Flashcards in Memory Deck (26)
WM v LTM
Short-term storage vs. long term retention
Seconds – 1 minute v several minutes – days
episodic v semantic/reference
My first day at University (What? Where? When?) vs knowing that Rescorla-Wagner is a model of associative learning (how the world works)
Semantic – context independent
early exps on relatively STM in animals
In simple conditioning, learning is seriously impaired if there is a gap between stimuli, or between response and reinforcer (delay of reinforcement)
In discrimination experiments with rats, Grice (1948) reckoned that impairment was evident with a 0.5-sec delay, and total with 10-sec delay
Similar results in discrimination experiments when there is a delay between stimulus and the possibility of response (delayed reaction) - often used with monkeys but also with rats (e.g. Smith 1951)
These results suggest a very short duration of working memory, however, much longer delays are tolerated with taste avoidance learning (Garcia & Koelling 1966 and many many others since). The important thing here is that we should not muddle the conditions affecting learning with those affecting memory.
Target memories could be lost because of
- Proactive interference from information acquired PRIOR to target
- Retroactive interference from information acquired AFTER target
- Decay due to the passage of time.
delayed matching to sample (DMTS)
Modern, Skinner-box technique for studying delayed reaction
Despite the name, does not necessarily require recognition of identity of sample and comparison stimuli
- oddity from sample - choose the comparison that does not match
- symbolic matching to sample (DSMTS): comparison stimuli are not the same as the sample, and the subject must learn the "code" connecting them
- multiple samples - in which case the design becomes a test of recognition memory or a list recognition task
Used with pigeons, monkeys, dolphins, etc – can all learn task
Grant and Roberts (1973) used ordinary DMTS on some trials to test proactive interference.
On other trials, two samples are presented; either 10s or 0s apart and the animal must respond on the basis of the last sample.
The results show that performance in the zero second condition, when one sample is immediately followed by the other, is worse, suggesting previously learned information is interfering with later learned.
This is good evidence for proactive interference in DMTS.
Grant (1988) used an ordinary DMTS to demonstrate retroactive interference.
On some trials, the brightness of the chamber is increased during the delay between sample and comparison.
It was found that the brighter the chamber, the worse the performance, suggesting evidence for retroactive interference; new information interferes with memory of the previous information.
the radial arm maze
Lots of experiments involving rats make use of the Radial Arm Maze, which is means of presenting a list of items to the animal.
In early experiments, animals were allowed to run down a certain number (of freely chosen) arms then the rats were removed for a delay interval and then returned on test.
Later experiments controlled the arms experienced on the study phase.
early work in the radial arm maze
Olton and Samuelson (1976) looked at choice accuracy in free choice procedure in an 8-arm radial maze.
If a rat re-entered a previously chosen arm, this was classed as an error.
The most striking aspect of the results is that performance was very good.
However, a free choice procedure suffers from the possibility that this is not so much due to memory as due to stereotyped response patterns on the part of the animal, e.g. after exiting an arm, turn left and take the next one.
forced the animal to visit a set of randomly chosen arms (blue).
The animal is given a choice between two arms, one visited and one novel, and is required to choose only one.
It was found that the rats are very good at this.
proactive interference exp
Hoffman and Maki (1986) conducted an experiment in which eight arms of the radial maze were portioned into two sets, A and B, of four arms each, then an initial phase of exposure to the arms in B was given, followed two hours later by a study phase in which the rats were forced to the arms in A.
After a further two hours a test phase of free choices amongst all eight arms was given, with choices of the set B rewarded.
A control group didn’t get the initial interference phase.
Experimental animals performed worse than controls, indicating a PI effect of the initial exposure to the B arms.
retroactive interference exp
Roberts (1981) conducted a different study, again where the eight arms were portioned into two sets, A and B, of four arms each, with another exposure to the arms in B given, followed by running three similar mazes in different rooms to the test maze.
Finally, the animal was returned to the test maze and its performance in choosing the A set of arms was found to be worse than that of controls who had experienced the same delay between the initial phase and test, but not run the intervening mazes.
Roberts and Smythe (1979) conducted a study focusing on decay.
They used the basic 2AFC procedure, where animals were forced to either one, three, five or seven arms of the mazer and then given a choice between a visited and unvisited arm.
The results were plotted in terms of backward serial position, where 1 = the last arm visited (hence the same delay between study and test in all conditions) and 2 = second to last arm visited.
Therefore, for the three-arm condition, three would be the first arm visited on that trial whereas for the five-arm condition it would be five.
This means that elapsed time is the only factor the backwards curves should superimpose, and the researchers found that they did.
For this, the animal was forced to one arm, returned to the centre of the maze, then confined there (the centre area was made opaque during the interval) for the same length of time that it had taken to run 7 arms.
