1

On the causative construction

*
Heidi Harley, University of Arizona
1 Affixal causatives and architectures
Japanese was the first language with morphological affixation of causative morphemes to
receive serious attention from generative grammarians (perhaps because it is the first
language of several important early generative theoreticians, e.g. Saito, Kuroda, and
Kuno). The typological differences between Japanese and English represented the first
exploration of the Universal Grammar hypothesis, which predicted the existence of
important similarities in the grammatical structure of languages from unrelated language
families. Consequently, its impact on generative linguistic theories in general and on
Principles-and-Parameters approaches specifically has been very significant.
The Japanese causative represents in a very pure form the problem of the
morphology/syntax interface. Consequently, the causative construction is one of the most
theoretically significant aspects of Japanese grammar, its three subtypes having attracted
more attention and inspired more theoretical proposals than almost any other
construction. Analyses of the causative have had a major influence on many foundational
aspects of syntactic theory, including control, case marking, clause structure, theta-theory
and argument structure, and the morphology-syntax interface.
All of these issues have received extensive treatment in the literature, and this
chapter will touch on many of them. We will, however, focus on the importance for
linguistic theory of a single problem posed by the construction: In what component of the
grammar are the various causatives constructed? What are the available morphological
possibilities a theory can explore in accounting for the three types of causative? It turns
2
out that the answers to many of the other syntactic questions posed by causatives depend
on the theoretical choices made in answering this one. In fact, it is not unreasonable to
say that the entire architecture of a given linguistic theory can be deduced from the
answer given to this one question. The best analysis, we presume, will provide a
theoretically satisfying, cross-linguistically consistent, and most importantly unified
treatment of the causative morpheme -(s)ase. Ill endeavor to show that such an analysis
exists, and that it demands a certain type of theoretical framework; indeed, such a unified
analysis could not exist in a framework configured differently in any significant way.
1.1 The Empirical Base
To create a causative expression in Japanese, the bisyllabic morpheme -(s)ase is attached
to what would be the embedded verb in an equivalent English causative construction, as
illustrated in the example in (1):
(1) Taroo-ga Hanako-o ik-ase-ta
Taro-N Hanako-A go-ase-PST
Taro made Hanako go.
The Causer (here, Taroo) is the nominative-marked subject of the whole sentence. The
logical subject of the root verb, referred to below as the Causee, is marked with
accusative or dative case (here, Hanako, using the accusative variant).
All V+sase combinations exhibit similar morphophonological properties,
indicating the indivisible nature of the single phonological word constructed by -sase
affixation. These are listed in (2) below:
(2) Properties of all -sase- causatives (many from Manning, Sag & Iida 1999)
a. V+sase is a single phonological word for stress, other word-size processes
(Kitagawa 1986, 1994)
3
b. -sase subject to phonological allomorphy depending last segment of V
(if its a vowel, then -sase-, if its a consonant, then -ase (Kuroda 1965a))
c. V+sase may feed (productive) nominalization with -kata, way of
d. -sase- is a bound morpheme; by itself it may not behave as a lexical verb (stem):
1
i. it may not reduplicated by itself to express repetition
ii. it may not bear focus intonation by itself
iii. it may not be inflected for subject honorification by itself.
v. it may not stand alone as an answer to a yes-no question
Despite their morphophonological similarity, however, certain subtypes of
V+sase combinations may be distinguished. The literature has, over time, identified two
main classes of V-(s)ase sequences in Japanese, the lexical (unproductive) causative (3)
and the syntactic (productive) causative (4). These two V+(s)ase combinations have
been shown to have distinct syntactic and semantic properites, although they are
morphophonologically very similar. Within the class of syntactic causatives, two further
subtypes have been identified, the make, -o-causative((4)a), and the let, -ni-causative
((4)b).
(3) (A subset of) Lexical causatives Miyagawa 1980, 1984
Taroo-ga zisyoku-o niow-ase-ta Jacobsen 1981, 1992
Taro-N resignation-A smell-ase-PST Matsumoto 2000
Taro hinted at resigation. (Lit: Taro made resignation smell.)
(4) Productive causatives
a. Make-causatives Kuroda 1965a, b; Kuno 1973
Hanako-wa Yoshi-o ik-ase-ta
Hanako-T Yoshi-A go-ase-past
Hanako made Yoshi go.
b. Let-causatives
Hanako-wa Yoshi-ni ik-ase-ta
Hanako-T Yoshi-D go-ase-past
Hanako allowed Yoshi to go/Hanako had Yoshi go.
The key problem of the causative construction has to do with a conflict between its
morphophonological status and its semantic status, which leads to significant problems in
its syntactic analysis. As noted above, the V+sase combination, together with any other
4
verbal suffixes that are attached to it, constitute a single phonological word. On the
assumption that phonological words are (syntactically simplex) terminal nodesthe
leaves of syntactic treesthe derived V+sase verb should head a single syntactic verb
phrase, and clauses containing such a verb phrase should behave in all respects like a
monoclausal construction.
For the lexical causatives, this does not lead to any serious difficulties. Lexical
causatives are monoclausal with respect to all relevant syntactic tests (see discussion
below). Further, they can undergo semantic drift, acquiring idiomatic readings in
combination with particular argument NPs (as illustrated by the examples in (3) above,
and discussed by Miyagawa 1980, 1984 and Zenno, 1985). Speakers have a sense that
these V+sase combinations are listed and non-productive. They feed non-productive
nominalization processes which can then independently undergo semantic drift (Volpe
2005). The arguments of a lexical causative are case-marked like the arguments of a
single clauseonly a single nominative case is possible, assigned to the Causer subject.
Finally, many lexical causative verbs (in fact, most such verbs) are formed with some
lexical causative morpheme other than -sase; choice of the causative allomorph is a
listed, arbitrary property for a given lexical causative verb root (Jacobsen 1981). In short,
lexical causatives behave syntactically, semantically, and morphophonologically like
single words: single verbs which head a single verb phrase.
Productive causatives, on the other hand, exhibit a number of biclausal properties,
most obviously, semantically: A productive V+sase combination refers to an event in
which an external Causer, X, acts to induce someone else, a Causee, to bring another
event or situation about, as described by the V. The best translation equivalent of a
5
Japanese productive causative in English involves embedding a clause headed by a bare
infinitive verb under a causative matrix verb (usually make but sometimes let or have).
Besides that intuitive biclausality, however, productive causatives like those in (4) exhibit
several other bicausal properties, listed in (5-8) below.
(5) Scopally, VP-modifying adverbials can be interpreted as modifying the caused
event or the causing event. Similarly, quantifiers on the object of the root
verb can take scope over just the caused event, or both the causing and caused
event (see Shibatani 1990:314).
(6) Subject-control -te adjuncts can be controlled either by the Causer subject or by the
Causee that is, by the subject of the embedded verb (see, among others
Terada 1991, Dubinsky 1994).
(7) The subject-oriented anaphor zibun can be anteceded either by the Causee or the
Causee, again, suggesting that the subject of the embedded verb is a true
subject (see Oshima 1979:433).
(8) Two separate events can be conjoined using the disjunct -ka, or, underneath a
single causative morpheme (see Kuroda 2003:455).
All of these properties together with full productivity and compositionality suggest
a syntactic combination of the V and the sase morpheme, and biclausality.
Productive causatives do exhibit several features that are typical of single clauses,
however. Besides being a single morphophonological word, a productive causative clause
is clearly a single case-marking domain, licensing only a single nominative argument.
Further, they obey the Double-o constraint (Harada 1973): causatives of intransitive Vs
may show accusative case on the Causee argument, as illustrated in (1) above, but
causatives of transitive Vs, requiring an accusative case for the object of the V, force the
Causee to receive the dative -ni marker, since within a single clause only a single
accusative argument is possible.
2,3
Similarly, a productive causative is only a single tense
domain. No independent tense marking is possible to distinguish the time of the caused
event from the time of the causing: the single tense morpheme on the end of the complex
6
verb must cover both. Finally, productive causative clauses behave as a single domain for
clause-mate negative polarity item licensing.
The main distinguishing properties of these two types of causatives are
summarized below:
(9) a. Lexical causative:
monoclausal by all tests (see below)
can have idiomatic interpretations
exhibit allomorphy with other lexical causative affixes
strong speaker sense of listedness, non-productivity
may feed (non-productive) nominalization
b. Productive causative:
Biclausal by tests involving scope, adverbial control, binding, disjunction
Monoclausal by tests involving negative polarity, tense
(Make-causative) monoclausal by tests involving case.
Causee must be animate/Agentive
Productive
There is also an interesting acquisition difference between lexical -sase and
syntactic -sase- (Murasugi et al. 2004): lexical -sase- appears first in the speech of
children, before productive sase (but not as early as zero-derived lexical causative uses
of verbs show up).
Within the set of causatives classified as productive, two subtypes have been
identified (Kuroda 1965a,b, Kuno 1973), where a difference in reading affects the case-
marking possibilities on the Causee of an intransitive verb. When the causative has a
make readingforcible or direct causationthe case-marker on the Causee of an
intransitive verb is accusative, as noted above ((4)a). When it has a reading more similar
to let permission, or indirect causationthe Causee receives dative -ni, even if the
verb is intransitive and the double-o constraint is not in effect ((4)b). Although this
distinction has received considerable attention in the literature, I will not discuss it here.
