Professional Documents
Culture Documents
Hayashi Shintaro Thesis
Hayashi Shintaro Thesis
The goal of this thesis is to argue for head movement in Japanese, which is
necessarily (and notoriously) ‘string-vacuous’ due to the head-final character of the
language. I show that a couple of discrepancies between complement te-clauses and
adjunct te-clauses cannot be satisfactorily captured unless head movement is resorted
to. It is also demonstrated that (what I refer to as) NJV clauses and VN clauses, which
have been traditionally distinguished from each other in the literature, exhibit
asymmetries under ellipsis phenomena. I argue that the asymmetries are rooted in
predicate raising, commonly referred to as verb movement in the literature.
HEAD MOVEMENT IN AN AGGLUTINATIVE SOV
LANGUAGE
By
Shintaro Hayashi
Advisory Committee:
Tomohiro Fujii, Associate Professor (Thesis Supervisor)
Tatsunori Mori, Professor
Naoyoshi Tamura, Professor
Roger Martin, Associate Professor
Kenshi Funakoshi, Project Assistant Professor
© Copyright by
Shintaro Hayashi
2015
To Toshinobu, Michiru and Sarasa
iii
Acknowledgements
Time is up. It is time to give up seeking for any further improvement and submit my
dissertation. Writing a dissertation is extremely tough work (as everyone knows), and
I have always been worring whether I could ever finish it. Very fortunately, I am
Fujii, Tasunori Mori, Naoyoshi Tamura, Roger Martin, and Kenshi Funakoshi, all of
syntacticians, Tomo, Roger and Kenshi, are owed a deep debt of gratitude. Tomo, my
thesis supervisor, is the person I can never thank enough. At every meeting with him,
hundrends of invaluable comments. I am 100% sure that this dissertation would never
have come into existence without his enormous intellectual support. Sincerely, I am
very privileged to have been directed by Tomo. Roger has constantly reminded me of
what syntactician living in the twenty-first century have to be aware of. I am grateful
to him for inviting me to Syntax Reading Group at Keio University as well, which
broadened my sight in so many ways. Chapter 3 of this dissertation would not have
committee; it is his dissertation that made me want to pursue the topics presentend
iv
there. Ever since we met first in 2014, I have been encouraged by not only his
Takashi Munakata, Koji Sugisaki, Osamu Sawada, Hisako Takahashi, Kohei Suzuki,
Jason Merchant, Karlos Arregi, Masaya Yoshida, Bum-Sik Park, Sayaka Goto and
Yoshiki Fujiwara.
Last but not least, I thank my parents, Toshinobu and Michiru, and my sister,
Sarasa, for their support, patience and love. I bet they never imagined that a punk in
Sapporo who never showed any interest in academic stuff at all would grow up to
v
Table of Contents
Acknowledgements...................................................................................................... iv
Table of Contents......................................................................................................... vi
1. Introduction
Section 1 String-vacuity of head movement in Japanese ................................1
Section 2 Otani and Whitman (1991) ..............................................................4
Section 3 Koizumi (2000)..............................................................................10
Section 4 Oku (1998) .....................................................................................12
Section 5 Fukui and Sakai (2003)..................................................................14
Section 6 Outline of the thesis .......................................................................16
References .................................................................................................................140
vi
Chapter 1: Introduction
One thing about head movement in Japanese that many syntacticians recognize is
grammar, is not expected to bring about any word order permutation. One traditional
way to diagnose movement in overt syntax is to check whether the surface position
that an element occupies in one environment is distinct from the one it appears in
English.
In the declarative clause (1a), the auxiliary will is placed after the subject John, while
in the interrogative clause (1b) it is placed before the subject. Facts like this have led
According to such a head movement analysis, the structure in (2a) is assigned to (1a),
in which the auxiliary is located in T. By contrast, (1b) has the structure in (2b), in
which the auxiliary has moved from T to C (see Adger 2003 for a comprehensive
review).
1
(2) a. [CP C[-Q] [TP John [T will] eat apples]]
that in the hierarchical structure we have two heads, X and Y, and that Y is
(3) Syntax
[YP … [XP ZP X] Y]
In PF, these two heads are always linearized next to each other in Japanese, as shown
in (4) (the symbol < means ‘precedes’). ZP, whether an argument or an adjunct, does
(4) PF
… ZP<X<Y …
Importantly, the PF representation derived from the structure in (5) below, in which X
undergoes head movement to Y, cannot be differentiated from the one derived from
(3) in which X does not move, due to the head-final character of the language. In
other words, both (3) and (5) are expected to yield the same PF representation in (4).
2
(5) Syntax
Recall that in the case of subject-auxiliary inversion in English in (2), we have John,
which intervenes between the two relevant heads (namely, T and C). This intervening
element can be taken to signal head movement; if the auxiliary precedes John in the
the derivation, nothing is linearly crossed on its way to Y. For this reason, head
nothing about word order can help us detect it, then the question is how one can
cogently argue for it (or against it, for that matter). Below, I briefly review arguments
for head movement in Japanese presented by Otani and Whitman (1991) and Koizumi
(2000) in section 1.2 and 1.3. To anticipate a bit, the claim of these authors is that
According to such a verb movement analysis, the sentence in (6), for instance, can be
derived in the way illustrated in (7) with English vocabulary (irrelevant details are
suppressed); the verb tabe ‘eat,’ which is generated as the head of the VP, undergoes
3
(6) Taro-wa tenpura-o tabe-ta.
After reviewing their ideas, in section 1.3 and 1.4 I also take up Oku’s (1998)
argument against Otani and Whitman and Fukui and Sakai’s (2003) against Koizumi.
Central to Otani and Whitman’s (1991) discussion is the null object construction
Mary-also discard-Past
In (8b), the object of the verb sute ‘discard’ is not pronounced, as indicated by [e].
The important observation they make is that when preceded by (8a), in which the
anaphor zibun ‘self’ is contained in the object, the NOC in (8b) displays ambiguity
4
with respect to the interpretation; it can be taken to mean either (9a) or (9b). The
former reading is referred to as the ‘strict’ reading, and the latter is called the ‘sloppy’
reading.
They first point out that it is not promising to analyze [e] in (8b) as a variable
As indicated in the translation, (10b) allows the sloppy reading. Huang’s analysis of
this fact is that there is an unpronounced topic in the initial position, and this null
topic binds a variable in the object position. To be more precise, (10b) is comparable
5
to (11) below in the relevant respect. In (11), ziji ‘self’ appears as a topic and binds a
variable [e], which gives rise to the sloppy reading, according to Huang. He further
argues that (10b) has an implicit topic phrase corresponding to ziji, which can be
At first sight, it might appear that Huang’s null topic analysis of (10) can be
extended to the NOC in Japanese in the following way. (8b) is structurally parallel to
(12) below, in which zibun-no tegami ‘self’s letters’ overtly appears as a topic and
binds a variable. The same binding relation obtains in (8b), but the topic happens to
be inaudible in that case. Since (12) allows the sloppy reading, so does (8b). This is
how the null topic analysis à la Huang captures the sloppy reading in (8b).
The reason that Otani and Whitman reject the null topic analysis of (8b) is that it
wrongly predicts that the sloppy reading is available in the case of (13).
6
(13) a. [Zibuni-no hatake-no ninzin]-ga Makuguregaa ozisani-no
daikoubutu desi-ta.
big.favorite be-Past
‘The carrots from selfi’s garden were Mr. McGregori’s big favorite.’
‘Peter was also very fond of the carrots from Mr. McGregor’s garden.’
Crucially, (13b) does not allow the sloppy reading ‘Peter was also very fond of the
carrots from his own garden,’ which seems unexpected if the null topic strategy is
available to derive the sloppy reading in (8b). As Otani and Whitman demonstrate,
the intended reading does obtain when there is an overt topic zibun-no hatake-no
daikoubutu desi-ta.
big.favorite be-Past
‘The carrots from selfi’s garden were Mr. McGregori’s big favorite.’
7
b. [Zibunj-no hatake-no ninzin]k-wa Piitaaj-mo [ek]
daisuki desi-ta.
very.fond.of be-Past
‘The carrots form selfj’s garden, Peterj was also very fond of.’
If the null topic strategy can be employed to derive the sloppy reading in the NOC in
(8b), then it is natural to predict that (13b) allows the sloppy reading by virtue of the
same strategy. However, this prediction is not born out, which leads Otani and
ambiguity (Sag 1976, Williams 1977); the sentence in (15b) which has a null VP is
ambiguous between the strict reading (16a) and the sloppy one (16b).
8
Otani and Whitman claim that the NOC in (8b) allows the sloppy reading because it
is derived by VP-ellipsis as well. In other words, they contend that the gap in (8b) is
actually a null VP, rather than a null object. The crucial point of their analysis is that
the verb sute ‘discard’ undergoes verb movement out of the VP, as shown in (17),
which makes the VP look similar to the one in the English example (15b) in that no
For expository purposes, let us interpret their VP-ellipsis analysis as follows. The
structure of (8a) is (18a), and (8b) is derived by deletion of the VP which is identical
to the antecedent VP in (18a), but crucially deletion does not affect the verb because
1
It should be mentioned that Otani and Whitman adopt the LF-copying analysis of
VP-ellipsis advocated by Williams (1977), and assume that interpretation of null VPs
is achieved by LF interpretive rules such as Derived VP Rule and VP Rule proposed
by Williams. Although in (17) I indicate their VP-ellipsis analysis of the NOC in
Japanese in terms of PF-deletion (Sag 1976, Merchant 2001) rather than LF-copying,
this is just for expository reasons, which does not affect the gist of their proposal.
9
If their analysis that attributes the sloppy reading in (8b) to VP-ellipsis is correct, then
it also accounts for why (13b) above lacks the sloppy reading. They claim that the
intended reading is absent in (13b) because it is not derived by VP-ellipsis, and that
VP-ellipsis is disallowed in (13b) for the same reason it is in the English example
in the discourse.
Thus, Otani and Whitman have argued for string-vacuous verb movement in Japanese
on the basis of the availability of the sloppy reading in the NOC. Their analysis which
incorporates verb movement and VP-ellipsis is empirically superior to the null topic
analysis, as the former is free from the problem of overgeneration the latter faces in
Overt verb movement in Japanese has also been argued for quite independently by
10
(20) [TP Subj [vP [VP IO [VP DO tV] tv] [[V+v]+T]]
scrambling, though I do not review all of his arguments. Rather, I focus on the one
based on coordination, because it is not only particularly revealing but also relatively
Koizumi takes up the example in (21), which involves coordination with the
‘(the fact that) Mary gave two apples to John, and three bananas to Bob.’
(Koizumi 2000:228)
The standard assumption about coordination is that only syntactic constituents can be
conjoined. If we maintain this assumption, then it must be the case that in (21) the
indirect objects John-ni ‘to John’ and Bob-ni ‘to Bob’ form a constituent with the
direct objects ringo-o futa-tu ‘two apples’ and banana-o san-bon ‘three apples,’
respectively. One question that immediately arises is why such unusual coordination
11
operation in syntax creates a constituent which is exclusively made up of an indirect
(22) [TP Mary [vP [VP [VP to John [VP two apples tgive] and
The crucial point of Koizumi’s analysis is that the verb age ‘give’ raises to v in an
moves to T. ATB verb movement in (22) thus makes the two VPs look superficially
verbless (‘remnant’ VPs in his terminology), which gives us the impression that the
indirect and direct objects make up a syntactic constituent. On the basis of the well-
In section 1.2, I have reviewed Otani and Whitman’s (1991) argument for verb
movement in Japanese. The primary motivation for them to propose verb movement
is the fact that the NOC in (8), repeated here as (23) for the sake of convenience,
12
analyzing the empty category [e] as a null VP, which is created by verb movement in
syntax.
Mary-also discard-Past
However, the validity of their argument has been challenged by Oku (1998),
omotteiru.
think
(Oku 1998:166)
13
There is a gap [e] in the subject position of the embedded clause in (24b). Oku
correctly points out that (24b) is ambiguous between the strict reading in (25a) and
The fact that (24b) allows the sloppy reading indicates that “the sloppy identity
ellipsis in English, which Otani and Whitman say that the NOC in Japanese is similar
to, does not create subject gaps. Since the sloppy reading does obtain in a context
where we cannot reasonably analyze the empty category to be a null VP, (24b)
Japanese2.
As quickly reviewed in section 1.3, Koizumi (2000) observes that the indirect and
direct object appear to form a syntactic constituent in (21) (repeated as (26) here),
which is conjoined by the particle to ‘and,’ assuming that only syntactic constituents
can be conjoined.
2
See also Hoji (1998), who independently argues against Otani and Whitman’s
analysis that equates the NOC in Japanese to VP-ellipsis in English.
14
(26) Mary-ga [[John-ni ringo-o futa-tu] to
‘(the fact that) Mary gave two apples to John, and three bananas to Bob.’
He then analyzes (26) as coordination of remnant VPs, and takes its well-formedness
to motivate verb movement in Japanese (in this particular case, ATB verb movement).
Fukui and Sakai (2003), who study coordination in some more detail and make the
following comment: “it seems premature to draw any definite conclusion from his
argument with respect to the existence of string vacuous V-to-T raising in Japanese”
(Fukui and Sakai 2003: 343)3. They provide the example of coordination in (27).
3
They also examine Koizumi’s arguments for verb movement in Japanese based on
scrambling and (pseudo-)clefts which I have skipped in section 1.3, and conclude that
they are not convincing enough to justify string-vacuous verb movement either. See
Fukui and Sakai (2003) for details. See also Takano (2002), who has questioned the
validity of Koizumi’s argument for verb movement based on (pseudo-)clefts.
15
(Fukui and Sakai 2003:346)
In (27), the connective particle to appears to conjoin the bracketed portions. These
bracketed portions contain arguments of the verb age ‘give’ on the one hand (namely,
Hanako-ni ‘to Hanako’ and Kumiko-ni ‘to Kumiko’) and arguments of the noun oyatu
‘snack’ on the other (namely, ringo(-o) mit-tu ‘three apples’ and banana(-o) ni-hon
‘two bananas’), although it is highly unlikely that an argument of the verb and that of
the noun form a syntactic constituent. Fukui and Sakai point out that even if one can
(27), verb movement does not seem responsible for such constituents, because
remnant VPs which result from ATB verb movement should not contain arguments of
nouns.
