پردازش کلمه غیرآگاهانه در یک پارادایم با پوشش آینه و باعث حواس پرتی توجه: یک ERP مطالعه

Abstract In this event-related potential (ERP) study a masking technique that prevents conscious perception of words and non-words through attentional distraction was used to reveal the temporal dynamics of word processing under non-conscious and conscious conditions. In the non-conscious condition, ERP responses differed between masked words and non-words from 112 to 160 ms after stimulus-onset over posterior brain areas. The early onset of the word–non-word differences was compatible with previous studies that reported non-conscious access to orthographic information within this time period. Moreover, source localisations provided evidence for automatic activation of prelexical phonological information, whereas no evidence for non-conscious semantic processing was found. When subjects were informed about the masking technique, lexical differences occurred at later time intervals, suggesting conscious access to additional word related information. These results indicate that early visual word processing does not depend entirely on attentional resources, but that non-conscious processing probably is restricted to rather lower-level linguistic information.

1. Introduction One of the most controversial issues in cognitive psychology research addresses the extent to which non-consciously perceived linguistic information is cognitively processed and therefore can influence human behaviour (Kouider & Dehaene, 2007). To elucidate this issue, it is important to know what brain structures and functions are differentially activated during conscious and non-conscious linguistic stimulus processing. The following event-related potential (ERP) study relates to both of these issues. Although models of conscious visual word processing vary in many aspects, there is evidence from various ERP studies suggesting that skilled word reading is based on hierarchically organised processing steps, engaging an anatomically distributed neural system, whereby subsequently specific types of information becomes available (Cohen and Dehaene, 2009, Coltheart et al., 1993, Fiez and Petersen, 1998 and Warrington and Shallice, 1980). Processing starts with an analysis of physical surface characteristics of words such as visual contrast, luminance, curves and lines that constitute single letters. Several studies have related this processing stage with an early electrophysiological activation between 50 and 150 ms after stimulus-onset over bilateral occipital regions (called P1 visual component), reflecting neuronal processing in striate and extra striate visual areas (Cornelissen et al., 2003, Dale et al., 2000, Khateb et al., 2002 and Sereno and Rayner, 2003). Next, various prelexical processes develop. At these stages, single letters are identified at abstract representation levels, information about letter position is analysed and automatic extraction of invariant orthographic regularities from alphabetic stimuli is initiated (Grossi & Coch, 2005). These prelexical stages of processing are electrophysiologically reflected by a negative activation mainly over left occipito-temporal scalp regions between 150 and 200 ms, referred to as linguistic N150 component (e.g., Cornelissen et al., 2003, Schendan et al., 1998 and Spironelli and Angrilli, 2007). Other studies demonstrated that alphabetic and non-alphabetic stimuli (words, legal and illegal non-words vs. faces or objects) elicit distinguishable N150 responses (Bentin et al., 1999, Cornelissen et al., 2003, Nobre et al., 1994 and Pammer et al., 2004), and there is also evidence for an early orthographic sensitivity in the N150 response (e.g., Hauk, Davis, Ford, Pulvermüller, & Marslen-Wilson, 2006). At the next processing stage, whole visual word forms are accessed. This lexical stage is characterised by an electrophysiological activity between 200 and 250 ms after stimulus-onset mainly over left-lateralised inferior temporal regions (Cohen et al., 2000, Pammer et al., 2004 and Proverbio et al., 2002), reflecting activation in the left mid fusiform gyrus, the so-called visual word form area (VWFA; Cohen et al., 2000; although VWFA is also engaged in prelexical orthographic processing, see Schurz et al., 2010). Several previous studies suggest that lexical processing is followed by phonological processing, that entails access to lexical phonological representations occurring slightly after the activation of orthographic representations, i.e., starting at around 300 ms after stimulus-onset over central and temporal areas (Bentin et al., 1999, Carreiras et al., 2009, Grainger et al., 2006 and Simon et al., 2004). The last stage in visual word processing involves the activation of semantic representations and is associated with ERP modulations around 400 ms after stimulus-onset over central electrode sites (Holcomb, 1993, Kiefer and Spitzer, 2000, Kutas and Hillyard, 1980, Proverbio et al., 2004 and Van Petten, 