Re: OT: Some thoughts abouth thinking


Jerry Avins wrote:
> Rune Allnor wrote:
>
> ...
>
> > Assumming that the main part of the "excess brain" in the
> > human is spent on treating speech and language, it would
> > mean that on the order of 0.5 kg of brain tissue is spent
> > on processing language.
> >
> > Based on that line of arguments, I find speech processing by
> > means of DSP a somewhat overwhelming task. Those "visions"
> > about an automatic translator tool included in the cell
> > phone SW seem a bit far fetched, to be diplomatic.
>
> ...
>
> How large is an orangutan's brain? Speech may assist thought, but it is
> not thought's essence. Zookeepers found an orangutan family (father,
> mother, child, and infant) out of their cage one morning and in the
> outdoor exercise yard adjacent to it. The zookeeper in charge was
> admonished for not having locked up properly. It happened again, after
> the director verified that the cage had been locked. A TV camera was set
> up to catch the culprit. It turned out to be the male orangutan*. He
> carried a wire shaped to fit his gum under his upper lip. After the park
> shut for the night, he fashioned it into a lockpick, opened the gate,
> restored its horseshoe shape, and put it back in its "pocket". No speech
> involved, but no lack of intelligence either. (Should we really cage
> such creatures?)

Thatæs the kind of stories that made insert a caveat about brain
size being a good measure for intelligence. There is more to it
than just size.

> Pretty solid theory has it that large brains evolve in part to master
> the complexities of communal interactions, including the nuances of
> deception. Dolphins provide an exotic example. General processing power
> is flexible and can be applied to activities not involved in its
> genesis. Dolphins seem to have a rich language.

The might well have. I did ponder that, too, during my insomnia.
Dolphins are rumoured to be eable to use their sonar to detect
and classify various objects in their environment. That capacity
reqires an ability to both modulate emitted sound and intepret
returns. Which in turn are the basic requirements to develop
language.

Food for thought in the sonar community.

> Hearing has been around for a long time. It was well enough established
> before speech developed to make the development useful, but our speech
> centers are now separate from hearing in general.

Well, yes. Hearing is part of the basic toolkit, at least what
mammals are concerned. I once saw an argument of the location of
the sense centra in the brain can serve as a clue to when they
were developed, the closer to the spinal cord the longer time
they have been in the brain.

In that context, the sense of smell is one of the oldest, as the
center for "smell processing" apparently is located at the joint
between the brain and the spinal cord. Which might explain why
certain scents just "kick off" a reaction of some kind, be it
perfume, rotten food or whatever.

> (The mechanics of
> speech generation entailed physical changes, also. Precious brain space
> is sacrificed to make room for a larger vocal cavity that protrudes
> upward. At about the age of two, the larynx descends, making it no
> longer possible to drink and breathe simultaneously. This further
> increases the size of the vocal cavity but also increases the risk of
> choking. Nursing infants don't need to pause for breath. None of my four
> children ever had a coughing fit while nursing.)
>
> Human speech processing seems to be divided into identifiable parts.
> 1) Sounds are classified as phonemes according to a "map" or look-up
> table. (Computer scientists weren't the first to do tokenizing.)
> 2) Phonemes are assembled to form words.
> 3) The words' meanings are checked for "reasonableness". This process
> resolves homonyms and may cause earlier steps to be repeated.
> 4) Individual meanings are assembled into concepts which both add to the
> context and depend on it. (Context influences meaning and resolves
> pronouns and elisions.)
>
> It is now possible for computers to achieve steps 1 and 2 with modest
> success. As far as I know, 3 and 4 are a long way off. Step 5 might be
> translation to another language. I believe that what seems like partial
> success in that area is illusory because it is mechanical. We will have
> good translation when text can be abstracted to concept and context
> which can then be expressed in any language "known" to the translator.

So much is based on association and context. Imagine you watch two
people meeting, and one shouts "***!" to the other. It may be two
friends expressing their joy of meeting after a long separation,
it may also be an obscenity. One just does not know from the
observed situation alone. Well, a human would be able to choose
between the above two interpretations based on body language and
intonation, but I would like to see the computer that has that
capacity.

> Jerry
>
> P.S. I attended a talk in chip design given by a foreigner who spoke
> deceptively good English. He referred several times to "resistor" in a
> context where it made no sense. With close attention, I noticed a slight
> oddity in how the 's' was pronounced in the accented second syllable.
> There seemed to be a hint of 'd' in front of it. That and more context
> cleared it up. He was saying "register", oddly pronouncing the 'g', and
> accenting the wrong syllable. For claims of machine speech recognition,
> I resistor a complaint.

I attended an in-house rehersal of a presentation that was to be
given at a conference. The talker consistently said "angel" where
she meant "angle". She was very happy about giving that presentation
in-house before going public...

Rune

.