Difference between revisions of "Vulcan/SystemPrototype"

From Knowitall
Jump to: navigation, search
(Status)
(Status)
Line 22: Line 22:
 
</blockquote>
 
</blockquote>
 
</blockquote>
 
</blockquote>
* <b>Debug information</b>
+
* <b>Debug information</b> Tuffy doesn't have much diagnostics built in.
 
<blockquote>
 
<blockquote>
* Need to know how the score is computed.  
+
* Need to know how the score is computed. <b>[To Do]</b>
* How to obtain the trace tree?
+
* Can compute ground facts (tuples) (including "hidden" i.e., new facts and their associated probabilites).
* Read through the output. See if there is useful information to extract already.
+
* We can currently print clauses (first-order rules) that are reachable from the query (KBMC style).  
 
+
* Re-constructing exact instantiations of rules will take some additional work.  
* <b>Couldn't find enough information in current debug output from Tuffy. Emailed Tuffy developers asking for help.</b>
 
 
</blockquote>
 
</blockquote>
 
* <b>Proposition generator</b> Check the output of Open IE 4.0 on the input sentence.  
 
* <b>Proposition generator</b> Check the output of Open IE 4.0 on the input sentence.  

Revision as of 18:26, 27 August 2013

Overview

The prototype is designed to work on three questions. We want the system to output the following:

  • Score for the input proposition.
  • New facts inferred.
  • Facts and rules used in scoring.

Status

  • Score The system outputs scores for each query predicate. If query is not in output then score is zero.
  • Unable to materialize Nested tuples. Skipping this until this becomes necessary. The main reason we'd need this is to score the nested tuple. For now we can compute this from the score of its components: Score(nested_tuple) = Score(top tuple) * Score (nested)
  • Knowledge: Worked out the facts and rules required.
  • Have a stub for evidence selection.
  • Run with baseline implementation.
  • Output debug information.
  • Debug information Tuffy doesn't have much diagnostics built in.
  • Need to know how the score is computed. [To Do]
  • Can compute ground facts (tuples) (including "hidden" i.e., new facts and their associated probabilites).
  • We can currently print clauses (first-order rules) that are reachable from the query (KBMC style).
  • Re-constructing exact instantiations of rules will take some additional work.
  • Proposition generator Check the output of Open IE 4.0 on the input sentence.
  • Run as is.
  • Run with modified tuple.