A BEHAVIORAL AND ELECTROPHYSIOLOGICAL INVESTIGATION
OF MASKED SEMANTIC PRIMING
Giordana Grossi
State University of New York at New Paltz, USA
People are faster in recognizing objects or words (targets) if these stimuli are preceded by semantically related items when compared to unrelated items (primes), a phenomenon called "semantic priming". Research has shown that semantic priming can also occur when the primes are presented very briefly and masked in order to prevent their overt identification (e.g., Marcel, 1983; Dehaene et al., 1998). This study is a behavioral and electrophysiological investigation of masked semantic priming effects. Specifically, we asked the following questions: first, do masked and unmasked semantic priming effects rely on same or non-identical neural systems? Secondly, what is the nature of masked priming effects?
To answer the first question, we employed Event Related Potentials (ERPs) in two lexical decision tasks (LDT) in which the relationship between pairs of English words was manipulated. Targets were preceded by primes that were identical (dog-DOG), semantically related (cat-DOG), unrelated (pen-DOG), or neutral (XXXXXXX-DOG). Nonwords were preceded by words and neutral primes. In the masked condition, primes were briefly flashed (67 ms) before the targets (500 ms) and preceded by a mask that prevented their identification. In the unmasked condition, both primes and targets were presented for 500 ms. Behavioral evidence for semantic priming was found in both tasks, although the effect was larger in the unmasked condition. Consistent with many previous studies, ERPs in the unmasked condition were more negative to unrelated than related targets starting 300 ms after the presentation of the targets (N400 effect); this effect was larger over the posterior sites and over the right hemisphere. In the masked condition, ERPs showed an N400-like effect that was smaller and later than the N400 effect in the unmasked condition and was broadly distributed across the scalp. In addition, earlier (150 ms) differences between related and unrelated words were observed at anterior sites. This data suggest that non-identical neural systems mediate semantic priming with and without awareness.
In a follow-up study, two hypotheses on the nature of masked priming effects were tested. According to Automatic Spreading Activation (ASA) theories (e.g., Loftus & Collins, 1975), the prime activates semantically related words through an automatic mechanism of spreading activation in long-term memory (LTM). If the target is among the words already activated by the prime, its identification will be facilitated in terms of reaction times (RT) when compared to unprimed targets (priming effect). In retrospective theories of priming (e.g., Whittlesea & Jacoby, 1990; Bodner & Masson, 1997, 2001, 2003), the prime creates an episodic representation that can be recruited to assist in the encoding of the target. No changes take place in LTM: in masked priming, the representation of the prime is still episodic, although unconscious.
ASA and retrospective theories of priming make different predictions regarding the effect of the proportion of related items in a stimulus list (relatedness proportion, RP) on priming. In unmasked conditions, priming effects are larger when most (e.g., 80%) pairs of words are related (high RP). These effects are generally explained in terms of automatic priming and expectancy: in an high RP context, subjects generate potential targets after the presentation of the primes, with the results that expected targets will be recognized faster than non-expected targets (see Neely, 1991, for a review). Accordingly, the N400 effect is also larger in high than low RP conditions (Holcomb, 1988). According to Bodner & Masson (1997, 2001, 2003), the cognitive system relies on the high validity of the prime in order to process the target even when the prime is not available for verbal (conscious) report, such as in masked conditions. No differences between masked and unmasked priming effects are predicted by retrospective theories. Bodner & Masson reported larger priming effects in high RP contexts for both repetition and semantic masked priming. According to this view, the N400 effect is predicted to be larger in the high than low RP condition in masked priming as well. Contrariwise, ASA theories do not predict RP effects in masked conditions: spreading activation, being an automatic mechanism (Posner & Snyder, 1975), is hypothesized to take place regardless of RP manipulations. Therefore, no differences are expected in terms of behavioral or electrophysiological effects between high and low RP conditions when subjects are not aware of the primes.
The first prediction was tested in a behavioral masked LDT. Forty subjects were randomly assigned to each RP condition (high=80% related words; low RP=20% related words). Semantically related and unrelated primes were briefly flashed (50 ms) before the targets (500 ms) and preceded by a mask that prevented their overt identification. In a second task, subjects were required to perform a LDT on the primes and try to identify them. Results showed that the priming effect was not affected by RP, even after excluding from the analyses subjects who were aware of the presence of the target.
The second prediction was tested in an ERP experiment, which differed from the behavioral experiment in two respects: subjects were tested in both RP conditions; moreover, in order to measure the N400 effect uncontaminated by motor-related potentials, subjects were required to perform a categorization task and press a button only when proper names were presented (therefore, only trials which did not required a button press were analyzed). Preliminary results indicate that a larger N400 was elicited by unrelated than related targets; moreover, a small earlier effect was observed at the anterior sites. Although these effects were larger in the high than low RP condition, they were not significantly modulated by RP. Importantly, a trend was present for order effects: in general, priming effects tended to be larger when participants received the high RP block first.
The present findings suggest the following conclusions. First, since an N400 effect was found in both masked (both high and low RP) and unmasked conditions, such effect seems to partially reflect ASA mechanisms, as other authors have suggested (e.g., Kiefer, 2002). Second, the presence of an early ERP priming effect suggests that the N400 might not be the only electrophysiological index of automatic semantic priming; the nature of this early ERP effect remains unclear and needs to be addressed in future studies. Third, given the lack of robust behavioral and electrophysiological RP effects, the current findings do not support retrospective accounts of masked semantic priming. Although, their preliminary nature warrants caution: the presence of order effects, if confirmed, would suggest that unconscious processes might be sensitive to contextual effects. This issue will be explored in future research.