Attentional modulation of unconscious automatic processes: Evidence from event-related potentials in a masked priming paradigm

Transcription

1 Journal of Cognitive Neuroscience in press Attentional modulation of unconscious automatic processes: Evidence from event-related potentials in a masked priming paradigm Markus Kiefer 1 and Doreen Brendel 1, 2 1 University of Ulm, Department of Psychiatry, Germany 2 Humboldt University at Berlin, Germany Running head: Attentional modulation of unconscious automatic processes Correspondence to: Markus Kiefer University of Ulm Department of Psychiatry Leimgrubenweg 12, Ulm, Germany Phone: Fax: Markus.Kiefer@medizin.uni-ulm.de URL:

2 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 2 Abstract Automatic processes are usually thought to occur independently of any cognitive resources. This traditional view has been recently challenged by showing that temporal attention to a target stimulus is a prerequisite for automatic response priming. The event-related potential (ERP) study reported here extends this research by pursuing a somewhat different approach. In two experiments, it was investigated whether masked semantic priming effects can be modulated by temporal attention to the prime using a cuing procedure. We hypothesized that masked priming is amplified when attention is directed to the stimulus stream in the time window of masked prime presentation, even in the absence of any prime awareness. ERPs were recorded while subjects performed a primed lexical decision task. Target words were preceded by semantically related or unrelated masked prime words, which were not consciously identified. A cue stimulus prompted subjects to direct their attention to the stimulus stream either shortly before the masked prime (short cue interval) or a long time interval before. Priming affected the amplitude of the N400 ERP component, an electrophysiological index of semantic processing. Unrelated prime-target pairs elicited a larger N400 than related pairs (N400 priming effect). Most importantly, this masked N400 priming effect was strongest when the cue interval and the SOA were short. The present results show that temporal attention to the prime is a prerequisite for obtaining masked N400 priming effects. They also demonstrate that unconscious automatic processes are susceptible to attentional modulation. Keywords: masked semantic priming, attention, unconscious perception, N400, event-related potentials

3 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 3 1. Introduction Unconscious automatic 1 processes are typically thought to be elicited autonomously and independently of any cognitive resources (Posner & Snyder, 1975; Schneider & Shiffrin, 1977). In classical theories of attentional control, independence of capacity-limited attention is a defining feature of automatic processes (Posner & Snyder, 1975). The classical view has been recently challenged by the observation that temporal attention to a target stimulus is a prerequisite for eliciting unconscious numerical response priming effects (Naccache, Blandin, & Dehaene, 2002). The study reported here extends this research by pursuing a somewhat different approach. Using event-related potentials (ERPs), we investigated whether masked semantic priming effects can be modulated by temporal attention. Applying a cuing procedure, we directly manipulated temporal attention to the masked prime. We were interested whether masked priming effects on the N400 ERP component, an electrophysiological correlate of semantic processing (Kutas & Hillyard, 1980), would depend on temporal attention to the unconsciously perceived prime. Semantic priming generally refers to the facilitation of a response to a target stimulus (e.g., a word) by a meaningfully related prime stimulus (Neely, 1991). In the masked priming procedure, conscious perception of the prime is eliminated by displaying a pattern mask (e.g., a random sequence of letters) before and after the prime. Unconscious semantic activation is demonstrated when the masked prime word facilitates the processing of the target stimulus. Semantic priming has been frequently observed in lexical decision tasks. When subjects have to decide whether a target word (e.g. "lemon") is a real word or a pseudoword, reactions are faster and more accurate if the target is preceded by a semantically related prime word (e.g. "sour") compared to an unrelated word (e.g. "house"). Two general cognitive mechanisms have been proposed to underlie this effect: Firstly, unconscious automatic spreading activation and, secondly, conscious strategic semantic processing (Posner & Snyder, 1975). According to the first cognitive mechanism, semantic priming reflects the automatic spread of activation in semantic networks. The presentation of a prime stimulus is thought to activate the corresponding conceptual representation in a semantic network, and activation spreads to semantically related nodes, hereby increasing

