Sommers, M.S. & Lewis, B.P. (1999). Who really lives next door: Creating false memories with phonological neighbors. Journal of Memory and Language, 40, 83-108.
Background
- Studying lists of semantically related words can produce false memories for words related to the theme (i.e. DRM paradigm).
- Sommers and Lewis were interested in whether similar high rates of false memories could be produced by studying lists of words that are all phonologically related to Critical Lures.
- One account of false memories in the DRM procedure is the Implicit Associative Response (IAR) account. The idea is that as you study a list of words, the words sometimes cause you to spontaneously think of related words (IARs). Later you become confused about the source of the IAR thinking it was actually presented on the list.
- Models of word recognition, such as the Neighborhood Activation Model (NAM), suggest that a process similar to the production of IARs occurs when people recognize words.
- In NAM a words neighborhood includes all the words that can be created by adding, deleting or substituting a single phoneme. The model says that presentation of words activates words in the neighborhood as a function of the number of matching phonemes. So if you hear the word dog then dog and dig, don, hog (woooooo pig souuuieeee) etc get activated. Of course, dog gets more activation because more of its phonemes match, but the other items get activated too to a degree.
- So the prediction is, if you present a bunch of items, all phonologically similar to a single nonpresented item, people are likely to recall and/or recognize the nonpresented item.
Experiment 1
Okay, basically in Experiment 1 they wanted to show that they could produce the false memory effect with phonological associates the way other folks have with semantic associates.
Methods.
- First thing they needed to do was to generate lists of words that are all phonologically related to a common nonpresented word. To do that they calculated a thing called the Frequency Weighted Neighbor Probability (FWNP) between each critical lure and its neighbors.
- The FWNP is a measure of how phonologically similar two words are. It's a weighted average of the similarity of each phoneme in the words. So how similar is the first phoneme of dog and hog (not very), how about the second phoneme (yes, very similar), and so on. Similarity of the phoneme is determined by how confusable the two phoneme's are. For instance, given that you hear the "d" sound how likely are you to mistakenly say you heard the "h" sound. In the appendix this probability is denoted as
p(PNi | PSi )
- This is the probability of thinking you heard the ith (e.g. 3rd) phoneme of the neighbor given that you actually heard the ith phoneme of the stimulus.
You then take all of those probabilities (for each phoneme) and you multiply them together. Finally you multiply the whole thing by the frequency of the neighbor. So the whole equation is...
FWNP = [
Pp(PNi | PSi )]Freqj
- So, anyway, for each neighbor of each critical lure, they calculated this FWNP thingamajig, which is basically just a measure of how similar each neighbor is to its critical lure, and then they formed lists for each CL starting with the most similar neighbor and then progressively going down until they had the fifteen most similar neighbors.
- They had a male speaker record these lists, and a group of participants listened to the recordings to provide intelligibility measures, to make sure that the words weren't subject to being misperceived (as opposed to being misremembered).
- Participants were randomly assigned to a Recall+Recognition group or a Math+Recognition only group. The Recall+ Recognition group, recalled each list immediately after it was presented. The Math+Recognition group, did math problems after each list was presented. Both groups then took an auditorially presented recognition test.
Results
- False Recall:
Participants were as likely to recall CLs (M = .54) as they were Targets (M = .58) and they recalled both much more frequently than they recalled unrelated items (M = .06).
- False Recognition:
Participants were more likely to recognize presented items (M = .77 R+R; M = .74 M+R) than CLs (M = .69 R+R; M = .64 M+R) and both of these were more common than false recognition of unrelated items (M = .19 R+R; M = .18 M+R).
- So, overall, there were very high rates of false recall and false recognition of CLs, but the recall/math manipulation didn't seem to make much of difference.
Experiment 2
In Experiment 2 they were interested in Source Monitoring Accounts of false memories. And so what they did was have either one speaker say all the words, or had a variety of speakers say the words. SMF would seem to predict that source discriminations would be better if you have more variability.
Methods
- Three conditions
- Single Talker
: A single talker reads all the lists
- Mixed Talker Blocked:
A single talker reads all the items on a given list, but different lists are read by different speakers
- Mixed Talker Random:
Different talkers read different words within a given list.
- Items are presented at a relatively slow rate (4 sec/word) to given people sufficient time to encode distinctive characteristics of the speaker's voices.
Results
- Recall:
The Mixed Talker conditions produced better recall of targets than the Single Talker condition but had no effect on recall of CLs or unrelated items.
|
|
True Recall |
False Recall |
Unrelated Recall |
|
Single |
62% |
61% |
15% |
|
Mixed Blocked |
71% |
64% |
21% |
|
Mixed Random |
70% |
63% |
17% |
- Recognition:
The Mixed Talker conditions also produced better recognition of targets than the Single Talker condition but did not influence recognition of CLs or unrelated items.
- Source Memory:
For a majority of the falsely recognized CLs participants said that they could "hear the voice saying the word" in their mind. As in previous studies, people's voice judgments were fairly accurate and were related to confidence for true memories.
Expeiment 3
In Experiment 3 they also included lists with the 15 least confusable neighbors. These, of course, should produce fewer false memories than the lists made up of the most confusable neighbors.
Results
- Recall:
Whether most confusable or least confusable lists were studied didn't influence true recall very much but had a big influence on recall of the CLs.
|
|
True Recall |
False Recall |
Unrelated |
|
Most Confusable |
57% |
53% |
11% |
|
Least Confusable |
62% |
33% |
9% |
- Recognition:
The same general pattern held up for recognition. Recognition of targets and unrelated distractors was uninfluenced by which type of list was studied, but more CLs were recognized when the most confusable lists were studied rather than the least confusable lists.
Important Findings
- Its possible to produce false memories using phonological associates as well as semantic associates.
- Making the voices more distinctive didn't seem to influence false recognition or recall only true recognition and recall. The authors argue that this is problematic for the Source Monitoring Framework.
- They argue that the results are most consistent with NAM and the IAR account.