[osmt] generation/388: skolemize HOOK of semantically vacuous lexical entries

oe at csli.stanford.edu oe at csli.stanford.edu
Mon Aug 28 11:29:04 CEST 2006


>Number:         388
>Notify-List:    osmt at emmtee.net
>Category:       generation
>Synopsis:       skolemize HOOK of semantically vacuous lexical entries
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    oe
>State:          open
>Class:          feature
>Submitter-Id:   remote
>Arrival-Date:   Mon Aug 28 11:29:02 +0200 2006
>Originator:     oe at csli.stanford.edu (Stephan Oepen)
>Release:        
>Organization:
>Environment:
>Description:
for improved generation efficiency, it would be helpful to associate the
trigger rules for semantically vacuous lexical entries with information
about the LBL and INDEX (and ideally XARG, i understand) of the entry to
be activated.  then, even for semantically vacuous chart entries, there
could be some Skolemization to constrain their combinatories; grammars
would have to decide on how to use this facility, e.g. for an auxiliary
it might be feasible to have it share the LBL and INDEX of its main verb
complement, thus Skolemizing would effectively `tie down' the auxiliary
to only combine with the appropriate main verb.  now, if we were to do
this, the additional complication arises that we might have to add more
than a single instance of a semantically vacuous LE into the chart, just
as we do for semantically non-empty rules already: say the auxiliary
trigger fired against multiple targets in an input MRS, thus giving rise
to multiple, distinct HOOK values, we would need several copies of the
lexical entry in question.
>How-To-Repeat:
>Fix:
>Unformatted:



More information about the logon mailing list