(Up to: Brains Aren TComputers )

Originally ranted at TS::BlogComments/112027943750703490 but am mirroring it here, so I can build upon it and tie it into Fractals.

From the original article...

For decades, the cognitive and neural sciences have treated mental processes as though they involved passing discrete packets of information in a strictly feed-forward fashion from one cognitive module to the next or in a string of individuated binary symbols -- like a digital computer," said Spivey. "More recently, however, a growing number of studies, such as ours, support dynamical-systems approaches to the mind. In this model, perception and cognition are mathematically described as a continuous trajectory through a high-dimensional mental space; the neural activation patterns flow back and forth to produce nonlinear, self-organized, emergent properties -- like a biological organism."

Wow, so under all that academico-speak [*], he's basically saying that this ain't no single pipeline. Heavens. For starters, computers haven't done single-"trajectory" work for years now - if ever. Under the hood these days, the graphics work gets farmed out to a separate processor (GPU), sound is obviously it's own component, and threaded processors/parallel pipelining is pretty much the norm. If, as he says, cognitive models have been bound up in factory-production-line-style models, then I think they could at least bring in a couple of first-year geeks to explain how tech really works...

having said that, I'm going to bring fractals up again, because I love them. A fractal model of linguistic processing makes sense to me - word processing is merely one level of granularity in recognising the concepts that language is merely a shadow of. The fact that the students in the experiment described are being shown random words without context completely (well, mostly) scraps the larger contexts into which linguistic processing fits - is "designed" for, as it were. I think that when we're processing language, we're actually doing it on at least 3 different levels:

• "Contextual", i.e. what we're expecting to input based on the current "topic". If you're having a discussion on food, and suddenly the other person starts talking about tennis, then you get thrown.
• "Conversational" - what we're expecting in terms of response/conversation flow. If someone answers a question you asked 5 minutes ago in the conversation, then you'll be thrown (unless they precede it with "Going back to your question from 5 minutes ago...") This is also why you can be having a conversation with someone, ask them a question when they're expecting another one, and they'll answer the one they were expecting (e.g. where, not when something happened.)
• "Verbal", for want of a better name - this is the actual words used, and where the experiment comes in. If we have the 2 layers above, then we can define a pretty narrow (although probably still pretty large, in NLP and computer memory terms) range of expected words, relative to the all the words that we actually know. Thus, as soon as we know a very small part of a word, we can refine this set even further, very quickly. Using expected patterns (e.g. we instinctively know which words/letters are more popular, so will expect these words more than others and can jump to, or "confirm" them faster), these assumptions are a good compromise between speed, probability and accuracy. This stage is also why you can jmuble up the lettres wihtin wrods and still raed them pretty qucilky - because context is more important.

Maybe this is just another temporary "problem" with emerging from an era of industrio-scientific growth. The human race will have advanced significantly when it realises that liberally-driven emergence is always going to produce more efficiency than design stemming from a single mind. This is one thing that free markets gets right.

[*] Oh, how I'm looking forward to more of this... ;)