2.1 Unpassivizability

Below are examples of experiencer subject causatives.

At first glance, the subjects in these sentences appear similar to those in experiencer-subject psych-verb constructions and accidental agent constructions, as illustrated by (8) and (9), respectively.

However, they are not identical. Experiencer subject causatives differ from the other two constructions in terms of passivization. Experiencer subject causatives cannot undergo passivization (Amano 1987; 2002). When passivized, the experiencer interpretation disappears, leaving only the agent interpretation, as demonstrated by the passivized forms of (7) in (10). In these sentences, the ni ‘by’ phrases must be interpreted as an agent.

In contrast, experiencer subject psych-verb constructions and accidental agent constructions can be passivized without changing their basic meaning. (11) and (12) are the passive forms of (8) and (9), respectively.

According to Bruening (2013), passivization is a morphosyntactic operation that prevents the external argument from being realized in a usual way. Asami (2024) argues that this is also true of passives in Japanese (cf. Jo & Seo 2023). In light of this view, the inability of experiencer subject causatives to be passivized indicates that their subject is not an external argument, unlike those in experiencer-subject psych-verb and accidental agent constructions.

2.2 Possessor–possessum relation between subject and object

In experiencer subject causatives, a subject and object establish a possessor–possessum relation (Oehrle & Nishio 1981; Nishio 1982; Amano 1987; 2002; Pylkkänen 2000; Hasegawa 2001; 2004; 2007; 2016; Akimoto 2017; Wood & Marantz 2017; Yasuhara 2017). Consider the sentences in (13). These examples illustrate that the possessor–possessum relation can be either inalienable or alienable. In (13a), Taro inalienably possesses an arm; in (13b), Hanako alienably possesses a house.

In (13), if the possessor of the possessum is distinct from the referent of the subject, the experiencer reading of that subject becomes impossible. As a result, the sentences are construed as agentive (e.g., ‘Hanako broke someone else’s arm’ and ‘Hanako burned down someone else’s house’ in (13a) and (13b), respectively).

It is worth noting that the following sentences are unacceptable under the experiencer reading when uttered without context (I will discuss why the sentences are marked with ‘#’ to indicate a pragmatic oddity shortly). In these cases, the possessor position of the object NP is occupied by an overt element that is not co-referential with the subject. As a result, the possessor of the possessum is interpreted as distinct from the referent of the subject by default.

Importantly, as pointed out by Akimoto (2017), seemingly unacceptable cases like (14a) and (14b) become acceptable with supporting contexts. Specifically, the contexts in (15a) and (15b), inspired by Akimoto (2017: 3–4), make the sentences in (14a) and (14b) acceptable. These contexts facilitate the interpretation of the complex object as the possessum of the subject.

Therefore, the occupation of the possessor position within the object does not necessarily prohibit the experiencer reading of the subject, as long as an appropriate context is provided.

2.3 At-issue and not-at-issue meanings

I now turn to the types of meanings contributed by experiencer subject causatives, with special attention to the distinction between at-issue and not-at-issue meanings (Karttunen 1973; Karttunen & Peters 1979; Potts 2005). Unlike at-issue meanings, not-at-issue meanings project beyond entailment-canceling operators such as negation, questions, and conditionals. Based on this criterion, I will demonstrate that in experiencer subject causatives, the experiential interpretation contributes the not-at-issue meaning, while the experiencer subject contributes the at-issue meaning.

I will begin with the experiential reading. The experiencer subject causative is embedded under negation in (16). As indicated by (i), this sentence can mean that the event of Kenji’s arm breaking did not occur, while still conveying the interpretation that Kenji would have been affected if the event had happened. This interpretation shows that the experiential reading can survive under negation. By contrast, as demonstrated by the unacceptable reading in (ii), it is impossible to negate the experiential event without also negating the arm-breaking event. I will discuss (iii) shortly when addressing the experiencer subject.

Next, the sentence in (17) presents the experiencer subject causative embedded in a yes/no question. In this sentence, the speaker can ask the addressee whether or not Kenji broke his arm, implying that if he did, he would have been affected. The experience is implied but not questioned; therefore, the addressee cannot simply answer “no” if they know that Kenji’s arm did break, but it did not matter to him.

Finally, let us consider the experiencer subject causative embedded in a conditional in (18). In this case, the experiential interpretation does not alter the condition under which the speaker gives Kenji 10,000 yen. That is, regardless of whether Kenji is affected or not, the speaker will give him 10,000 yen if his arm breaks.

To sum up, the sentences in (16), (17), and (18) demonstrate that the experiential reading of experiencer subject causatives projects beyond negation, question, and conditional operators, respectively. Therefore, it is not-at-issue.

In contrast, the experiencer subject can be targeted by negation, question, and conditional operators. In the sentence in (16), the subject can be the target of negation, as indicated by the possible reading in (iii). In the sentence in (18), the experiencer subject contributes to the truth condition of the conditional. The speaker does not have to pay 10,000 yen to Kenji if someone else’s arm breaks. An example of the yes/no question is provided in (19) below.⁶ In this question, the speaker can ask the hearer whether or not it is Kenji who had his arm broken.

Based on these data, I conclude that the experiencer subject under discussion is at an at-issue tier.

2.4 Lexical idiosyncrasy

Prior studies have stated that experiencer subject causatives can be formed only if a verb has an inchoative counterpart (Harley 1995a; Pylkkänen 2000; Wood & Marantz 2017). I provide some relevant pairs below.

This generalization correctly accounts for the impossible experiencer reading in (21) below. The verbs used in these sentences (taberu ‘eat’ and sagasu ‘search’) are activity verbs and do not have inchoative counterparts in Japanese.

