[From Bruce Nevin (980415.1202 EDT)]
I've been thinking about associative memory. It has seemed to be orthogonal
to the control hierarchy, not explicitly included in the present HPCT
model. I think perhaps some aspects of memory--the associative aspects, and
to some degree addressing and retrieval--are already modelled in HPCT. What
is not modelled is storage itself.
In B:CP (p. 287) remembering is defined as "replay of stored perceptual
signals as present-time perceptual signals, in combinations that actually
occurred at some time in the past." Imagining is defined identically,
except that the perceptual signals are replayed in combinations that never
occurred in the past. More recent research has showed that memories are
constructs and not records of events. This would blur the distinction that
is made here between memory and imagination; or, rather, compels us to find
some additional basis for distinguishing memory from imagination, other
than the supposed fidelity of memory to "actual events". I don't have any
specific proposals about this at present, I'm just pointing it out.
There are two aspects of memory, then: storage and association. Some have
proposed that memory is stored throughout the body and recovered through
specific bodily sensations plus associative processes similar to
imagination. I don't have anything to say about storage, addressing, and
retrieval.
But about association it occurs to me that there may be a plausible
mechanism in the model proposed for category perception.
>----- panting
>
>----- wagging tail
>
<dog>-+----- barking
>
>----- "dog" (the word)
>
>----- tail
>
>----- 4 legs
>
>----- fur
>
>----- growling
Given some of the perceptual inputs (sound of a bark, glimpse of fur) we
perceive the category <dog>. We may imagine one or more of the other
inputs. We may perhaps think of the word "dog" or anticipate the dog
growling if it does not know us. "Good dog, easy boy" we might say.
This looks like associative memory. Memory storage is not yet modelled.
Addressing and retrieval plausibly might depend in part upon association.
The association part might be through the perceptual input function (PIF)
connections at the category level.
It is commonly asserted that early childhood memories are recoverable after
the child has language as means for storing them. Perhaps it is the
development of the category level that is the gating factor.
The category <dog> is associated with the category <cat> by having many
common inputs to their respective PIFs.
>----- panting
>
>----- wagging tail
>
<dog>-+----- barking
>
>----- "dog" (the word)
\
>>---- tail
>>
>>---- 4 legs
>>
>>---- fur
>>
>>---- growling
/
>----- "cat" (the word)
>
>----- purring
>
<cat>-+----- miaowing
>
>----- retractable claws
>
>----- climbing trees
The taxonomic hierarchies of categories like dog-cat, mammal, animal follow
from their sharing of inputs rather than from any layering of sub-levels
within the category level.
Bruce Nevin