On the causative and passive morphology in Japanese and Korean

2.1 Adversity by exclusion

It is often stated that indirect passives in Japanese automatically induce adversity by referring to examples like (4) (Washio 1993: 51).

In (4), the dominant interpretation of kodomo “child” is John’s child; and since the passive subject John is not an argument of the verb stem ker “kick,” it is usually taken to be an instance of indirect passive. However, the apparent adversity on the part of John is, according to Washio, due to the pragmatic force caused by the lexical semantics of the verb stem ker, which describes a rather violent action like kicking John’s child. Washio (1993: 51–2) instead claims that the real nature of adversity shows itself in one reading of examples like (5) (Washio 1993: 52).

In (5), zibun “self” is ambiguously interpreted; either John or Mary may be its antecedent. If it is John, the reading in (5a) results; and if it is Mary, that in (5b) results. Note that the verb stem home “praise” in (5) does not have such lexical semantics that induces adversity on its own as ker “kick” in (4) does. In the reading in (5a), the affectedness upon John may be taken to be neutral.^[2] On the other hand, the reading in (5b) forces the interpretation where John is adversely affected by the event of Mary’s praising her own child. Washio (1993: 52) argues that the obligatory adversity on the passive subject in (5b) is due to its exclusion from the event that is depicted by the verb stem home “praise.” More concretely, John is not excluded from the event of praising in (5a) in the sense that the object of praising is in his possession, i.e., his own child, but John is excluded from the relevant event in (5b) because he has nothing to do with Mary’s praising her own child. Hence, Washio (1993) refers to the obligatory adversity exhibited in (5b) as adversity by exclusion.^[3]

Inasmuch as Washio’s (1993) adversity-by-exclusion account is tenable, it is expected that adversity will necessarily be induced on the subject of passives whose verb stems are intransitives. This is indeed borne out as illustrated in (6).

In (6), the verb stem is nak “cry,” an unergative intransitive verb, and John does not have anything to do with Mary’s crying. In other words, John is excluded from the event where Mary is the sole participant. Hence, adversity results as expected.

Furthermore, since, according to Washio (1993), passive of the exclusion type is impossible in Korean, the Korean equivalent of (6) should be ungrammatical.

Since (7) is totally ungrammatical under the intended reading, the expectation is also borne out.^[4]

2.2 Causative passive

Washio’s (1993) adversity-by-exclusion account has such a wide empirical coverage that it accounts for the obligatory adversity reading on the passive subject in cases like (5b) and (6) above. However, this is not the whole story. As widely known, Japanese allows causative passive very productively as shown in (8c) as well as (3a).

(8a) is a simple active transitive sentence, and (8b) is the causative derived from (8a). In either case, kodomo “child” can be taken for John’s. This possessive relation may be maintained in (8c), which is the passive form of the causative sentence in (8b).^[5] Interestingly, although John in (8c) should not be excluded from the causative event depicted by (8b) (i.e., the event to be passivized in (8c)) because it has a possessive relation with kodomo, adversity reading is forced on John. This is simply unexpected by Washio’s (1993) adversity-by-exclusion account; hence, it calls for an explanation.

3.1 The architecture of verb phrase in Japanese

Since the seminal work by Cinque (1999), it has widely been recognized that the structure under T, as well as that in the C-area (Rizzi 1997), is much richer than conventionally assumed, as shown in (9). Following many authors, e.g., Pylkkänen (2000, 2008), McGinnis (2001), Borer (2005), MacDonald (2008), Travis (2010), and Fukuda (2012), among others, I will assume that the fully articulated structure under T in Japanese may include functional heads above and below Voice in the sense of Kratzer (1996). As I have proposed earlier, High Applicative (H-Appl) may appear above Voice (Aoyagi 2010),^[6] and Cause, if present, should be distinct from Voice^[7] (Pylkkänen 2008; Aoyagi 2017, 2020). Simultaneously, I will adopt the basic tenets of Distributed Morphology (e.g., Halle and Marantz 1993; Marantz 1997, 2001; Embick and Noyer 2007; among many others), but I will assume with Basilico (2008: 738–9) that an acategorial root (√R) should be categorized by a category-determining functional head (i.e., v, n, or a) before an internal argument (IA) is merged. As a result, there is no root phrase (√P), and vP corresponds to “VP” in the traditional framework. In the following is shown our “layered verb phrase hypothesis.”

