From: "Furukawa Yukio" <yfurukawa@tokai.or.jp>
Date: Sat, 13 Dec 1997 01:19:05 +0900 
// revised from 11 Dec

Title: Comments on "A Comparative Analysis of SVCs and Korean V-V
Compounds"     (Corrected Version)
Author: Yukio Furukawa 
(the Graduate School of Kanda University of International Studies, Japan)
e-mail: yfurukawa@tokai.or.jp


Japanese, like Korean, is a head final language and also has V-V compounds,
as Kang pointed out through citing Nishiyama(1996). The crucial difference
between KVVC and JVVC is that Japanese has the resultative SVCs even though
it's a head final language. This might be a big problem both for Kang and
Collins. According to Kang's analysis, in the head final languages like
Korean, the head of the compound, V2, which has to be preceded by the event
of V1 temporally, can neither implicate VP1, nor be a predicate of the
direct argument of V2. However, Japanese accepts this resultative
compounds, and V1 can be the predicate of the internal argument of V2. The
crucial evidence of this is that unaccusative verbs come to the position of
V2 even though V1 is unergative or transitive.

(1)    sagashi-mawaru     (search-go around)    'go around with searching'
       motome-yuku        (seek-go)                  'go seeking'
       tsure-saru         (take-leave)              'take away/kidnap'
       hashiri-saru       (run-leave)                'run away'
       uchi-agaru         (hit-go up)         'be hit high up (in the air)'

The point is the internal argument of the higher V2 is also the external
argument of the lower V1. For example, in (2), the internal argument of V2
'sar'(leave), i.e. 'John' is also the external argument of V1
'hashir'(run).

(2)   John-ga      hashir-i    sar-ta
       John-Nom  run-Infl    leave-Past
       'John ran away'

In the framework of the Minimalist Program, every external argument has to
belong to the Spec of vP to get its external theta-role. What I'm worrying
here is (a) this vP is not for the head V2 but for V1 as far as the
external theta-role is assigned in (2), and (b) even if this assigning is
possible, both V1 and V2 have to adjoin to the head of vP. In that case, V2
will turn into transitive, or will get a transitive morpheme, at least.
However, 'sar' in (2) still remains as an unaccusative verb. Thus, V2 does
not move overtly, while V1 does, even though it also has to adjoin V2 for
HMC.

There is also an interesting fact in Japanese, i.e. V2 must be in the form
of transitive iff an accusative object appears.

(3) a.    John-ga     rocket-o     uch-i    ag-e-ta
           John-Nom  rocket-Acc hit-Infl  go up-Tr.-Past
           'John launched the rocket'
    b.    *John-ga     rocket-o      uch-i    ag-ar-ta
           John-Nom   rocket-Acc  hit-Infl  go up-Unacc.-Past
    c.    Rocket-ga       uch-i    ag-ar-ta
           Rocket-Nom  hit-Infl  go up-Unacc.-Past
           'The rocket was launched'
    d.    *Rocket-ga     uch-i    ag-e-ta
           Rocket-Nom  hit-Infl  go up-Tr.-Past

Then, a possible solution for these two problems above is that vP need not
come to the top of the VP shells. If vP is established between VP1 and VP2
in (3c), the external theta-role will be assigned to the implicit agent and
only V1 will be raised to the head of this intermediate vP. In (3a), vP
must appear on the top of the shells, and the transitivity of both two
verbs will be checked in it head. (Here, I have to assume that unergative
verbs are also transitive as is pointed out by Hale & Keyser.)

My proposal is quite radical to the excellent jobs done by Collins,
Chomsky, and Hale & Keyser, but as far as JVVC is concerned, it is
reasonable to assume the intermediate vP hypothesis.