9 Lab 9. Recall, Recognition, and Encoding Specificity: I’ve Seen You Before, But I Can’t Remember Where

Lab 9. Recall, Recognition, and Encoding Specificity: I’ve Seen You Before, But I Can’t Remember Where

COGLAB Exercise 28

Introduction

Psychologists who study memory generally recognize three stages in memory:  1) an encoding phase, where the information is first learned and prepared to be remembered; 2) a storage or consolidation phase, in which the information is allowed to “gel” in the brain.  Drugs, alcohol, and head trauma can all disrupt this consolidation phase; and, 3) a retrieval phase, where the stored information is brought back for use.

Much of what we call “forgetting” is really just an encoding failure.  In a recent study, about half of the people in Britain did not know which way the figures on their coins faced.  Did they “forget?”  Not really.  They just never encoded that information to begin with.[1] In this lab we’re interested in looking at retrieval failures.  We all have times (usually during tests) where we know we have the answer to a question, but also know we can’t retrieve it right now.  Sometimes, this is because the retrieval conditions aren’t right, as this lab will demonstrate.

The two primary ways psychologists measure memory accuracy are recall and recognition.  Recall is the act of trying to retrieve information given no cues to help you search.  A type of recall called cued recall is similar, but in a cued recall task you are given some information to use as clues, to aid your recall.  A recognition test involves giving you several alternatives, and asking you to pick the correct answer from among the alternatives.  If I asked you “Which actor played Jack Sparrow in the “Pirates of the Caribbean” series?”, your subsequent memory search would be a recall test.  If I said, he’s also portrayed Edward Scissorhands, Willy Wonka, and the Mad Hatter, I’d be giving you a cued recall test.  If I put the names, George Clooney / Tom Cruise / Johnny Depp / Brad Pitt, in front of you, and had you select the correct choice, this would be a recognition test[2].

This experiment will allow us to test two different hypotheses concerning memory.  The first is a theory that has been around for a long time–the association hypothesis.  This basically says that memories are formed by making associations between words or concepts, and anything that enhances those associations will improve memory.  The second hypothesis is known as the encoding specificity principle.  As we’ll see later, this looks both at the conditions under which the original memory is first learned, and also the conditions under which that memory is retrieved.

The association hypothesis basically assumes that memory is dependent upon the associative connections between the kinds of units which exist prior to the experiment.  That is, words and concepts that are associated prior to the experiment will still be associated after the experiment.  If you have an association between “chicken” and “egg” before the experiment starts, that association is presumably unaffected by whatever happens during the experiment.

By contrast, the encoding specificity principle says that conditions under which you form your memories are of importance.  If you learn something in one set of conditions, you’ll remember it better if we can reconstruct those conditions during retrieval.  If you learn something while you are in one physiological or emotional state, you’ll better retrieve those items if we can reconstruct those original physiological or emotional states.

To illustrate the two hypotheses, let’s look at two examples.  The words “DOG” and “CAT” are already strongly associated in your memory.  You’ve heard those two words together a number of times.  The association hypothesis says that because you’ve heard those two words together many times, they are already associated in your memory.  Thus, during retrieval, “CAT” would be a good cue to help your retrieve “DOG.” However, say that during the encoding phase, you were thinking about the word “FLEAS” while you were exposed to the word “DOG.”  (We could set you up to think about  “FLEAS” in a number of ways–either by telling you that a dog was scratching, or that a dog was not allowed in the house–or by telling you directly that you should associate “DOG” and “FLEAS” for this trial.)  If that were the case, then the things you were thinking about while you were learning the new association should be better cues.  If I gave you the cue “ITCH” you should do better, despite the fact that the words “DOG” and “ITCH” typically only have a weak association.

The concept of state dependent learning is a good illustration of the encoding specificity principle.  If you learn something in class while you are very depressed, for example, you’ll better retrieve that same information while you are also depressed.  If you drink two pots of coffee while studying for an exam, you’ll do better on the exam if you also had coffee that morning.  The coffee itself won’t help you perform better on the test–but if you DO drink that much coffee while you study, you’ll likely do better if you recreate those conditions during the test.

