The N400 is modulated by unconsciously perceived

Cognitive Brain Research 13 (2002) 27–39 www.elsevier.com / locate / bres

Research report

The N400 is modulated by unconsciously perceived masked words: further evidence for an automatic spreading activation account of N400 priming effects Markus Kiefer* University of Ulm, Department of Psychiatry, Leimgrubenweg 12, 89075 Ulm, Germany Accepted 2 August 2001

Abstract It is a matter of debate whether the N400 component of the event-related brain potential (ERP) is sensitive to unconscious automatic priming mechanisms or to strategic mechanisms only. Recent studies demonstrated N400 modulation by masked primes at a short SOA supporting an automatic spreading activation account. However, it cannot be ruled out that strategic mechanisms based upon partial prime identification contributed to the observed priming effects. The present study was set up to substantiate masked N400 priming effects as an index of automatic spreading activation. It was assessed whether partial identification of the masked words due to backward priming could have supported strategic priming to occur. In experiment 1, ERPs were recorded while subjects performed lexical decisions on targets preceded by masked and unmasked primes at an SOA of 67 ms. Masked words, which were not consciously perceived, as well as visible words were shown to modulate the N400 to meaningfully related target words. Experiment 2 required subjects to perform decisions on visual, lexical and semantic features of masked words presented with or without semantically related context words. Subjects performed at chance in all tasks. Furthermore, the results exclude the possibility that backward priming has rendered the masked words partially visible. The present study therefore demonstrates that N400 priming effects can be reliably obtained from unconsciously perceived masked words at a very short SOA and strengthens the notion that the N400 is modulated by automatic spreading activation and not exclusively by strategic semantic processes.  2002 Elsevier Science B.V. All rights reserved. Theme: Neural basis of behavior Topic: Cognition Keywords: Masked semantic priming; Unconscious perception; N400; Event-related potential

1. Introduction The automaticity of semantic word processing and its neurophysiological correlates have been intensively and controversially debated (e.g., [13]). Behavioral studies reliably show that the meaning of unconsciously perceived stimuli modulates responses to subsequently presented target words suggesting unconscious automatic semantic processes to have occurred [22,23,38]. In contrast, the studies directly comparing the neurophysiological correlates of conscious and unconscious semantic processing have frequently failed to obtain semantic brain activation elicited by unconsciously perceived or unattended words *Tel.: 149-731-502-1455; fax: 149-731-502-6751. E-mail address: [email protected] (M. Kiefer).

(e.g., [4,7,45], but see [12,15,29,46]). Therefore, it has been questioned that unconscious automatic processing modulates brain activity in semantic areas. In event-related brain potential (ERP) research on semantic processing, it is a matter of debate whether or not the N400 ERP component is modulated by controlled semantic processes only or by automatic processes as well [1,7,12,13,25,29,30,48]. The N400 is a negative ERP deflection over the centro-parietal scalp, which specifically reflects semantic processing [32]. 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., [19,33]) and the word level (e.g., [5,11,27,32,35,41]). At the word level, semantic processes can be investigated using the semantic priming paradigm [39]: Be-

0926-6410 / 02 / $ – see front matter  2002 Elsevier Science B.V. All rights reserved. PII: S0926-6410( 01 )00085-4

28

M. Kiefer / Cognitive Brain Research 13 (2002) 27 – 39

havioral responses to a target stimulus (e.g. a word or picture) are facilitated when it is preceded by a meaningfully related prime stimulus (semantic priming effect). For instance, in lexical decision tasks, when subjects have to decide whether a target word (e.g. ‘lemon’) is a real word or a pseudoword, responses are faster and more accurate if the target is presented following a semantically related prime word (e.g. ‘sour’) compared to an unrelated word (e.g. ‘house’). In ERP studies, N400 amplitude to targets is attenuated for semantically related word pairs compared to unrelated word pairs, the so-called N400 priming effect (e.g., [2,3,5,9,11,14,26,30]). Semantic priming effects are explained by two general classes of cognitive mechanisms: Firstly, unconscious automatic spreading activation and, secondly, conscious strategic semantic processing [43]. According to the first cognitive mechanism, the presentation of a prime stimulus is thought to activate the corresponding conceptual representation in a semantic network, and activation automatically spreads to semantically related nodes, thereby increasing their activation level [10]. As a consequence the response to a semantically related target is facilitated. According to the second class of cognitive mechanisms (strategic semantic processing), semantic priming is the result of controlled attentional processes. Several controlled mechanisms have been suggested to account for priming effects (for an overview, see [39]). Regarding the processing nature of the N400, the semantic matching account has been the most influential one (e.g., [7,9,24]). It refers to a strategy, in which semantic similarity between prime and target is matched after lexical access has been completed [40]. A meaningful relation between the prime and the target readily indicates that the target is a word which facilitates the word-response in a lexical decision task. For unrelated pairs, in contrast, a semantic mismatch is detected, and additional linguistic analysis is required. Therefore, the response to unrelated pairs is slowed down. Controlled priming mechanisms are generally acting more effective at relatively long intervals (e.g., .500 ms) between the onset of the prime and of the target (stimulus onset asynchrony, SOA) whereas automatic spreading activation is thought to be the dominant mechanism at short SOAs [39]. Nevertheless, there is evidence that semantic matching strategies can be employed at SOAs as short as 150 ms [31]. Thus, with clearly visible words both automatic spreading activation and controlled mechanisms usually contribute to semantic priming effects. In order to investigate automatic semantic priming mechanisms in isolation, visual awareness of the prime stimulus can be eliminated: 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. 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 exclu-

