ARTICLE IN PRESS Biological Psychology xxx (2012) xxx xxx

Transcription

1 G Model ARTICLE IN PRESS Biological Psychology xxx (2012) xxx xxx Contents lists available at SciVerse ScienceDirect Biological Psychology journa l h o me page: Emotion, Etmnooi, or Emitoon? Faster lexical access to emotional than to neutral words during reading Johanna Kissler a,b,, Cornelia Herbert a,c a Department of Psychology, University of Konstanz, Konstanz, Germany b Department of Psychology, University of Bielefeld, Bielefeld, Germany c Department of Psychology, University of Würzburg, Würzburg, Germany a r t i c l e i n f o Article history: Received 3 August 2011 Accepted 26 September 2012 Available online xxx Keywords: Emotion Attention Word processing Lexical access ERPs Early posterior negativity a b s t r a c t Cortical processing of emotional words differs from that of neutral words. Using EEG event-related potentials (ERPs), the present study examines the functional stage(s) of this differentiation. Positive, negative, and neutral nouns were randomly mixed with pseudowords and letter strings derived from words within each valence and presented for reading while participants EEG was recorded. Results indicated emotion effects in the N1 ( ms), early posterior negativity (EPN, ) and late positive potential (LPP, ms) time windows. Across valence, orthographic word-form effects occurred from about 180 ms after stimulus presentation. Crucially, in emotional words, lexicality effects (real words versus pseudowords) were identified from 216 ms, words being more negative over posterior cortex, coinciding with EPN effects, whereas neutral words differed from pseudowords only after 320 ms. Emotional content affects word processing at pre-lexical, lexical and post-lexical levels, but remarkably lexical access to emotional words is faster than access to neutral words Elsevier B.V. All rights reserved. 1. Introduction Visual processing of emotional words differs from visual processing of neutral words. Using electroencephalographic eventrelated potentials (ERP) emotional neutral differences have been identified at various temporal stages following word onset in lexical-decision, evaluation, or reading tasks (for review see Kissler et al., 2006). Several ERP studies have described more negativegoing ERPs for emotional words over occipital cortex between 200 and 300 ms (Herbert et al., 2008; Kissler et al., 2007; Kissler et al., 2009; Scott et al., 2009) or between 300 and 400 ms (Palazova et al., 2011; Schacht and Sommer, 2009a, 2009b) after word presentation. Collectively, these effects are referred to as emotion driven early posterior negativities (EPN). Morphologically analogous effects have been reported in emotional face (Schupp et al., 2004), scene (Junghofer et al., 2001), or gesture processing (Flaisch et al., 2011), and in explicit object-based attention tasks (Hillyard and Anllo-Vento, 1998; Schoenfeld et al., 2007), implying that early emotion effects in this time window reflect attentional highlighting of emotional stimuli in general (Schupp et al., 2006). Corresponding author at: Department of Psychology, University of Bielefeld, Postfach , Bielefeld, Germany. address: johanna.kissler@uni-bielefeld.de (J. Kissler). Larger parietal positivities arising around 500 ms after word onset have also been reported in emotion word processing (Fischler and Bradley, 2006; Herbert et al., 2006; Schacht and Sommer, 2009b), as well as in the processing of emotional scenes, faces, or gestures. Again, parallel effects are found in explicit attention tasks. Occasionally, effects of emotional content on the N400 component, a classic index of contextual semantic integration, have been found (Herbert et al., 2008; Kiehl et al., 1999), suggesting facilitated semantic integration of emotional words after a first initial attentional processing stage. Also, emotion effects arising earlier than 200 ms after word onset have been documented (Begleiter and Platz, 1969; Hofmann et al., 2009; Ortigue et al., 2004; Skrandies, 1998; Skrandies et al., 1998). These very early effects, in particular, support the view that emotional processing can operate even pre-attentively, outside the boundaries set by other informationprocessing mechanisms, as proposed by automatic vigilance (Pratto and John, 1991) or automatic evaluation models (Zajonc, 1980). In the context of language processing, such very early effects suggest the possibility of pre-lexical responses to emotional content. At least under conditions of low to moderate competing cognitive load, emotional stimuli, including words, are preferentially processed. Still, it is debated whether ERP emotion effects are specific to a particular category or dimension of emotion. Whereas much data suggest a major impact of emotional intensity, i.e. stimulus arousal, on emotion word processing (Fischler and Bradley, 2006; Kissler et al., 2007; Schacht and Sommer, 2009b), some /$ see front matter 2012 Elsevier B.V. All rights reserved.