Crucially, the experiencer reading on the subject is not fully productive with alternating verbs, as noted by Oehrle & Nishio (1981: 167) and Miyagawa (1989: 129). A significant number of causative/inchoative verbs are not permitted in experiencer subject causatives, as shown below. The (a) sentences are intended to be experiencer subject causatives, while the (b) sentences are their inchoative counterparts.⁷

These data are far from exhaustive, as many more examples with other alternating verbs can easily be generated. These examples underscore the fact that having inchoative counterparts is a necessary but not sufficient condition for verbs to allow experiencer subject causatives.

The situations that the unacceptable (a) sentences above are intended to express are easy to imagine as experiential events. Yet, they sound unacceptable as experiencer subject causatives. Thus, pragmatic appropriateness alone does not determine the acceptability of experiencer subject causatives. At this point, I do not observe any predictable factor governing which alternating verbs can appear in experiencer subject causatives. Therefore, I consider this observation an instance of lexical idiosyncrasy.

2.5 Summary

To sum up, this section has reviewed the following four characteristics of experiencer subject causatives.

3.1 Theoretical assumption

The analysis I propose adopts the concept of the Affect head (Bosse et al. 2012; Jo & Seo 2023). To understand how Affect works, I will first go over key aspects of Jo and Seo’s (2023) analysis of indirect passives in Japanese (see also Asami (2024)). In indirect passives, such as (27), the subject is interpreted as an experiencer indirectly affected by the event described by the rest of the sentence (e.g., Taro’s praising of his daughter).⁸

To explain indirect passives, Jo & Seo (2023) adopt the functional head Affect ((28); Bosse et al. 2012: 1210). Affect takes an event property of type ⟨st⟩, introduces an experiencing event, and takes an individual to fill its experiencer argument. It also conventionally implicates the event denoted by its complement as the source of the experiencing event, as indicated by what follows the colon in the denotation.

Developing Bosse et al. (2012)’s high AffectP analysis, where AffectP occurs hierarchically above VoiceP, Jo & Seo (2023) propose that indirect passives, such as (27), have the syntax and semantics like (29).⁹ The combination of Voice and its complement is achieved by Event Identification (Kratzer 1996). All other semantic compositions are done via Function Application (Heim & Kratzer 1998). A functional head, Passive, syntactically selects an agentive VoiceP with an unsaturated external argument and existentially closes that external argument (Bruening 2013). If existential closure does not occur, Passive inherits the unsaturated external argument from the Voice head, allowing an oblique phrase to saturate it, as seen in the current example. The Affect head syntactically selects PassiveP and projects an experiencer argument in its specifier position.

3.2 Low AffectP analysis of experiencer subject causatives

I propose that the Affect head can syntactically select not only PassiveP but also CauseP. The functional head Cause introduces an implicit event that brings about the eventuality denoted by its complement (Pylkkänen 2008). If Affect combines with PassiveP, the indirect passive results, as argued by Jo & Seo (2023); if it combines with CauseP, the experiencer subject causative obtains. My proposed low AffectP analysis of experiencer subject causatives is illustrated in (30). In this analysis, AffectP is positioned below semantically contentless expletive VoiceP (Schäfer 2008; Alexiadou et al. 2015). The expletive Voice is projected to indicate the phase edge.

A reviewer raises the possibility that expletive Voice first combines with CauseP, followed by Affect; consequently, the hierarchical structure would be CauseP < expletive VoiceP < AffectP. While this alternative produces the same semantics as my proposal, since expletive Voice is semantically contentless, it faces some challenges. Specifically, it complicates the explanation for why experiencer subject causatives are possible only with causative verbs. My proposal captures this fact more effectively by framing it as a matter of selection: Affect selects CauseP, thereby preventing non-causative verbs, such as activity verbs, from allowing experiencer subject causatives. Additionally, if we assume that Affect combines with expletive VoiceP, the resulting hierarchical structure would be expletive VoiceP < AffectP, which mirrors the structure for indirect passives (PassiveP < AffectP), given that Passive is a type of Voice head. Since indirect passives can occur with any agentive verb, this alternative fails to explain why experiencer subject causatives are not as productive as indirect passives. For these reasons, I conclude that the correct hierarchical relation is CauseP < AffectP < expletive VoiceP.

My low AffectP analysis captures the first and third properties of experiencer subject causatives, repeated below:

First, it accounts for (31a) in conjunction with the view that Passive is an independent syntactic head, which selects an agentive VoiceP that has not yet projected an external argument in its specifier (Bruening 2013).¹⁰ Passive cannot combine with the VoiceP in experiencer subject causatives, since it is of expletive type. Consequently, experiencer subject causatives cannot be passivized.

It is worth addressing Yasuhara’s (2017) analysis here. According to his analysis, the experiencer subject in the construction under discussion is introduced in Spec of expletive VoiceP, as shown in the structure (32), which is based on Yasuhara (2017: 461, (13)). Yasuhara (2017: 462) suggests that the impossibility of passivized experiencer subject causatives stems from there being “no thematic argument to be absorbed by the passive morphology.” This conjecture itself aligns well with my current analysis. However, I will not adopt Yasuhara’s (2017) analysis because it does not clearly explain how the experiencer subject and the experiential interpretation contribute at-issue and not-at-issue meanings, which I will address next.

The not-at-issue status of the experiential interpretation and the at-issue status of the experiencer subject (31b) follow from the meanings contributed by Affect. According to the denotation in (28), the experiential meaning is conventionally implicated, while the experiencer is introduced as an at-issue meaning. The distinction in at-issueness between the experiential meaning and the experiencer role is independently supported by what Bosse et al. (2012) describe as affected experiencer constructions across various languages.