Based on this observation, Fukui and Sakai conclude that somewhat exotic
Koizumi proposes, since equally (or even more) exotic coordination is permitted in
Japanese has been extensively discussed in the literature, it seems fair to say that
16
additional pieces of empirical evidence for the presence of head movement in the
grammar of Japanese.
boundary.
single clause. I first introduce two kinds of verbal predicates in Japanese, native
Japanese verbs and verbal nouns, and demonstrate that they behave differently in
ellipsis contexts. It is proposed that asymmetries they exhibit under ellipsis are rooted
in predicate raising. I argue that native Japanese verbs undergo predicate raising
17
Chapter 2: Head Movement out of Te-clauses
2.1 Introduction
The goal of this chapter is to argue for the presence of string-vacuous head movement
I argue that in (1) the string of the embedded verb tukut ‘cook’ and the morpheme -te
(hereafter V-te complex, or V-te for short) undergoes head movement to the higher
verb morat ‘get’ (see Nakau 1973: Ch. 7, Inoue 1976, Harada 1977/2000, Shibatani
1978, McCawley and Momoi 1986, and Nakatani 2004: Ch. 7 for relevant discussion
of V-te; see also Kuroda 1965: Ch. 6, Miyagawa 1987b, Terada 1990: Ch. 5,
Kitagawa and Kuroda 1992, Tsujimura 1993, Hoshi 1994: Ch. 4, Matsumoto 1996:
Ch. 3, Takahashi 2012, and Nishigauchi 2013 for structural proposals given to
The novelty of the study presented in this chapter is twofold. First, I provide
new arguments in favor of the classical head movement analysis of complement V-te,
based on hitherto unnoticed or less acknowledged facts. I also bring adjunct te-
clauses in perspective (see Tamori 1977, Nakatani 2004, Uchimaru 2006, Hayashi
18
2013 for discussion of adjunct V-te), demonstrating that adjunct V-te, unlike
complement V-te, never moves. Second, although the idea that complement V-te
moves has been found in the literature (see the references cited above), a unified
analysis of te-complements and te-adjuncts has not been fully worked out. I attempt
clauses, and argue that these properties can straightforwardly be accounted for by
head movement, and propose that head movement out of the te-complement takes
place in order to eliminate the uninterpretable tense feature on the higher predicate it
adjoins to. In Section 2.3 I take up some possible alternatives, which do not appeal to
head movement, to the analysis presented here. Section 2.4 explores a theoretical
that it does not feed word formation. It is suggested that this unusual result of head
This section introduces properties of te-clauses and shows that an analysis in terms of
head movement captures the data in a principled manner. Alternatives to the head
19
2.2.1 Cross-clausal head movement out of te-clauses
The purpose of this section is to lay out properties of te-clauses and show that
the properties naturally follow if V-te moves into the next higher clause in the
complement construction but not in its adjunct counterpart. Sections 2.2.2 and 2.2.3
deal with asymmetries that complement and adjunct te-clauses exhibit. In section
(2) and (3) below illustrate complement and adjunct te-clauses, respectively.1
1
As Matsumoto (1996) discusses, predicates that can take te-complements are
limited; e.g. ar ‘be,’ age ‘give,’ ik ‘go,’ ku ‘come,’ ok ‘put,’ mi ‘see,’ simaw ‘finish,’
etc.
20
(3) Taro-ga [piza-o tukut-te] okane-o morat-ta.
Complement te-clauses may involve various types of complementation. (2a) and (2b)
respectively. The structures of (2c) and of (2d) seem more difficult to identify. (2c),
naku ‘hardly get any customers’ is embedded under the desiderative adjective hosiku
this study. Rather, what is more important for us is that the properties examined in
2
This possibility is shown in (i) below. The representation in (i) seems plausible
because the verb morat ‘get’ can take a dative or ablative phrase as its source
argument when it does not take a te-complement, as in (ii).
The choice between (2c) and (i) does not affect the discussion made in the text in any
significant way. See Harada (1977/2000) for discussion on this matter.
21
this section are invariably observed across these te-complements selected by different
main predicates.
fashion, and in (5), where the V-te stays inside the adjunct clause.
Before going on to look at the core data, two points need to be made. First, regarding
the categorical status of -te, I adopt Nakatani’s (2004:129) analysis cited in (6).
otherwise.
According to (6), the morpheme -te is T, and te-clauses are nothing but TPs that are
not immediately dominated by CPs (see Nakatani 2004 for justification of (6)). So
when I say V-te moves up to the next higher clause, it means that T undergoes head
movement to the higher predicate. As for the question of how the embedded V and -te
are combined, I assume that head movement in syntax (viz. V-to-T movement) is also
chapter 3, the proposal concerning head movement of T made in this chapter can be
22
Next, I would like to demonstrate that the complement/adjunct bifurcation I
literature (see Nakau 1973, Tateishi 1994 and references cited therein), when the VP
proform soo su ‘do so’ is present, complements of verbs must disappear together with
the verbs, while adjuncts may or may not. For instance, (7b) is ungrammatical when
the direct object put inside the parentheses is pronounced, whereas (8b) is
piza-o tukut-ta.
pizza-Acc cook-Past
3
As pointed out by Tateishi (1994), examples like (7b) become considerably better if
a direct object occurring outside of the proform can be interpreted as a contrastive
focus. This factor must be controlled for, as Tateishi notes, when examining
constituent structures of VPs, and it actually is in the examples in the text.
23
b. Ziro-mo (razio-o kiki nagara) soo si-ta.
If we apply this VP-replacement test to te-clauses, the following results are obtained:
what I call ‘te-complements’ must disappear under soo su (see (9b)) and what I call
morat-ta.
get-Past
morat-ta.
get-Past
24
b. Ken-mo (piza-o tukut-te) soo si-ta.
Given that a contrast emerges between (9b) and (10b) when the material inside the
or absence of movement of V-te. They concern (i) mobility of te-clauses, (ii) their
2.2.2 Mobility
Let us begin with mobility of te-clauses. I show that complement and adjunct
te-clauses behave differently with respect to mobility, as stated in (11). This is a first
(11) Fact 1
never be.
This can be confirmed by the sharp contrast between (12) and (13) below. (12)
25
the other hand, (13), which is derived by moving the te-complement across the
Related observations have been made by McCawley and Momoi (1986), Miyagawa
(1987b), Sells (1990), Terada (1990), Matsumoto (1996), and Nakatani (2004).
However, it should be remarked that none of these authors, as far as I can see,
explicitly states that there is a discrepancy between complement and adjunct te-
clauses, as I do in (11). Rather, they observe that in the case of (what I refer to as) te-
complements, V-te and the main clause predicate must be linearly adjacent to each
other, without drawing the distinction between complement and adjunct te-clauses.
The head movement analysis in (4) and (5) above correctly captures this
te-clause always affects the V-te as well as other elements in the clause because it
stays inside the larger constituent that undergoes fronting, as shown in (14).
(14) Taro [TP PRO [vP[VP pizza tV ] tv] [T [cook-v]-te]] money got
26
As for complement te-clauses, however, the V-te is required to move to the matrix
verb under the present analysis. As a result, the V-te cannot occur inside the moved
TP, as shown in (15), and therefore (13) cannot be derived unless the T head fails to
(15) Taro [TP Ziro [vP[VP pizza tV ] tv] tT] [[T [cook-v]-te]+get]-PAST
The account of (11) I have just presented can be extended to the behavior of
(16b) and (17b), respectively. Similar but completely different observations have
been made by McCawley and Momoi (1986) and Sells (1990), who study gapping (or
morat-ta no?
get-Past Q
27
b. Iya, [suupu-o tukut-te] da.
morat-ta no?
get-Past Q
The syntax of fragment answers has been a topic of great interest (e.g. Merchant
2004). Here, I adopt Nishigauchi’s (2006) proposal that fragment answers (or ‘short
through focus movement to the specifier of FocP headed by the copula da, followed
(see Nishigauchi 2006 for details; see also Hiraiwa and Ishihara 2002 for general
discussion of no da sentences).
28
Coupled with this specific analysis of fragments, the head movement analysis pursued
moved to [Spec, FocP], and the moved te-clause can survive deletion, remaining as a
fragment. A complement te-clause, however, cannot move to [Spec, FocP] with the
V-te complex inside. As shown in (19) below, V-te in the complement construction
always ends up being inside the elided FinP as a result of head movement into the
(19) [FocP [TP Ziro soup tT ] i [FinP [TP Taro ti [T [cook-v]-te]+get-Past] no] da]
expected.
nominalizations headed by the suffix -kata ‘way.’ The subsection has three parts.
well to account for the distribution of no in simple kata-nominals. Second, I use the
behavior of the two kinds of te-clause in nominals suggests that complement V-te
moves to the higher predicate, while adjunct V-te does not. Finally, I compare the
29
analysis of kata-nominals presented here and the fully-developed analysis by
think that any adequate analysis of the phenomenon must capture (see Sugioka 1992,
Kageyama 1993, Hoshi 2002, and Kishimoto 2006 for general properties of kata-
nominals). The examples in (21) are examined to see how these generalizations are
motivated4. (As will be stated below, I assume with Kishimoto 2006 that -kata
marking of XP.
4
The status of the particle no which appears in nominals seems massively ambiguous,
especially if Watanabe (2010) is correct in saying that there are two types of no, a
genitive case particle and a morphological linking element (see also Tsujioka 2011).
Throughout my discussion of kata-nominals, I simply gloss the particle as no, in
order to avoid complications.
30
(21) a. [NP [vP Taro-no [boonasu-o moratta toki]-no
mono-no kai(*-no)]-kata]
thing-no buy-no-way
‘the way in which Taro buys things when he gets a bonus payment’
(20a) is illustrated by the fact that the subject, the adverbial clause, and the object
associated with the verb kai ‘buy’ are all no-marked in (21a). (20b) maintains that no-
phrase that dominates it is no-marked. In (21b), nothing can be no-marked inside the
toki-clause since the clause itself is no-marked. Finally, if (20c), i.e. that neither vP
nor VP is a barrier for no-marking, did not hold, Taro in (21a) for example would not
Now I would like to introduce the machinery to account for these properties. I
follow Kishimoto (2006) in assuming that the nominalizing suffix -kata ‘way’ is N,
that it takes a vP complement, as noted above, and that v and V overtly raise up to N
(22) [NP [vP Taro [VP [Adjunct bonus got when] thing tV] tv] [N [buy-v]-way]]
5
Kishimoto (2006) argues that it is the nominalizing suffix -kata that triggers the
series of head movement, and in clausal domains, where vPs are typically selected by
T in lieu of -kata, no such head movement takes place. I do not agree with him on this
point, although I assume with Kishimoto that verb movement is involved in the
derivation of kata-nominals. In chapter 3, I provide evidence to show that verb
movement out of vPs takes place in clausal domains as well.
31
The structure in (22) where -kata takes a vP complement has been motivated by
Sugioka (1992), who provides compelling evidence. In (23a) we have the VP idiom
asi-o arau, whose literal interpretation is ‘to wash a leg.’ Sugioka’s crucial
observation is that the VP idiom can occur inside the kata-nominal, as in (23b).
‘the way in which one cuts the connection with a gang group’
(Sugioka 1992:59)
(in D-Structure or LF, see Larson 1988, Koopman and Sportiche 1991, Richards 2001,
and references cited therein), then the fact that idiomaticity obtains not only in the
clausal (or verbal) domain (23a) but also in the kata-nominal (23b) strongly suggests
not plausible to analyze the structure of (23b) as something like (24), in which the
nominalizing suffix is attached to the verb in the lexicon, because in such a structure
the verb wash and the object leg do not make up a syntactic constituent.
32
(24) [NP from gang group [NP leg [N wash-way]]]
(25) (see also Kitagawa and Ross 1982, Saito and Murasugi 1990, Watanabe 2010 for
It should be stressed that this is not a place to discuss how to translate this machinery
into a framework that does not appeal to government. Given my current purpose of
33
This said, let us see how the proposed structure in (22) with the assumptions
in (25)-(26) work to capture the properties listed in (20). First, (20a) follows under the
present system that incorporates the GTC (26d). In (22), -kata only governs vP if
nothing happens. When V-to-v movement takes place, the GTC first allows v to
govern the object NP, in addition to the subject NP (which v already governs without
V-movement). v-to-N movement next allows N to govern whatever v governs. So, the
subject and the object are both governed by N. Virtually the same applies to
government of the adjunct in (22), whichever projection inside vP the adjunct phrase
may be adjoined to. Note that why (20c) holds is already answered. After V-to-v-to-N
movement takes place, N can govern the domain that v and V are originally able to
Next, how is the locality effect mentioned in (20b) captured? Note fist that the
presence of the maximal projection headed by toki prevents N from governing the NP
boonasu. Notice then the GTC does not appear to make tokiP transparent for
government because there is no evidence that the head of the adjunct, toki, undergoes
head movement to V or v. In fact, such movement should not be possible, given that
an adjunct is an island. So it is very unlikely for N to govern into the adjunct clause.
order to make the proposed analysis complete. It should be asked why vP and VP,
34
(27) *[NP [vP Taro-no [VP … mono-no tV] -no tv] -no kai-kata]
Examples like this lead us to add (25c), i.e. that a phrase headed by a trace cannot be
that (25c) is perhaps due to the fact that no, being a suffix, needs a phonologically
non-null host and requires the host to be in a strictly local relation to it.
complement and adjunct te-clauses. We can use the analysis introduced above to
clauses inside have been studied to a certain extent (e.g. Kageyama 1993, Kishimoto
2006), there is an important fact that has eluded attention and a theoretical analysis;
complement and adjunct te-clauses behave differently in the manner described in (28),
(28) Fact 2
35
(29) Distribution of no in kata-nominalizations containing adjunct te-clauses
morai-kata]
get-way
In (29a), the te-adjunct is no-marked, but the object piza ‘pizza’ embedded in the te-
clause is not. (29b) represents an opposite pattern; while the adjunct te-clause itself is
not accompanied by no, the object inside it is no-marked. (29c) is minimally different
from (29a) in that the adjunct te-clause lacks no-marking, and (29d) is an instance in
which not only the te-adjunct but also the object inside it receives no-marking.
