1

Nicholas Primrose
Thesis Adviser: J orge Hankamer
04/09/13


J apanese Numerals and Numeral Quantifiers


I. Introduction

In this paper, J apanese numerals and numeral quantifiers will be under investigation.
J apanese numerals are a complex unit: a combination of a number and a classifier. These
classifiers are always attached to a number and the classifier specifies the type of noun being
modified by the number. In English, numbers and quantifiers can appear in a large variety of
positions that make it difficult to tell exactly where these elements go in the structure. In
J apanese, numbers and quantifiers can appear in an even greater number of positions. The main
point of this paper will be to discover the positions that the numerals and numeral quantifiers can
be in. Since it appears that the numerals and numeral quantifiers can appear in an incredible
number of positions, these facts have caused researchers to propose a wide variety of possible
structures for the nominal domain. Typically, these proposals have a large number of functional
projections and a wide variety of movements around the nominal domain. Certain things, like
J apanese not having an overt determiner, make figuring out the actual structure very difficult. I
seek to understand the nominal domain better in order to pinpoint the positions of these numerals
and numeral quantifiers. What I will end up doing is positing a few positions within this
structure where the numeral and numeral quantifier can appear.
In the end, I will argue that there are three positions that numerals and numeral
quantifiers can appear. I will argue that numerals and numeral quantifiers participate in a word
formation process that superficially appears to be [#-GEN N]. Actually, the element that
superficially appears to be the GEN is actually a LINK that semantically relates any two
elements. Numerals and numeral quantifiers can also appear as an adjective, appearing as a right
modifier to NP. And finally, I will argue that these numerals and numeral quantifiers can be left
in interesting places resulting from their interesting interaction with the copy theory of
movement. These three positions will be argued mainly from properties of J apanese and the
syntax of J apanese, but at some points appeals will be made to other languages to help
understand the parts of the structure that are not easily defined, like the lack of an overt D in
J apanese.


The layout will be as follows:

Section 2 will explore the main data under investigation and the generalizations that need
to be explained under any theory of the syntax of the J apanese nominal domain. Section 3 will
show a basic picture of the problem. Section 4 will explore some basic elements that are
involved in J apanese numerals and numeral quantifiers and explicate their basic properties.
Section 5 will explore the morpheme no that attaches to numerals and numeral quantifiers and
2

show that that morpheme is the LINK, which is used in a word formation process to create Ns.
Section 6 will explore the numeral exclusive position, the position that the numeral appears in
that the numeral quantifier does not, and argue, mainly through other languages, that that
exclusive position must be a right modifier to NP. Section 7 will attempt to understand how
floating quantifiers fit into this theory. And Section 8 will sum up whats been learned and pose
further questions.



II. The Basic Data


Its important to go slow through this data, as this pattern is the core of the problem and
its important to keep this data in mind while looking at the rest of this paper. To understand it
better, heres a chart of where these numerals and quantifiers are able to appear in this simple
example.




(1) a. J ohn-wa san-satu hon-o kat-ta.
J ohn-TOP 3-CL book-ACC buy-PST

b. J ohn-wa san-satu-no hon-o kat-ta.
J ohn-TOP 3-CL-LINK book-ACC buy-PST

c. J ohn-wa hon san-satu-o kat-ta.
J ohn-TOP book 3-CL-ACC buy-PST

d. J ohn-wa hon-o san-satu kat-ta.
J ohn-TOP book-ACC 3-CL buy-PST

J ohn bought three books. (Watanabe 2006: 3)
12


1
Abbreviations are as follows: +SUBJ , subject feature; +Q, question marker; +WH, Wh-feature; #, numeral; ACC,
accusative; CL, classifier; D, determiner; DAT, dative; DP, determiner phrase; EPP, extended projection principle;
Prenominal

Prenominal
with no
N Between N
and K
K After K

san-satu










takusan










3

(2) a. J ohn-wa takusan hon-o kat-ta.
J ohn-TOP many book-ACC buy-PST

b. J ohn-wa takusan-no hon-o kat-ta.
J ohn-TOP many-LINK book-ACC buy-PST

c. *J ohn-wa hon takusan-o kat-ta.
J ohn-TOP book many-ACC buy-PST

d. J ohn-wa hon-o takusan kat-ta.
J ohn-TOP book-ACC many buy-PST

J ohn bought many books. (Watanabe 2006: 70)


In addition to appearing in those various positions, numerals and numeral quantifiers are
also able to float, and by that I mean they are able to appear in trace positions of things they can
attach to. Notice (3) and (4):

(3) Yuube, kuruma-ga doroboo ni 2-dai nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PST
Last night, two cars were stolen by a thief. (Miyagawa 1989: 38)

(4) Yuube, kuruma-ga doroboo ni takusan nusum-are-ta.
last.night cars-NOM thief by a.lot steal-PASS-PST
Last night, a lot of cars were stolen by a thief. (Miyagawa 1989: 21)

When kuruma raises up to specTP, the numerals and numeral quantifiers are able to be
left behind. These floaters can also appear when scrambling happens:

(5) Hon-o gakusei-ga san-satu kat-ta.
book-ACC student-NOM three-CL buy-PST
A/the student(s) bought three books. (Nakanishi 2b)

When hon scrambles up to the front of the sentence, numerals and numeral quantifiers
can be left behind. These additional floating positions will be difficult to describe and will be the
subject of section 6.


