Transcript Using LG and WordNet..

Using Link Grammar and
WordNet on Fact Extraction
for the Travel Domain
Problem Statement



Extract facts from English sentences
Reasoning can be done on extracted facts
How to extract facts?


Using the connectivity feature of Link Grammar to
associate relations between extracted facts
Using WordNet to associate semantic meanings
to extracted facts
WordNet

What is WordNet?



Developed by the Cognitive Science Laboratory
at Princeton University
Synonym sets of nouns, verbs and adjectives
Various semantic relations connect the synonym
sets



car is_a motor vehicle
breathe entails inhale
present and absent are antonyms
An example of is_a hierarchy

airplane, aeroplane, plane -=> heavier-than-air craft
=> aircraft
=> craft
=> vehicle
=> conveyance, transport
=> artifact, artefact
=> object, physical object
=> entity
=> whole, whole thing, unit
=> object, physical object
=> entity
Fact Extraction

Prerequisites:

knowledge base for travel objects





plane is_a transportation
ticket is_a document
knowledge base for geographic locations
Input: a simple English sentence related to
the travel domain
Output: facts extracted from Link Grammar
output and association with WordNet
Example

John took a plane from Paris to Baghdad.
event(e1,took)
parameter(e1,from,paris)
actor(e1,john)
object(e1,plane)
parameter(e1,to,baghdad)
place(paris)
place(baghdad)
Associate WordNet Sense

For object(e1,plane), the word “plane” exists in the
travel object knowledge base.




plane is_a transportation
Stem the word transportation => transport.
For event(e1,took), do WordNet search on “took”
and parse the search result to match the word
transport.
There is a match in the parsing, so associate
event(e1,took) to sense 11 as event(e1,took,11).
Associate WordNet Sense

There is a match in the parsing, so associate event(e1,took) to sense 11 as
event(e1,took,11).
Extracted Facts

John took a plane from Paris to Baghdad.







actor(e1,john)
parameter(e1,from,paris)
parameter(e1,to,baghdad)
place(paris)
place(baghdad)
object(e1,plane)
event(e1,took,11)
Future Work



Currently only utilizes the noun and verb
parts of WordNet.
Expand current work by utilizing the adjective
part of WordNet and the semantic relations.
Work on more examples to generalize the
algorithm.