Langston, Cognitive Psychology, Notes 6

Langston, Cognitive Psychology, Notes 6 -- Episodic Long Term Memory

Note: Some of the memory demonstrations that we will do in class will be messed up if you have advance knowledge of them. I encourage you to preview the notes before class, but try to skip the parts about the memory demonstrations.

I. Goals.
A. Where we are/themes.
B. The processes.
C. The parts.
D. Segue into a new idea.

II. Where we are/themes.
A. Here are some situations:
1. Why is it sometimes so hard to remember things after you study them? Why do you remember some things really well, but other things not as much when you study them in the same session? (I have some questions for you about this class.)
2. There's a lot of stuff that I know for a fact I learned at some point in my life. For example, I got a B in Calculus. I must have known some calculus back then. Now, I can’t remember a thing. Why?
B. Where we are. Remember, we’re working our way through this box model of the mind. We’ve talked about the sensory register. The register holds information briefly, and pattern recognition figures out what the information is. We also looked at attention. Last time we looked at brief memories in short-term memory. Now we turn our attention to long-term memory. What is long-term memory and how do you use it?
Information processing model

Long-term memory: A permanent memory store with unlimited capacity that acts as a store for everything you know.
C. Themes.
1. One long-term memory or many? The evidence indicates that there are three or four. These could be different kinds of memory, or different kinds of processes. We’ll discuss the evidence for that.

Top

III. The processes. Two basic things: Encoding (getting stuff in) and retrieval (getting stuff out). In a way, these two aren’t really independent (as in “if it’s not encoded it can’t be retrieved”). Let’s treat them as separate anyway.

CogLab: To start things off, we'll look at the data from our serial position demonstration.

A. Encoding: What affects it?
1. What you do with the material. You have lots of control processes that you can use to get material from short-term memory to long-term memory. Different strategies work better for different people, and different strategies work better in different situations.
a. Rehearsal (repetition). This is probably everyone’s favorite mode of learning a list. As you hear each item you repeat it and some other items over and over. This leads to two predictions. First, the more times you rehearse something, the better you should remember it. Second, the more items that come along after the last rehearsal of an item the worse memory should be (Atkinson & Shifrin, 1968). If rehearsal is the primary mechanism of getting material from short-term memory to long-term memory then we should see effects of manipulating rehearsal.
1) Glanzer and Cunitz (1966) again. Last time we looked at how a 30 second period of counting backwards messed up the recency part of a serial position curve. Speeding up the presentation of the list caused primacy to go down.
2) Rundus (1971). As people learned a list of words, Rundus asked them to rehearse out loud. They could rehearse any words they wanted to, but it had to be out loud. Rundus compared the number of rehearsals to recall probability and found that more frequent rehearsal led to better recall.
3) Fischler, Rundus, and Atkinson (1970). Make people rehearse only the item being presented. Primacy goes away because every item gets the same amount of rehearsal.
b. Mnemonics (coding). Besides rehearsal, you can try a number of other memory tricks to get information to transfer (some of these were discussed in the imagery unit). You might know that HOMES tells you the name of the great lakes. I still remember the number of Metro in Houston because in 1986 when I was helping my brother move they kept playing a commercial saying the number was “Dixie, drive your cows in” (635-4000). My locker combination is “toothpicks are so dirty” (26-0-30). In the next unit of the class, we’ll look at other techniques to improve transfer and see why they work.
c. Visual images. Some people are good verbal learners, some are better with images. There is evidence that bizarre imagery can have a large effect on improving memory. The trick is to imagine each item in a list interacting with other items in some creative way (again, check the imagery notes).
2. How you structure your control processes. There are three basic tasks involved: Allocate attention to the task, do some processing, decide when you’ve learned it (when to stop). Atkinson (1972) had three kinds of learners. One group received a randomized list of items. Each time a German word was presented, they wrote the English translation. Since the list was random, some of the things they rehearsed were already well-learned, and some were not learned at all. A second group got to choose what to study. If people optimize their learning strategy, then they’ll spend more time studying what they don’t know.
The interesting group for Atkinson was one for which a computer controlled the training. Atkinson theorized that there are three kinds of items when you’re in the process of learning. Some items you don’t know at all (unlearned), some you really know (learned), and some are in transition (temporary). If you rehearse, you should rehearse what’s temporary or unlearned, not what you know. The computer was programmed to pick only what was not learned. Atkinson wanted to compare this group (optimal) to people choosing for themselves and people rehearsing at random.
You can see the results here. The optimal group did the worst in training, but had the best recall one week later. The random group did the best in training, but had the worst recall. Why? The optimal group only studied what they didn’t know. This means they made a lot of mistakes in training, but they were learning more. The random group wasted a lot of rehearsal, and they didn’t learn as much. The group that chose their own rehearsals was in between. They did better than random, but not as good as optimal.
You might keep this in mind as you study for the exam. Try to learn what you don’t know yet, spend less time on what’s learned. This is also related to when you should stop studying. A lot of people stop too soon. Glenberg, Sanocki, Epstein, and Morris (1987) looked at calibration of comprehension. This is how well people think they know material compared to how well they actually know it. When people read a story and were then asked to rate how confident they were that they could recognize a statement from that story, there was no relationship between confidence and accuracy. This phenomenon is prevalent with a wide range of learning tasks and tests. What it means is that people rarely know whether what they know will be good enough for a test. Something else to consider as you study.
In answer to our first question, the reason you remember some things better might have something to do with the amount of rehearsal. The timing of the rehearsal also matters (don’t study things after you know them). If you study and study and don’t remember all of the material equally well, you might be using a poor strategy for rehearsal. There will be more to the answer to this soon.
B. Retrieval: What affects it? There are two kinds of remembering. In recall, you have to completely reconstruct the information (like an essay exam). For recognition, the information is presented with new information, and you have to pick it out (like a multiple choice exam). Obviously, recall tasks are harder. However, the same memory mechanisms might be used for both.
Let’s consider a rudimentary model of retrieval from long-term memory. There are automatic components of memory. One of these is frequency. People generally have a good idea how frequently they’ve encountered something even when they weren’t paying attention to frequency. Likewise, people are pretty good at remembering where things were located when they were encountered. For example, people might not know which page number an experiment was on, but they can remember that it was on the left page, about halfway down. People also seem to automatically encode some index of familiarity. This might be a result of processing. For example, try to resolve the meaning of this sentence:

(1) The old man boats made of wood.

If you encounter this again, you shouldn’t need to spend so much time figuring it out. This ease of processing might make you think the sentence is more familiar than this one:

(2) The horse raced past the barn fell.

The same thing applies to reading words, but it’s less dramatic. Resolving the sentences above takes time that’s measurable in seconds. Getting the meaning of “artichoke” when it’s presented at random might happen in half a second, but it’s harder than retrieving the meaning of “the.” So, if you see “artichoke” again, it should feel more familiar because it’s less difficult to access (you just figured it out a short time ago).
Familiarity is automatically encoded, and it’s automatically accessed. That’s the first step in our model. You compute familiarity and see what the result is. If familiarity is high, then you have to do more work to retrieve the item (high familiarity means you might be able to get something). If familiarity is low, there’s no need to try retrieval because there’s probably nothing there.
The second step is effortful retrieval. If it’s recall, you have to construct the cues you use to find the item. These cues can be things like “we heard about it on the day it snowed,” or they can be related to the meaning of what you’re trying to retrieve “this person tested the duration of short-term memory.” You put all of your cues together, and then pass them through memory to see if anything similar is in there. If you find something, that’s what you recall. To the extent that you get good cues, recall will be successful.
You can improve recall by mixing up the cues. A popular technique in interviewing witnesses is the cognitive interview. People are asked to reinstate the context (the room, the weather, their thoughts, etc.), report everything, no matter how trivial (this provides more potential cues), recall in a variety of orders, and take different perspectives. This technique produces 47% more information (Fisher, Geiselman, & Amador, 1989). Why? All of these tricks make more cues to use in searching during the effortful retrieval process.
For recognition, someone else constructs the cue (it’s the actual item you saw before). You need to get the familiarity for each item and then try the item as a retrieval cue. If we make an item seem more familiar, you’re likely to “recognize” it later, even if it’s the wrong item, because of its increased familiarity. This happens when police show people mug shots and then line-ups. The people in mug shots look familiar, and can be falsely recognized later.
Elizabeth Loftus (as in Loftus, Burns, & Miller, 1978) has made a career out of messing with witnesses. She’ll present a slide show of a car driving down a road. Then she’ll ask a bunch of questions. In one question, she’ll throw in a barn that wasn’t in the scene (“How fast was the car going when it passed the barn?”). People who get this question are more likely to remember a barn later. Its increased familiarity makes it easier to recognize, even though it was never there.
One last touch of this. When people construct multiple choice exams, they try to mess with your familiarity. The lures (wrong answers) will resemble the right answer to equate familiarity. This forces you to rely instead on the effortful retrieval component, and that is a better index of how much you’ve learned. For example, consider these two questions:

Question 1	Question 2
Peterson and Peterson found that the duration of short-term memory is:	Peterson and Peterson found that the duration of short-term memory is:
a. 16 seconds	a. 1 minute
b. 17 seconds	b. 1 hour
c. 18 seconds	c. 18 seconds
d. 19 seconds	d. 1 day

Guess which one’s harder?
We will see in a moment that the automatic component (familiarity) is spared when people have head injuries that damage memory, but the effortful component (retrieval) is what gets damaged.

Demonstration: You can see retrieval in action by doing the following. Write down as many of the 50 states as you can. Put a line on the page when you stop just slapping them down and start having to think about them. You should see some strategies emerge. The first few will probably be familiarity or frequency based (Tennessee). Then you might try an alphabetical strategy or a geographic strategy.

Retrieval is also studied in the laboratory with tip of the tongue states. You hear a definition (“What is the name of the instrument that uses the position of the sun and the stars to navigate?”). Sometimes, people will know they know the word, but can’t say it. This is a tip of the tongue state. When this happens, we can try to cue them to see if we can get retrieval over the hump. I might say “It sounds like sixteen.” Or, “It starts with ‘S’.” Spelling and sound cues are most effective for people in this state. (The word was “sextant”.)
So far, our discussion of retrieval has been driven by this two-stage model (familiarity and retrieval). This should lead to predictions about how certain variables affect memory. In the next unit, we’ll do a lot more here, but let’s just get a flavor.
1. Retention interval. The longer you wait for the test, the less you’ll remember. Ebbinghaus memorized lists of nonsense syllables (“NAX,” “POR,” “WEQ,” etc.). He found that most forgetting has happened after about five days. So, you forget pretty much all you’re going to forget in the first five days. Unfortunately, you forget almost everything in the first five days, so there’s not a lot to be proud of. Our model would claim that this is due to familiarity. With more time the effort you spent on the first attempt to process has faded (due to interference or decay), so you don’t have that to go by. Also, it gets harder to construct good cues for the effortful part.
2. List length. All things being equal, the more you try to learn, the less you’ll remember. This could be due to build-up of proactive interference, which might cause decreased familiarity (because there’s so much similar stuff). Or, the cues might be less effective because it’s hard to get a unique cue for each item (cue overload).
I'm going to discuss long term recency as an illustration of the "too much for each cue" problem and how context and cues can make a difference.

Demonstration: This is based on Bjork and Whitten (1974). Present a list of 10 word pairs for two seconds per pair. Instruct participants to rehearse only the current word pair. Have 12 seconds of multiplication problems before the first item and between all items. Have 20 seconds after the last item. Free recall all 20 words. Plot recall to see if we get a primacy and recency effect.

Bjork and Whitten found that the long-term recency effect was different from regular recency. This will take some thinking-back. Glanzer and Cunitz showed that having a 30 second distraction period after the list eliminated recency. Bjork and Whitten replicated this condition in their study. The 0+0 people had a regular free recall experiment. The 0+30 people had a regular experiment, plus 30 seconds of distraction. The 0+30 people had no recency, the 0+0 people did. The bottom half shows the time in the recall task the people wrote down the words. Lower numbers means earlier. The 0+0 people recalled the recency part really early, as if they were dumping out short-term memory. The 0+30 people didn't write the recency words earlier, and didn't remember them as well.
The real question has to do with the other condition. In 12+0, people had 12 seconds of distraction after each item, except there was no distraction after the last item. In 12+30, people had 12 seconds between items, plus another 30 seconds at the end. In both cases, you get a recency effect . Long-term recency isn't affected by a 30 second counting backwards period. Regular recency is.
Whether or not we expect a recency effect is a function of the interpresentation interval (IPI) (the time between items) and the retention interval (RI) (how long you wait to recall). Here's the function:

