John F. Kihlstrom
University of Michigan
University of Arizona
A full-fledged episodic memory can be described as a bundle of features containing three different sorts of information: a factual description of the event itself, a characterization of the spatiotemporal context in which it occurred, and a reference to the rememberer him- or herself as the agent or experiencer of the event. Thus, such a memory takes the following form:
I SAW A HIPPIE KISS A DEBUTANTE IN THE PARK LAST THURSDAY.
A generic associative network theory of memory would link a node representing the fact
A HIPPIE KISSED A DEBUTANTE.
to one or more nodes representing the context,
[THIS HAPPENED] IN THE PARK LAST THURSDAY
, as well as a node representing the self,
I SAW [This HAPPEN].
(There may be other contextual nodes as well, representing mood and other aspects of the internal state of the person at the time the event occurred, and perhaps the frequency with which the event has occurred, but we leave these complications to other meetings.) Most theories of memory have devoted primary attention to the information represented by the fact node, and the processes by which it is encoded and retrieved. Relatively little attention has been given to the context node(s), and still less to the self node.
However, beginning with the work of Hasher and Zacks (1979) on automatic and effortful processes in memory, and continuing with the recent upsurge of interest in mood-state and environmental-context effects on memory (Bjork & Richardson-Klavehn, 1988; Davies & Thomson, 1988), there has been increased concern with the manner in which contextual features are encoded, stored, and retrieved (for a recent review, see Naveh-Benjamin, 1987). Hasher and Zacks, of course, proposed that information concerning spatial and temporal context is encoded automatically and effortlessly. However, the apparent unreliability of environmental context effects in memory suggests that this is not the case. If context were encoded automatically, it should serve as a more reliable retrieval cue than it apparently does. However, if the encoding of context depends on the subjects' processing goals, and where they are directing their attention, it is not so surprising that context effects are weak and unreliable. Information that is not well encoded cannot be effectively utilized at the point of retrieval.
In this paper, we report two experiments that bear on the automaticity issue -- one pertaining to spatial context, the other to temporal context. Our experiment, like most others in this area, is based on one reported by Jean Mandler, Seegmuller, & Day (1979), and it represents a conceptual replication and extension of research reported at these meetings in 1984 by William Heindel and myself.
This figure presents the basic results of the Heindel study. Panel A shows the overall proportion of list items correctly recognized, and the proportion for which the context was correctly specified. As can be seen, overall memory for targets was somewhat better in the two intentional conditions than in the incidental one. But for context the effects of intentionality were quite substantial: although there was no difference between the two intentional conditions, memory for context was markedly inferior in the truly incidental condition. Recall that the truly incidental condition contained a levels-of-processing manipulation. Panel B shows that memory for targets, and for contexts, were both better for items presented in the semantic condition than in the orthographic or phonemic conditions. Because memory for context varied as a function of both intentionality and orienting task, Heindel and Kihlstrom concluded that spatial location context is not encoded automatically -- a position that has been supported in a recent paper by Naveh-Benjamin (1987).
One problem with the Heindel study was that the intentional conditions presented most list items twice, while the incidental condition presented all list items only once. That alone might have produced better memory for target items, and if memory for targets is better it is not surprising that memory for context is better as well. Accordingly, the present experiments retained the comparison of Item + Context, Item Only, and True Incidental Conditions, but embedded a levels-of-processing manipulation within each.
Basically, our studies involve three between-subjects conditions: Item + Context Intentional, in which subjects are specifically instructed to remember both the list items and the (spatial or temporal) context in which they were presented; Item Only Intentional, in which subjects are instructed to remember the items, but no mention is made of context; and True Incidental, in which subjects are not required to remember anything at all. Following the study phase, the subjects received either an old-new recognition test or a perceptual identification test, and were asked to indicate the context in which the items had been presented.
Experiment 1 involved memory for spatial location context. The subjects were presented with 48 pairs of words, cues paired with targets, 12 of which appeared in each quadrant of the screen of an IBM PC computer. A total of 60 subjects were randomly assigned to one of the three study conditions, two intentional and one incidental. All three conditions incorporated a levels-of-processing manipulation: for 16 items, the graphemic question was whether cue and target shared a letter in common (in fact, for positive items the two words shared exactly two letters in common); for another 16, the phonemic question was whether the two words rhymed; for the remaining 16, the semantic question whether they were from the same taxonomic category. In order to get memory off the floor, each cue-target pair was presented twice, with four pairs between repetitions. Following the study phase, half the subjects in each condition received an old-new recognition test consisting of the 48 targets plus an equivalent number of lures. The remaining subjects received a perceptual identification test, in which targets and lures were presented for approximately 16 milliseconds. Regardless of the type of test, after presentation of each target the subjects were asked to specify the quadrant in which the item had been presented.