GEN, genitive case; HON, honorifics; K, case; KP, case phrase; LINK, linker; LOC, locative; N, noun; NMLZ,
nominalizer; NOM, nominative; NP, noun phrase; NPST, non-past; NQ, numeral quantifier; PASS, passive; POL,
polite element; POSS, possessive; PRG, progressive; PST, past; Q, quantifier; QP, quantifier phrase; T, tense; TOP,
topic; V, verb; VP, verb phrase.
2
I would like to thank Sakae Fujita and Mariko Tajima for their grammaticality judgments throughout this paper.
Examples from Turkish are from p.c. with Jorge Hankamer. I would like to thank Jorge Hankamer, Anie Thompson,
and Jim McCloskey for their helpful comments.
4

III. A Basic Picture of the Problem

Since this paper will be delving into the nominal domain, I will be assuming the
following for its structure:


There are three main positions to understand: the floated position, the numeral exclusive
position, and the #-no position. First off, I will argue that the floated position for numerals and
numeral quantifiers is created from the numerals and numeral quantifiers having a special
property that allows them to be left behind whereas the rest of the copy they attach to must
delete.






5

Secondly, I argue that the numeral exclusive position is a right modifier to NP and thus
appears in the structure as the following:

Finally, I will argue that the #-no position is created by a N word formation process using
the morpheme no, the LINK. Thus, the position in the structure would be:


6

IV. Some Basic Things

Before moving on to more interesting things, I want to take the time to define numerals
and numeral quantifiers.
NQ
I will be defining numeral quantifiers (NQ) as subete (all) and takusan (many). These
can appear in all of the same positions as the numerals except for the position between N and K.
Some examples are below:
(6) takusan-no hito
a.lot-LINK people
a lot of people
(7) subete-no hon
all-LINK book
all books
(8) kyoukasyo-ga takusan
textbook-NOM a.lot
a lot of textbooks
(9) kokuban-o subete
blackboard-ACC all
all blackboards

#
I will also be investigating numerals. In J apanese, numerals come as a package deal with
both the number and a classifier that specifies the general kind of thing the object is being
counted as. Various classifiers followed by examples of each are shown below:
Hiki: used with small animals (mouse, squirrel, small dogs, etc.)
(10) san-biki-no risu
3-CL-LINK squirrel
3 squirrels
7

(11) go-hiki mausu
5-CL mouse
5 mice

Hon: used with long objects (bottles, etc.)
(12) biiru san-bon
Beer 3-CL
3 beers
(13) kyuu-hon-no tai
9-CL-LINK tie
9 ties

Mai: used with flat objects (sheets of paper, CDs, etc.)
(14) ichi-mai CD
1-CL CD
1 CD
(15) san-byaku-mai-no repooto
3-hundred-CL-LINK report
Three hundred page report

Satsu: used with bound volumes (books, magazines, comic books, etc.)
(16) juu-satu-no manga
10-CL-LINK comic.book
10 comic books
(17) san-satu zassi
3-CL magazine
3 magazines
8

Dai: used with electronics and large vehicles (computers, phones, cars, roller coasters, etc.)
(18) ni-dai-no compyuuta
2-CL-LINK computer
2 computers
(19) kuruma-ga hati-dai
car-NOM 8-CL
8 cars

There are many more classifiers.


V. NO

In this part, I will be dealing with the morpheme no that can be attached to numerals and
numeral quantifiers. This is the linker. It is used in a word formation process that is pretty
extensive in the language. LINK combines two elements to form a noun and semantically relates
them. The exact nature of the relation is difficult to describe and will not be the focus here.
What the focus of this part will be carefully separating the LINK apart from the GEN and
showing that the morpheme no that is attaching to numerals and numeral quantifiers has got to be
LINK. To do that, first, there will be some basic examples with no attaching to the numeral and
numeral quantifier. Then, the 4 morphemes homophonous with no in J apanese will be explored.
Two of them will quickly be rejected based on those morphemes acting in very different
environments from the no that attaches to numerals and numeral quantifiers. Then, the section
will narrow down to comparing the GEN and LINK. It will be shown that the morpheme in
question is LINK through three main methods. One method will be showing that are some basic
syntactic reasons to suggest that LINK is in a different place than GEN. The second method will
be showing that the morpheme in question works in an unexpected way when stacked if its
assumed that its the GEN. The third method will be showing that GEN, unlike LINK, can
license NPE.




9

To start with, some basic examples of the no attaching to numerals and numeral
quantifiers.
(1b) J ohn-wa san-satu-no hon-o kat-ta.
J ohn-TOP 3-CL-LINK book-ACC buy-PST
J ohn bought three books. (Watanabe 2006: 3b)

(2b) J ohn-wa takusan-no hon-o kat-ta.
J ohn-TOP many-LINK book-ACC buy-PST
J ohn bought many books. (Watanabe 2006: 70b)



The four morphemes homophonous with no are +Q, NMLZ, GEN and LINK. First, the
+Q no. There is another +Q morpheme that is much more commonly used than this morpheme,
ka. Ka and no both appear in the same position and cannot co-occur. The difference between ka
and no is that the +Q no is said more often by women.
(20) toshokan-ni it-ta-no?
library-to go-PST-+Q
Did you go to the library?
(21) yuusyoku-o tabe-ta-no?
Dinner-ACC eat-PST-+Q
Did you eat dinner?
(22) nani-o yom-u-no?
What-ACC read-NPST-+Q
What will you read?/ What are you reading?