Task	IPI	RI	Ratio (IPI/RI)	Expect recency?
Standard free recall	Short (3 s)	Short (3 s)	Large	Yes
Glanzer and Cunitz (1966) increased spacing	Longer (6-9 s)	Short (3 s?)	Large	Yes
Glanzer and Cunitz (1966) 30 s counting backward	Short (3 s)	Long (30 s)	small	No
Bjork and Whitten (1974) 12 + 0	Long (12 s)	Short (approx. 0)	Large	Yes
Bjork and Whitten (1974) 12 + 30	Long (12 s)	Long (30 s)	Moderate	Yes

Glenberg, Bradley, Kraus, and Renzaglia (1983) present a chart showing this relationship. Why does it work like this? Glenberg, et al. argue in favor of a contextual explanation. The idea is that there is a continually changing context in the experiment. The context can be from the environment (PowerPoint, blower), the cognitive and affective state (here to get test grade back, bored), and the experimental task (I wish this was Psychosex). Some parts of the context are changing faster than others. When you have a short IPI and short RI, the recent items share context with the test and that context can be used to help retrieve them. When the IPI is short and the RI gets longer the test context is less similar and is a less good cue. So, short IPI short RI gives you recency, short IPI long RI reduces it. When IPI is long a similar pattern emerges (I'll go over the chart in class).
All of this is to get to the point that the context can matter. Longer lists at one sitting provide less variable contexts and give you fewer potential retrieval cues.
3. Serial position. Remember the Glanzer and Cunitz work with free recall. The position in the list you learn will affect memory. First items are pretty good (primacy) and last items are really good (recency), the rest are pretty crappy. Primacy is due to rehearsal, which could affect familiarity by giving you more practice. Recency is discused above.
4. Type of test. Recognition tests are easier than recall tests. Recognition tests provide you with the cues. Recall tests don’t. Constructing good cues is most of the work.
One last bit here: Glenberg and colleagues have looked at the effect of suppression on recall. They asked participants to recall obscure facts (“Is Kiev in the Ukraine?”). They surreptitiously videotaped the participants to see if they closed their eyes or looked away. The harder the recall (indexed by more time trying or errors), the more likely people were to look away. The hypothesis is that memory is driven most by what’s presently in the environment (what it’s good for is telling you what to do now). To kick memory out of your present environment you have to suppress the environment to make new cues. So, getting in a quiet place with few distractions may also help.
I can add some to the answer to our questions:
1. Why is it hard to learn some stuff after studying? These factors. You have to have a good chance of experiencing familiarity or constructing a good cue. Some stuff doesn't seem to get in because you failed to do that.
2. Where did Calculus go? It's probably in there, but the right cue is hard to come by.

Top

IV. The parts. We’ve been treating long-term memory as a unitary thing. In reality, it’s probably not. Tulving (starting in 1972) is an advocate of multiple memories. He looks at neuropsychological data and data from experiments to conclude that types of memory are separate. A partial model is presented below:
A. Declarative memory. You can declare these (verbalize them). Also called explicit memory.
1. Semantic memory: Facts. They’re processed, compiled knowledge. You don’t know where you heard them, you just know them.
2. Episodic memory: Autobiographical information. Organized around time and around you as the experiencer of them. Everything starts here and some stuff can migrate into semantic. So far this is what we've been discussing.
B. Non-declarative memory. You can’t declare these, you might not even know about them. Also called implicit memory.
1. Procedural: How to do things.
2. Implicit: The residual effects of processing stuff. This would be familiarity in our model.
C. Declarative vs. non-declarative: Declarative memories come with a feeling of remembering, you know when you’re having them, and they can be damaged by brain damage. Non-declarative memories happen without awareness, last a long time, and can be spared even when declarative is wiped out. Note the similarity between non-declarative memory and automatic processing.
D. Are there different memories or different processes? Tulving likes different memories. Jacoby argues for process (Jacoby & Dallas, 1981). Explicit memories (recall and recognition) are conceptually driven, and are accessed with recall strategies. They’re likely to be episodic memory. Implicit memories are procedural, data driven (familiarity) and are accessed by indirect tasks (fragment completion).
Affect familiarity by manipulating modality. If you heard the words, the written form won’t be as familiar and vice versa. This should affect indirect memory, but not direct memory. It does.
Affect retrieval by having people solve anagrams (EMTLA) or read (METAL). The anagram should improve direct measures over reading, but shouldn’t affect indirect measures. It does just this.
So, instead of two kinds of memory, it looks like two kinds of processing.
Evidence for separate kinds of memory comes from amnesia. People can lose declarative but not procedural. If it’s just processing, this is hard to explain. Squire has shown that damage to the hippocampus is what makes it hard to form new declarative memories without harming the formation of procedural memories (HM).