Following the spirit of Reinhart (2002, 2016), I will propose that conventional thematic role labels like “agent” and “theme” should be specified by features. According to Reinhart (2002, 2016), two significant features are [cause change] and [mental state] (hereafter abbreviated as [c] and [m], respectively). The [c] feature indicates the presence or absence of a change-of-state component in the event described. A paradigmatic example is a causative sentence with -sase “cause” where the “causer” or the major participant of the causing event brings about a change that leads to the caused event. It is also involved in (simple) change-of-state predicates like sime “close(vt)” and simar “close(vi)” that exhibit transitive–unaccusative alternation because, in either case, the IA often referred to as “theme” undergoes a change of state and finally arrives at a closed state. On the other hand, the [m] feature indicates whether the mental capacity of a participant is relevant in the event described. For instance, while the “experiencer” of an event needs its mental capacity, the “cause” or “causer” of a causative event does not. Reinhart (2002, 2016) explicates that [+m] entails animacy, but not vice versa, and that a [+c] argument can be either animate or inanimate (see (10) and (11) below).

Furthermore, I will add a third feature [a(ctor)] (hereafter often labeled as Actor) which is assigned to the external argument (EA) of a transitive predicate whose semantics does not necessarily entail a change of state. This is exemplified by verbs of physical contact like tatak “beat (against)” and nade “caress” as shown in (12) below.

In what follows, the EA that is assigned [+m], [+c], and [+a] will be referred to as Sentient, Cause, and Actor, respectively, as indicated in (9) for an expository purpose. I will assume that the strictest version of the theta criterion in the sense of Chomsky (1981) should be abandoned, and an EA may assume more than one theta features.^[8]

According to Reinhart (2002, 2016), transitive–unaccusative alternation targets and dethematizes [+c], which means that the EA of a transitive verb that undergoes the alternation can be either animate or inanimate. This is by and large borne out as shown in the transitive and unaccusative pair of sentences in (10).^[9]

The transitive sime “close(vt)” in (10a) alternates with its unaccusative counterpart simar “close(vi)” in (10b) because the EA of the transitive sentence in (10a) is [+c], which is neutral with respect to animacy. If it is targeted and dethematized (as a result, it may be demoted to an adjunct), the unaccusative-intransitive sentence in (10b) results.

(11a, b) are on a par with (10a, b), respectively. However, as is often pointed out in the literature (see Inoue 1976: 61ff, for instance), the subject of transitive verbs of a certain class may render non-Actor, non-Cause reading. (11c) is such an example whereby Taro can be taken as Sentient (or “experiencer” in the conventional sense) due to [+m] assigned by H-Appl.

The dominant reading of (11c), where the cause kazi “fire” is expressed independently of the subject, is such that Taro was adversely affected by the event that his house was burned down by fire for which he was not responsible at all. In our terms, this reading is due to the presence of H-Appl which assigns Sentient to its Spec, where Taro in (11c) is accommodated.

Finally, the need for [+a], in addition to [+c] and [+m], is evidenced by sentences like (12a, b).

First of all, verbs of physical contact like tatak “beat (against)” in (12a) and nade “caress” in (12b) do not require their IA to undergo a change of state. This is evidenced by the fact that secondary resultative predicates do not co-occur with these verbs. Hence, [+c] is not involved. This is also the reason that neither tatak nor nade may undergo transitive–unaccusative alternation; i.e., neither of them has a morphologically related unaccusative counterpart. Next, the fact that their subjects can be either animate or inanimate indicates that [+m] is not involved, either. Then if the subject of a verb of physical contact is to be interpreted, a third thematic feature should be posited. This is the reason that I propose [+a] (or Actor as a convenient label) which is assigned by Voice as indicated in (9).

3.2 Loci of cause

Among the three EA-introducing functional heads under T, i.e., Cause, Voice, and H-Appl, only Voice is obligatory. This is because the type of voice of a sentence must be determined anyway. Furthermore, H-Appl, if present, is always located above Voice (Aoyagi 2010). On the other hand, Cause has much wider freedom in distribution. According to Pylkkänen (2008: 85), Cause is classified into three types depending on its selectional properties as shown in (13).

All three types are attested in Japanese. Root-selecting Cause in (13a) is exemplified by “denominal” verbs like modo-s “origin-caus” (return(tr))^[11] and wata-s “ocean-caus” (pass(tr) (across the ocean)). As for vP-selecting Cause in (13b), the change-of-state verbs (or “lexical” causative verbs) like kusar-as(e) “spoil(vi)-caus” (spoil(vt)) and sin-as(e) “die-caus” (make/let die), which fill in the gap of morphologically related transitive verbs (Miyagawa 1984; Kuroda 1993), provide evidence for it. Finally, phase-selecting Cause in (13c) is always spelled out as -sase as illustrated in (14).

I will assume with Pylkkänen (2008: 85) that an EA-introducing head is defined as a phase head. The verb stems yom “read” in (14a) and nak “cry” in (14b) are transitive and unergative intransitive, respectively. In either case, Mary is interpreted as Actor of the event depicted by the verb and may antecede zibun “self,”^[12] so that it qualifies as EA. Hence, α is identified as Voice. Thus, -sase in (14a, b) instantiates the head of phase-selecting Cause in (13c). After all, Cause may appear either higher or lower than Voice in Japanese.^[13]

3.3 Types of voice

As stated above, only Voice is obligatory among the three functional heads under T and above v since every single sentence must somehow indicate its voice, e.g., active, passive, middle/unaccusative, etc. I will assume that Voice is classified into three distinct types by two binary features [±passive] and [±EA]^[14] as shown in (15).

Given that [+passive] Voice is that of direct passive, its EA is dethematized by definition, and so [+EA] and [+passive] are mutually exclusive.^[15] First of all, the [+EA] Voice in (15.i) which assigns Actor to its Spec is exemplified by sentences with verbs of physical contact in (12) above. Next, the [+passive] Voice in (15.ii) is found in direct passive sentences with an inanimate subject as shown in (16).

In (16a), the IA doa “door” is assigned the nominative case because it becomes the sole argument due to John, which is potentially Cause/Actor, being dethematized and demoted to an adjunct. The same is true of (16b). Finally, the neutral [−EA, −passive] Voice in (15.iii) is attested in unaccusative sentences like (10b) and (11b) as well as “non-agentive” transitive sentences like (11c). The difference between (10b) and (11b), on the one hand, and (11c), on the other, is the presence of H-Appl that assigns Sentient ([+m]) to its Spec in the latter.

3.4 Voice–cause bundling

It is rather evident from our discussion about unaccusative sentences in (10b) and (11b) and “non-agentive” transitive sentences like (11c) that Cause, as well as Voice, may be [−EA], so that it cannot assign Cause to its Spec. In such a case, Cause is demoted to an adjunct or simply implied.

Pylkkänen (2008: 84) states that while Cause is bundled with Voice in English, Voice–Cause bundling does not take place in Japanese.^[16] However, I will claim on the contrary that Voice is always bundled with vP-selecting Cause immediately below it in Japanese.^{[17] [18]} In our terms, Voice–Cause bundling takes place in the following four separate cases.

The first two cases are straightforward. (18a) is what happens in sentences with change-of-state transitive verbs that undergo transitive–unaccusative alternation as in (10a) and (11a), and (18b) is called for in the corresponding unaccusative sentences like (10b) and (11b). On the other hand, the latter two cases are rather obscure. However, they are existent, or it seems that their existence is innocuous. (18c) arguably instantiates a change-of-state transitive event in which the EA (Actor) acts on something else than the IA (“theme”). This is illustrated by the following example, which is modified on (10a).

Now the relevant reading of (19) is that in which John simply pressed a button without acting on the door directly, and the first-order cause of the door-closing event is a mechanical system that is instigated by the button-pressing event. In this construal, John is only the Actor of the latter event, not necessarily be taken for the Cause of the former event. Finally, (18d) can be argued to instantiate a change-of-state event where the EA is an inanimate Cause as in (20) (modified on (11a)).

