W3C home > Mailing lists > Public > public-semweb-lifesci@w3.org > March 2009

Re: Weak semantic links [was: semantic dissonance in uniprot]

From: Pat Hayes <phayes@ihmc.us>
Date: Mon, 30 Mar 2009 14:50:40 -0500
Cc: w3c semweb HCLS <public-semweb-lifesci@w3.org>
Message-Id: <CCC54FD9-CB6C-4D2E-8BAF-E960BB0C8318@ihmc.us>
To: "John F. Madden MD, PhD" <john.madden@duke.edu>

On Mar 28, 2009, at 1:54 PM, John F. Madden MD, PhD wrote:

> Pat,
>> I think its fine for the SWeb to include 'weak' semantic links that  
>> don't (yet?) have tight definitions that can support machine  
>> inference, but still convey useful information to users and maybe  
>> even tool developers. (I know saying this runs the risk of opening  
>> the old 'social meaning' can of worms, but those worms aren't going  
>> to go away :-)
> If you're willing to run with this a bit, I'd be curious about your  
> thoughts. Turns out, this theme has been popping up in HCLS a fair  
> amount recently.

And elsewhere. A few of us spent a whole day discussing it intensely  
in Cambridge recently. Increased bandwidth does not however lead to  
greater mutual illumination.

> Sometimes, it arises in the context of knowledge-capture, viz.,  how  
> do you get experts who are not ontology-savvy to disgorge their  
> knowledge in an ontologically useful way? Here, you might resort to  
> loose semantics because the skill of the modelers does not support  
> the precision that is desired.  In other words, the knowledge is  
> actually there and could in principle could be modeled more  
> elaborately, but demanding precision yields diminishing returns  
> because amateurish modeling errors proliferate.

Yes, well, you can't expect to get professional results for a whole  
ontology this way. But 'loose identity' seems to be a very useful  
special case (or maybe cases)
> Other times it arises in the context of garden-variety uncertainty:  
> there is no ripe knowledge to be harvested, just a bunch of hunches  
> and intuitions. But for a particular community, these hunches and  
> intuitions might have value. (Usually, they don't have value--and  
> may even be toxic, in the sense of "large doses may kill you"--for  
> the world-at-large, because intuitions by their nature rely heavily  
> on background/contextual understanding.

So, drawing those out is the best way to proceed. Mutual  
incomprehension is one way to at least show up such background/ 
unstated/contextual issues.

  guess, is one aspect of 'social meaning'.)
> Anyway, the former problem (seems to me) is a human-engineering  
> problem. If we could figure out cleverer, more assistive modeling  
> tools and better educational techniques for ontology training, we  
> might be able to fix it.

Yes, this is what people here call the 'knowledge extraction' game,  
getting people experts to say their expertise in a form that it  
beginning to be formalizable.  We have local experts at this stuff,  
they are all psychologists rather than engineers. It is not a fully  
automatable skill at present, but they do use software tools to help  
them. Also it is _highly_ interactive, done best with teams.

> The second problem, I think, could only be helped by keeping these  
> assertions "inside-the-fence" of the community that had any use for  
> them. But this is a problem, because the semantic web isn't supposed  
> to have any fences, i.e. "anybody can say anything about anything".  
> What to do?

I don't think we need fences around the terminology, just its informal  
semantics. But that is easy. So if some community is using  
sameProteinAs, and they tell me that it has no formal meaning but they  
find it useful, then what's the harm in that? If I can grok their  
(informal) semantics, I can maybe use it in ways that they will find  
useful, which is fine. We still get the (not especially semantic, but  
Webbish) advantages of IRIs being globally unique and providing access  
to source documents, etc..

> I've been hoping that Named Graphs would solve this problem. I'm  
> curious if you think NG's can support this use case of segregating  
> potentially toxic 'knowledge' on the SW.

It would allow one ontology to say explicilty that some other ontology  
isn't for public use, or something similar. But in the case we are  
talking about, surely we wouldn't have a genuine ontology to say such  
bad things about, right?


> The other solution I can think of is the solution that enterprises  
> use for privacy: set up private webs, intranets, yadda yadda. Which  
> seems scary.
> John

IHMC                                     (850)434 8903 or (650)494 3973
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes
Received on Monday, 30 March 2009 19:51:25 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 17:20:41 UTC