7
For extensive discussion of the make/let distinction, see Dubinsky 1994 and Miyagawa
1999 and citations therein.
Examples illustrating each individual property described above are not provided
here for space reasons, and because similar summaries have been provided in multiple
other publications elsewhere. For useful summaries exemplifying most of these
properties, see Kitagawa 1986, 1994 and Manning, Sag and Iida 1999. For surveys of
many previous analyses, see Cipollone 2001 and Kuroda 2003.
1.2 Theoretical approaches
This constellation of properties really force one to face ones theoretical priorities. The
productive V+-sase forms pose serious architectural issues, even without considering the
lexical causatives. How should a theoretical framework be configured to allow it to
accommodate a construction which appears to be headed by a single morphological verb
and is monoclausal with respect to case, tense, and NPI licensing, but appears to be
biclausal with respect to binding, scope, control and disjunction? Resolving these issues
usually involves radical replumbing of grammatical architectures. Consequently, the
influence of Japanese causatives on linguistic theory couldnt be bigger.
4
1.2.1 Lexicalist treatments of V+sase: HPSG
Lexicalist frameworks take it as axiomatic that single morphophonological words
correspond to terminal nodes in the syntax. It follows that productive causatives must be
treated as syntactically monoclausal: only one morphophonological verb, therefore only
one clause.
8
Consequently, the apparent multiclausal properties of causative constructions
must arise from the (productive) operation affixing the causative morpheme in the
lexicon, producing a complex syntactic and semantic word. That is, it then follows that
binding relations, adverbial scope, quantifier scope, and adverbial control are phenomena
that depend on lexical operations, not syntactic structurethese phenomena are not
properly syntactic phenomena at all. This position is thoroughly presented in the
proposal of Manning, Sag and Iida 1999, treating causatives within the Head-driven
Phrase Structure Grammar framework. There, the key replumbing of the architecture is
the inclusion of adjunction and quantifier scope as lexical operations. Syntactic
constituency is no longer at issue in treating these phenomena in HPSG.
The most serious challenge to this approach to causatives, within the terms of
HPSG, comes from the availability of disjunction of two VPs under a single causative
morpheme, as discussed by Kuroda 2003:455.
5
Kurodas examples showing disjunction
of VPs under -sase are given in (10) below.
(10) a. Hanako-ga [[Masao-ni uti-o soozisuru]-ka [heya-dai-o haraw]]-aseru koto ni sita
Hanako-N [[Masao-D house-A clean]-OR [room-rent-A pay]]-sase that D do
Hanako decided to make Masao clean the house or pay room rent.
Reading: sase scopes over OR; Masao has a choice.
b. Hanako-ga [[Masao-ni uti-o soozis-aseru]-ka [heya-dai-o haraw-aseru]] koto ni sita
H.-N M.-D house-A clean-sase-OR room-rent-A pay-sase that D do
"Hanako decided to make Masao clean the house or she decided to make him pay
room rent"
Reading: OR scopes over sase; Masao wont have a choice.
The availability of disjunction of the verb phrase without the -sase affix is a significant
challenge to the lexicalist treatment of -sase, since in the phrase-structure grammars
employed for the syntactic component by these framewoks, disjunction is treated
syntactically, not lexically.
6
Treating the adjunction of adverbs as lexical, as well, raises
9
issues concerning how to capture syntactic adjunct/argument asymmetries within HPSG,
as discussed by Cipollone 2001; if both argument structure-altering operations and
adjunction operations are lexically implemented, it is not clear how their different
behaviors with respect to extraction etc. may be captured.
1.2.2 Principles and Parameters: Logical Form from Syntax
In Principles and Parameters-type approaches, on the other hand, a different set of
priorities are in force. The ultimate syntactic representation of a clause is taken to be
(isomorphic to) its Logical Form. Semantic properties such as scope assignment of
quantifiers must therefore be syntactically represented. Further, the notion of subject is
famously a configurational one in P&P; consequently the assignment of antecedents for
subject-oriented reflexives or of controller for adjoined -te phrases must be (at least
partially) syntactically determined. The consequence of these assumptions, then, is that
the causative morpheme and the verb to which it is affixed must each head a separate
syntactic projection, creating different constituents which can independently be used to
construct scopal or subject properties. In order to account for the ways in which
causatives have syntactically monoclausal properties, then, P&P frameworks propose that
the embedded clausal structure is deficient in some waynot a full CP or TP clause, but
some reduced yet argumentally complete clause is embedded by the causative morpheme.
The absent intermediate projections account for the monoclausal behavior in the relevant
domains.
The inescapable conclusion given this set of priorities is that morphological and
phonological words are not in a one-to-one relationship with syntactic terminal nodes.
The biggest problems to be faced by the P&P approach, then, are the following: Where
10
are words made, before or after syntax, or both? What is the constituent structure of the
embedded phrase?
There have been many proposals in the literature within these broad lines. They
are outlined below in roughly chronological order, followed by their key analytical
property and the locus of word-formation in the model:
(11) a. Predicate Raising (e.g. Kuno 1973): Biclausal D-structure collapses to
monoclausal S-structure; Syntax feeds word-formation.
b. Parallel monoclausal and biclausal trees. Word-formation feeds syntax (e.g.
Miyagawa 1984).
c. LF-excorporation and projection. Word-formation feeds syntax, which then
deconstructs complex words to project (covert) biclausal structure (Kitagawa
1986, 1994) (This proposal could be understood as a variant of Chomskys 1993
lexicalist checking-theory.)
d. Incorporation (Baker 1988). Syntax manipulates morphemes, feeds word-
formation.
Bakers Incorporation account became the most familiar P&P analysis, and still
represents the core idea behind most current approaches in the literature. It itself was an
updated version of Kunos Predicate Raising approach; the updating involved
understanding how the collapse of the biclausal structure is only apparent: the V which
heads the lower VP simply head-moves to adjoin to the -sase morpheme, which is a V in
its own right, projecting the matrix VP. On this approach, the input to the syntax is not
morphophonological words, but rather individual morphemes. Productive morphology is
affixed to its host by syntactic operations such as head-movement. Because head-
movement leaves a trace, there is no collapse of the lower clause when this happens;
rather, the entire structure remains present and interpreted at LFit is merely
unpronounced. An illustration of this account is provided below in (12).
11
(12) Derivation of Hanako-ga Taroo-ni pizza-o tabe-sase-ta in a Baker-style
Incorporation accont:
7
Input to the syntax: {Hanako
N
, Taroo
N
, pizza
N
, -ga
K
, -ni
K
, -o
K
, tabe
V
, -sase
V
, -ta
I
}
IP
KP
i
I
NP K VP I
Hanako ga t
i
V ta
VP V
KP V sase
NP K KP V
Taroo ni NP K tabe
pizza o
Several ingredients are needed to make an Incorporation account of productive
causatives work, each of which has major theoretical consequences for the rest of the
framework.
Since it is ungrammatical to attach a separate tense or complementizer morpheme
to the verb stem before affixing -sase, the proposal is that the constituent embedded
under -sase is VP, rather than IP (TP) or CP. This accounts for the absence of separate
tense domains for the two clauses. The theory can then explain the availability of NPIs in
the embedded VP by assuming that the clause-mate condition on Japanese NPI licensing
is sensitive to the TP domain, not the VP domain: because there is only one TP in a
productive causative, NPIs in the embedded VP will meet the clausemate requirement.
Domain for case-marking,
negative polarity licensing.
Only one IP, hence only one
such domain
Domain for subject-oriented
reflexive binding, condition
B, adverbial control,
quantifier scope. Two VPs,
hence two such domains.
Note VP-internal subjects.
12
The VP-internal subject hypothesis is then also necessary, so that the embedded
subject argument can be introduced in the lower VP, allowing for the presence of the
Causee in the structure without an embedded TP.
It must also be the case that clausal conditions on case assignment like the
Double-o Constraint are sensitive to the TP domain, rather than VP. To capture this, a
theory of abstract Case checking is needed in which clausal Case domains are bounded
by a TP projectiona Dependent Case case theory, of the Marantz 1991 type; see, e.g.
Miyagawa 1999). Such an account predicts that the transitivity of the embedded VP can
affect the case assigned in the whole clause.
Similarly, it is necessary to have in place a theory of scope that allows quantifiers
to scope at the VP level as well as the CP level, in order to account for lower-clause
quantifier scope.
Finally, and most importantly, the approach entails a partial rejection of the
Lexicalist Hypothesis: the account only works if the syntax manipulates bound
morphemes, as well as free ones.
8
In other words, productive inflectional and derivational
affixes must be considered to be input to the syntax. Rather than being presyntactically
attached to their host stems in the lexicon, such affixes are attached to their hosts either in
or following the syntactic component.
What of the lexical causatives? Recall that they are irregular, stem-specific,
semantically idiosyncratic, and non-productive. Nonproductive affixes are not input to
the syntax in this approach; they come pre-attached to their stems in a presyntactic
morphological component (the locus of irregularity). This explains a) their
nonproductivity, since syntax is understood as the domain of productivity, and b) the
13
monoclausal behavior of lexical causatives; one V in the numeration, one VP in the
derivation.
The end result is a type of hybrid account, where productive causatives are
combined with their verbs in the syntax, but lexical causatives are treated in a separate,
pre-syntactic part of the grammar.
