CogComp/cogcomp-nlp

bug in WNSim

Opened this issue · 0 comments

duplicate call for two supposedly different scores:
https://github.com/CogComp/cogcomp-nlp/blob/master/similarity/src/main/java/edu/illinois/cs/cogcomp/wsim/wordnet/PathFinder.java#L163
presumably, supposed to consider both directions for the relationship by default (i.e. rel(w1, w2) and rel(w2, w1) which makes sense if the wn relations specified for the call are asymmetric -- which they presumably are.

  1. verify intent of wnsim() behavior.
  2. evaluate change on relevant test set.
  3. parameterize to make bidirectional comparison a choice (right now there isn't and it is OK because of the bug described above; but client should have the choice of asymmetric/symmetric function)