2 ARTICLE IN PRESS G Model 2 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx results indicate specific effects for negative (Bernat et al., 2001) or positive words (Kissler and Koessler, 2011). These discrepancies are reflected in different theoretical accounts. According to one model, rapid automatic allocation of attention specifically to negative stimuli is biologically adaptive in facilitating rapid withdrawal from potentially dangerous environments. Therefore this mechanism may by-pass other cognitive processes (Pratto and John, 1991). Alternatively, attentional orienting to emotionally arousing stimuli in general may be needed to mobilize appropriate approach or avoidance behavior (Lang et al., 1997). Finally, appraisal theory proposes a cascade of processing steps where sequential checks influenced by situational demands determine the patterning of emotional responses, allowing for more flexibility and perhaps reconciling some empirical discrepancies (Grandjean and Scherer, 2008; Scherer, 2009). As outlined above, ERP effects of emotion in vision are often assumed to reflect spontaneous attentional highlighting of motivationally relevant emotional stimuli. This assumption is supported by analogous ERP effects in object-based selective attention tasks and in line with the general concept of emotionally motivated attention (Lang et al., 1997). Attention and emotion can affect the processing of perceptual objects, including words, at various temporal stages (Luck and Hillyard, 1999; Ruz and Nobre, 2008; Schupp et al., 2007; Vogel et al., 2005; Ziegler et al., 1997). Still, it is controversial, how much perceptual and cognitive processing has to be carried out, before a stimulus emotional significance is identified and how early in the processing stream emotion effects can occur. In other words, an important question still is: How much, if any, inference do preferences need? In emotion word processing the question can be put as: When do emotion effects occur in relation to different processing stages in the mental lexicon? To answer this question, emotion effects can be pitched against a time-line of word recognition derived from models of visual word processing. According to classic serial word processing models (e.g. Coltheart et al., 2001; McClelland and Rumelhart, 1981), increasingly abstract information, including orthographic word form and phonological and lexical properties of a word, is extracted from the visual percept as activation travels from primary visual to higher order, multimodal, association areas of the brain, where information is combined to achieve full comprehension of a word s meaning (for review see Dien, 2009). Word-processing stages are assumed to partly overlap and to be organized in an interactive and cascaded fashion, but the extent of parallel processing is a matter of considerable debate (Barber and Kutas, 2007; Grainger and Holcomb, 2009; Pulvermuller et al., 2009). Traditional models agree that some perceptual invariance has to be extracted from the physical signal, before a word s meaning can be accessed. It is traditionally assumed that orthographic processes take place within the first 250 ms, followed by lexical and semantic access from about 300 ms after stimulus onset, culminating in full semantic contextual integration around 400 ms (Grainger and Holcomb, 2009). However, some models argue for a considerably higher speed of word recognition with parallel or near parallel processing of different attributes and present evidence for semantic processing even within the first 200 ms (e.g. Pulvermuller et al., 2009). Because models and data differ regarding the speed and stages of word processing itself, identification of the locus of emotion effects along an empirically determined time-line of word processing is needed to inform models of word recognition and emotional processing. From a linguistic perspective, the extent to which emotional content can accelerate or by-pass other stages of word processing has implications for models of lexical access. Acceleration of word processing by emotion is suggested by consistent evidence of faster lexical decisions to emotional than to neutral words (Kousta et al., 2009; Schacht and Sommer, 2009b), but in lexical-decision emotional response facilitation may also play a role (Kissler and Koessler, 2011). Moreover, models such as the automatic vigilance model, developed from experiments with word stimuli, assume that emotion processing can operate preattentively (Pratto and John, 1991), suggesting the existence of short-cut routes, by-passing stages of perceptual analysis. A linguistic challenge is to account for such effects within models of word recognition and to determine whether they are unique to emotion words or also occur for other semantic classes, and whether they could be mimicked by attention manipulations. From the perspective of emotion theory, data will provide information regarding the temporal evolution of emotion cognition interactions and the direction of emotion effects at distinct processing stages. On-line measures such as ERPs can reveal such sequences that might be obscured in lexical-decision reaction times alone, which represent a compound measure, integrating the results of many different operations. So far, two ERP studies investigated the pre-, peri-, or postlexical status of emotion effects in word processing. Scott et al. (Scott et al., 2009) varied emotional content within high- and lowfrequency words in a lexical-decision task. Since frequency effects in visual word processing are considered indicative of lexical access, word frequency by emotion interactions was identified to determine the functional status of emotion effects. Interactions occurred on the N1 around 150 ms and in the subsequent EPN window ( ms). These N1 and EPN effects were assumed to indicate modulations of lexical access by emotional content. An earlier P1 amplitude reduction for negative words was interpreted as a pre-lexical perceptual defense mechanism. Palazova et al. (2011) examined ERP emotion effects in high- and low-frequency adjectives, verbs, and nouns, likewise using a lexical-decision task. Results confirmed an early frequency effect around ms which interacted with emotion. Word pseudoword differences were found between 250 and 550 ms. Crucially, main effects of emotion were reflected by an EPN potential and a centro-parietal positivity (LPP), whose temporal onset largely coincided (EPN) or followed (LPP) word pseudoword differentiation, supporting a lexical or post-lexical status of emotion effects. The present study further investigates the status of emotion effects along the time-line of word processing using a silent reading paradigm: For adults, silent reading is a highly automatic and very natural process (e.g. Kahneman and Chajczyk, 1983). Simultaneously measuring on-going brain activity during reading can reveal emotional modulations irrespective of additional effects of response preparation and execution, which may differ from wordprocessing-related activity itself. Lexical-decision reaction times are consistently faster for emotion words, with some controversy surrounding the relative role of valence versus arousal (Estes and Adelman, 2008; Kousta et al., 2009). However, in lexical decision many different processes could be affected by emotion. Moreover, lexical decision, while highly successful in identifying word nonword differences, draws explicit attention to a stimulus lexical status, thereby potentially affecting lexicality or emotion effects. In fact, Ziegler et al. (1997) demonstrated that the time course of lexical and semantic activation in word processing is task-dependent, with faster ERP word pseudoword differentiation in a semantic categorization task than during letter search. Here, emotion effects in reading are investigated in relation to effects of orthography and lexicality to determine the temporal and functional stage(s) of effects. We present positive, negative, and neutral nouns intermixed with letter strings and pseudowords to identify successively more refined processing stages. Under a serial model, differentiation of orthographically legal word forms (words or pseudowords) from illegal words forms (letter strings) should precede differentiation of real words (with a corresponding lexical