sively arises from automatic spreading activation. Such masked priming effects have been reliably demonstrated in several behavioral studies (e.g., [22,23,37,38]. In contrast to these robust behavioral masked priming effects, evidence regarding electrophysiological correlates of unconscious automatic semantic processing is rather mixed. In fact, the view that N400 modulation reflects only strategic semantic matching processes is mainly based upon observations that conscious or attentive processing of the prime is a prerequisite for N400 priming effects (for a review, see [13]): In an earlier study by Brown and Hagoort using the masked priming paradigm, N400 amplitude was modulated only by visible and not by masked primes although behavioral priming effects were obtained in both conditions [7]. N400 priming effects were found in a dichotic listening task for attended, but not for ignored prime words [4]. 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 [9]. In contrast to these earlier findings, there is recent evidence that the N400 potential is also modulated by masked words, which were not consciously perceived [12,29] and by words which were not available for report because they are presented during the attentional blink [36,46]. The results of these more recent studies suggest that N400 modulation also reflects automatic spread of activation. It has been proposed that masked N400 priming effects strongly depend on the SOA and that the use of the long SOA of 500 ms in the Brown and Hagoort study [7] is one possible explanation for their failure to detect masked N400 priming effects [12,13,29]. 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 [29]. 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. However, the interpretation of the masked N400 priming effects observed in the earlier studies is limited due to some methodological problems. Some of their subjects were excluded from analysis due to missing N400 modulation in the unmasked condition [12] and due to a high prime identification rate in the masked condition [12,29]. Furthermore, the reported masked N400 priming effects cannot be unequivocally interpreted as reflecting unconscious automatic spreading activation: In the test for the visibility of the masked words, it was not explicitly investigated whether backward priming from the target to the prime has rendered the masked primes partially identifiable. For instance, subjects may have recognized the masked prime ‘table’ in the context of the semantically related target ‘chair’ after identifying only a few letters (e.g.,‘ ]abl]’). Thus, in addition to spreading activation semantic matching strategies could have contributed to the observed masked N400 priming effects.


As the issue of the electrophysiological correlates of unconscious automatic semantic processing is highly controversial and of great theoretical importance, the present study was set up to further substantiate masked N400 priming effects as an electrophysiological reflection of unconscious automatic spreading activation in semantic networks. The goal of the study was twofold: Firstly, it aimed at replicating masked priming effects on the N400 at a short SOA using an improved masking procedure which is expected to prevent all subjects from consciously identifying the masked primes (Experiment 1). Secondly, it was explicitly tested whether backward priming from the target to the prime could have rendered the masked stimuli partially identifiable thereby allowing subjects to apply strategic priming mechanisms (Experiment 2). This experiment controls for that the masked priming effects from Experiment 1 are exclusively the result of unconscious spreading activation and are not affected by strategic priming.

2. Experiment 1 In the first experiment, masked priming effects on ERPs were assessed in a comparatively large sample (n524) with a modified design from our earlier study [29]. The use of a shorter stimulus duration (33.5 ms instead of 50 ms) and of a prolonged postmask duration (33.5 ms instead of 16.7 ms) was expected to make masking more effective and to prevent all tested subjects from consciously perceiving the primes, thereby avoiding a subject selection bias. As the SOA of 67 ms (and not the SOA of 200 ms) was the critical condition for obtaining masked priming effects, only this short SOA was realized.

2.1. Material and methods 2.1.1. Subjects Twenty-four healthy, right-handed, native German speakers with normal or corrected-to-normal vision served as subjects in the experiment (male / female: 13 / 11, average age: 31.2 years, range 18–55 years). Handedness was assessed using a translated version of the Edinburgh Handedness Inventory [42]. 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. 2.1.2. Material and procedure Subjects were first presented with a fixation cross for 750 ms and then with a prime word for 33.5 ms (Fig. 1). Thereafter, a target 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

29

Fig. 1. Temporal sequence of one trial. The prime word was presented either visible (A) or masked using a random letter pattern mask (B).

middle finger, respectively. In the masked condition, a random pattern mask consisting of 10 letters was presented for 100 ms before and for 33.5 ms after the prime word. Subjects were not informed of the presence of the prime behind the mask. In the unmasked condition, only the fixation cross was displayed before the prime and a blank screen for 33.5 ms thereafter. Hence, the prime target SOA was 67 ms in all conditions. All stimuli were displayed in white font against a black background on a computer monitor synchronous with the screen refresh (refresh rate516.67 ms). Stimuli were the same as in the earlier priming study [29] 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 2.58 in width and 0.88 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’). Prime-target combinations were equated in word length and frequency [47] of the primes as well as those of the targets across conditions (pseudowords were only matched in length). Trials of the masked and unmasked conditions were presented blockwise. Trial order within each block was randomized. Prime-target combinations were divided into two lists. The assignment of a list to a given experimental condition 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 visibility was obtained thereafter. Subjects had to perform lexical decisions on 40 masked words and pseudowords, respectively. Pseudowords were constructed from real words by exchanging one or two letters. Instructions stressed accuracy over response speed. Subjects were also

30


requested to make the best guess when they felt not confident about the correct response. All subjects performed in the prime visibility test within the 95% confidence interval of chance level (65% correct). Performance was distributed around the accuracy level of 50% which is expected by mere guessing (see also Fig. 5). Average accuracy was 49.9% (range 41–62%). Moreover, sensitivity measures d’ were calculated from hit (correct responses to words) and false alarm rates (incorrect responses to pseudowords) according to [21]. Average d’ was 20.12 (range 21.11–0.72; stdv50.43), which did not deviate significantly from zero. Thus, the masking procedure was successful, and subjects were not able to consciously identify the masked primes.

