Re: ARG-S and Binding

Stefan Mueller (Stefan.Mueller@dfki.de)
Sat, 17 Oct 1998 14:09:05 +0200


Hi everybody,

> Does anyone know of any work done on the possibility of composing
> the Argument Structure lists of different (verbal) heads?

I suggested such an approach in order to do case assignment on ARG-S. I
wanted to assign case on ARG-S since Complement Extraction Lexical Rules
change the position of the elements on COMPS, i.e. the element that is
at the second position in a lexical entry is at the first position if
the CELR extracts the first argument.

If you do not use the CELR you don't need to assign case on ARG-S and it
depends on your version of the Binding Theory whether you want things
like Argument Attraction to take place on ARG-S.

> As I said, does anyone know if/where this has been suggested and
> whether there would be any problematic consequence arising from the
> propagation and composition of argument structures of different heads?
>

with the totally nonconfigurational BT (PS94, Chapter 6.8.3).

Take for instance the German sentence

(1) Der Sheriff wird den Dieb  sich "uberlassen.
the sheriff will the thief self leave.to.his.one.devices
The sheriff will the thief leave to his one devices.'

Since Hinrichs and Nakazawa (1989) such constructions are handled with
argument attraction, i.e. the future auxiliary 'wird' attracts all
arguments of the verb '"uberlassen'. If you do this on the list that is
used for BT you get a paradox situation.

Verb        ARG-S

wird        Sheriff, Dieb, sich, "uberlassen
"uberlassen Sheriff, Dieb, sich

The table above shows the two argument structures of the verbs.
'"uberlassen' takes a subject (Sheriff), an accusative object (Dieb) and
the dative object (sich). All arguments of '"uberlassen' are attracted
by 'wird'. Therefore the ARG-S of 'wird' contains the same elements as
the one of '"uberlassen' plus '"uberlassen' itself.

The definition of o-command is as follows (PS94, p. 279):

Let Y and Z be synsem objects, with distinct LOCAL values, Y
referential. Then Y o-commands Z just in case either:

i.   Y is less oblique than Z; or
ii.  Y o-commands some X that subcategorizes for Z; or
iii. Y o-commands some X that is a projection of Z (i.e. the HEAD values
of X and Z are token-identical).

In (1) 'Dieb' locally o-commands 'sich' since 'sich' is more oblique.
'sich' o-commands '"uberlassen', and '"uberlassen' subcategorizes for
'Dieb'. Because of (ii) 'sich' o-commands 'Dieb', i.e. 'Dieb' and 'sich'
o-command each other. Sentences like (1) are ruled out by principle C
since 'sich' o-binds 'Dieb'.

This is independent of the order you choose for the Obliqueness
Hierarchy.

> Otherwise one needs an additional mechanism to permit
> the link to take place.

I think that additional mechanisms are needed since the notion of local
o-command cannot account for sentences like (2).

(2) Martin Walser versucht, sich und die Nation zu verstehen.    taz 12.10.98, S. 1
Martin Walser tries     self and the nation to understand   (http://www.taz.de)
Martin Walser tries to understand himself and the nation.'

For explanatory purposes, I take a simpler example. As I am not a native
speaker of English, I am using German examples, but I think the English
translations behave very similar.

(3) Er sorgt fuer [sich und seine Familie]
he cares for   self and his   family

Coordinated noun phrases show plural agreement.

(4) John and Mary know/*knows London.

If agreement is done semantically as in PS94, the index of 'John and
Mary' is plural, which is semantically plausible as well. For our
example above this means that we have a plural index for [sich und seine
Familie]. This means that 'sich' in (3) is not local to 'er' and
according to PS94, it is an exempt anaphor, i.e. it can be bound to
whatever we like (PS94, p. 266), which is not quite true.
Furthermore bindings like in (5) are allowed, since 'ihn' is not local
to 'er'.

(5) Er_i sorgt fuer [ihn_*i und seine Familie].
he   cares for   him   and his   family

Did I miss something? Are there solutions to this?

Greetings

Stefan

P.S. These things and other problems with BT are discussed
in (Mueller, 1999) in some more detail. I am about to write
these things down in English, so if somebody is interested
I could provide a draft in some weeks.

St.

@incollection{HN89,
Author    = {Erhard W. Hinrichs and Tsuneko Nakazawa},
title     = {Subcategorization and {VP} Structure in {German}},
Booktitle = {Aspects of German VP Structure},
Publisher = {Eberhard-Karls-Uni\-ver\-si\-t{\"a}t T\"ubingen},
Series    = {SfS-Report-01-93},
YEAR      = 1989
}

@techreport{Mueller97d,
author      = {Stefan M{\"u}ller},
title       = {Complement Extraction Lexical Rules and Argument
Attraction},
note        = {A slightly different version appeared in
{\it Natural Language Processing and Speech Technology.
Results of the 3rd KONVENS Conference, Bielefeld,
October 1996\/}.
http://www.dfki.de/~stefan/Pub/e_case_celr.html},
type        = {{Research Report}},
number      = {{RR-97-08}},
institution = {Deutsches Forschungszentrum f\"ur K\"unstliche
Intelligenz},
year        = 1997
}

@book{Mueller99,
author    = {Stefan M{\"u}ller},
title     = {{Deutsche Syntax deklarativ.
Head-Driven Phrase Structure Grammar f\"ur das
Deutsche}},
publisher = {Max Niemeyer Verlag},
series    = {Linguistische Arbeiten},
number    = 394,
note      = {http://www.dfki.de/~stefan/Pub/e_hpsg.html ISBN 3-484-30394-8},
year      = "To appear"
}

@book{PS94,
AUTHOR    = {Carl J. Pollard and Ivan A. Sag},
title     = {Head-Driven Phrase Structure Grammar},
Publisher = {University of Chicago Press},
SERIES    = {Studies in Contemporary Linguistics},
Year      = 1994
}

--
DFKI GmbH, FB Sprachtechnologie   Tel.: (+49 - 681) 302 - 5295
Stuhlsatzenhausweg 3              Fax:  (+49 - 681) 302 - 5338
D-66123 Saarbruecken              http://www.dfki.de/~stefan/

http://www.dfki.de/~stefan/Babel/Interaktiv/Babajava/


This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:38:22 PST