share interactive transcript request transcript/captions live captions
download
|
MyPlaylist
SPEAKER 1: This is a presentation by Human Development Outreach and Extension at Cornell University.
SPEAKER 2: This is a collaborative project between folks above and below the equator. The people above the equator are the first and last authors, and the ones below are in the middle. I'm from the north end of campus, which means we've been very much concerned with the quality of evidence, the accuracy of testimony over the years. Probably the work that people know the most about our work is work that's concerned with vulnerable witnesses, such as children, and also with the effects of suggestion on the accuracy of testimony and other forms of evidence.
I suppose the overriding idea here about evidence in criminal cases is it ain't CSI. The great bulk of the evidence in criminal cases, of course, is what comes out of people's mouths. And that means memory reports and the decisions that people have made about the content of their memories.
So we would like to study the kinds of variables that are most important in determining how distorted witnesses' reports are. A common feature, of course, of memory reports about crimes is that some sort of emotion is involved. And that's the kind of topic that I'm going to talk about today. Criminal events are emotional in and of themselves, but also the events of crimes take place against that contextual backdrop of high levels of emotion.
And emotion can also occur in the circumstances in which you get information from witnesses and victims and suspects, too. Trial testimony, interrogations, police interviews, and so on are inherently emotional. So we want to know to what extent does emotional content distort memory.
Let me just mention a few basic scientific distinctions that are fairly common in research on the effects of motion on memory. One is the two-dimensional kind of model of emotion, that it has at least a couple of important components. That is to say emotional events tend to be valence. That is to say they're definitely positive or negative. And they also tend to be physically arousing.
They're also known, the valence and arousal components, to affect various kinds of memory performance differently. So they're dissociated in their effects on memory performance. Of the various kinds of memory tasks, there have been neuroscience studies suggesting that there's different areas of the brain that are activated by changes in valence and changes in arousal. As far as arousal is concerned, not surprisingly, the left and right amygdala. As far as valence is concerned, right and left dorsolateral-- ventrolateral prefrontal cortex and medial PFC, as well.
There's also been some recent research that has suggested that specific kinds of emotion, such as anger, and happiness, and sadness, and so on may have very specific cognitive effects of their own, but that's not a topic I'm going to talk about today. Let's talk about some classic ideas about what the effects of emotion might be. One is the old Dr. Johnson effect. When a man knows he's to be hanged in a fortnight, it concentrates his mind wonderfully.
The scientific version of this hypothesis is that when you have negative content, it somehow focuses attention on target events, and really improves your verbatim memory for those events, and kind of suppresses distortion. Another kind of idea about memory is what I'll call the go with the flow idea. It's about positively valence events. You can find it discussed in a 2005 article on PsycNET by Storbeck and Clore.
A couple of years ago I, also attributed this idea to our colleague Alice [? Eisen ?] here at Cornell. She became apoplectic about that, so I don't attribute it to her anymore. The basic idea is that positive content enhances meaning extraction. You sort of relax. You sort of plumb the meaning of things, become creative, and so on. At the level of memory performance, of course, what that means is it's going to distort your memory. It's going to create false memories, but it probably won't affect your true memory for events that actually happen.
So what these couple of classic ideas add up to is that memory will probably be enhanced by negative emotion, and it'll probably be distorted by positive emotion. I'll just add here down at the bottom that something that's missing, of course, here is, so what's the role of arousal in these hypotheses? It's sort of missing in action.
So we have this theory about the relationship between memory and reasoning that's called fuzzy trace theory. And most of our work is attempts to test hypotheses from this theory. So let me just say briefly what the theory is and then what its take on emotion is. It's a dual trace theory of memory. It assumes that we store two basic traces of our experience, verbatim traces of the surface form of our experience. So we episodically code the surface form of our experience. But we also process the meaning content of our experience, and we episodically code the meaning content of our experience, as well. Just traces.
So if you need a handle on this, just that these two are, in fact, different things. Just imagine you study a categorized list of words consisting of maybe 10 animal names, and 10 color names, and maybe 10 articles of furniture. And then after you do that, we come back in about a month and we ask you, well, what did you study. And you can't remember a single word, but you'd probably be able to remember all the categories. So that's the distinction between episodic coding of the gist of the experience and its surface form.
Our take on emotion is that valence is mainly a gist dimension. So the effect of valence is mainly to impact gist processing. And it's conceptual or gisty in two senses. First of all, positive and negative valence are concepts in of themselves. Your junior prom was the pits, for example. But valence may also stimulate you to process the semantic content of events, as well.