2.1.3. EEG-recording, signal extraction and statistical analysis Scalp voltages were recorded using a equidistant montage of 64 sintered Ag /AgCl-electrodes mounted in a cap (Easy Cap, Falk Minow Systems). An electrode between Fpz and Fz was connected to the ground, and an electrode between Cz and FCz was used as recording reference. Eye movements were monitored with supra- and infra-orbital electrodes and with electrodes on the external canthi. Electrode impedance was kept below 5 kV. Electrical signals were amplified with Synamps amplifiers (low-pass filter570 Hz, 24 dB / octave attenuation; 50 Hz notch filter) and continuously recorded (digitization rate5250 Hz)., digitally band-pass filtered (high cut-off: 16 Hz, 24 dB / octave attenuation; low cut-off: 0.1 Hz, 12 dB / octave attenuation) and segmented (450 ms before to 800 ms after the onset of the target). EEG data were corrected to a 150 ms baseline prior to the onset of the prime (unmasked condition) and prior to the onset of the mask (masked condition), respectively. Artifact-free EEG segments to trials with correct responses were averaged separately for each experimental condition. In order to obtain a reference independent estimation of scalp voltage, the average-reference transformation was applied to the ERP data [6,28]. As N400 peak latency varied between the masked and unmasked conditions, mean voltages were analyzed statistically in two time windows, which were centered around the peak of the N400 deflection in these experimental conditions. The first time window (350–449 ms after onset of the target) covered the N400 peak in the unmasked condition and the second time window the N400 peak in the masked condition (450–549 ms). As in the previous study [29], two scalp regions of interest, each of them being represented by four pairs of contralateral electrodes, were selected for analysis: fronto–central (electrode sites: F1 / F2, FC3 / FC4, FC1 / FC2, C3 / C4) and parieto–occipital (electrode sites: P1 / 2, PO3 / PO4, PO1 / PO2, O1 / O2). Repeated measures ANOVAs were performed separately for each scalp region and time window with semantic priming, masking, hemisphere and electrode site as withinsubjects factors (P-level of 0.05). When appropriate,

degrees of freedom were adjusted according to the method of Greenhouse–Geisser, and the Greenhouse–Geisser e as well as the corrected significance levels are reported. In addition, multiple two-tailed t-tests were applied to each time sample at each electrode site in order to reveal the time course of ERP differences between semantically related and unrelated prime-target pairs in the different experimental conditions (P-level of 0.01). Finally, in order to test whether masked priming effects depended on prime visibility a regression analysis on masked priming effects (behavioral and ERP data) and sensitivity measures d’ was performed. The height at which the regression function crosses the Y-axes (the regression intercept) estimates the magnitude of priming at a d’ of zero (i.e., absence of conscious prime identification). A regression intercept significantly greater than zero indicates priming in the absence of masked prime identification. This regression intercept criterion has been proposed by [17,23] as a critical test of the hypothesis that priming has occurred unconsciously. As the regression intercept can only be meaningfully interpreted when correlations are significant, it was only determined in those analyses with significant correlation between identification measures and priming.

2.2. Results 2.2.1. Behavioral results Repeated measures ANOVAs were carried out on reaction time (RT) to the target words and on error rate (ER) with semantic relatedness and masking 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 (1.2% of the data set). Analysis of the RT data revealed reliable priming effects (main effect semantic relatedness, F(1,23) 5 71.441, MSe 5 531.3, P , 0.0001). Reactions to related word pairs were faster than to unrelated pairs (Fig. 2). Although masked priming was smaller than unmasked priming (interaction between semantic relatedness and masking, F(1,23) 5 12.958, MSe 5 292.0, P , 0.01), Newman–Keuls tests showed reliable RT differences between related and unrelated word pairs in all conditions (all P,0.001). Masked and unmasked priming effects were also obtained for ER. Subjects performed lexical decisions more accurately on trials with related word pairs (main effect semantic relatedness, F(1,23) 5 22.044, MSe 5 4.8, P 5 0.0001). Masking conditions did not significantly influence priming effects on ER (Fig. 2). 2.2.2. Electrophysiological results 2.2.2.1. 350 – 449 ms after target presentation. At parieto–occipital electrodes, the main effect of semantic relatedness was only marginally significant (F(1,23) 5 4.172, MSe 5 4.0, P 5 0.053). However, the significant


31

Fig. 2. Lexical decision latencies (lines, left y-axes) and error rates (bars, right y-axes) as a function of semantic relatedness and masking. The vertical lines depict the standard error of means of each condition.

interaction between semantic relatedness and masking indicated that the ERP priming effect differed between masked and unmasked conditions (F(1,23) 5 4.768, MSe 5 4.1, P , 0.05). Newman–Keuls tests showed that visible primes (unmasked condition) produced a significant N400 priming effect, i.e. ERPs to unrelated targets were more negative than to related targets (P,0.01). For masked primes, the corresponding mean ERP difference was not significant. At fronto–central electrodes, semantic relatedness did not affect ERPs (see Fig. 3).