Crucially, the distinction between at-issue and not-at-issue meanings helps differentiate my analysis from Akimoto’s (2017) Appl(icative)P analysis. At first glance, my low AffectP analysis appears similar to the ApplP analysis, which is shown below:¹¹

Under this analysis, NP in Spec of ApplP is interpreted as the experiencer of a change-of-state event expressed by vP. Setting aside details, Akimoto (2017) argues that this interpretation is achieved through the abstract HAVE relation between the subject NP and vP, mediated by Appl. However, this analysis immediately encounters a problem regarding the not-at-issue experiential interpretation of experiencer subject causatives. The ApplP analysis suggests that the experiential reading is due to the abstract HAVE relation between the subject and the event, which would predict a parallel between experiencer subject causatives and possessive sentences. However, this prediction is not confirmed. The possessive meaning contributed by possessive sentences is at-issue, as it can be negated, which is at odds with the not-at-issue experiential interpretation of experiencer subject causatives (see Section 2.3).

We could salvage the ApplP analysis by stipulating that the abstract HAVE relation expressed by Appl is distinct from the canonical possessive meaning contributed by verbs like motu ‘have.’ However, such a stipulation would significantly weaken the analogy between the possessive relation and the experiencer interpretation. In contrast, my current analysis does not derive the experiential meaning from the HAVE relation and, therefore, does not encounter this problem. I conclude that the ApplP analysis is not a compelling account of experiencer subject causatives.

The next subsection turns to the fourth property, lexical idiosyncrasy.

3.3 Modeling lexical idiosyncrasy

The fourth property of subject experiencer causatives involves lexical idiosyncrasy, as reiterated in (35).

As discussed in the preceding section, non-causative verbs, such as activity verbs, do not project CauseP and are therefore incompatible with Affect. This explains why experiencer subject causatives can only be formed with causative verbs. However, it remains unclear why not all causative verbs permit this construction.

To account for the idiosyncrasy, I adopt the view that lexical entries for each verb can include syntactic structures (Bruening 2021) (see also Ramchand (2008; 2018)). According to this view, for instance, the lexical entry for or ‘break’ includes a structure like (36).

Causative verbs that cannot form experiencer subject causatives, such as hodok- ‘untie,’ simply do not have a structure like (36) in their lexical entries.

The current view contrasts with another perspective, which holds that lexical entries or roots contain no syntactically relevant information (Marantz 1997; Borer 2005). A comprehensive argument for one view over the other is beyond the scope of this paper. However, it is worth noting that the view I adopt is motivated by the prevalence of idiosyncrasies within the verbal domain (e.g., the flexibility of the spray/load alternation (Iwata 2008), the flexibility of dative alternation (Hovav & Levin 2008), the flexibility of the selectional properties of verbs (Ramchand 2008), and the licensing and interpretation of implicit arguments (Bruening 2021), among others). The idiosyncrasy of experiencer subject causatives adds to these cases. It is unclear how these idiosyncrasies can be adequately captured if all the syntactic information is severed from lexical entries or roots.¹²

A reviewer raises questions about the implications of the current assumption for individual differences. Specifically, they ask whether language learners must learn, based on experience without negative evidence, whether a particular causative verb can appear in an experiencer subject causative frame. The reviewer also suggests that, if this is the case, individual differences in the acceptability of specific experiencer subject causatives would arise. It seems that the reviewer is on the right track. The same reviewer finds the particular experiencer subject causative in (23a) acceptable, while the speakers I have consulted find the same sentence unacceptable. Harley (2008: 48, note 11) observes that some Japanese speakers do not accept certain experiencer subject causatives, crediting this observation to Yosuke Sato (personal communication). I also find that participants’ acceptability of experiencer subject causatives in my experiment, reported in Section 6, ranges from 25.0 to 100.0% (mean = 81.5%; SD = 38.9). Since no predictable factor determines which causative verbs can form experiencer subject causatives, these individual differences likely stem from variations in linguistic experience. Further research is required to explore individual differences in the acceptability of experiencer subject causatives (as well as other constructions mentioned above).

4.1 Overt possessor

Hasegawa (2001; 2007) bases the possessor raising analysis on the observation that experiencer subject causatives prohibit an overt realization of the possessor that is coreferential with the subject. This is illustrated by the following examples, which include either a reflexive (zibun-no ‘self’s’) or a pronoun (kare-no ‘his’ or kanozyo-no ‘her’). These examples are acceptable only under the agentive reading, not the experiential reading.

Hasegawa (2001; 2007) argue that the unacceptability arises from competition between the trace of the subject and the overt element in the same possessor position.¹³

However, I argue that the apparent impossibility of the experiential reading in (43) is due to pragmatics rather than syntax. To my ears, these sentences, when presented out of the blue, convey a contrastive meaning such that Hanako broke her own bone, but not someone else’s, and Taro burned down his own house, but not someone else’s. Based on this intuition, I propose that the prominent agentive reading in (43) is influenced by the Gricean Maxim of Manner (Grice 1975), which states that speakers should avoid ambiguous constructions by using unambiguous alternatives. Under this analysis, the speaker uses the overt possessor co-referential with the subject to avoid ambiguity caused by a potential alternative. This alternative would involve a possessor that is not co-referential with the subject, as in (44). These sentences allow only an agentive reading. Consequently, the sentences in (43), which are alternatives of (44), are also interpreted as agentive.

The same reasoning can be applied to cases with the overt pronominal possessor kare-no ‘his’ or kanozyo-no ‘her’ in (43).

It is well-established that the Gricean Maxim of Manner can be canceled or rendered irrelevant in certain contexts. This predicts that the overt realization of the possessor does not necessarily block the experiential reading if appropriate contexts are provided. This prediction is indeed borne out.

