Two types of Intransitive Vs: FRAMEWORK The Unaccusative ...
Transitivity of intransitives and causative constructions in … · 7 The choice is free for...
Transcript of Transitivity of intransitives and causative constructions in … · 7 The choice is free for...
2
1. Introduction
Syntactic causative sentences in Japanese are derived by associating causative morpheme -(s)ase
with stem verbs.1 Consider the following examples:
(1) a. [Hanako-ga hon-o yom] ta.2
Hanako-NOM book-ACC read PAST
‘Hanako read a book.’
b. Taroo-ga [Hanako-ni hon-o yom]-ase ta.
Taroo-NOM -DAT -CAUS PAST
‘Taro made/let Hanako read a book.’
(2) a. [Hanako-ga odorok] ta.
Hanako-NOM be.surprised PAST
‘Hanako was surprised.'
b. Taroo-ga [Hanako-o odorok]-ase ta.
Taroo-NOM -ACC CAUS PAST
‘Taro surprised Hanako.’
(1a) and (2a) illustrate that the highest argument in a propositional phrase is marked with
nominative case. When the proposition is selected by -(s)ase as in (1b) and (2b), the highest
argument, called CAUSEE henceforth, is marked with either dative -ni or accusative -o.
In the generative literature it has been commonly assumed that case-marking on Causees is
semantically constrained. That is, the ni-causative is assigned a permissive interpretation, and the
o-causative is assigned a coercive interpretation. It has also been pointed out, however, that this
familiar dichotomy is descriptively inadequate. This paper provides a syntactic account for the
case marking in causative constructions, showing that it gives a more precise explanation to their
3
semantic properties as well.
The organization of this paper is as follows. In section 2 I briefly review previous analyses and
point out their problems. A new analysis is given in section 3. Section 4 shows how the new
analysis accounts for the relevant data. In section 5 I make my proposals more precise and
consider further consequences. Section 6 concludes the discussion.
2. Previous Analyses
2.1. Observations
Causative constructions have been eagerly discussed in the generative literature (e.g. Kuroda 1965,
Harada 1973, Kuno 1973, Shibatani 1973, 1976, 1978, Tonoike 1978, Terada 1990, Dubinsky
1994, Harley 1995, Miyagawa 1999). The Causee may be marked with either -ni or -o, and that
the possible marking is determined for each stem verb. Sentences (3a-c) exemplify this.
(3) a. Taroo-ga [Hanako{-ni/ *-o} hon-o yom] -ase ta.
Taroo-NOM Hanako{-DAT/ *-ACC} book-ACC read CAUS PAST
‘Taro made/let Hanako read a book.’
b. Taroo-ga [Hanako{-o/ *-ni} odorok] -ase ta.
Taroo-NOM Hanako{-ACC/ *-DAT} be.surprised CAUS PAST
‘Taro surprised Hanako.’
c. Taroo-ga [Hanako{-ni/ -o} odor] -ase ta.
Taroo-NOM Hanako{-DAT/ -ACC} dance CAUS PAST
‘Taro made/let Hanako dance.’
The stem verb yom ‘read’ in (3a) is transitive. (3a) illustrates that ni-causative is obligatory when a
transitive verb is causativized. In (3b), on the other hand, the stem verb odorok ‘be.surprized’ is
4
unaccusative. The sentence illustrates that o-causative is obligatory for an unaccusative stem verb.
When the stem verb is unergative, as odor ‘dance’ in (3c), either ni- or o-causative is allowed.
According to Kuroda (1965), the ni-causative denotes permissive causation while the o-causative
denotes coercive causation. When Hanako in (3c) is marked with -ni, for example, Hanako wants
to dance and Taro permits her to dance. When Hanako is marked with -o, on the other hand,
Hanako does not want to dance but Taro forces her to dance. Many others have since followed this
line of explanation (e.g. Harada 1973, Kuno 1973, Shibatani 1976, Dubinsky 1994, Tsujimura
1996, Hasegawa 1999, Miyagawa 1999). Several researchers have provided counterexamples but
generally they have been either considered exceptional or simply ignored.
The above observations are summarized in (4) and (5).
(4) Stem verb Possible case for the Causee
a. Transitive -ni
b. Unaccusative -o
c. Unergative -ni or -o
(5) Causative Interpretation
a. Ni-causative Permissive
b. O-causative Coercive
2.2. Two types of analysis
In the literature various analyses have been proposed for causative constructions. Ignoring minor
differences, Tonoike (1978) categorizes them into two types: the o-extra analysis and the ni-extra
analysis. I briefly explain each type of analysis below.
The o-extra analysis is proposed by Kuroda (1965), Kuno (1973), Shibatani (1978) among
others. They analyze the two types of causatives as follows:
5
(6) a. Ni-causative: [IP Causer [IP Causee-ni ... V]-(s)ase ...]
b. O-causative: [IP Causer Causee1-o [IP (Causee1)/PRO1 ...V]-(s)ase ...]
The o-causative is different from the ni-causative in that it contains a matrix object selected and
case-marked by -(s)ase. The o-extra analysis was popular in the EST era since it reflects the
intuition that the o-causative conveys more direct (i.e. coercive) causation than the ni-causative.
The ni-extra analysis is proposed by Tonoike (1978), Terada (1990), Harley (1995), Miyagawa
(1999) and many others. They assign just the opposite structures to the two causatives:
(7) a. Ni-causative: [IP Causer Causee1-ni [IP (Causee1)/PRO1 ...V]-(s)ase ...]
b. O-causative: [IP Causer [IP Causee-o … V]-(s)ase ...]
It is the ni-causative that contains a matrix object selected and case-marked by -(s)ase. Miyagawa
(1999) provides evidence for the ni-extra analysis. Here I give one piece of his evidence. In (8a, b)
a PP (kooen-e ‘to the park’) is scrambled to the left of the Causee (kodomo ‘child’).
(8) a.???Taroo-ga kooen-e1 kodomo-ni t1 ik-ase ta.
Taroo-NOM park-to child-DAT go CAUS PAST
b. Taroo-ga kooen-e1 kodomo-o t1 ik-ase ta.
-ACC
‘Taro made/let his child go to the park.’ (Miyagawa 1999: 249)
The examples above show that scrambling of the PP to the left of the Causee is not allowed in the
ni-causative, but OK in the o-causative. Under the ni-extra analysis this contrast is given a simple
6
account. Since the ni-Causee is object of the matrix clause, the PP in (8a) moves into the matrix
clause, presumably to the VP-adjoined position. In (8b), on the other hand, the scrambling of the
PP may be local since the o-Causee is in the embedded IP. Presumably the PP is adjoined to the
embedded IP. Under the ni-extra analysis the scrambling in (8a) is illegitimate because, according
to Miyagawa, VP is not a legitimate adjunction site.
The contrast in (8) would be hard to explain under the o-extra analysis as illustrated in (6a). It
would have to claim that local, IP-adjoining scrambling in (8a) is deviant, and non-local,
VP-adjoining scrambling in (8b) is legitimate. This is obviously incompatible with the general
theory of movement. The contrast in (8) suggests that the ni-extra analysis should be the better
one.
2.3. Shared assumptions
With all the differences between the two approaches, they share the following assumptions:
(9) Syntax does not constrain the choice between ni- and o-causatives.
(10) The ni-causative denotes permissive causation; the o-causative denotes coercive causation.
(11) THE DOUBLE-O CONSTRAINT (DOC) (cf. Harada 1973: 138, Shibatani 1978: 328)
A clause cannot contain more than one occurrence of o-marked object.
Given these assumptions, the constraints on case-marking as shown in (3a-c) are explained in the
following manner.
Either ni- or o-causative is available for unergative stem verbs, as in (3c). The choice is free in
syntax, and each causative is assigned a distinct interpretation. When Hanako is willing to dance,
ni- (i.e. permissive) causative is compatible with the situation. When Hanako is reluctant to dance,
on the other hand, o- (i.e. coercive) causative is compatible with the situation.
7
The choice is free for unaccusative stem verbs as well. But the ni-causative as in (3b)
generates a strange interpretation: Hanako is willing to get surprised and Taro allows her to do so.
Unaccusative-stem ni-causatives are thus deviant in semantics. The Theme, lacking volition in the
denoted eventuality, should be marked with –o when causativized.