Among these four possibilities, only (29a) produces a well-formed string. The
following paradigm given in (30), on the other hand, represents four patterns of no-
36
(30) Distribution of no in kata-nominalizations containing complement te-clauses
morai-kata]
get-way
In (30a), the complement te-clause lacks no, whereas the object inside it is no-marked.
In (30b), the te-complement itself, but not the object inside it, is suffixed by no. (30c)
is a case where neither the te-complement nor the object inside it is accompanied by
no, and in (30d) both the te-complement and the object embedded in it is no-marked.
This time, only (30a) qualifies as a grammatical nominal expression among these four.
We can see from this fact that complement te-clauses are an exception to (20a) since
nevertheless.
proposed captures Fact 2 if complement V-te undergoes head movement all the way
37
(31) a. [NP[vP …[Complement TP … pizza-no tV tv tT ] tV tv ]
[N cook-v-te get-v-way]]
[N get-v-way]]
If complement V-te moves to the higher V, we predict that the verbal complex further
moves to N, and this series of head movements allows N to no-mark elements internal
to the complement TP, as shown in (31a). If adjunct V-te stays inside TP, we
Let me make some additional comments before leaving this subsection. Recall
that a prominent feature of the present analysis of the distribution of no is making use
of head movement to answer the following question: why is it that certain phrases (i.e.
vP, VP and complement te-clauses) neither bear no nor block no-marking into them?
There is an alternative answer to this question. Kishimoto’s (2006) work is one of the
have a feature (call it F) that must be licensed by the suffix -kata under c-command.
How does the feature-licensing approach answer the question above? As far as I can
tell, the answer seems to be that vP, VP and complement te-clauses do not have F.
This answer raises a next question: how can we determine, on the basis of empirical
data, whether an item has this feature or not, independently of the distribution of no?
38
make a prediction about the distribution of no without looking at the actual
adjunct te-clauses have an F while complement te-clauses do not, but this prediction
is hard to test against data other than the distribution of no. By contrast, the analysis
proposed here is free from this problem. This is so precisely because the gist of the
present analysis is that Fact 2 is tied to presence or absence of V-te movement, and
looking at the mobility of a te-clause of the relevant kind, for example. Thus, I
In sum, I have shown that the analysis of kata-nominals introduced above and
Fact 2 lead us to conclude that complement V-te, unlike adjunct V-te, moves to the
matrix clause.
The claim that the T head of a complement te-clause undergoes movement can be
further reinforced by its behavior in Argument Ellipsis (Oku 1998; see also Takahashi
2008 for a state of the art summary). Of importance here is the fact in (32), which can
be confirmed by the sheer unacceptability of (33b) on the one hand and the
39
(32) Fact 3
I-Top get-Neg-Past
The dative argument Ziro-ni, the adjunct mayoneezu-de ‘with mayonnaise,’ and the
accusative argument susi-o, all of which are associated with tabe-te ‘eat-te,’ are
missing in both (33b) and (34b). Such omission is possible only if the V-te remains
40
overt, as in (34b). McCawley and Momoi (1986) make a similar observation based on
question-answer pairs. They point out that V-te cannot be omitted in answers in b-
sentences in (35)-(36).
morai-masi-ta ka?
get-Pol-Past Q
‘Yes, I did.’
‘Yes, I did.’
clauses. In other words, Argument Ellipsis applies to sister nodes of predicates that
select te-clauses.
41
The apparent ‘unelidability’ of the te-clause in (33b) is exactly what the
present analysis predicts; since the V-te has moved to the next higher clause, it does
not reside in the constituent that can be deleted by Argument Ellipsis any longer. On
the other hand, we predict that Argument Ellipsis of the TP produces an acceptable
output if the V-te remains overt as a remnant. Under my analysis, when Argument
Ellipsis applies to the te-clause, the arguments and adjunct inside the te-clause are
deleted but the V-te is obligatorily stranded, because of head movement to the higher
The well-formedness of the elliptical clause in (34b), which exhibits what I call V-te
6
One might argue that the well-formed string in (34b) obtains by eliding the
embedded constituents Ziro, with mayonnaise, and sushi separately, rather than
eliding the TP as a whole. Such a possibility is, however, untenable because it
necessitates ellipsis of the adjunct with mayonnaise: as discussed in some detail in
chapter 3, adjuncts in Japanese cannot be independently targeted by ellipsis. See
sections 3.4 and 3.7 for unpronounced adjuncts.
7
As Tomohiro Fujii has correctly pointed out, one tacit assumption behind my
account of Fact 3 is that TP complements of the predicates in question are susceptible
to Argument Ellipsis. I take this to be the null hypothesis, for Argument Ellipsis
applies to NP complements of these predicates (see footnote 2) without any problem;
having (ia) as its antecedent, not only (ib) but also (ic), in which the object NP ringo-
o ‘apple’ of the verb moraw ‘get’ is left pronounced, is perfectly grammatical.
42
Interestingly, if we take the kind of complement clause whose head does not
move to the next higher clause, Argument Ellipsis of the entire clause causes no
meizi-ta kedo,
order-Past but
(Tanaka 2008:14)
So long as we can maintain that this assumption is the null hypothesis, it can be
concluded that the set of data concerning Argument Ellipsis in (33)-(34) does not
contradict the prediction of the head movement analysis of V-te at least, whereas the
asymmetry between (33) on the one hand and (i) on the other seems to be unexpected
if V-te of a te-complement does not undergo movement. Other things being equal, the
proposed analysis therefore should be favored over such an alternative analysis which
does not incorporate head movement of V-te on empirical grounds.
43
That no head movement takes place out of yooni-clauses can be confirmed by using
meizi-ta.
order-Past
are closely correlated with each other. In this regard, the head movement derivation
I have so far established that the string vacuous head movement analysis of te-clauses,
why head movement of the V-te complex is required in complement cases in (4) but
not possible in adjunct cases in (5), an important question that I have not so far dealt
with.
44
(5) Taro [Adjunct te-clause Ziro pizza [cook-te]] money get-PAST
I propose that head movement of V-te is triggered by predicates that take te-
complements. To be more precise, I assume that when such predicates select te-
clauses as their internal argument, they enter the derivation with an uninterpretable
tense feature [uT], and this uninterpretable feature has an ‘EPP property’ that triggers
head movement.8 Following Pesetsky and Torrego (2001), I assume that an EPP
property of a feature is a subfeature of that feature. The verb moraw ‘get,’ one of the
predicates that allow te-complements, for example, has the following two lexical
What is indicated by (41) is the following: when moraw takes a te-clause (TP) as its
internal argument it is accompanied by the [uT] feature with the EPP property (41a),
whereas when the verb selects a source NP and a theme NP, it is stipulated that the
[uT] feature is simply absent (41b). Head movement of the complex T head from the
te-clause, which has an interpretable tense feature, is required to satisfy the EPP
property of [uT] and eliminate that uninterpretable feature. These features, if not
checked, lead the derivation to a violation of Full Interpretation. The relevant stages
8
For various proposals about the relationship between head movement and the EPP,
see Alexiadou and Anagnostopoulou (1998), Pesetsky and Torrego (2001), and
Roberts (2010).
45
of the derivation of the representative case of te-complements in (42) are sketched in
(43a) illustrates that the verb get enters the derivation with [uT], and the te-clause is
merged with the verb as its internal argument. Next, the EPP property of [uT] triggers
overt head movement of the V-te complex to the higher verb in a cross-clausal
As for the absence of head movement out of adjunct te-clauses, the analysis
runs as follows. The first case to consider is the case in which the matrix predicate
does not have the [uT] feature, i.e. (41b). In this case, V-te does not move simply
because nothing triggers the operation, and te-adjuncts attach to the matrix clause
freely. In the ‘complement’ case, i.e. (41a), one might ask why the following does not
happen: the head of the te-adjunct moves to check the EPP property of the matrix
predicate whereas the head of the te-complement stays inside the clause, as in (44).
46
(44) [Adjunct … ti] [Complement … V-te] V-tei+get[uT]
Given this configuration, the impossibility of V-te movement out of the adjunct te-
clause follows if movement obeys the Adjunct Island Condition. Thus, under no
This subsection has been devoted to instantiating the present analysis of the
theory.
In this section I take up three potential alternative accounts of te-clauses which do not
postulate head movement of the V-te complex. It is shown that none of them is
empirically sufficient.
One alternative analysis has to do with the notion of phase, so let us call it the
phases and the latter are not. Given that te-adjuncts are mobile but te-complements
are not (see Fact 1 in (11) above), it would be tempting to hypothesize that adjunct te-
(2004) idea which maintains that te-clauses are invariably TPs. More concretely, in
this analysis the complement te-clause in (1) and the adjunct one in (3), repeated
below, are analyzed as a non-phase (being a TP) and a phase (being a CP),
respectively,
47
(1) Taro-ga [Ziro-ni piza-o tukut-te] morat-ta.
apparently resist being elided (see Fact 3 in (32)) if Bošković (2014) is right that only
analyzed as TPs, are neither phases nor complements of phase heads and therefore
asymmetries between two kinds of te-clauses would boil down to the TP/CP
dichotomy, which would consequently eliminate the need for head movement of V-te
(28) above) is taken into consideration. Recall that te-adjuncts are no-marked in kata-
phase-based analysis would have to propose that only phases (including adjunct te-
clauses, under this analysis) serve as barriers for no-marking and receive no for some
reason (whatever it is). This is, however, at odds with the very fact observed in (20c)
that a vP, which is commonly conceived as a phase, is not a barrier for no-marking in
48
nominalizations. While this analysis can be probably rescued by assuming that head
movement of a phase head X may change XP into a non-phase (see den Dikken 2007,
than the analysis I have proposed, according to which transparency of vPs and te-
least unclear whether the phase-based analysis can adequately account for the facts
Let us turn to the second alternative, which is concerned with the process
restructuring complements, and this line of analysis in fact has been proposed in the
literature: complement te-clauses, in particular those of motion verbs, fall under the
immobility and unelidabilty (see Rizzi 1978, Cinque 2006:13ff, 36ff, Wurmbrand and
Germanic languages). Facts 1 and 3 thus can be taken to suggest that te-complements
incompatible with our proposal. In fact, restructuring has sometimes been analyzed in
terms of head movement (Kayne 1989). The important question for us is not whether
49
whether saying that te-complements are ‘reduced’ is sufficient to explain all the
properties seen above. To be more concrete, let us take two major clause-reduction
particular, Wurmbrand (2001) and Takahashi (2012) for attempts to deal with
te-adjuncts accounted for, by saying (i) that te-complements in general are merely vPs
constitute full clauses but are mere complements of functional heads contained in a
single clause (as instances of functional restructuring)? Immobility (i.e. Fact 1) and
unelidability (i.e. Fact 3) may be, as noted above, accounted for under either
hypothesis. However, the data points involving nominalizations (i.e. Fact 2 and the
general properties of the construction given in (20)) seem hard for these approaches
approach that does not appeal to head movement has to offer an adequate analysis of
that vPs, VPs and te-complements are all transparent for no-marking, assuming head
movement for vP and VP on one hand and restructuring for te-complements on the
other to capture this transparency effect gives rise to redundancy. As long as that has
50
not been done yet, it seems that we have no reason to discard the head movement
for now that even if te-complements are syntactically ‘reduced,’ any adequate
analysis of the data has to incorporate head movement. (See also Nakatani 2004:50
for arguments against the idea that te-complements in general are restructuring
complements.)
(45), for instance, is analyzed as making up a complex head from the beginning,
which is formed either lexically (Kitagawa 1986: Ch. 1, Sells 1995) or syntactically
(Saito and Hoshi 1998, Hoshi 2002, Saito 2012, see also Shimada 2007 and Tonoike
2009), as in (46a). The object piza ‘pizza’ is merged with this complex head, rather
(46) a. [[eat]-Caus]
b. [pizza [[eat]-Caus]]
The base-generation approach has been proposed in various forms in the literature,
and it is not clear whether any version of it actually predicts that V-te and a higher
51
predicate are always dominated by a single node in syntax before anything else is
introduced into the derivation, just like causatives. But for the sake of discussion, let
us assume that they are, and see how such an analysis goes. Under this alternative,
tukut-te morat ‘cook-te get’ in (1) above would be analyzed as in (47a), and the
embedded object pizza would be merged with this comlex head, as in (47b).
As is obvious, the V-te complex and the embedded object pizza do not form a
movement of V-te.
However, this very advantage of accounting for Fact 1 raises a problem for the
52
(48) a. Taro-wa [Ziro-ni mayoneezu-de susi-o
I-Top get-Neg-Past
I point out that the base-generation analysis seems to have difficulties in legitimately
deriving (48c). Under this analysis, the derivation of (48c) is assumed to proceed in
the steps given in (49). First, we have a complex head made up of tabe-te ‘eat-te’ and
moraw ‘get’ in (49a). The (embedded) object susi ‘sushi’ is merged with this head, as
mayonnaise,’ and the (embedded) subject Ziro, as shown in (49c-d). (Labels of the
constituents formed by merge of these dependents of eat-te are not important in the
following discussion, so they are simply labeled as α, β and γ just for convenience.)
53
c. [β with mayonnaise [α sushi [[eat-te] get]]]
The problem here is that there is no constituent which exclusively dominates the
dependents of eat-te introduced into the derivation in (49b-d). Given the widely held
view that ellipsis applies only to constituents on the one hand and the (partial)
constituent structure in (49d) on the other, it follows that the analysis of this sort fails
to rule in (48c), which is far from satisfactory. (As mentioned in footnote 6, the
elliptical clause in (48c) cannot be analyzed as a result of eliding the three dependents
the complex head dominating eat-te and get somehow moves out of γ, but this
appears to be essentially the same as the head movement analysis of V-te, which
makes the base-generation analysis no more successful than the analysis I have
approach does not qualify as a compelling alternative to the analysis developed in this
chapter either.