4 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 4 their activation level. Hence, if a word denoting a related concept is presented, its recognition is facilitated. According to Posner and Snyder (1975) automatic spread of activation does not depend on capacity-limited attentional processes. According to the second class of cognitive mechanisms (strategic semantic processing), semantic priming is the result of controlled attentional processes such as semantic matching or semantic expectation (for an overview, see Neely, 1991). By definition, strategic semantic processing depends on capacity-limited attentional resources (Posner & Snyder, 1975). With visible prime stimuli, both automatic spreading activation and controlled priming processes usually contribute. As for strategic semantic processing to occur, subjects must be aware of the presentation of the prime stimulus, semantic priming elicited by unconsciously perceived masked words exclusively arises from automatic spreading activation. Behavioral masked priming effects have been reliably demonstrated in several studies (e.g., Greenwald, Draine, & Abrams, 1996; Kiefer, 2002; Marcel, 1983; Merikle & Daneman, 1998). In ERP research on semantic processing, semantic priming effects are reflected by an amplitude modulation of the N400 ERP component. The N400 is a negative ERP deflection over the centro-parietal scalp, which specifically reflects semantic processing (Kutas & Hillyard, 1980). Studies using intracranial electrodes have suggested a generator in the anterior fusiform gyrus (Nobre & McCarthy, 1995). The significance of this brain area for semantic memory processes has also been shown in neuroimaging studies (e.g., Vandenberghe, Price, Wise, Josephs, & Frackowiak, 1996). The N400 has been shown to be sensitive to semantic deviations with larger N400 amplitudes for semantically incongruent words compared to congruent words at both the sentence (e.g., Friederici, Hahne, & Mecklinger, 1996; Kutas & Hillyard, 1984) and the word level (e.g., Bentin, McCarthy, & Wood, 1985; Kiefer, 2001, 2005) Using semantic priming paradigms, N400 amplitude to targets is attenuated for semantically related word pairs compared to unrelated word pairs, the so called N400 priming effect (e.g., Bentin et al., 1985; Holcomb & Neville, 1990; Kiefer, Weisbrod, Kern, Maier, & Spitzer, 1998). There is evidence that the N400 potential is reliably modulated by masked words, which were not consciously perceived (Deacon, Hewitt, Chien-Ming, & Nagata, 2000; Kiefer, 2002; Kiefer & Spitzer, 2000) and by words which were not available for report because they are presented

5 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 5 during the attentional blink (Luck, Vogel, & Shapiro, 1996; Rolke, Heil, Streb, & Henninghausen, 2001; Vogel, Luck, & Shapiro, 1998). The results of these recent studies suggest that N400 modulation also reflect automatic spread of activation. These findings are in contrast to results from some earlier studies, which suggested that N400 amplitude is exclusively modulated by strategic semantic processing. In fact, there is some evidence that conscious or attentive processing of the prime is a prerequisite for N400 priming effects (for a review, see Deacon & Shelley-Tremblay, 2000): In an earlier masked priming study by Brown & Hagoort (1993), N400 priming effects were only obtained for visible, but not for masked primes, although behavioral priming effects were obtained in both conditions. N400 priming effects were found in a dichotic listening task for attended, but not for ignored prime words (Bentin, Kutas, & Hillyard, 1995). Finally, N400 priming effects were obtained only when an orienting task required semantic processing of the prime, but not when the task asked for visual processing of word features (Chwilla, Brown, & Hagoort, 1995). It has been proposed that masked N400 priming effects strongly depend on the interval between the onset of the prime word and the target (stimulus onset asynchrony, SOA) and that the use of the long SOA of 500 ms in the Brown & Hagoort (1993) study is one possible explanation for their failure to detect masked N400 priming effects (Deacon et al., 2000; Kiefer, 2002; Kiefer & Spitzer, 2000). In fact, when varying the SOA systematically, masked N400 priming effects were found at an SOA of 67 ms, but not at an SOA of 200 ms (Kiefer & Spitzer, 2000). Unmasked N400 priming effects, in contrast, increased at the longer SOA. This study shows that masked priming on the N400 ERP component can be readily obtained, but decays rapidly within about 200 ms. In all masked priming studies described so far, subjects typically attended to the stimulation stream during the time windows of prime and target presentation. For that reason these earlier studies are not suitable to assess the influence of attention on automatic processes. Naccache et al. (2002) manipulated in a numerical response priming paradigm the allocation of temporal attention to the target. In the numerical response priming paradigm (Dehaene, Naccache, LeClec'H, Koechlin, Mueller, Dehaene-Lambertz, van de Moortele, & LeBihan, 1998), subjects had to compare target numbers to a fixed reference of five. Numbers

6 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 6 smaller and larger than five were assigned to different response hands. Subjects were faster when the masked prime and the target number fell on the same side of five, and therefore called for the same motor response than they did not (response priming effect; see also Vorberg, Mattler, Heinecke, Schmidt, & Schwarzbach, 2003). In order to manipulate the allocation of temporal attention, Naccache et al. (2002) presented subjects with a continuous stream of visual masks, within which the primes and targets appeared at varying time points after trial onset. They compared the amount of priming on the same trials, depending on whether the time of target occurrence was blocked and therefore predictable (implicit cueing) or was variable and therefore unpredictable (experiment 1). They found response priming effects only when the onset of the target was predictable. In two more experiments temporal attention was explicitly cued yielding identical results as with the implicit cuing procedure. Naccache et al. (2002) propose that a temporal window of attention is open for a few hundreds of milliseconds when subjects focus their attention on the predicted appearance of the target. Temporal attention is assumed to amplify the processing of the masked primes that were presented temporally close to the targets. This top-down attentional amplification of the unconsciously perceived masked number primes enhances, in turn, response priming effects on the targets (see also Dehaene & Naccache, 2001). Naccache et al. (2002) conclude that the concept of automaticity has to be refined since unconscious, automatic processes appear to be modulated by top-down strategic control. However, unconscious processing of the targets is automatic inasmuch it cannot serve as source of information for determining strategic processing steps (Merikle, Joordens, & Stolz, 1995). Similarly, Neumann (1984) proposed that automatic processes depend on a person s current intentions and direction of attention. In line with these authors, we suggest that attention and intentions configure the cognitive/neural system in a specific way or, as Neumann (1984) call it, specify process parameters. According to our view, a given attentional (or intentional) state might be necessary for unconscious stimuli to trigger further processes. These processes are not under intentional control once initiated and in that sense automatic. The assumption of top-down attentional influences on unconscious automatic processing also follows from a model of visual masking (Di Lollo, Enns, & Rensink, 2000; Enns & Di Lollo, 2000), which is based on re-entrant processing of visual stimuli. Di Lollo

