I was a bit … weird … as a teenager. As part of an Air Force family we moved frequently, and like most teenagers trying to distinguish themselves from their peers, I tried to use my strengths – an active intellect and an ease at working with abstractions – as a way of establishing myself in the new schools I constantly found myself in. I was the “smart kid”, the one who took to carrying around large books with titles such as “Principia Mathematica” by Bertrand Russell and Whitehead Alfred North in order to impress people with my intelligence (okay, so perhaps my social intelligence was not quite as well developed at that stage).
The danger of course in taking this particular strategy is that you actually have to read the books in question in order not to be caught out by teachers who were remarkably attuned to calling one’s bluff, so I spent much of a summer working my way through at least the first volume of the Principia. For those of you not familiar with the work, the Principia Mathematica was a book on logic, but its real value lay in the fact that the logic in question was explored not by making assertions in English but rather by attempting to construct all possible logical propositions using what would today be called symbolic logic or propositional calculus.
For instance, De Morgan’s Theorem, which states that for two propositions p and q, the negation of (p and q) is equivalent to the negation of p or the negation of q, can be stated in symbolic notation as
North and Russell attempted to do this for every possible logical statement possible – in essence trying to ground mathematics in logic. As an intellectual achievement it was truly stunning. As a resource, it was practically useless, and was, even more ironically, doomed to failure even before it began. Twenty years after the publication of the last of the three part Principia, a quiet, somewhat antisocial Austrian mathematician by the name of Kurt Gödel came along and proved that there was in fact no way that the Principia could be simultaneously consistent and complete, and to make matters worse, that there was no system of mathematics that could be completely described logically.
Gödel’s incompleteness theorems, like Heisenberg’s famous uncertainty principle from a few year’s before, rocked the world of mathematics and led to a sense of deep existential angst that the field spent decades recovering from. In essence, what Gödel asserted was that at its foundations, mathematics itself is illogical, that no model, no matter how convincing, could ever completely duplicate reality.
Of course, what I eventually discovered was that the practice of carrying around heavy books filled with expressions of predicate calculus did not in fact make people respect me; rather, that people, when faced with that which they don’t understand, tend to deprecate it, minimize it, move it as far out of their particular models of reality as they can … and then they shoot the messenger.
My dance card stayed empty through much of high school and well into my college years until I found that the love of logic actually translated pretty nicely to programming computers, and that occasionally taking a shower, wearing decent clothes, being polite and kindly (if somewhat puppy-dog innocent) and taking the occasional risk in asking out women worked far better than trying to look like a genius.
A few decades later, I have to wonder if the entire Semantic Web movement could use a few dating tips (or perhaps an entire manual). Most of the concepts embodied within such things as RDF triples are not in fact all that new – a triple is in many ways almost the archetypal propositional predicate, something that Bertrand Russell would easily have understood a hundred years ago. Indeed, advances such as the SPARQL language can be thought of as the Principia Mathematica as channeled by Ted Codd. For this reason alone, there’s a lot to hope for from the Semantic Web – the marriage of semantics and semiotics with computer processing should likely open up a fairly radical new way of understanding logical systems, should provide ways of viewing information through lenses that, like the emergence of radio astronomy, result in a vastly different perspective about how language and linguistic concepts exist and interact with one another.
Yet there are also some dangers for the practitioners of the Semantic Web. One of the first is the sin of hubris – of assuming that the semantic web framework as it is now being drawn is all-encompassing and complete, and that there are in fact no better ways of expressing this information. This is the specialists’ fallacy, and it tends to occur quite often in this field in particular.
This was driven home to me at an IW3 conference in Chiba, Japan a few years back. A number of W3C pleniary sessions were also held at the same time, and because of that it made sense to have an “open-mike” night where participants could ask questions of the various W3C members. I remember at one point where a gentleman (obviously an academic) came up to the mike and asked “It’s apparent to me that everything one can do in HTML one can do better in RDF. When is the W3C going to phase out HTML in favor of RDF?”
There was a long, drawn out pause, before one of the speakers (it may have been Liam Quinn) spoke quietly into the floating microphone on the table “Next.”
RDF as a technology has the potential to build models of incredibly high fidelity, but at the cost of a cohesive structure that can be efficiently utilized. Indeed, this is our own uncertainty principle at work – there comes a point in any model where you can only increase efficiency by decreasing flexibility and vice versa. You can in fact replace HTML with RDF, but it would be a hideously inefficient way of building web pages compared to a mechanism that’s already been optimized by nearly two decades of development, consensus and stress testing.
This hubris also manifests itself in classification systems and ontologies. Classification is a big part of linguistic semantics, and as such it should in theory be an area in which RDF/OWL works remarkably well. Yet the complexity of RDF (and even of alternate triples syntax, such as Turtle) for discussion of meaningful expressions beyond the Hello, World variety means that there is a disturbingly wide gulf between Semantic Web theorists and people that deal on a daily basis with classification systems. The conceptual framework is, in the vernacular, not user friendly.
What this means is the fact that (besides making it harder for Semanticists to get dates) people who might otherwise find the tools that RDF and OWL provide useful in their day-to-day classification needs are creating their own arbitrary schemes and mechanisms because RDF is so painful to use well. This isn’t just a matter of developing better tools, though that will help some. It is a matter of recognizing that better education needs to be done in this field, that there are whole sets of tools (such as XQuery) that have the potential to provide a significant no form factor to RDF/SPARQL manipulation, and that classification and data navigation are becoming inextricably linked.
Finally, the Semantic Web community needs to learn to be less insular as a whole. Entity extraction and content enrichment would seem a natural space for RDF (or better, RDFa), yet its surprising how few enrichment services speak to SemWeb friendly processes. There’s a similar niche within publication of syndicated feeds such as Atom where semantic web concepts would make a huge amount of sense, but again no one is speaking to the syndicators on this. Geospatial information systems would seem to be tailor made for annotative semantic content, but because of the myopia within the Semantic Web community, most of the semantics involved are GML-related, not RDF. Similar examples can be given in nearly every domain where XML is used, from health care to business reporting.
What makes this problem so damaging to the Semantic Web community is that once domain-specific semantic notation and terminology becomes entrenched, it is far more difficult for a general-purpose language to dislodge it – which means that each market that develops its own semantics represents one less market for the SemWeb community to promote standardized tools into. The Semantic Web is all about language and communication, and sometimes, in the quest to establish SemWeb as a legitimate “scientific” domain, what gets lost is that communication occurs between real live people.
So perhaps it is time for the Semantic Web to loosen up a bit, take off the white oxford shirt, dorky red bowtie and security glasses and replace them with a daring black tie on black shirt and a pair of contacts, and give that cute little Ruby number in marketing a call to set up a date. Who knows, it might even be fun?