Re: meta information

Daniel W. Connolly (connolly@hal.com)
Thu, 02 Jun 1994 12:46:56 -0500


In message <199406021656.JAA11931@nova.unix.portal.com>, Nick Arnett/Multimedia
Computing Corp. writes:
>
>Hmm. I'm already doing that kind of tagging -- creating links to
>background documents and such, using URLs. I'm interested in creating tags
>that *aren't* links, for the kinds of purposes that META is also intended
>for -- automatic creation of new indexes, etc. In other words, not
>anchors, just semantic tags.
>
>This would allow, for example, an investment company to have a proprietary
>file on Microsoft (don't we all?) that's automatically linked to incoming
>tagged news articles. More important, the system that processes incoming
>articles could see that a company is mentioned for which there is no
>backgrounder, which would trigger a search of public sources.

(Have you seen Topic from Verity? It's a nice engine for doing
just this sort of thing... it's natural-language based.)

>In any event, I *really* would like to see explicit support for this sort
>of thing in HTML.

Why? It seems like a separate data format might work better for
your app. A client could get and display the HTML, but it could
also ask for the document in application/x-people-places-things form,
(be sure to check with Apple before using that name... :-)
which might look like:

(PERSONS (GATES WILLIAM)
(CLINTON WILLIAM)
)
(CORPORATIONS MICROSOFT)
(COUNTRIES USA)

>BTW, one of the things I might stick into META elements would be lists of
>all of the companies mentioned in an article, all of the people, all of the
>products, all of the fizzbins, etc. Getting that sort of information --
>the names of the META elements and the lists of items in them -- in
>response to the HEAD request would be of great benefit to navigation
>automation.

I can imagine a set of navigation mechanisms based on this sort
of information. I don't see that HTML is the most convenient
representation of the information, though.

If you've got a specific way to attack the resource discovery problem
with these semantic tags, I'd be very interested.

But if you just want a general mechanism to express machine-readable
semantic information, you're barking up a non-existent tree. From
what I've learned from the knowledge-representation researchers,
the best way to exchange arbitrary semantic information between
domains is to write it out in the natural language. Then the
domain-specific parser extracts the parts it's interested in.

Dan