7 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 7 and Enns propose that visual stimuli are consciously perceived when re-entrant processing of stimuli in multiple visual subsystems results in a stable representation after several processing cycles. As the mask interferes with the processing of the stimulus, a stable representation is never reached even after many processing cycles. Enns & Di Lollo (2000) suggest that in addition to the amount of interference caused by competing stimuli (i.e., masks) attention is a crucial factor whether or not re-entrant processing leads to a stable activation pattern representing the stimulus. Attention is thought to amplify the activation of the stimulus representation. It is plausible to assume that attention amplifies the activation of a stimulus representation irrespective of whether or not a stable representation is achieved after several processing cycles. We therefore propose that attention is able to enhance the processing of both consciously and unconsciously perceived stimuli. In support of this view, Kentridge, Heywood, and Weiskrantz (2004) observed in patients with blindsight that spatial cueing improved discrimination performance without awareness (see also Kentridge, Heywood, & Weiskrantz, 1999). Thus, attention and conscious experience are functionally independent to some extent and should not be equated (but see, Merikle & Joordens, 1997; Velmans, 1991). The Naccache et al. (2002) study provides supportive evidence for an attentional modulation of unconscious, automatic processes, but also has some limitations. First of all, attention was only cued to the appearance of the target. As primes and targets were presented in close temporal proximity, the prime also was attended to. However, attention to the prime and to the target are confounded so that it is not safe to conclude that temporal attention enhanced response priming effects by amplifying processing of the masked prime. Secondly, Naccache et al. (2002) investigated the effects of temporal attention on response priming. It has been debated in the response priming literature whether masked response priming effects are mainly due to direct motor specifications without mediation through semantic processes (Abrams & Greenwald, 2000; Damian, 2001). For that reason, it is open whether the Naccache et al. (2002) results also hold for semantic priming. There is at least some evidence that unconscious behavioral semantic priming does not depend on spatial attention (Fuentes, Carmona, & Agis, 1994). However, this previous study only assessed behavioral priming, but did not record ERPs, so that it is open whether neurophysiological measures would be more

8 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 8 sensitive to detect top-down attentional modulation of unconscious, automatic semantic priming. Figure 1 about here The present ERP study was set up to test within a masked semantic priming paradigm whether temporal attention to the masked primes modulates behavioral and N400 priming effects. For the masked semantic priming paradigm, we adopted the design from our earlier studies (Kiefer, 2002; Kiefer & Spitzer, 2000): Subjects performed lexical decisions on target stimuli (words and pseudowords), which were preceded by briefly presented (33.5 ms) masked prime words, which could not be consciously identified. In order to track the time course of masked priming, the prime-target SOA was either short (67 ms) or long (200 ms). In the first experiment, a cuing procedure was applied (see Fig. 1) in order to prompt subjects to attend to the stimulation stream of masks either during the time window of masked prime presentation (short cue prime interval, CPI: 200 ms) or 1 s before masked prime presentation (long CPI: 800 ms). Filler trials with an intermediate CPI were also included in order to have smoother transitions between trial lengths. In the long CPI condition, as a long period of time had elapsed after cue presentation subjects should have disengaged attention when the masked prime is finally presented. The combinations of CPI/SOA conditions were presented in a randomized sequence in order to prevent subjects to predict the occurrence of the prime. Thus, in contrast to the Naccache et al. (2002) study, attention to the masked prime and not to the target was manipulated. In the second experiment, it was controlled for whether possible interactions between masked priming and CPI did depend on attentional cuing to the prime or were merely the result of the different trial lengths.

