[osmt] mle/412: robustness to reconstruction errors

[oe at csli.stanford.edu]

>Number:         412
>Notify-List:    osmt at emmtee.net
>Category:       mle
>Synopsis:       robustness to reconstruction errors
>Confidential:   no
>Severity:       critical
>Priority:       high
>Responsible:    oe
>State:          open
>Class:          feature
>Submitter-Id:   remote
>Arrival-Date:   Wed Nov 15 21:30:43 +0100 2006
>Originator:     oe at csli.stanford.edu (Stephan Oepen)
>Release:        
>Organization:
>Environment:
>Description:
while creating the feature cache, it must not be the case that items are
entered into the cache for which individual derivations failed to rebuild.
otherwise, there will be a mis-match of events vs. results downstream, and
the reading-in of scores and everything gets confused.  i think the best
thing we can do in the light of mismatches between the treebank and the
grammar is ignore complete items (and complain bitterly, this should not
happen in principle).

>How-To-Repeat:
>Fix:
>Unformatted: