Toglia, M.P., Neuschatz, J.S. & Goodwin, K.A. (1999). Recall accuracy and illusory memories: When more is less. Memory, 7, 233-256.

• This article looks at factors that influence false recall in the DRM (Deese/Roediger & McDermott paradigm).
• To remind everyone, the DRM paradigm is a word learning paradigm in which subjects are presented with lists of words in which every word is an associate of a critical non-presented word. For instance people are presented with bed, rest, awake, tired, dream, snooze, blanket, doze, slumber, snore, nap, peace, but are never presented with the word that is related to all of these words sleep. The DRM paradigm produces high levels of false recognition and false recall of the critical nonpresented word (e.g. sleep).
• The more is less theme of this article is basically the idea that having subjects focus on the meanings of words tends to increase true memories (this is the classic levels of processing effect) but should also increase false memories which preserves the meaning of the presented items. Thus you get more true memories but overall memory is less accurate.
• The authors argue that the Source Monitoring Framework is consistent with their predicted pattern of results because SMF predicts that false memories will occur when their is an overlap between internally generated and externally presented memories in terms of their qualitative characteristics and qualitative characteristics can include overlap of semantic features.
• The authors point out that one way that false memories might be generated is through implicit associative responses an idea proposed by Underwood a while back. An IAR occurs when an item automatically gives rise to the thought of an associated item (bed-> sleep)
• Another approach consistent with the author's predicted pattern of results is Fuzzy Trace Theory. Fuzzy Trace Theory posits two types of memory representations, verbatim and gist. Verbatim representations preserve item level information and gist preserves more general senses and meanings. Semantic processing should increase gist extraction because gist is meaning based. 

Experiment 1

• Subjects listened to 6 lists of 12 DRM lists presented by means of videotape. There was a 4 second pause between each item.
• Subjects performed either a semantic task (pleasantness rating) or a graphemic task (does the word contain the letter "a").
• Subjects performed an immediate recall test after each list was presented and were given 2 minutes per list for this test.

 Results

• As is typically found, semantic processing increased memory for presented words (targets) relative to graphemic processing. This is the standard levels of processing effect.
• False recall was also greater for semantic processing than for graphemic processing consistent with the author's hypothesis.

• Sometimes subjects recalled items that were not targets and were not the critical non-presented word but were still systematically related to the list.

• Subjects listened to 5 lists of 12 DRM lists from Experiment 1 were combined into one long 60 item list, again presented by means of videotape with a 4 second pause between each item.
• The list was either blocked by theme meaning, for instance, first subjects heard the sleep list, then the king list, etc. or in a quasi-random order (two items from the same list were always separated by at least two words).
• Subjects were given 5 minutes to recall as many items as they could and were tested either immediately or after a one week or three week retention interval.
• After the recall test subjects went back to the items they recalled and provided confidence ratings using a 3 point confidence scale.

• More true memories in blocked condition than in random condition and decline in true memories with delay.
• More false memories in blocked than in random condition and false memories were persistent, not greatly influenced by the delay.

• For true memories confidence declined across the delay but this was not true for false memories. For false memories, but not for true memories, confidence was higher in the blocked condition than in the random condition.
• Its worth noting that if one looks only at the random condition, confidence decreased for true and false memories at equivalent rates although the relevant interaction was not significant.

General Discussion

 Important Findings of This Article

• Factors that increased true recall (and presumably semantic processing) also increased false recall
• False memories were highly persistent across the three week delay (at least in the blocked condition)

• These findings are broadly consistent with Fuzzy Trace Theory because they show that false memories remain persistent over time. A core assumption of FTT is that gist is forgotten only slowly relative to verbatim. The findings are also consistent with FTT in that processing that increase semantic processing increased false recall.
• Source monitoring theory is also broadly consistent with findings. Consider that if subjects do experience an IAR when presented with items that this is more likely to occur if meaning is emphasized than if the surface form of the items is emphasized.
• Slower forgetting of CLs than targets is also consistent with SMF in that source discrimination should become more difficult over time. Because accurate source discrimination tends to favor the production of true memories relative to false memories one might expect false memories to increase and true memories to decline over time.