9 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 9 2. Experiment 1 In the first experiment, the effects of the manipulation of the CPI and of the SOA on masked behavioral and N400 priming effects were assessed. We assumed that masked priming depend on temporal attention to the masked prime and on the SOA. It was expected that masked priming effects would only be observed when the masked primes are attended and when the SOA is short. Results Behavioral results Masked word identification test Visibility of the masked primes was assessed in an identification test following the priming phase. Data of three subjects had to be excluded from further analysis because identification rate of these participants exceeded the confidence interval of chance performance (accuracy greater than 65 %). For the remaining 16 participants, identification performance was distributed around the chance level of 50 % which is expected by mere guessing. Accuracy of word identification did not differ between the CPI conditions (short: 49.45%, long: %; p >.95). In order to assess whether backward priming from the target to the prime may have facilitated identification of the masked primes, d sensitivity measures for the semantically related and unrelated conditions, respectively, were calculated from hit rates (correct responses to words) and false alarm rates (erroneous responses to letter strings) according to Green and Swets (1966). A repeated-measures ANOVA on d measures with the withinsubject factors semantic relatedness and CPI revealed no significant differences between conditions. Thus, it can be excluded that backward priming rendered the masked prime words partially recognizable. Figure 2 about here

10 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 10 Masked Priming task Repeated-measures ANOVAs were carried out on reaction time (RT) to the target words and on error rate (ER) with semantic relatedness, cue prime interval (CPI) and prime target SOA as within-subject factors. For RT-analysis, mean RT of the correct responses was calculated for each condition, responses longer than twice the individual mean were treated as outliers and not considered in the analysis (1.48 % of the data set). Analysis of the RT data yielded reliable priming effects (main effect semantic relatedness, F(1,15) = , MSe = 948, p <.0001): Reactions to related word pairs were faster than to unrelated pairs (Fig. 2). A main effect of CPI indicated that lexical decisions at a short CPI were slower than at a long CPI (F(1,15) = 4.966, MSe = 1069, p <.05). This effect of the CPI shows that participants followed our instructions to attend to the stimulation stream upon cue presentation and needed more time to process the target if the CPI was short. Analysis also yielded a main effect of SOA which was due to faster reaction at the long compared with those at the short SOA (F(1,15) = , MSe = 651, p <.0001). Analysis of ER yielded a main effect of semantic relatedness (F(1,23) = 6.517, MSe = 3.304, p <.05): Subjects performed lexical decisions more accurately on trials with related word pairs (.97 %) than with unrelated word pairs (1.79 %). There were also significant interactions between semantic relatedness and SOA as well as SOA and CPI, but mean differences between conditions were not significant in Fisher LSD post-hoc tests. Electrophysiological results Target words elicited P1 and N1 ERP components peaking at about 100 and 200 ms after stimulus onset, respectively, which are related to visual sensory processing. Thereafter, a negative deflection to semantically unrelated targets was observed at centro-parietal electrodes, which started at about 200 ms and peaked at about 550 ms (Fig. 3). According to its sensitivity to semantic relatedness, polarity, latency and topography, this potential was identified as N400 ERP component (Kutas & Hillyard, 1980). It was assessed statistically whether the centro-parietal N400 priming effect (more negative potential to unrelated than to

11 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 11 related targets) is modulated by the cue prime interval (CPI) and by the prime target SOA. For the rational of time window and electrode site selection see the methods section. As only effects involving the factor semantic relatedness are of theoretical importance, only main effects and interactions involving this factor are reported. Figure 3 about here ms after target presentation (descending part of the N400): A main effect of semantic relatedness was obtained (F(1, 15) = , MSe =.736, p <.01), which showed that ERPs to unrelated targets were more negative than to related targets. However, this main effect was qualified by significant interactions with SOA and CPI (Fig. 4). An interaction between semantic relatedness and SOA indicated that the N400 priming effect was larger at the short than at the long SOA (F(1,15) = 3.089, MSe =.359, p <.05). Post hoc tests revealed that the ERP difference between unrelated and related targets was significant only at the short SOA at electrode sites PO1/PO2. Most importantly, semantic relatedness interacted with CPI and SOA (semantic relatedness X CPI X SOA X electrode site; (F(3,45) = 2.878, MS e =.395, ε =.955, p <.05), which was due to the fact that N400 priming was only present when the CPI and the SOA were short. In this condition, reliable N400 priming effects were found in post-hoc tests at all electrode sites besides CP1/CP2. In the other conditions, ERP priming effects did not reach significance. Figure 4 about here ms after target presentation (peak of the N400): Analysis again yielded a highly significant main effect of semantic relatedness (F(1,15) = , MSe = 17.83, p <.0001). ERPs to unrelated targets were more negative than to related targets. As in the previous time window, semantic relatedness interacted with CPI, SOA and electrode site (F(3,45) = 3.120, MSe =.26, ε =.955, p <.05) indicating that N400 priming effect were largest when CPI and