The choice is also free for transitive stem verbs. The o-causative in (3a) is OK in syntax, but
excluded at PF by the DOC (11): Hanako-o cannot occur with hon-o. According to Shibatani
(1973, 1976, 1978) and Miyagawa (1999), the DOC violation can be evaded by changing case
morphology at PF:
(12) Syntax: Taroo-ga [Hanako-o hon-o yom]-ase ta. (=(3a))
↓
PF: Hanako-ni
The spelled-out sentence in (12) is ni-causative, but o-causative in syntax and in semantics. Hence
(12) is assigned coercive interpretation. Consequently, transitive-stem ni-causatives are ambiguous
between permissive and coercive.
2.4. Problems with the previous analyses
The previous studies have revealed a various properties of causative constructions, providing more
sophisticated analyses in a gradual way. However, the analyses suggested so far are problematic in
that they rely on dubious assumptions.
The first problem concerns the observational adequacy of (10). The permissive vs. coercive
dichotomy does not always hold. It is evident with transitive stem causatives: the ni-causative can
be either permissive or coercive. This problem may be evaded by assuming (12), but
case-changing operation at PF is itself a problem unless it is supported by any independent
8
evidence. Neither does this dichotomy deal with unergative stem causatives. Admittedly, in (3c)
we do sense that Taro should be more considerate of Hanako in the ni-causative than in the
o-causative. However, we sense such difference ONLY WHEN WE COMPARE A MINIMAL PAIR. When
we examine causative sentences separately, any causative is ambiguous between permissive and
coercive, which has been repeatedly pointed out in the literature (e.g. Harada 1973, Tonoike 1978,
Teramura 1982). The following examples are given by Tonoike (1978: 6):
(13) Kare-wa [iyagaru imooto-ni beddo-de ne] sase ta.
he-TOP reluctant sister-DAT bed-in sleep CAUS PAST
‘He made his reluctant sister sleep in the bed.’
(14) (Moo uma-o turete kaeru zikan datta ga, amari yukaisoo-ni kakoi-no naka-o hasitte iru
node,)
(‘The time had come for Taro to take the horse back, but, because the horse was running so
joyously in the arena, ...’)
Taroo-wa sono mama [uma-o hasir]-ase ta.
Taroo-TOP without stopping horse-ACC run CAUS PAST
‘Taro let the horse run for a little while.’
Each causative sentence contains an unergative verb, ne ‘sleep’ in (13) and hasir ‘run’ in (14). The
ni-causative (13) is obviously coercive, and the o-causative (14) permissive. This fact suggests that
permissive/coercive reading should be assigned by the context, not by the structure of the
sentence.
To sum, either ni- or o-causative sentence is ambiguous. The permissive vs. coercive
dichotomy as in (10) is sensed only when we compare a minimal pair. It follows that
transitive-stem ni-causative sentences are always ambiguous since they have no o-causative
9
counterparts.
A second problem with the previous analyses is the DOC (11). The DOC is nothing but a
description of data. It is a tautology to account for the data (the deviance of two o-phrases) in
terms of the DOC (the ban of two o-phrases).3 In minimalist inquiries it would be preferable if we
explained the absence of transitive o-causatives without recourse to the DOC.
3. A proposal
In this section I propose a new analysis of ni- and o-causatives. The syntactic facts to be explained
are quite simple:
(15) a. Ni-causative is obligatory for transitive stem verbs.
b. O-causative is obligatory for unaccusative stem verbs.
c Either causative is OK for unergative stem verbs.
My proposal is also a simple one:
(16) a. -(S)ase[+ni] takes a v*P.
b. -(S)ase[+o] takes a vP.
Following Hasegawa (1999) and Inoue et al. (1999), I assume that a causative morpheme takes a
verbal complement. Several other researchers have assumed that -(s)ase embeds S (or IP/TP) (e.g.
Shibatani 1976, 1978, Tonoike 1978, and Miyagawa 1999). Given the Predicate-Internal Subject
Hypothesis, however, the minimal propositional phrase should be v(*)P. It therefore seems
reasonable to assume v(*)P-embedding unless there is evidence that T must be present in the
embedded proposition.
10
Given (16a, b), the transitive-stem causative in (3a) and the unaccusative-stem causative in
(3b) are assigned the structures (17a) and (17b), respectively.
(17) a. Taroo-ga [v*P Hanako-ni [VP hon-o yom] v*]-ase[+ni] ta.
Taroo-NOM Hanako-DAT book-ACC read CAUS PAST
‘Taro made/let Hanako read a book.’
b. Taroo-ga [vP [Hanako-o odorok] v] -ase[+o] ta.
Taroo-NOM Hanako-ACC be.surprised CAUS PAST
‘Taro surprised Hanako.’
Different case assignment reflects different embedding. Since a transitive verb constitutes a v*P, it
must be selected by ni-assigning –(s)ase ((16a)). Ni-causative is hence obligatory for transitive
stem verbs. On the other hand, an unaccusative verb constitutes a vP. Since a vP is selected by
o-assigning –(s)ase ((16b)), o-causative is obligatory for unaccusative stem verbs. The
case-marking facts in (15a) and (15b) are thus given a simple syntactic account.
An apparent problem arises with regard to (15c). An unergative verb should constitute a v*P.
Hence it is predicted that only ni-causative is derivable with an unergative verb, as illustrated in
(18).
(18) Taroo-ga [v*P Hanako-ni [VP odor] v*] -ase[+ni] ta.
Taroo-NOM Hanako-DAT dance CAUS PAST
‘Taro made/let Hanako dance.’
If we wanted to maintain (16a, b), we would have to assume the following structure for an
unergative-stem o-causative sentence:
11
(19) Taroo-ga [vP [VP Hanako-o odor] v] -ase[+o] ta.
Namely, an unergative verb (odor ‘dance’ in this case) takes the Agent argument in the internal
argument position, constituting a vP. Henceforth I call such verbs UNACCUSATIVIZED. Apparently
ridiculous as it may seem, I argue that this is in fact the case. In section 4 I show how well this
proposal accounts for a variety of syntactic and semantic properties of causative constructions.
The notion of UNACCUSATIVIZATION will be made precise, but not until section 5. For the time
being let us see how far we can go with this analysis.
4. Explanation
4.1. Syntactic properties
4.1.1. Licensing of non-adjacent floating numeral quantifiers
As illustrated in (18) and (19), the ni-Causee is base-generated in SPEC of v* whereas the
o-Causee is base-generated in COMP of V. From this assumption the asymmetry in (20) naturally
follows.
(20) a. ?*Taroo-ga [gakusei-ni kinoo tosyokan-de 3-nin hatarak]-ase ta.
Taroo-NOM student-DAT yesterday library-at 3-CL work CAUS PAST
‘Taro made/let three students work at the library.’
b. Taroo-ga [gakusei-o kinoo tosyokan-de 3-nin hatarak]-ase ta.
-ACC
The ni-Causee in (20a) cannot license the non-adjacent floating numeral quantifier (FNQ) whereas
the o-Causee in (20b) can. Under the suggested analysis the embedded clauses in (20a, b) are
12
assigned the following structures (21a) and (21b), respectively.
(21) a. [v*P gakusei-ni [VP kinoo tosyokan-de 3-nin hatarak] v*]
b. [vP [VP gakusei-o1 kinoo tosyokan-de t1 3-nin hatarak] v]
The FNQ 3-nin, put between the locative adjunct and the verb, should be in the VP. In (21a) the
ni-Causee is base-generated in SPEC of v*. Hence the FNQ and the quantified NP do not form a
constituent at any point of the derivation, which renders the sentence deviant (cf. Miyagawa 1989,
Watanabe 2006). In (21b), on the other hand, the o-Causee is base-generated in COMP of V. It is
scrambled to the left of kinoo ‘yesterday’ later in the derivation. But its copy in the base position
(t1) forms a constituent with the FNQ. The FNQ is thus licensed by the o-Causee.4
4.1.2. Double causativization
Let us then consider the constraint on double causativization. Shibatani (1976) observes that
causativization can take place more than once.
(22) Taroo-ga Ziroo-ni Itiroo-o aruk-ase ta.