1993). In sum, in this sequential view of word processing it is assumed that analysis of whole visual word forms under conscious conditions is completed at around 200–250 ms after stimulus-onset, reflected by electrophysiological activity over posterior electrode sites. This posterior activity precedes more anterior activity characterising processing of phonologic and semantic information. However, it is important to note that results from some ERP studies on the time course and the functional relationship of the various processes during reading are not without controversy. For example, in contrast to the strictly sequential and modular accounts some ERP studies provide evidence that the speed of processing under conscious conditions might be substantial faster than has traditionally been proposed (Foxe and Simpson, 2002 and Thorpe et al., 1996), possibly leading to automatic and parallel access to orthographic, phonological and even semantic information already within 150 ms (e.g., Braun et al., 2009, Pulvermüller et al., 2001 and Sereno et al., 2003). Moreover, such immediate and parallel access to linguistic information has been shown not only in visual, but also in spoken language processing and has been referred to as immediacy principle (e.g., Hagoort, 2008). With respect to non-conscious processing, research over the last three decades using temporal masking paradigms (i.e., very short stimulus presentation combined with preceding and/or following visual pattern masks) and paradigms controlling attention in a way that prevents awareness of information that is visible in principle has demonstrated non-conscious cognitive processing of visually presented linguistic as well as non-linguistic stimuli, but the literature on how far processing under such conditions extends is still inconclusive. Previous behavioural and neurophysiological research suggests that visual information can become non-consciously available at various processing levels, including perceptual (e.g., Dehaene et al., 2001), orthographic (Dehaene et al., 2001, Fairhall et al., 2007 and Grossi and Coch, 2005), phonological (Brysbaert, 2001, Carreiras et al., 2009, Ferrand and Grainger, 1992, Humphreys et al., 1982 and Luo et al., 1998), morphological (Rastle, Davis, Marslen-Wilson, & Tyler, 2000) and even semantic processing stages (e.g., Dehaene et al., 1998, Devlin et al., 2004, Greenwald et al., 1996, Kiefer, 2002, Kiefer and Spitzer, 2000, Mack and Rock, 1998, Marcel, 1983 and Rolke et al., 2001). Nevertheless, while most studies showed non-conscious perceptual and orthographical processing, previous results are still inconsistent not only concerning the phonological level (e.g., Kouider, Dehaene, Jobert, & Le Bihan, 2007), but particularly the semantic level, and some studies convincingly postulated non-semantic interpretations of these latter effects (Abrams and Greenwald, 2000, Damian, 2001 and Eckstein and Perrig, 2007). Aside from the above mentioned debate about the depth of non-conscious word processing, only few neuropsychological studies directly compared the temporal dynamics or localisation of brain activation during conscious and non-conscious word processing. These studies showed mixed results. While some studies report comparable patterns of brain activation (e.g., Kiefer & Spitzer, 2000), others support the view that cognitive processes under conscious and non-conscious conditions are qualitatively separable. For example, Fecteau, Kingstone, and Enns (2004) demonstrated that during non-conscious word reading, right hemispheric brain structures are much more involved compared to conscious word reading. Concerning the time course of brain activation, a recent ERP study by Fairhall et al. (2007) showed non-conscious word–non-word differences that occurred substantially faster than under conscious conditions, suggesting distinct signatures of non-conscious word reading. Whereas most studies on non-conscious word processing use temporal masking techniques, Perrig and Eckstein (2005) employed a spatial mirror-masking technique that prevents conscious perception of letter strings through attentional distraction processes. Masked word and non-word primes used by Perrig and Eckstein consisted of non-descending letters that were mirrored at their baseline. An example of a mirror-masked word can be seen in Fig. 1. This kind of spatial masking results in unfamiliar, nonsense geometric like figures, in which words are no more identifiable as long as they are not expected, i.e., when subjects are not informed about the mirror-masking technique. Therefore it is assumed that mirror-masking causes a state of inattentional blindness. Example of the mirror-masked word house. The principle of the mirror-masking ... Fig. 1. Example of the mirror-masked word house. The principle of the mirror-masking technique is shown on the right side of the figure. Figure options Here, the term “inattentional blindness” does not refer to a specific