Information Technology Reference
In-Depth Information
longer time scale, result in the formation of seman-
tic/procedural memories (McClelland & Rumelhart,
1986). Thus, these priming effects provide a behav-
ioral window into the neural mechanisms of learning,
in addition to playing an important role in cognition by
providing a trace of recent experience.
Importantly, many forms of long-term priming are in-
tact (at normal levels) even in patients with hippocam-
pal system lesions (Graf, Squire, & Mandler, 1984),
which supports the idea that this priming does indeed
reflect cortical learning, and not learning in the hip-
pocampus. Thus, although one might have expected
based on our previous characterization that the hip-
pocampus is solely responsible for rapid learning ef-
fects, intact long-term priming in hippocampally le-
sioned patients suggests that the effects of a single
stimulus presentation can in fact be mediated by the
relatively small weight changes in the cortex that are
consistent with the slow overall acquisition of seman-
tic/procedural knowledge.
probe stimulus. A special case of this dimension is
called repetition priming, where the prime and the probe
are identical (e.g., the example of reading a list of words
and then reading those same words again a second time
more rapidly). More generally, similarity-based prim-
ing involves the spread of activation along associative
links (e.g., along semantic associations, or according
to visual similarity), resulting in facilitation for probes
that are similar to the prime (e.g., bread priming but-
ter). This spreading activation can also produce weight
changes, so that one might expect both short-term and
long-term priming in this case. Indeed, long-term prim-
ing of this form has been recently demonstrated and ex-
plained using just this type of model (Becker, Moscov-
itch, Behrmann, & Joordens, 1997). In what follows,
we will focus on the simpler case of repetition priming
in the lexical content dimension, but we assume that the
same basic principles apply across all examples of long-
term priming.
One of the main behavioral methodologies for study-
ing long-term repetition priming is stem completion ,
in which participants first study a list of words, and
they are then presented with a list of initial word frag-
ments (stems) and asked to generate words that com-
plete these stems. Unbeknownst to the subjects, many
of these stems can be completed with the previously
studied words. For example, a participant might read
“reason” initially, and then later be asked to complete
the stem “rea .” Because the stems are constructed to
have several possible completions and are pre-tested to
ensure these possible completions are roughly equally
likely to be produced, any increased probability of com-
ing up with the studied word (relative to the pre-test lev-
els) can be taken as an indication of some residual ef-
fects of processing the word. In other words, the initial
studying of the word has primed the subsequent pro-
cessing of that word. The fact that the list study is sepa-
rated from the stem completion task by a relatively long
time period (tens of minutes) and the study list contains
more words than could be maintained in active memory
suggests that a relatively long-lasting mechanism is at
work (i.e., long-term priming , as compared to the much
more transient short-term priming produced by residual
activations).
9.2.1
Long-Term Priming
The term priming generally refers to any facilitory ef-
fect of prior processing on subsequent processing. For
example, if people read a list of words out loud and
then read that same list of words again, they will be
faster the second time around. Interestingly, priming
can also be seen across semantically related words, so
that reading “bread” facilitates the reading of “butter,”
for example. Many different forms of priming have
been described (Schacter, 1987), but from our mech-
anistic perspective we will consider three cross-cutting
dimensions for delineating different types of priming:
(1) duration, (2) content, and (3) similarity. The du-
ration dimension has been discussed already, and in-
volves the distinction between short-term (activation-
based) and long-term (weight-based) priming. We will
discuss activation-based priming at greater length in
section 9.4, and focus on weight-based priming here.
The content dimension reflects the nature of the repre-
sentations involved, for example including visual, lexi-
cal (word-based), and semantic representations.
The similarity dimension reflects the similarity be-
tween the prior prime stimulus and the subsequent
Search WWH ::




Custom Search