Taroo-NOM Ziroo-DAT Itiroo-ACC walk CAUS PAST
‘Taro made/let Jiro make Ichiro walk.’ (Shibatani 1976: 244)
Causativization is applied twice in the derivation of (22), although only one causative morpheme
appears on the surface. Following Shibatani, I assume that one of the iterated morphemes is
reduced at PF. The derivation proceeds roughly as follows:
13
(23) a. [Itiroo aruk]
b. O-causativization: [Ziroo [Itiroo-o aruk] –ase]
c. Ni-causativization: [Taroo [Ziroo-ni [Itiroo-o aruk] –ase] –(s)ase] (o ni)
The verb first undergoes o-causativization, and then ni-causativization, which I call “o ni”
causativization. The lower Causee Itiroo is marked with -o and the higher Causee Ziroo is marked
with -ni. The double causative structure denotes double causation: Ichiro was caused to walk by
Jiro, and the causation was caused by Taro.
It is strictly restricted which type of causativization should be applied first and second.
Consider the following examples. Each sentence is derived through two applications of
causativization.
(24) a. *Taroo-ga [Ziroo-o [Itiroo-o aruk] ase ] (s)ase ta. (*o o)
Taroo-NOM Ziroo-ACC Itiroo-ACC walk CAUS CAUS PAST
b. *Taroo-ga [Ziroo-o [Itiroo-ni aruk] ase] (s)ase ta. (*ni o)
c. ??Taroo-ga [Ziroo-ni [Itiroo-ni aruk] ase] (s)ase ta. (??ni ni)
The awkwardness of (24c) disappears if the two ni-phrases are not adjacent, as the following
example shows.
(25) Taroo-ga [Ziroo-ni [hamabe-de Itiroo-ni aruk] ase] (s)ase] ta.
beach-LOC
‘Taro made/let Jiro make Ichiro walk on the beach.’
14
Awkwardness of (24c) probably has to do with phonology, i.e. the iteration of phrases of the same
case. The structure (24c) is legitimate in syntax. The syntactically legitimate patterns are, therefore,
o ni as in (23) and ni ni as in (24c).
The constraint on double causativization has been unnoticed in the literature. Actually, the
previous analyses may not be able to deal with the above facts since they do not assume any
syntactic constraints on the derivation of causative sentences. They may exclude (24a) as the
violation of the DOC, but referring to the DOC is itself a problem. What is worse, there is no way
to exclude (24b). When the first causativization is applied, the following structure is obtained:
(26) [Ziroo [Itiroo-ni aruk] –ase]
Since they do not contain any o-DP, the previous analyses would predict that either ni- or
o-causativization is applicable to (26). In reality, however, the application of o-causativization as in
(24b) leads to deviance.5
The suggested analysis gives a simple account for why o ni and ni ni are the only
legitimate combinations. Suppose that ni-causativization applies first. The derivation proceeds
roughly as follows:
(27) a. [v*P Itiroo [VP aruk-] v*]
b. Ni-causativization: [v*P Ziroo [v*P Itiroo-ni [VP aruk-] v*] ase[+ni] ]
c. Ni-causativization: [v*P Taroo [v*P Ziroo-ni [v*P Itiroo-ni [VP aruk-] v*] ase] (s)ase[+ni] ]
Since aruk ‘walk’ is unergative, it constitutes a v*P (as in (27a)). When the first causativization
applies, the v*P must be selected by -(s)ase[+ni] ((16a)). The application generates another v*P in
which the causative predicate -(s)ase takes a Causer argument in SEPC of v*. When the second
15
causativization applies, the v*P must be selected by -(s)ase[+ni] as well. When ni-causativization
applies first, therefore, it must be followed by ni-causativization (ni ni).
Consider next what if o-causativization applies first. Recall that an unergative verb can be
UNACCUSATIVIZED. In that case the derivation proceeds as follows:
(28) a. [vP [VP Itiroo aruk] v]
b. O-causativization: [v*P Ziroo [vP [VP Itiroo-o aruk] v] ase[+o] ]
c. Ni-causativization: [v*P Taroo [v*P Ziroo-ni [vP [VP Itiroo-o aruk] v] ase] (s)ase[+ni] ]
In (28a) the stem verb aruk ‘walk’ has been unaccusativized, which means that the verb merges its
Agent Itiroo in its internal argument position. Lacking an external argument, the verb constitutes
a vP. When the first causativization applies, the vP must be selected by -(s)ase[+o] ((16b)). The
obtained structure is a v*P as in (28b) since -(s)ase bears a Causer as its external argument. When
the second causativization applies, the v*P must be selected by -(s)ase[+ni]. When o-causativization
applies first, therefore, it must be followed by ni-causativization (o ni).
The above two are the only possible derivations. The suggested analysis thus explains why
only ni ni and o ni are legitimate combinations for double causativization.
4.1.3. Scrambling over Causee arguments
Let us then consider how the suggested analysis accounts for the asymmetry in scrambling
discussed in section 2.2. The relevant examples (8a, b) are repeated here as (29a, b), respectively.
(29) a.???Taroo-ga kooen-e1 kodomo-ni t1 ik-ase ta.
Taroo-NOM park-to child-DAT go CAUS PAST
16
b. Taroo-ga kooen-e1 kodomo-o t1 ik-ase ta.
-ACC
‘Taro made/let his child go to the park.’
In the deviant (29a) the PP (kooen-e ‘park-to’) is scrambled to the left of the ni-Causee whereas in
the legitimate (29b) the PP is scrambled to the left of the o-Causee. As shown in section 2.2,
Miyagawa (2007) accounts for this asymmetry under the ni-extra analysis. In (29a) the ni-Causee
is object of the matrix clause. The landing site for the scrambled phrase is therefore considered to
be a VP-adjoined position, which, according to Miyagawa, is not an appropriate landing site for
scrambling.
The suggested analysis explains this asymmetry in a similar way. (29a, b) are assigned the
structures (30a, b), respectively.
(30) a. ???Taroo-ga [v*P kooen-e1 kodomo-ni [VP t1 ik] v*]-ase[+ni] ta.
b. Taroo-ga [vP [VP kooen-e1 kodomo-o t1 ik] v]-ase[+o] ta.
In (30a) the ni-Causee occupies SPEC of v*. Hence the scrambled phrase is assumed to be raised
at least to a v*P-adjoined position. In (30b), the o-Causee occupies COMP of V. Hence the
scrambled phrase is assumed to be raised at least to a VP-adjoined position. The suggested
analysis thus attributes the difference in acceptability between (29a) and (29b) (or, (30a) and
(30b)) to the difference in adjunction sites: scrambled phrases can be adjoined to a VP, but not to a
v*P.
Notice that Miyagawa’s (2007) “VP” means a maximal projection of an extended verb. Hence
17
it is substantially equal to the “vP/v*P” in this paper. Hence the explanation given here is virtually
the same as Miyagawa’s: the scrambled phrase cannot adjoin to a maximal projection of an
extended verb. (v*P). (30a) is excluded since the PP adjoins to a v*P. On the other hand, (30b) is
acceptable since the PP adjoins to a VP, which is not considered a maximal projection of an
extended verb .6
4.2. Semantic properties
4.2.1. Semantic difference between ni-causatives and o-causatives
Let us now turn to the semantics of causatives. The familiar assumption is that the ni-causative is
permissive while the o-causative is coercive. However, this is not an adequate generalization. As
pointed out in section 2.4, we sense the permissive vs. coercive difference when and only when
we compare a minimal pair like (31a, b).
(31) a. Taroo-ga [Hanako-ni odor] -ase ta.
Taroo-NOM Hanako-DAT dance CAUS PAST
b. Taroo-ga [Hanako-o odor] -ase ta
-ACC
‘Taro made/let Hanako dance.’
When examined separately, the ni-causative can be coercive or permissive. The o-causative can
also be coercive or permissive. How are the two causatives different in semantics, then? Why do
we sense the permissive vs. coercive difference when we compare a minimal pair?
Under the suggested analysis, the semantic difference between the two causatives will be
explained as follows:
18
(32) a. In the ni-causative, the Causer is responsible for the Causee’s making up his/her MIND to
do so.
b. In the o-causative, the Causer is responsible for the Causee’s doing so.
Notice that persuading is involved in (32a), but not in (32b). It does not mean that the causation of
(32b) is coercive. It simply means that THE SENTENCE DOES NOT TELL US HOW THE CAUSER WORKS
ON THE CAUSEE. In the following I explain the semantics of each causative in turn.
Let us first consider the denotation of ni-causatives, taking (31a) for example. It is
uncontroversial that the ni-Causee (Hanako) is understood as the Agent of odor ‘dance’: she
volitionally initiates and controls the dancing event. Less uncontroversial is the following claim:
Hanako should be a Theme as well. Hanako can be considered a Theme because in the dancing
event she herself is affected. This idea has been implied by several researchers as we will see in
section 5.1. If this is correct, the denotation of an unergative eventuality will be like (33).
