[developers] Bug in LUI interactive unification

Dan Flickinger danf at stanford.edu
Wed Mar 9 20:48:45 CET 2011


Glenn Slayden at UW reports the following misbehavior with LUI when doing interactive unification, where in one order of unifying the two daughters of a rule, unification wrongly succeeds, while reversing the order in which the two daughters are unified in, unification fails as intended.  Below is Glenn's recipe for reproducing the problem, which I just did reproduce on the trunk version of LOGON.

 Dan

-----------
1. Loaded ERG trunk svn rev. 8962 (2/25/2011) into Trollet/LKB/LUI
2. Entered the following into the input area, obtaining one parse.
        Towards Solheimslien the path on the eastern edge is followed.
3. menu: Parse/Show Parse Chart
4. from second row, first cell, click on FHNI and then collapse the avm
5. from bottom row, last cell, click on SCPS and then collapse the avm
6. menu: View/Grammar Rule/HD-AJ_SCP_C, collapse so that ARGS are visible
7. drag FHNI into ARGS[0] of the rule
8. drag SCPS into ARGS[1] of the unification result

--> this gives a successful unification of the case you described in detail below, but as you point out, it should have failed at anti-synsem. This successful unification is correctly not in the parse chart, so the interactive unification should also have failed (in which case I wouldn't have been confused in the first place!).

9. If you instead reverse the order of steps 7 and 8 (above), the unification does fail.



More information about the developers mailing list