storage

Brian MacWhinney macw at CMU.EDU
Mon Oct 5 17:20:55 UTC 1998


Bill,
  Yes, psychology is rich with evidence regarding the storage-retrieval
issue.  I suppose that one could easily locate 500 recent articles on this
topic.  Going to this literature for specific references to cite is fine,
but perhaps you also want to get a lay of the land.  Consider the issue of
reading a new unknown word vs reading a known word.  For the known word, we
may do some phoneme-grapheme analysis, but we have a well-greased lexically
specific route to the word's sound.  The more frequent the word, the faster
we read it, etc.  My guess is that half of the papers in the area of word
recognition report significant frequency effects.
  Consider a simple result from Stemberger and MacWhinney (1986, Memory and
Cognition).  We found that high frequency regular past tense forms (wanted)
are more resistant to irregularizing speech errors than are low frequency
regular past tense forms (panted).  We believe that this is due to the fact
that high frequency forms are more thoroughly stored for unitized rote
retrieval.
  More broadly, psychologists tend to view this in terms of the power law
of learning.  Any process that is done repeatedly gets stronger according
to a power function.  Typically, this is understood as occuring through
chunking.  When lots of pieces tend to occur together frequently, the whole
starts to functions as a single chunk.  There is a lot of research and
modeling literature on this.  People at CMU like Anderson, Simon, Newell,
Just, etc. have done lots on this.  One twist on the power law emphasizes
how learning of a given form leads to entrenchment.  Joan Bybee's work
certainly builds on such findings and elaborates them.
  Another way of viewing this is in terms of the strategy-selection model
as developed by Reder, Siegler, and others, mostly for math, reasoning,
etc..  When confronted with a given problem, we have to chose whether to
retrieve or analyze. We usually apply a quick filter on the problem that
decides which of these two approaches will be most useful and then go from
there.  For language, this framework might be useful for high level
strategy choices in complex syntax and pragmatics.
   This is an enormous area.  Reading the recent work in this area is
somewhat complicated by the fact that current models emphasize
connectionist modeling which tends to distract from the issue you are
asking.  For this reason, you might find textbooks from the early 80s or
even 70s clearer on these issues  than some recent textbooks.  However, if
you stick with models like John Anderson's ACT-R model as reported in his
recent books and textbooks, the role of frequency is clearly highlighted.
  By the way, none of these remarks have anything to do with the really
really complex symbol manipulation models linguists often propose.
Instead, for psycholinguists, the dichotomy is usually between extremely
trivial rules like "add -ed" or "ph sounds like f".  We pretty much
discarded any belief in the psychological reality of really complex formal
linguistic rules in the 1970s.  This is not to say that psycholinguists are
not interested in abstract categories.  Some are still playing around with
traces, universal parts of speech, abstract syntactic structure, and the
like, but this work seldom gets grounded on really complex and abstract
rule systems.

--Brian MacWhinney



More information about the Funknet mailing list