(33) Agent Theme denoted change
Hanako volition Hanako dancing
The Agent Hanako controls the Patient Hanako by her volition, as a result of which the Patient
Hanako dances. This is the denotation of the embedded v*P. The v*P is then selected by -(s)ase.
The sentence (31a) therefore denotes that (33) is caused by the Causer Taroo.
(34) Agent (=Causer) [ Agent (=Causee) Theme denoted change]
Taro Hanako volition Hanako dancing
Since Hanako is the Agent, i.e. a volitional participant, Taro must works on Hanako’s MIND so as
19
for her to have such volition. In this sense the causation must be persuading.
It should be noted that it is not specified in (34) how Taro persuades Hanako. He may ask,
allow or threaten her to dance. Nor is it specified whether Hanako is willing or reluctant when she
makes up her mind to dance. She may be willing to dance. Or she may decide to dance reluctantly
by being urged by Taro. We do not know any of them because they are not denoted in the sentence.
Such information is, if present, in the context. What is denoted in the ni-causative, again, is that
what the Causer does is responsible for the ni-Causee’s making up his/her mind to do the denoted
event.
Let us next consider the semantics of o-causatives, taking (31b) for example. The caused
eventuality would be like (33) if nothing happened to the unergative stem verb. In the o-causative,
however, the unergative stem verb has been unaccusativized, as a result of which the Agent is
generated as an internal argument. The application of unaccusativization must have effect on the
denotation. Namely, the Agent role of Hanako disappears from (33). The resulting denotation is
represented in (35).
(35) Theme denoted change
Hanako dancing
Only Hanako’s bodily change is denoted in (35). It does not mean that Hanako’s agentivity is
absent in the event. It means that it is absent in the DENOTATION. Put differently, by
unaccusativization the speaker describes the dancing event as “phenomenal,” interested only in
Hanako’s bodily change.
The unaccusativized verb constitutes a vP since it does not take an external argument. The vP
is then selected by -(s)ase, as a result of which it is denoted that (35) is caused by the Causer Taroo.
The causative relation is schematically illustrated below:
20
(36) Agent (=Causer) [ Theme (=Causee) denoted change]
Taro Hanako dancing
It shows that Hanako’s body is moved in such a way as we describe as “dancing,” and Taro is
responsible for the change. Again, it is not specified in (36) how Taro brings about the change. He
may ask or force Hanako to dance. Nor is it specified whether Hanako wants to dance or not. They
are not denoted in the sentence. What is denoted in the o-causative (31b) is that what Taro does is
responsible for Hanako’s change.
To recap, the causal relationships between the participants in the ni-causative (31a) and in the
o-causative (31b) are schematically illustrated as in (37a) and (37b), respectively.
(37) Agent (=Causer) [ Agent (=Causee) Theme denoted change]
a. Ni-causative Taro Hanako volition Hanako dancing
b. O-causative: Taro Hanako dancing
The difference between the ni- and the o-causative has to do with how THE SPEAKER describes the
event. When the speaker has some interest in the Causee’s making up his/her mind to do the
denoted action, the event is described by the ni-causative. When the speaker is more interested in
the Causee’s change, on the other hand, the event is described by the o-causative. Hanako’s
agentivity is absent in (37b). It does not mean that Hanako’s volition is absent in the event. It is
absent in the denotation. The speaker ignores it since it is not of his/her concern.
This much given, we can now answer the question of why we sense the permissive vs.
coercive difference when we compare a minimal pair like (31a, b)/(37a, b). When compared, the
presence of Hanako’s volition in the ni-causative stands out. It makes us feel that Hanako’s
21
volition is more respected in the ni-causative than in the o-causative. Put differently, by
comparison we may notice that Hanako’s volition is not mentioned in the o-causative. We may
then wonder why the sentence avoids reference to Hanako’s volition, and guess that Hanako’s will
is disregarded in the event.
To sum, it is THE SPEAKER who ignores the Causee’s will when s/he describes the situation by
an o-causative sentence. The subject (Causer) may or may not respect the Causee’s will. We do
not know which since the speaker does not tell us about it. When the speaker describes the
situation by a ni-causative, on the other hand, we know at least that the subject (Causer) works on
the mind of the Causee. But we do not know how the subject works on the Causee. It may be
permissive or coercive. Therefore, either causative can be permissive or coercive. When
comparing a minimal pair, however, we feel that it is THE SUBJECT (Causer) who ignores the
o-Causee’s will. It is because the speaker’s indifference to the Causee’s will is mixed up with the
subject’s indifference to the Causee’s will. This is how we sense the permissive vs. coercive
difference when we compare a minimal pair of o- and ni-causatives.
4.2.2. The Peer Constraint on Ni-Causees
The above explanation opens up a way to account for the following semantic constraint on
ni-Causees. Terada (1990) observes that the ni-causative is awkward unless the ni-Causee is a
person or a “personified” entity (p.235).
(38) Hanako-ga omotya-no robotto{-o/ (*)-ni} aruk-ase ta.
Hanako-NOM toy robot {-ACC/ -(*)DAT} walk CAUS PAST
‘Hanako made/let the toy robot walk.’
The ni-causative in (38) is awkward since the Causee (the toy robot) is not a person. Given a
22
context in which the robot is described as a personified entity, however, the ni-causative is
acceptable. According to Terada, the notion “personified” is to be assimilated to “volitional.”
Following the familiar permissive vs. coercive dichotomy, Terada argues that the ni-Causee should
want to do the denoted action. Therefore the ni-Causee must be an entity that has mind to want
something.
Her explanation, however, is not satisfactory. First, let us examine example (14) again, which
is reproduced as (39).
(39) (Moo uma-o turete kaeru zikan datta ga, amari yukaisoo-ni kakoi-no naka-o hasitte iru
node,)
(‘The time had come for Taro to take the horse back, but, because the horse was running so
joyously in the arena, ...’)
Taroo-wa sono mama [uma-o hasir]-ase ta.
Taroo-TOP without stopping horse-ACC run CAUS PAST
‘Taro let the horse run for a little while.’
The context obviously conveys that (the speaker assumes that) the horse has a volition to run. But
the situation cannot be described by ni-causative as in (40).
(40) ??Taroo-wa sono mama [uma-ni hasir]-ase ta.
Precisely, an unergative-stem ni-causative sounds odd unless the speaker regards the Causee as
an equal to him/her. Consider the following examples:
23
(41) a. Taroo-wa [inu{-o/ ??-ni} sukinadake hoe] -sase ta.
Taroo-TOP dog{-ACC/ ??-DAT} to.the.full bark CAUS PAST
‘Taro made/let the dog bark as much as it wanted.’
b. Taroo-wa [kodomo{-o/ ??-ni} zibun-no heya-de asob] –ase ta.
Taroo-TOP child{-ACC/ ??-DAT} self-GEN room-LOC play CAUS PAST
‘Taro made/let his child play in his room.’
The animal/child Causee is not compatible with ni-marking when it is called by the genre it
belongs to. When the same referent is mentioned by its name, however, ni-marking is totally
acceptable. Compare (41a, b) with (42a, b).
(42) a. Taroo-wa [inu-no Hanako{-o/ -ni} sukinadake hoe] -sase ta.
dog-GEN Hanako{-ACC/ -DAT}
‘Taro made/let the dog Hanako bark as much as she wanted.’
b. Taroo-wa [kodomo-no Hanako{-o/ -ni} zibun-no heya-de asob] -ase ta.
child-GEN Hanako{-ACC/ -DAT}
‘Taro made/let his child Hanako play in his room.’
By calling the Causee by her name the speaker may regard the Causee as an equal to him/her, e.g.
as a member of the family. The contrast in grammaticality between (41) and (42) suggests that the
problem does not have to do with the identity of the ni-Causee, but with the way how the speaker
regards the Causee. I call this THE PEER CONSTRAINT ON NI-CAUSEES.
(43) THE PEER CONSTRAINT ON NI-CAUSEES
The speaker must regard a ni-Causee as an equal to him/her.
24
In (41a, b) the speaker, who calls the Causee by its genre, does not regard the Causee as an equal
to him/her, but as someone/something under guardianship/control. Hence the Causees in (41a, b)
are not compatible with ni-marking.