paradigm. Rather it refers to a phenomenon demonstrated by a number of studies showing that stimuli, visible in principle, are not consciously perceived if attention is distracted and therefore do not receive task-related attention (e.g., Grimes, 1996, Kouider et al., 2007, Mack and Rock, 1998, Neisser, 1979 and Simons and Chabris, 1999). Of course, the status of the subjects’ phenomenological awareness of such stimuli has to be debated and investigated in each study anew. In contrast to temporal masking paradigms, in which establishing real subliminality entails various methodological difficulties (Bernat, Shevrin, & Snodgrass, 2001), distracting attention has the advantage that the unattended stimulus can be presented for a much longer time (Mack and Rock, 1998 and Perrig and Eckstein, 2005). Therefore mirror-masking is an ideal technique to study non-conscious word processing. While conscious perception of visual information is guaranteed, critical information becomes only reportable when expected and attended to. Perrig and Eckstein (2005) implemented in their study a word stem completion task, in which mirror-masked words and non-words were presented supraliminally as primes. Subjects, being ignorant about the masking-technique, demonstrated the expected non-conscious perceptual priming effects. Moreover, when subjects were informed about the nature of the masking-technique and therefore searched for the hidden words, they showed additional semantic priming effects. Thus, with the mirror-masking technique, the influence of non-consciously processed linguistic information on human behaviour can easily be assessed and compared to identical but consciously processed information. To elucidate if linguistic information is processed similarly under conscious and non-conscious conditions, it is beneficial to identify brain structures involved in conscious and non-conscious linguistic stimulus processing. This ERP study relates to both of these issues. Extending the findings of Perrig and Eckstein (2005), the present study was designed to identify brain states that are correlated with different levels of subjects’ phenomenal awareness during mirror-masked word processing. Particularly, we investigated differences in the temporal dynamics and scalp distributions of ERPs under two conditions: First, when letter strings were mirror-masked and subjects were ignorant about the mirror-masking technique, a situation of inattentional blindness (i.e., ignorant condition). Second, when letter strings were masked and subjects were informed about the mirror-masking technique (i.e., when subjects know there might be words; informed condition). In both conditions subjects had to perform lexical decisions while ERPs were recorded. Thus, the conditions differed only with respect to the informational status of the subjects. Based on the findings of the aforementioned studies on non-conscious word processing, we expected to observe differences between words and non-words in the ignorant condition at rather early processing stages around 200 ms over posterior areas, reflecting automatic extraction of orthographic information from the hidden letter strings. For the informed condition, we additionally expected more pronounced ERP differences between words and non-words at later time periods, reflecting conscious access to lexico-semantic properties of mirror-masked words, as the semantic priming effects reported by Perrig and Eckstein (2005) suggest. Comparing the ignorant and the informed condition, it is important to note that although being informed about the mirror-masking technique and therefore expecting the occurrence of words, the initially unattended critical information is not as easily detectable as known from inattention paradigms (e.g., Mack & Rock, 1998). As outlined above, processing words written in a familiar visual format relies on the left occipito-temporal cortex as a part of the ventral visual pathway. However, recent research suggests that reading words presented in a visually degraded form resulting from rotation or spacing involves additional activation of the posterior parietal cortex, reflecting involvement of supplementary visual attentional resources (Cohen et al., 2008 and Rosazza et al., 2009). Therefore we expected to see stronger activations in these areas when subjects were informed about the masking technique compared to the ignorant condition. Since no precise information about the nature of the expected effects was available, we used global (across all electrodes) topographical comparisons, reducing the possibility of false positive results due to redundant testing. Distributed source localisation methods were used to localise effects in brain space. To identify and compare differences in the timing of processing, we parsed the ERPs into microstates, which are periods of quasi-stable field topography assumingly corresponding to different processing steps (Michel et al., 2001 and Pascual-Marqui et al., 1995).

