[developers] More ACE/LKB divergences (underspecified generation)

[Francis Bond]

G'day,

another difference between ACE and the LKB is in the generation of
underspecified properties.   In this case, we make substantial use of the
LKB's behaviour, and would prefer ACE to behave in the same way (possibly
with an optional choice).

The question is one of what happens when you specify a type in the middle
of a hierarchy.

Consider the sense 'PRON sleeps'
If  PRON has PERS: 3 NUM: sg and no gender, LKB and ACE both generate
He sleeps
She sleeps
It sleeps
They sleep.  [androgynous they]


If we restrict the gender to m-or-f [PERS: 3 NUM: sg GEN: m-or-f]

Then the LKB generates everything subsumed by this:
He sleeps
She sleeps
They sleep

ACE generates only things with exactly the same semantics:
They sleep
(i actually generates the others then rejects them with the heart-warming
message:
NICE TRY! `She sleeps.' has incompatible semantics:
NICE TRY! `He sleeps.' has incompatible semantics:

When translating from Japanese to English, we often have partial
information about some property, and would like to specify to generate that
and *get everything subsumed by it*.   Not being able to do this is greatly
reducing the coverage of JAEN using ACE: we often just set NUM to num, and
this results in nothing being generated (as the number is always specified
to something lower).

I am happy to also specifically submit this as an ACE feature request, but
am trying to send things to developers by default.

In the deep dark past Woodley said:
> The question of how to filter generation results in a useful way, though,
is perhaps still an open one.  Strictly checking for subsumption by the
input MRS is the right thing in most situations, I think.  The LKB has long
implemented the option (which you presumably have been using) to normally
only output the subsumed subset, but output all the results if none of them
are subsumed.  While a handy hack, this strikes me as a unprincipled.  Some
kind of ranking of the outputs by how severely they fail to be subsumed by
the input would be better (sounds a little like optimality theory).

It seems to me that currently ACE is checking for equality, not subsumption.


See also: http://moin.delph-in.net/SaarlandMrsSubsumptionDiscussion

-- 
Francis Bond <http://www3.ntu.edu.sg/home/fcbond/>
Division of Linguistics and Multilingual Studies
Nanyang Technological University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20150305/90abf7ec/attachment.html>