9
The remainder of this paper will be taken up with
laying out whats wrong with this picture, and what the implications for linguistic theory
are.
2 Lexical causatives
Like many languages, Japanese is rich in semantically related
inchoative/causative pairs of verbs, with overt causativizing (and/or inchoativizing)
morphology attached to a common root. These pairs have been extensively documented
by Jacobsen 1992; the first two examples of each class of pairs he identifies are given in
the table in (13). (None of these pairs involve -sase).
Class/#
10
! Intr Tr Rough ! gloss
(13) I: e/ hag hag-e-ru hag--u peel off
30 pairs hirak hirak-e-ru hirak--u open
11
II: /e ak ak--u ak-e-ru open
44 pairs hikkom hikkom--u hikkom-e-ru draw back
III: ar/e ag ag-ar-u ag-e-ru rise
71 pairs aratam aratam-ar-u aratam-e-ru improve
IV: ar/ hasam hasam-ar-u hasam-u catch between
8 pairs husag husag-ar-u husag-u obstruct (clog, jam?)
V: r/s ama ama-r-u ama-s-u remain
27 pairs hita hita-r-u hita-s-u soak
VI: re/s arawa arawa-re-ru arawa-s-u show (up)
18 pairs hana hana-re-ru hana-s-u separate from
VII: ri/s ka ka-ri-ru ka-s-u borrow/(lend)
2 pairs ta ta-ri-ru ta-s-u suffice/(supplement)
14
VIII: /as hekom hekom--u hekom-as-u dent
38 pairs her her--u her-as-u decrease
IX: e/as bak bak-e-ru bak-as-u turn into/bewitch
45 pairs bar bar-e-ru bar-as-u come/bring to light
X: i/as ak ak-i-ru ak-as-u tire
8 pairs dek dek-i-ru dek-as-u come/bring into
existence
XI: i/os horob horob-i-ru horob-os-u (fall to) ruin
6 pairs ok ok-i-ru ok-os-u get up
XII: /se abi abi-ru abi-se-ru pour over (self/other)
6 pairs ki ki-ru kise-ru put on (self/other)
XIII: e/akas obi obi-e-ru obi-(y)akas-u take fright/frighten
4 pairs hagur hagur-e-ru hagur-akas-u stray/evade
XIV: or/e kom kom-or-u kom-e-ru be fully present/fill
2 pairs nukum nukum-or-u nukum-e-ru warm
XV: are/e sut sut-are-ru sut-e-ru fall into disuse/discard
3 pairs wak wak-are-ru wak-e-ru divide
XVI: Misc nigiwa nigiwa--u nigiwa-s-u (make) prosper
25 pairs nob nob-i-ru nob-e-ru extend
2.1 Syntactic and semantic properties of non-sase lexical causatives
The causative member of these pairs has one more argument than its intransitive
counterpart, and bears a roughly causative reading with respect to it (sometimes one or
the other member of the pair having undergone some semantic drift), but shows no
obvious symptoms of a multiclausal syntactic structure. For example, compare the
available controllers for a -te- phrase in a syntactic vs. a lexical causative:
(14) Basic intransitive verb and its syntactic causative:
a. Hanako-wa arui-te it-ta
Hanako-Top walk-te go-PST
Hanako, walking, went.
b. Taroo-wa arui-te Hanako-o ik-ase-ta
Taroo-Top walk-te Hanako-A walk-sase-PST
Readings: "Taro made Hanako go, walking"
Taro, walking, made Hanako go
15
In the syntactic causative in (14)b, the phrase arui-te, walking, can be controlled either
by Hanako, the Causee subject of the embedded verb (who is the controller in the
noncausative sentence in (14)a), or by Taroo, the Causer subject of the causative -sase.
An identical pair of sentences is given for a lexical causative formed with the suffix -as-
in (15) below.
(15) Inchoative intransitive and its lexical causative:
a. Hanako-wa nure-te hi-e-ta
Hanako-T wet-te cool-inch-PST
Hanako (s body), getting wet, cooled.
b. Taroo-wa nure-te Hanako-o hi-(y)as-ita
Taro-T wet-te Hanako-A cool-caus-PST
Reading: Taroo, getting wet, cooled Hanako.
Impossible: Taroo cooled Hanako, (Hanako) getting wet.
Even though the notion of someone becoming cool by getting wet is semantically
sensible (as shown by the inchoative (15)a), the only available controller of the -te phrase
nure-te, getting wet, in (15)b, is the Causer, Taroo, rather than the Causee who is
becoming cool, Hanako. Lexical causatives, like underived transitive verbs, are
monoclausal with respect to this and all the other tests for biclausality listed above.
As shown by Miyagawa (1980, 1984, 1989, 1994, 1998) and Zenno (1985),
lexical causatives share another property with underived transitive verbs: they may
appear as part of an idiom. Sometimes their inchoative counterpart also participates (i.e.
the idiom alternates), as in (17), sometimes not, as in (16). (The examples below are from
Miyagawa 1989:126-127; they are given as V+object only, not in sentential uses.)
(16) Lexical causatives in idioms by themselves:
a. kama-o kake- (intr. kak-ar does not participate in this idiom)
sickle-A splash on
trick into confessing
16
b. zibara-o kir- (intr. kire not in this idiom)
my.stomach-A cut
pay out of ones own pocket
c. tenoura-o kaes- (intr. kaer not in this idiom)
palm-A return
change ones attitude suddenly
(17) Lexical causatives in alternating idioms:
a. te-ga kuwawar- te-o kuwae-
hand-N join hand-A add
be altered alter
b. hone-ga ore- hone-o or-
bone-N break
intr
bone-A break
tr
require hard work exert oneself
c. mune-ga itam- mune-o itame-
heart-N ache heart-A hurt
be worried worry (oneself)
Another test, developed by Oerhle and Nishio (1981), showed that lexical
causatives can participate in adversity readings, like simple transitive verbs and unlike
productive causatives (example in (18) taken from Miyagawa 1989:130).
(18) a. Simple transitive with adversity reading:
Taroo-ga ie-o yai-ta.
Taro-N house-A burn-PST
Taro burned his house.
Taros house burned, and he was adversely affected (he didnt cause it.)
b. Lexical causative with adversity reading:
Boku-wa kodomo-o gake kara ot-os-ita
I-T child-A cliff from drop-caus-PST
I dropped the child from the cliff.
The child dropped from the cliff, and I was adversely affected.
2.2 V+sase: The same properties as lexical causatives? or not?
The examples above applied these tests to unambiguously lexical causatives,
formed with causative affixes other than -sase. As noted in the first section, Miyagawa
argues that some V+sase combinations behave like the other lexical causatives above.
12
17
They participate in idioms, sometimes with ((19)c-d) and sometimes without ((19)a-b)
their intransitive counterpart:
(19) Lexical V+sase causatives in idioms:
a. tikara-o aw-ase-
power together-sase-
pull together
b. mimi-o sum-ase-
ear-A clear-sase
listen carefully
c. hana-ga saku- hana-o sak-ase-
flower-N bloom flower-A bloom-sase
be done heatedly engage in heatedly
d. hara-ga her- hara-o her-ase-
stomach-N lessen stomach-A lessen-sase
get hungry fast/wait for a meal
Some such V+sase forms also allow adversity causative interpretations:
(20) V+sase forms in adversity causatives (examples from Miyagawa 1989:129).
13
a. Taroo-ga yasai-o kusar-ase-ta
Taroo-N vegetable-A rot-sase-PST
Taroo spoiled the vegetables.
The vegetables rotted, and Taro was adversely affected.
b. Taroo-ga kaisya-o toosans-ase-ta
Taro-N company-A bankrupt-sase-PST
Taro bankrupted the company.
The company went bankrupt, and Taro was adversely affected.
But most V+sase combinations do not exhibit these propertiesmost V+sase
combinations are productive, not lexical. For instance, there is no adversity causative
interpretation available for the V+sase forms below (Miyagawa 1989:130):
14
(21) a. Boku-wa kodomo-o gake kara oti-sase-ta
I-T child-A cliff from drop-sase-PST
I caused the child to drop from the cliff.
Impossible: The child dropped from the cliff, and I was adversely
affected.
18
b. Kotosi-wa dekinai gakusei-o hue-sase-ta
This.year-T poor students-o increase-sase-PST
This year, we caused (the number of) poor students to increase.
Impossible: This year, the number of poor students increased, and we
were adversely affected.
c. Taroo-wa niku-o koge-sase-ta
Taro-T meat-A scorch-sase-PST
Taro caused the meat to scorch Pylkkanen 2002
Impossible: The meat scorched, and Taro was adversely affected.
Similarly, given an intransitive verb that participates in an idiom, like the examples in
(19)c-d above, a V+sase combination formed on the intransitive is not guaranteed to also
participate in the idiom (Miyagawa 1989:126):
(22) a. kiai-ga hair- *kiai-o hair-ase-
spirit-N enter spirit-A enter-sase
be full of spirit *inspire/put spirit into
b. hakusya-ga kakar- *hakusya-o kakar-ase-
spur-N splash.on spur-A splash.on.sase
spur on
intr
spur on
tr

These verbs have lexical causative forms with nonsase causative affixes, ir-e-ru and
kak-e-ru (they are members of Jacobsens class III alternators). The difference between
the verbs in (22), which do not allow an idiomatic interpretation with -sase, and the verbs
in (19), which do allow such an interpretation, is that the verbs in (19) have no other
lexical causative form. This is Miyagawas central observationthe only verbs which
show lexical causative behavior with -sase are the verbs which have no other
idiosyncratic lexical causative suffix of their own. In other words, lexical behavior of -
sase is only possible in cases where it is not blocked by a more specific causative suffx.