These examples allow the experiential reading even with the overt possessor. The possessor raising analysis incorrectly predicts that the overt possessor would compete with the trace or copy of the experiencer subject in the same possessor position within the object, thus prohibiting the experiential reading, contrary to fact. In contrast, under my pragmatic analysis, nothing inherently prevents the overt realization of the possessor, allowing the relevant sentences to be felicitous.

4.2 Scrambling

Hasegawa (2001) provides another piece of apparent evidence for the possessor raising analysis, based on scrambling. She observes that the experiential reading becomes impossible when the object is scrambled to a position in front of the subject.

Hasegawa argues that the unacceptability of the experiential reading in (47b) results from a violation of the proper binding condition: a moved element must c-command its trace (Fiengo 1977). Under the possessor raising analysis, the structure of the sentence in (47b) would resemble (48). In this structure, the trace t₁missing is not c-commanded by Subject₁ because the object containing that trace has been scrambled to the sentence-initial position. As a result, the proper binding condition is violated.

However, I contest this analysis. First, the validity of the proper binding condition has been challenged (e.g., Hiraiwa 2010). More importantly, the low acceptability does not necessarily stem from a purely grammatical factor. Various grammar-external factors, such as working memory demands, marked meaning, and discourse context, are known to influence the acceptability of a given sentence (Myers 2009: 412).

Although it is difficult to assess the relative influence of each factor on the observed acceptability, the most relevant factor in this case is discourse context. Kuno (1978) argues that when scrambling occurs, the scrambled accusative object tends to represent old information, while the nominative subject introduces new information, following the old–new information ordering. According to this discourse–pragmatic analysis of scrambling, a scrambled sentence presented out of the blue cannot maintain the proper old–new information flow. As a result, it is degraded, if not ungrammatical.

Given this background, let us consider (49). In this example, the appropriate context precedes the experiencer subject causative with the scrambled object, making the experiential reading acceptable. The accusative objects hizi-o ‘elbow’ and kata-o ‘shoulder’ in (49b) are explicitly referred to in the preceding context (49a), and the nominative subjects in (49b) represent new information. As a consequence, pragmatic felicity results.

If the unacceptability of sentence (47b) were due to a violation of some grammatical constraint, we would not expect to see an improvement in acceptability in (49). For instance, contextual support cannot improve the low acceptability caused by the violation of a grammatical rule, such as question formation in English (e.g., *Has the man who t₁ bought chicken has played in the park?). Therefore, I take the improved acceptability in (49b) as evidence that the low acceptability in (47b) does not arise from a purely grammatical factor.

It is worth noting that the experiencer subject causative with a scrambled object, when presented out of the blue, is unacceptable under the intended experiential reading but acceptable under the agentive reading.¹⁴ Consider (50). In these examples, the subject is interpreted as an agent.

This fact can also be explained under a discourse–pragmatic account. When these sentences are presented out of the blue, they are interpreted as follows. As mentioned above, a transitive sentence with a scrambled object tends to follow a given–new order (Kuno 1978). Under this consideration, for instance, one forces themselves to infer that the scrambled accusative phrase hone-o ‘bone’ in (50a) represents old information, while the nominative phrase Taro-ga new information. On the other hand, for the experiencer reading of Taro to be felicitous, Taro must be the possessor of the referent of the scrambled phrase hone-o ‘bone.’ This leads to a conflict: the old information hone ‘bone’ in the sentence-initial position implies that Taro is the possessor, but Taro is introduced as new information after the scrambled phrase. This conflict results in a processing cost. In contrast, the agentive reading does not cause such a conflict because hone ‘bone’ can refer to someone else’s bone, allowing Taro to be introduced as new information in the second position without any issue. I suggest that this is why the agentive reading is preferred over the experiential reading in scrambled experiencer subject causatives by default. The same reasoning applies to (50b).

Needless to say, the scrambled experiencer subject causatives are completely felicitous under the experiential reading if embedded under the appropriate context. Below is an example.

Also note that when a possessor co-referential with a nominative subject is overtly pronounced within the scrambled object, an agentive reading is preferred over an experiential reading (when the relevant sentences are used out of the blue).¹⁵ Consider (52). In these cases, there is a conflict of the sort mentioned earlier, regardless of whether one attempts to derive the experiential or agentive reading.

This preference is explained by the general assumption that, all else being equal, an unmarked meaning is preferred over a marked meaning. For instance, when one hears the word sentence, the most salient meaning would be a string of words because this is its most common, unmarked meaning, but it would be unlikely that a legal judgment or punishment imposed by a court comes first because it is its less frequent, marked meaning. The experiential reading of causatives is marked, as it is permitted only in a subset of causative verbs. In contrast, the agentive reading is unmarked. Consequently, the agentive reading is preferred over the experiential reading by default.

I conclude that the effect of scrambling does not constitute strong evidence for the possessor raising analysis.

4.3 Sloppy pronoun

I now offer evidence against the possessor raising analysis based on the sloppy identity of a pronominal object. Overt pronouns are known to allow sloppy identity in certain contexts (Karttunen 1969; Tomioka 2014; Sato 2020). The following is an example from Tomioka (2014: 254, (8)): In the second sentence of the example, the possessor of the overt pronoun sore ‘it’ is understood to be the subject of the same sentence, Kazuki, rather than the subject of the first sentence, Kazuki igai-no subete-no kodomo ‘every child except Kazuki.’ Hence, the picture that Kazuki tore up and threw away is the one that he drew.

Crucially, the sloppy reading of the overt pronoun also occurs in experiencer subject causatives. In the following examples, B’s sentences include an overt pronominal object with the object in the preceding sentence as its antecedent. What is of interest here is that the pronoun allows a sloppy interpretation, where the possessor of the entity expressed by that pronoun is the subject in B’s sentences but not in A’s sentences.