. Results 3.1. Behavioural data In the naïve part of the experiment, three subjects (12%) recognised at least one word in the geometric like figures and therefore were excluded from all further analysis. For the remaining subjects (N = 22), mean reaction times and accuracy for all conditions are summarised in Table 1. Table 1. Mean reaction times in milliseconds and accuracy (SD) for all conditions. Naïve (N = 22) Informed (N = 22) High performers (N = 12) Reaction times Accuracy Reaction times Accuracy Reaction times Accuracy Masked non-words 451 (69) .99 (.01)⁎ 838 (223) .85 (.15) 908 (156) .80 (.13) Masked words 451 (64) .99 (.02) 934 (207) .26 (.22) 973 (167) .41 (.19) ⁎ Note that the accuracy of .99 for masked words in the naïve part of the experiment means that the participants simply classified the masked words correctly as “not words”. Table options For masked stimuli the 2 × 2 repeated-measure ANOVA with the within-subject factors “status of information” (ignorant, informed) and “lexical status” (word, non-word) yielded significant main effects for both factors as well as a significant interaction for reaction times and accuracy (F-values between 20.11 and 444.23, all p-values < .01). As long as the subjects were ignorant about the masking technique, there was no difference in reaction times and accuracy for masked words and non-words (F(1, 21) < 1, p = .98 resp. F(1, 21) = .88, p = .36). The effect of lexical status was only present after informing the subjects, resulting in slower reaction times and reduced accuracy for masked words compared to masked non-words (F(1, 21) = 19.91, p < .01 resp. F(1, 21) = 74.16, p < .01). Importantly, after informing the subjects, mean accuracy (hit-rate) for masked words dropped to only .26 (SD = .22; i.e., on average 13.09 correctly classified masked words). Thus, after informing subjects about the mirror-masking technique (i.e., in the informed part) reaction times were about 430 ms slower and accuracy was poorer compared to the naïve part, reflecting the changed task requirements. In order to get reliable ERP-differences for correctly classified masked words (hits) and non-words (correct rejections) in the informed part of the experiment, data from a subgroup of subjects (N = 12; “high performers”) with at least 15 hits and correct rejections were analysed separately. For this “high performers”, mean accuracy for masked words and non-words was .41 and .80 respectively (SD = .19 and .13). 3.2. ERP-analysis For the microstate analysis, the different topographies of grand mean ERPs are shown in the upper part of Fig. 2. Based on a silhouette plot analysis, an optimal number of eight different microstate classes was obtained. These microstates were ordered in sequence of their appearance and labelled from A to H. (Upper part) Microstate topographies in the sequence of occurrence. Head seen ... Fig. 2. (Upper part) Microstate topographies in the sequence of occurrence. Head seen from above, red indicates positive, blue negative values, referred to average reference. (Lower part) Microstate assignment across time in the naïve part of the experiment (N = 22) for masked words (upper graph) and masked non-words (lower graph) plotted over electric field strength (GFP). Colours refer to the microstate topographies shown in the upper part of the figure. The grey bar indicates the time segment of significant topographic differences between the two conditions as revealed by TANOVA (112–160 ms after stimulus presentation). Figure options Microstate topography A showed the characteristic P100 distribution with positive potentials over bilateral occipital areas. Microstate B reflected a N150-like component as shown by the bilateral occipito-temporal negativity. The microstates from C to E displayed a P300-like posterior positivity that slightly shifts from parietal to central locations and an anterior negativity shifting from a pre-central to more ventral locations. Starting with microstate E, the later microstates showed a bilateral occipital negativity (microstate F) extending to bilateral temporal areas (microstates G and H), while the positivity shifts from central to bilateral frontal electrodes. The lower part of Fig. 2 shows the sequences of microstates for masked words and non-words plotted over the mean GFP. Confirming visual inspection, statistical comparisons revealed no significant differences in microstate sequences, onset latencies, durations and GFPs for words and non-words. 3.2.1. Masked condition ignorant Randomisation tests on microstate onset latencies and durations as well as analysis of mean GFP within each microstate displayed no significant differences between masked words and non-words. However, the TANOVA identified significantly different global topographical distributions between the two conditions in the N150-like microstate (microstate B), lasting from 112 to 160 ms (p < .05; grey bar in Fig. 2). The onset of this difference coincided with the transition of the P1-like microstate A to the N150-like microstate B and continued to the GFP peak for masked words at around 160 ms. Concerning local differences for this critical time period, the t-map of averaged ERPs contrasting words against non-words showed a stronger a right-lateralised posterior negativity and a stronger bilateral frontal positivity for words (t-min = −2.56, p < .05 at electrode PO10; and t-max = 3.10, p < .05 at electrode F5; upper part of Fig. 3). (Upper part) Interpolated grand mean maps for masked words (left) and masked ... Fig. 3. (Upper part) Interpolated grand mean maps for masked words (left) and masked non-words (right) 112–160 ms after stimulus presentation