Re: One cheer for XML ...

From: Marshall <>
Date: 29 Mar 2007 11:33:13 -0700
Message-ID: <>

On Mar 29, 9:50 am, "DBMS_Plumber" <> wrote:
> On Mar 29, 10:35 am, "Marshall" <> wrote:
> All of which is true, and raises interesting questions. How can we
> integrate the markup tags with the rest of the schema? Indeed, should
> we? Many of the 'judgements' made by these annotators are
> probabilistic, while SQL (indeed relational) semantics are (naively
> expressed) boolean. Can we bridge that gap?

The answer is not at all clear.

My job for the past two years has been working on a system with exactly this sort of "probabilistic" schema thingey. It's quite challenging, and I don't believe anyone has cracked it yet, if indeed there is any "there" there after all.

One possible interpretation of the situation is that using tags without an established schema is simply deferring the job of modeling, which will necessarily have to be done eventually. And it may be the case that doing it later is necessarily more work than doing it up front.

Even if this is true, there maybe applications where that's the right tradeoff. OTOH maybe making schema evolution easier, and making client schemas less coupled to server schemas is the better way.

Also, there may be applications where probabilistic approximations are good enough, and less effort. Not banking, for sure, but maybe information retrieval.

Marshall Received on Thu Mar 29 2007 - 20:33:13 CEST

Original text of this message