Constructive Memory and PDP Models

224
Chapter 6
Memory
2000). We still need to identify the processes behind such distortion. In addition, we
are not yet sure why false memories can seem as real to us as our memories of actual
events. Perhaps the more frequently we recall an event (as when students were allowed
to rehearse some lists), the stronger is our belief that we have accurately recalled it.
There is also evidence that merely thinking about, hearing sounds, or seeing photos
associated with certain objects or events appears to make false memories of them more
likely (Garry & Gerrie, 2005; Henkel, Franklin, & Johnson, 2000). Questions about how
false memories are created lead to even deeper questions about the degree to which our
imperfect memory processes might distort our experiences of reality. Is there an objective reality, or do we each experience our own version of reality?
Constructive Memory and PDP Models
PDP MODELS AND CONSTRUCTIVE
MEMORY If you hear that “our basket-
ball team won last night,” your schema
about basketball might prompt you to
encode, and later retrieve, the fact that
the players were men. Such spontaneous,
though often incorrect, generalizations
associated with PDP models of memory
help explain the appearance of constructive memories.
Parallel distributed processing models of memory offer one way of explaining how
semantic and episodic information become integrated in constructive memories. As mentioned earlier, PDP models suggest that newly learned facts alter our general knowledge
of the world. In these network models, learned associations between specific facts come
together. Let’s say, for example, that your own network “knows” that your friend Joe is a
male European American business major. It also “knows” that Claudia is a female African
American student, but it has never learned her major. Now suppose that every other student you know is a business major. In this case, the connection between “students you
know” and “business majors” would be so strong that you would conclude that Claudia
is a business major, too. You would be so confident in this belief that it would take overwhelming evidence for you to change your mind (Rumelhart & McClelland, 1986). In
other words, you would have constructed a memory about Claudia.
PDP networks also produce spontaneous generalizations. So if your friend tells you
that she just bought a new car, you would know without asking that, like all other cars
you have experienced, it has four wheels. This is a spontaneous generalization from
your knowledge base. Spontaneous generalizations are obviously helpful, but they can
also create significant errors if the network is based on limited or biased experience
with a class of objects or people.
If it occurs to you that ethnic prejudice can result from spontaneous generalization
errors, you are right (Greenwald & Banaji, 1995). Researchers are actually encouraged
by this prejudicial aspect of PDP networks, though, because it accurately reflects human
thought and memory. Virtually all of us make spontaneous generalizations about males,
females, European Americans, African Americans, the young, the old, and many other
categories (Rudman et al., 1999). Is prejudice, then, a process that we have no choice
in or control over? Not necessarily. Relatively unprejudiced people tend to recognize
that they are making generalizations and consciously try to ignore or suppress them
(Monteith, Sherman, & Devine, 1998).
PDP models also help us understand constructive memory by explaining
the operation of the schemas that guide it. As described in the chapters on social psychology and on thought, language, and intelligence, schemas are mental representations of categories of objects, places, events, and people. For example, most North
Americans have a schema for baseball game, so simply hearing these words is likely to
activate whole clusters of information in long-term memory, including the rules of the
game and images of players, bats, balls, a green field, summer days, and perhaps hot
dogs and stadiums. The generalized knowledge contained in schemas provides a basis
for making inferences about new information during the encoding stage. So if you hear
that a baseball player was injured, your schema about baseball might prompt you to
encode the incident as game related, even though the cause was not mentioned. Later,
you are likely to recall the injury as having occurred during a game (see Figure 6.8 for
another example).
Schemas
schemas Mental representations of
categories of objects, places, events,
and people.