3 G Model ARTICLE IN PRESS J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx 3 Table 1 Means and standard deviations (in brackets) of different stimulus dimensions. Word stimuli Positive Neutral Negative Valence 7.46 (.92) 5.20 (.36) 2.02 (.72) Arousal 5.84 (1.02) 2.18 (.77) 5.72 (.79) Concreteness 4.55 (1.16) 4.05 (2.04) 4.20 (1.42) Word Length 7.28 (2.70) 7.13 (1.97) 6.83 (1.95) Word Frequency (144.01) (20.91) (173.05) entry) from pseudowords (without lexical entry). Emotion effects will be assessed along the time-line of such effects. Our previous studies have used the rapid serial visual presentation (RSVP) technique to study emotion effects in word processing. Although effects have been largely replicated in lexical-decision tasks (Hinojosa et al., 2010; Palazova et al., 2011; Schacht and Sommer, 2009a, 2009b; Scott et al., 2009), RSVP with its absence of baseline periods and inherent conceptual masking, may create a special experimental situation. Here, we examine emotion effects in single word reading outside RSVP. In particular, we address the timing of emotion effects in relation to orthographic word-form analysis and lexical access in positive, negative, and neutral words, investigating the possibility that emotional content may accelerate or by-pass some of these processes. 2. Methods 2.1. Participants Twenty-four native German students (12 women) from the University of Konstanz, Germany, took part in the experiment. Mean age was (SE =.62) years. All were right-handed according to the Edinburgh Handedness Inventory (Oldfield, 1971). They reported no history of neurological or psychiatric disease and their vision was normal or corrected to normal. Subjects signed written informed consent forms and received either a financial bonus of 7.50 D ( $) or course credit for their participation Material 138 German nouns (46 high arousing positive, 46 high arousing negative, 46 neutral) were used. From these nouns, for each of the three emotion categories, 46 pseudowords and 46 letter strings were created by within- and between-word permutation, resulting in a total of 138 words, 138 pseudowords and 138 letter strings as experimental stimuli. Positive and negative nouns differed in valence but not in arousal. They differed from neutral nouns in both valence and arousal. Words were matched across emotion categories for concreteness, length and word frequency and did not differ in any of these. Arousal, valence, and concreteness values were based on self-collected nine-point scale ratings given by 45 students who did not take part in this experiment. Frequency of written words was determined from the standardized word-database CELEX (Baayen et al., 1995) (see Table 1 for a summary of stimulus characteristics). Pseudowords were based on the original words within each emotion category and were generated by letter permutations within and between words to preclude simple perceptual repair and at the same time maintain compliance to orthographic and phonotactic rules of German. Original word length was never altered. Letter strings were likewise generated by within- and between-word letter permutation, but now violated German orthographic and phonotactic rules. The words, pseudowords, and letter strings, used in the experiment are listed in Appendix A Procedure Participants were familiarized with the laboratory, handedness was assessed (Oldfield, 1971) and they were asked about their past and current health using a standardized questionnaire. They were then seated in an electrically shielded room and a geodesic net containing 256 EEG electrodes was positioned on their head (GSN 200 v2.0; EGI: Electrical Geodesics, Inc., Eugene, Oregon). Participants were informed that they were taking part in a word-processing study and that their task was to read a sequence of words and word-like stimuli while their EEG was being recorded. They were instructed to refrain from head and eye movements during stimulation as much as possible and attend to each stimulus for its entire presentation time. Stimuli were presented for 600 ms, separated by a pseudo-randomly varying inter-stimulus interval of ms during which a fixation cross was shown. Stimulus sequence was constrained such that overall transition probability was equal across stimuli and that no more than three exemplars of the same Stimulus Type (word, pseudoword, letter string) appeared in sequence EEG recording and analysis EEG was recorded from 256 channels using EGI amplifiers and Netstation software (GSN 200 v2.0; EGI: Electrical Geodesics, Inc., Eugene, Oregon). Recording bandwidth was Hz; sampling rate was 250 Hz. To reduce mains interference, a 50 Hz notch filter was used. Impedance was held beneath 50 k and Cz was used as recording reference. Off-line, data were re-referenced to an average reference and band-pass filtered between 0.1 and 35 Hz. Eye movement and blink artifacts were corrected using the correction algorithm implemented in BESA (Brain Electrical Source Analysis, MEGIS Software GmbH). Remaining artifacts were reduced by individual channel-interpolation or the epochs were rejected (max. 10%). Data were segmented from 100 to 700 ms, baseline corrected using the first 100 ms, and averaged according to condition EEG analysis The Matlab-based EMEGs v2.2 package (Peyk et al., 2011) was used for ERP visualization and analysis. To specify time windows and regions of interest, point-wise ANOVAs were conducted across all time-points and sensors. In a first step, differences between positive, negative, and neutral words were determined to replicate previous effects. Then, effects of Stimulus Type (word, pseudoword, letter string) were examined within each individual valence to assess orthographic and lexical processing stages for each valence, fully controlling for basic perceptual characteristics (note that pseudowords and letter strings were created by permutation within individual valence). Separate analyses were conducted to account for the possibility that the temporal dynamics of lexical processes differ between neutral, negative, and positive words. Results were considered meaningful, if effects remained significant at a significance level of p <.01 for at least 8 sample points (32 ms) and could be seen, without additional spatial or temporal smoothing, in a cluster of at least 10 electrodes (Kissler and Koessler, 2011). Main effects were decomposed using linear contrasts or quadratic trends, depending on the hypothesis to be tested. Statistical maps for contrasts were interpreted only in time windows that first revealed significant topographically overlapping effects in the ANOVA. Such effects are marked with black boxes in the figures, although the statistical maps for the contrasts are shown across the entire time window to provide full information. Additionally, based on regions and time windows of interest typically reported in the literature and identified in the point-wise ANOVAs, effects were confirmed using ANOVAs assessing the effect of Valence (positive, negative, neutral) in words and the effect of Stimulus Type (Words, Pseudowords, Letter strings) within valence. ANOVAs compared mean activity in clusters of sensors covering regions and time windows of interests and were decomposed using t-test. The same clusters were used across valences, and sensor clusters were