CHAPTER 8: ASSOCIATIVE THEORIES OF LTM
THE NETWORK NOTION
Memory connections provide much more than retrieval paths; instead, connections are our memories.
e.g. you know your mother's name because it's recorded in your memory.
this is literally represented by a memory connection
a connection between some memory content representing your mother and some memory content
representing the sound pattern of her name
for all other facts you know, so that all of knowledge is represented via a sprawling network of these connections, a
vast set of associations.
HOW MIGHT THE NETWORK WORK?
first we need some means of representing individual ideas
these representations will be the nodes within the network
these nodes are then tied to each other via connections that we call associations or associative links
search through memory is to begin at one node and to travel via the connections until the target info is reached.
not all associations are of equal strength.
e.g. when asked what your birthday is, you respond easily and quickly; this is because there is a strong
connection between the "my birthday" node and the node representing a specific date.
this connection has frequently been thought about in conjunction to the nodes, creating an easily traveled
path from one to the other.
some active intellectual engagement is needed to create (or strengthen) the connections; the nature of this is also
thinking about the material in several different ways will create multiple connections, each of which can later
be used as a retrieval path
the learner will also be helped if he or she thinks about the material in distinctive ways.
e.g. an association between "the list I studied earlier" and "the opposite of love" is distinctive and will help you
what travels through the associative links is akin to energy or fuel, and the associations themselves can be thought of
as "activation carriers"
once a node has been activated, it can in turn activate other nodes
energy will spread out from the just-activated node via its associations, and this will activate nodes connected
to the just-activated node.
nodes receive activation from their neighbours and as more and more activation arrives at a particular node, the
activation level for that node increases.
eventually, the activation levels will reach the node's response threshold (we say this fires).
response threshold: the quantity of info, or quantity of activation needed in order to trigger a response
this firing has several effects, including:
the node will itself be a source of activation
node sends energy to its neighbours and so activating them
firing the node will summon attention to that node
subthreshold activation: activation levels below the response threshold
activation is assumed to accumulate, so that two subthreshold inputs may add together or summate, and
bring the node to threshold.
if a node has been partially activated recently, it is in effect already warmed up, so that even a weak input will
be sufficient to bring the node to threshold.
spreading activation: as each node becomes activated and fires, it serves as a source for further activation,
spreading onward through the network.
activation spreads out from its starting point in all directions simultaneously, flowing through whatever connections
are in place.
some associative links are particularly effective; others are less so; some associations are "built in"-are innately strong EVIDENCE FAVORING THE NETWORK
Why do hints help us remember? E.g. If asked, "what is the capital of South Dakota?", but given a hint of "Is it perhaps
a man's name?"
mention of South Dakota will activate the nodes in memory that represent your knowledge about this state
activation will then spread outward from these nodes, eventually reaching nodes that represent the name
it's possible to have a weak connection between "South Dakota" nodes and the nodes representing "Pierre"
not familiar with South Dakota, or you haven't thought about this state, or its capital
insufficient activation will flow into the "Pierre" nodes, and these nodes won't reach threshold and won't be "found".
the nodes for "Pierre" will receive two sources (other being "Man's name") simultaneously, and this will probably be
enough to lift the nodes' activation to threshold levels.
recall, memory is best if the state you're in during memory retrieval is the same as the state you were in during
the logic here is the same as it for hints
the nodes representing the thoughts may be linked to the nodes representing the learned material
MORE DIRECT TESTS OF THE NETWORK CLAIMS
The network approach makes it own predictions about memory, allowing a more direct test of the claim that memory does
indeed have an associative basis.
SPREAD OF ACTIVATION AND PRIMING
subthreshold activation can accumulate, so that insufficient activation received from one source can add to the
insufficient activation received from another source.
recall: lexical-decision task
participants' perform this task by 'looking up" these letter strings in their "mental dictionary" and they base
their response on whether they find the string in the dictionary or not
we can use their speed of response in this task as n index of how quickly they can locate the word in their
they response faster if the stimulus words were related so that the first word could prime the second
search through the network is like travel and so the farther one must travel, the longer it should take to reach one's
Collins and Quillian explored the process of using a sentence verification task.
participants were shown sentences on a computer screen
mixed together with these obviously true sentences were a variety of false ones
participants had to hit a true or false button as quickly as they could
we expect slower responses to sentences that require a "Two-step" connection than to sentences that require
a single connection
response times depend heavily on the number of associative steps that must be traversed to support a
in addition, some connections can be traversed more quickly than others
the more "typical" the exemplar, the faster the response times
DEGREE OF FAN
degree of fan: the number of associative links radiating out from a node. If there are more links (high DOF), then
each of the associative nodes will receive only a small fraction of the activation flowing outward from the node.
the activation spreading outward from a higher degree of fan will be much more thinly divided, and so each link will
receive a smaller share of the total.
the more ways divided, the less activation to each recipient
the less activation to each recipient, the longer it takes for the target node to be fully activated, and the longer for the
response to be chosen RETRIEVING INFO FROM A NETWORK
SEARCHING THROUGH THE NETWORK VIA ASSOCIATIVE LINKS
the network that supports your memory works the same with associative links guiding search through the network
just as hyperlinks guide search through the internet
it is also important that activation can spread from more than one source simultaneously (relying on hints and context
it is equivalent of looking up one Web page and finding its set of links
in your memory, this process is achieved automatically: activation will simply spread out from both of the entry points
and, with no intervention and no guidance, will converge on the sought-after node.
FINDING ENTRY NODES
Some nodes within the net are input nodes: they receive activation via associative links; once triggered, they send activation
to other nodes
what is special about this is that they receive most of their input activation from appropriate detectors, with these in
turn being connected to the eyes, ears, and so on.
detectors for words might be triggered by detectors for the appropriate letter combo
these in turn would be triggered by letter detectors, which themselves be triggered by feature detectors.
memory network are seamlessly joined, with detectors (nodes that receive their input from the outside world)
sending activation to memory nodes, and in that fashion allowing cues from the outside to trigger events in the
UNPACKING THE NODES
The simpler we keep each node's informational content, the less we need to rely on some interpretive device.
DIFFERENT TYPES OF ASSOCIATIVE LINKS
Early theorizing introduced different types of associative links, with some links representing equivalence ( or partial
equivalence) relations and other links representing possessive relations.
these links were termed:
isa links, as in, "Sam isa dog"
hasa links, as in, "Sam hasa dog"