I have so far justified the head movement analysis of te-clauses on various empirical
grounds. In this section, we turn to a theoretical issue about head movement and its
relevance to morphology. It is shown that the result of head movement of the V-te
complex is somewhat exotic in the sense that it does not feed word formation, and
54
Matushansky proposes that so-called ‘head movement’ is not an atomic
head X to a local domain of another head Y in syntax on the one hand, and a post-
syntactic operation that combines X and Y into a single morphological unit (‘m-
merger’ in her terminology) on the other hand. According to her proposal, ‘head
applies to X and Y under adjacency, which collapses the two into one word.
b. Movement of X in syntax
[X Y [WP tX ZP]]
What is particularly important about Matushansky’s proposal for the current purposes
and its host in morphology. Since syntactic head movement and m-merger are distinct
display one without the other. Matushansky argues that Danish definiteness marking
55
in (52) below presents an environment in which m-merger of two syntactic heads
(52) a. hest-en
horse-Def
‘the horse’
b. *den hest
Def horse
(Matushansky 2006:88)
The examples in (52) demonstrate that in Danish the definite suffix -en appears when
the noun is bare (i.e. it is not modified), and the definite article den is used when the
noun is modified. She argues that what triggers lexical insertion of hest-en is not N-
morphology, and claims that N and D are in-situ in syntax; were N-to-D movement in
(52a), because røde ‘red,’ being an adjunct, is not predicted to block N-to-D
56
(53) a. No N-to-D movement in syntax
[DP D [NP N ]]
hest-en
What I would like to suggest in the remainder of this section is that te-complements
In order to see what m-merger does in Japanese, let us first consider complex
wa and -sae cannot show up between component predicates tukuri ‘cook’ and hazime
Let us assume that if two morphemes next to each other allow an adverbial particle to
be inserted in between the two, then there is a morphological word boundary between
the two morphemes, and if they do not, they form a morphologically inseparable unit
(see also Sugioka 1984 and Matsumoto 1996 for the relation between focus particles
57
and wordhood). Given this diagnostic for morphological wordhood, the sequence of
verbs in the complex predicate in (54) makes up a single morphological word. Let us
further assume that m-merger is responsible for this tight connection of the two verbs.
Then, the derivation of the complex predicate in (54) will be (55) below.
(55) a. Syntax
let us go back to te-clauses. Recall that the analysis of complement te-clauses I have
so far argued for is (56): the V-te complex undergoes head movement to a higher verb.
Importantly, the V-te can be freely separated from the higher predicate by an
which means in the present context that the V-te and the higher verb do not form one
9
Shibatani (2009) explicitly states that although the V-te complex and the higher verb
make up a ‘phonological word,’ they do not form a morphological word, showing no
morphological integrity, whereas component predicates in V-V compounds (see (54))
make up words in both phonological and morphological senses.
58
(58) Taro-ga [Ziro-ni piza-o tukut-te] {-wa/-sae}
morat-ta.
get-Past
acceptability of (58) suggests that although the V-te complex undergoes head
movement to the higher verb in syntax, the two heads do not get collapsed into one
b. No m-merger in morphology
… [[V-te] V] …
10
The correleation between adverbial particles and head movement (verb movement,
in particular) has long been under considerable debate: see chapter 4. As for the
derivation of (58), I tentatively assume that (i) the adverbial particle is adjoined to -te
(viz. T), and (ii) head movement of the V-te pied-pipes the particle on its way to the
higher predicate.
59
Hence, I would like to conclude that the case of te-complements in Japanese displays
the V-te complex that has undergone head movement and the higher verb do not
2.5 Conclusion
derived by string-vacuous head movement of the V-te complex. I have argued that the
that te-clauses display, and have proposed that movement of the V-te complex is
triggered by the uninterpretable tense feature [uT] that the higher predicate bears.
examined, and I concluded that none of them is as successful as the proposed analysis.
Also, I have shown that the result of head movement of V-te is more or less exotic in
that the V-te complex and the higher verb do not get spelled out as a single word,
60
Chapter 3: Predicate Raising within a Single Clause in Japanese
3.1 Introduction
In chapter 2, I have empirically motivated the claim that the grammar of Japanese
must be equipped with head movement, albeit string-vacuous and therefore inaudible,
by investigating te-clauses. The analysis proposed and defended there is that in the
clausal boundary (i.e. T-to-V movement). In this chapter I turn to head movement
within a single clause, and discuss verb movement in Japanese, a subcase of head
movement which has probably been most intensively disputed in previous works, as
the grammar of Japanese, as researchers such as Otani and Whitman (1991) and
Koizumi (2000) have maintained. I provide empirical support for this position by
This chapter is organized as follows: in section 3.2 I first introduce two kinds
of verbal predicates. In section 3.3 and 3.4 it is demonstrated that they behave
differently under ellipsis phenomena. In section 3.5 I propose that the set of ellipsis
facts in question should be captured by vP-ellipsis plus PR. Section 3.6 discusses
61
3.2 Two classes of verbal predicates: NJVs and VNs
Verbal predicates in Japanese can be divided into two classes, at least: native
Japanese verbs (NJVs) on the one hand and what Martin (1975) refers to as verbal
nouns (VNs) on the other hand1. To date, VNs in Japanese have been extensively
studied by a number of scholars in various ways (see Kageyama 1977, 1982, 1993,
Miyagawa 1987a, Iida 1987, Grimshaw and Mester 1988, Poser 1989, Tsujimura
1990, 1992, Terada 1990: Ch. 3, Manning 1993, Matsumoto 1996, Dubinsky 1997,
Saito and Hoshi 1998, 2000, Takahashi 2000, and references cited therein). I first
briefly review some of their descriptive properties that are not shared by NJVs.
In (1), the predicate is an NJV ki ‘come,’ while the one in (2) is a VN kikoku
‘returning.’
1
Predicates of VN clauses are often referred to as Sino-Japanese verbs as well
(Kuroda 1965, Kageyama 1977). As this term indicates, the great majority of VNs,
including kikoku in (2), are of Chinese origin, but there are VNs that are of English
origin and of Japanese origin as well. See Kageyama (1982), Poser (1989) and
Takahashi (2000).
62
One obvious difference between (1) and (2) is that in the former the past tense suffix -
ta is directly attached to the NJV, whereas in the latter it is carried by su ‘do,’ which
itself is an NJV (and spelled out as si when followed by -ta). As often discussed in the
literature, when they appear as predicates in clausal domains, VNs are unable to
directly support verbal affixes on their own, contrary to NJVs (Kageyama 1977, 1982,
Takahashi 2000). For example, Takahashi (2000) gives (3) as a partial list of verbal
-na ‘negation’
-(s)ase ‘causative’
-(r)are ‘passive’
-(rar)e ‘potential’
-ta ‘desiderative’
-tei ‘progressive’
(Takahashi 2000:28)
63
This being so, if su in (2) is removed, we get (4), in which the tense suffix happens to
It should be remarked in this regard that su cannot be employed in NJV clauses in the
ungrammatical strings obtain, as shown in (6) and (7) (see Kuroda 1965 and Aoyagi
2006 for related discussion). I will return to this point in section 3.8.
64
(7) Taro-ga Nihon-e ko-(*si)-nakat-ta.
Another well-acknowledged fact about VNs is their ability to form nominals without
the help of nominalizing suffixes (hence the name). For instance, the VN kikoku
heads a noun phrase in (8), which receives the nominative Case -ga.
NJVs, on the other hand, are clearly unable to create nominals on their own, as the ill-
nominalizing suffixes such as -kata ‘way’ are employed, as in (10).2 See section 2.3
2
What Poser (1989) refers to as renyoomeisi can be regarded as nominals derived by
NJVs without overt affixation. His description of renyoomeisi is that it “consists
segmentally of the bare verb stem” (Poser 1989:14-15), which is exemplified by the
well-formed nominal (ib), derived from the NJV kari ‘borrow’ in (ia).
(i) a. kari-ru
borrow-Pres
‘to borrow’
b. kari
‘borrowing’
65
(9) *[Nihon-e-no ki](-ga/-o)
Japan-to-no come-Nom/-Acc
ki-kata]-o osie-ta.
come-way-Acc tell-Past
NJVs and VNs is robust. I do not attempt to offer an account of the differences
mentioned above, since it is not my primary concern. Rather, in this chapter I show
that another discrepancy between these two classes of predicates emerges in ellipsis
contexts, and then argue that their behaviors in ellipsis phenomena empirically
assume, throughout this work, that the predicate in (2) is the VN kikoku, rather than
after (bare) VNs is a supporting verb (or a dummy verb), which does not participate
in θ-marking (see Grimshaw and Mester 1988 for related discussion). This
As Poser notes, however, the number of possible renyoomeisi is not infinite, and its
formation is far less productive than nominalization triggered by -kata, which is
100% productive.
66
(11) a. Taro-ga [PP L.A.-kara] kikoku si-ta.
The point here is that the source PP L.A.-kara ‘from L.A.’ is permitted in (11a), but
excluded in (11b). This sharp contrast does not seem to be predicted if one analyzes
su as a predicate, because both examples involve su, differing only in the choice of
the VN. Moreover, exactly the same pattern can be observed even in nominal
This strongly indicates that the availability of from L.A. in (11) and that of (12) are
governed by one and the same factor. That factor then cannot be su since it is
67
employed in (11) but not in (12). I thus conclude that the VNs, rather than su, qualify
as predicates in (11), and suggest that the VN sentaku ‘washing' is unable to take the
well.
su-verbs that are characterized as ‘truly lexicalized’ by Poser (1989) like those in (13).
(13) a. ai-su-ru
‘to love’
b. nes-su-ru
‘to heat’
At first blush, they might look more or less similar to what I refer to as VNs here such
as kikoku in (2), in that (i) they have Sino-Japanese nominal part ai ‘love’ and netu
‘heat’ right before su as well (see footnote 1), and (ii) nominal expressions in (14) are
well-formed.
mother-no love
‘a mother’s love’
sun-no heat
68
However similar they initially look, such predicates do not behave like VNs in a
Dubinsky (1997). For instance, while su following VNs can be replaced by the
One difference between NJVs and VNs I would like to discuss has to do with
predicate ellipsis, a descriptive property of which can be stated as Fact 1 in (16) (see
69
(16) Fact 1
To see the details of (16), let us first take a look at (17), where the NJV ki is
employed. (17a) is the antecedent clause for the clauses in (17b-c). I refer to elliptical
Ziro-Top come-Neg-Past
Ziro-Top Neg-Past
Both target clauses contain a gap in the object position (indicated by [Obj e]), and they
minimally differ with respect to the presence of the NJV in the surface string. While
the target clause in (17b) with the NJV pronounced is well-formed, the one in (17c),
which contains another gap in the position where the NJV ko is supposed to appear
70
Clauses with VN predicates, on the other hand, strikingly contrast with those
with NJVs in the relevant respect. The VN clause (18a) is the antecedent for the target
clauses in (18b-c). Again, there is a gap in the object position in both target clauses,
and the only difference between the two is that the VN kikoku is pronounced in (18b)
but omitted in (18b), as indicated by [VN e]. Omission of the VN in addition to the
Ziro-Top do-Neg-Past
Contrary to what we have seen in (17) with the NJV, the VN may be freely omitted
Kageyama (1982) and Poser (1989). We can thus see from (17) and (18) that
At this point, one might argue that (17c) is bad for the same reason as (19b)
below is. What (19b) shows is that when the adverbial particle -sae ‘even’ is attached
71
to the verb, the negation nakat ends up in the position where it is not linearly adjacent
such an ill-formed string can be ameliorated by inserting su right before the negation,
Similarly, the negation in (17c) has no adjacent verbal stem due to the omission of the
NJV, whereas in (18c) the adjacency requirement of the negation can be satisfied by
su even after the omission of the VN. Hence, if it turns out that what makes (19b) bad
is also responsible for (17c), then the grammatical status of (17c) may not have
anything to bear on the issue of predicate ellipsis per se, since in (19b) the predicate
ki is not omitted.
(17c) should not be equated with that of (19b). Importantly, the clause in (20), which
72
is derived from (17c) by adding su before the negation, is still bad as an elliptical
clause anteceded by (17a) (see Takahashi 2000 and Hoshi 2009 for related
observations).
Ziro-Top do-Neg-Past
This fact shows that (17c) cannot be remedied in the same way as (19b), which seems
requirement the negation imposes, just like (19b) is. Given this observation, I
conclude that what we have seen in (17) indeed has to do with predicate ellipsis.
73
(21) Funakoshi’s generalization (FG)
In Japanese, adjuncts can be null only if the clause-mate object (or other
(Funakoshi 2014:302-303)
Although I believe that FG is essentially correct, I show that FG holds for NJV
Let us quickly review what has been understood about null adjuncts in the
literature, before examining the core data. One consensus is that adjuncts, contrary to
arguments, can never be independently omitted (Oku 1998, Takita 2011). (22)
represents a case of elision of arguments in NJV clauses; when anteceded by (22), the
target clause (22) can be naturally understood as meaning Ziro didn’t come to Japan,
3
It should be mentioned that this is a preliminary version of his generalization.
Funakoshi provides a more accurate one as the final version on the basis of facts
about intransitive clauses and contrastive focus. See Funakoshi (2014: Ch. 6) for
details. Since neither intransitive clauses nor contrastive focus will be discussed in
this chapter, I take (21) as FG here.
4
Simpson, Choudhury and Menon (2013:111ff) report that a similar generalization
holds for unpronounced adjuncts in Bangla, Hindi and Malayalam. I am grateful to
Kenshi Funakoshi for bringing their work to my attention.
74
b. Ziro-wa [Obj e] ko-nakat-ta.
Ziro-Top come-Neg-Past
On the other hand, it is not possible to interpret (24b) as an elliptical clause in which
the adjunct L.A.-keeyude ‘via L.A.’ is understood though not pronounced; in other
75
(25) a. Ziro didn’t come to Japan anyhow.