19
2.3 The blocking effect
The hybrid P&P account outlined in section 1.2.2 above simply divided V+sase
combinations into productive, regular, compositional forms (created in the syntax), and
nonproductive, noncompositional forms (listed in the lexicon). This captures the
distinctions between the two types of forms enumerated above, but it does not predict that
there should be any systematic relationship within the lexicon between lexical V+sase
and the other lexical causative forms, nor that there should be any systematic relationship
between lexical V+sase and syntactic V+sase. If there is any such systematic
relationship, then the hybrid account is seriously flawed.
In fact, such a systematic relationship does exist. As noted above, Miyagawa
(1980 et seq) and Zenno (1985) show that there is a simple way to predict when a V+sase
combination can behave like other lexical causatives and when it may only behave as an
productive causative, with no noncompositional interpretation and no adversity causative:
Only intransitive roots with no other transitive form can behave lexically with -sase.
That is, lexical interpretations of -sase are possible only if the root to which it is
attached does not have a transitive form derived in another way.
2.4 Miyagawas 1984 treatment: Paradigmmatic Structure
The sensitivity of lexical V+sase to the (non)availability of another derived form
with the same meaning is a classic example of morphological blocking, seen in both
derivational and inflectional morphology cross-linguistically. A simple case is the
English past tense. Some verbs do not have a past tense formed with -ed: *runned,
*writed, *feeled, *hitted. The reason is that they have an independently formed, irregular
past tense, which blocks the regular form: ran, wrote, felt, hit.
15
20
The same phenomenon is argued to occur in derivational morphology. Many
English adjectives have a negative form in un-, but some do not: *unpossible,
*unconsiderate, *uncoherent. These are blocked by the irregular negative forms,
impossible, inconsiderate, incoherent.
The grammatical mechanism that is responsible for blocking effects, in many
theories of morphology (for instance, Paradigm-Function Morphology, as recently
discussed in, e.g., Stump 2001), is the n-dimensional grammatical space of a paradigm.
For English verbs, for example, blocking is captured in the following way. Every verbal
word-form is understood to be attached to a paradigm space, defined by the inflectional
features of English verbs: past & present participle, 1, 2, 3, sg, pl. Some verbs come with
their paradigm space partially filled in, lexically as it were for instance, in the past
tense space for write, the form wrote is already enteredbut empty slots, such as for the
progressive participle, are filled in by a default affix for that slot: write+ing.
16
(23) Paradigm in the lexicon for write
V: WRITE write
infinitive
present ppl
past ppl written
Before lexical items are sent off to the syntax, empty paradigm spaces are filled in
by default morphology (bolded in the tables below).
(24) Paradigm in the lexicon for write
V: WRITE write
infinitive write
present ppl writing
past ppl written
To apply such an analysis to derivational morphology, one has to allow
derivational features to define a paradigm space, such as [negative] for the
21
impossible/*unpossible pairs, or [nominal] for their nominalizations. Words with
special negative or nominal forms will have already filled in their relevant paradigm
slots, blocking the productive insertion of the default form un- in the negative slot, or the
default form -ness in the nominalization slot, illustrated in (25):
(25)
A: POSSIBLE possible
negative impossible
nominal possibility
A: LIKELY likely
negative unlikely
nominal likelihood
A: HAPPY happy
negative unhappy
nominal happiness
Miyagawa (1980, 1984, 1989) treated the blocking effect in Japanese causatives
with such a derivational paradigmatic structure, defined by a feature [transitive];
without it, the blocking effect couldnt be captured. In terms of its position and function
in the model, Miyagawas level of Paradigmatic Structure is the same level of structure
that paradigm-function morphologists work with, although Miyagawa used it
independently of that framework.
He proposed a paradigm space defined by intransitive, transitive, and ditransitive
features. For many verb stems, an irregular form already occupied the transitive or
ditransitive slot in the paradigm. Only if an irregular form did not occupy that slot could
a default -sase form be constructed fill up the gap.
(26)
V: AG agar rise
Intr agar-
Tr age-
Ditr
22
(27)
V: AG sak- bloom
Intr sak-
Tr sak-ase
Ditr
To account for the systematicity of the lexical V-sase forms, then, Miyagawa
proposed to adopt an extra layer of lexical structure. However, his theory went beyond
the lexical causatives, including the syntactic causatives as well.
Miyagawa argued that it cannot be a coincidence that these default V-sase
combinations are morphophonologically indistinguishable from productive causatives.
That is, according to their morphophonological properties, a lexical causative formed
with -sase is exactly the same as a productive causative formed with -sase. He reasoned
that syntactic causatives are spelled out as -sase because -sase is just the elsewhere,
default form for a causative meaning: the lexical causative suffix -sase and the productive
causative suffix -sase are the same suffix. If lexical causatives had nothing to do with
syntactic causatives, there would be no reason for the same morpheme to be involved in
both.
Consequently, Miyagawa 1984 concluded that syntactic causatives had to be
created in the lexicon, in the paradigmatic structure, as well. However, all the questions
discussed above then arose for his analysis, concerning how to capture the biclausal
properties of the productive causatives within a lexicalist approach, leading to his
proposal that causatives are associated with parallel monoclausal and biclausal structures.
The theory became ever more complex.
23
2.5 Theoretical options
We now are in a position to summarize the state of affairs systematically. V+sase
combinations can be lexical or productive. If productive, they behave biclausally with
respect to binding, control, scope and idiom interpretation. If lexical, they behave
monoclausally. The lexical V+sase combination is in complementary distribution with
the other lexical causative morphemes discussed by Jacobsen, such as V+e, V+s, V+os,
etc, with -sase acting as the default suffix for lexical causative formation when no other
form exists. The default lexical -sase is morphophonologically identical to the productive
-sase.
Three possible courses of analysis seem open at this point:
A: Treat the lexical and syntactic causatives completely separately. On this
approach, the V+sase lexical causatives would be relegated to the lexicon with
the rest of the lexical causatives. The morphological identity between the
default lexical causative morpheme and the syntactic causative morpheme
would be irrelevant. That is: Jacobsen just missed class XVI: /-sase.
B: Unify the lexical and syntactic causatives by treating them both in the lexicon.
On this approach, something other than in the lexicon has to distinguish the
syntactic and lexical causatives.
C: Unify the lexical and syntactic causatives by treating them both in the syntax.
On this approach, a theory of post-syntactic morphology would be needed.
Again something other than in the syntax has to distinguish the two types.
Enter Distributed Morphology, Hale and Keysers v, and Minimalism.
3 Late Insertion, the Elsewhere condition, vPs and phases
17
In this section, we will see how independently motivated theoretical proposals in
distinct domains of research can naturally provide a unified account of lexical and
24
syntactic Japanese causatives. First the distinct proposals are introduced and explained,
and then we will see how they fit together.
3.1 Distributed Morphology, Late Insertion and the Elsewhere Principle
In Bakers Incorporation account and later work inspired by it, the syntax
manipulates and combines the lexical entry of complete morphemes, fully specified for
phonological, morphological, syntactic and semantic properties. In this section, we will
see how adopting a Late Insertion approach, according to which phonological
information is only inserted to realize syntactic terminal nodes later in the derivation,
allows the capture of paradigmatic blocking effects without the use of paradigms.
In Distributed Morphology (Halle and Marantz 1993 et seq.), the syntax
manipulates and combines abstract feature bundles, selected by the grammar of the
individual language learner from an inventory provided by UG on the basis of positive
evidence. These feature bundles are the input to, and terminal nodes of, a syntactic
derivation.
After the syntax has completed its derivation, (via the Agree, Merge, and Copy,
operations, as per Minimalist theory) and Spell-Out is reached, the syntactic structure,
with (possibly slightly changed) feature bundles in its terminal nodes, are sent off to
PF/LF for interpretation.
An early step on the PF-side is Lexical Insertion, at which the abstract bundles are
given phonological clothing prepatory to pronunciation. Vocabulary Items (VIs)
phonological strings identified as expressing certain featurescompete to realize the
terminal nodes that the syntactic derivation has made available. At each terminal node,
there may be many VIs whose feature specification is compatible with the feature content
25
of the terminal node. The VI with the most compatible features (but no incompatible
ones) realizes that node it wins the competition and blocks the other compatible VIs
from occupying the node. When no other VI is available, the default VI is insertedthe
elsewhere VI. A system that chooses a morpheme based on feature specification this
way is said to obey the Elsewhere Principle.
Here is a (syntactically very simplified) example derivation. Imagine an initial
Numeration consisting of feature bundles such as those listed in (28):
(28) { [
D
+1,+pl, +NOM], [
T
+past, +NOM], [
D
+pl, +ACC], [
V
KEEP, +ACC] }
The syntax merges and moves these feature bundles to create a syntactic tree, in
which all the necessary feature-checking has been accomplished. After the syntax is done
with it, the (simplified) tree in (29) is handed off to Spell-Out. The Vocabulary Items I,
we, it, kep-, keep, -ed, and them are all compatible with the available positions, but only
the most highly specified VI at each slot succeeds in actually realizing the terminal node.
