(As I am rapidly losing time, I am providing you with the following notes, so that I can talk about other aspects of memory that I think you should know about. I've included the outline headings to help keep this organized.) A. Introduction 1. The Brown-Peterson Paradigm 2. Baddeley's Coding Study B. Atkinson & Shiffrin's Multistore Model {we already talked } 1. Overview of Stores & Processes {about these in the } 2. Sensory Memory {last lecture } 3. Short-Term Store a. Neurological Evidence b. Characteristics c. Search & Retrieval in Short-Term Memory If STS is a memory system, then we can also inquire about how we locate information that currently resides within this system. That was the point of a famous experiment by Sternberg. Sternberg asked a very simple question: Does the time it takes to locate information in STS depend on how many memory traces are currently there? To start with, he talked about two possibilities: That it would matter (a serial search model), and that it wouldn't (a parallel model). Here's how these models operate: The serial self-terminating model is the easiest to understand. According to this model, you check the memories in STS one by one until you find the memory you're looking for. So, on average, the more memories there are in STM, the longer it'll take you to find what you're looking for. This is called a serial model because you have to check things serially: one at a time. It is also called a self-terminating model because you stop once you've found the right memory. So, to give you an example, if you are looking for a particular book on your text shelf at home, you start scanning the titles of the books from left to right, until you come across the title you wanted. At that point, you stop searching since, obviously, you've found the desired book. What about the parallel search model? According to this model, you can check all your memories at the same time, instead of checking them one by one. So, it shouldn't matter how many memories there are in STM, it's going to take you the same amount of time regardless of whether there are one or seven. As an example, we do parallel searches when recognizing a friend's face: You can see the eyes, lips, nose, hairline, etc., all at the same time; you do not need to first check to see if it's the right pair of eyes, and then go on to check if the nose is ok, and after that, the lips. Now that we have our two models, how do we set up an experiment to test which is right? Sternberg set up the experiment by using a probe technique. In this task, you are seated in front of a screen, and see -- on a given trial -- anywhere from one to six letters. They come at you at something like one letter every second. These letters constitute what is called the memory set. Your job is to keep these in STS. Presumably, if you have six letters, then you are loading six memories into STS, whereas if you have only a single letter, then on that trial you need keep only one memory in STS. In any case, after seeing the memory set, you get a warning cue to let you know a memory test is about to happen. That test involves presenting another letter, called the probe. All you have to do is decide whether it's a member of your memory set or not. If it's one of the ones that was in your STS, then you press a "true" button; if it's not, then you press a "false" button. So, the probe can be one of two sorts: a target (one of the items we asked you to memorize), or a distractor (a letter not in the memory set). What does the serial self-terminating model predict in this case? Let's start with targets. Generally, the more items there are in your memory set, the longer you should take to hit the "true" key. The reason is, on average, you ought to find the memory you're looking for about half way through your search. So, for a memory set of one item, you have to check that one item to see if it matches. For a memory set of two items, half the time you'll find a match after checking the first item, and half the time you'll find a match only after checking the second item, as well. Thus, on average, you find a match after checking 1.5 items. For a memory set of 3 items, of course, you should generally find a match after checking 2 items. And so on. But, if you think about what happens with distractors, you'll see a very different prediction: Since the distractor is not in the memory set, by definition, you need to check all of the items. So, to look at what should happen, let's make a chart of the average number of searches per memory set: Size of Memory Set 1 2 3 4 5 6 ---------------------------------------------------------------- Average Number of Searches for: Target Probe 1 1.5 2 2.5 3 3.5 Distractor Probe 1 2 3 4 5 6 ---------------------------------------------------------------- Assuming that it takes a bit of time for each search, what you can see is that reaction time will increase for both targets and distractors as memory set size increases, but it will increase much more for the distractors. What about parallel searches? By definition, it doesn't matter how many items you have in STS if you do a parallel search. So, regradless of the size of the memory set, you ought to have the same time for targets. By the same token, we can also predict that you should have the same time for all distractors, although maybe distractors will take you a bit longer than targets (since it's generally easier