These two hypotheses allow a strong test of how memory works.  If you’ll look back at the conditions in the experiment, you’ll see that in the encoding stage, we always presented the weak associates.  That means that there was some association of the cue and target words before the experiment began, but only a weak association.  However, during retrieval, sometimes we gave you the strong associates.  That’s where the real difference between the hypotheses emerges.  If the association hypothesis is correct, then you’ll do best when given strong associates during retrieval.  If the encoding specificity principle is correct, though, you’ll do best when we recreate the same conditions under which you learned the items initially.  Therefore, you’ll do best with the weak associates during retrieval, since those are the conditions in which you learned the target items to begin with.

One final prediction of the encoding specificity principle is the comparison of conditions 2 (STRONG CUES) and 3 (NO CUES).  While the association hypothesis assumes that strong associates are always better cues for retrieval, the encoding specificity principle predicts that since those cues were not stored during encoding, they won’t help during retrieval.  A similar experiment was first performed by Tulving and Thompson (Tulving & Thompson, 1973).  Our experiment is more similar to that performed by Watkins and Tulving (1975).

Tulving’s and Thompson’s experiment (1973) involved four phases, shown in Table 9.1.  The first two sets of 24 pairs were really just designed to get you familiar with the situation, and also to induce you into a certain strategy–subjects were supposed to try to remember the pairs by forming associations between the cues and targets.  The first phase, then, was the encoding phase, where subjects learned the third list of cue-target pairs.

After the first phase, the encoding phase, subjects were given certain words and asked to generate associates to those words. They picked those words carefully, though.  They chose words that would be very likely to produce, as associates, words that had been target words in phase I.  For example, subjects were given as a cue-target pair the words HEAD-LIGHT, and were asked to remember the word LIGHT.  During the free association phase, they gave subjects the word DARK.  DARK and LIGHT are very strong associates, so subjects were likely to generate the word LIGHT.

After the generation task, subjects were told that some of the words they had generated had actually been studied in Phase I.  They were told to recognize those words from the list they had generated.

Finally, subjects were given a cued-recall test.  They were given the cue words from Phase I, and asked to generate the appropriate target word.  The steps in this procedure are outlined in Table 9.1 Most people assume that recognition is easier than recall.  This experiment was specifically designed to show you that this is not always the case!

Background of the Experiment

Tulving (1972) has distinguished between semantic memory and episodic memory[3].  Both are considered to be long-term memories, but they differ in a number of respects.  Some of the differences between semantic and episodic memory are shown in Table 9.2.

You are certainly familiar with the differences in these kinds of memories.  Episodic memories are your “personal” memories, where you are a vital part of that memory.  You remember your first kiss, the day of your graduation, your first day at Baylor, etc.  All of these memories are organized around temporal events, and the chronological order of those events is very important.

Table 9.2.  Comparison of Semantic and Episodic Memory (from Tulving, 1983)

Semantic memories, by contrast, are memories that are now free from context.  For example, you know that George Washington was the first President of the United States, and that Bill Clinton was President before George W. Bush.  You also know how to compute the average of a set of numbers, as well as the name of the person who came up with the term “semantic memory.”  These are impersonal memories–ones relatively free from the context in which you learned them.  They are not organized in a temporal fashion, but in some kind of conceptual way.  Did you learn the name of Montana or Idaho first?  You probably don’t know, and it also doesn’t matter.  On the other hand, knowing that you went out with person X before you went out with person X’s roommate is important, as the Jerry Springer show readily demonstrates.

For our purposes, one of the most important distinctions between the two kinds of memory is the form of their organization.  As was mentioned before, information in episodic memory is temporally organized.  If two episodic memories are psychologically “close” it’s probably because they occurred close together in time, like your first class at Baylor and your first fraternity or sorority meeting.  Semantic memory, by contrast, is organized in conceptual terms–on the basis of meaning, or appearance, or function.  Lincoln and Kennedy may be “close” in your semantic memory because both were assassinated, or because both were heroic presidents.  SMU and TCU may be close together because they are both (expensive) private colleges in the Dallas-Fort Worth area.  (They are probably “close” for a number of other reasons, too).