The sloppy readings in B’s sentences are problematic for the possessor raising analysis because the pronouns used here are incompatible with overt possessors (e.g., ??Taro-no soko ‘Taro’s there’ and ??Hanako-no sore ‘Hanako’s it’). If possessor raising were a necessary condition for the formation of experiencer subject causatives, B’s sentences in question would be unacceptable under the intended sloppy reading, contrary to fact. Under my pragmatic analysis, however, the sloppy identity of the pronouns can be explained in the same way as previously observed examples, such as (53). A detailed account of the sloppy identity of the overt pronoun must be left for future research.

4.4 Against a possible binding analysis

Let me now address another possible syntactic account, namely, a binding analysis, as schematized below. In this analysis, a subject binds a phonetically null pronoun or anaphor in the possessor position of the object.

Although I have not found any concrete proposals using the binding analysis in the literature, it is worth addressing this alternative because it does not encounter the same challenges as the possessor raising analysis. I will demonstrate that the binding analysis has both theoretical and empirical issues.

A theoretical problem concerns the requirement for binding. It is unclear what necessitates that the subject bind the silent anaphor or pronoun within the object for the well-formedness of experiencer subject causatives.

An empirical issue concerns the difference between experiencer subject causatives and their inchoative versions regarding the meaning of temporary possession.¹⁶ Consider (57). Given the context in (57), (57a) is felicitous as an experiencer subject causative, while is its inchoative version in (57b) is not. In (57a), the subject appears as the overt possessor of the surface subject. What is notable is that the relationship between Taro and the cap differs in these two sentences. The context indicates that Taro temporarily holds his friend’s cap, and this temporary possession is inferable in (57a). In contrast, the same temporary relation is difficult to infer in (57b). This sentence is acceptable only if the possession relation is not temporary. It is this unavailability of the temporary possession meaning that leads to the infelicity in (57b).

This subtle but significant difference poses a challenge to the binding analysis. The reason is that it predicts no interpretative difference between (57a) and (57b) in terms of possession meaning, since experiencer subject causatives and their inchoative counterparts share the same syntax concerning possession. Specifically, the possession meaning is expected to arise from the complex NP consisting of the possessor NP and its host NP.

A reviewer points out that the temporary possession reading becomes available if we paraphrase (57) into (58), where the possession meaning is expressed by the predicate motu ‘hold’ within the relative clause.

The contrast between (57b) and (58) can be attributed to the difference between attributive possession and predicative possession (Heine 1997). In attributive possession, possession is expressed within a complex NP consisting of a genitive phrase and its host NP. In predicative possession, it is expressed by a predicate, such as motu ‘hold’ in a clause. Akimoto (2017) independently observes that the temporary possession relation, analogous to my example, can be expressed by predicative possession but not by attributive possession. This observed contrast strongly suggests that the possessor–possessum relation in experiencer subject causatives does not arise from the attributive possession assumed in the binding analysis (or the possessor raising analysis).¹⁷

4.5 Summary

In summary, only my pragmatic analysis is compatible with all the empirical data regarding the possessor–possessum relation in experiencer subject causatives (i.e., the overt possessor, scrambling, sloppy pronoun, and temporary possession meaning).

5.1 Interpretation of -de phrases

I list some nouns that can be used as causer adjuncts below, categorized into two broad types: eventualities and natural forces.¹⁸

Nouns denoting eventualities, as in (61a), are compatible with Vendler’s (1967) narrow container verbs, such as occur (okoru in Japanese).¹⁹

The following example shows that the natural force nouns in (61b) are also compatible with okoru ‘occur’.

These facts suggest that -de ‘due to’ takes an eventuality-denoting noun.

Two reviewers note that okoru ‘occur’ does not sound entirely natural with nouns such as ooyuki ‘heavy snow,’ ooame ‘heavy rain,’ kaze ‘wind,’ or ame ‘rain.’ Some Japanese speakers I consulted share this judgment. However, they also mention that the acceptability improves when these collocations appear in academic contexts, such as science textbooks or lectures. This intuition is empirically supported by the following acceptable example.

I find this example completely acceptable, as do the other Japanese speakers I have consulted. I speculate that this is because sanseeu ‘acid rain’ is a technical term, making it easier to envision the academic context in which the collocation sanseeu-ga okoru ‘acid rain occurs’ is used.

Relatedly, one of the same two reviewers notes a difference between ooyuki ‘heavy snow’ and ooame ‘heavy rain,’ on the one hand, and kaze ‘wind,’ yuki ‘snow,’ and ame ‘rain,’ on the other. Specifically, the reviewer finds that the former two are more compatible with okoru ‘occur’ than the latter three. A similar contrast is also observed between kyoohuu ‘strong wind’ and kaze ‘wind.’ Two reviewers find the former acceptable with okoru, but not the latter. The only apparent semantic difference in these pairs seems to be the intensity of the eventuality denoted by them (e.g., kyoohuu ‘strong wind’ is stronger than kaze ‘wind’). Thus, the acceptability difference cannot be attributed to kaze ‘wind,’ yuki ‘snow,’ and ame ‘rain’ being non-eventuality-denoting, because it would be unreasonable to assume that only their stronger counterparts denote eventualities. Thus, I suggest that the acceptability difference arises from extra-linguistic factors and does not undermine my conclusion that all the nouns under discussion denote eventualities.

Note that -de ‘due to’ cannot take an animate noun denoting an agent, either in experiencer subject causatives or inchoatives (I will come back to these cases later).

