Language processing as pattern matching

Post Reply New Topic

There have been several threads where processing language as pattern matching without any true comprehension as been mentioned. I can't seem to grasp how this works. It seems to me that all language is a form of pattern matching. Basically, I learn patterns in speech, attached meaning to those patterns, and then generate a pattern that is consistent with that meaning. But maybe I'm using "pattern matching" in a different way.

A form of pattern matching that has no actual comprehension of the language is different than the above patterns. What I'm thinking is pattern matching in this context is essentially for a given a string of symbols, the pattern matching "algorithm" selects an appropriate response from another list of symbols. There is no meaning attached to the symbols, just a mapping between input and output. Sort of a column a, column b thing. So pure symbolic pattern matching would be for a given input:

Input = ABABABA

There may be on or more "acceptable" outputs

Output = CDADAD or FGSGS or ASADAF

In real language, the possible variations of appropriate answers create issues that would make me think that PURE pattern matching in this sense is impossible. If, as described, the pattern matching is good enough to passably navigate a conversation, then even something as simple as "Is the dog barking?" creates problems. Even if the appropriate response list contains only yes or no, half the time, pure pattern matching will give the wrong response. So there must be at least a much more complex system of matching patterns or at least some level of the meaning of the words.

What I am finding interesting is that my social interactions tend towards more pure pattern matching. I basically have to search for an appropriate response to basic conversational things. Once found, I apply it to the conversation, but without any connection to the underlying SOCIAL meaning or flow. I can accurately say it's raining, but it's matching the pattern "What's the weather like?" without any real understanding that the conversation flow isn't really about the weather so much as underlying social connections be forged through small talk.

Make sense?

Not symbols so much as sounds.

For me, there is no difference between words as sounds and words as symbols on the page. They are all discreet "things" that I attach meaning to. I don't do this easily in certain contexts.But the discreet vocalization "dog" is still a chunk of sensory input that becomes part of a longer series of chunks.

It also seems to me that the patterns in sounds are even more complex than those in just words. I can understand echolalia as simply repeating back a pattern. But what you are describing is beyond simple echolalia and requires some additional processing, even if actual comprehension of the patterns is present. What I don't see is how a complete disconnect between the patterns and the meanings could ever be effective on any level.

Initially I didn't learn words individually, but strings of words. I can't explain why it worked for me, and for the many other people I know with this issue, but it did.

Wavefreak58, were you hyperlexic? This can lead to things like expressive language exceeding comprehension, as you pick up a lot of words and phrases without also picking up their meanings.

My expressive language did exceed my comprehension when I was a child, and there were many times when someone would say something or something would happen that I would respond with a phrase that I knew fit the situation, without knowing what the phrase actually meant. This actually got me into trouble a few times.

I am not sure where my vocabulary was or how much I understood around me, but as I was apparently using complete sentences by 13 months, I am sure it was not typical either, and that I probably did understand at least a good amount of what I heard. I know that I spent a lot of time guessing what words meant without looking them up, and using them where seemed appropriate (or on occasion misusing them). I don't think I had the same issues that anbuend did, but I can sort of relate to them because of my own experiences.

Since it has been suggested to me that I use language in a manner similar to someone who did have those issues, however, I have been trying (and failing) to recall more than I have to work it out.

Last edited by Verdandi on 04 Apr 2011, 12:28 pm, edited 1 time in total.

It's a tough one, for sure. I can't help but wondering if your meaning was sensory in and of itself. I'm speculating of course, but the sounds that made up "is the dog barking" had (has?) a sensory "meaning" that was consistent with dogs barking or not barking, even if it was not attached to actual words. So your responses were not this pure abstract pattern matching I mentioned above so much as responding to a sensory input with a sensory output based on meaning built directly on those senses. You were actually speaking in a sensory language, not a verbal one.

There was some attachment to situation sometimes, but it was in an incredibly idiosyncratic way, and that came later, before linguistic meaning came through. But I think you underestimate the ability of the human mind (or some human minds anyway) to match patterns without meaning (and the amount that I screwed up sometimes and confused people, and the amount of time it took me to learn how to do it in a way that didn't sound like simple repeat-after-me echolalia). It doesn't have to make sense to you, it just happens. Most of the world doesn't make sense to me and I assume it gets on fine without me.

I guess I'm just comparing experiences. Your description of language more as mimicry tickles my brain as if it is reminding me of things. Especially about the way I deal with social language as opposed to technical language.



LOL. Now this makes perfect sense to me. I am always astonished about how little I truly understand.

I think I may have been hyperlexic, but I have no contact with family (for many years now) and no childhood records. But your description seems to fit. I was very wordy when young but pretty much completely shut down by third grade. Then I would talk as little as possible, except maybe at home.

Not symbols or sounds, but "idea".
Symbols are an association between memories of a particular pattern and the context under which they have been seen or used before. That does not qualify as language because we want the meaning of the symbols, and the meaning will change depending on context.

What is needed for language is a link between meaning and context, then the pattern can be matched to context. The meaning of a word is derived from the idea that it was made from. In order to understand a word, we must find out the idea that was used to make the word. This is done by following the pattern of the idiom or native language; we must generate several ideas loosely associated with each word, and selectively pick and choose the ideas that will flow in a rational manner to make them all work together, obeying the rules of grammar. Deriving meaning from words then is like a mini- evolutionary process of randomly seeing which ideas can coexists together to make all the words not be in conflict with each other, or seem rational.

The point is to see that neither the sound nor the symbol is the meaning. Language is the means that intelligent things exchange ideas. The idea that generated the word is more important that the sounds or symbols.

The pronunciation of sounds is not part of language. it's just a motor function to express language. One can just as easily express language with eye blinks, or hand signals.

For me, at least when this started, there was no "idea".

Wavefreak: I think what you are looking for, is the idea of a hidden markov model.

In a regular markov model, one is faced with a system which can be in a number of states, each transition between states dependent only on the current state.

In a hidden markov model, one is faced with a markov chain system (each state transitions according only to the probabilities associated with the current state) only the output of which is visible, each output state somehow correlated with an internal state.

Here's an example, pulled right from Wikipedia:

Alice's problem is to correlate the observed state (Bob's actions) with the hidden state (the weather).

Lots of work has been done on the analysis of hidden markov models, specifically in language recognition and communication.

Consider the analogy, now, with nonverbal communication. An aspie may identify observed states- expressions, body language, and such - but must correlate them to their most likely internal state:



http://en.wikipedia.org/wiki/Hidden_Markov_model
http://en.wikipedia.org/wiki/Viterbi_algorithm

Interesting. I'll need to cogitate on this for a bit.

You sig is tricky. I never meta mistake I didn't like.

Makes sense to me, wavefreak58, I understand language as a system of patterns, too. I could write at a university level long before I could explain what an adjective was--I would just match words to the appropriate patterns....didn't mean I consciously thought about the patterns or had any ability to talk about them. (social interaction doesn't seem to have clear patterns to me, although I've looked for them)

Part of this is related to me trying to figure out why I can be really good at some math and really bad at others. I did great in calculus. But it boiled down to learning rules. Then you just apply those rules. Such 'algorithmic' math is easy for me (I do have some memory issues, but that's another story). But as soon as I get into more abstract math, it gets a lot harder. Not incrementally harder, but hugely. I need to chew on an abstraction for an inordinately long period of time. Calculus was symbol manipulation without deep meaning.

Interesting.

I had issues with the "book learning" of English and grammar, in that I tended not to do very well in English classes, but my writing was grammatically sound even when I couldn't explain what made it grammatically sound.

Do you ever find you just learn stuff without thinking about it or trying to?