Default unification: online vs. offline

[Raquel Fernandez]

Hi,

Rob Malouf wrote:

> In the case where conflicting default constraints are inherited
> from orthogonal dimensions of the hierarchy, the result is the
> generalization
> of the conflicting constraints. For example, in a Nixon diamond, where
> something inherits the constraint [ F / a ] (i.e., the value of the feature
> F
> is, by default, an object of type a) from one dimension and the constraint
> [F / b ] from another orthogonal dimension, the effect will be that the
> value
> of F will be constrained to be the least upper bound of a and b. It's true
> that strictly speaking constraints can't be overridden by conflicting
> constraints from an orthogonal dimension, but the result is certainly
> well-defined and, in my opinion at least, pretty much what you want. Or is
> there some other problem you had in mind?

Ann Copestake wrote:

> wrt Raquel's point - Rob's right that YADU is fine with a
> multidimensional type hierarchy in the sense that it gives a
> well-defined result etc.  If you want to ensure that one side of a
> multiple inheritance gets preferred over the other and you have a
> well-defined notion of priority, it'd be trivial to amend the YADU
> definition to do that.  But as far as I can see, it'd have to be an
> external or stipulative notion of priority: there's nothing that you
> can sensibly compute on a hierarchy alone that'll give you an answer to
> Nixon Diamonds.

It's true that the result is well defined, although I think it is not the best
result you can get since in some cases you would loose too much
information (I'm thinking in examples likes the GHFP conflicting with some
hypothetic default constraint in the Clausality dimension).
My point was that the `problem' with this kind of hierarchies is that the type
hierarchy can not be used to define a priority order.
As Ann Copestake says,  it's necessary to stipulate an additional ordering
aside from the hierarchy in order to compute default
unification in an interesting way.

Cheers,
    Raquel


--
Raquel Fernández Rovira

NLP Group,
Dept of Computer Science
King's College London
The Strand,London WC2R 2LS,UK

phone: +44 (0)20 7848 2476
raquel at dcs.kcl.ac.uk


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/hpsg-l/attachments/20011030/d392cf37/attachment.htm>