The unacceptability of these sentences follows from the standard assumption that animate nouns are individual-denoting. Note that these sentences may sound very marginally acceptable only if the animate nouns are coerced to express a property or event related to the referent of the animate noun (e.g., Hanako’s fault causes Taro’s arm to break in (66a)).

5.2 Instrument-like -de phrases

The -de phrases present one complexity: the homophonous postposition -de can also head an instrument adjunct in agentive sentences.

What is of interest here is that these phrases also seem to appear in experiencer subject causatives and inchoatives, as shown in (68) and (69), respectively. (In these examples, I gloss -de as ‘due to’ rather than ‘with,’ because I will argue shortly that the NPs it introduces are eventuality-denoting.)

The nouns introduced by -de in (68) and (69) fail to pass Vendler’s (1967) narrow container test.

Additionally, they are incompatible with Vendler’s (1967) loose container verbs, such as expect (yokisuru in Japanese), which are known to take fact-denoting (and eventuality-denoting) nouns.²⁰

The results of the narrow and loose container tests indicate that the nouns introduced by -de in (68) and (69) align with neither eventuality-denoting nor fact-denoting nouns, similar to the canonical instruments in (67), which also fail to pass either of these tests.²¹

The incompatibility of the -de phrases in (68) and (69) with narrow/loose containers points to the possibility that they are individual-denoting, similar to canonical instruments in (67). However, I argue that this is not the case. First, the presence of instruments typically entails the presence of agents. Yet, neither experiencer subject causatives nor inchoatives permit agents in any form. This is demonstrated by the fact that they do not allow overt agent-denoting adjuncts or rational clauses, even when instrument-like adjuncts are present, as shown in (74) and (75), respectively.

Furthermore, unlike instruments, the referents of the instrument-like phrases under discussion do not need to be actively used in the events expressed by experiencer subject causatives or inchoatives. Consider (76). In (76a) and (76b), the branch falls without Taro’s intervention.

This suggests that the -de phrases in examples like (76) are not instruments in the strict sense. Note that they fail to pass the narrow or loose container tests, as shown below.

We now find ourselves in a contradictory situation: the instrument-like phrases in experiencer subject causatives and inchoatives seem to behave differently from both eventuality-denoting nouns and canonical instruments. A set of suggestive data to resolve this issue comes from paraphrasability. The sentences in (79) and (80) are paraphrased versions of (68) and (69), respectively. In these paraphrased sentences, the instrument-like phrases in the original sentences are replaced with clauses introduced by either okage-de or sei-de. Okage-de and sei-de introduce an adjunct clause expressing the cause of the eventuality described by the main clause. The former typically introduces a beneficial event, while the latter introduces an adverse event. Since this distinction is not relevant here, I gloss both as ‘because.’

The canonical instruments in the agentive sentences in (67) cannot be paraphrased in the same way.

The paraphrasability of the instrument-like phrases suggests that they are eventuality-denoting, similar to those in (61), rather than individual-denoting so that they can be paraphrased into a clause describing a causing event.²² One question that arises here is why they are incompatible with narrow or loose container tests. I propose that while these tests can demonstrate that a noun is eventuality-denoting, they cannot definitively show that a noun is never eventuality-denoting. This is supported by the observation that animate nouns, which are canonically considered individual-denoting, can behave like eventuality-denoting nouns in certain cases. We observe this pattern in the examples in (65) and (66), repeated here as (82) and (83), respectively. These sentences seem very marginally acceptable if the animate nouns are interpreted as events or properties associated with their referents.

Another piece of evidence supporting the current view comes from experiencer object psych verbs (e.g., Martin et al. 2023b). Consider (84). This sentence is ambiguous between an agentive reading and a non-agentive reading. In the agentive reading, Hanako intentionally bothers Taro (e.g., by making a lot of noise). In the non-agentive reading, something about Hanako, such as her bad behavior, bothers Taro, and Hanako does not need to act intentionally.

The second reading suggests that Hanako can be eventuality-denoting in this sentence, despite the fact that it fails to pass the narrow or loose container tests.

The subject in (84) under the non-agentive reading must be eventuality-denoting, as it can be paraphrased into a clause (Cheung & Larson 2015).

The point is that a canonical individual-denoting noun like Hanako can behave like an eventuality-denoting noun under certain circumstances. I take this to suggest that while the narrow/loose container tests can demonstrate that a particular noun can be eventuality- or fact-denoting, they cannot definitively show that it cannot be.

I conclude that the instrument-like -de phrases in experiencer subject causatives and inchoatives are eventuality-denoting, rather than individual-denoting, unlike canonical instruments.

5.3 Analysis

Let us now turn to a formal analysis of causer -de phrases. I adopt Martin et al. (2023a; b)’s view that causers are divided into two semantic types, type s or type e, depending on what they denote. Following Martin et al. (2023a; b), I assume that eventuality- or fact-denoting NPs are of type s, and I use the variable i, which ranges over the domain of situations. This domain encompasses the union of eventualities (v), facts (f), and other states of affairs. In contrast, individual-denoting nouns are of type e. Based on the evidence discussed so far, nouns introduced by -de ‘due to’ in experiencer subject causatives and inchoatives are all eventuality-denoting, and thus are of type s.

I propose that the causer-introducing postposition -de ‘due to’ has the denotation in (88). It first takes an argument of type s as its argument. Assuming that eventuality-denoting nouns like ziko ‘accident’ are of type s, the resulting -de phrase is defined as something like (89). It then takes an event property of type ⟨st⟩ as a second argument, identifying the causing event denoted by its sister. (The event variable itself is still open so it must be closed off by Aspect or Tense later (von Stechow & Beck 2015).)

