1

April, 2005


On the Syntax of Old Japanese
S.-Y. Kuroda
University of California, San Diego and Internaional Institute for Advanced Studies, Kyoto

1. Introduction
In the late nineteen eighties, I introduced a parameter Agreement which take one of the two
values, Forced or Non-Forced.
1
This parameter was assumed to control a variety of grammatical
phenomena such as wh-movement, subject and object Abstract Case marking. I maintained that
English is a Forced Agreement language, while Japanese is a Non-Forced Agreement language,
and claimed that the opposite value settings for this parameter account for major typological
differences between two languages.
In this paper, I would like to demonstrate that Old Japanese was a Forced Agreement
language. Old Japanese, I claim, had obligatory Abstract Case marking, both for subjects and
objects. It had obligatory Focus Movement; as a consequence, we can also recognize the
existence of an obligatory wh-movement as a special case of this focus movement. Old Japanese,
unlike Modern Japanese, lacked head internal relative clauses; this fact can also be accounted for
by means of obligatory movement of the relative clause head, as opposed to the optionality of
this movement in Modern Japanese. In all these respects, Old Japanese was rather like English.
However, the Case system in Old Japanese was not an accusative system like in English. I would
like to claim that the Case system in Old Japanese was neither accusative nor ergative, but rather
a system that conceptually neutralizes this division, a system unmarked as to accusative or
ergative.

1
See Kuroda 1986, 1988 (cited henceforth as WhorN). This paper is a much revised version of a paper presented at
The Second Linguistics Seminar-International Symposium: the History and Structure of Japanese, held at St
Catherine's College, Oxford, Kobe Institute, September 29, 2004. I would like to thank the organizers of this
conference for providing me with a chance to work on the history of Japanese. I would also like to express my
sincere gratitude to Satoshi Kinsui, Yasuhiro Kond, Takashi Nomura, Akira Watanabe, Janick Wrona, Yuko
Yanagida and in particular Shigeo Tonoike, who helped me in various ways during the preparation of this paper.
2
Akira Watanabe in a series of recent papers claimed that Old Japanese had wh-movement
and more generally a focus movement.
2
In this respect this paper follows his work and
incorporates it as a special case of a more general claim. However, his claim on wh-movement is
contingent on the assumption that bare noun phrases that precede wh-phrases in Old Japanese are
topics, but he did not provide a formal proof of this assumption. I will show that such bare noun
phrases are indeed topics. The proof of this fact is intimately connected to the claim that the Case
system of Old Japanese was unmarked.
Another difference between this work and Watanabe's on wh-questions in Old Japanese
must be noted. I recognize two different wh-question constructions in Old Japanese. One type
involved obligatory movement, as special case of focus movement. There was another type of
wh-questions in Old Japanese which did not involve movement: wh-phrases in this construction
stayed in situ and, I assume, was licensed directly by binding. In fact, I would assume that the
wh-question construction in Modern Japanese is a descendant of this type of wh-question.
I will first review the hypotheses and the claims made in [WhOrN] about Modern
Japanese and add one phenomenon that was not discussed in it but can also be accounted for by
the same hypotheses. Then, I propose the hypothesis according to which the Agreement
parameter was set for Old Japanese at the value opposite to the one for Modern Japanese. For
this paper, I mostly follow a rather classical framework of the so-called Government and Binding
theory as a vehicle of exposition.
For ease of later reference, I will formulate the major hypotheses that I maintain hold for
English and Old and Modern Japanese.

(1) Main Hypotheses
[1] The Forced Agreement Hypothesis for English
English is a Forced Agreement language.
[2] The Non-forced Agreement Hypothesis for Modern Japanese
Modern Japanese is a Non-Forced Agreement language.

2
See Watanabe 2001, 2002a, b, 2003.
3
[3] The Forced Agreement Hypothesis for Old Japanese
Old Japanese was a forced Agreement language.

By Old Japanese, I designate the language recorded in such documents as Nihonshoki and Kojiki,
the two earliest collections of chronicles, and Manysh , an anthology of poems, all compiled in
the 8th Century. I use the term Classical Japanese to refer to a later stage of Japanese as recorded
in the Heian literature, the language served as the model of the literary style of writing until the
mid 19th century.

2. Modern Japanese as a Non-Forced Agreement Language
2.1. General Remark 1: Agreement and Movement
Agreement, as it was conceived in [WhOrN], is a specified interaction between a head X,
in the sense of X-bar theory, and a maximal category YP governed by X, either at an external
(i..e., Specifier or subject) or an internal (i.e., object) position. As conceptual possibilities, the
interaction involved in Agreement could be formalized either in the transformationalist approach
in terms of derivation or in the lexicalist approach in terms of constraint, i.e., either as
derivational feature specification or as selectional feature constraint, acted on a maximal
category by an element in a head position. For ease of exposition I use the transformationnalist
terminology and call the head and the maximal category involved in Agreement the trigger and
the target of Agreement.
For Agreement to take place, a target must be found at an expected position. However, it
may not originally be generated at this position. Under such circumstances, Forced Agreement
appears to take the form of obligatory Movement; one might equivocate Forced Agreement and
obligatory Movement. But Agreement and Movement are conceptually independent. Movement
is in general subject to certain constraints by the principles of grammar; then, if obligatory
Movement fails due to constraints imposed on it, so does Forced Agreement, and the derivation
clashes with the consequence that the expected sentence form is doomed as ill-formed.

2.2. Wh-questions
4
In English a wh-phrase must be put at the beginning of an interrogative clause while in
Modern Japanese wh-phrases can remain in situ. We capture this difference in terms of
Agreement between the wh-phrase in Spec(C) and the interrogative complementiser in C. In
English, the Agreement is forced, hence a wh-phrase must be present at Spec(C); in
consequence, a wh-phrase must be moved from an argument or adjunct position to Spec(C). In
Japanese, Agreement is not forced, and as a consequence, as is well known, a wh-phrase may
remain in situ at an argument or adjunct position.
Since Agreement is forced in English, wh-Agreement entails the existence of obligatory
wh-movement. In contrast, Non-Forced Agreement does not entail the existence of non-
obligatory wh-movement. Wh-movement, if it exists in Japanese, must be optional, but whether
it exists or not is an empirical question. At the level of simple observation, the fact is consistent
with either the existence or the absence of wh-movement: wh-XP can be fronted, but this
fronting can be achieved by scrambling, a process required independently of accounting for wh
questions. Consequently, either of these possibilities, the existence or the absence of wh-
movement, is open for argument. Indeed, arguments have been presented for both possibilities in
the literature: Takahashi (1993) argues for the existence of wh-movement in Japanese on the
basis of the radical reconstruction phenomenon observed with scrambling by Saito (1989);
Kitagawa and Deguchi (2002) question Takahashi's argument on the basis of their analysis of
prosody associated with wh-questions.

2.3. General Remark 2: VP Internal Subject Hypothesis
I now wish to move on to the issue of Case Marking as Agreement, but before going into
specifics, I need to add another preliminary general remark. In [whOrN], I introduced the VP
Internal Subject Hypothesis. It assumes that the verb, and in general, any of the lexical categories
N, P, A, as well, has its own Spec position, where its subject, if there is any, is originally
generated and theta-marked by it. This hypothesis contrasts with the classical as well as the
present standard assumption on the structure of simple clauses. In the classical standard
assumption, V is defective and lacks a Spec position; the subject of V is generated and theta-
marked at Spec(I). The VP Internal Subject Hypothesis removed this asymmetry. In the later
5
development of the Government and Binding theory, the V-shell scheme was introduced; the V-
shell partially incorporated the idea that motivated the VP-Internal Subject Hypothesis and may
have been taken as a substitute for it.
The V-shell scheme went hand in hand with another later development in the
Government and Binding theory, according to which object case marking is also executed at
Spec of an Agr position. This expanded Agr scheme appears to provide our Agreement approach
with a conceptual advantage: we can assume that Agreement is uniformly an operation between
an Agr head and its Spec position.
The V-shell scheme appears to have some functional plausibility, to the extent that
transitive verbs are syntactically decomposed into the unergative and unaccusative components.
But not all transitive verbs are functionally causative. Nor is the lexical decomposition of
causative transitive verbs, even if desirable, inconsistent with the VP-Internal Subject Hypothesis
in the original form; it is simply a matter of how causative transitive verbs are analyzed. For
now, then, I take a conservative stand and will steer clear of the V-shell and the expanded Agr
scheme and stay with the VP-Internal Subject Hypothesis in the original form for this
presentation of the Agreement Hypotheses.

2.4. Case Marking
2.4.1. English Case marking
With the VP-Internal Subject Hypothesis at hand, then, the direct object of a transitive verb is
Case Marked as its internal argument. This Object Case Marking is obligatory in English,
though phonetically visible only for personal pronouns. I take it as an instance of Forced
Agreement, between a transitive verb and its internal argument.
The subject of a verb is Case Marked as a subject only in finite, tensed clauses, at
Spec(I), not at Spec(V). In English, this Subject Case Marking is obligatory, as it is manifest in
personal pronouns as subjects of tensed verbs; it is taken as another instance of Forced
Agreement, between a finite Infl and Spec(I). Forced Subject Agreement entails the obligatory
movement of a subject DP from Spec(V) to Spec(I) in a finite clause.

6
2.4.2. Japanese Case Marking
2.4.2.1. Bare DPs vs. ga/o marked DPs: Case Particle "Drop"
Turning our attention to Modern Japanese, in conformity with Non-Forced Agreement
Hypothesis for Modern Japanese, I claim that Subject and Object Case Marking is not forced in
Japanese. But for a proper understanding of this hypothesis, it is in order here to explain what I
mean by Case marking.
The particle ga and o are usually identified as nominative and accusative case markers in
the literature of Japanese linguistics. It is also common that Japanese linguists talk about "case
drop": ga and o may optionally drop and as a consequence subjects and objects may be bare DPs,
without ga or o attached to them. It appears that subject and object case marking is optional in
Japanese. A plausible step to take for implementing the Non-Forced Agreement Hypothesis for
Modern Japanese might appear to be the assumption that ga and o are the targets of Agreement,
but since Agreement is not forced, DPs can appear as subjects and objects without these
particles. But this is not my claim.
I distinguish abstract Case and morphological case. Abstract Case in Japanese is
phonologically null; DPs marked as subjects or objects are bare, without particles, like English
common nouns. The particle ga and o are morphological case markers. Agreement concerns
Abstract Case; bare DPs are targets of Agreement by the finite Infl and the transitive V,
respectively. Morphological case marking is a separate process that licenses DPs as arguments.
Since Case Agreement is not forced, a subject or object DP may fail to be licensed as a bare DP.
But it may still be licensed by morphological case by means of ga or o.
If DPs' being bare or marked with ga or o is simply a matter of case drop, ga and o
marked DPs should freely alternate with bare DPs. In other words, ga and o as subject and
object markers must appear to "drop freely". This prediction, however, does not hold; the fact is
that we observe more constrained distribution for bare DPs as subjects or objects than for ga or o
marked DPs, as we will see below.

2.4.2.2. The morphological case ga
7
Certain subordinate clauses appear to abhor bare subjects: Observe
3
:

(2) mosi asita paatii ni Masao *(ga) ku-reba
if tomorrow party to Masao ga come-if
'if Masao comes to the party tomorrow'
(3) oya *(ga) kane (o) mookete i-nagara
parent ga money o make be-though
'even though a parent is making money'

This skewed distribution of bare DPs is precisely what is predicted if we assume that a bare DP
is licensed as a subject at Spec(I) by a finite Infl in Japanese, as in English. The subordinate
clauses in the above examples are plausibly non-finite. In these clauses ga phrases function as
subjects. But this situation should not surprise us. In English, subjects are licensed by genitive
case or by preposition of in gerund phrases and derived nominals, not by abstract Case. It is in
fact worthy of note that historically ga was a genitive case marker.
Two sets of facts seem to counter this analogy between Japanese and English. First of
all, subordinate clauses that disallow bare DP subjects extend beyond what is usually taken as
tenseless clauses: Observe:
(4) Hanako wa Masao *(ga) gohan (o) tukuru-to yorokobu
Hanako wa Masao ga meal o make-if is-pleased
'if Masao cooks Hanako is pleased'
(5) Hanako wa dareka *(ga) yaki-imo (o) katte kita-node yorokonda
Hanako wa someone ga baked-yam o buy come-since was-pleased
'Hanako was pleased since Masao bought and brought baked yams..'

In these clauses, the verbs are in the so-called present and past forms. They are apparently
tensed, but ga may not "drop." In earlier work, however, I pointed out that there is some fact,

3
I will generally not gloss case markers, particles and auxiliaries. Not only is glossing them cumbersome, but also
casual tags for grammatical morphemes could be misleading. I decide that providing translations to examples would
suffice for the understanding of the role of functional items in examples.
8
marginal though it may be, that suggests that the so-called present ending is not a finite tense
marker; see Kuroda (1990, 1993). As far as the so-called past ending ta is concerned, it is well
known that its function is not easy to characterize. It is also well known that ta has an aspectual
character. Aspect markers are not incompatible with non-finite clauses. Then, it is possible that
the clauses in the above examples with verbs in ru and ta endings are non-finite subordinate
clauses and their Infls are incapable of licensing bare DPs with abstract subject Case.
Secondly, we face an opposite problem. Not only subordinate clauses but main clauses
may have ga phrases as subjects. Observe the following contrast exhibited by subordinate and
main clauses:

(6) mosi asita paatii ni Hanako *(ga) ku-reba, Masao (ga) kitto yorokobu yo
if tomorrow party to Hanako ga come-if Masao ga surely pleased yo
'if Hanako comes to the party tomorrow, Masao will surely be pleased
(7) gakusei *(ga) yaki-imo (o) katte kita-node, sensei (ga) yorokonde sore (o) tabete iru
student ga baked-yam o buy come since teacher ga pleased it o eat is
'since students bought and brought baked yams, the teacher is please and eating it.'

We compared ga phrases in Japanese subordinate clauses with genitive subjects in
English gerundive clauses. This comparison breaks down here, but the comparison happens to be
a superficial analogy. A more significant way of describing a parallelism between the situations
in English and Japanese would be this: the English gerundive clause cannot Case mark its
subject by its Infl, but the genitive case is available to license its subject. Likewise, in Japanese,
the non-finite clause cannot mark its subject by abstract Case but morphological case ga is
available to license its subject. Unlike English, however, even in the finite sentence its Infl may
leave its subject unmarked since Agreement is not Forced, But, then, the morphological case ga
is available to license it. Hence the possibility of main clauses with ga marked DPs is accounted
for.

