False memories and lexical decision: even twelve primes do not cause long-term semantic priming

Abstract

Semantic priming effects are usually obtained only if the prime is presented shortly before the target stimulus. Recent evidence obtained with the so-called false memory paradigm suggests, however, that in both explicit and implicit memory tasks semantic relations between words can result in long-lasting effects when multiple `primes' are presented. The aim of the present study was to investigate whether these effects would generalize to lexical decision. In four experiments we showed that even as many as 12 primes do not cause long-term semantic priming. In all experiments, however, a repetition priming effect was obtained. The present results are consistent with a number of other results showing that semantic information plays a minimal role in long-term priming in visual word recognition.

Introduction

A well-known finding, often reported in the literature, is that a response to a word (e.g., lion) is faster and more accurate if the target word is presented in the immediate context of a related word, the prime (e.g., tiger), than if it is presented in the context of an unrelated word (e.g., chair). This semantic priming effect was first obtained by Meyer and Schvaneveldt (1971) and has been replicated many times (e.g., Balota and Lorch, 1986, McNamara, 1992, Zeelenberg et al., 1998). Another well-known finding is the repetition priming effect. Responses to words are faster and more accurate for recently studied words than for words that have not been studied recently (e.g., Ratcliff et al., 1985, Scarborough et al., 1977, Wagenmakers et al., 2000).

One remarkable difference between semantic priming and repetition priming concerns the time interval over which both effects can be obtained. The semantic priming effect has been shown to be an extremely short-lived phenomenon. In the standard semantic priming paradigm, the prime is presented immediately prior to the presentation of the target. A number of studies have shown that the priming effect is eliminated if one or more unrelated words intervene between the presentation of the prime and the target (Bentin and Feldman, 1990, Dannenbring and Briand, 1982, Kirsner et al., 1984, Masson, 1995). Some studies (Joordens and Besner, 1992, McNamara, 1992) have found that priming can survive the presentation of one intervening unrelated word, but even in those cases the size of the priming effect was reduced dramatically by the presentation of an intervening word. For example, McNamara obtained a 30-ms priming effect when no words intervened between the prime and target, a 21-ms priming effect with one intervening word and no effect (i.e., a nonsignificant – 2-ms effect) with two intervening words. In contrast to semantic priming, repetition priming is obtained even when long periods of time and numerous unrelated items intervene between the first and second presentation of the target word. Several reports indicate that repetition priming can be obtained even when the first and second presentation of a word are one or more days apart (e.g., Jacoby and Dallas, 1981, Scarborough et al., 1977).

In contrast to the results mentioned above, Becker, Moscovitch, Behrmann, and Joordens (1997) recently obtained evidence for long-term semantic priming using an animacy decision task. In accordance with the results of previous studies, they obtained no long-term semantic priming, however, in a lexical decision task. Becker et al. proposed that a single mechanism underlies both repetition priming and long-term semantic priming and described a distributed connectionist type model for word recognition to account for their results. In this model, presentation of a prime causes learning of the pattern associated with that word by strengthening the connections between activated nodes. This learning speeds up later processing not only of the same word, but also of words that are similar to the prime because these words have a large part of their pattern in common with the prime. In this model, long-term semantic priming is due to the overlap of the semantic patterns of prime and target. Becker et al. argued that the different results in animacy decision and lexical decision are due to the different extents to which performance in both tasks relies on semantic processing. They argued that performance in animacy decision relies primarily on semantic processing whereas performance in lexical decision relies primarily on orthographic processing. Therefore, they argued that in a lexical decision task semantic similarity is not expected to result in long-term priming.

In a follow-up study, Joordens and Becker (1997) did obtain long-term semantic priming in lexical decision. More specifically, they obtained semantic priming over a lag of eight intervening stimuli. This is the only study that we are aware of that obtained semantic priming in a lexical decision task over a lag of more than two items. These results seem to conflict with the absence of long-term semantic priming generally observed in lexical decision. It should be noted, however, that the study of Joordens and Becker was specifically designed to encourage semantic processing of the stimuli. Semantic processing was promoted by including pseudohomophones as nonword stimuli (a pseudohomophone is a nonword that sounds like an existing word, for example brane), making the word/nonword decision more difficult. Joordens and Becker argued that the inclusion of pseudohomophones results in a higher degree of semantic processing. In other words, it turns the lexical decision task into a more semantic task, explaining the occurrence of long-term semantic priming.