to execute a "true" response than a "false" response). So, now that we have a method, we can go out and do the experiment (which I do have on computer, if you want to put yourself through this to get a clearer feel for what is going on). And when Sternberg does, it turns out that neither model adequately describes what subjects do. As memory set size increases, so do the reaction times, as predicted by a serial model. However, unlike the prediction from the chart above, distractors don't show a bigger effect than targets: the reaction times for distractors and targets are identical. Thus, Sternberg found results claiming people do a serial exhaustive search, rather than a self-terminating one. It's as if, once you locate the book you want on your shelf, you still go ahead and check the other titles. Why would anyone want to waste time checking additional memories? That result strikes a number of people as counter-intuitive. To explain these results, Sternberg hypothesized that it takes very little time to check a probe against a memory in STS, but it takes a long time to decide what to do about the results (whether to go on to the next item, or to press one or the other button). So, he argued that it may actually be more efficient to check the probe against all of the memories in STS, and just make one decision at the end, rather than a decision after each item. That's one explanation for this otherwise seemingly odd result. There's another, however, and it can be seen by considering the times involved in this process. What Sternberg found was that each additional item in the memory set added 38 msec to the response time. Let's round that up to 40 msec, and let's assume that we have our full set of six items in STS. In that case, we're talking about 240 msec to check all six items. If we round that figure up to 250 msec, you can see that you can check six items in under a quarter of a second. That is such a fast time that maybe the mind doesn't have enough time to signal the search process to stop once it's found a match. Consisten with this argument, memory sets that involve many more com- parisons than six items tend to show self-terminating searches. Two final points on the issue of searching for information in STS. First, Clifton & Trask repeated Sternberg's experiment using single-syllable words rather than letters. They found exactly the same results he did. That is exactly what we would expect from a claim that STS holds about 7 chunks. In Sternberg's experiment, the chunks were letters; in Clifton & Trask's experi- ment, they were words. Thus, this replicates Murdock's finding that single- syllable words give the same result as single-syllable letters in a Brown- Peterson task. Second. Schneider & Shiffrin, who had the theory of automati- city we looked at in the unit on Attention, find that with consistent mapping, people start doing parallel searches. Thus, while we normally do serial exhaustive searches of indormation from STS, if we are overloaded or if we have automated, other types of searches are possible. 4. Long-Term Memory Finally, we can briefly talk about LTM. Obviously, in terms of characteristics, LTM has a huge capacity, has a potentially long duration (years & years, we hope!), and codes in a variety of formats, although Baddeley's experiment suggests that semantic coding will be a very important format. As for why we forget information from LTM, it may be because of decay (although most people don't believe decay is important in LTM), interference, failure to use the proper retrieval cue, or simply searching for something in all the wrong places. As for how information is stored in LTM, it is also stored in chunks, although we tend to call these chunks episodes. Thus, Wickens, Dow, & Moody decided to redo Sternberg's experiment, but included a condition where people were sometimes subject to a distractor task after seeing the memory set, to force them to put it in LTM, and to wipe out their STS. The question they asked was, would the chunks people use in STS be the same as the chunks in LTM? To simplify the experiment, Wickens et al. used a memory set of 2 or 4 items. And what they found was, for the comparisons involving STS (those without the distractor task), people in their experiment showed exactly the same comparison times as Sternberg's subjects. But, when the distractor task was used, people were generally about one tenth of a second slower. And that was true both for a memory set of two items and a memory set of four items. Why was this an important finding? In the STS experiment, it looks like you store 2 or four chunks depending on the size of the memory set, and it looks like each chunk is a single letter. If you store single letters in LTM, then when you have to search a memory set of two items, you will first have to locate each item in LTM, retrieve it into STS, and then do a normal search. But when you have four items, you will have to locate and retrieve four items. In general, the model is: locate the desired information in LTM, copy it to STS (where we do our active thinking), and then search STS. So, according to this model, your response times should be as indicated in the chart: Size of Memory Set 2 4 ---------------------------------------------------------------- Number of LTM Retrievals: 2 4 Number of STS Comparisons: 2 4 ---------------------------------------------------------------- So, if 2 