(26) entails that Ziro came to Japan, which clearly contradicts the reading in (25a).
observation is that adjuncts can be null if clause-mate objects are also null. Let us
begin with (27) below, where the NJV ki is used; when anteceded by (27a), the target
clause (27b) allows the interpretation (25b), despite the fact the adjunct L.A.-keeyude
is not pronounced there. Thus, (27b) can be followed by (26) with no contradiction.
Ziro-Top come-Neg-Past
76
I refer to the interpretation in which the adjunct is implicitly understood (like the one
availability of null adjuncts in NJV clauses. However, I would like to point out that
clauses, which Funakoshi seems to have overlooked. The hitherto unnoticed fact is
(28) Fact 2
In VN clauses, null adjuncts are possible only when VNs as well as clause-
In order to see null adjunct phenomena in VN clauses, let us examine (29), where the
VN kikoku is employed. Crucially, when anteceded by (29a), the target clause (29b)
with an object gap does not yield the null adjunct reading Ziro didn’t do returning to
Japan via L.A.. In order to make that reading available, the VN predicate kikoku, in
si-ta kedo,
do-Past but
77
b. Ziro-wa [Obj e] kikoku si-nakat-ta.
Ziro-Top do-Neg-Past
This discrepancy between NJVs and VNs concerning the null adjunct reading can be
with (30), whereas a perfectly coherent utterance results when (29c) is followed by
(30).
kikoku si-ta.
returning do-Past
The same point can be made by the following dialogue. Under the context given in
78
(31) (Context: Ziro used to live in Japan, but now he lives abroad. He went back to
of the VN dramatically affects the availability of the null adjunct reading, let me add
one more set of examples in (32). In this case, the VN san-oofuku ‘three laps,’ which
79
(32) a. Taro-wa ikitugi-naside [nijuugo-meetoru puuru-o]
‘Taro swam three laps of the 25-meter pool without breathing, but…’
Ziro-Top do-Neg-Past
‘Ziro didn’t.’
(32a) is the antecedent for the target clauses in (32b-c). Both target clauses contain an
object gap that corresponds to nijuugo-meetoru puuru ’25-meter pool.’ Again, the
crucial point is that the target clause in (32b) with the overt VN unambiguously
receives the interpretation in (33a), whereas the one in (32c) can be readily
understood as meaning (33b). (If you swim three laps in a 25-meter pool, you swim
150 meters in total.) That is to say, only (32c) can be understood as having the
b. Ziro did swim 150 meters, but unlike Taro, he took a breath while he
was swimming.
80
The difference regarding the null adjunct reading can be confirmed by the following
situation. A coach ordered both Taro and Ziro to swim three laps of the 25-meter pool
without breathing. While Taro swam 150 meters in the way the coach had told him to
do so, Ziro happened to forget exactly what the coach had ordered. So he did swim
150 meters but he took a breath. In this situation, (32c) can be true but (32b) is false,
Facts like these tell us that object drop is not sufficient to make the null
recalled here is that the omission of VNs is only optional, as we have already seen in
(18) above. The question is, then, why the VN must be omitted along with the object
to derive the intended null adjunct reading in (29), (31) and (32).
I argue that behaviors of NJVs and VNs under ellipsis should be attributed to their
asymmetry with respect to predicate raising (PR)5. To be more specific, I propose that
while NJVs undergo PR to v in overt syntax on the one hand, VNs are not subject to
5
I eschew the term verb movement here because the categorial status of VNs seems
contentious, due to the ‘mixed’ properties of VNs I have introduced in 3.2. While
Hasegawa (1991) and Takahashi (2000) argue that they are Vs, Miyagawa (1987) and
Terada (1990) propose to analyze them as Ns. If VNs are actually Ns rather than Vs,
then it would make no sense to say ‘verb’ movement applies (or does not apply) to
VNs. Identification of the categorial status of VNs is far beyond the scope of this
study, and I use the term predicate raising in lieu of verb movement, because it
sounds more neutral and therefore can be used without bringing in such complications.
81
PR on the other, as schematically shown in (34) and (35) below6. I assume that v
overtly raises to the next higher head F in both NJV and VN clauses. Therefore, NJVs
are located outside, but VNs remain inside, vPs. Furthermore, I also assume that the
Let us first see how Fact 1 can be handled by making reference to PR. The
point was that NJVs cannot, but VNs may, be omitted along with objects; the relevant
Ziro-Top come-Neg-Past
6
By F in (34) and (35), I mean various heads that take a vP complement, including
(but not limited to) T and Neg.
82
c. *Ziro-wa [Obj e] [NJV e] nakat-ta.
Ziro-Top Neg-Past
Ziro-Top do-Neg-Past
(36)-(37):
I propose that predicate ellipsis under discussion involves vP-ellipsis, and the
(36) and (37). Under the current analysis, the form (36) simply cannot be derived by
vP-ellipsis. This is so because NJVs do not stay inside vPs due to PR, and therefore
83
they cannot be elided by virtue of vP-ellipsis, as sketched in (38) (elided material is
represented by strikethrough)7.
By contrast, vP-ellipsis in VN clauses successfully gives rise to the form (37). Since
VNs do not undergo PR and hence remain inside vPs, they can be left unpronounced
observation was that NJV and VN clauses differ in how they treat null adjuncts. The
case of NJV clauses is illustrated again by (39); the intended null adjunct reading Ziro
didn’t come to Japan via L.A. is unavailable unless the object is dropped.
7
As far as I am aware, the gist of the proposed analysis is not affected in any
significant way whether ellipsis is conceived as PF-deletion (Sag 1976, Merchant
2001) or as LF-copying (Williams 1977, Fiengo and May 1994). Although I indicate
vP-ellipsis in terms of PF-deletion here, it should be emphasized that this is for
expository reasons only, and that I am not arguing for the PF-deletion approach to
ellipsis.
84
b. Ziro-wa Nihon-e ko-nakat-ta.
Ziro-Top come-Neg-Past
My account of this contrast runs as follows. First of all, (39b) cannot be derived
through vP-ellipsis, because the object is pronounced. It should be recalled here that
structure of (39b) must be (40) below, in which the adjunct via L.A. is not generated
from the start. If we follow the standard assumption that (meaningful) elements can
be interpreted in LF if and only if they are structurally present (but see Ginzburg and
Sag 2000), then the lack of the null adjunct reading in (39b) straightforwardly follows.
On the other hand, (39c) can be assigned the structure in which the adjunct in
question is implicitly present. vP-ellipsis renders the adjunct as well as the object null,
while stranding the NJV located outside the vP due to PR, as depicted in (41). The
8
This account is virtually the same as Funakoshi’s (2014) explanation of FG in (11),
although he is not concerned with the distinction between NJV and VN clauses, as I
have already mentioned.
85
(41) … [NegP [vP tZiro via L.A. [VP to Japan tcome ] tv ] come-v-Neg] …
What about null adjuncts in VN clauses? The fact to be accounted for is that
the null adjunct reading ‘Ziro didn’t do return to Japan via L.A.’ is absent in (42b),
although the object is dropped there. For that particular reading to obtain, the VN as
si-ta kedo,
do-Past but
Ziro-Top do-Neg-Past
I argue that (42b), albeit elliptical, is not derived by vP-ellipsis. This argument holds
precisely because the fact that the VN, to which PR does not apply, is pronounced
necessarily entails the absence of vP-ellipsis under the current analysis. Hence, (42b)
86
is unambiguously assigned the structure in (43), in which the adjunct via L.A. is not
generated to begin with (recall the ban on adjunct drop in (24)). Given the structure in
As for (42c), on the other hand, I claim that it allows the intended reading because it
has the structure in which the adjunct is indeed generated. The adjunct is elided by
vP-ellipsis along with the object, just as in the case of the NJV clause (39c). However,
in this case vP-ellipsis obligatorily elides the predicate as well, since the VN, contrary
to the NJV, remains inside the vP, as depicted in (44). We can thus capture the
(44) … [NegP [vP tZiro via L.A. [VNP to Japan returning] tv ] v+su-Neg] …
To summarize the discussion so far, I have proposed that the set of data
9
As indicated by strikethrough in (43), the null object in (42b) is analyzed in terms of
Argument Ellipsis in the sense of Oku (1998), though the present analysis is not
affected if we assume that the null object in question is simply the empty pronoun pro.
87
3.6 A potential alternative: Miyagawa (1987a)
I have argued that su-stranding in VN clauses like (45b) involves predicate ellipsis,
and proposed that the empty category [e] in (45b) corresponds to the vP.
Ziro-Top do-Neg-Past
some detail.
(1982). Here, the VN hatugen ‘speaking out’ appears before su, and the second clause
88
(46) Gakkai-de, amerikajin-wa yoku hatugen su-ru ga,
‘At academic meetings, Americans always speak out, but the Japanese seldom
do.’
(Kageyama 1982:247)
Miyagawa proposes that the underlying source for the second clause in (46) is (47), in
which the VN is marked with accusative Case -o. As for the status of su, he claims
that it serves as the predicate of the clause, taking the o-marked VN as its object.
From this underlying structure, which is perfectly grammatical, the second clause of
(46) obtains by deleting the o-marked VN in the object position. Therefore, under
Miyagawa’s analysis, the gap in (46) corresponds to the object, as shown in (48).
89
His analysis is plausible to the extent that (i) VNs can be Case-marked and function
as arguments and (ii) the lexicon of Japanese seems to possess the verb su which is
associated with its own θ-grid (‘free-occurring suru’ in Miyagawa’s term), apart from
the supporting verb su. As we have already seen in section 3.2, the VN in (8),
argument.
The example in (49) can be taken to illustrate that su acts as the predicate of the
clause here, which more or less resembles to the English main verb do.
Thus, it seems that we have no reason to immediately reject his analysis of su-
stranding.
What should be remarked at this point is that for Miyagawa, su-stranding does
not involve predicate ellipsis at all. This is so precisely because he regards su as the
90
analyzed as an object gap. Hence, if his object ellipsis analysis of su-stranding is
correct, we can no longer maintain that NJV and VN clauses behave differently with
stranding, and argue that the object ellipsis analysis is empirically inadequate. To be
Sato (2010) shows that some VNs give rise to idiomatic readings when they
appear in the VN su form, which are totally absent in the VN-o su form (see also
Barrie 2012). For instance, let us take (50), where the VN ryoori ‘cooking’ is
and the literal reading to cook, whereas (50b), in which the VN is o-marked, receives
cooking do-Pres
b. ryoori-o su-ru
cooking-Acc do-Pres
91
Likewise, Sato observes that in (51), where the VN bekyoo ‘study’ is used, the literal-
idiomatic ambiguity arises in (51a), while in (51b) the literal reading is the only
possibility10.
study do-Pres
b. benkyoo-o su-ru
study-Acc do-Pres
With this in mind, let us turn to (52). The point of interest here is that the target clause
ryoori, just like the antecedent clause (52a); it can be interpreted as Spider-Man
10
That (51a) can be understood as an idiomatic expression may not be recognized by
every native speaker of Japanese. I was not aware of its idiomaticity until recently,
and I have encountered a couple of speakers who are not familiar with its idiomatic
reading. On the other hand, many speakers have informed me that (51b) is not
ambiguous in the way (51a) is, which conforms to Sato’s observation.
92
(52) a. Hulk-wa teki-o ryoori si-ta kedo,
Spider-Man-Top do-Neg-Past
I would like to point out that the object ellipsis analysis Miyagawa advocates has
difficulties in accounting for this fact. Under his analysis, (52b) will be derived from
the underlying source in (53), in which the verb su takes the o-marked VN as its
object. If the VN in the object position is deleted, the surface string in (52b) results.
Crucially, however, (53) obviously lacks the idiomatic reading that (52b) has. The
for the fact in the following way. The underlying source of (52b) is (54), in which we
have the bare VN ryoori as the predicate, rather than the o-marked one as the object.
93
vP-ellipsis applies in the way illustrated in (55), which gives rise to the surface form
in (52b).
In this case, the idiomatic reading in question does obtain in the underlying source in
(54). Hence, its availability in the derived form in (52b) follows without additional
stipulation.
To sum up, I have shown that Miyagawa’s analysis of su-stranding faces the
problem of undergeneration in the case of (52b), being unable to derive the idiomatic
on the other hand, can accommodate the relevant fact in a principled way. It should
nevertheness be stressed that the above discussion does not necessarily entail that the
grammar does not resort to object ellipsis at all to derive su-stranding. Rather, it
suggests that the grammar must be able to derive su-stranding at least by vP-ellipsis,
The claim I have so far motivated on the basis of facts about predicate ellipsis and
null adjuncts is that NJVs and VNs behave differently than each other regarding
94
predicate raising (PR): NJVs are susceptible to PR on the one hand, whereas VNs are
not on the other. This analysis can be schematically depicted in (34)-(35), repeated
below.
In section 2.3 of chapter 2, I proposed that the mechanism behind no-marking is (56)-
marking takes place under government by the nominalizing suffix -kata ‘way.’
95
d. The Government Transparency Corollary (GTC): A lexical item that
Armed with this set of assumptions about no-marking, the proposed analysis
illustrated in (34)-(35) predicts that NJVs and VNs exhibit yet another discrepancy in
kata-nominalizations. That is to say, NJVs resist, but VNs must receive, no-marking
because the current analysis should derive the structures of kata-nominals given in
(58) and (59), if we continue to assume that -kata takes a vP as its complement (as
argued by Kishimoto 2006; see also Sugioka 1992) and therefore qualifies as a
member of F in (34)-(35).
