Dear Richard,
On Wed, Sep 3, 2008 at 10:44 PM, Richard Hudson <[log in to unmask]> wrote:
> But my approach is monotonic because it always starts at the
> bottom, so instead of retracting older propositions, you never inherit those
> propositions because the younger ones always come first.
>
Actually no, our approach is deterministic: G(A) = G'(A) for any
saturated derivations G and G', but not (generally) monotonic: A < B
(B contains A) doesn't imply G(A) < G(B); unless there's a structural
constraint over A and B (which you state as starting at the bottom).
> The problem that worries me is the one you mention in your last sentence,
> where you suggest adding 'disjoint' links to relations that are mutually
> exclusive such as 'after' and 'before'. This is very similar to Lyne's
> suggestion, and I feel uncomfortable with it for the same reasons. One of my
> main objections is that this problem only arises when two relations link the
> same pair of entities, as in:
> [1: word dependent X]
> [2: word before X]
> [3: word subject Y]
> [4: word after Y]
> [5: subject is-a dependent]
> Here the conflict only arises because of [5], whereas most inheritance
> conflicts arise because one entity is-a some other entity. This is why I'm
> trying to re-frame the problem in terms of how relations (in contrast with
> entities) inherit properties. I haven't got there yet, but I'm still
> hopeful!
>
I thought that we've been talking about how relations are being
inherited based on their hierarchy (so that the "subtyping" is checked
on all three dimensions of relation triples). So that [W is-a word]
implies [W subject' Z] which overrides [W dependent' Z] (dependent' is
retracted, or if you prefer, not inferred in the first place). Then [W
after' Z] and not [W before' Z]. (Notation: for relations R, [R is-a
R'].)
The problematic situation is when (forgive me my ignorance and
artificial example):
[6: word topic V]
[7: word before V]
[8: topic is-a dependent]
If it so happens that [W topic' Z], then again both [W after' Z] and
[W before' Z]. Only now if this situation is undesirable, we have to
introduce properties for relations.
Kindest Regards, Łukasz
|