It should be noted that the deviance of ni-marking has nothing to do with the Causee’s volition
or willingness. All the ni-Causees in (40) and (41) must enjoy doing the denoted action. But
ni-marking is not compatible with them. The previous analyses would not be able to account for
why ni-marking is deviant in these examples.
Under the suggested analysis, the above constraint is accounted for as follows. Recall that in
the ni-causative the speaker has interest in how the Causer works on the mind of the Causee, while
in the o-causative the speaker is interested in the phenomenal change on the Causee. If the speaker
does not regard the Causee as an equal to him/her, he/she may well care less about the Causee’s
mind. Calling a Causee by its genre reflects the speaker’s lack of fellow feeling with the Causee.
Consequently, such a speaker should describe the situation by o-causative. If the speaker mentions
a Causee by its name, on the other hand, s/he may have some interest in the Causee’s mind.
Therefore such a Causee, whatever its identity is, is compatible with ni-marking.
The suggested analysis thus gives an account for the semantic constraint on unergative-stem
ni-causatives. However, it leaves a problem unsolved: the Peer Constraint does not hold for
transitive-stem ni-causatives, as exemplified by the following sentences:
(44) a. Taroo-wa [inu-ni sinbun-o totte ko] -sase ta.
Taroo-TOP dog-DAT newspaper-ACC bring CAUS PAST
‘Taro made/let the dog bring him a newspaper.’
b. Taroo-wa [kodomo-ni osoku-made terebi-o mi] -sase ta.
Taroo-TOP child-DAT late-till TV-ACC watch CAUS PAST
25
‘Taro made/let his child watch TV till late.’
In either example the speaker refers to the Causee by its genre, which reflects his/her lack of
fellow feelings with the referent. The acceptability of ni-marking indicates that the Peer Constraint
is not relevant to transitive-stem ni-causatives. Before we answer why this is the case, we need to
make more precise the difference between the denotation of unergatives and that of transitives. In
the next section I get this work done first and then answer why the Peer Constraint does not hold
for transitive-stem causatives.
5. Unaccusativization
5.1. The argument structure of unergatives
Since Hale and Keyser (1993) it has been a familiar view that unergatives are “hidden” transitives
(cf. Chomsky 1995). There is considerable evidence to show that unergative verbs should have an
implicit object. To pick up some: Kayne (1993) and Watanabe (1996), based on auxiliary selection
patterns in Romance languages, claim that transitive and unergative verbs should be of the same
kind, i.e. transitives. Branigan and Collins (1993) observe that an unergative verb can optionally
move out of VP, and attribute it to Case-checking with a null object in Spec-AgrOP. Burzio (1986)
and Levin and Rappaport Hovav (1995) study resultative constructions and argue that unergative
verbs should be a potential accusative assigner.
Here arises a question. Why is the object “hidden” in an unergative sentence? Recall that I
have suggested in section 4.2.1 that in an unergative event, the Agent doubles as a Theme. When
Hanako dances, for example, Hanako is the Agent whose volition initiates and controls the event.
The same person must be a Theme too, for the Agent Hanako controls her own body to dance. In
this sense the unergative event, such as dancing, walking, crying and swimming, is INTRINSICALLY
REFLEXIVE (cf. Reinhart and Reuland 1993). I assume the following Lexical Conceptual Structure
26
(LCS) for unergative eventualities:
(45) LCS: [[x ACT] cause BECOME [x BE AT-(DANCING) STATE]]
I have already presented an informal illustration of the above LCS in section 4.2.1: (33) is repeated
here as (46).
(46) Agent Theme denoted change
Hanako volition Hanako dancing
In the LCS (45)/(46) the same variable is assigned to the Agent and the Theme, by which
unergative verbs necessarily denotes a reflexive eventuality.
It is a common assumption that an argument structure (AS) reflects the information in an
LCS. The AS of an intrinsically reflexive event like (45)/(46) should therefore be like (47).
(47) (x, <x> )
The item in angle brackets represents an internal argument, and the item outside an external
argument. Notice that the two arguments are represented by the same variable. The AS of an
unergative verb is thus intrinsically reflexive as well.
A syntactic structure is constructed so as to satisfy the selectional requirement of an AS.
However, as suggested by Reinhart and Reuland (1993), “languages differ in whether they allow,
require, or forbid the patient [Theme] role of their intrinsically reflexive verbs to be realized
syntactically” (p.666). In Germanic languages such as Norwegian and Danish the intrinsically
reflexive internal argument is realized as a special kind of anaphor (SE). In English, in contrast,
27
the intrinsically reflexive internal argument never surfaces. The (apparent) absence may be
interpreted in either of the following ways. First, the internal argument of an AS is not realized in
English syntax when it is intrinsically reflexive (cf. Reinhart and Reuland 1993). Or, the SE-type
anaphor is phonetically null (Ø) in English (cf. Bergeton 2004). I tentatively adopt the latter idea
but it does not affect the discussion below if we adopt the former one.
With this in mind, let us consider unergative verbs in Japanese. It can be assumed that the LCS
and the AS of unergative verbs are invariant cross-linguistically. Namely, unergative verbs are
intrinsically reflexive in Japanese as well. Then the apparent absence of object in Japanese
unergative sentences should be explained in the same way as in English unergative sentences: the
internal argument of an unergative verb is realized as Ø. The relation between the LCS, the AS
and the syntax of an unergative verb is illustrated below.
(48) a. LCS: [[x ACT] cause BECOME [x BE AT-(DANCING) STATE]]
b. AS: ( x, <x> )
c. Syntax: Hanako1-ga Ø1 odor- ta.
Hanako-NOM dance PAST
Given the above discussions, we can now answer the question of why unergatives are “hidden”
transitives: it is because their intrinsically reflexive objects are realized as Ø.
Let us then consider the causal relationship between the Agent and the Theme in an unergative
event. Apparently unergative and transitive eventualities differ minimally as to the identity of the
Theme: the Agent and the Theme must be the same entity in an unergative eventuality (as in
(49a)) but not in a transitive eventuality (as in (49b)).
28
(49) a. LCS (unergative): [[x ACT] cause BECOME [x BE AT-SOME STATE]]
b. LCS (transitive): [[x ACT] cause BECOME [y BE AT-SOME STATE]]
(It should be noted that the x and the y in (49b) can refer to the same entity. What makes the
difference is whether reflexivity is lexically designated or not.)
According to Levin and Rappaport Hovav (1995), the two types of eventualities are different
in the type of causation. They claim that an unergative verb denotes AN INTERNALLY CAUSED
EVENTUALITY:
With an intransitive verb describing an internally caused eventualities, some property
inherent to the argument of the verb is “responsible” for bringing about the eventuality. For
agentive verbs such as play and speak, this property is the will or volition of the agent who
performs the activity. (Levin and Rappaport Hovav 1995: 91)
When Hanako dances, for example, the causation is internal since (i) her volition moves her own
body to dance and (ii) her volition is the only force to move her own body as denoted.
Given the above explanation, a transitive eventuality can be called EXTERNALLY CAUSED. But
what does “external causation” mean? Does it mean the Agent’s volition is absent in a transitive
eventuality? Not likely: John must be volitional when he paints a wall, for example. It means that
the Agent’s volition is absent in the DENOTATION of a transitive eventuality. What is described by
“John painted the wall,” for example, is that John’s physical force affects the state of the wall. His
volition is not described in the denotation. We may well sense his volition, but it comes from the
context or from our common sense. In this sense a transitive verb denotes an externally caused
eventuality.
29
The difference in the type of causation is schematically illustrated below:
(50) Unergative (internally caused) eventuality:
Agent Theme denoted change
Hanako volition Hanako dancing
(51) Transitive (externally caused) eventuality:
Agent Theme denoted change
John the wall painted
The volition of the Agent is present in the denotation of the unergative eventuality, but not in the
transitive eventuality.
To recap, unergative verbs are different from transitive verbs in the following two respects:
(52) a. The internal argument is “hidden” (realized as Ø) because it is intrinsically reflexive.
b. The Agent’s volition is involved in the denotation.
With these assumptions in mind, let us then turn to the set-aside question: What is
unaccusativization?
5.2. Unaccusativization
I have explained in section 3 that in an unergative-stem o-causative sentence the stem verb has
been unaccusativized, which means that its Agent is generated in the internal argument position as
in (53a). Lacking the external argument, the unaccusativized verb constitutes a vP. When
causativization is applied, the vP is selected by –(s)ase[+o] as in (53b).
30
(53) a. [vP [VP Hanako odor] v]
b. Taroo-ga [vP [VP Hanako-o odor] v] -ase[+o] ta.
