[developers] What does it mean in ACE for a predicate to be 'covered' in generation?

Michael Wayne Goodman goodmami at uw.edu
Thu Dec 21 21:42:46 CET 2017

Thanks for the explanation, Dan,

On Thu, Dec 21, 2017 at 2:24 AM, Dan Flickinger <danf at stanford.edu> wrote:

> Hi Mike,
> In looking at the github list of ERG predicates not covered, almost all of
> them look like instances of what Woodley described, where there is a
> lexical entry that introduces  multiple EPs (such as `here' or `now'), and
> where the Ja-En transfer rules apparently are not getting that nontrivial
> collection of EPs described exactly right in order to satisfy the
> generator's lexical lookup.

I fixed (in JaEn) a handful of them that were single outdated predicates.
In order to get the multi-EP ones I have a few ideas, but I probably won't
have time to implement them before my defense. At the least, I could make
sure JaEn has hand-written rules for the closed class ones, but I have to
then also prevent the automatic rules from doing a preemptive strike on
those predicates and thus making them unavailable to the hand-rules (maybe
by applying these hand-rules first, i.e., a pre-preemptive-strike-strike).

>   One element in particular that might have changed in recent memory is
> that the ERG pays a little more attention now to the variable properties of
> the ARG0 for the locative EP introduced by `here' and `now', so the
> generator trigger rules can (for efficiency) only introduce the
> semantically empty copula `be' when these guys are to be used
> predicatively, as in "people are here".  The Ja-En transfer rules might not
> be getting that ARG0 quite right, maybe.

I'm not doing anything with variable properties, so I might have to just
take a hit here.

>  Dan
> ------------------------------
> *From:* developers-bounces at emmtee.net <developers-bounces at emmtee.net> on
> behalf of Michael Wayne Goodman <goodmami at uw.edu>
> *Sent:* Tuesday, December 12, 2017 11:04 PM
> *To:* Woodley Packard
> *Cc:* developers
> *Subject:* Re: [developers] What does it mean in ACE for a predicate to
> be 'covered' in generation?
> Thanks Woodley,
> On Tue, Dec 12, 2017 at 9:43 PM, Woodley Packard <sweaglesw at sweaglesw.org>
> wrote:
> Hi Mike,
> An EP is covered if a grammar entity (lexeme or rule) can be found that
> supplies that EP while not introducing any other EPs that can't be
> reconciled to the input semantics.  So possibly what you are observing is
> that while place_n and time_n exist in the grammar (and SEM-I), if memory
> serves, every way to introduce them also introduces some other EP, e.g
> "when" is which_q time_n I think.  In this situation, if which_q were not
> in the input then "when" would not be activated, so time_n would risk being
> not covered (although there are other combinations that can cover it too).
> Might this be consistent with your experience?
> Yes, that makes sense, and mostly fits what I'm seeing. Some other
> predicates that ACE says aren't covered (I have a partial list of things
> that are "not covered" here: https://github.com/delph-in/JaEn/issues/3)
> are the ones that go along with, e.g., time_n or place_n, such as _soon_p
> and _here_a_1, so I'll have to figure out why (i'm assuming) my
> automatically extracted transfer rules are not putting them together in a
> way that ACE+ERG can make sense of.
> Woodley
> > On Dec 12, 2017, at 7:21 PM, Michael Wayne Goodman <goodmami at uw.edu>
> wrote:
> >
> > Hi Developers,
> >
> > I'm trying to generate with the ERG and ACE from semantics transfered
> from Jacy via JaEn. I'm finding that the JaEn grammar has gone stale in
> parts compared to the predicates used in the current Jacy and ERG (e.g.,
> coord vs coord_c in Jacy, and _good_a_at-for_rel vs _good_a_at-for-of_rel
> in the ERG). I'm updating those that are simple (like the above). Some
> predicates, however, appear to be valid already, but ACE still gives me
> messages like this:
> >
> >     WARNING: EP 'time_n' is not covered
> >     NOTE: EP 'time_n' is unknown in the semantic index
> >
> > time_n does in fact exist in the SEM-I (in etc/abstract.smi), although
> it is also on the "block" list of etc/patches.lisp. Another example is
> place_n, which occurs in etc/abstract.smi and is not blocked, but I get a
> similar message from ACE.
> >
> > How does ACE determine if a predicate is covered or not?
> >
> > --
> > Michael Wayne Goodman
> > Ph.D. Candidate, UW Linguistics
> --
> Michael Wayne Goodman
> Ph.D. Candidate, UW Linguistics

Michael Wayne Goodman
Ph.D. Candidate, UW Linguistics
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20171221/a4af157f/attachment.html>

More information about the developers mailing list