In (20), Meireki-no taika “the big fire of Meireki (period)” can be taken for Cause.^[19]

5.1 Passive–causatives??

As we have seen above, causative–passives are highly productive in Japanese. On the other hand, passive–causatives are highly marked if possible at all. The passive–causative in (25b) and its underlying passive in (25a) are from Saito (1982: 92).

Factual evaluation of examples like (25b) varies among Japanese linguists. For instance, while Hoshi (1999: 204) cites (25b) as an acceptable example, Nakamura and Hashimoto (1994) flatly reject it. Ishizuka (2012: 30, footnote 5) states that while the causative–passive sase-rare is canonical, the inverse passive–causative rare-sase is definitely marked by drawing the following pair of examples.^[22] (26a, b) are from Ishizuka (2012: 30), and the judgments are hers.

As a Japanese speaker, I believe that examples like (25b) and (26b) are highly awkward at best. However, if the passive subject is inanimate, the resultant passive–causative becomes even worse as exemplified below.

(27a) and (28a) are instances of direct passives with a [–animate] subject. If such a direct passive is embedded under -sase as in (27b) and (28b), the result is devastatingly worse than examples like (25b) and (26b). This contrast is rather sharp among the Japanese speakers who I have consulted. Even those who reject examples like (25b) and (26b) report that they are relatively better than (27b) or (28b). This contrast, if real, calls for an account.

In our terms, since the passive subject in (25a) and that of the underlying passive of (26b), i.e., Naomi-ga John-ni tazune-rare-ta “Naomi was visited by John,” are [+animate]; hence, they may be accommodated in Spec of H-Appl. If, as a marked option, Cause is allowed to be merged with H-ApplP, the representation of (25b) will look like (29).

In (29), Cause is merged with H-ApplP. I claim that this is marginally possible because H-Appl assigns Sentient ([+m]) to EA (John) in its Spec; thus, Cause in (29) counts as an instance of phase-selecting Cause (see (13c) above). This is simply impossible for (27b) or (28b). Since the subject of the direct passive clause embedded under -sase is an inanimate entity, it cannot be accommodated in Spec of H-Appl. Due to the lack of H-ApplP, Cause should be merged with [+passive] VoiceP. However, it is banned because the latter does not constitute a phase due to dethematization of its Spec as shown in (30).^[23]

In (30), VoiceP* is offensive because, its head being specified as [+passive], it does not count as a phase.^[24] Thus, Cause in (30) cannot be an instance of phase-selecting Cause in (13c). Nor does it qualify as Root-selecting Cause in (13a) or vP-selecting Cause in (13b), either.

5.2 Circumfixing double causatives

Our account of the marginal status of passive–causatives like (25b) is further supported by examples like (31b) and (32b).

(31a) and (32a) are examples of the highly productive causative–passive, and they are both perfectly grammatical, as expected. In (31b) and (32b), a causative–passive is further embedded under -sase. As a result, the passive -(r)are is circumfixed by two occurrences of Cause. It is true that both (31b) and (32b) are less than perfect at best. However, many Japanese speakers find that they are no worse than passive causatives like (25b).

As noted by Kuroda (1993), while some instances of the suffix -(s)as are phonologically reduced forms of the full-fledged Cause -(s)ase (e.g., hatarak-as < hatarak-ase “work-caus (cause to work)),” others are part of the so-called “lexical” causatives (e.g., ugok-as “move(vt)” < ugok “move(vi)”). In our terms, while the former is an instance of phase-selecting Cause, the latter is that of vP-selecting Cause (see (13) above).^[25] According to Kuroda (1993: 8–10), one touch stone that distinguishes between them is the double causative test; if double causative is possible, it is part of a “lexical” causative; otherwise, it is a phonologically reduced form of the productive -(s)ase (i.e., the spell out of phase-selecting Cause), as exhibited in the contrast between (33a) and (33b) drawn from Kuroda (1993: 9) with slight modifications.

While (33a) passes the double causative test, (33b) does not. This means that -as in the former is part of a “lexical” causative (i.e., root-selecting Cause in our terms), but that in the latter is a reduced form of the phase-selecting -(s)ase.