kept constant across effects as much as possible. Where appropriate according to previous reports, Hemisphere was included as a factor, comparing ERPs at two symmetrical left and right hemispheric electrode clusters. Sensor clusters are listed in Appendix B. 3. Results 3.1. Emotion in words Point-wise ANOVA Fig. 1 displays p-maps of the time course and topography for the main effect of Emotion in nouns (Fig. 1a), for the quadratic trend comparing emotional (positive and negative) with neutral nouns, testing the hypothesis that arousal accounts for emotional neutral differences in reading (Fig. 1b), and the linear contrast testing for valence-specific differences by comparing positive and negative nouns (Fig. 1c). ERP difference topographies for these effects are shown in Fig. 2 for negative minus neutral (a) and positive minus neutral (b) and positive minus negative nouns (c). ANOVA revealed a first significant effect between 108 and 140 ms over left centro-occipital scalp. This effect is apparent both in the ANOVA (Fig. 1a) and in the quadratic trend comparing emotional with neutral words (Fig. 1 b), but not in the linear contrast between positive and negative stimuli (Fig. 1c), indicating mainly an arousal effect. Still, the difference topographies show it to be driven mainly by negative nouns eliciting more negative-going ERPs than neutral nouns between 108 and 140 ms (Fig. 2a). Further emotion effects evolve from 216 ms after word onset, extending throughout the entire analysis window (Fig. 1a and b). They start over left lateral scalp for both negative and positive nouns compared to neutral ones and are due to a more pronounced posterior negativity and concomitant frontal positivity for

4 ARTICLE IN PRESS G Model 4 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx Fig. 1. Probability maps from 0 to 680 ms in steps of 32 ms (on average p <.01 for at least 32 ms) depicting the effects of emotional content on ERPs during reading. (a) Point-wise ANOVA comparing positive, negative, and neutral words. (b) Quadratic trends comparing ERPs elicited by positive and negative with neutral words. (c) t-tests contrasting positive and negative words. Black frames mark time windows that were significant already in the ANOVA and can therefore be interpreted in the post hoc comparison. emotional than for neutral words (Fig. 2a and b). This process mirrors the previously described EPN effect for emotional words. Directly comparing ERPs to positive and to negative words (Figs. 1c and 2c) reveals that from about 400 ms positive words elicit more right occipito-parietal positivity than negative or neutral words. This replicates the previously documented LPP effect, presently restricted to positive content Spatio-temporal cluster ANOVA: emotion effect in words ms Emotion was significant at two bilateral occipito-parietal electrode groups (F(2, 46) = 4.15, p <.05) due to more negative-going brain potentials for negative than for neutral words (p <.05). The other categories did not differ and although the effect appeared somewhat left lateralized, the interaction was far from significant (F < 1) ms Further Emotion effects (F(2, 46) = 4.84, p <.05) arose at two groups of occipito-parietal electrodes between 216 and 320 ms with both positive and negative words differing from neutral ones (ps <.05), but not from each other (F < 1). ERPs were generally more negative over the left hemisphere (F(1, 23) = 6.53, p <.05), but this did not interact with valence (F(2, 46) = 1.28, p <.19) ms Emotion further impacted on visual processing between 324 and 392 ms (F(2, 46) = 3.12, p =.05) due to both negative (p <.05) and positive (p <.05) words differing from neutral words. There was no effect of Hemisphere or an interaction between Emotion and Hemisphere (both p >.2) ms A centro-parietal effect of emotion (F(2, 46) = 4.72, p <.05) emerged between 468 and 608 ms. This was due to more positivegoing ERPs for positive than for negative words (p <.05). The effect differed between the hemispheres (F(2, 46) = 4.64, p <.05). Over the left hemisphere, both positive and neutral words had more positive-going ERPs than negative words (ps <.05), whereas over the right hemisphere positive words had more positive going ERPs than both positive and neutral words (ps <.01), the latter also eliciting more positivity than negative words (p <.05) Effects of Stimulus Type: neutral valence Point-wise ANOVA To determine the functional stage(s) at which emotion effects occur in neutral, negative, or positive words, the effect of Stimulus Type (Word, Pseudoword, Letter string) was analyzed within each valence. Fig. 3a displays the result of a point-wise ANOVA

5 G Model ARTICLE IN PRESS J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx 5 Fig. 2. Topographic difference maps of ERPs elicited by words with different emotional content from 0 to 680 ms in steps of 32 ms. (a) Negative words minus neutral words. (b) Positive words minus neutral words. (c) Positive words minus negative words. Black frames mark time windows where significant effects were found in the respective statistical comparisons. comparing ERPs elicited by neutral words with pseudowords and letter strings Fig. 3b and c display the p-maps for the linear contrasts. First effects of Stimulus Type are seen over inferior occipital scalp between 180 and 248 ms. Further effects arise over right occipital scalp from 288 ms, extending bilaterally and more superior from 324 ms. Centro-parietal effects of Stimulus Type are seen from around 432 ms. The underlying difference topographies are shown in Figs. 6a and 7a Pseudowords versus letter strings. Differentiation between pseudowords and letter strings, reflecting a formal analysis stage, is first seen between 180 and 249 ms after stimulus onset (Fig. 3c). Pseudowords elicit more positive-going ERPs over bilateral inferior occipital scalp (Fig. 7a). A second significant difference between pseudowords and letter strings arises from 288 ms as a prolonged right occipital negativity for pseudowords compared to letter strings successively extending over the entire back of the brain (Fig. 7a). Finally, from 540 ms a left frontal positivity accompanies the retreating right occipital negativity (Fig. 7a) Words versus pseudowords. Linear contrasts reveal a difference between neutral words and pseudowords starting from 324 ms (Fig. 3b) as a left occipital negativity and right frontal positivity (Fig. 6a). This difference extends until 500 ms (Fig. 3b), evolving from a left occipito-temporal negativity into a central positivity for neutral words, resembling the N400 effect. This process extends for several hundred milliseconds, retreating over occipital cortex, where it reaches significance again, between 576 and 608 ms. Finally, a slightly left frontal negativity develops for words compared to pseudowords, significant between 648 and 680 ms after stimulus onset Spatio-temporal cluster ANOVA: effect of Stimulus Type for neutral valence To further validate the above effects, a region of interest analysis was conducted for individual time windows for the first 500 ms. Figs. 6a and 7a depict the underlying topographical differences. For consistency and because of considerable spatial overlap in the effects, all posterior effects were analyzed at the same channel groups as the above emotion effects ms An effect of Stimulus Type (F(2, 46) = 9.68, p <.01) was identified at two bilateral inferior-occipital electrode groups. It was due to more positive-going brain potentials for neutral words and pseudowords than for letter strings (p <.01 and p =.01, respectively). Words and pseudowords did not differ (p >.5) ms Stimulus Type (F(2, 46) = 5.25, p <.01) was significant: Pseudowords, and in tendency also words, were more positive than letter strings (p <.05 and p =.07, respectively, Fig. 7a).