2.4.2.3. The morphological case o
9
In contrast to subjects, object DPs are Case-marked in situ as an internal argument of V. Whether
a clause is finite or non-finite does not matter for object Case Agreement; abstract Case marking
is always available in situ. For this reason, o cannot be shown to be a morphological case marker
in a simple way as ga can as a matter determined by the type of Infl. Is there any construction
where bare DP objects and o-marked objects do not alternate?
I contend that bare DP objects cannot be moved to sentence initial position by
scrambling. However, some caution is necessary for confirming this fact. For, bare DPs may be
moved to sentence initial position by topicalization. As a consequence, we cannot rule out the
possibility that a sentence has a sentence initial bare DP object. We need to look for some
conditions under which we can formally differentiate effects of bare DP topicalization and
scrambling.
First, observe the following forms:

(8) a. Masao Harvard e Yamada sensei ga suisen-sita
b. Masao o Harvard e Yamada sensei ga suisen-sita
'Teacher Yamada recommended Yamada to Harvard'
(9) a Ma no Yama MIT no gakusei ga san-nin katte itta
b. Ma no Yama o MIT no gakusei ga san-nin katte itta
'Three MIT students bought Magic Mountains'

Both forms are possible. I contend that Masao and Ma no Yama in the a-sentences must be
topics. Indefinite nouns are hard to topicalize. The prediction is that if we substitute dare 'who'
for Masao and nani 'what' for Ma no Yama, respectively, we should get a contrast between the a-
and b-sentences and this predication is confirmed:

(10) a *Dare Harvard e Yamada sensei ga suisen-sita?
b Dare-o Harvard e Yamada ga sensei suisen-sita?
'who did Yamada sensei recommend to Harvard?'
(11) a *nani MIT no gakusei ga san-nin katte itta?
10
b. nani o MIT no gakusei ga san-nin katte itta?
'what did three MIT students buy?'

Secondly, relative clauses cannot have topics. The prediction is that bare DP objects cannot be
preposed inside relative clauses. Observe:

(12) a *Harvard ga Masao Yamada sensei ga siusen sita daigaku da
b Harvard ga Masao o Yamada sensei ga suisen-sita daigaku da
'Harvard is the university to which Yamada-sensei recommended Masao'
(13) a *Coop ga Ma no Yama MIT no gakusei ga katte itta honya da
b Coop ga Ma no Yama o MIT no gakusei ga katte itta honya da
'Coop is the bookstore where MIT students bought Magic Mountain'

In sum, if we filter out the effect of topicalization, sentence-initial position is where we can
observe distributional asymmetry between bare and o-marked objects.
In my earlier work, I accounted for this asymmetry as a consequence from the following
hypotheses: (i) the landing site of scrambling is Spec(I); (ii) bare DPs get the subject Case at
Spec(I) and the object Case as an internal complement of V; (iii) Case-marked DPs may not
move to Case position. From these hypotheses it follows that a bare DP object may not
scramble. The motivation behind (iii) is to prevent a DP to get two Cases. However, I now
assume, as I stated above, that there are clauses usually taken as tensed, for example relative
clauses, that are non-finite and whose Spec, hence, is not a position for subject-Case marking.
But o-marked objects, but not bare objects, may scramble in such clauses, too; see (12), (13)
above. We can still keep the hypothesis (iii) with the understanding that Spec(I) is a Case-
position independently of whether I is finite (and hence Case-marks) or not. Doing so, however,
invalidates the original motivation behind (iii), that is, the prohibition against double Case
marking. There is, however, an alternative account of the asymmetry between bare and o-marked
objects we have observed above: the adjacency condition. The object DP must be adjacent to the
verb in order to be Case-marked. Exactly how to formulate this condition and make it viable is
11
somewhat of a problem. Be that as it may, I take it that the asymmetry in scramblability is a good
indication that bare objects and o-marked objects are licensed by different mechanisms, abstract
Case and morphological case, respectively. Thus, since bare DPs and o-marked DPs are
alternants as direct objects, I conclude, Agreement is not Forced for object Case marking, either.

2.4.2.4 Concluding remarks
To recapitulate, I maintain that bare DPs are licensed by Agreement, by finite Infl as
subjects and by transitive verbs as objects. But unlike in English, this Agreement (Case
assignment) is not Forced. Morphological case may also license DPs as subejects and objects
with ga and o, respectively.
The particles ga and o may not "drop" in formal style of writing and speech. I assume
that this is a constraint imposed on the language from outside of core grammar, largely a matter
of style or register management.
Incidentally, bare DPs may also be derived from topic wa phrases by "dropping" wa.
Thus, one could dispute my above argument by claiming, as Kuno (1973a:223ff) did, that all
bare DP subjects are derived by "dropping" wa. Indeed, my point above that ga may not
alternate distribtionally with zero in subordinate clauses might even be cited as evidence for the
claim that wa may, but ga may not be deleted. It is hard to counter this claim purely on
formal/distributional grounds. In order to argue against this objection, we would have to use
informal arguments based on intuitive judgment about a proper use of a given example in a
plausible context: in the suggested context, the bare DP in question may be replaced by a ga
phrase. Masunaga (1988) argued against Kuno's generalization in this manner effectively,
though I do not necessarily agree with Masunaga's judgments for all the examples she discusses.

2.5. Relativization
Basically following Kuno's (1973b: 254) insight, I assume that wa topicalization and
relativization are derivationally related, sharing a common step. However, I do not assume, as
Kuno did, that relative clauses are derived from wa-topicalized sentences. I assume rather that
relative clauses as well as wa-topicalized sentences involve a movement, as in the English
12
relative clause formation, the movement of an argument/adjunct from a clause internal position
to a clause external position. See Kuroda (1992: Ch9). In our present approach, we can assume
that this movement is effectuated by Agreement between a Complementiser and its Specifier
position, Spec(C).
But according to the Non-Forced Agreement Hypothesis for Modern Japanese,
Agreement is not forced, that is, the assumed movement must be optional. This prediction is
indeed borne out for relativization. As is well known, unlike English, Japanese allows Head
Internal Relative clauses. Details of exactly how to implement this idea need not concern us
here. Let me just note that the idea was already suggested by Harada (1973).
The Non-forced Agreement Hypothesis appears to fail for wa-topicalization, since the
wa-topic must be at sentence-initial position. One might take this as a vindication of the base-
generation approach, as opposed to the movement approach, according to which wa topics are
base generated in situ. But let me note that the forced movement for wa-topics has a basis of its
own independently of the Forced/Non-Forced Agreement parameter. The Spec(C) position can
and must be assumed to be semantically not vacuous for topics, even though it is theta-
theoretically vacuous. The movement of a topic wa phrase to Spec(C) position is forced on
semantic grounds in order for the sentence to be properly interpreted as intended.

3. Old Japanese
3.1. Sentence types and kakari musubi
Excluding imperative sentences, there are five types of clauses, with different conjugation forms
of main predicates:

Shshi conclusive
Rentai adnominal
Izen realis/presuppositional
Mizen irrealis
Reny adpredicative

13
Independent clauses can take either the shshi, rentai or izen form. Adnominal subordinate
clauses, in particular, relative clauses, take the rentai form. Hence the name rentai; cf taigen
'substantive' as opposed to ygen 'predicative'.
The default independent sentence form is the shshi form. However, certain particles
select the rentai or izen form. This phenomenon is commonly known as kakari-musubi
(suspense-resolution). The particles involved are called kakari particles.

(14) Kakari particles
With rentai resolution:
so/zo (focus/emphasis), namu (focus/emphasis), ya (question), ka (question)
With izen resolution:
koso

We are concerned primarily with ka. But some of the properties we discuss below are shared by
other kakari particles with rentai resolution. To follow what follows, it is sufficient to be aware
of the existence of this phenomenon. However, since, kakari musubi is one of the major issues
in the traditional grammar, I add an appendix on this matter at the end of this paper.

3.2. Case Marking
3.2.1. Bare DPs as Case Marked Subjects and Objects by Agreement
What prediction does the Forced Agreement Hypothesis make about Case marking? We have
maintained above that Agreement manifests itself in the form of bare DPs as subjects and objects
in Modern Japanese, not as in the form of ga or o marked DPs. To recall, one may not "drop"
morphological case markers ga and o in Modern Standard (formal) Japanese. One does, only in
informal, colloquial speech. We may get the impression that "case dropping" is a symptom of
corrupt speech. But bare DPs are common as subjects and objects in bungo, the literary style
based on Classical Japanese and used in the texts of Japanese literature from the Heian period
until the modern times. The status of bare DPs as subjects and objects is solid and invariant
through the history of the Japanese language. Thus, the null hypothesis would be that subject and
14
object Case marking has realized in the form of bare DP, as it does in English with common DPs,
through the history of Japanese:

(15) The Case Hypothesis
Both subject and object Case actualize as bare DPs through the history of Japanese.

In contrast, the status of ga and wo/o through the history of the Japanese language is the issue
we need to concern ourselves with. We need to deal with ga and wo separately.

3.2.2. ga Marked Subjects
It is well known that ga marked subjects occur only in limited contexts in Old Japanese. In
general they do not occur in clauses with predicates in shshi form. Most prominently, they
occur in "adnominal" contexts, where the predicates take the rentai form. Furthermore, unlike in
Modern Japanese, in Old as well as Classical Japanese, ga and no shared the grammatical
function of genitive, though apparently they are not free alternants in all contexts,
distributionally and in terms of pragmatic/semantic functions. This sharing of a grammatical, if
not pragmatic, function by ga and no extends to main clauses with predicates in the rentai form.
In general no may mark subjects where ga may, whether in main or subordinate clauses.

3.2.2.1. ga Marked Subjects and the Non-Finiteness Assumption
A parallelism between English and an older stage of Japanese suggests itself. In English the
genitive case functions not only as a marker of a broad relation that holds between two DPs, but
also as a subject marker in clauses where the predicates take a non-finite gerundive form. Let us
introduce the following assumption for Old Japanese:


(16) Hypothesis OJ-1. Finite and non-finite clauses
[a] Clauses whose predicates are in the shshi form are finite.
[b] Clauses whose predicates are in the rentai form are non-finite.

15
These assumptions, with the Forced Agreement hypothesis, make the following four correct
predications:

(17)
[i] A bare DP can be the subject of a clause with a shshi predicate.
[ii] A bare DP cannot be the subject of a clause with a rentai predicate.
[iii] A ga marked DP cannot be the subject of a clause with a shshi predicate.
[iv] A ga/no marked DP can be the subject of a clause with a rentai predicate.

I take [i], [iii] and [iv] non-controversial. [ii] appears to be contradicted by many
counterexamples. The next section addresses this issue.

3.2.2.2. Bare subjects in rentai clauses are topics
Watanabe (2001.7:100) cites Nomura's (1993) count of 30 examples of (yes-no or wh-)
questions in which a bare DP subject precedes a ka-phrase. Such a sentence ends in a rentai
predicate due to kakari musubi, hence is a counterexample to (17)[ii]. Watanabe states that he is
"tempted to conclude that [these examples] are similar to those where subjects are topicalized
[by ha] and the particle has dropped from the subjects, but I for now leave this matter for future
studies." I would like to argue that Watanabe was in fact tempted to the right direction. I will
argue for the following hypothesis;

(18) Hypothesis OJ-2.
Bare subjects in rentai sentences are topics.

Let us first confirm that the subject of a sentence with a rentai predicate can be topicalized by
ha:
4


(19) M154
Sasanami no Oho-yama-mori ha ta ga tame ka yama ni sime yuhu kimi mo ara-naku ni

4
Old Japanese examples examples are from Manysh unless otherwise indicated. I transcribe Old Japanese texts
by Hepburn Romanization of the kana reading given to the original by the editors of the texts I consulted. Thus, the
Romanization is not phonetically faithful. The OJ particles ha and wo correspond to Modern Japanese wa and o,
respectively.
16
Sasanami no Big-mountain-guardian ha who ga for ka mountain at sign tie you mo are-neg sfx
'For whom does the guardian of Sasanami's mountains put sacred signs, now that you, my Load,
are no longer?'

In this example, the ha phrase Sasanami no Oho-yama-mori ha is a topic.
Hypothesis OJ-2 (18) claims that a bare DP can also be a topic in a rentai clause.
Assume, to the contrary, that a bare DP can be a subject in a rentai clause without being a topic.
In other words, a bare DP can be licensed as a subject without being a topic in a rentai clause as
in a shshi clause. Then, we would expect that such a bare subject can also appear in a relative
clause, which is a rentai clause. Hence, we can take Hypothesis OJ-2 as in effect equivalent to
the following hypothesis:

(20) Hypothesis OJ-2'.
No relative clause has a bare subject.

I will claim that this hypothesis is upheld.
A look at Old Japanese texts would suffice to give us the impression that this hypothesis
cannot be correct. However, remarkably, a closer examination leads us to the following
generalization:

(21) No relative clause whose main verb is transitive has a bare subject.

In other words, if a relative clause whose main verb is transitive has a subject in it, the subject is
marked by ga/no.
Before proceeding further, let us agree on a terminological point. An intransitive
predicate can have more than one argument; one of them gets case- or Case-marked as a subject
at the surface level, and others, if there are any, are inherently case-marked or appear as
prepositional phrases. Let us call the former as the UNMARKED argument of an intransitive verb.
By intransitive predicate, I mean intransitive verb or adjective. This terminological convention
17
allows us to refer to this argument at an observational level without committing ourselves to any
analysis.
In contrast to (21), if the main verb of a relative clause is intransitive, we observe the
following:

(22) If the predicate of a relative clause is intransitive, its unmarked argument, if it appears
inside the relative clause, is either a bare DP or a ga/no marked DP.

Hypothesis OJ-2 (18) predicts Hypothesis OJ-2' (21) but apparently is contradicted by (22).
However, as far as I know (21) has virtually no exceptions. I have examined Nihonshoki and
Kojiki Kay and Manysh Volumes 1, 5, 14, 15, 17, 18 and 20. The following table shows the
number of (possible) exceptions to (21) and the number of relative clauses with transitive verbs
whose subjects are marked with no or ga.
5


(23) Nihonshoki 2 18
Kojiki 1 20
Manysh 0 33


5
The two possible exceptions in Nihonshoki are poems no. 79 and 80. Nihonshoki 79 is given below following the
interpretation in Nihon Koten Bungaku Taikei Vol 3:

Yamanobe no Kosimako yue ni hito-derahu uma no ya-tugi ha wosikeku mo nasi
'For the sake of Yamanobe no Kosimako, I don't feel sorry to lose eight horses that a man is showing off/proud of'.

