017):

The causee Hanako(-no koto) in (12) cannot receive accusative Case. Notably, (12) is unacceptable with or without no koto on the accusative causee. Importantly, violations of the deep DoC cannot be circumvented by clefting (Harada 1975):

Hanako(-no koto) in (13a) and sono hon ‘the book’ in (13b) are in the focus position of cleft construction. Contrary to (8a) and (8b), (13a) and (13b) are still unacceptable.4

We now consider the distribution of accusative (no koto) subjects with ForceP. I show that while the surface DoC alone is responsible for the case of the accusative no koto subject (6b), the deep DoC cannot be responsible for either (6a) or (6b). Interestingly, while (6a) does not improve with clefting, (6b) becomes more acceptable with clefting:

The accusative subject in (14a) and the accusative no koto subject in (14b) are both clefted. Example (14b) is considerably more acceptable than (14a). The unacceptability of (14a) suggests that the surface DoC is at least insufficient to exclude (6a) (i.e., it is not clear if the surface DoC is operative here (cf. Hiraiwa 2010b)). Furthermore, the acceptability of (14b) indicates that only the surface DoC is operative in (6b). Importantly, (14b) demonstrates that two accusative phrases are, in principle, permitted in this construction. The difference between (6a)/(14a) and (6b)/(14b) is further motivated by cases of other strategies to avoid violations of the surface DoC. These violations can be avoided when the two occurrences of accusative Case markers in a certain domain are separated by strategies such as (i) dislocation of one of the two accusative phrases via scrambling (Hale & Kitagawa 1976–1977; Hiraiwa 2010b) or (ii) replacement of one of the accusative Case markers with particles (Shibatani 1978; Hiraiwa 2010b). The following examples illustrate the scrambling strategy:

Example (15a) is a case of surface DoC violation caused by the presence of the two accusative phrases Hanako-no koto and atama ‘head’. In (15b), Hanako-no koto is scrambled to the sentence-initial position. Example (15b) is more acceptable than (15a). The presence or absence of no koto on Hanako does not affect the acceptability of these examples. Returning to the case of the accusative (no koto) subject with ForceP, the scrambling strategy works only with the accusative no koto subject:

The accusative subject Hanako in (16a) and the accusative no koto subject Hanako-no koto in (16b) are placed after the adverb oogoede ‘loudly’. Significantly, while scrambling of Hanako to the sentence-initial position does not result in the acceptability of (16b), scrambling of Hanako-no koto in (17b) does, which corroborates the distinction under consideration. The replacement strategy is illustrated below:

Atama ‘head’ in (18b) is accompanied by the focus particle -mo ‘also’, which suppresses the accusative Case maker -o. Example (18a) is more acceptable than (18b). The presence or absence of no koto on Hanako does not affect the acceptability of these examples. Returning to the case of the accusative subject with ForceP, the replacement strategy works only when the accusative subject is accompanied by no koto:

The ForceP complement in (19b) and (20b) is followed by the focus particle -mo ‘also’, which suppresses the accusative Case maker -o. While (19a) involving Hanako is unacceptable, (19b) involving Hanako-no koto is acceptable, which corroborates the distinction under consideration.

The unacceptability of (6a)/(14a) suggests that (6a)/(14a) violates the deep DoC, which means that (14a) could be treated on par with (12)/(13). However, if the deep DoC were responsible for (6a)/(14a), it should incorrectly rule out (6b)/(14b) as well. Recall that the causative construction with two accusative phrases is completely unacceptable regardless of the presence of no koto on the causee argument (12)/(13). Thus, if an account based on the deep DoC were to exclude (6a)/(14a), it should similarly exclude (6b)/(14b). If the deep DoC is formulated to prohibit more than one (accusative) “objects” (cf. Harada 1975; Williams 1981; Poser 2002; Takano 2004), the contrast between (6a)/(14a) and (6b)/(14b) is mysterious. In particular, the (relative) acceptability of (6b)/(14b) does not seem to receive a principled account because the first accusative phrase in (6b)/(14b) is accompanied by no koto, which marks transitive objects (2)/(3). If the deep DoC is formulated to prohibit multiple accusative Case assignment (Saito 1982; 2020), the relative acceptability of (6b)/(14b) remains mysterious as (6b)/(14b) contains two accusative phrases. Therefore, I conclude that the deep DoC is not responsible for the ban on accusative subjects with ForceP.

