Tip of the Tongue and Retrieval-Induced Forgetting: Selection and Inhibition Accounts

Have you ever momentarily forgotten the name of a specific place, or person, despite being able to recall many things about the name (for example the first few letters, or the number of syllables)? Chances are, if you've experienced this "tip-of-the-tongue" phenomenon, you've also had the word spontaneously occur to you minutes or hours later. One explanation for this fascinating failure of memory is retrieval-induced forgetting, in which the retrieval of closely related concepts and words actually competes with the word or concept you intended to retrieve (discussed previously). The intended item becomes available only after the residual activity among the incorrectly retrieved items has decayed.

This is the basic phenomenon investigated by a 2004 article by Johnson & Anderson in a slightly different context. The authors begin by reviewing the tip-of-the-tongue (TOT) phenomena and how it may relate to retrieval-induced forgetting (RIF). Although "blocking," "interference" and "unlearning" mechanisms can explain the basic findings in TOT and RIF, according to Johnson & Anderson they cannot explain why the unretrieved items are also inaccessible when completely different cues are provided (known as cue-independence, discussed previously). For this kind of effect, Johnson & Anderson suggest that an active process of inhibition must exist.

The same trends can be found in more semantic tasks, in which (for example) subjects might be asked to name a certain number of types of fish, before being asked to recall a specific previously studied type of fish. If subjects are asked to name only one additional exemplar from a category before recalling the previously studied target exemplar, they do so faster than if they'd named an exemplar from some unrelated category. In other words, they were "primed" by the retrieval of a related semantic prime.

On the other hand, if subjects are asked to recall four additional examplars from the same category, this reaction time benefit disappears - as though the repeated retrieval of related items actually resulted in a suppression of the target item in memory.

[Interestingly, it appears necessary for these additional exemplars to be generated by the subject, rather than merely read or viewed. This has been used to suggest that the inhibition is recall-specifc, although it's also possible that the inhibition engaged by recall is much stronger than that engaged by mere exposure, and is therefore easier to measure. This could be an important direction for future research.]

In the current study, Johnson & Anderson replicated this finding with a few twists. 160 subjects participated in one of two experiments. In the first experiment, subjects had to complete one, four or eight word fragments that were related to the less-frequent meaning of each of 24 homographs - words with two meanings (i.e., "prune," which can either be a fruit or a verb meaning "trim"). Afterwards, subjects had to complete a word fragment that was related to the dominant meaning of each homograph (for example, in the case of "prune" subjects might see "yogurt: f____" and be required to answer "fruit.") (The frequency, dominance, and length of all of these words was controlled for).

The results from this experiment showed that subjects were better at completing the word fragment associated with the homograph's dominant meaning (e.g., completing "f___" with "fruit" as paired with "yogurt") after completing a single word fragment related to the homograph's nondominant meaning (e.g., completing "t___" with "trim," as paired with "prune"). On the other hand, if they had completed more than one word fragment for the homograph's nondominant meaning, they were later impaired at completing the word fragment for the homograph's dominant meaning!

These results suggest that retrieval of a single word (i.e., a single word fragment completion) has an initial "priming" benefit on the retrieval of related words, but that with additional retrievals this benefit actually disappears.

In a second experiment, subjects completed word fragments where the words were low-frequency examples of a certain category (e.g., nutmeg is a low-frequency word, so one such pair might be "SEASONING - NU____", where subjects would need to provide the word "NUTMEG"). As in the first experiment, at the end of this task subjects were asked to complete a word fragment where the obvious answer was one that had been not been practiced in the previous phase (e.g., subjects might see "POPCORN - SA__", which they should complete with "salt".)

The results from the second experiment differed from the first in that retrieval of low-frequency exemplars always harmed the later retrieval of the high-frequency exemplar (i.e., there was no priming-related benefit as was seen for the completion of a single word fragment in the first experiment). Instead, recall of high-frequency exemplars was less likely with additional retrievals of low-frequency exemplars. These results can be interpreted to show that the high-frequency items were actively inhibited in order to complete the low-frequency word fragments, and were thus less likely to be retrieved in the end even though they are the more obvious associations!

Why should there be a benefit of retrieving competing items in the case of experiment 1 (with homographs) but not in the case of experiment 2 (with category-exemplar pairs)? Johnson & Anderson argue that inhibition is not powerful enough to counter the spreading activation caused by the initial retrieval of a competitor; additional subsequent retrievals provide time for the inhibition mechanism to act and the correct item to be selected.

In conclusion, the authors suggest that the left inferior prefrontal cortex may actually engage this inhibitory process in the service of selecting items from memory. To the contrary, some theories suggest that left inferior prefrontal cortex may actually modulate the diffuseness or spread of activity in other cortical regions, such that the semantic representations become increasingly focused (and therefore more likely to exclude alternative meanings or higher-frequency exemplars) with repeated retrievals from memory.

According to this perspective, the completion of a low-frequency word fragment would require tightly focused representations in left inferior PFC, which may preclude any priming related benefit to higher-frequency words. In other words, to even complete an unusual word fragment, you must figuratively "tighten" or "specialize" the kinds of memory cues you are using to the point where even moderately related concepts do not receive spreading activation. Instead, they may actually receive lateral inhibition, such that they are more difficult to access later.

On the other hand, semantic associates of the nondominant meaning of a homograph may not initially require such tightly focused representations. In other words, the two meanings of the word "prune" may not initially compete with each other as strongly as "salt" does with other seasonings, and therefore the activation of one meaning may be more likely to spread and activate a second meaning, at least initially. However, as more aspects of one meaning are retrieved in the completion of word fragments, the representations in left inferior PFC may become increasingly "tightened" or specific, and thus also exert lateral inhibition on the dominant meaning of a homograph.

This alternative interpretation motivates several predictions about the nature of inhibition in retrieval induced forgetting paradigms. First, although lateral inhibition is not under conscious control, the relative spread of activity in semantic networks may be consciously controllable by lateral inferior prefrontal regions. According to this hypothesis, if subjects are asked to generate a very large number of words related to the nondominant meaning of a homograph, at some point they may voluntarily loosen the representations in liPFC in order to generate more words. At this point, one might see a return of the benefit to completion of word fragments related to the dominant meaning of the homograph.

Another prediction is that as a result of this representational "tightening," subjects may be less likely to notice unexpected or infrequent events in their environment - for example, if subjects are completing the same task but are required to stomp their foot whenever a three-syllable word occurs, they may be less successful at this when engaged in a task that involves the retrieval of low-frequency or nondominant information relative to that involving more dominant or high-frequency information.