As should seem pretty clear, this cant possibly be the morpheme attaching to numerals
and numeral quantifiers. It has a completely different meaning and appears in a very different
syntactic location.


The second no that I will investigate is the NMLZ no. This morpheme attaches to verbs
in their bare NPST form without any politeness and turns those verbs into related nouns. The
basic examples are below:




10

(23) Watasi-wa hanasu no - ga suki.
I-TOP to.speak-NMLZ-NOM like
I like speaking.
(24) Mitiko-san-ga yakyuu-o suru-no-ga joozu-ni nari-tai.
Michiko-HON-NOM baseball-ACC do-NMLZ-NOM good-at become-want
Michiko wants to become good at playing baseball.
(25) Hikooki-no yoyaku suru-no-o si-tei-masu-ka?
Airport-LINK reservations do-NMLZ-ACC know-PRG-POL-+Q
Do you know how to make airplane reservations?
(Lit. Do you know doing airplane reservations)

One possible analysis of this morpheme is that it is really the LINK morpheme starting
out something like (26) and then undergoing NPE.
(26) suru-no-koto-o
do-NO-thing-ACC

If this analysis was true, then it would undergo NPE and result in (27):
(27) suru-no-o
do-NO-ACC

But this cant be right. As I will argue just a little bit later in this section, the LINK cant license
NPE but the GEN can. Which means the problem for this analysis is that if this use of no was the LINK,
then you would predict it to not license NPE in the way the analysis requires. And if this use of no was
the GEN, then you would predict the verb to be related to the dropped object by means of possession.
Since it makes no sense for verbs to possess anything, then this morpheme has got to be separate.
Because of these reasons, this no is a NMLZ that is used in a word formation process taking a V and
turning it into a N. This is also why this verb has to appear in the bare form. If you notice (28a) and
(28b), the verb cant take negation or tense:



11

(28) a. *Watasi-wa hanas-anai no - ga suki.
I-TOP speak-NEG-NMLZ-NOM like
I like not speaking.
b. *Watasi-wa hanasi-ta no - ga suki.
I-TOP speak-PST-NMLZ-NOM like
?I like spokening.
So this no attaches to Vs and makes Ns. This cant be the no that attaches to numerals
and numeral quantifiers.


Here, I will try to understand the similarities and differences between GEN and LINK to
prove that the morpheme no that attaches to numerals and numeral quantifiers is the LINK and
not the GEN. There will be three main points that I will use to prove this. The first is that the
syntactic position of LINK and GEN must be different. The second is that the semantics of
LINK differs in an interesting way from the semantics of GEN when there are stacked elements
and semantics of the numeral/numeral quantifier no is like LINK and not GEN. The third is that
GEN can license NPE whereas LINK cant. In terms of how this section is ordered, first, the
basic differences that Im positing between the two morphemes will appear in a list. Then, there
will be some basic examples of the two morphemes. And then, the three main points will be
attacked one by one.

This section will conclude that the GEN works as following:
Relates a possessor and possessee
Appears in the structure before adjectives
When stacked in structures like [NP
1
-GEN NP
2
-GEN- NP
3
], NP
1
must be a possessor that
possesses NP
2
and NP
2
must be a possessor that possesses NP
3
. This means the semantic
stacking applies linearly left-to-right.
Can license NPE

This section will conclude that LINK works as following:
Relates two meaningful elements
Appears in the structure after adjectives
When stacked in structures like [NP
1
-LINK NP
2
-LINK- NP
3
], both NP
1
and NP
2
are
meaningfully related to NP
3
. It is not the case that NP
1
is related to NP
2
and NP
2
is
related to NP
3
. This means the semantic stacking applies right-to-left
Cant license NPE
12

Basic examples of GEN look like the following:

(29) Mitiko-no tabemono
Michiko-GEN food
Michikos food
(30) Mitiko-no kaban
Michiko-GEN bag
Michikos bag
(31) Siniti-san-no titi-no ie
Shinichi-HON-GEN father-GEN house
Mr. Shinichis dads house

Basic examples of LINK look like the following:
(32) ame-no hi
rain-LINK day
rainy day (Saito & Murasugi 50a)
(33) san-satsu-no Chomsky-nitsuite-no hon
3-CL-LINK Chomsky-about-LINK book
three books about Chomsky (Watanabe 2006 22a)
(34) warui-no syuumatu
bad-LINK weekend
bad weekend

The syntactic position of GEN must be different than LINK because adjectives appear
before LINK but after GEN. First, here are some examples of adjectives appearing after GEN:
(35) a. Mitiko-no oisii tabemono
Michiko-GEN delicious food
Michikos delicious food
13

b. *oisii Mitiko-no tabemono
delicious Michiko-GEN food
Michikos delicious food
(36) a. Tanaka-san-no kibisi titi
Tanaka-HON-GEN strict father
Mr. Tanakas strict father
b. *kibisi Tanaka-san-no titi
strict Tanaka-HON-GEN father
Mr. Tanakas strict father

Without a doubt, adjectives must be after GEN. But also, without a doubt, adjectives
must be before LINK.
(37) a. oisii kankoku-no tabemono
delicious Korea-LINK food
delicious Korean food
b. *kankoku-no oisii tabemono
Korea-LINK delicious food
delicious Korean food
(38) a. kibisi bengosi-no otoko
strict lawyer-LINK man
strict lawyer man
b. *bengosi-no kibisi otoko
lawyer-LINK strict man
strict lawyer man

