To illustrate the rate of progress in neuroscience, I'd like to quote
two discussions about perception: (01)
1. A summary of a paper that was published in the journal _Neuron_
in January 2006. (02)
2. Three paragraphs about similar processes in Chapter 2 of my book
_Conceptual Structures_ (published late 1983, copyright 1984). (03)
If you compare the two, you may notice that there is nothing in the
2006 article that contradicts anything in my summary from 1983, which
cited psychological studies from 1960, 1967, 1970, and 1978. (04)
John Sowa
______________________________________________________________________ (05)
Source: http://www.sciencedaily.com/releases/2006/01/060125082426.htm (06)
"Vision doesn't happen in the eye," Connor said. "It happens at multiple
processing stages in the brain. We study how objects are signaled or
encoded by large populations of neurons at higher-level stages in the
object-processing part of the brain." (07)
The report, based on recordings of nerve cells in the visual cortex of
macaque monkeys, reveals that neurons in the higher-level visual cortex
at first respond to a visual stimulus "somewhat indiscriminately,"
signaling all the individual features within a shape to which they are
sensitive. For instance, a particular neuron may respond to objects with
either a concave fragment at the top or a convex fragment at the bottom.
At this point, the neural signals are ambiguous; the brain doesn't know
whether the concavity, the convexity or both are present. (08)
Milliseconds later, however, neurons begin to react exclusively to
combinations of shape fragments, rather than to individual fragments. In
other words, the brain begins to put the pieces together to form larger
sections, in the same way that an artisan might fasten discrete shards
of stained glass to create a design. (09)
"Humans do a rough categorization of objects very quickly," Connor said.
"For instance, in just a tenth of a second, we can recognize whether
something we see is an animal or not. Our results show that this
immediate, rough impression probably depends on recognizing just one or
more individual parts of what we see. Fine discriminations – such as
recognizing individual faces – take longer to happen, and our study
suggests that this delay depends upon emerging signals for combinations
of shape fragments. In a sense, the brain has to construct an internal
representation of an object from disparate pieces." (010)
______________________________________________________________________ (011)
Source: _Conceptual Structures: Information Processing in Mind and
Machine_, by John F. Sowa, Addison-Wesley, Reading, MA, 1984. (012)
From Chapter 2 "Psychological Evidence" pp. 29-30: (013)
Phonemes form syllables, syllables form words, and words form phrases
and sentences. An important question is whether perception is a
bottom-up process that first matches smaller units and combines them
into larger ones or whether it is a top-down process that first matches
large units and then fills in smaller details. Gestalt psychology is the
classic example of a top-down theory. Figure 2.1 is a schematic face
that illustrates the Gestalt phenomena. When reading the characters V
and O in a line of text, a person normally interprets them as letters.
But in the context of a face, they are interpreted as a nose and eyes.
Even caricatures with just a few obvious features are easily recognized.
When New York magazine printed a picture of an eggplant with pendulous
jowls and an extra lobe jutting out for a nose, it was instantly
recognized as Richard Nixon. Perceptual mechanisms find a schema that
fits the overall stimulus before they analyze finer details. (014)
In the context of a word, a reader does not normally see letters as
separate units. Especially in speed reading, letters pass before the
eyes at a faster rate than people can recognize them. Konorski (1967)
found some patients with brain lesions who could read words, but they
could not analyze words into letters. One patient "was able to read
known words fairly well. However, when he was asked to spell them, he
failed completely. He was also totally unable to read nonsensical words.
When a word was presented to him in which one letter was changed, he
usually did not notice the difference and read the word as if it were
written correctly. For example, seeing the word okilary he read it as
okulary (in Polish, spectacles) not being able to discover that one
letter was changed. When more letters were changed, he failed completely
to read the word" (p. 121). Konorski concluded that these patients saw
the words as unitary percepts, not as composite structures. (015)
Studies of the speed of perception also support top-down theories. When
a picture of a man is briefly flashed on a screen, a person first
identifies the general form and later notices the details of face,
hands, shirt, and tie. Bever (1970) found that people could recognize
spoken syllables faster than they could recognize single phonemes: when
subjects were told to listen for a word starting with the syllable bo-,
they responded 70 milliseconds faster than when they were listening for
a word beginning with the phoneme /b/. Furthermore, people can recognize
sentences faster than single words: subjects responded consistently
faster when they were listening for the sentence "Boys like girls" than
when they were listening for the first sentence beginning with the word
boys. For recognizing large vocabularies by computer, White (1978)
proposed a top-down method of matching templates for complete words
instead of first trying to recognize individual phonemes. He estimated
that the processing time would be reduced by a factor of 10 to 100. (016)
Multiple levels of percepts can help a person deal with novelty. A
common phrase like United States may be matched by a single percept, but
a less common phrase like United Fruit may require two separate
percepts. For an unfamiliar word like antidisestablishmentarianism,
perception may carry the analysis down to single syllables or groups of
syllables like anti-. And for a rare syllable like Omsk, it may go down
to the level of individual phonemes. The mechanisms for handling
percepts are innate, but the particular stock of percepts must be
learned. Some percepts are learned by rote memory, but the most
productive source is the construction of compound percepts from smaller
ones. Even though adults read a complete word or phrase as one unit,
children learn to build up words from combinations of letters. (017)
Some evidence favors a top-down approach, and other evidence favors a
bottom-up approach. In their work on stabilized images, Pritchard et al.
(1960) found evidence both for the stability of the whole, as Gestalt
theory maintains, and for the existence of parts for building the whole.
They concluded that both approaches "are valid and complement one
another." This psychological conclusion has also emerged as a basic
principle of AI: top-down reasoning is a goal-directed process that
imposes a tightly controlled organization; bottom-up reasoning is a
data-directed or stimulus-directed process that leads to more diffuse
chains of associations. The two approaches may be combined in
bidirectional reasoning, which is originally triggered by some stimulus
in the data, but which then invokes a high-level goal that controls the
rest of the process. Some problems are more naturally handled by one
approach or the other, but the combination of the two is especially
powerful. (018)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (019)
|