In defense of semantic value (a 2-in-1 rant) | Hacker News Reader

In defense of semantic value (a 2-in-1 rant)

onderhond.com

16 pointsonderhond8 comments

My objection to "the semantic web" is:

   1. Semantics only have meaning in a given context.
   2. "Global context" is an oxymoron.

And most of the song-and-dance routine about how wonderfully fantastic the semantic web will be implicitly assumes the creation of a global context, in which we all reasonably precisely agree on all the semantics of everything, and that's total pie-in-the-sky.

Recommended reading (which has the advantage of being a war story and not an exercise in theoretical handwaving): http://reprog.wordpress.com/2011/04/19/semantic-mapping-is-h... Also, follow those first two links down too ("the MARC format" and "Dublin Core") for full comprehension. You can't even get agreement on being what an author is between two contexts, and you think you're going to build a glorious semantic web‽ The semantic issues aren't going to get easier after that attribute....

Great comment.

I worked in the library industry for a very short while. An industry whose specialists are all trained metadata experts can't figure this stuff out. Of course, that is also part of the problem, but it isn't going to get easier in industries where people don't understand metadata.

I lost interest when the OP hand-waved managing the metadata as just a trivial practical concern that somebody would surely magic into existence. These problems are why companies like Attensity[0] exist today.

[0] http://www.attensity.com/

This is a red herring. The shared context of two documents is the schema that they both declare and nothing more. The promise of linked data is that a lot of the relations will (and does now for example in music) allow discovery of interesting relations.

As for library science, yes, it is a hard problem.

If declaring a schema meant that the two documents actually shared a schema, rather than just claiming they shared a schema, I wouldn't have had to post my comment, because the problem would have been long since solved and we wouldn't be awaiting the Semantic Web because it would have been here for about 10 years now.

Semantics is about more than slathering tags and attributes on things; semantics is about actual, factual agreement. Merely claimed agreement isn't enough when it's generally not true. If professionals in library sciences can't get it right, the closest thing to a discipline devoted to the study of metadata, who can or will?

I don't think it's just an issue of people wanting a "now" solution as the article suggests. Whenever I spend time with semantic web stuff it always reeks of the problematic "over-engineered solution in search of a problem". For instance OWL starts off by defining a computationally intractable model for reasoning. That would be fine if you wanted to use it as a framework for abstractly reasoning about a subject, but no one trying to solve an actual problem starts this way.

The other major warning sign (and indication of over-engineering on it's way) is a complete lack of historical context. For example rdf triples are a more verbose, and yet limited form of Prolog clauses (in a similar way that XML is essentially verbose, restricted s-exps), yet, in my experience, the vast majority of rdf 'experts' have never written a single line of Prolog, or are even loosely familiar with the idea of logic programming. Additionally if you were really trying to solve a problem you'd already have a working prototype kludged together with something that looked a lot like Prolog.

The combination of a general ignorance of the problem space (knowledge representation and reasoning, logic programming etc) and a poorly defined problem imho can lead to absolutely no good, and are pretty strong evidence that most semantic web project will go nowhere.

I agree completely with the second rant, but I don't really get what's difficult about marking up the various tidbits of data. Most websites use HTML templates; it's just a matter of adding certain classes to the elements.

It's also funny to notice the lack of semantic metadata in that page - you can't tell me that adding rel=me and rel=author to the links, as Scott Hanselman describes in his blog[1], is difficult.

[1]: http://www.hanselman.com/blog/EmbraceAuthorshipTheImportance...

Actually, the meta data you're talking about is there. You can even test the page to see that it should work (though it never appears like that in Google search results for me): http://www.google.com/webmasters/tools/richsnippets?url=http...

As for complexity, some classes are hard to grasp without a good reference list (like wtf is FN (formatted name)? I'm not saying it's rocket science, but it's still a lot more work (front-end and back-end) than adding class="contact person" to your base tag.

Hmm, sorry. I looked at the source and missed it.

Reply on news.ycombinator.com