retrievals takes .1 sec, then 4 retrievals from LTM should take double that amount (.2 sec). But of course, this didn't happen. And the reason is, apparently people created new chunks/episodes in LTM. Specifically, the entire list in STS became a single chunk stored in LTM, so that all you had to do was locate that one chunk (your list of 2 items for the memory set 2 condition, and your list of 4 items for the memory set 4 condition), copy it to STS, and then search STS normally. In this experiment, the time it took to locate and retrieve a list from LTM was .1 second. So, STS and LTM do NOT necessarily chunk information in the same sorts of ways. Given this, let's briefly close out this section by considering several studies that involve a very important finding that seems to require both STS and LTM, the serial position effect. a. The Serial Position Effect One of the major findings (for which there really still is no explanation that we all agree on) is the serial position effect (SPE). This has to do with your memory for a relatively long list of words, long enough (say 16 or 25 or something like that) that they can't all possible fit into STS. The SPE is the finding that memory is best for the first couple of words (the primacy effect), and the last couple of words (the recency effect), but is so-so for the words in the middle (asymptote). Specifically, you'll typically show better memory for the first word than the second, and you'll have better memory for the second word than for the third, etc. In analogous fashion, your memory for the last word will usually be better than your memory for the next-to-last word, which will usually be better than your memory for the word that came before it. Why is this important? According to multistore theorists like Glanzer, the SPE is really two different results that reflect the operation of two different memory systems. Specifically, primacy and asymptote are thought to be due to LTM, and recency is thought to be due to STS. So, when the list finishes, you still have a few words echoing around in your STS, and you tend to dump those out first (accounting for recency). As for primacy and asymptote, we need to add in one more wrinkle to the multistore model to explain how they work. That one more wrinkle is the concept of rehearsal. In Atkinson & Shiffrin's model, rehearsal is the mechanism necessary for creating permanent memories in LTM. Specifically, the more you rehearse an item in STS, the greater the probability of forming a LTM. (This should be a familiar concept to you from Wagner's model of classical condition: Remember the experiments with post- trial episodes where Wagner tried to disrupt rehearsal?). So, when we add in a model of rehearsal, primacy and asymptote start to make sense. Why? Because when you hear the first word, you start rehearsing it over and over. As it is the only word in STS, it is getting all the rehearsals and all the attention. Then, the second word comes in, and it starts getting rehearsed, as well. But, the rehearsals are now being split two ways, since there are two words in STS. That means that the second word is going to get fewer rehearsals that the first word, and that, in turn, means that the first word should have a greater probability of making it to LTM, or of being stronger in LTM. And of course, once the third word comes in, you split your rehearsals three ways so that the third word gets even fewer rehearsals, and is weaker. By the time you get to the words in the middle, STS is filled up, so that each word can only get a few rehearsals before you rehearse something else. Thus, the words in the middle are the baseline probability that stuff has gotten into LTM; they correspond to the fewest number of rehearsals, which is why they are referred to as asymptote. Consistent with this explanation of the SPE, asking subjects for free recall versus serial recall makes a difference. In free recall, you're allowed to remember the words in whatever order you want. Free recall typically shows a strong recency effect, and the words that are normally first written down are the last few words. In serial recall, however, you have to remember the words in the order in which they were spoken. That means that STS (which is holding the last few words) has a bit of time to decay. And as you would expect, we find a weaker recency effect with serial recall. b. Rundus' study using the overt rehearsal technique So how good, really, is this explanation of the SPE? Rundus tried to asnwer this question by using what is called the Overt Rehearsal Technique. He gave his subjects a long list of words, and told them to rehearse out loud, instead of silently. What he found was a strong correlation between primacy and asymp- tote, and the number of repetitions each word got: The first word got the most repetitions/rehearsals, but after 5 or so words, all the others got only a few repetitions. So, in his study, rehearsal was related to primacy and asymptote, as predicted by the theory, but it was not related to recency (a finding which is also consistent with the theory). c. Murdock's experiment using fast versus slow rates A second experiment involved using dissociation logic. In this procedure, if the SPE is really two different curves due to different systems, you try to dissociate the curves by manipulating one system, but not the other. So, something that should influence ONLY STS