Given the structure in (58), we predict that in the case of NJV kata-nominals, the
subject and object receive no (thanks to the GTC in (57d)), but the NJV itself is not
definition given in (57). On the other hand, we predict from the structure in (59) that
in the case of VN kata-nominals, the VN (as well as the subject) must be no-marked,
96
because the maximal projection VNP is overtly headed by the VN, which, contrary to
the NJV in (58), remains in situ. (I assume that the object in (59) can be no-marked
by the VN, because it is independently able to no-mark its object, as we have already
This prediction is indeed born out. First of all, the NJV kata-nominal (60) and
By contrast, the NJV kata-nominal given in (62), in which the NJV ki ‘come’ receives
no-marking, is sheerly ungrammatical. (Notice that (62) is still bad even if the
supporting verb su is added before the suffix -kata to provide it with a lexical support,
97
Similarly, the VN kata-nominal in (63) is bad, where the VN kikoku ‘returning’
The following examples from Poser (1989) and Kageyama (1993) illustrate the same
point: the VNs unten ‘driving’ in (64) and hassei ‘occurrence’ in (65) must be no-
marked:
driving-no do-way
11
In this regard, it is fair to mention that some speakers seem to accept the kata-
nominalization in which the VN is not no-marked like the one in (63). For instance, it
appears that Sugioka (1992) does not reject the following example, in which the VN
yuuwaku ‘temptation’ lacks no-marking.
Contrary to what these authors independently report, both (i) and (ii) sound utterly
unacceptable to my ear, unless the VNs are no-marked. Likewise, all native speakers I
have consulted so far uniformly rejected these examples. At this point, I am unable to
offer a satisfactory account of this variation among speakers, and I will put aside the
grammar of those who accept VN kata-nominals such as these.
98
b. *[NP unten si-kata]
(Poser 1989:15-16)
(Kageyama 1993:260)
summarized in (66).
This paradigm is not, to the extent that the mechanism of no-marking I adopt here can
NJVs from VNs with respect to PR. If both NJVs and VNs are subject to PR, then it
is predicted that not only NJVs but also VNs are exempt from no-marking. Hence, the
NJV kata-nominal (66a) and the VN counterpart (66c) are predicted to be equally
Likewise, if neither NJVs nor VNs are susceptible to PR, then the prediction is that
99
they both require no-marking; the NJV kata-nominal (66b) should be as good as the
VN counterpart (66d). This is a false prediction as well, since the former is not an
Given the current analysis, on the other hand, the paradigm in (66) can be
understood as a natural consequence. NJVs and VNs behave differently with respect
asymmetries regarding predicate ellipsis and null adjuncts: the former are raised by
PR in syntax but the latter remain in situ. Therefore, only the forms in (66a) and (66d)
The crucial observation made in section 2.4 of the last chapter that te-complements
I-Top get-Neg-Past
100
Also, I mentioned that the clause in (68b), in which tebe-te ‘eat-te’ is pronounced, is
On the basis of this observation, I argued there that when Argument Ellipsis applies
to a te-complement (e.g. the sister node of a predicate that selects a te-clause), the V-
Interestingly, the null adjunct reading discussed in section 3.4 obtains in V-te
stranding ellipsis as well. For instance, the elliptical clause (68b) can be understood
as (70), which is not expected unless the implicit counterpart of mayoneezu-de ‘with
101
(70) I did have Ziro eat sushi, but not with mayonnaise.
The availability of the null adjunct reading in question can be confirmed by the fact
tabe-te morat-ta.
eat-te get-Past
Notice that the predicate of the te-complement in (68) is an NJV, tabe ‘eat.’ What is
noteworthy here is the way the null adjunct reading arises in V-te stranding ellipsis
take a look at (72), where the VN san-oofuku ‘three laps’ is used as the predicate of
the te-complement.12 (72a) serves as the antecedent for target clauses in (72b) and
(72c).
12
In fairness, a near-synonymous NJV counterpart of the discourse in (72) is given in
(i).
102
(72) a. Taro-wa [Ziro-ni ikitugi-naside nijuugo-meetoru
‘Taro had Ziro to swim three laps of the 25-meter pool without
breathing, but…’
‘Ken didn’t have Ziro swim three laps of the 25-meter pool.’
‘Ken didn’t.’
The VN is overtly expressed in (72b) but is missing in (72c). Although both of them
are perfectly well-formed target clauses, they radically differ from each other with
respect to the intended null adjunct reading; importantly, while (72b) is interpreted
unambiguously as (73a), (72c) can be easily understood as (73b), which indicates that
When anteceded by (ia), (ib) allows for the null adjunct reading in (ii), and can be
truthfully uttered in a situation where Ken ordered Ziro to breath carefully in
swimming the pool, unlike Taro.
(ii) Ken did have Ziro swim the 25 meter pool, but not without breathing.
103
the null adjunct corresponding to ikitugi-naside ‘without breathing’ is structurally
(73) a. Ken didn’t have Ziro swim 150 meters in any way.
b. Ken had Ziro swim 150 meters, but not without breathing.
Suppose the following situation. Both Taro and Ken had Ziro swim three laps of the
25-meter pool. While Taro asked him to swim without breathing, Ken asked him to
take a breath because he wanted to see how smoothly Ziro can breathe when he
swims. In this situation, (72b) is rejected but (73c) can be true. Obviously, this is
I propose that this set of facts about the null adjunct reading in V-te stranding
ellipsis can be reduced to the analysis proposed in this chapter. In chapter 2, it was
assumed that a verb and -te (viz. T) in a te-clause make up a complex head by virtue
the verb is carried along as part of the derived head. The data concerning V-te
stranding ellipsis suggests that this is indeed true. I argue that the distinction between
as well: te-complements whose predicates are NJVs (which I refer to as NJV te-
whose predicates are VNs (which I call VN te-complements) looks like (75).
104
(74) NJV te-complements
(75) VN te-complements
manner as it does in the case of simple clauses discussed in section 3.5. As a result of
head movement of T out of the te-complement, which I have argued in chapter 2, the
NJV ends up in being outside the TP. On the other hand, the VN in (75) is not subject
to PR; it therefore remains inside the TP until the end of the derivation.
This structural difference with regard to PR enables us account for how null
adjuncts obtain in V-te stranding ellipsis in the following way. In the case of (68b),
structurally present.
[T eat-v-te]+get-Neg-Past
Argument Ellipsis of the TP in (76) elides the adjunct with mayonnaise (as well as
arguments associated with eat-te), while stranding the NJV eat, which results in the
surface string of (68b). Hence, the null adjunct reading arises in (68b).
105
(77) I [TP Ziro [vP with mayonnaise [VP sushi teat ] tv ] tT ]
[T eat-v-te]+get-Neg-Past
Let us turn to VN te-complements. The asymmetry between (72b) and (72c) in the
null adjunct reading can be accounted for as well. As for (72b), in which the VN san-
oofuku ‘three laps’ is pronounced, it lacks the intended null adjunct reading because
the TP is not elided. This is so because the overt VN automatically entails that the TP,
which dominates the VN, is not elided, given the structure in (75). Besides, we know
independently that adjuncts cannot be dropped (see section 3.4). So I propose that the
structure of (72b) looks like (78) below, in which the adjunct ikitugi-naside ‘without
breathing’ is not structurally present. The arguments Ziro and 25-meter pool are left
two (or they may be null pronouns). Then, it comes as no surprise that the null
(78) Ken [TP Ziro [vP [VP 25-meter pool three laps] tv ] tT ]
[T v+su-te]+get-Neg-Past
As for (72c), its compatibility with the null adjunct reading can be captured as
follows. The structure given in (75) above enables us to analyze (72c) as an elliptical
clause derived by Argument Ellipsis of the whole TP, since the VN is not pronounced
there. Then the representation in (79) can be assigned to (72c), in which the adjunct in
106
question is generated but elided along with the VN and the arguments as a result of
TP-ellipsis. Hence, contrary to (72b), (72c) allows the null adjunct reading.
(79) Ken [TP Ziro [vP without breathing [VP 25-meter pool three laps] tv ] tT ]
[T v+su-te]+get-Neg-Past
complements treat null adjuncts in essentially the same way; overt VNs invariably
block the null adjunct reading in both environments. This fact would be considered as
merely accidental if the way PR affects NJVs and VNs in simple clauses did not hold
13
As Tomohiro Fujii has critically pointed out, the asymmetry between NJV and VN
te-complements regarding null adjuncts examined here tells us little about the validity
of the head movement analysis of V-te proposed in chapter 2. This is so precisely
because the relevant contrast can be equally captured by vP-ellipsis within te-
complements, coupled with obligatory PR of NJVs (and the assumption that subjects
of te-complements can remain inside vPs). To be precise, we seem to have a way to
analyze (68b) and (72c) as being derived as in (i) and (ii), respectively. Notice that
neither involves head movement to the higher predicate get.
(i) I [TP [vP Ziro with mayonnaise [VP sushi teat ] tv ] [T eat-v-te]] get-Neg-Past
(ii) Ken [TP [vP Ziro without breathing [VP 25-meter pool three laps] tv ]
[T v+su-te]] get-Neg-Past
Embedded vP-ellipsis (indicated in (i) and (ii)) seems indistinguishable from TP-
ellipsis following head movement to get (indicated in (77) and (79)), especially in that
both strategies are expected to equally cover the range of null adjuncts inside te-
complements. The facts in (68) and (72) therefore cannot be taken to support the TP-
ellipsis derivation of (NJ)V-te stranding ellipsis. At the same time, he also mentions
that these facts do not contradict the prediction of the head movement analysis of V-te
either.
107
3.8 The trigger of predicate raising and the nature of su-insertion
is time to discuss its trigger. Recall that my account of the discrepancies between
NJV and VN clauses is that (i) NJVs must undergo PR to v while VNs are not subject
to PR, and (ii) in both NJV and VN clauses, v constantly raises up to a higher head F
which takes a vP as its complement. If this is indeed the case, then the question that
must be answered is that why only NJVs are susceptible to PR to v. To put differently,
what triggers PR in a rather selective manner so that it applies to NJVs but not to
VNs?
predicates. To be precise, I make the following set of assumptions about the lexicon
of Japanese:
(80) a. NJVs bear an interpretable feature [iNative], whereas VNs are not
What is meant by (80a) is lexical specifications of NJVs and VNs. As I have briefly
mentioned in section 3.2, NJVs are native vocabularies (hence the name) but VNs are
typically non-native ones (see also footnote 1). The feature [iNative] is intended to
108
formally represent this distinction. The assumption in (80b) is concerned with a
with [uNative], and this uninterpretable feature has an EPP property in the sense of
Pesetsky and Torrego (2001), attracting a head with the matching feature [iNative].
Finally, I assume that heads that take a vP complement (e.g. T and Neg) have [uv],
and the EPP property of this feature attracts v, as stated in (80c). Here, the two
uninterpretable features [uNative] and [uv] introduced in (80) are both conceived as
Let us first see how derivations of NJV clauses proceed, given the
here as (81).
The relevant steps of the derivation of this example are sketched in (82), ignoring
irrelevant details.
109
c. PR of the NJV to v
e. Head movement of v to T
First, the NJV ki ‘come’ is merged with the object Nihon-e ‘to Japan.’ As indicated in
(82a), the NJV enters the derivation with [iNative]. Next, a vP is formed by merging
v, which contains [uNative], with the VP. The crucial step of this derivation is (82c).
The EPP property of [uNative] on v triggers PR (i.e. head movement) of the NJV,
by merging the vP and T, which possesses [uv] (see (82d)). Head movement of v to T
is triggered by the EPP property of this feature, which successfully deletes [uv] that T
NJV, which has raised to v in the previous stage of the derivation. The derivation
results in the structure in which not only v but also the NJV is located outside the vP.
What about derivations of VN clauses, then? I argue that the derivation of the
VN clause in (2), repeated here as (83), proceeds in the steps given (84).
110
(83) Taro-ga Nihon-e kikoku si-ta.
d. Head movement of v to T
Notice first that in (84a) the VN does not come with the feature [iNative], unlike the
NJV in (82). This is assumed to be a lexical property of VNs, as stated in (80a). In the
next step (84b), v is merged with the VNP. Just as in the case of the NJV clause, v has
[uNative] with the EPP property here, too. However, unlike in (82b), there is no
matching feature for [uNative] in the structure in (84b), since the VN does not bear
[iNative]. PR does not apply to the VN exactly for this reason. As a consequence, the
introduced into the derivation with the uninterpretable feature [uv], and in (84d) head
111
The result of this derivation is the structure in which v is placed outside the vP, but
At this point, notice that the derivation of the VN clause in (84) seems
structure (minus the VNP, if one assumes cyclic Spell-Out) in (84d) is sent to PF,
clearly contrasts with the derivation of the NJV clause in (82), in which [uNative] on
If the VN clause (83) is indeed derived as in (84), as I claim here, then it must be
asked why the remaining uninterpretable feature on v does not cause a PF crash.
immediately after VNs is an NJV. So it should make sense to assume that it has the
otherwise leads to a PF crash. In the case of the VN clause (83), after the structure in
(84d) is sent to PF, su is adjoined to v and deletes [uNative] that survived the
(85) Insertion of su to v in PF
… [T [v v[uNative]+su[iNative] ]-T[uv]]]
112
It has often been suggested in the literature that a potential PF crash can be bypassed
with the help of a PF operation. For instance, assuming that ellipsis is a PF process,
scholars such as Lasnik (1999) and Merchant (2008) maintain that ellipsis is capable
Merchant’s on the other. In light of this, the idea that [uNative] on v, an illegitimate
considered to be invalid.
What should be recalled here is the fact that the supporting verb is excluded in
NJV clauses but obligatory in VN clauses, which I have noted in section 3.2. The
In the present context, this contrast concerning the supporting verb can be understood
clause, the same strategy is somehow unavailable in the NJV clause. I suggest that
113
this is because [uNative] that v bears in the NJV clause can be deleted by virtue of PR
of the NJV to v (see (82c) above), and that su-insertion in PF, which does the same
(1991), according to which operations specified as part of UG are less costly than
further assume that in the case of the NJV clause, [uNative] on v can in principle be
converge. If so, then we can reduce the obligatory absence of su in (86) to the
economy condition. The derivation which employs PR is preferred over the one
which does su-insertion in PF, since the former operation is less costly than the latter
one. Note in passing that in the case of the VN clause, the economy condition does
Summarizing this secton, I have proposed that PR applies to NJVs but not to
VNs because only the former bear [iNative], which can eliminate [uNative] on v. I
There has been much debate over the analysis of elliptical clauses like (88b).
114
(88) a. Taro-wa L.A.-keeyude Nihon-e ki-ta kedo,
Ziro-Top come-Neg-Past
‘Ziro didn’t.’
