engineering vs. science

[max...@umiacs.umd.edu]

For some reason, I can post a new msg, but I can't reply to an
existing one; it keeps asking me to join the group.

So I'm starting a new posting, but it's in response to an earlier msg
by Rob Freeman , which was in turn a response to one of mine, which
was in turn...you get the idea.

For some background, Rob had written:

> You can model long distance dependencies using raw word associations
> (collocations), Mike. I did an implementation of this. All you need to
> do is carry the (collocation) features through until they are needed to
> make a selection.

I called this an engineering approach. Rob responded:
> You've invented something called the "engineering approach",
> given it to me, and then proceeded to argue it does not work.

Needless to say, that's not quite how I see it. The approach is
Rob's, and I did label it an engineering approach; but far from
arguing that it did not work, I said it probably did work most of the
time.

Rob also wrote:
> I said the method I use works much like feature unification in HPSG.

The question is just how feature unification is done. If it is done
by looking to the right and left for neighboring words--which is what
I assumed Rob meant by collocation--then this is entirely different
from feature unification in HPSG, which depends on explicit
hierarchical syntactic structure, which can and often does result in
feature unification that passes over linearly intervening syntactic
nodes. So the fact that both HPSG and Rob's approach do feature
unification means very little, unless we know how that unification is
done.

Finally, Rob said:
> Can you first look at what I am proposing
> and _then_ dismiss it out of hand

Sorry, I can't until you say more about how it works. Without knowing
the algorithm for feature unification, all I have to go on is some
general statements, like "each parent node can be made to inherit the
(collocation) features of its daughters, until the relevant dependency
is encountered, and the selection is made"--but it isn't clear where
those nodes come from. If there's a syntactic pre-parse that gives
the nodes, then that's a different thing; but then this is no longer
about "raw word associations (collocations)." And if there is a
syntactic pre-parse, then I would think that my earlier point about
the separation of syntax and semantics was validated. But as I say,
it's hard to know without knowing what the algorithm is, so I'll stop
here.

Mike Maxwell

[David J Brooks]

> Don't know why it would distinguish on the basis of replies, though.

There are settings to allow "managers only", "members only" and "anyone"
to post. I have it set to "members only" until I find out a little more
about moderation.

D
--
David Brooks
Teaching Instructor
http://www.cs.bham.ac.uk/~djb