So you could think of valence in that sense as being a kind of levels of processing manipulation that stimulate semantic coding. And then we think of arousal as being mainly a verbatim kind of interference manipulation. So this is sort of the opposite of the Dr. Johnson idea, that heating up the amygdala kind of generates off-task noise. You might want to think of it on analogy to dual-task deficits as a kind of dual-task deficit.
So just to summarize, a negative valence, we think, stimulates semantic processing relative to positive or neutral valence, producing stronger gist memories of experience. So the idea about memory distortion that we're going to look at, actually, today in some experiments later on is that memory distortion is really maximized by negative valence, and that this is a semantic kind of effect. But later on at the very end of the talk, we're going to have to have some caveats about a couple of vulnerable populations, namely the elderly and children that will limit this generalization.
So how do you study memory distortion in the memory lab? There's essentially two kinds of ways. You put the emotion in the context. And then you have people perform a neutral memory task of some sort. Now, we're interested in memory distortion. So that means that all of our memory tasks have to measure false memory, as well as true memory, and the emphasis is going to be on false memory.
So one thing you can do is you can put the emotion in the context by inducing a mood of some sort, often, say, with music. And then you can shoot people into a neutral memory task, and measure true and false memory, and look at the effect of the mood that you put them in. Another kind of technique, which is the one we're going to talk about today, is to put the emotion in the information that has to be remembered, which is like it is in a crime, for example.
The work I'm going to talk about today is going to consist of a couple of experiments, one of which was done in Brazil, in Puerto Allegri, in Portuguese, and a parallel experiment that was done in English with American undergraduates. The basic design is that subjects study a series of what are called emotional DRM lists for Deese-Roediger-McDermott lists, which are a technique that I'll talk about in a minute that induces very high levels of false memory in the lab, and then to respond to a certain kind of recognition test that's called a conjoint recognition test, which has the feature that allows you to generate a data matrix over which you can define a mathematical model. And then you can extract measures of what's actually going on at the level of memory processing.
What are emotional DRM lists? Well, a DRM list is simply a list in which you take a familiar word, like window or chair, and then you go to norms of word association, and you find its first 10, 15 forward associates. Then you have people study the associates. You don't show them the cue word. And what happens is if you then give them a recall test or a recognition test, they remember that missing cue word at very high levels. So if you have people listen to nurse, hospital, sick, ill, physician, stethoscope, what are they going to think of when they do a recall test?
AUDIENCE: Doctor.
SPEAKER 2: Doctor. Bingo. But it wasn't there. Those are the lists. The riff on these when we talk about emotional DRM lists is we use exactly that. They're constructed in exactly the same way, and the missing words are the measures of false memory. But we select words that differ in valence. So some negative words, generating words like hungry, lie, thief. They're forward associates.
Some neutral words like fruit, teacher, window, they're forward associates. Or some positive words, like baby, god, hug. They're forward associates. The people study the forward associates, then we measure memory for both what they studied and for the missing items, as well. Arousal in this particular experiment today, these two experiments, what we're going to be doing is controlling arousal, rather than leaving it flopping around and looking just at the effects of valence with arousal control.
Conjoint recognition tests-- what are they? They're like standard recognition tests. In standard recognition tests, you shoot probes at people and you say, are they old or are they new? And in a conjoint recognition test, what you do is you do that, but you also ask two other kinds of questions for other kinds of probes. You say, is this an item that's not new, but is related to something you studied on the list? That would be like physician. If you studied physician, doctor would be a word that's related to physician.
And another kind of question-- that's called a gist question-- another kind of question that you use is what's called a verbatim plus gist question, where a probe comes up and you say, is it either something you studied or something that has the same meaning, is related to something that you studied? This allows you to generate a three-by-three matrix, three kinds of probes. Things you studied, things you didn't study, but are related to what you studied, things you didn't study that are unrelated to what you studied, and then these three kinds of questions.
And that gives you enough data to generate a mathematical model that allows you to measure memory processes. Specifically for true memories and false memories, the relative contributions of the goodness of your verbatim memory and the goodness of your gist memory for the semantic content of what you study.
Now, what I did was, since I didn't want to blind you with a lot of slides shown at fusion speed, was I just pooled the data of these two experiments in which to remind you again we had people study a lot of emotional DRM lists that differ in valence-- negative, positive, neutral-- with arousal levels controlled. And what we want to look at is the effects of valence on true and false memory.