12 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 12 SOA were short (Fig. 4). Post-hoc tests revealed that in the CPI-short/SOA-short condition, ERP differences between unrelated and related targets were significant at all electrode sites. In the CPI-long/SOA-short condition, mean differences were reliable at all electrodes besides PO1/PO2. In the CPI-short/SOA-long condition, N400 priming was only significant at the more anterior electrodes P1/P2 and CP1/CP2. Finally, in the CPI-long/SOA-long condition significant N400 priming effects were obtained at all electrode sites besides PO3/PO4. 3. Experiment 2 In the second experiment, it was investigated whether the modulation of masked N400 priming effects as a function of the cuing procedure observed in experiment 1 did indeed depend on enhanced attention to the stimulation stream during the time window of masked prime presentation. It should be ruled out that it was merely an effect of the different lengths of the interval between onset of the trial and target presentation. We therefore conducted an control experiment, in which the experimental design was the same as in experiment 1 except that the cue stimulus was not task-relevant and did not prompt subjects to attend to the stimulation stream. Results Masked word identification test Data of three subjects had to be excluded from further analysis because identification rate exceeded the confidence interval of chance performance. For the remaining 16 participants, masked word identification was distributed around the chance level of 50 % correct responses and did not significantly differ between the CPI conditions (short: 52.34%, long: %; p >.24). D identification measures for the semantically related and unrelated conditions were subjected to a repeated-measurement ANOVA with the within-subject factor semantic relatedness and CPI, which did not reveal significant effects. There was only a non-significant trend towards better prime identification for related (d =.28) than for unrelated (d =.13)

13 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 13 prime-context word combinations (F(1,15) = 3.398, MSe =.10, p =.09). Although this effect was not significant and none of our subjects verbally reported awareness of the masked primes in the priming experiment, it cannot be entirely excluded that some backward priming might have occurred, which could have modulated masked priming effects. We addressed this issue by a regression analysis relating masked priming effects to d identification measures (see below). Figure 5 about here Masked priming task Repeated-measures ANOVAs were carried out on RT to the target words and on ER with semantic relatedness, cue prime interval (CPI) and prime target SOA as within-subject factors. For RT-analysis, mean RT of the correct responses was calculated for each condition, responses longer than twice the individual mean were treated as outliers and not considered in the analysis (.71 % of the data set). As in experiment 1, analysis of the RT data revealed reliable priming effects (main effect semantic relatedness, F(1,15) = , MSe = 384, p <.0001). Reactions to related word pairs were faster than to unrelated pairs (Fig. 5). A main effect of SOA showed that reactions at the long SOA were faster than at the short SOA (F(1,15) = , MSe = 270, p <.0001). In contrast to experiment 1, the main effect CPI did not reach significance (F < 1). There was only a trend towards an interaction between CPI and SOA (F(1,15) = 2.889, MSe = 567, p =.11) indicating that at the short SOA RT was slightly increased for the long CPI, whereas at the long SOA RT was slightly decreased for the long CPI. Hence, unlike in experiment 1 the manipulation of the CPI did not significantly and uniformly affect RT across SOA conditions. The differential effects of the CPI in experiments 1 and 2 suggest that the cuing procedure of experiment 1 was successful to enhance subjects attention to the stimulation stream during masked prime presentation. Analysis of ER only revealed an significant main effect of semantic relatedness (F(1,15) = , MSe = , p <.001). Lexical decisions to semantically related targets

14 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 14 were more accurate (.94 %) than to unrelated targets (2.34 %). No other effects approached significance. Figure 6 about here Electrophysiological results Target stimuli elicited P1, N1 and N400 ERP components as in experiment 1. However, inspection of figure 6 shows that centro-parietal N400 priming effects were less pronounced than in experiment 1. Statistical analysis of the ERPs in experiment 2 was identical with that of experiment 1. It was assessed whether the centro-parietal N400 priming effect is modulated by the cue prime interval (CPI) and prime target SOA in the absence of any instruction to the subjects to attend to the stimulation stream ms after target presentation (descending part of the N400): Analysis only yielded an significant interaction between semantic relatedness and hemisphere (F(1, 15) = 4.985, MSe =.576, p <.05). ERPs to unrelated targets were slightly more negative than to related targets, particularly over the left hemisphere, but mean differences between semantic relatedness conditions were not significant over either hemisphere. All effects involving the factor CPI did not reach significance (all Fs < 2, all ps >.18) ms after target presentation (peak and ascending part of the N400): There was only a marginally significant main effect of semantic relatedness (F(1, 15) = 4.511, MSe = 5.25, p =.051) indicating that unrelated targets elicited slightly more negative ERPs than related targets. However, all effects involving the factor CPI were not significant (all Fs < 1.05, all ps >.38). Thus, compared to experiment 1, N400 priming effects were considerably reduced and were not modulated by the CPI.

