mental data structures

[gusl]

I've been thinking: how do people memorize long things? My proposal is that they use something like trinary trees.

(chunk-type node slot1 slot2 slot3)

(add-dm
  (c0 ISA node
     slot1 c1
     slot2 c2
     slot3 c3)

  (c1 ISA node
     slot1 Sun
     slot2 Mercury
     slot3 Venus)

  (c2 ISA node
     slot1 af31
     slot2 dda4
     slot3 0783)

  (c3 ISA node
     slot1 rats
     slot2 people
     slot3 elephants)
)

7+-2 is a rather well-accepted "information-theoretic" limit on short-term memory: this is how many "chunks" one can keep in there. This is not modeled explicitly in ACT-R: there is only one "declarative memory".

One effective strategy is to put more information in each chunk. But it is obvious that one can only do this so much. Niels has said that in ACT-R, chunks with more information heavier, and become this way harder to retrieve.

Meaningful chunks, of course, are easier to remember (see c1, c3). The relationship between the slots means that once one of them is found, retrieving the other 2 will be easy (through the associations). Also, since they have links to many concepts (that's what "meaningful" means), it's easier to retrieve them. And I'm not even talking about truth constraints yet (in case you're trying to remember a set of true sentences).

Comments

[jcreed.livejournal.com]

Where does the number three come from?

[gustavolacerda.livejournal.com]

my ass.

[gustavolacerda.livejournal.com]

but I bet that there is something meaningful that it optimizes.

[altamira16.livejournal.com]

Mnemonic devices

My
e
r
c
u
r
y

Very
e
n
u
s

Excellent
a
r
t
h

Mother
a
r
s

Just
u
p
i
t
e
r

Served
a
t
u
r
n

Us
r
a
n
u
s

Nine
e
p
t
u
n
e

Pizzas
l
u
t
o

Representation and Understanding

[http://users.livejournal.com/_greg/]

According to Don Norman, STM provides 4 general-purpose registers and a few modality-specific registers. Therefore, we can hold onto 4 arbitrary chunks plus a visual chunk, plus an auditory chunk, etc. Skill at encoding and re-encoding allows us to simulate having more registers. Using modality-specific registers may clash with the current task, etc. Unfortunately my only reference here is "personal communication", but read on.

The key reference I have for you is the foundational everyone-must-read book Representation and Understanding: Studies in Cognitive Science edited by Daniel Bobrow and Allan Collins. This classic (1975) collection is still a must-read as it contains the original and very readable articles on the foundations of semantic networks, frames and schemas.

_Greg

Re: Representation and Understanding

[gustavolacerda.livejournal.com]

Thanks very much for the ref. I have a notebook (mostly a stub) on Understanding, precisely about this issue. I am particularly interested in the case of mathematics.