This means that LINK must be in a different syntactic position from GEN. In my story of
the nominal domain, this means GEN must be in spec DP and LINK must be within the NP.
Since I will argue that LINK is part of a N word formation process, this data makes perfect
sense.
14

Secondly, there are some interesting semantic affects that are impossible to capture if the
no that attaches to numerals and numeral quantifiers is really the GEN. When stacked, the GEN
looks like the following:

(31) Siniti-san-no titi-no ie
Shinichi-HON-GEN father-GEN house
Mr. Shinichis dads house

(39) Mitiko-no inu-no kao
Michiko-GEN dog-GEN face
Michikos dogs face

In these constructions, the form is [NP
1
-GEN NP
2
-GEN- NP
3
]. NP
1
must be a possessor
that possesses NP
2
and NP
2
must be a possessor that possesses NP
3
. This means that (31) means
Mr. Shinichis dads house and it is not the case that that house is owned by Mr. Shinichi.
Similarly, (39) means Michikos dogs face and not Michikos face with some additional
info about her dog. This seems obvious. But this is not how the LINK semantically builds
elements together. Notice:

(40) oisii-no kankoku-no tabemono
delicious-LINK Korea-LINK food
delicious Korean food

(33) san-satsu-no Chomsky-nitsuite-no hon
3-CL-LINK Chomsky-about-LINK book
three books about Chomsky (Watanabe 2006 22a)

In the form, [NP
1
-LINK NP
2
-LINK- NP
3
], both NP
1
and NP
2
are meaningfully related to
NP
3
. It is not the case that NP
1
is related to NP
2
and NP
2
is related to NP
3
. In (40), it means that
the food is both related to Korea and related to deliciousness. Crucially, it does not mean that
Korea is delicious and that delicious Korea is related to food. That is what one would expect if
one were to assume that the no in the above examples was the GEN. Also notice (33), in which
15

it means that books are related to being about Chomsky and related to 3-ness. It does not mean
that there are 3-Chomsky-abouts that are related to a book.
Its possible that this is just to do with semantic restrictions. Perhaps the reason theres
this switch where both elements refer to the main N rather than to each other in a chain is
because of the examples being used. Perhaps the only reason theyre bad is because Korea cant
be delicious and you cant count abstract concepts like being about something. But if you notice
(41), its possible that crazy is linking with either rain or day. It could be a crazy day thats rainy
or crazy rain on this day. If this no was the GEN, you would expect it to mean crazy rain on that
day. If this no was working differently, you would expect it to mean a crazy day that was rainy.
And it means that theres a crazy day that was rainy.
(41) kureejii-no ame-no hi
crazy-LINK rain-LINK day
Crazy, rainy day

Therefore, when LINK stacks elements, all the stacked elements relate to the main N,
which seems to make sense for a N building process. For GEN, each element relates to the next
one in the chain, which seems to make sense for GEN based on the structure of how multiple
GENs come about. This is another proof towards the fact that the no that attaches to numerals
and numeral quantifiers is LINK and not GEN.

The final part of this section will be about NPE. This idea is based on Saito and
Murasugi (1990) and the appendix to Watanabe (2009). Saito and Murasugi argue that the GEN
allows for NP ellipsis whereas the LINK does not. Compare (42a) and (42b).
(42) a. [Rooma-no hakai
1
]-wa [Kyooto-no e
1
] -yorimo hisan datta.
Rome-GEN destruction-TOP Kyoto-GEN than horrible was
Romes destruction was more horrible than Kyotos.
b. *Saikin-wa [hare-no hi
1
]-ga [ame-no e
1
] yorimo ooi.
recently-TOP clear-LINK day-NOM rain-LINK than many
Recently, there have been more clear days than rainy days.
(Saito & Murasugi 49a and 51a)
In (42a), the GEN is able to license the NP ellipsis. In (42b), the LINK cant. Its clear
that Saito and Murasugis distinction between the two no homophonous elements is true. But
can the no that attaches to numeral quantifiers and numerals license NP ellipsis? For this, they
turn to (43):
(43) *[huka-kire-no hamu
1
]-wa yuusyoku-ni naru ga, [hito-kire-no e
1
]-wa nara-nai
Two-CL-LINK ham-TOP supper-to make but 1-CL-LINK -TOP make-NEG
Two slices of ham make up a supper, but one slice of ham does not.
(Saito &Murasugi 52b)
16


This example, however, is faulty. In general, numbers cannot be in topics. See (44) and
(45). The only difference between (44) and (45) is TOP vs. NOM. The number being in the
topic makes the sentence fail.

(44) *gakusei san-nin-wa honya-de hon-o kat-ta.
student 3-CL-TOP bookstore-LOC book-ACC buy-PST
(45) gakusei san-nin-ga honya-de hon-o kat-ta
Student 3-CL-NOM bookstore-LOC book-ACC buy-PST
3 students bought books at a bookstore.
But this doesnt mean that Saito & Murasugi are wrong. The no that attaches to numerals
and quantifiers still cant perform NP ellipsis.

(46) *Taroo-wa iti-niti-ni [san-satsu-no hon]-o yomu ga, Hanako-wa
Taroo-TOP 1-day-in three-CL-LINK book-ACC read but Hanako-TOP

go-satsu-no-o yomu
five-CL-LINK-ACC bought
Taroo reads 3 books in a day, but Hanako reads five. (Saito 2008: 22b)
This is another proof that the no that attaches to numerals and quantifiers is the LINK and
not the GEN. Syntactic position, semantic stacking, and interaction with NPE all prove that the
no that attaches to numerals and quantifiers must be LINK.