15 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 15 Regression analysis As we observed a non-significant trend towards superior masked prime identification for related than for unrelated prime-context word pairs in the masked prime identification test, a regression analysis on masked priming effects (behavioral and ERP data) and d identification measures was performed. This analysis relates masked behavioral priming (RT difference between unrelated and related targets) and masked N400 priming (ERP difference in the ms time window between related and unrelated targets at centro-parietal electrodes collapsed over electrode positions) to identification measures (d ) from the masked prime identification test for related and unrelated prime-context word combinations, respectively. We performed this analysis in order to assess the possibility that the absence of masked priming effects in experiment 2 was due to partial visibility of masked primes for semantically related prime-target combinations (backward priming). If prime visibility had influenced masked priming effects, d identification measures and masked priming measures should exhibit a significant correlation. However, for both behavioral and N400 priming effects, regression analysis did not yield any significant correlations, neither at the short CPIs (all ps >.15) nor at the long CPIs (all ps >.09). Hence, it is very unlikely that the results in experiment 2 were affected by backward priming. Joint analysis of experiments 1 and 2 The data of experiments 1 and 2 was subjected to a joint analysis in order to test whether the different effects of CPI on masked N400 priming as a function of cue relevance were statistically reliable. Cue relevance (relevant cue: experiment 1; irrelevant cue: experiment 2) was therefore entered as a between-subject variable. A repeated-measures ANOVA was performed with semantic relatedness, CPI, SOA, hemisphere and electrode site as withinsubjects factors and cue relevance as between-subject factor. In order to reduce complexity of this analysis, we only report the relevant interactions between semantic relatedness and cue relevance. In the ms time window, cue relevance interacted with semantic relatedness, SOA and electrode site (F(3,90) = 3.005, MSe =.54, ε =.794, p <.05) indexing that N400 priming was larger at the short than at the long SOA, but only when the cue was task-relevant. Most importantly, in the subsequent ms time window, the interaction

16 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 16 between cue relevance, semantic relatedness, CPI, SOA and electrode site was significant. (F(3,90) = 3.551, MSe =.28, ε = 1.0, p <.05). The joint analysis shows that masked N400 priming was largest at the short prime-target SOA when temporal attention was directed to the masked prime (short CPI when the cue was task-relevant, experiment 1). 4. Discussion In the present study, the assumption that automatic semantic processing depends on temporal attention was tested within a masked semantic priming paradigm by recoding ERPs. In line with our predictions, masked N400 priming effects had an earlier onset (descending part of the N400: ms after target onset) and were stronger around the peak of the N400 ( ms) when primes were presented within the attended time window (short CPI) and when the prime-target SOA was short (67 ms) compared with the other conditions (experiment 1). At the long SOA of 200 ms and when the prime was unattended, masked priming effect had a later onset. Moreover, N400 priming was generally smaller than in the short SOA/short CPI condition. As in the previous masked priming study (Kiefer, 2002), the peak of N400 priming effects was somewhat delayed (at about 550 ms) possibly due to the short stimulus durations. In experiment 2, when subjects were instructed to focus upon the target, masked N400 priming was generally reduced and did not reach statistical significance at all. Taken these results together, our study provides strong evidence that attention to an unconsciously perceived masked stimulus is a prerequisite for N400 ERP priming effects to occur. The present data therefore support the view that unconsciously perceived, masked stimuli require attentional amplification to elicit automatic processes (Dehaene & Naccache, 2001; Naccache et al., 2002). It should be noted that in earlier masked priming studies (Deacon et al., 2000; Kiefer, 2002; Kiefer & Spitzer, 2000), in which masked N400 priming effects were obtained, participants attended to the prime because the prime was presented shortly after the fixation cross and in close temporal proximity to the target. The dependency of masked N400 priming effects on the prime-target SOA in experiment 1 is in line with findings from earlier ERP priming studies demonstrating reliable

17 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 17 masked N400 priming effects at very short SOAs (i.e., < 200 ms) (Deacon et al., 2000; Kiefer & Spitzer, 2000), but not at SOAs of 200 ms and longer (Brown & Hagoort, 1993; Kiefer & Spitzer, 2000). Thus, the present results provide further evidence for the notion that the N400 is also sensitive to automatic spreading of activation and not exclusively to strategic postlexical semantic processes. We were able to identify two important boundary conditions for obtaining reliable N400 priming effects: prime-target SOA and attention to the prime. Whatever the precise semantic process is that is indexed by N400 amplitude modulation, e.g., automatic spread of activation in semantic networks, it also occurs under automatic processing conditions (in addition to strategic processing conditions), decays fast over time when elicited by masked stimuli and requires temporal attention to the eliciting stimuli. The boundary conditions for masked N400 priming effects identified in the present data may help to reconcile some discrepant findings in the literature regarding the processing nature of the N400. On the one hand, N400 amplitude has been shown to be modulated by unconsciously perceived masked words (Deacon et al., 2000; Kiefer, 2002; Kiefer & Spitzer, 2000) and by words not available for verbal report during the attentional blink (Luck et al., 1996; Rolke et al., 2001; Vogel et al., 1998) suggesting that the N400 ERP component is sensitive to automatic priming processes. On the other hand, N400 priming effects have been only found for attended, but not for ignored prime words (e.g., Bentin et al., 1995; Kellenbach & Michie, 1996). These latter findings have been taken as evidence that the N400 only reflects strategic post-lexical matching processes, but not automatic priming (e.g., automatic spreading of activation). Our data allows to resolve this discrepancy. The present observation of attentional modulation of unconscious masked N400 priming effects demonstrates that also automatic and not only strategic N400 priming requires the prime stimulus to be attended. Unlike in the Naccache et al. (2002) number response priming study, temporal attention did not modulate behavioral semantic priming effects in the present study: Masked behavioral priming was significant, but did not differ as a function of the cue prime interval. The present behavioral findings are therefore in line with earlier findings showing that unconscious behavioral semantic priming does not depend on spatial attention (Fuentes et al., 1994). Perhaps, semantic associations, which are stored in long-term memory, are more readily activated by unconsciously perceived stimuli compared to arbitrary stimulus-response