The episodic/semantic distinction has been the focus of a great deal of research over the past few decades.  In fact, it may be among the most hotly debated topics in all of memory research.  Regardless of the final outcome of the debate, the episodic/semantic distinction has certainly helped along our understanding of a number of phenomena.  For example, when we talk about somebody suffering from amnesia, or having “lost their memory,” what we really mean is that they have lost their episodic memory.  We would not be too surprised if someone wandered into the hospital and said “I don’t know who I am or where I live.”  By contrast, we would be very surprised if they came into the hospital and said, “I’m John Smith of Pittsburgh, but I’ve forgotten my multiplication tables and state capitols.”  The first case would be an example of someone losing their episodic memories, while the second case would be an example of someone losing their semantic memory.  We see people of the first sort, but have yet to see someone of the second sort.

Mitchell (1989) has used the semantic/episodic memory distinction to explain differential effects on memory in the aging process.  Mitchell’s research shows that memory loss in the elderly is typically restricted to a mild impairment in the formation of new episodic memories.  In fact, Mitchell’s work suggests that if anything, semantic memory actually improves throughout the lifespan!

Episodic memories are quite context-sensitive.  When you learned something in one set of circumstances, you typically remember that information more easily when you recreate those original circumstances.  Have you ever met someone in an unfamiliar place and not recognized them?  Say, you know a person quite well in a classroom situation, but have never seen them out of that context.  Then, one day you run into that person in the airport, or in the mall, and you can’t remember their name.  Or even if you do have a vague sense of knowing them, you can’t remember where you know them from.  This happens quite a bit to professors.  We are used to seeing our students in the classroom, and recognize them quickly and easily if we meet them on campus.  However, if we run into those same individuals in another situation, say in a restaurant, we’ll often forget that person’s name.  The idea of the “absent-minded professor” has some grounding in cognitive theory!

This is explained nicely by the encoding specificity principle.  Remember, this principle says that we not only encode new information, but we also encode the context in which we learned that information.  Retrieval is enhanced if we can recreate those same conditions.  If I met you in class, and learned your name in those circumstances, but then see you in the airport, I have a different set of cues at retrieval than I did at encoding.  My memory for your name will be less accessible.

Virtually all memory theories can account for context effects like these.  However, they differ in exactly how they represent context in memory.  This lab was designed to test two of these theories, called tagging theory and episodic theory.  Episodic theory is a direct extension of encoding specificity developed by Tulving.  Tagging theory was actually a precursor of episodic theory, and presents what was at the time a more conventional approach to the representation of context in memory.  Tagging theory says that the way an item’s context is stored in memory is by attaching some kind of marker–a “tag”–that stores that information.  If you learned the item LIGHT in the third list, you’ll have a tag that says, “This item was learned in the third list” accompanying that memory.  As you can see, there is no distinction between semantic and episodic memories with tagging theory.

Episodic theory assumes that every new encounter with an item is stored by its own individual memory trace.  You can learn the item LIGHT a number of times, but each time you do, you’ll have stored a new memory trace for that event.  These memory traces can actually be thought of as a subset of pre-existing semantic traces.  You knew the concept LIGHT before the experiment (it was in your semantic memory), but on this particular trial, this semantic item was learned–you would therefore form a new episodic trace of that event. The cue-target pair DARK-LIGHT might bring to mind things like sunshine, light bulbs, or shades of clothing.  If you study the same word LIGHT in the context of a cue like HEAVY-LIGHT your representation would be very different.  Now, your episodic trace might have things like “carry” or “weight” with it.

Because of the differences in how items are stored, the two theories make different predictions about retrieval.  According to the encoding specificity principle, you will do best during retrieval if we can recreate the conditions under which you first learned something.  This will be true regardless of whether the test is a recall, recognition, or cued recall test.  Episodic theory makes no distinction between the memory tasks.  Since the retrieval process is thought to involve utilizing information available during a retrieval episode, and that information is the same for both recall and recognition, there should be no differences.