Two major analyses have been proposed in the literature to date: the Argument
Ellipsis (AE) analysis and the vP-ellipsis analysis. These two approaches crucially
differ from each other in how they treat the gap [e]. According to the former, the gap
in (88b) corresponds to the object Nihon-ni ‘to Japan’ (Oku 1998, Saito 2007,
According to the latter, on the other hand, the gap is analyzed as a null vP which
results from vP-ellipsis. The verb is still pronounced there because it is placed outside
of the elided domain, due to PR in overt syntax (Otani & Whitman 1991, Funakoshi
115
As for the current analysis, it has a way to analyze the gap in question as a null vP
derived by vP-ellipsis plus PR, precisely because the NJV come is used as the
predicate in (88), as discussed in section 3.5. It therefore partly belongs to the latter
type of approach.
At this point, recall from section 3.4 that (88b) allows the null adjunct reading
Ziro didn’t come to Japan via L.A., when anteceded by (88a). It therefore can be
Crucially, the AE analysis seems unable to capture this fact, as Funakoshi (2014)
For that particular reading to arise, the adjunct L.A.-keeyude ‘via L.A.’ must
be structurally present in (88b). The problem is that while the object may be
legitimately rendered null by AE, via L.A. should not be elided in the same way, since
it is not an argument:
*AE AE
116
Consequently, it would be predicted that (88b) cannot be interpreted as having an
implicit counterpart of via L.A. and therefore lacks the null adjunct reading, which is
strongly disconfirmed.
stipulating that the object and the adjunct actually make up a constituent, and AE
applies to this complex object, as roughly sketched in (93). This idea may not be as
exotic as it sounds, for the possibility of such complex object formation has been
suggested by Sohn (1994), quite independently of ellipis discussed here (see also
Takano 2002).
(93) [TP Ziro [Obj [Adj via L.A.] [Obj to Japan]] come-v-Neg-Past]
Unfortunately, this line of analysis is untenable when the discrepancy between NJV
and VN clauses discussed in section 3.4 is taken into consideration. The core
observation was that the null adjunct reading in the VN clause with an object gap is
blocked by pronunciation of the VN. The relevant examples in (32) are repeated here
‘Taro swam three laps of the 25-meter pool without breathing, but…’
117
b. Ziro-wa [Obj e] san-oofuku si-nakat-ta.
‘Ziro didn’t swim three laps of the 25-meter pool.’ (*null adjunct
reading)
Ziro-Top do-Neg-Past
The point here is that if the null adjunct reading in the elliptical NJV clause (88b)
were derived as in (93), it would remain mysterious why the VN clause (94b), in
which object drop coincides with pronunciation of the VN, cannot receive the null
adjunct reading by virtue of AE of the complex object in the way illustrated in (95).
account for the range of null adjunct phenomena. By contrast, the proposed analysis
can adequately handle the facts; as discussed in section 3.5, null adjuncts are derived
by vP-ellipsis, and the asymmetry between NJV and VN clauses regarding null
adjuncts can be reduced to the difference in PR. I therefore conclude that vP-ellipsis
must be available as an option to derive NJV clauses with (apparent) object gaps.
118
If this conclusion is correct, then Japanese should be considered as another
language that instantiates so-called verb-stranding vP-ellipsis, which has been argued
and Russian (Gribanova 2013). Below, I show that Japanese and Hebrew exhibit
certain similarities.
Goldberg (2005) first shows that although Hebrew allows direct objects (DOs)
to be left unpronounced, they can be legitimately omitted only when they are
inanimate. For instance, in the answer to the question in (96), the DO which refers to
la-'universita?
to.the-university
no we buy from-the-store
(Goldberg 2005:50)
In the answer to the queation in (97), on the other hand, the DO, which is supposed to
119
(97) Q: (Ha-'im) Miryam hisi'a et Dvora
la-makolet?
to.the-grocery.store
(Goldberg 2005:53)
Curiously, she reports that aminate DOs can be omitted if indirect objects (IOs) are
la-makolet?
(Goldberg 2005:53)
120
Goldberg then proposes that (98A) is derived by verb movement into the inflectional
domain followed by ellipsis of the verbal domain, represented as VP in her own work,
(99) [TP she drive [VP t [Dvora] [to the grocery store]]]
As a result of ellipsis illustrated in (99), the DO and the IO, both of which are
Hebrew in that both languages permit elision of elements which cannot be dropped by
themselves (adjuncts in Japanese on the one hand and animate DOs in Hebrew on the
other) only if other vP-internal materials are left unpronounced. This similarity is not
surprising at all, since not only null animate DOs in Hebrew but also null adjuncts in
(NJV clauses in) Japanese can be reduced to the verb-stranding vP-ellipsis strategy.
This is the conclusion Funakoshi (2014) reaches on the basis of null adjuncts he
investigates.14 The study presented in this chapter with special attention to the
distinction between NJV and VN clauses should serve to reinforce this conclusion.
14
An interesting condition on verb-stranding vP-ellipsis in Hebrew is what Goldberg
(2005) calls the verbal identity requirement, according to which a verb in an
antecedent vP and the one in an a target vP must be identical to each other (see also
Doron 1990 and Potsdam 1997 for such a requirement). Goldberg (2005: Ch. 4)
empirically motivates this requirement by demonstrating that animate DOs in Hebrew
cannot be dropped if verbs in the antecedent and target vPs are not identical.
Although I have not included discussion of the verbal identity requirement in the
present study of vP-ellipsis in Japanese, see Funakoshi (2014: Ch. 6) where he
observes that null adjuncts in Japanese are sensitive to identity of verbs and concludes
that verb-stranding vP-ellipsis in Japanese obeys the verbal identity requirement as
well.
121
I close this section with some remarks on extraction of phrasal constituents
from vP-ellipsis sites. I have argued above that the null adjunct reading is consistently
absent in NJV clauses with overt objects, because pronunciation of objects entails the
Strikingly, (100b) clearly lacks the null adjunct reading. This fact seems problematic
to the proposed analysis, because (100b) may be incorrectly predicted to permit that
reading; as Karlos Arregi has correctly pointed out, so long as it can be derived by
scrambling of the object from the vP followed by vP-ellipsis, as sketched in (101), the
null adjunct reading should obtain even with the overt object.15
(101) [TP [Obj to Japan] Ziro [vP via L.A. tObj tcome tv ] come-v-Neg-Past]
15
Takita (2011) makes a related observation, quite independently of null adjuncts
under discussion here. He shows that elliptical clauses in Japanese that could be
analyzed as being successfully derived by scrambling and verb movement followed
by vP-ellipsis are ill-formed; see Takita (2011) for details.
122
This problem of overgeneration might be avoided by positing a ban on phrasal
movement out of elided vPs in Japanese. Unfortunately, this idea immediately raises
another question; it is known that English, which everyone admits displays vP-ellipsis,
demonstrate that topicalization and wh-movement can take place out of vP-ellipsis
sites.
(Merchant 2008:140)
Which films did he refuse to see, and which films did he agree to [vP e]?
(Merchant 2013:539)
Then, the fact that the null adjunct reading is not compatible with the scrambled
object in (100b) might be taken to cast serious doubt on my attempt to attribute the
permitted. One might even say that the set of Japanese data under discussion is more
(104) *Which films did he refuse to see, and which films did he agree [e]?
(ibid.:539)
123
(105) *Mary wondered which conference talk Tommy refused to attend and Susan
(Depiatnte 2001:210)
These examples show that phrasal movement out of NCA is entirely prohibited,
which makes sense if NCA is an instance of deep anaphora in the sense of Hankamer
and Sag (1976) and therefore can be assigned no structure at all, as commonly
assumed.
derived by scrambling of the object as far as syntax is concerned, but it still lacks the
relevant reading for some independent reason which I do not fully understand at the
moment16. In other words, I argue that the absence of the null adjunct reading in
(100b) does not necessarily entail that phrasal movement out of [e] in elliptical
Ziro-Top come-Neg-Past
‘Ziro didn’t.’
16
See Funakoshi (2014), who, being aware of the problem associated with clauses
containing scrambled objects like (100b), suggests that they cannot receive the null
adjunct reading due to a violation of a discourse constraint.
124
In the remainder of this section, I provide a piece of evidence that phrasal movement
tenpura-o ‘tempura-Acc’ and tomato-o ‘tomato-Acc’ appear as the focus followed the
copula da in (106a) and (106b), respecticely. The focus element is associated with the
gap inside the presuppositional clause headed by the complementizer no. I refer to
clefts in which a Case-marked element appears in the focus position like these as
Case-marked clefts.
tenpura-o da.
tempura-Acc Cop
The key point to observe is that the Case-marked cleft in (106b) allows the null
adjunct reading it is tomatoes that he didn’t eat with salt, when anteceded by (106a).
Therefore, (106a-b) can be true in a situation in which Taro ate both tempura and
125
This fact is relevant in the present context, as it has been independently
presuppositional clause, and moves to its [Spec, CP], as depicted in (107). See also
Hiraiwa and Ishihara (2002) and Takahashi (2006), where Case-marked clefts are
discussed.
that (i) they allow long-distance dependency between the gap in the presuppositional
clause and the focus, as shown in (108), and (ii) they exhibit island-sensitivity, as
126
(109) *[CP Taro-ga [NP [CP [ei] tabe-ta] hito]-o home-ta
Lit. ‘It is blowfish that Taro praise the person who ate.’
Now, the availability of the null adjunct reading tells us that it is very unlikely that
the gap [e] in (106b) is merely a trace left behind by operator movement. Rather, it
should be more plausible to regard it as a null vP containing the adjunct sio-de ‘with
salt’ inside, given the discussion on null adjuncts so far. If so, then the
(110) [CP Opi [TP pro [vP with salt [VP ti teat ] tv ] eat-v-Neg-Past] no]
Here, Op moves from the object position inside the vP. The adjunct with salt is
actually present in the structure, but it ends up being unpronounced due to vP-ellipsis
in the presuppositional clause, which gives rise to the null adjunct reading, as desired.
That the null adjunct reading arises in (106b) seems hard to account for if [e]
know that extraction of phrasal constituents out of NCA is forbidden (see (104) and
(105)). To the extent that Case-marked clefts are derived by operator movement, as
Hoji (1990) convincingly argues, we can never reasonably say that the gap is NCA.
127
The fact about the Case-marked cleft (106b) thus serves to indicate that not only
English but also Japanese permits phrasal movement out of elided vPs.
3.10 Conclusion
In this chapter, I discussed asymmetries between NJV and VN clauses with respect to
predicate ellipsis and null adjuncts. Although the literature on verb movement, or
predicate raising (PR), in Japanese abounds, no attempt has been made so far to argue
for (or against) PR in the language by making reference to the distinction between
NJVs and VNs, to the best of my knowledge. I proposed that NJVs and VNs crucially
contrast with each other regarding predicate raising (PR), and the asymmetries can be
reduced to the difference in PR; NJVs are subject to PR on the one hand, but VNs are
not on the other. As for the trigger of PR, I proposed that the uninterpretable feature
feature [iNative] whereas VNs do not, PR applies only to the former. I showed that
the proposed analysis accounts for a wide range of facts, and argued that NJV clauses
128
Chapter 4: Remaining Issues
In the preceding chapters, I argued that there are good empirical reasons to believe
that, its string-vacuity notwithstanding, Japanese has overt head movement in syntax.
problems with the view I advocated in chapter 3. The proposal I made there is that
prediacte raising (PR) applies to NJVs whereas VNs are not subject to PR. Two
adverbial particles such as -mo, -sae and -dake, put forward by Sakai (1998, 2000)
and Aoyagi (1998, 2006) (see also Kuroda 1965, Sells 1995 for discussion of these
particles). Both Sakai and Aoyagi argue against verb movement in Japanese on the
basis of examples like (1b), in which the verb tabe ‘eat’ is suffixed with these
particles, and the supporting verb su (spelled out as si here) is obligatorily inserted in
between the adverbial particle and the tense suffix. These authors assume that su is
required in (1b) in order to provide the tense suffix, which is linearly separated from
1
Whether or not su employed in (1b) should be distinguished from su obligatorily
employed in VN clauses discussed in chapter 3 is a highly contentious issue, although
both instances of su seem void of meaning. I will not go into further details of this
topic; see Miyara (1991) where it is argued that they are identical, and Kageyama
(1992) who refutes Miyara’s proposal.
129
(1) a. Taro-wa tenpura-o tabe-ta.
The gist of their proposals is that adverbial particles immediately after verbs are VP-
adjuncts. To be more precise, these authors argue that the base structure of (1b) looks
like (2), in which the particle is adjoined to VP, rather than (3), in which the particle
takes the VP as its complement and heads its own projection (PartP) (irrelevant
details suppressed).
Given that the particle in (2) is simply adjoined to the VP, the verb eat should in
particle on its way, since adjuncts are not expected to block head movement. If so,
then it should be possible that the structure in (4) can be derived from (2), if verb
130
(4) [TP Taro [VP [VP tempura teat] Particle] [eat+Past]]
Notice that in the surface string which is expected to result from (4), the verb and the
past tense suffix would be linearized next to each other, just like they are in (1a).
Sakai and Aoyagi point out that if such a derivation were permitted, the empirical fact
that (1b) requires the supporting verb while (1a) does not would remain mysterious.
They hence conlude that verb movement is not available at all in Japanese.
dependent upon the status of adverbial particles as VP-adjuncts; the particle is not
selected by anything in the structure in (2). Aoyagi and Sakai each motivate this
treatment by observing that these particles are transparent for selection. Aoyagi
demonstrates that the selectional relations between verbs and their complements are
never interrupted by the presence of the adverbial particles. For instance, he shows
that the verb mi ‘see’ is required to take a te-clause when interpreted as meaning try,
See chapter 2 and references cited therein for discussion of te-clauses, and Miyagawa
131
b. *Taroi-wa [[ei] tenpura-o tabe-ni] mi-ta.
This contrast continues to show up when particles intervene between the te- and ni-
clauses and the verb, too; (6a) is fine whereas (6b) sounds gibberish. See also Sells
mi-ta.
see-Past
mi-ta.
see-Past
Aoyagi thus takes this set of facts to indicate that adverbial particles are transparent
for selection.