VI. Numeral Exclusive Position

This section seeks to understand where the position exclusive to the numeral is. The
relevant contrast is below:
(1c) J ohn-wa hon san-satu-o kat-ta.
J ohn-TOP book 3-CL-ACC buy-PST
J ohn bought three books.

(2c) *J ohn-wa hon takusan-o kat-ta.
J ohn-TOP book many-ACC buy-PST
J ohn bought many books.

17

This section will review the various possibilities and reject them one by one. First, this
section will consider the position being a right specifier for NP, but some reasonable previously
held assumptions and cross-linguistic facts make this option unlikely if not impossible. Second,
this section will consider the #being the D head, which seems like a good spot when looking at
J apanese, but cross-linguistically, this seems unreasonable. And finally this section will consider
the #being a right modifier to NP. This proposal will be accepted as this seems best both for
J apanese and cross-linguistically. Though it would be better to not have to rely on cross-
linguistic data to prove the points in this section, J apanese not having an overt D makes relying
on cross-linguistic data necessary.


First, I will consider the #as being the right specifier of NP. That would look like:

This is possible and would get the order of elements correctly, but there are a number of
problems with this hypothesis. This hypothesis goes against previously held assumptions. The
first assumption that this goes against is a very simple but important assumption. That
assumption is that the notion that the ordering of specifiers, heads, and complements is largely
consistent within a language. In general, we find all the heads to be right or left, we find the
specifiers to be right or left, and we find the complements to be right or left. There are
exceptions where this is not the case, for example heads in German, but it seems to be a good
overall trend for the direction of these elements to stay consistent. This assumption seems like
one that is easy to break, but only for good reason. Since there are no other specifier-like
elements that come to the right, its hard to say why this specifier would come to the right. And
since I dont have a great reason to break that assumption, why break it at all? The second
assumption that this goes against is a NP-internal genitive raising hypothesis. It is very likely
that one wants the genitive to begin in spec NP, then for D[POSS] to probe down to find that DP,
and raise it up to spec DP. That would be schematized as follows:
18


This movement seems good for a number of reasons. It gives a good place for that DP to
start in the structure. It makes the nominal and verbal domain look more parallel because then
the D probing down looks a lot like the T probing down. And it explains GEN-POSS agreement
that you find in many languages similar to phi-feature agreement in the verbal domain. If such a
movement existed, then the NP would already have a specifier. It seems outlandish to suppose
theres a right and left specifier without some external evidence for this. The third problem with
this hypothesis is that this hypothesis seems bad cross-linguistically. Look at (47) and (48):
(47) a. The three big smelly pigs
b. *The big three smelly pigs
c. *The big smelly three pigs
(48) a. The seven squawking birds
b. *The squawking seven birds
Trying to take this result over to English gives a bad result. For English, it seems insane
that these number elements are in spec NP. Since there are multiple places for numbers in
J apanese, one might wonder if this is simply a different position. But notice (49):
(49) a. The three dogs
b. *The many dogs
This appears to be the same position that is being investigated in J apanese. It seems
crazy to try to take this idea about #being specNP cross-linguistic. Since these two important
assumptions would be broken and there would be massive problems for trying to figure out the
English data, the #cant be in spec NP.
19



Second, I will consider this #as a D head. That would look like the following:

This seems like a rather natural place for the numeral to go if one looks at J apanese. Its
a really clear position and since there seems to be no overt D, this would be the only overt D and
nothing would conflict with it. But trying to make this cross-linguistic seems to fail horribly.
Look at English and Turkish:
(47) a. The three big smelly pigs
b. *The big three smelly pigs
c. *The big smelly three pigs
(48) a. The seven squawking birds
b. *The squawking seven birds

In English, it seems ridiculous for the #to be a D head. It can co-occur with D heads, so
why would it be a D head?

(50) bir maymun
a monkey

(51) mutlu bir maymun
A happy monkey


20

(52) Cok mutlu bir maymun
A very happy monkey

(53) bir muz
A banana

(54) pek pahalI bir araba
too expensive a car

(55) biraz pahalI bir araba
a somewhat expensive car

(56) *bir mutlu maymun
a happy monkey

(57) *bir Cok gUzel kIz
a very beautiful girl

(58) *u bir pahalI maymun
That one expensive monkey

Turkish has a very similar structure to J apanese. But this bir element, which means 1, is
quite elusive. This bir element is closer to the noun than demonstratives and adjectives, which
makes it seem like specNP. Since one would want NP internal GEN raising for Turkish,
especially since one gets GEN-POSS agreement, the specNP position is taken. As I will argue
later, this means it is an adjective itself. But it definitely cant be a D.
Proving that #is not a D head seems very difficult within J apanese and requires one to
look outside of J apanese to other languages. In other languages, #being a D head is absolutely
ridiculous and for that reason, #is not a D head in J apanese.









21

Finally, I will consider the #as a right adjective. That would look like:

This would correctly get the order. One might wonder if this violates the general
direction of modifiers that one finds in J apanese. But modifiers seem different from heads,
specifiers, and complements in that modifier order doesnt seem as strict as any of the other
elements orders. Its quite common to find modifiers appearing on both sides, where any of the
other elements appearing on both sides might be quite strange. So if there isnt a problem with
some modifiers in J apanese being on the right, one might wonder about cross-linguistic data.
And here we return to English and Turkish.
In English, it seems quite good to have the #as an adjective. It appears between D and N
and before all the other adjectives.
(47) a. The three big smelly pigs
b. *The big three smelly pigs
c. *The big smelly three pigs
(59) The two large red brick houses

In English then, the #appears to be an adjective ordered before all other adjectives. Lets
return to Turkish.