Taroo-NOM Hanako-ACC dance CAUS PAST
‘Taro made/let Hanako dance.’
The above explanation apparently runs against the assumption held firmly in the generative
literature: the Agent must be generated in the external argument position. However, they are
compatible if we take into consideration what has been discussed in section 5.1.
Recall that the LCS for an unergative verb is intrinsically reflexive, i.e. the Agent doubles as
the Theme. The Theme is “hidden” (realized as Ø) in syntax since it is intrinsically reflexive.
Recall also that in the discussion of the semantics of o-causatives in section 4.2.1 I have claimed
that the Agent of a stem verb is removed from the denotation. With these arguments given, it can
be made precise what unaccusativization does to the mapping from an AS to syntax:
(54) a. UNACCUSATIVIZATION: Suppress the Agent in the AS.
b. Mapping from an unergative AS to syntax:
(i) When no unaccusativization applies:
. odor ‘dance’: (x, <x>) [v*P Hanako1 [VP Ø1 odor] v*]
(ii) When unaccusativization applies:
odor ‘dance’: (x, <x>) [vP [VP Hanako odor] v]
When no unaccusativization is applied, an unergative verb takes an Agent argument in the external
argument position as in (54bi). The intrinsically reflexive internal argument is “hidden” in syntax.
(54bii), on the other hand, illustrates what happens when unaccusativization applies.
Unaccusativization suppresses the external argument of the AS. (The shade represents the
31
suppression.) As a result of suppression, the internal argument is no longer “intrinsically
reflexive.” That is, it has no antecedent in the AS. Consequently there is no reason the Theme is
“hidden” in syntax. The internal argument is realized as an overt DP. Since it is the internal
argument it occupies the internal argument position. Hence Hanako occupies COMP of V in
(54bii). (The unaccusativized vP needs to be licensed. In this case –(s)ase[+o] serves as a licensor.
See section 5.4 for more details.)
In conclusion, unaccusativization is the suppression of the external argument of the AS. When
unaccusativization applies to an unergative verb, the “hidden” internal argument must be realized
in syntax. This is why and how the Causee appears in COMP of V in unergative-stem o-causative
sentences.
5.3. The Peer Constraint revisited
With the above discussion of unergatives in mind, let us again consider the Peer Constraint
discussed in section 4.2.2. I have accounted for why non-peer Causees cannot be ni-marked in
unergative causatives (e.g. (41a) repeated here as (55)), but left unanswered why the constraint
does not hold for transitive causatives (e.g. (44a) repeated here as (56)).
(55) Taroo-wa [inu{-o/ ??-ni} sukinadake hoe] -sase ta.
Taroo-TOP dog{-ACC/ ??-DAT} to.the.full bark CAUS PAST
‘Taro made/let the dog bark as much as it wanted.’
(56) Taroo-wa [inu-ni sinbun-o totte ko] -sase ta.
Taroo-TOP dog-DAT newspaper-ACC bring CAUS PAST
‘Taro made/let the dog bring him a newspaper.’
Given the discussion in section 5.1, the causal relationship of (55) is informally illustrated as in
32
(57a). Since barking is an unergative (hence intrinsically reflexive) eventuality, the dog should be
the Agent and the Theme of the barking event. Notice that the Agent’s volition is denoted in (57a)
since it is essential in an internally caused eventuality. In (56), on the other hand, the caused
eventuality is directed externally, i.e. from a dog to a newspaper. In section 5.1 I have assumed that
the Agent’s volition is not involved in the denotation of an externally caused eventuality. Hence
the causal relations in (56) will be schematically illustrated as in (57b).
(57) Causer Agent (ni-Causee) Theme
a. (55): ?? Taro [ the dog volition the dog ]
b. (56): Taro [ the dog a newspaper ]
The volition of the Causee is involved in (57a). That is, the Causer (Taro) influences on the dog’s
MIND in such a way that the dog comes to have volition to bark. In this sense the speaker assumes
the presence of the dog’s mind. But it contradicts the way s/he refers to the Causee. If the speaker
cared for the presence of the Causee’s mind, s/he would refer to the Causee not by its genre but in
some other way. The contradiction makes the sentence sound unnatural.
In (57b), the relation between the Causee (the dog) and the Theme (the newspaper) is
phenomenal: the dog’s physical force to the newspaper. Taro is responsible only for the dog’s
doing so, not for the dog’s making up its mind to do so. The transitive-stem ni-causative sentence
does not mention the existence of the Causee’s mind. Hence it is compatible with the speaker’s
referring to the Causee by its genre, which shows his/her lack of fellow feeling with the Causee.
To sum, the Peer Constraint is not relevant in transitive ni-causatives because the Causee’s
mind/volition is not involved in their denotation. They only denote that the subject (Causer) is
responsible for the ni-Causee’s external action to the Theme argument. The speaker can therefore
refer to the ni-Causee in whatever way. Unergative ni-causatives, on the other hand, denote the
33
existence of the ni-Causee’s mind/volition. But if the speaker calls the ni-Causee by its genre, it
shows that s/he ignores the presence of the Causee’s mind. Consequently, ni-marking on the
Causee and referring to the Causee by its genre are contradictory to each other.
5.4. Unaccusativization and passivization
At first sight, unaccusativization (53) is identical to passivization. In fact, passivization consists of
(i) unaccusativization and (ii) licensing by passive morpheme -(r)are. Take naos ‘repair’ for
example. The verb bears the argument structure of (x, <y>)-type, taking two DISTINCT arguments.
If unaccusativized, the external argument is suppressed and only the internal argument is selected.
(58) a. naos: (x, <y>) [v*P Hanako [VP tokei naos] v*]
↓ Unaccusativization applied
b. naos: (x, <y>) [vP [VP tokei naos] v]
The unaccusativized vP needs to be licensed by some licensor. Passive morpheme -(r)are is one of
such licensors. A passive sentence is thus derived.
(59) [TP [vP [vP [VP tokei-ga naos] v] -are] ta]
watch-NOM repar PASS PAST
A note is in order. It is not a passive morpheme that triggers unaccusativization.
Unaccusativization applies freely, but needs licensing. In (59) the vP is licensed by a passive
morpheme. If this is correct, causative morpheme -(s)ase[+o] can also be regarded as a licensor:
34
(60) a. odor ‘dance’ (unaccusativized): (x, <x>) [vP [VP Hanako odor] v]
b. Licensing by -(s)ase[+o] [v*P Taroo [vP [VP Hanako-o odor] v] -ase]
c. O-causative derived [TP [v*P Taroo-ga [vP [VP Hanako-o odor] v] –ase] ta]
Passivization and o-causativization therefore share the process of unaccusativization. They only
differ as to what morpheme licenses the unaccusativized vP.
Under the suggested analysis, then, we can make the following two predictions. First, an
o-causative is derivable with a transitive verb as well. Second, a passive sentence is derivable with
an unergative verb as well. In the remainder of this section I will examine each prediction.
5.4.1. Prediction (I): Transitive o-causatives
The suggested analysis predicts that o-causativization is possible for transitive verbs as well. Take
naos ‘repair’ for example. If unaccusativization is applied to naos, its external argument is
suppressed, and the verb constitutes a vP:
(61) naos (x, <y>) [vP [VP tokei naos] v] (=(58b))
The vP needs licensing. If the vP is licensed by -(r)are, the passive sentence (59) is obtained.
Similarly, if the vP is licensed by -(s)ase[+o], an o-causative sentence should be obtained. The
suggested analysis thus predicts the fifth type of causative: transitive-stem o-causative as shown in
(62).
(62) [TP [v*P Taroo-ga [vP [VP tokei-o naos] v] -ase] ta]
Taroo-NOM watch-ACC repar CAUS PAST
‘Taro got the watch repaired.’
35
The sentence does not denote a person who Taro gets to repair the watch. It only denotes that the
resulting state (the watch repaired) was caused by Taro.
Apparently, the sentence (62) is hard to distinguish from (63), a ni-causative the Causee of
which is realized as a null pronoun (pro).
(63) [TP [v*P Taroo-ga [v*P pro [VP tokei-o naos] v*] –ase[+ni] ta.
Taroo-NOM watch-ACC repair CAUS PAST
‘Taro made/let someone repair the watch.’
The v*P-taking -(s)ase assigns -ni to the Causee, but the case is not realized when the Causee is
pro. Put differently, the suggested analysis predicts that the sentence (64a) can be construed either
as in (64b) or (64c). How can it be attested that there really are two possible structures?