Nihon Koten Bungaku Zensh, however, has hito nerau uma 'eight horses that men are pursuing', instead of hito-
derau uma 'a man is showing off/proud of'. There are two points worthy of note. First, if we follow the reading of
Nihon Koten Bungaku Taikei, the rendaku (sequential voicing) form -derau of the verb in question may indicate that
we have here a compound word; then we do not have an exception to (21). If we follow Nihon Koten Bungaku
Zensh and take the verb to mean 'pursue', would it not be possible to interpret hito as the object of the verb which
was intended by the poet to refer to himself? Then we would have a plausible interpretation in the context in which
the poem was composed: 'Because of Yamanobe no Kosimako (whom I violated), [they] pursue me. I would not feel
sorry to lose eight horses[, if that is a deal they want]'. With this interpretation, poem no.79 is not an exception to
(21). (I am grateful to Janick Wrona for bringing my attention to this poem.)
Poem no. 80 may not be a counterexample, either, but I am not going to discuss it here.
The only possible exception in Kojiki is found in song no. 47. But it is said to be a song by an indigenous ethnic
group in Yoshino.
18
On the strength of (21) we must conclude that Hypothesis OJ-2' is upheld, in spite of the
apparent conflict of this conclusion with (22). I conclude that predicates in the rentai form do
not assign Case to their subjects.
Thus, I take how to account for (22) as a fundamental issue in the syntax of Old Japanese.
Now, two questions arise:

(24)
[i] In a relative clause whose predicate is intransitive, what determines the choice between a bare
and a ga/no-marked DP for the unmarked argument if it appears in it?
[ii] What licenses a bare DP argument in such a relative clause?

As for [i], I cannot rule out the possibility that more detailed philological study of texts than I
have done so far may reveal some interesting lexical, stylistic and/or diachronic variations that
are relevant for the distribution of the two types of DPs in question.
6
For now, however, I assume
that they are distributionally free alternants and address myself to the question [ii].
A plausible answer to question [ii] is to assume that the bare DP in question is licensed
not as a subject, that is, as an external argument, but as an internal argument of the verb; this
condition is satisfied if we assume that Old Japanese is an ergative language. Let us pursue this
line of thought.

6
I give below a tentative count of bare and ga/no subjects in relative clauses with intransitive predicates. My count
is very incomplete but there is some intriguing variations in the ratio of bare subjects and ga/no subjects in different
volumes of Manysh. I am not in a position to make any comment on this matter, however.
Vol Bare ga/no
Nihonshoki 6 5
Kojiki 3 4
Manysh 1 10 7
5 7 10
14 7 9
15 10 8
17 13 23
18 12 9
20 8 18

19
When we have discussed the Case marking of subjects and objects in the shshi form of
sentences, we take it for granted that the language has an accusative Case system. However,
abstract Case is actualized by bare DP both for external and internal arguments, and no overt
trace is left for the distinction between subject and object Case. Hence, as far as the Case
marking mechanism in shshi clauses is concerned, there is no empirical evidence for
determining whether the unmarked argument of an intransitive verb is Case marked as an
external argument or an internal argument. The ergativity assumption is compatible with the fact
about the subject Case marking in transitive clauses as well. We can thus account for bare DPs
in rentai (i.e., non-finite) clauses as well as shshi (i.e., finite) clauses by assuming that Old
Japanese is an ergative language.
Let us now return to the ga/no case marking mechanism. Recall that ga/no is a genitive case
marker. We can entertain two alternatives, an accusative or an ergative case system: we can
assume that the genitive case is assigned either to external arguments (the accusative hypothesis)
or to absolutive arguments (the ergative hypothesis).
It might at first appear that the latter alternative should be chosen: we could then assume
that the language is uniformly ergative: both the subject and object Case marking and the
genitive marking in rentai clauses are uniformly assumed as ergative. But I would argue to the
contrary that the accusative hypothesis for the ga/no marking provides us with a simpler
grammar for two reasons.
First of all, under the ergative hypothesis the unmarked argument of an intransitive verb
in a relative clause gets generated as an internal argument whether it is licensed by the abstract
Case and appears bare or is licensed by the genitive case and appears as a ga/no marked phrase, a
symptom of indeterminacy abhorred by Forced Agreement. Anticipating that the Forced
Agreement Hypothesis is vindicated for Old Japanese for several other grammatical phenomena,
the ergative hypothesis should be rejected.
In contrast, under the accusative hypothesis for the ga/no case marking, we would have to
assume that the unmarked argument of an intransitive predicate is generated either internally and
gets licensed by Case as a bare DP or is generated externally and gets marked genitively as a
ga/no phrase. We seem to be simply trading the indeterminacy of Case/case marking at an
20
internal position with the indeterminacy of where the unmarked arguments of intransitive
predicates get generated. Nonetheless, I would argue that the latter is a more innocent kind of
indeterminacy. I need to proceed to the second reason for choosing the accusative hypothesis for
the ga/no marking.
To recall, for shshi finite clauses, either an accusative or an ergative Case system will
do. But, then, it may be assumed instead that an accusative and an ergative Case system coexist
as free alternants for shshi finite clauses. If we apply the accusative system for the ga/no
marking, we can also assume that the unmarked argument of an intransitive clause may be
generated either as an external or an internal argument. The unmarked argument of an
intransitive clause may be generated either as an external or an internal argument. Thus, we can
conclude that Old Japanese is at the same time an accusative and an ergative language. If a
clause is shshi and hence finite, the argument is actualized as a bare DP in any case. If a clause
is rentai and non-finite, then it is actualized in the genitive case or bare, depending on whether it
is generated as an external or an internal argument. In this sense, the language might be
characterized as a mixed Case/case system.
However, I would like to take a step further and claim that the situation in question
deserves a better characterization. What we have is a situation where the opposition of an
accusative and an ergative system is neutralized. If a grammar is not specified either accusative
or ergative, there is no constraint as to where the unmarked argument of an intransitive predicate
be generated, either as an external or an internal argument. This is a minimally specified system.
Put it another way, it is an arche-system, representing a stage conceptually before an accusative
and an ergative system diverge from it. Let me formally introduce the notion of arche-Case/case
system and formulate a hypothesis on Old Japanese:

(25) A language has an Arche-Case/case system if the unmarked argument of an intransitive
predicate may be generated either as an external or internal argument.

(26) Hypothesis OJ-3.
Old Japanese is an Arche-Case/case language.
21

To sum up, with Hypothesis OJ-3, we can finally account for (22), and in consequence have
proved Hypothesis OJ-2' (20), and hence Hypothesis OJ-2 (18).

3.2.2.3. Conclusion: Subjects in Old Japanese
I maintain that the Forced Agreement Hypothesis (1) [3] and Hypothesis OJ-1 (16) make the
right predictions for Old Japanese. The subjects in shshi clauses are licensed by Agreement and
bare. Subjects in rentai clauses are marked by ga/no. The Case marking of subjects by Forced
Agreement in Old Japanese was eventually replaced by what we now have in Modern Japanese;
there emerged a double system where in finite clauses the subject DP may be bare, licensed at
Spec(I) by Agreement, or morphologically case marked by ga, inside VP, as I have maintained
above for Modern Japanese.
7


3.2.3. wo Marked Objects
If Agreement is forced in Old Japanese, the objects of transitive verbs must also be exclusively
licensed by Agreement in the form of bare DPs. But it appears, contrary to the Forced Agreement
Hypothesis, as though a DP was licensed as an object either in the bare form or in the form of a
wo phrase in Old Japanese as in Modern Japanese.

3.2.3.1. The functions commonly attributed to wo
The object case marking, however, is not the only function attributed to wo by traditional
Japanese scholarship. Commonly three functions of wo are recognized in Japanese grammar: wo
is assumed to belong to three different subcategories of particles. It is a kaku-joshi 'case particle',
a kant joshi 'interjective particle' and a setsuzoku joshi 'conjunctive particle'. The conjunctive
use of wo is a later development in a later stage of Old Japanese or in an earlier Classical
Japanese period but it has eventually faded by the time of Modern Japanese; we need not
concern ourselves with wo as a conjunctive particle. As a case particle, wo in Old Japanese has

7
I am not in a position at present to make any statement as to the status of bare DP subjects in clauses in the izen
form. Perhaps, they are structurally ambiguous between finite and non-finite, but I have to leave this issue aside for
now.
22
been commonly taken as an object case marker. As for the interjective use of wo, its semantic
effect is hard to characterize; it apparently adds some emotional or emphatic effect of a greater or
lesser degree of clarity. The interjective use of wo was prominent in Old Japanese but it has also
faded away from the language.

3.2.3.2. What the Forced Agreement Hypothesis predicts and is confirmed by philological
scholarship
If Agreement Case-marks bare DPs as objects in Old Japanese as it does in Modern Japanese,
the Forced Agreement Hypothesis must exclude wo from the category of kaku-joshi in Old
Japanese. We must assign only the function of kant joshi to wo, contrary to the traditional
description: a wo marked direct object is licensed as a direct object in the status of a bare DP and
wo is added only for the interjective function.
This situation is much like a topic ha/wa phrase functioning as a direct object, rather a
familiar situation we have both in Old and Modern Japanese. But we do not call ha/wa an object
case marker just because there are ha phrases that function as direct objects for a good reason. In
fact, not only can ha/wa phrases function as subjects or objects, but ha may also attach to DPs
marked with an inherent case marker like dative ni, or to adjunct PPs like DP to and even to verb
stems in the reny form. This distributional property is not a characteristic of only ha/wa; it is
shared by kakari particles in Old Japanese and focus particles like mo and sae in Modern
Japanese.
Then, we can say that the Forced Agreement Hypothesis predicts that the distribution of
wo must be like ha/wa or a kakari/focus particle. This is exactly what has since long been
recognized by Japanese philological scholarship. Not only do we find wo attached to locative ni
and he or comitative to: DP-ni-wo, DP-to-wo, DP-he-wo , but also we do find wo phrases
functioning as subjects or attached to verb stems (Konoshima 1973: 67f, 439).
To sum up, the Forced Agreement Hypothesis predicts a distribution pattern of the
particle wo in Old Japanese much different from the one we have of o in Modern Japanese. This
distribution has long been recognized at least among Japanese philologists. What principle might
be behind this distribution is a different matter, a question not raised by traditional philologists or
23
grammarians.

3.2.3.3. Stronger evidence the Forced Agreement Hypothesis: the Kinsui-Yanagida
generalization
What we have seen shows us that wo shares a property of kakari particles, but it cannot yet by
itself rule out the possibility that there are also instances of wo functioning simply as a case
marker, contradicting the Forced Agreement Hypothesis.
Let me clarify the point I am now addressing myself to. We have confirmed that the
distribution of wo is much like that of ha or a kakari particle. And yet I am still asking whether
wo can also be a case particle. Then, on the same token, should we not also ask if ha or for that
matter, any kakari particle, is functionally ambiguous between a kakari particle and a case
particle? Formally that is correct. But we don't double-classify ha and kakari particles and put
them also in the category of case particles. For a good reason. We know sufficiently well what
semantic function ha and kakari particles have. When we see a ha phrase or a kakari phrase used
as an object, we assume we can separate distinctly the function added to the object by that
particle. In contrast, the function of wo as an "interjective" particle is too uncertain to give us
confidence to rule out the presence of wo without it.
What is predicted by a hypothesis is after all only a necessary condition for it to be true.
The distribution of wo phrases satisfies a necessary condition for the Forced Agreement
Hypothesis for object Case marking; wo cannot simply be a case particle. But it is not a sufficient
condition to rule out the possibility that wo can also function purely as a case particle. Instead
assume that we find evidence that is not predicted by the hypothesis and yet can be accounted for
by it. Such evidence would much strengthen the case for the hypothesis.
Recently, Satoru Kinsui and Yuko Yanagida independently made an observation to this
effect for the Forced Agreement hypothesis. Let us first note that Nomura (1993, cited in Kinsui
2002: Observation 5) observed that in Manysh, kakari phrases, i.e., phrases marked as foci by
kakari particles, must precede ga and no marked subjects. This fact suggests that kakari phrases
are moved out of VP. I will return to this important finding later. Kinsui (2002: Observation 6)
and Yanagida (2004) observed that wo marked objects may precede karari phrases, indicating
24
that wo phrases can also be placed out of VP. Furthermore Kinsui (Observation 1a, Observation
2 ) observes, as does Yanagida (2003, 2004), that wo phrases precede bare and ga/no marked
subjects and not follow them, as shown in the following example.

(27) M2082
Amanokawa kahato yaso ari iduku ni ka kimi ga mi-hune wo a ga mati wora-mu
Milky-way ferry-port many be where at ka you ga sfx-boat wo I ga wait-aux
'There are many ferry docks along the Milky Way. Where shall I wait for your boat?'

Kinsui and Yanagida's generalization shows that independently of their relative order with kakari
phrases, wo phrases must move out of VP. Note that the hypothesis that wo is not a case marker
does not require wo to move out of VP; nor, for that matter, the Forced Agreement Hypothesis,
either. Whatever forces wo out of VP, this fact can be taken as good evidence that wo does not
function purely as a case marker. The Nomura's and Kinsui and Yamagiwa's findings provide us
with strong support for the claim that no instance of wo counts as a case marker pure and simple,
and consequently support the Forced Agreement Hypothesis for object Case marking.

3.2.3.4. Counterexmaples to the wo phrase movement
3.2.3.4.1. Counterexamples to the Kinsui-Yanagida generalization: *DP-ga/no > DP-wo
There are three counterexamples to the Kinsui-Yanagida generalization I am aware of, that is,
examples where wo-marked objects follow ga/no marked subjects.

(28) M872
Yama no na to ihi-tuge to ka mo Sayo-hime ga kono yama no uhe ni hire wo furi-kemu
Mountain no name as say-tell quote ka mo S-princess ga this mntn no upon at scarf wo wave-aux
'Was it for transmitting as the name of the mountain that Princess Sayo waved her scarf on this
mountain?'

Yanagida (2003) lists this example as a possible counterexample, but suggests a different reading
25
of the text: hire wo should read hire-wo, a compound noun 'scarf-tip'. This reinterpretation
makes this example conforming to the Kinsui-Yanagida generalization. However, I take this
reinterpretation problematic. For one thing, it happens that this poem is among a group of poems
with the same theme: Sayohime waving hire (M868, 871, 872, 873, 874, 875). Only in M872 is
hire followed by a character representing the syllable wo. Secondly, this character, en/on in the
on reading and tooi in the modern kun reading, is not a character usually used to represent the
particle wo in Manysh; it is an idiosyncratic property of a limited part of Manysh that this
particular character is used to represent the particle wo. From these considerations, I cannot
accept Yanagida's reinterprtation of the poem and must take it as a coutnerexample. However, the
part of Manyosh in question is thought to originate in the collection of poems kept by
YAMANOUE Okura, one of the prominent poets of Manysh, who, however, is believed to be a
non-native speaker of Japanese. This fact may be relevant for this counterexample.