In addition, I assume with Alexiadou et al. (2015) that the causer adjuncts are syntactically affiliated to the projection that introduces a causing event. I hypothesize that the relevant projections here are CauseP and BecomeP for experiencer subject causatives and inchoatives, respectively. This assumption follows from the view that adjuncts syntactically select the category that they modify (e,g., Pollard & Sag 1994).

Under the current analysis, the experiencer subject causatives with the causer adjunct with -de ‘due to’ have the syntax and semantics in (90).

Below are the syntax and semantics of the inchoative with the causer adjunct. Following Alexiadou et al. (2015), I assume that the inchoative construction has the same event complexity as its causative counterpart, in the sense that both include a causing event in their semantics. However, departing from Alexiadou et al. (2015), I adopt the view that the inchoative construction projects BecomeP rather than CauseP or its equivalent (cf. Harley 1995b; 2008). This syntactic distinction arises from the fact that the causative morpheme and the inchoative morpheme differ in their Case-assigning properties in Japanese. The former can assign an accusative Case, whereas the latter cannot (e.g., Hasegawa 2001).

5.4 Summary

To summarize, I first argued that the inanimate causers in experiencer subject causatives and inchoatives are eventuality-denoting. I then provided a formal analysis of the causer-introducing postposition -de. According to this analysis, -de takes an eventuality-denoting NP as its argument, and the resulting -de phrase adjoins to a projection that introduces a causing event (i.e., CauseP or BecomeP). The consequences of this syntactic analysis will be explored in the next section.

6.1 Two competing analyses of the syntactic position of the inanimate causer

The analysis proposed in Section 5.3 claims that a causer adjoins to CauseP which is above VP. This analysis predicts that the causer precedes the theme. I refer to this as the high causer analysis.

In contrast, a series of studies by Hasegawa (Hasegawa 2001; 2004; 2007; 2016) argues that the causer adjoins to V′ below the theme which occupies Spec of VP.²³ Consequently, the theme precedes the causer. I refer to this as the low causer analysis.

Hasegawa bases this structure on her syntactic analysis of experiencer-object constructions in (94), where the causer subject first merges with V′ and moves up to Spec of TP via the Spec of vP, while the experiencer object originates and remains in Spec of VP.

Based on this analysis, Hasegawa argues that the causer adjunct also occurs below the object in experiencer subject causatives.

According to Hasegawa, the structure in (94) is motivated by an inverse c-command relationship between the causer and the experiencer in experiencer-object constructions. First, an anaphor within the causer subject can be bound by the experiencer object, as shown in (95a) (e.g., Akatsuka 1976; Belletti & Rizzi 1988; Pesetsky 1995). Taking Binding Principle A as a ‘somewhere’ condition (Belletti & Rizzi 1988), the acceptability of (95a) indicates that the experiencer c-commands the causer at some point in the derivation. The agentive subject blocks this backward binding, as shown in (95b).²⁴

Second, a scope-bearing element in the object position can take scope over another in the causer subject position, as shown in (96a) (see also Homma (2004)). Japanese is well-known as a scope-rigid language, where the surface word order reflects the scope relation unless scrambling or movement occurs (Kuroda 1970; Kuno 1973). Therefore, the possible inverse scope relation in (96a) suggests that the experiencer object c-commands the causer subject at some point in the derivation. Notably, the agentive subject prevents the inverse scope relation, as shown in (96b).

Note that I consider densha ‘train’ in the object position as an experiencer, following Folli and Harley’s (2008) conception of teleological capability: an inherent ability of an entity to participate in a certain event. The train is capable of moving, and its movement can be delayed by some external cause, just like other movable animate entities. Therefore, I treat the sentence in (96a) as an instance of the experiencer-object construction.

These two observations lead Hasegawa to claim that the causer subject originates below the direct object in experiencer-object constructions, as shown in (94). The experiencer underlyingly c-commands the causer, accounting for the inverse c-command relation. Note that this type of analysis was first developed by Belletti & Rizzi (1988) and later extended by Cheung & Larson (2015). Based on her analysis of the object experiencer construction, Hasegawa argues that the causer adjunct in experiencer subject causatives also originates below the direct object, as illustrated in (93) above.

However, it is important to point out that the observed inverse c-command relation does not necessarily imply that the inanimate causer subject must originate below the experiencer object. Alternatively, one could propose that the experiencer originally occurs below the causer but moves over the causer later. Several authors have formalized this possibility. Specifically, Fujita (1996), Sato & Kishida (2009), and Alexiadou & Anagnostopoulou (2020) argue that the experiencer originates below the causer but moves up to Spec of AgrOP, Spec of Point of View Phrase, and somewhere within TP, respectively. Furthermore, one could postulate a direct extension of the Affect head to experiencer-object constructions, as shown in (97). In this structure, the causer appears within CauseP, while the experiencer originates in Spec of VP and moves up to Spec of AffectP, either covertly or overtly. After this movement, the experiencer c-commands the copy/trace of the causer, resulting in the observed inverse c-command relation.

Since this is beyond the scope of the paper, I will not provide a detailed discussion on the validity of each alternative analysis here. It suffices to say that in all four of these alternative analyses, the inverse c-command relation in experiencer-object constructions can still be captured, even if the causer subject originates above the experiencer object. All of these analyses are consistent with my high causer analysis of experiencer subject causatives. Therefore, the inverse c-command relation does not provide decisive evidence for the low causer analysis.

To sum up, the high causer analysis and the low causer analysis of experiencer subject causatives posit different hierarchical relations between the causer and the theme. The low causer analysis is motivated by the inverse c-command relation in experiencer-object constructions. However, the high causer analysis can also account for this relation. Therefore, it is difficult, if not impossible, to determine which analysis is preferable purely on theoretical grounds. For this reason, I turn to a psycholinguistic approach.