2.2.2.2. 450 – 549 ms after target presentation. The main effect semantic relatedness was highly significant at parieto–occipital electrodes (F(1,23) 5 29.576, MSe 5 3.0, P , 0.0001). ERPs to unrelated targets were more negative than to related targets. The interaction between semantic relatedness and masking was marginally significant (F(1,23) 5 4.248, MSe 5 4.0, P 5 0.0508) indicating that masked N400 priming effects were smaller than those in the unmasked condition. In order to substantiate that N400 priming effects were reliable for both masked and unmasked primes, separate ANOVAs were performed for these conditions. Significant main effects of semantic relatedness in the masked (F(1,23) 5 5.482, MSe 5 2.6, P , 0.05) as well as in the unmasked condition (F(1,23) 5 20.633 MSe 5 4.5, P , 0.001) confirmed parieto–occipital ERP priming effects independent of masking (Fig. 3). At fronto–central electrodes, semantic relatedness interacted with electrodes site (F(1,23) 5 7.800, MSe 5 0.32, e 5 0.692, P , 0.01), but mean differences between the related and unrelated conditions were not statistically reliable. 2.2.2.3. Sample-by-sample t-tests. Consecutive t-tests revealed statistically reliable differences between related and

unrelated targets starting at 400 ms in the unmasked condition and at 480 ms in the masked conditions (Fig. 4). Thus, ERP priming effects started about 80 ms later in the masked compared to the unmasked condition.

2.2.2.4. Regression analysis. In order to assess whether or not masked priming effects depended on prime visibility, a linear regression analysis was performed which relates masked behavioral priming (RT difference between unrelated–related targets) and masked N400 priming (ERP difference between related–unrelated-targets at parieto– occipital electrodes collapsed over electrode positions)1 to identification measures (d’) in the prime visibility test (Fig. 5). Behavioral masked priming was negatively correlated with prime visibility (r5 20.45, P,0.05). N400 priming effects (450–549 ms time window, in which significant effects were obtained) did not correlate significantly with the identification measures (r5 20.23, P50.29). For the unmasked condition, correlations were not significant (all P.0.37). For the RT data, in which the correlation between identification measures and priming was significant, the regression intercept was determined which estimates magnitude of masked priming at a d’ measure of zero (i.e., at zero identifiability of the masked primes; for the rationale of this technique see the method section). The regression intercept was significantly greater than zero indicating priming in the absence of conscious prime identification (intercept524 ms, P,0.0001). 1

For calculating ERP priming effects, the related minus unrelated subtraction was used so that a positive value indicates a reduction in N400 amplitude, i.e. an N400 priming effect, analogue to the RT priming effects.

32


Fig. 3. Average-referenced grand-averaged voltage data (n524) from parieto–occipital and fronto–central electrodes as a function of semantic relatedness in the unmasked (A) and in the masked condition (B). The onset of the target is indicated by the long vertical line. The small vertical line in the lowest plots shows the onset of the prime in the unmasked conditions and the onset of the forward mask in the masked condition. Negative potentials are plotted downwards.


33

Fig. 3. (continued)

3. Experiment 2 In the second experiment, it was investigated whether or not backward priming occurs under the stimulation conditions of Experiment 1 making the masked words available for conscious identification. Furthermore, it was tested

how the precise task in the visibility test influences identification of the masked stimuli. It was assessed at which level of analysis (visual, lexical or semantic) features of the masked words can be identified. This experiment should control for that the masked priming effects in Experiment 1 exclusively arose from spreading

34


Fig. 4. Results of the sample-by-sample t-tests at selected electrodes.

activation and were not contaminated by strategic semantic mechanisms because of partial prime identification. Visibility of the masked stimuli was tested in four different tasks: (i) lexical decision on single masked words and pseudowords as in experiment 1 (lexical decision without context), (ii) lexical decision on masked stimuli followed by an unmasked context word which was either semantically related or unrelated to the previously shown masked words (lexical decision with context), (iii) visual discrimination between masked words and letter strings (e.g., AAAAAAA) followed either by a semantically related or unrelated unmasked context word (visual discrimination with context), (iv) semantic relatedness judgement between a masked word and a subsequently presented unmasked word (semantic judgement). If masked word identification depends on the word features required to be detected in the different tasks, identification measures should be highest in the visual discrimination task with context and lowest in the semantic judgement task which depends on word analysis far beyond simple visual feature analysis. The occurrence of backward priming can be directly tested by comparing identification performance in the lexical decision and the visual discrimination tasks, both with context word. Identification rate should be largest in the semantically related condition of the visual discrimination task. In this task, backward priming is more beneficial for completing a partially recognized word because correct responses do not necessarily depend on the identification of each single letter of the word. In the lexical decision task, in contrast, backward priming does not improve performance that much because pseudowords were constructed by exchanging one or two letters from real words thus requiring identification of each single letter. Conversely, if backward priming does not play a role in masked word identification, identification measures should not differ between the different tasks and the semantic relatedness conditions.

3.1. Material and methods 3.1.1. Subjects Twenty healthy native German speakers (19 right-, 1

left-handed) with normal or corrected-to-normal vision served as subjects in the experiment (male / female: 8 / 12, average age: 28.6 years, range 19–44 years). None of the subjects participated in Experiment 1. Handedness was assessed using a translated version of the Edinburgh Handedness Inventory [42]. 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.

3.1.2. Material and procedure Subjects were presented with masked stimuli, which they had to identify under four different tasks. Timing of the events and masking procedure were identical to Experiment 1. The first task was identical with the visibility test of Experiment 1 (lexical decision without context). Subjects had to perform lexical decisions on single masked words and pseudowords (40 trials each). In the other three tasks, masked stimuli to be identified were always followed by an unmasked context word resembling the sequence of events in the priming paradigm of Experiment 1. But in contrast to the priming experiment, subjects had to respond to the first masked stimulus and not to the subsequently presented context word. The context word remained on the screen until the subject’s response. Stimuli were drawn from Experiment 1 and divided into four lists which were assigned to the different tasks. Lists did not differ significantly in average word frequency and word length. In the second task (lexical decision with context), masked stimuli consisted of 80 words and 80 pseudowords. One half of the words was semantically related to the subsequently presented unmasked words, the other half was semantically unrelated. Subjects had again to perform lexical decisions on the masked stimuli. In the third task (visual discrimination with context), 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 the subsequently presented unmasked context words (40 trials of each condition). Subjects were told to decide whether the masked stimulus


35

Fig. 5. Plots of (A) masked behavioral and (B) masked parieto–occipital ERP priming effects (time window of 450–549 ms) as a function of the sensitivity measure d’ in the masked visibility test. The plots also show the linear regression function.