22

(52) Cok mutlu bir maymun
A very happy monkey

(53) bir muz
A banana

(54) pek pahalI bir araba
too expensive a car

In Turkish, the #appears to be an adjective that is ordered after all other adjectives. This
gives it the appearance of specNP, but its not. Since this proposal works for J apanese and gets
the cross-linguistic data best, the exclusive position for the #must be a right modifier of the NP.


VII. Floating
This section will turn now to understanding the two pairs of sentences below. These pairs
of sentences are instances of what has traditionally been called floating quantifiers/numerals.
They seem to appear in a wide range of positions, many of which one might not expect them in,
making their distribution difficult to describe. There have been two main accounts to deal with
this data: a stranding and an adverbial analysis. These will both be shown to be inadequate to
deal with the data. The stranding analysis will be shown to be incorrect because of the difficulty
in deriving examples in which the quantified expression moves multiple times. Additionally, the
stranding analysis seems to suppose a [+SUBJ ] feature. The adverbial theory will be shown to
be faulty because of how late the process would need to apply and how difficult making sure that
this late process actually gets the right form would be. Instead, a copy theory of movement
analysis will be shown to be superior in understanding the data under investigation. To start, the
data:

(1) c. J ohn-wa hon-o san-satu kat-ta.
J ohn-TOP book-ACC 3-CL buy-PST
J ohn bought three books.
d. J ohn-wa san-satu hon-o kat-ta.
J ohn-TOP 3-CL book-ACC buy-PST
J ohn bought three books. (Watanabe 2006: 3c&d)

(2) c. J ohn-wa hon-o takusan kat-ta.
J ohn-TOP book-ACC many buy-PST
J ohn bought many books.
d. J ohn-wa takusan hon-o kat-ta.
J ohn-TOP many book-ACC buy-PST
J ohn bought many books. (Watanabe 2006: 70c&d)

23

The difficulty in describing these floaters is because they are able to be left behind by
movement operations. In (5), the numeral is left behind after scrambling moves hon-o to the
front and in (60), the numeral quantifier is left behind in the same manner.

(5) Hon-o gakusei-ga san-satu kat-ta.
book-ACC student-NOM three-CL buy-PAST
A/the student(s) bought three books. (Nakanishi 2b)
(60) Hon-o gakusei-ga takusan kat-ta
Book-ACC student-NOM a.lot buy-PST
The student bought a lot of books.


Numerals and quantifiers can also be floated in a sentence with passive:

(3) a. Yuube, kuruma-ga doroboo ni ni-dai nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PAST
Last night, two cars were stolen by a thief.
b. Kuruma-ga doroboo ni takusan nusum-are-ta.
cars-NOM thief by a.lot steal-PASS-PST
'A lot of cars were stolen by a thief. (Miyagawa 1989: 21, 38)

The two main analyses for this data have traditionally been the stranding analysis
(Sportiche (88), Kawashima (98), Boskovic (2004), etc.) and the adverbial analysis (Miyagawa
(89), Bobaljik (95), etc.) The difference is whether the floated quantifier is left there when the
movement operation happens, in the stranding analysis, or whether the floated quantifier is a
special adverb that can attach only to traces of the moved phrase, in the adverbial analysis. The
two possibilities are schematized below:


STRANDING:







24

ADVERBIAL:


In the stranding analysis, a subpart of the nominal domain (KP in the picture above)
moves out of the QP, leaving behind such information as quantifiers and numerals. This is
usually a product of feature percolation. Feature percolation either percolates the feature
maximally up to the QP, or stays down lower at the KP. Whichever phrase is highest and gets the
necessary feature is the one that moves on up.
In the adverbial analysis, the #P acts as an adverb for traces of QPs that it could
semantically refer to. Bobaljik, in particular, has the process putting #P adverbs into the
structure very late in the syntax. Bobaljik needs the process to be so late in order to know where
all the traces are going to be.

I will take a detour to a variety of English to argue that these floating constructions
cannot possibly be derived via the stranding analysis. McCloskey (2000) argues that all can be
left behind in WH trace locations in a dialect of English known as West Ulster English (WUE).
The pertinent data is as follows:

(61) a. What all did you get t for Christmas?
b. Who all did you meet t when you were in Derry?
c. Where all did they go t for their holidays? (McCloskey 2000: 1)

(62) a. What did you get all for Christmas?
b. Who did you meet all when you were in Derry?
c. Where did they go all for their holidays? (McCloskey 2000: 3)

All can either move up with the WH-word, or it can stay behind. In fact, all can get left
behind in-situ or in any of the CPs that the WH-word moves through. Notice (63):

(63) a. What all do you think (that) hell say (that) we should buy t?
b. What do you think all (that) hell say (that) we should buy t?
c. What do you think (that) hell say all (that) we should buy t?
d. What do you think (that) hell say (that) we should buy all? (McCloskey 2000: 12)
3


3
In case one is worried that this WUE data is faulty or is some small pocket of language, there is a remarkably
similar phenomenon that happens in Standard English (noticed by McCloskey (2000) on pg. 63 Ft. 8):
(i) What exactly did he say that he wanted?
(ii) What did he say exactly that he wanted?
(iii) What did he say that he wanted exactly?
25