(29) M2831
Misago wiru su ni wiru hune no yuhu-siwo wo matu-ramu yori ha ware koso masare
Misago be beach at be boat no evening-tide wo wait aux than ha I emph best
'I wait (for you) more than the boat on the beach where there are misago birds would be waiting
for evening tides'

It is likely that in this example, the no subject phrase misago wiru su ni wiru hune no is moved to
the left of the wo phrase yuhu-siwo wo by the heavy noun phrase shift to the left.

(30) M3689
Ihatano ni yadori suru kimi ihebito no idurato ware wo toha-ba ikani ihamu
Ihatano at stay do you family no where I wo ask-if how say-aux
'You, sleeping (buried) at Ihatano! if your family should ask me where (you are), how would I
respond?'

Formally it is not impossible to take ihebito no as the subject inside the embedded question; then,
26
the violation of the Kinsui-Yanagida generalization is removed. The poem, then, should be
glossed as 'You, sleeping (buried) at Ihatano! if you should ask me where your family is, how
would I respond?' But this is a less likely interpretation of the poem.
The verb tohu 'ask' is subcategorized for a dative ni argument in Modern Japanese. A
number of such verbs took wo phrases as objects in Old Japanese. It is conceivable that wo in
such phrases was exceptional and functioned as an inherent case marker. If that is the case
M3689 is not a counterexample. But I have no empirical support for this speculation.

3.2.3.4.2. A counterexample to a generalization of the Kinsui/Yanagida observation: *bare DP >
PP-wo
Kinsui and Yamagiwa were concerned with wo-marked objects. But, as I pointed out earlier, wo
as an interjective particle can be attached to other types of arguments and adjuncts. We have to
be also concerned with such uses of wo. In this widened perspective, we need to examine the
relative word order of bare objects and non-object wo phrases, a task not faced by Kinsui or
Yanagida. From the Kinsui-Yamagiwa generalization, it would be natural to draw a broader
generalization than suggested by them. If wo is taken as responsible for the movement of an
object out of VP when wo is attached to it, we would expect that wo moves other arguments and
adjuncts out of VP as well when it attaches to them. I have not done extensive research on this
issue, but there is one conterexample I am aware of:

(31) M3584
Wakare-na-ba ura-ganasi-kemu a ga koromo sita ni wo ki-mase tada ni ahu made ni
Part-aux-if suf-sad-aux I ga clothes under ni wo wear aux directly ni meet till at
'if we part, we would feel sad. Please wear my clothes under until we see face-to face'

Here, wo is attached to a ni phrase sita ni wo, which follows a bare object a ga koromo.

3.3. wh-Questions and kakari musubi constructions
3.3.1. Two types of wh-questions in Old Japanese
27
In Modern Japanese the indeterminates like dare and nani are directly followed by a case marker
or a postposition, as in dare ga, nani ni etc. and the question particle ka is put at clause final
position. In informal speech, this ka may be elided:

(32) Masao wa nani o tabe(masi)ta ka?
(33) Masao wa nani o tabeta (ka)?
(34) Masao wa nani o tabe ta no (desu ka)?
Masao wa what o ate
'What did Masao eat?'

If ka is attached to an indeterminate, we get an indefinite/existential reading, not a wh-question:

(35) Masao wa nani-ka (o) tabe masita (ka?)
'Masao ate something (did Masao eat anything?)'

In Old Japanese, unlike Modern Japanese, if ka is attached to an indeterminate, we have a wh-
question. But indeterminates can also occur without ka attached to them and form wh questions
in Old Japanese. We must thus recognize two different wh-constructions in Old Japanese, the ka-
wh question and the bare-wh question. These two types of questions have the following different
properties:

(36) The ka-wh question:
[k1] The particle ka is attached to an indeterminate phrase.
[k2] The predicate is in the rentai form.
[k3] The subject is a ga/no-marked DP phrase, unless it is at the same time a topic.
[k4] The indeterminate phrase with ka is moved out of VP.

(37) The bare-wh question:
[b1] It contains an indeterminate phrase without ka attached to it.
[b2] The predicate is either in the shushi or izen form.
28
[b3] The subject is a bare DP.
[b4] The indeterminate phrase remains in situ.

The following two pairs show the contrast between these two constructions:

(38)
M3749 (ka-wh question)
Hito-kuni ni kimi wo imasete itu-made ka a ga kohi-wo-ramu toki no siranaku ni
Foreign land to you wo let-go when-till ka I ga miss-be-aux time no know-neg
'letting you go to a foreign land, till when will I miss you, not knowing the time (when)?'
M3742 (bare-wh question)
Aha-mu hi wo sono hi to sira-zu toko-yami ni idure no hi made are kohi wo-ramu
See-aux day wo the day as know-neg eternal-dark at which no day till I miss be-aux
'without knowing the day we will see, until when will I miss you in pitch darkness?'
(39)
M795 (ka-wh question)
Ihe ni yukite ikani ka a ga se-mu makuraduku tuma-ya sabusiku omohoyu-besi mo
Home at go how ka I ga do-aux pillowed spouse-house sad feel aux emotive
'What would I do after returning home? Our bedroom with pillows would look sad.'
M4046 (bare-wh question)
Kamusaburu Taruhime-no-saki kogi-meguri mire domo aka-zu ikani ware se-mu
Awesome Taruhime-Point row-around see though bore-neg how I do-aux
'After having rowed around it while seeing it, the awesome Taruhime Point never gets boring.
What would I do?'


3.3.2. Wh-question type 1: the ka-wh question
The ka-wh question, to begin with, is a subtype of ka-focus questions with an indeterminate
phrase as a focus; the ka-focus question, in turn, is a subtype of the rentai kakari musubi
29
construction. The properties [k1] - [k4] listed above are properties the ka-focus question and the
rentai kakari musubi construction in general:

(40) The rentai kakari musubi construction
[k/m1] A kakari particle is attached
(i) to a focused phrase, if there is any;
(ii) to the sentence final predicate, if there is no focused phrase.
[k/m2] The predicate is in the rentai form.
[k/m3] The subject is a ga/nomarked DP phrase, unless it is at the same time a topic.
[k/m4] The focus phrase, if there is any, is moved out of VP. [Nomura's generalization]

The rentai kakari particles are emphatic so/zo and nam, and interrogative ya and ka. Only ka can
attach to an indeterminate and form a wh-question. Ka may also attach to a non-indeterminate
phrase and form a focused yes-no question:

(41) M 1742 (cited in Jidaibetsu Kokugo Jiten)
tada hitori i-watarasu ko ha waka-kusa no tuma ka aru ramu kasi-no-mi no hitori ka neramu
only one prf-cross girl ha young-glass no spouse ka be-aux acorn no one ka sleep
'does the girl who is crossing [the bridge] alone have a new husband? Does she sleep alone like
an acorn?'

[k/m1](i) and [k/m2] together are in effect the traditional definition of the rentai kakari musubi.
According to Mizutani (1974:31), however, examples where ka is attached to a non-
indeterminate phrase and where the predicates are unambigously in the rentai form are relatively
few: he cites M220, 290, 712, 2525, 2917. [k3], in particular, and [k/m3], in general, mean that
we find not only bare DPs but also ga/no marked DPs that function as subjects in ka-wh
questions and rentai kakari musubi sentences. We have already established that bare DP subjects
in rentai clauses are topics. [k/m3] thus follows from [k/m2].

30
3.3.3. Kakari particles as functional elements and the kakari phrase movement
[k/m1](ii) corresponds to the observation that "kakari particles in Old Japanese have the function
of final particles at sentence final position in addition to their kakari function at sentence internal
position". (Konoshima 1966:357) Konoshima attributes this generalization to YAMADA Yoshio.
Note that the "interjective" wo also has the sentence final function. The following example
illustrates this point concerning wo as well as [k/m1](ii) with the interrogative particle ka:

(42) M 2899
Nakanakani moda mo ara-masi wo adukinaku ahi-misomete mo are ha kohuru ka
rather silent mo be-aux wo in vain sfx-fall-in-love mo be ha be-in-love ka
'I should rather have kept silent. Having fallen in love in vain, am I yet in love? '

Note that an indeterminate in a ka-wh question can be assumed as focused. [k1] is hence
a special case of [k/m1](i). It follows that a ka-wh question cannot have ka at sentence-final
position. From this fact, we can draw an important analytic conclusion. We can account for this
apparent complementary distribution of ka if we assume that ka is a functional element and
marks a clause as interrogative at clause-final position but attaches to an indeterminate phrase if
there is one and forms a clause internal wh-phrase. I will leave aside for now and return later to
the question as to where ka is generated and how it gets attached to an indeterminate phrase. For
the moment it suffices to keep in mind that this attachment is obligatory, in conformity with the
Forced Agreement Hypothesis.
We can extend this attachment of ka to an indeterminate phrase in a wh-question to the
attachment of ka to a focused constituent in a ka-yes-no question. We can further generalize this
attachment to kakari musubi clauses in general Let us formulate our observation in the form of a
descriptive generalization.

(43) Kakari Particle Attachment. A kakari particle is a functional element. It is forced to
attach to a focused phrase, if there is any.

31
(43) is a restatement of a traditionally well-known fact that kakari particles have both clause-
final and clause-internal functions as shjoshi (final particles) and kakari joshi, respectively. I
have brought out the sense of the Forced Agreement parameter in this formulation.

3.3.3.1. Pied-piping
The kakari-particle is attached to an argument or adjunct that contains an indeterminate pronoun,
not necessarily to an indeterminate pronoun itself, a familiar pied-piping phenomenon. Note that
ka is attached to ta ga ta ni in the following example.

(44) M4141
Haru makete mono-ganasikini sa-yo hukete ha-buki naku sigi ta ga ta ni ka sumu
spring wait sfx-sad sfx-night get-dar wing-flap sing sandpiper who ga paddy ka live
'Being melancholy as spring has come, in whose paddy, I wonder, does a sandpiper live flapping
and singing as the night wears?'

3.3.3.2. Counterexmples (Apparent) to Kakari Particle Attachment
3.3.3.2.1. Predicate nominal indeterminates
There are examples where ta/tare is (part of) a predicative nominal followed by a copula to
which ka is attached:

(45) M 776
koto desi ha ta ga koto naru ka wo-yama-da no nahasiro midu no naka yodo ni site
word put-out ha who ga word be ka sfx-hill-paddy no seedling-bed water no in pool
'whose is the first word? And now like a stagnant pool of water in seedling beds of a hillside
paddy.

The poem consists of two sentences. The first sentence koto desi ha ta ga koto naru ka, where
koto desi ha 'what is put out' is a topic, ta ga koto 'whose words' is a predicative nominal and
naru is a copula in the rentai form. The wh question means something like 'whose words is what
is put out in words?' It appears that ka is at clause final position and fails to move. But we cam
32
assume that a predicative nominal forms a constituent with a copula and ka moves and pied-
pipes this constituent. The point is more apparent with the following example where tare 'who'
by itself is a predicate nominal followed by a copula:

(46) M2916
Tamakatuma aha-mu to ihu ha tare naru ka aheru toki sahe omo kakusi suru
Epithet meet-aux quot say ha who be ka meet time emph face hide do
'Who said 'let's meet'? You hide your face even when we meet.'

In the following example, however, the copula nari is reanalyzed and reverts to its etymological
origin -ni aru. ka pied-pipes a predicative nominal ni-phrase ta ga sono no ume ni 'plum flowers
of whose garden':

(47) M2327
Ta ga sono no ume ni ka ari-kemu kokodakumo sakini-keru ka mo mi ga hosi made ni
Who ga garden no plum ni ka be-aux this-much bloom-aux ka no see ga want till at
'(From) plum trees of whose garden were [these blanches], I wonder? They are blooming this
much! To the extent that I wish I can see the trees myself.'

3.3.3.2.2. A counterexample 2: M259
I am aware of one counterexample to (43):

(48) M259
Itu no ma mo kami-sabi keru ka Kagu-yama no hokosugi ga ure ni koke musu made ni
When no during mo godly aux ka Kagu-mountain no cedar ga top ni moss grow made ni
'when has it turned so godly aged that moss covers up to its top, the cedar tree of Kagu-
Mountain?'

The translation given above follows the interpretation of Shin Nihon Koten Bungaku Taikei. But
here the indeterminate phrase itu no ma 'during what time' is followed by mo. One conceivable
33
way out of this apparent counterexample is to take mo with the force of universal quantification.
Then, the poem should read 'has it been turning so godly aged during all times that moss crept up
to its top, the cedar of Kagu Mountain?' For related matters, see section 3.3.4.1.

3.3.3.3. Counterexamples to Nomura's generalization [k/m4]
So far I have concentrated on the phenomenon of how kakari particles attach to focused phrases.
But the real grammatical significance of kakari musubi was hidden until an important finding
was recently made in philology. According to Nomura (1993a), as already mentioned above,
kakari phrases must precede ga or no marked subjects in Manysh. I take this fact as strong
evidence for [k/m4], that is, that kakari phrases, in particular, ka-wh phrases, must be moved out
of VP. I would indeed take [k/m4] as an analytical equivalent of the distributional observation
formulated in Nomura's generalization.
8
I will later propose to account for this movement and the
attachment of kakari particles to focused constituents in terms of Agreement at Spec(I) in section
3.4.2. For now, I will consider the problem of apparent counterexamples to Nomura's
generalization as it is understood in the form of [k/m4].
Any constituent found to the left of a kakari phrase is a prima facie counterexample to
[k/m4]. However, there are factors other than kakari movement that are responsible for placing
phrases to the left of VP. We need to sort them out first. To begin with, topics obviously may
precede kakari phrases. In particular, we must remember that there are examples where bare DPs
function as subjects and precede kakari phrases. I have already claimed that such bare subject
DPs are at the same time topics. Hence they don't count as counterexamples to [k/m4].
Secondly, I distinguish bare-wh questions from ka-wh questions. Nomura's generalization
does not apply to wh-phrases without ka. I am not claiming that wh-movement in the usual sense
existed in Old Japanese; [k/m4] does not entail such a claim. (56) below, where a subject

8
Many examples of ga/no subjects found in Manysh are adjacent to main verbs. Lest one wonder if ga/no phrases
move to the right to adjoin to verbs instead of kakari phrases moving to the left of ga/no phrases, let me note that
this fact must be largely due to the lengh constraint imposed by the format of tanka 'short poems'. It is easier to find
ga/no subjects not adjacent to verbs in choka 'long poems', though we can find relevant examples in tanka, too, for
example, in M1919: Kunisu ra ga wakana tumu-ramu Siba no No 'the Siba Field where the Kunisu people pick
young grasses'.
34
precedes such a wh-phrase, would be a counterexample to wh-movement in the usual sense, but
not a counterexample to [k/m4]; see Tonoike (2002) on this point.
Thirdly, we have seen above that wo phrases move out of VP. The object of a verb is
often a wo phrase; if such an object precedes a kakari phrase, as in the following example, cited
in Tonoike (2002:87), it would counts as a counterexample to wh movement in the usual sense,
but it does not count as a counterexample to [k/m4].