36

was a word or a letter string. In the forth task (semantic judgement), subjects were presented with masked words, which were either semantically related or unrelated to the subsequently presented unmasked word (40 trials of each condition). Subjects had to indicate whether the word pair was semantically related or not. In all tasks, instructions stressed accuracy over response speed. Subjects were also requested to make the best guess when they were not confident about the correct response. The four tasks were presented blockwise in a counterbalanced fashion across all subjects. Within each block, trial order was randomized. As a measure for word identification, d’ values were calculated from hit and false alarm rates [21]. In the lexical decision tasks and in the visual discrimination task, correct responses to words were considered as hits and incorrect responses to pseudowords and letter strings, respectively, as false alarms. In the semantic judgement task, correct responses to semantically related words were counted as hits and incorrect responses to unrelated word pairs as false alarms.

and the context visual discrimination tasks. In a repeated measures ANOVA with task and semantic relatedness as factors no significant effects were obtained (all P.0.10). If anything, d’ was numerically smaller in the semantically related than in the unrelated condition. Therefore, the findings do not just reflect a lack of statistical power to detect possibly existing differences. The results of this experiment demonstrate that subjects were not able to identify masked words neither at the visual (visual discrimination task) nor at the lexical (lexical decision task) nor at the semantic level (semantic judgement task). Furthermore, the findings exclude the possibility that backward priming from the second unmasked word to the first masked word improves the identification of the latter, at least under the present stimulation conditions. The results therefore corroborate the automatic spreading activation account of the masked N400 priming effects observed in Experiment 1 and render the contribution of strategic priming mechanisms based upon partial masked prime identification very unlikely.

3.2. Results 4. Discussion All subjects achieved an accuracy in masked word identification (collapsed over all conditions within each task), which did not exceed the 95% interval of chance performance in all conditions (65% correct). Accuracy of individual subjects was distributed around the 50% chance level with an average accuracy of about 50% in all tasks (Table 1). In order to assess possible accuracy differences across the four tasks, a one-way ANOVA was performed. Tasks did not differ significantly from each other (P. 0.28). Further statistical analysis was based on d’ sensitivity measures. Average d’ measures in all tasks and context conditions did not deviate significantly from zero demonstrating that masked words were not identified. A one-way repeated measures ANOVA with experimental conditions as a factor (lexical decision with semantically related and unrelated context words, visual discrimination with semantically related and unrelated context words, semantic judgement) revealed no significant differences (P.0.11). In a further analysis, the issue of backward priming was directly tested by including only the semantically related and the unrelated conditions of the context lexical decision

In this study, masked words, which were not consciously perceived, and unmasked words modulated behavioral responses as well as the N400 ERP component (Experiment 1). In line with findings from other behavioral masked priming experiments, unconsciously perceived prime words facilitated responses to a subsequently presented, semantically related target word [17,23,37,38]. The present results also replicate findings from earlier ERP priming studies demonstrating masked N400 priming effects at very short SOAs (i.e., ,200 ms) [12,29]. The use of the relatively long SOA of 500 ms is therefore the most likely reason for the failure to observe masked N400 priming effects in the Brown and Hagoort study [7] (see [29] for a missing N400 priming at an SOA of 200 ms). In Experiment 1, subjects were not aware of the presence of the prime as shown by the verbal reports and the performance in the visibility test. Furthermore, a regression analysis, which related priming effects (behavioral and N400 priming) to the masked prime identification measure d’, showed that masked priming did not

Table 1 Identification measures for the masked stimuli as a function of task and semantic context (standard deviations in parentheses) Lexical decision without context

Lexical decision with context

Visual discrimination with context

Semantic judgement

Average accuracy in %

50.8 (4.4) range 43.8–63.8

49.4 (2.9) range 44.4–56.3

49.9 (2.4) range 44.4–53.8

51.9 (5.7) range 44.4–65.0

Average d’

