Language processing as pattern matching

Post Reply New Topic

After further consideration, I can see that this could be a useful model but it seems to austere an abstraction. I have trouble accepting the human mind is a state machine, let alone one that the next state is only dependent on the current state and none of the previous ones. I suppose if time is quantized, we would perhaps be state machines. Of course that invokes quantum mechanics and unless you observe the state you can't really determine what it is. Which begs the question, is human communication a series of state observations?

Sorry, I went a bit far afield.

Wavefreak: the beauty of Hidden Markov Models is that they don't concern themselves with micro details. The only thing the model cares about is that if the system is in State A, there is X probability of transitioning to State B, X probability of transitioning to State C, and so on.

In the example of the weather, for instance, one need not go into the meteorological detail, but instead consider which meteorological states are more or less likely to follow each other.

Now take some rough emotional states such as anger, arousal, joy, and sadness. For each of them, you can probably associate which are more or less likely to immediately follow it.

Is it likely that a person who has been expressing sadness, will next express arousal? Is it less likely they will continue to show sadness, than that they will start expressing anger? One can indeed estimate these probabilities, without needing to know the exact neurological processes involved.

Since the model need only consider states and the probability of transition between a given state and another, it does not depend on internal mechanisms.

The reason why hidden markov models have been studied so much with regard to language recognition is precisely their ability to work with black box mechanisms. So long as the black box produces output which can be probabilistically modelled, and has internal states which can be recognized as different, an HMM can aid in understanding its behaviour.

Does this model assume discrete states? I think it is a very interesting model, but doesn't it assume the current state can be completely defined and that there is also a set of discrete next states, that while there is a probabilistic function that maps to those states, each is also completely defined? So using this to model language might have great utility, but it begins to break down when the symbol "dog" doesn't even map to the same idea in the same individual twice in a row. Further, I would suspect that there is no state in the human mind ever visited more than once.

I wish I had time to research this. I really need to go back to school.

Wavefreak: This is an interesting point.

If you're interested in seeing how people are working on modelling cognitive states using such models, try to find a copy of Peter Norvig's Artificial Intelligence: A Modern Approach.

This qualifies as a special interest that I have no time to indulge in. It is extraordinarily frustrating. Even this little conversation makes me get cranky because I can't pursue it. I've never been able to do things in small doses. If I started REALLY running this down it would become all consuming.

Sigh ...

Sorry that it's so hard to explain.

Part of it, for me, is a translation problem. I'm sure that other people could have different ways of feigning comprehension, but in my case it was going so much off of a method of cognition I can barely write about at the best of times, that it's extraordinarily hard for me to describe how I do it. I know that with some people I can kind of describe it vaguely and they'll go "oh yeah THAT!" But I can't do a full description. I'm not even sure it's possible to come up with a model of how I did it because the form of cognition that I used to do it defies concepts in general let alone a whole system of concepts.

Don't apologize on my behalf. It doesn't surprise me at all that you find it hard to describe. Basically you are trying to describe using words a way of using words that can't be explained with words because if you could explain it with words the experience would be entirely different and we wouldn't be talking about it.



Or something like that.

For some reason that reminds me of a quote from this article:

http://replay.waybackmachine.org/200412 ... /snail.htm

(It's from Jim Sinclair's page on being intersexed, not xyr autism stuff. As background, it's a fictional conversation between Jim and a snail named Fibonacci.)

""You mean the people are named after their sex organs?" Fibonacci found that very amusing. Then xe had a startling thought. "Wait a minute. Do you mean that humans not only don't have all of their parts, but they don't even all have the same parts of the parts that they don't have all of?" Fibonacci tends to talk in spirals when xe gets excited."

Jim's writing is excellent. I keep forgetting where to find it on the archive. (bookmarking)