Working with web technologies, one cannot escape all the hype being built up around semantic technologies, Powerset, CYC, natural language technologies in general. We have more or less gotten used to this set of fancy new buzzword compliant companies raising obscene amounts of dollars for products, I personally believe are unachievable and are bound to dissapoint a whole generation of users.
These guys are making a dangerous bet, that they will be able to formally describe the world in a rational structure of well defined entities and relations. This sounds incredibly similar to what the structuralists were claiming 40 years ago, and failed.
These guys are smart, they recognized that top-bottom approach to building an ontology of the world will fail, so they bet everything on understanding natural language and deducing the meaning of the message with help of grammar as a translation layer for their rational representation. Their bet is, that the natural language grammar is rational enough, that they can follow it’s structures and translate them into mashine-readable format. 100 yars ago Wittgenstein and de Saussure started a movement around premisses like these, but unlike these contemporary nerds, they knew what their objectives and limits were. Wittgenstein described formal relation beetween language and mind, Saussure practical one. None of them was prudent enough to think he could invent the sense of the world.
The sense of the message is encoded into the context surrounding the individual words, not into their grammatical relations. Or better said, there is a substantial amount of sensful information that cannot be recognized using rational analysis, because language is not a rational structure. Human beings learn to communicate with observation and repeating after other. To know a language is mainly about interpolating it from the aggregated behavior of others. The interpolation is done according to ‘rules’ deSaussure and structural linguistics recognized, and has logical limits Wittgenstein and logical positivismus described. Actual ‘grammatical’ rules are the result of each individuals constantly growing interpolation and are never fixed. The message is always ‘in-between’ words and language rules, simply because humans don’t relly on language rules in order to tranfer the message, but rather adapt to current receiver’s language code.
Anyway, main reason for writing this article was seeing Powerset demo, with handpicked use cases, none of which worked. And some computer scientists around me simply wouldn’t accept the fact, that philosophy has already tried and abandoned the hope of rationally sorting out the world. And the only true power of computers lies in the abbility to crunch numbers at a scale so large, that the practical statistical analisys starts to give usefull results. Statistics is just like interpolation, and preserves the human approach to content; while logical deduction simply doesn’t have the infromations avaliable, on which it could reliably run. So it tends to take wrong turns and end up on entirel wrong part of the tree.