0 (0.42) range 21.34–0.74

related: 20.15 (0.37) range 21.34–0.42 unrelated: 0 (0.24) range 20.42–0.39

related: 0 (0.25) range 20.55–0.39 unrelated: 0 (0.16) range 20.32–0.39

0.14 (0.36) range 20.41–0.89


increase with prime visibility. For the behavioral data, priming effects even increased with decreasing prime visibility (although all subjects performed in the visibility test within the 95% confidence interval of chance performance), and the regression intercept, which estimates priming at a d’ of zero (i.e. in the absence of conscious prime identification), was significantly greater than zero. If anything, masked primes were more effective in modulating the processing of the targets the less information could be extracted (for a similar negative relation between visibility measures and masked priming effects see [8]). Thus, although the masked primes could not be consciously identified they nevertheless reliably modulated behavioral responses as well as the N400 ERP component. Most importantly, Experiment 2 excludes the possibility that identification of masked stimuli in the visibility test depended on the particular task. Using the same stimulation parameters as in Experiment 1 masked word identification was absent across a variety of tasks: Identification measures were the same in a lexical decision task with single stimuli, in a lexical decision task and in a visual discrimination task with semantically related and unrelated word pairs, and in a semantic judgement task. Thus, masked words could not be identified neither at a visual, a lexical nor at a semantic level. As masked word identification measures were not larger for related than for unrelated word pairs, it could be ruled out that backward priming from the target to the masked prime had enhanced the visibility of the masked stimuli. Therefore, it is safe to conclude that the observed masked behavioral and N400 priming effects are not contaminated by strategic semantic processes and that automatic spreading activation, which does not depend on conscious identification of the prime is the most likely mechanism underlying the present priming effects. The present study therefore substantiates earlier findings suggesting that the N400 is also sensitive to automatic semantic processes and not only to strategic processes such as semantic matching. There are demonstrations of N400 priming effects elicited by words during the attentional blink phenomenon which cannot be consciously identified [36,46]: In a rapidly presented stream of visual stimuli, subjects have to detect multiple targets. Identification for a second target is severely impaired (attentional blink) when the first target is correctly identified and the second target is shortly presented after the first target [44]. If a prime word is placed within the attentional blink and cannot be identified it nevertheless modulates the N400 [46]. Furthermore, N400 priming effects are reported at short SOAs (e.g., [1,30]) and for indirect (e.g., ‘lemon’–‘sweet’) semantic relations [48], i.e. under conditions when controlled semantic processes are minimized. Although all these studies report N400 priming effects presumably reflecting automatic spreading activation, there is an apparent discrepancy regarding the estimated temporal decay function of automatic spreading activation be-

37

tween these studies. Priming effects on the N400 in masking experiments decay when the prime-target SOA is 200 ms and longer [7,29]. In the attentional blink paradigm, in contrast, significant priming effects on the N400 were observed with an SOA in exceed of 500 ms [46]. It seems that the lifetime of spreading activation elicited by unconsciously perceived prime words may depend on the precise experimental procedure used to make primes unavailable for conscious identification. Di Lollo and colleagues suggest in their model of visual processing that both masking and attentional blink phenomena arise from an interrupted perceptual consolidation process within the visual system [16,18]. This consolidation process, which relies on a recurrent activation of representations in multiple visual subsystems over several processing cycles, results in a stable distributed visual representation, which is associated with conscious stimulus identification. However, visual masking and the attentional blink paradigm are thought to interfere with the stimulus consolidation process in different ways. This may in turn differentially affect the temporal decay of spreading activation elicited by these unconsciously perceived stimuli: Visual masking is assumed to interfere with the consolidation of a visual representation by overwriting the stimulus representation in an early visual module before the representation reaches a stable state. In the attentional blink paradigm, in contrast, consolidation is prevented by a missing attentional amplification of the stimulus representation. It is possible that visual masking interferes more strongly with the perceptual consolidation process than the attentional blink paradigm does because the prime representation is replaced by the mask representation in the early visual module after a short period of time. The visual representation of the prime, which is not sufficiently consolidated in both conditions to subserve stimulus identification, may therefore exhibit a shorter lifetime and / or less activation strength in masking than in the attentional blink. As a consequence, automatic spreading activation in semantic networks elicited by masked primes would decay faster. This explanation of the different decay functions of automatic N400 priming effects obtained within different paradigms is clearly speculative and deserves further investigation. Future research should clarify in which way these experimental techniques differentially affect visual representations and subsequent semantic processing (for different effects during masking and the attentional blink, see [20]). In contrast to previous ERP priming studies (e.g., [7,12,29,30]) the present N400 to unrelated targets and the corresponding N400 priming effects started and peaked relatively late, particularly in the masked condition (400 ms in the unmasked and 480 ms in the masked conditions). Furthermore, masked N400 priming effects were smaller than unmasked priming effects unlike in our earlier study [29]. This result was not predicted and can be explained

38


only post-hoc. Possibly, specific parameters of the experimental design may affect the size and peak latency of the N400 effect. The relatively short prime duration may have temporally delayed and weakened the activation within semantic networks, particularly for the masked words. The later onset of the masked priming effects in the consecutive t-tests may also reflect the smaller ERP amplitude differences between related and unrelated words in the masked condition (i.e., the smaller effect size) although the observed variations in peak latency cannot be attributed to this factor. At present it is not known how prime duration differentially affects latency and size of N400 priming under masked and unmasked conditions. At least, there is evidence that physical characteristics of target stimuli such as degradation influence N400 latency [25]. Besides these latency differences, masked and unmasked N400 priming effects were indistinguishable in topography since N400 priming did not interact with topographical factors (hemisphere, electrode site) depending on the masking condition. This suggests that the same semantic processes were involved in the masked and unmasked condition. In order to address the concern that variations in the P300 may have in part contributed to the observed masked N400 effects, ERPs in the 450–549 ms time window (in which masked priming effects were significant) were compared for fast and slow responding subjects by dividing participants into two groups based on a median split of general response speed across all experimental conditions. As P300 latency is known to correlate with reaction times at least under instructions stressing both speed and accuracy as in the present experiment [34], a possible overlap of N400 and P300 effects of semantic relatedness should be manifested in an interaction with response speed. In the present experiment, P300 latency was significantly shorter for fast (557 ms) than for slow responding subjects (621 ms; t(1,22) 5 3.605, P , 0.01). However, inclusion of the between-subject factor ‘response speed’ in the analysis of the parieto–occipital ERPs did not yield significant interactions with semantic relatedness. Thus, the observed masked N400 priming effects are most likely not contaminated by variations in the P300. The present study strengthens the spreading activation account of masked N400 priming effects and extends earlier demonstrations in several ways. Firstly, N400 priming effects were obtained in a relatively large sample of 24 subjects without the necessity to reject any subjects due to a missing N400 or due to high prime visibility as in previous studies. In fact, all of the subjects participating in the present study showed N400 priming effects in the unmasked condition. This excludes the possibility of any selection bias and demonstrates the stability of masked N400 priming effects. Secondly, regression analysis showed that masked N400 priming effects did not depend on prime visibility. Thirdly, in contrast to earlier studies it

