NepoMUCK? Anything ending in "MUCK" doesn't sound like a good product. The concept is very interesting but the name isn't the best I've seen.

I'm glad that they don't prefix everything with K though.

Yes, I know that Nepomuk means "Networked Environment for Personalized, Ontology-based Management of Unified Knowledge" as stated in the article.

.haeger

"Semantics" is information about meaning (whereas syntax is information about form). Semantic tools try to provide meaning by describing relationships between information atoms. The goal is to create systems which can answer questions like "how old is the president's oldest child?" with just the age, instead of listing all documents which contain the words "old" "president" "oldest" and "child".

Personally I think that metadata/tag based systems are the wrong road for semantic analysis of web pages. As soon as the semantics of a thing is decided by additional information added to describe that thing, its open to abuse.

The only advantage is its faster than what should be done, which is using good old maths to extract the true 'meaning' of a document or object.

Its not hard. Well, ok, its a little hard. Oh ok, its really rather difficult, but there are plenty of places you can get example code or libraries to make things easier.

I've got a better reason why it failed that doesn't require delving into first year philosophy.

People are lazy. Look at any image database and figure out why it's difficult to find something. Because people don't want to spend 20 minutes filling in tags for a single image they just want to show off to their friends.

Now expand that to every other form of data type, and its easy to see why the semantic web never did, and never will take off without significant AI involvement.

And I'll tell you why.

The Nepomuk Web site wants to make me chew my own arm off. Now, I'm familar with the Semantic Web, I'm excited by the idea of semantic organisation. But this site is the epitome of grim, lifeless European research-ese. It completely fails to convey the technological approach, how it works, or why you should give a damn. I get the impression that the team was more interested in the EC funding then actually developing a disruptive technology.

Why why can't researchers spend 15 minutes thinking about how to convey the importance and excitement of what they are trying to do in terms of practical examples.

I'm afraid you'll probably have to wait until some enterprising 3rd party to grab the source and build some of the technology into a different product.

Yep, that was my first thought as well. Quickly followed by wondering if 'into a collaboration environment which supports both the personal information management and the sharing and exchange across social and organizational relations' was some kind of euphemism for, eh, group pr0n of some kind.

Oh, well, either they have much less dirty minds than mine, or someone's desire for well-indexed pr0n browsing has gotten slightly out of hand.

Whoosh

Didn't you get the memo [slashdot.org]?
That should read:

"You may have Frontotemporal Dementia [wikipedia.org]. Please see your physician."

Actually, I'd say it's too early to say that the Semantic Web has failed. What has clearly failed for now is the vision for how the technology was to be used.

For one thing, it turned out that really, really clever textual matching is a lot more powerful than anybody thought possible. Twenty years or so ago, you'd have thought that you'd need to have some kind of sophisticated metadata to do the kinds of stuff we take for granted in Google today. I turns out that a technology that turns a needle in a haystack into a box of needles with some straw mixed in is pretty darned useful. Human intelligence picks the needle of meaning from the straw of superficial matches pretty effectively.

But what about non-human intelligence?

Well, here is another failure of the vision. Clearly, a semantic web is much more friendly to non-human agents. However, the whole agent philosophy of software design is extremely failure prone. A project which makes a resource easier to use for people is a safer bet than one which tries to replace human reason.

That said, you have the wrong end of the stick, philosophically. It is because meaning is not an attribute of data that we need semantic technology, It might be less contentious and pretentious if we simply call it "metadata".

If I want to find the rate of a certain disease in each county, the numerator is quite easy: I count all the instances of the disease. But the denominator turns out to be tricky, because of what I call the curious case of the dog barking in the night: some counties don't report any cases because they don't have any, others lack the technical capability to detect it.

Consider a county that can't detect the disease. I ought to exclude that county from the denominator in my rate calculations. On the other hand, a county which can detect ought to be included in the denominator, even if it reports no cases. However, since it found no cases, what we usually have is an absence of data which looks identical to the absence in counties that aren't capable.

You have to have the metadata to tell these cases apart. You have to have a model saying such and such a lab protocol is capable of detecting such and so set of infectious agents, and then you need metadata linking each data set to the appropriate model. You can do it by hand, manually discarding the data for counties you know you can't use, but this is really quite awkward when you cosider that the situation can change from year to year, or even within a year.

The model aspect presents a considerable can of worms. For any purpose, you want enough model, but no more than that. This is akin to the situation of novice designers who set out to create object frameworks before the have defined the software application. For us to share data we have to have some common model of things (although our terminology may differ). On the other hand it is certain our models disagree with each other; we want enough shared model to work together without forcing our entire model on each other, which is impractical.

The point is that you can't guess all the kinds of uses that future users as yet unknown might want to put data to, what kind of meaning they might extract from it. That's why search engine technology works so well: you put your stuff on the web and it gets spidered by Google: no guesswork needed. The Semantic Web, on the other hand, requires anticipating how the data will be used, which limits its usefulness. The "limits" here are, however ones of scope; the Semantic Web can't do everything, it certainly can't take the place of Google. Within the scope of its potential applications, it could be very useful indeed.

Semantic web is also about accessibility. Take a blind person for example surfing the web using a screen reader, do you have any idea how horrible his/her browsing experience would be like in the web today?

Click on a different page, and there you go listening to the same headers _again_. It can get very frustrating.

Without semantics there's no easy way for a screen reader (or other accessibility enabling devices) to successfully translate a document to something intelligible and usable.

You can see how a photograph (or even worse, an image with text that conveys something) can be completely hidden from a blind user without proper meta data that describe it. Or how a mildly complicated table would be read completely out of order if the reader couldn't distinguish between header rows and content rows, etc. (That's why designing using tables is a horrible idea).

The example I gave above is solved *cough*hacked*cough* today by adding two anchor links at the top of the page (skip to contents and skip to navigation) then hiding these links from regular browsers using CSS (Note: It happens to also be a valid [not hack] solution to the problem of scrolling past long navigation links on mobile devices).

I think you can see how a reader could easily identify which parts of the document are important and what should be skipped over or highlighted had it been served a semantic and valid [x]HTML document.