in the naïve part of the experiment (N = 22). The map in the middle shows the t-map of the mean differences between masked words and non-words. Head seen from above, red indicates positive, blue negative values, referred to average reference. (Lower part) sLORETA source estimates of ERP differences between masked words and non-words (112–160 ms after stimulus presentation). The left side displays a lateral view of the right hemisphere of the brain. The right part shows a sagittal view of the brain (view on the right medial hemisphere). RH = Right hemisphere; SV = Sagittal view. Figure options The sLORETA source estimates for the critical time period revealed greater current density for masked words compared to masked non-words mainly in restricted areas of the right parietal lobe (lower part of Fig. 3). Maximal current density for masked words was located in the right inferior supramarginal gyrus (BA 40; t = 3.51, p < .05), while they elicited less current density in medial frontal areas, particularly in the anterior cingulate cortex of the right hemisphere (BA 32; t = −3.28, p < .05) and in the medial frontal gyrus (BA 6 and 9; t = −2.55, p < .05). Further significant differences in activation were found in right posterior parietal (BA 7; t = 2.54, p < .05) and bilateral pre-central (BA 6) t = 2.47, p < .05) regions. Higher activations for masked words were also observable in left occipito-temporal areas (BA 37; t = 2.05, p > .05), although these differences did not reach significance. 3.2.2. Masked condition informed After informing subjects about the mirror-masking technique the microstate analysis for the “high performers” confirmed a similar sequence of microstates for correctly classified masked words and non-words (Fig. 4). Microstate assignment across time in the informed part of the experiment for ... Fig. 4. Microstate assignment across time in the informed part of the experiment for correctly classified masked words (upper graph) and masked non-words (lower graph) plotted over electric field strength (GFP) (N = 12). The upper part of the figure shows microstate topographies in the sequence of occurrence similar to those in Fig. 2. Figure options Additionally, randomization tests revealed a shorter duration of microstate F for non-words (p < .01). Furthermore the onset latency of microstate G was significantly shorter for non-words (p < .01). Mean GFP comparisons showed significantly higher amplitudes for words than non-words in microstate A and, in microstate F between 548 and 724 ms as well as between 724 and 880 ms (resp. microstate G for non-words; all ps < .01). TANOVA yielded significant topographical differences in the following three time periods: 332–360 ms, 492–528 ms and 700–804 ms (all ps < .01). In contrast, no significant topographical differences between words and non-words were found in the low performing subgroup of subjects. The t-map for the time period from 332 to 360 ms (left part of Fig. 5) showed a parieto-occipital negativity in the left hemisphere for masked words (p < .05). T-maps for the time periods from 492 to 528 ms and from 700 to 804 ms displayed a right-lateralised occipito-temporal negativity accompanied by a slightly right-lateralised central positivity for masked words. (Left part) Interpolated grand mean maps for correctly classified masked words ... Fig. 5. (Left part) Interpolated grand mean maps for correctly classified masked words and masked non-words in three different time segments (informed part of the experiment; N = 12). Between these voltage maps, the t-maps of the mean differences are displayed. Head seen from above, red indicates positive, blue negative values, referred to average reference. (Right part) sLORETA source estimates of differences between correctly classified masked words and masked non-words. BV = Basal view; RH = Right hemisphere; SV = Sagittal view. Figure options According to sLORETA solutions (right part of Fig. 5, differences for the 332–360 ms segment were localised in the left fusiform gyrus (BA 19 and 37; t = 2.55, p < .05), posterior middle temporal gyrus (BA 39; t = 2.49, p < .05), left postcentral gyrus (BA 43; t = 3.69, p < .05), right supramarginal gyrus (BA 40; t = −4.57, p < .05) and bilateral inferior frontal (BA 47; t = −3.43, p < .05) areas. During the 492–528 ms and 700–804 ms time segments, significant differences were found amongst others in left inferior parietal (BA 40; t = 2.72, p < .05), bilateral temporal (BA 21; t = 2.60, p < .05) and bilateral medial frontal (BA 6; t = −3.15, p < .05) areas. 3.2.3. General effect of “status of information” TANOVA for all subjects (N = 22) between masked stimuli (both words and non-words) presented before and after the information about the mirror-masking technique yielded significant topographical differences between 8 and 88 ms as well as between 176 and 804 ms. SLORETA solutions within this latter time segment yielded greater current density for masked stimuli in the informed part compared to masked stimuli in the naïve part of the experiment in bilateral posterior parieto-occipital areas (i.e., left cuneus; t-max = 6.14, p < .05; Fig. 6). (Upper part) Interpolated grand mean maps for masked stimuli in the informed ... Fig. 6. (Upper part) Interpolated grand mean maps for masked stimuli in the informed part (left) and the naïve part (right) of the experiment (N = 22). The map in the middle shows the t-map of the mean differences between the two conditions. Head seen from above, red indicates positive, blue negative values, referred to average reference. (Lower part) sLORETA source estimates of ERP differences between the two conditions. DV = Dorsal view.