(49) M 2396
tamasaka ni wa ga mi-si hito wo ikanaramu yosi wo motite ka mata hito-me mi-mu
by-chance ni I ga see-aux person wo what fate wo by again one-sight see-aux
'a person I happened to see, by what chance will I see again?'

Furthermore, as I spell out later, I assume that kakari movement lands kakari phrases at
Spec(I), not at Spec(C). I assume also that adjuncts can generally be generated outside of VP;
they can be adjoined to IP. It follows that examples in which an adjunct precedes a kakari phrase
may not be taken counterexamples to [k/m4], either.
What remain as possible counterexamples are those in which we find a bare DP
functioning as an object or a postpositonal phrase which is an internal complement of the verb.
According to Tonoike's (2002:88) count, there are 67 examples in Manysh where objects and
other complements "precede wh phrases". On the one hand, this count does not exclude those
with wo phrase objects like M2396 above (Tonoike's (6)b), nor does it exclude examples with
bare wh phrases like (58) below. On the other hand, the count concerns only wh phrases and not
kakari focus phrases in general. Given Tonoike's count, I am not in a position to estimate the
size of the type of possible counterexamples we have to be concerned with. But let me add
couple of remarks here.
3.3.3.3.1. Counterexamples (Apparent) 1. Heavy noun phrases preceding a kakari adjunct
In section 3.2.3.4.2, we noted an example where an adjunct wo phrase follows a bare object, a
counterexample to the claim that wo phrases move out of VP. There are similar counterexamples
35
with kakari particles, too. As Tonoike (2002) pointed out, there is an instance where a bare object
phrase precedes a ka marked adjunct:

(50) M83
Wata no soko okitu sira-nami Tatutayama itu-ka koe-na-mu imo ga atari mi-mu
ocean no bottom offing white-wave Tatutayama when cross-aux-aux wife neighborhood see aux
'When would I go over Tatuta mountain? And look at where my wife lives?'

Here the ka-phrase itu-ka is preceded by the direct object Tatutayama. The phrase wata no soko
okitu sira-nami 'white waves rising from the ocean bottom' modifies Mount Tatuta: rising high
like white waves do from the bottom of the ocean. Tonoike's is not an isolated such
counterexample. There seem to be quite a few examples of this type, but most of those that have
come to my attention so far all have heavy noun phrases as direct objects. Let me add a couple of
more examples:

(51) M3966.
uguhisu no naki tirasu-ramu haru no hana itusika kimi to taori kazasa-mu
bush-warbler no sing disperse-aux spring no flower when you with hand-break-aux
'When could I with you break and put on the head (twigs with) flowers that bush warblers would
now dispersing while singing?'

(52) M279
Wagimoko ni Winano ha mise-tu Nasugi-yama Tuno-no-Matuhara itu ka simesa-mu
My-wife to Winano ha show-aux Nasugi-mountain Tuno-Pine- when ka let-see-aux
'I let my wife see Wagino. When could I show her the Nasugi Mountain and Tuno Pines?'

In each of these examples, an object precedes a place or time adverbial ka-wh phrase. If an
adjunct is adjoined to IP, however, we expect that there must be some reason the object is put
before the ka phrase. In the above examples, the objects are heavy noun phrases.
36
Heavy noun phrase shift must be responsible for the apparent violation of [k/m4].

3.3.3.3.2. A Counterexample (Apparent) 2: A heavy noun phrase preceding a kakari subject
phrase; an example of philological interest

(53) M 840
Haru-yanagi kadura ni orisi ume no hana tare ka ukabesi sakaduki no he ni
Spring willow hair-pin ni break plum no flower who ka float sake-cup no upon ni
'who set afloat the plum flowers in sake cup that we picked for putting in willow hairpins?'

This example also involves a heavy noun phrase like those in the preceding subsection; the
kakari phrase preceded by it, however, is the subject. Besides, M 840 may be of some
philological interest, because the grammatical issue we are concerned with may be relevant to
choosing a proper reading from among variants in manuscripts. M840 as given in Nihon Koten
Bungaku Taikei as well as in Shin Nihon Koten Bungaku Taikei is an apparent counterexample to
[k/m4], since a ka wh subject follows a bare direct object. However, it is significant that Nihon
Koten Bungaku Taikei gives crucial variant readings in manuscripts. The third line of this poem
ume no hana given above should read ume no haru, according to three manuscripts, and ume
noru ha, according to one manuscript. What draws our attention is the second variant ume noru
ha. The poem then should read as follows:

M 840
Haru-yanagi kadura ni orisi ume noru ha tare ka ukabesi sakaduki no he ni

This reading makes the first three lines (17 syllables) read not as a noun phrase but as a rentai
clause followed by the topic ha. The verb noru means 'be/ride in/on a boat/car/horse etc'. The
clause can be interpreted as describing an event 'a plum [flower] picked for hair decoration is [as
if] on a boat'. In the poem this event description is made a topic: 'A plum flower being on board,
who has made this [scene] float on sake in a cup?' Then, M840 is not a counterexample to [k/m4].
37
It may even be tempting to interpret the first three lines as a topicalised head internal relative
clause: 'a flower [that is as if] on a boat, who has set [it] afloat in a sake cup?' For topicalised
rentai clauses, see below section 3.5.
The first variant reading ume no haru makes both interpretation and analysis much harder;
I will not be concerned with it, but it might be the result of a desperate attempt by a later copier
of the manuscript for correcting a perceived or real violation of the kakari phrase movement.

3.3.3.3.3. Summary; one counterexample to [k/m4]
To sum up, crucial possible counterexamples to [k/m4] are such ones in which a "light" bare
object or a "light" complement precedes a kakari phrase, in particular ka-wh phrase. My
admittedly not complete but fairly extensive search has so far come up with only one such
example:

(54) M187
Ture mo naki Sada no woka-he ni kaheri wi-ba sima no mi-hasi ni tare ka sumaha-mu
company mo neg no hill-side at return be-if island no sfx-bridge at who ka live-aux
'If we have returned to the hill-side of Sada where there are no acquaintance, who will reside on
the bridge-house of the island?'

Here, sima no mi-hasi ni must count as the complement selected by sumahu 'reside', but it
precedes tare ka 'who'. Let us recall that we have also one such relevant counterexample to the
extended Kinsui-Yanagida generalisation that wo phrases move out of VP; see (31) where a
"light" bare object precedes a wo phrase.
An obvious way to account for such apparent counterexamples as these if they turn up in
a significant number is to have recourse to scrambling. For this reason, they deserve our attention
of their own merits. For, the process of scrambling could have different theoretical significance
in languages where Agreement, in particular, Case marking, is forced or not; it should hence be
of particular interest to us to find out if and how scrambling worked in Old Japanese. In any
event we need more research on this matter.
38

3.3.4. Wh Questions Type 2: Bare-wh Questions
In contrast to the type of wh-questions we have discussed so far, there are cases where
indeterminate phrases are not accompanied by ka attached to it. This type of wh-question does
not contain ka at all, either attached to an indeterminate phrase or at clause-final position. For
tare/ta 'who' my tentative count based on Manysh Ssakuin come up with about 30 examples
of this type in Manysh, slightly less than examples of type 1 wh questions.
Let me repeat four properties I listed in (37).

[b1] It contains an indeterminate phrase without ka attached to it.
[b2] The predicate is either in the shshi or izen form.
[b3] The subject is a bare DP.
[b4] The indeterminate phrase remains in situ.

[b1] is the definition of this type of wh-question. In many examples the predicates are in a form
that is morphologically ambiguous between shshi and rentai, as in M3742 and M4046 in (38)
and (39). But none has a predicate unambiguously in the rentai form, except for those which
contain kakari particles other than ka, for example, M2654 and M3791. We can find a precious
few that contain predicates definitely in the shshi form:

(55) M869
Tarasihime kami no mikoto no na turasu to mi-tatasi-seri-si isi wo tare mi-ki
Tarasihime god no honorific no fish fish quote sfx-stand aux-aux rock who see-aux
'Who saw the rock where Goddess Tarasihime stood in order to fish?'

There are also examples with predicates in the izen form:

(56) M1389
Iso-no-ura ni ki-yoru sira-nami kaheri-tutu sugi-kate-naku ha tare ni tayutahe
39
Iso-no-cove at come-in white-wave return-sfx pass-can -neg ha who to waver
'White waves not being able to return back, to whom is it wavering?'

Some examples end in a wh-phrase functioning as a predicative nominal:

(57) M4397
miwatase-ba mukatu wo no he no hana nihohi terite tateru ha hasi-ki ta ga tuma
look-around-as yonder peak no upon no flower scent shine stand ha dear who wife
'as I look around [I wonder], whose wife is [she], flowers on hills yonder blossoming and
reflecting sunshine?'

From [b2] we can conclude that kakari musubi is not involved in this type of wh-questions not
only because there is no kakari particle attached to an indeterminate phrase, but also because the
predicate is not in the rentai form.
[b3] was noted by Saeki (1963: 6). As far as the case where the predicate is in the shshi
form, [b3] follows from the hypothesis that the subject Case marking is forced. As I mentioned
earlier, I leave izen clauses for future studies.
I maintain [b4] on three grounds. First of all, there is no overt evidence that movement of
any kind is involved with bare-wh constructions. Secondly, bare-wh phrases may be in
subordinate clauses, even in conditional clauses as in the second example below, a clear violation
of an island condition if a movement is involved:

(58) M 4070
Hito-moto no nadesiko uwesi sono kokoro tare ni mise-mu to omohisome-kemu
One stem no pink plant that heart who to show-aux quote think-aux
'who did I think of letting know my thought of planting a pink stem?'
(59) M 2263
nagatuki no sigure no ame no yama-giri no ibuseki a ga mune ta wo mire ba yamu
September no shower no rain mountain-fog no melancholy I ga heart who wo see if stop
'will my melancholy heart like mountain fogs in September rain heal if I see whom?'
40

Thirdly, wh questions in Modern Japanese do not involve movement; wh-phrases are licensed
directly by binding. I assume that the wh-question in modern Japanese is a descendant of the
bare-wh question of Old Japanese. The kakari musubi construction has transformed into the
modern focus particle construction, and in particular has been incorporated in the syntax of
indeterminates, but the ka-wh question was removed from this development; only the bare-wh
construction has survived in Modern Japanese. See Kitagawa and Deguchi (2002) and Kuroda
(forthcoming).
.
3.3.4.1. Negative concord and universal quantifier indeterminates
We can find a few examples of indeterminates accompanied by mo with the force of negative
concord or universal quantification in Manysh. Such indeterrminates are of course functionally
different from bare ones in bare-wh questions.

(60) M2628
Inisihe no situhata obi wo musubi tare tare to iu hito mo kimi ni ha masa-zi
Old-times no situhata obi wo tie hang who quote say person mo you to ha superior-neg
'Nobody is superior to you, with a situhata obi tied and hanging around'
(61) M2782
Sa-ne-gani ha tare to mo ne-me do okitumo no nabikisi kimi ga koto matu ware wo
Sfx-sleep-? ha who with mo sleep though offing seaweed no bend you word wait I wo
'whomever I might sleep with, I wait for your words, you to whom I bend like seaweeds in the
offing'
(62) M2783
Wagimoko ga nani to mo ware wo omowa-ne-ba huhumeru hana no ho ni saki-nu-besi
My-wife ga what quote mo I wo think-neg ba bud flower no spike ni bloom-aux-aux
41
'As my wife thinks of me nohow, a budding flower would open on the spike'

In M2628 and M2783, tare and nani are accompanied by mo and followed by negation. The
indeterminate tare in M2782 functions in the same way as free choice indeterminates in
temo/demo clauses in Modern Japanese. See no (1993:29) for similar examples, though no
takes the indeterminates in his examples as gimon-shi 'interrogative words'.

3.3.5. Conclusion
To conclude, there were two wh-interrogative constructions in Old Japanese, the ka-wh question
and the bare-wh question. The ka-wh question is a subtype of the rentai kakari construction and
as such involves forced movement. Bare wh-questions do not involve movement, and I assume
that indeterminates in them that function as wh-words are licensed in situ by binding.

3.4. Functional categories and movements
Let me now formulate an account of the descriptive claims made above. I assume a classical GB
framework based on the internal subject hypothesis. I introduce only two functional categories,
I(nfl) and C(omp), and leave open the possibility of finer distinctions. In addition, I assume that
as far as functional categories are concerned, the phrase structure is free of linear order. I assume
without going into any discussion that morphological and other factors determine linearization.
For example, both (63) and (64) are possible linearisations of IP:

(63) Spec(C)-[[Spec(I)-[I-[VP]]-C]
(64) Spec(C)-[[Spec(I)-[ [VP]-I]-C]

3.4.1. C(omp) and I(nfl)
Shshi clauses are finite and rentai clauses are non-finite (Hypothesis OJ-1 (16)). I introduce two
complementisers, C
s
for shshi clauses and C
r
for rentai clauses. C
s
selects Agr, a finite empty
Infl. Following the standard assumption, the subject DP moves from Spec(V) to Spec(I) and
42
Agrees with Agr to get subject Case. Thus, bare DPs are licensed as subjects in shshi clauses.
We have the following linearisation.

(65) [Spec(C) [[DP-[Agr-[[t] [X V]
V'
]
VP
]
I'
]
IP
C
s
]
C'
]

In contrast, C
r
selects a non-finite Infl. The subject DP of a rentai clause does not move to
Spec(I). It must get genitive case and is marked by ga/no in situ inside VP.

(66) [Spec(C) [[Spec(I) [Infl [DP-ga/no-[X V]
V'
]
VP
]
I'
]
IP
C
r
]
C'
].

3.4.2. Rentai kakari clauses: kakari phrase raising
I assume that kakari particles so/zo, nam, ya and ka are non-finite Infls and can be selected by C
r
.
A kakari particle Agrees with a focus constituent, if there is any. Note that since the subject does
not move to Spec(I) to get Case in rentai clauses, Spec(I) is left empty; a focus prhase, and in
particular, an indeterminate phrase, if there is any, can, and indeed must, move to Spec(I) and
gets the kakari particle attached to it. Take, for example, interrogative ka and assume that a non-
subject XP is focused. Then, we will get the following structure:

(67) [Spec(C) [[XP-[ka [DP-ga/no [..[t]..V]
V'
]
VP
]
I'
]
IP
C
r
]
C'
].

Here, Infl ka is actualized at the left edge of I', so that it might directly follow XP in Spec(I) and
Agree with it. If, on the other hand, the clause does not contain focussed constituent, Spec(I) is
left empty. The particle ka is enclitic. Looking for a host inside IP, it forces the Infl node to
actualize at the right edge of I':

(68) [Spec(C) [[Spec(I) [[DP-ga/no [X V]
V'
]
VP
[ka]
I
]
I'
]
IP
C
r
]
C'
].