should also influence the recency portion of the SPE, but not primacy or asymptote. In contrast, something that should influence ONLY LTM should affect primacy and asymptote, but not recency. Murdock took the approach of finding a factor that would presumably influence just LTM. Since LTM depends on rate of rehearsals, he presented the same list of words to two groups. One group heard the words at a fairly fast rate, but the other group heard the words at a fairly slow rate. Th point, of course, is that with the slow rate, each word shold get more rehearsals. Note that with this procedure, there should be no effect on STS: You only have so much room in STS, and this doesn't really depend on whether you hear things at a fast rate or a slow rate. So, if you have room for 5 words, for example, the last 5 words ought to be there in each group, and each group should thus first spit out these words. In line with the rpedictions, Murdock found that a fast rate lowered primacy and asymptote, but didn't affect recency. Thus, he appeared successful in showing that the primacy and asymptote portion of the SPE is from LTM. d. Glanzer & Cunitz's experiment using a distractor task A related approach using dissociation logic was performed by Glanzer & Cunitz. They also had several groups, but they used the same rates for all. How their groups differed was that one had to count backwards for 30 seconds before recalling (a variety of the Brown-Peterson distractor task meant to wipe out STS), whereas another recalled without being distracted. As you would expect from the theory, both groups showed the same primacy and asymptote effects, but the distractor group had no recency effect. This type of procedure has also been replicated by Postman & Phillips. C. Problems with the MultiStore Model So, in summary, there was a lot of evidence consistent with the multistore model's view of memory. But things were about to change. It didn't take long for a number of people (including some of the ones who ran the experiments mentioned above) to find problems with the model. To keep things simple, I'll focus on STM and LTM (ignoring the sensory memories and the issue of attention). 1. Interference vs. Decay in STM The Petersons had claimed that STM or STS lost information because of decay, and that was one of the reasons it was different from LTM (which lost informa- tion because of interference). Later studies showed that decay was not terribly important for STS, and that Brown and the Petersons had overlooked an important intereference possibility. a. Keppel & Underwood's study on PI Keppel & Underwood decided to re-open the issue of proactive interference (PI). If you recall, the Petersons had separated their trials into blocks of 12 trials to check for PI (i.e., interference due to response competition from previous trials). However, Keppel & Underwood asked whether 12 trials was the right size to look for PI. Specifically, they claimed that maybe PI built up very rapidly, within only a few trials. If that is so, then the reason the Petersons didn't see any PI is because it was already there in full strength in the first block. So, Keppel & Underwood re-ran a version of the Brown-Peterson experiment, but with the addition of checking for recall after every single trial. They arranged their experiment so that on the very first trial, some people would count backwards for 0 seconds, others for 3 seconds, others for 9 seconds, etc. Same for the second, third, and fourth trials. And what they found was that there was absolutely NO forgetting on the first trial! Even if you had counted backwards for 18 seconds, you still knew what the three letters were with nearly 100% accuracy. On Trial 2, however, there was evidence of some forgetting with time spent counting backwards, though it was not nearly so dramatic as in the Petersons' experiment. And by Trial 3, their results were essentially like the Petersons. So, they found that PI built up very quickly - within the first three trials. What happens as you count backwards is that you start getting response competition from letters you had on previous trials: You get confused about which letters are the ones you heard on the current trial, and so you are more likely to recall the wrong letters. b. Waugh & Norman's study on RI Waugh & Norman took another approach. They asked whether there might not also be retroactive interference in STS, although they knew that this would not account for the Petersons' results, since letters and numbers are not similar enough for the numbers to interfere with the letters. In RI, of course, something you learn later interferes with stuff you had learned earlier. Waugh & Norman used a variation of the probe technique we looked at earlier, in Sternberg's experiment. People got a list of numbers, but they never knew how long the list was going to be. At some point, they got a probe consisting of a number they had heard earlier in the list. All they had to do, once they got the probe, was try to remember the number that had come AFTER that number in the list. What Waugh & Norman also did was run a fast group and a slow group. The fast group heard numbers at the rate of 4 per second, while the slow group just heard one number per second. Waugh & Norman argued that their procedure allowed comparing two potential mechanisms of forgetting in STS: RI and decay. The reason their experiment can lok at decay