Again, if delay is all that matters, then performance on test should be similar to backward position 7 in the 7-arm condition, but if number of arms visited is the crucial factor, then it should be similar to backward position one.
The results show that there are more errors when the rat had been down more arms, suggesting their memory had faded as they had done other arms since.
The same result with controls as with the seven-arm condition suggests decay has occurred.
the nature of the code in memory
The nature of the code in memory can be retrospective or prospective.
Retrospective is, for example, remembering the arm you just visited, remembering the past.
Prospective is, for example, remembering the arms you are yet to visit.
Animals can definitely use retrospective encoding, but can they employ a prospective code and can they switch between the two codes in a flexible manner?
delayed symbolic matching to sample
In the delayed symbolic matching to sample experiments, the issue of the identity of the sample and comparison stimuli doesn’t arise.
There are two different discriminations (between samples and between comparison stimuli), which was exploited by Roitblat.
They can learn by trial and error.
DMTS Roitblat (1980)
Roitblat (1980) discovered that animals make more mistakes – more confusions – when Blue and Orange are the sample.
They tend to give the correct response for Orange to Blue sample and vice versa.
They make less confusions on trials where the sample is Red, even though some confusion with Orange might be expected, as 0 and 12.5 are very similar.
This suggests the use of a prospective code.
Cook, Brown and Riley (1985)
produced evidence that suggested that rats could switch flexibly between retrospective and prospective codes using a twelve-arm maze.
Rats were forced to either two, four or ten arms t=and then taken out for 15 minutes and tested on a visited v unvisited arm.
When less arms are previously visited, it is easy to learn which arms you have been down (retrospective), but when more arms are previously visited, it is easier to remember which arms you did not go down (prospective).
results from an eight-arm radial maze with a 15 minute retention interval.
More errors are made in the four-arm condition than the six-arm condition.
Cook et al. (1985) reported a similar findings coupled with a serial position effect in the conditions that showed the highest error rate, whereby the arms entered early in a trial were more likely to be (incorrectly) revisited after the delay.
They argued that this indicated a switch from a retrospective code to prospective code at the appropriate list length.
This has also been reliably demonstrated in humans.
The main issues when considering LTM are what transformations in the memory trace take place between original encoding and final storage, and whether memories that are forgotten are lost completely or merely become inaccessible.
Forgetting should be considered as a failure to retrieve, the memory is still there, it’s just that the animal is unable to access it under present conditions.
Consolidation refers to the idea that the initial memory trace is fragile and relatively short term and needs to be transformed (consolidated) into a more durable trace for the long-term.
Many studies have used avoidance learning in the shuttle box.
In this paradigm, the rate is trained to move from one side to the other to avoid shock.
When the rat stays on
one side too long, it gets a foot shock, and so the rats learn to keep switching.
phys disruption of consolidation
Several types of physiological trauma prevent learning if and only if they occur shortly after a stimulus or reward.
Examples include ECS (electroconvulsive shock: Duncan, 1949), sudden heating or cooling, drugs, concussive impact (American football) – often don’t remember anything from/before event – trauma prevents consolidation.
Therefore, after learning a shuttle response to avoid shock, ECS given shortly after training will disrupt performance after a delay, whereas ECS given after that delay and just before test will not be as effective.
The animal has to retrieve a memory to make use of it.
Many studies indicate that most forgetting should be viewed as a failure to retrieve rather than a loss of storage.
The most impressive of these is the class of study that falls under the heading of ‘reactivation’. In these experiments memories that are apparently ‘lost’ are successfully retrieved followed a reminder treatment.
This type of study poses difficulties for consolidation theory.
If a memory has not been consolidated, and hence is lost, how can it later be retrieved?
Gordon’s (1981) paper reports this reactivation study in the shuttle box.
Times are in seconds.
The delay between training and test was three days.
Reactivation can also restore memories after ECS (Gordon and Mowrer, 1980).
The results show that quite a bit is forgotten after three days.
If you hold them there for too long, it is way past the time when they would normally have a shock, so it reverses learning.
Similar effects have also be shown in humans.
capacity of animal LTM
Vaughan and Greene conducted a picture recognition experiment with pigeons.
They did a random pseudoconcept task, where slides of tress were arbitrarily divided into positive and negative, with original training with 40 positives and 40 negatives.
New slides were introduced once old ones mastered.
There was successful training up to 640 slides. Birds trained with 320 slides were then rested for 2 years; on re-testing there was some forgetting but quick recovery.
are animals stuck in time?
It is much less clear whether we can claim episodic memory in animals.
Roberts (2002) claims that animals are “stuck in time” – they cannot do what Suddendorf calls “Mental time travel”.
Clayton and Dickinson have famously argued that scrub jays can, in fact, do just this.