Let us return to tat-as “stand(vi)-caus” in (31) and ne-sas “lie.down-caus” in (32). First of all, each has a morphologically related transitive form, tat-e “stand(vt)” and ne-kas “lay.down,” respectively, distinct from tat-as in (31) and ne-sas in (32). Since each of them consists of an intranstive verb stem and the causative suffix -(s)as, they come so close to transitives semantically. However, they are not simple transitives (or “lexical” causatives). This is evidenced by the fact that neither of them passes Kuroda’s double causative test as shown in (34).

Neither *tat-as-ase “stand(vi)-CAUS-CAUS” in (34a) nor *ne-sas-ase “lie.down-CAUS-CAUS” in (34b) is permissible.

Furthermore, the instransitive verb stems tat “stand” in (31) and ne “lie.down” in (32) are unergatives when their subjects are animate as evidenced by the following examples.

In either case, Mary may be modified by the so-called “agent-oriented” adverb wazato “on purpose” and may be the antecedent of zibun “self.” This is a very strong indication that Mary is located in Spec of [+EA] Voice which assigns Actor to its Spec (with the possibility that it is further raised to Spec of H-Appl for Sentient).

Given these qualifications, the structure of (31b), for instance, can be represented as follows.

In (36), the lower Cause is the commonest type that is merged with VoiceP whose head assigns Actor to its Spec, but the use of the higher Cause, which is located higher than H-Appl, is a marked option as we have claimed in Section 5.1 (see (29)). Thus, the less-than-perfect status of the passive -rare sandwiched by two causative morphemes in (31b) and (32b) is on a par with that of passive–causatives like (25b).

6.1 The ambiguity of /Hi/ in Korean

It is widely recognized that the four {i, hi, li, ki} out of the seven causative morphemes {i, hi, li, ki, wu, kwu, chwu} can also be used as passive morphemes (e.g., Lee 1973; Yang 1979; Washio 1993; Kim 1994, 1998; Yeon 2002, 2003; Oshima 2006; Aoyagi 2007, among many others) as illustrated in the following examples.^{[27] [28]}

(37a, b) can be interpreted as either causative or passive. However, as indicated in the English glosses, the interpretive freedom of the (retained) accusative object in passives is more narrowly restricted than in causatives. This is because, unlike in Japanese, the subject of morphological passive in Korean may not be excluded from the event depicted by the verb stem. Hence, passives can be formed on intransitive verbs in Japanese, but not in Korean, as shown in (38a, b) reproduced from (6) and (7), respectively.

Since Mary is the sole participant of the crying event where John is not involved, Japanese allows the passive in (38a), but its Korean counterpart in (38b) is impossible. In addition, (37b) should be contrasted with the following Japanese counterpart.

While (39) in Japanese allows the accusative object (=hair) to be either John’s as in (39.i) or Mary’s (or even someone else’s) as in (39.ii), (37b) in Korean does not allow the latter interpretation under its passive reading. Aoyagi (2010) argues that H-Appl is located above Voice in Japanese, and it is available for the purpose of accommodating either benefactive or malefactive (or adversative) Sentient argument (assigned [+m]); however, H-Appl above Voice is not available in Korean.^[29] Thus, while -rare as the exponent of malefactive H-Appl makes the indirect passive in (36a) possible, the Korean /Hi/ may not occupy the same location.

However, the question still remains as to why /Hi/ can be ambiguous between causative and passive as shown in (37a, b) above. In what follows, we will argue that /Hi/ in Korean is univocally an exponent of Cause, and its passive sense comes as the result of Voice–Cause bundling.

6.2 The loci of /Hi/ in Korean

As we have mentioned in Section 3.2, Pylkkänen (2008: 85) proposes that Cause may appear at three different locations as indicated in (40) (reproduced from (13)).

As we have noted, all three instances of Cause are attested in Japanese. By null hypothesis, it is not surprising if all three types of Cause are also available in Korean. Indeed this seems to be the case.

First of all, root-selecting Cause in (40a) is instantiated by “de-adjectival” verbs like nelp-hi “widen(vt)” < nelp “wide(adj)” and noph-i “heighten(vt)” < noph “high(adj),” where /Hi/ can simultaneously be taken as the exponent of the category-determining v (Aoyagi 2017).

Next, vP-selecting Cause is exemplified by sentences with change-of-state verbs like the following.

