Difference between revisions of "Vulcan/TupleRepresentation"

From Knowitall
Jump to: navigation, search
Line 1: Line 1:
* In designing output for extraction from definition sentences, it is useful to use both
+
* In designing output for extraction from definition sentences, it is useful to use both nested and n-ary tuples. Here are some examples.
nested and n-ary tuples.  
 
 
 
Here are some examples.
 
  
 
<pre>
 
<pre>
Line 31: Line 28:
  
  
* For deep understanding and reasoning, we want high coverage of the information in each sentence.
+
* For deep understanding and reasoning, we want high coverage of the information in each sentence. Current Open IE often misses much of the information in a given sentence.   
Current Open IE often misses much of the information in a given sentence.   
 
  
 
* The nested and n-ary representation will make inference rules more compact.
 
* The nested and n-ary representation will make inference rules more compact.

Revision as of 17:56, 3 September 2013

  • In designing output for extraction from definition sentences, it is useful to use both nested and n-ary tuples. Here are some examples.
Example1:
Adaptation helps an organism survive in its environment
Pattern: X rel1 Y rel2 Z

Open IE 4.0 represents this as triples:
(adaptation; helps; organism survive in its environment)
(organism; survive; in its environment)

Should we make the nested tuple explicit:
(adaptation; helps; (organism; survive; in its environment))

Note that the nested form can be recovered from the pair of triples.

Example2:
The kidneys filter waste materials out of the blood
Pattern: X rel Y prep Z

Open IE 4.0 represents this as  an n-ary tuple:
(The kidneys; filter; waste materials; out of the blood)


Note that additional arguments in n-ary tuples always(?) begin with a
preposition.


  • For deep understanding and reasoning, we want high coverage of the information in each sentence. Current Open IE often misses much of the information in a given sentence.
  • The nested and n-ary representation will make inference rules more compact.

Without nesting, a single sentence is represented as two separate tuples; without n-ary, a single sentence becomes multiple tuples, some of which don't make much sense on their own.

  • We expect inference rules will be easier to express and easier to learn

if they operate on nested and n-ary tuples, rather than the sets of binary tuples that are equivalent to the nested or n-ary tuples.

  • The sample Vulcan propositions we are trying to prove are often nested or

n-ary.

Examples 2 and 3 are nested; 1 has a complex nesting; 4 is n-ary.
Example 1 also has sentence internal coreference to get (turtle; take in;
nutrients).

	1. "A turtle eating worms is an example of taking in nutrients"
	((turtle, eat, worms);  is;  example of (turtle; take in;
nutrients))

	2. "Growth causes the leaves of a plant to become larger"
	(growth;  cause;  (leaves of plant;  become;  larger))    

	3. "A sense of smell helps a fox find food"
	(sense of smell;  help;  (fox;  find;  food))

	4. "Gravity pulls the marble to the bottom of the glass"
	(gravity;  pull;  marble;  to bottom of glass)