(64) a. Taroo-ga tokei-o naos-ase ta.
Taroo-NOM watch-ACC repair CAUS PAST
b. Taroo-ga [vP [VP tokei-o naos] v] -ase[+o] ta [O-causative]
c. Taroo-ga [v*P pro [VP tokei-o naos] v*] -ase[+ni] ta [Ni-causative]
Before answering this question, I point out several properties of pro. Consider the following
examples:
(65) a. Maiasa pro sinbun-o yom -u.
every morning newspaper-ACC read PRES.
‘I/he/she/they read(s) a newspaper every day.’
36
b. Taroo-ga pro pro age ta.
Taroo-NOM give PAST.
‘Taro gave it to him/her/them.’
Pro in Japanese has several properties. First, pro can appear in any argument position. Second, it
never triggers morphological change on the verb. The content of pro is recovered from the context.
Third, and most importantly, pro does not have an arbitrary reading. (65a) cannot be interpreted as
“Someone reads a newspaper every morning,” nor can (65b) be interpreted as “Taro gave
something to someone.” What pro refers to must be specified in the preceding context. With this in
mind, let us return to the above question.
Suppose that Taro’s watch was broken and you want to know if Taro took any action for it. In
this context, neither (66a) nor (66b) is an appropriate answer.
(66) a. #Syuuri-no tame Taroo-wa tokei-o pro okur ta.
repair-GEN purpose Taroo-TOP watch-ACC send PAST
‘Taro sent the watch (to someone/to some shop) for repair.’
b. #Pro Taroo-no tame-ni tokei-o naos-i te kure ta.
Taroo-GEN sake-for watch-ACC repair give PAST
‘(Someone) generously repaired the watch for Taro.’
The deviance concerns the recoverability of pro. In (66a) okur ‘send’ should take a Goal, and in
(66b) naos ‘repair’ should take an Agent. However, the given context does not contain information
on a repair shop or a repairperson. Hence such arguments cannot be realized as pro. The oddness
of (66a, b) thus results.
In the same context, however, sentence (64a) makes an appropriate answer. Notice that the
37
context does not contain information on a repairperson. Hence the sentence (64a) cannot be
construed as in (64c), which contains a null Causee who repairs the watch. Under the suggested
analysis, (64a) can be construed as (64b) as well. It makes an appropriate answer since it does not
contain any non-recoverable pro.
5.4.2. Prediction (II): Unergative passives
Let us now consider the second prediction. The suggested analysis predicts, contrary to fact, that a
passive sentence is derivable with an unergative verb, as shown in (67).
(67) a. odor ‘dance’ (unaccusativized): (x, <x>) [vP [VP Hanako odor] v]
b. Licensing by -(r)are [vP [vP [VP Hanako odor] v] -are]
c. Passive derived *[TP [vP [vP [VP Hanako-ga odor] v] -are] ta]
‘(Lit.) Hanako was danced.’
I have assumed that unaccusativization is applied freely but needs licensing, and that -(s)ase[+o]
and -(r)are are both such licensors. Hence it would be predicted that the unaccusativized vP of an
unergative verb can be licensed by passive -(r)are, deriving a passive sentence like (67c). The
prediction is not borne out. But I do not assume that it undermines the suggested analysis. I
suspect that (67c) is OK in syntax, but excluded in semantics.
According to Shibatani (2006), passive voice is selected when the “discourse relevance” of the
agent is low relative to the patient.
(68) ACTIVE/PASSIVE OPPOSITION:
Does the action originate with an agent extremely low in the discourse relevance, or at least
lower relative to the patient?
38
Yes passive
No active (Shibatani 2006: 248)
When the agent is not relevant in the discourse, it is de-ranked to an optional element of a sentence,
and the more discourse-relevant patient is promoted to the subject of the sentence. This is how and
why a passive sentence is used. It must be evaluated somewhere in semantics whether the
selection of passive voice observes the rule like (68).
With this in mind, let us consider the passivization of unergative verbs. Recall that in the LCS
of an unergative verb the Agent and the Theme must refer to the same entity. Being the same entity,
it does not make sense to ask which argument is more relevant in the discourse. Therefore, passive
voice is unable to be selected for unergative verbs. The deviance of (67c) results from the violation
of the voice selection rule (68).7
6. Conclusion
In this paper I have provided a new analysis of causative constructions in Japanese. Causatives are
derived by either v*P-taking -(s)ase[+ni] or vP-taking -(s)ase[+o]. Transitive verbs and unergative
verbs constitute either v*P or vP. The latter structure is formed when the verb is unaccusativized,
i.e. when the verb’s external argument is suppressed. Hence either ni- or o-causative is derivable
with unergatives and transitives. Unaccusative verbs, on the other hand, necessarily constitute a vP.
Hence o-causative is obligatory for unaccusatives. The structure of each causative is illustrated as
follows:
39
Syntactic structures: Assigned interpretation:
(69) Transitive-stem causatives:
a. X-ga [v*P Y-ni [VP Z-o V] v*] -(s)ase[+ni] X Y Z
b. X-ga [vP [VP Z-o V] v] -(s)ase[+o] X Z
(70) Unergative-stem causatives:
a. X-ga [v*P Y1-ni [VP Ø1 V] v*] -(s)ase[+ni] X Y volition Y
b. X-ga [vP [VP Z-o V] v] -(s)ase[+o] X Z
(71) Unaccusative-stem causatives:
X-ga [vP [VP Z-o V] v] -(s)ase[+o] X Z
Syntactically, the ni-Causee is generated in SPEC of v* while the o-Causee is generated in COMP
of V. Semantically, only unergative ni-causatives (as in (70a)) denote the Causee’s volition. From
this a variety of syntactic and semantic properties of causative sentences follow straightforwardly.
In the course of discussion I have made two assumptions. First, unergative verbs have a
“hidden” internal argument which is internally reflexive. Second, unergative as well as transitive
verbs can undergo unaccusativization, i.e. the suppression of the external argument. Apparently
problematic as they may seem, I have shown that neither of them is an ad-hoc stipulation. Based
on these assumptions, I have shown that various syntactic/semantic properties of causative
constructions are given a simple account. Moreover, the suggested analysis correctly predicts the
existence of the fifth type of causative construction: transitive-stem o-causatives as shown in
(69b).
40
Notes
1 -Sase is associated with a verb stem that ends with a vowel, and -ase with a verb stem that ends
with a consonant.
2 The abbreviations used in this paper are as follows: ACC (accusative), CAUS (causative), CL
(classifier), COP (copula), DAT (dative), GEN (genitive), LOC (locative), NOM (nominative), PASS
(passive), PRES (present), TOP (topic).
3 It should be noted that there are two kinds of DOC proposed in the literature. An example of the
other DOC violation is given below:
(i) *Masao-ga ano uma-o ano mon-o toos i ta.
Masao-NOM that horse-ACC that gate-ACC pass PAST
‘(Intended.) Masao passed the horse through the gate.’ (Kuroda 1988: 23)
The other DOC, which I call the DOC2, is different from the DOC (11) in the following respects.
First, the DOC2 constrains the occurrence of o-marked object and o-marked ADJUNCT. Second, the
DOC2 violation is tolerable to some speakers. Shibatani (1978: 282) finds such an example in a
fiction:
(ii) Syoonen-wa ame-no naka-o saka-o nobori kiri, ...
boy-TOP rain-GEN inside-ACC slope-ACC climb up
‘The boy climbed up the hill in the rain, ...’ [my translation]
Even for speakers who would not accept (i) or (ii), the DOC2 violation is alleviated if the two
o-phrases are not adjacent.
(iii) ?Kono mon-o Taroo-wa muriyari kono kuruma-o t1 toos i ta.
this gate-ACC Taroo- TOP forcibly this cart- ACC pass PAST
‘Taro forcibly passed this cart through the gate.’ (Dubinsky 1994: 54)
Third, the DOC2 violation is not rescued by changing case morphology. Example (i) remains
deviant if one of the two o-phrases is changed to a ni-phrase:
41
(iv) *Masao-ga ano uma-ni ano mon-o toos i ta.
None of the above holds for the DOC-violating sentences, i.e. transitive o-causatives. First, no
native Japanese speakers would tolerate the transitive o-causative (3a). Second, the deviance of
(3a) is not alleviated if the two o-phrases are separated as in (v).