To account for this data, McCloskey proposes that the structure for who all is as follows:



Therefore, either the [+WH] KP or the [+WH] QP moves up by WH movement. But this
runs straight into a number of problems. The first problem would be how to deal with a similar
issue. Consider not the WUE data, but the normal English float data:

(64) a. *The carpets all will have been being dusted for two hours.
b. The carpets will all have been being dusted for two hours.
c. The carpets will have all been being dusted for two hours.
d. The carpets will have been all being dusted for two hours.
e. *The carpets will have been being all dusted for two hours. (Sportiche 24)
4


The correlate structure of the phrase [all the carpets] would look like this before
movement:

I do not want to concern myself at this moment whether that KP moves up to spec QP or
not. McCloskey (2000)s story about WUE makes sense. The [+WH] feature either remains on
the KP or pied-pipes to the QP. Whichever one is highest and gets the [+WH] feature is the one
that raises. But what about all the carpets? What feature is raising to give optionality in this
situation? To get the data correctly, you would need the following:

4
I disagree with Sportiches grammaticality judgments. Sportiche marked (e) as good.
26


Under McCloskeys story, you need a [+SUBJ ] feature to get the normal floating data.
Either the [+SUBJ ] feature would percolate up to the QP or not. This would give you the
optionality. This seems far from plausible. For WH movement, it makes sense why that [+WH]
feature is there; it has a WH word. But why is the [+SUBJ ] feature there? What motivates an
actual feature for [+SUBJ ] rather than subject being defined by position? Is this [+SUBJ ] feature
different from TP EPP in any meaningful way? What gives the [+SUBJ ] feature in the deep
structure? This cant be right.
The second problem with McCloskeys story is how all gets left behind in the middle
trace positions. It makes perfect sense in the in-situ position. Lets run through the story and try
to derive (65):

(65) What do you think (that) hell say all (that) we should buy t? (McCloskey 12c)

So in the base position we have the following structure:


(66)


What matters is that the feature, at this point, will have the option to pied-pipe up. If the
feature does not pied-pipe, then we get the structure in (66). This will give up the in-situ all. If
the WH feature does pied-pipe, it gives:






27

(67)


If you get the structure (67), then this QP will have raised successfully to the next
position. Now that it has arrived at the next trace position, can it optionally pied-pipe again?
The structure at this point looks like (67), but somehow, it needs to be able to pied-pipe or not.
How is this possible? Its already pied-piped! McCloskeys story cant be correct.
Trying to avoid some sort of [+SUBJ ] feature and some sort of re-pied-piping process
leads me to reject the stranding hypothesis. So what about the adverbial hypothesis? As I
mentioned earlier, Bobaljik requires the placement of the adverbs to come in at a later time. It
must either be extremely late in the syntax or post-syntactic. But how is this late syntax/post-
syntactic operation able to correctly match up these numeral adverbials with the proper host? To
see that, lets take an example:

(3) Yuube, kuruma-ga doroboo ni ni-dai nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PAST
Last night, two cars were stolen by a thief. (Miyagawa 1989: 21)

You somehow need the CL to match up properly to the trace thats been left behind. The
difficulty of this is partially based upon what you think traces contain. If they are nothing other
than a marker that something has moved out, then it seems very difficult to properly match up
numerals with the proper host. If you think that, then there seems to be no explanation as to why
(68) is bad:


(68) *Yuube, kuruma-ga doroboo ni futari nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PAST
Last night, two cars were stolen by a thief.

In (68), instead of the proper classifier (dai), the classifier for people is put in its place.
And this results in ungrammaticality. If traces were just markers that something moved, this
would be unexplained. There needs to be at least some information so the floated numeral
knows that its cars and not people its attaching to. And that information cant be present
because traces are simply markers that things have moved.
So, the stranding analysis doesnt work because it needs a [+SUBJ ] feature and the ability
to repied-pipe. The adverbial analysis doesnt work because the insertion of the adverbial
quantifier/numeral has to be extremely late syntactically to know where the traces are going to be
and by that point in the syntax the trace will be nothing more than a marker. What one needs to
solve this problem is a theory thats not the feature percolation story and that is similar to the
28

adverbial theory, but where the traces have more inside them. If traces had more internal
structure, then the numeral/numeral quantifier could be properly linked to them. The solution to
this is a copy theory of movement.
In a copy theory of movement, nothing really moves. Everywhere where one would posit
a movement would instead be a copy process, taking the element that would have moved, and
copying to the place it would have moved to. Under this story, subject raising would instead be
subject copying, copying the highest DP into spec TP. Then, there would be some general
principle that if there were multiple copies of the same element, only the highest one would be
pronounced. So, why does this theory fix the problem? Lets see a derivation:


(3a) Yuube, kuruma-ga doroboo ni ni-dai nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PAST
Last night, two cars were stolen by a thief.

Because of VP internal subject raising hypothesis, kuruma would start down below:

(3a) Yuube, doroboo ni ni-dai-no kuruma nusum-are-ta.
last.night thief by 2-CL-LINK car steal-PASS-PAST
Last night, two cars were stolen by a thief.

Then, kuruma would be copied into spec TP:

(3a) Yuube, ni-dai-no kuruma-ga doroboo ni ni-dai-no kuruma nusum-are-ta.
last.night 2-CL-LINK cars-NOM thief by 2-CL-LINK car steal-PASS-PST
Last night, two cars were stolen by a thief.