A recent study by McDermott (1997), however, suggests that long-term semantic priming might be obtained even in tasks that are usually assumed to rely primarily on the processing of orthographic or perceptual information. McDermott investigated long-term priming in word stem completion and word fragment completion and used a procedure analogous to the false memory paradigm (Deese, 1959, Roediger and McDermott, 1995). In this paradigm, a study list (e.g., garage, drive, transportation, crash, chauffeur, freeway, parking, wheel, bus, jeep, race, taxi) consisting of words that are all related to a critical item, the lure (e.g., car), is presented and subsequently memory is tested. In explicit memory paradigms such as free recall and recognition, this procedure leads to a high percentage of false memories (i.e., a high percentage of intrusions of the critical lure in free recall and a high percentage of false alarms in recognition). McDermott showed that this procedure also affects performance in implicit memory tasks. Of particular interest for the present study are the results obtained in word stem completion and word fragment completion. In these tasks subjects are given a word stem (e.g., co_) or a word fragment (e.g., _o_d) that they have to complete with the first word that comes to mind (e.g., cold). For the critical lures McDermott obtained a marginally significant priming effect in word stem completion and a significant priming effect in word fragment completion. These results contrast not only with the previous studies that failed to obtain long-term priming in lexical decision, but also with studies that failed to obtain long-term semantic priming in word fragment completion (Lombardi, 1997, Roediger and Challis, 1992).

Two factors may be responsible for McDermott's (1997) success in obtaining long-term priming. First, in the McDermott study as many as 10 related primes were presented for each target word. In contrast, all but one of the studies that failed to obtain long-term semantic priming have presented only one related prime for each target. The only study (Becker et al., 1997) that presented more than one prime also failed to obtain significant long-term semantic priming. However, there was a 10-ms effect in the expected direction and although Becker et al. presented five related primes for each target, this is still considerably less than the 10 primes presented in the McDermott study. A second possibly important factor is that in the McDermott study the primes were probably processed more deeply than in previous studies that failed to obtain long-term semantic priming in lexical decision. All published studies on long-term semantic priming in lexical decision have presented both the prime and the target in a lexical decision task. Because response times in lexical decision are usually quite fast (i.e., in the order of 500–600 ms) it is likely that subjects do not fully access the semantic information of a word and hence not much semantic information may be stored in memory (see Becker et al., for a similar argument). McDermott, however, presented the primes for 5 s each during the study phase and subjects were instructed to study them carefully because they would later be asked questions about the words.

The results of McDermott (1997) are of interest not only because they contrast with the absence of long-term priming in many studies but also because they challenge several views of long-term priming. One such view is proposed by Schacter (1994). Schacter argued that priming is mediated by a Perceptual Representation System (PRS). The PRS exists of three subsystems: the visual-word-form system, the auditory-word-form system and the structural-description system. Although the subsystems process different kinds of information, they are assumed to share common features and principles of operation. For example, the three subsystems support unconscious contributions to performance (i.e., implicit memory phenomena) and operate at a level that does not involve access to the meaning of words. In this view, long-term priming in visual priming tasks such as word fragment completion depends on the storage of perceptual (and not semantic) information. Therefore, the PRS account of long-term priming predicts that prior study of semantically related words should not affect the visual encoding of the target stimulus, and hence no long-term semantic priming should be obtained. Other researchers have also argued that long-term priming in visual word recognition does not depend on semantic processes. For example, Bowers, 1999, Bowers, 2000 argues that long-term priming depends on the strengthening of orthographic codes. In fact, the absence of long-term semantic priming is one of the arguments used by Bowers (1999) for an orthographic basis of long-term priming.

It may, however, be premature to take the results of McDermott (1997) as strong evidence against the view that long-term priming in visual word recognition depends on perceptual or orthographic processes. Several researchers have argued that performance in word stem and word fragment completion may be contaminated by explicit retrieval strategies (e.g., Reingold and Goshen-Gottstein, 1996, Toth et al., 1994). Responses in word stem and fragment completion are usually quite slow. In the McDermott study, subjects were allowed up to 20 s for responding, leaving plenty of time for explicit retrieval. Especially in word fragment completion, which is a difficult task, subjects might try to think back to the study phase in order to come up with a correct completion for the word fragment. If the results were indeed due to such a contamination by explicit retrieval strategies, then the data do not provide evidence against theories that attribute priming in visual word recognition to the strengthening of perceptual or orthographic codes.

The aim of the present study was to determine if long-term semantic priming can be obtained under conditions that reduce or eliminate the possibility of contamination by explicit retrieval attempts. Therefore, a lexical decision task was used in the present study. Because response times in lexical decision are very fast, it is unlikely that an explicit retrieval strategy will be effective. A finding of long-term semantic priming in the present study would be problematic for theories that attribute long-term priming in lexical decision to the storage of perceptual or orthographic information.¹