Sakai discusses the correlation between the insertion of the supporting verb
and the polite negative suffix masen. He shows that whereas verbal predicates as tabe
‘eat’ in (7) can be selected by masen, adjectival predicates like oisi(-ku) ‘delicious’ in
(8) cannot.
132
(7) a. Taro-wa tenpura-o tabe-ta.
tempura-Top delicious-Pres
‘Tempura is delicious.’
b. *Tenpura-wa oisiku-masen.
tempura-Top delicious-Neg.Pol
In order to get a well-formed result with an adjective, the supporting verb aru ‘be’
must be inserted right before the negative suffix, as in (9) (aru is spelled out as ari
before masen).
133
He then shows that masen invariably requires aru when adverbial particles attach to
adjectives as well, as shown in (10). He goes on to argue that it is higly unlikely that
the particles trigger insertion of aru, because in (11), where the verb is suffixed with
On the basis of this observation, Sakai also concludes that particles are transparent for
selection; if they headed their own projections and selected verbs or adjectives on
their left, the fact that insertion of aru takes place in the case of adjectives regardless
134
Although I believe that the core insight that adverbial particles are transparent
for selection per se is correct and far from trivial, it seems rather unclear whether the
for selectional relations that hold between verbs and nouns, as Baltin (1989) mentions.
The definite determiner the in (12), for instance, seems transparent in that the verb
gather requires a plural nominal as its object, ignoring the determiner; if it were the
case that the satisfied the selectional requirement of the verb in (12a), (12b) should be
Following the logic of Aoyagi and Sakai, one might want to say that determiners are
adjuncts located somewhere inside NP projections, but this obviously runs counter to
far beyond the scope of the present study. But so long as we can keep assuming the
may not oblige us to analyze them as adjuncts, which makes the structure in (2)
135
defended by Aoyagi and Sakai (repeted below for the sake of convenience) look less
plausible.2
Besides, it should be worth pointing out that it is one thing to analyze the
adverbial particles as adjuncts (on the basis of their transparency for selection), but it
is quite another to analyze them as being adjoined to phrases (VPs in the present
context). To put differently, we can think of the structure in which the adverbal
particle is adjoined to the verb itself, rather than the projection of the verb. For
instance, the sentence in (1b), reproduced below, can be assumed to have the base
particles. See also Sells (1995) for a similar (yet slightly different) idea.
2
But see Bruening (2009) for arguments against the traditional DP hypothesis.
136
(13) Particles adjoined to V
The particle in (13) does not project either, being adjoined to the verb. Now, suppose
that the comlex verb in (13) undergoes verb movement to T. We predict that the
In the PF representation derived from (14), the verb and the past tense morpheme will
not be linearized next to each other, because the former will be suffixed with the
particle. Hence, it follows that even if we assume that (i) Japanese has verb
movement and (ii) the particles are adjuncts, the PF representation which requires the
supporting verb su results, as desired. I thus conclude the view that the verb
movement analysis fails to account for why su is obligatory in (1b) held by Aoyagi
and Sakai may not be necessarily correct, being contingent upon the treatment of the
3
It is fair to note here that the structure in (14) seems unsustainable, since there is a
good reason to believe that a verb suffixed with an adverbial particle forms a
constituent with its object; the string made up of the verb suffixed with the particle
tabe-mo ‘eat-Particle’ and the object tenpura-o ‘tempura-Acc’ in (ia) can be
legitimately preposed, as in (ib). See Hoji, Miyagawa and Tada (1989) and Tateishi
(1994) for ‘VP-preposing’ of this sort.
137
To sum up the discussion so far, the argument provided by Aoyagi and Sakai
may not imperil the claim that verb movement (or predicate raising of NJVs, in my
own analysis) is available in Japanese at least in a serious way, for the structure in (2)
provided by Takano (2004). He examines examples like (15) with special attention to
the modifiers tigau, kotonaru, and betubetuno, all of which are glossed as ‘different.’
fairu si-ta.
file do-Past
(Takano 2004:175)
On the basis of the observation that the interpretation of the prenominal modifiers in
(15) is similar to that of the English adjective different which Carlson (1987)
discusses, he concludes that (i) the clause in (15) has coordinated verbs connected by
the phonetically unrealized conjunction, and (ii) the tense morpheme is syntactically
The grammaticality of (ib) can be construed as a clear indication that the verb is not
raised to T in (i). The structure in (14) is hence incompatible with this classic
observation. In order to nullify the argument against verb movement offered by
Aoyagi and Sakai, we therefore need to come up with a way to block raising of the
complex verb in (13), which enables us to rule in (ib). Exactly how to implement this
is left for future work.
138
separated from the second verb it morphologically attaches to. His analysis can be
problem for the predicate raising analysis I am pursuing here; the interaction between
predicate raising and the coordinate structure constraint must be dealt with.
Regrettably though, I have been unable to include such cases of coordination in the
139
References
Abney, Steven Paul (1987) The English noun phrase in its sentential aspect.
Aoyagi, Hiroshi (1998) Particles as adjunct clitics. In: Pius Tmanji and Kiyomi
Baker, Mark C. (1996) The polysynthesis parameters. New York: Oxford University
Press.
Baltin, Mark (1989) Heads and projections. In: Mark Baltin and Anthony S. Kroch
of Chicago Press.
Bošković, Željko (2014) Now I’m a phase, now I’m not a phase: on the variability of
140
Pennsylvania working papers in linguistics 15: Issue 1, 27-35. University of
Pennsylvania.
Carlson, Greg (1987) Same and different: some consequences for syntax and
Chomsky, Noam (1991) Some notes on economy of derivation and representation. In:
Chomsky, Noam (1993) A minimalist program for linguistic theory. In: Kenneth Hale
and Samuel Jay Keyser (eds.) The views from Building 20: Essays in
Dikken, Marcel den (2007) Phase extension: contours of a theory of the role of head
Doron, Edit (1990) V-movement and VP-ellipsis. Ms., The Hebrew University of
Jerusalem.
Fiengo, Robert and Robert May (1994) Indices and identity. Cambridge, MA: MIT
Press.
141
Fukui, Naoki and Hiromu Sakai (2003) The visibility guideline for functional
categories: verb raising in Japanese and related issues. Lingua 113: 321-375.
Funakoshi, Kenshi (2014) Syntactic head movement and its consequences. Doctoral
CSLI Publications.
Gribanova, Vera (2013) Verb-stranding verb phrase ellipsis and the structure of the
Russian verbal complex. Natural Language and Linguistic Theory 31: 91-136.
Grimshaw, Jane and Armin Mester (1988) Light verbs and θ-marking. Linguistic
Han, Chung-hye, Jeffrey Lidz and Julien Musolino (2007) V-raising and grammar
Hankamer, Jorge and Ivan Sag (1976) Deep and surface anaphora. Linguistic Inquiry
7: 391-426.
Descriptive and Applied Linguistics 10: 131-146. Reprinted in: Naoki Fukui (
142
Hasegawa, Nobuko (1991) On head movement in Japanese: the case of verbal nouns.
Japanese and the C-T relation. In: Enrico Boone, Martin Kohlberger and
Leiden University.
Hiraiwa, Ken and Shinichiro Ishihara (2002) Missing links: clefts, sluicing and “no
Nevins (eds.) MIT Working Papers in Linguistics 43, 35-54. Cambridge, MA:
MITWPL.
Hoji, Hajime (1990) Theories of anaphora and aspects of Japanese syntax. Ms.,
Hoji, Hajime (1998) Null objects and sloppy identity in Japanese. Linguistic Inquiry
29: 127-152.
MIT. http://web.mit.edu/miyagawa/www/pdfs/Hoji%20Miyagawa%20Ta
da%201989.pdf
Hoshi, Hiroto (1994) Passive, causative, and light verbs: A study on theta role
Hoshi, Hiroto (2002) Domains for theta marking. In: Proceedings of the 2002
143
Hoshi, Koji (2009) On the lack of VP-ellipsis in Japanese and its remifications in a
Keio University.
Linguistics 5: 117-155.
Kageyama, Taro (1992) Dummy su- and lexical su-: a reply to Miyara. Gengo
Kageyama, Taro (1993) Bunpoo to gokeisei [Grammar and word formation]. Tokyo:
Hituzi Syobo.
Kayne, Richard S. (1989) Null subjects and clitic climbing. In: Osvaldo Jaeggli and
Ken Safir (eds.) The null subject parameter, 239-261. Dordrecht: Kluwer
Academic Publishers.
Kenstowics, Michael (ed.) (2001) Ken Hale: A life in language. Cambridge, MA:
MIT Press.
144
Kishimoto, Hideki (2001) Binding of indeterminate pronounds and clause structure in
Kitagawa, Yoshihisa and S.-Y. Kuroda (1992) Passive in Japanese. Ms., Indiana
Koizumi, Masatoshi (2000) String vacuous overt verb-raising. Journal of East Asian
Linguistics 9: 227-285.
Koopman, Hilda and Dominique Sportiche (1991) The position of subjects. Lingua
85: 211-258.
Kuno, Susumu (1976) Subject raising. In: Masayoshi Shibatani (ed.) Syntax and
Press.
Larson, Richard K (1988) On the double object construction. Linguistic Inquiry 19:
335-391.
145
Manning, Christopher (1993) Analyzing the verbal noun: internal and external
Stanford: CSLI.
University Press.
study of the notion ‘word.’ Stanford and Tokyo: CSLI and Kurosio.
Matushansky, Ora (2006) Head movement in linguistic theory. Linguistic Inquiry 37:
69-109.
McCawley, James David and Katsuhiko Momoi (1986) The constituent of -te
Merchant, Jason (2004) Fragments and ellipsis. Linguistics and Philosophy 27: 661-
738.
Merchant, Jason (2008) Variable island repair under ellipsis. In: Kyle Johnson (ed.)
Merchant, Jason (2013) Diagnosing ellipsis. In: Lisa Lai-Shen and Norbert Corver
Miyagawa, Shigeru (1987b) Restructuring in Japanese. In: Takashi Imai and Mamoru
Miyagawa, Shigeru (2001) The EPP, scrambling, and wh-in-situ. In: Kenstowicz
(2001), 293-338.
146
Miyara, Shinsho (1991) On the insertion of /s/-form (suru) in Japanese. Gengo
Otani, Kazuyo and John Whitman (1991) V-raising and VP-ellipsis. Linguistic
Pesetsky, David and Esher Torrego (2001) T-to-C movement: causes and
Pollock, Jean-Yves (1989) Verb movement, universal grammar, and the structure of
Poser, William (1989) Japanese periphrastic verbs and noun incorporation. Ms.,
Stanford University.
Potsdam, Eric (1997) English verbal morphology and VP ellipsis. In: Kiyomi
147
Richards, Norvin (2003) Why there is an EPP. Gengo Kenkyu 123: 221-256
Rizzi, Luigi (1978) A restructuring rule in Italian syntax. In: Samuel Jey Keyser (ed.)
Roberts, Ian (2010) Agreement and head movement: clitics, incorporation, and
Saito, Mamoru (2007) Notes on East Asian argument ellipsis. Language Research 43:
203-227.
Saito, Mamoru and Hiroto Hoshi (1998) Control in complex predicates. In: Report of
Saito, Mamoru and Hiroto Hoshi (2000) Japanese light verb construction. In: Roger
Martin, David Michaels and Juan Uriagereka (eds.) Step by step: Essays on
Press.
Saito, Mamoru and Keiko Murasugi (1990) N’-deletion in Japanese. In: Hajime Hoji
Sakai, Hiromu (1998) Feature checking and morphological merger. In: David James
University.
148
Sato, Yosuke (2010) Bare verbal nouns, idiomatization and incorporation in Japanese.
Sells, Peter (1990) VP in Japanse: evidence from -te complements. In: Hajime Hoji
Sells, Peter (1995) Korean and Japanese morphology from a lexical perspective.
Shin, Stavros Skopeteas and Elisabeth Verhoeven (eds.) Form and function in
Mouton de Gruyter.
Shimada, Junri (2007) Head movement, binding theory, and phrase structure. Ms.,
MIT.
ellipsis and the licensing of covert nominals in Bangla, Hindi and Malayalam.
Sohn, Keun-Won (1994) Adjunction to argument, free ride and a minimalist program.
149
Sugioka, Yoko (1992) On the role of argument structure in nominalization. Language,
Takahashi, Daiko (2006) Apparent parasitic gaps and null arguments in Japanese.
Takahashi, Daiko (2008) Noun phrase ellipsis. In: Shigeru Miyagawa and Mamoru
Takahashi, Mari (2000) The syntax and morphology of Japanese verbal nouns.
Takano, Yuji (2002) Surprising constituents. Journal of East Asian Linguistics 11:
243-301.
Takano, Yuji (2004) Coordination of verbs and two types of verbal inflection.
Takita, Kensuke (2011) An argument for argument ellipsis from -sika NPIs. In: Suzi
Lima, Kevin Mullin and Brian Smith (eds.) Proceedings of NELS 39, 771-784.
Tamori, Ikuhiro (1976) The semantics and syntax of the Japanese gerundive and
Tateishi, Koichi (1994) The syntax of ‘subjects’. Stanford and Tokyo: CSLI and
Kurosio.
150
Terada, Michiko (1990) Incorporation and argument structure in Japanese. Doctoral
Tsujimura, Natsuko (1992) Licensing nominal clauses: the case of deverbal nominals
Tsujimura, Natsuko (1993) Adjuncts and event argument in restructuring. In: Soonja
Tsujioka, Takae (2011) Idioms, mixed marking in nominalization, and the base-
Watanabe, Akira (2010) Notes on nominal ellipsis and the nature of no and classifiers
Williams, Edwin (1977) Discourse and logical form. Linguistic Inquiry 8: 101-139.
Mouton de Gruyter.
151
Wurmbrand, Susi and Jonathan David Bobaljik (2005) Adjacency, PF, and
extraposition. In: Hans Broekhuls, Nobert Corver and Riny Huybregts (eds.)
152