6.2 Predictions

The two competing analyses predict different basic word orders in experiencer subject causatives. In the high causer analysis, the causer occurs hierarchically above the theme. Assuming the general linearization algorithm, they are linearized as causer–theme. The low causer analysis, in contrast, posits the reverse hierarchical order; as a result, they are linearized as theme–causer. Given that Japanese allows scrambling, the high causer analysis predicts that the theme–causer order is derived by scrambling the theme over the causer. In contrast, the low causer analysis would claim that the causer–theme order is derived by scrambling the causer over the theme. The basic and scrambled word orders predicted by each analysis are summarized below:

Crucially, the distinction between the two analyses regarding the basic and scrambled word orders can serve as psycholinguistic hypotheses under the linking hypothesis between structural complexity and cognitive load, as stated in (100).

The higher cognitive load associated with complex structures can be reflected in indices such as reaction times (RTs), eye movements, and changes in physiological measures such as ERPs. In what follows, I focus on RTs, because my experiment uses them as the primary dependent measure.

The linking hypothesis (100) effectively captures the so-called scrambling effect in sentence processing: scrambling increases RTs (see Koizumi (2015) for a review). Some languages allow flexible word order, and Japanese is one of them. For instance, a simple transitive sentence may allow either SOV or OSV word order.

It is generally assumed that the SOV order (101a) is canonical, while the OSV order (101b) is derived by scrambling (e.g., Nemoto 1999). Specifically, in the OSV sentence, the object undergoes scrambling from its original position to the sentence-initial position, as shown in (102b), while no such movement occurs in the SOV sentence, as shown in (102a).²⁵

Notice that the OSV word order exhibits a more complex representation (i.e., requiring a gap–filling parsing) than the SOV word order. Under the linking hypothesis (100), we should then observe an additional processing load for the OSV sentences relative to their SOV counterparts. This is indeed what previous studies find (Chujo 1983; Tamaoka et al. 2005). Additionally, other studies show that derived word order sentences lead to longer RTs than their canonical counterparts across different languages and different constructions (see Koizumi 2015 for a review).

Note that, as mentioned in Section 4.2, scrambled OSV sentences tend to occur when they follow given–new information flow (Kuno 1978). Crucially, Koizumi & Imamura (2017) find that pragmatically licensed scrambling (i.e., when the scrambled object represents given information and the nominative subject new information) alleviates the scrambling effect (see Kaiser & Trueswell 2004 for the original findings in Finnish). However, scrambled sentences still take longer to process than their canonical counterparts in Japanese. Therefore, I set aside the contextual effect and assume that the scrambling effect occurs regardless of context, as far as RTs are concerned (cf. Yano & Koizumi 2018).

Given this background, let us now return to the high causer analysis and the low causer analysis. These two analyses predict different basic and derived word orders, as shown in (98) and (99), respectively, as discussed above. According to the linking hypothesis (100), the high causer analysis predicts that the theme–causer order takes longer to process than the causer–theme order, while the low causer analysis predicts the opposite pattern. These two distinct predictions are summarized below:

Testing these two predictions can help determine which analysis provides a better explanation. For this reason, I conducted a sentence-processing experiment.

6.3 Methods

6.3.3 Procedure

The experiment was run using PCIbex (Zehr & Schwarz 2018). I employed a speeded forced-choice judgment task (e.g., Chujo 1983; Tamaoka et al. 2005; Koizumi & Tamaoka 2010). In this task, participants were instructed to decide whether a visually presented sentence was a possible or impossible sentence in Japanese as quickly and accurately as possible via button press. “Possible” and “impossible” correspond to “acceptable” and “unacceptable”, respectively, in experimental linguistic terms. We did not use the latter two terms in our instruction to avoid confusion, as all participants were unfamiliar with technical terminology. The trial structure is represented in Figure 1. An eye fixation element (“++++++”) first appeared on the screen for 1,200 ms. A whole sentence was then presented visually. Participants were required to respond as quickly and accurately as possible, deciding whether the sentence was a possible sentence in Japanese by pressing ‘J’ for ‘possible’ or ‘F’ for ‘impossible.’ RT and accuracy for each sentence were recorded. An optional short break was provided midway through the experiment. Every participant completed twelve practice sentences before the actual experiment. The entire experiment lasted approximately 10–15 minutes.

Figure 1

Trial structure.

6.4 Results

6.5 Discussion

The RT results revealed that the theme–causer order took longer to process than the causer–theme order in experiencer subject causatives (i.e., causer–theme < theme–causer). This result confirms the prediction of the high causer analysis proposed in this paper but not the prediction of the low causer analysis. Therefore, I conclude that the high causer analysis is correct: the causer adjunct originates above the theme object in experiencer subject causatives.

Crucially, this conclusion has broader implications for the long-standing issue of the original position of the causer. As mentioned in Section 6.1, there are two schools of thought on the causer’s position. One posits that it occurs underlyingly below the object (e.g., Belletti & Rizzi 1988; Cheung & Larson 2015), while the other argues for the opposite relation (e.g., Fujita 1996; Sato & Kishida 2009; Alexiadou et al. 2015). The current results support the second view. The choice of analysis for the causer’s position is crucial for the development of the theory of experiencer-object constructions. As discussed in Section 6.1, while the inverse c-command relation in experiencer-object constructions seemingly supports the view that the causer originally occurs below the experiencer (Belletti & Rizzi 1988; Hasegawa 2001; 2004; 2007; 2016; Cheung & Larson 2015), there are several possible accounts of this relation under the view that the causer originates above the experiencer (Fujita 1996; Sato & Kishida 2009; Alexiadou & Anagnostopoulou 2020). It remains to be seen which analysis adopting the high causer position provides the best account of experiencer-object constructions.