Note that the ban on the accusative subject in (6a)/(14a) cannot be attributed to the lexical property of tazune ‘ask’; one cannot simply say that tazune ‘ask’ disallows accusative subjects. Not all verbs allow accusative subjects (Kuno 1972; 1976; cf. Kobayashi & Maki 2002):5

The matrix verb in (21) is donat- ‘scream’. The embedded subject Hanako can receive nominative Case, but not accusative Case. Thus, it is conceivable that tazune ‘ask’ simply forbids accusative subjects, which would exclude (6a)/(14a). However, this is not the case; tazune ‘ask’ allows accusative subjects when the Force head ka is followed by the Report head to (Kuno 1972; 1976; Kobayashi & Maki 2002; Takeuchi 2010):6

Example (22) involves both tazune ‘ask’ and the embedded ForceP. Interestingly, contrary to (6a)/(14a), (22) is acceptable. We therefore cannot account for the ban on accusative subjects with ForceP based on the lexical property of tazune ‘ask’ (see Section 5 for analysis of (22)). The main observations in this section are summarized below:

After discussing phasal transfer, anti-locality, and the KP hypothesis, in this section, I demonstrate how their interactions elucidate the observations summarized above.

This section discusses cases where the accusative subject is blocked in the (apparent) absence of accusative Case marking on ForcePs. I argue that in such cases, ForcePs do bear (inaudible) Case, projecting a KP. I also show that the apparently Case-less ForcePs behave differently from ReportPs, which genuinely do not bear Case.

Note that accusative Case marking on the ForceP complement is optional (Fukui 1986; Hiraiwa 2010a; Saito 2010; 2012; Yoshida 2019; Tomioka 2020; Nasu et al. 2024):16

Interestingly, the contrast between the accusative subject and accusative no koto remains even if the Force head ka does not morphologically realize accusative Case:

While the ForceP with the accusative no koto subject is acceptable, the ForceP with the accusative subject is not. The acceptability of (52) with the accusative no koto subject is not surprising. As I have claimed above, the distribution of the accusative no koto subject is constrained by the surface DoC (47). As two accusative Case markers are not present, the surface DoC is not violated in (52). Example (52) with the accusative no koto subject is thus expected to be acceptable. Further, (52) with the accusative subject is interesting. I have argued above that the ban on the accusative subject with the (overtly) Case-marked ForceP arises from an interaction between phasal transfer and anti-locality, which crucially relies on the presence of the KP projection dominating the ForceP complement (46). Given that (52) with the accusative subject is unacceptable, similar to (44a), I claim (i) that the ForceP in (52) is dominated by a KP and (ii) that the K head does not have to be phonetically realized when it is accusative.

The claim that ForcePs occurring as a complement of tazune ‘ask’ always bear accusative Case is supported by some observations concerning the distribution of ForcePs in the causative construction. As we saw in (12), when the causative predicate -(s)ase ‘cause’ selects a transitive verb, the embedded object cannot receive accusative Case in the presence of the accusative causee, which is attributed to the deep DoC (Kuroda 1965; 1978; Harada 1973; 1975; Williams 1982; Saito 1982; 2020; Poser 2002; Takano 2004; Ochi 2009; Miyagawa 2017):

Although a principled explanation of the deep DoC is beyond the scope of this paper, I adopt this constraint to assess whether an element bears accusative Case. Note that if the embedded object receives dative Case, it can occur with the accusative causee:

Example (54) shows that aw- ‘meet’ must take a dative object. While the presence of the embedded dative object and the dative causee results in slight awkwardness (55a) (cf. Ochi 2009), the embedded dative object can co-occur with the accusative causee without severe degradation (55b).

Returning to the clausal complements under consideration, note first that ForcePs that morphologically realize accusative Case cannot appear with the accusative causee:

The embedded ForceP in (56) can appear with the dative causee but not with the accusative causee. Importantly, the contrast in (56) persists even if the accusative Case marker is dropped from the ForceP:

The unacceptability of (57) with the accusative causee supports the claim that apparently Case-less ForcePs indeed bear accusative Case, projecting a KP. By contrast, ReportPs, which cannot bear Case, can appear with either the dative or accusative causee (Poser 2002):

The acceptability of (58) with the accusative causee aligns with the assumption that ReportPs do not bear Case. Given that multiple accusatives are prohibited in the causative construction, we predict that while (58) with the dative causee allows an accusative subject, (58) with the accusative causee does not. This prediction is borne out (Poser 2002):

To summarize, I have shown that the apparently Case-less ForcePs do, in fact, bear Case, which is supported by the distribution of ForcePs and ReportPs in the causative construction.17

In this section, I discuss cases where the Force head ka does not bear Case when it co-occurs with the Report head to. I argue that such cases allow accusative subjects when they involve external pair-merge of the complementizers.