It's the first pass here. What we're going to do is we're going to compute signal detection statistics on the Brazilian and the American data. So we're going to compute d prime values and c values. The d prime values are the bars that are above the line, and the c values are the bars that are below the line. I will not talk about criterion setting. I won't talk about c. I'll just say, if you look, you'll see that people's criterion setting is somewhat liberal in this particular experiment.
What I want you to do is just look at the d prime values for the three kinds of valence for the two kinds of items. True memory items, things you actually studied. False memory items, things that weren't there, but you think might have been there. And I've got another slide on that just to bring that up. And this is just the top half of that slide. What you see is if you look at the black bars, what the black bars are is d prime values for true memory. The accuracy of your true memory jumps as you move from positive valence to neutral and negative valence. The difference between positive and neutral and positive and negative is reliable, but not between neutral and negative.
Now, if you look at false memory, what this d prime value here measures is your tendency to think that a positive, or a neutral, or a negative word that wasn't there, but is related to what was presented, was, in fact, there. What you'll see is that that tendency just increases smoothly as you move from positive, to neutral, to negative.
Now, I want you to notice two other things. If you look at the pairs of bars and their relative positioning, what that tells you is what's the relative accuracy of true and false memory, or what's net accuracy? And what you see is that for positive valence and for neutral valence, memory is net accurate. False memory d prime is lower than true memory d prime. But for negative valence, the opposite is true. If flip-flops for these materials.
So here's a summary of what the d prime data show about valence-induced memory distortion. First of all, false memory is elevated by negative valence relative to neutral valence. False memory is lowered by positive valence relative to neutral valence. Net memory accuracy, the d primes for true and false is increased by positive valence relative to neutral. Net memory accuracy is decreased by negative valence relative to neutral. And when valence is negative, memory for these materials is net inaccurate. The sign of the difference between the d prime for true and false memory is, in fact, negative.
I've got a couple of minutes left? Yeah, a couple of minutes left. OK, you'll all breathe the sigh of relief. Let me just skip quickly through the modeling data. The mathematical modeling data show a very simple story, that on the false memory side, what's happening is gist memory, your memory for the semantic content of experience, which supports errors is increasing considerably as we move from positive, to neutral, to negative materials.
What's also increasing is-- probably isn't a good way to put it, but what's decreasing as you move from positive, to neutral, to negative is your ability to get in and use your verbatim memory to suppress false memory responses. True memory valence has much less of an effect than it does on false memory. The strength of gist memory tends to increase as we move from positive to neutral probes, but not after that.
And there's basically kind of inconsistent effects on verbatim memory. So there's consistent effects on gist memory and verbatim memory if you're talking about false information, talking about distortion, but small and more limited effects on the true memory side. So the story so far, at least in our lab for negative valence and memory distortion goes like this. Negative valence increases false memory hugely and decreases net accuracy. And that's because true memory doesn't increase nearly enough to compensate for the increases in false memory.
It's mainly a gist memory effect. So negative valence is like a really good gist. And you're really processing meaning content in negative valence. And the memory distortion effects, though, from valence in our data are valence-specific, though. Positive valence actually reduces false memory. It actually protects you against false memory. The distortion is a negative valence specific effect in our data.
Arousal obviously needs to be folded into this whole picture of valence effect. We obviously want to know if arousal interacts with valence, amplifies or suppresses valence effect. We're working on that. We've got a big norming project underway with these kinds of materials, in which we are able to effectively manipulate arousal against valence. And right now, it looks like arousal-- what is that?
AUDIENCE: Go ahead. Finish your sentence.
SPEAKER 2: OK. I couldn't read it, sorry. It looks like arousal does have an effect when you're dealing with positively valenced materials. It will tend to determine the influence of positive valenced materials, but not negatively valenced ones.
I just want to end with two caveats. These kinds of negative distortion effects in young adults, there's two kinds of populations that we worry about as being vulnerable ones in the law-- children and the elderly. These kinds of effects probably won't generalize to those problems. We know that older adults tend to preferentially process positive information, rather than negative information.
So the kinds of effects that I've reported here for college students are probably going to flip-flop in the aged. And in our lab and in Joe Michael's lab here, we, in fact, have preliminary data that, in fact, that is true for the aged. With children, you probably won't get these kinds of valence effects to any great degree at all, because they're semantic kinds of effects.