This figure shows the essential results of the memory test, in terms of the number of target items correctly recognized. In a signal-detection analysis, we computed the ratio of correct to incorrect responses. A 3x3 mixed-design analysis of variance, with study condition as a between-groups factor and orienting task as a within-subjects factor, was performed both on the raw ratio scores and their logarithms. In both cases there was a significant main effect of level of processing, with about twice as many items remembered in the semantic and phonemic conditions than in the graphemic one. But there was no main effect of study condition, nor was the two-way interaction significant.
More to the point, of course, is the subjects' specification of the context in which the targets were presented. Of course, a straightforward comparison across orienting tasks would be misleading, because of the different levels of recognition. It is unfair to ask subjects where the targets appeared, if they don't remember the targets themselves. Accordingly, we considered only those targets that had been correctly recognized, and computed the ratio of those for which the quadrant was correctly respecified to those for which it was not. The slide shows these results. Each bar represents the total number of items correctly recognized: the blue portion of each bar shows correct respecification, the green portion incorrect. A 3x3 ANOVA of the log ratios again showed a big effect of orienting task, but no effect of study condition and no interaction. Fully 68% of items recognized in the semantic condition were correctly respecified, compared to only 47% and 41%, respectively, in the phonemic and graphemic conditions.
This figure shows the comparable results of the perceptual identification test. There was, of course, a significant priming effect, such that targets were identified more accurately than lures under these conditions of brief presentation. But, again, the major point of the experiment concerns the subjects' ability to specify the quadrant in which identified targets appeared. A 3x3 ANOVA on the log ratios revealed, again, a significant effect of orienting task but not of study condition. Items subject to semantic analysis were correctly respecified significantly more often, 55% of the time, than those subject to phonemic or graphemic analyses (39% and 31%, respectively).
So Experiment 1 gives fairly convincing evidence that processing of spatial context is not automatic, in that respecification depends on the amount of elaborative processing received by the items at the time of encoding. With these results in hand, we conducted a second experiment investigating the processing of information pertaining to temporal rather than spatial context. The basic design of the experiment was the same, including the three study conditions and the three orienting tasks; and the items were the same, too, except that all were presented in the center of the computer screen. The 48 items were randomly divided into four sublists of 12 items each, separated by rest intervals of one minute between lists. As before, the subjects completed recognition or perceptual identification tests, followed by a respecification test.
In principle, then, this experiment was completely analogous to Experiment 1, except that four temporal locations, reflecting sublist membership, were substituted for the four spatial locations. Unfortunately for our principles, it proved too difficult for our subjects to make such a fine-grained temporal distinction: performance on the respecification portion of the experiment was only at chance levels. Accordingly, we were forced to adopt post-hoc a coarser criterion for temporal respecification, comparing the first two lists to the second two.
This figure shows the essential results of the memory test, in terms of the number of target items correctly recognized. The signal-detection analysis again yielded a significant main effect of level of processing, but no effect of study condition and no interaction.
With respect to respecification of the temporal context in which the targets were presented, again we considered only those targets that had been correctly recognized, and computed the ratio of those for which the list was correctly respecified to those for which it was not. The figure shows these results. The dark portion of each bar shows correct respecification, the shaded portion incorrect. A 3x3 ANOVA of the log ratios showed a small but significant effect of orienting task, as before. There was no significant effect of study condition, but there was a significant two-way interaction: respecification of graphemic items was boosted in the Item + Context intentional condition.
This figure shows the comparable results for temporal respecification after the perceptual identification test. Naturally, there was a big priming effect, such that targets were identified more accurately than lures. But, again, the major point of the experiment concerns the subjects' ability to specify the list on which identified targets appeared. In this case, the 3x3 ANOVA failed to turn up any significant effects of either orienting task or study condition.