The competition is illustrated at the bottom of the tree.
26
(29) TP
D
i
T
+1
+pl T VP
+NOM [+past]
[+NOM] D
i
V
V D
KEEP +pl
+ACC +ACC
_
we -ed kep them
I keep it
it
+ Adjacency:
18
We kep -ed them
+ morphophonology We kep-t em
The theoretical attraction of such an approach is that it allows a natural account of mirror
principle effects, provides a straightforward relationship between syntax and
morphology, and most importantly calls for only a single generative engine it requires
no generative mechanisms in the lexicon. That is, there is no need for a separate level of
paradigmatic structure to generate inflected and/or derived word forms, or to capture the
blocking effect. The blocking effect is captured by the Elsewhere principle the process
of competition of compatible VIs. The default, elsewhere VI will only win the
competition if no more specific VI is compatible with that slot.
3.2 (Modified) Hale and Keyser (1993, 2002)-type vPs for causative/inchoative
alternations
In the Distributed Morphology conception of blocking, the VIs in a given
competition must be competing to realize a slot which corresponds to a terminal node in a
Winning VIs
Competing but losing
VIseligible for
insertion but not the
most highly specified
Spell-out slots for
terminal nodes
27
syntactic tree. With respect to the lexical causative/inchoative pairs, which behave
syntactically like monomorphemic simplex verbs, there had been no previous suggestion
in the literature that their syntactic representation should be any more complex than
simply V. However, given DM assumptions, if the causative morphemes in lexical
causatives are competing with each other, there must be a syntactic terminal node within
the root+suffix complex for the suffix alonethat is, the syntactic representation of the
lexical causative verb must involve one verbal projection for the root, and a separate
projection for the suffix.
Hale and Keyser (1993, 2002) independently proposed that all transitive
verbseven morphologically simplex onesare made up of two separate heads: The
main V introduces the internal arguments of the verb and projects to VP, and the
external argument is introduced in the specifier of a v, which takes the VP as its
complement. In a slight revision to their account, Harley 1995 and Marantz 1997
proposed that a v was also present in inchoative constructions, but that it was a distinct
v which selected no external argument.
19
The lower root ! will head-move to attach to
its c-commanding v head, creating a syntactically complex head-adjunction structure
with two terminal nodes. The two resulting structures for inchoative (unaccusative
intransitive) verbs and causative, agentive, transitive verbs are given in (30) below.
(30) a. Unaccusative verbs b. Causative verbs.
vP vP
v !P DP
Agent
v
BECOME
DP ! John v !P
CAUS
the door open DP !
the door open
28
The relevance to the problem of where in the syntax to locate the
inchoative/causative suffixal morphology of Japanese, documented so extensively by
Jacobsen, is clear. That morphology is a realization of the two types of v head illustrated
above.
3.3 Late insertion and lexical causatives
The treatment of the blocking phenomenon in lexical causatives suggested by this
set of assumptions should now be clear. In a derivation which will contain a lexical
causative,
20
all the various causative morphemes compete to realize the v
CAUS
head in the
syntactic tree.
21
Depending on the class membership of the causative root, one particular
causative morpheme will winthe one specified for co-occurrence with roots of that
particular class. If no class is specified for a given rootthe elsewhere casethen the
default -sase morpheme will step in to fill the gap. The list of morphemes competing to
realize v
CAUS
is given in (31); for completeness the list of morphemes competing in the
inchoative case to realize v
BECOME
is given in (32).
(31) Morphemes competing to realize v
CAUS
in Japanese
-- ! CAUS / [ !
I+IV
___
v
] (38 Jacobsen roots on the list for --)
-e- ! CAUS / [ !
II+III+XIV+XV
____
v
] (120 roots on list)
-s- ! CAUS / [ !
V+VI+VII
____
v
] (47 roots on list)
-as- ! CAUS / [ !
VII+IX+X
____
v
] (91 roots on list)
-os- ! CAUS / [ !
XI
____
v
] (6 roots on list)
-se- ! CAUS / [ !
XII
____
v
] (6 roots on list)
-akas- ! CAUS / [ !
XIII
____
v
] (4 roots on list)
-sase- ! CAUS / Elsewhere (no roots on list) Blocking effect!
(32) Morphemes competing to realize v
BECOME
in Japanese:
-e- ! BECOME / [ !
I+IX+XII
___
v
] (79 Jacobsen roots on the list)
-ar- ! BECOME / [ !
III+IV
___
v
] (79 roots on list)
-r- ! BECOME / [ !
V
___
v
] (27 roots on list)
-re- ! BECOME / [ !
VI
___
v
] (18 roots on list)
-ri- ! BECOME / [ !
VII
___
v
] (2 roots on list)
-i- ! BECOME / [ !
X+XI
___
v
] (14 roots on list)
29
-or- ! BECOME / [ !
XIV
___
v
] (2 roots on list)
-are- ! BECOME / [ !
XV
___
v
] (3 roots on list) (Elsewhere? See n. 22)
-- ! BECOME / [ !
II+VII+XII
____
v
] (88 roots on list) (Elsewhere?)
So it is possible to treat the lexical causative as subject to syntactic decomposition
and thus capture the blocking effect. How does this help with the productive causative?
And how are the other distinctions between the two causatives to be captured, in this all-
syntax approach?
3.4 Implications for syntactic causatives
If -sase- is simply an Elsewhere form of the Agent-introducing v
CAUS
, and if all syntactic
causatives are realized with -sase-, then syntactic causatives are also a realization of the
Agent-introducing v
CAUS
. The syntactic version of this v
CAUS
, however, does not take a !P
headed by a verb root as its complement, but rather an argument-structurally complete
complementin fact, its complement is another vP, with its own independent agent
argument. This is illustrated for a productive causative of the simple transitive verb tabe-,
eat in (33). Like all agentive transitive verbs in Hale and Keysers approach, tabe- is
itself a realization of a root plus an external-argument introducing v head. In the case of
tabe-, we assume that the v which introduces its external argument is realized by a null
morpheme. (In a syntactic causative of a lexically causative verb, that lower v slot would
be filled by whatever causative morpheme was appropriate to the lexical causative root,
of course, as in, e.g. kow-as-ase, [[break-CAUS]
vP1
-CAUS]
vP2
, cause (someone) to break
(something).)
22
30
(33) vP2
DP v
Taroo vP1 v
DP v sase
Hanako !P v
DP !
pizza tabe
(Taroo-ga Hanako-ni pizza-o tabe-sase-ta)
Taro-N Hanako-D pizza-A eat-caus-past
Taro made Hanako eat pizza.
In a syntactic causative, the matrix CAUS v will not meet the structural
description for any special root-conditioned allomorphs of CAUS. The matrix CAUS will
be insulated from the root by one layer of bracketingit is separated from the root by the
embedded v.
23
Consequently, in productive causatives, the prediction is always that the
default realization of the v
CAUS
morpheme will win the competitionproductive
causatives will always be spelled out by the VI -sase-.
(34) (matrix v after head-to-head movement): [ [ !
TABE
____
v
]

_____
v
]
We are now in a position to propose clear definitional criteria that will distinguish lexical
and productive V+sase combinations in this framework:
(35) a. lexical causative: a CAUS v that is immediately adjacent to a root.
b. productive causative: a CAUS v that is not adjacent to a root (i.e. one which
embeds a vP).
Compare the lexical and syntactic causative structures in (36) below:
31
(36) a. vP b. vP
DP v DP v
Taro-ga Taro-ga
!P v vP v
-s -ase
DP ! DP v
tenoura-o kae Hanako-ni
!P v

DP !
hansai-o tutae
a. Taro-ga tenoura-o kae-s b. Taroo-wa Hanako-ni hanasi-o tutae-sase-ta
Taro-N palm-A return-CAUS Taro-T Hanako-D story-A convey-CAUS-PST
Taro changed his attitude suddenly "Taro made Hanako convey a story"
(=(16)c above)
To distinguish between the syntactic properties of lexical and productive causatives, then,
it suffices to identify vP as the locus of the relevant syntactic properties that suggest a
biclausal approach. It has long been assumed that vP is a locus for successive-cyclic A-
bar movement, and hence a possible target constituent for quantifier scope. Since vP
introduces the external argument, it is natural to associate subject-oriented binding
preferences with vP, as well as subject control into adverbials, perhaps supplemented
with a c-command restriction. And finally, since vP is the modern equivalent of the
former simplex VP projection, it is natural to think of VP adverbials having two loci for
scope in productive causatives but only one in lexical causatives. In short, by ascribing
these properties to a particular functional projection, we are able to appeal to the same
type of explanation for their absence in the lexical causative as we appealed to to explain
the single-clause effects on case-assignment and NPI licensing in the productive
causative. The culprit is the absence of two instances of the relevant syntactic projection
in each case TP in the case of case-assignment and NPI licensing in productive
32
causatives, vP in case of subject control, adverbial modification, quantification and
binding in lexical causatives.
24
3.5 Why not a v
BECOME
layer in lexical causatives?