is that decay is essentially an explanation that claims forgetting occurs with nothing more than the passage of time: The longer it's been, the less you remember. So, all things being equal, the slow group should have heard stuff longer ago than the fast group, and should have more trouble with that stuff if decay is important. Since this is a bit abstract, let's look at two sample trials to see what Waugh & Norman's experiment was capable of. In the first below, we will deliberately make PI the same in both groups, and we will also make RI the same. The only thing that differs should thus be decay: fast group slow group time (seconds) 1st second 1 4 7 9 1 2nd second 5 8 2 6 4 3rd second 1 7 4th second 9 5th second 5 6th second 8 7th second 2 8th second 6 9th second 1 In the experiment above, 1 is the probe, since 1 is the repeated number. So, the number both groups are trying to remember is the number 4. Note that PI is identical: each group had only one item before 4 (the same item, in fact). Similarly, note that RI is identical: each group has 7 numbers that came after (including the probe; again, the numbers were identical for these groups). The ONLY difference is one of time: The fast group heard 4 about 2 seconds ago, the slow group heard 4 about 7 seconds ago. So, if decay is operating like Brown and the Petersons and the multistore model claim, the fast group should have better memory. But in fact, what Waugh & Norman found was equal memory for the two groups in this type of setup. Now consider another type of setup: fast group slow group time (seconds) 1st second 1 4 7 9 1 2nd second 5 8 2 6 4 3rd second 1 7 4th second 1 As before, both groups have to remember 4. As before, PI is equal for both groups. Note that decay is now also equal, since each group heard 4 about 2 seconds ago. But, they are no longer equal on RI: The fast group heard 7 numbers after 4, but the slow group heard only 2 numbers after 4. So, if RI operates in STS, there should be more opportunity for response competition (due to more potential responses causing confusion) in the fast group. And in fact, the upshot of Waugh & Norman's experiment was that forgetting seemed to be strictly a function of RI: The more numbers there were that came after the one you were trying to remember, the worse your memory was. So, not only does STS have PI, it also has RI, just like LTM. c. Reitman's study on decay The studies we've looked at so far suggest that short-term memory primarily loses information due to interference and response competition. So, the question to be asked is whether there is any evidence of decay, as originally claimed by Brown and the Petersons. The answer is that there is *some* evidence of decay, but it's not a major source of loss. This involves a study by Reitman. She gave her subjects a distractor task that shouldn't involve PI or RI: trying to listen to a series of "DOH" syllables in a staticky back- ground, and pressing a key whenever a "TOH" would appear. Obviously, as these aren't real words, response competition should not be a problem. She also made this a hard task: She arranged the level of static so that people would be correct in detecting the "TOHs" only half the time. AND, she deliberately asked her subjects *not* to rehearse. At the end of the experiment, she checked with them to see whether they had rehearsed or not. Those who said they were able to rehearse showed no forgetting. However, the non-rehearsers did show a bit of forgetting, consistent with a decay theory claim. So, there is some decay operating, but it is not a major factor. 2. Auditory vs. Other Coding in STM The multistore model claimed that STM coded primarily for sound, and LTM coded primarily for meaning. A number of studies quickly challenged this: a. Wickens' study on semantic codes (release from PI) Wickens had people listen to short lists of words and try to immediately recall them in order. When the lists all involved items from the same semantic category, he found that recall got worse with later lists, due to proactive interference building up (as in Keppel & Underwood). But, when later lists involved a *change* in semantic category (from names of animals to flowers, for example), then there was no proactive interference. Since these lists seemed to be small enough to be kept in short-term memory, he argued that this result showed that semantic meaning was capable of causing inter- ference in STM. For that to be the case, STM would also have to hold semantic codes, and not just acoustic or sound codes. b. Shulman's study on homonym and synonym probes of the SPE Shulman did a clever experiment using the serial position effect. If you will recall, the recency portion of that effect is supposedly due to STM, while everything else is due to recall from LTM. He reasoned that if STM specialized in sound codes and LTM specialized in semantic codes, then people ought to be especially good at processing meanings on primacy and asymptote words, and sounds on recency words. So, after hearing a long list of words, his subjects got a probe task. There were three types of probes. *Identity* probes presented a word, and the subject's task was to say whether it had been on the list. These showed the normal serial position effect. *Homonym* probes presented words that were not on the list. However, half of these sounded like list words, and subjects had to indicate which were the