Saturday, May 28, 2005

Digitality, Goodman, and Greece

I've spent a few days in Greece with a good friend of mine, and while on the airplane finished re-reading Goodman, producing some notes for perusal by Henry. For those who are interested they are here. The most interesting thing about Goodman is his notion of a "notation" system and his strict separation from semantics and syntax, and he defines a notation as something where every syntactic marking has a distinct semantic marking, much like either music or computer programs (Goodman being more interested in music than computer programs). However, both fail on a level as notation systems, since the syntax of both can be ambiguous. While his general concept of notational system may be simply too fanciful for any real system, the ideas he propounds are wonderfully suited to an analysis of the Web and representation in general.

His idea of notation is fascinating. A notation is a symbolic language where every syntactic marking has a distinct semantic class, much like either music or computer programs (Goodman being more interested in music than computer programs). It is also a fairly sensible idea that has never been carried out in practice. Even in music and computer programming, one of the few fields where syntax has a clear incarnation in the physical world (this quarter note means play this, this piece of code means move this to a register, write something in standard output!), there is still syntactic forms of code with the same semantics (for and while loops) and in music (two half-note rests and a full rest). However, I think his ideas do apply very well to digitality.


The Weather in Thessaloniki

Saturday, May 07, 2005

W3C Rules Workshop

Just spent the last week and a half in the United States at the W3C Rule Workshop, and had a great time. My input was a position paper about data integration, in which I make the elementary yet important observation that people are using XML primarily as a syntax for data transfer - and the natural corollary for data transfer is data integration. To do data integration, people need a coherent sense of global identity, and the Semantic Web (via its use of URIs) is one good attempt at providing it - although I am the first to point out the philosophical problems. However, to determine if two URIs can be merged or related, we need rules - after all, how else are we to determine the criteria?



Tim Berners-Lee rapping on the Semantic Web


The W3C Rule Workshop was the usual odd mixture of academics, W3C Web heads, and members of the business community. Large companies such as Oracle and iLOG seemed very pleased at a W3C standard for rules, and were somewhat surprisingly interested in the Semantic Web There was two main poles: those such as KIF and the gigantic (and perhaps overly complex but nonetheless impressive) RuleML effort, who were using XML to encode first-order logic and slight variants, and efforts such as SWRL and N3 who are trying to build a rule-language directly on top of OWL and RDF. The main points of confusion and debate seemed to be if RDF should just be an encoding format for rules, or should somehow the formal semantics of FOL could be grafted on top of RDF just as OWL-DL grafted description logics on RDF. Tim Berners-Lee noted I was "making trouble up in the front" on these points, but I'm not clear exactly what is the best path forward. I prefer the latter, but am unclear if it is possible. And does the Web really need Prolog 2.0 with URIs? I actually think the Semantic Web needs a programming language in the style of logical-functional programming, and am hoping this is a step in the right direction.



Pat Hayes presenting KIF


The second debate centered around negation-as-failure (to negate a proposition we try to prove it via executing the proposition, and if it is proved, then its negation fails. Conversely, if the proposition fails during execution, it's negation will succeed). TimBL believes that since the Web is essentially open, and negation-as-failure works operates via a closed world assumption, that the Web should not use negation-as-failure. However, the academic community and anyone who actually implements rule engines knows that negation-as-failure is useful, since although its empiricism violates the a priori concept of negation, it is unclear how else one could possibly implement negation. What TimBL needs to realize is that every inference secretly encodes the closed-world assumption, since every logical operation is operating with a limited knowledge-base. Only in non-temporal monotonic reasoning without negation the closed-world assumption does not practically matter. With negation, it matters. What is more important than the "closed vs. open world" is that each proof should carry via its provenance its full proof and links to the knowledge-base that made it possible. While more ambitious than named graphs, with FOL for the SW this is actually possible.

Lastly, had dinner with Pat Hayes and discussed how a more example-driven approach to solving httpRange-14 would work and could clarify the use owl:import.

Also submitted the infamous Web Proper Names proposal to ISWC 2005, and had such a great time at this W3C workshop that I submitted another position paper to the next workshop on semantics for Web Services.

Managed to sneak in a quick trip to visit my family as well in North Carolina as well, and sorted out various horrific visa issues with the embassy.



My brother, sister and I