Miyagawa 1994, 1998, proposes that there is also an inchoative v embedded under the
causative v of a lexical causative, adopting a structure like that given in (37) below:
(37) vP (Miyagawa 1994, 1998)
DP
Agent
v
John v vP
CAUS
v !P
BECOME
DP !
the door open
In order to capture the fact that inchoative morphology disappears in the lexical
causative member of the causative/inchoative pair for the vast majority of cases,
Miyagawa proposes that the lexical causative morphemes realize a complex segment, the
v
CAUS
+v
BECOME
heads together. In order to accomplish this, the v
CAUS
and v
BECOME
morphemes must fuse into a single terminal node prior to insertion. In one case he
discusses, the lexical causative meaning bother, iya-gar-sase, it appears as if inchoative
morphology -gar- is indeed embedded under the lexical causative morpheme -sase;
Miyagawa assumes that fusion must then have failed in this one case, justifying the
approach.
On the present approach, where the lexical causative v and the inchoative v are
interchangeable, rather than simultaneously present, we would have to assume that -gar is
not the spellout of the inchoative v head, but some other morpheme. This is necessary
33
given the logic of the analysis above. If lexical causatives embedded an inchoative v
rather than a bare !P, it would become impossible to distinguish between syntactic
causatives of inchoatives, and lexical causatives. Compare the structures, under the
inchoative-inside-lexical-causatives hypothesis, for the following two sentences, from
Miyagawa 1989:130, ex. 43a/b. The availability of the adversity reading for (38)a, as
well as the irregular causativizer -os-, indicates that ot-os- is a lexical causative; the
absence of the adversity reading in (38)b, along with the default causativizer -sase-,
indicates that ot-i-sase is a productive causative.
(38) a. Boku-wa kodomo-o gake kara ot-os-ita
I-T child-A cliff-from drop-CAUS-PST
I dropped the child from the cliff."
The child dropped from the cliff, and I was adversely affected." Lexical
b. Boku-wa kodomo-o gake-kara ot-i-sase-ta
I-T child-A cliff from drop-BECOME-CAUS-PST
I caused the child to drop from the cliff.
#The child dropped from the cliff, and I was adversely affected Productive
On Miyagawas 1994/1998 structures, where the lexical causative embeds the inchoative,
these two sentences would be represented as in (39):
(39) a. vP b. vP
DP v DP v
Boku-wa Boku-wa
vP v
CAUS
vP v
CAUS
-sase
!P v
BECOME
-os- !P v
BECOME
-i-
DP ! DP !
kodomo-o kodomo-o
PP ! PP !
gake kara ot- gake kara ot-
If the lexical causative ot-os includes a v
BECOME
in its structure, then the only
difference between the lexical causative and the productive causative is whether or not
34
Fusion (a post-syntactic operation) has applied to the v
BECOME
and v
CAUS
roots to ensure
that they are spelled out by the single -os- morpheme. This type of post-syntactic
operation cannot account for the syntactic distinctions observed above between lexical
and productive causatives, in terms of adverbial scope, control possibilities, availability
of adversative readings, etc. The lexical/productive distinction must be more categorical
than a mere postsyntactic morphological diacritic, since it has such strong consequences
for meaning. The distinction must be represented at LF. Consequently, it is preferable to
treat the lexical causative as directly embedding the !P, hence lacking the intervening
vP
BECOME
.
The notion that -gar in iya-gar-sase, bother, is not the inchoative but some other
type of morpheme is supported by the fact that it seems to appear in psychological
predicates only, such as kuyasi-garu, dumb-gar, lit. regret, samu-garu, cold-gar, lit.
feel cold, atsu-garu, hot-gar, lit. feel hot, hoshi-garu, want-gar, lit. feel like and
kowa-garu, fear-gar, lit fear (Yosuke Sato, p.c.). Indeed, Miyagawa 1989:157 notes
that -gar- affixation only appears with these adjectives when they have a non-first person
subjectin the first person, the -gar- affix is not needed. This suggests perhaps some
connection of -gar- to evidentiality, rather than to the BECOME v predicate, since
presumably the event/argument structure of experiencing these emotions is identical no
matter what the person of the subject is.
3.6 Correlations with other proposals in the literature
One significant distinction between the lexical causative and syntactic causative
that we have not discussed much above is the possibility of idiomatization in the former,
and the impossibility of it in the latter. In fact, this pattern fits well with independent
35
proposals in the literature by Kratzer 1996 and Marantz 1997, according to which the
agent-selecting vP is the boundary of a domain for special meaning specificationno
projection outside a vP can participate in an idiomatic specification that depends on the
root. This proposal was introduced to account for a pattern of facts first observed in
Marantz 1984: idiomatic meanings for verbs are often conditioned by the object selected
for by a transitive verb (kill the afternoon/kill the bottle/kill an audience), but are
seldom or never conditioned by the subject associated with a transitive verb. Assuming
that agent-introducting v is an interpretive boundary of this type can account for those
facts, and also predicts the pattern of facts about idiomatization observed here: lexical
causatives, with nothing intervening between the causative head and the root, often
idiomatize, while productive causatives, with a vP between the causative head and the
root, never do.
Another proposal has to do with the conditioning of allomorphy by the root only.
So far, nothing we have said would prevent a particular root in combination with a
particular v from conditioning causative allomorphy for a productive causative which
took that vP as its complement. However, that does not seem to occur: productive
causatives are always realized with the default causative morpheme -sase.
Arad 2002 has claimed, based on evidence from Hebrew, that the domain for
special meanings, i.e. the vP, is also a boundary domain for allomorphic conditioning of
this type: roots can only condition specific allomorphs of morphemes which are
syntactically directly adjacent to them. If this is true, it also predicts that Japanese
productive causatives could never be allomorphically conditioned by the roots with
which they occur.
25
36
Both the proposal concerning vP as boundary for idiomatic semantic
interpretation and the proposal taking vP to be a boundary for allomorphic conditioning
correlate well with Chomsky (2001)s claim that vP is a phase edgea boundary in the
syntactic derivation at which the v complement is sent for interpretation to LF and PF,
and which is subsequently impenetrable to later syntactic operations. Since in lexical
causatives the complement to v consists of a ! and its selected object, this constituent
corresponds exactly to the V+O sequence which Marantzs generalization claims is the
locus of idiomatization. The notion of interpretation by phase, then, can provide a
theoretical basis for the claim that !+O may idomatize while Agent+! (without O) may
not.
3.7 The beginning of the High/Low Attachment Analysis
This was one of the first proposals in a very fruitful line of inquiry which I will
call high/low attachment analyses. Since then, many analyses have appealed to the idea
that attachment of a morpheme to a higher functional projection results in regular
morphology and compositional meaning, while attachment of the same morpheme to a
lower projection (often the !), results in some allomorphy and potential meaning drift.
Other early examples of such an analysis are the approach to English of-ing and acc-ing
gerunds presented in Kratzer 1996, and the approach to Chichewa statives and passives
sketched in Marantz 1997. Since then, such approaches have been extremely fruitful in
looking at all kinds of morphology on the derivational/inflectional,
unproductive/productive cusp, in many unrelated languages: Travis 2000 used such an
approach to treat Malagasy lexical and syntactic causatives, in very much the same spirit
as outlined here for Japanese. Embick 2004 adopts the idea to treat the distinction
37
between stative, resultative, and passive participle formation in English. Fortin 2004
applies it to Minnangkabu causatives. Jackson 2005 shows it applies to statives and
resultatives in Pima. Alexiadou and Anagnostopoulou 2005 use it to treat adjectival
participles in Greek. Svenonius 2005 continues the trend, discussing high/low treatments
for causatives in several languages
4 Conclusion
I hope to have shown above that Japanese causatives have been extremely influential in
shaping developments in syntactic theory, indeed, in many syntactic theories. Japanese
causativeseven omitting the lexical oneseither force one to do more syntax in the
lexicon (Manning, Sag & Iida), or more morphology in the syntax (Baker).
It was argued that a careful examination of lexical causatives forces one to figure out a
way to unify traditional idiosyncratic, irregular word-formation with regular,
compositional syntax, and yet maintain a principled distinction between the two. A post-
syntactic morphologya late insertion approachwith recursive vPs, allows a simple,
unified treatment of all three types of lexical causatives, with a principled understanding
of the nature of the distinction between lexical and syntactic causatives.
The analysis could in general be taken as additional evidence for the phasal status of vP,
and for successive-cyclic QR through vP, among other things.
38
References
Alexiadou, A. and E. Anagnostopoulou. 2005. On the syntax and morphology of Greek
participles. Talk presented at the Workshop on the Morphosyntax of Modern
Greek, LSA Institute, July 2005.
Arad, Maya. 2003. Locality Constraints on the Interpretation of Roots: the Case of
Hebrew Denominal Verbs. Natural Language and Linguistic Theory 21, p. 737-
778
Baker, M. 1988. Incorporation. Chicago, IL: University of Chicago Press.
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In Kenneth Hale and
Samuel Jay Keyser, eds., The View from Building 20: Essays in Linguistics in
Honor of Sylvain Bromberger, 1-52. Cambridge, Mass.: The MIT Press
Chomsky, Noam. 2001. Derivation by Phase. In Michael Kenstowicz, ed, Ken Hale: A
Life in Language. Cambridge, MA: MIT Press, 1-52.
Cipollone, Domenic. 2001. Morphologically complex predicates in Japanese and what
they tell us about grammar architechture. In OSU Working Papers in Lingusitics
56, ed. by Michael Daniels, David Dowty, Anna Feldman and Vanessa Metcalf.