Tagging theory, though, is based on a different assumption.  Tagging theory basically puts forth what is known as a generate-recognize model of retrieval.  In this theory, recall is a two-stage process:  the person, trying to recall information, must first get that information out of memory (the “generation” task).  After the information is generated, the person must then decide if the information is correct (the “recognize” process).  Notice, though, that a recognition test involves only the “recognition” stage.  For example, on a multiple choice test you do not have to generate the correct answer, you just have to recognize it as being among the possible choices.  On an essay test, you generate your own answer, and then presumably recognize it as the correct one.  (Or perhaps an elaborate guess).  The generate-recognize model of retrieval would make the very strong prediction, then, that you should be able to recognize everything you can recall.  If you can recall something, then both the generation and the recognition processes have succeeded.  If you are only asked to do the second phase, the recognition process–in other words, perform a recognition task–that should be equally successful.  There may also be some times when the generation process fails, but the item can still be recognized.

This experiment tests this prediction.  We made the recognition test very difficult, and the recall test maximally efficient by presenting during retrieval the cues that you learned in the encoding phase.  Tagging theory would predict that the recognition test should show better performance–all you are asked to do is the editing task.  Presumably, there may be some items you cannot generate, but if you can generate it, you should be able to recognize it.

Episodic theory, though, makes very different predictions.  Because of the context-sensitivity of the process, you may find that the recognition test is more difficult than the recall test.  After all, the recognition test provides you with a different set of retrieval cues, cues that were not present during encoding.  The cued-recall test, though, presents to you the same cues at retrieval that you had during encoding.

Questions for Lab 9

1. What are the independent variables? What are the dependent variables?

2. Analyze the graph of your results, both individual and class – Do they agree with the predictions of tagging theory or encoding specificity theory?

3. Do cues always help memory study and recall? Explain your answer.

4. Using the findings surrounding encoding specificity, what suggestions about studying would you give someone who wanted to improve his/her performance on tests?

5. Students sometimes claim to study differently for different types of exams. That is, if they know they will be given a multiple choice test (a recognition test) they say they study differently than if the test is going to be an essay exam (a recall test). Do you think there is any truth to this?  Why or why not?  Do you think it is possible that the results of this experiment are due to different types of study for the different kinds of tests?  Why or why not?  Is this consistent with encoding

specificity?

6. To get a driver’s license, one usually must pass a written exam as well as an in-car driving test. From what you know about encoding specificity, why is the in-car test so important?
7. Morris, Bransford, and Franks (Morris, Bransford, & Franks, 1977), in a study designed to test the Levels of Processing (LOP) hypothesis, found some intriguing and counter-intuitive results. Like we discussed in class, Morris et al. had subjects perform either a semantic encoding task or a rhyming encoding task.  After this encoding phase, one half the subjects were given a standard recognition test.  The other half of the subjects were given a rhyme recognition test; instead of recognizing the previously learned items from a list of distracters, these subjects had to recognize all the items that rhymed with the original items.  If, for example, the word TOY was one of the target words studied in Phase I, the subject was to pick out a word like BOY in the rhyme recognition test, since BOY rhymes with TOY.  As you would guess, the standard recognition test produced results consistent with the LOP predictions:  subjects who processed words at the semantic level demonstrated better performance.  However, in the rhyme recognition test, the results were exactly opposite–those who had done the rhyme encoding task did better on the rhyme recognition test.

Given what you know about the LOP hypothesis, and the encoding specificity principle, provide an account of these results.  What implications do these results have for the LOP hypothesis?  What does this say about “depth” of encoding?

Data Sheet for Lab 9

Encoding Specificity

Name:

Report Mean Percent Recalled:

Lure: __________

Graphs for Lab 9

Recall, Recognition, and Encoding Specificity

Individual Data

NAME:

Recall, Recognition, and Encoding Specificity

Group Data

Turn this graph in along with your lab

NAME: