Roediger and McDermott (1995) introduced a new laboratory technique for studying false recall and false recognition. Their consequent DRM lists, which were inspired by Deese’s (1959) procedure, are a reliable way of inducing false recall and false recognition. The present experiment aimed to better understand the underlying mechanisms of false recall (FR). The authors desired to empirically demonstrate that more than one process motivates FR. They posited that certain parameter manipulations (e.g. presentation duration) could produce a U-shaped relation with the probability of FR. For instance, fast presentation duration should yield a positive correlation between accurate and false recall (both should increase). Slower presentation duration should cause FR to decline. If this U-shaped pattern occurs, then it can be reasonably argued that more than one process underlies FR. The authors address discrepant findings of previous research (e.g. some studies show an increase in FR as a function of increased presentation duration; others studies do not demonstrate this pattern). According to McDermott and Watson, the contrasting findings of past research are due to differing presentation parameters. Wider parameters are needed. Two types of word lists are compared: semantic lists (e.g. BED REST) and phonological lists (e.g. SWEET SLEEP). The authors compare the 2 types of lists in order to answer the questions: 1) Does FR produce an inverted U relation with presentation duration (incr. FR w/ incr..pres. dur/ decr. FR w/ decr..pres durn)? 2) Will a robust amount of FR be found for phonologically related lists?

Method

350 recruits in their 6^th week of basic training at Lackland Air Force Base participated in the experiment. Presentation duration was varied between subjects with 5 levels (20, 250, 1000, 3000, 5000 ms). There were 2 types of lists: semantic and phonological. 31 subjects were tested in 10 conditions. 72, 16 word lists were used. Subjects studied 36 lists. In order to minimize perceptual misperceptions, some words were presented in all "lowercase" letters, while others were presented in all "UPPERCASE" letters. Subjects were tested via computer. The were shown 36 lists (16 words each, one word presented at a time). They were instructed to write down recalled words prior to each list. To ensure understanding, subjects took a practice test. They were told that the upper/lowercase words were not important.

Results

Recall probabilities. FR of critical lure in semantic list gave an inverted U shape (see Figure 1). FR increased with increased presentation duration (PD). FR fell with further increase in presentation duration. Both list types were similar, EXCEPT recall of studied items steadily increased with PD for the list types. Semantic lists led to better recall. Extreme differences were found at the fastest PD. Semantic lists rise in FR from 20-250 ms. Phonological lists decline in FR from 20-250 ms. Look at my rendition of the result patterns below:

Semantic lists. Studied items were recalled more than critical items. 20 ms PD led to low levels of recall for both lists. Probability of accurate recall increased with increased PD. Probability of FR varied with increased duration.

Phonological lists. Critical items were recalled with greater probability than studied items. As PD increased, probability of accurate recall increased; probability of FR decreased. However, probability of FR happened 4 times more than accurate recall at 20 ms.

Serial position analysis. See Figure 2 for Serial position of curves Semantic lists produced better accurate recall than phonological lists. The curves were similar. They showed both primacy and recency effects for accurate recall. FR probabilities were near accurate recall probabilities for items in the middle of the list (PD influenced this).

Output position analysis. See Figure 3 Output positions were examined for systematicy.

Output position: in all cases, critical item that was inaccurately recalled was examined

Analyzed: only critical items that were recalled and created a ratio for each item (output position/total # recalled)

Quintiles: proportion of ratios that fell between .0 and .20 (1^st quintile) and .20 and .40 (2^nd quintile)

It was concluded that there was some systematicy in output position. Subjects tended to recall critical items towards the end of the output sequence.

If more critical items were recalled towards the beginning of the output sequence, it would look like this:

At 5000 ms output positions were nonsystematic. At 1000 ms phonological and semantic outputs differed most with respect to the output position. This finding was most similar to Roediger and McDermott’s (1995) procedure (i.e. critical item was towards end of output sequence). Generally, phonological intrusions appeared in less systematic output positions than semantic intrusions – but, dramatic differences were not found.

Discussion

*Principal findings* For semantic lists, probability of FR varied and produced a U shape across PD. For short durations, an increased PD led to increased FR probabilities. For longer durations, increased PD led to decreased FR probabilities. Phonological lists were both similar and different from semantic lists. Phonological lists gave robust FR with fast presentation (20 ms). But, FR declined across all presentation durations. There were no marked differences between serial and output positions.

The authors give several explanations of the mechanisms underlying FR. They propose that for semantic lists, a dual process seems to influence probability that lists will produce false memories. The 2 processes include: activation and monitoring. The present paper states that patterns found for semantic lists are due to a buildup in semantic activation (20-250 ms). Single process theories cannot account for U shape (See Figure 4). They have objections to list length being considered a mechanism. The authors state that activation is the mechanism of both list length and presentation duration. Semantic activation could play a role in both encoding and retrieval processes. Monitoring, on the other hand, is a controlled process. It influences whether activation is translated into a later false memory. Monitoring takes place during encoding and retrieval. Subjects have to decide if activated concept refers to previously encountered event. The concept thus influences the likelihood of FR. For shorter durations, activation builds up with increased duration. With decreased time, monitoring breaks down – and activation is then used. For longer durations, subjects can use strategies that oppose activation.

Phonological lists have not been widely studied. The current model that explains their mechanisms is called Neighborhood Activation Model (NAM). NAM states that phonologically similar words reside in the "same neighborhood." Overall, semantic and phonological lists were similar (except for 20 ms condition for phonological lists – perceptual illusions?). The similarities suggest that semantic and phonological lists employ similar mechanisms.