18 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 18 mappings (Neumann, 1990). Alternatively, although a significant RT effect of the cue-prime intervals suggest an effective manipulation of attention, our cueing procedure could have been less powerful to disengage temporal attention from the masked prime compared to the unattended condition in the Naccache et al. (2002) study. In contrast to behavioral priming effects, the present N400 priming effects were more sensitive to the manipulation of attention. This was particularly evident in experiment 2, in which we obtained highly significant behavioral priming effects, but only a trend for N400 priming effects. Dissociations between behavioral and ERP measures similar to those obtained in the present study (significant behavioral priming, but missing N400 priming effect) have been observed in several studies (e.g., Bentin et al., 1995; Brown & Hagoort, 1993; Kiefer & Spitzer, 2000), but the reason for it remains to be determined. It is possible that dependent variables, whether neurophysiological or behavioral, are differentially sensitive to experimental manipulations or capture different aspects of information processing. Electrical brain activity must be considerably strong to be measured as ERP deflection on the scalp surface (Regan, 1989). If, in some experimental conditions, semantic priming is not accompanied by a large change in the intensity of the electrical current, but rather by a change of other parameters of neural activity such as synchronization (Fries, Roelfsema, Engel, König, & Singer, 1997) or oscillatory frequency (Pulvermüller, Lutzenberger, & Preissl, 1999), these changes would not show up in the scalp ERP. Nevertheless, behavioral measures could still capture differences in the output of the neural machinery. Perhaps, processes that are reflected by the modulation of the N400 ERP component show a faster decay and depend more strongly on attentional amplification compared to processes captured by RT measures. It is also noteworthy that even for behavioral measures differences emerge across studies. Greenwald et al. (1996) using error rate (ER) within a response deadline procedure as dependent measure found a fast decay of masked priming, while in our work only ERP, but neither RT nor ER measures showed signs for such a fast decay (see also Kiefer & Spitzer, 2000). We speculate that different dependent measures might reflect different aspects of information processing and might therefore be differentially sensitive to experimental manipulations. For that reason, it is advantageous to combine behavioral and neurophysiological measures in studying human cognition.

19 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 19 N400 priming effects elicited by masked primes and by primes placed during the attentional blink demonstrate unconscious, automatic semantic priming processes. However, automatic priming effects appear to differentially depend on attention to the prime stimulus in these paradigms. In the attentional blink paradigm, subjects have to detect multiple targets within a rapidly presented stream of visual stimuli (usually with an SOA around 100 ms). Identification for a second target is severely impaired (attentional blink) when the first target is correctly identified, when the second target is shortly presented ( ms SOA) after the first target (Raymond, Shapiro, & Arnell, 1992), and if each target is followed by at least one distractor item within the stream. If a prime word is placed within the attentional blink period and cannot be identified it nevertheless modulates the N400 (Luck et al., 1996; Rolke et al., 2001; Vogel et al., 1998). It has been proposed that the attentional blink is caused by missing attentional resources for processing the second target because the cognitive system is still engaged with the first target (Chun & Potter, 1995; Enns & Di Lollo, 2000; for a review see, Shapiro, Arnell, & Raymond, 1997). Priming effects within the attentional blink have been therefore interpreted as semantic processing in the absence of attention. This interpretation seems to conflict with the present results demonstrating that masked N400 priming effects depend on temporal attention to the prime. Based on the model of visual masking proposed by Di Lollo and colleagues (see the introduction section), we propose that visual masking may interfere more strongly with perceptual processing than the attentional blink paradigm does because the prime representation is replaced by the mask representation after a short period of time (Di Lollo et al., 2000). Perceptual representations of masked stimuli may be therefore less stable than those of stimuli placed during the attention blink. Hence, N400 priming effects elicited by masked words may be more susceptible to the disengagement of attention. We also note that in the attentional blink paradigm, subjects always attend to the stimulation stream. The second target may still receive some temporal attention even if it is not sufficient for its conscious identification. Future research is clearly needed in order to elucidate the differences between masking and attentional blink paradigms (for a further discussion, see Giesbrecht & Di Lollo, 1998; Kiefer, 2002). In conclusion, the present study has important implications for the research on the relation of attention and automaticity in several respects. Firstly, we could show for the first