And what we know with other kinds of semantic memory distortion in kids that when it's a form of memory distortion that involves penetrating the meaning of things and basing your memory performance on that, kids just don't do that to the same level as high school students and adults. So probably valence effects, whatever they may be, will be very limited in children. OK, thank you.
AUDIENCE: Yeah, discuss it [INAUDIBLE].
SPEAKER 2: We do run a lot. This is just an outtake from a large program of research, and we do run out lots of other paradigms, narrative stuff, pictures, videos, and so on. And our theoretical ideas are based on common findings from all those paradigms. So they vary all the way from these kinds of wordless tasks that I talked about today to things that are very cognitive and very comprehension-oriented and so on. So that's the issue of paradigms.
There has been some work done on-- you mentioned the issue of motivation to suppress false memories. There has been some work done on that. And it works very marginally, particularly if you motivate people and encourage them to suppress false memories on the memory tasks themselves. So you get an effect, but it's a small effect.
Where you get a much larger effect is if you warn people that they're going to be apt to generate false memories in a particular task before they encode the materials. And then you look at how they perform on memory tests later on. Once again, that's not a huge effect, but it's a much larger effect. And it appears that they code the information differently in those situations. So that has been looked at.
And you mentioned social psychology. Some of these paradigms are very similar to what social psychologists work with, but I now want to jump farther out yet and say with respect to the issue of real crime scenes and that sort of thing, from maybe everybody's perspective, those are much more costly kinds of experiments, and more complex, and more resource-consumptive kinds of experiments to run.
So our tendency is to try and get our facts right in the laboratory under very precisely controlled conditions, and then move in the direction of the forensic situation to see what we've learned in the laboratory that can be applied to help us understand what's going on forensically, and to do that in two kinds of ways.
First way is simply explanatory. Are there things that we learn in the laboratory that allow us to explain salient findings out there in the forensic literature, but then do they allow us to conduct new kinds of experiments in forensic situations as well? So it's always bench to bedside, as my wife likes to say. So this is the bench stuff.
What you were talking about with respect to emotional memory for crime scenes, I need to draw some distinctions. The effect that you were talking about is really what we call the distinction between central and peripheral events. And people tend to remember central events of crime scenes better than peripheral events. The question is whether that's somehow got something to do with crime, because that is true in general. In any kind of memory situation that has a gist to it, that has a story to it of some sort, people remember the central events better than the peripheral events.
With respect to my interest here in emotion, although those situations are emotional, in some sense, there's no systematic attempt to manipulate the variables, like we were talking about in the talk yesterday morning, the important components of emotion. So what the components are that are producing those kinds of effects is not exactly known unless they're manipulated.
Then finally, it's not clear-- when people remembered central events better than peripheral events, for example, it's not clear that that's not somehow a different and more accurate memory than a false memory or a memory distortion. What do I mean by that? People have argued, like Ralph Adophs argues, for example, that that memory superiority is, in fact, a gist superiority is what it is, that you remember the central events very well not because you necessarily have a vivid, verbatim memory of them, but because they fit so well with the gist, that like the word doctor on the doctor list, you just basically wouldn't remember it.
If you measure memory for other kinds of events in that situation, they will be falsified. And also, you can falsify memory for central events, as well. It wasn't thought that you could do that for some years, but actually, Professor Ceci has done a number of experiments on falsifying memory for the central content of events. So let's see if I had anything else. No.
SPEAKER 3: We have time for one more question, if there are any.
AUDIENCE: He also asked about the net true memory which is false memory and what the implication of that was. And it was zero, apparently.
AUDIENCE: Actually, [INAUDIBLE]. It's just you can decide to say it's more or less [INAUDIBLE] or you can decide to say--
AUDIENCE: But what's the practical significance, for example, of saying their false memories outweigh their true? What would that mean? If you want to respect [INAUDIBLE] testimony or not?
SPEAKER 2: Yeah. Well, you won't like this answer. I mean, let's be honest. That particular effect is specific to those specific materials. Is this something that you would expect to get in general for negatively valenced materials? Not necessarily. The point is the relative magnitudes of those effects. The point is that the net accuracy of memory, whichever direction it is, is getting poorer and poorer the more that you move in that direction, and that it appears to be a semantic effect.
SPEAKER 1: This has been a presentation by Human Development Outreach and Extension at Cornell University.
Watch the full presentation with speaker footage and slides .
Human Development professor Charles Brainerd discusses research on how emotion can distort memory, and how this impacts the legal system.
