Database Reference
In-Depth Information
5.3.4 R
ELATING SUMMARY
car accident
TO TEXT
member proplets
owner proplets
...
⎡
⎣
⎤
⎦
noun: accident
mdr: (car 1)
prn: 67
...
core: accident
...
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
...
⎡
⎣
⎤
⎦
noun: car
fnc: hit
prn: 1
noun: (car 1)
fnc: speed
prn: 2
noun: (car 1)
mdd: accident
prn: 67
...
core: car
...
...
⎡
⎣
⎤
⎦
...
verb: hit
arg: car tree
prn: 1
...
core: hit
...
⎡
⎤
verb: speed
arg: (car 1)
prn: 2
...
core: speed
⎣
⎦
...
...
The connection between propositions 1 and 2 and summary 67 is coreference-
by-address. It is based on the original
car
value in proposition 1, the corre-
sponding address value
(car 1)
in proposition 2, and similarly in summary 67.
The summary consists of the
accident
proplet and a
car
pointer proplet. They
share the
prn
value
67
, and are connected by the modifier-modified relation.
The summary and the original text are connected by the address value
(car 1)
,
which serves as the core value of the second
car
proplet as well as the
mdr
(modifier) value of the
accident
proplet.
How is the summary automatically derived from the text? The summary-
creating inference deriving the new content with the
prn
value
67
is formally
defined as the following D(eductor) inference rule, shown with the sample
input and output of 5.3.4 at the Word Bank (content) level:
5.3.5 S
UMMARY
-
CREATING
D
INFERENCE
antecedent
consequent
⎡
⎤
⎡
⎤
⎡
⎤
⎡
⎤
⎡
⎤
noun:
α
fnc: hit
prn: K
verb: hit
arg:
αβ
prn: K
noun:
β
fnc: hit
prn: K
noun: (
α
K)
mdd: accident
prn: K+M
noun: accident
mdr: (
α
K)
prn: K+M
rule
level
⎣
⎦
⎣
⎦
⎣
⎦
⎣
⎦
⎣
⎦
sum
where
α
{car, truck, boat, ship, plane, ...} and
β
{tree, rock, wall, mountain, ...}
∪ α
matching and binding
⇑
⇓
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
⎡
⎣
⎤
⎦
noun: car
fnc: hit
prn: 1
verb: hit
arg: car tree
prn: 1
noun: tree
fnc: hit
prn: 1
noun: (car 1)
mdd: accident
prn: 67
noun: accident
mdr: (car 1)
prn: 67
Wo rd
Bank
input
output
Antecedent and consequent are related by the connective
sum
(marize).
Search WWH ::
Custom Search