(v) *Hanako-o1 Taroo-ga t1 hon-o yom -ase ta.
Third, (according to Shibatani and Miyagawa,) the deviance of (3a) disappears by changing case
morphology as in (12). With this much given, the two DOCs should be considered distinct
constraints. The existence of DOC2, if real, does not lend support to the stipulation of the DOC.
4 One might suspect that the ni-Causee cannot license FNQs because it constitutes a PP headed by
-ni. This is not the case, as the following example shows:
(i) Asako-ga gakusei{-o/ -ni} 3-nin tyuugoku-e ik ase ta.
Asako-NOM students{-ACC/-DAT} 3-CL China-to go CAUS PAST
‘Asako made/let three of her students go to China.’ (Hasegawa 1999:142)
The ni-Causee does license the FNQ, which suggests that the ni-Causee is not a PP, but a DP.
Incidentally, Miyagawa (1999) provides an example very similar to Hasegawa’s but with the
opposite judgment.
(ii) Taroo-ga kodomo{-o/*-ni} 2-ri kooen-e ik ase ta.
Taroo- NOM kids{- ACC/*- DAT} 2-CL park-to go CAUS PAST
‘Taro made/let two kids go to the park.’ (Miyagawa 1999: 251)
Miyagawa assumes that the ni-Causee cannot license an FNQ even if they are adjacent.
Considering the fact that (i) and (ii) are virtually the same in syntax, I assume that the deviance of
ni-causatives in (ii) should have to do with semantics. The ni-causative in (ii) may be excluded for
the violation of the Peer Constraint. See section 4.2.2.
5 One might suspect that o-causativization cannot be applied when the stem verb contains two
arguments (which I call quasi-DOC). But this idea cannot be maintained. O-causativization is
42
indeed applicable to a two-place predicate if it does not contain any o-phrase. Suppose that
causativization is about to apply to (i).
(i) [Hanako uma-ni nor]
Hanako horse-DAT ride
Nor ‘ride’ takes two arguments. The internal argument is assigned dative -ni case. As the following
example shows, o-causativization is applicable to (i).
(ii) Taroo-ga [Hanako-o uma-ni nor]- ase ta.
Taroo-NOM -ACC CAUS PAST
‘Taro made/let Hanako see Jiro.’
The observation above argues against the existence of the quasi-DOC.
6 I provide the following example as a piece of evidence that a scrambled phrase can never adjoins
to a v*P:
(i) Aru gakusei-ga dono hito-mo1 Taroo-ni t1 syookaisi ta.
some student-NOM every person Taroo-DAT introduce PAST
‘Some student introduced everyone to Taro.’ (some > every; *every > some)
According to Hoji (1985), the indirect object precedes the direct object in the basic word order.
Hence the direct object (dono hito-mo ‘everyone’) in (i) should have been scrambled to the left of
the indirect object (Taroo-ni ‘to Taro’).
Suppose that the scrambled phrase may be adjoined either to a VP or to a v*P. In the former
case, (i) is construed as in (iia). In the latter case, the word order of (i) is obtained by subject
movement of the external argument as in (iib). (Subject movement optionally takes place in
Japanese (cf. Miyagawa 2001, 2007)).
(ii) a. [TP [v*P Subject [VP DO1 [VP IO t1 V] v*] T] (VP-adjoining scrambling)
b. [TP Subject2 [v*P DO1 [v*P t2 [VP IO t1 V] v*] T] (v*P-adjoining scrambling)
Here we make a prediction as to possible scope interpretations between the subject and the DO in
43
(i). If the DO adjoins to a VP as in (iia), the subject must take scope over the DO. If the DO
adjoins to a v*P, on the other hand, the DO can take either wide or narrow scope since it binds the
trace of the subject (t2). The fact is that the DO cannot take scope over the subject in (i). It follows
that that a scrambled phrase can adjoin to a VP, but not to a v*P when it undergoes local (v*P
phase-internal) scrambling.
7 The present discussion has no relevance to impersonal passives.
(i) Es wird getanzt. (German)
it is danced.
‘There is dancing.’
In (i), both external and internal arguments have been suppressed. Hence it is not subject to the
rule (68). Impersonal passives should be derived by another kind of suppressing operation.
44
References
Bergeton, Uffe. 2004. The independence of binding and intensification. Doctoral dissertation,
University of Southern California.
Branigan, Philip and Collins, Chris. 1993. Verb movement and the quotative constructions in
English. MIT Working Papers in Linguistics 18: 1-14.
Burzio, Luigi. 1986. Italian syntax: A Government-Binding approach. Dordrecht: Reidel.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Dubinsky, Stanley. 1994. Predicate union and the syntax of Japanese causatives. Journal of
Linguistics 30: 43-79.
Hale, Ken and Keyser, Samuel J. 1993. On argument structure and the lexical expression of
syntactic relations. In: Hale, Ken and Keyser, Samuel J. (eds.), The view from building 20,
53-109. Cambridge, MA: MIT Press.
Harada, Shin-Ichi. 1973. Counter Equi NP Deletion. Annual Bulletin of the Research Institute of
Logopedics and Phoniatrics 7: 113-147.
Harley, Heidi. 1995. Subject, events, and licensing. Doctoral dissertation, MIT.
Hasegawa, Nobuko. 1999. Seisei nihongogaku nyuumon [An introduction to generative Japanese
linguistics]. Tokyo: Taishukan.
Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese. Doctoral
dissertation, University of Washington.
Inoue, Kazuko; Harada, Kazuko; and Abe, Yasuaki. 1999. Seisei gengogaku nyuumon [An
introduction to generative linguistics]. Tokyo: Taishukan.
Kayne, Richard S. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica 47:
3-31.
Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge, MA: MIT Press.
Kuroda, S.-Y. 1965. Generative grammatical studies in the Japanese language. Doctoral
45
dissertation, MIT.
Kuroda, S.-Y. 1988. Whether we agree or not: A comparative syntax of English and Japanese.
Lingvisticae Investigationes 12: 1-47.
Levin, Beth and Rappaport Hovav, Malka. 1995. Unaccusativity: At the syntax-lexical interface.
Cambridge, MA: MIT Press.
Miyagawa, Shigeru. 1989. Syntax and semantics 22: Structure and Case marking in Japanese. San
Diego: Academic Press.
Miyagawa, Shigeru. 1999. Causatives. In: Tsujimura, Natsuko (ed.), The handbook of Japanese
linguistics, 236-2698. Malden, MA: Blackwell.
Miyagawa, Shigeru. 2001. The EPP, scrambling, and wh-in-situ. In: Kenstowicz, Michael (ed.),
Ken Hale: A life in language, 293-338. Cambridge, MA: MIT Press.
Miyagawa, Shigeru. 2007. Unifying agreement and agreement-less languages. MIT Working
Papers in Linguistics 54: 47-66.
Reinhart, Tanya and Reuland, Eric. 1993. Reflexivity. Linguistic Inquiry 24: 657-720.
Shibatani, Masayoshi. 1973. Semantics of Japanese causativization. Foundations of Language 9:
327-373.
Shibatani, Masayoshi. 1976. Causativization. In: Shibatani, Masayoshi (ed.), Syntax and
semantics 5: Japanese generative grammar, 239-294. New York: Academic Press.
Shibatani, Masayoshi. 1978. Nihongo-no bunseki – Seisei bumpoo-no hoohoo [The analysis of
Japanese language: A generative approach]. Tokyo: Taishukan.
Shibatani, Masayoshi. 2006. On the conceptual framework for voice phenomena. Linguistics 44:
217-269.
Terada, Michiko. 1990. Incorporation and argument structure in Japanese. Doctoral dissertation,
University of Massachusetts.
Teramura, Hideo. 1982. Nihongo-no syntax-to imi (I) [Syntax and semantics of the Japanese
46
language (I)]. Tokyo: Kurosio Publishers.
Tonoike, Shigeo. 1978. On the causative constructions in Japanese. In: Hinds, John and Howard,
Irwin (eds.), Problems in Japanese syntax and semantics, 3-29. Tokyo: Kaitakusha.
Tsujimura, Natsuko. 1996. An introduction to Japanese linguistics. Malden, MA: Blackwell.
Watanabe, Akira. 1996. Case absorption and Wh-agreement. Dordrecht: Kluwer.
Watanabe, Akira. 2006. Functional projections of nominals in Japanese: Syntax of classifiers.
Natural Language and Linguistic Theory 24: 241-306.