could be ruled out that measures of masked word identification depended on the task applied in the visibility test and that backward priming from the target to the masked prime may have enabled subjects to apply strategic mechanisms. Even in the simple visual discrimination task (compared to lexical decision and semantic judgement), subjects performed at chance level. It is, therefore, very unlikely that strategic semantic processes based upon partial prime identification contributed to the observed masked priming effects.

5. Conclusions In conclusion, the present study shows that masked priming effects on the N400 can be reliably obtained at a very short SOA (67 ms). These masked N400 priming effects were observed under conditions in which even partial identification of the masked primes could be ruled out. Hence, the observed masked priming effects must arise from automatic spreading activation. The results, therefore, provide strong evidence for the notion that the N400 is also sensitive to automatic spreading of activation and not exclusively to strategic post-lexical semantic processes.

Acknowledgements Supported by grants from the University of Ulm Medical School (P.506, P.638). The author thanks Jutta Weik¨ ¨ mann, Robert Muller and Sabine Gunter for their help during data acquisition as well as Matthias Weisbrod, Manfred Spitzer and two anonymous referees for helpful comments on an earlier version of this manuscript.

References [1] J.E. Anderson, P.J. Holcomb, Auditory and visual semantic priming using different stimulus onset asynchronies: an event-related potential study, Psychophysiology 32 (1995) 177–190. [2] S. Bentin, Event-related potentials, semantic processes, and expectancy factors in word recognition, Brain Lang. 31 (1987) 308–327. [3] S. Bentin, Electrophysiological studies of visual word perception, lexical organization and semantic processing: a tutorial review, Lang. Speech 32 (1989) 205–220. [4] S. Bentin, M. Kutas, S.A. Hillyard, Semantic processing and memory for attended and unattended words in dichotic listening: behavioral and electrophysiological evidence, J. Exp. Psychol. Hum. Percept. Perform. 21 (1995) 54–67. [5] S. Bentin, G. McCarthy, C.C. Wood, Event-related potentials, lexical decision, and semantic priming, Electroenceph. Clin. Neurophysiol. 60 (1985) 343–355. [6] O. Bertrand, F. Perrin, J.A. Pernier, A theoretical justification of the average reference in topographic evoked potential studies, Electroenceph. Clin. Neurophysiol. 62 (1985) 462–464. [7] C.M. Brown, P. Hagoort, The processing nature of the N400: evidence from masked priming, J. Cogn. Neurosci. 5 (1993) 33–44.

M. Kiefer / Cognitive Brain Research 13 (2002) 27 – 39 [8] T.H. Carr, D. Dagenbach, Semantic priming and repetition priming from masked words: evidence for a center-surround attentional mechanism in perceptual recognition, J. Exp. Psychol. Learn. Mem. Cogn. 16 (1990) 341–350. [9] D.J. Chwilla, C.M. Brown, P. Hagoort, The N400 as a function of the level of processing, Psychophysiology 32 (1995) 274–285. [10] A.M. Collins, E.F. Loftus, A spreading–activation theory of semantic processing, Psychol. Rev. 82 (1975) 407–428. [11] D. Deacon, A. Mehta, C. Tinsley, J.M. Nousak, Variation in the latencies and amplitudes of N400 and NA as a function of semantic priming, Psychophysiology 32 (1995) 560–570. [12] D. Deacon, S. Hewitt, Y. Chien-Ming, M. Nagata, Event-related potential indices of semantic priming using masked and unmasked words: evidence that the N400 does not reflect a post-lexical process, Cogn. Brain Res. 9 (2000) 137–146. [13] D. Deacon, J. Shelley-Tremblay, How automatically is meaning accessed: a review of the effects of attention on semantic processing, Front. Biosci. 5 (2000) E82–94. [14] D. Deacon, T.J. Uhm, W. Ritter, S. Hewitt, A. Dynowska, The lifetime of automatic semantic priming effects may exceed two seconds, Cogn. Brain Res. 7 (1999) 465–472. [15] S. Dehaene, L. Naccache, G. LeClec’H, E. Koechlin, M. Mueller, G. Dehaene-Lambertz, P.-F. van de Moortele, D. LeBihan, Imaging unconscious priming, Nature 395 (1998) 597–600. [16] V. Di Lollo, J.T. Enns, R.A. Rensink, Competition for consciousness among visual events: the psychophysics of reentrant visual processes, J. Exp. Psychol. Gen. 129 (2000) 481–507. [17] S.C. Draine, A.G. Greenwald, Replicable unconscious semantic priming, J. Exp. Psychol. Gen. 127 (1998) 286–303. [18] J.T. Enns, V. Di Lollo, What’s new in visual masking, Trends Cogn. Sci. 4 (2000) 345–352. [19] A.D. Friederici, A. Hahne, A. Mecklinger, Temporal structure of syntactic parsing: early and late event-related potentials, J. Exp. Psychol. Learn. Mem. Cogn. 22 (1996) 1219–1248. [20] B. Giesbrecht, V. Di Lollo, Beyond the attentional blink: visual masking by object substitution, J. Exp. Psychol. Hum. Percept. Perform. 24 (1998) 1454–1466. [21] D.M. Green, J.A. Swets, Signal Detection Theory and Psychophysics, Wiley, New York, 1966. [22] A.G. Greenwald, S.C. Draine, Distinguishing unconscious from conscious cognition — reasonable assumptions and replicable findings: reply to Merikle and Reingold (1998) and Dosher (1998), J. Exp. Psychol. Gen. 127 (1998) 320–324. [23] A.G. Greenwald, S.C. Draine, R.L. Abrams, Three cognitive markers of unconscious semantic activation, Science 273 (1996) 1699–1702. [24] P. Hagoort, C.M. Brown, T.Y. Swaab, Lexical-semantic event-related potential effects in patients with left hemisphere lesions and aphasia, and patients with right hemisphere lesions without aphasia, Brain 119 (1996) 627–649. [25] P.J. Holcomb, Semantic priming and stimulus degradation. Implications for the role of the N400 in language processing, Psychophysiology 30 (1993) 47–61. [26] P.J. Holcomb, H.J. Neville, Auditory and visual semantic priming in lexical decision: a comparison using event-related potentials, Lang. Cogn. Process 5 (1990) 281–312. [27] M. Kiefer, Perceptual and semantic sources of category-specific effects in object categorization: event-related potentials during picture and word categorization, Mem. Cogn. 29 (2001) 100–116. [28] M. Kiefer, F. Marzinzik, M. Weisbrod, M. Scherg, M. Spitzer, The time course of brain activations during response inhibition: evidence from event-related potentials in a Go / Nogo task, NeuroReport 9 (1998) 765–770.