Dubinsky, Stan. 1994. Predicate union and the syntax of Japanese causatives. Journal of
Linguistics 30, 43-79.
Embick, David. 2004. On the structure of resultative participles in English. Linguistic
Inquiry 35.3, 355-392.
Folli, Rafaella, Heidi Harley and Simin Karimi. 2005. Determinants of event structure in
Persian complex predicates, Lingua 115.10, 1365-1401
Fortin, Catharine R. 2004. Minangkabau Causatives: Evidence for the l-syntax/s-syntax
39
division. Talk presented at AFLA 2004, April 2004, ZAS Berlin.
Harada, S. I. 1973. "Counter Equi NP Deletion," Annual Bulletin 7, The Research
Institute of Logopedics and Phoniatrics, University of Tokyo.
Hale, K. and Keyser, S.J. 1993. On argument structure and the lexical expression of
syntactic relations. In Hale, K. and S.J. Keyser, eds. The view from Building 20:
Essays in honor of Sylvain Bromberger. Cambridge: MIT Press
Hale, K. and S. J. Keyser. 2002. Prolegomena to a theory of argument structure.
Cambridge, MA:MIT Press.
Harley, H. 1995a. Subjects, Events and Licensing. PhD. Dissertation, MIT.
Harley, H. 1995b. Sase bizarre: The structure of Japanese causatives. In Proceedings of
the Canadian Linguistic Society Meeting, ed. by P. Koskinen.
Harley, H. and R. Noyer. 2000. Licensing in the non-lexicalist lexicon. In B. Peeters, ed.,
The Lexicon/Encyclopedia Interface, 349-374, Amsterdam: Elsevier Press.
Jackson, Eric. 2005c. Derived statives in Pima. Paper presented at the SSILA Annual
Meeting, 7 January 2005.
http://www.linguistics.ucla.edu/people/grads/ejackson/SSILA05PimaDerivedStati
ves.pdf
Jacobsen, W. M. 1981. Transitivity in the Japanese Verbal System. Ph D. dissertation,
University of Chicago.
Jacobsen, W.M. 1992. The transitive structure of events in Japanese. Studies in Japanese
Linguistics I, Tokyo: Kurosio Publishers
Kitagawa, Y. 1986. Subjects in Japanese and English. Doctoral dissertation, University
of Massachusetts, Amherst.
40
Kitagawa, Y. 1994. Subjects in Japanese and English. Outstanding Dissertations in
Lingusitics series. New York: Garland.
Kratzer, Angelika. 1996. Severing the external argument from its verb. In J. Rooryck and
L. Zaring eds., Phrase Structure and the Lexicon, 109-137. Dordrecht: Kluwer.
Kuno, S. 1973. The Structure of the Japanese Language. Cambridge, MA: MIT Press.
Kuroda, S.-Y., 1965a. Causative forms in Japanese. Foundations of Language 1.30-50.
Kuroda, S-Y. 1965b. Generative Grammatical Studies in the Japanese Language.
Doctoral dissertation, MIT.
Kuroda, S.-Y., 1981. Some recent issues in linguistic theory and Japanese syntax.
Coyote Papers 2.103-122.
Kuroda, S.-Y., 1990. Shieki no jodoshi no jiritsu-sei ni tsuite. Bunpo to imi no aida:
Kunihiro Tetsuya kyoju kanreki taikan kinen ronbunshu. 93-104. Tokyo: Kurosio.
Kuroda, Y (1994) Lexical and Productive Causatives in Japanese: an examination of the
theory of paradigmatic structure, in Journal of Japanese Linguistics
Kuroda, S-Y. 2003. Complex predicates and predicate raising. In N. Fukui (ed.), Formal
Japanese syntax and universal grammar: The past 20 years, Lingua.4-6:447-480
Manning, C, I. Sag and M. Iida. 1999. The lexical integrity of Japanese causatives. In
Levine, R.D. and Green, G. (eds.) Studies in contemporary phrase structure
grammar. Cambridge: CUP.
Marantz, Alec. 1991. Case and licensing. In ESCOL 91: Proceedings of the Eighth
Eastern States Conference on Linguistics, ed. German Westphal, Benjamin Ao,
and Hee-Rahk Chae, 234253. CLC publications.
41
Marantz, A. 1997. No escape from syntax: Dont try a morphological analysis in the
privacy of your own lexicon In A. Dimitriadis, L. Siegel, et. al., eds. University
of Pennsylvania Working Papers in Linguistics, vol. 4.2, Proceedings of the 21st
Annual Penn Linguistics Colloquium, 1997, pp. 201-225.
Matsumoto, Yo. 2000. On the Crosslinguistic Parameterization of Causative Predicates:
Implications from Japanese and Other Languages, In Miriam Butt and Tracy
Holloway King (eds.), Argument Realization. Stanford: CSLI Publications, 135-
169.
Miyagawa, S. 1980. Complex verbs and the lexicon. PhD. dissertation, University of
Arizona.
Miyagawa, S. 1984. Blocking and Japanese causatives. Lingua 64, 177-207.
Miyagawa, S. 1989. Structure and Case Marking in Japanese. New York: Academic
Press.
Miyagawa, S. 1994. "(S)ase as an Elsewhere Causative," Program of the Conference on
Theoretical Linguistics and Japanese Langauge Teaching, Tsuda University, pp.
61-76.
Miyagawa, S. 1998. "(S)ase as an Elsewhere Causative and the Syntactic Nature of
Words," Journal of Japanese Linguistics 16, pp. 67-110
Miyagawa, S. 1999. "Causatives," Natsuko Tsujimura, ed., The Handbook of Japanese
Linguistics, Blackwell, pp. 236-268.
Murasugi, Keiko, Tomoko Hasimoto , and Sachiko Kato. 2004. On the Acquisition of
Causatives in Japanese*. BU Conference on Language Development 28
Proceedings Supplement,
http://www.bu.edu/linguistics/APPLIED/BUCLD/supp.html
42
Nishiyama, Kunio. This volume. Morphology and predicate formation.
Pylkkanen, L. 2002. Introducing Arguments. Ph. D Thesis, MIT.
Sadakane, N. and M. Koizumi. 1995. On the nature of dative particle -ni in Japanese.
Linguistics 33:5-33.
Shibatani, Masayoshi. 1990. The Languages in Japan. Cambridge: Cambridge University
Press.
Svenonius, P. 2005. Two domains of causatives. Talk presented at CASTL, March 10,
2005.
http://www.hum.uit.no/a/svenonius/papers/Svenonius05TwoDomains.pdf
Takano, Yuji. 2004. Coordination of verbs and two types verbal inflection. Linguistic
Inquiry 35-1: 168-178.
Terada, M. 1991. Incorporation and Argument Structure in Japanese, PhD dissertation,
University of Massachusetts, Amherst, Amherst:GLSA
Travis, Lisa. 2000. Event structure in syntax. In Events as Grammatical Objects: The
Converging Perspectives of Lexical Semantics and Syntax, edited by Carol Tenny
and James Pustejovsky, pp. 145-185. CSLI, Stanford
Volpe, Mark. 2005. Japanese Morphology and Its Theoretical Consequences:
Derivational Morphology in Distributed Morphology. PhD. Dissertation, SUNY
Stony Brook.
Zenno, Y. 1985. Paradigmatic Structure and Japanese Idioms. Masters thesis, Ohio
State University, Columbus.
Heidi Harley
Department of Linguistics
University of Arizona
Tucson, AZ 85721-0028
43
hharley@email.arizona.edu
44
*
I wish to express my gratitude to the workshop organizers, Shigeru Miyagawa and
Mamoru Saito, for inviting me to present at the Workshop on Linguistic Theory and the
Japanese Language in July, 2005 at Harvard University, and for their valuable feedback,
as well as that of the audience there. Thanks also to Yosuke Sato and Kunio Nishiyama
for their detailed comments, which have improved the manuscript greatly. Caroline
Heycock and Akira Watanabe also provided important comments. Takaomi Kato, Yosuke
Sato, Hironobu Kasai and Kazutoshi Ohno helped me with some of the Japanese example
sentences herein. Any and all remaining shortcomings and mistakes are of course my
responsibility entirely.
1
Kuroda 1981, 1990 (as cited in Kuroda 2003) presents some examples from negation
and intervening particles to suggest that -sase- does have an independent existence as a
verbal morpheme; Miyagawa 1989:115f, and Kitagawa 1994:184f., followed by
Manning, Sag and Iida 1999:47, argue that in fact these are examples of the -ase-
allomorph suffixed to light verb s-, do. Kuroda (2003 n. 14) disagrees, ascribing
Kitagawas position to grammaticality judgment differences.
2
In the case of the make-reading of a causative of a transitive verb, it seems that the
dative -ni on the Causee is the structural case-marker -ni, not postpositional -ni (Sadakane
& Koizumi 1995). In let-reading causatives, with both transitive and intransitive root
verbs, the -ni on the Causee seems to be P -ni
3
An interesting piece of evidence showing the monoclausal nature of the case-marking of
productive causatives is noted by Manning, Sag and Iida: when a productive causative is
suffixed with the potential/abilitative morpheme -e-, the embedded accusative argument
45

may be optionally marked with nominative case, rather than accusativejust like the
accusative arguments of monoclausal, morphologically simple transitive verbs suffixed
with the same morpheme. An example illustrating this phenomenon is given below:
(i) Sensei-ga Taroo-ni tsukue-ga/o soujis-ase-re-naka-tta (koto)
teacher-N Taroo-D desk-N/A clean-CAUSE-ABL-NEG-PAST (fact)
The teacher was not able to make Taroo clean his desk.