6 ARTICLE IN PRESS G Model 6 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx Fig. 3. Probability maps from 0 to 680 ms in steps of 32 ms (on average p <.01 for at least 32 ms) depicting the effects of Stimulus Type on ERPs during reading of neutral words, pseudoword, or letter strings derived from neutral words. (a) Point-wise ANOVA comparing words, pseudowords, and letter strings. (b) t-tests comparing words and pseudowords. (c) t-tests contrasting pseudowords and letter strings. Black frames mark time windows that were significant already in the ANOVA and can therefore be interpreted in the post hoc comparison ms A significant effect of Hemisphere was due to more negativegoing potentials over the left hemisphere (F(1, 23) = 5.99, p <.05), but no effects of Stimulus Type (F < 1) or interactions were found (F < 1) ms A main effect of Stimulus Type (F(2, 46) = 8.86, p <.01) occurred with more negative-going potentials for words than for letter strings (p <.01). Pseudowords were likewise somewhat more negative than letter strings (p =.08) and did not differ from words (p >.2). ERPs were somewhat more negative over the left hemisphere (F(1, 23) = 2.82, p =.1), but the interaction was far from significant (F <.1) ms Between 324 and 392 ms a further effect of Stimulus Type emerged (F(2,46) = 13.43; p <.01), now differentiating words from pseudowords (Fig. 6a). Words were more negative than both pseudowords (p <.05) and letter strings (p <.01). The latter did not differ. Effects were somewhat larger over left than over right occipital areas (F(2, 46) = 2.64, p =.08) ms Finally, a centro-parietal effect was tested in the typical N400 time range. This yielded a main effect of Stimulus Type (F(2,46) = 10.12; p < 0.01). Figs. 6a and 7a) reveal that pseudowords were associated with more negative-going ERPs than both words (p <.01) and letter strings (p <.05). In tendency, letter strings were also more negative than words (p =.06) Effects of Stimulus Type: negative valence Point-wise ANOVA Fig. 4a displays the results of a point-wise ANOVA comparing ERPs for negative words and pseudowords and letter strings. First effects of Stimulus Type are seen over right inferior occipital scalp from 180 ms. From 216 ms these effect extend over the left hemisphere, spreading over bilateral occipital and frontal scalp by 392 ms. From 396 ms effects are seen fronto-centrally, lasting until about 540 ms. Underlying topographical differences are shown in Figs. 6b and 7b Pseudowords versus letter strings. Pseudowords and letter strings first differ right-occipitally from 180 after stimulus onset (Fig. 4 c). By 284 ms, more positive going ERPs are elicited by pseudowords, primarily over right inferior occipital scalp (Fig. 7 b). From 288 ms a prolonged bilateral occipital negativity for pseudowords compared to letter strings successively extends over the entire back of the brain (Fig. 7b). Finally, from 540 ms a left frontal positivity accompanies the retreating right occipital negativity (Fig. 7a).

7 G Model ARTICLE IN PRESS J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx 7 Fig. 4. Probability maps from 0 to 680 ms in steps of 32 ms (on average p <.01 for at least 32 ms) depicting the effects of Stimulus Type on ERPs during reading of negative words, pseudoword, or letter strings derived from negative words. (a) Point-wise ANOVA comparing words, pseudowords, and letter strings. (b) t-tests comparing words and pseudowords. (c) t-tests contrasting pseudowords and letter strings. Black frames mark time windows that were significant already in the ANOVA and can therefore be interpreted in the post hoc comparison Words versus pseudowords. Differentiation between negative words and pseudowords starts from 252 ms (Fig. 4 b) as a left occipital negativity and left frontal positivity (Fig. 6b) lasting until 356 ms. From 360 ms a left frontal positivity develops that gradually extends posterior, spreading over much of the central scalp by 432 ms and by 536 ms retreating to left frontal regions. This process resembles the N400 effect Spatio-temporal cluster ANOVAs: effect of Stimulus Type for negative valence To further validate the above effects, a regions of interest analysis was conducted for the same sensor clusters and time windows as above. Topographical differences underlying these effects are shown in Figs. 6b and 7b ms A first effect of Stimulus Type (F(2, 46) = 3.79, p <.05) was due to somewhat more negative-going brain potentials for letter strings than for negative words and pseudowords (p <.1 and p <.2, respectively). Words and pseudowords did not differ (F < 1) ms No clear effects of Stimulus Type (F(2, 46) = 2.18, p =.12) or Hemisphere (F(2, 23) = 2.82, p =.11) and no interaction between the two (F(2, 46) = 1.4, p =.26) were observed ms A main effect of Stimulus Type (F(2, 46) = 3.93, p <.05) was found, with more negative-going potentials for words than for pseudowords (p <.05). The effect of Hemisphere (F(2, 23) = 2.49, p =.13) or the interaction between Stimulus Type and Hemisphere (F(2, 46) = 1.49, p =.23) were not significant ms ERPs differed according to Stimulus Type (F(2, 46) = 14.26, p <.01) with more negative-going potentials for words than for pseudowords (p <.05) and letter strings (p <.01). The effect of Hemisphere (F(2, 23) = 2.50, p =.13) or the interaction between Stimulus Type and Hemisphere (F < 1) were not significant ms Stimulus Type was significant (F(2,46) = 18.38; p <.01) due to words being more negative than both pseudowords (p <.05) and letter strings (p <.01). Pseudowords were also more negative than letter strings (p <.05). The hemispheres did not differ significantly (F(1, 23) = 2.64, p =.08) and the interaction was far from significant (F < 1) ms Finally, a centro-parietal, N400-like effect, was tested. This yielded a main effect of Stimulus Type (F(2,46) = 5.65; p < 0.01). Pseudowords were associated with more negative-going ERPs than

8 ARTICLE IN PRESS G Model 8 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx Fig. 5. Probability maps from 0 to 680 ms in steps of 32 ms (on average p <.01 for at least 32 ms) depicting the effects of Stimulus Type on ERPs during reading of positive words, pseudoword, or letter strings derived from negative words. (a) Point-wise ANOVA comparing words, pseudowords, and letter strings. (b) t-tests comparing words and pseudowords. (c) t-tests contrasting pseudowords and letter strings. Black frames mark time windows that were significant already in the ANOVA and can therefore be interpreted in the post hoc comparison. words (p <.01) and letter strings (p =.01). The latter did not differ (p >.2) Effects of Stimulus Type: positive valence Point-wise ANOVA Fig. 5a displays the result of a point-wise ANOVA comparing the cortical processing of positive words and pseudowords and letter strings. Fig. 6b illustrates the underlying topographic differences. Initial effects of Stimulus Type are seen over right frontal scalp from 216 ms. From 252 ms additional effects are found over left occipital and right parietal