"Working Memory"

[Christian Schunn]

I agree with Wayne that Soar did make a mistake in using the word
'chunk'---their notion of chunk has almost nothing to do with the
historical, psychological sense of the word. However, I don't think we
should generalize to the conclusion that no architecture should use the
word 'chunk'. While the declarative memory element in ACT-R may have
some subtle differences from the general meaning of chunk, it is very
close in meaning:

-Chunks refer to declarative knowledge stored in long-term memory
-Chunks are built up through experience
-Chunks have direct implications for memory span and retrieval times

While it is true that previous incarnations of ACT-R lead to somewhat
arbitrary uses of the old WMEs so that they frequently did not
correspond to the common meaning of chunk, I think this is becoming less
and less true as we think more seriously about the origins of knowledge
in ACT-R models. The visual interface is one big step in that direction.

Perhaps the best resolution of this issue is to ask non-users of ACT-R
what they think of our use of 'chunk'. For long-time ACT-R users, it
does not really matter what term we use---it is the novice user or
reader for which this is an issue. My expectation is that psychologists
will treat DMEs as a hacker's invention; whereas they will bring a lot
of useful and relevant baggage to bear if they see the word 'chunk'.

And on a more amusing note, 'dimes' reminds me of drug dealers.

 -Chris