The verb stem in (41a) is nal “fly(vi),” an intransitive, and the exponent of Cause is /Hi/. On the other hand, the verb stem in (41b) is the transitive tat “close(vt);” hence, the exponent of Cause is ø in the same way as its Japanese counterpart sime “close(vt)” (see (10a) in Section 3.1 above). Either of the verbs may co-occur with a secondary resultative predicate, which indicates that they are indeed change-of-state verbs. Furthermore, the presence of Cause which assigns [+c] to its Spec is evidenced by the fact that the subject can be either animate or inanimate.

Finally, phase-selecting Cause is instantiated by causatives formed on transitive verbs like (42b) as well as (37a, b) under their causative interpretations.

In (42b), the transitive clause with Actor EA (=Mary) in (42a) is embedded under /Hi/ as a causative morpheme. Thus, all three types of Cause are found in Korean as well as in Japanese.

6.3 Passive as the result of Voice–Cause bundling

Let us return to the ambiguity of /Hi/ between causative and passive as noted in Section 6.1. To begin with, the causative sense of (37b), for instance, can result from the following construction.^[30]

The verb stem kkakk “cut” in (37b) is transitive, and [+EA] Voice assigns Actor to Mary in its Spec. Since VoiceP counts as a phase, phase-selecting Cause in (40c) can be employed. If it is [+EA], it assigns Cause to John in its Spec. Inasmuch as (37b) is represented as a causative sentence as in (43), the accusative marked IA hair need not be John’s, and it can be Mary’s or even somebody else’s.

Now let us explicate how the passive sense of (37a, b) obtains. Suppose that Cause can be [–EA] in Korean as well as in Japanese (see (17) in Section 3.4 above) and that Voice can also be [+passive], hence, [–EA]. Then Voice–Cause bundling should take place as indicated below.

(44a) is illicit for the same reason as (30) in Section 5.1 is. Since VoiceP without an EA does not constitute a phase, the selectional property of Cause is not satisfied in (44a) (see (40)). However, if bundling applies, Cause will be able to select for vP as indicated in (44b) (see (40b)).^[31] In our terms, the passive sense of (37a, b) is not solely due to /Hi/, but it results from the combination of [+passive] Voice and /HI/ as Cause without an EA (=“causer”). Furthermore, if the possessor of IA John in (44b) can be moved to Spec of T (e.g., for an EPP reason), the possessor passive in (45) will be derived.^[32]

Movement of the possessor John restricts the interpretation of the retained object hair to his own.^{[33] , [34]} Thus, our assumption that /Hi/ is univocally causative, and bundling takes place when both Cause and Voice are [–EA] successfully accounts for the apparent ambiguity of /Hi/ between causative and passive. In our terms, the passive sense in examples like (37a, b) is derived from causative. This derivational direction is supported widely in the literature (e.g., Keenan and Dryer 1985; Haspelmath 1990; Washio 1993; Yeon 2002, 2003, among others).

6.4 An implication to /Hi/ vs /Hu/

As noted in Section 6.1, only the first four, represented as /Hi/, among the seven causative morphemes {i, hi, li, ki, wu, kwu, chwu} can also be used as passives; in other words, the latter three, represented as /Hu/, are exclusively causatives. Indeed, this contrast between /Hi/ and /Hu/ follows from their selectional properties.

As shown in (46), /Hi/ can be attached to dyadic stems as well as monadic stems (i.e., adjectives and intransitives). On the other hand, /Hu/ can be attached only to monadic stems as shown in (47).^[35]

On the one hand, (46a–c) suggest that /Hi/ can instantiate root-selecting, vP-selecting, and phase-selecting Cause, respectively. On the other hand, (47) suggests that /Hu/ may instantiate root-selecting and vP-selecting Cause but not phase-selecting Cause (i.e., Cause that selects VoiceP with EA).^[36] This state of affairs is shown in (48) below.

As shown in (48), both /Hi/ and /Hu/ may appear as root-selecting and vP-selecting Cause. However, very significantly, /Hi/ but not /Hu/ may appear as phase-selecting Cause. If the passive sense is derived by bundling of Cause and immediately lower Voice as envisaged in (44), only /Hi/ may be involved.