The Force head ka and Report head to can co-occur within a single clause (Saito 2010; 2012):

Interestingly, as we have seen in (22), the “complex” ReportP/ForceP allows accusative subjects (Kuno 1972; 1976; Kobayashi & Maki 2002; Takeuchi 2010):

The accusative subject Hanako in (61) precedes the matrix adverb suguni ‘immediately’, which indicates that Hanako is in the matrix VP. As tazune ‘ask’ cannot select Hanako as the object (45), the accusative subject must originate from the ReportP/ForceP complement. Note that the Force head ka and Report head to in (60)/(61) do not bear Case, which suggests that the clausal complement in question does not project a KP:

However, recall that the phasal transfer hypothesis (24b) dictates that “a phase is transferred upon the completion of the next phase up” (Saito 2017a: 64). Given that CPs are phases (Chomsky 2000; 2013; 2015), we predict that the ForceP is transferred when the ReportP is completed. This means that the embedded subject cannot stay within the complement clause, otherwise the embedded subject in (61) would be transferred together with the ForceP before the former receives accusative Case from the matrix v/V:

The embedded subject KP is first merged with the ReportP complement before the ForceP is transferred. Here, I assume the following: given that ReportP can undergo scrambling just like KP arguments, the Report head to also has an anti-labeling property. Consequently, as both the embedded KP and ReportP complement have an anti-labeling property, the merger of the embedded KP subject and the ReportP complement yields an unlabeled syntactic object [??] (cf. (34)). This unlabeled status thus forces the KP subject to move further. However, the movement from [??] into the matrix VP should be ruled out by anti-locality (i.e., the movement does not cross a labeled projection distinct from the matrix VP) and the ban on improper movement in (43) (i.e., the movement from an A′ position (Spec, CP) to an A-position (Spec, VP) is impossible).

However, the problems are resolved once we assume that the complex ReportP/ForceP forms a single phase, which makes it possible to analyze (61) on par with (40). It has been proposed that when a head is externally pair-merged with a phase head, the latter becomes syntactically “invisible” (Epstein et al. 2016; Saito 2020). Takahashi (2023) argues that in the potential construction in Japanese, where an object can receive either accusative or nominative Case, two phase heads can be externally pair-merged (cf. Bošković 2015), showing that the Case alternation of the object reflects the optionality of the pair-merge operation (Morikate 2022). As is well known, a transitive object can receive nominative Case when a verb is accompanied by a potential suffix, which is a stative predicate (Kuno 1973; Saito 1982; Takezawa 1987):

Tuzuke ‘continue’ in (64) is accompanied by the potential suffix -rare ‘can’, and the object can receive either accusative or nominative Case. Following insights from past literature (Takezawa 1988; Ura 1999; Takahashi 2011; Moritake 2022), Takahashi (2023) assumes that the Case alternation in (64) reflects the optionality of external pair-merge of two phase heads: the potential suffix (v_can) and the embedded v:

In (65a), which corresponds to (64) with the accusative object, the embedded v assigns accusative Case and projects the vP phase. The potential suffix (v_can) selects the vP complement. Assuming the phasal transfer hypothesis (24), the vP phase is transferred upon the completion of the v_canP phase. In (65b), which corresponds to (64) with the nominative object, the potential suffix (v_can) is pair-merged with the embedded v (i.e., the v head is adjoined to the potential suffix (v_can)); consequently, the embedded v ceases to be a phase head that can assign accusative Case. As there is only one “visible” phase head (i.e., v-v_can), there is no transfer upon the completion of the v_canP phase. Thus, the object can receive nominative Case from Tense and move into the TP domain (Koizumi 1998; Nomura 2005; Takahashi 2023).

Against this backdrop, I argue that when the movement of the embedded subject takes place out of the complex ReportP/ForceP, the two phase heads (i.e., complementizers) are externally pair-merged. In particular, I propose that the Report head to is externally pair-merged with the Force head ka (i.e., the Force head ka is adjoined to the Report head to). This cancels the phasehood of the ForceP complement, and the movement of the embedded subject takes place, as shown in (40).

The ReportP/ForceP in (66) is not transferred until the completion of the matrix vP. The accusative subject thus moves directly from the embedded Spec, TP into the matrix VP, crossing the embedded TP and CP. The problems concerning improper movement and anti-locality that were pointed out above therefore do not arise.