20 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 20 time that temporal attention to the prime is a prerequisite for obtaining masked N400 priming effects. Together with the behavioral results by Naccache et al. (2002), our findings demonstrate that unconscious automatic processes depend on and are susceptible to attentional modulation. These results therefore challenge the classical view of automatic processes to occur autonomously and independent of attentional influences (Posner & Snyder, 1975; Schneider & Shiffrin, 1977). Secondly, our data suggest that attention should not be equated with conscious awareness. Although representations of unconsciously perceived masked stimuli required some attentional amplification to elicit automatic N400 priming processes, the masked prime stimuli themselves were not consciously perceived. Finally, the present results may help to reconcile some discrepant findings regarding the sensitivity of the N400 ERP component to automatic priming processes. In previous masked priming studies, the N400 has been shown to be modulated by automatic priming processes. In contrast, other studies failed to observe N400 priming to unattended visible primes. Originally, these latter results have been taken as support for the notion that the N400 is modulated by controlled semantic processes only. The present data, however, suggest that N400 priming effects may depend on attentional enhancement of the prime whether the priming processes are automatic or controlled. 5 Material and methods 5.1 Experiment 1 Subjects Nineteen healthy, right-handed, native German speakers with normal or corrected-to-normal vision served as subjects in the experiment. Data of three subjects had to be excluded from analysis because of better than chance performance in the masked prime identification test. The remaining sixteen subjects (male/female: 6/10) had an average age of 25.1 years (range years). Handedness was assessed using a translated version of the Edinburgh

21 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 21 Handedness Inventory (Oldfield, 1971). All subjects signed a written consent after the nature and the consequences of the experiment had been explained. The experiment was conducted in accordance with the Declaration of Helsinki. Material and procedure In each trial, subjects were first presented with a fixation cross for 750 ms and then with a random letter string comprised of 10 letters for 200 ms (Fig. 1). Thereafter, a cue consisting of 10 red capital Xs was displayed for 200 ms. Subjects were instructed to focus their attention on the stimulation stream as soon as this cue appeared. After the cue, a random letter string (forward mask) was presented for 200 ms in one condition (cue prime interval, CPI, short), in the other condition (CPI long), the random letter string remained on the screen for 800 ms. We also included filler trials with an intermediate cue prime interval, in which the random letter string was presented for 500 ms. In either case, the random letter string was followed by the prime word, which was shown for 33.5 ms. After prime presentation, another random letter string was presented for 33.5 ms, which served as backward mask. Thereafter, the target stimulus was displayed that either formed a real word or a pronounceable pseudoword. Subjects were told to decide as fast and as accurately as possible whether the target was a real word or not and had to respond by pressing one of two buttons with the right index and the middle finger, respectively. Subjects were not informed of the presence of the prime behind the masks. In one condition, the target was immediately displayed after the backward mask resulting in a prime-target SOA of 67 ms (SOA short). In the other condition, the screen went black for 133 ms before the target appeared, which yielded an SOA of 200 ms (SOA long). The target remained on the screen until the subject had responded. Thereafter, three hash marks were presented, which prompted the subject to initiate the next trial by a button press. All stimuli were displayed in white font against a black background on a computer monitor synchronous with the screen refresh (refresh rate = ms). Critical stimuli were

22 Markus Kiefer and Doreen Brendel: Attentional modulation of automatic processes 22 the same as in the earlier priming studies (Kiefer, 2002; Kiefer & Spitzer, 2000) and consisted of 320 German word-word and 320 word-pseudoword pairs. Primes and targets were on average 5 letters long (range 3-9) and subtended at a viewing distance of 90 cm a visual angle of about 2.5 in width and.8 in height. The word-pseudoword pairs served as distractors and were not further analyzed. The word-word combinations consisted of 160 semantically related pairs ( hen - egg ) and 160 semantically unrelated pairs ( car - leaf ). Critical prime-target combinations were equated in word length and frequency (Ruoff, 1990) of the primes as well as those of the targets across conditions (pseudowords were only matched in length). For the filler trials, 160 new prime-target combinations were selected. Half of the filler trials consisted of word-word pairs (40 related, 40 unrelated), the other half consisted of wordpseudoword pairs. The total number of 800 trials (640 experimental and 160 filler trials) was divided in five blocks of 160 trials each. Between the blocks the subjects were allowed to rest. Each block started with 6 practice trials similar to the experimental trials. Trial order within each block was randomized. Prime-target combinations were divided into four lists. The assignment of a list to a given experimental condition (CPI/SOA) and the block presentation order was counterbalanced across subjects. After the priming experiment, subjects were informed of the presence of the prime behind the mask and were questioned as to whether they had recognized that prime words had been presented. None of the subjects reported awareness of the primes. An objective measure of prime identification was obtained thereafter (for details, see Kiefer, 2002). In a visual discrimination task, masked stimuli consisted of 80 words and 80 letter strings. Each letter string was comprised of 9 repetitions of the identical capital letter (e.g., AAAAAAAAA), which was randomly selected in each trial. Masked words were either semantically related or unrelated to a subsequently presented unmasked context word (40 trials of each condition). Sequence of events and stimulation parameters were identical with the main experiment. Subjects were also instructed to focus attention on the stimulation steam when the cue was presented. The only difference to the main experiment was that only the SOA long condition (onset difference between masked word and context = target word) was realized. This condition should provide a liberal estimation of masked prime identification for the short SOA because at the long SOA the masking influence of the context/target word is reduced.