ones that sounded like the originals. If recency words are coded for sound, people ought to be particularly good at making homonym judgments for the recency words. Finally, *synonym* probes were non-list words, but half of these were words with similar meanings (EVENING versus NIGHT, for example). In this condition, subjects had to indicate which were the words with similar meanings. If primacy and asymptote words are coded for meaning, then people ought to be particularly good at synonym judgments for these two types of words. The result: There was no evidence that a given probe type was better for one portion of the serial position curve than for another. This implies that sound AND meaning information are available for both the primacy words (in LTM) and the recency words (in STM). Thus, like the work on interference, this study claimed that STM and LTM held very similar types of codes or information. c. Shepard's study on visual-like imagery codes in STM Another set of experiments that challenged the idea of there being only auditory codes in STM was done by Shepard and his colleagues. Your text mentions several of these. In one, Shepard gave people the task of deciding whether a letter, if presented upright, would be facing the correct direction. For example, you might see an "R" at different angles from upright, or you might see the reversed "R" at different angles. What Shepard found was that the further from upright a letter was, the longer it took to decide if it was facing to the left or the right. He claimed that this was evidence of imagery. Why? Because according to his model, you placed the image of the letter in your STM and you then had to rotate it to upright to see which way it faced. So, in rotating the letter, you had to make it go through all intermediate points, which took time. The alternative, according to Shepard, would have been to form a description of the letter and compare it to your description of an "R". If that is what you did, all descriptions of the backwards-facing R should have been identical to one another, and it should not have made any difference how far from upright the backwards R was. Similarly, all descriptions of the forward-facing R should have been identical. So, the finding that distance from upright has an effect seems to strongly suggest that people are capable of rotating mental images. These types of studies are referred to as MENTAL ROTATION studies. d. Kosslyn, Ball, & Reiser's experiment on imagery in STM Another study that supported the notion of imagery was done by Kosslyn et al. They had their subjects learn and study a map of an island. The island had many different locations on it (well, pond, hut, rocks, beach, dunes, etc.). To prove that you had learned the map, you had to reproduce it accurately in a drawing. Anyway, once people had learned the map, they were told to imagine being at a certain spot on the map (like the Well). They were then given a second location, and timed while they decided whether the second location was on the map or not. Basically, the finding in this MENTAL SCANNING study was that it took longer to decide when things were further away. Like Shepard, Kosslyn et al. claimed that this meant people had to mentally scan along all the intervening points, so that it took longer when there were more of these. To give you a feel for this, quickly imagine going from Girard Hall to the following locations: Martin Hall; Griffith; Cajundome; the Mall. Your sense ought to be that you mentally 'arrive' at the nearer locations a bit sooner. The alternative theory is again a descriptions-based theory that claims that your map basically is stored in STM as a list of spots, so that all you have to do is look down your list to find whether something is on it or not. That type of theory doesn't seem to predict this effect, according to Kosslyn. e. Kerr's work on motor/spatial codes in STM Finally, we can ask what it means exactly to have imagery codes in STM. Does this mean that we carry around something like a picture, a visual scene? Our own experiences seem to suggest that this is indeed the case, but an experiment by Kerr suggests otherwise. Basically, Kerr repeated Shepard's mental rotation experiments. He also got the same results. So, just what is the problem here? Only something very simple: The people who did his experiment had been blind from birth! The letters they got were Braille letters that were at various angles. Clearly, in this case, these subjects weren't creating visual pictures. But, they do have codes that tell them about locations and directions in space. Whatever imagery they created, it apparently involved these types of spatial (your text calls them "motor") codes rather than visual codes 3. Rehearsal as the means of encoding stuff into LTM Finally, the Atkinson-Shiffrin model claimed that rehearsal was the primary mechanism for coding stuff into LTM. A number of studies very quickly challenged this claim. They showed cases where rehearsal just plain failed to have an effect, or where other mechanisms proved more important. Your text lists a bunch of these, but I'll go through a few important studies. a. Craik & Watkins' study on phoneme monitoring Craik and Watkins did a clever experiment where they took a group of subjects and told them they would hear a list of words. Their task was to remember the last word that started with a certain phoneme or sound -- say, the "g" sound. So, as a subject, you might hear a list like the following: girl desk goat gorilla chalk cat pen phone lab gate... If you ask yourself what you would do in this situation, the answer is something like the following: As soon as you hear a target word, put it in STM and start rehearsing it. Listen for the next target word. When you hear that, stop rehearsing the first, and start rehearsing the second. And so on, until the experimenter finally asks you what the last word that started with a "g" was. Now go back to the list: GOAT will not get very many rehearsals, since a word starting with "g" came immediately after. GIRL should get more rehearsals than GOAT, and GORILLA should get many more rehearsals than GIRL, since there is a long stretch in there without target words (words starting with a "g"). After the subjects had finished the experiment and had started another task, they got a surprise request to list ALL the g-words. The rehearsal theory predicts that words that got a lot of rehearsals should be more likely to be remembered. However, this did not happen. In this experiment where subjects did NOT think they would have to remember the words, the memories for the g-words were all at about the same level. So, what this shows is that it is not rehearsal that causes the LTM to form, it is rehearsal AND intention to learn! Somehow, the intention to learn is causing something else to happen that plain rehearsal by itself isn't. b. Rundus's experiment using the Distractor Recall Paradigm Another study showing the same finding was by Rundus. He did a variation of the Brown-Peterson study. Subjects got a list of numbers, and the distractor task involved repeating several words out loud for a certain amount of time. Then, they tried to remember the numbers, and went on to the next trial with a new set of numbers, and a new set of words. However, at the end of it all, Rundus asked his subjects to try to remember the distractor words. The rehearsal theory predicts that words that were repeated a lot should have had a better chance of getting into LTM. However, Rundus found that amount of time spent repeating a word had no effect. People didn't think they would have to recall the distractor items, and these simply did not make it into LTM. c. Bower's study on interaction Bower looked directly at the effects of rehearsal instructions. He had three groups of subjects. Each got the same list. Basically, each knew it had to learn the list for a later test of memory. On the list were pairs of words. One group was told that the best way to learn these pairs was by rehearsing them -- repeating them over and over. A second group was told that the best way to learn them was by forming an image for each word. The third group was told that the best way to learn these words was by forming an image for each PAIR of words -- an image in which the two objects would be interacting. If your pair was CUP-TIGER, for example, you might have formed an image of a tiger holding a cup. The results were that the imagery and rehearsal groups were about equally good. But, the interacting imagery group was far superior to the other two. So, tieing stuff together has a stronger effect on remembering than simply repeating individual things. d. Bower et al.'s study on Organization Finally, another relevant study by Bower, Clark, Lesgold, and Winzenz looked at the effects of organization. To give you the short version of this study, people briefly saw about 112 words. The words appeared on 4 cards. On average, they would have seen each word for only a few seconds. After seeing the cards, people tried to remember the words. In this study, there were two groups of subjects who differed not in which words they saw, but in how the words were organized. For one group, the four cards corresponded to four different categories. At the top of the card was a word naming a category (like MINERALS). The words on that card would look something like the following: MINERALS METALS STONES RARE COMMON ALLOYS PRECIOUS MASONRY platinum aluminum bronze sapphire limestone silver copper steel emerald granite gold lead brass diamond marble iron ruby slate Then, they would get another card representing another category such as TRANSPORTATION or ANIMALS. The other group got four cards too, and even the same words. But in their case, the words were all scrambled up across the four cards. And of course, what you will have noticed about the card example above is that it is organized in a logical, hierarchical fashion. As you might have guessed, the basic result is that the organized group remembered most of the words, but the disorganized group remembered very little. Thus, how you organize things has a very powerful influence on how well you will remember them. (One reason why I give you an explicit outline, in fact!) SUMMARY So, the bottom line: There were a lot of studies that suggested the multi- store model was too simplistic. Rehearsal didn't seem effective, and a lot of studies started suggesting that how you *encoded* stuff was going to be more important than how long you rehearsed it. This led to the final batch of models that we'll talk about in class today - the processing models. Unlike the store models, they make the claim that you can process or encode stuff in a variety of ways, and that some of these will lead to longer lasting or easier retrieved memories than others. .