Then something interesting happens. Unlike normal copy theory, the entirety of the
lower copy doesnt have to be deleted. It could all get deleted. But theres another possibility in
which the numeral and numeral quantifier gets pronounced at the lower copy and is, thus, forced
to not be pronounced at the upper copy. This would result in the final result:

(3a) Yuube, kuruma-ga doroboo ni ni-dai nusum-are-ta.
last.night cars-NOM thief by 2-CL steal-PASS-PAST
Last night, two cars were stolen by a thief.

Why numerals and numeral quantifiers are marked specially to undergo this is still
strange, but this story makes a lot more sense than positing a [+SUBJ ] feature, a repied-piping
option, or an adverbial story in which the numerals and numeral quantifiers know more about the
semantics of the moved thing than should be possible with traces. Though this seems like a nice
way to solve the problem, it still doesnt escape the fact that this story doesnt explain the basic
data still. Lets return to (1) and (2):




29

(1) c. J ohn-wa hon-o san-satu kat-ta.
J ohn-TOP book-ACC 3-CL buy-PST
J ohn bought three books.
d. J ohn-wa san-satu hon-o kat-ta.
J ohn-TOP 3-CL book-ACC buy-PST
J ohn bought three books. (Watanabe 2006: 3c&d)

(2) c. J ohn-wa hon-o takusan kat-ta.
J ohn-TOP book-ACC many buy-PST
J ohn bought many books.
d. J ohn-wa takusan hon-o kat-ta.
J ohn-TOP many book-ACC buy-PST
J ohn bought many books. (Watanabe 2006: 70c&d)

The problem that still remains is why can the numeral/numeral quantifier show up in two
different positions. The solution to this is that theres copying, even in this simple case. There is
independent evidence for an object movement in J apanese that takes the ACC marked DP and
moves it higher. The evidence comes from the following:

(69) a. Sensei-ga Mitiko-san-ni Maeda-san-o syoukai-si-ta
Sensei-NOM Michiko-HON-DAT Maeda-HON-ACC introduce-do-PST
b. Sensei-ga Maeda-san-o Mitiko-san-ni syoukai-si-ta
Sensei-NOM Maeda-HON-ACC Michiko-HON-DAT introduce-do-PST
Sensei introduced Maeda to Michiko.

This shows that there can be object movement even in simple cases like (1) and (2),
which means that there are two positions for the numeral/numeral quantifier to attach to.

If you follow the stranding story, you need some sort of [+SUBJ ] feature and you need
some sort of re-pied piping ability, both of which seem outlandish. If you follow the adverbial
story, you need late insertion attaching these elements onto traces, but that seems ridiculous
considering how difficult properly matching the semantics of these numerals/numeral quantifiers
to their traces is. Instead, the copy theory of movement has been shown to be superior and
derives all the data.


VIII. Conclusion:

In this paper, I have investigated numerals and numeral quantifiers in J apanese. I have
concluded that there are three positions that numerals and numeral quantifiers can appear in: they
can be left behind in trace positions by acting special with regard to copy theory, they can appear
as right modifiers to NP, and they can appear as a subpart of N as part of a N word building
process. The biggest question that remains is why. Why is the numeral able to appear as a right
modifier to NP but not the numeral quantifier? Why are numerals and numeral quantifiers so
special when it comes to copy theory? Why numerals and numeral quantifiers are able to appear
in this noun formation process seems easy enough. This word formation process is extremely
30

extensive in the language and semantically-meaningful elements can be built with other such
elements and create nouns. Areas for further research include how to better understand those
questions of why and to try to see how extensive this three-way distinction in numeral/numeral
quantifier positions is cross-linguistically.



References:

Bobaljik, J onathan. 1995. Morphosyntax: The Syntax of Verb Inflection, Doctoral dissertation,
MIT.

Boskovic, Zeljko. 2004. Be Careful When You Float Your Quantifiers, Natural Language and
Linguistics Theory 22, 453-480.

Kawashima, Ruriko. 1998. The Structure of Extended Nominal Phrases: The Scrambling of
Numerals, Approximate Numerals, and Quantifiers in J apanese. Journal of East Asian
Linguistics 7, 126.

McCloskey, J ames. 2000. Quantifier float and wh-movement in an Irish English. Linguistic
Inquiry 31: 5784.

Miyagawa, Shigeru. 1989. Light Verbs and the Ergative Hypothesis. Linguistic Inquiry 20:659-
668.

Nakanishi, Kimiko. 2008. The syntax and semantics of floating numeral quantifiers. In The
Oxford handbook of J apanese linguistics, ed. Shigeru Miyagawa and Mamoru Saito, 287319.
Oxford: Oxford University Press.

Saito, Mamoru, T.-H. J onah Lin, and Keiko Murasugi. 2008. N'-Ellipsis and the structure of
noun phrases in Chinese and J apanese, Journal of East Asian Linguistics 17.247-271.

Saito, M. & K. Murasugi. 1990. N-deletion in J apanese: A preliminary study. In
J apanese/Korean linguistics 1, ed. H. Hoji, 285301. Stanford, CA: CSLI Publications.

Sportiche, D. 1988. A theory of floating quantifiers and its corollaries for constituent structure.
Linguistic Inquiry 19:425-449.

Watanabe, Akira. 2006. Functional Projections of Nominals in J apanese: Syntax of Classifiers.
Natural Language and Linguistic Theory 24: 241-306.

Watanabe, A. 2009. Vague quantity, numerals, and natural numbers. Syntax 13:3777.