Re: Southampton Pub data as linked open data

On Jul 30, 2008, at 8:44 AM, Chris Bizer wrote:

> Hi Bijan and Richard,
>
> I think it would be helpful for this discussion to distinguish a  
> bit between the different use cases of Semantic Web technologies.
>
> Looking back at the developments over the last years, I think there  
> are two general types of use cases:

I really don't accept this analytical framework. It has more than a  
bit of the just so story about it.

> 1. Sophisticated, reasoning-focused applications which use an  
> expressive ontology language and which require sound formal  
> semantics and consistent ontologies in order to deliver their  
> benefits to the user. In order to keep things consistent, these  
> applications usually only work with data from a small set of data  
> sources. In order to be able to apply sophisticated reasoning  
> mechanisms, these applications usually only work with small datasets.

The last two are just not true and very unhelpful. Can we stop there?

(Consider the general work on tractable fragments. Consider the SHER  
work.)

> 2. The general open Web use case where many information providers  
> use Semantic Web technologies to publish and interlink structured  
> data on the Web. Within this use case, the benefits for the user  
> mainly come from the large amounts of Web-accessible data and the  
> ability to discover related information from other data sources by  
> following RDF links.

I don't think I believe this either. And least not in the substantive  
claims that follow.

(Check out the Tambis and the DL Lite papers. Tambis is *exactly*  
about using ontologies to provide better *user* experience of  
navigating over multiple, independent data sources. DL Lite is  
*designed* to handle the data integration case wherein you can leave  
your data in the original RDBMSs and yet have a view that is more  
"conceptual" (and intuitive) which also allows for proper distributed  
query.)

> For each type of the use cases, there is usually a different set of  
> technologies applied. OWL and classic heavy-weight reasoning for  
> the first use case. HTTP, RDF, RDFS and light-weight smushing  
> techniques for the second use case.

A more useful, imho, categorization is that there are several  
different sorts of task, e.g.,:

1) determining what to present (and how)
2) modeling
3) determining how to navigate
4) mapping and otherwise meshing

All of these occur in ontology land. Indeed, 2 is usually *in  
service* of the other ones. (Please check out Alan's slides I  
referenced earlier.)

Forget OWL exists. I think I would *still* have a problem with the  
merge of all these roles that you see with things like subpropertying  
rdfs:label.

> In the first use case, people think in terms of "ontologies", for  
> instance a basic concept in OWL2 are ontologies. In the second use  
> case, classes and properties are mixed from different vocabularies  
> as people see fit and are related to each other by RDF links.

I don't see how the latter follows from anything about the second use  
case. At least not in a non-tendentious manner.

> The second use case is inspired by the Web 2.0 movement and aims at  
> extending the web with a
> data commons into which *many* people publish data.

Many people publish Web data without understanding or even knowing  
the existence of HTML. Most people who *publish* data in the world (a  
lot of people!) having considerable data modeling and data format  
skills. I really don't see that RDF is so much better off than many  
of these (having worked with very non-data-literate people).

> As it is not very likely that all these people will be logicians  
> and understand (or are interested in) the formal semantics behind  
> the things they do,

Most people writing OWL ontologies are not logicians. It's kind of  
hard to take your analysis too seriously if you elide this basic fact.

> people (including me) working on the second use case are often a  
> bit critical about too tight formal semantics and extended public  
> discussions about minor details that arise from some specs.

Funny, I've been trying over and over to move this discussion into  
"What would be a more useful presentation mechanism" or "What are the  
real requirements of a labeling system". Why not engage *that*?

> These discussions have been a mayor obstacle to deploying the  
> Semantic Web over the last years as they drive away people away  
> from using the technologies.
[snip]

*Really*. Are you kidding? On what evidence do you base this?

More of an obstacle than "RDF is magic and so much better than XML  
because it does magic merging" lines? Do you follow the HTML world  
and how it views RDF (based on, for example, the RSS and Mozilla  
experiences)?

I mean, a good chunk of the Web world spends its time arguing about  
the semantics of elements in fussy detail. (When do you use abbr?  
What is an alt text exactly?)

So, sorry, I've experienced your line as a *constant drumbeat* since  
I started working on the semantic web and it just seems to be a "shut  
up and don't think" line rather than a serious empirical analysis.  
There's nothing novel about it, afaict, nor is there any substantive,  
grounded content to it, afaict.

Note this *does not mean* I am, in fact, the straw man people seem to  
want me to be. I'm *not* arguing that OWL or the logical core of OWL  
is the be all and end all of *anything*. I try hard to find specific  
descriptions of when OWL is a large win for its users and try to  
steer people away from OWL (or RDF or XML) when it's not right for them.

I've been spending a *huge* amount of time trying to get decent  
annotations into OWL and do you know what the biggest, the single  
biggest, barrier is? Serializing them into RDF. Just trying to get  
annotations on annotations (of arbitrary depth) is a real nightmare  
(yet trivial in XML :().

Sigh. Perhaps it's time to unsubscribe to this list. The reflexive  
OWL bashing is getting to me. As is the implicit  and sometimes  
explicit) arguments from I'm-in-the-cool-group-and-you're-not. I  
don't know why it's so hard to grasp that someone can be both a  
logician *AND* seriously engaged in human factors, but here I am.

Or not. If you'd like further replies from me, please send to my  
address directly.

Cheers,
Bijan.

Received on Wednesday, 30 July 2008 08:54:35 UTC