scalp, gradually extending in frontal and parietal regions until 356 ms. In the following time windows, effects gradually diminish and finally disappear by 612 ms Pseudowords versus Letter strings. Differentiation between pseudowords and letter strings, reflecting a formal analysis stage, first starts right-frontally from 216 ms (Fig. 5c). Although the occipital negativity for letter strings observed for the neutral and negative valences is likewise visible (Fig. 7c), here it falls short of significance. Instead, a left frontal negativity is significant until 284 ms. Subsequently, an initially right dominant posterior negativity for pseudowords appears ( ms) gradually becoming left-dominant, and moving more anterior by 500 ms (Figs. 5c and 7c) Words versus pseudowords. Linear contrasts reveal that the differentiation between positive words and pseudowords starts from 216 ms (Fig. 5 b). It is first significant as a frontal positivity (Fig. 6 b). The concomitant left occipital negativity is visible in this time window, but only reaches significance from 252 ms, lasting until 392 ms (Fig. 6b). From 432 ms the frontal positivity starts to extend centrally, reverting to frontal regions by 572 ms, this process reflecting the N400 effect. An additional brief occipital positivity is seen between 576 and 608 ms Spatio-temporal cluster ANOVA: effect of Stimulus Type for positive valence To further validate the above effects, the same regionsof-interest analysis as above was conducted. Topographical differences underlying the reported effects are shown in Figs. 6c and 7c ms Although over the right hemisphere words and pseudowords were somewhat more positive-going than letter strings, neither a clear effect of Stimulus Type (F(2, 46) = 1.7, p <.2), nor of Hemisphere (p >.4) or an interaction (F(2, 46) = 2.13, p <.2) were found ms No effects of Stimulus Type (F(2, 46) = 1.42, p <.2) or Hemisphere (F(2, 23) = 1.81, p =.19) were found, but an interaction between

9 G Model ARTICLE IN PRESS J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx 9 Fig. 6. Topographic difference maps of ERPs elicited by words versus pseudowords in different emotion categories from 0 to 680 ms in steps of 32 ms. (a) Neutral words minus pseudowords. (b) Negative words minus pseudowords. (c) Positive words minus pseudowords. Black frames mark time windows where significant effects were found in the respective statistical comparisons. Stimulus Type X Hemisphere (F(2, 46) = 5.1, p =.01 emerged. It indicated that over the left hemisphere words were more negative than pseudowords (p <.01), whereas over the right hemisphere words were more positive than letter strings (p <.05), not differing from pseudowords. Words were also more negative over the left than over the right hemisphere (p <.01), whereas pseudowords or letter strings showed no asymmetry ms A main effect of Stimulus Type (F(2, 46) = 4.86, p =.01), with more negative-going potentials for words than for pseudowords (p <.05), as well as a significant effect of Hemisphere (F(1, 23) = 6.93, p <.05), with more negativity over the left hemisphere, were identified. The interaction between Stimulus Type and Hemisphere (F(2, 46) = 6.07, p <.01) indicated more negative-going potentials over the left hemisphere elicited by words than by pseudowords or letter strings (p <.01). No such difference was observed over the right hemisphere (p >.2) ms A main effect of Stimulus Type (F(2, 46) = 16.93, p <.01) with more negative-going potentials for words than for pseudowords (p =.05) and letter strings (p <.01) as well as more negative-going potentials for pseudowords than for words (p <.05) were found. The effect of Hemisphere or the interaction between Stimulus Type and Hemisphere were not significant (F < 1) ms ERPs differed depending on Stimulus Types (F(2,46) = 16.48; p <.01), words being more negative than both pseudowords (p <.01) and letter strings (p <.01). The latter were not significantly different. The hemispheres did not differ (F(1, 23) = 1.60, p >.2) and the interaction was far from significant (F < 1) ms Finally, a centro-parietal N400-like effect yielded a main effect of Stimulus Type (F(2,46) = 4.52; p < 0.05). Pseudowords were associated with more negative-going ERPs than words (p <.05). Letter strings fell between words and pseudowords but did not differ significantly from either (both p >.1). 4. Discussion The present study investigated the time course and topography of emotion and lexicality effects in silent reading. It replicates and extends several findings about neural mechanisms of word processing, in general, and emotion word processing in particular. But in combination of the two, the study yields a remarkable new finding: Word pseudoword differentiation is faster for emotional than for neutral words, indicating faster lexical access to emotional than to neutral words. Replicating previous research, an EPN to emotional compared to neutral words was found between 200 and 300 ms after stimulus onset (Herbert et al., 2008; Hinojosa et al., 2010; Kissler et al.,

10 ARTICLE IN PRESS G Model 10 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx Fig. 7. Topographic difference maps of ERPs elicited by pseudowords versus letter strings derived from different emotion categories from 0 to 680 ms in steps of 32 ms. (a) Pseudowords and letter strings derived from neutral words. (b) Pseudowords and letter strings derived from negative words. (c) Pseudowords and letter strings derived from positive words. Black frames mark time windows where significant effects were found in the respective statistical comparisons. 2007; Scott et al., 2009). Also, a second posterior emotion effect was identified between 300 and 400 ms, similar to the effects previously reported by Schacht and her colleagues (Schacht and Sommer, 2009a, 2009b). Perhaps, this later part of the effect is attenuated and obscured in RSVP designs. Regarding the time-line of reading, an occipital positivity and a concomitant left frontal negativity were found for pseudowords from about 180 ms after stimulus onset, reflecting orthographical word-form processing. Orthographical word-form processing was assessed comparing pseudowords with letter strings so as not to confound it further with emotion effects carried by the words. Pseudoword letter string differentiation occurred as a bilateral posterior negativity for pseudowords by about 200 ms lasting until about 500 ms. There was some variation in the onset of the orthographical differentiation in that the effect became significant later (from 216 ms) for stimuli derived from positive words than from negative or neutral words (from 180 ms). The reason for this is currently unclear, but because the topographic differences between pseudowords and letter strings were similar for all three valences (Fig. 7), it could be due to random noise effects. Crucially, however, the differentiation between words and pseudowords, that is, between word forms that possess an actual lexical entry, and word forms that do not, was found to differ according to emotional content, occurring in the early EPN window ( ms) for emotional words, but only later for neutral words. In detail, in neutral words, word pseudoword differentiation occurred as a left occipito-temporal negativity from 324 ms after stimulus onset. By contrast, in negative words a left occipito-temporal negativity significantly differentiated word from pseudowords from 252 ms. In positive words the effect was significant already from 216 ms. Since pseudowords and letter strings were derived from the words within one valence, an influence of simple perceptual features is unlikely, since physical content did not vary between stimulus types (words, pseudowords, letter strings) within one valence. Thus, lexical access in reading was found to be faster in emotional than in neutral words. The timing and topography of this effect suggests that, at least in silent reading, a timing-difference in lexical access between emotional and neutral stimuli drives at least the early ( ms) emotion-word EPN effect. In other words, emotional content accelerates lexical access. This acceleration may contribute to the faster lexical decisions for emotional words seen in lexical-decision tasks (Kousta et al., 2009; Schacht and Sommer, 2009b). Whether the difference in onset between positive and negative words is meaningful and replicable is open to future research. While valence-dependent effects are sometimes found in lexical decisions, occasionally also with faster reaction times for positive words (e.g. Kissler and Koessler, 2011), arousal-dependent effects with acceleration for both positive and negative valence are more typical (Kousta et al., 2009). In any case, comparing the topographies of the effects, the differences between positive and negative words and pseudowords are clearly more similar regarding timing and topography than the difference between neutral words and pseudowords. Fig. 8 illustrates this major result of this experiment as ERP time course at

11 G Model ARTICLE IN PRESS J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx 11 Fig. 8. ERP time course at selected sensors for emotional words (red), neutral words (blue), pseudowords (black solid), and letter strings (black dashed). To facilitate inspection of major effects, ERPs are collapsed across valence (positive and negative) for emotional words. Pseudowords and letter strings are collapsed across valences of the original words. six representative sensors, collapsing across negative and positive words. Pre- and post-lexical effects of emotional content were also observed: A posterior negativity driven primarily by negative words was observed already from about 110 ms after word onset. This early emotion effect may replicate the posterior N1 effect for negative words reported by Scott et al. (2009). N1 emotion effects are reported less consistently than later effects but are in line with findings of a negativity bias towards high arousing material (Cacioppo, 2004; Ito and Cacioppo, 2000). This negativity bias is also compatible with Pratto and John s (1991) automatic vigilance model. From an emotion-theory perspective, it would support the long-held view that emotional responses or preferences, need no inferences (Zajonc, 1980), or at least fewer inferences than other types of processing, suggesting that emotional processing can operate outside the boundaries of cognitive information-processing. At any rate, recent work by Schacht and her group actually provide evidence that very early, presumably pre-lexical, emotion effects in word processing can be induced by associating a previously meaningless symbol with reward value in an operant learning procedure (Schacht et al., 2012). However, since the negative words have somewhat, although not significantly, lower frequencies than positive and neutral words, a covert frequency effect also needs to be considered. In this regard, Scott et al. (2009) report stronger P1 and N1 effects for high frequency than for low frequency words, suggesting that subtle differences in word frequency, with somewhat less frequent negative words should have attenuated, rather than accentuated these effects. Consistent with previous literature, emotion effects were also found in a later time window, the LPP window, and in the present, as well as in some previous studies (Herbert et al., 2008; Kissler et al., 2009), the more pronounced posterior-parietal positivity was restricted to positive contents. Again, the possibility of subtle imbalances in non-emotional stimulus properties needs to be considered, precluding firm conclusions. However the entire timing pattern of effects lends itself to an interpretation in terms of a plausible adaptive processing sequence, perhaps most consistent with appraisal theory (Grandjean and Scherer, 2008): It may be advantageous to initially (even pre-lexically) rapidly and selectively process negative stimuli. At a mid-latency stage, which here has been shown to coincide with lexical access to emotional words (EPN), the organism may process arousing stimuli regardless of valence. Finally, at a post-lexical stage, pleasant stimuli may enjoy an advantage. Whereas rapid processing of negative contents may be survival critical, later extended evaluation of pleasurable material may be important for subjective well being. The most important finding of the present study is that lexical access for emotional words is faster than lexical access for neutral words. This merits further discussion in the context of the word-processing literature in general. There is wide agreement about the existence of distinct orthographic, lexical, and

12 ARTICLE IN PRESS G Model 12 J. Kissler, C. Herbert / Biological Psychology xxx (2012) xxx xxx semantic processing levels, presently modeled by pseudoword versus letter string, word versus pseudoword, and emotional content manipulations, respectively. Considerable debate, however, concerns the timing of different processing steps in visual word processing and the degree to which these steps occur in sequence, in parallel, or in interaction. Perhaps the most influential view assumes interactive and cascaded processing, allowing for both serial processing steps and their interaction via feedback mechanisms (Coltheart et al., 2001). Empirically, some data support very rapid initial activation of the mental lexicon, chiefly supported by eye-movement and ERP evidence from lexical decision. Word frequency effects around 150 ms are taken as perhaps the first indicators of lexical access (Sereno and Rayner, 2003; Sereno et al., 1998). Early word frequency by emotion interactions have also been reported in two previous studies (Palazova et al., 2011; Scott et al., 2009), but the present study did not manipulate this variable. Although the mental lexicon clearly uses perceptual frequency as an organizational principle to access its content, this is a rather low-level principle that may not be specific to words. For instance, orthographic typicality effects are highly similar for words and pseudowords and can also