The two structures are illustrated by the following examples.

(69) M40
[[[[[Aminoura ni huna-nori su-ramu wotome-ra ga tama-mo no suso ni [siho mitu-
ramu]
VP
]
VP
[ka]
I
]
I'
]
IP
]
CP
43
Aminoura at boat-board do-aux girl-pl ga sfx-robe no hem at tide rise-aux ka
'Would the tide arise to the hemlines of the clothes of the girls who were to ride in the boat at
Aminoura?'
(70) M3891
[Aratu no umi siho hi siho miti toki ha are-do [[idure no toki]-[[ka]
I
[wa ga [t] kohi-za-
ramu]
VP
]
I'
]
IP
]
CP
Aratu no sea tide ebb tide risse time ka be-sfx when no time ka I ga long-neg-aux
'The tides of Aratu ebb and rise at times, but when will I not long for you?'

3.4.2.1. The target of kakari movement
Recall that ka in general pied-pipes the constituent that contains an indeterminate. That is, what
moves to Spec(I) is not necessarily an indeterminate pronoun but an argument or adjunct that
contains an indeterminate pronoun. Schematically, we have the following structure where Wh-
pro is an indeterminate pronoun and [t] is a trace of an argument or adjunct that contains wh-pro
and has moved to Spec(I):

(71) S = [[.Wh-pro.. ]
XP
-[[ka]
I
[[t]]
VP
]
I'
]
IP


3.4.2.1.1. Pied piped subordinate clauses
The attachment mechanism of focus particles to various constituents in Modern Japanese must
be a descendent of the kakari particle attachment in Old Japanese, even though much change has
taken place as to which particles participate in the process and with what function. See Apendix.
The interrogative ka is not movable any more in Modern Japanese. It is remarkable that ka in Old
Japanese attaches to, and thus pied-pipes kinds of constituents that we would not expect from the
syntactic behavior of focus particles in Modern Japanese. Old Japanese ka pied-pipes
subordinate adverbial clauses. First, observe that ka pied-pipes a ba conditional clause in the
following example:

(72) M4392
44
Ame-tusi no idure no kami wo inoraba ka utukusi haha ni mata koto tohamu
heaven-earth no which no god wo pray-ba ka dear mother to again word ask
'Can I talk with my dear mother again if I pray to which god of the universe?'

We find a similar example in M1784. Furthermore, in Old Japanese a presuppositional-
conditional clause can be formed without ba, and even such a clause can be pied-piped, as shown
in the following example:

(73) M2380
hasikiyasi ta ga sahure-ka-mo, tamahoko no miti mi-wasurete kimi ga ki-masa-nu
ah, ah who ga interfere-ka-mo epithet no way see-forgot you ga come-aux-neg
'Ah, with who interfering have you not come forgetting the way?'

In this example the indeterminate ta 'who' is in a clause with the main verb sahure followed by a
clause-final ka(-mo). We have a couple of interesting philological and grammatical issues. To
begin with, the verb sahure is represented ideographically in the original text. Thus, it is up to
philological interpretation how to read this verb and to assign a proper morphological form. The
traditional reading reproduced above takes the verb as in the izen (realis conditional) form
sahure. It is conceivable to take the verb as in the rentai form. Then, the poem would consist of
two sentences: "Ah, who is interfering? You have lost the way and have not come." But, then,
we would have a violation of the kakari particle attachment. The reading assigned in the
philological tradition rightly stays clear of this violation. Secondly, even with the izen reading
assigned to the verb in question, it would still be conceivable that the poem is taken as consisting
of two sentences. For, independent sentences, though rare, can take verbs in the izen form. But,
then, again, we would have a violation of the kakari particle attachment. So, we are led to take
hasikiyasi ta ga sahure-ka-mo as a pied-piped adverbial clause. But, then, thirdly, this pied-piped
kakari phrase must be resolved by a predicate in the rentai form. However, the final predicate ki-
masa-nu is a form morphologically ambiguous between shshi and rentai and it appears that the
reading we are trying to justify, though not inconsistent with this verb form, is not determined
unambiguously from the text. The subject of this predicate kimi 'you' is also ideographically
45
represented in the original text without any indication of a particle attached to it. But we cannot
have kimi as a bare DP here prosodically; we need a three mora word to make the last line of this
waka poem seven mora long. The traditional reading supplies ga. This decision entails that the
predicate is taken as rentai, resolving the kakari suspense introduced by a pied-piping ka.
In sum, the traditional interpretation of M2380 is a convergence of a number of
philological considerations; if we follow it, as I do, M2380 provides us with an interesting
example of pied-piped subordinate clause.

3.4.2.2. The landing site of kakari movement
Thanks to the pied-piping of subordinate clauses of the type demonstrated above, an
indeterminate pronoun, a kakari trigger, may be located in an adjunct island and separated from
the rentai predicate that resolves its kakari suspense. The pied piping ka so to speak serves as a
ferry to the island. If kakari momevement were to apply to the indeterminate directly, the island
constraint would be violated. The question that naturally arises, however, is whether kakari
movement can be long distance from inside a complement of a bridge type, or it is inherently
clause-bound. This is a difficult question to settle due to the limited quality and quantity of the
data available. But if we follow Watanabe (2002:82), kakari movement is a long-distance
movement. His claim, according to him, is based on a lone example he has found in Manysh ,
but the example deserves close attention, since it raises interesting questions.

(74) M2573
Kokoro sahe matureru kimi ni nani wo ka mo iha-zu ihi-si to wa ga nusuma-ha-mu
Heart even offer you to what wo ka mo say-neg say-aux quote I ga cheat-aux-aux
'What would I (say) to you, and keep cheating you, that I said without having said it, to you to
whom I have offered my heart?'

The poem is not easy to understand. The verb nusumu means 'steal' in Modern Japanese but in
Old Japanese it can also be used with the meaning 'do something stealthily', a type of verb
unfamiliar in the Modern Japanese vocabulary and the argument structure of this use can only be
surmised. According to Jidaibetsu Kokugo Daijiten two usage types are exemplified:
46

(75) Ono ga inoti wo nusumi sise-mu
I ga life wo steal kill-aux
(76) Ono ga inoti wo sisemu to nusum[u]'
I ga life wo kill-aux to steal

Both should mean something like 'pro will sneakily kill me (my life)'. In (75) nusumu is used as
the first member of a serial verb, while in (76) nusumu takes a quotative to complement. Our
example (74) conforms to the second pattern. There are two points that call our attention, one
grammatical and the other philological. Grammatically it is significant that the predicate inside
the to complement, ihi-si 'said' is a rentai form. Thus, kakari triggered by the indeterminate nani
is resolved inside the to complement clause. The matrix predicate nusuma-ha-mu, on the other
hand, is morphologically ambiguous between shshi and rentai. In addition, its subject, the first
person pronoun is only ideographically represented. It could be read either as wa ga, as given in
(74), or as ware, 'I', a bare DP. From a strictly morphological and philological standpoint, an
alternative reading is conceivable for M2573:

(77) M2573
Kokoro sahe matureru kimi ni nani wo ka mo iha-zu ihi-si to ware nusuma-ha-mu
'I would sneakily keep asking you, whom I have offered my heart: what did I say without having
said it to you'

With this reading the kakari triggered by ka is resolved inside the subordinate clause; long
distance movement is not ivolved.
The commentators of both Nihon Koten Bungaku Taikei and Shin Nihon Koten Bungaku
Taikei choose the reading wa ga. This choice implies that the main predicate nusuma-ha-mu is
taken as in the rentai form, and hence can also be taken as the musubi corresponding to the
kakari phrase nani wo ka. With this interpretation, the kakari phrase is presumably raised long-
distance to the matrix clause, as Watanabe determined. With this interpretation, it is noteworthy
47
that the kakari phrase movement has left a chain of rentai forms <ihi-si, nusuma-ha-mu>,
indicating that it obeys subjacency. I do not know whether the philologists have a good reason to
exclude the other interpretation, which is consistent with the hypothesis that kakari movement is
clause-bound.

3.4.3. Topics
Bare DPs as well as wa phrases can be topics. I have earlier claimed that bare subjects of
transitive verbs in rentai kakari clauses must be topics. The bare subjects of intransitive predicate
may not necessarily be topics in rentai clauses, and indeed cannot in relative clauses; they can be
Case marked by the intransitive predicate. But bare DPs can of course be topics in intransitive
clauses, too. The following variant forms of the same waka are good evidence for these points:

(78) M232/M234
Mikasa-yama nobe yuku miti ha kokidakumo sizini aretaru ka hisa ni ara-naku ni
Mikasa-yama nobe yu yuku miti kokidakumo sizini aretaru ka hisa ni ara-naku ni
Mikasa-mountain field (from) go road (ha) very thickly overgrown ka long be-neg ni
'Is the road through the fields of Makasa Mountain very overgrown with weeds? Not long since
[he] has gone.'

The bare noun Mikasa-yama modifies nobe and form a DP Mikasa-yama nobe. This DP is bare
in M232; it is accompanied by yu in M234 as an argument of the verb yuku. Our main concern
here is the bare DP Mikasa-yama nobe yu yuku miti in M234. This bare DP in M234 must be a
topic, as the wa phrase Mikasa-yama nobe yuku miti ha in M232 is.
Incidentally, a bare object may also become a topic:

(79) M4238
Kimi ga yuki mosi hisani ara-ba ume-yanagi tare to tomo ni ka wa ga kadura-kamu
You ga go if long be-sfx plum-willow who with ka wa ga make-ornament
'if your journey lasts long, with whom will I make hairpins with plum and willow branches?'
48

I assume that ume yanagi is a topic.
I assume that topics are moved to Spec(C). There are two types of apparent
counterexamples to this assumption. First of all, there are examples where wa topics are
preceded by kakari phrases. According to Sasaki (1992), such examples are very few as far as ka
and ya are concerned. It would be fair to assume, following Sasaki that such kakari phrases are
dislocated to the left and adjoined to CP. In contrast, there are a substantial number of such
examples with so/zo. Many of such examples can be explained away by means of heavy noun
phrase shift, but not all of them. I leave the issue with so/zo for future studies.
Secondly, there are cases where bare subjects are preceded by kakari phrases. Watanabe
(2001:100) cites the following counts in Manysh from Nomura (1993):

(80) Word order: ka > bare subject 13 exampels
(81) Word order: bare subject > ka 30 examples

Watanabe is concerned with wh-movement, not with case marking; for him (81) is problematic
but not (80). In contrast, for me (80), not (81), is a problem. I have already demonstrated that the
bare subject in (81) is a topic. (80), however, appears to be problematic; it shows wrong word
order, since I have claimed that kakari phrases move to Spec(I). However, in all the 13 examples
Nomura counts for (80) the predicates are either intransitive verbs or adjectives, as in the
following example:

(82) M2206
Maso-kagami Minabuti-yama ha kehu mo ka mo siratuyu okite momiti tiru-ramu
Sfx-mirror Minabuti-mountain ha today mo ka mo white-dew lay maple fall-aux
'At Minabuti Mountain today, too, are there dews and are maple leaves falling?'

Thus, we can conclude that the bare subject in (80) is not a topic but Case marked by the verb.
Not only are these examples not counterexamples, but they are rather supporting evidence of our
49
hypotheses, since their existence is exactly what our hypotheses predict.

3.4.4. The landing sites of wo phrases
Following the Kinsui-Yanagida generalization, I have claimed earlier that wo phrases move out
of VP. As I have also already mentioned, wo can be attached to adjunct postpositional phrases,
and can also be a clause-final particle. In these two respects, wo may appear to share
characteristics of kakari particles. However, wo is not a kakari particle.
First of all, the predicate of a clause that contain a wo phrase can be in the shshi form:

(83) M 126
miyabi wo to ware ha kikeru wo yado kasa-zu ware wo kahese-ri oso no miyabi wo
courtly wo quote I ha hear-aux wo shelter offer-neg I wo turn/away-aux dull courtly man
'Though I have heard you were a courtly person, you turned me away, you a dull courtier!'
(84) M 869 (= (55))
Tarasihime kami no mikoto no na turasu to mi-tatasi-seri-si isi wo tare mi-ki
Tarasihime god no honorific no fish fish quot sfx-stand aux-aux rock who see-aux
'Who saw the rock where Goddess Tarasihime stood in order to fish?'

Luckily, we have here the predicate kahese-ri 'turned away' in M126 and mi-ki 'saw' in M869,
unambiguous shshi forms.
The distribution of the wo phrase corroborates this conclusion. Unlike kakari phrases, wo
phrases can occupy any position relative to bare subjects (but not ga/no marked subjects), kakari
phrases, and topics. I maintain, then, that wo phrases can be adjoined freely to VP, IP or CP:
9


(85) XP-ka > DP-wo Adjunction to VP
M 3153
mi-yuki huru Kosi no Oo-yama yuki-sugite idure no hi ni ka wa ga sato wo mimu
sfx-snow fall Kosi no big-mountain go-pass where no day ka I ga homeland wo see-aux

9
Yanagida (2004) also drew a similar conclusion.
50
'When would I see my homeland by passing Oyama Moutain of Koshi where it snows?'

Here idure no hi ni ka occupies Spec(I), hence wa ga sato wo 'my homeland' must be adjoioned
to VP.

(86) DP-wo > bare subject Adjunction to IP in a finite clause
See (84) above. Note that (84) is a bare-wh question. The indeterminate tare is bare, but as an
interrogative pronoun, it cannot be a topic; it is a subject pure and simple of a finite clause and
occupies Spec(I).

(87) M1486
Wa ga yado no hana-tatibana wo hototogisu ki naka-zu tuti ni tirasite-mu to ka
I ga lodge no flower-orange wo cuckoo come sing-neg ground disperse-aux quote ka
'the [flowers of] orange trees in blossom in my house cuckoos, without having come and sung,
let fall, do they? '

The quotative particle to takes a predicate in the shshi form. Hence, the bare noun hototogisu is
the subject of a finite clause in Spec(I).