Top

V. Segue into a new idea. Let’s begin the transition to a more modern approach to the cognitive system.
A. Segue. Return your attention to the old box model. We’ve now gone over all of the boxes. But, we’ve diversified a bit. Our new model looks something like this:

Attention can operate pretty much anywhere. Processes (like pattern recognition) take place primarily in working memory. This is a complicated system, and we have pretty good support for each of the divisions. But, does that mean memory really works this way?
A different approach is to look at processing (instead of representation). What you do with information has a big impact on how well it’s remembered. Let’s consider for a moment the extreme of this viewpoint: none of these memory divisions are real. Instead, memory is entirely a function of processing. If you do very low level tasks (basic sensation, a little perception), then you get what looks like a sensory register. If you do more, you get short-term memory, etc.
Levels of processing: Material can be processed at a shallow level (attention to physical features) up to a deep level (attention to meaning). Generally, the deeper the processing the better your memory will be.
B. Kinds of processing.
1. We’ve looked at the role of rehearsal in memory. Rundus’ data indicates that more rehearsal leads to better memory, and this rehearsal seems to be the source of primacy effects. But, is all rehearsal the same? No. There are two types.
a. Type I rehearsal: This is called maintenance rehearsal. This is what you use when you’re trying to hold something in the articulatory loop temporarily (like a phone number). This type of rehearsal is not likely to lead to lasting memory. When you stop maintaining the item, it goes away.
b. Type II rehearsal: This is called elaborative rehearsal. Instead of merely repeating the item, you try to elaborate it and increase the number of memory cues that can access it. You might use a mnemonic device or mental imagery. This kind of rehearsal leads to lasting memories.
Let’s look at the effectiveness of maintenance rehearsal.

Demonstration: I'll present a memory task here to look at the impact of increasing rehearsal.

Usually, more maintenance rehearsal won’t improve recall. Any rehearsal is better than no rehearsal, but varying the amount won’t help.
Our demonstration was a replication of a study by Craik and Watkins (1973). They varied the rehearsal time from 0 to 12 seconds. Over the course of the experiment, participants saw 27 lists. For the overall recall, 0 seconds of rehearsal led to 12% recall, 12 seconds led to 15% (not different). Again, increasing the amount of time that you’re merely maintaining information is not going to improve memory.
We can add more to the answer to the first question. No matter how much you study, if it’s just maintenance rehearsal, then you won’t get much out of it.
2. If more maintenance isn’t going to improve memory, what will? Elaborative rehearsal.

CogLab: We'll look at our data from the levels of processing demonstration.

Demonstration: I’m going to present a list of words. I’ll also distribute a task sheet. For each word, you circle “yes” or “no” based on the question on your task sheet. We'll look at the specific nature of the recall to see what we can find out about the different approaches.

This demonstration replicates a study by Hyde and Jenkins (1969). They had people make these same types of judgments. Their meaning participants recalled 16.3 words out of 24. The ‘e’ people only got 9.4. We know meaning people really attended to meaning because of the clustering data. Meaning people had 68% clustering, ‘e’ people only had 26%. (Note: Compare to our data.)
In the next set of notes we’ll fully explore this focus on processing over representation. We’ll try to develop a processing account of memory.

Top

Cognitive Psychology Notes 6
Will Langston

Back to Langston's Cognitive Psychology Page