4
See Nishiyama (this volume) for discussion of this problem with respect to other types
of morphologically complex words in Japanese.
5
Akira Watanabe (p.c.) notes that this same problem for lexicalist treatments of complex
predicates arises with the affixal light-verb constructions in Japanese.
6
Interestingly, it seems that this structure is not available for examples involving true
coordination, rather than disjunction (thanks to Takaomi Kato for these examples and
discussion):
(i). *Ken-wa [Naomi-ni [kesa hurui huku-o sute]
K.-Top [N.-D [this.morning old clothes-A throw:away]
[sakuban kuroozetto-o soozi]]-sase-ta.
[last:night closet-acc cleaning]]-sase-PST
"Ken made Naomi throw away her old clothes this morning and clean out the closet
last night."
(ii) *Yamada kyoozyu-wa [betubetu-no gakusei-ni [toogoron-no ronbun-o yomi ]
Yamada professor-Top [different-G student-D [syntax-G paper-A read ]
[oninron-no ronbun-o kak]]-ase-ta.
[phonology-G paper-A write]]-sase-PST.
46

Prof. Yamada made different students read a paper on sytax and write a paper on
phonology.
In (i), temporal modifiers are included which make the first VP temporally follow the
second; in (ii), betubetu different is attached to the Causee argument. Both these
controls are intended to enforce a true coordination structure, and the sentences are
ungrammatical. (See Takano 2004 on relevance of betubetu for ensuring that a given
sentence with a bare V-stem VP contains a true coordination structure, rather than an
adjoined, VP-modifier position for the first VP). Similar examples without the temporal
modifiers or betebetu seem clearly to involve adjunction, as argued by Manning, Sag &
Iida; they get a sequential, After Xing, Y, or X, then Y reading. Interestingly, the
same is true in Korean, which, unlike Japanese, has an overt coordination particle. Its
quite mysterious why coordination is not allowed under -sase but disjunction
isespecially since, as Takano 2004 shows, coordination in Japanese can apply to bare
verb stems, below, e.g., a single tense morpheme. See XX below for speculation.
7
Since I have said that such an approach should treat all productive morphology (esp.
inflectional morphology) as syntactically attached, I have adopted a KaseP hypothesis
for Japanese case particles in this tree. For discussion of how these case morphemes can
be licensed by particular case-marking heads, see Miyagawa 1999.
8
Because productive causatives can undergo nominalization, this theory has to allow the
syntax to derive -kata way of nominals, too.
9
This basic picture once established, many questions remain to be solved, concerning the
make/let distinction, the role of unergativity, unaccusativity and agentivity, psych-
47

predicate causatives, restructuring effects, and more. See the survey literature cited above
and references therein.
10
The number of pairs does not include other pairs derived from a root already on the list
even when these are not transparently semantically related; the number of listemes on
each list, then, is likely somewhat larger.
11
Mamoru Saito and Yosuke Sato (p.c.) inform me that the forms listed as meaning
open here, hirakeru~hiraku, are not used (the pair from class II, aku~akeru, is the
appropriate one). Some other particular items in Jacobsens lists also seem to not
currently be in use, for example bakasu, dekasu, and nukumeru.
12
See also Matsumoto 2000.
13
The adversity interpretation is not available for these examples for at least some
speakers (Yosuke Sato, p.c.).
14
Kunio Nishiyamia (p.c.) notes that some clearly monoclausal (and hence lexical)
causatives nonetheless do not have an adversity interpretation, such as sir-ase-ru,
inform, lit know-cause. The idiomatization and adversity causative implications are
thus unidirectional: If a causative has an idiomatic interpretation or an adversative
interpretation, it is lexical; if it does not, the case is still underdetermined. Tests involving
the full constellation of monoclausal/biclausal properties listed above is necessary to
establish lexical vs. syntactic status in many individual cases. The testing situation is
complicated in cases where a ! and the unaccusative/inchoative form of the verb are
phonologically identical, however, since both the lexical reading (embedding the root)
and the syntactic reading (embedding the inchoative vP) may be available in such cases.
48

15
Another clear example of blocking arises in the case of verb-specific honorific
morphology in Japanese, as for tabe, eat. It cannot be marked with the productive
honorific *o-tabe-ni naru. Rather its irregular honorific form must be used: mesiagaru,
eat.HON. For discussion, see Volpe 2005.
16
In such theories, there is an important distinction between the root lexeme itself, which
is the real lexical entry, and its word forms, which merely occupy the slots of its
paradigm. In theories which allow paradigms to attach derivational morphology, like
Miyagawas, the word forms which fill the slots in a derivational paradigm are also
themselves lexemes which have their own inflectional paradigms attached paradigms
within paradigms, as it were.
17
Most of the following is a mildly revised version of Miyagawas 1994, 1998 analysis,
which appeared in Harley 1995. It has been updated to take recent Minimalist
terminology into account, but the crucial mechanisms remain the same.
18
Though cf. Takano 2004.
19
In the DM account of Harley 1995 and Marantz 1997, there is no constituent V, and
hence no VP: a verb is created in the syntax by combining a ! and a v head. The
distinction between a light verb and a main verb, then, comes down to whether the v
element has had a ! element combined with it or not: main verbs are v+! combinations;
light verbs like -sase- are pure v. The question of whether any given verb in a language
is a realization of just v or of v+!, then, is an empirical question. For instance, English
make might be a v, like -sase-, or a v+! (=V) combination, like coerce or permit. See
also footnote 22.
49

20
Although, in the analysis about to be described, the class of causatives we have been
calling lexical are formed in the syntax, and no pre-syntactic word-forming lexical
component is assumed, the major conceptual advantage of the analysis. For
terminological consistency, however, Ill continue referring to these monoclausal
causatives as lexical.
21
To make the accounts of the English and Japanese lexical causatives as parallel as
possible, we will assume that the v head has had the ! node adjoined to it by head-
movement. Because Japanese is head-final, however, the head-movement is not
necessary to get all the morphemes in the correct order; they could remain in situ but
simply merge under adjacency.
22
Sometimes, of course, the appropriate v causative morpheme in a lexical causative is
itself the null v morpheme, as for Jacobsens classes I and IV above, e.g. ak--U,
open.tr. The question of why the root !ak, open, may also be embedded under
v
BECOME
(allomorph -e-) but the root !tabe, eat, apparently may not, despite also taking
the allomorph of the external-argument-selecting v, can be answered in one of two
ways. If -rare-, the passive morpheme, is the Elsewhere form of v
BECOME
, then it is
perhaps the case that !tabe does alternate, with its passive form tabe-rare involving
replacing the v
CAUS
morpheme for a v
BECOME
, -rare-, thus explaining the suppression
of the external argument, similar to the analysis of Persian light verbs presented in Folli,
Harley and Karimi 2005. Alternatively (if the productive passive is assumed to involve
embedding vP, rather than !P, like the productive causative), the failure of !tabe to
alternate can be taken as an indication that the immediate licensing environment for
!tabe insertion does not allow a v
BECOME
; for discussion on how to restrict the
50

productivity of inchoative/causative alternations in a syntactic approach to verb-
formation, see Harley and Noyer 2000.
23
This is the case no matter whether the ! and embedded v head move to the matrix v,
or if they are attached just by virtue of their head-final adjacency in Japanese. In either
case, the same remarks obtain: a productive causative v will never be adjacent to a root,
but will always be separated from it by another v, the one that comes lexically with that
!.
24
There are still generalizations about the various causatives which remain uncaptured on
this analysis. For instance, why does lexical -sase- always alternate with an inchoative
form whose inchoative suffix is ? In principle, a root could be on a special list for an
unaccusative morpheme like -r- or -e-, but not for a causative morpheme, and hence -r-
or -e- could in principle alternate with -sase-, yet this does not seem to occur. See
Miyagawa 1998 for a proposal. Similarly, the haplological restriction on multiple
occurrences of -sase are unexplainedone would expect that inifinite recursion should
be possibleas with John made Mary make Joe make Sue eat pizzabut more than one
syntactic -sase embedding is ungrammatical (though Kuroda 1994 shows that a lexical
-sase may combine with one syntactic -sase).
25
There are interesting reasons to question this precise formulation of both of these
claims, brought up by Volpe (2004). Lexical causatives can be the input to
nominalization, with nominalizing morphemes conditioned by the particular lexical
causative involved, and subject to subsequent meaning drift. For example, the lexical
causative chir-as-u, to scatter, when suffixed by nominalizing -i (chirashi), refers
idiomatically to leaflets, not to any scattered item. Yet this idiomatic nominalization
51

clearly includes the causative suffix -as-, which should have been a barrier to special
meaning assignment, under the Marantz/Arad hypothesis. Pylkkanen (2002) proposes
that the causative head and agent-introducing heads are distinct, with the latter selecting
the former as its complement, and that only the agent-introducing head is the boundary
for special meaning. Its possible that her approach could resolve the conflict between
Volpes idiomatic nominalization facts (including v
CAUS
but not the agent head) and
Miyagawas verbal idiomatization facts (assuming the productive causative embeds the
agent-introducing phrase, not just v
CAUS
P).