Veridical and false memory were examined in lists that contained 12 words that all converged onto the same meaning of a critical nonpresented word (e.g., snooze, wake, bedroom, slumber…, for SLEEP) or lists that contained 6 words that converged onto one meaning and 6 words that converged onto a different meaning of a homograph (e.g., stumble, season, trip, autumn…, for FALL). Associative strength from the list items to the critical item was equated across the two types of lists. In Experiments 1–5, patterns of veridical memory differed across the two types of lists; however, false memory of the critical item did not differ. This same pattern occurred regardless of whether the words diverging onto the two meanings of the homograph were presented blocked or intermixed, whether each list item was presented for 80 ms, 200 or 1200 ms during encoding, and whether a recall or recognition test was given. In Experiment 6, critical nonpresented items that followed lists converging onto one meaning were judged as more strongly related to the list. These results suggest that false memory in the DRM paradigm largely reflects lexical/associative activation, rather than the formation of a meaningful thematic representation.
Aristotle speculated that humans organize the world into a coherent mental representation through the formation of links between related experiences. In particular, his laws of association governed that such links are likely to be formed between concepts that are similar, opposites, or follow one another closely in time. More recently, cognitive psychologists have attempted to formalize a mechanism through which such extensive associative networks could (1) represent a vast amount of world knowledge and (2) access such knowledge to answer general knowledge questions, predict upcoming events, and make inferences during comprehension (Anderson, 1983; Collins & Loftus, 1975; Kintsch, 1974; Ratcliff & McKoon, 1978).
The most common procedure for investigating the organization of such networks of related information is the semantic priming paradigm ( Anderson, 1983). Using this paradigm, researchers have discovered that responding to a target word such as “cat” is faster (in naming and lexical decision tasks) following a semantically related prime (e.g., dog) than following an unrelated prime (e.g., table). Because relatedness exerts an influence in these simple tasks, some researchers have suggested that semantic priming reflects an automatic spreading activation mechanism in which, while reading or hearing a word, activation automatically spreads from the semantic representation (node) of that word to the representations (nodes) of semantically associated neighbors ( Neely, 1977; Posner & Snyder, 1975).
Demonstrations of semantic relatedness have also been obtained using episodic memory tasks. For instance, Underwood (1965) noticed that the presence of a word such as “table” in a study list increased people’s likelihood of falsely recognizing a related word such as “chair” during a later recognition test, relative to an unrelated word such as “screen.” In a more powerful procedure, known as the Deese–Roediger–McDermott (DRM) false memory paradigm (after Deese, 1959; Roediger & McDermott, 1995), participants see or hear lists that include the first 15 associates for a given target word and are then given a recall or recognition test. The robust finding from these studies is that the nonpresented target word is falsely remembered at very high levels (see Gallo & Roediger, 2002; Roediger, Balota, & Watson, 2001; Roediger & Gallo, 2003; for recent reviews). In fact, in some circumstances, these words are recalled or recognized as often (Roediger & McDermott, 1995) or even more often (Brainerd & Reyna, 1998; McDermott, 1996; Watson, Balota, & Roediger III, 2003) than items actually presented.
The question of meaning
For both semantic priming and false memory paradigms, one fundamental question has centered on whether the effect reflects lexical associative activation from the prime (or studied items) to the target (or critical nonpresented item) or is due to the extraction of meaning from the prime which then facilitates the processing of the target. The difficulty in answering this question stems from the fact that both priming studies and false memory studies rely heavily on stimuli obtained from word association norms. A vast majority of such associated pairs contain a large overlap in semantic features (see Table 1 from Hutchison, 2003). For instance, the words “cat” and “dog” are both associated (in that they typically co-occur in language) and semantically related (in that they are part of the same PET category and share many semantic features such as “fur” and “claws”). As a result, priming effects from such items could be due to either lexical association, semantic feature overlap, or both.