(88) DP-wo > XP-ka Adjunction to IP in a non-finite clause
10


10
M2396 below is cited in Watanabe (2002:81). Watanabe takes wo as an object marker and is concerned about
this example as a possible counterexample to wh-movement. He suggests that this wo phrase is a topic and
wonders if the original case marker reappears when the topic marker is dropped in Old Japanese. We do not
take wo as an object marker, nor do we take the wo phrase as a topic.
Incidentally, I have taken resort to heavy noun phrase shift for explaining apparent counterexamples on
previous occasions; see sections 3.2.3.4.1, 3.3.3.3.1 and 3.43. We cannot doubt the existence of heavy DP shift
as a non-core device. Consequently, in order to substantiate a claim about word order X > Y, it is important to
find examples where X can qualify as a "light X". Once such examples are found, examples with "heavy X"
should count as legitimate examples, too. Here, I take M2722 as an example with a "light" DP that legitimize
other examples of the word order we are interested in.
51
M 106
hutari yuke-do yuki-sugi gataki aki-yama wo ikani ka kimi ga hitori koe-namu
Two go-though go-pass hard autumn wo how ka you ga one-person cross-aux
'mountains in autumn, which are hard to pass even when we go together, how will you cross over
by yourself?'
M 2396
tamasaka ni wa ga mi-si hito wo ikanaramu yosi wo motite ka mata hito-me mi-mu
by-chance ni I ga see-aux person wo what fate wo by again one-sight see-aux
'a person I happened to see, by what chance will I see again?'
M2722
Ootomo-no-mitu no tomari ni hune hatete Tatuta no yama wo itu ka koe ika-mu
Ootomo-no-mitu no port ni boat anchor Tatuta no Mountain wo when ka cross go-aux
'When will I cross Tatuta Mountain after anchoring the boat at Ootomo-no-mitu?'

(89) DP-wo > bare subject Adjunction to CP in a non-finite clause
M132
Iwami no ya Takatuno-yama no ko no ma yori wa ga huru sode wo imo mi-tu-ramu ka
Iwami no ya Takatuno-mountain no tree no between from I ga wave sleeve wo now see-aux ka
'Would my wife see the sleeves I wave from between trees on Takatuno Mountain in Iwami?'

In M132 a wo phrase precedes a bare subject of a transitive verb in a rentai clause. The bare
subject must be a topic, hence in Spec(C).

(90) DP-wo > DP-ha
M193
Hatakora ga yo hiru to iha-zu yuku miti wo ware ha koto-goto miyadi ni zo suru
Farmer ga night day-time quote say-neg go road wo I ha all road-to-palace ni zo make
'We all take to the palace the road where farmers go day and night '

52
M193 contains a kakari particle zo; the main preidcate suru is in the rentai form.

(91) DP-wo > DP-ha Adjunction to CP in a finite clause
M808
Tatu no ma wo are ha motome-mu ao-ni-yosi nara no miyako ni ko-mu hito no tani
Dragon no horse wo I ha find-aux blue-in-beautiful no capital in come-aux person no for
'I will find a dragon horse for a person who will come to the beautiful capital Nara'
M4501
Yati-kusa no hana ha uturohu tokiha naru matu no sa-eda wo ware ha musuba na
8000-kinds no flower ha change evergreen be pine no sfx-branch wo I wa tie na
'All kinds of flowers come and go. I will tie branches of evergreeen pines'

In M808, the predicate motomemu is morphologically ambiguous, but there is no reason why it is
not in the shshi form. M4501 ends in a suffix, which requires a verb in the mizen form. I
presume that both M808 and M4501 are examples of finite clauses.
We have confirmed that wo phrases can be adjoined to any type of clausal maximal
categories. In addition, wo phrases can be move into Spec(I) and Spec(C) by kakari movement
and topicalization, respectively. The following sentence is a ka-wh question; the wh-phrase is a
wo phrase:

(92) DP-wo-ka
M 439
kaheru-beki toki ha nari-keri miyako nite ta ga temoto wo ka wa ga makura-kamu
return-aux time ha become-aux capital at who ga arm wo ka I ga pillow
'The time to return has come. Whose arms will I pillow in the capital?'

In the next examples wo phrases are topics. The ba that follows wo in M2766 is commonly
considered as a rendaku (sequential voicing) form of ha:

53
(93) DP-wo-ha
M423
Kimi wo ba asu yu soto ni ka mo mi-mu
You wo ha tomorrow out at ka mo see-aux
'I will see you in the other world from tomorrow'
11

M2766
Misimae no irie no komo wo kari ni koso ware wo ba kimi ha omohi-tari-kere
Misima no inlet no komo-plant wo cut/casually koso I wo ba you ha
'Did you care for me only casually (as if cutting komo plants at Misima Inlet)?'

From these observations I draw the following conclusion:

(94) A wo phrase may be the target of wa topicalization and kakari focusing. If not focused or
topicalised, it moves and adjoins to VP, IP or CP.

The obligatory movement of wo phrases substantiates the claim that wo is not a case marker and
cannot license object DPs. Subjects and objects must be licensed by abstract Case in Old
Japanese. The movement of wo phrases confirms that Old Japanese is a Forced Agreement
language. However, the nature of this movement is not certain. I don't claim that all movements
are under the control of the Agreement parameter; other factors could also motivate and
contribute to movements. I have suggested earlier that wa topicalization is obligatory in Modern
Japanese, a Non-Forced Agreement language. It is not strictly controlled by the Agreement
parameter. In the case of wo movement in Old Japanese, it is an obligatory movement in a
Forced Agreement language, and yet the lack of restrictions of its landing sites suggests that a
different force is behind this movement. I leave this issue for future studies.

3.5. Relative clauses
To recall, Modern Japanese has both head internal and head external relativization,

11
It is of some interest to note that a variation recorded in the original text has kimi wo asu yu ha, instead of Kimi
wo ba asu yu.
54
evidence for the claim that Agreement is not forced in Modern Japanese, as I have pointed out
above. According to Kond (1981; 2000:343f), head internal relative clauses were recorded only
from around the beginning of the Heian Period (the 9th to 12th century). They apparently did not
exist in Old Japanese, as the Forced Agreement Hypothesis correctly predicts. A couple of
remarks may be in order to supplement this point.
First of all, there is one possible exception in Manysh:

(95) M4429
Umaya naru nawa tatu koma no okuru ga he imo ga ihi-si wo okite kanasi mo
Stable is rope cut horse no left-behind ka ha wife ga say-aux wo left sad mo
'I am sad as I left my wife behind who said she would not be left behind like a horse that would
cut rope in the stable'

This poem was composed by a soldier recruit in the Eastern dialect, and poses some difficulty in
interpreting it. For example, the commentators conjecture that the sequence ga he found in the
poem as a dialect variant of ka ha. But umaya naru nawa tatu koma no okuru ga he imo ga ihi-si
is presumably a head inernal relative clause.
12

Secondly, we can find examples in Manysh which one might, but should not, take as
examples of head internal relative clauses. Consider, for example, earlier examples (56) and (57)
which I repeat here:

M1389
Iso no ura ni kiyoru siranami kaheri tutu sugikatenaku ha tare ni tayutahe
M4397
Miwataseba mukatu wo no he no hana niohi terite tateru ha hasiki ta ga tuma

In M1389, siranami 'white waves' might be taken as the internal head of a head internal relative
clause followed by the topic marker ha. In M4397, similarly, hana 'flower' might be taken as the

12
Janick Wrona brought this example to my attention.
55
head of a head internal relative clause. It is not clear exactly how the commentators of Nihon
Koten Bungaku Taikei and Shin Nihon Koten Bungaku Taikei structurally analyze these examples.
But the clauses preceding the topic marker can be taken as nominalised clauses referring to
events, as I did above; then the topic marker in effect functions to connect two event descriptions,
rather than a topicalised subject in the form of a head internal relative clause and the rest of the
sentence.
We can in fact find an interesting pair in Manysh . Compare (89), which I repeat here,
with (96):

(89) M 132
iwami naru takatuno yama no ko no ma yori wa ga furu sode wo imo mituru ka mo
(96) M 134
iwami naru takatuno yama no ko no ma yu mo wa ga sode furu wo imo mikemu ka mo

M132 and M134 are almost identical, except that in the former the object of the verb see is a
sleeve while in the latter it is the event of [my] waving a sleeve. In fact, if we were to assume
that head internal relative clauses had existed in Old Japanese, we could say that M134 is
structurally ambiguous between two interpretations. For ease of reference, let us call this type of
examples head internal relative lookalikes.
Thirdly, it is important to distinguish clearly between head internal relative clauses and
no-relative clauses.
13
The latter type of relative clause must be analysed as headed by a no-
phrase, not as a head internal relative clause; I take it as a left-headed relative clause. Examples
that may be taken as no-relatives existed in Old Japanese. See Ishigaki (1955) and Kond (2000:
341).
Finally, however, Ishigaki (1955:25) in his seminal work makes intriguing comments on
this construction. He cites examples of the following structures from Senmyau ''imperial order' :

13
I did not distinguish these two types of relative clauses in Kuroda (1974) but recognized no-relatives (no-
introduced relatives) as a separate category in Kuroda (1975/76). Incidentally, keijsei meishiku, the term/concept
well-known and much discussed among Japanese grammarians, is sometimes understood as denoting the head
internal relative clause. But the term must be taken as denoting exclusively the no-relative in Ishigaki (1941), where
it was originally introduced. See Kond (2000:341ff) for related comments.
56

(97)
[NP no ...Predicate
1
] ha ...Predicate
2

[NP no ...Predicate
1
] wo ...Predicate
2
,

where Predicate
1
is in the rentai form and Predicate
2
is in the shshi form. Ishigaki interprets
these examples by supplying the elided head mono 'one (who)'. Ishigaki's examples have ha or
wo following Predicate
1
. No example is given with ga or other particles/postpositions. In fact,
Ishigaki took a trouble of explicitly stating about this construction that "there is no case where
the subject case particle ga follows a predicate [ygen]" in Old Japanese (op. cit.: 20). It is
significant that the examples of head internal relative lookalikes I cited above are also followed
by ha or wo. We might paraphrase Ishigaki's comment thus: Left-headed no-relative clauses, if
existed, should have given rise to examples that would contradict this claim by Ishigaki.
Speculating about the development of head internal relative clauses we might say: clauses
were first juxtaposed by means of the topic marker ha or the interjective marker wo, to express a
connection between two denoted events, as in the above examples from Manysh. This manner
of connecting two clauses by a topic or interjective marker was extended to no-headed relative
clause lookalikes, as seen in Ishigaki's Senmyau examples. Head internal relative clauses, and
perhaps even genuine no relative clauses, did not appear before the Heian period, as Ishigaki's
comment mentioned above suggests.
14
We might say that the stage had been set and well
prepared by the Nara period for the emergence of head internal relative clauses at the beginning
of the Heian period, once the Forced Agreement parameter was reset to the opposite value.
Finally, examples like those from Manysh given above might at first appear to provide
evidence against the claim that head internal relative clauses did not exist in Old Japanese, but I
have argued against this conclusion. Once this counterargument is accepted, however, these

14
There is, however, one example that may be taken as an exception to this statement. M3752 contains a phrase haru
no hi no uraganasiki ni: if one take ni as the temporal postposition 'at, on', one can interpret the phrase as a no
relative clause with the head harun no hi 'on a sad spring day'. But ni following a rentai predicate can also suggest a
vague connection of two events; thus one can take it as an adverbial clause 'a spring day being sad, ..' Examples like
this deserve special attention, as they might have served as an origin of the more prevalent use of no relatives. This
example has also been brought to my attention by Janick Wrona.
57
examples of lookalikes from Manysh could, in an ironical way, be turned into data favorable
for the position that head internal relative clauses did not exist in Old Japanese. For one might
think that short poems, the main source of the Old Japanese data, would be a disadvantageous
vehicle for head internal relative clauses; head internal relative clauses seem to have particularly
thrived in the stylistic environment provided by the Heian prose literature that favored rather
convoluted syntax; poems are too short to accommodate head internal relative clauses. But
lookalike examples like those shown above from Manysh would suggest that there is no reason
why Many poets would not have exploited head internal relative clauses, had they existed for
their disposal.

4. Conclusion
I have argued for the Agreement Hypothesis for Old Japanese:

[1] Old Japanese was a Forced Agreement language.

I have claimed that this hypothesis accounts for the claims/generalizations [2]-[6]:

[2] Subjects in shshi clauses must be Case-marked as bare DPs.
[3] Subjects (external arguments) in rentai clauses cannot be bare unless they are at the same
time topics; they must be genitive and marked by ga/no.
[4] Objects must Case-marked by verbs as bare DPs.
[5] Kakari musubi focus phrases, in particular ka-wh-phrases, move to Spec(I).
[6] Head internal relative clauses did not exist.

[2] is a restatement of a classically well-known fact. [3] derives from the following
generalizations ([7], [8]) that I have drawn and the claim [9] that I have made on the basis of [2]
and [8]:

[7] The subject of a relative clause with a transitive verb was marked by ga/no.
58
[8] The unmarked argument of an intransitive predicate in a rentai clause is either bare or
genitive and marked by ga/no.
[9] The Case system is an UNMARKED, arche-Case system, a system that transcends the
accusative-ergative distinction.

[4] is dependent on [10], which is a refined restatement of the Kinsui-Yanagida generalization.

[10] Wo phrases must move and adjoin to CP, IP or VP

I assume that if a wo phrase functions as a direct object, it is Case marked (in the status of a bare
DP) by a transitive verb in situ and moves out of VP due to wo. [5] is a restatement of Nomura's
generalization. [6] is a fact reported by Kond.
I presume that the Agreement parameter setting had changed to the opposite value by the
time the Heian prose literature flourished. Kakari movement was lost in the Heian period; see
Watanabe (2002b,2003). Head internal relative clauses were abundantly documented in the
Heian literature; see Kitayama (1951), Kuroda (1974). I am not in a position to make definite and
precise statements on the fate of Case marking. However, it would seem fair to assume that the
Kinsui-Yanagida generalization was lost, perhaps, early in the Heian period.
15
This indicates that
wo started functioning as a case marker as well. As wo became an alternative for object marking,
the unmarked case system must have settled to an accusative system. For the subject marking,
the choice between bare and ga/no marked DPs depended on the finite/non-finite distinction
encoded in the shshi/rentai contrast in Old Japanese. This distinction started to collapse during
the late Heian period. Perhaps ga/no subjects did not become alternative subject marker in finite
clauses before this collapse. But this later shift does not contradict the hypothesis that the
Agreement parameter had already switched to Non-Forced earlier in the Heian period; the
parameter switch prepared a way for the emergence of an alternative, morphological subject
marking.
What triggered the parameter switch is a question I have to leave open for future studies.

15
On the basis of Yanagida, personal communication, October, 2004.
59
Watanabe (2003:551) proposes that "learnability considerations hold the key" for why wh-
movement was lost. He hypothesizes that the increase in subject topicalization deprived children
of the crucial word order cue for wh-movement. I am concerned with a parameter resetting, not
the change in a single grammatical process. However, it is, I assume, not necessarily the case that
all the grammatical processes that are controlled by a parameter must undergo changes together
simultaneously. A parameter value represents a stable state. Circumstantial and contingent factors
could drive one or more grammatical processes to undergo changes of their own; before the
parameter is reset such changes could instead trigger other changes and drive the language to
converge to a stable state. I earlier indicated that favorable environments already existed in Old
Japanese for the emergence of head internal relative clauses. Much empirical research would be
needed before we understand how Old Japanese transformed into a Non-Forced Agreement
language.

