When Burkina Faso does not prevent birds from flying backwards: Output effects in knowledge tests
Thursday, May 07, 2015
What’s the capital of Burkina Faso? Which bird can fly backwards? What was the first human invention to break the sound barrier? Questions of this type probe what is called “semantic memory”, the repository of our general knowledge, which is known to be organized by meaning rather than by autobiographical cues. For example, you may know that Ouagadougou is the capital of Burkina Faso without having any recollection of how, when, and where you acquired that fact—you have the “semantic” knowledge but not a memory of its acquisition.
By contrast, if you were given a random list of words, such as hanson – unreorganised – semainier – borderline – barnacled, and were then asked to recall it later, then that information would be retrieved from your “episodic memory”, which is the repository of our autobiographical memories. To retrieve from episodic memory, mere knowledge of the word semainier is insufficient; you also have to remember that it was on the list of words you just tried to remember.
Those two different types of information—semantic and episodic—are obviously not independent. For example, I can still remember that I Googled “Burkina Faso” 5 minutes ago to find out the name of its capital, and I remember searching for the stimuli with obscure facts on my disk so I could determine that hummingbirds are the only birds that can fly backwards. Conversely, semantic information can aid in the retrieval of episodic information—for example, if all items you are trying to remember belong to the same category (e.g., “fruit”) then you can exploit your semantic knowledge and recall “apples” and “pears” from your shopping list more readily.
A recent article in the Psychonomic Bulletin & Review by William Aue, Amy Criss, and Melissa Prince examined one particular type of interaction between episodic and semantic information; namely the adverse effects of retrieval known as output interference. Output interference (OI from here on) refers to the well-established finding that the retrieval of episodic information interferes with further recall, all other factors being held constant. The act of recalling “hanson”, “unreorganised”, and “semainier” makes borderline and barnacled less retrievable, presumably because the recalled items are encoded into memory and therefore create additional interference with the remaining memories.
Curiously, however, previous research has shown that tests of semantic knowledge do not contribute to OI for episodic recall. Thus, asking people to make simple semantic judgments during a memory test (“is this letter string a word or not: kumquat”) does not impair episodic list recall for other words. What has been unknown until now is whether purely semantic tests create OI—does recall of Ouagadougou and hummingbirds to the first two fact questions from above prevent you from retrieving the first human invention to have cracked the sound barrier?
Aue and colleagues set out to answer this question by presenting people with 150 general-knowledge questions on two consecutive occasions. During the first test, corrective feedback was provided after each item, although participants did not expect there to be a second test of the same items. During the second test, feedback was withheld, but participants were encouraged to consider their preceding responses and feedback to answer the questions during this next round.
All test items consisted of a general-knowledge question, such as “what is the capital of Australia?” that was accompanied by a multiple-choice question involving 4 alternatives (“Sydney-Melbourne-Canberra-Brisbane”). Participants selected one of the 4 response options and if it was incorrect, then during the first test they would get corrective feedback (e.g., “Incorrect! The correct answer is Canberra”).
Emphasis during analysis was on performance across trials throughout the 150-item test sequence. If OI were present, then this would express itself as a decline in performance over trials. The results for the two tests are shown in the figure below:
The pattern is quite obvious: Performance did not decline over trials (plotted here aggregated into blocks of 10 trials) for the first test, but it did decline during test 2. This permits an obvious conclusion: There was no OI when performance was supported by semantic knowledge alone—putting aside the remote possibility that one or the other participant might still remember when they learned that Canberra was the capital of Australia, performance during the first test relied entirely on semantic knowledge rather than episodic memory.
The results were quite different for the second test: Although performance was better overall, reflecting the combined effects of practice and the feedback during test 1, it declined over trials. This decline is the classic signature pattern of OI, suggesting that the involvement of episodic memory—that is, recollection of the corrective feedback from test 1—was subject to the OI that is routinely observed with other episodic test.
To pin down this possibility further, Aue and colleagues split responses during test 2 into two subsets: Those that were answered correctly during test 1, and those that were not. The idea was that if people answered items correctly the first time round based on their semantic knowledge, then they could presumably do this again during the second test, in which case no OI would be obtained. Conversely, if they failed on test 1, then the corrective feedback would still be available in episodic memory, but episodic memory should exhibit the usual OI.
The figure below shows that this is precisely what happened:
There was no evidence of OI for items that participants were known to have knowledge of, whereas OI was very strong for items that participants had not known but for which they might remember the feedback from the first test.
Aue and colleagues conclude that, in line with much prior research, once knowledge has been acquired its representations in semantic memory are stable and hence can be accessed without affecting other representations. If you know that Ouagadougou is the capital of Burkina Faso, retrieving that knowledge will not impair your ability to tell me that hummingbirds can fly backwards. Episodic traces, by contrast, contain contextual and autobiographical features, and hence if you have to remember the answer Ouagadougou from earlier in this post, the retrieval of those contextual and autobiographical features make it harder to retrieve further answers associated with the same features.
In a nutshell, knowledge is more resilient than memories.
So if you want to know that whips cracked the sound barrier long before Chuck Yeager, make sure you rehearse and use this information in many different contexts until your memory of me just telling you the answer has faded.