The analysis is supported by certain observations regarding the distribution of wh-phrases. Note that when a head X is externally pair-merged with another head Y (i.e., Y is adjoined to X), Y becomes syntactically “invisible.” For instance, when a head X is externally pair-merged with a transitive v, the latter ceases to be a phase head that licenses its dependents (i.e., phrases that require Case) (Epstein et al. 2016; Saito 2020; Moritake 2022; Takahashi 2023). I suggest extending this idea to complementizers. Given that the Force head ka licenses wh-phrases, we expect that the Force head ka cannot license wh-phrases when it becomes “invisible” via external pair-merge.18 Specifically, when the embedded subject moves out of the complex ReportP/ForceP complement as in (66), which entails that the Report head to is externally pair-merged with the Force head ka, wh-phrases fail to be licensed by the Force head ka. This prediction is borne out (Takemura 1975–1976; Ueda 1988):

In (67a), the embedded nominative subject dare ‘who’ is associated with the Force head ka. In (67b), the embedded accusative subject dare ‘who’ fails to be associated with the Force head ka.19

Note that, as a reviewer pointed out, the contrast between (67a) and (67b) may simply show that the accusative subject is outside the scope (i.e., the c-command domain) of the Force head ka, rather than reflecting the invisibility of the Force head ka: (67b) may be unacceptable because the accusative subject is positioned above the Force head ka. Importantly, the contrast in question is not limited to the accusative wh-subject. This suggests that the Force head ka indeed becomes invisible with the accusative subject. Embedded non-accusative wh-phrases also fail to be licensed when an accusative subject is present (Ueda 1988):

In (68a), the embedded dative argument nani ‘what’ is associated with the Force head ka, whereas in (68b), the embedded dative argument nani ‘what’ fails to be associated with the Force head ka.20

Interestingly, the embedded wh-phrases can be licensed by a matrix complementizer regardless of the Case of the embedded subject:

Example (69a) involves the embedded nominative subject. The embedded dative object nani ‘what’ in (69a) takes either the embedded or matrix scope (Yoshida 2019). Example (69b) involves the embedded accusative subject. Nani ‘what’ in (69b) only takes the matrix scope. Example (69b) is unambiguous because the embedded Force head ka, with which the Report head to is pair-merged, cannot be associated with the embedded wh-phrase nani ‘what’. I suggest that (69a) is structurally ambiguous because pair-merge of complementizers is optional in this case. Note that as the embedded subject in (69a) receives nominative Case, the Report head to does not have to be pair-merged with the Force head ka, though nothing blocks the pair-merge operation in question. When the pair-merge operation occurs, the Force head ka fails to license the embedded wh-phrase. Therefore, the only complementizer that can be associated with the embedded wh-phrase is the matrix complementizer no, which results in the matrix scope interpretation. When the pair-merge operation does not occur, the closest complementizer ka licenses the embedded wh-phrase, resulting in the embedded scope interpretation.21^,22

Recall that the accusative no koto KP is base-generated in the matrix VP (41). The former can also appear with the complex ForceP/ReportP:

Note that (70) does not involve the accusative subject, which is associated with the external pair-merge of the two complementizers. In other words, the presence of the accusative no koto KP in (70) indicates that in contrast to (66), external pair-merge of the Report head to and Force head ka is not required here. Thus, in cases such as (70), embedded wh-phrases can take either the embedded or matrix scope. This prediction is borne out:

The embedded wh-phrase nani ‘what’ in (71) can take either the matrix or embedded scope, which results from the optionality of the pair-merge of complementizers. We also predict that when the complement CP contains only the Force head ka, the embedded wh-phrase must take the embedded scope even in the presence of the matrix accusative no koto KP. This prediction is borne out (I assume that the K head cannot be pair-merged with the Force head ka (see footnote 22)):

The embedded wh-phrase nani ‘what’ in (72) can only take embedded scope because there is no Report head to, which can be pair-merged with the force head ka. As the Force head ka is closer to the matrix complementizer no in (72), the former must license the embedded wh-phrase (Nishigauchi 1990; Watanabe 1992).

To summarize, I have argued that when accusative subjects are allowed with the complex ReportP/ForceP, the Report head to is pair-merged with the Force head ka. This claim is supported by the distribution of wh-phrases.

In this paper, I have argued that the distribution of accusative (no koto) subjects with ReportP and ForceP can be explained through the interaction of phasal transfer and anti-locality. Accusative subjects, which must originate in complement CPs, can move out of ReportP without violating the ban on improper movement. However, they cannot move out of ForceP due to the interaction of phasal transfer and anti-locality. Accusative no koto subjects, which are base-generated within matrix VPs, are allowed with both ReportP and ForceP because their derivation does not violate constraints on movement. I have also demonstrated that some seemingly Case-less ForcePs that disallow accusative subjects are selected by the K head. Finally, I have suggested that instances where accusative subjects are permitted with the complex ReportP/ForceP, which truly lacks Case, involve external pair-merge of the two complementizers. This analysis is corroborated by observations regarding the distribution of wh-phrases.