Appendix. Kakari Musubi
A.1. Predicate Conjugation
Japanese verbs and adjectives are traditionally described as taking one of the five forms, mi-zen
'pre-realis' or 'irrealis', ren-y 'ad-predicative', shshi 'conclusive', ren-tai 'ad nominal' and i-zen
'post-realis' or 'presuppositional/conditional'. A clause ends in a predicate complex, consisting of
a verb or adjective stem with or without one or more enclitic predicate stems. The last predicate
stem determines the conjugation form of the predicate complex. The predicate complex may
further be suffixed with an enclitic particle. The enclitic suffix, if there is any, selects the
conjugation form of the predicate complex. The syntactic context determines the conjugation
form of the word. For example, the main clause, in the default case, ends in a predicate complex
in the shshi form.

A.2. Kakari musubi in the narrower sense
The kakari particles drew particular attention of Japanese grammarians because they are
responsible for making exceptions to this default rule for terminating a main clause. Instead, the
rule of kakari-musubi intervenes: if a main clause contains an argument or adjunct with one of
60
the emphatic particles so/zo and nam or with the interrogative ya or ka, it terminates with a
predicate complex in the rentai form; on the other hand, if a sentence contains an argument or
adjunct with the emphatic/selective koso, it terminates in a predicate-complex in the izen form.
We might call kakari-musubi 'suspense and resolution'; a kakari particle introduces a suspense
and it is resolved by a non-shshi form of a predicate.

A.3. Kakari musubi in the wider sense: MOTOORI Norinaga
Exactly speaking, however, this is a description of kakari-musubi in a narrower and commonly
understood sense of the term. In more specific contexts of Japanese linguistic scholarship, the
term may be understood more widely, including cases in which sentences terminate with a
predicate complex in the shshi form; in this wider sense, as conceived by MOTOORI Norinaga,
the great philologist of the 18th century, kakari-musubi specifies the ways sentences are
terminated in general depending on what constituents they contain. Thus, sentences in general
are conceived of containing arguments with special particles that determine how the sentences
end. The particle that determines the form of the predicate complex is a kakari and the predicate
complex that is selected by the kakari is a musubi. Norinaga noted, for example, that the particle
wa, among others, terminates a sentence in the shshi form. As Mizutani (1974:23ff) pointed out,
Norinaga even introduced in his theory of kakari musubi an empty category tada 'plain' as a
kakari, in order to account for cases where bare subjects must be taken as selecting the forms of
predicate complexes.

A.4. Kakari joshi 'kakari particle' in the modern sense: a residue of Norinaga's thought
Norinaga's theory of kakari musubi was not understood and did not properly influence modern
Japanese scholarship; see Mizutani (1974) on this point. However, it indirectly influenced how
the term kakari-joshi is used. Kakari joshi could be defined formally and narrowly, as those that
trigger kakari musubi in the narrow sense. The term is commonly extended to cover those
particles that are not original kakari particles but that share certain syntactic properties with
them. The kakari particle, in the narrow sense, have a functional characteristic in common: they
mark focused constituents; they hence also share certain syntactic characteristics. After the
61
demise of kakari musubi in the late Classical Japanese period, the kakari particles survived with
this function and were joined by new particles with similar functions. As a matter of fact, the old
kakari particles that participated in the classical kakari musubi, except for koso, eventually fell
out of this category of particles. But the term kakari joshi is used even by scholars of Modern
Japanese to refer to FOCUS particles such as sae, dake. We do not have to be concerned with
Norinaga's theory of kakari musubi for our present purposes, but this so to speak anachronistic
use of the term kakari joshi is not irrelevant to the study of Modern Japanese.

A.5. The aftermath of the kakari musubi
The kakari musubi scheme started to collapse during the era of Classical Japanese. We must note
that, as I account for it in section 3.4, kakari musubi is not to be taken as a singular process. We
need to factor it out in two processes and to trace the aftermath of kakari musubi in two lines
separately.
A kakari particle, for example, ka, is selected by the rentai Comp as an Infl. A focused
argument or adjunct moves to Spec(I). The phrase structure is linearised so that the kakari
particle, which is enclitic, might be adjacent to the focused constituent and eventually be
attached to it. Thus, two processes are separated at the observational level: the focus movement
and the particle attachment.
The demise of kakari musubi is commonly understood as the disappearance of clause-
final predicates in the rentai and izen forms. Another relevant fact is that the morphological
distinction between the shshi and rentai forms all but started to disappear during the Heian
period. Besides there is another relevant factor that has not been paid attention to until recently:
the obligatory focus movement and its loss before or early in the Heian period. It is plausible to
assume that these changes were related, one change contributing to causing others. But this is not
the place to engage ourselves in speculating on the question of which started what and how.
What is relevant to our immediate concern is the effect of the loss of obligatory focus movement.
In my account, obligatory focus movement and a proper linearisation of Infl made it
possible for an enclitic kakari particle, for example, ka, to attach to the lexical category and
make a phonological word. But if focus movement is not obligatory, the actualization of this
62
morpho-phonological process is not assured. Two alternatives are conceivable to rescue relevant
derivations from crashing.
First of all, the grammar may let kakari particles stay in situ at Infl and eventually attach
to the clause-final predicates with a proper linearization and make phonological words. The
attachment to a predicate is in effect what happened in Old Japanese when the clause did not
contain a focused element and a kakari particle functioned as a sh-joshi 'clause-final particle'.
Then, understanding a particular constituent as a focus is a matter of interpretation, perhaps aided
by prosody, but not a matter of overt syntax.
Secondly, the grammar may let kakari particles somehow DESCEND DOWNWARD AND
ATTACH to the focused constituents, which remain in situ. Downward attachment can be taken as
an actualization of Agreement between a kakari particle and a focused constituent.
Before proceeding further, let me at this point clarify what were and what have been
members of the class of kakari particles. As I have mentioned earlier, the particles that
participated in kakari musubi in Old Japanese have all disappeared from the language except for
ka. Furthermore, though kakari particles as a class have survived as a class, ka has not stayed in
the class. New comers such as sae have been added to the class and sustained it. Besides, there
are two that are usually taken as kakari joshi but that did not participate in kakari musubi in the
narrow sense and have survived inside the class to the modern day: ha/wa and mo.
With this background, let us return to the discussion of the aftermath of kakari musubi
and consider the second alternative above with the Agreement parameter set as Non-Forced, as
in Modern Japanese. Since Agreement is not forced, a kakari particle may stay at clause final
position (with the understanding of the clause being left to interpretation) or may descend and
attach to a focused element. This is exactly what I assumed for the account of the syntax of
kakari particles in Kuroda (1965) by means of transformations I called ATTACHMENT
transformations, with the caveat that if a particle does not descend, it must eventually be attached
to verb stem to make a phonological word; a light verb suru must be inserted to carry tense and
63
modal morphemes.
16
Whether one takes it as theoretically viable or not, this description gives a
convenient frame of reference as a correct description of the aftermath, if it is, of kakari musubi
in Modern Japanese.
Let me return to Old Japanese. As I have mentioned above, ha/wa and mo are commonly
put in the class of kakari joshi. Of these, ha was not used as a sentence final particle, nor was it
attached to indeterminate phrases. In contrast, mo shares with ka two distinctive properties: (i) it
can be at sentence final position; (ii) it can be attached to indeterminate phrases. I also stated
above that mo as well as ha did not participate in kakari musubi. By saying this, however, I had
in mind the traditional characterisation of kakari musubi: resolving the kakari in the rentai or
izen form. Clauses containing mo phrases do not have such musubi restrictions. Nor did mo as a
sentence final particle take the rentai form of a predicate as ka did. But now we know another
characteristic property of kakari musubi. The kakari phrase must move out of VP. Did mo
phrases also move? To wit, does Nomura's generalization extend to mo phrases? I am not in a
position to answer this question one way or other

References
Harada, S. I. 1973. 'Remarks on relativization'. Annual Bulletin Research Institute of Logopedics
and Phoniatrucs 8.133-144. Reprinted in Harada 2000, pp.237-248.
Harada, S.-I. 2000. Shintakkusu to Imi. Tokyo: Taishukan.
Ishigaki, Kenji. 1942. 'Saysei ygen hanpatsu no hsoku'. Kokugo to Kokubungaku.
Ishigaki, Kenji. 1959 Joshi no Rekishiteki Kenky. Tokyo: Iwanami Shoten.
Kinsui, Satoshi. 2002. 'On the nominative and accusative case markers in Manysh.' Handout.
Kitagawa, Yoshihisa and Masanori Deguchi. 2002. Prosody in syntactic analysis. Ms. Indiana
University.
Kitayama, Keita. 1951. Genji Monogatari no goh. Tokyo: Tk Shoin.

16
I used the term huku-joshi in Kuroda (1965) instead of kakari joshi, following the then standard Hashimoto
grammar. Incidentally, downward movement was generally abhorred in the transformational grammar, perhaps one
reason why the attachment transformation has not been accepted by the mainstream generative grammar. In the
minimalist approach, however, a descriptive equivalent of the attachment transformation could be accommodated in
the form of upward movement.

64
Kji, Kazuteru. 1980. Manysh Joshi no Kenky, Tokyo: Meiji Shoin.
Konoshima, Masatoshi. 1966. Kokugo Joshi no Kenky : Joshishi no Soby. Tokyo: fsha.
Kond, Yasuhiro. 1981. 'Chko-go no rentai kz nitsuite'. Kokugo to Kokubungaku.
Kond, Yasuhiro. 2000. Nihon-go Kijutsu Bunp no Riron. Kasugabe: Hitsuji Shoh.
Kuno, Susumu. 1973a. Nihon Bunp Kenk. Tokyo: Taishkan.
Kuno, Susumu. 1973b. The Structure of the Japanese Language. Cambridge, MA: MIT Press.
Kuroda, S.-Y. 1974. 'Pivot-independent relative clauses in Japanese, I'. Papers in Japanese
Linguistics 3.59-93. Reprinted in Kuroda 1992, pp.118-145.
Kuroda, S.-Y. 1986. '(D')Accord ou pas d'accord: Quelques ides gnrales concernant une
grammaire comparative de l'anglais et du japonais'. Pierre Pica (Ed.) Recherche Linguistique
14: Hommage Mitsou Ronat. Paris: Presse de l'Universit de Paris VIII, pp. 189-206.
Kuroda, S.-Y. 1988. 'Whether we agree or not: a comparative grammar of English and Japanese',
in Linguisticae Investigationes. 12.2-44. Reprinted in W. Poser (Ed.) Working Papers from
the Second SDF Workshop in Japanese Syntax, Stanford: CSLI, 1988, pp.103-144 and in
Kuroda 1992, pp. 315-157.
Kuroda, S.-Y. 1990. 'Shieki jodshi no jiritsusei ni tsuite'. in Shigeru Tsuchida et al. (Eds.),
Bunp to Imi no Aida, Tokyo: Kurosio Shuppan, pp. 93-104.
Kuroda, S.-Y. 1992. Japanese Syntax and Semantics. Dordrecht: Kluwer Academic Publishers.
Kuroda, S.-Y. 1993. 'La syntaxe du causatif japonais et l'organisation de trois composantes
syntaxiques de la grammaire', Recherches Linguistiques de Vincennes 22, 75-93.
Kuroda, S.-Y. Forthcoming. Prosody and the syntax of indeterminates. D. Rohers, O-H. Kim
and Y. Kitgawa (Eds.). Indiana University Working Papers in Linguistics-Syntax and
Beyond.
Masunaga, Kiyoko. 1988. 'Case deletion and discourse context'. William J. Poser (Ed.), Papers
from the Second International Workshop on Japanese Syntax. Stanford: CSLI, pp.145-192.
Mizutani, Shizuo. 1974. Kokugogaku Itsutsu no Hakken, Sai-hakken. Tokyo: Sbunsha.
Nomura, Takashi. 1993. 'Jdaigo no no to ga ni tsuite.' Kokugo-Kokubun 62.524-541.
no, Susumu. 1993. Kakari Musubi no Kenky. Tokyo: Iwanami Shoten.
Saeki, Umetomo. 1963. Many Goh Kenky. Tokyo: Yhd.
Saito, Mamoru. 1989. 'Scrambling as Semantically Vacuous A'-Movement'. In Mark Baltin and
65
Anthony Kroch (Eds.) Alternative Conceptions of Phrase Structure. Chicago: University of
Chicago Press.
Sasaki, Takashi. 1992. Nihongo no Kbun to Hyki, Kasugabe: Hitsuji Shoh.
Takahashi, Daiko. 1993. 'Movement of wh-phase in Japanese'. Natural Language and Linguistic
Theory 11.655-678.
Tonoike, Shigeo. 2002. 'Kodai Nihongo ni sah wh id wa atta ka', Gengo 31.3.86-91.
Tonoike, Shigeo. 2004. Seisei Bunp to kakari musubi. Handouts. Lectures at the University of
Tokyo.
Watanabe, Akira. 2001. 'Minimarisuto purogugamu nymon'. 6: Kodai Nihongo o miru', Gengo
30.6.98-95.
Watanabe, Akira. 2002a. 'Nihongo ni okeru sah wh id saisetsu'. Gengo 31.8.78-81.
Watanabe, Akira. 2002b. 'Loss of overt wh-movement in Old Japanese'. D. W. Lightfoot (Ed.)
Syntactic Effects of Morphological Change. Oxford: Oxford University Press, pp. 179-195.
Watanabe, Akira. 2003. 'Wh and operator constructions in Japanese', Lingua 113.519-558.
Yanagida, Yuko. 2003. 'Obligatory movement and head parameter: evidence from early Old
Japanese'. Sophia Linguistica 50.103-16.
Yanagida, Yuko. 2004. Topic, focus and clause structure in early Old Japanese, ms. The
University of Tsukuba.

Texts and Dictionaries
Nihon Koten Bungaku Taikei. Vol. 3. Kodai Kay Sh. Tokyo: Iwanami Shoten. 1957.
Nihon Koten Bungaku Taikei. Vol. 4-7. Manysh. Tokyo: Iwanami Shoten. 1957-1962
Nihon Koten Bungaku Zensh. Vol. 1. Kojiki, Jdai Kay. Tokyo: Shgakkan. 1973.
Shin Nihon Koten Bungaku Taikei Vol. 1-4. Manysh. Tokyo: Iwanami Shoten. 1999-2003.
Jidaibetsu Kokugo Daijiten: Jdai Hen. Tokyo: Sanseid. 1977.
Manysh Ssakuin. Tokyo: Heibonsha. 1974.