Jones, T. C., Jacoby, L. L, & Gellis, L. A. (2001). Cross-modal feature and conjunction errors in recognition memory. Journal of Memory and Language, 44, 131-152.  

Jones et al. outline two theoretical accounts that have been given for why memory conjunction errors occur. Memory conjunction errors occur when discrete pieces of information (i.e., features) from two distinct memory events are inappropriately bound together producing a memorial event that was never experienced in the past. For example, a person might study blackmail and jailbird, but when they later try to remember what they saw they remember blackbird. Blackbird was never experienced but both features black and bird were studied.  

The representational account proposes that we retain specific types of memory representations for stimuli experienced in the past. For each event that is represented in memory there are thought to be two types of representations encoded, featural and configural.  

Features are basic level representations (e.g., shape, sound, color, form). These representations are bound to together at retrieval to form a coherent global representation of a past event. This binding process occurring at retrieval is guided by the configural representation, which contains the location of features in relation to one another. Memory conjunction errors occur when there is a failure to retrieve configural information at the time of test. This lack of configural information limits a person’s ability to accurately remember how features were originally put together. In addition, configural information may not be encoded at study, and thus, would not be available at the time of test to appropriately piece features back together. According to this explanation memory conjunction errors can result from failures during encoding and retrieval.  

Memory conjunction errors can also occur when the configural information is encoded inaccurately (Kroll, Knight, Metcalfe, Wood, & Tulving, 1996). In other words, conjunction errors can result from a failure at the time of encoding to accurately store the relationship of the different features to one another. This notion assumes that features presented close in time to one another can become inappropriately bound together, leaving an inaccurate configural trace. This notion is also similar to more recent work by Hannigan and Reinitz (2000) and Reinitz and Hannigan (2001). According to the arguments given by Kroll et al and Reinitz and Hannigan (2001) time is a fundamental component to inappropriately binding features together at study.  

The featural explanation can account for memory conjunction errors quit well, but it does not account for feature errors. A feature error happens when a person is presented with something (e.g., jailbird), and they inaccurately accept a distractor item with jail in it (e.g., jailhouse). However, the second part of this pair was never presented during encoding (e.g, house). Therefore, there should not be a feature representation for this second part.  

The procedural account focuses on the processes evoked by the task and does not implicate individual representations. Instead Jones, et al. contend that conjunction errors can be accounted for by a familiarity component of a dual process model of memory. Familiarity is a fast judgment of the likelihood that something presented at test was presented at study. Recollection is the second component and is a slower, controlled process. When a person recollects something they are aware of what happened in the past. By the logic of dual process models, hits on a recognition test result from both the familiarity and the recollection processes. But feature errors and conjunction errors occur only when recollection fails and there is reliance upon familiarity. If a person was to recall the context of the feature in a feature distractor or the set of features in a conjunction distractor the person would reject the distractor as old. The assumption made by these authors is that memory conjunction errors occur when there is familiarity for a conjunction distractor in the absence of recollection.  

The following experiments were designed to test the influence of shifts in the modality of the presentation of study words and test words on rates of memory conjunction errors.  

Experiment 1

Three questions addressed by the first experiment 

  1. Does changing the presentation modality from study to test affect recognition of old words? 
  2. Is the familiarity underlying conjunction errors sensitive to shifts in study to test modality? 
  3. Does mixing the presentation modality of the study word primes reduce the probability of conjunction errors?
Experimental design:
Half of the words were presented aurally and half of the words were presented visually during study. All words were presented visually on the recognition memory test. There were six different types of recognition test items (see Table 1). 
  

 Recognition Test Items 

  1. There were targets that were presented aurally at study. 
  2. There were targets that were presented visually at study 
  3. There were conjunction items that both primes were presented aurally. 
  4. There were conjunction items that both primes were presented visually. 
  5. There were mixed conjunction items that one prime was presented aurally and one prime was presented visually at study. 
  6. There were new words that had not been studied previously.
Results:
As can be seen in figure 1 no modality effects were obtained in experiment 1. There was no difference in rates of false alarms to the different types of conjunction items.
  

Experiment 2

Experiment 2 differed from experiment 1 with the addition of feature distractor items. These items were compound words that had one word that was studied previous but the other word in the compound word had not been studied previously. In addition, there was the addition of a between participant manipulation of test modality. Half of the participants completed a recognition test in which the test words were presented audibly and the other half of the participants completed a recognition test in which the words were presented visually.  

Results:
The authors predicted and observed the following pattern of recognition responding, TARGET > CONJUNCTION > FEATURE > NEW WORDS. Again the authors did not observe any modality effects. As can seen in Table 2 rates of hits and false alarms did not different between the different test modalities or the different study modalities.
  

Experiment 3

The researchers failed to observe a modality effect in the first two experiments, but this could be the result of recollection masking rates of responding to conjunction and feature items based on familiarity. Therefore, Jones et al. manipulated the ability of participants to rely on recollection by limiting the amount of time that participants had to make their recognition responses. Speeded recognition responding decreases the ability of participants to use the recollection process. This should allow for potential differences in levels of familiarity resulting from differences in modalities between study and test to be observed. 

  

Results:
As predicted by the authors the following differences were observed in rates of responding old to the different types of recognition items, TARGET > CONJUNCTION > FEATURE > NEW WORDS. Again there was no modality effect observed. Although there was a marginally significant decrease in hits to targets when there was a mismatch between modalities at study and test. 

Experiment 4

Experiment 4 was similar to the previous two experiments except for the addition of a within participant manipulation of response time allowed at test, and a between participants manipulation of study modality. Unlike the previous experiments all study words were either presented aurally or visually to participants.   

Results: There was a main effect for modality and a main effect the amount of time allowed to make a decision. False alarms more prevalent in the auditory study group and false alarms were also more likely to commit false alarms in the short deadline responding group.   

The mean acceptances of targets, conjunction lures and feature lures were corrected by subtracting out the rate of accepting completely new words. These means demonstrated yet again that there was no modality effect. However, the speeded response did decrease rates of responding to targets, but not rates of responding to conjunction and feature lures.  

From the results of the final experiment the researchers concluded that responding to targets is based on both recollection and familiarity but responding to conjunction and features is based entirely upon familiarity. Therefore the conclusion is that there is no need to introduce new mechanisms to account for false alarms to conjunction lures on recognition memory tests.
 
University of Arkansas
Department of Psychology
Graduate Program in Experimental Psychology
Lampinen Lab
False Memory Reading Group
False Memory Reading Group Fall 2001