Set Constraints In Computational Linguisti s

Description

Set Constraints In Computational Linguisti s | Solving Tree Des riptions | Denys Du hier University of the Saarland du hierps.uni-sb.de June 30, 1999 Computational linguisti s has always been a priviledged appli ation domain for onstraint programming. While attention has traditionally fo used on fea- ture onstraints, set onstraints are now emerging as aording espe ially elegant and on ise de larative formulations that naturally obtain very eÆ ient opera- tional semanti s using onstraint propagation. In [DG99, DN99? we show how to reformulate, in terms of set onstraints, the problem of nding minimal models of tree des riptions expressed in an extended version of dominan e logi . In [Du 99? we provide an a ount of parsing, in the framework of dependen y grammar, using set onstraints, that is both a su in t de larative spe i ation of the problem as a CSP and a very eÆ ient implementation when regarded as a on urrent onstraint program. Both appli ations an be viewed as highly ombinatorial onguration problems and set onstraints allow elegant de larative spe i ations that are also eÆ ient onstraint programs for enumerating solutions. In this paper, we propose to des ribe our appli ation of set onstraints to the problem of nding solutions of tree des riptions.

• ied representation


• trees rather


• formedness onstraint


• semanti underspe


• onstraint


• rather than


• support pre eden

Informations

• Publié par: pefav
• Nombre de lectures: 16
• Langue: English

Extrait

Set
Constrain
ts
parsing,
sets
in
In
er,
Computational
℄
Linguistics
as
|
and
Solving
problem
T
supp
ree
it
Descriptions
terms
|
e
Den

ys
programs

our
hier
sets
Univ
h
ersit
in
y
T
of
of
the
mited
Saarland
paren

VS95


June
and
30,
biguities,
1999

Computational
solutions.
linguistics
ose
has

alw
of
a

ys
in
b
t
een
een
a
Oz
priviledged
3
application
Linguistics
domain
useful
for
trees

whic
t
ab
programming.
relation
While
formal
atten
[R
tion
tree
has
man
traditionally
[MHF83
fo
R

[Mus95
on
and
fea-
℄
ture
are


ts,
en
set
this

e
ts
describ
are
of
no
to
w
nding
emerging
descriptions.
as
w
aording
are
esp
non-negativ
ecially
for
elegan
ery
t
t
and
has

elop

programming
e
,
form
the
ulations
ears.
that
Descriptions
naturally

obtain
frequen
v
manipulate
ery
rather

es.
t

op
p
era-
sp
tional
trees
seman
the

than
using
relation.

ork
t
orated
propagation.
,
In
Since
[DG99,
ha
℄
applications
w
areas
e
linguistics:
sho
for
w
[VS92
ho
for
w
grammars,
to
for
reform
seman
ulate,
e
in
h
terms
to
of
that
set
also

t
ts,
t
the
for
problem
umerating
of
In
nding
pap
minimal
w
mo
prop
dels
to
of
e
tree
application
descriptions
set
expressed
ts
in
the
an
of
extended
solutions
v
tree
ersion
The
of
that
dominance
e

sider
In
nite
[Duc99
of
℄
e
w
tegers,
e
whic
pro
v
vide

an


programming
t
ort
of
b
parsing,
dev
in
ed
the
the
framew
language
ork
[Smo95
of
MC98
dep
for
endency
past
grammar,
y
using
1
set
ree

in
ts,
In
that
linguistics
is
is
b
tly
oth
to
a
descriptions

trees

than
e
themselv
sp
℄

tro
of
D-Theory
the
h
problem
er-
as
to
a
eak
CSP
out
and
in
a
of
v
domination
ery
rather

the
t
t
implemen
The
tation
framew
when
w
regarded
elab
as
in
a
VS92

BR
t
℄

then,
t
descriptions
program.
v
Both
found
applications
in

y
b
of
e
tational
view
in
ed
℄
as

highly
in

,
binatorial
VSW95]

tree-adjoining
problems
D-tree
and
in
set
℄

undersp
ts
in
allo

w
scop
elegan
am
t
whic

[EKNR99
e
extended
sp
parallelism

1ellipsis,
and
in
[GW98]
x
v

for
x
discourse
relationships.
represen
x
tation.
6
Y
loves
et
C
little
tion
w
x
as
x
kno
;
wn


:

℄
the
b

there
t
realize

x
treatmen
12
t
^
of
)
tree
(
descriptions.
^

w
tly
:
,
x
t
12
w
ecied
o
Since
algorithms
p
w
ould
ere
y
prop
This
osed
14
for
adv
solving
Seman
dominance
of

y
ts:
r
[KNT98

℄
C
prop

osed
u
a
:
saturation-based
x
metho
x
d
(
motiv
and
ated
^
b
10
y
u
an
^
application
x
to
11
undersp
13
ecied
u
seman
Figure

tation
while
Solving
[DG99
ts
℄
exp
describ
it.
ed
test
an
alternativ

ula
ding

in
x
to

set
ariables

3
strain
hnique
ts

for
del
a
Seman
treatmen
t
t
utterance
of
or
discourse.
10
W
11
e

prop
r
ose
13
to
x
describ
x
e
0
the
2

:
e
x
axiomatization
x
in
v
to
)
set
:

4
ts
)
rst
:
prop
6
osed
7
in
x
[DG99].
9
This
8
axiomatiza-
(
tion
^
has
va
a
x

r

5
reading
11
as
C
a
^

love
t
;

^
t
va
program
x
that
r
implememen
Under
ts

a
\every
v
woman
ery
Constrain

dominance
t
NP-complete
solv
e
er.
an
T
algorithm
ree
na
Descriptions
generate
with
d
Dominance
umerate
Constrain
giv
ts.
the
Dominance
x

x
ts
_
describ
x
e
y
relations

b
Figure
et
14
w
hence
een
14
the
p
no
our
des
es
of
tage
a
propagation
tree.
e
[KNT98
Application
℄
Undersp
denes
undersp
them
to
as
ossible

natural
junctions
an
of
equal
form

ulae
x
x
love
C
x

va
y
u
stating
x
that
va
a
v
no
x
de

x
9
dominates
0
a

no
1
de
x
y
C
,
x
1
^
and
1
x
8
:
(
f
3
(
^
y
2
1
9
:
(
:
7
:
^
y
3
n
imply
)
x
stating
;
that
5
x
^
is
4
a
man
no
x
de
)
lab
x
eled
:
with
(
n
8
-ary
x
sym
)
b
x
ol
:
f
oman
and
x
a
)
se-
x

:
of
r
immediate
^
daugthers
10
y
va
1
v
;
x
:
C
:
x
:
^
y
9
n

.
11
Dominance
x

:
t
(
ha
12
v
x
e
)
trees
x
as
:
mo
r
dels,
^
i.e.
13
ev
va
ery
v
v
1:
ariable
sp
denotes
seman
a
represen
no
of
de
man
of
a
a
"
tree.
Dominance
Figure
ts.
1
solving
displa

ys
is
an
[KNT98
exam-
w
ple

of
ect
a
y
tree
olynomial
description
for
with
A
dominance


e
ts:
and
it
metho
sho
w
ws
en
b
the
oth
es
a
en
form
y
ula
form
and
8
a
8

:

C
of
y
its
y


ts.
_

k
x
.
0
is
8
not
u
in

1
x
are
1
v
!
and

3
x

3
=
man
196

ossible
x
Instead
4

va
tak
r
full
u
an

of
x
t
6
to

eectiv
x
mo
5
elimination.
9
to
v



x

2

^
aims

represen
x
p
7
meanings
w
a
oman
language

in
x
1
8
is
va
to,
r
and
v
of

2undersp
eci