In summary, four aspects of these experiments warrant comment. First, with respect to either recognition or respecification accuracy, there is the failure to find a significant effect of study condition, intentional vs. incidental -- though the trends are always in the right direction. This is a little unusual, but perhaps it is explained by the test-wiseness of our experimental subjects. It used to be that you could do surprise recall tests early in the semester, before students in the introductory course got to social psychology and learned all about deception. Now that surprise is an important element in some of the classic experiments in cognitive psychology that students encounter early in the course, it may be more difficult to carry out intentional vs. incidental learning comparisons. On the other hand, observations of our subjects, and their comments during postexperimental interviews, indicated that the orienting tasks may have distracted the intentional subjects from their memory task. That is, they may have been so busy answering questions that they had less time than optimal to encode them properly for later remembering.
Another slight anomaly was the apparent effect of orienting task on perceptual identification, at least in the spatial context experiment. This is pretty interesting, because perceptual identification, as a measure of implicit memory (Kihlstrom, 1987; Schacter, 1987), is supposed to be independent of such encoding variables. That is, in normals, orienting task affects measures of explicit memory such as recall and recognition, as it did in our experiments, but usually does not affect measures of implicit memory such as perceptual identification or word-fragment completion. We suspect that the consequences of orienting task in the present experiment reflect top-down effects of subjects' explicit memory for the targets presented during the study phase. That is, once subjects have recognized a couple of stimuli in the perceptual identification task as targets from the study phase, they may retrieve other targets from memory to prepare them for what they might see next. And since explicit memory is partly a function of elaboration at the time of processing, it is not surprising that whatever effects explicit memory has on perceptual identification are also related to orienting condition.
Nevertheless, we did observe a clear effect of orienting task on recognition memory, and on memory for spatial context as well. Even though spatial respecification did not vary with study condition, the fact that it did vary with orienting task is inconsistent with Hasher and Zacks' automaticity hypothesis, which holds that variations in encoding strategies should not affect memory for spatial location. Considered in the context of a long line of converging research, from Mandler to Naveh-Benjamin, it would seem that information about spatial context is encoded effortfully rather than automatically.
With respect to encoding of temporal context, we think the results are not yet in. McCormack (1984), in an experiment closely resembling ours, found better temporal respecification in Item + Context than Item Only study conditions, but did not include a True Incidental control group. As we noted at the outset, the temporal discrimination required of our subjects was extremely difficult, and the effects of study condition or orienting task may have been obscured by floor effects. In fact, respecification was far above chance in the spatial experiment, about 50% against a base rate of 25%, but only slightly better than chance in the temporal experiment, about 59% against a base rate of 50%. In retrospect, we think that the temporal boundaries between lists in Experiment 2 were not quite so clear as their spatial counterparts in Experiment 1: if so, then there may not have been much temporal context to encode in the first place. Even so, the different outcomes of our two studies, force us to entertain the hypothesis that context is not context is not context, and that the processes that encode spatial context differ in some respects from those that encode temporal context. Time will tell.
Paper presented at the 29th annual meeting of the Psychonomic Society, Chicago, November 1988. This research was supported in part by a National Science Foundation Graduate Fellowship to WF; and by a Biomedical Research Support Grant from the University of Wisconsin, and by Grant #MH-35856 from the National Institute of Mental Health, both to JFK. We thank Carol Cook, Lucy Canter Kihlstrom, Daniel L. Schacter, Keith Smith, Edward E. Smith, Betsy A. Tobias, and Douglas J. Tataryn for their comments at various stages.
Bjork, R.A., & Richardson-Klavehn, A. (1989). On the puzzling relationship between environmental context and human memory. In C. Izawa (Ed.), Current issues in cognitive processes: The Tulane Floweree Symposium. Hillsdale, N.J.: Erlbaum, in press.
Davies, G.M., & Thomson, D.M. (Eds.) (1988). Memory in context: Context in memory. Chichester: Wiley.
Hasher, L., & Zacks, R.T. (1979). Automatic and effortful processes in memory. Journal of Experimental Psychology: General, 108, 356-388.
Kihlstrom, J.F., & Heindel, W.C. (1984). Automatic and effortful encoding of context. Paper presented at the 25th annual meeting of the Psychonomic Society, San Antonio, Texas. Link to text and illustrations.
Naveh-Benjamin, M. (1987). Coding of spatial location information: an automatic process? Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 595-605.
This page last modified 02/10/06 03:28:18 PM.