39

[29] M. Kiefer, M. Spitzer, Time course of conscious and unconscious semantic brain activation, NeuroReport 11 (2000) 2401–2407. [30] M. Kiefer, M. Weisbrod, I. Kern, S. Maier, M. Spitzer, Right hemisphere activation during indirect semantic priming: evidence from event-related potentials, Brain Lang. 64 (1998) 377–408. [31] M. Koivisto, Categorical priming in the cerebral hemispheres: automatic in the left hemisphere, postlexical in the right hemisphere?, Neuropsychologia 36 (1998) 661–668. [32] M. Kutas, S.A. Hillyard, Reading senseless sentences: brain potentials reflect semantic incongruity, Science 207 (1980) 203–205. [33] M. Kutas, S.A. Hillyard, Brain potentials during reading reflect word expectancy and semantic association, Nature 307 (1984) 1161– 1163. [34] M. Kutas, G. McCarthy, E. Donchin, Augmenting mental chronometry: the P300 as a measure of stimulus evaluation time, Science 197 (1977) 792–795. [35] M. Kutas, C.K. Van Petten, Psycholinguistics electrified, in: M.A. Gernsbacher (Ed.), Handbook of Psycholinguistics, Academic Press, New York, 1994, pp. 83–143. [36] S.J. Luck, E.K. Vogel, K.L. Shapiro, Word meanings can be accessed but not reported during the attentional blink, Nature 383 (1996) 616–618. [37] A.J. Marcel, Conscious and unconscious perception: experiments on visual masking and word recognition, Cogn. Psychol. 15 (1983) 197–237. [38] P.M. Merikle, M. Daneman, Psychological investigations of unconscious perception, J. Consci. Stud. 5 (1998) 5–18. [39] J.H. Neely, Semantic priming effects in visual word recognition: a selective review of current findings and theories, in: D. Besner, G.W. Humphreys (Eds.), Basic Progresses in Reading — Visual Word Recognition, Lawrence Erlbaum Associates, Hillsdale, N.J, 1991, pp. 264–333. [40] J.H. Neely, D.E. Keefe, Semantic context effects on visual word processing: a hybrid prospective–retrospective processing theory, in: G.H. Bower (Ed.), The Psychology of Learning and Motivation: Advances in Research and Theory, Vol. 24, Academic Press, New York, 1989, pp. 207–248. [41] A.C. Nobre, G. McCarthy, Language-related ERPs: scalp distributions and modulation by word type and semantic priming, J. Cogn. Neurosci. 6 (1994) 233–255. [42] R. Oldfield, The assessment and analysis of handedness: the Edinburgh inventory, Neuropsychologia 9 (1971) 97–113. [43] M.I. Posner, C.R.R. Snyder, Attention and cognitive control, in: R.L. Solso (Ed.), Information Processing and Cognition: The Loyota Symposium, Lawrence Erlbaum Associates, Hillsdale, 1975, pp. 55–85. [44] J.E. Raymond, K.L. Shapiro, K.M. Arnell, Temporary suppression of visual processing in an RSVP task: an attentional blink?, J. Exp. Psychol. Hum. Percept. Perform. 18 (1992) 849–860. [45] G. Rees, C. Russell, C.D. Frith, J. Driver, Inattentional blindness versus inattentional amnesia for fixated but ignored words, Science 286 (1999) 2504–2507. [46] B. Rolke, M. Heil, J. Streb, E. Henninghausen, Missed prime words within the attentional blink evoke an N400 semantic priming effect, Psychophysiology 38 (2001) 165–174. ¨ ¨ [47] A. Ruoff, Haufigkeitsworterbuch gesprochener Sprache, 2nd Edi¨ tion, Niemeyer, Tubingen, 1990. [48] M. Weisbrod, M. Kiefer, S. Winkler, S. Maier, H. Hill, D. RoeschEly, M. Spitzer, Electrophysiological correlates of direct versus indirect semantic priming in normal volunteers, Cogn. Brain Res. 8 (1999) 289–298.