been found around 100 ms (Hauk et al., 2006). Using other experimental manipulations, magnetoencephalography (MEG) data indicate lexical processing considerably later, only around 350 ms after word onset (Pylkkanen and Marantz, 2003; Pylkkanen et al., 2002). The present results found lexicality effects as reflected by word pseudoword differentiation for neutral words from about 324 ms onwards, similar to the above MEG evidence. However, effects occurred for negative words from 252 ms and for positive words already from 216 ms. Palazova et al. (2011) report word pseudoword differentiation for nouns from 270 ms in a lexical-decision task, but included the emotional words in their word pseudoword comparison. If averaged across valence, both studies would yield a similar timing of lexical access. Present results suggest that part of the variability of reports on the timing of lexical access may be due to differential lexical access for different semantic classes. It is open to future research, whether such effects are restricted to emotional content or might also occur for other semantic classes, if these are explicitly attended to. Future studies will also elucidate, whether faster lexical access for emotional words is specific to reading designs. There is the possibility that intrinsically motivated attention to emotion speeds up processing, but that this acceleration disappears when attention is explicitly drawn to the lexicality dimension, as in lexical-decision tasks, consistent with evidence that the timing of word pseudoword differentiation in the ERP can vary with the task at hand (Ziegler et al., 1997). Alternatively, faster lexical access to emotional words may be a fairly general phenomenon that at least partly accounts for faster lexical-decision times for emotional words. The present results support, at least during initial processing phases, sequentially initiated, partly overlapping processing stages with an initial orthographic (from 180 ms) and a later lexical phase (from about 220 ms, depending on valence). A similar sequence was described across various tasks by Bentin et al. (Bentin et al., 1999). They found in early time windows major differences between words and pseudowords on the one hand and letter strings on the other hand and only comparatively late identified word pseudoword differences, peaking around 400 ms, when semantic integration is also assumed to occur. Previous reports have linked the emotion-word EPN effect to semantic processing and indeed, the time window that carries the lexicality effect in emotional words also, and by virtue of this, comprises an emotion-neutral differentiation. Indeed, responses to stimulus semantics could in general coincide with or precede other forms of lexical analysis or task requirements might alter the time course and perhaps also the general occurrence of attribute-specific activations in word processing. On the one hand, distinct responses to words of different semantic categories have been reported between 200 and 300 ms after word onset (Dehaene, 1995; Hauk et al., 2008; Hinojosa et al., 2004). On the other hand, under certain experimental manipulations that limit attentional resources even a total absence of semantic processing, which is otherwise assumed to be automatic during reading, has been observed (Smith et al., 2001). Because at least in non-competitive situations all kinds of emotional stimuli attract attention, and because emotional stimuli often signal the need for rapid action, they may result in differential brain responses even in situations where otherwise apparently very little higher level analysis occurs early on. Experimental test should demonstrate whether the EPN effect represents early semantic analysis and determine the degree to which it depends on attentional engagement and current motivational demands. Demands imposed by the experimenter, for instance in the form of attention to a specific semantic category, might be able to mimic effects currently found for emotional words that possess an intrinsic attentional and motivational relevance. Still, present results demonstrate that emotional content accelerates lexical access in silent reading tasks. Several later effects of both stimulus type and emotional content also appeared, again replicating and extending previous findings. Regarding stimulus type, from about 360 to 500 ms after stimulus presentation, more negative-going centro-parietal ERPs were observed for pseudowords, essentially replicating the well-known N400 effect, presumably indicating extended lexical search processes for word-like stimuli for which no lexical entry has yet been found. Finally, although not the focus of the present analyses, around 600 ms post stimulus, words elicited relatively more positivity over the left hemisphere than letter strings and pseudowords which appears consistent with a P600 effect in word reading. Although the P600 is mostly seen in syntactic tasks, a general attentional interpretation of late positive potentials would suggest that words, because they contain more information than pseudowords and letter strings attract more attentional resources, resulting in a larger P600. Also, effects of semantics on the P600 have been observed (van Herten et al., 2005). In parallel with the later effects of stimulus type, further effects of emotional content were also observed. A second posterior negativity for emotional words appeared between 320 and 420 ms. It showed a similar topography as the previous EPN and was likewise driven by arousal, differentiating between positive and negative words on the one hand and neutral words on the other hand. Regarding its timing and topography, this effect may resemble the EPN effect described by Schacht and colleagues who described a posterior negativity for emotional words emerging in lexicaldecision or evaluation tasks, similar to the EPN (Palazova et al., 2011; Schacht and Sommer, 2009a, 2009b). Perhaps this effect is obscured in the RSVP designs that we have previously used. At any rate, this second EPN effect parallels word pseudoword differentiation. Finally, a right-lateralized late positive potential revealed an emotion effect around 500 ms. Replicating previous research (Herbert et al., 2008; Herbert et al., 2006; Kissler et al., 2009; Schapkin et al., 2000), this positivity was more pronounced for positive words and differentiated them from both neutral and negative ones. In this time window a frontal positivity was also identified, being larger for emotionally arousing than for neutral words. The effect, which to our knowledge has not been described before, may reflect the frontal cortex s well-documented role in the evaluation of emotional stimuli (for review, see Etkin et al., 2011). Its identification may be due to the higher spatial sampling of the present compared to previous studies. In sum, the present emotion effects replicate previous reports of N1, EPN and LPP modulations in single word reading outside the RSVP design. Remarkably, against a time-line of structural and